[2.4.23-pre3] Possible bug in fs/buffer.c

[2.4.23-pre3] Possible bug in fs/buffer.c

[Rolf Eike Beer]

This is __put_unused_buffer_head from fs/buffer.c, lines 1156 to 1171:


static void __put_unused_buffer_head(struct buffer_head * bh)
{
	if (unlikely(buffer_attached(bh)))
		BUG();
	if (nr_unused_buffer_heads >= MAX_UNUSED_BUFFERS) {
		kmem_cache_free(bh_cachep, bh);
	} else {
		bh->b_dev = B_FREE;
===>		bh->b_blocknr = -1;		<===
		bh->b_this_page = NULL;

		nr_unused_buffer_heads++;
		bh->b_next_free = unused_list;
		unused_list = bh;
	}
}

In include/linux/fs.h "struct buffer_head" is defined this way:

struct buffer_head {
        /* First cache line: */
        struct buffer_head *b_next;     /* Hash queue list */
        unsigned long b_blocknr;        /* block number */
...

So this line (and line 1205, which is the same) is either ugly (and someone
meant ~0UL or something similar) or completely bogus. Same way in 
2.6.0-test4-bk10/fs/buffer.c, line 1031 (b_blocknr is a sector_t, which is an
unsigned long).

Comments?

Eike

Re: [2.4.23-pre3] Possible bug in fs/buffer.c

[Andreas Schwab]

Rolf Eike Beer <eike-kernel@sf-tec.de> writes:

> This is __put_unused_buffer_head from fs/buffer.c, lines 1156 to 1171:
>
>
> static void __put_unused_buffer_head(struct buffer_head * bh)
> {
> 	if (unlikely(buffer_attached(bh)))
> 		BUG();
> 	if (nr_unused_buffer_heads >= MAX_UNUSED_BUFFERS) {
> 		kmem_cache_free(bh_cachep, bh);
> 	} else {
> 		bh->b_dev = B_FREE;
> ===>		bh->b_blocknr = -1;		<===
> 		bh->b_this_page = NULL;
>
> 		nr_unused_buffer_heads++;
> 		bh->b_next_free = unused_list;
> 		unused_list = bh;
> 	}
> }
>
> In include/linux/fs.h "struct buffer_head" is defined this way:
>
> struct buffer_head {
>         /* First cache line: */
>         struct buffer_head *b_next;     /* Hash queue list */
>         unsigned long b_blocknr;        /* block number */
> ...
>
> So this line (and line 1205, which is the same) is either ugly (and someone
> meant ~0UL or something similar) or completely bogus.

It's neither ugly, nor bogus.  The only 100% reliable way to assign the
maximum value to an unsigned integer is to use -1.

Andreas.

-- 
Andreas Schwab, SuSE Labs, schwab@suse.de
SuSE Linux AG, Deutschherrnstr. 15-19, D-90429 Nürnberg
Key fingerprint = 58CA 54C7 6D53 942B 1756  01D3 44D5 214B 8276 4ED5
"And now for something completely different."

Re: [2.4.23-pre3] Possible bug in fs/buffer.c

[Alan Cox]

On Llu, 2003-09-08 at 16:42, Andreas Schwab wrote:
> It's neither ugly, nor bogus.  The only 100% reliable way to assign the
> maximum value to an unsigned integer is to use -1.

Its not 100% reliable either 8). Properly you should use the limits.h
values. The kernel assumes 2's complement so just adding a cast would
probably keep gcc happy

Re: [2.4.23-pre3] Possible bug in fs/buffer.c

[Andreas Schwab]

Alan Cox <alan@lxorguk.ukuu.org.uk> writes:

> On Llu, 2003-09-08 at 16:42, Andreas Schwab wrote:
>> It's neither ugly, nor bogus.  The only 100% reliable way to assign the
>> maximum value to an unsigned integer is to use -1.
>
> Its not 100% reliable either 8).

Could you please elaborate?  Casting -1 to an unsigned type is guaranteed
to yield the maximum value for that type, at least since C89, but I think
even K&R C did get it right.

Andreas.

-- 
Andreas Schwab, SuSE Labs, schwab@suse.de
SuSE Linux AG, Deutschherrnstr. 15-19, D-90429 Nürnberg
Key fingerprint = 58CA 54C7 6D53 942B 1756  01D3 44D5 214B 8276 4ED5
"And now for something completely different."

Re: [2.4.23-pre3] Possible bug in fs/buffer.c

[J.A. Magallon]

On 09.08, Andreas Schwab wrote:
> Alan Cox <alan@lxorguk.ukuu.org.uk> writes:
> 
> > On Llu, 2003-09-08 at 16:42, Andreas Schwab wrote:
> >> It's neither ugly, nor bogus.  The only 100% reliable way to assign the
> >> maximum value to an unsigned integer is to use -1.
> >
> > Its not 100% reliable either 8).
> 
> Could you please elaborate?  Casting -1 to an unsigned type is guaranteed
> to yield the maximum value for that type, at least since C89, but I think
> even K&R C did get it right.
> 

Would not be much cleaner to do use ~0UL ?

-- 
J.A. Magallon <jamagallon@able.es>      \            Software is like sex:
werewolf.able.es                         \      It's better when it's free
Mandrake Linux release 9.2 (Cooker) for i586
Linux 2.4.23-pre2-jam1m (gcc 3.3.1 (Mandrake Linux 9.2 3.3.1-1mdk))

Re: [2.4.23-pre3] Possible bug in fs/buffer.c

[Alan Cox]

On Llu, 2003-09-08 at 17:04, Andreas Schwab wrote:
> > Its not 100% reliable either 8).
> 
> Could you please elaborate?  Casting -1 to an unsigned type is guaranteed
> to yield the maximum value for that type, at least since C89, but I think
> even K&R C did get it right.

My error - its ~0 that is unreliable.

Re: [2.4.23-pre3] Possible bug in fs/buffer.c

[Rolf Eike Beer]

Am Dienstag, 9. September 2003 02:53 schrieb Alan Cox:
> On Llu, 2003-09-08 at 17:04, Andreas Schwab wrote:
> > > Its not 100% reliable either 8).
> >
> > Could you please elaborate?  Casting -1 to an unsigned type is guaranteed
> > to yield the maximum value for that type, at least since C89, but I think
> > even K&R C did get it right.
>
> My error - its ~0 that is unreliable.

Uh-oh:

eike@bilbo:/mnt/kernel/linux-2.4.23-pre3> grep -r "~0[^xX]" *| wc -l
    713

Eike

Re: [2.4.23-pre3] Possible bug in fs/buffer.c

[Rolf Eike Beer]

Am Montag, 8. September 2003 17:58 schrieb Alan Cox:
> On Llu, 2003-09-08 at 16:42, Andreas Schwab wrote:
> > It's neither ugly, nor bogus.  The only 100% reliable way to assign the
> > maximum value to an unsigned integer is to use -1.
>
> Its not 100% reliable either 8). Properly you should use the limits.h
> values. The kernel assumes 2's complement so just adding a cast would
> probably keep gcc happy

gcc didn't even find that. He complained about a line slightly above this one. 
In limits.h there is no value equal to -1UL.

Eike

Re: [2.4.23-pre3] Possible bug in fs/buffer.c

[Alan Cox]

On Llu, 2003-09-08 at 17:12, Rolf Eike Beer wrote:
> gcc didn't even find that. He complained about a line slightly above this one. 
> In limits.h there is no value equal to -1UL.

UINT_MAX in a standard C limits.h. Right now I dont think the kernel
defines it but  it could and ANSI C deals with this stuff.