[RFC PATCH ghak100 V1 1/2] audit: avoid fcaps on MNT_FORCE

[Richard Guy Briggs]

Don't fetch fcaps when umount2 is called with MNT_FORCE to avoid a
process hang while it waits for the missing resource to (possibly never)
re-appear.

Note the comment above user_path_mountpoint_at():
* A umount is a special case for path walking. We're not actually interested
* in the inode in this situation, and ESTALE errors can be a problem. We
* simply want track down the dentry and vfsmount attached at the mountpoint
* and avoid revalidating the last component.

This can happen on ceph, cifs, 9p, lustre, fuse (gluster) or NFS.

Please see the github issue tracker
https://github.com/linux-audit/audit-kernel/issues/100

Signed-off-by: Richard Guy Briggs <***@redhat.com>
---
fs/namei.c | 2 +-
fs/namespace.c | 3 +++
include/linux/audit.h | 8 ++++++--
kernel/audit.c | 5 +++--
kernel/audit.h | 2 +-
kernel/auditsc.c | 6 +++---
6 files changed, 17 insertions(+), 9 deletions(-)

diff --git a/fs/namei.c b/fs/namei.c
index 0cab6494978c..5ac8410b704c 100644
--- a/fs/namei.c
+++ b/fs/namei.c
@@ -2720,7 +2720,7 @@ int user_path_at_empty(int dfd, const char __user *name, unsigned flags,
if (unlikely(error == -ESTALE))
error = path_mountpoint(&nd, flags | LOOKUP_REVAL, path);
if (likely(!error))
- audit_inode(name, path->dentry, 0);
+ audit_inode(name, path->dentry, flags & LOOKUP_NO_REVAL);
restore_nameidata();
putname(name);
return error;
diff --git a/fs/namespace.c b/fs/namespace.c
index 99186556f8d3..5bae5bbd4e1f 100644
--- a/fs/namespace.c
+++ b/fs/namespace.c
@@ -1636,6 +1636,9 @@ int ksys_umount(char __user *name, int flags)
if (!(flags & UMOUNT_NOFOLLOW))
lookup_flags |= LOOKUP_FOLLOW;

+ if (!(flags & MNT_FORCE))
+ lookup_flags |= LOOKUP_NO_REVAL;
+
retval = user_path_mountpoint_at(AT_FDCWD, name, lookup_flags, &path);
if (retval)
goto out;
diff --git a/include/linux/audit.h b/include/linux/audit.h
index 9334fbef7bae..503f1710c9d0 100644
--- a/include/linux/audit.h
+++ b/include/linux/audit.h
@@ -25,6 +25,7 @@

#include <linux/sched.h>
#include <linux/ptrace.h>
+#include <linux/namei.h> /* LOOKUP_* */
#include <uapi/linux/audit.h>

#define AUDIT_INO_UNSET ((unsigned long)-1)
@@ -229,6 +230,7 @@ extern void __audit_syscall_entry(int major, unsigned long a0, unsigned long a1,

#define AUDIT_INODE_PARENT 1 /* dentry represents the parent */
#define AUDIT_INODE_HIDDEN 2 /* audit record should be hidden */
+#define AUDIT_INODE_NOREVAL 4 /* audit record incomplete */
extern void __audit_inode(struct filename *name, const struct dentry *dentry,
unsigned int flags);
extern void __audit_file(const struct file *);
@@ -289,11 +291,13 @@ static inline void audit_getname(struct filename *name)
}
static inline void audit_inode(struct filename *name,
const struct dentry *dentry,
- unsigned int parent) {
+ unsigned int lflags) {
if (unlikely(!audit_dummy_context())) {
unsigned int flags = 0;
- if (parent)
+ if (lflags & LOOKUP_PARENT)
flags |= AUDIT_INODE_PARENT;
+ if (lflags & LOOKUP_NO_REVAL)
+ flags |= AUDIT_INODE_NOREVAL;
__audit_inode(name, dentry, flags);
}
}
diff --git a/kernel/audit.c b/kernel/audit.c
index 2a8058764aa6..45ca6d15ce89 100644
--- a/kernel/audit.c
+++ b/kernel/audit.c
@@ -2097,7 +2097,7 @@ static inline int audit_copy_fcaps(struct audit_names *name,

/* Copy inode data into an audit_names. */
void audit_copy_inode(struct audit_names *name, const struct dentry *dentry,
- struct inode *inode)
+ struct inode *inode, unsigned int flags)
{
name->ino = inode->i_ino;
name->dev = inode->i_sb->s_dev;
@@ -2106,7 +2106,8 @@ void audit_copy_inode(struct audit_names *name, const struct dentry *dentry,
name->gid = inode->i_gid;
name->rdev = inode->i_rdev;
security_inode_getsecid(inode, &name->osid);
- audit_copy_fcaps(name, dentry);
+ if (!(flags & AUDIT_INODE_NOREVAL))
+ audit_copy_fcaps(name, dentry);
}

/**
diff --git a/kernel/audit.h b/kernel/audit.h
index 214e14948370..7db09151c2d0 100644
--- a/kernel/audit.h
+++ b/kernel/audit.h
@@ -212,7 +212,7 @@ struct audit_context {

extern void audit_copy_inode(struct audit_names *name,
const struct dentry *dentry,
- struct inode *inode);
+ struct inode *inode, unsigned int flags);
extern void audit_log_cap(struct audit_buffer *ab, char *prefix,
kernel_cap_t *cap);
extern void audit_log_name(struct audit_context *context,
diff --git a/kernel/auditsc.c b/kernel/auditsc.c
index b2d1f043f17f..d39a7fbaf944 100644
--- a/kernel/auditsc.c
+++ b/kernel/auditsc.c
@@ -1846,7 +1846,7 @@ void __audit_inode(struct filename *name, const struct dentry *dentry,
n->type = AUDIT_TYPE_NORMAL;
}
handle_path(dentry);
- audit_copy_inode(n, dentry, inode);
+ audit_copy_inode(n, dentry, inode, flags & AUDIT_INODE_NOREVAL);
}

void __audit_file(const struct file *file)
@@ -1947,7 +1947,7 @@ void __audit_inode_child(struct inode *parent,
n = audit_alloc_name(context, AUDIT_TYPE_PARENT);
if (!n)
return;
- audit_copy_inode(n, NULL, parent);
+ audit_copy_inode(n, NULL, parent, 0);
}

if (!found_child) {
@@ -1966,7 +1966,7 @@ void __audit_inode_child(struct inode *parent,
}

if (inode)
- audit_copy_inode(found_child, dentry, inode);
+ audit_copy_inode(found_child, dentry, inode, 0);
else
found_child->ino = AUDIT_INO_UNSET;
}

[Richard Guy Briggs]

Like 42d5e37654e4 ("audit: filter PATH records keyed on filesystem magic")

Any user or remote filesystem could become unavailable and effectively
block on a forced unmount.

-a always,exit -S umount2 -F key=umount2

Provide a method to ignore these user and remote filesystems to prevent
them from being impossible to unmount.

Extend the "AUDIT_FILTER_FS" filter that uses the field type
AUDIT_FSTYPE keying off the filesystem 4-octet hexadecimal magic
identifier to filter specific filesystems to cover audit_inode() to address
this blockage.

An example rule would look like:
-a never,filesystem -F fstype=0x517B -F key=ignore_smb
-a never,filesystem -F fstype=0x6969 -F key=ignore_nfs

Arguably the better way to address this issue is to disable auditing
processes that touch removable filesystems.

Please see the github issue tracker
https://github.com/linux-audit/audit-kernel/issues/100

Signed-off-by: Richard Guy Briggs <***@redhat.com>
---
kernel/auditsc.c | 23 +++++++++++++++++++++++
1 file changed, 23 insertions(+)

diff --git a/kernel/auditsc.c b/kernel/auditsc.c
index d39a7fbaf944..59d6d3fbc00e 100644
--- a/kernel/auditsc.c
+++ b/kernel/auditsc.c
@@ -1777,10 +1777,33 @@ void __audit_inode(struct filename *name, const struct dentry *dentry,
struct inode *inode = d_backing_inode(dentry);
struct audit_names *n;
bool parent = flags & AUDIT_INODE_PARENT;
+ struct audit_entry *e;
+ struct list_head *list = &audit_filter_list[AUDIT_FILTER_FS];
+ int i;

if (!context->in_syscall)
return;

+ rcu_read_lock();
+ if (!list_empty(list)) {
+ list_for_each_entry_rcu(e, list, list) {
+ for (i = 0; i < e->rule.field_count; i++) {
+ struct audit_field *f = &e->rule.fields[i];
+
+ if (f->type == AUDIT_FSTYPE) {
+ if (audit_comparator(inode->i_sb->s_magic,
+ f->op, f->val)) {
+ if (e->rule.action == AUDIT_NEVER) {
+ rcu_read_unlock();
+ return;
+ }
+ }
+ }
+ }
+ }
+ }
+ rcu_read_unlock();
+
if (!name)
goto out_alloc;

[Miklos Szeredi]

Post by Richard Guy Briggs
Don't fetch fcaps when umount2 is called with MNT_FORCE to avoid a
process hang while it waits for the missing resource to (possibly never)
re-appear.

The patch would be pretty good if the dependence on MNT_FORCE wasn't
added. As it is, it's buggy in more ways than one:

- It does the opposite of the above (i.e. skips fcaps *unless*
MNT_FORCE is set)
- sets LOOKUP_NO_REVAL from caller of path lookup, which is invalid
(LOOKUP_NO_REVAL is used only internally by path lookup)
- the fact that *_path_mountpoint_at() shouldn't touch the mount root
is independent of MNT_FORCE

I still don't quite understand what audit is trying to do here, but
apparently it's okay to skip getxattr in the MNT_FORCE case. So why
is it not okay to skip it in the non-MNT_FORCE case?

Thanks,
Miklos

[Richard Guy Briggs]

Post by Miklos Szeredi

Post by Richard Guy Briggs
Don't fetch fcaps when umount2 is called with MNT_FORCE to avoid a
process hang while it waits for the missing resource to (possibly never)
re-appear.

The patch would be pretty good if the dependence on MNT_FORCE wasn't
- It does the opposite of the above (i.e. skips fcaps *unless*
MNT_FORCE is set)

I agree it looks wrong now that I look at it. It turns out my test case
didn't trigger it properly since "umount -l" doesn't set MNT_FORCE while
it needs "-f" to do so. This is unacceptable since "-l" needs to work
in this situation too.

Post by Miklos Szeredi
- sets LOOKUP_NO_REVAL from caller of path lookup, which is invalid
(LOOKUP_NO_REVAL is used only internally by path lookup)

Fair enough. 949a852e46dd viro 2016-03-14 ("namei: teach lookup_slow()
to skip revalidate") needs a comment update.

Maybe my patch was interacting with this one and changing the behaviour I
expected.

Post by Miklos Szeredi
- the fact that *_path_mountpoint_at() shouldn't touch the mount root
is independent of MNT_FORCE

We don't entirely agree here since I'm still aiming for a best effort to
collect this information for the PATH record, but that may be misleading
at best.

Post by Miklos Szeredi
I still don't quite understand what audit is trying to do here, but
apparently it's okay to skip getxattr in the MNT_FORCE case. So why
is it not okay to skip it in the non-MNT_FORCE case?

The simple answer is that the audit PATH record format expects the four
cap_f* fields to be there and a best effort is being attempted to fill
in that information in an expected way with meaningful values. Perhaps
better to accept that it is unreasonable to expect any fcaps on any
umount operation and simply ignore those fields in the PATH record for
umount syscall events.

This is really a problem the audit folks have backed ourselves into.
This was introduced by 851f7ff56d9c ("This patch will print
cap_permitted and cap_inheritable data in the PATH...")
The fcaps are only really needed for the case of an event that changes
fcaps. In that case, the fcaps should have been added as a seperate
audit record to accompany this event as necessary, rather than included
in the PATH record that is shared by multiple event types, most of which
do not change the fcaps. There has been significant and ongoing effort
to normalize all audit record types so that they contain predictable
fields in a predictable order without any fields that swing in and out
since this makes userspace audit record parsers faster and more
reliable.

My preferred solution would be to in fact remove these four cap_f*
fields from the PATH record and put them in a new record that was only
included when the event is relevant and the values are non-zero.
This isn't an option with current upstream kernel audit devel policy.

Thanks Miklos for taking the time to provide feedback on this patch.

Post by Miklos Szeredi
Thanks,
Miklos

- RGB

--
Richard Guy Briggs <***@redhat.com>
Sr. S/W Engineer, Kernel Security, Base Operating Systems
Remote, Ottawa, Red Hat Canada
IRC: rgb, SunRaycer
Voice: +1.647.777.2635, Internal: (81) 32635

[Miklos Szeredi]

Post by Richard Guy Briggs
The simple answer is that the audit PATH record format expects the four
cap_f* fields to be there and a best effort is being attempted to fill
in that information in an expected way with meaningful values. Perhaps
better to accept that it is unreasonable to expect any fcaps on any
umount operation and simply ignore those fields in the PATH record for
umount syscall events.

When there's a mount there are in fact two objects belonging to the
exact same path, each having completely independent metadata: the
mount point and the root of the mount. For example:

stat /mnt
umount /mnt
stat /mnt

The first stat will show the root of the mount, the second one will
show the mount point.
Which one is the relevant for audit?

Not saying audit should be doing getxattr on any of them, just trying
to see more clearly.

Thanks,
Miklos

[Richard Guy Briggs]

Post by Miklos Szeredi

Post by Richard Guy Briggs
The simple answer is that the audit PATH record format expects the four
cap_f* fields to be there and a best effort is being attempted to fill
in that information in an expected way with meaningful values. Perhaps
better to accept that it is unreasonable to expect any fcaps on any
umount operation and simply ignore those fields in the PATH record for
umount syscall events.

When there's a mount there are in fact two objects belonging to the
exact same path, each having completely independent metadata: the
stat /mnt
umount /mnt
stat /mnt
The first stat will show the root of the mount, the second one will
show the mount point.
Which one is the relevant for audit?

It would be the root of the mount, the one that is visible to processes
in that mount namespace.

Obviously, once that mount has been unmounted, it would be the mount
point (no longer in use as such at that point) that is of interest.

On mounting, I'm guessing both would be of interest if the fcaps changed
for that process-visible path in that mount namespace, so this provides
an additional operation that would need recording aside from the case
of a simple attribute change.

Post by Miklos Szeredi
Not saying audit should be doing getxattr on any of them, just trying
to see more clearly.
Thanks,
Miklos

- RGB

--
Richard Guy Briggs <***@redhat.com>
Sr. S/W Engineer, Kernel Security, Base Operating Systems
Remote, Ottawa, Red Hat Canada
IRC: rgb, SunRaycer
Voice: +1.647.777.2635, Internal: (81) 32635

[Steve Grubb]

Post by Richard Guy Briggs

Post by Miklos Szeredi

Post by Richard Guy Briggs
The simple answer is that the audit PATH record format expects the four
cap_f* fields to be there and a best effort is being attempted to fill
in that information in an expected way with meaningful values. Perhaps
better to accept that it is unreasonable to expect any fcaps on any
umount operation and simply ignore those fields in the PATH record for
umount syscall events.

When there's a mount there are in fact two objects belonging to the
exact same path, each having completely independent metadata: the
stat /mnt
umount /mnt
stat /mnt
The first stat will show the root of the mount, the second one will
show the mount point.
Which one is the relevant for audit?

It would be the root of the mount, the one that is visible to processes
in that mount namespace.
Obviously, once that mount has been unmounted, it would be the mount
point (no longer in use as such at that point) that is of interest.
On mounting, I'm guessing both would be of interest if the fcaps changed
for that process-visible path in that mount namespace, so this provides
an additional operation that would need recording aside from the case
of a simple attribute change.

fcaps are on files. Mountpoints are directories. Would fcaps changes be
possible?

-Steve

Post by Richard Guy Briggs

Post by Miklos Szeredi
Not saying audit should be doing getxattr on any of them, just trying
to see more clearly.
Thanks,
Miklos

- RGB
--
Sr. S/W Engineer, Kernel Security, Base Operating Systems
Remote, Ottawa, Red Hat Canada
IRC: rgb, SunRaycer
Voice: +1.647.777.2635, Internal: (81) 32635