io_submit、io_setup和io_getevents示例

io_submit、io_setup和io_getevents和LINUX上的AIO系统调用。这有一个非常特别注意的地方——传递给io_setup的aio_context参数必须初始化为0，在它的man手册里其实有说明，但容易被忽视，我就犯了这个错误，man说明如下：

ctxp must not point to an AIO context that already exists, and must be initialized to 0 prior to the call

完整示例如下：// 包含必须头文件
#include <errno.h>
#include <stdio.h>
#include <fcntl.h>
#include <unistd.h>
#include <string.h>
#include <stdlib.h>
#include <sys/stat.h>
#include <sys/types.h>
#include <libaio.h>

int main()
{
      io_context_t ctx;
      unsigned nr_events = 10;
      memset(&ctx, 0, sizeof(ctx));  // It's necessary，这里一定要的
      int errcode = io_setup(nr_events, &ctx);
      if (errcode == 0)
            printf("io_setup success\n");
      else
            printf("io_setup error: :%d:%s\n", errcode, strerror(-errcode));

      // 如果不指定O_DIRECT，则io_submit操作和普通的read/write操作没有什么区别了，将来的LINUX可能
      // 可以支持不指定O_DIRECT标志
      int fd = open("./direct.txt", O_CREAT|O_DIRECT|O_WRONLY, S_IRWXU|S_IRWXG|S_IROTH);
      printf("open: %s\n", strerror(errno));

      char* buf;
      errcode = posix_memalign((void**)&buf, sysconf(_SC_PAGESIZE), sysconf(_SC_PAGESIZE));
      printf("posix_memalign: %s\n", strerror(errcode));

      strcpy(buf, "hello xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx");

      struct iocb *iocbpp = (struct iocb *)malloc(sizeof(struct iocb));
      memset(iocbpp, 0, sizeof(struct iocb));

      iocbpp[0].data          = buf;
      iocbpp[0].aio_lio_opcode = IO_CMD_PWRITE;
      iocbpp[0].aio_reqprio = 0;
      iocbpp[0].aio_fildes    = fd;

      iocbpp[0].u.c.buf = buf;
      iocbpp[0].u.c.nbytes = page_size;//strlen(buf); // 这个值必须按512字节对齐
      iocbpp[0].u.c.offset = 0; // 这个值必须按512字节对齐

      // 提交异步操作，异步写磁盘
      int n = io_submit(ctx, 1, &iocbpp);
      printf("==io_submit==: %d:%s\n", n, strerror(-n));

      struct io_event events[10];
      struct timespec timeout = {1, 100};
      // 检查写磁盘情况，类似于epoll_wait或select
      n = io_getevents(ctx, 1, 10, events, &timeout);
      printf("io_getevents: %d:%s\n", n, strerror(-n));

      close(fd);
      io_destroy(ctx);
      return 0;
}

测试环境：Linux 2.6.16，SUSE Linux Enterprise Server 10 (x86_64)

struct iocb {
   /* these are internal to the kernel/libc. */
   __u64 aio_data;    /* data to be returned in event\'s data */用来返回异步IO事件信息的空间，类似于epoll中的ptr。
   __u32 PADDED(aio_key, aio_reserved1); /* the kernel sets aio_key to the req # */
   /* common fields */
   __u16 aio_lio_opcode; /* see IOCB_CMD_ above */
   __s16 aio_reqprio;    // 请求的优先级
   __u32 aio_fildes;       //  文件描述符
   __u64 aio_buf;          // 用户态缓冲区
   __u64 aio_nbytes;    // 文件操作的字节数
   __s64 aio_offset;    // 文件操作的偏移量

   /* extra parameters */
   __u64 aio_reserved2;  /* TODO: use this for a (struct sigevent *) */
   __u64 aio_reserved3;
}; /* 64 bytes */

struct io_event {
   __u64          data;       /* the data field from the iocb */ // 类似于epoll_event中的ptr
   __u64          obj;          /* what iocb this event came from */ // 对应的用户态iocb结构体指针
   __s64          res;          /* result code for this event */ // 操作的结果，类似于read/write的返回值
   __s64          res2;       /* secondary result */
};

系统调用功能原型io_setup为当前进程初始化一个异步IO上下文int io_setup(unsigned nr_events,aio_context_t *ctxp);io_submit提交一个或者多个异步IO操作int io_submit(aio_context_t ctx_id,long nr, struct iocb **iocbpp);io_getevents获得未完成的异步IO操作的状态int io_getevents(aio_context_t ctx_id, long min_nr, long nr, struct io_event *events, struct timespec *timeout);io_cancel取消一个未完成的异步IO操作int io_cancel(aio_context_t ctx_id, struct iocb *iocb, struct io_event *result);io_destroy从当前进程删除一个异步IO上下文int io_destroy(aio_context_t ctx);