Skip to contentSkip to conversation

bpf: local storage helpers should check nullness of owner ptr passed

PATCH In review KP Singh 4 participants 7 messages Jan 7, 2021 – Jan 11, 2021 BPF

lore
Share & tools
lore

KP Singh

author
#0 lore raw

PATCH 1/1 bpf: local storage helpers should check nullness of owner ptr passed

The verifier allows ARG_PTR_TO_BTF_ID helper arguments to be NULL, so helper implementations need to check this before dereferencing them. This was already fixed for the socket storage helpers but not for task and inode.

The issue can be reproduced by attaching an LSM program to inode_rename hook (called when moving files) which tries to get the inode of the new file without checking for its nullness and then trying to move an existing file to a new path:

mv existing_file new_file_does_not_exist

The report including the sample program and the steps for reproducing the bug:

https://lore.kernel.org/bpf/CANaYP3HWkH91SN=wTNO9FL_2ztHfqcXKX38SSE-JJ2voh+vssw@mail.gmail.com

Fixes 4cf1bc1f1045 ("bpf: Implement task local storage")
Fixes 8ea636848aca ("bpf: Implement bpf_local_storage for inodes")
---
 kernel/bpf/bpf_inode_storage.c | 5 ++++-
 kernel/bpf/bpf_task_storage.c  | 5 ++++-
 2 files changed, 8 insertions(+), 2 deletions(-)
kernel/bpf/bpf_inode_storage.c +4 -1
index 6edff97ad594..dbc1dbdd2cbf 100644
@@ -176,7 +176,7 @@ BPF_CALL_4(bpf_inode_storage_get, struct bpf_map *, map, struct inode *, inode,
 	 * bpf_local_storage_update expects the owner to have a
 	 * valid storage pointer.
 	 */
-	if (!inode_storage_ptr(inode))
+	if (!inode || !inode_storage_ptr(inode))
 		return (unsigned long)NULL;
 
 	sdata = inode_storage_lookup(inode, map, true);
@@ -200,6 +200,9 @@ BPF_CALL_4(bpf_inode_storage_get, struct bpf_map *, map, struct inode *, inode,
 BPF_CALL_2(bpf_inode_storage_delete,
 	   struct bpf_map *, map, struct inode *, inode)
 {
+	if (!inode)
+		return -EINVAL;
+
 	/* This helper must only called from where the inode is gurranteed
 	 * to have a refcount and cannot be freed.
 	 */
kernel/bpf/bpf_task_storage.c +4 -1
index 4ef1959a78f2..e0da0258b732 100644
@@ -218,7 +218,7 @@ BPF_CALL_4(bpf_task_storage_get, struct bpf_map *, map, struct task_struct *,
 	 * bpf_local_storage_update expects the owner to have a
 	 * valid storage pointer.
 	 */
-	if (!task_storage_ptr(task))
+	if (!task || !task_storage_ptr(task))
 		return (unsigned long)NULL;
 
 	sdata = task_storage_lookup(task, map, true);
@@ -243,6 +243,9 @@ BPF_CALL_4(bpf_task_storage_get, struct bpf_map *, map, struct task_struct *,
 BPF_CALL_2(bpf_task_storage_delete, struct bpf_map *, map, struct task_struct *,
 	   task)
 {
+	if (!task)
+		return -EINVAL;
+
 	/* This helper must only be called from places where the lifetime of the task
 	 * is guaranteed. Either by being refcounted or by being protected
 	 * by an RCU read-side critical section.
Signature · 1 line
2.30.0.284.gd98b1dd5eaa7-goog
#1 lore raw
Quoted 6 lines · KP Singh

The verifier allows ARG_PTR_TO_BTF_ID helper arguments to be NULL, so helper implementations need to check this before dereferencing them. This was already fixed for the socket storage helpers but not for task and inode.

Quoted 6 lines

The issue can be reproduced by attaching an LSM program to inode_rename hook (called when moving files) which tries to get the inode of the new file without checking for its nullness and then trying to move an existing file to a new path:

mv existing_file new_file_does_not_exist

Seems like it's simple to write a selftest for this then?

Quoted 20 lines

The report including the sample program and the steps for reproducing the bug:

https://lore.kernel.org/bpf/CANaYP3HWkH91SN=wTNO9FL_2ztHfqcXKX38SSE-JJ2voh+vssw@mail.gmail.com

Fixes 4cf1bc1f1045 ("bpf: Implement task local storage")
Fixes 8ea636848aca ("bpf: Implement bpf_local_storage for inodes")
---
 kernel/bpf/bpf_inode_storage.c | 5 ++++-
 kernel/bpf/bpf_task_storage.c  | 5 ++++-
 2 files changed, 8 insertions(+), 2 deletions(-)
kernel/bpf/bpf_inode_storage.c +0 -0
index 6edff97ad594..dbc1dbdd2cbf 100644

kernel/bpf/bpf_inode_storage.c BPF_CALL_4(bpf_inode_storage_get, struct bpf_map *, map, struct inode *, inode, · line 179

Quoted 5 lines
kernel/bpf/bpf_inode_storage.c +1 -1
@@ -176,7 +176,7 @@ BPF_CALL_4(bpf_inode_storage_get, struct bpf_map *, map, struct inode *, inode,
         * bpf_local_storage_update expects the owner to have a
         * valid storage pointer.
         */
-       if (!inode_storage_ptr(inode))
+       if (!inode || !inode_storage_ptr(inode))

would it be bad to move !inode check inside inode_storage_ptr itself? same for task_storage_ptr() below.

kernel/bpf/bpf_inode_storage.c BPF_CALL_4(bpf_inode_storage_get, struct bpf_map *, map, struct inode *, inode, · line 206

Quoted 11 lines
kernel/bpf/bpf_inode_storage.c +3 -0
                return (unsigned long)NULL;
 
        sdata = inode_storage_lookup(inode, map, true);
@@ -200,6 +200,9 @@ BPF_CALL_4(bpf_inode_storage_get, struct bpf_map *, map, struct inode *, inode,
 BPF_CALL_2(bpf_inode_storage_delete,
           struct bpf_map *, map, struct inode *, inode)
 {
+       if (!inode)
+               return -EINVAL;
+
        /* This helper must only called from where the inode is gurranteed

Gmail highlights a typo in "gurranteed" ;)

Quoted 28 lines
kernel/bpf/bpf_inode_storage.c +0 -0
         * to have a refcount and cannot be freed.
         */
kernel/bpf/bpf_task_storage.c +4 -2
index 4ef1959a78f2..e0da0258b732 100644
@@ -218,7 +218,7 @@ BPF_CALL_4(bpf_task_storage_get, struct bpf_map *, map, struct task_struct *,
         * bpf_local_storage_update expects the owner to have a
         * valid storage pointer.
         */
-       if (!task_storage_ptr(task))
+       if (!task || !task_storage_ptr(task))
                return (unsigned long)NULL;
 
        sdata = task_storage_lookup(task, map, true);
@@ -243,6 +243,9 @@ BPF_CALL_4(bpf_task_storage_get, struct bpf_map *, map, struct task_struct *,
 BPF_CALL_2(bpf_task_storage_delete, struct bpf_map *, map, struct task_struct *,
           task)
 {
+       if (!task)
+               return -EINVAL;
+
        /* This helper must only be called from places where the lifetime of the task
         * is guaranteed. Either by being refcounted or by being protected
         * by an RCU read-side critical section.
--

2.30.0.284.gd98b1dd5eaa7-goog

#2 lore raw
Quoted 32 lines · Andrii Nakryiko
Quoted 6 lines · KP Singh

The verifier allows ARG_PTR_TO_BTF_ID helper arguments to be NULL, so helper implementations need to check this before dereferencing them. This was already fixed for the socket storage helpers but not for task and inode.

Quoted 6 lines

The issue can be reproduced by attaching an LSM program to inode_rename hook (called when moving files) which tries to get the inode of the new file without checking for its nullness and then trying to move an existing file to a new path:

mv existing_file new_file_does_not_exist

Seems like it's simple to write a selftest for this then?

Quoted 15 lines

The report including the sample program and the steps for reproducing the bug:

https://lore.kernel.org/bpf/CANaYP3HWkH91SN=wTNO9FL_2ztHfqcXKX38SSE-JJ2voh+vssw@mail.gmail.com

Fixes 4cf1bc1f1045 ("bpf: Implement task local storage")
Fixes 8ea636848aca ("bpf: Implement bpf_local_storage for inodes")
---
 kernel/bpf/bpf_inode_storage.c | 5 ++++-
 kernel/bpf/bpf_task_storage.c  | 5 ++++-
 2 files changed, 8 insertions(+), 2 deletions(-)
Quoted 13 lines

kernel/bpf/bpf_inode_storage.c BPF_CALL_4(bpf_inode_storage_get, struct bpf_map *, map, struct inode *, inode, · line 179

Quoted 10 lines
kernel/bpf/bpf_inode_storage.c +1 -1
index 6edff97ad594..dbc1dbdd2cbf 100644
@@ -176,7 +176,7 @@ BPF_CALL_4(bpf_inode_storage_get, struct bpf_map *, map, struct inode *, inode,
         * bpf_local_storage_update expects the owner to have a
         * valid storage pointer.
         */
-       if (!inode_storage_ptr(inode))
+       if (!inode || !inode_storage_ptr(inode))

would it be bad to move !inode check inside inode_storage_ptr itself? same for task_storage_ptr() below.

And for deletes, inode_storage_delete calls into inode_storage_lookup(), which also seems like a reasonable place to check for null? Even better, inode_storage_lookup() shares logic with inode_storage_ptr(), so if we make sure that all code calls inode_storage_ptr(), then we need to check for NULL just in inode_storage_ptr().

I totally might be missing some subtleties, of course.

Quoted 43 lines

kernel/bpf/bpf_inode_storage.c BPF_CALL_4(bpf_inode_storage_get, struct bpf_map *, map, struct inode *, inode, · line 206

Quoted 12 lines
kernel/bpf/bpf_inode_storage.c +3 -0
                return (unsigned long)NULL;
 
        sdata = inode_storage_lookup(inode, map, true);
@@ -200,6 +200,9 @@ BPF_CALL_4(bpf_inode_storage_get, struct bpf_map *, map, struct inode *, inode,
 BPF_CALL_2(bpf_inode_storage_delete,
           struct bpf_map *, map, struct inode *, inode)
 {
+       if (!inode)
+               return -EINVAL;
+
        /* This helper must only called from where the inode is gurranteed

Gmail highlights a typo in "gurranteed" ;)

Quoted 28 lines
kernel/bpf/bpf_inode_storage.c +0 -0
         * to have a refcount and cannot be freed.
         */
kernel/bpf/bpf_task_storage.c +4 -2
index 4ef1959a78f2..e0da0258b732 100644
@@ -218,7 +218,7 @@ BPF_CALL_4(bpf_task_storage_get, struct bpf_map *, map, struct task_struct *,
         * bpf_local_storage_update expects the owner to have a
         * valid storage pointer.
         */
-       if (!task_storage_ptr(task))
+       if (!task || !task_storage_ptr(task))
                return (unsigned long)NULL;
 
        sdata = task_storage_lookup(task, map, true);
@@ -243,6 +243,9 @@ BPF_CALL_4(bpf_task_storage_get, struct bpf_map *, map, struct task_struct *,
 BPF_CALL_2(bpf_task_storage_delete, struct bpf_map *, map, struct task_struct *,
           task)
 {
+       if (!task)
+               return -EINVAL;
+
        /* This helper must only be called from places where the lifetime of the task
         * is guaranteed. Either by being refcounted or by being protected
         * by an RCU read-side critical section.
--

2.30.0.284.gd98b1dd5eaa7-goog

author
#3 lore raw
Quoted 11 lines · Andrii Nakryiko
Quoted 8 lines · Andrii Nakryiko
Quoted 6 lines · KP Singh

The verifier allows ARG_PTR_TO_BTF_ID helper arguments to be NULL, so helper implementations need to check this before dereferencing them. This was already fixed for the socket storage helpers but not for task and inode.

Quoted 8 lines
Quoted 8 lines
Quoted 6 lines

The issue can be reproduced by attaching an LSM program to inode_rename hook (called when moving files) which tries to get the inode of the new file without checking for its nullness and then trying to move an existing file to a new path:

mv existing_file new_file_does_not_exist

Seems like it's simple to write a selftest for this then?

Sure, I will send in a separate patch for selftest and also for the typo.

Quoted 30 lines
Quoted 16 lines
Quoted 16 lines

The report including the sample program and the steps for reproducing the bug:

https://lore.kernel.org/bpf/CANaYP3HWkH91SN=wTNO9FL_2ztHfqcXKX38SSE-JJ2voh+vssw@mail.gmail.com

Fixes 4cf1bc1f1045 ("bpf: Implement task local storage")
Fixes 8ea636848aca ("bpf: Implement bpf_local_storage for inodes")
---
 kernel/bpf/bpf_inode_storage.c | 5 ++++-
 kernel/bpf/bpf_task_storage.c  | 5 ++++-
 2 files changed, 8 insertions(+), 2 deletions(-)
Quoted 13 lines

kernel/bpf/bpf_inode_storage.c BPF_CALL_4(bpf_inode_storage_get, struct bpf_map *, map, struct inode *, inode, · line 179

Quoted 10 lines
kernel/bpf/bpf_inode_storage.c +1 -1
index 6edff97ad594..dbc1dbdd2cbf 100644
@@ -176,7 +176,7 @@ BPF_CALL_4(bpf_inode_storage_get, struct bpf_map *, map, struct inode *, inode,
         * bpf_local_storage_update expects the owner to have a
         * valid storage pointer.
         */
-       if (!inode_storage_ptr(inode))
+       if (!inode || !inode_storage_ptr(inode))

would it be bad to move !inode check inside inode_storage_ptr itself? same for task_storage_ptr() below.

Quoted 8 lines

And for deletes, inode_storage_delete calls into inode_storage_lookup(), which also seems like a reasonable place to check for null? Even better, inode_storage_lookup() shares logic with inode_storage_ptr(), so if we make sure that all code calls inode_storage_ptr(), then we need to check for NULL just in inode_storage_ptr().

I totally might be missing some subtleties, of course.

All these are good candidates for nullness checks too (I also thought about bpf_inode and
bpf_task having a null check).

I kind of like the explicit check / input validation in the helper before it does anything with the pointer.
It's a reminder that the value cannot be assumed to be NULL.

FWIW, we do a similar explicit check in the socket storage code as well.

[...]

Quoted 2 lines
Quoted 2 lines

Gmail highlights a typo in "gurranteed" ;)

Thanks, and thanks gmail ;)

Quoted 29 lines
Quoted 29 lines
Quoted 29 lines
kernel/bpf/bpf_inode_storage.c +0 -0
         * to have a refcount and cannot be freed.
         */
kernel/bpf/bpf_task_storage.c +4 -2
index 4ef1959a78f2..e0da0258b732 100644
@@ -218,7 +218,7 @@ BPF_CALL_4(bpf_task_storage_get, struct bpf_map *, map, struct task_struct *,
         * bpf_local_storage_update expects the owner to have a
         * valid storage pointer.
         */
-       if (!task_storage_ptr(task))
+       if (!task || !task_storage_ptr(task))
                return (unsigned long)NULL;
 
        sdata = task_storage_lookup(task, map, true);
@@ -243,6 +243,9 @@ BPF_CALL_4(bpf_task_storage_get, struct bpf_map *, map, struct task_struct *,
 BPF_CALL_2(bpf_task_storage_delete, struct bpf_map *, map, struct task_struct *,
           task)
 {
+       if (!task)
+               return -EINVAL;
+
        /* This helper must only be called from places where the lifetime of the task
         * is guaranteed. Either by being refcounted or by being protected
         * by an RCU read-side critical section.
--

2.30.0.284.gd98b1dd5eaa7-goog

maintainer
#4 lore raw
Quoted 16 lines · KP Singh
Quoted 14 lines · Andrii Nakryiko
Quoted 12 lines · Andrii Nakryiko
Quoted 11 lines · KP Singh

The verifier allows ARG_PTR_TO_BTF_ID helper arguments to be NULL, so helper implementations need to check this before dereferencing them. This was already fixed for the socket storage helpers but not for task and inode.

The issue can be reproduced by attaching an LSM program to inode_rename hook (called when moving files) which tries to get the inode of the new file without checking for its nullness and then trying to move an existing file to a new path:

Quoted 5 lines
Quoted 3 lines
Quoted 3 lines
Quoted 1 line

mv existing_file new_file_does_not_exist

Seems like it's simple to write a selftest for this then?

Sure, I will send in a separate patch for selftest and also for the typo.

If it's small or trivial to add a selftest for the fix, I'd suggest to add it as part of this series for 'ease of logistics' as it would otherwise be a bit odd to i) either have a stand-alone patch against bpf tree with just a selftest or ii) having to wait until bpf syncs into bpf-next and then send one against bpf-next where for the latter there's risk that it gets forgotten in meantime as it might take a while.

Thanks,
Daniel

author
#5 lore raw
Quoted 22 lines · Daniel Borkmann
Quoted 20 lines · KP Singh
Quoted 17 lines · Andrii Nakryiko
Quoted 15 lines · Andrii Nakryiko
Quoted 6 lines · KP Singh

The verifier allows ARG_PTR_TO_BTF_ID helper arguments to be NULL, so helper implementations need to check this before dereferencing them. This was already fixed for the socket storage helpers but not for task and inode.

Quoted 6 lines

The issue can be reproduced by attaching an LSM program to inode_rename hook (called when moving files) which tries to get the inode of the new file without checking for its nullness and then trying to move an existing file to a new path:

mv existing_file new_file_does_not_exist

Seems like it's simple to write a selftest for this then?

Quoted 8 lines
Quoted 1 line

Sure, I will send in a separate patch for selftest and also for the typo.

If it's small or trivial to add a selftest for the fix, I'd suggest to add it as part of this series for 'ease of logistics' as it would otherwise be a bit odd to i) either have a stand-alone patch against bpf tree with just a selftest or ii) having to wait until bpf syncs into bpf-next and then send one against bpf-next where for the latter there's risk that it gets forgotten in meantime as it might take a while.

Sounds good, I completely missed the fact the fix would take some time to sync from bpf to bpf-next and until then the selftest will keep crashing.

I updated the selftest and will send in a v2 and while we are at it, will also fix
the typo Andrii found.

- KP

Quoted 3 lines

Thanks,
Daniel