afs: Fix dynamic lookup to fail on cell lookup failure

When a process tries to access an entry in /afs, normally what happens is
that an automount dentry is created by ->lookup() and then triggered, which
jumps through the ->d_automount() op.  Currently, afs_dynroot_lookup() does
not do cell DNS lookup, leaving that to afs_d_automount() to perform -
however, it is possible to use access() or stat() on the automount point,
which will always return successfully, have briefly created an afs_cell
record if one did not already exist.

This means that something like:

        test -d "/afs/.west" && echo Directory exists

will print "Directory exists" even though no such cell is configured.  This
breaks the "west" python module available on PIP as it expects this access
to fail.

Now, it could be possible to make afs_dynroot_lookup() perform the DNS[*]
lookup, but that would make "ls --color /afs" do this for each cell in /afs
that is listed but not yet probed.  kafs-client, probably wrongly, preloads
the entire cell database and all the known cells are then listed in /afs -
and doing ls /afs would be very, very slow, especially if any cell supplied
addresses but was wholly inaccessible.

 [*] When I say "DNS", actually read getaddrinfo(), which could use any one
     of a host of mechanisms.  Could also use static configuration.

To fix this, make the following changes:

 (1) Create an enum to specify the origination point of a call to
     afs_lookup_cell() and pass this value into that function in place of
     the "excl" parameter (which can be derived from it).  There are six
     points of origination:

        - Cell preload through /proc/net/afs/cells
        - Root cell config through /proc/net/afs/rootcell
        - Lookup in dynamic root
        - Automount trigger
        - Direct mount with mount() syscall
        - Alias check where YFS tells us the cell name is different

 (2) Add an extra state into the afs_cell state machine to indicate a cell
     that's been initialised, but not yet looked up.  This is separate from
     one that can be considered active and has been looked up at least
     once.

 (3) Make afs_lookup_cell() vary its behaviour more, depending on where it
     was called from:

     If called from preload or root cell config, DNS lookup will not happen
     until we definitely want to use the cell (dynroot mount, automount,
     direct mount or alias check).  The cell will appear in /afs but stat()
     won't trigger DNS lookup.

     If the cell already exists, dynroot will not wait for the DNS lookup
     to complete.  If the cell did not already exist, dynroot will wait.

     If called from automount, direct mount or alias check, it will wait
     for the DNS lookup to complete.

 (4) Make afs_lookup_cell() return an error if lookup failed in one way or
     another.  We try to return -ENOENT if the DNS says the cell does not
     exist and -EDESTADDRREQ if we couldn't access the DNS.

Reported-by: Markus Suvanto <markus.suvanto@gmail.com>
Closes: https://bugzilla.kernel.org/show_bug.cgi?id=220685
Signed-off-by: David Howells <dhowells@redhat.com>
Link: https://patch.msgid.link/1784747.1761158912@warthog.procyon.org.uk
Fixes: 1d0b929fc070 ("afs: Change dynroot to create contents on demand")
Tested-by: Markus Suvanto <markus.suvanto@gmail.com>
cc: Marc Dionne <marc.dionne@auristor.com>
cc: linux-afs@lists.infradead.org
Signed-off-by: Christian Brauner <brauner@kernel.org>

0 files changed, 0 insertions, 0 deletions