Daiki Ueno [Tue, 28 Feb 2017 15:06:50 +0000 (16:06 +0100)]
build: Remove systemd unit files for now
Given that the remote proxy service shall be only used by NetworkManager
and not generally useful, revert commit a4fb2bb587fb1a0146cf97f039b671d3258488f9 for now.
Once the necessary command that runs the proxy module is implemented in
p11-kit, maybe NetworkManager itself could install those files.
In common/unix-peer.c, we are checking if HAVE_GETPEEREID is defined,
however, we never actually check if getpeereid() is available, so
fix that by checking this function using AC_CHECK_FUNCS().
Daiki Ueno [Fri, 17 Feb 2017 15:18:21 +0000 (16:18 +0100)]
trust: Honor "modifiable" setting in persist file
Previously, all objects read from p11-kit persist files are marked as
modifiable when parsing, regardless of the explicit "modifiable: false"
setting in the file.
Reported by Kai Engert in:
https://bugs.freedesktop.org/show_bug.cgi?id=99797
Daiki Ueno [Wed, 11 Jan 2017 08:32:19 +0000 (09:32 +0100)]
rpc: Add PKCS#11 module that connects to socket
This patch adds a PKCS#11 module that connects to the p11-kit server
exposed on the filesystem. The filename of the socket is determined in
the following order:
- $P11_KIT_SERVER_ADDRESS, if the envvar is available
- $XDG_RUNTIME_DIR/p11-kit/pkcs11, if the envvar is available
- /run/$(id -u)/p11-kit/pkcs11, if /run/$(id -u) exists
- /var/run/$(id -u)/p11-kit/pkcs11, if /var/run/$(id -u) exists
- ~/.cache/p11-kit/pkcs11.
Note that the program loading this module may have called setuid() and
secure_getenv() which we use for fetching envvars could return NULL.
Daiki Ueno [Wed, 15 Feb 2017 11:07:31 +0000 (12:07 +0100)]
rpc: Port exec transport to Windows
On Windows, use _spawnv() to create a subprocess and two unidirectional
pipe created with _pipe() to communicate with it. If we can assume
WinSock, it might be simpler to use a socketpair() replacement from:
https://github.com/ncm/selectable-socketpair.
Kai Engert [Thu, 2 Feb 2017 15:01:01 +0000 (16:01 +0100)]
Support loading new NSS attribute CKA_NSS_MOZILLA_CA_POLICY from .p11-kit files.
See also NSS bug https://bugzilla.mozilla.org/show_bug.cgi?id=1334976
and p11-kit bug https://bugs.freedesktop.org/show_bug.cgi?id=99453
Daiki Ueno [Tue, 14 Feb 2017 15:36:23 +0000 (16:36 +0100)]
virtual: Move mutex into p11_library_init()
We used to provide p11_virtual_fixed_{,un}init() to only initialize a
mutex used in virtual.c. That required all the tests calling virtual
functions to call p11_virtual_fixed_{,un}init() in main().
For simplicity, move the mutex variable initialization into
p11_library_init().
Daiki Ueno [Mon, 13 Feb 2017 15:32:01 +0000 (16:32 +0100)]
trust: Revert to the original 'extract' behavior
Since commit f4384a40, due to a missing ex->flags setting, the 'trust
extract' command didn't retrieve correlation between related objects and
that was causing assertion failure when writing PEM files.
Daiki Ueno [Sat, 17 Dec 2016 06:11:36 +0000 (07:11 +0100)]
iter: Enable iteration over slots/tokens/modules
While PKCS#11 URI can identify slots/tokens/modules, P11KitIter is only
capable of iterating over objects.
This patch adds new behaviors to P11KitIter to support iterations over
slots/tokens/modules, using the C coroutine trick as described in:
http://www.chiark.greenend.org.uk/~sgtatham/coroutines.html
Daiki Ueno [Mon, 23 Jan 2017 09:02:56 +0000 (10:02 +0100)]
uri: Relax pin-* parsing for compatibility
While 'pin-source' and 'pin-value' are defined as query atttribute, they
were defined as path attribute in earlier drafts, and some
implementations still stick to it.
For backward compatibility, accept those in path attributes when
parsing (but not when formatting).
Reported by Andreas Metzler in:
https://lists.freedesktop.org/archives/p11-glue/2017-January/000637.html
Stef Walter [Sun, 29 Jan 2017 14:10:37 +0000 (15:10 +0100)]
trust: Implement a 'trust dump' command
This dumps all the PKCS#11 objects in the internal .p11-kit
persistence format.
This is part of the trust command and tooling, even though
at some point it could go in the p11-kit command. The reason
for this is that the code related to the internal .p11-kit
objects is in the trust code, and consumed solely by the
trust related modules.
Daiki Ueno [Fri, 26 Aug 2016 15:20:06 +0000 (17:20 +0200)]
build: Make libffi closure optional
libffi's closure support is not available on all platforms and may fail
at run time if running under a stricter SELinux policy. Fallback to
pre-compiled closures if it is not usable.
* common/compat.c: Fix "implicit declaration of function 'issetugid'"
warning. On FreeBSD, it's required to define __BSD_VISIBLE to make
issetugid(2) visible
* common/test-message.c: Fix "implicit declaration of function
'asprintf'" by including <stdio.h>
* p11-kit/test-iter.c: Fix "format '%lu' expects argument of
type 'long unsigned int', but argument 3 has type 'int'" by
changing format string to "%d"
Lubomir Rintel [Thu, 8 Dec 2016 17:20:37 +0000 (18:20 +0100)]
systemd: add per-user remoting socket
This allows daemons outside user's session to use per-user PKCS#11
modules. Useful for letting VPN daemons or wpa_supplicant use
certificates stored in user's GNOME keyring, etc.
Lubomir Rintel [Tue, 3 Nov 2015 07:11:39 +0000 (08:11 +0100)]
common: use recursive pthread mutex for library lock
This allows us to do nested locking within one thread avoiding a lockup
when remoting the p11-kit-proxy.so module:
#0 0x00007f190f35838d in __lll_lock_wait () from /lib64/libpthread.so.0
#1 0x00007f190f351e4d in pthread_mutex_lock () from /lib64/libpthread.so.0
#2 0x00007f190f98657f in C_GetFunctionList (list=0x7ffe7ec3f798) at p11-kit/proxy.c:2355
#3 0x00007f190f993cc9 in dlopen_and_get_function_list (funcs=0x7ffe7ec3f798, path=0x7ffe7ec40926 "/usr/local/lib/p11-kit-proxy.so", mod=0x249e3d0) at p11-kit/modules.c:337
#4 load_module_from_file_inlock (name=name@entry=0x0, path=path@entry=0x7ffe7ec40926 "/usr/local/lib/p11-kit-proxy.so", result=result@entry=0x7ffe7ec3f7e8) at p11-kit/modules.c:382
#5 0x00007f190f99587f in p11_kit_module_load (module_path=module_path@entry=0x7ffe7ec40926 "/usr/local/lib/p11-kit-proxy.so", flags=flags@entry=0) at p11-kit/modules.c:2427
#6 0x0000000000401c4b in serve_module_from_file (file=0x7ffe7ec40926 "/usr/local/lib/p11-kit-proxy.so") at p11-kit/remote.c:105
#7 main (argc=1, argv=<optimized out>) at p11-kit/remote.c:169
The Windows NT mutex is aready recursive by default.
Daiki Ueno [Mon, 16 Jan 2017 13:06:30 +0000 (14:06 +0100)]
uri: Avoid typecasting confusion on s390x
Like memcpy(), the 'void *' argument of p11_buffer_add() points to the
memory area ordered in host's endianness. Add typecast of int->char to
avoid the confusion.
Reported by Andreas Metzler in:
https://lists.freedesktop.org/archives/p11-glue/2017-January/000633.html
Lubomir Rintel [Wed, 28 Dec 2016 15:11:21 +0000 (16:11 +0100)]
uri: fix the query attribute parsing
The pin-* attributes belong to the query part. We should not parse them
until we see a '?' and they're separated with a '&'.
This might be an important thing -- some of the query attributes may
have security implications reaching outside scope of the token itself, to the
host system itself. E.g. a pin-source may cause the consumer to access a file
or module-path (unimplemented) execute code. The user may want to just chop the
attribute part off if they want the consumer access the token and not take the
security considerations into account.
- Current command for creation of the p11-kit-proxy symlink
uses shell brace expansion that isn't supported by all
the shells (e.g. FreeBSD's /bin/sh does not support that).
Replace it with the old-fashioned 'for' loop
- Match extension of the source and the target, i.e. so links
to so, dylib links to dylib (previously dylib linked to so)
- Add an uninstall-local target to clean up the symlink
This fixes issues pointed in:
https://bugzilla.redhat.com/show_bug.cgi?id=985445
except for p11-kit/conf.c:read_config_file(), which was rewritten using
mmap() and thus length calculation is no longer needed.
Previously p11-kit-trust.so tried to interpret certificate as PEM format
first. This could cause potential conflict if the certificate were
actually in DER format and contained a PEM marker strings.
This test hasn't been working since the removal of the pthread_atfork()
deinit code. To properly clean up, the child process needs to call
C_Initialize() and C_Finalize(), and it is already tested by
/proxy/initialize-child.
GCC's asan spotted this:
Direct leak of 338 byte(s) in 13 object(s) allocated from:
#0 0x7f54f03fee20 in malloc (/lib64/libasan.so.3+0xc6e20)
#1 0x445e8c in p11_path_build ../common/path.c:222
#2 0x4385bd in expand_tempdir ../common/test.c:334
#3 0x43869c in p11_test_directory ../common/test.c:361
#4 0x4033e3 in setup_temp ../trust/test-token.c:79
Reset mod->init_count when forkid has changed. Otherwise C_Finalize
does not get called.
GCC's asan spotted this:
Direct leak of 48 byte(s) in 1 object(s) allocated from:
#0 0x7f89bc7bfe20 in malloc (/lib64/libasan.so.3+0xc6e20)
#1 0x7f89bc47a1f1 in p11_dict_new ../common/dict.c:278
#2 0x7f89bc42143d in managed_C_Initialize ../p11-kit/modules.c:1477
#3 0x7f89bc464c72 in binding_C_Initialize ../p11-kit/virtual.c:121
#4 0x7f89bc1b0a51 in ffi_closure_unix64_inner (/lib64/libffi.so.6+0x5a51)
#5 0x7f89bc1b0dbf in ffi_closure_unix64 (/lib64/libffi.so.6+0x5dbf)
#6 0x7f89bc44f9e8 in rpc_C_Initialize ../p11-kit/rpc-server.c:691
Make sure to call p11_virtual_uninit() on managed module. Otherwise the
associated lower_module will not be released.
GCC's asan spotted this:
Direct leak of 56 byte(s) in 1 object(s) allocated from:
#0 0x7f6c5368dfe0 in calloc (/lib64/libasan.so.3+0xc6fe0)
#1 0x4436ba in p11_rpc_client_init ../p11-kit/rpc-client.c:2082
#2 0x42c147 in p11_rpc_transport_new ../p11-kit/rpc-transport.c:850
#3 0x415d95 in setup_module_for_remote_inlock ../p11-kit/modules.c:411
GCC's asan spotted this:
Direct leak of 120 byte(s) in 1 object(s) allocated from:
#0 0x7f8d4f221fe0 in calloc (/lib64/libasan.so.3+0xc6fe0)
#1 0x427f55 in rpc_socket_new ../p11-kit/rpc-transport.c:100
#2 0x42bc1b in rpc_exec_connect ../p11-kit/rpc-transport.c:767
However, on Mac OS X the library is named libp11-kit.dylib so
in the above command the source of the link resolves to nothing,
the destination becomes the source and the link to a non-existent
file is created in the working directory.
Daiki Ueno [Fri, 12 Aug 2016 12:27:46 +0000 (14:27 +0200)]
test: Make test-module work --without-trust-module
The test-module program currently depends on TRUST_PATHS, which is
determined by the configure script and normally points to a resource
outside of the build tree. To make the test system-independent, use
a crafted path for testing.