Changes to procmail:
- Regression bugs from 3.20:
- Broke compilation with K&R compilers
- procmail -p -m was overridding PATH
- maildir delivery included garbage in filenames
- Mismatched HOST in last rcfile didn't discard the message
- COMSAT wasn't turned off by an rcfile on the command line
- Catch overly long command line variable assignments
- If a command expansion is truncated, set PROCMAIL_OVERFLOW
and don't trim trailing (really middle) newlines
- If the comsat host can't be resolved, set COMSAT to "no"
Some fixes to the man pages
More paranoia: start to use strlcpy()
Generate safe temp and maildir filenames when the hostname
contains / or : by mapping them to \ooo
I've also seen corrupted From lines (missing F) when I upgraded to 3.21.
Someone reported in the Esd-l mailing list (see
http://www.spconnect.com/pipermail/esd-l/
for archives) that 3.22 solved the problem for him.
- Incorrect prototype broke compilation on Tru64 UNIX
- INCLUDERC was broken by trying to be fancy and not fully
succeeding when support for multiple -a options was added
Also set USE_BUILDLINK_ONLY.
Changes to procmail:
- SECURITY: don't do unsafe things from signal handlers:
- ignore TRAP when terminating because of a signal
- resolve the host and protocol of COMSAT when it is set
- save the absolute path form of $LASTFOLDER for the comsat
message when it is set
- only use the log buffer if it's safe
- Support LMTP for delivery mode (not enabled by default)
- Preliminary support for using mmap() for `large' messages
(this doesn't work yet)
- SWITCHRC=zero-length-file didn't always abort the current
rcfile
- A race to create the mailspool would bounce one of the
messages due to an internal error
- LC_ in KEEPENV would preserve only the first LC_foo variable
- Strip runtime linker variables (LD_*) from environment on
all platforms
- Drop duplicate and malformed environment entries
- Multiple -a options will now set $2, $3, etc
- Command line assignments to INCLUDERC and SWITCHRC no longer
have any effect
- Worked around AIX 4.3.3 xlc compiler (incorrect file-scope
variable initialization)
- When delivering to a maildir, don't force the message to end
with an empty line
- Be more paranoid about leaking information between recipients
- Unset LOCKFILE if we can't actually lock it
- Set MAILDIR to '.' if the chdir fails
- LASTFOLDER was sometimes set by '?' conditions
- Buffer the log more efficiently
- Use the `standard' format for maildir filenames and retry
on name collision
- Rename by linking to prevent lossage
Changes to lockfile:
- Include the system mailbox lockfile path in the -v output
- Resist attempts to use a setuid lockfile command
- Fix infinite loop on -l, -r, or -s flag with no value.
Documented formail's treatment of >From_ lines as continuations
of the From_ line and warned of problems caused by non-RFC822
field names like 'Old-From '
Clarified procmail's treatment of $@ and $#
Fixed a man page formatting problem
Use long, not off_t, with fseek()/ftell()
Increase our paranoia: start to use strlcat()
The default MAILDIR is now configurable separately from the
default rcfile location
Include an RPM spec file in the examples directory for automated
builds
Include and use mkinstalldirs
Called nice() when shouldn't have (this time for sure!)
Some of the more interesting changes:
- A race to create the mailspool would bounce one of the
messages due to an internal error
- LC_ in KEEPENV would preserve only the first LC_foo variable
- Strip runtime linker variables (LD_*) from environment on
all platforms
- Drop duplicate and malformed environment entries
- Fixed a man page formatting problem
