*/
/*
* This file is part of Lustre, http://www.lustre.org/
- * Lustre is a trademark of Sun Microsystems, Inc.
*
* lustre/osd/osd_lproc.c
*
#include "osd_internal.h"
-#ifdef CONFIG_PROC_FS
-
void osd_brw_stats_update(struct osd_device *osd, struct osd_iobuf *iobuf)
{
- struct brw_stats *s = &osd->od_brw_stats;
+ struct brw_stats *bs = &osd->od_brw_stats;
sector_t *last_block = NULL;
- struct page **pages = iobuf->dr_pages;
- struct page *last_page = NULL;
- unsigned long discont_pages = 0;
- unsigned long discont_blocks = 0;
+ struct page **pages = iobuf->dr_pages;
+ struct page *last_page = NULL;
+ unsigned long discont_pages = 0;
+ unsigned long discont_blocks = 0;
sector_t *blocks = iobuf->dr_blocks;
- int i, nr_pages = iobuf->dr_npages;
- int blocks_per_page;
- int rw = iobuf->dr_rw;
+ int i, nr_pages = iobuf->dr_npages;
+ int blocks_per_page;
+ int rw = iobuf->dr_rw;
- if (unlikely(nr_pages == 0))
- return;
+ if (unlikely(nr_pages == 0))
+ return;
blocks_per_page = 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);
-}
-
-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: %lld.%09ld (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);
-
- 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], 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]);
+ lprocfs_oh_tally_log2(&bs->bs_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++;
+ }
+ }
- return len;
+ lprocfs_oh_tally(&bs->bs_hist[BRW_R_DISCONT_PAGES+rw], discont_pages);
+ lprocfs_oh_tally(&bs->bs_hist[BRW_R_DISCONT_BLOCKS+rw], discont_blocks);
}
-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++)
- spin_lock_init(&osd->od_brw_stats.hist[i].oh_lock);
+ int result = -ENOMEM;
+ ENTRY;
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);
-
+ if (osd->od_stats) {
lprocfs_counter_init(osd->od_stats, LPROC_OSD_GET_PAGE,
LPROCFS_CNTR_AVGMINMAX|LPROCFS_CNTR_STDDEV,
"get_page", "usec");
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;
+ result = 0;
+ }
-out:
- RETURN(result);
+ ldebugfs_register_osd_stats(osd->od_dt_dev.dd_debugfs_entry,
+ &osd->od_brw_stats, osd->od_stats);
+
+ RETURN(result);
}
static ssize_t fstype_show(struct kobject *kobj, struct attribute *attr,
if (rc)
return rc;
- osd->od_read_cache = val;
+ osd->od_read_cache = !!val;
return count;
}
LUSTRE_RW_ATTR(read_cache_enable);
if (rc)
return rc;
- osd->od_writethrough_cache = val;
+ osd->od_writethrough_cache = !!val;
return count;
}
LUSTRE_RW_ATTR(writethrough_cache_enable);
+static ssize_t fallocate_zero_blocks_show(struct kobject *kobj,
+ struct attribute *attr,
+ char *buf)
+{
+ struct dt_device *dt = container_of(kobj, struct dt_device,
+ dd_kobj);
+ struct osd_device *osd = osd_dt_dev(dt);
+
+ LASSERT(osd);
+ if (unlikely(!osd->od_mnt))
+ return -EINPROGRESS;
+
+ return scnprintf(buf, PAGE_SIZE, "%d\n", osd->od_fallocate_zero_blocks);
+}
+
+/*
+ * Set how fallocate() interacts with the backing filesystem:
+ * -1: fallocate is disabled and returns -EOPNOTSUPP
+ * 0: fallocate allocates unwritten extents (like ext4)
+ * 1: fallocate zeroes allocated extents on disk
+ */
+static ssize_t fallocate_zero_blocks_store(struct kobject *kobj,
+ struct attribute *attr,
+ const char *buffer, size_t count)
+{
+ struct dt_device *dt = container_of(kobj, struct dt_device,
+ dd_kobj);
+ struct osd_device *osd = osd_dt_dev(dt);
+ long val;
+ int rc;
+
+ LASSERT(osd);
+ if (unlikely(!osd->od_mnt))
+ return -EINPROGRESS;
+
+ rc = kstrtol(buffer, 0, &val);
+ if (rc)
+ return rc;
+
+ if (val < -1 || val > 1)
+ return -EINVAL;
+
+ osd->od_fallocate_zero_blocks = val;
+ return count;
+}
+LUSTRE_RW_ATTR(fallocate_zero_blocks);
+
ssize_t force_sync_store(struct kobject *kobj, struct attribute *attr,
const char *buffer, size_t count)
{
if (unlikely(!dev->od_mnt))
return -EINPROGRESS;
- return sprintf(buf, "%lld\n", dev->od_auto_scrub_interval);
+ return scnprintf(buf, PAGE_SIZE, "%lld\n",
+ dev->od_scrub.os_scrub.os_auto_scrub_interval);
}
static ssize_t auto_scrub_store(struct kobject *kobj, struct attribute *attr,
if (rc)
return rc;
- dev->od_auto_scrub_interval = val;
+ dev->od_scrub.os_scrub.os_auto_scrub_interval = val;
return count;
}
LUSTRE_RW_ATTR(auto_scrub);
}
LUSTRE_RW_ATTR(full_scrub_threshold_rate);
+static ssize_t extent_bytes_allocation_show(struct kobject *kobj,
+ struct attribute *attr, char *buf)
+{
+ struct dt_device *dt = container_of(kobj, struct dt_device,
+ dd_kobj);
+ struct osd_device *dev = osd_dt_dev(dt);
+ int i;
+ unsigned int min = (unsigned int)(~0), cur;
+
+ for_each_online_cpu(i) {
+ cur = *per_cpu_ptr(dev->od_extent_bytes_percpu, i);
+ if (cur < min)
+ min = cur;
+ }
+ return snprintf(buf, PAGE_SIZE, "%u\n", min);
+}
+LUSTRE_RO_ATTR(extent_bytes_allocation);
+
static int ldiskfs_osd_oi_scrub_seq_show(struct seq_file *m, void *data)
{
struct osd_device *dev = osd_dt_dev((struct dt_device *)m->private);
struct seq_file *m = file->private_data;
struct dt_device *dt = m->private;
struct osd_device *osd = osd_dt_dev(dt);
+ char kernbuf[22] = "";
u64 val;
int rc;
if (unlikely(osd->od_mnt == NULL))
return -EINPROGRESS;
- rc = lprocfs_str_with_units_to_u64(buffer, count, &val, '1');
- if (rc)
+ if (count >= sizeof(kernbuf))
+ return -EINVAL;
+
+ if (copy_from_user(kernbuf, buffer, count))
+ return -EFAULT;
+ kernbuf[count] = 0;
+
+ rc = sysfs_memparse(kernbuf, count, &val, "B");
+ if (rc < 0)
return rc;
osd->od_readcache_max_filesize = val > OSD_MAX_CACHE_SIZE ?
LDEBUGFS_SEQ_FOPS(ldiskfs_osd_readcache);
+static int ldiskfs_osd_readcache_max_io_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_mnt == NULL))
+ return -EINPROGRESS;
+
+ seq_printf(m, "%lu\n", osd->od_readcache_max_iosize >> 20);
+ return 0;
+}
+
+static ssize_t
+ldiskfs_osd_readcache_max_io_seq_write(struct file *file,
+ const char __user *buffer,
+ size_t count, loff_t *off)
+{
+ struct seq_file *m = file->private_data;
+ struct dt_device *dt = m->private;
+ struct osd_device *osd = osd_dt_dev(dt);
+ char kernbuf[22] = "";
+ u64 val;
+ int rc;
+
+ LASSERT(osd != NULL);
+ if (unlikely(osd->od_mnt == NULL))
+ return -EINPROGRESS;
+
+ if (count >= sizeof(kernbuf))
+ return -EINVAL;
+
+ if (copy_from_user(kernbuf, buffer, count))
+ return -EFAULT;
+ kernbuf[count] = 0;
+
+ rc = sysfs_memparse(kernbuf, count, &val, "MiB");
+ if (rc < 0)
+ return rc;
+
+ if (val > PTLRPC_MAX_BRW_SIZE)
+ return -ERANGE;
+ osd->od_readcache_max_iosize = val;
+ return count;
+}
+
+LDEBUGFS_SEQ_FOPS(ldiskfs_osd_readcache_max_io);
+
+static int ldiskfs_osd_writethrough_max_io_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_mnt == NULL))
+ return -EINPROGRESS;
+
+ seq_printf(m, "%lu\n", osd->od_writethrough_max_iosize >> 20);
+ return 0;
+}
+
+static ssize_t
+ldiskfs_osd_writethrough_max_io_seq_write(struct file *file,
+ const char __user *buffer,
+ size_t count, loff_t *off)
+{
+ struct seq_file *m = file->private_data;
+ struct dt_device *dt = m->private;
+ struct osd_device *osd = osd_dt_dev(dt);
+ char kernbuf[22] = "";
+ u64 val;
+ int rc;
+
+ LASSERT(osd != NULL);
+ if (unlikely(osd->od_mnt == NULL))
+ return -EINPROGRESS;
+
+ if (count >= sizeof(kernbuf))
+ return -EINVAL;
+
+ if (copy_from_user(kernbuf, buffer, count))
+ return -EFAULT;
+ kernbuf[count] = 0;
+
+ rc = sysfs_memparse(kernbuf, count, &val, "MiB");
+ if (rc < 0)
+ return rc;
+
+ if (val > PTLRPC_MAX_BRW_SIZE)
+ return -ERANGE;
+ osd->od_writethrough_max_iosize = val;
+ return count;
+}
+
+LDEBUGFS_SEQ_FOPS(ldiskfs_osd_writethrough_max_io);
+
#if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(3, 0, 52, 0)
static ssize_t index_in_idif_show(struct kobject *kobj, struct attribute *attr,
char *buf)
}
LUSTRE_RW_ATTR(index_backup);
-struct lprocfs_vars lprocfs_osd_obd_vars[] = {
+struct ldebugfs_vars ldebugfs_osd_obd_vars[] = {
{ .name = "oi_scrub",
.fops = &ldiskfs_osd_oi_scrub_fops },
{ .name = "readcache_max_filesize",
.fops = &ldiskfs_osd_readcache_fops },
+ { .name = "readcache_max_io_mb",
+ .fops = &ldiskfs_osd_readcache_max_io_fops },
+ { .name = "writethrough_max_io_mb",
+ .fops = &ldiskfs_osd_writethrough_max_io_fops },
{ NULL }
};
&lustre_attr_writethrough_cache_enable.attr,
&lustre_attr_fstype.attr,
&lustre_attr_mntdev.attr,
+ &lustre_attr_fallocate_zero_blocks.attr,
&lustre_attr_force_sync.attr,
&lustre_attr_nonrotational.attr,
&lustre_attr_index_backup.attr,
&lustre_attr_pdo.attr,
&lustre_attr_full_scrub_ratio.attr,
&lustre_attr_full_scrub_threshold_rate.attr,
+ &lustre_attr_extent_bytes_allocation.attr,
NULL,
};
LASSERT(name);
LASSERT(type);
- LCONSOLE_INFO("osd-ldiskfs create tunables for %s\n", name);
+ CDEBUG(D_CONFIG, "%s: register osd-ldiskfs tunable parameters\n", name);
/* put reference taken by class_search_type */
kobject_put(&type->typ_kobj);
osd->od_dt_dev.dd_ktype.default_attrs = ldiskfs_attrs;
rc = dt_tunables_init(&osd->od_dt_dev, type, name,
- lprocfs_osd_obd_vars);
+ ldebugfs_osd_obd_vars);
if (rc) {
CERROR("%s: cannot setup sysfs / debugfs entry: %d\n",
name, rc);
return dt_tunables_fini(&osd->od_dt_dev);
}
-#endif