Re: [PATCH] make last_inode counter in new_inode 32-bit on kernels that offer x86 compatability

[Eric Dumazet <dada1@cosmosbay.com>]

Jörn Engel a écrit :
> On Mon, 6 November 2006 13:47:23 -0500, Jeff Layton wrote:
>> On Mon, 2006-11-06 at 11:22 -0700, Matthew Wilcox wrote:
>>> On Mon, Nov 06, 2006 at 01:12:05PM -0500, Jeff Layton wrote:
>>>> The attached patch remedies this by making the last_inode counter be an
>>>> unsigned int on kernels that have ia32 compatability mode enabled.
>>> ... and this only happens on ia64/x86_64 kernels, not sparc64, ppc64,
>>> s390x, parisc64 or mips64?
>> Here's a new (untested) patch that replaces the ia32 specific
>> compatability mode defines with CONFIG_COMPAT, as suggested by Matthew.
> 
> While you're at it, how about making last_ino per-sb instead of
> system-wide?  ino collisions after a wrap are just as bad as inos
> beyond 32bit.  And this should be a fairly simple method to reduce the
> risk.
> 
> Also, do you have a testcase that can actually force the wrap?

while (1) {
	int fd[2];
	pipe(fd);
	close(fd[0]);
	close(fd[1]);
}