curl/packages/OS400
Peter Wu 970c22f970 libcurl: add UNIX domain sockets support
The ability to do HTTP requests over a UNIX domain socket has been
requested before, in Apr 2008 [0][1] and Sep 2010 [2]. While a
discussion happened, no patch seems to get through. I decided to give it
a go since I need to test a nginx HTTP server which listens on a UNIX
domain socket.

One patch [3] seems to make it possible to use the
CURLOPT_OPENSOCKETFUNCTION function to gain a UNIX domain socket.
Another person wrote a Go program which can do HTTP over a UNIX socket
for Docker[4] which uses a special URL scheme (though the name contains
cURL, it has no relation to the cURL library).

This patch considers support for UNIX domain sockets at the same level
as HTTP proxies / IPv6, it acts as an intermediate socket provider and
not as a separate protocol. Since this feature affects network
operations, a new feature flag was added ("unix-sockets") with a
corresponding CURL_VERSION_UNIX_SOCKETS macro.

A new CURLOPT_UNIX_SOCKET_PATH option is added and documented. This
option enables UNIX domain sockets support for all requests on the
handle (replacing IP sockets and skipping proxies).

A new configure option (--enable-unix-sockets) and CMake option
(ENABLE_UNIX_SOCKETS) can disable this optional feature. Note that I
deliberately did not mark this feature as advanced, this is a
feature/component that should easily be available.

 [0]: http://curl.haxx.se/mail/lib-2008-04/0279.html
 [1]: http://daniel.haxx.se/blog/2008/04/14/http-over-unix-domain-sockets/
 [2]: http://sourceforge.net/p/curl/feature-requests/53/
 [3]: http://curl.haxx.se/mail/lib-2008-04/0361.html
 [4]: https://github.com/Soulou/curl-unix-socket

Signed-off-by: Peter Wu <peter@lekensteyn.nl>
2014-12-04 02:52:19 +01:00
..
ccsidcurl.c Adjust OS400-specific support to last release 2014-11-05 12:42:24 +01:00
ccsidcurl.h OS400: coding style standards 2013-10-28 12:00:22 +01:00
curl.inc.in libcurl: add UNIX domain sockets support 2014-12-04 02:52:19 +01:00
initscript.sh OS400: make it compilable again. Make RPG binding up to date. 2014-06-18 20:22:44 +02:00
make-include.sh OS400: sync RPG wrapper, zlib support, fix header file names, ... 2013-10-25 18:37:37 +02:00
make-lib.sh vtls: remove QsoSSL 2014-10-13 16:33:47 +02:00
make-src.sh remove the CVSish $Id$ lines 2010-03-24 11:02:54 +01:00
make-tests.sh OS400: sync RPG wrapper, zlib support, fix header file names, ... 2013-10-25 18:37:37 +02:00
makefile.sh config-os400.h: enable system strdup(), strcmpi(), etc. 2013-07-15 18:38:17 +02:00
os400sys.c OS400: fix bugs in curl_*escape_ccsid() and reduce variables scope 2014-10-14 15:43:25 +02:00
os400sys.h OS400: sync RPG wrapper, zlib support, fix header file names, ... 2013-10-25 18:37:37 +02:00
README.OS400 vtls: remove QsoSSL 2014-10-13 16:33:47 +02:00

Implementation notes:

  This is a true OS/400 implementation, not a PASE implementation (for PASE,
use AIX implementation).

  The biggest problem with OS/400 is EBCDIC. Libcurl implements an internal
conversion mechanism, but it has been designed for computers that have a
single native character set. OS/400 default native character set varies
depending on the country for which it has been localized. And more, a job
may dynamically alter its "native" character set.
  Several characters that do not have fixed code in EBCDIC variants are
used in libcurl strings. As a consequence, using the existing conversion
mechanism would have lead in a localized binary library - not portable across
countries.
  For this reason, and because libcurl was originally designed for ASCII based
operating systems, the current OS/400 implementation uses ASCII as internal
character set. This has been accomplished using the QADRT library and
include files, a C and system procedures ASCII wrapper library. See IBM QADRT
description for more information.
  This then results in libcurl being an ASCII library: any function string
argument is taken/returned in ASCII and a C/C++ calling program built around
QADRT may use libcurl functions as on any other platform.
  QADRT does not define ASCII wrappers for all C/system procedures: the
OS/400 configuration header file and an additional module (os400sys.c) define
some more of them, that are used by libcurl and that QADRT left out.
  To support all the different variants of EBCDIC, non-standard wrapper
procedures have been added to libcurl on OS/400: they provide an additional
CCSID (numeric Coded Character Set ID specific to OS/400) parameter for each
string argument. String values passed to callback procedures are NOT converted,
so text gathered this way is (probably !) ASCII.

  Another OS/400 problem comes from the fact that the last fixed argument of a
vararg procedure may not be of type char, unsigned char, short or unsigned
short. Enums that are internally implemented by the C compiler as one of these
types are also forbidden. Libcurl uses enums as vararg procedure tagfields...
Happily, there is a pragma forcing enums to type "int". The original libcurl
header files are thus altered during build process to use this pragma, in
order to force libcurl enums of being type int (the pragma disposition in use
before inclusion is restored before resuming the including unit compilation).

  Secure socket layer is provided by the IBM GSKit API: unlike other SSL
implementations, GSKit is based on "certificate stores" or keyrings
rather than individual certificate/key files. Certificate stores, as well as
"certificate labels" are managed by external IBM-defined applications.
  There are two ways to specify an SSL context:
- By an application identifier.
- By a keyring file pathname and (optionally) certificate label.
  To identify an SSL context by application identifier, use option
SETOPT_SSLCERT to specify the application identifier.
  To address an SSL context by keyring and certificate label, use CURLOPT_CAINFO
to set-up the keyring pathname, CURLOPT_SSLCERT to define the certificate label
(omitting it will cause the default certificate in keyring to be used) and
CURLOPT_KEYPASSWD to give the keyring password. If SSL is used without
defining any of these options, the default (i.e.: system) keyring is used for
server certificate validation.

  Non-standard EBCDIC wrapper prototypes are defined in an additional header
file: ccsidcurl.h. These should be self-explanatory to an OS/400-aware
designer. CCSID 0 can be used to select the current job's CCSID.
  Wrapper procedures with variable arguments are described below:

_ curl_easy_setopt_ccsid()
  Variable arguments are a string pointer and a CCSID (unsigned int) for
options:
        CURLOPT_CAINFO
        CURLOPT_CAPATH
        CURLOPT_COOKIE
        CURLOPT_COOKIEFILE
        CURLOPT_COOKIEJAR
        CURLOPT_COOKIELIST
        CURLOPT_COPYPOSTFIELDS
        CURLOPT_CRLFILE
        CURLOPT_CUSTOMREQUEST
        CURLOPT_DNS_SERVERS
        CURLOPT_EGDSOCKET
        CURLOPT_ENCODING
        CURLOPT_FTP_ACCOUNT
        CURLOPT_FTP_ALTERNATIVE_TO_USER
        CURLOPT_FTPPORT
        CURLOPT_INTERFACE
        CURLOPT_ISSUERCERT
        CURLOPT_KEYPASSWD
        CURLOPT_KRBLEVEL
        CURLOPT_LOGIN_OPTIONS
        CURLOPT_MAIL_FROM
        CURLOPT_MAIL_AUTH
        CURLOPT_NETRC_FILE
        CURLOPT_NOPROXY
        CURLOPT_PASSWORD
        CURLOPT_PINNEDPUBLICKEY
        CURLOPT_PROXY
        CURLOPT_PROXYPASSWORD
        CURLOPT_PROXYUSERNAME
        CURLOPT_PROXYUSERPWD
        CURLOPT_RANDOM_FILE
        CURLOPT_RANGE
        CURLOPT_REFERER
        CURLOPT_RTSP_SESSION_UID
        CURLOPT_RTSP_STREAM_URI
        CURLOPT_RTSP_TRANSPORT
        CURLOPT_SOCKS5_GSSAPI_SERVICE
        CURLOPT_SSH_HOST_PUBLIC_KEY_MD5 Note: SSH not available on OS400.
        CURLOPT_SSH_KNOWNHOSTS          Note: SSH not available on OS400.
        CURLOPT_SSH_PRIVATE_KEYFILE     Note: SSH not available on OS400.
        CURLOPT_SSH_PUBLIC_KEYFILE      Note: SSH not available on OS400.
        CURLOPT_SSLCERT
        CURLOPT_SSLCERTTYPE
        CURLOPT_SSL_CIPHER_LIST
        CURLOPT_SSLENGINE
        CURLOPT_SSLKEY
        CURLOPT_SSLKEYTYPE
        CURLOPT_TLSAUTH_PASSWORD
        CURLOPT_TLSAUTH_TYPE
        CURLOPT_TLSAUTH_USERNAME
        CURLOPT_URL
        CURLOPT_USERAGENT
        CURLOPT_USERNAME
        CURLOPT_USERPWD
        CURLOPT_XOAUTH2_BEARER
  Else it is the same as for curl_easy_setopt().
  Note that CURLOPT_ERRORBUFFER is not in the list above, since it gives the
address of an (empty) character buffer, not the address of a string.
CURLOPT_POSTFIELDS stores the address of static binary data (of type void *) and
thus is not converted. If CURLOPT_COPYPOSTFIELDS is issued after
CURLOPT_POSTFIELDSIZE != -1, the data size is adjusted according to the
CCSID conversion result length.

_ curl_formadd_ccsid()
  In the variable argument list, string pointers should be followed by a (long)
CCSID for the following options:
        CURLFORM_FILENAME
        CURLFORM_CONTENTTYPE
        CURLFORM_BUFFER
        CURLFORM_FILE
        CURLFORM_FILECONTENT
        CURLFORM_COPYCONTENTS
        CURLFORM_COPYNAME
        CURLFORM_PTRNAME
  If taken from an argument array, an additional array entry must follow each
entry containing one of the above option. This additional entry holds the CCSID
in its value field, and the option field is meaningless.
  It is not possible to have a string pointer and its CCSID across a function
parameter/array boundary.
  Please note that CURLFORM_PTRCONTENTS and CURLFORM_BUFFERPTR are considered
unconvertible strings and thus are NOT followed by a CCSID.

_ curl_easy_getinfo_ccsid
  The following options are followed by a 'char * *' and a CCSID. Unlike
curl_easy_getinfo(), the value returned in the pointer should be freed after
use:
        CURLINFO_EFFECTIVE_URL
        CURLINFO_CONTENT_TYPE
        CURLINFO_FTP_ENTRY_PATH
        CURLINFO_REDIRECT_URL
        CURLINFO_PRIMARY_IP
        CURLINFO_RTSP_SESSION_ID
        CURLINFO_LOCAL_IP
  Likewise, the following options are followed by a struct curl_slist * * and a
CCSID.
        CURLINFO_SSL_ENGINES
        CURLINFO_COOKIELIST
Lists returned should be released with curl_slist_free_all() after use.
  Option CURLINFO_CERTINFO is followed by a struct curl_certinfo * * and a
CCSID. Returned structures sould be free'ed using curl_certinfo_free_all() after
use.
  Other options are processed like in curl_easy_getinfo().

  Standard compilation environment does support neither autotools nor make;
in fact, very few common utilities are available. As a consequence, the
config-os400.h has been coded manually and the compilation scripts are
a set of shell scripts stored in subdirectory packages/OS400.

  The "curl" command and the test environment are currently not supported on
OS/400.


Protocols currently implemented on OS/400:
_ DICT
_ FILE
_ FTP
_ FTPS
_ FTP with secure transmission
_ GOPHER
_ HTTP
_ HTTPS
_ IMAP
_ IMAPS
_ IMAP with secure transmission
_ LDAP
_ POP3
_ POP3S
_ POP3 with secure transmission
_ RTSP
_ SMTP
_ SMTPS
_ SMTP with secure transmission
_ TELNET
_ TFTP



Compiling on OS/400:

  These instructions targets people who knows about OS/400, compiling, IFS and
archive extraction. Do not ask questions about these subjects if you're not
familiar with.

_ As a prerequisite, QADRT development environment must be installed.
_ Install the curl source directory in IFS.
_ Enter shell (QSH)
_ Change current directory to the curl installation directory
_ Change current directory to ./packages/OS400
_ Edit file iniscript.sh. You may want to change tunable configuration
  parameters, like debug info generation, optimisation level, listing option,
  target library, etc.
_ Copy any file in the current directory to makelog (i.e.:
  cp initscript.sh makelog): this is intended to create the makelog file with
  an ASCII CCSID!
_ Enter the command "sh makefile.sh > makelog 2>&1'
_ Examine the makelog file to check for compilation errors.

  Leaving file initscript.sh unchanged, this will produce the following OS/400
objects:
_ Library CURL. All other objects will be stored in this library.
_ Modules for all libcurl units.
_ Binding directory CURL_A, to be used at calling program link time for
  statically binding the modules (specify BNDSRVPGM(QADRTTS QGLDCLNT QGLDBRDR)
  when creating a program using CURL_A).
_ Service program CURL.<soname>, where <soname> is extracted from the
  lib/Makefile.am VERSION variable. To be used at calling program run-time
  when this program has dynamically bound curl at link time.
_ Binding directory CURL. To be used to dynamically bind libcurl when linking a
  calling program.
_ Source file H. It contains all the include members needed to compile a C/C++
  module using libcurl, and an ILE/RPG /copy member for support in this
  language.
_ Standard C/C++ libcurl include members in file H.
_ CCSIDCURL member in file H. This defines the non-standard EBCDIC wrappers for
  C and C++.
_ CURL.INC member in file H. This defines everything needed by an ILE/RPG
  program using libcurl.
_ LIBxxx modules and programs. Although the test environment is not supported
  on OS/400, the libcurl test programs are compiled for manual tests.



Special programming consideration:

QADRT being used, the following points must be considered:
_ If static binding is used, service program QADRTTS must be linked too.
_ The EBCDIC CCSID used by QADRT is 37 by default, NOT THE JOB'S CCSID. If
  another EBCDIC CCSID is required, it must be set via a locale through a call
  to setlocale_a (QADRT's setlocale() ASCII wrapper) with category LC_ALL or
  LC_CTYPE, or by setting environment variable QADRT_ENV_LOCALE to the locale
  object path before executing the program.
_ Do not use original source include files unless you know what you are doing.
  Use the installed members instead (in /QSYS.LIB/CURL.LIB/H.FILE).



ILE/RPG support:

  Since 95% of the OS/400 programmers use ILE/RPG exclusively, a definition
  /COPY member is provided for this language. To include all libcurl
  definitions in an ILE/RPG module, line

     h bnddir('CURL/CURL')

must figure in the program header, and line

     d/copy curl/h,curl.inc

in the global data section of the module's source code.

  No vararg procedure support exists in ILE/RPG: for this reason, the following
considerations apply:
_ Procedures curl_easy_setopt_long(), curl_easy_setopt_object(),
  curl_easy_setopt_function() and curl_easy_setopt_offset() are all alias
  prototypes to curl_easy_setopt(), but with different parameter lists.
_ Procedures curl_easy_getinfo_string(), curl_easy_getinfo_long(),
  curl_easy_getinfo_double() and curl_easy_getinfo_slist() are all alias
  prototypes to curl_easy_getinfo(), but with different parameter lists.
_ Procedures curl_multi_setopt_long(), curl_multi_setopt_object(),
  curl_multi_setopt_function() and curl_multi_setopt_offset() are all alias
  prototypes to curl_multi_setopt(), but with different parameter lists.
_ The prototype of procedure curl_formadd() allows specifying a pointer option
  and the CURLFORM_END option. This makes possible to use an option array
  without any additional definition. If some specific incompatible argument
  list is used in the ILE/RPG program, the latter must define a specialised
  alias. The same applies to curl_formadd_ccsid() too.

  Since RPG cannot cast a long to a pointer, procedure curl_form_long_value()
is provided for that purpose: this allows storing a long value in the curl_forms
array.