jbd-stats-2.6-sles10.patch

非常经典的一个分布式系统

Index: linux-2.6.16.46-0.14/include/linux/jbd.h
===================================================================
--- linux-2.6.16.46-0.14.orig/include/linux/jbd.h
+++ linux-2.6.16.46-0.14/include/linux/jbd.h
@@ -422,6 +422,16 @@ struct handle_s 
 };
 
 
+/*
+ * Some stats for checkpoint phase
+ */
+struct transaction_chp_stats_s {
+	unsigned long		cs_chp_time;
+	unsigned long		cs_forced_to_close;
+	unsigned long		cs_written;
+	unsigned long		cs_dropped;
+};
+
 /* The transaction_t type is the guts of the journaling mechanism.  It
  * tracks a compound transaction through its various states:
  *
@@ -553,6 +563,21 @@ struct transaction_s 
 	spinlock_t		t_handle_lock;
 
 	/*
+	 * Longest time some handle had to wait for running transaction
+	 */
+	unsigned long		t_max_wait;
+
+	/*
+	 * When transaction started
+	 */
+	unsigned long		t_start;
+
+	/*
+	 * Checkpointing stats [j_checkpoint_sem]
+	 */
+	struct transaction_chp_stats_s t_chp_stats;
+
+	/*
 	 * Number of outstanding updates running on this transaction
 	 * [t_handle_lock]
 	 */
@@ -592,6 +617,57 @@ struct transaction_s 
 	struct list_head	t_jcb;
 };
 
+struct transaction_run_stats_s {
+	unsigned long		rs_wait;
+	unsigned long		rs_running;
+	unsigned long		rs_locked;
+	unsigned long		rs_flushing;
+	unsigned long		rs_logging;
+
+	unsigned long		rs_handle_count;
+	unsigned long		rs_blocks;
+	unsigned long		rs_blocks_logged;
+};
+
+struct transaction_stats_s
+{
+	int 			ts_type;
+	unsigned long		ts_tid;
+	union {
+		struct transaction_run_stats_s run;
+		struct transaction_chp_stats_s chp;
+	} u;
+};
+
+#define JBD_STATS_RUN		1
+#define JBD_STATS_CHECKPOINT	2
+
+#define ts_wait			u.run.rs_wait
+#define ts_running		u.run.rs_running
+#define ts_locked		u.run.rs_locked
+#define ts_flushing		u.run.rs_flushing
+#define ts_logging		u.run.rs_logging
+#define ts_handle_count		u.run.rs_handle_count
+#define ts_blocks		u.run.rs_blocks
+#define ts_blocks_logged	u.run.rs_blocks_logged
+
+#define ts_chp_time		u.chp.cs_chp_time
+#define ts_forced_to_close	u.chp.cs_forced_to_close
+#define ts_written		u.chp.cs_written
+#define ts_dropped		u.chp.cs_dropped
+
+#define CURRENT_MSECS		(jiffies_to_msecs(jiffies))
+
+static inline unsigned int
+jbd_time_diff(unsigned int start, unsigned int end)
+{
+	if (unlikely(start > end))
+		end = end + (~0UL - start);
+	else
+		end -= start;
+	return end;
+}
+
 /**
  * struct journal_s - The journal_s type is the concrete type associated with
  *     journal_t.
@@ -845,6 +921,16 @@ struct journal_s
 	pid_t			j_last_sync_writer;
 
 	/*
+	 *
+	 */
+	struct transaction_stats_s *j_history;
+	int			j_history_max;
+	int			j_history_cur;
+	spinlock_t		j_history_lock;
+	struct proc_dir_entry	*j_proc_entry;
+	struct transaction_stats_s j_stats;
+
+	/*
 	 * An opaque pointer to fs-private information.  ext3 puts its
 	 * superblock pointer here
 	 */
Index: linux-2.6.16.46-0.14/fs/jbd/transaction.c
===================================================================
--- linux-2.6.16.46-0.14.orig/fs/jbd/transaction.c
+++ linux-2.6.16.46-0.14/fs/jbd/transaction.c
@@ -60,6 +60,8 @@ get_transaction(journal_t *journal, tran
 
 	J_ASSERT(journal->j_running_transaction == NULL);
 	journal->j_running_transaction = transaction;
+	transaction->t_max_wait = 0;
+	transaction->t_start = CURRENT_MSECS;
 
 	return transaction;
 }
@@ -86,6 +88,7 @@ static int start_this_handle(journal_t *
 	int nblocks = handle->h_buffer_credits;
 	transaction_t *new_transaction = NULL;
 	int ret = 0;
+	unsigned long ts = CURRENT_MSECS;
 
 	if (nblocks > journal->j_max_transaction_buffers) {
 		printk(KERN_ERR "JBD: %s wants too many credits (%d > %d)\n",
@@ -219,6 +222,12 @@ repeat_locked:
 	/* OK, account for the buffers that this operation expects to
 	 * use and add the handle to the running transaction. */
 
+	if (time_after(transaction->t_start, ts)) {
+		ts = jbd_time_diff(ts, transaction->t_start);
+		if (ts > transaction->t_max_wait)
+			transaction->t_max_wait= ts;
+	}
+
 	handle->h_transaction = transaction;
 	transaction->t_outstanding_credits += nblocks;
 	transaction->t_updates++;
Index: linux-2.6.16.46-0.14/fs/jbd/journal.c
===================================================================
--- linux-2.6.16.46-0.14.orig/fs/jbd/journal.c
+++ linux-2.6.16.46-0.14/fs/jbd/journal.c
@@ -36,6 +36,7 @@
 #include <asm/uaccess.h>
 #include <asm/page.h>
 #include <linux/proc_fs.h>
+#include <linux/seq_file.h>
 
 EXPORT_SYMBOL(journal_start);
 EXPORT_SYMBOL(journal_restart);
@@ -637,6 +638,300 @@ struct journal_head *journal_get_descrip
 	return journal_add_journal_head(bh);
 }
 
+struct jbd_stats_proc_session {
+	journal_t *journal;
+	struct transaction_stats_s *stats;
+	int start;
+	int max;
+};
+
+static void *jbd_history_skip_empty(struct jbd_stats_proc_session *s,
+					struct transaction_stats_s *ts,
+					int first)
+{
+	if (ts == s->stats + s->max)
+		ts = s->stats;
+	if (!first && ts == s->stats + s->start)
+		return NULL;
+	while (ts->ts_type == 0) {
+		ts++;
+		if (ts == s->stats + s->max)
+			ts = s->stats;
+		if (ts == s->stats + s->start)
+			return NULL;
+	}
+	return ts;
+
+}
+
+static void *jbd_seq_history_start(struct seq_file *seq, loff_t *pos)
+{
+	struct jbd_stats_proc_session *s = seq->private;
+	struct transaction_stats_s *ts;
+	int l = *pos;
+
+	if (l == 0)
+		return SEQ_START_TOKEN;
+	ts = jbd_history_skip_empty(s, s->stats + s->start, 1);
+	if (!ts)
+		return NULL;
+	while (--l && (ts = jbd_history_skip_empty(s, ++ts, 0)) != NULL);
+	return ts;
+}
+
+static void *jbd_seq_history_next(struct seq_file *seq, void *v, loff_t *pos)
+{
+	struct jbd_stats_proc_session *s = seq->private;
+	struct transaction_stats_s *ts = v;
+
+	++*pos;
+	if (v == SEQ_START_TOKEN)
+		return jbd_history_skip_empty(s, s->stats + s->start, 1);
+	else
+		return jbd_history_skip_empty(s, ++ts, 0);
+}
+
+static int jbd_seq_history_show(struct seq_file *seq, void *v)
+{
+	struct transaction_stats_s *ts = v;
+	if (v == SEQ_START_TOKEN) {
+		seq_printf(seq, "%-4s %-5s %-5s %-5s %-5s %-5s %-5s %-6s %-5s "
+				"%-5s %-5s %-5s %-5s %-5s\n", "R/C", "tid",
+				"wait", "run", "lock", "flush", "log", "hndls",
+				"block", "inlog", "ctime", "write", "drop",
+				"close");
+		return 0;
+	}
+	if (ts->ts_type == JBD_STATS_RUN)
+		seq_printf(seq, "%-4s %-5lu %-5lu %-5lu %-5lu %-5lu %-5lu "
+				"%-6lu %-5lu %-5lu\n", "R", ts->ts_tid,
+				ts->ts_wait, ts->ts_running, ts->ts_locked,
+				ts->ts_flushing, ts->ts_logging,
+				ts->ts_handle_count, ts->ts_blocks,
+				ts->ts_blocks_logged);
+	else if (ts->ts_type == JBD_STATS_CHECKPOINT)
+		seq_printf(seq, "%-4s %-5lu %48s %-5lu %-5lu %-5lu %-5lu\n",
+				"C", ts->ts_tid, " ", ts->ts_chp_time,
+				ts->ts_written, ts->ts_dropped,
+				ts->ts_forced_to_close);
+	else
+		J_ASSERT(0);
+	return 0;
+}
+
+static void jbd_seq_history_stop(struct seq_file *seq, void *v)
+{
+}
+
+static struct seq_operations jbd_seq_history_ops = {
+	.start  = jbd_seq_history_start,
+	.next   = jbd_seq_history_next,
+	.stop   = jbd_seq_history_stop,
+	.show   = jbd_seq_history_show,
+};
+
+static int jbd_seq_history_open(struct inode *inode, struct file *file)
+{
+	journal_t *journal = PDE(inode)->data;
+	struct jbd_stats_proc_session *s;
+	int rc, size;
+
+	s = kmalloc(sizeof(*s), GFP_KERNEL);
+	if (s == NULL)
+		return -EIO;
+	size = sizeof(struct transaction_stats_s) * journal->j_history_max;
+	s->stats = kmalloc(size, GFP_KERNEL);
+	if (s->stats == NULL) {
+		kfree(s);
+		return -EIO;
+	}
+	spin_lock(&journal->j_history_lock);
+	memcpy(s->stats, journal->j_history, size);
+	s->max = journal->j_history_max;
+	s->start = journal->j_history_cur % s->max;
+	spin_unlock(&journal->j_history_lock);
+
+	rc = seq_open(file, &jbd_seq_history_ops);
+	if (rc == 0) {
+		struct seq_file *m = (struct seq_file *)file->private_data;
+		m->private = s;
+	} else {
+		kfree(s->stats);
+		kfree(s);
+	}
+	return rc;
+
+}
+
+static int jbd_seq_history_release(struct inode *inode, struct file *file)
+{
+	struct seq_file *seq = (struct seq_file *)file->private_data;
+	struct jbd_stats_proc_session *s = seq->private;
+	kfree(s->stats);
+	kfree(s);
+	return seq_release(inode, file);
+}
+
+static struct file_operations jbd_seq_history_fops = {
+	.owner		= THIS_MODULE,
+	.open           = jbd_seq_history_open,
+	.read           = seq_read,
+	.llseek         = seq_lseek,
+	.release        = jbd_seq_history_release,
+};
+
+static void *jbd_seq_info_start(struct seq_file *seq, loff_t *pos)
+{
+	return *pos ? NULL : SEQ_START_TOKEN;
+}
+
+static void *jbd_seq_info_next(struct seq_file *seq, void *v, loff_t *pos)
+{
+	return NULL;
+}
+
+static int jbd_seq_info_show(struct seq_file *seq, void *v)
+{
+	struct jbd_stats_proc_session *s = seq->private;
+	if (v != SEQ_START_TOKEN)
+		return 0;
+	seq_printf(seq, "%lu transaction, each upto %u blocks\n",
+			s->stats->ts_tid,
+			s->journal->j_max_transaction_buffers);
+	if (s->stats->ts_tid == 0)
+		return 0;
+	seq_printf(seq, "average: \n  %lums waiting for transaction\n",
+			s->stats->ts_wait / s->stats->ts_tid);
+	seq_printf(seq, "  %lums running transaction\n",
+			s->stats->ts_running / s->stats->ts_tid);
+	seq_printf(seq, "  %lums transaction was being locked\n",
+			s->stats->ts_locked / s->stats->ts_tid);
+	seq_printf(seq, "  %lums flushing data (in ordered mode)\n",
+			s->stats->ts_flushing / s->stats->ts_tid);
+	seq_printf(seq, "  %lums logging transaction\n",
+			s->stats->ts_logging / s->stats->ts_tid);
+	seq_printf(seq, "  %lu handles per transaction\n",
+			s->stats->ts_handle_count / s->stats->ts_tid);
+	seq_printf(seq, "  %lu blocks per transaction\n",
+			s->stats->ts_blocks / s->stats->ts_tid);
+	seq_printf(seq, "  %lu logged blocks per transaction\n",
+			s->stats->ts_blocks_logged / s->stats->ts_tid);
+	return 0;
+}
+
+static void jbd_seq_info_stop(struct seq_file *seq, void *v)
+{
+}
+
+static struct seq_operations jbd_seq_info_ops = {
+	.start  = jbd_seq_info_start,
+	.next   = jbd_seq_info_next,
+	.stop   = jbd_seq_info_stop,
+	.show   = jbd_seq_info_show,
+};
+
+static int jbd_seq_info_open(struct inode *inode, struct file *file)
+{
+	journal_t *journal = PDE(inode)->data;
+	struct jbd_stats_proc_session *s;
+	int rc, size;
+
+	s = kmalloc(sizeof(*s), GFP_KERNEL);
+	if (s == NULL)