[1637] in SIPB_Linux_Development
Re: (fwd) Linux patch to fs/inode.c, mark #3
daemon@ATHENA.MIT.EDU (Theodore Y. Ts'o)
Mon Mar 31 15:34:50 1997
Date: Mon, 31 Mar 1997 15:34:32 -0500
From: "Theodore Y. Ts'o" <tytso@MIT.EDU>
To: Derek Atkins <warlord@MIT.EDU>
Cc: linux-dev@MIT.EDU
In-Reply-To: Derek Atkins's message of 31 Mar 1997 14:34:07 -0500,
<sjmd8sfq2j4.fsf@bobbi-harlow.MIT.EDU>
From: Derek Atkins <warlord@MIT.EDU>
Date: 31 Mar 1997 14:34:07 -0500
I spent a while with Ted today trying to track down the Linux-AFS
cache corruption.. We went through the AFS cache code, and he seemed
to think that there wasn't much of a problem. Under a system that is
heavily loaded, he thinks that there is a race condition in the Linux
kernel where it is possible for two opened files to get the same inode
structure.
Try this patch instead. I've refined the patch a bit more since my
meeting with Derek this morning.... As before, it's against 2.0.29
- Ted
--- fs/inode.c 1997/03/31 17:05:03 1.1
+++ fs/inode.c 1997/03/31 20:02:49
@@ -147,7 +147,8 @@
static inline void lock_inode(struct inode * inode)
{
- wait_on_inode(inode);
+ while (inode->i_lock)
+ wait_on_inode(inode);
inode->i_lock = 1;
}
@@ -173,8 +174,8 @@
{
struct wait_queue * wait;
+ lock_inode(inode);
truncate_inode_pages(inode, 0);
- wait_on_inode(inode);
if (IS_WRITABLE(inode)) {
if (inode->i_sb && inode->i_sb->dq_op)
inode->i_sb->dq_op->drop(inode);
@@ -186,6 +187,7 @@
nr_free_inodes++;
memset(inode,0,sizeof(*inode));
((volatile struct inode *) inode)->i_wait = wait;
+ wake_up(&inode->i_wait);
insert_inode_free(inode);
}
@@ -512,6 +514,7 @@
sleep_on(&inode_wait);
goto repeat;
}
+found_good:
if (best->i_lock) {
wait_on_inode(best);
goto repeat;
@@ -522,7 +525,7 @@
}
if (best->i_count)
goto repeat;
-found_good:
+
clear_inode(best);
best->i_count = 1;
best->i_nlink = 1;
