freebsd-nq/module/zfs/zap_micro.c

1680 lines
41 KiB
C
Raw Normal View History

2008-11-20 20:01:55 +00:00
/*
* CDDL HEADER START
*
* The contents of this file are subject to the terms of the
* Common Development and Distribution License (the "License").
* You may not use this file except in compliance with the License.
*
* You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
* or http://www.opensolaris.org/os/licensing.
* See the License for the specific language governing permissions
* and limitations under the License.
*
* When distributing Covered Code, include this CDDL HEADER in each
* file and include the License file at usr/src/OPENSOLARIS.LICENSE.
* If applicable, add the following below this CDDL HEADER, with the
* fields enclosed by brackets "[]" replaced with your own identifying
* information: Portions Copyright [yyyy] [name of copyright owner]
*
* CDDL HEADER END
*/
2008-11-20 20:01:55 +00:00
/*
* Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
* Copyright (c) 2011, 2016 by Delphix. All rights reserved.
* Copyright (c) 2014 Spectra Logic Corporation, All rights reserved.
* Copyright 2017 Nexenta Systems, Inc.
2008-11-20 20:01:55 +00:00
*/
#include <sys/zio.h>
2008-11-20 20:01:55 +00:00
#include <sys/spa.h>
#include <sys/dmu.h>
#include <sys/zfs_context.h>
#include <sys/zap.h>
#include <sys/refcount.h>
#include <sys/zap_impl.h>
#include <sys/zap_leaf.h>
#include <sys/avl.h>
#include <sys/arc.h>
Illumos 5027 - zfs large block support 5027 zfs large block support Reviewed by: Alek Pinchuk <pinchuk.alek@gmail.com> Reviewed by: George Wilson <george.wilson@delphix.com> Reviewed by: Josef 'Jeff' Sipek <josef.sipek@nexenta.com> Reviewed by: Richard Elling <richard.elling@richardelling.com> Reviewed by: Saso Kiselkov <skiselkov.ml@gmail.com> Reviewed by: Brian Behlendorf <behlendorf1@llnl.gov> Approved by: Dan McDonald <danmcd@omniti.com> References: https://www.illumos.org/issues/5027 https://github.com/illumos/illumos-gate/commit/b515258 Porting Notes: * Included in this patch is a tiny ISP2() cleanup in zio_init() from Illumos 5255. * Unlike the upstream Illumos commit this patch does not impose an arbitrary 128K block size limit on volumes. Volumes, like filesystems, are limited by the zfs_max_recordsize=1M module option. * By default the maximum record size is limited to 1M by the module option zfs_max_recordsize. This value may be safely increased up to 16M which is the largest block size supported by the on-disk format. At the moment, 1M blocks clearly offer a significant performance improvement but the benefits of going beyond this for the majority of workloads are less clear. * The illumos version of this patch increased DMU_MAX_ACCESS to 32M. This was determined not to be large enough when using 16M blocks because the zfs_make_xattrdir() function will fail (EFBIG) when assigning a TX. This was immediately observed under Linux because all newly created files must have a security xattr created and that was failing. Therefore, we've set DMU_MAX_ACCESS to 64M. * On 32-bit platforms a hard limit of 1M is set for blocks due to the limited virtual address space. We should be able to relax this one the ABD patches are merged. Ported-by: Brian Behlendorf <behlendorf1@llnl.gov> Closes #354
2014-11-03 20:15:08 +00:00
#include <sys/dmu_objset.h>
2008-11-20 20:01:55 +00:00
#ifdef _KERNEL
#include <sys/sunddi.h>
#endif
extern inline mzap_phys_t *zap_m_phys(zap_t *zap);
static int mzap_upgrade(zap_t **zapp,
void *tag, dmu_tx_t *tx, zap_flags_t flags);
2008-11-20 20:01:55 +00:00
uint64_t
zap_getflags(zap_t *zap)
{
if (zap->zap_ismicro)
return (0);
return (zap_f_phys(zap)->zap_flags);
}
2008-11-20 20:01:55 +00:00
int
zap_hashbits(zap_t *zap)
2008-11-20 20:01:55 +00:00
{
if (zap_getflags(zap) & ZAP_FLAG_HASH64)
return (48);
else
return (28);
}
2008-11-20 20:01:55 +00:00
uint32_t
zap_maxcd(zap_t *zap)
{
if (zap_getflags(zap) & ZAP_FLAG_HASH64)
return ((1<<16)-1);
else
return (-1U);
}
2008-11-20 20:01:55 +00:00
static uint64_t
zap_hash(zap_name_t *zn)
{
zap_t *zap = zn->zn_zap;
uint64_t h = 0;
2008-11-20 20:01:55 +00:00
if (zap_getflags(zap) & ZAP_FLAG_PRE_HASHED_KEY) {
ASSERT(zap_getflags(zap) & ZAP_FLAG_UINT64_KEY);
h = *(uint64_t *)zn->zn_key_orig;
} else {
h = zap->zap_salt;
ASSERT(h != 0);
ASSERT(zfs_crc64_table[128] == ZFS_CRC64_POLY);
if (zap_getflags(zap) & ZAP_FLAG_UINT64_KEY) {
int i;
const uint64_t *wp = zn->zn_key_norm;
ASSERT(zn->zn_key_intlen == 8);
for (i = 0; i < zn->zn_key_norm_numints; wp++, i++) {
int j;
uint64_t word = *wp;
for (j = 0; j < zn->zn_key_intlen; j++) {
h = (h >> 8) ^
zfs_crc64_table[(h ^ word) & 0xFF];
word >>= NBBY;
}
}
} else {
int i, len;
const uint8_t *cp = zn->zn_key_norm;
/*
* We previously stored the terminating null on
* disk, but didn't hash it, so we need to
* continue to not hash it. (The
* zn_key_*_numints includes the terminating
* null for non-binary keys.)
*/
len = zn->zn_key_norm_numints - 1;
ASSERT(zn->zn_key_intlen == 1);
for (i = 0; i < len; cp++, i++) {
h = (h >> 8) ^
zfs_crc64_table[(h ^ *cp) & 0xFF];
}
}
}
2008-11-20 20:01:55 +00:00
/*
* Don't use all 64 bits, since we need some in the cookie for
* the collision differentiator. We MUST use the high bits,
* since those are the ones that we first pay attention to when
* choosing the bucket.
2008-11-20 20:01:55 +00:00
*/
h &= ~((1ULL << (64 - zap_hashbits(zap))) - 1);
2008-11-20 20:01:55 +00:00
return (h);
2008-11-20 20:01:55 +00:00
}
static int
zap_normalize(zap_t *zap, const char *name, char *namenorm, int normflags)
2008-11-20 20:01:55 +00:00
{
size_t inlen, outlen;
int err;
ASSERT(!(zap_getflags(zap) & ZAP_FLAG_UINT64_KEY));
2008-11-20 20:01:55 +00:00
inlen = strlen(name) + 1;
outlen = ZAP_MAXNAMELEN;
err = 0;
(void) u8_textprep_str((char *)name, &inlen, namenorm, &outlen,
normflags | U8_TEXTPREP_IGNORE_NULL | U8_TEXTPREP_IGNORE_INVALID,
U8_UNICODE_LATEST, &err);
2008-11-20 20:01:55 +00:00
return (err);
}
boolean_t
zap_match(zap_name_t *zn, const char *matchname)
{
ASSERT(!(zap_getflags(zn->zn_zap) & ZAP_FLAG_UINT64_KEY));
if (zn->zn_matchtype & MT_NORMALIZE) {
2008-11-20 20:01:55 +00:00
char norm[ZAP_MAXNAMELEN];
if (zap_normalize(zn->zn_zap, matchname, norm,
zn->zn_normflags) != 0)
2008-11-20 20:01:55 +00:00
return (B_FALSE);
return (strcmp(zn->zn_key_norm, norm) == 0);
2008-11-20 20:01:55 +00:00
} else {
return (strcmp(zn->zn_key_orig, matchname) == 0);
2008-11-20 20:01:55 +00:00
}
}
void
zap_name_free(zap_name_t *zn)
{
kmem_free(zn, sizeof (zap_name_t));
}
zap_name_t *
zap_name_alloc(zap_t *zap, const char *key, matchtype_t mt)
2008-11-20 20:01:55 +00:00
{
zap_name_t *zn = kmem_alloc(sizeof (zap_name_t), KM_SLEEP);
2008-11-20 20:01:55 +00:00
zn->zn_zap = zap;
zn->zn_key_intlen = sizeof (*key);
zn->zn_key_orig = key;
zn->zn_key_orig_numints = strlen(zn->zn_key_orig) + 1;
2008-11-20 20:01:55 +00:00
zn->zn_matchtype = mt;
zn->zn_normflags = zap->zap_normflags;
/*
* If we're dealing with a case sensitive lookup on a mixed or
* insensitive fs, remove U8_TEXTPREP_TOUPPER or the lookup
* will fold case to all caps overriding the lookup request.
*/
if (mt & MT_MATCH_CASE)
zn->zn_normflags &= ~U8_TEXTPREP_TOUPPER;
2008-11-20 20:01:55 +00:00
if (zap->zap_normflags) {
/*
* We *must* use zap_normflags because this normalization is
* what the hash is computed from.
*/
if (zap_normalize(zap, key, zn->zn_normbuf,
zap->zap_normflags) != 0) {
2008-11-20 20:01:55 +00:00
zap_name_free(zn);
return (NULL);
}
zn->zn_key_norm = zn->zn_normbuf;
zn->zn_key_norm_numints = strlen(zn->zn_key_norm) + 1;
2008-11-20 20:01:55 +00:00
} else {
if (mt != 0) {
2008-11-20 20:01:55 +00:00
zap_name_free(zn);
return (NULL);
}
zn->zn_key_norm = zn->zn_key_orig;
zn->zn_key_norm_numints = zn->zn_key_orig_numints;
2008-11-20 20:01:55 +00:00
}
zn->zn_hash = zap_hash(zn);
if (zap->zap_normflags != zn->zn_normflags) {
/*
* We *must* use zn_normflags because this normalization is
* what the matching is based on. (Not the hash!)
*/
if (zap_normalize(zap, key, zn->zn_normbuf,
zn->zn_normflags) != 0) {
zap_name_free(zn);
return (NULL);
}
zn->zn_key_norm_numints = strlen(zn->zn_key_norm) + 1;
}
return (zn);
}
zap_name_t *
zap_name_alloc_uint64(zap_t *zap, const uint64_t *key, int numints)
{
zap_name_t *zn = kmem_alloc(sizeof (zap_name_t), KM_SLEEP);
ASSERT(zap->zap_normflags == 0);
zn->zn_zap = zap;
zn->zn_key_intlen = sizeof (*key);
zn->zn_key_orig = zn->zn_key_norm = key;
zn->zn_key_orig_numints = zn->zn_key_norm_numints = numints;
zn->zn_matchtype = 0;
zn->zn_hash = zap_hash(zn);
2008-11-20 20:01:55 +00:00
return (zn);
}
static void
mzap_byteswap(mzap_phys_t *buf, size_t size)
{
int i, max;
buf->mz_block_type = BSWAP_64(buf->mz_block_type);
buf->mz_salt = BSWAP_64(buf->mz_salt);
buf->mz_normflags = BSWAP_64(buf->mz_normflags);
max = (size / MZAP_ENT_LEN) - 1;
for (i = 0; i < max; i++) {
buf->mz_chunk[i].mze_value =
BSWAP_64(buf->mz_chunk[i].mze_value);
buf->mz_chunk[i].mze_cd =
BSWAP_32(buf->mz_chunk[i].mze_cd);
}
}
void
zap_byteswap(void *buf, size_t size)
{
uint64_t block_type;
block_type = *(uint64_t *)buf;
if (block_type == ZBT_MICRO || block_type == BSWAP_64(ZBT_MICRO)) {
/* ASSERT(magic == ZAP_LEAF_MAGIC); */
mzap_byteswap(buf, size);
} else {
fzap_byteswap(buf, size);
}
}
static int
mze_compare(const void *arg1, const void *arg2)
{
const mzap_ent_t *mze1 = arg1;
const mzap_ent_t *mze2 = arg2;
Performance optimization of AVL tree comparator functions perf: 2.75x faster ddt_entry_compare() First 256bits of ddt_key_t is a block checksum, which are expected to be close to random data. Hence, on average, comparison only needs to look at first few bytes of the keys. To reduce number of conditional jump instructions, the result is computed as: sign(memcmp(k1, k2)). Sign of an integer 'a' can be obtained as: `(0 < a) - (a < 0)` := {-1, 0, 1} , which is computed efficiently. Synthetic performance evaluation of original and new algorithm over 1G random keys on 2.6GHz Intel(R) Xeon(R) CPU E5-2660 v3: old 6.85789 s new 2.49089 s perf: 2.8x faster vdev_queue_offset_compare() and vdev_queue_timestamp_compare() Compute the result directly instead of using conditionals perf: zfs_range_compare() Speedup between 1.1x - 2.5x, depending on compiler version and optimization level. perf: spa_error_entry_compare() `bcmp()` is not suitable for comparator use. Use `memcmp()` instead. perf: 2.8x faster metaslab_compare() and metaslab_rangesize_compare() perf: 2.8x faster zil_bp_compare() perf: 2.8x faster mze_compare() perf: faster dbuf_compare() perf: faster compares in spa_misc perf: 2.8x faster layout_hash_compare() perf: 2.8x faster space_reftree_compare() perf: libzfs: faster avl tree comparators perf: guid_compare() perf: dsl_deadlist_compare() perf: perm_set_compare() perf: 2x faster range_tree_seg_compare() perf: faster unique_compare() perf: faster vdev_cache _compare() perf: faster vdev_uberblock_compare() perf: faster fuid _compare() perf: faster zfs_znode_hold_compare() Signed-off-by: Gvozden Neskovic <neskovic@gmail.com> Signed-off-by: Richard Elling <richard.elling@gmail.com> Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov> Closes #5033
2016-08-27 18:12:53 +00:00
int cmp = AVL_CMP(mze1->mze_hash, mze2->mze_hash);
if (likely(cmp))
return (cmp);
return (AVL_CMP(mze1->mze_cd, mze2->mze_cd));
2008-11-20 20:01:55 +00:00
}
static void
mze_insert(zap_t *zap, int chunkid, uint64_t hash)
2008-11-20 20:01:55 +00:00
{
mzap_ent_t *mze;
ASSERT(zap->zap_ismicro);
ASSERT(RW_WRITE_HELD(&zap->zap_rwlock));
mze = kmem_alloc(sizeof (mzap_ent_t), KM_SLEEP);
2008-11-20 20:01:55 +00:00
mze->mze_chunkid = chunkid;
mze->mze_hash = hash;
mze->mze_cd = MZE_PHYS(zap, mze)->mze_cd;
ASSERT(MZE_PHYS(zap, mze)->mze_name[0] != 0);
2008-11-20 20:01:55 +00:00
avl_add(&zap->zap_m.zap_avl, mze);
}
static mzap_ent_t *
mze_find(zap_name_t *zn)
{
mzap_ent_t mze_tofind;
mzap_ent_t *mze;
avl_index_t idx;
avl_tree_t *avl = &zn->zn_zap->zap_m.zap_avl;
ASSERT(zn->zn_zap->zap_ismicro);
ASSERT(RW_LOCK_HELD(&zn->zn_zap->zap_rwlock));
mze_tofind.mze_hash = zn->zn_hash;
mze_tofind.mze_cd = 0;
2008-11-20 20:01:55 +00:00
mze = avl_find(avl, &mze_tofind, &idx);
if (mze == NULL)
mze = avl_nearest(avl, idx, AVL_AFTER);
for (; mze && mze->mze_hash == zn->zn_hash; mze = AVL_NEXT(avl, mze)) {
ASSERT3U(mze->mze_cd, ==, MZE_PHYS(zn->zn_zap, mze)->mze_cd);
if (zap_match(zn, MZE_PHYS(zn->zn_zap, mze)->mze_name))
2008-11-20 20:01:55 +00:00
return (mze);
}
2008-11-20 20:01:55 +00:00
return (NULL);
}
static uint32_t
mze_find_unused_cd(zap_t *zap, uint64_t hash)
{
mzap_ent_t mze_tofind;
mzap_ent_t *mze;
avl_index_t idx;
avl_tree_t *avl = &zap->zap_m.zap_avl;
uint32_t cd;
ASSERT(zap->zap_ismicro);
ASSERT(RW_LOCK_HELD(&zap->zap_rwlock));
mze_tofind.mze_hash = hash;
mze_tofind.mze_cd = 0;
2008-11-20 20:01:55 +00:00
cd = 0;
for (mze = avl_find(avl, &mze_tofind, &idx);
mze && mze->mze_hash == hash; mze = AVL_NEXT(avl, mze)) {
if (mze->mze_cd != cd)
2008-11-20 20:01:55 +00:00
break;
cd++;
}
return (cd);
}
Fix ENOSPC in "Handle zap_add() failures in ..." Commit cc63068 caused ENOSPC error when copy a large amount of files between two directories. The reason is that the patch limits zap leaf expansion to 2 retries, and return ENOSPC when failed. The intent for limiting retries is to prevent pointlessly growing table to max size when adding a block full of entries with same name in different case in mixed mode. However, it turns out we cannot use any limit on the retry. When we copy files from one directory in readdir order, we are copying in hash order, one leaf block at a time. Which means that if the leaf block in source directory has expanded 6 times, and you copy those entries in that block, by the time you need to expand the leaf in destination directory, you need to expand it 6 times in one go. So any limit on the retry will result in error where it shouldn't. Note that while we do use different salt for different directories, it seems that the salt/hash function doesn't provide enough randomization to the hash distance to prevent this from happening. Since cc63068 has already been reverted. This patch adds it back and removes the retry limit. Also, as it turn out, failing on zap_add() has a serious side effect for mzap_upgrade(). When upgrading from micro zap to fat zap, it will call zap_add() to transfer entries one at a time. If it hit any error halfway through, the remaining entries will be lost, causing those files to become orphan. This patch add a VERIFY to catch it. Reviewed-by: Sanjeev Bagewadi <sanjeev.bagewadi@gmail.com> Reviewed-by: Richard Yao <ryao@gentoo.org> Reviewed-by: Tony Hutter <hutter2@llnl.gov> Reviewed-by: Albert Lee <trisk@forkgnu.org> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Reviewed by: Matthew Ahrens <mahrens@delphix.com> Signed-off-by: Chunwei Chen <david.chen@nutanix.com> Closes #7401 Closes #7421
2018-04-18 21:19:50 +00:00
/*
* Each mzap entry requires at max : 4 chunks
* 3 chunks for names + 1 chunk for value.
*/
#define MZAP_ENT_CHUNKS (1 + ZAP_LEAF_ARRAY_NCHUNKS(MZAP_NAME_LEN) + \
ZAP_LEAF_ARRAY_NCHUNKS(sizeof (uint64_t)))
/*
* Check if the current entry keeps the colliding entries under the fatzap leaf
* size.
*/
static boolean_t
mze_canfit_fzap_leaf(zap_name_t *zn, uint64_t hash)
{
zap_t *zap = zn->zn_zap;
mzap_ent_t mze_tofind;
mzap_ent_t *mze;
avl_index_t idx;
avl_tree_t *avl = &zap->zap_m.zap_avl;
uint32_t mzap_ents = 0;
mze_tofind.mze_hash = hash;
mze_tofind.mze_cd = 0;
for (mze = avl_find(avl, &mze_tofind, &idx);
mze && mze->mze_hash == hash; mze = AVL_NEXT(avl, mze)) {
mzap_ents++;
}
/* Include the new entry being added */
mzap_ents++;
return (ZAP_LEAF_NUMCHUNKS_DEF > (mzap_ents * MZAP_ENT_CHUNKS));
}
2008-11-20 20:01:55 +00:00
static void
mze_remove(zap_t *zap, mzap_ent_t *mze)
{
ASSERT(zap->zap_ismicro);
ASSERT(RW_WRITE_HELD(&zap->zap_rwlock));
avl_remove(&zap->zap_m.zap_avl, mze);
kmem_free(mze, sizeof (mzap_ent_t));
}
static void
mze_destroy(zap_t *zap)
{
mzap_ent_t *mze;
void *avlcookie = NULL;
while ((mze = avl_destroy_nodes(&zap->zap_m.zap_avl, &avlcookie)))
2008-11-20 20:01:55 +00:00
kmem_free(mze, sizeof (mzap_ent_t));
avl_destroy(&zap->zap_m.zap_avl);
}
static zap_t *
mzap_open(objset_t *os, uint64_t obj, dmu_buf_t *db)
{
zap_t *winner;
zap_t *zap;
int i;
uint64_t *zap_hdr = (uint64_t *)db->db_data;
uint64_t zap_block_type = zap_hdr[0];
uint64_t zap_magic = zap_hdr[1];
2008-11-20 20:01:55 +00:00
ASSERT3U(MZAP_ENT_LEN, ==, sizeof (mzap_ent_phys_t));
zap = kmem_zalloc(sizeof (zap_t), KM_SLEEP);
rw_init(&zap->zap_rwlock, NULL, RW_DEFAULT, NULL);
2008-11-20 20:01:55 +00:00
rw_enter(&zap->zap_rwlock, RW_WRITER);
zap->zap_objset = os;
zap->zap_object = obj;
zap->zap_dbuf = db;
if (zap_block_type != ZBT_MICRO) {
mutex_init(&zap->zap_f.zap_num_entries_mtx, 0, MUTEX_DEFAULT,
0);
zap->zap_f.zap_block_shift = highbit64(db->db_size) - 1;
if (zap_block_type != ZBT_HEADER || zap_magic != ZAP_MAGIC) {
winner = NULL; /* No actual winner here... */
goto handle_winner;
}
2008-11-20 20:01:55 +00:00
} else {
zap->zap_ismicro = TRUE;
}
/*
* Make sure that zap_ismicro is set before we let others see
* it, because zap_lockdir() checks zap_ismicro without the lock
* held.
*/
dmu_buf_init_user(&zap->zap_dbu, zap_evict_sync, NULL, &zap->zap_dbuf);
winner = dmu_buf_set_user(db, &zap->zap_dbu);
2008-11-20 20:01:55 +00:00
if (winner != NULL)
goto handle_winner;
2008-11-20 20:01:55 +00:00
if (zap->zap_ismicro) {
zap->zap_salt = zap_m_phys(zap)->mz_salt;
zap->zap_normflags = zap_m_phys(zap)->mz_normflags;
2008-11-20 20:01:55 +00:00
zap->zap_m.zap_num_chunks = db->db_size / MZAP_ENT_LEN - 1;
avl_create(&zap->zap_m.zap_avl, mze_compare,
sizeof (mzap_ent_t), offsetof(mzap_ent_t, mze_node));
for (i = 0; i < zap->zap_m.zap_num_chunks; i++) {
mzap_ent_phys_t *mze =
&zap_m_phys(zap)->mz_chunk[i];
2008-11-20 20:01:55 +00:00
if (mze->mze_name[0]) {
zap_name_t *zn;
zap->zap_m.zap_num_entries++;
zn = zap_name_alloc(zap, mze->mze_name, 0);
mze_insert(zap, i, zn->zn_hash);
2008-11-20 20:01:55 +00:00
zap_name_free(zn);
}
}
} else {
zap->zap_salt = zap_f_phys(zap)->zap_salt;
zap->zap_normflags = zap_f_phys(zap)->zap_normflags;
2008-11-20 20:01:55 +00:00
ASSERT3U(sizeof (struct zap_leaf_header), ==,
2*ZAP_LEAF_CHUNKSIZE);
/*
* The embedded pointer table should not overlap the
* other members.
*/
ASSERT3P(&ZAP_EMBEDDED_PTRTBL_ENT(zap, 0), >,
&zap_f_phys(zap)->zap_salt);
2008-11-20 20:01:55 +00:00
/*
* The embedded pointer table should end at the end of
* the block
*/
ASSERT3U((uintptr_t)&ZAP_EMBEDDED_PTRTBL_ENT(zap,
1<<ZAP_EMBEDDED_PTRTBL_SHIFT(zap)) -
(uintptr_t)zap_f_phys(zap), ==,
2008-11-20 20:01:55 +00:00
zap->zap_dbuf->db_size);
}
rw_exit(&zap->zap_rwlock);
return (zap);
handle_winner:
rw_exit(&zap->zap_rwlock);
rw_destroy(&zap->zap_rwlock);
if (!zap->zap_ismicro)
mutex_destroy(&zap->zap_f.zap_num_entries_mtx);
kmem_free(zap, sizeof (zap_t));
return (winner);
2008-11-20 20:01:55 +00:00
}
static int
zap_lockdir_impl(dmu_buf_t *db, void *tag, dmu_tx_t *tx,
2008-11-20 20:01:55 +00:00
krw_t lti, boolean_t fatreader, boolean_t adding, zap_t **zapp)
{
dmu_object_info_t doi;
2008-11-20 20:01:55 +00:00
zap_t *zap;
krw_t lt;
objset_t *os = dmu_buf_get_objset(db);
uint64_t obj = db->db_object;
2008-11-20 20:01:55 +00:00
ASSERT0(db->db_offset);
*zapp = NULL;
2008-11-20 20:01:55 +00:00
dmu_object_info_from_db(db, &doi);
if (DMU_OT_BYTESWAP(doi.doi_type) != DMU_BSWAP_ZAP)
return (SET_ERROR(EINVAL));
2008-11-20 20:01:55 +00:00
zap = dmu_buf_get_user(db);
if (zap == NULL) {
2008-11-20 20:01:55 +00:00
zap = mzap_open(os, obj, db);
if (zap == NULL) {
/*
* mzap_open() didn't like what it saw on-disk.
* Check for corruption!
*/
return (SET_ERROR(EIO));
}
}
2008-11-20 20:01:55 +00:00
/*
* We're checking zap_ismicro without the lock held, in order to
* tell what type of lock we want. Once we have some sort of
* lock, see if it really is the right type. In practice this
* can only be different if it was upgraded from micro to fat,
* and micro wanted WRITER but fat only needs READER.
*/
lt = (!zap->zap_ismicro && fatreader) ? RW_READER : lti;
rw_enter(&zap->zap_rwlock, lt);
if (lt != ((!zap->zap_ismicro && fatreader) ? RW_READER : lti)) {
/* it was upgraded, now we only need reader */
ASSERT(lt == RW_WRITER);
ASSERT(RW_READER ==
((!zap->zap_ismicro && fatreader) ? RW_READER : lti));
2008-11-20 20:01:55 +00:00
rw_downgrade(&zap->zap_rwlock);
lt = RW_READER;
}
zap->zap_objset = os;
if (lt == RW_WRITER)
dmu_buf_will_dirty(db, tx);
ASSERT3P(zap->zap_dbuf, ==, db);
ASSERT(!zap->zap_ismicro ||
zap->zap_m.zap_num_entries <= zap->zap_m.zap_num_chunks);
if (zap->zap_ismicro && tx && adding &&
zap->zap_m.zap_num_entries == zap->zap_m.zap_num_chunks) {
uint64_t newsz = db->db_size + SPA_MINBLOCKSIZE;
if (newsz > MZAP_MAX_BLKSZ) {
dprintf("upgrading obj %llu: num_entries=%u\n",
obj, zap->zap_m.zap_num_entries);
*zapp = zap;
int err = mzap_upgrade(zapp, tag, tx, 0);
if (err != 0)
rw_exit(&zap->zap_rwlock);
return (err);
2008-11-20 20:01:55 +00:00
}
VERIFY0(dmu_object_set_blocksize(os, obj, newsz, 0, tx));
2008-11-20 20:01:55 +00:00
zap->zap_m.zap_num_chunks =
db->db_size / MZAP_ENT_LEN - 1;
}
*zapp = zap;
return (0);
}
OpenZFS 7004 - dmu_tx_hold_zap() does dnode_hold() 7x on same object Using a benchmark which has 32 threads creating 2 million files in the same directory, on a machine with 16 CPU cores, I observed poor performance. I noticed that dmu_tx_hold_zap() was using about 30% of all CPU, and doing dnode_hold() 7 times on the same object (the ZAP object that is being held). dmu_tx_hold_zap() keeps a hold on the dnode_t the entire time it is running, in dmu_tx_hold_t:txh_dnode, so it would be nice to use the dnode_t that we already have in hand, rather than repeatedly calling dnode_hold(). To do this, we need to pass the dnode_t down through all the intermediate calls that dmu_tx_hold_zap() makes, making these routines take the dnode_t* rather than an objset_t* and a uint64_t object number. In particular, the following routines will need to have analogous *_by_dnode() variants created: dmu_buf_hold_noread() dmu_buf_hold() zap_lookup() zap_lookup_norm() zap_count_write() zap_lockdir() zap_count_write() This can improve performance on the benchmark described above by 100%, from 30,000 file creations per second to 60,000. (This improvement is on top of that provided by working around the object allocation issue. Peak performance of ~90,000 creations per second was observed with 8 CPUs; adding CPUs past that decreased performance due to lock contention.) The CPU used by dmu_tx_hold_zap() was reduced by 88%, from 340 CPU-seconds to 40 CPU-seconds. Sponsored by: Intel Corp. Signed-off-by: Matthew Ahrens <mahrens@delphix.com> Signed-off-by: Ned Bass <bass6@llnl.gov> Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov> OpenZFS-issue: https://www.illumos.org/issues/7004 OpenZFS-commit: https://github.com/openzfs/openzfs/pull/109 Closes #4641 Closes #4972
2016-07-20 22:42:13 +00:00
static int
zap_lockdir_by_dnode(dnode_t *dn, dmu_tx_t *tx,
krw_t lti, boolean_t fatreader, boolean_t adding, void *tag, zap_t **zapp)
{
dmu_buf_t *db;
int err;
err = dmu_buf_hold_by_dnode(dn, 0, tag, &db, DMU_READ_NO_PREFETCH);
if (err != 0) {
return (err);
}
err = zap_lockdir_impl(db, tag, tx, lti, fatreader, adding, zapp);
if (err != 0) {
dmu_buf_rele(db, tag);
}
return (err);
}
int
zap_lockdir(objset_t *os, uint64_t obj, dmu_tx_t *tx,
krw_t lti, boolean_t fatreader, boolean_t adding, void *tag, zap_t **zapp)
{
dmu_buf_t *db;
int err;
err = dmu_buf_hold(os, obj, 0, tag, &db, DMU_READ_NO_PREFETCH);
if (err != 0)
return (err);
err = zap_lockdir_impl(db, tag, tx, lti, fatreader, adding, zapp);
if (err != 0)
dmu_buf_rele(db, tag);
return (err);
}
2008-11-20 20:01:55 +00:00
void
zap_unlockdir(zap_t *zap, void *tag)
2008-11-20 20:01:55 +00:00
{
rw_exit(&zap->zap_rwlock);
dmu_buf_rele(zap->zap_dbuf, tag);
2008-11-20 20:01:55 +00:00
}
static int
mzap_upgrade(zap_t **zapp, void *tag, dmu_tx_t *tx, zap_flags_t flags)
2008-11-20 20:01:55 +00:00
{
mzap_phys_t *mzp;
int i, sz, nchunks;
int err = 0;
2008-11-20 20:01:55 +00:00
zap_t *zap = *zapp;
ASSERT(RW_WRITE_HELD(&zap->zap_rwlock));
sz = zap->zap_dbuf->db_size;
mzp = vmem_alloc(sz, KM_SLEEP);
2008-11-20 20:01:55 +00:00
bcopy(zap->zap_dbuf->db_data, mzp, sz);
nchunks = zap->zap_m.zap_num_chunks;
if (!flags) {
err = dmu_object_set_blocksize(zap->zap_objset, zap->zap_object,
1ULL << fzap_default_block_shift, 0, tx);
if (err) {
vmem_free(mzp, sz);
return (err);
}
2008-11-20 20:01:55 +00:00
}
dprintf("upgrading obj=%llu with %u chunks\n",
zap->zap_object, nchunks);
/* XXX destroy the avl later, so we can use the stored hash value */
mze_destroy(zap);
fzap_upgrade(zap, tx, flags);
2008-11-20 20:01:55 +00:00
for (i = 0; i < nchunks; i++) {
mzap_ent_phys_t *mze = &mzp->mz_chunk[i];
zap_name_t *zn;
if (mze->mze_name[0] == 0)
continue;
dprintf("adding %s=%llu\n",
mze->mze_name, mze->mze_value);
zn = zap_name_alloc(zap, mze->mze_name, 0);
Fix ENOSPC in "Handle zap_add() failures in ..." Commit cc63068 caused ENOSPC error when copy a large amount of files between two directories. The reason is that the patch limits zap leaf expansion to 2 retries, and return ENOSPC when failed. The intent for limiting retries is to prevent pointlessly growing table to max size when adding a block full of entries with same name in different case in mixed mode. However, it turns out we cannot use any limit on the retry. When we copy files from one directory in readdir order, we are copying in hash order, one leaf block at a time. Which means that if the leaf block in source directory has expanded 6 times, and you copy those entries in that block, by the time you need to expand the leaf in destination directory, you need to expand it 6 times in one go. So any limit on the retry will result in error where it shouldn't. Note that while we do use different salt for different directories, it seems that the salt/hash function doesn't provide enough randomization to the hash distance to prevent this from happening. Since cc63068 has already been reverted. This patch adds it back and removes the retry limit. Also, as it turn out, failing on zap_add() has a serious side effect for mzap_upgrade(). When upgrading from micro zap to fat zap, it will call zap_add() to transfer entries one at a time. If it hit any error halfway through, the remaining entries will be lost, causing those files to become orphan. This patch add a VERIFY to catch it. Reviewed-by: Sanjeev Bagewadi <sanjeev.bagewadi@gmail.com> Reviewed-by: Richard Yao <ryao@gentoo.org> Reviewed-by: Tony Hutter <hutter2@llnl.gov> Reviewed-by: Albert Lee <trisk@forkgnu.org> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Reviewed by: Matthew Ahrens <mahrens@delphix.com> Signed-off-by: Chunwei Chen <david.chen@nutanix.com> Closes #7401 Closes #7421
2018-04-18 21:19:50 +00:00
/* If we fail here, we would end up losing entries */
VERIFY0(fzap_add_cd(zn, 8, 1, &mze->mze_value, mze->mze_cd,
tag, tx));
2008-11-20 20:01:55 +00:00
zap = zn->zn_zap; /* fzap_add_cd() may change zap */
zap_name_free(zn);
}
vmem_free(mzp, sz);
2008-11-20 20:01:55 +00:00
*zapp = zap;
Fix ENOSPC in "Handle zap_add() failures in ..." Commit cc63068 caused ENOSPC error when copy a large amount of files between two directories. The reason is that the patch limits zap leaf expansion to 2 retries, and return ENOSPC when failed. The intent for limiting retries is to prevent pointlessly growing table to max size when adding a block full of entries with same name in different case in mixed mode. However, it turns out we cannot use any limit on the retry. When we copy files from one directory in readdir order, we are copying in hash order, one leaf block at a time. Which means that if the leaf block in source directory has expanded 6 times, and you copy those entries in that block, by the time you need to expand the leaf in destination directory, you need to expand it 6 times in one go. So any limit on the retry will result in error where it shouldn't. Note that while we do use different salt for different directories, it seems that the salt/hash function doesn't provide enough randomization to the hash distance to prevent this from happening. Since cc63068 has already been reverted. This patch adds it back and removes the retry limit. Also, as it turn out, failing on zap_add() has a serious side effect for mzap_upgrade(). When upgrading from micro zap to fat zap, it will call zap_add() to transfer entries one at a time. If it hit any error halfway through, the remaining entries will be lost, causing those files to become orphan. This patch add a VERIFY to catch it. Reviewed-by: Sanjeev Bagewadi <sanjeev.bagewadi@gmail.com> Reviewed-by: Richard Yao <ryao@gentoo.org> Reviewed-by: Tony Hutter <hutter2@llnl.gov> Reviewed-by: Albert Lee <trisk@forkgnu.org> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Reviewed by: Matthew Ahrens <mahrens@delphix.com> Signed-off-by: Chunwei Chen <david.chen@nutanix.com> Closes #7401 Closes #7421
2018-04-18 21:19:50 +00:00
return (0);
2008-11-20 20:01:55 +00:00
}
/*
* The "normflags" determine the behavior of the matchtype_t which is
* passed to zap_lookup_norm(). Names which have the same normalized
* version will be stored with the same hash value, and therefore we can
* perform normalization-insensitive lookups. We can be Unicode form-
* insensitive and/or case-insensitive. The following flags are valid for
* "normflags":
*
* U8_TEXTPREP_NFC
* U8_TEXTPREP_NFD
* U8_TEXTPREP_NFKC
* U8_TEXTPREP_NFKD
* U8_TEXTPREP_TOUPPER
*
* The *_NF* (Normalization Form) flags are mutually exclusive; at most one
* of them may be supplied.
*/
void
mzap_create_impl(objset_t *os, uint64_t obj, int normflags, zap_flags_t flags,
dmu_tx_t *tx)
2008-11-20 20:01:55 +00:00
{
dmu_buf_t *db;
mzap_phys_t *zp;
VERIFY0(dmu_buf_hold(os, obj, 0, FTAG, &db, DMU_READ_NO_PREFETCH));
2008-11-20 20:01:55 +00:00
#ifdef ZFS_DEBUG
{
dmu_object_info_t doi;
dmu_object_info_from_db(db, &doi);
ASSERT3U(DMU_OT_BYTESWAP(doi.doi_type), ==, DMU_BSWAP_ZAP);
2008-11-20 20:01:55 +00:00
}
#endif
dmu_buf_will_dirty(db, tx);
zp = db->db_data;
zp->mz_block_type = ZBT_MICRO;
zp->mz_salt = ((uintptr_t)db ^ (uintptr_t)tx ^ (obj << 1)) | 1ULL;
zp->mz_normflags = normflags;
dmu_buf_rele(db, FTAG);
if (flags != 0) {
zap_t *zap;
/* Only fat zap supports flags; upgrade immediately. */
VERIFY(0 == zap_lockdir(os, obj, tx, RW_WRITER,
B_FALSE, B_FALSE, FTAG, &zap));
VERIFY3U(0, ==, mzap_upgrade(&zap, FTAG, tx, flags));
zap_unlockdir(zap, FTAG);
}
2008-11-20 20:01:55 +00:00
}
int
zap_create_claim(objset_t *os, uint64_t obj, dmu_object_type_t ot,
dmu_object_type_t bonustype, int bonuslen, dmu_tx_t *tx)
{
Implement large_dnode pool feature Justification ------------- This feature adds support for variable length dnodes. Our motivation is to eliminate the overhead associated with using spill blocks. Spill blocks are used to store system attribute data (i.e. file metadata) that does not fit in the dnode's bonus buffer. By allowing a larger bonus buffer area the use of a spill block can be avoided. Spill blocks potentially incur an additional read I/O for every dnode in a dnode block. As a worst case example, reading 32 dnodes from a 16k dnode block and all of the spill blocks could issue 33 separate reads. Now suppose those dnodes have size 1024 and therefore don't need spill blocks. Then the worst case number of blocks read is reduced to from 33 to two--one per dnode block. In practice spill blocks may tend to be co-located on disk with the dnode blocks so the reduction in I/O would not be this drastic. In a badly fragmented pool, however, the improvement could be significant. ZFS-on-Linux systems that make heavy use of extended attributes would benefit from this feature. In particular, ZFS-on-Linux supports the xattr=sa dataset property which allows file extended attribute data to be stored in the dnode bonus buffer as an alternative to the traditional directory-based format. Workloads such as SELinux and the Lustre distributed filesystem often store enough xattr data to force spill bocks when xattr=sa is in effect. Large dnodes may therefore provide a performance benefit to such systems. Other use cases that may benefit from this feature include files with large ACLs and symbolic links with long target names. Furthermore, this feature may be desirable on other platforms in case future applications or features are developed that could make use of a larger bonus buffer area. Implementation -------------- The size of a dnode may be a multiple of 512 bytes up to the size of a dnode block (currently 16384 bytes). A dn_extra_slots field was added to the current on-disk dnode_phys_t structure to describe the size of the physical dnode on disk. The 8 bits for this field were taken from the zero filled dn_pad2 field. The field represents how many "extra" dnode_phys_t slots a dnode consumes in its dnode block. This convention results in a value of 0 for 512 byte dnodes which preserves on-disk format compatibility with older software. Similarly, the in-memory dnode_t structure has a new dn_num_slots field to represent the total number of dnode_phys_t slots consumed on disk. Thus dn->dn_num_slots is 1 greater than the corresponding dnp->dn_extra_slots. This difference in convention was adopted because, unlike on-disk structures, backward compatibility is not a concern for in-memory objects, so we used a more natural way to represent size for a dnode_t. The default size for newly created dnodes is determined by the value of a new "dnodesize" dataset property. By default the property is set to "legacy" which is compatible with older software. Setting the property to "auto" will allow the filesystem to choose the most suitable dnode size. Currently this just sets the default dnode size to 1k, but future code improvements could dynamically choose a size based on observed workload patterns. Dnodes of varying sizes can coexist within the same dataset and even within the same dnode block. For example, to enable automatically-sized dnodes, run # zfs set dnodesize=auto tank/fish The user can also specify literal values for the dnodesize property. These are currently limited to powers of two from 1k to 16k. The power-of-2 limitation is only for simplicity of the user interface. Internally the implementation can handle any multiple of 512 up to 16k, and consumers of the DMU API can specify any legal dnode value. The size of a new dnode is determined at object allocation time and stored as a new field in the znode in-memory structure. New DMU interfaces are added to allow the consumer to specify the dnode size that a newly allocated object should use. Existing interfaces are unchanged to avoid having to update every call site and to preserve compatibility with external consumers such as Lustre. The new interfaces names are given below. The versions of these functions that don't take a dnodesize parameter now just call the _dnsize() versions with a dnodesize of 0, which means use the legacy dnode size. New DMU interfaces: dmu_object_alloc_dnsize() dmu_object_claim_dnsize() dmu_object_reclaim_dnsize() New ZAP interfaces: zap_create_dnsize() zap_create_norm_dnsize() zap_create_flags_dnsize() zap_create_claim_norm_dnsize() zap_create_link_dnsize() The constant DN_MAX_BONUSLEN is renamed to DN_OLD_MAX_BONUSLEN. The spa_maxdnodesize() function should be used to determine the maximum bonus length for a pool. These are a few noteworthy changes to key functions: * The prototype for dnode_hold_impl() now takes a "slots" parameter. When the DNODE_MUST_BE_FREE flag is set, this parameter is used to ensure the hole at the specified object offset is large enough to hold the dnode being created. The slots parameter is also used to ensure a dnode does not span multiple dnode blocks. In both of these cases, if a failure occurs, ENOSPC is returned. Keep in mind, these failure cases are only possible when using DNODE_MUST_BE_FREE. If the DNODE_MUST_BE_ALLOCATED flag is set, "slots" must be 0. dnode_hold_impl() will check if the requested dnode is already consumed as an extra dnode slot by an large dnode, in which case it returns ENOENT. * The function dmu_object_alloc() advances to the next dnode block if dnode_hold_impl() returns an error for a requested object. This is because the beginning of the next dnode block is the only location it can safely assume to either be a hole or a valid starting point for a dnode. * dnode_next_offset_level() and other functions that iterate through dnode blocks may no longer use a simple array indexing scheme. These now use the current dnode's dn_num_slots field to advance to the next dnode in the block. This is to ensure we properly skip the current dnode's bonus area and don't interpret it as a valid dnode. zdb --- The zdb command was updated to display a dnode's size under the "dnsize" column when the object is dumped. For ZIL create log records, zdb will now display the slot count for the object. ztest ----- Ztest chooses a random dnodesize for every newly created object. The random distribution is more heavily weighted toward small dnodes to better simulate real-world datasets. Unused bonus buffer space is filled with non-zero values computed from the object number, dataset id, offset, and generation number. This helps ensure that the dnode traversal code properly skips the interior regions of large dnodes, and that these interior regions are not overwritten by data belonging to other dnodes. A new test visits each object in a dataset. It verifies that the actual dnode size matches what was stored in the ztest block tag when it was created. It also verifies that the unused bonus buffer space is filled with the expected data patterns. ZFS Test Suite -------------- Added six new large dnode-specific tests, and integrated the dnodesize property into existing tests for zfs allow and send/recv. Send/Receive ------------ ZFS send streams for datasets containing large dnodes cannot be received on pools that don't support the large_dnode feature. A send stream with large dnodes sets a DMU_BACKUP_FEATURE_LARGE_DNODE flag which will be unrecognized by an incompatible receiving pool so that the zfs receive will fail gracefully. While not implemented here, it may be possible to generate a backward-compatible send stream from a dataset containing large dnodes. The implementation may be tricky, however, because the send object record for a large dnode would need to be resized to a 512 byte dnode, possibly kicking in a spill block in the process. This means we would need to construct a new SA layout and possibly register it in the SA layout object. The SA layout is normally just sent as an ordinary object record. But if we are constructing new layouts while generating the send stream we'd have to build the SA layout object dynamically and send it at the end of the stream. For sending and receiving between pools that do support large dnodes, the drr_object send record type is extended with a new field to store the dnode slot count. This field was repurposed from unused padding in the structure. ZIL Replay ---------- The dnode slot count is stored in the uppermost 8 bits of the lr_foid field. The bits were unused as the object id is currently capped at 48 bits. Resizing Dnodes --------------- It should be possible to resize a dnode when it is dirtied if the current dnodesize dataset property differs from the dnode's size, but this functionality is not currently implemented. Clearly a dnode can only grow if there are sufficient contiguous unused slots in the dnode block, but it should always be possible to shrink a dnode. Growing dnodes may be useful to reduce fragmentation in a pool with many spill blocks in use. Shrinking dnodes may be useful to allow sending a dataset to a pool that doesn't support the large_dnode feature. Feature Reference Counting -------------------------- The reference count for the large_dnode pool feature tracks the number of datasets that have ever contained a dnode of size larger than 512 bytes. The first time a large dnode is created in a dataset the dataset is converted to an extensible dataset. This is a one-way operation and the only way to decrement the feature count is to destroy the dataset, even if the dataset no longer contains any large dnodes. The complexity of reference counting on a per-dnode basis was too high, so we chose to track it on a per-dataset basis similarly to the large_block feature. Signed-off-by: Ned Bass <bass6@llnl.gov> Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov> Closes #3542
2016-03-17 01:25:34 +00:00
return (zap_create_claim_dnsize(os, obj, ot, bonustype, bonuslen,
0, tx));
}
int
zap_create_claim_dnsize(objset_t *os, uint64_t obj, dmu_object_type_t ot,
dmu_object_type_t bonustype, int bonuslen, int dnodesize, dmu_tx_t *tx)
{
return (zap_create_claim_norm_dnsize(os, obj,
0, ot, bonustype, bonuslen, dnodesize, tx));
2008-11-20 20:01:55 +00:00
}
int
zap_create_claim_norm(objset_t *os, uint64_t obj, int normflags,
dmu_object_type_t ot,
dmu_object_type_t bonustype, int bonuslen, dmu_tx_t *tx)
Implement large_dnode pool feature Justification ------------- This feature adds support for variable length dnodes. Our motivation is to eliminate the overhead associated with using spill blocks. Spill blocks are used to store system attribute data (i.e. file metadata) that does not fit in the dnode's bonus buffer. By allowing a larger bonus buffer area the use of a spill block can be avoided. Spill blocks potentially incur an additional read I/O for every dnode in a dnode block. As a worst case example, reading 32 dnodes from a 16k dnode block and all of the spill blocks could issue 33 separate reads. Now suppose those dnodes have size 1024 and therefore don't need spill blocks. Then the worst case number of blocks read is reduced to from 33 to two--one per dnode block. In practice spill blocks may tend to be co-located on disk with the dnode blocks so the reduction in I/O would not be this drastic. In a badly fragmented pool, however, the improvement could be significant. ZFS-on-Linux systems that make heavy use of extended attributes would benefit from this feature. In particular, ZFS-on-Linux supports the xattr=sa dataset property which allows file extended attribute data to be stored in the dnode bonus buffer as an alternative to the traditional directory-based format. Workloads such as SELinux and the Lustre distributed filesystem often store enough xattr data to force spill bocks when xattr=sa is in effect. Large dnodes may therefore provide a performance benefit to such systems. Other use cases that may benefit from this feature include files with large ACLs and symbolic links with long target names. Furthermore, this feature may be desirable on other platforms in case future applications or features are developed that could make use of a larger bonus buffer area. Implementation -------------- The size of a dnode may be a multiple of 512 bytes up to the size of a dnode block (currently 16384 bytes). A dn_extra_slots field was added to the current on-disk dnode_phys_t structure to describe the size of the physical dnode on disk. The 8 bits for this field were taken from the zero filled dn_pad2 field. The field represents how many "extra" dnode_phys_t slots a dnode consumes in its dnode block. This convention results in a value of 0 for 512 byte dnodes which preserves on-disk format compatibility with older software. Similarly, the in-memory dnode_t structure has a new dn_num_slots field to represent the total number of dnode_phys_t slots consumed on disk. Thus dn->dn_num_slots is 1 greater than the corresponding dnp->dn_extra_slots. This difference in convention was adopted because, unlike on-disk structures, backward compatibility is not a concern for in-memory objects, so we used a more natural way to represent size for a dnode_t. The default size for newly created dnodes is determined by the value of a new "dnodesize" dataset property. By default the property is set to "legacy" which is compatible with older software. Setting the property to "auto" will allow the filesystem to choose the most suitable dnode size. Currently this just sets the default dnode size to 1k, but future code improvements could dynamically choose a size based on observed workload patterns. Dnodes of varying sizes can coexist within the same dataset and even within the same dnode block. For example, to enable automatically-sized dnodes, run # zfs set dnodesize=auto tank/fish The user can also specify literal values for the dnodesize property. These are currently limited to powers of two from 1k to 16k. The power-of-2 limitation is only for simplicity of the user interface. Internally the implementation can handle any multiple of 512 up to 16k, and consumers of the DMU API can specify any legal dnode value. The size of a new dnode is determined at object allocation time and stored as a new field in the znode in-memory structure. New DMU interfaces are added to allow the consumer to specify the dnode size that a newly allocated object should use. Existing interfaces are unchanged to avoid having to update every call site and to preserve compatibility with external consumers such as Lustre. The new interfaces names are given below. The versions of these functions that don't take a dnodesize parameter now just call the _dnsize() versions with a dnodesize of 0, which means use the legacy dnode size. New DMU interfaces: dmu_object_alloc_dnsize() dmu_object_claim_dnsize() dmu_object_reclaim_dnsize() New ZAP interfaces: zap_create_dnsize() zap_create_norm_dnsize() zap_create_flags_dnsize() zap_create_claim_norm_dnsize() zap_create_link_dnsize() The constant DN_MAX_BONUSLEN is renamed to DN_OLD_MAX_BONUSLEN. The spa_maxdnodesize() function should be used to determine the maximum bonus length for a pool. These are a few noteworthy changes to key functions: * The prototype for dnode_hold_impl() now takes a "slots" parameter. When the DNODE_MUST_BE_FREE flag is set, this parameter is used to ensure the hole at the specified object offset is large enough to hold the dnode being created. The slots parameter is also used to ensure a dnode does not span multiple dnode blocks. In both of these cases, if a failure occurs, ENOSPC is returned. Keep in mind, these failure cases are only possible when using DNODE_MUST_BE_FREE. If the DNODE_MUST_BE_ALLOCATED flag is set, "slots" must be 0. dnode_hold_impl() will check if the requested dnode is already consumed as an extra dnode slot by an large dnode, in which case it returns ENOENT. * The function dmu_object_alloc() advances to the next dnode block if dnode_hold_impl() returns an error for a requested object. This is because the beginning of the next dnode block is the only location it can safely assume to either be a hole or a valid starting point for a dnode. * dnode_next_offset_level() and other functions that iterate through dnode blocks may no longer use a simple array indexing scheme. These now use the current dnode's dn_num_slots field to advance to the next dnode in the block. This is to ensure we properly skip the current dnode's bonus area and don't interpret it as a valid dnode. zdb --- The zdb command was updated to display a dnode's size under the "dnsize" column when the object is dumped. For ZIL create log records, zdb will now display the slot count for the object. ztest ----- Ztest chooses a random dnodesize for every newly created object. The random distribution is more heavily weighted toward small dnodes to better simulate real-world datasets. Unused bonus buffer space is filled with non-zero values computed from the object number, dataset id, offset, and generation number. This helps ensure that the dnode traversal code properly skips the interior regions of large dnodes, and that these interior regions are not overwritten by data belonging to other dnodes. A new test visits each object in a dataset. It verifies that the actual dnode size matches what was stored in the ztest block tag when it was created. It also verifies that the unused bonus buffer space is filled with the expected data patterns. ZFS Test Suite -------------- Added six new large dnode-specific tests, and integrated the dnodesize property into existing tests for zfs allow and send/recv. Send/Receive ------------ ZFS send streams for datasets containing large dnodes cannot be received on pools that don't support the large_dnode feature. A send stream with large dnodes sets a DMU_BACKUP_FEATURE_LARGE_DNODE flag which will be unrecognized by an incompatible receiving pool so that the zfs receive will fail gracefully. While not implemented here, it may be possible to generate a backward-compatible send stream from a dataset containing large dnodes. The implementation may be tricky, however, because the send object record for a large dnode would need to be resized to a 512 byte dnode, possibly kicking in a spill block in the process. This means we would need to construct a new SA layout and possibly register it in the SA layout object. The SA layout is normally just sent as an ordinary object record. But if we are constructing new layouts while generating the send stream we'd have to build the SA layout object dynamically and send it at the end of the stream. For sending and receiving between pools that do support large dnodes, the drr_object send record type is extended with a new field to store the dnode slot count. This field was repurposed from unused padding in the structure. ZIL Replay ---------- The dnode slot count is stored in the uppermost 8 bits of the lr_foid field. The bits were unused as the object id is currently capped at 48 bits. Resizing Dnodes --------------- It should be possible to resize a dnode when it is dirtied if the current dnodesize dataset property differs from the dnode's size, but this functionality is not currently implemented. Clearly a dnode can only grow if there are sufficient contiguous unused slots in the dnode block, but it should always be possible to shrink a dnode. Growing dnodes may be useful to reduce fragmentation in a pool with many spill blocks in use. Shrinking dnodes may be useful to allow sending a dataset to a pool that doesn't support the large_dnode feature. Feature Reference Counting -------------------------- The reference count for the large_dnode pool feature tracks the number of datasets that have ever contained a dnode of size larger than 512 bytes. The first time a large dnode is created in a dataset the dataset is converted to an extensible dataset. This is a one-way operation and the only way to decrement the feature count is to destroy the dataset, even if the dataset no longer contains any large dnodes. The complexity of reference counting on a per-dnode basis was too high, so we chose to track it on a per-dataset basis similarly to the large_block feature. Signed-off-by: Ned Bass <bass6@llnl.gov> Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov> Closes #3542
2016-03-17 01:25:34 +00:00
{
return (zap_create_claim_norm_dnsize(os, obj, normflags, ot, bonustype,
bonuslen, 0, tx));
}
int
zap_create_claim_norm_dnsize(objset_t *os, uint64_t obj, int normflags,
dmu_object_type_t ot, dmu_object_type_t bonustype, int bonuslen,
int dnodesize, dmu_tx_t *tx)
2008-11-20 20:01:55 +00:00
{
int err;
Implement large_dnode pool feature Justification ------------- This feature adds support for variable length dnodes. Our motivation is to eliminate the overhead associated with using spill blocks. Spill blocks are used to store system attribute data (i.e. file metadata) that does not fit in the dnode's bonus buffer. By allowing a larger bonus buffer area the use of a spill block can be avoided. Spill blocks potentially incur an additional read I/O for every dnode in a dnode block. As a worst case example, reading 32 dnodes from a 16k dnode block and all of the spill blocks could issue 33 separate reads. Now suppose those dnodes have size 1024 and therefore don't need spill blocks. Then the worst case number of blocks read is reduced to from 33 to two--one per dnode block. In practice spill blocks may tend to be co-located on disk with the dnode blocks so the reduction in I/O would not be this drastic. In a badly fragmented pool, however, the improvement could be significant. ZFS-on-Linux systems that make heavy use of extended attributes would benefit from this feature. In particular, ZFS-on-Linux supports the xattr=sa dataset property which allows file extended attribute data to be stored in the dnode bonus buffer as an alternative to the traditional directory-based format. Workloads such as SELinux and the Lustre distributed filesystem often store enough xattr data to force spill bocks when xattr=sa is in effect. Large dnodes may therefore provide a performance benefit to such systems. Other use cases that may benefit from this feature include files with large ACLs and symbolic links with long target names. Furthermore, this feature may be desirable on other platforms in case future applications or features are developed that could make use of a larger bonus buffer area. Implementation -------------- The size of a dnode may be a multiple of 512 bytes up to the size of a dnode block (currently 16384 bytes). A dn_extra_slots field was added to the current on-disk dnode_phys_t structure to describe the size of the physical dnode on disk. The 8 bits for this field were taken from the zero filled dn_pad2 field. The field represents how many "extra" dnode_phys_t slots a dnode consumes in its dnode block. This convention results in a value of 0 for 512 byte dnodes which preserves on-disk format compatibility with older software. Similarly, the in-memory dnode_t structure has a new dn_num_slots field to represent the total number of dnode_phys_t slots consumed on disk. Thus dn->dn_num_slots is 1 greater than the corresponding dnp->dn_extra_slots. This difference in convention was adopted because, unlike on-disk structures, backward compatibility is not a concern for in-memory objects, so we used a more natural way to represent size for a dnode_t. The default size for newly created dnodes is determined by the value of a new "dnodesize" dataset property. By default the property is set to "legacy" which is compatible with older software. Setting the property to "auto" will allow the filesystem to choose the most suitable dnode size. Currently this just sets the default dnode size to 1k, but future code improvements could dynamically choose a size based on observed workload patterns. Dnodes of varying sizes can coexist within the same dataset and even within the same dnode block. For example, to enable automatically-sized dnodes, run # zfs set dnodesize=auto tank/fish The user can also specify literal values for the dnodesize property. These are currently limited to powers of two from 1k to 16k. The power-of-2 limitation is only for simplicity of the user interface. Internally the implementation can handle any multiple of 512 up to 16k, and consumers of the DMU API can specify any legal dnode value. The size of a new dnode is determined at object allocation time and stored as a new field in the znode in-memory structure. New DMU interfaces are added to allow the consumer to specify the dnode size that a newly allocated object should use. Existing interfaces are unchanged to avoid having to update every call site and to preserve compatibility with external consumers such as Lustre. The new interfaces names are given below. The versions of these functions that don't take a dnodesize parameter now just call the _dnsize() versions with a dnodesize of 0, which means use the legacy dnode size. New DMU interfaces: dmu_object_alloc_dnsize() dmu_object_claim_dnsize() dmu_object_reclaim_dnsize() New ZAP interfaces: zap_create_dnsize() zap_create_norm_dnsize() zap_create_flags_dnsize() zap_create_claim_norm_dnsize() zap_create_link_dnsize() The constant DN_MAX_BONUSLEN is renamed to DN_OLD_MAX_BONUSLEN. The spa_maxdnodesize() function should be used to determine the maximum bonus length for a pool. These are a few noteworthy changes to key functions: * The prototype for dnode_hold_impl() now takes a "slots" parameter. When the DNODE_MUST_BE_FREE flag is set, this parameter is used to ensure the hole at the specified object offset is large enough to hold the dnode being created. The slots parameter is also used to ensure a dnode does not span multiple dnode blocks. In both of these cases, if a failure occurs, ENOSPC is returned. Keep in mind, these failure cases are only possible when using DNODE_MUST_BE_FREE. If the DNODE_MUST_BE_ALLOCATED flag is set, "slots" must be 0. dnode_hold_impl() will check if the requested dnode is already consumed as an extra dnode slot by an large dnode, in which case it returns ENOENT. * The function dmu_object_alloc() advances to the next dnode block if dnode_hold_impl() returns an error for a requested object. This is because the beginning of the next dnode block is the only location it can safely assume to either be a hole or a valid starting point for a dnode. * dnode_next_offset_level() and other functions that iterate through dnode blocks may no longer use a simple array indexing scheme. These now use the current dnode's dn_num_slots field to advance to the next dnode in the block. This is to ensure we properly skip the current dnode's bonus area and don't interpret it as a valid dnode. zdb --- The zdb command was updated to display a dnode's size under the "dnsize" column when the object is dumped. For ZIL create log records, zdb will now display the slot count for the object. ztest ----- Ztest chooses a random dnodesize for every newly created object. The random distribution is more heavily weighted toward small dnodes to better simulate real-world datasets. Unused bonus buffer space is filled with non-zero values computed from the object number, dataset id, offset, and generation number. This helps ensure that the dnode traversal code properly skips the interior regions of large dnodes, and that these interior regions are not overwritten by data belonging to other dnodes. A new test visits each object in a dataset. It verifies that the actual dnode size matches what was stored in the ztest block tag when it was created. It also verifies that the unused bonus buffer space is filled with the expected data patterns. ZFS Test Suite -------------- Added six new large dnode-specific tests, and integrated the dnodesize property into existing tests for zfs allow and send/recv. Send/Receive ------------ ZFS send streams for datasets containing large dnodes cannot be received on pools that don't support the large_dnode feature. A send stream with large dnodes sets a DMU_BACKUP_FEATURE_LARGE_DNODE flag which will be unrecognized by an incompatible receiving pool so that the zfs receive will fail gracefully. While not implemented here, it may be possible to generate a backward-compatible send stream from a dataset containing large dnodes. The implementation may be tricky, however, because the send object record for a large dnode would need to be resized to a 512 byte dnode, possibly kicking in a spill block in the process. This means we would need to construct a new SA layout and possibly register it in the SA layout object. The SA layout is normally just sent as an ordinary object record. But if we are constructing new layouts while generating the send stream we'd have to build the SA layout object dynamically and send it at the end of the stream. For sending and receiving between pools that do support large dnodes, the drr_object send record type is extended with a new field to store the dnode slot count. This field was repurposed from unused padding in the structure. ZIL Replay ---------- The dnode slot count is stored in the uppermost 8 bits of the lr_foid field. The bits were unused as the object id is currently capped at 48 bits. Resizing Dnodes --------------- It should be possible to resize a dnode when it is dirtied if the current dnodesize dataset property differs from the dnode's size, but this functionality is not currently implemented. Clearly a dnode can only grow if there are sufficient contiguous unused slots in the dnode block, but it should always be possible to shrink a dnode. Growing dnodes may be useful to reduce fragmentation in a pool with many spill blocks in use. Shrinking dnodes may be useful to allow sending a dataset to a pool that doesn't support the large_dnode feature. Feature Reference Counting -------------------------- The reference count for the large_dnode pool feature tracks the number of datasets that have ever contained a dnode of size larger than 512 bytes. The first time a large dnode is created in a dataset the dataset is converted to an extensible dataset. This is a one-way operation and the only way to decrement the feature count is to destroy the dataset, even if the dataset no longer contains any large dnodes. The complexity of reference counting on a per-dnode basis was too high, so we chose to track it on a per-dataset basis similarly to the large_block feature. Signed-off-by: Ned Bass <bass6@llnl.gov> Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov> Closes #3542
2016-03-17 01:25:34 +00:00
err = dmu_object_claim_dnsize(os, obj, ot, 0, bonustype, bonuslen,
dnodesize, tx);
2008-11-20 20:01:55 +00:00
if (err != 0)
return (err);
mzap_create_impl(os, obj, normflags, 0, tx);
2008-11-20 20:01:55 +00:00
return (0);
}
uint64_t
zap_create(objset_t *os, dmu_object_type_t ot,
dmu_object_type_t bonustype, int bonuslen, dmu_tx_t *tx)
{
return (zap_create_norm(os, 0, ot, bonustype, bonuslen, tx));
}
Implement large_dnode pool feature Justification ------------- This feature adds support for variable length dnodes. Our motivation is to eliminate the overhead associated with using spill blocks. Spill blocks are used to store system attribute data (i.e. file metadata) that does not fit in the dnode's bonus buffer. By allowing a larger bonus buffer area the use of a spill block can be avoided. Spill blocks potentially incur an additional read I/O for every dnode in a dnode block. As a worst case example, reading 32 dnodes from a 16k dnode block and all of the spill blocks could issue 33 separate reads. Now suppose those dnodes have size 1024 and therefore don't need spill blocks. Then the worst case number of blocks read is reduced to from 33 to two--one per dnode block. In practice spill blocks may tend to be co-located on disk with the dnode blocks so the reduction in I/O would not be this drastic. In a badly fragmented pool, however, the improvement could be significant. ZFS-on-Linux systems that make heavy use of extended attributes would benefit from this feature. In particular, ZFS-on-Linux supports the xattr=sa dataset property which allows file extended attribute data to be stored in the dnode bonus buffer as an alternative to the traditional directory-based format. Workloads such as SELinux and the Lustre distributed filesystem often store enough xattr data to force spill bocks when xattr=sa is in effect. Large dnodes may therefore provide a performance benefit to such systems. Other use cases that may benefit from this feature include files with large ACLs and symbolic links with long target names. Furthermore, this feature may be desirable on other platforms in case future applications or features are developed that could make use of a larger bonus buffer area. Implementation -------------- The size of a dnode may be a multiple of 512 bytes up to the size of a dnode block (currently 16384 bytes). A dn_extra_slots field was added to the current on-disk dnode_phys_t structure to describe the size of the physical dnode on disk. The 8 bits for this field were taken from the zero filled dn_pad2 field. The field represents how many "extra" dnode_phys_t slots a dnode consumes in its dnode block. This convention results in a value of 0 for 512 byte dnodes which preserves on-disk format compatibility with older software. Similarly, the in-memory dnode_t structure has a new dn_num_slots field to represent the total number of dnode_phys_t slots consumed on disk. Thus dn->dn_num_slots is 1 greater than the corresponding dnp->dn_extra_slots. This difference in convention was adopted because, unlike on-disk structures, backward compatibility is not a concern for in-memory objects, so we used a more natural way to represent size for a dnode_t. The default size for newly created dnodes is determined by the value of a new "dnodesize" dataset property. By default the property is set to "legacy" which is compatible with older software. Setting the property to "auto" will allow the filesystem to choose the most suitable dnode size. Currently this just sets the default dnode size to 1k, but future code improvements could dynamically choose a size based on observed workload patterns. Dnodes of varying sizes can coexist within the same dataset and even within the same dnode block. For example, to enable automatically-sized dnodes, run # zfs set dnodesize=auto tank/fish The user can also specify literal values for the dnodesize property. These are currently limited to powers of two from 1k to 16k. The power-of-2 limitation is only for simplicity of the user interface. Internally the implementation can handle any multiple of 512 up to 16k, and consumers of the DMU API can specify any legal dnode value. The size of a new dnode is determined at object allocation time and stored as a new field in the znode in-memory structure. New DMU interfaces are added to allow the consumer to specify the dnode size that a newly allocated object should use. Existing interfaces are unchanged to avoid having to update every call site and to preserve compatibility with external consumers such as Lustre. The new interfaces names are given below. The versions of these functions that don't take a dnodesize parameter now just call the _dnsize() versions with a dnodesize of 0, which means use the legacy dnode size. New DMU interfaces: dmu_object_alloc_dnsize() dmu_object_claim_dnsize() dmu_object_reclaim_dnsize() New ZAP interfaces: zap_create_dnsize() zap_create_norm_dnsize() zap_create_flags_dnsize() zap_create_claim_norm_dnsize() zap_create_link_dnsize() The constant DN_MAX_BONUSLEN is renamed to DN_OLD_MAX_BONUSLEN. The spa_maxdnodesize() function should be used to determine the maximum bonus length for a pool. These are a few noteworthy changes to key functions: * The prototype for dnode_hold_impl() now takes a "slots" parameter. When the DNODE_MUST_BE_FREE flag is set, this parameter is used to ensure the hole at the specified object offset is large enough to hold the dnode being created. The slots parameter is also used to ensure a dnode does not span multiple dnode blocks. In both of these cases, if a failure occurs, ENOSPC is returned. Keep in mind, these failure cases are only possible when using DNODE_MUST_BE_FREE. If the DNODE_MUST_BE_ALLOCATED flag is set, "slots" must be 0. dnode_hold_impl() will check if the requested dnode is already consumed as an extra dnode slot by an large dnode, in which case it returns ENOENT. * The function dmu_object_alloc() advances to the next dnode block if dnode_hold_impl() returns an error for a requested object. This is because the beginning of the next dnode block is the only location it can safely assume to either be a hole or a valid starting point for a dnode. * dnode_next_offset_level() and other functions that iterate through dnode blocks may no longer use a simple array indexing scheme. These now use the current dnode's dn_num_slots field to advance to the next dnode in the block. This is to ensure we properly skip the current dnode's bonus area and don't interpret it as a valid dnode. zdb --- The zdb command was updated to display a dnode's size under the "dnsize" column when the object is dumped. For ZIL create log records, zdb will now display the slot count for the object. ztest ----- Ztest chooses a random dnodesize for every newly created object. The random distribution is more heavily weighted toward small dnodes to better simulate real-world datasets. Unused bonus buffer space is filled with non-zero values computed from the object number, dataset id, offset, and generation number. This helps ensure that the dnode traversal code properly skips the interior regions of large dnodes, and that these interior regions are not overwritten by data belonging to other dnodes. A new test visits each object in a dataset. It verifies that the actual dnode size matches what was stored in the ztest block tag when it was created. It also verifies that the unused bonus buffer space is filled with the expected data patterns. ZFS Test Suite -------------- Added six new large dnode-specific tests, and integrated the dnodesize property into existing tests for zfs allow and send/recv. Send/Receive ------------ ZFS send streams for datasets containing large dnodes cannot be received on pools that don't support the large_dnode feature. A send stream with large dnodes sets a DMU_BACKUP_FEATURE_LARGE_DNODE flag which will be unrecognized by an incompatible receiving pool so that the zfs receive will fail gracefully. While not implemented here, it may be possible to generate a backward-compatible send stream from a dataset containing large dnodes. The implementation may be tricky, however, because the send object record for a large dnode would need to be resized to a 512 byte dnode, possibly kicking in a spill block in the process. This means we would need to construct a new SA layout and possibly register it in the SA layout object. The SA layout is normally just sent as an ordinary object record. But if we are constructing new layouts while generating the send stream we'd have to build the SA layout object dynamically and send it at the end of the stream. For sending and receiving between pools that do support large dnodes, the drr_object send record type is extended with a new field to store the dnode slot count. This field was repurposed from unused padding in the structure. ZIL Replay ---------- The dnode slot count is stored in the uppermost 8 bits of the lr_foid field. The bits were unused as the object id is currently capped at 48 bits. Resizing Dnodes --------------- It should be possible to resize a dnode when it is dirtied if the current dnodesize dataset property differs from the dnode's size, but this functionality is not currently implemented. Clearly a dnode can only grow if there are sufficient contiguous unused slots in the dnode block, but it should always be possible to shrink a dnode. Growing dnodes may be useful to reduce fragmentation in a pool with many spill blocks in use. Shrinking dnodes may be useful to allow sending a dataset to a pool that doesn't support the large_dnode feature. Feature Reference Counting -------------------------- The reference count for the large_dnode pool feature tracks the number of datasets that have ever contained a dnode of size larger than 512 bytes. The first time a large dnode is created in a dataset the dataset is converted to an extensible dataset. This is a one-way operation and the only way to decrement the feature count is to destroy the dataset, even if the dataset no longer contains any large dnodes. The complexity of reference counting on a per-dnode basis was too high, so we chose to track it on a per-dataset basis similarly to the large_block feature. Signed-off-by: Ned Bass <bass6@llnl.gov> Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov> Closes #3542
2016-03-17 01:25:34 +00:00
uint64_t
zap_create_dnsize(objset_t *os, dmu_object_type_t ot,
dmu_object_type_t bonustype, int bonuslen, int dnodesize, dmu_tx_t *tx)
{
return (zap_create_norm_dnsize(os, 0, ot, bonustype, bonuslen,
dnodesize, tx));
}
2008-11-20 20:01:55 +00:00
uint64_t
zap_create_norm(objset_t *os, int normflags, dmu_object_type_t ot,
dmu_object_type_t bonustype, int bonuslen, dmu_tx_t *tx)
{
Implement large_dnode pool feature Justification ------------- This feature adds support for variable length dnodes. Our motivation is to eliminate the overhead associated with using spill blocks. Spill blocks are used to store system attribute data (i.e. file metadata) that does not fit in the dnode's bonus buffer. By allowing a larger bonus buffer area the use of a spill block can be avoided. Spill blocks potentially incur an additional read I/O for every dnode in a dnode block. As a worst case example, reading 32 dnodes from a 16k dnode block and all of the spill blocks could issue 33 separate reads. Now suppose those dnodes have size 1024 and therefore don't need spill blocks. Then the worst case number of blocks read is reduced to from 33 to two--one per dnode block. In practice spill blocks may tend to be co-located on disk with the dnode blocks so the reduction in I/O would not be this drastic. In a badly fragmented pool, however, the improvement could be significant. ZFS-on-Linux systems that make heavy use of extended attributes would benefit from this feature. In particular, ZFS-on-Linux supports the xattr=sa dataset property which allows file extended attribute data to be stored in the dnode bonus buffer as an alternative to the traditional directory-based format. Workloads such as SELinux and the Lustre distributed filesystem often store enough xattr data to force spill bocks when xattr=sa is in effect. Large dnodes may therefore provide a performance benefit to such systems. Other use cases that may benefit from this feature include files with large ACLs and symbolic links with long target names. Furthermore, this feature may be desirable on other platforms in case future applications or features are developed that could make use of a larger bonus buffer area. Implementation -------------- The size of a dnode may be a multiple of 512 bytes up to the size of a dnode block (currently 16384 bytes). A dn_extra_slots field was added to the current on-disk dnode_phys_t structure to describe the size of the physical dnode on disk. The 8 bits for this field were taken from the zero filled dn_pad2 field. The field represents how many "extra" dnode_phys_t slots a dnode consumes in its dnode block. This convention results in a value of 0 for 512 byte dnodes which preserves on-disk format compatibility with older software. Similarly, the in-memory dnode_t structure has a new dn_num_slots field to represent the total number of dnode_phys_t slots consumed on disk. Thus dn->dn_num_slots is 1 greater than the corresponding dnp->dn_extra_slots. This difference in convention was adopted because, unlike on-disk structures, backward compatibility is not a concern for in-memory objects, so we used a more natural way to represent size for a dnode_t. The default size for newly created dnodes is determined by the value of a new "dnodesize" dataset property. By default the property is set to "legacy" which is compatible with older software. Setting the property to "auto" will allow the filesystem to choose the most suitable dnode size. Currently this just sets the default dnode size to 1k, but future code improvements could dynamically choose a size based on observed workload patterns. Dnodes of varying sizes can coexist within the same dataset and even within the same dnode block. For example, to enable automatically-sized dnodes, run # zfs set dnodesize=auto tank/fish The user can also specify literal values for the dnodesize property. These are currently limited to powers of two from 1k to 16k. The power-of-2 limitation is only for simplicity of the user interface. Internally the implementation can handle any multiple of 512 up to 16k, and consumers of the DMU API can specify any legal dnode value. The size of a new dnode is determined at object allocation time and stored as a new field in the znode in-memory structure. New DMU interfaces are added to allow the consumer to specify the dnode size that a newly allocated object should use. Existing interfaces are unchanged to avoid having to update every call site and to preserve compatibility with external consumers such as Lustre. The new interfaces names are given below. The versions of these functions that don't take a dnodesize parameter now just call the _dnsize() versions with a dnodesize of 0, which means use the legacy dnode size. New DMU interfaces: dmu_object_alloc_dnsize() dmu_object_claim_dnsize() dmu_object_reclaim_dnsize() New ZAP interfaces: zap_create_dnsize() zap_create_norm_dnsize() zap_create_flags_dnsize() zap_create_claim_norm_dnsize() zap_create_link_dnsize() The constant DN_MAX_BONUSLEN is renamed to DN_OLD_MAX_BONUSLEN. The spa_maxdnodesize() function should be used to determine the maximum bonus length for a pool. These are a few noteworthy changes to key functions: * The prototype for dnode_hold_impl() now takes a "slots" parameter. When the DNODE_MUST_BE_FREE flag is set, this parameter is used to ensure the hole at the specified object offset is large enough to hold the dnode being created. The slots parameter is also used to ensure a dnode does not span multiple dnode blocks. In both of these cases, if a failure occurs, ENOSPC is returned. Keep in mind, these failure cases are only possible when using DNODE_MUST_BE_FREE. If the DNODE_MUST_BE_ALLOCATED flag is set, "slots" must be 0. dnode_hold_impl() will check if the requested dnode is already consumed as an extra dnode slot by an large dnode, in which case it returns ENOENT. * The function dmu_object_alloc() advances to the next dnode block if dnode_hold_impl() returns an error for a requested object. This is because the beginning of the next dnode block is the only location it can safely assume to either be a hole or a valid starting point for a dnode. * dnode_next_offset_level() and other functions that iterate through dnode blocks may no longer use a simple array indexing scheme. These now use the current dnode's dn_num_slots field to advance to the next dnode in the block. This is to ensure we properly skip the current dnode's bonus area and don't interpret it as a valid dnode. zdb --- The zdb command was updated to display a dnode's size under the "dnsize" column when the object is dumped. For ZIL create log records, zdb will now display the slot count for the object. ztest ----- Ztest chooses a random dnodesize for every newly created object. The random distribution is more heavily weighted toward small dnodes to better simulate real-world datasets. Unused bonus buffer space is filled with non-zero values computed from the object number, dataset id, offset, and generation number. This helps ensure that the dnode traversal code properly skips the interior regions of large dnodes, and that these interior regions are not overwritten by data belonging to other dnodes. A new test visits each object in a dataset. It verifies that the actual dnode size matches what was stored in the ztest block tag when it was created. It also verifies that the unused bonus buffer space is filled with the expected data patterns. ZFS Test Suite -------------- Added six new large dnode-specific tests, and integrated the dnodesize property into existing tests for zfs allow and send/recv. Send/Receive ------------ ZFS send streams for datasets containing large dnodes cannot be received on pools that don't support the large_dnode feature. A send stream with large dnodes sets a DMU_BACKUP_FEATURE_LARGE_DNODE flag which will be unrecognized by an incompatible receiving pool so that the zfs receive will fail gracefully. While not implemented here, it may be possible to generate a backward-compatible send stream from a dataset containing large dnodes. The implementation may be tricky, however, because the send object record for a large dnode would need to be resized to a 512 byte dnode, possibly kicking in a spill block in the process. This means we would need to construct a new SA layout and possibly register it in the SA layout object. The SA layout is normally just sent as an ordinary object record. But if we are constructing new layouts while generating the send stream we'd have to build the SA layout object dynamically and send it at the end of the stream. For sending and receiving between pools that do support large dnodes, the drr_object send record type is extended with a new field to store the dnode slot count. This field was repurposed from unused padding in the structure. ZIL Replay ---------- The dnode slot count is stored in the uppermost 8 bits of the lr_foid field. The bits were unused as the object id is currently capped at 48 bits. Resizing Dnodes --------------- It should be possible to resize a dnode when it is dirtied if the current dnodesize dataset property differs from the dnode's size, but this functionality is not currently implemented. Clearly a dnode can only grow if there are sufficient contiguous unused slots in the dnode block, but it should always be possible to shrink a dnode. Growing dnodes may be useful to reduce fragmentation in a pool with many spill blocks in use. Shrinking dnodes may be useful to allow sending a dataset to a pool that doesn't support the large_dnode feature. Feature Reference Counting -------------------------- The reference count for the large_dnode pool feature tracks the number of datasets that have ever contained a dnode of size larger than 512 bytes. The first time a large dnode is created in a dataset the dataset is converted to an extensible dataset. This is a one-way operation and the only way to decrement the feature count is to destroy the dataset, even if the dataset no longer contains any large dnodes. The complexity of reference counting on a per-dnode basis was too high, so we chose to track it on a per-dataset basis similarly to the large_block feature. Signed-off-by: Ned Bass <bass6@llnl.gov> Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov> Closes #3542
2016-03-17 01:25:34 +00:00
return (zap_create_norm_dnsize(os, normflags, ot, bonustype, bonuslen,
0, tx));
}
uint64_t
zap_create_norm_dnsize(objset_t *os, int normflags, dmu_object_type_t ot,
dmu_object_type_t bonustype, int bonuslen, int dnodesize, dmu_tx_t *tx)
{
uint64_t obj = dmu_object_alloc_dnsize(os, ot, 0, bonustype, bonuslen,
dnodesize, tx);
2008-11-20 20:01:55 +00:00
mzap_create_impl(os, obj, normflags, 0, tx);
return (obj);
}
uint64_t
zap_create_flags(objset_t *os, int normflags, zap_flags_t flags,
dmu_object_type_t ot, int leaf_blockshift, int indirect_blockshift,
dmu_object_type_t bonustype, int bonuslen, dmu_tx_t *tx)
{
Implement large_dnode pool feature Justification ------------- This feature adds support for variable length dnodes. Our motivation is to eliminate the overhead associated with using spill blocks. Spill blocks are used to store system attribute data (i.e. file metadata) that does not fit in the dnode's bonus buffer. By allowing a larger bonus buffer area the use of a spill block can be avoided. Spill blocks potentially incur an additional read I/O for every dnode in a dnode block. As a worst case example, reading 32 dnodes from a 16k dnode block and all of the spill blocks could issue 33 separate reads. Now suppose those dnodes have size 1024 and therefore don't need spill blocks. Then the worst case number of blocks read is reduced to from 33 to two--one per dnode block. In practice spill blocks may tend to be co-located on disk with the dnode blocks so the reduction in I/O would not be this drastic. In a badly fragmented pool, however, the improvement could be significant. ZFS-on-Linux systems that make heavy use of extended attributes would benefit from this feature. In particular, ZFS-on-Linux supports the xattr=sa dataset property which allows file extended attribute data to be stored in the dnode bonus buffer as an alternative to the traditional directory-based format. Workloads such as SELinux and the Lustre distributed filesystem often store enough xattr data to force spill bocks when xattr=sa is in effect. Large dnodes may therefore provide a performance benefit to such systems. Other use cases that may benefit from this feature include files with large ACLs and symbolic links with long target names. Furthermore, this feature may be desirable on other platforms in case future applications or features are developed that could make use of a larger bonus buffer area. Implementation -------------- The size of a dnode may be a multiple of 512 bytes up to the size of a dnode block (currently 16384 bytes). A dn_extra_slots field was added to the current on-disk dnode_phys_t structure to describe the size of the physical dnode on disk. The 8 bits for this field were taken from the zero filled dn_pad2 field. The field represents how many "extra" dnode_phys_t slots a dnode consumes in its dnode block. This convention results in a value of 0 for 512 byte dnodes which preserves on-disk format compatibility with older software. Similarly, the in-memory dnode_t structure has a new dn_num_slots field to represent the total number of dnode_phys_t slots consumed on disk. Thus dn->dn_num_slots is 1 greater than the corresponding dnp->dn_extra_slots. This difference in convention was adopted because, unlike on-disk structures, backward compatibility is not a concern for in-memory objects, so we used a more natural way to represent size for a dnode_t. The default size for newly created dnodes is determined by the value of a new "dnodesize" dataset property. By default the property is set to "legacy" which is compatible with older software. Setting the property to "auto" will allow the filesystem to choose the most suitable dnode size. Currently this just sets the default dnode size to 1k, but future code improvements could dynamically choose a size based on observed workload patterns. Dnodes of varying sizes can coexist within the same dataset and even within the same dnode block. For example, to enable automatically-sized dnodes, run # zfs set dnodesize=auto tank/fish The user can also specify literal values for the dnodesize property. These are currently limited to powers of two from 1k to 16k. The power-of-2 limitation is only for simplicity of the user interface. Internally the implementation can handle any multiple of 512 up to 16k, and consumers of the DMU API can specify any legal dnode value. The size of a new dnode is determined at object allocation time and stored as a new field in the znode in-memory structure. New DMU interfaces are added to allow the consumer to specify the dnode size that a newly allocated object should use. Existing interfaces are unchanged to avoid having to update every call site and to preserve compatibility with external consumers such as Lustre. The new interfaces names are given below. The versions of these functions that don't take a dnodesize parameter now just call the _dnsize() versions with a dnodesize of 0, which means use the legacy dnode size. New DMU interfaces: dmu_object_alloc_dnsize() dmu_object_claim_dnsize() dmu_object_reclaim_dnsize() New ZAP interfaces: zap_create_dnsize() zap_create_norm_dnsize() zap_create_flags_dnsize() zap_create_claim_norm_dnsize() zap_create_link_dnsize() The constant DN_MAX_BONUSLEN is renamed to DN_OLD_MAX_BONUSLEN. The spa_maxdnodesize() function should be used to determine the maximum bonus length for a pool. These are a few noteworthy changes to key functions: * The prototype for dnode_hold_impl() now takes a "slots" parameter. When the DNODE_MUST_BE_FREE flag is set, this parameter is used to ensure the hole at the specified object offset is large enough to hold the dnode being created. The slots parameter is also used to ensure a dnode does not span multiple dnode blocks. In both of these cases, if a failure occurs, ENOSPC is returned. Keep in mind, these failure cases are only possible when using DNODE_MUST_BE_FREE. If the DNODE_MUST_BE_ALLOCATED flag is set, "slots" must be 0. dnode_hold_impl() will check if the requested dnode is already consumed as an extra dnode slot by an large dnode, in which case it returns ENOENT. * The function dmu_object_alloc() advances to the next dnode block if dnode_hold_impl() returns an error for a requested object. This is because the beginning of the next dnode block is the only location it can safely assume to either be a hole or a valid starting point for a dnode. * dnode_next_offset_level() and other functions that iterate through dnode blocks may no longer use a simple array indexing scheme. These now use the current dnode's dn_num_slots field to advance to the next dnode in the block. This is to ensure we properly skip the current dnode's bonus area and don't interpret it as a valid dnode. zdb --- The zdb command was updated to display a dnode's size under the "dnsize" column when the object is dumped. For ZIL create log records, zdb will now display the slot count for the object. ztest ----- Ztest chooses a random dnodesize for every newly created object. The random distribution is more heavily weighted toward small dnodes to better simulate real-world datasets. Unused bonus buffer space is filled with non-zero values computed from the object number, dataset id, offset, and generation number. This helps ensure that the dnode traversal code properly skips the interior regions of large dnodes, and that these interior regions are not overwritten by data belonging to other dnodes. A new test visits each object in a dataset. It verifies that the actual dnode size matches what was stored in the ztest block tag when it was created. It also verifies that the unused bonus buffer space is filled with the expected data patterns. ZFS Test Suite -------------- Added six new large dnode-specific tests, and integrated the dnodesize property into existing tests for zfs allow and send/recv. Send/Receive ------------ ZFS send streams for datasets containing large dnodes cannot be received on pools that don't support the large_dnode feature. A send stream with large dnodes sets a DMU_BACKUP_FEATURE_LARGE_DNODE flag which will be unrecognized by an incompatible receiving pool so that the zfs receive will fail gracefully. While not implemented here, it may be possible to generate a backward-compatible send stream from a dataset containing large dnodes. The implementation may be tricky, however, because the send object record for a large dnode would need to be resized to a 512 byte dnode, possibly kicking in a spill block in the process. This means we would need to construct a new SA layout and possibly register it in the SA layout object. The SA layout is normally just sent as an ordinary object record. But if we are constructing new layouts while generating the send stream we'd have to build the SA layout object dynamically and send it at the end of the stream. For sending and receiving between pools that do support large dnodes, the drr_object send record type is extended with a new field to store the dnode slot count. This field was repurposed from unused padding in the structure. ZIL Replay ---------- The dnode slot count is stored in the uppermost 8 bits of the lr_foid field. The bits were unused as the object id is currently capped at 48 bits. Resizing Dnodes --------------- It should be possible to resize a dnode when it is dirtied if the current dnodesize dataset property differs from the dnode's size, but this functionality is not currently implemented. Clearly a dnode can only grow if there are sufficient contiguous unused slots in the dnode block, but it should always be possible to shrink a dnode. Growing dnodes may be useful to reduce fragmentation in a pool with many spill blocks in use. Shrinking dnodes may be useful to allow sending a dataset to a pool that doesn't support the large_dnode feature. Feature Reference Counting -------------------------- The reference count for the large_dnode pool feature tracks the number of datasets that have ever contained a dnode of size larger than 512 bytes. The first time a large dnode is created in a dataset the dataset is converted to an extensible dataset. This is a one-way operation and the only way to decrement the feature count is to destroy the dataset, even if the dataset no longer contains any large dnodes. The complexity of reference counting on a per-dnode basis was too high, so we chose to track it on a per-dataset basis similarly to the large_block feature. Signed-off-by: Ned Bass <bass6@llnl.gov> Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov> Closes #3542
2016-03-17 01:25:34 +00:00
return (zap_create_flags_dnsize(os, normflags, flags, ot,
leaf_blockshift, indirect_blockshift, bonustype, bonuslen, 0, tx));
}
uint64_t
zap_create_flags_dnsize(objset_t *os, int normflags, zap_flags_t flags,
dmu_object_type_t ot, int leaf_blockshift, int indirect_blockshift,
dmu_object_type_t bonustype, int bonuslen, int dnodesize, dmu_tx_t *tx)
{
uint64_t obj = dmu_object_alloc_dnsize(os, ot, 0, bonustype, bonuslen,
dnodesize, tx);
ASSERT(leaf_blockshift >= SPA_MINBLOCKSHIFT &&
Illumos 5027 - zfs large block support 5027 zfs large block support Reviewed by: Alek Pinchuk <pinchuk.alek@gmail.com> Reviewed by: George Wilson <george.wilson@delphix.com> Reviewed by: Josef 'Jeff' Sipek <josef.sipek@nexenta.com> Reviewed by: Richard Elling <richard.elling@richardelling.com> Reviewed by: Saso Kiselkov <skiselkov.ml@gmail.com> Reviewed by: Brian Behlendorf <behlendorf1@llnl.gov> Approved by: Dan McDonald <danmcd@omniti.com> References: https://www.illumos.org/issues/5027 https://github.com/illumos/illumos-gate/commit/b515258 Porting Notes: * Included in this patch is a tiny ISP2() cleanup in zio_init() from Illumos 5255. * Unlike the upstream Illumos commit this patch does not impose an arbitrary 128K block size limit on volumes. Volumes, like filesystems, are limited by the zfs_max_recordsize=1M module option. * By default the maximum record size is limited to 1M by the module option zfs_max_recordsize. This value may be safely increased up to 16M which is the largest block size supported by the on-disk format. At the moment, 1M blocks clearly offer a significant performance improvement but the benefits of going beyond this for the majority of workloads are less clear. * The illumos version of this patch increased DMU_MAX_ACCESS to 32M. This was determined not to be large enough when using 16M blocks because the zfs_make_xattrdir() function will fail (EFBIG) when assigning a TX. This was immediately observed under Linux because all newly created files must have a security xattr created and that was failing. Therefore, we've set DMU_MAX_ACCESS to 64M. * On 32-bit platforms a hard limit of 1M is set for blocks due to the limited virtual address space. We should be able to relax this one the ABD patches are merged. Ported-by: Brian Behlendorf <behlendorf1@llnl.gov> Closes #354
2014-11-03 20:15:08 +00:00
leaf_blockshift <= SPA_OLD_MAXBLOCKSHIFT &&
indirect_blockshift >= SPA_MINBLOCKSHIFT &&
Illumos 5027 - zfs large block support 5027 zfs large block support Reviewed by: Alek Pinchuk <pinchuk.alek@gmail.com> Reviewed by: George Wilson <george.wilson@delphix.com> Reviewed by: Josef 'Jeff' Sipek <josef.sipek@nexenta.com> Reviewed by: Richard Elling <richard.elling@richardelling.com> Reviewed by: Saso Kiselkov <skiselkov.ml@gmail.com> Reviewed by: Brian Behlendorf <behlendorf1@llnl.gov> Approved by: Dan McDonald <danmcd@omniti.com> References: https://www.illumos.org/issues/5027 https://github.com/illumos/illumos-gate/commit/b515258 Porting Notes: * Included in this patch is a tiny ISP2() cleanup in zio_init() from Illumos 5255. * Unlike the upstream Illumos commit this patch does not impose an arbitrary 128K block size limit on volumes. Volumes, like filesystems, are limited by the zfs_max_recordsize=1M module option. * By default the maximum record size is limited to 1M by the module option zfs_max_recordsize. This value may be safely increased up to 16M which is the largest block size supported by the on-disk format. At the moment, 1M blocks clearly offer a significant performance improvement but the benefits of going beyond this for the majority of workloads are less clear. * The illumos version of this patch increased DMU_MAX_ACCESS to 32M. This was determined not to be large enough when using 16M blocks because the zfs_make_xattrdir() function will fail (EFBIG) when assigning a TX. This was immediately observed under Linux because all newly created files must have a security xattr created and that was failing. Therefore, we've set DMU_MAX_ACCESS to 64M. * On 32-bit platforms a hard limit of 1M is set for blocks due to the limited virtual address space. We should be able to relax this one the ABD patches are merged. Ported-by: Brian Behlendorf <behlendorf1@llnl.gov> Closes #354
2014-11-03 20:15:08 +00:00
indirect_blockshift <= SPA_OLD_MAXBLOCKSHIFT);
VERIFY(dmu_object_set_blocksize(os, obj,
1ULL << leaf_blockshift, indirect_blockshift, tx) == 0);
mzap_create_impl(os, obj, normflags, flags, tx);
2008-11-20 20:01:55 +00:00
return (obj);
}
int
zap_destroy(objset_t *os, uint64_t zapobj, dmu_tx_t *tx)
{
/*
* dmu_object_free will free the object number and free the
* data. Freeing the data will cause our pageout function to be
* called, which will destroy our data (zap_leaf_t's and zap_t).
*/
return (dmu_object_free(os, zapobj, tx));
}
void
zap_evict_sync(void *dbu)
2008-11-20 20:01:55 +00:00
{
zap_t *zap = dbu;
2008-11-20 20:01:55 +00:00
rw_destroy(&zap->zap_rwlock);
if (zap->zap_ismicro)
mze_destroy(zap);
else
mutex_destroy(&zap->zap_f.zap_num_entries_mtx);
kmem_free(zap, sizeof (zap_t));
}
int
zap_count(objset_t *os, uint64_t zapobj, uint64_t *count)
{
zap_t *zap;
int err;
err = zap_lockdir(os, zapobj, NULL, RW_READER, TRUE, FALSE, FTAG, &zap);
2008-11-20 20:01:55 +00:00
if (err)
return (err);
if (!zap->zap_ismicro) {
err = fzap_count(zap, count);
} else {
*count = zap->zap_m.zap_num_entries;
}
zap_unlockdir(zap, FTAG);
2008-11-20 20:01:55 +00:00
return (err);
}
/*
* zn may be NULL; if not specified, it will be computed if needed.
* See also the comment above zap_entry_normalization_conflict().
*/
static boolean_t
mzap_normalization_conflict(zap_t *zap, zap_name_t *zn, mzap_ent_t *mze)
{
mzap_ent_t *other;
int direction = AVL_BEFORE;
boolean_t allocdzn = B_FALSE;
if (zap->zap_normflags == 0)
return (B_FALSE);
again:
for (other = avl_walk(&zap->zap_m.zap_avl, mze, direction);
other && other->mze_hash == mze->mze_hash;
other = avl_walk(&zap->zap_m.zap_avl, other, direction)) {
if (zn == NULL) {
zn = zap_name_alloc(zap, MZE_PHYS(zap, mze)->mze_name,
MT_NORMALIZE);
2008-11-20 20:01:55 +00:00
allocdzn = B_TRUE;
}
if (zap_match(zn, MZE_PHYS(zap, other)->mze_name)) {
2008-11-20 20:01:55 +00:00
if (allocdzn)
zap_name_free(zn);
return (B_TRUE);
}
}
if (direction == AVL_BEFORE) {
direction = AVL_AFTER;
goto again;
}
if (allocdzn)
zap_name_free(zn);
return (B_FALSE);
}
/*
* Routines for manipulating attributes.
*/
int
zap_lookup(objset_t *os, uint64_t zapobj, const char *name,
uint64_t integer_size, uint64_t num_integers, void *buf)
{
return (zap_lookup_norm(os, zapobj, name, integer_size,
num_integers, buf, 0, NULL, 0, NULL));
2008-11-20 20:01:55 +00:00
}
static int
zap_lookup_impl(zap_t *zap, const char *name,
2008-11-20 20:01:55 +00:00
uint64_t integer_size, uint64_t num_integers, void *buf,
matchtype_t mt, char *realname, int rn_len,
boolean_t *ncp)
{
int err = 0;
2008-11-20 20:01:55 +00:00
mzap_ent_t *mze;
zap_name_t *zn;
zn = zap_name_alloc(zap, name, mt);
if (zn == NULL)
return (SET_ERROR(ENOTSUP));
2008-11-20 20:01:55 +00:00
if (!zap->zap_ismicro) {
err = fzap_lookup(zn, integer_size, num_integers, buf,
realname, rn_len, ncp);
} else {
mze = mze_find(zn);
if (mze == NULL) {
err = SET_ERROR(ENOENT);
2008-11-20 20:01:55 +00:00
} else {
if (num_integers < 1) {
err = SET_ERROR(EOVERFLOW);
2008-11-20 20:01:55 +00:00
} else if (integer_size != 8) {
err = SET_ERROR(EINVAL);
2008-11-20 20:01:55 +00:00
} else {
*(uint64_t *)buf =
MZE_PHYS(zap, mze)->mze_value;
2008-11-20 20:01:55 +00:00
(void) strlcpy(realname,
MZE_PHYS(zap, mze)->mze_name, rn_len);
2008-11-20 20:01:55 +00:00
if (ncp) {
*ncp = mzap_normalization_conflict(zap,
zn, mze);
}
}
}
}
zap_name_free(zn);
return (err);
}
int
zap_lookup_norm(objset_t *os, uint64_t zapobj, const char *name,
uint64_t integer_size, uint64_t num_integers, void *buf,
matchtype_t mt, char *realname, int rn_len,
boolean_t *ncp)
{
zap_t *zap;
int err;
err = zap_lockdir(os, zapobj, NULL, RW_READER, TRUE, FALSE, FTAG, &zap);
if (err != 0)
return (err);
err = zap_lookup_impl(zap, name, integer_size,
num_integers, buf, mt, realname, rn_len, ncp);
zap_unlockdir(zap, FTAG);
2008-11-20 20:01:55 +00:00
return (err);
}
int
zap_prefetch(objset_t *os, uint64_t zapobj, const char *name)
{
zap_t *zap;
int err;
zap_name_t *zn;
err = zap_lockdir(os, zapobj, NULL, RW_READER, TRUE, FALSE, FTAG, &zap);
if (err)
return (err);
zn = zap_name_alloc(zap, name, 0);
if (zn == NULL) {
zap_unlockdir(zap, FTAG);
return (SET_ERROR(ENOTSUP));
}
fzap_prefetch(zn);
zap_name_free(zn);
zap_unlockdir(zap, FTAG);
return (err);
}
OpenZFS 7004 - dmu_tx_hold_zap() does dnode_hold() 7x on same object Using a benchmark which has 32 threads creating 2 million files in the same directory, on a machine with 16 CPU cores, I observed poor performance. I noticed that dmu_tx_hold_zap() was using about 30% of all CPU, and doing dnode_hold() 7 times on the same object (the ZAP object that is being held). dmu_tx_hold_zap() keeps a hold on the dnode_t the entire time it is running, in dmu_tx_hold_t:txh_dnode, so it would be nice to use the dnode_t that we already have in hand, rather than repeatedly calling dnode_hold(). To do this, we need to pass the dnode_t down through all the intermediate calls that dmu_tx_hold_zap() makes, making these routines take the dnode_t* rather than an objset_t* and a uint64_t object number. In particular, the following routines will need to have analogous *_by_dnode() variants created: dmu_buf_hold_noread() dmu_buf_hold() zap_lookup() zap_lookup_norm() zap_count_write() zap_lockdir() zap_count_write() This can improve performance on the benchmark described above by 100%, from 30,000 file creations per second to 60,000. (This improvement is on top of that provided by working around the object allocation issue. Peak performance of ~90,000 creations per second was observed with 8 CPUs; adding CPUs past that decreased performance due to lock contention.) The CPU used by dmu_tx_hold_zap() was reduced by 88%, from 340 CPU-seconds to 40 CPU-seconds. Sponsored by: Intel Corp. Signed-off-by: Matthew Ahrens <mahrens@delphix.com> Signed-off-by: Ned Bass <bass6@llnl.gov> Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov> OpenZFS-issue: https://www.illumos.org/issues/7004 OpenZFS-commit: https://github.com/openzfs/openzfs/pull/109 Closes #4641 Closes #4972
2016-07-20 22:42:13 +00:00
int
zap_lookup_by_dnode(dnode_t *dn, const char *name,
uint64_t integer_size, uint64_t num_integers, void *buf)
{
return (zap_lookup_norm_by_dnode(dn, name, integer_size,
num_integers, buf, 0, NULL, 0, NULL));
OpenZFS 7004 - dmu_tx_hold_zap() does dnode_hold() 7x on same object Using a benchmark which has 32 threads creating 2 million files in the same directory, on a machine with 16 CPU cores, I observed poor performance. I noticed that dmu_tx_hold_zap() was using about 30% of all CPU, and doing dnode_hold() 7 times on the same object (the ZAP object that is being held). dmu_tx_hold_zap() keeps a hold on the dnode_t the entire time it is running, in dmu_tx_hold_t:txh_dnode, so it would be nice to use the dnode_t that we already have in hand, rather than repeatedly calling dnode_hold(). To do this, we need to pass the dnode_t down through all the intermediate calls that dmu_tx_hold_zap() makes, making these routines take the dnode_t* rather than an objset_t* and a uint64_t object number. In particular, the following routines will need to have analogous *_by_dnode() variants created: dmu_buf_hold_noread() dmu_buf_hold() zap_lookup() zap_lookup_norm() zap_count_write() zap_lockdir() zap_count_write() This can improve performance on the benchmark described above by 100%, from 30,000 file creations per second to 60,000. (This improvement is on top of that provided by working around the object allocation issue. Peak performance of ~90,000 creations per second was observed with 8 CPUs; adding CPUs past that decreased performance due to lock contention.) The CPU used by dmu_tx_hold_zap() was reduced by 88%, from 340 CPU-seconds to 40 CPU-seconds. Sponsored by: Intel Corp. Signed-off-by: Matthew Ahrens <mahrens@delphix.com> Signed-off-by: Ned Bass <bass6@llnl.gov> Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov> OpenZFS-issue: https://www.illumos.org/issues/7004 OpenZFS-commit: https://github.com/openzfs/openzfs/pull/109 Closes #4641 Closes #4972
2016-07-20 22:42:13 +00:00
}
int
zap_lookup_norm_by_dnode(dnode_t *dn, const char *name,
uint64_t integer_size, uint64_t num_integers, void *buf,
matchtype_t mt, char *realname, int rn_len,
boolean_t *ncp)
{
zap_t *zap;
int err;
err = zap_lockdir_by_dnode(dn, NULL, RW_READER, TRUE, FALSE,
FTAG, &zap);
if (err != 0)
return (err);
err = zap_lookup_impl(zap, name, integer_size,
num_integers, buf, mt, realname, rn_len, ncp);
zap_unlockdir(zap, FTAG);
return (err);
}
int
zap_prefetch_uint64(objset_t *os, uint64_t zapobj, const uint64_t *key,
int key_numints)
{
zap_t *zap;
int err;
zap_name_t *zn;
err = zap_lockdir(os, zapobj, NULL, RW_READER, TRUE, FALSE, FTAG, &zap);
if (err)
return (err);
zn = zap_name_alloc_uint64(zap, key, key_numints);
if (zn == NULL) {
zap_unlockdir(zap, FTAG);
return (SET_ERROR(ENOTSUP));
}
fzap_prefetch(zn);
zap_name_free(zn);
zap_unlockdir(zap, FTAG);
return (err);
}
int
zap_lookup_uint64(objset_t *os, uint64_t zapobj, const uint64_t *key,
int key_numints, uint64_t integer_size, uint64_t num_integers, void *buf)
{
zap_t *zap;
int err;
zap_name_t *zn;
err = zap_lockdir(os, zapobj, NULL, RW_READER, TRUE, FALSE, FTAG, &zap);
if (err)
return (err);
zn = zap_name_alloc_uint64(zap, key, key_numints);
if (zn == NULL) {
zap_unlockdir(zap, FTAG);
return (SET_ERROR(ENOTSUP));
}
err = fzap_lookup(zn, integer_size, num_integers, buf,
NULL, 0, NULL);
zap_name_free(zn);
zap_unlockdir(zap, FTAG);
return (err);
}
int
zap_contains(objset_t *os, uint64_t zapobj, const char *name)
{
int err = zap_lookup_norm(os, zapobj, name, 0,
0, NULL, 0, NULL, 0, NULL);
if (err == EOVERFLOW || err == EINVAL)
err = 0; /* found, but skipped reading the value */
return (err);
}
2008-11-20 20:01:55 +00:00
int
zap_length(objset_t *os, uint64_t zapobj, const char *name,
uint64_t *integer_size, uint64_t *num_integers)
{
zap_t *zap;
int err;
mzap_ent_t *mze;
zap_name_t *zn;
err = zap_lockdir(os, zapobj, NULL, RW_READER, TRUE, FALSE, FTAG, &zap);
2008-11-20 20:01:55 +00:00
if (err)
return (err);
zn = zap_name_alloc(zap, name, 0);
2008-11-20 20:01:55 +00:00
if (zn == NULL) {
zap_unlockdir(zap, FTAG);
return (SET_ERROR(ENOTSUP));
2008-11-20 20:01:55 +00:00
}
if (!zap->zap_ismicro) {
err = fzap_length(zn, integer_size, num_integers);
} else {
mze = mze_find(zn);
if (mze == NULL) {
err = SET_ERROR(ENOENT);
2008-11-20 20:01:55 +00:00
} else {
if (integer_size)
*integer_size = 8;
if (num_integers)
*num_integers = 1;
}
}
zap_name_free(zn);
zap_unlockdir(zap, FTAG);
2008-11-20 20:01:55 +00:00
return (err);
}
int
zap_length_uint64(objset_t *os, uint64_t zapobj, const uint64_t *key,
int key_numints, uint64_t *integer_size, uint64_t *num_integers)
{
zap_t *zap;
int err;
zap_name_t *zn;
err = zap_lockdir(os, zapobj, NULL, RW_READER, TRUE, FALSE, FTAG, &zap);
if (err)
return (err);
zn = zap_name_alloc_uint64(zap, key, key_numints);
if (zn == NULL) {
zap_unlockdir(zap, FTAG);
return (SET_ERROR(ENOTSUP));
}
err = fzap_length(zn, integer_size, num_integers);
zap_name_free(zn);
zap_unlockdir(zap, FTAG);
return (err);
}
2008-11-20 20:01:55 +00:00
static void
mzap_addent(zap_name_t *zn, uint64_t value)
{
int i;
zap_t *zap = zn->zn_zap;
int start = zap->zap_m.zap_alloc_next;
uint32_t cd;
ASSERT(RW_WRITE_HELD(&zap->zap_rwlock));
#ifdef ZFS_DEBUG
for (i = 0; i < zap->zap_m.zap_num_chunks; i++) {
ASSERTV(mzap_ent_phys_t *mze);
ASSERT(mze = &zap_m_phys(zap)->mz_chunk[i]);
ASSERT(strcmp(zn->zn_key_orig, mze->mze_name) != 0);
2008-11-20 20:01:55 +00:00
}
#endif
cd = mze_find_unused_cd(zap, zn->zn_hash);
/* given the limited size of the microzap, this can't happen */
ASSERT(cd < zap_maxcd(zap));
2008-11-20 20:01:55 +00:00
again:
for (i = start; i < zap->zap_m.zap_num_chunks; i++) {
mzap_ent_phys_t *mze = &zap_m_phys(zap)->mz_chunk[i];
2008-11-20 20:01:55 +00:00
if (mze->mze_name[0] == 0) {
mze->mze_value = value;
mze->mze_cd = cd;
(void) strlcpy(mze->mze_name, zn->zn_key_orig,
sizeof (mze->mze_name));
2008-11-20 20:01:55 +00:00
zap->zap_m.zap_num_entries++;
zap->zap_m.zap_alloc_next = i+1;
if (zap->zap_m.zap_alloc_next ==
zap->zap_m.zap_num_chunks)
zap->zap_m.zap_alloc_next = 0;
mze_insert(zap, i, zn->zn_hash);
2008-11-20 20:01:55 +00:00
return;
}
}
if (start != 0) {
start = 0;
goto again;
}
cmn_err(CE_PANIC, "out of entries!");
2008-11-20 20:01:55 +00:00
}
static int
zap_add_impl(zap_t *zap, const char *key,
2008-11-20 20:01:55 +00:00
int integer_size, uint64_t num_integers,
const void *val, dmu_tx_t *tx, void *tag)
2008-11-20 20:01:55 +00:00
{
int err = 0;
2008-11-20 20:01:55 +00:00
mzap_ent_t *mze;
const uint64_t *intval = val;
zap_name_t *zn;
zn = zap_name_alloc(zap, key, 0);
2008-11-20 20:01:55 +00:00
if (zn == NULL) {
zap_unlockdir(zap, tag);
return (SET_ERROR(ENOTSUP));
2008-11-20 20:01:55 +00:00
}
if (!zap->zap_ismicro) {
err = fzap_add(zn, integer_size, num_integers, val, tag, tx);
2008-11-20 20:01:55 +00:00
zap = zn->zn_zap; /* fzap_add() may change zap */
} else if (integer_size != 8 || num_integers != 1 ||
Fix ENOSPC in "Handle zap_add() failures in ..." Commit cc63068 caused ENOSPC error when copy a large amount of files between two directories. The reason is that the patch limits zap leaf expansion to 2 retries, and return ENOSPC when failed. The intent for limiting retries is to prevent pointlessly growing table to max size when adding a block full of entries with same name in different case in mixed mode. However, it turns out we cannot use any limit on the retry. When we copy files from one directory in readdir order, we are copying in hash order, one leaf block at a time. Which means that if the leaf block in source directory has expanded 6 times, and you copy those entries in that block, by the time you need to expand the leaf in destination directory, you need to expand it 6 times in one go. So any limit on the retry will result in error where it shouldn't. Note that while we do use different salt for different directories, it seems that the salt/hash function doesn't provide enough randomization to the hash distance to prevent this from happening. Since cc63068 has already been reverted. This patch adds it back and removes the retry limit. Also, as it turn out, failing on zap_add() has a serious side effect for mzap_upgrade(). When upgrading from micro zap to fat zap, it will call zap_add() to transfer entries one at a time. If it hit any error halfway through, the remaining entries will be lost, causing those files to become orphan. This patch add a VERIFY to catch it. Reviewed-by: Sanjeev Bagewadi <sanjeev.bagewadi@gmail.com> Reviewed-by: Richard Yao <ryao@gentoo.org> Reviewed-by: Tony Hutter <hutter2@llnl.gov> Reviewed-by: Albert Lee <trisk@forkgnu.org> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Reviewed by: Matthew Ahrens <mahrens@delphix.com> Signed-off-by: Chunwei Chen <david.chen@nutanix.com> Closes #7401 Closes #7421
2018-04-18 21:19:50 +00:00
strlen(key) >= MZAP_NAME_LEN ||
!mze_canfit_fzap_leaf(zn, zn->zn_hash)) {
err = mzap_upgrade(&zn->zn_zap, tag, tx, 0);
if (err == 0) {
err = fzap_add(zn, integer_size, num_integers, val,
tag, tx);
}
2008-11-20 20:01:55 +00:00
zap = zn->zn_zap; /* fzap_add() may change zap */
} else {
mze = mze_find(zn);
if (mze != NULL) {
err = SET_ERROR(EEXIST);
2008-11-20 20:01:55 +00:00
} else {
mzap_addent(zn, *intval);
}
}
ASSERT(zap == zn->zn_zap);
zap_name_free(zn);
if (zap != NULL) /* may be NULL if fzap_add() failed */
zap_unlockdir(zap, tag);
return (err);
}
int
zap_add(objset_t *os, uint64_t zapobj, const char *key,
int integer_size, uint64_t num_integers,
const void *val, dmu_tx_t *tx)
{
zap_t *zap;
int err;
err = zap_lockdir(os, zapobj, tx, RW_WRITER, TRUE, TRUE, FTAG, &zap);
if (err != 0)
return (err);
err = zap_add_impl(zap, key, integer_size, num_integers, val, tx, FTAG);
/* zap_add_impl() calls zap_unlockdir() */
return (err);
}
int
zap_add_by_dnode(dnode_t *dn, const char *key,
int integer_size, uint64_t num_integers,
const void *val, dmu_tx_t *tx)
{
zap_t *zap;
int err;
err = zap_lockdir_by_dnode(dn, tx, RW_WRITER, TRUE, TRUE, FTAG, &zap);
if (err != 0)
return (err);
err = zap_add_impl(zap, key, integer_size, num_integers, val, tx, FTAG);
/* zap_add_impl() calls zap_unlockdir() */
2008-11-20 20:01:55 +00:00
return (err);
}
int
zap_add_uint64(objset_t *os, uint64_t zapobj, const uint64_t *key,
int key_numints, int integer_size, uint64_t num_integers,
const void *val, dmu_tx_t *tx)
{
zap_t *zap;
int err;
zap_name_t *zn;
err = zap_lockdir(os, zapobj, tx, RW_WRITER, TRUE, TRUE, FTAG, &zap);
if (err)
return (err);
zn = zap_name_alloc_uint64(zap, key, key_numints);
if (zn == NULL) {
zap_unlockdir(zap, FTAG);
return (SET_ERROR(ENOTSUP));
}
err = fzap_add(zn, integer_size, num_integers, val, FTAG, tx);
zap = zn->zn_zap; /* fzap_add() may change zap */
zap_name_free(zn);
if (zap != NULL) /* may be NULL if fzap_add() failed */
zap_unlockdir(zap, FTAG);
return (err);
}
2008-11-20 20:01:55 +00:00
int
zap_update(objset_t *os, uint64_t zapobj, const char *name,
int integer_size, uint64_t num_integers, const void *val, dmu_tx_t *tx)
{
zap_t *zap;
mzap_ent_t *mze;
const uint64_t *intval = val;
zap_name_t *zn;
int err;
#ifdef ZFS_DEBUG
uint64_t oldval;
/*
* If there is an old value, it shouldn't change across the
* lockdir (eg, due to bprewrite's xlation).
*/
if (integer_size == 8 && num_integers == 1)
(void) zap_lookup(os, zapobj, name, 8, 1, &oldval);
#endif
err = zap_lockdir(os, zapobj, tx, RW_WRITER, TRUE, TRUE, FTAG, &zap);
2008-11-20 20:01:55 +00:00
if (err)
return (err);
zn = zap_name_alloc(zap, name, 0);
2008-11-20 20:01:55 +00:00
if (zn == NULL) {
zap_unlockdir(zap, FTAG);
return (SET_ERROR(ENOTSUP));
2008-11-20 20:01:55 +00:00
}
if (!zap->zap_ismicro) {
err = fzap_update(zn, integer_size, num_integers, val,
FTAG, tx);
2008-11-20 20:01:55 +00:00
zap = zn->zn_zap; /* fzap_update() may change zap */
} else if (integer_size != 8 || num_integers != 1 ||
strlen(name) >= MZAP_NAME_LEN) {
dprintf("upgrading obj %llu: intsz=%u numint=%llu name=%s\n",
zapobj, integer_size, num_integers, name);
err = mzap_upgrade(&zn->zn_zap, FTAG, tx, 0);
if (err == 0) {
2008-11-20 20:01:55 +00:00
err = fzap_update(zn, integer_size, num_integers,
val, FTAG, tx);
}
2008-11-20 20:01:55 +00:00
zap = zn->zn_zap; /* fzap_update() may change zap */
} else {
mze = mze_find(zn);
if (mze != NULL) {
ASSERT3U(MZE_PHYS(zap, mze)->mze_value, ==, oldval);
MZE_PHYS(zap, mze)->mze_value = *intval;
2008-11-20 20:01:55 +00:00
} else {
mzap_addent(zn, *intval);
}
}
ASSERT(zap == zn->zn_zap);
zap_name_free(zn);
if (zap != NULL) /* may be NULL if fzap_upgrade() failed */
zap_unlockdir(zap, FTAG);
2008-11-20 20:01:55 +00:00
return (err);
}
int
zap_update_uint64(objset_t *os, uint64_t zapobj, const uint64_t *key,
int key_numints,
int integer_size, uint64_t num_integers, const void *val, dmu_tx_t *tx)
{
zap_t *zap;
zap_name_t *zn;
int err;
err = zap_lockdir(os, zapobj, tx, RW_WRITER, TRUE, TRUE, FTAG, &zap);
if (err)
return (err);
zn = zap_name_alloc_uint64(zap, key, key_numints);
if (zn == NULL) {
zap_unlockdir(zap, FTAG);
return (SET_ERROR(ENOTSUP));
}
err = fzap_update(zn, integer_size, num_integers, val, FTAG, tx);
zap = zn->zn_zap; /* fzap_update() may change zap */
zap_name_free(zn);
if (zap != NULL) /* may be NULL if fzap_upgrade() failed */
zap_unlockdir(zap, FTAG);
return (err);
}
2008-11-20 20:01:55 +00:00
int
zap_remove(objset_t *os, uint64_t zapobj, const char *name, dmu_tx_t *tx)
{
return (zap_remove_norm(os, zapobj, name, 0, tx));
2008-11-20 20:01:55 +00:00
}
static int
zap_remove_impl(zap_t *zap, const char *name,
2008-11-20 20:01:55 +00:00
matchtype_t mt, dmu_tx_t *tx)
{
mzap_ent_t *mze;
zap_name_t *zn;
int err = 0;
2008-11-20 20:01:55 +00:00
zn = zap_name_alloc(zap, name, mt);
if (zn == NULL)
return (SET_ERROR(ENOTSUP));
2008-11-20 20:01:55 +00:00
if (!zap->zap_ismicro) {
err = fzap_remove(zn, tx);
} else {
mze = mze_find(zn);
if (mze == NULL) {
err = SET_ERROR(ENOENT);
2008-11-20 20:01:55 +00:00
} else {
zap->zap_m.zap_num_entries--;
bzero(&zap_m_phys(zap)->mz_chunk[mze->mze_chunkid],
2008-11-20 20:01:55 +00:00
sizeof (mzap_ent_phys_t));
mze_remove(zap, mze);
}
}
zap_name_free(zn);
return (err);
}
int
zap_remove_norm(objset_t *os, uint64_t zapobj, const char *name,
matchtype_t mt, dmu_tx_t *tx)
{
zap_t *zap;
int err;
err = zap_lockdir(os, zapobj, tx, RW_WRITER, TRUE, FALSE, FTAG, &zap);
if (err)
return (err);
err = zap_remove_impl(zap, name, mt, tx);
zap_unlockdir(zap, FTAG);
return (err);
}
int
zap_remove_by_dnode(dnode_t *dn, const char *name, dmu_tx_t *tx)
{
zap_t *zap;
int err;
err = zap_lockdir_by_dnode(dn, tx, RW_WRITER, TRUE, FALSE, FTAG, &zap);
if (err)
return (err);
err = zap_remove_impl(zap, name, 0, tx);
zap_unlockdir(zap, FTAG);
2008-11-20 20:01:55 +00:00
return (err);
}
int
zap_remove_uint64(objset_t *os, uint64_t zapobj, const uint64_t *key,
int key_numints, dmu_tx_t *tx)
{
zap_t *zap;
int err;
zap_name_t *zn;
err = zap_lockdir(os, zapobj, tx, RW_WRITER, TRUE, FALSE, FTAG, &zap);
if (err)
return (err);
zn = zap_name_alloc_uint64(zap, key, key_numints);
if (zn == NULL) {
zap_unlockdir(zap, FTAG);
return (SET_ERROR(ENOTSUP));
}
err = fzap_remove(zn, tx);
zap_name_free(zn);
zap_unlockdir(zap, FTAG);
return (err);
}
2008-11-20 20:01:55 +00:00
/*
* Routines for iterating over the attributes.
*/
void
zap_cursor_init_serialized(zap_cursor_t *zc, objset_t *os, uint64_t zapobj,
uint64_t serialized)
{
zc->zc_objset = os;
zc->zc_zap = NULL;
zc->zc_leaf = NULL;
zc->zc_zapobj = zapobj;
zc->zc_serialized = serialized;
zc->zc_hash = 0;
zc->zc_cd = 0;
2008-11-20 20:01:55 +00:00
}
void
zap_cursor_init(zap_cursor_t *zc, objset_t *os, uint64_t zapobj)
{
zap_cursor_init_serialized(zc, os, zapobj, 0);
}
void
zap_cursor_fini(zap_cursor_t *zc)
{
if (zc->zc_zap) {
rw_enter(&zc->zc_zap->zap_rwlock, RW_READER);
zap_unlockdir(zc->zc_zap, NULL);
2008-11-20 20:01:55 +00:00
zc->zc_zap = NULL;
}
if (zc->zc_leaf) {
rw_enter(&zc->zc_leaf->l_rwlock, RW_READER);
zap_put_leaf(zc->zc_leaf);
zc->zc_leaf = NULL;
}
zc->zc_objset = NULL;
}
uint64_t
zap_cursor_serialize(zap_cursor_t *zc)
{
if (zc->zc_hash == -1ULL)
return (-1ULL);
if (zc->zc_zap == NULL)
return (zc->zc_serialized);
ASSERT((zc->zc_hash & zap_maxcd(zc->zc_zap)) == 0);
ASSERT(zc->zc_cd < zap_maxcd(zc->zc_zap));
/*
* We want to keep the high 32 bits of the cursor zero if we can, so
* that 32-bit programs can access this. So usually use a small
* (28-bit) hash value so we can fit 4 bits of cd into the low 32-bits
* of the cursor.
*
* [ collision differentiator | zap_hashbits()-bit hash value ]
*/
return ((zc->zc_hash >> (64 - zap_hashbits(zc->zc_zap))) |
((uint64_t)zc->zc_cd << zap_hashbits(zc->zc_zap)));
2008-11-20 20:01:55 +00:00
}
int
zap_cursor_retrieve(zap_cursor_t *zc, zap_attribute_t *za)
{
int err;
avl_index_t idx;
mzap_ent_t mze_tofind;
mzap_ent_t *mze;
if (zc->zc_hash == -1ULL)
return (SET_ERROR(ENOENT));
2008-11-20 20:01:55 +00:00
if (zc->zc_zap == NULL) {
int hb;
2008-11-20 20:01:55 +00:00
err = zap_lockdir(zc->zc_objset, zc->zc_zapobj, NULL,
RW_READER, TRUE, FALSE, NULL, &zc->zc_zap);
2008-11-20 20:01:55 +00:00
if (err)
return (err);
/*
* To support zap_cursor_init_serialized, advance, retrieve,
* we must add to the existing zc_cd, which may already
* be 1 due to the zap_cursor_advance.
*/
ASSERT(zc->zc_hash == 0);
hb = zap_hashbits(zc->zc_zap);
zc->zc_hash = zc->zc_serialized << (64 - hb);
zc->zc_cd += zc->zc_serialized >> hb;
if (zc->zc_cd >= zap_maxcd(zc->zc_zap)) /* corrupt serialized */
zc->zc_cd = 0;
2008-11-20 20:01:55 +00:00
} else {
rw_enter(&zc->zc_zap->zap_rwlock, RW_READER);
}
if (!zc->zc_zap->zap_ismicro) {
err = fzap_cursor_retrieve(zc->zc_zap, zc, za);
} else {
mze_tofind.mze_hash = zc->zc_hash;
mze_tofind.mze_cd = zc->zc_cd;
2008-11-20 20:01:55 +00:00
mze = avl_find(&zc->zc_zap->zap_m.zap_avl, &mze_tofind, &idx);
if (mze == NULL) {
mze = avl_nearest(&zc->zc_zap->zap_m.zap_avl,
idx, AVL_AFTER);
}
if (mze) {
mzap_ent_phys_t *mzep = MZE_PHYS(zc->zc_zap, mze);
ASSERT3U(mze->mze_cd, ==, mzep->mze_cd);
2008-11-20 20:01:55 +00:00
za->za_normalization_conflict =
mzap_normalization_conflict(zc->zc_zap, NULL, mze);
za->za_integer_length = 8;
za->za_num_integers = 1;
za->za_first_integer = mzep->mze_value;
(void) strcpy(za->za_name, mzep->mze_name);
2008-11-20 20:01:55 +00:00
zc->zc_hash = mze->mze_hash;
zc->zc_cd = mze->mze_cd;
2008-11-20 20:01:55 +00:00
err = 0;
} else {
zc->zc_hash = -1ULL;
err = SET_ERROR(ENOENT);
2008-11-20 20:01:55 +00:00
}
}
rw_exit(&zc->zc_zap->zap_rwlock);
return (err);
}
void
zap_cursor_advance(zap_cursor_t *zc)
{
if (zc->zc_hash == -1ULL)
return;
zc->zc_cd++;
}
2008-11-20 20:01:55 +00:00
int
zap_get_stats(objset_t *os, uint64_t zapobj, zap_stats_t *zs)
{
int err;
zap_t *zap;
err = zap_lockdir(os, zapobj, NULL, RW_READER, TRUE, FALSE, FTAG, &zap);
2008-11-20 20:01:55 +00:00
if (err)
return (err);
bzero(zs, sizeof (zap_stats_t));
if (zap->zap_ismicro) {
zs->zs_blocksize = zap->zap_dbuf->db_size;
zs->zs_num_entries = zap->zap_m.zap_num_entries;
zs->zs_num_blocks = 1;
} else {
fzap_get_stats(zap, zs);
}
zap_unlockdir(zap, FTAG);
2008-11-20 20:01:55 +00:00
return (0);
}
2009-07-02 22:44:48 +00:00
Update build system and packaging Minimal changes required to integrate the SPL sources in to the ZFS repository build infrastructure and packaging. Build system and packaging: * Renamed SPL_* autoconf m4 macros to ZFS_*. * Removed redundant SPL_* autoconf m4 macros. * Updated the RPM spec files to remove SPL package dependency. * The zfs package obsoletes the spl package, and the zfs-kmod package obsoletes the spl-kmod package. * The zfs-kmod-devel* packages were updated to add compatibility symlinks under /usr/src/spl-x.y.z until all dependent packages can be updated. They will be removed in a future release. * Updated copy-builtin script for in-kernel builds. * Updated DKMS package to include the spl.ko. * Updated stale AUTHORS file to include all contributors. * Updated stale COPYRIGHT and included the SPL as an exception. * Renamed README.markdown to README.md * Renamed OPENSOLARIS.LICENSE to LICENSE. * Renamed DISCLAIMER to NOTICE. Required code changes: * Removed redundant HAVE_SPL macro. * Removed _BOOT from nvpairs since it doesn't apply for Linux. * Initial header cleanup (removal of empty headers, refactoring). * Remove SPL repository clone/build from zimport.sh. * Use of DEFINE_RATELIMIT_STATE and DEFINE_SPINLOCK removed due to build issues when forcing C99 compilation. * Replaced legacy ACCESS_ONCE with READ_ONCE. * Include needed headers for `current` and `EXPORT_SYMBOL`. Reviewed-by: Tony Hutter <hutter2@llnl.gov> Reviewed-by: Olaf Faaland <faaland1@llnl.gov> Reviewed-by: Matthew Ahrens <mahrens@delphix.com> Reviewed-by: Pavel Zakharov <pavel.zakharov@delphix.com> Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov> TEST_ZIMPORT_SKIP="yes" Closes #7556
2018-02-16 01:53:18 +00:00
#if defined(_KERNEL)
EXPORT_SYMBOL(zap_create);
Implement large_dnode pool feature Justification ------------- This feature adds support for variable length dnodes. Our motivation is to eliminate the overhead associated with using spill blocks. Spill blocks are used to store system attribute data (i.e. file metadata) that does not fit in the dnode's bonus buffer. By allowing a larger bonus buffer area the use of a spill block can be avoided. Spill blocks potentially incur an additional read I/O for every dnode in a dnode block. As a worst case example, reading 32 dnodes from a 16k dnode block and all of the spill blocks could issue 33 separate reads. Now suppose those dnodes have size 1024 and therefore don't need spill blocks. Then the worst case number of blocks read is reduced to from 33 to two--one per dnode block. In practice spill blocks may tend to be co-located on disk with the dnode blocks so the reduction in I/O would not be this drastic. In a badly fragmented pool, however, the improvement could be significant. ZFS-on-Linux systems that make heavy use of extended attributes would benefit from this feature. In particular, ZFS-on-Linux supports the xattr=sa dataset property which allows file extended attribute data to be stored in the dnode bonus buffer as an alternative to the traditional directory-based format. Workloads such as SELinux and the Lustre distributed filesystem often store enough xattr data to force spill bocks when xattr=sa is in effect. Large dnodes may therefore provide a performance benefit to such systems. Other use cases that may benefit from this feature include files with large ACLs and symbolic links with long target names. Furthermore, this feature may be desirable on other platforms in case future applications or features are developed that could make use of a larger bonus buffer area. Implementation -------------- The size of a dnode may be a multiple of 512 bytes up to the size of a dnode block (currently 16384 bytes). A dn_extra_slots field was added to the current on-disk dnode_phys_t structure to describe the size of the physical dnode on disk. The 8 bits for this field were taken from the zero filled dn_pad2 field. The field represents how many "extra" dnode_phys_t slots a dnode consumes in its dnode block. This convention results in a value of 0 for 512 byte dnodes which preserves on-disk format compatibility with older software. Similarly, the in-memory dnode_t structure has a new dn_num_slots field to represent the total number of dnode_phys_t slots consumed on disk. Thus dn->dn_num_slots is 1 greater than the corresponding dnp->dn_extra_slots. This difference in convention was adopted because, unlike on-disk structures, backward compatibility is not a concern for in-memory objects, so we used a more natural way to represent size for a dnode_t. The default size for newly created dnodes is determined by the value of a new "dnodesize" dataset property. By default the property is set to "legacy" which is compatible with older software. Setting the property to "auto" will allow the filesystem to choose the most suitable dnode size. Currently this just sets the default dnode size to 1k, but future code improvements could dynamically choose a size based on observed workload patterns. Dnodes of varying sizes can coexist within the same dataset and even within the same dnode block. For example, to enable automatically-sized dnodes, run # zfs set dnodesize=auto tank/fish The user can also specify literal values for the dnodesize property. These are currently limited to powers of two from 1k to 16k. The power-of-2 limitation is only for simplicity of the user interface. Internally the implementation can handle any multiple of 512 up to 16k, and consumers of the DMU API can specify any legal dnode value. The size of a new dnode is determined at object allocation time and stored as a new field in the znode in-memory structure. New DMU interfaces are added to allow the consumer to specify the dnode size that a newly allocated object should use. Existing interfaces are unchanged to avoid having to update every call site and to preserve compatibility with external consumers such as Lustre. The new interfaces names are given below. The versions of these functions that don't take a dnodesize parameter now just call the _dnsize() versions with a dnodesize of 0, which means use the legacy dnode size. New DMU interfaces: dmu_object_alloc_dnsize() dmu_object_claim_dnsize() dmu_object_reclaim_dnsize() New ZAP interfaces: zap_create_dnsize() zap_create_norm_dnsize() zap_create_flags_dnsize() zap_create_claim_norm_dnsize() zap_create_link_dnsize() The constant DN_MAX_BONUSLEN is renamed to DN_OLD_MAX_BONUSLEN. The spa_maxdnodesize() function should be used to determine the maximum bonus length for a pool. These are a few noteworthy changes to key functions: * The prototype for dnode_hold_impl() now takes a "slots" parameter. When the DNODE_MUST_BE_FREE flag is set, this parameter is used to ensure the hole at the specified object offset is large enough to hold the dnode being created. The slots parameter is also used to ensure a dnode does not span multiple dnode blocks. In both of these cases, if a failure occurs, ENOSPC is returned. Keep in mind, these failure cases are only possible when using DNODE_MUST_BE_FREE. If the DNODE_MUST_BE_ALLOCATED flag is set, "slots" must be 0. dnode_hold_impl() will check if the requested dnode is already consumed as an extra dnode slot by an large dnode, in which case it returns ENOENT. * The function dmu_object_alloc() advances to the next dnode block if dnode_hold_impl() returns an error for a requested object. This is because the beginning of the next dnode block is the only location it can safely assume to either be a hole or a valid starting point for a dnode. * dnode_next_offset_level() and other functions that iterate through dnode blocks may no longer use a simple array indexing scheme. These now use the current dnode's dn_num_slots field to advance to the next dnode in the block. This is to ensure we properly skip the current dnode's bonus area and don't interpret it as a valid dnode. zdb --- The zdb command was updated to display a dnode's size under the "dnsize" column when the object is dumped. For ZIL create log records, zdb will now display the slot count for the object. ztest ----- Ztest chooses a random dnodesize for every newly created object. The random distribution is more heavily weighted toward small dnodes to better simulate real-world datasets. Unused bonus buffer space is filled with non-zero values computed from the object number, dataset id, offset, and generation number. This helps ensure that the dnode traversal code properly skips the interior regions of large dnodes, and that these interior regions are not overwritten by data belonging to other dnodes. A new test visits each object in a dataset. It verifies that the actual dnode size matches what was stored in the ztest block tag when it was created. It also verifies that the unused bonus buffer space is filled with the expected data patterns. ZFS Test Suite -------------- Added six new large dnode-specific tests, and integrated the dnodesize property into existing tests for zfs allow and send/recv. Send/Receive ------------ ZFS send streams for datasets containing large dnodes cannot be received on pools that don't support the large_dnode feature. A send stream with large dnodes sets a DMU_BACKUP_FEATURE_LARGE_DNODE flag which will be unrecognized by an incompatible receiving pool so that the zfs receive will fail gracefully. While not implemented here, it may be possible to generate a backward-compatible send stream from a dataset containing large dnodes. The implementation may be tricky, however, because the send object record for a large dnode would need to be resized to a 512 byte dnode, possibly kicking in a spill block in the process. This means we would need to construct a new SA layout and possibly register it in the SA layout object. The SA layout is normally just sent as an ordinary object record. But if we are constructing new layouts while generating the send stream we'd have to build the SA layout object dynamically and send it at the end of the stream. For sending and receiving between pools that do support large dnodes, the drr_object send record type is extended with a new field to store the dnode slot count. This field was repurposed from unused padding in the structure. ZIL Replay ---------- The dnode slot count is stored in the uppermost 8 bits of the lr_foid field. The bits were unused as the object id is currently capped at 48 bits. Resizing Dnodes --------------- It should be possible to resize a dnode when it is dirtied if the current dnodesize dataset property differs from the dnode's size, but this functionality is not currently implemented. Clearly a dnode can only grow if there are sufficient contiguous unused slots in the dnode block, but it should always be possible to shrink a dnode. Growing dnodes may be useful to reduce fragmentation in a pool with many spill blocks in use. Shrinking dnodes may be useful to allow sending a dataset to a pool that doesn't support the large_dnode feature. Feature Reference Counting -------------------------- The reference count for the large_dnode pool feature tracks the number of datasets that have ever contained a dnode of size larger than 512 bytes. The first time a large dnode is created in a dataset the dataset is converted to an extensible dataset. This is a one-way operation and the only way to decrement the feature count is to destroy the dataset, even if the dataset no longer contains any large dnodes. The complexity of reference counting on a per-dnode basis was too high, so we chose to track it on a per-dataset basis similarly to the large_block feature. Signed-off-by: Ned Bass <bass6@llnl.gov> Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov> Closes #3542
2016-03-17 01:25:34 +00:00
EXPORT_SYMBOL(zap_create_dnsize);
EXPORT_SYMBOL(zap_create_norm);
Implement large_dnode pool feature Justification ------------- This feature adds support for variable length dnodes. Our motivation is to eliminate the overhead associated with using spill blocks. Spill blocks are used to store system attribute data (i.e. file metadata) that does not fit in the dnode's bonus buffer. By allowing a larger bonus buffer area the use of a spill block can be avoided. Spill blocks potentially incur an additional read I/O for every dnode in a dnode block. As a worst case example, reading 32 dnodes from a 16k dnode block and all of the spill blocks could issue 33 separate reads. Now suppose those dnodes have size 1024 and therefore don't need spill blocks. Then the worst case number of blocks read is reduced to from 33 to two--one per dnode block. In practice spill blocks may tend to be co-located on disk with the dnode blocks so the reduction in I/O would not be this drastic. In a badly fragmented pool, however, the improvement could be significant. ZFS-on-Linux systems that make heavy use of extended attributes would benefit from this feature. In particular, ZFS-on-Linux supports the xattr=sa dataset property which allows file extended attribute data to be stored in the dnode bonus buffer as an alternative to the traditional directory-based format. Workloads such as SELinux and the Lustre distributed filesystem often store enough xattr data to force spill bocks when xattr=sa is in effect. Large dnodes may therefore provide a performance benefit to such systems. Other use cases that may benefit from this feature include files with large ACLs and symbolic links with long target names. Furthermore, this feature may be desirable on other platforms in case future applications or features are developed that could make use of a larger bonus buffer area. Implementation -------------- The size of a dnode may be a multiple of 512 bytes up to the size of a dnode block (currently 16384 bytes). A dn_extra_slots field was added to the current on-disk dnode_phys_t structure to describe the size of the physical dnode on disk. The 8 bits for this field were taken from the zero filled dn_pad2 field. The field represents how many "extra" dnode_phys_t slots a dnode consumes in its dnode block. This convention results in a value of 0 for 512 byte dnodes which preserves on-disk format compatibility with older software. Similarly, the in-memory dnode_t structure has a new dn_num_slots field to represent the total number of dnode_phys_t slots consumed on disk. Thus dn->dn_num_slots is 1 greater than the corresponding dnp->dn_extra_slots. This difference in convention was adopted because, unlike on-disk structures, backward compatibility is not a concern for in-memory objects, so we used a more natural way to represent size for a dnode_t. The default size for newly created dnodes is determined by the value of a new "dnodesize" dataset property. By default the property is set to "legacy" which is compatible with older software. Setting the property to "auto" will allow the filesystem to choose the most suitable dnode size. Currently this just sets the default dnode size to 1k, but future code improvements could dynamically choose a size based on observed workload patterns. Dnodes of varying sizes can coexist within the same dataset and even within the same dnode block. For example, to enable automatically-sized dnodes, run # zfs set dnodesize=auto tank/fish The user can also specify literal values for the dnodesize property. These are currently limited to powers of two from 1k to 16k. The power-of-2 limitation is only for simplicity of the user interface. Internally the implementation can handle any multiple of 512 up to 16k, and consumers of the DMU API can specify any legal dnode value. The size of a new dnode is determined at object allocation time and stored as a new field in the znode in-memory structure. New DMU interfaces are added to allow the consumer to specify the dnode size that a newly allocated object should use. Existing interfaces are unchanged to avoid having to update every call site and to preserve compatibility with external consumers such as Lustre. The new interfaces names are given below. The versions of these functions that don't take a dnodesize parameter now just call the _dnsize() versions with a dnodesize of 0, which means use the legacy dnode size. New DMU interfaces: dmu_object_alloc_dnsize() dmu_object_claim_dnsize() dmu_object_reclaim_dnsize() New ZAP interfaces: zap_create_dnsize() zap_create_norm_dnsize() zap_create_flags_dnsize() zap_create_claim_norm_dnsize() zap_create_link_dnsize() The constant DN_MAX_BONUSLEN is renamed to DN_OLD_MAX_BONUSLEN. The spa_maxdnodesize() function should be used to determine the maximum bonus length for a pool. These are a few noteworthy changes to key functions: * The prototype for dnode_hold_impl() now takes a "slots" parameter. When the DNODE_MUST_BE_FREE flag is set, this parameter is used to ensure the hole at the specified object offset is large enough to hold the dnode being created. The slots parameter is also used to ensure a dnode does not span multiple dnode blocks. In both of these cases, if a failure occurs, ENOSPC is returned. Keep in mind, these failure cases are only possible when using DNODE_MUST_BE_FREE. If the DNODE_MUST_BE_ALLOCATED flag is set, "slots" must be 0. dnode_hold_impl() will check if the requested dnode is already consumed as an extra dnode slot by an large dnode, in which case it returns ENOENT. * The function dmu_object_alloc() advances to the next dnode block if dnode_hold_impl() returns an error for a requested object. This is because the beginning of the next dnode block is the only location it can safely assume to either be a hole or a valid starting point for a dnode. * dnode_next_offset_level() and other functions that iterate through dnode blocks may no longer use a simple array indexing scheme. These now use the current dnode's dn_num_slots field to advance to the next dnode in the block. This is to ensure we properly skip the current dnode's bonus area and don't interpret it as a valid dnode. zdb --- The zdb command was updated to display a dnode's size under the "dnsize" column when the object is dumped. For ZIL create log records, zdb will now display the slot count for the object. ztest ----- Ztest chooses a random dnodesize for every newly created object. The random distribution is more heavily weighted toward small dnodes to better simulate real-world datasets. Unused bonus buffer space is filled with non-zero values computed from the object number, dataset id, offset, and generation number. This helps ensure that the dnode traversal code properly skips the interior regions of large dnodes, and that these interior regions are not overwritten by data belonging to other dnodes. A new test visits each object in a dataset. It verifies that the actual dnode size matches what was stored in the ztest block tag when it was created. It also verifies that the unused bonus buffer space is filled with the expected data patterns. ZFS Test Suite -------------- Added six new large dnode-specific tests, and integrated the dnodesize property into existing tests for zfs allow and send/recv. Send/Receive ------------ ZFS send streams for datasets containing large dnodes cannot be received on pools that don't support the large_dnode feature. A send stream with large dnodes sets a DMU_BACKUP_FEATURE_LARGE_DNODE flag which will be unrecognized by an incompatible receiving pool so that the zfs receive will fail gracefully. While not implemented here, it may be possible to generate a backward-compatible send stream from a dataset containing large dnodes. The implementation may be tricky, however, because the send object record for a large dnode would need to be resized to a 512 byte dnode, possibly kicking in a spill block in the process. This means we would need to construct a new SA layout and possibly register it in the SA layout object. The SA layout is normally just sent as an ordinary object record. But if we are constructing new layouts while generating the send stream we'd have to build the SA layout object dynamically and send it at the end of the stream. For sending and receiving between pools that do support large dnodes, the drr_object send record type is extended with a new field to store the dnode slot count. This field was repurposed from unused padding in the structure. ZIL Replay ---------- The dnode slot count is stored in the uppermost 8 bits of the lr_foid field. The bits were unused as the object id is currently capped at 48 bits. Resizing Dnodes --------------- It should be possible to resize a dnode when it is dirtied if the current dnodesize dataset property differs from the dnode's size, but this functionality is not currently implemented. Clearly a dnode can only grow if there are sufficient contiguous unused slots in the dnode block, but it should always be possible to shrink a dnode. Growing dnodes may be useful to reduce fragmentation in a pool with many spill blocks in use. Shrinking dnodes may be useful to allow sending a dataset to a pool that doesn't support the large_dnode feature. Feature Reference Counting -------------------------- The reference count for the large_dnode pool feature tracks the number of datasets that have ever contained a dnode of size larger than 512 bytes. The first time a large dnode is created in a dataset the dataset is converted to an extensible dataset. This is a one-way operation and the only way to decrement the feature count is to destroy the dataset, even if the dataset no longer contains any large dnodes. The complexity of reference counting on a per-dnode basis was too high, so we chose to track it on a per-dataset basis similarly to the large_block feature. Signed-off-by: Ned Bass <bass6@llnl.gov> Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov> Closes #3542
2016-03-17 01:25:34 +00:00
EXPORT_SYMBOL(zap_create_norm_dnsize);
EXPORT_SYMBOL(zap_create_flags);
Implement large_dnode pool feature Justification ------------- This feature adds support for variable length dnodes. Our motivation is to eliminate the overhead associated with using spill blocks. Spill blocks are used to store system attribute data (i.e. file metadata) that does not fit in the dnode's bonus buffer. By allowing a larger bonus buffer area the use of a spill block can be avoided. Spill blocks potentially incur an additional read I/O for every dnode in a dnode block. As a worst case example, reading 32 dnodes from a 16k dnode block and all of the spill blocks could issue 33 separate reads. Now suppose those dnodes have size 1024 and therefore don't need spill blocks. Then the worst case number of blocks read is reduced to from 33 to two--one per dnode block. In practice spill blocks may tend to be co-located on disk with the dnode blocks so the reduction in I/O would not be this drastic. In a badly fragmented pool, however, the improvement could be significant. ZFS-on-Linux systems that make heavy use of extended attributes would benefit from this feature. In particular, ZFS-on-Linux supports the xattr=sa dataset property which allows file extended attribute data to be stored in the dnode bonus buffer as an alternative to the traditional directory-based format. Workloads such as SELinux and the Lustre distributed filesystem often store enough xattr data to force spill bocks when xattr=sa is in effect. Large dnodes may therefore provide a performance benefit to such systems. Other use cases that may benefit from this feature include files with large ACLs and symbolic links with long target names. Furthermore, this feature may be desirable on other platforms in case future applications or features are developed that could make use of a larger bonus buffer area. Implementation -------------- The size of a dnode may be a multiple of 512 bytes up to the size of a dnode block (currently 16384 bytes). A dn_extra_slots field was added to the current on-disk dnode_phys_t structure to describe the size of the physical dnode on disk. The 8 bits for this field were taken from the zero filled dn_pad2 field. The field represents how many "extra" dnode_phys_t slots a dnode consumes in its dnode block. This convention results in a value of 0 for 512 byte dnodes which preserves on-disk format compatibility with older software. Similarly, the in-memory dnode_t structure has a new dn_num_slots field to represent the total number of dnode_phys_t slots consumed on disk. Thus dn->dn_num_slots is 1 greater than the corresponding dnp->dn_extra_slots. This difference in convention was adopted because, unlike on-disk structures, backward compatibility is not a concern for in-memory objects, so we used a more natural way to represent size for a dnode_t. The default size for newly created dnodes is determined by the value of a new "dnodesize" dataset property. By default the property is set to "legacy" which is compatible with older software. Setting the property to "auto" will allow the filesystem to choose the most suitable dnode size. Currently this just sets the default dnode size to 1k, but future code improvements could dynamically choose a size based on observed workload patterns. Dnodes of varying sizes can coexist within the same dataset and even within the same dnode block. For example, to enable automatically-sized dnodes, run # zfs set dnodesize=auto tank/fish The user can also specify literal values for the dnodesize property. These are currently limited to powers of two from 1k to 16k. The power-of-2 limitation is only for simplicity of the user interface. Internally the implementation can handle any multiple of 512 up to 16k, and consumers of the DMU API can specify any legal dnode value. The size of a new dnode is determined at object allocation time and stored as a new field in the znode in-memory structure. New DMU interfaces are added to allow the consumer to specify the dnode size that a newly allocated object should use. Existing interfaces are unchanged to avoid having to update every call site and to preserve compatibility with external consumers such as Lustre. The new interfaces names are given below. The versions of these functions that don't take a dnodesize parameter now just call the _dnsize() versions with a dnodesize of 0, which means use the legacy dnode size. New DMU interfaces: dmu_object_alloc_dnsize() dmu_object_claim_dnsize() dmu_object_reclaim_dnsize() New ZAP interfaces: zap_create_dnsize() zap_create_norm_dnsize() zap_create_flags_dnsize() zap_create_claim_norm_dnsize() zap_create_link_dnsize() The constant DN_MAX_BONUSLEN is renamed to DN_OLD_MAX_BONUSLEN. The spa_maxdnodesize() function should be used to determine the maximum bonus length for a pool. These are a few noteworthy changes to key functions: * The prototype for dnode_hold_impl() now takes a "slots" parameter. When the DNODE_MUST_BE_FREE flag is set, this parameter is used to ensure the hole at the specified object offset is large enough to hold the dnode being created. The slots parameter is also used to ensure a dnode does not span multiple dnode blocks. In both of these cases, if a failure occurs, ENOSPC is returned. Keep in mind, these failure cases are only possible when using DNODE_MUST_BE_FREE. If the DNODE_MUST_BE_ALLOCATED flag is set, "slots" must be 0. dnode_hold_impl() will check if the requested dnode is already consumed as an extra dnode slot by an large dnode, in which case it returns ENOENT. * The function dmu_object_alloc() advances to the next dnode block if dnode_hold_impl() returns an error for a requested object. This is because the beginning of the next dnode block is the only location it can safely assume to either be a hole or a valid starting point for a dnode. * dnode_next_offset_level() and other functions that iterate through dnode blocks may no longer use a simple array indexing scheme. These now use the current dnode's dn_num_slots field to advance to the next dnode in the block. This is to ensure we properly skip the current dnode's bonus area and don't interpret it as a valid dnode. zdb --- The zdb command was updated to display a dnode's size under the "dnsize" column when the object is dumped. For ZIL create log records, zdb will now display the slot count for the object. ztest ----- Ztest chooses a random dnodesize for every newly created object. The random distribution is more heavily weighted toward small dnodes to better simulate real-world datasets. Unused bonus buffer space is filled with non-zero values computed from the object number, dataset id, offset, and generation number. This helps ensure that the dnode traversal code properly skips the interior regions of large dnodes, and that these interior regions are not overwritten by data belonging to other dnodes. A new test visits each object in a dataset. It verifies that the actual dnode size matches what was stored in the ztest block tag when it was created. It also verifies that the unused bonus buffer space is filled with the expected data patterns. ZFS Test Suite -------------- Added six new large dnode-specific tests, and integrated the dnodesize property into existing tests for zfs allow and send/recv. Send/Receive ------------ ZFS send streams for datasets containing large dnodes cannot be received on pools that don't support the large_dnode feature. A send stream with large dnodes sets a DMU_BACKUP_FEATURE_LARGE_DNODE flag which will be unrecognized by an incompatible receiving pool so that the zfs receive will fail gracefully. While not implemented here, it may be possible to generate a backward-compatible send stream from a dataset containing large dnodes. The implementation may be tricky, however, because the send object record for a large dnode would need to be resized to a 512 byte dnode, possibly kicking in a spill block in the process. This means we would need to construct a new SA layout and possibly register it in the SA layout object. The SA layout is normally just sent as an ordinary object record. But if we are constructing new layouts while generating the send stream we'd have to build the SA layout object dynamically and send it at the end of the stream. For sending and receiving between pools that do support large dnodes, the drr_object send record type is extended with a new field to store the dnode slot count. This field was repurposed from unused padding in the structure. ZIL Replay ---------- The dnode slot count is stored in the uppermost 8 bits of the lr_foid field. The bits were unused as the object id is currently capped at 48 bits. Resizing Dnodes --------------- It should be possible to resize a dnode when it is dirtied if the current dnodesize dataset property differs from the dnode's size, but this functionality is not currently implemented. Clearly a dnode can only grow if there are sufficient contiguous unused slots in the dnode block, but it should always be possible to shrink a dnode. Growing dnodes may be useful to reduce fragmentation in a pool with many spill blocks in use. Shrinking dnodes may be useful to allow sending a dataset to a pool that doesn't support the large_dnode feature. Feature Reference Counting -------------------------- The reference count for the large_dnode pool feature tracks the number of datasets that have ever contained a dnode of size larger than 512 bytes. The first time a large dnode is created in a dataset the dataset is converted to an extensible dataset. This is a one-way operation and the only way to decrement the feature count is to destroy the dataset, even if the dataset no longer contains any large dnodes. The complexity of reference counting on a per-dnode basis was too high, so we chose to track it on a per-dataset basis similarly to the large_block feature. Signed-off-by: Ned Bass <bass6@llnl.gov> Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov> Closes #3542
2016-03-17 01:25:34 +00:00
EXPORT_SYMBOL(zap_create_flags_dnsize);
EXPORT_SYMBOL(zap_create_claim);
EXPORT_SYMBOL(zap_create_claim_norm);
Implement large_dnode pool feature Justification ------------- This feature adds support for variable length dnodes. Our motivation is to eliminate the overhead associated with using spill blocks. Spill blocks are used to store system attribute data (i.e. file metadata) that does not fit in the dnode's bonus buffer. By allowing a larger bonus buffer area the use of a spill block can be avoided. Spill blocks potentially incur an additional read I/O for every dnode in a dnode block. As a worst case example, reading 32 dnodes from a 16k dnode block and all of the spill blocks could issue 33 separate reads. Now suppose those dnodes have size 1024 and therefore don't need spill blocks. Then the worst case number of blocks read is reduced to from 33 to two--one per dnode block. In practice spill blocks may tend to be co-located on disk with the dnode blocks so the reduction in I/O would not be this drastic. In a badly fragmented pool, however, the improvement could be significant. ZFS-on-Linux systems that make heavy use of extended attributes would benefit from this feature. In particular, ZFS-on-Linux supports the xattr=sa dataset property which allows file extended attribute data to be stored in the dnode bonus buffer as an alternative to the traditional directory-based format. Workloads such as SELinux and the Lustre distributed filesystem often store enough xattr data to force spill bocks when xattr=sa is in effect. Large dnodes may therefore provide a performance benefit to such systems. Other use cases that may benefit from this feature include files with large ACLs and symbolic links with long target names. Furthermore, this feature may be desirable on other platforms in case future applications or features are developed that could make use of a larger bonus buffer area. Implementation -------------- The size of a dnode may be a multiple of 512 bytes up to the size of a dnode block (currently 16384 bytes). A dn_extra_slots field was added to the current on-disk dnode_phys_t structure to describe the size of the physical dnode on disk. The 8 bits for this field were taken from the zero filled dn_pad2 field. The field represents how many "extra" dnode_phys_t slots a dnode consumes in its dnode block. This convention results in a value of 0 for 512 byte dnodes which preserves on-disk format compatibility with older software. Similarly, the in-memory dnode_t structure has a new dn_num_slots field to represent the total number of dnode_phys_t slots consumed on disk. Thus dn->dn_num_slots is 1 greater than the corresponding dnp->dn_extra_slots. This difference in convention was adopted because, unlike on-disk structures, backward compatibility is not a concern for in-memory objects, so we used a more natural way to represent size for a dnode_t. The default size for newly created dnodes is determined by the value of a new "dnodesize" dataset property. By default the property is set to "legacy" which is compatible with older software. Setting the property to "auto" will allow the filesystem to choose the most suitable dnode size. Currently this just sets the default dnode size to 1k, but future code improvements could dynamically choose a size based on observed workload patterns. Dnodes of varying sizes can coexist within the same dataset and even within the same dnode block. For example, to enable automatically-sized dnodes, run # zfs set dnodesize=auto tank/fish The user can also specify literal values for the dnodesize property. These are currently limited to powers of two from 1k to 16k. The power-of-2 limitation is only for simplicity of the user interface. Internally the implementation can handle any multiple of 512 up to 16k, and consumers of the DMU API can specify any legal dnode value. The size of a new dnode is determined at object allocation time and stored as a new field in the znode in-memory structure. New DMU interfaces are added to allow the consumer to specify the dnode size that a newly allocated object should use. Existing interfaces are unchanged to avoid having to update every call site and to preserve compatibility with external consumers such as Lustre. The new interfaces names are given below. The versions of these functions that don't take a dnodesize parameter now just call the _dnsize() versions with a dnodesize of 0, which means use the legacy dnode size. New DMU interfaces: dmu_object_alloc_dnsize() dmu_object_claim_dnsize() dmu_object_reclaim_dnsize() New ZAP interfaces: zap_create_dnsize() zap_create_norm_dnsize() zap_create_flags_dnsize() zap_create_claim_norm_dnsize() zap_create_link_dnsize() The constant DN_MAX_BONUSLEN is renamed to DN_OLD_MAX_BONUSLEN. The spa_maxdnodesize() function should be used to determine the maximum bonus length for a pool. These are a few noteworthy changes to key functions: * The prototype for dnode_hold_impl() now takes a "slots" parameter. When the DNODE_MUST_BE_FREE flag is set, this parameter is used to ensure the hole at the specified object offset is large enough to hold the dnode being created. The slots parameter is also used to ensure a dnode does not span multiple dnode blocks. In both of these cases, if a failure occurs, ENOSPC is returned. Keep in mind, these failure cases are only possible when using DNODE_MUST_BE_FREE. If the DNODE_MUST_BE_ALLOCATED flag is set, "slots" must be 0. dnode_hold_impl() will check if the requested dnode is already consumed as an extra dnode slot by an large dnode, in which case it returns ENOENT. * The function dmu_object_alloc() advances to the next dnode block if dnode_hold_impl() returns an error for a requested object. This is because the beginning of the next dnode block is the only location it can safely assume to either be a hole or a valid starting point for a dnode. * dnode_next_offset_level() and other functions that iterate through dnode blocks may no longer use a simple array indexing scheme. These now use the current dnode's dn_num_slots field to advance to the next dnode in the block. This is to ensure we properly skip the current dnode's bonus area and don't interpret it as a valid dnode. zdb --- The zdb command was updated to display a dnode's size under the "dnsize" column when the object is dumped. For ZIL create log records, zdb will now display the slot count for the object. ztest ----- Ztest chooses a random dnodesize for every newly created object. The random distribution is more heavily weighted toward small dnodes to better simulate real-world datasets. Unused bonus buffer space is filled with non-zero values computed from the object number, dataset id, offset, and generation number. This helps ensure that the dnode traversal code properly skips the interior regions of large dnodes, and that these interior regions are not overwritten by data belonging to other dnodes. A new test visits each object in a dataset. It verifies that the actual dnode size matches what was stored in the ztest block tag when it was created. It also verifies that the unused bonus buffer space is filled with the expected data patterns. ZFS Test Suite -------------- Added six new large dnode-specific tests, and integrated the dnodesize property into existing tests for zfs allow and send/recv. Send/Receive ------------ ZFS send streams for datasets containing large dnodes cannot be received on pools that don't support the large_dnode feature. A send stream with large dnodes sets a DMU_BACKUP_FEATURE_LARGE_DNODE flag which will be unrecognized by an incompatible receiving pool so that the zfs receive will fail gracefully. While not implemented here, it may be possible to generate a backward-compatible send stream from a dataset containing large dnodes. The implementation may be tricky, however, because the send object record for a large dnode would need to be resized to a 512 byte dnode, possibly kicking in a spill block in the process. This means we would need to construct a new SA layout and possibly register it in the SA layout object. The SA layout is normally just sent as an ordinary object record. But if we are constructing new layouts while generating the send stream we'd have to build the SA layout object dynamically and send it at the end of the stream. For sending and receiving between pools that do support large dnodes, the drr_object send record type is extended with a new field to store the dnode slot count. This field was repurposed from unused padding in the structure. ZIL Replay ---------- The dnode slot count is stored in the uppermost 8 bits of the lr_foid field. The bits were unused as the object id is currently capped at 48 bits. Resizing Dnodes --------------- It should be possible to resize a dnode when it is dirtied if the current dnodesize dataset property differs from the dnode's size, but this functionality is not currently implemented. Clearly a dnode can only grow if there are sufficient contiguous unused slots in the dnode block, but it should always be possible to shrink a dnode. Growing dnodes may be useful to reduce fragmentation in a pool with many spill blocks in use. Shrinking dnodes may be useful to allow sending a dataset to a pool that doesn't support the large_dnode feature. Feature Reference Counting -------------------------- The reference count for the large_dnode pool feature tracks the number of datasets that have ever contained a dnode of size larger than 512 bytes. The first time a large dnode is created in a dataset the dataset is converted to an extensible dataset. This is a one-way operation and the only way to decrement the feature count is to destroy the dataset, even if the dataset no longer contains any large dnodes. The complexity of reference counting on a per-dnode basis was too high, so we chose to track it on a per-dataset basis similarly to the large_block feature. Signed-off-by: Ned Bass <bass6@llnl.gov> Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov> Closes #3542
2016-03-17 01:25:34 +00:00
EXPORT_SYMBOL(zap_create_claim_norm_dnsize);
EXPORT_SYMBOL(zap_destroy);
EXPORT_SYMBOL(zap_lookup);
EXPORT_SYMBOL(zap_lookup_by_dnode);
EXPORT_SYMBOL(zap_lookup_norm);
EXPORT_SYMBOL(zap_lookup_uint64);
EXPORT_SYMBOL(zap_contains);
EXPORT_SYMBOL(zap_prefetch);
EXPORT_SYMBOL(zap_prefetch_uint64);
EXPORT_SYMBOL(zap_add);
EXPORT_SYMBOL(zap_add_by_dnode);
EXPORT_SYMBOL(zap_add_uint64);
EXPORT_SYMBOL(zap_update);
EXPORT_SYMBOL(zap_update_uint64);
EXPORT_SYMBOL(zap_length);
EXPORT_SYMBOL(zap_length_uint64);
EXPORT_SYMBOL(zap_remove);
EXPORT_SYMBOL(zap_remove_by_dnode);
EXPORT_SYMBOL(zap_remove_norm);
EXPORT_SYMBOL(zap_remove_uint64);
EXPORT_SYMBOL(zap_count);
EXPORT_SYMBOL(zap_value_search);
EXPORT_SYMBOL(zap_join);
EXPORT_SYMBOL(zap_join_increment);
EXPORT_SYMBOL(zap_add_int);
EXPORT_SYMBOL(zap_remove_int);
EXPORT_SYMBOL(zap_lookup_int);
EXPORT_SYMBOL(zap_increment_int);
EXPORT_SYMBOL(zap_add_int_key);
EXPORT_SYMBOL(zap_lookup_int_key);
EXPORT_SYMBOL(zap_increment);
EXPORT_SYMBOL(zap_cursor_init);
EXPORT_SYMBOL(zap_cursor_fini);
EXPORT_SYMBOL(zap_cursor_retrieve);
EXPORT_SYMBOL(zap_cursor_advance);
EXPORT_SYMBOL(zap_cursor_serialize);
EXPORT_SYMBOL(zap_cursor_init_serialized);
EXPORT_SYMBOL(zap_get_stats);
#endif