Skip to contentSkip to conversation

fix sys_unshare()+SEM_UNDO: perform an implicit CLONE_SYSVSEM in CLONE_NEWIPC

PATCH In review Manfred Spraul 2 participants 5 messages Apr 13, 2008 – Apr 15, 2008

Subsystems (3)
lore
Share & tools
lore
#0 lore raw

PATCH 2/2 fix sys_unshare()+SEM_UNDO: perform an implicit CLONE_SYSVSEM in CLONE_NEWIPC

sys_unshare(CLONE_NEWIPC) doesn't handle the undo lists properly, this can cause a kernel memory corruption. CLONE_NEWIPC must detach from the existing undo lists.
Fix, part 2: perform an implicit CLONE_SYSVSEM in CLONE_NEWIPC. CLONE_NEWIPC creates a new IPC namespace, the task cannot access the existing semaphore arrays after the unshare syscall. Thus the task can/must detach from the existing undo list entries, too.

This fixes the kernel corruption, because it makes it impossible that undo records from two different namespaces are in sysvsem.undo_list.

I'm not sure who's the right maintainer, Andrew, could you merge it?

---
 kernel/fork.c |    6 +++++-
 1 files changed, 5 insertions(+), 1 deletions(-)
kernel/fork.c +5 -1
index 7f242b0..6f956a7 100644
@@ -1735,9 +1735,13 @@ asmlinkage long sys_unshare(unsigned long unshare_flags)
 		goto bad_unshare_cleanup_semundo;
 
 	if (new_fs ||  new_mm || new_fd || new_ulist || new_nsproxy) {
-		if (unshare_flags & CLONE_SYSVSEM) {
+		if (unshare_flags & (CLONE_NEWIPC|CLONE_SYSVSEM)) {
 			/*
 			 * CLONE_SYSVSEM is equivalent to sys_exit().
+			 *
+			 * CLONE_NEWIPC must also detach from the undolist: after switching
+			 * to a new ipc namespace, the semaphore arrays from the old
+			 * namespace are unreachable.
 			 */
 			exit_sem(current);
 		}
Signature · 1 line
1.5.4.1
#1 lore raw
Quoted 10 lines · Manfred Spraul

sys_unshare(CLONE_NEWIPC) doesn't handle the undo lists properly, this can cause a kernel memory corruption. CLONE_NEWIPC must detach from the existing undo lists.
Fix, part 2: perform an implicit CLONE_SYSVSEM in CLONE_NEWIPC. CLONE_NEWIPC creates a new IPC namespace, the task cannot access the existing semaphore arrays after the unshare syscall. Thus the task can/must detach from the existing undo list entries, too.

This fixes the kernel corruption, because it makes it impossible that undo records from two different namespaces are in sysvsem.undo_list.

So this was never an issue with clone(CLONE_NEWIPC|CLONE_SYSVSEM), which should have had the same result as unshare(CLONE_NEWIPC&~CLONE_SYSVSEM)?

Quoted 28 lines

I'm not sure who's the right maintainer, Andrew, could you merge it?

---
 kernel/fork.c |    6 +++++-
 1 files changed, 5 insertions(+), 1 deletions(-)
kernel/fork.c +5 -1
index 7f242b0..6f956a7 100644
@@ -1735,9 +1735,13 @@ asmlinkage long sys_unshare(unsigned long unshare_flags)
 		goto bad_unshare_cleanup_semundo;
 
 	if (new_fs ||  new_mm || new_fd || new_ulist || new_nsproxy) {
-		if (unshare_flags & CLONE_SYSVSEM) {
+		if (unshare_flags & (CLONE_NEWIPC|CLONE_SYSVSEM)) {
 			/*
 			 * CLONE_SYSVSEM is equivalent to sys_exit().
+			 *
+			 * CLONE_NEWIPC must also detach from the undolist: after switching
+			 * to a new ipc namespace, the semaphore arrays from the old
+			 * namespace are unreachable.
 			 */
 			exit_sem(current);
 		}
Signature · 1 line
1.5.4.1
#2 lore raw
Quoted 10 lines · Serge E. Hallyn
Quoted 8 lines · Manfred Spraul

sys_unshare(CLONE_NEWIPC) doesn't handle the undo lists properly, this can cause a kernel memory corruption. CLONE_NEWIPC must detach from the existing undo lists.
Fix, part 2: perform an implicit CLONE_SYSVSEM in CLONE_NEWIPC. CLONE_NEWIPC creates a new IPC namespace, the task cannot access the existing semaphore arrays after the unshare syscall. Thus the task can/must detach from the existing undo list entries, too.

Quoted 8 lines
Quoted 3 lines

This fixes the kernel corruption, because it makes it impossible that undo records from two different namespaces are in sysvsem.undo_list.

So this was never an issue with clone(CLONE_NEWIPC|CLONE_SYSVSEM), which should have had the same result as unshare(CLONE_NEWIPC&~CLONE_SYSVSEM)?

Actually, the story is slightly different:
unshare(CLONE_NEWIPC|CLONE_SYSVSEM) returns -EINVAL right now.

Thus all apps right now call unshare(CLONE_NEWIPC|&~CLONE_SYSVSEM).
This combination doesn't make much sense. Even worse - it easily causes a kernel oops. Thus my fix is twofold:
- add support for unshare(CLONE_SYSVSEM).
- implicitely add CLONE_SYSVSEM to all calls that set CLONE_NEWIPC.

It's not really pretty: If a pivot_namespace syscall is ever added, then CLONE_NEWIPC&~CLONE_SYSVSEM would make sense again. What do you think? Can we break backward compatibility and add
if ( (unshare_flags & CLONE_NEWIPC) && !(unshare_flags &
CLONE_SYSVSEM) )

      return -EINVAL;

into sys_unshare()?
I have decided against that, it breaks the current ABI.
And we gain virtually nothing - most if not all unshare users will be single threaded apps that do not use sysvsem at all, and even most sysvsem users do not use SEM_UNDO.

--
Manfred

#3 lore raw
Quoted 17 lines · Manfred Spraul
Quoted 16 lines · Serge E. Hallyn
Quoted 13 lines · Manfred Spraul

sys_unshare(CLONE_NEWIPC) doesn't handle the undo lists properly, this can
cause a kernel memory corruption. CLONE_NEWIPC must detach from the existing
undo lists.
Fix, part 2: perform an implicit CLONE_SYSVSEM in CLONE_NEWIPC. CLONE_NEWIPC creates a new IPC namespace, the task cannot access the existing semaphore arrays after the unshare syscall. Thus the task can/must detach from the existing undo list entries, too.

This fixes the kernel corruption, because it makes it impossible that undo records from two different namespaces are in sysvsem.undo_list.

Quoted 6 lines
Quoted 4 lines

So this was never an issue with clone(CLONE_NEWIPC|CLONE_SYSVSEM), which should have had the same result as unshare(CLONE_NEWIPC&~CLONE_SYSVSEM)?

Actually, the story is slightly different:
unshare(CLONE_NEWIPC|CLONE_SYSVSEM) returns -EINVAL right now.

Right, but it's feasible with clone (where CLONE_SYSVSEM has the opposite meaning of with unshare) right? According to the comment, CLONE_SYSVSEM was disabled mainly out of fear that it would be confusing to enable it since it does have the inverse meaning in unshare as in clone.

Quoted 6 lines

Thus all apps right now call unshare(CLONE_NEWIPC|&~CLONE_SYSVSEM). This combination doesn't make much sense. Even worse - it easily causes a kernel oops.
Thus my fix is twofold:
- add support for unshare(CLONE_SYSVSEM).
- implicitely add CLONE_SYSVSEM to all calls that set CLONE_NEWIPC.

Right but your fix ignores the fact that you can achieve the same result as unshare(CLONE_NEWIPC&~CLONE_SYSVSEM) by doing clone(CLONE_NEWIPC|CLONE_SYSVSEM).

Quoted 7 lines

It's not really pretty: If a pivot_namespace syscall is ever added, then CLONE_NEWIPC&~CLONE_SYSVSEM would make sense again.
What do you think? Can we break backward compatibility and add
if ( (unshare_flags & CLONE_NEWIPC) && !(unshare_flags & CLONE_SYSVSEM)
)

      return -EINVAL;

into sys_unshare()?

I don't think we need to, I think what you're doing makes perfect sense. If you asked for CLONE_NEWIPC, then you wanted to do CLONE_SYSVSEM (uh, or, in clone parlance, !CLONE_SYSVSEM :)

Quoted 4 lines

I have decided against that, it breaks the current ABI.
And we gain virtually nothing - most if not all unshare users will be single threaded apps that do not use sysvsem at all, and even most sysvsem users do not use SEM_UNDO.

And most importantly sharing your semundo list but not your sems with your parent is silly!

-serge

#4 lore raw
Quoted 8 lines · Serge E. Hallyn
Quoted 7 lines

Thus all apps right now call unshare(CLONE_NEWIPC|&~CLONE_SYSVSEM).
This combination doesn't make much sense. Even worse - it easily causes a kernel oops. Thus my fix is twofold:
- add support for unshare(CLONE_SYSVSEM).
- implicitely add CLONE_SYSVSEM to all calls that set CLONE_NEWIPC.

Quoted 4 lines

Right but your fix ignores the fact that you can achieve the same result as unshare(CLONE_NEWIPC&~CLONE_SYSVSEM) by doing clone(CLONE_NEWIPC|CLONE_SYSVSEM).

Duh, I overlooked that part. You are right.
Hmm - what's the best fix? Implicitely add CLONE_SYSVSEM or return -EINVAL if CLONE_SYSVSEM is not set? I don't care.
Could you write a patch? I probably won't have enough time until the next weekend.

Quoted 6 lines
Quoted 3 lines

I have decided against that, it breaks the current ABI.
And we gain virtually nothing - most if not all unshare users will be single threaded apps that do not use sysvsem at all, and even most sysvsem users do not use SEM_UNDO.

And most importantly sharing your semundo list but not your sems with your parent is silly!

It's only 99% silly: a task might want to have not-yet undone changes in its parent's namespace. As soon as the task exit, they are undone. Probably the most complex way to implement wait4().

--
Manfred