truncate(1): Add hole-punching support

[khng]

Summary

This commit adds hole-punching support to the truncate(1) utility. If
the option -d is specified, truncate(1) performs zeroing. In case the
operation is supported by the underlying file systems of the specified files,
hole-punching is performed instead of zeroing.

Sponsored by: The FreeBSD Foundation

[khng]

Remove MACHINE-specific from Makefile.depend

[kib]

Did you considered adding some more options to truncate(1) instead?

usr.bin/fspacectl/fspacectl.c

64 ↗

(On Diff #93747)

The body of the loop clearly deserves {} around it

92 ↗

(On Diff #93747)

!do_dealloc

106–107 ↗

(On Diff #93747)

return ()

[khng]

In D31556#711650, @kib wrote:

Did you considered adding some more options to truncate(1) instead?

This sounds fine to me.

[markj]

In D31556#711650, @kib wrote:

Did you considered adding some more options to truncate(1) instead?

Or trim(8), which already performs a similar operation on device files.

[khng]

Move the functionality to truncate(1) instead.

[imp]

In D31556#711674, @markj wrote:

In D31556#711650, @kib wrote:

Did you considered adding some more options to truncate(1) instead?

Or trim(8), which already performs a similar operation on device files.

truncate is a better place for it, imho, though Linux's man pages make no mention of this. It also has no trim command. all imho.

dd also supports the BIO_DELETE operations. I've not checked to see if gnu's coreutil's dd can operate on files or not. But it seems a bit wrong to add it there.

The truncate implementation looks good. I have but one minor nit beyond noting history here.

usr.bin/truncate/truncate.1

4

Nit: No need for this 'blank line' between the copyright notices.

[kib]

usr.bin/truncate/truncate.1

38

47

Might be remove list of suffixes from all sizes, just typeset it as a meta-symbol.
Then note somewhere that that size is a number with following accepted suffixes.

I hope you see what I mean.

usr.bin/truncate/truncate.c

72

Any reason to not include do_dealloc into next line?

132

Where is this enforced?

[markj]

usr.bin/truncate/truncate.1

166

Isn't it useful to be able to omit the length and create a hole from the specified offset until EOF?

[khng]

Fix copyright notice, styles. The argument checking is rearranged to make it more obvious. Remove duplication of size/offset/length suffix documentation and group them together.

[khng]

usr.bin/truncate/truncate.1

166

I thought it could be easily misused to accidentally turn a file into a completely sparse when you forgot to type the preceding dash for instance.

usr.bin/truncate/truncate.c

132

Around line 145, but I reordered it to be in the same line below this comment block.

[khng]

• Disallow users calling with length == 0.
• Resume from signal interruption.

[kib]

usr.bin/truncate/truncate.c

207

Why? A signal most likely comes from user trying to stop the operation. Instead, the utility would try to restart.

In fact I do not think that EINTR is possible, because you do not install any signal handlers, but still I believe that this is simply wrong to do.

[khng]

Restore the old one.

[kib]

usr.bin/truncate/truncate.c

205

The ';' should be on separate line.

209

I do not understand this at all.

[khng]

I don't even think the loop for fspacectl call is required.

[kib]

usr.bin/truncate/truncate.c

210

I think it would be cleaner if you added 'do_truncate' control var, set to 1 by default. This 'else' is quite confusing, I must admit.

[khng]

Add do_truncate.

[kib]

usr.bin/truncate/truncate.c

204

You do not need 'else'

[jlduran]

I believe this commit has broken the cross-builds on GitHub.

[jlduran]

usr.bin/truncate/truncate.c

203

At a quick glance, it looks like SPACECTL_DEALLOC is only defined if __BSD_VISIBLE, hence throwing:

/Users/runner/work/freebsd-src/freebsd-src/usr.bin/truncate/truncate.c:206:22: error: use of undeclared identifier 'SPACECTL_DEALLOC'
                        r = fspacectl(fd, SPACECTL_DEALLOC, &sr, 0, &sr);

[khng]

In D31556#712959, @jlduran_gmail.com wrote:

I believe this commit has broken the cross-builds on GitHub.

A basic background for this: usr.bin/truncate is currently a listed target in _basic_bootstrap_tools.