ceph: add IO size metrics support

This will collect IO's total size and then calculate the average
size, and also will collect the min/max IO sizes.

The debugfs will show the size metrics in bytes and will let the
userspace applications to switch to what they need.

URL: https://tracker.ceph.com/issues/49913
Signed-off-by: Xiubo Li <xiubli@redhat.com>
Reviewed-by: Jeff Layton <jlayton@kernel.org>
Signed-off-by: Ilya Dryomov <idryomov@gmail.com>

5 files changed, 119 insertions, 28 deletions

diff --git a/fs/ceph/addr.c b/fs/ceph/addr.c
index 9c17b3a30880..a1e2813731d1 100644
--- a/fs/ceph/addr.c
+++ b/fs/ceph/addr.c
@@ -218,7 +218,7 @@ static void finish_netfs_read(struct ceph_osd_request *req)
 	int err = req->r_result;
 
 	ceph_update_read_metrics(&fsc->mdsc->metric, req->r_start_latency,
-				 req->r_end_latency, err);
+				 req->r_end_latency, osd_data->length, err);
 
 	dout("%s: result %d subreq->len=%zu i_size=%lld\n", __func__, req->r_result,
 	     subreq->len, i_size_read(req->r_inode));
@@ -552,7 +552,7 @@ static int writepage_nounlock(struct page *page, struct writeback_control *wbc)
 		err = ceph_osdc_wait_request(osdc, req);
 
 	ceph_update_write_metrics(&fsc->mdsc->metric, req->r_start_latency,
-				  req->r_end_latency, err);
+				  req->r_end_latency, len, err);
 
 	ceph_osdc_put_request(req);
 	if (err == 0)
@@ -627,6 +627,7 @@ static void writepages_finish(struct ceph_osd_request *req)
 	struct ceph_snap_context *snapc = req->r_snapc;
 	struct address_space *mapping = inode->i_mapping;
 	struct ceph_fs_client *fsc = ceph_inode_to_client(inode);
+	unsigned int len = 0;
 	bool remove_page;
 
 	dout("writepages_finish %p rc %d\n", inode, rc);
@@ -639,9 +640,6 @@ static void writepages_finish(struct ceph_osd_request *req)
 		ceph_clear_error_write(ci);
 	}
 
-	ceph_update_write_metrics(&fsc->mdsc->metric, req->r_start_latency,
-				  req->r_end_latency, rc);
-
 	/*
 	 * We lost the cache cap, need to truncate the page before
 	 * it is unlocked, otherwise we'd truncate it later in the
@@ -658,6 +656,7 @@ static void writepages_finish(struct ceph_osd_request *req)
 
 		osd_data = osd_req_op_extent_osd_data(req, i);
 		BUG_ON(osd_data->type != CEPH_OSD_DATA_TYPE_PAGES);
+		len += osd_data->length;
 		num_pages = calc_pages_for((u64)osd_data->alignment,
 					   (u64)osd_data->length);
 		total_pages += num_pages;
@@ -688,6 +687,9 @@ static void writepages_finish(struct ceph_osd_request *req)
 		release_pages(osd_data->pages, num_pages);
 	}
 
+	ceph_update_write_metrics(&fsc->mdsc->metric, req->r_start_latency,
+				  req->r_end_latency, len, rc);
+
 	ceph_put_wrbuffer_cap_refs(ci, total_pages, snapc);
 
 	osd_data = osd_req_op_extent_osd_data(req, 0);
@@ -1703,7 +1705,7 @@ int ceph_uninline_data(struct file *filp, struct page *locked_page)
 		err = ceph_osdc_wait_request(&fsc->client->osdc, req);
 
 	ceph_update_write_metrics(&fsc->mdsc->metric, req->r_start_latency,
-				  req->r_end_latency, err);
+				  req->r_end_latency, len, err);
 
 out_put:
 	ceph_osdc_put_request(req);
diff --git a/fs/ceph/debugfs.c b/fs/ceph/debugfs.c
index 425f3356332a..38b78b45811f 100644
--- a/fs/ceph/debugfs.c
+++ b/fs/ceph/debugfs.c
@@ -127,7 +127,7 @@ static int mdsc_show(struct seq_file *s, void *p)
 	return 0;
 }
 
-#define CEPH_METRIC_SHOW(name, total, avg, min, max, sq) {		\
+#define CEPH_LAT_METRIC_SHOW(name, total, avg, min, max, sq) {		\
 	s64 _total, _avg, _min, _max, _sq, _st;				\
 	_avg = ktime_to_us(avg);					\
 	_min = ktime_to_us(min == KTIME_MAX ? 0 : min);			\
@@ -140,6 +140,12 @@ static int mdsc_show(struct seq_file *s, void *p)
 		   name, total, _avg, _min, _max, _st);			\
 }
 
+#define CEPH_SZ_METRIC_SHOW(name, total, avg, min, max, sum) {		\
+	u64 _min = min == U64_MAX ? 0 : min;				\
+	seq_printf(s, "%-14s%-12lld%-16llu%-16llu%-16llu%llu\n",	\
+		   name, total, avg, _min, max, sum);			\
+}
+
 static int metric_show(struct seq_file *s, void *p)
 {
 	struct ceph_fs_client *fsc = s->private;
@@ -147,6 +153,7 @@ static int metric_show(struct seq_file *s, void *p)
 	struct ceph_client_metric *m = &mdsc->metric;
 	int nr_caps = 0;
 	s64 total, sum, avg, min, max, sq;
+	u64 sum_sz, avg_sz, min_sz, max_sz;
 
 	sum = percpu_counter_sum(&m->total_inodes);
 	seq_printf(s, "item                               total\n");
@@ -170,7 +177,7 @@ static int metric_show(struct seq_file *s, void *p)
 	max = m->read_latency_max;
 	sq = m->read_latency_sq_sum;
 	spin_unlock(&m->read_metric_lock);
-	CEPH_METRIC_SHOW("read", total, avg, min, max, sq);
+	CEPH_LAT_METRIC_SHOW("read", total, avg, min, max, sq);
 
 	spin_lock(&m->write_metric_lock);
 	total = m->total_writes;
@@ -180,7 +187,7 @@ static int metric_show(struct seq_file *s, void *p)
 	max = m->write_latency_max;
 	sq = m->write_latency_sq_sum;
 	spin_unlock(&m->write_metric_lock);
-	CEPH_METRIC_SHOW("write", total, avg, min, max, sq);
+	CEPH_LAT_METRIC_SHOW("write", total, avg, min, max, sq);
 
 	spin_lock(&m->metadata_metric_lock);
 	total = m->total_metadatas;
@@ -190,7 +197,29 @@ static int metric_show(struct seq_file *s, void *p)
 	max = m->metadata_latency_max;
 	sq = m->metadata_latency_sq_sum;
 	spin_unlock(&m->metadata_metric_lock);
-	CEPH_METRIC_SHOW("metadata", total, avg, min, max, sq);
+	CEPH_LAT_METRIC_SHOW("metadata", total, avg, min, max, sq);
+
+	seq_printf(s, "\n");
+	seq_printf(s, "item          total       avg_sz(bytes)   min_sz(bytes)   max_sz(bytes)  total_sz(bytes)\n");
+	seq_printf(s, "----------------------------------------------------------------------------------------\n");
+
+	spin_lock(&m->read_metric_lock);
+	total = m->total_reads;
+	sum_sz = m->read_size_sum;
+	avg_sz = total > 0 ? DIV64_U64_ROUND_CLOSEST(sum_sz, total) : 0;
+	min_sz = m->read_size_min;
+	max_sz = m->read_size_max;
+	spin_unlock(&m->read_metric_lock);
+	CEPH_SZ_METRIC_SHOW("read", total, avg_sz, min_sz, max_sz, sum_sz);
+
+	spin_lock(&m->write_metric_lock);
+	total = m->total_writes;
+	sum_sz = m->write_size_sum;
+	avg_sz = total > 0 ? DIV64_U64_ROUND_CLOSEST(sum_sz, total) : 0;
+	min_sz = m->write_size_min;
+	max_sz = m->write_size_max;
+	spin_unlock(&m->write_metric_lock);
+	CEPH_SZ_METRIC_SHOW("write", total, avg_sz, min_sz, max_sz, sum_sz);
 
 	seq_printf(s, "\n");
 	seq_printf(s, "item          total           miss            hit\n");
diff --git a/fs/ceph/file.c b/fs/ceph/file.c
index d51af3698032..707102f5cad9 100644
--- a/fs/ceph/file.c
+++ b/fs/ceph/file.c
@@ -903,7 +903,7 @@ static ssize_t ceph_sync_read(struct kiocb *iocb, struct iov_iter *to,
 		ceph_update_read_metrics(&fsc->mdsc->metric,
 					 req->r_start_latency,
 					 req->r_end_latency,
-					 ret);
+					 len, ret);
 
 		ceph_osdc_put_request(req);
 
@@ -1035,12 +1035,12 @@ static void ceph_aio_complete_req(struct ceph_osd_request *req)
 	struct ceph_aio_request *aio_req = req->r_priv;
 	struct ceph_osd_data *osd_data = osd_req_op_extent_osd_data(req, 0);
 	struct ceph_client_metric *metric = &ceph_sb_to_mdsc(inode->i_sb)->metric;
+	unsigned int len = osd_data->bvec_pos.iter.bi_size;
 
 	BUG_ON(osd_data->type != CEPH_OSD_DATA_TYPE_BVECS);
 	BUG_ON(!osd_data->num_bvecs);
 
-	dout("ceph_aio_complete_req %p rc %d bytes %u\n",
-	     inode, rc, osd_data->bvec_pos.iter.bi_size);
+	dout("ceph_aio_complete_req %p rc %d bytes %u\n", inode, rc, len);
 
 	if (rc == -EOLDSNAPC) {
 		struct ceph_aio_work *aio_work;
@@ -1058,9 +1058,9 @@ static void ceph_aio_complete_req(struct ceph_osd_request *req)
 	} else if (!aio_req->write) {
 		if (rc == -ENOENT)
 			rc = 0;
-		if (rc >= 0 && osd_data->bvec_pos.iter.bi_size > rc) {
+		if (rc >= 0 && len > rc) {
 			struct iov_iter i;
-			int zlen = osd_data->bvec_pos.iter.bi_size - rc;
+			int zlen = len - rc;
 
 			/*
 			 * If read is satisfied by single OSD request,
@@ -1077,8 +1077,7 @@ static void ceph_aio_complete_req(struct ceph_osd_request *req)
 			}
 
 			iov_iter_bvec(&i, READ, osd_data->bvec_pos.bvecs,
-				      osd_data->num_bvecs,
-				      osd_data->bvec_pos.iter.bi_size);
+				      osd_data->num_bvecs, len);
 			iov_iter_advance(&i, rc);
 			iov_iter_zero(zlen, &i);
 		}
@@ -1088,10 +1087,10 @@ static void ceph_aio_complete_req(struct ceph_osd_request *req)
 	if (req->r_start_latency) {
 		if (aio_req->write)
 			ceph_update_write_metrics(metric, req->r_start_latency,
-						  req->r_end_latency, rc);
+						  req->r_end_latency, len, rc);
 		else
 			ceph_update_read_metrics(metric, req->r_start_latency,
-						 req->r_end_latency, rc);
+						 req->r_end_latency, len, rc);
 	}
 
 	put_bvecs(osd_data->bvec_pos.bvecs, osd_data->num_bvecs,
@@ -1299,10 +1298,10 @@ ceph_direct_read_write(struct kiocb *iocb, struct iov_iter *iter,
 
 		if (write)
 			ceph_update_write_metrics(metric, req->r_start_latency,
-						  req->r_end_latency, ret);
+						  req->r_end_latency, len, ret);
 		else
 			ceph_update_read_metrics(metric, req->r_start_latency,
-						 req->r_end_latency, ret);
+						 req->r_end_latency, len, ret);
 
 		size = i_size_read(inode);
 		if (!write) {
@@ -1476,7 +1475,7 @@ ceph_sync_write(struct kiocb *iocb, struct iov_iter *from, loff_t pos,
 			ret = ceph_osdc_wait_request(&fsc->client->osdc, req);
 
 		ceph_update_write_metrics(&fsc->mdsc->metric, req->r_start_latency,
-					  req->r_end_latency, ret);
+					  req->r_end_latency, len, ret);
 out:
 		ceph_osdc_put_request(req);
 		if (ret != 0) {
diff --git a/fs/ceph/metric.c b/fs/ceph/metric.c
index afa5b1f8730f..9577c71e645d 100644
--- a/fs/ceph/metric.c
+++ b/fs/ceph/metric.c
@@ -20,6 +20,8 @@ static bool ceph_mdsc_send_metrics(struct ceph_mds_client *mdsc,
 	struct ceph_opened_files *files;
 	struct ceph_pinned_icaps *icaps;
 	struct ceph_opened_inodes *inodes;
+	struct ceph_read_io_size *rsize;
+	struct ceph_write_io_size *wsize;
 	struct ceph_client_metric *m = &mdsc->metric;
 	u64 nr_caps = atomic64_read(&m->total_caps);
 	u32 header_len = sizeof(struct ceph_metric_header);
@@ -31,7 +33,8 @@ static bool ceph_mdsc_send_metrics(struct ceph_mds_client *mdsc,
 
 	len = sizeof(*head) + sizeof(*cap) + sizeof(*read) + sizeof(*write)
 	      + sizeof(*meta) + sizeof(*dlease) + sizeof(*files)
-	      + sizeof(*icaps) + sizeof(*inodes);
+	      + sizeof(*icaps) + sizeof(*inodes) + sizeof(*rsize)
+	      + sizeof(*wsize);
 
 	msg = ceph_msg_new(CEPH_MSG_CLIENT_METRICS, len, GFP_NOFS, true);
 	if (!msg) {
@@ -132,6 +135,26 @@ static bool ceph_mdsc_send_metrics(struct ceph_mds_client *mdsc,
 	inodes->total = cpu_to_le64(sum);
 	items++;
 
+	/* encode the read io size metric */
+	rsize = (struct ceph_read_io_size *)(inodes + 1);
+	rsize->header.type = cpu_to_le32(CLIENT_METRIC_TYPE_READ_IO_SIZES);
+	rsize->header.ver = 1;
+	rsize->header.compat = 1;
+	rsize->header.data_len = cpu_to_le32(sizeof(*rsize) - header_len);
+	rsize->total_ops = cpu_to_le64(m->total_reads);
+	rsize->total_size = cpu_to_le64(m->read_size_sum);
+	items++;
+
+	/* encode the write io size metric */
+	wsize = (struct ceph_write_io_size *)(rsize + 1);
+	wsize->header.type = cpu_to_le32(CLIENT_METRIC_TYPE_WRITE_IO_SIZES);
+	wsize->header.ver = 1;
+	wsize->header.compat = 1;
+	wsize->header.data_len = cpu_to_le32(sizeof(*wsize) - header_len);
+	wsize->total_ops = cpu_to_le64(m->total_writes);
+	wsize->total_size = cpu_to_le64(m->write_size_sum);
+	items++;
+
 	put_unaligned_le32(items, &head->num);
 	msg->front.iov_len = len;
 	msg->hdr.version = cpu_to_le16(1);
@@ -226,6 +249,9 @@ int ceph_metric_init(struct ceph_client_metric *m)
 	m->read_latency_max = 0;
 	m->total_reads = 0;
 	m->read_latency_sum = 0;
+	m->read_size_min = U64_MAX;
+	m->read_size_max = 0;
+	m->read_size_sum = 0;
 
 	spin_lock_init(&m->write_metric_lock);
 	m->write_latency_sq_sum = 0;
@@ -233,6 +259,9 @@ int ceph_metric_init(struct ceph_client_metric *m)
 	m->write_latency_max = 0;
 	m->total_writes = 0;
 	m->write_latency_sum = 0;
+	m->write_size_min = U64_MAX;
+	m->write_size_max = 0;
+	m->write_size_sum = 0;
 
 	spin_lock_init(&m->metadata_metric_lock);
 	m->metadata_latency_sq_sum = 0;
@@ -312,7 +341,7 @@ static inline void __update_stdev(ktime_t total, ktime_t lsum,
 
 void ceph_update_read_metrics(struct ceph_client_metric *m,
 			      ktime_t r_start, ktime_t r_end,
-			      int rc)
+			      unsigned int size, int rc)
 {
 	ktime_t lat = ktime_sub(r_end, r_start);
 	ktime_t total;
@@ -322,7 +351,11 @@ void ceph_update_read_metrics(struct ceph_client_metric *m,
 
 	spin_lock(&m->read_metric_lock);
 	total = ++m->total_reads;
+	m->read_size_sum += size;
 	m->read_latency_sum += lat;
+	METRIC_UPDATE_MIN_MAX(m->read_size_min,
+			      m->read_size_max,
+			      size);
 	METRIC_UPDATE_MIN_MAX(m->read_latency_min,
 			      m->read_latency_max,
 			      lat);
@@ -333,7 +366,7 @@ void ceph_update_read_metrics(struct ceph_client_metric *m,
 
 void ceph_update_write_metrics(struct ceph_client_metric *m,
 			       ktime_t r_start, ktime_t r_end,
-			       int rc)
+			       unsigned int size, int rc)
 {
 	ktime_t lat = ktime_sub(r_end, r_start);
 	ktime_t total;
@@ -343,7 +376,11 @@ void ceph_update_write_metrics(struct ceph_client_metric *m,
 
 	spin_lock(&m->write_metric_lock);
 	total = ++m->total_writes;
+	m->write_size_sum += size;
 	m->write_latency_sum += lat;
+	METRIC_UPDATE_MIN_MAX(m->write_size_min,
+			      m->write_size_max,
+			      size);
 	METRIC_UPDATE_MIN_MAX(m->write_latency_min,
 			      m->write_latency_max,
 			      lat);
diff --git a/fs/ceph/metric.h b/fs/ceph/metric.h
index 1bdc45409daa..0133955a3c6a 100644
--- a/fs/ceph/metric.h
+++ b/fs/ceph/metric.h
@@ -17,8 +17,10 @@ enum ceph_metric_type {
 	CLIENT_METRIC_TYPE_OPENED_FILES,
 	CLIENT_METRIC_TYPE_PINNED_ICAPS,
 	CLIENT_METRIC_TYPE_OPENED_INODES,
+	CLIENT_METRIC_TYPE_READ_IO_SIZES,
+	CLIENT_METRIC_TYPE_WRITE_IO_SIZES,
 
-	CLIENT_METRIC_TYPE_MAX = CLIENT_METRIC_TYPE_OPENED_INODES,
+	CLIENT_METRIC_TYPE_MAX = CLIENT_METRIC_TYPE_WRITE_IO_SIZES,
 };
 
 /*
@@ -34,6 +36,8 @@ enum ceph_metric_type {
 	CLIENT_METRIC_TYPE_OPENED_FILES,	\
 	CLIENT_METRIC_TYPE_PINNED_ICAPS,	\
 	CLIENT_METRIC_TYPE_OPENED_INODES,	\
+	CLIENT_METRIC_TYPE_READ_IO_SIZES,	\
+	CLIENT_METRIC_TYPE_WRITE_IO_SIZES,	\
 						\
 	CLIENT_METRIC_TYPE_MAX,			\
 }
@@ -103,6 +107,20 @@ struct ceph_opened_inodes {
 	__le64 total;
 } __packed;
 
+/* metric read io size header */
+struct ceph_read_io_size {
+	struct ceph_metric_header header;
+	__le64 total_ops;
+	__le64 total_size;
+} __packed;
+
+/* metric write io size header */
+struct ceph_write_io_size {
+	struct ceph_metric_header header;
+	__le64 total_ops;
+	__le64 total_size;
+} __packed;
+
 struct ceph_metric_head {
 	__le32 num;	/* the number of metrics that will be sent */
 } __packed;
@@ -119,6 +137,9 @@ struct ceph_client_metric {
 
 	spinlock_t read_metric_lock;
 	u64 total_reads;
+	u64 read_size_sum;
+	u64 read_size_min;
+	u64 read_size_max;
 	ktime_t read_latency_sum;
 	ktime_t read_latency_sq_sum;
 	ktime_t read_latency_min;
@@ -126,6 +147,9 @@ struct ceph_client_metric {
 
 	spinlock_t write_metric_lock;
 	u64 total_writes;
+	u64 write_size_sum;
+	u64 write_size_min;
+	u64 write_size_max;
 	ktime_t write_latency_sum;
 	ktime_t write_latency_sq_sum;
 	ktime_t write_latency_min;
@@ -173,10 +197,10 @@ static inline void ceph_update_cap_mis(struct ceph_client_metric *m)
 
 extern void ceph_update_read_metrics(struct ceph_client_metric *m,
 				     ktime_t r_start, ktime_t r_end,
-				     int rc);
+				     unsigned int size, int rc);
 extern void ceph_update_write_metrics(struct ceph_client_metric *m,
 				      ktime_t r_start, ktime_t r_end,
-				      int rc);
+				      unsigned int size, int rc);
 extern void ceph_update_metadata_metrics(struct ceph_client_metric *m,
 				         ktime_t r_start, ktime_t r_end,
 					 int rc);