Fix issues like:
<inline asm>:1:1: error: unknown directive
.symver cil_build_policydb_pdb, cil_build_policydb@LIBSEPOL_1.0
Which was caused by the DISABLE_SYMVER define not being defined
for static, Mac or Android builds.
Acked-by: Joshua Brindle <joshua.brindle@crunchydata.com>
Signed-off-by: William Roberts <william.c.roberts@intel.com>
Currently a constraint `t1 == t2` gets converted to the invalid cil syntax `(mlsconstrain (class_name (perm_name)) (eq t1 ))` and fails to be loaded into the kernel.
Fixes: 893851c0a1 ("policycoreutils: add a HLL compiler to convert policy packages (.pp) to CIL")
Signed-off-by: Christian Göttsche <cgzones@googlemail.com>
Acked-by: James Carter <jwcart2@gmail.com>
The iteration over the set ebitmap bits is not implemented very
efficiently in libsepol. It is slowing down the policy optimization
quite significantly, so convert the type_map from an array of ebitmaps
to an array of simple ordered vectors, which can be traveresed more
easily. The worse space efficiency of the vectors is less important than
the speed in this case.
After this change the duration of semodule -BN decreased from 6.4s to
5.5s on Fedora Rawhide x86_64 (and from 6.1s to 5.6s with the unconfined
module disabled).
Signed-off-by: Ondrej Mosnacek <omosnace@redhat.com>
Acked-by: Stephen Smalley <stephen.smalley.work@gmail.com>
Only attributes can be a superset of another attribute, so we can skip
non-attributes right away.
Signed-off-by: Ondrej Mosnacek <omosnace@redhat.com>
Acked-by: Stephen Smalley <stephen.smalley.work@gmail.com>
I copy-pasted it from a different part of the code, which had to deal
with policydb that isn't final yet. Since we only deal with the final
kernel policy here, we can skip the check for the type datum being NULL.
Signed-off-by: Ondrej Mosnacek <omosnace@redhat.com>
Acked-by: Stephen Smalley <stephen.smalley.work@gmail.com>
With the old hidden_def and hidden_proto DSO infrastructure removed,
correctness of the map file becomes paramount, as it is what filters out
public API. Because of this, the wild cards should not be used, as it
lets some functions through that should not be made public API. Thus
remove the wild cards, and sort the list.
Additionally, verify that nothing changed in external symbols as well:
This was checked by generating an old export map (from master):
nm --defined-only -g ./src/libsepol.so | cut -d' ' -f 3-3 | grep -v '^_' > old.map
Then creating a new one for this library after this patch is applied:
nm --defined-only -g ./src/libsepol.so | cut -d' ' -f 3-3 | grep -v '^_' > new.map
And diffing them:
diff old.map new.map
Fixes: #165Fixes: #204
Acked-by: Stephen Smalley <sds@tycho.nsa.gov>
Signed-off-by: William Roberts <william.c.roberts@intel.com>
Add -fno-semantic-interposition to CFLAGS. This will restore
the DSO infrastructures protections to insure internal callers
of exported symbols call into libselinux and not something loading first
in the library list.
Clang has this enabled by default.
Acked-by: Stephen Smalley <sds@tycho.nsa.gov>
Signed-off-by: William Roberts <william.c.roberts@intel.com>
libsepol already has a linker script controlling it's exports, so this
patch has a net 0 affect, with the exception that internal callers of
external routines, which there could be 0 of, could potentially call a
non-libsepol routine depending on library load order.
NOTE A FEW SYMBOLS ARE EXPORTED THAT NORMALLY WOULDN'T BE
- sepol_context_to_sid
- sepol_ibendport_sid
- sepol_ibpkey_sid
- sepol_msg_default_handler
- sepol_node_sid
- sepol_port_sid
A subsequent map update will follow.
This list was generated by generating an old export map (from master):
nm --defined-only -g ./src/libsepol.so | cut -d' ' -f 3-3 | grep -v '^_' > old.map
Then creating a new one for this library after this patch is applied:
nm --defined-only -g ./src/libsepol.so | cut -d' ' -f 3-3 | grep -v '^_' > new.map
And diffing them:
diff old.map new.map
Acked-by: Stephen Smalley <sds@tycho.nsa.gov>
Signed-off-by: William Roberts <william.c.roberts@intel.com>
libsepol carried its own (outdated) copy of flask.h with the generated
security class and initial SID values for use by the policy
compiler and the forked copy of the security server code
leveraged by tools such as audit2why. Convert libsepol and
checkpolicy entirely to looking up class values from the policy,
remove the SECCLASS_* definitions from its flask.h header, and move
the header with its remaining initial SID definitions private to
libsepol. While we are here, fix the sepol_compute_sid() logic to
properly support features long since added to the policy and kernel,
although there are no users of it other than checkpolicy -d (debug)
and it is not exported to users of the shared library. There
are still some residual differences between the kernel logic and
libsepol.
Signed-off-by: Stephen Smalley <sds@tycho.nsa.gov>
Acked-by: Petr Lautrbach <plautrba@redhat.com>
The value attrs_expand_size == 1 removes all empty attributes, but it
also makes sense to expand all attributes that have only one type. This
removes some redundant rules (there is sometimes the same rule for the
type and the attribute) and reduces the number of attributes that the
kernel has to go through when looking up rules.
Signed-off-by: Ondrej Mosnacek <omosnace@redhat.com>
Acked-by: James Carter <jwcart2@gmail.com>
A parameter of a macro was only considered to be a duplicate if it
matched both the name and flavor of another parameter. While it is
true that CIL is able to differentiate between those two parameters,
there is no reason to use the same name for two macro parameters and
it is better to return an error for what is probably an error.
Remove the check of the flavors when checking for duplicate parameters.
Signed-off-by: James Carter <jwcart2@gmail.com>
Acked-by: Ondrej Mosnacek <omosnace@redhat.com>
Type transition file names are stored in a symbol table. Before the
name is added, the symbol table is searched to see if the name had
already been inserted. If it has, then the already existing datum is
returned. If it has not, then the name is added if either the
typetransition rule does not occur in a macro or the name is not one
of the macro parameters.
Checking for a previous insertion before checking if the name is a
macro parameter can cause a macro parameter to be treated as the
actual name if a previous type transition file name is the same as
the parameter.
Now check the name to see if it a macro paramter before checking for
its existence in the symbol table.
Signed-off-by: James Carter <jwcart2@gmail.com>
Acked-by: Ondrej Mosnacek <omosnace@redhat.com>
This reverts commit 542e878690.
After 6968ea9775 ("libsepol: make ebitmap_cardinality() of linear
complexity"), the caching only saves ~0.06 % of total semodule -BN
running time (on x86_64 without using the POPCNT instruction), so it's
no longer worth the added complexity.
Signed-off-by: Ondrej Mosnacek <omosnace@redhat.com>
Acked-by: Stephen Smalley <sds@tycho.nsa.gov>
Create the macro ebitmap_is_empty() to check if an ebitmap is empty.
Use ebitmap_is_empty(), instead of ebitmap_cardinality() or
ebitmap_length(), to check whether or not an ebitmap is empty.
Signed-off-by: James Carter <jwcart2@gmail.com>
Acked-by: Ondrej Mosnacek <omosnace@redhat.com>
As ebitmap_get_bit() complexity is linear in the size of the bitmap, the
complexity of ebitmap_cardinality() is quadratic. This can be optimized
by browsing the nodes of the bitmap directly in ebitmap_cardinality().
While at it, use built-in function __builtin_popcountll() to count the
ones in the 64-bit value n->map for each bitmap node. This seems better
suited than "count++". This seems to work on gcc and clang on x86,
x86_64, ARM and ARM64 but if it causes compatibility issues with some
compilers or architectures (or with older versions of gcc or clang),
the use of __builtin_popcountll() can be replaced by a C implementation
of a popcount algorithm.
Signed-off-by: Nicolas Iooss <nicolas.iooss@m4x.org>
Detect when the hashtab's load factor gets too high and try to grow it
and rehash it in such case. If the reallocation fails, just keep the
hashtab at its current size, since this is not a fatal error (it will
just be slower).
This speeds up semodule -BN on Fedora from ~8.9s to ~7.2s (1.7 seconds
saved).
Signed-off-by: Ondrej Mosnacek <omosnace@redhat.com>
hashtab_replace() and hashtab_map_remove_on_error() aren't used
anywhere, no need to keep them around...
Signed-off-by: Ondrej Mosnacek <omosnace@redhat.com>
Acked-by: Stephen Smalley <sds@tycho.nsa.gov>
According to profiling of semodule -BN, ebitmap_cardinality() is called
quite often and contributes a lot to the total runtime. Cache its result
in the ebitmap struct to reduce this overhead. The cached value is
invalidated on most modifying operations, but ebitmap_cardinality() is
usually called once the ebitmap doesn't change any more.
After this patch, the time to do 'semodule -BN' on Fedora Rawhide has
decreased from ~10.9s to ~8.9s (2s saved).
Signed-off-by: Ondrej Mosnacek <omosnace@redhat.com>
[sds@tycho.nsa.gov: correct times per follow-up on list]
Acked-by: Stephen Smalley <sds@tycho.nsa.gov>
The filename_- and range_trans_table ancillary hash tables in
cil_binary.c just duplicate the final policydb content and can be simply
removed.
Signed-off-by: Ondrej Mosnacek <omosnace@redhat.com>
The classperms associated with each map class permission and with each
classpermissionset are verified in __cil_verify_classperms() which had
multiple problems with how it did the verification.
1) Verification was short-circuited when the first normal class is found.
The second classpermissionset statement below would not have been
verified.
(classpermission cp1)
(classpermissionset cp1 (CLASS (PERM)))
(classpermissionset cp1 cp2)
2) The classperms of a map class permission and classpermissionset were
not checked for being NULL before the function recursively called itself.
This would result in a segfault if the missing map or set was referred to
before the classmap or classpermission occured. This error was reported by
Dominick Grift (dominick.grift@defensec.nl).
These rules would cause a segfault.
(classmap cm1 (mp1))
(classmapping cm1 mp1 (cm2 (mp2)))
(classmap cm2 (mp2))
But an error would be produced for these rules.
(classmap cm1 (mp1))
(classmap cm2 (mp2))
(classmapping cm2 mp2 (cm1 (mp1)))
3) The loop detection logic was incomplete and could only detect a loop
with a certain statement ordering.
These rules would cause a stack overflow.
(classmap cm1 (mp1))
(classmapping cm1 mp1 (cm2 (mp2)))
(classmap cm2 (mp2))
(classmapping cm2 mp2 (cm3 (mp3)))
(classmap cm3 (mp3))
(classmapping cm3 mp3 (cm2 (mp2)))
Rewrote __cil_verify_classperms() to fix these errors.
Signed-off-by: James Carter <jwcart2@tycho.nsa.gov>
Add support for new SELinux policy capability genfs_seclabel_symlinks.
With this capability enabled symlinks on kernel filesystems will receive
contexts based on genfscon statements, like directories and files,
and not be restricted to the respective filesystem root sid.
Signed-off-by: Christian Göttsche <cgzones@googlemail.com>
Remove restrictions in libsepol and checkpolicy that required all
declared initial SIDs to be assigned a context. With this patch,
it is possible to build and load a policy that drops the sid <sidname>
<context> declarations for the unused initial SIDs. It is still
required to retain the sid <sidname> declarations (in the flask
definitions) in order to preserve the initial SID ordering/values.
The unused initial SIDs can be renamed, e.g. to add an unused_
prefix or similar, if desired, since the names used in the policy
are not stored in the kernel binary policy.
In CIL policies, the (sid ...) and (sidorder (...)) statements
must be left intact for compatibility but the (sidcontext ...)
statements for the unused initial SIDs can be omitted after this change.
With current kernels, if one removes an unused initial SID context
from policy, builds policy with this change applied and loads the
policy into the kernel, cat /sys/fs/selinux/initial_contexts/<sidname>
will show the unlabeled context. With the kernel patch to remove unused
initial SIDs, the /sys/fs/selinux/initial_contexts/<sidname>
file will not be created for unused initial SIDs in the first place.
NB If an unused initial SID was assigned a context different from
the unlabeled context in existing policy, then it is not safe to
remove that initial SID context from policy and reload policy on
the running kernel that was booted with the original policy. This
is because that kernel may have assigned that SID to various kernel
objects already and those objects will then be treated as having
the unlabeled context after the removal. In refpolicy, examples
of such initial SIDs are the "fs" SID and the "sysctl" SID. Even
though these initial SIDs are not directly used (in code) by the current
kernel, their contexts are being applied to filesystems and sysctl files by
policy and therefore the SIDs are being assigned to objects.
NB The "sysctl" SID was in use by the kernel up until
commit 8e6c96935fcc1ed3dbebc96fddfef3f2f2395afc ("security/selinux:
fix /proc/sys/ labeling) circa v2.6.39. Removing its context from
policy will cause sysctl(2) or /proc/sys accesses to end up
performing permission checks against the unlabeled context and
likely encounter denials for kernels < 2.6.39.
Signed-off-by: Stephen Smalley <sds@tycho.nsa.gov>
Commit 4459d635b8 ("libsepol: Remove cil_mem_error_handler() function
pointer") replaced cil_mem_error_handler usage with inline contents of
the default handler. However, it left over the header declaration and
two callers. Convert these as well and remove the header declaration.
This also fixes a build failure with -fno-common.
Fixes: 4459d635b8 ("libsepol: Remove cil_mem_error_handler() function pointer")
Signed-off-by: Ondrej Mosnacek <omosnace@redhat.com>
GCC 10 comes with -fno-common enabled by default - fix the CIL_KEY_*
global variables to be defined only once in cil.c and declared in the
header file correctly with the 'extern' keyword, so that other units
including the file don't generate duplicate definitions.
Signed-off-by: Ondrej Mosnacek <omosnace@redhat.com>
When copying an avrule with extended permissions (permx) in
cil_copy_avrule(), the check for a named permx checks the new permx
instead of the old one, so the check will always fail. This leads to a
segfault when trying to copy a named permx because there will be an
attempt to copy the nonexistent permx struct instead of the name of
the named permx.
Check whether the original is a named permx instead of the new one.
Signed-off-by: James Carter <jwcart2@tycho.nsa.gov>
Since failing to resolve a statement in an optional block is normal,
only display messages about the statement failing to resolve and the
optional block being disabled at the highest verbosity level.
These messages are now only at log level CIL_INFO instead of CIL_WARN.
Signed-off-by: James Carter <jwcart2@tycho.nsa.gov>
There's a typo in commit b8213acff8 ("libsepol: add a function to optimize
kernel policy") which added new function sepol_policydb_optimize(), but there's
sepol_optimize_policy in libsepol.map.
LIBSEPOL_3.0 is used to follow the next release version libsepol-3.0
Signed-off-by: Petr Lautrbach <plautrba@redhat.com>
Acked-by: Ondrej Mosnacek <omosnace@redhat.com>
Acked-by: Stephen Smalley <sds@tycho.nsa.gov>
This improves commit b8213acf (libsepol: add a function to optimize
kernel policy) by Ondrej Mosnacek <omosnace@redhat.com> by always
removing redundant conditional rules which have an identical rule
in the unconditional policy.
Add a flag called not_cond to is_avrule_redundant(). When checking
unconditional rules against the avtab (which stores the unconditional
rules) we need to skip the actual rule that we are checking (otherwise
a rule would be determined to be redundant with itself and bad things
would happen), but when checking a conditional rule against the avtab
we do not want to skip an identical rule (which is what currently
happens), we want to remove the redundant permissions in the conditional
rule.
A couple of examples to illustrate when redundant condtional rules
are not removed.
Example 1
allow t1 t2:class1 perm1;
if (bool1) {
allow t1 t2:class1 perm1;
}
The conditional rule is clearly redundant, but without this change it
will not be removed, because of the check for an identical rule.
Example 2
typeattribute t1 a1;
allow t1 t2:class1 perm1;
allow a1 t2:class1 perm1;
if (bool1) {
allow t1 t2:class1 perm1;
}
The conditional rule is again clearly redundant, but now the order of
processing during the optimization will determine whether or not the
rule is removed. Because a1 contains only t1, a1 and t1 are considered
to be supersets of each other. If the rule with the attribute is
processed first, then it will be determined to be redundant and
removed, so the conditional rule will not be removed. But if the rule
with the type is processed first, then it will be removed and the
conditional rule will be determined to be redundant with the rule with
the attribute and removed as well.
The change reduces the size of policy a bit more than the original
optimization. Looking at the change in number of allow rules, there is
about a 10% improvement over the old optimization.
orig old new
Refpolicy 113284 82467 78053
Fedora 106410 64015 60008
Signed-off-by: James Carter <jwcart2@tycho.nsa.gov>
Several static analyzers (clang's one, Facebook Infer, etc.) warn about
NULL pointer dereferences after a call to CU_ASSERT_PTR_NOT_NULL_FATAL()
in the test code written using CUnit framework. This is because this
CUnit macro is too complex for them to understand that the pointer
cannot be NULL: it is translated to a call to CU_assertImplementation()
with an argument as TRUE in order to mean that the call is fatal if the
asserted condition failed (cf.
http://cunit.sourceforge.net/doxdocs/group__Framework.html).
A possible solution could consist in replacing the
CU_ASSERT_..._FATAL() calls by assert() ones, as most static analyzers
know about assert(). Nevertheless this seems to go against CUnit's API.
An alternative solution consists in overriding CU_ASSERT_..._FATAL()
macros in order to expand to assert() after a call to the matching
CU_ASSERT_...() non-fatal macro. This appears to work fine and to remove
many false-positive warnings from various static analyzers.
As this substitution should only occur when using static analyzer, put
it under #ifdef __CHECKER__, which is the macro used by sparse when
analyzing the Linux kernel.
Signed-off-by: Nicolas Iooss <nicolas.iooss@m4x.org>
Use codespell (https://github.com/codespell-project/codespell) in order
to find many common misspellings that are present in English texts.
Signed-off-by: Nicolas Iooss <nicolas.iooss@m4x.org>
In test_attr_types, the pointer decl is allowed to be NULL in the
beginning, but is dereferenced to produce a helpful message right before
a CU_ASSERT_FATAL. Make this derefence not happen if the pointer is
NULL.
This issue has been found using clang's static analyzer.
Signed-off-by: Nicolas Iooss <nicolas.iooss@m4x.org>
semodule-utils/semodule_link/semodule_link.c contains:
static sepol_module_package_t *load_module(char *filename)
{
/* ... */
if (sepol_module_package_create(&p)) {
/* ... */
goto bad;
/* ... */
bad:
sepol_module_package_free(p);
When sepol_module_package_create() fails while having successfully
allocated p, it currently frees p without setting it back to NULL. This
causes a use-after-free in load_module().
Prevent this use-after-free by setting sepol_module_package_create's
argument back to NULL when an error happens.
This issue has been found using Infer static analyzer.
Signed-off-by: Nicolas Iooss <nicolas.iooss@m4x.org>
Doing this looks wrong:
len = scope->decl_ids_len;
if (scope == NULL) {
/* ... */
Move the dereferencing of scope after the NULL check.
This issue has been found using Infer static analyzer.
Signed-off-by: Nicolas Iooss <nicolas.iooss@m4x.org>
When strs_stack_init(&stack) fails to allocate memory and stack is still
NULL, it should not be dereferenced with strs_stack_pop(stack).
This issue has been found using Infer static analyzer.
Signed-off-by: Nicolas Iooss <nicolas.iooss@m4x.org>
As reported by Nicolas Iooss (nicolas.iooss@m4x.org), static analyzers
have problems understanding that the default memory error handler does
not return since it is called through the cil_mem_error_handler()
function pointer. This results in a number of false positive warnings
about null pointer dereferencing.
Since the ability to set the cil_mem_error_handler() is only through
the function cil_set_mem_error_handler() which is never used and whose
definition is not in any header file, remove that function, remove the
use of cil_mem_error_handler() and directly in-line the contents of
the default handler, cil_default_mem_error_handler().
Signed-off-by: James Carter <jwcart2@tycho.nsa.gov>
This patch is loosely based on a patch by Yuli Khodorkovskiy
<yuli@crunchydata.com> from June 13th, 2019.
Since any permission used in the policy should be defined, CIL
should return an error if it cannot resolve a permission used
in a policy. This was the original behavior of CIL.
The behavior was changed over three commits from July to November
2016 (See commits 46e157b47, da51020d6, and 2eefb20d8). The change
was motivated by Fedora trying to remove permissions from its
policy that were never upstreamed (ex/ process ptrace_child and
capability2 compromise_kernel). Local or third party modules
compiled with those permissions would break policy updates.
After three years it seems unlikely that we need to worry about
those local and third party modules and it is time for CIL to
give an error like it should.
Signed-off-by: James Carter <jwcart2@tycho.nsa.gov>
Policy developers can set a default_range default to glblub and
computed contexts will be the intersection of the ranges of the
source and target contexts. This can be used by MLS userspace
object managers to find the range of clearances that two contexts
have in common. An example usage is computing a transition between
the network context and the context of a user logging into an MLS
application.
For example, one can add a default with
this cil:
(defaultrange db_table glblub)
or in te (base module only):
default_range db_table glblub;
and then test using the compute_create utility:
$ ./compute_create system_u:system_r:kernel_t:s0:c1,c2,c5-s0:c1.c20 system_u:system_r:kernel_t:s0:c0.c20-s0:c0.c36 db_table
system_u:object_r:kernel_t:s0:c1,c2,c5-s0:c1.c20
Some example range transitions are:
User Permitted Range | Network Device Label | Computed Label
---------------------|----------------------|----------------
s0-s1:c0.c12 | s0 | s0
s0-s1:c0.c12 | s0-s1:c0.c1023 | s0-s1:c0.c12
s0-s4:c0.c512 | s1-s1:c0.c1023 | s1-s1:c0.c512
s0-s15:c0,c2 | s4-s6:c0.c128 | s4-s6:c0,c2
s0-s4 | s2-s6 | s2-s4
s0-s4 | s5-s8 | INVALID
s5-s8 | s0-s4 | INVALID
Signed-off-by: Joshua Brindle <joshua.brindle@crunchydata.com>
Remove legacy local boolean and user code, and to preserve API/ABI
compatibility the following functions int values should be set to '0'
as they are no longer used:
selinux_mkload_policy(int preservebools)
security_set_boolean_list(.... int permanent)
and the following are now no-op and return '-1':
security_load_booleans()
sepol_genusers()
sepol_set_delusers()
sepol_genbools()
sepol_genbools_array()
and these still return their paths for compatibility, however they are
marked as deprecated:
selinux_booleans_path()
selinux_users_path()
These have been removed as they are local functions only:
sepol_genusers_policydb()
sepol_genbools_policydb()
Also "SETLOCALDEFS" removed from SELinux config file and code.
Signed-off-by: Richard Haines <richard_c_haines@btinternet.com>
In module_to_cil.c, sepol_ppfile_to_module_package() calls functions
from module.c without including the internal header. This makes building
libsepol with "gcc -flto -fuse-ld=gold" fails when linking libsepol.so:
/tmp/ccHYAKVZ.ltrans21.ltrans.o:<artificial>:function
sepol_ppfile_to_module_package: error: undefined reference to
'sepol_module_package_free'
/tmp/ccHYAKVZ.ltrans21.ltrans.o:<artificial>:function
sepol_ppfile_to_module_package: error: undefined reference to
'sepol_module_package_create'
/tmp/ccHYAKVZ.ltrans21.ltrans.o:<artificial>:function
sepol_ppfile_to_module_package: error: undefined reference to
'sepol_module_package_create'
collect2: error: ld returned 1 exit status
Fixes: https://github.com/SELinuxProject/selinux/issues/165
Signed-off-by: Nicolas Iooss <nicolas.iooss@m4x.org>
Add sepol_policydb_optimize(), which checks a kernel policy for
redundant rules (i.e. those that are covered by an existing more general
rule) and removes them.
Results on Fedora 29 policy:
WITHOUT OPTIMIZATION:
# time semodule -B
real 0m21,280s
user 0m18,636s
sys 0m2,525s
$ wc -c /sys/fs/selinux/policy
8692158 /sys/fs/selinux/policy
$ seinfo (edited)
Allow: 113159
Dontaudit: 10297
Total: 123156
WITH OPTIMIZATION ENABLED:
# time semodule -B
real 0m22,825s
user 0m20,178s
sys 0m2,520s
$ wc -c /sys/fs/selinux/policy
8096158 /sys/fs/selinux/policy
$ seinfo (edited)
Allow: 66334
Dontaudit: 7480
Total: 73814
Signed-off-by: Ondrej Mosnacek <omosnace@redhat.com>
