Skip to contentSkip to conversation

Fix for Integrity subsystem null pointer deref

GIT PULL James Morris 5 participants 12 messages Oct 29, 2014 – Nov 8, 2014

lore
Share & tools
lore
#0 lore raw

These changes fix a bug in xattr handling, where the evm and ima inode_setxattr() functions do not check for empty xattrs being passed from userspace (leading to user-triggerable null pointer dereferences).

Please pull.

The following changes since commit 9f76628da20f96a179ca62b504886f99ecc29223:

Merge branch 'for-3.18' of git://linux-nfs.org/~bfields/linux (2014-10-28 13:32:06 -0700)

are available in the git repository at:

git://git.kernel.org/pub/scm/linux/kernel/git/jmorris/linux-security.git for-linus

Dmitry Kasatkin (2):
ima: check xattr value length and type in the ima_inode_setxattr()
evm: check xattr value length and type in evm_inode_setxattr()

James Morris (1):
Merge branch 'for-linus' of git://git.kernel.org/.../zohar/linux-integrity into for-linus

 security/integrity/evm/evm_main.c     |    9 ++++++---
 security/integrity/ima/ima_appraise.c |    2 ++
 security/integrity/integrity.h        |    1 +
 3 files changed, 9 insertions(+), 3 deletions(-)
#1 lore raw
Quoted 9 lines · James Morris

These changes fix a bug in xattr handling, where the evm and ima inode_setxattr() functions do not check for empty xattrs being passed from userspace (leading to user-triggerable null pointer dereferences).

Please pull.

The following changes since commit 9f76628da20f96a179ca62b504886f99ecc29223:

Quoted 8 lines

Merge branch 'for-3.18' of git://linux-nfs.org/~bfields/linux (2014-10-28 13:32:06 -0700)

are available in the git repository at:

git://git.kernel.org/pub/scm/linux/kernel/git/jmorris/linux-security.git for-linus

Dmitry Kasatkin (2):
ima: check xattr value length and type in the ima_inode_setxattr()

I haven't read this one, but:

Quoted 1 line

evm: check xattr value length and type in evm_inode_setxattr()

const struct evm_ima_xattr_data *xattr_data = xattr_value;
- if ((strcmp(xattr_name, XATTR_NAME_EVM) == 0)
- && (xattr_data->type == EVM_XATTR_HMAC))
- return -EPERM;
+ if (strcmp(xattr_name, XATTR_NAME_EVM) == 0) {
+ if (!xattr_value_len)
+ return -EINVAL;
+ if (xattr_data->type != EVM_IMA_XATTR_DIGSIG)
+ return -EPERM;
+ }

Huh? (Sorry about severe whitespace damage.)

Shouldn't there be something like if (xattr_value_len < sizeof(struct evm_ima_xattr_data)) return -EINVAL?

--Andy

Quoted 8 lines

James Morris (1):
Merge branch 'for-linus' of git://git.kernel.org/.../zohar/linux-integrity into for-linus

 security/integrity/evm/evm_main.c     |    9 ++++++---
 security/integrity/ima/ima_appraise.c |    2 ++
 security/integrity/integrity.h        |    1 +
 3 files changed, 9 insertions(+), 3 deletions(-)
Signature · 2 lines
Andy Lutomirski
AMA Capital Management, LLC
#2 lore raw
Quoted 34 lines · Andy Lutomirski
Quoted 9 lines · James Morris

These changes fix a bug in xattr handling, where the evm and ima inode_setxattr() functions do not check for empty xattrs being passed from userspace (leading to user-triggerable null pointer dereferences).

Please pull.

The following changes since commit 9f76628da20f96a179ca62b504886f99ecc29223:

Quoted 8 lines

Merge branch 'for-3.18' of git://linux-nfs.org/~bfields/linux (2014-10-28 13:32:06 -0700)

are available in the git repository at:

git://git.kernel.org/pub/scm/linux/kernel/git/jmorris/linux-security.git for-linus

Dmitry Kasatkin (2):
ima: check xattr value length and type in the ima_inode_setxattr()

I haven't read this one, but:

Quoted 1 line

evm: check xattr value length and type in evm_inode_setxattr()

const struct evm_ima_xattr_data *xattr_data = xattr_value;
- if ((strcmp(xattr_name, XATTR_NAME_EVM) == 0)
- && (xattr_data->type == EVM_XATTR_HMAC))
- return -EPERM;
+ if (strcmp(xattr_name, XATTR_NAME_EVM) == 0) {
+ if (!xattr_value_len)
+ return -EINVAL;
+ if (xattr_data->type != EVM_IMA_XATTR_DIGSIG)
+ return -EPERM;
+ }

Quoted 4 lines

Huh? (Sorry about severe whitespace damage.)

Shouldn't there be something like if (xattr_value_len < sizeof(struct evm_ima_xattr_data)) return -EINVAL?

Prior to commit 2fb1c9a "evm: prohibit userspace writing 'security.evm' HMAC value", a process with CAP_SYS_ADMIN could write either an HMAC or signature. As the HMAC key should only be known to the kernel, only signatures are now allowed. Instead of "struct evm_ima_xattr_data", the code should reflect this change and use "struct signature_v2_hdr". We'll clean up this code for the next release. For now, this patch prevents the oops.

thanks,

Mimi

#3 lore raw
Quoted 41 lines · Mimi Zohar
Quoted 34 lines · Andy Lutomirski
Quoted 9 lines · James Morris

These changes fix a bug in xattr handling, where the evm and ima inode_setxattr() functions do not check for empty xattrs being passed from userspace (leading to user-triggerable null pointer dereferences).

Please pull.

The following changes since commit 9f76628da20f96a179ca62b504886f99ecc29223:

Quoted 8 lines

Merge branch 'for-3.18' of git://linux-nfs.org/~bfields/linux (2014-10-28 13:32:06 -0700)

are available in the git repository at:

git://git.kernel.org/pub/scm/linux/kernel/git/jmorris/linux-security.git for-linus

Dmitry Kasatkin (2):
ima: check xattr value length and type in the ima_inode_setxattr()

I haven't read this one, but:

Quoted 1 line

evm: check xattr value length and type in evm_inode_setxattr()

const struct evm_ima_xattr_data *xattr_data = xattr_value;
- if ((strcmp(xattr_name, XATTR_NAME_EVM) == 0)
- && (xattr_data->type == EVM_XATTR_HMAC))
- return -EPERM;
+ if (strcmp(xattr_name, XATTR_NAME_EVM) == 0) {
+ if (!xattr_value_len)
+ return -EINVAL;
+ if (xattr_data->type != EVM_IMA_XATTR_DIGSIG)
+ return -EPERM;
+ }

Quoted 4 lines

Huh? (Sorry about severe whitespace damage.)

Shouldn't there be something like if (xattr_value_len < sizeof(struct evm_ima_xattr_data)) return -EINVAL?

Quoted 8 lines

Prior to commit 2fb1c9a "evm: prohibit userspace writing 'security.evm' HMAC value", a process with CAP_SYS_ADMIN could write either an HMAC or signature. As the HMAC key should only be known to the kernel, only signatures are now allowed. Instead of "struct evm_ima_xattr_data", the code should reflect this change and use "struct signature_v2_hdr". We'll clean up this code for the next release. For now, this patch prevents the oops.

I have no idea what the semantics are. All I'm saying is that it looks like the code still accesses memory past the end of the buffer. The buffer isn't a null pointer, so the symptom is different, but it may still be a security bug.

--Andy

Quoted 4 lines

thanks,

Mimi

#4 lore raw
Quoted 51 lines · Andy Lutomirski
Quoted 41 lines · Mimi Zohar
Quoted 34 lines · Andy Lutomirski
Quoted 9 lines · James Morris

These changes fix a bug in xattr handling, where the evm and ima inode_setxattr() functions do not check for empty xattrs being passed from userspace (leading to user-triggerable null pointer dereferences).

Please pull.

The following changes since commit 9f76628da20f96a179ca62b504886f99ecc29223:

Quoted 8 lines

Merge branch 'for-3.18' of git://linux-nfs.org/~bfields/linux (2014-10-28 13:32:06 -0700)

are available in the git repository at:

git://git.kernel.org/pub/scm/linux/kernel/git/jmorris/linux-security.git for-linus

Dmitry Kasatkin (2):
ima: check xattr value length and type in the ima_inode_setxattr()

I haven't read this one, but:

Quoted 1 line

evm: check xattr value length and type in evm_inode_setxattr()

const struct evm_ima_xattr_data *xattr_data = xattr_value;
- if ((strcmp(xattr_name, XATTR_NAME_EVM) == 0)
- && (xattr_data->type == EVM_XATTR_HMAC))
- return -EPERM;
+ if (strcmp(xattr_name, XATTR_NAME_EVM) == 0) {
+ if (!xattr_value_len)
+ return -EINVAL;
+ if (xattr_data->type != EVM_IMA_XATTR_DIGSIG)
+ return -EPERM;
+ }

Quoted 4 lines

Huh? (Sorry about severe whitespace damage.)

Shouldn't there be something like if (xattr_value_len < sizeof(struct evm_ima_xattr_data)) return -EINVAL?

Quoted 8 lines

Prior to commit 2fb1c9a "evm: prohibit userspace writing 'security.evm' HMAC value", a process with CAP_SYS_ADMIN could write either an HMAC or signature. As the HMAC key should only be known to the kernel, only signatures are now allowed. Instead of "struct evm_ima_xattr_data", the code should reflect this change and use "struct signature_v2_hdr". We'll clean up this code for the next release. For now, this patch prevents the oops.

Quoted 4 lines

I have no idea what the semantics are. All I'm saying is that it looks like the code still accesses memory past the end of the buffer. The buffer isn't a null pointer, so the symptom is different, but it may still be a security bug.

There's no accessing of data here, just writing the data out as an extended attribute, which requires CAP_SYS_ADMIN privilege.

Mimi

#5 lore raw
In reply to Andy Lutomirski
Quoted 6 lines · Andy Lutomirski

I have no idea what the semantics are. All I'm saying is that it looks like the code still accesses memory past the end of the buffer. The buffer isn't a null pointer, so the symptom is different, but it may still be a security bug.

--Andy

It only reads one byte into the struct "xattr_data->type" so checking for non-zero is sufficient and the patch is fine.

I fixed that exact same bug in lustre last week where the xattr size is not zero but it's less than the size of the struct. So this seems like maybe it could be a common anti-pattern though.

regards,
dan carpenter

#6 lore raw
Quoted 10 lines · Dan Carpenter
Quoted 6 lines · Andy Lutomirski

I have no idea what the semantics are. All I'm saying is that it looks like the code still accesses memory past the end of the buffer. The buffer isn't a null pointer, so the symptom is different, but it may still be a security bug.

--Andy

It only reads one byte into the struct "xattr_data->type" so checking for non-zero is sufficient and the patch is fine.

Indeed. Still... eww. I don't like code that, upon local inspection, is apparently wrong, even though it's coincidentally correct due to some other far away condition.

--Andy

Quoted 8 lines

I fixed that exact same bug in lustre last week where the xattr size is not zero but it's less than the size of the struct. So this seems like maybe it could be a common anti-pattern though.

regards,
dan carpenter

Signature · 2 lines
Andy Lutomirski
AMA Capital Management, LLC
#7 lore raw
Quoted 8 lines · Andy Lutomirski
Quoted 6 lines · Dan Carpenter
Quoted 5 lines · Andy Lutomirski

I have no idea what the semantics are. All I'm saying is that it looks like the code still accesses memory past the end of the buffer. The buffer isn't a null pointer, so the symptom is different, but it may still be a security bug.

Quoted 8 lines
Quoted 4 lines
Quoted 1 line

--Andy

It only reads one byte into the struct "xattr_data->type" so checking for non-zero is sufficient and the patch is fine.

Indeed. Still... eww. I don't like code that, upon local inspection, is apparently wrong, even though it's coincidentally correct due to some other far away condition.

No, the code may be incomplete, but definitely not wrong.

Mimi

#8 lore raw
Quoted 12 lines · Mimi Zohar
Quoted 10 lines · Andy Lutomirski
Quoted 8 lines · Dan Carpenter
Quoted 6 lines · Andy Lutomirski

I have no idea what the semantics are. All I'm saying is that it looks like the code still accesses memory past the end of the buffer. The buffer isn't a null pointer, so the symptom is different, but it may still be a security bug.

--Andy

Quoted 8 lines
Quoted 6 lines
Quoted 2 lines

It only reads one byte into the struct "xattr_data->type" so checking for non-zero is sufficient and the patch is fine.

Indeed. Still... eww. I don't like code that, upon local inspection, is apparently wrong, even though it's coincidentally correct due to some other far away condition.

No, the code may be incomplete, but definitely not wrong.

I said "apparently wrong" instead of "wrong" for a reason :)

Quoted 3 lines

Mimi

#9 lore raw
Quoted 16 lines · Andy Lutomirski
Quoted 15 lines · Mimi Zohar
Quoted 13 lines · Andy Lutomirski
Quoted 10 lines · Dan Carpenter
Quoted 6 lines · Andy Lutomirski

I have no idea what the semantics are. All I'm saying is that it looks like the code still accesses memory past the end of the buffer. The buffer isn't a null pointer, so the symptom is different, but it may still be a security bug.

--Andy

It only reads one byte into the struct "xattr_data->type" so checking for non-zero is sufficient and the patch is fine.

Quoted 7 lines
Quoted 5 lines
Quoted 3 lines

Indeed. Still... eww. I don't like code that, upon local inspection, is apparently wrong, even though it's coincidentally correct due to some other far away condition.

No, the code may be incomplete, but definitely not wrong.

I said "apparently wrong" instead of "wrong" for a reason :)

I see there is a long discussion about this patch.

Actually using something like this "if (xattr_value_len < sizeof(struct evm_ima_xattr_data))" or using sizeof (struct signature_v2_hdr)
in this function does not give too much.
xattr value is variable length value and having that statement false absolutely does not mean that the value is correct.
It can be even a random garbage of the correct size.
This particular function checks the first byte only so the test is good enough.

- Dmitry

Quoted 8 lines
Quoted 4 lines

Mimi

--
To unsubscribe from this list: send the line "unsubscribe linux-security-module" in the body of a message to [email protected]
More majordomo info at http://vger.kernel.org/majordomo-info.html

Signature · 2 lines
Thanks,
Dmitry
#10 lore raw
Quoted 27 lines · Dmitry Kasatkin
Quoted 16 lines · Andy Lutomirski
Quoted 15 lines · Mimi Zohar
Quoted 13 lines · Andy Lutomirski
Quoted 10 lines · Dan Carpenter
Quoted 6 lines · Andy Lutomirski

I have no idea what the semantics are. All I'm saying is that it looks like the code still accesses memory past the end of the buffer. The buffer isn't a null pointer, so the symptom is different, but it may still be a security bug.

--Andy

It only reads one byte into the struct "xattr_data->type" so checking for non-zero is sufficient and the patch is fine.

Quoted 7 lines
Quoted 5 lines
Quoted 3 lines

Indeed. Still... eww. I don't like code that, upon local inspection, is apparently wrong, even though it's coincidentally correct due to some other far away condition.

No, the code may be incomplete, but definitely not wrong.

I said "apparently wrong" instead of "wrong" for a reason :)

I see there is a long discussion about this patch.

Quoted 9 lines

Actually using something like this "if (xattr_value_len < sizeof(struct evm_ima_xattr_data))" or using sizeof (struct signature_v2_hdr)
in this function does not give too much.
xattr value is variable length value and having that statement false absolutely does not mean that the value is correct.
It can be even a random garbage of the correct size.
This particular function checks the first byte only so the test is good enough.

My point is that there's no possible way to tell that only the first byte is read just by reading the function.

--Andy

Quoted 16 lines

- Dmitry

Quoted 8 lines
Quoted 4 lines

Mimi

--
To unsubscribe from this list: send the line "unsubscribe linux-security-module" in the body of a message to [email protected]
More majordomo info at http://vger.kernel.org/majordomo-info.html

--
Thanks,
Dmitry

Signature · 2 lines
Andy Lutomirski
AMA Capital Management, LLC
#11 lore raw
In reply to Dan Carpenter
Quoted 3 lines · Dan Carpenter

I fixed that exact same bug in lustre last week where the xattr size is not zero but it's less than the size of the struct. So this seems like maybe it could be a common anti-pattern though.

It must not be very common. I wrote a Smatch script which finds both the lustre and the ima bugs but it doesn't find anything else major.

Apparently parsing vmcores is buggy, for example and I reported a couple other small bugs to other lists.

fs/proc/vmcore.c:547 update_note_header_size_elf64() warn: is 'notes_section' large enough for 'struct elf64_note'? fs/proc/vmcore.c:733 update_note_header_size_elf32() warn: is 'notes_section' large enough for 'struct elf32_note'?

regards,
dan carpenter