zio_compress: introduce compress_threshold

Now default compression is lz4, which can stop
compression process by itself on incompressible data.
If there are additional size checks -
we will only make our compressratio worse.

New usable compression thresholds are:
- less than BPE_PAYLOAD_SIZE (embedded_data feature);
- at least one saved sector.

Old 12.5% threshold is left to minimize affect
on existing user expectations of CPU utilization.

If data wasn't compressed - it will be saved as
ZIO_COMPRESS_OFF, so if we really need to recompress
data without ashift info and check anything -
we can just compress it with zero threshold.
So, we don't need a new feature flag here!

Signed-off-by: George Melikov <[email protected]>

include/sys/zio_compress.h

@@ -25,6 +25,7 @@
  * Copyright (c) 2019, Klara Inc.
  * Use is subject to license terms.
  * Copyright (c) 2015, 2016 by Delphix. All rights reserved.
+ * Copyright (c) 2021 by George Melikov. All rights reserved.
  */
 
 #ifndef _SYS_ZIO_COMPRESS_H
@@ -184,7 +185,7 @@ extern int lz4_decompress_zfs(void *src, void *dst, size_t s_len, size_t d_len,
  * Compress and decompress data if necessary.
  */
 extern size_t zio_compress_data(enum zio_compress c, abd_t *src, void **dst,
-    size_t s_len, uint8_t level);
+    size_t s_len, uint8_t level, int compress_threshold);
 extern int zio_decompress_data(enum zio_compress c, abd_t *src, void *dst,
     size_t s_len, size_t d_len, uint8_t *level);
 extern int zio_decompress_data_buf(enum zio_compress c, void *src, void *dst,

man/man7/zfsprops.7

@@ -917,14 +917,24 @@ zeroes (the NUL byte).
 When a zero-filled block is detected, it is stored as
 a hole and not compressed using the indicated compression algorithm.
 .Pp
-Any block being compressed must be no larger than 7/8 of its original size
-after compression, otherwise the compression will not be considered worthwhile
-and the block saved uncompressed.
-Note that when the logical block is less than
-8 times the disk sector size this effectively reduces the necessary compression
-ratio; for example, 8 KiB blocks on disks with 4 KiB disk sectors must compress
-to 1/2
-or less of their original size.
+All blocks are allocated as a whole number of sectors
+.Pq chunks of 2^ Ns Sy ashift No bytes , e.g . Sy 512B No or Sy 4KB .
+Compression may result in a non-sector-aligned size, which will be rounded up
+to a whole number of sectors.
+If compression saves less than one whole sector,
+the block will be stored uncompressed.
+Therefore, blocks whose logical size is a small number of sectors will
+experience less compression
+(e.g. for
+.Sy recordsize Ns = Ns Sy 16K
+with
+.Sy 4K
+sectors, which have 4 sectors per block,
+compression needs to save at least 25% to actually save space on disk).
+.Pp
+There is
+.Sy 12.5%
+default compression threshold in addition to sector rounding.
 .It Xo
 .Sy context Ns = Ns Sy none Ns | Ns
 .Ar SELinux-User : Ns Ar SELinux-Role : Ns Ar SELinux-Type : Ns Ar Sensitivity-Level

module/zfs/arc.c

@@ -29,6 +29,7 @@
  * Copyright (c) 2019, Klara Inc.
  * Copyright (c) 2019, Allan Jude
  * Copyright (c) 2020, The FreeBSD Foundation [1]
+ * Copyright (c) 2021 by George Melikov. All rights reserved.
  *
  * [1] Portions of this software were developed by Allan Jude
  *     under sponsorship from the FreeBSD Foundation.
@@ -1786,7 +1787,7 @@ arc_hdr_authenticate(arc_buf_hdr_t *hdr, spa_t *spa, uint64_t dsobj)
 	    !HDR_COMPRESSION_ENABLED(hdr)) {
 
 		csize = zio_compress_data(HDR_GET_COMPRESS(hdr),
-		    hdr->b_l1hdr.b_pabd, &tmpbuf, lsize, hdr->b_complevel);
+		    hdr->b_l1hdr.b_pabd, &tmpbuf, lsize, hdr->b_complevel, 0);
 		ASSERT3P(tmpbuf, !=, NULL);
 		ASSERT3U(csize, <=, psize);
 		abd = abd_get_from_buf(tmpbuf, lsize);
@@ -8945,7 +8946,7 @@ l2arc_apply_transforms(spa_t *spa, arc_buf_hdr_t *hdr, uint64_t asize,
 		tmp = abd_borrow_buf(cabd, bufsize);
 
 		psize = zio_compress_data(compress, to_write, &tmp, size,
-		    hdr->b_complevel);
+		    hdr->b_complevel, 0);
 
 		if (psize >= asize) {
 			psize = HDR_GET_PSIZE(hdr);
@@ -8957,6 +8958,7 @@ l2arc_apply_transforms(spa_t *spa, arc_buf_hdr_t *hdr, uint64_t asize,
 				abd_zero_off(to_write, psize, asize - psize);
 			goto encrypt;
 		}
+
 		ASSERT3U(psize, <=, HDR_GET_PSIZE(hdr));
 		if (psize < asize)
 			memset((char *)tmp + psize, 0, bufsize - psize);
@@ -10442,7 +10444,8 @@ l2arc_log_blk_commit(l2arc_dev_t *dev, zio_t *pio, l2arc_write_callback_t *cb)
 
 	/* try to compress the buffer */
 	psize = zio_compress_data(ZIO_COMPRESS_LZ4,
-	    abd_buf->abd, (void **) &tmpbuf, sizeof (*lb), 0);
+	    abd_buf->abd, (void **) &tmpbuf, sizeof (*lb), 0,
+	    1 << dev->l2ad_vdev->vdev_ashift);
 
 	/* a log block is never entirely zero */
 	ASSERT(psize != 0);

module/zfs/zio.c

@@ -26,6 +26,7 @@
  * Copyright (c) 2019, Klara Inc.
  * Copyright (c) 2019, Allan Jude
  * Copyright (c) 2021, Datto, Inc.
+ * Copyright (c) 2021 by George Melikov. All rights reserved.
  */
 
 #include <sys/sysmacros.h>
@@ -1849,7 +1850,7 @@ zio_write_compress(zio_t *zio)
 	    !(zio->io_flags & ZIO_FLAG_RAW_COMPRESS)) {
 		void *cbuf = NULL;
 		psize = zio_compress_data(compress, zio->io_abd, &cbuf, lsize,
-		    zp->zp_complevel);
+		    zp->zp_complevel, spa->spa_min_alloc);
 		if (psize == 0) {
 			compress = ZIO_COMPRESS_OFF;
 		} else if (psize >= lsize) {
@@ -1914,7 +1915,7 @@ zio_write_compress(zio_t *zio)
 		 * to a hole.
 		 */
 		psize = zio_compress_data(ZIO_COMPRESS_EMPTY,
-		    zio->io_abd, NULL, lsize, zp->zp_complevel);
+		    zio->io_abd, NULL, lsize, zp->zp_complevel, 0);
 		if (psize == 0 || psize >= lsize)
 			compress = ZIO_COMPRESS_OFF;
 	} else if (zio->io_flags & ZIO_FLAG_RAW_COMPRESS &&

module/zfs/zio_compress.c

@@ -22,15 +22,11 @@
 /*
  * Copyright 2009 Sun Microsystems, Inc.  All rights reserved.
  * Use is subject to license terms.
- */
-/*
  * Copyright (c) 2013 by Saso Kiselkov. All rights reserved.
- */
-
-/*
  * Copyright (c) 2013, 2018 by Delphix. All rights reserved.
  * Copyright (c) 2019, Klara Inc.
  * Copyright (c) 2019, Allan Jude
+ * Copyright (c) 2021 by George Melikov. All rights reserved.
  */
 
 #include <sys/zfs_context.h>
@@ -124,9 +120,12 @@ zio_compress_zeroed_cb(void *data, size_t len, void *private)
 	return (0);
 }
 
+/*
+ * Set compress_threshold=0 for reproducibility of ARC checks.
+ */
 size_t
 zio_compress_data(enum zio_compress c, abd_t *src, void **dst, size_t s_len,
-    uint8_t level)
+    uint8_t level, int compress_threshold)
 {
 	size_t c_len, d_len;
 	uint8_t complevel;
@@ -145,8 +144,19 @@ zio_compress_data(enum zio_compress c, abd_t *src, void **dst, size_t s_len,
 	if (c == ZIO_COMPRESS_EMPTY)
 		return (s_len);
 
-	/* Compress at least 12.5% */
-	d_len = s_len - (s_len >> 3);
+	/*
+	 * Usable compression thresholds are:
+	 * - less than BPE_PAYLOAD_SIZE (embedded_data feature)
+	 * - at least one saved sector
+	 * - minimun 12.5% should be saved (legacy value)
+	 */
+	if (compress_threshold > 0) {
+		d_len = MAX(BPE_PAYLOAD_SIZE,
+		    s_len - MAX(s_len % compress_threshold, s_len >> 3));
+	} else {
+		/* Special case for reproducibility of ARC checks */
+		d_len = s_len;
+	}
 
 	complevel = ci->ci_level;
 
@@ -174,7 +184,6 @@ zio_compress_data(enum zio_compress c, abd_t *src, void **dst, size_t s_len,
 	if (c_len > d_len)
 		return (s_len);
 
-	ASSERT3U(c_len, <=, d_len);
 	return (c_len);
 }