mirrors/cosmopolitan
1
0
Fork 0
mirror of https://github.com/jart/cosmopolitan.git synced 2025-01-31 11:37:35 +00:00
Code Issues Projects Releases Packages Wiki Activity Actions
cosmopolitan/tool/net/help.txt
Justine Tunney 625aa365f1 Refactor pledge() to be more configurable 
The earlier iterations did too much guesswork when it came to things
like stderr logging and syscall origin verification. This change will
make things more conformant to existing practices. The __pledge_mode
extension now can be configured in a better way.

There's also a new `-q` flag added to pledge.com, e.g.

    o//tool/build/pledge.com -qv. ls

Is a good way to disable warnings about `tty` access attempts.
2022-08-11 11:35:30 -07:00

5282 lines
194 KiB
Text
Raw Blame History

This file contains ambiguous Unicode characters

This file contains Unicode characters that might be confused with other characters. If you think that this is intentional, you can safely ignore this warning. Use the Escape button to reveal them.

SYNOPSIS
redbean.com [-?BVabdfghjkmsuvz] [-p PORT] [-D DIR] [-- SCRIPTARGS...]
DESCRIPTION
redbean - single-file distributable web server
OVERVIEW
redbean makes it possible to share web applications that run offline
as a single-file Actually Portable Executable PKZIP archive which
contains your assets. All you need to do is download the redbean.com
program below, change the filename to .zip, add your content in a zip
editing tool, and then change the extension back to .com.
redbean can serve 1 million+ gzip encoded responses per second on a
cheap personal computer. That performance is thanks to zip and gzip
using the same compression format, which enables kernelspace copies.
Another reason redbean goes fast is that it's a tiny static binary,
which makes fork memory paging nearly free.
redbean is also easy to modify to suit your own needs. The program
itself is written as a single .c file. It embeds the Lua programming
language and SQLite which let you write dynamic pages.
FEATURES
- Lua v5.4
- SQLite 3.35.5
- TLS v1.2 / v1.1 / v1.0
- HTTP v1.1 / v1.0 / v0.9
- Chromium-Zlib Compression
- Statusz Monitoring Statistics
- Self-Modifying PKZIP Object Store
- Linux + Windows + Mac + FreeBSD + OpenBSD + NetBSD
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 eval Lua code in arg
-F eval Lua code in file
-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]
-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-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
────────────────────────────────────────────────────────────────────────────────
USAGE
This executable is also a ZIP file that contains static assets.
You can run redbean interactively in your terminal as follows:
./redbean.com -vvvmbag # starts server verbosely
open http://127.0.0.1:8080/ # shows zip listing page
CTRL-C # 1x: graceful shutdown
CTRL-C # 2x: forceful shutdown
You can override the default listing page by adding:
zip redbean.com index.lua # lua server pages take priority
zip redbean.com index.html # default page for 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.
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
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.
$ 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
Audio video content should not be compressed in your ZIP files.
Uncompressed assets enable browsers to send Range HTTP request.
On the other hand compressed assets are best for gzip encoding.
zip redbean.com index.html # adds file
zip -0 redbean.com video.mp4 # adds without compression
You can have redbean run as a daemon by doing the following:
sudo ./redbean.com -vvdp80 -p443 -L redbean.log -P redbean.pid
kill -TERM $(cat redbean.pid) # 1x: graceful shutdown
kill -TERM $(cat redbean.pid) # 2x: forceful shutdown
redbean currently has a 32kb limit on request messages and 64kb
including the payload. redbean will grow to whatever the system
limits allow. Should fork() or accept() fail redbean will react
by going into "meltdown mode" which closes lingering workers.
You can trigger this at any time using:
kill -USR2 $(cat redbean.pid)
Another failure condition is running out of disk space in which
case redbean reacts by truncating the log file. Lastly, redbean
does the best job possible reporting on resource usage when the
logger is in debug mode noting that NetBSD is the best at this.
Your redbean is an actually portable executable, that's able to
run on six different operating systems. To do that, it needs to
extract a 4kb loader program to ${TMPDIR:-${HOME:-.}}/.ape that'll
map your redbean into memory. It does however check to see if `ape`
is on the system path beforehand. You can also "assimilate" any
redbean into the platform-local executable format by running:
$ file redbean.com
redbean.com: DOS/MBR boot sector
$ ./redbean.com --assimilate
$ file redbean.com
redbean.com: ELF 64-bit LSB executable
────────────────────────────────────────────────────────────────────────────────
SECURITY
redbean uses a protocol polyglot for serving HTTP and HTTPS on
the same port numbers. For example, both of these are valid:
http://127.0.0.1:8080/
https://127.0.0.1:8080/
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
Redbean supports sandboxing flags on Linux and OpenBSD.
-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.
See http://redbean.dev for further details.
────────────────────────────────────────────────────────────────────────────────
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 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
$ ./redbean.com --assimilate
$ 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
- redbean supports a printf modulus operator, like Python. For
example, you can say `"hello %s" % {"world"}` instead of
`string.format("hello %s", "world")`.
- redbean supports a string multiply operator, like Python. For
example, you can say `"hi" * 2` instead of `string.rep("hi", 2)`.
- redbean supports octal (base 8) integer literals. For example
`0644 == 420` is the case in redbean, whereas in upstream Lua
`0644 == 644` would be the case.
- redbean supports binary (base 2) integer literals. For example
`0b1010 == 10` is the case in redbean, whereas in upstream Lua
`0b1010` would result in an error.
- redbean supports the GNU syntax for the ASCII ESC character in
string literals. For example, `"\e"` is the same as `"\x1b"`.
────────────────────────────────────────────────────────────────────────────────
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.com -v arg1 arg2
Then your `/.init.lua` file will have the `arg` array like:
arg[-1] = '/usr/bin/redbean.com'
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'
────────────────────────────────────────────────────────────────────────────────
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
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.
────────────────────────────────────────────────────────────────────────────────
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.
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`, redbean will close the connection without calling fork.
OnLogLatency(reqtimeus:int,contimeus:int)
If this function is defined it'll be called from the main process
each time redbean completes handling of a request, but before the
response is sent. The handler received 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. If this hook is defined, then
`/.heartbeat.lua` is not called.
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.
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 this 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 Content-Range and
Date, 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):
- Expires: sets the maximum lifetime of the cookie as an HTTP-date
timestamp. Can be specified as a Date in the RFC1123 (string)
format or as a UNIX timestamp (number of seconds).
- MaxAge: sets number of seconds until the cookie expires. A zero
or negative number will expire the cookie immediately. If both
Expires and MaxAge are set, MaxAge has precedence.
- Domain: sets the host to which the cookie will be sent.
- Path: sets the path that must be present in the request URL, or
the client will not send the Cookie header.
- Secure: (bool) requests the cookie to be only send to the
server when a request is made with the https: scheme.
- HttpOnly: (bool) forbids JavaScript from accessing the cookie.
- SameSite: (Strict, Lax, or None) controls whether a cookie is
sent with cross-origin requests, providing some protection
against cross-site request forgery attacks.
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. 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.
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.
DecodeLatin1(iso-8859-1:str) → utf-8:str
Turns ISO-8859-1 string into UTF-8.
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
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:
- useoutput: (bool=false) encodes the result directly to the
output buffer and returns `nil` value. This option is
ignored if used outside of request handling code.
- sorted: (bool=true) Lua uses hash tables so the order of
object keys is lost in a Lua table. So, by default, we use
`strcmp` to impose a deterministic output order. If you
don't care about ordering then setting `sorted=false`
should yield a performance boost in serialization.
- pretty: (bool=false) Setting this option to true will
cause tables with more than one entry to be formatted
across multiple lines for readability.
- indent: (str=" ") This option controls the indentation of
pretty formatting. This field is ignored if `pretty` isn't
true.
- maxdepth: (int=64) This option controls the maximum amount
of recursion the serializer is allowed to perform. The max
is 32767. You might not be able to set it that high if
there isn't enough C stack memory. Your serializer checks
for this and will return an error rather than crashing.
This function will return an error if:
- `value` is cyclic
- `value` has depth greater than 64
- `value` contains functions, user data, or threads
- `value` is table that blends string / non-string keys
- Your serializer runs out of C heap memory (setrlimit)
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:
- useoutput: (bool=false) encodes the result directly to the
output buffer and returns `nil` value. This option is
ignored if used outside of request handling code.
- sorted: (bool=true) Lua uses hash tables so the order of
object keys is lost in a Lua table. So, by default, we use
`strcmp` to impose a deterministic output order. If you
don't care about ordering then setting `sorted=false`
should yield a performance boost in serialization.
- pretty: (bool=false) Setting this option to true will
cause tables with more than one entry to be formatted
across multiple lines for readability.
- indent: (str=" ") This option controls the indentation of
pretty formatting. This field is ignored if `pretty` isn't
true.
- maxdepth: (int=64) This option controls the maximum amount
of recursion the serializer is allowed to perform. The max
is 32767. You might not be able to set it that high if
there isn't enough C stack memory. Your serializer checks
for this and will return an error rather than crashing.
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 representing 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
-/?.~_@:!$&'()*+,;=0-9A-Za-z and everything else gets %XX encoded.
Please note that '& can still break HTML and that '() can still
break CSS URLs. This function is charset agnostic and will not
canonicalize overlong encodings. It is assumed that a UTF-8 string
will be supplied. See kescapefragment.c.
EscapeHost(str) → str
Escapes URL host. See kescapeauthority.c
EscapeLiteral(str) → str
Escapes JavaScript or JSON string literal content. The caller is
responsible for adding the surrounding quotation marks. This
implementation \uxxxx sequences for all non-ASCII sequences. HTML
entities are also encoded, so the output doesn't need EscapeHtml.
This function assumes UTF-8 input. Overlong encodings are
canonicalized. Invalid input sequences are assumed to be
ISO-8859-1. The output is UTF-16 since that's what JavaScript
uses. For example, some individual codepoints such as emoji
characters will encode as multiple \uxxxx sequences. Ints that are
impossible to encode as UTF-16 are substituted with the \xFFFD
replacement character. See escapejsstringliteral.c.
EscapeParam(str) → str
Escapes URL parameter name or value. The allowed characters are
-.*_0-9A-Za-z and everything else gets %XX encoded. This function
is charset agnostic and will not canonicalize overlong encodings.
It is assumed that a UTF-8 string will be supplied. See
kescapeparam.c.
EscapePass(str) → str
Escapes URL password. See kescapeauthority.c.
EscapePath(str) → str
Escapes URL path. This is the same as EscapeSegment except slash
is allowed. The allowed characters are -.~_@:!$&'()*+,;=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.c.
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.c.
EscapeUser(str) → str
Escapes URL username. See kescapeauthority.c.
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.
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 request method, body, and headers, as
well as some other options:
- method (default = "GET"): sets the method to be used for the
request. The specified method is converted to uppercase.
- body (default = ""): sets the body value to be sent.
- headers: sets headers for the request using the key/value pairs
from this table. Only string keys are used and all the values are
converted to strings.
- followredirect (default = true): forces temporary and permanent
redirects to be followed. This behavior can be disabled by
passing `false`.
- maxredirects (default = 5): sets the number of allowed redirects
to minimize looping due to misconfigured servers. When the number
is exceeded, the last response is returned.
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.
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).
If both a file and a ZIP asset are present, then the file is used.
GetAssetSize(path:str) → int
Returns byte size of uncompressed contents of 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.
GetBody() → str
Returns the request message body if present or an empty string.
Also available as GetPayload (deprecated).
GetCookie(name:str) → str
Returns cookie value.
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.
GetRemoteAddr() → ip:uint32,port:uint16
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, only if
IsPrivateIp or IsLoopbackIp return true. When multiple addresses
are present in the header, the last/right-most address is used.
GetResponseBody() → str
Returns the response message body if present or an empty string.
Also returns an empty string during streaming.
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.
GetClientFd() → int
Returns file descriptor being used for client connection.
This is useful for scripts that want to use unix:fork().
IsClientUsingSsl() → bool
Returns true if client connection has begun being managed by
the MbedTLS security layer. This is an important thing to
consider if a script is taking control of GetClientFd()
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:
- `"LINUX"`
- `"METAL"`
- `"WINDOWS"`
- `"XNU"`
- `"NETBSD"`
- `"FREEBSD"`
- `"OPENBSD"`
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.
GetSslIdentity() → str
Returns certificate subject or PSK identity from the current SSL
session. `nil` is returned for regular (non-SSL) connections.
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 as GetVersion (deprecated).
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).
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 as IsCompressed (deprecated).
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, returning object having the following fields: scheme,
user, pass, host, port, path, params, fragment. 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 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.
ProgramAddr(ip:int)
ProgramAddr(host:str)
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.
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)
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. This should only
be called from /.init.lua.
ProgramCertificate(pem:str)
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.
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 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.
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)
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.
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 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.
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:str)
May be used to disable the the verification of certificates
for remote hosts when using the Fetch() API. This function is
not available in unsecure mode.
ProgramSslClientVerify(enabled:str)
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:str)
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)
See https://redbean.dev/ for further details.
IsDaemon() → bool
Returns true if -d flag was passed to redbean.
ProgramUid(int)
Same as the -U flag if called from .init.lua for setuid()
ProgramGid(int)
Same as the -G flag if called from .init.lua for setgid()
ProgramDirectory(str)
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.
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(str)
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')
Sleep(seconds:number)
Sleeps the specified number of seconds (can be fractional). The
smallest interval is a microsecond.
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.
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.
Note that the asset needs to have "read other" permissions;
otherwise this function logs a warning and returns 403 Forbidden.
If this is undesirable, use GetAssetMode and ServeAsset to bypass
the check.
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 other Serve* functions.
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
other Serve* functions.
ServeRedirect(code:int,location:str)
Instructs redbean to return the specified redirect code along with
the Location header set. This function is mutually exclusive with
SetStatus and other Serve* functions.
SetLogLevel(level:int)
Sets logger verbosity. Reasonable values for level are kLogDebug >
kLogVerbose > kLogInfo > kLogWarn > kLogError > kLogFatal. This is
reset at the end of the http request, so it can be used to disable
access log and message logging.
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.
Crc32(initial:int,data:str) → int
Computes 32-bit CRC-32 used by zip/zlib/gzip/etc.
Crc32c(initial:int,data:str) → int
Computes 32-bit Castagnoli Cyclic Redundancy Check.
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.
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.
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(data) → int
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.
────────────────────────────────────────────────────────────────────────────────
CONSTANTS
kLogDebug
Integer for debug logging level. See Log.
kLogVerbose
Integer for verbose logging level, which is less than kLogDebug.
kLogInfo
Integer for info logging level, which is less than kLogVerbose.
kLogWarn
Integer for warn logging level, which is less than kLogVerbose.
kLogError
Integer for error logging level, which is less than kLogWarn.
kLogFatal
Integer for fatal logging level, which is less than kLogError.
Logging anything at this level will result in a backtrace and
process exit.
────────────────────────────────────────────────────────────────────────────────
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
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 = 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:
- `re.BASIC`
- `re.ICASE`
- `re.NEWLINE`
- `re.NOSUB`
- `re.NOTBOL`
- `re.NOTEOL`
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]) → re.Regex
├─→ preg:re.Regex
└─→ nil, re.Errno
Compiles regular expression.
`flags` defaults to zero and may have any of:
- `re.BASIC`
- `re.ICASE`
- `re.NEWLINE`
- `re.NOSUB`
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
re.Regex:search(text:str[, flags:int])
├─→ match:str[, group1:str, ...]
└─→ nil, re.Errno
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:
- `re.NOTBOL`
- `re.NOTEOL`
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 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
────────────────────────────────────────────────────────────────────────────────
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 re.Regex: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 re.Regex:search.
────────────────────────────────────────────────────────────────────────────────
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(path: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.
path.islink(path:str)
└─→ bool
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
For further details, 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:
- `"LINUX"`
- `"WINDOWS"`
- `"XNU"`
- `"NETBSD"`
- `"FREEBSD"`
- `"OPENBSD"`
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.
- E: End of Option list
- N: No-Operation
- M: Maxmimum Segment Size
- K: Window Scale
- O: SACK Permitted
- A: SACK
- e: Echo (obsolete)
- r: Echo reply (obsolete)
- T: Timestamps
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:int[, 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.variants.argon2_id` blend of other two methods [default]
- `argon2.variants.argon2_i` maximize resistance to side-channel attacks
- `argon2.variants.argon2_d` maximize resistance to gpu cracking attacks
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:
- `O_RDONLY`: open for reading (default)
- `O_WRONLY`: open for writing
- `O_RDWR`: open for reading and writing
The following values may also be OR'd into `flags`:
- `O_CREAT` create file if it doesn't exist
- `O_TRUNC` automatic ftruncate(fd,0) if exists
- `O_CLOEXEC` automatic close() upon execve()
- `O_EXCL` exclusive access (see below)
- `O_APPEND` open file for append only
- `O_NONBLOCK` asks read/write to fail with EAGAIN rather than block
- `O_DIRECT` it's complicated (not supported on Apple and OpenBSD)
- `O_DIRECTORY` useful for stat'ing (hint on UNIX but required on NT)
- `O_TMPFILE` try to make temp more secure (Linux and Windows only)
- `O_NOFOLLOW` fail if it's a symlink (zero on Windows)
- `O_DSYNC` it's complicated (zero on non-Linux/Apple)
- `O_RSYNC` it's complicated (zero on non-Linux/Apple)
- `O_PATH` it's complicated (zero on non-Linux)
- `O_VERIFY` it's complicated (zero on non-FreeBSD)
- `O_SHLOCK` it's complicated (zero on non-BSD)
- `O_EXLOCK` it's complicated (zero on non-BSD)
- `O_NOATIME` don't record access time (zero on non-Linux)
- `O_RANDOM` hint random access intent (zero on non-Windows)
- `O_SEQUENTIAL` hint sequential access intent (zero on non-Windows)
- `O_COMPRESSED` ask fs to abstract compression (zero on non-Windows)
- `O_INDEXED` turns on that slow performance (zero on non-Windows)
There are three regular combinations for the above flags:
- `O_RDONLY`: Opens existing file for reading. If it doesn't
exist then nil is returned and errno will be `ENOENT` (or in
some other cases `ENOTDIR`).
- `O_WRONLY|O_CREAT|O_TRUNC`: Creates file. If it already
exists, then the existing copy is destroyed and the opened
file will start off with a length of zero. This is the
behavior of the traditional creat() system call.
- `O_WRONLY|O_CREAT|O_EXCL`: Create file only if doesn't exist
already. If it does exist then `nil` is returned along with
`errno` set to `EEXIST`.
`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:
Linux 5.4 fork l: 97,200𝑐 31,395𝑛𝑠 [metal]
FreeBSD 12 fork l: 236,089𝑐 78,841𝑛𝑠 [vmware]
Darwin 20.6 fork l: 295,325𝑐 81,738𝑛𝑠 [metal]
NetBSD 9 fork l: 5,832,027𝑐 1,947,899𝑛𝑠 [vmware]
OpenBSD 6.8 fork l: 13,241,940𝑐 4,422,103𝑛𝑠 [vmware]
Windows10 fork l: 18,802,239𝑐 6,360,271𝑛𝑠 [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 divided 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.
`EAGAIN` is returned if you've enforced a max number of
processes using `setrlimit(RLIMIT_NPROC)`.
unix.dup(oldfd:int[, newfd:int[, flags:int]])
├─→ newfd:int
└─→ nil, unix.Errno
Duplicates file descriptor.
`newfd` defaults to the lowest number available file descriptor.
If the new number is specified and it's already open, then it'll
be silently closed before the duplication happens.
`flags` can have `O_CLOEXEC` which means the returned file
descriptors will be automatically closed upon execve().
unix.pipe([flags:int])
├─→ reader:int, writer:int
└─→ nil, unix.Errno
Creates fifo which enables communication between processes.
`flags` may have any combination (using bitwise OR) of:
- `O_CLOEXEC`: Automatically close file descriptor upon execve()
- `O_NONBLOCK`: Request `EAGAIN` be raised rather than blocking
- `O_DIRECT`: Enable packet mode w/ atomic reads and writes, so long
as they're no larger than `PIPE_BUF` (guaranteed to be 512+ bytes)
with support limited to Linux, Windows NT, FreeBSD, and NetBSD.
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(err))
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:
- `pid > 0` signals one process by id
- `== 0` signals all processes in current process group
- `-1` signals all processes possible (except init)
- `< -1` signals all processes in -pid process group
`sig` can be:
- `0` checks both if pid exists and we can signal it
- `SIGINT` sends ctrl-c keyboard interrupt
- `SIGQUIT` sends backtrace and exit signal
- `SIGTERM` sends shutdown signal
- etc.
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:
- `AT_SYMLINK_NOFOLLOW`: do not follow symbolic links.
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`.
unix.unlink(path:str[, dirfd:int])
├─→ true
└─→ nil, unix.Errno
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.
unix.link(existingpath:str, newpath:str[, flags:int[, olddirfd, newdirfd]])
├─→ true
└─→ nil, unix.Errno
Creates hard link, so your underlying inode has two names.
unix.symlink(target:str, linkpath:str[, newdirfd:int])
├─→ true
└─→ nil, unix.Errno
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.
unix.readlink(path:str[, dirfd:int])
├─→ content:str
└─→ nil, unix.Errno
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.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:
- `unix.UTIME_NOW`: Fill this timestamp with current time. This
feature is not available on old versions of Linux, e.g. RHEL5.
- `unix.UTIME_OMIT`: Do not alter this timestamp. This feature is
not available on old versions of Linux, e.g. RHEL5.
`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
- `AT_SYMLINK_NOFOLLOW`: Do not follow symbolic links. This makes it
possible to edit the timestamps on the symbolic link itself,
rather than the file it points to.
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:
- `unix.UTIME_NOW`: Fill this timestamp with current time.
- `unix.UTIME_OMIT`: Do not alter this timestamp.
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.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.fcntl(fd:int, cmd:int, ...)
├─→ ...
└─→ nil, unix.Errno
Manipulates file descriptor.
Setting `cmd` to `F_GETFD`, `F_SETFD`, `F_GETFL` or `F_SETFL`
lets you query and/or change the status of file descriptors. For
example, it's possible using this to change `FD_CLOEXEC`.
[work in progress] POSIX advisory locks can be controlled by setting
`cmd` to `F_UNLCK`, `F_RDLCK`, `F_WRLCK`, `F_SETLK`, or `F_SETLKW`.
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.setfsuid(uid:int)
├─→ true
└─→ nil, unix.Errno
Sets user id for file system ops.
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:
- `CLOCK_REALTIME`: universally supported
- `CLOCK_REALTIME_FAST`: ditto but faster on freebsd
- `CLOCK_MONOTONIC`: universally supported
- `CLOCK_MONOTONIC_FAST`: ditto but faster on freebsd
- `CLOCK_MONOTONIC_RAW`: nearly universally supported
- `CLOCK_PROCESS_CPUTIME_ID`: linux and bsd
- `CLOCK_THREAD_CPUTIME_ID`: linux and bsd
- `CLOCK_REALTIME_COARSE`: : linux and openbsd
- `CLOCK_MONOTONIC_COARSE`: linux
- `CLOCK_PROF`: linux and netbsd
- `CLOCK_BOOTTIME`: linux and openbsd
- `CLOCK_REALTIME_ALARM`: linux-only
- `CLOCK_BOOTTIME_ALARM`: linux-only
- `CLOCK_TAI`: linux-only
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:
- `SEEK_SET`: Sets the file position to `offset` [default]
- `SEEK_CUR`: Sets the file position to `position + offset`
- `SEEK_END`: Sets the file position to `filesize + offset`
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
`family` defaults to `AF_INET` and can be:
- `AF_INET`: Creates Internet Protocol Version 4 (IPv4) socket.
- `AF_UNIX`: Creates local UNIX domain socket. On the New Technology
this requires Windows 10 and only works with `SOCK_STREAM`.
`type` defaults to `SOCK_STREAM` and can be:
- `SOCK_STREAM`
- `SOCK_DGRAM`
- `SOCK_RAW`
- `SOCK_RDM`
- `SOCK_SEQPACKET`
You may bitwise OR any of the following into `type`:
- `SOCK_CLOEXEC`
- `SOCK_NONBLOCK`
`protocol` may be any of:
- `0` to let kernel choose [default]
- `IPPROTO_TCP`
- `IPPROTO_UDP`
- `IPPROTO_RAW`
- `IPPROTO_IP`
- `IPPROTO_ICMP`
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:
- `SOCK_STREAM`
- `SOCK_DGRAM`
- `SOCK_SEQPACKET`
You may bitwise OR any of the following into `type`:
- `SOCK_CLOEXEC`
- `SOCK_NONBLOCK`
`protocol` defaults to `0`.
unix.bind(fd:int[, ip:uint32, port:uint16])
unix.bind(fd:int[, unixpath:str])
├─→ 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))
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
- `SOL_SOCKET`, `SO_TYPE`
- `SOL_SOCKET`, `SO_DEBUG`
- `SOL_SOCKET`, `SO_ACCEPTCONN`
- `SOL_SOCKET`, `SO_BROADCAST`
- `SOL_SOCKET`, `SO_REUSEADDR`
- `SOL_SOCKET`, `SO_REUSEPORT`
- `SOL_SOCKET`, `SO_KEEPALIVE`
- `SOL_SOCKET`, `SO_DONTROUTE`
- `SOL_TCP`, `TCP_NODELAY`
- `SOL_TCP`, `TCP_CORK`
- `SOL_TCP`, `TCP_QUICKACK`
- `SOL_TCP`, `TCP_FASTOPEN_CONNECT`
- `SOL_TCP`, `TCP_DEFER_ACCEPT`
- `SOL_IP`, `IP_HDRINCL`
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
- `SOL_SOCKET`, `SO_SNDBUF`
- `SOL_SOCKET`, `SO_RCVBUF`
- `SOL_SOCKET`, `SO_RCVLOWAT`
- `SOL_SOCKET`, `SO_SNDLOWAT`
- `SOL_TCP`, `TCP_KEEPIDLE`
- `SOL_TCP`, `TCP_KEEPINTVL`
- `SOL_TCP`, `TCP_FASTOPEN`
- `SOL_TCP`, `TCP_KEEPCNT`
- `SOL_TCP`, `TCP_MAXSEG`
- `SOL_TCP`, `TCP_SYNCNT`
- `SOL_TCP`, `TCP_NOTSENT_LOWAT`
- `SOL_TCP`, `TCP_WINDOW_CLAMP`
- `SOL_IP`, `IP_TOS`
- `SOL_IP`, `IP_MTU`
- `SOL_IP`, `IP_TTL`
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
- `SOL_SOCKET`, `SO_RCVTIMEO`: If this option is specified then
your stream socket will have a read() / recv() timeout. If the
specified interval elapses without receiving data, then EAGAIN
shall be returned by read. If this option is used on listening
sockets, it'll be inherited by accepted sockets. Your redbean
already does this for GetClientFd() based on the `-t` flag.
- `SOL_SOCKET`, `SO_SNDTIMEO`: This is the same as `SO_RCVTIMEO`
but it applies to the write() / send() functions.
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])
├─→ {[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:
- `POLLIN` (events, revents): There is data to read.
- `POLLOUT` (events, revents): Writing is now possible, although may
still block if available space in a socket or pipe is exceeded
(unless `O_NONBLOCK` is set).
- `POLLPRI` (events, revents): There is some exceptional condition
(for example, out-of-band data on a TCP socket).
- `POLLRDHUP` (events, revents): Stream socket peer closed
connection, or shut down writing half of connection.
- `POLLERR` (revents): Some error condition.
- `POLLHUP` (revents): Hang up. When reading from a channel such as
a pipe or a stream socket, this event merely indicates that the
peer closed its end of the channel.
- `POLLNVAL` (revents): Invalid request.
`timeoutms` is the number of milliseconds to block. If this is set to
-1 then that means block as long as it takes until there's an event
or an interrupt. If the timeout expires, an empty table is returned.
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
├─→ clientfd:int, unixpath:str
└─→ nil, unix.Errno
Accepts new client socket descriptor for a listening tcp socket.
`flags` may have any combination (using bitwise OR) of:
- `SOCK_CLOEXEC`
- `SOCK_NONBLOCK`
unix.connect(fd:int, ip:uint32, port:uint16)
unix.connect(fd:int, unixpath:str)
├─→ 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
├─→ unixpath:str
└─→ nil, unix.Errno
Retrieves the local address of a socket.
unix.getpeername(fd:int)
├─→ ip:uint32, port:uint16
├─→ unixpath:str
└─→ nil, unix.Errno
Retrieves the remote address of a socket.
This operation will either fail on `AF_UNIX` sockets or return an
empty string.
unix.recv(fd:int[, bufsiz:int[, flags:int]])
├─→ data:str
└─→ nil, unix.Errno
`flags` may have any combination (using bitwise OR) of:
- `MSG_WAITALL`
- `MSG_DONTROUTE`
- `MSG_PEEK`
- `MSG_OOB`
unix.recvfrom(fd:int[, bufsiz:int[, flags:int]])
├─→ data:str, ip:uint32, port:uint16
├─→ data:str, unixpath:str
└─→ nil, unix.Errno
`flags` may have any combination (using bitwise OR) of:
- `MSG_WAITALL`
- `MSG_DONTROUTE`
- `MSG_PEEK`
- `MSG_OOB`
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 combination (using bitwise OR) of:
- `MSG_NOSIGNAL`: Don't SIGPIPE on EOF
- `MSG_OOB`: Send stream data through out of bound channel
- `MSG_DONTROUTE`: Don't go through gateway (for diagnostics)
- `MSG_MORE`: Manual corking to belay nodelay (0 on non-Linux)
unix.sendto(fd:int, data:str, ip:uint32, port:uint16[, flags:int])
unix.sendto(fd:int, data:str, unixpath:str[, flags:int])
├─→ sent:int
└─→ nil, unix.Errno
This is useful for sending messages over UDP sockets to specific
addresses.
`flags` may have any combination (using bitwise OR) of:
- `MSG_OOB`
- `MSG_DONTROUTE`
- `MSG_NOSIGNAL`
unix.shutdown(fd:int, how:int)
├─→ true
└─→ nil, unix.Errno
Partially closes socket.
`how` is set to one of:
- `SHUT_RD`: sends a tcp half close for reading
- `SHUT_WR`: sends a tcp half close for writing
- `SHUT_RDWR`
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:unix.Sigset)
├─→ oldmask:unix.Sigset
└─→ nil, unix.Errno
Manipulates bitset of signals blocked by process.
`how` can be one of:
- `SIG_BLOCK`: applies `mask` to set of blocked signals using bitwise OR
- `SIG_UNBLOCK`: removes bits in `mask` from set of blocked signals
- `SIG_SETMASK`: replaces process signal mask with `mask`
`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:unix.Sigset]]])
├─→ oldhandler:func|int, flags:int, mask:unix.Sigset
└─→ nil, unix.Errno
`sig` can be one of:
- `unix.SIGINT`
- `unix.SIGQUIT`
- `unix.SIGTERM`
- etc.
`handler` can be:
- Lua function
- `unix.SIG_IGN`
- `unix.SIG_DFL`
`flags` can have:
- `unix.SA_RESTART`: Enables BSD signal handling semantics. Normally
i/o entrypoints check for pending signals to deliver. If one gets
delivered during an i/o call, the normal behavior is to cancel the
i/o operation and return -1 with `EINTR` in errno. If you use the
`SA_RESTART` flag then that behavior changes, so that any function
that's been annotated with @restartable will not return `EINTR`
and will instead resume the i/o operation. This makes coding
easier but it can be an anti-pattern if not used carefully, since
poor usage can easily result in latency issues. It also requires
one to do more work in signal handlers, so special care needs to
be given to which C library functions are @asyncsignalsafe.
- `unix.SA_RESETHAND`: Causes signal handler to be single-shot. This
means that, upon entry of delivery to a signal handler, it's reset
to the `SIG_DFL` handler automatically. You may use the alias
`SA_ONESHOT` for this flag, which means the same thing.
- `unix.SA_NODEFER`: Disables the reentrancy safety check on your signal
handler. Normally that's a good thing, since for instance if your
`SIGSEGV` signal handler happens to segfault, you're going to want
your process to just crash rather than looping endlessly. But in
some cases it's desirable to use `SA_NODEFER` instead, such as at
times when you wish to `longjmp()` out of your signal handler and
back into your program. This is only safe to do across platforms
for non-crashing signals such as `SIGCHLD` and `SIGINT`. Crash
handlers should use Xed instead to recover execution, because on
Windows a `SIGSEGV` or `SIGTRAP` crash handler might happen on a
separate stack and/or a separate thread. You may use the alias
`SA_NOMASK` for this flag, which means the same thing.
- `unix.SA_NOCLDWAIT`: Changes `SIGCHLD` so the zombie is gone and
you can't call wait() anymore; similar but may still deliver the
SIGCHLD.
- `unix.SA_NOCLDSTOP`: Lets you set `SIGCHLD` handler that's only
notified on exit/termination and not notified on `SIGSTOP`,
`SIGTSTP`, `SIGTTIN`, `SIGTTOU`, or `SIGCONT`.
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:Sigmask])
└─→ 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)
`intns` needs to be on the interval `[0,1000000000)`
`valuens` needs to be on the interval `[0,1000000000)`
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:
- `RLIMIT_AS` limits the size of the virtual address space. This
will work on all platforms. It's emulated on XNU and Windows which
means it won't propagate across execve() currently.
- `RLIMIT_CPU` causes `SIGXCPU` to be sent to the process when the
soft limit on CPU time is exceeded, and the process is destroyed
when the hard limit is exceeded. It works everywhere but Windows
where it should be possible to poll getrusage() with setitimer().
- `RLIMIT_FSIZE` causes `SIGXFSZ` to sent to the process when the
soft limit on file size is exceeded and the process is destroyed
when the hard limit is exceeded. It works everywhere but Windows.
- `RLIMIT_NPROC` limits the number of simultaneous processes and it
should work on all platforms except Windows. Please be advised it
limits the process, with respect to the activities of the user id
as a whole.
- `RLIMIT_NOFILE` limits the number of open file descriptors and it
should work on all platforms except Windows (TODO).
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:
- `RUSAGE_SELF`: current process
- `RUSAGE_THREAD`: current thread
- `RUSAGE_CHILDREN`: not supported on Windows NT
- `RUSAGE_BOTH`: not supported on non-Linux
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.
`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).
rpath
Allows chdir, getcwd, open, stat, fstat, access, readlink, chmod,
chmod, fchmod.
wpath
Allows getcwd, open, stat, fstat, access, readlink, chmod, fchmod.
cpath
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, tiocgwinsz, tcgets, tcsets, tcsetsw, tcsetsf.
inet
Allows socket (AF_INET), listen, bind, connect, 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.
id
Allows setuid, setreuid, setresuid, setgid, setregid, setresgid,
setgroups, setrlimit, getpriority, setpriority.
settime
Allows settimeofday and clock_adjtime.
unveil
Allows unveil().
exec
Allows execve.
If the executable in question needs a loader, then you will need
"rpath prot_exec" too. With APE, security is strongest when you
assimilate your binaries beforehand, using the --assimilate flag,
or the o//tool/build/assimilate.com program. On OpenBSD this is
mandatory.
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:
- `unix.PLEDGE_PENALTY_KILL_THREAD` causes the violating thread to
be killed. This is the default on Linux. It's effectively the
same as killing the process, since redbean has no threads. The
termination signal can't be caught and will be either `SIGSYS`
or `SIGABRT`. Consider enabling stderr logging below so you'll
know why your program failed. Otherwise check the system log.
- `unix.PLEDGE_PENALTY_KILL_PROCESS` causes the process and all
its threads to be killed. This is always the case on OpenBSD.
- `unix.PLEDGE_PENALTY_RETURN_EPERM` causes system calls to just
return an `EPERM` error instead of killing. This is a gentler
solution that allows code to display a friendly warning. Please
note this may lead to weird behaviors if the software being
sandboxed is lazy about checking error results.
`mode` may optionally bitwise or the following flags:
- `unix.PLEDGE_STDERR_LOGGING` enables friendly error message
logging letting you know which promises are needed whenever
violations occur. Without this, violations will be logged to
`dmesg` on Linux if the penalty is to kill the process. You
would then need to manually look up the system call number and
then cross reference it with the cosmopolitan libc pledge()
documentation. You can also use `strace -ff` which is easier.
This is ignored OpenBSD, which already has a good system log.
Turning on stderr logging (which uses SECCOMP trapping) also
means that the `unix.WTERMSIG()` on your killed processes will
always be `unix.SIGABRT` on both Linux and OpenBSD. Otherwise,
Linux prefers to raise `unix.SIGSYS`.
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(nil,nil)`
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(nil,nil)` 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].
`path` is the file or directory to unveil
`permissions` is a string consisting of zero or more of the
following characters:
- 'r' makes `path` available for read-only path operations,
corresponding to the pledge promise "rpath".
- `w` makes `path` available for write operations, corresponding
to the pledge promise "wpath".
- `x` makes `path` available for execute operations,
corresponding to the pledge promises "exec" and "execnative".
- `c` allows `path` to be created and removed, corresponding to
the pledge promise "cpath".
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.
- `mon` 1 ≤ mon ≤ 12
- `mday` 1 ≤ mday ≤ 31
- `hour` 0 ≤ hour ≤ 23
- `min` 0 ≤ min ≤ 59
- `sec` 0 ≤ sec ≤ 60
- `gmtoff` ±93600 seconds
- `wday` 0 ≤ wday ≤ 6
- `yday` 0 ≤ yday ≤ 365
- `dst` 1 if daylight savings, 0 if not, -1 if not unknown
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() which has further details.
Your redbean ships with a subset of the time zone database.
- `/zip/usr/share/zoneinfo/Honolulu` Z-10
- `/zip/usr/share/zoneinfo/Anchorage` Z -9
- `/zip/usr/share/zoneinfo/GST` Z -8
- `/zip/usr/share/zoneinfo/Boulder` Z -6
- `/zip/usr/share/zoneinfo/Chicago` Z -5
- `/zip/usr/share/zoneinfo/New_York` Z -4
- `/zip/usr/share/zoneinfo/UTC` Z +0
- `/zip/usr/share/zoneinfo/GMT` Z +0
- `/zip/usr/share/zoneinfo/London` Z +1
- `/zip/usr/share/zoneinfo/Berlin` Z +2
- `/zip/usr/share/zoneinfo/Israel` Z +3
- `/zip/usr/share/zoneinfo/India` Z +5
- `/zip/usr/share/zoneinfo/Beijing` Z +8
- `/zip/usr/share/zoneinfo/Japan` Z +9
- `/zip/usr/share/zoneinfo/Sydney` Z+10
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:
- `AT_SYMLINK_NOFOLLOW`: do not follow symbolic links.
`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.
`fd` should be a file descriptor that was opened using
`unix.open(path, O_RDONLY|O_DIRECTORY)`.
`flags` may have any of:
- `AT_SYMLINK_NOFOLLOW`: do not follow symbolic links.
`dirfd` defaults to to `unix.AT_FDCWD` and may optionally be set to
a directory file descriptor to which `path` is relative.
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.opendir(path:str)
├─→ state:unix.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:unix.Dir
└─→ nil, unix.Errno
Opens directory for listing its contents, via an fd.
`fd` should be created by `open(path, O_RDONLY|O_DIRECTORY)`. The
returned unix.Dir ownership 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 pseudoteletypewriter.
unix.tiocgwinsz(fd:int)
├─→ rows:int, cols:int
└─→ nil, unix.Errno
Returns cellular dimensions of pseudoteletypewriter display.
────────────────────────────────────────────────────────────────────────────────
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. Or error, `nil` will
be returned and `errno` will be non-nil.
`kind` can be any of:
- `DT_REG`: file is a regular file
- `DT_DIR`: file is a directory
- `DT_BLK`: file is a block device
- `DT_LNK`: file is a symbolic link
- `DT_CHR`: file is a character device
- `DT_FIFO`: file is a named pipe
- `DT_SOCK`: file is a named socket
- `DT_UNKNOWN`
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.
unid.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 unix.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 unix.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.
unis.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 unix.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.S_ISREG(st:mode())` means regular file
- `unix.S_ISDIR(st:mode())` means directory
- `unix.S_ISLNK(st:mode())` means symbolic link
- `unix.S_ISCHR(st:mode())` means character device
- `unix.S_ISBLK(st:mode())` means block device
- `unix.S_ISFIFO(st:mode())` means fifo or pipe
- `unix.S_ISSOCK(st:mode())` means socket
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 whether or not compression is being 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 set to
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 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 MAGNUMS
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 MAGNUMS
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.
unix.ENOENT
No such file or directory.
Raised by access, bind, chdir, chmod, chown, chroot, clock_getres,
execve, opendir, inotify_add_watch, link, mkdir, mknod, open,
readlink, rename, rmdir, stat, swapon, 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, inotify_add_watch.
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, pciconfig_read, 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, sched_setattr.
unix.ENOEXEC
Exec format error.
Raised by execve, uselib.
unix.ECHILD
No child process.
Raised by wait, waitpid, waitid, wait3, wait4.
unix.ESRCH
No such process.
Raised by getpriority, getrlimit, getsid, ioprio_set, kill, setpgid,
tkill, 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, inotify_add_watch,
inotify_rm_watch, 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,
timerfd_create, 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, timerfd_create, tkill,
write,
unix.EPIPE
Broken pipe. Returned by write, send. 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.
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, inotify_add_watch, link, mkdir, mknod, open, readlink,
rename, rmdir, stat, symlink, truncate, u unlink, utimensat.
unix.EACCES
Permission denied.
Raised by access, bind, chdir, chmod, chown, chroot, clock_getres,
connect, execve, fcntl, getpriority, inotify_add_watch, 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, fanotify_init, fork, getgroups, getrlimit,
inotify_add_watch, inotify_init, ioperm, link, mbind, mincore,
mkdir, mknod, mlock, mmap, mprotect, mremap, msync, open, poll,
readlink, recv, rename, rmdir, select, send, sigaltstack, splice,
stat, subpage_prot, swapon, symlink, sync_file_range, tee,
timer_create, timerfd_create, unlink.
unix.EPERM
Operation not permitted.
Raised by accept, chmod, chown, chroot, copy_file_range, execve,
fallocate, fanotify_init, fcntl, futex, get_robust_list,
getdomainname, getgroups, gethostname, getpriority, getrlimit,
getsid, gettimeofday, idle, init_module, io_submit, ioctl_console,
ioctl_ficlonerange, ioctl_fideduperange, ioperm, iopl, ioprio_set,
keyctl, kill, link, lookup_dcookie, madvise, mbind, membarrier,
migrate_pages, mkdir, mknod, mlock, mmap, mount, move_pages, msgctl,
nice, open, open_by_handle_at, pciconfig_read, perf_event_open,
pidfd_getfd, pidfd_send_signal, pivot_root, prctl, process_vm_readv,
ptrace, quotactl, reboot, rename, request_key, rmdir,
rt_sigqueueinfo, sched_setaffinity, sched_setattr, sched_setparam,
sched_setscheduler, seteuid, setfsgid, setfsuid, setgid, setns,
setpgid, setresuid, setreuid, setsid, setuid, setup, setxattr,
sigaltstack, spu_create, stime, swapon, 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 inotify_add_watch, 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 arch_prctl, mmap, open, prctl, timerfd_create.
unix.EISDIR
Is a 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, inotify_init, mmap, open, pipe, socket,
socketpair, swapon, timerfd_create, uselib, userfaultfd.
unix.EMFILE
Too many open files.
Raised by accept, dup, execve, fanotify_init, fcntl, inotify_init,
open, pipe, socket, socketpair, timerfd_create.
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, inotify_add_watch, 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.
unix.EMLINK
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.ENOMSG
Raised by msgop.
unix.EIDRM
Identifier removed.
Raised by msgctl.
unix.ETIME
Timer expired; timer expired.
Raised by connect.
unix.EPROTO
Raised by accept, connect, socket, socketpair.
unix.EOVERFLOW
Raised by copy_file_range, fanotify_init, lseek, mmap,
open, stat, statfs
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 MAGNUMS
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.
────────────────────────────────────────────────────────────────────────────────
LEGAL
redbean contains software licensed ISC, MIT, BSD-2, BSD-3, zlib
which makes it a permissively licensed gift to anyone who might
find it useful. The transitive closure of legalese can be found
inside the binary structure, because notice licenses require we
distribute the license along with any software that uses it. By
putting them in the binary, compliance in automated and no need
for further action on the part of the user who is distributing.
────────────────────────────────────────────────────────────────────────────────
SEE ALSO
https://redbean.dev/
https://news.ycombinator.com/item?id=26271117
Reference in a new issue View git blame Copy permalink