"Michael Kerrisk (man-pages)" <mtk.manpages@xxxxxxxxx> writes:
> Jeff,
>
> On Thu, Mar 28, 2013 at 6:28 PM, Jeff Moyer <jmoyer@xxxxxxxxxx> wrote:
>> Hi,
>>
>> The alignment text for O_DIRECT is slightly misleading:
>>
>> Under Linux 2.4, transfer sizes, and the alignment of the user buffer
>> and the file offset must all be multiples of the logical block size of
>> the file system. Under Linux 2.6, alignment to 512-byte boundaries
>> suffices.
>>
>> The last sentence is incorrect. You cannot perform O_DIRECT I/O in
>> sizes smaller than the underlying logical block size (for block devices
>> and block-based file systems; nfs is a different beast). I've attached
>> my proposed change. Comments and word-smithing are most welcome.
>
> Could you provide some detail on how you determined, verified, or
> tested this detail? I'm not saying you're wrong, but that text has
> been there a long time (since before I was maintainer), in the days
> when there was no VCS and changelog. It's even possible that I sent
> the patch to Andries that added this piece, since I have an idea that
> I tested this detail on Linux 2.6 at some point in the distant past.
> Indeed, I just now did a (very) quick test using an old program that
> suggest that you can do O_DIRECT I/O with block sizes smaller than the
> LBS.
Details on the test?
Devices do not accept I/O that is smaller than a single addressable
logical block. Thus, you cannot perform direct I/O in sizes smaller
than that (there is no read-modify-write in O_DIRECT). The code that
enforces the limitation is in fs/direct-io.c:
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)
{
...
unsigned i_blkbits = ACCESS_ONCE(inode->i_blkbits);
unsigned blkbits = i_blkbits;
unsigned blocksize_mask = (1 << blkbits) - 1;
ssize_t retval = -EINVAL;
...
if (offset & blocksize_mask) { // offset is not a multiple of the file system block size
if (bdev)
blkbits = blksize_bits(bdev_logical_block_size(bdev));
blocksize_mask = (1 << blkbits) - 1;
if (offset & blocksize_mask)
goto out; // offset is not a multiple of the logical block size
}
// and the check is done for the address and length of each segment:
/* 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;
}
}
...
out:
return retval;
}
Cheers,
Jeff
--
To unsubscribe from this list: send the line "unsubscribe linux-man" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
