linux-kernel - Re: [PATCH v2 1/8] lib/printbuf: New data structure for heap-allocated strings

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite for Android: free password hash cracker in your pocket

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Message-ID: <20220425041909.hcyirjphrkhxz6hx@moria.home.lan>
Date:   Mon, 25 Apr 2022 00:19:09 -0400
From:   Kent Overstreet <kent.overstreet@...il.com>
To:     Matthew Wilcox <willy@...radead.org>
Cc:     Joe Perches <joe@...ches.com>, linux-kernel@...r.kernel.org,
        linux-mm@...ck.org, linux-fsdevel@...r.kernel.org, hch@....de,
        hannes@...xchg.org, akpm@...ux-foundation.org,
        linux-clk@...r.kernel.org, linux-tegra@...r.kernel.org,
        linux-input@...r.kernel.org, roman.gushchin@...ux.dev
Subject: Re: [PATCH v2 1/8] lib/printbuf: New data structure for
 heap-allocated strings

On Mon, Apr 25, 2022 at 03:44:34AM +0100, Matthew Wilcox wrote:
> On Sun, Apr 24, 2022 at 04:46:03PM -0700, Joe Perches wrote:
> > > + * pr_human_readable_u64, pr_human_readable_s64: Print an integer with human
> > > + * readable units.
> > 
> > Why not extend vsprintf for this using something like %pH[8|16|32|64] 
> > or %pH[c|s|l|ll|uc|us|ul|ull] ?
> 
> The %pX extension we have is _cute_, but ultimately a bad idea.  It
> centralises all kinds of unrelated things in vsprintf.c, eg bdev_name()
> and clock() and ip_addr_string().

And it's not remotely discoverable. I didn't realize we had bdev_name()
available as a format string until just now or I would've been using it!

> Really, it's working around that we don't have something like Java's
> StringBuffer (which I see both seq_buf and printbuf as attempting to
> be).  So we have this primitive format string hack instead of exposing
> methods like:
> 
> void dentry_string(struct strbuf *, struct dentry *);

Exactly!

> as an example,
>                 if (unlikely(ino == dir->i_ino)) {
>                         EXT4_ERROR_INODE(dir, "'%pd' linked to parent dir",
>                                          dentry);
>                         return ERR_PTR(-EFSCORRUPTED);
>                 }
> 
> would become something like:
> 
> 		if (unlikely(ino == dir->i_ino)) {
> 			struct strbuf strbuf;
> 			strbuf_char(strbuf, '\'');
> 			dentry_string(strbuf, dentry);
> 			strbuf_string(strbuf, "' linked to parent dir");
> 			EXT4_ERROR_INODE(dir, strbuf);
> 			return ERR_PTR(-EFSCORRUPTED);
> 		}
> 
> which isn't terribly nice, but C has sucky syntax for string
> construction.  Other languages have done this better, including Rust.

Over IRC just now you proposed "%p(%p)", dentry_name, dentry - I'm _really_
liking this idea, especially if we can get glibc to take it.

Then your ext4 example becomes just 

	if (unlikely(ino == dir->i_ino)) {
		EXT4_ERROR_INODE(dir, "'%p(%p)' linked to parent dir",
				 dentry_name, dentry);
		return ERR_PTR(-EFSCORRUPTED);
	}

And you can cscope to the pretty-printer! And dentry_name becomes just

void dentry_name(struct printbuf *out, struct dentry *dentry)
{
	...
}

Which is quite a bit simpler than the current definition.

Sweeeeeet.