On Tue, Dec 05, 2017 at 04:52:58PM +0800, Anand Jain wrote:
> Support for a new command is being added here:
> btrfs dev ignore <dev>
> Which shall undo the effects of the command
> btrfs dev scan <dev>
>
> This cli/ioctl is needed as there is no way to continue to mount in
> degraded mode if the device is already scanned, which is required to
> recover from the split brain raid conditions.
>
> This patch proposes to use ioctl #5 as it was empty.
> IOW(BTRFS_IOCTL_MAGIC, 5, ..)
> If #5 is reserved for some other purpose, I think I should change this.
I think 5 is free for use.
> Signed-off-by: Anand Jain <anand.j...@oracle.com>
> ---
> v2: Use -EBUSY instead of -ENOENT
> Since now delete_device_from_list() holds device_list_mutex
> so dont hold device_list_mutex in its parent. Reword and indent
> pr_err/info.
> v3: Send to correct ML
> v4: no change.
>
> fs/btrfs/super.c | 4 +++
> fs/btrfs/volumes.c | 64
> ++++++++++++++++++++++++++++++++++++++++++++++
> fs/btrfs/volumes.h | 2 ++
> include/uapi/linux/btrfs.h | 2 ++
> 4 files changed, 72 insertions(+)
>
> diff --git a/fs/btrfs/super.c b/fs/btrfs/super.c
> index f443517fa2f8..fcc4a6ef4795 100644
> --- a/fs/btrfs/super.c
> +++ b/fs/btrfs/super.c
> @@ -2212,6 +2212,10 @@ static long btrfs_control_ioctl(struct file *file,
> unsigned int cmd,
> ret = btrfs_scan_one_device(vol->name, FMODE_READ,
> &btrfs_fs_type, &fs_devices);
> break;
> + case BTRFS_IOC_IGNORE_DEV:
> + ret = btrfs_ignore_one_device(vol->name, FMODE_READ,
> + &btrfs_fs_type, &fs_devices);
> + break;
> case BTRFS_IOC_DEVICES_READY:
> ret = btrfs_scan_one_device(vol->name, FMODE_READ,
> &btrfs_fs_type, &fs_devices);
> diff --git a/fs/btrfs/volumes.c b/fs/btrfs/volumes.c
> index 5deda80316f0..2bae2ccd262d 100644
> --- a/fs/btrfs/volumes.c
> +++ b/fs/btrfs/volumes.c
> @@ -1205,6 +1205,70 @@ static int btrfs_read_disk_super(struct block_device
> *bdev, u64 bytenr,
> return 0;
> }
>
> +static int device_list_remove(struct btrfs_super_block *disk_super, u64
> devid)
> +{
> + int ret = 0;
> + struct btrfs_fs_devices *fs_devices;
> + struct btrfs_device *device;
> +
> + fs_devices = find_fsid(disk_super->fsid);
Don't we need uuid mutext to call find_fsid? All other users do that.
> + if (!fs_devices)
> + return -ENOENT;
> +
> + if (fs_devices->opened)
> + return -EBUSY;
> +
> + mutex_lock(&uuid_mutex);
> +
> + device = find_device(fs_devices, devid, disk_super->dev_item.uuid);
> + if (device)
> + delete_device_from_list(device);
> +
> + mutex_unlock(&uuid_mutex);
> +
> + return ret;
> +}
> +
> +int btrfs_ignore_one_device(const char *path, fmode_t flags, void *holder,
> + struct btrfs_fs_devices **fs_devices_ret)
> +{
> + struct btrfs_super_block *disk_super;
> + struct block_device *bdev;
> + struct page *page;
> + int ret = -EINVAL;
Please move EINVAL to the point where this happens (ie. after the
btrfs_read_disk_super call). This is the common pattern and makes
reading the code smooth.
> + u64 devid;
> + u64 bytenr;
> +
> + bytenr = btrfs_sb_offset(0);
> + flags |= FMODE_EXCL;
> +
> + bdev = blkdev_get_by_path(path, flags, holder);
> + if (IS_ERR(bdev)) {
> + ret = PTR_ERR(bdev);
> + goto error;
> + }
> +
> + if (btrfs_read_disk_super(bdev, bytenr, &page, &disk_super))
> + goto error_bdev_put;
> +
> + devid = btrfs_stack_device_id(&disk_super->dev_item);
> +
> + ret = device_list_remove(disk_super, devid);
> + if (ret)
> + pr_err("BTRFS: %pU device %s devid %llu failed to ignore: %d\n",
> + disk_super->fsid, path, devid, ret);
So we can't easily use btrfs_printk here due to lack of fs_info that
would appear as "<unknown>" in place of the device. Ok.
> + else
> + pr_info("BTRFS: %pU device %s devid %llu ignored\n",
> + disk_super->fsid, path, devid);
> +
> + btrfs_release_disk_super(page);
> +
> +error_bdev_put:
> + blkdev_put(bdev, flags);
> +error:
> + return ret;
> +}
> +
> /*
> * Look for a btrfs signature on a device. This may be called out of the
> mount path
> * and we are not allowed to call set_blocksize during the scan. The
> superblock
> diff --git a/fs/btrfs/volumes.h b/fs/btrfs/volumes.h
> index 7acfd61611aa..08d3425bc880 100644
> --- a/fs/btrfs/volumes.h
> +++ b/fs/btrfs/volumes.h
> @@ -423,6 +423,8 @@ int btrfs_open_devices(struct btrfs_fs_devices
> *fs_devices,
> fmode_t flags, void *holder);
> int btrfs_scan_one_device(const char *path, fmode_t flags, void *holder,
> struct btrfs_fs_devices **fs_devices_ret);
> +int btrfs_ignore_one_device(const char *path, fmode_t flags, void *holder,
> + struct btrfs_fs_devices **fs_devices_ret);
> int btrfs_close_devices(struct btrfs_fs_devices *fs_devices);
> void btrfs_close_extra_devices(struct btrfs_fs_devices *fs_devices, int
> step);
> void btrfs_assign_next_active_device(struct btrfs_fs_info *fs_info,
> diff --git a/include/uapi/linux/btrfs.h b/include/uapi/linux/btrfs.h
> index ce615b75e855..bb13bad41b30 100644
> --- a/include/uapi/linux/btrfs.h
> +++ b/include/uapi/linux/btrfs.h
> @@ -744,6 +744,8 @@ enum btrfs_err_code {
> struct btrfs_ioctl_vol_args)
> #define BTRFS_IOC_SCAN_DEV _IOW(BTRFS_IOCTL_MAGIC, 4, \
> struct btrfs_ioctl_vol_args)
> +#define BTRFS_IOC_IGNORE_DEV _IOW(BTRFS_IOCTL_MAGIC, 5, \
> + struct btrfs_ioctl_vol_args)
While reading the patches, I realized we may want to extend the ioctl to
unregister/forge all devices that are not currently mounted. For that
purpose using the btrfs_ioctl_vol_args_v2 would be suitable as it has
more struct members.
Another extension is to unregister only stale devices (when there's no
device node under /dev), eg. after the device is unuplugged and readded
by another name.
--
To unsubscribe from this list: send the line "unsubscribe linux-btrfs" in
the body of a message to majord...@vger.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
