-Change log file for Exim from version 4.21
--------------------------------------------
+This document describes *changes* to previous versions, that might
+affect Exim's operation, with an unchanged configuration file. For new
+options, and new features, see the NewStuff file next to this ChangeLog.
+
+
+Exim version 4.94
+-----------------
+
+JH/01 Avoid costly startup code when not strictly needed. This reduces time
+ for some exim process initialisations. It does mean that the logging
+ of TLS configuration problems is only done for the daemon startup.
+
+JH/02 Early-pipelining support code is now included unless disabled in Makefile.
+
+JH/03 DKIM verification defaults no long accept sha1 hashes, to conform to
+ RFC 8301. They can still be enabled, using the dkim_verify_hashes main
+ option.
+
+JH/04 Support CHUNKING from an smtp transport using a transport_filter, when
+ DKIM signing is being done. Previously a transport_filter would always
+ disable CHUNKING, falling back to traditional DATA.
+
+JH/05 Regard command-line receipients as tainted.
+
+JH/06 Bug 340: Remove the daemon pid file on exit, whe due to SIGTERM.
+
+JH/07 Bug 2489: Fix crash in the "pam" expansion condition. It seems that the
+ PAM library frees one of the arguments given to it, despite the
+ documentation. Therefore a plain malloc must be used.
+
+JH/08 Bug 2491: Use tainted buffers for the transport smtp context. Previously
+ on-stack buffers were used, resulting in a taint trap when DSN information
+ copied from a received message was written into the buffer.
+
+JH/09 Bug 2493: Harden ARC verify against Outlook, whick has been seen to mix
+ the ordering of its ARC headers. This caused a crash.
+
+JH/10 Bug 2492: Use tainted memory for retry record when needed. Previously when
+ a new record was being constructed with information from the peer, a trap
+ was taken.
+
+JH/11 Bug 2494: Unset the default for dmarc_tld_file. Previously a naiive
+ installation would get error messages from DMARC verify, when it hit the
+ nonexistent file indicated by the default. Distros wanting DMARC enabled
+ should both provide the file and set the option.
+ Also enforce no DMARC verification for command-line sourced messages.
+
+JH/12 Fix an uninitialised flag in early-pipelining. Previously connections
+ could, depending on the platform, hang at the STARTTLS response.
+
+JH/13 Bug 2498: Reset a counter used for ARC verify before handling another
+ message on a connection. Previously if one message had ARC headers and
+ the following one did not, a crash could result when adding an
+ Authentication-Results: header.
+
+JH/14 Bug 2500: Rewind some of the common-coding in string handling between the
+ Exim main code and Exim-related utities. The introduction of taint
+ tracking also did many adjustments to string handling. Since then, eximon
+ frequently terminated with an assert failure.
+
+JH/15 When PIPELINING, synch after every hundred or so RCPT commands sent and
+ check for 452 responses. This slightly helps the inefficieny of doing
+ a large alias-expansion into a recipient-limited target. The max_rcpt
+ transport option still applies (and at the current default, will override
+ the new feature). The check is done for either cause of synch, and forces
+ a fast-retry of all 452'd recipients using a new MAIL FROM on the same
+ connection. The new facility is not tunable at this time.
+
+JH/16 Fix the variables set by the gsasl authenticator. Previously a pointer to
+ library live data was being used, so the results became garbage. Make
+ copies while it is still usable.
+
+JH/17 Logging: when the deliver_time selector ise set, include the DT= field
+ on delivery deferred (==) and failed (**) lines (if a delivery was
+ attemtped). Previously it was only on completion (=>) lines.
+
+JH/18 Authentication: the gsasl driver not provides the $authN variables in time
+ for the expansion of the server_scram_iter and server_scram_salt options.
+
+WB/01 SPF: DNS lookups for the obsolete SPF RR type done by the libspf2 library
+ are now specifically given a NO_DATA response without hitting the system
+ resolver. The library goes on to do the now-standard TXT lookup.
+ Use of dnsdb lookups is not affected.
+
+JH/19 Bug 2507: Modules: on handling a dynamic-module (lookups) open failure,
+ only retrieve the errormessage once. Previously two calls to dlerror()
+ were used, and the second one (for mainlog/paniclog) retrieved null
+ information.
+
+JH/20 Taint checking: disallow use of tainted data for
+ - the appendfile transport file and directory options
+ - the pipe transport command
+ - the autoreply transport file, log and once options
+ - file names used by the redirect router (including filter files)
+ - named-queue names
+ Previously this was permitted.
+
+JH/21 Bug 2501: Fix init call in the heimdal authenticator. Previously it
+ adjusted the size of a major service buffer; this failed because the
+ buffer was in use at the time. Change to a compile-time increase in the
+ buffer size, when this authenticator is compiled into exim.
+
+JH/22 Taint-checking: move to safe-mode taint checking on all platforms. The
+ previous fast-mode was untenable in the face of glibs using mmap to
+ support larger malloc requests.
+
+PP/01 Update the openssl_options possible values through OpenSSL 1.1.1c.
+ New values supported, if defined on system where compiled:
+ allow_no_dhe_kex, cryptopro_tlsext_bug, enable_middlebox_compat,
+ no_anti_replay, no_encrypt_then_mac, prioritize_chacha, tlsext_padding
+
+JH/23 Performance improvement in the initial phase of a two-pass queue run. By
+ running a limited number of proceses in parallel, a benefit is gained. The
+ amount varies with the platform hardware and load. The use of the option
+ queue_run_in_order means we cannot do this, as ordering becomes
+ indeterminate.
+
+JH/24 Bug 2524: fix the cyrus_sasl auth driver gssapi usage. A previous fix
+ had introduced a string-copy (for ensuring NUL-termination) which was not
+ appropriate for that case, which can include embedded NUL bytes in the
+ block of data. Investigation showed the copy to actually be needless, the
+ data being length-specified.
+
+JH/25 Fix use of concurrent TLS connections under GnuTLS. When a callout was
+ done during a receiving connection, and both used TLS, global info was
+ used rather than per-connection info for tracking the state of data
+ queued for transmission. This could result in a connection hang.
+
+JH/26 Fix use of the SIZE parameter on MAIL commands, on continued connections.
+ Previously, when delivering serveral messages down a single connection
+ only the first would provide a SIZE. This was due to the size information
+ not being properly tracked.
+
+JH/27 Bug 2530: When operating in a timezone with sub-minute offset, such as
+ TAI (at 37 seconds currently), pretend to be in UTC for time-related
+ expansion and logging. Previously, spurious values such as a future
+ minute could be seen.
+
+JH/28 Bug 2533: Fix expansion of ${tr } item. When called in some situations
+ it could crash from a null-deref. This could also affect the
+ ${addresses: } operator and ${readsock } item.
+
+JH/29 Bug 2537: Fix $mime_part_count. When a single connection had a non-mime
+ message following a mime one, the variable was not reset.
+
+JH/30 When an pipelined-connect fails at the first response, assume incorrect
+ cached capability (perhaps the peer reneged?) and immediately retry in
+ non-pipelined mode.
+
+JH/31 Fix spurious detection of timeout while writing to transport filter.
+
+
+Exim version 4.93
+-----------------
+
+JH/01 OpenSSL: With debug enabled output keying information sufficient, server
+ side, to decode a TLS 1.3 packet capture.
+
+JH/02 OpenSSL: Suppress the sending of (stateful) TLS1.3 session tickets.
+ Previously the default library behaviour applied, sending two, each in
+ its own TCP segment.
+
+JH/03 Debug output for ACL now gives the config file name and line number for
+ each verb.
+
+JH/04 The default received_header_text now uses the RFC 8314 tls cipher clause.
+
+JH/05 DKIM: ensure that dkim_domain elements are lowercased before use.
+
+JH/06 Fix buggy handling of autoreply bounce_return_size_limit, and a possible
+ buffer overrun for (non-chunking) other transports.
+
+JH/07 GnuTLS: Our use of late (post-handshake) certificate verification, under
+ TLS1.3, means that a server rejecting a client certificate is not visible
+ to the client until the first read of encrypted data (typically the
+ response to EHLO). Add detection for that case and treat it as a failed
+ TLS connection attempt, so that the normal retry-in-clear can work (if
+ suitably configured).
+
+JB/01 Bug 2375: fix expansions of 822 addresses having comments in local-part
+ and/or domain. Found and fixed by Jason Betts.
+
+JH/08 Add hardening against SRV & TLSA lookups the hit CNAMEs (a nonvalid
+ configuration). If a CNAME target was not a wellformed name pattern, a
+ crash could result.
+
+JH/09 Logging: Fix initial listening-on line for multiple ports for an IP when
+ the OS reports them interleaved with other addresses.
+
+JH/10 OpenSSL: Fix aggregation of messages. Previously, when PIPELINING was
+ used both for input and for a verify callout, both encrypted, SMTP
+ responses being sent by the server could be lost. This resulted in
+ dropped connections and sometimes bounces generated by a peer sending
+ to this system.
+
+JH/11 Harden plaintext authenticator against a badly misconfigured client-send
+ string. Previously it was possible to cause undefined behaviour in a
+ library routine (usually a crash). Found by "zerons".
+
+JH/12 Bug 2384: fix "-bP smtp_receive_timeout". Previously it returned no
+ output.
+
+JH/13 Bug 2386: Fix builds with Dane under LibreSSL 2.9.0 onward. Some old
+ API was removed, so update to use the newer ones.
+
+JH/14 Bug 1891: Close the log file if receiving a non-smtp message, without
+ any timeout set, is taking a long time. Previously we would hang on to a
+ rotated logfile "forever" if the input was arriving with long gaps
+ (a previous attempt to fix addressed lack, for a long time, of initial
+ input).
+
+HS/01 Bug 2390: Use message_id for tempfile creation to avoid races in a
+ shared (NFS) environment. The length of the tempfile name is now
+ 4 + 16 ("hdr.$message_exim_id") which might break on file
+ systems which restrict the file name length to lower values.
+ (It was "hdr.$pid".)
+
+HS/02 Bug 2390: Use message_id for tempfile creation to avoid races in a
+ shared (NFS) environment.
+
+HS/03 Bug 2392: exigrep does case sensitive *option* processing (as it
+ did for all versions <4.90). Notably -M, -m, --invert, -I may be
+ affected.
+
+JH/15 Use unsigned when creating bitmasks in macros, to avoid build errors
+ on some platforms for bit 31.
+
+JH/16 GnuTLS: rework ciphersuite strings under recent library versions. Thanks
+ to changes apparently associated with TLS1.3 handling some of the APIs
+ previously used were either nonfunctional or inappropriate. Strings
+ like TLS1.3:ECDHE_SECP256R1__RSA_PSS_RSAE_SHA256__AES_256_GCM__AEAD:256
+ and TLS1.2:ECDHE_SECP256R1__RSA_SHA256__AES_128_CBC__SHA256:128 replace
+ the previous TLS1.2:ECDHE_RSA_AES_256_GCM_SHA384:256 .
+ This affects log line X= elements, the $tls_{in,out}_cipher variables,
+ and the use of specific cipher names in the encrypted= ACL condition.
+
+JH/17 OpenSSL: the default openssl_options now disables ssl_v3.
+
+JH/18 GnuTLS: fix $tls_out_ocsp under hosts_request_ocsp. Previously the
+ verification result was not updated unless hosts_require_ocsp applied.
+
+JH/19 Bug 2398: fix listing of a named-queue. Previously, even with the option
+ queue_list_requires_admin set to false, non-admin users were denied the
+ facility.
+
+JH/20 Bug 2389: fix server advertising of usable certificates, under GnuTLS in
+ directory-of-certs mode. Previously they were advertised despite the
+ documentation.
+
+JH/21 The smtp transport option "hosts_noproxy_tls" is now unset by default.
+ A single TCP connection by a client will now hold a TLS connection open
+ for multiple message deliveries, by default. Previoud the default was to
+ not do so.
+
+JH/22 The smtp transport option "hosts_try_dane" now enables all hosts by
+ default. If built with the facility, DANE will be used. The facility
+ SUPPORT_DANE is now enabled in the prototype build Makefile "EDITME".
+
+JH/23 The build default is now for TLS to be included; the SUPPORT_TLS define
+ is replaced with DISABLE_TLS. Either USE_GNUTLS or (the new) USE_OPENSSL
+ must be defined and you must still, unless you define DISABLE_TLS, manage
+ the the include-dir and library-file requirements that go with that
+ choice. Non-TLS builds are still supported.
+
+JH/24 Fix duplicated logging of peer name/address, on a transport connection-
+ reject under TFO.
+
+JH/25 The smtp transport option "hosts_try_fastopen" now enables all hosts by
+ default. If the platform supports and has the facility enabled, it will
+ be requested on all coneections.
+
+JH/26 The PIPE_CONNECT facility is promoted from experimental status and is now
+ controlled by the build-time option SUPPORT_PIPE_CONNECT.
+
+PP/01 Unbreak heimdal_gssapi, broken in 4.92.
+
+JH/27 Bug 2404: Use the main-section configuration option "dsn_from" for
+ success-DSN messages. Previously the From: header was always the default
+ one for these; the option was ignored.
+
+JH/28 Fix the timeout on smtp response to apply to the whole response.
+ Previously it was reset for every read, so a teergrubing peer sending
+ single bytes within the time limit could extend the connection for a
+ long time. Credit to Qualsys Security Advisory Team for the discovery.
+
+JH/29 Fix DSN Final-Recipient: field. Previously it was the post-routing
+ delivery address, which leaked information of the results of local
+ forwarding. Change to the original envelope recipient address, per
+ standards.
+
+JH/30 Bug 2411: Fix DSN generation when RFC 3461 failure notification is
+ requested. Previously not bounce was generated and a log entry of
+ error ignored was made.
+
+JH/31 Avoid re-expansion in ${sort } expansion. (CVE-2019-13917)
+
+JH/32 Introduce a general tainting mechanism for values read from the input
+ channel, and values derived from them. Refuse to expand any tainted
+ values, to catch one form of exploit.
+
+JH/33 Bug 2413: Fix dkim_strict option. Previously the expansion result
+ was unused and the unexpanded text used for the test. Found and
+ fixed by Ruben Jenster.
+
+JH/34 Fix crash after TLS shutdown. When the TCP/SMTP channel was left open,
+ an attempt to use a TLS library read routine dereffed a nul pointer,
+ causing a segfault.
+
+JH/35 Bug 2409: filter out-of-spec chars from callout response before using
+ them in our smtp response.
+
+JH/36 Have the general router option retry_use_local_part default to true when
+ any of the restrictive preconditions are set (to anything). Previously it
+ was only for check_local user. The change removes one item of manual
+ configuration which is required for proper retries when a remote router
+ handles a subset of addresses for a domain.
+
+JH/37 Appendfile: when evaluating quota use (non-quota_size_regex) take the file
+ link count into consideration.
+
+HS/04 Fix handling of very log lines in -H files. If a -<key> <value> line
+ caused the extension of big_buffer, the following lines were ignored.
+
+JH/38 Bug 1395: Teach the DNS negative-cache about TTL value from the SOA in
+ accordance with RFC 2308. Previously there was no expiry, so a longlived
+ receive process (eg. due to ACL delays) versus a short SOA value could
+ surprise.
+
+HS/05 Handle trailing backslash gracefully. (CVE-2019-15846)
+
+JH/39 Promote DMARC support to mainline.
+
+JH/40 Bug 2452: Add a References: header to DSNs.
+
+JH/41 With GnuTLS 3.6.0 (and later) do not attempt to manage Diffie-Hellman
+ parameters. The relevant library call is documented as "Deprecated: This
+ function is unnecessary and discouraged on GnuTLS 3.6.0 or later. Since
+ 3.6.0, DH parameters are negotiated following RFC7919."
+
+HS/06 Change the default of dnssec_request_domains to "*"
+
+JH/42 Bug 2545: Fix CHUNKING for all RCPT commands rejected. Previously we
+ carried on and emitted a BDAT command, even when PIPELINING was not
+ active.
+
+JH/43 Bug 2465: Fix taint-handling in dsearch lookup. Previously a nontainted
+ buffer was used for the filename, resulting in a trap when tainted
+ arguments (eg. $domain) were used.
+
+JH/44 With OpenSSL 1.1.1 (onwards) disable renegotiation for TLS1.2 and below;
+ recommended to avoid a possible server-load attack. The feature can be
+ re-enabled via the openssl_options main cofiguration option.
+
+JH/45 local_scan API: documented the current smtp_printf() call. This changed
+ for version 4.90 - adding a "more data" boolean to the arguments.
+ Bumped the ABI version number also, this having been missed previously;
+ release versions 4.90 to 4.92.3 inclusive were effectively broken in
+ respect of usage of smtp_printf() by either local_scan code or libraries
+ accessed via the ${dlfunc } expansion item. Both will need coding
+ adjustment for any calls to smtp_printf() to match the new function
+ signature; a FALSE value for the new argument is always safe.
+
+JH/46 FreeBSD: fix use of the sendfile() syscall. The shim was not updating
+ the file-offset (which the Linux syscall does, and exim expects); this
+ resulted in an indefinite loop.
+
+JH/47 ARC: fix crash in signing, triggered when a configuration error failed
+ to do ARC verification. The Authentication-Results: header line added
+ by the configuration then had no ARC item.
+
+
+Exim version 4.92
+-----------------
+
+JH/01 Remove code calling the customisable local_scan function, unless a new
+ definition "HAVE_LOCAL_SCAN=yes" is present in the Local/Makefile.
+
+JH/02 Bug 1007: Avoid doing logging from signal-handlers, as that can result in
+ non-signal-safe functions being used.
+
+JH/03 Bug 2269: When presented with a received message having a stupidly large
+ number of DKIM-Signature headers, disable DKIM verification to avoid
+ a resource-consumption attack. The limit is set at twenty.
+
+JH/04 Add variables $arc_domains, $arc_oldest_pass for ARC verify. Fix the
+ report of oldest_pass in ${authres } in consequence, and separate out
+ some descriptions of reasons for verification fail.
+
+JH/05 Bug 2273: Cutthrough delivery left a window where the received messsage
+ files in the spool were present and unlocked. A queue-runner could spot
+ them, resulting in a duplicate delivery. Fix that by doing the unlock
+ after the unlink. Investigation by Tim Stewart. Take the opportunity to
+ add more error-checking on spoolfile handling while that code is being
+ messed with.
+
+PP/01 Refuse to open a spool data file (*-D) if it's a symlink.
+ No known attacks, no CVE, this is defensive hardening.
+
+JH/06 Bug 2275: The MIME ACL unlocked the received message files early, and
+ a queue-runner could start a delivery while other operations were ongoing.
+ Cutthrough delivery was a common victim, resulting in duplicate delivery.
+ Found and investigated by Tim Stewart. Fix by using the open message data
+ file handle rather than opening another, and not locally closing it (which
+ releases a lock) for that case, while creating the temporary .eml format
+ file for the MIME ACL. Also applies to "regex" and "spam" ACL conditions.
+
+JH/07 Bug 177: Make a random-recipient callout success visible in ACL, by setting
+ $sender_verify_failure/$recipient_verify_failure to "random".
+
+JH/08 When generating a selfsigned cert, use serial number 1 since zero is not
+ legitimate.
+
+JH/09 Bug 2274: Fix logging of cmdline args when starting in an unlinked cwd.
+ Previously this would segfault.
+
+JH/10 Fix ARC signing for case when DKIM signing failed. Previously this would
+ segfault.
+
+JH/11 Bug 2264: Exim now only follows CNAME chains one step by default. We'd
+ like zero, since the resolver should be doing this for us, But we need one
+ as a CNAME but no MX presence gets the CNAME returned; we need to check
+ that doesn't point to an MX to declare it "no MX returned" rather than
+ "error, loop". A new main option is added so the older capability of
+ following some limited number of chain links is maintained.
+
+JH/12 Add client-ip info to non-pass iprev ${authres } lines.
+
+JH/13 For receent Openssl versions (1.1 onward) use modern generic protocol
+ methods. These should support TLS 1.3; they arrived with TLS 1.3 and the
+ now-deprecated earlier definitions used only specified the range up to TLS
+ 1.2 (in the older-version library docs).
+
+JH/14 Bug 2284: Fix DKIM signing for body lines starting with a pair of dots.
+
+JH/15 Rework TLS client-side context management. Stop using a global, and
+ explicitly pass a context around. This enables future use of TLS for
+ connections to service-daemons (eg. malware scanning) while a client smtp
+ connection is using TLS; with cutthrough connections this is quite likely.
+
+JH/16 Fix ARC verification to do AS checks in reverse order.
+
+JH/17 Support a "tls" option on the ${readsocket } expansion item.
+
+JH/18 Bug 2287: Fix the protocol name (eg utf8esmtp) for multiple messages
+ using the SMTPUTF8 option on their MAIL FROM commands, in one connection.
+ Previously the "utf8" would be re-prepended for every additional message.
+
+JH/19 Reject MAIL FROM commands with SMTPUTF8 when the facility was not advertised.
+ Previously thery were accepted, resulting in issues when attempting to
+ forward messages to a non-supporting MTA.
+
+PP/02 Let -n work with printing macros too, not just options.
+
+JH/20 Bug 2296: Fix cutthrough for >1 address redirection. Previously only
+ one parent address was copied, and bogus data was used at delivery-logging
+ time. Either a crash (after delivery) or bogus log data could result.
+ Discovery and analysis by Tim Stewart.
+
+PP/03 Make ${utf8clean:} expansion operator detect incomplete final character.
+ Previously if the string ended mid-character, we did not insert the
+ promised '?' replacement.
+
+PP/04 Documentation: current string operators work on bytes, not codepoints.
+
+JH/21 Change as many as possible of the global flags into one-bit bitfields; these
+ should pack well giving a smaller memory footprint so better caching and
+ therefore performance. Group the declarations where this can't be done so
+ that the byte-sized flag variables are not interspersed among pointer
+ variables, giving a better chance of good packing by the compiler.
+
+JH/22 Bug 1896: Fix the envelope from for DMARC forensic reports to be possibly
+ non-null, to avoid issues with sites running BATV. Previously reports were
+ sent with an empty envelope sender so looked like bounces.
+
+JH/23 Bug 2318: Fix the noerror command within filters. It wasn't working.
+ The ignore_error flag wasn't being returned from the filter subprocess so
+ was not set for later routers. Investigation and fix by Matthias Kurz.
+
+JH/24 Bug 2310: Raise a msg:fail:internal event for each undelivered recipient,
+ and a msg:complete for the whole, when a message is manually removed using
+ -Mrm. Developement by Matthias Kurz, hacked on by JH.
+
+JH/25 Avoid fixed-size buffers for pathnames in DB access. This required using
+ a "Gnu special" function, asprintf() in the DB utility binary builds; I
+ hope that is portable enough.
+
+JH/26 Bug 2311: Fix DANE-TA verification under GnuTLS. Previously it was also
+ requiring a known-CA anchor certificate; make it now rely entirely on the
+ TLSA as an anchor. Checking the name on the leaf cert against the name
+ on the A-record for the host is still done for TA (but not for EE mode).
+
+JH/27 Fix logging of proxy address. Previously, a pointless "PRX=[]:0" would be
+ included in delivery lines for non-proxied connections, when compiled with
+ SUPPORT_SOCKS and running with proxy logging enabled.
+
+JH/28 Bug 2314: Fire msg:fail:delivery event even when error is being ignored.
+ Developement by Matthias Kurz, tweaked by JH. While in that bit of code,
+ move the existing event to fire before the normal logging of message
+ failure so that custom logging is bracketed by normal logging.
+
+JH/29 Bug 2322: A "fail" command in a non-system filter (file) now fires the
+ msg:fail:internal event. Developement by Matthias Kurz.
+
+JH/30 Bug 2329: Increase buffer size used for dns lookup from 2k, which was
+ far too small for todays use of crypto signatures stored there. Go all
+ the way to the max DNS message size of 64kB, even though this might be
+ overmuch for IOT constrained device use.
+
+JH/31 Fix a bad use of a copy function, which could be used to pointlessly
+ copy a string over itself. The library routine is documented as not
+ supporting overlapping copies, and on MacOS it actually raised a SIGABRT.
+
+JH/32 For main options check_spool_space and check_inode_space, where the
+ platform supports 64b integers, support more than the previous 2^31 kB
+ (i.e. more than 2 TB). Accept E, P and T multipliers in addition to
+ the previous G, M, k.
+
+JH/33 Bug 2338: Fix the cyrus-sasl authenticator to fill in the
+ $authenticated_fail_id variable on authentication failure. Previously
+ it was unset.
+
+JH/34 Increase RSA keysize of autogen selfsign cert from 1024 to 2048. RHEL 8.0
+ OpenSSL didn't want to use such a weak key. Do for GnuTLS also, and for
+ more-modern GnuTLS move from GNUTLS_SEC_PARAM_LOW to
+ GNUTLS_SEC_PARAM_MEDIUM.
+
+JH/35 OpenSSL: fail the handshake when SNI processing hits a problem, server
+ side. Previously we would continue as if no SNI had been received.
+
+JH/36 Harden the handling of string-lists. When a list consisted of a sole
+ "<" character, which should be a list-separator specification, we walked
+ off past the nul-terimation.
+
+JH/37 Bug 2341: Send "message delayed" warning MDNs (restricted to external
+ causes) even when the retry time is not yet met. Previously they were
+ not, meaning that when (say) an account was over-quota and temp-rejecting,
+ and multiple senders' messages were queued, only one sender would get
+ notified on each configured delay_warning cycle.
+
+JH/38 Bug 2351: Log failures to extract envelope addresses from message headers.
+
+JH/39 OpenSSL: clear the error stack after an SSL_accept(). With anon-auth
+ cipher-suites, an error can be left on the stack even for a succeeding
+ accept; this results in impossible error messages when a later operation
+ actually does fail.
+
+AM/01 Bug 2359: GnuTLS: repeat lowlevel read and write operations while they
+ return error codes indicating retry. Under TLS1.3 this becomes required.
+
+JH/40 Fix the feature-cache refresh for EXPERIMENTAL_PIPE_CONNECT. Previously
+ it only wrote the new authenticators, resulting in a lack of tracking of
+ peer changes of ESMTP extensions until the next cache flush.
+
+JH/41 Fix the loop reading a message header line to check for integer overflow,
+ and more-often against header_maxsize. Previously a crafted message could
+ induce a crash of the recive process; now the message is cleanly rejected.
+
+JH/42 Bug 2366: Fix the behaviour of the dkim_verify_signers option. It had
+ been totally disabled for all of 4.91. Discovery and fix by "Mad Alex".
+
+
+Exim version 4.91
+-----------------
+
+GF/01 DEFER rather than ERROR on redis cluster MOVED response.
+ When redis_servers is set to a list of > 1 element, and the Redis servers
+ in that list are in cluster configuration, convert the REDIS_REPLY_ERROR
+ case of MOVED into a DEFER case instead, thus moving the query onto the
+ next server in the list. For a cluster of N elements, all N servers must
+ be defined in redis_servers.
+
+GF/02 Catch and remove uninitialized value warning in exiqsumm
+ Check for existence of @ARGV before looking at $ARGV[0]
+
+JH/01 Replace the store_release() internal interface with store_newblock(),
+ which internalises the check required to safely use the old one, plus
+ the allocate and data copy operations duplicated in both (!) of the
+ extant use locations.
+
+JH/02 Disallow '/' characters in queue names specified for the "queue=" ACL
+ modifier. This matches the restriction on the commandline.
+
+JH/03 Fix pgsql lookup for multiple result-tuples with a single column.
+ Previously only the last row was returned.
+
+JH/04 Bug 2217: Tighten up the parsing of DKIM signature headers. Previously
+ we assumed that tags in the header were well-formed, and parsed the
+ element content after inspecting only the first char of the tag.
+ Assumptions at that stage could crash the receive process on malformed
+ input.
+
+JH/05 Bug 2215: Fix crash associated with dnsdb lookup done from DKIM ACL.
+ While running the DKIM ACL we operate on the Permanent memory pool so that
+ variables created with "set" persist to the DATA ACL. Also (at any time)
+ DNS lookups that fail create cache records using the Permanent pool. But
+ expansions release any allocations made on the current pool - so a dnsdb
+ lookup expansion done in the DKIM ACL releases the memory used for the
+ DNS negative-cache, and bad things result. Solution is to switch to the
+ Main pool for expansions.
+ While we're in that code, add checks on the DNS cache during store_reset,
+ active in the testsuite.
+ Problem spotted, and debugging aided, by Wolfgang Breyha.
+
+JH/06 Fix issue with continued-connections when the DNS shifts unreliably.
+ When none of the hosts presented to a transport match an already-open
+ connection, close it and proceed with the list. Previously we would
+ queue the message. Spotted by Lena with Yahoo, probably involving
+ round-robin DNS.
+
+JH/07 Bug 2214: Fix SMTP responses resulting from non-accept result of MIME ACL.
+ Previously a spurious "250 OK id=" response was appended to the proper
+ failure response.
+
+JH/08 The "support for" informational output now, which built with Content
+ Scanning support, has a line for the malware scanner interfaces compiled
+ in. Interface can be individually included or not at build time.
+
+JH/09 The "aveserver", "kavdaemon" and "mksd" interfaces are now not included
+ by the template makefile "src/EDITME". The "STREAM" support for an older
+ ClamAV interface method is removed.
+
+JH/10 Bug 2223: Fix mysql lookup returns for the no-data case (when the number of
+ rows affected is given instead).
+
+JH/11 The runtime Berkeley DB library version is now additionally output by
+ "exim -d -bV". Previously only the compile-time version was shown.
+
+JH/12 Bug 2230: Fix cutthrough routing for nonfirst messages in an initiating
+ SMTP connection. Previously, when one had more receipients than the
+ first, an abortive onward connection was made. Move to full support for
+ multiple onward connections in sequence, handling cutthrough connection
+ for all multi-message initiating connections.
+
+JH/13 Bug 2229: Fix cutthrough routing for nonstandard port numbers defined by
+ routers. Previously, a multi-recipient message would fail to match the
+ onward-connection opened for the first recipient, and cause its closure.
+
+JH/14 Bug 2174: A timeout on connect for a callout was also erroneously seen as
+ a timeout on read on a GnuTLS initiating connection, resulting in the
+ initiating connection being dropped. This mattered most when the callout
+ was marked defer_ok. Fix to keep the two timeout-detection methods
+ separate.
+
+JH/15 Relax results from ACL control request to enable cutthrough, in
+ unsupported situations, from error to silently (except under debug)
+ ignoring. This covers use with PRDR, frozen messages, queue-only and
+ fake-reject.
+
+HS/01 Fix Buffer overflow in base64d() (CVE-2018-6789)
+
+JH/16 Fix bug in DKIM verify: a buffer overflow could corrupt the malloc
+ metadata, resulting in a crash in free().
+
+PP/01 Fix broken Heimdal GSSAPI authenticator integration.
+ Broken in f2ed27cf5, missing an equals sign for specified-initialisers.
+ Broken also in d185889f4, with init system revamp.
+
+JH/17 Bug 2113: Fix conversation closedown with the Avast malware scanner.
+ Previously we abruptly closed the connection after reading a malware-
+ found indication; now we go on to read the "scan ok" response line,
+ and send a quit.
+
+JH/18 Bug 2239: Enforce non-usability of control=utf8_downconvert in the mail
+ ACL. Previously, a crash would result.
+
+JH/19 Speed up macro lookups during configuration file read, by skipping non-
+ macro text after a replacement (previously it was only once per line) and
+ by skipping builtin macros when searching for an uppercase lead character.
+
+JH/20 DANE support moved from Experimental to mainline. The Makefile control
+ for the build is renamed.
+
+JH/21 Fix memory leak during multi-message connections using STARTTLS. A buffer
+ was allocated for every new TLS startup, meaning one per message. Fix
+ by only allocating once (OpenSSL) or freeing on TLS-close (GnuTLS).
+
+JH/22 Bug 2236: When a DKIM verification result is overridden by ACL, DMARC
+ reported the original. Fix to report (as far as possible) the ACL
+ result replacing the original.
+
+JH/23 Fix memory leak during multi-message connections using STARTTLS under
+ OpenSSL. Certificate information is loaded for every new TLS startup,
+ and the resources needed to be freed.
+
+JH/24 Bug 2242: Fix exim_dbmbuild to permit directoryless filenames.
+
+JH/25 Fix utf8_downconvert propagation through a redirect router. Previously it
+ was not propagated.
+
+JH/26 Bug 2253: For logging delivery lines under PRDR, append the overall
+ DATA response info to the (existing) per-recipient response info for
+ the "C=" log element. It can have useful tracking info from the
+ destination system. Patch from Simon Arlott.
+
+JH/27 Bug 2251: Fix ldap lookups that return a single attribute having zero-
+ length value. Previously this would segfault.
+
+HS/02 Support Avast multiline protoocol, this allows passing flags to
+ newer versions of the scanner.
+
+JH/28 Ensure that variables possibly set during message acceptance are marked
+ dead before release of memory in the daemon loop. This stops complaints
+ about them when the debug_store option is enabled. Discovered specifically
+ for sender_rate_period, but applies to a whole set of variables.
+ Do the same for the queue-runner and queue-list loops, for variables set
+ from spool message files. Do the same for the SMTP per-message loop, for
+ certain variables indirectly set in ACL operations.
+
+JH/29 Bug 2250: Fix a longstanding bug in heavily-pipelined SMTP input (such
+ as a multi-recipient message from a mailinglist manager). The coding had
+ an arbitrary cutoff number of characters while checking for more input;
+ enforced by writing a NUL into the buffer. This corrupted long / fast
+ input. The problem was exposed more widely when more pipelineing of SMTP
+ responses was introduced, and one Exim system was feeding another.
+ The symptom is log complaints of SMTP syntax error (NUL chars) on the
+ receiving system, and refused recipients seen by the sending system
+ (propating to people being dropped from mailing lists).
+ Discovered and pinpointed by David Carter.
+
+JH/30 The (EXPERIMENTAL_DMARC) variable $dmarc_ar_header is withdrawn, being
+ replaced by the ${authresults } expansion.
+
+JH/31 Bug 2257: Fix pipe transport to not use a socket-only syscall.
+
+HS/03 Set a handler for SIGTERM and call exit(3) if running as PID 1. This
+ allows proper process termination in container environments.
+
+JH/32 Bug 2258: Fix spool_wireformat in combination with LMTP transport.
+ Previously the "final dot" had a newline after it; ensure it is CR,LF.
+
+JH/33 SPF: remove support for the "spf" ACL condition outcome values "err_temp"
+ and "err_perm", deprecated since 4.83 when the RFC-defined words
+ "temperror" and "permerror" were introduced.
+
+JH/34 Re-introduce enforcement of no cutthrough delivery on transports having
+ transport-filters or DKIM-signing. The restriction was lost in the
+ consolidation of verify-callout and delivery SMTP handling.
+ Extend the restriction to also cover ARC-signing.
+
+JH/35 Cutthrough: for a final-dot response timeout (and nonunderstood responses)
+ in defer=pass mode supply a 450 to the initiator. Previously the message
+ would be spooled.
+
+PP/02 DANE: add dane_require_tls_ciphers SMTP Transport option; if unset,
+ tls_require_ciphers is used as before.
+
+HS/03 Malware Avast: Better match the Avast multiline protocol. Add
+ "pass_unscanned". Only tmpfails from the scanner are written to
+ the paniclog, as they may require admin intervention (permission
+ denied, license issues). Other scanner errors (like decompression
+ bombs) do not cause a paniclog entry.
+
+JH/36 Fix reinitialisation of DKIM logging variable between messages.
+ Previously it was possible to log spurious information in receive log
+ lines.
+
+JH/37 Bug 2255: Revert the disable of the OpenSSL session caching. This
+ triggered odd behaviour from Outlook Express clients.
+
+PP/03 Add util/renew-opendmarc-tlds.sh script for safe renewal of public
+ suffix list.
+
+JH/38 DKIM: accept Ed25519 pubkeys in SubjectPublicKeyInfo-wrapped form,
+ since the IETF WG has not yet settled on that versus the original
+ "bare" representation.
+
+JH/39 Fix syslog logging for syslog_timestamp=no and log_selector +millisec.
+ Previously the millisecond value corrupted the output.
+ Fix also for syslog_pid=no and log_selector +pid, for which the pid
+ corrupted the output.
+
+
+Exim version 4.90
+-----------------
+
+JH/01 Rework error string handling in TLS interface so that the caller in
+ more cases is responsible for logging. This permits library-sourced
+ string to be attached to addresses during delivery, and collapses
+ pairs of long lines into single ones.
+
+PP/01 Allow PKG_CONFIG_PATH to be set in Local/Makefile and use it correctly
+ during configuration. Wildcards are allowed and expanded.
+
+JH/02 Rework error string handling in DKIM to pass more info back to callers.
+ This permits better logging.
+
+JH/03 Rework the transport continued-connection mechanism: when TLS is active,
+ do not close it down and have the child transport start it up again on
+ the passed-on TCP connection. Instead, proxy the child (and any
+ subsequent ones) for TLS via a unix-domain socket channel. Logging is
+ affected: the continued delivery log lines do not have any DNSSEC, TLS
+ Certificate or OCSP information. TLS cipher information is still logged.
+
+JH/04 Shorten the log line for daemon startup by collapsing adjacent sets of
+ identical IP addresses on different listening ports. Will also affect
+ "exiwhat" output.
+
+PP/02 Bug 2070: uClibc defines __GLIBC__ without providing glibc headers;
+ add noisy ifdef guards to special-case this sillyness.
+ Patch from Bernd Kuhls.
+
+JH/05 Tighten up the checking in isip4 (et al): dotted-quad components larger
+ than 255 are no longer allowed.
+
+JH/06 Default openssl_options to include +no_ticket, to reduce load on peers.
+ Disable the session-cache too, which might reduce our load. Since we
+ currrectly use a new context for every connection, both as server and
+ client, there is no benefit for these.
+ GnuTLS appears to not support tickets server-side by default (we don't
+ call gnutls_session_ticket_enable_server()) but client side is enabled
+ by default on recent versions (3.1.3 +) unless the PFS priority string
+ is used (3.2.4 +).
+
+PP/03 Add $SOURCE_DATE_EPOCH support for reproducible builds, per spec at
+ <https://reproducible-builds.org/specs/source-date-epoch/>.
+
+JH/07 Fix smtp transport use of limited max_rcpt under mua_wrapper. Previously
+ the check for any unsuccessful recipients did not notice the limit, and
+ erroneously found still-pending ones.
+
+JH/08 Pipeline CHUNKING command and data together, on kernels that support
+ MSG_MORE. Only in-clear (not on TLS connections).
+
+JH/09 Avoid using a temporary file during transport using dkim. Unless a
+ transport-filter is involved we can buffer the headers in memory for
+ creating the signature, and read the spool data file once for the
+ signature and again for transmission.
+
+JH/10 Enable use of sendfile in Linux builds as default. It was disabled in
+ 4.77 as the kernel support then wasn't solid, having issues in 64bit
+ mode. Now, it's been long enough. Add support for FreeBSD also.
+
+JH/11 Bug 2104: Fix continued use of a transport connection with TLS. In the
+ case where the routing stage had gathered several addresses to send to
+ a host before calling the transport for the first, we previously failed
+ to close down TLS in the old transport process before passing the TCP
+ connection to the new process. The new one sent a STARTTLS command
+ which naturally failed, giving a failed delivery and bloating the retry
+ database. Investigation and fix prototype from Wolfgang Breyha.
+
+JH/12 Fix check on SMTP command input synchronisation. Previously there were
+ false-negatives in the check that the sender had not preempted a response
+ or prompt from Exim (running as a server), due to that code's lack of
+ awareness of the SMTP input buffering.
+
+PP/04 Add commandline_checks_require_admin option.
+ Exim drops privileges sanely, various checks such as -be aren't a
+ security problem, as long as you trust local users with access to their
+ own account. When invoked by services which pass untrusted data to
+ Exim, this might be an issue. Set this option in main configuration
+ AND make fixes to the calling application, such as using `--` to stop
+ processing options.
+
+JH/13 Do pipelining under TLS. Previously, although safe, no advantage was
+ taken. Now take care to pack both (client) MAIL,RCPT,DATA, and (server)
+ responses to those, into a single TLS record each way (this usually means
+ a single packet). As a side issue, smtp_enforce_sync now works on TLS
+ connections.
+
+PP/05 OpenSSL/1.1: use DH_bits() for more accurate DH param sizes. This
+ affects you only if you're dancing at the edge of the param size limits.
+ If you are, and this message makes sense to you, then: raise the
+ configured limit or use OpenSSL 1.1. Nothing we can do for older
+ versions.
+
+JH/14 For the "sock" variant of the malware scanner interface, accept an empty
+ cmdline element to get the documented default one. Previously it was
+ inaccessible.
+
+JH/15 Fix a crash in the smtp transport caused when two hosts in succession
+ are unsuable for non-message-specific reasons - eg. connection timeout,
+ banner-time rejection.
+
+JH/16 Fix logging of delivery remote port, when specified by router, under
+ callout/hold.
+
+PP/06 Repair manualroute's ability to take options in any order, even if one
+ is the name of a transport.
+ Fixes bug 2140.
+
+HS/01 Cleanup, prevent repeated use of -p/-oMr (CVE-2017-1000369)
+
+JH/17 Change the list-building routines interface to use the expanding-string
+ triplet model, for better allocation and copying behaviour.
+
+JH/18 Prebuild the data-structure for "builtin" macros, for faster startup.
+ Previously it was constructed the first time a possibly-matching string
+ was met in the configuration file input during startup; now it is done
+ during compilation.
+
+JH/19 Bug 2141: Use the full-complex API for Berkeley DB rather than the legacy-
+ compatible one, to avoid the (poorly documented) possibility of a config
+ file in the working directory redirecting the DB files, possibly correpting
+ some existing file. CVE-2017-10140 assigned for BDB.
+
+JH/20 Bug 2147: Do not defer for a verify-with-callout-and-random which is not
+ cache-hot. Previously, although the result was properly cached, the
+ initial verify call returned a defer.
+
+JH/21 Bug 2151: Avoid using SIZE on the MAIL for a callout verify, on any but
+ the main verify for receipient in uncached-mode.
+
+JH/22 Retire historical build files to an "unsupported" subdir. These are
+ defined as "ones for which we have no current evidence of testing".
+
+JH/23 DKIM: enforce the DNS pubkey record "h" permitted-hashes optional field,
+ if present. Previously it was ignored.
+
+JH/24 Start using specified-initialisers in C structure init coding. This is
+ a C99 feature (it's 2017, so now considered safe).
+
+JH/25 Use one-bit bitfields for flags in the "addr" data structure. Previously
+ if was a fixed-sized field and bitmask ops via macros; it is now more
+ extensible.
+
+PP/07 GitHub PR 56: Apply MariaDB build fix.
+ Patch provided by Jaroslav Škarvada.
+
+PP/08 Bug 2161: Fix regression in sieve quoted-printable handling introduced
+ during Coverity cleanups [4.87 JH/47]
+ Diagnosis and fix provided by Michael Fischer v. Mollard.
+
+JH/26 Fix DKIM bug: when the pseudoheader generated for signing was exactly
+ the right size to place the terminating semicolon on its own folded
+ line, the header hash was calculated to an incorrect value thanks to
+ the (relaxed) space the fold became.
+
+HS/02 Fix Bug 2130: large writes from the transport subprocess were chunked
+ and confused the parent.
+
+JH/27 Fix SOCKS bug: an unitialized pointer was deref'd by the transport process
+ which could crash as a result. This could lead to undeliverable messages.
+
+JH/28 Logging: "next input sent too soon" now shows where input was truncated
+ for log purposes.
+
+JH/29 Fix queue_run_in_order to ignore the PID portion of the message ID. This
+ matters on fast-turnover and PID-randomising systems, which were getting
+ out-of-order delivery.
+
+JH/30 Fix a logging bug on aarch64: an unsafe routine was previously used for
+ a possibly-overlapping copy. The symptom was that "Remote host closed
+ connection in response to HELO" was logged instead of the actual 4xx
+ error for the HELO.
+
+JH/31 Fix CHUNKING code to properly flush the unwanted chunk after an error.
+ Previously only that bufferd was discarded, resulting in SYMTP command
+ desynchronisation.
+
+JH/32 DKIM: when a message has multiple signatures matching an identity given
+ in dkim_verify_signers, run the dkim acl once for each. Previously only
+ one run was done. Bug 2189.
+
+JH/33 Downgrade an unfound-list name (usually a typo in the config file) from
+ "panic the current process" to "deliberately defer". The panic log is
+ still written with the problem list name; the mail and reject logs now
+ get a temp-reject line for the message that was being handled, saying
+ something like "domains check lookup or other defer". The SMTP 451
+ message is still "Temporary local problem".
+
+JH/34 Bug 2199: Fix a use-after-free while reading smtp input for header lines.
+ A crafted sequence of BDAT commands could result in in-use memory beeing
+ freed. CVE-2017-16943.
+
+HS/03 Bug 2201: Fix checking for leading-dot on a line during headers reading
+ from SMTP input. Previously it was always done; now only done for DATA
+ and not BDAT commands. CVE-2017-16944.
+
+JH/35 Bug 2201: Flush received data in BDAT mode after detecting an error fatal
+ to the message (such as an overlong header line). Previously this was
+ not done and we did not exit BDAT mode. Followon from the previous item
+ though a different problem.
+
+
+Exim version 4.89
+-----------------
+
+JH/01 Bug 1922: Support IDNA2008. This has slightly different conversion rules
+ than -2003 did; needs libidn2 in addition to libidn.
+
+JH/02 The path option on a pipe transport is now expanded before use.
+
+PP/01 GitHub PR 50: Do not call ldap_start_tls_s on ldapi:// connections.
+ Patch provided by "Björn", documentation fix added too.
+
+JH/03 Bug 2003: fix Proxy Protocol v2 handling: the address size field was
+ missing a wire-to-host endian conversion.
+
+JH/04 Bug 2004: fix CHUNKING in non-PIPELINEING mode. Chunk data following
+ close after a BDAT command line could be taken as a following command,
+ giving a synch failure. Fix by only checking for synch immediately
+ before acknowledging the chunk.
+
+PP/02 GitHub PR 52: many spelling fixes, which include fixing parsing of
+ no_require_dnssec option and creation of _HAVE_TRANSPORT_APPEND_MAILDIR
+ macro. Patches provided by Josh Soref.
+
+JH/05 Have the EHLO response advertise VRFY, if there is a vrfy ACL defined.
+ Previously we did not; the RFC seems ambiguous and VRFY is not listed
+ by IANA as a service extension. However, John Klensin suggests that we
+ should.
+
+JH/06 Bug 2017: Fix DKIM verification in -bh test mode. The data feed into
+ the dkim code may be unix-mode line endings rather than smtp wire-format
+ CRLF, so prepend a CR to any bare LF.
+
+JH/07 Rationalise the coding for callout smtp conversations and transport ones.
+ As a side-benfit, callouts can now use PIPELINING hence fewer round-trips.
+
+JH/08 Bug 2016: Fix DKIM verification vs. CHUNKING. Any BDAT commands after
+ the first were themselves being wrongly included in the feed into dkim
+ processing; with most chunk sizes in use this resulted in an incorrect
+ body hash calculated value.
+
+JH/09 Bug 2014: permit inclusion of a DKIM-Signature header in a received
+ DKIM signature block, for verification. Although advised against by
+ standards it is specifically not ruled illegal.
+
+JH/10 Bug 2025: Fix reception of (quoted) local-parts with embedded spaces.
+
+JH/11 Bug 2029: Fix crash in DKIM verification when a message signature block is
+ missing a body hash (the bh= tag).
+
+JH/12 Bug 2018: Re-order Proxy Protocol startup versus TLS-on-connect startup.
+ It seems that HAProxy sends the Proxy Protocol information in clear and
+ only then does a TLS startup, so do the same.
+
+JH/13 Bug 2027: Avoid attempting to use TCP Fast Open for non-transport client
+ TCP connections (such as for Spamd) unless the daemon successfully set
+ Fast Open mode on its listening sockets. This fixes breakage seen on
+ too-old kernels or those not configured for Fast Open, at the cost of
+ requiring both directions being enabled for TFO, and TFO never being used
+ by non-daemon-related Exim processes.
+
+JH/14 Bug 2000: Reject messages recieved with CHUNKING but with malformed line
+ endings, at least on the first header line. Try to canonify any that get
+ past that check, despite the cost.
+
+JH/15 Angle-bracket nesting (an error inserted by broken sendmails) levels are
+ now limited to an arbitrary five deep, while parsing addresses with the
+ strip_excess_angle_brackets option enabled.
+
+PP/03 Bug 2018: For Proxy Protocol and TLS-on-connect, do not over-read and
+ instead leave the unprompted TLS handshake in socket buffer for the
+ TLS library to consume.
+
+PP/04 Bug 2018: Also handle Proxy Protocol v2 safely.
+
+PP/05 FreeBSD compat: handle that Ports no longer create /usr/bin/perl
+
+JH/16 Drop variables when they go out of scope. Memory management drops a whole
+ region in one operation, for speed, and this leaves assigned pointers
+ dangling. Add checks run only under the testsuite which checks all
+ variables at a store-reset and panics on a dangling pointer; add code
+ explicitly nulling out all the variables discovered. Fixes one known
+ bug: a transport crash, where a dangling pointer for $sending_ip_address
+ originally assigned in a verify callout, is re-used.
+
+PP/06 Drop '.' from @INC in various Perl scripts.
+
+PP/07 Switch FreeBSD iconv to always use the base-system libc functions.
+
+PP/08 Reduce a number of compilation warnings under clang; building with
+ CC=clang CFLAGS+=-Wno-dangling-else -Wno-logical-op-parentheses
+ should be warning-free.
+
+JH/17 Fix inbound CHUNKING when DKIM disabled at runtime.
+
+HS/01 Fix portability problems introduced by PP/08 for platforms where
+ realloc(NULL) is not equivalent to malloc() [SunOS et al].
+
+HS/02 Bug 1974: Fix missing line terminator on the last received BDAT
+ chunk. This allows us to accept broken chunked messages. We need a more
+ general solution here.
+
+PP/09 Wrote util/chunking_fixqueue_finalnewlines.pl to help recover
+ already-broken messages in the queue.
+
+JH/18 Bug 2061: Fix ${extract } corrupting an enclosing ${reduce } $value.
+
+JH/19 Fix reference counting bug in routing-generated-address tracking.
+
+
+Exim version 4.88
+-----------------
+
+JH/01 Use SIZE on MAIL FROM in a cutthrough connection, if the destination
+ supports it and a size is available (ie. the sending peer gave us one).
+
+JH/02 The obsolete acl condition "demime" is removed (finally, after ten
+ years of being deprecated). The replacements are the ACLs
+ acl_smtp_mime and acl_not_smtp_mime.
+
+JH/03 Upgrade security requirements imposed for hosts_try_dane: previously
+ a downgraded non-dane trust-anchor for the TLS connection (CA-style)
+ or even an in-clear connection were permitted. Now, if the host lookup
+ was dnssec and dane was requested then the host is only used if the
+ TLSA lookup succeeds and is dnssec. Further hosts (eg. lower priority
+ MXs) will be tried (for hosts_try_dane though not for hosts_require_dane)
+ if one fails this test.
+ This means that a poorly-configured remote DNS will make it incommunicado;
+ but it protects against a DNS-interception attack on it.
+
+JH/04 Bug 1810: make continued-use of an open smtp transport connection
+ non-noisy when a race steals the message being considered.
+
+JH/05 If main configuration option tls_certificate is unset, generate a
+ self-signed certificate for inbound TLS connections.
+
+JH/06 Bug 165: hide more cases of password exposure - this time in expansions
+ in rewrites and routers.
+
+JH/07 Retire gnutls_require_mac et.al. These were nonfunctional since 4.80
+ and logged a warning sing 4.83; now they are a configuration file error.
+
+JH/08 Bug 1836: Fix crash in VRFY handling when handed an unqualified name
+ (lacking @domain). Apply the same qualification processing as RCPT.
+
+JH/09 Bug 1804: Avoid writing msglog files when in -bh or -bhc mode.
+
+JH/10 Support ${sha256:} applied to a string (as well as the previous
+ certificate).
+
+JH/11 Cutthrough: avoid using the callout hints db on a verify callout when
+ a cutthrough deliver is pending, as we always want to make a connection.
+ This also avoids re-routing the message when later placing the cutthrough
+ connection after a verify cache hit.
+ Do not update it with the verify result either.
+
+JH/12 Cutthrough: disable when verify option success_on_redirect is used, and
+ when routing results in more than one destination address.
+
+JH/13 Cutthrough: expand transport dkim_domain option when testing for dkim
+ signing (which inhibits the cutthrough capability). Previously only
+ the presence of an option was tested; now an expansion evaluating as
+ empty is permissible (obviously it should depend only on data available
+ when the cutthrough connection is made).
+
+JH/14 Fix logging of errors under PIPELINING. Previously the log line giving
+ the relevant preceding SMTP command did not note the pipelining mode.
+
+JH/15 Fix counting of empty lines in $body_linecount and $message_linecount.
+ Previously they were not counted.
+
+JH/16 DANE: treat a TLSA lookup response having all non-TLSA RRs, the same
+ as one having no matching records. Previously we deferred the message
+ that needed the lookup.
+
+JH/17 Fakereject: previously logged as a normal message arrival "<="; now
+ distinguished as "(=".
+
+JH/18 Bug 1867: make the fail_defer_domains option on a dnslookup router work
+ for missing MX records. Previously it only worked for missing A records.
+
+JH/19 Bug 1850: support Radius libraries that return REJECT_RC.
+
+JH/20 Bug 1872: Ensure that acl_smtp_notquit is run when the connection drops
+ after the data-go-ahead and data-ack. Patch from Jason Betts.
+
+JH/21 Bug 1846: Send DMARC forensic reports for reject and quarantine results,
+ even for a "none" policy. Patch from Tony Meyer.
+
+JH/22 Fix continued use of a connection for further deliveries. If a port was
+ specified by a router, it must also match for the delivery to be
+ compatible.
+
+JH/23 Bug 1874: fix continued use of a connection for further deliveries.
+ When one of the recipients of a message was unsuitable for the connection
+ (has no matching addresses), we lost track of needing to mark it
+ deferred. As a result mail would be lost.
+
+JH/24 Bug 1832: Log EHLO response on getting conn-close response for HELO.
+
+JH/25 Decoding ACL controls is now done using a binary search; the source code
+ takes up less space and should be simpler to maintain. Merge the ACL
+ condition decode tables also, with similar effect.
+
+JH/26 Fix problem with one_time used on a redirect router which returned the
+ parent address unchanged. A retry would see the parent address marked as
+ delivered, so not attempt the (identical) child. As a result mail would
+ be lost.
+
+JH/27 Fix a possible security hole, wherein a process operating with the Exim
+ UID can gain a root shell. Credit to http://www.halfdog.net/ for
+ discovery and writeup. Ubuntu bug 1580454; no bug raised against Exim
+ itself :(
+
+JH/28 Enable {spool,log} filesystem space and inode checks as default.
+ Main config options check_{log,spool}_{inodes,space} are now
+ 100 inodes, 10MB unless set otherwise in the configuration.
+
+JH/29 Fix the connection_reject log selector to apply to the connect ACL.
+ Previously it only applied to the main-section connection policy
+ options.
+
+JH/30 Bug 1897: fix callouts connection fallback from TLS to cleartext.
+
+PP/01 Changed default Diffie-Hellman parameters to be Exim-specific, created
+ by me. Added RFC7919 DH primes as an alternative.
+
+PP/02 Unbreak build via pkg-config with new hash support when crypto headers
+ are not in the system include path.
+
+JH/31 Fix longstanding bug with aborted TLS server connection handling. Under
+ GnuTLS, when a session startup failed (eg because the client disconnected)
+ Exim did stdio operations after fclose. This was exposed by a recent
+ change which nulled out the file handle after the fclose.
+
+JH/32 Bug 1909: Fix OCSP proof verification for cases where the proof is
+ signed directly by the cert-signing cert, rather than an intermediate
+ OCSP-signing cert. This is the model used by LetsEncrypt.
+
+JH/33 Bug 1914: Ensure socket is nonblocking before draining after SMTP QUIT.
+
+HS/01 Fix leak in verify callout under GnuTLS, about 3MB per recipient on
+ an incoming connection.
+
+HS/02 Bug 1802: Do not half-close the connection after sending a request
+ to rspamd.
+
+HS/03 Use "auto" as the default EC curve parameter. For OpenSSL < 1.0.2
+ fallback to "prime256v1".
+
+JH/34 SECURITY: Use proper copy of DATA command in error message.
+ Could leak key material. Remotely exploitable. CVE-2016-9963.
Exim version 4.87
-----------------
+
JH/01 Bug 1664: Disable OCSP for GnuTLS library versions at/before 3.3.16
and 3.4.4 - once the server is enabled to respond to an OCSP request
it does even when not requested, resulting in a stapling non-aware
HS/02 Add the Exim version string to the process info. This way exiwhat
gives some more detail about the running daemon.
-JH/06 Bug 1395: time-limit cacheing of DNS lookups, to the TTL value. This may
+JH/06 Bug 1395: time-limit caching of DNS lookups, to the TTL value. This may
matter for fast-change records such as DNSBLs.
JH/07 Bug 1678: Always record an interface option value, if set, as part of a
JH/18 Bug 1709: When built with TLS support, the tls_advertise_hosts option now
defaults to "*" (all hosts). The variable is now available when not built
- with TLS, default unset, mainly to enable keeping the testuite sane.
+ with TLS, default unset, mainly to enable keeping the testsuite sane.
If a server certificate is not supplied (via tls_certificate) an error is
logged, and clients will find TLS connections fail on startup. Presumably
they will retry in-clear.
in transport context, after the attempt, and per-recipient. The latter type
is per host attempted. The event data is the error message, and the errno
information encodes the lookup type (A vs. MX) used for the (first) host,
- and the trailing two digits of the smtp 4xx reponse.
+ and the trailing two digits of the smtp 4xx response.
GF/01 Bug 1715: Fix for race condition in exicyclog, where exim could attempt
to write to mainlog (or rejectlog, paniclog) in the window between file
"pri" and "weight". Note that the previous implicit priority given by the
list order is no longer honoured.
-JH/22 Bugs 963, 1721: Fix some corner cases in message body canonicalisation
+JH/22 Bugs 963, 1721: Fix some corner cases in message body canonicalization
for DKIM processing.
JH/23 Move SOCKS5 support from Experimental to mainline, enabled for a build
by defining SUPPORT_SOCKS.
+JH/26 Move PROXY support from Experimental to mainline, enabled for a build
+ by defining SUPPORT_PROXY. Note that the proxy_required_hosts option
+ is renamed to hosts_proxy, and the proxy_{host,target}_{address,port}.
+ variables are renamed to proxy_{local,external}_{address,port}.
+
+JH/27 Move Internationalisation support from Experimental to mainline, enabled
+ for a build by defining SUPPORT_I18N
+
+JH/28 Bug 1745: Fix redis lookups to handle (quoted) spaces embedded in parts
+ of the query string, and make ${quote_redis:} do that quoting.
+
+JH/29 Move Events support from Experimental to mainline, enabled by default
+ and removable for a build by defining DISABLE_EVENT.
+
+JH/30 Updated DANE implementation code to current from Viktor Dukhovni.
+
+JH/31 Fix bug with hosts_connection_nolog and named-lists which were wrongly
+ cached by the daemon.
+
+JH/32 Move Redis support from Experimental to mainline, enabled for a build
+ by defining LOOKUP_REDIS. The libhiredis library is required.
+
+JH/33 Bug 1748: Permit ACL dnslists= condition in non-smtp ACLs if explicit
+ keys are given for lookup.
+
+JH/34 Bug 1192: replace the embedded copy of PolarSSL RSA routines in the DKIM
+ support, by using OpenSSL or GnuTLS library ones. This means DKIM is
+ only supported when built with TLS support. The PolarSSL SHA routines
+ are still used when the TLS library is too old for convenient support.
+
+JH/35 Require SINGLE_DH_USE by default in OpenSSL (main config option
+ openssl_options), for security. OpenSSL forces this from version 1.1.0
+ server-side so match that on older versions.
+
+JH/36 Bug 1778: longstanding bug in memory use by the ${run } expansion: A fresh
+ allocation for $value could be released as the expansion processing
+ concluded, but leaving the global pointer active for it.
+
+JH/37 Bug 1769: Permit a VRFY ACL to override the default 252 response,
+ and to use the domains and local_parts ACL conditions.
+
+JH/38 Fix cutthrough bug with body lines having a single dot. The dot was
+ incorrectly not doubled on cutthrough transmission, hence seen as a
+ body-termination at the receiving system - resulting in truncated mails.
+ Commonly the sender saw a TCP-level error, and retransmitted the message
+ via the normal store-and-forward channel. This could result in duplicates
+ received - but deduplicating mailstores were liable to retain only the
+ initial truncated version.
+
+JH/39 Bug 1781: Fix use of DKIM private-keys having trailing '=' in the base-64.
+
+JH/40 Fix crash in queryprogram router when compiled with EXPERIMENTAL_SRS.
+
+JH/41 Bug 1792: Fix selection of headers to sign for DKIM: bottom-up. While
+ we're in there, support oversigning also; bug 1309.
+
+JH/42 Bug 1796: Fix error logged on a malware scanner connection failure.
+
+HS/04 Add support for keep_environment and add_environment options.
+
+JH/43 Tidy coding issues detected by gcc --fsanitize=undefined. Some remain;
+ either intentional arithmetic overflow during PRNG, or testing config-
+ induced overflows.
+
+JH/44 Bug 1800: The combination of a -bhc commandline option and cutthrough
+ delivery resulted in actual delivery. Cancel cutthrough before DATA
+ stage.
+
+JH/45 Fix cutthrough, when connection not opened by verify and target hard-
+ rejects a recipient: pass the reject to the originator.
+
+JH/46 Multiple issues raised by Coverity. Some were obvious or plausible bugs.
+ Many were false-positives and ignorable, but it's worth fixing the
+ former class.
+
+JH/47 Fix build on HP-UX and older Solaris, which need (un)setenv now also
+ for the new environment-manipulation done at startup. Move the routines
+ from being local to tls.c to being global via the os.c file.
+
+JH/48 Bug 1807: Fix ${extract } for the numeric/3-string case. While preparsing
+ an extract embedded as result-arg for a map, the first arg for extract
+ is unavailable so we cannot tell if this is a numbered or keyed
+ extraction. Accept either.
+
Exim version 4.86
-----------------
+
JH/01 Bug 1545: The smtp transport option "retry_include_ip_address" is now
expanded.
now have the list separator specified.
JH/19 Bug 392: spamd_address, and clamd av_scanner, now support retry
- option values.
+ option values.
JH/20 Bug 1571: Ensure that $tls_in_peerdn is set, when verification fails
under OpenSSL.
JH/24 Verification callouts now attempt to use TLS by default.
-HS/01 DNSSEC options (dnssec_require_domains, dnssec_request_domains)
+HS/01 DNSSEC options (dnssec_require_domains, dnssec_request_domains)
are generic router options now. The defaults didn't change.
JH/25 Bug 466: Add RFC2322 support for MIME attachment filenames.
HS/02 Bug 1575: exigrep falls back to autodetection of compressed
files if ZCAT_COMMAND is not executable.
-JH/26 Bug 1539: Add timout/retry options on dnsdb lookups.
+JH/26 Bug 1539: Add timeout/retry options on dnsdb lookups.
JH/27 Bug 286: Support SOA lookup in dnsdb lookups.
JH/28 Bug 1588: Do not use the A lookup following an AAAA for setting the FQDN.
Normally benign, it bites when the pair was led to by a CNAME;
- modern usage is to not canoicalize the domain to a CNAME target
+ modern usage is to not canonicalize the domain to a CNAME target
(and we were inconsistent anyway for A-only vs AAAA+A).
JH/29 Bug 1632: Removed the word "rejected" from line logged for ACL discards.
JH/36 Bug 1659: Guard checking of input smtp commands again pseudo-command
added for tls authenticator.
+HS/03 Add perl_taintmode main config option
+
Exim version 4.85
-----------------
+
TL/01 When running the test suite, the README says that variables such as
no_msglog_check are global and can be placed anywhere in a specific
test's script, however it was observed that placement needed to be near
with certificate use, exposed issues where response data items split
over buffer boundaries were not parsed properly. This eventually
resulted in duplicates being sent. This issue only became common enough
- to notice due to the introduction of conection certificate information,
+ to notice due to the introduction of connection certificate information,
the item size being so much larger. Found and fixed by Wolfgang Breyha.
JH/06 Bug 1533: Fix truncation of items in headers_remove lists. A fixed
JH/07 Add support for directories of certificates when compiled with a GnuTLS
version 3.3.6 or later.
-JH/08 Rename the TPDA expermimental facility to Event Actions. The #ifdef
+JH/08 Rename the TPDA experimental facility to Event Actions. The #ifdef
is EXPERIMENTAL_EVENT, the main-configuration and transport options
both become "event_action", the variables become $event_name, $event_data
and $event_defer_errno. There is a new variable $verify_mode, usable in
JH/15 Updates and fixes to the EXPERIMENTAL_DSN feature.
-JH/16 Fix string representation of time values on 64bit time_t anchitectures.
+JH/16 Fix string representation of time values on 64bit time_t architectures.
Bug 1561.
JH/17 Fix a null-indirection in certextract expansions when a nondefault
return.
JH/01 Bug 1513: Fix parsing of quoted parameter values in MIME headers.
- This was a regression intruduced in 4.83 by another bugfix.
+ This was a regression introduced in 4.83 by another bugfix.
JH/02 Fix broken compilation when EXPERIMENTAL_DSN is enabled.
advertises the facility. If the client requests PRDR a new
acl_data_smtp_prdr ACL is called once for each recipient, after
the body content is received and before the acl_smtp_data ACL.
- The client is controlled by bolth of: a hosts_try_prdr option
+ The client is controlled by both of: a hosts_try_prdr option
on the smtp transport, and the server advertisement.
Default client logging of deliveries and rejections involving
PRDR are flagged with the string "PRDR".
JH/15 AUTH support on callouts (and hence cutthrough-deliveries).
Bugzilla 321, 823.
-TF/04 Added udpsend ACL modifer and hexquote expansion operator
+TF/04 Added udpsend ACL modifier and hexquote expansion operator
PP/21 Fix eximon continuous updating with timestamped log-files.
Broken in a format-string cleanup in 4.80, missed when I repaired the
Analysis and variant patch by Todd Lyons.
NM/04 Bugzilla 1237 - fix cases where printf format usage not indicated
- Bug report from Lars Müller <lars@samba.org> (via SUSE),
+ Bug report from Lars Müller <lars@samba.org> (via SUSE),
Patch from Dirk Mueller <dmueller@suse.com>
PP/13 tls_peerdn now print-escaped for spool files.
diagnostics.
Report and patch from Dmitry Banschikov.
-PP/16 Removed "dont_insert_empty_fragments" fron "openssl_options".
+PP/16 Removed "dont_insert_empty_fragments" from "openssl_options".
Removed SSL_clear() after SSL_new() which led to protocol negotiation
failures. We appear to now support TLS1.1+ with Exim.
has clearer semantics. The /leaky, /strict, and /readonly update modes
are mutually exclusive. The update mode is no longer included in the
database key; it just determines when the database is updated. (This
- means that when you upgrde Exim will forget old rate measurements.)
+ means that when you upgrade Exim will forget old rate measurements.)
Exim now checks that the per_* options are used with an update mode that
makes sense for the current ACL. For example, when Exim is processing a
Exim version 4.75
-----------------
-NM/01 Workround for PCRE version dependency in version reporting
+NM/01 Workaround for PCRE version dependency in version reporting
Bugzilla 1073
TF/01 Update valgrind.h and memcheck.h to copies from valgrind-3.6.0.
variable declaration deep within a block. Bug and patch from
Dennis Davis.
-PP/15 lookups-Makefile IRIX compatibilty coercion.
+PP/15 lookups-Makefile IRIX compatibility coercion.
PP/16 Make DISABLE_DKIM build knob functional.
colons if the lookup type is iplsearch. This is not incompatible, because
previously such lookups could never work.
- The situation is now rather anomolous, since one *can* have colons in
+ The situation is now rather anomalous, since one *can* have colons in
ordinary lsearch keys. However, making the change in all cases is
incompatible and would probably break a number of configurations.
PH/20 Added hosts_avoid_pipelining to the smtp transport.
PH/21 Long custom messages for fakedefer and fakereject are now split up
- into multiline reponses in the same way that messages for "deny" and
+ into multiline responses in the same way that messages for "deny" and
other ACL rejections are.
PH/22 Applied Jori Hamalainen's speed-up changes and typo fixes to exigrep,
runs only) independently of the message's sender address. This meant
that, if the 4xx error was in fact related to the sender, a different
message to the same recipient with a different sender could confuse
- things. In particualar, this can happen when sending to a greylisting
+ things. In particular, this can happen when sending to a greylisting
server, but other circumstances could also provoke similar problems.
I have changed the default so that the retry time for these errors is now
based a combination of the sender and recipient addresses. This change
JJ/03 exipick.20061117.2, made header handling as similar to exim as possible
(added [br]h_ prefixes, implemented RFC2047 decoding. Fixed
- whitesspace changes from 4.64-PH/27
+ whitespace changes from 4.64-PH/27
JJ/04 exipick.20061117.2, fixed format and added $message_headers_raw to
match 4.64-PH/13
(a) Failures to set uid/gid, the current directory, or a process leader
in a subprocess such as that created by queryprogram now generate
- suitable debugging ouput when -d is set.
+ suitable debugging output when -d is set.
(b) The queryprogram router detects when it is not running as root,
outputs suitable debugging information if -d is set, and then runs
and most important:
o fixes a bug in processing the envelope test (when testing
- multiple envelope elements, the last element determinted the
+ multiple envelope elements, the last element determined the
result)
PH/10 Exim was violating RFC 3834 ("Recommendations for Automatic Responses to
Auto-submitted: auto-generated
in the messages that it generates (bounce messages and others, such as
- warnings). In the case of bounce messages for non-SMTP mesages, there was
+ warnings). In the case of bounce messages for non-SMTP messages, there was
also a typo: it was using "Auto_submitted" (underscore instead of
hyphen). Since every message generated by Exim is necessarily in response
to another message, thes have all been changed to:
PH/05 There's a shambles in IRIX6 - it defines EX_OK in unistd.h which conflicts
with the definition in sysexits.h (which is #included earlier).
Fortunately, Exim does not actually use EX_OK. The code used to try to
- preserve the sysexits.h value, by assumimg that macro definitions were
+ preserve the sysexits.h value, by assuming that macro definitions were
scanned for macro replacements. I have been disabused of this notion,
so now the code just undefines EX_OK before #including unistd.h.
SC/01 Eximstats: added -xls and the ability to specify output files
(patch written by Frank Heydlauf).
-SC/02 Eximstats: use FileHandles for outputing results.
+SC/02 Eximstats: use FileHandles for outputting results.
SC/03 Eximstats: allow any combination of xls, txt, and html output.
58. When a "warn" ACL statement has a log_message modifier, the message is
remembered, and not repeated. This is to avoid a lot of repetition when a
message has many recipients that cause the same warning to be written.
- Howewer, Exim was preserving the list of already written lines for an
+ However, Exim was preserving the list of already written lines for an
entire SMTP session, which doesn't seem right. The memory is now reset if a
new message is started.
the list was checked. (An example that provoked this was putting <; in the
middle of a list instead of at the start.) If this happened during a DATA
ACL check, a -D file could be left lying around. This kind of configuration
- error no longer causes Exim to die; instead it causes a defer errror. The
+ error no longer causes Exim to die; instead it causes a defer error. The
incident is still logged to the main and panic logs.
74. Buglet left over from Exim 3 conversion. The message "too many messages
systems (e.g. Solaris), it also passes back the IP address string as the
"host name". However, on others (e.g. Linux), it passes back an empty
string. Exim wasn't checking for this, and was changing the host name to an
- empty string, assuming it had been canonicized.
+ empty string, assuming it had been canonicalized.
5. Although rare, it is permitted to have more than one PTR record for a given
IP address. I thought that gethostbyaddr() or getipnodebyaddr() always gave
13. The install script calls Exim with "-C /dev/null" in order to find the
version number. If ALT_CONFIG_PREFIX was set, this caused an error message
- to be output. Howeve, since Exim outputs its version number before the
+ to be output. However, since Exim outputs its version number before the
error, it didn't break the script. It just looked ugly. I fixed this by
always allowing "-C /dev/null" if the caller is root.
34. Testing for a connection timeout using "timeout_connect" in the retry rules
did not work. The code looks as if it has *never* worked, though it appears
- to have been documented since at least releast 1.62. I have made it work.
+ to have been documented since at least release 1.62. I have made it work.
35. The "timeout_DNS" error in retry rules, also documented since at least
1.62, also never worked. As it isn't clear exactly what this means, and
16. Check for letters, digits, hyphens, and dots in the names of dnslist
domains, and warn by logging if others are found.
-17. At least on BSD, alignment is not guarenteed for the array of ifreq's
+17. At least on BSD, alignment is not guaranteed for the array of ifreq's
returned from GIFCONF when Exim is trying to find the list of interfaces on
a host. The code in os.c has been modified to copy each ifreq to an aligned
structure in all cases.
24. Ignore Sendmail's -Ooption=value command line item.
25. When execve() failed while trying to run a command in a pipe transport,
- Exim was returning EX_UNAVAILBLE (69) from the subprocess. However, this
+ Exim was returning EX_UNAVAILABLE (69) from the subprocess. However, this
could be confused with a return value of 69 from the command itself. This
has been changed to 127, the value the shell returns if it is asked to run
a non-existent command. The wording for the related log line suggests a
47. Change 50 for 4.20 was a heap of junk. I don't know what I was thinking
when I implemented it. It didn't allow for the fact that some option values
- may legitimatetly be negative (e.g. size_addition), and it didn't even do
+ may legitimately be negative (e.g. size_addition), and it didn't even do
the right test for positive values.
48. Domain names in DNS records are case-independent. Exim always looks them up
git://git.exim.org / users / jgh / exim.git / blobdiff