zio_compress: introduce max size 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!

Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Reviewed-by: Tony Hutter <hutter2@llnl.gov>
Reviewed-by: Alexander Motin <mav@FreeBSD.org>
Signed-off-by: George Melikov <mail@gmelikov.ru>
Closes #9416

diff --git a/cmd/zstream/zstream_recompress.c b/cmd/zstream/zstream_recompress.c
index ae2c56320b2acc0cc6acd100fb90c4dc34d3aa1c..586ac5623aa57b9e5a08ddcf9e650bc762210b99 100644 (file)
--- a/cmd/zstream/zstream_recompress.c
+++ b/cmd/zstream/zstream_recompress.c
@@ -288,7 +288,8 @@ zstream_do_recompress(int argc, char *argv[])
                                abd_t *pabd =
                                    abd_get_from_buf_struct(&abd, buf, bufsz);
                                size_t csize = zio_compress_data(ctype, &dabd,
-                                   &pabd, drrw->drr_logical_size, level);
+                                   &pabd, drrw->drr_logical_size,
+                                   drrw->drr_logical_size, level);
                                size_t rounded =
                                    P2ROUNDUP(csize, SPA_MINBLOCKSIZE);
                                if (rounded >= drrw->drr_logical_size) {

diff --git a/include/sys/zio.h b/include/sys/zio.h
index 628416e98eb964e40aff9d14e4708f528efd81d2..9e3b66dc01db65892327a1103d9ebfd19e20249f 100644 (file)
--- a/include/sys/zio.h
+++ b/include/sys/zio.h
@@ -29,6 +29,7 @@
  * Copyright (c) 2019, Allan Jude
  * Copyright (c) 2019, 2023, 2024, Klara Inc.
  * Copyright (c) 2019-2020, Michael Niewöhner
+ * Copyright (c) 2024 by George Melikov. All rights reserved.
  */
 
 #ifndef _ZIO_H
@@ -603,6 +604,8 @@ extern int zio_alloc_zil(spa_t *spa, objset_t *os, uint64_t txg,
 extern void zio_flush(zio_t *zio, vdev_t *vd);
 extern void zio_shrink(zio_t *zio, uint64_t size);
 
+extern size_t zio_get_compression_max_size(uint64_t gcd_alloc,
+    uint64_t min_alloc, size_t s_len);
 extern int zio_wait(zio_t *zio);
 extern void zio_nowait(zio_t *zio);
 extern void zio_execute(void *zio);

diff --git a/include/sys/zio_compress.h b/include/sys/zio_compress.h
index 31602039a150dd713d26335df346f418c340dd2c..ceef757abd20250595fcd97c9efd4d6615ec7d41 100644 (file)
--- a/include/sys/zio_compress.h
+++ b/include/sys/zio_compress.h
@@ -25,6 +25,7 @@
  * Copyright (c) 2019, 2024, Klara, Inc.
  * Use is subject to license terms.
  * Copyright (c) 2015, 2016 by Delphix. All rights reserved.
+ * Copyright (c) 2021, 2024 by George Melikov. All rights reserved.
  */
 
 #ifndef _SYS_ZIO_COMPRESS_H
@@ -174,7 +175,7 @@ extern int zfs_lz4_decompress(abd_t *src, abd_t *dst, size_t s_len,
  * Compress and decompress data if necessary.
  */
 extern size_t zio_compress_data(enum zio_compress c, abd_t *src, abd_t **dst,
-    size_t s_len, uint8_t level);
+    size_t s_len, size_t d_len, uint8_t level);
 extern int zio_decompress_data(enum zio_compress c, abd_t *src, abd_t *abd,
     size_t s_len, size_t d_len, uint8_t *level);
 extern int zio_compress_to_feature(enum zio_compress comp);

diff --git a/man/man7/zfsprops.7 b/man/man7/zfsprops.7
index fa228e9bd7feec3c3e201c9cffe43a9459b2658c..a3ae4d877f7909c7107ac73b602a306df0965b86 100644 (file)
--- a/man/man7/zfsprops.7
+++ b/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

diff --git a/module/zfs/arc.c b/module/zfs/arc.c
index b5bcd367b247fda773cb7b55e7943b29a47b81e1..8e513a4e60fd15c51637fa16f2e61f5894075279 100644 (file)
--- a/module/zfs/arc.c
+++ b/module/zfs/arc.c
@@ -29,6 +29,7 @@
  * Copyright (c) 2019, 2023, Klara Inc.
  * Copyright (c) 2019, Allan Jude
  * Copyright (c) 2020, The FreeBSD Foundation [1]
+ * Copyright (c) 2021, 2024 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)) {
                abd = NULL;
                csize = zio_compress_data(HDR_GET_COMPRESS(hdr),
-                   hdr->b_l1hdr.b_pabd, &abd, lsize, hdr->b_complevel);
+                   hdr->b_l1hdr.b_pabd, &abd, lsize, lsize, hdr->b_complevel);
                ASSERT3P(abd, !=, NULL);
                ASSERT3U(csize, <=, psize);
                abd_zero_off(abd, csize, psize - csize);
@@ -9029,8 +9030,8 @@ l2arc_apply_transforms(spa_t *spa, arc_buf_hdr_t *hdr, uint64_t asize,
        if (compress != ZIO_COMPRESS_OFF && !HDR_COMPRESSION_ENABLED(hdr)) {
                cabd = abd_alloc_for_io(MAX(size, asize), ismd);
                uint64_t csize = zio_compress_data(compress, to_write, &cabd,
-                   size, hdr->b_complevel);
-               if (csize > psize) {
+                   size, MIN(size, psize), hdr->b_complevel);
+               if (csize >= size || csize > psize) {
                        /*
                         * We can't re-compress the block into the original
                         * psize.  Even if it fits into asize, it does not
@@ -10521,9 +10522,11 @@ l2arc_log_blk_commit(l2arc_dev_t *dev, zio_t *pio, l2arc_write_callback_t *cb)
         */
        list_insert_tail(&cb->l2wcb_abd_list, abd_buf);
 
-       /* try to compress the buffer */
+       /* try to compress the buffer, at least one sector to save */
        psize = zio_compress_data(ZIO_COMPRESS_LZ4,
-           abd_buf->abd, &abd, sizeof (*lb), 0);
+           abd_buf->abd, &abd, sizeof (*lb),
+           zio_get_compression_max_size(dev->l2ad_vdev->vdev_ashift,
+           dev->l2ad_vdev->vdev_ashift, sizeof (*lb)), 0);
 
        /* a log block is never entirely zero */
        ASSERT(psize != 0);

diff --git a/module/zfs/dmu_recv.c b/module/zfs/dmu_recv.c
index a1752650f3bae0595f6a039b28befd7068698af9..9b3da032f35458d98ab82752f9ff861accfeb42c 100644 (file)
--- a/module/zfs/dmu_recv.c
+++ b/module/zfs/dmu_recv.c
@@ -1408,7 +1408,7 @@ do_corrective_recv(struct receive_writer_arg *rwa, struct drr_write *drrw,
                abd_t *cabd = abd_alloc_linear(BP_GET_PSIZE(bp),
                    B_FALSE);
                uint64_t csize = zio_compress_data(BP_GET_COMPRESS(bp),
-                   abd, &cabd, abd_get_size(abd),
+                   abd, &cabd, abd_get_size(abd), BP_GET_PSIZE(bp),
                    rwa->os->os_complevel);
                abd_zero_off(cabd, csize, BP_GET_PSIZE(bp) - csize);
                /* Swap in newly compressed data into the abd */

diff --git a/module/zfs/zio.c b/module/zfs/zio.c
index 2e38d47c99b1544173751ff8d62e249c9fd9bc11..4ba9260819c5550d67a1fa8556a14805f8f467fa 100644 (file)
--- a/module/zfs/zio.c
+++ b/module/zfs/zio.c
@@ -26,6 +26,7 @@
  * Copyright (c) 2019, 2023, 2024, Klara Inc.
  * Copyright (c) 2019, Allan Jude
  * Copyright (c) 2021, Datto, Inc.
+ * Copyright (c) 2021, 2024 by George Melikov. All rights reserved.
  */
 
 #include <sys/sysmacros.h>
@@ -1704,6 +1705,21 @@ zio_roundup_alloc_size(spa_t *spa, uint64_t size)
        return (spa->spa_min_alloc);
 }
 
+size_t
+zio_get_compression_max_size(uint64_t gcd_alloc, uint64_t min_alloc,
+    size_t s_len)
+{
+       size_t d_len;
+
+       /* minimum 12.5% must be saved (legacy value, may be changed later) */
+       d_len = s_len - (s_len >> 3);
+
+       d_len = d_len - d_len % gcd_alloc;
+       if (d_len < min_alloc)
+               return (BPE_PAYLOAD_SIZE);
+       return (d_len);
+}
+
 /*
  * ==========================================================================
  * Prepare to read and write logical blocks
@@ -1885,7 +1901,10 @@ zio_write_compress(zio_t *zio)
                        psize = lsize;
                else
                        psize = zio_compress_data(compress, zio->io_abd, &cabd,
-                           lsize, zp->zp_complevel);
+                           lsize,
+                           zio_get_compression_max_size(spa->spa_gcd_alloc,
+                           spa->spa_min_alloc, lsize),
+                           zp->zp_complevel);
                if (psize == 0) {
                        compress = ZIO_COMPRESS_OFF;
                } else if (psize >= lsize) {

diff --git a/module/zfs/zio_compress.c b/module/zfs/zio_compress.c
index 9182917f75ebf901690c807474f4bd5b77c2754a..10c4825738623b97c45579ad92e7fc90b8acf3eb 100644 (file)
--- a/module/zfs/zio_compress.c
+++ b/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, 2024, Klara, Inc.
  * Copyright (c) 2019, Allan Jude
+ * Copyright (c) 2021, 2024 by George Melikov. All rights reserved.
  */
 
 #include <sys/zfs_context.h>
@@ -129,9 +125,9 @@ zio_compress_select(spa_t *spa, enum zio_compress child,
 
 size_t
 zio_compress_data(enum zio_compress c, abd_t *src, abd_t **dst, size_t s_len,
-    uint8_t level)
+    size_t d_len, uint8_t level)
 {
-       size_t c_len, d_len;
+       size_t c_len;
        uint8_t complevel;
        zio_compress_info_t *ci = &zio_compress_table[c];
 
@@ -156,15 +152,11 @@ zio_compress_data(enum zio_compress c, abd_t *src, abd_t **dst, size_t s_len,
        if (*dst == NULL)
                *dst = abd_alloc_sametype(src, s_len);
 
-       /* Compress at least 12.5%, but limit to the size of the dest abd. */
-       d_len = MIN(s_len - (s_len >> 3), abd_get_size(*dst));
-
        c_len = ci->ci_compress(src, *dst, s_len, d_len, complevel);
 
        if (c_len > d_len)
                return (s_len);
 
-       ASSERT3U(c_len, <=, d_len);
        return (c_len);
 }