*
* 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
+ * http://www.gnu.org/licenses/gpl-2.0.html
*
* GPL HEADER END
*/
/*
- * Copyright (c) 2012, 2014, Intel Corporation.
+ * Copyright (c) 2012, 2016, Intel Corporation.
* Use is subject to license terms.
*
* Author: Niu Yawei <niu@whamcloud.com>
#define DEBUG_SUBSYSTEM S_CLASS
-
#include <obd_class.h>
#include <lprocfs_status.h>
-#include <lustre/lustre_idl.h>
#ifdef CONFIG_PROC_FS
*/
struct job_stat {
- struct hlist_node js_hash;
- struct list_head js_list;
- atomic_t js_refcount;
- char js_jobid[LUSTRE_JOBID_SIZE];
- time_t js_timestamp; /* seconds */
- struct lprocfs_stats *js_stats;
- struct obd_job_stats *js_jobstats;
+ struct hlist_node js_hash; /* hash struct for this jobid */
+ struct list_head js_list; /* on ojs_list, with ojs_lock */
+ atomic_t js_refcount; /* num users of this struct */
+ char js_jobid[LUSTRE_JOBID_SIZE]; /* job name + NUL*/
+ ktime_t js_time_init; /* time of initial stat*/
+ ktime_t js_time_latest; /* time of most recent stat*/
+ struct lprocfs_stats *js_stats; /* per-job statistics */
+ struct obd_job_stats *js_jobstats; /* for accessing ojs_lock */
};
-static unsigned job_stat_hash(cfs_hash_t *hs, const void *key, unsigned mask)
+static unsigned
+job_stat_hash(struct cfs_hash *hs, const void *key, unsigned mask)
{
return cfs_hash_djb2_hash(key, strlen(key), mask);
}
return hlist_entry(hnode, struct job_stat, js_hash);
}
-static void job_stat_get(cfs_hash_t *hs, struct hlist_node *hnode)
+static void job_stat_get(struct cfs_hash *hs, struct hlist_node *hnode)
{
struct job_stat *job;
job = hlist_entry(hnode, struct job_stat, js_hash);
static void job_free(struct job_stat *job)
{
LASSERT(atomic_read(&job->js_refcount) == 0);
- LASSERT(job->js_jobstats);
+ LASSERT(job->js_jobstats != NULL);
write_lock(&job->js_jobstats->ojs_lock);
list_del_init(&job->js_list);
job_free(job);
}
-static void job_stat_put_locked(cfs_hash_t *hs, struct hlist_node *hnode)
+static void job_stat_put_locked(struct cfs_hash *hs, struct hlist_node *hnode)
{
struct job_stat *job;
job = hlist_entry(hnode, struct job_stat, js_hash);
job_putref(job);
}
-static void job_stat_exit(cfs_hash_t *hs, struct hlist_node *hnode)
+static void job_stat_exit(struct cfs_hash *hs, struct hlist_node *hnode)
{
CERROR("should not have any items\n");
}
-static cfs_hash_ops_t job_stats_hash_ops = {
+static struct cfs_hash_ops job_stats_hash_ops = {
.hs_hash = job_stat_hash,
.hs_key = job_stat_key,
.hs_keycmp = job_stat_keycmp,
.hs_exit = job_stat_exit,
};
-static int job_iter_callback(cfs_hash_t *hs, cfs_hash_bd_t *bd,
- struct hlist_node *hnode, void *data)
+/**
+ * Jobstats expiry iterator to clean up old jobids
+ *
+ * Called for each job_stat structure on this device, it should delete stats
+ * older than the specified \a oldest_time in seconds. If \a oldest_time is
+ * in the future then this will delete all statistics (e.g. during shutdown).
+ *
+ * \param[in] hs hash of all jobids on this device
+ * \param[in] bd hash bucket containing this jobid
+ * \param[in] hnode hash structure for this jobid
+ * \param[in] data pointer to stats expiry time in seconds
+ */
+static int job_cleanup_iter_callback(struct cfs_hash *hs,
+ struct cfs_hash_bd *bd,
+ struct hlist_node *hnode, void *data)
{
- time_t oldest = *((time_t *)data);
+ ktime_t oldest_time = *((ktime_t *)data);
struct job_stat *job;
job = hlist_entry(hnode, struct job_stat, js_hash);
- if (!oldest || job->js_timestamp < oldest)
+ if (ktime_before(job->js_time_latest, oldest_time))
cfs_hash_bd_del_locked(hs, bd, hnode);
return 0;
}
-static void lprocfs_job_cleanup(struct obd_job_stats *stats, bool force)
+/**
+ * Clean up jobstats that were updated more than \a before seconds ago.
+ *
+ * Since this function may be called frequently, do not scan all of the
+ * jobstats on each call, only twice per cleanup interval. That means stats
+ * may be on average around cleanup_interval / 4 older than the cleanup
+ * interval, but that is not considered harmful.
+ *
+ * The value stored in ojs_cleanup_interval is how often to perform a cleanup
+ * scan, and 1/2 of the maximum age of the individual statistics. This is
+ * done rather than dividing the interval by two each time, because it is
+ * much easier to do the division when the value is initially set (in seconds)
+ * rather than after it has been converted to ktime_t, and maybe a bit faster.
+ *
+ * If \a clear is true then this will force clean up all jobstats
+ * (e.g. at shutdown).
+ *
+ * If there is already another thread doing jobstats cleanup, don't try to
+ * do this again in the current thread unless this is a force cleanup.
+ *
+ * \param[in] stats stucture tracking all job stats for this device
+ * \param[in] clear clear all job stats if true
+ */
+static void lprocfs_job_cleanup(struct obd_job_stats *stats, bool clear)
{
- time_t oldest, now;
+ ktime_t cleanup_interval = stats->ojs_cleanup_interval;
+ ktime_t now = ktime_get();
+ ktime_t oldest;
- if (stats->ojs_cleanup_interval == 0)
- return;
+ if (likely(!clear)) {
+ /* ojs_cleanup_interval of zero means never clean up stats */
+ if (ktime_to_ns(cleanup_interval) == 0)
+ return;
+
+ if (ktime_before(now, ktime_add(stats->ojs_cleanup_last,
+ cleanup_interval)))
+ return;
+
+ if (stats->ojs_cleaning)
+ return;
+ }
- now = cfs_time_current_sec();
- if (!force && now < stats->ojs_last_cleanup +
- stats->ojs_cleanup_interval)
+ write_lock(&stats->ojs_lock);
+ if (!clear && stats->ojs_cleaning) {
+ write_unlock(&stats->ojs_lock);
return;
+ }
- oldest = now - stats->ojs_cleanup_interval;
- cfs_hash_for_each_safe(stats->ojs_hash, job_iter_callback,
+ stats->ojs_cleaning = true;
+ write_unlock(&stats->ojs_lock);
+
+ /* Can't hold ojs_lock over hash iteration, since it is grabbed by
+ * job_cleanup_iter_callback()
+ * ->cfs_hash_bd_del_locked()
+ * ->job_putref()
+ * ->job_free()
+ *
+ * Holding ojs_lock isn't necessary for safety of the hash iteration,
+ * since locking of the hash is handled internally, but there isn't
+ * any benefit to having multiple threads doing cleanup at one time.
+ *
+ * Subtract twice the cleanup_interval, since it is 1/2 the maximum age.
+ */
+ oldest = ktime_sub(now, ktime_add(cleanup_interval, cleanup_interval));
+ cfs_hash_for_each_safe(stats->ojs_hash, job_cleanup_iter_callback,
&oldest);
- stats->ojs_last_cleanup = cfs_time_current_sec();
+
+ write_lock(&stats->ojs_lock);
+ stats->ojs_cleaning = false;
+ stats->ojs_cleanup_last = ktime_get();
+ write_unlock(&stats->ojs_lock);
}
static struct job_stat *job_alloc(char *jobid, struct obd_job_stats *jobs)
{
struct job_stat *job;
- LASSERT(jobs->ojs_cntr_num && jobs->ojs_cntr_init_fn);
-
OBD_ALLOC_PTR(job);
if (job == NULL)
return NULL;
return NULL;
}
- jobs->ojs_cntr_init_fn(job->js_stats);
+ jobs->ojs_cntr_init_fn(job->js_stats, 0);
- memcpy(job->js_jobid, jobid, LUSTRE_JOBID_SIZE);
- job->js_timestamp = cfs_time_current_sec();
+ memcpy(job->js_jobid, jobid, sizeof(job->js_jobid));
+ job->js_time_init = ktime_get();
+ job->js_time_latest = job->js_time_init;
job->js_jobstats = jobs;
INIT_HLIST_NODE(&job->js_hash);
INIT_LIST_HEAD(&job->js_list);
struct job_stat *job, *job2;
ENTRY;
- LASSERT(stats && stats->ojs_hash);
+ LASSERT(stats != NULL);
+ LASSERT(stats->ojs_hash != NULL);
- lprocfs_job_cleanup(stats, false);
+ if (event >= stats->ojs_cntr_num)
+ RETURN(-EINVAL);
- if (!jobid || !strlen(jobid))
+ if (jobid == NULL || strlen(jobid) == 0)
RETURN(-EINVAL);
if (strlen(jobid) >= LUSTRE_JOBID_SIZE) {
if (job)
goto found;
+ lprocfs_job_cleanup(stats, false);
+
job = job_alloc(jobid, stats);
if (job == NULL)
RETURN(-ENOMEM);
found:
LASSERT(stats == job->js_jobstats);
- LASSERT(stats->ojs_cntr_num > event);
- job->js_timestamp = cfs_time_current_sec();
+ job->js_time_latest = ktime_get();
lprocfs_counter_add(job->js_stats, event, amount);
job_putref(job);
+
RETURN(0);
}
EXPORT_SYMBOL(lprocfs_job_stats_log);
void lprocfs_job_stats_fini(struct obd_device *obd)
{
struct obd_job_stats *stats = &obd->u.obt.obt_jobstats;
- time_t oldest = 0;
if (stats->ojs_hash == NULL)
return;
- cfs_hash_for_each_safe(stats->ojs_hash, job_iter_callback, &oldest);
+
+ lprocfs_job_cleanup(stats, true);
cfs_hash_putref(stats->ojs_hash);
stats->ojs_hash = NULL;
LASSERT(list_empty(&stats->ojs_list));
* Example of output on MDT:
*
* job_stats:
- * - job_id: test_id.222.25844
- * snapshot_time: 1322494486
- * open: { samples: 3, unit: reqs }
- * close: { samples: 3, unit: reqs }
+ * - job_id: dd.4854
+ * snapshot_time: 1322494486.123456789
+ * start_time: 1322494476.012345678
+ * elapsed_time: 10.111111111
+ * open: { samples: 1, unit: reqs }
+ * close: { samples: 1, unit: reqs }
* mknod: { samples: 0, unit: reqs }
* link: { samples: 0, unit: reqs }
* unlink: { samples: 0, unit: reqs }
* mkdir: { samples: 0, unit: reqs }
* rmdir: { samples: 0, unit: reqs }
- * rename: { samples: 1, unit: reqs }
- * getattr: { samples: 7, unit: reqs }
+ * rename: { samples: 0, unit: reqs }
+ * getattr: { samples: 1, unit: reqs }
* setattr: { samples: 0, unit: reqs }
* getxattr: { samples: 0, unit: reqs }
* setxattr: { samples: 0, unit: reqs }
* Example of output on OST:
*
* job_stats:
- * - job_id 4854
- * snapshot_time: 1322494602
- * read: { samples: 0, unit: bytes, min: 0, max: 0, sum: 0 }
- * write: { samples: 1, unit: bytes, min: 10, max: 10, sum: 10 }
- * setattr: { samples: 0, unit: reqs }
- * punch: { samples: 0, unit: reqs }
- * sync: { samples: 0, unit: reqs }
+ * - job_id dd.4854
+ * snapshot_time: 1322494602.123456789
+ * start_time: 1322494592.987654321
+ * elapsed_time: 9.135802468
+ * read: { samples: 0, unit: bytes, min: 0, max: 0, sum: 0 }
+ * write: { samples: 1, unit: bytes, min: 4096, max: 4096, sum: 4096 }
+ * setattr: { samples: 0, unit: reqs }
+ * punch: { samples: 0, unit: reqs }
+ * sync: { samples: 0, unit: reqs }
*/
static const char spaces[] = " ";
static int lprocfs_jobstats_seq_show(struct seq_file *p, void *v)
{
- struct job_stat *job = v;
- struct lprocfs_stats *s;
- struct lprocfs_counter ret;
- struct lprocfs_counter_header *cntr_header;
- int i;
+ struct job_stat *job = v;
+ struct lprocfs_stats *s;
+ struct lprocfs_counter ret;
+ struct lprocfs_counter_header *cntr_header;
+ int i;
if (v == SEQ_START_TOKEN) {
seq_printf(p, "job_stats:\n");
return 0;
}
- seq_printf(p, "- %-16s %s\n", "job_id:", job->js_jobid);
- seq_printf(p, " %-16s %ld\n", "snapshot_time:", job->js_timestamp);
+ /* Replace the non-printable character in jobid with '?', so
+ * that the output of jobid will be confined in single line. */
+ seq_printf(p, "- %-16s ", "job_id:");
+ for (i = 0; i < strlen(job->js_jobid); i++) {
+ if (isprint(job->js_jobid[i]) != 0)
+ seq_putc(p, job->js_jobid[i]);
+ else
+ seq_putc(p, '?');
+ }
+ seq_putc(p, '\n');
+
+ lprocfs_stats_header(p, job->js_time_latest, job->js_time_init, 16,
+ ":", true);
s = job->js_stats;
for (i = 0; i < s->ls_num; i++) {
cntr_header = &s->ls_cnt_header[i];
lprocfs_stats_collect(s, i, &ret);
- seq_printf(p, " %s:%.*s { samples: %11"LPF64"u",
+ seq_printf(p, " %s:%.*s { samples: %11llu",
cntr_header->lc_name,
width(cntr_header->lc_name, 15), spaces,
ret.lc_count);
seq_printf(p, ", unit: %5s", cntr_header->lc_units);
if (cntr_header->lc_config & LPROCFS_CNTR_AVGMINMAX) {
- seq_printf(p, ", min:%8"LPF64"u, max:%8"LPF64"u,"
- " sum:%16"LPF64"u",
+ seq_printf(p, ", min: %8llu, max: %8llu, sum: %16llu",
ret.lc_count ? ret.lc_min : 0,
ret.lc_count ? ret.lc_max : 0,
ret.lc_count ? ret.lc_sum : 0);
}
if (cntr_header->lc_config & LPROCFS_CNTR_STDDEV) {
- seq_printf(p, ", sumsq: %18"LPF64"u",
+ seq_printf(p, ", sumsq: %18llu",
ret.lc_count ? ret.lc_sumsquare : 0);
}
}
static const struct seq_operations lprocfs_jobstats_seq_sops = {
- start: lprocfs_jobstats_seq_start,
- stop: lprocfs_jobstats_seq_stop,
- next: lprocfs_jobstats_seq_next,
- show: lprocfs_jobstats_seq_show,
+ .start = lprocfs_jobstats_seq_start,
+ .stop = lprocfs_jobstats_seq_stop,
+ .next = lprocfs_jobstats_seq_next,
+ .show = lprocfs_jobstats_seq_show,
};
static int lprocfs_jobstats_seq_open(struct inode *inode, struct file *file)
struct seq_file *seq;
int rc;
- rc = LPROCFS_ENTRY_CHECK(inode);
- if (rc < 0)
- return rc;
-
rc = seq_open(file, &lprocfs_jobstats_seq_sops);
if (rc)
return rc;
struct seq_file *seq = file->private_data;
struct obd_job_stats *stats = seq->private;
char jobid[LUSTRE_JOBID_SIZE];
- int all = 0;
struct job_stat *job;
if (len == 0 || len >= LUSTRE_JOBID_SIZE)
return -EINVAL;
+ if (stats->ojs_hash == NULL)
+ return -ENODEV;
+
if (copy_from_user(jobid, buf, len))
return -EFAULT;
jobid[len] = 0;
if (jobid[len - 1] == '\n')
jobid[len - 1] = 0;
- if (strcmp(jobid, "clear") == 0)
- all = 1;
+ if (strcmp(jobid, "clear") == 0) {
+ lprocfs_job_cleanup(stats, true);
- LASSERT(stats->ojs_hash);
- if (all) {
- time_t oldest = 0;
- cfs_hash_for_each_safe(stats->ojs_hash, job_iter_callback,
- &oldest);
return len;
}
- if (!strlen(jobid))
+ if (strlen(jobid) == 0)
return -EINVAL;
job = cfs_hash_lookup(stats->ojs_hash, jobid);
return len;
}
-static const struct file_operations lprocfs_jobstats_seq_fops = {
- .owner = THIS_MODULE,
- .open = lprocfs_jobstats_seq_open,
- .read = seq_read,
- .write = lprocfs_jobstats_seq_write,
- .llseek = seq_lseek,
- .release = lprocfs_seq_release,
+/**
+ * Clean up the seq file state when the /proc file is closed.
+ *
+ * This also expires old job stats from the cache after they have been
+ * printed in case the system is idle and not generating new jobstats.
+ *
+ * \param[in] inode struct inode for seq file being closed
+ * \param[in] file struct file for seq file being closed
+ *
+ * \retval 0 on success
+ * \retval negative errno on failure
+ */
+static int lprocfs_jobstats_seq_release(struct inode *inode, struct file *file)
+{
+ struct seq_file *seq = file->private_data;
+ struct obd_job_stats *stats = seq->private;
+
+ lprocfs_job_cleanup(stats, false);
+
+ return lprocfs_seq_release(inode, file);
+}
+
+static const struct proc_ops lprocfs_jobstats_seq_fops = {
+ PROC_OWNER(THIS_MODULE)
+ .proc_open = lprocfs_jobstats_seq_open,
+ .proc_read = seq_read,
+ .proc_write = lprocfs_jobstats_seq_write,
+ .proc_lseek = seq_lseek,
+ .proc_release = lprocfs_jobstats_seq_release,
};
int lprocfs_job_stats_init(struct obd_device *obd, int cntr_num,
LASSERT(obd->obd_proc_entry != NULL);
LASSERT(obd->obd_type->typ_name);
- if (strcmp(obd->obd_type->typ_name, LUSTRE_MDT_NAME) &&
- strcmp(obd->obd_type->typ_name, LUSTRE_OST_NAME)) {
- CERROR("Invalid obd device type.\n");
+ if (cntr_num <= 0)
+ RETURN(-EINVAL);
+
+ if (init_fn == NULL)
+ RETURN(-EINVAL);
+
+ /* Currently needs to be a target due to the use of obt_jobstats. */
+ if (strcmp(obd->obd_type->typ_name, LUSTRE_MDT_NAME) != 0 &&
+ strcmp(obd->obd_type->typ_name, LUSTRE_OST_NAME) != 0) {
+ CERROR("%s: invalid device type %s for job stats: rc = %d\n",
+ obd->obd_name, obd->obd_type->typ_name, -EINVAL);
RETURN(-EINVAL);
}
stats = &obd->u.obt.obt_jobstats;
rwlock_init(&stats->ojs_lock);
stats->ojs_cntr_num = cntr_num;
stats->ojs_cntr_init_fn = init_fn;
- stats->ojs_cleanup_interval = 600; /* 10 mins by default */
- stats->ojs_last_cleanup = cfs_time_current_sec();
+ /* Store 1/2 the actual interval, since we use that the most, and
+ * it is easier to work with.
+ */
+ stats->ojs_cleanup_interval = ktime_set(600 / 2, 0); /* default 10 min*/
+ stats->ojs_cleanup_last = ktime_get();
entry = lprocfs_add_simple(obd->obd_proc_entry, "job_stats", stats,
&lprocfs_jobstats_seq_fops);
RETURN(0);
}
EXPORT_SYMBOL(lprocfs_job_stats_init);
+#endif /* CONFIG_PROC_FS*/
-int lprocfs_job_interval_seq_show(struct seq_file *m, void *data)
+ssize_t job_cleanup_interval_show(struct kobject *kobj, struct attribute *attr,
+ char *buf)
{
- struct obd_device *obd = m->private;
+ struct obd_device *obd = container_of(kobj, struct obd_device,
+ obd_kset.kobj);
struct obd_job_stats *stats;
+ struct timespec64 ts;
- LASSERT(obd != NULL);
stats = &obd->u.obt.obt_jobstats;
- return seq_printf(m, "%d\n", stats->ojs_cleanup_interval);
+ ts = ktime_to_timespec64(stats->ojs_cleanup_interval);
+
+ return scnprintf(buf, PAGE_SIZE, "%lld\n", (long long)ts.tv_sec * 2);
}
-EXPORT_SYMBOL(lprocfs_job_interval_seq_show);
+EXPORT_SYMBOL(job_cleanup_interval_show);
-ssize_t
-lprocfs_job_interval_seq_write(struct file *file, const char *buffer,
- size_t count, loff_t *off)
+ssize_t job_cleanup_interval_store(struct kobject *kobj,
+ struct attribute *attr,
+ const char *buffer, size_t count)
{
- struct obd_device *obd = ((struct seq_file *)file->private_data)->private;
+ struct obd_device *obd = container_of(kobj, struct obd_device,
+ obd_kset.kobj);
struct obd_job_stats *stats;
- int val, rc;
+ unsigned int val;
+ int rc;
- LASSERT(obd != NULL);
stats = &obd->u.obt.obt_jobstats;
- rc = lprocfs_write_helper(buffer, count, &val);
+ rc = kstrtouint(buffer, 0, &val);
if (rc)
return rc;
- stats->ojs_cleanup_interval = val;
- lprocfs_job_cleanup(stats, true);
+ stats->ojs_cleanup_interval = ktime_set(val / 2, 0);
+ lprocfs_job_cleanup(stats, false);
+
return count;
}
-EXPORT_SYMBOL(lprocfs_job_interval_seq_write);
-#endif /* CONFIG_PROC_FS*/
+EXPORT_SYMBOL(job_cleanup_interval_store);