@Kusalananda What is this?

What happens in Linux (or any other Unix) if a directory that has a mount point within it (a subdirectory) is renamed? Can you unmount using the new or old path of the mount point?

... and good morning!

Just wondering because on OpenBSD you no longer can unmount that drive if that happens, but there's a proposed patch that makes it possible.

@FaheemMitha That's the "locate the cursor" feature of iTerm on macOS.

@Kusalananda Oh. Handy, possibly.

@FaheemMitha Possibly, if one has many panes an tiny font maybe.

Though it's actually quite easy to mislay the cursor.

@Kusalananda You can unmount using the new path, and not the old

@MichaelHomer Ok. I just hadn't thought about it ever being an issue. Thanks.

I just checked what happened, neither had I

I suppose it's keyed on device/inode?

Does a directory correspond to an inode?

I sort of remember that a directory is a special file. With a fixed size.

@FaheemMitha Yes.

But no, mounting does the path, it seems

@Kusalananda Is that a standard thing across all Unix-like systems?

@FaheemMitha Directories don't have a fixed size. The file grows as you create new entries in it.

At least, two hard-linked files don't both show the same bind mount on them

@Kusalananda Oh.

I don't think that's testable with a normal mount

@FaheemMitha Some filesystems only allows directories to grow and will never truncate them, no matter if their directory entires are deleted. This would be for performance reasons.

@FaheemMitha It would be the same on all Unix filesystems.

@MichaelHomer Yes. I thought it may be possible that the old path may have been stored somewhere as a sort of handle for the mount, to use with umount, but that is obviously not the case.

I have prodded a lot and I can't figure out exactly how it's tracking the rename

My two guesses are either that it intercepts every rename call to check, which seems like a performance nightmare, or that the kernel mount table is keyed on device/inode of parent directory and name

But I don't see a way to distinguish the two and I don't want to read the source

@MichaelHomer It seems likely that the mount is tied to the device+inode

Not of the mount point though!

At least, not for bind mounts on files

I would have thought the mount was tied to the inode. What is the device here?

@FaheemMitha The parent filesystem.

@Kusalananda Oh.

@FaheemMitha Well yes. An inode on a given system is not guaranteed to be unique if it's not specified what device (filesystem) it's an inode on.

In practice almost guaranteed they are non-unique overall, given the way most filesystems allocate their inodes

Why would connecting to a DB in perl affect how gzip behaves?

@terdon The DBI code changes some form of state. I would not know what state, but it may be the contents of ARGV or some file descriptor or something.

The error may also actually be produced by the non-DBI code, but not printed for whatever reason.

@Kusalananda I doubt that. Perl's pretty verbose when run with the use warnings (as I did). But no, @StephenKitt pointed me in the right direction. I can't reproduce this with newer versions, so it looks like it was just a bug that has since been fixed.

Thanks Stephen!

@terdon you’re welcome!

@terdon @StephenKitt Found it! Writing up an answer...

@derobert Oh? And it isn't a version thing? I can't reproduce on the two other machines with newer software

It's probably a version thing, but of MySQL (or maybe dbd-mysql), not Perl.

@terdon answer posted

@terdon BTW, open has a version taking a list, which is preferable so you don't have to worry about escaping things for the shell

open (my $fileHandle, "-|", "/bin/zcat", $ARGV[0]);

@MichaelHomer I would guess it's tied to a dentry; kernel.org/doc/Documentation/filesystems/vfs.txt

@derobert AH, nice thanks.

And thank you so much for figuring that out. I've been bashing my head against the wall since last night trying to figure it out. This was part of a very complex program which included sshing to another machine to run commands and it took ages to figure out that the problem was with the SQL.

Still, another user claimed he can reproduce with the same perl version. Whatever, I'm sorted :)

@terdon There are other things that might be happening too, e.g., maybe the buffer sizes are slightly different. There are both buffers in the pipe and in perl's read; if those absorb all of the zcat output, then it won't get SIGPIPE

You'd have to compare e.g., strace output beyond that.

@derobert hmm, point

@StephenKitt I don't know if I'd call SE "general appeal". Some of the sites are pretty niche.

TeX, for example. It's pretty specific.

I agree that the model, though at first sight it looks very simple-minded and even a bit childish (points and badges) does work well, and it's effective at keeping out spammers, which is a huge problem in general.

Hi

Could I get some help with intel x86 assembly?

Probably not here

Where should I go?

Presumably SO

@JBis We're not kicking you out, it's just that the people who hang out here don't tend to know about that sort of thing.