This function has to ensure it doesn't follow a symlink that could be used
to escape the virtfs directory. This could be easily achieved if fchmodat()
on linux honored the AT_SYMLINK_NOFOLLOW flag as described in POSIX, but
it doesn't. There was a tentative to implement a new fchmodat2() syscall
with the correct semantics:

https://patchwork.kernel.org/patch/9596301/

but it didn't gain much momentum. Also it was suggested to look at an O_PATH
based solution in the first place.

The current implementation covers most use-cases, but it notably fails if:
- the target path has access rights equal to 0000 (openat() returns EPERM),
  => once you've done chmod(0000) on a file, you can never chmod() again
- the target path is UNIX domain socket (openat() returns ENXIO)
  => bind() of UNIX domain sockets fails if the file is on 9pfs

The solution is to use O_PATH: openat() now succeeds in both cases, and we
can ensure the path isn't a symlink with fstat(). The associated entry in
"/proc/self/fd" can hence be safely passed to the regular chmod() syscall.

The previous behavior is kept for older systems that don't have O_PATH.
Signed-off-by: NGreg Kurz <groug@kaod.org>
Reviewed-by: NEric Blake <eblake@redhat.com>
Tested-by: NZhi Yong Wu <zhiyong.wu@ucloud.cn>
Acked-by: NPhilippe Mathieu-Daudé <f4bug@amsat.org>

In mapped security modes, files are created with very restrictive
permissions (600 for files and 700 for directories). This makes
file sharing between virtual machines and users on the host rather
complicated. Imagine eg. a group of users that need to access data
produced by processes on a virtual machine. Giving those users access
to the data will be difficult since the group access mode is always 0.

This patch makes the default mode for both files and directories
configurable. Existing setups that don't know about the new parameters
keep using the current secure behavior.
Signed-off-by: NTobias Schramm <tobleminer@gmail.com>
Signed-off-by: NGreg Kurz <groug@kaod.org>

Commit a0e640a8 introduced a path processing error.
Pass fstatat the dirpath based path component instead
of the entire path.
Signed-off-by: NBruce Rogers <brogers@suse.com>
Signed-off-by: NGreg Kurz <groug@kaod.org>

When using the mapped-file security, credentials are stored in a metadata
directory located in the parent directory. This is okay for all paths with
the notable exception of the root path, since we don't want and probably
can't create a metadata directory above the virtfs directory on the host.

This patch introduces a dedicated metadata file, sitting in the virtfs root
for this purpose. It relies on the fact that the "." name necessarily refers
to the virtfs root.

As for the metadata directory, we don't want the client to see this file.
The current code only cares for readdir() but there are many other places
to fix actually. The filtering logic is hence put in a separate function.

Before:

# ls -ld
drwxr-xr-x. 3 greg greg 4096 May  5 12:49 .
# chown root.root .
chown: changing ownership of '.': Is a directory
# ls -ld
drwxr-xr-x. 3 greg greg 4096 May  5 12:49 .

After:

# ls -ld
drwxr-xr-x. 3 greg greg 4096 May  5 12:49 .
# chown root.root .
# ls -ld
drwxr-xr-x. 3 root root 4096 May  5 12:50 .

and from the host:

ls -al .virtfs_metadata_root
-rwx------. 1 greg greg 26 May  5 12:50 .virtfs_metadata_root
$ cat .virtfs_metadata_root
virtfs.uid=0
virtfs.gid=0
Reported-by: NLeo Gaspard <leo@gaspard.io>
Signed-off-by: NGreg Kurz <groug@kaod.org>
Reviewed-by: NEric Blake <eblake@redhat.com>
Tested-by: NLeo Gaspard <leo@gaspard.io>
[groug: work around a patchew false positive in
        local_set_mapped_file_attrat()]

The logic to open a path currently sits between local_open_nofollow() and
the relative_openat_nofollow() helper, which has no other user.

For the sake of clarity, this patch moves all the code of the helper into
its unique caller. While here we also:
- drop the code to skip leading "/" because the backend isn't supposed to
  pass anything but relative paths without consecutive slashes. The assert()
  is kept because we really don't want a buggy backend to pass an absolute
  path to openat().
- use strchrnul() to get a simpler code. This is ok since virtfs is for
  linux+glibc hosts only.
- don't dup() the initial directory and add an assert() to ensure we don't
  return the global mountfd to the caller. BTW, this would mean that the
  caller passed an empty path, which isn't supposed to happen either.
Signed-off-by: NGreg Kurz <groug@kaod.org>
Reviewed-by: NEric Blake <eblake@redhat.com>
[groug: fixed typos in changelog]

When using the mapped-file security mode, the creds of a path /foo/bar
are stored in the /foo/.virtfs_metadata/bar file. This is okay for all
paths unless they end with '.' or '..', because we cannot create the
corresponding file in the metadata directory.

This patch ensures that '.' and '..' are resolved in all paths.

The core code only passes path elements (no '/') to the backend, with
the notable exception of the '/' path, which refers to the virtfs root.
This patch preserves the current behavior of converting it to '.' so
that it can be passed to "*at()" syscalls ('/' would mean the host root).
Signed-off-by: NGreg Kurz <groug@kaod.org>
Reviewed-by: NEric Blake <eblake@redhat.com>

When trying to remove a file from a directory, both created in non-mapped
mode, the file remains and EBADF is returned to the guest.

This is a regression introduced by commit "df4938a6 9pfs: local:
unlinkat: don't follow symlinks" when fixing CVE-2016-9602. It changed the
way we unlink the metadata file from

    ret = remove("$dir/.virtfs_metadata/$name");
    if (ret < 0 && errno != ENOENT) {
         /* Error out */
    }
    /* Ignore absence of metadata */

to

    fd = openat("$dir/.virtfs_metadata")
    unlinkat(fd, "$name")
    if (ret < 0 && errno != ENOENT) {
         /* Error out */
    }
    /* Ignore absence of metadata */

If $dir was created in non-mapped mode, openat() fails with ENOENT and
we pass -1 to unlinkat(), which fails in turn with EBADF.

We just need to check the return of openat() and ignore ENOENT, in order
to restore the behaviour we had with remove().
Signed-off-by: NGreg Kurz <groug@kaod.org>
Reviewed-by: NEric Blake <eblake@redhat.com>
[groug: rewrote the comments as suggested by Eric]

When using the mapped-file security mode, we shouldn't let the client mess
with the metadata. The current code already tries to hide the metadata dir
from the client by skipping it in local_readdir(). But the client can still
access or modify it through several other operations. This can be used to
escalate privileges in the guest.

Affected backend operations are:
- local_mknod()
- local_mkdir()
- local_open2()
- local_symlink()
- local_link()
- local_unlinkat()
- local_renameat()
- local_rename()
- local_name_to_path()

Other operations are safe because they are only passed a fid path, which
is computed internally in local_name_to_path().

This patch converts all the functions listed above to fail and return
EINVAL when being passed the name of the metadata dir. This may look
like a poor choice for errno, but there's no such thing as an illegal
path name on Linux and I could not think of anything better.

This fixes CVE-2017-7493.
Reported-by: NLeo Gaspard <leo@gaspard.io>
Signed-off-by: NGreg Kurz <groug@kaod.org>
Reviewed-by: NEric Blake <eblake@redhat.com>

The local backend was recently converted to using "at*()" syscalls in order
to ensure all accesses happen below the shared directory. This requires that
we only pass relative paths, otherwise the dirfd argument to the "at*()"
syscalls is ignored and the path is treated as an absolute path in the host.
This is actually the case for paths in all fids, with the notable exception
of the root fid, whose path is "/". This causes the following backend ops to
act on the "/" directory of the host instead of the virtfs shared directory
when the export root is involved:
- lstat
- chmod
- chown
- utimensat

ie, chmod /9p_mount_point in the guest will be converted to chmod / in the
host for example. This could cause security issues with a privileged QEMU.

All "*at()" syscalls are being passed an open file descriptor. In the case
of the export root, this file descriptor points to the path in the host that
was passed to -fsdev.

The fix is thus as simple as changing the path of the export root fid to be
"." instead of "/".

This is CVE-2017-7471.

Cc: qemu-stable@nongnu.org
Reported-by: NLéo Gaspard <leo@gaspard.io>
Signed-off-by: NGreg Kurz <groug@kaod.org>
Reviewed-by: NEric Blake <eblake@redhat.com>
Signed-off-by: NPeter Maydell <peter.maydell@linaro.org>

We should pass O_NOFOLLOW otherwise openat() will follow symlinks and make
QEMU vulnerable.

While here, we also fix local_unlinkat_common() to use openat_dir() for
the same reasons (it was a leftover in the original patchset actually).

This fixes CVE-2016-9602.
Signed-off-by: NGreg Kurz <groug@kaod.org>
Reviewed-by: NDaniel P. Berrange <berrange@redhat.com>
Reviewed-by: NEric Blake <eblake@redhat.com>

The name argument can never be an empty string, and dirfd always point to
the containing directory of the file name. AT_EMPTY_PATH is hence useless
here. Also it breaks build with glibc version 2.13 and older.

It is actually an oversight of a previous tentative patch to implement this
function. We can safely drop it.
Reported-by: NMark Cave-Ayland <mark.cave-ayland@ilande.co.uk>
Signed-off-by: NGreg Kurz <groug@kaod.org>
Tested-by: NMark Cave-Ayland <mark.cave-ayland@ilande.co.uk>
Reviewed-by: NEric Blake <eblake@redhat.com>

If we cannot open the given path, we can return right away instead of
passing -1 to fstatfs() and close(). This will make Coverity happy.

(Coverity issue CID1371729)
Signed-off-by: NGreg Kurz <groug@kaod.org>
Reviewed-by: NDaniel P. berrange <berrange@redhat.com>
Reviewed-by: NEric Blake <eblake@redhat.com>
Reviewed-by: NPhilippe Mathieu-Daudé <f4bug@amsat.org>

Coverity issue CID1371731
Signed-off-by: NGreg Kurz <groug@kaod.org>
Reviewed-by: NDaniel P. Berrange <berrange@redhat.com>
Reviewed-by: NPhilippe Mathieu-Daudé <f4bug@amsat.org>

This was spotted by Coverity as a fd leak. This is certainly true, but also
local_remove() would always return without doing anything, unless the fd is
zero, which is very unlikely.

(Coverity issue CID1371732)
Signed-off-by: NGreg Kurz <groug@kaod.org>
Reviewed-by: NEric Blake <eblake@redhat.com>

Now that the all callbacks have been converted to use "at" syscalls, we
can drop this code.
Signed-off-by: NGreg Kurz <groug@kaod.org>
Reviewed-by: NStefan Hajnoczi <stefanha@redhat.com>

The local_open2() callback is vulnerable to symlink attacks because it
calls:

(1) open() which follows symbolic links for all path elements but the
    rightmost one
(2) local_set_xattr()->setxattr() which follows symbolic links for all
    path elements
(3) local_set_mapped_file_attr() which calls in turn local_fopen() and
    mkdir(), both functions following symbolic links for all path
    elements but the rightmost one
(4) local_post_create_passthrough() which calls in turn lchown() and
    chmod(), both functions also following symbolic links

This patch converts local_open2() to rely on opendir_nofollow() and
mkdirat() to fix (1), as well as local_set_xattrat(),
local_set_mapped_file_attrat() and local_set_cred_passthrough() to
fix (2), (3) and (4) respectively. Since local_open2() already opens
a descriptor to the target file, local_set_cred_passthrough() is
modified to reuse it instead of opening a new one.

The mapped and mapped-file security modes are supposed to be identical,
except for the place where credentials and file modes are stored. While
here, we also make that explicit by sharing the call to openat().

This partly fixes CVE-2016-9602.
Signed-off-by: NGreg Kurz <groug@kaod.org>
Reviewed-by: NStefan Hajnoczi <stefanha@redhat.com>

The local_mkdir() callback is vulnerable to symlink attacks because it
calls:

(1) mkdir() which follows symbolic links for all path elements but the
    rightmost one
(2) local_set_xattr()->setxattr() which follows symbolic links for all
    path elements
(3) local_set_mapped_file_attr() which calls in turn local_fopen() and
    mkdir(), both functions following symbolic links for all path
    elements but the rightmost one
(4) local_post_create_passthrough() which calls in turn lchown() and
    chmod(), both functions also following symbolic links

This patch converts local_mkdir() to rely on opendir_nofollow() and
mkdirat() to fix (1), as well as local_set_xattrat(),
local_set_mapped_file_attrat() and local_set_cred_passthrough() to
fix (2), (3) and (4) respectively.

The mapped and mapped-file security modes are supposed to be identical,
except for the place where credentials and file modes are stored. While
here, we also make that explicit by sharing the call to mkdirat().

This partly fixes CVE-2016-9602.
Signed-off-by: NGreg Kurz <groug@kaod.org>
Reviewed-by: NStefan Hajnoczi <stefanha@redhat.com>

The local_mknod() callback is vulnerable to symlink attacks because it
calls:

(1) mknod() which follows symbolic links for all path elements but the
    rightmost one
(2) local_set_xattr()->setxattr() which follows symbolic links for all
    path elements
(3) local_set_mapped_file_attr() which calls in turn local_fopen() and
    mkdir(), both functions following symbolic links for all path
    elements but the rightmost one
(4) local_post_create_passthrough() which calls in turn lchown() and
    chmod(), both functions also following symbolic links

This patch converts local_mknod() to rely on opendir_nofollow() and
mknodat() to fix (1), as well as local_set_xattrat() and
local_set_mapped_file_attrat() to fix (2) and (3) respectively.

A new local_set_cred_passthrough() helper based on fchownat() and
fchmodat_nofollow() is introduced as a replacement to
local_post_create_passthrough() to fix (4).

The mapped and mapped-file security modes are supposed to be identical,
except for the place where credentials and file modes are stored. While
here, we also make that explicit by sharing the call to mknodat().

This partly fixes CVE-2016-9602.
Signed-off-by: NGreg Kurz <groug@kaod.org>
Reviewed-by: NStefan Hajnoczi <stefanha@redhat.com>

The local_symlink() callback is vulnerable to symlink attacks because it
calls:

(1) symlink() which follows symbolic links for all path elements but the
    rightmost one
(2) open(O_NOFOLLOW) which follows symbolic links for all path elements but
    the rightmost one
(3) local_set_xattr()->setxattr() which follows symbolic links for all
    path elements
(4) local_set_mapped_file_attr() which calls in turn local_fopen() and
    mkdir(), both functions following symbolic links for all path
    elements but the rightmost one

This patch converts local_symlink() to rely on opendir_nofollow() and
symlinkat() to fix (1), openat(O_NOFOLLOW) to fix (2), as well as
local_set_xattrat() and local_set_mapped_file_attrat() to fix (3) and
(4) respectively.

This partly fixes CVE-2016-9602.
Signed-off-by: NGreg Kurz <groug@kaod.org>
Reviewed-by: NStefan Hajnoczi <stefanha@redhat.com>

The local_chown() callback is vulnerable to symlink attacks because it
calls:

(1) lchown() which follows symbolic links for all path elements but the
    rightmost one
(2) local_set_xattr()->setxattr() which follows symbolic links for all
    path elements
(3) local_set_mapped_file_attr() which calls in turn local_fopen() and
    mkdir(), both functions following symbolic links for all path
    elements but the rightmost one

This patch converts local_chown() to rely on open_nofollow() and
fchownat() to fix (1), as well as local_set_xattrat() and
local_set_mapped_file_attrat() to fix (2) and (3) respectively.

This partly fixes CVE-2016-9602.
Signed-off-by: NGreg Kurz <groug@kaod.org>
Reviewed-by: NStefan Hajnoczi <stefanha@redhat.com>

The local_chmod() callback is vulnerable to symlink attacks because it
calls:

(1) chmod() which follows symbolic links for all path elements
(2) local_set_xattr()->setxattr() which follows symbolic links for all
    path elements
(3) local_set_mapped_file_attr() which calls in turn local_fopen() and
    mkdir(), both functions following symbolic links for all path
    elements but the rightmost one

We would need fchmodat() to implement AT_SYMLINK_NOFOLLOW to fix (1). This
isn't the case on linux unfortunately: the kernel doesn't even have a flags
argument to the syscall :-\ It is impossible to fix it in userspace in
a race-free manner. This patch hence converts local_chmod() to rely on
open_nofollow() and fchmod(). This fixes the vulnerability but introduces
a limitation: the target file must readable and/or writable for the call
to openat() to succeed.

It introduces a local_set_xattrat() replacement to local_set_xattr()
based on fsetxattrat() to fix (2), and a local_set_mapped_file_attrat()
replacement to local_set_mapped_file_attr() based on local_fopenat()
and mkdirat() to fix (3). No effort is made to factor out code because
both local_set_xattr() and local_set_mapped_file_attr() will be dropped
when all users have been converted to use the "at" versions.

This partly fixes CVE-2016-9602.
Signed-off-by: NGreg Kurz <groug@kaod.org>
Reviewed-by: NStefan Hajnoczi <stefanha@redhat.com>

The local_link() callback is vulnerable to symlink attacks because it calls:

(1) link() which follows symbolic links for all path elements but the
    rightmost one
(2) local_create_mapped_attr_dir()->mkdir() which follows symbolic links
    for all path elements but the rightmost one

This patch converts local_link() to rely on opendir_nofollow() and linkat()
to fix (1), mkdirat() to fix (2).

This partly fixes CVE-2016-9602.
Signed-off-by: NGreg Kurz <groug@kaod.org>
Reviewed-by: NStefan Hajnoczi <stefanha@redhat.com>

When using the mapped-file security model, we also have to create a link
for the metadata file if it exists. In case of failure, we should rollback.

That's what this patch does.
Signed-off-by: NGreg Kurz <groug@kaod.org>
Reviewed-by: NStefan Hajnoczi <stefanha@redhat.com>

The local_rename() callback is vulnerable to symlink attacks because it
uses rename() which follows symbolic links in all path elements but the
rightmost one.

This patch simply transforms local_rename() into a wrapper around
local_renameat() which is symlink-attack safe.

This partly fixes CVE-2016-9602.
Signed-off-by: NGreg Kurz <groug@kaod.org>
Reviewed-by: NStefan Hajnoczi <stefanha@redhat.com>

The local_renameat() callback is currently a wrapper around local_rename()
which is vulnerable to symlink attacks.

This patch rewrites local_renameat() to have its own implementation, based
on local_opendir_nofollow() and renameat().

This partly fixes CVE-2016-9602.
Signed-off-by: NGreg Kurz <groug@kaod.org>
Reviewed-by: NStefan Hajnoczi <stefanha@redhat.com>

The local_lstat() callback is vulnerable to symlink attacks because it
calls:

(1) lstat() which follows symbolic links in all path elements but the
    rightmost one
(2) getxattr() which follows symbolic links in all path elements
(3) local_mapped_file_attr()->local_fopen()->openat(O_NOFOLLOW) which
    follows symbolic links in all path elements but the rightmost
    one

This patch converts local_lstat() to rely on opendir_nofollow() and
fstatat(AT_SYMLINK_NOFOLLOW) to fix (1), fgetxattrat_nofollow() to
fix (2).

A new local_fopenat() helper is introduced as a replacement to
local_fopen() to fix (3). No effort is made to factor out code
because local_fopen() will be dropped when all users have been
converted to call local_fopenat().

This partly fixes CVE-2016-9602.
Signed-off-by: NGreg Kurz <groug@kaod.org>
Reviewed-by: NStefan Hajnoczi <stefanha@redhat.com>

The local_readlink() callback is vulnerable to symlink attacks because it
calls:

(1) open(O_NOFOLLOW) which follows symbolic links for all path elements but
    the rightmost one
(2) readlink() which follows symbolic links for all path elements but the
    rightmost one

This patch converts local_readlink() to rely on open_nofollow() to fix (1)
and opendir_nofollow(), readlinkat() to fix (2).

This partly fixes CVE-2016-9602.
Signed-off-by: NGreg Kurz <groug@kaod.org>
Reviewed-by: NStefan Hajnoczi <stefanha@redhat.com>

The local_truncate() callback is vulnerable to symlink attacks because
it calls truncate() which follows symbolic links in all path elements.

This patch converts local_truncate() to rely on open_nofollow() and
ftruncate() instead.

This partly fixes CVE-2016-9602.
Signed-off-by: NGreg Kurz <groug@kaod.org>
Reviewed-by: NStefan Hajnoczi <stefanha@redhat.com>

The local_statfs() callback is vulnerable to symlink attacks because it
calls statfs() which follows symbolic links in all path elements.

This patch converts local_statfs() to rely on open_nofollow() and fstatfs()
instead.

This partly fixes CVE-2016-9602.
Signed-off-by: NGreg Kurz <groug@kaod.org>
Reviewed-by: NStefan Hajnoczi <stefanha@redhat.com>

The local_utimensat() callback is vulnerable to symlink attacks because it
calls qemu_utimens()->utimensat(AT_SYMLINK_NOFOLLOW) which follows symbolic
links in all path elements but the rightmost one or qemu_utimens()->utimes()
which follows symbolic links for all path elements.

This patch converts local_utimensat() to rely on opendir_nofollow() and
utimensat(AT_SYMLINK_NOFOLLOW) directly instead of using qemu_utimens().
It is hence assumed that the OS supports utimensat(), i.e. has glibc 2.6
or higher and linux 2.6.22 or higher, which seems reasonable nowadays.

This partly fixes CVE-2016-9602.
Signed-off-by: NGreg Kurz <groug@kaod.org>
Reviewed-by: NStefan Hajnoczi <stefanha@redhat.com>

The local_remove() callback is vulnerable to symlink attacks because it
calls:

(1) lstat() which follows symbolic links in all path elements but the
    rightmost one
(2) remove() which follows symbolic links in all path elements but the
    rightmost one

This patch converts local_remove() to rely on opendir_nofollow(),
fstatat(AT_SYMLINK_NOFOLLOW) to fix (1) and unlinkat() to fix (2).

This partly fixes CVE-2016-9602.
Signed-off-by: NGreg Kurz <groug@kaod.org>
Reviewed-by: NStefan Hajnoczi <stefanha@redhat.com>

The local_unlinkat() callback is vulnerable to symlink attacks because it
calls remove() which follows symbolic links in all path elements but the
rightmost one.

This patch converts local_unlinkat() to rely on opendir_nofollow() and
unlinkat() instead.

Most of the code is moved to a separate local_unlinkat_common() helper
which will be reused in a subsequent patch to fix the same issue in
local_remove().

This partly fixes CVE-2016-9602.
Signed-off-by: NGreg Kurz <groug@kaod.org>
Reviewed-by: NStefan Hajnoczi <stefanha@redhat.com>
Reviewed-by: NStefan Hajnoczi <stefanha@redhat.com>

The local_open() and local_opendir() callbacks are vulnerable to symlink
attacks because they call:

(1) open(O_NOFOLLOW) which follows symbolic links in all path elements but
    the rightmost one
(2) opendir() which follows symbolic links in all path elements

This patch converts both callbacks to use new helpers based on
openat_nofollow() to only open files and directories if they are
below the virtfs shared folder

This partly fixes CVE-2016-9602.
Signed-off-by: NGreg Kurz <groug@kaod.org>
Reviewed-by: NStefan Hajnoczi <stefanha@redhat.com>

This patch opens the shared folder and caches the file descriptor, so that
it can be used to do symlink-safe path walk.
Signed-off-by: NGreg Kurz <groug@kaod.org>
Reviewed-by: NStefan Hajnoczi <stefanha@redhat.com>

If these functions fail, they should not change *fs. Let's use local
variables to fix this.
Signed-off-by: NGreg Kurz <groug@kaod.org>
Reviewed-by: NStefan Hajnoczi <stefanha@redhat.com>

If this function fails, it should not modify *ctx.
Signed-off-by: NGreg Kurz <groug@kaod.org>
Reviewed-by: NStefan Hajnoczi <stefanha@redhat.com>

This patchset adds the throttle support for the 9p-local driver.
For now this functionality can be enabled only through qemu cli options.
QMP interface and support to other drivers need further extensions.
To make it simple for other 9p drivers, the throttle code has been put in
separate files.
Signed-off-by: NPradeep Jagadeesh <pradeep.jagadeesh@huawei.com>
Reviewed-by: NAlberto Garcia <berto@igalia.com>
(pass extra NULL CoMutex * argument to qemu_co_queue_wait(),
 added options to qemu-options.hx, Greg Kurz)
Signed-off-by: NGreg Kurz <groug@kaod.org>

Signed-off-by: NGreg Kurz <groug@kaod.org>

This helper is similar to v9fs_string_sprintf(), but it includes the
terminating NUL character in the size field.

This is to avoid doing v9fs_string_sprintf((V9fsString *) &path) and
then bumping the size.

Affected users are changed to use this new helper.
Signed-off-by: NGreg Kurz <groug@kaod.org>
Reviewed-by: NCédric Le Goater <clg@kaod.org>

This patch changes the 9p code to use readdir() again instead of
readdir_r(), which is deprecated in glibc 2.24.

All the locking was put in place by a previous patch.
Reviewed-by: NEric Blake <eblake@redhat.com>
Signed-off-by: NGreg Kurz <gkurz@linux.vnet.ibm.com>