Re: [PATCH 4/4] loop: Add LOOP_SET_FD_AND_STATUS ioctl.

[Martijn Coenen]

On Wed, Apr 22, 2020 at 8:19 AM Christoph Hellwig <hch@xxxxxx> wrote:
>
> On Mon, Apr 20, 2020 at 10:04:09AM +0200, Martijn Coenen wrote:
> > This allows userspace to completely setup a loop device with a single
> > ioctl, removing the in-between state where the device can be partially
> > configured - eg the loop device has a backing file associated with it,
> > but is reading from the wrong offset.
> >
> > Besides removing the intermediate state, another big benefit of this
> > ioctl is that LOOP_SET_STATUS can be slow; the main reason for this
> > slowness is that LOOP_SET_STATUS(64) calls blk_mq_freeze_queue() to
> > freeze the associated queue; this requires waiting for RCU
> > synchronization, which I've measured can take about 15-20ms on this
> > device on average.
> >
> > Here's setting up ~70 regular loop devices with an offset on an x86
> > Android device, using LOOP_SET_FD and LOOP_SET_STATUS:
> >
> > vsoc_x86:/system/apex # time for i in `seq 30 100`;
> > do losetup -r -o 4096 /dev/block/loop$i com.android.adbd.apex; done
> >     0m03.40s real     0m00.02s user     0m00.03s system
> >
> > Here's configuring ~70 devices in the same way, but using a modified
> > losetup that uses the new LOOP_SET_FD_AND_STATUS ioctl:
> >
> > vsoc_x86:/system/apex # time for i in `seq 30 100`;
> > do losetup -r -o 4096 /dev/block/loop$i com.android.adbd.apex; done
> >     0m01.94s real     0m00.01s user     0m00.01s system
> >
> > Signed-off-by: Martijn Coenen <maco@xxxxxxxxxxx>
> > ---
> >  drivers/block/loop.c      | 47 ++++++++++++++++++++++++++++++---------
> >  include/uapi/linux/loop.h |  6 +++++
> >  2 files changed, 42 insertions(+), 11 deletions(-)
> >
> > diff --git a/drivers/block/loop.c b/drivers/block/loop.c
> > index 6e656390b285..e1dbd70d6d6e 100644
> > --- a/drivers/block/loop.c
> > +++ b/drivers/block/loop.c
> > @@ -1065,8 +1065,9 @@ loop_set_from_status(struct loop_device *lo, const struct loop_info64 *info)
> >       return 0;
> >  }
> >
> > -static int loop_set_fd(struct loop_device *lo, fmode_t mode,
> > -                    struct block_device *bdev, unsigned int arg)
> > +static int loop_set_fd_and_status(struct loop_device *lo, fmode_t mode,
> > +                               struct block_device *bdev, unsigned int fd,
> > +                               const struct loop_info64 *info)
> >  {
> >       struct file     *file;
> >       struct inode    *inode;
> > @@ -1081,7 +1082,7 @@ static int loop_set_fd(struct loop_device *lo, fmode_t mode,
> >       __module_get(THIS_MODULE);
> >
> >       error = -EBADF;
> > -     file = fget(arg);
> > +     file = fget(fd);
> >       if (!file)
> >               goto out;
> >
> > @@ -1090,7 +1091,7 @@ static int loop_set_fd(struct loop_device *lo, fmode_t mode,
> >        * here to avoid changing device under exclusive owner.
> >        */
> >       if (!(mode & FMODE_EXCL)) {
> > -             claimed_bdev = bd_start_claiming(bdev, loop_set_fd);
> > +             claimed_bdev = bd_start_claiming(bdev, loop_set_fd_and_status);
> >               if (IS_ERR(claimed_bdev)) {
> >                       error = PTR_ERR(claimed_bdev);
> >                       goto out_putf;
> > @@ -1117,9 +1118,24 @@ static int loop_set_fd(struct loop_device *lo, fmode_t mode,
> >               lo_flags |= LO_FLAGS_READ_ONLY;
> >
> >       error = -EFBIG;
> > -     size = get_loop_size(lo, file);
> > +     if (info)
> > +             size = get_size(info->lo_offset, info->lo_sizelimit,
> > +                             file);
> > +     else
> > +             size = get_loop_size(lo, file);
> >       if ((loff_t)(sector_t)size != size)
> >               goto out_unlock;
> > +
> > +     if (info) {
> > +             error = loop_set_from_status(lo, info);
> > +             if (error)
> > +                     goto out_unlock;
> > +     } else {
> > +             lo->transfer = NULL;
> > +             lo->ioctl = NULL;
> > +             lo->lo_sizelimit = 0;
> > +             lo->lo_offset = 0;
> > +     }
>
> Just curious:  Can't we just pass in an on-stack info for the legacy
> case and avoid all these conditionals?

Yeah, that is actually much nicer. I will rework it to that.

> Can you throw in another prep patch that adds a:
>
>         void __user *argp = (void __user *)arg;
>
> line at the top of lo_compat_ioctl, and switches the LOOP_SET_STATUS
> and LOOP_GET_STATUS case to it?

Did you mean in regular lo_ioctl()? eg something like this:

@@ -1671,6 +1671,7 @@ static int lo_ioctl(struct block_device *bdev,
fmode_t mode,
        unsigned int cmd, unsigned long arg)
 {
        struct loop_device *lo = bdev->bd_disk->private_data;
+       void __user *argp = (void __user *) arg;
        int err;

        switch (cmd) {
@@ -1694,21 +1695,19 @@ static int lo_ioctl(struct block_device *bdev,
fmode_t mode,
        case LOOP_SET_STATUS:
                err = -EPERM;
                if ((mode & FMODE_WRITE) || capable(CAP_SYS_ADMIN)) {
-                       err = loop_set_status_old(lo,
-                                       (struct loop_info __user *)arg);
+                       err = loop_set_status_old(lo, argp);
                }
                break;
        case LOOP_GET_STATUS:
-               return loop_get_status_old(lo, (struct loop_info __user *) arg);
+               return loop_get_status_old(lo, argp);
        case LOOP_SET_STATUS64:
                err = -EPERM;
                if ((mode & FMODE_WRITE) || capable(CAP_SYS_ADMIN)) {
-                       err = loop_set_status64(lo,
-                                       (struct loop_info64 __user *) arg);
+                       err = loop_set_status64(lo, argp);
                }
                break;
        case LOOP_GET_STATUS64:
-               return loop_get_status64(lo, (struct loop_info64 __user *) arg);
+               return loop_get_status64(lo, argp);


> > +struct loop_fd_and_status {
> > +     struct loop_info64      info;
> > +     __u32                   fd;
>
> This should grow a
>
>         __u32   __pad;
>
> to avoid different struct sizes on x86-32 vs x86-64.

will do, thanks!