[PATCH] add path-oriented proc_mkdir_path() function to /proc

[PATCH] add path-oriented proc_mkdir_path() function to /proc

[Edward Falk]

Hi all; apologies if there's a maintainer I should be sending this to
as well; couldn't find one in the maintainers list or in the source code.

This patch adds the function proc_mkdir_path() to fs/proc/generic.c.
This allows kernel code to create e.g. "/proc/foo/bar/baz" without needing
to check the hard way if /proc/foo/ and /proc/foo/bar/ already exist.

This patch works with both 2.6.6-rc3 and 2.6.6

	-ed falk, efalk@google.com



diff -urN linux-2.6.6-rc3/fs/proc/generic.c linux/fs/proc/generic.c
--- linux-2.6.6-rc3/fs/proc/generic.c	2004-05-08 15:24:03.000000000 -0700
+++ linux/fs/proc/generic.c	2004-05-08 15:47:04.000000000 -0700
@@ -612,6 +612,40 @@
 	return proc_mkdir_mode(name, S_IRUGO | S_IXUGO, parent);
 }
 
+
+struct proc_dir_entry *proc_mkdir_path(const char *name,
+				struct proc_dir_entry *parent)
+{
+	const char		*next;
+	int			len;
+	struct proc_dir_entry	*de;
+
+	/* Search top level for first component of name.  If not found,
+	 * create it.  Keep doing this until we run out of name components.
+	 */
+
+	for(;;) {
+		next = strchr(name, '/');
+		len = next != NULL ? next - name : strlen(name);
+		for (de = parent->subdir; de != NULL; de = de->next) {
+			if (proc_match(len, name, de))
+				break;
+		}
+		if (de == NULL) {
+			char fname[NAME_MAX+1];
+			memcpy(fname, name, len); fname[len] = '\0';
+			de = proc_mkdir(fname, parent);
+		}
+		if (next == NULL)
+			break;
+		parent = de;
+		name = next + 1;
+	}
+	return de;
+}
+
+
+
 struct proc_dir_entry *create_proc_entry(const char *name, mode_t mode,
 					 struct proc_dir_entry *parent)
 {
diff -urN linux-2.6.6-rc3/include/linux/proc_fs.h linux/include/linux/proc_fs.h
--- linux-2.6.6-rc3/include/linux/proc_fs.h	2004-05-08 15:24:07.000000000 -0700
+++ linux/include/linux/proc_fs.h	2004-05-08 15:46:06.000000000 -0700
@@ -142,6 +142,8 @@
 extern struct proc_dir_entry *proc_mkdir(const char *,struct proc_dir_entry *);
 extern struct proc_dir_entry *proc_mkdir_mode(const char *name, mode_t mode,
 			struct proc_dir_entry *parent);
+extern struct proc_dir_entry *proc_mkdir_path(const char *path,
+			struct proc_dir_entry *parent);
 
 static inline struct proc_dir_entry *create_proc_read_entry(const char *name,
 	mode_t mode, struct proc_dir_entry *base, 

Re: [PATCH] add path-oriented proc_mkdir_path() function to /proc

[Valdis.Kletnieks]

[-- Attachment #1: Type: text/plain, Size: 1018 bytes --]

On Mon, 10 May 2004 15:09:08 PDT, Edward Falk <efalk@google.com>  said:
> Hi all; apologies if there's a maintainer I should be sending this to
> as well; couldn't find one in the maintainers list or in the source code.
> 
> This patch adds the function proc_mkdir_path() to fs/proc/generic.c.
> This allows kernel code to create e.g. "/proc/foo/bar/baz" without needing
> to check the hard way if /proc/foo/ and /proc/foo/bar/ already exist.

Hmm.. Looks like a useful utility function (there's certainly enough deep
trees in my /proc/sys tree), but I wonder...

1) Do we have cases where code should be implementing "it had *better* exist"
checks?  This may be important if an intermediate directory "should have" been
created by sysctl or something, and has special permission needs..

2) Alternatively, does using this open up accidental collisions where we should
have checked something *doesnt* exist already, and complain if it does?

(Feel free to address either one by adding a "Dont do that then" comment ;)

[-- Attachment #2: Type: application/pgp-signature, Size: 226 bytes --]

Re: [PATCH] add path-oriented proc_mkdir_path() function to /proc

[Paul Jackson]

Edward wrote:
> ... adds the function proc_mkdir_path() ...

Who uses this?

-- 
                          I won't rest till it's the best ...
                          Programmer, Linux Scalability
                          Paul Jackson <pj@sgi.com> 1.650.933.1373

Re: [PATCH] add path-oriented proc_mkdir_path() function to /proc

[Edward Falk]

[patch to allow creation of e.g. /proc/foo/bar/ with one function call.]

> Hmm.. Looks like a useful utility function (there's certainly enough deep
> trees in my /proc/sys tree), but I wonder...
> 
> 1) Do we have cases where code should be implementing "it had *better* exist"
> checks?  This may be important if an intermediate directory "should have" been
> created by sysctl or something, and has special permission needs..
> 
> 2) Alternatively, does using this open up accidental collisions where we should
> have checked something *doesnt* exist already, and complain if it does?
> 
> (Feel free to address either one by adding a "Dont do that then" comment ;)

Don't do that, then.  :)

OK, long answer:

1 & 2) are beyond the scope of my patch, but it seems to me that
additional functionality could be added -- perhaps in the form of O_CREAT,
O_EXCL flags -- if demand warranted it.  Perhaps this could be done in
a later patch?

	-ed falk, efalk@google.com

Re: [PATCH] add path-oriented proc_mkdir_path() function to /proc

[Edward Falk]

>> ... adds the function proc_mkdir_path() ...
>
>Who uses this?

Well, almost nobody yet :)

I'm very shortly going to submit a diagnostic utility that creates
nested entries in /proc.

I originally wrote proc_mkdir_path() to make the nested entries, and
someone suggested that it be submitted as a seperate patch since it's
of general usefulness.

	-ed falk, falk@google.com