Each cloud storage system is slightly different. Rclone attempts to provide a unified interface to them, but some underlying differences show through.
Here is an overview of the major features of each cloud storage system.
|Name||Hash||ModTime||Case Insensitive||Duplicate Files||MIME Type||Metadata|
|Akamai Netstorage||MD5, SHA256||R/W||No||No||R||-|
|Amazon S3 (or S3 compatible)||MD5||R/W||No||No||R/W||RWU|
|Enterprise File Fabric||-||R/W||Yes||No||R/W||-|
|Google Cloud Storage||MD5||R/W||No||No||R/W||-|
|Internet Archive||MD5, SHA1, CRC32||R/W ¹¹||No||No||-||RWU|
|Mail.ru Cloud||Mailru ⁶||R/W||Yes||No||-||-|
|Microsoft Azure Blob Storage||MD5||R/W||No||No||R/W||-|
|Microsoft OneDrive||SHA1 ⁵||R/W||Yes||No||R||-|
|pCloud||MD5, SHA1 ⁷||R||No||No||W||-|
|SFTP||MD5, SHA1 ²||R/W||Depends||No||-||-|
|WebDAV||MD5, SHA1 ³||R ⁴||Depends||No||-||-|
|The local filesystem||All||R/W||Depends||No||-||RWU|
¹ Dropbox supports its own custom hash. This is an SHA256 sum of all the 4 MiB block SHA256s.
² SFTP supports checksums if the same login has shell access and
sha1sum as well as
echo are in the remote's PATH.
³ WebDAV supports hashes when used with Owncloud and Nextcloud only.
⁴ WebDAV supports modtimes when used with Owncloud and Nextcloud only.
⁵ Microsoft OneDrive Personal supports SHA1 hashes, whereas OneDrive for business and SharePoint server support Microsoft's own QuickXorHash.
⁶ Mail.ru uses its own modified SHA1 hash
⁷ pCloud only supports SHA1 (not MD5) in its EU region
⁸ Opendrive does not support creation of duplicate files using
their web client interface or other stock clients, but the underlying
storage platform has been determined to allow duplicate files, and it
is possible to create them with
rclone. It may be that this is a
mistake or an unsupported feature.
⁹ QingStor does not support SetModTime for objects bigger than 5 GiB.
¹⁰ FTP supports modtimes for the major FTP servers, and also others if they advertised required protocol extensions. See this for more details.
¹¹ Internet Archive requires option
wait_archive to be set to a non-zero value
for full modtime support.
¹² HiDrive supports its own custom hash. It combines SHA1 sums for each 4 KiB block hierarchically to a single top-level sum.
The cloud storage system supports various hash types of the objects.
The hashes are used when transferring data as an integrity check and
can be specifically used with the
--checksum flag in syncs and in
To use the verify checksums when transferring between cloud storage systems they must support a common hash type.
Almost all cloud storage systems store some sort of timestamp
on objects, but several of them not something that is appropriate
to use for syncing. E.g. some backends will only write a timestamp
that represent the time of the upload. To be relevant for syncing
it should be able to store the modification time of the source
object. If this is not the case, rclone will only check the file
size by default, though can be configured to check the file hash
--checksum flag). Ideally it should also be possible to
change the timestamp of an existing file without having to re-upload it.
Storage systems with a
- in the ModTime column, means the
modification read on objects is not the modification time of the
file when uploaded. It is most likely the time the file was uploaded,
or possibly something else (like the time the picture was taken in
Storage systems with a
R (for read-only) in the ModTime column,
means the it keeps modification times on objects, and updates them
when uploading objects, but it does not support changing only the
modification time (
SetModTime operation) without re-uploading,
possibly not even without deleting existing first. Some operations
in rclone, such as
sync commands, will automatically
SetModTime support and re-upload if necessary to keep
the modification times in sync. Other commands will not work
SetModTime support, e.g.
touch command on an existing
file will fail, and changes to modification time only on a files
mount will be silently ignored.
Storage systems with
R/W (for read/write) in the ModTime column,
means they do also support modtime-only operations.
If a cloud storage systems is case sensitive then it is possible to
have two files which differ only in case, e.g.
FILE.txt. If a cloud storage system is case insensitive then that
This can cause problems when syncing between a case insensitive system and a case sensitive system. The symptom of this is that no matter how many times you run the sync it never completes fully.
The local filesystem and SFTP may or may not be case sensitive depending on OS.
Most of the time this doesn't cause any problems as people tend to avoid files whose name differs only by case even on case sensitive systems.
If a cloud storage system allows duplicate files then it can have two objects with the same name.
This confuses rclone greatly when syncing - use the
command to rename or remove duplicates.
Some cloud storage systems might have restrictions on the characters
that are usable in file or directory names.
rclone detects such a name during a file upload, it will
transparently replace the restricted characters with similar looking
Unicode characters. To handle the different sets of restricted characters
for different backends, rclone uses something it calls encoding.
This process is designed to avoid ambiguous file names as much as possible and allow to move files between many cloud storage systems transparently.
The name shown by
rclone to the user or during log output will only
contain a minimal set of replaced characters
to ensure correct formatting and not necessarily the actual name used
on the cloud storage.
This transformation is reversed when downloading a file or parsing
rclone arguments. For example, when uploading a file named
to Onedrive, it will be displayed as
my file?.txt on the console, but
my file？.txt to Onedrive (the
? gets replaced by the similar
？ character, the so-called "fullwidth question mark").
The reverse transformation allows to read a file
from Google Drive, by passing the name
unusual／name.txt on the command line
/ needs to be replaced by the similar looking
The filename encoding system works well in most cases, at least where file names are written in English or similar languages. You might not even notice it: It just works. In some cases it may lead to issues, though. E.g. when file names are written in Chinese, or Japanese, where it is always the Unicode fullwidth variants of the punctuation marks that are used.
On Windows, the characters
? are examples of restricted
characters. If these are used in filenames on a remote that supports it,
Rclone will transparently convert them to their fullwidth Unicode
： when downloading to Windows, and back again
when uploading. This way files with names that are not allowed on Windows
can still be stored.
However, if you have files on your Windows system originally with these same
Unicode characters in their names, they will be included in the same conversion
process. E.g. if you create a file in your Windows filesystem with name
： is the Unicode fullwidth colon symbol, and use
rclone to upload it to Google Drive, which supports regular
question mark), rclone will replace the fullwidth
: with the
: and store the file as
Test:1.jpg in Google Drive. Since
both Windows and Google Drive allows the name
Test：1.jpg, it would
probably be better if rclone just kept the name as is in this case.
With the opposite situation; if you have a file named
in your Google Drive, e.g. uploaded from a Linux system where
: is valid
in file names. Then later use rclone to copy this file to your Windows
computer you will notice that on your local disk it gets renamed
Test：1.jpg. The original filename is not legal on Windows, due to
:, and rclone therefore renames it to make the copy possible.
That is all good. However, this can also lead to an issue: If you already
had a different file named
Test：1.jpg on Windows, and then use rclone
to copy either way. Rclone will then treat the file originally named
Test:1.jpg on Google Drive and the file originally named
on Windows as the same file, and replace the contents from one with the other.
Its virtually impossible to handle all cases like these correctly in all situations, but by customizing the encoding option, changing the set of characters that rclone should convert, you should be able to create a configuration that works well for your specific situation. See also the example below.
(Windows was used as an example of a file system with many restricted characters, and Google drive a storage system with few.)
The table below shows the characters that are replaced by default.
When a replacement character is found in a filename, this character
will be escaped with the
‛ character to avoid ambiguous file names.
(e.g. a file named
␀.txt would shown as
Each cloud storage backend can use a different set of characters, which will be specified in the documentation for each backend.
The default encoding will also encode these file names as they are problematic with many cloud storage systems.
Some backends only support a sequence of well formed UTF-8 bytes as file or directory names.
In this case all invalid UTF-8 bytes will be replaced with a quoted
representation of the byte value to allow uploading a file to such a
backend. For example, the invalid byte
0xFE will be encoded as
A common source of invalid UTF-8 bytes are local filesystems, that store names in a different encoding than UTF-8 or UTF-16, like latin1. See the local filenames section for details.
Most backends have an encoding option, specified as a flag
backend is the name of the backend, or as
a config parameter
encoding (you'll need to select the Advanced
rclone config to see it).
This will have default value which encodes and decodes characters in such a way as to preserve the maximum number of characters (see above).
However this can be incorrect in some scenarios, for example if you
have a Windows file system with Unicode fullwidth characters
：, that you want to remain as those characters on the
remote rather than being translated to regular (halfwidth)
--backend-encoding flags allow you to change that. You can
disable the encoding completely with
--backend-encoding None or set
encoding = None in the config file.
Encoding takes a comma separated list of encodings. You can see the
list of all possible values by passing an invalid value to this
--local-encoding "help". The command
rclone help flags encoding
will show you the defaults for the backends.
|CrLf||CR 0x0D, LF 0x0A||
|Ctl||All control characters 0x00-0x1F||
|InvalidUtf8||An invalid UTF-8 character (e.g. latin1)||
|LeftCrLfHtVt||CR 0x0D, LF 0x0A, HT 0x09, VT 0x0B on the left of a string||
|LeftSpace||SPACE on the left of a string||
|None||No characters are encoded|
|RightCrLfHtVt||CR 0x0D, LF 0x0A, HT 0x09, VT 0x0B on the right of a string||
|RightSpace||SPACE on the right of a string||
To take a specific example, the FTP backend's default encoding is
However, let's say the FTP server is running on Windows and can't have any of the invalid Windows characters in file names. You are backing up Linux servers to this FTP server which do have those characters in file names. So you would add the Windows set which are
to the existing ones, giving:
This can be specified using the
--ftp-encoding flag or using an
encoding parameter in the config file.
As a nother example, take a Windows system where there is a file with
： is the Unicode fullwidth colon symbol.
When using rclone to copy this to a remote which supports
the regular (halfwidth) colon (such as Google Drive), you will notice
that the file gets renamed to
To avoid this you can change the set of characters rclone should convert
for the local filesystem, using command-line argument
Rclone's default behavior on Windows corresponds to
If you want to use fullwidth characters
？ in your filenames
without rclone changing them when uploading to a remote, then set the same as
the default value but without
Alternatively, you can disable the conversion of any characters with
Instead of using command-line argument
--local-encoding, you may also set it
as environment variable
or configure a remote of type
local in your config,
and set the
encoding option there.
The risk by doing this is that if you have a filename with the regular (halfwidth)
? in your cloud storage, and you try to download
it to your Windows filesystem, this will fail. These characters are not
valid in filenames on Windows, and you have told rclone not to work around
this by converting them to valid fullwidth variants.
MIME types (also known as media types) classify types of documents
using a simple text classification, e.g.
Some cloud storage systems support reading (
R) the MIME type of
objects and some support writing (
W) the MIME type of objects.
The MIME type can be important if you are serving files directly to HTTP from the storage system.
If you are copying from a remote which supports reading (
R) to a
remote which supports writing (
W) then rclone will preserve the MIME
types. Otherwise they will be guessed from the extension, or the
remote itself may assign the MIME type.
Backends may or may support reading or writing metadata. They may support reading and writing system metadata (metadata intrinsic to that backend) and/or user metadata (general purpose metadata).
The levels of metadata support are
||Read only System Metadata|
||Read and write System Metadata|
||Read and write System Metadata and read and write User Metadata|
See the metadata docs for more info.
All rclone remotes support a base command set. Other features depend upon backend-specific capabilities.
|Amazon S3 (or S3 compatible)||No||Yes||No||No||Yes||Yes||Yes||Yes||No||No|
|Enterprise File Fabric||Yes||Yes||Yes||Yes||Yes||No||No||No||No||Yes|
|Google Cloud Storage||Yes||Yes||No||No||No||Yes||Yes||No||No||No|
|Microsoft Azure Blob Storage||Yes||Yes||No||No||No||Yes||Yes||No||No||No|
|OpenStack Swift||Yes †||Yes||No||No||No||Yes||Yes||No||Yes||No|
|The local filesystem||Yes||No||Yes||Yes||No||No||Yes||No||Yes||Yes|
This deletes a directory quicker than just deleting all the files in the directory.
† Note Swift, Hubic, and Storj implement this in order to delete directory markers but they don't actually have a quicker way of deleting files other than deleting them individually.
‡ StreamUpload is not supported with Nextcloud
Used when copying an object to and from the same remote. This known
as a server-side copy so you can copy a file without downloading it
and uploading it again. It is used if you use
rclone copy or
rclone move if the remote doesn't support
If the server doesn't support
Copy directly then for copy operations
the file is downloaded then re-uploaded.
Used when moving/renaming an object on the same remote. This is known
as a server-side move of a file. This is used in
rclone move if the
server doesn't support
If the server isn't capable of
Move then rclone simulates it with
Copy then delete. If the server doesn't support
Copy then rclone
will download the file and re-upload it.
This is used to implement
rclone move to move a directory if
possible. If it isn't then it will use
Move on each file (which
falls back to
Copy then download and upload - see
This is used for emptying the trash for a remote by
If the server can't do
rclone cleanup will return an
‡‡ Note that while Box implements this it has to delete every file individually so it will be slower than emptying the trash via the WebUI
The remote supports a recursive list to list all the contents beneath
a directory quickly. This enables the
--fast-list flag to work.
See the rclone docs for more details.
Some remotes allow files to be uploaded without knowing the file size
in advance. This allows certain operations to work without spooling the
file to local disk first, e.g.
Sets the necessary permissions on a file or folder and prints a link that allows others to access them, even if they don't have an account on the particular cloud provider.
about prints quota information for a remote. Typical output
includes bytes used, free, quota and in trash.
If a remote lacks about capability
rclone about remote:returns
Backends without about capability cannot determine free space for an
rclone mount, or use policy
mfs (most free space) as a member of an
rclone union remote.
The remote supports empty directories. See Limitations for details. Most Object/Bucket-based remotes do not support this.