-/* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
- * vim:expandtab:shiftwidth=8:tabstop=8:
- *
+/*
* GPL HEADER START
*
* DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
* GPL HEADER END
*/
/*
- * Copyright 2008 Sun Microsystems, Inc. All rights reserved
+ * Copyright (c) 2008, 2010, Oracle and/or its affiliates. All rights reserved.
* Use is subject to license terms.
+ *
+ * Copyright (c) 2011, Whamcloud, Inc.
*/
/*
* This file is part of Lustre, http://www.lustre.org/
#include "osd_internal.h"
#ifdef LPROCFS
-enum {
- LPROC_OSD_NR
-};
-static const char *osd_counter_names[LPROC_OSD_NR] = {
+void osd_brw_stats_update(struct osd_device *osd, struct osd_iobuf *iobuf)
+{
+ struct brw_stats *s = &osd->od_brw_stats;
+ unsigned long *last_block = NULL;
+ struct page **pages = iobuf->dr_pages;
+ struct page *last_page = NULL;
+ unsigned long discont_pages = 0;
+ unsigned long discont_blocks = 0;
+ unsigned long *blocks = iobuf->dr_blocks;
+ int i, nr_pages = iobuf->dr_npages;
+ int blocks_per_page;
+ int rw = iobuf->dr_rw;
+
+ if (unlikely(nr_pages == 0))
+ return;
+
+ blocks_per_page = CFS_PAGE_SIZE >> osd_sb(osd)->s_blocksize_bits;
+
+ lprocfs_oh_tally_log2(&s->hist[BRW_R_PAGES+rw], nr_pages);
+
+ while (nr_pages-- > 0) {
+ if (last_page && (*pages)->index != (last_page->index + 1))
+ discont_pages++;
+ last_page = *pages;
+ pages++;
+ for (i = 0; i < blocks_per_page; i++) {
+ if (last_block && *blocks != (*last_block + 1))
+ discont_blocks++;
+ last_block = blocks++;
+ }
+ }
+
+ lprocfs_oh_tally(&s->hist[BRW_R_DISCONT_PAGES+rw], discont_pages);
+ lprocfs_oh_tally(&s->hist[BRW_R_DISCONT_BLOCKS+rw], discont_blocks);
+}
+
+#define pct(a, b) (b ? a * 100 / b : 0)
+
+static void display_brw_stats(struct seq_file *seq, char *name, char *units,
+ struct obd_histogram *read, struct obd_histogram *write, int scale)
+{
+ unsigned long read_tot, write_tot, r, w, read_cum = 0, write_cum = 0;
+ int i;
+
+ seq_printf(seq, "\n%26s read | write\n", " ");
+ seq_printf(seq, "%-22s %-5s %% cum %% | %-11s %% cum %%\n",
+ name, units, units);
+
+ read_tot = lprocfs_oh_sum(read);
+ write_tot = lprocfs_oh_sum(write);
+ for (i = 0; i < OBD_HIST_MAX; i++) {
+ r = read->oh_buckets[i];
+ w = write->oh_buckets[i];
+ read_cum += r;
+ write_cum += w;
+ if (read_cum == 0 && write_cum == 0)
+ continue;
+
+ if (!scale)
+ seq_printf(seq, "%u", i);
+ else if (i < 10)
+ seq_printf(seq, "%u", scale << i);
+ else if (i < 20)
+ seq_printf(seq, "%uK", scale << (i-10));
+ else
+ seq_printf(seq, "%uM", scale << (i-20));
+
+ seq_printf(seq, ":\t\t%10lu %3lu %3lu | %4lu %3lu %3lu\n",
+ r, pct(r, read_tot), pct(read_cum, read_tot),
+ w, pct(w, write_tot), pct(write_cum, write_tot));
+
+ if (read_cum == read_tot && write_cum == write_tot)
+ break;
+ }
+}
+
+static void brw_stats_show(struct seq_file *seq, struct brw_stats *brw_stats)
+{
+ struct timeval now;
+
+ /* this sampling races with updates */
+ cfs_gettimeofday(&now);
+ seq_printf(seq, "snapshot_time: %lu.%lu (secs.usecs)\n",
+ now.tv_sec, now.tv_usec);
+
+ display_brw_stats(seq, "pages per bulk r/w", "rpcs",
+ &brw_stats->hist[BRW_R_PAGES],
+ &brw_stats->hist[BRW_W_PAGES], 1);
+
+ display_brw_stats(seq, "discontiguous pages", "rpcs",
+ &brw_stats->hist[BRW_R_DISCONT_PAGES],
+ &brw_stats->hist[BRW_W_DISCONT_PAGES], 0);
+
+ display_brw_stats(seq, "discontiguous blocks", "rpcs",
+ &brw_stats->hist[BRW_R_DISCONT_BLOCKS],
+ &brw_stats->hist[BRW_W_DISCONT_BLOCKS], 0);
+
+ display_brw_stats(seq, "disk fragmented I/Os", "ios",
+ &brw_stats->hist[BRW_R_DIO_FRAGS],
+ &brw_stats->hist[BRW_W_DIO_FRAGS], 0);
+
+ display_brw_stats(seq, "disk I/Os in flight", "ios",
+ &brw_stats->hist[BRW_R_RPC_HIST],
+ &brw_stats->hist[BRW_W_RPC_HIST], 0);
+
+ display_brw_stats(seq, "I/O time (1/1000s)", "ios",
+ &brw_stats->hist[BRW_R_IO_TIME],
+ &brw_stats->hist[BRW_W_IO_TIME], 1000 / CFS_HZ);
+
+ display_brw_stats(seq, "disk I/O size", "ios",
+ &brw_stats->hist[BRW_R_DISK_IOSIZE],
+ &brw_stats->hist[BRW_W_DISK_IOSIZE], 1);
+}
+
+#undef pct
+
+static int osd_brw_stats_seq_show(struct seq_file *seq, void *v)
+{
+ struct osd_device *osd = seq->private;
+
+ brw_stats_show(seq, &osd->od_brw_stats);
+
+ return 0;
+}
+
+static ssize_t osd_brw_stats_seq_write(struct file *file, const char *buf,
+ size_t len, loff_t *off)
+{
+ struct seq_file *seq = file->private_data;
+ struct osd_device *osd = seq->private;
+ int i;
+
+ for (i = 0; i < BRW_LAST; i++)
+ lprocfs_oh_clear(&osd->od_brw_stats.hist[i]);
+
+ return len;
+}
+
+LPROC_SEQ_FOPS(osd_brw_stats);
+
+static int osd_stats_init(struct osd_device *osd)
+{
+ int i, result;
+ ENTRY;
+
+ for (i = 0; i < BRW_LAST; i++)
+ cfs_spin_lock_init(&osd->od_brw_stats.hist[i].oh_lock);
+
+ osd->od_stats = lprocfs_alloc_stats(LPROC_OSD_LAST, 0);
+ if (osd->od_stats != NULL) {
+ result = lprocfs_register_stats(osd->od_proc_entry, "stats",
+ osd->od_stats);
+ if (result)
+ GOTO(out, result);
+
+ lprocfs_counter_init(osd->od_stats, LPROC_OSD_GET_PAGE,
+ LPROCFS_CNTR_AVGMINMAX|LPROCFS_CNTR_STDDEV,
+ "get_page", "usec");
+ lprocfs_counter_init(osd->od_stats, LPROC_OSD_NO_PAGE,
+ LPROCFS_CNTR_AVGMINMAX,
+ "get_page_failures", "num");
+ lprocfs_counter_init(osd->od_stats, LPROC_OSD_CACHE_ACCESS,
+ LPROCFS_CNTR_AVGMINMAX,
+ "cache_access", "pages");
+ lprocfs_counter_init(osd->od_stats, LPROC_OSD_CACHE_HIT,
+ LPROCFS_CNTR_AVGMINMAX,
+ "cache_hit", "pages");
+ lprocfs_counter_init(osd->od_stats, LPROC_OSD_CACHE_MISS,
+ LPROCFS_CNTR_AVGMINMAX,
+ "cache_miss", "pages");
+#if OSD_THANDLE_STATS
+ lprocfs_counter_init(osd->od_stats, LPROC_OSD_THANDLE_STARTING,
+ LPROCFS_CNTR_AVGMINMAX,
+ "thandle starting", "usec");
+ lprocfs_counter_init(osd->od_stats, LPROC_OSD_THANDLE_OPEN,
+ LPROCFS_CNTR_AVGMINMAX,
+ "thandle open", "usec");
+ lprocfs_counter_init(osd->od_stats, LPROC_OSD_THANDLE_CLOSING,
+ LPROCFS_CNTR_AVGMINMAX,
+ "thandle closing", "usec");
+#endif
+ lprocfs_seq_create(osd->od_proc_entry, "brw_stats",
+ 0444, &osd_brw_stats_fops, osd);
+ } else
+ result = -ENOMEM;
+
+out:
+ RETURN(result);
+}
+
+static const char *osd_counter_names[] = {
+#if OSD_THANDLE_STATS
+ [LPROC_OSD_THANDLE_STARTING] = "thandle starting",
+ [LPROC_OSD_THANDLE_OPEN] = "thandle open",
+ [LPROC_OSD_THANDLE_CLOSING] = "thandle closing"
+#endif
};
int osd_procfs_init(struct osd_device *osd, const char *name)
/* Find the type procroot and add the proc entry for this device */
lprocfs_osd_init_vars(&lvars);
osd->od_proc_entry = lprocfs_register(name, type->typ_procroot,
- lvars.obd_vars, osd);
+ lvars.obd_vars, &osd->od_dt_dev);
if (IS_ERR(osd->od_proc_entry)) {
rc = PTR_ERR(osd->od_proc_entry);
CERROR("Error %d setting up lprocfs for %s\n",
rc = lu_time_init(&osd->od_stats,
osd->od_proc_entry,
osd_counter_names, ARRAY_SIZE(osd_counter_names));
+
+ rc = osd_stats_init(osd);
+
EXIT;
out:
if (rc)
-int lprocfs_osd_rd_blksize(char *page, char **start, off_t off, int count,
- int *eof, void *data)
+static int lprocfs_osd_rd_fstype(char *page, char **start, off_t off, int count,
+ int *eof, void *data)
{
- struct osd_device *osd = data;
- int rc = osd_statfs(NULL, &osd->od_dt_dev, &osd->od_kstatfs);
- if (!rc) {
- *eof = 1;
- rc = snprintf(page, count, "%ld\n", osd->od_kstatfs.f_bsize);
- }
- return rc;
+ struct obd_device *osd = data;
+
+ LASSERT(osd != NULL);
+ return snprintf(page, count, "ldiskfs\n");
}
-int lprocfs_osd_rd_kbytestotal(char *page, char **start, off_t off, int count,
- int *eof, void *data)
+static int lprocfs_osd_rd_mntdev(char *page, char **start, off_t off, int count,
+ int *eof, void *data)
{
- struct osd_device *osd = data;
- int rc = osd_statfs(NULL, &osd->od_dt_dev, &osd->od_kstatfs);
- if (!rc) {
- __u32 blk_size = osd->od_kstatfs.f_bsize >> 10;
- __u64 result = osd->od_kstatfs.f_blocks;
+ struct osd_device *osd = osd_dt_dev(data);
- while (blk_size >>= 1)
- result <<= 1;
+ LASSERT(osd != NULL);
+ if (unlikely(osd->od_mount == NULL))
+ return -EINPROGRESS;
- *eof = 1;
- rc = snprintf(page, count, LPU64"\n", result);
- }
- return rc;
+ LASSERT(mnt_get_devname(osd->od_mount->lmi_mnt));
+ *eof = 1;
+
+ return snprintf(page, count, "%s\n",
+ mnt_get_devname(osd->od_mount->lmi_mnt));
}
-int lprocfs_osd_rd_kbytesfree(char *page, char **start, off_t off, int count,
+#ifdef HAVE_LDISKFS_PDO
+static int lprocfs_osd_rd_pdo(char *page, char **start, off_t off, int count,
int *eof, void *data)
{
- struct osd_device *osd = data;
- int rc = osd_statfs(NULL, &osd->od_dt_dev, &osd->od_kstatfs);
- if (!rc) {
- __u32 blk_size = osd->od_kstatfs.f_bsize >> 10;
- __u64 result = osd->od_kstatfs.f_bfree;
-
- while (blk_size >>= 1)
- result <<= 1;
+ *eof = 1;
- *eof = 1;
- rc = snprintf(page, count, LPU64"\n", result);
- }
- return rc;
+ return snprintf(page, count, "%s\n", ldiskfs_pdo ? "ON" : "OFF");
}
-int lprocfs_osd_rd_kbytesavail(char *page, char **start, off_t off, int count,
- int *eof, void *data)
+static int lprocfs_osd_wr_pdo(struct file *file, const char *buffer,
+ unsigned long count, void *data)
{
- struct osd_device *osd = data;
- int rc = osd_statfs(NULL, &osd->od_dt_dev, &osd->od_kstatfs);
- if (!rc) {
- __u32 blk_size = osd->od_kstatfs.f_bsize >> 10;
- __u64 result = osd->od_kstatfs.f_bavail;
+ int pdo;
+ int rc;
- while (blk_size >>= 1)
- result <<= 1;
+ rc = lprocfs_write_helper(buffer, count, &pdo);
+ if (rc != 0)
+ return rc;
- *eof = 1;
- rc = snprintf(page, count, LPU64"\n", result);
- }
- return rc;
+ ldiskfs_pdo = !!pdo;
+
+ return count;
}
+#endif
-int lprocfs_osd_rd_filestotal(char *page, char **start, off_t off, int count,
- int *eof, void *data)
+static int lprocfs_osd_rd_auto_scrub(char *page, char **start, off_t off,
+ int count, int *eof, void *data)
{
- struct osd_device *osd = data;
- int rc = osd_statfs(NULL, &osd->od_dt_dev, &osd->od_kstatfs);
- if (!rc) {
- *eof = 1;
- rc = snprintf(page, count, LPU64"\n", osd->od_kstatfs.f_files);
- }
+ struct osd_device *dev = data;
- return rc;
-}
+ LASSERT(dev != NULL);
+ if (unlikely(dev->od_mount == NULL))
+ return -EINPROGRESS;
-int lprocfs_osd_rd_filesfree(char *page, char **start, off_t off, int count,
- int *eof, void *data)
-{
- struct osd_device *osd = data;
- int rc = osd_statfs(NULL, &osd->od_dt_dev, &osd->od_kstatfs);
- if (!rc) {
- *eof = 1;
- rc = snprintf(page, count, LPU64"\n", osd->od_kstatfs.f_ffree);
- }
- return rc;
+ *eof = 1;
+ return snprintf(page, count, "%d\n", !dev->od_scrub.os_no_scrub);
}
-int lprocfs_osd_rd_fstype(char *page, char **start, off_t off, int count,
- int *eof, void *data)
+static int lprocfs_osd_wr_auto_scrub(struct file *file, const char *buffer,
+ unsigned long count, void *data)
{
- struct obd_device *osd = data;
+ struct osd_device *dev = data;
+ int val, rc;
- LASSERT(osd != NULL);
- return snprintf(page, count, "ldiskfs\n");
+ LASSERT(dev != NULL);
+ if (unlikely(dev->od_mount == NULL))
+ return -EINPROGRESS;
+
+ rc = lprocfs_write_helper(buffer, count, &val);
+ if (rc)
+ return rc;
+
+ dev->od_scrub.os_no_scrub = !val;
+ return count;
}
-static int lprocfs_osd_rd_mntdev(char *page, char **start, off_t off, int count,
- int *eof, void *data)
+static int lprocfs_osd_rd_oi_scrub(char *page, char **start, off_t off,
+ int count, int *eof, void *data)
{
- struct osd_device *osd = data;
+ struct osd_device *dev = data;
- LASSERT(osd != NULL);
- LASSERT(osd->od_mount->lmi_mnt->mnt_devname);
- *eof = 1;
+ LASSERT(dev != NULL);
+ if (unlikely(dev->od_mount == NULL))
+ return -EINPROGRESS;
- return snprintf(page, count, "%s\n",
- osd->od_mount->lmi_mnt->mnt_devname);
+ *eof = 1;
+ return osd_scrub_dump(dev, page, count);
}
struct lprocfs_vars lprocfs_osd_obd_vars[] = {
{ "filesfree", lprocfs_osd_rd_filesfree, 0, 0 },
{ "fstype", lprocfs_osd_rd_fstype, 0, 0 },
{ "mntdev", lprocfs_osd_rd_mntdev, 0, 0 },
- { 0 }
+#ifdef HAVE_LDISKFS_PDO
+ { "pdo", lprocfs_osd_rd_pdo, lprocfs_osd_wr_pdo, 0 },
+#endif
+ { "auto_scrub", lprocfs_osd_rd_auto_scrub,
+ lprocfs_osd_wr_auto_scrub, 0 },
+ { "oi_scrub", lprocfs_osd_rd_oi_scrub, 0, 0 },
+ { 0 }
};
struct lprocfs_vars lprocfs_osd_module_vars[] = {