Add support for parsing UCL config files to newsyslog

[allanjude]

Summary

Missing man page updates
Requires newer version of UCL than is included in base (remove hack in Makefile once base is updated, in meantime use port)
Requires much review, I am a very novice C programmer

[allanjude]

etc/newsyslog.ucl

1

This 'sentinel' lets newsyslog know that this file is in FreeBSD UCL version 0.1. If it does not see this at the top of the file, it treats it as a legacy .conf file in the original format.

This will allow the new format to keep the original filename, newsyslog.conf
It also allows for 'versioning' of the config schema, so we can easily keep backwards compatibility if we change the config syntax again in the future.

3

This is a comment, of the original line from newsyslog.conf

26

A new format for the time specification would likely be very welcome

203

Working with Vsevolod Stakhov (cebka@) to come up with a way to do opaque strip splitting (https://github.com/vstakhov/libucl/issues/36)

so this could be: flags = binary, nosignal
or something similar

usr.sbin/newsyslog/Makefile

9

Dirty hack until contrib/libucl is updated. Not in a rush for this as there are some more features to be added to libucl that would be useful for this project.

usr.sbin/newsyslog/newsyslog.c

1086

taste the first line of the file. If it isn't what we want, rewind() and continue as before

Otherwise, pass it off to the new config parser

1488

This is the map that uses the callbacks. It ended up different than I originally pictures, but I think it is still nicer than a giant strcmp() if ladder

1692

This is mostly just verbatim from the original code. It can likely be improved, but I wanted to try to avoid introducing too many new bugs at once

[cse_cem_gmail_com]

Not a super thorough review, just passed over briefly. I'm not an expert on UCL nor newsyslog.

etc/newsyslog.ucl

5

Note to other reviewers: UCL doesn't implicitly understand 'mode' is octal, we just parse this field specially later in the patch.

usr.sbin/newsyslog/newsyslog.c

1082

Using double for this is odd. Usually we just multiply by a power of 10 (or 0x10).

1085

Check the return value of fgets(3) before using line.

1091

style(9): The body of the if() statement goes on a line by itself.

1092–1095

Floating point equality is somewhat fragile. Use integers for this.

1505

You probably want to lseek(2) the fd back to the beginning — fgets(3) and other stdio functions may buffer large chunks of the file and the fd may be well past the beginning of the UCL stuff.

1511

extra space (style)

1517

extra space

1523–1525

There's lots of this, so I'm not going to mark it every time. But, in style(9) it is suggested that if/while/for braces be omitted when they are not required (with exceptions for complicated bodies). printf is not complicated.

1659

style(9): return (true);

1667–1668

Ah, I see, you force octal here. Ok.

1702–1703

these should be on the same line (style(9))

[allanjude]

Update with feedback from cem

[allanjude]

More style(9)

[allanjude]

usr.sbin/newsyslog/newsyslog.c

1082

So make 'uclver' an int, and multiply the result of strtod(3) by 10?

[cse_cem_gmail_com]

usr.sbin/newsyslog/newsyslog.c

1082

That'd work. You could also just use an integer version (e.g. 1) and strtol or whatever. (This is just versioning the config schema, right?)

[cse_cem_gmail_com]

Changes between 3218 and 3220 LGTM.

usr.sbin/newsyslog/newsyslog.c

1695–1696

These on the same line too

1737

extra space

[cse_cem_gmail_com]

Looks pretty good!

usr.sbin/newsyslog/newsyslog.c

212

Don't need the backslash. typedef isn't part of the preprocessor.

213

It looks like the 3rd parameter here should be struct conf_entry * based on later functions.

1086–1090

You can drop the lineptr variable entirely. You would use strtod like:

uclver = strtod(&line[5], NULL);

1490–1501

I think you can drop the (conf_helper_func) casts if you change the function typedef from void * to struct conf_entry *.

1554

The usual pattern for this is:

if ((working->flags & CE_NOSIGNAL) != 0) {

1609–1610

You can drop result entirely.

target->uid = ucl_object_toint(obj);

1615

Use atol or strtol. uid_t is unsigned and yes, somebody has >2 billion uids (or at least uses the high numbers).

1616

target->uid = atol(val); or similar

1627–1628

target->uid = pwd->pw_uid;

1644

extra parens, you don't need them around function calls

1652

Probably atol here too.

1664

Probably: if (sscanf(...) != 1) {.

sscanf may return EOF (-1) as well as (0) depending on what went wrong. Anything other than 1 conversion is failure for us.

1678

(same re: sscanf(...) != 1.)

1715

No 'n' needed for xz, I think?

1796–1797

same line

1799–1800

also same line

1803–1804

same

1808–1814

braces can be dropped here.

1828–1829

same

1889

You can drop the comment, the pattern is well understood.

Alternatively, you can include sys/cdefs.h (if it isn't already) and use __unused like:

conf_set_unknown(..., struct conf_entry *target __unused)

3129

extra space

3142

If you want to reduce indentation, you can do:

if (strcasecmp(...) != 0)
  continue;

And then the rest of the loop body can be unindented.

3144–3146

Can drop NULL check entirely. Just don't put a NULL callback in your table :-).

3150

if (!found && ...)

3156

(ret)

[allanjude]

Update with more feedback from cem

[allanjude]

usr.sbin/newsyslog/newsyslog.c

213

I originally had this code in libucl, so it was generalized to work in places other than just newsyslog. I just dumped it into newsyslog for now.

1086–1090

dropped

1490–1501

see response to https://reviews.freebsd.org/D1548#inline-9104

1505

I moved the rewind(cf) that was done only in the case of a non-ucl config earlier, so it should resolve this

1554

Not my code, but since it is inside my function, fixing it

1609–1610

ok

1615

this case doesn't need to be here, it is left over. If the user entered a number, it will by type UCL_INT, unless they put a decimal in it or something silly, in which case they deserve an error anyway.

1644

those were existing, but removed

1715

It is done so it allows the user to do 'xzip' as well. same reason bzip is cut to 4 chars, so bzip or bzip2 both are accepted.

3144–3146

This is there on purpose. We have a NULL callback for the 'file' key, because it is handled elsewhere, and we don't want that key to trip the 'default' callback. See line #1486

This should be documented better, my goal is to make something generic to speed up the process of converting other utilities to UCL.

[cse_cem_gmail_com]

LGTM.

usr.sbin/newsyslog/newsyslog.c

1715

Ah, I got bzip/bzip2 but hadn't seen xzip before. Seems reasonable to me.

3144–3146

Ok.

3150

Typo — I think you want !found?

[allanjude]

usr.sbin/newsyslog/newsyslog.c

1715

This question made me think about it more, and improve it further
use strncasecmp(val, ..., 2) in all cases:

so gzip accepts: gz gzip
bzip2: bz bz2 bzip bzip2
xz: xz xzip yes
none: no none

(added yes to xz since we accept no)

3150

good catch, thank you

[allanjude]

Further expand the accepted input for the compression parameter

[cse_cem_gmail_com]

LGTM.

usr.sbin/newsyslog/newsyslog.c

1707

Maybe just strcasecmp() for "yes"?

[allanjude]

Feedback from cem

[mat]

Mmm, newsyslog.conf has:

<include> /etc/newsyslog.conf.d/*
<include> /usr/local/etc/newsyslog.conf.d/*

This doesn't seem to have been included here.

[allanjude]

In D1548#23, @mat wrote:

Mmm, newsyslog.conf has:

<include> /etc/newsyslog.conf.d/*
<include> /usr/local/etc/newsyslog.conf.d/*

This doesn't seem to have been included here.

Correct, I've not added that yet. I'll get that done shortly.

[eadler]

overall reasonable, just some quick comments.

usr.sbin/newsyslog/newsyslog.c

1082

style(9) prefers not having init inline

1093

theoretical overflow not handled, but is likely fine

1520

TODO comment :)

1543

if (!success) {} ?

1785

if ul == 0, check errno?

[allanjude]

Address Feedback

[allanjude]

Update based on feedback

[emaste]

What's next for this patch?