Forum OpenACS Development: Announcement: NaviServer 4.99.24 available
I am pleased to announce the availability of NaviServer 4.99.24 [1,2]. This release contains various fixes and enhancements. In short, it contains improved security features (reject invalid UTF-8), fallback charsets, Cookie support avoiding potential problems with more restrictive browsers, better network filtering support, improved Oracle support for new Oracle versions, etc.
The summary of changes is listed below. This release was tested with Ubuntu 20.04, Rocky Linux 8.5, FreeBSD 13.0, macOS 11.6.6.
The following people have contributed to this release:
- Antonio Pisano
- David Osborne
- Gustaf Neumann
- Zoran Vasiljevic
The upgrade to 4.99.24 is recommended. The installation script install-ns is updated to point to the new version.
-gustaf neumann
[1] https://sourceforge.net/projects/naviserver/files/naviserver/4.99.24/
[2] https://bitbucket.org/naviserver/naviserver/
======================================= NaviServer 4.99.24, released 2022-06-14 ======================================= 77 files changed, 3242 insertions(+), 1100 deletions(-) New Features: - Improved security * Added protection against certain attacks in ns_dbquotevalue Due to the corrected conversion to external UTF-8 in db-output, new potential attack vectors appeared that were protected earlier via the Tcl-internal 'modified UTF-8'. E.g., the binary null character is stored as an overlong (two-byte) encoding of null (0xc0 0x80), so that an actual (embedded) null byte (0x00) never appears in the string. Due to the conversion, the internal representation is translated back to the binary null character. Embedded null byte characters can lead to non-terminated string literals via ns_dbquotevalue. In the updated version of NaviServer, ns_dbquotevalue raises an exception when this occurs. Therefore, the function can be used as well as an input checker (together with "try"). * Raise an exception when trying to use "ns_urldecode" to produce invalid UTF-8 Background: several (external) functions expect valid UTF-8 to be passed in and crash if this is not the case. One such example is tDOM. These nasty byte sequences are used more intensively by vulnerability scanners. Therefore, ns_urldecode raises now an exception, when it tries to convert to invalid UTF-8. It is still possible to use ns_urldecode to convert to other charsets. ns_urldecode -charset iso8859-1 -part path "/mot%C3or" When urldecode() is called internally and would produce invalid UTF-8, it truncates the string (and writes a warning to the system log). Note that the new fallback charset feature might avoid such exceptions (see below). - Fallback charsets In case, a conversion to UTF-8 fails due to invalid byte sequences, one can now provide a fallback charset for a second attempt of decoding this data. This feature is useful for websites that have to deal with requests containing invalid (form) data, typically from legacy applications. The fallback charset can be provided as optional parameter "-fallbackcharset" for the command "ns_getform", "ns_parsequery" and "ns_urldecode": ns_getform ?-fallbackcharset fallbackcharset? ?charset? ns_parsequery ?-charset charset? ?-fallbackcharset fallbackcharset? ?--? querystring ns_urldecode ?-charset charset? ?-fallbackcharset fallbackcharset? ?-part part? ?--? string In case, the parameter is not specified, it can be also be provided to the form-processing commands "ns_getform" and "ns_parsequery" via configuration variables: * per-server configuration parameter "formFallbackCharset" (in the section "ns/server/$server"), or as * global server configuration parameter "formFallbackCharset" (in the section "ns/parameters"). The highest precedence has the optional parameter, followed by the per-server configuration parameter and the global configuration parameter. - Provide a hint when cache-entry is too large for caching Background: the size of the entry is typically determined after the execution of a potentially expensive query. During the eval of the command, the cache entry is locked and forces a serialization. However, this means that in these cases the situation is worse than without a cache, where some queries can be executed in parallel. We faced the situation of an expected slowdown of the server with many "create entry collision", where due to application matters, an entry was becoming too large. This situation is not easy to debug, especially in stress situations. The log message would have helped a log to identify the cause. - Added support for multibyte numeric entities This change supports conversion of numeric entities representing multibyte characters into HTML in "ns_striphtml" and "ns_unquotehtml". Technically, the numeric entities represent Unicode code points, which are transformed into UTF-8 serialization. Every entity represents a single code point; The values can be provided in decimal or hexadecimal notation. Before this change, only single byte numeric entities were supported. ASCII control characters (decimal 0-31) are ignored as before. - New and extended commands: * ns_unquotehtml /text/ This command is the inverse operation of "ns_quotehtml". It replaces the named and numeric entities in the provided string with the native values. The command is similar to "ns_striphtml", but "ns_striphtml" removes as well other HTML markup which might not be desired in all cases. This change fixes as well a bug with numeric entities (the old code assumed, these are starting directly with a number after the ampersand) and it adds support for numeric entities with hexadecimal values (so far with the same value range as for decimal numeric entities). * ns_subnetmatch /subnet/ /ipaddr/ Determine, if a provided IP address (IPv4 or IPv6) is included in a subnet specification, which is provided in CIDR notation. The command makes internal NaviServer functionality available at the Tcl level. The regression test was extended to cover the functionality. The command ns_subnetmatch validates the provided subnet specification (IPv4 or IPv6 address followed by slash and number of significant bits) and the provided IP address and tests whether the IP address is in the implied range. The command returns a boolean value as the result. When comparing an IPv4 and IPv6 address/CIDR specification or vice versa, the result is always false. The function can be use when e.g. restricting access to certain functionality to some subnets. The function can be used as well to check, whether an IP address is an IPv4 or IPv6 address. Examples: % ns_subnetmatch 137.208.0.0/16 137.208.116.31 1 % ns_subnetmatch 137.208.0.0/16 112.207.16.33 0 % ns_subnetmatch 2001:628:404:74::31/64 [ns_conn peeraddr] ... # Is IP address a valid IPv6 address? % ns_subnetmatch ::/0 $ip # Is IP address a valid IPv4 address? % ns_subnetmatch 0.0.0.0/0 $ip * ns_connchan: Added new subcommand "ns_connchan connect" "ns_connchan connect" is similar to "ns_connchan open", except that it does not send an HTTP request (HTTP method, URL, and header fields) but just opens the connection. It can be used for some non-HTTP communication over TCP and TLS over the ns_connchan infrastructure. * ns_parseheader, Ns_ParseHeader(): return the field number (index) of the parsed entry Previously, there was no explicit feedback, what field of an "ns_set" has been parsed by "ns_parseheader". Now, in success cases, the function returns the index of the new/modified entry. This function made it possible to generalize and simplify the Tcl-level parsing of "multipart/form-data" significantly. Additionally, a new optional argument "-prefix" was added. When specified, it adds the specified prefix to the key. * ns_setcookie, ns_deletecookie Mozilla and Chrome changed the default value for SameSite of cookies from "none" to "lax" in February and Aug 2020. Cookies that explicitly set SameSite=None must also set the "Secure" attribute. In order to mirror this change of policy in NaviServer and to reduce necessary code changes, the default behavior for setting or deleting cookie is now samesite "lax" (when "-samesite" is not explicitly specified). When trying to set a cookie with "-samesite none" without the "-secure" flag, a warning is generated, and the "-samesite lax" is assumed, since major browsers announced that these will reject these cookies soon. API changes: - ns_getform, ns_parsequery, and ns_urldecode New optional parameter "-fallbackcharset". See above for details. - ns_parsequery: added option "-charset" and raise exception on failure The new option "-charset" can be used to add a charset for the result encoding of the passed-in HTTP query. In case the charset is UTF8 (default on most platforms), and the content is invalid UTF-8, an exception is raised (similar to ns_urldecode). This can be addressed by parameter "-fallbackcharset" (See above for details). - ns_deletecookie: added support "-samesite" flag for ns_deletecookie. Since "ns_deletecookie" sets internally a cookie, some browsers might ignore in the future certain cookie requests (e.g. when "-samesite" is not used or set to "none" on non-secure connections). - ns_trim enhancements: The new option "-prefix ..." can be used to strip a string (such as ">> ") from every line starting with it. - Potential incompatibilities * "ns_urldecode" and "ns_getform" will raise an exception when invalid UTF-8 data is tried to be interpreted as UTF-8 and no fallback charset is provided. Invalid UTF-8 data causes trouble with external components such as Tdom or databases and opens vulnerability vectors. Performance Improvements: - Improve "cachingmode none" "ns_cache_eval" works as follows: 1) create a temporary cache entry for the key 2) lock the cache-key (to avoid multiple parallel executions) 3) execute the query 4) store the result for the entry on success 5) unlock the cache-key Previously, "cachingmode none" was simply avoiding to store the cached values (step 4), but was serializing calls for a cache key as in default caching modes. This was leading easily to cache entry collisions. Now, "cachingmode none" is avoiding all steps 1..5 (therefore no serialization and no cache collisions). See also: https://openacs.org/forums/message-view?message_id=5665480 Bug Fixes: - Improved robustness of "ns_parseurl" for handling query parameters and fragments for partial URLs * fix over-eager collecting of URL components in tail * extended regression test - Fixed Ns_ResetFileVec NOT to invalidate residual Ns_FileVec buffer.q (caused problems under Windows). - ns_striphtml: Fixed probably very old bug for markup immediately after an entity This bug fix handles cases, where e.g. two entities are in a text right next to each other, like e.g. in the string "hello<>world". The old code was correctly decoding the first entity, but output the second one literally. - Fixed compilation for C++, which was introduced in 4.99.23 to avoid usage of reserved C identifiers Many thanks to Brendan Graves for reporting the problem. - Added missing named entities "apos" and "quote". These have been missing since ages. - Provide an error message when the configured locale is not installed on the host. This change causes NaviServer to abort, when the configured locale is not installed on the host. Typically, this locale is e.g. used by "ns_strcoll" for determining the default collating order. The configuration file for the regression testing sets the environment variable LANG to "en_US.UTF-8". This means that for running the stock regression test, this locale must be installed on the system. Before this change, NaviServer could crash at runtime when trying to access the default locale (as e.g. in "ns_strcoll") - Added support for "_charset_" field for default charset in multipart/form-data (RFC 7578, section 4.6) RFC 7578 (July 2015) defines an optional "_charset_" entry in the form (typically provided as hidden form field) to specify the charset of text entries. This is now supported as well by NaviServer. This is apparently a seldom used feature. Documentation improvements: --------------------------- - Improved the following man pages: doc/src/manual/admin-install.man doc/src/naviserver/ns_conn.man doc/src/naviserver/ns_connchan.man doc/src/naviserver/ns_cookie.man doc/src/naviserver/ns_crypto.man doc/src/naviserver/ns_getform.man doc/src/naviserver/ns_http.man doc/src/naviserver/ns_httptime.man doc/src/naviserver/ns_log.man doc/src/naviserver/ns_parseheader.man doc/src/naviserver/ns_parsequery.man doc/src/naviserver/ns_parseurl.man doc/src/naviserver/ns_rlimit.man doc/src/naviserver/ns_urldecode.man doc/src/naviserver/ns_urlencode.man doc/src/naviserver/ns_valid_utf8.man doc/src/naviserver/textutil-cmds.man nsdb/doc/mann/ns_db.man Configuration Changes: ---------------------- - Updated OpenACS sample configuration file * reflect recent Oracle (tested with Oracle 19c) * added documentation for "StaticCSP", "CookieNamespace", "NsShutdownWithNonZeroExitCode", "LogIncludeUserId" Code Changes: ------------- - Set Tcl error code "NS_INVALID_UTF8" for errors due to invalid UTF-8 - Changed Tcl error code "NSCACHE" to "NS_CACHE". Now all NaviServer-specific error codes start with the prefix "NS_". - Extended regression test - Improve Tcl version compatibility * Removed -DTCL_NO_DEPRECATED from default CFLAGS to cope with recent deprecation in Tcl 8.7a5 - Code Cleanup . Do not declare reserved C identifiers . Improved type cleanness . Refactored file-based multipart form parser to make logic explicit (Tcl code) - Improved comments, fixed typos - Marked "ns_set_precision" as deprecated, since there is no reason why not setting the Tcl variable ::tcl_precision directly. - Don't hard-wire port for HTTPS testing to 8443 The setup code looks now for a free port for HTTPS connections starting with 8443, and remembers the free port in the configuration value "tls_listenport" and "tls_listenurl". This is now fully analogous to the setup of the plain HTTP testing (setting "listenport" and "listenurl") - Silence warning with recent versions of gcc when certain values of _FORTIFY_SOURCE/-Wstringop-overflow are set. Changes in NaviServer Modules: ============================== 22 files changed, 8447 insertions(+), 1429 deletions(-) - general: Updated obsolete ChangeLog files and replaced these by automatically generated ones. nsdbpg: ------- - Raise exception when a value for a bind variable contains a NUL character. This value is explicitly forbidden in text strings passed to PostgreSQL. - Let "ns_pg" report available subcommands even when handle is not specified. This makes the command compatible with the "icanuse" feature in OpenACS. nsoracle: --------- - Added support for output columns of type SQLT_TIMESTAMP or SQLT_TIMESTAMP_TZ This change fixes a bug, where SQL queries of the form SELECT TO_TIMESTAMP(sysdate) FROM dual lead to errors for the form: Database operation "getrow" failed (exception 1406, "nsoracle.c:3659:Ns_OracleGetRow: error in `OCIStmtFetch ()': ORA-01406: fetched column value was truncated The driver needs for several output types special rules, where the timestamp cases were not supported so far. It is also recommended to set the according environment variables specifying the output format in the configuration server of NaviServer, such as e.g. set ::env(NLS_TIMESTAMP_FORMAT) "YYYY-MM-DD HH24:MI:SS.FF6" set ::env(NLS_TIMESTAMP_TZ_FORMAT) "YYYY-MM-DD HH24:MI:SS.FF6 TZH:TZM" For testing in you local Oracle installation, you might test the output formats (and the required sizes with the following snippet for sqlplus: COLUMN localtimestamp format a40 COLUMN systimestamp format a40 COLUMN ts_bytes format a80 alter session set nls_timestamp_format='YYYY-MM-DD HH24:MI:SS.FF6'; select localtimestamp, length(localtimestamp), dump(localtimestamp) ts_bytes from dual; alter session set nls_timestamp_tz_format='YYYY-MM-DD HH24:MI:SS.FF6 TZH:TZM'; select systimestamp, length(systimestamp), dump(systimestamp) ts_bytes from dual; alter session set nls_timestamp_tz_format='YYYY-MM-DD HH24:MI:SS.FF6 TZR'; select systimestamp, length(systimestamp), dump(systimestamp) ts_bytes from dual; - Code cleanup, ensure silent compilation with standard compiler settings - Improved spelling ECDSA letsencrypt: ------------ - Added option to produce certificates with ECDSA Prior to this change, all certificates were using RSA keys. Since a while, keys based on elliptic curves became the preference of letsencrypt. - Improved spelling