[9fans] Fossil thinks my free space is a wee bit more than it is

[9fans] Fossil thinks my free space is a wee bit more than it is

[Anthony Sorace]

While untarring a big-ish file to a small-ish hard drive, I was
running df on my fossil console to see if I was going to run out of
space.  The output from these successive runs is... not correct:

prompt: fsys main df
	main: 1,644,601,344 used + 2,588,672 free = 1,647,190,016 (99% used)
prompt: fsys main df
	main: 1,652,891,648 used + 35,184,366,387,200 free = 1,647,190,016 (100% used)
prompt: fsys main df
	main: 1,666,473,984 used + 35,184,352,804,864 free = 1,647,190,016 (101% used)
prompt: fsys main df
	main: 1,693,908,992 used + 35,184,325,369,856 free = 1,647,190,016 (102% used)

Overdraft protection? For reference:

:; ls -l /dev/sdC0/fossil
--rw-r----- S 0 a a 1650169856 Aug  1 22:57 /dev/sdC0/fossil

Things seem to be operating more-or-less normally so this isn't a huge
issue, but it's clearly a reporting bug.

Anthony

Re: [9fans] Fossil thinks my free space is a wee bit more than it is

[erik quanstrom]

> prompt: fsys main df
> 	main: 1,644,601,344 used + 2,588,672 free = 1,647,190,016 (99% used)
> prompt: fsys main df
> 	main: 1,652,891,648 used + 35,184,366,387,200 free = 1,647,190,016 (100% used)
> prompt: fsys main df
> 	main: 1,666,473,984 used + 35,184,352,804,864 free = 1,647,190,016 (101% used)
> prompt: fsys main df
> 	main: 1,693,908,992 used + 35,184,325,369,856 free = 1,647,190,016 (102% used)
> 
> Overdraft protection? For reference:
> 
> :; ls -l /dev/sdC0/fossil
> --rw-r----- S 0 a a 1650169856 Aug  1 22:57 /dev/sdC0/fossil
> 
> Things seem to be operating more-or-less normally so this isn't a huge
> issue, but it's clearly a reporting bug.
> 
> Anthony

let's see.  fossil uses bsize=8*1024.  35184366387200/8192 = 4294966600.
that looks like a small negative [sic] u32int to me.

i see where fl->used is incremented, but i don't see where it's decremented.
maybe that's correct, but it looks fishy.

- erik

Re: [9fans] Fossil thinks my free space is a wee bit more than it is

[Russ Cox]

The fix should be to edit the bottom of
/sys/src/cmd/fossil/cache.c:/^doRemoveLink
changing

	l.epochClose = p->epoch;
	blockSetLabel(b, &l, 0);

to

	l.epochClose = p->epoch;
	if(l.epochClose == l.epoch){
		vtLock(c->fl->lk);
		if(l.epoch == c->fl->epochlow)
			c->fl->nused--;
		blockSetLabel(b, &l, 0);
		vtUnlock(c->fl->lk);
	}else
		blockSetLabel(b, &l, 0);

Untested.

Treating l.epoch == l.epochClose as meaning the block is free
is an optimization that allows fossil to reclaim space allocated
and then discarded in the same snapshot.  The nused tracking
code was not changed when the optimization got added, 
causing the misreported use numbers.

Also, you can delete u32int nfree; from struct FreeList.
It is never accessed.

Russ