Description of problem:
Use of __thread variables is great for creating a thread-safe variable, but
only insofar as the contents of that variable can safely be abandoned on
pthread_exit(). The moment you store malloc()d data into a __thread void*
variable, you have leaked memory when the thread exits, since there is no way
to associate a destructor with __thread variables.
The _only_ safe way to use thread-local caching of malloc()d data is to use
pthread_key_create, and associate a destructor that will call free() on the
resulting data when the thread exits.
libselinux is guilty of abusing __thread variables to store malloc()d data as a
form of a cache, to minimize computation by reusing earlier results from the
same thread. As a result of this memory leak, repeated starting and stopping
of domains via libvirt can result in the OOM killer triggering, since libvirt
fires up a thread per domain, and each thread uses selinux calls such as
fgetfilecon.
Version-Release number of selected component (if applicable):
libselinux-2.0.94-2.el6.x86_64
libvirt-0.8.1-27.el6.x86_64
How reproducible:
100%
Steps to Reproduce:
0. These steps are run as root, assuming hardware kvm support and existence of
a VM named fedora (adjust the steps below as appropriate); if desired, I can
reduce this to a simpler test case that does not rely on libvirt, by using a
single .c file that links against libselinux and repeatedly spawns threads.
1. service libvirtd stop
2. valgrind --quiet --leak-check=full /usr/sbin/libvirtd& pid=$!
3. virsh start fedora
4. kill $pid
Actual results:
The biggest leak reported is due to libselinux' abuse of __thread:
==26696== 829,730 (40 direct, 829,690 indirect) bytes in 1 blocks are
definitely lost in loss record 500 of 500
==26696== at 0x4A0515D: malloc (vg_replace_malloc.c:195)
==26696== by 0x3022E0D48C: selabel_open (label.c:165)
==26696== by 0x3022E11646: matchpathcon_init_prefix (matchpathcon.c:296)
==26696== by 0x3022E1190D: matchpathcon (matchpathcon.c:317)
==26696== by 0x3033ED7FB5: SELinuxRestoreSecurityFileLabel (security_selinux.c:381)
==26696== by 0x3033ED8539: SELinuxRestoreSecurityAllLabel (security_selinux.c:749)
==26696== by 0x459153: qemuSecurityStackedRestoreSecurityAllLabel (qemu_security_stacked.c:257)
==26696== by 0x43F0C5: qemudShutdownVMDaemon (qemu_driver.c:4311)
==26696== by 0x4555C9: qemudStartVMDaemon (qemu_driver.c:4234)
==26696== by 0x458416: qemudDomainObjStart (qemu_driver.c:7268)
==26696== by 0x45896F: qemudDomainStart (qemu_driver.c:7308)
==26696== by 0x3033E75412: virDomainCreate (libvirt.c:4881)
==26696==
Basically, libvirt created a thread that used matchpathcon during 'virsh start
fedora', and matchpathcon stuffed over 800k of malloc'd data into:
static __thread char **con_array;
which are then inaccessible when libvirt exits the thread as part of shutting
down on SIGTERM.
Expected results:
valgrind should not report any memory leaks related to libselinux.
Signed-off-by: Eamon Walsh <ewalsh@tycho.nsa.gov>
Reported-by: Eric Blake <eblake@redhat.com>
Tested-by: Eric Blake <eblake@redhat.com>
The majority of the patch is just handling the case of memory
allocation failures and making sure things get cleaned up correctly in
those cases.
This also moves duplicate code in parse_ebitmap() and parse_raw() into
parse_category(), and also updates the parse function to ensure the
config files are in the correct format.
Signed-off-by: Steve Lawrence <slawrence@tresys.com>
SELinux Project contribution of mcstrans. mcstrans is a userland package
specific to SELinux which allows system administrators to define
sensitivity levels and categories and provides a daemon for their
translation into human readable form. This version is a merge of Joe
Nalls git tree ( http://github.com/joenall/mcstrans) and patches
supplied by Dan Walsh and others at RedHat.
Ted
Signed-off-by: Steve Lawrence <slawrence@tresys.com>
Email: kaigai@ak.jp.nec.com
Subject: libselinux APIs should take "const" qualifier?
Date: Tue, 23 Mar 2010 11:56:36 +0900
(2010/03/19 22:32), Stephen Smalley wrote:
> On Fri, 2010-03-19 at 16:52 +0900, KaiGai Kohei wrote:
>> Right now, security_context_t is an alias of char *, declared in selinux.h.
>>
>> Various kind of libselinux API takes security_context_t arguments,
>> however, it is inconvenience in several situations.
>>
>> For example, the following query is parsed, then delivered to access
>> control subsystem with the security context as "const char *" cstring.
>>
>> ALTER TABLE my_tbl SECURITY LABEL TO 'system_u:object_r:sepgsql_table_t:SystemHigh';
>> const char *<---- ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
>>
>> In this case, we want to call selinux_trans_to_raw_context() to translate
>> the given security context into raw format. But it takes security_context_t
>> argument for the source context, although this pointer is read-only.
>> In the result, compiler raises warnings because we gave "const char *" pointer
>> into functions which take security_context_t (= char *).
>>
>> Any comments?
>>
>> It seems to me the following functions' prototype should be qualified by
>> "const".
>
> That seems reasonable and should have no impact on library ABI.
> On the other hand, others have pointed out that security_context_t is
> not a properly encapsulated data type at all, and perhaps should be
> deprecated and replaced with direct use of char*/const char* throughout.
>
> There are other library API issues as well that have come up in the
> past, such as lack of adequate namespacing (with approaches put forth),
> but we don't ever seem to get a round tuit.
At first, I tried to add const qualifiers read-only security_context_t
pointers, but didn't replace them by char */const char * yet, right now.
BTW, I could find out the following code:
int security_compute_create(security_context_t scon,
security_context_t tcon,
security_class_t tclass,
security_context_t * newcon)
{
int ret;
security_context_t rscon = scon;
security_context_t rtcon = tcon;
security_context_t rnewcon;
if (selinux_trans_to_raw_context(scon, &rscon))
return -1;
if (selinux_trans_to_raw_context(tcon, &rtcon)) {
freecon(rscon);
return -1;
}
:
In this case, scon and tcon can be qualified by const, and the first
argument of selinux_trans_to_raw_context() can take const pointer.
But it tries to initialize rscon and tscon by const pointer, although
these are used to store raw security contexts.
The selinux_trans_to_raw_context() always set dynamically allocated
text string on the second argument, so we don't need to initialize it
anyway. I also removed these initializations in this patch.
Does the older mcstrans code could return without allocation of raw
format when the given scon is already raw format? I don't know why
these are initialized in this manner.
Thanks.
--
KaiGai Kohei <kaigai@ak.jp.nec.com>
Signed-off-by: Chad Sellers <csellers@tresys.com>
Email: slawrence@tresys.com
Subject: Minor fixup of checkmodule man page.
Date: Fri, 11 Jun 2010 15:25:58 -0400
On Mon, 2010-05-03 at 13:45 -0400, Daniel J Walsh wrote:
> Quality Engineering is going through all commands on the system looking
> for mismatches between man page/usage and actual code.
>
> It found that checkmodule had a -d option that is unused and undocumented -h
Reviewed-by: Steve Lawrence <slawrence@tresys.com>
I'd just add the long --help option to the man page for completeness:
Signed-off-by: Chad Sellers <csellers@tresys.com>
Email: dwalsh@redhat.com
Subject: Minor fixup of checkmodule man page.
Date: Mon, 03 May 2010 13:45:30 -0400
-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1
Quality Engineering is going through all commands on the system looking
for mismatches between man page/usage and actual code.
It found that checkmodule had a -d option that is unused and undocumented -h
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v2.0.14 (GNU/Linux)
Comment: Using GnuPG with Fedora - http://enigmail.mozdev.org/
iEYEARECAAYFAkvfC7oACgkQrlYvE4MpobNPrACg0uP02CWYPs9YcdU87jts9YqT
hMAAn2QA1UWZpGLvvU4yxStmhUU1Kg1+
=topF
-----END PGP SIGNATURE-----
Signed-off-by: Chad Sellers <csellers@tresys.com>
Email: slawrence@tresys.com
Subject: Updated sandbox patch.
Date: Mon, 07 Jun 2010 17:53:41 -0400
On Thu, 2010-05-27 at 08:57 -0400, Daniel J Walsh wrote:
> -----BEGIN PGP SIGNED MESSAGE-----
> Hash: SHA1
>
> On 05/26/2010 04:06 PM, Steve Lawrence wrote:
> > On Wed, 2010-05-19 at 15:59 -0400, Daniel J Walsh wrote:
> > Fixed patch that handles Spaces in homedir.
>
> > The following patch makes a few updates the the sandbox patch, though I
> > have a question:
>
> > Is the sandbox.init script needed anymore? It looks like seunshare was
> > changed to now bind mount and make private the necessary directories.
> > The only thing that seems missing is making root rshared. Also, if the
> > init script is obsolete, do the mounts also need the MS_REC flag for
> > recursive bind/private like they are mounted in the init script? e.g.
>
> The init script is needed for the xguest package/more specifically
> pam_namespace, but also needed for
> mount --make-rshared /
>
> Whether the init script belongs in policycoreutils is questionable though.
>
>
> > mount(dst, dst, NULL, (MS_BIND | MS_REC), NULL)
> > mount(dst, dst, NULL, (MS_PRIVATE | MS_REC), NULL)
>
> We probably should add these. Although it is not likely.
>
> > Changes the following patch makes:
>
> > sandbox.py
> > - Removes unused 'import commands'
> > - Fixes the chcon function, and replaces the deprecated os.path.walk
> > with os.walk. I think this way is a bit easier to read too.
>
> I think chcon should be added to libselinux python bindings and then
> leave the recursive flag. (restorecon is currently in python bindings._
>
> > - Removes the 'yum install seunshare' message. This tool is not specific
> > to RPM based distros.
>
> People are using seunshare without X now that I have added the -M flag.
> So I will move it from the -gui package to the base package with
> sandbox and then this should not be necessary.
> > - Remove try/except around -I include to be consistent with the -i
> > option. If we can't include a file, then this should bail, no matter
> > if it's being included via -i or -I.
>
> Ok, I was thinking you could list a whole bunch of files in the -I case
> and if one does not exist, allow it to continue. But I don't really care.
> > - Fix homedir/tmpdir typo in chcon call
>
> > sandbox.init (maybe obsoleted?)
> > - Fix restart so it stops and starts
> > - unmount the bind mounts when stopped
> I doubt this will work. Two many locks in /tmp /home
> > - Abort with failure if any mounts fail
>
> > seunshare.c
> > - Define the mount flag MS_PRIVATE if it isn't already. The flag is only
> > defined in the latest glibc but has been in the kernel since 2005.
> > - Simplify an if-statment. Also, I'm not sure the purpose of the
> > strncmmp in that conditional, so maybe I've oversimplified.
> This is wrong. The problem comes about when you mount within the same
> directory.
>
> seunshare -t /home/dwalsh/sanbox/tmp -h /home/dwalsh/sandbox/home ...
>
> seunshare -t /tmp/sandbox/tmp -h /tmp/sandbox/home
>
> If you do not have the check one of the above will fail.
>
> In the first example if Homedir is mounted first,
> /home/dwalsh/sanbox/tmp will no longer exist when seunshare attempts to
> mount it on /tmp.
>
> Similarly, if /tmp is mounted first in the second example.
> /tmp/sandbox/home will no longer exist.
>
> You have to check to make sure one of the directories is not included in
> the other.
>
> It seems
> > like maybe an error should be thrown if tmpdir_s == pw_dir or
> > homedir_s == "/tmp", but maybe I'm missing something.
>
> See above.
>
> I was blowing up because I use
>
> ~/sandbox/tmp and ~/sandbox/home for my mountpoints.
<snip>
Below is an updated patch that makes a few changes the the latest
Sandbox Patch [1]. This requires the chcon patch [2].
Changes this patch makes:
sandbox.py
- Remove unused 'import commands'
- Uses new chcon method in libselinux [2]
- Removes the 'yum install seunshare' message
- Converts an IOError to a string for printing a warning if a file
listed in -I does not exist
sandbox.init
- Print the standard Starting/Stoping messages with the appropriate
OK/FAIL
- Abort with failure if any mounts fail
seunshare.c
- Add the MS_REC flag during mounts to perform recursive mounts
- Define the mount flags MS_PRIVATE and MS_REC if they aren't already.
The flags are only defined in the latest glibc but have been in the
kernel since 2005.
- Calls realpath(3) on tmpdir_s and homedir_s. If relative paths are
used, it wouldn't correctly detect that tmpdir is inside homedir and
change the mount order. This fixes that.
[1] http://marc.info/?l=selinux&m=127429948731841&w=2
[2] http://marc.info/?l=selinux&m=127594712200878&w=2
Signed-off-by: Chad Sellers <csellers@tresys.com>
Email: slawrence@tresys.com
Subject: Add chcon method to libselinux python bindings
Date: Mon, 7 Jun 2010 17:40:05 -0400
Adds a chcon method to the libselinux python bindings to change the
context of a file/directory tree.
Signed-off-by: Chad Sellers <csellers@tresys.com>
This patch simply removes duplicate slashes (meaning "//") from
pathnames passed into selabel_lookup. It does not do a full
realpath() calculation (e.g. following symlinks, etc.), as the
client should really do that before calling into libselinux.
Signed-off-by: Chad Sellers <csellers@tresys.com>
This test must have been disabled a very long time ago, before attributes were present in the kernel policy. Since the attributes are now present this unit test should be turned back on, unless I'm missing something pretty major (it looks reasonable and is successful when run).
Signed-off-by: Joshua Brindle <jbrindle@tresys.com>
Sepolgen has long not recovered from parsing errors, leading to
a blacklist of none bad modules in the source. I finally tracked
down the problem (lexer state) and this patch fixes the problem
by causing the lexer to be rebuilt on error.
Acked-by: Joshua Brindle <jbrindle@tresys.com>
for the given database object identified by its name and object class.
It is necessary to implement a feature something like the restorecon on databases.
The specfile shall be described as follows:
------------------------
#
# The specfile for database objects
# (for SE-PostgreSQL)
#
# <object class> <object name> <security context>
#
db_database * system_u:object_r:sepgsql_db_t:s0
db_schema *.pg_catalog system_u:obejct_r:sepgsql_sys_schema_t:s0
db_schema *.* system_u:object_r:sepgsql_schema_t:s0
db_table *.pg_catalog.* system_u:object_r:sepgsql_sysobj_t:s0
db_table *.*.* system_u:object_r:sepgsql_table_t:s0
------------------------
- All the characters after the '#' are ignored.
- Wildcards ('*' and '?') are available.
- It returns the first match security context.
Note that hierarchy of the namespace of database objects depends on RDBMS.
So, author of the specfile needs to write correct patterns which are suitable
for the target RDBMS. The patched selabel_*() interfaces don't have any
heuristics for the namespace hierarchy to be suitable for widespread RDBMSs.
In the case of SE-PgSQL, when we lookup an expected security context for the
'my_table' table in the 'public' schema and 'postgres' database, the caller
shall provide 'postgres.public.my_table' as a key.
In the default, it tries to read a specfile which maps database objects and security
context from the /etc/selinux/$POLICYTYPE/contexts/sepgsql_contexts.
Note that when another RDBMS uses this interface, it needs to give an explicit
SELABEL_OPT_PATH option on the selabel_open().
Signed-off-by: KaiGai Kohei <kaigai@ak.jp.nec.com>
Acked-by: Eamon Walsh <ewalsh@tycho.nsa.gov>
On 03/08/2010 11:11 AM, Karl MacMillan wrote:
> Accidentally sent this straight to Josh.
>
> Karl
>
> On Thu, Mar 4, 2010 at 4:46 PM, Karl MacMillan<karlwmacmillan@gmail.com> wrote:
>
>> I meant this - I don't want to pass around a boolean flag when we have
>> a flag for rule type. This allows cleanly adding support for, say,
>> generating both allow rules and auditallow rules at the same time.
>>
>>
<snip>
Ok this one only adds a flag to the policygenerator to tell it to
generate dontaudit rules.
No passing of args.
Acked-by: Karl MacMillan <karlwmacmillan@gmail.com>
avc_open() creates the netlink socket in nonblocking mode. If the
application later takes control of the netlink socket with
avc_netlink_acquire_fd() and then calls avc_netlink_loop(), it
will fail with EWOULDBLOCK.
To remedy this, remove the O_NONBLOCK flag from the netlink socket
at the start of avc_netlink_loop(). Also, with this fix, there is
no need for avc_open() to ever create a blocking socket, so change
that and update the man page.
-v2: use poll() in avc_netlink_check_nb(). This makes both
avc_netlink_loop() and avc_netlink_check_nb() independent of the
O_NONBLOCK flag.
-v3: move poll() to avc_receive() internal function; patch by
KaiGai Kohei <kaigai@kaigai.gr.jp>
Signed-off-by: Eamon Walsh <ewalsh@tycho.nsa.gov>
On 02/24/2010 02:24 PM, Daniel J Walsh wrote:
>
Ignore the first patch it was missing pc.in files.
Acked-by: Eamon Walsh <ewalsh@tycho.nsa.gov>
Signed-off-by: Joshua Brindle <method@manicmethod.com>
Email: dwalsh@redhat.com
Subject: Fix memory leak on disabled selinux machines.
Date: Wed, 24 Feb 2010 14:15:31 -0500
I think this patch originally came from Eric Paris and was updated by
others but has not been adopted yet. Not sure why.
Always free buf on exit.
Signed-off-by: Joshua Brindle <method@manicmethod.com>
I want to change the default of libsemanage to not look for home
directories in getpwent. This patch allows you to set the flag
usepasswd=false in the semanage.conf file. and genhomedircon will only
setup the labeling of /home, /export/home and any confined users homedirs.
If this patch is not acceptable because libsemanage is being rewritten,
I would like the functionality to be added to the new libsemanage.
On Mon, 2010-02-15 at 14:19 -0800, Justin Mattock wrote:
> this is new:
>
>
> make[2]: Leaving directory `/home/kernel/selinux/libselinux/include'
> make -C src install
> make[2]: Entering directory `/home/kernel/selinux/libselinux/src'
> cc -Werror -Wall -W -Wundef -Wshadow -Wmissing-noreturn
> -Wmissing-format-attribute -I../include -I/usr/include -D_GNU_SOURCE
> -D_FILE_OFFSET_BITS=64 -c -o label_file.o label_file.c
> cc1: warnings being treated as errors
> label_file.c: In function 'init':
> label_file.c:434: error: implicit declaration of function 'fstat'
> label_file.c:436: error: implicit declaration of function 'S_ISREG'
> make[2]: *** [label_file.o] Error 1
> make[2]: Leaving directory `/home/kernel/selinux/libselinux/src'
> make[1]: *** [install] Error 2
> make[1]: Leaving directory `/home/kernel/selinux/libselinux'
> make: *** [install] Error 1
>
> three areas where this could of been created
> update glibc
> updated kernel
> update userspace(altohugh there was not vary many commits in the pull).
Newer glibc headers expose a failure to #include the required headers
for stat(2). Also exposes a conflict in redefining close() in that
file. Patch below should fix.
Only audit the permissions specified by the policy, excluding any
permissions specified via dontaudit or not specified via auditallow.
This only shows up when a single avc_has_perm() call is made with
multiple permissions where some of those permissions are dontaudit'd or
auditallow'd while others are not. The corresponding kernel patch has
already been applied, see:
http://git.kernel.org/?p=linux/kernel/git/jmorris/security-testing-2.6.git;a=commit;h=b6cac5a30b325e14cda425670bb3568d3cad0aa8
Signed-off-by: Stephen D. Smalley <sds@tycho.nsa.gov>
