Following the launch of version 4.2 of the official apps, Telegram client apps may be required to download popular files that were published in public channels with more than 100,000 members from secondary Content Delivery Network data centers. These CDN DCs are located in regions with significant Telegram traffic where we wouldn't want to place Telegram servers for various reasons.
The CDN DCs are not a part of the Telegram cloud and should be considered enemy territory. For this reason, each file that is to be sent to these CDN DCs is encrypted with a unique key using AES-256-CTR encryption. The CDN can't access the data it stores because these keys are only accessible to the main MTProto server and to the authorized client.
When a file from a public channel with ~100,000 members becomes popular in a particular region, the Telegram server may encrypt this file with a unique AES-256-CTR key and send it to a relevant CDN DC for storage.
When a file is stored in a CDN DC close to the end user, the download speed will be much higher because the data needs to travel smaller distances and will likely avoid many bottlenecks that exist between regions.
This is secure because CDN DCs are treated the same way as internet providers / random third parties:
CDNs are very limited when it comes to communication: the master data center only uploads encrypted files for storage and will accept no data from the CDN. The client apps only download encrypted files and accept no other updates. The client apps obtain the keys necessary to decrypt the file from the main Telegram server and verify the integrity of the file by its hash, which means that the CDN may only supply the correct file – anything different will be immediately discarded by the client.
CDN DCs do not store files on hard disks – only in memory. When a CDN server runs out of memory, a simple LRU algorithm is used to replace the least popular files with new ones.
The API may return the upload.fileCdnRedirect constructor after an upload.getFile query. In this case, the client must request the required file from a CDN DC. The dc_id in the response is the id of the new CDN. The IP address for the connection will be available in help.getConfig, same as with the master DCs. The corresponding dcOption will have the flag cdn:flags.3?true.
Once a successful connection to the CDN-dc_id is established, the client must generate an auth_key (after confirming that the public RSA MTProto key of the CDN DC matches one from the list returned in help.getCdnConfig). Then the client must perform an upload.getCdnFile for each offset. For files of an unknown size it is necessary to repeat the query until an empty reply is returned.
upload.getCdnFile may return the upload.cdnFileReuploadNeeded constructor. In this case, the client needs to send an upload.reuploadCdnFile request to the DC that got the original upload.getFile request. Once upload.reuploadCdnFile is successfull, the app needs to request the file from the CDN DC again.
The main DC for a file is the DC where its main copy is stored (not to be confused with the main DC of the user) – either userProfilePhoto.dc_id, chatPhoto.dc_id, photo.dc_id, or document.dc_id.
In upload.fileCdnRedirect, the server sends a decryption key and IV for the file (the fields
Having received a portion of encrypted data from the CDN DC inside upload.cdnFile, the client must decrypt this data using
AES-256-CTR. For IV, it should use the value of
encryption_iv, modified in the following manner: for each
offset replace the last 4 bytes of the
offset / 16 in big-endian. This allows to effectively decrypt a file and to use random access to a file's content (e.g., for streaming).
In order to confirm that the CDN DC passed an untampered file, clients must verify hashes for each downloaded part. upload.fileCdnRedirect, upload.reuploadCdnFile and upload.getCdnFileHashes contain FileHash constructors. Each of these constructors contains the SHA-256 hash of a part of the file that starts with
offset and takes
Before saving each portion of the data received from the CDN DC into the file, the client must confirm that its hash matches the hash that was received from the master DC. If missing a hash for any file part, client developers must use the upload.getCdnFileHashes method to obtain the missing hash.
fileHash#6242c773 offset:int limit:int hash:bytes = FileHash; upload.fileCdnRedirect#f18cda44 dc_id:int file_token:bytes encryption_key:bytes encryption_iv:bytes file_hashes:Vector<FileHash> = upload.File; upload.cdnFileReuploadNeeded#eea8e46e request_token:bytes = upload.CdnFile; upload.cdnFile#a99fca4f bytes:bytes = upload.CdnFile; dcOption#18b7a10d flags:# ipv6:flags.0?true media_only:flags.1?true tcpo_only:flags.2?true cdn:flags.3?true static:flags.4?true id:int ip_address:string port:int secret:flags.10?bytes = DcOption; cdnPublicKey#c982eaba dc_id:int public_key:string = CdnPublicKey; cdnConfig#5725e40a public_keys:Vector<CdnPublicKey> = CdnConfig; ---functions--- // CDN DC upload.getCdnFile#2000bcc3 file_token:bytes offset:int limit:int = upload.CdnFile; // Master DC upload.reuploadCdnFile#9b2754a8 file_token:bytes request_token:bytes = Vector<FileHash>; upload.getCdnFileHashes#4da54231 file_token:bytes offset:int = Vector<FileHash>; help.getCdnConfig#52029342 = CdnConfig;
offsetmust be divisible by 4096 bytes
limitmust be divisible by 4096 bytes
offset / (1024 * 1024) == (offset + limit - 1) / (1024 * 1024)
|upload.getCdnFile||FILE_TOKEN_INVALID||The CDN DC did not accept the
|upload.reuploadCdnFile||FILE_TOKEN_INVALID||The master DC did not accept the
|upload.reuploadCdnFile||REQUEST_TOKEN_INVALID||The master DC did not accept the