Better notes about handling IPv6 on old machines

[rsync/rsync.git] / TODO

-*- indented-text -*-

BUGS ---------------------------------------------------------------

rsync-url barfs on upload

  rsync foo rsync://localhost/transfer/

  Fix the parser.


There seems to be a bug with hardlinks

  mbp/2 build$ ls -l /tmp/a /tmp/b -i
  /tmp/a:
  total 32
  2568307 -rw-rw-r--    3 mbp      mbp            29 Mar 25 17:30 a1
  2568307 -rw-rw-r--    3 mbp      mbp            29 Mar 25 17:30 a2
  2568307 -rw-rw-r--    3 mbp      mbp            29 Mar 25 17:30 a3
  2568310 -rw-rw-r--    5 mbp      mbp            29 Mar 25 17:30 a4
  2568310 -rw-rw-r--    5 mbp      mbp            29 Mar 25 17:30 a5
  2568310 -rw-rw-r--    5 mbp      mbp            29 Mar 25 17:30 b1
  2568310 -rw-rw-r--    5 mbp      mbp            29 Mar 25 17:30 b2
  2568310 -rw-rw-r--    5 mbp      mbp            29 Mar 25 17:30 b3

  /tmp/b:
  total 32
  2568309 -rw-rw-r--    3 mbp      mbp            29 Mar 25 17:30 a1
  2568309 -rw-rw-r--    3 mbp      mbp            29 Mar 25 17:30 a2
  2568309 -rw-rw-r--    3 mbp      mbp            29 Mar 25 17:30 a3
  2568311 -rw-rw-r--    5 mbp      mbp            29 Mar 25 17:30 a4
  2568311 -rw-rw-r--    5 mbp      mbp            29 Mar 25 17:30 a5
  2568311 -rw-rw-r--    5 mbp      mbp            29 Mar 25 17:30 b1
  2568311 -rw-rw-r--    5 mbp      mbp            29 Mar 25 17:30 b2
  2568311 -rw-rw-r--    5 mbp      mbp            29 Mar 25 17:30 b3
  mbp/2 build$ rm -r /tmp/b && ./rsync -avH /tmp/a/ /tmp/b
  building file list ... done
  created directory /tmp/b
  ./
  a1
  a4
  a2 => a1
  a3 => a2
  wrote 350 bytes  read 52 bytes  804.00 bytes/sec
  total size is 232  speedup is 0.58
  mbp/2 build$ rm -r /tmp/b
  mbp/2 build$ ls -l /tmp/b
  ls: /tmp/b: No such file or directory
  mbp/2 build$ rm -r /tmp/b && ./rsync -avH /tmp/a/ /tmp/b
  rm: cannot remove `/tmp/b': No such file or directory
  mbp/2 build$ rm -f -r /tmp/b && ./rsync -avH /tmp/a/ /tmp/b
  building file list ... done
  created directory /tmp/b
  ./
  a1
  a4
  a2 => a1
  a3 => a2
  wrote 350 bytes  read 52 bytes  804.00 bytes/sec
  total size is 232  speedup is 0.58
  mbp/2 build$ ls -l /tmp/b
  total 32
  -rw-rw-r--    3 mbp      mbp            29 Mar 25 17:30 a1
  -rw-rw-r--    3 mbp      mbp            29 Mar 25 17:30 a2
  -rw-rw-r--    3 mbp      mbp            29 Mar 25 17:30 a3
  -rw-rw-r--    5 mbp      mbp            29 Mar 25 17:30 a4
  -rw-rw-r--    5 mbp      mbp            29 Mar 25 17:30 a5
  -rw-rw-r--    5 mbp      mbp            29 Mar 25 17:30 b1
  -rw-rw-r--    5 mbp      mbp            29 Mar 25 17:30 b2
  -rw-rw-r--    5 mbp      mbp            29 Mar 25 17:30 b3
  mbp/2 build$ ls -l /tmp/a
  total 32
  -rw-rw-r--    3 mbp      mbp            29 Mar 25 17:30 a1
  -rw-rw-r--    3 mbp      mbp            29 Mar 25 17:30 a2
  -rw-rw-r--    3 mbp      mbp            29 Mar 25 17:30 a3
  -rw-rw-r--    5 mbp      mbp            29 Mar 25 17:30 a4
  -rw-rw-r--    5 mbp      mbp            29 Mar 25 17:30 a5
  -rw-rw-r--    5 mbp      mbp            29 Mar 25 17:30 b1
  -rw-rw-r--    5 mbp      mbp            29 Mar 25 17:30 b2
  -rw-rw-r--    5 mbp      mbp            29 Mar 25 17:30 b3


Progress indicator can produce corrupt output when transferring directories:

  main/binary-arm/
  main/binary-arm/admin/
  main/binary-arm/base/
  main/binary-arm/comm/8.56kB/s    0:00:52
  main/binary-arm/devel/
  main/binary-arm/doc/
  main/binary-arm/editors/
  main/binary-arm/electronics/s    0:00:53
  main/binary-arm/games/
  main/binary-arm/graphics/
  main/binary-arm/hamradio/
  main/binary-arm/interpreters/
  main/binary-arm/libs/6.61kB/s    0:00:54
  main/binary-arm/mail/
  main/binary-arm/math/
  main/binary-arm/misc/


lchmod
  I don't think we handle this properly on systems that don't have the
  call.  Are there any such?


Cross-test versions
  Part of the regression suite should be making sure that we don't
  break backwards compatibility: old clients vs new servers and so
  on.  Ideally we would test the cross product of versions.  

  It might be sufficient to test downloads from well-known public
  rsync servers running different versions of rsync.  This will give
  some testing and also be the most common case for having different
  versions and not being able to upgrade.

--no-blocking-io might be broken

  in the same way as --no-whole-file; somebody needs to check.

Do not rely on having a group called "nobody"

  http://www.linuxbase.org/spec/refspecs/LSB_1.1.0/gLSB/usernames.html

  On Debian it's "nogroup"

DAEMON --------------------------------------------------------------

server-imposed bandwidth limits

rsyncd over ssh

  There are already some patches to do this.

  BitKeeper uses a server whose login shell is set to bkd.  That's
  probably a reasonable approach.


FEATURES ------------------------------------------------------------


--dry-run is too dry

  Mark Santcroos points out that -n fails to list files which have
  only metadata changes, though it probably should.  

  There may be a Debian bug about this as well.


use chroot

  If the platform doesn't support it, then don't even try.

  If running as non-root, then don't fail, just give a warning.
  (There was a thread about this a while ago?)

    http://lists.samba.org/pipermail/rsync/2001-August/thread.html
    http://lists.samba.org/pipermail/rsync/2001-September/thread.html


--files-from

  Avoids traversal.  Better option than a pile of --include statements
  for people who want to generate the file list using a find(1)
  command or a script.


supplementary groups

  Perhaps allow supplementary groups to be specified in rsyncd.conf;
  then make the first one the primary gid and all the rest be
  supplementary gids.


File list structure in memory

  Rather than one big array, perhaps have a tree in memory mirroring
  the directory tree.  

  This might make sorting much faster!  (I'm not sure it's a big CPU
  problem, mind you.)  

  It might also reduce memory use in storing repeated directory names
  -- again I'm not sure this is a problem.

Performance

  Traverse just one directory at a time.  Tridge says it's possible.

  At the moment rsync reads the whole file list into memory at the
  start, which makes us use a lot of memory and also not pipeline
  network access as much as we could.


Handling duplicate names

  We need to be careful of duplicate names getting into the file list.
  See clean_flist().  This could happen if multiple arguments include
  the same file.  Bad.

  I think duplicates are only a problem if they're both flowing
  through the pipeline at the same time.  For example we might have
  updated the first occurrence after reading the checksums for the
  second.  So possibly we just need to make sure that we don't have
  both in the pipeline at the same time.  

  Possibly if we did one directory at a time that would be sufficient.

  Alternatively we could pre-process the arguments to make sure no
  duplicates will ever be inserted.  There could be some bad cases
  when we're collapsing symlinks.

  We could have a hash table.

  The root of the problem is that we do not want more than one file
  list entry referring to the same file.  At first glance there are
  several ways this could happen: symlinks, hardlinks, and repeated
  names on the command line.

  If names are repeated on the command line, they may be present in
  different forms, perhaps by traversing directory paths in different
  ways, traversing paths including symlinks.  Also we need to allow
  for expansion of globs by rsync.

  At the moment, clean_flist() requires having the entire file list in
  memory.  Duplicate names are detected just by a string comparison.

  We don't need to worry about hard links causing duplicates because
  files are never updated in place.  Similarly for symlinks.

  I think even if we're using a different symlink mode we don't need
  to worry.

  Unless we're really clever this will introduce a protocol
  incompatibility, so we need to be able to accept the old format as
  well.


Memory accounting

  At exit, show how much memory was used for the file list, etc.

  Also we do a wierd exponential-growth allocation in flist.c.  I'm
  not sure this makes sense with modern mallocs.  At any rate it will
  make us allocate a huge amount of memory for large file lists.


Hard-link handling

  At the moment hardlink handling is very expensive, so it's off by
  default.  It does not need to be so.  

  Since most of the solutions are rather intertwined with the file
  list it is probably better to fix that first, although fixing
  hardlinks is possibly simpler.

  We can rule out hardlinked directories since they will probably
  screw us up in all kinds of ways.  They simply should not be used.

  At the moment rsync only cares about hardlinks to regular files.  I
  guess you could also use them for sockets, devices and other beasts,
  but I have not seen them.

  When trying to reproduce hard links, we only need to worry about
  files that have more than one name (nlinks>1 && !S_ISDIR). 

  The basic point of this is to discover alternate names that refer to
  the same file.  All operations, including creating the file and
  writing modifications to it need only to be done for the first name.
  For all later names, we just create the link and then leave it
  alone.

  If hard links are to be preserved:

    Before the generator/receiver fork, the list of files is received
    from the sender (recv_file_list), and a table for detecting hard
    links is built.

    The generator looks for hard links within the file list and does
    not send checksums for them, though it does send other metadata.

    The sender sends the device number and inode with file entries, so
    that files are uniquely identified.

    The receiver goes through and creates hard links (do_hard_links)
    after all data has been written, but before directory permissions
    are set.

  At the moment device and inum are sent as 4-byte integers, which
  will probably cause problems on large filesystems.  On Linux the
  kernel uses 64-bit ino_t's internally, and people will soon have
  filesystems big enough to use them.  We ought to follow NFS4 in
  using 64-bit device and inode identification, perhaps with a
  protocol version bump.

  Once we've seen all the names for a particular file, we no longer
  need to think about it and we can deallocate the memory.

  We can also have the case where there are links to a file that are
  not in the tree being transferred.  There's nothing we can do about
  that.  Because we rename the destination into place after writing,
  any hardlinks to the old file are always going to be orphaned.  In
  fact that is almost necessary because otherwise we'd get really
  confused if we were generating checksums for one name of a file and
  modifying another.

  At the moment the code seems to make a whole second copy of the file
  list, which seems unnecessary.

  We should have a test case that exercises hard links.  Since it
  might be hard to compare ./tls output where the inodes change we
  might need a little program to check whether several names refer to
  the same file.



Handling IPv6 on old machines

  The KAME IPv6 patch is nice in theory but has proved a bit of a
  nightmare in practice.  The basic idea of their patch is that rsync
  is rewritten to use the new getaddrinfo()/getnameinfo() interface,
  rather than gethostbyname()/gethostbyaddr() as in rsync 2.4.6.
  Systems that don't have the new interface are handled by providing
  our own implementation in lib/, which is selectively linked in.

  The problem with this is that it is really hard to get right on
  platforms that have a half-working implementation, so redefining
  these functions clashes with system headers, and leaving them out
  breaks.  This affects at least OSF/1, RedHat 5, and Cobalt, which
  are moderately improtant.

  Perhaps the simplest solution would be to have two different files
  implementing the same interface, and choose either the new or the
  old API.  This is probably necessary for systems that e.g. have
  IPv6, but gethostbyaddr() can't handle it.  The Linux manpage claims
  this is currently the case.

  In fact, our internal sockets interface (things like
  open_socket_out(), etc) is much narrower than the getaddrinfo()
  interface, and so probably simpler to get right.  In addition, the
  old code is known to work well on old machines.

  We could drop the rather large lib/getaddrinfo files.


Other IPv6 stuff:
  
  Implement suggestions from http://www.kame.net/newsletter/19980604/
  and ftp://ftp.iij.ad.jp/pub/RFC/rfc2553.txt

  If a host has multiple addresses, then listen try to connect to all
  in order until we get through.  (getaddrinfo may return multiple
  addresses.)  This is kind of implemented already.

  Possibly also when starting as a server we may need to listen on
  multiple passive addresses.  This might be a bit harder, because we
  may need to select on all of them.  Hm.

  Define a syntax for IPv6 literal addresses.  Since they include
  colons, they tend to break most naming systems, including ours.
  Based on the HTTP IPv6 syntax, I think we should use
 
     rsync://[::1]/foo/bar
     [::1]::bar

  which should just take a small change to the parser code.


Errors

  If we hang or get SIGINT, then explain where we were up to.  Perhaps
  have a static buffer that contains the current function name, or
  some kind of description of what we were trying to do.  This is a
  little easier on people than needing to run strace/truss.

  "The dungeon collapses!  You are killed."  Rather than "unexpected
  eof" give a message that is more detailed if possible and also more
  helpful.  

  If we get an error writing to a socket, then we should perhaps
  continue trying to read to see if an error message comes across
  explaining why the socket is closed.  I'm not sure if this would
  work, but it would certainly make our messages more helpful.

  What happens if a directory is missing -x attributes.  Do we lose
  our load?  (Debian #28416)  Probably fixed now, but a test case
  would be good.


File attributes

  Device major/minor numbers should be at least 32 bits each.  See
  http://lists.samba.org/pipermail/rsync/2001-November/005357.html

  Transfer ACLs.  Need to think of a standard representation.
  Probably better not to even try to convert between NT and POSIX.
  Possibly can share some code with Samba.

Empty directories

  With the current common --include '*/' --exclude '*' pattern, people
  can end up with many empty directories.  We might avoid this by
  lazily creating such directories.


zlib

  Perhaps don't use our own zlib.  

  Advantages:
   
    - will automatically be up to date with bugfixes in zlib

    - can leave it out for small rsync on e.g. recovery disks

    - can use a shared library

    - avoids people breaking rsync by trying to do this themselves and
      messing up

  Should we ship zlib for systems that don't have it, or require
  people to install it separately?

  Apparently this will make us incompatible with versions of rsync
  that use the patched version of rsync.  Probably the simplest way to
  do this is to just disable gzip (with a warning) when talking to old
  versions.


logging

  Perhaps flush stdout after each filename, so that people trying to
  monitor progress in a log file can do so more easily.  See
  http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=48108

  At the connections that just get a list of modules are not logged,
  but they should be.

  If a child of the rsync daemon dies with a signal, we should notice
  that when we reap it and log a message.

  Keep stderr and stdout properly separated (Debian #23626)

  After we get the @RSYNCD greeting from the server, we know it's
  version but we have not yet sent the command line, so we could just
  remove the -z option if the server is too old.  

  For ssh invocation it's not so simple, because we actually use the
  command line to start the remote process.  However, we only actually
  do compression in token.c, and we could therefore once we discover
  the remote version emit an error if it's too old.  I'm not sure if
  that's a good tradeoff or not.


rsyncd over ssh

  There are already some patches to do this.

proxy authentication

  Allow RSYNC_PROXY to be http://user:pass@proxy.foo:3128/, and do
  HTTP Basic Proxy-Authentication.  

  Multiple schemes are possible, up to and including the insanity that
  is NTLM, but Basic probably covers most cases.

SOCKS

  Add --with-socks, and then perhaps a command-line option to put them
  on or off.  This might be more reliable than LD_PRELOAD hacks.

FAT support

  rsync to a FAT partition on a Unix machine doesn't work very well
  at the moment.  I think we get errors about invalid filenames and 
  perhaps also trying to do atomic renames.

  I guess the code to do this is currently #ifdef'd on Windows; perhaps
  we ought to intelligently fall back to it on Unix too.


Better statistics:

  <Rasmus> mbp: hey, how about an rsync option that just gives you the
  summary without the list of files?  And perhaps gives more
  information like the number of new files, number of changed,
  deleted, etc. ?
  <mbp> Rasmus: nice idea
  <mbp> there is --stats
  <mbp> but at the moment it's very tridge-oriented
  <mbp> rather than user-friendly
  <mbp> it would be nice to improve it
  <mbp> that would also work well with --dryrun

TDB:

  Rather than storing the file list in memory, store it in a TDB.

  This *might* make memory usage lower while building the file list.

  Hashtable lookup will mean files are not transmitted in order,
  though... hm.

  This would neatly eliminate one of the major post-fork shared data
  structures.


chmod:

  On 12 Mar 2002, Dave Dykstra <dwd@bell-labs.com> wrote:
  > If we would add an option to do that functionality, I would vote for one
  > that was more general which could mask off any set of permission bits and
  > possibly add any set of bits.  Perhaps a chmod-like syntax if it could be
  > implemented simply.

  I think that would be good too.  For example, people uploading files   
  to a web server might like to say

  rsync -avzP --chmod a+rX ./ sourcefrog.net:/home/www/sourcefrog/

  Ideally the patch would implement as many of the gnu chmod semantics
  as possible.  I think the mode parser should be a separate function
  that passes back something like (mask,set) description to the rest of
  the program.  For bonus points there would be a test case for the  
  parser.

  Possibly also --chown

  (Debian #23628)


--diff

  Allow people to specify the diff command.  (Might want to use wdiff,
  gnudiff, etc.)

  Just diff the temporary file with the destination file, and delete
  the tmp file rather than moving it into place.

  Interaction with --partial.

  Security interactions with daemon mode?

  (Suggestion from david.e.sewell)


Incorrect timestamps (Debian #100295)

  A bit hard to believe, but apparently it happens.


Check "refuse options works"

  We need a test case for this...

  Was this broken when we changed to popt?


PERFORMANCE ----------------------------------------------------------

MD4 file_sum

  If we're doing a local transfer, or using -W, then perhaps don't
  send the file checksum.  If we're doing a local transfer, then
  calculating MD4 checksums uses 90% of CPU and is unlikely to be
  useful.

  Indeed for transfers over zlib or ssh we can also rely on the
  transport to have quite strong protection against corruption.

  Perhaps we should have an option to disable this, analogous to
  --whole-file, although it would default to disabled.  The file
  checksum takes up a definite space in the protocol -- we can either
  set it to 0, or perhaps just leave it out.

MD4

  Perhaps borrow an assembler MD4 from someone?

  Make sure we call MD4 with properly-sized blocks whenever possible
  to avoid copying into the residue region?

String area code

  Test whether this is actually faster than just using malloc().  If
  it's not (anymore), throw it out.
          

PLATFORMS ------------------------------------------------------------

Win32

  Don't detach, because this messes up --srvany.

  http://sources.redhat.com/ml/cygwin/2001-08/msg00234.html

  According to "Effective TCP/IP Programming" (??) close() on a socket
  has incorrect behaviour on Windows -- it sends a RST packet to the
  other side, which gives a "connection reset by peer" error.  On that
  platform we should probably do shutdown() instead.  However, on Unix
  we are correct to call close(), because shutdown() discards
  untransmitted data.


DEVELOPMENT ----------------------------------------------------------

Splint

  Build rsync with SPLINT to try to find security holes.  Add
  annotations as necessary.  Keep track of the number of warnings
  found initially, and see how many of them are real bugs, or real
  security bugs.  Knowing the percentage of likely hits would be
  really interesting for other projects.

Torture test

  Something that just keeps running rsync continuously over a data set
  likely to generate problems.

Cross-testing

  Run current rsync versions against significant past releases.

Memory debugger

  jra recommends Valgrind:

    http://devel-home.kde.org/~sewardj/

Release script
  
  Update spec files

  Build tar file; upload

  Send announcement to mailing list and c.o.l.a.
  
  Make freshmeat announcement

  Update web site



TESTING --------------------------------------------------------------

Cross-test versions

  Part of the regression suite should be making sure that we don't
  break backwards compatibility: old clients vs new servers and so
  on.  Ideally we would test both up and down from the current release
  to all old versions.

  We might need to omit broken old versions, or versions in which
  particular functionality is broken

  It might be sufficient to test downloads from well-known public
  rsync servers running different versions of rsync.  This will give
  some testing and also be the most common case for having different
  versions and not being able to upgrade.


Test on kernel source

  Download all versions of kernel; unpack, sync between them.  Also
  sync between uncompressed tarballs.  Compare directories after
  transfer.

  Use local mode; ssh; daemon; --whole-file and --no-whole-file.

  Use awk to pull out the 'speedup' number for each transfer.  Make
  sure it is >= x.


Test large files

  Sparse and non-sparse

Mutator program

  Insert bytes, delete bytes, swap blocks, ...

configure option to enable dangerous tests

If tests are skipped, say why.

Test daemon feature to disallow particular options.

Pipe program that makes slow/jerky connections.

Versions of read() and write() that corrupt the stream, or abruptly fail

Separate makefile target to run rough tests -- or perhaps just run
them every time?

Test "refuse options" works

  What about for --recursive?

  If you specify an unrecognized option here, you should get an error.


DOCUMENTATION --------------------------------------------------------

Update README

Keep list of open issues and todos on the web site

Update web site from CVS


Perhaps redo manual as SGML

  The man page is getting rather large, and there is more information
  that ought to be added.

  TexInfo source is probably a dying format.

  Linuxdoc looks like the most likely contender.  I know DocBook is
  favoured by some people, but it's so bloody verbose, even with emacs
  support.


BUILD FARM -----------------------------------------------------------

Add machines

  AMDAHL UTS (Dave Dykstra)

  Cygwin (on different versions of Win32?)

  HP-UX variants (via HP?)

  SCO


LOGGING --------------------------------------------------------------

  Perhaps flush stdout after each filename, so that people trying to
  monitor progress in a log file can do so more easily.  See
  http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=48108

  At the connections that just get a list of modules are not logged,
  but they should be.

  If a child of the rsync daemon dies with a signal, we should notice
  that when we reap it and log a message.

  Keep stderr and stdout properly separated (Debian #23626)

  Use a separate function for reporting errors; prefix it with
  "rsync:" or "rsync(remote)", or perhaps even "rsync(local
  generator): ".

verbose output
  
  Indicate whether files are new, updated, or deleted

  At end of transfer, show how many files were or were not transferred
  correctly.

-vv

  Explain *why* every file is transferred or not (e.g. "local mtime
  123123 newer than 1283198")


debugging of daemon

  Add an rsyncd.conf parameter to turn on debugging on the server.



NICE -----------------------------------------------------------------

--no-detach and --no-fork options

  Very useful for debugging.  Also good when running under a
  daemon-monitoring process that tries to restart the service when the
  parent exits.

hang/timeout friendliness

internationalization

  Change to using gettext().  Probably need to ship this for platforms
  that don't have it.  

  Solicit translations.

  Does anyone care?  Before we bother modifying the code, we ought to
  get the manual translated first, because that's possibly more useful
  and at any rate demonstrates desire.

rsyncsh 

   Write a small emulation of interactive ftp as a Pythonn program
   that calls rsync.  Commands such as "cd", "ls", "ls *.c" etc map
   fairly directly into rsync commands: it just needs to remember the
   current host, directory and so on.  We can probably even do
   completion of remote filenames.


RELATED PROJECTS -----------------------------------------------------

http://rsync.samba.org/rsync-and-debian/

rsyncable gzip patch

  Exhaustive, tortuous testing

  Cleanups?

rsyncsplit as alternative to real integration with gzip?

reverse rsync over HTTP Range

  Goswin Brederlow suggested this on Debian; I think tridge and I
  talked about it previous in relation to rproxy.