Kernel Development

From SELinux Wiki

(Difference between revisions)
Jump to: navigation, search
Revision as of 06:50, 2 December 2008 (edit)
JamesMorris (Talk | contribs)

← Previous diff		 Current revision (23:05, 20 April 2010) (edit) (undo)
JamesMorris (Talk | contribs)
(→To Do List)
 
(22 intermediate revisions not shown.)
Line 1: Line 1:
== To Do List == == To Do List ==
-* Handle setxattr() consistently on filesystems that do not support labeling.<sup>6</sup>+* Fix error propagation for policydb_read to reflect actual errors rather than the default of -EINVAL.
-* Remove ae.used field and address avc_has_perm* scalability.<sup>7</sup>+* Reduce SELinux memory footprint, both static and dynamic.
-* Audit for correctness when CONFIG_BUG=n.+* Improve scripts/selinux/mdp to generate a more useful minimal policy as a starting point. Also make it more flexible.
-* Add LTP tests for recent kernel changes.+* Dynamic discovery of initial SIDs, similar to dynamic discovery of classes/perms. Map kernel initial SIDs to policy initial SIDs by string name rather than requiring identical index values, handle unknown initial SIDs cleanly (map to unlabeled), and allow future extensibility without causing problems (start regular SIDs at some fixed offset, e.g. 100, or start from the highest legal value and decrement, so that policy reload that changes the number of initial SIDs won't affect them).
-* Remove all unnecessary secondary_ops calls that have no corresponding capability hook (legacy of old attempts to support other security modules as secondary)+* Distinguish access(2) from open(2) auditing (see
 +https://bugzilla.redhat.com/show_bug.cgi?id=495211).
-* Possibly convert all security_ops calls to direct cap_ calls for greater efficiency and clarity, although we would still retain secondary_ops so that we can reset security_ops in selinux_disable(), although technically that would be cleaner if handled by the framework rather than direct manipulation by selinux.+* Audit hooks to ensure that we don't have any more cases where DAC can be weakened.
-* the message inode_doinit_with_dentry: context_to_sid(blah:blah:bah) returned 22 is meaningless to all but 2 people in the world and isn't uncommon (and maybe printk_ratelimit() it+* Fix the performance issue apparently related to syscall audit which Linus keeps whining about.
-* avc_init() calls kmem_cache_create() but this is not freed if we call selinux_disable() from init+* Add support for SCTP (see https://bugzilla.redhat.com/show_bug.cgi?id=517676).
-* Possibly add a permission to allow multi-threaded dynamic transitions.<sup>5</sup>+* Fix signal inheritance controls (possibly drop some or all, or only enforce in policy for certain domains).
-* Reduce size of critical sections and use of GFP_ATOMIC.+* UBIFS support.
-* Remove load_mutex mutex. [patch queued]+* Add LTP tests for recent kernel changes.
-* Open code POLICY_RDLOCK and friends (per suggestion from akpm) [patches queued]+* Eliminate the need for secondary_ops altogether by providing LSM support for reverting to the original (capability) security module.
 + 
 +* Reduce size of critical sections and use of GFP_ATOMIC.
* Investigate security policy for cgroups. * Investigate security policy for cgroups.
- 
-* Labeling for loopback traffic (in progress HP). 
- 
-* Reduce memory usage of selinux structs: pahole (eparis RH BZ#235284) 
* Add a 'map' check on mmap and mprotect so that we can distinguish memory mapped access (since it has different implications for revocation). * Add a 'map' check on mmap and mprotect so that we can distinguish memory mapped access (since it has different implications for revocation).
- 
-* btrfs support (fs needs to be wired up to xattr code) 
* crfs ? * crfs ?
- 
-* cap_override class<sup>2</sup> (rfc patch posted, needs re-base and extension for 64-bit caps) 
- 
-* Compile out LSM hooks & allow SELinux to be linked directly. 
* Automate checking for new syscalls in kernels (-mm, -rc etc). * Automate checking for new syscalls in kernels (-mm, -rc etc).
-* remove secondary module stacking code (eparis RH BZ#231890)+* fine-grained labeling for usbfs and other pseudo filesystems of interest
- +
-* fine grained enforcement of sysfs objects (RH BZ#228902)+
- +
-* ditto for usbfs and other pseudo filesystems of interest+
- +
-* additional support of a security netfilter table for secmark/net forwarding (RH: merged to nf repo)+
- +
-* Namespacing of SELinux global functions and variables.<sup>4</sup>+
* NFSv4 support (in progress) * NFSv4 support (in progress)
- 
-* Linux hv controls (in progress Tresys?) 
* Revoke memory-mapped file access upon policy change or setxattr. * Revoke memory-mapped file access upon policy change or setxattr.
Line 68: Line 51:
* CIFS support for single-context clients (also has xattrs & Karl says it's better than NFS). * CIFS support for single-context clients (also has xattrs & Karl says it's better than NFS).
- 
-* Investigate integration with integrity measurement (in progress IBM and NSA) 
* Crypto policy for domains & object handling * Crypto policy for domains & object handling
Line 81: Line 62:
* move *mem permissions to new memprotect class. Bump policy version. * move *mem permissions to new memprotect class. Bump policy version.
-* discovery of class and permission offsets <sup>3</sup>+* better support for FS whose labelling behaviour is not specified in policy. If nothing from policy just test for xattr support and use it if it is there (RH in progress, patch reverted due to fuse deadlocks).
- +
-* better support for FS whose labelling behaviour is not specified in policy. If nothing from policy just test for xattr support and use it if it is there.+
- +
-''Notes:''+
-<p>+
-</p><p>+
-<sup>2</sup>Allow SELinux to selectively grant capabilities authoritatively based on SELinux domain. Executables could be made privileged w/o needing to be setuid root, all via SELinux without needing yet another mechanism like file capabilities. Eliminate the need for filesystem capabilities support (which will be a nightmare to manage, as they are per-file bitmaps vs. per-type access+
-vectors).</p>+
-<p>+
-<sup>3</sup> Make the hooks/avc layer request class/perm offsets from security server so that static offsets are no longer necessary and obsolete kernel classes can be purged.</p>+
-<p>+
-<sup>4</sup><pre>We have a lot of global functions and variables, some with no prefix at+
-all, some with prefixes that are not clearly scoped to selinux.+
- +
-I'd at least:+
-- rename the security server functions from security_ to selinux_ to+
-avoid confusion/conflicts with LSM.+
-- rename ss_initialized to selinux_ss_initialized.+
-- rename the policydb and sidtab variables to selinux_policydb and+
-selinux_sidtab (and/or wrap them in a single container structure with a+
-single active policy pointer to it, with the intent of ultimately+
-refcounting it and introducing _get and _put functions).+
- +
-Then you've got the generic data structures and their functions, like+
-hashtab_, symtab_, etc, which could either be taken to lib/ or given+
-selinux_ prefixes.</pre>+
-</p>+
-<p>+
-<sup>5</sup> We may wish to consider removing this restriction altogether, and/or+
-making it subject to a permission check. Per-thread context can be+
-useful for multi-threaded server effectively acting as a userspace+
-object manager but wanting to set kernel context to avoid race+
-conditions on file accesses, ala the samba file server case.+
-</p>+
-<p>+
-<sup>6</sup> Seems like we ought to handle setxattr() consistently on filesystems+
-that do not support labeling. Mountpoint labeling triggers EOPNOTSUPP.+
-Genfscon does not.+
- +
-Lack of a separate kernel interfaces for getxattr vs. getfilecon()+
-prevents us from probing for actual attribute on disk. Deferred context+
-support will let you see the raw context value from disk if present but+
-doesn't indicate absence of an attribute.+
-</p>+
-<p>+
-<sup>7</sup> [http://lkml.org/lkml/2008/11/17/351 lkml post]+
- +
-== Known Bugs ==+
- +
-== Done ==+
- +
-* Finalize NFS binary mount support: ensure new hooks are called.+
- +
-* Review ''Netlink link creation API'' code for security hook coverage.+
- +
-* Remove obsolete object backpointers.+
- +
-* Fix context_struct_compute_av latency issue raised by Ingo Molnar ([http://marc.info/?l=linux-kernel&m=118095653422494&w=2 lkml post])+
- +
-* Better support for sys_splice and related syscalls+
- +
-* change Kconfig to use select instead of depends (eparis RH BZ# 228899)+
- +
-* allow undefined classes and permissions in kernel (eparis RH BZ#235280)+
- +
-* explicitly set i_ino on all creations in selinuxfs (eparis RH BZ#235248)+
- +
-* Review sys_fallocate if/when it is merged+
- +
-* Labeling for forwarded traffic (done: HP)+
- +
-* security_file_permission callsite consolidation<sup>1</sup> (done: RH)+
- +
-* Add hook for filesystems with binary mount data per requests by fsdevel folk (done: RH)+
- +
-* add NFSv4 support for command line mount options. (done: RH)+
- +
-* Support for 64-bit capabilities (sds of the NSA)+
- +
-* Display LSM mount options in /proc/mounts (done: RH)+
- +
-* Permissive domains (done: RH).+
- +
-* printk prefixes and error message cleanup (done: RH)+
- +
-* open permission (done: RH)+
- +
-* security_port_sid optimization (done: HP, netport cache)+
- +
-* Normalize SELinux in-kernel API (obsolete: converted to LSM hooks)+
- +
-* Support for setting down unknown file contexts for package managers and filesystem restore (done: NSA, deferred mapping of contexts patch).+
-* Finer-grained proc checking so that we don't require full ptrace permission just to read process state (done: NSA, split proc ptrace checking into read vs. attach).+* memory leak detector pops on policy reload. probably due to {new,old}policydb being on stack and memcpy'd into the data section policydb (aka probably false positive)
-* Improve/fix ioctl checking (done: NSA, simplify ioctl checking).+* better validation of classes/perms on policy reload. Warn if any permissions are defined in a kernel class in the policy that are not defined in the kernel's classmap.
''Notes:'' ''Notes:''
<p> <p>
-<sup>1</sup> Provide a static inline helper for all FMODE_READ/FMODE_WRITE checks that also includes the corresponding security_file_permission() call to help ensure that they always happen together in the future. Possibly even rolling up rw_verify_area() checking as well into it. 
</p> </p>

Current revision

[edit] To Do List

  • Fix error propagation for policydb_read to reflect actual errors rather than the default of -EINVAL.
  • Reduce SELinux memory footprint, both static and dynamic.
  • Improve scripts/selinux/mdp to generate a more useful minimal policy as a starting point. Also make it more flexible.
  • Dynamic discovery of initial SIDs, similar to dynamic discovery of classes/perms. Map kernel initial SIDs to policy initial SIDs by string name rather than requiring identical index values, handle unknown initial SIDs cleanly (map to unlabeled), and allow future extensibility without causing problems (start regular SIDs at some fixed offset, e.g. 100, or start from the highest legal value and decrement, so that policy reload that changes the number of initial SIDs won't affect them).
  • Distinguish access(2) from open(2) auditing (see

https://bugzilla.redhat.com/show_bug.cgi?id=495211).

  • Audit hooks to ensure that we don't have any more cases where DAC can be weakened.
  • Fix the performance issue apparently related to syscall audit which Linus keeps whining about.
  • Fix signal inheritance controls (possibly drop some or all, or only enforce in policy for certain domains).
  • UBIFS support.
  • Add LTP tests for recent kernel changes.
  • Eliminate the need for secondary_ops altogether by providing LSM support for reverting to the original (capability) security module.
  • Reduce size of critical sections and use of GFP_ATOMIC.
  • Investigate security policy for cgroups.
  • Add a 'map' check on mmap and mprotect so that we can distinguish memory mapped access (since it has different implications for revocation).
  • crfs ?
  • Automate checking for new syscalls in kernels (-mm, -rc etc).
  • fine-grained labeling for usbfs and other pseudo filesystems of interest
  • NFSv4 support (in progress)
  • Revoke memory-mapped file access upon policy change or setxattr.
  • Real device labeling and access control (i.e. bind a label to a device in the kernel irrespective of what device node is used to access it so that a process that can create any device nodes at all can't effectively bypass all device access controls just by creating an arbitrary node to any device in a type accessible to it),
  • Full APIs for getting and setting security contexts of sockets and IPC objects. Ensure that socket context is kept consistent on socket inode and sock structures when changed.
  • Polyinstantiated ports
  • Increased granularity for Generic Netlink
  • CIFS support for single-context clients (also has xattrs & Karl says it's better than NFS).
  • Crypto policy for domains & object handling
  • Expand LTP as a full regression testuite for every permission & class
  • Redo performance testing & profiling
  • Better controls for posix message queues (?)
  • move *mem permissions to new memprotect class. Bump policy version.
  • better support for FS whose labelling behaviour is not specified in policy. If nothing from policy just test for xattr support and use it if it is there (RH in progress, patch reverted due to fuse deadlocks).
  • memory leak detector pops on policy reload. probably due to {new,old}policydb being on stack and memcpy'd into the data section policydb (aka probably false positive)
  • better validation of classes/perms on policy reload. Warn if any permissions are defined in a kernel class in the policy that are not defined in the kernel's classmap.

Notes:

[edit] Resources

Retrieved from "http://selinuxproject.org/page/Kernel_Development"
Personal tools