zinject: "no-op" error injection

When injected, this causes the matching IO to appear to succeed, but the
actual work is never submitted to the physical device. This can be used
to simulate a write-back cache servicing a write, but the backing device
has failed and the cache cannot complete the operation in the
background.

Sponsored-by: Klara, Inc.
Sponsored-by: Wasabi Technology, Inc.
Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Signed-off-by: Rob Norris <rob.norris@klarasystems.com>
Closes #16085 

4 files changed, 19 insertions, 6 deletions

diff --git a/cmd/zinject/zinject.c b/cmd/zinject/zinject.c
index a1afa4a63feb..e9141fb4ba55 100644
--- a/cmd/zinject/zinject.c
+++ b/cmd/zinject/zinject.c
@@ -221,6 +221,7 @@ static const struct errstr errstrtable[] = {
 	{ ENXIO,	"nxio" },
 	{ ECHILD,	"dtl" },
 	{ EILSEQ,	"corrupt" },
+	{ ENOSYS,	"noop" },
 	{ 0, NULL },
 };
 
@@ -269,8 +270,8 @@ usage(void)
 	    "\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', 'dtl', or\n"
-	    "\t\t'corrupt' (bit flip).\n"
+	    "\t\t'errno' can be 'nxio' (the default), 'io', 'dtl',\n"
+	    "\t\t'corrupt' (bit flip), or 'noop' (successfully do nothing).\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"
@@ -889,7 +890,7 @@ main(int argc, char **argv)
 			if (error < 0) {
 				(void) fprintf(stderr, "invalid error type "
 				    "'%s': must be one of: io decompress "
-				    "decrypt nxio dtl corrupt\n",
+				    "decrypt nxio dtl corrupt noop\n",
 				    optarg);
 				usage();
 				libzfs_fini(g_zfs);
diff --git a/man/man8/zinject.8 b/man/man8/zinject.8
index 817dcb7fe32a..f67b5e378dc3 100644
--- a/man/man8/zinject.8
+++ b/man/man8/zinject.8
@@ -211,9 +211,11 @@ to flip a bit in the data after a read,
 .It Sy dtl
 for an ECHILD error,
 .It Sy io
-for an EIO error where reopening the device will succeed, or
+for an EIO error where reopening the device will succeed,
 .It Sy nxio
-for an ENXIO error where reopening the device will fail.
+for an ENXIO error where reopening the device will fail, or
+.It Sy noop
+to drop the IO without executing it, and return success.
 .El
 .Pp
 For EIO and ENXIO, the "failed" reads or writes still occur.
diff --git a/module/zfs/zio.c b/module/zfs/zio.c
index 8d8523038e60..414e3d4e93bd 100644
--- a/module/zfs/zio.c
+++ b/module/zfs/zio.c
@@ -4058,6 +4058,16 @@ zio_vdev_io_start(zio_t *zio)
 	    zio->io_type == ZIO_TYPE_WRITE ||
 	    zio->io_type == ZIO_TYPE_TRIM)) {
 
+		if (zio_handle_device_injection(vd, zio, ENOSYS) != 0) {
+			/*
+			 * "no-op" injections return success, but do no actual
+			 * work. Just skip the remaining vdev stages.
+			 */
+			zio_vdev_io_bypass(zio);
+			zio_interrupt(zio);
+			return (NULL);
+		}
+
 		if ((zio = vdev_queue_io(zio)) == NULL)
 			return (NULL);
 
diff --git a/tests/zfs-tests/tests/functional/cli_root/zinject/zinject_args.ksh b/tests/zfs-tests/tests/functional/cli_root/zinject/zinject_args.ksh
index f8a8ffbb7b0e..dd9ef9ddd229 100755
--- a/tests/zfs-tests/tests/functional/cli_root/zinject/zinject_args.ksh
+++ b/tests/zfs-tests/tests/functional/cli_root/zinject/zinject_args.ksh
@@ -47,7 +47,7 @@ function cleanup
 
 function test_device_fault
 {
-	typeset -a errno=("io" "decompress" "decrypt" "nxio" "dtl" "corrupt")
+	typeset -a errno=("io" "decompress" "decrypt" "nxio" "dtl" "corrupt" "noop")
 	for e in ${errno[@]}; do
 		log_must eval \
 		    "zinject -d $DISK1 -e $e -T read -f 0.001 $TESTPOOL"