All users now just use '->iterate_shared()', which only takes the
directory inode lock for reading.

Filesystems that never got convered to shared mode now instead use a
wrapper that drops the lock, re-takes it in write mode, calls the old
function, and then downgrades the lock back to read mode.

This way the VFS layer and other callers no longer need to care about
filesystems that never got converted to the modern era.

The filesystems that use the new wrapper are ceph, coda, exfat, jfs,
ntfs, ocfs2, overlayfs, and vboxsf.

Honestly, several of them look like they really could just iterate their
directories in shared mode and skip the wrapper entirely, but the point
of this change is to not change semantics or fix filesystems that
haven't been fixed in the last 7+ years, but to finally get rid of the
dual iterators.

Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Signed-off-by: Christian Brauner <brauner@kernel.org>

Commit db1d1e8b9867 ("IMA: use vfs_getattr_nosec to get the i_version")
partially closed an IMA integrity issue when directly modifying a file
on the lower filesystem.  If the overlay file is first opened by a user
and later the lower backing file is modified by root, but the extended
attribute is NOT updated, the signature validation succeeds with the old
original signature.

Update the super_block s_iflags to SB_I_IMA_UNVERIFIABLE_SIGNATURE to
force signature reevaluation on every file access until a fine grained
solution can be found.

Signed-off-by: Eric Snowberg <eric.snowberg@oracle.com>
Signed-off-by: Mimi Zohar <zohar@linux.ibm.com>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

While initially I thought that we couldn't move all new mount api
handling into params.{c,h} it turns out it is possible. So this just
moves a good chunk of code out of super.c and into params.{c,h}.

Signed-off-by: Christian Brauner <brauner@kernel.org>
Signed-off-by: Amir Goldstein <amir73il@gmail.com>

Pull overlayfs update from Amir Goldstein:

 - fix two NULL pointer deref bugs (Zhihao Cheng)

 - add support for "data-only" lower layers destined to be used by
   composefs

 - port overlayfs to the new mount api (Christian Brauner)

* tag 'ovl-update-6.5' of git://git.kernel.org/pub/scm/linux/kernel/git/overlayfs/vfs: (26 commits)
  ovl: add Amir as co-maintainer
  ovl: reserve ability to reconfigure mount options with new mount api
  ovl: modify layer parameter parsing
  ovl: port to new mount api
  ovl: factor out ovl_parse_options() helper
  ovl: store enum redirect_mode in config instead of a string
  ovl: pass ovl_fs to xino helpers
  ovl: clarify ovl_get_root() semantics
  ovl: negate the ofs->share_whiteout boolean
  ovl: check type and offset of struct vfsmount in ovl_entry
  ovl: implement lazy lookup of lowerdata in data-only layers
  ovl: prepare for lazy lookup of lowerdata inode
  ovl: prepare to store lowerdata redirect for lazy lowerdata lookup
  ovl: implement lookup in data-only layers
  ovl: introduce data-only lower layers
  ovl: remove unneeded goto instructions
  ovl: deduplicate lowerdata and lowerstack[]
  ovl: deduplicate lowerpath and lowerstack[]
  ovl: move ovl_entry into ovl_inode
  ovl: factor out ovl_free_entry() and ovl_stack_*() helpers
  ...

Pull splice updates from Jens Axboe:
 "This kills off ITER_PIPE to avoid a race between truncate,
  iov_iter_revert() on the pipe and an as-yet incomplete DMA to a bio
  with unpinned/unref'ed pages from an O_DIRECT splice read. This causes
  memory corruption.

  Instead, we either use (a) filemap_splice_read(), which invokes the
  buffered file reading code and splices from the pagecache into the
  pipe; (b) copy_splice_read(), which bulk-allocates a buffer, reads
  into it and then pushes the filled pages into the pipe; or (c) handle
  it in filesystem-specific code.

  Summary:

   - Rename direct_splice_read() to copy_splice_read()

   - Simplify the calculations for the number of pages to be reclaimed
     in copy_splice_read()

   - Turn do_splice_to() into a helper, vfs_splice_read(), so that it
     can be used by overlayfs and coda to perform the checks on the
     lower fs

   - Make vfs_splice_read() jump to copy_splice_read() to handle
     direct-I/O and DAX

   - Provide shmem with its own splice_read to handle non-existent pages
     in the pagecache. We don't want a ->read_folio() as we don't want
     to populate holes, but filemap_get_pages() requires it

   - Provide overlayfs with its own splice_read to call down to a lower
     layer as overlayfs doesn't provide ->read_folio()

   - Provide coda with its own splice_read to call down to a lower layer
     as coda doesn't provide ->read_folio()

   - Direct ->splice_read to copy_splice_read() in tty, procfs, kernfs
     and random files as they just copy to the output buffer and don't
     splice pages

   - Provide wrappers for afs, ceph, ecryptfs, ext4, f2fs, nfs, ntfs3,
     ocfs2, orangefs, xfs and zonefs to do locking and/or revalidation

   - Make cifs use filemap_splice_read()

   - Replace pointers to generic_file_splice_read() with pointers to
     filemap_splice_read() as DIO and DAX are handled in the caller;
     filesystems can still provide their own alternate ->splice_read()
     op

   - Remove generic_file_splice_read()

   - Remove ITER_PIPE and its paraphernalia as generic_file_splice_read
     was the only user"

* tag 'for-6.5/splice-2023-06-23' of git://git.kernel.dk/linux: (31 commits)
  splice: kdoc for filemap_splice_read() and copy_splice_read()
  iov_iter: Kill ITER_PIPE
  splice: Remove generic_file_splice_read()
  splice: Use filemap_splice_read() instead of generic_file_splice_read()
  cifs: Use filemap_splice_read()
  trace: Convert trace/seq to use copy_splice_read()
  zonefs: Provide a splice-read wrapper
  xfs: Provide a splice-read wrapper
  orangefs: Provide a splice-read wrapper
  ocfs2: Provide a splice-read wrapper
  ntfs3: Provide a splice-read wrapper
  nfs: Provide a splice-read wrapper
  f2fs: Provide a splice-read wrapper
  ext4: Provide a splice-read wrapper
  ecryptfs: Provide a splice-read wrapper
  ceph: Provide a splice-read wrapper
  afs: Provide a splice-read wrapper
  9p: Add splice_read wrapper
  net: Make sock_splice_read() use copy_splice_read() by default
  tty, proc, kernfs, random: Use copy_splice_read()
  ...

Using the old mount api to remount an overlayfs superblock via
mount(MS_REMOUNT) all mount options will be silently ignored. For
example, if you create an overlayfs mount:

        mount -t overlay overlay -o lowerdir=/mnt/a:/mnt/b,upperdir=/mnt/upper,workdir=/mnt/work /mnt/merged

and then issue a remount via:

        # force mount(8) to use mount(2)
        export LIBMOUNT_FORCE_MOUNT2=always
        mount -t overlay overlay -o remount,WOOTWOOT,lowerdir=/DOESNT-EXIST /mnt/merged

with completely nonsensical mount options whatsoever it will succeed
nonetheless. This prevents us from every changing any mount options we
might introduce in the future that could reasonably be changed during a
remount.

We don't need to carry this issue into the new mount api port. Similar
to FUSE we can use the fs_context::oldapi member to figure out that this
is a request coming through the legacy mount api. If we detect it we
continue silently ignoring all mount options.

But for the new mount api we simply report that mount options cannot
currently be changed. This will allow us to potentially alter mount
properties for new or even old properties. It any case, silently
ignoring everything is not something new apis should do.

Signed-off-by: Christian Brauner <brauner@kernel.org>
Signed-off-by: Amir Goldstein <amir73il@gmail.com>

We ran into issues where mount(8) passed multiple lower layers as one
big string through fsconfig(). But the fsconfig() FSCONFIG_SET_STRING
option is limited to 256 bytes in strndup_user(). While this would be
fixable by extending the fsconfig() buffer I'd rather encourage users to
append layers via multiple fsconfig() calls as the interface allows
nicely for this. This has also been requested as a feature before.

With this port to the new mount api the following will be possible:

        fsconfig(fs_fd, FSCONFIG_SET_STRING, "lowerdir", "/lower1", 0);

        /* set upper layer */
        fsconfig(fs_fd, FSCONFIG_SET_STRING, "upperdir", "/upper", 0);

        /* append "/lower2", "/lower3", and "/lower4" */
        fsconfig(fs_fd, FSCONFIG_SET_STRING, "lowerdir", ":/lower2:/lower3:/lower4", 0);

        /* turn index feature on */
        fsconfig(fs_fd, FSCONFIG_SET_STRING, "index", "on", 0);

        /* append "/lower5" */
        fsconfig(fs_fd, FSCONFIG_SET_STRING, "lowerdir", ":/lower5", 0);

Specifying ':' would have been rejected so this isn't a regression. And
we can't simply use "lowerdir=/lower" to append on top of existing
layers as "lowerdir=/lower,lowerdir=/other-lower" would make
"/other-lower" the only lower layer so we'd break uapi if we changed
this. So the ':' prefix seems a good compromise.

Users can choose to specify multiple layers at once or individual
layers. A layer is appended if it starts with ":". This requires that
the user has already added at least one layer before. If lowerdir is
specified again without a leading ":" then all previous layers are
dropped and replaced with the new layers. If lowerdir is specified and
empty than all layers are simply dropped.

An additional change is that overlayfs will now parse and resolve layers
right when they are specified in fsconfig() instead of deferring until
super block creation. This allows users to receive early errors.

It also allows users to actually use up to 500 layers something which
was theoretically possible but ended up not working due to the mount
option string passed via mount(2) being too large.

This also allows a more privileged process to set config options for a
lesser privileged process as the creds for fsconfig() and the creds for
fsopen() can differ. We could restrict that they match by enforcing that
the creds of fsopen() and fsconfig() match but I don't see why that
needs to be the case and allows for a good delegation mechanism.

Plus, in the future it means we're able to extend overlayfs mount
options and allow users to specify layers via file descriptors instead
of paths:

        fsconfig(FSCONFIG_SET_PATH{_EMPTY}, "lowerdir", "lower1", dirfd);

        /* append */
        fsconfig(FSCONFIG_SET_PATH{_EMPTY}, "lowerdir", "lower2", dirfd);

        /* append */
        fsconfig(FSCONFIG_SET_PATH{_EMPTY}, "lowerdir", "lower3", dirfd);

        /* clear all layers specified until now */
        fsconfig(FSCONFIG_SET_STRING, "lowerdir", NULL, 0);

This would be especially nice if users create an overlayfs mount on top
of idmapped layers or just in general private mounts created via
open_tree(OPEN_TREE_CLONE). Those mounts would then never have to appear
anywhere in the filesystem. But for now just do the minimal thing.

We should probably aim to move more validation into ovl_fs_parse_param()
so users get errors before fsconfig(FSCONFIG_CMD_CREATE). But that can
be done in additional patches later.

This is now also rebased on top of the lazy lowerdata lookup which
allows the specificatin of data only layers using the new "::" syntax.

The rules are simple. A data only layers cannot be followed by any
regular layers and data layers must be preceeded by at least one regular
layer.

Parsing the lowerdir mount option must change because of this. The
original patchset used the old lowerdir parsing function to split a
lowerdir mount option string such as:

        lowerdir=/lower1:/lower2::/lower3::/lower4

simply replacing each non-escaped ":" by "\0". So sequences of
non-escaped ":" were counted as layers. For example, the previous
lowerdir mount option above would've counted 6 layers instead of 4 and a
lowerdir mount option such as:

        lowerdir="/lower1:/lower2::/lower3::/lower4:::::::::::::::::::::::::::"

would be counted as 33 layers. Other than being ugly this didn't matter
much because kern_path() would reject the first "\0" layer. However,
this overcounting of layers becomes problematic when we base allocations
on it where we very much only want to allocate space for 4 layers
instead of 33.

So the new parsing function rejects non-escaped sequences of colons
other than ":" and "::" immediately instead of relying on kern_path().

Link: https://github.com/util-linux/util-linux/issues/2287
Link: https://github.com/util-linux/util-linux/issues/1992
Link: https://bugs.archlinux.org/task/78702
Link: https://lore.kernel.org/linux-unionfs/20230530-klagen-zudem-32c0908c2108@brauner
Signed-off-by: Christian Brauner <brauner@kernel.org>
Signed-off-by: Amir Goldstein <amir73il@gmail.com>

Overlayfs creates the real underlying files with fake f_path, whose
f_inode is on the underlying fs and f_path on overlayfs.

Those real files were open with FMODE_NONOTIFY, because fsnotify code was
not prapared to handle fsnotify hooks on files with fake path correctly
and fanotify would report unexpected event->fd with fake overlayfs path,
when the underlying fs was being watched.

Teach fsnotify to handle events on the real files, and do not set real
files to FMODE_NONOTIFY to allow operations on real file (e.g. open,
access, modify, close) to generate async and permission events.

Because fsnotify does not have notifications on address space
operations, we do not need to worry about ->vm_file not reporting
events to a watched overlayfs when users are accessing a mapped
overlayfs file.

Acked-by: Jan Kara <jack@suse.cz>
Signed-off-by: Amir Goldstein <amir73il@gmail.com>
Message-Id: <20230615112229.2143178-6-amir73il@gmail.com>
Signed-off-by: Christian Brauner <brauner@kernel.org>

Overlayfs uses open_with_fake_path() to allocate internal kernel files,
with a "fake" path - whose f_path is not on the same fs as f_inode.

Allocate a container struct backing_file for those internal files, that
is used to hold the "fake" ovl path along with the real path.

backing_file_real_path() can be used to access the stored real path.

Signed-off-by: Amir Goldstein <amir73il@gmail.com>
Message-Id: <20230615112229.2143178-5-amir73il@gmail.com>
Signed-off-by: Christian Brauner <brauner@kernel.org>

Overlayfs and cachefiles use vfs_open_tmpfile() to open a tmpfile
without accounting for nr_files.

Rename this helper to kernel_tmpfile_open() to better reflect this
helper is used for kernel internal users.

Signed-off-by: Amir Goldstein <amir73il@gmail.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Message-Id: <20230615112229.2143178-2-amir73il@gmail.com>
Signed-off-by: Christian Brauner <brauner@kernel.org>