2005-12-15 11:33:52 -07:00
|
|
|
/* auditfilter.c -- filtering of audit events
|
|
|
|
*
|
|
|
|
* Copyright 2003-2004 Red Hat, Inc.
|
|
|
|
* Copyright 2005 Hewlett-Packard Development Company, L.P.
|
|
|
|
* Copyright 2005 IBM Corporation
|
|
|
|
*
|
|
|
|
* This program is free software; you can redistribute it and/or modify
|
|
|
|
* it under the terms of the GNU General Public License as published by
|
|
|
|
* the Free Software Foundation; either version 2 of the License, or
|
|
|
|
* (at your option) any later version.
|
|
|
|
*
|
|
|
|
* This program is distributed in the hope that it will be useful,
|
|
|
|
* but WITHOUT ANY WARRANTY; without even the implied warranty of
|
|
|
|
* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
|
|
|
|
* GNU General Public License for more details.
|
|
|
|
*
|
|
|
|
* You should have received a copy of the GNU General Public License
|
|
|
|
* along with this program; if not, write to the Free Software
|
|
|
|
* Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
|
|
|
|
*/
|
|
|
|
|
|
|
|
#include <linux/kernel.h>
|
|
|
|
#include <linux/audit.h>
|
|
|
|
#include <linux/kthread.h>
|
[PATCH] audit: path-based rules
In this implementation, audit registers inotify watches on the parent
directories of paths specified in audit rules. When audit's inotify
event handler is called, it updates any affected rules based on the
filesystem event. If the parent directory is renamed, removed, or its
filesystem is unmounted, audit removes all rules referencing that
inotify watch.
To keep things simple, this implementation limits location-based
auditing to the directory entries in an existing directory. Given
a path-based rule for /foo/bar/passwd, the following table applies:
passwd modified -- audit event logged
passwd replaced -- audit event logged, rules list updated
bar renamed -- rule removed
foo renamed -- untracked, meaning that the rule now applies to
the new location
Audit users typically want to have many rules referencing filesystem
objects, which can significantly impact filtering performance. This
patch also adds an inode-number-based rule hash to mitigate this
situation.
The patch is relative to the audit git tree:
http://kernel.org/git/?p=linux/kernel/git/viro/audit-current.git;a=summary
and uses the inotify kernel API:
http://lkml.org/lkml/2006/6/1/145
Signed-off-by: Amy Griffis <amy.griffis@hp.com>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
2006-04-07 14:55:56 -06:00
|
|
|
#include <linux/mutex.h>
|
|
|
|
#include <linux/fs.h>
|
|
|
|
#include <linux/namei.h>
|
2005-12-15 11:33:52 -07:00
|
|
|
#include <linux/netlink.h>
|
[PATCH] audit: path-based rules
In this implementation, audit registers inotify watches on the parent
directories of paths specified in audit rules. When audit's inotify
event handler is called, it updates any affected rules based on the
filesystem event. If the parent directory is renamed, removed, or its
filesystem is unmounted, audit removes all rules referencing that
inotify watch.
To keep things simple, this implementation limits location-based
auditing to the directory entries in an existing directory. Given
a path-based rule for /foo/bar/passwd, the following table applies:
passwd modified -- audit event logged
passwd replaced -- audit event logged, rules list updated
bar renamed -- rule removed
foo renamed -- untracked, meaning that the rule now applies to
the new location
Audit users typically want to have many rules referencing filesystem
objects, which can significantly impact filtering performance. This
patch also adds an inode-number-based rule hash to mitigate this
situation.
The patch is relative to the audit git tree:
http://kernel.org/git/?p=linux/kernel/git/viro/audit-current.git;a=summary
and uses the inotify kernel API:
http://lkml.org/lkml/2006/6/1/145
Signed-off-by: Amy Griffis <amy.griffis@hp.com>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
2006-04-07 14:55:56 -06:00
|
|
|
#include <linux/sched.h>
|
|
|
|
#include <linux/inotify.h>
|
Audit: use new LSM hooks instead of SELinux exports
Stop using the following exported SELinux interfaces:
selinux_get_inode_sid(inode, sid)
selinux_get_ipc_sid(ipcp, sid)
selinux_get_task_sid(tsk, sid)
selinux_sid_to_string(sid, ctx, len)
kfree(ctx)
and use following generic LSM equivalents respectively:
security_inode_getsecid(inode, secid)
security_ipc_getsecid*(ipcp, secid)
security_task_getsecid(tsk, secid)
security_sid_to_secctx(sid, ctx, len)
security_release_secctx(ctx, len)
Call security_release_secctx only if security_secid_to_secctx
succeeded.
Signed-off-by: Casey Schaufler <casey@schaufler-ca.com>
Signed-off-by: Ahmed S. Darwish <darwish.07@gmail.com>
Acked-by: James Morris <jmorris@namei.org>
Reviewed-by: Paul Moore <paul.moore@hp.com>
2008-03-01 12:54:38 -07:00
|
|
|
#include <linux/security.h>
|
2005-12-15 11:33:52 -07:00
|
|
|
#include "audit.h"
|
|
|
|
|
[PATCH] audit: path-based rules
In this implementation, audit registers inotify watches on the parent
directories of paths specified in audit rules. When audit's inotify
event handler is called, it updates any affected rules based on the
filesystem event. If the parent directory is renamed, removed, or its
filesystem is unmounted, audit removes all rules referencing that
inotify watch.
To keep things simple, this implementation limits location-based
auditing to the directory entries in an existing directory. Given
a path-based rule for /foo/bar/passwd, the following table applies:
passwd modified -- audit event logged
passwd replaced -- audit event logged, rules list updated
bar renamed -- rule removed
foo renamed -- untracked, meaning that the rule now applies to
the new location
Audit users typically want to have many rules referencing filesystem
objects, which can significantly impact filtering performance. This
patch also adds an inode-number-based rule hash to mitigate this
situation.
The patch is relative to the audit git tree:
http://kernel.org/git/?p=linux/kernel/git/viro/audit-current.git;a=summary
and uses the inotify kernel API:
http://lkml.org/lkml/2006/6/1/145
Signed-off-by: Amy Griffis <amy.griffis@hp.com>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
2006-04-07 14:55:56 -06:00
|
|
|
/*
|
|
|
|
* Locking model:
|
|
|
|
*
|
|
|
|
* audit_filter_mutex:
|
|
|
|
* Synchronizes writes and blocking reads of audit's filterlist
|
|
|
|
* data. Rcu is used to traverse the filterlist and access
|
|
|
|
* contents of structs audit_entry, audit_watch and opaque
|
2008-03-01 13:01:11 -07:00
|
|
|
* LSM rules during filtering. If modified, these structures
|
[PATCH] audit: path-based rules
In this implementation, audit registers inotify watches on the parent
directories of paths specified in audit rules. When audit's inotify
event handler is called, it updates any affected rules based on the
filesystem event. If the parent directory is renamed, removed, or its
filesystem is unmounted, audit removes all rules referencing that
inotify watch.
To keep things simple, this implementation limits location-based
auditing to the directory entries in an existing directory. Given
a path-based rule for /foo/bar/passwd, the following table applies:
passwd modified -- audit event logged
passwd replaced -- audit event logged, rules list updated
bar renamed -- rule removed
foo renamed -- untracked, meaning that the rule now applies to
the new location
Audit users typically want to have many rules referencing filesystem
objects, which can significantly impact filtering performance. This
patch also adds an inode-number-based rule hash to mitigate this
situation.
The patch is relative to the audit git tree:
http://kernel.org/git/?p=linux/kernel/git/viro/audit-current.git;a=summary
and uses the inotify kernel API:
http://lkml.org/lkml/2006/6/1/145
Signed-off-by: Amy Griffis <amy.griffis@hp.com>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
2006-04-07 14:55:56 -06:00
|
|
|
* must be copied and replace their counterparts in the filterlist.
|
|
|
|
* An audit_parent struct is not accessed during filtering, so may
|
|
|
|
* be written directly provided audit_filter_mutex is held.
|
|
|
|
*/
|
|
|
|
|
|
|
|
/*
|
|
|
|
* Reference counting:
|
|
|
|
*
|
|
|
|
* audit_parent: lifetime is from audit_init_parent() to receipt of an IN_IGNORED
|
|
|
|
* event. Each audit_watch holds a reference to its associated parent.
|
|
|
|
*
|
|
|
|
* audit_watch: if added to lists, lifetime is from audit_init_watch() to
|
|
|
|
* audit_remove_watch(). Additionally, an audit_watch may exist
|
|
|
|
* temporarily to assist in searching existing filter data. Each
|
|
|
|
* audit_krule holds a reference to its associated watch.
|
|
|
|
*/
|
|
|
|
|
|
|
|
struct audit_parent {
|
|
|
|
struct list_head ilist; /* entry in inotify registration list */
|
|
|
|
struct list_head watches; /* associated watches */
|
|
|
|
struct inotify_watch wdata; /* inotify watch data */
|
|
|
|
unsigned flags; /* status flags */
|
|
|
|
};
|
|
|
|
|
|
|
|
/*
|
|
|
|
* audit_parent status flags:
|
|
|
|
*
|
|
|
|
* AUDIT_PARENT_INVALID - set anytime rules/watches are auto-removed due to
|
|
|
|
* a filesystem event to ensure we're adding audit watches to a valid parent.
|
|
|
|
* Technically not needed for IN_DELETE_SELF or IN_UNMOUNT events, as we cannot
|
|
|
|
* receive them while we have nameidata, but must be used for IN_MOVE_SELF which
|
|
|
|
* we can receive while holding nameidata.
|
|
|
|
*/
|
|
|
|
#define AUDIT_PARENT_INVALID 0x001
|
|
|
|
|
|
|
|
/* Audit filter lists, defined in <linux/audit.h> */
|
2005-12-15 11:33:52 -07:00
|
|
|
struct list_head audit_filter_list[AUDIT_NR_FILTERS] = {
|
|
|
|
LIST_HEAD_INIT(audit_filter_list[0]),
|
|
|
|
LIST_HEAD_INIT(audit_filter_list[1]),
|
|
|
|
LIST_HEAD_INIT(audit_filter_list[2]),
|
|
|
|
LIST_HEAD_INIT(audit_filter_list[3]),
|
|
|
|
LIST_HEAD_INIT(audit_filter_list[4]),
|
|
|
|
LIST_HEAD_INIT(audit_filter_list[5]),
|
|
|
|
#if AUDIT_NR_FILTERS != 6
|
|
|
|
#error Fix audit_filter_list initialiser
|
|
|
|
#endif
|
|
|
|
};
|
|
|
|
|
[PATCH] audit: watching subtrees
New kind of audit rule predicates: "object is visible in given subtree".
The part that can be sanely implemented, that is. Limitations:
* if you have hardlink from outside of tree, you'd better watch
it too (or just watch the object itself, obviously)
* if you mount something under a watched tree, tell audit
that new chunk should be added to watched subtrees
* if you umount something in a watched tree and it's still mounted
elsewhere, you will get matches on events happening there. New command
tells audit to recalculate the trees, trimming such sources of false
positives.
Note that it's _not_ about path - if something mounted in several places
(multiple mount, bindings, different namespaces, etc.), the match does
_not_ depend on which one we are using for access.
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
2007-07-22 06:04:18 -06:00
|
|
|
DEFINE_MUTEX(audit_filter_mutex);
|
[PATCH] audit: path-based rules
In this implementation, audit registers inotify watches on the parent
directories of paths specified in audit rules. When audit's inotify
event handler is called, it updates any affected rules based on the
filesystem event. If the parent directory is renamed, removed, or its
filesystem is unmounted, audit removes all rules referencing that
inotify watch.
To keep things simple, this implementation limits location-based
auditing to the directory entries in an existing directory. Given
a path-based rule for /foo/bar/passwd, the following table applies:
passwd modified -- audit event logged
passwd replaced -- audit event logged, rules list updated
bar renamed -- rule removed
foo renamed -- untracked, meaning that the rule now applies to
the new location
Audit users typically want to have many rules referencing filesystem
objects, which can significantly impact filtering performance. This
patch also adds an inode-number-based rule hash to mitigate this
situation.
The patch is relative to the audit git tree:
http://kernel.org/git/?p=linux/kernel/git/viro/audit-current.git;a=summary
and uses the inotify kernel API:
http://lkml.org/lkml/2006/6/1/145
Signed-off-by: Amy Griffis <amy.griffis@hp.com>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
2006-04-07 14:55:56 -06:00
|
|
|
|
|
|
|
/* Inotify events we care about. */
|
|
|
|
#define AUDIT_IN_WATCH IN_MOVE|IN_CREATE|IN_DELETE|IN_DELETE_SELF|IN_MOVE_SELF
|
|
|
|
|
|
|
|
void audit_free_parent(struct inotify_watch *i_watch)
|
|
|
|
{
|
|
|
|
struct audit_parent *parent;
|
|
|
|
|
|
|
|
parent = container_of(i_watch, struct audit_parent, wdata);
|
|
|
|
WARN_ON(!list_empty(&parent->watches));
|
|
|
|
kfree(parent);
|
|
|
|
}
|
|
|
|
|
|
|
|
static inline void audit_get_watch(struct audit_watch *watch)
|
|
|
|
{
|
|
|
|
atomic_inc(&watch->count);
|
|
|
|
}
|
|
|
|
|
|
|
|
static void audit_put_watch(struct audit_watch *watch)
|
|
|
|
{
|
|
|
|
if (atomic_dec_and_test(&watch->count)) {
|
|
|
|
WARN_ON(watch->parent);
|
|
|
|
WARN_ON(!list_empty(&watch->rules));
|
|
|
|
kfree(watch->path);
|
|
|
|
kfree(watch);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
static void audit_remove_watch(struct audit_watch *watch)
|
|
|
|
{
|
|
|
|
list_del(&watch->wlist);
|
|
|
|
put_inotify_watch(&watch->parent->wdata);
|
|
|
|
watch->parent = NULL;
|
|
|
|
audit_put_watch(watch); /* match initial get */
|
|
|
|
}
|
|
|
|
|
2006-02-07 10:05:27 -07:00
|
|
|
static inline void audit_free_rule(struct audit_entry *e)
|
2005-12-15 11:33:52 -07:00
|
|
|
{
|
2006-03-10 17:14:06 -07:00
|
|
|
int i;
|
[PATCH] audit: path-based rules
In this implementation, audit registers inotify watches on the parent
directories of paths specified in audit rules. When audit's inotify
event handler is called, it updates any affected rules based on the
filesystem event. If the parent directory is renamed, removed, or its
filesystem is unmounted, audit removes all rules referencing that
inotify watch.
To keep things simple, this implementation limits location-based
auditing to the directory entries in an existing directory. Given
a path-based rule for /foo/bar/passwd, the following table applies:
passwd modified -- audit event logged
passwd replaced -- audit event logged, rules list updated
bar renamed -- rule removed
foo renamed -- untracked, meaning that the rule now applies to
the new location
Audit users typically want to have many rules referencing filesystem
objects, which can significantly impact filtering performance. This
patch also adds an inode-number-based rule hash to mitigate this
situation.
The patch is relative to the audit git tree:
http://kernel.org/git/?p=linux/kernel/git/viro/audit-current.git;a=summary
and uses the inotify kernel API:
http://lkml.org/lkml/2006/6/1/145
Signed-off-by: Amy Griffis <amy.griffis@hp.com>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
2006-04-07 14:55:56 -06:00
|
|
|
|
|
|
|
/* some rules don't have associated watches */
|
|
|
|
if (e->rule.watch)
|
|
|
|
audit_put_watch(e->rule.watch);
|
2006-03-10 17:14:06 -07:00
|
|
|
if (e->rule.fields)
|
|
|
|
for (i = 0; i < e->rule.field_count; i++) {
|
|
|
|
struct audit_field *f = &e->rule.fields[i];
|
2008-04-18 17:59:43 -06:00
|
|
|
kfree(f->lsm_str);
|
|
|
|
security_audit_rule_free(f->lsm_rule);
|
2006-03-10 17:14:06 -07:00
|
|
|
}
|
2006-02-07 10:05:27 -07:00
|
|
|
kfree(e->rule.fields);
|
2006-06-14 16:45:21 -06:00
|
|
|
kfree(e->rule.filterkey);
|
2006-02-07 10:05:27 -07:00
|
|
|
kfree(e);
|
|
|
|
}
|
|
|
|
|
[PATCH] audit: watching subtrees
New kind of audit rule predicates: "object is visible in given subtree".
The part that can be sanely implemented, that is. Limitations:
* if you have hardlink from outside of tree, you'd better watch
it too (or just watch the object itself, obviously)
* if you mount something under a watched tree, tell audit
that new chunk should be added to watched subtrees
* if you umount something in a watched tree and it's still mounted
elsewhere, you will get matches on events happening there. New command
tells audit to recalculate the trees, trimming such sources of false
positives.
Note that it's _not_ about path - if something mounted in several places
(multiple mount, bindings, different namespaces, etc.), the match does
_not_ depend on which one we are using for access.
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
2007-07-22 06:04:18 -06:00
|
|
|
void audit_free_rule_rcu(struct rcu_head *head)
|
2006-02-07 10:05:27 -07:00
|
|
|
{
|
|
|
|
struct audit_entry *e = container_of(head, struct audit_entry, rcu);
|
|
|
|
audit_free_rule(e);
|
|
|
|
}
|
|
|
|
|
[PATCH] audit: path-based rules
In this implementation, audit registers inotify watches on the parent
directories of paths specified in audit rules. When audit's inotify
event handler is called, it updates any affected rules based on the
filesystem event. If the parent directory is renamed, removed, or its
filesystem is unmounted, audit removes all rules referencing that
inotify watch.
To keep things simple, this implementation limits location-based
auditing to the directory entries in an existing directory. Given
a path-based rule for /foo/bar/passwd, the following table applies:
passwd modified -- audit event logged
passwd replaced -- audit event logged, rules list updated
bar renamed -- rule removed
foo renamed -- untracked, meaning that the rule now applies to
the new location
Audit users typically want to have many rules referencing filesystem
objects, which can significantly impact filtering performance. This
patch also adds an inode-number-based rule hash to mitigate this
situation.
The patch is relative to the audit git tree:
http://kernel.org/git/?p=linux/kernel/git/viro/audit-current.git;a=summary
and uses the inotify kernel API:
http://lkml.org/lkml/2006/6/1/145
Signed-off-by: Amy Griffis <amy.griffis@hp.com>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
2006-04-07 14:55:56 -06:00
|
|
|
/* Initialize a parent watch entry. */
|
|
|
|
static struct audit_parent *audit_init_parent(struct nameidata *ndp)
|
|
|
|
{
|
|
|
|
struct audit_parent *parent;
|
|
|
|
s32 wd;
|
|
|
|
|
|
|
|
parent = kzalloc(sizeof(*parent), GFP_KERNEL);
|
|
|
|
if (unlikely(!parent))
|
|
|
|
return ERR_PTR(-ENOMEM);
|
|
|
|
|
|
|
|
INIT_LIST_HEAD(&parent->watches);
|
|
|
|
parent->flags = 0;
|
|
|
|
|
|
|
|
inotify_init_watch(&parent->wdata);
|
|
|
|
/* grab a ref so inotify watch hangs around until we take audit_filter_mutex */
|
|
|
|
get_inotify_watch(&parent->wdata);
|
2008-02-14 20:34:32 -07:00
|
|
|
wd = inotify_add_watch(audit_ih, &parent->wdata,
|
|
|
|
ndp->path.dentry->d_inode, AUDIT_IN_WATCH);
|
[PATCH] audit: path-based rules
In this implementation, audit registers inotify watches on the parent
directories of paths specified in audit rules. When audit's inotify
event handler is called, it updates any affected rules based on the
filesystem event. If the parent directory is renamed, removed, or its
filesystem is unmounted, audit removes all rules referencing that
inotify watch.
To keep things simple, this implementation limits location-based
auditing to the directory entries in an existing directory. Given
a path-based rule for /foo/bar/passwd, the following table applies:
passwd modified -- audit event logged
passwd replaced -- audit event logged, rules list updated
bar renamed -- rule removed
foo renamed -- untracked, meaning that the rule now applies to
the new location
Audit users typically want to have many rules referencing filesystem
objects, which can significantly impact filtering performance. This
patch also adds an inode-number-based rule hash to mitigate this
situation.
The patch is relative to the audit git tree:
http://kernel.org/git/?p=linux/kernel/git/viro/audit-current.git;a=summary
and uses the inotify kernel API:
http://lkml.org/lkml/2006/6/1/145
Signed-off-by: Amy Griffis <amy.griffis@hp.com>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
2006-04-07 14:55:56 -06:00
|
|
|
if (wd < 0) {
|
|
|
|
audit_free_parent(&parent->wdata);
|
|
|
|
return ERR_PTR(wd);
|
|
|
|
}
|
|
|
|
|
|
|
|
return parent;
|
|
|
|
}
|
|
|
|
|
|
|
|
/* Initialize a watch entry. */
|
|
|
|
static struct audit_watch *audit_init_watch(char *path)
|
|
|
|
{
|
|
|
|
struct audit_watch *watch;
|
|
|
|
|
|
|
|
watch = kzalloc(sizeof(*watch), GFP_KERNEL);
|
|
|
|
if (unlikely(!watch))
|
|
|
|
return ERR_PTR(-ENOMEM);
|
|
|
|
|
|
|
|
INIT_LIST_HEAD(&watch->rules);
|
|
|
|
atomic_set(&watch->count, 1);
|
|
|
|
watch->path = path;
|
|
|
|
watch->dev = (dev_t)-1;
|
|
|
|
watch->ino = (unsigned long)-1;
|
|
|
|
|
|
|
|
return watch;
|
|
|
|
}
|
|
|
|
|
2006-03-10 17:14:06 -07:00
|
|
|
/* Initialize an audit filterlist entry. */
|
|
|
|
static inline struct audit_entry *audit_init_entry(u32 field_count)
|
|
|
|
{
|
|
|
|
struct audit_entry *entry;
|
|
|
|
struct audit_field *fields;
|
|
|
|
|
|
|
|
entry = kzalloc(sizeof(*entry), GFP_KERNEL);
|
|
|
|
if (unlikely(!entry))
|
|
|
|
return NULL;
|
|
|
|
|
|
|
|
fields = kzalloc(sizeof(*fields) * field_count, GFP_KERNEL);
|
|
|
|
if (unlikely(!fields)) {
|
|
|
|
kfree(entry);
|
|
|
|
return NULL;
|
|
|
|
}
|
|
|
|
entry->rule.fields = fields;
|
|
|
|
|
|
|
|
return entry;
|
|
|
|
}
|
|
|
|
|
2006-02-07 10:05:27 -07:00
|
|
|
/* Unpack a filter field's string representation from user-space
|
|
|
|
* buffer. */
|
[PATCH] audit: watching subtrees
New kind of audit rule predicates: "object is visible in given subtree".
The part that can be sanely implemented, that is. Limitations:
* if you have hardlink from outside of tree, you'd better watch
it too (or just watch the object itself, obviously)
* if you mount something under a watched tree, tell audit
that new chunk should be added to watched subtrees
* if you umount something in a watched tree and it's still mounted
elsewhere, you will get matches on events happening there. New command
tells audit to recalculate the trees, trimming such sources of false
positives.
Note that it's _not_ about path - if something mounted in several places
(multiple mount, bindings, different namespaces, etc.), the match does
_not_ depend on which one we are using for access.
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
2007-07-22 06:04:18 -06:00
|
|
|
char *audit_unpack_string(void **bufp, size_t *remain, size_t len)
|
2006-02-07 10:05:27 -07:00
|
|
|
{
|
|
|
|
char *str;
|
|
|
|
|
|
|
|
if (!*bufp || (len == 0) || (len > *remain))
|
|
|
|
return ERR_PTR(-EINVAL);
|
|
|
|
|
|
|
|
/* Of the currently implemented string fields, PATH_MAX
|
|
|
|
* defines the longest valid length.
|
|
|
|
*/
|
|
|
|
if (len > PATH_MAX)
|
|
|
|
return ERR_PTR(-ENAMETOOLONG);
|
|
|
|
|
|
|
|
str = kmalloc(len + 1, GFP_KERNEL);
|
|
|
|
if (unlikely(!str))
|
|
|
|
return ERR_PTR(-ENOMEM);
|
|
|
|
|
|
|
|
memcpy(str, *bufp, len);
|
|
|
|
str[len] = 0;
|
|
|
|
*bufp += len;
|
|
|
|
*remain -= len;
|
|
|
|
|
|
|
|
return str;
|
|
|
|
}
|
|
|
|
|
[PATCH] audit: path-based rules
In this implementation, audit registers inotify watches on the parent
directories of paths specified in audit rules. When audit's inotify
event handler is called, it updates any affected rules based on the
filesystem event. If the parent directory is renamed, removed, or its
filesystem is unmounted, audit removes all rules referencing that
inotify watch.
To keep things simple, this implementation limits location-based
auditing to the directory entries in an existing directory. Given
a path-based rule for /foo/bar/passwd, the following table applies:
passwd modified -- audit event logged
passwd replaced -- audit event logged, rules list updated
bar renamed -- rule removed
foo renamed -- untracked, meaning that the rule now applies to
the new location
Audit users typically want to have many rules referencing filesystem
objects, which can significantly impact filtering performance. This
patch also adds an inode-number-based rule hash to mitigate this
situation.
The patch is relative to the audit git tree:
http://kernel.org/git/?p=linux/kernel/git/viro/audit-current.git;a=summary
and uses the inotify kernel API:
http://lkml.org/lkml/2006/6/1/145
Signed-off-by: Amy Griffis <amy.griffis@hp.com>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
2006-04-07 14:55:56 -06:00
|
|
|
/* Translate an inode field to kernel respresentation. */
|
|
|
|
static inline int audit_to_inode(struct audit_krule *krule,
|
|
|
|
struct audit_field *f)
|
|
|
|
{
|
|
|
|
if (krule->listnr != AUDIT_FILTER_EXIT ||
|
[PATCH] audit: watching subtrees
New kind of audit rule predicates: "object is visible in given subtree".
The part that can be sanely implemented, that is. Limitations:
* if you have hardlink from outside of tree, you'd better watch
it too (or just watch the object itself, obviously)
* if you mount something under a watched tree, tell audit
that new chunk should be added to watched subtrees
* if you umount something in a watched tree and it's still mounted
elsewhere, you will get matches on events happening there. New command
tells audit to recalculate the trees, trimming such sources of false
positives.
Note that it's _not_ about path - if something mounted in several places
(multiple mount, bindings, different namespaces, etc.), the match does
_not_ depend on which one we are using for access.
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
2007-07-22 06:04:18 -06:00
|
|
|
krule->watch || krule->inode_f || krule->tree)
|
[PATCH] audit: path-based rules
In this implementation, audit registers inotify watches on the parent
directories of paths specified in audit rules. When audit's inotify
event handler is called, it updates any affected rules based on the
filesystem event. If the parent directory is renamed, removed, or its
filesystem is unmounted, audit removes all rules referencing that
inotify watch.
To keep things simple, this implementation limits location-based
auditing to the directory entries in an existing directory. Given
a path-based rule for /foo/bar/passwd, the following table applies:
passwd modified -- audit event logged
passwd replaced -- audit event logged, rules list updated
bar renamed -- rule removed
foo renamed -- untracked, meaning that the rule now applies to
the new location
Audit users typically want to have many rules referencing filesystem
objects, which can significantly impact filtering performance. This
patch also adds an inode-number-based rule hash to mitigate this
situation.
The patch is relative to the audit git tree:
http://kernel.org/git/?p=linux/kernel/git/viro/audit-current.git;a=summary
and uses the inotify kernel API:
http://lkml.org/lkml/2006/6/1/145
Signed-off-by: Amy Griffis <amy.griffis@hp.com>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
2006-04-07 14:55:56 -06:00
|
|
|
return -EINVAL;
|
|
|
|
|
|
|
|
krule->inode_f = f;
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
|
|
|
/* Translate a watch string to kernel respresentation. */
|
|
|
|
static int audit_to_watch(struct audit_krule *krule, char *path, int len,
|
|
|
|
u32 op)
|
|
|
|
{
|
|
|
|
struct audit_watch *watch;
|
|
|
|
|
|
|
|
if (!audit_ih)
|
|
|
|
return -EOPNOTSUPP;
|
|
|
|
|
|
|
|
if (path[0] != '/' || path[len-1] == '/' ||
|
|
|
|
krule->listnr != AUDIT_FILTER_EXIT ||
|
|
|
|
op & ~AUDIT_EQUAL ||
|
[PATCH] audit: watching subtrees
New kind of audit rule predicates: "object is visible in given subtree".
The part that can be sanely implemented, that is. Limitations:
* if you have hardlink from outside of tree, you'd better watch
it too (or just watch the object itself, obviously)
* if you mount something under a watched tree, tell audit
that new chunk should be added to watched subtrees
* if you umount something in a watched tree and it's still mounted
elsewhere, you will get matches on events happening there. New command
tells audit to recalculate the trees, trimming such sources of false
positives.
Note that it's _not_ about path - if something mounted in several places
(multiple mount, bindings, different namespaces, etc.), the match does
_not_ depend on which one we are using for access.
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
2007-07-22 06:04:18 -06:00
|
|
|
krule->inode_f || krule->watch || krule->tree)
|
[PATCH] audit: path-based rules
In this implementation, audit registers inotify watches on the parent
directories of paths specified in audit rules. When audit's inotify
event handler is called, it updates any affected rules based on the
filesystem event. If the parent directory is renamed, removed, or its
filesystem is unmounted, audit removes all rules referencing that
inotify watch.
To keep things simple, this implementation limits location-based
auditing to the directory entries in an existing directory. Given
a path-based rule for /foo/bar/passwd, the following table applies:
passwd modified -- audit event logged
passwd replaced -- audit event logged, rules list updated
bar renamed -- rule removed
foo renamed -- untracked, meaning that the rule now applies to
the new location
Audit users typically want to have many rules referencing filesystem
objects, which can significantly impact filtering performance. This
patch also adds an inode-number-based rule hash to mitigate this
situation.
The patch is relative to the audit git tree:
http://kernel.org/git/?p=linux/kernel/git/viro/audit-current.git;a=summary
and uses the inotify kernel API:
http://lkml.org/lkml/2006/6/1/145
Signed-off-by: Amy Griffis <amy.griffis@hp.com>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
2006-04-07 14:55:56 -06:00
|
|
|
return -EINVAL;
|
|
|
|
|
|
|
|
watch = audit_init_watch(path);
|
2008-04-29 02:03:09 -06:00
|
|
|
if (IS_ERR(watch))
|
[PATCH] audit: path-based rules
In this implementation, audit registers inotify watches on the parent
directories of paths specified in audit rules. When audit's inotify
event handler is called, it updates any affected rules based on the
filesystem event. If the parent directory is renamed, removed, or its
filesystem is unmounted, audit removes all rules referencing that
inotify watch.
To keep things simple, this implementation limits location-based
auditing to the directory entries in an existing directory. Given
a path-based rule for /foo/bar/passwd, the following table applies:
passwd modified -- audit event logged
passwd replaced -- audit event logged, rules list updated
bar renamed -- rule removed
foo renamed -- untracked, meaning that the rule now applies to
the new location
Audit users typically want to have many rules referencing filesystem
objects, which can significantly impact filtering performance. This
patch also adds an inode-number-based rule hash to mitigate this
situation.
The patch is relative to the audit git tree:
http://kernel.org/git/?p=linux/kernel/git/viro/audit-current.git;a=summary
and uses the inotify kernel API:
http://lkml.org/lkml/2006/6/1/145
Signed-off-by: Amy Griffis <amy.griffis@hp.com>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
2006-04-07 14:55:56 -06:00
|
|
|
return PTR_ERR(watch);
|
|
|
|
|
|
|
|
audit_get_watch(watch);
|
|
|
|
krule->watch = watch;
|
|
|
|
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
2006-07-01 01:56:16 -06:00
|
|
|
static __u32 *classes[AUDIT_SYSCALL_CLASSES];
|
|
|
|
|
|
|
|
int __init audit_register_class(int class, unsigned *list)
|
|
|
|
{
|
|
|
|
__u32 *p = kzalloc(AUDIT_BITMASK_SIZE * sizeof(__u32), GFP_KERNEL);
|
|
|
|
if (!p)
|
|
|
|
return -ENOMEM;
|
|
|
|
while (*list != ~0U) {
|
|
|
|
unsigned n = *list++;
|
|
|
|
if (n >= AUDIT_BITMASK_SIZE * 32 - AUDIT_SYSCALL_CLASSES) {
|
|
|
|
kfree(p);
|
|
|
|
return -EINVAL;
|
|
|
|
}
|
|
|
|
p[AUDIT_WORD(n)] |= AUDIT_BIT(n);
|
|
|
|
}
|
|
|
|
if (class >= AUDIT_SYSCALL_CLASSES || classes[class]) {
|
|
|
|
kfree(p);
|
|
|
|
return -EINVAL;
|
|
|
|
}
|
|
|
|
classes[class] = p;
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
2006-08-31 17:26:40 -06:00
|
|
|
int audit_match_class(int class, unsigned syscall)
|
|
|
|
{
|
2007-05-16 16:45:42 -06:00
|
|
|
if (unlikely(syscall >= AUDIT_BITMASK_SIZE * 32))
|
2006-08-31 17:26:40 -06:00
|
|
|
return 0;
|
|
|
|
if (unlikely(class >= AUDIT_SYSCALL_CLASSES || !classes[class]))
|
|
|
|
return 0;
|
|
|
|
return classes[class][AUDIT_WORD(syscall)] & AUDIT_BIT(syscall);
|
|
|
|
}
|
|
|
|
|
2007-05-15 13:37:10 -06:00
|
|
|
#ifdef CONFIG_AUDITSYSCALL
|
2007-03-29 16:01:04 -06:00
|
|
|
static inline int audit_match_class_bits(int class, u32 *mask)
|
|
|
|
{
|
|
|
|
int i;
|
|
|
|
|
|
|
|
if (classes[class]) {
|
|
|
|
for (i = 0; i < AUDIT_BITMASK_SIZE; i++)
|
|
|
|
if (mask[i] & classes[class][i])
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
return 1;
|
|
|
|
}
|
|
|
|
|
|
|
|
static int audit_match_signal(struct audit_entry *entry)
|
|
|
|
{
|
|
|
|
struct audit_field *arch = entry->rule.arch_f;
|
|
|
|
|
|
|
|
if (!arch) {
|
|
|
|
/* When arch is unspecified, we must check both masks on biarch
|
|
|
|
* as syscall number alone is ambiguous. */
|
|
|
|
return (audit_match_class_bits(AUDIT_CLASS_SIGNAL,
|
|
|
|
entry->rule.mask) &&
|
|
|
|
audit_match_class_bits(AUDIT_CLASS_SIGNAL_32,
|
|
|
|
entry->rule.mask));
|
|
|
|
}
|
|
|
|
|
|
|
|
switch(audit_classify_arch(arch->val)) {
|
|
|
|
case 0: /* native */
|
|
|
|
return (audit_match_class_bits(AUDIT_CLASS_SIGNAL,
|
|
|
|
entry->rule.mask));
|
|
|
|
case 1: /* 32bit on biarch */
|
|
|
|
return (audit_match_class_bits(AUDIT_CLASS_SIGNAL_32,
|
|
|
|
entry->rule.mask));
|
|
|
|
default:
|
|
|
|
return 1;
|
|
|
|
}
|
|
|
|
}
|
2007-05-15 13:37:10 -06:00
|
|
|
#endif
|
2007-03-29 16:01:04 -06:00
|
|
|
|
2006-02-07 10:05:27 -07:00
|
|
|
/* Common user-space to kernel rule translation. */
|
|
|
|
static inline struct audit_entry *audit_to_entry_common(struct audit_rule *rule)
|
|
|
|
{
|
|
|
|
unsigned listnr;
|
|
|
|
struct audit_entry *entry;
|
|
|
|
int i, err;
|
|
|
|
|
|
|
|
err = -EINVAL;
|
|
|
|
listnr = rule->flags & ~AUDIT_FILTER_PREPEND;
|
|
|
|
switch(listnr) {
|
|
|
|
default:
|
|
|
|
goto exit_err;
|
|
|
|
case AUDIT_FILTER_USER:
|
|
|
|
case AUDIT_FILTER_TYPE:
|
|
|
|
#ifdef CONFIG_AUDITSYSCALL
|
|
|
|
case AUDIT_FILTER_ENTRY:
|
|
|
|
case AUDIT_FILTER_EXIT:
|
|
|
|
case AUDIT_FILTER_TASK:
|
|
|
|
#endif
|
|
|
|
;
|
|
|
|
}
|
2006-05-22 23:36:13 -06:00
|
|
|
if (unlikely(rule->action == AUDIT_POSSIBLE)) {
|
|
|
|
printk(KERN_ERR "AUDIT_POSSIBLE is deprecated\n");
|
|
|
|
goto exit_err;
|
|
|
|
}
|
|
|
|
if (rule->action != AUDIT_NEVER && rule->action != AUDIT_ALWAYS)
|
2006-02-07 10:05:27 -07:00
|
|
|
goto exit_err;
|
|
|
|
if (rule->field_count > AUDIT_MAX_FIELDS)
|
|
|
|
goto exit_err;
|
|
|
|
|
|
|
|
err = -ENOMEM;
|
2006-03-10 17:14:06 -07:00
|
|
|
entry = audit_init_entry(rule->field_count);
|
|
|
|
if (!entry)
|
2006-02-07 10:05:27 -07:00
|
|
|
goto exit_err;
|
|
|
|
|
|
|
|
entry->rule.flags = rule->flags & AUDIT_FILTER_PREPEND;
|
|
|
|
entry->rule.listnr = listnr;
|
|
|
|
entry->rule.action = rule->action;
|
|
|
|
entry->rule.field_count = rule->field_count;
|
|
|
|
|
|
|
|
for (i = 0; i < AUDIT_BITMASK_SIZE; i++)
|
|
|
|
entry->rule.mask[i] = rule->mask[i];
|
|
|
|
|
2006-07-01 01:56:16 -06:00
|
|
|
for (i = 0; i < AUDIT_SYSCALL_CLASSES; i++) {
|
|
|
|
int bit = AUDIT_BITMASK_SIZE * 32 - i - 1;
|
|
|
|
__u32 *p = &entry->rule.mask[AUDIT_WORD(bit)];
|
|
|
|
__u32 *class;
|
|
|
|
|
|
|
|
if (!(*p & AUDIT_BIT(bit)))
|
|
|
|
continue;
|
|
|
|
*p &= ~AUDIT_BIT(bit);
|
|
|
|
class = classes[i];
|
|
|
|
if (class) {
|
|
|
|
int j;
|
|
|
|
for (j = 0; j < AUDIT_BITMASK_SIZE; j++)
|
|
|
|
entry->rule.mask[j] |= class[j];
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2006-02-07 10:05:27 -07:00
|
|
|
return entry;
|
|
|
|
|
|
|
|
exit_err:
|
|
|
|
return ERR_PTR(err);
|
|
|
|
}
|
|
|
|
|
|
|
|
/* Translate struct audit_rule to kernel's rule respresentation.
|
|
|
|
* Exists for backward compatibility with userspace. */
|
|
|
|
static struct audit_entry *audit_rule_to_entry(struct audit_rule *rule)
|
|
|
|
{
|
|
|
|
struct audit_entry *entry;
|
2008-04-27 03:39:56 -06:00
|
|
|
struct audit_field *ino_f;
|
2006-02-07 10:05:27 -07:00
|
|
|
int err = 0;
|
2005-12-15 11:33:52 -07:00
|
|
|
int i;
|
|
|
|
|
2006-02-07 10:05:27 -07:00
|
|
|
entry = audit_to_entry_common(rule);
|
|
|
|
if (IS_ERR(entry))
|
|
|
|
goto exit_nofree;
|
|
|
|
|
|
|
|
for (i = 0; i < rule->field_count; i++) {
|
|
|
|
struct audit_field *f = &entry->rule.fields[i];
|
|
|
|
|
|
|
|
f->op = rule->fields[i] & (AUDIT_NEGATE|AUDIT_OPERATORS);
|
|
|
|
f->type = rule->fields[i] & ~(AUDIT_NEGATE|AUDIT_OPERATORS);
|
|
|
|
f->val = rule->values[i];
|
|
|
|
|
[PATCH] audit: path-based rules
In this implementation, audit registers inotify watches on the parent
directories of paths specified in audit rules. When audit's inotify
event handler is called, it updates any affected rules based on the
filesystem event. If the parent directory is renamed, removed, or its
filesystem is unmounted, audit removes all rules referencing that
inotify watch.
To keep things simple, this implementation limits location-based
auditing to the directory entries in an existing directory. Given
a path-based rule for /foo/bar/passwd, the following table applies:
passwd modified -- audit event logged
passwd replaced -- audit event logged, rules list updated
bar renamed -- rule removed
foo renamed -- untracked, meaning that the rule now applies to
the new location
Audit users typically want to have many rules referencing filesystem
objects, which can significantly impact filtering performance. This
patch also adds an inode-number-based rule hash to mitigate this
situation.
The patch is relative to the audit git tree:
http://kernel.org/git/?p=linux/kernel/git/viro/audit-current.git;a=summary
and uses the inotify kernel API:
http://lkml.org/lkml/2006/6/1/145
Signed-off-by: Amy Griffis <amy.griffis@hp.com>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
2006-04-07 14:55:56 -06:00
|
|
|
err = -EINVAL;
|
|
|
|
switch(f->type) {
|
2006-06-05 06:15:59 -06:00
|
|
|
default:
|
2006-03-10 17:14:06 -07:00
|
|
|
goto exit_free;
|
2006-06-05 06:15:59 -06:00
|
|
|
case AUDIT_PID:
|
|
|
|
case AUDIT_UID:
|
|
|
|
case AUDIT_EUID:
|
|
|
|
case AUDIT_SUID:
|
|
|
|
case AUDIT_FSUID:
|
|
|
|
case AUDIT_GID:
|
|
|
|
case AUDIT_EGID:
|
|
|
|
case AUDIT_SGID:
|
|
|
|
case AUDIT_FSGID:
|
|
|
|
case AUDIT_LOGINUID:
|
|
|
|
case AUDIT_PERS:
|
|
|
|
case AUDIT_MSGTYPE:
|
2006-08-26 12:06:20 -06:00
|
|
|
case AUDIT_PPID:
|
2006-06-05 06:15:59 -06:00
|
|
|
case AUDIT_DEVMAJOR:
|
|
|
|
case AUDIT_DEVMINOR:
|
|
|
|
case AUDIT_EXIT:
|
|
|
|
case AUDIT_SUCCESS:
|
2007-06-04 15:00:14 -06:00
|
|
|
/* bit ops are only useful on syscall args */
|
|
|
|
if (f->op == AUDIT_BIT_MASK ||
|
|
|
|
f->op == AUDIT_BIT_TEST) {
|
|
|
|
err = -EINVAL;
|
|
|
|
goto exit_free;
|
|
|
|
}
|
|
|
|
break;
|
2006-06-05 06:15:59 -06:00
|
|
|
case AUDIT_ARG0:
|
|
|
|
case AUDIT_ARG1:
|
|
|
|
case AUDIT_ARG2:
|
|
|
|
case AUDIT_ARG3:
|
|
|
|
break;
|
2006-09-28 15:46:21 -06:00
|
|
|
/* arch is only allowed to be = or != */
|
|
|
|
case AUDIT_ARCH:
|
|
|
|
if ((f->op != AUDIT_NOT_EQUAL) && (f->op != AUDIT_EQUAL)
|
|
|
|
&& (f->op != AUDIT_NEGATE) && (f->op)) {
|
|
|
|
err = -EINVAL;
|
|
|
|
goto exit_free;
|
|
|
|
}
|
2007-03-29 16:01:04 -06:00
|
|
|
entry->rule.arch_f = f;
|
2006-09-28 15:46:21 -06:00
|
|
|
break;
|
2006-08-31 17:26:40 -06:00
|
|
|
case AUDIT_PERM:
|
|
|
|
if (f->val & ~15)
|
|
|
|
goto exit_free;
|
|
|
|
break;
|
2008-04-28 02:15:49 -06:00
|
|
|
case AUDIT_FILETYPE:
|
|
|
|
if ((f->val & ~S_IFMT) > S_IFMT)
|
|
|
|
goto exit_free;
|
|
|
|
break;
|
[PATCH] audit: path-based rules
In this implementation, audit registers inotify watches on the parent
directories of paths specified in audit rules. When audit's inotify
event handler is called, it updates any affected rules based on the
filesystem event. If the parent directory is renamed, removed, or its
filesystem is unmounted, audit removes all rules referencing that
inotify watch.
To keep things simple, this implementation limits location-based
auditing to the directory entries in an existing directory. Given
a path-based rule for /foo/bar/passwd, the following table applies:
passwd modified -- audit event logged
passwd replaced -- audit event logged, rules list updated
bar renamed -- rule removed
foo renamed -- untracked, meaning that the rule now applies to
the new location
Audit users typically want to have many rules referencing filesystem
objects, which can significantly impact filtering performance. This
patch also adds an inode-number-based rule hash to mitigate this
situation.
The patch is relative to the audit git tree:
http://kernel.org/git/?p=linux/kernel/git/viro/audit-current.git;a=summary
and uses the inotify kernel API:
http://lkml.org/lkml/2006/6/1/145
Signed-off-by: Amy Griffis <amy.griffis@hp.com>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
2006-04-07 14:55:56 -06:00
|
|
|
case AUDIT_INODE:
|
|
|
|
err = audit_to_inode(&entry->rule, f);
|
|
|
|
if (err)
|
|
|
|
goto exit_free;
|
|
|
|
break;
|
2006-03-10 17:14:06 -07:00
|
|
|
}
|
|
|
|
|
2006-02-07 10:05:27 -07:00
|
|
|
entry->rule.vers_ops = (f->op & AUDIT_OPERATORS) ? 2 : 1;
|
2006-02-16 12:40:01 -07:00
|
|
|
|
|
|
|
/* Support for legacy operators where
|
|
|
|
* AUDIT_NEGATE bit signifies != and otherwise assumes == */
|
2006-02-07 10:05:27 -07:00
|
|
|
if (f->op & AUDIT_NEGATE)
|
2006-02-16 12:40:01 -07:00
|
|
|
f->op = AUDIT_NOT_EQUAL;
|
|
|
|
else if (!f->op)
|
|
|
|
f->op = AUDIT_EQUAL;
|
|
|
|
else if (f->op == AUDIT_OPERATORS) {
|
|
|
|
err = -EINVAL;
|
|
|
|
goto exit_free;
|
|
|
|
}
|
2005-12-15 11:33:52 -07:00
|
|
|
}
|
2006-02-07 10:05:27 -07:00
|
|
|
|
2008-04-27 03:39:56 -06:00
|
|
|
ino_f = entry->rule.inode_f;
|
|
|
|
if (ino_f) {
|
|
|
|
switch(ino_f->op) {
|
[PATCH] audit: path-based rules
In this implementation, audit registers inotify watches on the parent
directories of paths specified in audit rules. When audit's inotify
event handler is called, it updates any affected rules based on the
filesystem event. If the parent directory is renamed, removed, or its
filesystem is unmounted, audit removes all rules referencing that
inotify watch.
To keep things simple, this implementation limits location-based
auditing to the directory entries in an existing directory. Given
a path-based rule for /foo/bar/passwd, the following table applies:
passwd modified -- audit event logged
passwd replaced -- audit event logged, rules list updated
bar renamed -- rule removed
foo renamed -- untracked, meaning that the rule now applies to
the new location
Audit users typically want to have many rules referencing filesystem
objects, which can significantly impact filtering performance. This
patch also adds an inode-number-based rule hash to mitigate this
situation.
The patch is relative to the audit git tree:
http://kernel.org/git/?p=linux/kernel/git/viro/audit-current.git;a=summary
and uses the inotify kernel API:
http://lkml.org/lkml/2006/6/1/145
Signed-off-by: Amy Griffis <amy.griffis@hp.com>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
2006-04-07 14:55:56 -06:00
|
|
|
case AUDIT_NOT_EQUAL:
|
|
|
|
entry->rule.inode_f = NULL;
|
|
|
|
case AUDIT_EQUAL:
|
|
|
|
break;
|
|
|
|
default:
|
2006-08-01 15:52:26 -06:00
|
|
|
err = -EINVAL;
|
[PATCH] audit: path-based rules
In this implementation, audit registers inotify watches on the parent
directories of paths specified in audit rules. When audit's inotify
event handler is called, it updates any affected rules based on the
filesystem event. If the parent directory is renamed, removed, or its
filesystem is unmounted, audit removes all rules referencing that
inotify watch.
To keep things simple, this implementation limits location-based
auditing to the directory entries in an existing directory. Given
a path-based rule for /foo/bar/passwd, the following table applies:
passwd modified -- audit event logged
passwd replaced -- audit event logged, rules list updated
bar renamed -- rule removed
foo renamed -- untracked, meaning that the rule now applies to
the new location
Audit users typically want to have many rules referencing filesystem
objects, which can significantly impact filtering performance. This
patch also adds an inode-number-based rule hash to mitigate this
situation.
The patch is relative to the audit git tree:
http://kernel.org/git/?p=linux/kernel/git/viro/audit-current.git;a=summary
and uses the inotify kernel API:
http://lkml.org/lkml/2006/6/1/145
Signed-off-by: Amy Griffis <amy.griffis@hp.com>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
2006-04-07 14:55:56 -06:00
|
|
|
goto exit_free;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2006-02-07 10:05:27 -07:00
|
|
|
exit_nofree:
|
|
|
|
return entry;
|
|
|
|
|
|
|
|
exit_free:
|
|
|
|
audit_free_rule(entry);
|
|
|
|
return ERR_PTR(err);
|
2005-12-15 11:33:52 -07:00
|
|
|
}
|
|
|
|
|
2006-02-07 10:05:27 -07:00
|
|
|
/* Translate struct audit_rule_data to kernel's rule respresentation. */
|
|
|
|
static struct audit_entry *audit_data_to_entry(struct audit_rule_data *data,
|
|
|
|
size_t datasz)
|
2005-12-15 11:33:52 -07:00
|
|
|
{
|
2006-02-07 10:05:27 -07:00
|
|
|
int err = 0;
|
|
|
|
struct audit_entry *entry;
|
2008-04-27 03:39:56 -06:00
|
|
|
struct audit_field *ino_f;
|
2006-02-07 10:05:27 -07:00
|
|
|
void *bufp;
|
2006-03-10 17:14:06 -07:00
|
|
|
size_t remain = datasz - sizeof(struct audit_rule_data);
|
2005-12-15 11:33:52 -07:00
|
|
|
int i;
|
2006-03-10 17:14:06 -07:00
|
|
|
char *str;
|
2005-12-15 11:33:52 -07:00
|
|
|
|
2006-02-07 10:05:27 -07:00
|
|
|
entry = audit_to_entry_common((struct audit_rule *)data);
|
|
|
|
if (IS_ERR(entry))
|
|
|
|
goto exit_nofree;
|
2005-12-15 11:33:52 -07:00
|
|
|
|
2006-02-07 10:05:27 -07:00
|
|
|
bufp = data->buf;
|
|
|
|
entry->rule.vers_ops = 2;
|
|
|
|
for (i = 0; i < data->field_count; i++) {
|
|
|
|
struct audit_field *f = &entry->rule.fields[i];
|
|
|
|
|
|
|
|
err = -EINVAL;
|
|
|
|
if (!(data->fieldflags[i] & AUDIT_OPERATORS) ||
|
|
|
|
data->fieldflags[i] & ~AUDIT_OPERATORS)
|
|
|
|
goto exit_free;
|
|
|
|
|
|
|
|
f->op = data->fieldflags[i] & AUDIT_OPERATORS;
|
|
|
|
f->type = data->fields[i];
|
2006-03-10 17:14:06 -07:00
|
|
|
f->val = data->values[i];
|
2008-04-18 17:59:43 -06:00
|
|
|
f->lsm_str = NULL;
|
|
|
|
f->lsm_rule = NULL;
|
2006-02-07 10:05:27 -07:00
|
|
|
switch(f->type) {
|
2006-06-05 06:15:59 -06:00
|
|
|
case AUDIT_PID:
|
|
|
|
case AUDIT_UID:
|
|
|
|
case AUDIT_EUID:
|
|
|
|
case AUDIT_SUID:
|
|
|
|
case AUDIT_FSUID:
|
|
|
|
case AUDIT_GID:
|
|
|
|
case AUDIT_EGID:
|
|
|
|
case AUDIT_SGID:
|
|
|
|
case AUDIT_FSGID:
|
|
|
|
case AUDIT_LOGINUID:
|
|
|
|
case AUDIT_PERS:
|
|
|
|
case AUDIT_MSGTYPE:
|
|
|
|
case AUDIT_PPID:
|
|
|
|
case AUDIT_DEVMAJOR:
|
|
|
|
case AUDIT_DEVMINOR:
|
|
|
|
case AUDIT_EXIT:
|
|
|
|
case AUDIT_SUCCESS:
|
|
|
|
case AUDIT_ARG0:
|
|
|
|
case AUDIT_ARG1:
|
|
|
|
case AUDIT_ARG2:
|
|
|
|
case AUDIT_ARG3:
|
|
|
|
break;
|
2007-03-29 16:01:04 -06:00
|
|
|
case AUDIT_ARCH:
|
|
|
|
entry->rule.arch_f = f;
|
|
|
|
break;
|
2006-06-29 15:56:39 -06:00
|
|
|
case AUDIT_SUBJ_USER:
|
|
|
|
case AUDIT_SUBJ_ROLE:
|
|
|
|
case AUDIT_SUBJ_TYPE:
|
|
|
|
case AUDIT_SUBJ_SEN:
|
|
|
|
case AUDIT_SUBJ_CLR:
|
2006-06-29 15:57:08 -06:00
|
|
|
case AUDIT_OBJ_USER:
|
|
|
|
case AUDIT_OBJ_ROLE:
|
|
|
|
case AUDIT_OBJ_TYPE:
|
|
|
|
case AUDIT_OBJ_LEV_LOW:
|
|
|
|
case AUDIT_OBJ_LEV_HIGH:
|
2006-03-10 17:14:06 -07:00
|
|
|
str = audit_unpack_string(&bufp, &remain, f->val);
|
|
|
|
if (IS_ERR(str))
|
|
|
|
goto exit_free;
|
|
|
|
entry->rule.buflen += f->val;
|
|
|
|
|
2008-03-01 13:01:11 -07:00
|
|
|
err = security_audit_rule_init(f->type, f->op, str,
|
2008-04-18 17:59:43 -06:00
|
|
|
(void **)&f->lsm_rule);
|
2006-03-10 17:14:06 -07:00
|
|
|
/* Keep currently invalid fields around in case they
|
|
|
|
* become valid after a policy reload. */
|
|
|
|
if (err == -EINVAL) {
|
2008-03-01 13:01:11 -07:00
|
|
|
printk(KERN_WARNING "audit rule for LSM "
|
2006-03-10 17:14:06 -07:00
|
|
|
"\'%s\' is invalid\n", str);
|
|
|
|
err = 0;
|
|
|
|
}
|
|
|
|
if (err) {
|
|
|
|
kfree(str);
|
|
|
|
goto exit_free;
|
|
|
|
} else
|
2008-04-18 17:59:43 -06:00
|
|
|
f->lsm_str = str;
|
2006-03-10 17:14:06 -07:00
|
|
|
break;
|
[PATCH] audit: path-based rules
In this implementation, audit registers inotify watches on the parent
directories of paths specified in audit rules. When audit's inotify
event handler is called, it updates any affected rules based on the
filesystem event. If the parent directory is renamed, removed, or its
filesystem is unmounted, audit removes all rules referencing that
inotify watch.
To keep things simple, this implementation limits location-based
auditing to the directory entries in an existing directory. Given
a path-based rule for /foo/bar/passwd, the following table applies:
passwd modified -- audit event logged
passwd replaced -- audit event logged, rules list updated
bar renamed -- rule removed
foo renamed -- untracked, meaning that the rule now applies to
the new location
Audit users typically want to have many rules referencing filesystem
objects, which can significantly impact filtering performance. This
patch also adds an inode-number-based rule hash to mitigate this
situation.
The patch is relative to the audit git tree:
http://kernel.org/git/?p=linux/kernel/git/viro/audit-current.git;a=summary
and uses the inotify kernel API:
http://lkml.org/lkml/2006/6/1/145
Signed-off-by: Amy Griffis <amy.griffis@hp.com>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
2006-04-07 14:55:56 -06:00
|
|
|
case AUDIT_WATCH:
|
|
|
|
str = audit_unpack_string(&bufp, &remain, f->val);
|
|
|
|
if (IS_ERR(str))
|
|
|
|
goto exit_free;
|
|
|
|
entry->rule.buflen += f->val;
|
|
|
|
|
|
|
|
err = audit_to_watch(&entry->rule, str, f->val, f->op);
|
|
|
|
if (err) {
|
|
|
|
kfree(str);
|
|
|
|
goto exit_free;
|
|
|
|
}
|
|
|
|
break;
|
[PATCH] audit: watching subtrees
New kind of audit rule predicates: "object is visible in given subtree".
The part that can be sanely implemented, that is. Limitations:
* if you have hardlink from outside of tree, you'd better watch
it too (or just watch the object itself, obviously)
* if you mount something under a watched tree, tell audit
that new chunk should be added to watched subtrees
* if you umount something in a watched tree and it's still mounted
elsewhere, you will get matches on events happening there. New command
tells audit to recalculate the trees, trimming such sources of false
positives.
Note that it's _not_ about path - if something mounted in several places
(multiple mount, bindings, different namespaces, etc.), the match does
_not_ depend on which one we are using for access.
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
2007-07-22 06:04:18 -06:00
|
|
|
case AUDIT_DIR:
|
|
|
|
str = audit_unpack_string(&bufp, &remain, f->val);
|
|
|
|
if (IS_ERR(str))
|
|
|
|
goto exit_free;
|
|
|
|
entry->rule.buflen += f->val;
|
|
|
|
|
|
|
|
err = audit_make_tree(&entry->rule, str, f->op);
|
|
|
|
kfree(str);
|
|
|
|
if (err)
|
|
|
|
goto exit_free;
|
|
|
|
break;
|
[PATCH] audit: path-based rules
In this implementation, audit registers inotify watches on the parent
directories of paths specified in audit rules. When audit's inotify
event handler is called, it updates any affected rules based on the
filesystem event. If the parent directory is renamed, removed, or its
filesystem is unmounted, audit removes all rules referencing that
inotify watch.
To keep things simple, this implementation limits location-based
auditing to the directory entries in an existing directory. Given
a path-based rule for /foo/bar/passwd, the following table applies:
passwd modified -- audit event logged
passwd replaced -- audit event logged, rules list updated
bar renamed -- rule removed
foo renamed -- untracked, meaning that the rule now applies to
the new location
Audit users typically want to have many rules referencing filesystem
objects, which can significantly impact filtering performance. This
patch also adds an inode-number-based rule hash to mitigate this
situation.
The patch is relative to the audit git tree:
http://kernel.org/git/?p=linux/kernel/git/viro/audit-current.git;a=summary
and uses the inotify kernel API:
http://lkml.org/lkml/2006/6/1/145
Signed-off-by: Amy Griffis <amy.griffis@hp.com>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
2006-04-07 14:55:56 -06:00
|
|
|
case AUDIT_INODE:
|
|
|
|
err = audit_to_inode(&entry->rule, f);
|
|
|
|
if (err)
|
|
|
|
goto exit_free;
|
|
|
|
break;
|
2006-06-14 16:45:21 -06:00
|
|
|
case AUDIT_FILTERKEY:
|
|
|
|
err = -EINVAL;
|
|
|
|
if (entry->rule.filterkey || f->val > AUDIT_MAX_KEY_LEN)
|
|
|
|
goto exit_free;
|
|
|
|
str = audit_unpack_string(&bufp, &remain, f->val);
|
|
|
|
if (IS_ERR(str))
|
|
|
|
goto exit_free;
|
|
|
|
entry->rule.buflen += f->val;
|
|
|
|
entry->rule.filterkey = str;
|
|
|
|
break;
|
2006-08-31 17:26:40 -06:00
|
|
|
case AUDIT_PERM:
|
|
|
|
if (f->val & ~15)
|
|
|
|
goto exit_free;
|
|
|
|
break;
|
2008-04-28 02:15:49 -06:00
|
|
|
case AUDIT_FILETYPE:
|
|
|
|
if ((f->val & ~S_IFMT) > S_IFMT)
|
|
|
|
goto exit_free;
|
|
|
|
break;
|
2006-06-05 06:15:59 -06:00
|
|
|
default:
|
|
|
|
goto exit_free;
|
[PATCH] audit: path-based rules
In this implementation, audit registers inotify watches on the parent
directories of paths specified in audit rules. When audit's inotify
event handler is called, it updates any affected rules based on the
filesystem event. If the parent directory is renamed, removed, or its
filesystem is unmounted, audit removes all rules referencing that
inotify watch.
To keep things simple, this implementation limits location-based
auditing to the directory entries in an existing directory. Given
a path-based rule for /foo/bar/passwd, the following table applies:
passwd modified -- audit event logged
passwd replaced -- audit event logged, rules list updated
bar renamed -- rule removed
foo renamed -- untracked, meaning that the rule now applies to
the new location
Audit users typically want to have many rules referencing filesystem
objects, which can significantly impact filtering performance. This
patch also adds an inode-number-based rule hash to mitigate this
situation.
The patch is relative to the audit git tree:
http://kernel.org/git/?p=linux/kernel/git/viro/audit-current.git;a=summary
and uses the inotify kernel API:
http://lkml.org/lkml/2006/6/1/145
Signed-off-by: Amy Griffis <amy.griffis@hp.com>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
2006-04-07 14:55:56 -06:00
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2008-04-27 03:39:56 -06:00
|
|
|
ino_f = entry->rule.inode_f;
|
|
|
|
if (ino_f) {
|
|
|
|
switch(ino_f->op) {
|
[PATCH] audit: path-based rules
In this implementation, audit registers inotify watches on the parent
directories of paths specified in audit rules. When audit's inotify
event handler is called, it updates any affected rules based on the
filesystem event. If the parent directory is renamed, removed, or its
filesystem is unmounted, audit removes all rules referencing that
inotify watch.
To keep things simple, this implementation limits location-based
auditing to the directory entries in an existing directory. Given
a path-based rule for /foo/bar/passwd, the following table applies:
passwd modified -- audit event logged
passwd replaced -- audit event logged, rules list updated
bar renamed -- rule removed
foo renamed -- untracked, meaning that the rule now applies to
the new location
Audit users typically want to have many rules referencing filesystem
objects, which can significantly impact filtering performance. This
patch also adds an inode-number-based rule hash to mitigate this
situation.
The patch is relative to the audit git tree:
http://kernel.org/git/?p=linux/kernel/git/viro/audit-current.git;a=summary
and uses the inotify kernel API:
http://lkml.org/lkml/2006/6/1/145
Signed-off-by: Amy Griffis <amy.griffis@hp.com>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
2006-04-07 14:55:56 -06:00
|
|
|
case AUDIT_NOT_EQUAL:
|
|
|
|
entry->rule.inode_f = NULL;
|
|
|
|
case AUDIT_EQUAL:
|
|
|
|
break;
|
|
|
|
default:
|
2006-08-01 15:52:26 -06:00
|
|
|
err = -EINVAL;
|
[PATCH] audit: path-based rules
In this implementation, audit registers inotify watches on the parent
directories of paths specified in audit rules. When audit's inotify
event handler is called, it updates any affected rules based on the
filesystem event. If the parent directory is renamed, removed, or its
filesystem is unmounted, audit removes all rules referencing that
inotify watch.
To keep things simple, this implementation limits location-based
auditing to the directory entries in an existing directory. Given
a path-based rule for /foo/bar/passwd, the following table applies:
passwd modified -- audit event logged
passwd replaced -- audit event logged, rules list updated
bar renamed -- rule removed
foo renamed -- untracked, meaning that the rule now applies to
the new location
Audit users typically want to have many rules referencing filesystem
objects, which can significantly impact filtering performance. This
patch also adds an inode-number-based rule hash to mitigate this
situation.
The patch is relative to the audit git tree:
http://kernel.org/git/?p=linux/kernel/git/viro/audit-current.git;a=summary
and uses the inotify kernel API:
http://lkml.org/lkml/2006/6/1/145
Signed-off-by: Amy Griffis <amy.griffis@hp.com>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
2006-04-07 14:55:56 -06:00
|
|
|
goto exit_free;
|
2006-02-07 10:05:27 -07:00
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
exit_nofree:
|
|
|
|
return entry;
|
|
|
|
|
|
|
|
exit_free:
|
|
|
|
audit_free_rule(entry);
|
|
|
|
return ERR_PTR(err);
|
|
|
|
}
|
|
|
|
|
|
|
|
/* Pack a filter field's string representation into data block. */
|
[PATCH] audit: watching subtrees
New kind of audit rule predicates: "object is visible in given subtree".
The part that can be sanely implemented, that is. Limitations:
* if you have hardlink from outside of tree, you'd better watch
it too (or just watch the object itself, obviously)
* if you mount something under a watched tree, tell audit
that new chunk should be added to watched subtrees
* if you umount something in a watched tree and it's still mounted
elsewhere, you will get matches on events happening there. New command
tells audit to recalculate the trees, trimming such sources of false
positives.
Note that it's _not_ about path - if something mounted in several places
(multiple mount, bindings, different namespaces, etc.), the match does
_not_ depend on which one we are using for access.
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
2007-07-22 06:04:18 -06:00
|
|
|
static inline size_t audit_pack_string(void **bufp, const char *str)
|
2006-02-07 10:05:27 -07:00
|
|
|
{
|
|
|
|
size_t len = strlen(str);
|
|
|
|
|
|
|
|
memcpy(*bufp, str, len);
|
|
|
|
*bufp += len;
|
|
|
|
|
|
|
|
return len;
|
|
|
|
}
|
|
|
|
|
|
|
|
/* Translate kernel rule respresentation to struct audit_rule.
|
|
|
|
* Exists for backward compatibility with userspace. */
|
|
|
|
static struct audit_rule *audit_krule_to_rule(struct audit_krule *krule)
|
|
|
|
{
|
|
|
|
struct audit_rule *rule;
|
|
|
|
int i;
|
|
|
|
|
2006-12-06 21:38:51 -07:00
|
|
|
rule = kzalloc(sizeof(*rule), GFP_KERNEL);
|
2006-02-07 10:05:27 -07:00
|
|
|
if (unlikely(!rule))
|
2006-05-02 13:06:01 -06:00
|
|
|
return NULL;
|
2006-02-07 10:05:27 -07:00
|
|
|
|
|
|
|
rule->flags = krule->flags | krule->listnr;
|
|
|
|
rule->action = krule->action;
|
|
|
|
rule->field_count = krule->field_count;
|
|
|
|
for (i = 0; i < rule->field_count; i++) {
|
|
|
|
rule->values[i] = krule->fields[i].val;
|
|
|
|
rule->fields[i] = krule->fields[i].type;
|
|
|
|
|
|
|
|
if (krule->vers_ops == 1) {
|
|
|
|
if (krule->fields[i].op & AUDIT_NOT_EQUAL)
|
|
|
|
rule->fields[i] |= AUDIT_NEGATE;
|
|
|
|
} else {
|
|
|
|
rule->fields[i] |= krule->fields[i].op;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
for (i = 0; i < AUDIT_BITMASK_SIZE; i++) rule->mask[i] = krule->mask[i];
|
|
|
|
|
|
|
|
return rule;
|
|
|
|
}
|
2005-12-15 11:33:52 -07:00
|
|
|
|
2006-02-07 10:05:27 -07:00
|
|
|
/* Translate kernel rule respresentation to struct audit_rule_data. */
|
|
|
|
static struct audit_rule_data *audit_krule_to_data(struct audit_krule *krule)
|
|
|
|
{
|
|
|
|
struct audit_rule_data *data;
|
|
|
|
void *bufp;
|
|
|
|
int i;
|
|
|
|
|
|
|
|
data = kmalloc(sizeof(*data) + krule->buflen, GFP_KERNEL);
|
|
|
|
if (unlikely(!data))
|
2006-05-02 13:06:01 -06:00
|
|
|
return NULL;
|
2006-02-07 10:05:27 -07:00
|
|
|
memset(data, 0, sizeof(*data));
|
|
|
|
|
|
|
|
data->flags = krule->flags | krule->listnr;
|
|
|
|
data->action = krule->action;
|
|
|
|
data->field_count = krule->field_count;
|
|
|
|
bufp = data->buf;
|
|
|
|
for (i = 0; i < data->field_count; i++) {
|
|
|
|
struct audit_field *f = &krule->fields[i];
|
|
|
|
|
|
|
|
data->fields[i] = f->type;
|
|
|
|
data->fieldflags[i] = f->op;
|
|
|
|
switch(f->type) {
|
2006-06-29 15:56:39 -06:00
|
|
|
case AUDIT_SUBJ_USER:
|
|
|
|
case AUDIT_SUBJ_ROLE:
|
|
|
|
case AUDIT_SUBJ_TYPE:
|
|
|
|
case AUDIT_SUBJ_SEN:
|
|
|
|
case AUDIT_SUBJ_CLR:
|
2006-06-29 15:57:08 -06:00
|
|
|
case AUDIT_OBJ_USER:
|
|
|
|
case AUDIT_OBJ_ROLE:
|
|
|
|
case AUDIT_OBJ_TYPE:
|
|
|
|
case AUDIT_OBJ_LEV_LOW:
|
|
|
|
case AUDIT_OBJ_LEV_HIGH:
|
2006-03-10 17:14:06 -07:00
|
|
|
data->buflen += data->values[i] =
|
2008-04-18 17:59:43 -06:00
|
|
|
audit_pack_string(&bufp, f->lsm_str);
|
2006-03-10 17:14:06 -07:00
|
|
|
break;
|
[PATCH] audit: path-based rules
In this implementation, audit registers inotify watches on the parent
directories of paths specified in audit rules. When audit's inotify
event handler is called, it updates any affected rules based on the
filesystem event. If the parent directory is renamed, removed, or its
filesystem is unmounted, audit removes all rules referencing that
inotify watch.
To keep things simple, this implementation limits location-based
auditing to the directory entries in an existing directory. Given
a path-based rule for /foo/bar/passwd, the following table applies:
passwd modified -- audit event logged
passwd replaced -- audit event logged, rules list updated
bar renamed -- rule removed
foo renamed -- untracked, meaning that the rule now applies to
the new location
Audit users typically want to have many rules referencing filesystem
objects, which can significantly impact filtering performance. This
patch also adds an inode-number-based rule hash to mitigate this
situation.
The patch is relative to the audit git tree:
http://kernel.org/git/?p=linux/kernel/git/viro/audit-current.git;a=summary
and uses the inotify kernel API:
http://lkml.org/lkml/2006/6/1/145
Signed-off-by: Amy Griffis <amy.griffis@hp.com>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
2006-04-07 14:55:56 -06:00
|
|
|
case AUDIT_WATCH:
|
|
|
|
data->buflen += data->values[i] =
|
|
|
|
audit_pack_string(&bufp, krule->watch->path);
|
|
|
|
break;
|
[PATCH] audit: watching subtrees
New kind of audit rule predicates: "object is visible in given subtree".
The part that can be sanely implemented, that is. Limitations:
* if you have hardlink from outside of tree, you'd better watch
it too (or just watch the object itself, obviously)
* if you mount something under a watched tree, tell audit
that new chunk should be added to watched subtrees
* if you umount something in a watched tree and it's still mounted
elsewhere, you will get matches on events happening there. New command
tells audit to recalculate the trees, trimming such sources of false
positives.
Note that it's _not_ about path - if something mounted in several places
(multiple mount, bindings, different namespaces, etc.), the match does
_not_ depend on which one we are using for access.
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
2007-07-22 06:04:18 -06:00
|
|
|
case AUDIT_DIR:
|
|
|
|
data->buflen += data->values[i] =
|
|
|
|
audit_pack_string(&bufp,
|
|
|
|
audit_tree_path(krule->tree));
|
|
|
|
break;
|
2006-06-14 16:45:21 -06:00
|
|
|
case AUDIT_FILTERKEY:
|
|
|
|
data->buflen += data->values[i] =
|
|
|
|
audit_pack_string(&bufp, krule->filterkey);
|
|
|
|
break;
|
2006-02-07 10:05:27 -07:00
|
|
|
default:
|
|
|
|
data->values[i] = f->val;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
for (i = 0; i < AUDIT_BITMASK_SIZE; i++) data->mask[i] = krule->mask[i];
|
|
|
|
|
|
|
|
return data;
|
|
|
|
}
|
|
|
|
|
|
|
|
/* Compare two rules in kernel format. Considered success if rules
|
|
|
|
* don't match. */
|
|
|
|
static int audit_compare_rule(struct audit_krule *a, struct audit_krule *b)
|
|
|
|
{
|
|
|
|
int i;
|
|
|
|
|
|
|
|
if (a->flags != b->flags ||
|
|
|
|
a->listnr != b->listnr ||
|
|
|
|
a->action != b->action ||
|
|
|
|
a->field_count != b->field_count)
|
2005-12-15 11:33:52 -07:00
|
|
|
return 1;
|
|
|
|
|
|
|
|
for (i = 0; i < a->field_count; i++) {
|
2006-02-07 10:05:27 -07:00
|
|
|
if (a->fields[i].type != b->fields[i].type ||
|
|
|
|
a->fields[i].op != b->fields[i].op)
|
2005-12-15 11:33:52 -07:00
|
|
|
return 1;
|
2006-02-07 10:05:27 -07:00
|
|
|
|
|
|
|
switch(a->fields[i].type) {
|
2006-06-29 15:56:39 -06:00
|
|
|
case AUDIT_SUBJ_USER:
|
|
|
|
case AUDIT_SUBJ_ROLE:
|
|
|
|
case AUDIT_SUBJ_TYPE:
|
|
|
|
case AUDIT_SUBJ_SEN:
|
|
|
|
case AUDIT_SUBJ_CLR:
|
2006-06-29 15:57:08 -06:00
|
|
|
case AUDIT_OBJ_USER:
|
|
|
|
case AUDIT_OBJ_ROLE:
|
|
|
|
case AUDIT_OBJ_TYPE:
|
|
|
|
case AUDIT_OBJ_LEV_LOW:
|
|
|
|
case AUDIT_OBJ_LEV_HIGH:
|
2008-04-18 17:59:43 -06:00
|
|
|
if (strcmp(a->fields[i].lsm_str, b->fields[i].lsm_str))
|
2006-03-10 17:14:06 -07:00
|
|
|
return 1;
|
|
|
|
break;
|
[PATCH] audit: path-based rules
In this implementation, audit registers inotify watches on the parent
directories of paths specified in audit rules. When audit's inotify
event handler is called, it updates any affected rules based on the
filesystem event. If the parent directory is renamed, removed, or its
filesystem is unmounted, audit removes all rules referencing that
inotify watch.
To keep things simple, this implementation limits location-based
auditing to the directory entries in an existing directory. Given
a path-based rule for /foo/bar/passwd, the following table applies:
passwd modified -- audit event logged
passwd replaced -- audit event logged, rules list updated
bar renamed -- rule removed
foo renamed -- untracked, meaning that the rule now applies to
the new location
Audit users typically want to have many rules referencing filesystem
objects, which can significantly impact filtering performance. This
patch also adds an inode-number-based rule hash to mitigate this
situation.
The patch is relative to the audit git tree:
http://kernel.org/git/?p=linux/kernel/git/viro/audit-current.git;a=summary
and uses the inotify kernel API:
http://lkml.org/lkml/2006/6/1/145
Signed-off-by: Amy Griffis <amy.griffis@hp.com>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
2006-04-07 14:55:56 -06:00
|
|
|
case AUDIT_WATCH:
|
|
|
|
if (strcmp(a->watch->path, b->watch->path))
|
|
|
|
return 1;
|
|
|
|
break;
|
[PATCH] audit: watching subtrees
New kind of audit rule predicates: "object is visible in given subtree".
The part that can be sanely implemented, that is. Limitations:
* if you have hardlink from outside of tree, you'd better watch
it too (or just watch the object itself, obviously)
* if you mount something under a watched tree, tell audit
that new chunk should be added to watched subtrees
* if you umount something in a watched tree and it's still mounted
elsewhere, you will get matches on events happening there. New command
tells audit to recalculate the trees, trimming such sources of false
positives.
Note that it's _not_ about path - if something mounted in several places
(multiple mount, bindings, different namespaces, etc.), the match does
_not_ depend on which one we are using for access.
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
2007-07-22 06:04:18 -06:00
|
|
|
case AUDIT_DIR:
|
|
|
|
if (strcmp(audit_tree_path(a->tree),
|
|
|
|
audit_tree_path(b->tree)))
|
|
|
|
return 1;
|
|
|
|
break;
|
2006-06-14 16:45:21 -06:00
|
|
|
case AUDIT_FILTERKEY:
|
|
|
|
/* both filterkeys exist based on above type compare */
|
|
|
|
if (strcmp(a->filterkey, b->filterkey))
|
|
|
|
return 1;
|
|
|
|
break;
|
2006-02-07 10:05:27 -07:00
|
|
|
default:
|
|
|
|
if (a->fields[i].val != b->fields[i].val)
|
|
|
|
return 1;
|
|
|
|
}
|
2005-12-15 11:33:52 -07:00
|
|
|
}
|
|
|
|
|
|
|
|
for (i = 0; i < AUDIT_BITMASK_SIZE; i++)
|
|
|
|
if (a->mask[i] != b->mask[i])
|
|
|
|
return 1;
|
|
|
|
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
[PATCH] audit: path-based rules
In this implementation, audit registers inotify watches on the parent
directories of paths specified in audit rules. When audit's inotify
event handler is called, it updates any affected rules based on the
filesystem event. If the parent directory is renamed, removed, or its
filesystem is unmounted, audit removes all rules referencing that
inotify watch.
To keep things simple, this implementation limits location-based
auditing to the directory entries in an existing directory. Given
a path-based rule for /foo/bar/passwd, the following table applies:
passwd modified -- audit event logged
passwd replaced -- audit event logged, rules list updated
bar renamed -- rule removed
foo renamed -- untracked, meaning that the rule now applies to
the new location
Audit users typically want to have many rules referencing filesystem
objects, which can significantly impact filtering performance. This
patch also adds an inode-number-based rule hash to mitigate this
situation.
The patch is relative to the audit git tree:
http://kernel.org/git/?p=linux/kernel/git/viro/audit-current.git;a=summary
and uses the inotify kernel API:
http://lkml.org/lkml/2006/6/1/145
Signed-off-by: Amy Griffis <amy.griffis@hp.com>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
2006-04-07 14:55:56 -06:00
|
|
|
/* Duplicate the given audit watch. The new watch's rules list is initialized
|
|
|
|
* to an empty list and wlist is undefined. */
|
|
|
|
static struct audit_watch *audit_dupe_watch(struct audit_watch *old)
|
|
|
|
{
|
|
|
|
char *path;
|
|
|
|
struct audit_watch *new;
|
|
|
|
|
|
|
|
path = kstrdup(old->path, GFP_KERNEL);
|
|
|
|
if (unlikely(!path))
|
|
|
|
return ERR_PTR(-ENOMEM);
|
|
|
|
|
|
|
|
new = audit_init_watch(path);
|
2008-04-29 02:03:09 -06:00
|
|
|
if (IS_ERR(new)) {
|
[PATCH] audit: path-based rules
In this implementation, audit registers inotify watches on the parent
directories of paths specified in audit rules. When audit's inotify
event handler is called, it updates any affected rules based on the
filesystem event. If the parent directory is renamed, removed, or its
filesystem is unmounted, audit removes all rules referencing that
inotify watch.
To keep things simple, this implementation limits location-based
auditing to the directory entries in an existing directory. Given
a path-based rule for /foo/bar/passwd, the following table applies:
passwd modified -- audit event logged
passwd replaced -- audit event logged, rules list updated
bar renamed -- rule removed
foo renamed -- untracked, meaning that the rule now applies to
the new location
Audit users typically want to have many rules referencing filesystem
objects, which can significantly impact filtering performance. This
patch also adds an inode-number-based rule hash to mitigate this
situation.
The patch is relative to the audit git tree:
http://kernel.org/git/?p=linux/kernel/git/viro/audit-current.git;a=summary
and uses the inotify kernel API:
http://lkml.org/lkml/2006/6/1/145
Signed-off-by: Amy Griffis <amy.griffis@hp.com>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
2006-04-07 14:55:56 -06:00
|
|
|
kfree(path);
|
|
|
|
goto out;
|
|
|
|
}
|
|
|
|
|
|
|
|
new->dev = old->dev;
|
|
|
|
new->ino = old->ino;
|
|
|
|
get_inotify_watch(&old->parent->wdata);
|
|
|
|
new->parent = old->parent;
|
|
|
|
|
|
|
|
out:
|
|
|
|
return new;
|
|
|
|
}
|
|
|
|
|
2008-04-18 17:59:43 -06:00
|
|
|
/* Duplicate LSM field information. The lsm_rule is opaque, so must be
|
2006-03-10 17:14:06 -07:00
|
|
|
* re-initialized. */
|
2008-03-01 13:01:11 -07:00
|
|
|
static inline int audit_dupe_lsm_field(struct audit_field *df,
|
2006-03-10 17:14:06 -07:00
|
|
|
struct audit_field *sf)
|
|
|
|
{
|
|
|
|
int ret = 0;
|
2008-04-18 17:59:43 -06:00
|
|
|
char *lsm_str;
|
2006-03-10 17:14:06 -07:00
|
|
|
|
2008-04-18 17:59:43 -06:00
|
|
|
/* our own copy of lsm_str */
|
|
|
|
lsm_str = kstrdup(sf->lsm_str, GFP_KERNEL);
|
|
|
|
if (unlikely(!lsm_str))
|
2006-12-22 02:10:02 -07:00
|
|
|
return -ENOMEM;
|
2008-04-18 17:59:43 -06:00
|
|
|
df->lsm_str = lsm_str;
|
2006-03-10 17:14:06 -07:00
|
|
|
|
2008-04-18 17:59:43 -06:00
|
|
|
/* our own (refreshed) copy of lsm_rule */
|
|
|
|
ret = security_audit_rule_init(df->type, df->op, df->lsm_str,
|
|
|
|
(void **)&df->lsm_rule);
|
2006-03-10 17:14:06 -07:00
|
|
|
/* Keep currently invalid fields around in case they
|
|
|
|
* become valid after a policy reload. */
|
|
|
|
if (ret == -EINVAL) {
|
2008-03-01 13:01:11 -07:00
|
|
|
printk(KERN_WARNING "audit rule for LSM \'%s\' is "
|
2008-04-18 17:59:43 -06:00
|
|
|
"invalid\n", df->lsm_str);
|
2006-03-10 17:14:06 -07:00
|
|
|
ret = 0;
|
|
|
|
}
|
|
|
|
|
|
|
|
return ret;
|
|
|
|
}
|
|
|
|
|
|
|
|
/* Duplicate an audit rule. This will be a deep copy with the exception
|
2008-03-01 13:01:11 -07:00
|
|
|
* of the watch - that pointer is carried over. The LSM specific fields
|
2006-03-10 17:14:06 -07:00
|
|
|
* will be updated in the copy. The point is to be able to replace the old
|
[PATCH] audit: path-based rules
In this implementation, audit registers inotify watches on the parent
directories of paths specified in audit rules. When audit's inotify
event handler is called, it updates any affected rules based on the
filesystem event. If the parent directory is renamed, removed, or its
filesystem is unmounted, audit removes all rules referencing that
inotify watch.
To keep things simple, this implementation limits location-based
auditing to the directory entries in an existing directory. Given
a path-based rule for /foo/bar/passwd, the following table applies:
passwd modified -- audit event logged
passwd replaced -- audit event logged, rules list updated
bar renamed -- rule removed
foo renamed -- untracked, meaning that the rule now applies to
the new location
Audit users typically want to have many rules referencing filesystem
objects, which can significantly impact filtering performance. This
patch also adds an inode-number-based rule hash to mitigate this
situation.
The patch is relative to the audit git tree:
http://kernel.org/git/?p=linux/kernel/git/viro/audit-current.git;a=summary
and uses the inotify kernel API:
http://lkml.org/lkml/2006/6/1/145
Signed-off-by: Amy Griffis <amy.griffis@hp.com>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
2006-04-07 14:55:56 -06:00
|
|
|
* rule with the new rule in the filterlist, then free the old rule.
|
|
|
|
* The rlist element is undefined; list manipulations are handled apart from
|
|
|
|
* the initial copy. */
|
|
|
|
static struct audit_entry *audit_dupe_rule(struct audit_krule *old,
|
|
|
|
struct audit_watch *watch)
|
2006-03-10 17:14:06 -07:00
|
|
|
{
|
|
|
|
u32 fcount = old->field_count;
|
|
|
|
struct audit_entry *entry;
|
|
|
|
struct audit_krule *new;
|
2006-06-14 16:45:21 -06:00
|
|
|
char *fk;
|
2006-03-10 17:14:06 -07:00
|
|
|
int i, err = 0;
|
|
|
|
|
|
|
|
entry = audit_init_entry(fcount);
|
|
|
|
if (unlikely(!entry))
|
|
|
|
return ERR_PTR(-ENOMEM);
|
|
|
|
|
|
|
|
new = &entry->rule;
|
|
|
|
new->vers_ops = old->vers_ops;
|
|
|
|
new->flags = old->flags;
|
|
|
|
new->listnr = old->listnr;
|
|
|
|
new->action = old->action;
|
|
|
|
for (i = 0; i < AUDIT_BITMASK_SIZE; i++)
|
|
|
|
new->mask[i] = old->mask[i];
|
|
|
|
new->buflen = old->buflen;
|
[PATCH] audit: path-based rules
In this implementation, audit registers inotify watches on the parent
directories of paths specified in audit rules. When audit's inotify
event handler is called, it updates any affected rules based on the
filesystem event. If the parent directory is renamed, removed, or its
filesystem is unmounted, audit removes all rules referencing that
inotify watch.
To keep things simple, this implementation limits location-based
auditing to the directory entries in an existing directory. Given
a path-based rule for /foo/bar/passwd, the following table applies:
passwd modified -- audit event logged
passwd replaced -- audit event logged, rules list updated
bar renamed -- rule removed
foo renamed -- untracked, meaning that the rule now applies to
the new location
Audit users typically want to have many rules referencing filesystem
objects, which can significantly impact filtering performance. This
patch also adds an inode-number-based rule hash to mitigate this
situation.
The patch is relative to the audit git tree:
http://kernel.org/git/?p=linux/kernel/git/viro/audit-current.git;a=summary
and uses the inotify kernel API:
http://lkml.org/lkml/2006/6/1/145
Signed-off-by: Amy Griffis <amy.griffis@hp.com>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
2006-04-07 14:55:56 -06:00
|
|
|
new->inode_f = old->inode_f;
|
|
|
|
new->watch = NULL;
|
2006-03-10 17:14:06 -07:00
|
|
|
new->field_count = old->field_count;
|
[PATCH] audit: watching subtrees
New kind of audit rule predicates: "object is visible in given subtree".
The part that can be sanely implemented, that is. Limitations:
* if you have hardlink from outside of tree, you'd better watch
it too (or just watch the object itself, obviously)
* if you mount something under a watched tree, tell audit
that new chunk should be added to watched subtrees
* if you umount something in a watched tree and it's still mounted
elsewhere, you will get matches on events happening there. New command
tells audit to recalculate the trees, trimming such sources of false
positives.
Note that it's _not_ about path - if something mounted in several places
(multiple mount, bindings, different namespaces, etc.), the match does
_not_ depend on which one we are using for access.
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
2007-07-22 06:04:18 -06:00
|
|
|
/*
|
|
|
|
* note that we are OK with not refcounting here; audit_match_tree()
|
|
|
|
* never dereferences tree and we can't get false positives there
|
|
|
|
* since we'd have to have rule gone from the list *and* removed
|
|
|
|
* before the chunks found by lookup had been allocated, i.e. before
|
|
|
|
* the beginning of list scan.
|
|
|
|
*/
|
|
|
|
new->tree = old->tree;
|
2006-03-10 17:14:06 -07:00
|
|
|
memcpy(new->fields, old->fields, sizeof(struct audit_field) * fcount);
|
|
|
|
|
2008-04-18 17:59:43 -06:00
|
|
|
/* deep copy this information, updating the lsm_rule fields, because
|
2006-03-10 17:14:06 -07:00
|
|
|
* the originals will all be freed when the old rule is freed. */
|
|
|
|
for (i = 0; i < fcount; i++) {
|
|
|
|
switch (new->fields[i].type) {
|
2006-06-29 15:56:39 -06:00
|
|
|
case AUDIT_SUBJ_USER:
|
|
|
|
case AUDIT_SUBJ_ROLE:
|
|
|
|
case AUDIT_SUBJ_TYPE:
|
|
|
|
case AUDIT_SUBJ_SEN:
|
|
|
|
case AUDIT_SUBJ_CLR:
|
2006-06-29 15:57:08 -06:00
|
|
|
case AUDIT_OBJ_USER:
|
|
|
|
case AUDIT_OBJ_ROLE:
|
|
|
|
case AUDIT_OBJ_TYPE:
|
|
|
|
case AUDIT_OBJ_LEV_LOW:
|
|
|
|
case AUDIT_OBJ_LEV_HIGH:
|
2008-03-01 13:01:11 -07:00
|
|
|
err = audit_dupe_lsm_field(&new->fields[i],
|
2006-03-10 17:14:06 -07:00
|
|
|
&old->fields[i]);
|
2006-06-14 16:45:21 -06:00
|
|
|
break;
|
|
|
|
case AUDIT_FILTERKEY:
|
|
|
|
fk = kstrdup(old->filterkey, GFP_KERNEL);
|
|
|
|
if (unlikely(!fk))
|
|
|
|
err = -ENOMEM;
|
|
|
|
else
|
|
|
|
new->filterkey = fk;
|
2006-03-10 17:14:06 -07:00
|
|
|
}
|
|
|
|
if (err) {
|
|
|
|
audit_free_rule(entry);
|
|
|
|
return ERR_PTR(err);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
[PATCH] audit: path-based rules
In this implementation, audit registers inotify watches on the parent
directories of paths specified in audit rules. When audit's inotify
event handler is called, it updates any affected rules based on the
filesystem event. If the parent directory is renamed, removed, or its
filesystem is unmounted, audit removes all rules referencing that
inotify watch.
To keep things simple, this implementation limits location-based
auditing to the directory entries in an existing directory. Given
a path-based rule for /foo/bar/passwd, the following table applies:
passwd modified -- audit event logged
passwd replaced -- audit event logged, rules list updated
bar renamed -- rule removed
foo renamed -- untracked, meaning that the rule now applies to
the new location
Audit users typically want to have many rules referencing filesystem
objects, which can significantly impact filtering performance. This
patch also adds an inode-number-based rule hash to mitigate this
situation.
The patch is relative to the audit git tree:
http://kernel.org/git/?p=linux/kernel/git/viro/audit-current.git;a=summary
and uses the inotify kernel API:
http://lkml.org/lkml/2006/6/1/145
Signed-off-by: Amy Griffis <amy.griffis@hp.com>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
2006-04-07 14:55:56 -06:00
|
|
|
if (watch) {
|
|
|
|
audit_get_watch(watch);
|
|
|
|
new->watch = watch;
|
|
|
|
}
|
|
|
|
|
2006-03-10 17:14:06 -07:00
|
|
|
return entry;
|
|
|
|
}
|
|
|
|
|
[PATCH] audit: path-based rules
In this implementation, audit registers inotify watches on the parent
directories of paths specified in audit rules. When audit's inotify
event handler is called, it updates any affected rules based on the
filesystem event. If the parent directory is renamed, removed, or its
filesystem is unmounted, audit removes all rules referencing that
inotify watch.
To keep things simple, this implementation limits location-based
auditing to the directory entries in an existing directory. Given
a path-based rule for /foo/bar/passwd, the following table applies:
passwd modified -- audit event logged
passwd replaced -- audit event logged, rules list updated
bar renamed -- rule removed
foo renamed -- untracked, meaning that the rule now applies to
the new location
Audit users typically want to have many rules referencing filesystem
objects, which can significantly impact filtering performance. This
patch also adds an inode-number-based rule hash to mitigate this
situation.
The patch is relative to the audit git tree:
http://kernel.org/git/?p=linux/kernel/git/viro/audit-current.git;a=summary
and uses the inotify kernel API:
http://lkml.org/lkml/2006/6/1/145
Signed-off-by: Amy Griffis <amy.griffis@hp.com>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
2006-04-07 14:55:56 -06:00
|
|
|
/* Update inode info in audit rules based on filesystem event. */
|
|
|
|
static void audit_update_watch(struct audit_parent *parent,
|
|
|
|
const char *dname, dev_t dev,
|
|
|
|
unsigned long ino, unsigned invalidating)
|
|
|
|
{
|
|
|
|
struct audit_watch *owatch, *nwatch, *nextw;
|
|
|
|
struct audit_krule *r, *nextr;
|
|
|
|
struct audit_entry *oentry, *nentry;
|
|
|
|
|
|
|
|
mutex_lock(&audit_filter_mutex);
|
|
|
|
list_for_each_entry_safe(owatch, nextw, &parent->watches, wlist) {
|
2006-06-08 21:19:31 -06:00
|
|
|
if (audit_compare_dname_path(dname, owatch->path, NULL))
|
[PATCH] audit: path-based rules
In this implementation, audit registers inotify watches on the parent
directories of paths specified in audit rules. When audit's inotify
event handler is called, it updates any affected rules based on the
filesystem event. If the parent directory is renamed, removed, or its
filesystem is unmounted, audit removes all rules referencing that
inotify watch.
To keep things simple, this implementation limits location-based
auditing to the directory entries in an existing directory. Given
a path-based rule for /foo/bar/passwd, the following table applies:
passwd modified -- audit event logged
passwd replaced -- audit event logged, rules list updated
bar renamed -- rule removed
foo renamed -- untracked, meaning that the rule now applies to
the new location
Audit users typically want to have many rules referencing filesystem
objects, which can significantly impact filtering performance. This
patch also adds an inode-number-based rule hash to mitigate this
situation.
The patch is relative to the audit git tree:
http://kernel.org/git/?p=linux/kernel/git/viro/audit-current.git;a=summary
and uses the inotify kernel API:
http://lkml.org/lkml/2006/6/1/145
Signed-off-by: Amy Griffis <amy.griffis@hp.com>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
2006-04-07 14:55:56 -06:00
|
|
|
continue;
|
|
|
|
|
|
|
|
/* If the update involves invalidating rules, do the inode-based
|
|
|
|
* filtering now, so we don't omit records. */
|
2007-06-23 18:16:47 -06:00
|
|
|
if (invalidating && current->audit_context &&
|
[PATCH] audit: path-based rules
In this implementation, audit registers inotify watches on the parent
directories of paths specified in audit rules. When audit's inotify
event handler is called, it updates any affected rules based on the
filesystem event. If the parent directory is renamed, removed, or its
filesystem is unmounted, audit removes all rules referencing that
inotify watch.
To keep things simple, this implementation limits location-based
auditing to the directory entries in an existing directory. Given
a path-based rule for /foo/bar/passwd, the following table applies:
passwd modified -- audit event logged
passwd replaced -- audit event logged, rules list updated
bar renamed -- rule removed
foo renamed -- untracked, meaning that the rule now applies to
the new location
Audit users typically want to have many rules referencing filesystem
objects, which can significantly impact filtering performance. This
patch also adds an inode-number-based rule hash to mitigate this
situation.
The patch is relative to the audit git tree:
http://kernel.org/git/?p=linux/kernel/git/viro/audit-current.git;a=summary
and uses the inotify kernel API:
http://lkml.org/lkml/2006/6/1/145
Signed-off-by: Amy Griffis <amy.griffis@hp.com>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
2006-04-07 14:55:56 -06:00
|
|
|
audit_filter_inodes(current, current->audit_context) == AUDIT_RECORD_CONTEXT)
|
|
|
|
audit_set_auditable(current->audit_context);
|
|
|
|
|
|
|
|
nwatch = audit_dupe_watch(owatch);
|
2008-04-29 02:03:09 -06:00
|
|
|
if (IS_ERR(nwatch)) {
|
[PATCH] audit: path-based rules
In this implementation, audit registers inotify watches on the parent
directories of paths specified in audit rules. When audit's inotify
event handler is called, it updates any affected rules based on the
filesystem event. If the parent directory is renamed, removed, or its
filesystem is unmounted, audit removes all rules referencing that
inotify watch.
To keep things simple, this implementation limits location-based
auditing to the directory entries in an existing directory. Given
a path-based rule for /foo/bar/passwd, the following table applies:
passwd modified -- audit event logged
passwd replaced -- audit event logged, rules list updated
bar renamed -- rule removed
foo renamed -- untracked, meaning that the rule now applies to
the new location
Audit users typically want to have many rules referencing filesystem
objects, which can significantly impact filtering performance. This
patch also adds an inode-number-based rule hash to mitigate this
situation.
The patch is relative to the audit git tree:
http://kernel.org/git/?p=linux/kernel/git/viro/audit-current.git;a=summary
and uses the inotify kernel API:
http://lkml.org/lkml/2006/6/1/145
Signed-off-by: Amy Griffis <amy.griffis@hp.com>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
2006-04-07 14:55:56 -06:00
|
|
|
mutex_unlock(&audit_filter_mutex);
|
|
|
|
audit_panic("error updating watch, skipping");
|
|
|
|
return;
|
|
|
|
}
|
|
|
|
nwatch->dev = dev;
|
|
|
|
nwatch->ino = ino;
|
|
|
|
|
|
|
|
list_for_each_entry_safe(r, nextr, &owatch->rules, rlist) {
|
|
|
|
|
|
|
|
oentry = container_of(r, struct audit_entry, rule);
|
|
|
|
list_del(&oentry->rule.rlist);
|
|
|
|
list_del_rcu(&oentry->list);
|
|
|
|
|
|
|
|
nentry = audit_dupe_rule(&oentry->rule, nwatch);
|
2008-04-29 02:03:09 -06:00
|
|
|
if (IS_ERR(nentry))
|
[PATCH] audit: path-based rules
In this implementation, audit registers inotify watches on the parent
directories of paths specified in audit rules. When audit's inotify
event handler is called, it updates any affected rules based on the
filesystem event. If the parent directory is renamed, removed, or its
filesystem is unmounted, audit removes all rules referencing that
inotify watch.
To keep things simple, this implementation limits location-based
auditing to the directory entries in an existing directory. Given
a path-based rule for /foo/bar/passwd, the following table applies:
passwd modified -- audit event logged
passwd replaced -- audit event logged, rules list updated
bar renamed -- rule removed
foo renamed -- untracked, meaning that the rule now applies to
the new location
Audit users typically want to have many rules referencing filesystem
objects, which can significantly impact filtering performance. This
patch also adds an inode-number-based rule hash to mitigate this
situation.
The patch is relative to the audit git tree:
http://kernel.org/git/?p=linux/kernel/git/viro/audit-current.git;a=summary
and uses the inotify kernel API:
http://lkml.org/lkml/2006/6/1/145
Signed-off-by: Amy Griffis <amy.griffis@hp.com>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
2006-04-07 14:55:56 -06:00
|
|
|
audit_panic("error updating watch, removing");
|
|
|
|
else {
|
|
|
|
int h = audit_hash_ino((u32)ino);
|
|
|
|
list_add(&nentry->rule.rlist, &nwatch->rules);
|
|
|
|
list_add_rcu(&nentry->list, &audit_inode_hash[h]);
|
|
|
|
}
|
|
|
|
|
|
|
|
call_rcu(&oentry->rcu, audit_free_rule_rcu);
|
|
|
|
}
|
|
|
|
|
2008-01-07 15:09:31 -07:00
|
|
|
if (audit_enabled) {
|
|
|
|
struct audit_buffer *ab;
|
|
|
|
ab = audit_log_start(NULL, GFP_KERNEL,
|
|
|
|
AUDIT_CONFIG_CHANGE);
|
|
|
|
audit_log_format(ab,
|
|
|
|
"op=updated rules specifying path=");
|
|
|
|
audit_log_untrustedstring(ab, owatch->path);
|
|
|
|
audit_log_format(ab, " with dev=%u ino=%lu\n",
|
|
|
|
dev, ino);
|
|
|
|
audit_log_format(ab, " list=%d res=1", r->listnr);
|
|
|
|
audit_log_end(ab);
|
|
|
|
}
|
[PATCH] audit: path-based rules
In this implementation, audit registers inotify watches on the parent
directories of paths specified in audit rules. When audit's inotify
event handler is called, it updates any affected rules based on the
filesystem event. If the parent directory is renamed, removed, or its
filesystem is unmounted, audit removes all rules referencing that
inotify watch.
To keep things simple, this implementation limits location-based
auditing to the directory entries in an existing directory. Given
a path-based rule for /foo/bar/passwd, the following table applies:
passwd modified -- audit event logged
passwd replaced -- audit event logged, rules list updated
bar renamed -- rule removed
foo renamed -- untracked, meaning that the rule now applies to
the new location
Audit users typically want to have many rules referencing filesystem
objects, which can significantly impact filtering performance. This
patch also adds an inode-number-based rule hash to mitigate this
situation.
The patch is relative to the audit git tree:
http://kernel.org/git/?p=linux/kernel/git/viro/audit-current.git;a=summary
and uses the inotify kernel API:
http://lkml.org/lkml/2006/6/1/145
Signed-off-by: Amy Griffis <amy.griffis@hp.com>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
2006-04-07 14:55:56 -06:00
|
|
|
audit_remove_watch(owatch);
|
|
|
|
goto add_watch_to_parent; /* event applies to a single watch */
|
|
|
|
}
|
|
|
|
mutex_unlock(&audit_filter_mutex);
|
|
|
|
return;
|
|
|
|
|
|
|
|
add_watch_to_parent:
|
|
|
|
list_add(&nwatch->wlist, &parent->watches);
|
|
|
|
mutex_unlock(&audit_filter_mutex);
|
|
|
|
return;
|
|
|
|
}
|
|
|
|
|
|
|
|
/* Remove all watches & rules associated with a parent that is going away. */
|
|
|
|
static void audit_remove_parent_watches(struct audit_parent *parent)
|
|
|
|
{
|
|
|
|
struct audit_watch *w, *nextw;
|
|
|
|
struct audit_krule *r, *nextr;
|
|
|
|
struct audit_entry *e;
|
|
|
|
|
|
|
|
mutex_lock(&audit_filter_mutex);
|
|
|
|
parent->flags |= AUDIT_PARENT_INVALID;
|
|
|
|
list_for_each_entry_safe(w, nextw, &parent->watches, wlist) {
|
|
|
|
list_for_each_entry_safe(r, nextr, &w->rules, rlist) {
|
|
|
|
e = container_of(r, struct audit_entry, rule);
|
2008-01-07 15:09:31 -07:00
|
|
|
if (audit_enabled) {
|
|
|
|
struct audit_buffer *ab;
|
|
|
|
ab = audit_log_start(NULL, GFP_KERNEL,
|
|
|
|
AUDIT_CONFIG_CHANGE);
|
|
|
|
audit_log_format(ab, "op=remove rule path=");
|
|
|
|
audit_log_untrustedstring(ab, w->path);
|
|
|
|
if (r->filterkey) {
|
|
|
|
audit_log_format(ab, " key=");
|
|
|
|
audit_log_untrustedstring(ab,
|
|
|
|
r->filterkey);
|
|
|
|
} else
|
|
|
|
audit_log_format(ab, " key=(null)");
|
|
|
|
audit_log_format(ab, " list=%d res=1",
|
|
|
|
r->listnr);
|
|
|
|
audit_log_end(ab);
|
|
|
|
}
|
[PATCH] audit: path-based rules
In this implementation, audit registers inotify watches on the parent
directories of paths specified in audit rules. When audit's inotify
event handler is called, it updates any affected rules based on the
filesystem event. If the parent directory is renamed, removed, or its
filesystem is unmounted, audit removes all rules referencing that
inotify watch.
To keep things simple, this implementation limits location-based
auditing to the directory entries in an existing directory. Given
a path-based rule for /foo/bar/passwd, the following table applies:
passwd modified -- audit event logged
passwd replaced -- audit event logged, rules list updated
bar renamed -- rule removed
foo renamed -- untracked, meaning that the rule now applies to
the new location
Audit users typically want to have many rules referencing filesystem
objects, which can significantly impact filtering performance. This
patch also adds an inode-number-based rule hash to mitigate this
situation.
The patch is relative to the audit git tree:
http://kernel.org/git/?p=linux/kernel/git/viro/audit-current.git;a=summary
and uses the inotify kernel API:
http://lkml.org/lkml/2006/6/1/145
Signed-off-by: Amy Griffis <amy.griffis@hp.com>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
2006-04-07 14:55:56 -06:00
|
|
|
list_del(&r->rlist);
|
|
|
|
list_del_rcu(&e->list);
|
|
|
|
call_rcu(&e->rcu, audit_free_rule_rcu);
|
|
|
|
}
|
|
|
|
audit_remove_watch(w);
|
|
|
|
}
|
|
|
|
mutex_unlock(&audit_filter_mutex);
|
|
|
|
}
|
|
|
|
|
|
|
|
/* Unregister inotify watches for parents on in_list.
|
|
|
|
* Generates an IN_IGNORED event. */
|
|
|
|
static void audit_inotify_unregister(struct list_head *in_list)
|
|
|
|
{
|
|
|
|
struct audit_parent *p, *n;
|
|
|
|
|
|
|
|
list_for_each_entry_safe(p, n, in_list, ilist) {
|
|
|
|
list_del(&p->ilist);
|
|
|
|
inotify_rm_watch(audit_ih, &p->wdata);
|
|
|
|
/* the put matching the get in audit_do_del_rule() */
|
|
|
|
put_inotify_watch(&p->wdata);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
/* Find an existing audit rule.
|
|
|
|
* Caller must hold audit_filter_mutex to prevent stale rule data. */
|
|
|
|
static struct audit_entry *audit_find_rule(struct audit_entry *entry,
|
|
|
|
struct list_head *list)
|
|
|
|
{
|
|
|
|
struct audit_entry *e, *found = NULL;
|
|
|
|
int h;
|
|
|
|
|
|
|
|
if (entry->rule.watch) {
|
|
|
|
/* we don't know the inode number, so must walk entire hash */
|
|
|
|
for (h = 0; h < AUDIT_INODE_BUCKETS; h++) {
|
|
|
|
list = &audit_inode_hash[h];
|
|
|
|
list_for_each_entry(e, list, list)
|
|
|
|
if (!audit_compare_rule(&entry->rule, &e->rule)) {
|
|
|
|
found = e;
|
|
|
|
goto out;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
goto out;
|
|
|
|
}
|
|
|
|
|
|
|
|
list_for_each_entry(e, list, list)
|
|
|
|
if (!audit_compare_rule(&entry->rule, &e->rule)) {
|
|
|
|
found = e;
|
|
|
|
goto out;
|
|
|
|
}
|
|
|
|
|
|
|
|
out:
|
|
|
|
return found;
|
|
|
|
}
|
|
|
|
|
|
|
|
/* Get path information necessary for adding watches. */
|
|
|
|
static int audit_get_nd(char *path, struct nameidata **ndp,
|
|
|
|
struct nameidata **ndw)
|
|
|
|
{
|
|
|
|
struct nameidata *ndparent, *ndwatch;
|
|
|
|
int err;
|
|
|
|
|
|
|
|
ndparent = kmalloc(sizeof(*ndparent), GFP_KERNEL);
|
|
|
|
if (unlikely(!ndparent))
|
|
|
|
return -ENOMEM;
|
|
|
|
|
|
|
|
ndwatch = kmalloc(sizeof(*ndwatch), GFP_KERNEL);
|
|
|
|
if (unlikely(!ndwatch)) {
|
|
|
|
kfree(ndparent);
|
|
|
|
return -ENOMEM;
|
|
|
|
}
|
|
|
|
|
|
|
|
err = path_lookup(path, LOOKUP_PARENT, ndparent);
|
|
|
|
if (err) {
|
|
|
|
kfree(ndparent);
|
|
|
|
kfree(ndwatch);
|
|
|
|
return err;
|
|
|
|
}
|
|
|
|
|
|
|
|
err = path_lookup(path, 0, ndwatch);
|
|
|
|
if (err) {
|
|
|
|
kfree(ndwatch);
|
|
|
|
ndwatch = NULL;
|
|
|
|
}
|
|
|
|
|
|
|
|
*ndp = ndparent;
|
|
|
|
*ndw = ndwatch;
|
|
|
|
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
|
|
|
/* Release resources used for watch path information. */
|
|
|
|
static void audit_put_nd(struct nameidata *ndp, struct nameidata *ndw)
|
|
|
|
{
|
|
|
|
if (ndp) {
|
2008-02-14 20:34:35 -07:00
|
|
|
path_put(&ndp->path);
|
[PATCH] audit: path-based rules
In this implementation, audit registers inotify watches on the parent
directories of paths specified in audit rules. When audit's inotify
event handler is called, it updates any affected rules based on the
filesystem event. If the parent directory is renamed, removed, or its
filesystem is unmounted, audit removes all rules referencing that
inotify watch.
To keep things simple, this implementation limits location-based
auditing to the directory entries in an existing directory. Given
a path-based rule for /foo/bar/passwd, the following table applies:
passwd modified -- audit event logged
passwd replaced -- audit event logged, rules list updated
bar renamed -- rule removed
foo renamed -- untracked, meaning that the rule now applies to
the new location
Audit users typically want to have many rules referencing filesystem
objects, which can significantly impact filtering performance. This
patch also adds an inode-number-based rule hash to mitigate this
situation.
The patch is relative to the audit git tree:
http://kernel.org/git/?p=linux/kernel/git/viro/audit-current.git;a=summary
and uses the inotify kernel API:
http://lkml.org/lkml/2006/6/1/145
Signed-off-by: Amy Griffis <amy.griffis@hp.com>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
2006-04-07 14:55:56 -06:00
|
|
|
kfree(ndp);
|
|
|
|
}
|
|
|
|
if (ndw) {
|
2008-02-14 20:34:35 -07:00
|
|
|
path_put(&ndw->path);
|
[PATCH] audit: path-based rules
In this implementation, audit registers inotify watches on the parent
directories of paths specified in audit rules. When audit's inotify
event handler is called, it updates any affected rules based on the
filesystem event. If the parent directory is renamed, removed, or its
filesystem is unmounted, audit removes all rules referencing that
inotify watch.
To keep things simple, this implementation limits location-based
auditing to the directory entries in an existing directory. Given
a path-based rule for /foo/bar/passwd, the following table applies:
passwd modified -- audit event logged
passwd replaced -- audit event logged, rules list updated
bar renamed -- rule removed
foo renamed -- untracked, meaning that the rule now applies to
the new location
Audit users typically want to have many rules referencing filesystem
objects, which can significantly impact filtering performance. This
patch also adds an inode-number-based rule hash to mitigate this
situation.
The patch is relative to the audit git tree:
http://kernel.org/git/?p=linux/kernel/git/viro/audit-current.git;a=summary
and uses the inotify kernel API:
http://lkml.org/lkml/2006/6/1/145
Signed-off-by: Amy Griffis <amy.griffis@hp.com>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
2006-04-07 14:55:56 -06:00
|
|
|
kfree(ndw);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
/* Associate the given rule with an existing parent inotify_watch.
|
|
|
|
* Caller must hold audit_filter_mutex. */
|
|
|
|
static void audit_add_to_parent(struct audit_krule *krule,
|
|
|
|
struct audit_parent *parent)
|
|
|
|
{
|
|
|
|
struct audit_watch *w, *watch = krule->watch;
|
|
|
|
int watch_found = 0;
|
|
|
|
|
|
|
|
list_for_each_entry(w, &parent->watches, wlist) {
|
|
|
|
if (strcmp(watch->path, w->path))
|
|
|
|
continue;
|
|
|
|
|
|
|
|
watch_found = 1;
|
|
|
|
|
|
|
|
/* put krule's and initial refs to temporary watch */
|
|
|
|
audit_put_watch(watch);
|
|
|
|
audit_put_watch(watch);
|
|
|
|
|
|
|
|
audit_get_watch(w);
|
|
|
|
krule->watch = watch = w;
|
|
|
|
break;
|
|
|
|
}
|
|
|
|
|
|
|
|
if (!watch_found) {
|
|
|
|
get_inotify_watch(&parent->wdata);
|
|
|
|
watch->parent = parent;
|
|
|
|
|
|
|
|
list_add(&watch->wlist, &parent->watches);
|
|
|
|
}
|
|
|
|
list_add(&krule->rlist, &watch->rules);
|
|
|
|
}
|
|
|
|
|
|
|
|
/* Find a matching watch entry, or add this one.
|
|
|
|
* Caller must hold audit_filter_mutex. */
|
|
|
|
static int audit_add_watch(struct audit_krule *krule, struct nameidata *ndp,
|
|
|
|
struct nameidata *ndw)
|
|
|
|
{
|
|
|
|
struct audit_watch *watch = krule->watch;
|
|
|
|
struct inotify_watch *i_watch;
|
|
|
|
struct audit_parent *parent;
|
|
|
|
int ret = 0;
|
|
|
|
|
|
|
|
/* update watch filter fields */
|
|
|
|
if (ndw) {
|
2008-02-14 20:34:32 -07:00
|
|
|
watch->dev = ndw->path.dentry->d_inode->i_sb->s_dev;
|
|
|
|
watch->ino = ndw->path.dentry->d_inode->i_ino;
|
[PATCH] audit: path-based rules
In this implementation, audit registers inotify watches on the parent
directories of paths specified in audit rules. When audit's inotify
event handler is called, it updates any affected rules based on the
filesystem event. If the parent directory is renamed, removed, or its
filesystem is unmounted, audit removes all rules referencing that
inotify watch.
To keep things simple, this implementation limits location-based
auditing to the directory entries in an existing directory. Given
a path-based rule for /foo/bar/passwd, the following table applies:
passwd modified -- audit event logged
passwd replaced -- audit event logged, rules list updated
bar renamed -- rule removed
foo renamed -- untracked, meaning that the rule now applies to
the new location
Audit users typically want to have many rules referencing filesystem
objects, which can significantly impact filtering performance. This
patch also adds an inode-number-based rule hash to mitigate this
situation.
The patch is relative to the audit git tree:
http://kernel.org/git/?p=linux/kernel/git/viro/audit-current.git;a=summary
and uses the inotify kernel API:
http://lkml.org/lkml/2006/6/1/145
Signed-off-by: Amy Griffis <amy.griffis@hp.com>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
2006-04-07 14:55:56 -06:00
|
|
|
}
|
|
|
|
|
|
|
|
/* The audit_filter_mutex must not be held during inotify calls because
|
|
|
|
* we hold it during inotify event callback processing. If an existing
|
|
|
|
* inotify watch is found, inotify_find_watch() grabs a reference before
|
|
|
|
* returning.
|
|
|
|
*/
|
|
|
|
mutex_unlock(&audit_filter_mutex);
|
|
|
|
|
2008-02-14 20:34:32 -07:00
|
|
|
if (inotify_find_watch(audit_ih, ndp->path.dentry->d_inode,
|
|
|
|
&i_watch) < 0) {
|
[PATCH] audit: path-based rules
In this implementation, audit registers inotify watches on the parent
directories of paths specified in audit rules. When audit's inotify
event handler is called, it updates any affected rules based on the
filesystem event. If the parent directory is renamed, removed, or its
filesystem is unmounted, audit removes all rules referencing that
inotify watch.
To keep things simple, this implementation limits location-based
auditing to the directory entries in an existing directory. Given
a path-based rule for /foo/bar/passwd, the following table applies:
passwd modified -- audit event logged
passwd replaced -- audit event logged, rules list updated
bar renamed -- rule removed
foo renamed -- untracked, meaning that the rule now applies to
the new location
Audit users typically want to have many rules referencing filesystem
objects, which can significantly impact filtering performance. This
patch also adds an inode-number-based rule hash to mitigate this
situation.
The patch is relative to the audit git tree:
http://kernel.org/git/?p=linux/kernel/git/viro/audit-current.git;a=summary
and uses the inotify kernel API:
http://lkml.org/lkml/2006/6/1/145
Signed-off-by: Amy Griffis <amy.griffis@hp.com>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
2006-04-07 14:55:56 -06:00
|
|
|
parent = audit_init_parent(ndp);
|
|
|
|
if (IS_ERR(parent)) {
|
|
|
|
/* caller expects mutex locked */
|
|
|
|
mutex_lock(&audit_filter_mutex);
|
|
|
|
return PTR_ERR(parent);
|
|
|
|
}
|
|
|
|
} else
|
|
|
|
parent = container_of(i_watch, struct audit_parent, wdata);
|
|
|
|
|
|
|
|
mutex_lock(&audit_filter_mutex);
|
|
|
|
|
|
|
|
/* parent was moved before we took audit_filter_mutex */
|
|
|
|
if (parent->flags & AUDIT_PARENT_INVALID)
|
|
|
|
ret = -ENOENT;
|
|
|
|
else
|
|
|
|
audit_add_to_parent(krule, parent);
|
|
|
|
|
|
|
|
/* match get in audit_init_parent or inotify_find_watch */
|
|
|
|
put_inotify_watch(&parent->wdata);
|
|
|
|
return ret;
|
|
|
|
}
|
|
|
|
|
|
|
|
/* Add rule to given filterlist if not a duplicate. */
|
2006-02-07 10:05:27 -07:00
|
|
|
static inline int audit_add_rule(struct audit_entry *entry,
|
[PATCH] audit: path-based rules
In this implementation, audit registers inotify watches on the parent
directories of paths specified in audit rules. When audit's inotify
event handler is called, it updates any affected rules based on the
filesystem event. If the parent directory is renamed, removed, or its
filesystem is unmounted, audit removes all rules referencing that
inotify watch.
To keep things simple, this implementation limits location-based
auditing to the directory entries in an existing directory. Given
a path-based rule for /foo/bar/passwd, the following table applies:
passwd modified -- audit event logged
passwd replaced -- audit event logged, rules list updated
bar renamed -- rule removed
foo renamed -- untracked, meaning that the rule now applies to
the new location
Audit users typically want to have many rules referencing filesystem
objects, which can significantly impact filtering performance. This
patch also adds an inode-number-based rule hash to mitigate this
situation.
The patch is relative to the audit git tree:
http://kernel.org/git/?p=linux/kernel/git/viro/audit-current.git;a=summary
and uses the inotify kernel API:
http://lkml.org/lkml/2006/6/1/145
Signed-off-by: Amy Griffis <amy.griffis@hp.com>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
2006-04-07 14:55:56 -06:00
|
|
|
struct list_head *list)
|
2005-12-15 11:33:52 -07:00
|
|
|
{
|
2006-02-07 10:05:27 -07:00
|
|
|
struct audit_entry *e;
|
[PATCH] audit: path-based rules
In this implementation, audit registers inotify watches on the parent
directories of paths specified in audit rules. When audit's inotify
event handler is called, it updates any affected rules based on the
filesystem event. If the parent directory is renamed, removed, or its
filesystem is unmounted, audit removes all rules referencing that
inotify watch.
To keep things simple, this implementation limits location-based
auditing to the directory entries in an existing directory. Given
a path-based rule for /foo/bar/passwd, the following table applies:
passwd modified -- audit event logged
passwd replaced -- audit event logged, rules list updated
bar renamed -- rule removed
foo renamed -- untracked, meaning that the rule now applies to
the new location
Audit users typically want to have many rules referencing filesystem
objects, which can significantly impact filtering performance. This
patch also adds an inode-number-based rule hash to mitigate this
situation.
The patch is relative to the audit git tree:
http://kernel.org/git/?p=linux/kernel/git/viro/audit-current.git;a=summary
and uses the inotify kernel API:
http://lkml.org/lkml/2006/6/1/145
Signed-off-by: Amy Griffis <amy.griffis@hp.com>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
2006-04-07 14:55:56 -06:00
|
|
|
struct audit_field *inode_f = entry->rule.inode_f;
|
|
|
|
struct audit_watch *watch = entry->rule.watch;
|
[PATCH] audit: watching subtrees
New kind of audit rule predicates: "object is visible in given subtree".
The part that can be sanely implemented, that is. Limitations:
* if you have hardlink from outside of tree, you'd better watch
it too (or just watch the object itself, obviously)
* if you mount something under a watched tree, tell audit
that new chunk should be added to watched subtrees
* if you umount something in a watched tree and it's still mounted
elsewhere, you will get matches on events happening there. New command
tells audit to recalculate the trees, trimming such sources of false
positives.
Note that it's _not_ about path - if something mounted in several places
(multiple mount, bindings, different namespaces, etc.), the match does
_not_ depend on which one we are using for access.
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
2007-07-22 06:04:18 -06:00
|
|
|
struct audit_tree *tree = entry->rule.tree;
|
2007-07-16 19:25:01 -06:00
|
|
|
struct nameidata *ndp = NULL, *ndw = NULL;
|
|
|
|
int h, err;
|
2006-07-10 06:29:24 -06:00
|
|
|
#ifdef CONFIG_AUDITSYSCALL
|
|
|
|
int dont_count = 0;
|
|
|
|
|
|
|
|
/* If either of these, don't count towards total */
|
|
|
|
if (entry->rule.listnr == AUDIT_FILTER_USER ||
|
|
|
|
entry->rule.listnr == AUDIT_FILTER_TYPE)
|
|
|
|
dont_count = 1;
|
|
|
|
#endif
|
[PATCH] audit: path-based rules
In this implementation, audit registers inotify watches on the parent
directories of paths specified in audit rules. When audit's inotify
event handler is called, it updates any affected rules based on the
filesystem event. If the parent directory is renamed, removed, or its
filesystem is unmounted, audit removes all rules referencing that
inotify watch.
To keep things simple, this implementation limits location-based
auditing to the directory entries in an existing directory. Given
a path-based rule for /foo/bar/passwd, the following table applies:
passwd modified -- audit event logged
passwd replaced -- audit event logged, rules list updated
bar renamed -- rule removed
foo renamed -- untracked, meaning that the rule now applies to
the new location
Audit users typically want to have many rules referencing filesystem
objects, which can significantly impact filtering performance. This
patch also adds an inode-number-based rule hash to mitigate this
situation.
The patch is relative to the audit git tree:
http://kernel.org/git/?p=linux/kernel/git/viro/audit-current.git;a=summary
and uses the inotify kernel API:
http://lkml.org/lkml/2006/6/1/145
Signed-off-by: Amy Griffis <amy.griffis@hp.com>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
2006-04-07 14:55:56 -06:00
|
|
|
|
|
|
|
if (inode_f) {
|
|
|
|
h = audit_hash_ino(inode_f->val);
|
|
|
|
list = &audit_inode_hash[h];
|
|
|
|
}
|
|
|
|
|
|
|
|
mutex_lock(&audit_filter_mutex);
|
|
|
|
e = audit_find_rule(entry, list);
|
|
|
|
mutex_unlock(&audit_filter_mutex);
|
|
|
|
if (e) {
|
|
|
|
err = -EEXIST;
|
[PATCH] audit: watching subtrees
New kind of audit rule predicates: "object is visible in given subtree".
The part that can be sanely implemented, that is. Limitations:
* if you have hardlink from outside of tree, you'd better watch
it too (or just watch the object itself, obviously)
* if you mount something under a watched tree, tell audit
that new chunk should be added to watched subtrees
* if you umount something in a watched tree and it's still mounted
elsewhere, you will get matches on events happening there. New command
tells audit to recalculate the trees, trimming such sources of false
positives.
Note that it's _not_ about path - if something mounted in several places
(multiple mount, bindings, different namespaces, etc.), the match does
_not_ depend on which one we are using for access.
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
2007-07-22 06:04:18 -06:00
|
|
|
/* normally audit_add_tree_rule() will free it on failure */
|
|
|
|
if (tree)
|
|
|
|
audit_put_tree(tree);
|
[PATCH] audit: path-based rules
In this implementation, audit registers inotify watches on the parent
directories of paths specified in audit rules. When audit's inotify
event handler is called, it updates any affected rules based on the
filesystem event. If the parent directory is renamed, removed, or its
filesystem is unmounted, audit removes all rules referencing that
inotify watch.
To keep things simple, this implementation limits location-based
auditing to the directory entries in an existing directory. Given
a path-based rule for /foo/bar/passwd, the following table applies:
passwd modified -- audit event logged
passwd replaced -- audit event logged, rules list updated
bar renamed -- rule removed
foo renamed -- untracked, meaning that the rule now applies to
the new location
Audit users typically want to have many rules referencing filesystem
objects, which can significantly impact filtering performance. This
patch also adds an inode-number-based rule hash to mitigate this
situation.
The patch is relative to the audit git tree:
http://kernel.org/git/?p=linux/kernel/git/viro/audit-current.git;a=summary
and uses the inotify kernel API:
http://lkml.org/lkml/2006/6/1/145
Signed-off-by: Amy Griffis <amy.griffis@hp.com>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
2006-04-07 14:55:56 -06:00
|
|
|
goto error;
|
|
|
|
}
|
2005-12-15 11:33:52 -07:00
|
|
|
|
[PATCH] audit: path-based rules
In this implementation, audit registers inotify watches on the parent
directories of paths specified in audit rules. When audit's inotify
event handler is called, it updates any affected rules based on the
filesystem event. If the parent directory is renamed, removed, or its
filesystem is unmounted, audit removes all rules referencing that
inotify watch.
To keep things simple, this implementation limits location-based
auditing to the directory entries in an existing directory. Given
a path-based rule for /foo/bar/passwd, the following table applies:
passwd modified -- audit event logged
passwd replaced -- audit event logged, rules list updated
bar renamed -- rule removed
foo renamed -- untracked, meaning that the rule now applies to
the new location
Audit users typically want to have many rules referencing filesystem
objects, which can significantly impact filtering performance. This
patch also adds an inode-number-based rule hash to mitigate this
situation.
The patch is relative to the audit git tree:
http://kernel.org/git/?p=linux/kernel/git/viro/audit-current.git;a=summary
and uses the inotify kernel API:
http://lkml.org/lkml/2006/6/1/145
Signed-off-by: Amy Griffis <amy.griffis@hp.com>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
2006-04-07 14:55:56 -06:00
|
|
|
/* Avoid calling path_lookup under audit_filter_mutex. */
|
|
|
|
if (watch) {
|
|
|
|
err = audit_get_nd(watch->path, &ndp, &ndw);
|
|
|
|
if (err)
|
|
|
|
goto error;
|
|
|
|
}
|
|
|
|
|
|
|
|
mutex_lock(&audit_filter_mutex);
|
|
|
|
if (watch) {
|
|
|
|
/* audit_filter_mutex is dropped and re-taken during this call */
|
|
|
|
err = audit_add_watch(&entry->rule, ndp, ndw);
|
|
|
|
if (err) {
|
|
|
|
mutex_unlock(&audit_filter_mutex);
|
|
|
|
goto error;
|
|
|
|
}
|
|
|
|
h = audit_hash_ino((u32)watch->ino);
|
|
|
|
list = &audit_inode_hash[h];
|
2005-12-15 11:33:52 -07:00
|
|
|
}
|
[PATCH] audit: watching subtrees
New kind of audit rule predicates: "object is visible in given subtree".
The part that can be sanely implemented, that is. Limitations:
* if you have hardlink from outside of tree, you'd better watch
it too (or just watch the object itself, obviously)
* if you mount something under a watched tree, tell audit
that new chunk should be added to watched subtrees
* if you umount something in a watched tree and it's still mounted
elsewhere, you will get matches on events happening there. New command
tells audit to recalculate the trees, trimming such sources of false
positives.
Note that it's _not_ about path - if something mounted in several places
(multiple mount, bindings, different namespaces, etc.), the match does
_not_ depend on which one we are using for access.
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
2007-07-22 06:04:18 -06:00
|
|
|
if (tree) {
|
|
|
|
err = audit_add_tree_rule(&entry->rule);
|
|
|
|
if (err) {
|
|
|
|
mutex_unlock(&audit_filter_mutex);
|
|
|
|
goto error;
|
|
|
|
}
|
|
|
|
}
|
2005-12-15 11:33:52 -07:00
|
|
|
|
|
|
|
if (entry->rule.flags & AUDIT_FILTER_PREPEND) {
|
|
|
|
list_add_rcu(&entry->list, list);
|
2006-06-02 11:16:01 -06:00
|
|
|
entry->rule.flags &= ~AUDIT_FILTER_PREPEND;
|
2005-12-15 11:33:52 -07:00
|
|
|
} else {
|
|
|
|
list_add_tail_rcu(&entry->list, list);
|
|
|
|
}
|
2006-07-10 06:29:24 -06:00
|
|
|
#ifdef CONFIG_AUDITSYSCALL
|
|
|
|
if (!dont_count)
|
|
|
|
audit_n_rules++;
|
2007-03-29 16:01:04 -06:00
|
|
|
|
|
|
|
if (!audit_match_signal(entry))
|
|
|
|
audit_signals++;
|
2006-07-10 06:29:24 -06:00
|
|
|
#endif
|
[PATCH] audit: path-based rules
In this implementation, audit registers inotify watches on the parent
directories of paths specified in audit rules. When audit's inotify
event handler is called, it updates any affected rules based on the
filesystem event. If the parent directory is renamed, removed, or its
filesystem is unmounted, audit removes all rules referencing that
inotify watch.
To keep things simple, this implementation limits location-based
auditing to the directory entries in an existing directory. Given
a path-based rule for /foo/bar/passwd, the following table applies:
passwd modified -- audit event logged
passwd replaced -- audit event logged, rules list updated
bar renamed -- rule removed
foo renamed -- untracked, meaning that the rule now applies to
the new location
Audit users typically want to have many rules referencing filesystem
objects, which can significantly impact filtering performance. This
patch also adds an inode-number-based rule hash to mitigate this
situation.
The patch is relative to the audit git tree:
http://kernel.org/git/?p=linux/kernel/git/viro/audit-current.git;a=summary
and uses the inotify kernel API:
http://lkml.org/lkml/2006/6/1/145
Signed-off-by: Amy Griffis <amy.griffis@hp.com>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
2006-04-07 14:55:56 -06:00
|
|
|
mutex_unlock(&audit_filter_mutex);
|
2005-12-15 11:33:52 -07:00
|
|
|
|
2007-07-16 19:25:01 -06:00
|
|
|
audit_put_nd(ndp, ndw); /* NULL args OK */
|
[PATCH] audit: path-based rules
In this implementation, audit registers inotify watches on the parent
directories of paths specified in audit rules. When audit's inotify
event handler is called, it updates any affected rules based on the
filesystem event. If the parent directory is renamed, removed, or its
filesystem is unmounted, audit removes all rules referencing that
inotify watch.
To keep things simple, this implementation limits location-based
auditing to the directory entries in an existing directory. Given
a path-based rule for /foo/bar/passwd, the following table applies:
passwd modified -- audit event logged
passwd replaced -- audit event logged, rules list updated
bar renamed -- rule removed
foo renamed -- untracked, meaning that the rule now applies to
the new location
Audit users typically want to have many rules referencing filesystem
objects, which can significantly impact filtering performance. This
patch also adds an inode-number-based rule hash to mitigate this
situation.
The patch is relative to the audit git tree:
http://kernel.org/git/?p=linux/kernel/git/viro/audit-current.git;a=summary
and uses the inotify kernel API:
http://lkml.org/lkml/2006/6/1/145
Signed-off-by: Amy Griffis <amy.griffis@hp.com>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
2006-04-07 14:55:56 -06:00
|
|
|
return 0;
|
|
|
|
|
|
|
|
error:
|
2007-07-16 19:25:01 -06:00
|
|
|
audit_put_nd(ndp, ndw); /* NULL args OK */
|
[PATCH] audit: path-based rules
In this implementation, audit registers inotify watches on the parent
directories of paths specified in audit rules. When audit's inotify
event handler is called, it updates any affected rules based on the
filesystem event. If the parent directory is renamed, removed, or its
filesystem is unmounted, audit removes all rules referencing that
inotify watch.
To keep things simple, this implementation limits location-based
auditing to the directory entries in an existing directory. Given
a path-based rule for /foo/bar/passwd, the following table applies:
passwd modified -- audit event logged
passwd replaced -- audit event logged, rules list updated
bar renamed -- rule removed
foo renamed -- untracked, meaning that the rule now applies to
the new location
Audit users typically want to have many rules referencing filesystem
objects, which can significantly impact filtering performance. This
patch also adds an inode-number-based rule hash to mitigate this
situation.
The patch is relative to the audit git tree:
http://kernel.org/git/?p=linux/kernel/git/viro/audit-current.git;a=summary
and uses the inotify kernel API:
http://lkml.org/lkml/2006/6/1/145
Signed-off-by: Amy Griffis <amy.griffis@hp.com>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
2006-04-07 14:55:56 -06:00
|
|
|
if (watch)
|
|
|
|
audit_put_watch(watch); /* tmp watch, matches initial get */
|
|
|
|
return err;
|
2005-12-15 11:33:52 -07:00
|
|
|
}
|
|
|
|
|
[PATCH] audit: path-based rules
In this implementation, audit registers inotify watches on the parent
directories of paths specified in audit rules. When audit's inotify
event handler is called, it updates any affected rules based on the
filesystem event. If the parent directory is renamed, removed, or its
filesystem is unmounted, audit removes all rules referencing that
inotify watch.
To keep things simple, this implementation limits location-based
auditing to the directory entries in an existing directory. Given
a path-based rule for /foo/bar/passwd, the following table applies:
passwd modified -- audit event logged
passwd replaced -- audit event logged, rules list updated
bar renamed -- rule removed
foo renamed -- untracked, meaning that the rule now applies to
the new location
Audit users typically want to have many rules referencing filesystem
objects, which can significantly impact filtering performance. This
patch also adds an inode-number-based rule hash to mitigate this
situation.
The patch is relative to the audit git tree:
http://kernel.org/git/?p=linux/kernel/git/viro/audit-current.git;a=summary
and uses the inotify kernel API:
http://lkml.org/lkml/2006/6/1/145
Signed-off-by: Amy Griffis <amy.griffis@hp.com>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
2006-04-07 14:55:56 -06:00
|
|
|
/* Remove an existing rule from filterlist. */
|
2006-02-07 10:05:27 -07:00
|
|
|
static inline int audit_del_rule(struct audit_entry *entry,
|
2005-12-15 11:33:52 -07:00
|
|
|
struct list_head *list)
|
|
|
|
{
|
|
|
|
struct audit_entry *e;
|
[PATCH] audit: path-based rules
In this implementation, audit registers inotify watches on the parent
directories of paths specified in audit rules. When audit's inotify
event handler is called, it updates any affected rules based on the
filesystem event. If the parent directory is renamed, removed, or its
filesystem is unmounted, audit removes all rules referencing that
inotify watch.
To keep things simple, this implementation limits location-based
auditing to the directory entries in an existing directory. Given
a path-based rule for /foo/bar/passwd, the following table applies:
passwd modified -- audit event logged
passwd replaced -- audit event logged, rules list updated
bar renamed -- rule removed
foo renamed -- untracked, meaning that the rule now applies to
the new location
Audit users typically want to have many rules referencing filesystem
objects, which can significantly impact filtering performance. This
patch also adds an inode-number-based rule hash to mitigate this
situation.
The patch is relative to the audit git tree:
http://kernel.org/git/?p=linux/kernel/git/viro/audit-current.git;a=summary
and uses the inotify kernel API:
http://lkml.org/lkml/2006/6/1/145
Signed-off-by: Amy Griffis <amy.griffis@hp.com>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
2006-04-07 14:55:56 -06:00
|
|
|
struct audit_field *inode_f = entry->rule.inode_f;
|
|
|
|
struct audit_watch *watch, *tmp_watch = entry->rule.watch;
|
[PATCH] audit: watching subtrees
New kind of audit rule predicates: "object is visible in given subtree".
The part that can be sanely implemented, that is. Limitations:
* if you have hardlink from outside of tree, you'd better watch
it too (or just watch the object itself, obviously)
* if you mount something under a watched tree, tell audit
that new chunk should be added to watched subtrees
* if you umount something in a watched tree and it's still mounted
elsewhere, you will get matches on events happening there. New command
tells audit to recalculate the trees, trimming such sources of false
positives.
Note that it's _not_ about path - if something mounted in several places
(multiple mount, bindings, different namespaces, etc.), the match does
_not_ depend on which one we are using for access.
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
2007-07-22 06:04:18 -06:00
|
|
|
struct audit_tree *tree = entry->rule.tree;
|
[PATCH] audit: path-based rules
In this implementation, audit registers inotify watches on the parent
directories of paths specified in audit rules. When audit's inotify
event handler is called, it updates any affected rules based on the
filesystem event. If the parent directory is renamed, removed, or its
filesystem is unmounted, audit removes all rules referencing that
inotify watch.
To keep things simple, this implementation limits location-based
auditing to the directory entries in an existing directory. Given
a path-based rule for /foo/bar/passwd, the following table applies:
passwd modified -- audit event logged
passwd replaced -- audit event logged, rules list updated
bar renamed -- rule removed
foo renamed -- untracked, meaning that the rule now applies to
the new location
Audit users typically want to have many rules referencing filesystem
objects, which can significantly impact filtering performance. This
patch also adds an inode-number-based rule hash to mitigate this
situation.
The patch is relative to the audit git tree:
http://kernel.org/git/?p=linux/kernel/git/viro/audit-current.git;a=summary
and uses the inotify kernel API:
http://lkml.org/lkml/2006/6/1/145
Signed-off-by: Amy Griffis <amy.griffis@hp.com>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
2006-04-07 14:55:56 -06:00
|
|
|
LIST_HEAD(inotify_list);
|
|
|
|
int h, ret = 0;
|
2006-07-10 06:29:24 -06:00
|
|
|
#ifdef CONFIG_AUDITSYSCALL
|
|
|
|
int dont_count = 0;
|
|
|
|
|
|
|
|
/* If either of these, don't count towards total */
|
|
|
|
if (entry->rule.listnr == AUDIT_FILTER_USER ||
|
|
|
|
entry->rule.listnr == AUDIT_FILTER_TYPE)
|
|
|
|
dont_count = 1;
|
|
|
|
#endif
|
[PATCH] audit: path-based rules
In this implementation, audit registers inotify watches on the parent
directories of paths specified in audit rules. When audit's inotify
event handler is called, it updates any affected rules based on the
filesystem event. If the parent directory is renamed, removed, or its
filesystem is unmounted, audit removes all rules referencing that
inotify watch.
To keep things simple, this implementation limits location-based
auditing to the directory entries in an existing directory. Given
a path-based rule for /foo/bar/passwd, the following table applies:
passwd modified -- audit event logged
passwd replaced -- audit event logged, rules list updated
bar renamed -- rule removed
foo renamed -- untracked, meaning that the rule now applies to
the new location
Audit users typically want to have many rules referencing filesystem
objects, which can significantly impact filtering performance. This
patch also adds an inode-number-based rule hash to mitigate this
situation.
The patch is relative to the audit git tree:
http://kernel.org/git/?p=linux/kernel/git/viro/audit-current.git;a=summary
and uses the inotify kernel API:
http://lkml.org/lkml/2006/6/1/145
Signed-off-by: Amy Griffis <amy.griffis@hp.com>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
2006-04-07 14:55:56 -06:00
|
|
|
|
|
|
|
if (inode_f) {
|
|
|
|
h = audit_hash_ino(inode_f->val);
|
|
|
|
list = &audit_inode_hash[h];
|
|
|
|
}
|
2005-12-15 11:33:52 -07:00
|
|
|
|
[PATCH] audit: path-based rules
In this implementation, audit registers inotify watches on the parent
directories of paths specified in audit rules. When audit's inotify
event handler is called, it updates any affected rules based on the
filesystem event. If the parent directory is renamed, removed, or its
filesystem is unmounted, audit removes all rules referencing that
inotify watch.
To keep things simple, this implementation limits location-based
auditing to the directory entries in an existing directory. Given
a path-based rule for /foo/bar/passwd, the following table applies:
passwd modified -- audit event logged
passwd replaced -- audit event logged, rules list updated
bar renamed -- rule removed
foo renamed -- untracked, meaning that the rule now applies to
the new location
Audit users typically want to have many rules referencing filesystem
objects, which can significantly impact filtering performance. This
patch also adds an inode-number-based rule hash to mitigate this
situation.
The patch is relative to the audit git tree:
http://kernel.org/git/?p=linux/kernel/git/viro/audit-current.git;a=summary
and uses the inotify kernel API:
http://lkml.org/lkml/2006/6/1/145
Signed-off-by: Amy Griffis <amy.griffis@hp.com>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
2006-04-07 14:55:56 -06:00
|
|
|
mutex_lock(&audit_filter_mutex);
|
|
|
|
e = audit_find_rule(entry, list);
|
|
|
|
if (!e) {
|
|
|
|
mutex_unlock(&audit_filter_mutex);
|
|
|
|
ret = -ENOENT;
|
|
|
|
goto out;
|
|
|
|
}
|
|
|
|
|
|
|
|
watch = e->rule.watch;
|
|
|
|
if (watch) {
|
|
|
|
struct audit_parent *parent = watch->parent;
|
|
|
|
|
|
|
|
list_del(&e->rule.rlist);
|
|
|
|
|
|
|
|
if (list_empty(&watch->rules)) {
|
|
|
|
audit_remove_watch(watch);
|
|
|
|
|
|
|
|
if (list_empty(&parent->watches)) {
|
|
|
|
/* Put parent on the inotify un-registration
|
|
|
|
* list. Grab a reference before releasing
|
|
|
|
* audit_filter_mutex, to be released in
|
|
|
|
* audit_inotify_unregister(). */
|
|
|
|
list_add(&parent->ilist, &inotify_list);
|
|
|
|
get_inotify_watch(&parent->wdata);
|
|
|
|
}
|
2005-12-15 11:33:52 -07:00
|
|
|
}
|
|
|
|
}
|
[PATCH] audit: path-based rules
In this implementation, audit registers inotify watches on the parent
directories of paths specified in audit rules. When audit's inotify
event handler is called, it updates any affected rules based on the
filesystem event. If the parent directory is renamed, removed, or its
filesystem is unmounted, audit removes all rules referencing that
inotify watch.
To keep things simple, this implementation limits location-based
auditing to the directory entries in an existing directory. Given
a path-based rule for /foo/bar/passwd, the following table applies:
passwd modified -- audit event logged
passwd replaced -- audit event logged, rules list updated
bar renamed -- rule removed
foo renamed -- untracked, meaning that the rule now applies to
the new location
Audit users typically want to have many rules referencing filesystem
objects, which can significantly impact filtering performance. This
patch also adds an inode-number-based rule hash to mitigate this
situation.
The patch is relative to the audit git tree:
http://kernel.org/git/?p=linux/kernel/git/viro/audit-current.git;a=summary
and uses the inotify kernel API:
http://lkml.org/lkml/2006/6/1/145
Signed-off-by: Amy Griffis <amy.griffis@hp.com>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
2006-04-07 14:55:56 -06:00
|
|
|
|
[PATCH] audit: watching subtrees
New kind of audit rule predicates: "object is visible in given subtree".
The part that can be sanely implemented, that is. Limitations:
* if you have hardlink from outside of tree, you'd better watch
it too (or just watch the object itself, obviously)
* if you mount something under a watched tree, tell audit
that new chunk should be added to watched subtrees
* if you umount something in a watched tree and it's still mounted
elsewhere, you will get matches on events happening there. New command
tells audit to recalculate the trees, trimming such sources of false
positives.
Note that it's _not_ about path - if something mounted in several places
(multiple mount, bindings, different namespaces, etc.), the match does
_not_ depend on which one we are using for access.
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
2007-07-22 06:04:18 -06:00
|
|
|
if (e->rule.tree)
|
|
|
|
audit_remove_tree_rule(&e->rule);
|
|
|
|
|
[PATCH] audit: path-based rules
In this implementation, audit registers inotify watches on the parent
directories of paths specified in audit rules. When audit's inotify
event handler is called, it updates any affected rules based on the
filesystem event. If the parent directory is renamed, removed, or its
filesystem is unmounted, audit removes all rules referencing that
inotify watch.
To keep things simple, this implementation limits location-based
auditing to the directory entries in an existing directory. Given
a path-based rule for /foo/bar/passwd, the following table applies:
passwd modified -- audit event logged
passwd replaced -- audit event logged, rules list updated
bar renamed -- rule removed
foo renamed -- untracked, meaning that the rule now applies to
the new location
Audit users typically want to have many rules referencing filesystem
objects, which can significantly impact filtering performance. This
patch also adds an inode-number-based rule hash to mitigate this
situation.
The patch is relative to the audit git tree:
http://kernel.org/git/?p=linux/kernel/git/viro/audit-current.git;a=summary
and uses the inotify kernel API:
http://lkml.org/lkml/2006/6/1/145
Signed-off-by: Amy Griffis <amy.griffis@hp.com>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
2006-04-07 14:55:56 -06:00
|
|
|
list_del_rcu(&e->list);
|
|
|
|
call_rcu(&e->rcu, audit_free_rule_rcu);
|
|
|
|
|
2006-07-10 06:29:24 -06:00
|
|
|
#ifdef CONFIG_AUDITSYSCALL
|
|
|
|
if (!dont_count)
|
|
|
|
audit_n_rules--;
|
2007-03-29 16:01:04 -06:00
|
|
|
|
|
|
|
if (!audit_match_signal(entry))
|
|
|
|
audit_signals--;
|
2006-07-10 06:29:24 -06:00
|
|
|
#endif
|
[PATCH] audit: path-based rules
In this implementation, audit registers inotify watches on the parent
directories of paths specified in audit rules. When audit's inotify
event handler is called, it updates any affected rules based on the
filesystem event. If the parent directory is renamed, removed, or its
filesystem is unmounted, audit removes all rules referencing that
inotify watch.
To keep things simple, this implementation limits location-based
auditing to the directory entries in an existing directory. Given
a path-based rule for /foo/bar/passwd, the following table applies:
passwd modified -- audit event logged
passwd replaced -- audit event logged, rules list updated
bar renamed -- rule removed
foo renamed -- untracked, meaning that the rule now applies to
the new location
Audit users typically want to have many rules referencing filesystem
objects, which can significantly impact filtering performance. This
patch also adds an inode-number-based rule hash to mitigate this
situation.
The patch is relative to the audit git tree:
http://kernel.org/git/?p=linux/kernel/git/viro/audit-current.git;a=summary
and uses the inotify kernel API:
http://lkml.org/lkml/2006/6/1/145
Signed-off-by: Amy Griffis <amy.griffis@hp.com>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
2006-04-07 14:55:56 -06:00
|
|
|
mutex_unlock(&audit_filter_mutex);
|
|
|
|
|
|
|
|
if (!list_empty(&inotify_list))
|
|
|
|
audit_inotify_unregister(&inotify_list);
|
|
|
|
|
|
|
|
out:
|
|
|
|
if (tmp_watch)
|
|
|
|
audit_put_watch(tmp_watch); /* match initial get */
|
[PATCH] audit: watching subtrees
New kind of audit rule predicates: "object is visible in given subtree".
The part that can be sanely implemented, that is. Limitations:
* if you have hardlink from outside of tree, you'd better watch
it too (or just watch the object itself, obviously)
* if you mount something under a watched tree, tell audit
that new chunk should be added to watched subtrees
* if you umount something in a watched tree and it's still mounted
elsewhere, you will get matches on events happening there. New command
tells audit to recalculate the trees, trimming such sources of false
positives.
Note that it's _not_ about path - if something mounted in several places
(multiple mount, bindings, different namespaces, etc.), the match does
_not_ depend on which one we are using for access.
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
2007-07-22 06:04:18 -06:00
|
|
|
if (tree)
|
|
|
|
audit_put_tree(tree); /* that's the temporary one */
|
[PATCH] audit: path-based rules
In this implementation, audit registers inotify watches on the parent
directories of paths specified in audit rules. When audit's inotify
event handler is called, it updates any affected rules based on the
filesystem event. If the parent directory is renamed, removed, or its
filesystem is unmounted, audit removes all rules referencing that
inotify watch.
To keep things simple, this implementation limits location-based
auditing to the directory entries in an existing directory. Given
a path-based rule for /foo/bar/passwd, the following table applies:
passwd modified -- audit event logged
passwd replaced -- audit event logged, rules list updated
bar renamed -- rule removed
foo renamed -- untracked, meaning that the rule now applies to
the new location
Audit users typically want to have many rules referencing filesystem
objects, which can significantly impact filtering performance. This
patch also adds an inode-number-based rule hash to mitigate this
situation.
The patch is relative to the audit git tree:
http://kernel.org/git/?p=linux/kernel/git/viro/audit-current.git;a=summary
and uses the inotify kernel API:
http://lkml.org/lkml/2006/6/1/145
Signed-off-by: Amy Griffis <amy.griffis@hp.com>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
2006-04-07 14:55:56 -06:00
|
|
|
|
|
|
|
return ret;
|
2005-12-15 11:33:52 -07:00
|
|
|
}
|
|
|
|
|
2006-02-07 10:05:27 -07:00
|
|
|
/* List rules using struct audit_rule. Exists for backward
|
|
|
|
* compatibility with userspace. */
|
2006-05-21 23:09:24 -06:00
|
|
|
static void audit_list(int pid, int seq, struct sk_buff_head *q)
|
2005-12-15 11:33:52 -07:00
|
|
|
{
|
2006-05-21 23:09:24 -06:00
|
|
|
struct sk_buff *skb;
|
2005-12-15 11:33:52 -07:00
|
|
|
struct audit_entry *entry;
|
|
|
|
int i;
|
|
|
|
|
[PATCH] audit: path-based rules
In this implementation, audit registers inotify watches on the parent
directories of paths specified in audit rules. When audit's inotify
event handler is called, it updates any affected rules based on the
filesystem event. If the parent directory is renamed, removed, or its
filesystem is unmounted, audit removes all rules referencing that
inotify watch.
To keep things simple, this implementation limits location-based
auditing to the directory entries in an existing directory. Given
a path-based rule for /foo/bar/passwd, the following table applies:
passwd modified -- audit event logged
passwd replaced -- audit event logged, rules list updated
bar renamed -- rule removed
foo renamed -- untracked, meaning that the rule now applies to
the new location
Audit users typically want to have many rules referencing filesystem
objects, which can significantly impact filtering performance. This
patch also adds an inode-number-based rule hash to mitigate this
situation.
The patch is relative to the audit git tree:
http://kernel.org/git/?p=linux/kernel/git/viro/audit-current.git;a=summary
and uses the inotify kernel API:
http://lkml.org/lkml/2006/6/1/145
Signed-off-by: Amy Griffis <amy.griffis@hp.com>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
2006-04-07 14:55:56 -06:00
|
|
|
/* This is a blocking read, so use audit_filter_mutex instead of rcu
|
|
|
|
* iterator to sync with list writers. */
|
2005-12-15 11:33:52 -07:00
|
|
|
for (i=0; i<AUDIT_NR_FILTERS; i++) {
|
2006-02-07 10:05:27 -07:00
|
|
|
list_for_each_entry(entry, &audit_filter_list[i], list) {
|
|
|
|
struct audit_rule *rule;
|
|
|
|
|
|
|
|
rule = audit_krule_to_rule(&entry->rule);
|
|
|
|
if (unlikely(!rule))
|
|
|
|
break;
|
2006-05-21 23:09:24 -06:00
|
|
|
skb = audit_make_reply(pid, seq, AUDIT_LIST, 0, 1,
|
2006-02-07 10:05:27 -07:00
|
|
|
rule, sizeof(*rule));
|
2006-05-21 23:09:24 -06:00
|
|
|
if (skb)
|
|
|
|
skb_queue_tail(q, skb);
|
2006-02-07 10:05:27 -07:00
|
|
|
kfree(rule);
|
|
|
|
}
|
2005-12-15 11:33:52 -07:00
|
|
|
}
|
[PATCH] audit: path-based rules
In this implementation, audit registers inotify watches on the parent
directories of paths specified in audit rules. When audit's inotify
event handler is called, it updates any affected rules based on the
filesystem event. If the parent directory is renamed, removed, or its
filesystem is unmounted, audit removes all rules referencing that
inotify watch.
To keep things simple, this implementation limits location-based
auditing to the directory entries in an existing directory. Given
a path-based rule for /foo/bar/passwd, the following table applies:
passwd modified -- audit event logged
passwd replaced -- audit event logged, rules list updated
bar renamed -- rule removed
foo renamed -- untracked, meaning that the rule now applies to
the new location
Audit users typically want to have many rules referencing filesystem
objects, which can significantly impact filtering performance. This
patch also adds an inode-number-based rule hash to mitigate this
situation.
The patch is relative to the audit git tree:
http://kernel.org/git/?p=linux/kernel/git/viro/audit-current.git;a=summary
and uses the inotify kernel API:
http://lkml.org/lkml/2006/6/1/145
Signed-off-by: Amy Griffis <amy.griffis@hp.com>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
2006-04-07 14:55:56 -06:00
|
|
|
for (i = 0; i < AUDIT_INODE_BUCKETS; i++) {
|
|
|
|
list_for_each_entry(entry, &audit_inode_hash[i], list) {
|
|
|
|
struct audit_rule *rule;
|
|
|
|
|
|
|
|
rule = audit_krule_to_rule(&entry->rule);
|
|
|
|
if (unlikely(!rule))
|
|
|
|
break;
|
|
|
|
skb = audit_make_reply(pid, seq, AUDIT_LIST, 0, 1,
|
|
|
|
rule, sizeof(*rule));
|
|
|
|
if (skb)
|
|
|
|
skb_queue_tail(q, skb);
|
|
|
|
kfree(rule);
|
|
|
|
}
|
|
|
|
}
|
2006-05-21 23:09:24 -06:00
|
|
|
skb = audit_make_reply(pid, seq, AUDIT_LIST, 1, 1, NULL, 0);
|
|
|
|
if (skb)
|
|
|
|
skb_queue_tail(q, skb);
|
2005-12-15 11:33:52 -07:00
|
|
|
}
|
|
|
|
|
2006-02-07 10:05:27 -07:00
|
|
|
/* List rules using struct audit_rule_data. */
|
2006-05-21 23:09:24 -06:00
|
|
|
static void audit_list_rules(int pid, int seq, struct sk_buff_head *q)
|
2006-02-07 10:05:27 -07:00
|
|
|
{
|
2006-05-21 23:09:24 -06:00
|
|
|
struct sk_buff *skb;
|
2006-02-07 10:05:27 -07:00
|
|
|
struct audit_entry *e;
|
|
|
|
int i;
|
|
|
|
|
[PATCH] audit: path-based rules
In this implementation, audit registers inotify watches on the parent
directories of paths specified in audit rules. When audit's inotify
event handler is called, it updates any affected rules based on the
filesystem event. If the parent directory is renamed, removed, or its
filesystem is unmounted, audit removes all rules referencing that
inotify watch.
To keep things simple, this implementation limits location-based
auditing to the directory entries in an existing directory. Given
a path-based rule for /foo/bar/passwd, the following table applies:
passwd modified -- audit event logged
passwd replaced -- audit event logged, rules list updated
bar renamed -- rule removed
foo renamed -- untracked, meaning that the rule now applies to
the new location
Audit users typically want to have many rules referencing filesystem
objects, which can significantly impact filtering performance. This
patch also adds an inode-number-based rule hash to mitigate this
situation.
The patch is relative to the audit git tree:
http://kernel.org/git/?p=linux/kernel/git/viro/audit-current.git;a=summary
and uses the inotify kernel API:
http://lkml.org/lkml/2006/6/1/145
Signed-off-by: Amy Griffis <amy.griffis@hp.com>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
2006-04-07 14:55:56 -06:00
|
|
|
/* This is a blocking read, so use audit_filter_mutex instead of rcu
|
|
|
|
* iterator to sync with list writers. */
|
2006-02-07 10:05:27 -07:00
|
|
|
for (i=0; i<AUDIT_NR_FILTERS; i++) {
|
|
|
|
list_for_each_entry(e, &audit_filter_list[i], list) {
|
|
|
|
struct audit_rule_data *data;
|
|
|
|
|
|
|
|
data = audit_krule_to_data(&e->rule);
|
|
|
|
if (unlikely(!data))
|
|
|
|
break;
|
2006-05-21 23:09:24 -06:00
|
|
|
skb = audit_make_reply(pid, seq, AUDIT_LIST_RULES, 0, 1,
|
[PATCH] audit: path-based rules
In this implementation, audit registers inotify watches on the parent
directories of paths specified in audit rules. When audit's inotify
event handler is called, it updates any affected rules based on the
filesystem event. If the parent directory is renamed, removed, or its
filesystem is unmounted, audit removes all rules referencing that
inotify watch.
To keep things simple, this implementation limits location-based
auditing to the directory entries in an existing directory. Given
a path-based rule for /foo/bar/passwd, the following table applies:
passwd modified -- audit event logged
passwd replaced -- audit event logged, rules list updated
bar renamed -- rule removed
foo renamed -- untracked, meaning that the rule now applies to
the new location
Audit users typically want to have many rules referencing filesystem
objects, which can significantly impact filtering performance. This
patch also adds an inode-number-based rule hash to mitigate this
situation.
The patch is relative to the audit git tree:
http://kernel.org/git/?p=linux/kernel/git/viro/audit-current.git;a=summary
and uses the inotify kernel API:
http://lkml.org/lkml/2006/6/1/145
Signed-off-by: Amy Griffis <amy.griffis@hp.com>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
2006-04-07 14:55:56 -06:00
|
|
|
data, sizeof(*data) + data->buflen);
|
|
|
|
if (skb)
|
|
|
|
skb_queue_tail(q, skb);
|
|
|
|
kfree(data);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
for (i=0; i< AUDIT_INODE_BUCKETS; i++) {
|
|
|
|
list_for_each_entry(e, &audit_inode_hash[i], list) {
|
|
|
|
struct audit_rule_data *data;
|
|
|
|
|
|
|
|
data = audit_krule_to_data(&e->rule);
|
|
|
|
if (unlikely(!data))
|
|
|
|
break;
|
|
|
|
skb = audit_make_reply(pid, seq, AUDIT_LIST_RULES, 0, 1,
|
|
|
|
data, sizeof(*data) + data->buflen);
|
2006-05-21 23:09:24 -06:00
|
|
|
if (skb)
|
|
|
|
skb_queue_tail(q, skb);
|
2006-02-07 10:05:27 -07:00
|
|
|
kfree(data);
|
|
|
|
}
|
|
|
|
}
|
2006-05-21 23:09:24 -06:00
|
|
|
skb = audit_make_reply(pid, seq, AUDIT_LIST_RULES, 1, 1, NULL, 0);
|
|
|
|
if (skb)
|
|
|
|
skb_queue_tail(q, skb);
|
2006-02-07 10:05:27 -07:00
|
|
|
}
|
|
|
|
|
2006-06-14 16:45:21 -06:00
|
|
|
/* Log rule additions and removals */
|
2008-04-18 08:09:25 -06:00
|
|
|
static void audit_log_rule_change(uid_t loginuid, u32 sessionid, u32 sid,
|
|
|
|
char *action, struct audit_krule *rule,
|
|
|
|
int res)
|
2006-06-14 16:45:21 -06:00
|
|
|
{
|
|
|
|
struct audit_buffer *ab;
|
|
|
|
|
2008-01-07 15:09:31 -07:00
|
|
|
if (!audit_enabled)
|
|
|
|
return;
|
|
|
|
|
2006-06-14 16:45:21 -06:00
|
|
|
ab = audit_log_start(NULL, GFP_KERNEL, AUDIT_CONFIG_CHANGE);
|
|
|
|
if (!ab)
|
|
|
|
return;
|
2008-04-18 08:09:25 -06:00
|
|
|
audit_log_format(ab, "auid=%u ses=%u", loginuid, sessionid);
|
2006-06-14 16:45:21 -06:00
|
|
|
if (sid) {
|
|
|
|
char *ctx = NULL;
|
|
|
|
u32 len;
|
Audit: use new LSM hooks instead of SELinux exports
Stop using the following exported SELinux interfaces:
selinux_get_inode_sid(inode, sid)
selinux_get_ipc_sid(ipcp, sid)
selinux_get_task_sid(tsk, sid)
selinux_sid_to_string(sid, ctx, len)
kfree(ctx)
and use following generic LSM equivalents respectively:
security_inode_getsecid(inode, secid)
security_ipc_getsecid*(ipcp, secid)
security_task_getsecid(tsk, secid)
security_sid_to_secctx(sid, ctx, len)
security_release_secctx(ctx, len)
Call security_release_secctx only if security_secid_to_secctx
succeeded.
Signed-off-by: Casey Schaufler <casey@schaufler-ca.com>
Signed-off-by: Ahmed S. Darwish <darwish.07@gmail.com>
Acked-by: James Morris <jmorris@namei.org>
Reviewed-by: Paul Moore <paul.moore@hp.com>
2008-03-01 12:54:38 -07:00
|
|
|
if (security_secid_to_secctx(sid, &ctx, &len))
|
2006-06-14 16:45:21 -06:00
|
|
|
audit_log_format(ab, " ssid=%u", sid);
|
Audit: use new LSM hooks instead of SELinux exports
Stop using the following exported SELinux interfaces:
selinux_get_inode_sid(inode, sid)
selinux_get_ipc_sid(ipcp, sid)
selinux_get_task_sid(tsk, sid)
selinux_sid_to_string(sid, ctx, len)
kfree(ctx)
and use following generic LSM equivalents respectively:
security_inode_getsecid(inode, secid)
security_ipc_getsecid*(ipcp, secid)
security_task_getsecid(tsk, secid)
security_sid_to_secctx(sid, ctx, len)
security_release_secctx(ctx, len)
Call security_release_secctx only if security_secid_to_secctx
succeeded.
Signed-off-by: Casey Schaufler <casey@schaufler-ca.com>
Signed-off-by: Ahmed S. Darwish <darwish.07@gmail.com>
Acked-by: James Morris <jmorris@namei.org>
Reviewed-by: Paul Moore <paul.moore@hp.com>
2008-03-01 12:54:38 -07:00
|
|
|
else {
|
2006-06-14 16:45:21 -06:00
|
|
|
audit_log_format(ab, " subj=%s", ctx);
|
Audit: use new LSM hooks instead of SELinux exports
Stop using the following exported SELinux interfaces:
selinux_get_inode_sid(inode, sid)
selinux_get_ipc_sid(ipcp, sid)
selinux_get_task_sid(tsk, sid)
selinux_sid_to_string(sid, ctx, len)
kfree(ctx)
and use following generic LSM equivalents respectively:
security_inode_getsecid(inode, secid)
security_ipc_getsecid*(ipcp, secid)
security_task_getsecid(tsk, secid)
security_sid_to_secctx(sid, ctx, len)
security_release_secctx(ctx, len)
Call security_release_secctx only if security_secid_to_secctx
succeeded.
Signed-off-by: Casey Schaufler <casey@schaufler-ca.com>
Signed-off-by: Ahmed S. Darwish <darwish.07@gmail.com>
Acked-by: James Morris <jmorris@namei.org>
Reviewed-by: Paul Moore <paul.moore@hp.com>
2008-03-01 12:54:38 -07:00
|
|
|
security_release_secctx(ctx, len);
|
|
|
|
}
|
2006-06-14 16:45:21 -06:00
|
|
|
}
|
2006-12-14 09:48:47 -07:00
|
|
|
audit_log_format(ab, " op=%s rule key=", action);
|
2006-06-14 16:45:21 -06:00
|
|
|
if (rule->filterkey)
|
|
|
|
audit_log_untrustedstring(ab, rule->filterkey);
|
|
|
|
else
|
|
|
|
audit_log_format(ab, "(null)");
|
|
|
|
audit_log_format(ab, " list=%d res=%d", rule->listnr, res);
|
|
|
|
audit_log_end(ab);
|
|
|
|
}
|
|
|
|
|
2005-12-15 11:33:52 -07:00
|
|
|
/**
|
|
|
|
* audit_receive_filter - apply all rules to the specified message type
|
|
|
|
* @type: audit message type
|
|
|
|
* @pid: target pid for netlink audit messages
|
|
|
|
* @uid: target uid for netlink audit messages
|
|
|
|
* @seq: netlink audit message sequence (serial) number
|
|
|
|
* @data: payload data
|
2006-02-07 10:05:27 -07:00
|
|
|
* @datasz: size of payload data
|
2005-12-15 11:33:52 -07:00
|
|
|
* @loginuid: loginuid of sender
|
2008-05-19 16:09:21 -06:00
|
|
|
* @sessionid: sessionid for netlink audit message
|
2006-04-01 16:29:34 -07:00
|
|
|
* @sid: SE Linux Security ID of sender
|
2005-12-15 11:33:52 -07:00
|
|
|
*/
|
|
|
|
int audit_receive_filter(int type, int pid, int uid, int seq, void *data,
|
2008-04-18 08:09:25 -06:00
|
|
|
size_t datasz, uid_t loginuid, u32 sessionid, u32 sid)
|
2005-12-15 11:33:52 -07:00
|
|
|
{
|
|
|
|
struct task_struct *tsk;
|
2006-05-21 23:09:24 -06:00
|
|
|
struct audit_netlink_list *dest;
|
2006-02-07 10:05:27 -07:00
|
|
|
int err = 0;
|
|
|
|
struct audit_entry *entry;
|
2005-12-15 11:33:52 -07:00
|
|
|
|
|
|
|
switch (type) {
|
|
|
|
case AUDIT_LIST:
|
2006-02-07 10:05:27 -07:00
|
|
|
case AUDIT_LIST_RULES:
|
2005-12-15 11:33:52 -07:00
|
|
|
/* We can't just spew out the rules here because we might fill
|
|
|
|
* the available socket buffer space and deadlock waiting for
|
|
|
|
* auditctl to read from it... which isn't ever going to
|
|
|
|
* happen if we're actually running in the context of auditctl
|
|
|
|
* trying to _send_ the stuff */
|
2007-10-18 04:06:06 -06:00
|
|
|
|
2006-05-21 23:09:24 -06:00
|
|
|
dest = kmalloc(sizeof(struct audit_netlink_list), GFP_KERNEL);
|
2005-12-15 11:33:52 -07:00
|
|
|
if (!dest)
|
|
|
|
return -ENOMEM;
|
2006-05-21 23:09:24 -06:00
|
|
|
dest->pid = pid;
|
|
|
|
skb_queue_head_init(&dest->q);
|
2005-12-15 11:33:52 -07:00
|
|
|
|
[PATCH] audit: path-based rules
In this implementation, audit registers inotify watches on the parent
directories of paths specified in audit rules. When audit's inotify
event handler is called, it updates any affected rules based on the
filesystem event. If the parent directory is renamed, removed, or its
filesystem is unmounted, audit removes all rules referencing that
inotify watch.
To keep things simple, this implementation limits location-based
auditing to the directory entries in an existing directory. Given
a path-based rule for /foo/bar/passwd, the following table applies:
passwd modified -- audit event logged
passwd replaced -- audit event logged, rules list updated
bar renamed -- rule removed
foo renamed -- untracked, meaning that the rule now applies to
the new location
Audit users typically want to have many rules referencing filesystem
objects, which can significantly impact filtering performance. This
patch also adds an inode-number-based rule hash to mitigate this
situation.
The patch is relative to the audit git tree:
http://kernel.org/git/?p=linux/kernel/git/viro/audit-current.git;a=summary
and uses the inotify kernel API:
http://lkml.org/lkml/2006/6/1/145
Signed-off-by: Amy Griffis <amy.griffis@hp.com>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
2006-04-07 14:55:56 -06:00
|
|
|
mutex_lock(&audit_filter_mutex);
|
2006-02-07 10:05:27 -07:00
|
|
|
if (type == AUDIT_LIST)
|
2006-05-21 23:09:24 -06:00
|
|
|
audit_list(pid, seq, &dest->q);
|
2006-02-07 10:05:27 -07:00
|
|
|
else
|
2006-05-21 23:09:24 -06:00
|
|
|
audit_list_rules(pid, seq, &dest->q);
|
[PATCH] audit: path-based rules
In this implementation, audit registers inotify watches on the parent
directories of paths specified in audit rules. When audit's inotify
event handler is called, it updates any affected rules based on the
filesystem event. If the parent directory is renamed, removed, or its
filesystem is unmounted, audit removes all rules referencing that
inotify watch.
To keep things simple, this implementation limits location-based
auditing to the directory entries in an existing directory. Given
a path-based rule for /foo/bar/passwd, the following table applies:
passwd modified -- audit event logged
passwd replaced -- audit event logged, rules list updated
bar renamed -- rule removed
foo renamed -- untracked, meaning that the rule now applies to
the new location
Audit users typically want to have many rules referencing filesystem
objects, which can significantly impact filtering performance. This
patch also adds an inode-number-based rule hash to mitigate this
situation.
The patch is relative to the audit git tree:
http://kernel.org/git/?p=linux/kernel/git/viro/audit-current.git;a=summary
and uses the inotify kernel API:
http://lkml.org/lkml/2006/6/1/145
Signed-off-by: Amy Griffis <amy.griffis@hp.com>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
2006-04-07 14:55:56 -06:00
|
|
|
mutex_unlock(&audit_filter_mutex);
|
2006-05-21 23:09:24 -06:00
|
|
|
|
|
|
|
tsk = kthread_run(audit_send_list, dest, "audit_send_list");
|
2005-12-15 11:33:52 -07:00
|
|
|
if (IS_ERR(tsk)) {
|
2006-05-21 23:09:24 -06:00
|
|
|
skb_queue_purge(&dest->q);
|
2005-12-15 11:33:52 -07:00
|
|
|
kfree(dest);
|
|
|
|
err = PTR_ERR(tsk);
|
|
|
|
}
|
|
|
|
break;
|
|
|
|
case AUDIT_ADD:
|
2006-02-07 10:05:27 -07:00
|
|
|
case AUDIT_ADD_RULE:
|
|
|
|
if (type == AUDIT_ADD)
|
|
|
|
entry = audit_rule_to_entry(data);
|
|
|
|
else
|
|
|
|
entry = audit_data_to_entry(data, datasz);
|
|
|
|
if (IS_ERR(entry))
|
|
|
|
return PTR_ERR(entry);
|
|
|
|
|
|
|
|
err = audit_add_rule(entry,
|
|
|
|
&audit_filter_list[entry->rule.listnr]);
|
2008-04-18 08:09:25 -06:00
|
|
|
audit_log_rule_change(loginuid, sessionid, sid, "add",
|
|
|
|
&entry->rule, !err);
|
2006-01-09 07:48:17 -07:00
|
|
|
|
|
|
|
if (err)
|
2006-02-07 10:05:27 -07:00
|
|
|
audit_free_rule(entry);
|
2005-12-15 11:33:52 -07:00
|
|
|
break;
|
|
|
|
case AUDIT_DEL:
|
2006-02-07 10:05:27 -07:00
|
|
|
case AUDIT_DEL_RULE:
|
|
|
|
if (type == AUDIT_DEL)
|
|
|
|
entry = audit_rule_to_entry(data);
|
|
|
|
else
|
|
|
|
entry = audit_data_to_entry(data, datasz);
|
|
|
|
if (IS_ERR(entry))
|
|
|
|
return PTR_ERR(entry);
|
|
|
|
|
|
|
|
err = audit_del_rule(entry,
|
|
|
|
&audit_filter_list[entry->rule.listnr]);
|
2008-04-18 08:09:25 -06:00
|
|
|
audit_log_rule_change(loginuid, sessionid, sid, "remove",
|
|
|
|
&entry->rule, !err);
|
2006-01-09 07:48:17 -07:00
|
|
|
|
2006-02-07 10:05:27 -07:00
|
|
|
audit_free_rule(entry);
|
2005-12-15 11:33:52 -07:00
|
|
|
break;
|
|
|
|
default:
|
|
|
|
return -EINVAL;
|
|
|
|
}
|
|
|
|
|
|
|
|
return err;
|
|
|
|
}
|
|
|
|
|
|
|
|
int audit_comparator(const u32 left, const u32 op, const u32 right)
|
|
|
|
{
|
|
|
|
switch (op) {
|
|
|
|
case AUDIT_EQUAL:
|
|
|
|
return (left == right);
|
|
|
|
case AUDIT_NOT_EQUAL:
|
|
|
|
return (left != right);
|
|
|
|
case AUDIT_LESS_THAN:
|
|
|
|
return (left < right);
|
|
|
|
case AUDIT_LESS_THAN_OR_EQUAL:
|
|
|
|
return (left <= right);
|
|
|
|
case AUDIT_GREATER_THAN:
|
|
|
|
return (left > right);
|
|
|
|
case AUDIT_GREATER_THAN_OR_EQUAL:
|
|
|
|
return (left >= right);
|
2007-06-04 15:00:14 -06:00
|
|
|
case AUDIT_BIT_MASK:
|
|
|
|
return (left & right);
|
|
|
|
case AUDIT_BIT_TEST:
|
|
|
|
return ((left & right) == right);
|
2005-12-15 11:33:52 -07:00
|
|
|
}
|
2006-02-16 12:40:01 -07:00
|
|
|
BUG();
|
|
|
|
return 0;
|
2005-12-15 11:33:52 -07:00
|
|
|
}
|
|
|
|
|
[PATCH] audit: path-based rules
In this implementation, audit registers inotify watches on the parent
directories of paths specified in audit rules. When audit's inotify
event handler is called, it updates any affected rules based on the
filesystem event. If the parent directory is renamed, removed, or its
filesystem is unmounted, audit removes all rules referencing that
inotify watch.
To keep things simple, this implementation limits location-based
auditing to the directory entries in an existing directory. Given
a path-based rule for /foo/bar/passwd, the following table applies:
passwd modified -- audit event logged
passwd replaced -- audit event logged, rules list updated
bar renamed -- rule removed
foo renamed -- untracked, meaning that the rule now applies to
the new location
Audit users typically want to have many rules referencing filesystem
objects, which can significantly impact filtering performance. This
patch also adds an inode-number-based rule hash to mitigate this
situation.
The patch is relative to the audit git tree:
http://kernel.org/git/?p=linux/kernel/git/viro/audit-current.git;a=summary
and uses the inotify kernel API:
http://lkml.org/lkml/2006/6/1/145
Signed-off-by: Amy Griffis <amy.griffis@hp.com>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
2006-04-07 14:55:56 -06:00
|
|
|
/* Compare given dentry name with last component in given path,
|
|
|
|
* return of 0 indicates a match. */
|
2006-06-08 21:19:31 -06:00
|
|
|
int audit_compare_dname_path(const char *dname, const char *path,
|
|
|
|
int *dirlen)
|
[PATCH] audit: path-based rules
In this implementation, audit registers inotify watches on the parent
directories of paths specified in audit rules. When audit's inotify
event handler is called, it updates any affected rules based on the
filesystem event. If the parent directory is renamed, removed, or its
filesystem is unmounted, audit removes all rules referencing that
inotify watch.
To keep things simple, this implementation limits location-based
auditing to the directory entries in an existing directory. Given
a path-based rule for /foo/bar/passwd, the following table applies:
passwd modified -- audit event logged
passwd replaced -- audit event logged, rules list updated
bar renamed -- rule removed
foo renamed -- untracked, meaning that the rule now applies to
the new location
Audit users typically want to have many rules referencing filesystem
objects, which can significantly impact filtering performance. This
patch also adds an inode-number-based rule hash to mitigate this
situation.
The patch is relative to the audit git tree:
http://kernel.org/git/?p=linux/kernel/git/viro/audit-current.git;a=summary
and uses the inotify kernel API:
http://lkml.org/lkml/2006/6/1/145
Signed-off-by: Amy Griffis <amy.griffis@hp.com>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
2006-04-07 14:55:56 -06:00
|
|
|
{
|
|
|
|
int dlen, plen;
|
|
|
|
const char *p;
|
|
|
|
|
|
|
|
if (!dname || !path)
|
|
|
|
return 1;
|
|
|
|
|
|
|
|
dlen = strlen(dname);
|
|
|
|
plen = strlen(path);
|
|
|
|
if (plen < dlen)
|
|
|
|
return 1;
|
|
|
|
|
|
|
|
/* disregard trailing slashes */
|
|
|
|
p = path + plen - 1;
|
|
|
|
while ((*p == '/') && (p > path))
|
|
|
|
p--;
|
|
|
|
|
|
|
|
/* find last path component */
|
|
|
|
p = p - dlen + 1;
|
|
|
|
if (p < path)
|
|
|
|
return 1;
|
|
|
|
else if (p > path) {
|
|
|
|
if (*--p != '/')
|
|
|
|
return 1;
|
|
|
|
else
|
|
|
|
p++;
|
|
|
|
}
|
2005-12-15 11:33:52 -07:00
|
|
|
|
2006-06-08 21:19:31 -06:00
|
|
|
/* return length of path's directory component */
|
|
|
|
if (dirlen)
|
|
|
|
*dirlen = p - path;
|
[PATCH] audit: path-based rules
In this implementation, audit registers inotify watches on the parent
directories of paths specified in audit rules. When audit's inotify
event handler is called, it updates any affected rules based on the
filesystem event. If the parent directory is renamed, removed, or its
filesystem is unmounted, audit removes all rules referencing that
inotify watch.
To keep things simple, this implementation limits location-based
auditing to the directory entries in an existing directory. Given
a path-based rule for /foo/bar/passwd, the following table applies:
passwd modified -- audit event logged
passwd replaced -- audit event logged, rules list updated
bar renamed -- rule removed
foo renamed -- untracked, meaning that the rule now applies to
the new location
Audit users typically want to have many rules referencing filesystem
objects, which can significantly impact filtering performance. This
patch also adds an inode-number-based rule hash to mitigate this
situation.
The patch is relative to the audit git tree:
http://kernel.org/git/?p=linux/kernel/git/viro/audit-current.git;a=summary
and uses the inotify kernel API:
http://lkml.org/lkml/2006/6/1/145
Signed-off-by: Amy Griffis <amy.griffis@hp.com>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
2006-04-07 14:55:56 -06:00
|
|
|
return strncmp(p, dname, dlen);
|
|
|
|
}
|
2005-12-15 11:33:52 -07:00
|
|
|
|
|
|
|
static int audit_filter_user_rules(struct netlink_skb_parms *cb,
|
2006-02-07 10:05:27 -07:00
|
|
|
struct audit_krule *rule,
|
2005-12-15 11:33:52 -07:00
|
|
|
enum audit_state *state)
|
|
|
|
{
|
|
|
|
int i;
|
|
|
|
|
|
|
|
for (i = 0; i < rule->field_count; i++) {
|
2006-02-07 10:05:27 -07:00
|
|
|
struct audit_field *f = &rule->fields[i];
|
2005-12-15 11:33:52 -07:00
|
|
|
int result = 0;
|
|
|
|
|
2006-02-07 10:05:27 -07:00
|
|
|
switch (f->type) {
|
2005-12-15 11:33:52 -07:00
|
|
|
case AUDIT_PID:
|
2006-02-07 10:05:27 -07:00
|
|
|
result = audit_comparator(cb->creds.pid, f->op, f->val);
|
2005-12-15 11:33:52 -07:00
|
|
|
break;
|
|
|
|
case AUDIT_UID:
|
2006-02-07 10:05:27 -07:00
|
|
|
result = audit_comparator(cb->creds.uid, f->op, f->val);
|
2005-12-15 11:33:52 -07:00
|
|
|
break;
|
|
|
|
case AUDIT_GID:
|
2006-02-07 10:05:27 -07:00
|
|
|
result = audit_comparator(cb->creds.gid, f->op, f->val);
|
2005-12-15 11:33:52 -07:00
|
|
|
break;
|
|
|
|
case AUDIT_LOGINUID:
|
2006-02-07 10:05:27 -07:00
|
|
|
result = audit_comparator(cb->loginuid, f->op, f->val);
|
2005-12-15 11:33:52 -07:00
|
|
|
break;
|
|
|
|
}
|
|
|
|
|
|
|
|
if (!result)
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
switch (rule->action) {
|
|
|
|
case AUDIT_NEVER: *state = AUDIT_DISABLED; break;
|
|
|
|
case AUDIT_ALWAYS: *state = AUDIT_RECORD_CONTEXT; break;
|
|
|
|
}
|
|
|
|
return 1;
|
|
|
|
}
|
|
|
|
|
2008-05-19 19:13:02 -06:00
|
|
|
int audit_filter_user(struct netlink_skb_parms *cb)
|
2005-12-15 11:33:52 -07:00
|
|
|
{
|
2007-02-10 02:46:09 -07:00
|
|
|
enum audit_state state = AUDIT_DISABLED;
|
2005-12-15 11:33:52 -07:00
|
|
|
struct audit_entry *e;
|
|
|
|
int ret = 1;
|
|
|
|
|
|
|
|
rcu_read_lock();
|
|
|
|
list_for_each_entry_rcu(e, &audit_filter_list[AUDIT_FILTER_USER], list) {
|
|
|
|
if (audit_filter_user_rules(cb, &e->rule, &state)) {
|
|
|
|
if (state == AUDIT_DISABLED)
|
|
|
|
ret = 0;
|
|
|
|
break;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
rcu_read_unlock();
|
|
|
|
|
|
|
|
return ret; /* Audit by default */
|
|
|
|
}
|
|
|
|
|
|
|
|
int audit_filter_type(int type)
|
|
|
|
{
|
|
|
|
struct audit_entry *e;
|
|
|
|
int result = 0;
|
2007-10-18 04:06:06 -06:00
|
|
|
|
2005-12-15 11:33:52 -07:00
|
|
|
rcu_read_lock();
|
|
|
|
if (list_empty(&audit_filter_list[AUDIT_FILTER_TYPE]))
|
|
|
|
goto unlock_and_return;
|
|
|
|
|
|
|
|
list_for_each_entry_rcu(e, &audit_filter_list[AUDIT_FILTER_TYPE],
|
|
|
|
list) {
|
|
|
|
int i;
|
2006-02-07 10:05:27 -07:00
|
|
|
for (i = 0; i < e->rule.field_count; i++) {
|
|
|
|
struct audit_field *f = &e->rule.fields[i];
|
|
|
|
if (f->type == AUDIT_MSGTYPE) {
|
|
|
|
result = audit_comparator(type, f->op, f->val);
|
2005-12-15 11:33:52 -07:00
|
|
|
if (!result)
|
|
|
|
break;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
if (result)
|
|
|
|
goto unlock_and_return;
|
|
|
|
}
|
|
|
|
unlock_and_return:
|
|
|
|
rcu_read_unlock();
|
|
|
|
return result;
|
|
|
|
}
|
2006-03-10 17:14:06 -07:00
|
|
|
|
2008-04-18 17:59:43 -06:00
|
|
|
/* This function will re-initialize the lsm_rule field of all applicable rules.
|
2008-03-01 13:01:11 -07:00
|
|
|
* It will traverse the filter lists serarching for rules that contain LSM
|
2006-03-10 17:14:06 -07:00
|
|
|
* specific filter fields. When such a rule is found, it is copied, the
|
2008-03-01 13:01:11 -07:00
|
|
|
* LSM field is re-initialized, and the old rule is replaced with the
|
2006-03-10 17:14:06 -07:00
|
|
|
* updated rule. */
|
2008-03-01 13:01:11 -07:00
|
|
|
int audit_update_lsm_rules(void)
|
2006-03-10 17:14:06 -07:00
|
|
|
{
|
|
|
|
struct audit_entry *entry, *n, *nentry;
|
[PATCH] audit: path-based rules
In this implementation, audit registers inotify watches on the parent
directories of paths specified in audit rules. When audit's inotify
event handler is called, it updates any affected rules based on the
filesystem event. If the parent directory is renamed, removed, or its
filesystem is unmounted, audit removes all rules referencing that
inotify watch.
To keep things simple, this implementation limits location-based
auditing to the directory entries in an existing directory. Given
a path-based rule for /foo/bar/passwd, the following table applies:
passwd modified -- audit event logged
passwd replaced -- audit event logged, rules list updated
bar renamed -- rule removed
foo renamed -- untracked, meaning that the rule now applies to
the new location
Audit users typically want to have many rules referencing filesystem
objects, which can significantly impact filtering performance. This
patch also adds an inode-number-based rule hash to mitigate this
situation.
The patch is relative to the audit git tree:
http://kernel.org/git/?p=linux/kernel/git/viro/audit-current.git;a=summary
and uses the inotify kernel API:
http://lkml.org/lkml/2006/6/1/145
Signed-off-by: Amy Griffis <amy.griffis@hp.com>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
2006-04-07 14:55:56 -06:00
|
|
|
struct audit_watch *watch;
|
[PATCH] audit: watching subtrees
New kind of audit rule predicates: "object is visible in given subtree".
The part that can be sanely implemented, that is. Limitations:
* if you have hardlink from outside of tree, you'd better watch
it too (or just watch the object itself, obviously)
* if you mount something under a watched tree, tell audit
that new chunk should be added to watched subtrees
* if you umount something in a watched tree and it's still mounted
elsewhere, you will get matches on events happening there. New command
tells audit to recalculate the trees, trimming such sources of false
positives.
Note that it's _not_ about path - if something mounted in several places
(multiple mount, bindings, different namespaces, etc.), the match does
_not_ depend on which one we are using for access.
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
2007-07-22 06:04:18 -06:00
|
|
|
struct audit_tree *tree;
|
2006-03-10 17:14:06 -07:00
|
|
|
int i, err = 0;
|
|
|
|
|
[PATCH] audit: path-based rules
In this implementation, audit registers inotify watches on the parent
directories of paths specified in audit rules. When audit's inotify
event handler is called, it updates any affected rules based on the
filesystem event. If the parent directory is renamed, removed, or its
filesystem is unmounted, audit removes all rules referencing that
inotify watch.
To keep things simple, this implementation limits location-based
auditing to the directory entries in an existing directory. Given
a path-based rule for /foo/bar/passwd, the following table applies:
passwd modified -- audit event logged
passwd replaced -- audit event logged, rules list updated
bar renamed -- rule removed
foo renamed -- untracked, meaning that the rule now applies to
the new location
Audit users typically want to have many rules referencing filesystem
objects, which can significantly impact filtering performance. This
patch also adds an inode-number-based rule hash to mitigate this
situation.
The patch is relative to the audit git tree:
http://kernel.org/git/?p=linux/kernel/git/viro/audit-current.git;a=summary
and uses the inotify kernel API:
http://lkml.org/lkml/2006/6/1/145
Signed-off-by: Amy Griffis <amy.griffis@hp.com>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
2006-04-07 14:55:56 -06:00
|
|
|
/* audit_filter_mutex synchronizes the writers */
|
|
|
|
mutex_lock(&audit_filter_mutex);
|
2006-03-10 17:14:06 -07:00
|
|
|
|
|
|
|
for (i = 0; i < AUDIT_NR_FILTERS; i++) {
|
|
|
|
list_for_each_entry_safe(entry, n, &audit_filter_list[i], list) {
|
2008-03-01 13:01:11 -07:00
|
|
|
if (!security_audit_rule_known(&entry->rule))
|
2006-03-10 17:14:06 -07:00
|
|
|
continue;
|
|
|
|
|
[PATCH] audit: path-based rules
In this implementation, audit registers inotify watches on the parent
directories of paths specified in audit rules. When audit's inotify
event handler is called, it updates any affected rules based on the
filesystem event. If the parent directory is renamed, removed, or its
filesystem is unmounted, audit removes all rules referencing that
inotify watch.
To keep things simple, this implementation limits location-based
auditing to the directory entries in an existing directory. Given
a path-based rule for /foo/bar/passwd, the following table applies:
passwd modified -- audit event logged
passwd replaced -- audit event logged, rules list updated
bar renamed -- rule removed
foo renamed -- untracked, meaning that the rule now applies to
the new location
Audit users typically want to have many rules referencing filesystem
objects, which can significantly impact filtering performance. This
patch also adds an inode-number-based rule hash to mitigate this
situation.
The patch is relative to the audit git tree:
http://kernel.org/git/?p=linux/kernel/git/viro/audit-current.git;a=summary
and uses the inotify kernel API:
http://lkml.org/lkml/2006/6/1/145
Signed-off-by: Amy Griffis <amy.griffis@hp.com>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
2006-04-07 14:55:56 -06:00
|
|
|
watch = entry->rule.watch;
|
[PATCH] audit: watching subtrees
New kind of audit rule predicates: "object is visible in given subtree".
The part that can be sanely implemented, that is. Limitations:
* if you have hardlink from outside of tree, you'd better watch
it too (or just watch the object itself, obviously)
* if you mount something under a watched tree, tell audit
that new chunk should be added to watched subtrees
* if you umount something in a watched tree and it's still mounted
elsewhere, you will get matches on events happening there. New command
tells audit to recalculate the trees, trimming such sources of false
positives.
Note that it's _not_ about path - if something mounted in several places
(multiple mount, bindings, different namespaces, etc.), the match does
_not_ depend on which one we are using for access.
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
2007-07-22 06:04:18 -06:00
|
|
|
tree = entry->rule.tree;
|
[PATCH] audit: path-based rules
In this implementation, audit registers inotify watches on the parent
directories of paths specified in audit rules. When audit's inotify
event handler is called, it updates any affected rules based on the
filesystem event. If the parent directory is renamed, removed, or its
filesystem is unmounted, audit removes all rules referencing that
inotify watch.
To keep things simple, this implementation limits location-based
auditing to the directory entries in an existing directory. Given
a path-based rule for /foo/bar/passwd, the following table applies:
passwd modified -- audit event logged
passwd replaced -- audit event logged, rules list updated
bar renamed -- rule removed
foo renamed -- untracked, meaning that the rule now applies to
the new location
Audit users typically want to have many rules referencing filesystem
objects, which can significantly impact filtering performance. This
patch also adds an inode-number-based rule hash to mitigate this
situation.
The patch is relative to the audit git tree:
http://kernel.org/git/?p=linux/kernel/git/viro/audit-current.git;a=summary
and uses the inotify kernel API:
http://lkml.org/lkml/2006/6/1/145
Signed-off-by: Amy Griffis <amy.griffis@hp.com>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
2006-04-07 14:55:56 -06:00
|
|
|
nentry = audit_dupe_rule(&entry->rule, watch);
|
2008-04-29 02:03:09 -06:00
|
|
|
if (IS_ERR(nentry)) {
|
2006-03-10 17:14:06 -07:00
|
|
|
/* save the first error encountered for the
|
|
|
|
* return value */
|
|
|
|
if (!err)
|
|
|
|
err = PTR_ERR(nentry);
|
2008-03-01 13:01:11 -07:00
|
|
|
audit_panic("error updating LSM filters");
|
[PATCH] audit: path-based rules
In this implementation, audit registers inotify watches on the parent
directories of paths specified in audit rules. When audit's inotify
event handler is called, it updates any affected rules based on the
filesystem event. If the parent directory is renamed, removed, or its
filesystem is unmounted, audit removes all rules referencing that
inotify watch.
To keep things simple, this implementation limits location-based
auditing to the directory entries in an existing directory. Given
a path-based rule for /foo/bar/passwd, the following table applies:
passwd modified -- audit event logged
passwd replaced -- audit event logged, rules list updated
bar renamed -- rule removed
foo renamed -- untracked, meaning that the rule now applies to
the new location
Audit users typically want to have many rules referencing filesystem
objects, which can significantly impact filtering performance. This
patch also adds an inode-number-based rule hash to mitigate this
situation.
The patch is relative to the audit git tree:
http://kernel.org/git/?p=linux/kernel/git/viro/audit-current.git;a=summary
and uses the inotify kernel API:
http://lkml.org/lkml/2006/6/1/145
Signed-off-by: Amy Griffis <amy.griffis@hp.com>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
2006-04-07 14:55:56 -06:00
|
|
|
if (watch)
|
|
|
|
list_del(&entry->rule.rlist);
|
2006-03-10 17:14:06 -07:00
|
|
|
list_del_rcu(&entry->list);
|
|
|
|
} else {
|
[PATCH] audit: path-based rules
In this implementation, audit registers inotify watches on the parent
directories of paths specified in audit rules. When audit's inotify
event handler is called, it updates any affected rules based on the
filesystem event. If the parent directory is renamed, removed, or its
filesystem is unmounted, audit removes all rules referencing that
inotify watch.
To keep things simple, this implementation limits location-based
auditing to the directory entries in an existing directory. Given
a path-based rule for /foo/bar/passwd, the following table applies:
passwd modified -- audit event logged
passwd replaced -- audit event logged, rules list updated
bar renamed -- rule removed
foo renamed -- untracked, meaning that the rule now applies to
the new location
Audit users typically want to have many rules referencing filesystem
objects, which can significantly impact filtering performance. This
patch also adds an inode-number-based rule hash to mitigate this
situation.
The patch is relative to the audit git tree:
http://kernel.org/git/?p=linux/kernel/git/viro/audit-current.git;a=summary
and uses the inotify kernel API:
http://lkml.org/lkml/2006/6/1/145
Signed-off-by: Amy Griffis <amy.griffis@hp.com>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
2006-04-07 14:55:56 -06:00
|
|
|
if (watch) {
|
|
|
|
list_add(&nentry->rule.rlist,
|
|
|
|
&watch->rules);
|
|
|
|
list_del(&entry->rule.rlist);
|
[PATCH] audit: watching subtrees
New kind of audit rule predicates: "object is visible in given subtree".
The part that can be sanely implemented, that is. Limitations:
* if you have hardlink from outside of tree, you'd better watch
it too (or just watch the object itself, obviously)
* if you mount something under a watched tree, tell audit
that new chunk should be added to watched subtrees
* if you umount something in a watched tree and it's still mounted
elsewhere, you will get matches on events happening there. New command
tells audit to recalculate the trees, trimming such sources of false
positives.
Note that it's _not_ about path - if something mounted in several places
(multiple mount, bindings, different namespaces, etc.), the match does
_not_ depend on which one we are using for access.
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
2007-07-22 06:04:18 -06:00
|
|
|
} else if (tree)
|
|
|
|
list_replace_init(&entry->rule.rlist,
|
|
|
|
&nentry->rule.rlist);
|
2006-03-10 17:14:06 -07:00
|
|
|
list_replace_rcu(&entry->list, &nentry->list);
|
|
|
|
}
|
|
|
|
call_rcu(&entry->rcu, audit_free_rule_rcu);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
[PATCH] audit: path-based rules
In this implementation, audit registers inotify watches on the parent
directories of paths specified in audit rules. When audit's inotify
event handler is called, it updates any affected rules based on the
filesystem event. If the parent directory is renamed, removed, or its
filesystem is unmounted, audit removes all rules referencing that
inotify watch.
To keep things simple, this implementation limits location-based
auditing to the directory entries in an existing directory. Given
a path-based rule for /foo/bar/passwd, the following table applies:
passwd modified -- audit event logged
passwd replaced -- audit event logged, rules list updated
bar renamed -- rule removed
foo renamed -- untracked, meaning that the rule now applies to
the new location
Audit users typically want to have many rules referencing filesystem
objects, which can significantly impact filtering performance. This
patch also adds an inode-number-based rule hash to mitigate this
situation.
The patch is relative to the audit git tree:
http://kernel.org/git/?p=linux/kernel/git/viro/audit-current.git;a=summary
and uses the inotify kernel API:
http://lkml.org/lkml/2006/6/1/145
Signed-off-by: Amy Griffis <amy.griffis@hp.com>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
2006-04-07 14:55:56 -06:00
|
|
|
mutex_unlock(&audit_filter_mutex);
|
2006-03-10 17:14:06 -07:00
|
|
|
|
|
|
|
return err;
|
|
|
|
}
|
[PATCH] audit: path-based rules
In this implementation, audit registers inotify watches on the parent
directories of paths specified in audit rules. When audit's inotify
event handler is called, it updates any affected rules based on the
filesystem event. If the parent directory is renamed, removed, or its
filesystem is unmounted, audit removes all rules referencing that
inotify watch.
To keep things simple, this implementation limits location-based
auditing to the directory entries in an existing directory. Given
a path-based rule for /foo/bar/passwd, the following table applies:
passwd modified -- audit event logged
passwd replaced -- audit event logged, rules list updated
bar renamed -- rule removed
foo renamed -- untracked, meaning that the rule now applies to
the new location
Audit users typically want to have many rules referencing filesystem
objects, which can significantly impact filtering performance. This
patch also adds an inode-number-based rule hash to mitigate this
situation.
The patch is relative to the audit git tree:
http://kernel.org/git/?p=linux/kernel/git/viro/audit-current.git;a=summary
and uses the inotify kernel API:
http://lkml.org/lkml/2006/6/1/145
Signed-off-by: Amy Griffis <amy.griffis@hp.com>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
2006-04-07 14:55:56 -06:00
|
|
|
|
|
|
|
/* Update watch data in audit rules based on inotify events. */
|
|
|
|
void audit_handle_ievent(struct inotify_watch *i_watch, u32 wd, u32 mask,
|
|
|
|
u32 cookie, const char *dname, struct inode *inode)
|
|
|
|
{
|
|
|
|
struct audit_parent *parent;
|
|
|
|
|
|
|
|
parent = container_of(i_watch, struct audit_parent, wdata);
|
|
|
|
|
|
|
|
if (mask & (IN_CREATE|IN_MOVED_TO) && inode)
|
|
|
|
audit_update_watch(parent, dname, inode->i_sb->s_dev,
|
|
|
|
inode->i_ino, 0);
|
|
|
|
else if (mask & (IN_DELETE|IN_MOVED_FROM))
|
|
|
|
audit_update_watch(parent, dname, (dev_t)-1, (unsigned long)-1, 1);
|
|
|
|
/* inotify automatically removes the watch and sends IN_IGNORED */
|
|
|
|
else if (mask & (IN_DELETE_SELF|IN_UNMOUNT))
|
|
|
|
audit_remove_parent_watches(parent);
|
|
|
|
/* inotify does not remove the watch, so remove it manually */
|
|
|
|
else if(mask & IN_MOVE_SELF) {
|
|
|
|
audit_remove_parent_watches(parent);
|
|
|
|
inotify_remove_watch_locked(audit_ih, i_watch);
|
|
|
|
} else if (mask & IN_IGNORED)
|
|
|
|
put_inotify_watch(i_watch);
|
|
|
|
}
|