On Fri, Aug 11, 2023 at 01:04:33PM +0200, Jan Kara wrote:
> @@ -478,7 +478,7 @@ blk_mode_t file_to_blk_mode(struct file *file)
> mode |= BLK_OPEN_READ;
> if (file->f_mode & FMODE_WRITE)
> mode |= BLK_OPEN_WRITE;
> - if (file->private_data)
> + if (file->f_flags & O_EXCL)
> mode |= BLK_OPEN_EXCL;
> if (file->f_flags & O_NDELAY)
> mode |= BLK_OPEN_NDELAY;
> index 3be11941fb2d..47f216d8697f 100644
> --- a/block/ioctl.c
> +++ b/block/ioctl.c
> @@ -575,7 +575,7 @@ long blkdev_ioctl(struct file *file, unsigned cmd, unsigned long arg)
> {
> struct block_device *bdev = I_BDEV(file->f_mapping->host);
> void __user *argp = (void __user *)arg;
> - blk_mode_t mode = file_to_blk_mode(file);
> + blk_mode_t mode = ((struct bdev_handle *)file->private_data)->mode;
Take a look at sd_ioctl() and note that fcntl(2) can be used to set/clear O_NDELAY.
The current variant works since we recalculate mode every time; this one will end up
stuck with whatever we had at open time. Note that Christoph's series could do this
in blkdev_ioctl()
- /*
- * O_NDELAY can be altered using fcntl(.., F_SETFL, ..), so we have
- * to updated it before every ioctl.
- */
- if (file->f_flags & O_NDELAY)
- mode |= FMODE_NDELAY;
- else
- mode &= ~FMODE_NDELAY;
precisely because his file_to_blk_mode() picks O_NDELAY from flags when blkdev_ioctl()
calls it.
The same goes for compat counterpart of that thing. Both need to deal with that
scenario - you need something that would pick the O_NDELAY updates from ->f_flags.
Al, trying to catch up on the awful pile of mail that has accumulated over the 3 months
of being net.dead...
