On Wed, Oct 26, 2022 at 10:30 AM Nicolas Bouchinet <nicolas.bouchinet@xxxxxxxxxxx> wrote: > > From: Nicolas Bouchinet <nicolas.bouchinet@xxxxxxxxxxx> > > Fixes a NULL pointer dereference occurring in the > `evm_protected_xattr_common` function of the EVM LSM. The bug is > triggered if a `inode_init_security` hook returns 0 without initializing > the given `struct xattr` fields (which is the case of BPF) and if no > other LSM overrides thoses fields after. This also leads to memory > leaks. > > The `call_int_hook_xattr` macro has been inlined into the > `security_inode_init_security` hook in order to check hooks return > values and skip ones who doesn't init `xattrs`. > > Modify `evm_init_hmac` function to init the EVM hmac using every > entry of the given xattr array. > > The `MAX_LSM_EVM_XATTR` value is now based on the security modules > compiled in, which gives room for SMACK, SELinux, Apparmor, BPF and > IMA/EVM security attributes. > > Changes the default return value of the `inode_init_security` hook > definition to `-EOPNOTSUPP`. > > Changes the hook documentation to match the behavior of the LSMs using > it (only xattr->value is initialised with kmalloc and thus is the only > one that should be kfreed by the caller). > > Cc: roberto.sassu@xxxxxxxxxxxxxxx > Signed-off-by: Nicolas Bouchinet <nicolas.bouchinet@xxxxxxxxxxx> > --- > Changes since v3: > https://lore.kernel.org/linux-integrity/Y1fu4jofqLHVDprT@archlinux/ > > * Fixes compilation error reported by the kernel test robot. > --- > include/linux/lsm_hook_defs.h | 2 +- > include/linux/lsm_hooks.h | 4 ++-- > security/integrity/evm/evm.h | 1 + > security/integrity/evm/evm_crypto.c | 9 +++++++-- > security/integrity/evm/evm_main.c | 7 ++++--- > security/security.c | 31 ++++++++++++++++++++++------- > 6 files changed, 39 insertions(+), 15 deletions(-) ... > diff --git a/security/security.c b/security/security.c > index 14d30fec8a00..79524f8734f1 100644 > --- a/security/security.c > +++ b/security/security.c > @@ -30,7 +30,11 @@ > #include <linux/msg.h> > #include <net/flow.h> > > -#define MAX_LSM_EVM_XATTR 2 > +#define MAX_LSM_EVM_XATTR \ > + ((IS_ENABLED(CONFIG_EVM) ? 1 : 0) + \ > + (IS_ENABLED(CONFIG_SECURITY_SELINUX) ? 1 : 0) + \ > + (IS_ENABLED(CONFIG_SECURITY_SMACK) ? 1 : 0) + \ > + (IS_ENABLED(CONFIG_BPF_LSM) ? 1 : 0)) ... > @@ -1091,9 +1095,11 @@ int security_inode_init_security(struct inode *inode, struct inode *dir, > const struct qstr *qstr, > const initxattrs initxattrs, void *fs_data) > { > + int i = 0; > + int ret = -EOPNOTSUPP; > struct xattr new_xattrs[MAX_LSM_EVM_XATTR + 1]; > struct xattr *lsm_xattr, *evm_xattr, *xattr; > - int ret; > + struct security_hook_list *hook_ptr; > > if (unlikely(IS_PRIVATE(inode))) > return 0; > @@ -1103,15 +1109,26 @@ int security_inode_init_security(struct inode *inode, struct inode *dir, > dir, qstr, NULL, NULL, NULL); > memset(new_xattrs, 0, sizeof(new_xattrs)); > lsm_xattr = new_xattrs; > - ret = call_int_hook(inode_init_security, -EOPNOTSUPP, inode, dir, qstr, > - &lsm_xattr->name, > - &lsm_xattr->value, > - &lsm_xattr->value_len); > + hlist_for_each_entry(hook_ptr, &security_hook_heads.inode_init_security, > + list) { > + ret = hook_ptr->hook.inode_init_security(inode, dir, qstr, > + &lsm_xattr->name, > + &lsm_xattr->value, > + &lsm_xattr->value_len); > + if (ret == -EOPNOTSUPP) > + continue; > + if (WARN_ON_ONCE(i >= MAX_LSM_EVM_XATTR)) > + ret = -ENOMEM; It would really like to see us get rid of the MAX_LSM_EVM_XATTR macro and determine the array size similar to what we do with the security blob sizes. The macro definition is a kludgy hack that is bound to get out of sync at some point and this extra checking inside the hook is something we should work to remove. > + if (ret != 0) > + break; > + lsm_xattr++; > + i++; > + } > if (ret) > goto out; > > evm_xattr = lsm_xattr + 1; > - ret = evm_inode_init_security(inode, lsm_xattr, evm_xattr); > + ret = evm_inode_init_security(inode, new_xattrs, evm_xattr); > if (ret) > goto out; > ret = initxattrs(inode, new_xattrs, fs_data); > -- > 2.38.1 -- paul-moore.com
