Illumos #3236

3236 zio nop-write
Reviewed by: Matt Ahrens <[email protected]>
Reviewed by: Adam Leventhal <[email protected]>
Reviewed by: Christopher Siden <[email protected]>
Approved by: Garrett D'Amore <[email protected]>

References:
  illumos/illumos-gate@80901aea8e78a2c20751f61f01bebd1d5b5c2ba5
  https://www.illumos.org/issues/3236

Porting Notes

1. This patch is being merged dispite an increased instance of
   https://www.illumos.org/issues/3113 being triggered by ztest.

Ported-by: Brian Behlendorf <[email protected]>
Issue #1489

4 files changed, 96 insertions, 19 deletions

diff --git a/include/sys/dbuf.h b/include/sys/dbuf.h
index 85e967d0f..3140665ab 100644
--- a/include/sys/dbuf.h
+++ b/include/sys/dbuf.h
@@ -20,6 +20,7 @@
  */
 /*
  * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 2013 by Delphix. All rights reserved.
  * Copyright (c) 2013 by Saso Kiselkov. All rights reserved.
  */
 
@@ -131,6 +132,7 @@ typedef struct dbuf_dirty_record {
 			blkptr_t dr_overridden_by;
 			override_states_t dr_override_state;
 			uint8_t dr_copies;
+			boolean_t dr_nopwrite;
 		} dl;
 	} dt;
 } dbuf_dirty_record_t;
diff --git a/include/sys/dmu.h b/include/sys/dmu.h
index 0dd7e2835..fdd9923e6 100644
--- a/include/sys/dmu.h
+++ b/include/sys/dmu.h
@@ -498,6 +498,11 @@ void dmu_evict_user(objset_t *os, dmu_buf_evict_func_t *func);
 void *dmu_buf_get_user(dmu_buf_t *db);
 
 /*
+ * Returns the blkptr associated with this dbuf, or NULL if not set.
+ */
+struct blkptr *dmu_buf_get_blkptr(dmu_buf_t *db);
+
+/*
  * Indicate that you are going to modify the buffer's data (db_data).
  *
  * The transaction (tx) must be assigned to a txg (ie. you've called
diff --git a/include/sys/zio.h b/include/sys/zio.h
index f5a128e0b..b505ca1e6 100644
--- a/include/sys/zio.h
+++ b/include/sys/zio.h
@@ -196,7 +196,9 @@ enum zio_flag {
 	ZIO_FLAG_GANG_CHILD	= 1 << 22,
 	ZIO_FLAG_DDT_CHILD	= 1 << 23,
 	ZIO_FLAG_GODFATHER	= 1 << 24,
-	ZIO_FLAG_FASTWRITE      = 1 << 25
+	ZIO_FLAG_NOPWRITE	= 1 << 25,
+	ZIO_FLAG_REEXECUTED	= 1 << 26,
+	ZIO_FLAG_FASTWRITE	= 1 << 27
 };
 
 #define	ZIO_FLAG_MUSTSUCCEED		0
@@ -296,8 +298,9 @@ typedef struct zio_prop {
 	dmu_object_type_t	zp_type;
 	uint8_t			zp_level;
 	uint8_t			zp_copies;
-	uint8_t			zp_dedup;
-	uint8_t			zp_dedup_verify;
+	boolean_t		zp_dedup;
+	boolean_t		zp_dedup_verify;
+	boolean_t		zp_nopwrite;
 } zio_prop_t;
 
 typedef struct zio_cksum_report zio_cksum_report_t;
@@ -466,7 +469,8 @@ extern zio_t *zio_rewrite(zio_t *pio, spa_t *spa, uint64_t txg, blkptr_t *bp,
     void *data, uint64_t size, zio_done_func_t *done, void *private,
     int priority, enum zio_flag flags, zbookmark_t *zb);
 
-extern void zio_write_override(zio_t *zio, blkptr_t *bp, int copies);
+extern void zio_write_override(zio_t *zio, blkptr_t *bp, int copies,
+    boolean_t nopwrite);
 
 extern void zio_free(spa_t *spa, uint64_t txg, const blkptr_t *bp);
 
diff --git a/include/sys/zio_impl.h b/include/sys/zio_impl.h
index 2d062d091..787079d2c 100644
--- a/include/sys/zio_impl.h
+++ b/include/sys/zio_impl.h
@@ -38,6 +38,70 @@ extern "C" {
 #endif
 
 /*
+ * XXX -- Describe ZFS I/O pipleine here. Fill in as needed.
+ *
+ * The ZFS I/O pipeline is comprised of various stages which are defined
+ * in the zio_stage enum below. The individual stages are used to construct
+ * these basic I/O operations: Read, Write, Free, Claim, and Ioctl.
+ *
+ * I/O operations: (XXX - provide detail for each of the operations)
+ *
+ * Read:
+ * Write:
+ * Free:
+ * Claim:
+ * Ioctl:
+ *
+ * Although the most common pipeline are used by the basic I/O operations
+ * above, there are some helper pipelines (one could consider them
+ * sub-pipelines) which are used internally by the ZIO module and are
+ * explained below:
+ *
+ * Interlock Pipeline:
+ * The interlock pipeline is the most basic pipeline and is used by all
+ * of the I/O operations. The interlock pipeline does not perform any I/O
+ * and is used to coordinate the dependencies between I/Os that are being
+ * issued (i.e. the parent/child relationship).
+ *
+ * Vdev child Pipeline:
+ * The vdev child pipeline is responsible for performing the physical I/O.
+ * It is in this pipeline where the I/O are queued and possibly cached.
+ *
+ * In addition to performing I/O, the pipeline is also responsible for
+ * data transformations. The transformations performed are based on the
+ * specific properties that user may have selected and modify the
+ * behavior of the pipeline. Examples of supported transformations are
+ * compression, dedup, and nop writes. Transformations will either modify
+ * the data or the pipeline. This list below further describes each of
+ * the supported transformations:
+ *
+ * Compression:
+ * ZFS supports three different flavors of compression -- gzip, lzjb, and
+ * zle. Compression occurs as part of the write pipeline and is performed
+ * in the ZIO_STAGE_WRITE_BP_INIT stage.
+ *
+ * Dedup:
+ * Dedup reads are handled by the ZIO_STAGE_DDT_READ_START and
+ * ZIO_STAGE_DDT_READ_DONE stages. These stages are added to an existing
+ * read pipeline if the dedup bit is set on the block pointer.
+ * Writing a dedup block is performed by the ZIO_STAGE_DDT_WRITE stage
+ * and added to a write pipeline if a user has enabled dedup on that
+ * particular dataset.
+ *
+ * NOP Write:
+ * The NOP write feature is performed by the ZIO_STAGE_NOP_WRITE stage
+ * and is added to an existing write pipeline if a crypographically
+ * secure checksum (i.e. SHA256) is enabled and compression is turned on.
+ * The NOP write stage will compare the checksums of the current data
+ * on-disk (level-0 blocks only) and the data that is currently being written.
+ * If the checksum values are identical then the pipeline is converted to
+ * an interlock pipeline skipping block allocation and bypassing the
+ * physical I/O.  The nop write feature can handle writes in either
+ * syncing or open context (i.e. zil writes) and as a result is mutually
+ * exclusive with dedup.
+ */
+
+/*
  * zio pipeline stage definitions
  */
 enum zio_stage {
@@ -50,27 +114,29 @@ enum zio_stage {
 
 	ZIO_STAGE_CHECKSUM_GENERATE	= 1 << 5,	/* -W--- */
 
-	ZIO_STAGE_DDT_READ_START	= 1 << 6,	/* R---- */
-	ZIO_STAGE_DDT_READ_DONE		= 1 << 7,	/* R---- */
-	ZIO_STAGE_DDT_WRITE		= 1 << 8,	/* -W--- */
-	ZIO_STAGE_DDT_FREE		= 1 << 9,	/* --F-- */
+	ZIO_STAGE_NOP_WRITE		= 1 << 6,	/* -W--- */
+
+	ZIO_STAGE_DDT_READ_START	= 1 << 7,	/* R---- */
+	ZIO_STAGE_DDT_READ_DONE		= 1 << 8,	/* R---- */
+	ZIO_STAGE_DDT_WRITE		= 1 << 9,	/* -W--- */
+	ZIO_STAGE_DDT_FREE		= 1 << 10,	/* --F-- */
 
-	ZIO_STAGE_GANG_ASSEMBLE		= 1 << 10,	/* RWFC- */
-	ZIO_STAGE_GANG_ISSUE		= 1 << 11,	/* RWFC- */
+	ZIO_STAGE_GANG_ASSEMBLE		= 1 << 11,	/* RWFC- */
+	ZIO_STAGE_GANG_ISSUE		= 1 << 12,	/* RWFC- */
 
-	ZIO_STAGE_DVA_ALLOCATE		= 1 << 12,	/* -W--- */
-	ZIO_STAGE_DVA_FREE		= 1 << 13,	/* --F-- */
-	ZIO_STAGE_DVA_CLAIM		= 1 << 14,	/* ---C- */
+	ZIO_STAGE_DVA_ALLOCATE		= 1 << 13,	/* -W--- */
+	ZIO_STAGE_DVA_FREE		= 1 << 14,	/* --F-- */
+	ZIO_STAGE_DVA_CLAIM		= 1 << 15,	/* ---C- */
 
-	ZIO_STAGE_READY			= 1 << 15,	/* RWFCI */
+	ZIO_STAGE_READY			= 1 << 16,	/* RWFCI */
 
-	ZIO_STAGE_VDEV_IO_START		= 1 << 16,	/* RW--I */
-	ZIO_STAGE_VDEV_IO_DONE		= 1 << 17,	/* RW--I */
-	ZIO_STAGE_VDEV_IO_ASSESS	= 1 << 18,	/* RW--I */
+	ZIO_STAGE_VDEV_IO_START		= 1 << 17,	/* RW--I */
+	ZIO_STAGE_VDEV_IO_DONE		= 1 << 18,	/* RW--I */
+	ZIO_STAGE_VDEV_IO_ASSESS	= 1 << 19,	/* RW--I */
 
-	ZIO_STAGE_CHECKSUM_VERIFY	= 1 << 19,	/* R---- */
+	ZIO_STAGE_CHECKSUM_VERIFY	= 1 << 20,	/* R---- */
 
-	ZIO_STAGE_DONE			= 1 << 20	/* RWFCI */
+	ZIO_STAGE_DONE			= 1 << 21	/* RWFCI */
 };
 
 #define	ZIO_INTERLOCK_STAGES			\