Re: Revert "aio: fix aio request leak when events are reaped by user space"

[Jeff Moyer]

Dan Aloni <dan@xxxxxxxxxxxx> writes:

> I like your extension to the test program. I have a suggestion for
> the integration inside libaio's harness, since the original issue
> depended on the size of the ring buffer, let's have max_ios be picked
> from {ring->nr + 1, ring->nr - 1, ring->nr}*{1,2,3,4} for the various
> test cases. I guess Jeff who is maintaining it will have some ideas
> too.

I can't think of anything in addition to what you've mentioned.
Attached is a version that I've tested.  The only modifications were to
make it work with the libaio test harness.  If you have time to review
it, that would be great.  I've tested it on both working and failing
kernels.

Cheers,
Jeff

/*
 * aio-test-ring-buffer-overflow.c
 *
 * Copyright (C) 2014, Dan Aloni, Kernelim Ltd.
 * Copyright (C) 2014, Benjamin LaHaise <bcrl@xxxxxxxxx>.
 * Copyright (C) 2014, Jeff Moyer <jmoyer@xxxxxxxxxx>
 *
 * This program is free software; you can redistribute it and/or modify
 * it under the terms of the GNU General Public License version 2 as
 * published by the Free Software Foundation.
 *
 * This program is distributed in the hope that it will be useful,
 * but WITHOUT ANY WARRANTY; without even the implied warranty of
 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
 * General Public License for more details.
 */
#ifndef _GNU_SOURCE
#define _GNU_SOURCE 1
#endif
#include <assert.h>
#include <errno.h>
#include <libaio.h>
#include <stdbool.h>
#include <stdio.h>
#include <stdlib.h>
#include <sys/param.h>
#include <sys/stat.h>
#include <sys/types.h>
#include <sys/uio.h>
#include <unistd.h>
#include <sys/types.h>
#include <sys/stat.h>
#include <fcntl.h>
#include <sys/mman.h>
#include <sys/wait.h>

const int max_events = 32;
const int io_size = 0x1000;
struct iocb *io;
struct iocb **iops;
struct iovec *iovecs;
struct io_event *events;
char *data;

long submitted = 0;
long completed = 0;
long pending = 0;

#define SYS_IO_GETEVENTS 0
#define USER_GETEVENTS 1

static volatile sig_atomic_t done = 0;

struct aio_ring {
	unsigned	id;	/* kernel internal index number */
	unsigned	nr;	/* number of io_events */
	volatile unsigned	head;
	volatile unsigned	tail;

	unsigned	magic;
	unsigned	compat_features;
	unsigned	incompat_features;
	unsigned	header_length;	/* size of aio_ring */

	struct io_event io_events[0];
};

int get_ring_size(int nr_events)
{
	io_context_t ctx;
	int ret, ring_size;
	struct aio_ring *ring;

	memset(&ctx, 0, sizeof(ctx));
	ret = io_setup(nr_events, &ctx);
	assert(!ret);

	ring = (void *)ctx;
	ring_size = ring->nr;

	ret = io_destroy(ctx);
	assert(!ret);

	return ring_size;
}

int user_getevents(io_context_t ctx, int nr_events, struct io_event *event)
{
	struct aio_ring *ring = (void *)ctx;
	int completed = 0;
	while ((completed < nr_events) && (ring->head != ring->tail)) {
		unsigned new_head = ring->head;
		*event = ring->io_events[new_head];
		new_head += 1;
		new_head %= ring->nr;
		ring->head = new_head;
		completed++;
	}
	return completed;
}

void prune(io_context_t io_ctx, int max_ios, int getevents_type)
{
	int ret;

	if (getevents_type == USER_GETEVENTS)
        	ret = user_getevents(io_ctx, max_ios, events);
	else
        	ret = io_getevents(io_ctx, pending, max_ios, events, NULL);
	if (ret > 0) {
		printf("Completed: %d\n", ret);
		completed += ret;
		pending -= ret;
	}
}

void run_test(int max_ios, int getevents_type)
{
	int fd, ret;
	long i, to_submit;
	struct iocb **iocb_sub;
	io_context_t io_ctx;
	const char *filename = "testfile";

	memset(&io_ctx, 0, sizeof(io_ctx));
	ret = io_setup(max_events, &io_ctx);
	assert(!ret);

	io = calloc(max_ios, sizeof(*io));
	iops = calloc(max_ios, sizeof(*iops));
	iovecs = calloc(max_ios, sizeof(*iovecs));
	events = calloc(max_ios, sizeof(*events));

	unlink(filename);
	fd = open(filename, O_CREAT | O_RDWR | O_DIRECT, 0644);
	assert(fd >= 0);

	ret = ftruncate(fd, max_ios * io_size);
	assert(!ret);

	data = mmap(NULL, io_size * max_ios, PROT_READ | PROT_WRITE,
		    MAP_PRIVATE | MAP_ANONYMOUS, -1, 0);
	assert(data != MAP_FAILED);

	for (i = 0; i < max_ios; i++) {
		iops[i] = &io[i];
		io[i].data = io;
		iovecs[i].iov_base = &data[io_size * i];
		iovecs[i].iov_len = io_size;
		io_prep_preadv(&io[i], fd, &iovecs[i], 1, 0);
	}

	submitted = completed = pending = 0;

	to_submit = max_ios;
	iocb_sub = iops;

	while (submitted < max_ios) {
		printf("Submitting: %ld\n", to_submit);

		ret = io_submit(io_ctx, to_submit, iocb_sub);
		if (ret >= 0) {
			printf("Submitted: %d\n", ret);
			submitted += ret;
			iocb_sub += ret;
			pending += ret;
			to_submit -= ret;
		} else {
			if (ret == -EAGAIN) {
				printf("Submitted too much, that's okay\n");
				prune(io_ctx, max_ios, getevents_type);
			}
		}
	}

	prune(io_ctx, max_ios, getevents_type);
	io_destroy(io_ctx);
	close(fd);
	ret = munmap(data, io_size * max_ios);
	assert(!ret);

	printf("Verifying...\n");

	assert(completed == submitted);

	printf("OK\n");
}

void run_child(void)
{
	int ring_size;

	ring_size = get_ring_size(max_events);

	run_test(ring_size-1, SYS_IO_GETEVENTS);
	run_test(ring_size, SYS_IO_GETEVENTS);
	run_test(ring_size+1, SYS_IO_GETEVENTS);
	run_test(ring_size*2, SYS_IO_GETEVENTS);
	run_test(ring_size*4, SYS_IO_GETEVENTS);

	run_test(ring_size-1, USER_GETEVENTS);
	run_test(ring_size, USER_GETEVENTS);
	run_test(ring_size+1, USER_GETEVENTS);
	run_test(ring_size*2, USER_GETEVENTS);
	run_test(ring_size*4, USER_GETEVENTS);

	exit(0);
}

void sighandler(int signo)
{
	assert(signo == SIGCHLD);
	done = 1;
}

int test_main(void)
{
	unsigned int ret;
	sighandler_t oldhandler;
	pid_t child;

	switch (child = fork()) {
	case 0: /* child */
		run_child();
		break;
	case -1:
		perror("fork");
		exit(1);
	default:
		oldhandler = signal(SIGCHLD, sighandler);
		assert(oldhandler != SIG_ERR);
		break;
	}

	ret = sleep(10);
	if (ret != 0) {
		pid_t pid;
		int status;

		assert(done);

		pid = wait(&status);
		if (pid != child) {
			perror("wait");
			exit(1);
		}

		return WEXITSTATUS(status);
	}

	return 1; /* failed */
}
/*
 * Local Variables:
 *  mode: c
 *  c-basic-offset: 8
 * End:
 */
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/