dss.git
7 years agoUse semaphore locking to avoid starting dss multiple times.
Andre Noll [Sat, 13 Nov 2010 19:31:10 +0000 (20:31 +0100)]
Use semaphore locking to avoid starting dss multiple times.

It's trickier than one might expect but it is hopefully also much
better than any pidfile-based approach.

This patch adds ipc.c and ipc.h containing the public lock_dss()
function which acquires a semaphore-based lock whose key is based
on the hash of the resolved path name of the dss config file. This
allows different instances of dss to coexist.

All semaphore operations are called with both the SEM_UNDO and the
IPC_NOWAIT flag. SEM_UNDO guarantees that no stale lock remains after
dss was killed by SIGKIlL while IPC_NOWAIT makes the call to lock_dss()
fail if another process is already holding the lock.

The prune/create/run commands simply take the lock at startup and
exit if it could not be acquired.

The underlying semaphore set contains two semaphores. This is necessary
to implement the --kill subcommand which is done in a subsequent patch.

7 years agoIntroduce get_config_file_name().
Andre Noll [Sat, 13 Nov 2010 19:21:04 +0000 (20:21 +0100)]
Introduce get_config_file_name().

ATM, the config file name is computed in parse_config(). However, for the ipc
stuff we'll need that information as well, so move the computation to a separate
helper function().

7 years agodoc: Add a second example config file.
Andre Noll [Mon, 30 Apr 2012 19:30:35 +0000 (21:30 +0200)]
doc: Add a second example config file.

This commit provides another example for the INSTALL file. It
illustrates the use of some more sophisticated config options of dss.

Many thanks to Ivo Welch who suggested this change.

9 years agoNEWS update.
Andre Noll [Thu, 31 Mar 2011 21:54:37 +0000 (23:54 +0200)]
NEWS update.

9 years agoNEWS update
Andre Noll [Mon, 7 Feb 2011 16:58:58 +0000 (17:58 +0100)]
NEWS update

9 years agoMerge branch 't/recycling'
Andre Noll [Mon, 7 Feb 2011 16:57:40 +0000 (17:57 +0100)]
Merge branch 't/recycling'

9 years agoMerge branch 't/logfile_only_for_daemon'
Andre Noll [Mon, 7 Feb 2011 16:56:36 +0000 (17:56 +0100)]
Merge branch 't/logfile_only_for_daemon'

9 years agoMinor documentation improvements.
Andre Noll [Sun, 14 Nov 2010 11:17:38 +0000 (12:17 +0100)]
Minor documentation improvements.

9 years agoInclude the NEWS file in the web page.
Andre Noll [Sat, 13 Nov 2010 20:33:03 +0000 (21:33 +0100)]
Include the NEWS file in the web page.

Reformat the release headers in the NEWS file to make the html variant
produced by grutatxt nicer.

9 years agoINSTALL: Fix typo in example.
Andre Noll [Wed, 10 Nov 2010 14:30:20 +0000 (15:30 +0100)]
INSTALL: Fix typo in example.

The dss command line option is called --ls rather than --list. Thanks to Henry
Vogt who spotted this documentation bug.

9 years agoMention that inode monitoring works also for ext2 and ext4.
Andre Noll [Tue, 9 Nov 2010 20:58:04 +0000 (21:58 +0100)]
Mention that inode monitoring works also for ext2 and ext4.

9 years agoChange default program for all hooks from /bin/true to true.
Andre Noll [Tue, 9 Nov 2010 17:43:49 +0000 (18:43 +0100)]
Change default program for all hooks from /bin/true to true.

At least on Mac OS, true is /usr/bin/true, not /bin/true. So the old default
/bin/true causes all hooks to fail on these systems. Since we execute external
programs via execvp() anyway, there is no need to hardcode the path.

Thanks to Henry Vogt for pointing out this flaw.

9 years agodss-0.1.4. v0.1.4
Andre Noll [Mon, 8 Nov 2010 09:52:18 +0000 (10:52 +0100)]
dss-0.1.4.

9 years agoAdd NEWS file.
Andre Noll [Mon, 8 Nov 2010 09:48:10 +0000 (10:48 +0100)]
Add NEWS file.

It contains the release date and the some notes about what has
changed. The text is identical to what is listed as "Changes" on
the freshmeat pages.

9 years agoChange copyright year from 2009 to 2010.
Andre Noll [Sat, 6 Nov 2010 14:14:35 +0000 (15:14 +0100)]
Change copyright year from 2009 to 2010.

Better late than never..

9 years agoIntroduce snapshot recycling.
Andre Noll [Fri, 14 May 2010 12:43:51 +0000 (14:43 +0200)]
Introduce snapshot recycling.

When snapshotting large file systems whose contents do not change much between
snapshots, we end up removing large amounts of files just to recreate (hard
links to) most of them afterwards. This patch changes snapshot creation so that
outdated, redundant and orphaned snapshots are reused as the basis for new
snapshots. Only if no existing snapshot is suitable for recycling, a new one is
created.

9 years agoUse logfile only for --run --daemon.
Andre Noll [Sat, 17 Apr 2010 12:43:00 +0000 (14:43 +0200)]
Use logfile only for --run --daemon.

In all other cases it does not really make sense, so it is
better to write log output to stderr in these cases, even if
--logfile was given.

9 years agoDump config and internal state on SIGHUP.
Andre Noll [Mon, 5 Apr 2010 22:32:47 +0000 (00:32 +0200)]
Dump config and internal state on SIGHUP.

This patch changes the signal handling of SIGHUP so information about
the internal state and the config are written to the log file.

Both the old and the reloaded configuration are written to make it
easier to investigate what has changed.

Internal state output consist of the value of all global variables,
inclusing hook state, PIDs, free disk space and path information.

The patch also adds a one-line description of each possible hook state,
and this description is also printed on SIGHUP.

9 years agoMerge branch 't/unconditional_hooks'
Andre Noll [Wed, 12 May 2010 09:04:36 +0000 (11:04 +0200)]
Merge branch 't/unconditional_hooks'

9 years agoUnify sending of signals.
Andre Noll [Wed, 12 May 2010 09:00:55 +0000 (11:00 +0200)]
Unify sending of signals.

This patch introduces dss_kill(), a wrapper for kill(2) which
prints a nice log message and checks the return value of the
the underlying call to kill().

9 years agoSkip unnecessary check whether next snapshot is due.
Andre Noll [Wed, 12 May 2010 08:23:18 +0000 (10:23 +0200)]
Skip unnecessary check whether next snapshot is due.

If the status of the create hook is PRE_SUCCESS, we _know_ that
the next snapshot is due.

9 years agoDon't check whether hooks are given.
Andre Noll [Mon, 5 Apr 2010 19:39:12 +0000 (21:39 +0200)]
Don't check whether hooks are given.

Just use the default /bin/true if no hook was given.

9 years agoMake all hooks default to /bin/true.
Andre Noll [Mon, 5 Apr 2010 19:58:30 +0000 (21:58 +0200)]
Make all hooks default to /bin/true.

This way, we can always execute the hook which should simplify
the code a bit.

9 years agoRemove a noisy debug log message.
Andre Noll [Fri, 16 Apr 2010 11:46:18 +0000 (13:46 +0200)]
Remove a noisy debug log message.

This message was not very useful at all and spammed the log quite a bit.

9 years agoInvalidate create_pid if create process has died.
Andre Noll [Fri, 16 Apr 2010 11:39:28 +0000 (13:39 +0200)]
Invalidate create_pid if create process has died.

We're checking create_pid against zero at several places, for example before
sending a signal to the create process. So set create_pid is zero in
handle_sigchld() if the create process just died.

10 years agoLog a message when sending signals to the create process.
Andre Noll [Mon, 5 Apr 2010 18:37:14 +0000 (20:37 +0200)]
Log a message when sending signals to the create process.

10 years agopre_remove_hook: Really pass full path of the snapshot.
Andre Noll [Mon, 5 Apr 2010 18:35:47 +0000 (20:35 +0200)]
pre_remove_hook: Really pass full path of the snapshot.

We computed and logged the correct command, but executed the
wrong one.

10 years agoReuse old rsync argv if rsync has to be restarted.
Andre Noll [Thu, 25 Mar 2010 13:49:47 +0000 (14:49 +0100)]
Reuse old rsync argv if rsync has to be restarted.

If rsync must be restarted due to an exit code of 12 or 13,
create_rsync_argv() was called even if the old rsync_argv should
be reused in this case. This (correctly) triggers the assertion

assert(!name_of_reference_snapshot);

in create_rsync_argv(). Fix this by not calling create_rsync_argv()
if there is a reference snapshot.

10 years agoDon't invalidate current_snapshot_time during compute_next_snapshot_time().
Andre Noll [Thu, 25 Mar 2010 13:43:20 +0000 (14:43 +0100)]
Don't invalidate current_snapshot_time during compute_next_snapshot_time().

This bug could lead to snapshots with creation time Jan 1 1970.

10 years agoAvoid busy loop on rsync exit status 12 or 13.
Andre Noll [Fri, 12 Mar 2010 14:47:07 +0000 (15:47 +0100)]
Avoid busy loop on rsync exit status 12 or 13.

Although we set the next snapshot time to now + 60 seconds in case
rsync exits with exit status 12 or 13, we miss to check this time
barrier in case the snapshot creation status is HS_NEEDS_RESTART.

Fix this by adding an additional check in the switch() statement
of the select loop. As this change would trigger the assertion

snapshot_creation_status == HS_READY

in next_snapshot_is_due(), remove this assertion.

10 years agoFix create mode in case no post-create hook is given.
Andre Noll [Fri, 12 Mar 2010 13:32:17 +0000 (14:32 +0100)]
Fix create mode in case no post-create hook is given.

Without the patch, dss would hang forever waiting for the post-create
hook to die.

10 years agoIntroduce --no-resume.
Andre Noll [Mon, 1 Feb 2010 09:21:35 +0000 (10:21 +0100)]
Introduce --no-resume.

If the dss daemon (or the rsync process) is killed while a snaphot
is being created, e.g. because of a server shutdown, the latest
snapshot remains incomplete until it is removed by the usual shapshot
pruning mechanism.

This patch changes the snapshot creation behaviour if the
most recently created snapshot happens to be incomplete and the
new --no-resume option is not given. In this case the directory
of the incomplete snapshot is reused as the destination directory
for the the new snapshot.

This change saves disk space and reduces the snapshot creation time,
depending of course on how far the previous rsync process got before
it was interrupted.

10 years agoFix two typos in comments.
Andre Noll [Thu, 3 Sep 2009 19:29:29 +0000 (21:29 +0200)]
Fix two typos in comments.

10 years agoProperly invalidate create_pid also for the post-create hook.
Andre Noll [Fri, 28 Aug 2009 13:23:57 +0000 (15:23 +0200)]
Properly invalidate create_pid also for the post-create hook.

If the process associated with the create_pid dies, handle_sigchld()
investigates snapshot_creation_status to tell whether the pre-create
hook, the rsync process or the post-create hook has died.

In the first two cases, handle_pre_create_hook_exit() and
handle_rsync_exit() are called, respectively. Both functions correctly
invalidate create_pid (by resetting it to zero).

However, the post-create hook handling code misses to reset
create_pid. This causes dss to send SIGTERM to this pid on exit,
which might be fatal as the pid might have been reassigned to some
unrelated process in the meanwhile.

Fix this bug by moving the invalidation of create_pid to the end of
the "if (pid == create_pid)" clause, which even saves a line of code.

Many thanks to Sebastian Stark who pointed out that bug.

10 years agoDump the dss configuration on startup and on SIGHUP.
Andre Noll [Fri, 28 Aug 2009 11:42:51 +0000 (13:42 +0200)]
Dump the dss configuration on startup and on SIGHUP.

This makes it possible to look at the current dss configuration
(which may differ from what is contained in the config file).

This feature is handy for the users of dss and it helps to debug the
dss code.

10 years agoImprove error diagnostics.
Andre Noll [Fri, 28 Aug 2009 09:28:57 +0000 (11:28 +0200)]
Improve error diagnostics.

When parsing the command line options we must not error out if a
required option was not given because that option might be specified
in the config file. Therefore we have to call cmdline_parser_ext()
with params->check_required = 0.

However, if --config-file is not given and the default config file
(~/.dssrc) does not exist, we end up with no check for required
options at all.

In particular, if the required --dest-dir option is not given,
conf.dest_dir is NULL and we call chdir(NULL) which returns EBADADRESS
at least on Linux. This causes dss to print the error message

Aug 28 11:35:07 main: Bad address

which is not really helpful. Fix this shortcoming by calling
cmdline_parser_ext() _again_ if no config file was read by
parse_config_file(). This second call uses params->check_required =
1, so that a proper error message is printed if any required options
are missing.

In the above example, the output changes to

./dss: '--source-dir' option required
./dss: '--dest-dir' option required

which is much better.

10 years agoImprove next_snapshot_is_due().
Andre Noll [Fri, 28 Aug 2009 09:12:30 +0000 (11:12 +0200)]
Improve next_snapshot_is_due().

Currently it's a bit weird how next_snapshot_is_due() decides whether
the next snapshot time has to be (re-)computed:

On startup, next_snapshot_time is zero as it is declared
static.

next_snapshot_is_due() checks whether next_snapshot_time is
greater than the current time. If yes, then next_snapshot_time
needs not be updated and the function returns false.

Otherwise (e.g. if it is called for the first time),
next_snapshot_time is recomputed, next_snapshot_is_due()
checks again if it is greater than the current time and
returns false if it is, true otherwise.

Consequently, dss computes the next snapshot time twice per snapshot.
Moreover, it compares next_snapshot_time twice against the current time
where one comparison would suffice. The code is thus less efficient
and harder to understand than necessary. This patch addresses both
issues. It introduces the two trivial helper functions

next_snapshot_time_is_valid() and invalidate_next_snapshot_time().

The former function simply tests next_snapshot_time against zero. It
is called from next_snapshot_is_due(). If it returns false, the new
compute_next_snapshot_time() is called (which makes next_snapshot_time
valid). Next, the usual comparison against the current time is
performed.

invalidate_next_snapshot_time() sets next_snapshot_time to zero. It
is called from pre_create_hook() and from handle_sighup(), the latter
call is necessary because changes in the config file might lead to
different snapshot creation times.

10 years agoSimplify computation of next snapshot time.
Andre Noll [Thu, 27 Aug 2009 14:27:29 +0000 (16:27 +0200)]
Simplify computation of next snapshot time.

Using an int64_t rather than a struct timeval for the next snapshot time
makes the code simpler and more readable as we don't have to use the
tv_xxx() functions to perform manipulations.

10 years agoFix off-by-one bug in find_outdated_snapshot().
Andre Noll [Thu, 27 Aug 2009 12:55:02 +0000 (14:55 +0200)]
Fix off-by-one bug in find_outdated_snapshot().

The man page sayeth:

"dss removes any snapshots older than n times u",

where n is the number of unit intervals and u is the duration of
a unit interval. As intervals count from zero, this means that a
snapshot should be considered outdated if its interval number
is greater _or equal_ than n.

However, the current code only removes snapshots in intervals
strictly greater than n. Fix this bug and clarify the documentation.

10 years agoMake snapshot creation trump snapshot removal.
Andre Noll [Wed, 26 Aug 2009 11:24:12 +0000 (13:24 +0200)]
Make snapshot creation trump snapshot removal.

This patch changes the behaviour of dss in case the following three
conditions are all met:

(1) There is at least one snapshot which could be deleted
    (orphaned, redundant, outdated).
(2) Disk space is not low.
(3) A new snapshot is due.

A common case where these conditions are fulfilled is when dss is
started after it was not running for some time, for example due to
a server crash on Friday evening...

In this situation the old code created a new snapshot only after all
orphaned/redundant/outdated have been removed, which can take hours
and is not what one usually wants to happen in the above mentioned
scenario. Instead, it is desirable to create a new snapshot ASAP,
and only after this snapshot has been created, the removal of old
snapshots should take place.

This patch implements this behaviour and goes even one step further:
If disk space is not low and a new snapshot is due or being created,
dss won't trigger snapshot removal any more.

Another positive side effect of this change is that snapshot creation
times become more stable since the rsync process will only be
interrupted by a rm process if disk space is low.

10 years agoMove check for low disk space to try_to_free_disk_space().
Andre Noll [Mon, 24 Aug 2009 08:34:46 +0000 (10:34 +0200)]
Move check for low disk space to try_to_free_disk_space().

The value is only used there, so move the check to this function.

10 years agodss-0.1.3. v0.1.3
Andre Noll [Mon, 6 Jul 2009 13:55:13 +0000 (15:55 +0200)]
dss-0.1.3.

10 years agoChange copyright year from 2008 to 2009.
Andre Noll [Fri, 3 Jul 2009 15:03:39 +0000 (17:03 +0200)]
Change copyright year from 2008 to 2009.

A bit late, but better late than never.

10 years agoTry harder to avoid removing the reference snapshot.
Andre Noll [Mon, 29 Jun 2009 09:36:25 +0000 (11:36 +0200)]
Try harder to avoid removing the reference snapshot.

With the old code, it was possible that dss decided to remove the
snapshot which is currently being used as the hardlink directory for
the current rsync process. This patch changes the behaviour so that
reference snapshots are never removed.

The downside of this approach is of course that it is now easier to
fill up the disk..

10 years agoDon't remove the last snapshot unless disk space is low.
Andre Noll [Mon, 8 Jun 2009 06:15:18 +0000 (08:15 +0200)]
Don't remove the last snapshot unless disk space is low.

10 years agoDon't busy-loop if pre-create hook returns non-zero.
Andre Noll [Mon, 8 Jun 2009 05:57:30 +0000 (07:57 +0200)]
Don't busy-loop if pre-create hook returns non-zero.

Sleep at least one minute before retrying.

10 years agoFix removal of orphaned snapshots.
Andre Noll [Sun, 7 Jun 2009 19:44:50 +0000 (21:44 +0200)]
Fix removal of orphaned snapshots.

These were detected but not removed due to the goto jumping to the
wrong label.

10 years agoImprove documentation of the remove hooks.
Andre Noll [Sun, 7 Jun 2009 19:16:41 +0000 (21:16 +0200)]
Improve documentation of the remove hooks.

10 years agoClean up snapshot removal logic.
Andre Noll [Thu, 4 Jun 2009 11:47:35 +0000 (13:47 +0200)]
Clean up snapshot removal logic.

Replace the remove_xxx_snapshot() functions by find_xxx_snapshot()
and start the actual removal in the caller. This simplifies the
code a bit and makes it much more readable. It also simplifies the
--dry-run handling for com_prune().

10 years agoRemove orphaned snapshots first if disk space is low.
Andre Noll [Fri, 29 May 2009 20:09:06 +0000 (22:09 +0200)]
Remove orphaned snapshots first if disk space is low.

If the dss process gets killed, an orphaned snapshot might result.
Detect this case and prefer to remove such orphaned snapshots before
resorting to remove the oldest snapshot.

10 years agoFix some typos.
Andre Noll [Fri, 29 May 2009 20:03:19 +0000 (22:03 +0200)]
Fix some typos.

10 years agoRestart rsync also on exit value 12.
Andre Noll [Fri, 29 May 2009 11:49:13 +0000 (13:49 +0200)]
Restart rsync also on exit value 12.

An exit value of 12 means "Error in rsync protocol data stream",
and it happens regularly for rsync-3 without apparent reason. So
restart the rsync process also in this case.

10 years agoFix return value of remove_oldest_snapshot().
Andre Noll [Tue, 5 May 2009 08:53:43 +0000 (10:53 +0200)]
Fix return value of remove_oldest_snapshot().

We _must_ return non-zero if the remove was initiated. The old code
would exit with

try_to_free_disk_space: uhuhu: not enough disk space for a single snapshot

even though there are plenty of snapshots available that can be deleted.

10 years agoFix rm-hooks in case no post-remove hook was specified.
Andre Noll [Mon, 27 Apr 2009 18:19:11 +0000 (20:19 +0200)]
Fix rm-hooks in case no post-remove hook was specified.

Without the patch the following assertion was triggered upon
removal of the second snapshot:

dss: dss.c:229: pre_remove_hook: Assertion `!snapshot_currently_being_removed' failed.

Thanks to Sebastian Stark who pointed out the issue.

11 years agoImplement rm-hooks.
Andre Noll [Wed, 8 Apr 2009 13:23:01 +0000 (15:23 +0200)]
Implement rm-hooks.

This adds calls for the pre-remove and the post-remove hooks, similar to
the pre-create/post-create hooks. If the pre-remove hook fails, snapshot
deletion is deferred until the hook succeeds.

11 years agoRename hook_status enum.
Andre Noll [Tue, 7 Apr 2009 14:50:32 +0000 (16:50 +0200)]
Rename hook_status enum.

We'd like to use only one set of states for both creation/removal.

11 years agoMerge branch 'master' into rm_hook
Andre Noll [Tue, 7 Apr 2009 14:17:01 +0000 (16:17 +0200)]
Merge branch 'master' into rm_hook

11 years agocom_create: Abort if pre-create hook fails.
Andre Noll [Tue, 7 Apr 2009 14:15:01 +0000 (16:15 +0200)]
com_create: Abort if pre-create hook fails.

handle_pre_create_hook_exit() returns zero if the hook failed.

11 years agoAdd pre-remove and post-remove hooks to dss.ggo.
Andre Noll [Tue, 7 Apr 2009 13:23:34 +0000 (15:23 +0200)]
Add pre-remove and post-remove hooks to dss.ggo.

These are not yet used in the code.

11 years agoRename rm_pid to remove_pid.
Andre Noll [Tue, 7 Apr 2009 13:09:59 +0000 (15:09 +0200)]
Rename rm_pid to remove_pid.

This variable will contain the pid of rm-hooks or the rm process.

11 years agoMerge branch 'master' into rm_hook
Andre Noll [Tue, 7 Apr 2009 12:18:14 +0000 (14:18 +0200)]
Merge branch 'master' into rm_hook

11 years agoMerge commit 'meins/master'
Andre Noll [Mon, 30 Mar 2009 10:50:35 +0000 (12:50 +0200)]
Merge commit 'meins/master'

11 years agoFix the pre-create hook.
Andre Noll [Mon, 30 Mar 2009 10:49:03 +0000 (12:49 +0200)]
Fix the pre-create hook.

Returning non-zero from the pre-create hook caused dss to exit with
"unexpected exit code" rather than waiting until the hook returns
zero.

Fix this bug and also reduce the verbostity of the log messages caused
by executing the pre-create hook: It should be enough to tell the user
only once per hour that no more snapshots are going to be created.

11 years agoClarify usage of --rsync-option.
Andre Noll [Sun, 22 Mar 2009 02:42:07 +0000 (03:42 +0100)]
Clarify usage of --rsync-option.

11 years agoMerge commit 'fml/master'
Andre Noll [Mon, 16 Mar 2009 21:05:32 +0000 (22:05 +0100)]
Merge commit 'fml/master'

11 years agoBump version number to 0.1.2.
Andre Noll [Mon, 16 Mar 2009 16:30:01 +0000 (17:30 +0100)]
Bump version number to 0.1.2.

Oops, this should have been done just before tagging the 0.1.2 release..

11 years agoIntroduce the snapshot removal status.
Andre Noll [Thu, 5 Mar 2009 09:38:39 +0000 (10:38 +0100)]
Introduce the snapshot removal status.

It is not yet used, but we need something like that for the remove-hooks.

11 years agoTrivial typo fix.
Andre Noll [Mon, 16 Mar 2009 16:19:28 +0000 (17:19 +0100)]
Trivial typo fix.

11 years agoUse only one global variable for snapshot creation pids.
Andre Noll [Mon, 16 Mar 2009 15:57:56 +0000 (16:57 +0100)]
Use only one global variable for snapshot creation pids.

There's no need to have pre_create_hook_pid, rsync_pid and
post_create_hook_pid because only one of them can be running at
any point in time. We can always tell which it is by examining the
snapshot_creation_status.

So replace these three variables by the single create_pid variable.

Besides of killing two global variables, this change also fixes a
real bug: If the dss process catches SIGINT or SIGTERM, the old code
would only kill a running rsync process but not the pre-create or
post-create hook. However, the new code kills whatever create process
is currently running, which is the right thing to do.

11 years agoMerge commit 'meins/master'
Andre Noll [Fri, 13 Mar 2009 17:06:04 +0000 (18:06 +0100)]
Merge commit 'meins/master'

11 years agoMerge commit 'fml/master'
Andre Noll [Fri, 13 Mar 2009 08:24:21 +0000 (09:24 +0100)]
Merge commit 'fml/master'

11 years agoInclude the reason why a snapshot gets removed in the log message. v0.1.2
Andre Noll [Thu, 5 Mar 2009 09:16:10 +0000 (10:16 +0100)]
Include the reason why a snapshot gets removed in the log message.

11 years agoNever remove the snapshot that is currently being created.
Andre Noll [Thu, 5 Mar 2009 09:13:12 +0000 (10:13 +0100)]
Never remove the snapshot that is currently being created.

Thanks to Sebastian Stark for pointing out this bug.

11 years agoMerge commit 'fml/master'
Andre Noll [Wed, 11 Feb 2009 22:51:44 +0000 (23:51 +0100)]
Merge commit 'fml/master'

11 years agoMerge commit 'stark/master'
Andre Noll [Mon, 2 Feb 2009 18:23:33 +0000 (19:23 +0100)]
Merge commit 'stark/master'

11 years agoInvalidate all rsync options on SIGHUP.
Andre Noll [Mon, 2 Feb 2009 18:23:20 +0000 (19:23 +0100)]
Invalidate all rsync options on SIGHUP.

gengetopt's params->override option does not do what its name suggests,
see

http://www.opensubscriber.com/message/help-gengetopt@gnu.org/10980633.html

or
http://www.mail-archive.com/help-gengetopt@gnu.org/msg00507.html

11 years agofix post_create_hook: really pass full path to last complete snapshot
Sebastian Stark [Sat, 31 Jan 2009 09:04:28 +0000 (10:04 +0100)]
fix post_create_hook: really pass full path to last complete snapshot

11 years agoMerge commit 'fml/master'
Andre Noll [Tue, 23 Dec 2008 22:14:34 +0000 (23:14 +0100)]
Merge commit 'fml/master'

11 years agoAdd some comments to compute_next_snapshot_time().
Andre Noll [Thu, 11 Dec 2008 13:07:49 +0000 (14:07 +0100)]
Add some comments to compute_next_snapshot_time().

Also, move the call to gettimeofday() down to where it is needed.

11 years agoMerge commit 'sio/master'
Andre Noll [Mon, 8 Dec 2008 16:21:51 +0000 (17:21 +0100)]
Merge commit 'sio/master'

11 years agoFix check when to use rsync locally.
Andre Noll [Mon, 8 Dec 2008 16:17:50 +0000 (17:17 +0100)]
Fix check when to use rsync locally.

We can do this if (a) remote_host_arg is "localhost" and (b)
remote_user_arg is the same as logname. The old code only
looked at the logname and thus tried to use rsync locally even
if a remote_host_arg was specified.

11 years agoFix a typo in README.
Andre Noll [Mon, 1 Dec 2008 17:38:49 +0000 (18:38 +0100)]
Fix a typo in README.

11 years agoINSTALL: Mention "make man".
Andre Noll [Sat, 22 Nov 2008 22:41:59 +0000 (23:41 +0100)]
INSTALL: Mention "make man".

11 years agoClarify the --keep-redundant option.
Andre Noll [Sun, 16 Nov 2008 12:19:32 +0000 (13:19 +0100)]
Clarify the --keep-redundant option.

dss cares about the number of free inodes, not the inode ratio.

11 years agodss-0.1.1. v0.1.1
Andre Noll [Wed, 12 Nov 2008 16:52:20 +0000 (17:52 +0100)]
dss-0.1.1.

11 years agoPrevent busy loops on rsync exit code 13.
Andre Noll [Thu, 6 Nov 2008 09:32:01 +0000 (10:32 +0100)]
Prevent busy loops on rsync exit code 13.

We restart the rsync process in case it returned with exit code 13
which unfortunately happens for some unknown reasons even with a
valid configuration.

This may lead to a busy loop, so wait at least one minute before
restarting rsync.

11 years agoIgnore any snapshot directory with creation time > completion time.
Andre Noll [Thu, 6 Nov 2008 08:52:28 +0000 (09:52 +0100)]
Ignore any snapshot directory with creation time > completion time.

Without this fix, dss would abort du to an assertion in dss.c:

dss.c:171: compute_next_snapshot_time: Assertion `x >= 0' failed.

Thanks to Sebastian Stark for pointing out this flaw.

11 years agouse C99-compliant format strings for consistency reasons
Sebastian Stark [Wed, 22 Oct 2008 15:14:30 +0000 (17:14 +0200)]
use C99-compliant format strings for consistency reasons

11 years agoopen /dev/null for reading AND writing when executing rsync.
Sebastian Stark [Wed, 22 Oct 2008 12:59:11 +0000 (14:59 +0200)]
open /dev/null for reading AND writing when executing 

This is needed for child processes to be able to write to fd 2 without failing.

For example, rsync will not be able to write an error message because of "Bad
file descriptor" which in turn leads to rsync exiting with meaningless exit
code 13 ("Errors with program diagnostics"), masking the actual error and exit
code.

The fact that rsync uses exit code 13 in that case makes this bug particularly
painful since 13 is interpreted by dss as a temporary rsync error that can be
"fixed" by simply restarting rsync. This can lead to an infinite loop,
obviously.

11 years agoshow human readable snapshot creation duration when listing snapshots.
Sebastian Stark [Wed, 22 Oct 2008 11:31:42 +0000 (13:31 +0200)]
show human readable snapshot creation duration when listing snapshots.

The PRId64 macro is defined in the C99 standard. See

  http://www.opengroup.org/onlinepubs/009695399/basedefs/inttypes.h.html

11 years agoAdd dss.1 and dss.1.html to .gitignore.
Andre Noll [Mon, 7 Jul 2008 14:29:56 +0000 (16:29 +0200)]
Add dss.1 and dss.1.html to .gitignore.

11 years agoRestart the rsync process if it returned with exit code 13.
Andre Noll [Mon, 7 Jul 2008 14:29:38 +0000 (16:29 +0200)]
Restart the rsync process if it returned with exit code 13.

Unfortunately this happens frequently for no apparent reason.

11 years agoFix the exit hook.
Andre Noll [Wed, 21 May 2008 14:36:13 +0000 (16:36 +0200)]
Fix the exit hook.

As dss_exec_cmdline_pid() uses the space character as a separator
to split the command line, the words of the error message were
passed as separate parameters to the exit hook.

Use dss_exec() directly to avoid this flaw, i.e. to pass the full
error message as $1 to the exit hook.

11 years agoOffer a direct link to the automatically generated tarball.
Andre Noll [Fri, 16 May 2008 09:27:48 +0000 (11:27 +0200)]
Offer a direct link to the automatically generated tarball.

As suggested by Sebastian Stark.

11 years agoNAME section for manpage
Sebastian Stark [Wed, 14 May 2008 14:08:25 +0000 (16:08 +0200)]
NAME section for manpage

11 years agoadd SEE ALSO section to manpage
Sebastian Stark [Wed, 14 May 2008 14:03:31 +0000 (16:03 +0200)]
add SEE ALSO section to manpage

11 years agodss-0.1.0. v0.1.0
Andre Noll [Wed, 14 May 2008 13:16:08 +0000 (15:16 +0200)]
dss-0.1.0.

11 years agosnap.c: Add missing include to make it compile on FreeBSD.
Andre Noll [Wed, 14 May 2008 12:22:40 +0000 (14:22 +0200)]
snap.c: Add missing include to make it compile on FreeBSD.

11 years agoNetBSD Compile fixes.
Andre Noll [Wed, 14 May 2008 12:09:41 +0000 (14:09 +0200)]
NetBSD Compile fixes.

Add a missing include and the dss_isdigit macro.

11 years agoFix paths and URLs, improve INSTALL text.
Andre Noll [Tue, 13 May 2008 14:47:29 +0000 (16:47 +0200)]
Fix paths and URLs, improve INSTALL text.