static void *ileaf_resize(struct btree *btree, tuxkey_t inum, vleaf *base, unsigned newsize)
{
assert(ileaf_sniff(btree, base));
struct ileaf *leaf = base;
assert(inum >= ibase(leaf));
be_u16 *dict = base + btree->sb->blocksize;
unsigned at = inum - ibase(leaf);
if (at >= btree->entries_per_leaf)
return NULL;
unsigned extend_empty = at < icount(leaf) ? 0 : at - icount(leaf) + 1;
unsigned offset = at && icount(leaf) ? from_be_u16(*(dict - (at < icount(leaf) ? at : icount(leaf)))) : 0;
unsigned size = at < icount(leaf) ? from_be_u16(*(dict - at - 1)) - offset : 0;
int more = newsize - size;
if (more > 0 && sizeof(*dict) * extend_empty + more > ileaf_free(btree, leaf))
return NULL;
for (; extend_empty--; leaf->count = to_be_u16(from_be_u16(leaf->count) + 1))
*(dict - icount(leaf) - 1) = to_be_u16(atdict(dict, icount(leaf)));
assert(icount(leaf));
unsigned itop = from_be_u16(*(dict - icount(leaf)));
void *attrs = leaf->table + offset;
printf("resize inum 0x%Lx at 0x%x from %x to %x\n", (L)inum, offset, size, newsize);
assert(itop >= offset + size);
memmove(attrs + newsize, attrs + size, itop - offset - size);
for (int i = at + 1; i <= icount(leaf); i++)
add_idict(dict - i, more);
return attrs;
}
/*
* Find free inum
* (callback for btree_traverse())
*
* return value:
* 1 - found
* 0 - not found
*/
int ileaf_find_free(struct btree *btree, tuxkey_t key_bottom,
tuxkey_t key_limit, void *leaf,
tuxkey_t key, u64 len, void *data)
{
tuxkey_t at = key - ibase(leaf);
unsigned count = icount(leaf);
key_limit = min(key_limit, key + len);
if (at < count) {
__be16 *dict = ileaf_dict(btree, leaf);
unsigned limit, offset = atdict(dict, at);
while (at < count) {
at++;
limit = __atdict(dict, at);
if (offset == limit) {
at--;
break;
}
offset = limit;
}
}
if (ibase(leaf) + at < key_limit) {
*(inum_t *)data = ibase(leaf) + at;
return 1;
}
return 0;
}
static tuxkey_t ileaf_split(struct btree *btree, tuxkey_t hint,
void *from, void *into)
{
assert(ileaf_sniff(btree, from));
struct ileaf *leaf = from, *dest = into;
__be16 *dict = ileaf_dict(btree, from);
__be16 *destdict = ileaf_dict(btree, into);
#ifdef SPLIT_AT_INUM
/*
* This is to prevent to have same ibase on both of from and into
* FIXME: we would want to split at better position.
*/
if (hint == ibase(leaf))
hint++;
trace("split at inum 0x%Lx", hint);
unsigned at = min_t(tuxkey_t, hint - ibase(leaf), icount(leaf));
#else
/* binsearch inum starting nearest middle of block */
unsigned at = 1, hi = icount(leaf);
while (at < hi) {
int mid = (at + hi) / 2;
if (*(dict - mid) < (btree->sb->blocksize / 2))
at = mid + 1;
else
hi = mid;
}
#endif
/* should trim leading empty inodes on copy */
unsigned split = atdict(dict, at), free = atdict(dict, icount(leaf));
trace("split at %x of %x", at, icount(leaf));
trace("copy out %x bytes at %x", free - split, split);
assert(free >= split);
memcpy(dest->table, leaf->table + split, free - split);
dest->count = cpu_to_be16(icount(leaf) - at);
veccopy(destdict - icount(dest), dict - icount(leaf), icount(dest));
for (int i = 1; i <= icount(dest); i++)
add_idict(destdict - i, -split);
#ifdef SPLIT_AT_INUM
/* round down to multiple of 64 above ibase */
inum_t round = hint & ~(inum_t)(btree->entries_per_leaf - 1);
dest->ibase = cpu_to_be64(round > ibase(leaf) + icount(leaf) ? round : hint);
#else
dest->ibase = cpu_to_be64(ibase(leaf) + at);
#endif
leaf->count = cpu_to_be16(at);
memset(leaf->table + split, 0, (char *)(dict - icount(leaf)) - (leaf->table + split));
ileaf_trim(btree, leaf);
return ibase(dest);
}
inum_t find_empty_inode(struct btree *btree, struct ileaf *leaf, inum_t goal)
{
assert(goal >= ibase(leaf));
goal -= ibase(leaf);
//printf("find empty inode starting at %Lx, base %Lx\n", (L)goal, (L)ibase(leaf));
be_u16 *dict = (void *)leaf + btree->sb->blocksize;
unsigned i, offset = goal && goal < icount(leaf) ? from_be_u16(*(dict - goal)) : 0;
for (i = goal; i < icount(leaf); i++) {
unsigned limit = from_be_u16(*(dict - i - 1));
if (offset == limit)
break;
offset = limit;
}
return i + ibase(leaf);
}
/*
* Chop inums
* return value:
* < 0 - error
* 1 - modified
* 0 - not modified
*/
static int ileaf_chop(struct btree *btree, tuxkey_t start, u64 len, void *leaf)
{
struct ileaf *ileaf = leaf;
__be16 *dict = ileaf_dict(btree, leaf);
tuxkey_t base = ibase(ileaf);
unsigned count = icount(ileaf);
tuxkey_t at = start - base;
void *startp, *endp, *tailp;
unsigned size;
if (at + 1 > count)
return 0;
len = min_t(u64, len, count - at);
startp = ileaf->table + atdict(dict, at);
endp = ileaf->table + atdict(dict, at + len);
if (startp == endp)
return 0;
/* Remove data */
tailp = ileaf->table + atdict(dict, count);
memmove(startp, endp, tailp - endp);
/* Adjust dict */
size = endp - startp;
while (at < count) {
at++;
add_idict(dict - at, -size);
}
ileaf_trim(btree, leaf);
return 1;
}
void *ileaf_lookup(struct btree *btree, inum_t inum, struct ileaf *leaf, unsigned *result)
{
assert(inum >= ibase(leaf));
tuxkey_t at = inum - ibase(leaf), size = 0;
void *attrs = NULL;
trace("lookup inode 0x%Lx, %Lx + %Lx", inum, ibase(leaf), at);
if (at < icount(leaf)) {
__be16 *dict = ileaf_dict(btree, leaf);
unsigned offset = atdict(dict, at);
if ((size = __atdict(dict, at + 1) - offset))
attrs = leaf->table + offset;
}
*result = size;
return attrs;
}
static int ileaf_merge(struct btree *btree, void *vinto, void *vfrom)
{
struct ileaf *into = vinto, *from = vfrom;
unsigned fromcount = icount(from);
/* If "from" is empty, does nothing */
if (!fromcount)
return 1;
assert(ibase(from) > ibase(into));
tuxkey_t fromibase = ibase(from);
unsigned count = icount(into);
int hole = fromibase - ibase(into) + count;
__be16 *dict = ileaf_dict(btree, into);
__be16 *fromdict = ileaf_dict(btree, from);
int need_size = hole * sizeof(*dict) + ileaf_need(btree, from);
if (ileaf_free(btree, into) < need_size)
return 0;
/* Fill hole of dict until from_ibase */
unsigned limit = atdict(dict, count);
__be16 __limit = cpu_to_be16(limit);
while (hole--) {
count++;
*(dict - count) = __limit;
}
/* Copy data from "from" */
unsigned fromlimit = atdict(fromdict, fromcount);
memcpy(into->table + limit, from->table, fromlimit);
/* Adjust copying fromdict */
if (limit) {
int i;
for (i = 1; i <= fromcount; i++)
add_idict(dict - i, limit);
}
veccopy(dict - count - fromcount, fromdict - fromcount, fromcount);
into->count = cpu_to_be16(count + fromcount);
return 1;
}
int ileaf_purge(struct btree *btree, inum_t inum, struct ileaf *leaf)
{
if (inum < ibase(leaf) || inum - ibase(leaf) >= btree->entries_per_leaf)
return -EINVAL;
be_u16 *dict = (void *)leaf + btree->sb->blocksize;
unsigned at = inum - ibase(leaf);
unsigned offset = atdict(dict, at);
unsigned size = from_be_u16(*(dict - at - 1)) - offset;
printf("delete inode %Lx from %p[%x/%x]\n", (L)inum, leaf, at, size);
if (!size)
return -ENOENT;
unsigned free = from_be_u16(*(dict - icount(leaf))), tail = free - offset - size;
assert(offset + size + tail <= free);
memmove(leaf->table + offset, leaf->table + offset + size, tail);
for (int i = at + 1; i <= icount(leaf); i++)
add_idict(dict - i, -size);
ileaf_trim(btree, leaf);
return 0;
}
static tuxkey_t ileaf_split_hint(struct btree *btree, struct ileaf *ileaf,
tuxkey_t key, int size)
{
/*
* FIXME: make sure there is space for size.
* FIXME: better split position?
*/
tuxkey_t base = ibase(ileaf);
unsigned count = icount(ileaf);
if (key >= base + count)
return key & ~(btree->entries_per_leaf - 1);
return base + count / 2;
}
static void ileaf_dump(struct btree *btree, void *vleaf)
{
if (!tux3_trace)
return;
struct ileaf_attr_ops *attr_ops = btree->ops->private_ops;
struct ileaf *leaf = vleaf;
inum_t inum = ibase(leaf);
__be16 *dict = ileaf_dict(btree, leaf);
unsigned offset = 0;
trace_on("inode table block 0x%Lx/%i (%x bytes free)",
ibase(leaf), icount(leaf), ileaf_free(btree, leaf));
for (int i = 0; i < icount(leaf); i++, inum++) {
int limit = __atdict(dict, i + 1), size = limit - offset;
if (!size)
continue;
if (size < 0)
trace_on(" 0x%Lx: <corrupt>\n", inum);
else if (!size)
trace_on(" 0x%Lx: <empty>\n", inum);
else if (attr_ops == &iattr_ops) {
/* FIXME: this doesn't work in kernel */
struct tux3_inode tuxnode = {};
struct inode *inode = &tuxnode.vfs_inode;
void *attrs = leaf->table + offset;
inode->i_sb = vfs_sb(btree->sb),
attr_ops->decode(btree, inode, attrs, size);
free_xcache(inode);
}
offset = limit;
}
}
static void ileaf_dump(struct btree *btree, vleaf *vleaf)
{
struct sb *sb = btree->sb;
struct ileaf *leaf = vleaf;
inum_t inum = ibase(leaf);
be_u16 *dict = vleaf + sb->blocksize;
unsigned offset = 0;
printf("inode table block 0x%Lx/%i (%x bytes free)\n", (L)ibase(leaf), icount(leaf), ileaf_free(btree, leaf));
//hexdump(dict - icount(leaf), icount(leaf) * 2);
for (int i = -1; -i <= icount(leaf); i--, inum++) {
int limit = from_be_u16(dict[i]), size = limit - offset;
if (!size)
continue;
printf(" 0x%Lx: ", (L)inum);
//printf("[%x] ", offset);
if (size < 0)
printf("<corrupt>\n");
else if (!size)
printf("<empty>\n");
else {
/* FIXME: this doesn't work in kernel */
struct inode inode = { .i_sb = vfs_sb(btree->sb) };
unsigned xsize = decode_xsize(&inode, leaf->table + offset, size);
tux_inode(&inode)->xcache = xsize ? new_xcache(xsize) : NULL;
decode_attrs(&inode, leaf->table + offset, size);
dump_attrs(&inode);
xcache_dump(&inode);
free(tux_inode(&inode)->xcache);
}
offset = limit;
}
}
void *ileaf_lookup(struct btree *btree, inum_t inum, struct ileaf *leaf, unsigned *result)
{
assert(inum >= ibase(leaf));
assert(inum < ibase(leaf) + btree->entries_per_leaf);
unsigned at = inum - ibase(leaf), size = 0;
void *attrs = NULL;
printf("lookup inode 0x%Lx, %Lx + %x\n", (L)inum, (L)ibase(leaf), at);
if (at < icount(leaf)) {
be_u16 *dict = (void *)leaf + btree->sb->blocksize;
unsigned offset = atdict(dict, at);
if ((size = from_be_u16(*(dict - at - 1)) - offset))
attrs = leaf->table + offset;
}
*result = size;
return attrs;
}
/*
* Enumerate inum
* (callback for btree_traverse())
*/
int ileaf_enumerate(struct btree *btree, tuxkey_t key_bottom,
tuxkey_t key_limit, void *leaf,
tuxkey_t key, u64 len, void *data)
{
struct ileaf *ileaf = leaf;
__be16 *dict = ileaf_dict(btree, ileaf);
struct ileaf_enumrate_cb *cb = data;
tuxkey_t at, base = ibase(ileaf);
unsigned count;
at = key - base;
count = min_t(u64, key + len - base, icount(ileaf));
if (at < count) {
unsigned offset = atdict(dict, at);
for (; at < count; at++) {
unsigned size, limit;
inum_t inum;
void *attrs;
int err;
limit = __atdict(dict, at + 1);
if (limit <= offset)
continue;
attrs = ileaf->table + offset;
size = limit - offset;
inum = base + at;
err = cb->callback(btree, inum, attrs, size, cb->data);
if (err)
return err;
offset = limit;
}
}
return 0;
}
static tuxkey_t ileaf_split(struct btree *btree, tuxkey_t inum, vleaf *from, vleaf *into)
{
assert(ileaf_sniff(btree, from));
struct ileaf *leaf = from, *dest = into;
be_u16 *dict = from + btree->sb->blocksize, *destdict = into + btree->sb->blocksize;
#ifdef SPLIT_AT_INUM
printf("split at inum 0x%Lx\n", (L)inum);
assert(inum >= ibase(leaf));
unsigned at = inum - ibase(leaf) < icount(leaf) ? inum - ibase(leaf) : icount(leaf);
#else
/* binsearch inum starting nearest middle of block */
unsigned at = 1, hi = icount(leaf);
while (at < hi) {
int mid = (at + hi) / 2;
if (*(dict - mid) < (btree->sb->blocksize / 2))
at = mid + 1;
else
hi = mid;
}
#endif
/* should trim leading empty inodes on copy */
unsigned split = atdict(dict, at), free = from_be_u16(*(dict - icount(leaf)));
printf("split at %x of %x\n", at, icount(leaf));
printf("copy out %x bytes at %x\n", free - split, split);
assert(free >= split);
memcpy(dest->table, leaf->table + split, free - split);
dest->count = to_be_u16(icount(leaf) - at);
veccopy(destdict - icount(dest), dict - icount(leaf), icount(dest));
for (int i = 1; i <= icount(dest); i++)
add_idict(destdict - i, -split);
#ifdef SPLIT_AT_INUM
/* round down to multiple of 64 above ibase */
inum_t round = inum & ~(inum_t)(btree->entries_per_leaf - 1);
dest->ibase = to_be_u64(round > ibase(leaf) + icount(leaf) ? round : inum);
#else
dest->ibase = to_be_u64(ibase(leaf) + at);
#endif
leaf->count = to_be_u16(at);
memset(leaf->table + split, 0, (char *)(dict - icount(leaf)) - (leaf->table + split));
ileaf_trim(btree, leaf);
return ibase(dest);
}
static void *ileaf_resize(struct btree *btree, tuxkey_t inum, void *vleaf,
int newsize)
{
struct ileaf *ileaf = vleaf;
__be16 *dict = ileaf_dict(btree, ileaf);
unsigned count = icount(ileaf);
tuxkey_t at = inum - ibase(ileaf);
int extend_dict, offset, size;
assert(inum >= ibase(ileaf));
/* Get existent attributes, and calculate expand/shrink size */
if (at + 1 > count) {
/* Check size roughly to avoid overflow int */
if ((at + 1) * sizeof(*dict) >= btree->sb->blocksize)
goto overflow;
/* Need to extend dict */
extend_dict = (at + 1 - count) * sizeof(*dict);
offset = atdict(dict, count);
size = 0;
} else {
/* "at" is in dict, so get attr size */
extend_dict = 0;
offset = atdict(dict, at);
size = __atdict(dict, at + 1) - offset;
}
if (ileaf_free(btree, ileaf) < newsize - size + extend_dict) {
overflow:
return NULL;
}
/* Extend dict */
if (extend_dict) {
__be16 limit = cpu_to_be16(atdict(dict, count));
while (count < at + 1) {
count++;
*(dict - count) = limit;
}
ileaf->count = cpu_to_be16(count);
}
void *attrs = ileaf->table + offset;
if (newsize != size) {
/* Expand/Shrink attr space */
unsigned limit = __atdict(dict, count);
assert(limit >= offset + size);
memmove(attrs + newsize, attrs + size, limit - offset - size);
/* Adjust dict */
int diff = newsize - size;
at++;
while (at <= count) {
add_idict(dict - at, diff);
at++;
}
}
return attrs;
}