Re: [PATCH v2] audit: add FSCONFIG auxiliary record to log filesystem configuration

[Steve Grubb <sgrubb@redhat.com>]

On Tuesday, July 28, 2026 4:35:20 PM Eastern Daylight Time Paul Moore wrote:
> On Jul 27, 2026 Ricardo Robaina <rrobaina@redhat.com> wrote:
> > Modern mount tools (util-linux >= 2.39.1) use the new mount API
> > (fsopen, fsconfig, fsmount, move_mount) instead of the legacy mount(2)
> > syscall. The generic SYSCALL audit record logs the fsconfig syscall but
> > does not capture the configuration parameters, creating an audit gap for
> > critical mount information such as the device being mounted.
> > 
> > Add an FSCONFIG auxiliary record that logs the command type, parameter
> > name (key), parameter value, and aux parameter passed to fsconfig(2).
> > 
> >  ----
> >  type=SYSCALL : syscall=fsconfig ... a1=FSCONFIG_SET_STRING ...
> >  type=FSCONFIG : fs_cmd=1 fs_key=source fs_val="tmpfs" fs_aux=0
> >  ----
> >  type=SYSCALL : syscall=fsconfig ... a1=FSCONFIG_CMD_CREATE ...
> >  type=FSCONFIG : fs_cmd=6 fs_key=(null) fs_val=(null) fs_aux=0
> >  ----
> >  type=SYSCALL : syscall=fsconfig ... a1=FSCONFIG_SET_BINARY ...
> >  type=FSCONFIG : fs_cmd=2 fs_key=hidepid fs_val="<binary>" fs_aux=4
> > 
> > Link: https://github.com/linux-audit/audit-kernel/issues/153
> > Acked-by: Christian Brauner <brauner@kernel.org>
> > Signed-off-by: Ricardo Robaina <rrobaina@redhat.com>
> > ---
> > Changes in v2:
> > - Fixed null-check for fs_key field.
> > - Clarified trimmed SYSCALL output in commit message examples.
> > 
> >  fs/fsopen.c                |  7 +++++++
> >  include/linux/audit.h      | 13 +++++++++++++
> >  include/uapi/linux/audit.h |  1 +
> >  kernel/auditsc.c           | 27 +++++++++++++++++++++++++++
> >  4 files changed, 48 insertions(+)
> > 
> > diff --git a/fs/fsopen.c b/fs/fsopen.c
> > index ae19e5136598..9b3c02f59df4 100644
> > --- a/fs/fsopen.c
> > +++ b/fs/fsopen.c
> > @@ -15,6 +15,7 @@
> > 
> >  #include <linux/namei.h>
> >  #include <linux/file.h>
> >  #include <uapi/linux/mount.h>
> > 
> > +#include <linux/audit.h>
> > 
> >  #include "internal.h"
> >  #include "mount.h"
> > 
> > @@ -357,6 +358,7 @@ SYSCALL_DEFINE5(fsconfig,
> > 
> >  	struct fs_context *fc;
> >  	int ret;
> >  	int lookup_flags = 0;
> > 
> > +	const char *value_str = NULL;
> > 
> >  	struct fs_parameter param = {
> >  	
> >  		.type	= fs_value_is_undefined,
> > 
> > @@ -423,6 +425,7 @@ SYSCALL_DEFINE5(fsconfig,
> > 
> >  			goto out_key;
> >  		
> >  		}
> >  		param.size = strlen(param.string);
> > 
> > +		value_str = param.string;
> > 
> >  		break;
> >  	
> >  	case FSCONFIG_SET_BINARY:
> >  		param.type = fs_value_is_blob;
> > 
> > @@ -432,6 +435,7 @@ SYSCALL_DEFINE5(fsconfig,
> > 
> >  			ret = PTR_ERR(param.blob);
> >  			goto out_key;
> >  		
> >  		}
> > 
> > +		value_str = "<binary>";
> 
> Is there a reason why we're not logging the binary data?  We might want
> to impose a size limit to truncate the logged data (although we have
> provisions to log rather large binary chunks), but I don't see a reason
> why we couldn't log the binary data as a hex string.

I did a search for any user of this. There are no known consumers. Last 
February, fsparam_blob() / fs_param_is_blob parser support was removed. That 
means we don't really have a use case. If, for example, someone decided to 
use this API to insert a private key used for decryption or something, we 
probably don't want that in logs. So, without an actual use case, it's hard 
to say if logging this would be a liability or something needed. But yes, it 
could be done with the hex string encoding if required.

-Steve


> >  		break;
> >  	
> >  	case FSCONFIG_SET_PATH_EMPTY:
> >  		lookup_flags = LOOKUP_EMPTY;
> > 
> > @@ -445,6 +449,7 @@ SYSCALL_DEFINE5(fsconfig,
> > 
> >  		}
> >  		param.dirfd = aux;
> >  		param.size = strlen(param.name->name);
> > 
> > +		value_str = param.name->name;
> > 
> >  		break;
> >  	
> >  	case FSCONFIG_SET_FD:
> >  		param.type = fs_value_is_file;
> > 
> > @@ -458,6 +463,8 @@ SYSCALL_DEFINE5(fsconfig,
> > 
> >  		break;
> >  	
> >  	}
> > 
> > +	audit_log_fsconfig(cmd, param.key, value_str, aux);
> > +
> > 
> >  	ret = mutex_lock_interruptible(&fc->uapi_mutex);
> >  	if (ret == 0) {
> >  	
> >  		ret = vfs_fsconfig_locked(fc, cmd, &param);
> 
> ...
> 
> > diff --git a/kernel/auditsc.c b/kernel/auditsc.c
> > index 6610e667c728..fefc5c5dd4aa 100644
> > --- a/kernel/auditsc.c
> > +++ b/kernel/auditsc.c
> > @@ -2882,6 +2882,33 @@ void __audit_log_nfcfg(const char *name, u8 af,
> > unsigned int nentries,> 
> >  }
> >  EXPORT_SYMBOL_GPL(__audit_log_nfcfg);
> > 
> > +void __audit_log_fsconfig(unsigned int cmd, const char *key,
> > +			  const char *value, int aux)
> > +{
> > +	struct audit_buffer *ab;
> > +
> > +	ab = audit_log_start(audit_context(), GFP_KERNEL, AUDIT_FSCONFIG);
> > +	if (!ab)
> > +		return;
> > +
> > +	audit_log_format(ab, "fs_cmd=%u", cmd);
> > +	audit_log_format(ab, " fs_key=");
> 
> Calling into audit_log_format() is expensive due to the string
> processing, we should combine this into a single audit_log_format()
> call:
> 
>  audit_log_format(ab, "fs_cmd=%u fs_key=", cmd);
> 
> > +	if (key)
> > +		audit_log_untrustedstring(ab, key);
> > +	else
> > +		audit_log_format(ab, "(null)");
> > +
> > +	audit_log_format(ab, " fs_val=");
> > +	if (value)
> > +		audit_log_untrustedstring(ab, value);
> > +	else
> > +		audit_log_format(ab, "(null)");
> > +
> > +	audit_log_format(ab, " fs_aux=%d", aux);
> > +
> > +	audit_log_end(ab);
> > +}
> > +
> > 
> >  static void audit_log_task(struct audit_buffer *ab)
> >  {
> >  
> >  	kuid_t auid, uid;
> 
> --
> paul-moore.com