Optimize small random numbers generation
In all places except two spa_get_random() is used for small values, and the consumers do not require well seeded high quality values. Switch those two exceptions directly to random_get_pseudo_bytes() and optimize spa_get_random(), renaming it to random_in_range(), since it is not related to SPA or ZFS in general. On FreeBSD directly map random_in_range() to new prng32_bounded() KPI added in FreeBSD 13. On Linux and in user-space just reduce the type used to uint32_t to avoid more expensive 64bit division. Reviewed-by: Ryan Moeller <ryan@iXsystems.com> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Signed-off-by: Alexander Motin <mav@FreeBSD.org> Sponsored-By: iXsystems, Inc. Closes #12183
This commit is contained in:
parent
ba91311561
commit
29274c9f6d
@ -30,6 +30,9 @@
|
||||
#define _OPENSOLARIS_SYS_RANDOM_H_
|
||||
|
||||
#include_next <sys/random.h>
|
||||
#if __FreeBSD_version >= 1300108
|
||||
#include <sys/prng.h>
|
||||
#endif
|
||||
|
||||
static inline int
|
||||
random_get_bytes(uint8_t *p, size_t s)
|
||||
@ -45,4 +48,23 @@ random_get_pseudo_bytes(uint8_t *p, size_t s)
|
||||
return (0);
|
||||
}
|
||||
|
||||
static inline uint32_t
|
||||
random_in_range(uint32_t range)
|
||||
{
|
||||
#if __FreeBSD_version >= 1300108
|
||||
return (prng32_bounded(range));
|
||||
#else
|
||||
uint32_t r;
|
||||
|
||||
ASSERT(range != 0);
|
||||
|
||||
if (range == 1)
|
||||
return (0);
|
||||
|
||||
(void) random_get_pseudo_bytes((void *)&r, sizeof (r));
|
||||
|
||||
return (r % range);
|
||||
#endif
|
||||
}
|
||||
|
||||
#endif /* !_OPENSOLARIS_SYS_RANDOM_H_ */
|
||||
|
@ -36,4 +36,19 @@ random_get_bytes(uint8_t *ptr, size_t len)
|
||||
|
||||
extern int random_get_pseudo_bytes(uint8_t *ptr, size_t len);
|
||||
|
||||
static __inline__ uint32_t
|
||||
random_in_range(uint32_t range)
|
||||
{
|
||||
uint32_t r;
|
||||
|
||||
ASSERT(range != 0);
|
||||
|
||||
if (range == 1)
|
||||
return (0);
|
||||
|
||||
(void) random_get_pseudo_bytes((void *)&r, sizeof (r));
|
||||
|
||||
return (r % range);
|
||||
}
|
||||
|
||||
#endif /* _SPL_RANDOM_H */
|
||||
|
@ -1065,7 +1065,6 @@ extern spa_t *spa_by_guid(uint64_t pool_guid, uint64_t device_guid);
|
||||
extern boolean_t spa_guid_exists(uint64_t pool_guid, uint64_t device_guid);
|
||||
extern char *spa_strdup(const char *);
|
||||
extern void spa_strfree(char *);
|
||||
extern uint64_t spa_get_random(uint64_t range);
|
||||
extern uint64_t spa_generate_guid(spa_t *spa);
|
||||
extern void snprintf_blkptr(char *buf, size_t buflen, const blkptr_t *bp);
|
||||
extern void spa_freeze(spa_t *spa);
|
||||
|
@ -638,6 +638,21 @@ extern int lowbit64(uint64_t i);
|
||||
extern int random_get_bytes(uint8_t *ptr, size_t len);
|
||||
extern int random_get_pseudo_bytes(uint8_t *ptr, size_t len);
|
||||
|
||||
static __inline__ uint32_t
|
||||
random_in_range(uint32_t range)
|
||||
{
|
||||
uint32_t r;
|
||||
|
||||
ASSERT(range != 0);
|
||||
|
||||
if (range == 1)
|
||||
return (0);
|
||||
|
||||
(void) random_get_pseudo_bytes((void *)&r, sizeof (r));
|
||||
|
||||
return (r % range);
|
||||
}
|
||||
|
||||
extern void kernel_init(int mode);
|
||||
extern void kernel_fini(void);
|
||||
extern void random_init(void);
|
||||
|
@ -437,7 +437,7 @@ arc_available_memory(void)
|
||||
int64_t lowest = INT64_MAX;
|
||||
|
||||
/* Every 100 calls, free a small amount */
|
||||
if (spa_get_random(100) == 0)
|
||||
if (random_in_range(100) == 0)
|
||||
lowest = -1024;
|
||||
|
||||
return (lowest);
|
||||
@ -458,7 +458,7 @@ arc_all_memory(void)
|
||||
uint64_t
|
||||
arc_free_memory(void)
|
||||
{
|
||||
return (spa_get_random(arc_all_memory() * 20 / 100));
|
||||
return (random_in_range(arc_all_memory() * 20 / 100));
|
||||
}
|
||||
|
||||
void
|
||||
|
@ -5070,7 +5070,7 @@ metaslab_alloc_dva(spa_t *spa, metaslab_class_t *mc, uint64_t psize,
|
||||
* damage can result in extremely long reconstruction times. This
|
||||
* will also test spilling from special to normal.
|
||||
*/
|
||||
if (psize >= metaslab_force_ganging && (spa_get_random(100) < 3)) {
|
||||
if (psize >= metaslab_force_ganging && (random_in_range(100) < 3)) {
|
||||
metaslab_trace_add(zal, NULL, NULL, psize, d, TRACE_FORCE_GANG,
|
||||
allocator);
|
||||
return (SET_ERROR(ENOSPC));
|
||||
|
@ -523,9 +523,9 @@ mmp_write_uberblock(spa_t *spa)
|
||||
mutex_exit(&mmp->mmp_io_lock);
|
||||
|
||||
offset = VDEV_UBERBLOCK_OFFSET(vd, VDEV_UBERBLOCK_COUNT(vd) -
|
||||
MMP_BLOCKS_PER_LABEL + spa_get_random(MMP_BLOCKS_PER_LABEL));
|
||||
MMP_BLOCKS_PER_LABEL + random_in_range(MMP_BLOCKS_PER_LABEL));
|
||||
|
||||
label = spa_get_random(VDEV_LABELS);
|
||||
label = random_in_range(VDEV_LABELS);
|
||||
vdev_label_write(zio, vd, label, ub_abd, offset,
|
||||
VDEV_UBERBLOCK_SIZE(vd), mmp_write_done, mmp,
|
||||
flags | ZIO_FLAG_DONT_PROPAGATE);
|
||||
|
@ -20,9 +20,6 @@
|
||||
#include <sys/multilist.h>
|
||||
#include <sys/trace_zfs.h>
|
||||
|
||||
/* needed for spa_get_random() */
|
||||
#include <sys/spa.h>
|
||||
|
||||
/*
|
||||
* This overrides the number of sublists in each multilist_t, which defaults
|
||||
* to the number of CPUs in the system (see multilist_create()).
|
||||
@ -275,7 +272,7 @@ multilist_get_num_sublists(multilist_t *ml)
|
||||
unsigned int
|
||||
multilist_get_random_index(multilist_t *ml)
|
||||
{
|
||||
return (spa_get_random(ml->ml_num_sublists));
|
||||
return (random_in_range(ml->ml_num_sublists));
|
||||
}
|
||||
|
||||
/* Lock and return the sublist specified at the given index */
|
||||
|
@ -3175,7 +3175,7 @@ spa_activity_check(spa_t *spa, uberblock_t *ub, nvlist_t *config)
|
||||
import_delay = spa_activity_check_duration(spa, ub);
|
||||
|
||||
/* Add a small random factor in case of simultaneous imports (0-25%) */
|
||||
import_delay += import_delay * spa_get_random(250) / 1000;
|
||||
import_delay += import_delay * random_in_range(250) / 1000;
|
||||
|
||||
import_expire = gethrtime() + import_delay;
|
||||
|
||||
@ -4619,7 +4619,7 @@ spa_ld_checkpoint_rewind(spa_t *spa)
|
||||
vdev_t *svd[SPA_SYNC_MIN_VDEVS] = { NULL };
|
||||
int svdcount = 0;
|
||||
int children = rvd->vdev_children;
|
||||
int c0 = spa_get_random(children);
|
||||
int c0 = random_in_range(children);
|
||||
|
||||
for (int c = 0; c < children; c++) {
|
||||
vdev_t *vd = rvd->vdev_child[(c0 + c) % children];
|
||||
@ -9111,7 +9111,7 @@ spa_sync_rewrite_vdev_config(spa_t *spa, dmu_tx_t *tx)
|
||||
vdev_t *svd[SPA_SYNC_MIN_VDEVS] = { NULL };
|
||||
int svdcount = 0;
|
||||
int children = rvd->vdev_children;
|
||||
int c0 = spa_get_random(children);
|
||||
int c0 = random_in_range(children);
|
||||
|
||||
for (int c = 0; c < children; c++) {
|
||||
vdev_t *vd =
|
||||
|
@ -1494,32 +1494,21 @@ spa_strfree(char *s)
|
||||
kmem_free(s, strlen(s) + 1);
|
||||
}
|
||||
|
||||
uint64_t
|
||||
spa_get_random(uint64_t range)
|
||||
{
|
||||
uint64_t r;
|
||||
|
||||
ASSERT(range != 0);
|
||||
|
||||
if (range == 1)
|
||||
return (0);
|
||||
|
||||
(void) random_get_pseudo_bytes((void *)&r, sizeof (uint64_t));
|
||||
|
||||
return (r % range);
|
||||
}
|
||||
|
||||
uint64_t
|
||||
spa_generate_guid(spa_t *spa)
|
||||
{
|
||||
uint64_t guid = spa_get_random(-1ULL);
|
||||
uint64_t guid;
|
||||
|
||||
if (spa != NULL) {
|
||||
while (guid == 0 || spa_guid_exists(spa_guid(spa), guid))
|
||||
guid = spa_get_random(-1ULL);
|
||||
do {
|
||||
(void) random_get_pseudo_bytes((void *)&guid,
|
||||
sizeof (guid));
|
||||
} while (guid == 0 || spa_guid_exists(spa_guid(spa), guid));
|
||||
} else {
|
||||
while (guid == 0 || spa_guid_exists(guid, 0))
|
||||
guid = spa_get_random(-1ULL);
|
||||
do {
|
||||
(void) random_get_pseudo_bytes((void *)&guid,
|
||||
sizeof (guid));
|
||||
} while (guid == 0 || spa_guid_exists(guid, 0));
|
||||
}
|
||||
|
||||
return (guid);
|
||||
@ -2888,7 +2877,6 @@ EXPORT_SYMBOL(spa_maxdnodesize);
|
||||
EXPORT_SYMBOL(spa_guid_exists);
|
||||
EXPORT_SYMBOL(spa_strdup);
|
||||
EXPORT_SYMBOL(spa_strfree);
|
||||
EXPORT_SYMBOL(spa_get_random);
|
||||
EXPORT_SYMBOL(spa_generate_guid);
|
||||
EXPORT_SYMBOL(snprintf_blkptr);
|
||||
EXPORT_SYMBOL(spa_freeze);
|
||||
|
@ -726,7 +726,7 @@ space_map_write_impl(space_map_t *sm, range_tree_t *rt, maptype_t maptype,
|
||||
length > SM_RUN_MAX ||
|
||||
vdev_id != SM_NO_VDEVID ||
|
||||
(zfs_force_some_double_word_sm_entries &&
|
||||
spa_get_random(100) == 0)))
|
||||
random_in_range(100) == 0)))
|
||||
words = 2;
|
||||
|
||||
space_map_write_seg(sm, rs_get_start(rs, rt), rs_get_end(rs,
|
||||
|
@ -1572,7 +1572,7 @@ vdev_indirect_splits_enumerate_randomly(indirect_vsd_t *iv, zio_t *zio)
|
||||
indirect_child_t *ic = list_head(&is->is_unique_child);
|
||||
int children = is->is_unique_children;
|
||||
|
||||
for (int i = spa_get_random(children); i > 0; i--)
|
||||
for (int i = random_in_range(children); i > 0; i--)
|
||||
ic = list_next(&is->is_unique_child, ic);
|
||||
|
||||
ASSERT3P(ic, !=, NULL);
|
||||
@ -1736,7 +1736,7 @@ vdev_indirect_reconstruct_io_done(zio_t *zio)
|
||||
* Known_good will be TRUE when reconstruction is known to be possible.
|
||||
*/
|
||||
if (zfs_reconstruct_indirect_damage_fraction != 0 &&
|
||||
spa_get_random(zfs_reconstruct_indirect_damage_fraction) == 0)
|
||||
random_in_range(zfs_reconstruct_indirect_damage_fraction) == 0)
|
||||
known_good = (vdev_indirect_splits_damage(iv, zio) == 0);
|
||||
|
||||
/*
|
||||
|
@ -496,7 +496,7 @@ vdev_mirror_preferred_child_randomize(zio_t *zio)
|
||||
int p;
|
||||
|
||||
if (mm->mm_root) {
|
||||
p = spa_get_random(mm->mm_preferred_cnt);
|
||||
p = random_in_range(mm->mm_preferred_cnt);
|
||||
return (vdev_mirror_dva_select(zio, p));
|
||||
}
|
||||
|
||||
|
@ -205,8 +205,10 @@ zil_init_log_chain(zilog_t *zilog, blkptr_t *bp)
|
||||
{
|
||||
zio_cksum_t *zc = &bp->blk_cksum;
|
||||
|
||||
zc->zc_word[ZIL_ZC_GUID_0] = spa_get_random(-1ULL);
|
||||
zc->zc_word[ZIL_ZC_GUID_1] = spa_get_random(-1ULL);
|
||||
(void) random_get_pseudo_bytes((void *)&zc->zc_word[ZIL_ZC_GUID_0],
|
||||
sizeof (zc->zc_word[ZIL_ZC_GUID_0]));
|
||||
(void) random_get_pseudo_bytes((void *)&zc->zc_word[ZIL_ZC_GUID_1],
|
||||
sizeof (zc->zc_word[ZIL_ZC_GUID_1]));
|
||||
zc->zc_word[ZIL_ZC_OBJSET] = dmu_objset_id(zilog->zl_os);
|
||||
zc->zc_word[ZIL_ZC_SEQ] = 1ULL;
|
||||
}
|
||||
|
@ -201,7 +201,7 @@ zio_decompress_data(enum zio_compress c, abd_t *src, void *dst,
|
||||
* in non-ECC RAM), we handle this error (and test it).
|
||||
*/
|
||||
if (zio_decompress_fail_fraction != 0 &&
|
||||
spa_get_random(zio_decompress_fail_fraction) == 0)
|
||||
random_in_range(zio_decompress_fail_fraction) == 0)
|
||||
ret = SET_ERROR(EINVAL);
|
||||
|
||||
return (ret);
|
||||
|
@ -117,7 +117,7 @@ freq_triggered(uint32_t frequency)
|
||||
*/
|
||||
uint32_t maximum = (frequency <= 100) ? 100 : ZI_PERCENTAGE_MAX;
|
||||
|
||||
return (spa_get_random(maximum) < frequency);
|
||||
return (random_in_range(maximum) < frequency);
|
||||
}
|
||||
|
||||
/*
|
||||
@ -347,12 +347,12 @@ zio_inject_bitflip_cb(void *data, size_t len, void *private)
|
||||
{
|
||||
zio_t *zio __maybe_unused = private;
|
||||
uint8_t *buffer = data;
|
||||
uint_t byte = spa_get_random(len);
|
||||
uint_t byte = random_in_range(len);
|
||||
|
||||
ASSERT(zio->io_type == ZIO_TYPE_READ);
|
||||
|
||||
/* flip a single random bit in an abd data buffer */
|
||||
buffer[byte] ^= 1 << spa_get_random(8);
|
||||
buffer[byte] ^= 1 << random_in_range(8);
|
||||
|
||||
return (1); /* stop after first flip */
|
||||
}
|
||||
@ -493,7 +493,7 @@ zio_handle_ignored_writes(zio_t *zio)
|
||||
}
|
||||
|
||||
/* Have a "problem" writing 60% of the time */
|
||||
if (spa_get_random(100) < 60)
|
||||
if (random_in_range(100) < 60)
|
||||
zio->io_pipeline &= ~ZIO_VDEV_IO_STAGES;
|
||||
break;
|
||||
}
|
||||
|
Loading…
Reference in New Issue
Block a user