[pdm-devel] [PATCH proxmox v2 4/4] access-control: allow reading all acls of the current authid

Shannon Sterz s.sterz at proxmox.com
Thu Oct 23 13:32:59 CEST 2025


On Thu Oct 23, 2025 at 11:31 AM CEST, Dominik Csapak wrote:
> High level question:
>
> in proxmox-backup we already have a 'list_permissions' api call
>
> would that (or something like that) not be better suited for the purpose?
> (i don't know if it would be easy to refactor that api call here;
> probably not because it's pbs specific)
>
> since with this api call, we'd only have the 'raw' acl entries and
> must to calculate things like propagation, etc. on the frontend?
>
> Or am I wrong with that assumption?

the endpoint also exists in pdm, but there are couple of things that
prevented me from using it:

- it's not yet in a common crate making it more brittle to use in
  yew-comp. though that can of course be changed easily.
- it gives you permissions, not roles. the acl tree, however, works with
  roles. the idea behind this patch series was to re-use the logic of
  the acl tree as much as possible. using permissions instead of roles
  would mean we:
    - either need to implement another datastructure on the front-end
      that queries something like an acl tree, but with permissions.
      duplicating quite a bit of logic and also being more error prone.
      we still need to traverse the tree after all to make sure that we
      have the correct path.
    - adapting the acl tree to work with permissions directly as well.
      this would mean quite a bit of churn in the acl tree, we need to
      make sure that none of the current use sites break.
- the performance impact of doing it this way wasn't to big in my
  testing. the actual acl tree in the front-end only consists of the
  users roles. so it isn't too big by itself. i doubt the propagation
  logic by itself costs us that much, as it often can just act as a
  short-circuit (i.e. meaning we return immediately if a role
  propagates).

i can however look into stress testing this a bit more to see if it
would cause issues given a very complex acl tree for one specific user.

> On 10/22/25 3:11 PM, Shannon Sterz wrote:
>> adds a parameter to the `API_METHOD_READ_ACL` endpoint to allow
>> listing all ACL entries of the currently authenticated Authid.
>> allowing a user to see their own ACLs does not really exposes any
>> additional confidential information. however, being able to query this
>> information allows us, for example, to adapt ui components to a users
>> capabilities.
>>
>> Signed-off-by: Shannon Sterz <s.sterz at proxmox.com>
>> ---
>>   proxmox-access-control/src/api/acl.rs | 37 ++++++++++++++++++++++-----
>>   1 file changed, 31 insertions(+), 6 deletions(-)
>>
>> diff --git a/proxmox-access-control/src/api/acl.rs b/proxmox-access-control/src/api/acl.rs
>> index 0194d517..07222939 100644
>> --- a/proxmox-access-control/src/api/acl.rs
>> +++ b/proxmox-access-control/src/api/acl.rs
>> @@ -23,6 +23,12 @@ use crate::CachedUserInfo;
>>                   optional: true,
>>                   default: false,
>>               },
>> +            "exact-authid": {
>> +                description: "Whether to return ACL entries for the exact current authid only.",
>> +                type: bool,
>> +                optional: true,
>> +                default: false,
>> +            }
>>           },
>>       },
>>       returns: {
>> @@ -34,13 +40,17 @@ use crate::CachedUserInfo;
>>       },
>>       access: {
>>           permission: &Permission::Anybody,
>> -        description: "Returns all ACLs if user has sufficient privileges on this endpoint, otherwise it is limited to the user's API tokens.",
>> +        description: "Returns all ACLs if a user has sufficient privileges on this endpoint. \
>> +            Otherwise it is limited to the user's API tokens. However, if `exact-authid` is \
>> +            specified, all ACLs of the current Auhtid will be returned, whether the Authid has \
>> +            privileges to list other ACLs here or not.",
>>       },
>>   )]
>>   /// Get ACL entries, can be filter by path.
>>   pub fn read_acl(
>>       path: Option<String>,
>>       exact: bool,
>> +    exact_authid: bool,
>>       rpcenv: &mut dyn RpcEnvironment,
>>   ) -> Result<Vec<AclListItem>, Error> {
>>       let auth_id = rpcenv
>> @@ -58,7 +68,11 @@ pub fn read_acl(
>>           )
>>           .is_err();
>>
>> -    let filter = if filter_entries { Some(auth_id) } else { None };
>> +    let filter = if filter_entries || exact_authid {
>> +        Some(auth_id)
>> +    } else {
>> +        None
>> +    };
>>
>>       let (mut tree, digest) = crate::acl::config()?;
>>
>> @@ -74,7 +88,13 @@ pub fn read_acl(
>>
>>       rpcenv["digest"] = hex::encode(digest).into();
>>
>> -    Ok(extract_acl_node_data(node, path.as_deref(), exact, &filter))
>> +    Ok(extract_acl_node_data(
>> +        node,
>> +        path.as_deref(),
>> +        exact_authid,
>> +        exact,
>> +        &filter,
>> +    ))
>>   }
>>
>>   #[api(
>> @@ -241,7 +261,8 @@ pub fn update_acl(
>>   fn extract_acl_node_data(
>>       node: &AclTreeNode,
>>       path: Option<&str>,
>> -    exact: bool,
>> +    exact_authid: bool,
>> +    exact_path: bool,
>>       auth_id_filter: &Option<Authid>,
>>   ) -> Vec<AclListItem> {
>>       // tokens can't have tokens, so we can early return
>> @@ -259,7 +280,11 @@ fn extract_acl_node_data(
>>
>>           for (user, roles) in &node.users {
>>               if let Some(auth_id_filter) = auth_id_filter {
>> -                if !user.is_token() || user.user() != auth_id_filter.user() {
>> +                if exact_authid {
>> +                    if user != auth_id_filter {
>> +                        continue;
>> +                    }
>> +                } else if !user.is_token() || user.user() != auth_id_filter.user() {
>>                       continue;
>>                   }
>>               }
>> @@ -291,7 +316,7 @@ fn extract_acl_node_data(
>>               }
>>           }
>>
>> -        if !exact {
>> +        if !exact_path {
>>               nodes.extend(
>>                   node.children
>>                       .iter()





More information about the pdm-devel mailing list