Just to throw a can of turquoise paint into the bike-shed works,
this could be handy with a list of devids to say *where* to create the
new block group.
This allows for some truly horrible things to happen if abused, but
could also allow for some kind of poor-mans directed-balance: Create a
new block group on the devices you want, balance away one block group
on device(s) you don't want -- data should end up going to the new
empty block group in preference to another new one being automatically
allocated.
(Alternatively, ignore this suggestion, and I'll just wait for a
proper "move this BG to these devices" ioctl...)
Hugo.
On Mon, Aug 05, 2019 at 08:24:23PM +0800, Qu Wenruo wrote:
>
>
> On 2019/8/3 上午12:10, Josef Bacik wrote:
> > In testing block group removal it's sometimes handy to be able to create
> > block groups on demand. Add an ioctl to allow us to force allocation
> > from userspace.
>
> Not sure if we should add another ioctl just for debug purpose.
>
> Although I see the usefulness in such debug feature, can we move it to
> something like sysfs so we can hide it more easily?
>
> >
> > Signed-off-by: Josef Bacik <josef@xxxxxxxxxxxxxx>
> > ---
> > fs/btrfs/ioctl.c | 30 ++++++++++++++++++++++++++++++
> > include/uapi/linux/btrfs.h | 1 +
> > 2 files changed, 31 insertions(+)
> >
> > diff --git a/fs/btrfs/ioctl.c b/fs/btrfs/ioctl.c
> > index d0743ec1231d..f100def53c29 100644
> > --- a/fs/btrfs/ioctl.c
> > +++ b/fs/btrfs/ioctl.c
> > @@ -5553,6 +5553,34 @@ static int _btrfs_ioctl_send(struct file *file, void __user *argp, bool compat)
> > return ret;
> > }
> >
> > +static long btrfs_ioctl_alloc_chunk(struct file *file, void __user *arg)
> > +{
> > + struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
> > + struct btrfs_trans_handle *trans;
> > + u64 flags;
> > + int ret;
> > +
> > + if (!capable(CAP_SYS_ADMIN))
> > + return -EPERM;
> > +
> > + if (copy_from_user(&flags, arg, sizeof(flags)))
> > + return -EFAULT;
> > +
> > + /* We can only specify one type at a time. */
> > + if (flags != BTRFS_BLOCK_GROUP_DATA &&
> > + flags != BTRFS_BLOCK_GROUP_METADATA &&
> > + flags != BTRFS_BLOCK_GROUP_SYSTEM)
> > + return -EINVAL;
>
> It looks like MIXED bg get less and less love.
>
> > +
> > + trans = btrfs_start_transaction(root, 0);
> > + if (IS_ERR(trans))
> > + return PTR_ERR(trans);
> > +
> > + ret = btrfs_chunk_alloc(trans, flags, CHUNK_ALLOC_FORCE);
>
> And the flags lacks the profile bits, thus default to SINGLE.
> Is it designed or you'd better use btrfs_force_chunk_alloc()?
>
> Thanks,
> Qu
>
> > + btrfs_end_transaction(trans);
> > + return ret < 0 ? ret : 0;
> > +}
> > +
> > long btrfs_ioctl(struct file *file, unsigned int
> > cmd, unsigned long arg)
> > {
> > @@ -5699,6 +5727,8 @@ long btrfs_ioctl(struct file *file, unsigned int
> > return btrfs_ioctl_get_subvol_rootref(file, argp);
> > case BTRFS_IOC_INO_LOOKUP_USER:
> > return btrfs_ioctl_ino_lookup_user(file, argp);
> > + case BTRFS_IOC_ALLOC_CHUNK:
> > + return btrfs_ioctl_alloc_chunk(file, argp);
> > }
> >
> > return -ENOTTY;
> > diff --git a/include/uapi/linux/btrfs.h b/include/uapi/linux/btrfs.h
> > index c195896d478f..3a6474c34ad0 100644
> > --- a/include/uapi/linux/btrfs.h
> > +++ b/include/uapi/linux/btrfs.h
> > @@ -943,5 +943,6 @@ enum btrfs_err_code {
> > struct btrfs_ioctl_get_subvol_rootref_args)
> > #define BTRFS_IOC_INO_LOOKUP_USER _IOWR(BTRFS_IOCTL_MAGIC, 62, \
> > struct btrfs_ioctl_ino_lookup_user_args)
> > +#define BTRFS_IOC_ALLOC_CHUNK _IOR(BTRFS_IOCTL_MAGIC, 63, __u64)
> >
> > #endif /* _UAPI_LINUX_BTRFS_H */
> >
>
--
Hugo Mills | We don't just borrow words; on occasion, English has
hugo@... carfax.org.uk | pursued other languages down alleyways to beat them
http://carfax.org.uk/ | unconscious and rifle their pockets for new
PGP: E2AB1DE4 | vocabulary. James D. Nicoll
Attachment:
signature.asc
Description: PGP signature
