summaryrefslogtreecommitdiffstats
path: root/module/zfs
diff options
context:
space:
mode:
Diffstat (limited to 'module/zfs')
-rw-r--r--module/zfs/dsl_pool.c2
-rw-r--r--module/zfs/metaslab.c39
-rw-r--r--module/zfs/spa_misc.c2
-rw-r--r--module/zfs/zfs_debug.c7
-rw-r--r--module/zfs/zio.c30
5 files changed, 65 insertions, 15 deletions
diff --git a/module/zfs/dsl_pool.c b/module/zfs/dsl_pool.c
index f1ab29c62..aa2f743b1 100644
--- a/module/zfs/dsl_pool.c
+++ b/module/zfs/dsl_pool.c
@@ -20,7 +20,7 @@
*/
/*
* Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
- * Copyright (c) 2013 by Delphix. All rights reserved.
+ * Copyright (c) 2011, 2014 by Delphix. All rights reserved.
* Copyright (c) 2013 Steven Hartland. All rights reserved.
*/
diff --git a/module/zfs/metaslab.c b/module/zfs/metaslab.c
index 098d73f95..d41a570c5 100644
--- a/module/zfs/metaslab.c
+++ b/module/zfs/metaslab.c
@@ -65,6 +65,21 @@ uint64_t metaslab_gang_bang = SPA_MAXBLOCKSIZE + 1; /* force gang blocks */
int zfs_condense_pct = 200;
/*
+ * Condensing a metaslab is not guaranteed to actually reduce the amount of
+ * space used on disk. In particular, a space map uses data in increments of
+ * MAX(1 << ashift, SPACE_MAP_INITIAL_BLOCKSIZE), so a metaslab might use the
+ * same number of blocks after condensing. Since the goal of condensing is to
+ * reduce the number of IOPs required to read the space map, we only want to
+ * condense when we can be sure we will reduce the number of blocks used by the
+ * space map. Unfortunately, we cannot precisely compute whether or not this is
+ * the case in metaslab_should_condense since we are holding ms_lock. Instead,
+ * we apply the following heuristic: do not condense a spacemap unless the
+ * uncondensed size consumes greater than zfs_metaslab_condense_block_threshold
+ * blocks.
+ */
+int zfs_metaslab_condense_block_threshold = 4;
+
+/*
* The zfs_mg_noalloc_threshold defines which metaslab groups should
* be eligible for allocation. The value is defined as a percentage of
* free space. Metaslab groups that have more free space than
@@ -1633,6 +1648,8 @@ metaslab_group_preload(metaslab_group_t *mg)
* times the size than the free space range tree representation
* (i.e. zfs_condense_pct = 110 and in-core = 1MB, minimal = 1.1.MB).
*
+ * 3. The on-disk size of the space map should actually decrease.
+ *
* Checking the first condition is tricky since we don't want to walk
* the entire AVL tree calculating the estimated on-disk size. Instead we
* use the size-ordered range tree in the metaslab and calculate the
@@ -1643,13 +1660,21 @@ metaslab_group_preload(metaslab_group_t *mg)
* To determine the second criterion we use a best-case estimate and assume
* each segment can be represented on-disk as a single 64-bit entry. We refer
* to this best-case estimate as the space map's minimal form.
+ *
+ * Unfortunately, we cannot compute the on-disk size of the space map in this
+ * context because we cannot accurately compute the effects of compression, etc.
+ * Instead, we apply the heuristic described in the block comment for
+ * zfs_metaslab_condense_block_threshold - we only condense if the space used
+ * is greater than a threshold number of blocks.
*/
static boolean_t
metaslab_should_condense(metaslab_t *msp)
{
space_map_t *sm = msp->ms_sm;
range_seg_t *rs;
- uint64_t size, entries, segsz;
+ uint64_t size, entries, segsz, object_size, optimal_size, record_size;
+ dmu_object_info_t doi;
+ uint64_t vdev_blocksize = 1 << msp->ms_group->mg_vd->vdev_ashift;
ASSERT(MUTEX_HELD(&msp->ms_lock));
ASSERT(msp->ms_loaded);
@@ -1674,9 +1699,15 @@ metaslab_should_condense(metaslab_t *msp)
entries = size / (MIN(size, SM_RUN_MAX));
segsz = entries * sizeof (uint64_t);
- return (segsz <= space_map_length(msp->ms_sm) &&
- space_map_length(msp->ms_sm) >= (zfs_condense_pct *
- sizeof (uint64_t) * avl_numnodes(&msp->ms_tree->rt_root)) / 100);
+ optimal_size = sizeof (uint64_t) * avl_numnodes(&msp->ms_tree->rt_root);
+ object_size = space_map_length(msp->ms_sm);
+
+ dmu_object_info_from_db(sm->sm_dbuf, &doi);
+ record_size = MAX(doi.doi_data_block_size, vdev_blocksize);
+
+ return (segsz <= object_size &&
+ object_size >= (optimal_size * zfs_condense_pct / 100) &&
+ object_size > zfs_metaslab_condense_block_threshold * record_size);
}
/*
diff --git a/module/zfs/spa_misc.c b/module/zfs/spa_misc.c
index abe313de4..bb740e569 100644
--- a/module/zfs/spa_misc.c
+++ b/module/zfs/spa_misc.c
@@ -20,7 +20,7 @@
*/
/*
* Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
- * Copyright (c) 2013 by Delphix. All rights reserved.
+ * Copyright (c) 2011, 2014 by Delphix. All rights reserved.
* Copyright 2011 Nexenta Systems, Inc. All rights reserved.
*/
diff --git a/module/zfs/zfs_debug.c b/module/zfs/zfs_debug.c
index 47b7834f5..e7f266b5f 100644
--- a/module/zfs/zfs_debug.c
+++ b/module/zfs/zfs_debug.c
@@ -20,7 +20,7 @@
*/
/*
* Copyright (c) 2010, Oracle and/or its affiliates. All rights reserved.
- * Copyright (c) 2013 by Delphix. All rights reserved.
+ * Copyright (c) 2012, 2014 by Delphix. All rights reserved.
*/
#include <sys/zfs_context.h>
@@ -136,7 +136,10 @@ zfs_dbgmsg_fini(void)
* echo ::zfs_dbgmsg | mdb -k
*
* Monitor these messages by running:
- * dtrace -q -n 'zfs-dbgmsg{printf("%s\n", stringof(arg0))}'
+ * dtrace -qn 'zfs-dbgmsg{printf("%s\n", stringof(arg0))}'
+ *
+ * When used with libzpool, monitor with:
+ * dtrace -qn 'zfs$pid::zfs_dbgmsg:probe1{printf("%s\n", copyinstr(arg1))}'
*/
void
zfs_dbgmsg(const char *fmt, ...)
diff --git a/module/zfs/zio.c b/module/zfs/zio.c
index 0ba167c62..9d70b3e59 100644
--- a/module/zfs/zio.c
+++ b/module/zfs/zio.c
@@ -889,8 +889,8 @@ zio_read_phys(zio_t *pio, vdev_t *vd, uint64_t offset, uint64_t size,
ASSERT3U(offset + size, <=, vd->vdev_psize);
zio = zio_create(pio, vd->vdev_spa, 0, NULL, data, size, done, private,
- ZIO_TYPE_READ, priority, flags, vd, offset, NULL,
- ZIO_STAGE_OPEN, ZIO_READ_PHYS_PIPELINE);
+ ZIO_TYPE_READ, priority, flags | ZIO_FLAG_PHYSICAL, vd, offset,
+ NULL, ZIO_STAGE_OPEN, ZIO_READ_PHYS_PIPELINE);
zio->io_prop.zp_checksum = checksum;
@@ -910,8 +910,8 @@ zio_write_phys(zio_t *pio, vdev_t *vd, uint64_t offset, uint64_t size,
ASSERT3U(offset + size, <=, vd->vdev_psize);
zio = zio_create(pio, vd->vdev_spa, 0, NULL, data, size, done, private,
- ZIO_TYPE_WRITE, priority, flags, vd, offset, NULL,
- ZIO_STAGE_OPEN, ZIO_WRITE_PHYS_PIPELINE);
+ ZIO_TYPE_WRITE, priority, flags | ZIO_FLAG_PHYSICAL, vd, offset,
+ NULL, ZIO_STAGE_OPEN, ZIO_WRITE_PHYS_PIPELINE);
zio->io_prop.zp_checksum = checksum;
@@ -2642,7 +2642,9 @@ zio_vdev_io_start(zio_t *zio)
align = 1ULL << vd->vdev_top->vdev_ashift;
- if (P2PHASE(zio->io_size, align) != 0) {
+ if (!(zio->io_flags & ZIO_FLAG_PHYSICAL) &&
+ P2PHASE(zio->io_size, align) != 0) {
+ /* Transform logical writes to be a full physical block size. */
uint64_t asize = P2ROUNDUP(zio->io_size, align);
char *abuf = zio_buf_alloc(asize);
ASSERT(vd == vd->vdev_top);
@@ -2653,8 +2655,22 @@ zio_vdev_io_start(zio_t *zio)
zio_push_transform(zio, abuf, asize, asize, zio_subblock);
}
- ASSERT(P2PHASE(zio->io_offset, align) == 0);
- ASSERT(P2PHASE(zio->io_size, align) == 0);
+ /*
+ * If this is not a physical io, make sure that it is properly aligned
+ * before proceeding.
+ */
+ if (!(zio->io_flags & ZIO_FLAG_PHYSICAL)) {
+ ASSERT0(P2PHASE(zio->io_offset, align));
+ ASSERT0(P2PHASE(zio->io_size, align));
+ } else {
+ /*
+ * For physical writes, we allow 512b aligned writes and assume
+ * the device will perform a read-modify-write as necessary.
+ */
+ ASSERT0(P2PHASE(zio->io_offset, SPA_MINBLOCKSIZE));
+ ASSERT0(P2PHASE(zio->io_size, SPA_MINBLOCKSIZE));
+ }
+
VERIFY(zio->io_type != ZIO_TYPE_WRITE || spa_writeable(spa));
/*
generated by cgit v1.2.3 (git 2.39.1) at 2024-12-15 14:49:44 +0000