netCDF Authorization Support

Author: Dennis Heimbigner

Address: https://www.unidata.ucar.edu/staff/dmh/

Draft: 11/21/2014

Last Revised: 10/24/2015

Table of Contents

  1. Introduction
  2. URL-Based Authentication
  3. RC File Authentication
  4. Redirection-Based Authentication
  5. URL Constrained RC File Entries
  6. Client-Side Certificates
  7. Appendix A. All RC-File Keys
  8. Appendix B. ESG Access in Detail

Introduction

netCDF can support user authorization using the facilities provided by the curl library. This includes basic password authentication as well as certificate-based authorization.

With some exceptions (e.g. see the section on redirection) The libcurl authorization mechanisms can be accessed in two ways

  1. Inserting the username and password into the url, or
  2. Accessing information from a so-called rc file named either .daprc or .dodsrc

URL-Based Authentication

For simple password based authentication, it is possible to directly insert the username and the password into a url in this form.
    http://username:password@host/...
This username and password will be used if the server asks for authentication. Note that only simple password authentication is supported in this format. Specifically note that redirection based authorization will not work with this because the username and password will only be used on the initial request, not the redirection

RC File Authentication

The netcdf library supports an rc file mechanism to allow the passing of a number of parameters to libnetcdf and libcurl.

The file must be called one of the following names: ".daprc" or ".dodsrc" If both .daprc and .dodsrc exist, then the .daprc file will take precedence.

The rc file is searched for first in the current directory and then in the home directory (as defined by the HOME environment variable).

The rc file format is a series of lines of the general form:

[<host:port>]<key>=<value>
where the bracket-enclosed host:port is optional and will be discussed subsequently.

The currently defined set of authorization-related keys are as follows. The second column is the affected curl_easy_setopt option(s), if any.
KeyAffected curl_easy_setopt OptionsNotes
HTTP.COOKIEJARCURLOPT_COOKIEJAR
HTTP.COOKIEFILECURLOPT_COOKIEJARAlias for CURLOPT_COOKIEJAR
HTTP.PROXY_SERVERCURLOPT_PROXY, CURLOPT_PROXYPORT, CURLOPT_PROXYUSERPWD
HTTP.SSL.CERTIFICATECURLOPT_SSLCERT
HTTP.SSL.KEYCURLOPT_SSLKEY
HTTP.SSL.KEYPASSWORDCURLOPT_KEYPASSWORD
HTTP.SSL.CAINFOCURLOPT_SSLCAINFO
HTTP.SSL.CAPATHCURLOPT_SSLCAPATH
HTTP.SSL.VERIFYPEERCURLOPT_SSL_VERIFYPEER
HTTP.SSL.VALIDATECURLOPT_SSL_VERIFYPEER, CURLOPT_SSL_VERIFYHOST
HTTP.CREDENTIALS.USERPASSWORDCURLOPT_USERPASSWORD
HTTP.NETRCN.A.Specify path of the .netrc file

Password Authentication

The key HTTP.CREDENTIALS.USERPASSWORD can be used to set the simple password authentication. This is an alternative to setting it in the url. The value must be of the form "username:password". See redirection authorization for important additional information.

Cookie Jar

The HTTP.COOKIEJAR key specifies the name of file from which to read cookies (CURLOPT_COOKIEJAR) and also the file into which to store cookies (CURLOPT_COOKIEFILE). The same value is used for both CURLOPT values. It defaults to in-memory storage. See redirection authorization for important additional information.

Certificate Authentication

HTTP.SSL.CERTIFICATE specifies a file path for a file containing a PEM cerficate. This is typically used for client-side authentication.

HTTP.SSL.KEY is essentially the same as HTTP.SSL.CERTIFICATE and should always have the same value.

HTTP.SSL.KEYPASSWORD specifies the password for accessing the HTTP.SSL.CERTIFICAT/HTTP.SSL.key file.

HTTP.SSL.CAPATH specifies the path to a directory containing trusted certificates for validating server certificates.

HTTP.SSL.VALIDATE is a boolean (1/0) value that if true (1) specifies that the client should verify the server's presented certificate.

HTTP.PROXY_SERVER specifies the url for accessing the proxy: e.g. http://[username:password@]host[:port]

HTTP.NETRC specifies the absolute path of the .netrc file. See redirection authorization for information about using .netrc.

Redirection-Based Authentication

Some sites provide authentication by using a third party site to do the authentication. Examples include ESG and URS.

The process is usually as follows.

  1. The client contacts the server of interest (SOI), the actual data provider using, typically http protocol.
  2. The SOI sends a redirect to the client to connect to the e.g. URS system using the 'https' protocol (note the use of https instead of http).
  3. The client authenticates with URS.
  4. URS sends a redirect (with authorization information) to send the client back to the SOI to actually obtain the data.

It turns out that libcurl uses the password in the .daprc file — or from the url — only for the initial connection. This causes problems because the redirected connection is the one that actually requires the password. This is where .netrc comes in. Libcurl will use .netrc for the redirected connection. It is possible to cause libcurl to use the .daprc password always, but this introduces a security hole because it may send the initial user+pwd to the redirection site. In summary, if you are using redirection, then you must create a .netrc file to hold the password for the site to which the redirection is sent.

The format of this .netrc file will contain content that typically look like this.

machine mmmmmm login xxxxxx password yyyyyy
where the machine, mmmmmm, is the hostname of the machine to which the client is redirected for authorization, and the login and password are those needed to authenticate on that machine.

The .netrc file can be specified by putting the following line in your .daprc/.dodsrc file.

HTTP.NETRC=<path to netrc file>

One final note. In using this, it is almost certain that you will need to specify a real cookie jar file (HTTP.COOKIEJAR) so that the redirect site can pass back authorization information.

URL Constrained RC File Entries

Each line of the rc file can begin with a host+port enclosed in square brackets. The form is "host:port". If the port is not specified then the form is just "host". The reason that more of the url is not used is that libcurl's authorization grain is not any finer than host level.

Examples.

[remotetest.unidata.ucar.edu]HTTP.VERBOSE=1
or
[fake.ucar.edu:9090]HTTP.VERBOSE=0
If the url request from, say, the netcdf_open method has a host+port matching one of the prefixes in the rc file, then the corresponding entry will be used, otherwise ignored.

For example, the URL

http://remotetest.unidata.ucar.edu/thredds/dodsC/testdata/testData.nc
will have HTTP.VERBOSE set to 1.

Similarly,

http://fake.ucar.edu:9090/dts/test.01
will have HTTP.VERBOSE set to 0.

Client-Side Certificates

Some systems, notably ESG (Earth System Grid), requires the use of client-side certificates, as well as being re-direction based. This requires setting the following entries: Note that the first two are to support re-direction based authentication.

Appendix A. All RC-File Keys

For completeness, this is the list of all rc-file keys. If this documentation is out of date with respect to the actual code, the code is definitive.
Keycurl_easy_setopt Option
HTTP.DEFLATECUROPT_DEFLATE
with value "deflate,gzip"
HTTP.VERBOSE CUROPT_VERBOSE
HTTP.TIMEOUTCUROPT_TIMEOUT
HTTP.USERAGENTCUROPT_USERAGENT
HTTP.COOKIEJARCUROPT_COOKIEJAR
HTTP.COOKIE_JARCUROPT_COOKIEJAR
HTTP.PROXY_SERVERCURLOPT_PROXY,
CURLOPT_PROXYPORT,
CURLOPT_PROXYUSERPWD
HTTP.SSL.CERTIFICATECUROPT_SSLCERT
HTTP.SSL.KEYCUROPT_SSLKEY
HTTP.SSL.KEYPASSWORDCUROPT_KEYPASSWORD
HTTP.SSL.CAINFOCUROPT_SSLCAINFO
HTTP.SSL.CAPATHCUROPT_SSLCAPATH
HTTP.SSL.VERIFYPEERCUROPT_SSL_VERIFYPEER
HTTP.CREDENTIALS.USERPASSWORDCUROPT_USERPASSWORD
HTTP.NETRCCURLOPT_NETRC,CURLOPT_NETRC_FILE

Appendix B. URS Access in Detail

It is possible to use the NASA Earthdata Login System (URS) with netcdf by using using the process specified in the redirection based authorization section. In order to access URS controlled datasets, however, it is necessary to register as a user with NASA at the https://uat.urs.earthdata.nasa.gov/ website.

Appendix C. ESG Access in Detail

It is possible to access Earth Systems Grid (ESG) datasets from ESG servers through the netCDF API using the techniques described in the section on Client-Side Certificates.

In order to access ESG datasets, however, it is necessary to register as a user with ESG and to setup your environment so that proper authentication is established between an netcdf client program and the ESG data server. Specifically, it is necessary to use what is called "client-side keys" to enable this authentication. Normally, when a client accesses a server in a secure fashion (using "https"), the server provides an authentication certificate to the client. With client-side keys, the client must also provide a certificate to the server so that the server can know with whom it is communicating.

The netcdf library uses the curl library and it is that underlying library that must be properly configured.

Terminology

The key elements for client-side keys requires the constructions of two "stores" on the client side. The server actually has a similar set of stores, but the client need not be concerned with those.

Initial Steps

The first step is to obtain authorization from ESG. Note that this information may evolve over time, and may be out of date. This discussion is in terms of BADC and NCSA. You will need to substitute as necessary.
  1. Register at http://badc.nerc.ac.uk/register to obtain access to badc and to obtain an openid, which will looks something like:
    https://ceda.ac.uk/openid/Firstname.Lastname
  2. Ask BADC for access to whatever datasets are of interest.

  3. Obtain short term credentials at http://grid.ncsa.illinois.edu/myproxy/MyProxyLogon/ You will need to download and run the MyProxyLogon program. This will create a keyfile in, typically, the directory ".globus". The keyfile will have a name similar to this: "x509up_u13615" The other elements in ".globus" are certificates to use in validating the certificate your client gets from the server.

  4. Obtain the program source ImportKey.java from this location: http://www.agentbob.info/agentbob/79-AB.html (read the whole page, it will help you understand the remaining steps).

Building the KeyStore

You will have to modify the keyfile in the previous step and then create a keystore and install the key and a certificate. The commands are these:
    openssl pkcs8 -topk8 -nocrypt -in x509up_u13615 -inform PEM -out key.der -outform DER

    openssl x509 -in x509up_u13615 -inform PEM -out cert.der -outform DER

    java -classpath  -Dkeypassword="" -Dkeystore=./ key.der cert.der
Note, the file names "key.der" and "cert.der" can be whatever you choose. It is probably best to leave the .der extension, though.

Building the TrustStore

Building the truststore is a bit tricky because as provided, the certificates in ".globus" need some massaging. See the script below for the details. The primary command is this, which is executed for every certificate, c, in globus. It sticks the certificate into the file named "truststore"
  keytool -trustcacerts -storepass "password" -v -keystore "truststore"  -importcert -file "${c}"

Running the C Client

Refer to the section on Client-Side Certificates. The keys specified there must be set in the rc file to support ESG access. Of course, the file paths above are suggestions only; you can modify as needed. The HTTP.SSL.CERTIFICATE and HTTP.SSL.KEY entries should have same value, which is the file path for the certificate produced by MyProxyLogon. The HTTP.SSL.CAPATH entry should be the path to the "certificates" directory produced by MyProxyLogon.

As noted, also uses re-direction based authentication. So, when it receives an initial connection from a client, it redirects to a separate authentication server. When that server has authenticated the client, it redirects back to the original url to complete the request.

Script for creating Stores

The following script shows in detail how to actually construct the key and trust stores. It is specific to the format of the globus file as it was when ESG support was first added. It may have changed since then, in which case, you will need to seek some help in fixing this script. It would help if you communicated what you changed to the author so this document can be updated.
#!/bin/sh -x
KEYSTORE="esgkeystore"
TRUSTSTORE="esgtruststore"
GLOBUS="globus"
TRUSTROOT="certificates"
CERT="x509up_u13615"
TRUSTROOTPATH="$GLOBUS/$TRUSTROOT"
CERTFILE="$GLOBUS/$CERT"
PWD="password"

D="-Dglobus=$GLOBUS"
CCP="bcprov-jdk16-145.jar" 
CP="./build:${CCP}" 
JAR="myproxy.jar"

# Initialize needed directories
rm -fr build
mkdir build
rm -fr $GLOBUS
mkdir $GLOBUS
rm -f $KEYSTORE
rm -f $TRUSTSTORE

# Compile MyProxyCmd and ImportKey
javac -d ./build -classpath "$CCP" *.java
javac -d ./build ImportKey.java

# Execute MyProxyCmd
java -cp "$CP myproxy.MyProxyCmd

# Build the keystore
openssl pkcs8 -topk8 -nocrypt -in $CERTFILE -inform PEM -out key.der -outform DER
openssl x509 -in $CERTFILE -inform PEM -out cert.der -outform DER
java -Dkeypassword=$PWD -Dkeystore=./${KEYSTORE} -cp ./build ImportKey key.der cert.der

# Clean up the certificates in the globus directory
for c in ${TRUSTROOTPATH}/*.0 ; do
    alias=`basename $c .0`
    sed -e '0,/---/d' <$c >/tmp/${alias}
    echo "-----BEGIN CERTIFICATE-----" >$c       
    cat /tmp/${alias} >>$c
done
 
# Build the truststore
for c in ${TRUSTROOTPATH}/*.0 ; do
    alias=`basename $c .0`
    echo "adding: $TRUSTROOTPATH/${c}"
    echo "alias: $alias"
    yes | keytool -trustcacerts -storepass "$PWD" -v -keystore ./$TRUSTSTORE -alias $alias -importcert -file "${c}"
done
exit