expire-iso8601: report and implement oldest first

Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>

expire-iso8601: compensate for insufficient density

Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>

expire-iso8601: break out $use->()

Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>

expire-iso8601: Set $|=1 to clarify output order

Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>

expire-iso8601: include #0 too

Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>

Revert "expire-iso8601: Use date(1)"

It's not faster.

With DateTime::Format::Strptime, whole program runtime on my
test case:
  real    0m1.371s
  user    0m1.254s
  sys     0m0.117s

With date(1):
  real    0m5.408s
  user    0m1.834s
  sys     0m3.683s

This reverts commit 2d3da03e893d0c59e236224f049215fab8e81e8e.

Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>

expire-iso8601: Use date(1)

-davenant:~> time perl -e 'use DateTime::Format::Strptime;'

real    0m1.269s
user    0m1.175s
sys     0m0.093s

Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>

expire-iso8601: Fix a dependency

Actually this is huge so make it Recommends.

Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>

expire-iso8601: Fix a dependency

This is just a perl lib so ask for it.

Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>

changelog: document expire-8601 changes

expire-iso8601: Replace with new implementation

This fixes the design problem and is also much more maintainable and
comprehensible.

Closes: #862897
Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>

expire-8601: new version fixes

Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>

expire-8601: new version seems good now

Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>

expire-8601: wip new version, err on other side

Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>

expire-8601: wip new version, sort of works, need adj

Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>

expire-8601: wip new version

Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>

expire-8601: wip new version

Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>

expire-8601: wip new version

Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>

expire-8601: wip new version

Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>

expire-8601: wip new version

Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>

expire-8601: wip new version

Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>

changelog: document changes and finalise 6.1.2

Ready for upload, which will involve wearing also my other hat.

Signed-off-by: Ian Jackson <ian.jackson@eu.citrix.com>

git-cache-proxy: Do not update timestamp due to housekeeping gc

Otherwise gc will always reset the timestamp and nothing will be
cleaned up.

Signed-off-by: Ian Jackson <ian.jackson@eu.citrix.com>

changelog: Start ~citrix3~

Signed-off-by: Ian Jackson <ian.jackson@eu.citrix.com>

git-cache-proxy: Tolerate failure to write gc stamp

This can happen if the directory no longer exists.

Signed-off-by: Ian Jackson <ian.jackson@eu.citrix.com>

git-cache-proxy: Recheck gc timestamp after taking lock

Otherwise we can repeatedly gc the same thing.

Signed-off-by: Ian Jackson <ian.jackson@eu.citrix.com>

git-cache-proxy: gc: Break out $gccheck

No functional change.

Signed-off-by: Ian Jackson <ian.jackson@eu.citrix.com>

changelog: Document changes and finalise 6.1.2~citrix2~

Signed-off-by: Ian Jackson <ian.jackson@eu.citrix.com>

git-cache-proxy: Use open-coded fork/exec for git gc

This allows us to handle ENOENT from chdir specially.  We want to
tolerate this, because it can happen if the previous lockholder
cleaned it all up.

Signed-off-by: Ian Jackson <ian.jackson@eu.citrix.com>

git-cache-proxy: Abandon fetch attempt if client disconnects

The git client sends the command and then waits for our response.  We
can tell if it has disconnected, because our stdin (the socket) will
be readable.

This is a good idea because otherwise if we have clients with their
own (formal or informal) timeouts, we can build up a backlog of
queued useless fetch attempts.

Signed-off-by: Ian Jackson <ian.jackson@eu.citrix.com>

git-cache-proxy: Move $SIG{ALRM} setting to global

This is actually used in multiple places.  It should be set globally.
No overall functional change with the current code.

Signed-off-by: Ian Jackson <ian.jackson@eu.citrix.com>

changelog: Start 6.1.2~citrix2~

Use 2~ since there was a 1~ floating about on mariner...

Signed-off-by: Ian Jackson <ian.jackson@eu.citrix.com>

Merge remote-tracking branch 'dgit/dgit/sid'

Signed-off-by: Ian Jackson <ian.jackson@citrix.com>

changelog: Bump version to 6.1.1 and finalise

I found some 6.1.1~citrix1~ on mariner, so go higher than that.

Signed-off-by: Ian Jackson <ian.jackson@eu.citrix.com>

changelog: Document Citrix changes

Signed-off-by: Ian Jackson <ian.jackson@eu.citrix.com>

Merge remote-tracking branch 'mariner/master'

Signed-off-by: Ian Jackson <ian.jackson@citrix.com>

git-cache-proxy: Document some options in the head comment

Signed-off-by: Ian Jackson <ian.jackson@eu.citrix.com>

git-cache-proxy: Periodically run `git gc --quiet'

git has a feature which is supposed to automatically run git gc.

But if your repository accumulates enough loose objects, then this
occurs:

 $ git gc --auto
 Auto packing the repository in background for optimum performance.
 See "git help gc" for manual housekeeping.
 error: The last gc run reported the following. Please correct the root cause
 and remove gc.log.
 Automatic cleanup will not be performed until the file is removed.

 warning: There are too many unreachable loose objects; run 'git prune' to remove them.

Removing $GIT_DIR/gc.log generally simply causes git gc --auto to
print, again, the message:

 warning: There are too many unreachable loose objects; run 'git prune' to remove them.

Ie, a repository can get into a state where it needs gc so badly that
git will not gc it automatically.  This is, of course, mad.  It seems
to have been a misguided attempt at a safety catch.  Unfortunately
there does not even appear to be a configuration knob to set this `too
many' limit to infinity.

Work around this problem by running `git gc --quiet' (every 10 days,
by default).  This is not ideal because it runs even if we haven't
updated the tree, but we would rather do it out of housekeeping rather
than in the middle of fetch/clone.

`git gc' without --auto does not refuse to prune the objects that need
pruning, and fixes the repository.

We leave the gc.auto config option enabled since I think it is largely
harmless, and disabling it would be extra work.

Signed-off-by: Ian Jackson <ian.jackson@eu.citrix.com>

git-cache-proxy: Refactor to prepare for 2nd kind of housekeeping

Generalise the housekeeping action code.
Introduce new variables $mode_what, $mode_locknb, and $mode_action.
Move the actual cleanup functionality into the $mode_action value.

No overall functional change.

Signed-off-by: Ian Jackson <ian.jackson@eu.citrix.com>

changelog: start ~citrix2

Signed-off-by: Ian Jackson <ian.jackson@eu.citrix.com>

changelog: finalise 6.0.5~citrix1

fishdescriptor: Use Python "errno" module

Use Python "errno" module, not "os.errno" (which is not in modern
python3 apparently).

Signed-off-by: Ian Jackson <ian.jackson@eu.citrix.com>

changelog: Start 6.0.5~citrix1

Signed-off-by: Ian Jackson <ian.jackson@eu.citrix.com>

git-branchmove: rewrite in perl

Closes: #914398, #914399
Signed-off-by: Sean Whitton <spwhitton@spwhitton.name>

changelog: finalise 6.0.5~iwj1

Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>

changelog: Document rc.local change

Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>

really: Default config: Writeability of rc.local suffices, too

This makes it work on systems not using sysvinit and which therefore
don't have /etc/inittab.

Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>

really: Support REALLY_CHECK_FILE_2 for checking multiple files

Only works if REALLY_CHECK_FILE is defined too.

Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>

really: Invert logic in REALLY_CHECK_FILE checkroot

This will make it possible to check for a 2nd file.

No functional change.

Reviewed-by: Clare Boothby <clareb@chiark.greenend.org.uk>
Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>

changelog: start 6.0.5

Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>

changelog: finalise 6.0.4

Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>

changelog: Document bug number for bugfix in 6.0.4~citrix1.

Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>

sync-accounts: Fix perl syntax error.  Closes:#865985.

Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>

changelog: document fishdescriptor errno fix

Signed-off-by: Ian Jackson <ian.jackson@eu.citrix.com>

fishdescriptor: cast __errno_location correctly

The return value from __errno_location() is an int*.  My syntax had
erroneously specified that it returned an int.  Nothing spotted this
bug because the expression is evaluated by gdb whose C interpreter is
very strange.  In particular, gdb lets you dereference an int, even on
a platform where ints are 32-bit and pointers are 64-bit.

If you are on a 32-bit platform, this does not matter.  Likewise if
you are on a 64-bit platform and the address of errno happens, by
luck, to be within the 32-bit addressable part of the space.

If you are mildly lucky the result of this is an error like this:
  gdb.MemoryError: Cannot access memory at address 0x2f24ef10
buried in stack traces from fishdescriptor.

If you are unlucky, fishdescriptor will successfully access some wrong
location.  This means it does not actually save and restore errno,
since it saves and restores somewhere else instead.  So fishdescriptor
will corrupt the errno value of the thread that it happens to
be (ab)using, overwriting it with the errno from fishdescriptor's own
calls, possibly causing the target program to become confused about
the error(s) from system call(s) it is making.

If you are very unlucky, fishdescriptor will successfully access some
wrong location which is actively in use by something outside the
target process (eg, direct IO, shared memory0.  fishdescriptor will
save the value (a single int) and then restore it a bit later.  This
might in principle cause any kind of arbitrary lossage.

Signed-off-by: Ian Jackson <ian.jackson@eu.citrix.com>

changelog: finalise 6.0.3

Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>

changelog: make 6.0.3~citrix1 entry

gbp dch, and edited details, version and trailer line

Signed-off-by: Ian Jackson <ian.jackson@eu.citrix.com>

fishdescriptor: When looking for run_dir, look in target root

This is correct, since we are operating in our environment but we are
going to use the path in the target's root.

Signed-off-by: Ian Jackson <ian.jackson@eu.citrix.com>

fishdescriptor: Calculate target_root earlier and pass to _sock_dir

No functional change.

We are going to use this in a moment.

Signed-off-by: Ian Jackson <ian.jackson@eu.citrix.com>

fishdescriptor: Cope if donor is in a chroot without /proc

`our_sockname' contains something like
  /proc/DONOR/root/run/user/UID/fishdescriptor/SOCKET

When I tested this I did it in a chroot where /proc was mounted.  In
such a chroot /proc/DONOR/root -> /.  So this works.

But if there is no /proc, it does not work.

Instead, we can can and should assume that the path in the donor is
relative to its own root.  Ie, we can just use `sockname'.

CC: George Dunlap <george.dunlap@citrix.com>
Signed-off-by: Ian Jackson <ian.jackson@eu.citrix.com>

git-branchmove: Avoid transporting tags about

Signed-off-by: Ian Jackson <ian.jackson@eu.citrix.com>

changelog: Prep for Rebuild with arch:all packages included in the upload.

changelog: finalise 6.0.1

changelog: discuss fishdescriptor bugs

fishdescriptor: Actually tidy up actual2intended when closing

Otherwise we will probably try to move the new-defunct open-file
somewhere when we try to make the descriptor refer to the intended
open-file.

Signed-off-by: Ian Jackson <Ian.Jackson@eu.citrix.com>

fishdescriptor: Work if child is not running as root, but we are

We make a socket, normally in /var/run/user for the victim process,
which is therefore accessible only to the victim uid (and to root).
If we are running as root, the socket will typically not be writeable
by the user, and this will fail.  We need to chmod it.

Signed-off-by: Ian Jackson <Ian.Jackson@eu.citrix.com>

fishdescriptor: Fix a tuple bug in fd permutation

The fd permutation algorithm would fail in nontrivial cases because it
tries to update individual members of the tupes in the fdmap.  That is
not permitted in Python.  Make the tuples into lists.

Signed-off-by: Ian Jackson <Ian.Jackson@eu.citrix.com>

fishdescriptor: Fix a typo

Signed-off-by: Ian Jackson <Ian.Jackson@eu.citrix.com>

start 6.0.1~

Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>

finalise 6.0.0

Bumping version to 6 because of fishdescriptor

Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>

Merge remote-tracking branch 'mariner/fishdescriptor'

fishdescriptor: Re-finalise changelog

Signed-off-by: Ian Jackson <Ian.Jackson@eu.citrix.com>

fishdescriptor: Provide copyright notices and licence statements

Signed-off-by: Ian Jackson <Ian.Jackson@eu.citrix.com>

fishdescriptor: Minor control file improvement

Signed-off-by: Ian Jackson <Ian.Jackson@eu.citrix.com>

fishdescriptor: Fix typo in help message

Signed-off-by: Ian Jackson <Ian.Jackson@eu.citrix.com>

xbatmon-simple: Minor improvements to the .txt doc.

Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>

fishdescriptor: build system (nugatory)

Signed-off-by: Ian Jackson <Ian.Jackson@eu.citrix.com>

Revert "fishdescriptor: for revert: attempt at setuptools"

This reverts commit 15b1fbea6eaf22d148764b352f2d86b5fc7efb1c.

fishdescriptor: for revert: attempt at setuptools

This is too hard.

Signed-off-by: Ian Jackson <Ian.Jackson@eu.citrix.com>

fishdescriptor: mention in changelog

Signed-off-by: Ian Jackson <Ian.Jackson@eu.citrix.com>

fishdescriptor: fix error handling

Signed-off-by: Ian Jackson <Ian.Jackson@eu.citrix.com>

fishdescriptor: sort out debugging output

Signed-off-by: Ian Jackson <Ian.Jackson@eu.citrix.com>

fishdescriptor: bugfixes

Signed-off-by: Ian Jackson <Ian.Jackson@eu.citrix.com>

fishdescriptor: bugfixes

Signed-off-by: Ian Jackson <Ian.Jackson@eu.citrix.com>

fishdescriptor: hideous errno workaround

Signed-off-by: Ian Jackson <Ian.Jackson@eu.citrix.com>

Revert "fishdescriptor: try parse_eval_via_print"

This reverts commit 33545a21597767a50443a717b015fc9e3c8dd553.

fishdescriptor: try parse_eval_via_print

Actually this works for reading errno but not for setting it.

Signed-off-by: Ian Jackson <Ian.Jackson@eu.citrix.com>

fishdescriptor: wip utility, bugfixes

Signed-off-by: Ian Jackson <Ian.Jackson@eu.citrix.com>

fishdescriptor: wip utility, bugfixes

Signed-off-by: Ian Jackson <Ian.Jackson@eu.citrix.com>

fishdescriptor: wip utility

Signed-off-by: Ian Jackson <Ian.Jackson@eu.citrix.com>

fishdescriptor: wip utility, before rename various variables

Signed-off-by: Ian Jackson <Ian.Jackson@eu.citrix.com>

fishdescriptor: wip utility

Signed-off-by: Ian Jackson <Ian.Jackson@eu.citrix.com>

fishdescriptor: Make Donor.pid a public member

Signed-off-by: Ian Jackson <Ian.Jackson@eu.citrix.com>

fishdescriptor: before redo option parsing

Signed-off-by: Ian Jackson <Ian.Jackson@eu.citrix.com>

fishdescriptor: use lookup_type for the field list

Signed-off-by: Ian Jackson <ian.jackson@eu.citrix.com>

fishdescriptor: bugfix

Signed-off-by: Ian Jackson <ian.jackson@eu.citrix.com>

fishdescriptor: bugfixes, seems to get fds now!

Signed-off-by: Ian Jackson <ian.jackson@eu.citrix.com>

fishdescriptor: bugfixes

Signed-off-by: Ian Jackson <ian.jackson@eu.citrix.com>

fishdescriptor: bugfixes

Signed-off-by: Ian Jackson <ian.jackson@eu.citrix.com>

fishdescriptor: debugging

Signed-off-by: Ian Jackson <Ian.Jackson@eu.citrix.com>