Message-id: <20080915053937.GG4090@webber.adilger.int>
Date:	Sun, 14 Sep 2008 22:39:37 -0700
From:	Andreas Dilger <adilger@....com>
To:	Abhijit Paithankar <apaithan@...mai.com>
Cc:	linux-fsdevel@...r.kernel.org, linux-ext4@...r.kernel.org
Subject: Re: [RFC PATCH] Filesystem Journal Notifications

On Sep 14, 2008  19:31 -0700, Andreas Dilger wrote:
> We had a much more flexible journal commit notification mechanism.  The
> caller would register a callback with the journal layer that contained
> a caller-private callback function and callback data.  This avoided
> exposing the journal transaction IDs outside of the filesystem, and the
> caller could use any identifier desired.
> 
> There was also no limit on the number/type of callbacks registered with
> the filesystem.  The basic data structure was a linked list of callbacks
> registered on the transaction:
> 
> struct journal_callback {
>        struct list_head jcb_list;
>        void (*jcb_func)(struct journal_callback *jcb, int error);
>        /* caller data goes here */
> };
> 
> and in journal_commit_transaction() the callbacks are called:
> 
>                list_for_each_entry_safe(jcb, n, &commit_transaction->t_jcb,
> 					jcb_list) {
>                        list_del_init(&jcb->jcb_list);
>                        jcb->jcb_func(jcb, error);
>                }

Here is the patch (against 2.6.18 kernels, but I don't think anything has
changed in this area in a long time):


Index: linux-2.6/include/linux/jbd.h
===================================================================
--- linux-2.6.orig/include/linux/jbd.h	2006-07-15 16:08:35.000000000 +0800
+++ linux-2.6/include/linux/jbd.h	2006-07-15 16:13:01.000000000 +0800
@@ -356,6 +356,27 @@ static inline void jbd_unlock_bh_journal
 	bit_spin_unlock(BH_JournalHead, &bh->b_state);
 }
 
+#define HAVE_JOURNAL_CALLBACK_STATUS
+/**
+ *   struct journal_callback - Base structure for callback information
+ *   @jcb_list: list information for other callbacks attached to the same handle
+ *   @jcb_func: Function to call with this callback structure
+ *
+ *   This struct is a 'seed' structure for a using with your own callback
+ *   structs. If you are using callbacks you must allocate one of these
+ *   or another struct of your own definition which has this struct 
+ *   as it's first element and pass it to journal_callback_set().
+ *
+ *   This is used internally by jbd to maintain callback information.
+ *
+ *   See journal_callback_set for more information.
+ **/
+struct journal_callback {
+	struct list_head jcb_list;		/* t_jcb_lock */
+	void (*jcb_func)(struct journal_callback *jcb, int error);
+	/* caller data goes here */
+};
+
 struct jbd_revoke_table_s;
 
 /**
@@ -364,6 +385,7 @@ struct jbd_revoke_table_s;
  * @h_transaction: Which compound transaction is this update a part of?
  * @h_buffer_credits: Number of remaining buffers we are allowed to dirty.
  * @h_ref: Reference count on this handle
+ * @h_jcb: List of application registered callbacks for this handle.
  * @h_err: Field for caller's use to track errors through large fs operations
  * @h_sync: flag for sync-on-close
  * @h_jdata: flag to force data journaling
@@ -389,6 +411,13 @@ struct handle_s 
 	/* operations */
 	int			h_err;
 
+	/*
+	 * List of application registered callbacks for this handle. The
+	 * function(s) will be called after the transaction that this handle is
+	 * part of has been committed to disk. [t_jcb_lock]
+	 */
+	struct list_head	h_jcb;
+
 	/* Flags [no locking] */
 	unsigned int	h_sync:		1;	/* sync-on-close */
 	unsigned int	h_jdata:	1;	/* force data journaling */
@@ -430,6 +459,8 @@ struct handle_s 
  *    j_state_lock
  *    ->j_list_lock			(journal_unmap_buffer)
  *
+ *    t_handle_lock
+ *    ->t_jcb_lock
  */
 
 struct transaction_s 
@@ -559,6 +590,15 @@ struct transaction_s 
 	 */
 	int t_handle_count;
 
+	/*
+	 * Protects the callback list
+	 */
+	spinlock_t		t_jcb_lock;
+	/*
+	 * List of registered callback functions for this transaction.
+	 * Called when the transaction is committed. [t_jcb_lock]
+	 */
+	struct list_head	t_jcb;
 };
 
 /**
@@ -906,6 +946,10 @@ extern void	 journal_invalidatepage(jour
 extern int	 journal_try_to_free_buffers(journal_t *, struct page *, gfp_t);
 extern int	 journal_stop(handle_t *);
 extern int	 journal_flush (journal_t *);
+extern void	 journal_callback_set(handle_t *handle,
+				      void (*fn)(struct journal_callback *,int),
+				      struct journal_callback *jcb);
+
 extern void	 journal_lock_updates (journal_t *);
 extern void	 journal_unlock_updates (journal_t *);
 
Index: linux-2.6/fs/jbd/checkpoint.c
===================================================================
--- linux-2.6.orig/fs/jbd/checkpoint.c	2006-07-15 16:08:36.000000000 +0800
+++ linux-2.6/fs/jbd/checkpoint.c	2006-07-15 16:13:01.000000000 +0800
@@ -688,6 +688,7 @@ void __journal_drop_transaction(journal_
 	J_ASSERT(transaction->t_checkpoint_list == NULL);
 	J_ASSERT(transaction->t_checkpoint_io_list == NULL);
 	J_ASSERT(transaction->t_updates == 0);
+	J_ASSERT(list_empty(&transaction->t_jcb));
 	J_ASSERT(journal->j_committing_transaction != transaction);
 	J_ASSERT(journal->j_running_transaction != transaction);
 
Index: linux-2.6/fs/jbd/commit.c
===================================================================
--- linux-2.6.orig/fs/jbd/commit.c	2006-07-15 16:08:36.000000000 +0800
+++ linux-2.6/fs/jbd/commit.c	2006-07-15 16:13:01.000000000 +0800
@@ -708,6 +708,21 @@ wait_for_iobuf:
            transaction can be removed from any checkpoint list it was on
            before. */
 
+	/* Call any callbacks that had been registered for handles in this
+	 * transaction.  It is up to the callback to free any allocated
+	 * memory.  No locking is required, since this is the only process
+	 * that is processing this transaction anymore. */
+	if (!list_empty(&commit_transaction->t_jcb)) {
+		struct journal_callback *jcb, *n;
+		int error = is_journal_aborted(journal);
+
+		list_for_each_entry_safe(jcb, n, &commit_transaction->t_jcb,
+					 jcb_list) {
+			list_del_init(&jcb->jcb_list);
+			jcb->jcb_func(jcb, error);
+		}
+	}
+
 	jbd_debug(3, "JBD: commit phase 7\n");
 
 	J_ASSERT(commit_transaction->t_sync_datalist == NULL);
Index: linux-2.6/fs/jbd/journal.c
===================================================================
--- linux-2.6.orig/fs/jbd/journal.c	2006-07-15 16:08:36.000000000 +0800
+++ linux-2.6/fs/jbd/journal.c	2006-07-15 16:13:01.000000000 +0800
@@ -58,6 +58,7 @@ EXPORT_SYMBOL(journal_sync_buffer);
 #endif
 EXPORT_SYMBOL(journal_flush);
 EXPORT_SYMBOL(journal_revoke);
+EXPORT_SYMBOL(journal_callback_set);
 
 EXPORT_SYMBOL(journal_init_dev);
 EXPORT_SYMBOL(journal_init_inode);
Index: linux-2.6/fs/jbd/transaction.c
===================================================================
--- linux-2.6.orig/fs/jbd/transaction.c	2006-07-15 16:08:35.000000000 +0800
+++ linux-2.6/fs/jbd/transaction.c	2006-07-15 16:13:01.000000000 +0800
@@ -50,6 +50,8 @@ get_transaction(journal_t *journal, tran
 	transaction->t_state = T_RUNNING;
 	transaction->t_tid = journal->j_transaction_sequence++;
 	transaction->t_expires = jiffies + journal->j_commit_interval;
+	INIT_LIST_HEAD(&transaction->t_jcb);
+	spin_lock_init(&transaction->t_jcb_lock);
 	spin_lock_init(&transaction->t_handle_lock);
 
 	/* Set up the commit timer for the new transaction. */
@@ -241,6 +243,7 @@ static handle_t *new_handle(int nblocks)
 	memset(handle, 0, sizeof(*handle));
 	handle->h_buffer_credits = nblocks;
 	handle->h_ref = 1;
+	INIT_LIST_HEAD(&handle->h_jcb);
 
 	return handle;
 }
@@ -1291,6 +1294,36 @@ drop:
 }
 
 /**
+ * void journal_callback_set() -  Register a callback function for this handle.
+ * @handle: handle to attach the callback to.
+ * @func: function to callback.
+ * @jcb:  structure with additional information required by func() , and
+ *        some space for jbd internal information.
+ * 
+ * The function will be called when the transaction that this handle is
+ * part of has been committed to disk with the original callback data
+ * struct and the error status of the journal as parameters.  There is no
+ * guarantee of ordering between handles within a single transaction, nor
+ * between callbacks registered on the same handle.
+ *
+ * The caller is responsible for allocating the journal_callback struct.
+ * This is to allow the caller to add as much extra data to the callback
+ * as needed, but reduce the overhead of multiple allocations.  The caller
+ * allocated struct must start with a struct journal_callback at offset 0,
+ * and has the caller-specific data afterwards.
+ */
+void journal_callback_set(handle_t *handle,
+			  void (*func)(struct journal_callback *jcb, int error),
+			  struct journal_callback *jcb)
+{
+	jcb->jcb_func = func;
+	spin_lock(&handle->h_transaction->t_jcb_lock);
+	list_add_tail(&jcb->jcb_list, &handle->h_jcb);
+	spin_unlock(&handle->h_transaction->t_jcb_lock);
+}
+
+/**
  * int journal_stop() - complete a transaction
  * @handle: tranaction to complete.
  * 
@@ -1363,6 +1396,11 @@ int journal_stop(handle_t *handle)
 			wake_up(&journal->j_wait_transaction_locked);
 	}
 
+	/* Move callbacks from the handle to the transaction. */
+	spin_lock(&transaction->t_jcb_lock);
+	list_splice(&handle->h_jcb, &transaction->t_jcb);
+	spin_unlock(&transaction->t_jcb_lock);
+
 	/*
 	 * If the handle is marked SYNC, we need to set another commit
 	 * going!  We also want to force a commit if the current


Cheers, Andreas
--
Andreas Dilger
Sr. Staff Engineer, Lustre Group
Sun Microsystems of Canada, Inc.

--
To unsubscribe from this list: send the line "unsubscribe linux-ext4" in
the body of a message to majordomo@...r.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

Powered by blists - more mailing lists