The Caddyfile has a way for you to specify options that apply globally. Some options act as default values; others customize HTTP servers and don't apply to just one particular site; while yet others customize the behavior of the Caddyfile [adapter](/docs/config-adapters).
Enables debug mode, which sets the log level to `DEBUG` for the [default logger](#log). This reveals more details that can be useful when troubleshooting (and is very verbose in production). We ask that you enable this before asking for help on the [community forums](https://caddy.community). For example, at the top of your Caddyfile, if you have no other global options:
**For internal use only**; does not change the HTTP port for clients. This is typically used if within your internal network, you needed to port forward `80` to a different port (e.g. `8080`) before it reaches Caddy, for routing purposes.
**For internal use only**; does not change the HTTPS port for clients. This is typically used if within your internal network, you needed to port forward `443` to a different port (e.g. `8443`) before it reaches Caddy, for routing purposes.
The default bind address(es) to be used for all sites, if the [`bind` directive](/docs/caddyfile/directives/bind) is not used in the site. Default: empty, which binds to all interfaces.
Keep in mind that this will only apply to servers which are generated by the Caddyfile; this means that the HTTP server created by [Automatic HTTPS](/docs/automatic-https) for HTTP-to-HTTPS redirects will not inherit these bind addresses. To work around this, make sure to declare an `http://` site (it can be empty, with no directives) so that it exists when the Caddyfile is adapted, to receive the bind addresses.
Assigns an order to HTTP handler directive(s). As HTTP handlers execute in a sequential chain, it is necessary for the handlers to be executed in the right order. Standard directives have [a pre-defined order](/docs/caddyfile/directives#directive-order), but if using third-party HTTP handler modules, you'll need to define the order explicitly by either using this option or placing the directive in a [`route` block](/docs/caddyfile/directives/route). Ordering can be described absolutely (`first` or `last`), or relatively (`before` or `after`) to another directive.
For example, to use the [`replace-response` plugin](https://github.com/caddyserver/replace-response), you'd want to make sure its directive is ordered after `encode` so that it can perform replacements before the response is encoded (because responses flow up the handler chain, not down):
Configures Caddy's storage mechanism. The default is [`file_system`](/docs/json/storage/file_system/). There are many other available [storage modules](/docs/json/storage/) provided as plugins.
Customizing the storage module is typically needed when syncing Caddy's storage across multiple instances of Caddy to make sure they all use the same certificates and keys. See the [Automatic HTTPS section on storage](/docs/automatic-https#storage) for more details.
How often to scan storage units for old or expired assets and remove them. These scans exert lots of reads (and list operations) on the storage module, so choose a longer interval for large deployments. Accepts [duration values](/docs/conventions#durations).
Storage will always be cleaned when the process first starts. Then, a new cleaning will be started this duration after the previous cleaning started if the previous cleaning finished in less than half the time of this interval (otherwise next start will be skipped).
Customizes the [admin API endpoint](/docs/api). Accepts placeholders. Takes [network addresses](/docs/conventions#network-addresses).
Default: `localhost:2019`, unless the `CADDY_ADMIN` environment variable is set.
If set to `off`, then the admin endpoint will be disabled. When disabled, **config changes will be impossible** without stopping and starting the server, since the [`caddy reload` command](/docs/command-line#caddy-reload) uses the admin API to push the new config to the running server.
Remember to use the `--address` CLI flag with compatible [commands](/docs/command-line) to specify the current admin endpoint, if the address of the running server was changed from the default.
Also supports these sub-options:
- **origins** configures the list of origins that are allowed to connect to the endpoint.
A default is intelligently chosen:
- if the listen address is loopback (e.g. `localhost` or a loopback IP, or a unix socket) then the allowed origins are `localhost`, `::1` and `127.0.0.1`, joined with the listen address port (so `localhost:2019` is a valid origin).
- if the listen address is not loopback, then the allowed origin is the same as the listen address.
If the listen address host is not a wildcard interface (wildcards include: empty string, or `0.0.0.0`, or `[::]`), then `Host` header enforcement is performed. Effectively, this means that by default, the `Host` header is validated to be in `origins`, since the interface is `localhost`. But for an address like `:2020` which has a wildcard interface, `Host` header validation is not performed.
- **enforce_origin** enables enforcement of the `Origin` request header.
This is most useful when the listen address is a wildcard interface (since `Host` is not validated), and the admin API is exposed to the public internet. It enables CORS preflight checks and ensures that the `Origin` header is validated against the `origins` list. Only use this if you're running Caddy on your development machine and need to access the admin API from a web browser.
For example, to expose the admin API on a different port, on all interfaces — ⚠️ this port **should not be exposed publicly**, otherwise anyone can control your server; consider enabling origin enforcement if you need it to be public:
```caddy
{
admin :2020
}
```
To turn off the admin API — ⚠️ this makes **config reloads impossible** without stopping and starting the server:
Controls whether the current JSON config should be persisted to the [configuration directory](/docs/conventions#configuration-directory), to avoid losing config changes performed via the admin API. Currently, only the `off` option is supported. By default, the config is persisted.
The name can be passed to indicate a specific logger for which to customize the behavior. If no name is specified, the behavior of the `default` logger is modified. You can read more about the `default` logger and an explanation of [how logging works in Caddy](/docs/logging).
Multiple loggers with different names can be configured by using the `log` multiple times.
The differs from the [`log` directive](/docs/caddyfile/directives/log), which only configures HTTP request logging (also known as access logs). The `log` global option shares its configuration structure with the directive (except for `include` and `exclude`), and complete documentation can be found on the directive's page.
Defines the grace period for shutting down HTTP servers (i.e. during config changes or when Caddy is stopping).
During the grace period, no new connections are accepted, idle connections are closed, and active connections are impatiently waited upon to finish their requests. If clients do not finish their requests within the grace period, the server will be forcefully terminated to allow the reload to complete and free up resources. Accepts [duration values](/docs/conventions#durations).
By default, the grace period is eternal, which means connections are never forcefully closed.
_before_ the [grace period](#grace_period) during which a server that is going to be stopped continues to operate normally, except the `{http.shutting_down}` placeholder evaluates to `true` and `{http.time_until_shutdown}` gives the time until the grace period begins.
This causes a delay if any server is being shut down as part of a config change, and effectively schedules the change for a later time. It is useful for announcing to health checkers of this server's impending doom and to give time for a load balancer to move it out of the rotation; for example:
```caddy
{
shutdown_delay 30s
}
example.com {
handle /health-check {
@goingDown vars {http.shutting_down} true
respond @goingDown "Bye-bye in {http.time_until_shutdown}" 503
Configures [Automatic HTTPS](/docs/automatic-https), which is the feature that enables Caddy to automate certificate management and HTTP-to-HTTPS redirects for your sites.
There are a few modes to choose from:
-`off`: Disables both certificate automation and HTTP-to-HTTPS redirects.
-`ignore_loaded_certs`: Automate certificates even for names which appear on manually-loaded certificates. Useful if you specified a certificate using the [`tls` directive](/docs/caddyfile/directives/tls) which contains names (or wildcards) that you instead want to be managed automatically.
<asideclass="tip">
This option does not affect Caddy's default protocol, which is always HTTPS, when a site address has a valid domain name. This means that `auto_https off` will not cause your site to be served over HTTP, it will only disable automatic certificate management and redirects.
This means that if you wish to serve your site over HTTP, you should change your [site address](/docs/caddyfile/concepts#addresses) to be prefixed with `http://` or suffixed with `:80` (or the [`http_port` option](#http_port)).
Your email address. Mainly used when creating an ACME account with your CA, and is highly recommended in case there are problems with your certificates.
Keep in mind that Let's Encrypt may send you emails about your certificate nearing expiry, but this may be misleading because Caddy may have chosen to use a different issuer (e.g. ZeroSSL) when renewing. Check your logs and/or the certificate itself (in your browser for example) to see which issuer was used, and that its expiry is still valid; if so, you may safely ignore the email from Let's Encrypt.
If configured, the fallback becomes the TLS ServerName in the ClientHello if the original ServerName doesn't match any certificates in the cache.
The uses for this are very niche; typically if a client is a CDN and passes through the ServerName of the downstream handshake but can accept a certificate with the origin's hostname instead, then you would set this as your origin's hostname. Note that Caddy must be managing a certificate for this name.
Causes **all** certificates to be issued internally by default, rather than through a (public) ACME CA such as Let's Encrypt. This is useful as a quick toggle in development environments.
Specifies the URL to the ACME CA's directory. It is strongly recommended to set this to Let's Encrypt's [staging endpoint <img src="/old/resources/images/external-link.svg" class="external-link">](https://letsencrypt.org/docs/staging-environment/) for testing or development. Default: ZeroSSL and Let's Encrypt's production endpoints.
Note that a globally-configured ACME CA may not apply to all sites; see the [hostname requirements](/docs/automatic-https#hostname-requirements) for using the default ACME issuer(s).
Configures the [ACME DNS challenge](/docs/automatic-https#dns-challenge) provider to use for all ACME transactions.
Requires a custom build of Caddy with a plugin for your DNS provider.
The tokens following the name of the provider set up the provider the same as if specified in the [`tls` directive's `acme` issuer](/docs/caddyfile/directives/tls#acme).
Configures [On-Demand TLS](/docs/automatic-https#on-demand-tls) where it is enabled, but does not enable it (to enable it, use the [`on_demand` subdirective of the `tls` directive](/docs/caddyfile/directives/tls#syntax)). Required for use in production environments, to prevent abuse.
- **ask** will cause Caddy to make an HTTP request to the given URL, asking whether a domain is allowed to have a certificate issued.
The request has a query string of `?domain=` containing the value of the domain name.
If the endpoint returns a `2xx` status code, Caddy will be authorized to obtain a certificate for that name. Any other status code will result in cancelling issuance of the certificate and erroring the TLS handshake.
The ask endpoint should return _as fast as possible_, in a few milliseconds, ideally. Typically, your endpoint should do a constant-time lookup in an database with an index by domain name; avoid loops. Avoid making DNS queries or other network requests.
Defines the issuer (or source) of TLS certificates.
This allows configuring issuers globally, instead of per-site as you would with the [`tls` directive's `issuer` subdirective](/docs/caddyfile/directives/tls#issuer).
May be repeated if you wish to configure more than one issuer to try. They will be tried in the order they are defined.
```caddy
{
cert_issuer acme {
...
}
cert_issuer zerossl {
...
}
}
```
##### `renew_interval`
How often to scan all loaded, managed certificates for expiration, and trigger renewal if expired.
Default: `10m`
```caddy
{
renew_interval 30m
}
```
##### `ocsp_interval`
How often to check if [OCSP staples <img src="/old/resources/images/external-link.svg" class="external-link">](https://en.wikipedia.org/wiki/OCSP_stapling) need updating.
- **root_common_name** is a list of one or more common names; Caddy will choose the first chain that has a root that matches with at least one of the specified common names.
- **any_common_name** is a list of one or more common names; Caddy will choose the first chain that has an issuer that matches with at least one of the specified common names.
Note that specifying `preferred_chains` as a global option will affect all issuers if there isn't any [overriding issuer level config](/docs/caddyfile/directives/tls#acme).
Customizes [HTTP servers](/docs/json/apps/http/servers/) with settings that potentially span multiple sites, and thus can't be rightly configured in site blocks. These options affect the listener/socket or other facilities beneath the HTTP layer.
Can be specified more than once with different `listener_address` values to configure different options per server. For example, `servers :443` will only apply to the server that is bound to the listener address `:443`. Omitting the listener address will apply the options to any remaining server.
When using `servers`, it will **only** apply to servers that **actually appear** in your Caddyfile (i.e. are produced by a site block). Remember, [Automatic HTTPS](/docs/automatic-https) will create a server listening on port `80` (or the [`http_port` option](#http_port)), to serve HTTP->HTTPS redirects and to solve the ACME HTTP challenge; this happens at runtime, i.e. _after_ the Caddyfile adapter applies `servers`. So in other words, this means that `servers`**will not** apply to `:80` unless you explicitly declare a site block like `http://` or `:80`.
If you are using the [`bind` directive](/docs/caddyfile/directives/bind) or the [`default_bind` global option](/docs/caddyfile/options#default_bind), the `listener_address`*MUST* match the bind address combined with the port of the site block, otherwise the settings won't be applied. For example:
A custom name to assign to this server. Usually helpful to identify a server by its name in logs and metrics. If not set, Caddy will define it dynamically using a `srvX` pattern, where `X` starts with `0` and increments based on the number of servers in the config.
Keep in mind that only servers produced by site blocks in your config will have settings applied. [Automatic HTTPS](/docs/automatic-https) creates an `:80` server (or [`http_port`](#http_port)) at runtime, so if you want rename it, you'll need at least an empty `http://` site block.
Allows configuring [listener wrappers](/docs/json/apps/http/servers/listener_wrappers/), which can modify the behaviour of the socket listener. They are applied in the given order.
There is a special no-op [`tls`](/docs/json/apps/http/servers/listener_wrappers/tls/) listener wrapper provided as a standard module which marks where TLS should be handled in the chain of listener wrappers. It should only be used if another listener wrapper must be placed in front of the TLS handshake. This _does not_ enable TLS for a server; e.g. if this is used on your `:80` HTTP server, it will still act as a no-op.
The included [`http_redirect`](/docs/json/apps/http/servers/listener_wrappers/http_redirect/) listener wrapper can look at the first few bytes of an incoming request to determine if it's likely HTTP (instead of TLS), and trigger an HTTP-to-HTTPS redirect on the same port but using the `https://` scheme. This is most useful when serving HTTPS on a non-standard port (other than `443`), since browsers will try HTTP unless the scheme is specified. Don't use this on an HTTP server. It must be placed _before_ the `tls` listener wrapper. For example:
Also included is the [`proxy_protocol`](/docs/json/apps/http/servers/listener_wrappers/proxy_protocol/) listener wrapper (prior to v2.7.0 it was only available via a plugin), which enables [PROXY protocol](https://github.com/haproxy/haproxy/blob/master/doc/proxy-protocol.txt) parsing (popularized by HAProxy). This must be used _before_ the `tls` listener wrapper since it parses plaintext data at the start of the connection:
- **read_body** is a [duration value](/docs/conventions#durations) that sets how long to allow a read from a client's upload. Setting this to a short, non-zero value can mitigate slowloris attacks, but may also affect legitimately slow clients. Defaults to no timeout.
- **read_header** is a [duration value](/docs/conventions#durations) that sets how long to allow a read from a client's request headers. Defaults to no timeout.
- **write** is a [duration value](/docs/conventions#durations) that sets how long to allow a write to a client. Note that setting this to a small value when serving large files may negatively affect legitimately slow clients. Defaults to no timeout.
- **idle** is a [duration value](/docs/conventions#durations) that sets the maximum time to wait for the next request when keep-alives are enabled. Defaults to 5 minutes to help avoid resource exhaustion.
Enabling this causes trusted requests to have the _real_ client IP parsed from HTTP headers (by default, `X-Forwarded-For`; see [`client_ip_headers`](#client-ip-headers) to configure other headers). If trusted, the client IP is added to [access logs](/docs/caddyfile/directives/log), is available as a `{client_ip}` [placeholder](/docs/caddyfile/concepts#placeholders), and allows the use of the [`client_ip` matcher](/docs/caddyfile/matchers#client-ip). If the request is not from a trusted proxy, then the client IP is set to the remote IP address of the direct incoming connection.
Some matchers or handlers may use the trust status of the request to make decisions. For example, if trusted, the [`reverse_proxy`](/docs/caddyfile/directives/reverse_proxy#defaults) handler will proxy and augment the sensitive `X-Forwarded-*` request headers.
Currently, only the `static` [IP source module](/docs/json/apps/http/servers/trusted_proxies/) is included with the standard distribution of Caddy, but this can be [extended](/docs/extending-caddy) with plugins to maintain a dynamic list of IP ranges.
As a shortcut, `private_ranges` can be used to match all private IPv4 and IPv6 ranges. It's the same as specifying all of these ranges: `192.168.0.0/16 172.16.0.0/12 10.0.0.0/8 127.0.0.1/8 fd00::/8 ::1`.
Pairing with [`trusted_proxies`](#trusted-proxies), allows configuring which headers to use to determine the client's IP address. By default, only `X-Forwarded-For` is considered. Multiple header fields can be specified, in which case the first non-empty header value is used.
Enables Prometheus metrics collection; necessary before scraping metrics. Note that metrics reduce performance on really busy servers. (Our community is working on improving this. Please get involved!)
The maximum size to parse from a client's HTTP request headers. If the limit is exceeded, the server will respond with HTTP status `431 Request Header Fields Too Large`. It accepts all formats supported by [go-humanize](https://github.com/dustin/go-humanize/blob/master/bytes.go). By default, the limit is `1MB`.
Enable full-duplex communication for HTTP/1 requests. Only has an effect if Caddy was built with Go 1.21 or later.
For HTTP/1 requests, the Go HTTP server by default consumes any unread portion of the request body before beginning to write the response, preventing handlers from concurrently reading from the request and writing the response. Enabling this option disables this behavior and permits handlers to continue to read from the request while concurrently writing the response.
Test thoroughly with your HTTP clients, as some older clients may not support full-duplex HTTP/1 which can cause them to deadlock. See [golang/go#57786](https://github.com/golang/go/issues/57786) for more info.
⚠️ This is an experimental feature. Subject to change or removal.
Since Caddy v2.5, by default, headers with potentially sensitive information (`Cookie`, `Set-Cookie`, `Authorization` and `Proxy-Authorization`) will be logged with empty values in access logs (see the [`log` directive](/docs/caddyfile/directives/log)).
If you wish to _not_ have these headers redacted, you may enable the `log_credentials` option.
Currently, enabling HTTP/2 (including H2C) necessarily implies enabling HTTP/1.1 because the Go standard library does not let us disable HTTP/1.1 when using its HTTP server. However, either HTTP/1.1 or HTTP/3 can be enabled independently.
Note that H2C ("Cleartext HTTP/2" or "H2 over TCP") and HTTP/3 are not implemented by the Go standard library, so some functionality or features may be limited. We recommend against enabling H2C unless it is absolutely necessary for your application.
Enabling this requires that a request's `Host` header matches the value of the `ServerName` sent by the client's TLS ClientHello, a necessary safeguard when using TLS client authentication. If there's a mismatch, HTTP status `421 Misdirected Request` response is written to the client.
This option will automatically be turned on if [client authentication](/docs/caddyfile/directives/tls#client_auth) is configured. This disallows TLS client auth bypass (domain fronting) which could otherwise be exploited by sending an unprotected SNI value during a TLS handshake, then putting a protected domain in the Host header after establishing connection. This behavior is a safe default, but you may explicitly turn it off with `insecure_off`; for example in the case of running a proxy where domain fronting is desired and access is not restricted based on hostname.
The PKI (Public Key Infrastructure) app is the foundation for Caddy's [Local HTTPS](/docs/automatic-https#local-https) and [ACME server](/docs/caddyfile/directives/acme_server) features. The app defines certificate authorities (CAs) which are capable of signing certificates.
The default CA ID is `local`. If the ID is omitted when configuring the `ca`, then `local` is assumed.
The [duration](/docs/conventions#durations) for which intermediate certificates are valid. This value **must** be less than the lifetime of the root cert (`3600d` or 10 years).
Default: `7d`. It is _not recommended_ to change this, unless absolutely necessary.
A key pair (certificate and private key) to use as the root for the CA. If not specified, one will be generated and managed automatically.
- **format** is the format in which the certificate and private key are provided. Currently, only `pem_file` is supported, which is the default, so this field is optional.
- **cert** is the certificate. This should be the path to a PEM file, when using `pem_file` format.
- **key** is the private key. This should be the path to a PEM file, when using `pem_file` format.
##### `intermediate`
A key pair (certificate and private key) to use as the intermediate for the CA. If not specified, one will be generated and managed automatically.
- **format** is the format in which the certificate and private key are provided. Currently, only `pem_file` is supported, which is the default, so this field is optional.
- **cert** is the certificate. This should be the path to a PEM file, when using `pem_file` format.
- **key** is the private key. This should be the path to a PEM file, when using `pem_file` format.
Caddy modules emit events when interesting things happen (or are about to happen).
Events typically include a metadata payload. The best way to learn about events and their payloads is from each module's documentation, but you may also see the events and their data payloads by enabling the [`debug` global option](#debug) and reading the logs.
For example, to run a command after a certificate is obtained ([third-party plugin <img src="/old/resources/images/external-link.svg" class="external-link">](https://github.com/mholt/caddy-events-exec) required), with a part of the event payload being passed to the script using a placeholder:
