If trying to unlock a v2-encrypted directory fails because the kernel
lacks support for v2 policies, show a better error message.  This can
happen if someone downgrades their kernel or tries to access encrypted
directories on removable storage from a computer with an older kernel.

Detecting this case is difficult since all we have to go with is EACCES
when opening the directory.  Implement a heuristic where if get EACCES,
we actually have read access to the directory, and the kernel doesn't
support v2 policies, we show the improved error message.

Before:

  # fscrypt unlock dir
  [ERROR] fscrypt unlock: open dir: permission denied

After:

  # fscrypt unlock dir
  [ERROR] fscrypt unlock: open dir: permission denied

  This may be caused by the directory using a v2 encryption policy and
  the current kernel not supporting it. If indeed the case, then this
  directory can only be used on kernel v5.4 and later. You can create
  directories accessible on older kernels by changing policy_version to
  1 in /etc/fscrypt.conf.

canonicalizePath() is now only used by an error path in
getMountFromLink(), which we can make use getDeviceName() instead.

Currently, fscrypt treats bind mounts as separate filesystems.  This is
broken because fscrypt will look for a directory's encryption policy in
different places depending on which mount it's accessed through.  This
forces users to create an fscrypt metadata directory at every bind
mount, and to copy fscrypt metadata around between mounts.

Fix this by storing fscrypt metadata only at the root of the filesystem.

To accomplish this:

- Make mountsByDevice store only a single Mount per filesystem, rather
  than multiple.  For this Mount, choose a mount of the full filesystem
  if available, preferably a read-write mount.  If the filesystem has
  only bind mounts, store a nil entry in mountsByDevice so we can show a
  proper error message later.

- Change FindMount() and GetMount() to look up the Mount by device
  number rather than by path, so that they don't return different Mounts
  depending on which path is used.

- Change AllFilesystems() to not return bind mounts.

- Due to the above changes, the mountsByPath map is no longer needed
  outside of loadMountInfo().  So make it a local variable there.

Resolves https://github.com/google/fscrypt/issues/59

A root filesystem mounted via the kernel command line always has a
source of "/dev/root", which isn't a real device node.  This makes
fscrypt think this filesystem doesn't have a source device, which breaks
creating login passphrase-protected directories on other filesystems:

    fscrypt encrypt: filesystem /: no device for mount "/": system error: cannot create filesystem link

This also makes 'fscrypt status' show a blank source device:

    MOUNTPOINT  DEVICE          FILESYSTEM  ENCRYPTION     FSCRYPT
    /                           ext4        supported      Yes

To fix this case, update loadMountInfo() to map the device number to the
device name via sysfs rather than use the mount source field.

Add a utility type and functions for handling device numbers.

If the user has set a restrictive umask, e.g. 0077, then
/etc/fscrypt.conf would be created without the world-readable bit set.
Fix it by overriding the umask when creating the file.

Resolves https://github.com/google/fscrypt/issues/151

These were found by a combination of manual review and a custom script
that checks for common errors.

Also removed an outdated sentence from the comment for setupBefore().

This commit changes the error handing for the crypto, filesystem,
metadata, pam, and util packages to use the error handling library
github.com/pkg/errors. This means elimination of the FSError type, an
increased use of wrapping errors (as opposed to logging), switching
on the Cause() of an error (as opposed to its value), and improving our
integration tests involving TEST_FILESYSTEM_ROOT.

This commit also fixes a few bugs with the keyring code to ensure that
our {Find|Remove|Insert}PolicyKey functions are always operating on the
same keyring. The check for filesystem support has been moved from the
filesystem package to the metadata package. Finally, the API for the
filesystem package has been slightly modified:
	* filesystem.AllFilesystems() now returns all the filesystems in
	  sorted order
	* certain path methods are now public

O_SYNC is also removed for writing the metadata. We don't get that much
from syncing the metadata, as the actual file data could also be
corrupted by and IO error. The sync operation is also occasionally very
slow (~3 seconds) and can be unfriendly to battery life.

Change-Id: I392c2655141714b16dfdbc84ac09780072be2cf0

This commit adds in the filesystem subpackage. The goal of this package
is to provide and interface for adding to and removing from the metadata
storage for a given filesystem. This is primarily done in filesystem.go.
To facilitate this functionality, mountpoint.go exposes an interface for
querying the system about the current mounted filesystems and their
information. Note that this operation is done with a lazy loading
mechanism.

To refer to other filesystems, we use link files that can be parsed by
libblkid. The README is also updated to account for this new dependancy.

This package uses the FSError type under the hood so that error messages
will include the filesystem name, but callers can still check for
specific error instances.

Change-Id: I74fe4e84b8e3a5b73f1337c35307ffe0bf7cdea9