redbean
single-file distributable web server

redbean is an open source webserver in a single-file that runs natively on six OSes for both AMD64 and ARM64. Basic idea is if you want to build a web app that runs anywhere, then you download the redbean.com file, put your .html and .lua files inside it using the zip command, and you've got a hermetic app you deploy and share.

redbean embeds Lua, SQLite, and MbedTLS into a fork() driven application server that benchmarks at 5.3 million qps on Threadripper. It's got a live bestline REPL with code completion and a UNIX module too, that lets you directly use the Cosmopolitan Libc system call interface. redbean furthermore provides sandboxing and system call tracing for security. This makes redbean a great fit for when you want to build an app that's vertically integrated into a single tiny file that runs on nearly all PCs and servers.

table of contents

documentation

modules

reference

community

download   [Linux] [Windows] [MacOS] [FreeBSD] [OpenBSD] [NetBSD]

redbean v3.0.0 was released on 2024-08-17.

build from source   [Linux]

# build once run anywhere using linux
git clone https://github.com/jart/cosmopolitan
cd cosmopolitan
make -j8 o//tool/net/redbean
o//tool/net/redbean -vv

supplementary tools

On cosmo.zip you'll find other command line tools, built as APE binaries, that are essential when developing software with redbean.

features

installation

redbean is designed to not need to be installed. Since there's no dependencies, all you need to do is download the binary and run it.

curl https://redbean.dev/redbean-demo-latest.com >redbean-demo.com
chmod +x redbean-demo.com
./redbean-demo.com -v

That'll start up a command line web server with a REPL. If you're on a desktop, it'll also launch a tab in your browser, visiting your redbean web server. If you go to http://localhost:8080/ then you should be able to see a listing of the content, with many demo scripts.

Having trouble? Here are some more specific instructions for various platforms, in case the above commands don't work.

MacOS

On macOS with Apple Silicon you need to have Xcode Command Line Tools installed for redbean to be able to bootstrap itself.

If you use zsh and have trouble running redbean, try saying sh -c ./redbean-demo.com. This is due to a bug that was fixed in zsh 5.9+. The same is the case for Python subprocess, old versions of Fish, etc.

If you get the MacOS error "... cannot be opened because the developer cannot be verified" then:

  1. Immediately launch System Settings, then go to Privacy & Security. llamafile should be listed at the bottom, with a button to Allow.
  2. If not, then change your command in the Terminal to be sudo spctl --master-disable; [llama launch command]; sudo spctl --master-enable. This is because --master-disable disables _all_ checking, so you need to turn it back on after quitting llama.

Linux

On some Linux systems, you might get errors relating to `run-detectors` or WINE. This is due to `binfmt_misc` registrations. You can fix that by adding an additional registration for the APE file format redbean uses:

sudo wget -O /usr/bin/ape https://cosmo.zip/pub/cosmos/bin/ape-$(uname -m).elf
sudo chmod +x /usr/bin/ape
sudo sh -c "echo ':APE:M::MZqFpD::/usr/bin/ape:' >/proc/sys/fs/binfmt_misc/register"
sudo sh -c "echo ':APE-jart:M::jartsr::/usr/bin/ape:' >/proc/sys/fs/binfmt_misc/register"

Windows

Generally speaking with Windows, you just need to make sure the APE binary has a .exe or .com extension. To make things simple for you:

curl -o zip.exe https://cosmo.zip/pub/cosmos/bin/zip
curl -o redbean-demo.com https://redbean.dev/redbean-demo-latest.com
.\redbean-demo.com

On WSL, there are many possible gotchas. One thing that helps solve them is this:

[Unit]
Description=cosmopolitan APE binfmt service
After=wsl-binfmt.service

[Service]
Type=oneshot
ExecStart=/bin/sh -c "echo ':APE:M::MZqFpD::/usr/bin/ape:' >/proc/sys/fs/binfmt_misc/register"

[Install]
WantedBy=multi-user.target

Put that in /etc/systemd/system/cosmo-binfmt.service. You also need to download ape-x86_64.elf and copy it to /usr/bin/ape with chmod +x.

Then run sudo systemctl enable cosmo-binfmt.

Another thing that's helped WSL users who experience issues, is to disable the WIN32 interop feature:

sudo sh -c "echo -1 > /proc/sys/fs/binfmt_misc/WSLInterop"

In the instance of getting a Permission Denied on disabling interop through CLI, it can be permanently disabled by adding the following in /etc/wsl.conf.

[interop]
enabled=false

usage

Here's basic overview of how you'd go about using redbean alongside InfoZIP which is a free tool that comes included with most UNIX systems.

echo '<b>hello</b>' >hello.html
zip redbean.com hello.html

# in first terminal
$ ./redbean.com -vv
I2022-04-29T07:17:28+000767:redbean] (srvr) listen http://127.0.0.1:8080
>: waiting for command...

# in second terminal
$ curl http://127.0.0.1:8080/hello.html
<b>hello</b>
$ curl -k https://127.0.0.1:8080/hello.html
<b>hello</b>

# add some stuff to the zip executable
$ echo 'Write("<b>hello</b>")' >hello.lua
$ zip redbean.com hello.lua

# perform a naked http request
$ printf 'GET /hello.lua\n\n' | nc 127.0.0.1 8080
<b>hello</b>

One GUI-like feature redbean itself offers is the ability to LaunchBrowser() at startup, which can be called from your your /.init.lua file.

source build

You have greater power to customize your redbean build if you build from source. For example:

git clone https://github.com/jart/cosmopolitan && cd cosmopolitan
make -j8 MODE=tiny o/tiny/tool/net
ls o/tiny/tool/net/redbean*.com
o/tiny/tool/net/redbean.com -vv
make -j8 MODE=tiny o/tiny/tool/net
o/tiny/tool/net/redbean.com -vv
make -j8 MODE=tinylinux o/tinylinux/tool/net
o/tinylinux/tool/net/redbean-original.com -vv

Some of the supported build modes are: MODE= (default), MODE=opt, rel, dbg, tiny, asan, optlinux, tinylinux, tinylinuxbsd, and tinysysv.

details

You can launch redbean in your terminal in the most verbose way possible by doing the following:

./redbean.com -vvmbag         # starts server verbosely
open http://127.0.0.1:8080/   # shows zip listing page
CTRL-C                        # 1x: graceful shutdown
CTRL-C                        # 2x: forceful shutdown

Assets can be added to the zip archive as follows:

zip redbean.com index.html                  # adds file
zip -r redbean.com mirrored-website         # adds directory
echo comment | zip -c redbean.com foo.html  # adds file w/ comment

By default, anything you add to the archive gets compressed. Sometimes you don't want that to happen. A good example is video files. The web browser will want to send HTTP range requests to seek in the video, in which case redbean requires that the asset be uncompressed.

zip -0 redbean.com video.mp4  # adds file without compression

You can run redbean interactively in your terminal as follows:

redbean.com -vv
CTRL-C                        # 1x: graceful shutdown
CTRL-C                        # 2x: forceful shutdown

The index.lua and index.html names are special since they're used to automatically figure out how to serve directories. Such files can appear in any directory, however the root directory is special. The default action for / is to show a listing page showing the contents of your zip central directory.

The listing page only applies to the root directory. However the default index page applies to subdirectories too. In order for it to work, there needs to be an empty directory entry in the zip. That should already be the default practice of your zip editor.

redbean normalizes the trailing slash for you automatically:

$ printf 'GET /a.example HTTP/1.0\n\n' | nc 127.0.0.1 8080
HTTP/1.0 307 Temporary Redirect
Location: /a.example/

Virtual hosting is accomplished this way too. The Host is simply prepended to the path, and if it doesn't exist, it gets removed.

$ printf 'GET / HTTP/1.1\nHost:a.example\n\n' | nc 127.0.0.1 8080
HTTP/1.1 200 OK
Link: <http://127.0.0.1/a.example/index.html>; rel=\"canonical\"

If you mirror a lot of websites within your redbean then you can actually tell your browser that redbean is your proxy server, in which redbean will act as your private version of the Internet.

wget                     \
  --mirror               \
  --convert-links        \
  --adjust-extension     \
  --page-requisites      \
  --no-parent            \
  --no-if-modified-since \
  http://a.example/index.html
zip -r redbean.com a.example/  # default page for directory
printf 'GET http://a.example HTTP/1.0\n\n' | nc 127.0.0.1 8080
HTTP/1.0 200 OK
Link: <http://127.0.0.1/a.example/index.html>; rel="canonical"

If you use a reverse proxy, then redbean recognizes the following provided that the proxy forwards requests over the local network:

X-Forwarded-For: 203.0.113.42:31337
X-Forwarded-Host: foo.example:80

There's a text/plain statistics page called /statusz that makes it easy to track and monitor the health of your redbean:

printf 'GET /statusz\n\n' | nc 127.0.0.1 8080

redbean will display an error page using the /redbean.png logo by default, embedded as a bas64 data uri. You can override the custom page for various errors by adding files to the zip root.

zip redbean.com 404.html      # custom not found page

flags

-h or -? help
-d daemonize
-u uniprocess
-z print port
-m log messages
-i interpreter mode
-b log message bodies
-a log resource usage
-g log handler latency
-E show crash reports to public ips
-j enable ssl client verify
-k disable ssl fetch verify
-Z log worker system calls
-f log worker function calls
-B only use stronger cryptography
-X disable ssl server and client support
-* permit self-modification of executable
-J disable non-ssl server and client support
-% hasten startup by not generating an rsa key
-s increase silence [repeatable]
-v increase verbosity [repeatable]
-V increase ssl verbosity [repeatable]
-S increase pledge sandboxing [repeatable]
-e CODE eval Lua code in arg [repeatable]
-F PATH eval Lua code in file [repeatable]
-H K:V sets http header globally [repeatable]
-D DIR overlay assets in local directory [repeatable]
-r /X=/Y redirect X to Y [repeatable]
-R /X=/Y rewrites X to Y [repeatable]
-K PATH tls private key path [repeatable]
-C PATH tls certificate(s) path [repeatable]
-A PATH add assets with path (recursive) [repeatable]
-M INT tunes max message payload size [def. 65536]
-t INT timeout ms or keepalive sec if <0 [def. 60000]
-p PORT listen port [def. 8080; repeatable]
-l ADDR listen addr [def. 0.0.0.0; repeatable]
-c SEC configures static cache-control
-W TTY use tty path to monitor memory pages
-L PATH log file location
-P PATH pid file location
-U INT daemon set user id
-G INT daemon set group id
-w PATH launch browser on startup
--strace enables system call tracing (see also -Z)
--ftrace enables function call tracing (see also -f)

keyboard

ctrl-d exit
ctrl-c ctrl-c exit
ctrl-e end
ctrl-a start
ctrl-b back
ctrl-f forward
ctrl-l clear
ctrl-h backspace
ctrl-d delete
ctrl-n next history
ctrl-p previous history
ctrl-r search history
ctrl-g cancel search
alt-< beginning of history
alt-> end of history
alt-f forward word
alt-b backward word
ctrl-right forward word
ctrl-left backward word
ctrl-k kill line forwards
ctrl-u kill line backwards
alt-h kill word backwards
ctrl-w kill word backwards
ctrl-alt-h kill word backwards
alt-d kill word forwards
ctrl-y yank
alt-y rotate kill ring and yank again
ctrl-t transpose
alt-t transpose word
alt-u uppercase word
alt-l lowercase word
alt-c capitalize word
ctrl-\ quit process
ctrl-s pause output
ctrl-q unpause output (if paused)
ctrl-q escaped insert
ctrl-alt-f forward expr
ctrl-alt-b backward expr
alt-right forward expr
alt-left backward expr
alt-shift-b barf expr
alt-shift-s slurp expr
ctrl-space set mark
ctrl-x ctrl-x goto mark
ctrl-z suspend process
alt-\ squeeze adjacent whitespace
protip remap caps lock to ctrl

lua server pages

Any files with the extension .lua will be dynamically served by redbean. Here's the simplest possible example:

Write('<b>Hello World</b>')

The Lua Server Page above should be able to perform at 700,000 responses per second on a Core i9, without any sort of caching. If you want a Lua handler that can do 1,000,000 responses per second, then try adding the following global handler to your /.init.lua file:

function OnHttpRequest()
   Write('<b>Hello World</b>')
end

Here's an example of a more typical workflow for Lua Server Pages using the redbean API:

SetStatus(200)
SetHeader('Content-Type', 'text/plain; charset=utf-8')
Write('<p>Hello ')
Write(EscapeHtml(GetParam('name')))

We didn't need the first two lines in the previous example, because they're implied by redbean automatically if you don't set them. Responses are also buffered until the script finishes executing. That enables redbean to make HTTP as easy as possible. In the future, API capabilities will be expanded to make possible things like websockets.

redbean embeds the Lua standard library. You can use packages such as io to persist and share state across requests and connections, as well as the StoreAsset function, and the lsqlite3 module.

Your Lua interpreter begins its life in the main process at startup in the .init.lua, which is likely where you'll want to perform all your expensive one-time operations like importing modules. Then, as requests roll in, isolated processes are cloned from the blueprint you created.

repl

Your redbean displays a Read-Eval-Print-Loop that lets you modify the state of the main server process while your server is running. Any changes will propagate into forked clients.

Your REPL is displayed only when redbean is run as a non-daemon in a Unix terminal or the Windows 10 command prompt or PowerShell. Since the REPL is a Lua REPL it's not included in a redbean-static builds.

redbean uses the same keyboard shortcuts as GNU Readline and Emacs. Some of its keyboard commands (listed in a previous section) were inspired by Paredit.

A history of your commands is saved to ~/.redbean_history.

If you love the redbean repl and want to use it as your language interpreter then you can pass the -i flag to put redbean into interpreter mode.

redbean.com -i binarytrees.lua 15

When the -i flag is passed (for interpreter mode), redbean won't start a web server and will instead functions like the lua command. The first command line argument becomes the script you want to run. If you don't supply a script, then the repl without a web server is displayed. This is useful for testing since redbean extensions and modules for the Lua language, are still made available. You can also write redbean scripts with shebang lines:

#!/usr/bin/redbean -i
print('hello world')

However UNIX operating systems usually require that interpreters be encoded in its preferred executable format. You can assimilate your redbean into the local format using the following commands:

$ file redbean.com
redbean.com: DOS/MBR boot sector
$ curl -o assimilate https://cosmo.zip/pub/cosmos/bin/assimilate
$ chmod +x assimilate
$ ./assimilate ./redbean.com
$ file redbean.com
redbean.com: ELF 64-bit LSB executable
$ sudo cp redbean.com /usr/bin/redbean

By following the above steps, redbean can be installed systemwide for multiple user accounts. It's also possible to chmod the binary to have setuid privileges. Please note that, if you do this, the UNIX section provides further details on APIs like unix.setuid that will help you remove root privileges from the process in the appropriate manner.

lua enhancements

We've made some enhancements to the Lua language that should make it more comfortable for C/C++ and Python developers. Some of these

special paths

/
redbean will generate a zip central directory listing for this page, and this page only, but only if there isn't an /index.lua or /index.html file defined.
/.init.lua
This script is run once in the main process at startup. This lets you modify the state of the Lua interpreter before connection processes are forked off. For example, it's a good idea to do expensive one-time computations here. You can also use this file to call the ProgramFOO() functions below. The init module load happens after redbean's arguments and zip assets have been parsed, but before calling functions like socket() and fork(). Note that this path is a hidden file so that it can't be unintentionally run by the network client.
/.reload.lua (deprecated; use OnServerReload instead)
This script is run from the main process when SIGHUP is received. This only applies to redbean when running in daemon mode. Any changes that are made to the Lua interpreter state will be inherited by future forked connection processes. Note that this path is a hidden file so that it can't be unintentionally run by the network client.
/.lua/...
Your Lua modules go in this directory. The way it works is redbean sets Lua's package.path to /zip/.lua/?.lua;/zip/.lua/?/init.lua by default. Cosmopolitan Libc lets system calls like open read from the ZIP structure, if the filename is prefixed with /zip/. So this works like magic.
/redbean.png
If it exists, it'll be used as the / listing page icon, embedded as a base64 URI.
/usr/share/zoneinfo
This directory contains a subset of the timezone database. Your TZ environment variable controls which one of these files is used by functions such as unix.localtime().
/usr/share/ssl/root
This directory contains your root certificate authorities. It is needed so the Fetch() HTTPS client API can verify that a remote certificate was signed by a third party. You can add your own certificate files to this directory within the ZIP executable. If you enable HTTPS client verification then redbean will check that HTTPS clients (a) have a certificate and (b) it was signed.
/.args
Specifies default command-line arguments.

There's one argument per line. Trailing newline is ignored. If the special argument ... is not encountered, then the replacement will only happen if no CLI args are specified. If the special argument ... is encountered, then it'll be replaced with whatever CLI args were specified by the user.

For example, you might want to use redbean.com in interpreter mode, where your script file is inside the zip. Then, if your redbean is run, what you want is to have the default behavior be running your script. In that case, you might:

$ cat <<'EOF' >.args
-i
/zip/hello.lua
EOF

$ cat <<'EOF' >hello.lua
print("hello world")
EOF

$ zip redbean.com .args hello.lua
$ ./redbean.com
hello world

Please note that if you ran:

$ ./redbean.com -vv

Then the default mode of redbean will kick back in. To prevent that from happening, simply add the magic arg ... to the end of your .args file.

globals

arg: array[str]
Array of command line arguments, excluding those parsed by getopt() in the C code, which stops parsing at the first non-hyphenated arg. In some cases you can use the magic -- argument to delimit C from Lua arguments.

For example, if you launch your redbean as follows:

redbean -v arg1 arg2

Then your /.init.lua file will have the arg array like:

arg[-1] = '/usr/bin/redbean'
arg[ 0] = '/zip/.init.lua'
arg[ 1] = 'arg1'
arg[ 2] = 'arg2'

If you launch redbean in interpreter mode (rather than web server) mode, then an invocation like this:

./redbean.com -i script.lua arg1 arg2

Would have an arg array like this:

arg[-1] = './redbean.com'
arg[ 0] = 'script.lua'
arg[ 1] = 'arg1'
arg[ 2] = 'arg2'

hooks

OnHttpRequest()
If this function is defined in the global scope by your /.init.lua then redbean will call it at the earliest possible moment to hand over control for all messages (with the exception of OPTIONS *). See functions like Route which asks redbean to do its default thing from the handler.
OnError(status:int, message:string)
If this function is defined and if any errors occurs in OnHttpRequest() then this method will be called instead of displaying the default error page. Useful if you need to display the error page using your specific code or send it to any tier service.
OnClientConnection(ip:int, port:int, serverip:int, serverport:int) → bool
If this function is defined it'll be called from the main process each time redbean accepts a new client connection. If it returns true then redbean will close the connection without calling fork.
OnLogLatency(reqtimeus:int, contimeus:int)
If this function is defined it'll be called from the child worker process each time redbean completes the handling of a request, but before the response is sent. The handler receives the time (in µs) since the request handling and connection handling started.
OnProcessCreate(pid:int, ip:int, port:int, serverip:int, serverport:int)
If this function is defined it'll be called from the main process each time redbean forks a connection handler worker process. The ip/port of the remote client is provided, along with the ip/port of the listening interface that accepted the connection. This may be used to create a server activity dashboard, in which case the data provider handler should set SetHeader('Connection','Close'). This won't be called in uniprocess mode.
OnProcessDestroy(pid:int)
If this function is defined it'll be called from the main process each time redbean reaps a child connection process using wait4(). This won't be called in uniprocess mode.
OnServerHeartbeat()
If this function is defined it'll be called from the main process on each server heartbeat. The heartbeat interval is configurable with ProgramHeartbeatInterval.
OnServerListen(socketdescriptor:int, serverip:int, serverport:int) → bool
If this function is defined it'll be called from the main process before redbean starts listening on a port. This hook can be used to modify socket configuration to set SO_REUSEPORT, for example. If it returns true, redbean will not listen to that ip/port.
OnServerReload(reindex:bool)
If this function is defined it'll be called from the main process on each server reload triggered by SIGHUP (for daemonized) and SIGUSR1 (for all) redbean instances. reindex indicates if redbean assets have been re-indexed following the signal.
OnServerStart()
If this function is defined it'll be called from the main process right before the main event loop starts.
OnServerStop()
If this function is defined it'll be called from the main process after all the connection processes have been reaped and exit() is ready to be called.
OnWorkerStart()
If this function is defined it'll be called from the child worker process after it's been forked and before messages are handled. This won't be called in uniprocess mode.
OnWorkerStop()
If this function is defined it'll be called from the child worker process once _exit() is ready to be called. This won't be called in uniprocess mode.

functions

Write(data:str)
Appends data to HTTP response payload buffer. This is buffered independently of headers.
SetStatus(code:int[, reason:str])
Starts an HTTP response, specifying the parameters on its first line. reason is optional since redbean can fill in the appropriate text for well-known magic numbers, e.g. 200, 404, etc. This method will reset the response and is therefore mutually exclusive with ServeAsset and other Serve* functions. If a status setting function isn't called, then the default behavior is to send 200 OK.
SetHeader(name:str, value:str)
Appends HTTP header to response header buffer. name is case-insensitive and restricted to non-space ASCII. value is a UTF-8 string that must be encodable as ISO-8859-1. Leading and trailing whitespace is trimmed automatically. Overlong characters are canonicalized. C0 and C1 control codes are forbidden, with the exception of tab. This function automatically calls SetStatus(200, "OK") if a status has not yet been set. As SetStatus and Serve* functions reset the response, SetHeader needs to be called after SetStatus and Serve* functions are called. The header buffer is independent of the payload buffer. Neither is written to the wire until the Lua Server Page has finished executing. This function disallows the setting of certain headers such as and Content-Range which are abstracted by the transport layer. In such cases, consider calling ServeAsset.
SetCookie(name:str, value:str[, options:table])
Appends Set-Cookie HTTP header to the response header buffer. Several Set-Cookie headers can be added to the same response. __Host- and __Secure- prefixes are supported and may set or overwrite some of the options (for example, specifying __Host- prefix sets the Secure option to true, sets the path to "/", and removes the Domain option). The following options can be used (their lowercase equivalents are supported as well):
GetParam(name:str) → value:str
Returns first value associated with name. name is handled in a case-sensitive manner. This function checks Request-URL parameters first. Then it checks application/x-www-form-urlencoded from the message body, if it exists, which is common for HTML forms sending POST requests. If a parameter is supplied matching name that has no value, e.g. foo in ?foo&bar=value, then the returned value will be nil, whereas for ?foo=&bar=value it would be "". To differentiate between no-equal and absent, use the HasParam function. The returned value is decoded from ISO-8859-1 (only in the case of Request-URL) and we assume that percent-encoded characters were supplied by the client as UTF-8 sequences, which are returned exactly as the client supplied them, and may therefore may contain overlong sequences, control codes, NUL characters, and even numbers which have been banned by the IETF. It is the responsibility of the caller to impose further restrictions on validity, if they're desired.
EscapeHtml(str) → str
Escapes HTML entities: The set of entities is &><"' which become &amp;&gt;&lt;&quot;&#39;. This function is charset agnostic and will not canonicalize overlong encodings. It is assumed that a UTF-8 string will be supplied. See escapehtml.c.
LaunchBrowser([path:str])
Launches web browser on local machine with URL to this redbean server. It is the responsibility of the caller to escape the path with EscapePath if needed, as it's not escaped automatically. This function may be called from your /.init.lua.
CategorizeIp(ip:uint32) → str
Returns a string describing an IP address. This is currently Class A granular. It can tell you if traffic originated from private networks, ARIN, APNIC, DOD, etc.
DecodeLatin1(iso-8859-1:str) → utf-8:str
Turns ISO-8859-1 string into UTF-8.
EncodeHex(binary:str) → ascii:str
Turns binary into ASCII base-16 hexadecimal lowercase string.
DecodeHex(ascii:str) → binary:str
Turns ASCII base-16 hexadecimal byte string into binary string, case-insensitively. Non-hex characters may not appear in string.
DecodeBase32(ascii:str[, alphabet:str]) → binary:str
Turns ASCII into binary using provided alphabet. The default decoding uses Crockford's base32 alphabet in a permissive way that ignores whitespaces and dash ('-') and stops at the first character outside of the alphabet.
EncodeBase32(binary:str[, alphabet:str]) → ascii:str
Turns binary into ASCII using provided alphabet (using Crockford's base32 encoding by default). Any alphabet that has a power of 2 length (up to 128) may be supplied for encoding and decoding, which allows to provide alternative base32 encodings.
DecodeBase64(ascii:str) → binary:str
Turns ASCII into binary, in a permissive way that ignores characters outside the base64 alphabet, such as whitespace. See decodebase64.c.
EncodeBase64(binary:str) → ascii:str
Turns binary into ASCII. This can be used to create HTML data: URIs that do things like embed a PNG file in a web page. See encodebase64.c.
DecodeJson(input:str)
├─→ int64
├─→ string
├─→ double
├─→ array
├─→ object
├─→ false
├─→ true
├─→ nil
└─→ nil, error:str
Turns JSON string into a Lua data structure.

This is a generally permissive parser, in the sense that like v8, it permits scalars as top-level values. Therefore we must note that this API can be thought of as special, in the sense

val = assert(DecodeJson(str))

will usually do the right thing, except in cases where false or null are the top-level value. In those cases, it's needed to check the second value too in order to discern from error

val, err = DecodeJson(str)
if not val then
   if err then
      print('bad json', err)
   elseif val == nil then
      print('val is null')
   elseif val == false then
      print('val is false')
   end
end

This parser supports 64-bit signed integers. If an overflow happens, then the integer is silently coerced to double, as consistent with v8. If a double overflows into Infinity, we coerce it to null since that's what v8 does, and the same goes for underflows which, like v8, are coerced to 0.0.

When objects are parsed, your Lua object can't preserve the original ordering of fields. As such, they'll be sorted by EncodeJson() and may not round-trip with original intent

This parser has perfect conformance with JSONTestSuite.

This parser validates utf-8 and utf-16.

EncodeJson(value[, options:table])
├─→ json:str
├─→ true [if useoutput]
└─→ nil, error:str
Turns Lua data structure into JSON string.

Since Lua tables are both hashmaps and arrays, we use a simple fast algorithm for telling the two apart. Tables with non-zero length (as reported by #) are encoded as arrays, and any non-array elements are ignored. For example:

>: EncodeJson({2})
"[2]"
>: EncodeJson({[1]=2, ["hi"]=1})
"[2]"

If there are holes in your array, then the serialized array will exclude everything after the first hole. If the beginning of your array is a hole, then an error is returned.

>: EncodeJson({[1]=1, [3]=3})
"[1]"
>: EncodeJson({[2]=1, [3]=3})
"[]"
>: EncodeJson({[2]=1, [3]=3})
nil     "json objects must only use string keys"

If the raw length of a table is reported as zero, then we check for the magic element [0]=false. If it's present, then your table will be serialized as empty array []. An entry is inserted by DecodeJson() automatically, only when encountering empty arrays, and it's necessary in order to make empty arrays round-trip. If raw length is zero and [0]=false is absent, then your table will be serialized as an iterated object.

>: EncodeJson({})
"{}"
>: EncodeJson({[0]=false})
"[]"
>: EncodeJson({["hi"]=1})
"{\"hi\":1}"
>: EncodeJson({["hi"]=1, [0]=false})
"[]"
>: EncodeJson({["hi"]=1, [7]=false})
nil     "json objects must only use string keys"

The following options may be used:

This function will return an error if:

We assume strings in value contain UTF-8. This serializer currently does not produce UTF-8 output. The output format is right now ASCII. Your UTF-8 data will be safely transcoded to \uXXXX sequences which are UTF-16. Overlong encodings in your input strings will be canonicalized rather than validated.

NaNs are serialized as null and Infinities are null which is consistent with the v8 behavior.

EncodeLua(value[, options:table])
├─→ luacode:str
├─→ true [if useoutput]
└─→ nil, error:str
Turns Lua data structure into Lua code string.

Since Lua uses tables as both hashmaps and arrays, tables will only be serialized as an array with determinate order, if it's an array in the strictest possible sense.

  1. for all 𝑘=𝑣 in table, 𝑘 is an integer ≥1
  2. no holes exist between MIN(𝑘) and MAX(𝑘)
  3. if non-empty, MIN(𝑘) is 1

In all other cases, your table will be serialized as an object which is iterated and displayed as a list of (possibly) sorted entries that have equal signs.

>: EncodeLua({3, 2})
"{3, 2}"
>: EncodeLua({[1]=3, [2]=3})
"{3, 2}"
>: EncodeLua({[1]=3, [3]=3})
"{[1]=3, [3]=3}"
>: EncodeLua({["hi"]=1, [1]=2})
"{[1]=2, hi=1}"

The following options may be used:

If a user data object has a __repr or __tostring meta method, then that'll be used to encode the Lua code.

This serializer is designed primarily to describe data. For example, it's used by the REPL where we need to be able to ignore errors when displaying data structures, since showing most things imperfectly is better than crashing. Therefore this isn't the kind of serializer you'd want to use to persist data in prod. Try using the JSON serializer for that purpose.

Non-encodable value types (e.g. threads, functions) will be represented as a string literal with the type name and pointer address. The string description is of an unspecified format that could most likely change. This encoder detects cyclic tables; however instead of failing, it embeds a string of unspecified layout describing the cycle.

Integer literals are encoded as decimal. However if the int64 number is ≥256 and has a population count of 1 then we switch to representating the number in hexadecimal, for readability. Hex numbers have leading zeroes added in order to visualize whether the number fits in a uint16, uint32, or int64. Also some numbers can only be encoded expressionally. For example, NaNs are serialized as 0/0, and Infinity is math.huge.

>: 7000
7000
>: 0x100
0x0100
>: 0x10000
0x00010000
>: 0x100000000
0x0000000100000000
>: 0/0
0/0
>: 1.5e+9999
math.huge
>: -9223372036854775807 - 1
-9223372036854775807 - 1

The only failure return condition currently implemented is when C runs out of heap memory.

EncodeLatin1(utf-8:str[, flags:int]) → iso-8859-1:str
Turns UTF-8 into ISO-8859-1 string.
EscapeFragment(str) → str
Escapes URL #fragment. The allowed characters are -/?.~_@:!$&'()*+,;=0-9A-Za-z and everything else gets %XX encoded. Please note that '& can still break HTML and that '() can still break CSS URLs. This function is charset agnostic and will not canonicalize overlong encodings. It is assumed that a UTF-8 string will be supplied. See kescapefragment.S.
EscapeHost(str) → str
Escapes URL host. See kescapeauthority.S
EscapeLiteral(str) → str
Escapes JavaScript or JSON string literal content. The caller is responsible for adding the surrounding quotation marks. This implementation \uxxxx sequences for all non-ASCII sequences. HTML entities are also encoded, so the output doesn't need EscapeHtml. This function assumes UTF-8 input. Overlong encodings are canonicalized. Invalid input sequences are assumed to be ISO-8859-1. The output is UTF-16 since that's what JavaScript uses. For example, some individual codepoints such as emoji characters will encode as multiple \uxxxx sequences. Ints that are impossible to encode as UTF-16 are substituted with the \xFFFD replacement character. See escapejsstringliteral.c.
EscapeParam(str) → str
Escapes URL parameter name or value. The allowed characters are -.*_0-9A-Za-z and everything else gets %XX encoded. This function is charset agnostic and will not canonicalize overlong encodings. It is assumed that a UTF-8 string will be supplied. See kescapeparam.S.
EscapePass(str) → str
Escapes URL password. See kescapeauthority.S.
EscapePath(str) → str
Escapes URL path. This is the same as EscapeSegment except slash is allowed. The allowed characters are -.~_@:!$&'()*+,;=0-9A-Za-z/ and everything else gets %XX encoded. Please note that '& can still break HTML, so the output may need EscapeHtml too. Also note that '() can still break CSS URLs. This function is charset agnostic and will not canonicalize overlong encodings. It is assumed that a UTF-8 string will be supplied. See kescapepath.S.
EscapeSegment(str) → str
Escapes URL path segment. This is the same as EscapePath except slash isn't allowed. The allowed characters are -.~_@:!$&'()*+,;=0-9A-Za-z and everything else gets %XX encoded. Please note that '& can still break HTML, so the output may need EscapeHtml too. Also note that '() can still break CSS URLs. This function is charset agnostic and will not canonicalize overlong encodings. It is assumed that a UTF-8 string will be supplied. See kescapesegment.S.
EscapeUser(str) → str
Escapes URL username. See kescapeauthority.S.
EvadeDragnetSurveillance(bool)
If this option is programmed then redbean will not transmit a Server Name Indicator (SNI) when performing Fetch() requests. This function is not available in unsecure mode.
UuidV4() -> str
Returns a uuid v4 string.
UuidV7() -> str
Returns a uuid v7 string.
Fetch(url:str[, body:str|{method=value:str, body=value:str, headers=table,...,...}])
├─→ status:int, {header:str=value:str,...}, body:str
└─→ nil, error:str
Sends an HTTP/HTTPS request to the specified URL. If only the URL is provided, then a GET request is sent. If both URL and body parameters are specified, then a POST request is sent. If any other method needs to be specified (for example, PUT or DELETE), then passing a table as the second value allows setting method and body values as well other options: When the redirect is being followed, the same method and body values are being sent in all cases except when 303 status is returned. In that case the method is set to GET and the body is removed before the redirect is followed. Note that if these (method/body) values are provided as table fields, they will be modified in place.
FormatHttpDateTime(seconds:int) → rfc1123:str
Converts UNIX timestamp to an RFC1123 string that looks like this: Mon, 29 Mar 2021 15:37:13 GMT. See formathttpdatetime.c.
FormatIp(uint32) → str
Turns integer like 0x01020304 into a string like 1.2.3.4. See also ParseIp for the inverse operation.
Curve25519(secret:str, public_or_basepoint:str) -> str
Returns the shared key for the specified secret and public values. The second argument can be either a public key or a basepoint, used for generating a shared or public key.

Here a small usage sample:

          >: secret1 = GetRandomBytes(32)
          >: public1 = Curve25519(secret1, "\9")
          >: secret2 = GetRandomBytes(32)
          >: public2 = Curve25519(secret2, "\9")
          >: shared_key1 = Curve25519(secret1, public2)
          >: shared_key2 = Curve25519(secret2, public1)
          >: shared_key1 == shared_key2
          true
GetRemoteAddr()
├─→ ip:uint32,port:uint16
└─→ nil
Returns client ip4 address and port, e.g. 0x01020304,31337 would represent 1.2.3.4:31337. This is the same as GetClientAddr except it will use the ip:port from the X-Forwarded-For header when the IP returned by GetClientAddr is a trusted IP (returns true when checked against IsTrustedIp). When multiple addresses are present in the header, the last/right-most address is used. Returns nil if the address doesn't parse as IPv4 value (e.g. because it is misformatted or an IPv6 address).
GetResponseBody()
├─→ body:str
└─→ nil, error:str
Returns the (uncompressed) response message body if present or an empty string. May also return a partial or empty string during streaming, as the full content may not be known at the call time. Returns an error when decompression fails.
GetClientAddr() → ip:uint32, port:uint16
Returns client socket ip4 address and port, e.g. 0x01020304,31337 would represent 1.2.3.4:31337. Please consider using GetRemoteAddr instead, since the latter takes into consideration reverse proxy scenarios.
GetServerAddr() → ip:uint32, port:uint16
Returns address to which listening server socket is bound, e.g. 0x01020304,8080 would represent 1.2.3.4:8080. If -p 0 was supplied as the listening port, then the port in this string will be whatever number the operating system assigned.
GetDate() → seconds:int
Returns date associated with request that's used to generate the Date header, which is now, give or take a second. The returned value is a UNIX timestamp.
GetHeader(name:str) → value:str
Returns HTTP header. name is case-insensitive. The header value is returned as a canonical UTF-8 string, with leading and trailing whitespace trimmed, which was decoded from ISO-8859-1, which is guaranteed to not have C0/C1 control sequences, with the exception of the tab character. Leading and trailing whitespace is automatically removed. In the event that the client suplies raw UTF-8 in the HTTP message headers, the original UTF-8 sequence can be losslessly restored by counter-intuitively recoding the returned string back to Latin1. If the requested header is defined by the RFCs as storing comma-separated values (e.g. Allow, Accept-Encoding) and the field name occurs multiple times in the message, then this function will fold those multiple entries into a single string.
GetHeaders() → {name:str=value:str,...}
Returns HTTP headers as dictionary mapping header key strings to their UTF-8 decoded values. The ordering of headers from the request message is not preserved. Whether or not the same key can repeat depends on whether or not it's a standard header, and if so, if it's one of the ones that the RFCs define as repeatable. See khttprepeatable.c. Those headers will not be folded. Standard headers which aren't on that list, will be overwritten with the last-occurring one during parsing. Extended headers are always passed through exactly as they're received. Please consider using GetHeader API if possible since it does a better job abstracting these issues.
GetLogLevel() → int
Returns logger verbosity level. Likely return values are kLogDebug > kLogVerbose > kLogInfo > kLogWarn > kLogError > kLogFatal.
GetHost() → str
Returns host associated with request. This will be the Host header, if it's supplied. Otherwise it's the bind address.
GetHostOs() → str
Returns string that describes the host OS.

This can return:

GetHostIsa() → str
Returns string describing host instruction set architecture.

This can return:

GetMonospaceWidth(str|char) → int
Returns monospace display width of string. This is useful for fixed-width formatting. For example, CJK characters typically take up two cells. This function takes into consideration combining characters, which are discounted, as well as control codes and ANSI escape sequences.
GetMethod() → str
Returns HTTP method. Normally this will be GET, HEAD, or POST in which case redbean normalizes this value to its uppercase form. Anything else that the RFC classifies as a "token" string is accepted too, which might contain characters like &".
GetParams() → {{name:str[, value:str]},...}
Returns name=value parameters from Request-URL and application/x-www-form-urlencoded message body in the order they were received. This may contain duplicates. The inner array will have either one or two items, depending on whether or not the equals sign was used.
GetPath() → str
Returns the Request-URL path. This is guaranteed to begin with "/". It is further guaranteed that no "//" or "/." exists in the path. The returned value is returned as a UTF-8 string which was decoded from ISO-8859-1. We assume that percent-encoded characters were supplied by the client as UTF-8 sequences, which are returned exactly as the client supplied them, and may therefore may contain overlong sequences, control codes, NUL characters, and even numbers which have been banned by the IETF. redbean takes those things into consideration when performing path safety checks. It is the responsibility of the caller to impose further restrictions on validity, if they're desired.
GetEffectivePath() → str
Returns path as it was resolved by the routing algorithms, which might contain the virtual host prepended if used.
GetScheme() → str
Returns scheme from Request-URL, if any.
GetPayload() → str
Returns the request message payload, or empty string if there isn't one.
GetStatus() → int
Returns current status (as set by an earlier SetStatus call) or nil if the status hasn't been set yet.
GetTime() → seconds:number
Returns current time as a UNIX timestamp with 0.0001s precision.
GetUrl() → str
Returns the effective Request-URL as an ASCII string, where illegal characters or UTF-8 is guaranteed to be percent encoded, and has been normalized to include either the Host or X-Forwarded-Host headers, if they exist, and possibly a scheme too if redbean is being used as an HTTP proxy server. In the future this API might change to return an object instead.
GetHttpVersion() → int
Returns the request HTTP protocol version, which can be 9 for HTTP/0.9, 10 for HTTP/1.0, or 11 for HTTP/1.1.
also available under the deprecated name GetVersion
GetHttpReason(code:int) → str
Returns a string describing the HTTP reason phrase. See gethttpreason.c
GetRandomBytes([length:int]) → str
Returns string with the specified number of random bytes (1..256). If no length is specified, then a string of length 16 is returned.
GetRedbeanVersion() → int
Returns the Redbean version in the format 0xMMmmpp, with major (MM), minor (mm), and patch (pp) versions encoded. The version value 1.4 would be represented as 0x010400.
GetZipPaths([prefix:str]) → {path:str,...}
Returns paths of all assets in the zip central directory, prefixed by a slash. If prefix parameter is provided, then only paths that start with the prefix (case sensitive) are returned.
HasParam(name:str) → bool
Returns true if parameter with name was supplied in either the Request-URL or an application/x-www-form-urlencoded message body.
HidePath(prefix:str)
Programs redbean / listing page to not display any paths beginning with prefix. This function should only be called from /.init.lua.
IsHiddenPath(path:str) → bool
Returns true if the prefix of the given path is set with HidePath.
IsPublicIp(uint32) → bool
Returns true if IP address is not a private network (10.0.0.0/8, 172.16.0.0/12, 192.168.0.0/16) and is not localhost (127.0.0.0/8). Note: we intentionally regard TEST-NET IPs as public.
IsPrivateIp(uint32) → bool
Returns true if IP address is part of a private network (10.0.0.0/8, 172.16.0.0/12, 192.168.0.0/16).
IsLoopbackClient() → bool
Returns true if the client IP address (returned by GetRemoteAddr) is part of the localhost network (127.0.0.0/8).
IsLoopbackIp(uint32) → bool
Returns true if IP address is part of the localhost network (127.0.0.0/8).
IsAssetCompressed(path:str) → bool
Returns true if ZIP artifact at path is stored on disk using DEFLATE compression. also available under the deprecated name IsCompressed
IndentLines(str[, int]) → str
Adds spaces to beginnings of multiline string. If the int parameter is not supplied then 1 space will be added.
LoadAsset(path:str) → str
Returns contents of file as string. The asset may be sourced from either the zip (decompressed) or the local filesystem if the -D flag was used. If slurping large file into memory is a concern, then consider using ServeAsset which can serve directly off disk.
StoreAsset(path:str, data:str[, mode:int])
Stores asset to executable's ZIP central directory. This currently happens in an append-only fashion and is still largely in the proof-of-concept stages. Currently only supported on Linux, XNU, and FreeBSD. In order to use this feature, the -* flag must be passed.
Log(level:int, message:str)
Emits message string to log, if level is less than or equal to GetLogLevel. If redbean is running in interactive mode, then this will log to the console. If redbean is running as a daemon or the -L LOGFILE flag is passed, then this will log to the file. Reasonable values for level are kLogDebug > kLogVerbose > kLogInfo > kLogWarn > kLogError > kLogFatal. The logger emits timestamps in the local timezone with microsecond precision. If log entries are emitted more frequently than once per second, then the log entry will display a delta timestamp, showing how much time has elapsed since the previous log entry. This behavior is useful for quickly measuring how long various portions of your code take to execute.
ParseHttpDateTime(rfc1123:str) → seconds:int
Converts RFC1123 string that looks like this: Mon, 29 Mar 2021 15:37:13 GMT to a UNIX timestamp. See parsehttpdatetime.c.
ParseUrl(str) → URL
Parses URL.

An object containing the following fields is returned:

- scheme is a string, e.g. "http" - user is the username string, or nil if absent - pass is the password string, or nil if absent - host is the hostname string, or nil if url was a path - port is the port string, or nil if absent - path is the path string, or nil if absent - params is the URL paramaters, e.g. /?a=b&c would be represented as the data structure {{"a", "b"}, {"c"}, ...} - fragment is the stuff after the # character

flags may have:

- kUrlPlus to turn + into space - kUrlLatin1 to transcode ISO-8859-1 input into UTF-8

This parser is charset agnostic. Percent encoded bytes are decoded for all fields. Returned values might contain things like NUL characters, spaces, control codes, and non-canonical encodings. Absent can be discerned from empty by checking if the pointer is set.

There's no failure condition for this routine. This is a permissive parser. This doesn't normalize path segments like . or .. so use IsAcceptablePath() to check for those. No restrictions are imposed beyond that which is strictly necessary for parsing. All the data that is provided will be consumed to the one of the fields. Strict conformance is enforced on some fields more than others, like scheme, since it's the most non-deterministically defined field of them all.

Please note this is a URL parser, not a URI parser. Which means we support everything the URI spec says we should do except for the things we won't do, like tokenizing path segments into an array and then nesting another array beneath each of those for storing semicolon parameters. So this parser won't make SIP easy. What it can do is parse HTTP URLs and most URIs like data:opaque, better in fact than most things which claim to be URI parsers.

IsAcceptablePath(str) → bool
Returns true if path doesn't contain ".", ".." or "//" segments See isacceptablepath.c
IsReasonablePath(str) → bool
Returns true if path doesn't contain "." or ".." segments See isreasonablepath.c
EncodeUrl(URL) → str
This function is the inverse of ParseUrl. The output will always be correctly formatted. The exception is if illegal characters are supplied in the scheme field, since there's no way of escaping those. Opaque parts are escaped as though they were paths, since many URI parsers won't understand things like an unescaped question mark in path.
ParseIp(str) → int
Converts IPv4 address string to integer, e.g. "1.2.3.4" → 0x01020304, or returns -1 for invalid inputs. See also FormatIp for the inverse operation.
GetAssetComment(path:str) → str
Returns comment text associated with asset in the ZIP central directory. Also available as GetComment (deprecated).
GetAssetLastModifiedTime(path:str) → seconds:number
Returns UNIX timestamp for modification time of a ZIP asset (or local file if the -D flag is used). If both a file and a ZIP asset are present, then the file is used. Also available as GetLastModifiedTime (deprecated).
GetAssetMode(path:str) → int
Returns UNIX-style octal mode for ZIP asset (or local file if the -D flag is used)
GetAssetSize(path:str) → int
Returns byte size of uncompressed contents of ZIP asset (or local file if the -D flag is used)
GetBody() → str
Returns the request message body if present or an empty string.
also available under the deprecated name GetPayload
GetCookie(name:str) → str
Returns cookie value.
Md5(str) → str
Computes MD5 checksum, returning 16 bytes of binary.
Sha1(str) → str
Computes SHA1 checksum, returning 20 bytes of binary.
Sha224(str) → str
Computes SHA224 checksum, returning 28 bytes of binary.
Sha256(str) → str
Computes SHA256 checksum, returning 32 bytes of binary.
Sha384(str) → str
Computes SHA384 checksum, returning 48 bytes of binary.
Sha512(str) → str
Computes SHA512 checksum, returning 64 bytes of binary.
GetCryptoHash(name:str, payload:str[, key:str]) → str
Returns value of the specified cryptographic hash function. If the key is provided, then HMAC value of the same function is returned. The name can be one of the following strings: MD5, SHA1, SHA224, SHA256, SHA384, SHA512, and BLAKE2B256.
IsDaemon() → bool
Returns true if -d flag was passed to redbean.
ProgramAddr(str)
ProgramAddr(ip:int)
Configures the address on which to listen. This can be called multiple times to set more than one address. If an integer is provided then it should be a word-encoded IPv4 address, such as the ones returned by ResolveIp(). If a string is provided, it will first be passed to ParseIp() to see if it's an IPv4 address. If it isn't, then a HOSTS.TXT lookup is performed, with fallback to the system-configured DNS resolution service. Please note that in MODE=tiny the HOSTS.TXT and DNS resolution isn't included, and therefore an IP must be provided.
ProgramGid(int)
Same as the -G flag if called from .init.lua for setgid()
ProgramDirectory([directory:str]) → {directory, ...}
Same as the -D flag if called from .init.lua for overlaying local file system directories. This may be called multiple times. The first directory programmed is preferred. These currently do not show up in the index page listing.

This call also modifies package.path value to either prepend the added directory in front of the default path (if found) or to append it (in all other cases). If no directory is provided, then a table with previously set directories is returned.

ProgramLogMessages(bool)
Same as the -m flag if called from .init.lua for logging message headers only.
ProgramLogBodies(bool)
Same as the -b flag if called from .init.lua for logging message bodies as part of POST / PUT / etc. requests.
ProgramLogPath(bool)
Same as the -L flag if called from .init.lua for setting the log file path on the local file system. It's created if it doesn't exist. This is called before de-escalating the user / group id. The file is opened in append only mode. If the disk runs out of space then redbean will truncate the log file if has access to change the log file after daemonizing.
ProgramPidPath(str)
Same as the -P flag if called from .init.lua for setting the pid file path on the local file system. It's useful for reloading daemonized redbean using kill -HUP $(cat /var/run/redbean.pid) or terminating redbean with kill $(cat /var/run/redbean.pid) which will gracefully terminate all clients. Sending the TERM signal twice will cause a forceful shutdown, which might make someone with a slow internet connection who's downloading big files unhappy.
ProgramUniprocess([bool]) → bool
Same as the -u flag if called from .init.lua. Can be used to configure the uniprocess mode. The current value is returned.
Slurp(filename:str[, i:int[, j:int]])
├─→ data:str
└─→ nil, unix.Errno
Reads all data from file the easy way.

This function reads file data from local file system. Zip file assets can be accessed using the `/zip/...` prefix.

i and j may be used to slice a substring in filename. These parameters are 1-indexed and behave consistently with Lua's string.sub() API. For example:

assert(Barf('x.txt', 'abc123'))
assert(assert(Slurp('x.txt', 2, 3)) == 'bc')

This function is uninterruptible so unix.EINTR errors will be ignored. This should only be a concern if you've installed signal handlers. Use the UNIX API if you need to react to it.

Barf(filename:str, data:str[, mode:int[, flags:int[, offset:int]]])
├─→ true
└─→ nil, unix.Errno
Writes all data to file the easy way.

This function writes to the local file system.

mode defaults to 0644. This parameter is ignored when flags doesn't have unix.O_CREAT.

flags defaults to unix.O_TRUNC | unix.O_CREAT.

offset is 1-indexed and may be used to overwrite arbitrary slices within a file when used in conjunction with flags=0. For example:

assert(Barf('x.txt', 'abc123'))
assert(Barf('x.txt', 'XX', 0, 0, 3))
assert(assert(Slurp('x.txt', 1, 6)) == 'abXX23')
ProgramContentType(ext:str[, contenttype:str]) → str
Sets or returns content type associated with a file extension.
ProgramHeader(name:str, value:str)
Appends HTTP header to the header buffer for all responses (whereas SetHeader only appends a header to the current response buffer). name is case-insensitive and restricted to non-space ASCII. value is a UTF-8 string that must be encodable as ISO-8859-1. Leading and trailing whitespace is trimmed automatically. Overlong characters are canonicalized. C0 and C1 control codes are forbidden, with the exception of tab. The header buffer is independent of the payload buffer. This function disallows the setting of certain headers such as Content-Range and Date, which are abstracted by the transport layer.
ProgramHeartbeatInterval([milliseconds:int])
Sets the heartbeat interval (in milliseconds). 5000ms is the default and 100ms is the minimum. If milliseconds is not specified, then the current interval is returned.
ProgramTimeout(milliseconds:int|seconds:int)
Default timeout is 60000ms. Minimal value of timeout is 10(ms). Negative values (<0) sets the keepalive in seconds. This function should only be called from /.init.lua.
ProgramSslTicketLifetime(seconds:int)
Defaults to 86400 (24 hours). This may be set to ≤0 to disable SSL tickets. It's a good idea to use these since it increases handshake performance 10x and eliminates a network round trip. This function is not available in unsecure mode.
ProgramBrand(str)
Changes HTTP Server header, as well as the <h1> title on the / listing page. The brand string needs to be a UTF-8 value that's encodable as ISO-8859-1. If the brand is changed to something other than redbean, then the promotional links will be removed from the listing page too. This function should only be called from /.init.lua.
ProgramCache(seconds:int[, directive:string])
Configures Cache-Control and Expires header generation for static asset serving. A negative value will disable the headers. Zero means don't cache. Greater than zero asks public proxies and browsers to cache for a given number of seconds. The directive value is added to the Cache-Control header when specified (with "must-revalidate" provided by default) and can be set to an empty string to remove the default value. This function should only be called from /.init.lua.
ProgramPort(uint16)
Hard-codes the port number on which to listen, which can be any number in the range 1..65535, or alternatively 0 to ask the operating system to choose a port, which may be revealed later on by GetServerAddr or the -z flag to stdout.
ProgramMaxPayloadSize(int)
Sets the maximum HTTP message payload size in bytes. The default is very conservatively set to 65536 so this is something many people will want to increase. This limit is enforced at the transport layer, before any Lua code is called, because right now redbean stores and forwards messages. (Use the UNIX API for raw socket streaming.) Setting this to a very high value can be useful if you're less concerned about rogue clients and would rather have your Lua code be granted more control to bounce unreasonable messages. If a value less than 1450 is supplied, it'll automatically be increased to 1450, since that's the size of ethernet frames. This function can only be called from .init.lua.
ProgramRedirect(code:int, src:str, location:str)
Configures fallback routing for paths which would otherwise return 404 Not Found. If code is 0 then the path is rewritten internally as an accelerated redirect. If code is 301, 302, 307, or 308 then a redirect response will be sent to the client. This function should only be called from /.init.lua.
ProgramCertificate(pem:str)
This function is the same as the -C flag if called from .init.lua, e.g. ProgramCertificate(LoadAsset("/.sign.crt")) for zip loading or ProgramCertificate(Slurp("/etc/letsencrypt.lol/fullchain.pem")) for local file system only. This function is not available in unsecure mode.
ProgramMaxWorkers(int)
Limits the number of workers forked by redbean. If that number is reached, the server continues polling until the number of workers is reduced or the value is updated. Setting it to 0 removes the limit (this is the default).
ProgramPrivateKey(pem:str)
This function is the same as the -K flag if called from .init.lua, e.g. ProgramPrivateKey(LoadAsset("/.sign.key")) for zip loading or ProgramPrivateKey(Slurp("/etc/letsencrypt/privkey.pem")) for local file system only. This function is not available in unsecure mode.
ProgramSslPresharedKey(key:str, identity:str)
This function can be used to enable the PSK ciphersuites which simplify SSL and enhance its performance in controlled environments. key may contain 1..32 bytes of random binary data and identity is usually a short plaintext string. The first time this function is called, the preshared key will be added to both the client and the server SSL configs. If it's called multiple times, then the remaining keys will be added to the server, which is useful if you want to assign separate keys to each client, each of which needs a separate identity too. If this function is called multiple times with the same identity string, then the latter call will overwrite the prior. If a preshared key is supplied and no certificates or key-signing-keys are programmed, then redbean won't bother auto-generating any serving certificates and will instead use only PSK ciphersuites. This function is not available in unsecure mode.
ProgramSslFetchVerify(enabled:bool)
May be used to disable the verification of certificates for remote hosts when using the Fetch() API. This function is not available in unsecure mode.
ProgramSslClientVerify(enabled:bool)
Enables the verification of certificates supplied by the HTTP clients that connect to your redbean. This has the same effect as the -j flag. Tuning this option alone does not preclude the possibility of unsecured HTTP clients, which can be disabled using ProgramSslRequired(). This function can only be called from .init.lua. This function is not available in unsecure mode.
ProgramSslRequired(mandatory:bool)
Enables the blocking of HTTP so that all inbound clients and must use the TLS transport layer. This has the same effect as the -J flag. Fetch() is still allowed to make outbound HTTP requests. This function can only be called from .init.lua. This function is not available in unsecure mode.
ProgramSslCiphersuite(name:str)
This function may be called multiple times to specify the subset of available ciphersuites you want to use in both the HTTPS server and the Fetch() client. The default list, ordered by preference, is as follows:
ECDHE-ECDSA-AES256-GCM-SHA384
ECDHE-ECDSA-AES128-GCM-SHA256
ECDHE-ECDSA-CHACHA20-POLY1305-SHA256
ECDHE-PSK-AES256-GCM-SHA384
ECDHE-PSK-AES128-GCM-SHA256
ECDHE-PSK-CHACHA20-POLY1305-SHA256
ECDHE-RSA-AES256-GCM-SHA384
ECDHE-RSA-AES128-GCM-SHA256
ECDHE-RSA-CHACHA20-POLY1305-SHA256
DHE-RSA-AES256-GCM-SHA384
DHE-RSA-AES128-GCM-SHA256
DHE-RSA-CHACHA20-POLY1305-SHA256
ECDHE-ECDSA-AES128-CBC-SHA256
ECDHE-RSA-AES256-CBC-SHA384
ECDHE-RSA-AES128-CBC-SHA256
DHE-RSA-AES256-CBC-SHA256
DHE-RSA-AES128-CBC-SHA256
ECDHE-PSK-AES256-CBC-SHA384
ECDHE-PSK-AES128-CBC-SHA256
ECDHE-ECDSA-AES256-CBC-SHA
ECDHE-ECDSA-AES128-CBC-SHA
ECDHE-RSA-AES256-CBC-SHA
ECDHE-RSA-AES128-CBC-SHA
DHE-RSA-AES256-CBC-SHA
DHE-RSA-AES128-CBC-SHA
ECDHE-PSK-AES256-CBC-SHA
ECDHE-PSK-AES128-CBC-SHA
RSA-AES256-GCM-SHA384
RSA-AES128-GCM-SHA256
RSA-AES256-CBC-SHA256
RSA-AES128-CBC-SHA256
RSA-AES256-CBC-SHA
RSA-AES128-CBC-SHA
PSK-AES256-GCM-SHA384
PSK-AES128-GCM-SHA256
PSK-CHACHA20-POLY1305-SHA256
PSK-AES256-CBC-SHA384
PSK-AES128-CBC-SHA256
PSK-AES256-CBC-SHA
PSK-AES128-CBC-SHA
ECDHE-RSA-3DES-EDE-CBC-SHA
DHE-RSA-3DES-EDE-CBC-SHA
ECDHE-PSK-3DES-EDE-CBC-SHA
RSA-3DES-EDE-CBC-SHA
PSK-3DES-EDE-CBC-SHA

When redbean is run on an old (or low-power) CPU that doesn't have the AES-NI instruction set (Westmere c. 2010) then the default ciphersuite is tuned automatically to favor the ChaCha20 Poly1305 suites.

The names above are canonical to redbean. They were programmatically simplified from the official IANA names. This function will accept the IANA names too. In most cases it will accept the OpenSSL and GnuTLS naming convention as well.

This function is not available in unsecure mode.

Route([host:str,[path:str]])
Instructs redbean to follow the normal HTTP serving path. This function is useful when writing an OnHttpRequest handler, since that overrides the serving path entirely. So if the handler decides it doesn't want to do anything, it can simply call this function, to hand over control back to the redbean core. By default, the host and path arguments are supplied from the resolved GetUrl value. This handler always resolves, since it will generate a 404 Not Found response if redbean couldn't find an appropriate endpoint.
Sleep(seconds:number)
Sleeps the specified number of seconds (can be fractional). The smallest interval is a microsecond.
RouteHost([host:str,[path:str]]) → bool
This is the same as Route, except it only implements the subset of request routing needed for serving virtual-hosted assets, where redbean tries to prefix the path with the hostname when looking up a file. This function returns true if the request was resolved. If it was resolved, then your OnHttpRequest request handler can still set additional headers.
RoutePath([path:str]) → bool
This is the same as Route, except it only implements the subset of request routing needed for serving assets. This function returns true if the request was resolved. If it was resolved, then your OnHttpRequest request handler can still set additional headers.
ServeAsset(path:str)
Instructs redbean to serve static asset at path. This function causes what would normally happen outside a dynamic handler to happen. The asset can be sourced from either the zip or local filesystem if -D is used. This function is mutually exclusive with SetStatus and ServeError.
ServeError(code:int[, reason:str])
Instructs redbean to serve a boilerplate error page. This takes care of logging the error, setting the reason phrase, and adding a payload. This function is mutually exclusive with SetStatus and ServeAsset.
SetLogLevel(level:int)
Sets logger verbosity. Reasonable values for level are kLogDebug > kLogVerbose > kLogInfo > kLogWarn > kLogError > kLogFatal.
VisualizeControlCodes(str) → str
Replaces C0 control codes and trojan source characters with descriptive UNICODE pictorial representation. This function also canonicalizes overlong encodings. C1 control codes are replaced with a JavaScript-like escape sequence.
Underlong(str) → str
Canonicalizes overlong encodings.
Bsf(x:int) → int
Returns position of first bit set. Passing 0 will raise an error. Same as the Intel x86 instruction BSF.
Bsr(x:int) → int
Returns binary logarithm of 𝑥. Passing 0 will raise an error. Same as the Intel x86 instruction BSR.
Crc32(initial:int, data:str) → int
Computes Phil Katz CRC-32 used by zip/zlib/gzip/etc.
Crc32c(initial:int, data:str) → int
Computes 32-bit Castagnoli Cyclic Redundancy Check.
Popcnt(x:int) → int
Returns number of bits set in integer.
Rdtsc() → int
Returns CPU timestamp counter.
Lemur64() → int
Returns fastest pseudorandom non-cryptographic random number. This linear congruential generator passes practrand and bigcrush.
Rand64() → int
Returns nondeterministic pseudorandom non-cryptographic number. This linear congruential generator passes practrand and bigcrush. This generator is safe across fork(), threads, and signal handlers.
Rdrand() → int
Returns 64-bit hardware random integer from RDRND instruction, with automatic fallback to getrandom() if not available.
Rdseed() → int
Returns 64-bit hardware random integer from RDSEED instruction, with automatic fallback to RDRND and getrandom() if not available.
GetCpuCount() → int
Returns CPU core count or 0 if it couldn't be determined.
GetCpuCore() → int
Returns 0-indexed CPU core on which process is currently scheduled.
GetCpuNode() → int
Returns 0-indexed NUMA node on which process is currently scheduled.
Decimate(str) → str
Shrinks byte buffer in half using John Costella's magic kernel. This downscales data 2x using an eight-tap convolution, e.g.
>: Decimate('\xff\xff\x00\x00\xff\xff\x00\x00\xff\xff\x00\x00')
"\xff\x00\xff\x00\xff\x00"

This is very fast if SSSE3 is available (Intel 2004+ / AMD 2011+).

MeasureEntropy(data) → float
Returns Shannon entropy of array. This gives you an idea of the density of information. Cryptographic random should be in the ballpark of 7.9 whereas plaintext will be more like 4.5.
Deflate(uncompressed:str[, level:int])
├─→ compressed:str
└─→ nil, error:str
Compresses data.
>: Deflate("hello")
"\xcbH\xcd\xc9\xc9\x07\x00"
>: Inflate("\xcbH\xcd\xc9\xc9\x07\x00", 5)
"hello"

The output format is raw DEFLATE that's suitable for embedding into formats like a ZIP file. It's recommended that, like ZIP, you also store separately a Crc32() checksum in addition to the original uncompressed size.

level is the compression level, which defaults to 7. The max is 9. Lower numbers go faster (4 for instance is a sweet spot) and higher numbers go slower but have better compression.

Inflate(compressed:str, maxoutsize:int)
├─→ uncompressed:str
└─→ nil, error:str
Decompresses data.

This function performs the inverse of Deflate(). It's recommended that you perform a Crc32() check on the output string after this function succeeds.

maxoutsize is the uncompressed size, which should be known. However, it is permissable (although not advised) to specify some large number in which case (on success) the byte length of the output string may be less than maxoutsize.

Benchmark(func[, count[, maxattempts]])
└─→ nanos:real, ticks:int, overhead-ticks:int, tries:int
Performs microbenchmark.

The first value returned is the average number of nanoseconds that func needed to execute. Nanoseconds are computed from RDTSC tick counts, using an approximation that's measured beforehand with the unix.clock_gettime() function.

The ticks result is the canonical average number of clock ticks.

This subroutine will subtract whatever the overhead happens to be for benchmarking a function that does nothing. This overhead value will be reported in the result.

tries indicates if your microbenchmark needed to be repeated, possibly because your system is under load and the benchmark was preempted by the operating system, or moved to a different core.

oct(int) → str
Formats string as octal integer literal string. If the provided value is zero, the result will be "0". Otherwise the resulting value will be the zero-prefixed octal string. The result is currently modulo 2^64. Negative numbers are converted to unsigned.
hex(int) → str
Formats string as hexadecimal integer literal string. If the provided value is zero, the result will be "0". Otherwise the resulting value will be the "0x"-prefixed hex string. The result is currently modulo 2^64. Negative numbers are converted to unsigned.
bin(int) → str
Formats string as binary integer literal string. If the provided value is zero, the result will be "0". Otherwise the resulting value will be the "0b"-prefixed binary str. The result is currently modulo 2^64. Negative numbers are converted to unsigned.
ResolveIp(hostname:str)
├─→ ip:uint32
└─→ nil, error:str
Gets IP address associated with hostname.

This function first checks if hostname is already an IP address, in which case it returns the result of ParseIp. Otherwise, it checks HOSTS.TXT on the local system and returns the first IPv4 address associated with hostname. If no such entry is found, a DNS lookup is performed using the system configured (e.g. /etc/resolv.conf) DNS resolution service. If the service returns multiple IN A records then only the first one is returned.

The returned address is word-encoded in host endian order. For example, 1.2.3.4 is encoded as 0x01020304. The FormatIp function may be used to turn this value back into a string.

If no IP address could be found, then nil is returned alongside a string of unspecified format describing the error. Calls to this function may be wrapped in assert() if an exception is desired.

IsTrustedIp(ip:int)
└─→ bool
Returns true if IP address is trustworthy.

If the ProgramTrustedIp() function has NOT been called then redbean will consider the networks 127.0.0.0/8, 10.0.0.0/8, 172.16.0.0/12, and 192.168.0.0/16 to be trustworthy too. If ProgramTrustedIp() HAS been called at some point earlier in your redbean's lifecycle, then it'll trust the IPs and network subnets you specify instead.

The network interface addresses used by the host machine are always considered trustworthy, e.g. 127.0.0.1. This may change soon, if we decide to export a GetHostIps() API which queries your NIC devices.

ProgramTrustedIp(ip:int[, cidr:int])
Trusts an IP address or network.

This function may be used to configure the IsTrustedIp() function which is how redbean determines if a client is allowed to send us headers like X-Forwarded-For (cf GetRemoteAddr vs. GetClientAddr) without them being ignored. Trusted IPs is also how redbean turns off token bucket rate limiting selectively, so be careful. Here's an example of how you could trust all of Cloudflare's IPs:

ProgramTrustedIp(ParseIp("103.21.244.0"), 22);
ProgramTrustedIp(ParseIp("103.22.200.0"), 22);
ProgramTrustedIp(ParseIp("103.31.4.0"), 22);
ProgramTrustedIp(ParseIp("104.16.0.0"), 13);
ProgramTrustedIp(ParseIp("104.24.0.0"), 14);
ProgramTrustedIp(ParseIp("108.162.192.0"), 18);
ProgramTrustedIp(ParseIp("131.0.72.0"), 22);
ProgramTrustedIp(ParseIp("141.101.64.0"), 18);
ProgramTrustedIp(ParseIp("162.158.0.0"), 15);
ProgramTrustedIp(ParseIp("172.64.0.0"), 13);
ProgramTrustedIp(ParseIp("173.245.48.0"), 20);
ProgramTrustedIp(ParseIp("188.114.96.0"), 20);
ProgramTrustedIp(ParseIp("190.93.240.0"), 20);
ProgramTrustedIp(ParseIp("197.234.240.0"), 22);
ProgramTrustedIp(ParseIp("198.41.128.0"), 17);

Although you might want consider trusting redbean's open source freedom embracing solution to DDOS protection instead!

ProgramTokenBucket([replenish:num[, cidr:int[, reject:int[, ignore:int[, ban:int]]]]])
Enables DDOS protection.

Imagine you have 232 buckets, one for each IP address. Each bucket can hold about 127 tokens. Every second a background worker puts one token in each bucket. When a TCP client socket is opened, it takes a token from its bucket, and then proceeds. If the bucket holds only a third of its original tokens, then redbean sends them a 429 warning. If the client ignores this warning and keeps sending requests, until there's no tokens left, then the banhammer finally comes down.

function OnServerStart()
    ProgramTokenBucket()
    ProgramTrustedIp(ParseIp('x.x.x.x'), 32)
    assert(unix.setrlimit(unix.RLIMIT_NPROC, 1000, 1000))
end

This model of network rate limiting generously lets people "burst" a tiny bit. For example someone might get a strong craving for content and smash the reload button in Chrome 64 times in a few seconds. But since the client only get 1 new token per second, they'd better cool their heels for a few minutes after doing that. This amount of burst can be altered by choosing the reject / ignore / ban threshold arguments. For example, if the reject parameter is set to 126 then no bursting is allowed, which probably isn't a good idea.

redbean is programmed to acquire a token immediately after accept() is called from the main server process, which is well before fork() or read() or any Lua code happens. redbean then takes action, based on the token count, which can be accept / reject / ignore / ban. If redbean determines a ban is warrented, then 4-byte datagram is sent to the unix domain socket /var/run/blackhole.sock which should be operated using the blackholed program we distribute separately.

The trick redbean uses on Linux for example is insert rules in your raw prerouting table. redbean is very fast at the application layer so the biggest issue we've encountered in production is are kernels themselves, and programming the raw prerouting table dynamically is how we solved that.

replenish is the number of times per second a token should be added to each bucket. The default value is 1 which means one token is granted per second to all buckets. The minimum value is 1/3600 which means once per hour. The maximum value for this setting is 1e6, which means once every microsecond.

cidr is the specificity of judgement. Since creating 2^32 buckets would need 4GB of RAM, redbean defaults this value to 24 which means filtering applies to class c network blocks (i.e. x.x.x.*), and your token buckets only take up 2^24 bytes of RAM (16MB). This can be set to any number on the inclusive interval [8,32], where having a lower number means you use less ram/cpu, but splash damage applies more to your clients; whereas higher numbers means more ram/cpu usage, while ensuring rate limiting only applies to specific compromised actors.

reject is the token count or treshold at which redbean should send 429 Too Many Request warnings to the client. Permitted values can be anywhere between -1 and 126 inclusively. The default value is 30 and -1 means to disable (assuming AcquireToken() will be used).

ignore is the token count or treshold, at which redbean should try simply ignoring clients and close the connection without logging any kind of warning, and without sending any response. The default value for this setting is MIN(reject / 2, 15). This must be less than or equal to the reject setting. Allowed values are [-1,126] where you can use -1 as a means of disabling ignore.

ban is the token count at which redbean should report IP addresses to the blackhole daemon via a unix-domain socket datagram so they'll get banned in the kernel routing tables. redbean's default value for this setting is MIN(ignore / 10, 1). Permitted values are [-1,126] where -1 may be used as a means of disabling the ban feature.

This function throws an exception if the constraints described above are not the case. Warnings are logged should redbean fail to connect to the blackhole daemon, assuming it hasn't been disabled. It's safe to use load balancing tools when banning is enabled, since you can't accidentally ban your own network interface addresses, loopback ips, or ProgramTrustedIp() addresses where these rate limits don't apply.

It's assumed will be called from the .init.lua global scope although it could be used in interpreter mode, or from a forked child process in which case the only processes that'll have ability to use it will be that same process, and any descendent processes. This function is only able to be called once.

This feature is not available in unsecure mode.

AcquireToken([ip:uint32])
└─→ int8
Atomically acquires token.

This routine atomically acquires a single token for an ip address. The return value is the token count before the subtraction happened. No action is taken based on the count, since the caller will decide.

ip should be an IPv4 address and this defaults to GetClientAddr(), although other interpretations of its meaning are possible.

Your token buckets are stored in shared memory so this can be called from multiple forked processes. which operate on the same values.

Blackhole([ip:uint32])
└─→ bool
Sends IP address to blackholed service.

ProgramTokenBucket() needs to be called beforehand. The default settings will blackhole automatically, during the accept() loop based on the banned threshold. However if your Lua code calls AcquireToken() manually, then you'll need this function to take action on the returned values.

This function returns true if a datagram could be sent sucessfully. Otherwise false is returned, which can happen if blackholed isn't running, or if a lot of processes are sending messages to it and the operation would have blocked.

It's assumed that the blackholed service is running locally in the background.

lsqlite3 module

Please refer to the LuaSQLite3 Documentation.

For example, you could put the following in your /.init.lua file:

sqlite3 = require 'lsqlite3'
db = sqlite3.open_memory()
db:exec[[
  CREATE TABLE test (
    id INTEGER PRIMARY KEY,
    content TEXT
  );
  INSERT INTO test (content) VALUES ('Hello World');
  INSERT INTO test (content) VALUES ('Hello Lua');
  INSERT INTO test (content) VALUES ('Hello Sqlite3');
]]

Then, your Lua server pages or OnHttpRequest handler may perform SQL queries by accessing the db global. The performance is good too, at about 400k qps.

for row in db:nrows("SELECT * FROM test") do
   Write(row.id.." "..row.content.."<br>")
end

Warning: Please read How To Corrupt An SQLite Database File because those same rules apply to redbean. One recommendation worth paying attention to is that SQLite connections can't cross fork() boundaries, and redbean is a forking web server. The simplest solution to solving this, is to create your database handles inside your OnWorkerStart() hook.

The above example is only appropriate for read-only databases. When setting up a mutable SQLite database, it's important to consider that redbean is a forking web server, and SQLite handles shouldn't cross fork boundaries. That means you need to create your SQLite database object lazily from within the client process, and SQLite makes this very fast. One approach that's worked well for us on Linux, is to do that using the write-ahead log. For example:

re = require 're'
sqlite3 = require 'lsqlite3'
reNumberPath = re.compile[[^/([0-9][0-9]*)$]]
function SetupSql()
   if not db then
      db = sqlite3.open('redbean.sqlite3')
      db:busy_timeout(1000)
      db:exec[[PRAGMA journal_mode=WAL]]
      db:exec[[PRAGMA synchronous=NORMAL]]
      getBarStmt = db:prepare[[
         SELECT
            foo
         FROM
            Bar
         WHERE
            id = ?
      ]]
   end
end
local function GetBar(id)
   if not getBarStmt then
      Log(kLogWarn, 'prepare failed: ' .. db:errmsg())
      return nil
   end
   getBarStmt:reset()
   getBarStmt:bind(1, id)
   for bar in getBarStmt:nrows() do
      return bar
   end
   return nil
end
function OnHttpRequest()
   SetupSql()
   _, id = reNumberPath:search(GetPath())
   if id then
      bar = GetBar(id)
      SetHeader('Content-Type', 'text/plain; charset=utf-8')
      Write(string(bar.foo))
      return
   end
   Route()
   SetHeader('Content-Language', 'en-US')
end

redbean supports a subset of what's defined in the upstream LuaSQLite3 project. Most of the unsupported APIs relate to pointers and database notification hooks.

re module

This module exposes an API for POSIX regular expressions which enable you to validate input, search for substrings, extract pieces of strings, etc. Here's a usage example:

# Example IPv4 Address Regular Expression (see also ParseIP)
p = assert(re.compile([[^([0-9]{1,3})\.([0-9]{1,3})\.([0-9]{1,3})\.([0-9]{1,3})$]]))
m,a,b,c,d = assert(p:search(𝑠))
if m then
  print("ok", tonumber(a), tonumber(b), tonumber(c), tonumber(d))
else
  print("not ok")
end
re.search(regex:str, text:str[, flags:int])
├─→ match:str[, group1:str, ...]
└─→ nil, re.Errno
Searches for regular expression match in text.

This is a shorthand notation roughly equivalent to:

preg = re.compile(regex)
patt = preg:search(re, text)

flags defaults to zero and may have any of:

This has exponential complexity. Please use re.compile() to compile your regular expressions once from /.init.lua. This API exists for convenience. This isn't recommended for prod.

This uses POSIX extended syntax by default.

re.compile(regex:str[, flags:int])
├─→ preg:re.Regex
└─→ nil, re.Errno
Compiles regular expression.

flags defaults to zero and may have any of:

This has an O(2^𝑛) cost. Consider compiling regular expressions once from your /.init.lua file.

If regex is an untrusted user value, then unix.setrlimit should be used to impose cpu and memory quotas for security.

This uses POSIX extended syntax by default.

re.Regex Object

Executes precompiled regular expression.

Returns nothing (nil) if the pattern doesn't match anything. Otherwise it pushes the matched substring and any parenthesis-captured values too. Flags may contain re.NOTBOL or re.NOTEOL to indicate whether or not text should be considered at the start and/or end of a line.

`flags` defaults to zero and may have any of:

This has an O(𝑛) cost.

re.Errno Object

re.Errno:errno()
└─→ errno:int
Returns regex error number.
re.Errno:doc()
└─→ description:str
Returns English string describing error code.
re.Errno:__tostring()
└─→ str
Delegates to re.Errno:doc()

re flags

re.BASIC
Use this flag if you prefer the default POSIX regex syntax. We use extended regex notation by default. For example, an extended regular expression for matching an IP address might look like ([0-9]*)\.([0-9]*)\.([0-9]*)\.([0-9]*) whereas with basic syntax it would look like \([0-9]*\)\.\([0-9]*\)\.\([0-9]*\)\.\([0-9]*\). This flag may only be used with re.compile and re.search.
re.ICASE
Use this flag to make your pattern case ASCII case-insensitive. This means [a-z] will mean the same thing as [A-Za-z]. This flag may only be used with re.compile and re.search.
re.NEWLINE
Use this flag to change the handling of NEWLINE (\x0a) characters. When this flag is set, (1) a NEWLINE shall not be matched by a "." or any form of a non-matching list, (2) a "^" shall match the zero-length string immediately after a NEWLINE (regardless of re.NOTBOL), and (3) a "$" shall match the zero-length string immediately before a NEWLINE (regardless of re.NOTEOL).
re.NOSUB
Causes re.search to only report success and failure. This is reported via the API by returning empty string for success. This flag may only be used with re.compile and re.search.
re.NOTBOL
The first character of the string pointed to by string is not the beginning of the line. This flag may only be used with re.search and regex_t*:search.
re.NOTEOL
The last character of the string pointed to by string is not the end of the line. This flag may only be used with re.search and regex_t*:search.

re errors

re.NOMATCH
No match
re.BADPAT
Invalid regex
re.ECOLLATE
Unknown collating element
re.ECTYPE
Unknown character class name
re.EESCAPE
Trailing backslash
re.ESUBREG
Invalid back reference
re.EBRACK
Missing ]
re.EPAREN
Missing )
re.EBRACE
Missing }
re.BADBR
Invalid contents of {}
re.ERANGE
Invalid character range.
re.ESPACE
Out of memory
re.BADRPT
Repetition not preceded by valid expression

path module

The path module may be used to manipulate unix paths.

Note that we use unix paths on Windows. For example, if you have a path like C:\foo\bar then it should be /c/foo/bar with redbean. It should also be noted the unix module is more permissive when using Windows paths, where translation to win32 is very light.

path.dirname(str)
└─→ str
Strips final component of path, e.g.
path      │ dirname
───────────────────
.         │ .
..        │ .
/         │ /
usr       │ .
/usr/     │ /
/usr/lib  │ /usr
/usr/lib/ │ /usr
path.basename(str)
└─→ str
Returns final component of path, e.g.
path      │ basename
─────────────────────
.         │ .
..        │ ..
/         │ /
usr       │ usr
/usr/     │ usr
/usr/lib  │ lib
/usr/lib/ │ lib
path.join(str)
└─→ str
Concatenates path components, e.g.
x         │ y        │ joined
─────────────────────────────────
/         │ /        │ /
/usr      │ lib      │ /usr/lib
/usr/     │ lib      │ /usr/lib
/usr/lib  │ /lib     │ /lib

You may specify 1+ arguments.

Specifying no arguments will raise an error. If nil arguments are specified, then they're skipped over. If exclusively nil arguments are passed, then nil is returned. Empty strings behave similarly to nil, but unlike nil may coerce a trailing slash.

path.exists(path:str)
└─→ bool
Returns true if path exists.

This function is inclusive of regular files, directories, and special files. Symbolic links are followed are resolved. On error, false is returned.

path.isfile(path:str)
└─→ bool
Returns true if path exists and is regular file.

Symbolic links are not followed. On error, false is returned.

path.isdir(path:str)
└─→ bool
Returns true if path exists and is directory.

Symbolic links are not followed. On error, false is returned.

Returns true if path exists and is symbolic link.

Symbolic links are not followed. On error, false is returned.

maxmind module

This module may be used to get city/country/asn/etc from IPs, e.g.

-- .init.lua
maxmind = require 'maxmind'
asndb = maxmind.open('/usr/local/share/maxmind/GeoLite2-ASN.mmdb')

-- request handler
as = asndb:lookup(GetRemoteAddr())
if as then
    asnum = as:get('autonomous_system_number')
    asorg = as:get('autonomous_system_organization')
    Write(EscapeHtml(asnum))
    Write(' ')
    Write(EscapeHtml(asorg))
end

The database file is distributed by MaxMind. You need to sign up on their website to get a free copy. The database has a generalized structure. For a concrete example of how this module may be used, please see maxmind.lua in redbean-demo.com.

finger module

This is an experimental module that, like the maxmind module, gives you insight into what kind of device is connecting to your redbean. This module can help you protect your redbean because it provides tools for identifying clients that misrepresent themselves. For example the User-Agent header might report itself as a Windows computer when the SYN packet says it's a Linux computer.

function OnServerListen(fd, ip, port)
   unix.setsockopt(fd, unix.SOL_TCP, unix.TCP_SAVE_SYN, true)
   return false
end

function OnClientConnection(ip, port, serverip, serverport)
   fd = GetClientFd()
   syn = unix.getsockopt(fd, unix.SOL_TCP, unix.TCP_SAVED_SYN)
end

function OnHttpRequest()
   Log(kLogInfo, "client is running %s and reports %s" % {
          finger.GetSynFingerOs(finger.FingerSyn(syn)),
          GetHeader('User-Agent')})
   Route()
end

The following functions are provided.

finger.FingerSyn(syn_packet_bytes:str)
├─→ synfinger:uint32
└─→ nil, error:str
Fingerprints IP+TCP SYN packet.

This returns a hash-like magic number that reflects the SYN packet structure, e.g. ordering of options, maximum segment size, etc. We make no guarantees this hashing algorithm won't change as we learn more about the optimal way to fingerprint, so be sure to save your syn packets too if you're using this feature, in case they need to be rehashed in the future.

This function is nil/error propagating.

finger.GetSynFingerOs(synfinger:uint32)
├─→ osname:str
└─→ nil, error:str
Fingerprints IP+TCP SYN packet.

If synfinger is a known hard-coded magic number, then one of the following strings may be returned:

If this function returns nil, then one thing you can do to help is file an issue and share with us your SYN packet specimens. The way we prefer to receive them is in EncodeLua(syn_packet_bytes) format along with details on the operating system which you must know.

finger.DescribeSyn(syn_packet_bytes:str)
├─→ description:str
└─→ nil, error:str
Describes IP+TCP SYN packet.

The layout looks as follows:

TTL:OPTIONS:WSIZE:MSS

The TTL, WSIZE, and MSS fields are unsigned decimal fields.

The OPTIONS field communicates the ordering of the commonly used subset of tcp options. The following character mappings are defined. TCP options not on this list will be ignored.

This function is nil/error propagating.

argon2 module

This module implements a password hashing algorithm based on blake2b that won the Password Hashing Competition.

It can be used to securely store user passwords in your SQLite database, in a way that destroys the password, but can be verified by regenerating the hash again the next time the user logs in. Destroying the password is important, since if your database is compromised, the bad guys won't be able to use rainbow tables to recover the plain text of the passwords.

Argon2 achieves this security by being expensive to compute. Care should be taken in choosing parameters, since an HTTP endpoint that uses Argon2 can just as easily become a denial of service vector. For example, you may want to consider throttling your login endpoint.

argon2.hash_encoded(pass:str, salt:str[, config:table])
├─→ ascii:str
└─→ nil, error:str
Hashes password.

This is consistent with the README of the reference implementation:

>: assert(argon2.hash_encoded("password", "somesalt", {
       variant = argon2.variants.argon2_i,
       m_cost = 65536,
       hash_len = 24,
       parallelism = 4,
       t_cost = 2,
     }))
"$argon2i$v=19$m=65536,t=2,p=4$c29tZXNhbHQ$RdescudvJCsgt3ub+b+dWRWJTmaaJObG"

pass is the secret value to be encoded.

salt is a nonce value used to hash the string.

config.m_cost is the memory hardness in kibibytes, which defaults to 4096 (4 mibibytes). It's recommended that this be tuned upwards.

config.t_cost is the number of iterations, which defaults to 3.

config.parallelism is the parallelism factor, which defaults to 1.

config.hash_len is the number of desired bytes in hash output, which defaults to 32.

config.variant may be:

argon2.verify(encoded:str, pass:str)
├─→ ok:bool
└─→ nil, error:str
Verifies password, e.g.
>: argon2.verify("$argon2i$v=19$m=65536,t=2,p=4$c29tZXNhbHQ$RdescudvJCsgt3ub+b+dWRWJTmaaJObG", "password")
true

unix module

This module exposes the low-level System Five system call interface. This module works on all supported platforms, including Windows NT.

unix.open(path:str[, flags:int[, mode:int[, dirfd:int]]])
├─→ fd:int
└─→ nil, unix.Errno
Opens file.

Returns a file descriptor integer that needs to be closed, e.g.

fd = assert(unix.open('/etc/passwd', unix.O_RDONLY))
print(unix.read(fd))
unix.close(fd)

flags should have one of:

The following values may also be OR'd into flags:

There are three regular combinations for the above flags:

dirfd defaults to to unix.AT_FDCWD and may optionally be set to a directory file descriptor to which path is relative.

Returns ENOENT if path doesn't exist.

Returns ENOTDIR if path contained a directory component that wasn't a directory.

unix.close(fd:int)
├─→ true
└─→ nil, unix.Errno
Closes file descriptor.

This function should never be called twice for the same file descriptor, regardless of whether or not an error happened. The file descriptor is always gone after close is called. So it technically always succeeds, but that doesn't mean an error should be ignored. For example, on NFS a close failure could indicate data loss.

Closing does not mean that scheduled i/o operations have been completed. You'd need to use fsync() or fdatasync() beforehand to ensure that. You shouldn't need to do that normally, because our close implementation guarantees a consistent view, since on systems where it isn't guaranteed (like Windows) close will implicitly sync.

File descriptors are automatically closed on exit().

Returns EBADF if fd wasn't valid.

Returns EINTR possibly maybe.

Returns EIO if an i/o error occurred.

unix.read(fd:int[, bufsiz:str[, offset:int]])
├─→ data:str
└─→ nil, unix.Errno
Reads from file descriptor.

This function returns empty string on end of file. The exception is if bufsiz is zero, in which case an empty returned string means the file descriptor works.

unix.write(fd:int, data:str[, offset:int])
├─→ wrotebytes:int
└─→ nil, unix.Errno
Writes to file descriptor.
unix.exit([exitcode:int])
└─→ ⊥
Invokes _Exit(exitcode) on the process. This will immediately halt the current process. Memory will be freed. File descriptors will be closed. Any open connections it owns will be reset. This function never returns.
unix.environ()
└─→ {str,...}
Returns raw environment variables.

This allocates and constructs the C/C++ environ variable as a Lua table consisting of string keys and string values.

This data structure preserves casing. On Windows NT, by convention, environment variable keys are treated in a case-insensitive way. It is the responsibility of the caller to consider this.

This data structure preserves valueless variables. It's possible on both UNIX and Windows to have an environment variable without an equals, even though it's unusual.

This data structure preserves duplicates. For example, on Windows, there's some irregular uses of environment variables such as how the command prompt inserts multiple environment variables with empty string as keys, for its internal bookkeeping.

unix.fork()
├─┬─→ 0
│ └─→ childpid:int
└─→ nil, unix.Errno
Creates a new process mitosis style.

This system call returns twice. The parent process gets the nonzero pid. The child gets zero.

Here's a simple usage example of creating subprocesses, where we fork off a child worker from a main process hook callback to do some independent chores, such as sending an HTTP request back to redbean.

-- as soon as server starts, make a fetch to the server
-- then signal redbean to shutdown when fetch is complete
local onServerStart = function()
   if assert(unix.fork()) == 0 then
      local ok, headers, body = Fetch('http://127.0.0.1:8080/test')
      unix.kill(unix.getppid(), unix.SIGTERM)
      unix.exit(0)
   end
end
OnServerStart = onServerStart

We didn't need to use wait() here, because (a) we want redbean to go back to what it was doing before as the Fetch() completes, and (b) redbean's main process already has a zombie collector. However it's a moot point, since once the fetch is done, the child process then asks redbean to gracefully shutdown by sending SIGTERM its parent.

This is actually a situation where we *must* use fork, because the purpose of the main redbean process is to call accept() and create workers. So if we programmed redbean to use the main process to send a blocking request to itself instead, then redbean would deadlock and never be able to accept() the client.

While deadlocking is an extreme example, the truth is that latency issues can crop up for the same reason that just cause jitter instead, and as such, can easily go unnoticed. For example, if you do something that takes longer than a few milliseconds from inside your redbean heartbeat, then that's a few milliseconds in which redbean is no longer concurrent, and tail latency is being added to its ability to accept new connections. fork() does a great job at solving this.

If you're not sure how long something will take, then when in doubt, fork off a process. You can then report its completion to something like SQLite. Redbean makes having lots of processes cheap. On Linux they're about as lightweight as what heavyweight environments call greenlets. You can easily have 10,000 Redbean workers on one PC.

Here's some benchmarks for fork() performance across platforms:

                         cycles      nanos  environ
Linux 5.4 fork           97,200     31,395  [metal]
FreeBSD 12 fork         236,089     78,841  [vmware]
Darwin 20.6 fork        295,325     81,738  [metal]
NetBSD 9 fork         5,832,027  1,947,899  [vmware]
OpenBSD 6.8 fork     13,241,940  4,422,103  [vmware]
Windows10 fork       18,802,239  6,360,271  [metal]

One of the benefits of using fork() is it creates an isolation barrier between the different parts of your app. This can lead to enhanced reliability and security. For example, redbean uses fork so it can wipe your ssl keys from memory before handing over control to request handlers that process untrusted input. It also ensures that if your Lua app crashes, it won't take down the server as a whole. Hence it should come as no surprise that fork() would go slower on operating systems that have more security features. So depending on your use case, you can choose the operating system that suits you.

unix.commandv(prog:str)
├─→ path:str
└─→ nil, unix.Errno
Performs $PATH lookup of executable.
unix = require 'unix'
prog = assert(unix.commandv('ls'))
unix.execve(prog, {prog, '-hal', '.'}, {'PATH=/bin'})
unix.exit(127)

We automatically suffix .com and .exe for all platforms when path searching. By default, the current directory is not on the path. If prog is an absolute path, then it's returned as-is. If prog contains slashes then it's not path searched either and will be returned if it exists.

unix.execve(prog:str[, args:List<*>, env:List<*>])
└─→ nil, unix.Errno
Exits current process, replacing it with a new instance of the specified program. prog needs to be an absolute path, see commandv(). env defaults to to the current environ(). Here's a basic usage example:
unix.execve('/bin/ls', {'/bin/ls', '-hal'}, {'PATH=/bin'})
unix.exit(127)

prog needs to be the resolved pathname of your executable. You can use commandv() to search your PATH.

args is a string list table. The first element in args should be prog. Values are coerced to strings. This parameter defaults to {prog}.

env is a string list table. Values are coerced to strings. No ordering requirement is imposed. By convention, each string has its key and value separated by an equals sign without spaces. If this parameter is not specified, it'll default to the C/C++ environ variable which is inherited from the shell that launched redbean. It's the responsibility of the user to supply a sanitized environ when spawning untrusted processes.

execve() is normally called after fork() returns 0. If that isn't the case, then your redbean worker will be destroyed.

This function never returns on success.

unix.EAGAIN is returned if you've enforced a max number of processes using setrlimit(unix.RLIMIT_NPROC).

unix.dup(oldfd:int[, newfd:int[, flags:int[, lowest:int]]])
├─→ newfd:int
└─→ nil, unix.Errno
Duplicates file descriptor.

newfd may be specified to choose a specific number for the new file descriptor. If it's already open, then the preexisting one will be silently closed. EINVAL is returned if newfd equals oldfd.

flags can have unix.O_CLOEXEC which means the returned file descriptors will be automatically closed upon execve().

lowest defaults to zero and defines the lowest numbered file descriptor that's acceptable to use. If newfd is specified then lowest is ignored. For example, if you wanted to duplicate standard input, then:

stdin2 = assert(unix.dup(0, nil, unix.O_CLOEXEC, 3))

Will ensure that, in the rare event standard output or standard error are closed, you won't accidentally duplicate standard input to those numbers.

unix.pipe([flags:int])
├─→ reader:int, writer:int
└─→ nil, unix.Errno
Creates fifo which enables communication between processes.

flags can have any of

Returns two file descriptors: one for reading and one for writing.

Here's an example of how pipe(), fork(), dup(), etc. may be used to serve an HTTP response containing the output of a subprocess.

local unix = require 'unix'
ls = assert(unix.commandv('ls'))
reader, writer = assert(unix.pipe())
if assert(unix.fork()) == 0 then
   unix.close(1)
   unix.dup(writer)
   unix.close(writer)
   unix.close(reader)
   unix.execve(ls, {ls, '-Shal'})
   unix.exit(127)
else
   unix.close(writer)
   SetHeader('Content-Type', 'text/plain')
   while true do
      data, err = unix.read(reader)
      if data then
         if data ~= '' then
            Write(data)
         else
            break
         end
      elseif err:errno() ~= EINTR then
         Log(kLogWarn, tostring(err))
         break
      end
   end
   assert(unix.close(reader))
   assert(unix.wait())
end
unix.wait([pid:int[, options:int]])
├─→ pid:int, wstatus:int, unix.Rusage
└─→ nil, unix.Errno
Waits for subprocess to terminate.

pid defaults to -1 which means any child process. Setting pid to 0 is equivalent to -getpid(). If pid < -1 then that means wait for any pid in the process group -pid. Then lastly if pid > 0 then this waits for a specific process id

Options may have WNOHANG which means don't block, check for the existence of processes that are already dead (technically speaking zombies) and if so harvest them immediately.

Returns the process id of the child that terminated. In other cases, the returned pid is nil and errno is non-nil.

The returned wstatus contains information about the process exit status. It's a complicated integer and there's functions that can help interpret it. For example:

-- wait for zombies
-- traditional technique for SIGCHLD handlers
while true do
   pid, status = unix.wait(-1, unix.WNOHANG)
   if pid then
      if unix.WIFEXITED(status) then
         print('child', pid, 'exited with',
               unix.WEXITSTATUS(status))
      elseif unix.WIFSIGNALED(status) then
         print('child', pid, 'crashed with',
               unix.strsignal(unix.WTERMSIG(status)))
      end
   elseif status:errno() == unix.ECHILD then
      Log(kLogDebug, 'no more zombies')
      break
   else
      Log(kLogWarn, tostring(status))
      break
   end
end
unix.WIFEXITED(wstatus:int)
└─→ bool
Returns true if process exited cleanly.
unix.WEXITSTATUS(wstatus:int)
└─→ exitcode:uint8
Returns code passed to exit() assuming WIFEXITED(wstatus) is true.
unix.WIFSIGNALED(wstatus:int)
└─→ bool
Returns true if process terminated due to a signal.
unix.WTERMSIG(wstatus:int)
└─→ sig:uint8
Returns signal that caused process to terminate assuming WIFSIGNALED(wstatus) is true.
unix.getpid()
└─→ pid:int
Returns process id of current process.

This function does not fail.

unix.getppid()
└─→ pid:int
Returns process id of parent process.

This function does not fail.

unix.kill(pid:int, sig:int)
├─→ true
└─→ nil, unix.Errno
Sends signal to process(es).

The impact of this action can be terminating the process, or interrupting it to request something happen.

pid can be:

sig can be:

Windows NT only supports the kill() signals required by the ANSI C89 standard, which are SIGINT and SIGQUIT. All other signals on the Windows platform that are sent to another process via kill() will be treated like SIGKILL.

unix.raise(sig:int)
├─→ rc:int
└─→ nil, unix.Errno
Triggers signal in current process.

This is pretty much the same as kill(getpid(), sig).

unix.access(path:str, how:int[, flags:int[, dirfd:int]])
├─→ true
└─→ nil, unix.Errno
Checks if effective user of current process has permission to access file. how can be R_OK, W_OK, X_OK, or F_OK to check for read, write, execute, and existence respectively.

flags may have any of:

unix.mkdir(path:str[, mode:int[, dirfd:int]])
├─→ true
└─→ nil, unix.Errno
Makes directory.

path is the path of the directory you wish to create.

mode is octal permission bits, e.g. 0755.

Fails with EEXIST if path already exists, whether it be a directory or a file.

Fails with ENOENT if the parent directory of the directory you want to create doesn't exist. For making a/really/long/path/ consider using makedirs() instead.

Fails with ENOTDIR if a parent directory component existed that wasn't a directory.

Fails with EACCES if the parent directory doesn't grant write permission to the current user.

Fails with ENAMETOOLONG if the path is too long.

unix.makedirs(path:str[, mode:int])
├─→ true
└─→ nil, unix.Errno
Makes directories.

Unlike mkdir() this convenience wrapper will automatically create parent parent directories as needed. If the directory already exists then, unlike mkdir() which returns EEXIST, the makedirs() function will return success.

path is the path of the directory you wish to create.

mode is octal permission bits, e.g. 0755.

unix.chdir(path:str)
├─→ true
└─→ nil, unix.Errno
Changes current directory to path.
Removes file at path.

If path refers to a symbolic link, the link is removed.

Returns EISDIR if path refers to a directory. See rmdir().

unix.rmdir(path:str[, dirfd:int])
├─→ true
└─→ nil, unix.Errno
Removes empty directory at path.

Returns ENOTDIR if path isn't a directory, or a path component in path exists yet wasn't a directory.

unix.rename(oldpath:str, newpath:str[, olddirfd:int, newdirfd:int])
├─→ true
└─→ nil, unix.Errno
Renames file or directory.
Creates hard link, so your underlying inode has two names.
Creates symbolic link.

On Windows NT a symbolic link is called a "reparse point" and can only be created from an administrator account. Your redbean will automatically request the appropriate permissions.

Reads contents of symbolic link.

Note that broken links are supported on all platforms. A symbolic link can contain just about anything. It's important to not assume that content will be a valid filename.

On Windows NT, this function transliterates \ to / and furthermore prefixes //?/ to WIN32 DOS-style absolute paths, thereby assisting with simple absolute filename checks in addition to enabling one to exceed the traditional 260 character limit.

unix.rmrf(path:str)
├─→ true
└─→ nil, unix.Errno
Recursively removes filesystem path.

Like unix.makedirs() this function isn't actually a system call but rather is a Libc convenience wrapper. It's intended to be equivalent to using the UNIX shell's rm -rf path command.

path is the file or directory path you wish to destroy.

unix.realpath(path:str)
├─→ path:str
└─→ nil, unix.Errno
Returns absolute path of filename, with . and .. components removed, and symlinks will be resolved.
unix.utimensat(path[, asecs, ananos, msecs, mnanos[, dirfd[, flags]]])
├─→ 0
└─→ nil, unix.Errno
Changes access and/or modified timestamps on file.

path is a string with the name of the file.

The asecs and ananos parameters set the access time. If they're none or nil, the current time will be used.

The msecs and mnanos parameters set the modified time. If they're none or nil, the current time will be used.

The nanosecond parameters (ananos and mnanos) must be on the interval [0,1000000000) or unix.EINVAL is raised. On XNU this is truncated to microsecond precision. On Windows NT, it's truncated to hectonanosecond precision. These nanosecond parameters may also be set to one of the following special values:

dirfd is a file descriptor integer opened with O_DIRECTORY that's used for relative path names. It defaults to unix.AT_FDCWD.

flags may have have any of the following flags bitwise or'd

unix.futimens(fd:int[, asecs, ananos, msecs, mnanos])
├─→ 0
└─→ nil, unix.Errno
Changes access and/or modified timestamps on file descriptor.

fd is the file descriptor of a file opened with unix.open.

The asecs and ananos parameters set the access time. If they're none or nil, the current time will be used.

The msecs and mnanos parameters set the modified time. If they're none or nil, the current time will be used.

The nanosecond parameters (ananos and mnanos) must be on the interval [0,1000000000) or unix.EINVAL is raised. On XNU this is truncated to microsecond precision. On Windows NT, it's truncated to hectonanosecond precision. These nanosecond parameters may also be set to one of the following special values:

This system call is currently not available on very old versions of Linux, e.g. RHEL5.

unix.chown(path:str, uid:int, gid:int[, flags:int[, dirfd:int]])
├─→ true
└─→ nil, unix.Errno
Changes user and group on file.

Returns ENOSYS on Windows NT.

unix.chmod(path:str, mode:int[, flags:int[, dirfd:int]])
├─→ true
└─→ nil, unix.Errno
Changes mode bits on file.

On Windows NT the chmod system call only changes the read-only status of a file.

unix.getcwd()
├─→ path:str
└─→ nil, unix.Errno
Returns current working directory.

On Windows NT, this function transliterates \ to / and furthermore prefixes //?/ to WIN32 DOS-style absolute paths, thereby assisting with simple absolute filename checks in addition to enabling one to exceed the traditional 260 character limit.

unix.fcntl(fd:int, unix.F_GETFD)
├─→ flags:int
└─→ nil, unix.Errno
Returns file descriptor flags.

The returned flags may include any of:

Returns EBADF if fd isn't open.

unix.fcntl(fd:int, unix.F_SETFD, flags:int)
├─→ true
└─→ nil, unix.Errno
Sets file descriptor flags.

flags may include any of:

Returns EBADF if fd isn't open.

unix.fcntl(fd:int, unix.F_GETFL)
├─→ flags:int
└─→ nil, unix.Errno
Returns file descriptor status flags.

flags & unix.O_ACCMODE includes one of:

Examples of values flags & ~unix.O_ACCMODE may include:

Examples of values flags & ~unix.O_ACCMODE won't include:

Returns EBADF if fd isn't open.

unix.fcntl(fd:int, unix.F_SETFL, flags:int)
├─→ true
└─→ nil, unix.Errno
Changes file descriptor status flags.

Examples of values flags may include:

These values should be ignored:

Returns EBADF if fd isn't open.

unix.fcntl(fd:int, unix.F_SETLK[, type[, start[, len[, whence]]]])
unix.fcntl(fd:int, unix.F_SETLKW[, type[, start[, len[, whence]]]])
├─→ true
└─→ nil, unix.Errno
Acquires lock on file interval.

POSIX Advisory Locks allow multiple processes to leave voluntary hints to each other about which portions of a file they're using.

The command may be:

fd is file descriptor of open() file.

type may be one of:

start is 0-indexed byte offset into file. The default is zero.

len is byte length of interval. Zero is the default and it means until the end of the file.

whence may be one of:

Returns EAGAIN if lock couldn't be acquired. POSIX says this theoretically could also be EACCES but we haven't seen this behavior on any of our supported platforms.

Returns EBADF if fd wasn't open.

unix.fcntl(fd:int, unix.F_GETLK[, type[, start[, len[, whence]]]])
├─→ unix.F_UNLCK
├─→ type, start, len, whence, pid
└─→ nil, unix.Errno
Acquires information about POSIX advisory lock on file.

This function accepts the same parameters as fcntl(F_SETLK) and tells you if the lock acquisition would be successful for a given range of bytes. If locking would have succeeded, then F_UNLCK is returned. If the lock would not have succeeded, then information about a conflicting lock is returned.

Returned type may be F_RDLCK or F_WRLCK.

Returned pid is the process id of the current lock owner.

This function is currently not supported on Windows.

Returns EBADF if fd wasn't open.

unix.getsid(pid:int)
├─→ sid:int
└─→ nil, unix.Errno
Gets session id.
unix.getpgrp()
├─→ pgid:int
└─→ nil, unix.Errno
Gets process group id.
unix.setpgrp()
├─→ pgid:int
└─→ nil, unix.Errno
Sets process group id. This is the same as setpgid(0,0).
unix.setpgid(pid:int, pgid:int)
├─→ true
└─→ nil, unix.Errno
Sets process group id the modern way.
unix.getpgid(pid:int)
├─→ pgid:int
└─→ nil, unix.Errno
Gets process group id the modern way.
unix.setsid()
├─→ sid:int
└─→ nil, unix.Errno
Sets session id.

This function can be used to create daemons.

Fails with ENOSYS on Windows NT.

unix.getuid()
└─→ uid:int
Gets real user id.

On Windows this system call is polyfilled by running GetUserNameW() through Knuth's multiplicative hash.

This function does not fail.

unix.getgid()
└─→ gid:int
Sets real group id.

On Windows this system call is polyfilled as getuid().

This function does not fail.

unix.geteuid()
└─→ uid:int
Gets effective user id.

For example, if your redbean is a setuid binary, then getuid() will return the uid of the user running the program, and geteuid() shall return zero which means root, assuming that's the file owning user.

On Windows this system call is polyfilled as getuid().

This function does not fail.

unix.getegid()
└─→ gid:int
Gets effective group id.

On Windows this system call is polyfilled as getuid().

This function does not fail.

unix.chroot(path:str)
├─→ true
└─→ nil, unix.Errno
Changes root directory.

Returns ENOSYS on Windows NT.

unix.setuid(uid:int)
├─→ true
└─→ nil, unix.Errno
Sets user id.

One use case for this function is dropping root privileges. Should you ever choose to run redbean as root and decide not to use the -G and -U flags, you can replicate that behavior in the Lua processes you spawn as follows:

ok, err = unix.setgid(1000)  -- check your /etc/groups
if not ok then Log(kLogFatal, tostring(err)) end
ok, err = unix.setuid(1000)  -- check your /etc/passwd
if not ok then Log(kLogFatal, tostring(err)) end

If your goal is to relinquish privileges because redbean is a setuid binary, then things are more straightforward:

ok, err = unix.setgid(unix.getgid())
if not ok then Log(kLogFatal, tostring(err)) end
ok, err = unix.setuid(unix.getuid())
if not ok then Log(kLogFatal, tostring(err)) end

See also the setresuid() function and be sure to refer to your local system manual about the subtleties of changing user id in a way that isn't restorable.

Returns ENOSYS on Windows NT if uid isn't getuid().

unix.setgid(gid:int)
├─→ true
└─→ nil, unix.Errno
Sets group id.

Returns ENOSYS on Windows NT if gid isn't getgid().

unix.setresuid(real:int, effective:int, saved:int)
├─→ true
└─→ nil, unix.Errno
Sets real, effective, and saved user ids.

If any of the above parameters are -1, then it's a no-op.

Returns ENOSYS on Windows NT. Returns ENOSYS on Macintosh and NetBSD if saved isn't -1.

unix.setresgid(real:int, effective:int, saved:int)
├─→ true
└─→ nil, unix.Errno
Sets real, effective, and saved group ids.

If any of the above parameters are -1, then it's a no-op.

Returns ENOSYS on Windows NT. Returns ENOSYS on Macintosh and NetBSD if saved isn't -1.

unix.umask(newmask:int)
└─→ oldmask:int
Sets file permission mask and returns the old one.

This is used to remove bits from the mode parameter of functions like open() and mkdir(). The masks typically used are 027 and 022. Those masks ensure that, even if a file is created with 0666 bits, it'll be turned into 0640 or 0644 so that users other than the owner can't modify it.

To read the mask without changing it, try doing this:

mask = unix.umask(027)
unix.umask(mask)

On Windows NT this is a no-op and mask is returned.

This function does not fail.

unix.syslog(priority:int, msg:str)
Generates a log message, which will be distributed by syslogd.

priority is a bitmask containing the facility value and the level value. If no facility value is ORed into priority, then the default value set by openlog() is used. If set to NULL, the program name is used. Level is one of LOG_EMERG, LOG_ALERT, LOG_CRIT, LOG_ERR, LOG_WARNING, LOG_NOTICE, LOG_INFO, LOG_DEBUG.

This function currently works on Linux, Windows, and NetBSD. On WIN32 it uses the ReportEvent() facility.

unix.clock_gettime([clock:int])
├─→ seconds:int, nanos:int
└─→ nil, unix.Errno
Returns nanosecond precision timestamp from system, e.g.
>: unix.clock_gettime()
1651137352      774458779
>: Benchmark(unix.clock_gettime)
126     393     571     1

clock can be any one of of:

Returns EINVAL if clock isn't supported on platform.

This function only fails if clock is invalid.

This function goes fastest on Linux and Windows.

unix.nanosleep(seconds:int[, nanos:int])
├─→ remseconds:int, remnanos:int
└─→ nil, unix.Errno
Sleeps with nanosecond precision.

Returns EINTR if a signal was received while waiting.

unix.sync()
unix.fsync(fd:int)
├─→ true
└─→ nil, unix.Errno
unix.fdatasync(fd:int)
├─→ true
└─→ nil, unix.Errno
These functions are used to make programs slower by asking the operating system to flush data to the physical medium.
unix.lseek(fd:int, offset:int[, whence:int])
├─→ newposbytes:int
└─→ nil, unix.Errno
Seeks to file position.

whence can be one of:

Returns the new position relative to the start of the file.

unix.truncate(path:str[, length:int])
├─→ true
└─→ nil, unix.Errno
Reduces or extends underlying physical medium of file. If file was originally larger, content >length is lost.

length defaults to zero.

unix.ftruncate(fd:int[, length:int])
├─→ true
└─→ nil, unix.Errno
Reduces or extends underlying physical medium of open file. If file was originally larger, content >length is lost.

length defaults to zero.

unix.socket([family:int[, type:int[, protocol:int]]])
├─→ fd:int
└─→ nil, unix.Errno
Creates socket endpoint for process communication.

family defaults to AF_INET and can be:

type defaults to SOCK_STREAM and can be:

You may bitwise or any of the following into type:

protocol defaults to IPPROTO_TCP and can be:
unix.socketpair([family:int[, type:int[, protocol:int]]])
├─→ fd1:int, fd2:int
└─→ nil, unix.Errno
Creates bidirectional pipe.

family defaults to AF_UNIX.

type defaults to SOCK_STREAM and can be:

You may bitwise or any of the following into type:

protocol defaults to 0.

unix.bind(fd:int[, ip:uint32, port:uint16])
├─→ true
└─→ nil, unix.Errno
Binds socket.

ip and port are in host endian order. For example, if you wanted to listen on 1.2.3.4:31337 you could do any of these

unix.bind(sock, 0x01020304, 31337)
unix.bind(sock, ParseIp('1.2.3.4'), 31337)
unix.bind(sock, 1 << 24 | 0 << 16 | 0 << 8 | 1, 31337)

ip and port both default to zero. The meaning of bind(0, 0) is to listen on all interfaces with a kernel-assigned ephemeral port number, that can be retrieved and used as follows:

sock = assert(unix.socket())  -- create ipv4 tcp socket
assert(unix.bind(sock))       -- all interfaces ephemeral port
ip, port = assert(unix.getsockname(sock))
print('listening on ip', FormatIp(ip), 'port', port)
assert(unix.listen(sock))
assert(unix.accept(sock))
while true do
   client, clientip, clientport = assert(unix.accept(sock))
   print('got client ip', FormatIp(clientip), 'port', clientport)
   unix.close(client)
end

Further note that calling unix.bind(sock) is equivalent to not calling bind() at all, since the above behavior is the default.

unix.siocgifconf()
├─→ {{name:str, ip:uint32, netmask:uint32}, ...}
└─→ nil, unix.Errno
Returns list of network adapter addresses.
unix.getsockopt(fd:int, level:int, optname:int) → ...
unix.setsockopt(fd:int, level:int, optname:int, ...) → ok:bool, unix.Errno

Tunes networking parameters.

level and optname may be one of the following pairs. The ellipses type signature above changes depending on which options are used.

optname is the option feature magic number. The constants for these will be set to 0 if the option isn't supported on the host platform.

Raises ENOPROTOOPT if your level / optname combination isn't valid, recognized, or supported on the host platform.

Raises ENOTSOCK if fd is valid but isn't a socket.

Raises EBADF if fd isn't valid.

unix.getsockopt(fd:int, level:int, optname:int)
    ├─→ value:int
    └─→ nil, unix.Errno
unix.setsockopt(fd:int, level:int, optname:int, value:bool)
    ├─→ true
    └─→ nil, unix.Errno
unix.getsockopt(fd:int, level:int, optname:int)
    ├─→ value:int
    └─→ nil, unix.Errno
unix.setsockopt(fd:int, level:int, optname:int, value:int)
    ├─→ true
    └─→ nil, unix.Errno
unix.getsockopt(fd:int, level:int, optname:int)
    ├─→ secs:int, nsecs:int
    └─→ nil, unix.Errno
unix.setsockopt(fd:int, level:int, optname:int, secs:int[, nanos:int])
    ├─→ true
    └─→ nil, unix.Errno
unix.getsockopt(fd:int, unix.SOL_SOCKET, unix.SO_LINGER)
    ├─→ seconds:int, enabled:bool
    └─→ nil, unix.Errno
unix.setsockopt(fd:int, unix.SOL_SOCKET, unix.SO_LINGER, secs:int, enabled:bool)
    ├─→ true
    └─→ nil, unix.Errno
This SO_LINGER parameter can be used to make close() a blocking call. Normally when the kernel returns immediately when it receives close(). Sometimes it's desirable to have extra assurance on errors happened, even if it comes at the cost of performance.
unix.setsockopt(serverfd:int, unix.SOL_TCP, unix.TCP_SAVE_SYN, enabled:int)
    ├─→ true
    └─→ nil, unix.Errno
unix.getsockopt(clientfd:int, unix.SOL_TCP, unix.TCP_SAVED_SYN)
    ├─→ syn_packet_bytes:str
    └─→ nil, unix.Errno

This TCP_SAVED_SYN option may be used to retrieve the bytes of the TCP SYN packet that the client sent when the connection for fd was opened. In order for this to work, TCP_SAVE_SYN must have been set earlier on the listening socket. This is Linux-only. You can use the OnServerListen hook to enable SYN saving in your Redbean. When the TCP_SAVE_SYN option isn't used, this may return empty string.

unix.poll({[fd:int]=events:int, ...}[, timeoutms:int[, mask:unix.Sigset]])
├─→ {[fd:int]=revents:int, ...}
└─→ nil, unix.Errno
Checks for events on a set of file descriptors.

The table of file descriptors to poll uses sparse integer keys. Any pairs with non-integer keys will be ignored. Pairs with negative keys are ignored by poll(). The returned table will be a subset of the supplied file descriptors.

events and revents may be any combination (using bitwise OR) of:

timeoutms is the number of milliseconds to block. The default is -1 which means block indefinitely until there's an event or an interrupt. If the timeout elapses without any such events, an empty table is returned. A timeout of zero means non-blocking.

mask serves the purpose of enabling poll to listen for both file descriptor events and signals. It's equivalent to saying:

oldmask = unix.sigprocmask(unix.SIG_SETMASK, mask);
unix.poll(fds, timeout);
unix.sigprocmask(unix.SIG_SETMASK, oldmask);

Except it'll happen atomically on supported platforms. The only exceptions are MacOS and NetBSD where this behavior is simulated by the polyfill. Atomicity is helpful for unit testing signal behavior.

EINTR is returned if the kernel decided to deliver a signal to a signal handler instead during your call. This is a @norestart system call that always returns EINTR even if SA_RESTART is in play.

unix.gethostname()
├─→ host:str
└─→ nil, unix.Errno
Returns hostname of system.
unix.listen(fd:int[, backlog:int])
├─→ true
└─→ nil, unix.Errno
Begins listening for incoming connections on a socket.
unix.accept(serverfd:int[, flags:int])
├─→ clientfd:int, ip:uint32, port:uint16
└─→ nil, unix.Errno
Accepts new client socket descriptor for a listening tcp socket.

flags can have any of:

unix.connect(fd:int, ip:uint32, port:uint16)
├─→ true
└─→ nil, unix.Errno
Connects a TCP socket to a remote host.

With TCP this is a blocking operation. For a UDP socket it simply remembers the intended address so that send() or write() may be used rather than sendto().

unix.getsockname(fd:int)
├─→ ip:uint32, port:uint16
└─→ nil, unix.Errno
Retrieves the local address of a socket.
unix.getpeername(fd:int)
├─→ ip:uint32, port:uint16
└─→ nil, unix.Errno
Retrieves the remote address of a socket.
unix.recv(fd:int[, bufsiz:int[, flags:int]])
├─→ data:str
└─→ nil, unix.Errno
Receives message from a socket.

flags can have:

unix.recvfrom(fd:int[, bufsiz:int[, flags:int]])
├─→ data:str, ip:uint32, port:uint16
└─→ nil, unix.Errno
Receives message from a socket.

flags can have:

unix.send(fd:int, data:str[, flags:int])
├─→ sent:int
└─→ nil, unix.Errno
This is the same as write except it has a flags argument that's intended for sockets.

flags may have any of:

unix.sendto(fd:int, data:str, ip:uint32, port:uint16[, flags:int])
├─→ sent:int
└─→ nil, unix.Errno
This is useful for sending messages over UDP sockets to specific addresses.

flags may have any of:

unix.shutdown(fd:int, how:int)
├─→ true
└─→ nil, unix.Errno
Partially closes socket.

how is set to one of:

This system call currently has issues on Macintosh, so portable code should log rather than assert failures reported by shutdown().

unix.sigprocmask(how:int, newmask:Sigset)
├─→ oldmask:unix.Sigset
└─→ nil, unix.Errno
Manipulates bitset of signals blocked by process.

how can be one of:

mask is a unix.Sigset() object (see section below).

For example, to temporarily block SIGTERM and SIGINT so critical work won't be interrupted, sigprocmask() can be used as follows:

newmask = unix.Sigset(unix.SIGTERM)
oldmask = assert(unix.sigprocmask(unix.SIG_BLOCK, newmask))
-- do something...
assert(unix.sigprocmask(unix.SIG_SETMASK, oldmask))
unix.sigaction(sig:int[, handler:func|int[, flags:int[, mask:Sigset]]])
├─→ oldhandler:func|int, flags:int, mask:unix.Sigset
└─→ nil, unix.Errno
Changes action taken upon receipt of a specific signal.

sig can be one of:

handler can be:

flags can have:

Example:

function OnSigUsr1(sig)
  gotsigusr1 = true
end
gotsigusr1 = false
oldmask = assert(unix.sigprocmask(unix.SIG_BLOCK, unix.Sigset(unix.SIGUSR1)))
assert(unix.sigaction(unix.SIGUSR1, OnSigUsr1))
assert(unix.raise(unix.SIGUSR1))
assert(not gotsigusr1)
ok, err = unix.sigsuspend(oldmask)
assert(not ok)
assert(err:errno() == unix.EINTR)
assert(gotsigusr1)
assert(unix.sigprocmask(unix.SIG_SETMASK, oldmask))

It's a good idea to not do too much work in a signal handler.

unix.sigsuspend([mask:unix.Sigset])
└─→ nil, unix.Errno
Waits for signal to be delivered.

The signal mask is temporarily replaced with mask during this system call. mask specifies which signals should be blocked.

unix.setitimer(which[, intervalsec, intns, valuesec, valuens])
├─→ intervalsec:int, intervalns:int, valuesec:int, valuens:int
└─→ nil, unix.Errno
Causes SIGALRM signals to be generated at some point(s) in the future. The which parameter should be ITIMER_REAL.

Here's an example of how to create a 400 ms interval timer:

ticks = 0
assert(unix.sigaction(unix.SIGALRM, function(sig)
   print('tick no. %d' % {ticks})
   ticks = ticks + 1
end))
assert(unix.setitimer(unix.ITIMER_REAL, 0, 400e6, 0, 400e6))
while true do
   unix.sigsuspend()
end

Here's how you'd do a single-shot timeout in 1 second:

unix.sigaction(unix.SIGALRM, MyOnSigAlrm, unix.SA_RESETHAND)
unix.setitimer(unix.ITIMER_REAL, 0, 0, 1, 0)
unix.strsignal(sig:int) → str
Turns platform-specific sig code into its symbolic name.

For example:

>: unix.strsignal(9)
"SIGKILL"
>: unix.strsignal(unix.SIGKILL)
"SIGKILL"

Please note that signal numbers are normally different across supported platforms, and the constants should be preferred.

unix.setrlimit(resource:int, soft:int[, hard:int])
├─→ true
└─→ nil, unix.Errno
Changes resource limit.

resource may be one of:

If a limit isn't supported by the host platform, it'll be set to 127. On most platforms these limits are enforced by the kernel and as such are inherited by subprocesses.

hard defaults to whatever was specified in soft.

unix.getrlimit(resource:int)
├─→ soft:int, hard:int
└─→ nil, unix.Errno
Returns information about resource limits for current process.
unix.getrusage([who:int])
├─→ unix.Rusage
└─→ nil, unix.Errno
Returns information about resource usage for current process, e.g.
>: unix.getrusage()
{utime={0, 53644000}, maxrss=44896, minflt=545, oublock=24, nvcsw=9}

who defaults to RUSAGE_SELF and can be any of:

See the unix.Rusage section below for details on returned fields.

unix.pledge([promises:str[, execpromises:str[, mode:int]]])
├─→ true
└─→ nil, unix.Errno
Restrict system operations.

This can be used to sandbox your redbean workers. It allows finer customization compared to the -S flag.

Pledging causes most system calls to become unavailable. If a forbidden system call is used, then the process will be killed. In that case, on OpenBSD, your system log will explain which promise you need. On Linux, we report the promise to stderr, with one exception: reporting is currently not possible if you pledge exec.

Using pledge is irreversible. On Linux it causes PR_SET_NO_NEW_PRIVS to be set on your process.

By default exit and exit_group are always allowed. This is useful for processes that perform pure computation and interface with the parent via shared memory.

Once pledge is in effect, the chmod functions (if allowed) will not permit the sticky/setuid/setgid bits to change. Linux will EPERM here and OpenBSD should ignore those three bits rather than crashing.

User and group IDs also can't be changed once pledge is in effect. OpenBSD should ignore the chown functions without crashing. Linux will just EPERM.

Root access isn't required. Support is limited to OpenBSD and Linux 2.6.23+ (i.e. RHEL6 c. 2012) so long as Redbean is running directly on the host system, i.e. not running in a userspace emulator like Blink or Qemu. If your environment isn't supported, then pledge() will return 0 and do nothing, rather than raising ENOSYS, so the apps you share with others will err on the side of not breaking. If a functionality check is needed, please use unix.pledge(nil, nil) which is a no-op that will fail appropriately when the necessary system support isn't available to impose security restrictions.

promises is a string that may include any of the following groups delimited by spaces. This list has been curated to focus on the system calls for which this module provides wrappers. See the Cosmopolitan Libc pledge() documentation for a comprehensive and authoritative list of raw system calls. Having the raw system call list may be useful if you're executing foreign programs.

stdio
Allows read, write, send, recv, recvfrom, close, clock_getres, clock_gettime, dup, fchdir, fstat, fsync, fdatasync, ftruncate, getdents, getegid, getrandom, geteuid, getgid, getgroups, getitimer, getpgid, getpgrp, getpid, hgetppid, getresgid, getresuid, getrlimit, getsid, gettimeofday, getuid, lseek, madvise, brk, mmap/mprotect (PROT_EXEC isn't allowed), msync, munmap, gethostname, nanosleep, pipe, pipe2, poll, setitimer, shutdown, sigaction, sigsuspend, sigprocmask, socketpair, umask, wait4, getrusage, ioctl(FIONREAD), ioctl(FIONBIO), ioctl(FIOCLEX), ioctl(FIONCLEX), fcntl(F_GETFD), fcntl(F_SETFD), fcntl(F_GETFL), fcntl(F_SETFL), raise, kill(getpid()).
rpath (read path ops)
Allows chdir, getcwd, open, stat, fstat, access, readlink, chmod, chmod, fchmod.
wpath (write path ops)
Allows getcwd, open, stat, fstat, access, readlink, chmod, fchmod.
cpath (create path ops)
Allows rename, link, symlink, unlink, mkdir, rmdir.
fattr
Allows chmod, fchmod, utimensat, futimens.
flock
Allows flock, fcntl(F_GETLK), fcntl(F_SETLK), fcntl(F_SETLKW).
tty
Allows isatty, tcgetwinsize, tcgets, tcsets, tcsetsw, tcsetsf.
inet
Allows socket (AF_INET), listen, bind, connect, accept, getpeername, getsockname, setsockopt, getsockopt.
anet
Allows socket (AF_INET), listen, bind, accept, getpeername, getsockname, setsockopt, getsockopt.
unix
Allows socket (AF_UNIX), listen, bind, connect, accept, getpeername, getsockname, setsockopt, getsockopt.
dns
Allows sendto, recvfrom, socket(AF_INET), connect.
recvfd
Allows recvmsg, recvmmsg.
sendfd
Allows sendmsg, sendmmsg.
proc
Allows fork, vfork, clone, kill, tgkill, getpriority, setpriority, setrlimit, setpgid, setsid.
settime
Allows settimeofday and clock_adjtime.
chown
Allows chown.
unveil
Allows unveil.
exec
Allows execve.
prot_exec
Allows mmap(PROT_EXEC) and mprotect(PROT_EXEC).

This may be needed to launch non-static non-native executables, such as non-assimilated APE binaries, or programs that link dynamic shared objects, i.e. most Linux distro binaries.

execpromises only matters if "exec" is specified in promises. In that case, this specifies the promises that'll apply once execve() happens. If this is NULL then the default is used, which is unrestricted. OpenBSD allows child processes to escape the sandbox (so a pledged OpenSSH server process can do things like spawn a root shell). Linux however requires monotonically decreasing privileges. This function will will perform some validation on Linux to make sure that execpromises is a subset of promises. Your libc wrapper for execve() will then apply its SECCOMP BPF filter later. Since Linux has to do this before calling sys_execve(), the executed process will be weakened to have execute permissions too.

mode if specified should specify one penalty:

mode may optionally bitwise or the following flags:

unix.unveil(path:str, permissions:str)
├─→ true
└─→ nil, unix.Errno
Restricts filesystem operations, e.g.
unix.unveil(".", "r");     -- current dir + children visible
unix.unveil("/etc", "r");  -- make /etc readable too
unix.unveil(nil, nil);     -- commit and lock policy
Unveiling restricts a thread's view of the filesystem to a set of allowed paths with specific privileges.

Once you start using unveil(), the entire file system is considered hidden. You then specify, by repeatedly calling unveil(), which paths should become unhidden. When you're finished, you call unveil(0,0) which commits your policy, after which further use is forbidden, in the current thread, as well as any threads or processes it spawns.

There are some differences between unveil() on Linux versus OpenBSD.

  1. Build your policy and lock it in one go. On OpenBSD, policies take effect immediately and may evolve as you continue to call unveil() but only in a more restrictive direction. On Linux, nothing will happen until you call unveil(0,0) which commits and locks.
  2. Try not to overlap directory trees. On OpenBSD, if directory trees overlap, then the most restrictive policy will be used for a given file. On Linux overlapping may result in a less restrictive policy and possibly even undefined behavior.
  3. OpenBSD and Linux disagree on error codes. On OpenBSD, accessing paths outside of the allowed set raises ENOENT, and accessing ones with incorrect permissions raises EACCES. On Linux, both these cases raise EACCES.
  4. Unlike OpenBSD, Linux does nothing to conceal the existence of paths. Even with an unveil() policy in place, it's still possible to access the metadata of all files using functions like stat() and open(O_PATH), provided you know the path. A sandboxed process can always, for example, determine how many bytes of data are in /etc/passwd, even if the file isn't readable. But it's still not possible to use opendir() and go fishing for paths which weren't previously known.

This system call is supported natively on OpenBSD and polyfilled on Linux using the Landlock LSM[1].

This function requires OpenBSD or Linux 5.13+ (2022+). If the kernel support isn't available (or we're in an emulator like Qemu or Blink) then zero is returned and nothing happens (instead of raising ENOSYS) because the files are still unveiled. Use unix.unveil("", nil) to feature check the host system, which is defined as a no-op that'll fail if the host system doesn't have the necessary features that allow unix.unveil() impose bona-fide security restrictions. Otherwise, if everything is good, a return value >=0 is returned, where 0 means OpenBSD, and >=1 means Linux with Landlock LSM, in which case the return code shall be the maximum supported Landlock ABI version.

path is the file or directory to unveil

permissions is a string consisting of zero or more of the following characters:

unix.gmtime(unixts:int)
├─→ year,mon,mday,hour,min,sec,gmtoffsec,wday,yday,dst:int,zone:str
└─→ nil, unix.Errno
Breaks down UNIX timestamp into Zulu Time numbers.

This function is like localtime() except it always returns Greenwich Mean Time irrespective of the TZ environment variable.

For example:

>: unix.gmtime(unix.clock_gettime())
2022    5       11      22      43      20      0       3       130     0       "GMT"

Here's how you might format a localized timestamp with nanoseconds:

>: unixsec, nanos = unix.clock_gettime()
>: year,mon,mday,hour,min,sec = unix.localtime(unixsec)
>: '%.4d-%.2d-%.2dT%.2d:%.2d:%.2d.%.9dZ' % {year,mon,mday,hour,min,sec,nanos}
"2022-05-11T15:46:32.160239978Z"

year is the year, where zero is defined as 0 A.D. This value may be on the interval -13.7e9 ≤ year ≤ 10e14 which is the time from the Big Bang, through most of the Stelliferous Era.

mon is the month of the year, on the interval 1 ≤ mon ≤ 12 in order to make printf style formatting easier.

mday is the day of the month, on the interval 1 ≤ mday ≤ 31 in order to make printf style formatting easier.

hour represent hours, on the interval 0 ≤ hour ≤ 23.

min represents minutes, on the interval 0 ≤ min ≤ 59.

sec represents seconds, on the interval 0 ≤ sec ≤ 60. Please note this is a 61 second interval in order to accommodate highly rare leap second events.

wday is the day of the week, on the interval 0 ≤ wday ≤ 6.

yday is the day of the year on the interval 0 ≤ yday ≤ 365.

gmtoff is the Zulu time offset in seconds, which should be on the interval ±93600 seconds.

dst will be 1 if daylight savings, 0 if not daylight savings, or -1 if it couldn't be determined.

unix.localtime(unixts:int)
├─→ year,mon,mday,hour,min,sec,gmtoffsec,wday,yday,dst:int,zone:str
└─→ nil, unix.Errno
Breaks down UNIX timestamp into local time numbers, e.g.
>: unix.localtime(unix.clock_gettime())
2022    4       28      2       14      22      -25200  4       117     1       "PDT"

This follows the same API as gmtime() except it takes the TZ environment variable into consideration to determine the most appropriate localization.

Please see the gmtime() function for documentaiton on the meaning of the various returned values.

Here's an example of how you might format a localized timestamp:

>: unixsec, nanos = unix.clock_gettime()
>: year, mon, mday, hour, min, sec, gmtoffsec = unix.localtime(unixsec)
>: '%.4d-%.2d-%.2dT%.2d:%.2d:%.2d.%.9d%+.2d%.2d' % {
        year, mon, mday, hour, min, sec, nanos,
        gmtoffsec / (60 * 60), math.abs(gmtoffsec) % 60}
"2022-05-11T15:46:32.160239978-0700"

Your redbean ships with a subset of the time zone database.

You can control which timezone is used using the TZ environment variable. If your time zone isn't included in the above list, you can simply copy it inside your redbean. The same is also the case for future updates to the database, which can be swapped out when needed, without having to recompile.

unix.stat(path:str[, flags:int[, dirfd:int]])
├─→ unix.Stat
└─→ nil, unix.Errno
Gets information about file or directory.

flags may have any of:

dirfd defaults to to unix.AT_FDCWD and may optionally be set to a directory file descriptor to which path is relative.

unix.fstat(fd:int)
├─→ unix.Stat
└─→ nil, unix.Errno
Gets information about opened file descriptor.

A common use for fstat() is getting the size of a file. For example:

fd = assert(unix.open('hello.txt', unix.O_RDONLY))
st = assert(unix.fstat(fd))
Log(kLogInfo, 'hello.txt is %d bytes in size' % {st:size()})
unix.close(fd)
unix.statfs(path:str)
├─→ unix.Statfs
└─→ nil, unix.Errno
Gets information about filesystem.

path is the path of a file or directory in the mounted filesystem.

unix.fstatfs(fd:int)
├─→ unix.Statfs
└─→ nil, unix.Errno
Gets information about filesystem.

fd is an open() file descriptor of a file or directory in the mounted filesystem.

unix.opendir(path:str)
├─→ state:Dir
└─→ nil, unix.Errno
Opens directory for listing its contents.

For example, to print a simple directory listing:

Write('<ul>\r\n')
for name, kind, ino, off in assert(unix.opendir(dir)) do
    if name ~= '.' and name ~= '..' then
       Write('<li>%s\r\n' % {EscapeHtml(name)})
    end
end
Write('</ul>\r\n')
unix.fdopendir(fd:int)
├─→ next:function, state:Dir
└─→ nil, unix.Errno
Opens directory for listing its contents, via an fd.

fd should be created by open(path, unix.O_RDONLY|unix.O_DIRECTORY). The returned unix.Dir takes ownership of the file descriptor and will close it automatically when garbage collected.

unix.isatty(fd:int)
├─→ true
└─→ nil, unix.Errno
Returns true if file descriptor is a teletypewriter. Otherwise nil with an Errno object holding one of the following values:

No other error numbers are possible.

unix.tcgetwinsize(fd:int)
├─→ rows:int, cols:int
└─→ nil, unix.Errno
Returns cellular dimensions of pseudoteletypewriter display.
unix.tmpfd()
├─→ fd:int
└─→ nil, unix.Errno
Returns file descriptor of open anonymous file.

This creates a secure temporary file inside $TMPDIR. If it isn't defined, then /tmp is used on UNIX and GetTempPath() is used on the New Technology. This resolution of $TMPDIR happens once, which is copied to the kTmpDir global.

Once close() is called, the returned file is guaranteed to be deleted automatically. On UNIX the file is unlink()'d before this function returns. On the New Technology it happens upon close().

On the New Technology, temporary files created by this function should have better performance, because kNtFileAttributeTemporary asks the kernel to more aggressively cache and reduce i/o ops.

unix.sched_yield()
Relinquishes scheduled quantum.
unix.mapshared(size:int)
└─→ unix.Memory()
Creates interprocess shared memory mapping.

This function allocates special memory that'll be inherited across fork in a shared way. By default all memory in Redbean is "private" memory that's only viewable and editable to the process that owns it. When unix.fork() happens, memory is copied appropriately so that changes to memory made in the child process, don't clobber the memory at those same addresses in the parent process. If you don't want that to happen, and you want the memory to be shared similar to how it would be shared if you were using threads, then you can use this function to achieve just that.

The memory object this function returns may be accessed using its methods, which support atomics and futexes. It's very low-level. For example, you can use it to implement scalable mutexes:

mem = unix.mapshared(8000 * 8)
LOCK = 0 -- pick an arbitrary word index for lock
-- From Futexes Are Tricky Version 1.1 § Mutex, Take 3;
-- Ulrich Drepper, Red Hat Incorporated, June 27, 2004.
function Lock()
    local ok, old = mem:cmpxchg(LOCK, 0, 1)
    if not ok then
        if old == 1 then
            old = mem:xchg(LOCK, 2)
        end
        while old > 0 do
            mem:wait(LOCK, 2)
            old = mem:xchg(LOCK, 2)
        end
    end
end
function Unlock()
    old = mem:fetch_add(LOCK, -1)
    if old == 2 then
        mem:store(LOCK, 0)
        mem:wake(LOCK, 1)
    end
end

It's possible to accomplish the same thing as unix.mapshared() using files and unix.fcntl() advisory locks. However this goes significantly faster. For example, that's what SQLite does and we recommend using SQLite for IPC in redbean. But, if your app has thousands of forked processes fighting for a file lock you might need something lower level than file locks, to implement things like throttling. Shared memory is a good way to do that since there's nothing that's faster.

The size parameter needs to be a multiple of 8. The returned memory is zero initialized. When allocating shared memory, you should try to get as much use out of it as possible, since the overhead of allocating a single shared mapping is 500 words of resident memory and 8000 words of virtual memory. It's because the Cosmopolitan Libc mmap() granularity is 2**16.

This system call does not fail. An exception is instead thrown if sufficient memory isn't available.

unix.Memory object

unix.Memory encapsulates memory that's shared across fork() and this module provides the fundamental synchronization primitives. These objects are created by unix.mapshared().

Redbean memory maps may be used in two ways:

  1. as an array of bytes a.k.a. a string
  2. as an array of words a.k.a. integers

They're aliased, union, or overlapped views of the same memory. For example if you write a string to your memory region, you'll be able to read it back as an integer.

Reads, writes, and word operations will throw an exception if a memory boundary error or overflow occurs.

unix.Memory:read([offset:int[, bytes:int]])
└─→ str
Reads bytes from memory region

offset is the starting byte index from which memory is copied, which defaults to zero.

If bytes is none or nil, then the nul-terminated string at `offset` is returned. You may specify bytes to safely read binary data.

This operation happens atomically. Each shared mapping has a single lock which is used to synchronize reads and writes to that specific map. To make it scale, create additional maps.

unix.Memory:write([offset:int,] data:str[, bytes:int]])
Writes bytes to memory region.

offset is the starting byte index to which memory is copied, which defaults to zero.

If bytes is none or nil, then an implicit nil-terminator will be included after your data so things like json can be easily serialized to shared memory.

This operation happens atomically. Each shared mapping has a single lock which is used to synchronize reads and writes to that specific map. To make it scale, create additional maps.

unix.Memory:load(word_index:int)
└─→ int
Loads word from memory region.

This operation is atomic and has relaxed barrier semantics.

unix.Memory:store(word_index:int, value:int)
Stores word from memory region.

This operation is atomic and has relaxed barrier semantics.

unix.Memory:xchg(word_index:int, value:int)
└─→ int
Exchanges value.

This sets word at word_index to value and returns the value previously held within that word.

This operation is atomic and provides the same memory barrier semantics as the aligned x86 LOCK XCHG instruction.

unix.Memory:cmpxchg(word_index:int, old:int, new:int)
└─→ success:bool, old:int
Compares and exchanges value.

This inspects the word at word_index and if its value is the same as old then it'll be replaced by the value new, in which case true shall be returned alongside old. If a different value was held at word, then false shall be returned along with its value.

This operation happens atomically and provides the same memory barrier semantics as the aligned x86 LOCK CMPXCHG instruction.

unix.Memory:fetch_add(word_index:int, value:int)
└─→ old:int
Fetches then adds value.

This method modifies the word at word_index to contain the sum of its value and the value parameter. This method then returns the value as it existed before the addition was performed.

This operation is atomic and provides the same memory barrier semantics as the aligned x86 LOCK XADD instruction.

unix.Memory:fetch_and(word_index:int, value:int)
└─→ int
Fetches and bitwise ands value.

This operation happens atomically and provides the same memory barrier ordering semantics as its x86 implementation.

unix.Memory:fetch_or(word_index:int, value:int)
└─→ int
Fetches and bitwise ors value.

This operation happens atomically and provides the same memory barrier ordering semantics as its x86 implementation.

unix.Memory:fetch_xor(word_index:int, value:int)
└─→ int
Fetches and bitwise xors value.

This operation happens atomically and provides the same memory barrier ordering semantics as its x86 implementation.

unix.Memory:wait(word_index:int, expect:int[, abs_deadline:int[, nanos:int]])
├─→ 0
├─→ nil, unix.Errno(unix.EINTR)
├─→ nil, unix.Errno(unix.EAGAIN)
└─→ nil, unix.Errno(unix.ETIMEDOUT)
Waits for word to have a different value.

This method asks the kernel to suspend the process until either the absolute deadline expires or we're woken up by another process that calls unix.Memory:wake().

The expect parameter is the value you expect the word to have and this function will return if that's not the case. Please note this parameter doesn't imply the kernel will poll the value for you, and you still need to call wake() when you know the memory's changed.

The default behavior is to wait until the heat death of the universe if necessary. You may alternatively specify an absolute deadline. If it's less than or equal to the value returned by clock_gettime, then this routine is non-blocking. Otherwise we'll block at most until the current time reaches the absolute deadline.

Futexes are supported natively on Linux, FreeBSD, and OpenBSD. When this interface is used on other platforms this method will manually poll the memory location with exponential backoff. Doing this works well enough that we're passing the *NSYNC unit tests, but is not as low latency as having kernel supported futexes.

EINTR if a signal is delivered while waiting on deadline. Callers should use futexes inside a loop that is able to cope with spurious wakeups. We don't actually guarantee the value at word has in fact changed when this returns.

EAGAIN is raised if, upon entry, the word at word_index had a different value than what's specified at expect.

ETIMEDOUT is raised when the absolute deadline expires.

unix.Memory:wake(index:int[, count:int])
└─→ woken:int
Wakes other processes waiting on word.

This method may be used to signal or broadcast to waiters. The count specifies the number of processes that should be woken, which defaults to INT_MAX.

The return value is the number of processes that were actually woken as a result of the system call. No failure conditions are defined.

unix.Dir object

unix.Dir objects are created by opendir() or fdopendir(). The following methods are available:

unix.Dir:close()
├─→ true
└─→ nil, unix.Errno
Closes directory stream object and associated its file descriptor.

This is called automatically by the garbage collector.

This may be called multiple times.

unix.Dir:read()
├─→ name:str, kind:int, ino:int, off:int
└─→ nil
Reads entry from directory stream.

Returns nil if there are no more entries. On error, nil will be returned and errno will be non-nil.

kind can be any of:

Note: This function also serves as the __call metamethod, so that unix.Dir objects may be used as a for loop iterator.

unix.Dir:fd()
├─→ fd:int
└─→ nil, unix.Errno
Returns file descriptor of open directory object.

Returns EOPNOTSUPP if using a /zip/... path. Returns EOPNOTSUPP if using Windows NT.

unix.Dir:tell()
├─→ off:int
└─→ nil, unix.Errno
Returns current arbitrary offset into stream.
unix.Dir:rewind()
Resets stream back to beginning.

unix.Rusage object

unix.Rusage objects are created by wait() or getrusage(). The following accessor methods are available.

unix.Rusage:utime()
└─→ seconds:int, nanos:int
Returns amount of CPU consumed in userspace.

It's always the case that 0 ≤ nanos < 1e9.

On Windows NT this is collected from GetProcessTimes().

unix.Rusage:stime()
└─→ seconds:int, nanos:int
Returns amount of CPU consumed in kernelspace.

It's always the case that 0 ≤ 𝑥 < 1e9.

On Windows NT this is collected from GetProcessTimes().

unix.Rusage:maxrss()
└─→ kilobytes:int
Returns amount of physical memory used at peak consumption.

On Windows NT this is collected from NtProcessMemoryCountersEx::PeakWorkingSetSize / 1024.

unix.Rusage:idrss()
└─→ integralkilobytes:int
Returns integral private memory consumption w.r.t. scheduled ticks.

If you chart memory usage over the lifetime of your process, then this would be the space filled in beneath the chart. The frequency of kernel scheduling is defined as CLK_TCK. Each time a tick happens, the kernel samples your process's memory usage, by adding it to this value. You can derive the average consumption from this value by computing how many ticks are in utime + stime.

Currently only available on FreeBSD and NetBSD.

unix.Rusage:ixrss()
└─→ integralkilobytes:int
Returns integral shared memory consumption w.r.t. scheduled ticks.

If you chart memory usage over the lifetime of your process, then this would be the space filled in beneath the chart. The frequency of kernel scheduling is defined as CLK_TCK. Each time a tick happens, the kernel samples your process's memory usage, by adding it to this value. You can derive the average consumption from this value by computing how many ticks are in utime + stime.

Currently only available on FreeBSD and NetBSD.

unix.Rusage:isrss()
└─→ integralkilobytes:int
Returns integral stack memory consumption w.r.t. scheduled ticks.

If you chart memory usage over the lifetime of your process, then this would be the space filled in beneath the chart. The frequency of kernel scheduling is defined as CLK_TCK. Each time a tick happens, the kernel samples your process's memory usage, by adding it to this value. You can derive the average consumption from this value by computing how many ticks are in utime + stime.

This is only applicable to redbean if its built with MODE=tiny, because redbean likes to allocate its own deterministic stack.

Currently only available on FreeBSD and NetBSD.

unix.Rusage:minflt()
└─→ count:int
Returns number of minor page faults.

This number indicates how many times redbean was preempted by the kernel to memcpy() a 4096-byte page. This is one of the tradeoffs fork() entails. This number is usually tinier, when your binaries are tinier.

Not available on Windows NT.

unix.Rusage:majflt()
└─→ count:int
Returns number of major page faults.

This number indicates how many times redbean was preempted by the kernel to perform i/o. For example, you might have used mmap() to load a large file into memory lazily.

On Windows NT this is NtProcessMemoryCountersEx::PageFaultCount.

unix.Rusage:nswap()
└─→ count:int
Returns number of swap operations.

Operating systems like to reserve hard disk space to back their RAM guarantees, like using a gold standard for fiat currency. When your system is under heavy memory load, swap operations may happen while redbean is working. This number keeps track of them.

Not available on Linux, Windows NT.

unix.Rusage:inblock()
└─→ count:int
Returns number of times filesystem had to perform input.

On Windows NT this is NtIoCounters::ReadOperationCount.

unix.Rusage:oublock()
└─→ count:int
Returns number of times filesystem had to perform output.

On Windows NT this is NtIoCounters::WriteOperationCount.

unix.Rusage:msgsnd()
└─→ count:int
Returns count of ipc messages sent.

Not available on Linux, Windows NT.

unix.Rusage:msgrcv()
└─→ count:int
Returns count of ipc messages received.

Not available on Linux, Windows NT.

unix.Rusage:nsignals()
└─→ count:int
Returns number of signals received.

Not available on Linux.

unix.Rusage:nvcsw()
└─→ count:int
Returns number of voluntary context switches.

This number is a good thing. It means your redbean finished its work quickly enough within a time slice that it was able to give back the remaining time to the system.

unix.Rusage:nivcsw()
└─→ count:int
Returns number of non-consensual context switches.

This number is a bad thing. It means your redbean was preempted by a higher priority process after failing to finish its work, within the allotted time slice.

unix.Stat object

unix.Stat objects are created by stat() or fstat(). The following accessor methods are available.

unix.Stat:size()
└─→ bytes:int
Size of file in bytes.
unix.Stat:mode()
└─→ mode:int
Contains file type and permissions.

For example, 0010644 is what you might see for a file and 0040755 is what you might see for a directory.

To determine the file type:

unix.Stat:uid()
└─→ uid:int
User ID of file owner.
unix.Stat:gid()
└─→ gid:int
Group ID of file owner.
unix.Stat:birthtim()
└─→ unixts:int, nanos:int
File birth time.

This field should be accurate on Apple, Windows, and BSDs. On Linux this is the minimum of atim/mtim/ctim. On Windows NT nanos is only accurate to hectonanoseconds.

Here's an example of how you might print a file timestamp:

st = assert(unix.stat('/etc/passwd'))
unixts, nanos = st:birthtim()
year,mon,mday,hour,min,sec,gmtoffsec = unix.localtime(unixts)
Write('%.4d-%.2d-%.2dT%.2d:%.2d:%.2d.%.9d%+.2d%.2d % {
   year, mon, mday, hour, min, sec, nanos,
   gmtoffsec / (60 * 60), math.abs(gmtoffsec) % 60})
unix.Stat:mtim()
└─→ unixts:int, nanos:int
Last modified time.
unix.Stat:atim()
└─→ unixts:int, nanos:int
Last access time.

Please note that file systems are sometimes mounted with noatime out of concern for i/o performance. Linux also provides O_NOATIME as an option for open().

On Windows NT this is the same as birth time.

unix.Stat:ctim()
└─→ unixts:int, nanos:int
Complicated time.

Means time file status was last changed on UNIX.

On Windows NT this is the same as birth time.

unix.Stat:blocks()
└─→ count512:int
Number of 512-byte blocks used by storage medium.

This provides some indication of how much physical storage a file actually consumes. For example, for small file systems, your system might report this number as being 8, which means 4096 bytes.

On Windows NT, if O_COMPRESSED is used for a file, then this number will reflect the size after compression. you can use:

st = assert(unix.stat("moby.txt"))
print('file size is %d bytes' % {st:size()})
print('file takes up %d bytes of space' % {st:blocks() * 512})
if GetHostOs() == 'WINDOWS' and st:flags() & 0x800 then
   print('thanks to file system compression')
end

To tell if compression is used on a file.

unix.Stat:blksize()
└─→ bytes:int
Block size that underlying device uses.

This field might be of assistance in computing optimal i/o sizes.

Please note this field has no relationship to blocks, as the latter is fixed at a 512 byte size.

unix.Stat:ino()
└─→ inode:int
Inode number.

This can be used to detect some other process used rename() to swap out a file underneath you, so you can do a refresh. redbean does it during each main process heartbeat for its own use cases.

On Windows NT this is set to NtByHandleFileInformation::FileIndex.

unix.Stat:dev()
└─→ dev:int
ID of device containing file.

On Windows NT this is NtByHandleFileInformation::VolumeSerialNumber.

unix.Stat:rdev()
└─→ rdev:int
Information about device type.

This value may be set to 0 or -1 for files that aren't devices, depending on the operating system. unix.major() and unix.minor() may be used to extract the device numbers.

unix.Statfs object

unix.Statfs objects are created by statfs() or fstatfs(). The following accessor methods are available.

unix.Statfs:fstypename()
└─→ str
Type of filesystem.

Here's some examples of likely values:

unix.Statfs:type()
└─→ int
Type of filesystem.

This is a platform-specific magic number. Consider using the unix.Statfs:fstypename() method instead. On Windows, this will actually be a Knuth multiplicative hash of the name.

unix.Statfs:bsize()
└─→ int
Optimal transfer block size.

This field serves two purposes:

  1. It tells you how to chunk i/o operations. For local disks, it'll likely be any value between 512 and 4096 depending on the operating system. For network filesystems it will likely be a much larger value, e.g. 512kb.
  2. It can be multiplied with the fields blocks, bfree, and bavail to obtain a byte count.
unix.Statfs:blocks()
└─→ int
Total data blocks in filesystem.

The size of a block is measured as unix.Statfs:bsize().

unix.Statfs:bfree()
└─→ int
Total free blocks in filesystem.

The size of a block is measured as unix.Statfs:bsize().

unix.Statfs:bavail()
└─→ int
Total free blocks available in filesystem to unprivileged users.

The size of a block is measured as unix.Statfs:bsize().

unix.Statfs:files()
└─→ int
Total file nodes in filesystem.

On Windows this is always the maximum integer value.

unix.Statfs:ffree()
└─→ int
Total free file nodes in filesystem.

On Windows this is always the maximum integer value.

unix.Statfs:fsid()
└─→ int
Filesystem id.
unix.Statfs:namelen()
└─→ int
Maximum length of filename components in bytes.
unix.Statfs:flags()
└─→ int
Filesystem flags.

The following flags are defined:

unix.Statfs:owner()
└─→ int
User id of owner of filesystem mount.

On Linux this is always 0 for root. On Windows this is always 0.

unix.Sigset object

The unix.Sigset class defines a mutable bitset that may currently contain 128 entries. See unix.NSIG to find out how many signals your operating system actually supports.

unix.Sigset(sig:int, ...)
└─→ unix.Sigset
Constructs new signal bitset object.
unix.Sigset:add(sig:int)
Adds signal to bitset.
unix.Sigset:remove(sig:int)
Removes signal from bitset.
unix.Sigset:fill()
Sets all bits in signal bitset to true.
unix.Sigset:clear()
Sets all bits in signal bitset to false.
unix.Sigset:contains(sig:int)
└─→ bool
Returns true if sig is member of signal bitset.
unix.Sigset:__repr()
unix.Sigset:__tostring()
Returns Lua code string that recreates object.

unix signal magic numbers

unix.SIGINT
Terminal CTRL-C keystroke.
unix.SIGQUIT
Terminal CTRL-\ keystroke.
unix.SIGHUP
Terminal hangup or daemon reload; auto-broadcasted to process group.
unix.SIGILL
Illegal instruction.
unix.SIGTRAP
INT3 instruction.
unix.SIGABRT
Process aborted.
unix.SIGBUS
Valid memory access that went beyond underlying end of file.
unix.SIGFPE
Illegal math.
unix.SIGKILL
Terminate with extreme prejudice.
unix.SIGUSR1
Do whatever you want.
unix.SIGUSR2
Do whatever you want.
unix.SIGSEGV
Invalid memory access.
unix.SIGPIPE
Write to closed file descriptor.
unix.SIGALRM
Sent by setitimer().
unix.SIGTERM
Terminate.
unix.SIGCHLD
Child process exited or terminated and is now a zombie (unless this is SIG_IGN or SA_NOCLDWAIT) or child process stopped due to terminal i/o or profiling/debugging (unless you used SA_NOCLDSTOP)
unix.SIGCONT
Child process resumed from profiling/debugging.
unix.SIGSTOP
Child process stopped due to profiling/debugging.
unix.SIGTSTP
Terminal CTRL-Z keystroke.
unix.SIGTTIN
Terminal input for background process.
unix.SIGTTOU
Terminal output for background process.
unix.SIGXCPU
CPU time limit exceeded.
unix.SIGXFSZ
File size limit exceeded.
unix.SIGVTALRM
Virtual alarm clock.
unix.SIGPROF
Profiling timer expired.
unix.SIGWINCH
Terminal resized.
unix.SIGPWR
Not implemented in most community editions of system five.

unix.Errno object

This object is returned by system calls that fail. We prefer returning an object because for many system calls, an error is part their normal operation. For example, it's often desirable to use the errno() method when performing a read() to check for EINTR.

unix.Errno:errno()
└─→ errno:int
Returns error magic number.

The error number is always different for different platforms. On UNIX systems, error numbers occupy the range [1,127] in practice. The System V ABI reserves numbers as high as 4095. On Windows NT, error numbers can go up to 65535.

unix.Errno:winerr()
└─→ errno:int
Returns Windows error number.

On UNIX systems this is always 0. On Windows NT this will normally be the same as errno(). Because Windows defines so many error codes, there's oftentimes a multimapping between its error codes and System Five. In those cases, this value reflect the GetLastError() result at the time the error occurred.

unix.Errno:name()
└─→ symbol:str
Returns string of symbolic name of System Five error code.

For example, this might return "EINTR".

unix.Errno:call()
└─→ symbol:str
Returns name of system call that failed.

For example, this might return "read" if read() was what failed.

unix.Errno:doc()
└─→ symbol:str
Returns English string describing System Five error code.

For example, this might return "Interrupted system call".

unix.Errno:__tostring()
└─→ str
Returns verbose string describing error.

Different information components are delimited by slash.

For example, this might return "EINTR/4/Interrupted system call".

On Windows NT this will include additional information about the Windows error (including FormatMessage() output) if the WIN32 error differs from the System Five error code.

unix error magic numbers

unix.EINVAL
Invalid argument.

Raised by [pretty much everything].

unix.ENOSYS
System call not available on this platform.

On Windows this is raised by chroot, setuid, setgid, getsid, setsid, and others we're doing our best to document.

unix.ENOENT
No such file or directory.

Raised by access, bind, chdir, chmod, chown, chroot, clock_getres, execve, opendir, link, mkdir, mknod, open, readlink, rename, rmdir, stat, symlink, truncate, unlink, utime, utimensat.

unix.ENOTDIR
Not a directory. This means that a directory component in a supplied path *existed* but wasn't a directory. For example, if you try to open("foo/bar") and foo is a regular file, then ENOTDIR will be returned.

Raised by open, access, chdir, chroot, execve, link, mkdir, mknod, opendir, readlink, rename, rmdir, stat, symlink, truncate, unlink, utimensat, bind, chmod, chown, fcntl, futimesat.

unix.EINTR
The greatest of all errnos; crucial for building real time reliable software.

Raised by accept, clock_nanosleep, close, connect, dup, fcntl, flock, getrandom, nanosleep, open, pause, poll, ptrace, read, recv, select, send, sigsuspend, sigwaitinfo, truncate, wait, write.

unix.EIO
Raised by access, acct, chdir, chmod, chown, chroot, close, copy_file_range, execve, fallocate, fsync, ioperm, link, madvise, mbind, ptrace, read, readlink, sendfile, statfs, symlink, sync_file_range, truncate, unlink, write.
unix.ENXIO
No such device or address.

Raised by lseek, open, prctl.

unix.E2BIG
Argument list too long.

Raised by execve.

unix.ENOEXEC
Exec format error.

Raised by execve.

unix.ECHILD
No child process.

Raised by wait.

unix.ESRCH
No such process.

Raised by getpriority, getrlimit, getsid, ioprio_set, kill, setpgid, utimensat.

unix.EBADF
Bad file descriptor; cf. EBADFD.

Raised by accept, access, bind, chdir, chmod, chown, close, connect, copy_file_range, dup, fcntl, flock, fsync, futimesat, opendir, getpeername, getsockname, getsockopt, ioctl, link, listen, llseek, lseek, mkdir, mknod, mmap, open, prctl, read, readahead, readlink, recv, rename, select, send, shutdown, splice, stat, symlink, sync, sync_file_range, truncate, unlink, utimensat, write.

unix.EAGAIN
Resource temporarily unavailable (e.g. SO_RCVTIMEO expired, too many processes, too much memory locked, read or write with O_NONBLOCK needs polling, etc.).

Raised by accept, connect, fcntl, fork, getrandom, mincore, mlock, mmap, mremap, poll, read, select, send, setresuid, setreuid, setuid, sigwaitinfo, splice, tee, timer_create, kill, write,

unix.EPIPE
Broken pipe.

This happens when you try to write data to a subprocess via a pipe but the reader end has already closed, possibly because the process died. Normally i/o routines only return this if SIGPIPE doesn't kill the process. Unlike default UNIX programs, redbean currently ignores SIGPIPE by default, so this error code is a distinct possibility when pipes or sockets are being used.

Returned by write, send.

unix.ENAMETOOLONG
Filename too long. Cosmopolitan Libc currently defines PATH_MAX as 1024 characters. On UNIX that limit should only apply to system call wrappers like realpath. On Windows NT it's observed by all system calls that accept a pathname.

Raised by access, bind, chdir, chmod, chown, chroot, execve, gethostname, link, mkdir, mknod, open, readlink, rename, rmdir, stat, symlink, truncate, unlink, utimensat.

unix.EACCES
Permission denied.

Raised by access, bind, chdir, chmod, chown, chroot, clock_getres, connect, execve, fcntl, getpriority, link, mkdir, mknod, mmap, mprotect, msgctl, open, prctl, ptrace, readlink, rename, rmdir, semget, send, setpgid, socket, stat, symlink, truncate, unlink, uselib, utime, utimensat.

unix.ENOMEM
We require more vespene gas.

Raised by access, bind, chdir, chmod, chown, chroot, clone, copy_file_range, execve, fork, getgroups, getrlimit, ioperm, link, mbind, mincore, mkdir, mknod, mlock, mmap, mprotect, mremap, msync, open, poll, readlink, recv, rename, rmdir, select, send, sigaltstack, splice, stat, symlink, sync_file_range, tee, unlink.

unix.EPERM
Operation not permitted.

Raised by accept, chmod, chown, chroot, copy_file_range, execve, fcntl, getdomainname, gethostname, getrlimit, getsid, ioperm, iopl, kill, link, mkdir, mknod, nice, open, rename, rmdir, sched_setaffinity, sched_setscheduler, seteuid, setfsgid, setfsuid, setgid, setns, setpgid, setresuid, setreuid, setsid, setuid, setup, setxattr, sigaltstack, spu_create, stime, symlink, syslog, truncate, unlink, utime, utimensat, write.

unix.ENOTBLK
Block device required.

Raised by umount.

unix.EBUSY
Device or resource busy.

Raised by dup, fcntl, msync, prctl, ptrace, rename, rmdir.

unix.EEXIST
File exists.

Raised by link, mkdir, mknod, mmap, open, rename, rmdir, symlink.

unix.EXDEV
Improper link.

Raised by copy_file_range, link, rename.

unix.ENODEV
No such device.

Raised by mmap, open.

unix.EISDIR
Is a directory.

Raised by copy_file_range, execve, open, read, rename, truncate, unlink.

unix.ENFILE
Too many open files in system.

Raised by accept, execve, mmap, open, pipe, socket, socketpair.

unix.EMFILE
Too many open files.

Raised by accept, dup, execve, fcntl, open, pipe, socket, socketpair.

unix.ENOTTY
Inappropriate i/o control operation.

Raised by ioctl.

unix.ETXTBSY
Won't open executable that's executing in write mode.

Raised by access, copy_file_range, execve, mmap, open, truncate.

unix.EFBIG
File too large.

Raised by copy_file_range, open, truncate, write.

unix.ENOSPC
No space left on device.

Raised by copy_file_range, fsync, link, mkdir, mknod, open, rename, symlink, sync_file_range, write.

unix.EDQUOT
Disk quota exceeded.

Raised by link, mkdir, mknod, open, rename, symlink, write.

unix.ESPIPE
Invalid seek.

Raised by lseek, splice, sync_file_range.

unix.EROFS
Read-only filesystem.

Raised by access, bind, chmod, chown, link, mkdir, mknod, open, rename, rmdir, symlink, truncate, unlink, utime, utimensat.

Too many links;

raised by link, mkdir, rename.

unix.ERANGE
Result too large.

Raised by prctl.

unix.EDEADLK
Resource deadlock avoided.

Raised by fcntl.

unix.ENOLCK
No locks available.

Raised by fcntl, flock.

unix.ENOTEMPTY
Directory not empty.

Raised by rmdir.

unix.ELOOP
Too many levels of symbolic links.

Raised by access, bind, chdir, chmod, chown, chroot, execve, link, mkdir, mknod, open, readlink, rename, rmdir, stat, symlink, truncate, unlink, utimensat.

unix.ETIME
Timer expired.

Raised by connect.

unix.EPROTO
Raised by accept, connect, socket, socketpair.
unix.EOVERFLOW
Raised by copy_file_range, lseek, mmap, open, stat.
unix.ENOTSOCK
Not a socket.

Raised by accept, bind, connect, getpeername, getsockname, getsockopt, listen, recv, send, shutdown.

unix.EDESTADDRREQ
Destination address required.

Raised by send, write.

unix.EMSGSIZE
Message too long.

Raised by send.

unix.EPROTOTYPE
Protocol wrong type for socket.

Raised by connect.

unix.ENOPROTOOPT
Protocol not available.

Raised by getsockopt, accept.

unix.EPROTONOSUPPORT
Protocol not supported.

Raised by socket, socketpair.

unix.ESOCKTNOSUPPORT
Socket type not supported.
unix.ENOTSUP
Operation not supported.

Raised by chmod, clock_getres, clock_nanosleep, timer_create.

unix.EOPNOTSUPP
Socket operation not supported.

Raised by accept, listen, mmap, prctl, readv, send, socketpair.

unix.EPFNOSUPPORT
Protocol family not supported.
unix.EAFNOSUPPORT
Address family not supported.

Raised by connect, socket, socketpair.

unix.EADDRINUSE
Address already in use.

Raised by bind, connect, listen.

unix.EADDRNOTAVAIL
Address not available.

Raised by bind, connect.

unix.ENETDOWN
Network is down.

Raised by accept.

unix.ENETUNREACH
Host is unreachable.

Raised by accept, connect.

unix.ENETRESET
Connection reset by network.
unix.ECONNABORTED
Connection reset before accept.

Raised by accept.

unix.ECONNRESET
Connection reset by client.

Raised by send.

unix.ENOBUFS
No buffer space available;

Raised by getpeername, getsockname, send.

unix.EISCONN
Socket is connected.

Raised by connect, send.

unix.ENOTCONN
Socket is not connected.

Raised by getpeername, recv, send, shutdown.

unix.ESHUTDOWN
Cannot send after transport endpoint shutdown; note that shutdown write is an EPIPE.
unix.ETOOMANYREFS
Too many references: cannot splice.

Raised by sendmsg.

unix.ETIMEDOUT
Connection timed out.

Raised by connect.

unix.ECONNREFUSED
System-imposed limit on the number of threads was encountered.

Raised by connect, listen, recv.

unix.EHOSTDOWN
Host is down.

Raised by accept.

unix.EHOSTUNREACH
Host is unreachable.

Raised by accept.

unix.EALREADY
Connection already in progress.

Raised by connect, send.

unix.ENODATA
No message is available in xsi stream or named pipe is being closed; no data available; barely in posix; returned by ioctl; very close in spirit to EPIPE?

unix miscellaneous magic numbers

unix.ARG_MAX
Returns maximum length of arguments for new processes.

This is the character limit when calling execve(). It's the sum of the lengths of argv and envp including any nul terminators and pointer arrays. For example to see how much your shell envp uses

$ echo $(($(env | wc -c) + 1 + ($(env | wc -l) + 1) * 8))
758

POSIX mandates this be 4096 or higher. On Linux this it's 128*1024. On Windows NT it's 32767*2 because CreateProcess lpCommandLine and environment block are separately constrained to 32,767 characters. Most other systems define this limit much higher.

unix.BUFSIZ
Returns default buffer size.

The UNIX module does not perform any buffering between calls.

Each time a read or write is performed via the UNIX API your redbean will allocate a buffer of this size by default. This current default would be 4096 across platforms.

unix.CLK_TCK
Returns the scheduler frequency.

This is granularity at which the kernel does work. For example, the Linux kernel normally operates at 100hz so its CLK_TCK will be 100.

This value is useful for making sense out of unix.Rusage data.

unix.PIPE_BUF
Returns maximum size at which pipe i/o is guaranteed atomic.

POSIX requires this be at least 512. Linux is more generous and allows 4096. On Windows NT this is currently 4096, and it's the parameter redbean passes to CreateNamedPipe().

unix.PATH_MAX
Returns maximum length of file path.

This applies to a complete path being passed to system calls.

POSIX.1 XSI requires this be at least 1024 so that's what most platforms support. On Windows NT, the limit is technically 260 characters. Your redbean works around that by prefixing //?/ to your paths as needed. On Linux this limit will be 4096, but that won't be the case for functions such as realpath that are implemented at the C library level; however such functions are the exception rather than the norm, and report enametoolong(), when exceeding the libc limit.

unix.NAME_MAX
Returns maximum length of file path component.

POSIX requires this be at least 14. Most operating systems define it as 255. It's a good idea to not exceed 253 since that's the limit on DNS labels.

unix.NSIG
Returns maximum number of signals supported by underlying system.

The limit for unix.Sigset is 128 to support FreeBSD, but most operating systems define this much lower, like 32. This constant reflects the value chosen by the underlying operating system.

constants

kLogDebug
Integer for debug logging level. See Log.
kLogVerbose
Integer for verbose logging level, which is less than kLogDebug. See Log.
kLogInfo
Integer for info logging level, which is less than kLogVerbose. See Log.
kLogWarn
Integer for warn logging level, which is less than kLogVerbose. See Log.
kLogError
Integer for error logging level, which is less than kLogWarn. See Log.
kLogFatal
Integer for fatal logging level, which is less than kLogError. See Log. Logging anything at this level will result in a backtrace and process exit.

operations

You can have redbean run as a daemon by doing the following:

redbean.com -vv -d -L redbean.log -P redbean.pid
kill -TERM $(cat redbean.pid) # 1x: graceful shutdown
kill -TERM $(cat redbean.pid) # 2x: forceful shutdown

It's possible to modify global interpreter state later on in the server's lifecycle. When running in daemon mode, using kill -HUP $(pidof redbean.com) will instruct redbean to run the code in .reload.lua from the main process, will will be lazily propagated to client connections.

You can modify the zip while redbean is running. The zip command by default will do this by replacing the inode. redbean will detect the changed inode within a second and broadcast SIGUSR1 to the process group so the new assets get indexed as soon as possible. It's also possible to modify the executable assets in place, while the executable is running. If you do that, then you need to be careful to not disturb the local file and central directory. What you would do instead is append changed or new files. Then append a new central directory, along with a new end of central directory record. Finally, memset(0) the old end of central directory record. redbean will detect it's gone and reindex. You can even modify local files in place too. The way you would do that is by clearing the PK♥♦ magic marker while the file memory is being mutated, and then putting it back. Any requests that arrive during the modification will result in a 503 Service Unavailable so your load balancer can failover.

redbean will grow to whatever number of processes your system limits and tcp stack configuration allow. Once functions like fork() and accept() start to fail, redbean will enter "meltdown mode" where it interrupts worker processes to immediately close idling and lagging connections. redbean may need to drop requests by sending 503 Service Unavailable until congestion subsides. So be sure your load balancer is configured to immediately failover to another instance in such cases.

There's a 64kb limit on request message size, where the header portion is further limited to 32kb. We do that to guarantee processes stay tiny. You can tune it using the -M flag. redbean spawns a process for each connection. redbean needs about 200kb of RAM per worker on average. Clients are encouraged to pipeline HTTP requests within the same connection.

redbean rejects requests for hidden files, i.e. any path containing the substring /. and requests with denormalized paths, e.g. /../../etc/passwd are also categorically rejected. Furthermore, redbean won't service requests that come in more than 32 fragments. Those few restrictions aside, redbean generally aims to follow Postel's maxim in the sense that it's liberal in what it accepts but conservative in what it sends.

If you want Rust-like promises then redbean can be compiled with ASAN memory safety. Cosmopolitan Libc has the only open source implementation of the Address Sanitizer runtime that's intended for production use. It causes redbean worker processes to crash and log a report should a bug like a buffer overrun, use-after-free, or stack smash occur. This has a marginal impact on performance. It can be useful in environments where interruptions in a service are more desirable than risking the system being compromised.

security

redbean doesn't secure your computer. It does however provide tools we hope will help you do it yourself. For further details on why things need to be this way, please see the disclaimer in the ISC license.

Some computing environments rely on physical security and redbean is a good fit for that. For example, if you need to run a web app on an air-gapped computer running an old version of some other operating system that can't be upgraded, then you load your redbean off a thumb drive provided that the system was installed after the year 2007 or more specifically runs x86_64 with Linux 2.6.18+ or Windows Vista+.

Some environments require that security be provided using existing infrastructure like SSL frontends. In that case, the "redbean-unsecure" download link might be the right choice for you, since it's a special build of redbean that leaves out the security code. That way, you can bolt the security on separately using a tool like stunnel.

ssl

redbean provides integrated SSL support based on MbedTLS. It's configured to offer 128 bits of security with modern clients, but will fall back to at minimum 112 bits of security depending on the preferences of the client. Both are secure based on public knowledge until 2030 according to NIST. If you'd rather restrict yourself to just 150+ bits of security but with the tradeoff of dropping support for old Internet Explorer and making embedded clients less happy, then pass the -B flag, which'll restrict redbean to a very short list of protocols, algorithms, and parameters that the NSA, NIST, and IANA all agree upon.

redbean's SSL implementation is tuned for performance. It uses hardware algorithms when available such as AES-NI, SHA-NI, and RDRAND. redbean does not use costly hardnening measures specific only to legacy clients like Internet Explorer if they increase denial of service risk for the server as a whole.

redbean is tuned for ease of use. Your redbean uses a protocol polyglot for serving HTTP and HTTPS on the same port numbers. Both the TLS hello and the SSLv2 hello are accepted, even though only TLS is supported. For example, both of these are valid:

http://127.0.0.1:8080/
https://127.0.0.1:8080/

By default, your redbean will automatically generate ephemeral self-signed ECDSA and RSA serving certificates. This causes browser warnings. The simplest option for making the warning go away is to give redbean a key signing key (KSK).

openssl req -x509 -newkey rsa:2048 \
  -keyout .ca.key -out .ca.crt -days 6570 -nodes \
  -subj '/C=US/ST=CA/O=Jane Doe/CN=My Root CA 1' \
  -addext 'keyUsage = critical,cRLSign,keyCertSign'
sudo ./redbean.com -C ca.crt -K .ca.key -p 80 -p 443

Your SSL root can then be installed on client machines as follows:

# linux
sudo cp ca.crt /usr/local/share/ca-certificates
sudo update-ca-certificates

# macos
sudo security add-trusted-cert -d -r trustRoot \
  -k /Library/Keychains/System.keychain ca.crt

# windows
certutil -addstore -f "ROOT" ca.crt

# notes
# firefox is special you have to use its settings

If your goal is to make SSL deploys easy, then it's possible to put the KSK inside the redbean.com file using the InfoZIP program. Be sure the key is a hidden file. It can be loaded using your .init.lua script with the LoadAsset, ProgramCertificate, and ProgramPrivateKey APIs. Please note, this is just an example of what you could do; we don't claim it's what you should do.

For a public-facing online service, the simplest way to use SSL is with Let's Encrypt. Let's say you're migrating from NGINX. In that case you'll likely want something like the following:

# commands subject to public monitoring
certbot certonly --nginx --key-type ecdsa \
  --cert-name redbean-ecdsa -d redbean.dev -d www.redbean.dev
certbot certonly --nginx --key-type rsa \
  --cert-name redbean-rsa -d redbean.dev -d www.redbean.dev

You can then program /var/www/html/.init.lua as such:

ProgramPrivateKey(Slurp('/etc/letsencrypt/live/redbean-ecdsa/privkey.pem'))
ProgramCertificate(Slurp('/etc/letsencrypt/live/redbean-ecdsa/fullchain.pem'))
ProgramPrivateKey(Slurp('/etc/letsencrypt/live/redbean-rsa/privkey.pem'))
ProgramCertificate(Slurp('/etc/letsencrypt/live/redbean-rsa/fullchain.pem'))
if IsDaemon() then
   ProgramUid(33)  # see `vipw` to get appropriate number
   ProgramGid(33)  # see `vigr` to get appropriate number
   ProgramPort(80)
   ProgramPort(443)
   ProgramLogPath('/var/log/redbean.log')
   ProgramPidPath('/var/run/redbean.pid')
end
function OnHttpRequest()
   path = GetPath()
   if path == '/favicon.ico' or
      path == '/site.webmanifest' or
      path == '/favicon-16x16.png' or
      path == '/favicon-32x32.png' or
      path == '/apple-touch-icon' then
      SetLogLevel(kLogWarn)
   end
   Route()
   SetHeader('Content-Language', 'en-US')
end

You'd then run redbean as follows:

redbean.com -dD /var/www/html

You can load as many public and private keys as you want. They can be specified as pem, der, concatenated ascii, bundles, or chains. If you don't specify specific chains then redbean will automatically infer it based on SUBJECT → ISSUER relationships. Your redbean won't serve the self-signed root certificate at the end of the chain where self-signed is defined as SUBJECT == ISSUER. Otherwise you can control when chains terminate by setting the max length constraint to zero.

Your redbean supports SSL virtual hosting. 99.76% of TLS clients send a Server Name Indicator (SNI), which is matched against DNS or IPs in Subject Alternative Names (SAN) or the Common Name (CN) of subject if SAN isn't used. This means you don't need to reveal your whole domain portfolio to each client just to have ssl. You can just use different certificates for each domain if you choose to do so. If redbean can't find an appropriate match, then the first certificate will be chosen.

SSL layer client verification is unusual, but some options are:

  1. Pass the -j to enable verification of HTTPS clients. Clients are verified based on the SSL roots you've provided. Those can be installed via the Lua API or placed in the ZIP executable folder usr/share/ssl/root.
  2. You can use preshared keys via the Lua API. Please note that PSK isn't mutually exclusive with normal PKI. Please read the relevant API documentation.

SSL verbosity is controlled as follows for troubleshooting:

-V       log ssl errors
-VV      log ssl state changes too
-VVV     log ssl informational messages too
-VVVV    log ssl verbose details too

That's in addition to existing flags like -vvvm.

ddos protection

Once you have SSL setup, the threats you're most likely to face are distributed denial of service attacks, or DDOS for short. redbean defeats this form of abuse using token buckets and blackholed. That lets us protect not only your web app, but the kernel itself.

The example below shows how you can configure a secure reverse proxy. Each block of 256 IPs is allowed to send one message per second, on average, with bursts of up to ~100 messages per second allowed for a short period of time. Once an IP exhausts its tokens, it gets blocked by the system firewall.

-- we'll be reverse proxying to a server running here
BACKEND = '127.0.0.1:8080'

-- put your home or office public ip here
-- so you don't accidentally ban yourself
-- you don't want to lose your ssh access
ProgramTrustedIp(ParseIp('1.2.3.4'), 32)

RELAY_HEADERS_TO_CLIENT = {
    'Access-Control-Allow-Origin',
    'Cache-Control',
    'Connection',
    'Content-Type',
    'Last-Modified',
    'Referrer-Policy',
}

function OnServerStart()
    -- enables ddos protection
    ProgramTokenBucket()
    -- place limit on number of forked workers
    -- this determines the point at which "meltdown" happens
    assert(unix.setrlimit(unix.RLIMIT_NPROC, 1000, 1000))
end

function OnWorkerStart()
    -- set limits on memory and cpu just in case
    assert(unix.setrlimit(unix.RLIMIT_RSS, 2*1024*1024))
    assert(unix.setrlimit(unix.RLIMIT_CPU, 2))
    -- we don't need filesystem access
    assert(unix.unveil(nil, nil))
    -- we only need minimal system calls
    assert(unix.pledge("stdio inet unix", nil, unix.PLEDGE_PENALTY_RETURN_EPERM))
end

function OnHttpRequest()
    local ip = GetClientAddr()
    if not IsTrustedIp(ip) then
        local tok = AcquireToken(ip)
        if tok < 2 then
            if Blackhole(ip) then
                Log(kLogWarn, "banned %s" % {FormatIp(ip)})
            else
                Log(kLogWarn, "failed to ban %s" % {FormatIp(ip)})
            end
        end
        if tok < 30 then
            ServeError(429)
            SetHeader('Connection', 'close')
            Log(kLogWarn, "warned %s who has %d tokens" % {FormatIp(ip), tok})
            return
        end
    end
    local url = 'http://' .. BACKEND .. EscapePath(GetPath())
    local name = GetParam('name')
    if name then
        url = url .. '?name=' .. EscapeParam(name)
    end
    local status, headers, body =
        Fetch(url,
              {method = GetMethod(),
               headers = {
                   ['Accept'] = GetHeader('Accept'),
                   ['CF-IPCountry'] = GetHeader('CF-IPCountry'),
                   ['If-Modified-Since'] = GetHeader('If-Modified-Since'),
                   ['Referer'] = GetHeader('Referer'),
                   ['Sec-CH-UA-Platform'] = GetHeader('Sec-CH-UA-Platform'),
                   ['User-Agent'] = GetHeader('User-Agent'),
                   ['X-Forwarded-For'] = FormatIp(ip)}})
    if status then
        SetStatus(status)
        for k,v in pairs(RELAY_HEADERS_TO_CLIENT) do
            SetHeader(v, headers[v])
        end
        Write(body)
    else
        local err = headers
        Log(kLogError, "proxy failed %s" % {err})
        ServeError(503)
    end
end

You'll notice in the code above that we're doing token bucket logic by hand. redbean actually does this automatically in the core code, but it's only enforced on accept()'d connections. If you aren't specifying SetHeader('Connection', 'close') on each response, then a each of those connections can technically send infinite HTTP messages. redbean core doesn't police messages, which is left to your Lua code to decide.

The trick blackholed uses is the raw prerouting table, which drops traffic before the kernel's conntrack mechanism kicks in. This works out of the box with zero configuration on Linux if you download the blackholed daemon from the supplementary tools section above. Its purpose is to permit unprivileged system accounts to safely ban IPv4 addresses via a UNIX named socket. redbean integrates with this service automatically when ProgramTokenBucket() is used.

sudo ./blackholed.com -d

Please note you can't use Cloudflare if you use redbean's DDOS protection. If someone launches a DDOS attack against your server through Cloudflare, then redbean can't blackhole the IPs of the attackers because your server isn't talking to the attackers. We recommend using redbean's DDOS protection instead, because it's done a much better job fending off attacks than Cloudflare was able to do, for the services we operate online.

memory safety

redbean provides hardened ASAN (Address Sanitizer) builds that proactively guard against any potential memory weaknesses that may be discovered, such as buffer overruns, use after free, etc. MODE=asan is recomended when serving on the public Internet.

sandboxing

redbean also supports robust sandboxing on Linux Kernel 5.13+ and OpenBSD. The recommended way to harden your redbean is to call the pledge() and unveil() functions.

Here's an example. If you have a SQLite app, then the key to using features like pledge() is to perform the permissions-demanding activities before calling pledge(). In this case, that would be connecting to the DB first.

function OnWorkerStart()
    db = sqlite3.open("db.sqlite3")
    db:busy_timeout(1000)
    db:exec[[PRAGMA journal_mode=WAL]]
    db:exec[[PRAGMA synchronous=NORMAL]]
    db:exec[[SELECT x FROM warmup WHERE x = 1]]
    assert(unix.setrlimit(unix.RLIMIT_RSS, 100 * 1024 * 1024))
    assert(unix.setrlimit(unix.RLIMIT_CPU, 4))
    assert(unix.unveil("/var/tmp", "rwc"))
    assert(unix.unveil("/tmp", "rwc"))
    assert(unix.unveil(nil, nil))
    assert(unix.pledge("stdio flock rpath wpath cpath", nil,
                       unix.PLEDGE_PENALTY_RETURN_EPERM))
end

What makes this technique interesting is redbean doesn't have file system access to the database file, and instead uses an inherited file descriptor that was opened beforehand. With SQLite the tmp access is only needed to support things like covering indexes. The -Z flag is also helpful to see where things go wrong, so you know which promises are needed to support your use case.

pledge() will work on all Linux kernels since RHEL6 since it uses SECCOMP BPF filtering. On the other hand, unveil() requires Landlock LSM which was only introduced in 2021. If you need unveil() then be sure to test the restrictions work. Most environments don't support unveil(), so it's designed to be a no-op in unsupported environments.

Alternatively, there's CLI flags which make it simple to get started:

-S (online policy)
This causes unix.pledge("stdio rpath inet dns id") to be called on workers after fork() is called. This permits read-only operations and APIs like Fetch() that let workers send and receive data with private and public Internet hosts. Access to the unix module is somewhat restricted, disallowing its more powerful APIs like exec.
-SS (offline policy)
This causes unix.pledge("stdio rpath id") to be called on workers after after fork() is called. This prevents workers from talking to the network (other than the client) and allows read-only file system access (e.g. -D DIR flag). The `id` group helps you to call other functions important to redbean security, such as the unix.setrlimit() function.
-SSS (contained policy)
This causes unix.pledge("stdio") to be called on workers after after fork() is called. This prevents workers from communicating with the network (other than the client connection) and prevents file system access (with some exceptions like logging). Redbean should only be able to serve from its own zip file in this mode. Lua script access to the unix module is highly restricted.

Unlike the unix.pledge() function, these sandboxing flags use a more permissive policy on Linux. Rather than killing the process, they'll cause system calls to fail with EPERM instead. Therefore these flags should be gentler when you want security errors to be recoverable.

legal

redbean contains software licensed ISC, MIT, BSD-2, BSD-3, zlib. The transitive closure of legal notices can be found inside the binary structure. We put the licenses inside the binary we believe that this satisfactorily automates legal compliance, for the redbean project and anyone who uses it.

benchmark

# Note: Benchmarked on an AMD Ryzen Threadripper PRO 7995WX 96-Cores
# Note: Use ./redbean.com -s
$ wrk --latency -t 256 -c 256 -H 'Accept-Encoding: gzip' \
    http://127.0.0.1:8080/favicon.ico
Running 10s test @ http://127.0.0.1:8080/tool/net/demo/index.html
  256 threads and 256 connections
  Thread Stats   Avg      Stdev     Max   +/- Stdev
    Latency     3.92ms   35.76ms 829.17ms   98.34%
    Req/Sec    25.72k     8.08k   39.35k    62.68%
  Latency Distribution
     50%   29.00us
     75%   42.00us
     90%   63.00us
     99%  159.62ms
  35413605 requests in 6.60s, 210.36GB read
Requests/sec: 5366223.78
Transfer/sec:     31.88GB

[United States of Lemuria - two dollar bill - all debts public and primate]

Funding for redbean was crowdsourced from Justine Tunney's GitHub sponsors and Patreon subscribers. Your support is what makes projects like redbean possible. Thank you.

see also

old versions