When decoding a lower file handle, we first call ovl_check_origin_fh()
with connected=false to get any real lower dentry for overlay inode
cache lookup.
If the real dentry is a disconnected dir dentry, ovl_check_origin_fh()
is called again with connected=true to get a connected real dentry
and find the lower layer the real dentry belongs to.
If the first call returned a connected real dentry, we use it to
lookup an overlay connected dentry, but the first ovl_check_origin_fh()
call with connected=false did not check that the found dentry is under
the root of the layer (see ovl_acceptable()), it only checked that
the found dentry super block matches the uuid of the lower file handle.
In case there are multiple lower layers on the same fs and the found
dentry is not from the top most lower layer, using the layer index
returned from the first ovl_check_origin_fh() is wrong and we end
up failing to decode the file handle.
Fix this by always calling ovl_check_origin_fh() with connected=true
if we got a directory dentry in the first call.
Fixes: 8b58924ad5 ("ovl: lookup in inode cache first when decoding...")
Cc: <[email protected]> # v4.17
Signed-off-by: Amir Goldstein <[email protected]>
Signed-off-by: Miklos Szeredi <[email protected]>
Theodore Ts'o reported a v4.19 regression with docker-dropbox:
https://marc.info/?l=linux-fsdevel&m=154070089431116&w=2
"I was rebuilding my dropbox Docker container, and it failed in 4.19
with the following error:
...
dpkg: error: error creating new backup file \
'/var/lib/dpkg/status-old': Invalid cross-device link"
The problem did not reproduce with metacopy feature disabled.
The error was caused by insufficient credentials to set
"trusted.overlay.redirect" xattr on link of a metacopy file.
Reproducer:
echo Y > /sys/module/overlay/parameters/redirect_dir
echo Y > /sys/module/overlay/parameters/metacopy
cd /tmp
mkdir l u w m
chmod 777 l u
touch l/foo
ln l/foo l/link
chmod 666 l/foo
mount -t overlay none -olowerdir=l,upperdir=u,workdir=w m
su fsgqa
ln m/foo m/bar
[ 21.455823] overlayfs: failed to set redirect (-1)
ln: failed to create hard link 'm/bar' => 'm/foo':\
Invalid cross-device link
Reported-by: Theodore Y. Ts'o <[email protected]>
Reported-by: Maciej Zięba <[email protected]>
Fixes: 4120fe64dc ("ovl: Set redirect on upper inode when it is linked")
Cc: <[email protected]> # v4.19
Signed-off-by: Amir Goldstein <[email protected]>
Acked-by: Vivek Goyal <[email protected]>
Signed-off-by: Miklos Szeredi <[email protected]>
When an event is reported on a sub-directory and the parent inode has
a mark mask with FS_EVENT_ON_CHILD|FS_ISDIR, the event will be sent to
fsnotify() even if the event type is not in the parent mark mask
(e.g. FS_OPEN).
Further more, if that event happened on a mount or a filesystem with
a mount/sb mark that does have that event type in their mask, the "on
child" event will be reported on the mount/sb mark. That is not
desired, because user will get a duplicate event for the same action.
Note that the event reported on the victim inode is never merged with
the event reported on the parent inode, because of the check in
should_merge(): old_fsn->inode == new_fsn->inode.
Fix this by looking for a match of an actual event type (i.e. not just
FS_ISDIR) in parent's inode mark mask and by not reporting an "on child"
event to group if event type is only found on mount/sb marks.
[backport hint: The bug seems to have always been in fanotify, but this
patch will only apply cleanly to v4.19.y]
Cc: <[email protected]> # v4.19
Signed-off-by: Amir Goldstein <[email protected]>
Signed-off-by: Jan Kara <[email protected]>
printk format used %*s instead of %.*s, so hostname_len does not limit
the number of bytes accessed from hostname.
Signed-off-by: Amir Goldstein <[email protected]>
Cc: [email protected]
Signed-off-by: J. Bruce Fields <[email protected]>
It just makes the interface strange without adding any significant value.
The only case where locked is false and return value is 0 is in
ovl_rename() when new is negative, so handle that case explicitly in
ovl_rename().
Signed-off-by: Amir Goldstein <[email protected]>
Signed-off-by: Miklos Szeredi <[email protected]>
We use uuid to associate an overlay lower file handle with a lower layer,
so we can accept lower fs with null uuid as long as all lower layers with
null uuid are on the same fs.
This change allows enabling index and nfs_export features for the setup of
single lower fs of type squashfs - squashfs supports file handles, but has
a null uuid. This change also allows enabling index and nfs_export features
for nested overlayfs, where the lower overlay has nfs_export enabled.
Enabling the index feature with single lower squashfs fixes the
unionmount-testsuite test:
./run --ov --squashfs --verify
As a by-product, if, like the lower squashfs, upper fs also uses the
generic export_encode_fh() implementation to export 32bit inode file
handles (e.g. ext4), then the xino_auto config/module/mount option will
enable unique overlay inode numbers.
Signed-off-by: Amir Goldstein <[email protected]>
Signed-off-by: Miklos Szeredi <[email protected]>
In an attempt to dedup ~100 LOC, we ended up creating a tangled call chain,
whose branches merge and diverge in several points according to the
immutable c->tmpfile copy up mode.
This call chain was hard to analyse for locking correctness because the
locking requirements for the c->tmpfile flow were very different from the
locking requirements for the !c->tmpfile flow (i.e. directory vs. regulare
file copy up).
Split the copy up helpers of the c->tmpfile flow from those of the
!c->tmpfile (i.e. workdir) flow and remove the c->tmpfile mode from copy up
context.
Suggested-by: Al Viro <[email protected]>
Signed-off-by: Amir Goldstein <[email protected]>
Signed-off-by: Miklos Szeredi <[email protected]>
linking a non-copied-up file into a non-copied-up parent results in a
nested call to mutex_lock_interruptible(&oi->lock). Fix this by copying up
target parent before ovl_nlink_start(), same as done in ovl_rename().
~/unionmount-testsuite$ ./run --ov -s
~/unionmount-testsuite$ ln /mnt/a/foo100 /mnt/a/dir100/
WARNING: possible recursive locking detected
--------------------------------------------
ln/1545 is trying to acquire lock:
00000000bcce7c4c (&ovl_i_lock_key[depth]){+.+.}, at:
ovl_copy_up_start+0x28/0x7d
but task is already holding lock:
0000000026d73d5b (&ovl_i_lock_key[depth]){+.+.}, at:
ovl_nlink_start+0x3c/0xc1
[SzM: this seems to be a false positive, but doing the copy-up first is
harmless and removes the lockdep splat]
Reported-by: [email protected]
Fixes: 5f8415d6b8 ("ovl: persistent overlay inode nlink for...")
Cc: <[email protected]> # v4.13
Signed-off-by: Amir Goldstein <[email protected]>
Signed-off-by: Miklos Szeredi <[email protected]>
Some anon_bdev filesystems (e.g. overlayfs, ceph) don't have s_blocksize
set. Returning zero from FIGETBSZ ioctl results in a Floating point
exception from the e2fsprogs utility filefrag, which divides the size of
the file with the value returned by FIGETBSZ.
Fix the interface by returning -EINVAL for these filesystems.
Fixes: d1d04ef857 ("ovl: stack file ops")
Cc: <[email protected]> # v4.19
Signed-off-by: Amir Goldstein <[email protected]>
Signed-off-by: Miklos Szeredi <[email protected]>
Kernel test robot reported [1] a 6% performance regression in a
concurrent unlink(2) workload on commit 60f7ed8c7c ("fsnotify: send
path type events to group with super block marks").
The performance test was run with no fsnotify marks at all on the
data set, so the only extra instructions added by the offending
commit are tests of the super_block fields s_fsnotify_{marks,mask}
and these tests happen on almost every single inode access.
When adding those fields to the super_block struct, we did not give much
thought of placing them on a hot cache lines (we just placed them at the
end of the struct).
Re-organize struct super_block to try and keep some frequently accessed
fields on the same cache line.
Move the frequently accessed fields s_fsnotify_{marks,mask} near the
frequently accessed fields s_fs_info,s_time_gran, while filling a 64bit
alignment hole after s_time_gran.
Move the seldom accessed fields s_id,s_uuid,s_max_links,s_mode near the
seldom accessed fields s_vfs_rename_mutex,s_subtype.
Rong Chen confirmed that this patch solved the reported problem.
[1] https://lkml.org/lkml/2018/9/30/206
Reported-by: kernel test robot <[email protected]>
Tested-by: kernel test robot <[email protected]>
Fixes: 1e6cb72399 ("fsnotify: add super block object type")
Signed-off-by: Amir Goldstein <[email protected]>
Signed-off-by: Jan Kara <[email protected]>
In order to identify which thread triggered the event in a
multi-threaded program, add the FAN_REPORT_TID flag in fanotify_init to
opt-in for reporting the event creator's thread id information.
Signed-off-by: nixiaoming <[email protected]>
Signed-off-by: Amir Goldstein <[email protected]>
Signed-off-by: Jan Kara <[email protected]>
Also define the FANOTIFY_EVENT_FLAGS consisting of the extra flags
FAN_ONDIR and FAN_ON_CHILD.
Signed-off-by: Amir Goldstein <[email protected]>
Signed-off-by: Jan Kara <[email protected]>
The BUG_ON() statements to verify number of bits in ALL_FSNOTIFY_BITS
and ALL_INOTIFY_BITS are converted to build time check of the constant.
Signed-off-by: Amir Goldstein <[email protected]>
Signed-off-by: Jan Kara <[email protected]>
We do not want to add new bits to the FAN_ALL_* uapi constants
because they have been exposed to userspace. If there are programs
out there using these constants, those programs could break if
re-compiled with modified FAN_ALL_* constants and run on an old kernel.
We deprecate the uapi constants FAN_ALL_* and define new FANOTIFY_*
constants for internal use to replace them. New feature bits will be
added only to the new constants.
Cc: <[email protected]>
Signed-off-by: Amir Goldstein <[email protected]>
Signed-off-by: Jan Kara <[email protected]>
fanotify mark add/remove code jumps through hoops to avoid setting the
FS_ISDIR in the commulative object mask.
That was just papering over a bug in fsnotify() handling of the FS_ISDIR
extra flag. This bug is now fixed, so all the hoops can be removed along
with the unneeded internal flag FAN_MARK_ONDIR.
Signed-off-by: Amir Goldstein <[email protected]>
Signed-off-by: Jan Kara <[email protected]>
FS_EVENT_ON_CHILD gets a special treatment in fsnotify() because it is
not a flag specifying an event type, but rather an extra flags that may
be reported along with another event and control the handling of the
event by the backend.
FS_ISDIR is also an "extra flag" and not an "event type" and therefore
desrves the same treatment. With inotify/dnotify backends it was never
possible to set FS_ISDIR in mark masks, so it did not matter.
With fanotify backend, mark adding code jumps through hoops to avoid
setting the FS_ISDIR in the commulative object mask.
Separate the constant ALL_FSNOTIFY_EVENTS to ALL_FSNOTIFY_FLAGS and
ALL_FSNOTIFY_EVENTS, so the latter can be used to test for specific
event types.
Signed-off-by: Amir Goldstein <[email protected]>
Signed-off-by: Jan Kara <[email protected]>
The new mark flag FAN_MARK_FILESYSTEMS collides with existing internal
flag FAN_MARK_ONDIR. Change internal flag value to avoid the collision.
Fixes: d54f4fba88 ("fanotify: add API to attach/detach super block mark")
Signed-off-by: Amir Goldstein <[email protected]>
Signed-off-by: Jan Kara <[email protected]>
This averts the need to re-generate flags in fanotify_show_fdinfo()
and sets the scene for addition of more upcoming flags without growing
new members to the fanotify_data struct.
Signed-off-by: Amir Goldstein <[email protected]>
Signed-off-by: Jan Kara <[email protected]>
Commit 031a072a0b ("vfs: call vfs_clone_file_range() under freeze
protection") created a wrapper do_clone_file_range() around
vfs_clone_file_range() moving the freeze protection to former, so
overlayfs could call the latter.
The more common vfs practice is to call do_xxx helpers from vfs_xxx
helpers, where freeze protecction is taken in the vfs_xxx helper, so
this anomality could be a source of confusion.
It seems that commit 8ede205541 ("ovl: add reflink/copyfile/dedup
support") may have fallen a victim to this confusion -
ovl_clone_file_range() calls the vfs_clone_file_range() helper in the
hope of getting freeze protection on upper fs, but in fact results in
overlayfs allowing to bypass upper fs freeze protection.
Swap the names of the two helpers to conform to common vfs practice
and call the correct helpers from overlayfs and nfsd.
Signed-off-by: Amir Goldstein <[email protected]>
Signed-off-by: Miklos Szeredi <[email protected]>
Tested by doing clone on overlayfs while upper xfs+reflink is frozen:
xfs_io -f /ovl/y
fsfreeze -f /xfs
xfs_io> reflink /ovl/x
Before the fix xfs_io enters xfs_reflink_remap_range() and blocks
in xfs_trans_alloc(). After the fix, xfs_io blocks outside xfs code
in ovl_clone_file_range().
Fixes: 8ede205541 ("ovl: add reflink/copyfile/dedup support")
Signed-off-by: Amir Goldstein <[email protected]>
Signed-off-by: Miklos Szeredi <[email protected]>
Tested by re-writing to an open overlayfs file while upper ext4 is frozen:
xfs_io -f /ovl/x
xfs_io> pwrite 0 4096
fsfreeze -f /ext4
xfs_io> pwrite 0 4096
WARNING: CPU: 0 PID: 1492 at fs/ext4/ext4_jbd2.c:53 \
ext4_journal_check_start+0x48/0x82
After the fix, the second write blocks in ovl_write_iter() and avoids
hitting WARN_ON(sb->s_writers.frozen == SB_FREEZE_COMPLETE) in
ext4_journal_check_start().
Fixes: 2a92e07edc ("ovl: add ovl_write_iter()")
Signed-off-by: Amir Goldstein <[email protected]>
Signed-off-by: Miklos Szeredi <[email protected]>
Add another mark type flag FAN_MARK_FILESYSTEM for add/remove/flush
of super block mark type.
A super block watch gets all events on the filesystem, regardless of
the mount from which the mark was added, unless an ignore mask exists
on either the inode or the mount where the event was generated.
Only one of FAN_MARK_MOUNT and FAN_MARK_FILESYSTEM mark type flags
may be provided to fanotify_mark() or no mark type flag for inode mark.
Cc: <[email protected]>
Signed-off-by: Amir Goldstein <[email protected]>
Signed-off-by: Jan Kara <[email protected]>
Send events to group if super block mark mask matches the event
and unless the same group has an ignore mask on the vfsmount or
the inode on which the event occurred.
Soon, fanotify backend is going to support super block marks and
fanotify backend only supports path type events.
Signed-off-by: Amir Goldstein <[email protected]>
Signed-off-by: Jan Kara <[email protected]>
Add the infrastructure to attach a mark to a super_block struct
and detach all attached marks when super block is destroyed.
This is going to be used by fanotify backend to setup super block
marks.
Signed-off-by: Amir Goldstein <[email protected]>
Signed-off-by: Jan Kara <[email protected]>
Commit 92183a4289 ("fsnotify: fix ignore mask logic in
send_to_group()") acknoledges the use case of ignoring an event on
an inode mark, because of an ignore mask on a mount mark of the same
group (i.e. I want to get all events on this file, except for the events
that came from that mount).
This change depends on correctly merging the inode marks and mount marks
group lists, so that the mount mark ignore mask would be tested in
send_to_group(). Alas, the merging of the lists did not take into
account the case where event in question is not in the mask of any of
the mount marks.
To fix this, completely remove the tests for inode and mount event masks
from the lists merging code.
Fixes: 92183a4289 ("fsnotify: fix ignore mask logic in send_to_group")
Signed-off-by: Amir Goldstein <[email protected]>
Signed-off-by: Jan Kara <[email protected]>
The implementation of readahead(2) syscall is identical to that of
fadvise64(POSIX_FADV_WILLNEED) with a few exceptions:
1. readahead(2) returns -EINVAL for !mapping->a_ops and fadvise64()
ignores the request and returns 0.
2. fadvise64() checks for integer overflow corner case
3. fadvise64() calls the optional filesystem fadvise() file operation
Unite the two implementations by calling vfs_fadvise() from readahead(2)
syscall. Check the !mapping->a_ops in readahead(2) syscall to preserve
documented syscall ABI behaviour.
Suggested-by: Miklos Szeredi <[email protected]>
Fixes: d1d04ef857 ("ovl: stack file ops")
Signed-off-by: Amir Goldstein <[email protected]>
Signed-off-by: Miklos Szeredi <[email protected]>
This is going to be used by overlayfs and possibly useful
for other filesystems.
Signed-off-by: Amir Goldstein <[email protected]>
Signed-off-by: Miklos Szeredi <[email protected]>
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 <[email protected]>
Fixes: f7c72396d0 ("ovl: add O_DIRECT support")
Signed-off-by: Amir Goldstein <[email protected]>
Signed-off-by: Miklos Szeredi <[email protected]>
nfsd and lockd call vfs_lock_file() to lock/unlock the inode
returned by locks_inode(file).
Many places in nfsd/lockd code use the inode returned by
file_inode(file) for lock manipulation. With Overlayfs, file_inode()
(the underlying inode) is not the same object as locks_inode() (the
overlay inode). This can result in "Leaked POSIX lock" messages
and eventually to a kernel crash as reported by Eddie Horng:
https://marc.info/?l=linux-unionfs&m=153086643202072&w=2
Fix all the call sites in nfsd/lockd that should use locks_inode().
This is a correctness bug that manifested when overlayfs gained
NFS export support in v4.16.
Reported-by: Eddie Horng <[email protected]>
Tested-by: Eddie Horng <[email protected]>
Cc: Jeff Layton <[email protected]>
Fixes: 8383f17488 ("ovl: wire up NFS export operations")
Cc: [email protected]
Signed-off-by: Amir Goldstein <[email protected]>
Signed-off-by: J. Bruce Fields <[email protected]>
Only upper dir can be impure, but if we are in the middle of
iterating a lower real dir, dir could be copied up and marked
impure. We only want the impure cache if we started iterating
a real upper dir to begin with.
Aditya Kali reported that the following reproducer hits the
WARN_ON(!cache->refcount) in ovl_get_cache():
docker run --rm drupal:8.5.4-fpm-alpine \
sh -c 'cd /var/www/html/vendor/symfony && \
chown -R www-data:www-data . && ls -l .'
Reported-by: Aditya Kali <[email protected]>
Tested-by: Aditya Kali <[email protected]>
Fixes: 4edb83bb10 ('ovl: constant d_ino for non-merge dirs')
Cc: <[email protected]> # v4.14
Signed-off-by: Amir Goldstein <[email protected]>
Signed-off-by: Miklos Szeredi <[email protected]>
Factor out helpers fanotify_add_mark() and fanotify_remove_mark()
to reduce duplicated code.
Signed-off-by: Amir Goldstein <[email protected]>
Signed-off-by: Jan Kara <[email protected]>
Use a helper to get the mask from the object (i.e. i_fsnotify_mask)
to generalize code of add/remove inode/vfsmount mark.
Signed-off-by: Amir Goldstein <[email protected]>
Signed-off-by: Jan Kara <[email protected]>
Make the code to attach/detach a connector to object more generic
by letting the fsnotify connector point to an abstract fsnotify_connp_t.
Code that needs to dereference an inode or mount object now uses the
helpers fsnotify_conn_{inode,mount}.
Signed-off-by: Amir Goldstein <[email protected]>
Signed-off-by: Jan Kara <[email protected]>
Instead of passing inode and vfsmount arguments to fsnotify_add_mark()
and its _locked variant, pass an abstract object pointer and the object
type.
The helpers fsnotify_obj_{inode,mount} are added to get the concrete
object pointer from abstract object pointer.
Signed-off-by: Amir Goldstein <[email protected]>
Signed-off-by: Jan Kara <[email protected]>
The object marks manipulation functions fsnotify_destroy_marks()
fsnotify_find_mark() and their helpers take an argument of type
struct fsnotify_mark_connector __rcu ** to dereference the connector
pointer. use a typedef to describe this type for brevity.
Signed-off-by: Amir Goldstein <[email protected]>
Signed-off-by: Jan Kara <[email protected]>
tmpfs uses the helper d_find_alias() to find a dentry from a decoded
inode, but d_find_alias() skips unhashed dentries, so unlinked files
cannot be decoded from a file handle.
This can be reproduced using xfstests test program open_by_handle:
$ open_by handle -c /tmp/testdir
$ open_by_handle -dk /tmp/testdir
open_by_handle(/tmp/testdir/file000000) returned 116 incorrectly on an unlinked open file!
To fix this, if d_find_alias() can't find a hashed alias, call
d_find_any_alias() to return an unhashed one.
Link: http://lkml.kernel.org/r/CAOQ4uxg+qSLP0KwdW+h1tcPqOCQd+_pGZVXiePQB1TXCMBMctQ@mail.gmail.com
Signed-off-by: Amir Goldstein <[email protected]>
Reviewed-by: NeilBrown <[email protected]>
Cc: Hugh Dickins <[email protected]>
Cc: Jeff Layton <[email protected]>
Cc: "J. Bruce Fields" <[email protected]>
Cc: Al Viro <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>
Currently, there is a small window where ovl_obtain_alias() can
race with ovl_instantiate() and create two different overlay inodes
with the same underlying real non-dir non-hardlink inode.
The race requires an adversary to guess the file handle of the
yet to be created upper inode and decode the guessed file handle
after ovl_creat_real(), but before ovl_instantiate().
This race does not affect overlay directory inodes, because those
are decoded via ovl_lookup_real() and not with ovl_obtain_alias().
This patch fixes the race, by using inode_insert5() to add a newly
created inode to cache.
If the newly created inode apears to already exist in cache (hashed
by the same real upper inode), we instantiate the dentry with the old
inode and drop the new inode, instead of silently not hashing the new
inode.
Signed-off-by: Amir Goldstein <[email protected]>
Signed-off-by: Miklos Szeredi <[email protected]>
Also used ovl_create_temp() in ovl_create_index() instead of calling
ovl_do_mkdir() directly, so now all callers of ovl_do_mkdir() are routed
through ovl_create_real(), which paves the way for Al's fix for non-hashed
result from vfs_mkdir().
Signed-off-by: Amir Goldstein <[email protected]>
Signed-off-by: Miklos Szeredi <[email protected]>
Overlayfs should cope with online changes to underlying layer
without crashing the kernel, which is what xfstest overlay/019
checks.
This test may sometimes trigger WARN_ON() in ovl_create_or_link()
when linking an overlay inode that has been changed on underlying
layer.
Remove those WARN_ON() to prevent the stress test from failing.
Signed-off-by: Amir Goldstein <[email protected]>
Signed-off-by: Miklos Szeredi <[email protected]>
David's tree is no longer maintained, so point to my maintained fork.
Add --verify flag to the run example, which enables all latest features
and provides test coverage for constant st_ino/st_dev.
Signed-off-by: Amir Goldstein <[email protected]>
Acked-by: David Howells <[email protected]>
Signed-off-by: Miklos Szeredi <[email protected]>
Before changing the arguments of the functions fsnotify_add_mark()
and fsnotify_add_mark_locked(), convert most callers to use a wrapper.
Signed-off-by: Amir Goldstein <[email protected]>
Signed-off-by: Jan Kara <[email protected]>
Use fsnotify_foreach_obj_type macros to generalize the code that filters
events by marks mask and ignored_mask.
This is going to be used for adding mark of super block object type.
Signed-off-by: Amir Goldstein <[email protected]>
Signed-off-by: Jan Kara <[email protected]>
Use fsnotify_foreach_obj_type macros to generalize the code that filters
events by marks mask and ignored_mask.
This is going to be used for adding mark of super block object type.
Signed-off-by: Amir Goldstein <[email protected]>
Signed-off-by: Jan Kara <[email protected]>
Make some code that handles marks of object types inode and vfsmount
generic, so it can handle other object types.
Introduce fsnotify_foreach_obj_type macro to iterate marks by object type
and fsnotify_iter_{should|set}_report_type macros to set/test report_mask.
This is going to be used for adding mark of another object type
(super block mark).
Signed-off-by: Amir Goldstein <[email protected]>
Signed-off-by: Jan Kara <[email protected]>
Introduce helpers fsnotify_iter_select_report_types() and
fsnotify_iter_next() to abstract the inode/vfsmount marks merged
list iteration.
This is a preparation patch before generalizing mark list
iteration to more mark object types (i.e. super block marks).
Signed-off-by: Amir Goldstein <[email protected]>
Signed-off-by: Jan Kara <[email protected]>
inode_mark and vfsmount_mark arguments are passed to handle_event()
operation as function arguments as well as on iter_info struct.
The difference is that iter_info struct may contain marks that should
not be handled and are represented as NULL arguments to inode_mark or
vfsmount_mark.
Instead of passing the inode_mark and vfsmount_mark arguments, add
a report_mask member to iter_info struct to indicate which marks should
be handled, versus marks that should only be kept alive during user
wait.
This change is going to be used for passing more mark types
with handle_event() (i.e. super block marks).
Signed-off-by: Amir Goldstein <[email protected]>
Signed-off-by: Jan Kara <[email protected]>
An fsnotify_mark_connector is referencing a single type of object
(either inode or vfsmount). Instead of storing a type mask in
connector->flags, store a single type id in connector->type to
identify the type of object.
When a connector object is detached from the object, its type is set
to FSNOTIFY_OBJ_TYPE_DETACHED and this object is not going to be
reused.
The function fsnotify_clear_marks_by_group() is the only place where
type mask was used, so use type flags instead of type id to this
function.
This change is going to be more convenient when adding a new object
type (super block).
Signed-off-by: Amir Goldstein <[email protected]>
Signed-off-by: Jan Kara <[email protected]>
If fs is frozen after mount and before the first file open, the
update of s_last_mounted bypasses freeze protection and prints out
a WARNING splat:
$ mount /vdf
$ fsfreeze -f /vdf
$ cat /vdf/foo
[ 31.578555] WARNING: CPU: 1 PID: 1415 at
fs/ext4/ext4_jbd2.c:53 ext4_journal_check_start+0x48/0x82
[ 31.614016] Call Trace:
[ 31.614997] __ext4_journal_start_sb+0xe4/0x1a4
[ 31.616771] ? ext4_file_open+0xb6/0x189
[ 31.618094] ext4_file_open+0xb6/0x189
If fs is frozen, skip s_last_mounted update.
[backport hint: to apply to stable tree, need to apply also patches
vfs: add the sb_start_intwrite_trylock() helper
ext4: factor out helper ext4_sample_last_mounted()]
Cc: [email protected]
Fixes: bc0b0d6d69 ("ext4: update the s_last_mounted field in the superblock")
Signed-off-by: Amir Goldstein <[email protected]>
Signed-off-by: Theodore Ts'o <[email protected]>
Reviewed-by: Jan Kara <[email protected]>
Needed by ext4 to test frozen fs before updating s_last_mounted.
Signed-off-by: Amir Goldstein <[email protected]>
Signed-off-by: Theodore Ts'o <[email protected]>
Reviewed-by: Jan Kara <[email protected]>
The comment claims that this helper will try not to loose bits, but for
64bit long it looses the high bits before hashing 64bit long into 32bit
int. Use the helper hash_long() to do the right thing for 64bit long.
For 32bit long, there is no change.
All the callers of end_name_hash() either assign the result to
qstr->hash, which is u32 or return the result as an int value (e.g.
full_name_hash()). Change the helper return type to int to conform to
its users.
[ It took me a while to apply this, because my initial reaction to it
was - incorrectly - that it could make for slower code.
After having looked more at it, I take back all my complaints about
the patch, Amir was right and I was mis-reading things or just being
stupid.
I also don't worry too much about the possible performance impact of
this on 64-bit, since most architectures that actually care about
performance end up not using this very much (the dcache code is the
most performance-critical, but the word-at-a-time case uses its own
hashing anyway).
So this ends up being mostly used for filesystems that do their own
degraded hashing (usually because they want a case-insensitive
comparison function).
A _tiny_ worry remains, in that not everybody uses DCACHE_WORD_ACCESS,
and then this potentially makes things more expensive on 64-bit
architectures with slow or lacking multipliers even for the normal
case.
That said, realistically the only such architecture I can think of is
PA-RISC. Nobody really cares about performance on that, it's more of a
"look ma, I've got warts^W an odd machine" platform.
So the patch is fine, and all my initial worries were just misplaced
from not looking at this properly. - Linus ]
Signed-off-by: Amir Goldstein <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>
There is alreay an entry for all the backends, but those entries do
not cover all the fsnotify files.
Signed-off-by: Amir Goldstein <[email protected]>
Signed-off-by: Jan Kara <[email protected]>
The ignore mask logic in send_to_group() does not match the logic
in fanotify_should_send_event(). In the latter, a vfsmount mark ignore
mask precedes an inode mark mask and in the former, it does not.
That difference may cause events to be sent to fanotify backend for no
reason. Fix the logic in send_to_group() to match that of
fanotify_should_send_event().
Signed-off-by: Amir Goldstein <[email protected]>
Signed-off-by: Jan Kara <[email protected]>
With mount option "xino=on", mounter declares that there are enough
free high bits in underlying fs to hold the layer fsid.
If overlayfs does encounter underlying inodes using the high xino
bits reserved for layer fsid, a warning will be emitted and the original
inode number will be used.
The mount option name "xino" goes after a similar meaning mount option
of aufs, but in overlayfs case, the mapping is stateless.
An example for a use case of "xino=on" is when upper/lower is on an xfs
filesystem. xfs uses 64bit inode numbers, but it currently never uses the
upper 8bit for inode numbers exposed via stat(2) and that is not likely to
change in the future without user opting-in for a new xfs feature. The
actual number of unused upper bit is much larger and determined by the xfs
filesystem geometry (64 - agno_log - agblklog - inopblog). That means
that for all practical purpose, there are enough unused bits in xfs
inode numbers for more than OVL_MAX_STACK unique fsid's.
Another use case of "xino=on" is when upper/lower is on tmpfs. tmpfs inode
numbers are allocated sequentially since boot, so they will practially
never use the high inode number bits.
For compatibility with applications that expect 32bit inodes, the feature
can be disabled with "xino=off". The option "xino=auto" automatically
detects underlying filesystem that use 32bit inodes and enables the
feature. The Kconfig option OVERLAY_FS_XINO_AUTO and module parameter of
the same name, determine if the default mode for overlayfs mount is
"xino=auto" or "xino=off".
Signed-off-by: Amir Goldstein <[email protected]>
Signed-off-by: Miklos Szeredi <[email protected]>
When overlay layers are not all on the same fs, but all inode numbers
of underlying fs do not use the high 'xino' bits, overlay st_ino values
are constant and persistent.
In that case, relax non-samefs constraint for consistent d_ino and always
iterate non-merge dir using ovl_fill_real() actor so we can remap lower
inode numbers to unique lower fs range.
Signed-off-by: Amir Goldstein <[email protected]>
Signed-off-by: Miklos Szeredi <[email protected]>
When overlay layers are not all on the same fs, but all inode numbers
of underlying fs do not use the high 'xino' bits, overlay st_ino values
are constant and persistent.
In that case, set i_ino value to the same value as st_ino for nfsd
readdirplus validator.
Signed-off-by: Amir Goldstein <[email protected]>
Signed-off-by: Miklos Szeredi <[email protected]>
On 64bit systems, when overlay layers are not all on the same fs, but
all inode numbers of underlying fs are not using the high bits, use the
high bits to partition the overlay st_ino address space. The high bits
hold the fsid (upper fsid is 0). This way overlay inode numbers are unique
and all inodes use overlay st_dev. Inode numbers are also persistent
for a given layer configuration.
Currently, our only indication for available high ino bits is from a
filesystem that supports file handles and uses the default encode_fh()
operation, which encodes a 32bit inode number.
Signed-off-by: Amir Goldstein <[email protected]>
Signed-off-by: Miklos Szeredi <[email protected]>
Instead of allocating an anonymous bdev per lower layer, allocate
one anonymous bdev per every unique lower fs that is different than
upper fs.
Every unique lower fs is assigned an fsid > 0 and the number of
unique lower fs are stored in ofs->numlowerfs.
The assigned fsid is stored in the lower layer struct and will be
used also for inode number multiplexing.
Signed-off-by: Amir Goldstein <[email protected]>
Signed-off-by: Miklos Szeredi <[email protected]>
A helper for ovl_getattr() to map the values of st_dev and st_ino
according to constant st_ino rules.
Signed-off-by: Amir Goldstein <[email protected]>
Signed-off-by: Miklos Szeredi <[email protected]>
When decoding a lower file handle, we need to check if lower file was
copied up and indexed and if it has a whiteout index, we need to check
if this is an unlinked but open non-dir before returning -ESTALE.
To find out if this is an unlinked but open non-dir we need to lookup
an overlay inode in inode cache by lower inode and that requires decoding
the lower file handle before looking in inode cache.
Before this change, if the lower inode turned out to be a directory, we
may have paid an expensive cost to reconnect that lower directory for
nothing.
After this change, we start by decoding a disconnected lower dentry and
using the lower inode for looking up an overlay inode in inode cache.
If we find overlay inode and dentry in cache, we avoid the index lookup
overhead. If we don't find an overlay inode and dentry in cache, then we
only need to decode a connected lower dentry in case the lower dentry is
a non-indexed directory.
The xfstests group overlay/exportfs tests decoding overlayfs file
handles after drop_caches with different states of the file at encode
and decode time. Overall the tests in the group call ovl_lower_fh_to_d()
89 times to decode a lower file handle.
Before this change, the tests called ovl_get_index_fh() 75 times and
reconnect_one() 61 times.
After this change, the tests call ovl_get_index_fh() 70 times and
reconnect_one() 59 times. The 2 cases where reconnect_one() was avoided
are cases where a non-upper directory file handle was encoded, then the
directory removed and then file handle was decoded.
To demonstrate the affect on decoding file handles with hot inode/dentry
cache, the drop_caches call in the tests was disabled. Without
drop_caches, there are no reconnect_one() calls at all before or after
the change. Before the change, there are 75 calls to ovl_get_index_fh(),
exactly as the case with drop_caches. After the change, there are only
10 calls to ovl_get_index_fh().
Signed-off-by: Amir Goldstein <[email protected]>
Signed-off-by: Miklos Szeredi <[email protected]>
On lookup of non directory, we try to decode the origin file handle
stored in upper inode. The origin file handle is supposed to be decoded
to a disconnected non-dir dentry, which is fine, because we only need
the lower inode of a copy up origin.
However, if the origin file handle somehow turns out to be a directory
we pay the expensive cost of reconnecting the directory dentry, only to
get a mismatch file type and drop the dentry.
Optimize this case by explicitly opting out of reconnecting the dentry.
Opting-out of reconnect is done by passing a NULL acceptable callback
to exportfs_decode_fh().
While the case described above is a strange corner case that does not
really need to be optimized, the API added for this optimization will
be used by a following patch to optimize a more common case of decoding
an overlayfs file handle.
Signed-off-by: Amir Goldstein <[email protected]>
Signed-off-by: Miklos Szeredi <[email protected]>
Rename ovl_encode_fh() to ovl_encode_real_fh() to differentiate from the
exportfs function ovl_encode_inode_fh() and change the latter to
ovl_encode_fh() to match the exportfs method name.
Rename ovl_decode_fh() to ovl_decode_real_fh() for consistency.
Signed-off-by: Amir Goldstein <[email protected]>
Signed-off-by: Miklos Szeredi <[email protected]>
For broken hardlinks, we do not return lower st_ino, so we should
also not return lower pseudo st_dev.
Fixes: a0c5ad307a ("ovl: relax same fs constraint for constant st_ino")
Cc: <[email protected]> #v4.15
Signed-off-by: Amir Goldstein <[email protected]>
Signed-off-by: Miklos Szeredi <[email protected]>
As of now if we encounter an opaque dir while looking for a dentry, we set
d->last=true. This means that there is no need to look further in any of
the lower layers. This works fine as long as there are no redirets or
relative redircts. But what if there is an absolute redirect on the
children dentry of opaque directory. We still need to continue to look into
next lower layer. This patch fixes it.
Here is an example to demonstrate the issue. Say you have following setup.
upper: /redirect (redirect=/a/b/c)
lower1: /a/[b]/c ([b] is opaque) (c has absolute redirect=/a/b/d/)
lower0: /a/b/d/foo
Now "redirect" dir should merge with lower1:/a/b/c/ and lower0:/a/b/d.
Note, despite the fact lower1:/a/[b] is opaque, we need to continue to look
into lower0 because children c has an absolute redirect.
Following is a reproducer.
Watch me make foo disappear:
$ mkdir lower middle upper work work2 merged
$ mkdir lower/origin
$ touch lower/origin/foo
$ mount -t overlay none merged/ \
-olowerdir=lower,upperdir=middle,workdir=work2
$ mkdir merged/pure
$ mv merged/origin merged/pure/redirect
$ umount merged
$ mount -t overlay none merged/ \
-olowerdir=middle:lower,upperdir=upper,workdir=work
$ mv merged/pure/redirect merged/redirect
Now you see foo inside a twice redirected merged dir:
$ ls merged/redirect
foo
$ umount merged
$ mount -t overlay none merged/ \
-olowerdir=middle:lower,upperdir=upper,workdir=work
After mount cycle you don't see foo inside the same dir:
$ ls merged/redirect
During middle layer lookup, the opaqueness of middle/pure is left in
the lookup state and then middle/pure/redirect is wrongly treated as
opaque.
Fixes: 02b69b284c ("ovl: lookup redirects")
Cc: <[email protected]> #v4.10
Signed-off-by: Amir Goldstein <[email protected]>
Signed-off-by: Vivek Goyal <[email protected]>
Signed-off-by: Miklos Szeredi <[email protected]>
Eddie Horng reported that readdir of an overlayfs directory that
was exported via NFSv3 returns entries with d_type set to DT_UNKNOWN.
The reason is that while preparing the response for readdirplus, nfsd
checks inside encode_entryplus_baggage() that a child dentry's inode
number matches the value of d_ino returns by overlayfs readdir iterator.
Because the overlayfs inodes use arbitrary inode numbers that are not
correlated with the values of st_ino/d_ino, NFSv3 falls back to not
encoding d_type. Although this is an allowed behavior, we can fix it for
the case of all overlayfs layers on the same underlying filesystem.
When NFS export is enabled and d_ino is consistent with st_ino
(samefs), set the same value also to i_ino in ovl_fill_inode() for all
overlayfs inodes, nfsd readdirplus sanity checks will pass.
ovl_fill_inode() may be called from ovl_new_inode(), before real inode
was created with ino arg 0. In that case, i_ino will be updated to real
upper inode i_ino on ovl_inode_init() or ovl_inode_update().
Reported-by: Eddie Horng <[email protected]>
Tested-by: Eddie Horng <[email protected]>
Signed-off-by: Amir Goldstein <[email protected]>
Fixes: 8383f17488 ("ovl: wire up NFS export operations")
Cc: <[email protected]> #v4.16
Signed-off-by: Miklos Szeredi <[email protected]>
When event on child inodes are sent to the parent inode mark and
parent inode mark was not marked with FAN_EVENT_ON_CHILD, the event
will not be delivered to the listener process. However, if the same
process also has a mount mark, the event to the parent inode will be
delivered regadless of the mount mark mask.
This behavior is incorrect in the case where the mount mark mask does
not contain the specific event type. For example, the process adds
a mark on a directory with mask FAN_MODIFY (without FAN_EVENT_ON_CHILD)
and a mount mark with mask FAN_CLOSE_NOWRITE (without FAN_ONDIR).
A modify event on a file inside that directory (and inside that mount)
should not create a FAN_MODIFY event, because neither of the marks
requested to get that event on the file.
Fixes: 1968f5eed5 ("fanotify: use both marks when possible")
Cc: stable <[email protected]>
Signed-off-by: Amir Goldstein <[email protected]>
Signed-off-by: Jan Kara <[email protected]>
This change relaxes copy up on encode of merge dir with lower layer > 1
and handles the case of encoding a merge dir with lower layer 1, where an
ancestor is a non-indexed merge dir. In that case, decode of the lower
file handle will not have been possible if the non-indexed ancestor is
redirected before or after encode.
Before encoding a non-upper directory file handle from real layer N, we
need to check if it will be possible to reconnect an overlay dentry from
the real lower decoded dentry. This is done by following the overlay
ancestry up to a "layer N connected" ancestor and verifying that all
parents along the way are "layer N connectable". If an ancestor that is
NOT "layer N connectable" is found, we need to copy up an ancestor, which
is "layer N connectable", thus making that ancestor "layer N connected".
For example:
layer 1: /a
layer 2: /a/b/c
The overlay dentry /a is NOT "layer 2 connectable", because if dir /a is
copied up and renamed, upper dir /a will be indexed by lower dir /a from
layer 1. The dir /a from layer 2 will never be indexed, so the algorithm
in ovl_lookup_real_ancestor() (*) will not be able to lookup a connected
overlay dentry from the connected lower dentry /a/b/c.
To avoid this problem on decode time, we need to copy up an ancestor of
/a/b/c, which is "layer 2 connectable", on encode time. That ancestor is
/a/b. After copy up (and index) of /a/b, it will become "layer 2 connected"
and when the time comes to decode the file handle from lower dentry /a/b/c,
ovl_lookup_real_ancestor() will find the indexed ancestor /a/b and decoding
a connected overlay dentry will be accomplished.
(*) the algorithm in ovl_lookup_real_ancestor() can be improved to lookup
an entry /a in the lower layers above layer N and find the indexed dir /a
from layer 1. If that improvement is made, then the check for "layer N
connected" will need to verify there are no redirects in lower layers above
layer N. In the example above, /a will be "layer 2 connectable". However,
if layer 2 dir /a is a target of a layer 1 redirect, then /a will NOT be
"layer 2 connectable":
layer 1: /A (redirect = /a)
layer 2: /a/b/c
Signed-off-by: Amir Goldstein <[email protected]>
Signed-off-by: Miklos Szeredi <[email protected]>
Commit 31747eda41 ("ovl: hash directory inodes for fsnotify")
fixed an issue of inotify watch on directory that stops getting
events after dropping dentry caches.
A similar issue exists for non-dir non-upper files, for example:
$ mkdir -p lower upper work merged
$ touch lower/foo
$ mount -t overlay -o
lowerdir=lower,workdir=work,upperdir=upper none merged
$ inotifywait merged/foo &
$ echo 2 > /proc/sys/vm/drop_caches
$ cat merged/foo
inotifywait doesn't get the OPEN event, because ovl_lookup() called
from 'cat' allocates a new overlay inode and does not reuse the
watched inode.
Fix this by hashing non-dir overlay inodes by lower real inode in
the following cases that were not hashed before this change:
- A non-upper overlay mount
- A lower non-hardlink when index=off
A helper ovl_hash_bylower() was added to put all the logic and
documentation about which real inode an overlay inode is hashed by
into one place.
The issue dates back to initial version of overlayfs, but this
patch depends on ovl_inode code that was introduced in kernel v4.13.
Cc: <[email protected]> #v4.13
Signed-off-by: Amir Goldstein <[email protected]>
Signed-off-by: Miklos Szeredi <[email protected]>
The time values in stat and inode may differ for overlayfs and stat time
values are the correct ones to use. This is also consistent with the fact
that fill_post_wcc() also stores stat time values.
This means introducing a stat call that could fail, where previously we
were just copying values out of the inode. To be conservative about
changing behavior, we fall back to copying values out of the inode in
the error case. It might be better just to clear fh_pre_saved (though
note the BUG_ON in set_change_info).
Signed-off-by: Amir Goldstein <[email protected]>
Signed-off-by: J. Bruce Fields <[email protected]>
The values of stat->mtime and inode->i_mtime may differ for overlayfs
and stat->mtime is the correct value to use when encoding getattr.
This is also consistent with the fact that other attr times are also
encoded from stat values.
Both callers of lease_get_mtime() already have the value of stat->mtime,
so the only needed change is that lease_get_mtime() will not overwrite
this value with inode->i_mtime in case the inode does not have an
exclusive lease.
Signed-off-by: Amir Goldstein <[email protected]>
Reviewed-by: Jeff Layton <[email protected]>
Signed-off-by: J. Bruce Fields <[email protected]>
Another fix for an issue reported by 0-day robot.
Reported-by: Dan Carpenter <[email protected]>
Fixes: 8ed5eec9d6 ("ovl: encode pure upper file handles")
Signed-off-by: Amir Goldstein <[email protected]>
Signed-off-by: Miklos Szeredi <[email protected]>
A re-factoring patch in NFS export series has passed the wrong argument
to ovl_get_inode() causing a regression in the very recent fix to
fsnotify of overlay merge dir.
The regression has caused merge directory inodes to be hashed by upper
instead of lower real inode, when NFS export and directory indexing is
disabled. That caused an inotify watch to become obsolete after directory
copy up and drop caches.
LTP test inotify07 was improved to catch this regression.
The regression also caused multiple redirect dirs to same origin not to
be detected on lookup with NFS export disabled. An xfstest was added to
cover this case.
Fixes: 0aceb53e73 ("ovl: do not pass overlay dentry to ovl_get_inode()")
Signed-off-by: Amir Goldstein <[email protected]>
Signed-off-by: Miklos Szeredi <[email protected]>
Commit 66f364649d ("xfs: remove if_rdev") moved storing of rdev
value for special inodes to VFS inodes, but forgot to preserve the
value of i_rdev when recycling a reclaimable xfs_inode.
This was detected by xfstest overlay/017 with inodex=on mount option
and xfs base fs. The test does a lookup of overlay chardev and blockdev
right after drop caches.
Overlayfs inodes hold a reference on underlying xfs inodes when mount
option index=on is configured. If drop caches reclaim xfs inodes, before
it relclaims overlayfs inodes, that can sometimes leave a reclaimable xfs
inode and that test hits that case quite often.
When that happens, the xfs inode cache remains broken (zere i_rdev)
until the next cycle mount or drop caches.
Fixes: 66f364649d ("xfs: remove if_rdev")
Signed-off-by: Amir Goldstein <[email protected]>
Reviewed-by: Christoph Hellwig <[email protected]>
Reviewed-by: Darrick J. Wong <[email protected]>
Signed-off-by: Darrick J. Wong <[email protected]>
Now that NFS export operations are implemented, enable overlayfs NFS
export support if the "nfs_export" feature is enabled.
Signed-off-by: Amir Goldstein <[email protected]>
Signed-off-by: Miklos Szeredi <[email protected]>
ovl_lookup_real() in lower layer walks back lower parents to find the
topmost indexed parent. If an indexed ancestor is found before reaching
lower layer root, ovl_lookup_real() is called recursively with upper
layer to walk back from indexed upper to the topmost connected/hashed
upper parent (or up to root).
ovl_lookup_real() in upper layer then walks forward to connect the topmost
upper overlay dir dentry and ovl_lookup_real() in lower layer continues to
walk forward to connect the decoded lower overlay dir dentry.
Signed-off-by: Amir Goldstein <[email protected]>
Signed-off-by: Miklos Szeredi <[email protected]>
Decoding a dir file handle requires walking backward up to layer root and
for lower dir also checking the index to see if any of the parents have
been copied up.
Lookup overlay ancestor dentry in inode/dentry cache by decoded real
parents to shortcut looking up all the way back to layer root.
Signed-off-by: Amir Goldstein <[email protected]>
Signed-off-by: Miklos Szeredi <[email protected]>
Non-indexed upper dirs are encoded as upper file handles. When NFS export
is enabled, hash non-indexed directory inodes by upper inode, so we can
find them in inode cache using the decoded upper inode.
When NFS export is disabled, directories are not indexed on copy up, so
hash non-indexed directory inodes by origin inode, the same hash key
that is used before copy up.
Signed-off-by: Amir Goldstein <[email protected]>
Signed-off-by: Miklos Szeredi <[email protected]>
Similar to decoding a pure upper dir file handle, decoding a pure lower
dir file handle is implemented by looking an overlay dentry of the same
path as the pure lower path and verifying that the overlay dentry's
real lower matches the decoded real lower file handle.
Unlike the case of upper dir file handle, the lookup of overlay path by
lower real path can fail or find a mismatched overlay dentry if any of
the lower parents have been copied up and renamed. To address this case
we will need to check if any of the lower parents are indexed.
Signed-off-by: Amir Goldstein <[email protected]>
Signed-off-by: Miklos Szeredi <[email protected]>
Decoding an indexed dir file handle is done by looking up the file handle
in index dir by name and then decoding the upper dir from the index origin
file handle. The decoded upper path is used to lookup an overlay dentry of
the same path.
Signed-off-by: Amir Goldstein <[email protected]>
Signed-off-by: Miklos Szeredi <[email protected]>
Lookup overlay inode in cache by origin inode, so we can decode a file
handle of an open file even if the index has a whiteout index entry to
mark this overlay inode was unlinked.
Signed-off-by: Amir Goldstein <[email protected]>
Signed-off-by: Miklos Szeredi <[email protected]>
Decoding an indexed non-dir file handle is similar to decoding a lower
non-dir file handle, but additionally, we lookup the file handle in index
dir by name to find the real upper inode.
Signed-off-by: Amir Goldstein <[email protected]>
Signed-off-by: Miklos Szeredi <[email protected]>
Decoding a lower non-dir file handle is done by decoding the lower dentry
from underlying lower fs, finding or allocating an overlay inode that is
hashed by the real lower inode and instantiating an overlay dentry with
that inode.
Signed-off-by: Amir Goldstein <[email protected]>
Signed-off-by: Miklos Szeredi <[email protected]>
For indexed or lower non-dir, encode a non-connectable lower file handle
from origin inode. For indexed or lower dir, when ofs->numlower == 1,
encode a lower file handle from lower dir.
Signed-off-by: Amir Goldstein <[email protected]>
Signed-off-by: Miklos Szeredi <[email protected]>
Decoding a merge dir, whose origin's parent is under a redirected
lower dir is not always possible. As a simple aproximation, we do
not encode lower dir file handles when overlay has multiple lower
layers and origin is below the topmost lower layer.
We should later relax this condition and copy up only the parent
that is under a redirected lower.
Signed-off-by: Amir Goldstein <[email protected]>
Signed-off-by: Miklos Szeredi <[email protected]>
We only need to encode origin if there is a chance that the same object was
encoded pre copy up and then we need to stay consistent with the same
encoding also after copy up.
In case a non-pure upper is not indexed, then it was copied up before NFS
export support was enabled. In that case, we don't need to worry about
staying consistent with pre copy up encoding and we encode an upper file
handle.
This mitigates the problem that with no index, we cannot find an upper
inode from origin inode, so we cannot decode a non-indexed upper from
origin file handle.
Signed-off-by: Amir Goldstein <[email protected]>
Signed-off-by: Miklos Szeredi <[email protected]>
Until this change, we decoded upper file handles by instantiating an
overlay dentry from the real upper dentry. This is sufficient to handle
pure upper files, but insufficient to handle merge/impure dirs.
To that end, if decoded real upper dir is connected and hashed, we
lookup an overlay dentry with the same path as the real upper dir.
If decoded real upper is non-dir, we instantiate a disconnected overlay
dentry as before this change.
Because ovl_fh_to_dentry() returns a connected overlay dir dentry,
exportfs never needs to call get_parent() and get_name() to reconnect an
upper overlay dir. Because connectable non-dir file handles are not
supported, exportfs will not be able to use fh_to_parent() and get_name()
methods to reconnect a disconnected non-dir to its parent. Therefore, the
methods get_parent() and get_name() are implemented just to print out a
sanity warning and the method fh_to_parent() is implemented to warn the
user that using the 'subtree_check' exportfs option is not supported.
An alternative approach could have been to implement instantiating of
an overlay directory inode from origin/index and implement get_parent()
and get_name() by calling into underlying fs operations and them
instantiating the overlay parent dir.
The reasons for not choosing the get_parent() approach were:
- Obtaining a disconnected overlay dir dentry would requires a
delicate re-factoring of ovl_lookup() to get a dentry with overlay
parent info. It was preferred to avoid doing that re-factoring unless
it was proven worthy.
- Going down the path of disconnected dir would mean that the (non
trivial) code path of d_splice_alias() could be traveled and that
meant writing more tests and introduces race cases that are very hard
to hit on purpose. Taking the path of connecting overlay dentry by
forward lookup is therefore the safe and boring way to avoid surprises.
The culprits of the chosen "connected overlay dentry" approach:
- We need to take special care to rename of ancestors while connecting
the overlay dentry by real dentry path. These subtleties are usually
handled by generic exportfs and VFS code.
- In a hypothetical workload, we could end up in a loop trying to connect,
interrupted by rename and restarting connect forever.
Signed-off-by: Amir Goldstein <[email protected]>
Signed-off-by: Miklos Szeredi <[email protected]>
Decoding an upper file handle is done by decoding the upper dentry from
underlying upper fs, finding or allocating an overlay inode that is
hashed by the real upper inode and instantiating an overlay dentry with
that inode.
Signed-off-by: Amir Goldstein <[email protected]>
Signed-off-by: Miklos Szeredi <[email protected]>
Amir Goldstein