DirectIO的对齐问题

最近在代码中使用了Linux AIO接口,其需要通过O_DIRECT方式打开文件,同时在IO时要求块大小对齐。

对于块大小对齐,找到了一个解释:Why does O_DIRECT require I/O to be 512-byte aligned?
O_DIRECT应该是要求buffer地址,传送数据大小,文件offset都是512-byte对齐,即ioctl BLKSSZGET 获取的大小,和mkfs时的block大小无关。

实际测试的结果也符合这个解释:

[root@localhost ~]# mkfs.ext4 -b 4096 /dev/sdd2
[root@localhost ~]# umpe2fs /dev/sdd2 | grep -i 'block size'
dumpe2fs 1.42.9 (28-Dec-2013)
Block size:               4096

[root@localhost 4096] ./readwrite.out 
1. test count.
write 4096 bytes, ret:4096.
write 2048 bytes, ret:2048.
write 1024 bytes, ret:1024.
write 512 bytes, ret:512.
write 256 bytes failed. ret:-1, Invalid argument

2. test offset.
write offset 4096, ret:512.
write offset 2048, ret:512.
write offset 1024, ret:512.
write offset 512, ret:512.
write offset 256 failed. ret:-1, Invalid argument

看内核代码是在do_blockdev_direct_IO函数中判断对齐并返回-EINVAL

static inline ssize_t
do_blockdev_direct_IO(int rw, struct kiocb *iocb, struct inode *inode,
    struct block_device *bdev, const struct iovec *iov, loff_t offset, 
    unsigned long nr_segs, get_block_t get_block, dio_iodone_t end_io,
    dio_submit_t submit_io, int flags)
{
    ssize_t retval = -EINVAL;

    /* 省略掉无关代码*/
    
    if (rw & WRITE)
        rw = WRITE_ODIRECT;

    /*
     * Avoid references to bdev if not absolutely needed to give
     * the early prefetch in the caller enough time.
     */

    if (offset & blocksize_mask) {
        if (bdev)
            blkbits = blksize_bits(bdev_logical_block_size(bdev));
        blocksize_mask = (1 << blkbits) - 1;
        if (offset & blocksize_mask)
            goto out;
    }
    
    /* Check the memory alignment.  Blocks cannot straddle pages */
    for (seg = 0; seg < nr_segs; seg++) {
        addr = (unsigned long)iov[seg].iov_base;
        size = iov[seg].iov_len;
        end += size;
        if (unlikely((addr & blocksize_mask) ||
                 (size & blocksize_mask))) {
            if (bdev)
                blkbits = blksize_bits(
                     bdev_logical_block_size(bdev));
            blocksize_mask = (1 << blkbits) - 1;
            if ((addr & blocksize_mask) || (size & blocksize_mask))
                goto out;
        }
    }   
    

通过kprobe可以确认是在这个函数中

cd /sys/kernel/debug/tracing
echo 'r:myretprobe do_blockdev_direct_IO $retval' > kprobe_events
echo 1 > events/kprobes/myretprobe/enable

➜  tracing more trace
# tracer: nop
#
# entries-in-buffer/entries-written: 9/9   #P:2
#
#                              _-----=> irqs-off
#                             / _----=> need-resched
#                            | / _---=> hardirq/softirq
#                            || / _--=> preempt-depth
#                            ||| /     delay
#           TASK-PID   CPU#  ||||    TIMESTAMP  FUNCTION
#              | |       |   ||||       |         |
           <...>-67363 [000] d..1 94189.878965: myretprobe: (__blockdev_direct_IO+0x55/0x60 <- do_blockdev_direct_IO) arg1=1000
           <...>-67363 [000] d..1 94189.880246: myretprobe: (__blockdev_direct_IO+0x55/0x60 <- do_blockdev_direct_IO) arg1=800
           <...>-67363 [000] d..1 94189.881208: myretprobe: (__blockdev_direct_IO+0x55/0x60 <- do_blockdev_direct_IO) arg1=400
           <...>-67363 [000] d..1 94189.882268: myretprobe: (__blockdev_direct_IO+0x55/0x60 <- do_blockdev_direct_IO) arg1=200
           <...>-67363 [000] d..1 94189.883792: myretprobe: (__blockdev_direct_IO+0x55/0x60 <- do_blockdev_direct_IO) arg1=200
           <...>-67363 [000] d..1 94189.884805: myretprobe: (__blockdev_direct_IO+0x55/0x60 <- do_blockdev_direct_IO) arg1=200
           <...>-67363 [000] d..1 94189.885609: myretprobe: (__blockdev_direct_IO+0x55/0x60 <- do_blockdev_direct_IO) arg1=200
           <...>-67363 [000] d..1 94189.887123: myretprobe: (__blockdev_direct_IO+0x55/0x60 <- do_blockdev_direct_IO) arg1=200
           <...>-67363 [000] d..1 94189.887222: myretprobe: (__blockdev_direct_IO+0x55/0x60 <- do_blockdev_direct_IO) arg1=ffffffffffffffea

完整测试代码:

#define _GNU_SOURCE
#include 
#include 
#include 
#include 
#include 
#include 
#include 

void test1(int fd) 
{
    char buf[8192] __attribute__((aligned(8192)));
    size_t count = 4096;
    ssize_t ret;

    while (count > 0) {
        ret = write(fd, buf, count);
        if (ret < 0) {
            printf("write %d bytes failed. ret:%d, %s\n",
                    count, ret, strerror(errno));
            break;
        }

        count /= 2;
    }

    return;
}


int test2(int fd) 
{
    char buf[8192] __attribute__((aligned(8192)));
    size_t count = 4096;
    ssize_t ret;

    while (count > 0) {
        ret = write(fd, buf + count, 512);
        if (ret < 0) {
            printf("write offset %d failed. ret:%d, %s\n",
                    count, ret, strerror(errno));
            break;
        }

        count /= 2;
    }

    return;
}


int main()
{
    int fd = open("testfile", O_RDWR | O_CREAT | O_DIRECT, S_IRWXU | S_IRWXG | S_IRWXO);
    if (fd < 0) {
        printf("open file failed.\n");
        return -1;
    }

    printf("1. test count.\n");
    test1(fd);

    printf("2. test offset.\n");
    test2(fd);

    close(fd);

    return 0;
}

通过ioctl BLKSSZGET获取设备的block大小

#include 
#include 
#include 
#include 
#include 
#include 
#include 

int main(int argc, char* argv[])
{
    int fd, rc;
    size_t bs;

    if (argc != 2) {
        printf("ioctl /dev/sad.\n");
        return -1;
    }

    fd = open(argv[1], O_RDONLY);
    if (fd < 0) {
        printf("open %s failed! ret %d, %s\n",
                argv[1], fd, strerror(errno));
        return -1;
    }
    
    rc = ioctl(fd, BLKSSZGET, &bs);
    if (rc < 0) {
        printf("open %s failed! ret %d, %s\n",
                argv[1], rc, strerror(errno));
    } else {
        printf("dev %s bs is %d\n", argv[1], bs);
    }

    close(fd);

    return 0;
}

你可能感兴趣的:(DirectIO的对齐问题)