Inject zinject(8) a percentage amount of dev errs

In the original form of device error injection, it was an all or nothing
situation.  To help simulate intermittent error conditions, you can now
specify a real number percentage value. This is also very useful for our
ZFS fault diagnosis testing and for injecting intermittent errors during
load testing.

Reviewed-by: Brian Behlendorf <[email protected]>
Signed-off-by: Don Brady <[email protected]>
Closes #6227 

4 files changed, 76 insertions, 20 deletions

diff --git a/cmd/zinject/zinject.c b/cmd/zinject/zinject.c
index 604554e71..ccd3534d0 100644
--- a/cmd/zinject/zinject.c
+++ b/cmd/zinject/zinject.c
@@ -21,6 +21,7 @@
 /*
  * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
  * Copyright (c) 2012, 2015 by Delphix. All rights reserved.
+ * Copyright (c) 2017, Intel Corporation.
  */
 
 /*
@@ -124,7 +125,7 @@
  * cache.
  *
  * The '-f' flag controls the frequency of errors injected, expressed as a
- * integer percentage between 1 and 100.  The default is 100.
+ * real number percentage between 0.0001 and 100.  The default is 100.
  *
  * The this form is responsible for actually injecting the handler into the
  * framework.  It takes the arguments described above, translates them to the
@@ -230,11 +231,13 @@ usage(void)
 	    "\t\tspa_vdev_exit() will trigger a panic.\n"
 	    "\n"
 	    "\tzinject -d device [-e errno] [-L <nvlist|uber|pad1|pad2>] [-F]\n"
-	    "\t    [-T <read|write|free|claim|all> pool\n"
+	    "\t    [-T <read|write|free|claim|all>] [-f frequency] pool\n"
 	    "\t\tInject a fault into a particular device or the device's\n"
 	    "\t\tlabel.  Label injection can either be 'nvlist', 'uber',\n "
 	    "\t\t'pad1', or 'pad2'.\n"
 	    "\t\t'errno' can be 'nxio' (the default), 'io', or 'dtl'.\n"
+	    "\t\t'frequency' is a value between 0.0001 and 100.0 that limits\n"
+	    "\t\tdevice error injection to a percentage of the IOs.\n"
 	    "\n"
 	    "\tzinject -d device -A <degrade|fault> -D <delay secs> pool\n"
 	    "\t\tPerform a specific action on a particular device.\n"
@@ -305,7 +308,7 @@ usage(void)
 	    "\t\t-u\tUnload the associated pool.  Can be specified with only\n"
 	    "\t\t\ta pool object.\n"
 	    "\t\t-f\tOnly inject errors a fraction of the time.  Expressed as\n"
-	    "\t\t\ta percentage between 1 and 100.\n"
+	    "\t\t\ta percentage between 0.0001 and 100.\n"
 	    "\n"
 	    "\t-t data\t\tInject an error into the plain file contents of a\n"
 	    "\t\t\tfile.  The object must be specified as a complete path\n"
@@ -645,6 +648,27 @@ parse_delay(char *str, uint64_t *delay, uint64_t *nlanes)
 	return (0);
 }
 
+static int
+parse_frequency(const char *str, uint32_t *percent)
+{
+	double val;
+	char *post;
+
+	val = strtod(str, &post);
+	if (post == NULL || *post != '\0')
+		return (EINVAL);
+
+	/* valid range is [0.0001, 100.0] */
+	val /= 100.0f;
+	if (val < 0.000001f || val > 1.0f)
+		return (ERANGE);
+
+	/* convert to an integer for use by kernel */
+	*percent = ((uint32_t)(val * ZI_PERCENTAGE_MAX));
+
+	return (0);
+}
+
 int
 main(int argc, char **argv)
 {
@@ -760,10 +784,12 @@ main(int argc, char **argv)
 			}
 			break;
 		case 'f':
-			record.zi_freq = atoi(optarg);
-			if (record.zi_freq < 1 || record.zi_freq > 100) {
-				(void) fprintf(stderr, "frequency range must "
-				    "be in the range (0, 100]\n");
+			ret = parse_frequency(optarg, &record.zi_freq);
+			if (ret != 0) {
+				(void) fprintf(stderr, "%sfrequency value must "
+				    "be in the range [0.0001, 100.0]\n",
+				    ret == EINVAL ? "invalid value: " :
+				    ret == ERANGE ? "out of range: " : "");
 				libzfs_fini(g_zfs);
 				return (1);
 			}
@@ -898,7 +924,8 @@ main(int argc, char **argv)
 		 * '-c' is invalid with any other options.
 		 */
 		if (raw != NULL || range != NULL || type != TYPE_INVAL ||
-		    level != 0 || record.zi_cmd != ZINJECT_UNINITIALIZED) {
+		    level != 0 || record.zi_cmd != ZINJECT_UNINITIALIZED ||
+		    record.zi_freq > 0) {
 			(void) fprintf(stderr, "cancel (-c) incompatible with "
 			    "any other options\n");
 			usage();
@@ -972,7 +999,8 @@ main(int argc, char **argv)
 
 	} else if (raw != NULL) {
 		if (range != NULL || type != TYPE_INVAL || level != 0 ||
-		    record.zi_cmd != ZINJECT_UNINITIALIZED) {
+		    record.zi_cmd != ZINJECT_UNINITIALIZED ||
+		    record.zi_freq > 0) {
 			(void) fprintf(stderr, "raw (-b) format with "
 			    "any other options\n");
 			usage();
@@ -1007,7 +1035,7 @@ main(int argc, char **argv)
 			error = EIO;
 	} else if (record.zi_cmd == ZINJECT_PANIC) {
 		if (raw != NULL || range != NULL || type != TYPE_INVAL ||
-		    level != 0 || device != NULL) {
+		    level != 0 || device != NULL || record.zi_freq > 0) {
 			(void) fprintf(stderr, "panic (-p) incompatible with "
 			    "other options\n");
 			usage();
diff --git a/include/sys/zfs_ioctl.h b/include/sys/zfs_ioctl.h
index 495cdea3a..c68b8770b 100644
--- a/include/sys/zfs_ioctl.h
+++ b/include/sys/zfs_ioctl.h
@@ -22,6 +22,7 @@
  * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
  * Copyright (c) 2012, 2015 by Delphix. All rights reserved.
  * Copyright 2016 RackTop Systems.
+ * Copyright (c) 2017, Intel Corporation.
  */
 
 #ifndef	_SYS_ZFS_IOCTL_H
@@ -338,6 +339,10 @@ typedef struct zinject_record {
 #define	ZEVENT_SEEK_START	0
 #define	ZEVENT_SEEK_END		UINT64_MAX
 
+/* scaled frequency ranges */
+#define	ZI_PERCENTAGE_MIN	4294UL
+#define	ZI_PERCENTAGE_MAX	UINT32_MAX
+
 typedef enum zinject_type {
 	ZINJECT_UNINITIALIZED,
 	ZINJECT_DATA_FAULT,
diff --git a/man/man8/zinject.8 b/man/man8/zinject.8
index ab22d4a5b..50fecfb64 100644
--- a/man/man8/zinject.8
+++ b/man/man8/zinject.8
@@ -76,7 +76,7 @@ create 3 lanes on the device; one lane with a latency
 of 10 ms and two lanes with a 25 ms latency.
 
 .TP
-.B "zinject \-d \fIvdev\fB [\-e \fIdevice_error\fB] [\-L \fIlabel_error\fB] [\-T \fIfailure\fB] [\-F] \fIpool\fB"
+.B "zinject \-d \fIvdev\fB [\-e \fIdevice_error\fB] [\-L \fIlabel_error\fB] [\-T \fIfailure\fB] [\-f \fIfrequency\fB] [\-F] \fIpool\fB"
 Force a vdev error.
 .TP
 .B "zinject \-I [\-s \fIseconds\fB | \-g \fItxgs\fB] \fIpool\fB"
@@ -113,8 +113,8 @@ Specify
 .BR "nxio" " for an ENXIO error where reopening the device will fail."
 .TP
 .BI "\-f" " frequency"
-Only inject errors a fraction of the time. Expressed as an integer
-percentage between 1 and 100.
+Only inject errors a fraction of the time. Expressed as a real number
+percentage between 0.0001 and 100.
 .TP
 .BI "\-F"
 Fail faster. Do fewer checks.
diff --git a/module/zfs/zio_inject.c b/module/zfs/zio_inject.c
index 0e8e9d932..4a4d431e3 100644
--- a/module/zfs/zio_inject.c
+++ b/module/zfs/zio_inject.c
@@ -21,6 +21,7 @@
 /*
  * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
  * Copyright (c) 2012, 2015 by Delphix. All rights reserved.
+ * Copyright (c) 2017, Intel Corporation.
  */
 
 /*
@@ -99,6 +100,26 @@ static kmutex_t inject_delay_mtx;
 static int inject_next_id = 1;
 
 /*
+ * Test if the requested frequency was triggered
+ */
+static boolean_t
+freq_triggered(uint32_t frequency)
+{
+	/*
+	 * zero implies always (100%)
+	 */
+	if (frequency == 0)
+		return (B_TRUE);
+
+	/*
+	 * Note: we still handle legacy (unscaled) frequecy values
+	 */
+	uint32_t maximum = (frequency <= 100) ? 100 : ZI_PERCENTAGE_MAX;
+
+	return (spa_get_random(maximum) < frequency);
+}
+
+/*
  * Returns true if the given record matches the I/O in progress.
  */
 static boolean_t
@@ -113,8 +134,7 @@ zio_match_handler(zbookmark_phys_t *zb, uint64_t type,
 	    record->zi_object == DMU_META_DNODE_OBJECT) {
 		if (record->zi_type == DMU_OT_NONE ||
 		    type == record->zi_type)
-			return (record->zi_freq == 0 ||
-			    spa_get_random(100) < record->zi_freq);
+			return (freq_triggered(record->zi_freq));
 		else
 			return (B_FALSE);
 	}
@@ -128,8 +148,7 @@ zio_match_handler(zbookmark_phys_t *zb, uint64_t type,
 	    zb->zb_blkid >= record->zi_start &&
 	    zb->zb_blkid <= record->zi_end &&
 	    error == record->zi_error)
-		return (record->zi_freq == 0 ||
-		    spa_get_random(100) < record->zi_freq);
+		return (freq_triggered(record->zi_freq));
 
 	return (B_FALSE);
 }
@@ -294,6 +313,12 @@ zio_handle_device_injection(vdev_t *vd, zio_t *zio, int error)
 
 			if (handler->zi_record.zi_error == error) {
 				/*
+				 * limit error injection if requested
+				 */
+				if (!freq_triggered(handler->zi_record.zi_freq))
+					continue;
+
+				/*
 				 * For a failed open, pretend like the device
 				 * has gone away.
 				 */
@@ -466,10 +491,8 @@ zio_handle_io_delay(zio_t *zio)
 		if (handler->zi_record.zi_cmd != ZINJECT_DELAY_IO)
 			continue;
 
-		if (handler->zi_record.zi_freq != 0 &&
-		    spa_get_random(100) >= handler->zi_record.zi_freq) {
+		if (!freq_triggered(handler->zi_record.zi_freq))
 			continue;
-		}
 
 		if (vd->vdev_guid != handler->zi_record.zi_guid)
 			continue;