[PATCH] GFS2: Fix unsafe dereference in dump_holder()

[PATCH] GFS2: Fix unsafe dereference in dump_holder()

[Tetsuo Handa]

>From efee681e4ca6152d9549a50e93ca932b00f03014 Mon Sep 17 00:00:00 2001
From: Tetsuo Handa <penguin-kernel@I-love.SAKURA.ne.jp>
Date: Tue, 24 Dec 2013 18:04:43 +0900
Subject: [PATCH] GFS2: Fix unsafe dereference in dump_holder()

Since we are not in RCU nor tasklist_lock held, we need to grab a reference on
the returned task_struct. Use get_pid_task()/get_task_comm() to safely read the
target task's comm name.

Signed-off-by: Tetsuo Handa <penguin-kernel@I-love.SAKURA.ne.jp>
---
 fs/gfs2/glock.c |   11 +++++++++--
 1 files changed, 9 insertions(+), 2 deletions(-)

diff --git a/fs/gfs2/glock.c b/fs/gfs2/glock.c
index c8420f7..7bda965 100644
--- a/fs/gfs2/glock.c
+++ b/fs/gfs2/glock.c
@@ -1654,15 +1654,22 @@ static int dump_holder(struct seq_file *seq, const struct gfs2_holder *gh)
 {
 	struct task_struct *gh_owner = NULL;
 	char flags_buf[32];
+	char name[TASK_COMM_LEN];
 
 	if (gh->gh_owner_pid)
-		gh_owner = pid_task(gh->gh_owner_pid, PIDTYPE_PID);
+		gh_owner = get_pid_task(gh->gh_owner_pid, PIDTYPE_PID);
+	if (gh_owner) {
+		get_task_comm(name, gh_owner);
+		put_task_struct(gh_owner);
+	} else {
+		strlcpy(name, "(ended)", sizeof(name));
+	}
 	gfs2_print_dbg(seq, " H: s:%s f:%s e:%d p:%ld [%s] %pS\n",
 		       state2str(gh->gh_state),
 		       hflags2str(flags_buf, gh->gh_flags, gh->gh_iflags),
 		       gh->gh_error,
 		       gh->gh_owner_pid ? (long)pid_nr(gh->gh_owner_pid) : -1,
-		       gh_owner ? gh_owner->comm : "(ended)",
+		       name,
 		       (void *)gh->gh_ip);
 	return 0;
 }
-- 
1.7.1

Re: [PATCH] GFS2: Fix unsafe dereference in dump_holder()

[Steven Whitehouse]

Hi,

On Tue, 2013-12-24 at 20:51 +0900, Tetsuo Handa wrote:
> >From efee681e4ca6152d9549a50e93ca932b00f03014 Mon Sep 17 00:00:00 2001
> From: Tetsuo Handa <penguin-kernel@I-love.SAKURA.ne.jp>
> Date: Tue, 24 Dec 2013 18:04:43 +0900
> Subject: [PATCH] GFS2: Fix unsafe dereference in dump_holder()
> 
> Since we are not in RCU nor tasklist_lock held, we need to grab a reference on
> the returned task_struct. Use get_pid_task()/get_task_comm() to safely read the
> target task's comm name.
> 
> Signed-off-by: Tetsuo Handa <penguin-kernel@I-love.SAKURA.ne.jp>
> ---

Well this is not entirely true... for 99.9% of calls to this function we
do have the rcu read lock held (when called from the seq file code). So
the only issue is when called from the GLOCK_BUG_ON() macro.

Since it appears that rcu_read_lock() will nest, then we could
presumably just add that around dump_holder or even around
gfs2_dump_glock ?

I think that is a neater solution than taking extra ref counts and
copying the string around,

Steve.

>  fs/gfs2/glock.c |   11 +++++++++--
>  1 files changed, 9 insertions(+), 2 deletions(-)
> 
> diff --git a/fs/gfs2/glock.c b/fs/gfs2/glock.c
> index c8420f7..7bda965 100644
> --- a/fs/gfs2/glock.c
> +++ b/fs/gfs2/glock.c
> @@ -1654,15 +1654,22 @@ static int dump_holder(struct seq_file *seq, const struct gfs2_holder *gh)
>  {
>  	struct task_struct *gh_owner = NULL;
>  	char flags_buf[32];
> +	char name[TASK_COMM_LEN];
>  
>  	if (gh->gh_owner_pid)
> -		gh_owner = pid_task(gh->gh_owner_pid, PIDTYPE_PID);
> +		gh_owner = get_pid_task(gh->gh_owner_pid, PIDTYPE_PID);
> +	if (gh_owner) {
> +		get_task_comm(name, gh_owner);
> +		put_task_struct(gh_owner);
> +	} else {
> +		strlcpy(name, "(ended)", sizeof(name));
> +	}
>  	gfs2_print_dbg(seq, " H: s:%s f:%s e:%d p:%ld [%s] %pS\n",
>  		       state2str(gh->gh_state),
>  		       hflags2str(flags_buf, gh->gh_flags, gh->gh_iflags),
>  		       gh->gh_error,
>  		       gh->gh_owner_pid ? (long)pid_nr(gh->gh_owner_pid) : -1,
> -		       gh_owner ? gh_owner->comm : "(ended)",
> +		       name,
>  		       (void *)gh->gh_ip);
>  	return 0;
>  }

Re: [PATCH] GFS2: Fix unsafe dereference in dump_holder()

[Tetsuo Handa]

Steven Whitehouse wrote:
> > Subject: [PATCH] GFS2: Fix unsafe dereference in dump_holder()
> > 
> > Since we are not in RCU nor tasklist_lock held, we need to grab a reference on
> > the returned task_struct. Use get_pid_task()/get_task_comm() to safely read the
> > target task's comm name.
> > 
> > Signed-off-by: Tetsuo Handa <penguin-kernel@I-love.SAKURA.ne.jp>
> > ---
> 
> Well this is not entirely true... for 99.9% of calls to this function we
> do have the rcu read lock held (when called from the seq file code). So
> the only issue is when called from the GLOCK_BUG_ON() macro.
> 
> Since it appears that rcu_read_lock() will nest, then we could
> presumably just add that around dump_holder or even around
> gfs2_dump_glock ?
> 
> I think that is a neater solution than taking extra ref counts and
> copying the string around,

OK. I didn't check what lock protects this dereference. I just noticed this
dereference when I was checking direct ->comm access which might fail to return
consistent comm name. I might propose changing ->comm access to RCU in the
future ( https://lkml.org/lkml/2013/12/25/21 ).

For now add only rcu_read_lock()/rcu_read_unlock(). Patch follows.
----------
>From 63f41a384636be1acd1575ee9f9e0de317cbc131 Mon Sep 17 00:00:00 2001
From: Tetsuo Handa <penguin-kernel@I-love.SAKURA.ne.jp>
Date: Thu, 2 Jan 2014 19:41:23 +0900
Subject: [PATCH v2] GFS2: Fix unsafe dereference in dump_holder()

GLOCK_BUG_ON() might call this function without RCU read lock. Make sure that
RCU read lock is held when using task_struct returned from pid_task().

Signed-off-by: Tetsuo Handa <penguin-kernel@I-love.SAKURA.ne.jp>
---
 fs/gfs2/glock.c |    2 ++
 1 files changed, 2 insertions(+), 0 deletions(-)

diff --git a/fs/gfs2/glock.c b/fs/gfs2/glock.c
index c8420f7..6f7a47c 100644
--- a/fs/gfs2/glock.c
+++ b/fs/gfs2/glock.c
@@ -1655,6 +1655,7 @@ static int dump_holder(struct seq_file *seq, const struct gfs2_holder *gh)
 	struct task_struct *gh_owner = NULL;
 	char flags_buf[32];
 
+	rcu_read_lock();
 	if (gh->gh_owner_pid)
 		gh_owner = pid_task(gh->gh_owner_pid, PIDTYPE_PID);
 	gfs2_print_dbg(seq, " H: s:%s f:%s e:%d p:%ld [%s] %pS\n",
@@ -1664,6 +1665,7 @@ static int dump_holder(struct seq_file *seq, const struct gfs2_holder *gh)
 		       gh->gh_owner_pid ? (long)pid_nr(gh->gh_owner_pid) : -1,
 		       gh_owner ? gh_owner->comm : "(ended)",
 		       (void *)gh->gh_ip);
+	rcu_read_unlock();
 	return 0;
 }
 
-- 
1.7.1

Re: [PATCH] GFS2: Fix unsafe dereference in dump_holder()

[Steven Whitehouse]

Hi,

Thanks - that looks good. I've applied it to the GFS2 -nmw and -fixes
trees,

Steve.

On Thu, 2014-01-02 at 19:52 +0900, Tetsuo Handa wrote:
> Steven Whitehouse wrote:
> > > Subject: [PATCH] GFS2: Fix unsafe dereference in dump_holder()
> > > 
> > > Since we are not in RCU nor tasklist_lock held, we need to grab a reference on
> > > the returned task_struct. Use get_pid_task()/get_task_comm() to safely read the
> > > target task's comm name.
> > > 
> > > Signed-off-by: Tetsuo Handa <penguin-kernel@I-love.SAKURA.ne.jp>
> > > ---
> > 
> > Well this is not entirely true... for 99.9% of calls to this function we
> > do have the rcu read lock held (when called from the seq file code). So
> > the only issue is when called from the GLOCK_BUG_ON() macro.
> > 
> > Since it appears that rcu_read_lock() will nest, then we could
> > presumably just add that around dump_holder or even around
> > gfs2_dump_glock ?
> > 
> > I think that is a neater solution than taking extra ref counts and
> > copying the string around,
> 
> OK. I didn't check what lock protects this dereference. I just noticed this
> dereference when I was checking direct ->comm access which might fail to return
> consistent comm name. I might propose changing ->comm access to RCU in the
> future ( https://lkml.org/lkml/2013/12/25/21 ).
> 
> For now add only rcu_read_lock()/rcu_read_unlock(). Patch follows.
> ----------
> >From 63f41a384636be1acd1575ee9f9e0de317cbc131 Mon Sep 17 00:00:00 2001
> From: Tetsuo Handa <penguin-kernel@I-love.SAKURA.ne.jp>
> Date: Thu, 2 Jan 2014 19:41:23 +0900
> Subject: [PATCH v2] GFS2: Fix unsafe dereference in dump_holder()
> 
> GLOCK_BUG_ON() might call this function without RCU read lock. Make sure that
> RCU read lock is held when using task_struct returned from pid_task().
> 
> Signed-off-by: Tetsuo Handa <penguin-kernel@I-love.SAKURA.ne.jp>
> ---
>  fs/gfs2/glock.c |    2 ++
>  1 files changed, 2 insertions(+), 0 deletions(-)
> 
> diff --git a/fs/gfs2/glock.c b/fs/gfs2/glock.c
> index c8420f7..6f7a47c 100644
> --- a/fs/gfs2/glock.c
> +++ b/fs/gfs2/glock.c
> @@ -1655,6 +1655,7 @@ static int dump_holder(struct seq_file *seq, const struct gfs2_holder *gh)
>  	struct task_struct *gh_owner = NULL;
>  	char flags_buf[32];
>  
> +	rcu_read_lock();
>  	if (gh->gh_owner_pid)
>  		gh_owner = pid_task(gh->gh_owner_pid, PIDTYPE_PID);
>  	gfs2_print_dbg(seq, " H: s:%s f:%s e:%d p:%ld [%s] %pS\n",
> @@ -1664,6 +1665,7 @@ static int dump_holder(struct seq_file *seq, const struct gfs2_holder *gh)
>  		       gh->gh_owner_pid ? (long)pid_nr(gh->gh_owner_pid) : -1,
>  		       gh_owner ? gh_owner->comm : "(ended)",
>  		       (void *)gh->gh_ip);
> +	rcu_read_unlock();
>  	return 0;
>  }
>