afs: Overhaul the callback handling

Overhaul the AFS callback handling by the following means:

 (1) Don't give up callback promises on vnodes that we are no longer using,
     rather let them just expire on the server or let the server break
     them.  This is actually more efficient for the server as the callback
     lookup is expensive if there are lots of extant callbacks.

 (2) Only give up the callback promises we have from a server when the
     server record is destroyed.  Then we can just give up *all* the
     callback promises on it in one go.

 (3) Servers can end up being shared between cells if cells are aliased, so
     don't add all the vnodes being backed by a particular server into a
     big FID-indexed tree on that server as there may be duplicates.

     Instead have each volume instance (~= superblock) register an interest
     in a server as it starts to make use of it and use this to allow the
     processor for callbacks from the server to find the superblock and
     thence the inode corresponding to the FID being broken by means of
     ilookup_nowait().

 (4) Rather than iterating over the entire callback list when a mass-break
     comes in from the server, maintain a counter of mass-breaks in
     afs_server (cb_seq) and make afs_validate() check it against the copy
     in afs_vnode.

     It would be nice not to have to take a read_lock whilst doing this,
     but that's tricky without using RCU.

 (5) Save a ref on the fileserver we're using for a call in the afs_call
     struct so that we can access its cb_s_break during call decoding.

 (6) Write-lock around callback and status storage in a vnode and read-lock
     around getattr so that we don't see the status mid-update.

This has the following consequences:

 (1) Data invalidation isn't seen until someone calls afs_validate() on a
     vnode.  Unfortunately, we need to use a key to query the server, but
     getting one from a background thread is tricky without caching loads
     of keys all over the place.

 (2) Mass invalidation isn't seen until someone calls afs_validate().

 (3) Callback breaking is going to hit the inode_hash_lock quite a bit.
     Could this be replaced with rcu_read_lock() since inodes are destroyed
     under RCU conditions.

Signed-off-by: David Howells <dhowells@redhat.com>

1 files changed, 60 insertions, 49 deletions

diff --git a/fs/afs/inode.c b/fs/afs/inode.c
index fbb441d25022..4822a2a50a61 100644
--- a/fs/afs/inode.c
+++ b/fs/afs/inode.c
@@ -23,11 +23,6 @@
 #include <linux/namei.h>
 #include "internal.h"
 
-struct afs_iget_data {
-	struct afs_fid		fid;
-	struct afs_volume	*volume;	/* volume on which resides */
-};
-
 static const struct inode_operations afs_symlink_inode_operations = {
 	.get_link	= page_get_link,
 	.listxattr	= afs_listxattr,
@@ -39,6 +34,7 @@ static const struct inode_operations afs_symlink_inode_operations = {
 static int afs_inode_map_status(struct afs_vnode *vnode, struct key *key)
 {
 	struct inode *inode = AFS_VNODE_TO_I(vnode);
+	bool changed;
 
 	_debug("FS: ft=%d lk=%d sz=%llu ver=%Lu mod=%hu",
 	       vnode->status.type,
@@ -47,6 +43,8 @@ static int afs_inode_map_status(struct afs_vnode *vnode, struct key *key)
 	       vnode->status.data_version,
 	       vnode->status.mode);
 
+	read_seqlock_excl(&vnode->cb_lock);
+
 	switch (vnode->status.type) {
 	case AFS_FTYPE_FILE:
 		inode->i_mode	= S_IFREG | vnode->status.mode;
@@ -63,9 +61,7 @@ static int afs_inode_map_status(struct afs_vnode *vnode, struct key *key)
 		if ((vnode->status.mode & 0777) == 0644) {
 			inode->i_flags |= S_AUTOMOUNT;
 
-			spin_lock(&vnode->lock);
 			set_bit(AFS_VNODE_MOUNTPOINT, &vnode->flags);
-			spin_unlock(&vnode->lock);
 
 			inode->i_mode	= S_IFDIR | 0555;
 			inode->i_op	= &afs_mntpt_inode_operations;
@@ -78,13 +74,11 @@ static int afs_inode_map_status(struct afs_vnode *vnode, struct key *key)
 		break;
 	default:
 		printk("kAFS: AFS vnode with undefined type\n");
+		read_sequnlock_excl(&vnode->cb_lock);
 		return -EBADMSG;
 	}
 
-#ifdef CONFIG_AFS_FSCACHE
-	if (vnode->status.size != inode->i_size)
-		fscache_attr_changed(vnode->cache);
-#endif
+	changed = (vnode->status.size != inode->i_size);
 
 	set_nlink(inode, vnode->status.nlink);
 	inode->i_uid		= vnode->status.owner;
@@ -97,13 +91,20 @@ static int afs_inode_map_status(struct afs_vnode *vnode, struct key *key)
 	inode->i_generation	= vnode->fid.unique;
 	inode->i_version	= vnode->status.data_version;
 	inode->i_mapping->a_ops	= &afs_fs_aops;
+
+	read_sequnlock_excl(&vnode->cb_lock);
+
+#ifdef CONFIG_AFS_FSCACHE
+	if (changed)
+		fscache_attr_changed(vnode->cache);
+#endif
 	return 0;
 }
 
 /*
  * iget5() comparator
  */
-static int afs_iget5_test(struct inode *inode, void *opaque)
+int afs_iget5_test(struct inode *inode, void *opaque)
 {
 	struct afs_iget_data *data = opaque;
 
@@ -237,8 +238,7 @@ struct inode *afs_iget(struct super_block *sb, struct key *key,
 
 	if (!status) {
 		/* it's a remotely extant inode */
-		set_bit(AFS_VNODE_CB_BROKEN, &vnode->flags);
-		ret = afs_vnode_fetch_status(vnode, NULL, key);
+		ret = afs_vnode_fetch_status(vnode, NULL, key, true);
 		if (ret < 0)
 			goto bad_inode;
 	} else {
@@ -249,16 +249,16 @@ struct inode *afs_iget(struct super_block *sb, struct key *key,
 			/* it's a symlink we just created (the fileserver
 			 * didn't give us a callback) */
 			vnode->cb_version = 0;
-			vnode->cb_expiry = 0;
 			vnode->cb_type = 0;
-			vnode->cb_expires = ktime_get_real_seconds();
+			vnode->cb_expires_at = 0;
 		} else {
 			vnode->cb_version = cb->version;
-			vnode->cb_expiry = cb->expiry;
 			vnode->cb_type = cb->type;
-			vnode->cb_expires = vnode->cb_expiry +
-				ktime_get_real_seconds();
+			vnode->cb_expires_at = cb->expiry;
+			set_bit(AFS_VNODE_CB_PROMISED, &vnode->flags);
 		}
+
+		vnode->cb_expires_at += ktime_get_real_seconds();
 	}
 
 	/* set up caching before mapping the status, as map-status reads the
@@ -320,25 +320,34 @@ void afs_zap_data(struct afs_vnode *vnode)
  */
 int afs_validate(struct afs_vnode *vnode, struct key *key)
 {
+	time64_t now = ktime_get_real_seconds();
+	bool valid = false;
 	int ret;
 
 	_enter("{v={%x:%u} fl=%lx},%x",
 	       vnode->fid.vid, vnode->fid.vnode, vnode->flags,
 	       key_serial(key));
 
-	if (vnode->cb_promised &&
-	    !test_bit(AFS_VNODE_CB_BROKEN, &vnode->flags) &&
-	    !test_bit(AFS_VNODE_MODIFIED, &vnode->flags) &&
-	    !test_bit(AFS_VNODE_ZAP_DATA, &vnode->flags)) {
-		if (vnode->cb_expires < ktime_get_real_seconds() + 10) {
-			_debug("callback expired");
-			set_bit(AFS_VNODE_CB_BROKEN, &vnode->flags);
-		} else {
-			goto valid;
+	/* Quickly check the callback state.  Ideally, we'd use read_seqbegin
+	 * here, but we have no way to pass the net namespace to the RCU
+	 * cleanup for the server record.
+	 */
+	read_seqlock_excl(&vnode->cb_lock);
+
+	if (test_bit(AFS_VNODE_CB_PROMISED, &vnode->flags)) {
+		if (vnode->cb_s_break != vnode->cb_interest->server->cb_s_break) {
+			vnode->cb_s_break = vnode->cb_interest->server->cb_s_break;
+		} else if (!test_bit(AFS_VNODE_DIR_MODIFIED, &vnode->flags) &&
+			   !test_bit(AFS_VNODE_ZAP_DATA, &vnode->flags) &&
+			   vnode->cb_expires_at - 10 > now) {
+				valid = true;
 		}
+	} else if (test_bit(AFS_VNODE_DELETED, &vnode->flags)) {
+		valid = true;
 	}
 
-	if (test_bit(AFS_VNODE_DELETED, &vnode->flags))
+	read_sequnlock_excl(&vnode->cb_lock);
+	if (valid)
 		goto valid;
 
 	mutex_lock(&vnode->validate_lock);
@@ -347,12 +356,16 @@ int afs_validate(struct afs_vnode *vnode, struct key *key)
 	 * a new promise - note that if the (parent) directory's metadata was
 	 * changed then the security may be different and we may no longer have
 	 * access */
-	if (!vnode->cb_promised ||
-	    test_bit(AFS_VNODE_CB_BROKEN, &vnode->flags)) {
+	if (!test_bit(AFS_VNODE_CB_PROMISED, &vnode->flags)) {
 		_debug("not promised");
-		ret = afs_vnode_fetch_status(vnode, NULL, key);
-		if (ret < 0)
+		ret = afs_vnode_fetch_status(vnode, NULL, key, false);
+		if (ret < 0) {
+			if (ret == -ENOENT) {
+				set_bit(AFS_VNODE_DELETED, &vnode->flags);
+				ret = -ESTALE;
+			}
 			goto error_unlock;
+		}
 		_debug("new promise [fl=%lx]", vnode->flags);
 	}
 
@@ -367,7 +380,7 @@ int afs_validate(struct afs_vnode *vnode, struct key *key)
 	if (test_and_clear_bit(AFS_VNODE_ZAP_DATA, &vnode->flags))
 		afs_zap_data(vnode);
 
-	clear_bit(AFS_VNODE_MODIFIED, &vnode->flags);
+	clear_bit(AFS_VNODE_DIR_MODIFIED, &vnode->flags);
 	mutex_unlock(&vnode->validate_lock);
 valid:
 	_leave(" = 0");
@@ -386,10 +399,17 @@ int afs_getattr(const struct path *path, struct kstat *stat,
 		u32 request_mask, unsigned int query_flags)
 {
 	struct inode *inode = d_inode(path->dentry);
+	struct afs_vnode *vnode = AFS_FS_I(inode);
+	int seq = 0;
 
 	_enter("{ ino=%lu v=%u }", inode->i_ino, inode->i_generation);
 
-	generic_fillattr(inode, stat);
+	do {
+		read_seqbegin_or_lock(&vnode->cb_lock, &seq);
+		generic_fillattr(inode, stat);
+	} while (need_seqretry(&vnode->cb_lock, seq));
+
+	done_seqretry(&vnode->cb_lock, seq);
 	return 0;
 }
 
@@ -416,13 +436,10 @@ void afs_evict_inode(struct inode *inode)
 
 	vnode = AFS_FS_I(inode);
 
-	_enter("{%x:%u.%d} v=%u x=%u t=%u }",
+	_enter("{%x:%u.%d}",
 	       vnode->fid.vid,
 	       vnode->fid.vnode,
-	       vnode->fid.unique,
-	       vnode->cb_version,
-	       vnode->cb_expiry,
-	       vnode->cb_type);
+	       vnode->fid.unique);
 
 	_debug("CLEAR INODE %p", inode);
 
@@ -431,18 +448,12 @@ void afs_evict_inode(struct inode *inode)
 	truncate_inode_pages_final(&inode->i_data);
 	clear_inode(inode);
 
-	afs_give_up_callback(vnode);
-
-	if (vnode->server) {
-		spin_lock(&vnode->server->fs_lock);
-		rb_erase(&vnode->server_rb, &vnode->server->fs_vnodes);
-		spin_unlock(&vnode->server->fs_lock);
-		afs_put_server(afs_i2net(inode), vnode->server);
-		vnode->server = NULL;
+	if (vnode->cb_interest) {
+		afs_put_cb_interest(afs_i2net(inode), vnode->cb_interest);
+		vnode->cb_interest = NULL;
 	}
 
 	ASSERT(list_empty(&vnode->writebacks));
-	ASSERT(!vnode->cb_promised);
 
 #ifdef CONFIG_AFS_FSCACHE
 	fscache_relinquish_cookie(vnode->cache, 0);