OpenZFS 6531 - Provide mechanism to artificially limit disk performance

Reviewed by: Paul Dagnelie <[email protected]>
Reviewed by: Matthew Ahrens <[email protected]>
Reviewed by: George Wilson <[email protected]>
Approved by: Dan McDonald <[email protected]>
Ported by: Tony Hutter <[email protected]>
Signed-off-by: Brian Behlendorf <[email protected]>

OpenZFS-issue: https://www.illumos.org/issues/6531
OpenZFS-commit: https://github.com/openzfs/openzfs/commit/97e8130

Porting notes:
- Added new IO delay tracepoints, and moved common ZIO tracepoint macros
  to a new trace_common.h file.
- Used zio_delay_taskq() in place of OpenZFS's timeout_generic() function.
- Updated zinject man page
- Updated zpool_scrub test files

1 files changed, 71 insertions, 0 deletions

diff --git a/module/zfs/zio.c b/module/zfs/zio.c
index 523a924d6..4063703ad 100644
--- a/module/zfs/zio.c
+++ b/module/zfs/zio.c
@@ -40,6 +40,7 @@
 #include <sys/blkptr.h>
 #include <sys/zfeature.h>
 #include <sys/time.h>
+#include <sys/trace_zio.h>
 
 /*
  * ==========================================================================
@@ -1390,6 +1391,76 @@ zio_interrupt(zio_t *zio)
 	zio_taskq_dispatch(zio, ZIO_TASKQ_INTERRUPT, B_FALSE);
 }
 
+void
+zio_delay_interrupt(zio_t *zio)
+{
+	/*
+	 * The timeout_generic() function isn't defined in userspace, so
+	 * rather than trying to implement the function, the zio delay
+	 * functionality has been disabled for userspace builds.
+	 */
+
+#ifdef _KERNEL
+	/*
+	 * If io_target_timestamp is zero, then no delay has been registered
+	 * for this IO, thus jump to the end of this function and "skip" the
+	 * delay; issuing it directly to the zio layer.
+	 */
+	if (zio->io_target_timestamp != 0) {
+		hrtime_t now = gethrtime();
+
+		if (now >= zio->io_target_timestamp) {
+			/*
+			 * This IO has already taken longer than the target
+			 * delay to complete, so we don't want to delay it
+			 * any longer; we "miss" the delay and issue it
+			 * directly to the zio layer. This is likely due to
+			 * the target latency being set to a value less than
+			 * the underlying hardware can satisfy (e.g. delay
+			 * set to 1ms, but the disks take 10ms to complete an
+			 * IO request).
+			 */
+
+			DTRACE_PROBE2(zio__delay__miss, zio_t *, zio,
+			    hrtime_t, now);
+
+			zio_interrupt(zio);
+		} else {
+			taskqid_t tid;
+			hrtime_t diff = zio->io_target_timestamp - now;
+			clock_t expire_at_tick = ddi_get_lbolt() +
+			    NSEC_TO_TICK(diff);
+
+			DTRACE_PROBE3(zio__delay__hit, zio_t *, zio,
+			    hrtime_t, now, hrtime_t, diff);
+
+			if (NSEC_TO_TICK(diff) == 0) {
+				/* Our delay is less than a jiffy - just spin */
+				zfs_sleep_until(zio->io_target_timestamp);
+			} else {
+				/*
+				 * Use taskq_dispatch_delay() in the place of
+				 * OpenZFS's timeout_generic().
+				 */
+				tid = taskq_dispatch_delay(system_taskq,
+				    (task_func_t *) zio_interrupt,
+				    zio, TQ_NOSLEEP, expire_at_tick);
+				if (!tid) {
+					/*
+					 * Couldn't allocate a task.  Just
+					 * finish the zio without a delay.
+					 */
+					zio_interrupt(zio);
+				}
+			}
+		}
+		return;
+	}
+#endif
+	DTRACE_PROBE1(zio__delay__skip, zio_t *, zio);
+	zio_interrupt(zio);
+}
+
 /*
  * Execute the I/O pipeline until one of the following occurs:
  * (1) the I/O completes; (2) the pipeline stalls waiting for