hulk inclusion category: feature feature: digest-lists
---------------------------
EVM is a module for the protection of the integrity of file metadata. It protects security-relevant extended attributes, and some file attributes such as the UID and the GID. It protects their integrity with an HMAC or with a signature.
What makes EVM different from other LSMs is that it makes a security decision depending on multiple pieces of information, which cannot be managed atomically by the system.
Example: cp -a file.orig file.dest
If security.selinux, security.ima and security.evm must be preserved, cp will invoke setxattr() for each xattr, and EVM performs a verification during each operation. The problem is that copying security.evm from file.orig to file.dest will likely break the following EVM verifications if some metadata still have to be copied. EVM has no visibility on the metadata of the source file, so it cannot determine when the copy can be considered complete.
On the other hand, EVM has to check metadata during every operation to ensure that there is no transition from corrupted metadata, e.g. after an offline attack, to valid ones after the operation. An HMAC update would prevent the corruption to be detected, as the HMAC on the new values would be correct. Thus, to avoid this issue, EVM has to return an error to the system call so that its execution will be interrupted.
A solution that would satisfy both requirements, not breaking user space applications and detecting corrupted metadata is to let metadata operations be completed successfully and to pass the result of the EVM verification from the pre hooks to the post hooks. In this way, the HMAC update can be avoided if the verification wasn't successful.
This approach will bring another important benefit: it is no longer required that every file has a valid HMAC or signature. Instead of always enforcing metadata integrity, even when it is not relevant for IMA, EVM will let IMA decide for files selected with the appraisal policy, depending on the result of the requested verification.
The main problem is that the result of the verification currently cannot be passed from the pre hooks to the post hooks, due to how the LSM API is defined. A possible solution would be to use integrity_iint_cache for this purpose, but it will increase the memory pressure, as new structures will be allocated also for metadata operations, not only for measurement, appraisal and audit. Another solution would be to extend the LSM API, but it seems not worthwhile as EVM would be the only module getting a benefit from this change.
Given that pre and post hooks are called from the same system call, a more efficient solution seems to move the hooks outside the LSM infrastructure, so that the return value of the pre hooks can be passed to the post hooks. A predefined error (-EAGAIN) will be used to signal to the system call to continue the execution. Otherwise, if the pre hooks return -EPERM, the system calls will behave as before and will immediately return before metadata are changed.
Overview of the changes:
evm_inode_init_security() LSM (no change) evm_inode_setxattr() LSM -> vfs_setxattr() evm_inode_post_setxattr() LSM -> vfs_setxattr() evm_inode_removexattr() LSM -> vfs_removexattr() evm_inode_post_removexattr() vfs_removexattr() (no change) evm_inode_setattr() LSM -> vfs_setattr() evm_inode_post_setattr() vfs_setattr() (no change) evm_verifyxattr() outside LSM (no change)
Signed-off-by: Roberto Sassu roberto.sassu@huawei.com --- fs/attr.c | 5 ++++- fs/xattr.c | 17 +++++++++++++++-- security/security.c | 18 +++--------------- 3 files changed, 22 insertions(+), 18 deletions(-)
diff --git a/fs/attr.c b/fs/attr.c index d22e8187477f..322e5e887ece 100644 --- a/fs/attr.c +++ b/fs/attr.c @@ -227,7 +227,7 @@ int notify_change(struct dentry * dentry, struct iattr * attr, struct inode **de { struct inode *inode = dentry->d_inode; umode_t mode = inode->i_mode; - int error; + int error, evm_error; struct timespec64 now; unsigned int ia_valid = attr->ia_valid;
@@ -326,6 +326,9 @@ int notify_change(struct dentry * dentry, struct iattr * attr, struct inode **de error = security_inode_setattr(dentry, attr); if (error) return error; + evm_error = evm_inode_setattr(dentry, attr); + if (evm_error) + return evm_error; error = try_break_deleg(inode, delegated_inode); if (error) return error; diff --git a/fs/xattr.c b/fs/xattr.c index 470ee0af3200..a0e4e77f78b1 100644 --- a/fs/xattr.c +++ b/fs/xattr.c @@ -182,6 +182,7 @@ int __vfs_setxattr_noperm(struct dentry *dentry, const char *name, fsnotify_xattr(dentry); security_inode_post_setxattr(dentry, name, value, size, flags); + evm_inode_post_setxattr(dentry, name, value, size); } } else { if (unlikely(is_bad_inode(inode))) @@ -221,7 +222,7 @@ __vfs_setxattr_locked(struct dentry *dentry, const char *name, struct inode **delegated_inode) { struct inode *inode = dentry->d_inode; - int error; + int error, evm_error;
error = xattr_permission(inode, name, MAY_WRITE); if (error) @@ -231,6 +232,12 @@ __vfs_setxattr_locked(struct dentry *dentry, const char *name, if (error) goto out;
+ evm_error = evm_inode_setxattr(dentry, name, value, size); + if (evm_error) { + error = evm_error; + goto out; + } + error = try_break_deleg(inode, delegated_inode); if (error) goto out; @@ -428,7 +435,7 @@ __vfs_removexattr_locked(struct dentry *dentry, const char *name, struct inode **delegated_inode) { struct inode *inode = dentry->d_inode; - int error; + int error, evm_error;
error = xattr_permission(inode, name, MAY_WRITE); if (error) @@ -438,6 +445,12 @@ __vfs_removexattr_locked(struct dentry *dentry, const char *name, if (error) goto out;
+ evm_error = evm_inode_removexattr(dentry, name); + if (evm_error) { + error = evm_error; + goto out; + } + error = try_break_deleg(inode, delegated_inode); if (error) goto out; diff --git a/security/security.c b/security/security.c index 9478444bf93f..17e2bed11bf7 100644 --- a/security/security.c +++ b/security/security.c @@ -706,14 +706,9 @@ int security_inode_permission(struct inode *inode, int mask)
int security_inode_setattr(struct dentry *dentry, struct iattr *attr) { - int ret; - if (unlikely(IS_PRIVATE(d_backing_inode(dentry)))) return 0; - ret = call_int_hook(inode_setattr, 0, dentry, attr); - if (ret) - return ret; - return evm_inode_setattr(dentry, attr); + return call_int_hook(inode_setattr, 0, dentry, attr); } EXPORT_SYMBOL_GPL(security_inode_setattr);
@@ -742,10 +737,7 @@ int security_inode_setxattr(struct dentry *dentry, const char *name, ret = cap_inode_setxattr(dentry, name, value, size, flags); if (ret) return ret; - ret = ima_inode_setxattr(dentry, name, value, size); - if (ret) - return ret; - return evm_inode_setxattr(dentry, name, value, size); + return ima_inode_setxattr(dentry, name, value, size); }
void security_inode_post_setxattr(struct dentry *dentry, const char *name, @@ -754,7 +746,6 @@ void security_inode_post_setxattr(struct dentry *dentry, const char *name, if (unlikely(IS_PRIVATE(d_backing_inode(dentry)))) return; call_void_hook(inode_post_setxattr, dentry, name, value, size, flags); - evm_inode_post_setxattr(dentry, name, value, size); }
int security_inode_getxattr(struct dentry *dentry, const char *name) @@ -786,10 +777,7 @@ int security_inode_removexattr(struct dentry *dentry, const char *name) ret = cap_inode_removexattr(dentry, name); if (ret) return ret; - ret = ima_inode_removexattr(dentry, name); - if (ret) - return ret; - return evm_inode_removexattr(dentry, name); + return ima_inode_removexattr(dentry, name); }
int security_inode_need_killpriv(struct dentry *dentry)