diff --git a/matter/src/data_model/core/mod.rs b/matter/src/data_model/core/mod.rs
index 84f67f0..ce593a5 100644
--- a/matter/src/data_model/core/mod.rs
+++ b/matter/src/data_model/core/mod.rs
@@ -228,7 +228,7 @@ pub mod subscribe;
 /// Type of Resume Request
 enum ResumeReq {
     Subscribe(subscribe::SubsCtx),
-    Read,
+    Read(read::ResumeReadReq),
 }
 
 impl objects::ChangeConsumer for DataModel {
@@ -263,8 +263,15 @@ impl InteractionConsumer for DataModel {
         trans: &mut Transaction,
         tw: &mut TLVWriter,
     ) -> Result<(), Error> {
-        let is_chunked = self.handle_read_attr_array(req, trans, tw)?;
-        if !is_chunked {
+        let resume = self.handle_read_attr_array(req, trans, tw)?;
+        if let Some(resume) = resume {
+            // This is a multi-hop read transaction, remember this read request
+            if !trans.exch.is_data_none() {
+                error!("Exchange data already set, and multi-hop read");
+                return Err(Error::InvalidState);
+            }
+            trans.exch.set_data_boxed(Box::new(ResumeReq::Read(resume)));
+        } else {
             tw.bool(TagType::Context(SupressResponse as u8), true)?;
             // Mark transaction complete, if not chunked
             trans.complete();
@@ -311,7 +318,7 @@ impl InteractionConsumer for DataModel {
     ) -> Result<(OpCode, ResponseRequired), Error> {
         if let Some(resume) = trans.exch.take_data_boxed::<ResumeReq>() {
             match *resume {
-                ResumeReq::Read => Ok((OpCode::Reserved, ResponseRequired::No)),
+                ResumeReq::Read(_) => Ok((OpCode::Reserved, ResponseRequired::No)),
                 ResumeReq::Subscribe(mut ctx) => {
                     let result = self.handle_subscription_confirm(trans, tw, &mut ctx)?;
                     trans.exch.set_data_boxed(resume);
diff --git a/matter/src/data_model/core/read.rs b/matter/src/data_model/core/read.rs
index 2a12d6b..106094b 100644
--- a/matter/src/data_model/core/read.rs
+++ b/matter/src/data_model/core/read.rs
@@ -16,6 +16,8 @@
  */
 
 use crate::data_model::{core::DataModel, objects::*};
+use crate::transport::packet::Packet;
+use crate::utils::writebuf::WriteBuf;
 use crate::{
     acl::{AccessReq, Accessor},
     error::*,
@@ -104,12 +106,7 @@ impl<'a, 'b, 'c> Encoder for AttrReadEncoder<'a, 'b, 'c> {
 pub struct ResumeReadReq {
     /// The Read Request Attribute Path that caused chunking, and this is the path
     /// that needs to be resumed.
-    ///
-    /// TODO: Ideally, the entire ReadRequest (with any subsequent AttrPaths) should also
-    /// be maintained. But for now, we just store the AttrPath that caused the overflow
-    /// and chunking. Hopefully, the other end requests any pending paths when it sees no
-    /// more chunks.
-    pending_path: GenericPath,
+    pending_req: Option<Packet<'static>>,
 
     /// The Attribute that couldn't be encoded because our buffer got full. The next chunk
     /// will start encoding from this attribute onwards.
@@ -198,7 +195,7 @@ impl DataModel {
         read_req: &ReadReq,
         trans: &mut Transaction,
         tw: &mut TLVWriter,
-    ) -> Result<bool, Error> {
+    ) -> Result<Option<ResumeReadReq>, Error> {
         let mut resume_read_req: ResumeReadReq = Default::default();
 
         let mut attr_encoder = AttrReadEncoder::new(tw);
@@ -226,19 +223,31 @@ impl DataModel {
                     &mut attr_details,
                     &mut resume_read_req.resume_encode,
                 );
-                if result.is_err() {
-                    resume_read_req.pending_path = attr_path.to_gp();
-                    break;
-                }
             }
             tw.end_container()?;
             if result.is_err() {
                 // If there was an error, indicate chunking. The resume_read_req would have been
                 // already populated from in the loop above.
                 tw.bool(TagType::Context(MoreChunkedMsgs as u8), true)?;
-                return Ok(true);
+                // Retain the entire request, because we need the data-filters, and subsequent attr-reads, if any
+                // when we resume this read in the next hop
+                resume_read_req.pending_req = Some(copy_read_req_to_packet(read_req)?);
+                return Ok(Some(resume_read_req));
             }
         }
-        Ok(false)
+        Ok(None)
     }
 }
+
+fn copy_read_req_to_packet(read_req: &ReadReq) -> Result<Packet<'static>, Error> {
+    let mut packet = Packet::new_rx()?;
+    let backup = packet.as_borrow_slice();
+    let backup_len = backup.len();
+    let mut wb = WriteBuf::new(backup, backup_len);
+    let mut tw = TLVWriter::new(&mut wb);
+    // TODO: This is unnecessarily wasteful, could directly copy &[u8] if accessible
+    read_req.to_tlv(&mut tw, TagType::Anonymous)?;
+    let data_len = wb.as_borrow_slice().len();
+    packet.get_parsebuf()?.set_len(data_len);
+    Ok(packet)
+}