1 |
Notes for TOMOYO Linux project |
2 |
|
3 |
This is a handy Mandatory Access Control patch for Linux kernels. |
4 |
This patch is released under the GPLv2. |
5 |
|
6 |
Project URL: http://tomoyo.sourceforge.jp/ |
7 |
|
8 |
The authors of this patch (hereafter, we) don't have much experience |
9 |
in kernel programming. We are worried that this patch would contain |
10 |
some mistakes such as missing hooks, improper location of hooks, |
11 |
potential deadlocks. There would be better way of implementation. |
12 |
All kinds of comments, pointing the errors and suggestions are welcome. |
13 |
|
14 |
We do hope this patch reduces the labor of server security management |
15 |
and you enjoy the life with Linux. |
16 |
|
17 |
This project was very inspired by the comic "Card Captor SAKURA", |
18 |
one of the CLAMP's masterworks. |
19 |
|
20 |
ChangeLog: |
21 |
|
22 |
Version 1.0 2005/11/11 First release. |
23 |
|
24 |
Fix 2005/11/18 |
25 |
|
26 |
@ Add setattr() missing hook in SYAORAN fs. |
27 |
|
28 |
setattr() checking for special inode was missing. |
29 |
|
30 |
Fix 2005/11/25 |
31 |
|
32 |
@ Allow initrd.img include /sbin/init . |
33 |
|
34 |
Since version 1.0 loads policy when /sbin/init is called |
35 |
for the first time, initrd.img without the policy directory |
36 |
mustn't start /sbin/init . This forced users not to use |
37 |
initrd.img that includes /sbin/init . |
38 |
I modified to delay loading policy if the policy directory |
39 |
doesn't exist and wait for /sbin/init being called again. |
40 |
|
41 |
Fix 2005/12/02 |
42 |
|
43 |
@ Use lookup_one_len() instead of lookup_hash(). |
44 |
|
45 |
Kernel 2.6.15 changed parameters for lookup_hash(). |
46 |
I modified to use lookup_one_len() to keep compatibility. |
47 |
|
48 |
Fix 2005/12/06 |
49 |
|
50 |
@ Add S_ISDIR() check in SYAORAN fs. |
51 |
|
52 |
Malicious configuration file that attempts to create an inode |
53 |
under non-directory inode caused segmentation fault. |
54 |
|
55 |
Version 1.0.1 2005/12/08 Minor update release. |
56 |
|
57 |
Fix 2006/01/04 |
58 |
|
59 |
@ Add CheckWritePermission() check in unix_bind(). |
60 |
|
61 |
I modified to check write permission in unix_bind(), for |
62 |
sys_mknod(S_IFSOCK) checks write permission. |
63 |
|
64 |
@ Show hook version in proc_misc_init(). |
65 |
|
66 |
The hook part of this patch depends on the kernel's version, |
67 |
while the rest part of this patch doesn't. |
68 |
I added the hook version so that the administrator can |
69 |
know the last modified date of the hooks. |
70 |
|
71 |
@ Move permission checks from filp_open() to open_namei(). |
72 |
|
73 |
I moved the location of checking MAC's permission |
74 |
from filp_open() to open_namei(). |
75 |
|
76 |
@ Fix an error in filp_open(). (only 2.6.15-rc5) |
77 |
|
78 |
This error was only in the patch 2.6.15-rc5 and |
79 |
was fixed in the patch for 2.6.15. |
80 |
|
81 |
Fix 2006/01/12 |
82 |
|
83 |
@ Add /proc/ccs/info/self_domain. |
84 |
|
85 |
I added /proc/ccs/info/self_domain so that the userland programs |
86 |
can know the name of domain they belong to if necessary. |
87 |
|
88 |
Fix 2006/01/13 |
89 |
|
90 |
@ Merge constants for CheckTaskCapability(). |
91 |
|
92 |
I merged *_INHERITABLE_* and *_LOCAL_* to avoid always |
93 |
calling CheckTaskCapability() with both constants. |
94 |
|
95 |
@ DropTaskCapability() returns -EAGAIN on success. |
96 |
|
97 |
DropTaskCapability() must not return 0 on success, for |
98 |
DropTaskCapability() is called from do_execve(). |
99 |
|
100 |
@ Fix an error for chroot() permission check. |
101 |
|
102 |
The chroot() restriction was not working due to the following mistake. |
103 |
CheckChRootPermission() || CheckTaskCapability() returns 0 or 1, while |
104 |
CheckChRootPermission() | CheckTaskCapability() returns 0 or -EPERM. |
105 |
|
106 |
Fix 2006/01/17 |
107 |
|
108 |
@ Suppress some of debug messages in TOMOYO. |
109 |
|
110 |
I added KERN_DEBUG to suppress some of debug messages. |
111 |
|
112 |
Fix 2006/01/19 |
113 |
|
114 |
@ Remove isRoot() checks in AddChrootACL() and AddMountACL(). |
115 |
|
116 |
I found a program that needs to chroot by non-root. |
117 |
So, I stopped checking uid=euid=0 for these functions so that |
118 |
"accept mode" can append ACLs. |
119 |
The isRoot() is checked at AddChrootPolicy() and AddMountPolicy(). |
120 |
|
121 |
@ Map NULL device name to "<NULL>" in AddMountACL(). |
122 |
|
123 |
VMware mounts vmware-hgfs with NULL device name. |
124 |
So I mapped NULL device name to "<NULL>". |
125 |
|
126 |
Fix 2006/01/20 |
127 |
|
128 |
@ Suppress some of debug messages in SAKURA. |
129 |
|
130 |
I added KERN_DEBUG to suppress some of debug messages. |
131 |
|
132 |
@ Call panic() if failed to load given profile. |
133 |
|
134 |
Call panic() if profile index was given via CCS= parameter |
135 |
but the profile doesn't exist. |
136 |
If CCS= parameter is not given, the kernel attempts to load |
137 |
profile 0, but it doesn't call panic() if profile 0 doesn't exist. |
138 |
|
139 |
Fix 2006/01/24 |
140 |
|
141 |
@ Use full_name_hash() for IsGloballyReadableFile(). |
142 |
|
143 |
I modified to use full_name_hash() for faster scan. |
144 |
|
145 |
@ Add signal checking condition in CheckSignalACL(). |
146 |
|
147 |
The documentation says "if the target domain's domainname |
148 |
starts with the source domain's domainname, it is always granted" |
149 |
but actually it isn't. I'll change the documentation instead of |
150 |
changing the source code. |
151 |
|
152 |
Also, checking for pid = -1 was missing. This error was fixed. |
153 |
|
154 |
Fix 2006/02/09 |
155 |
|
156 |
@ Use mutex_lock()/mutex_unlock instead of down()/up(). |
157 |
|
158 |
Kernel 2.6.16 changed members of "struct inode". |
159 |
I modified to use mutex_lock()/mutex_unlock() for after 2.6.16 |
160 |
and down()/up() for before 2.6.16. |
161 |
|
162 |
Version 1.0.2 2006/02/14 Many bug-fixes release. |
163 |
|
164 |
Fix 2006/02/21 |
165 |
|
166 |
@ Divide generic-write permission into individual write permissions. |
167 |
|
168 |
Write permission was divided into the following permissions. |
169 |
|
170 |
'mkdir' for creating directory. |
171 |
'rmdir' for deleting directory. |
172 |
'create' for creating regular file. |
173 |
'unlink' for deleting non-directory. |
174 |
'mksock' for creating UNIX domain socket. |
175 |
'mkfifo' for creating FIFO. |
176 |
'mkchar' for creating character device. |
177 |
'mkblock' for creating block device. |
178 |
'link' for creating hard link. |
179 |
'symlink' for creating symbolic link. |
180 |
'rename' for renaming directory or non-directory. |
181 |
'truncate' for truncating regular file. |
182 |
|
183 |
The permission check for opening files is done using |
184 |
conventional read/write/execute permission. |
185 |
|
186 |
@ Add /proc/ccs/info/mapping. |
187 |
|
188 |
I added /proc/ccs/info/mapping so that the userland programs |
189 |
can know the mapping of individual write permissions. |
190 |
|
191 |
Fix 2006/02/27 |
192 |
|
193 |
@ Fix handling of trailing '\*' in PathMatchesToPattern(). |
194 |
|
195 |
PathMatchesToPattern("/tmp/", "/tmp/\*") returned true |
196 |
because "\*" matches "zero or more repetitions of characters |
197 |
until '/' or end". But since this is a comparison between |
198 |
directory and non-directory, this should not match. |
199 |
|
200 |
This behavior causes the following security risks. |
201 |
In enforce mode, allowing "2 /tmp/\*" grants |
202 |
"mkdir /tmp/" and "rmdir /tmp/" which should be |
203 |
granted only when "2 /tmp/" is allowed. |
204 |
In accept mode, "mkdir /tmp/" or "rmdir /tmp/" appends |
205 |
"2 /tmp/\*" into the domain policy if "file_pattern /tmp/\*" |
206 |
is in the exception policy. |
207 |
|
208 |
I changed not to ignore trailing '\*' in the pattern |
209 |
if pathname ends with '/'. |
210 |
|
211 |
Fix 2006/03/01 |
212 |
|
213 |
@ Add missing spinlock in GetAbsolutePath(). |
214 |
|
215 |
vfsmount_lock was missing. |
216 |
|
217 |
Fix 2006/03/08 |
218 |
|
219 |
@ Add support for "shared subtree" mount operations. |
220 |
|
221 |
Kernel 2.6.15 introduced "shared subtree" functionality. |
222 |
But CheckMountPermission() couldn't recognize flags for |
223 |
do_change_type(). |
224 |
|
225 |
@ Add support for more mount flags. |
226 |
|
227 |
atime/noatime, diratime/nodiratime, recurse/norecurse flags |
228 |
are supported. |
229 |
|
230 |
Fix 2006/03/20 |
231 |
|
232 |
@ Check port numbers for only AF_INET/AF_INET6. |
233 |
|
234 |
CheckBindEntry() and CheckConnectEntry() should check port numbers |
235 |
only when the given address family is either AF_INET or AF_INET6, |
236 |
for address family such as AF_UNSPEC could be passed to bind() |
237 |
and connect() for PF_INET/PF_INET6 sockets. |
238 |
|
239 |
Fix 2006/03/27 |
240 |
|
241 |
@ Use /proc/self/ rather than /proc/\$/ for current process. |
242 |
|
243 |
GetAbsolutePath() now uses "self" instead of pid |
244 |
if current process refers to information related to itself. |
245 |
This exception violates the rule "TOMOYO Linux's pathnames don't |
246 |
contain symbolic links before the last '/'", but I think it worth |
247 |
to do so. The following are the merits gained by this exception. |
248 |
|
249 |
Prevent administrators from granting redundant permissions |
250 |
when a process needs to refer to only current process's information. |
251 |
|
252 |
Allow administrators make current process's information always |
253 |
readable using 'allow_read' directive. |
254 |
|
255 |
Version 1.1 2006/04/01 Functionality enhancement release. |
256 |
|
257 |
Fix 2006/04/03 |
258 |
|
259 |
@ Use queue instead of fixed sized array for audit log. |
260 |
|
261 |
WriteAuditLog() now uses queue to save statically allocated memory. |
262 |
Administrators can give any size for audit logs at runtime. |
263 |
|
264 |
@ Use kzalloc() instead of kmalloc() + memset(). |
265 |
|
266 |
kmalloc() + memset() were replaced with kzalloc(). |
267 |
|
268 |
Fix 2006/04/04 |
269 |
|
270 |
@ Support "delayed enforcing" mode. |
271 |
|
272 |
Until now, access request was immediately rejected |
273 |
if policy doesn't allow that access and the system is |
274 |
running in enforce mode. |
275 |
Sometimes, especially after updating softwares, |
276 |
some unexpected access requests arise from proper procedure. |
277 |
Such access requests should be granted because |
278 |
they are not caused by malicious attacks. |
279 |
So I introduced a mechanism to allow administrator some grace |
280 |
to decide to grant or reject such access requests. |
281 |
This mechanism is implemented in the following manner. |
282 |
"Don't return immediately if permission denied." |
283 |
"Sleep for a while waiting administrator's decision." |
284 |
"Return successfully if administrator tells to do so." |
285 |
|
286 |
Fix 2006/04/12 |
287 |
|
288 |
@ Fix handling of prefix in GetAbsolutePath(). |
289 |
|
290 |
Some objects doesn't have prefix "/". |
291 |
Pipe has prefix "pipe:" and socket has prefix "socket:". |
292 |
GetAbsolutePath() couldn't handle prefixes other than '/' properly. |
293 |
|
294 |
@ Remove IsCorrectPath() checks for File Access Control functions. |
295 |
|
296 |
File Access Control functions accepted only pathnames that start |
297 |
with '/' because these functions assumed pathnames returned by |
298 |
GetAbsolutePath() always start with '/'. |
299 |
However, I found a program that opens an unnamed pipe via |
300 |
(probably) /proc/PID/fd/ directory. (You can see entries like |
301 |
"pipe:[number]" if you run "ls -l /proc/*/fd/".) |
302 |
Now, File Access Control functions have to accept pathnames |
303 |
that don't start with '/'. So, I stopped checking IsCorrectPath(). |
304 |
|
305 |
Fix 2006/04/19 |
306 |
|
307 |
@ Fix handling of NULL nameidata in vfs_open(). |
308 |
|
309 |
In 2.6 kernels, NFS daemon and sys_mq_open() call |
310 |
vfs_create() with NULL nameidata. In such cases, |
311 |
CheckSingleWritePermission() must not be called. |
312 |
|
313 |
Version 1.1.1 2006/05/15 Functionality enhancement release. |
314 |
|
315 |
Fix 2006/05/16 |
316 |
|
317 |
@ Support program files aggregation. |
318 |
|
319 |
Until now, programs that have no fixed names and their |
320 |
parent programs had to be run in a trusted domain |
321 |
since it is impossible to use patterns for granting |
322 |
execute permission and defining domains. |
323 |
I introduced a mechanism to aggregate similar programs |
324 |
using 'aggregator' directive. |
325 |
Some examples: |
326 |
|
327 |
'aggregator /tmp/logrotate.\?\?\?\?\?\? /tmp/logrotate.tmp' |
328 |
to run all temporary programs for logrotate as /tmp/logrotate.tmp |
329 |
|
330 |
'aggregator /usr/bin/tac /bin/cat' |
331 |
to run /usr/bin/tac and /bin/cat as /bin/cat |
332 |
|
333 |
Fix 2006/05/18 |
334 |
|
335 |
@ Unlimit max count for audit log. |
336 |
|
337 |
I forgot to replace MAX_GRANT_LOG and MAX_REJECT_LOG with INT_MAX |
338 |
so that administrators can give any size for audit logs at runtime. |
339 |
|
340 |
Fix 2006/05/22 |
341 |
|
342 |
@ Support individual domain ACL removal. |
343 |
|
344 |
Until now, to remove ACLs from a domain, administrator had to |
345 |
once delete and recreate that domain, which wastes a lot of memory. |
346 |
I introduced a mechanism to remove domain ACL without deleting and |
347 |
recreating domains. |
348 |
Administrator can delete domains or remove ACLs from domains |
349 |
via /proc/ccs/policy/domain_policy . |
350 |
/proc/ccs/policy/delete_domain and /proc/ccs/policy/update_domain |
351 |
were removed. |
352 |
|
353 |
Fix 2006/05/30 |
354 |
|
355 |
@ Add missing spinlock in SAKURA_MayMount(). |
356 |
|
357 |
vfsmount_lock was missing. |
358 |
|
359 |
Version 1.1.2 2006/06/02 Functionality enhancement release. |
360 |
|
361 |
Fix 2006/06/13 |
362 |
|
363 |
@ Merge tomoyo_connect.c and tomoyo_bind.c into tomoyo_port.c |
364 |
|
365 |
I merged these files that have only difference CONNECT and BIND, |
366 |
that are likely to be enabled both or neither. |
367 |
|
368 |
@ Add CONFIG_TOMOYO_AUDIT option. |
369 |
|
370 |
I made auditing functions as optional because some Linux boxes |
371 |
may have not enough disk space to store audit logs. |
372 |
|
373 |
Fix 2006/06/15 |
374 |
|
375 |
@ Support use of symbolic links for program execution. |
376 |
|
377 |
Until now, domains for programs executed by dereferencing |
378 |
symbolic links were defined using dereferenced pathnames. |
379 |
This was inconvenient for some Linux boxes who use busybox but |
380 |
can't keep hard links of busybox. |
381 |
I introduced a mechanism to allow using pathnames of |
382 |
symbolic links using 'alias' directive. |
383 |
Some examples: |
384 |
|
385 |
'alias /sbin/busybox /bin/ls' to run /bin/ls |
386 |
(which is a symbolic link to /sbin/busybox) as /bin/ls |
387 |
if /bin/ls is executed. |
388 |
|
389 |
'alias /bin/bash /bin/sh' to run /bin/sh |
390 |
(which is a symbolic link to /bin/bash) as /bin/sh |
391 |
if /bin/sh is executed. |
392 |
|
393 |
Fix 2006/06/21 |
394 |
|
395 |
@ Use ccs_alloc() instead of kzalloc(). |
396 |
|
397 |
To detect memory leaks, |
398 |
I added a wrapper for tracing kmalloc() and kfree(). |
399 |
There is no way to detect memory leaks caused by ccs-*.txt . |
400 |
|
401 |
Version 1.1.3 2006/07/13 Functionality enhancement release. |
402 |
|
403 |
Fix 2006/07/14 |
404 |
|
405 |
@ Change behavior of pathname pattern matching. |
406 |
|
407 |
Until now, it was impossible to use patterns like "\*.txt" because |
408 |
"\*" matched zero or more repetitions of characters until next '/'. |
409 |
Now, "\*" matches zero or more repetitions of characters. |
410 |
|
411 |
Until now, it was impossible to use patterns like "\$00" |
412 |
because "\$" matched one or more repetitions of digits until next |
413 |
non digit character. |
414 |
Now, "\$" matches one or more repetitions of digits. |
415 |
|
416 |
Also, new patterns "\x" "\X" "\a" "\A" "\@" are added. |
417 |
|
418 |
Fix 2006/07/21 |
419 |
|
420 |
@ Add CONFIG_TOMOYO_NETWORK option. |
421 |
|
422 |
Until now, only port numbers for TCP and UDP were controllable. |
423 |
Now, the combination of IPv4/IPv6 address and port numbers |
424 |
for TCP and UDP is controllable. |
425 |
CONFIG_TOMOYO_NETWORKPORT became obsolete. |
426 |
|
427 |
Fix 2006/07/25 |
428 |
|
429 |
@ Change matching rule for CheckFileACL(). |
430 |
|
431 |
Until now, only first entry that matched the requested pathname |
432 |
was used for permission checking. For example, two entries |
433 |
|
434 |
"2 /tmp/file-\$.txt" |
435 |
"4 /tmp/fil\?-0.txt" |
436 |
|
437 |
are given in this order and requested pathname is "/tmp/file-0.txt", |
438 |
the "2 /tmp/file-\$.txt" is used. But if two entries |
439 |
|
440 |
"4 /tmp/fil\?-0.txt" |
441 |
"2 /tmp/file-\$.txt" |
442 |
|
443 |
are given in this order, the "4 /tmp/fil\?-0.txt" is used. |
444 |
This may potentially cause trouble because the result of |
445 |
permission checks depends on the order of entries. |
446 |
|
447 |
Now, all entries that matched the requested pathname |
448 |
are used for permission checking so that the result of |
449 |
permission checks doesn't depend on the order of entries. |
450 |
|
451 |
Fix 2006/07/27 |
452 |
|
453 |
@ Support RAW IPv4/IPv6 control. |
454 |
|
455 |
Some programs such as 'ping' and 'traceroute' use raw IP socket. |
456 |
Now, the combination of IPv4/IPv6 address and protocol numbers |
457 |
for IP is controllable. |
458 |
|
459 |
Fix 2006/08/04 |
460 |
|
461 |
@ Add filename and argv[0] comparison check. |
462 |
|
463 |
The domain transition was done based on filename passed to do_execve(), |
464 |
while the behavior was defined based on argv[0]. |
465 |
There is no problem if the filename is argv[0]-unaware application. |
466 |
But if argv[0]-aware, access control bypassing happens if the process |
467 |
transits to trusted domain but behaves as different program. |
468 |
For example, when the administrator specifies domain for /bin/ls as |
469 |
trusted but both /bin/ls and /bin/cat are links to /sbin/busybox , |
470 |
a cracker can run /bin/cat in a trusted domain if the cracker |
471 |
succeeds to invoke do_execve() with filename = "/bin/ls" and |
472 |
argv[0] = "/bin/cat". |
473 |
|
474 |
I introduced a directive that permits the mismatch of |
475 |
basename of filename and argv[0]. |
476 |
|
477 |
Fix 2006/08/10 |
478 |
|
479 |
@ Support ID based condition checks. |
480 |
|
481 |
It was impossible to use process id (uid and gid and so on) for |
482 |
checking individual domain ACL. |
483 |
|
484 |
Now it became possible to use process id for checking individual |
485 |
domain ACL. For example, |
486 |
|
487 |
"1 /bin/sh if task.euid!=0" |
488 |
|
489 |
allows the domain to execute /bin/sh only when the process's euid |
490 |
is not 0, and |
491 |
|
492 |
"6 /home/\*/\* if task.uid=path1.uid" |
493 |
|
494 |
allows the domain to read-write user's home directory |
495 |
only when the file's owner matches the process's uid. |
496 |
|
497 |
Fix 2006/08/22 |
498 |
|
499 |
@ Fix ROUNDUP() in fs/realpath.c . |
500 |
|
501 |
Alignment using sizeof(int) may be inappropriate for 64bit environment. |
502 |
I changed to use the larger size of 'void *' and 'long' |
503 |
instead of 'int'. |
504 |
For environment where sizeof(int) = sizeof(long) = sizeof(void *), |
505 |
this change has no effect. |
506 |
|
507 |
Version 1.2 2006/09/03 Functionality enhancement release. |
508 |
|
509 |
Fix 2006/09/30 |
510 |
|
511 |
@ Fix CheckFilePerm() in fs/tomoyo_file.c . |
512 |
|
513 |
The location to call path_release() was too early. |
514 |
|
515 |
Fix 2006/10/02 |
516 |
|
517 |
@ Support per-domain profile. |
518 |
|
519 |
It became possible to assign different profiles for different domains. |
520 |
This will help administrators using building up approach. |
521 |
|
522 |
Fix 2006/10/05 |
523 |
|
524 |
@ Change parameters for CheckFilePerm(). |
525 |
|
526 |
I was re-resolving pathnames inside CheckFilePerm() even though |
527 |
the caller function already resolved them. |
528 |
So I changed to pass dentry and vfsmount instead of pathname, |
529 |
and removed changes made on 2006/09/30. |
530 |
|
531 |
Fix 2006/10/06 |
532 |
|
533 |
@ Support deny_rewrite and allow_rewrite permission. |
534 |
|
535 |
It became possible to make regular files append-only |
536 |
using "deny_rewrite" directive in exception policy and |
537 |
override it using "allow_rewrite" directive in domain policy. |
538 |
|
539 |
Regular files specified using "deny_rewrite" directive |
540 |
can't be open()ed with O_TRUNC or without O_APPEND, |
541 |
can't be truncate()ed or ftruncate()ed, |
542 |
can't be turned O_APPEND flag off using fcntl(F_SETFL) |
543 |
unless specified using "allow_rewrite" directive. |
544 |
|
545 |
Fix 2006/10/12 |
546 |
|
547 |
@ Enable configuration options by default for kernel config. |
548 |
|
549 |
CONFIG_SAKURA and CONFIG_TOMOYO are now 'y' by default |
550 |
and CONFIG_SYAORAN is now 'm' by default. |
551 |
|
552 |
Fix 2006/10/13 |
553 |
|
554 |
@ Use external policy loader. |
555 |
|
556 |
Until now, policies are loaded when /sbin/init starts and |
557 |
initial control levels are switched using CCS= parameter. |
558 |
But since some boxes have to fixate kernel command line options |
559 |
at compilation time, I think it will become more flexible |
560 |
by running external policy loader using init= parameter so that |
561 |
initial control levels can be specified before /sbin/init starts. |
562 |
|
563 |
Call panic() if initial control levels are not specified. |
564 |
|
565 |
Fix 2006/10/16 |
566 |
|
567 |
@ Add missing parameter in FindNextDomain(). |
568 |
|
569 |
'struct file' was needed for allowing 'if path1.*' checks. |
570 |
|
571 |
Fix 2006/10/23 |
572 |
|
573 |
@ Print error messages in CheckFlags(). |
574 |
|
575 |
Some users seem to have troubles picking up all necessary |
576 |
entries for the configuration file of SYAORAN filesystem |
577 |
since makesyaoranconf can't pick up entries that are |
578 |
nonexistent at the time. |
579 |
I added error message so that users can find missing entries |
580 |
using dmesg. |
581 |
|
582 |
Fix 2006/10/24 |
583 |
|
584 |
@ Change /proc/ccs/info/self_domain . |
585 |
|
586 |
I changed /proc/ccs/info/self_domain to return |
587 |
the domain of open time rather than first read time. |
588 |
This modification makes shell's redirection usage |
589 |
more convenient since redirection opens file |
590 |
but doesn't read at the time. |
591 |
|
592 |
'cat < /proc/ccs/info/self_domain' will return |
593 |
the domain of shell, and |
594 |
'cat /proc/ccs/info/self_domain' will return |
595 |
the domain of cat . |
596 |
|
597 |
Fix 2006/11/06 |
598 |
|
599 |
@ Replace MAX_ENFORCE_GRACE with ALLOW_ENFORCE_GRACE. |
600 |
|
601 |
Since it was inconvenient that requests that are waiting for |
602 |
supervisor's decision are rejected automatically when |
603 |
MAX_ENFORCE_GRACE seconds has elapsed, I modified WriteAnswer() |
604 |
reset timeout counter whenever a supervisor's decision is written |
605 |
and I modified ccs-queryd write a dummy decision every seconds |
606 |
so that the requests won't be rejected automatically as long as |
607 |
ccs-queryd is running. |
608 |
This change made MAX_ENFORCE_GRACE's meaning boolean. |
609 |
So I fixated MAX_ENFORCE_GRACE to 10 seconds and removed |
610 |
MAX_ENFORCE_GRACE parameter. |
611 |
To allow administrators selectively enable "delayed enforcing" |
612 |
mode, I added ALLOW_ENFORCE_GRACE parameter. |
613 |
The behavior of "delayed enforcing" mode is defined |
614 |
in the following order. |
615 |
|
616 |
(1) The requests are rejected immediately if ALLOW_ENFORCE_GRACE=0. |
617 |
(2) The requests are rejected immediately |
618 |
if nobody is opening /proc/ccs/policy/query interface. |
619 |
(3) The requests won't be rejected automatically |
620 |
if ALLOW_ENFORCE_GRACE=1 and ccs-queryd is running. |
621 |
(4) The requests will be rejected in 10 seconds |
622 |
if somebody other than ccs-queryd (such as less(1)) is |
623 |
opening /proc/ccs/policy/query interface, for |
624 |
such process doesn't write dummy decisions. |
625 |
|
626 |
Version 1.3 2006/11/11 First anniversary release. |
627 |
|
628 |
Fix 2006/11/13 |
629 |
|
630 |
@ Replace trust_domain with keep_domain. |
631 |
|
632 |
Since it was troublesome that there are two elements that can disable MAC |
633 |
(assigning a profile that doesn't enable MAC or registering domains |
634 |
with trust_domain directive), I removed trust_domain directive. |
635 |
Instead, I introduced keep_domain directive to not to transit domains |
636 |
unless a program registered with initializer directive is executed. |
637 |
This change has the following advantages. |
638 |
|
639 |
(1) Allows administrator use "enforce mode" for operations after login. |
640 |
Since it was difficult to know what commands and files are invoked |
641 |
and accessed in what sequences beforehand, we had to use trust_domain |
642 |
directive for such domain, allowing users invoke any commands and |
643 |
access any files in any sequence. |
644 |
But now, we can use keep_domain directive and assign a profile for |
645 |
"enforce mode" for such domain, forcing users invoke only allowed |
646 |
commands and access only allowed files in any sequence |
647 |
while these operations are kept under the control of "enforce mode". |
648 |
|
649 |
(2) Allows administrator determine easily whether the domain is |
650 |
under MAC or not because only the profile currently assigned to |
651 |
the domain determines it. |
652 |
|
653 |
(3) Saves total number of domains and memory. |
654 |
|
655 |
Fix 2006/11/22 |
656 |
|
657 |
@ Don't allow use of undefined profile. |
658 |
|
659 |
To avoid assigning undefined profile to domains by error, |
660 |
I added checks before assigning profiles to domains. |
661 |
Now, profiles have to be defined prior to assigning them to domains. |
662 |
|
663 |
Version 1.3.1 2006/12/08 Minor update release. |
664 |
|
665 |
Fix 2006/12/10 |
666 |
|
667 |
@ Allow pathname grouping. |
668 |
|
669 |
To reduce the labor of repeating '/\*' to allow access recursively, |
670 |
I introduced a macro 'path_group' to make group such pathnames. |
671 |
For example, you had to give like |
672 |
|
673 |
4 /var/www/html/\* |
674 |
4 /var/www/html/\*/\* |
675 |
4 /var/www/html/\*/\*/\* |
676 |
4 /var/www/html/\*/\*/\*/\* |
677 |
|
678 |
but now, you can give just |
679 |
|
680 |
4 @WEB-CONTENTS |
681 |
|
682 |
if you give |
683 |
|
684 |
path_group WEB-CONTENTS /var/www/html/\* |
685 |
path_group WEB-CONTENTS /var/www/html/\*/\* |
686 |
path_group WEB-CONTENTS /var/www/html/\*/\*/\* |
687 |
path_group WEB-CONTENTS /var/www/html/\*/\*/\*/\* |
688 |
|
689 |
in the exception policy. |
690 |
This macro will be useful when grouping different directories. |
691 |
|
692 |
Fix 2006/12/15 |
693 |
|
694 |
@ Use structured pathnames instead for simple 'char *'. |
695 |
|
696 |
To reduce the cost of strcmp(), I changed the return value of |
697 |
SaveName() from 'const char *' to 'const struct path_info *'. |
698 |
This change will speed up PathMatchesToPattern() comparison. |
699 |
|
700 |
Fix 2006/12/19 |
701 |
|
702 |
@ Allow registering policy managers using domainnames. |
703 |
|
704 |
It was difficult to restrict programs that can update policies |
705 |
via /proc/ccs/ interfaces using pathnames of these programs, for |
706 |
these programs could be unintendedly invoked. |
707 |
Now, it became possible to restrict domains that can update policies |
708 |
via /proc/ccs/ interfaces as well as programs. |
709 |
By restricting using domainnames, it becomes easier to avoid |
710 |
unintended invocation. |
711 |
|
712 |
Fix 2006/12/22 |
713 |
|
714 |
@ Add initialize_domain,no_initizlize_domain,no_keep_domain |
715 |
|
716 |
To control domain transitions more strictly, |
717 |
initialize_domain,no_initizlize_domain,no_keep_domain directives |
718 |
were introduced. |
719 |
|
720 |
"initialize_domain /some/program" means |
721 |
jump to "<kernel> /some/program" domain if /some/program is |
722 |
called from any domain. |
723 |
This is equivalent to conventional "initializer /some/program". |
724 |
|
725 |
"initialize_domain /some/program from some_domain" means |
726 |
jump to "<kernel> /some/program" domain only if /some/program is |
727 |
called from "some_domain" domain. |
728 |
|
729 |
"no_initialize_domain /some/program" means |
730 |
don't jump to "<kernel> /some/program" domain even if |
731 |
"initialize_domain /some/program" or |
732 |
"initialize_domain /some/program from some_domain" are given |
733 |
if /some/program is called from any domain. |
734 |
|
735 |
"no_initialize_domain /some/program from some_domain" means |
736 |
don't jump to "<kernel> /some/program" domain even if |
737 |
"initialize_domain /some/program" or |
738 |
"initialize_domain /some/program from some_domain" are given |
739 |
if /some/program is called from "some_domain" domain. |
740 |
|
741 |
"keep_domain some_domain" means don't jump to child domain |
742 |
if any programs are called from "some_domain" domain. |
743 |
|
744 |
"keep_domain /some/program from some_domain" means |
745 |
don't jump to child domain only if /some/program is |
746 |
called from "some_domain" domain. |
747 |
|
748 |
"no_keep_domain some_domain" means |
749 |
jump to child domain even if |
750 |
"keep_domain /some/program" or |
751 |
"keep_domain /some/program from some_domain" are given |
752 |
if any programs are called from "some_domain" domain. |
753 |
|
754 |
"no_keep_domain /some/program from some_domain" means |
755 |
jump to child domain even if |
756 |
"keep_domain /some/program" or |
757 |
"keep_domain /some/program from some_domain" are given |
758 |
if /some/program is called from "some_domain" domain. |
759 |
|
760 |
"some_domain" can be just the last component of domainname. |
761 |
For example, giving "/bin/mail" as "some_domain" matches |
762 |
all domains whose domainname ends with "/bin/mail". |
763 |
|
764 |
Fix 2007/01/19 |
765 |
|
766 |
@ Allow reuse of memory allocated for domain policy. |
767 |
|
768 |
Regarding domain policy, unlike other policies, didn't have |
769 |
"is_deleted" flag and new memory were allocated |
770 |
if the deleted entries are given again. |
771 |
But to allow administrators switch domain policy periodically, |
772 |
I introduced "is_deleted" flag. |
773 |
|
774 |
Writing "some_domain" to /proc/ccs/policy/domain_policy |
775 |
creates "some_domain" using new memory if it didn't exist. |
776 |
|
777 |
Writing "select some_domain" doesn't create "some_domain" |
778 |
if it didn't exist. |
779 |
|
780 |
Writing "delete some_domain" deletes "some_domain" |
781 |
but does not delete entries in "some_domain". |
782 |
|
783 |
Writing "undelete some_domain" undeletes "some_domain" |
784 |
if it was deleted by "delete some_domain". |
785 |
|
786 |
Fix 2007/01/22 |
787 |
|
788 |
@ Allow getting already deleted pathnames. |
789 |
|
790 |
To allow getting pathnames that are already deleted, |
791 |
I removed (IS_ROOT(dentry) || !d_unhashed(dentry)) check. |
792 |
|
793 |
Fix 2007/01/26 |
794 |
|
795 |
@ Limit string length to 4000. |
796 |
|
797 |
I was using PAGE_SIZE (4096 in many environments) |
798 |
as the max length of any string data. |
799 |
But for environments that have larger PAGE_SIZE, |
800 |
doing memset(ptr, 0, PAGE_SIZE) everytime is too wasteful. |
801 |
|
802 |
Fix 2007/01/29 |
803 |
|
804 |
@ Add garbage collector for domain policy. |
805 |
|
806 |
Writing "some_domain" to /proc/ccs/policy/domain_policy |
807 |
creates "some_domain" using new memory only if |
808 |
some process is staying at that deleted domain. |
809 |
If no process is staying at that deleted domain, |
810 |
"some_domain" is undeleted with all ACLs deleted. |
811 |
|
812 |
Version 1.3.2 2007/02/14 Usability enhancement release. |
813 |
|
814 |
Fix 2007/02/20 |
815 |
|
816 |
@ Allow address grouping. |
817 |
|
818 |
To reduce the labor of repeating similar IPv4/IPv6 addresses, |
819 |
I introduced a macro 'address_group' to make group such addresses. |
820 |
For example, you had to give like |
821 |
|
822 |
allow_network TCP accept 10.0.0.0-10.255.255.255 1024-65535 |
823 |
allow_network TCP accept 172.16.0.0-172.31.255.255 1024-65535 |
824 |
allow_network TCP accept 192.168.0.0-192.168.255.255 1024-65535 |
825 |
|
826 |
but now, you can give just |
827 |
|
828 |
allow_network TCP accept @localnet 1024-65535 |
829 |
|
830 |
if you give |
831 |
|
832 |
address_group localnet 10.0.0.0-10.255.255.255 |
833 |
address_group localnet 172.16.0.0-172.31.255.255 |
834 |
address_group localnet 192.168.0.0-192.168.255.255 |
835 |
|
836 |
in the exception policy. |
837 |
|
838 |
Fix 2007/03/03 |
839 |
|
840 |
@ Remove obsolete functions. |
841 |
|
842 |
@ Add some hooks. |
843 |
|
844 |
Read permission check is done if open_exec() |
845 |
is called from search_binary_handler(). |
846 |
Read permission check is not done if open_exec() |
847 |
is called from do_execve(), instead, |
848 |
execute permission check is done at |
849 |
search_binary_handler_with_transition(). |
850 |
|
851 |
I moved the location of calling CheckCapabilityACL() |
852 |
and CheckMountPermission() from sys_mount() to do_mount(). |
853 |
|
854 |
Fix 2007/03/07 |
855 |
|
856 |
@ Use 'unsigned int' for sscanf(). |
857 |
|
858 |
I compiled SYAORAN fs on x86_64 environment and found |
859 |
the compiler showing warning messages about size of data types. |
860 |
Since size of data types may mismatch for sscanf(), |
861 |
I replaced some types with 'unsigned int'. |
862 |
|
863 |
Version 1.4 2007/04/01 x86_64 support release. |
864 |
|
865 |
Fix 2007/04/18 |
866 |
|
867 |
@ Change argv[0] checking rule. |
868 |
|
869 |
I was comparing the basename of symbolic link's pathname and argv[0]. |
870 |
Since execute permission check and domain transition are done |
871 |
based on realpath while argv[0] check is done based on the symlink's |
872 |
pathname and argv[0], this specification will allow attackers behave |
873 |
as /bin/cat in the domain of /bin/ls if "/bin/ls and /bin/cat are |
874 |
links to /sbin/busybox" and "the attacker is permitted to create |
875 |
a symlink named ~/cat that points to /bin/ls" and "the attacker is |
876 |
permitted to run /bin/ls". |
877 |
So, I changed to compare the basename of realpath and argv[0]. |
878 |
Also, I moved the location to compare before processing |
879 |
"aggregator" directive so that |
880 |
"aggregator /tmp/logrotate.\?\?\?\?\?\? /tmp/logrotate.tmp" |
881 |
won't cause the mismatch of the basename of realpath and argv[0]. |
882 |
|
883 |
If /bin/ls is a symlink to /sbin/busybox, then |
884 |
creating a symlink named ~/cat that points to /bin/ls and |
885 |
executing ~/cat won't work as expected because permission check and |
886 |
domain transition are done using /sbin/busybox (realpath of /bin/ls) |
887 |
and will be rejected since the administrator won't grant |
888 |
"1 /sbin/busybox". |
889 |
|
890 |
Fix 2007/05/07 |
891 |
|
892 |
@ Support pathname subtraction. |
893 |
|
894 |
There was no way to exclude specific pathnames when granting |
895 |
permissions using wildcards. |
896 |
There would be a need to exclude specific files and directories. |
897 |
I introduced "\-" as subtraction operator. |
898 |
|
899 |
"A\-B" means "A" other than "B". |
900 |
"A\-B\-C" means "A" other than "B" and "C". |
901 |
"A\-B\-C\-D" means "A" other than "B" and "C" and "D". |
902 |
|
903 |
"A", "B", "C", "D" may contain wildcards. |
904 |
|
905 |
An example usage is "/home/\*/\*\-.ssh/\*", which means |
906 |
"/home/\*/\*/\*" other than "/home/\*/.ssh/\*". |
907 |
|
908 |
"A" should contain wildcards because subtraction from constants |
909 |
(e.g. "/usr\-usr/" or "/usr\-home/") is meaningless. |
910 |
|
911 |
Don't try "A\-B\+C" because "\+" is not addition operator. |
912 |
|
913 |
Fix 2007/05/24 |
914 |
|
915 |
@ Fix autobind hook. |
916 |
|
917 |
The location to call SAKURA_MayAutobind() in net/ipv4/udp.c |
918 |
and net/ipv6/udp.c were wrong. |
919 |
|
920 |
Fix 2007/06/03 |
921 |
|
922 |
@ Add a space in MakeMountOptions(). |
923 |
|
924 |
I forgot to add a space after "atime" and "noatime". |
925 |
|
926 |
Version 1.4.1 2007/06/05 Minor update release. |
927 |
|
928 |
Fix 2007/07/04 |
929 |
|
930 |
@ Fix ReadAddressGroupPolicy() bug. |
931 |
|
932 |
ReadAddressGroupPolicy() fails if both "path_group" and "address_group" |
933 |
are used because I forgot to set "head->read_var1 = NULL". |
934 |
|
935 |
Fix 2007/07/10 |
936 |
|
937 |
@ Add compat_sys_stime() hook. |
938 |
|
939 |
Some of 64bit kernels support compat_sys_stime() |
940 |
but permission check was missing. |
941 |
|
942 |
Version 1.4.2 2007/07/13 Bug fix release. |
943 |
|
944 |
Fix 2007/08/06 |
945 |
|
946 |
@ Remove mount-flags manipulation. |
947 |
|
948 |
Until now, administrator is permitted to turn on/off specific mount |
949 |
options regardless of mount options passed to kernel. |
950 |
I removed this feature because "exact option matching" sounds better than |
951 |
"automatic option enabler/disabler". |
952 |
|
953 |
@ Remove /proc/ccs/info/mapping . |
954 |
|
955 |
I removed /proc/ccs/info/mapping because nobody seems to use this |
956 |
feature. |
957 |
|
958 |
@ Call external policy loader automatically. |
959 |
|
960 |
Until now, users had to add init=/.init parameter to load policy |
961 |
before /sbin/init starts. |
962 |
I inserted call_usermodehelper() to call external policy loader when |
963 |
execve("/sbin/init") is requested and external policy loader exists. |
964 |
|
965 |
This change will remove init=/.init parameter from most environment, |
966 |
although call_usermodehelper() can't handle interactive operations. |
967 |
|
968 |
@ Move external policy loader from /.init to /sbin/ccs-init . |
969 |
|
970 |
Installing programs in / directory is not good for packaging. |
971 |
|
972 |
Fix 2007/08/13 |
973 |
|
974 |
@ Update external policy loader. |
975 |
|
976 |
It turned out that /sbin/ccs-init invoked via call_usermodehelper() |
977 |
can handle interactive operations by opening /dev/console . |
978 |
Now, there is no difference between init=/sbin/ccs-init and |
979 |
call_usermodehelper("/sbin/ccs-init"), and users no longer need to |
980 |
add init=/sbin/ccs-init parameter to load policy before /sbin/init |
981 |
starts. |
982 |
|
983 |
Fix 2007/08/14 |
984 |
|
985 |
@ Update recvmsg() hooks. |
986 |
|
987 |
Until now, it was impossible to apply network access control for |
988 |
incoming UDP and RAW packets if they are brought to userland using |
989 |
read() or recvmsg() with NULL address because address buffer is NULL. |
990 |
I moved hooks from sock_recvmsg() to skb_recv_datagram() so that |
991 |
network access control for incoming UDP and RAW packets always work. |
992 |
|
993 |
Fix 2007/08/16 |
994 |
|
995 |
@ Return appropriate error code for CheckMountPermission(). |
996 |
|
997 |
I was returning -EPERM if something is wrong with CheckMountPermission(). |
998 |
But SELinux determines whether selinuxfs is supported by kernel |
999 |
based on whether error code is -ENODEV or not. |
1000 |
So I stopped returning -EPERM unconditionally. |
1001 |
|
1002 |
Fix 2007/08/17 |
1003 |
|
1004 |
@ Remove initializer directive. |
1005 |
|
1006 |
Use "initialize_domain" instead of "initializer". |
1007 |
|
1008 |
Fix 2007/08/21 |
1009 |
|
1010 |
@ Fix "allow_argv0 ... if if ..." bug. |
1011 |
|
1012 |
It was impossible to use a word "if" to the second argument of |
1013 |
allow_argv0 if condition part is used. |
1014 |
|
1015 |
Fix 2007/08/24 |
1016 |
|
1017 |
@ Move /proc/ccs/\*/\* to /proc/ccs/\* . |
1018 |
|
1019 |
Some pathnames for /proc/ccs/ interface were changed. |
1020 |
|
1021 |
Fix 2007/09/05 |
1022 |
|
1023 |
@ Drop MSG_PEEK'ed message before skb_free_datagram(). |
1024 |
|
1025 |
I need to remove head message from unwanted source |
1026 |
from socket's receive queue so that the caller can pick up |
1027 |
next message from wanted source with MSG_PEEK flags. |
1028 |
|
1029 |
Version 1.5.0 2007/09/20 Usability enhancement release. |
1030 |
|
1031 |
Fix 2007/09/27 |
1032 |
|
1033 |
@ Avoid eating memory after quota exceeded. |
1034 |
|
1035 |
Although ACL entries in a domain won't be added if the domain's quota |
1036 |
has exceeded, SaveName() in AddFileACL() is called anyway. |
1037 |
This caused unneeded memory consumption. |
1038 |
|
1039 |
Now, quota checking is done before getting domain_acl_lock lock. |
1040 |
This may exceed quota by one or two entries, but that won't matter. |
1041 |
|
1042 |
Fix 2007/10/16 |
1043 |
|
1044 |
@ Add environment variable check. |
1045 |
|
1046 |
There are environment variables that may cause dangerous behavior |
1047 |
like LD_\* . |
1048 |
So I introduced 'allow_env' directive that allows specified |
1049 |
environment variable inherited to next domain. |
1050 |
Unlike other permissions, this check is done at execve() time |
1051 |
using next domain's ACL information. |
1052 |
|
1053 |
To manage commonly inherited environments like PATH , |
1054 |
you can use 'allow_env' directive in exception policy |
1055 |
to globally grant specified environment variable. |
1056 |
|
1057 |
Fix 2007/11/05 |
1058 |
|
1059 |
@ Replace semaphore with mutex. |
1060 |
|
1061 |
I replaced semaphore with mutex. |
1062 |
|
1063 |
@ Add missing down() in AddReservedEntry(). |
1064 |
|
1065 |
Mutex debugging capability told me that I had forgotten to call down() |
1066 |
since TOMOYO version 1.3.2 . |
1067 |
This function is not called by learning mode, |
1068 |
so the semaphore's counter will not overflow for normal usage. |
1069 |
|
1070 |
Fix 2005/11/27 |
1071 |
|
1072 |
@ Fix ReadTable() truncation bug. |
1073 |
|
1074 |
"snprintf(str, size, format, ...) >= size" means truncated. |
1075 |
But I was checking for "snprintf(str, size, format, ...) > size". |
1076 |
As a result, some entries might be dumped without '\n'. |
1077 |
|
1078 |
@ Purge direct "->prev"/"->next" manipulation. |
1079 |
|
1080 |
All list manipulations use "struct list_head" or "struct list1_head". |
1081 |
"struct list1_head" doesn't have "->prev" member to save memory usage. |
1082 |
|
1083 |
Fix 2007/11/29 |
1084 |
|
1085 |
@ Add missing semaphore in GetEXE(). |
1086 |
|
1087 |
mm->mmap_sem was missing. |
1088 |
|
1089 |
Fix 2007/12/17 |
1090 |
|
1091 |
@ Remove unused EXPORT_SYMBOL(). |
1092 |
|
1093 |
Mark some functions static. |
1094 |
|
1095 |
Fix 2007/12/18 |
1096 |
|
1097 |
@ Fix AddMountACL() rejection bug. |
1098 |
|
1099 |
To my surprise, "mount --bind source dest" accepts |
1100 |
not only "both source and dest are directory" |
1101 |
but also "both source and dest are non-directory". |
1102 |
I was rejecting if dest is not a directory in AddMountACL(). |
1103 |
|
1104 |
@ Change log format. |
1105 |
|
1106 |
Profile number and mode is added in audit logs. |
1107 |
|
1108 |
Fix 2008/01/03 |
1109 |
|
1110 |
@ Change directive for file's read/write/execute permission. |
1111 |
|
1112 |
Directives for file's read/write/execute permissions were |
1113 |
4/2/1 respectively. But for easier understanding, they are now |
1114 |
replaced by read/write/execute (e.g. "allow_read" instead of "4"). |
1115 |
But for easier inputting, 4/2/1 are still accepted instead of |
1116 |
allow_read/allow_write/allow_execute respectively. |
1117 |
|
1118 |
@ Change internal data structure. |
1119 |
|
1120 |
Since I don't have more than 16 types of file permissions, |
1121 |
I combined them using bit-fields. |
1122 |
|
1123 |
Each entry had a field for conditional permission support. |
1124 |
But since this field is unlikely used, I separated the field from |
1125 |
common part. |
1126 |
|
1127 |
These changes will reduce memory used by policy. |
1128 |
|
1129 |
Fix 2008/01/15 |
1130 |
|
1131 |
@ Add ptrace() hook. |
1132 |
|
1133 |
To prevent attackers from controlling important processes using |
1134 |
ptrace(), I added a hook for ptrace(). |
1135 |
Most programs (except strace(1) and gdb(1)) won't use ptrace(2). |
1136 |
|
1137 |
@ Fix sleep condition check in CheckSocketRecvDatagramPermission(). |
1138 |
|
1139 |
It seems that correct method to use is in_atomic() |
1140 |
rather than in_interrupt() because in_atomic() returns nonzero |
1141 |
whenever scheduling is not allowed. |
1142 |
|
1143 |
Fix 2008/02/05 |
1144 |
|
1145 |
@ Use find_task_by_vpid() instead of find_task_by_pid(). |
1146 |
|
1147 |
Kernel 2.6.24 introduced PID namespace. |
1148 |
To search PID given from userland, the kernel needs to use |
1149 |
find_task_by_vpid() instead of find_task_by_pid(). |
1150 |
|
1151 |
Fix 2008/02/14 |
1152 |
|
1153 |
@ Add execve() parameter checking. |
1154 |
|
1155 |
Until now, it was impossible to check argv[] and envp[] parameters |
1156 |
passed to execve(). |
1157 |
I expanded conditional permission syntax so that |
1158 |
{ argc, envc, argv[] , envp[] } parameters can be checked if needed. |
1159 |
This will allow administrator permit execution of /bin/sh only when |
1160 |
/bin/sh is invoked in the form of "/bin/sh -c" and environment variable |
1161 |
HOME is set by specifying |
1162 |
|
1163 |
allow_execute /bin/sh if exec.argv[1]="-c" exec.envp["HOME"]!=NULL |
1164 |
|
1165 |
in the policy. |
1166 |
This extension will make exploit codes difficult to start /bin/sh because |
1167 |
they unlikely set up environment variables and unlikely specify "-c" |
1168 |
option when invoking /bin/sh , whereas proper functions likely set up |
1169 |
environment variables and likely specify "-c" option. |
1170 |
|
1171 |
Fix 2008/02/18 |
1172 |
|
1173 |
@ Add process state checking. |
1174 |
|
1175 |
Until now, it was impossible to change ACL without executing program. |
1176 |
I added three variables for performing stateful checking within a domain. |
1177 |
You can set current process's state like: |
1178 |
|
1179 |
allow_network TCP accept @TRUSTED_HOSTS 1024-65535 ; set task.state[0]=1 |
1180 |
allow_network TCP accept @UNTRUSTED_HOSTS 1024-65535 ; set task.state[0]=0 |
1181 |
|
1182 |
and you can use the state like |
1183 |
|
1184 |
allow_read /path/to/important/file if task.state[0]=1 |
1185 |
|
1186 |
in the policy. |
1187 |
The state changes when the request was granted by the MAC's policy, |
1188 |
so please be careful with situations where the state has changed |
1189 |
successfully but the request was not processed because of other reasons |
1190 |
(e.g. out of memory). |
1191 |
|
1192 |
Fix 2008/02/26 |
1193 |
|
1194 |
@ Support /proc/ccs/ access by non-root user. |
1195 |
|
1196 |
Until now, only root user can access /proc/ccs/ interface. |
1197 |
But to permit /proc/ccs/ access by non-root user so that it won't require |
1198 |
ssh login by root user when administrating from remote host, |
1199 |
I made "(current->uid == 0 && current->euid == 0)" requirement optional. |
1200 |
If this requirement is disabled, only "conventional DAC permission |
1201 |
checks" and "/proc/ccs/manager checks" are used. |
1202 |
|
1203 |
Fix 2008/02/29 |
1204 |
|
1205 |
@ Add sleep_on_violation feature. |
1206 |
|
1207 |
Some exploit codes (e.g. trans2open for Samba) continue running |
1208 |
until it achieves the purpose of the exploit code (e.g. invoke /bin/sh). |
1209 |
|
1210 |
If such code is injected due to buffer overflow but the kernel |
1211 |
rejects the request, it triggers infinite "Permission denied" loop. |
1212 |
As a result, the CPU usage becomes 100% and gives bad effects to |
1213 |
the rest of processes. |
1214 |
This is a side effect of rejecting the request from the exploit code |
1215 |
which wouldn't happen if the request from the exploit code was granted. |
1216 |
|
1217 |
To avoid such CPU consumption, I added a penalty that forcibly |
1218 |
sleeps for specified period when a request is rejected. |
1219 |
|
1220 |
This penalty doesn't work if the exploit code does nothing but |
1221 |
continue running, but I think most exploit code's purpose is |
1222 |
to start some program rather than to slow down the target system. |
1223 |
|
1224 |
@ Add alt_exec feature. |
1225 |
|
1226 |
Since TOMOYO Linux's approach is "know all essential requests in advance |
1227 |
and create policy that permits only them", you can regard anomalous |
1228 |
requests as attacks (if you want to do so). |
1229 |
|
1230 |
Common MAC implementations merely reject requests that violate policy. |
1231 |
But I added a special handler for execve() to TOMOYO Linux. |
1232 |
|
1233 |
This handler is triggered when a process requested to execute a program |
1234 |
but the request was rejected by the policy. |
1235 |
This handler executes a program specified by the administrator |
1236 |
instead of a program requested by the process. |
1237 |
|
1238 |
Most attackers attempt to execute /bin/sh to start something malicious. |
1239 |
Attackers execute an exploit code using buffer overflow vulnerability |
1240 |
to steal control of a process. But this handler can get back control |
1241 |
if an exploit code requests execve() that is not permitted by policy. |
1242 |
|
1243 |
By default, this handler does nothing (i.e. merely reject execve() |
1244 |
request). You can specify any program to start what you want to do. |
1245 |
|
1246 |
You can redirect attackers to somewhere else (e.g. honey pot). |
1247 |
This makes it possible to act your Linux box as an on-demand honey pot |
1248 |
while keeping regular services for your usage. |
1249 |
|
1250 |
You can collect information of the attacker (e.g. IP address) and |
1251 |
update firewall configuration. |
1252 |
|
1253 |
You can silently terminate a process who requested execve() |
1254 |
that is not permitted by policy. |
1255 |
|
1256 |
Fix 2008/03/03 |
1257 |
|
1258 |
@ Add "force_alt_exec" directive. |
1259 |
|
1260 |
To be able to fully utilize "alt_exec" feature, |
1261 |
I added "force_alt_exec" directive so that |
1262 |
all execute requests are replaced by the execute request of a program |
1263 |
specified by alt_exec feature. |
1264 |
|
1265 |
If this directive is specified for a domain, the domain no longer |
1266 |
executes any programs regardless of the mode of file access control |
1267 |
(i.e. the domain won't execute even if MAC_FOR_FILE=0 ). |
1268 |
Instead, the domain executes the program specified by alt_exec feature |
1269 |
and the program specified by alt_exec feature validates the execute |
1270 |
request and executes it if it is appropriate to execute. |
1271 |
|
1272 |
If you can tolerate that there is no chance to return an error code |
1273 |
to the caller to tell the execute request was rejected, |
1274 |
this is more flexible approach than in-kernel execve() parameter |
1275 |
checking because we can do argv[] and envp[] checking easily. |
1276 |
|
1277 |
Fix 2008/03/04 |
1278 |
|
1279 |
@ Use string for access control mode. |
1280 |
|
1281 |
An integer expression for access control mode sometimes confuses |
1282 |
administrators because profile number is also an integer expression. |
1283 |
To avoid confusion between profile number and access control mode, |
1284 |
I introduced a string expression for access control mode. |
1285 |
|
1286 |
Modes which take an integer between 0 and 3. |
1287 |
|
1288 |
0 -> disabled |
1289 |
1 -> learning |
1290 |
2 -> permissive |
1291 |
3 -> enforcing |
1292 |
|
1293 |
Modes which take 0 or 1. |
1294 |
|
1295 |
0 -> disabled |
1296 |
1 -> enabled |
1297 |
|
1298 |
Fix 2008/03/10 |
1299 |
|
1300 |
@ Rename "force_alt_exec" directive to "execute_handler". |
1301 |
|
1302 |
To be able to use different programs for validating execve() parameters, |
1303 |
I moved the location to specify the program's pathname from profile |
1304 |
to domain policy. |
1305 |
|
1306 |
The "execute_handler" directive takes one pathname which is |
1307 |
invoked whenever execve() request is issued. Thus, any "allow_execute" |
1308 |
directives in a domain with "execute_handler" are ignored. |
1309 |
This directive is designed for validating expected/desirable execve() |
1310 |
requests in userspace, although there is no way to tell the caller |
1311 |
that the execve() request was rejected. |
1312 |
|
1313 |
@ Rename "alt_exec" directive to "denied_execute_handler". |
1314 |
|
1315 |
The "denied_execute_handler" directive takes one pathname which is |
1316 |
invoked only when execve() request was rejected. In other words, |
1317 |
this program is invoked only when the following conditions are met. |
1318 |
|
1319 |
(1) None of "allow_execute" directives in the domain matched. |
1320 |
(2) The execve() request was rejected in enforcing mode. |
1321 |
(3) "execute_handler" directive is not used by the domain. |
1322 |
|
1323 |
This directive is designed for handling unexpected/undesirable execve() |
1324 |
requests, to redirect the process issuing such requests to somewhere. |
1325 |
|
1326 |
Fix 2008/03/18 |
1327 |
|
1328 |
@ Fix wrong/redundant locks in pre-vfs functions. |
1329 |
|
1330 |
lock_kernel()/unlock_kernel() in pre_vfs_rename() were redundant for |
1331 |
2.6 kernels. |
1332 |
|
1333 |
Locking order in pre_vfs_link() and pre_vfs_unlink() for 2.4 kernels |
1334 |
after 2.4.33 were different from before 2.4.32 . |
1335 |
|
1336 |
Fix 2008/03/28 |
1337 |
|
1338 |
@ Disable execute handler loop. |
1339 |
|
1340 |
To be able to use "execute_handler" in a "keep_domain" domain, |
1341 |
ignore "execute_handler" and "denied_execute_handler" directives |
1342 |
if the current process is executing programs specified by |
1343 |
"execute_handler" or "denied_execute_handler" directive. |
1344 |
|
1345 |
This exception is needed to avoid infinite execute handler loop. |
1346 |
If a domain has both "keep_domain" and "execute_handler", |
1347 |
any execute request by that domain is handled by an execute handler, |
1348 |
and the execute handler attempts to process original execute request. |
1349 |
But the original execute request is handled by the same execute handler |
1350 |
unless the execute handler ignores "execute_handler". |
1351 |
|
1352 |
@ Update coding style. |
1353 |
|
1354 |
I rewrote the code to pass scripts/checkpatch.pl as much as possible. |
1355 |
Function names were changed to use only lower letters. |
1356 |
|
1357 |
Version 1.6.0 2008/04/01 Feature enhancement release. |
1358 |
|
1359 |
Fix 2008/04/14 |
1360 |
|
1361 |
@ Fix "Compilation failures" and "Initialization ordering bugs" |
1362 |
with kernels before 2.4.30/2.6.11 . |
1363 |
|
1364 |
2.6 kernels before 2.6.9 didn't have include/linux/hardirq.h , |
1365 |
resulting compilation error at #include <linux/hardirq.h> . |
1366 |
I added #elif condition. |
1367 |
|
1368 |
CentOS 4.6's 2.6.9 kernel calls do_execve() before initialization of |
1369 |
ccs_alloc(), resulting NULL pointer dereference. |
1370 |
I changed __initcall to core_initcall. |
1371 |
|
1372 |
CentOS 4.6's 2.6.9 kernel backported kzalloc() from 2.6.14 , |
1373 |
resulting compilation error at kzalloc(). |
1374 |
I modified prototype of kzalloc(). |
1375 |
|
1376 |
Fix 2008/04/20 |
1377 |
|
1378 |
@ Fix "Compilation failures" with kernels before 2.4.30/2.6.11 . |
1379 |
|
1380 |
Turbolinux 10 Server's 2.6.8 kernel backported kzalloc() as an inlined |
1381 |
function, resulting compilation error at kzalloc(). |
1382 |
I converted kzalloc() from an inlined function into a macro. |
1383 |
|
1384 |
Fix 2008/04/21 |
1385 |
|
1386 |
@ Add workaround for gcc 3.2.2's inline bug. |
1387 |
|
1388 |
RedHat Linux 9's gcc 3.2.2 generated a bad code |
1389 |
if ((var_of_u8 & 0x000000BF) & 0x80000000) { } |
1390 |
where the expected code is |
1391 |
if ((var_of_u8 & 0xBF) & 0x80) { } |
1392 |
when embedding ccs_acl_type2() into print_entry(), |
1393 |
resulting runtime BUG(). |
1394 |
I added the expected code explicitly as a workaround. |
1395 |
|
1396 |
Fix 2008/05/06 |
1397 |
|
1398 |
@ Add memory quota. |
1399 |
|
1400 |
1.5.x returns -ENOMEM when FindNextDomain() failed to create a new |
1401 |
domain, but I forgot to return -ENOMEM when find_next_domain() failed to |
1402 |
create a new domain. |
1403 |
|
1404 |
A domain is automatically created by find_next_domain() only if |
1405 |
the domain for the requested program doesn't exist. |
1406 |
This behavior is for the administrator's convenience. |
1407 |
The administrator needn't to know how many domains are needed for running |
1408 |
the whole programs in the system beforehand when developing the policy. |
1409 |
But the administrator does not want the kernel to reject execution of the |
1410 |
requested program when developing the policy. |
1411 |
|
1412 |
So, I think it is better to grant execution of programs even if |
1413 |
find_next_domain() failed to create a new domain than reject execution. |
1414 |
Thus, I decided not to return -ENOMEM when find_next_domain() failed to |
1415 |
create a new domain. This exception breaks the domain transition rules, |
1416 |
so I print "transition_failed" warning in /proc/ccs/domain_policy |
1417 |
when this exception happened. |
1418 |
|
1419 |
Also, to prevent the system from being halted by unexpectedly allocating |
1420 |
all kernel memory for the policy, I added memory quota. |
1421 |
This quota is configurable via /proc/ccs/meminfo like |
1422 |
|
1423 |
echo Shared: 1048576 > /proc/ccs/meminfo |
1424 |
echo Private: 1048576 > /proc/ccs/meminfo |
1425 |
|
1426 |
Version 1.6.1 2008/05/10 Bug fix release. |
1427 |
|
1428 |
Fix 2008/06/04 |
1429 |
|
1430 |
@ Check open mode of /proc/ccs/ interface. |
1431 |
|
1432 |
It turned out that I can avoid allocating memory for reading if |
1433 |
FMODE_READ is not set and memory for writing if FMODE_WRITE is not set. |
1434 |
|
1435 |
@ Wait for completion of /sbin/ccs-init . |
1436 |
|
1437 |
Since 2.4 kernel's call_usermodehelper() can't wait for termination of |
1438 |
the executed program, I was using the close() request of |
1439 |
/proc/ccs/meminfo to indicate that loading policy has finished. |
1440 |
But since /proc/ccs/meminfo could be accessed for setting memory quota |
1441 |
by /etc/ccs/ccs-post-init , I stopped using the close() request. |
1442 |
The policy loader no longer need to access /proc/ccs/meminfo to notify |
1443 |
the kernel that loading policy has finished. |
1444 |
|
1445 |
Fix 2008/06/05 |
1446 |
|
1447 |
@ Fix realpath for pipes and sockets. |
1448 |
|
1449 |
Kernel 2.6.22 and later use different method for calculating d_path(). |
1450 |
Since fs/realpath.c didn't notice the change, the realpath of pipes |
1451 |
appeared as "pipe:" rather than "pipe:[\$]" when they are opened via |
1452 |
/proc/PID/fd/ directory. |
1453 |
|
1454 |
@ Add process's information into /proc/ccs/query . |
1455 |
|
1456 |
While /proc/ccs/grant_log and /proc/ccs/reject_log contain process's |
1457 |
information, /proc/ccs/query doesn't contain it. |
1458 |
To be able to utilize ccs-queryd and ccs-notifyd more, I added it into |
1459 |
/proc/ccs/query . |
1460 |
|
1461 |
Fix 2008/06/10 |
1462 |
|
1463 |
@ Allow using patterns for globally readable files. |
1464 |
|
1465 |
To allow users specify locale specific files to globally readable files, |
1466 |
I relaxed checking in update_globally_readable_entry(). |
1467 |
|
1468 |
Fix 2008/06/11 |
1469 |
|
1470 |
@ Remove ALLOW_ENFORCE_GRACE parameter. |
1471 |
|
1472 |
Since unexpected requests caused by doing software updates can happen |
1473 |
in all profiles, users likely have to write ALLOW_ENFORCE_GRACE=enabled |
1474 |
to all profiles. And it makes meaningless to allow users to selectively |
1475 |
enable specific profile's ALLOW_ENFORCE_GRACE parameter. |
1476 |
So, I removed ALLOW_ENFORCE_GRACE parameter. |
1477 |
Now, the system behaves as if ALLOW_ENFORCE_GRACE=enabled is specified. |
1478 |
The behavior of "delayed enforcing" mode is defined in the following |
1479 |
order. |
1480 |
|
1481 |
(1) The requests are rejected immediately if nobody is opening |
1482 |
/proc/ccs/query interface. |
1483 |
(2) The requests will be rejected in 10 seconds if somebody other than |
1484 |
ccs-queryd (such as less(1)) is opening /proc/ccs/query interface, |
1485 |
for such process doesn't write dummy decisions. |
1486 |
|
1487 |
Fix 2008/06/22 |
1488 |
|
1489 |
@ Pass escaped pathname to audit_execute_handler_log(). |
1490 |
|
1491 |
I was passing unescaped pathname to audit_execute_handler_log() |
1492 |
which causes /proc/ccs/grant_log contain whitespace characters |
1493 |
if execute handler's pathname contains whitespace characters. |
1494 |
|
1495 |
Fix 2008/06/25 |
1496 |
|
1497 |
@ Return 0 when ccs_may_umount() succeeds. |
1498 |
|
1499 |
I forgot to clear error value in ccs_may_umount() when the requested |
1500 |
directory didn't match "deny_unmount" directive. As a result, any umount() |
1501 |
request with RESTRICT_UNMOUNT=enforcing returned -EPERM error. |
1502 |
|
1503 |
Version 1.6.2 2008/06/25 Usability enhancement release. |
1504 |
|
1505 |
Fix 2008/07/01 |
1506 |
|
1507 |
@ Fix "Compilation failure" with 2.4.20 kernel. |
1508 |
|
1509 |
RedHat Linux 9's 2.4.20 kernel backported O(1) scheduler patch, |
1510 |
resulting compilation error at ccs_load_policy(). |
1511 |
I added defined(TASK_DEAD) check. |
1512 |
|
1513 |
Fix 2008/07/08 |
1514 |
|
1515 |
@ Don't check permissions if vfsmount is NULL. |
1516 |
|
1517 |
Some filesystems (e.g. unionfs) pass NULL vfsmount. |
1518 |
I changed fs/tomoyo_file.c not to try to calculate pathnames |
1519 |
if vfsmount is NULL. |
1520 |
|
1521 |
Version 1.6.3 2008/07/15 Bug fix release. |
1522 |
|
1523 |
Fix 2008/08/21 |
1524 |
|
1525 |
@ Add workaround for gcc 4.3's bug. |
1526 |
|
1527 |
In some environments, fs/tomoyo_network.c could not be compiled |
1528 |
because of gcc 4.3's bug. |
1529 |
I modified save_ipv6_address() to use "integer literal" value |
1530 |
instead for "static const u8" variable. |
1531 |
|
1532 |
@ Change prototypes of some functions. |
1533 |
|
1534 |
To support 2.6.27 kernels, I replaced "struct nameidata" with |
1535 |
"struct path" for some functions. |
1536 |
|
1537 |
@ Detect distributor specific patches automatically. |
1538 |
|
1539 |
Since kernels with AppArmor patch applied is increasing, |
1540 |
I introduced a mechanism which determines whether specific patches |
1541 |
are applied or not, based on "#define" directives in the patches. |
1542 |
|
1543 |
Fix 2008/08/29 |
1544 |
|
1545 |
@ Remove "-ccs" suffix from Makefile's EXTRAVERSION. |
1546 |
|
1547 |
To reduce conflicts on Makefile's EXTRAVERSION, |
1548 |
I removed "-ccs" suffix from ccs-patch-2.\*.diff . |
1549 |
Those who build kernels without using specs/build-\*.sh , |
1550 |
please edit EXTRAVERSION tag manually so that original kernels |
1551 |
will not be overwritten by TOMOYO Linux kernels. |
1552 |
|
1553 |
Version 1.6.4 2008/09/03 Minor update release. |
1554 |
|
1555 |
Fix 2008/09/09 |
1556 |
|
1557 |
@ Add "try again" response to "delayed enforcing" mode. |
1558 |
|
1559 |
To be able to handle pathname changes caused by software updates, |
1560 |
"delayed enforcing" mode was introduced. It allows administrator to |
1561 |
grant access requests which are about to be rejected by the kernel. |
1562 |
|
1563 |
To be able to handle pathname changes caused by software updates better, |
1564 |
I introduced "try again" response. As "delayed enforcing" mode sleeps |
1565 |
a process which violated policy, administrator can update policy while |
1566 |
the process is sleeping. This "try again" response allows administrator |
1567 |
to restart policy checks from the beginning after updating policy. |
1568 |
|
1569 |
Fix 2008/09/11 |
1570 |
|
1571 |
@ Remember whether the process is allowed to write to /proc/ccs/ interface. |
1572 |
|
1573 |
Since programs for manipulating policy (e.g. ccs-queryd ) are installed |
1574 |
in the form of RPM/DEB packages, these programs lose the original |
1575 |
pathnames when they are updated by the package manager. The package |
1576 |
manager renames these programs before deleting these programs so that |
1577 |
the package manager can rollback the operation. |
1578 |
This causes a problem when the programs are listed into /proc/ccs/manager |
1579 |
using pathnames, as the programs will no longer be allowed to write to |
1580 |
/proc/ccs/ interface while the process of old version of the program is |
1581 |
alive. |
1582 |
|
1583 |
To solve this problem, I modified to remember the fact that the process |
1584 |
is once allowed to write to /proc/ccs/ interface until the process |
1585 |
attempts to execute a different program. |
1586 |
This change makes it impossible to revoke permission to write to |
1587 |
/proc/ccs/ interface without killing the process, but it will be better |
1588 |
than nonfunctioning ccs-queryd program. |
1589 |
|
1590 |
Fix 2008/09/19 |
1591 |
|
1592 |
@ Allow selecting a domain by PID. |
1593 |
|
1594 |
Sometimes we want to know what ACLs are given to specific PID, but |
1595 |
finding a domainname for that PID from /proc/ccs/.process_status and |
1596 |
reading ACLs from /proc/ccs/domain_policy by the domainname is very slow. |
1597 |
Thus, I modified /proc/ccs/domain_policy to allow selecting a domain by |
1598 |
PID. For example, to read domain ACL of current process from bash, |
1599 |
run as follows. |
1600 |
|
1601 |
# exec 100<>/proc/ccs/domain_policy |
1602 |
# echo select pid=$$ >&100 |
1603 |
# while read -u 100; do echo $REPLY; done |
1604 |
|
1605 |
If a domain is once selected by PID, reading /proc/ccs/domain_policy will |
1606 |
print only that domain if that PID exists or print nothing otherwise. |
1607 |
|
1608 |
@ Disallow concurrent /proc/ccs/ access using the same file descriptor. |
1609 |
|
1610 |
Until now, one process can read() from /proc/ccs/ while other process |
1611 |
that shares the file descriptor can write() to /proc/ccs/ . |
1612 |
But to implement "Allow selecting a domain by PID" feature, I disabled |
1613 |
concurrent read()/write() because the feature need to modify read buffer |
1614 |
while writing. |
1615 |
|
1616 |
Fix 2008/10/01 |
1617 |
|
1618 |
@ Add retry counter into /proc/ccs/query . |
1619 |
|
1620 |
To be able to handle some of queries from /proc/ccs/query without user's |
1621 |
interaction, I added retry counter for avoiding infinite loop caused by |
1622 |
"try again" response. |
1623 |
|
1624 |
Fix 2008/10/07 |
1625 |
|
1626 |
@ Don't transit to new domain until do_execve() succeeds. |
1627 |
|
1628 |
Until now, a process's domain was updated to new domain which the process |
1629 |
will belong to before do_execve() succeeds so that the kernel can do |
1630 |
permission checks for interpreters and environment variables based on |
1631 |
new domain. But this caused a subtle problem when other process sends |
1632 |
signals to the process, for the process returns to old domain if |
1633 |
do_execve() failed. |
1634 |
|
1635 |
So, I modified to pass new domain to functions so that I can avoid |
1636 |
modifying a process's domain before do_execve() succeeds. |
1637 |
|
1638 |
@ Use old task state for audit logs. |
1639 |
|
1640 |
Until now, audit logs were generated using the task state after |
1641 |
processing "; set task.state" part. But to generate accurate logs, |
1642 |
I modified to save the task state before processing "; set task.state" |
1643 |
part and use the saved state for audit logs. |
1644 |
|
1645 |
@ Use a structure for passing parameters. |
1646 |
|
1647 |
As the number of parameters is increasing, I modified to use a structure |
1648 |
for passing parameters. |
1649 |
|
1650 |
Fix 2008/10/11 |
1651 |
|
1652 |
@ Remove domain_acl_lock mutex. |
1653 |
|
1654 |
I noticed that I don't need to keep all functions that modify an ACL of |
1655 |
a domain mutually exclusive. Since each functions handles different type |
1656 |
of ACL, locking is needed only when they append an ACL to a domain. |
1657 |
So, I modified to use local locks. |
1658 |
|
1659 |
Fix 2008/10/14 |
1660 |
|
1661 |
@ Fix ccs_check_condition() bug. |
1662 |
|
1663 |
Due to a bug in ccs_check_condition(), it was impossible to use |
1664 |
task.state[0] task.state[1] task.state[2] inside condition part |
1665 |
if the ACL does not treat a pathname. For example, an ACL like |
1666 |
|
1667 |
allow_network TCP connect @HTTP_SERVERS 80 if task.state[0]=100 |
1668 |
|
1669 |
didn't work. |
1670 |
|
1671 |
Fix 2008/10/15 |
1672 |
|
1673 |
@ Show process information in /proc/ccs/.process_status . |
1674 |
|
1675 |
To be able to determine a process's type, I added a command "info PID" |
1676 |
which returns process information of the specified PID in |
1677 |
"PID manager=\* execute_handler=\* state[0]=\$ state[1]=\$ state[2]=\$" |
1678 |
format. |
1679 |
|
1680 |
Fix 2008/10/20 |
1681 |
|
1682 |
@ Use rcu_dereference() when walking the list. |
1683 |
|
1684 |
I was using "dependency ordering" for appending an element to a list |
1685 |
without asking the reader to take a lock. But "dependency ordering" |
1686 |
is not respected by DEC Alpha or by some aggressive value-speculation |
1687 |
compiler optimizations. |
1688 |
|
1689 |
On such environment, use of "dependency ordering" can lead to system |
1690 |
crash because the reader might read uninitialized value of newly |
1691 |
appended element. |
1692 |
|
1693 |
To prevent the reader from reading uninitialized value of newly appended |
1694 |
element, I inserted rcu_dereference() when walking the list. |
1695 |
|
1696 |
Fix 2008/11/04 |
1697 |
|
1698 |
@ Use sys_getpid() instead for current->pid. |
1699 |
|
1700 |
Kernel 2.6.24 introduced PID namespace. |
1701 |
|
1702 |
To compare PID given from userland, I can't use current->pid. |
1703 |
So, I modified to use sys_getpid() instead for current->pid. |
1704 |
|
1705 |
I modified to use task_tgid_nr_ns() for 2.6.25 and later instead for |
1706 |
current->tgid when checking /proc/self/ in get_absolute_path(). |
1707 |
|
1708 |
Fix 2008/11/07 |
1709 |
|
1710 |
@ Fix is_alphabet_char(). |
1711 |
|
1712 |
is_alphabet_char() should match 'A' - 'Z' and 'a' - 'z', |
1713 |
but was matching from 'A' - 'F' and 'a' - 'f'. |
1714 |
|
1715 |
@ Add /proc/ccs/.execute_handler . |
1716 |
|
1717 |
Process information became visible to userspace by |
1718 |
"Show process information in /proc/ccs/.process_status" feature. |
1719 |
However, programs specified by execute_handler directive may run as |
1720 |
non root user, making it impossible to see process information. |
1721 |
|
1722 |
So, I added a new interface that allows execute handler processes |
1723 |
to see process information. The content of /proc/ccs/.execute_handler is |
1724 |
identical to /proc/ccs/.process_status . |
1725 |
|
1726 |
Version 1.6.5 2008/11/11 Third anniversary release. |
1727 |
|
1728 |
Fix 2008/12/01 |
1729 |
|
1730 |
@ Introduce "task.type=execute_handler" condition. |
1731 |
|
1732 |
The execute_handler directive is very very powerful. You can use this |
1733 |
directive to do anything you want to do (e.g. logging and validating and |
1734 |
modifying command line parameters and environment variables, opening and |
1735 |
closing and redirecting files, creating pipes to implement antivirus and |
1736 |
spam filtering, deploying a DMZ between the ssh daemon and the login |
1737 |
shells). |
1738 |
|
1739 |
To be able to use this directive in a domain with keep_domain directive |
1740 |
while limiting access to resources needed for such purposes to only |
1741 |
programs invoked as an execute handler process, I added a new condition. |
1742 |
|
1743 |
In learning mode, "if task.type=execute_handler" condition part will be |
1744 |
automatically added for requests issued by an execute_handler process. |
1745 |
|
1746 |
@ Introduce file's type and permissions as conditions. |
1747 |
|
1748 |
To be able to limit file types a process can access, I added |
1749 |
new conditions for checking file's type and permissions. |
1750 |
For example, |
1751 |
|
1752 |
allow_read /etc/fstab if path1.type=file path1.perm=0644 |
1753 |
|
1754 |
will allow opening /etc/fstab for reading only if /etc/fstab is a regular |
1755 |
file and it's permission is 0644, and |
1756 |
|
1757 |
allow_write /dev/null if path1.type=char path1.dev_major=1 path1.dev_minor=3 |
1758 |
|
1759 |
will allow opening /dev/null for writing only if /dev/null is a character |
1760 |
device file with major=1 and minor=3 attributes. |
1761 |
|
1762 |
@ Add memory quota for temporary memory used for auditing. |
1763 |
|
1764 |
Although there are MAX_GRANT_LOG and MAX_REJECT_LOG parameters |
1765 |
which limit the number of entries for audit logs so that we can avoid |
1766 |
memory consumption by audit logs, it would be more convenient if we can |
1767 |
also limit the size in bytes. |
1768 |
Thus, I added a new quota line. |
1769 |
|
1770 |
echo Dynamic: 1048576 > /proc/ccs/meminfo |
1771 |
|
1772 |
This quota is not applied to temporary memory used for permission checks. |
1773 |
|
1774 |
Fix 2008/12/09 |
1775 |
|
1776 |
@ Fix ccs_can_save_audit_log() checks. |
1777 |
|
1778 |
Due to incorrect statement "if (ccs_can_save_audit_log() < 0)" |
1779 |
while ccs_can_save_audit_log() is boolean, MAX_GRANT_LOG and |
1780 |
MAX_REJECT_LOG were not working. |
1781 |
|
1782 |
This bug will trigger OOM killer if /usr/sbin/ccs-auditd is not working. |
1783 |
|
1784 |
Fix 2008/12/24 |
1785 |
|
1786 |
@ Add "ccs_" prefix. |
1787 |
|
1788 |
To be able to tell whether a symbol is TOMOYO Linux related or not, |
1789 |
I added "ccs_" prefix as much as possible. |
1790 |
|
1791 |
@ Fix ccs_check_flags() error message. |
1792 |
|
1793 |
I meant to print SYAORAN-ERROR: message when error == -EPERM, |
1794 |
but I was printing it when error == 0 since 1.6.0 . |
1795 |
|
1796 |
Fix 2009/01/05 |
1797 |
|
1798 |
@ Use kmap_atomic()/kunmap_atomic() for reading "struct linux_binprm". |
1799 |
|
1800 |
As remove_arg_zero() uses kmap_atomic(KM_USER0), I modified to use |
1801 |
kmap_atomic(KM_USER0) rather than kmap(). |
1802 |
|
1803 |
Fix 2009/01/28 |
1804 |
|
1805 |
@ Fix "allow_read" + "allow_write" != "allow_read/write" problem. |
1806 |
|
1807 |
Since 1.6.0 , due to a bug in ccs_update_single_path_acl(), |
1808 |
appending "allow_read/write" entry didn't update internal "allow_read" |
1809 |
and "allow_write" entries. As a result, attempt to open(O_RDWR) succeeds |
1810 |
but open(O_RDONLY) and open(O_WRONLY) fail. |
1811 |
|
1812 |
Workaround is to write an entry twice when newly appending that entry. |
1813 |
If written twice, internal "allow_read" and "allow_write" entries |
1814 |
are updated. |
1815 |
|
1816 |
Fix 2009/02/26 |
1817 |
|
1818 |
@ Fix profile read error. |
1819 |
|
1820 |
Incorrect profiles were shown in /proc/ccs/profile |
1821 |
if either CONFIG_SAKURA or CONFIG_TOMOYO is disabled. |
1822 |
|
1823 |
Fix 2009/03/02 |
1824 |
|
1825 |
@ Undelete CONFIG_TOMOYO_AUDIT option. |
1826 |
|
1827 |
While HDD-less systems can use profiles with MAX_GRANT_LOG=0 and |
1828 |
MAX_REJECT_LOG=0 , I undeleted CONFIG_TOMOYO_AUDIT option for saving |
1829 |
memory used for /proc/ccs/grant_log and /proc/ccs/reject_log interfaces. |
1830 |
|
1831 |
Fix 2009/03/13 |
1832 |
|
1833 |
@ Show only profile entry names ever specified. |
1834 |
|
1835 |
Even if an administrator specifies only COMMENT= and MAC_FOR_FILE= |
1836 |
entries for /proc/ccs/profile , all available profile entries are shown. |
1837 |
This was designed to help administrators to know what entries are |
1838 |
available, but sometimes makes administrators feel noisy because of |
1839 |
entries showing default values. |
1840 |
|
1841 |
Thus, I modified to show only profile entry names ever specified. |
1842 |
|
1843 |
Fix 2009/03/18 |
1844 |
|
1845 |
@ Add MAC_FOR_IOCTL functionality. |
1846 |
|
1847 |
To be able to restrict ioctl() requests, I added MAC_FOR_IOCTL |
1848 |
functionality. |
1849 |
|
1850 |
This functionality requires modification of ccs-patch-\*.diff . |
1851 |
|
1852 |
@ Use better name for socket's pathname. |
1853 |
|
1854 |
Until now, socket's pathname was represented as "socket:[\$]" format |
1855 |
where \$ is inode's number. But inode's number is useless for name based |
1856 |
access control. Therefore, I modified to represent socket's pathname as |
1857 |
"socket:[family=\$:type=\$:protocol=\$]" format. |
1858 |
|
1859 |
This will help administrator to control ioctl() against sockets more |
1860 |
precisely. |
1861 |
|
1862 |
@ Fix misplaced ccs_capable() call. (only 2.6.8-\* and 2.6.9-\*) |
1863 |
|
1864 |
Location to insert ccs_capable(TOMOYO_SYS_IOCTL) in sys_ioctl() was |
1865 |
wrong since version 1.1 . |
1866 |
|
1867 |
@ Insert ccs_check_ioctl_permission() call. |
1868 |
|
1869 |
To make MAC_FOR_IOCTL functionality working, I inserted |
1870 |
ccs_check_ioctl_permission() call into ccs-patch-\*.diff . |
1871 |
|
1872 |
Fix 2009/03/23 |
1873 |
|
1874 |
@ Move sysctl()'s check from ccs-patch-\*.diff to fs/tomoyo_file.c . |
1875 |
|
1876 |
Since try_parse_table() in kernel/sysctl.c is almost identical between |
1877 |
all versions, I moved that function to fs/tomoyo_file.c . |
1878 |
|
1879 |
@ Relocate definitions and functions. |
1880 |
|
1881 |
To reduce exposed symbols, I relocated some definitions and functions. |
1882 |
|
1883 |
Fix 2009/03/24 |
1884 |
|
1885 |
@ Add CONFIG_TOMOYO_BUILTIN_INITIALIZERS option. |
1886 |
|
1887 |
Some systems don't have /sbin/modprobe and /sbin/hotplug . |
1888 |
Thus, I made these pathnames configurable. |
1889 |
|
1890 |
Version 1.6.7 2009/04/01 Feature enhancement release. |
1891 |
|
1892 |
Fix 2009/04/06 |
1893 |
|
1894 |
@ Drop "undelete domain" command. |
1895 |
|
1896 |
I added "undelete domain" command on 2007/01/19, but never used by policy |
1897 |
management tools. The garbage collector I added on 2007/01/29 will |
1898 |
automatically reuse memory and allow administrators switch domain policy |
1899 |
periodically, provided that the administrator kills processes in old |
1900 |
domains before recreating new domains with the same domainnames. |
1901 |
|
1902 |
Thus, I dropped "undelete domain" command. |
1903 |
|
1904 |
@ Escape invalid characters in ccs_check_mount_permission2(). |
1905 |
|
1906 |
ccs_check_mount_permission2() was passing unencoded strings to printk() |
1907 |
and ccs_update_mount_acl() and ccs_check_supervisor(). This may cause |
1908 |
/proc/ccs/system_policy and /proc/ccs/query to contain invalid |
1909 |
characters within a string. |
1910 |
|
1911 |
Fix 2009/04/07 |
1912 |
|
1913 |
@ Fix IPv4's "address_group" handling error. |
1914 |
|
1915 |
Since 1.6.5 , due to lack of ntohl() (byte order conversion) in |
1916 |
ccs_update_address_group_entry(), "address_group" with IPv4 address was |
1917 |
not working. |
1918 |
|
1919 |
This problem happens on little endian platforms (e.g. x86). |
1920 |
|
1921 |
Fix 2009/05/08 |
1922 |
|
1923 |
@ Add condition for symlink's target pathname. |
1924 |
|
1925 |
Until now, "allow_symlink" keyword allows creation of a symlink but does |
1926 |
not check the symlink's target. Usually it is no problem because |
1927 |
permission checks are done using dereferenced pathname. But in some |
1928 |
cases, we should restrict the symlink's target. For example, |
1929 |
"ln -s .htpasswd /var/www/html/readme.html" by CGI program should be |
1930 |
blocked because we will allow Apache to read both |
1931 |
/var/www/html/readme.html and /var/www/html/.htpasswd . |
1932 |
|
1933 |
Thus, I added new condition, "symlink.target". |
1934 |
|
1935 |
allow_symlink /var/www/html/\*.html if symlink.target="\*.html" |
1936 |
|
1937 |
allow_symlink /var/www/html/\*\-.\* if symlink.target="\*\-.\*" |
1938 |
|
1939 |
@ Don't return -EAGAIN at ccs_socket_recvmsg_permission(). |
1940 |
|
1941 |
It turned out that it is not permitted for accept() and recvmsg() to |
1942 |
return -EAGAIN if poll() said connections/datagrams are ready. However, |
1943 |
recvmsg() may return -EAGAIN and potentially confuse some applications |
1944 |
because ccs_socket_recvmsg_permission() is returning -EAGAIN. |
1945 |
|
1946 |
Thus, I modified ccs_socket_recvmsg_permission() to return -ENOMEM |
1947 |
rather than -EAGAIN. |
1948 |
|
1949 |
Fix 2009/05/19 |
1950 |
|
1951 |
@ Don't call get_fs_type() with a mutex held. |
1952 |
|
1953 |
Until now, when ccs_update_mount_acl() is called with unsupported |
1954 |
filesystem, /sbin/modprobe is executed from get_fs_type() to load |
1955 |
filesystem module. And get_fs_type() does not return until /sbin/modprobe |
1956 |
finishes. |
1957 |
|
1958 |
This means that it will cause deadlock if /sbin/modprobe (which is |
1959 |
executed via get_fs_type() in ccs_update_mount_acl()) calls |
1960 |
ccs_update_mount_acl(); although it won't happen unless an administrator |
1961 |
inserts execute_handler to call mount() requests in learning mode or to |
1962 |
add "allow_mount" entries to /proc/ccs/system_policy . |
1963 |
|
1964 |
I modified to unlock the mutex before calling get_fs_type(). |
1965 |
|
1966 |
Fix 2009/05/20 |
1967 |
|
1968 |
@ Update recvmsg() hooks. |
1969 |
|
1970 |
Since 1.5.0, I was doing network access control for incoming UDP and RAW |
1971 |
packets inside skb_recv_datagram(). But to synchronize with LSM version, |
1972 |
I moved ccs_recv_datagram_permission() hook from skb_recv_datagram() to |
1973 |
udp_recvmsg()/udpv6_recvmsg()/raw_recvmsg()/rawv6_recvmsg() with name |
1974 |
change to ccs_recvmsg_permission(). |
1975 |
|
1976 |
Version 1.6.8 2009/05/28 Feature enhancement release. |
1977 |
|
1978 |
Fix 2009/07/03 |
1979 |
|
1980 |
@ Fix buffer overrun when used with CONFIG_SLOB=y . |
1981 |
|
1982 |
Since 1.6.7 , ccs_allocate_execve_entry() was requesting for only 4000 |
1983 |
bytes while the comment says it is 4096 bytes. This may lead to buffer |
1984 |
overrun when slob allocator is used, for slob allocator allocates exactly |
1985 |
4000 bytes whereas slab and slub allocators allocate 4096 bytes. |
1986 |
|
1987 |
Fix 2009/09/01 |
1988 |
|
1989 |
@ Add garbage collector support. |
1990 |
|
1991 |
Until now, it was impossible to release memory used by deleted policy. |
1992 |
I added SRCU based garbage collector so that memory used by deleted |
1993 |
policy will be automatically released. |
1994 |
|
1995 |
@ Remove word length limitation and line length limitation. |
1996 |
|
1997 |
Until now, the max length of a word is 4000 and the max length of a line |
1998 |
is 8192. To be able to handle longer pathnames, I removed these |
1999 |
limitations. Now, the max length (except the domainname and |
2000 |
argv[]/envp[]) is 128K (which is the max amount of memory kmalloc() |
2001 |
can allocate in most environments). |
2002 |
|
2003 |
@ Support more fine grained profile configuration. |
2004 |
|
2005 |
Profile was reconstructed. |
2006 |
|
2007 |
@ Support more fine grained parameters restrictions. |
2008 |
|
2009 |
"allow_create", "allow_mkdir", "allow_mkfifo", "allow_mksock" check |
2010 |
create mode. "allow_mkblock" and "allow_mkchar" check create mode and |
2011 |
major/minor device numbers. "allow_chmod" check new mode. "allow_chown" |
2012 |
checks new owner. "allow_chgrp" checks new group. |
2013 |
|
2014 |
@ Allow number grouping. |
2015 |
|
2016 |
To help specifying numeric values, a new directive "number_group" is |
2017 |
introduced. |
2018 |
|
2019 |
@ Remove "alias" directive and "allow_argv0" directive. |
2020 |
|
2021 |
Until now, "allow_execute" used dereferenced pathname if it is a symlink |
2022 |
unless explicitly specified by "alias" directive. |
2023 |
|
2024 |
Now, "allow_execute" uses symlink's pathname if it is a symlink. |
2025 |
"exec.realpath" in "if" clause checks the dereferenced pathname. |
2026 |
"exec.argv[0]" in "if" clause checks the invocation name. |
2027 |
|
2028 |
@ Remove /proc/ccs/system_policy and /etc/ccs/system_policy.conf . |
2029 |
|
2030 |
"deny_autobind" was moved to /proc/ccs/exception_policy and |
2031 |
/etc/ccs/exception_policy.conf . Other directives were moved to |
2032 |
/proc/ccs/domain_policy and /etc/ccs/domain_policy.conf . |
2033 |
|
2034 |
@ Remove syaoran filesystem. |
2035 |
|
2036 |
Since "allow_create"/"allow_mkdir"/"allow_mkfifo"/"allow_mksock"/ |
2037 |
"allow_mkblock"/"allow_mkchar"/"allow_chmod"/"allow_chown"/"allow_chgrp" |
2038 |
can restrict mode changes and owner/group changes, there is no need to |
2039 |
restrict these changes at filesystem level. |
2040 |
|
2041 |
Thus, I removed syaoran filesystem. |
2042 |
|
2043 |
@ Reduce spinlocks. |
2044 |
|
2045 |
Until now, TOMOYO was using own list for detecting memory leak. But as |
2046 |
kernel 2.6.31 introduced memory leak detection mechanism |
2047 |
( CONFIG_DEBUG_KMEMLEAK ), TOMOYO no longer needs to use own list. |
2048 |
|
2049 |
I removed the list to reduce use of spinlocks. |
2050 |
|
2051 |
@ Rewrite ccs-patch-2.\*.diff . |
2052 |
|
2053 |
ccs-patch-2.\*.diff was rewritten like LSM hooks. |
2054 |
|
2055 |
@ Don't check "allow_read/write" for open-for-ioctl-only. |
2056 |
|
2057 |
open(pathname, 3) means open for ioctl() only. |
2058 |
Until now, TOMOYO was checking "allow_read/write" for open(pathname, 3). |
2059 |
But since TOMOYO checks "allow_ioctl" for ioctl(), I modified not to |
2060 |
require "allow_read/write" for open(pathname, 3). |
2061 |
|
2062 |
@ Add missing sigqueue() and tgsigqueue() hooks. |
2063 |
|
2064 |
Until now, kill(), tkill(), tgkill() had hooks but sigqueue() and |
2065 |
tgsigqueue() didn't. |
2066 |
|
2067 |
@ Move files from fs/ to security/ccsecurity. |
2068 |
|
2069 |
Config menu section changed from "File systems" to "Security options". |
2070 |
|
2071 |
Kernel config symbols changed from CONFIG_SAKURA CONFIG_TOMOYO |
2072 |
CONFIG_SYAORAN to CONFIG_CCSECURITY . |
2073 |
|
2074 |
@ Add global PID to audit logs. |
2075 |
|
2076 |
ccs-queryd was using domainname for reaching the domain which the process |
2077 |
belongs to, but the domain could be deleted while ccs-queryd is handling |
2078 |
policy violation. If the domain is deleted, ccs-queryd no longer can |
2079 |
reach the domain by domainname. Thus, ccs-queryd now uses PID for |
2080 |
reaching the domain which the process belongs to. |
2081 |
|
2082 |
Kernel 2.6.24 introduced PID namespace. The PID in access logs generated |
2083 |
by a process inside a container is useless for ccs-queryd for reaching |
2084 |
the domain which the process belongs to. |
2085 |
|
2086 |
Thus, I added global PID in audit logs. |
2087 |
|
2088 |
@ Transit to new domain before do_execve() succeeds. |
2089 |
|
2090 |
Permission checks for interpreters and environment variables are |
2091 |
done using new domain. In order to be allow ccs-queryd to reach the new |
2092 |
domain via global PID, I reverted "Don't transit to new domain until |
2093 |
do_execve() succeeds." made on 2008/10/07. |
2094 |
|
2095 |
Version 1.7.0 2009/09/03 Feature enhancement release. |
2096 |
|
2097 |
Fix 2009/09/04 |
2098 |
|
2099 |
@ Fix wrong ccs_profile() calls. |
2100 |
|
2101 |
I can't call ccs_profile() for profile existence test because |
2102 |
ccs_profile() never returns NULL. |
2103 |
|
2104 |
Fix 2009/09/06 |
2105 |
|
2106 |
@ Fix wrong error code in ccs_try_alt_exec(). |
2107 |
|
2108 |
ccs_try_alt_exec() was returning ENOMEM when kmalloc() failed. |
2109 |
It needs to return -ENOMEM to fail. |
2110 |
|
2111 |
Fix 2009/09/10 |
2112 |
|
2113 |
@ Do not check umount() permission for mount(MS_MOVE) requests. |
2114 |
|
2115 |
Until 1.6.x , umount() restriction was black listing. In 1.7.0 , it is |
2116 |
white listing. This change caused "mount --move old new" requests to |
2117 |
require "allow_unmount old" permission in addition to |
2118 |
"allow_mount old new --move 0" permission. |
2119 |
But we don't want to allow umount(old) requests when we want to allow |
2120 |
only mount(old, new, MS_MOVE) requests. Thus, I modified not to check |
2121 |
"allow_unmount old" permission for mount(old, new, MS_MOVE) requests. |
2122 |
|
2123 |
Fix 2009/09/11 |
2124 |
|
2125 |
@ Support recursive match operators. |
2126 |
|
2127 |
Until now, ccs_path_matches_pattern() did not support recursive |
2128 |
comparison. Thus, users had to repeat "/\*" when they want to specify |
2129 |
recursively. |
2130 |
|
2131 |
I introduced "\{" and "\}" as repetition operator. |
2132 |
To ensure consistency with TOMOYO's '/'-tokenized pattern matching rules |
2133 |
and "\-" operator, only "/\{dir\}/" sequences (where dir does not contain |
2134 |
'/') is permitted. |
2135 |
|
2136 |
Fix 2009/09/24 |
2137 |
|
2138 |
@ Don't check chmod/chown capability for requests from kernel. |
2139 |
|
2140 |
Until now, ccs_setattr_permission() was inserted in notify_change(). |
2141 |
But notify_change() is also called by requests from kernel (e.g. UnionFS) |
2142 |
and it made difficult to use TOMOYO on UnionFS. |
2143 |
|
2144 |
Thus, I moved ccs_capable() checks from ccs_setattr_permission() to |
2145 |
ccs_chmod_permission() and ccs_chown_permission(), and removed |
2146 |
ccs_setattr_permission(). |
2147 |
|
2148 |
Fix 2009/09/25 |
2149 |
|
2150 |
@ Embed more information into audit logs. |
2151 |
|
2152 |
Until now, /proc/ccs/grant_log /proc/ccs/reject_log /proc/ccs/query were |
2153 |
not printing file's information (e.g. file's uid/gid/mode). |
2154 |
|
2155 |
Recently, users who started using "if" clause expect that the learning |
2156 |
mode automatically adds various conditions like "if task.uid=path1.uid". |
2157 |
|
2158 |
But the profile will become too complicated if I support all possible |
2159 |
conditions. Thus, I added all information which is enough to generate |
2160 |
"if" clause with all possible conditions from audit logs. |
2161 |
|
2162 |
Now, the learning mode got different usage. Users can specify |
2163 |
"CONFIG::learning={ max_entry=0 }" in the profile. All requests which |
2164 |
are not permitted by policy will be sent to /proc/ccs/reject_log with |
2165 |
"mode=learning" header lines. Users can selectively append conditions |
2166 |
and append to the policy using "/usr/sbin/ccs-loadpolicy -d". |
2167 |
The learning mode with "CONFIG::learning={ max_entry=0 }" is almost |
2168 |
the same with the permissive mode, only difference is "mode=learning" |
2169 |
and "mode=permissive". |
2170 |
|
2171 |
Fix 2009/10/05 |
2172 |
|
2173 |
@ Fix size truncation bug at ccs_memcmp(). |
2174 |
|
2175 |
ccs_memcmp() was using "u8" for size parameter by error. Therefore, when |
2176 |
size >= 256 was passed to ccs_memcmp(), it was doing partial comparison |
2177 |
(incorrect result) or read overrun (CPU stall). |
2178 |
|
2179 |
ccs_memcmp() should use "size_t" for size parameter because size of |
2180 |
"struct ccs_condition" may exceed 256 bytes if complicated condition was |
2181 |
given. |
2182 |
|
2183 |
Fix 2009/10/08 |
2184 |
|
2185 |
@ Add CONFIG_CCSECURITY_DEFAULT_LOADER option. |
2186 |
|
2187 |
I made the default policy loader's pathname ( /sbin/ccs-init ) |
2188 |
configurable. |
2189 |
|
2190 |
@ Add CONFIG_CCSECURITY_ALTERNATIVE_TRIGGER option. |
2191 |
|
2192 |
Some environments do not have /sbin/init . In such environments, we need |
2193 |
to use different program's pathname (e.g. /init or /linuxrc ) as |
2194 |
activation trigger. |
2195 |
|
2196 |
Thus, I made the alternative trigger ( /sbin/ccs-start ) configurable. |
2197 |
|
2198 |
Fix 2009/11/02 |
2199 |
|
2200 |
@ Fix buffer contention. |
2201 |
|
2202 |
A permission like |
2203 |
|
2204 |
allow_env PATH if exec.envp["PATH"]="/" |
2205 |
|
2206 |
was not working since I was using the same buffer for both environment |
2207 |
variable's name and value. |
2208 |
|
2209 |
Fix 2009/11/03 |
2210 |
|
2211 |
@ Fix memory leak in ccs_write_address_group_policy(). |
2212 |
|
2213 |
I forgot to call kfree() if same entry was added. |
2214 |
|
2215 |
@ Reduce mutexes. |
2216 |
|
2217 |
I was using mutex_lock()/mutex_unlock() so that I can use |
2218 |
atomic_dec_and_test() for removing an element from a list. |
2219 |
I moved that operation to garbage collector in order to reduce frequency |
2220 |
of mutex_lock()/mutex_unlock() calls. |
2221 |
|
2222 |
@ Escape from nested loops correctly. |
2223 |
|
2224 |
In ccs_read_address_group_policy(), I was escaping from nested loops |
2225 |
correctly. But in ccs_read_path_group_policy() and |
2226 |
ccs_read_number_group_policy(), I wasn't. |
2227 |
|
2228 |
As a result, reading path_group and number_group caused kernel oops |
2229 |
when they were not read atomically. |