The memory leak was detected by kmemleak when running xfstests
overlay/051,053
Fixes: caf70cb2ba ("ovl: cleanup orphan index entries")
Cc: <stable@vger.kernel.org> # v4.13
Signed-off-by: Amir Goldstein <amir73il@gmail.com>
Signed-off-by: Miklos Szeredi <mszeredi@redhat.com>
ovl_free_fs() dereferences ofs->workbasedir and ofs->upper_mnt in cases when
those might not have been initialized yet.
Fix the initialization order for these fields.
Reported-by: syzbot+c75f181dc8429d2eb887@syzkaller.appspotmail.com
Signed-off-by: Miklos Szeredi <mszeredi@redhat.com>
Cc: <stable@vger.kernel.org> # v4.15
Fixes: 95e6d4177c ("ovl: grab reference to workbasedir early")
Fixes: a9075cdb46 ("ovl: factor out ovl_free_fs() helper")
Since overlayfs implements stacked file operations, the underlying
filesystems are not supposed to be exposed to the overlayfs file,
whose f_inode is an overlayfs inode.
Assigning an overlayfs file to swap_file results in an attempt of xfs
code to dereference an xfs_inode struct from an ovl_inode pointer:
CPU: 0 PID: 2462 Comm: swapon Not tainted
4.18.0-xfstests-12721-g33e17876ea4e #3402
RIP: 0010:xfs_find_bdev_for_inode+0x23/0x2f
Call Trace:
xfs_iomap_swapfile_activate+0x1f/0x43
__se_sys_swapon+0xb1a/0xee9
Fix this by not assigning the real inode mapping to f_mapping, which
will cause swapon() to return an error (-EINVAL). Although it makes
sense not to allow setting swpafile on an overlayfs file, some users
may depend on it, so we may need to fix this up in the future.
Keeping f_mapping pointing to overlay inode mapping will cause O_DIRECT
open to fail. Fix this by installing ovl_aops with noop_direct_IO in
overlay inode mapping.
Keeping f_mapping pointing to overlay inode mapping will cause other
a_ops related operations to fail (e.g. readahead()). Those will be
fixed by follow up patches.
Suggested-by: Miklos Szeredi <mszeredi@redhat.com>
Fixes: f7c72396d0 ("ovl: add O_DIRECT support")
Signed-off-by: Amir Goldstein <amir73il@gmail.com>
Signed-off-by: Miklos Szeredi <mszeredi@redhat.com>
...otherwise there will be list corruption due to inode_sb_list_add() being
called for inode already on the sb list.
Signed-off-by: Miklos Szeredi <mszeredi@redhat.com>
Fixes: e950564b97 ("vfs: don't evict uninitialized inode")
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
All the bits are in patches before this. So it is time to enable the
metadata only copy up feature.
Signed-off-by: Vivek Goyal <vgoyal@redhat.com>
Reviewed-by: Amir Goldstein <amir73il@gmail.com>
Signed-off-by: Miklos Szeredi <mszeredi@redhat.com>
ovl_copy_up() by default will only do metadata only copy up (if enabled).
That means when ovl_real_ioctl() calls ovl_real_file(), it will still get
the lower file (as ovl_real_file() opens data file and not metacopy). And
that means "chattr +i" will end up modifying lower inode.
There seem to be two ways to solve this.
A. Open metacopy file in ovl_real_ioctl() and do operations on that
B. Force full copy up when FS_IOC_SETFLAGS is called.
I am resorting to option B for now as it feels little safer option. If
there are performance issues due to this, we can revisit it.
Signed-off-by: Vivek Goyal <vgoyal@redhat.com>
Reviewed-by: Amir Goldstein <amir73il@gmail.com>
Signed-off-by: Miklos Szeredi <mszeredi@redhat.com>
truncate should copy up full file (and not do metacopy only), otherwise it
will be broken. For example, use truncate to increase size of a file so
that any read beyong existing size will return null bytes. If we don't
copy up full file, then we end up opening lower file and read from it only
reads upto the old size (and not new size after truncate). Hence to avoid
such situations, copy up data as well when file size changes.
So far it was being done by d_real(O_WRONLY) call in truncate() path. Now
that patch has been reverted. So force full copy up in ovl_setattr() if
size of file is changing.
Signed-off-by: Vivek Goyal <vgoyal@redhat.com>
Reviewed-by: Amir Goldstein <amir73il@gmail.com>
Signed-off-by: Miklos Szeredi <mszeredi@redhat.com>
Right now we seem to check redirect only if upperdentry is found. But it
is possible that there is no upperdentry but later we found an index.
We need to check redirect on index as well and set it in
ovl_inode->redirect. Otherwise link code can assume that dentry does not
have redirect and place a new one which breaks things. In my testing
overlay/033 test started failing in xfstests. Following are the details.
For example do following.
$ mkdir lower upper work merged
- Make lower dir with 4 links.
$ echo "foo" > lower/l0.txt
$ ln lower/l0.txt lower/l1.txt
$ ln lower/l0.txt lower/l2.txt
$ ln lower/l0.txt lower/l3.txt
- Mount with index on and metacopy on.
$ mount -t overlay -o lowerdir=lower,upperdir=upper,workdir=work,\
index=on,metacopy=on none merged
- Link lower
$ ln merged/l0.txt merged/l4.txt
(This will metadata copy up of l0.txt and put an absolute redirect
/l0.txt)
$ echo 2 > /proc/sys/vm/drop/caches
$ ls merged/l1.txt
(Now l1.txt will be looked up. There is no upper dentry but there is
lower dentry and index will be found. We don't check for redirect on
index, hence ovl_inode->redirect will be NULL.)
- Link Upper
$ ln merged/l4.txt merged/l5.txt
(Lookup of l4.txt will use inode from l1.txt lookup which is still in
cache. It has ovl_inode->redirect NULL, hence link will put a new
redirect and replace /l0.txt with /l4.txt
- Drop caches.
echo 2 > /proc/sys/vm/drop_caches
- List l1.txt and it returns -ESTALE
$ ls merged/l0.txt
(It returns stale because, we found a metacopy of l0.txt in upper and it
has redirect l4.txt but there is no file named l4.txt in lower layer.
So lower data copy is not found and -ESTALE is returned.)
So problem here is that we did not process redirect on index. Check
redirect on index as well and then problem is fixed.
Signed-off-by: Vivek Goyal <vgoyal@redhat.com>
Reviewed-by: Amir Goldstein <amir73il@gmail.com>
Signed-off-by: Miklos Szeredi <mszeredi@redhat.com>
When we create a hardlink to a metacopy upper file, first the redirect on
that inode. Path based lookup will not work with newly created link and
redirect will solve that issue.
Also use absolute redirect as two hardlinks could be in different
directores and relative redirect will not work.
I have not put any additional locking around setting redirects while
introducing redirects for non-dir files. For now it feels like existing
locking is sufficient. If that's not the case, we will have add more
locking. Following is my rationale about why do I think current locking
seems ok.
Basic problem for non-dir files is that more than on dentry could be
pointing to same inode and in theory only relying on dentry based locks
(d->d_lock) did not seem sufficient.
We set redirect upon rename and upon link creation. In both the paths for
non-dir file, VFS locks both source and target inodes (->i_rwsem). That
means vfs rename and link operations on same source and target can't he
happening in parallel (Even if there are multiple dentries pointing to same
inode). So that probably means that at a time on an inode, only one call
of ovl_set_redirect() could be working and we don't need additional locking
in ovl_set_redirect().
ovl_inode->redirect is initialized only when inode is created new. That
means it should not race with any other path and setting
ovl_inode->redirect should be fine.
Reading of ovl_inode->redirect happens in ovl_get_redirect() path. And
this called only in ovl_set_redirect(). And ovl_set_redirect() already
seemed to be protected using ->i_rwsem. That means ovl_set_redirect() and
ovl_get_redirect() on source/target inode should not make progress in
parallel and is mutually exclusive. Hence no additional locking required.
Now, only case where ovl_set_redirect() and ovl_get_redirect() could race
seems to be case of absolute redirects where ovl_get_redirect() has to
travel up the tree. In that case we already take d->d_lock and that should
be sufficient as directories will not have multiple dentries pointing to
same inode.
So given VFS locking and current usage of redirect, current locking around
redirect seems to be ok for non-dir as well. Once we have the logic to
remove redirect when metacopy file gets copied up, then we probably will
need additional locking.
Signed-off-by: Vivek Goyal <vgoyal@redhat.com>
Reviewed-by: Amir Goldstein <amir73il@gmail.com>
Signed-off-by: Miklos Szeredi <mszeredi@redhat.com>
Set redirect on metacopy files upon rename. This will help find data
dentry in lower dirs.
Signed-off-by: Vivek Goyal <vgoyal@redhat.com>
Reviewed-by: Amir Goldstein <amir73il@gmail.com>
Signed-off-by: Miklos Szeredi <mszeredi@redhat.com>
If a dentry has copy up origin, we set flag OVL_PATH_ORIGIN. So far this
decision was easy that we had to check only for oe->numlower and if it is
non-zero, we knew there is copy up origin. (For non-dir we installed
origin dentry in lowerstack[0]).
But we don't create ORGIN xattr for broken hardlinks (index=off). And with
metacopy feature it is possible that we will install lowerstack[0] but
ORIGIN xattr is not there. It is data dentry of upper metacopy dentry
which has been found using regular name based lookup or using REDIRECT. So
with addition of this new case, just presence of oe->numlower is not
sufficient to guarantee that ORIGIN xattr is present.
So to differentiate between two cases, look at OVL_CONST_INO flag. If this
flag is set and upperdentry is there, that means it can be marked as type
ORIGIN. OVL_CONST_INO is not set if lower hardlink is broken or will be
broken over copy up.
Signed-off-by: Vivek Goyal <vgoyal@redhat.com>
Reviewed-by: Amir Goldstein <amir73il@gmail.com>
Signed-off-by: Miklos Szeredi <mszeredi@redhat.com>
Add an ovl_inode flag OVL_CONST_INO. This flag signifies if inode number
will remain constant over copy up or not. This flag does not get updated
over copy up and remains unmodifed after setting once.
Next patch in the series will make use of this flag. It will basically
figure out if dentry is of type ORIGIN or not. And this can be derived by
this flag.
ORIGIN = (upperdentry && ovl_test_flag(OVL_CONST_INO, inode)).
Suggested-by: Amir Goldstein <amir73il@gmail.com>
Signed-off-by: Vivek Goyal <vgoyal@redhat.com>
Reviewed-by: Amir Goldstein <amir73il@gmail.com>
Signed-off-by: Miklos Szeredi <mszeredi@redhat.com>
Right now OVL_PATH_MERGE is used only for merged directories. But
conceptually, a metacopy dentry (backed by a lower data dentry) is a merged
entity as well.
So mark metacopy dentries as OVL_PATH_MERGE and ovl_rename() makes use of
this property later to set redirect on a metacopy file.
Signed-off-by: Vivek Goyal <vgoyal@redhat.com>
Reviewed-by: Amir Goldstein <amir73il@gmail.com>
Signed-off-by: Miklos Szeredi <mszeredi@redhat.com>
Right now we rely on path based lookup for data origin of metacopy upper.
This will work only if upper has not been renamed. We solved this problem
already for merged directories using redirect. Use same logic for metacopy
files.
This patch just goes on to check redirects for metacopy files.
Signed-off-by: Vivek Goyal <vgoyal@redhat.com>
Reviewed-by: Amir Goldstein <amir73il@gmail.com>
Signed-off-by: Miklos Szeredi <mszeredi@redhat.com>
Move some directory related code in else block. This is pure code
reorganization and no functionality change.
Next patch enables redirect processing on metacopy files and needs this
change. By keeping non-functional changes in a separate patch, next patch
looks much smaller and cleaner.
Signed-off-by: Vivek Goyal <vgoyal@redhat.com>
Reviewed-by: Amir Goldstein <amir73il@gmail.com>
Signed-off-by: Miklos Szeredi <mszeredi@redhat.com>
Metacopy dentry/inode is internal to overlay and is never exposed outside
of it. Exception is metacopy upper file used for fsync(). Modify d_real()
to look for dentries/inode which have data, but also allow matching upper
inode without data for the fsync case.
Signed-off-by: Vivek Goyal <vgoyal@redhat.com>
Reviewed-by: Amir Goldstein <amir73il@gmail.com>
Signed-off-by: Miklos Szeredi <mszeredi@redhat.com>
ovl_open() should open file which contains data and not open metacopy
inode. With the introduction of metacopy inodes, with current
implementaion we will end up opening metacopy inode as well.
But there can be certain circumstances like ovl_fsync() where we want to
allow opening a metacopy inode instead.
Hence, change ovl_open_realfile() and and add extra parameter which
specifies whether to allow opening metacopy inode or not. If this
parameter is false, we look for data inode and open that.
This should allow covering both the cases.
Signed-off-by: Vivek Goyal <vgoyal@redhat.com>
Reviewed-by: Amir Goldstein <amir73il@gmail.com>
Signed-off-by: Miklos Szeredi <mszeredi@redhat.com>
Add an helper to retrieve real data inode associated with overlay inode.
This helper will ignore all metacopy inodes and will return only the real
inode which has data.
Signed-off-by: Vivek Goyal <vgoyal@redhat.com>
Reviewed-by: Amir Goldstein <amir73il@gmail.com>
Signed-off-by: Miklos Szeredi <mszeredi@redhat.com>
Right now ovl_inode stores inode pointer for lower inode. This helps with
quickly getting lower inode given overlay inode (ovl_inode_lower()).
Now with metadata only copy-up, we can have metacopy inode in middle layer
as well and inode containing data can be different from ->lower. I need to
be able to open the real file in ovl_open_realfile() and for that I need to
quickly find the lower data inode.
Hence store lower data inode also in ovl_inode. Also provide an helper
ovl_inode_lowerdata() to access this field.
Signed-off-by: Vivek Goyal <vgoyal@redhat.com>
Reviewed-by: Amir Goldstein <amir73il@gmail.com>
Signed-off-by: Miklos Szeredi <mszeredi@redhat.com>
If an inode has been copied up metadata only, then we need to query the
number of blocks from lower and fill up the stat->st_blocks.
We need to be careful about races where we are doing stat on one cpu and
data copy up is taking place on other cpu. We want to return
stat->st_blocks either from lower or stable upper and not something in
between. Hence, ovl_has_upperdata() is called first to figure out whether
block reporting will take place from lower or upper.
We now support metacopy dentries in middle layer. That means number of
blocks reporting needs to come from lowest data dentry and this could be
different from lower dentry. Hence we end up making a separate
vfs_getxattr() call for metacopy dentries to get number of blocks.
Signed-off-by: Vivek Goyal <vgoyal@redhat.com>
Reviewed-by: Amir Goldstein <amir73il@gmail.com>
Signed-off-by: Miklos Szeredi <mszeredi@redhat.com>
Now we have the notion of data dentry and metacopy dentry.
ovl_dentry_lower() will return uppermost lower dentry, but it could be
either data or metacopy dentry. Now we support metacopy dentries in lower
layers so it is possible that lowerstack[0] is metacopy dentry while
lowerstack[1] is actual data dentry.
So add an helper which returns lowest most dentry which is supposed to be
data dentry.
Signed-off-by: Vivek Goyal <vgoyal@redhat.com>
Reviewed-by: Amir Goldstein <amir73il@gmail.com>
Signed-off-by: Miklos Szeredi <mszeredi@redhat.com>
So far lower could not be a meta inode. So whenever it was time to copy up
data of a meta inode, we could copy it up from top most lower dentry.
But now lower itself can be a metacopy inode. That means data copy up
needs to take place from a data inode in metacopy inode chain. Find lower
data inode in the chain and use that for data copy up.
Introduced a helper called ovl_path_lowerdata() to find the lower data
inode chain.
Signed-off-by: Vivek Goyal <vgoyal@redhat.com>
Reviewed-by: Amir Goldstein <amir73il@gmail.com>
Signed-off-by: Miklos Szeredi <mszeredi@redhat.com>
This patch modifies ovl_lookup() and friends to lookup metacopy dentries.
It also allows for presence of metacopy dentries in lower layer.
During lookup, check for presence of OVL_XATTR_METACOPY and if not present,
set OVL_UPPERDATA bit in flags.
We don't support metacopy feature with nfs_export. So in nfs_export code,
we set OVL_UPPERDATA flag set unconditionally if upper inode exists.
Do not follow metacopy origin if we find a metacopy only inode and metacopy
feature is not enabled for that mount. Like redirect, this can have
security implications where an attacker could hand craft upper and try to
gain access to file on lower which it should not have to begin with.
Signed-off-by: Vivek Goyal <vgoyal@redhat.com>
Reviewed-by: Amir Goldstein <amir73il@gmail.com>
Signed-off-by: Miklos Szeredi <mszeredi@redhat.com>
Right now we use goto out_nomem which assumes error code is -ENOMEM. But
there are other errors returned like -ESTALE as well. So instead of
out_nomem, use out_err which will do ERR_PTR(err). That way one can put
error code in err and jump to out_err.
This just code reorganization and no change of functionality.
I am about to add more code and this organization helps laying more code
and error paths on top of it.
Signed-off-by: Vivek Goyal <vgoyal@redhat.com>
Reviewed-by: Amir Goldstein <amir73il@gmail.com>
Signed-off-by: Miklos Szeredi <mszeredi@redhat.com>
Now we will have the capability to have upper inodes which might be only
metadata copy up and data is still on lower inode. So add a new xattr
OVL_XATTR_METACOPY to distinguish between two cases.
Presence of OVL_XATTR_METACOPY reflects that file has been copied up
metadata only and and data will be copied up later from lower origin. So
this xattr is set when a metadata copy takes place and cleared when data
copy takes place.
We also use a bit in ovl_inode->flags to cache OVL_UPPERDATA which reflects
whether ovl inode has data or not (as opposed to metadata only copy up).
If a file is copied up metadata only and later when same file is opened for
WRITE, then data copy up takes place. We copy up data, remove METACOPY
xattr and then set the UPPERDATA flag in ovl_inode->flags. While all these
operations happen with oi->lock held, read side of oi->flags can be
lockless. That is another thread on another cpu can check if UPPERDATA
flag is set or not.
So this gives us an ordering requirement w.r.t UPPERDATA flag. That is, if
another cpu sees UPPERDATA flag set, then it should be guaranteed that
effects of data copy up and remove xattr operations are also visible.
For example.
CPU1 CPU2
ovl_open() acquire(oi->lock)
ovl_open_maybe_copy_up() ovl_copy_up_data()
open_open_need_copy_up() vfs_removexattr()
ovl_already_copied_up()
ovl_dentry_needs_data_copy_up() ovl_set_flag(OVL_UPPERDATA)
ovl_test_flag(OVL_UPPERDATA) release(oi->lock)
Say CPU2 is copying up data and in the end sets UPPERDATA flag. But if
CPU1 perceives the effects of setting UPPERDATA flag but not the effects of
preceding operations (ex. upper that is not fully copied up), it will be a
problem.
Hence this patch introduces smp_wmb() on setting UPPERDATA flag operation
and smp_rmb() on UPPERDATA flag test operation.
May be some other lock or barrier is already covering it. But I am not sure
what that is and is it obvious enough that we will not break it in future.
So hence trying to be safe here and introducing barriers explicitly for
UPPERDATA flag/bit.
Signed-off-by: Vivek Goyal <vgoyal@redhat.com>
Reviewed-by: Amir Goldstein <amir73il@gmail.com>
Signed-off-by: Miklos Szeredi <mszeredi@redhat.com>
There are couple of places where we need to know if file is already copied
up (in lockless manner). Right now its open coded and there are only two
conditions to check. Soon this patch series will introduce another
condition to check and Amir wants to introduce one more. So introduce a
helper instead to check this so that code is easier to read.
Signed-off-by: Vivek Goyal <vgoyal@redhat.com>
Reviewed-by: Amir Goldstein <amir73il@gmail.com>
Signed-off-by: Miklos Szeredi <mszeredi@redhat.com>
If it makes sense to copy up only metadata during copy up, do it. This is
done for regular files which are not opened for WRITE.
Right now ->metacopy is set to 0 always. Last patch in the series will
remove the hard coded statement and enable metacopy feature.
Signed-off-by: Vivek Goyal <vgoyal@redhat.com>
Reviewed-by: Amir Goldstein <amir73il@gmail.com>
Signed-off-by: Miklos Szeredi <mszeredi@redhat.com>
Just a little re-ordering of code. This helps with next patch where after
copying up metadata, we skip data copying step, if needed.
Signed-off-by: Vivek Goyal <vgoyal@redhat.com>
Reviewed-by: Amir Goldstein <amir73il@gmail.com>
Signed-off-by: Miklos Szeredi <mszeredi@redhat.com>
By default metadata only copy up is disabled. Provide a mount option so
that users can choose one way or other.
Also provide a kernel config and module option to enable/disable metacopy
feature.
metacopy feature requires redirect_dir=on when upper is present.
Otherwise, it requires redirect_dir=follow atleast.
As of now, metacopy does not work with nfs_export=on. So if both
metacopy=on and nfs_export=on then nfs_export is disabled.
Signed-off-by: Vivek Goyal <vgoyal@redhat.com>
Reviewed-by: Amir Goldstein <amir73il@gmail.com>
Signed-off-by: Miklos Szeredi <mszeredi@redhat.com>
Right now two copy up helpers are in inode.c. Amir suggested it might be
better to move these to copy_up.c.
There will one more related function which will come in later patch.
Signed-off-by: Vivek Goyal <vgoyal@redhat.com>
Reviewed-by: Amir Goldstein <amir73il@gmail.com>
Signed-off-by: Miklos Szeredi <mszeredi@redhat.com>
ovl_inode->redirect is an inode property and should be initialized in
ovl_get_inode() only when we are adding a new inode to cache. If inode is
already in cache, it is already initialized and we should not be touching
ovl_inode->redirect field.
As of now this is not a problem as redirects are used only for directories
which don't share inode. But soon I want to use redirects for regular
files also and there it can become an issue.
Hence, move ->redirect initialization in ovl_get_inode().
Signed-off-by: Vivek Goyal <vgoyal@redhat.com>
Reviewed-by: Amir Goldstein <amir73il@gmail.com>
Signed-off-by: Miklos Szeredi <mszeredi@redhat.com>
This was provided for debugging the ro/rw inconsistecy. The inconsitency
is now gone so this option is obsolete.
Signed-off-by: Miklos Szeredi <mszeredi@redhat.com>
Opening regular files on overlayfs is now handled via ovl_open(). Remove
the now unused "open_flags" argument from d_op->d_real() and the d_real()
helper.
Signed-off-by: Miklos Szeredi <mszeredi@redhat.com>
This partially reverts commit c568d68341.
Overlayfs files will now automatically get the correct locks, no need to
hack overlay support in VFS.
It is a partial revert, because it leaves the locks_inode() calls in place
and defines locks_inode() to file_inode(). We could revert those as well,
but it would be unnecessary code churn and it makes sense to document that
we are getting the inode for locking purposes.
Don't revert MS_NOREMOTELOCK yet since that has been part of the userspace
API for some time (though not in a useful way). Will try to remove
internal flags later when the dust around the new mount API settles.
Signed-off-by: Miklos Szeredi <mszeredi@redhat.com>
Acked-by: Jeff Layton <jlayton@kernel.org>
Implement stacked fiemap().
Need to split inode operations for regular file (which has fiemap) and
special file (which doesn't have fiemap).
Signed-off-by: Miklos Szeredi <mszeredi@redhat.com>
In the common case we can just use the real file cached in
file->private_data. There are two exceptions:
1) File has been copied up since open: in this unlikely corner case just
use a throwaway real file for the operation. If ever this becomes a
perfomance problem (very unlikely, since overlayfs has been doing most fine
without correctly handling this case at all), then we can deal with that by
updating the cached real file.
2) File's f_flags have changed since open: no need to reopen the cached
real file, we can just change the flags there as well.
Signed-off-by: Miklos Szeredi <mszeredi@redhat.com>
Implement file operations on a regular overlay file. The underlying file
is opened separately and cached in ->private_data.
It might be worth making an exception for such files when accounting in
nr_file to confirm to userspace expectations. We are only adding a small
overhead (248bytes for the struct file) since the real inode and dentry are
pinned by overlayfs anyway.
This patch doesn't have any effect, since the vfs will use d_real() to find
the real underlying file to open. The patch at the end of the series will
actually enable this functionality.
AV: make it use open_with_fake_path(), don't mess with override_creds
SzM: still need to mess with override_creds() until no fs uses
current_cred() in their open method.
Signed-off-by: Miklos Szeredi <mszeredi@redhat.com>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
Amir Goldstein