Skip to contentSkip to conversation

rcu_read_lock/unlock protect find_task_by_vpid call in migrate_pages

PATCH Dave Jones 3 participants 5 messages Nov 25, 2010

Subsystems (2)
lore
Share & tools
lore

Dave Jones

author
#0 lore raw

===================================================
[ INFO: suspicious rcu_dereference_check() usage. ]
---------------------------------------------------
kernel/pid.c:419 invoked rcu_dereference_check() without protection!

other info that might help us debug this:

rcu_scheduler_active = 1, debug_locks = 1
1 lock held by scrashme/11079:
#0: (tasklist_lock){.?.?..}, at: [<ffffffff81112702>] sys_migrate_pages+0xd6/0x29d

stack backtrace:
Pid: 11079, comm: scrashme Not tainted 2.6.37-rc3+ #2
Call Trace:
 [<ffffffff8107d095>] lockdep_rcu_dereference+0x9d/0xa5
 [<ffffffff81069dc8>] find_task_by_pid_ns+0x44/0x5d
 [<ffffffff81069e03>] find_task_by_vpid+0x22/0x24
 [<ffffffff8111270f>] sys_migrate_pages+0xe3/0x29d
 [<ffffffff8107c7c6>] ? trace_hardirqs_off_caller+0xa3/0x10b
 [<ffffffff8147b635>] ? retint_swapgs+0x13/0x1b
 [<ffffffff8107e2e7>] ? trace_hardirqs_on_caller+0x13f/0x172
 [<ffffffff8147a95c>] ? trace_hardirqs_on_thunk+0x3a/0x3f
 [<ffffffff81009cb2>] system_call_fastpath+0x16/0x1b
mm/mempolicy.c +2 -0
index 4a57f13..2f0f55b 100644
@@ -1308,6 +1308,7 @@ SYSCALL_DEFINE4(migrate_pages, pid_t, pid, unsigned long, maxnode,
 
 	/* Find the mm_struct */
 	read_lock(&tasklist_lock);
+	rcu_read_lock();
 	task = pid ? find_task_by_vpid(pid) : current;
 	if (!task) {
 		read_unlock(&tasklist_lock);
@@ -1315,6 +1316,7 @@ SYSCALL_DEFINE4(migrate_pages, pid_t, pid, unsigned long, maxnode,
 		goto out;
 	}
 	mm = get_task_mm(task);
+	rcu_read_unlock();
 	read_unlock(&tasklist_lock);
 
 	err = -EINVAL;
#1 lore raw

Hi

Quoted 15 lines
mm/mempolicy.c +1 -0
index 4a57f13..2f0f55b 100644
@@ -1308,6 +1308,7 @@ SYSCALL_DEFINE4(migrate_pages, pid_t, pid, unsigned long, maxnode,
 
 	/* Find the mm_struct */
 	read_lock(&tasklist_lock);
+	rcu_read_lock();
 	task = pid ? find_task_by_vpid(pid) : current;
 	if (!task) {
 		read_unlock(&tasklist_lock);

mm/mempolicy.c SYSCALL_DEFINE4(migrate_pages, pid_t, pid, unsigned long, maxnode, · line 1322

Quoted 7 lines
mm/mempolicy.c +1 -0
@@ -1315,6 +1316,7 @@ SYSCALL_DEFINE4(migrate_pages, pid_t, pid, unsigned long, maxnode,
 		goto out;
 	}
 	mm = get_task_mm(task);
+	rcu_read_unlock();
 	read_unlock(&tasklist_lock);
 
 	err = -EINVAL;

Thanks reporting. but mmotm seems already has the same patch.

author
#2 lore raw
Quoted 10 lines · KOSAKI Motohiro

mm/mempolicy.c SYSCALL_DEFINE4(migrate_pages, pid_t, pid, unsigned long, maxnode, · line 1322

Quoted 8 lines
mm/mempolicy.c +1 -0
@@ -1315,6 +1316,7 @@ SYSCALL_DEFINE4(migrate_pages, pid_t, pid, unsigned long, maxnode,
 		goto out;
 	}
 	mm = get_task_mm(task);
+	rcu_read_unlock();
 	read_unlock(&tasklist_lock);
 
 	err = -EINVAL;

Thanks reporting. but mmotm seems already has the same patch.

indeed. It's also better than mine, which missed the unlock on error path.

thanks,

Dave

#3 lore raw
In reply to KOSAKI Motohiro
Quoted 23 lines · KOSAKI Motohiro

Hi

Quoted 21 lines
mm/mempolicy.c +2 -0
index 4a57f13..2f0f55b 100644
@@ -1308,6 +1308,7 @@ SYSCALL_DEFINE4(migrate_pages, pid_t, pid, unsigned long, maxnode,
 
 	/* Find the mm_struct */
 	read_lock(&tasklist_lock);
+	rcu_read_lock();
 	task = pid ? find_task_by_vpid(pid) : current;
 	if (!task) {
 		read_unlock(&tasklist_lock);
@@ -1315,6 +1316,7 @@ SYSCALL_DEFINE4(migrate_pages, pid_t, pid, unsigned long, maxnode,
 		goto out;
 	}
 	mm = get_task_mm(task);
+	rcu_read_unlock();
 	read_unlock(&tasklist_lock);
 
Quoted 3 lines

mm/mempolicy.c SYSCALL_DEFINE4(migrate_pages, pid_t, pid, unsigned long, maxnode, · line 1322

Quoted 1 line
mm/mempolicy.c +0 -0
 	err = -EINVAL;

Thanks reporting. but mmotm seems already has the same patch.

Another one,... again:

Do we still need the tasklist_lock in this case?

Also, why is that think complaining, surely the tasklist_lock pins any and all PID objects?

#4 lore raw
Quoted 28 lines · KOSAKI Motohiro
Quoted 23 lines

Hi

Quoted 21 lines
mm/mempolicy.c +2 -0
index 4a57f13..2f0f55b 100644
@@ -1308,6 +1308,7 @@ SYSCALL_DEFINE4(migrate_pages, pid_t, pid, unsigned long, maxnode,
 
 	/* Find the mm_struct */
 	read_lock(&tasklist_lock);
+	rcu_read_lock();
 	task = pid ? find_task_by_vpid(pid) : current;
 	if (!task) {
 		read_unlock(&tasklist_lock);
@@ -1315,6 +1316,7 @@ SYSCALL_DEFINE4(migrate_pages, pid_t, pid, unsigned long, maxnode,
 		goto out;
 	}
 	mm = get_task_mm(task);
+	rcu_read_unlock();
 	read_unlock(&tasklist_lock);
 
Quoted 3 lines

mm/mempolicy.c SYSCALL_DEFINE4(migrate_pages, pid_t, pid, unsigned long, maxnode, · line 1322

Quoted 1 line
mm/mempolicy.c +0 -0
 	err = -EINVAL;

Thanks reporting. but mmotm seems already has the same patch.

Quoted 7 lines

Another one,... again:

Do we still need the tasklist_lock in this case?

Also, why is that think complaining, surely the tasklist_lock pins any and all PID objects?

(cc to Christoph and Oleg)

Good spotting.

As far as I understand, find_task_by_pid() did required tasklist_lock in old days. but It isn't now.

So I think we can remove tasklist_lock from here.

===========================================================================

Quoted 1 line

From 9de9f70f74e55d92b5e9057e22fc629405f63295 Mon Sep 17 00:00:00 2001

From: KOSAKI Motohiro <[email protected]>
Date: Thu, 16 Dec 2010 17:49:23 +0900
Subject: [PATCH] mempolicy: remove tasklist_lock from migrate_pages

Today, tasklist_lock in migrate_pages doesn't protect anything. rcu_read_lock() provide enough protection from pid hash walk.

---
 mm/mempolicy.c |    3 ---
 1 files changed, 0 insertions(+), 3 deletions(-)
mm/mempolicy.c +0 -3
index 11ff260..9064945 100644
@@ -1308,16 +1308,13 @@ SYSCALL_DEFINE4(migrate_pages, pid_t, pid, unsigned long, maxnode,
 
 	/* Find the mm_struct */
 	rcu_read_lock();
-	read_lock(&tasklist_lock);
 	task = pid ? find_task_by_vpid(pid) : current;
 	if (!task) {
-		read_unlock(&tasklist_lock);
 		rcu_read_unlock();
 		err = -ESRCH;
 		goto out;
 	}
 	mm = get_task_mm(task);
-	read_unlock(&tasklist_lock);
 	rcu_read_unlock();
 
 	err = -EINVAL;
Signature · 1 line
1.6.5.2