sysfs: fix deadlock

On Thu, Sep 11, 2008 at 10:27:10AM +0200, Ingo Molnar wrote:

> and it's working fine on most boxes. One testbox found this new locking
> scenario:
>
> PM: Adding info for No Bus:vcsa7
> EDAC DEBUG: MC0: i82860_check()
>
> =======================================================
> [ INFO: possible circular locking dependency detected ]
> 2.6.27-rc6-tip #1
> -------------------------------------------------------
> X/4873 is trying to acquire lock:
>  (&bb->mutex){--..}, at: [<c020ba20>] mmap+0x40/0xa0
>
> but task is already holding lock:
>  (&mm->mmap_sem){----}, at: [<c0125a1e>] sys_mmap2+0x8e/0xc0
>
> which lock already depends on the new lock.
>
>
> the existing dependency chain (in reverse order) is:
>
> -> #1 (&mm->mmap_sem){----}:
>        [<c017dc96>] validate_chain+0xa96/0xf50
>        [<c017ef2b>] __lock_acquire+0x2cb/0x5b0
>        [<c017f299>] lock_acquire+0x89/0xc0
>        [<c01aa8fb>] might_fault+0x6b/0x90
>        [<c040b618>] copy_to_user+0x38/0x60
>        [<c020bcfb>] read+0xfb/0x170
>        [<c01c09a5>] vfs_read+0x95/0x110
>        [<c01c1443>] sys_pread64+0x63/0x80
>        [<c012146f>] sysenter_do_call+0x12/0x43
>        [<ffffffff>] 0xffffffff
>
> -> #0 (&bb->mutex){--..}:
>        [<c017d8b7>] validate_chain+0x6b7/0xf50
>        [<c017ef2b>] __lock_acquire+0x2cb/0x5b0
>        [<c017f299>] lock_acquire+0x89/0xc0
>        [<c0d6f2ab>] __mutex_lock_common+0xab/0x3c0
>        [<c0d6f698>] mutex_lock_nested+0x38/0x50
>        [<c020ba20>] mmap+0x40/0xa0
>        [<c01b111e>] mmap_region+0x14e/0x450
>        [<c01b170f>] do_mmap_pgoff+0x2ef/0x310
>        [<c0125a3d>] sys_mmap2+0xad/0xc0
>        [<c012146f>] sysenter_do_call+0x12/0x43
>        [<ffffffff>] 0xffffffff
>
> other info that might help us debug this:
>
> 1 lock held by X/4873:
>  #0:  (&mm->mmap_sem){----}, at: [<c0125a1e>] sys_mmap2+0x8e/0xc0
>
> stack backtrace:
> Pid: 4873, comm: X Not tainted 2.6.27-rc6-tip #1
>  [<c017cd09>] print_circular_bug_tail+0x79/0xc0
>  [<c017d8b7>] validate_chain+0x6b7/0xf50
>  [<c017a5b5>] ? trace_hardirqs_off_caller+0x15/0xb0
>  [<c017ef2b>] __lock_acquire+0x2cb/0x5b0
>  [<c017f299>] lock_acquire+0x89/0xc0
>  [<c020ba20>] ? mmap+0x40/0xa0
>  [<c0d6f2ab>] __mutex_lock_common+0xab/0x3c0
>  [<c020ba20>] ? mmap+0x40/0xa0
>  [<c0d6f698>] mutex_lock_nested+0x38/0x50
>  [<c020ba20>] ? mmap+0x40/0xa0
>  [<c020ba20>] mmap+0x40/0xa0
>  [<c01b111e>] mmap_region+0x14e/0x450
>  [<c01afb88>] ? arch_get_unmapped_area_topdown+0xf8/0x160
>  [<c01b170f>] do_mmap_pgoff+0x2ef/0x310
>  [<c0125a3d>] sys_mmap2+0xad/0xc0
>  [<c012146f>] sysenter_do_call+0x12/0x43
>  [<c0120000>] ? __switch_to+0x130/0x220
>  =======================
> evbug.c: Event. Dev: input3, Type: 20, Code: 0, Value: 500
> warning: `sudo' uses deprecated v2 capabilities in a way that may be insecure.
>
> i've attached the config.
>
> at first sight it looks like a genuine bug in fs/sysfs/bin.c?

Yes, it is a real bug by the looks. bin.c takes bb->mutex under mmap_sem
when it is mmapped, and then does its copy_*_user under bb->mutex too.

Here is a basic fix for the sysfs lor.


From: Nick Piggin <npiggin@suse.de>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>

fs/sysfs/bin.c

@@ -61,6 +61,7 @@ read(struct file *file, char __user *userbuf, size_t bytes, loff_t *off)
 	int size = dentry->d_inode->i_size;
 	loff_t offs = *off;
 	int count = min_t(size_t, bytes, PAGE_SIZE);
+	char *temp;
 
 	if (size) {
 		if (offs > size)
@@ -69,23 +70,33 @@ read(struct file *file, char __user *userbuf, size_t bytes, loff_t *off)
 			count = size - offs;
 	}
 
+	temp = kmalloc(count, GFP_KERNEL);
+	if (!temp)
+		return -ENOMEM;
+
 	mutex_lock(&bb->mutex);
 
 	count = fill_read(dentry, bb->buffer, offs, count);
-	if (count < 0)
-		goto out_unlock;
+	if (count < 0) {
+		mutex_unlock(&bb->mutex);
+		goto out_free;
+	}
 
-	if (copy_to_user(userbuf, bb->buffer, count)) {
+	memcpy(temp, bb->buffer, count);
+
+	mutex_unlock(&bb->mutex);
+
+	if (copy_to_user(userbuf, temp, count)) {
 		count = -EFAULT;
-		goto out_unlock;
+		goto out_free;
 	}
 
 	pr_debug("offs = %lld, *off = %lld, count = %d\n", offs, *off, count);
 
 	*off = offs + count;
 
- out_unlock:
-	mutex_unlock(&bb->mutex);
+ out_free:
+	kfree(temp);
 	return count;
 }
 
@@ -118,6 +129,7 @@ static ssize_t write(struct file *file, const char __user *userbuf,
 	int size = dentry->d_inode->i_size;
 	loff_t offs = *off;
 	int count = min_t(size_t, bytes, PAGE_SIZE);
+	char *temp;
 
 	if (size) {
 		if (offs > size)
@@ -126,19 +138,27 @@ static ssize_t write(struct file *file, const char __user *userbuf,
 			count = size - offs;
 	}
 
-	mutex_lock(&bb->mutex);
+	temp = kmalloc(count, GFP_KERNEL);
+	if (!temp)
+		return -ENOMEM;
 
-	if (copy_from_user(bb->buffer, userbuf, count)) {
+	if (copy_from_user(temp, userbuf, count)) {
 		count = -EFAULT;
-		goto out_unlock;
+		goto out_free;
 	}
 
+	mutex_lock(&bb->mutex);
+
+	memcpy(bb->buffer, temp, count);
+
 	count = flush_write(dentry, bb->buffer, offs, count);
+	mutex_unlock(&bb->mutex);
+
 	if (count > 0)
 		*off = offs + count;
 
- out_unlock:
-	mutex_unlock(&bb->mutex);
+out_free:
+	kfree(temp);
 	return count;
 }