探索性能巅峰：io_uring用户态接口的神奇之处
【摘要】 这篇文章将带你深入探索io_uring用户态接口的神奇之处，它是一项引人注目的技术，能够显著提升IO操作的性能。我们将介绍io_uring的工作原理，并解释它为什么在性能方面与传统接口相比具有明显优势。你将了解到io_uring的异步特性是如何实现的，以及它如何减少了对内核的系统调用次数。我们还将探讨io_uring在实际应用中的潜力和使用场景，以及如何利用它来优化你的应用程序。
一、io_uring和epoll的区别
（1）epoll设置事件完成之后，以后只要不修改或删除事件，就可以一直等待IO事件触发。即事件驱动机制。
epoll对事件的管理使用的是红黑树。
（2）io_uring有两个队列，SQ和CQ，io_uring_submite之后，事件提交在SQ等待，事件达到后交给CQ，应用程序调用io_uring_peek_batch_cqe从CQ取出后，会调用io_uring_cq_advance将事件触发销毁，因此要想一直可以等待事件，需要从CQ取出后再次把事件加入SQ中。即异步机制。
io_uring对事件的管理采用两个队列：SQ（submition queue）和CQ（completion queue）。
二、io_uring 与epoll性能比较
2.1、安装rust_echo_bench测试工具
（1）安装cargo：
# ubuntu
sudo apt-get install cargo
# centos
sudo yum install cargo（2）下载rust_echo_bench：
git clone https://github.com/haraldh/rust_echo_bench.git（3）编译rust_echo_bench：
cd rust_echo_bench
cargo run --release -- --help执行过程：
Updating crates.io index
Downloaded unicode-width v0.1.9
Downloaded getopts v0.2.21
Downloaded 2 crates (35.2 KB) in 1.58s
Compiling unicode-width v0.1.9
Compiling getopts v0.2.21
Compiling echo_bench v0.2.0 (/home/user/rust_echo_bench)
Finished release [optimized] target(s) in 4m 05s
Running `target/release/echo_bench --help`
Echo benchmark.
Usage:
target/release/echo_bench [ -a <address> ] [ -l <length> ] [ -c <number> ] [ -t <duration> ]
target/release/echo_bench (-h | --help)
target/release/echo_bench --version
Options:
-h, --help Print this help.
-a, --address <address>
Target echo server address. Default: 127.0.0.1:12345
-l, --length <length>
Test message length. Default: 512
-t, --duration <duration>
Test duration in seconds. Default: 60
-c, --number <number>
Test connection number. Default: 50（3）运行
cargo run --release -- --address "192.168.7.233:9999" --number 1000 --duration 60 --length 512将相关参数修改即可。
2.2、测试比较
io_uring结果：
Finished release [optimized] target(s) in 0.00s
Running `target/release/echo_bench --address '192.168.7.233:9999' --number 1000 --duration 60 --length 512`
Benchmarking: 192.168.7.235:9999
1000 clients, running 512 bytes, 60 sec.
Speed: 8836 request/sec, 8836 response/sec
Requests: 530176
Responses: 530176epoll测试前，需要先调整ulimit大小：ulimit -n 1024567。
epoll结果：
Finished release [optimized] target(s) in 0.01s
Running `target/release/echo_bench --address '192.168.7.233:8888' --number 1000 --duration 60 --length 512`
Benchmarking: 192.168.7.233:8888
1000 clients, running 512 bytes, 60 sec.
Speed: 7908 request/sec, 7908 response/sec
Requests: 474517
Responses: 474516注意，此测试结果仅仅是某次数据或访问的比较，为自己编写的代码提供一种测试方案，不作为说明io_uring与epoll/poll/select的性能高低。
2.3、小结
io_uring在性能上不比reactor高多少，io_uring不一定会完全替代epoll，未来是io_uring与epoll并存，只是网络IO事件的处理方案多了一个选择。
三、实现封装io_uring用户态文件读写接口
io_uring提供三个系统调用接口：io_uring_submit()、io_uring_enter()、io_uring_register()。不使用liburing情况下，需要自己实现用户层的接口。io_uring除了可以处理网络IO，也可以处理磁盘IO事件。
3.1、系统调用
系统调用函数是syscall。调用syscall函数时，会触发一个0x80的中断。每个系统调用都有系统调用号，存放在sys_call_table数组中。
流程：调用syscall，触发0x80中断，将系统调用号赋给eax寄存器，参数赋给ebx、ecx等寄存器，然后执行system_call。
函数原型：
#define _GNU_SOURCE /* See feature_test_macros(7) */
#include <unistd.h>
#include <sys/syscall.h> /* For SYS_xxx definitions */
long syscall(long number, ...);使用示例：
#define _GNU_SOURCE
#include <unistd.h>
#include <sys/syscall.h>
#include <sys/types.h>
#include <signal.h>
int main(int argc, char *argv[])
{
pid_t tid;
tid = syscall(SYS_gettid);
tid = syscall(SYS_tgkill, getpid(), tid, SIGHUP);
}3.2、内存映射mmap
应用层和内核直接数据交互，可以通过映射内存块的方式。
函数原型：
#include <sys/mman.h>
void *mmap(void *addr, size_t length, int prot, int flags,
int fd, off_t offset);
int munmap(void *addr, size_t length);使用示例：
#include <sys/mman.h>
#include <sys/stat.h>
#include <fcntl.h>
#include <stdio.h>
#include <stdlib.h>
#include <unistd.h>
#define handle_error(msg) \
do { perror(msg); exit(EXIT_FAILURE); } while (0)
int
main(int argc, char *argv[])
{
char *addr;
int fd;
struct stat sb;
off_t offset, pa_offset;
size_t length;
ssize_t s;
if (argc < 3 || argc > 4) {
fprintf(stderr, "%s file offset [length]\n", argv[0]);
exit(EXIT_FAILURE);
}
fd = open(argv[1], O_RDONLY);
if (fd == -1)
handle_error("open");
if (fstat(fd, &sb) == -1) /* To obtain file size */
handle_error("fstat");
offset = atoi(argv[2]);
pa_offset = offset & ~(sysconf(_SC_PAGE_SIZE) - 1);
/* offset for mmap() must be page aligned */
if (offset >= sb.st_size) {
fprintf(stderr, "offset is past end of file\n");
exit(EXIT_FAILURE);
}
if (argc == 4) {
length = atoi(argv[3]);
if (offset + length > sb.st_size)
length = sb.st_size - offset;
/* Can't display bytes past end of file */
} else { /* No length arg ==> display to end of file */
length = sb.st_size - offset;
}
addr = mmap(NULL, length + offset - pa_offset, PROT_READ,
MAP_PRIVATE, fd, pa_offset);
if (addr == MAP_FAILED)
handle_error("mmap");
s = write(STDOUT_FILENO, addr + offset - pa_offset, length);
if (s != length) {
if (s == -1)
handle_error("write");
fprintf(stderr, "partial write");
exit(EXIT_FAILURE);
}
exit(EXIT_SUCCESS);
}3.3、SQ_RING、CQ_RING、SQES关系
SQ_RING和CQ_RING的各项偏移指向SQES，真正存储数据的是SQES。
3.4、代码实现示例
#include <stdio.h>
#include <stdlib.h>
#include <sys/stat.h>
#include <sys/ioctl.h>
#include <sys/syscall.h>
#include <sys/mman.h>
#include <sys/uio.h>
#include <linux/fs.h>
#include <fcntl.h>
#include <unistd.h>
#include <string.h>
#include <linux/io_uring.h>
#define URING_QUEUE_DEPTH 1024
#define BLOCK_SZ 1024
// 定义结构体，将内核数据mmap到用户层
struct app_io_sq_ring {
unsigned *head;
unsigned *tail;
unsigned *ring_mask;
unsigned *ring_entries;
unsigned *flags;
unsigned *array;
};
struct app_io_cq_ring {
unsigned *head;
unsigned *tail;
unsigned *ring_mask;
unsigned *ring_entries;
struct io_uring_cqe *cqes;
};
// 返回
struct submitter {
int ring_fd;
struct app_io_sq_ring sq_ring;
struct app_io_cq_ring cq_ring;
struct io_uring_sqe *sqes;
};
// file info
struct file_info {
off_t file_sz;
struct iovec iovecs[];
};
int io_uring_setup(unsigned entries, struct io_uring_params *p)
{
// io_uring.c
// SYSCALL_DEFINE2(io_uring_setup,u32,entries,struct io_uring_params __user *,params)
// 内核代码 io_uring_setup
return (int)syscall(__NR_io_uring_setup, entries, p);
}
int io_uring_enter(int ring_fd, unsigned int to_submit,
unsigned int min_complete, unsigned int flags)
{
return (int)syscall(__NR_io_uring_enter, ring_fd, to_submit, min_complete, flags, NULL, 0);
}
int app_setup_uring(struct submitter *s)
{
struct io_uring_params p;
memset(&p, 0, sizeof(p));
// 创建 CQ和SQ 内存空间
s->ring_fd = io_uring_setup(URING_QUEUE_DEPTH, &p);
if (s->ring_fd < 0)
return -1;
int sring_sz = p.sq_off.array + p.sq_entries * sizeof(unsigned);
int cring_sz = p.cq_off.cqes + p.cq_entries * sizeof(struct io_uring_cqe);
// 单映射，判断cq和sq是否公用一块内存
if (p.features & IORING_FEAT_SINGLE_MMAP)
{
if (cring_sz > sring_sz)
sring_sz = cring_sz;
cring_sz = sring_sz;
}
// mmap sq空间
void *sq_ptr = mmap(0, sring_sz, PROT_READ | PROT_WRITE,
MAP_SHARED | MAP_POPULATE, s->ring_fd, IORING_OFF_SQ_RING);
if (sq_ptr == MAP_FAILED)
return -1;
// 单映射，判断cq和sq是否公用一块内存
void *cq_ptr = NULL;
if (p.features & IORING_FEAT_SINGLE_MMAP)
{
cq_ptr = sq_ptr;
}
else
{
// mmap cq空间
cq_ptr = mmap(0, sring_sz, PROT_READ | PROT_WRITE,
MAP_SHARED | MAP_POPULATE, s->ring_fd, IORING_OFF_CQ_RING);
if (cq_ptr == MAP_FAILED)
return -1;
}
struct app_io_sq_ring *sring = &s->sq_ring;
struct app_io_cq_ring *cring = &s->cq_ring;
// sq赋值
sring->head = sq_ptr + p.sq_off.head;
sring->tail = sq_ptr + p.sq_off.tail;
sring->ring_mask= sq_ptr + p.sq_off.ring_mask;
sring->ring_entries = sq_ptr + p.sq_off.ring_entries;
sring->flags = sq_ptr + p.sq_off.flags;
sring->array = sq_ptr + p.sq_off.array;
// 将sqes映射到用户空间
s->sqes = mmap(0, p.sq_entries * sizeof(struct io_uring_sqe),PROT_READ | PROT_WRITE,
MAP_SHARED | MAP_POPULATE, s->ring_fd, IORING_OFF_SQES);
if (s->sqes == MAP_FAILED)
return -1;
// cq赋值
cring->head = cq_ptr + p.cq_off.head;
cring->tail = cq_ptr + p.cq_off.tail;
cring->ring_mask = cq_ptr + p.cq_off.ring_mask;
cring->ring_entries = cq_ptr + p.cq_off.ring_entries;
cring->cqes = sq_ptr + p.cq_off.cqes;
return 0;
}
// get file size
off_t get_file_size(int fd)
{
struct stat st;
if (fstat(fd, &st) < 0)
return -1;
if (S_ISBLK(st.st_mode))
{
unsigned long long bytes;
if (ioctl(fd, BLKGETSIZE64, &bytes) != 0)
return -1;
return bytes;
}
else if (S_ISREG(st.st_mode))
return st.st_size;
return -1;
}
void output_to_console(char *buf,int len)
{
while (len--)
{
fputc(*buf++, stdout);
}
}
void read_from_cq(struct submitter *s)
{
struct file_info *fi;
struct app_io_cq_ring *cring = &s->cq_ring;
struct io_uring_cqe *cqe;
unsigned head = *cring->head;
while (1)
{
if (head == *cring->tail)
break;
cqe = &cring->cqes[head&*s->cq_ring.ring_mask];
fi = (struct file_info*)cqe->user_data;
if (cqe->res < 0)
fprintf(stderr, "error: %d\n", cqe->res);
int blocks = fi->file_sz / BLOCK_SZ;
if (fi->file_sz%BLOCK_SZ)
blocks++;
int i = 0;
while (++i < blocks)
{
output_to_console(fi->iovecs[i].iov_base, fi->iovecs[i].iov_len);
printf("################################ i : %d, blocks : %d\n",i, blocks);
}
head++;
printf("head : %d,tail : %d, blocks: %d\n", head, *cring->tail,blocks);
}
*cring->head = head;
printf("read form cq end!\n");
}
int submit_to_sq(char *file_path,struct submitter *s)
{
int filefd = open(file_path, O_RDONLY);
if (filefd < 0)
return -1;
off_t filesz = get_file_size(filefd);
if (filesz < 0) {
close(filefd);
return -1;
}
struct app_io_sq_ring *sring = &s->sq_ring;
off_t bytes_remaining = filesz;
int blocks = filesz / BLOCK_SZ;
if (filesz%BLOCK_SZ)
blocks++;
struct file_info *fi = malloc(sizeof(struct file_info) + sizeof(struct iovec)*blocks);
if (!fi)
{
close(filefd);
return -2;
}
fi->file_sz = filesz;
unsigned current_block=0;
while (bytes_remaining)
{
off_t bytes_to_read = bytes_remaining;
if (bytes_to_read > BLOCK_SZ)
bytes_to_read = BLOCK_SZ;
fi->iovecs[current_block].iov_len = bytes_to_read;
void *buf;
if (posix_memalign(&buf, BLOCK_SZ, BLOCK_SZ))
{
close(filefd);
return 1;
}
fi->iovecs[current_block].iov_base = buf;
current_block++;
bytes_remaining -= bytes_to_read;
}
unsigned next_tail = 0, tail = 0, index = 0;
next_tail = tail = *sring->tail;
next_tail++;
index = tail &*s->sq_ring.ring_mask;
struct io_uring_sqe *sqe = &s->sqes[index];
sqe->fd = filefd;
sqe->flags = 0;
sqe->opcode = IORING_OP_READV;
sqe->addr = (unsigned long)fi->iovecs;
sqe->len = blocks;
sqe->off = 0;
sqe->user_data = (unsigned long long)fi;
sring->array[index] = index;
tail = next_tail;
if (*sring->tail != tail)
*sring->tail = tail;
int ret = io_uring_enter(s->ring_fd, 1, 1, IORING_ENTER_GETEVENTS);
if (ret < 0)
{
close(filefd);
return -1;
}
close(filefd);
return 0;
}
int main(int argc, char *argv[])
{
struct submitter *s = malloc(sizeof(struct submitter));
if (s == NULL)
{
perror("malloc fail");
return -1;
}
memset(s, 0, sizeof(struct submitter));
if (app_setup_uring(s))
return 1;
int i = 1;
for (i = 1; i < argc; i++)
{
if (submit_to_sq(argv[i], s))
return 1;
read_from_cq(s);
}
return 0;
}总结
io_uring比epoll好的点是io_uring使用共享内存，不仅仅共享了IO事件，需要的变量也通过共享内存共享到用户空间，像SQ和CQ队列。io_uring不仅可以处理网络IO，也可以处理磁盘IO。
封装io_uring处理磁盘IO事件：
（1）io_uring_setup，创建SQ和CQ以及SQES，mmap io_uring_cqe / io_uring_sqe / SQES到用户空间。
（2）submit 文件IO到SQ。
（3）从CQ中读取文件IO。
关注公众号《Lion 莱恩呀》随时随地学习。
【版权声明】本文为华为云社区用户原创内容，未经允许不得转载，如需转载请自行联系原作者进行授权。如果您发现本社区中有涉嫌抄袭的内容，欢迎发送邮件进行举报，并提供相关证据，一经查实，本社区将立刻删除涉嫌侵权内容，举报邮箱：
cloudbbs@huaweicloud.com
- 点赞
- 收藏
- 关注作者
评论（0）