1 Index: linux-2.6.9/include/linux/jbd.h
2 ===================================================================
3 --- linux-2.6.9.orig/include/linux/jbd.h
4 +++ linux-2.6.9/include/linux/jbd.h
5 @@ -422,6 +422,16 @@ struct handle_s
10 + * Some stats for checkpoint phase
12 +struct transaction_chp_stats_s {
13 + unsigned long cs_chp_time;
14 + unsigned long cs_forced_to_close;
15 + unsigned long cs_written;
16 + unsigned long cs_dropped;
19 /* The transaction_t type is the guts of the journaling mechanism. It
20 * tracks a compound transaction through its various states:
22 @@ -553,6 +563,21 @@ struct transaction_s
23 spinlock_t t_handle_lock;
26 + * Longest time some handle had to wait for running transaction
28 + unsigned long t_max_wait;
31 + * When transaction started
33 + unsigned long t_start;
36 + * Checkpointing stats [j_checkpoint_sem]
38 + struct transaction_chp_stats_s t_chp_stats;
41 * Number of outstanding updates running on this transaction
44 @@ -592,6 +617,57 @@ struct transaction_s
45 struct list_head t_jcb;
48 +struct transaction_run_stats_s {
49 + unsigned long rs_wait;
50 + unsigned long rs_running;
51 + unsigned long rs_locked;
52 + unsigned long rs_flushing;
53 + unsigned long rs_logging;
55 + unsigned long rs_handle_count;
56 + unsigned long rs_blocks;
57 + unsigned long rs_blocks_logged;
60 +struct transaction_stats_s
63 + unsigned long ts_tid;
65 + struct transaction_run_stats_s run;
66 + struct transaction_chp_stats_s chp;
70 +#define JBD_STATS_RUN 1
71 +#define JBD_STATS_CHECKPOINT 2
73 +#define ts_wait u.run.rs_wait
74 +#define ts_running u.run.rs_running
75 +#define ts_locked u.run.rs_locked
76 +#define ts_flushing u.run.rs_flushing
77 +#define ts_logging u.run.rs_logging
78 +#define ts_handle_count u.run.rs_handle_count
79 +#define ts_blocks u.run.rs_blocks
80 +#define ts_blocks_logged u.run.rs_blocks_logged
82 +#define ts_chp_time u.chp.cs_chp_time
83 +#define ts_forced_to_close u.chp.cs_forced_to_close
84 +#define ts_written u.chp.cs_written
85 +#define ts_dropped u.chp.cs_dropped
87 +#define CURRENT_MSECS (jiffies_to_msecs(jiffies))
89 +static inline unsigned int
90 +jbd_time_diff(unsigned int start, unsigned int end)
92 + if (unlikely(start > end))
93 + end = end + (~0UL - start);
100 * struct journal_s - The journal_s type is the concrete type associated with
102 @@ -828,6 +904,16 @@ struct journal_s
103 struct jbd_revoke_table_s *j_revoke_table[2];
108 + struct transaction_stats_s *j_history;
111 + spinlock_t j_history_lock;
112 + struct proc_dir_entry *j_proc_entry;
113 + struct transaction_stats_s j_stats;
116 * An opaque pointer to fs-private information. ext3 puts its
117 * superblock pointer here
119 Index: linux-2.6.9/fs/jbd/commit.c
120 ===================================================================
121 --- linux-2.6.9.orig/fs/jbd/commit.c
122 +++ linux-2.6.9/fs/jbd/commit.c
124 #include <linux/mm.h>
125 #include <linux/pagemap.h>
126 #include <linux/smp_lock.h>
127 +#include <linux/jiffies.h>
130 * Default IO end handler for temporary BJ_IO buffer_heads.
131 @@ -101,6 +102,7 @@ static int inverted_lock(journal_t *jour
133 void journal_commit_transaction(journal_t *journal)
135 + struct transaction_stats_s stats;
136 transaction_t *commit_transaction;
137 struct journal_head *jh, *new_jh, *descriptor;
138 struct buffer_head *wbuf[64];
139 @@ -147,6 +149,11 @@ void journal_commit_transaction(journal_
140 spin_lock(&journal->j_state_lock);
141 commit_transaction->t_state = T_LOCKED;
143 + stats.ts_wait = commit_transaction->t_max_wait;
144 + stats.ts_locked = CURRENT_MSECS;
145 + stats.ts_running = jbd_time_diff(commit_transaction->t_start,
148 spin_lock(&commit_transaction->t_handle_lock);
149 while (commit_transaction->t_updates) {
151 @@ -219,6 +226,9 @@ void journal_commit_transaction(journal_
153 journal_switch_revoke_table(journal);
155 + stats.ts_flushing = CURRENT_MSECS;
156 + stats.ts_locked = jbd_time_diff(stats.ts_locked, stats.ts_flushing);
158 commit_transaction->t_state = T_FLUSH;
159 journal->j_committing_transaction = commit_transaction;
160 journal->j_running_transaction = NULL;
161 @@ -365,6 +375,11 @@ write_out_data:
163 commit_transaction->t_state = T_COMMIT;
165 + stats.ts_logging = CURRENT_MSECS;
166 + stats.ts_flushing = jbd_time_diff(stats.ts_flushing, stats.ts_logging);
167 + stats.ts_blocks = commit_transaction->t_outstanding_credits;
168 + stats.ts_blocks_logged = 0;
172 while (commit_transaction->t_buffers) {
173 @@ -513,6 +528,7 @@ start_journal_io:
174 submit_bh(WRITE, bh);
177 + stats.ts_blocks_logged += bufs;
179 /* Force a new descriptor to be generated next
180 time round the loop. */
181 @@ -767,6 +783,7 @@ restart_loop:
182 cp_transaction = jh->b_cp_transaction;
183 if (cp_transaction) {
184 JBUFFER_TRACE(jh, "remove from old cp transaction");
185 + cp_transaction->t_chp_stats.cs_dropped++;
186 __journal_remove_checkpoint(jh);
189 @@ -816,6 +833,37 @@ restart_loop:
191 cond_resched_lock(&journal->j_list_lock);
194 + commit_transaction->t_start = CURRENT_MSECS;
195 + stats.ts_logging = jbd_time_diff(stats.ts_logging,
196 + commit_transaction->t_start);
199 + * File the transaction for history
201 + stats.ts_type = JBD_STATS_RUN;
202 + stats.ts_tid = commit_transaction->t_tid;
203 + stats.ts_handle_count = commit_transaction->t_handle_count;
204 + spin_lock(&journal->j_history_lock);
205 + memcpy(journal->j_history + journal->j_history_cur, &stats,
207 + if (++journal->j_history_cur == journal->j_history_max)
208 + journal->j_history_cur = 0;
211 + * Calculate overall stats
213 + journal->j_stats.ts_tid++;
214 + journal->j_stats.ts_wait += stats.ts_wait;
215 + journal->j_stats.ts_running += stats.ts_running;
216 + journal->j_stats.ts_locked += stats.ts_locked;
217 + journal->j_stats.ts_flushing += stats.ts_flushing;
218 + journal->j_stats.ts_logging += stats.ts_logging;
219 + journal->j_stats.ts_handle_count += stats.ts_handle_count;
220 + journal->j_stats.ts_blocks += stats.ts_blocks;
221 + journal->j_stats.ts_blocks_logged += stats.ts_blocks_logged;
222 + spin_unlock(&journal->j_history_lock);
224 spin_unlock(&journal->j_list_lock);
226 * This is a bit sleazy. We borrow j_list_lock to protect
227 Index: linux-2.6.9/fs/jbd/checkpoint.c
228 ===================================================================
229 --- linux-2.6.9.orig/fs/jbd/checkpoint.c
230 +++ linux-2.6.9/fs/jbd/checkpoint.c
231 @@ -166,6 +166,7 @@ static int __cleanup_transaction(journal
232 transaction_t *t = jh->b_transaction;
233 tid_t tid = t->t_tid;
235 + transaction->t_chp_stats.cs_forced_to_close++;
236 spin_unlock(&journal->j_list_lock);
237 jbd_unlock_bh_state(bh);
238 log_start_commit(journal, tid);
239 @@ -226,7 +227,7 @@ __flush_batch(journal_t *journal, struct
241 static int __flush_buffer(journal_t *journal, struct journal_head *jh,
242 struct buffer_head **bhs, int *batch_count,
244 + int *drop_count, transaction_t *transaction)
246 struct buffer_head *bh = jh2bh(jh);
248 @@ -247,6 +248,7 @@ static int __flush_buffer(journal_t *jou
249 set_buffer_jwrite(bh);
250 bhs[*batch_count] = bh;
251 jbd_unlock_bh_state(bh);
252 + transaction->t_chp_stats.cs_written++;
254 if (*batch_count == NR_BATCH) {
255 __flush_batch(journal, bhs, batch_count);
256 @@ -315,6 +317,8 @@ int log_do_checkpoint(journal_t *journal
259 transaction = journal->j_checkpoint_transactions;
260 + if (transaction->t_chp_stats.cs_chp_time == 0)
261 + transaction->t_chp_stats.cs_chp_time = CURRENT_MSECS;
262 this_tid = transaction->t_tid;
263 jh = transaction->t_checkpoint_list;
264 last_jh = jh->b_cpprev;
265 @@ -331,7 +335,8 @@ int log_do_checkpoint(journal_t *journal
269 - retry = __flush_buffer(journal, jh, bhs, &batch_count, &drop_count);
270 + retry = __flush_buffer(journal, jh, bhs, &batch_count,
271 + &drop_count, transaction);
272 } while (jh != last_jh && !retry);
275 @@ -597,6 +602,8 @@ void __journal_insert_checkpoint(struct
277 void __journal_drop_transaction(journal_t *journal, transaction_t *transaction)
279 + struct transaction_stats_s stats;
281 assert_spin_locked(&journal->j_list_lock);
282 if (transaction->t_cpnext) {
283 transaction->t_cpnext->t_cpprev = transaction->t_cpprev;
284 @@ -622,5 +629,25 @@ void __journal_drop_transaction(journal_
285 J_ASSERT(journal->j_running_transaction != transaction);
287 jbd_debug(1, "Dropping transaction %d, all done\n", transaction->t_tid);
290 + * File the transaction for history
292 + if (transaction->t_chp_stats.cs_written != 0 ||
293 + transaction->t_chp_stats.cs_chp_time != 0) {
294 + stats.ts_type = JBD_STATS_CHECKPOINT;
295 + stats.ts_tid = transaction->t_tid;
296 + stats.u.chp = transaction->t_chp_stats;
297 + if (stats.ts_chp_time)
298 + stats.ts_chp_time =
299 + jbd_time_diff(stats.ts_chp_time, CURRENT_MSECS);
300 + spin_lock(&journal->j_history_lock);
301 + memcpy(journal->j_history + journal->j_history_cur, &stats,
303 + if (++journal->j_history_cur == journal->j_history_max)
304 + journal->j_history_cur = 0;
305 + spin_unlock(&journal->j_history_lock);
310 Index: linux-2.6.9/fs/jbd/transaction.c
311 ===================================================================
312 --- linux-2.6.9.orig/fs/jbd/transaction.c
313 +++ linux-2.6.9/fs/jbd/transaction.c
314 @@ -60,6 +60,8 @@ get_transaction(journal_t *journal, tran
316 J_ASSERT(journal->j_running_transaction == NULL);
317 journal->j_running_transaction = transaction;
318 + transaction->t_max_wait = 0;
319 + transaction->t_start = CURRENT_MSECS;
323 @@ -86,6 +88,7 @@ static int start_this_handle(journal_t *
324 int nblocks = handle->h_buffer_credits;
325 transaction_t *new_transaction = NULL;
327 + unsigned long ts = CURRENT_MSECS;
329 if (nblocks > journal->j_max_transaction_buffers) {
330 printk(KERN_ERR "JBD: %s wants too many credits (%d > %d)\n",
331 @@ -219,6 +222,12 @@ repeat_locked:
332 /* OK, account for the buffers that this operation expects to
333 * use and add the handle to the running transaction. */
335 + if (time_after(transaction->t_start, ts)) {
336 + ts = jbd_time_diff(ts, transaction->t_start);
337 + if (ts > transaction->t_max_wait)
338 + transaction->t_max_wait= ts;
341 handle->h_transaction = transaction;
342 transaction->t_outstanding_credits += nblocks;
343 transaction->t_updates++;
344 Index: linux-2.6.9/fs/jbd/journal.c
345 ===================================================================
346 --- linux-2.6.9.orig/fs/jbd/journal.c
347 +++ linux-2.6.9/fs/jbd/journal.c
349 #include <asm/uaccess.h>
350 #include <asm/page.h>
351 #include <linux/proc_fs.h>
352 +#include <linux/seq_file.h>
354 EXPORT_SYMBOL(journal_start);
355 EXPORT_SYMBOL(journal_restart);
356 @@ -649,6 +650,300 @@ struct journal_head *journal_get_descrip
357 return journal_add_journal_head(bh);
360 +struct jbd_stats_proc_session {
361 + journal_t *journal;
362 + struct transaction_stats_s *stats;
367 +static void *jbd_history_skip_empty(struct jbd_stats_proc_session *s,
368 + struct transaction_stats_s *ts,
371 + if (ts == s->stats + s->max)
373 + if (!first && ts == s->stats + s->start)
375 + while (ts->ts_type == 0) {
377 + if (ts == s->stats + s->max)
379 + if (ts == s->stats + s->start)
386 +static void *jbd_seq_history_start(struct seq_file *seq, loff_t *pos)
388 + struct jbd_stats_proc_session *s = seq->private;
389 + struct transaction_stats_s *ts;
393 + return SEQ_START_TOKEN;
394 + ts = jbd_history_skip_empty(s, s->stats + s->start, 1);
397 + while (--l && (ts = jbd_history_skip_empty(s, ++ts, 0)) != NULL);
401 +static void *jbd_seq_history_next(struct seq_file *seq, void *v, loff_t *pos)
403 + struct jbd_stats_proc_session *s = seq->private;
404 + struct transaction_stats_s *ts = v;
407 + if (v == SEQ_START_TOKEN)
408 + return jbd_history_skip_empty(s, s->stats + s->start, 1);
410 + return jbd_history_skip_empty(s, ++ts, 0);
413 +static int jbd_seq_history_show(struct seq_file *seq, void *v)
415 + struct transaction_stats_s *ts = v;
416 + if (v == SEQ_START_TOKEN) {
417 + seq_printf(seq, "%-4s %-5s %-5s %-5s %-5s %-5s %-5s %-6s %-5s "
418 + "%-5s %-5s %-5s %-5s %-5s\n", "R/C", "tid",
419 + "wait", "run", "lock", "flush", "log", "hndls",
420 + "block", "inlog", "ctime", "write", "drop",
424 + if (ts->ts_type == JBD_STATS_RUN)
425 + seq_printf(seq, "%-4s %-5lu %-5lu %-5lu %-5lu %-5lu %-5lu "
426 + "%-6lu %-5lu %-5lu\n", "R", ts->ts_tid,
427 + ts->ts_wait, ts->ts_running, ts->ts_locked,
428 + ts->ts_flushing, ts->ts_logging,
429 + ts->ts_handle_count, ts->ts_blocks,
430 + ts->ts_blocks_logged);
431 + else if (ts->ts_type == JBD_STATS_CHECKPOINT)
432 + seq_printf(seq, "%-4s %-5lu %48s %-5lu %-5lu %-5lu %-5lu\n",
433 + "C", ts->ts_tid, " ", ts->ts_chp_time,
434 + ts->ts_written, ts->ts_dropped,
435 + ts->ts_forced_to_close);
441 +static void jbd_seq_history_stop(struct seq_file *seq, void *v)
445 +static struct seq_operations jbd_seq_history_ops = {
446 + .start = jbd_seq_history_start,
447 + .next = jbd_seq_history_next,
448 + .stop = jbd_seq_history_stop,
449 + .show = jbd_seq_history_show,
452 +static int jbd_seq_history_open(struct inode *inode, struct file *file)
454 + journal_t *journal = PDE(inode)->data;
455 + struct jbd_stats_proc_session *s;
458 + s = kmalloc(sizeof(*s), GFP_KERNEL);
461 + size = sizeof(struct transaction_stats_s) * journal->j_history_max;
462 + s->stats = kmalloc(size, GFP_KERNEL);
463 + if (s->stats == NULL) {
467 + spin_lock(&journal->j_history_lock);
468 + memcpy(s->stats, journal->j_history, size);
469 + s->max = journal->j_history_max;
470 + s->start = journal->j_history_cur % s->max;
471 + spin_unlock(&journal->j_history_lock);
473 + rc = seq_open(file, &jbd_seq_history_ops);
475 + struct seq_file *m = (struct seq_file *)file->private_data;
485 +static int jbd_seq_history_release(struct inode *inode, struct file *file)
487 + struct seq_file *seq = (struct seq_file *)file->private_data;
488 + struct jbd_stats_proc_session *s = seq->private;
491 + return seq_release(inode, file);
494 +static struct file_operations jbd_seq_history_fops = {
495 + .owner = THIS_MODULE,
496 + .open = jbd_seq_history_open,
498 + .llseek = seq_lseek,
499 + .release = jbd_seq_history_release,
502 +static void *jbd_seq_info_start(struct seq_file *seq, loff_t *pos)
504 + return *pos ? NULL : SEQ_START_TOKEN;
507 +static void *jbd_seq_info_next(struct seq_file *seq, void *v, loff_t *pos)
512 +static int jbd_seq_info_show(struct seq_file *seq, void *v)
514 + struct jbd_stats_proc_session *s = seq->private;
515 + if (v != SEQ_START_TOKEN)
517 + seq_printf(seq, "%lu transaction, each upto %u blocks\n",
519 + s->journal->j_max_transaction_buffers);
520 + if (s->stats->ts_tid == 0)
522 + seq_printf(seq, "average: \n %lums waiting for transaction\n",
523 + s->stats->ts_wait / s->stats->ts_tid);
524 + seq_printf(seq, " %lums running transaction\n",
525 + s->stats->ts_running / s->stats->ts_tid);
526 + seq_printf(seq, " %lums transaction was being locked\n",
527 + s->stats->ts_locked / s->stats->ts_tid);
528 + seq_printf(seq, " %lums flushing data (in ordered mode)\n",
529 + s->stats->ts_flushing / s->stats->ts_tid);
530 + seq_printf(seq, " %lums logging transaction\n",
531 + s->stats->ts_logging / s->stats->ts_tid);
532 + seq_printf(seq, " %lu handles per transaction\n",
533 + s->stats->ts_handle_count / s->stats->ts_tid);
534 + seq_printf(seq, " %lu blocks per transaction\n",
535 + s->stats->ts_blocks / s->stats->ts_tid);
536 + seq_printf(seq, " %lu logged blocks per transaction\n",
537 + s->stats->ts_blocks_logged / s->stats->ts_tid);
541 +static void jbd_seq_info_stop(struct seq_file *seq, void *v)
545 +static struct seq_operations jbd_seq_info_ops = {
546 + .start = jbd_seq_info_start,
547 + .next = jbd_seq_info_next,
548 + .stop = jbd_seq_info_stop,
549 + .show = jbd_seq_info_show,
552 +static int jbd_seq_info_open(struct inode *inode, struct file *file)
554 + journal_t *journal = PDE(inode)->data;
555 + struct jbd_stats_proc_session *s;
558 + s = kmalloc(sizeof(*s), GFP_KERNEL);
561 + size = sizeof(struct transaction_stats_s);
562 + s->stats = kmalloc(size, GFP_KERNEL);
563 + if (s->stats == NULL) {
567 + spin_lock(&journal->j_history_lock);
568 + memcpy(s->stats, &journal->j_stats, size);
569 + s->journal = journal;
570 + spin_unlock(&journal->j_history_lock);
572 + rc = seq_open(file, &jbd_seq_info_ops);
574 + struct seq_file *m = (struct seq_file *)file->private_data;
584 +static int jbd_seq_info_release(struct inode *inode, struct file *file)
586 + struct seq_file *seq = (struct seq_file *)file->private_data;
587 + struct jbd_stats_proc_session *s = seq->private;
590 + return seq_release(inode, file);
593 +static struct file_operations jbd_seq_info_fops = {
594 + .owner = THIS_MODULE,
595 + .open = jbd_seq_info_open,
597 + .llseek = seq_lseek,
598 + .release = jbd_seq_info_release,
601 +static struct proc_dir_entry *proc_jbd_stats = NULL;
603 +static void jbd_stats_proc_init(journal_t *journal)
607 + snprintf(name, sizeof(name) - 1, "%s", bdevname(journal->j_dev, name));
608 + journal->j_proc_entry = proc_mkdir(name, proc_jbd_stats);
609 + if (journal->j_proc_entry) {
610 + struct proc_dir_entry *p;
611 + p = create_proc_entry("history", S_IRUGO,
612 + journal->j_proc_entry);
614 + p->proc_fops = &jbd_seq_history_fops;
616 + p = create_proc_entry("info", S_IRUGO,
617 + journal->j_proc_entry);
619 + p->proc_fops = &jbd_seq_info_fops;
626 +static void jbd_stats_proc_exit(journal_t *journal)
630 + snprintf(name, sizeof(name) - 1, "%s", bdevname(journal->j_dev, name));
631 + remove_proc_entry("info", journal->j_proc_entry);
632 + remove_proc_entry("history", journal->j_proc_entry);
633 + remove_proc_entry(name, proc_jbd_stats);
636 +static void journal_init_stats(journal_t *journal)
640 + if (proc_jbd_stats == NULL)
643 + journal->j_history_max = 100;
644 + size = sizeof(struct transaction_stats_s) * journal->j_history_max;
645 + journal->j_history = kmalloc(size, GFP_KERNEL);
646 + if (journal->j_history == NULL) {
647 + journal->j_history_max = 0;
650 + memset(journal->j_history, 0, size);
651 + spin_lock_init(&journal->j_history_lock);
655 * Management for journal control blocks: functions to create and
656 * destroy journal_t structures, and to initialise and read existing
657 @@ -691,6 +986,9 @@ static journal_t * journal_init_common (
662 + journal_init_stats(journal);
667 @@ -733,6 +1031,7 @@ journal_t * journal_init_dev(struct bloc
668 journal->j_blk_offset = start;
669 journal->j_maxlen = len;
670 journal->j_blocksize = blocksize;
671 + jbd_stats_proc_init(journal);
673 bh = __getblk(journal->j_dev, start, journal->j_blocksize);
674 J_ASSERT(bh != NULL);
675 @@ -770,6 +1069,7 @@ journal_t * journal_init_inode (struct i
677 journal->j_maxlen = inode->i_size >> inode->i_sb->s_blocksize_bits;
678 journal->j_blocksize = inode->i_sb->s_blocksize;
679 + jbd_stats_proc_init(journal);
681 err = journal_bmap(journal, 0, &blocknr);
682 /* If that failed, give up */
683 @@ -1140,6 +1440,8 @@ void journal_destroy(journal_t *journal)
684 brelse(journal->j_sb_buffer);
687 + if (journal->j_proc_entry)
688 + jbd_stats_proc_exit(journal);
689 if (journal->j_inode)
690 iput(journal->j_inode);
691 if (journal->j_revoke)
692 @@ -1896,6 +2198,28 @@ static void __exit remove_jbd_proc_entry
696 +#if defined(CONFIG_PROC_FS)
698 +#define JBD_STATS_PROC_NAME "fs/jbd"
700 +static void __init create_jbd_stats_proc_entry(void)
702 + proc_jbd_stats = proc_mkdir(JBD_STATS_PROC_NAME, NULL);
705 +static void __exit remove_jbd_stats_proc_entry(void)
707 + if (proc_jbd_stats)
708 + remove_proc_entry(JBD_STATS_PROC_NAME, NULL);
713 +#define create_jbd_stats_proc_entry() do {} while (0)
714 +#define remove_jbd_stats_proc_entry() do {} while (0)
718 kmem_cache_t *jbd_handle_cache;
720 static int __init journal_init_handle_cache(void)
721 @@ -1950,6 +2274,7 @@ static int __init journal_init(void)
723 journal_destroy_caches();
724 create_jbd_proc_entry();
725 + create_jbd_stats_proc_entry();
729 @@ -1961,6 +2286,7 @@ static void __exit journal_exit(void)
730 printk(KERN_EMERG "JBD: leaked %d journal_heads!\n", n);
732 remove_jbd_proc_entry();
733 + remove_jbd_stats_proc_entry();
734 journal_destroy_caches();