freebsd-dev/module/zfs/sa.c

2075 lines
53 KiB
C
Raw Normal View History

/*
* 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
*/
/*
* Copyright (c) 2010, Oracle and/or its affiliates. All rights reserved.
* Copyright (c) 2013 by Delphix. All rights reserved.
* Copyright (c) 2014 Spectra Logic Corporation, All rights reserved.
*/
#include <sys/zfs_context.h>
#include <sys/types.h>
#include <sys/param.h>
#include <sys/systm.h>
#include <sys/sysmacros.h>
#include <sys/dmu.h>
#include <sys/dmu_impl.h>
#include <sys/dmu_objset.h>
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
#include <sys/dmu_tx.h>
#include <sys/dbuf.h>
#include <sys/dnode.h>
#include <sys/zap.h>
#include <sys/sa.h>
#include <sys/sunddi.h>
#include <sys/sa_impl.h>
#include <sys/dnode.h>
#include <sys/errno.h>
#include <sys/zfs_context.h>
/*
* ZFS System attributes:
*
* A generic mechanism to allow for arbitrary attributes
* to be stored in a dnode. The data will be stored in the bonus buffer of
* the dnode and if necessary a special "spill" block will be used to handle
* overflow situations. The spill block will be sized to fit the data
* from 512 - 128K. When a spill block is used the BP (blkptr_t) for the
* spill block is stored at the end of the current bonus buffer. Any
* attributes that would be in the way of the blkptr_t will be relocated
* into the spill block.
*
* Attribute registration:
*
* Stored persistently on a per dataset basis
* a mapping between attribute "string" names and their actual attribute
* numeric values, length, and byteswap function. The names are only used
* during registration. All attributes are known by their unique attribute
* id value. If an attribute can have a variable size then the value
* 0 will be used to indicate this.
*
* Attribute Layout:
*
* Attribute layouts are a way to compactly store multiple attributes, but
* without taking the overhead associated with managing each attribute
* individually. Since you will typically have the same set of attributes
* stored in the same order a single table will be used to represent that
* layout. The ZPL for example will usually have only about 10 different
* layouts (regular files, device files, symlinks,
* regular files + scanstamp, files/dir with extended attributes, and then
* you have the possibility of all of those minus ACL, because it would
* be kicked out into the spill block)
*
* Layouts are simply an array of the attributes and their
* ordering i.e. [0, 1, 4, 5, 2]
*
* Each distinct layout is given a unique layout number and that is whats
* stored in the header at the beginning of the SA data buffer.
*
* A layout only covers a single dbuf (bonus or spill). If a set of
* attributes is split up between the bonus buffer and a spill buffer then
* two different layouts will be used. This allows us to byteswap the
* spill without looking at the bonus buffer and keeps the on disk format of
* the bonus and spill buffer the same.
*
* Adding a single attribute will cause the entire set of attributes to
* be rewritten and could result in a new layout number being constructed
* as part of the rewrite if no such layout exists for the new set of
* attribues. The new attribute will be appended to the end of the already
* existing attributes.
*
* Both the attribute registration and attribute layout information are
* stored in normal ZAP attributes. Their should be a small number of
* known layouts and the set of attributes is assumed to typically be quite
* small.
*
* The registered attributes and layout "table" information is maintained
* in core and a special "sa_os_t" is attached to the objset_t.
*
* A special interface is provided to allow for quickly applying
* a large set of attributes at once. sa_replace_all_by_template() is
* used to set an array of attributes. This is used by the ZPL when
* creating a brand new file. The template that is passed into the function
* specifies the attribute, size for variable length attributes, location of
* data and special "data locator" function if the data isn't in a contiguous
* location.
*
* Byteswap implications:
*
* Since the SA attributes are not entirely self describing we can't do
* the normal byteswap processing. The special ZAP layout attribute and
* attribute registration attributes define the byteswap function and the
* size of the attributes, unless it is variable sized.
* The normal ZFS byteswapping infrastructure assumes you don't need
* to read any objects in order to do the necessary byteswapping. Whereas
* SA attributes can only be properly byteswapped if the dataset is opened
* and the layout/attribute ZAP attributes are available. Because of this
* the SA attributes will be byteswapped when they are first accessed by
* the SA code that will read the SA data.
*/
typedef void (sa_iterfunc_t)(void *hdr, void *addr, sa_attr_type_t,
uint16_t length, int length_idx, boolean_t, void *userp);
static int sa_build_index(sa_handle_t *hdl, sa_buf_type_t buftype);
static void sa_idx_tab_hold(objset_t *os, sa_idx_tab_t *idx_tab);
static void *sa_find_idx_tab(objset_t *os, dmu_object_type_t bonustype,
void *data);
static void sa_idx_tab_rele(objset_t *os, void *arg);
static void sa_copy_data(sa_data_locator_t *func, void *start, void *target,
int buflen);
static int sa_modify_attrs(sa_handle_t *hdl, sa_attr_type_t newattr,
sa_data_op_t action, sa_data_locator_t *locator, void *datastart,
uint16_t buflen, dmu_tx_t *tx);
arc_byteswap_func_t sa_bswap_table[] = {
byteswap_uint64_array,
byteswap_uint32_array,
byteswap_uint16_array,
byteswap_uint8_array,
zfs_acl_byteswap,
};
#define SA_COPY_DATA(f, s, t, l) \
{ \
if (f == NULL) { \
if (l == 8) { \
*(uint64_t *)t = *(uint64_t *)s; \
} else if (l == 16) { \
*(uint64_t *)t = *(uint64_t *)s; \
*(uint64_t *)((uintptr_t)t + 8) = \
*(uint64_t *)((uintptr_t)s + 8); \
} else { \
bcopy(s, t, l); \
} \
} else \
sa_copy_data(f, s, t, l); \
}
/*
* This table is fixed and cannot be changed. Its purpose is to
* allow the SA code to work with both old/new ZPL file systems.
* It contains the list of legacy attributes. These attributes aren't
* stored in the "attribute" registry zap objects, since older ZPL file systems
* won't have the registry. Only objsets of type ZFS_TYPE_FILESYSTEM will
* use this static table.
*/
sa_attr_reg_t sa_legacy_attrs[] = {
{"ZPL_ATIME", sizeof (uint64_t) * 2, SA_UINT64_ARRAY, 0},
{"ZPL_MTIME", sizeof (uint64_t) * 2, SA_UINT64_ARRAY, 1},
{"ZPL_CTIME", sizeof (uint64_t) * 2, SA_UINT64_ARRAY, 2},
{"ZPL_CRTIME", sizeof (uint64_t) * 2, SA_UINT64_ARRAY, 3},
{"ZPL_GEN", sizeof (uint64_t), SA_UINT64_ARRAY, 4},
{"ZPL_MODE", sizeof (uint64_t), SA_UINT64_ARRAY, 5},
{"ZPL_SIZE", sizeof (uint64_t), SA_UINT64_ARRAY, 6},
{"ZPL_PARENT", sizeof (uint64_t), SA_UINT64_ARRAY, 7},
{"ZPL_LINKS", sizeof (uint64_t), SA_UINT64_ARRAY, 8},
{"ZPL_XATTR", sizeof (uint64_t), SA_UINT64_ARRAY, 9},
{"ZPL_RDEV", sizeof (uint64_t), SA_UINT64_ARRAY, 10},
{"ZPL_FLAGS", sizeof (uint64_t), SA_UINT64_ARRAY, 11},
{"ZPL_UID", sizeof (uint64_t), SA_UINT64_ARRAY, 12},
{"ZPL_GID", sizeof (uint64_t), SA_UINT64_ARRAY, 13},
{"ZPL_PAD", sizeof (uint64_t) * 4, SA_UINT64_ARRAY, 14},
{"ZPL_ZNODE_ACL", 88, SA_UINT8_ARRAY, 15},
};
/*
* This is only used for objects of type DMU_OT_ZNODE
*/
sa_attr_type_t sa_legacy_zpl_layout[] = {
0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15
};
/*
* Special dummy layout used for buffers with no attributes.
*/
sa_attr_type_t sa_dummy_zpl_layout[] = { 0 };
static int sa_legacy_attr_count = 16;
static kmem_cache_t *sa_cache = NULL;
/*ARGSUSED*/
static int
sa_cache_constructor(void *buf, void *unused, int kmflag)
{
sa_handle_t *hdl = buf;
mutex_init(&hdl->sa_lock, NULL, MUTEX_DEFAULT, NULL);
return (0);
}
/*ARGSUSED*/
static void
sa_cache_destructor(void *buf, void *unused)
{
sa_handle_t *hdl = buf;
mutex_destroy(&hdl->sa_lock);
}
void
sa_cache_init(void)
{
sa_cache = kmem_cache_create("sa_cache",
sizeof (sa_handle_t), 0, sa_cache_constructor,
sa_cache_destructor, NULL, NULL, NULL, 0);
}
void
sa_cache_fini(void)
{
if (sa_cache)
kmem_cache_destroy(sa_cache);
}
static int
layout_num_compare(const void *arg1, const void *arg2)
{
const sa_lot_t *node1 = arg1;
const sa_lot_t *node2 = arg2;
if (node1->lot_num > node2->lot_num)
return (1);
else if (node1->lot_num < node2->lot_num)
return (-1);
return (0);
}
static int
layout_hash_compare(const void *arg1, const void *arg2)
{
const sa_lot_t *node1 = arg1;
const sa_lot_t *node2 = arg2;
if (node1->lot_hash > node2->lot_hash)
return (1);
if (node1->lot_hash < node2->lot_hash)
return (-1);
if (node1->lot_instance > node2->lot_instance)
return (1);
if (node1->lot_instance < node2->lot_instance)
return (-1);
return (0);
}
boolean_t
sa_layout_equal(sa_lot_t *tbf, sa_attr_type_t *attrs, int count)
{
int i;
if (count != tbf->lot_attr_count)
return (1);
for (i = 0; i != count; i++) {
if (attrs[i] != tbf->lot_attrs[i])
return (1);
}
return (0);
}
#define SA_ATTR_HASH(attr) (zfs_crc64_table[(-1ULL ^ attr) & 0xFF])
static uint64_t
sa_layout_info_hash(sa_attr_type_t *attrs, int attr_count)
{
int i;
uint64_t crc = -1ULL;
for (i = 0; i != attr_count; i++)
crc ^= SA_ATTR_HASH(attrs[i]);
return (crc);
}
static int
sa_get_spill(sa_handle_t *hdl)
{
int rc;
if (hdl->sa_spill == NULL) {
if ((rc = dmu_spill_hold_existing(hdl->sa_bonus, NULL,
&hdl->sa_spill)) == 0)
VERIFY(0 == sa_build_index(hdl, SA_SPILL));
} else {
rc = 0;
}
return (rc);
}
/*
* Main attribute lookup/update function
* returns 0 for success or non zero for failures
*
* Operates on bulk array, first failure will abort further processing
*/
int
sa_attr_op(sa_handle_t *hdl, sa_bulk_attr_t *bulk, int count,
sa_data_op_t data_op, dmu_tx_t *tx)
{
sa_os_t *sa = hdl->sa_os->os_sa;
int i;
int error = 0;
sa_buf_type_t buftypes;
buftypes = 0;
ASSERT(count > 0);
for (i = 0; i != count; i++) {
ASSERT(bulk[i].sa_attr <= hdl->sa_os->os_sa->sa_num_attrs);
bulk[i].sa_addr = NULL;
/* First check the bonus buffer */
if (hdl->sa_bonus_tab && TOC_ATTR_PRESENT(
hdl->sa_bonus_tab->sa_idx_tab[bulk[i].sa_attr])) {
SA_ATTR_INFO(sa, hdl->sa_bonus_tab,
SA_GET_HDR(hdl, SA_BONUS),
bulk[i].sa_attr, bulk[i], SA_BONUS, hdl);
if (tx && !(buftypes & SA_BONUS)) {
dmu_buf_will_dirty(hdl->sa_bonus, tx);
buftypes |= SA_BONUS;
}
}
if (bulk[i].sa_addr == NULL &&
((error = sa_get_spill(hdl)) == 0)) {
if (TOC_ATTR_PRESENT(
hdl->sa_spill_tab->sa_idx_tab[bulk[i].sa_attr])) {
SA_ATTR_INFO(sa, hdl->sa_spill_tab,
SA_GET_HDR(hdl, SA_SPILL),
bulk[i].sa_attr, bulk[i], SA_SPILL, hdl);
if (tx && !(buftypes & SA_SPILL) &&
bulk[i].sa_size == bulk[i].sa_length) {
dmu_buf_will_dirty(hdl->sa_spill, tx);
buftypes |= SA_SPILL;
}
}
}
if (error && error != ENOENT) {
return ((error == ECKSUM) ? EIO : error);
}
switch (data_op) {
case SA_LOOKUP:
if (bulk[i].sa_addr == NULL)
return (SET_ERROR(ENOENT));
if (bulk[i].sa_data) {
SA_COPY_DATA(bulk[i].sa_data_func,
bulk[i].sa_addr, bulk[i].sa_data,
bulk[i].sa_size);
}
continue;
case SA_UPDATE:
/* existing rewrite of attr */
if (bulk[i].sa_addr &&
bulk[i].sa_size == bulk[i].sa_length) {
SA_COPY_DATA(bulk[i].sa_data_func,
bulk[i].sa_data, bulk[i].sa_addr,
bulk[i].sa_length);
continue;
} else if (bulk[i].sa_addr) { /* attr size change */
error = sa_modify_attrs(hdl, bulk[i].sa_attr,
SA_REPLACE, bulk[i].sa_data_func,
bulk[i].sa_data, bulk[i].sa_length, tx);
} else { /* adding new attribute */
error = sa_modify_attrs(hdl, bulk[i].sa_attr,
SA_ADD, bulk[i].sa_data_func,
bulk[i].sa_data, bulk[i].sa_length, tx);
}
if (error)
return (error);
break;
default:
break;
}
}
return (error);
}
static sa_lot_t *
sa_add_layout_entry(objset_t *os, sa_attr_type_t *attrs, int attr_count,
uint64_t lot_num, uint64_t hash, boolean_t zapadd, dmu_tx_t *tx)
{
sa_os_t *sa = os->os_sa;
sa_lot_t *tb, *findtb;
int i;
avl_index_t loc;
ASSERT(MUTEX_HELD(&sa->sa_lock));
tb = kmem_zalloc(sizeof (sa_lot_t), KM_SLEEP);
tb->lot_attr_count = attr_count;
tb->lot_attrs = kmem_alloc(sizeof (sa_attr_type_t) * attr_count,
KM_SLEEP);
bcopy(attrs, tb->lot_attrs, sizeof (sa_attr_type_t) * attr_count);
tb->lot_num = lot_num;
tb->lot_hash = hash;
tb->lot_instance = 0;
if (zapadd) {
char attr_name[8];
if (sa->sa_layout_attr_obj == 0) {
sa->sa_layout_attr_obj = zap_create_link(os,
DMU_OT_SA_ATTR_LAYOUTS,
sa->sa_master_obj, SA_LAYOUTS, tx);
}
(void) snprintf(attr_name, sizeof (attr_name),
"%d", (int)lot_num);
VERIFY(0 == zap_update(os, os->os_sa->sa_layout_attr_obj,
attr_name, 2, attr_count, attrs, tx));
}
list_create(&tb->lot_idx_tab, sizeof (sa_idx_tab_t),
offsetof(sa_idx_tab_t, sa_next));
for (i = 0; i != attr_count; i++) {
if (sa->sa_attr_table[tb->lot_attrs[i]].sa_length == 0)
tb->lot_var_sizes++;
}
avl_add(&sa->sa_layout_num_tree, tb);
/* verify we don't have a hash collision */
if ((findtb = avl_find(&sa->sa_layout_hash_tree, tb, &loc)) != NULL) {
for (; findtb && findtb->lot_hash == hash;
findtb = AVL_NEXT(&sa->sa_layout_hash_tree, findtb)) {
if (findtb->lot_instance != tb->lot_instance)
break;
tb->lot_instance++;
}
}
avl_add(&sa->sa_layout_hash_tree, tb);
return (tb);
}
static void
sa_find_layout(objset_t *os, uint64_t hash, sa_attr_type_t *attrs,
int count, dmu_tx_t *tx, sa_lot_t **lot)
{
sa_lot_t *tb, tbsearch;
avl_index_t loc;
sa_os_t *sa = os->os_sa;
boolean_t found = B_FALSE;
mutex_enter(&sa->sa_lock);
tbsearch.lot_hash = hash;
tbsearch.lot_instance = 0;
tb = avl_find(&sa->sa_layout_hash_tree, &tbsearch, &loc);
if (tb) {
for (; tb && tb->lot_hash == hash;
tb = AVL_NEXT(&sa->sa_layout_hash_tree, tb)) {
if (sa_layout_equal(tb, attrs, count) == 0) {
found = B_TRUE;
break;
}
}
}
if (!found) {
tb = sa_add_layout_entry(os, attrs, count,
avl_numnodes(&sa->sa_layout_num_tree), hash, B_TRUE, tx);
}
mutex_exit(&sa->sa_lock);
*lot = tb;
}
static int
sa_resize_spill(sa_handle_t *hdl, uint32_t size, dmu_tx_t *tx)
{
int error;
uint32_t blocksize;
if (size == 0) {
blocksize = SPA_MINBLOCKSIZE;
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
} else if (size > SPA_OLD_MAXBLOCKSIZE) {
ASSERT(0);
return (SET_ERROR(EFBIG));
} else {
blocksize = P2ROUNDUP_TYPED(size, SPA_MINBLOCKSIZE, uint32_t);
}
error = dbuf_spill_set_blksz(hdl->sa_spill, blocksize, tx);
ASSERT(error == 0);
return (error);
}
static void
sa_copy_data(sa_data_locator_t *func, void *datastart, void *target, int buflen)
{
if (func == NULL) {
bcopy(datastart, target, buflen);
} else {
boolean_t start;
int bytes;
void *dataptr;
void *saptr = target;
uint32_t length;
start = B_TRUE;
bytes = 0;
while (bytes < buflen) {
func(&dataptr, &length, buflen, start, datastart);
bcopy(dataptr, saptr, length);
saptr = (void *)((caddr_t)saptr + length);
bytes += length;
start = B_FALSE;
}
}
}
/*
* Determine several different values pertaining to system attribute
* buffers.
*
* Return the size of the sa_hdr_phys_t header for the buffer. Each
* variable length attribute except the first contributes two bytes to
* the header size, which is then rounded up to an 8-byte boundary.
*
* The following output parameters are also computed.
*
* index - The index of the first attribute in attr_desc that will
* spill over. Only valid if will_spill is set.
*
* total - The total number of bytes of all system attributes described
* in attr_desc.
*
* will_spill - Set when spilling is necessary. It is only set when
* the buftype is SA_BONUS.
*/
static int
sa_find_sizes(sa_os_t *sa, sa_bulk_attr_t *attr_desc, int attr_count,
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
dmu_buf_t *db, sa_buf_type_t buftype, int full_space, int *index,
int *total, boolean_t *will_spill)
{
int var_size_count = 0;
int i;
int hdrsize;
2013-12-06 22:16:40 +00:00
int extra_hdrsize;
if (buftype == SA_BONUS && sa->sa_force_spill) {
*total = 0;
*index = 0;
*will_spill = B_TRUE;
return (0);
}
*index = -1;
*total = 0;
2013-12-06 22:16:40 +00:00
*will_spill = B_FALSE;
2013-12-06 22:16:40 +00:00
extra_hdrsize = 0;
hdrsize = (SA_BONUSTYPE_FROM_DB(db) == DMU_OT_ZNODE) ? 0 :
sizeof (sa_hdr_phys_t);
Fix rounding discrepancy in sa_find_sizes() A rounding discrepancy exists between how sa_build_layouts() and sa_find_sizes() calculate when the spill block needs to be kicked in. This results in a narrow size range where sa_build_layouts() believes there must be a spill block allocated but due to the discrepancy there isn't. A panic then occurs when the hdl->sa_spill NULL pointer is dereferenced. The following reproducer for this bug was isolated: truncate -s 128m /tmp/tank zpool create tank /tmp/tank zfs create -o xattr=sa tank/fish ln -s `perl -e 'print "z" x 41'` /tank/fish/z setfattr -hn trusted.foo -v`perl -e 'print "z"x45'` /tank/fish/z This test results in roughly the following system attribute (SA) layout: 176 bytes - "standard" SA's 41 bytes - name of symbolic link target 100 bytes - XDR encoded nvlist for xattr --- 317 bytes - total Because 317 is less than DN_MAX_BONUSLEN (320), sa_find_sizes() decides no spill block is needed. But sa_build_layouts() rounds 41 up to 48 when computing the space requirements so it tries to switch to the spill block. Note that we were only able to reproduce this bug using a combination of symbolic links and the Linux-specific xattr=sa dataset property. So while this issue is not technically Linux-specific, it may be difficult or impossible to hit the narrow size range needed to reproduce it on other platforms. To fix the discrepancy, round the running total in sa_find_sizes() up to an 8-byte boundary before accounting for each SA, since this is how they will be stored in the bonus and (possibly) spill buffers. To make the intent of the code more clear, explicitly assert key assumptions about expected alignment of data and whether spill-over will occur. Signed-off-by: Matthew Ahrens <mahrens@delphix.com Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov> Closes #1240
2013-01-29 23:49:15 +00:00
ASSERT(IS_P2ALIGNED(full_space, 8));
for (i = 0; i != attr_count; i++) {
boolean_t is_var_sz, might_spill_here;
int tmp_hdrsize;
Fix rounding discrepancy in sa_find_sizes() A rounding discrepancy exists between how sa_build_layouts() and sa_find_sizes() calculate when the spill block needs to be kicked in. This results in a narrow size range where sa_build_layouts() believes there must be a spill block allocated but due to the discrepancy there isn't. A panic then occurs when the hdl->sa_spill NULL pointer is dereferenced. The following reproducer for this bug was isolated: truncate -s 128m /tmp/tank zpool create tank /tmp/tank zfs create -o xattr=sa tank/fish ln -s `perl -e 'print "z" x 41'` /tank/fish/z setfattr -hn trusted.foo -v`perl -e 'print "z"x45'` /tank/fish/z This test results in roughly the following system attribute (SA) layout: 176 bytes - "standard" SA's 41 bytes - name of symbolic link target 100 bytes - XDR encoded nvlist for xattr --- 317 bytes - total Because 317 is less than DN_MAX_BONUSLEN (320), sa_find_sizes() decides no spill block is needed. But sa_build_layouts() rounds 41 up to 48 when computing the space requirements so it tries to switch to the spill block. Note that we were only able to reproduce this bug using a combination of symbolic links and the Linux-specific xattr=sa dataset property. So while this issue is not technically Linux-specific, it may be difficult or impossible to hit the narrow size range needed to reproduce it on other platforms. To fix the discrepancy, round the running total in sa_find_sizes() up to an 8-byte boundary before accounting for each SA, since this is how they will be stored in the bonus and (possibly) spill buffers. To make the intent of the code more clear, explicitly assert key assumptions about expected alignment of data and whether spill-over will occur. Signed-off-by: Matthew Ahrens <mahrens@delphix.com Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov> Closes #1240
2013-01-29 23:49:15 +00:00
*total = P2ROUNDUP(*total, 8);
*total += attr_desc[i].sa_length;
2013-12-06 22:16:40 +00:00
if (*will_spill)
continue;
is_var_sz = (SA_REGISTERED_LEN(sa, attr_desc[i].sa_attr) == 0);
if (is_var_sz)
var_size_count++;
/*
* Calculate what the SA header size would be if this
* attribute doesn't spill.
*/
tmp_hdrsize = hdrsize + ((is_var_sz && var_size_count > 1) ?
sizeof (uint16_t) : 0);
/*
* Check whether this attribute spans into the space
* that would be used by the spill block pointer should
* a spill block be needed.
*/
might_spill_here =
buftype == SA_BONUS && *index == -1 &&
(*total + P2ROUNDUP(tmp_hdrsize, 8)) >
(full_space - sizeof (blkptr_t));
if (is_var_sz && var_size_count > 1) {
2013-12-06 22:16:40 +00:00
if (buftype == SA_SPILL ||
tmp_hdrsize + *total < full_space) {
Fix mismatch between SA header size and layout When a system attribute layout is created an inconsistency may occur between the system attribute header (sa_hdr_phys_t) size and the variable-sized attribute count stored in the layout. The inconsistency results in the following failed assertion when SA_HDR_SIZE_MATCH_LAYOUT returns false: SPLError: 11315:0:(sa.c:1541:sa_find_idx_tab()) ASSERTION((IS_SA_BONUSTYPE(bonustype) && SA_HDR_SIZE_MATCH_LAYOUT(hdr, tb)) || !IS_SA_BONUSTYPE(bonustype) || (IS_SA_BONUSTYPE(bonustype) && hdr->sa_layout_info == 0)) failed The bug originates in this snippet from sa_find_sizes(). if (is_var_sz && var_size > 1) { if (P2ROUNDUP(hdrsize + sizeof (uint16_t), *total < full_space) { hdrsize += sizeof (uint16_t); This assumes that the current variable-sized attribute will be stored in the current buffer and accounts for the space needed to store its size in the sa_hdr_phys_t. However if the next attribute spills over we need to store a blkptr_t at the end of the bonus buffer to point to the spill block. If the current attribute is in the way of the blkptr_t then it too will be relocated into the spill block. But since we've already accounted for it in the header size we get the inconsistency described above. To avoid this, record the index of the last variable-sized attribute that prompted a hdrsize increase, and reverse the increase if we later determine that that attribute will be relocated to the spill block. Signed-off-by: Matthew Ahrens <mahrens@delphix.com> Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov> Closes #1250
2013-01-30 17:48:57 +00:00
/*
2013-12-06 22:16:40 +00:00
* Record the extra header size in case this
* increase needs to be reversed due to
* spill-over.
Fix mismatch between SA header size and layout When a system attribute layout is created an inconsistency may occur between the system attribute header (sa_hdr_phys_t) size and the variable-sized attribute count stored in the layout. The inconsistency results in the following failed assertion when SA_HDR_SIZE_MATCH_LAYOUT returns false: SPLError: 11315:0:(sa.c:1541:sa_find_idx_tab()) ASSERTION((IS_SA_BONUSTYPE(bonustype) && SA_HDR_SIZE_MATCH_LAYOUT(hdr, tb)) || !IS_SA_BONUSTYPE(bonustype) || (IS_SA_BONUSTYPE(bonustype) && hdr->sa_layout_info == 0)) failed The bug originates in this snippet from sa_find_sizes(). if (is_var_sz && var_size > 1) { if (P2ROUNDUP(hdrsize + sizeof (uint16_t), *total < full_space) { hdrsize += sizeof (uint16_t); This assumes that the current variable-sized attribute will be stored in the current buffer and accounts for the space needed to store its size in the sa_hdr_phys_t. However if the next attribute spills over we need to store a blkptr_t at the end of the bonus buffer to point to the spill block. If the current attribute is in the way of the blkptr_t then it too will be relocated into the spill block. But since we've already accounted for it in the header size we get the inconsistency described above. To avoid this, record the index of the last variable-sized attribute that prompted a hdrsize increase, and reverse the increase if we later determine that that attribute will be relocated to the spill block. Signed-off-by: Matthew Ahrens <mahrens@delphix.com> Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov> Closes #1250
2013-01-30 17:48:57 +00:00
*/
hdrsize = tmp_hdrsize;
if (*index != -1 || might_spill_here)
2013-12-06 22:16:40 +00:00
extra_hdrsize += sizeof (uint16_t);
} else {
2013-12-06 22:16:40 +00:00
ASSERT(buftype == SA_BONUS);
if (*index == -1)
*index = i;
*will_spill = B_TRUE;
continue;
}
}
/*
* Store index of where spill *could* occur. Then
* continue to count the remaining attribute sizes. The
* sum is used later for sizing bonus and spill buffer.
*/
if (might_spill_here)
*index = i;
if ((*total + P2ROUNDUP(hdrsize, 8)) > full_space &&
buftype == SA_BONUS)
*will_spill = B_TRUE;
}
2013-12-06 22:16:40 +00:00
if (*will_spill)
hdrsize -= extra_hdrsize;
Fix mismatch between SA header size and layout When a system attribute layout is created an inconsistency may occur between the system attribute header (sa_hdr_phys_t) size and the variable-sized attribute count stored in the layout. The inconsistency results in the following failed assertion when SA_HDR_SIZE_MATCH_LAYOUT returns false: SPLError: 11315:0:(sa.c:1541:sa_find_idx_tab()) ASSERTION((IS_SA_BONUSTYPE(bonustype) && SA_HDR_SIZE_MATCH_LAYOUT(hdr, tb)) || !IS_SA_BONUSTYPE(bonustype) || (IS_SA_BONUSTYPE(bonustype) && hdr->sa_layout_info == 0)) failed The bug originates in this snippet from sa_find_sizes(). if (is_var_sz && var_size > 1) { if (P2ROUNDUP(hdrsize + sizeof (uint16_t), *total < full_space) { hdrsize += sizeof (uint16_t); This assumes that the current variable-sized attribute will be stored in the current buffer and accounts for the space needed to store its size in the sa_hdr_phys_t. However if the next attribute spills over we need to store a blkptr_t at the end of the bonus buffer to point to the spill block. If the current attribute is in the way of the blkptr_t then it too will be relocated into the spill block. But since we've already accounted for it in the header size we get the inconsistency described above. To avoid this, record the index of the last variable-sized attribute that prompted a hdrsize increase, and reverse the increase if we later determine that that attribute will be relocated to the spill block. Signed-off-by: Matthew Ahrens <mahrens@delphix.com> Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov> Closes #1250
2013-01-30 17:48:57 +00:00
hdrsize = P2ROUNDUP(hdrsize, 8);
return (hdrsize);
}
#define BUF_SPACE_NEEDED(total, header) (total + header)
/*
* Find layout that corresponds to ordering of attributes
* If not found a new layout number is created and added to
* persistent layout tables.
*/
static int
sa_build_layouts(sa_handle_t *hdl, sa_bulk_attr_t *attr_desc, int attr_count,
dmu_tx_t *tx)
{
sa_os_t *sa = hdl->sa_os->os_sa;
uint64_t hash;
sa_buf_type_t buftype;
sa_hdr_phys_t *sahdr;
void *data_start;
sa_attr_type_t *attrs, *attrs_start;
int i, lot_count;
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
int dnodesize;
int spill_idx;
int hdrsize;
int spillhdrsize = 0;
int used;
dmu_object_type_t bonustype;
sa_lot_t *lot;
int len_idx;
int spill_used;
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
int bonuslen;
boolean_t spilling;
dmu_buf_will_dirty(hdl->sa_bonus, tx);
bonustype = SA_BONUSTYPE_FROM_DB(hdl->sa_bonus);
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
dmu_object_dnsize_from_db(hdl->sa_bonus, &dnodesize);
bonuslen = DN_BONUS_SIZE(dnodesize);
/* first determine bonus header size and sum of all attributes */
hdrsize = sa_find_sizes(sa, attr_desc, attr_count, hdl->sa_bonus,
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
SA_BONUS, bonuslen, &spill_idx, &used, &spilling);
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
if (used > SPA_OLD_MAXBLOCKSIZE)
return (SET_ERROR(EFBIG));
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
VERIFY0(dmu_set_bonus(hdl->sa_bonus, spilling ?
MIN(bonuslen - sizeof (blkptr_t), used + hdrsize) :
used + hdrsize, tx));
ASSERT((bonustype == DMU_OT_ZNODE && spilling == 0) ||
bonustype == DMU_OT_SA);
/* setup and size spill buffer when needed */
if (spilling) {
boolean_t dummy;
if (hdl->sa_spill == NULL) {
VERIFY(dmu_spill_hold_by_bonus(hdl->sa_bonus, NULL,
&hdl->sa_spill) == 0);
}
dmu_buf_will_dirty(hdl->sa_spill, tx);
spillhdrsize = sa_find_sizes(sa, &attr_desc[spill_idx],
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
attr_count - spill_idx, hdl->sa_spill, SA_SPILL,
hdl->sa_spill->db_size, &i, &spill_used, &dummy);
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
if (spill_used > SPA_OLD_MAXBLOCKSIZE)
return (SET_ERROR(EFBIG));
if (BUF_SPACE_NEEDED(spill_used, spillhdrsize) >
hdl->sa_spill->db_size)
VERIFY(0 == sa_resize_spill(hdl,
BUF_SPACE_NEEDED(spill_used, spillhdrsize), tx));
}
/* setup starting pointers to lay down data */
data_start = (void *)((uintptr_t)hdl->sa_bonus->db_data + hdrsize);
sahdr = (sa_hdr_phys_t *)hdl->sa_bonus->db_data;
buftype = SA_BONUS;
attrs_start = attrs = kmem_alloc(sizeof (sa_attr_type_t) * attr_count,
KM_SLEEP);
lot_count = 0;
for (i = 0, len_idx = 0, hash = -1ULL; i != attr_count; i++) {
uint16_t length;
Fix rounding discrepancy in sa_find_sizes() A rounding discrepancy exists between how sa_build_layouts() and sa_find_sizes() calculate when the spill block needs to be kicked in. This results in a narrow size range where sa_build_layouts() believes there must be a spill block allocated but due to the discrepancy there isn't. A panic then occurs when the hdl->sa_spill NULL pointer is dereferenced. The following reproducer for this bug was isolated: truncate -s 128m /tmp/tank zpool create tank /tmp/tank zfs create -o xattr=sa tank/fish ln -s `perl -e 'print "z" x 41'` /tank/fish/z setfattr -hn trusted.foo -v`perl -e 'print "z"x45'` /tank/fish/z This test results in roughly the following system attribute (SA) layout: 176 bytes - "standard" SA's 41 bytes - name of symbolic link target 100 bytes - XDR encoded nvlist for xattr --- 317 bytes - total Because 317 is less than DN_MAX_BONUSLEN (320), sa_find_sizes() decides no spill block is needed. But sa_build_layouts() rounds 41 up to 48 when computing the space requirements so it tries to switch to the spill block. Note that we were only able to reproduce this bug using a combination of symbolic links and the Linux-specific xattr=sa dataset property. So while this issue is not technically Linux-specific, it may be difficult or impossible to hit the narrow size range needed to reproduce it on other platforms. To fix the discrepancy, round the running total in sa_find_sizes() up to an 8-byte boundary before accounting for each SA, since this is how they will be stored in the bonus and (possibly) spill buffers. To make the intent of the code more clear, explicitly assert key assumptions about expected alignment of data and whether spill-over will occur. Signed-off-by: Matthew Ahrens <mahrens@delphix.com Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov> Closes #1240
2013-01-29 23:49:15 +00:00
ASSERT(IS_P2ALIGNED(data_start, 8));
attrs[i] = attr_desc[i].sa_attr;
length = SA_REGISTERED_LEN(sa, attrs[i]);
if (length == 0)
length = attr_desc[i].sa_length;
if (spilling && i == spill_idx) { /* switch to spill buffer */
VERIFY(bonustype == DMU_OT_SA);
if (buftype == SA_BONUS && !sa->sa_force_spill) {
sa_find_layout(hdl->sa_os, hash, attrs_start,
lot_count, tx, &lot);
SA_SET_HDR(sahdr, lot->lot_num, hdrsize);
}
buftype = SA_SPILL;
hash = -1ULL;
len_idx = 0;
sahdr = (sa_hdr_phys_t *)hdl->sa_spill->db_data;
sahdr->sa_magic = SA_MAGIC;
data_start = (void *)((uintptr_t)sahdr +
spillhdrsize);
attrs_start = &attrs[i];
lot_count = 0;
}
hash ^= SA_ATTR_HASH(attrs[i]);
attr_desc[i].sa_addr = data_start;
attr_desc[i].sa_size = length;
SA_COPY_DATA(attr_desc[i].sa_data_func, attr_desc[i].sa_data,
data_start, length);
if (sa->sa_attr_table[attrs[i]].sa_length == 0) {
sahdr->sa_lengths[len_idx++] = length;
}
data_start = (void *)P2ROUNDUP(((uintptr_t)data_start +
length), 8);
lot_count++;
}
sa_find_layout(hdl->sa_os, hash, attrs_start, lot_count, tx, &lot);
/*
* Verify that old znodes always have layout number 0.
* Must be DMU_OT_SA for arbitrary layouts
*/
VERIFY((bonustype == DMU_OT_ZNODE && lot->lot_num == 0) ||
(bonustype == DMU_OT_SA && lot->lot_num > 1));
if (bonustype == DMU_OT_SA) {
SA_SET_HDR(sahdr, lot->lot_num,
buftype == SA_BONUS ? hdrsize : spillhdrsize);
}
kmem_free(attrs, sizeof (sa_attr_type_t) * attr_count);
if (hdl->sa_bonus_tab) {
sa_idx_tab_rele(hdl->sa_os, hdl->sa_bonus_tab);
hdl->sa_bonus_tab = NULL;
}
if (!sa->sa_force_spill)
VERIFY(0 == sa_build_index(hdl, SA_BONUS));
if (hdl->sa_spill) {
sa_idx_tab_rele(hdl->sa_os, hdl->sa_spill_tab);
if (!spilling) {
/*
* remove spill block that is no longer needed.
*/
dmu_buf_rele(hdl->sa_spill, NULL);
hdl->sa_spill = NULL;
hdl->sa_spill_tab = NULL;
VERIFY(0 == dmu_rm_spill(hdl->sa_os,
sa_handle_object(hdl), tx));
} else {
VERIFY(0 == sa_build_index(hdl, SA_SPILL));
}
}
return (0);
}
static void
sa_free_attr_table(sa_os_t *sa)
{
int i;
if (sa->sa_attr_table == NULL)
return;
for (i = 0; i != sa->sa_num_attrs; i++) {
if (sa->sa_attr_table[i].sa_name)
kmem_free(sa->sa_attr_table[i].sa_name,
strlen(sa->sa_attr_table[i].sa_name) + 1);
}
kmem_free(sa->sa_attr_table,
sizeof (sa_attr_table_t) * sa->sa_num_attrs);
sa->sa_attr_table = NULL;
}
static int
sa_attr_table_setup(objset_t *os, sa_attr_reg_t *reg_attrs, int count)
{
sa_os_t *sa = os->os_sa;
uint64_t sa_attr_count = 0;
uint64_t sa_reg_count = 0;
int error = 0;
uint64_t attr_value;
sa_attr_table_t *tb;
zap_cursor_t zc;
zap_attribute_t za;
int registered_count = 0;
int i;
dmu_objset_type_t ostype = dmu_objset_type(os);
sa->sa_user_table =
kmem_zalloc(count * sizeof (sa_attr_type_t), KM_SLEEP);
sa->sa_user_table_sz = count * sizeof (sa_attr_type_t);
if (sa->sa_reg_attr_obj != 0) {
error = zap_count(os, sa->sa_reg_attr_obj,
&sa_attr_count);
/*
* Make sure we retrieved a count and that it isn't zero
*/
if (error || (error == 0 && sa_attr_count == 0)) {
if (error == 0)
error = SET_ERROR(EINVAL);
goto bail;
}
sa_reg_count = sa_attr_count;
}
if (ostype == DMU_OST_ZFS && sa_attr_count == 0)
sa_attr_count += sa_legacy_attr_count;
/* Allocate attribute numbers for attributes that aren't registered */
for (i = 0; i != count; i++) {
boolean_t found = B_FALSE;
int j;
if (ostype == DMU_OST_ZFS) {
for (j = 0; j != sa_legacy_attr_count; j++) {
if (strcmp(reg_attrs[i].sa_name,
sa_legacy_attrs[j].sa_name) == 0) {
sa->sa_user_table[i] =
sa_legacy_attrs[j].sa_attr;
found = B_TRUE;
}
}
}
if (found)
continue;
if (sa->sa_reg_attr_obj)
error = zap_lookup(os, sa->sa_reg_attr_obj,
reg_attrs[i].sa_name, 8, 1, &attr_value);
else
error = SET_ERROR(ENOENT);
switch (error) {
case ENOENT:
sa->sa_user_table[i] = (sa_attr_type_t)sa_attr_count;
sa_attr_count++;
break;
case 0:
sa->sa_user_table[i] = ATTR_NUM(attr_value);
break;
default:
goto bail;
}
}
sa->sa_num_attrs = sa_attr_count;
tb = sa->sa_attr_table =
kmem_zalloc(sizeof (sa_attr_table_t) * sa_attr_count, KM_SLEEP);
/*
* Attribute table is constructed from requested attribute list,
* previously foreign registered attributes, and also the legacy
* ZPL set of attributes.
*/
if (sa->sa_reg_attr_obj) {
for (zap_cursor_init(&zc, os, sa->sa_reg_attr_obj);
(error = zap_cursor_retrieve(&zc, &za)) == 0;
zap_cursor_advance(&zc)) {
uint64_t value;
value = za.za_first_integer;
registered_count++;
tb[ATTR_NUM(value)].sa_attr = ATTR_NUM(value);
tb[ATTR_NUM(value)].sa_length = ATTR_LENGTH(value);
tb[ATTR_NUM(value)].sa_byteswap = ATTR_BSWAP(value);
tb[ATTR_NUM(value)].sa_registered = B_TRUE;
if (tb[ATTR_NUM(value)].sa_name) {
continue;
}
tb[ATTR_NUM(value)].sa_name =
kmem_zalloc(strlen(za.za_name) +1, KM_SLEEP);
(void) strlcpy(tb[ATTR_NUM(value)].sa_name, za.za_name,
strlen(za.za_name) +1);
}
zap_cursor_fini(&zc);
/*
* Make sure we processed the correct number of registered
* attributes
*/
if (registered_count != sa_reg_count) {
ASSERT(error != 0);
goto bail;
}
}
if (ostype == DMU_OST_ZFS) {
for (i = 0; i != sa_legacy_attr_count; i++) {
if (tb[i].sa_name)
continue;
tb[i].sa_attr = sa_legacy_attrs[i].sa_attr;
tb[i].sa_length = sa_legacy_attrs[i].sa_length;
tb[i].sa_byteswap = sa_legacy_attrs[i].sa_byteswap;
tb[i].sa_registered = B_FALSE;
tb[i].sa_name =
kmem_zalloc(strlen(sa_legacy_attrs[i].sa_name) +1,
KM_SLEEP);
(void) strlcpy(tb[i].sa_name,
sa_legacy_attrs[i].sa_name,
strlen(sa_legacy_attrs[i].sa_name) + 1);
}
}
for (i = 0; i != count; i++) {
sa_attr_type_t attr_id;
attr_id = sa->sa_user_table[i];
if (tb[attr_id].sa_name)
continue;
tb[attr_id].sa_length = reg_attrs[i].sa_length;
tb[attr_id].sa_byteswap = reg_attrs[i].sa_byteswap;
tb[attr_id].sa_attr = attr_id;
tb[attr_id].sa_name =
kmem_zalloc(strlen(reg_attrs[i].sa_name) + 1, KM_SLEEP);
(void) strlcpy(tb[attr_id].sa_name, reg_attrs[i].sa_name,
strlen(reg_attrs[i].sa_name) + 1);
}
sa->sa_need_attr_registration =
(sa_attr_count != registered_count);
return (0);
bail:
kmem_free(sa->sa_user_table, count * sizeof (sa_attr_type_t));
sa->sa_user_table = NULL;
sa_free_attr_table(sa);
return ((error != 0) ? error : EINVAL);
}
int
sa_setup(objset_t *os, uint64_t sa_obj, sa_attr_reg_t *reg_attrs, int count,
sa_attr_type_t **user_table)
{
zap_cursor_t zc;
zap_attribute_t za;
sa_os_t *sa;
dmu_objset_type_t ostype = dmu_objset_type(os);
sa_attr_type_t *tb;
int error;
mutex_enter(&os->os_user_ptr_lock);
if (os->os_sa) {
mutex_enter(&os->os_sa->sa_lock);
mutex_exit(&os->os_user_ptr_lock);
tb = os->os_sa->sa_user_table;
mutex_exit(&os->os_sa->sa_lock);
*user_table = tb;
return (0);
}
sa = kmem_zalloc(sizeof (sa_os_t), KM_SLEEP);
mutex_init(&sa->sa_lock, NULL, MUTEX_DEFAULT, NULL);
sa->sa_master_obj = sa_obj;
os->os_sa = sa;
mutex_enter(&sa->sa_lock);
mutex_exit(&os->os_user_ptr_lock);
avl_create(&sa->sa_layout_num_tree, layout_num_compare,
sizeof (sa_lot_t), offsetof(sa_lot_t, lot_num_node));
avl_create(&sa->sa_layout_hash_tree, layout_hash_compare,
sizeof (sa_lot_t), offsetof(sa_lot_t, lot_hash_node));
if (sa_obj) {
error = zap_lookup(os, sa_obj, SA_LAYOUTS,
8, 1, &sa->sa_layout_attr_obj);
if (error != 0 && error != ENOENT)
goto fail;
error = zap_lookup(os, sa_obj, SA_REGISTRY,
8, 1, &sa->sa_reg_attr_obj);
if (error != 0 && error != ENOENT)
goto fail;
}
if ((error = sa_attr_table_setup(os, reg_attrs, count)) != 0)
goto fail;
if (sa->sa_layout_attr_obj != 0) {
uint64_t layout_count;
error = zap_count(os, sa->sa_layout_attr_obj,
&layout_count);
/*
* Layout number count should be > 0
*/
if (error || (error == 0 && layout_count == 0)) {
if (error == 0)
error = SET_ERROR(EINVAL);
goto fail;
}
for (zap_cursor_init(&zc, os, sa->sa_layout_attr_obj);
(error = zap_cursor_retrieve(&zc, &za)) == 0;
zap_cursor_advance(&zc)) {
sa_attr_type_t *lot_attrs;
uint64_t lot_num;
lot_attrs = kmem_zalloc(sizeof (sa_attr_type_t) *
za.za_num_integers, KM_SLEEP);
if ((error = (zap_lookup(os, sa->sa_layout_attr_obj,
za.za_name, 2, za.za_num_integers,
lot_attrs))) != 0) {
kmem_free(lot_attrs, sizeof (sa_attr_type_t) *
za.za_num_integers);
break;
}
VERIFY(ddi_strtoull(za.za_name, NULL, 10,
(unsigned long long *)&lot_num) == 0);
(void) sa_add_layout_entry(os, lot_attrs,
za.za_num_integers, lot_num,
sa_layout_info_hash(lot_attrs,
za.za_num_integers), B_FALSE, NULL);
kmem_free(lot_attrs, sizeof (sa_attr_type_t) *
za.za_num_integers);
}
zap_cursor_fini(&zc);
/*
* Make sure layout count matches number of entries added
* to AVL tree
*/
if (avl_numnodes(&sa->sa_layout_num_tree) != layout_count) {
ASSERT(error != 0);
goto fail;
}
}
/* Add special layout number for old ZNODES */
if (ostype == DMU_OST_ZFS) {
(void) sa_add_layout_entry(os, sa_legacy_zpl_layout,
sa_legacy_attr_count, 0,
sa_layout_info_hash(sa_legacy_zpl_layout,
sa_legacy_attr_count), B_FALSE, NULL);
(void) sa_add_layout_entry(os, sa_dummy_zpl_layout, 0, 1,
0, B_FALSE, NULL);
}
*user_table = os->os_sa->sa_user_table;
mutex_exit(&sa->sa_lock);
return (0);
fail:
os->os_sa = NULL;
sa_free_attr_table(sa);
if (sa->sa_user_table)
kmem_free(sa->sa_user_table, sa->sa_user_table_sz);
mutex_exit(&sa->sa_lock);
avl_destroy(&sa->sa_layout_hash_tree);
avl_destroy(&sa->sa_layout_num_tree);
mutex_destroy(&sa->sa_lock);
kmem_free(sa, sizeof (sa_os_t));
return ((error == ECKSUM) ? EIO : error);
}
void
sa_tear_down(objset_t *os)
{
sa_os_t *sa = os->os_sa;
sa_lot_t *layout;
void *cookie;
kmem_free(sa->sa_user_table, sa->sa_user_table_sz);
/* Free up attr table */
sa_free_attr_table(sa);
cookie = NULL;
while ((layout =
avl_destroy_nodes(&sa->sa_layout_hash_tree, &cookie))) {
sa_idx_tab_t *tab;
while ((tab = list_head(&layout->lot_idx_tab))) {
ASSERT(refcount_count(&tab->sa_refcount));
sa_idx_tab_rele(os, tab);
}
}
cookie = NULL;
while ((layout = avl_destroy_nodes(&sa->sa_layout_num_tree, &cookie))) {
kmem_free(layout->lot_attrs,
sizeof (sa_attr_type_t) * layout->lot_attr_count);
kmem_free(layout, sizeof (sa_lot_t));
}
avl_destroy(&sa->sa_layout_hash_tree);
avl_destroy(&sa->sa_layout_num_tree);
mutex_destroy(&sa->sa_lock);
kmem_free(sa, sizeof (sa_os_t));
os->os_sa = NULL;
}
void
sa_build_idx_tab(void *hdr, void *attr_addr, sa_attr_type_t attr,
uint16_t length, int length_idx, boolean_t var_length, void *userp)
{
sa_idx_tab_t *idx_tab = userp;
if (var_length) {
ASSERT(idx_tab->sa_variable_lengths);
idx_tab->sa_variable_lengths[length_idx] = length;
}
TOC_ATTR_ENCODE(idx_tab->sa_idx_tab[attr], length_idx,
(uint32_t)((uintptr_t)attr_addr - (uintptr_t)hdr));
}
static void
sa_attr_iter(objset_t *os, sa_hdr_phys_t *hdr, dmu_object_type_t type,
sa_iterfunc_t func, sa_lot_t *tab, void *userp)
{
void *data_start;
sa_lot_t *tb = tab;
sa_lot_t search;
avl_index_t loc;
sa_os_t *sa = os->os_sa;
int i;
uint16_t *length_start = NULL;
uint8_t length_idx = 0;
if (tab == NULL) {
search.lot_num = SA_LAYOUT_NUM(hdr, type);
tb = avl_find(&sa->sa_layout_num_tree, &search, &loc);
ASSERT(tb);
}
if (IS_SA_BONUSTYPE(type)) {
data_start = (void *)P2ROUNDUP(((uintptr_t)hdr +
offsetof(sa_hdr_phys_t, sa_lengths) +
(sizeof (uint16_t) * tb->lot_var_sizes)), 8);
length_start = hdr->sa_lengths;
} else {
data_start = hdr;
}
for (i = 0; i != tb->lot_attr_count; i++) {
int attr_length, reg_length;
uint8_t idx_len;
reg_length = sa->sa_attr_table[tb->lot_attrs[i]].sa_length;
if (reg_length) {
attr_length = reg_length;
idx_len = 0;
} else {
attr_length = length_start[length_idx];
idx_len = length_idx++;
}
func(hdr, data_start, tb->lot_attrs[i], attr_length,
idx_len, reg_length == 0 ? B_TRUE : B_FALSE, userp);
data_start = (void *)P2ROUNDUP(((uintptr_t)data_start +
attr_length), 8);
}
}
/*ARGSUSED*/
void
sa_byteswap_cb(void *hdr, void *attr_addr, sa_attr_type_t attr,
uint16_t length, int length_idx, boolean_t variable_length, void *userp)
{
sa_handle_t *hdl = userp;
sa_os_t *sa = hdl->sa_os->os_sa;
sa_bswap_table[sa->sa_attr_table[attr].sa_byteswap](attr_addr, length);
}
void
sa_byteswap(sa_handle_t *hdl, sa_buf_type_t buftype)
{
sa_hdr_phys_t *sa_hdr_phys = SA_GET_HDR(hdl, buftype);
dmu_buf_impl_t *db;
int num_lengths = 1;
int i;
ASSERTV(sa_os_t *sa = hdl->sa_os->os_sa);
ASSERT(MUTEX_HELD(&sa->sa_lock));
if (sa_hdr_phys->sa_magic == SA_MAGIC)
return;
db = SA_GET_DB(hdl, buftype);
if (buftype == SA_SPILL) {
arc_release(db->db_buf, NULL);
arc_buf_thaw(db->db_buf);
}
sa_hdr_phys->sa_magic = BSWAP_32(sa_hdr_phys->sa_magic);
sa_hdr_phys->sa_layout_info = BSWAP_16(sa_hdr_phys->sa_layout_info);
/*
* Determine number of variable lenghts in header
* The standard 8 byte header has one for free and a
* 16 byte header would have 4 + 1;
*/
if (SA_HDR_SIZE(sa_hdr_phys) > 8)
num_lengths += (SA_HDR_SIZE(sa_hdr_phys) - 8) >> 1;
for (i = 0; i != num_lengths; i++)
sa_hdr_phys->sa_lengths[i] =
BSWAP_16(sa_hdr_phys->sa_lengths[i]);
sa_attr_iter(hdl->sa_os, sa_hdr_phys, DMU_OT_SA,
sa_byteswap_cb, NULL, hdl);
if (buftype == SA_SPILL)
arc_buf_freeze(((dmu_buf_impl_t *)hdl->sa_spill)->db_buf);
}
static int
sa_build_index(sa_handle_t *hdl, sa_buf_type_t buftype)
{
sa_hdr_phys_t *sa_hdr_phys;
dmu_buf_impl_t *db = SA_GET_DB(hdl, buftype);
dmu_object_type_t bonustype = SA_BONUSTYPE_FROM_DB(db);
sa_os_t *sa = hdl->sa_os->os_sa;
sa_idx_tab_t *idx_tab;
sa_hdr_phys = SA_GET_HDR(hdl, buftype);
mutex_enter(&sa->sa_lock);
/* Do we need to byteswap? */
/* only check if not old znode */
if (IS_SA_BONUSTYPE(bonustype) && sa_hdr_phys->sa_magic != SA_MAGIC &&
sa_hdr_phys->sa_magic != 0) {
VERIFY(BSWAP_32(sa_hdr_phys->sa_magic) == SA_MAGIC);
sa_byteswap(hdl, buftype);
}
idx_tab = sa_find_idx_tab(hdl->sa_os, bonustype, sa_hdr_phys);
if (buftype == SA_BONUS)
hdl->sa_bonus_tab = idx_tab;
else
hdl->sa_spill_tab = idx_tab;
mutex_exit(&sa->sa_lock);
return (0);
}
/*ARGSUSED*/
static void
sa_evict(void *dbu)
{
panic("evicting sa dbuf\n");
}
static void
sa_idx_tab_rele(objset_t *os, void *arg)
{
sa_os_t *sa = os->os_sa;
sa_idx_tab_t *idx_tab = arg;
if (idx_tab == NULL)
return;
mutex_enter(&sa->sa_lock);
if (refcount_remove(&idx_tab->sa_refcount, NULL) == 0) {
list_remove(&idx_tab->sa_layout->lot_idx_tab, idx_tab);
if (idx_tab->sa_variable_lengths)
kmem_free(idx_tab->sa_variable_lengths,
sizeof (uint16_t) *
idx_tab->sa_layout->lot_var_sizes);
refcount_destroy(&idx_tab->sa_refcount);
kmem_free(idx_tab->sa_idx_tab,
sizeof (uint32_t) * sa->sa_num_attrs);
kmem_free(idx_tab, sizeof (sa_idx_tab_t));
}
mutex_exit(&sa->sa_lock);
}
static void
sa_idx_tab_hold(objset_t *os, sa_idx_tab_t *idx_tab)
{
ASSERTV(sa_os_t *sa = os->os_sa);
ASSERT(MUTEX_HELD(&sa->sa_lock));
(void) refcount_add(&idx_tab->sa_refcount, NULL);
}
void
sa_spill_rele(sa_handle_t *hdl)
{
mutex_enter(&hdl->sa_lock);
if (hdl->sa_spill) {
sa_idx_tab_rele(hdl->sa_os, hdl->sa_spill_tab);
dmu_buf_rele(hdl->sa_spill, NULL);
hdl->sa_spill = NULL;
hdl->sa_spill_tab = NULL;
}
mutex_exit(&hdl->sa_lock);
}
void
sa_handle_destroy(sa_handle_t *hdl)
{
dmu_buf_t *db = hdl->sa_bonus;
mutex_enter(&hdl->sa_lock);
(void) dmu_buf_remove_user(db, &hdl->sa_dbu);
if (hdl->sa_bonus_tab)
sa_idx_tab_rele(hdl->sa_os, hdl->sa_bonus_tab);
if (hdl->sa_spill_tab)
sa_idx_tab_rele(hdl->sa_os, hdl->sa_spill_tab);
dmu_buf_rele(hdl->sa_bonus, NULL);
if (hdl->sa_spill)
dmu_buf_rele((dmu_buf_t *)hdl->sa_spill, NULL);
mutex_exit(&hdl->sa_lock);
kmem_cache_free(sa_cache, hdl);
}
int
sa_handle_get_from_db(objset_t *os, dmu_buf_t *db, void *userp,
sa_handle_type_t hdl_type, sa_handle_t **handlepp)
{
int error = 0;
sa_handle_t *handle = NULL;
#ifdef ZFS_DEBUG
dmu_object_info_t doi;
dmu_object_info_from_db(db, &doi);
ASSERT(doi.doi_bonus_type == DMU_OT_SA ||
doi.doi_bonus_type == DMU_OT_ZNODE);
#endif
/* find handle, if it exists */
/* if one doesn't exist then create a new one, and initialize it */
if (hdl_type == SA_HDL_SHARED)
handle = dmu_buf_get_user(db);
if (handle == NULL) {
sa_handle_t *winner = NULL;
handle = kmem_cache_alloc(sa_cache, KM_SLEEP);
handle->sa_dbu.dbu_evict_func = NULL;
handle->sa_userp = userp;
handle->sa_bonus = db;
handle->sa_os = os;
handle->sa_spill = NULL;
handle->sa_bonus_tab = NULL;
handle->sa_spill_tab = NULL;
error = sa_build_index(handle, SA_BONUS);
if (hdl_type == SA_HDL_SHARED) {
dmu_buf_init_user(&handle->sa_dbu, sa_evict, NULL);
winner = dmu_buf_set_user_ie(db, &handle->sa_dbu);
}
if (winner != NULL) {
kmem_cache_free(sa_cache, handle);
handle = winner;
}
}
*handlepp = handle;
return (error);
}
int
sa_handle_get(objset_t *objset, uint64_t objid, void *userp,
sa_handle_type_t hdl_type, sa_handle_t **handlepp)
{
dmu_buf_t *db;
int error;
if ((error = dmu_bonus_hold(objset, objid, NULL, &db)))
return (error);
return (sa_handle_get_from_db(objset, db, userp, hdl_type,
handlepp));
}
int
sa_buf_hold(objset_t *objset, uint64_t obj_num, void *tag, dmu_buf_t **db)
{
return (dmu_bonus_hold(objset, obj_num, tag, db));
}
void
sa_buf_rele(dmu_buf_t *db, void *tag)
{
dmu_buf_rele(db, tag);
}
int
sa_lookup_impl(sa_handle_t *hdl, sa_bulk_attr_t *bulk, int count)
{
ASSERT(hdl);
ASSERT(MUTEX_HELD(&hdl->sa_lock));
return (sa_attr_op(hdl, bulk, count, SA_LOOKUP, NULL));
}
int
sa_lookup(sa_handle_t *hdl, sa_attr_type_t attr, void *buf, uint32_t buflen)
{
int error;
sa_bulk_attr_t bulk;
VERIFY3U(buflen, <=, SA_ATTR_MAX_LEN);
bulk.sa_attr = attr;
bulk.sa_data = buf;
bulk.sa_length = buflen;
bulk.sa_data_func = NULL;
ASSERT(hdl);
mutex_enter(&hdl->sa_lock);
error = sa_lookup_impl(hdl, &bulk, 1);
mutex_exit(&hdl->sa_lock);
return (error);
}
#ifdef _KERNEL
int
sa_lookup_uio(sa_handle_t *hdl, sa_attr_type_t attr, uio_t *uio)
{
int error;
sa_bulk_attr_t bulk;
bulk.sa_data = NULL;
bulk.sa_attr = attr;
bulk.sa_data_func = NULL;
ASSERT(hdl);
mutex_enter(&hdl->sa_lock);
if ((error = sa_attr_op(hdl, &bulk, 1, SA_LOOKUP, NULL)) == 0) {
error = uiomove((void *)bulk.sa_addr, MIN(bulk.sa_size,
uio->uio_resid), UIO_READ, uio);
}
mutex_exit(&hdl->sa_lock);
return (error);
}
#endif
void *
sa_find_idx_tab(objset_t *os, dmu_object_type_t bonustype, void *data)
{
sa_idx_tab_t *idx_tab;
sa_hdr_phys_t *hdr = (sa_hdr_phys_t *)data;
sa_os_t *sa = os->os_sa;
sa_lot_t *tb, search;
avl_index_t loc;
/*
* Deterimine layout number. If SA node and header == 0 then
* force the index table to the dummy "1" empty layout.
*
* The layout number would only be zero for a newly created file
* that has not added any attributes yet, or with crypto enabled which
* doesn't write any attributes to the bonus buffer.
*/
search.lot_num = SA_LAYOUT_NUM(hdr, bonustype);
tb = avl_find(&sa->sa_layout_num_tree, &search, &loc);
/* Verify header size is consistent with layout information */
ASSERT(tb);
ASSERT((IS_SA_BONUSTYPE(bonustype) &&
SA_HDR_SIZE_MATCH_LAYOUT(hdr, tb)) || !IS_SA_BONUSTYPE(bonustype) ||
(IS_SA_BONUSTYPE(bonustype) && hdr->sa_layout_info == 0));
/*
* See if any of the already existing TOC entries can be reused?
*/
for (idx_tab = list_head(&tb->lot_idx_tab); idx_tab;
idx_tab = list_next(&tb->lot_idx_tab, idx_tab)) {
boolean_t valid_idx = B_TRUE;
int i;
if (tb->lot_var_sizes != 0 &&
idx_tab->sa_variable_lengths != NULL) {
for (i = 0; i != tb->lot_var_sizes; i++) {
if (hdr->sa_lengths[i] !=
idx_tab->sa_variable_lengths[i]) {
valid_idx = B_FALSE;
break;
}
}
}
if (valid_idx) {
sa_idx_tab_hold(os, idx_tab);
return (idx_tab);
}
}
/* No such luck, create a new entry */
idx_tab = kmem_zalloc(sizeof (sa_idx_tab_t), KM_SLEEP);
idx_tab->sa_idx_tab =
kmem_zalloc(sizeof (uint32_t) * sa->sa_num_attrs, KM_SLEEP);
idx_tab->sa_layout = tb;
refcount_create(&idx_tab->sa_refcount);
if (tb->lot_var_sizes)
idx_tab->sa_variable_lengths = kmem_alloc(sizeof (uint16_t) *
tb->lot_var_sizes, KM_SLEEP);
sa_attr_iter(os, hdr, bonustype, sa_build_idx_tab,
tb, idx_tab);
sa_idx_tab_hold(os, idx_tab); /* one hold for consumer */
sa_idx_tab_hold(os, idx_tab); /* one for layout */
list_insert_tail(&tb->lot_idx_tab, idx_tab);
return (idx_tab);
}
void
sa_default_locator(void **dataptr, uint32_t *len, uint32_t total_len,
boolean_t start, void *userdata)
{
ASSERT(start);
*dataptr = userdata;
*len = total_len;
}
static void
sa_attr_register_sync(sa_handle_t *hdl, dmu_tx_t *tx)
{
uint64_t attr_value = 0;
sa_os_t *sa = hdl->sa_os->os_sa;
sa_attr_table_t *tb = sa->sa_attr_table;
int i;
mutex_enter(&sa->sa_lock);
if (!sa->sa_need_attr_registration || sa->sa_master_obj == 0) {
mutex_exit(&sa->sa_lock);
return;
}
if (sa->sa_reg_attr_obj == 0) {
sa->sa_reg_attr_obj = zap_create_link(hdl->sa_os,
DMU_OT_SA_ATTR_REGISTRATION,
sa->sa_master_obj, SA_REGISTRY, tx);
}
for (i = 0; i != sa->sa_num_attrs; i++) {
if (sa->sa_attr_table[i].sa_registered)
continue;
ATTR_ENCODE(attr_value, tb[i].sa_attr, tb[i].sa_length,
tb[i].sa_byteswap);
VERIFY(0 == zap_update(hdl->sa_os, sa->sa_reg_attr_obj,
tb[i].sa_name, 8, 1, &attr_value, tx));
tb[i].sa_registered = B_TRUE;
}
sa->sa_need_attr_registration = B_FALSE;
mutex_exit(&sa->sa_lock);
}
/*
* Replace all attributes with attributes specified in template.
* If dnode had a spill buffer then those attributes will be
* also be replaced, possibly with just an empty spill block
*
* This interface is intended to only be used for bulk adding of
* attributes for a new file. It will also be used by the ZPL
* when converting and old formatted znode to native SA support.
*/
int
sa_replace_all_by_template_locked(sa_handle_t *hdl, sa_bulk_attr_t *attr_desc,
int attr_count, dmu_tx_t *tx)
{
sa_os_t *sa = hdl->sa_os->os_sa;
if (sa->sa_need_attr_registration)
sa_attr_register_sync(hdl, tx);
return (sa_build_layouts(hdl, attr_desc, attr_count, tx));
}
int
sa_replace_all_by_template(sa_handle_t *hdl, sa_bulk_attr_t *attr_desc,
int attr_count, dmu_tx_t *tx)
{
int error;
mutex_enter(&hdl->sa_lock);
error = sa_replace_all_by_template_locked(hdl, attr_desc,
attr_count, tx);
mutex_exit(&hdl->sa_lock);
return (error);
}
/*
* add/remove/replace a single attribute and then rewrite the entire set
* of attributes.
*/
static int
sa_modify_attrs(sa_handle_t *hdl, sa_attr_type_t newattr,
sa_data_op_t action, sa_data_locator_t *locator, void *datastart,
uint16_t buflen, dmu_tx_t *tx)
{
sa_os_t *sa = hdl->sa_os->os_sa;
dmu_buf_impl_t *db = (dmu_buf_impl_t *)hdl->sa_bonus;
dnode_t *dn;
sa_bulk_attr_t *attr_desc;
void *old_data[2];
int bonus_attr_count = 0;
Implement SA based xattrs The current ZFS implementation stores xattrs on disk using a hidden directory. In this directory a file name represents the xattr name and the file contexts are the xattr binary data. This approach is very flexible and allows for arbitrarily large xattrs. However, it also suffers from a significant performance penalty. Accessing a single xattr can requires up to three disk seeks. 1) Lookup the dnode object. 2) Lookup the dnodes's xattr directory object. 3) Lookup the xattr object in the directory. To avoid this performance penalty Linux filesystems such as ext3 and xfs try to store the xattr as part of the inode on disk. When the xattr is to large to store in the inode then a single external block is allocated for them. In practice most xattrs are small and this approach works well. The addition of System Attributes (SA) to zfs provides us a clean way to make this optimization. When the dataset property 'xattr=sa' is set then xattrs will be preferentially stored as System Attributes. This allows tiny xattrs (~100 bytes) to be stored with the dnode and up to 64k of xattrs to be stored in the spill block. If additional xattr space is required, which is unlikely under Linux, they will be stored using the traditional directory approach. This optimization results in roughly a 3x performance improvement when accessing xattrs which brings zfs roughly to parity with ext4 and xfs (see table below). When multiple xattrs are stored per-file the performance improvements are even greater because all of the xattrs stored in the spill block will be cached. However, by default SA based xattrs are disabled in the Linux port to maximize compatibility with other implementations. If you do enable SA based xattrs then they will not be visible on platforms which do not support this feature. ---------------------------------------------------------------------- Time in seconds to get/set one xattr of N bytes on 100,000 files ------+--------------------------------+------------------------------ | setxattr | getxattr bytes | ext4 xfs zfs-dir zfs-sa | ext4 xfs zfs-dir zfs-sa ------+--------------------------------+------------------------------ 1 | 2.33 31.88 21.50 4.57 | 2.35 2.64 6.29 2.43 32 | 2.79 30.68 21.98 4.60 | 2.44 2.59 6.78 2.48 256 | 3.25 31.99 21.36 5.92 | 2.32 2.71 6.22 3.14 1024 | 3.30 32.61 22.83 8.45 | 2.40 2.79 6.24 3.27 4096 | 3.57 317.46 22.52 10.73 | 2.78 28.62 6.90 3.94 16384 | n/a 2342.39 34.30 19.20 | n/a 45.44 145.90 7.55 65536 | n/a 2941.39 128.15 131.32* | n/a 141.92 256.85 262.12* Legend: * ext4 - Stock RHEL6.1 ext4 mounted with '-o user_xattr'. * xfs - Stock RHEL6.1 xfs mounted with default options. * zfs-dir - Directory based xattrs only. * zfs-sa - Prefer SAs but spill in to directories as needed, a trailing * indicates overflow in to directories occured. NOTE: Ext4 supports 4096 bytes of xattr name/value pairs per file. NOTE: XFS and ZFS have no limit on xattr name/value pairs per file. NOTE: Linux limits individual name/value pairs to 65536 bytes. NOTE: All setattr/getattr's were done after dropping the cache. NOTE: All tests were run against a single hard drive. Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov> Issue #443
2011-10-24 23:55:20 +00:00
int bonus_data_size = 0;
int spill_data_size = 0;
int spill_attr_count = 0;
int error;
uint16_t length;
int i, j, k, length_idx;
sa_hdr_phys_t *hdr;
sa_idx_tab_t *idx_tab;
int attr_count;
int count;
ASSERT(MUTEX_HELD(&hdl->sa_lock));
/* First make of copy of the old data */
DB_DNODE_ENTER(db);
dn = DB_DNODE(db);
if (dn->dn_bonuslen != 0) {
bonus_data_size = hdl->sa_bonus->db_size;
old_data[0] = kmem_alloc(bonus_data_size, KM_SLEEP);
bcopy(hdl->sa_bonus->db_data, old_data[0],
hdl->sa_bonus->db_size);
bonus_attr_count = hdl->sa_bonus_tab->sa_layout->lot_attr_count;
} else {
old_data[0] = NULL;
}
DB_DNODE_EXIT(db);
/* Bring spill buffer online if it isn't currently */
if ((error = sa_get_spill(hdl)) == 0) {
spill_data_size = hdl->sa_spill->db_size;
old_data[1] = zio_buf_alloc(spill_data_size);
bcopy(hdl->sa_spill->db_data, old_data[1],
hdl->sa_spill->db_size);
spill_attr_count =
hdl->sa_spill_tab->sa_layout->lot_attr_count;
} else if (error && error != ENOENT) {
if (old_data[0])
kmem_free(old_data[0], bonus_data_size);
return (error);
} else {
old_data[1] = NULL;
}
/* build descriptor of all attributes */
attr_count = bonus_attr_count + spill_attr_count;
if (action == SA_ADD)
attr_count++;
else if (action == SA_REMOVE)
attr_count--;
attr_desc = kmem_zalloc(sizeof (sa_bulk_attr_t) * attr_count, KM_SLEEP);
/*
* loop through bonus and spill buffer if it exists, and
* build up new attr_descriptor to reset the attributes
*/
k = j = 0;
count = bonus_attr_count;
hdr = SA_GET_HDR(hdl, SA_BONUS);
idx_tab = SA_IDX_TAB_GET(hdl, SA_BONUS);
for (; k != 2; k++) {
/*
* Iterate over each attribute in layout. Fetch the
* size of variable-length attributes needing rewrite
* from sa_lengths[].
*/
for (i = 0, length_idx = 0; i != count; i++) {
sa_attr_type_t attr;
attr = idx_tab->sa_layout->lot_attrs[i];
length = SA_REGISTERED_LEN(sa, attr);
if (attr == newattr) {
if (length == 0)
++length_idx;
if (action == SA_REMOVE)
continue;
ASSERT(length == 0);
ASSERT(action == SA_REPLACE);
SA_ADD_BULK_ATTR(attr_desc, j, attr,
locator, datastart, buflen);
} else {
if (length == 0)
length = hdr->sa_lengths[length_idx++];
SA_ADD_BULK_ATTR(attr_desc, j, attr,
NULL, (void *)
(TOC_OFF(idx_tab->sa_idx_tab[attr]) +
(uintptr_t)old_data[k]), length);
}
}
if (k == 0 && hdl->sa_spill) {
hdr = SA_GET_HDR(hdl, SA_SPILL);
idx_tab = SA_IDX_TAB_GET(hdl, SA_SPILL);
count = spill_attr_count;
} else {
break;
}
}
if (action == SA_ADD) {
length = SA_REGISTERED_LEN(sa, newattr);
if (length == 0) {
length = buflen;
}
SA_ADD_BULK_ATTR(attr_desc, j, newattr, locator,
datastart, length);
}
error = sa_build_layouts(hdl, attr_desc, attr_count, tx);
if (old_data[0])
kmem_free(old_data[0], bonus_data_size);
if (old_data[1])
zio_buf_free(old_data[1], spill_data_size);
kmem_free(attr_desc, sizeof (sa_bulk_attr_t) * attr_count);
return (error);
}
static int
sa_bulk_update_impl(sa_handle_t *hdl, sa_bulk_attr_t *bulk, int count,
dmu_tx_t *tx)
{
int error;
sa_os_t *sa = hdl->sa_os->os_sa;
dmu_object_type_t bonustype;
dmu_buf_t *saved_spill;
ASSERT(hdl);
ASSERT(MUTEX_HELD(&hdl->sa_lock));
bonustype = SA_BONUSTYPE_FROM_DB(SA_GET_DB(hdl, SA_BONUS));
saved_spill = hdl->sa_spill;
/* sync out registration table if necessary */
if (sa->sa_need_attr_registration)
sa_attr_register_sync(hdl, tx);
error = sa_attr_op(hdl, bulk, count, SA_UPDATE, tx);
if (error == 0 && !IS_SA_BONUSTYPE(bonustype) && sa->sa_update_cb)
sa->sa_update_cb(hdl, tx);
/*
* If saved_spill is NULL and current sa_spill is not NULL that
* means we increased the refcount of the spill buffer through
* sa_get_spill() or dmu_spill_hold_by_dnode(). Therefore we
* must release the hold before calling dmu_tx_commit() to avoid
* making a copy of this buffer in dbuf_sync_leaf() due to the
* reference count now being greater than 1.
*/
if (!saved_spill && hdl->sa_spill) {
if (hdl->sa_spill_tab) {
sa_idx_tab_rele(hdl->sa_os, hdl->sa_spill_tab);
hdl->sa_spill_tab = NULL;
}
dmu_buf_rele((dmu_buf_t *)hdl->sa_spill, NULL);
hdl->sa_spill = NULL;
}
return (error);
}
/*
* update or add new attribute
*/
int
sa_update(sa_handle_t *hdl, sa_attr_type_t type,
void *buf, uint32_t buflen, dmu_tx_t *tx)
{
int error;
sa_bulk_attr_t bulk;
VERIFY3U(buflen, <=, SA_ATTR_MAX_LEN);
bulk.sa_attr = type;
bulk.sa_data_func = NULL;
bulk.sa_length = buflen;
bulk.sa_data = buf;
mutex_enter(&hdl->sa_lock);
error = sa_bulk_update_impl(hdl, &bulk, 1, tx);
mutex_exit(&hdl->sa_lock);
return (error);
}
int
sa_update_from_cb(sa_handle_t *hdl, sa_attr_type_t attr,
uint32_t buflen, sa_data_locator_t *locator, void *userdata, dmu_tx_t *tx)
{
int error;
sa_bulk_attr_t bulk;
VERIFY3U(buflen, <=, SA_ATTR_MAX_LEN);
bulk.sa_attr = attr;
bulk.sa_data = userdata;
bulk.sa_data_func = locator;
bulk.sa_length = buflen;
mutex_enter(&hdl->sa_lock);
error = sa_bulk_update_impl(hdl, &bulk, 1, tx);
mutex_exit(&hdl->sa_lock);
return (error);
}
/*
* Return size of an attribute
*/
int
sa_size(sa_handle_t *hdl, sa_attr_type_t attr, int *size)
{
sa_bulk_attr_t bulk;
int error;
bulk.sa_data = NULL;
bulk.sa_attr = attr;
bulk.sa_data_func = NULL;
ASSERT(hdl);
mutex_enter(&hdl->sa_lock);
if ((error = sa_attr_op(hdl, &bulk, 1, SA_LOOKUP, NULL)) != 0) {
mutex_exit(&hdl->sa_lock);
return (error);
}
*size = bulk.sa_size;
mutex_exit(&hdl->sa_lock);
return (0);
}
int
sa_bulk_lookup_locked(sa_handle_t *hdl, sa_bulk_attr_t *attrs, int count)
{
ASSERT(hdl);
ASSERT(MUTEX_HELD(&hdl->sa_lock));
return (sa_lookup_impl(hdl, attrs, count));
}
int
sa_bulk_lookup(sa_handle_t *hdl, sa_bulk_attr_t *attrs, int count)
{
int error;
ASSERT(hdl);
mutex_enter(&hdl->sa_lock);
error = sa_bulk_lookup_locked(hdl, attrs, count);
mutex_exit(&hdl->sa_lock);
return (error);
}
int
sa_bulk_update(sa_handle_t *hdl, sa_bulk_attr_t *attrs, int count, dmu_tx_t *tx)
{
int error;
ASSERT(hdl);
mutex_enter(&hdl->sa_lock);
error = sa_bulk_update_impl(hdl, attrs, count, tx);
mutex_exit(&hdl->sa_lock);
return (error);
}
int
sa_remove(sa_handle_t *hdl, sa_attr_type_t attr, dmu_tx_t *tx)
{
int error;
mutex_enter(&hdl->sa_lock);
error = sa_modify_attrs(hdl, attr, SA_REMOVE, NULL,
NULL, 0, tx);
mutex_exit(&hdl->sa_lock);
return (error);
}
void
sa_object_info(sa_handle_t *hdl, dmu_object_info_t *doi)
{
dmu_object_info_from_db((dmu_buf_t *)hdl->sa_bonus, doi);
}
void
sa_object_size(sa_handle_t *hdl, uint32_t *blksize, u_longlong_t *nblocks)
{
dmu_object_size_from_db((dmu_buf_t *)hdl->sa_bonus,
blksize, nblocks);
}
void
sa_set_userp(sa_handle_t *hdl, void *ptr)
{
hdl->sa_userp = ptr;
}
dmu_buf_t *
sa_get_db(sa_handle_t *hdl)
{
return ((dmu_buf_t *)hdl->sa_bonus);
}
void *
sa_get_userdata(sa_handle_t *hdl)
{
return (hdl->sa_userp);
}
void
sa_register_update_callback_locked(objset_t *os, sa_update_cb_t *func)
{
ASSERT(MUTEX_HELD(&os->os_sa->sa_lock));
os->os_sa->sa_update_cb = func;
}
void
sa_register_update_callback(objset_t *os, sa_update_cb_t *func)
{
mutex_enter(&os->os_sa->sa_lock);
sa_register_update_callback_locked(os, func);
mutex_exit(&os->os_sa->sa_lock);
}
uint64_t
sa_handle_object(sa_handle_t *hdl)
{
return (hdl->sa_bonus->db_object);
}
boolean_t
sa_enabled(objset_t *os)
{
return (os->os_sa == NULL);
}
int
sa_set_sa_object(objset_t *os, uint64_t sa_object)
{
sa_os_t *sa = os->os_sa;
if (sa->sa_master_obj)
return (1);
sa->sa_master_obj = sa_object;
return (0);
}
int
sa_hdrsize(void *arg)
{
sa_hdr_phys_t *hdr = arg;
return (SA_HDR_SIZE(hdr));
}
void
sa_handle_lock(sa_handle_t *hdl)
{
ASSERT(hdl);
mutex_enter(&hdl->sa_lock);
}
void
sa_handle_unlock(sa_handle_t *hdl)
{
ASSERT(hdl);
mutex_exit(&hdl->sa_lock);
}
#ifdef _KERNEL
EXPORT_SYMBOL(sa_handle_get);
EXPORT_SYMBOL(sa_handle_get_from_db);
EXPORT_SYMBOL(sa_handle_destroy);
EXPORT_SYMBOL(sa_buf_hold);
EXPORT_SYMBOL(sa_buf_rele);
EXPORT_SYMBOL(sa_spill_rele);
EXPORT_SYMBOL(sa_lookup);
EXPORT_SYMBOL(sa_update);
EXPORT_SYMBOL(sa_remove);
EXPORT_SYMBOL(sa_bulk_lookup);
EXPORT_SYMBOL(sa_bulk_lookup_locked);
EXPORT_SYMBOL(sa_bulk_update);
EXPORT_SYMBOL(sa_size);
EXPORT_SYMBOL(sa_update_from_cb);
EXPORT_SYMBOL(sa_object_info);
EXPORT_SYMBOL(sa_object_size);
EXPORT_SYMBOL(sa_get_userdata);
EXPORT_SYMBOL(sa_set_userp);
EXPORT_SYMBOL(sa_get_db);
EXPORT_SYMBOL(sa_handle_object);
EXPORT_SYMBOL(sa_register_update_callback);
EXPORT_SYMBOL(sa_setup);
EXPORT_SYMBOL(sa_replace_all_by_template);
EXPORT_SYMBOL(sa_replace_all_by_template_locked);
EXPORT_SYMBOL(sa_enabled);
EXPORT_SYMBOL(sa_cache_init);
EXPORT_SYMBOL(sa_cache_fini);
EXPORT_SYMBOL(sa_set_sa_object);
EXPORT_SYMBOL(sa_hdrsize);
EXPORT_SYMBOL(sa_handle_lock);
EXPORT_SYMBOL(sa_handle_unlock);
EXPORT_SYMBOL(sa_lookup_uio);
#endif /* _KERNEL */