kanidm/src/lib/entry.rs

// use serde_json::{Error, Value};
use super::proto_v1::Entry as ProtoEntry;
use audit::AuditScope;
use error::{OperationError, SchemaError};
use filter::{Filter, FilterValid};
use modify::{Modify, ModifyInvalid, ModifyList, ModifyValid};
use schema::{SchemaAttribute, SchemaClass, SchemaReadTransaction};
use server::{QueryServerReadTransaction, QueryServerWriteTransaction};
use std::collections::btree_map::{Iter as BTreeIter, IterMut as BTreeIterMut};
use std::collections::BTreeMap;
use std::collections::HashMap;
use std::iter::ExactSizeIterator;
use std::slice::Iter as SliceIter;

// make a trait entry for everything to adhere to?
//  * How to get indexs out?
//  * How to track pending diffs?

// Entry is really similar to serde Value, but limits the possibility
// of what certain types could be.
//
// The idea of an entry is that we have
// an entry that looks like:
//
// {
//    'class': ['object', ...],
//    'attr': ['value', ...],
//    'attr': ['value', ...],
//    ...
// }
//
// When we send this as a result to clients, we could embed other objects as:
//
// {
//    'attr': [
//        'value': {
//        },
//    ],
// }
//

pub struct EntryClasses<'a> {
    size: usize,
    inner: Option<SliceIter<'a, String>>,
    // _p: &'a PhantomData<()>,
}

impl<'a> Iterator for EntryClasses<'a> {
    type Item = &'a String;

    #[inline]
    fn next(&mut self) -> Option<(&'a String)> {
        match self.inner.iter_mut().next() {
            Some(i) => i.next(),
            None => None,
        }
    }

    #[inline]
    fn size_hint(&self) -> (usize, Option<usize>) {
        match self.inner.iter().next() {
            Some(i) => i.size_hint(),
            None => (0, None),
        }
    }
}

impl<'a> ExactSizeIterator for EntryClasses<'a> {
    fn len(&self) -> usize {
        self.size
    }
}

pub struct EntryAvas<'a> {
    inner: BTreeIter<'a, String, Vec<String>>,
}

impl<'a> Iterator for EntryAvas<'a> {
    type Item = (&'a String, &'a Vec<String>);

    #[inline]
    fn next(&mut self) -> Option<(&'a String, &'a Vec<String>)> {
        self.inner.next()
    }

    #[inline]
    fn size_hint(&self) -> (usize, Option<usize>) {
        self.inner.size_hint()
    }
}

pub struct EntryAvasMut<'a> {
    inner: BTreeIterMut<'a, String, Vec<String>>,
}

impl<'a> Iterator for EntryAvasMut<'a> {
    type Item = (&'a String, &'a mut Vec<String>);

    #[inline]
    fn next(&mut self) -> Option<(&'a String, &'a mut Vec<String>)> {
        self.inner.next()
    }

    #[inline]
    fn size_hint(&self) -> (usize, Option<usize>) {
        self.inner.size_hint()
    }
}

// This is a BE concept, so move it there!

// Entry should have a lifecycle of types. THis is Raw (modifiable) and Entry (verified).
// This way, we can move between them, but only certain actions are possible on either
// This means modifications happen on Raw, but to move to Entry, you schema normalise.
// Vice versa, you can for free, move to Raw, but you lose the validation.

// Because this is type system it's "free" in the end, and means we force validation
// at the correct and required points of the entries life.

// This is specifically important for the commit to the backend, as we only want to
// commit validated types.

#[derive(Clone, Copy, Serialize, Deserialize, Debug)]
pub struct EntryNew; // new
#[derive(Clone, Copy, Serialize, Deserialize, Debug)]
pub struct EntryCommitted; // It's been in the DB, so it has an id
                           // pub struct EntryPurged;

#[derive(Clone, Copy, Serialize, Deserialize, Debug)]
pub struct EntryValid; // Asserted with schema.
#[derive(Clone, Copy, Serialize, Deserialize, Debug)]
pub struct EntryInvalid; // Modified

#[derive(Serialize, Deserialize, Debug)]
pub struct Entry<VALID, STATE> {
    valid: VALID,
    state: STATE,
    pub id: Option<i64>,
    // Flag if we have been schema checked or not.
    // pub schema_validated: bool,
    attrs: BTreeMap<String, Vec<String>>,
}

impl Entry<EntryInvalid, EntryNew> {
    #[cfg(test)]
    pub fn new() -> Self {
        Entry {
            // This means NEVER COMMITED
            valid: EntryInvalid,
            state: EntryNew,
            id: None,
            attrs: BTreeMap::new(),
        }
    }

    // FIXME: Can we consume protoentry?
    pub fn from(
        audit: &mut AuditScope,
        e: &ProtoEntry,
        qs: &QueryServerWriteTransaction,
    ) -> Result<Self, OperationError> {
        // Why not the trait? In the future we may want to extend
        // this with server aware functions for changes of the
        // incoming data.

        // Somehow we need to take the tree of e attrs, and convert
        // all ref types to our types ...

        let map2: Result<BTreeMap<String, Vec<String>>, OperationError> = e
            .attrs
            .iter()
            .map(|(k, v)| {
                let nv: Result<Vec<_>, _> =
                    v.iter().map(|vr| qs.clone_value(audit, &k, vr)).collect();
                match nv {
                    Ok(mut nvi) => {
                        nvi.sort_unstable();
                        Ok((k.clone(), nvi))
                    }
                    Err(e) => Err(e),
                }
            })
            .collect();

        let x = match map2 {
            Ok(v) => v,
            Err(e) => return Err(e),
        };

        Ok(Entry {
            // For now, we do a straight move, and we sort the incoming data
            // sets so that BST works.
            state: EntryNew,
            valid: EntryInvalid,
            id: None,
            attrs: x,
        })
    }
}

impl<STATE> Entry<EntryInvalid, STATE> {
    pub fn validate(
        self,
        schema: &SchemaReadTransaction,
    ) -> Result<Entry<EntryValid, STATE>, SchemaError> {
        // We need to clone before we start, as well be mutating content.
        // We destructure:
        let Entry {
            valid: _,
            state,
            id,
            attrs,
        } = self;

        let schema_classes = schema.get_classes();
        let schema_attributes = schema.get_attributes();

        // This should never fail!
        let schema_attr_name = match schema_attributes.get("name") {
            Some(v) => v,
            None => {
                return Err(SchemaError::Corrupted);
            }
        };

        let mut new_attrs = BTreeMap::new();

        // First normalise
        for (attr_name, avas) in attrs.iter() {
            let attr_name_normal: String = schema_attr_name.normalise_value(attr_name);
            // Get the needed schema type
            let schema_a_r = schema_attributes.get(&attr_name_normal);

            let mut avas_normal: Vec<String> = match schema_a_r {
                Some(schema_a) => {
                    avas.iter()
                        .map(|av| {
                            // normalise those based on schema?
                            schema_a.normalise_value(av)
                        })
                        .collect()
                }
                None => avas.clone(),
            };

            // Ensure they are ordered property.
            avas_normal.sort_unstable();

            // Should never fail!
            let _ = new_attrs.insert(attr_name_normal, avas_normal);
        }

        let ne = Entry {
            valid: EntryValid,
            state: state,
            id: id,
            attrs: new_attrs,
        };
        // Now validate.

        // First look at the classes on the entry.
        // Now, check they are valid classes
        //
        // FIXME: We could restructure this to be a map that gets Some(class)
        // if found, then do a len/filter/check on the resulting class set?

        {
            // First, check we have class on the object ....
            if !ne.attribute_pres("class") {
                return Err(SchemaError::InvalidClass);
            }

            let entry_classes = ne.classes();
            let entry_classes_size = entry_classes.len();

            let classes: HashMap<String, &SchemaClass> = entry_classes
                .filter_map(|c| match schema_classes.get(c) {
                    Some(cls) => Some((c.clone(), cls)),
                    None => None,
                })
                .collect();

            if classes.len() != entry_classes_size {
                return Err(SchemaError::InvalidClass);
            };

            let extensible = classes.contains_key("extensibleobject");

            // What this is really doing is taking a set of classes, and building an
            // "overall" class that describes this exact object for checking

            //   for each class
            //      add systemmust/must and systemmay/may to their lists
            //      add anything from must also into may

            // Now from the set of valid classes make a list of must/may
            // FIXME: This is clone on read, which may be really slow. It also may
            // be inefficent on duplicates etc.
            let must: HashMap<String, &SchemaAttribute> = classes
                .iter()
                // Join our class systemmmust + must into one iter
                .flat_map(|(_, cls)| cls.systemmust.iter().chain(cls.must.iter()))
                .map(|s| {
                    // This should NOT fail - if it does, it means our schema is
                    // in an invalid state!
                    // TODO: Make this return Corrupted on failure.
                    (s.clone(), schema_attributes.get(s).unwrap())
                })
                .collect();

            // FIXME: Error needs to say what is missing
            // We need to return *all* missing attributes.

            // Check that all must are inplace
            //   for each attr in must, check it's present on our ent
            // FIXME: Could we iter over only the attr_name
            for (attr_name, _attr) in must {
                let avas = ne.get_ava(&attr_name);
                if avas.is_none() {
                    return Err(SchemaError::MissingMustAttribute(String::from(attr_name)));
                }
            }

            // Check that any other attributes are in may
            //   for each attr on the object, check it's in the may+must set
            for (attr_name, avas) in ne.avas() {
                match schema_attributes.get(attr_name) {
                    Some(a_schema) => {
                        // Now, for each type we do a *full* check of the syntax
                        // and validity of the ava.
                        let r = a_schema.validate_ava(avas);
                        // We have to destructure here to make type checker happy
                        match r {
                            Ok(_) => {}
                            Err(e) => return Err(e),
                        }
                    }
                    None => {
                        if !extensible {
                            return Err(SchemaError::InvalidAttribute);
                        }
                    }
                }
            }
        } // unborrow ne.

        // Well, we got here, so okay!
        Ok(ne)
    }
}

impl<VALID, STATE> Clone for Entry<VALID, STATE>
where
    VALID: Copy,
    STATE: Copy,
{
    // Dirty modifiable state. Works on any other state to dirty them.
    fn clone(&self) -> Entry<VALID, STATE> {
        Entry {
            valid: self.valid,
            state: self.state,
            id: self.id,
            attrs: self.attrs.clone(),
        }
    }
}

/*
 * A series of unsafe transitions allowing entries to skip certain steps in
 * the process to facilitate eq/checks.
 */
impl<VALID, STATE> Entry<VALID, STATE> {
    #[cfg(test)]
    pub unsafe fn to_valid_new(self) -> Entry<EntryValid, EntryNew> {
        Entry {
            valid: EntryValid,
            state: EntryNew,
            id: self.id,
            attrs: self.attrs,
        }
    }

    #[cfg(test)]
    pub unsafe fn to_valid_committed(self) -> Entry<EntryValid, EntryCommitted> {
        Entry {
            valid: EntryValid,
            state: EntryCommitted,
            id: self.id,
            attrs: self.attrs,
        }
    }

    // Both invalid states can be reached from "entry -> invalidate"
}

impl Entry<EntryValid, EntryNew> {
    pub fn compare(&self, rhs: &Entry<EntryValid, EntryCommitted>) -> bool {
        self.attrs == rhs.attrs
    }
}

impl Entry<EntryValid, EntryCommitted> {
    pub fn compare(&self, rhs: &Entry<EntryValid, EntryNew>) -> bool {
        self.attrs == rhs.attrs
    }

    pub fn to_tombstone(&self) -> Self {
        // Duplicate this to a tombstone entry.
        let uuid_ava = self
            .get_ava(&String::from("uuid"))
            .expect("Corrupted entry!");
        let class_ava = vec!["object".to_string(), "tombstone".to_string()];

        let mut attrs_new: BTreeMap<String, Vec<String>> = BTreeMap::new();

        attrs_new.insert("uuid".to_string(), uuid_ava.clone());
        attrs_new.insert("class".to_string(), class_ava);

        Entry {
            valid: EntryValid,
            state: EntryCommitted,
            id: self.id,
            attrs: attrs_new,
        }
    }
}

impl<STATE> Entry<EntryValid, STATE> {
    pub fn invalidate(self) -> Entry<EntryInvalid, STATE> {
        Entry {
            valid: EntryInvalid,
            state: self.state,
            id: self.id,
            attrs: self.attrs,
        }
    }

    pub fn seal(self) -> Entry<EntryValid, EntryCommitted> {
        Entry {
            valid: self.valid,
            state: EntryCommitted,
            id: self.id,
            attrs: self.attrs,
        }
    }

    // Assert if this filter matches the entry (no index)
    pub fn entry_match_no_index(&self, filter: &Filter<FilterValid>) -> bool {
        // Go through the filter components and check them in the entry.
        // This is recursive!!!!
        match filter {
            Filter::Eq(attr, value) => self.attribute_equality(attr.as_str(), value.as_str()),
            Filter::Sub(attr, subvalue) => {
                self.attribute_substring(attr.as_str(), subvalue.as_str())
            }
            Filter::Pres(attr) => {
                // Given attr, is is present in the entry?
                self.attribute_pres(attr.as_str())
            }
            Filter::Or(l) => l.iter().fold(false, |acc, f| {
                // Check with ftweedal about or filter zero len correctness.
                if acc {
                    acc
                } else {
                    self.entry_match_no_index(f)
                }
            }),
            Filter::And(l) => l.iter().fold(true, |acc, f| {
                // Check with ftweedal about and filter zero len correctness.
                if acc {
                    self.entry_match_no_index(f)
                } else {
                    acc
                }
            }),
            Filter::AndNot(f) => !self.entry_match_no_index(f),
            Filter::invalid(_) => {
                // TODO: Is there a better way to not need to match the phantom?
                unimplemented!()
            }
        }
    }

    pub fn filter_from_attrs(&self, attrs: &Vec<String>) -> Option<Filter<FilterValid>> {
        // Because we are a valid entry, a filter we create *must* be valid
        //
        // Generate a filter from the attributes requested and defined.
        // Basically, this is a series of nested and's (which will be
        // optimised down later: but if someone wants to solve flatten() ...)

        // Take name: (a, b), name: (c, d) -> (name, a), (name, b), (name, c), (name, d)

        let mut pairs: Vec<(String, String)> = Vec::new();

        for attr in attrs {
            match self.attrs.get(attr) {
                Some(values) => {
                    for v in values {
                        pairs.push((attr.clone(), v.clone()))
                    }
                }
                None => return None,
            }
        }

        // Now make this a filter?

        let eq_filters = pairs
            .into_iter()
            .map(|(attr, value)| Filter::Eq(attr, value))
            .collect();

        Some(Filter::And(eq_filters))
    }

    pub fn into(&self) -> ProtoEntry {
        // It's very likely that at this stage we'll need to apply
        // access controls, dynamic attributes or more.
        // As a result, this may not even be the right place
        // for the conversion as algorithmically it may be
        // better to do this from the outside view. This can
        // of course be identified and changed ...
        ProtoEntry {
            attrs: self.attrs.clone(),
        }
    }

    pub fn gen_modlist_assert(
        &self,
        schema: &SchemaReadTransaction,
    ) -> Result<ModifyList<ModifyInvalid>, SchemaError> {
        // Create a modlist from this entry. We make this assuming we want the entry
        // to have this one as a subset of values. This means if we have single
        // values, we'll replace, if they are multivalue, we present them.
        let mut mods = ModifyList::new();

        for (k, vs) in self.attrs.iter() {
            // Get the schema attribute type out.
            match schema.is_multivalue(k) {
                Ok(r) => {
                    if !r {
                        // As this is single value, purge then present to maintain this
                        // invariant
                        mods.push_mod(Modify::Purged(k.clone()));
                    }
                }
                // TODO: Do something with this error properly.
                Err(e) => return Err(e),
            }
            for v in vs {
                mods.push_mod(Modify::Present(k.clone(), v.clone()));
            }
        }

        Ok(mods)
    }
}

impl<VALID, STATE> Entry<VALID, STATE> {
    pub fn get_ava(&self, attr: &String) -> Option<&Vec<String>> {
        self.attrs.get(attr)
    }

    pub fn attribute_pres(&self, attr: &str) -> bool {
        // FIXME: Do we need to normalise attr name?
        self.attrs.contains_key(attr)
    }

    pub fn attribute_equality(&self, attr: &str, value: &str) -> bool {
        // we assume based on schema normalisation on the way in
        // that the equality here of the raw values MUST be correct.
        // We also normalise filters, to ensure that their values are
        // syntax valid and will correctly match here with our indexes.

        // FIXME: Make this binary_search

        self.attrs.get(attr).map_or(false, |v| {
            v.iter()
                .fold(false, |acc, av| if acc { acc } else { value == av })
        })
    }

    pub fn attribute_substring(&self, _attr: &str, _subvalue: &str) -> bool {
        unimplemented!();
    }

    pub fn classes(&self) -> EntryClasses {
        // Get the class vec, if any?
        // How do we indicate "empty?"
        // FIXME: Actually handle this error ...
        let v = self
            .attrs
            .get("class")
            .map(|c| c.len())
            .expect("INVALID STATE, NO CLASS FOUND");
        let c = self.attrs.get("class").map(|c| c.iter());
        EntryClasses { size: v, inner: c }
    }

    pub fn avas(&self) -> EntryAvas {
        EntryAvas {
            inner: self.attrs.iter(),
        }
    }
}

impl<STATE> Entry<EntryInvalid, STATE>
where
    STATE: Copy,
{
    // This should always work? It's only on validate that we'll build
    // a list of syntax violations ...
    // If this already exists, we silently drop the event? Is that an
    // acceptable interface?
    // Should value here actually be a &str?
    pub fn add_ava(&mut self, attr: String, value: String) {
        // get_mut to access value
        // How do we make this turn into an ok / err?
        self.attrs
            .entry(attr)
            .and_modify(|v| {
                // Here we need to actually do a check/binary search ...
                // FIXME: Because map_err is lazy, this won't do anything on release
                match v.binary_search(&value) {
                    // It already exists, done!
                    Ok(_) => {}
                    Err(idx) => {
                        // This cloning is to fix a borrow issue with the or_insert below.
                        // Is there a better way?
                        v.insert(idx, value.clone())
                    }
                }
            })
            .or_insert(vec![value]);
    }

    pub fn remove_ava(&mut self, attr: &String, value: &String) {
        self.attrs
            // TODO: Fix this clone ...
            .entry(attr.clone())
            .and_modify(|v| {
                // Here we need to actually do a check/binary search ...
                // FIXME: Because map_err is lazy, this won't do anything on release
                match v.binary_search(&value) {
                    // It exists, rm it.
                    Ok(idx) => {
                        v.remove(idx);
                    }
                    // It does not exist, move on.
                    Err(_) => {}
                }
            });
    }

    pub fn purge_ava(&mut self, attr: &String) {
        self.attrs.remove(attr);
    }

    // FIXME: Should this collect from iter instead?
    /// Overwrite the existing avas.
    pub fn set_avas(&mut self, attr: String, values: Vec<String>) {
        // Overwrite the existing value
        let _ = self.attrs.insert(attr, values);
    }

    pub fn avas_mut(&mut self) -> EntryAvasMut {
        EntryAvasMut {
            inner: self.attrs.iter_mut(),
        }
    }

    // Should this be schemaless, relying on checks of the modlist, and the entry validate after?
    pub fn apply_modlist(
        &self,
        modlist: &ModifyList<ModifyValid>,
    ) -> Result<Entry<EntryInvalid, STATE>, ()> {
        // Apply a modlist, generating a new entry that conforms to the changes.
        // This is effectively clone-and-transform

        // clone the entry
        let mut ne: Entry<EntryInvalid, STATE> = Entry {
            valid: self.valid,
            state: self.state,
            id: self.id,
            attrs: self.attrs.clone(),
        };

        // mutate
        for modify in modlist {
            match modify {
                Modify::Present(a, v) => ne.add_ava(a.clone(), v.clone()),
                Modify::Removed(a, v) => ne.remove_ava(a, v),
                Modify::Purged(a) => ne.purge_ava(a),
            }
        }

        // return it
        Ok(ne)
    }
}

impl<VALID, STATE> PartialEq for Entry<VALID, STATE> {
    fn eq(&self, rhs: &Entry<VALID, STATE>) -> bool {
        // FIXME: This is naive. Later it should be schema
        // aware checking.
        self.attrs == rhs.attrs
    }
}

#[derive(Serialize, Deserialize, Debug)]
enum Credential {
    Password {
        name: String,
        hash: String,
    },
    TOTPPassword {
        name: String,
        hash: String,
        totp_secret: String,
    },
    SshPublicKey {
        name: String,
        data: String,
    },
}

#[derive(Serialize, Deserialize, Debug)]
struct User {
    username: String,
    // Could this be derived from self? Do we even need schema?
    class: Vec<String>,
    displayname: String,
    legalname: Option<String>,
    email: Vec<String>,
    // uuid?
    // need to support deref later ...
    memberof: Vec<String>,
    sshpublickey: Vec<String>,

    credentials: Vec<Credential>,
}

#[cfg(test)]
mod tests {
    use super::{Entry, EntryInvalid, EntryNew};
    use modify::{Modify, ModifyList};
    use serde_json;

    #[test]
    fn test_entry_basic() {
        let mut e: Entry<EntryInvalid, EntryNew> = Entry::new();

        e.add_ava(String::from("userid"), String::from("william"));

        let _d = serde_json::to_string_pretty(&e).unwrap();
    }

    #[test]
    fn test_entry_dup_value() {
        // Schema doesn't matter here because we are duplicating a value
        // it should fail!

        // We still probably need schema here anyway to validate what we
        // are adding ... Or do we validate after the changes are made in
        // total?
        let mut e: Entry<EntryInvalid, EntryNew> = Entry::new();
        e.add_ava(String::from("userid"), String::from("william"));
        e.add_ava(String::from("userid"), String::from("william"));

        let values = e.get_ava(&String::from("userid")).unwrap();
        // Should only be one value!
        assert_eq!(values.len(), 1)
    }

    #[test]
    fn test_entry_pres() {
        let mut e: Entry<EntryInvalid, EntryNew> = Entry::new();
        e.add_ava(String::from("userid"), String::from("william"));

        assert!(e.attribute_pres("userid"));
        assert!(!e.attribute_pres("name"));
    }

    #[test]
    fn test_entry_equality() {
        let mut e: Entry<EntryInvalid, EntryNew> = Entry::new();

        e.add_ava(String::from("userid"), String::from("william"));

        assert!(e.attribute_equality("userid", "william"));
        assert!(!e.attribute_equality("userid", "test"));
        assert!(!e.attribute_equality("nonexist", "william"));
    }

    #[test]
    fn test_entry_apply_modlist() {
        // Test application of changes to an entry.
        let mut e: Entry<EntryInvalid, EntryNew> = Entry::new();
        e.add_ava(String::from("userid"), String::from("william"));

        let mods = unsafe {
            ModifyList::new_valid_list(vec![Modify::Present(
                String::from("attr"),
                String::from("value"),
            )])
        };

        let ne = e.apply_modlist(&mods).unwrap();

        // Assert the changes are there
        assert!(ne.attribute_equality("attr", "value"));

        // Assert present for multivalue
        // Assert purge on single/multi/empty value
        // Assert removed on value that exists and doesn't exist
    }
}