Quelle  ringbuf.c   Sprache: C

// SPDX-License-Identifier: GPL-2.0
#define _GNU_SOURCE
#include <linux/compiler.h>
#include <asm/barrier.h>
#include <test_progs.h>
#include <sys/mman.h>
#include <sys/epoll.h>
#include <time.h>
#include <sched.h>
#include <signal.h>
#include <pthread.h>
#include <sys/sysinfo.h>
#include <linux/perf_event.h>
#include <linux/ring_buffer.h>

#include "test_ringbuf.lskel.h"
#include "test_ringbuf_n.lskel.h"
#include "test_ringbuf_map_key.lskel.h"
#include "test_ringbuf_write.lskel.h"

#define EDONE 7777

static int duration = 0;

struct sample {
 int pid;
 int seq;
 long value;
 char comm[16];
};

static int sample_cnt;

static void atomic_inc(int *cnt)
{
 __atomic_add_fetch(cnt, 1, __ATOMIC_SEQ_CST);
}

static int atomic_xchg(int *cnt, int val)
{
 return __atomic_exchange_n(cnt, val, __ATOMIC_SEQ_CST);
}

static int process_sample(void *ctx, void *data, size_t len)
{
 struct sample *s = data;

 atomic_inc(&sample_cnt);

 switch (s->seq) {
 case 0:
  CHECK(s->value != 333, "sample1_value", "exp %ld, got %ld\n",
        333L, s->value);
  return 0;
 case 1:
  CHECK(s->value != 777, "sample2_value", "exp %ld, got %ld\n",
        777L, s->value);
  return -EDONE;
 default:
  /* we don't care about the rest */
  return 0;
 }
}

static struct test_ringbuf_map_key_lskel *skel_map_key;
static struct test_ringbuf_lskel *skel;
static struct ring_buffer *ringbuf;

static void trigger_samples()
{
 skel->bss->dropped = 0;
 skel->bss->total = 0;
 skel->bss->discarded = 0;

 /* trigger exactly two samples */
 skel->bss->value = 333;
 syscall(__NR_getpgid);
 skel->bss->value = 777;
 syscall(__NR_getpgid);
}

static void *poll_thread(void *input)
{
 long timeout = (long)input;

 return (void *)(long)ring_buffer__poll(ringbuf, timeout);
}

static void ringbuf_write_subtest(void)
{
 struct test_ringbuf_write_lskel *skel;
 int page_size = getpagesize();
 size_t *mmap_ptr;
 int err, rb_fd;

 skel = test_ringbuf_write_lskel__open();
 if (!ASSERT_OK_PTR(skel, "skel_open"))
  return;

 skel->maps.ringbuf.max_entries = 0x40000;

 err = test_ringbuf_write_lskel__load(skel);
 if (!ASSERT_OK(err, "skel_load"))
  goto cleanup;

 rb_fd = skel->maps.ringbuf.map_fd;

 mmap_ptr = mmap(NULL, page_size, PROT_READ | PROT_WRITE, MAP_SHARED, rb_fd, 0);
 if (!ASSERT_OK_PTR(mmap_ptr, "rw_cons_pos"))
  goto cleanup;
 *mmap_ptr = 0x30000;
 ASSERT_OK(munmap(mmap_ptr, page_size), "unmap_rw");

 skel->bss->pid = getpid();

 ringbuf = ring_buffer__new(rb_fd, process_sample, NULL, NULL);
 if (!ASSERT_OK_PTR(ringbuf, "ringbuf_new"))
  goto cleanup;

 err = test_ringbuf_write_lskel__attach(skel);
 if (!ASSERT_OK(err, "skel_attach"))
  goto cleanup_ringbuf;

 skel->bss->discarded = 0;
 skel->bss->passed = 0;

 /* trigger exactly two samples */
 syscall(__NR_getpgid);
 syscall(__NR_getpgid);

 ASSERT_EQ(skel->bss->discarded, 2, "discarded");
 ASSERT_EQ(skel->bss->passed, 0, "passed");

 test_ringbuf_write_lskel__detach(skel);
cleanup_ringbuf:
 ring_buffer__free(ringbuf);
cleanup:
 test_ringbuf_write_lskel__destroy(skel);
}

static void ringbuf_subtest(void)
{
 const size_t rec_sz = BPF_RINGBUF_HDR_SZ + sizeof(struct sample);
 pthread_t thread;
 long bg_ret = -1;
 int err, cnt, rb_fd;
 int page_size = getpagesize();
 void *mmap_ptr, *tmp_ptr;
 struct ring *ring;
 int map_fd;
 unsigned long avail_data, ring_size, cons_pos, prod_pos;

 skel = test_ringbuf_lskel__open();
 if (CHECK(!skel, "skel_open", "skeleton open failed\n"))
  return;

 skel->maps.ringbuf.max_entries = page_size;

 err = test_ringbuf_lskel__load(skel);
 if (CHECK(err != 0, "skel_load", "skeleton load failed\n"))
  goto cleanup;

 rb_fd = skel->maps.ringbuf.map_fd;
 /* good read/write cons_pos */
 mmap_ptr = mmap(NULL, page_size, PROT_READ | PROT_WRITE, MAP_SHARED, rb_fd, 0);
 ASSERT_OK_PTR(mmap_ptr, "rw_cons_pos");
 tmp_ptr = mremap(mmap_ptr, page_size, 2 * page_size, MREMAP_MAYMOVE);
 if (!ASSERT_ERR_PTR(tmp_ptr, "rw_extend"))
  goto cleanup;
 ASSERT_ERR(mprotect(mmap_ptr, page_size, PROT_EXEC), "exec_cons_pos_protect");
 ASSERT_OK(munmap(mmap_ptr, page_size), "unmap_rw");

 /* bad writeable prod_pos */
 mmap_ptr = mmap(NULL, page_size, PROT_WRITE, MAP_SHARED, rb_fd, page_size);
 err = -errno;
 ASSERT_ERR_PTR(mmap_ptr, "wr_prod_pos");
 ASSERT_EQ(err, -EPERM, "wr_prod_pos_err");

 /* bad writeable data pages */
 mmap_ptr = mmap(NULL, page_size, PROT_WRITE, MAP_SHARED, rb_fd, 2 * page_size);
 err = -errno;
 ASSERT_ERR_PTR(mmap_ptr, "wr_data_page_one");
 ASSERT_EQ(err, -EPERM, "wr_data_page_one_err");
 mmap_ptr = mmap(NULL, page_size, PROT_WRITE, MAP_SHARED, rb_fd, 3 * page_size);
 ASSERT_ERR_PTR(mmap_ptr, "wr_data_page_two");
 mmap_ptr = mmap(NULL, 2 * page_size, PROT_WRITE, MAP_SHARED, rb_fd, 2 * page_size);
 ASSERT_ERR_PTR(mmap_ptr, "wr_data_page_all");

 /* good read-only pages */
 mmap_ptr = mmap(NULL, 4 * page_size, PROT_READ, MAP_SHARED, rb_fd, 0);
 if (!ASSERT_OK_PTR(mmap_ptr, "ro_prod_pos"))
  goto cleanup;

 ASSERT_ERR(mprotect(mmap_ptr, 4 * page_size, PROT_WRITE), "write_protect");
 ASSERT_ERR(mprotect(mmap_ptr, 4 * page_size, PROT_EXEC), "exec_protect");
 ASSERT_ERR_PTR(mremap(mmap_ptr, 0, 4 * page_size, MREMAP_MAYMOVE), "ro_remap");
 ASSERT_OK(munmap(mmap_ptr, 4 * page_size), "unmap_ro");

 /* good read-only pages with initial offset */
 mmap_ptr = mmap(NULL, page_size, PROT_READ, MAP_SHARED, rb_fd, page_size);
 if (!ASSERT_OK_PTR(mmap_ptr, "ro_prod_pos"))
  goto cleanup;

 ASSERT_ERR(mprotect(mmap_ptr, page_size, PROT_WRITE), "write_protect");
 ASSERT_ERR(mprotect(mmap_ptr, page_size, PROT_EXEC), "exec_protect");
 ASSERT_ERR_PTR(mremap(mmap_ptr, 0, 3 * page_size, MREMAP_MAYMOVE), "ro_remap");
 ASSERT_OK(munmap(mmap_ptr, page_size), "unmap_ro");

 /* only trigger BPF program for current process */
 skel->bss->pid = getpid();

 ringbuf = ring_buffer__new(skel->maps.ringbuf.map_fd,
       process_sample, NULL, NULL);
 if (CHECK(!ringbuf, "ringbuf_create", "failed to create ringbuf\n"))
  goto cleanup;

 err = test_ringbuf_lskel__attach(skel);
 if (CHECK(err, "skel_attach", "skeleton attachment failed: %d\n", err))
  goto cleanup;

 trigger_samples();

 ring = ring_buffer__ring(ringbuf, 0);
 if (!ASSERT_OK_PTR(ring, "ring_buffer__ring_idx_0"))
  goto cleanup;

 map_fd = ring__map_fd(ring);
 ASSERT_EQ(map_fd, skel->maps.ringbuf.map_fd, "ring_map_fd");

 /* 2 submitted + 1 discarded records */
 CHECK(skel->bss->avail_data != 3 * rec_sz,
       "err_avail_size", "exp %ld, got %ld\n",
       3L * rec_sz, skel->bss->avail_data);
 CHECK(skel->bss->ring_size != page_size,
       "err_ring_size", "exp %ld, got %ld\n",
       (long)page_size, skel->bss->ring_size);
 CHECK(skel->bss->cons_pos != 0,
       "err_cons_pos", "exp %ld, got %ld\n",
       0L, skel->bss->cons_pos);
 CHECK(skel->bss->prod_pos != 3 * rec_sz,
       "err_prod_pos", "exp %ld, got %ld\n",
       3L * rec_sz, skel->bss->prod_pos);

 /* verify getting this data directly via the ring object yields the same
 * results
 */
 avail_data = ring__avail_data_size(ring);
 ASSERT_EQ(avail_data, 3 * rec_sz, "ring_avail_size");
 ring_size = ring__size(ring);
 ASSERT_EQ(ring_size, page_size, "ring_ring_size");
 cons_pos = ring__consumer_pos(ring);
 ASSERT_EQ(cons_pos, 0, "ring_cons_pos");
 prod_pos = ring__producer_pos(ring);
 ASSERT_EQ(prod_pos, 3 * rec_sz, "ring_prod_pos");

 /* poll for samples */
 err = ring_buffer__poll(ringbuf, -1);

 /* -EDONE is used as an indicator that we are done */
 if (CHECK(err != -EDONE, "err_done", "done err: %d\n", err))
  goto cleanup;
 cnt = atomic_xchg(&sample_cnt, 0);
 CHECK(cnt != 2, "cnt", "exp %d samples, got %d\n", 2, cnt);

 /* we expect extra polling to return nothing */
 err = ring_buffer__poll(ringbuf, 0);
 if (CHECK(err != 0, "extra_samples", "poll result: %d\n", err))
  goto cleanup;
 cnt = atomic_xchg(&sample_cnt, 0);
 CHECK(cnt != 0, "cnt", "exp %d samples, got %d\n", 0, cnt);

 CHECK(skel->bss->dropped != 0, "err_dropped", "exp %ld, got %ld\n",
       0L, skel->bss->dropped);
 CHECK(skel->bss->total != 2, "err_total", "exp %ld, got %ld\n",
       2L, skel->bss->total);
 CHECK(skel->bss->discarded != 1, "err_discarded", "exp %ld, got %ld\n",
       1L, skel->bss->discarded);

 /* now validate consumer position is updated and returned */
 trigger_samples();
 CHECK(skel->bss->cons_pos != 3 * rec_sz,
       "err_cons_pos", "exp %ld, got %ld\n",
       3L * rec_sz, skel->bss->cons_pos);
 err = ring_buffer__poll(ringbuf, -1);
 CHECK(err <= 0, "poll_err", "err %d\n", err);
 cnt = atomic_xchg(&sample_cnt, 0);
 CHECK(cnt != 2, "cnt", "exp %d samples, got %d\n", 2, cnt);

 /* start poll in background w/ long timeout */
 err = pthread_create(&thread, NULL, poll_thread, (void *)(long)10000);
 if (CHECK(err, "bg_poll", "pthread_create failed: %d\n", err))
  goto cleanup;

 /* turn off notifications now */
 skel->bss->flags = BPF_RB_NO_WAKEUP;

 /* give background thread a bit of a time */
 usleep(50000);
 trigger_samples();
 /* sleeping arbitrarily is bad, but no better way to know that
 * epoll_wait() **DID NOT** unblock in background thread
 */
 usleep(50000);
 /* background poll should still be blocked */
 err = pthread_tryjoin_np(thread, (void **)&bg_ret);
 if (CHECK(err != EBUSY, "try_join", "err %d\n", err))
  goto cleanup;

 /* BPF side did everything right */
 CHECK(skel->bss->dropped != 0, "err_dropped", "exp %ld, got %ld\n",
       0L, skel->bss->dropped);
 CHECK(skel->bss->total != 2, "err_total", "exp %ld, got %ld\n",
       2L, skel->bss->total);
 CHECK(skel->bss->discarded != 1, "err_discarded", "exp %ld, got %ld\n",
       1L, skel->bss->discarded);
 cnt = atomic_xchg(&sample_cnt, 0);
 CHECK(cnt != 0, "cnt", "exp %d samples, got %d\n", 0, cnt);

 /* clear flags to return to "adaptive" notification mode */
 skel->bss->flags = 0;

 /* produce new samples, no notification should be triggered, because
 * consumer is now behind
 */
 trigger_samples();

 /* background poll should still be blocked */
 err = pthread_tryjoin_np(thread, (void **)&bg_ret);
 if (CHECK(err != EBUSY, "try_join", "err %d\n", err))
  goto cleanup;

 /* still no samples, because consumer is behind */
 cnt = atomic_xchg(&sample_cnt, 0);
 CHECK(cnt != 0, "cnt", "exp %d samples, got %d\n", 0, cnt);

 skel->bss->dropped = 0;
 skel->bss->total = 0;
 skel->bss->discarded = 0;

 skel->bss->value = 333;
 syscall(__NR_getpgid);
 /* now force notifications */
 skel->bss->flags = BPF_RB_FORCE_WAKEUP;
 skel->bss->value = 777;
 syscall(__NR_getpgid);

 /* now we should get a pending notification */
 usleep(50000);
 err = pthread_tryjoin_np(thread, (void **)&bg_ret);
 if (CHECK(err, "join_bg", "err %d\n", err))
  goto cleanup;

 if (CHECK(bg_ret <= 0, "bg_ret", "epoll_wait result: %ld", bg_ret))
  goto cleanup;

 /* due to timing variations, there could still be non-notified
 * samples, so consume them here to collect all the samples
 */
 err = ring_buffer__consume(ringbuf);
 CHECK(err < 0, "rb_consume", "failed: %d\b", err);

 /* also consume using ring__consume to make sure it works the same */
 err = ring__consume(ring);
 ASSERT_GE(err, 0, "ring_consume");

 /* 3 rounds, 2 samples each */
 cnt = atomic_xchg(&sample_cnt, 0);
 CHECK(cnt != 6, "cnt", "exp %d samples, got %d\n", 6, cnt);

 /* BPF side did everything right */
 CHECK(skel->bss->dropped != 0, "err_dropped", "exp %ld, got %ld\n",
       0L, skel->bss->dropped);
 CHECK(skel->bss->total != 2, "err_total", "exp %ld, got %ld\n",
       2L, skel->bss->total);
 CHECK(skel->bss->discarded != 1, "err_discarded", "exp %ld, got %ld\n",
       1L, skel->bss->discarded);

 test_ringbuf_lskel__detach(skel);
cleanup:
 ring_buffer__free(ringbuf);
 test_ringbuf_lskel__destroy(skel);
}

/*
 * Test ring_buffer__consume_n() by producing N_TOT_SAMPLES samples in the ring
 * buffer, via getpid(), and consuming them in chunks of N_SAMPLES.
 */
#define N_TOT_SAMPLES 32
#define N_SAMPLES 4

/* Sample value to verify the callback validity */
#define SAMPLE_VALUE 42L

static int process_n_sample(void *ctx, void *data, size_t len)
{
 struct sample *s = data;

 ASSERT_EQ(s->value, SAMPLE_VALUE, "sample_value");

 return 0;
}

static void ringbuf_n_subtest(void)
{
 struct test_ringbuf_n_lskel *skel_n;
 int err, i;

 skel_n = test_ringbuf_n_lskel__open();
 if (!ASSERT_OK_PTR(skel_n, "test_ringbuf_n_lskel__open"))
  return;

 skel_n->maps.ringbuf.max_entries = getpagesize();
 skel_n->bss->pid = getpid();

 err = test_ringbuf_n_lskel__load(skel_n);
 if (!ASSERT_OK(err, "test_ringbuf_n_lskel__load"))
  goto cleanup;

 ringbuf = ring_buffer__new(skel_n->maps.ringbuf.map_fd,
       process_n_sample, NULL, NULL);
 if (!ASSERT_OK_PTR(ringbuf, "ring_buffer__new"))
  goto cleanup;

 err = test_ringbuf_n_lskel__attach(skel_n);
 if (!ASSERT_OK(err, "test_ringbuf_n_lskel__attach"))
  goto cleanup_ringbuf;

 /* Produce N_TOT_SAMPLES samples in the ring buffer by calling getpid() */
 skel_n->bss->value = SAMPLE_VALUE;
 for (i = 0; i < N_TOT_SAMPLES; i++)
  syscall(__NR_getpgid);

 /* Consume all samples from the ring buffer in batches of N_SAMPLES */
 for (i = 0; i < N_TOT_SAMPLES; i += err) {
  err = ring_buffer__consume_n(ringbuf, N_SAMPLES);
  if (!ASSERT_EQ(err, N_SAMPLES, "rb_consume"))
   goto cleanup_ringbuf;
 }

cleanup_ringbuf:
 ring_buffer__free(ringbuf);
cleanup:
 test_ringbuf_n_lskel__destroy(skel_n);
}

static int process_map_key_sample(void *ctx, void *data, size_t len)
{
 struct sample *s;
 int err, val;

 s = data;
 switch (s->seq) {
 case 1:
  ASSERT_EQ(s->value, 42, "sample_value");
  err = bpf_map_lookup_elem(skel_map_key->maps.hash_map.map_fd,
       s, &val);
  ASSERT_OK(err, "hash_map bpf_map_lookup_elem");
  ASSERT_EQ(val, 1, "hash_map val");
  return -EDONE;
 default:
  return 0;
 }
}

static void ringbuf_map_key_subtest(void)
{
 int err;

 skel_map_key = test_ringbuf_map_key_lskel__open();
 if (!ASSERT_OK_PTR(skel_map_key, "test_ringbuf_map_key_lskel__open"))
  return;

 skel_map_key->maps.ringbuf.max_entries = getpagesize();
 skel_map_key->bss->pid = getpid();

 err = test_ringbuf_map_key_lskel__load(skel_map_key);
 if (!ASSERT_OK(err, "test_ringbuf_map_key_lskel__load"))
  goto cleanup;

 ringbuf = ring_buffer__new(skel_map_key->maps.ringbuf.map_fd,
       process_map_key_sample, NULL, NULL);
 if (!ASSERT_OK_PTR(ringbuf, "ring_buffer__new"))
  goto cleanup;

 err = test_ringbuf_map_key_lskel__attach(skel_map_key);
 if (!ASSERT_OK(err, "test_ringbuf_map_key_lskel__attach"))
  goto cleanup_ringbuf;

 syscall(__NR_getpgid);
 ASSERT_EQ(skel_map_key->bss->seq, 1, "skel_map_key->bss->seq");
 err = ring_buffer__poll(ringbuf, -1);
 ASSERT_EQ(err, -EDONE, "ring_buffer__poll");

cleanup_ringbuf:
 ring_buffer__free(ringbuf);
cleanup:
 test_ringbuf_map_key_lskel__destroy(skel_map_key);
}

void test_ringbuf(void)
{
 if (test__start_subtest("ringbuf"))
  ringbuf_subtest();
 if (test__start_subtest("ringbuf_n"))
  ringbuf_n_subtest();
 if (test__start_subtest("ringbuf_map_key"))
  ringbuf_map_key_subtest();
 if (test__start_subtest("ringbuf_write"))
  ringbuf_write_subtest();
}