Re: [PATCH] configfs: fix a race in configfs_lookup()

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

 



> You lost all the original signed-off-by lines of the original, AND you
> lost the authorship of the original commit.  And you didn't cc: anyone
> involved in the original patch, to get their review, or objection to it
> being backported.
Sorry for the rookie mistakes. I drafted another version and it is
attached to the email. Can you please check whether it is OK?

> Also, how did you test this change?  is this something that you have
> actually hit in real life?
Yes. I encountered this when testing the latest v5.10.y branch. A
minimal proof-of-concept code looks like this:
~~~
#include <stdio.h>
#include <stdlib.h>
#include <fcntl.h>
#include <unistd.h>

int main(void)
{
	int fd = open("/sys/kernel/config", 0);

	if(!fork()) {
		while(1) lseek(fd, SEEK_CUR, 1);
	}
	while(1) unlinkat(fd, "file", 0);

	return 0;
}
~~~
>From 366d165e876a14a5b25ad741fdcd8658aa7e690d Mon Sep 17 00:00:00 2001
From: Kyle Zeng <zengyhkyle@xxxxxxxxx>
Date: Fri, 1 Sep 2023 17:41:14 -0700
Subject: [PATCH v5.10.y] configfs: fix a race in configfs_lookup()

commit c42dd069be8dfc9b2239a5c89e73bbd08ab35de0 upstream.

configfs: fix a race in configfs_lookup()
When configfs_lookup() is executing list_for_each_entry(),
it is possible that configfs_dir_lseek() is calling list_del().
Some unfortunate interleavings of them can cause a kernel NULL
pointer dereference error

Thread 1                  Thread 2
//configfs_dir_lseek()    //configfs_lookup()
list_del(&cursor->s_sibling);
                         list_for_each_entry(sd, ...)

Fix this by grabbing configfs_dirent_lock in configfs_lookup()
while iterating ->s_children.

Signed-off-by: Sishuai Gong <sishuai@xxxxxxxxxx>
Signed-off-by: Christoph Hellwig <hch@xxxxxx>
Signed-off-by: Kyle Zeng <zengyhkyle@xxxxxxxxx>
---
Please apply this patch to v5.10.y. This patch adapts the original patch
to v5.10.y considering codebase change.
The idea is to hold the configfs_dirent_lock when traversing
->s_children, which follows the core idea of the original patch.

This patch has been tested against the v5.10.y stable tree.

 fs/configfs/dir.c | 2 ++
 1 file changed, 2 insertions(+)

diff --git a/fs/configfs/dir.c b/fs/configfs/dir.c
index 12388ed4faa5..0b7e9ab517d5 100644
--- a/fs/configfs/dir.c
+++ b/fs/configfs/dir.c
@@ -479,6 +479,7 @@ static struct dentry * configfs_lookup(struct inode *dir,
 	if (!configfs_dirent_is_ready(parent_sd))
 		goto out;
 
+	spin_lock(&configfs_dirent_lock);
 	list_for_each_entry(sd, &parent_sd->s_children, s_sibling) {
 		if (sd->s_type & CONFIGFS_NOT_PINNED) {
 			const unsigned char * name = configfs_get_name(sd);
@@ -491,6 +492,7 @@ static struct dentry * configfs_lookup(struct inode *dir,
 			break;
 		}
 	}
+	spin_unlock(&configfs_dirent_lock);
 
 	if (!found) {
 		/*
-- 
2.34.1


[Index of Archives]     [Linux Kernel]     [Kernel Development Newbies]     [Linux USB Devel]     [Video for Linux]     [Linux Audio Users]     [Yosemite Hiking]     [Linux Kernel]     [Linux SCSI]

  Powered by Linux