|
|
|
|
@@ -1,26 +1,26 @@
|
|
|
|
|
package sys
|
|
|
|
|
package descriptor
|
|
|
|
|
|
|
|
|
|
import "math/bits"
|
|
|
|
|
|
|
|
|
|
// FileTable is a data structure mapping 32 bit file descriptor to file objects.
|
|
|
|
|
// Table is a data structure mapping 32 bit descriptor to items.
|
|
|
|
|
//
|
|
|
|
|
// The data structure optimizes for memory density and lookup performance,
|
|
|
|
|
// trading off compute at insertion time. This is a useful compromise for the
|
|
|
|
|
// use cases we employ it with: files are usually read or written a lot more
|
|
|
|
|
// often than they are opened, each operation requires a table lookup so we are
|
|
|
|
|
// better off spending extra compute to insert files in the table in order to
|
|
|
|
|
// use cases we employ it with: items are usually accessed a lot more often
|
|
|
|
|
// than they are inserted, each operation requires a table lookup so we are
|
|
|
|
|
// better off spending extra compute to insert items in the table in order to
|
|
|
|
|
// get cheaper lookups. Memory efficiency is also crucial to support scaling
|
|
|
|
|
// with programs that open thousands of files: having a high or non-linear
|
|
|
|
|
// with programs that maintain thousands of items: having a high or non-linear
|
|
|
|
|
// memory-to-item ratio could otherwise be used as an attack vector by malicous
|
|
|
|
|
// applications attempting to damage performance of the host.
|
|
|
|
|
type FileTable struct {
|
|
|
|
|
type Table[Key ~uint32, Item any] struct {
|
|
|
|
|
masks []uint64
|
|
|
|
|
files []*FileEntry
|
|
|
|
|
items []Item
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
// Len returns the number of files stored in the table.
|
|
|
|
|
func (t *FileTable) Len() (n int) {
|
|
|
|
|
// We could make this a O(1) operation if we cached the number of files in
|
|
|
|
|
// Len returns the number of items stored in the table.
|
|
|
|
|
func (t *Table[Key, Item]) Len() (n int) {
|
|
|
|
|
// We could make this a O(1) operation if we cached the number of items in
|
|
|
|
|
// the table. More state usually means more problems, so until we have a
|
|
|
|
|
// clear need for this, the simple implementation may be a better trade off.
|
|
|
|
|
for _, mask := range t.masks {
|
|
|
|
|
@@ -29,9 +29,9 @@ func (t *FileTable) Len() (n int) {
|
|
|
|
|
return n
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
// Grow ensures that t has enough room for n files, potentially reallocating the
|
|
|
|
|
// internal buffers if their capacity was too small to hold this many files.
|
|
|
|
|
func (t *FileTable) Grow(n int) {
|
|
|
|
|
// Grow ensures that t has enough room for n items, potentially reallocating the
|
|
|
|
|
// internal buffers if their capacity was too small to hold this many items.
|
|
|
|
|
func (t *Table[Key, Item]) Grow(n int) {
|
|
|
|
|
// Round up to a multiple of 64 since this is the smallest increment due to
|
|
|
|
|
// using 64 bits masks.
|
|
|
|
|
n = (n*64 + 63) / 64
|
|
|
|
|
@@ -40,20 +40,20 @@ func (t *FileTable) Grow(n int) {
|
|
|
|
|
masks := make([]uint64, n)
|
|
|
|
|
copy(masks, t.masks)
|
|
|
|
|
|
|
|
|
|
files := make([]*FileEntry, n*64)
|
|
|
|
|
copy(files, t.files)
|
|
|
|
|
items := make([]Item, n*64)
|
|
|
|
|
copy(items, t.items)
|
|
|
|
|
|
|
|
|
|
t.masks = masks
|
|
|
|
|
t.files = files
|
|
|
|
|
t.items = items
|
|
|
|
|
}
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
// Insert inserts the given file to the table, returning the fd that it is
|
|
|
|
|
// Insert inserts the given item to the table, returning the key that it is
|
|
|
|
|
// mapped to.
|
|
|
|
|
//
|
|
|
|
|
// The method does not perform deduplication, it is possible for the same file
|
|
|
|
|
// to be inserted multiple times, each insertion will return a different fd.
|
|
|
|
|
func (t *FileTable) Insert(file *FileEntry) (fd uint32) {
|
|
|
|
|
// The method does not perform deduplication, it is possible for the same item
|
|
|
|
|
// to be inserted multiple times, each insertion will return a different key.
|
|
|
|
|
func (t *Table[Key, Item]) Insert(item Item) (key Key) {
|
|
|
|
|
offset := 0
|
|
|
|
|
insert:
|
|
|
|
|
// Note: this loop could be made a lot more efficient using vectorized
|
|
|
|
|
@@ -63,10 +63,10 @@ insert:
|
|
|
|
|
if ^mask != 0 { // not full?
|
|
|
|
|
shift := bits.TrailingZeros64(^mask)
|
|
|
|
|
index += offset
|
|
|
|
|
fd = uint32(index)*64 + uint32(shift)
|
|
|
|
|
t.files[fd] = file
|
|
|
|
|
key = Key(index)*64 + Key(shift)
|
|
|
|
|
t.items[key] = item
|
|
|
|
|
t.masks[index] = mask | uint64(1<<shift)
|
|
|
|
|
return fd
|
|
|
|
|
return key
|
|
|
|
|
}
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
@@ -80,52 +80,53 @@ insert:
|
|
|
|
|
goto insert
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
// Lookup returns the file associated with the given fd (may be nil).
|
|
|
|
|
func (t *FileTable) Lookup(fd uint32) (file *FileEntry, found bool) {
|
|
|
|
|
if i := int(fd); i >= 0 && i < len(t.files) {
|
|
|
|
|
index := uint(fd) / 64
|
|
|
|
|
shift := uint(fd) % 64
|
|
|
|
|
// Lookup returns the item associated with the given key (may be nil).
|
|
|
|
|
func (t *Table[Key, Item]) Lookup(key Key) (item Item, found bool) {
|
|
|
|
|
if i := int(key); i >= 0 && i < len(t.items) {
|
|
|
|
|
index := uint(key) / 64
|
|
|
|
|
shift := uint(key) % 64
|
|
|
|
|
if (t.masks[index] & (1 << shift)) != 0 {
|
|
|
|
|
file, found = t.files[i], true
|
|
|
|
|
item, found = t.items[i], true
|
|
|
|
|
}
|
|
|
|
|
}
|
|
|
|
|
return
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
// InsertAt inserts the given `file` at the file descriptor `fd`.
|
|
|
|
|
func (t *FileTable) InsertAt(file *FileEntry, fd uint32) {
|
|
|
|
|
if diff := int(fd) - t.Len(); diff > 0 {
|
|
|
|
|
// InsertAt inserts the given `item` at the item descriptor `key`.
|
|
|
|
|
func (t *Table[Key, Item]) InsertAt(item Item, key Key) {
|
|
|
|
|
if diff := int(key) - t.Len(); diff > 0 {
|
|
|
|
|
t.Grow(diff)
|
|
|
|
|
}
|
|
|
|
|
index := uint(fd) / 64
|
|
|
|
|
shift := uint(fd) % 64
|
|
|
|
|
index := uint(key) / 64
|
|
|
|
|
shift := uint(key) % 64
|
|
|
|
|
t.masks[index] |= 1 << shift
|
|
|
|
|
t.files[fd] = file
|
|
|
|
|
t.items[key] = item
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
// Delete deletes the file stored at the given fd from the table.
|
|
|
|
|
func (t *FileTable) Delete(fd uint32) {
|
|
|
|
|
if index, shift := fd/64, fd%64; int(index) < len(t.masks) {
|
|
|
|
|
// Delete deletes the item stored at the given key from the table.
|
|
|
|
|
func (t *Table[Key, Item]) Delete(key Key) {
|
|
|
|
|
if index, shift := key/64, key%64; int(index) < len(t.masks) {
|
|
|
|
|
mask := t.masks[index]
|
|
|
|
|
if (mask & (1 << shift)) != 0 {
|
|
|
|
|
t.files[fd] = nil
|
|
|
|
|
var zero Item
|
|
|
|
|
t.items[key] = zero
|
|
|
|
|
t.masks[index] = mask & ^uint64(1<<shift)
|
|
|
|
|
}
|
|
|
|
|
}
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
// Range calls f for each file and its associated fd in the table. The function
|
|
|
|
|
// Range calls f for each item and its associated key in the table. The function
|
|
|
|
|
// f might return false to interupt the iteration.
|
|
|
|
|
func (t *FileTable) Range(f func(uint32, *FileEntry) bool) {
|
|
|
|
|
func (t *Table[Key, Item]) Range(f func(Key, Item) bool) {
|
|
|
|
|
for i, mask := range t.masks {
|
|
|
|
|
if mask == 0 {
|
|
|
|
|
continue
|
|
|
|
|
}
|
|
|
|
|
for j := uint32(0); j < 64; j++ {
|
|
|
|
|
for j := Key(0); j < 64; j++ {
|
|
|
|
|
if (mask & (1 << j)) == 0 {
|
|
|
|
|
continue
|
|
|
|
|
}
|
|
|
|
|
if fd := uint32(i)*64 + j; !f(fd, t.files[fd]) {
|
|
|
|
|
if key := Key(i)*64 + j; !f(key, t.items[key]) {
|
|
|
|
|
return
|
|
|
|
|
}
|
|
|
|
|
}
|
|
|
|
|
@@ -133,11 +134,12 @@ func (t *FileTable) Range(f func(uint32, *FileEntry) bool) {
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
// Reset clears the content of the table.
|
|
|
|
|
func (t *FileTable) Reset() {
|
|
|
|
|
func (t *Table[Key, Item]) Reset() {
|
|
|
|
|
for i := range t.masks {
|
|
|
|
|
t.masks[i] = 0
|
|
|
|
|
}
|
|
|
|
|
for i := range t.files {
|
|
|
|
|
t.files[i] = nil
|
|
|
|
|
var zero Item
|
|
|
|
|
for i := range t.items {
|
|
|
|
|
t.items[i] = zero
|
|
|
|
|
}
|
|
|
|
|
}
|