Add zstd support to zfs

This PR adds two new compression types, based on ZStandard:

- zstd: A basic ZStandard compression algorithm Available compression.
  Levels for zstd are zstd-1 through zstd-19, where the compression
  increases with every level, but speed decreases.

- zstd-fast: A faster version of the ZStandard compression algorithm
  zstd-fast is basically a "negative" level of zstd. The compression
  decreases with every level, but speed increases.

  Available compression levels for zstd-fast:
   - zstd-fast-1 through zstd-fast-10
   - zstd-fast-20 through zstd-fast-100 (in increments of 10)
   - zstd-fast-500 and zstd-fast-1000

For more information check the man page.

Implementation details:

Rather than treat each level of zstd as a different algorithm (as was
done historically with gzip), the block pointer `enum zio_compress`
value is simply zstd for all levels, including zstd-fast, since they all
use the same decompression function.

The compress= property (a 64bit unsigned integer) uses the lower 7 bits
to store the compression algorithm (matching the number of bits used in
a block pointer, as the 8th bit was borrowed for embedded block
pointers).  The upper bits are used to store the compression level.

It is necessary to be able to determine what compression level was used
when later reading a block back, so the concept used in LZ4, where the
first 32bits of the on-disk value are the size of the compressed data
(since the allocation is rounded up to the nearest ashift), was
extended, and we store the version of ZSTD and the level as well as the
compressed size. This value is returned when decompressing a block, so
that if the block needs to be recompressed (L2ARC, nop-write, etc), that
the same parameters will be used to result in the matching checksum.

All of the internal ZFS code ( `arc_buf_hdr_t`, `objset_t`,
`zio_prop_t`, etc.) uses the separated _compress and _complevel
variables.  Only the properties ZAP contains the combined/bit-shifted
value. The combined value is split when the compression_changed_cb()
callback is called, and sets both objset members (os_compress and
os_complevel).

The userspace tools all use the combined/bit-shifted value.

Additional notes:

zdb can now also decode the ZSTD compression header (flag -Z) and
inspect the size, version and compression level saved in that header.
For each record, if it is ZSTD compressed, the parameters of the decoded
compression header get printed.

ZSTD is included with all current tests and new tests are added
as-needed.

Per-dataset feature flags now get activated when the property is set.
If a compression algorithm requires a feature flag, zfs activates the
feature when the property is set, rather than waiting for the first
block to be born.  This is currently only used by zstd but can be
extended as needed.

Portions-Sponsored-By: The FreeBSD Foundation
Co-authored-by: Allan Jude <[email protected]>
Co-authored-by: Brian Behlendorf <[email protected]>
Co-authored-by: Sebastian Gottschall <[email protected]>
Co-authored-by: Kjeld Schouten-Lebbing <[email protected]>
Co-authored-by: Michael Niewöhner <[email protected]>
Signed-off-by: Allan Jude <[email protected]>
Signed-off-by: Allan Jude <[email protected]>
Signed-off-by: Brian Behlendorf <[email protected]>
Signed-off-by: Sebastian Gottschall <[email protected]>
Signed-off-by: Kjeld Schouten-Lebbing <[email protected]>
Signed-off-by: Michael Niewöhner <[email protected]>
Closes #6247
Closes #9024
Closes #10277
Closes #10278

Author: 5 people

AUTHORS

@@ -182,6 +182,7 @@ CONTRIBUTORS:
     Keith M Wesolowski <[email protected]>
     Kevin Tanguy <[email protected]>
     KireinaHoro <[email protected]>
+    Kjeld Schouten-Lebbing <[email protected]>
     Kohsuke Kawaguchi <[email protected]>
     Kyle Blatter <[email protected]>
     Kyle Fuller <[email protected]>
@@ -210,6 +211,7 @@ CONTRIBUTORS:
     Michael Gebetsroither <[email protected]>
     Michael Kjorling <[email protected]>
     Michael Martin <[email protected]>
+    Michael Niewöhner <[email protected]>
     Mike Gerdts <[email protected]>
     Mike Harsch <[email protected]>
     Mike Leddy <[email protected]>
@@ -258,6 +260,7 @@ CONTRIBUTORS:
     Saso Kiselkov <[email protected]>
     Scot W. Stevenson <[email protected]>
     Sean Eric Fagan <[email protected]>
+    Sebastian Gottschall <[email protected]>
     Sen Haerens <[email protected]>
     Serapheim Dimitropoulos <[email protected]>
     Seth Forshee <[email protected]>

cmd/dbufstat/dbufstat.in

@@ -343,7 +343,7 @@ def get_compstring(c):
                     "ZIO_COMPRESS_GZIP_6",  "ZIO_COMPRESS_GZIP_7",
                     "ZIO_COMPRESS_GZIP_8",  "ZIO_COMPRESS_GZIP_9",
                     "ZIO_COMPRESS_ZLE",     "ZIO_COMPRESS_LZ4",
-                    "ZIO_COMPRESS_FUNCTION"]
+                    "ZIO_COMPRESS_ZSTD",    "ZIO_COMPRESS_FUNCTION"]
 
     # If "-rr" option is used, don't convert to string representation
     if raw > 1:

cmd/zdb/zdb.c

@@ -27,6 +27,10 @@
  * Copyright (c) 2017, 2018 Lawrence Livermore National Security, LLC.
  * Copyright (c) 2015, 2017, Intel Corporation.
  * Copyright (c) 2020 Datto Inc.
+ * Copyright (c) 2020, The FreeBSD Foundation [1]
+ *
+ * [1] Portions of this software were developed by Allan Jude
+ *     under sponsorship from the FreeBSD Foundation.
  */
 
 #include <stdio.h>
@@ -71,6 +75,7 @@
 #include <sys/dsl_scan.h>
 #include <sys/btree.h>
 #include <zfs_comutil.h>
+#include <sys/zstd/zstd.h>
 
 #include <libnvpair.h>
 #include <libzutil.h>
@@ -834,6 +839,7 @@ usage(void)
 	    "work with dataset)\n");
 	(void) fprintf(stderr, "        -Y attempt all reconstruction "
 	    "combinations for split blocks\n");
+	(void) fprintf(stderr, "        -Z show ZSTD headers \n");
 	(void) fprintf(stderr, "Specify an option more than once (e.g. -bb) "
 	    "to make only that option verbose\n");
 	(void) fprintf(stderr, "Default is to dump everything non-verbosely\n");
@@ -2134,6 +2140,65 @@ blkid2offset(const dnode_phys_t *dnp, const blkptr_t *bp,
 	    dnp->dn_datablkszsec << SPA_MINBLOCKSHIFT);
 }
 
+static void
+snprintf_zstd_header(spa_t *spa, char *blkbuf, size_t buflen,
+    const blkptr_t *bp)
+{
+	abd_t *pabd;
+	void *buf;
+	zio_t *zio;
+	zfs_zstdhdr_t zstd_hdr;
+	int error;
+
+	if (BP_GET_COMPRESS(bp) != ZIO_COMPRESS_ZSTD)
+		return;
+
+	if (BP_IS_HOLE(bp))
+		return;
+
+	if (BP_IS_EMBEDDED(bp)) {
+		buf = malloc(SPA_MAXBLOCKSIZE);
+		if (buf == NULL) {
+			(void) fprintf(stderr, "out of memory\n");
+			exit(1);
+		}
+		decode_embedded_bp_compressed(bp, buf);
+		memcpy(&zstd_hdr, buf, sizeof (zstd_hdr));
+		free(buf);
+		zstd_hdr.c_len = BE_32(zstd_hdr.c_len);
+		zstd_hdr.raw_version_level = BE_32(zstd_hdr.raw_version_level);
+		(void) snprintf(blkbuf + strlen(blkbuf),
+		    buflen - strlen(blkbuf),
+		    " ZSTD:size=%u:version=%u:level=%u:EMBEDDED",
+		    zstd_hdr.c_len, zstd_hdr.version, zstd_hdr.level);
+		return;
+	}
+
+	pabd = abd_alloc_for_io(SPA_MAXBLOCKSIZE, B_FALSE);
+	zio = zio_root(spa, NULL, NULL, 0);
+
+	/* Decrypt but don't decompress so we can read the compression header */
+	zio_nowait(zio_read(zio, spa, bp, pabd, BP_GET_PSIZE(bp), NULL, NULL,
+	    ZIO_PRIORITY_SYNC_READ, ZIO_FLAG_CANFAIL | ZIO_FLAG_RAW_COMPRESS,
+	    NULL));
+	error = zio_wait(zio);
+	if (error) {
+		(void) fprintf(stderr, "read failed: %d\n", error);
+		return;
+	}
+	buf = abd_borrow_buf_copy(pabd, BP_GET_LSIZE(bp));
+	memcpy(&zstd_hdr, buf, sizeof (zstd_hdr));
+	zstd_hdr.c_len = BE_32(zstd_hdr.c_len);
+	zstd_hdr.raw_version_level = BE_32(zstd_hdr.raw_version_level);
+
+	(void) snprintf(blkbuf + strlen(blkbuf),
+	    buflen - strlen(blkbuf),
+	    " ZSTD:size=%u:version=%u:level=%u:NORMAL",
+	    zstd_hdr.c_len, zstd_hdr.version, zstd_hdr.level);
+
+	abd_return_buf_copy(pabd, buf, BP_GET_LSIZE(bp));
+}
+
 static void
 snprintf_blkptr_compact(char *blkbuf, size_t buflen, const blkptr_t *bp,
     boolean_t bp_freed)
@@ -2198,7 +2263,7 @@ snprintf_blkptr_compact(char *blkbuf, size_t buflen, const blkptr_t *bp,
 }
 
 static void
-print_indirect(blkptr_t *bp, const zbookmark_phys_t *zb,
+print_indirect(spa_t *spa, blkptr_t *bp, const zbookmark_phys_t *zb,
     const dnode_phys_t *dnp)
 {
 	char blkbuf[BP_SPRINTF_LEN];
@@ -2222,6 +2287,8 @@ print_indirect(blkptr_t *bp, const zbookmark_phys_t *zb,
 	}
 
 	snprintf_blkptr_compact(blkbuf, sizeof (blkbuf), bp, B_FALSE);
+	if (dump_opt['Z'] && BP_GET_COMPRESS(bp) == ZIO_COMPRESS_ZSTD)
+		snprintf_zstd_header(spa, blkbuf, sizeof (blkbuf), bp);
 	(void) printf("%s\n", blkbuf);
 }
 
@@ -2234,7 +2301,7 @@ visit_indirect(spa_t *spa, const dnode_phys_t *dnp,
 	if (bp->blk_birth == 0)
 		return (0);
 
-	print_indirect(bp, zb, dnp);
+	print_indirect(spa, bp, zb, dnp);
 
 	if (BP_GET_LEVEL(bp) > 0 && !BP_IS_HOLE(bp)) {
 		arc_flags_t flags = ARC_FLAG_WAIT;
@@ -3310,7 +3377,25 @@ dump_object(objset_t *os, uint64_t object, int verbosity,
 		    " (K=%s)", ZDB_CHECKSUM_NAME(doi.doi_checksum));
 	}
 
-	if (doi.doi_compress != ZIO_COMPRESS_INHERIT || verbosity >= 6) {
+	if (doi.doi_compress == ZIO_COMPRESS_INHERIT &&
+	    ZIO_COMPRESS_HASLEVEL(os->os_compress) && verbosity >= 6) {
+		const char *compname = NULL;
+		if (zfs_prop_index_to_string(ZFS_PROP_COMPRESSION,
+		    ZIO_COMPRESS_RAW(os->os_compress, os->os_complevel),
+		    &compname) == 0) {
+			(void) snprintf(aux + strlen(aux),
+			    sizeof (aux) - strlen(aux), " (Z=inherit=%s)",
+			    compname);
+		} else {
+			(void) snprintf(aux + strlen(aux),
+			    sizeof (aux) - strlen(aux),
+			    " (Z=inherit=%s-unknown)",
+			    ZDB_COMPRESS_NAME(os->os_compress));
+		}
+	} else if (doi.doi_compress == ZIO_COMPRESS_INHERIT && verbosity >= 6) {
+		(void) snprintf(aux + strlen(aux), sizeof (aux) - strlen(aux),
+		    " (Z=inherit=%s)", ZDB_COMPRESS_NAME(os->os_compress));
+	} else if (doi.doi_compress != ZIO_COMPRESS_INHERIT || verbosity >= 6) {
 		(void) snprintf(aux + strlen(aux), sizeof (aux) - strlen(aux),
 		    " (Z=%s)", ZDB_COMPRESS_NAME(doi.doi_compress));
 	}
@@ -4093,6 +4178,8 @@ dump_l2arc_log_entries(uint64_t log_entries,
 		    (u_longlong_t)L2BLK_GET_PSIZE((&le[j])->le_prop));
 		(void) printf("|\t\t\t\tcompr: %llu\n",
 		    (u_longlong_t)L2BLK_GET_COMPRESS((&le[j])->le_prop));
+		(void) printf("|\t\t\t\tcomplevel: %llu\n",
+		    (u_longlong_t)(&le[j])->le_complevel);
 		(void) printf("|\t\t\t\ttype: %llu\n",
 		    (u_longlong_t)L2BLK_GET_TYPE((&le[j])->le_prop));
 		(void) printf("|\t\t\t\tprotected: %llu\n",
@@ -4186,16 +4273,14 @@ dump_l2arc_log_blocks(int fd, l2arc_dev_hdr_phys_t l2dhdr,
 		switch (L2BLK_GET_COMPRESS((&lbps[0])->lbp_prop)) {
 		case ZIO_COMPRESS_OFF:
 			break;
-		case ZIO_COMPRESS_LZ4:
+		default:
 			abd = abd_alloc_for_io(asize, B_TRUE);
 			abd_copy_from_buf_off(abd, &this_lb, 0, asize);
 			zio_decompress_data(L2BLK_GET_COMPRESS(
 			    (&lbps[0])->lbp_prop), abd, &this_lb,
-			    asize, sizeof (this_lb));
+			    asize, sizeof (this_lb), NULL);
 			abd_free(abd);
 			break;
-		default:
-			break;
 		}
 
 		if (this_lb.lb_magic == BSWAP_64(L2ARC_LOG_BLK_MAGIC))
@@ -7684,9 +7769,9 @@ zdb_decompress_block(abd_t *pabd, void *buf, void *lbuf, uint64_t lsize,
 			VERIFY0(random_get_pseudo_bytes(lbuf2, lsize));
 
 			if (zio_decompress_data(*cfuncp, pabd,
-			    lbuf, psize, lsize) == 0 &&
+			    lbuf, psize, lsize, NULL) == 0 &&
 			    zio_decompress_data(*cfuncp, pabd,
-			    lbuf2, psize, lsize) == 0 &&
+			    lbuf2, psize, lsize, NULL) == 0 &&
 			    bcmp(lbuf, lbuf2, lsize) == 0)
 				break;
 		}
@@ -8078,7 +8163,7 @@ main(int argc, char **argv)
 	zfs_btree_verify_intensity = 3;
 
 	while ((c = getopt(argc, argv,
-	    "AbcCdDeEFGhiI:klLmMo:Op:PqRsSt:uU:vVx:XYy")) != -1) {
+	    "AbcCdDeEFGhiI:klLmMo:Op:PqRsSt:uU:vVx:XYyZ")) != -1) {
 		switch (c) {
 		case 'b':
 		case 'c':
@@ -8098,6 +8183,7 @@ main(int argc, char **argv)
 		case 'S':
 		case 'u':
 		case 'y':
+		case 'Z':
 			dump_opt[c]++;
 			dump_all = 0;
 			break;

configure.ac

@@ -136,6 +136,7 @@ AC_CONFIG_FILES([
 	include/sys/fs/Makefile
 	include/sys/lua/Makefile
 	include/sys/sysevent/Makefile
+	include/sys/zstd/Makefile
 	lib/Makefile
 	lib/libavl/Makefile
 	lib/libefi/Makefile
@@ -163,6 +164,7 @@ AC_CONFIG_FILES([
 	lib/libzfs_core/Makefile
 	lib/libzfs_core/libzfs_core.pc
 	lib/libzpool/Makefile
+	lib/libzstd/Makefile
 	lib/libzutil/Makefile
 	man/Makefile
 	man/man1/Makefile
@@ -180,6 +182,7 @@ AC_CONFIG_FILES([
 	module/unicode/Makefile
 	module/zcommon/Makefile
 	module/zfs/Makefile
+	module/zstd/Makefile
 	rpm/Makefile
 	rpm/generic/Makefile
 	rpm/generic/zfs-dkms.spec

include/sys/Makefile.am

@@ -1,4 +1,4 @@
-SUBDIRS = fm fs crypto lua sysevent
+SUBDIRS = fm fs crypto lua sysevent zstd
 
 COMMON_H = \
 	abd.h \

include/sys/arc.h

@@ -22,6 +22,8 @@
  * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
  * Copyright (c) 2012, 2016 by Delphix. All rights reserved.
  * Copyright (c) 2013 by Saso Kiselkov. All rights reserved.
+ * Copyright (c) 2019, Allan Jude
+ * Copyright (c) 2019, Klara Inc.
  */
 
 #ifndef	_SYS_ARC_H
@@ -252,18 +254,20 @@ void arc_convert_to_raw(arc_buf_t *buf, uint64_t dsobj, boolean_t byteorder,
 arc_buf_t *arc_alloc_buf(spa_t *spa, void *tag, arc_buf_contents_t type,
     int32_t size);
 arc_buf_t *arc_alloc_compressed_buf(spa_t *spa, void *tag,
-    uint64_t psize, uint64_t lsize, enum zio_compress compression_type);
+    uint64_t psize, uint64_t lsize, enum zio_compress compression_type,
+    uint8_t complevel);
 arc_buf_t *arc_alloc_raw_buf(spa_t *spa, void *tag, uint64_t dsobj,
     boolean_t byteorder, const uint8_t *salt, const uint8_t *iv,
     const uint8_t *mac, dmu_object_type_t ot, uint64_t psize, uint64_t lsize,
-    enum zio_compress compression_type);
+    enum zio_compress compression_type, uint8_t complevel);
+uint8_t arc_get_complevel(arc_buf_t *buf);
 arc_buf_t *arc_loan_buf(spa_t *spa, boolean_t is_metadata, int size);
 arc_buf_t *arc_loan_compressed_buf(spa_t *spa, uint64_t psize, uint64_t lsize,
-    enum zio_compress compression_type);
+    enum zio_compress compression_type, uint8_t complevel);
 arc_buf_t *arc_loan_raw_buf(spa_t *spa, uint64_t dsobj, boolean_t byteorder,
     const uint8_t *salt, const uint8_t *iv, const uint8_t *mac,
     dmu_object_type_t ot, uint64_t psize, uint64_t lsize,
-    enum zio_compress compression_type);
+    enum zio_compress compression_type, uint8_t complevel);
 void arc_return_buf(arc_buf_t *buf, void *tag);
 void arc_loan_inuse_buf(arc_buf_t *buf, void *tag);
 void arc_buf_destroy(arc_buf_t *buf, void *tag);

include/sys/arc_impl.h

@@ -269,12 +269,13 @@ typedef struct l2arc_log_ent_phys {
 	 */
 	uint64_t		le_prop;
 	uint64_t		le_daddr;	/* buf location on l2dev */
+	uint64_t		le_complevel;
 	/*
 	 * We pad the size of each entry to a power of 2 so that the size of
 	 * l2arc_log_blk_phys_t is power-of-2 aligned with SPA_MINBLOCKSHIFT,
 	 * because of the L2ARC_SET_*SIZE macros.
 	 */
-	const uint64_t		le_pad[3];	/* pad to 64 bytes	 */
+	const uint64_t		le_pad[2];	/* pad to 64 bytes	 */
 } l2arc_log_ent_phys_t;
 
 #define	L2ARC_LOG_BLK_MAX_ENTRIES	(1022)
@@ -460,6 +461,9 @@ struct arc_buf_hdr {
 	uint64_t		b_birth;
 
 	arc_buf_contents_t	b_type;
+	uint8_t			b_complevel;
+	uint8_t			b_reserved1; /* used for 4 byte alignment */
+	uint16_t		b_reserved2; /* used for 4 byte alignment */
 	arc_buf_hdr_t		*b_hash_next;
 	arc_flags_t		b_flags;
 

include/sys/dmu_objset.h

@@ -118,6 +118,7 @@ struct objset {
 	uint64_t os_dnodesize; /* default dnode size for new objects */
 	enum zio_checksum os_checksum;
 	enum zio_compress os_compress;
+	uint8_t os_complevel;
 	uint8_t os_copies;
 	enum zio_checksum os_dedup_checksum;
 	boolean_t os_dedup_verify;

include/sys/dsl_dataset.h

@@ -435,6 +435,8 @@ int dsl_dataset_set_refquota(const char *dsname, zprop_source_t source,
     uint64_t quota);
 int dsl_dataset_set_refreservation(const char *dsname, zprop_source_t source,
     uint64_t reservation);
+int dsl_dataset_set_compression(const char *dsname, zprop_source_t source,
+    uint64_t compression);
 
 boolean_t dsl_dataset_is_before(dsl_dataset_t *later, dsl_dataset_t *earlier,
     uint64_t earlier_txg);

include/sys/spa.h

@@ -28,6 +28,8 @@
  * Copyright 2017 Joyent, Inc.
  * Copyright (c) 2017, 2019, Datto Inc. All rights reserved.
  * Copyright (c) 2017, Intel Corporation.
+ * Copyright (c) 2019, Allan Jude
+ * Copyright (c) 2019, Klara Inc.
  */
 
 #ifndef _SYS_SPA_H
@@ -120,6 +122,7 @@ struct dsl_crypto_params;
 
 #define	SPA_COMPRESSBITS	7
 #define	SPA_VDEVBITS		24
+#define	SPA_COMPRESSMASK	((1U << SPA_COMPRESSBITS) - 1)
 
 /*
  * All SPA data is represented by 128-bit data virtual addresses (DVAs).

include/sys/zfs_context.h

@@ -748,6 +748,12 @@ extern int kmem_cache_reap_active(void);
 
 #define	____cacheline_aligned
 
+/*
+ * Kernel modules
+ */
+#define	__init
+#define	__exit
+
 #endif /* _KERNEL */
 
 #ifdef __cplusplus

include/sys/zfs_ioctl.h

@@ -105,7 +105,7 @@ typedef enum drr_headertype {
 #define	DMU_BACKUP_FEATURE_COMPRESSED		(1 << 22)
 #define	DMU_BACKUP_FEATURE_LARGE_DNODE		(1 << 23)
 #define	DMU_BACKUP_FEATURE_RAW			(1 << 24)
-/* flag #25 is reserved for the ZSTD compression feature */
+#define	DMU_BACKUP_FEATURE_ZSTD			(1 << 25)
 #define	DMU_BACKUP_FEATURE_HOLDS		(1 << 26)
 /*
  * The SWITCH_TO_LARGE_BLOCKS feature indicates that we can receive
@@ -132,7 +132,8 @@ typedef enum drr_headertype {
     DMU_BACKUP_FEATURE_RESUMING | DMU_BACKUP_FEATURE_LARGE_BLOCKS | \
     DMU_BACKUP_FEATURE_COMPRESSED | DMU_BACKUP_FEATURE_LARGE_DNODE | \
     DMU_BACKUP_FEATURE_RAW | DMU_BACKUP_FEATURE_HOLDS | \
-    DMU_BACKUP_FEATURE_REDACTED | DMU_BACKUP_FEATURE_SWITCH_TO_LARGE_BLOCKS)
+    DMU_BACKUP_FEATURE_REDACTED | DMU_BACKUP_FEATURE_SWITCH_TO_LARGE_BLOCKS | \
+    DMU_BACKUP_FEATURE_ZSTD)
 
 /* Are all features in the given flag word currently supported? */
 #define	DMU_STREAM_SUPPORTED(x)	(!((x) & ~DMU_BACKUP_FEATURE_MASK))