[9fans] 9P1 clients?

[9fans] 9P1 clients?

[geoff]

Hmm, no answer on plan9dev, so I'll try here⋯

Does anybody still have 9P1 clients (other than drawterm, which we
have 9P2000 versions of now)?  I ask because I've got a version of
Ken's file server that uses 64-bit file sizes, offsets and block
numbers.  It works just fine, but 9P1 isn't a great fit, since some of
the relevant fields are only 32 bits wide (e.g., qid.path).

Incidentally, here's a program to find the largest file size possible
on a given file system empirically.  With a 4kB block size and
triple-indirect blocks on my 63-bit file server, it prints

	542,811,039,744 (5.42811e+11)

which matches the theoretical limit.  Quadruple-indirect blocks may be
called for to reach 2⁶⁳ bytes.


# To unbundle, run this file
echo 9fslimit.c
sed 's/^X//' >9fslimit.c <<'!'
X/* 9fslimit >tempfile - perform a binary search to find largest file size */

X#include <u.h>
X#include <libc.h>

static int
iswritablebyte(vlong offset)
X{
X	char c;

X	return seek(1, offset, 0) >= 0 && write(1, &c, 1) == 1;
X}

static int
islastbyte(vlong offset)
X{
X	return iswritablebyte(offset) && !iswritablebyte(offset+1);
X}

X/*
X * sometimes limits are near powers of 2, but sometimes they are
X * determined by file system structure (e.g., indirect blocks).
X */
void
main(int argc, char **argv)
X{
X	int verbose = 0;
X	uvlong low = 0, high = ~0ULL, new = 0;

X	ARGBEGIN {
X	case 'v':
X		verbose = 1;
X		break;
X	default:
X		fprint(2, "usage: %s >junkfile\n", argv0);
X		exits("usage");
X	} ARGEND;
X	while (low <= high){
X		new = (high + low)/2;
X		if (verbose)
X			fprint(2, "new guess: %,lld\n", new);
X		if (islastbyte(new))
X			break;
X		if (iswritablebyte(new))
X			low = new + 1;
X		else
X			high = new - 1;
X	}
X	if (low > high)
X		fprint(2, "failed to converge\n");
X	else {
X		/*
X		 * new is last possible offset with zero-origin, so add 1 to
X		 * get size.
X		 */
X		++new;
X		fprint(2, "%,lld (%g)\n", new, (double)new);
X	}
X	exits(0);
X}
!

Re: [9fans] 9P1 clients?

[geoff]

Don't run 9fslimits on a file system that doesn't support sparse files
(holes).  Andrey ran it on fossil and it's apparently made fossil
misbehave.

Re: [9fans] 9P1 clients?

[Russ Cox]

fossil should have no problem with sparse files.

russ