Use of uninitialized value $ssh[11] in open at /usr/lib/ZnapZend/ZFS.pm line 212

[jimklimov]

In my dmesg I see a series of messages in SUBJ, grouped as a burst of 10 and another 7 half a minute later.

Current codebase has it a bit shifted by line numbers, but otherwise this area seems same as installed:

https://github.com/oetiker/znapzend/blob/27f965ff84fc496d5d2c93bd74b2a745666633fa/lib/ZnapZend/ZFS.pm#L209-L215

This is in listSnapshots, and looking at debug output and the code, I believe it is trying to parse a dataset name with colons (the only thing that apparently differs from successful logs) and splitHostDataSet fails to do it properly (can colons be treated as e.g. port separator for general-case SSH remote?) Colons in dataset names are injected by current OpenIndiana pkg handling of backup rootfs'es.

https://github.com/oetiker/znapzend/blob/27f965ff84fc496d5d2c93bd74b2a745666633fa/lib/ZnapZend/ZFS.pm#L40-L42

Afterwards per symptoms, $dataSet is undef and operations with it emit warnings (two actually, just one in dmesg though) until it fails in call to zfs program where it specifies no dataset to list:

[2022-08-24 18:11:18.11843] [29435] [debug] checking to clean up snapshots of source rpool/zones/nut/jenkins-worker/ROOT/openindiana-2022:03:15-backup-1
Use of uninitialized value $ssh[11] in join or string at /usr/lib/ZnapZend/ZFS.pm line 211.
# zfs list -H -o name -t snapshot -s creation -d 1
Use of uninitialized value $ssh[11] in open at /usr/lib/ZnapZend/ZFS.pm line 212.
cannot open '': empty component in name
[2022-08-24 18:11:18.12961] [29435] [debug] got nothing to clean in source rpool/zones/nut/jenkins-worker/ROOT/openindiana-2022:03:15-backup-1

[jimklimov]

Looking at the regex, it seems like trying to catch the port is indeed the culprit (for the local path, without any hostspec in the string)

[jimklimov]

Eh, not port, just a separator I guess. Playing at https://regex101.com/r/6wXZhu/1 this expression splits root@localhost:rpool/zones/nut/jenkins-worker/ROOT/openindiana-2022-03-15-backup-1 into two groups - root@localhost and the dataset path (note no colons in test path; regex fails if there are any).

Off-topic: Wondering now how it would pass a non-22 SSH port? Is there some option or just ~/.ssh/config should be used to define params for a "hostname"? Is that documented so then? (Just quickly wondering, did not check)

[jimklimov]

Commit 58fbfc4f50ff55160d232113a43d781b4c767e1e seems to have last substantially changed the regex as

bugfix: ':' character in snapshot name does now work

-my $splitHostDataSet     = sub { return ($_[0] =~ /^(?:([^:]+):)?([^:]+)$/); };
+my $splitHostDataSet     = sub { return ($_[0] =~ /^(?:([^:\/]+):)?([^:]+|[^:@]+\@.+)$/); };

[jimklimov]

First it seemed like simplifying this to

^(?:([^:\/]+):)?(.+)$

would help

But there is a corner case with a pool root dataset (no slashes) and tsformat with colons:

rpool@snaptime-12:34

How does one confidently tell this apart from user rpool on ssh server snaptime-12 accessing (root) dataset 34? :\

(more so if that pool also has colons in the name like 34:56?)

[jimklimov]

I think use of colon to separate $remote:$dataSet was not such a good idea :-D Can it be replaced by e.g. whitespace? Is it an internal implementation detail or something users (or third-party code) expect?

UPDATE: Silly me; it is the [[user@]host:]dataset documented for DST in znapzendzetup... But still, the earlier posted ideas remain:

^(?:([^\s\/]+)\s)?(.+)$

works to split correctly into two groups this string:

root@host rpool@snaptime-12:34:56

and even permits (syntactically) to add a port number, if we wished to handle it later on:

root@host:22 rpool@snaptime-12:34:56

and still makes an (undef, dataset) from plain rpool@snaptime-12:34:56 for local paths without a remote

It does collapse to one second group grabbing whole input string if there is a slash in first token, like "user@host:22/tcp" but:

• I have no idea how such token could be used? this is not HTTP Auth or something :)
• Dataset names AFAIK may not have spaces, so probably first group catcher can also be simplified to just match non-whitespace

[jimklimov]

Fiddling more with it, this seems better viable: https://regex101.com/r/ambWtC/1

^(?:(.+)\s)?([^\s]+)$

Parses this:

root@host -p 22 -id /tmp/key rpool@snaptime-12:34:56

into usable args for ssh: group 1 == root@host -p 22 -id /tmp/key (which is everything until last whitespace) and the dataset (group 2 which is just the last long non-whitespace token)

And it still works for just the local dataset name (then group1 = undef)

[jimklimov]

Scratch that last shot for now, has some issue to track down with specs like remotehost:pond/dataset/name :\ Also the premise of root@host -p 22 -id /tmp/key rpool@snaptime-12:34:56 in the later posts was wrong - ~/.ssh/config should pass the options, so the DST value to parse is as documented in e.g. znapzendzetup --help:

[[user@]host:]dataset

with complexity of what chars dataset may contain validly (and current regex does not cover that all), including obscene cases like say dataset being named c: (or pond/c:data) because why not, making it hard to discern from the user@host: prefix (especially when there is none and DST is local).

[jimklimov]

So the winner currently was a slight mod of the original:

 my $splitHostDataSet = sub {
-    return ($_[0] =~ /^(?:([^:\/]+):)?([^:]+|[^:@]+\@.+)$/);
+    return ($_[0] =~ /^(?:([^:\/]+):)?([^@\s]+|[^@\s]+\@[^@\s]+)$/);