[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <20190430030914.GF23075@ZenIV.linux.org.uk>
Date: Tue, 30 Apr 2019 04:09:14 +0100
From: Al Viro <viro@...iv.linux.org.uk>
To: Linus Torvalds <torvalds@...ux-foundation.org>
Cc: linux-fsdevel <linux-fsdevel@...r.kernel.org>,
Linux List Kernel Mailing <linux-kernel@...r.kernel.org>
Subject: Re: [RFC][PATCHSET] sorting out RCU-delayed stuff in
->destroy_inode()
On Tue, Apr 16, 2019 at 11:01:16AM -0700, Linus Torvalds wrote:
> On Tue, Apr 16, 2019 at 10:49 AM Al Viro <viro@...iv.linux.org.uk> wrote:
> >
> > 83 files changed, 241 insertions(+), 516 deletions(-)
>
> I think this single line is pretty convincing on its own. Ignoring
> docs and fs/inode.c, we have
>
> 80 files changed, 190 insertions(+), 494 deletions(-)
>
> IOW, just over 300 lines of boiler plate code removed.
>
> The additions are
>
> - Ten more lines of actual code in fs/inode.c (and that's not
> actually added complexity, it looks simpler if anything - most of it
> is the new "i_callback()" helper function)
>
> - 19 lines of doc updates.
>
> So it absolutely looks fine to me.
>
> I only skimmed through the actual filesystem (and one networking)
> patches, but they looked like trivial conversions to a better
> interface.
... except that this callback can (and always could) get executed after
freeing struct super_block. So we can't just dereference ->i_sb->s_op
and expect to survive; the table ->s_op pointed to will still be there,
but ->i_sb might very well have been freed, with all its contents overwritten.
We need to copy the callback into struct inode itself, unfortunately.
The following incremental fixes it; I'm going to fold it into the first
commit in there.
diff --git a/Documentation/filesystems/porting b/Documentation/filesystems/porting
index 9d80f9e0855e..b8d3ddd8b8db 100644
--- a/Documentation/filesystems/porting
+++ b/Documentation/filesystems/porting
@@ -655,3 +655,11 @@ in your dentry operations instead.
* if ->free_inode() is non-NULL, it gets scheduled by call_rcu()
* combination of NULL ->destroy_inode and NULL ->free_inode is
treated as NULL/free_inode_nonrcu, to preserve the compatibility.
+
+ Note that the callback (be it via ->free_inode() or explicit call_rcu()
+ in ->destroy_inode()) is *NOT* ordered wrt superblock destruction;
+ as the matter of fact, the superblock and all associated structures
+ might be already gone. The filesystem driver is guaranteed to be still
+ there, but that's it. Freeing memory in the callback is fine; doing
+ more than that is possible, but requires a lot of care and is best
+ avoided.
diff --git a/fs/inode.c b/fs/inode.c
index fb45590d284e..855dad43b11d 100644
--- a/fs/inode.c
+++ b/fs/inode.c
@@ -164,6 +164,7 @@ int inode_init_always(struct super_block *sb, struct inode *inode)
inode->i_wb_frn_avg_time = 0;
inode->i_wb_frn_history = 0;
#endif
+ inode->free_inode = sb->s_op->free_inode;
if (security_inode_alloc(inode))
goto out;
@@ -211,8 +212,8 @@ EXPORT_SYMBOL(free_inode_nonrcu);
static void i_callback(struct rcu_head *head)
{
struct inode *inode = container_of(head, struct inode, i_rcu);
- if (inode->i_sb->s_op->free_inode)
- inode->i_sb->s_op->free_inode(inode);
+ if (inode->free_inode)
+ inode->free_inode(inode);
else
free_inode_nonrcu(inode);
}
diff --git a/include/linux/fs.h b/include/linux/fs.h
index 2e9b9f87caca..5ed6b39e588e 100644
--- a/include/linux/fs.h
+++ b/include/linux/fs.h
@@ -718,6 +718,7 @@ struct inode {
#endif
void *i_private; /* fs or device private pointer */
+ void (*free_inode)(struct inode *);
} __randomize_layout;
static inline unsigned int i_blocksize(const struct inode *node)
Powered by blists - more mailing lists