Whamcloud - gitweb
LU-11771 osd: avoid use of HZ in brw_stats
[fs/lustre-release.git] / lustre / osd-zfs / osd_lproc.c
index 0854ce6..895eb13 100644 (file)
  *
  * You should have received a copy of the GNU General Public License
  * version 2 along with this program; If not, see
- * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
- *
- * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
- * CA 95054 USA or visit www.sun.com if you need additional information or
- * have any questions.
+ * http://www.gnu.org/licenses/gpl-2.0.html
  *
  * GPL HEADER END
  */
 /*
  * Copyright (c) 2008, 2010, Oracle and/or its affiliates. All rights reserved.
  * Use is subject to license terms.
- */
-/*
- * Copyright (c) 2012, Intel Corporation.
- * Use is subject to license terms.
+ *
+ * Copyright (c) 2012, 2017, Intel Corporation.
  */
 /*
  * This file is part of Lustre, http://www.lustre.org/
  * Author: Mike Pershin <tappro@whamcloud.com>
  */
 
-#define DEBUG_SUBSYSTEM S_CLASS
+#define DEBUG_SUBSYSTEM S_OSD
 
 #include <obd.h>
 #include <obd_class.h>
 #include <lprocfs_status.h>
-#include <lustre/lustre_idl.h>
+#include <lustre_scrub.h>
 
 #include "osd_internal.h"
 
-#ifdef LPROCFS
+#ifdef CONFIG_PROC_FS
+
+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 %3u %3u   | %4lu %3u %3u\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 timespec64 now;
+
+       /* this sampling races with updates */
+       ktime_get_real_ts64(&now);
+       seq_printf(seq, "snapshot_time:         %llu.%09lu (secs.nsecs)\n",
+                  (s64)now.tv_sec, now.tv_nsec);
+
+       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);
+#if 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);
+#endif
+       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], 1);
+
+       display_brw_stats(seq, "disk I/O size", "ios",
+                         &brw_stats->hist[BRW_R_DISK_IOSIZE],
+                         &brw_stats->hist[BRW_W_DISK_IOSIZE], 1);
+}
+
+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 __user *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 result;
+       int result, i;
        ENTRY;
 
+       for (i = 0; i < BRW_LAST; i++)
+               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",
@@ -99,6 +197,8 @@ static int osd_stats_init(struct osd_device *osd)
                                LPROCFS_CNTR_AVGMINMAX,
                                "thandle_closing", "usec");
 #endif
+               result = lprocfs_seq_create(osd->od_proc_entry, "brw_stats",
+                                           0644, &osd_brw_stats_fops, osd);
        } else {
                result = -ENOMEM;
        }
@@ -107,27 +207,77 @@ out:
        RETURN(result);
 }
 
-static int lprocfs_osd_rd_fstype(char *page, char **start, off_t off,
-                               int count, int *eof, void *data)
+static int zfs_osd_auto_scrub_seq_show(struct seq_file *m, void *data)
 {
-       return snprintf(page, count, "zfs\n");
+       struct osd_device *dev = osd_dt_dev((struct dt_device *)m->private);
+
+       LASSERT(dev != NULL);
+       if (!dev->od_os)
+               return -EINPROGRESS;
+
+       seq_printf(m, "%lld\n", dev->od_auto_scrub_interval);
+       return 0;
 }
 
-static int lprocfs_osd_rd_mntdev(char *page, char **start, off_t off, int count,
-                               int *eof, void *data)
+static ssize_t
+zfs_osd_auto_scrub_seq_write(struct file *file, const char __user *buffer,
+                            size_t count, loff_t *off)
 {
-       struct osd_device *osd = osd_dt_dev((struct dt_device *)data);
+       struct seq_file *m = file->private_data;
+       struct dt_device *dt = m->private;
+       struct osd_device *dev = osd_dt_dev(dt);
+       int rc;
+       __s64 val;
 
-       LASSERT(osd != NULL);
-       *eof = 1;
+       LASSERT(dev != NULL);
+       if (!dev->od_os)
+               return -EINPROGRESS;
+
+       rc = kstrtoull_from_user(buffer, count, 0, &val);
+       if (rc)
+               return rc;
+
+       dev->od_auto_scrub_interval = val;
+       return count;
+}
+LPROC_SEQ_FOPS(zfs_osd_auto_scrub);
+
+static int zfs_osd_oi_scrub_seq_show(struct seq_file *m, void *data)
+{
+       struct osd_device *dev = osd_dt_dev((struct dt_device *)m->private);
+
+       LASSERT(dev != NULL);
+       if (!dev->od_os)
+               return -EINPROGRESS;
+
+       scrub_dump(m, &dev->od_scrub);
+       return 0;
+}
+LPROC_SEQ_FOPS_RO(zfs_osd_oi_scrub);
+
+static int zfs_osd_fstype_seq_show(struct seq_file *m, void *data)
+{
+       seq_puts(m, "zfs\n");
+       return 0;
+}
+LPROC_SEQ_FOPS_RO(zfs_osd_fstype);
+
+static int zfs_osd_mntdev_seq_show(struct seq_file *m, void *data)
+{
+       struct osd_device *osd = osd_dt_dev((struct dt_device *)m->private);
 
-       return snprintf(page, count, "%s\n", osd->od_mntdev);
+       LASSERT(osd != NULL);
+       seq_printf(m, "%s\n", osd->od_mntdev);
+       return 0;
 }
+LPROC_SEQ_FOPS_RO(zfs_osd_mntdev);
 
-static int lprocfs_osd_wr_force_sync(struct file *file, const char *buffer,
-                                       unsigned long count, void *data)
+static ssize_t
+lprocfs_osd_force_sync_seq_write(struct file *file, const char __user *buffer,
+                               size_t count, loff_t *off)
 {
-       struct dt_device  *dt = data;
+       struct seq_file   *m = file->private_data;
+       struct dt_device  *dt = m->private;
        struct lu_env      env;
        int rc;
 
@@ -139,50 +289,115 @@ static int lprocfs_osd_wr_force_sync(struct file *file, const char *buffer,
 
        return rc == 0 ? count : rc;
 }
+LPROC_SEQ_FOPS_WR_ONLY(zfs, osd_force_sync);
+
+static int zfs_osd_index_backup_seq_show(struct seq_file *m, void *data)
+{
+       struct osd_device *dev = osd_dt_dev((struct dt_device *)m->private);
+
+       LASSERT(dev != NULL);
+       if (!dev->od_os)
+               return -EINPROGRESS;
+
+       seq_printf(m, "%d\n", dev->od_index_backup_policy);
+       return 0;
+}
 
-static int lprocfs_osd_rd_iused_est(char *page, char **start, off_t off, int count,
-                                       int *eof, void *data)
+static ssize_t zfs_osd_index_backup_seq_write(struct file *file,
+                                             const char __user *buffer,
+                                             size_t count, loff_t *off)
 {
-       struct osd_device *osd = osd_dt_dev((struct dt_device *)data);
+       struct seq_file *m = file->private_data;
+       struct dt_device *dt = m->private;
+       struct osd_device *dev = osd_dt_dev(dt);
+       int val;
+       int rc;
+
+       LASSERT(dev != NULL);
+       if (!dev->od_os)
+               return -EINPROGRESS;
+
+       rc = kstrtoint_from_user(buffer, count, 0, &val);
+       if (rc)
+               return rc;
+
+       dev->od_index_backup_policy = val;
+       return count;
+}
+LPROC_SEQ_FOPS(zfs_osd_index_backup);
+
+static int zfs_osd_readcache_seq_show(struct seq_file *m, void *data)
+{
+       struct osd_device *osd = osd_dt_dev((struct dt_device *)m->private);
+
        LASSERT(osd != NULL);
+       if (unlikely(osd->od_os == NULL))
+               return -EINPROGRESS;
 
-       return snprintf(page, count, "%d\n", osd->od_quota_iused_est);
+       seq_printf(m, "%llu\n", osd->od_readcache_max_filesize);
+       return 0;
 }
 
-static int lprocfs_osd_wr_iused_est(struct file *file, const char *buffer,
-                                       unsigned long count, void *data)
+static ssize_t
+zfs_osd_readcache_seq_write(struct file *file, const char __user *buffer,
+                           size_t count, loff_t *off)
 {
-       struct osd_device *osd = osd_dt_dev((struct dt_device *)data);
-       int                rc, val;
+       struct seq_file *m = file->private_data;
+       struct dt_device *dt = m->private;
+       struct osd_device *osd = osd_dt_dev(dt);
+       s64 val;
+       int rc;
 
        LASSERT(osd != NULL);
+       if (unlikely(osd->od_os == NULL))
+               return -EINPROGRESS;
 
-       rc = lprocfs_write_helper(buffer, count, &val);
+       rc = lprocfs_str_with_units_to_s64(buffer, count, &val, '1');
        if (rc)
                return rc;
+       if (val < 0)
+               return -ERANGE;
 
-       osd->od_quota_iused_est = !!val;
-
+       osd->od_readcache_max_filesize = val > OSD_MAX_CACHE_SIZE ?
+                                        OSD_MAX_CACHE_SIZE : val;
        return count;
 }
+LPROC_SEQ_FOPS(zfs_osd_readcache);
 
-struct lprocfs_vars lprocfs_osd_obd_vars[] = {
-       { "blocksize",          lprocfs_dt_rd_blksize,  0, 0 },
-       { "kbytestotal",        lprocfs_dt_rd_kbytestotal,      0, 0 },
-       { "kbytesfree",         lprocfs_dt_rd_kbytesfree,       0, 0 },
-       { "kbytesavail",        lprocfs_dt_rd_kbytesavail,      0, 0 },
-       { "filestotal",         lprocfs_dt_rd_filestotal,       0, 0 },
-       { "filesfree",          lprocfs_dt_rd_filesfree,        0, 0 },
-       { "fstype",          lprocfs_osd_rd_fstype,      0, 0 },
-       { "mntdev",          lprocfs_osd_rd_mntdev,      0, 0 },
-       { "force_sync",      0, lprocfs_osd_wr_force_sync     },
-       { "quota_iused_estimate",  lprocfs_osd_rd_iused_est,
-               lprocfs_osd_wr_iused_est,   0, 0 },
-       { 0 }
-};
+LPROC_SEQ_FOPS_RO_TYPE(zfs, dt_blksize);
+LPROC_SEQ_FOPS_RO_TYPE(zfs, dt_kbytestotal);
+LPROC_SEQ_FOPS_RO_TYPE(zfs, dt_kbytesfree);
+LPROC_SEQ_FOPS_RO_TYPE(zfs, dt_kbytesavail);
+LPROC_SEQ_FOPS_RO_TYPE(zfs, dt_filestotal);
+LPROC_SEQ_FOPS_RO_TYPE(zfs, dt_filesfree);
 
-struct lprocfs_vars lprocfs_osd_module_vars[] = {
-       { "num_refs",        lprocfs_rd_numrefs,         0, 0 },
+struct lprocfs_vars lprocfs_osd_obd_vars[] = {
+       { .name =       "blocksize",
+         .fops =       &zfs_dt_blksize_fops            },
+       { .name =       "kbytestotal",
+         .fops =       &zfs_dt_kbytestotal_fops        },
+       { .name =       "kbytesfree",
+         .fops =       &zfs_dt_kbytesfree_fops         },
+       { .name =       "kbytesavail",
+         .fops =       &zfs_dt_kbytesavail_fops        },
+       { .name =       "filestotal",
+         .fops =       &zfs_dt_filestotal_fops         },
+       { .name =       "filesfree",
+         .fops =       &zfs_dt_filesfree_fops          },
+       { .name =       "auto_scrub",
+         .fops =       &zfs_osd_auto_scrub_fops        },
+       { .name =       "oi_scrub",
+         .fops =       &zfs_osd_oi_scrub_fops          },
+       { .name =       "fstype",
+         .fops =       &zfs_osd_fstype_fops            },
+       { .name =       "mntdev",
+         .fops =       &zfs_osd_mntdev_fops            },
+       { .name =       "force_sync",
+         .fops =       &zfs_osd_force_sync_fops        },
+       { .name =       "index_backup",
+         .fops =       &zfs_osd_index_backup_fops      },
+       { .name =       "readcache_max_filesize",
+         .fops =       &zfs_osd_readcache_fops },
        { 0 }
 };
 
@@ -203,7 +418,8 @@ int osd_procfs_init(struct osd_device *osd, const char *name)
        LASSERT(type != NULL);
 
        osd->od_proc_entry = lprocfs_register(name, type->typ_procroot,
-                       lprocfs_osd_obd_vars, &osd->od_dt_dev);
+                                             lprocfs_osd_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, name);