linux-sg2042/fs/xfs/libxfs/xfs_iext_tree.c

1056 lines
23 KiB
C
Raw Normal View History

// SPDX-License-Identifier: GPL-2.0
xfs: use a b+tree for the in-core extent list Replace the current linear list and the indirection array for the in-core extent list with a b+tree to avoid the need for larger memory allocations for the indirection array when lots of extents are present. The current extent list implementations leads to heavy pressure on the memory allocator when modifying files with a high extent count, and can lead to high latencies because of that. The replacement is a b+tree with a few quirks. The leaf nodes directly store the extent record in two u64 values. The encoding is a little bit different from the existing in-core extent records so that the start offset and length which are required for lookups can be retreived with simple mask operations. The inner nodes store a 64-bit key containing the start offset in the first half of the node, and the pointers to the next lower level in the second half. In either case we walk the node from the beginninig to the end and do a linear search, as that is more efficient for the low number of cache lines touched during a search (2 for the inner nodes, 4 for the leaf nodes) than a binary search. We store termination markers (zero length for the leaf nodes, an otherwise impossible high bit for the inner nodes) to terminate the key list / records instead of storing a count to use the available cache lines as efficiently as possible. One quirk of the algorithm is that while we normally split a node half and half like usual btree implementations we just spill over entries added at the very end of the list to a new node on its own. This means we get a 100% fill grade for the common cases of bulk insertion when reading an inode into memory, and when only sequentially appending to a file. The downside is a slightly higher chance of splits on the first random insertions. Both insert and removal manually recurse into the lower levels, but the bulk deletion of the whole tree is still implemented as a recursive function call, although one limited by the overall depth and with very little stack usage in every iteration. For the first few extents we dynamically grow the list from a single extent to the next powers of two until we have a first full leaf block and that building the actual tree. The code started out based on the generic lib/btree.c code from Joern Engel based on earlier work from Peter Zijlstra, but has since been rewritten beyond recognition. Signed-off-by: Christoph Hellwig <hch@lst.de> Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com> Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
2017-11-04 01:34:46 +08:00
/*
* Copyright (c) 2017 Christoph Hellwig.
*/
#include <linux/cache.h>
#include <linux/kernel.h>
#include <linux/slab.h>
#include "xfs.h"
#include "xfs_format.h"
#include "xfs_bit.h"
#include "xfs_log_format.h"
#include "xfs_inode.h"
#include "xfs_inode_fork.h"
#include "xfs_trans_resv.h"
#include "xfs_mount.h"
#include "xfs_bmap.h"
xfs: use a b+tree for the in-core extent list Replace the current linear list and the indirection array for the in-core extent list with a b+tree to avoid the need for larger memory allocations for the indirection array when lots of extents are present. The current extent list implementations leads to heavy pressure on the memory allocator when modifying files with a high extent count, and can lead to high latencies because of that. The replacement is a b+tree with a few quirks. The leaf nodes directly store the extent record in two u64 values. The encoding is a little bit different from the existing in-core extent records so that the start offset and length which are required for lookups can be retreived with simple mask operations. The inner nodes store a 64-bit key containing the start offset in the first half of the node, and the pointers to the next lower level in the second half. In either case we walk the node from the beginninig to the end and do a linear search, as that is more efficient for the low number of cache lines touched during a search (2 for the inner nodes, 4 for the leaf nodes) than a binary search. We store termination markers (zero length for the leaf nodes, an otherwise impossible high bit for the inner nodes) to terminate the key list / records instead of storing a count to use the available cache lines as efficiently as possible. One quirk of the algorithm is that while we normally split a node half and half like usual btree implementations we just spill over entries added at the very end of the list to a new node on its own. This means we get a 100% fill grade for the common cases of bulk insertion when reading an inode into memory, and when only sequentially appending to a file. The downside is a slightly higher chance of splits on the first random insertions. Both insert and removal manually recurse into the lower levels, but the bulk deletion of the whole tree is still implemented as a recursive function call, although one limited by the overall depth and with very little stack usage in every iteration. For the first few extents we dynamically grow the list from a single extent to the next powers of two until we have a first full leaf block and that building the actual tree. The code started out based on the generic lib/btree.c code from Joern Engel based on earlier work from Peter Zijlstra, but has since been rewritten beyond recognition. Signed-off-by: Christoph Hellwig <hch@lst.de> Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com> Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
2017-11-04 01:34:46 +08:00
#include "xfs_trace.h"
/*
* In-core extent record layout:
*
* +-------+----------------------------+
* | 00:53 | all 54 bits of startoff |
* | 54:63 | low 10 bits of startblock |
* +-------+----------------------------+
* | 00:20 | all 21 bits of length |
* | 21 | unwritten extent bit |
* | 22:63 | high 42 bits of startblock |
* +-------+----------------------------+
*/
#define XFS_IEXT_STARTOFF_MASK xfs_mask64lo(BMBT_STARTOFF_BITLEN)
#define XFS_IEXT_LENGTH_MASK xfs_mask64lo(BMBT_BLOCKCOUNT_BITLEN)
#define XFS_IEXT_STARTBLOCK_MASK xfs_mask64lo(BMBT_STARTBLOCK_BITLEN)
struct xfs_iext_rec {
uint64_t lo;
uint64_t hi;
};
/*
* Given that the length can't be a zero, only an empty hi value indicates an
* unused record.
*/
static bool xfs_iext_rec_is_empty(struct xfs_iext_rec *rec)
{
return rec->hi == 0;
}
static inline void xfs_iext_rec_clear(struct xfs_iext_rec *rec)
{
rec->lo = 0;
rec->hi = 0;
}
static void
xfs_iext_set(
struct xfs_iext_rec *rec,
struct xfs_bmbt_irec *irec)
{
ASSERT((irec->br_startoff & ~XFS_IEXT_STARTOFF_MASK) == 0);
ASSERT((irec->br_blockcount & ~XFS_IEXT_LENGTH_MASK) == 0);
ASSERT((irec->br_startblock & ~XFS_IEXT_STARTBLOCK_MASK) == 0);
rec->lo = irec->br_startoff & XFS_IEXT_STARTOFF_MASK;
rec->hi = irec->br_blockcount & XFS_IEXT_LENGTH_MASK;
rec->lo |= (irec->br_startblock << 54);
rec->hi |= ((irec->br_startblock & ~xfs_mask64lo(10)) << (22 - 10));
if (irec->br_state == XFS_EXT_UNWRITTEN)
rec->hi |= (1 << 21);
}
static void
xfs_iext_get(
struct xfs_bmbt_irec *irec,
struct xfs_iext_rec *rec)
{
irec->br_startoff = rec->lo & XFS_IEXT_STARTOFF_MASK;
irec->br_blockcount = rec->hi & XFS_IEXT_LENGTH_MASK;
irec->br_startblock = rec->lo >> 54;
irec->br_startblock |= (rec->hi & xfs_mask64hi(42)) >> (22 - 10);
if (rec->hi & (1 << 21))
irec->br_state = XFS_EXT_UNWRITTEN;
else
irec->br_state = XFS_EXT_NORM;
}
enum {
NODE_SIZE = 256,
KEYS_PER_NODE = NODE_SIZE / (sizeof(uint64_t) + sizeof(void *)),
RECS_PER_LEAF = (NODE_SIZE - (2 * sizeof(struct xfs_iext_leaf *))) /
sizeof(struct xfs_iext_rec),
};
/*
* In-core extent btree block layout:
*
* There are two types of blocks in the btree: leaf and inner (non-leaf) blocks.
*
* The leaf blocks are made up by %KEYS_PER_NODE extent records, which each
* contain the startoffset, blockcount, startblock and unwritten extent flag.
* See above for the exact format, followed by pointers to the previous and next
* leaf blocks (if there are any).
*
* The inner (non-leaf) blocks first contain KEYS_PER_NODE lookup keys, followed
* by an equal number of pointers to the btree blocks at the next lower level.
*
* +-------+-------+-------+-------+-------+----------+----------+
* Leaf: | rec 1 | rec 2 | rec 3 | rec 4 | rec N | prev-ptr | next-ptr |
* +-------+-------+-------+-------+-------+----------+----------+
*
* +-------+-------+-------+-------+-------+-------+------+-------+
* Inner: | key 1 | key 2 | key 3 | key N | ptr 1 | ptr 2 | ptr3 | ptr N |
* +-------+-------+-------+-------+-------+-------+------+-------+
*/
struct xfs_iext_node {
uint64_t keys[KEYS_PER_NODE];
#define XFS_IEXT_KEY_INVALID (1ULL << 63)
void *ptrs[KEYS_PER_NODE];
};
struct xfs_iext_leaf {
struct xfs_iext_rec recs[RECS_PER_LEAF];
struct xfs_iext_leaf *prev;
struct xfs_iext_leaf *next;
};
inline xfs_extnum_t xfs_iext_count(struct xfs_ifork *ifp)
{
return ifp->if_bytes / sizeof(struct xfs_iext_rec);
}
static inline int xfs_iext_max_recs(struct xfs_ifork *ifp)
{
if (ifp->if_height == 1)
return xfs_iext_count(ifp);
return RECS_PER_LEAF;
}
static inline struct xfs_iext_rec *cur_rec(struct xfs_iext_cursor *cur)
{
return &cur->leaf->recs[cur->pos];
}
static inline bool xfs_iext_valid(struct xfs_ifork *ifp,
struct xfs_iext_cursor *cur)
{
if (!cur->leaf)
return false;
if (cur->pos < 0 || cur->pos >= xfs_iext_max_recs(ifp))
return false;
if (xfs_iext_rec_is_empty(cur_rec(cur)))
return false;
return true;
}
static void *
xfs_iext_find_first_leaf(
struct xfs_ifork *ifp)
{
struct xfs_iext_node *node = ifp->if_u1.if_root;
int height;
if (!ifp->if_height)
return NULL;
for (height = ifp->if_height; height > 1; height--) {
node = node->ptrs[0];
ASSERT(node);
}
return node;
}
static void *
xfs_iext_find_last_leaf(
struct xfs_ifork *ifp)
{
struct xfs_iext_node *node = ifp->if_u1.if_root;
int height, i;
if (!ifp->if_height)
return NULL;
for (height = ifp->if_height; height > 1; height--) {
for (i = 1; i < KEYS_PER_NODE; i++)
if (!node->ptrs[i])
break;
node = node->ptrs[i - 1];
ASSERT(node);
}
return node;
}
void
xfs_iext_first(
struct xfs_ifork *ifp,
struct xfs_iext_cursor *cur)
{
cur->pos = 0;
cur->leaf = xfs_iext_find_first_leaf(ifp);
}
void
xfs_iext_last(
struct xfs_ifork *ifp,
struct xfs_iext_cursor *cur)
{
int i;
cur->leaf = xfs_iext_find_last_leaf(ifp);
if (!cur->leaf) {
cur->pos = 0;
return;
}
for (i = 1; i < xfs_iext_max_recs(ifp); i++) {
if (xfs_iext_rec_is_empty(&cur->leaf->recs[i]))
break;
}
cur->pos = i - 1;
}
void
xfs_iext_next(
struct xfs_ifork *ifp,
struct xfs_iext_cursor *cur)
{
if (!cur->leaf) {
ASSERT(cur->pos <= 0 || cur->pos >= RECS_PER_LEAF);
xfs_iext_first(ifp, cur);
return;
}
ASSERT(cur->pos >= 0);
ASSERT(cur->pos < xfs_iext_max_recs(ifp));
cur->pos++;
if (ifp->if_height > 1 && !xfs_iext_valid(ifp, cur) &&
cur->leaf->next) {
cur->leaf = cur->leaf->next;
cur->pos = 0;
}
}
void
xfs_iext_prev(
struct xfs_ifork *ifp,
struct xfs_iext_cursor *cur)
{
if (!cur->leaf) {
ASSERT(cur->pos <= 0 || cur->pos >= RECS_PER_LEAF);
xfs_iext_last(ifp, cur);
return;
}
ASSERT(cur->pos >= 0);
ASSERT(cur->pos <= RECS_PER_LEAF);
recurse:
do {
cur->pos--;
if (xfs_iext_valid(ifp, cur))
return;
} while (cur->pos > 0);
if (ifp->if_height > 1 && cur->leaf->prev) {
cur->leaf = cur->leaf->prev;
cur->pos = RECS_PER_LEAF;
goto recurse;
}
}
static inline int
xfs_iext_key_cmp(
struct xfs_iext_node *node,
int n,
xfs_fileoff_t offset)
{
if (node->keys[n] > offset)
return 1;
if (node->keys[n] < offset)
return -1;
return 0;
}
static inline int
xfs_iext_rec_cmp(
struct xfs_iext_rec *rec,
xfs_fileoff_t offset)
{
uint64_t rec_offset = rec->lo & XFS_IEXT_STARTOFF_MASK;
uint32_t rec_len = rec->hi & XFS_IEXT_LENGTH_MASK;
xfs: use a b+tree for the in-core extent list Replace the current linear list and the indirection array for the in-core extent list with a b+tree to avoid the need for larger memory allocations for the indirection array when lots of extents are present. The current extent list implementations leads to heavy pressure on the memory allocator when modifying files with a high extent count, and can lead to high latencies because of that. The replacement is a b+tree with a few quirks. The leaf nodes directly store the extent record in two u64 values. The encoding is a little bit different from the existing in-core extent records so that the start offset and length which are required for lookups can be retreived with simple mask operations. The inner nodes store a 64-bit key containing the start offset in the first half of the node, and the pointers to the next lower level in the second half. In either case we walk the node from the beginninig to the end and do a linear search, as that is more efficient for the low number of cache lines touched during a search (2 for the inner nodes, 4 for the leaf nodes) than a binary search. We store termination markers (zero length for the leaf nodes, an otherwise impossible high bit for the inner nodes) to terminate the key list / records instead of storing a count to use the available cache lines as efficiently as possible. One quirk of the algorithm is that while we normally split a node half and half like usual btree implementations we just spill over entries added at the very end of the list to a new node on its own. This means we get a 100% fill grade for the common cases of bulk insertion when reading an inode into memory, and when only sequentially appending to a file. The downside is a slightly higher chance of splits on the first random insertions. Both insert and removal manually recurse into the lower levels, but the bulk deletion of the whole tree is still implemented as a recursive function call, although one limited by the overall depth and with very little stack usage in every iteration. For the first few extents we dynamically grow the list from a single extent to the next powers of two until we have a first full leaf block and that building the actual tree. The code started out based on the generic lib/btree.c code from Joern Engel based on earlier work from Peter Zijlstra, but has since been rewritten beyond recognition. Signed-off-by: Christoph Hellwig <hch@lst.de> Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com> Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
2017-11-04 01:34:46 +08:00
if (rec_offset > offset)
return 1;
if (rec_offset + rec_len <= offset)
return -1;
return 0;
}
static void *
xfs_iext_find_level(
struct xfs_ifork *ifp,
xfs_fileoff_t offset,
int level)
{
struct xfs_iext_node *node = ifp->if_u1.if_root;
int height, i;
if (!ifp->if_height)
return NULL;
for (height = ifp->if_height; height > level; height--) {
for (i = 1; i < KEYS_PER_NODE; i++)
if (xfs_iext_key_cmp(node, i, offset) > 0)
break;
node = node->ptrs[i - 1];
if (!node)
break;
}
return node;
}
static int
xfs_iext_node_pos(
struct xfs_iext_node *node,
xfs_fileoff_t offset)
{
int i;
for (i = 1; i < KEYS_PER_NODE; i++) {
if (xfs_iext_key_cmp(node, i, offset) > 0)
break;
}
return i - 1;
}
static int
xfs_iext_node_insert_pos(
struct xfs_iext_node *node,
xfs_fileoff_t offset)
{
int i;
for (i = 0; i < KEYS_PER_NODE; i++) {
if (xfs_iext_key_cmp(node, i, offset) > 0)
return i;
}
return KEYS_PER_NODE;
}
static int
xfs_iext_node_nr_entries(
struct xfs_iext_node *node,
int start)
{
int i;
for (i = start; i < KEYS_PER_NODE; i++) {
if (node->keys[i] == XFS_IEXT_KEY_INVALID)
break;
}
return i;
}
static int
xfs_iext_leaf_nr_entries(
struct xfs_ifork *ifp,
struct xfs_iext_leaf *leaf,
int start)
{
int i;
for (i = start; i < xfs_iext_max_recs(ifp); i++) {
if (xfs_iext_rec_is_empty(&leaf->recs[i]))
break;
}
return i;
}
static inline uint64_t
xfs_iext_leaf_key(
struct xfs_iext_leaf *leaf,
int n)
{
return leaf->recs[n].lo & XFS_IEXT_STARTOFF_MASK;
}
static void
xfs_iext_grow(
struct xfs_ifork *ifp)
{
struct xfs_iext_node *node = kmem_zalloc(NODE_SIZE, KM_NOFS);
int i;
if (ifp->if_height == 1) {
struct xfs_iext_leaf *prev = ifp->if_u1.if_root;
node->keys[0] = xfs_iext_leaf_key(prev, 0);
node->ptrs[0] = prev;
} else {
struct xfs_iext_node *prev = ifp->if_u1.if_root;
ASSERT(ifp->if_height > 1);
node->keys[0] = prev->keys[0];
node->ptrs[0] = prev;
}
for (i = 1; i < KEYS_PER_NODE; i++)
node->keys[i] = XFS_IEXT_KEY_INVALID;
ifp->if_u1.if_root = node;
ifp->if_height++;
}
static void
xfs_iext_update_node(
struct xfs_ifork *ifp,
xfs_fileoff_t old_offset,
xfs_fileoff_t new_offset,
int level,
void *ptr)
{
struct xfs_iext_node *node = ifp->if_u1.if_root;
int height, i;
for (height = ifp->if_height; height > level; height--) {
for (i = 0; i < KEYS_PER_NODE; i++) {
if (i > 0 && xfs_iext_key_cmp(node, i, old_offset) > 0)
break;
if (node->keys[i] == old_offset)
node->keys[i] = new_offset;
}
node = node->ptrs[i - 1];
ASSERT(node);
}
ASSERT(node == ptr);
}
static struct xfs_iext_node *
xfs_iext_split_node(
struct xfs_iext_node **nodep,
int *pos,
int *nr_entries)
{
struct xfs_iext_node *node = *nodep;
struct xfs_iext_node *new = kmem_zalloc(NODE_SIZE, KM_NOFS);
const int nr_move = KEYS_PER_NODE / 2;
int nr_keep = nr_move + (KEYS_PER_NODE & 1);
int i = 0;
/* for sequential append operations just spill over into the new node */
if (*pos == KEYS_PER_NODE) {
*nodep = new;
*pos = 0;
*nr_entries = 0;
goto done;
}
for (i = 0; i < nr_move; i++) {
new->keys[i] = node->keys[nr_keep + i];
new->ptrs[i] = node->ptrs[nr_keep + i];
node->keys[nr_keep + i] = XFS_IEXT_KEY_INVALID;
node->ptrs[nr_keep + i] = NULL;
}
if (*pos >= nr_keep) {
*nodep = new;
*pos -= nr_keep;
*nr_entries = nr_move;
} else {
*nr_entries = nr_keep;
}
done:
for (; i < KEYS_PER_NODE; i++)
new->keys[i] = XFS_IEXT_KEY_INVALID;
return new;
}
static void
xfs_iext_insert_node(
struct xfs_ifork *ifp,
uint64_t offset,
void *ptr,
int level)
{
struct xfs_iext_node *node, *new;
int i, pos, nr_entries;
again:
if (ifp->if_height < level)
xfs_iext_grow(ifp);
new = NULL;
node = xfs_iext_find_level(ifp, offset, level);
pos = xfs_iext_node_insert_pos(node, offset);
nr_entries = xfs_iext_node_nr_entries(node, pos);
ASSERT(pos >= nr_entries || xfs_iext_key_cmp(node, pos, offset) != 0);
ASSERT(nr_entries <= KEYS_PER_NODE);
if (nr_entries == KEYS_PER_NODE)
new = xfs_iext_split_node(&node, &pos, &nr_entries);
/*
* Update the pointers in higher levels if the first entry changes
* in an existing node.
*/
xfs: use a b+tree for the in-core extent list Replace the current linear list and the indirection array for the in-core extent list with a b+tree to avoid the need for larger memory allocations for the indirection array when lots of extents are present. The current extent list implementations leads to heavy pressure on the memory allocator when modifying files with a high extent count, and can lead to high latencies because of that. The replacement is a b+tree with a few quirks. The leaf nodes directly store the extent record in two u64 values. The encoding is a little bit different from the existing in-core extent records so that the start offset and length which are required for lookups can be retreived with simple mask operations. The inner nodes store a 64-bit key containing the start offset in the first half of the node, and the pointers to the next lower level in the second half. In either case we walk the node from the beginninig to the end and do a linear search, as that is more efficient for the low number of cache lines touched during a search (2 for the inner nodes, 4 for the leaf nodes) than a binary search. We store termination markers (zero length for the leaf nodes, an otherwise impossible high bit for the inner nodes) to terminate the key list / records instead of storing a count to use the available cache lines as efficiently as possible. One quirk of the algorithm is that while we normally split a node half and half like usual btree implementations we just spill over entries added at the very end of the list to a new node on its own. This means we get a 100% fill grade for the common cases of bulk insertion when reading an inode into memory, and when only sequentially appending to a file. The downside is a slightly higher chance of splits on the first random insertions. Both insert and removal manually recurse into the lower levels, but the bulk deletion of the whole tree is still implemented as a recursive function call, although one limited by the overall depth and with very little stack usage in every iteration. For the first few extents we dynamically grow the list from a single extent to the next powers of two until we have a first full leaf block and that building the actual tree. The code started out based on the generic lib/btree.c code from Joern Engel based on earlier work from Peter Zijlstra, but has since been rewritten beyond recognition. Signed-off-by: Christoph Hellwig <hch@lst.de> Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com> Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
2017-11-04 01:34:46 +08:00
if (node != new && pos == 0 && nr_entries > 0)
xfs_iext_update_node(ifp, node->keys[0], offset, level, node);
for (i = nr_entries; i > pos; i--) {
node->keys[i] = node->keys[i - 1];
node->ptrs[i] = node->ptrs[i - 1];
}
node->keys[pos] = offset;
node->ptrs[pos] = ptr;
if (new) {
offset = new->keys[0];
ptr = new;
level++;
goto again;
}
}
static struct xfs_iext_leaf *
xfs_iext_split_leaf(
struct xfs_iext_cursor *cur,
int *nr_entries)
{
struct xfs_iext_leaf *leaf = cur->leaf;
struct xfs_iext_leaf *new = kmem_zalloc(NODE_SIZE, KM_NOFS);
const int nr_move = RECS_PER_LEAF / 2;
int nr_keep = nr_move + (RECS_PER_LEAF & 1);
int i;
/* for sequential append operations just spill over into the new node */
if (cur->pos == RECS_PER_LEAF) {
xfs: use a b+tree for the in-core extent list Replace the current linear list and the indirection array for the in-core extent list with a b+tree to avoid the need for larger memory allocations for the indirection array when lots of extents are present. The current extent list implementations leads to heavy pressure on the memory allocator when modifying files with a high extent count, and can lead to high latencies because of that. The replacement is a b+tree with a few quirks. The leaf nodes directly store the extent record in two u64 values. The encoding is a little bit different from the existing in-core extent records so that the start offset and length which are required for lookups can be retreived with simple mask operations. The inner nodes store a 64-bit key containing the start offset in the first half of the node, and the pointers to the next lower level in the second half. In either case we walk the node from the beginninig to the end and do a linear search, as that is more efficient for the low number of cache lines touched during a search (2 for the inner nodes, 4 for the leaf nodes) than a binary search. We store termination markers (zero length for the leaf nodes, an otherwise impossible high bit for the inner nodes) to terminate the key list / records instead of storing a count to use the available cache lines as efficiently as possible. One quirk of the algorithm is that while we normally split a node half and half like usual btree implementations we just spill over entries added at the very end of the list to a new node on its own. This means we get a 100% fill grade for the common cases of bulk insertion when reading an inode into memory, and when only sequentially appending to a file. The downside is a slightly higher chance of splits on the first random insertions. Both insert and removal manually recurse into the lower levels, but the bulk deletion of the whole tree is still implemented as a recursive function call, although one limited by the overall depth and with very little stack usage in every iteration. For the first few extents we dynamically grow the list from a single extent to the next powers of two until we have a first full leaf block and that building the actual tree. The code started out based on the generic lib/btree.c code from Joern Engel based on earlier work from Peter Zijlstra, but has since been rewritten beyond recognition. Signed-off-by: Christoph Hellwig <hch@lst.de> Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com> Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
2017-11-04 01:34:46 +08:00
cur->leaf = new;
cur->pos = 0;
*nr_entries = 0;
goto done;
}
for (i = 0; i < nr_move; i++) {
new->recs[i] = leaf->recs[nr_keep + i];
xfs_iext_rec_clear(&leaf->recs[nr_keep + i]);
}
if (cur->pos >= nr_keep) {
cur->leaf = new;
cur->pos -= nr_keep;
*nr_entries = nr_move;
} else {
*nr_entries = nr_keep;
}
done:
if (leaf->next)
leaf->next->prev = new;
new->next = leaf->next;
new->prev = leaf;
leaf->next = new;
return new;
}
static void
xfs_iext_alloc_root(
struct xfs_ifork *ifp,
struct xfs_iext_cursor *cur)
{
ASSERT(ifp->if_bytes == 0);
ifp->if_u1.if_root = kmem_zalloc(sizeof(struct xfs_iext_rec), KM_NOFS);
ifp->if_height = 1;
/* now that we have a node step into it */
cur->leaf = ifp->if_u1.if_root;
cur->pos = 0;
}
static void
xfs_iext_realloc_root(
struct xfs_ifork *ifp,
struct xfs_iext_cursor *cur)
{
size_t new_size = ifp->if_bytes + sizeof(struct xfs_iext_rec);
void *new;
/* account for the prev/next pointers */
if (new_size / sizeof(struct xfs_iext_rec) == RECS_PER_LEAF)
new_size = NODE_SIZE;
new = kmem_realloc(ifp->if_u1.if_root, new_size, KM_NOFS);
memset(new + ifp->if_bytes, 0, new_size - ifp->if_bytes);
ifp->if_u1.if_root = new;
cur->leaf = new;
}
/*
* Increment the sequence counter if we are on a COW fork. This allows
* the writeback code to skip looking for a COW extent if the COW fork
* hasn't changed. We use WRITE_ONCE here to ensure the update to the
* sequence counter is seen before the modifications to the extent
* tree itself take effect.
*/
static inline void xfs_iext_inc_seq(struct xfs_ifork *ifp, int state)
{
if (state & BMAP_COWFORK)
WRITE_ONCE(ifp->if_seq, READ_ONCE(ifp->if_seq) + 1);
}
void
xfs_iext_insert(
struct xfs_inode *ip,
xfs: use a b+tree for the in-core extent list Replace the current linear list and the indirection array for the in-core extent list with a b+tree to avoid the need for larger memory allocations for the indirection array when lots of extents are present. The current extent list implementations leads to heavy pressure on the memory allocator when modifying files with a high extent count, and can lead to high latencies because of that. The replacement is a b+tree with a few quirks. The leaf nodes directly store the extent record in two u64 values. The encoding is a little bit different from the existing in-core extent records so that the start offset and length which are required for lookups can be retreived with simple mask operations. The inner nodes store a 64-bit key containing the start offset in the first half of the node, and the pointers to the next lower level in the second half. In either case we walk the node from the beginninig to the end and do a linear search, as that is more efficient for the low number of cache lines touched during a search (2 for the inner nodes, 4 for the leaf nodes) than a binary search. We store termination markers (zero length for the leaf nodes, an otherwise impossible high bit for the inner nodes) to terminate the key list / records instead of storing a count to use the available cache lines as efficiently as possible. One quirk of the algorithm is that while we normally split a node half and half like usual btree implementations we just spill over entries added at the very end of the list to a new node on its own. This means we get a 100% fill grade for the common cases of bulk insertion when reading an inode into memory, and when only sequentially appending to a file. The downside is a slightly higher chance of splits on the first random insertions. Both insert and removal manually recurse into the lower levels, but the bulk deletion of the whole tree is still implemented as a recursive function call, although one limited by the overall depth and with very little stack usage in every iteration. For the first few extents we dynamically grow the list from a single extent to the next powers of two until we have a first full leaf block and that building the actual tree. The code started out based on the generic lib/btree.c code from Joern Engel based on earlier work from Peter Zijlstra, but has since been rewritten beyond recognition. Signed-off-by: Christoph Hellwig <hch@lst.de> Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com> Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
2017-11-04 01:34:46 +08:00
struct xfs_iext_cursor *cur,
struct xfs_bmbt_irec *irec,
int state)
xfs: use a b+tree for the in-core extent list Replace the current linear list and the indirection array for the in-core extent list with a b+tree to avoid the need for larger memory allocations for the indirection array when lots of extents are present. The current extent list implementations leads to heavy pressure on the memory allocator when modifying files with a high extent count, and can lead to high latencies because of that. The replacement is a b+tree with a few quirks. The leaf nodes directly store the extent record in two u64 values. The encoding is a little bit different from the existing in-core extent records so that the start offset and length which are required for lookups can be retreived with simple mask operations. The inner nodes store a 64-bit key containing the start offset in the first half of the node, and the pointers to the next lower level in the second half. In either case we walk the node from the beginninig to the end and do a linear search, as that is more efficient for the low number of cache lines touched during a search (2 for the inner nodes, 4 for the leaf nodes) than a binary search. We store termination markers (zero length for the leaf nodes, an otherwise impossible high bit for the inner nodes) to terminate the key list / records instead of storing a count to use the available cache lines as efficiently as possible. One quirk of the algorithm is that while we normally split a node half and half like usual btree implementations we just spill over entries added at the very end of the list to a new node on its own. This means we get a 100% fill grade for the common cases of bulk insertion when reading an inode into memory, and when only sequentially appending to a file. The downside is a slightly higher chance of splits on the first random insertions. Both insert and removal manually recurse into the lower levels, but the bulk deletion of the whole tree is still implemented as a recursive function call, although one limited by the overall depth and with very little stack usage in every iteration. For the first few extents we dynamically grow the list from a single extent to the next powers of two until we have a first full leaf block and that building the actual tree. The code started out based on the generic lib/btree.c code from Joern Engel based on earlier work from Peter Zijlstra, but has since been rewritten beyond recognition. Signed-off-by: Christoph Hellwig <hch@lst.de> Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com> Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
2017-11-04 01:34:46 +08:00
{
struct xfs_ifork *ifp = xfs_iext_state_to_fork(ip, state);
xfs: use a b+tree for the in-core extent list Replace the current linear list and the indirection array for the in-core extent list with a b+tree to avoid the need for larger memory allocations for the indirection array when lots of extents are present. The current extent list implementations leads to heavy pressure on the memory allocator when modifying files with a high extent count, and can lead to high latencies because of that. The replacement is a b+tree with a few quirks. The leaf nodes directly store the extent record in two u64 values. The encoding is a little bit different from the existing in-core extent records so that the start offset and length which are required for lookups can be retreived with simple mask operations. The inner nodes store a 64-bit key containing the start offset in the first half of the node, and the pointers to the next lower level in the second half. In either case we walk the node from the beginninig to the end and do a linear search, as that is more efficient for the low number of cache lines touched during a search (2 for the inner nodes, 4 for the leaf nodes) than a binary search. We store termination markers (zero length for the leaf nodes, an otherwise impossible high bit for the inner nodes) to terminate the key list / records instead of storing a count to use the available cache lines as efficiently as possible. One quirk of the algorithm is that while we normally split a node half and half like usual btree implementations we just spill over entries added at the very end of the list to a new node on its own. This means we get a 100% fill grade for the common cases of bulk insertion when reading an inode into memory, and when only sequentially appending to a file. The downside is a slightly higher chance of splits on the first random insertions. Both insert and removal manually recurse into the lower levels, but the bulk deletion of the whole tree is still implemented as a recursive function call, although one limited by the overall depth and with very little stack usage in every iteration. For the first few extents we dynamically grow the list from a single extent to the next powers of two until we have a first full leaf block and that building the actual tree. The code started out based on the generic lib/btree.c code from Joern Engel based on earlier work from Peter Zijlstra, but has since been rewritten beyond recognition. Signed-off-by: Christoph Hellwig <hch@lst.de> Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com> Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
2017-11-04 01:34:46 +08:00
xfs_fileoff_t offset = irec->br_startoff;
struct xfs_iext_leaf *new = NULL;
int nr_entries, i;
xfs_iext_inc_seq(ifp, state);
xfs: use a b+tree for the in-core extent list Replace the current linear list and the indirection array for the in-core extent list with a b+tree to avoid the need for larger memory allocations for the indirection array when lots of extents are present. The current extent list implementations leads to heavy pressure on the memory allocator when modifying files with a high extent count, and can lead to high latencies because of that. The replacement is a b+tree with a few quirks. The leaf nodes directly store the extent record in two u64 values. The encoding is a little bit different from the existing in-core extent records so that the start offset and length which are required for lookups can be retreived with simple mask operations. The inner nodes store a 64-bit key containing the start offset in the first half of the node, and the pointers to the next lower level in the second half. In either case we walk the node from the beginninig to the end and do a linear search, as that is more efficient for the low number of cache lines touched during a search (2 for the inner nodes, 4 for the leaf nodes) than a binary search. We store termination markers (zero length for the leaf nodes, an otherwise impossible high bit for the inner nodes) to terminate the key list / records instead of storing a count to use the available cache lines as efficiently as possible. One quirk of the algorithm is that while we normally split a node half and half like usual btree implementations we just spill over entries added at the very end of the list to a new node on its own. This means we get a 100% fill grade for the common cases of bulk insertion when reading an inode into memory, and when only sequentially appending to a file. The downside is a slightly higher chance of splits on the first random insertions. Both insert and removal manually recurse into the lower levels, but the bulk deletion of the whole tree is still implemented as a recursive function call, although one limited by the overall depth and with very little stack usage in every iteration. For the first few extents we dynamically grow the list from a single extent to the next powers of two until we have a first full leaf block and that building the actual tree. The code started out based on the generic lib/btree.c code from Joern Engel based on earlier work from Peter Zijlstra, but has since been rewritten beyond recognition. Signed-off-by: Christoph Hellwig <hch@lst.de> Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com> Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
2017-11-04 01:34:46 +08:00
if (ifp->if_height == 0)
xfs_iext_alloc_root(ifp, cur);
else if (ifp->if_height == 1)
xfs_iext_realloc_root(ifp, cur);
nr_entries = xfs_iext_leaf_nr_entries(ifp, cur->leaf, cur->pos);
ASSERT(nr_entries <= RECS_PER_LEAF);
ASSERT(cur->pos >= nr_entries ||
xfs_iext_rec_cmp(cur_rec(cur), irec->br_startoff) != 0);
if (nr_entries == RECS_PER_LEAF)
new = xfs_iext_split_leaf(cur, &nr_entries);
/*
* Update the pointers in higher levels if the first entry changes
* in an existing node.
*/
xfs: use a b+tree for the in-core extent list Replace the current linear list and the indirection array for the in-core extent list with a b+tree to avoid the need for larger memory allocations for the indirection array when lots of extents are present. The current extent list implementations leads to heavy pressure on the memory allocator when modifying files with a high extent count, and can lead to high latencies because of that. The replacement is a b+tree with a few quirks. The leaf nodes directly store the extent record in two u64 values. The encoding is a little bit different from the existing in-core extent records so that the start offset and length which are required for lookups can be retreived with simple mask operations. The inner nodes store a 64-bit key containing the start offset in the first half of the node, and the pointers to the next lower level in the second half. In either case we walk the node from the beginninig to the end and do a linear search, as that is more efficient for the low number of cache lines touched during a search (2 for the inner nodes, 4 for the leaf nodes) than a binary search. We store termination markers (zero length for the leaf nodes, an otherwise impossible high bit for the inner nodes) to terminate the key list / records instead of storing a count to use the available cache lines as efficiently as possible. One quirk of the algorithm is that while we normally split a node half and half like usual btree implementations we just spill over entries added at the very end of the list to a new node on its own. This means we get a 100% fill grade for the common cases of bulk insertion when reading an inode into memory, and when only sequentially appending to a file. The downside is a slightly higher chance of splits on the first random insertions. Both insert and removal manually recurse into the lower levels, but the bulk deletion of the whole tree is still implemented as a recursive function call, although one limited by the overall depth and with very little stack usage in every iteration. For the first few extents we dynamically grow the list from a single extent to the next powers of two until we have a first full leaf block and that building the actual tree. The code started out based on the generic lib/btree.c code from Joern Engel based on earlier work from Peter Zijlstra, but has since been rewritten beyond recognition. Signed-off-by: Christoph Hellwig <hch@lst.de> Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com> Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
2017-11-04 01:34:46 +08:00
if (cur->leaf != new && cur->pos == 0 && nr_entries > 0) {
xfs_iext_update_node(ifp, xfs_iext_leaf_key(cur->leaf, 0),
offset, 1, cur->leaf);
}
for (i = nr_entries; i > cur->pos; i--)
cur->leaf->recs[i] = cur->leaf->recs[i - 1];
xfs_iext_set(cur_rec(cur), irec);
ifp->if_bytes += sizeof(struct xfs_iext_rec);
trace_xfs_iext_insert(ip, cur, state, _RET_IP_);
xfs: use a b+tree for the in-core extent list Replace the current linear list and the indirection array for the in-core extent list with a b+tree to avoid the need for larger memory allocations for the indirection array when lots of extents are present. The current extent list implementations leads to heavy pressure on the memory allocator when modifying files with a high extent count, and can lead to high latencies because of that. The replacement is a b+tree with a few quirks. The leaf nodes directly store the extent record in two u64 values. The encoding is a little bit different from the existing in-core extent records so that the start offset and length which are required for lookups can be retreived with simple mask operations. The inner nodes store a 64-bit key containing the start offset in the first half of the node, and the pointers to the next lower level in the second half. In either case we walk the node from the beginninig to the end and do a linear search, as that is more efficient for the low number of cache lines touched during a search (2 for the inner nodes, 4 for the leaf nodes) than a binary search. We store termination markers (zero length for the leaf nodes, an otherwise impossible high bit for the inner nodes) to terminate the key list / records instead of storing a count to use the available cache lines as efficiently as possible. One quirk of the algorithm is that while we normally split a node half and half like usual btree implementations we just spill over entries added at the very end of the list to a new node on its own. This means we get a 100% fill grade for the common cases of bulk insertion when reading an inode into memory, and when only sequentially appending to a file. The downside is a slightly higher chance of splits on the first random insertions. Both insert and removal manually recurse into the lower levels, but the bulk deletion of the whole tree is still implemented as a recursive function call, although one limited by the overall depth and with very little stack usage in every iteration. For the first few extents we dynamically grow the list from a single extent to the next powers of two until we have a first full leaf block and that building the actual tree. The code started out based on the generic lib/btree.c code from Joern Engel based on earlier work from Peter Zijlstra, but has since been rewritten beyond recognition. Signed-off-by: Christoph Hellwig <hch@lst.de> Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com> Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
2017-11-04 01:34:46 +08:00
if (new)
xfs_iext_insert_node(ifp, xfs_iext_leaf_key(new, 0), new, 2);
}
static struct xfs_iext_node *
xfs_iext_rebalance_node(
struct xfs_iext_node *parent,
int *pos,
struct xfs_iext_node *node,
int nr_entries)
{
/*
* If the neighbouring nodes are completely full, or have different
* parents, we might never be able to merge our node, and will only
* delete it once the number of entries hits zero.
*/
xfs: use a b+tree for the in-core extent list Replace the current linear list and the indirection array for the in-core extent list with a b+tree to avoid the need for larger memory allocations for the indirection array when lots of extents are present. The current extent list implementations leads to heavy pressure on the memory allocator when modifying files with a high extent count, and can lead to high latencies because of that. The replacement is a b+tree with a few quirks. The leaf nodes directly store the extent record in two u64 values. The encoding is a little bit different from the existing in-core extent records so that the start offset and length which are required for lookups can be retreived with simple mask operations. The inner nodes store a 64-bit key containing the start offset in the first half of the node, and the pointers to the next lower level in the second half. In either case we walk the node from the beginninig to the end and do a linear search, as that is more efficient for the low number of cache lines touched during a search (2 for the inner nodes, 4 for the leaf nodes) than a binary search. We store termination markers (zero length for the leaf nodes, an otherwise impossible high bit for the inner nodes) to terminate the key list / records instead of storing a count to use the available cache lines as efficiently as possible. One quirk of the algorithm is that while we normally split a node half and half like usual btree implementations we just spill over entries added at the very end of the list to a new node on its own. This means we get a 100% fill grade for the common cases of bulk insertion when reading an inode into memory, and when only sequentially appending to a file. The downside is a slightly higher chance of splits on the first random insertions. Both insert and removal manually recurse into the lower levels, but the bulk deletion of the whole tree is still implemented as a recursive function call, although one limited by the overall depth and with very little stack usage in every iteration. For the first few extents we dynamically grow the list from a single extent to the next powers of two until we have a first full leaf block and that building the actual tree. The code started out based on the generic lib/btree.c code from Joern Engel based on earlier work from Peter Zijlstra, but has since been rewritten beyond recognition. Signed-off-by: Christoph Hellwig <hch@lst.de> Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com> Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
2017-11-04 01:34:46 +08:00
if (nr_entries == 0)
return node;
if (*pos > 0) {
struct xfs_iext_node *prev = parent->ptrs[*pos - 1];
int nr_prev = xfs_iext_node_nr_entries(prev, 0), i;
if (nr_prev + nr_entries <= KEYS_PER_NODE) {
for (i = 0; i < nr_entries; i++) {
prev->keys[nr_prev + i] = node->keys[i];
prev->ptrs[nr_prev + i] = node->ptrs[i];
}
return node;
}
}
if (*pos + 1 < xfs_iext_node_nr_entries(parent, *pos)) {
struct xfs_iext_node *next = parent->ptrs[*pos + 1];
int nr_next = xfs_iext_node_nr_entries(next, 0), i;
if (nr_entries + nr_next <= KEYS_PER_NODE) {
/*
* Merge the next node into this node so that we don't
* have to do an additional update of the keys in the
* higher levels.
*/
xfs: use a b+tree for the in-core extent list Replace the current linear list and the indirection array for the in-core extent list with a b+tree to avoid the need for larger memory allocations for the indirection array when lots of extents are present. The current extent list implementations leads to heavy pressure on the memory allocator when modifying files with a high extent count, and can lead to high latencies because of that. The replacement is a b+tree with a few quirks. The leaf nodes directly store the extent record in two u64 values. The encoding is a little bit different from the existing in-core extent records so that the start offset and length which are required for lookups can be retreived with simple mask operations. The inner nodes store a 64-bit key containing the start offset in the first half of the node, and the pointers to the next lower level in the second half. In either case we walk the node from the beginninig to the end and do a linear search, as that is more efficient for the low number of cache lines touched during a search (2 for the inner nodes, 4 for the leaf nodes) than a binary search. We store termination markers (zero length for the leaf nodes, an otherwise impossible high bit for the inner nodes) to terminate the key list / records instead of storing a count to use the available cache lines as efficiently as possible. One quirk of the algorithm is that while we normally split a node half and half like usual btree implementations we just spill over entries added at the very end of the list to a new node on its own. This means we get a 100% fill grade for the common cases of bulk insertion when reading an inode into memory, and when only sequentially appending to a file. The downside is a slightly higher chance of splits on the first random insertions. Both insert and removal manually recurse into the lower levels, but the bulk deletion of the whole tree is still implemented as a recursive function call, although one limited by the overall depth and with very little stack usage in every iteration. For the first few extents we dynamically grow the list from a single extent to the next powers of two until we have a first full leaf block and that building the actual tree. The code started out based on the generic lib/btree.c code from Joern Engel based on earlier work from Peter Zijlstra, but has since been rewritten beyond recognition. Signed-off-by: Christoph Hellwig <hch@lst.de> Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com> Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
2017-11-04 01:34:46 +08:00
for (i = 0; i < nr_next; i++) {
node->keys[nr_entries + i] = next->keys[i];
node->ptrs[nr_entries + i] = next->ptrs[i];
}
++*pos;
return next;
}
}
return NULL;
}
static void
xfs_iext_remove_node(
struct xfs_ifork *ifp,
xfs_fileoff_t offset,
void *victim)
{
struct xfs_iext_node *node, *parent;
int level = 2, pos, nr_entries, i;
ASSERT(level <= ifp->if_height);
node = xfs_iext_find_level(ifp, offset, level);
pos = xfs_iext_node_pos(node, offset);
again:
ASSERT(node->ptrs[pos]);
ASSERT(node->ptrs[pos] == victim);
kmem_free(victim);
nr_entries = xfs_iext_node_nr_entries(node, pos) - 1;
offset = node->keys[0];
for (i = pos; i < nr_entries; i++) {
node->keys[i] = node->keys[i + 1];
node->ptrs[i] = node->ptrs[i + 1];
}
node->keys[nr_entries] = XFS_IEXT_KEY_INVALID;
node->ptrs[nr_entries] = NULL;
if (pos == 0 && nr_entries > 0) {
xfs_iext_update_node(ifp, offset, node->keys[0], level, node);
xfs: use a b+tree for the in-core extent list Replace the current linear list and the indirection array for the in-core extent list with a b+tree to avoid the need for larger memory allocations for the indirection array when lots of extents are present. The current extent list implementations leads to heavy pressure on the memory allocator when modifying files with a high extent count, and can lead to high latencies because of that. The replacement is a b+tree with a few quirks. The leaf nodes directly store the extent record in two u64 values. The encoding is a little bit different from the existing in-core extent records so that the start offset and length which are required for lookups can be retreived with simple mask operations. The inner nodes store a 64-bit key containing the start offset in the first half of the node, and the pointers to the next lower level in the second half. In either case we walk the node from the beginninig to the end and do a linear search, as that is more efficient for the low number of cache lines touched during a search (2 for the inner nodes, 4 for the leaf nodes) than a binary search. We store termination markers (zero length for the leaf nodes, an otherwise impossible high bit for the inner nodes) to terminate the key list / records instead of storing a count to use the available cache lines as efficiently as possible. One quirk of the algorithm is that while we normally split a node half and half like usual btree implementations we just spill over entries added at the very end of the list to a new node on its own. This means we get a 100% fill grade for the common cases of bulk insertion when reading an inode into memory, and when only sequentially appending to a file. The downside is a slightly higher chance of splits on the first random insertions. Both insert and removal manually recurse into the lower levels, but the bulk deletion of the whole tree is still implemented as a recursive function call, although one limited by the overall depth and with very little stack usage in every iteration. For the first few extents we dynamically grow the list from a single extent to the next powers of two until we have a first full leaf block and that building the actual tree. The code started out based on the generic lib/btree.c code from Joern Engel based on earlier work from Peter Zijlstra, but has since been rewritten beyond recognition. Signed-off-by: Christoph Hellwig <hch@lst.de> Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com> Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
2017-11-04 01:34:46 +08:00
offset = node->keys[0];
}
if (nr_entries >= KEYS_PER_NODE / 2)
return;
if (level < ifp->if_height) {
/*
* If we aren't at the root yet try to find a neighbour node to
* merge with (or delete the node if it is empty), and then
* recurse up to the next level.
*/
xfs: use a b+tree for the in-core extent list Replace the current linear list and the indirection array for the in-core extent list with a b+tree to avoid the need for larger memory allocations for the indirection array when lots of extents are present. The current extent list implementations leads to heavy pressure on the memory allocator when modifying files with a high extent count, and can lead to high latencies because of that. The replacement is a b+tree with a few quirks. The leaf nodes directly store the extent record in two u64 values. The encoding is a little bit different from the existing in-core extent records so that the start offset and length which are required for lookups can be retreived with simple mask operations. The inner nodes store a 64-bit key containing the start offset in the first half of the node, and the pointers to the next lower level in the second half. In either case we walk the node from the beginninig to the end and do a linear search, as that is more efficient for the low number of cache lines touched during a search (2 for the inner nodes, 4 for the leaf nodes) than a binary search. We store termination markers (zero length for the leaf nodes, an otherwise impossible high bit for the inner nodes) to terminate the key list / records instead of storing a count to use the available cache lines as efficiently as possible. One quirk of the algorithm is that while we normally split a node half and half like usual btree implementations we just spill over entries added at the very end of the list to a new node on its own. This means we get a 100% fill grade for the common cases of bulk insertion when reading an inode into memory, and when only sequentially appending to a file. The downside is a slightly higher chance of splits on the first random insertions. Both insert and removal manually recurse into the lower levels, but the bulk deletion of the whole tree is still implemented as a recursive function call, although one limited by the overall depth and with very little stack usage in every iteration. For the first few extents we dynamically grow the list from a single extent to the next powers of two until we have a first full leaf block and that building the actual tree. The code started out based on the generic lib/btree.c code from Joern Engel based on earlier work from Peter Zijlstra, but has since been rewritten beyond recognition. Signed-off-by: Christoph Hellwig <hch@lst.de> Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com> Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
2017-11-04 01:34:46 +08:00
level++;
parent = xfs_iext_find_level(ifp, offset, level);
pos = xfs_iext_node_pos(parent, offset);
ASSERT(pos != KEYS_PER_NODE);
ASSERT(parent->ptrs[pos] == node);
node = xfs_iext_rebalance_node(parent, &pos, node, nr_entries);
if (node) {
victim = node;
node = parent;
goto again;
}
} else if (nr_entries == 1) {
/*
* If we are at the root and only one entry is left we can just
* free this node and update the root pointer.
*/
xfs: use a b+tree for the in-core extent list Replace the current linear list and the indirection array for the in-core extent list with a b+tree to avoid the need for larger memory allocations for the indirection array when lots of extents are present. The current extent list implementations leads to heavy pressure on the memory allocator when modifying files with a high extent count, and can lead to high latencies because of that. The replacement is a b+tree with a few quirks. The leaf nodes directly store the extent record in two u64 values. The encoding is a little bit different from the existing in-core extent records so that the start offset and length which are required for lookups can be retreived with simple mask operations. The inner nodes store a 64-bit key containing the start offset in the first half of the node, and the pointers to the next lower level in the second half. In either case we walk the node from the beginninig to the end and do a linear search, as that is more efficient for the low number of cache lines touched during a search (2 for the inner nodes, 4 for the leaf nodes) than a binary search. We store termination markers (zero length for the leaf nodes, an otherwise impossible high bit for the inner nodes) to terminate the key list / records instead of storing a count to use the available cache lines as efficiently as possible. One quirk of the algorithm is that while we normally split a node half and half like usual btree implementations we just spill over entries added at the very end of the list to a new node on its own. This means we get a 100% fill grade for the common cases of bulk insertion when reading an inode into memory, and when only sequentially appending to a file. The downside is a slightly higher chance of splits on the first random insertions. Both insert and removal manually recurse into the lower levels, but the bulk deletion of the whole tree is still implemented as a recursive function call, although one limited by the overall depth and with very little stack usage in every iteration. For the first few extents we dynamically grow the list from a single extent to the next powers of two until we have a first full leaf block and that building the actual tree. The code started out based on the generic lib/btree.c code from Joern Engel based on earlier work from Peter Zijlstra, but has since been rewritten beyond recognition. Signed-off-by: Christoph Hellwig <hch@lst.de> Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com> Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
2017-11-04 01:34:46 +08:00
ASSERT(node == ifp->if_u1.if_root);
ifp->if_u1.if_root = node->ptrs[0];
ifp->if_height--;
kmem_free(node);
}
}
static void
xfs_iext_rebalance_leaf(
struct xfs_ifork *ifp,
struct xfs_iext_cursor *cur,
struct xfs_iext_leaf *leaf,
xfs_fileoff_t offset,
int nr_entries)
xfs: use a b+tree for the in-core extent list Replace the current linear list and the indirection array for the in-core extent list with a b+tree to avoid the need for larger memory allocations for the indirection array when lots of extents are present. The current extent list implementations leads to heavy pressure on the memory allocator when modifying files with a high extent count, and can lead to high latencies because of that. The replacement is a b+tree with a few quirks. The leaf nodes directly store the extent record in two u64 values. The encoding is a little bit different from the existing in-core extent records so that the start offset and length which are required for lookups can be retreived with simple mask operations. The inner nodes store a 64-bit key containing the start offset in the first half of the node, and the pointers to the next lower level in the second half. In either case we walk the node from the beginninig to the end and do a linear search, as that is more efficient for the low number of cache lines touched during a search (2 for the inner nodes, 4 for the leaf nodes) than a binary search. We store termination markers (zero length for the leaf nodes, an otherwise impossible high bit for the inner nodes) to terminate the key list / records instead of storing a count to use the available cache lines as efficiently as possible. One quirk of the algorithm is that while we normally split a node half and half like usual btree implementations we just spill over entries added at the very end of the list to a new node on its own. This means we get a 100% fill grade for the common cases of bulk insertion when reading an inode into memory, and when only sequentially appending to a file. The downside is a slightly higher chance of splits on the first random insertions. Both insert and removal manually recurse into the lower levels, but the bulk deletion of the whole tree is still implemented as a recursive function call, although one limited by the overall depth and with very little stack usage in every iteration. For the first few extents we dynamically grow the list from a single extent to the next powers of two until we have a first full leaf block and that building the actual tree. The code started out based on the generic lib/btree.c code from Joern Engel based on earlier work from Peter Zijlstra, but has since been rewritten beyond recognition. Signed-off-by: Christoph Hellwig <hch@lst.de> Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com> Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
2017-11-04 01:34:46 +08:00
{
/*
* If the neighbouring nodes are completely full we might never be able
* to merge our node, and will only delete it once the number of
* entries hits zero.
*/
if (nr_entries == 0)
goto remove_node;
xfs: use a b+tree for the in-core extent list Replace the current linear list and the indirection array for the in-core extent list with a b+tree to avoid the need for larger memory allocations for the indirection array when lots of extents are present. The current extent list implementations leads to heavy pressure on the memory allocator when modifying files with a high extent count, and can lead to high latencies because of that. The replacement is a b+tree with a few quirks. The leaf nodes directly store the extent record in two u64 values. The encoding is a little bit different from the existing in-core extent records so that the start offset and length which are required for lookups can be retreived with simple mask operations. The inner nodes store a 64-bit key containing the start offset in the first half of the node, and the pointers to the next lower level in the second half. In either case we walk the node from the beginninig to the end and do a linear search, as that is more efficient for the low number of cache lines touched during a search (2 for the inner nodes, 4 for the leaf nodes) than a binary search. We store termination markers (zero length for the leaf nodes, an otherwise impossible high bit for the inner nodes) to terminate the key list / records instead of storing a count to use the available cache lines as efficiently as possible. One quirk of the algorithm is that while we normally split a node half and half like usual btree implementations we just spill over entries added at the very end of the list to a new node on its own. This means we get a 100% fill grade for the common cases of bulk insertion when reading an inode into memory, and when only sequentially appending to a file. The downside is a slightly higher chance of splits on the first random insertions. Both insert and removal manually recurse into the lower levels, but the bulk deletion of the whole tree is still implemented as a recursive function call, although one limited by the overall depth and with very little stack usage in every iteration. For the first few extents we dynamically grow the list from a single extent to the next powers of two until we have a first full leaf block and that building the actual tree. The code started out based on the generic lib/btree.c code from Joern Engel based on earlier work from Peter Zijlstra, but has since been rewritten beyond recognition. Signed-off-by: Christoph Hellwig <hch@lst.de> Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com> Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
2017-11-04 01:34:46 +08:00
if (leaf->prev) {
int nr_prev = xfs_iext_leaf_nr_entries(ifp, leaf->prev, 0), i;
if (nr_prev + nr_entries <= RECS_PER_LEAF) {
for (i = 0; i < nr_entries; i++)
xfs: use a b+tree for the in-core extent list Replace the current linear list and the indirection array for the in-core extent list with a b+tree to avoid the need for larger memory allocations for the indirection array when lots of extents are present. The current extent list implementations leads to heavy pressure on the memory allocator when modifying files with a high extent count, and can lead to high latencies because of that. The replacement is a b+tree with a few quirks. The leaf nodes directly store the extent record in two u64 values. The encoding is a little bit different from the existing in-core extent records so that the start offset and length which are required for lookups can be retreived with simple mask operations. The inner nodes store a 64-bit key containing the start offset in the first half of the node, and the pointers to the next lower level in the second half. In either case we walk the node from the beginninig to the end and do a linear search, as that is more efficient for the low number of cache lines touched during a search (2 for the inner nodes, 4 for the leaf nodes) than a binary search. We store termination markers (zero length for the leaf nodes, an otherwise impossible high bit for the inner nodes) to terminate the key list / records instead of storing a count to use the available cache lines as efficiently as possible. One quirk of the algorithm is that while we normally split a node half and half like usual btree implementations we just spill over entries added at the very end of the list to a new node on its own. This means we get a 100% fill grade for the common cases of bulk insertion when reading an inode into memory, and when only sequentially appending to a file. The downside is a slightly higher chance of splits on the first random insertions. Both insert and removal manually recurse into the lower levels, but the bulk deletion of the whole tree is still implemented as a recursive function call, although one limited by the overall depth and with very little stack usage in every iteration. For the first few extents we dynamically grow the list from a single extent to the next powers of two until we have a first full leaf block and that building the actual tree. The code started out based on the generic lib/btree.c code from Joern Engel based on earlier work from Peter Zijlstra, but has since been rewritten beyond recognition. Signed-off-by: Christoph Hellwig <hch@lst.de> Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com> Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
2017-11-04 01:34:46 +08:00
leaf->prev->recs[nr_prev + i] = leaf->recs[i];
if (cur->leaf == leaf) {
cur->leaf = leaf->prev;
cur->pos += nr_prev;
}
goto remove_node;
}
}
if (leaf->next) {
int nr_next = xfs_iext_leaf_nr_entries(ifp, leaf->next, 0), i;
if (nr_entries + nr_next <= RECS_PER_LEAF) {
/*
* Merge the next node into this node so that we don't
* have to do an additional update of the keys in the
* higher levels.
*/
for (i = 0; i < nr_next; i++) {
leaf->recs[nr_entries + i] =
leaf->next->recs[i];
}
xfs: use a b+tree for the in-core extent list Replace the current linear list and the indirection array for the in-core extent list with a b+tree to avoid the need for larger memory allocations for the indirection array when lots of extents are present. The current extent list implementations leads to heavy pressure on the memory allocator when modifying files with a high extent count, and can lead to high latencies because of that. The replacement is a b+tree with a few quirks. The leaf nodes directly store the extent record in two u64 values. The encoding is a little bit different from the existing in-core extent records so that the start offset and length which are required for lookups can be retreived with simple mask operations. The inner nodes store a 64-bit key containing the start offset in the first half of the node, and the pointers to the next lower level in the second half. In either case we walk the node from the beginninig to the end and do a linear search, as that is more efficient for the low number of cache lines touched during a search (2 for the inner nodes, 4 for the leaf nodes) than a binary search. We store termination markers (zero length for the leaf nodes, an otherwise impossible high bit for the inner nodes) to terminate the key list / records instead of storing a count to use the available cache lines as efficiently as possible. One quirk of the algorithm is that while we normally split a node half and half like usual btree implementations we just spill over entries added at the very end of the list to a new node on its own. This means we get a 100% fill grade for the common cases of bulk insertion when reading an inode into memory, and when only sequentially appending to a file. The downside is a slightly higher chance of splits on the first random insertions. Both insert and removal manually recurse into the lower levels, but the bulk deletion of the whole tree is still implemented as a recursive function call, although one limited by the overall depth and with very little stack usage in every iteration. For the first few extents we dynamically grow the list from a single extent to the next powers of two until we have a first full leaf block and that building the actual tree. The code started out based on the generic lib/btree.c code from Joern Engel based on earlier work from Peter Zijlstra, but has since been rewritten beyond recognition. Signed-off-by: Christoph Hellwig <hch@lst.de> Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com> Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
2017-11-04 01:34:46 +08:00
if (cur->leaf == leaf->next) {
cur->leaf = leaf;
cur->pos += nr_entries;
xfs: use a b+tree for the in-core extent list Replace the current linear list and the indirection array for the in-core extent list with a b+tree to avoid the need for larger memory allocations for the indirection array when lots of extents are present. The current extent list implementations leads to heavy pressure on the memory allocator when modifying files with a high extent count, and can lead to high latencies because of that. The replacement is a b+tree with a few quirks. The leaf nodes directly store the extent record in two u64 values. The encoding is a little bit different from the existing in-core extent records so that the start offset and length which are required for lookups can be retreived with simple mask operations. The inner nodes store a 64-bit key containing the start offset in the first half of the node, and the pointers to the next lower level in the second half. In either case we walk the node from the beginninig to the end and do a linear search, as that is more efficient for the low number of cache lines touched during a search (2 for the inner nodes, 4 for the leaf nodes) than a binary search. We store termination markers (zero length for the leaf nodes, an otherwise impossible high bit for the inner nodes) to terminate the key list / records instead of storing a count to use the available cache lines as efficiently as possible. One quirk of the algorithm is that while we normally split a node half and half like usual btree implementations we just spill over entries added at the very end of the list to a new node on its own. This means we get a 100% fill grade for the common cases of bulk insertion when reading an inode into memory, and when only sequentially appending to a file. The downside is a slightly higher chance of splits on the first random insertions. Both insert and removal manually recurse into the lower levels, but the bulk deletion of the whole tree is still implemented as a recursive function call, although one limited by the overall depth and with very little stack usage in every iteration. For the first few extents we dynamically grow the list from a single extent to the next powers of two until we have a first full leaf block and that building the actual tree. The code started out based on the generic lib/btree.c code from Joern Engel based on earlier work from Peter Zijlstra, but has since been rewritten beyond recognition. Signed-off-by: Christoph Hellwig <hch@lst.de> Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com> Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
2017-11-04 01:34:46 +08:00
}
offset = xfs_iext_leaf_key(leaf->next, 0);
leaf = leaf->next;
goto remove_node;
}
}
return;
remove_node:
if (leaf->prev)
leaf->prev->next = leaf->next;
if (leaf->next)
leaf->next->prev = leaf->prev;
xfs_iext_remove_node(ifp, offset, leaf);
}
static void
xfs_iext_free_last_leaf(
struct xfs_ifork *ifp)
{
ifp->if_height--;
kmem_free(ifp->if_u1.if_root);
ifp->if_u1.if_root = NULL;
xfs: use a b+tree for the in-core extent list Replace the current linear list and the indirection array for the in-core extent list with a b+tree to avoid the need for larger memory allocations for the indirection array when lots of extents are present. The current extent list implementations leads to heavy pressure on the memory allocator when modifying files with a high extent count, and can lead to high latencies because of that. The replacement is a b+tree with a few quirks. The leaf nodes directly store the extent record in two u64 values. The encoding is a little bit different from the existing in-core extent records so that the start offset and length which are required for lookups can be retreived with simple mask operations. The inner nodes store a 64-bit key containing the start offset in the first half of the node, and the pointers to the next lower level in the second half. In either case we walk the node from the beginninig to the end and do a linear search, as that is more efficient for the low number of cache lines touched during a search (2 for the inner nodes, 4 for the leaf nodes) than a binary search. We store termination markers (zero length for the leaf nodes, an otherwise impossible high bit for the inner nodes) to terminate the key list / records instead of storing a count to use the available cache lines as efficiently as possible. One quirk of the algorithm is that while we normally split a node half and half like usual btree implementations we just spill over entries added at the very end of the list to a new node on its own. This means we get a 100% fill grade for the common cases of bulk insertion when reading an inode into memory, and when only sequentially appending to a file. The downside is a slightly higher chance of splits on the first random insertions. Both insert and removal manually recurse into the lower levels, but the bulk deletion of the whole tree is still implemented as a recursive function call, although one limited by the overall depth and with very little stack usage in every iteration. For the first few extents we dynamically grow the list from a single extent to the next powers of two until we have a first full leaf block and that building the actual tree. The code started out based on the generic lib/btree.c code from Joern Engel based on earlier work from Peter Zijlstra, but has since been rewritten beyond recognition. Signed-off-by: Christoph Hellwig <hch@lst.de> Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com> Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
2017-11-04 01:34:46 +08:00
}
void
xfs_iext_remove(
struct xfs_inode *ip,
struct xfs_iext_cursor *cur,
int state)
xfs: use a b+tree for the in-core extent list Replace the current linear list and the indirection array for the in-core extent list with a b+tree to avoid the need for larger memory allocations for the indirection array when lots of extents are present. The current extent list implementations leads to heavy pressure on the memory allocator when modifying files with a high extent count, and can lead to high latencies because of that. The replacement is a b+tree with a few quirks. The leaf nodes directly store the extent record in two u64 values. The encoding is a little bit different from the existing in-core extent records so that the start offset and length which are required for lookups can be retreived with simple mask operations. The inner nodes store a 64-bit key containing the start offset in the first half of the node, and the pointers to the next lower level in the second half. In either case we walk the node from the beginninig to the end and do a linear search, as that is more efficient for the low number of cache lines touched during a search (2 for the inner nodes, 4 for the leaf nodes) than a binary search. We store termination markers (zero length for the leaf nodes, an otherwise impossible high bit for the inner nodes) to terminate the key list / records instead of storing a count to use the available cache lines as efficiently as possible. One quirk of the algorithm is that while we normally split a node half and half like usual btree implementations we just spill over entries added at the very end of the list to a new node on its own. This means we get a 100% fill grade for the common cases of bulk insertion when reading an inode into memory, and when only sequentially appending to a file. The downside is a slightly higher chance of splits on the first random insertions. Both insert and removal manually recurse into the lower levels, but the bulk deletion of the whole tree is still implemented as a recursive function call, although one limited by the overall depth and with very little stack usage in every iteration. For the first few extents we dynamically grow the list from a single extent to the next powers of two until we have a first full leaf block and that building the actual tree. The code started out based on the generic lib/btree.c code from Joern Engel based on earlier work from Peter Zijlstra, but has since been rewritten beyond recognition. Signed-off-by: Christoph Hellwig <hch@lst.de> Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com> Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
2017-11-04 01:34:46 +08:00
{
struct xfs_ifork *ifp = xfs_iext_state_to_fork(ip, state);
xfs: use a b+tree for the in-core extent list Replace the current linear list and the indirection array for the in-core extent list with a b+tree to avoid the need for larger memory allocations for the indirection array when lots of extents are present. The current extent list implementations leads to heavy pressure on the memory allocator when modifying files with a high extent count, and can lead to high latencies because of that. The replacement is a b+tree with a few quirks. The leaf nodes directly store the extent record in two u64 values. The encoding is a little bit different from the existing in-core extent records so that the start offset and length which are required for lookups can be retreived with simple mask operations. The inner nodes store a 64-bit key containing the start offset in the first half of the node, and the pointers to the next lower level in the second half. In either case we walk the node from the beginninig to the end and do a linear search, as that is more efficient for the low number of cache lines touched during a search (2 for the inner nodes, 4 for the leaf nodes) than a binary search. We store termination markers (zero length for the leaf nodes, an otherwise impossible high bit for the inner nodes) to terminate the key list / records instead of storing a count to use the available cache lines as efficiently as possible. One quirk of the algorithm is that while we normally split a node half and half like usual btree implementations we just spill over entries added at the very end of the list to a new node on its own. This means we get a 100% fill grade for the common cases of bulk insertion when reading an inode into memory, and when only sequentially appending to a file. The downside is a slightly higher chance of splits on the first random insertions. Both insert and removal manually recurse into the lower levels, but the bulk deletion of the whole tree is still implemented as a recursive function call, although one limited by the overall depth and with very little stack usage in every iteration. For the first few extents we dynamically grow the list from a single extent to the next powers of two until we have a first full leaf block and that building the actual tree. The code started out based on the generic lib/btree.c code from Joern Engel based on earlier work from Peter Zijlstra, but has since been rewritten beyond recognition. Signed-off-by: Christoph Hellwig <hch@lst.de> Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com> Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
2017-11-04 01:34:46 +08:00
struct xfs_iext_leaf *leaf = cur->leaf;
xfs_fileoff_t offset = xfs_iext_leaf_key(leaf, 0);
int i, nr_entries;
trace_xfs_iext_remove(ip, cur, state, _RET_IP_);
xfs: use a b+tree for the in-core extent list Replace the current linear list and the indirection array for the in-core extent list with a b+tree to avoid the need for larger memory allocations for the indirection array when lots of extents are present. The current extent list implementations leads to heavy pressure on the memory allocator when modifying files with a high extent count, and can lead to high latencies because of that. The replacement is a b+tree with a few quirks. The leaf nodes directly store the extent record in two u64 values. The encoding is a little bit different from the existing in-core extent records so that the start offset and length which are required for lookups can be retreived with simple mask operations. The inner nodes store a 64-bit key containing the start offset in the first half of the node, and the pointers to the next lower level in the second half. In either case we walk the node from the beginninig to the end and do a linear search, as that is more efficient for the low number of cache lines touched during a search (2 for the inner nodes, 4 for the leaf nodes) than a binary search. We store termination markers (zero length for the leaf nodes, an otherwise impossible high bit for the inner nodes) to terminate the key list / records instead of storing a count to use the available cache lines as efficiently as possible. One quirk of the algorithm is that while we normally split a node half and half like usual btree implementations we just spill over entries added at the very end of the list to a new node on its own. This means we get a 100% fill grade for the common cases of bulk insertion when reading an inode into memory, and when only sequentially appending to a file. The downside is a slightly higher chance of splits on the first random insertions. Both insert and removal manually recurse into the lower levels, but the bulk deletion of the whole tree is still implemented as a recursive function call, although one limited by the overall depth and with very little stack usage in every iteration. For the first few extents we dynamically grow the list from a single extent to the next powers of two until we have a first full leaf block and that building the actual tree. The code started out based on the generic lib/btree.c code from Joern Engel based on earlier work from Peter Zijlstra, but has since been rewritten beyond recognition. Signed-off-by: Christoph Hellwig <hch@lst.de> Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com> Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
2017-11-04 01:34:46 +08:00
ASSERT(ifp->if_height > 0);
ASSERT(ifp->if_u1.if_root != NULL);
ASSERT(xfs_iext_valid(ifp, cur));
xfs_iext_inc_seq(ifp, state);
xfs: use a b+tree for the in-core extent list Replace the current linear list and the indirection array for the in-core extent list with a b+tree to avoid the need for larger memory allocations for the indirection array when lots of extents are present. The current extent list implementations leads to heavy pressure on the memory allocator when modifying files with a high extent count, and can lead to high latencies because of that. The replacement is a b+tree with a few quirks. The leaf nodes directly store the extent record in two u64 values. The encoding is a little bit different from the existing in-core extent records so that the start offset and length which are required for lookups can be retreived with simple mask operations. The inner nodes store a 64-bit key containing the start offset in the first half of the node, and the pointers to the next lower level in the second half. In either case we walk the node from the beginninig to the end and do a linear search, as that is more efficient for the low number of cache lines touched during a search (2 for the inner nodes, 4 for the leaf nodes) than a binary search. We store termination markers (zero length for the leaf nodes, an otherwise impossible high bit for the inner nodes) to terminate the key list / records instead of storing a count to use the available cache lines as efficiently as possible. One quirk of the algorithm is that while we normally split a node half and half like usual btree implementations we just spill over entries added at the very end of the list to a new node on its own. This means we get a 100% fill grade for the common cases of bulk insertion when reading an inode into memory, and when only sequentially appending to a file. The downside is a slightly higher chance of splits on the first random insertions. Both insert and removal manually recurse into the lower levels, but the bulk deletion of the whole tree is still implemented as a recursive function call, although one limited by the overall depth and with very little stack usage in every iteration. For the first few extents we dynamically grow the list from a single extent to the next powers of two until we have a first full leaf block and that building the actual tree. The code started out based on the generic lib/btree.c code from Joern Engel based on earlier work from Peter Zijlstra, but has since been rewritten beyond recognition. Signed-off-by: Christoph Hellwig <hch@lst.de> Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com> Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
2017-11-04 01:34:46 +08:00
nr_entries = xfs_iext_leaf_nr_entries(ifp, leaf, cur->pos) - 1;
for (i = cur->pos; i < nr_entries; i++)
leaf->recs[i] = leaf->recs[i + 1];
xfs_iext_rec_clear(&leaf->recs[nr_entries]);
ifp->if_bytes -= sizeof(struct xfs_iext_rec);
if (cur->pos == 0 && nr_entries > 0) {
xfs_iext_update_node(ifp, offset, xfs_iext_leaf_key(leaf, 0), 1,
leaf);
offset = xfs_iext_leaf_key(leaf, 0);
} else if (cur->pos == nr_entries) {
if (ifp->if_height > 1 && leaf->next)
cur->leaf = leaf->next;
else
cur->leaf = NULL;
cur->pos = 0;
}
if (nr_entries >= RECS_PER_LEAF / 2)
return;
if (ifp->if_height > 1)
xfs_iext_rebalance_leaf(ifp, cur, leaf, offset, nr_entries);
else if (nr_entries == 0)
xfs_iext_free_last_leaf(ifp);
}
/*
* Lookup the extent covering bno.
*
* If there is an extent covering bno return the extent index, and store the
* expanded extent structure in *gotp, and the extent cursor in *cur.
* If there is no extent covering bno, but there is an extent after it (e.g.
* it lies in a hole) return that extent in *gotp and its cursor in *cur
* instead.
* If bno is beyond the last extent return false, and return an invalid
* cursor value.
*/
bool
xfs_iext_lookup_extent(
struct xfs_inode *ip,
struct xfs_ifork *ifp,
xfs_fileoff_t offset,
struct xfs_iext_cursor *cur,
struct xfs_bmbt_irec *gotp)
{
XFS_STATS_INC(ip->i_mount, xs_look_exlist);
cur->leaf = xfs_iext_find_level(ifp, offset, 1);
if (!cur->leaf) {
cur->pos = 0;
return false;
}
for (cur->pos = 0; cur->pos < xfs_iext_max_recs(ifp); cur->pos++) {
struct xfs_iext_rec *rec = cur_rec(cur);
if (xfs_iext_rec_is_empty(rec))
break;
if (xfs_iext_rec_cmp(rec, offset) >= 0)
goto found;
}
/* Try looking in the next node for an entry > offset */
if (ifp->if_height == 1 || !cur->leaf->next)
return false;
cur->leaf = cur->leaf->next;
cur->pos = 0;
if (!xfs_iext_valid(ifp, cur))
return false;
found:
xfs_iext_get(gotp, cur_rec(cur));
return true;
}
/*
* Returns the last extent before end, and if this extent doesn't cover
* end, update end to the end of the extent.
*/
bool
xfs_iext_lookup_extent_before(
struct xfs_inode *ip,
struct xfs_ifork *ifp,
xfs_fileoff_t *end,
struct xfs_iext_cursor *cur,
struct xfs_bmbt_irec *gotp)
{
/* could be optimized to not even look up the next on a match.. */
if (xfs_iext_lookup_extent(ip, ifp, *end - 1, cur, gotp) &&
gotp->br_startoff <= *end - 1)
return true;
if (!xfs_iext_prev_extent(ifp, cur, gotp))
return false;
*end = gotp->br_startoff + gotp->br_blockcount;
return true;
}
void
xfs_iext_update_extent(
struct xfs_inode *ip,
int state,
struct xfs_iext_cursor *cur,
struct xfs_bmbt_irec *new)
{
struct xfs_ifork *ifp = xfs_iext_state_to_fork(ip, state);
xfs_iext_inc_seq(ifp, state);
xfs: use a b+tree for the in-core extent list Replace the current linear list and the indirection array for the in-core extent list with a b+tree to avoid the need for larger memory allocations for the indirection array when lots of extents are present. The current extent list implementations leads to heavy pressure on the memory allocator when modifying files with a high extent count, and can lead to high latencies because of that. The replacement is a b+tree with a few quirks. The leaf nodes directly store the extent record in two u64 values. The encoding is a little bit different from the existing in-core extent records so that the start offset and length which are required for lookups can be retreived with simple mask operations. The inner nodes store a 64-bit key containing the start offset in the first half of the node, and the pointers to the next lower level in the second half. In either case we walk the node from the beginninig to the end and do a linear search, as that is more efficient for the low number of cache lines touched during a search (2 for the inner nodes, 4 for the leaf nodes) than a binary search. We store termination markers (zero length for the leaf nodes, an otherwise impossible high bit for the inner nodes) to terminate the key list / records instead of storing a count to use the available cache lines as efficiently as possible. One quirk of the algorithm is that while we normally split a node half and half like usual btree implementations we just spill over entries added at the very end of the list to a new node on its own. This means we get a 100% fill grade for the common cases of bulk insertion when reading an inode into memory, and when only sequentially appending to a file. The downside is a slightly higher chance of splits on the first random insertions. Both insert and removal manually recurse into the lower levels, but the bulk deletion of the whole tree is still implemented as a recursive function call, although one limited by the overall depth and with very little stack usage in every iteration. For the first few extents we dynamically grow the list from a single extent to the next powers of two until we have a first full leaf block and that building the actual tree. The code started out based on the generic lib/btree.c code from Joern Engel based on earlier work from Peter Zijlstra, but has since been rewritten beyond recognition. Signed-off-by: Christoph Hellwig <hch@lst.de> Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com> Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
2017-11-04 01:34:46 +08:00
if (cur->pos == 0) {
struct xfs_bmbt_irec old;
xfs_iext_get(&old, cur_rec(cur));
if (new->br_startoff != old.br_startoff) {
xfs_iext_update_node(ifp, old.br_startoff,
new->br_startoff, 1, cur->leaf);
}
}
trace_xfs_bmap_pre_update(ip, cur, state, _RET_IP_);
xfs_iext_set(cur_rec(cur), new);
trace_xfs_bmap_post_update(ip, cur, state, _RET_IP_);
}
/*
* Return true if the cursor points at an extent and return the extent structure
* in gotp. Else return false.
*/
bool
xfs_iext_get_extent(
struct xfs_ifork *ifp,
struct xfs_iext_cursor *cur,
struct xfs_bmbt_irec *gotp)
{
if (!xfs_iext_valid(ifp, cur))
return false;
xfs_iext_get(gotp, cur_rec(cur));
return true;
}
/*
* This is a recursive function, because of that we need to be extremely
* careful with stack usage.
*/
static void
xfs_iext_destroy_node(
struct xfs_iext_node *node,
int level)
{
int i;
if (level > 1) {
for (i = 0; i < KEYS_PER_NODE; i++) {
if (node->keys[i] == XFS_IEXT_KEY_INVALID)
break;
xfs_iext_destroy_node(node->ptrs[i], level - 1);
}
}
kmem_free(node);
}
void
xfs_iext_destroy(
struct xfs_ifork *ifp)
{
xfs_iext_destroy_node(ifp->if_u1.if_root, ifp->if_height);
ifp->if_bytes = 0;
ifp->if_height = 0;
ifp->if_u1.if_root = NULL;
}