Linux and macOS only share some errno definitions with equal macro
name and value. In fact most mappings for errno are completely
different on the two systems.
This patch converts some important errno values from macOS host to
corresponding Linux errno values before eventually sending such error
codes along with 'Rlerror' replies (if 9p2000.L is used that is). Not
having translated errnos before violated the 9p2000.L protocol spec,
which says:
"
size[4] Rlerror tag[2] ecode[4]
... ecode is a numerical Linux errno.
"
https://github.com/chaos/diod/wiki/protocol#lerror----return-error-code
This patch fixes a bunch of misbehaviours when running a Linux client
on macOS host. For instance this patch fixes:
mount -t 9p -o posixacl ...
on Linux guest if security_mode=mapped was used for 9p server, which
refused to mount successfully, because macOS returned ENOATTR==93
when client tried to retrieve POSIX ACL xattrs, because errno 93
is defined as EPROTONOSUPPORT==93 on Linux, so Linux client believed
that xattrs were not supported by filesystem on host in general.
Signed-off-by: Christian Schoenebeck <qemu_oss@crudebyte.com>
Link: https://lore.kernel.org/qemu-devel/20220421124835.3e664669@bahia/
Reviewed-by: Greg Kurz <groug@kaod.org>
Reviewed-by: Akihiko Odaki <akihiko.odaki@gmail.com>
Message-Id: <b322ab298a62069e527d2b032028bdc9115afacd.1651228001.git.qemu_oss@crudebyte.com>
The 'rdev' field in 9p reponse 'Rgetattr' is of type dev_t,
which is actually a system dependant type and therefore both the
size and encoding of dev_t differ between macOS and Linux.
So far we have sent 'rdev' to guest in host's dev_t format as-is,
which caused devices to appear with wrong device numbers on
guests running on macOS hosts, eventually leading to various
misbehaviours on guest in conjunction with device files.
This patch fixes this issue by converting the device number from
host's dev_t format to Linux dev_t format. As 9p request
'Tgettattr' is exclusive to protocol version 9p2000.L, it should
be fair to assume that 'rdev' field is assumed to be in Linux dev_t
format by client as well.
Signed-off-by: Christian Schoenebeck <qemu_oss@crudebyte.com>
Link: https://lore.kernel.org/qemu-devel/20220421093056.5ab1e7ed@bahia/
Reviewed-by: Greg Kurz <groug@kaod.org>
Reviewed-by: Akihiko Odaki <akihiko.odaki@gmail.com>
Message-Id: <b3a430c2c382ba69a7405e04c0b090ab0d86f17e.1651228001.git.qemu_oss@crudebyte.com>
API doc comments in QEMU are supposed to be in kerneldoc format, so
convert API doc comments from Doxygen format to kerneldoc format.
Signed-off-by: Christian Schoenebeck <qemu_oss@crudebyte.com>
Reviewed-by: Greg Kurz <groug@kaod.org>
Message-Id: <dc1c4a85e233f5884ee5f6ec96b87db286083df7.1646314856.git.qemu_oss@crudebyte.com>
API doc comments in QEMU are supposed to be in kerneldoc format, so drop
occurrences of "@c" which is Doxygen format for fixed-width text.
Link: https://lore.kernel.org/qemu-devel/CAFEAcA89+ENOM6x19OEF53Kd2DWkhN5SN21Va0D7yepJSa3Jyg@mail.gmail.com/
Based-on: <E1nP9Oz-00043L-KJ@lizzy.crudebyte.com>
Signed-off-by: Christian Schoenebeck <qemu_oss@crudebyte.com>
Reviewed-by: Peter Maydell <peter.maydell@linaro.org>
Reviewed-by: Greg Kurz <groug@kaod.org>
Message-Id: <E1nPTwO-0006pl-Np@lizzy.crudebyte.com>
Function qemu_dirent_dup() is currently only used by 9pfs server, so move
it from project global header osdep.h to 9pfs specific header 9p-util.h.
Link: https://lore.kernel.org/qemu-devel/CAFEAcA_=HAUNomKD2wurSVaAHa5mrk22A1oHKLWUDjk7v6Khmg@mail.gmail.com/
Based-on: <20220227223522.91937-12-wwcohen@gmail.com>
Signed-off-by: Christian Schoenebeck <qemu_oss@crudebyte.com>
Reviewed-by: Peter Maydell <peter.maydell@linaro.org>
Message-Id: <E1nP9Oz-00043L-KJ@lizzy.crudebyte.com>
Darwin does not support mknodat. However, to avoid race conditions
with later setting the permissions, we must avoid using mknod on
the full path instead. We could try to fchdir, but that would cause
problems if multiple threads try to call mknodat at the same time.
However, luckily there is a solution: Darwin includes a function
that sets the cwd for the current thread only.
This should suffice to use mknod safely.
This function (pthread_fchdir_np) is protected by a check in
meson in a patch later in this series.
Signed-off-by: Keno Fischer <keno@juliacomputing.com>
Signed-off-by: Michael Roitzsch <reactorcontrol@icloud.com>
[Will Cohen: - Adjust coding style
- Replace clang references with gcc
- Note radar filed with Apple for missing syscall
- Replace direct syscall with pthread_fchdir_np and
adjust patch notes accordingly
- Declare pthread_fchdir_np with
- __attribute__((weak_import)) to allow checking for
its presence before usage
- Move declarations above cplusplus guard
- Add CONFIG_PTHREAD_FCHDIR_NP to meson and check for
presence in 9p-util
- Rebase to apply cleanly on top of the 2022-02-10
changes to 9pfs
- Fix line over 90 characters formatting error]
Signed-off-by: Will Cohen <wwcohen@gmail.com>
Message-Id: <20220227223522.91937-10-wwcohen@gmail.com>
Signed-off-by: Christian Schoenebeck <qemu_oss@crudebyte.com>
Reviewed-by: Christian Schoenebeck <qemu_oss@crudebyte.com>
On darwin `fgetxattr` takes two extra optional arguments,
and the l* variants are not defined (in favor of an extra
flag to the regular variants.
Signed-off-by: Keno Fischer <keno@juliacomputing.com>
[Michael Roitzsch: - Rebase for NixOS]
Signed-off-by: Michael Roitzsch <reactorcontrol@icloud.com>
Signed-off-by: Will Cohen <wwcohen@gmail.com>
Message-Id: <20220227223522.91937-9-wwcohen@gmail.com>
Signed-off-by: Christian Schoenebeck <qemu_oss@crudebyte.com>
Acked-by: Christian Schoenebeck <qemu_oss@crudebyte.com>
Darwin doesn't have either of these flags. Darwin does have
F_NOCACHE, which is similar to O_DIRECT, but has different
enough semantics that other projects don't generally map
them automatically. In any case, we don't support O_DIRECT
on Linux at the moment either.
Signed-off-by: Keno Fischer <keno@juliacomputing.com>
[Michael Roitzsch: - Rebase for NixOS]
Signed-off-by: Michael Roitzsch <reactorcontrol@icloud.com>
[Will Cohen: - Adjust coding style]
Signed-off-by: Will Cohen <wwcohen@gmail.com>
Message-Id: <20220227223522.91937-6-wwcohen@gmail.com>
[C.S.: - Fix compiler warning "unused label 'again'". ]
Link: https://lore.kernel.org/qemu-devel/11201492.CjeqJxXfGd@silver/
Signed-off-by: Christian Schoenebeck <qemu_oss@crudebyte.com>
Reviewed-by: Christian Schoenebeck <qemu_oss@crudebyte.com>
On darwin d_seekoff exists, but is optional and does not seem to
be commonly used by file systems. Use `telldir` instead to obtain
the seek offset and inject it into d_seekoff, and create a
qemu_dirent_off helper to call it appropriately when appropriate.
Signed-off-by: Keno Fischer <keno@juliacomputing.com>
[Michael Roitzsch: - Rebase for NixOS]
Signed-off-by: Michael Roitzsch <reactorcontrol@icloud.com>
[Will Cohen: - Adjust to pass testing
- Ensure that d_seekoff is filled using telldir
on darwin, and create qemu_dirent_off helper
to decide which to access]
[Fabian Franz: - Add telldir error handling for darwin]
Signed-off-by: Fabian Franz <fabianfranz.oss@gmail.com>
[Will Cohen: - Ensure that telldir error handling uses
signed int
- Cleanup of telldir error handling
- Remove superfluous error handling for
qemu_dirent_off
- Adjust formatting
- Use qemu_dirent_off in codir.c
- Declare qemu_dirent_off as static to prevent
linker error
- Move qemu_dirent_off above the end-of-file
endif to fix compilation]
Signed-off-by: Will Cohen <wwcohen@gmail.com>
Message-Id: <20220227223522.91937-5-wwcohen@gmail.com>
Signed-off-by: Christian Schoenebeck <qemu_oss@crudebyte.com>
Reviewed-by: Christian Schoenebeck <qemu_oss@crudebyte.com>
QEMU's local 9pfs server passes through O_NOATIME from the client. If
the QEMU process doesn't have permissions to use O_NOATIME (namely, it
does not own the file nor have the CAP_FOWNER capability), the open will
fail. This causes issues when from the client's point of view, it
believes it has permissions to use O_NOATIME (e.g., a process running as
root in the virtual machine). Additionally, overlayfs on Linux opens
files on the lower layer using O_NOATIME, so in this case a 9pfs mount
can't be used as a lower layer for overlayfs (cf.
dabfe19719/vmtest/onoatimehack.c
and https://github.com/NixOS/nixpkgs/issues/54509).
Luckily, O_NOATIME is effectively a hint, and is often ignored by, e.g.,
network filesystems. open(2) notes that O_NOATIME "may not be effective
on all filesystems. One example is NFS, where the server maintains the
access time." This means that we can honor it when possible but fall
back to ignoring it.
Acked-by: Christian Schoenebeck <qemu_oss@crudebyte.com>
Signed-off-by: Omar Sandoval <osandov@fb.com>
Message-Id: <e9bee604e8df528584693a4ec474ded6295ce8ad.1587149256.git.osandov@fb.com>
Signed-off-by: Greg Kurz <groug@kaod.org>
These functions will need custom implementations on Darwin. Since the
implementation is very similar among all of them, and 9p-util already
has the _nofollow version of fgetxattrat, let's move them all there.
Signed-off-by: Keno Fischer <keno@juliacomputing.com>
Signed-off-by: Greg Kurz <groug@kaod.org>
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: Greg Kurz <groug@kaod.org>
Reviewed-by: Eric Blake <eblake@redhat.com>
Tested-by: Zhi Yong Wu <zhiyong.wu@ucloud.cn>
Acked-by: Philippe Mathieu-Daudé <f4bug@amsat.org>
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: Greg Kurz <groug@kaod.org>
Reviewed-by: Eric Blake <eblake@redhat.com>
[groug: fixed typos in changelog]
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: Greg Kurz <groug@kaod.org>
Reviewed-by: Daniel P. Berrange <berrange@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
When O_PATH is used with O_DIRECTORY, it only acts as an optimization: the
openat() syscall simply finds the name in the VFS, and doesn't trigger the
underlying filesystem.
On systems that don't define O_PATH, because they have glibc version 2.13
or older for example, we can safely omit it. We don't want to deactivate
O_PATH globally though, in case it is used without O_DIRECTORY. The is done
with a dedicated macro.
Systems without O_PATH may thus fail to resolve names that involve
unreadable directories, compared to newer systems succeeding, but such
corner case failure is our only option on those older systems to avoid
the security hole of chasing symlinks inappropriately.
Signed-off-by: Greg Kurz <groug@kaod.org>
Reviewed-by: Eric Blake <eblake@redhat.com>
(added last paragraph to changelog as suggested by Eric Blake)
Signed-off-by: Greg Kurz <groug@kaod.org>
The local_lsetxattr() callback is vulnerable to symlink attacks because
it calls lsetxattr() which follows symbolic links in all path elements but
the rightmost one.
This patch introduces a helper to emulate the non-existing fsetxattrat()
function: it is implemented with /proc/self/fd which provides a trusted
path that can be safely passed to lsetxattr().
local_lsetxattr() is converted to use this helper and opendir_nofollow().
This partly fixes CVE-2016-9602.
Signed-off-by: Greg Kurz <groug@kaod.org>
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
The local_lgetxattr() callback is vulnerable to symlink attacks because
it calls lgetxattr() which follows symbolic links in all path elements but
the rightmost one.
This patch introduces a helper to emulate the non-existing fgetxattrat()
function: it is implemented with /proc/self/fd which provides a trusted
path that can be safely passed to lgetxattr().
local_lgetxattr() is converted to use this helper and opendir_nofollow().
This partly fixes CVE-2016-9602.
Signed-off-by: Greg Kurz <groug@kaod.org>
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
When using the passthrough security mode, symbolic links created by the
guest are actual symbolic links on the host file system.
Since the resolution of symbolic links during path walk is supposed to
occur on the client side. The server should hence never receive any path
pointing to an actual symbolic link. This isn't guaranteed by the protocol
though, and malicious code in the guest can trick the server to issue
various syscalls on paths whose one or more elements are symbolic links.
In the case of the "local" backend using the "passthrough" or "none"
security modes, the guest can directly create symbolic links to arbitrary
locations on the host (as per spec). The "mapped-xattr" and "mapped-file"
security modes are also affected to a lesser extent as they require some
help from an external entity to create actual symbolic links on the host,
i.e. another guest using "passthrough" mode for example.
The current code hence relies on O_NOFOLLOW and "l*()" variants of system
calls. Unfortunately, this only applies to the rightmost path component.
A guest could maliciously replace any component in a trusted path with a
symbolic link. This could allow any guest to escape a virtfs shared folder.
This patch introduces a variant of the openat() syscall that successively
opens each path element with O_NOFOLLOW. When passing a file descriptor
pointing to a trusted directory, one is guaranteed to be returned a
file descriptor pointing to a path which is beneath the trusted directory.
This will be used by subsequent patches to implement symlink-safe path walk
for any access to the backend.
Symbolic links aren't the only threats actually: a malicious guest could
change a path element to point to other types of file with undesirable
effects:
- a named pipe or any other thing that would cause openat() to block
- a terminal device which would become QEMU's controlling terminal
These issues can be addressed with O_NONBLOCK and O_NOCTTY.
Two helpers are introduced: one to open intermediate path elements and one
to open the rightmost path element.
Suggested-by: Jann Horn <jannh@google.com>
Signed-off-by: Greg Kurz <groug@kaod.org>
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
(renamed openat_nofollow() to relative_openat_nofollow(),
assert path is relative and doesn't contain '//',
fixed side-effect in assert, Greg Kurz)
Signed-off-by: Greg Kurz <groug@kaod.org>
Christian Schoenebeck
Keno Fischer
Omar Sandoval
Greg Kurz