Introduce lo_dirp_put() so that FUSE_RELEASEDIR does not cause
use-after-free races with other threads that are accessing lo_dirp.
Also make lo_releasedir() atomic to prevent FUSE_RELEASEDIR racing with
itself. This prevents double-frees.
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
Reviewed-by: Philippe Mathieu-Daudé <philmd@redhat.com>
Signed-off-by: Dr. David Alan Gilbert <dgilbert@redhat.com>
Hold the lock across both lo_map_get() and lo_map_remove() to prevent
races between two FUSE_RELEASE requests. In this case I don't see a
serious bug but it's safer to do things atomically.
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
Reviewed-by: Daniel P. Berrangé <berrange@redhat.com>
Signed-off-by: Dr. David Alan Gilbert <dgilbert@redhat.com>
Doing posix locks with-in guest kernel are not sufficient if a file/dir
is being shared by multiple guests. So we need the notion of daemon doing
the locks which are visible to rest of the guests.
Given posix locks are per process, one can not call posix lock API on host,
otherwise bunch of basic posix locks properties are broken. For example,
If two processes (A and B) in guest open the file and take locks on different
sections of file, if one of the processes closes the fd, it will close
fd on virtiofsd and all posix locks on file will go away. This means if
process A closes the fd, then locks of process B will go away too.
Similar other problems exist too.
This patch set tries to emulate posix locks while using open file
description locks provided on Linux.
Daemon provides two options (-o posix_lock, -o no_posix_lock) to enable
or disable posix locking in daemon. By default it is enabled.
There are few issues though.
- GETLK() returns pid of process holding lock. As we are emulating locks
using OFD, and these locks are not per process and don't return pid
of process, so GETLK() in guest does not reuturn process pid.
- As of now only F_SETLK is supported and not F_SETLKW. We can't block
the thread in virtiofsd for arbitrary long duration as there is only
one thread serving the queue. That means unlock request will not make
it to daemon and F_SETLKW will block infinitely and bring virtio-fs
to a halt. This is a solvable problem though and will require significant
changes in virtiofsd and kernel. Left as a TODO item for now.
Signed-off-by: Vivek Goyal <vgoyal@redhat.com>
Reviewed-by: Masayoshi Mizuma <m.mizuma@jp.fujitsu.com>
Signed-off-by: Dr. David Alan Gilbert <dgilbert@redhat.com>
Signed-off-by: Eric Ren <renzhen@linux.alibaba.com>
Reviewed-by: Daniel P. Berrangé <berrange@redhat.com>
Signed-off-by: Dr. David Alan Gilbert <dgilbert@redhat.com>
This offers an helper function for lo_data's cleanup.
Signed-off-by: Liu Bo <bo.liu@linux.alibaba.com>
Reviewed-by: Daniel P. Berrangé <berrange@redhat.com>
Signed-off-by: Dr. David Alan Gilbert <dgilbert@redhat.com>
valgrind reported that lo.source is leaked on quiting, but it was defined
as (const char*) as it may point to a const string "/".
Signed-off-by: Liu Bo <bo.liu@linux.alibaba.com>
Reviewed-by: Misono Tomohiro <misono.tomohiro@jp.fujitsu.com>
Signed-off-by: Dr. David Alan Gilbert <dgilbert@redhat.com>
Neither fuse_parse_cmdline() nor fuse_opt_parse() goes to the right place
to do cleanup.
Signed-off-by: Liu Bo <bo.liu@linux.alibaba.com>
Reviewed-by: Daniel P. Berrangé <berrange@redhat.com>
Signed-off-by: Dr. David Alan Gilbert <dgilbert@redhat.com>
Clear out our inodes and fd's on a 'destroy' - so we get rid
of them if we reboot the guest.
Signed-off-by: Dr. David Alan Gilbert <dgilbert@redhat.com>
Reviewed-by: Daniel P. Berrangé <berrange@redhat.com>
Signed-off-by: Dr. David Alan Gilbert <dgilbert@redhat.com>
Improve performance of inode lookup by using a hash table.
Signed-off-by: Miklos Szeredi <mszeredi@redhat.com>
Signed-off-by: Dr. David Alan Gilbert <dgilbert@redhat.com>
Signed-off-by: Liu Bo <bo.liu@linux.alibaba.com>
Reviewed-by: Daniel P. Berrangé <berrange@redhat.com>
Signed-off-by: Dr. David Alan Gilbert <dgilbert@redhat.com>
- Rename "cache=never" to "cache=none" to match 9p's similar option.
- Rename CACHE_NORMAL constant to CACHE_AUTO to match the "cache=auto"
option.
Signed-off-by: Miklos Szeredi <mszeredi@redhat.com>
Signed-off-by: Dr. David Alan Gilbert <dgilbert@redhat.com>
Reviewed-by: Daniel P. Berrangé <berrange@redhat.com>
Signed-off-by: Dr. David Alan Gilbert <dgilbert@redhat.com>
Inititialize the root inode in a single place.
Signed-off-by: Miklos Szeredi <mszeredi@redhat.com>
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
dgilbert:
with fix suggested by Misono Tomohiro <misono.tomohiro@jp.fujitsu.com>
Reviewed-by: Misono Tomohiro <misono.tomohiro@jp.fujitsu.com>
Signed-off-by: Dr. David Alan Gilbert <dgilbert@redhat.com>
The Linux file handle APIs (struct export_operations) can access inodes
that are not attached to parents because path name traversal is not
performed. Refuse if there is no parent in lo_do_lookup().
Also clean up lo_do_lookup() while we're here.
Signed-off-by: Miklos Szeredi <mszeredi@redhat.com>
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
Reviewed-by: Misono Tomohiro <misono.tomohiro@jp.fujitsu.com>
Signed-off-by: Dr. David Alan Gilbert <dgilbert@redhat.com>
Signed-off-by: Miklos Szeredi <mszeredi@redhat.com>
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
Reviewed-by: Daniel P. Berrangé <berrange@redhat.com>
Signed-off-by: Dr. David Alan Gilbert <dgilbert@redhat.com>
Signed-off-by: Miklos Szeredi <mszeredi@redhat.com>
Reviewed-by: Misono Tomohiro <misono.tomohiro@jp.fujitsu.com>
Signed-off-by: Dr. David Alan Gilbert <dgilbert@redhat.com>
...because the attributes sent in the READDIRPLUS reply would be discarded
anyway.
Signed-off-by: Miklos Szeredi <mszeredi@redhat.com>
Reviewed-by: Daniel P. Berrangé <berrange@redhat.com>
Signed-off-by: Dr. David Alan Gilbert <dgilbert@redhat.com>
virtiofsd has some threads, so we see a lot of logs with debug option.
It would be useful for debugging if we can see the timestamp.
Add nano second timestamp, which got by get_clock(), to the log with
FUSE_LOG_DEBUG level if the syslog option isn't set.
The log is like as:
# ./virtiofsd -d -o vhost_user_socket=/tmp/vhostqemu0 -o source=/tmp/share0 -o cache=auto
...
[5365943125463727] [ID: 00000002] fv_queue_thread: Start for queue 0 kick_fd 9
[5365943125568644] [ID: 00000002] fv_queue_thread: Waiting for Queue 0 event
[5365943125573561] [ID: 00000002] fv_queue_thread: Got queue event on Queue 0
Signed-off-by: Masayoshi Mizuma <m.mizuma@jp.fujitsu.com>
Reviewed-by: Philippe Mathieu-Daudé <philmd@redhat.com>
Signed-off-by: Dr. David Alan Gilbert <dgilbert@redhat.com>
virtiofsd has some threads, so we see a lot of logs with debug option.
It would be useful for debugging if we can identify the specific thread
from the log.
Add ID, which is got by gettid(), to the log with FUSE_LOG_DEBUG level
so that we can grep the specific thread.
The log is like as:
]# ./virtiofsd -d -o vhost_user_socket=/tmp/vhostqemu0 -o source=/tmp/share0 -o cache=auto
...
[ID: 00000097] unique: 12696, success, outsize: 120
[ID: 00000097] virtio_send_msg: elem 18: with 2 in desc of length 120
[ID: 00000003] fv_queue_thread: Got queue event on Queue 1
[ID: 00000003] fv_queue_thread: Queue 1 gave evalue: 1 available: in: 65552 out: 80
[ID: 00000003] fv_queue_thread: Waiting for Queue 1 event
[ID: 00000071] fv_queue_worker: elem 33: with 2 out desc of length 80 bad_in_num=0 bad_out_num=0
[ID: 00000071] unique: 12694, opcode: READ (15), nodeid: 2, insize: 80, pid: 2014
[ID: 00000071] lo_read(ino=2, size=65536, off=131072)
Signed-off-by: Masayoshi Mizuma <m.mizuma@jp.fujitsu.com>
Signed-off-by: Dr. David Alan Gilbert <dgilbert@redhat.com>
added rework as suggested by Daniel P. Berrangé during review
Reviewed-by: Philippe Mathieu-Daudé <philmd@redhat.com>
Signed-off-by: Dr. David Alan Gilbert <dgilbert@redhat.com>
Introduce "-o log_level=" command line option to specify current log
level (priority), valid values are "debug info warn err", e.g.
./virtiofsd -o log_level=debug ...
So only log priority higher than "debug" will be printed to
stderr/syslog. And the default level is info.
The "-o debug"/"-d" options are kept, and imply debug log level.
Signed-off-by: Eryu Guan <eguan@linux.alibaba.com>
dgilbert: Reworked for libfuse's log_func
Signed-off-by: Dr. David Alan Gilbert <dgilbert@redhat.com>
with fix by:
Signed-off-by: Xiao Yang <yangx.jy@cn.fujitsu.com>
Reviewed-by: Daniel P. Berrangé <berrange@redhat.com>
Signed-off-by: Dr. David Alan Gilbert <dgilbert@redhat.com>
Sometimes collecting output from stderr is inconvenient or does not fit
within the overall logging architecture. Add syslog(3) support for
cases where stderr cannot be used.
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
dgilbert: Reworked as a logging function
Signed-off-by: Dr. David Alan Gilbert <dgilbert@redhat.com>
Reviewed-by: Daniel P. Berrangé <berrange@redhat.com>
Signed-off-by: Dr. David Alan Gilbert <dgilbert@redhat.com>
virtiofsd can exceed the default open file descriptor limit easily on
most systems. Take advantage of the fact that it runs as root to raise
the limit.
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
Reviewed-by: Daniel P. Berrangé <berrange@redhat.com>
Reviewed-by: Philippe Mathieu-Daudé <philmd@redhat.com>
Signed-off-by: Dr. David Alan Gilbert <dgilbert@redhat.com>
If client requested killing setuid/setgid bits on file being written, drop
CAP_FSETID capability so that setuid/setgid bits are cleared upon write
automatically.
pjdfstest chown/12.t needs this.
Signed-off-by: Vivek Goyal <vgoyal@redhat.com>
dgilbert: reworked for libcap-ng
Reviewed-by: Misono Tomohiro <misono.tomohiro@jp.fujitsu.com>
Reviewed-by: Sergio Lopez <slp@redhat.com>
Signed-off-by: Dr. David Alan Gilbert <dgilbert@redhat.com>
libcap-ng reads /proc during capng_get_caps_process, and virtiofsd's
sandboxing doesn't have /proc mounted; thus we have to do the
caps read before we sandbox it and save/restore the state.
Signed-off-by: Dr. David Alan Gilbert <dgilbert@redhat.com>
Reviewed-by: Daniel P. Berrangé <berrange@redhat.com>
Signed-off-by: Dr. David Alan Gilbert <dgilbert@redhat.com>
Only allow system calls that are needed by virtiofsd. All other system
calls cause SIGSYS to be directed at the thread and the process will
coredump.
Restricting system calls reduces the kernel attack surface and limits
what the process can do when compromised.
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
with additional entries by:
Signed-off-by: Ganesh Maharaj Mahalingam <ganesh.mahalingam@intel.com>
Signed-off-by: Masayoshi Mizuma <m.mizuma@jp.fujitsu.com>
Signed-off-by: Misono Tomohiro <misono.tomohiro@jp.fujitsu.com>
Signed-off-by: piaojun <piaojun@huawei.com>
Signed-off-by: Vivek Goyal <vgoyal@redhat.com>
Signed-off-by: Eric Ren <renzhen@linux.alibaba.com>
Signed-off-by: Dr. David Alan Gilbert <dgilbert@redhat.com>
virtiofsd needs access to /proc/self/fd. Let's move to a new pid
namespace so that a compromised process cannot see another other
processes running on the system.
One wrinkle in this approach: unshare(CLONE_NEWPID) affects *child*
processes and not the current process. Therefore we need to fork the
pid 1 process that will actually run virtiofsd and leave a parent in
waitpid(2). This is not the same thing as daemonization and parent
processes should not notice a difference.
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
Reviewed-by: Daniel P. Berrangé <berrange@redhat.com>
Signed-off-by: Dr. David Alan Gilbert <dgilbert@redhat.com>
If the process is compromised there should be no network access. Use an
empty network namespace to sandbox networking.
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
Reviewed-by: Daniel P. Berrangé <berrange@redhat.com>
Signed-off-by: Dr. David Alan Gilbert <dgilbert@redhat.com>
Use a mount namespace with the shared directory tree mounted at "/" and
no other mounts.
This prevents symlink escape attacks because symlink targets are
resolved only against the shared directory and cannot go outside it.
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
Signed-off-by: Peng Tao <tao.peng@linux.alibaba.com>
Reviewed-by: Daniel P. Berrangé <berrange@redhat.com>
Signed-off-by: Dr. David Alan Gilbert <dgilbert@redhat.com>
Sandboxing will remove /proc from the mount namespace so we can no
longer build string paths into "/proc/self/fd/...".
Keep an O_PATH file descriptor so we can still re-open fds via
/proc/self/fd.
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
Reviewed-by: Philippe Mathieu-Daudé <philmd@redhat.com>
Signed-off-by: Dr. David Alan Gilbert <dgilbert@redhat.com>
Construct a fake dirent for the root directory's ".." entry. This hides
the parent directory from the FUSE client.
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
Reviewed-by: Sergio Lopez <slp@redhat.com>
Signed-off-by: Dr. David Alan Gilbert <dgilbert@redhat.com>
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
Reviewed-by: Sergio Lopez <slp@redhat.com>
Signed-off-by: Dr. David Alan Gilbert <dgilbert@redhat.com>
Several FUSE requests contain single path components. A correct FUSE
client sends well-formed path components but there is currently no input
validation in case something went wrong or the client is malicious.
Refuse ".", "..", and paths containing '/' when we expect a path
component.
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
Reviewed-by: Daniel P. Berrangé <berrange@redhat.com>
Signed-off-by: Dr. David Alan Gilbert <dgilbert@redhat.com>
We have two operations that cannot be done race-free on a symlink in
certain cases: utimes and link.
Add racy fallback for these if the race-free method doesn't work. We do
our best to avoid races even in this case:
- get absolute path by reading /proc/self/fd/NN symlink
- lookup parent directory: after this we are safe against renames in
ancestors
- lookup name in parent directory, and verify that we got to the original
inode, if not retry the whole thing
Both utimes(2) and link(2) hold i_lock on the inode across the operation,
so a racing rename/delete by this fuse instance is not possible, only from
other entities changing the filesystem.
If the "norace" option is given, then disable the racy fallbacks.
Signed-off-by: Miklos Szeredi <mszeredi@redhat.com>
Reviewed-by: Masayoshi Mizuma <m.mizuma@jp.fujitsu.com>
Signed-off-by: Dr. David Alan Gilbert <dgilbert@redhat.com>
Do not expose file descriptor numbers to clients. This prevents the
abuse of internal file descriptors (like stdin/stdout).
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
Fix from:
Signed-off-by: Xiao Yang <yangx.jy@cn.fujitsu.com>
dgilbert:
Added lseek
Reviewed-by: Masayoshi Mizuma <m.mizuma@jp.fujitsu.com>
Signed-off-by: Dr. David Alan Gilbert <dgilbert@redhat.com>
Do not expose lo_dirp pointers to clients.
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
Reviewed-by: Philippe Mathieu-Daudé <philmd@redhat.com>
Signed-off-by: Dr. David Alan Gilbert <dgilbert@redhat.com>
Do not expose lo_inode pointers to clients.
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
Reviewed-by: Masayoshi Mizuma <m.mizuma@jp.fujitsu.com>
Signed-off-by: Dr. David Alan Gilbert <dgilbert@redhat.com>
A layer of indirection is needed because passthrough_ll cannot expose
pointers or file descriptor numbers to untrusted clients. Malicious
clients could send invalid pointers or file descriptors in order to
crash or exploit the file system daemon.
lo_map provides an integer key->value mapping. This will be used for
ino and fh fields in the patches that follow.
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
Reviewed-by: Masayoshi Mizuma <m.mizuma@jp.fujitsu.com>
Signed-off-by: Dr. David Alan Gilbert <dgilbert@redhat.com>
We need to create files in the caller's context. Otherwise after
creating a file, the caller might not be able to do file operations on
that file.
Changed effective uid/gid to caller's uid/gid, create file and then
switch back to uid/gid 0.
Use syscall(setresuid, ...) otherwise glibc does some magic to change EUID
in all threads, which is not what we want.
Signed-off-by: Vivek Goyal <vgoyal@redhat.com>
Signed-off-by: Miklos Szeredi <mszeredi@redhat.com>
Signed-off-by: Dr. David Alan Gilbert <dgilbert@redhat.com>
Add the --print-capabilities option as per vhost-user.rst "Backend
programs conventions". Currently there are no advertised features.
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
Reviewed-by: Daniel P. Berrangé <berrange@redhat.com>
Signed-off-by: Dr. David Alan Gilbert <dgilbert@redhat.com>
Listen on our unix socket for the connection from QEMU, when we get it
initialise vhost-user and dive into our own loop variant (currently
dummy).
Signed-off-by: Dr. David Alan Gilbert <dgilbert@redhat.com>
Reviewed-by: Daniel P. Berrangé <berrange@redhat.com>
Signed-off-by: Dr. David Alan Gilbert <dgilbert@redhat.com>
The -o source=PATH option will be used by most command-line invocations.
Let's document it!
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
Reviewed-by: Daniel P. Berrangé <berrange@redhat.com>
Signed-off-by: Dr. David Alan Gilbert <dgilbert@redhat.com>
If caller has not sent file handle in request, then using inode, retrieve
the fd opened using O_PATH and use that to open file again and issue
fsync. This will be needed when dax_flush() calls fsync. At that time
we only have inode information (and not file).
Signed-off-by: Vivek Goyal <vgoyal@redhat.com>
Reviewed-by: Daniel P. Berrangé <berrange@redhat.com>
Signed-off-by: Dr. David Alan Gilbert <dgilbert@redhat.com>
1) Use correct CONFIG_FALLOCATE macro to check if fallocate() is supported.(i.e configure
script sets CONFIG_FALLOCATE intead of HAVE_FALLOCATE if fallocate() is supported)
2) Replace HAVE_POSIX_FALLOCATE with CONFIG_POSIX_FALLOCATE.
Signed-off-by: Xiao Yang <yangx.jy@cn.fujitsu.com>
Signed-off-by: Dr. David Alan Gilbert <dgilbert@redhat.com>
Merged from two of Xiao Yang's patches
All of the fuse files include config.h and define GNU_SOURCE
where we don't have either under our build - remove them.
Fixup path to the kernel's fuse.h in the QEMUs world.
Signed-off-by: Dr. David Alan Gilbert <dgilbert@redhat.com>
Reviewed-by: Philippe Mathieu-Daudé <philmd@redhat.com>
Tested-by: Philippe Mathieu-Daudé <philmd@redhat.com>
Signed-off-by: Dr. David Alan Gilbert <dgilbert@redhat.com>
Signed-off-by: Xiao Yang <yangx.jy@cn.fujitsu.com>
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
Signed-off-by: Dr. David Alan Gilbert <dgilbert@redhat.com>
Classic FUSE file system daemons take a mountpoint argument but
virtiofsd exposes a vhost-user UNIX domain socket instead. The
mountpoint argument is not used by virtiofsd but the user is still
required to pass a dummy argument on the command-line.
Remove the mountpoint argument to clean up the command-line.
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
Reviewed-by: Daniel P. Berrangé <berrange@redhat.com>
Signed-off-by: Dr. David Alan Gilbert <dgilbert@redhat.com>
There's a lot of the original fuse code we don't need; trim them down.
Signed-off-by: Dr. David Alan Gilbert <dgilbert@redhat.com>
with additional trimming by:
Signed-off-by: Misono Tomohiro <misono.tomohiro@jp.fujitsu.com>
Reviewed-by: Daniel P. Berrangé <berrange@redhat.com>
Reviewed-by: Xiao Yang <yangx.jy@cn.fujitsu.com>
Signed-off-by: Dr. David Alan Gilbert <dgilbert@redhat.com>
passthrough_ll is one of the examples in the upstream fuse project
and is the main part of our daemon here. It passes through requests
from fuse to the underlying filesystem, using syscalls as directly
as possible.
From libfuse fuse-3.8.0
Signed-off-by: Dr. David Alan Gilbert <dgilbert@redhat.com>
Fixed up 'GPL' to 'GPLv2' as per Dan's comments and consistent
with the 'LICENSE' file in libfuse; patch sent to libfuse to fix
it upstream.
Reviewed-by: Daniel P. Berrangé <berrange@redhat.com>
Signed-off-by: Dr. David Alan Gilbert <dgilbert@redhat.com>