mounting arbitrary directories

mounting arbitrary directories

[C Anthony Risinger]

i have read just recently and in the past that btrfs supports COW for
_any_ file/directory (this is reflinks, yes?), and today i
accidentally noticed that i can mount a directory as well (if it's in
the top level volume at least).

eg. if i have a "regular" directory (not a subvolume) in the top-level:

/__boot

i can mount it with:

mount -o subvol=__boot /dev/sda /mnt

is this supposed to be supported via the "subvol" mount option?
though it sounds like i've answered my own question, i ask because
playing around with this caused my kernel to oops for the first time
in a long, long time (on .35 but just about to upgrade to .36 in about
10 min).

i am working on an update to my initramfs hook that will utilize
extlinux, and this property to provide seamless kernel level system
rollbacks, and i want to make sure it's safe to do this.

also, is there a way to target an arbitrary directory?  does "subvol"
support paths yet, or maybe via "subvolid" somehow?  essentially i
just want to mount a directory inside a snapshot at /boot, so when
users upgrade there kernels, the images are visible to extlinux (which
cannot yet peek inside or use subvolumes, so it has to be a regular
directory in the top-level volume)

any insight is much appreciated.

C Anthony

[RFC-PATCH] Re: mounting arbitrary directories

[Calvin Walton]

On Sat, 2010-11-27 at 21:19 -0600, C Anthony Risinger wrote:
> i have read just recently and in the past that btrfs supports COW for
> _any_ file/directory (this is reflinks, yes?), and today i
> accidentally noticed that i can mount a directory as well (if it's in
> the top level volume at least).
> 
> eg. if i have a "regular" directory (not a subvolume) in the top-level:
> 
> /__boot
> 
> i can mount it with:
> 
> mount -o subvol=__boot /dev/sda /mnt

The 'subvol' option actually works using the same mechanism as a bind
mount. The fact that it works using a directory is purely a coincidence
- I would not expect it to be officially supported, and you shouldn't
rely on it.

> i am working on an update to my initramfs hook that will utilize
> extlinux, and this property to provide seamless kernel level system
> rollbacks, and i want to make sure it's safe to do this.

To handle system rollbacks, you really should be using subvolumes and
snapshots, not regular directories.

> also, is there a way to target an arbitrary directory?  does "subvol"
> support paths yet, or maybe via "subvolid" somehow?  essentially i

I don't think that it would be very hard to make subvol= support a path
instead of only one level deep. Actually, I think I could make a patch
for that myself... I've included it here. Mildly tested, but I'm not
really a kernel programmer and might have missed something -
particularly with regards to the locking.

> just want to mount a directory inside a snapshot at /boot, so when
> users upgrade there kernels, the images are visible to extlinux (which
> cannot yet peek inside or use subvolumes, so it has to be a regular
> directory in the top-level volume)

Ah, this is the first I've heard that extlinux doesn't support reading
files in subvolumes. That's an unfortunate limitation :/

------------------------8<----------8<---------------------

From: Calvin Walton <calvin.walton@gmail.com>
Date: Sat, 27 Nov 2010 23:17:55 -0500
Subject: [PATCH] Btrfs: Allow mounting sub-sub(-sub...)-volumes using subvol=a/b/c

Currently you can only mount subvolumes which are direct children of the
'default' root. I.e. with

	ID 258 top level 5 path a
	ID 259 top level 5 path a/b

you could mount with "-o subvol=a" but not "-o subvol=a/b"

This patch fixes that by recursively following the path to the subvolume.

Signed-off-by: Calvin Walton <calvin.walton@gmail.com>
---
 fs/btrfs/super.c |   41 +++++++++++++++++++++++------------------
 1 files changed, 23 insertions(+), 18 deletions(-)

diff --git a/fs/btrfs/super.c b/fs/btrfs/super.c
index 8299a25..5e78c86 100644
--- a/fs/btrfs/super.c
+++ b/fs/btrfs/super.c
@@ -646,26 +646,31 @@ static struct dentry *btrfs_mount(struct file_system_type *fs_type, int flags,
 	/* if they gave us a subvolume name bind mount into that */
 	if (strcmp(subvol_name, ".")) {
 		struct dentry *new_root;
-		mutex_lock(&root->d_inode->i_mutex);
-		new_root = lookup_one_len(subvol_name, root,
-				      strlen(subvol_name));
-		mutex_unlock(&root->d_inode->i_mutex);
-
-		if (IS_ERR(new_root)) {
-			deactivate_locked_super(s);
-			error = PTR_ERR(new_root);
-			dput(root);
-			goto error_free_subvol_name;
-		}
-		if (!new_root->d_inode) {
+		char *subvol_name_next = subvol_name;
+		char *subvol_name_part;
+
+		while ((subvol_name_part = strsep(&subvol_name_next, "/"))) {
+			mutex_lock(&root->d_inode->i_mutex);
+			new_root = lookup_one_len(subvol_name, root,
+						strlen(subvol_name));
+			mutex_unlock(&root->d_inode->i_mutex);
+
+			if (IS_ERR(new_root)) {
+				deactivate_locked_super(s);
+				error = PTR_ERR(new_root);
+				dput(root);
+				goto error_free_subvol_name;
+			}
+			if (!new_root->d_inode) {
+				dput(root);
+				dput(new_root);
+				deactivate_locked_super(s);
+				error = -ENXIO;
+				goto error_free_subvol_name;
+			}
 			dput(root);
-			dput(new_root);
-			deactivate_locked_super(s);
-			error = -ENXIO;
-			goto error_free_subvol_name;
+			root = new_root;
 		}
-		dput(root);
-		root = new_root;
 	}
 
 	kfree(subvol_name);
-- 
1.7.3.2

-- 
Calvin Walton <calvin.walton@gmail.com>

Re: [RFC-PATCH] Re: mounting arbitrary directories

[C Anthony Risinger]

On Sat, Nov 27, 2010 at 10:22 PM, Calvin Walton <calvin.walton@gmail.co=
m> wrote:
> On Sat, 2010-11-27 at 21:19 -0600, C Anthony Risinger wrote:
>> i have read just recently and in the past that btrfs supports COW fo=
r
>> _any_ file/directory (this is reflinks, yes?), and today i
>> accidentally noticed that i can mount a directory as well (if it's i=
n
>> the top level volume at least).
>>
>> eg. if i have a "regular" directory (not a subvolume) in the top-lev=
el:
>>
>> /__boot
>>
>> i can mount it with:
>>
>> mount -o subvol=3D__boot /dev/sda /mnt
>
> The 'subvol' option actually works using the same mechanism as a bind
> mount. The fact that it works using a directory is purely a coinciden=
ce
> - I would not expect it to be officially supported, and you shouldn't
> rely on it.

ok, good to know.

>> i am working on an update to my initramfs hook that will utilize
>> extlinux, and this property to provide seamless kernel level system
>> rollbacks, and i want to make sure it's safe to do this.
>
> To handle system rollbacks, you really should be using subvolumes and
> snapshots, not regular directories.

i do use snapshots, but the kernels cannot be read from them by
extlinux.  i must manually sync to either the top level btrfs root or
separate boot partition(s); this is required to handle kernel
upgrades... else you end up running a kernel against modules/etc.
built against an older/newer version.  the only other way i could
think of was to use a permanent intermediary kernel with the sole
purpose of mounting the btrfs device, and kexec'ing the correct kernel
within a snapshot.

>> also, is there a way to target an arbitrary directory? =A0does "subv=
ol"
>> support paths yet, or maybe via "subvolid" somehow? =A0essentially i
>
> I don't think that it would be very hard to make subvol=3D support a =
path
> instead of only one level deep. Actually, I think I could make a patc=
h
> for that myself... I've included it here. Mildly tested, but I'm not
> really a kernel programmer and might have missed something -
> particularly with regards to the locking.

nice, something like this is a good addition!  if it's really that
simple it should definitely be included.

>> just want to mount a directory inside a snapshot at /boot, so when
>> users upgrade there kernels, the images are visible to extlinux (whi=
ch
>> cannot yet peek inside or use subvolumes, so it has to be a regular
>> directory in the top-level volume)
>
> Ah, this is the first I've heard that extlinux doesn't support readin=
g
> files in subvolumes. That's an unfortunate limitation :/

well i actually thought it was working... but i ended up wasting
several hours only to learn that extlinux could "kind of" see them.
basically, usually the first snapshot appeared correct, but it was
really a recursive link back on itself.  sometimes snapshots would be
missing pieces -- seems to be related to COW; i think extlinux could
only see the chunks in snapshots that were still shared with the
top-level volume... but that's a observational guess.

needless to say, it really tripped me up for awhile until i remembered
the "rosh.c32" module and manually peeked around a bit :-)

C Anthony
--
To unsubscribe from this list: send the line "unsubscribe linux-btrfs" =
in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

Re: [RFC-PATCH] Re: mounting arbitrary directories

[C Anthony Risinger]

On Nov 27, 2010, at 10:22 PM, Calvin Walton <calvin.walton@gmail.com>
wrote:

> On Sat, 2010-11-27 at 21:19 -0600, C Anthony Risinger wrote:
>> i have read just recently and in the past that btrfs supports COW for
>> _any_ file/directory (this is reflinks, yes?), and today i
>> accidentally noticed that i can mount a directory as well (if it's in
>> the top level volume at least).
>>
>> eg. if i have a "regular" directory (not a subvolume) in the top-
>> level:
>>
>> /__boot
>>
>> i can mount it with:
>>
>> mount -o subvol=__boot /dev/sda /mnt
>
> The 'subvol' option actually works using the same mechanism as a bind
> mount. The fact that it works using a directory is purely a
> coincidence
> - I would not expect it to be officially supported, and you shouldn't
> rely on it.
>
>> i am working on an update to my initramfs hook that will utilize
>> extlinux, and this property to provide seamless kernel level system
>> rollbacks, and i want to make sure it's safe to do this.
>
> To handle system rollbacks, you really should be using subvolumes and
> snapshots, not regular directories.
>
>> also, is there a way to target an arbitrary directory?  does "subvol"
>> support paths yet, or maybe via "subvolid" somehow?  essentially i
>
> I don't think that it would be very hard to make subvol= support a
> path
> instead of only one level deep. Actually, I think I could make a patch
> for that myself... I've included it here. Mildly tested, but I'm not
> really a kernel programmer and might have missed something -
> particularly with regards to the locking.
>
>> just want to mount a directory inside a snapshot at /boot, so when
>> users upgrade there kernels, the images are visible to extlinux
>> (which
>> cannot yet peek inside or use subvolumes, so it has to be a regular
>> directory in the top-level volume)
>
> Ah, this is the first I've heard that extlinux doesn't support reading
> files in subvolumes. That's an unfortunate limitation :/

Hrrrm... Well I thought I'd be clever and use this little trick one
time to update my kernel... Anyways I oops out like 3 times in a row
(last func was <something>.clone in each IIRC) and now I'm stuck with
only my mobile since my server isn't set up yet and my  SSD just
failed on my netbook yesterday :-(

Sooooo, if anyone can help me recover this partition long enough to
grab a few things... I would be most grateful :-) I think I have
everything critical, but I'd rather take another look :-) Right now I
can't mount; mount is failing w/bad superblock.

/me promises to never recklessly sabotage himself after being warned <
6 hrs earlier

C Anthony [mobile]

Re: [RFC-PATCH] Re: mounting arbitrary directories

[C Anthony Risinger]

On Sun, Nov 28, 2010 at 4:07 AM, C Anthony Risinger <anthony@extof.me> =
wrote:
> On Nov 27, 2010, at 10:22 PM, Calvin Walton <calvin.walton@gmail.com>
> wrote:
>> On Sat, 2010-11-27 at 21:19 -0600, C Anthony Risinger wrote:
>>
>>> eg. if i have a "regular" directory (not a subvolume) in the top-
>>> level:
>>>
>>> /__boot
>>>
>>> i can mount it with:
>>>
>>> mount -o subvol=3D__boot /dev/sda /mnt
>>
>> The 'subvol' option actually works using the same mechanism as a bin=
d
>> mount. The fact that it works using a directory is purely a
>> coincidence
>> - I would not expect it to be officially supported, and you shouldn'=
t
>> rely on it.
>
> Hrrrm... Well I thought I'd be clever and use this little trick one
> time to update my kernel... Anyways I oops out like 3 times in a row
> (last func was <something>.clone in each IIRC) and now I'm stuck with
> only my mobile since my server isn't set up yet and my =A0SSD just
> failed on my netbook yesterday :-(
>
> Sooooo, if anyone can help me recover this partition long enough to
> grab a few things... I would be most grateful :-) I think I have
> everything critical, but I'd rather take another look :-) Right now I
> can't mount; mount is failing w/bad superblock.
>
> /me promises to never recklessly sabotage himself after being warned =
<
> 6 hrs earlier

any suggestions for me?  i dd'ed the corrupt partition to an external
disk because i need the machine back up and running, but if possible
i'd like to get the image running again.  i mounted it via loopback
and as expected got the same errors:

(will post the dmesg output next message -- left at home)
open_ctree_fd failed....
wanted XXXX found YYYY instead

the mount command itself fails with "bad superblock or ..."

i have seen others withe similar crash reports and IIRC correctly were
able to recover it (seems like something just didn't get updated right
before it locked...)

any ideas?

C Anthony
--
To unsubscribe from this list: send the line "unsubscribe linux-btrfs" =
in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

Re: [RFC-PATCH] Re: mounting arbitrary directories

[C Anthony Risinger]

On Mon, Nov 29, 2010 at 11:42 AM, C Anthony Risinger <anthony@extof.me>=
 wrote:
> On Sun, Nov 28, 2010 at 4:07 AM, C Anthony Risinger <anthony@extof.me=
> wrote:
>> On Nov 27, 2010, at 10:22 PM, Calvin Walton <calvin.walton@gmail.com=
>
>> wrote:
>>> On Sat, 2010-11-27 at 21:19 -0600, C Anthony Risinger wrote:
>>>
>>>> eg. if i have a "regular" directory (not a subvolume) in the top-
>>>> level:
>>>>
>>>> /__boot
>>>>
>>>> i can mount it with:
>>>>
>>>> mount -o subvol=3D__boot /dev/sda /mnt
>>>
>>> The 'subvol' option actually works using the same mechanism as a bi=
nd
>>> mount. The fact that it works using a directory is purely a
>>> coincidence
>>> - I would not expect it to be officially supported, and you shouldn=
't
>>> rely on it.
>>
>> Hrrrm... Well I thought I'd be clever and use this little trick one
>> time to update my kernel... Anyways I oops out like 3 times in a row
>> (last func was <something>.clone in each IIRC) and now I'm stuck wit=
h
>> only my mobile since my server isn't set up yet and my =A0SSD just
>> failed on my netbook yesterday :-(
>>
>> Sooooo, if anyone can help me recover this partition long enough to
>> grab a few things... I would be most grateful :-) I think I have
>> everything critical, but I'd rather take another look :-) Right now =
I
>> can't mount; mount is failing w/bad superblock.
>>
>> /me promises to never recklessly sabotage himself after being warned=
 <
>> 6 hrs earlier
>
> any suggestions for me? =A0i dd'ed the corrupt partition to an extern=
al
> disk because i need the machine back up and running, but if possible
> i'd like to get the image running again. =A0i mounted it via loopback
> and as expected got the same errors:
>
> (will post the dmesg output next message -- left at home)
> open_ctree_fd failed....
> wanted XXXX found YYYY instead
>
> the mount command itself fails with "bad superblock or ..."
>
> i have seen others withe similar crash reports and IIRC correctly wer=
e
> able to recover it (seems like something just didn't get updated righ=
t
> before it locked...)
>
> any ideas?

dmesg output was:

------------------------------------------------------
device label root devid 1 transid 61235 /dev/loop0
parent transid verify failed on 11335634944 wanted 61235 found 61237
parent transid verify failed on 11335634944 wanted 61235 found 61237
parent transid verify failed on 11335634944 wanted 61235 found 61237
btrfs: open_ctree failed
------------------------------------------------------

i tried btrfsck as suggested in the recent thread:

------------------------------------------------------
# sudo losetup /dev/loop0 /mnt/btrfs.img

# sudo btrfsck -s 1 /dev/loop0
using SB copy 1, bytenr 67108864
parent transid verify failed on 11335634944 wanted 61235 found 61237
parent transid verify failed on 11335634944 wanted 61235 found 61237
parent transid verify failed on 11335634944 wanted 61235 found 61237
btrfsck: disk-io.c:739: open_ctree_fd: Assertion `!(!tree_root->node)' =
failed.

# sudo btrfsck -s 2 /dev/loop0
using SB copy 2, bytenr 274877906944
No valid Btrfs found on /dev/loop0
------------------------------------------------------

soooo... does this mean i'm totally boned for ever mounting this image
again?  or should i keep it around for later?  or anything else i can
try/do?

C Anthony
--
To unsubscribe from this list: send the line "unsubscribe linux-btrfs" =
in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html