From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: from userp1040.oracle.com ([156.151.31.81]:44792 "EHLO userp1040.oracle.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1751288AbaEUDoO (ORCPT ); Tue, 20 May 2014 23:44:14 -0400 Message-ID: <537C21C1.9040809@oracle.com> Date: Wed, 21 May 2014 11:47:13 +0800 From: Anand Jain MIME-Version: 1.0 To: Qu Wenruo , linux-btrfs@vger.kernel.org Subject: Re: [RFC PATCH 1/2] btrfs: Add missing device check in dev_info/rm_dev ioctl References: <1399357993-9254-1-git-send-email-quwenruo@cn.fujitsu.com> <1399357993-9254-2-git-send-email-quwenruo@cn.fujitsu.com> In-Reply-To: <1399357993-9254-2-git-send-email-quwenruo@cn.fujitsu.com> Content-Type: text/plain; charset=ISO-8859-1; format=flowed Sender: linux-btrfs-owner@vger.kernel.org List-ID: Hi Qu, Either we need to determine missing disk when disk disappear or when we hit by -ENODEV during IO. Discovering the same at the time when "user" run 'btrfs fi show' or 'btrfs dev del', is just "not correct". You might use interface as proposed here. [PATCH RFC] btrfs: revamp /sys/fs/btrfs//devices further to this I will be sending another patch which will call the above sysfs interlace when we get the kobject notification when disk disappear. On 05/06/2014 02:33 PM, Qu Wenruo wrote: > Old btrfs can't find a missing btrfs device since there is no > mechanism for block layer to inform fs layer. > > But we can use a workaround that only check status(by using > request_queue->queue_flags) of every device in a btrfs > filesystem when calling dev_info/rm_dev ioctl, since other ioctls > do not really cares about missing device. > Cc: Anand Jain > Signed-off-by: Qu Wenruo > --- > fs/btrfs/ioctl.c | 1 + > fs/btrfs/volumes.c | 25 ++++++++++++++++++++++++- > fs/btrfs/volumes.h | 2 ++ > 3 files changed, 27 insertions(+), 1 deletion(-) > > diff --git a/fs/btrfs/ioctl.c b/fs/btrfs/ioctl.c > index 0401397..7680a40 100644 > --- a/fs/btrfs/ioctl.c > +++ b/fs/btrfs/ioctl.c > @@ -2606,6 +2606,7 @@ static long btrfs_ioctl_dev_info(struct btrfs_root *root, void __user *arg) > goto out; > } > > + btrfs_check_dev_missing(root, dev, 1); > di_args->devid = dev->devid; > di_args->bytes_used = dev->bytes_used; > di_args->total_bytes = dev->total_bytes; > diff --git a/fs/btrfs/volumes.c b/fs/btrfs/volumes.c > index d241130a..c7d7908 100644 > --- a/fs/btrfs/volumes.c > +++ b/fs/btrfs/volumes.c > @@ -1548,9 +1548,10 @@ int btrfs_rm_device(struct btrfs_root *root, char *device_path) > * is held. > */ > list_for_each_entry(tmp, devices, dev_list) { > + btrfs_check_dev_missing(root, tmp, 0); > if (tmp->in_fs_metadata && > !tmp->is_tgtdev_for_dev_replace && > - !tmp->bdev) { > + (!tmp->bdev || tmp->missing)) { > device = tmp; > break; > } > @@ -6300,3 +6301,25 @@ int btrfs_scratch_superblock(struct btrfs_device *device) > > return 0; > } > + > +/* If need_lock is set, uuid_mutex will be used */ > +int btrfs_check_dev_missing(struct btrfs_root *root, struct btrfs_device *dev, > + int need_lock) > +{ > + struct request_queue *q; > + > + if (unlikely(!dev || !dev->bdev || !dev->bdev->bd_queue)) > + return -ENOENT; > + q = dev->bdev->bd_queue; > + > + if (need_lock) > + mutex_lock(&uuid_mutex); > + if (test_bit(QUEUE_FLAG_DEAD, &q->queue_flags) || > + test_bit(QUEUE_FLAG_DYING, &q->queue_flags)) { > + dev->missing = 1; > + root->fs_info->fs_devices->missing_devices++; > + } > + if (need_lock) > + mutex_unlock(&uuid_mutex); > + return 0; > +} > diff --git a/fs/btrfs/volumes.h b/fs/btrfs/volumes.h > index 80754f9..47a44af 100644 > --- a/fs/btrfs/volumes.h > +++ b/fs/btrfs/volumes.h > @@ -356,6 +356,8 @@ unsigned long btrfs_full_stripe_len(struct btrfs_root *root, > int btrfs_finish_chunk_alloc(struct btrfs_trans_handle *trans, > struct btrfs_root *extent_root, > u64 chunk_offset, u64 chunk_size); > +int btrfs_check_dev_missing(struct btrfs_root *root, struct btrfs_device *dev, > + int need_lock); > static inline void btrfs_dev_stat_inc(struct btrfs_device *dev, > int index) > { >