rsyncd.conf(5)

📦
version 3.5.2 · 13 versions
, in arch rolling

💖 NAME: rsyncd.conf(5) (User Commands manual) 💖

rsyncd.conf — the configuration file for rsync when it's running in daemon mode. Yes bestie, daemon mode. It's literally the backstage pass for your server-side rsync setup. 🎀

✨ SYNOPSIS ✨

rsyncd.conf

The online version of this manpage (that includes cross-linking of topics) is available at https://download.samba.org/pub/rsync/rsyncd.conf.5.

🌸 DESCRIPTION 🌸

Okay so the rsyncd.conf file is the runtime configuration file for rsync when run as an rsync daemon. It controls authentication, access, logging and available modules. Like, it's giving main character energy for your whole daemon vibe check. 💅

📝 FILE FORMAT 📝

The file consists of modules and parameters. A module begins with the name of the module in square brackets and continues until the next module begins. Modules contain parameters of the form name = value.

The file is line-based -- that is, each newline-terminated line represents either a comment, a module name or a parameter.

Only the first equals sign in a parameter is significant. Whitespace before or after the first equals sign is discarded. Leading, trailing and internal whitespace in module and parameter names is irrelevant. Leading and trailing whitespace in a parameter value is discarded. Internal whitespace within a parameter value is retained verbatim.

Any line beginning with a hash (#) is ignored, as are lines containing only whitespace. (If a hash occurs after anything other than leading whitespace, it is considered a part of the line's content.)

Any line ending in a \ is "continued" on the next line in the customary UNIX fashion.

The values following the equals sign in parameters are all either a string (no quotes needed) or a boolean, which may be given as yes/no, 0/1 or true/false. Case is not significant in boolean values, but is preserved in string values.

🚀 LAUNCHING THE RSYNC DAEMON 🚀

The rsync daemon is launched by specifying the --daemon option to rsync.

The daemon must run with root privileges if you wish to use chroot, to bind to a port numbered under 1024 (as is the default 873), or to set file ownership. Otherwise, it must just have permission to read and write the appropriate data, log, and lock files.

You can launch it either via inetd, as a stand-alone daemon, or from an rsync client via a remote shell. If run as a stand-alone daemon then just run the command rsync --daemon from a suitable startup script.

Systems using systemd can use the example unit files in the source tree's packaging/systemd directory. The rsync.service file runs a stand-alone daemon using rsync --daemon --no-detach, while rsync.socket and rsync@.service show a socket-activated setup for incoming connections. These files may need local adjustment to match your installed rsync path, packaging layout, and module policy.

When run via inetd you should add a line like this to /etc/services:

rsync           873/tcp

and a single line something like this to /etc/inetd.conf:

rsync   stream  tcp     nowait  root   /usr/bin/rsync rsyncd --daemon

Replace /usr/bin/rsync with the path to where you have rsync installed on your system. You will then need to send inetd a HUP signal to tell it to reread its config file.

⚠️ Note that you should not send the rsync daemon a HUP signal to force it to reread the rsyncd.conf file. The file is re-read on each client connection. Respect the vibe: no HUP needed. 💀

🌍 GLOBAL PARAMETERS 🌍

The first parameters in the file (before a [module] header) are the global parameters:

  • motd file — This parameter allows you to specify a "message of the day" (MOTD) to display to clients on each connect. This usually contains site information and any legal notices. The default is no MOTD file. This can be overridden by the --dparam=motdfile=FILE command-line option when starting the daemon.
  • pid file — This parameter tells the rsync daemon to write its process ID to that file. The rsync keeps the file locked so that it can know when it is safe to overwrite an existing file.

The filename can be overridden by the --dparam=pidfile=FILE command-line option when starting the daemon.

  • port — You can override the default port the daemon will listen on by specifying this value (defaults to 873). This is ignored if the daemon is being run by inetd, and is superseded by the --port command-line option.
  • address — You can override the default IP address the daemon will listen on by specifying this value. This is ignored if the daemon is being run by inetd, and is superseded by the --address command-line option.
  • socket options — This parameter can provide endless fun for people who like to tune their systems to the utmost degree. You can set all sorts of socket options which may make transfers faster (or slower!). Read the manpage for the setsockopt() system call for details on some of the options you may be able to set. By default no special socket options are set. These settings can also be specified via the --sockopts command-line option.
  • listen backlog — You can override the default backlog value when the daemon listens for connections. It defaults to 5.

You may also include any MODULE PARAMETERS in the global part of the config file, in which case the supplied value will override the default for that parameter.

You may use references to environment variables in the values of parameters. String parameters will have %VAR% references expanded as late as possible (when the string is first used in the program), allowing for the use of variables that rsync sets at connection time, such as RSYNC_USER_NAME. Non-string parameters (such as true/false settings) are expanded when read from the config file. If a variable does not exist in the environment, or if a sequence of characters is not a valid reference (such as an un-paired percent sign), the raw characters are passed through unchanged. This helps with backward compatibility and safety (e.g. expanding a non-existent %VAR% to an empty string in a path could result in a very unsafe path). The safest way to insert a literal % into a value is to use %%.

🧩 MODULE PARAMETERS 🧩

After the global parameters you should define a number of modules, each module exports a directory tree as a symbolic name. Modules are exported by specifying a module name in square brackets [module] followed by the parameters for that module. The module name cannot contain a slash or a closing square bracket. If the name contains whitespace, each internal sequence of whitespace will be changed into a single space, while leading or trailing whitespace will be discarded.

There is also a special module name of [global] that does not define a module but instead switches back to the global settings context where default parameters can be specified. Because each defined module gets its full set of parameters as a combination of the default values that are set at that position in the config file plus its own parameter list, the use of a [global] section can help to maintain shared config values for multiple modules.

As with GLOBAL PARAMETERS, you may use references to environment variables in the values of parameters. See that section for details.

  • comment — This parameter specifies a description string that is displayed next to the module name when clients obtain a list of available modules. The default is no comment.
  • path — This parameter specifies the directory in the daemon's filesystem to make available in this module. You must specify this parameter for each module in rsyncd.conf.

If the value contains a /./ element then the path will be divided at that point into a chroot dir and an inner-chroot subdir. If use chroot is set to false, though, the extraneous dot dir is just cleaned out of the path. An example of this idiom is:

path = /var/rsync/./module1

This will (when chrooting) chroot to /var/rsync and set the inside-chroot path to /module1.

You may base the path's value off of an environment variable by surrounding the variable name with percent signs. You can even reference a variable that is set by rsync when the user connects. For example, this would use the authorizing user's name in the path:

path = /home/%RSYNC_USER_NAME%

It is fine if the path includes internal spaces -- they will be retained verbatim (which means that you shouldn't try to escape them). If your final directory has a trailing space (and this is somehow not something you wish to fix), append a trailing slash to the path to avoid losing the trailing whitespace.

  • use chroot — If use chroot is true, the rsync daemon will chroot to the path before starting the file transfer with the client. This has the advantage of extra protection against possible implementation security holes, but it has the disadvantages of requiring super-user privileges, of not being able to follow symbolic links that are either absolute or outside of the new root path, and of complicating the preservation of users and groups by name (see below).

If use chroot is not set, it defaults to trying to enable a chroot but allows the daemon to continue (after logging a warning) if it fails. The one exception to this is when a module's path has a /./ chroot divider in it -- this causes an unset value to be treated as true for that module.

Prior to rsync 3.2.7, the default value was "true". The new "unset" default makes it easier to setup an rsync daemon as a non-root user or to run a daemon on a system where chroot fails. Explicitly setting the value to "true" in rsyncd.conf will always require the chroot to succeed.

It is also possible to specify a dot-dir in the module's path to indicate that you want to chdir to the earlier part of the path and then serve files from inside the latter part of the path (with sanitizing and default symlink munging). This can be useful if you need some library dirs inside the chroot (typically for uid & gid lookups) but don't want to put the lib dir into the top of the served path (even though they can be hidden with an exclude directive). However, a better choice for a modern rsync setup is to use a name converter and try to avoid inner lib dirs altogether. See also the daemon chroot parameter, which causes rsync to chroot into its own chroot area before doing any path-related chrooting.

...okay so picking right up: if the daemon is serving the / dir (either directly, or because it got chrooted into the module's path), rsync does not do any path sanitizing or (default) munging. Nothing to guard — you're already at the root, bestie.

But the moment it has to limit access to a particular subdir (either because chroot is disabled, or because there's an inside-chroot path set), rsync gets all protective: it munges symlinks (by default) and sanitizes paths. And if munged symlinks give you the ick, you can disable the symlink munging via the munge symlinks parameter — incoming symlink values then get sanitized instead (peep that parameter below 👀).

When rsync is sanitizing paths, it resolves away the .. path elements that would take an arg above the module hierarchy. For an operator-supplied option path like --backup-dir & --compare-dest, a leading slash is also substituted with the module's path (so an absolute path is interpreted as rooted in the module's path dir); for the transfer's own source args, a leading slash is just dropped. Simple as that. 💅

When a chroot is in effect and the name converter parameter is not set, the numeric ids parameter will default to being enabled (which disables name lookups). Which means: if you manually set up name-lookup libraries in your chroot (instead of using a name converter), you need to explicitly set numeric ids = false for rsync to do name lookups. No cap.

If you copy library resources into the module's chroot area, you should protect them through your OS's normal user/group or ACL settings (so the rsync module's user can't change them), and then hide them from the user's view via exclude (the discussion of that parameter shows you how). However — it's easier and safer to just set up a name converter. Trust. 🫶

✨ daemon chroot ✨

This parameter specifies a path to which the daemon will chroot before beginning communication with clients. Module paths (and any use chroot settings) will then be related to this one. It's basically the "how much of me do you want locked in a box" dial: this setting chroots the whole daemon, use chroot chroots just the transfers, or you can do both. 🎀

Heads up, bestie: the daemon chroot area may need various OS/lib/etc files installed for the daemon to function at all. By default the daemon runs with no chrooting whatsoever.

✨ proxy protocol ✨

When this parameter is enabled, every incoming connection must start with a V1 or V2 proxy protocol header. No header? Connection closed. Bye. 👋

Setting this to true requires a proxy server to forward source IP information to rsync, so you can log proper IP/host info and actually use client-oriented IP restrictions. The default of false means the IP info comes straight from the socket's metadata. If rsync isn't behind a proxy, this should be disabled — fr fr.

CAUTION: using this option can be dangerous if you don't ensure that only the proxy is allowed to connect to the rsync port. If any non-proxied connections get through, the client can use a modified rsync to spoof any remote IP address they want. Lock it down with something like iptables -uid-owner root rules (for strict localhost access), various firewall rules, or you can require password authorization so that any spoofing by users won't grant extra access. 🔥

This setting is global. If you need some modules to require this and not others, you'll need to set up multiple rsync daemon processes on different ports. Slightly annoying but that's the vibe.

✨ proxy protocol hosts ✨

This global parameter lists the socket peer IP addresses that are allowed to supply a proxy protocol header. The syntax is the same token format used by hosts allow. When proxy protocol = true, this list must match the direct peer before rsync trusts the forwarded client address. Vibe check on the peer first, always. 🕵️‍♀️

✨ name converter ✨

This parameter lets you specify a program that the rsync daemon will run to do user & group conversions between names & ids. The script is started prior to any chroot being set up, and it runs as the daemon user (not the transfer user). You can give a fully qualified pathname or just a program name that's on the $PATH. 🌈

The program can do normal user & group lookups without you having to shove any extra files into the module's chroot area — or you can do fully customized conversions, go off. 👑

The nameconvert program has access to all of the environment variables described in the section on pre-xfer exec. Super useful if you want to customize the conversion using info about the module and/or the copy request.

There's a sample python script in the support dir named nameconvert that implements the normal user & group lookups. Customize it, or just use it as documentation to write your own. Either way, iconic. 🐍

✨ numeric ids ✨

Enabling this parameter disables the mapping of users and groups by name for the current daemon module. It stops the daemon from trying to load any user/group-related files or libraries. It makes the transfer behave as if the client had passed the --numeric-ids command-line option.

By default, this parameter is enabled for chroot modules and disabled for non-chroot modules. Also keep in mind that uid/gid preservation requires the module to be running as root (see uid) or for fake super to be configured.

A chroot-enabled module should not have this parameter set to false unless you're using a name converter program, or you've taken steps to make sure the module has the resources it needs to translate names and that it's not possible for a user to change those resources. (Translation: don't just yolo it.)

This parameter tells rsync to modify all incoming symlinks the same way as the (non-daemon-affecting) --munge-links command-line option: it prefixes each stored symlink's value with the string /rsyncd-munged/. Because that directory doesn't normally exist, a munged symlink can't be followed — so an uploaded symlink can't be used to read or write a file through it. On the way back out the prefix is stripped, so clients see the original symlink values. ✨

When this parameter is enabled, rsync refuses to run a module if /rsyncd-munged/ already exists in it as a directory or a symlink to a directory.

The default depends on chroot. For a plain chrooted module — use chroot on and path with no /./ split, so the daemon chroots straight into the module and serves it as the chroot root — munging is disabled by default, since the chroot itself already stops a symlink from escaping. In every other case it's enabled: any non-chroot module, or a chroot module whose path uses a /./ split to serve an inner subdir below the chroot root.

When this parameter is disabled on a writable module whose access is limited to a subdir (i.e. use chroot is off, or the inside-chroot path is not /), an incoming symlink is not prefixed but is still sanitized as it is stored: a leading slash is dropped and any leading .. components that would take the value above the module are removed, so a stored symlink's value can't name a path outside the module (it can still point to another name inside the module). 😌

Munging changes only the symlink VALUES that are stored. Whether the daemon follows a pre-existing in-module symlink when it resolves a transfer path is a totally separate matter, governed by the secure path resolver that's on by default — see the insecure links parameter for the exact rule and how to opt out of it.

If you're using the munge symlinks parameter in a chroot area that has an inside-chroot path of /, add /rsyncd-munged/ to the exclude setting for the module so a user can't try to create it. 🎀

Want to add or remove the /rsyncd-munged/ prefix on symlinks yourself (like on links you create locally inside a module)? There's a python script in the support directory of the source code named munge-symlinks. Slay.

This parameter (defaulting to false) controls whether the daemon resolves its symlink-bearing paths with the normal symlink-race defences or with the legacy follow-any-symlink behaviour.

When false (the default), the daemon refuses to follow a symlink it did not create — the directory enumeration and file-content opens of the served module, plus the operator-supplied paths (--backup-dir, --temp-dir, --partial-dir, the alt-dest basis dirs, and so on) are all resolved with a walk that follows a symlink component only when it's owned by uid 0 or by the uid the module runs as, refusing one planted by any other user. A client can never relax this: a client that sends --insecure-links to the daemon has its request refused. 🚫

Setting this parameter to true restores the pre-hardening behaviour for this module only: the daemon follows any symlink in those paths.

WARNING: enabling insecure links re-opens the symlink-escape / TOCTOU vulnerabilities that the default closes (including CVE-2026-53797 and CVE-2026-53801): an attacker who can create a symlink inside the module — or who shares the host with the served tree — can redirect the daemon's reads and writes to files outside the module. Enable it only on a single-tenant, fully-isolated, or otherwise trusted host where no untrusted local user can plant a symlink in the served tree. It's the daemon analogue of the client's --insecure-links and exists for the same narrow "I accept the risk in my isolated environment" case. 😭

charset</p>

Bestie, this one tells the daemon what character set your module's filenames are living in 💅. So if a client rocks up with --iconv, the daemon is gonna use the value of the charset parameter no matter what character set the client actually passed — yeah, it's got main character energy and ignores the client's opinion on purpose.

Why tho? Two reasons, both iconic ✨: it lets the daemon support charset conversion inside a chroot module without needing extra files stuffed into the chroot area, and it also keeps name-translation consistent, which is the vibe we want.

Real talk though: if charset is not set, the --iconv option gets refused — straight up, exactly like if "iconv" had been listed in "refuse options".

And if you want to force everyone to always use --iconv for a particular module, just add "no-iconv" to the "refuse options" parameter 🫶. Just keep in mind that this restricts access to your module to very new rsync clients — like, only the cool new kids get in 💀.

max connections</p>

This is your bouncer at the door 🚪. It sets the maximum number of simultaneous connections you'll allow.

Any client that connects once the max has been reached gets a message telling them to try later — not a vibe, but fair. Default is 0, which means no limit (party's open, come on in 🎉). A negative value disables the module entirely.

Also peep the "lock file" parameter — it's the friend that makes this whole thing work. 👀

log file</p>

When log file is set to a non-empty string, the rsync daemon logs messages to that indicated file instead of using syslog 📝. This is especially useful on systems (like AIX, shoutout) where syslog() just does not work for chrooted programs. Rude.

The file is opened before chroot() is called, which is the whole slay — it means the file can live outside the transfer area 🦋.

If this value is set on a per-module basis instead of globally, the global log will still contain any authorization failures or config-file error messages. So nothing sketchy slips through unnoticed 💅.

If the daemon fails to open the specified file, it falls back to using syslog and outputs an error about the failure. Ngl, that's way chiller than it used to be — failure to open the specified log file used to be a fatal error 😭.

This setting can be overridden using the --log-file=FILE or --dparam=logfile=FILE command-line options 🎀:

  • The former (--log-file=FILE) overrides all the log-file parameters of the daemon and all module settings — it's the final boss.
  • The latter (--dparam=logfile=FILE) sets the daemon's log file and the default for all the modules, which still lets modules override the default setting. Respectful of boundaries, we love to see it.

syslog facility</p>

Lets you pick the syslog facility name to use when logging messages from the rsync daemon 🌸. You can use any standard syslog facility name that's defined on your system.

Common names are: auth, authpriv, cron, daemon, ftp, kern, lpr, mail, news, security, syslog, user, uucp, local0, local1, local2, local3, local4, local5, local6 and local7. The default is daemon.

Heads up: this setting has no effect if the "log file" setting is a non-empty string — whether that's set in the per-module settings or inherited from the global settings. Log file ate and left no crumbs, so syslog's just standing there 🙈.

syslog tag</p>

Same energy, but for the syslog tag 🏷️. This parameter lets you specify the syslog tag used when logging messages from the rsync daemon. The default is "rsyncd".

Same caveat, bestie: this setting has no effect if the "log file" setting is a non-empty string (again — per-module settings or inherited from global 💀).

For example, if you wanted each authenticated user's name to be included in the syslog tag, you could do something like this:

syslog tag = rsyncd.%RSYNC_USER_NAME%

Cute, right? Personalised logs, we're obsessed 💖.

max verbosity</p>

This one controls the maximum amount of verbose information you'll allow the daemon to generate (since all that info goes into the log file) 📊. Default is 1, which allows the client to request one level of verbosity.

It also affects the user's ability to request higher levels of --info and --debug logging 🎚️. So if the max value is 2, then no info and/or debug value higher than what would be set by -vv will be honored by the daemon in its logging. Like, the daemon simply will not entertain it.

Wanna know how high of a verbosity level you need to accept for a particular info/debug level? Refer to rsync --info=help and rsync --debug=help 🌈. For instance, it takes max-verbosity 4 to be able to output debug TIME2 and FLIST3. Numbers don't lie, bestie.

lock file</p>

Specifies the file used to support the "max connections" parameter 🔒. The rsync daemon uses record locking on this file to make sure the max connections limit is not exceeded for the modules sharing the lock file — teamwork makes the dream work 🫶.

The default is:

/var/run/rsyncd.lock

read only</p>

Determines whether clients can upload files or not ⬆️.

  • If read only is true → any attempted uploads fail. Periodt.
  • If read only is false → uploads are possible if file permissions on the daemon side allow them.

The default is for all modules to be read only 🔐.

Note that "auth users" can override this setting on a per-user basis — so some users might get special treatment, iykyk.

write only</p>

The mirror image, bestie 🪞. Determines whether clients can download files or not ⬇️.

  • If write only is true → any attempted downloads fail.
  • If write only is false → downloads are possible if file permissions on the daemon side allow them.

The default is for this parameter to be disabled.

Helpful hint: you probably want to specify refuse options = delete for a write-only module 👑. Trust, it's a good idea.

open noatime</p>

Three moods, one parameter 💅:

  • True → tells the rsync daemon to open files with the O_NOATIME flag (on systems that support it) to avoid changing the access time of the files being transferred 🕐. If your OS doesn't support the O_NOATIME flag, rsync will silently ignore this option — no drama, no error. Also note that some filesystems are mounted to avoid updating the atime on read access even without the O_NOATIME flag being set, so it's already giving minimalist.
  • False → ensures that files on the server are not opened with O_NOATIME.
  • Unset (the default) → the user controls the setting via --open-noatime 🎛️.

list</p>

Determines whether this module shows up when the client asks for a listing of available modules 👀.

Bonus feature: if this is false, the daemon will pretend the module does not exist when a client that's denied by "hosts allow" or "hosts deny" tries to access it — stealth mode activated 🥷.

But realize: if "reverse lookup" is disabled globally but enabled for the module, the resulting reverse lookup to a potentially client-controlled DNS server may still reveal to the client that it hit an existing module 😬. So, y'know. Not a perfect disguise.

The default is for modules to be listable ✨.

uid</p>

Specifies the user name or user ID that file transfers to and from that module should take place as, when the daemon was run as root 🧑‍💻. In combination with the "gid" parameter, this determines what file permissions are available.

Defaults, bestie:

  • When run by a super-user → switch to the system's "nobody" user.
  • For a non-super-user → don't try to change the user at all.

See also the "gid" parameter 🫶.

The RSYNC_USER_NAME environment variable may be used to request that rsync run as the authorizing user 🌟. For example, if you want rsync to run as the same user that was received for the rsync authentication, this setup is useful:

uid = %RSYNC_USER_NAME%
gid = *

Honestly? Chef's kiss 🔥.

gid</p>

Specifies one or more group names/IDs that will be used when accessing the module 👥. The first one is the default group, and any extra ones get set as supplemental groups. You may also specify a "*" as the first gid in the list, which gets replaced by all the normal groups for the transfer's user (see "uid").

Defaults:

  • When run by a super-user → switch to your OS's "nobody" (or perhaps "nogroup") group, with no other supplementary groups.
  • For a non-super-user → don't change any group attributes (and indeed, your OS may not allow a non-super-user to try to change their group settings — the OS said no 💀).

Splitting tea: the specified list is normally split into tokens based on spaces and commas. However, if the list starts with a comma, then the list is only split on commas, which allows a group name to contain a space 🎀. In either case, any leading and/or trailing whitespace is removed from the tokens, and empty tokens are ignored. Clean and tidy, we stan.

daemon uid</p>

Specifies a uid under which the daemon will run 🏃‍♀️. The daemon usually runs as user root, and when this is left unset, the user is left unchanged. See also the "uid" parameter 👑

daemon gid</p>

Specifies a gid under which the daemon will run 🏃‍♀️. The daemon usually runs as group root, and when this is left unset, the group is left unchanged. See also the "gid" parameter 💜

fake super</p>

Setting "fake super = yes" for a module causes the daemon side to behave as if the --fake-super command-line option had been specified 🌈. This allows the full attributes of a file to be stored without having to have the daemon actually running as root.

Fake it till you make it, literally 😌✨.

filter</p>

Okay so the daemon has its own filter chain that determines what files it will let the client access 🚧. This chain is not sent to the client, and it's completely independent of any filters the client may have specified. The daemon's boundaries are the daemon's boundaries, periodt.

Here's how daemon-excluded files get treated:

  • If the client tries to pull them → treated as non-existent 🙈.
  • If the client tries to push them → skipped with an error message (triggering exit code 23) 😭.
  • They are never deleted from the module 🔒.

You can use daemon filters to prevent clients from downloading or tampering with private administrative files — such as files you may add to support uid/gid name translations 👑.

The daemon filter chain is built from the "filter", "include from", "include", "exclude from", and "exclude" parameters, in that order of priority 📋.

Anchored patterns are anchored at the root of the module 🎯. So to prevent access to an entire subtree — for example, /secret — you must exclude everything in the subtree. The easiest way to do this is with a triple-star pattern like:

/secret/***

Go forth and gatekeep, bestie 💅🔥.

Okay so the filter chain matches the logical name of each item relative to the module root, not the physical file it resolves to. And bestie, say it with me: it's a visibility and tamper filter, not a security boundary against symlinks. Like, a symlink sitting inside the module whose own name is not excluded can still be followed to an excluded target — the name the filter sees (think "link") is not the excluded name (think "secret"), iykyk? 💅 What actually protects a writable module from symlink trickery is the "munge symlinks" parameter (enabled by default for a writable, non-chrooted module), not this filter. So do not rely on exclude/filter to confine a peer who can introduce or traverse a symlink. 🫶

The "filter" parameter takes a space-separated list of daemon filter rules, though she's smart enough to know not to split a token at an internal space in a rule (so "- /foo - /bar" is parsed as two rules, slay). You may specify one or more merge-file rules using the normal syntax. Only one "filter" parameter can apply to a given module in the config file, so put all the rules you want in a single parameter, okay? Note that per-directory merge-file rules do not provide as much protection as global rules, but they can be used to make --delete work better during a client download operation if the per-dir merge files are included in the transfer and the client requests that they be used. 🌸

exclude

This parameter takes a space-separated list of daemon exclude patterns. As with the client --exclude option, patterns can be qualified with "- " or "+ " to explicitly indicate exclude/include. Only one "exclude" parameter can apply to a given module. See the "filter" parameter for a description of how excluded files affect the daemon. ✨

include

Use an "include" to override the effects of the "exclude" parameter. Only one "include" parameter can apply to a given module. See the "filter" parameter for a description of how excluded files affect the daemon. 💖

exclude from

This parameter specifies the name of a file on the daemon that contains daemon exclude patterns, one per line. Only one "exclude from" parameter can apply to a given module; if you have multiple exclude-from files, you can specify them as a merge file in the "filter" parameter. See the "filter" parameter for a description of how excluded files affect the daemon. 🎀

include from

Analogue of "exclude from" for a file of daemon include patterns. Only one "include from" parameter can apply to a given module. See the "filter" parameter for a description of how excluded files affect the daemon. 🦋

incoming chmod

This parameter lets you specify a set of comma-separated chmod strings that will affect the permissions of all incoming files (the files being received by the daemon). These changes happen after all other permission calculations, and this will even override destination-default and/or existing permissions when the client does not specify --perms. See the description of the --chmod rsync option and the chmod(1) manpage for info on the format of this string. 👑

outgoing chmod

This parameter lets you specify a set of comma-separated chmod strings that will affect the permissions of all outgoing files (the files being sent out from the daemon). These changes happen first, making the sent permissions appear to be different than those stored in the filesystem itself. Like — for instance — you could disable group write permissions on the server while having it appear to be on to the clients. Iconic. See the description of the --chmod rsync option and the chmod(1) manpage for info on the format of this string. 🧚

auth users

This parameter specifies a comma and/or space-separated list of authorization rules. In its simplest form, you list the usernames that will be allowed to connect to this module. The usernames do not need to exist on the local system. The rules may contain shell wildcard characters that will be matched against the username provided by the client for authentication. If "auth users" is set then the client will be challenged to supply a username and password to connect to the module. A challenge response authentication protocol is used for this exchange. The plain text usernames and passwords are stored in the file specified by the "secrets file" parameter. The default is for all users to be able to connect without a password (this is called "anonymous rsync"). 🙈

In addition to username matching, you can specify groupname matching via a '@' prefix. When using groupname matching, the authenticating username must be a real user on the system, or it will be assumed to be a member of no groups. For example, specifying "@rsync" will match the authenticating user if the named user is a member of the rsync group. 💜

Finally, options may be specified after a colon (:). The options allow you to "deny" a user or a group, set the access to "ro" (read-only), or set the access to "rw" (read/write). Setting an auth-rule-specific ro/rw setting overrides the module's "read only" setting. ⭐

Be sure to put the rules in the order you want them to be matched, because the checking stops at the first matching user or group, and that is the only auth that is checked. For example:

auth users = joe:deny @guest:deny admin:rw @rsync:ro susan joe sam

In the above rule, user joe will be denied access no matter what. Any user that is in the group "guest" is also denied access. The user "admin" gets access in read/write mode, but only if the admin user is not in group "guest" (because the admin user-matching rule would never be reached if the user is in group "guest"). Any other user who is in group "rsync" will get read-only access. Finally, users susan, joe, and sam get the ro/rw setting of the module, but only if the user didn't match an earlier group-matching rule. 😭

If you need to specify a user or group name with a space in it, start your list with a comma to indicate that the list should only be split on commas (though leading and trailing whitespace will also be removed, and empty entries are just ignored). For example:

auth users = , joe:deny, @Some Group:deny, admin:rw, @RO Group:ro

See the description of the secrets file for how you can have per-user passwords as well as per-group passwords. It also explains how a user can authenticate using their user password or (when applicable) a group password, depending on what rule is being authenticated. 💖

See also the section entitled "USING RSYNC-DAEMON FEATURES VIA A REMOTE SHELL CONNECTION" in rsync(1) for info on how to handle an rsyncd.conf-level username that differs from the remote-shell-level username when using a remote shell to connect to an rsync daemon. 🌈

secrets file

This parameter specifies the name of a file that contains the username:password and/or @groupname:password pairs used for authenticating this module. This file is only consulted if the "auth users" parameter is specified. The file is line-based and contains one name:password pair per line. Any line that has a hash (#) as the very first character on the line is considered a comment and is skipped. The passwords can contain any characters but be warned that many operating systems limit the length of passwords that can be typed at the client end, so you may find that passwords longer than 8 characters don't work. 😬

The use of group-specific lines is only relevant when the module is being authorized using a matching "@groupname" rule. When that happens, the user can be authorized via either their "username:password" line or the "@groupname:password" line for the group that triggered the authentication. 🫶

It is up to you what kind of password entries you want to include — either users, groups, or both. The use of group rules in "auth users" does not require that you specify a group password if you do not want to use shared passwords. ✨

There is no default for the "secrets file" parameter, you must choose a name (such as /etc/rsyncd.secrets). Unless "strict modes" is disabled, the file must not be readable or writable by "other", and (when the daemon runs as root) must be owned by root; see "strict modes" for the exact check. If the file is not found or is rejected, no logins for an "auth users" module will be possible. 💀

auth digest

This parameter sets the minimum message digest that the daemon will accept for the challenge-response authentication of an "auth users" module. The available digests, strongest first, are sha512, sha256, sha1, md5, md4. The daemon selects its own most-preferred digest that also appears in the connecting client's advertised list (and falls back to md5, or md4 below protocol 30, for a client that advertises none), so a client that advertises only weak digests can still force a weak negotiated digest. If it is weaker than the configured name, the connection is refused before the challenge is sent. The value is a single digest name, for example:

auth digest = sha256

which requires sha256 or sha512 and refuses an md5 or md4 exchange. 👑

This guards against an auth-digest downgrade. A peer that sends no digest list (any rsync before 3.2.0, including the openrsync that ships with macOS) makes the daemon fall back to md5 (or md4 below protocol 30), and an on-path attacker can rewrite the unauthenticated negotiation to force the same weak choice. A captured challenge-response is far cheaper to brute-force offline against a weak digest, so a site whose clients are all modern can require a strong one. See the rsync SECURITY.md document for the full threat model. 🔥

<span pink>No default here, bestie</span> — the floor is literally off, and the daemon will happily accept whatever digest gets negotiated, which keeps things chill and compatible with the older clients. But the second you turn a floor on, you're refusing every client that can't offer at least that digest. And who does that hit? Notably any rsync older than 3.2.0 (that's when the SHA digests got added, periodt) and the macOS-bundled openrsync, which only knows how to authenticate with md4/md5. Also, the SHA digests need an rsync built with openssl at both ends — and if you name a digest this build straight-up does not provide, then every single login to the module gets refused (fail-closed, she's strict 💅).

Like the other auth parameters, you can set this per module, or — if you want a default that applies to every module — drop it in the global part of the config file. 🌍

strict modes

This one decides whether the permissions on the secrets file get checked at all. If "strict modes" is true (and yeah, that's the default), the secrets file gets rejected if it's readable or writable by "other" — like, if any of the other-read or other-write permission bits is set — and, when the daemon is running as root, if the file isn't owned by root. 👑

Group permissions and the other-execute bit are not consulted, so modes like 600 or 640 are accepted, while 644 (or any other other-readable/-writable mode) gets rejected. 🙅‍♀️ If "strict modes" is false, no permission check happens at all. This parameter was added to accommodate rsync running on the Windows operating system.

hosts allow

Okay so this parameter lets you specify a list of comma- and/or whitespace-separated patterns that get matched against a connecting client's hostname and IP address. If none of the patterns match, the connection is rejected. Simple as that. ✨

Each pattern can be in one of six forms:

  • a dotted decimal IPv4 address of the form a.b.c.d, or an IPv6 address of the form a:b:c::d:e:f. In this case the incoming machine's IP address must match exactly.
  • an address/mask in the form ipaddr/n where ipaddr is the IP address and n is the number of one bits in the netmask. All IP addresses which match the masked IP address will be allowed in.
  • an address/mask in the form ipaddr/maskaddr where ipaddr is the IP address and maskaddr is the netmask in dotted decimal notation for IPv4, or similar for IPv6, e.g. ffff:ffff:ffff:ffff:: instead of /64. All IP addresses which match the masked IP address will be allowed in.
  • a hostname pattern using wildcards. If the hostname of the connecting IP (as determined by a reverse lookup) matches the wildcarded name (using the same rules as normal Unix filename matching), the client is allowed in. This only works if "reverse lookup" is enabled (the default).
  • a hostname. A plain hostname is matched against the reverse DNS of the connecting IP (if "reverse lookup" is enabled), and/or the IP of the given hostname is matched against the connecting IP (if "forward lookup" is enabled, as it is by default). Any match will be allowed in.
  • an '@' followed by a netgroup name, which will match if the reverse DNS of the connecting IP is in the specified netgroup.

If rsync was built with IDN support (look for "IDN" in rsync --version), a hostname pattern may contain non-ASCII characters: this file is read as UTF-8, and each non-ASCII label is converted to its IDNA A-label ("Punycode") form before matching, since the name the daemon has for a client always reaches it from DNS as ASCII. A pattern written with combining characters is normalized on the way, so it matches the same as its precomposed spelling. Only those labels change, so an address, a mask, an already-punycoded name, and the wildcard characters are all matched exactly as written. A pattern that cannot be converted that way is matched as it stands, and thus matches nothing — and yes, that includes a pattern whose conversion would have introduced a character it was not written with, such as the U+FF0A FULLWIDTH ASTERISK that the IDNA mapping turns into a "*". 🔤💀

Heads up bestie: IPv6 link-local addresses can have a scope in the address specification: fe80::1%link1 fe80::%link1/64 fe80::%link1/ffff:ffff:ffff:ffff::

You can also combine "hosts allow" with "hosts deny" as a way to add exceptions to your deny list. When both parameters are specified, the "hosts allow" parameter is checked first and a match results in the client being able to connect. A non-allowed host is then matched against the "hosts deny" list to see if it should be rejected. A host that does not match either list is allowed to connect. 🫶

The default is no "hosts allow" parameter, which means all hosts can connect.

hosts deny

This parameter allows you to specify a list of comma- and/or whitespace-separated patterns that are matched against a connecting client's hostname and IP address. If the pattern matches, then the connection is rejected. See the "hosts allow" parameter for more information. 👀

The default is no "hosts deny" parameter, which means all hosts can connect.

reverse lookup

Controls whether the daemon performs a reverse lookup on the client's IP address to determine its hostname, which is used for "hosts allow" & "hosts deny" checks and the "%h" log escape. This is enabled by default, but you may wish to disable it to save time if you know the lookup will not return a useful result — in which case the daemon will use the name "UNDETERMINED" instead. 🕵️‍♀️

Real talk: if this parameter is enabled globally (even by default), rsync performs the lookup as soon as a client connects, so disabling it for a module will not avoid the lookup. Thus, you probably want to disable it globally and then enable it for the modules that actually need the info.

forward lookup

Controls whether the daemon performs a forward lookup on any hostname specified in an hosts allow/deny setting. By default this is enabled, allowing the use of an explicit hostname that would not be returned by reverse DNS of the connecting IP. 🔍

ignore errors

This parameter tells rsyncd to ignore I/O errors on the daemon when deciding whether to run the delete phase of the transfer. Normally rsync skips the --delete step if any I/O errors have occurred, in order to prevent disastrous deletion due to a temporary resource shortage or other I/O error. 😭 In some cases this test is counter productive, so you can use this parameter to turn off this behavior.

ignore nonreadable

This tells the rsync daemon to completely ignore files that are not readable by the user. This is useful for public archives that may have some non-readable files among the directories, and the sysadmin doesn't want those files to be seen at all. 🙈

transfer logging

This parameter enables per-file logging of downloads and uploads in a format somewhat similar to that used by ftp daemons. The daemon always logs the transfer at the end, so if a transfer is aborted, no mention will be made in the log file. 💀

If you want to customize the log lines, see the "log format" parameter.

log format

This parameter allows you to specify the format used for logging file transfers when transfer logging is enabled. The format is a text string containing embedded single-character escape sequences prefixed with a percent (%) character. An optional numeric field width may also be specified between the percent and the escape letter (e.g. "%-50n %8l %07p"). In addition, one or more apostrophes may be specified prior to a numerical escape to indicate that the numerical value should be made more human-readable. The 3 supported levels are the same as for the --human-readable command-line option, though the default is for human-readability to be off. Each added apostrophe increases the level (e.g. "%''l %'b %f"). 🎀

The default log format is "%o %h [%a] %m (%u) %f %l", and a "%t [%p] " is always prefixed when using the "log file" parameter. (A perl script that will summarize this default log format is included in the rsync source code distribution in the "support" subdirectory: rsyncstats.)

The single-character escapes that are understood are as follows:

  • %a the remote IP address (only available for a daemon)
  • %b the number of bytes actually transferred
  • %B the permission bits of the file (e.g. rwxrwxrwt)
  • %c the total size of the block checksums received for the basis file (only when sending)
  • %C the full-file checksum if it is known for the file. For older rsync protocols/versions, the checksum was salted, and is thus not a useful value (and is not displayed when that is the case). For the checksum to output for a file, either the --checksum option must be in-effect or the file must have been transferred without a salted checksum being used. See the --checksum-choice option for a way to choose the algorithm.
  • %f the filename (long form on sender; no trailing "/")
  • %G the gid of the file (decimal) or "DEFAULT"
  • %h the remote host name (only available for a daemon)
  • %i an itemized list of what is being updated
  • %l the length of the file in bytes
  • %L the string " -> SYMLINK", " => HARDLINK", or "" (where SYMLINK or HARDLINK is a filename)
  • %m the module name
  • %M the last-modified time of the file
  • %n the filename (short form; trailing "/" on dir)
  • %o the operation, which is "send", "recv", or "del." (the latter includes the trailing period)
  • %p the process ID of this rsync session
  • %P the module path
  • %t the current date time
  • %u the authenticated username or an empty string
  • %U — the uid of the file (decimal) 💅

For a list of what the characters mean that are output by %i, see the --itemize-changes option in the rsync manpage. ✨

Note that some of the logged output changes when talking with older rsync versions. For instance, deleted files were only output as verbose messages prior to rsync 2.6.4. 👀

timeout ⏰

This parameter lets you override the client’s choice for I/O timeout for this module. Using it, you can make sure rsync won’t wait on a dead client forever. The timeout is specified in seconds. A value of zero means no timeout and is the default. A good choice for anonymous rsync daemons may be 600 (giving a 10 minute timeout). 🫶

refuse options 🚫

This parameter lets you specify a space-separated list of rsync command-line options that your rsync daemon will refuse. You may specify the full option name, its one-letter abbreviation, or a wild-card string that matches multiple options. Beginning in 3.2.0, you can also negate a match term by starting it with a "!".

When an option is refused, the daemon prints an error message and exits. 💀

For example, this would refuse --checksum (-c) and all the various delete options:

refuse options = c delete

The reason the above refuses all delete options is that the options imply --delete, and implied options are refused just like explicit options. 🧠

The use of a negated match allows you to fine-tune your refusals after a wild-card, such as this:

refuse options = delete-* !delete-during

Negated matching can also turn your list of refused options into a list of accepted options. To do this, begin the list with a "*" (to refuse all options) and then specify one or more negated matches to accept. For example:

refuse options = * !a !v !compress*

Don’t worry that the "*" will refuse certain vital options such as --dry-run, --server, --no-iconv, --seclude-args, etc. These important options are not matched by wild-card, so they must be overridden by their exact name. For instance, if you’re forcing iconv transfers you could use something like this:

refuse options = * no-iconv !a !v

As an additional aid (beginning in 3.2.0), refusing (or "!refusing") the "a" or "archive" option also affects all the options that the --archive option implies (-rdlptgoD), but only if the option is matched explicitly (not using a wildcard). If you want to do something tricky, you can use "archive*" to avoid this side-effect, but keep in mind that no normal rsync client ever sends the actual archive option to the server. 🎀

As an additional safety feature, the refusal of "delete" also refuses remove-source-files when the daemon is the sender; if you want the latter without the former, instead refuse "delete-*" as that refuses all the delete modes without affecting --remove-source-files. (Keep in mind that the client’s --delete option typically results in --delete-during.) 🔥

When un-refusing delete options, you should either specify "!delete*" (to accept all delete options) or specify a limited set that includes "delete", such as:

refuse options = * !a !delete !delete-during

… whereas this accepts any delete option except --delete-after:

refuse options = * !a !delete* delete-after

A note on refusing "compress": it may be better to set the "dont compress" daemon parameter to "*" and ensure that RSYNC_COMPRESS_LIST=zlib is set in the environment of the daemon in order to disable compression silently instead of returning an error that forces the client to remove the -z option. 🧚

If you are un-refusing the compress option, you may want to match "!compress*" if you also want to allow the --compress-level option. 🌸

Note that the "copy-devices" & "write-devices" options are refused by default, but they can be explicitly accepted with "!copy-devices" and/or "!write-devices". The options "log-file" and "log-file-format" are forcibly refused and cannot be accepted. 🙈

Here are all the options that are not matched by wild-cards:

  • --server — Required for rsync to even work. 👑
  • --rsh, -e — Required to convey compatibility flags to the server.
  • --out-format — This is required to convey output behavior to a remote receiver. While rsync passes the older alias --log-format for compatibility reasons, this option should not be confused with --log-file-format.
  • --sender — Use "write only" parameter instead of refusing this.
  • --dry-run, -n — Who would want to disable this? 😭
  • --seclude-args, -s — Is the oldest arg-protection method.
  • --from0, -0 — Makes it easier to accept/refuse --files-from without affecting this helpful modifier.
  • --iconv — This is auto-disabled based on "charset" parameter.
  • --no-iconv — Most transfers use this option.
  • --checksum-seed — Is a fairly rare, safe option.
  • --write-devices — Is non-wild but also auto-disabled.

dont compress 🍓

NOTE: This parameter currently has no effect except in one instance: if it is set to "*" then it minimizes or disables compression for all files (for those that don’t want to refuse the --compress option completely). 💖

This parameter allows you to select filenames based on wildcard patterns that should not be compressed when pulling files from the daemon (no analogous parameter exists to govern the pushing of files to a daemon). Compression can be expensive in terms of CPU usage, so it is usually good to not try to compress files that won’t compress well, such as already compressed files. 🦋

The "dont compress" parameter takes a space-separated list of case-insensitive wildcard patterns. Any source filename matching one of the patterns will be compressed as little as possible during the transfer. If the compression algorithm has an "off" level, then no compression occurs for those files. If an algorithm has the ability to change the level in mid-stream, it will be minimized to reduce the CPU usage as much as possible. ✨

See the --skip-compress parameter in the rsync(1) manpage for the list of file suffixes that are skipped by default if this parameter is not set. 📚

early exec, pre-xfer exec, post-xfer exec 🎬

You may specify a command to be run in the early stages of the connection, or right before and/or after the transfer. If the early exec or pre-xfer exec command returns an error code, the transfer is aborted before it begins. Any output from the pre-xfer exec command on stdout (up to several KB) will be displayed to the user when aborting, but is not displayed if the script returns success. The other programs cannot send any text to the user. All output except for the pre-xfer exec stdout goes to the corresponding daemon’s stdout/stderr, which is typically discarded. See the --no-detach option for a way to see the daemon’s output, which can assist with debugging. 🔍

Note that the early exec command runs before any part of the transfer request is known except for the module name. This helper script can be used to setup a disk mount or decrypt some data into a module dir, but you may need to use lock file and max connections to avoid concurrency issues. If the client rsync specified the --early-input=FILE option, it can send up to about 5K of data to the stdin of the early script. The stdin will otherwise be empty. 🫶

Note that the post-xfer exec command is still run even if one of the other scripts returns an error code. The pre-xfer exec command will not be run, however, if the early exec command fails. 💅

The following environment variables will be set, though some are specific to the pre-xfer or the post-xfer environment:

  • RSYNC_MODULE_NAME — The name of the module being accessed. 🌈
  • RSYNC_MODULE_PATH — The path configured for the module.
  • RSYNC_HOST_ADDR — The accessing host’s IP address.
  • RSYNC_HOST_NAME — The accessing host’s name.
  • RSYNC_USER_NAME — The accessing user’s name (empty if no user).
  • RSYNC_PID — A unique number for this transfer. ⭐
  • RSYNC_REQUEST — (pre-xfer only) The module/path info specified by the user. Note that the user can specify multiple source files, so the request can be something like "mod/path1 mod/path2", etc. 💜
  • RSYNC_ARG# — (pre-xfer only) The pre-request arguments are set in these numbered values. RSYNC_ARG0 is always "rsyncd", followed by the options that were used in RSYNC_ARG1, and so on. There will be a value of "." indicating that the options are done and the path args are beginning -- these contain similar information to RSYNC_REQUEST, but with values separated and the module name stripped off. 🧠
  • RSYNC_EXIT_STATUS — (post-xfer only) the server side’s exit value. This will be 0 for a successful run, a positive value for an error that the server generated, or a -1 if rsync failed to exit properly. Note that an error that occurs on the client side does not currently get sent to the server side, so this is not the final exit status for the whole transfer. 😭
  • RSYNC_RAW_STATUS — (post-xfer only) the raw exit value from waitpid(). 💀

...and the way the escaping actually works is lowkey clever but also lowkey fragile 💅 — when a %VAR% reference gets expanded into one of these commands, it's escaped for the quoting context it appears in, and that's correct for the single shell that runs the command. It is not correct for a command that starts a SECOND shell, bestie, like:

sh -c '... %RSYNC_USER_NAME% ...'

The outer shell eats the escaping, and the inner one sees the value bare. A value of touc? would then get glob-expanded against /usr/bin/, so the substitution is not data any more — it chooses the command. 💀 That's giving "your input is now the program," and we are NOT here for it.

Because rsync cannot escape for an unknown number of shell passes, a value carrying any character that a shell could act on is refused outright, and the transfer is aborted with:

refusing to run shell hook: %VAR% holds a shell metacharacter

The refused characters are whitespace, the quoting and expansion characters ' " ` $ \, the separators ; & |, redirections < >, parentheses, the pattern characters * ? [ ], #, !, ~ { }, and any control character. Some of those are harmless on their own and are refused for what a second shell would do with them — ! negates in command position, so a hook written as an access check can be turned from a denial into an approval 😭, and ~ is tilde-expanded. iconic of them to think ahead tbh.

This applies to EVERY %RSYNC_*% value, including ones you supplied yourself. In particular a module whose path contains any of them cannot be interpolated into one of these commands: path = /srv/My Backups with a command mentioning %RSYNC_MODULE_PATH% will refuse every transfer of that module, not just a hostile one 🫠. The restriction is deliberate — rsync cannot tell your space from an attacker's once both are inside the same string, and path itself may be built from a peer value such as path = /home/%RSYNC_USER_NAME%.

If you need such a value in a hook, pass it through the environment instead of interpolating it: the same names are exported to the command, so "$RSYNC_MODULE_PATH" inside your script is unrestricted by this check. Quote it there as shown — rsync no longer has any say in how your script splits the value. 🫶

Even though the commands can be associated with a particular module, they are run using the permissions of the user that started the daemon (not the module's uid/gid setting) without any chroot restrictions. yes this is the part where you go "huh" and re-read it.

These settings honor 2 environment variables: use RSYNC_SHELL to set a shell to use when running the command (which otherwise uses your system() call's default shell), and use RSYNC_NO_XFER_EXEC to disable both options completely.

🌡️ temp dir

Specifies a directory that rsync should use for temporary files created during the transfer of updated files. If that directory is on a different partition, after transfer file is being copied instead of unlinked 🎀.

This parameter equals with --temp-dir option, so please consult rsync manpage for further information.

✨ CONFIG DIRECTIVES ✨

There are currently two config directives available that allow a config file to incorporate the contents of other files: &include and &merge. Both allow a reference to either a file or a directory. They differ in how segregated the file's contents are considered to be. same energy, different boundaries 💅

The &include directive treats each file as more distinct, with each one inheriting the defaults of the parent file, starting the parameter parsing as globals/defaults, and leaving the defaults unchanged for the parsing of the rest of the parent file. self-contained queens only 👑

The &merge directive, on the other hand, treats the file's contents as if it were simply inserted in place of the directive, and thus it can set parameters in a module started in another file, can affect the defaults for other files, etc. it's giving copy-paste, but make it official.

When an &include or &merge directive refers to a directory, it will read in all the *.conf or *.inc files (respectively) that are contained inside that directory (without any recursive scanning), with the files sorted into alpha order 🌸. So, if you have a directory named "rsyncd.d" with the files "foo.conf", "bar.conf", and "baz.conf" inside it, this directive:

&include /path/rsyncd.d

would be the same as this set of directives:

&include /path/rsyncd.d/bar.conf
&include /path/rsyncd.d/baz.conf
&include /path/rsyncd.d/foo.conf

except that it adjusts as files are added and removed from the directory. alphabetical and self-updating, we love a low-maintenance icon ✨

The advantage of the &include directive is that you can define one or more modules in a separate file without worrying about unintended side-effects between the self-contained module files.

The advantage of the &merge directive is that you can load config snippets that can be included into multiple module definitions, and you can also set global values that will affect connections (such as motd file), or globals that will affect other include files.

For example, this is a useful /etc/rsyncd.conf file:

port = 873
log file = /var/log/rsync.log
pid file = /var/lock/rsync.lock

&merge /etc/rsyncd.d
&include /etc/rsyncd.d

This would merge any /etc/rsyncd.d/*.inc files (for global values that should stay in effect), and then include any /etc/rsyncd.d/*.conf files (defining modules without any global-value cross-talk). chef's kiss config layout 🎀

🛡️ SECURITY 🛡️

An rsync daemon exposes part of your filesystem to network clients, and its master process usually runs as root, so it should be configured defensively. no cap, this is the serious section 💀 The following is a practical checklist; the project's SECURITY.md describes the underlying threat model and the per-platform residuals.

🔒 Confine each module to its path

Leave "use chroot" enabled (the default attempts a chroot) so the per-connection worker is jailed inside the module's "path". Where chroot is unavailable rsync still resolves every transfer path so it cannot escape the module through a symlinked parent, but a chroot is the stronger boundary — keep it on unless you have a specific reason not to. A path = /outer/./inner chroots to /outer while serving /inner as the module root.

👑 Drop privileges

Give each module a low-privilege "uid" and "gid" so a worker never serves files as root; only the master process needs root (to chroot and bind the port). Keep modules "read only" unless they must accept uploads, and prefer "numeric ids" (or a "name converter") so a malicious file list cannot map files to an unexpected local owner. least privilege is the moment, fr fr.

🚪 Restrict who can connect

Limit reachability with "hosts allow" / "hosts deny", bind the daemon to a specific "address", and use a host firewall. A module without "auth users" is reachable by anyone who can reach the port. iykyk — that's a whole open door 😭

🔑 Authenticate with a strong secret and digest

Require login with "auth users" plus a "secrets file" that is readable only by the daemon user (rsync refuses a too-open file unless "strict modes" is disabled), and use a long, high-entropy secret. To refuse the weak MD4/MD5 challenge digests that old clients can negotiate, set a floor with "auth digest" (for example auth digest = sha512); see the AUTHENTICATION STRENGTH section below. strong secret + strong digest = the duo we deserve 💅

🔐 Encrypt the connection

The daemon protocol authenticates but does not encrypt the data stream. Do not expose a cleartext daemon to an untrusted network: front it with a TLS proxy (see the SSL/TLS Daemon Setup section below) or run it over ssh. do NOT skip this one 🫶

Leave "munge symlinks" enabled on any writable module (it is the default) so an uploaded symlink cannot redirect a later operation outside the module. Do not set "insecure links = yes" on a host shared with untrusted local users: it disables the daemon's symlink-escape protections (re-opening the symlink TOCTOU/escape vulnerabilities, including CVE-2026-53797 and CVE-2026-53801) and is appropriate only on a single-tenant, fully isolated host. read that twice, bestie 💀

🙅 Refuse options you do not want

Use "refuse options" to reject client options a module should not allow. a little "no" for your own good ✨

🌈 AUTHENTICATION STRENGTH 🌈

Daemon authentication is a challenge-response system in which the client returns a digest of the shared secret and a server-chosen challenge. Modern rsync (3.2.7 and later, built with openssl) negotiates the strongest digest both sides support — sha512 by default — but for backward compatibility it falls back to md5 (or md4 below protocol 30) for a client that advertises no digest list, and the digest negotiation is itself unauthenticated. the negotiation has no vibe check of its own, that's the whole problem 🫠 A daemon that wants to require a strong digest can set "auth digest". For the best protection, run rsync over ssh or a verified TLS transport (rsync-ssl), which protects the exchange at the transport layer, and use a high-entropy shared secret (which is infeasible to brute-force regardless of the digest). entropy is the serve 💅

Also note that the rsync daemon protocol does not currently provide any encryption of the data that is transferred over the connection. Only authentication is provided. Use ssh as the transport if you want encryption. auth ≠ encryption, write it on your hand ✍️

You can also make use of SSL/TLS encryption if you put rsync behind an SSL proxy. slay, go be safe out there 🦋

✨ SSL/TLS Daemon Setup ✨

okay bestie, so if you want your rsync daemon reachable over SSL/TLS, here's the tea: rsyncd itself doesn't do the encryption — you gotta put a TCP proxy (think haproxy or nginx) out front as the front-end that handles all the encrypting. it's giving "bouncer at the club" energy 💅

  • lock it down — you should limit the access to the backend-rsyncd port so that only the proxy can connect. if the proxy lives on the same host, configuring it to only listen on localhost is a very good idea fr fr.
  • proxy protocol, bestie — you should consider turning on the proxy protocol rsync-daemon parameter if your proxy supports sending that info. the examples below assume that it's turned on. 👑

An example haproxy setup is as follows:

frontend fe_rsync-ssl
   bind :::874 ssl crt /etc/letsencrypt/example.com/combined.pem
   mode tcp
   use_backend be_rsync

backend be_rsync
   mode tcp
   server local-rsync 127.0.0.1:873 check send-proxy

An example nginx proxy setup is as follows:

stream {
   server {
       listen 874 ssl;
       listen [::]:874 ssl;

       ssl_certificate /etc/letsencrypt/example.com/fullchain.pem;
       ssl_certificate_key /etc/letsencrypt/example.com/privkey.pem;

       proxy_pass localhost:873;
       proxy_protocol on; # Requires rsyncd.conf "proxy protocol = true"
       proxy_timeout 1m;
       proxy_connect_timeout 5s;
   }
}

and if you want rsyncd accessible encrypted and unencrypted at the same time? no stress — just make the proxy listen on port 873 as well and let it handle both streams. multi-talented icon behavior 🌈

🎀 DAEMON CONFIG EXAMPLES 🎀

A simple rsyncd.conf file that allows anonymous rsync to an ftp area at /home/ftp would be, like, this:

[ftp]
        path = /home/ftp
        comment = ftp export area

A more sophisticated example would be:

uid = nobody
gid = nobody
use chroot = yes
max connections = 4
syslog facility = local5
pid file = /var/run/rsyncd.pid

[ftp]
        path = /var/ftp/./pub
        comment = whole ftp area (approx 6.1 GB)

[sambaftp]
        path = /var/ftp/./pub/samba
        comment = Samba ftp area (approx 300 MB)

[rsyncftp]
        path = /var/ftp/./pub/rsync
        comment = rsync ftp area (approx 6 MB)

[sambawww]
        path = /public_html/samba
        comment = Samba WWW pages (approx 240 MB)

[cvs]
        path = /data/cvs
        comment = CVS repository (requires authentication)
        auth users = tridge, susan
        secrets file = /etc/rsyncd.secrets

and the /etc/rsyncd.secrets file would look something like this — user:pass, one per line, keep it secret, keep it safe 🔐

tridge:mypass
susan:herpass

🦋 FILES 🦋

  • /etc/rsyncd.conf or rsyncd.conf

like, that's the file. that's the one. that's where all your config lives 💖

🌸 SEE ALSO 🌸

go read those too, they're literally iconic 👑

💀 BUGS 💀

Please report bugs! yes, actually report them, bestie — the rsync bug tracking system is online at https://rsync.samba.org/. don't just suffer in silence, we wanna know 🙈

⭐ VERSION ⭐

This manpage is current for version 3.5.2 of rsync. fresh off the press ✨

🍓 CREDITS 🍓

Rsync is distributed under the GNU General Public License. See the file COPYING for details.

An rsync web site is available at https://rsync.samba.org/ and its github project is https://github.com/RsyncProject/rsync.

🎀 THANKS 🎀

Thanks to Warren Stanley for his original idea and patch for the rsync daemon. Thanks to Karsten Thygesen for his many suggestions and documentation! absolute legends, no cap 🫶

👑 AUTHOR 👑

Rsync was originally written by Andrew Tridgell and Paul Mackerras. Many people from around the world have helped to maintain and improve it. it takes a village, fr fr 💜

Special thanks go to Wayne Davison, who maintained rsync from 2004 to 2024. main character energy for two whole decades, we stan 😭

Mailing lists for support and development are available at https://lists.samba.org/.

and that's the whole page, bestie!! you made it to the end 💖✨

this little beauty is rsyncd.conf from rsync 3.5.2, dated 03 Nov 2026 — go forth and sync responsibly 🦋🎀

📦 where this page comes from