[PATCH v5 40/40] 9p: Use netfslib read/write_iter

Thu May 23 00:44:17 PDT 2024

Hi Andrea,

Note that there are changes to the netfslib write-side upstream and you might
also want to apply the attached.

In https://bugs.launchpad.net/ubuntu/+source/autopkgtest/+bug/2056461 you say:

| It seems that kernel 6.8 introduced a regression in the 9pfs related to
| caching and netfslib, that can cause some user-space apps to read content
| from files that is not up-to-date (when they are used in a producer/consumer
| fashion).

Can you clarify how these files are being used?

David
---
commit 39302c160390441ed5b4f4f7ad480c44eddf0962
Author: David Howells <dhowells at redhat.com>
Date:   Wed May 22 17:30:22 2024 +0100

    netfs, 9p: Fix race between umount and async request completion
    
    There's a problem in 9p's interaction with netfslib whereby a crash occurs
    because the 9p_fid structs get forcibly destroyed during client teardown
    (without paying attention to their refcounts) before netfslib has finished
    with them.  However, it's not a simple case of deferring the clunking that
    p9_fid_put() does as that requires the client.
    
    The problem is that netfslib has to unlock pages and clear the IN_PROGRESS
    flag before destroying the objects involved - including the pid - and, in
    any case, nothing checks to see if writeback completed barring looking at
    the page flags.
    
    Fix this by keeping a count of outstanding I/O requests (of any type) and
    waiting for it to quiesce during inode eviction.
    
    Signed-off-by: David Howells <dhowells at redhat.com>
    cc: Eric Van Hensbergen <ericvh at kernel.org>
    cc: Latchesar Ionkov <lucho at ionkov.net>
    cc: Dominique Martinet <asmadeus at codewreck.org>
    cc: Christian Schoenebeck <linux_oss at crudebyte.com>
    cc: Jeff Layton <jlayton at kernel.org>
    cc: Steve French <sfrench at samba.org>
    cc: v9fs at lists.linux.dev
    cc: linux-afs at lists.infradead.org
    cc: linux-cifs at vger.kernel.org
    cc: netfs at lists.linux.dev
    cc: linux-fsdevel at vger.kernel.org

diff --git a/fs/9p/vfs_inode.c b/fs/9p/vfs_inode.c
index 8c9a896d691e..57cfa9f65046 100644
--- a/fs/9p/vfs_inode.c
+++ b/fs/9p/vfs_inode.c
@@ -354,6 +354,7 @@ void v9fs_evict_inode(struct inode *inode)
 		version = cpu_to_le32(v9inode->qid.version);
 		netfs_clear_inode_writeback(inode, &version);
 
+		netfs_wait_for_outstanding_io(inode);
 		clear_inode(inode);
 		filemap_fdatawrite(&inode->i_data);
 
@@ -361,8 +362,10 @@ void v9fs_evict_inode(struct inode *inode)
 		if (v9fs_inode_cookie(v9inode))
 			fscache_relinquish_cookie(v9fs_inode_cookie(v9inode), false);
 #endif
-	} else
+	} else {
+		netfs_wait_for_outstanding_io(inode);
 		clear_inode(inode);
+	}
 }
 
 struct inode *
diff --git a/fs/afs/inode.c b/fs/afs/inode.c
index 94fc049aff58..c831e711a4ac 100644
--- a/fs/afs/inode.c
+++ b/fs/afs/inode.c
@@ -652,6 +652,7 @@ void afs_evict_inode(struct inode *inode)
 
 	afs_set_cache_aux(vnode, &aux);
 	netfs_clear_inode_writeback(inode, &aux);
+	netfs_wait_for_outstanding_io(inode);
 	clear_inode(inode);
 
 	while (!list_empty(&vnode->wb_keys)) {
diff --git a/fs/netfs/objects.c b/fs/netfs/objects.c
index c90d482b1650..f4a642727479 100644
--- a/fs/netfs/objects.c
+++ b/fs/netfs/objects.c
@@ -72,6 +72,7 @@ struct netfs_io_request *netfs_alloc_request(struct address_space *mapping,
 		}
 	}
 
+	atomic_inc(&ctx->io_count);
 	trace_netfs_rreq_ref(rreq->debug_id, 1, netfs_rreq_trace_new);
 	netfs_proc_add_rreq(rreq);
 	netfs_stat(&netfs_n_rh_rreq);
@@ -124,6 +125,7 @@ static void netfs_free_request(struct work_struct *work)
 {
 	struct netfs_io_request *rreq =
 		container_of(work, struct netfs_io_request, work);
+	struct netfs_inode *ictx = netfs_inode(rreq->inode);
 	unsigned int i;
 
 	trace_netfs_rreq(rreq, netfs_rreq_trace_free);
@@ -142,6 +144,9 @@ static void netfs_free_request(struct work_struct *work)
 		}
 		kvfree(rreq->direct_bv);
 	}
+
+	if (atomic_dec_and_test(&ictx->io_count))
+		wake_up_var(&ictx->io_count);
 	call_rcu(&rreq->rcu, netfs_free_request_rcu);
 }
 
diff --git a/fs/smb/client/cifsfs.c b/fs/smb/client/cifsfs.c
index ec5b639f421a..21c9e173ea9a 100644
--- a/fs/smb/client/cifsfs.c
+++ b/fs/smb/client/cifsfs.c
@@ -435,6 +435,7 @@ cifs_evict_inode(struct inode *inode)
 	if (inode->i_state & I_PINNING_NETFS_WB)
 		cifs_fscache_unuse_inode_cookie(inode, true);
 	cifs_fscache_release_inode_cookie(inode);
+	netfs_wait_for_outstanding_io(inode);
 	clear_inode(inode);
 }
 
diff --git a/include/linux/netfs.h b/include/linux/netfs.h
index d2d291a9cdad..3ca3906bb8da 100644
--- a/include/linux/netfs.h
+++ b/include/linux/netfs.h
@@ -68,6 +68,7 @@ struct netfs_inode {
 	loff_t			remote_i_size;	/* Size of the remote file */
 	loff_t			zero_point;	/* Size after which we assume there's no data
 						 * on the server */
+	atomic_t		io_count;	/* Number of outstanding reqs */
 	unsigned long		flags;
 #define NETFS_ICTX_ODIRECT	0		/* The file has DIO in progress */
 #define NETFS_ICTX_UNBUFFERED	1		/* I/O should not use the pagecache */
@@ -474,6 +475,7 @@ static inline void netfs_inode_init(struct netfs_inode *ctx,
 	ctx->remote_i_size = i_size_read(&ctx->inode);
 	ctx->zero_point = LLONG_MAX;
 	ctx->flags = 0;
+	atomic_set(&ctx->io_count, 0);
 #if IS_ENABLED(CONFIG_FSCACHE)
 	ctx->cache = NULL;
 #endif
@@ -517,4 +519,20 @@ static inline struct fscache_cookie *netfs_i_cookie(struct netfs_inode *ctx)
 #endif
 }
 
+/**
+ * netfs_wait_for_outstanding_io - Wait for outstanding I/O to complete
+ * @ctx: The netfs inode to wait on
+ *
+ * Wait for outstanding I/O requests of any type to complete.  This is intended
+ * to be called from inode eviction routines.  This makes sure that any
+ * resources held by those requests are cleaned up before we let the inode get
+ * cleaned up.
+ */
+static inline void netfs_wait_for_outstanding_io(struct inode *inode)
+{
+	struct netfs_inode *ictx = netfs_inode(inode);
+
+	wait_var_event(&ictx->io_count, atomic_read(&ictx->io_count) == 0);
+}
+
 #endif /* _LINUX_NETFS_H */


