static loff_t unatom_dict_write(struct inode *atable, atom_t atom, loff_t where)
{
unsigned delta = tux3_get_current_delta();
struct buffer_head *buffer, *clone;
loff_t old;
unsigned offset;
buffer = blockread_unatom(atable, atom, &offset);
if (!buffer)
return -EIO;
/*
* The atable is protected by i_mutex for now.
* blockdirty() should never return -EAGAIN.
* FIXME: need finer granularity locking
*/
clone = blockdirty(buffer, delta);
if (IS_ERR(clone)) {
assert(PTR_ERR(clone) != -EAGAIN);
blockput(buffer);
return PTR_ERR(clone);
}
__be64 *unatom_dict = bufdata(clone);
old = be64_to_cpu(unatom_dict[offset]);
unatom_dict[offset] = cpu_to_be64(where);
mark_buffer_dirty_non(clone);
blockput(clone);
return old;
}
tux_dirent *tux_find_entry(struct inode *dir, const char *name, unsigned len,
struct buffer_head **result, loff_t size)
{
struct sb *sb = tux_sb(dir->i_sb);
unsigned reclen = TUX_REC_LEN(len);
block_t block, blocks = size >> sb->blockbits;
int err = -ENOENT;
for (block = 0; block < blocks; block++) {
struct buffer_head *buffer = blockread(mapping(dir), block);
if (!buffer) {
err = -EIO; // need ERR_PTR for blockread!!!
goto error;
}
tux_dirent *entry = bufdata(buffer);
tux_dirent *limit = (void *)entry + sb->blocksize - reclen;
while (entry <= limit) {
if (entry->rec_len == 0) {
blockput(buffer);
tux_zero_len_error(dir, block);
err = -EIO;
goto error;
}
if (tux_match(entry, name, len)) {
*result = buffer;
return entry;
}
entry = next_entry(entry);
}
blockput(buffer);
}
error:
*result = NULL; /* for debug */
return ERR_PTR(err);
}
/* Modify buffer of refcount, then release buffer */
static int update_refcount(struct sb *sb, struct buffer_head *buffer,
unsigned offset, u16 val)
{
unsigned delta = tux3_get_current_delta();
struct buffer_head *clone;
__be16 *refcount;
/*
* The atable is protected by i_mutex for now.
* blockdirty() should never return -EAGAIN.
* FIXME: need finer granularity locking
*/
clone = blockdirty(buffer, delta);
if (IS_ERR(clone)) {
assert(PTR_ERR(clone) != -EAGAIN);
blockput(buffer);
return PTR_ERR(clone);
}
refcount = bufdata(clone);
refcount[offset] = cpu_to_be16(val);
mark_buffer_dirty_non(clone);
blockput(clone);
return 0;
}
int tux_readdir(struct file *file, void *state, filldir_t filldir)
{
loff_t pos = file->f_pos;
#ifdef __KERNEL__
struct inode *dir = file->f_dentry->d_inode;
#else
struct inode *dir = file->f_inode;
#endif
int revalidate = file->f_version != dir->i_version;
struct sb *sb = tux_sb(dir->i_sb);
unsigned blockbits = sb->blockbits;
block_t block, blocks = dir->i_size >> blockbits;
unsigned offset = pos & sb->blockmask;
assert(!(dir->i_size & sb->blockmask));
for (block = pos >> blockbits ; block < blocks; block++) {
struct buffer_head *buffer = blockread(mapping(dir), block);
if (!buffer)
return -EIO;
void *base = bufdata(buffer);
if (revalidate) {
if (offset) {
tux_dirent *entry = base + offset;
tux_dirent *p = base + (offset & sb->blockmask);
while (p < entry && p->rec_len)
p = next_entry(p);
offset = (void *)p - base;
file->f_pos = (block << blockbits) + offset;
}
file->f_version = dir->i_version;
revalidate = 0;
}
tux_dirent *limit = base + sb->blocksize - TUX_REC_LEN(1);
for (tux_dirent *entry = base + offset; entry <= limit; entry = next_entry(entry)) {
if (entry->rec_len == 0) {
blockput(buffer);
tux_zero_len_error(dir, block);
return -EIO;
}
if (!is_deleted(entry)) {
unsigned type = (entry->type < TUX_TYPES) ? filetype[entry->type] : DT_UNKNOWN;
int lame = filldir(
state, entry->name, entry->name_len,
(block << blockbits) | ((void *)entry - base),
be64_to_cpu(entry->inum), type);
if (lame) {
blockput(buffer);
return 0;
}
}
file->f_pos += tux_rec_len_from_disk(entry->rec_len);
}
blockput(buffer);
offset = 0;
}
return 0;
}
static struct dentry *tux3_lookup(struct inode *dir, struct dentry *dentry,
unsigned int flags)
{
struct sb *sb = tux_sb(dir->i_sb);
struct buffer_head *buffer;
struct inode *inode;
tux_dirent *entry;
inum_t inum;
entry = tux_find_dirent(dir, &dentry->d_name, &buffer);
if (IS_ERR(entry)) {
if (PTR_ERR(entry) != -ENOENT)
return ERR_CAST(entry);
inode = NULL;
goto out;
}
inum = be64_to_cpu(entry->inum);
blockput(buffer);
inode = tux3_iget(sb, inum);
if (IS_ERR(inode) && PTR_ERR(inode) == -ENOENT)
tux3_warn(sb, "%s: inum %Lu not found", __func__, inum);
out:
return d_splice_alias(inode, dentry);
}
static int bitmap_test(struct sb *sb, block_t start, block_t count, int set)
{
struct inode *bitmap = sb->bitmap;
unsigned mapshift = sb->blockbits + 3;
unsigned mapsize = 1 << mapshift;
unsigned mapmask = mapsize - 1;
unsigned mapoffset = start & mapmask;
block_t mapblock, mapblocks = (start + count + mapmask) >> mapshift;
int (*test)(u8 *, unsigned, unsigned) = set ? all_set : all_clear;
for (mapblock = start >> mapshift; mapblock < mapblocks; mapblock++) {
struct buffer_head *buffer;
unsigned len;
int ret;
buffer = blockget(mapping(bitmap), mapblock);
assert(buffer);
len = min_t(block_t, mapsize - mapoffset, count);
ret = test(bufdata(buffer), mapoffset, len);
blockput(buffer);
if (!ret)
return 0;
mapoffset = 0;
count -= len;
}
return 1;
}
static inline void cursor_pop_blockput(struct cursor *cursor)
{
if(DEBUG_MODE_K==1)
{
printf("\t\t\t\t%25s[K] %25s %4d #in\n",__FILE__,__func__,__LINE__);
}
blockput(cursor_pop(cursor));
}
static void tux_update_entry(struct buffer_head *buffer, tux_dirent *entry,
inum_t inum, umode_t mode)
{
entry->inum = cpu_to_be64(inum);
entry->type = tux_type_by_mode[(mode & S_IFMT) >> STAT_SHIFT];
mark_buffer_dirty_non(buffer);
blockput(buffer);
}
/* Unpin logblocks. */
static void replay_unpin_logblocks(struct sb *sb, unsigned i, unsigned logcount)
{
if(DEBUG_MODE_K==1)
{
printk(KERN_INFO"%25s %25s %4d #in\n",__FILE__,__func__,__LINE__);
}
struct buffer_head *logbuf;
while (i < logcount) {
logbuf = blockget(mapping(sb->logmap), i);
assert(logbuf != NULL);
blockput(logbuf);
blockput(logbuf); /* Unpin */
i++;
}
}
/* This is called for the freeing block on volmap */
static void __blockput_free(struct sb *sb, struct buffer_head *buffer,
unsigned delta)
{
/* FIXME: buffer was freed, so we would like to free cache */
tux3_clear_buffer_dirty(buffer, delta);
tux3_try_cancel_dirty_page(buffer->b_page);
blockput(buffer);
}
/* Convert atom to name */
static int unatom(struct inode *atable, atom_t atom, char *name, unsigned size)
{
if(DEBUG_MODE_K==1)
{
printf("\t\t\t\t%25s[K] %25s %4d #in\n",__FILE__,__func__,__LINE__);
}
struct sb *sb = tux_sb(atable->i_sb);
struct buffer_head *buffer;
int err;
loff_t where = unatom_dict_read(atable, atom);
if (where < 0) {
err = where;
goto error;
}
buffer = blockread(mapping(atable), where >> sb->blockbits);
if (!buffer) {
err = -EIO;
goto error;
}
tux_dirent *entry = bufdata(buffer) + (where & sb->blockmask);
if (entry_atom(entry) != atom) {
tux3_fs_error(sb, "atom %x reverse entry broken", atom);
err = -EIO;
goto error_blockput;
}
unsigned len = entry->name_len;
if (size) {
if (len > size) {
err = -ERANGE;
goto error_blockput;
}
memcpy(name, entry->name, len);
}
blockput(buffer);
return len;
error_blockput:
blockput(buffer);
error:
return err;
}
int tux_delete_entry(struct inode *dir, struct buffer_head *buffer,
tux_dirent *entry)
{
unsigned delta = tux3_get_current_delta();
tux_dirent *prev = NULL, *this = bufdata(buffer);
struct buffer_head *clone;
void *olddata;
while ((char *)this < (char *)entry) {
if (this->rec_len == 0) {
blockput(buffer);
tux_zero_len_error(dir, bufindex(buffer));
return -EIO;
}
prev = this;
this = next_entry(this);
}
/*
* The directory is protected by i_mutex.
* blockdirty() should never return -EAGAIN.
*/
olddata = bufdata(buffer);
clone = blockdirty(buffer, delta);
if (IS_ERR(clone)) {
assert(PTR_ERR(clone) != -EAGAIN);
blockput(buffer);
return PTR_ERR(clone);
}
entry = ptr_redirect(entry, olddata, bufdata(clone));
prev = ptr_redirect(prev, olddata, bufdata(clone));
if (prev)
prev->rec_len = tux_rec_len_to_disk((void *)next_entry(entry) - (void *)prev);
memset(entry->name, 0, entry->name_len);
entry->name_len = entry->type = 0;
entry->inum = 0;
mark_buffer_dirty_non(clone);
blockput(clone);
return 0;
}
/* Modify atom refcount */
static int atomref(struct inode *atable, atom_t atom, int use)
{
if(DEBUG_MODE_K==1)
{
printf("\t\t\t\t%25s[K] %25s %4d #in\n",__FILE__,__func__,__LINE__);
}
struct sb *sb = tux_sb(atable->i_sb);
unsigned shift = sb->blockbits - ATOMREF_BLKBITS;
unsigned block = sb->atomref_base + ATOMREF_SIZE * (atom >> shift);
unsigned offset = atom & ~(-1 << shift), kill = 0;
struct buffer_head *buffer;
__be16 *refcount;
int err;
buffer = blockread(mapping(atable), block);
if (!buffer)
return -EIO;
refcount = bufdata(buffer);
int low = be16_to_cpu(refcount[offset]) + use;
trace("inc atom %x by %d, offset %x[%x], low = %d",
atom, use, block, offset, low);
/* This releases buffer */
err = update_refcount(sb, buffer, offset, low);
if (err)
return err;
if (!low || (low & (-1 << 16))) {
buffer = blockread(mapping(atable), block + 1);
if (!buffer)
return -EIO;
refcount = bufdata(buffer);
int high = be16_to_cpu(refcount[offset]);
if (!low)
blockput(buffer);
else {
trace("carry %d, offset %x[%x], high = %d",
(low >> 16), block, offset, high);
high += (low >> 16);
assert(high >= 0); /* paranoia check */
/* This releases buffer */
err = update_refcount(sb, buffer, offset, high);
if (err) {
/* FIXME: better set a flag that atomref broke
* or something! */
return err;
}
}
kill = !(low | high);
}
void
ventiproc(void *dummy)
{
int i;
Block *db;
u32int bno;
u64int bsize;
USED(dummy);
proccreate(vtsendproc, z, STACK);
proccreate(vtrecvproc, z, STACK);
writechan = chancreate(sizeof(WriteReq), 0);
for(i=0; i<nwritethread; i++)
threadcreate(writethread, nil, STACK);
vtcachesetwrite(zcache, myvtwrite);
bsize = fsys->blocksize;
vtfilelock(vfile, -1);
while((db = qread(qventi, &bno)) != nil){
if(nop){
blockput(db);
continue;
}
if(vtfilewrite(vfile, db->data, bsize, bno*bsize) != bsize)
sysfatal("ventiproc vtfilewrite: %r");
if(vtfileflushbefore(vfile, (bno+1)*bsize) < 0)
sysfatal("ventiproc vtfileflushbefore: %r");
blockput(db);
}
vtfileunlock(vfile);
vtcachesetwrite(zcache, nil);
for(i=0; i<nwritethread; i++)
send(writechan, nil);
chanfree(writechan);
if(statustime)
print("# %T venti proc exiting - nsend %d nrecv %d\n", nsend, nrecv);
runlock(&endlk);
}
int tux_dir_is_empty(struct inode *dir)
{
struct sb *sb = tux_sb(dir->i_sb);
block_t block, blocks = dir->i_size >> sb->blockbits;
__be64 self = cpu_to_be64(tux_inode(dir)->inum);
struct buffer_head *buffer;
for (block = 0; block < blocks; block++) {
buffer = blockread(mapping(dir), block);
if (!buffer)
return -EIO;
tux_dirent *entry = bufdata(buffer);
tux_dirent *limit = bufdata(buffer) + sb->blocksize - TUX_REC_LEN(1);
for (; entry <= limit; entry = next_entry(entry)) {
if (!entry->rec_len) {
blockput(buffer);
tux_zero_len_error(dir, block);
return -EIO;
}
if (is_deleted(entry))
continue;
if (entry->name[0] != '.')
goto not_empty;
if (entry->name_len > 2)
goto not_empty;
if (entry->name_len < 2) {
if (entry->inum != self)
goto not_empty;
} else if (entry->name[1] != '.')
goto not_empty;
}
blockput(buffer);
}
return 0;
not_empty:
blockput(buffer);
return -ENOTEMPTY;
}
static void level_replace_blockput(struct cursor *cursor, int level,
struct buffer_head *buffer,
struct index_entry *next)
{
#ifdef CURSOR_DEBUG
assert(buffer);
assert(level <= cursor->level);
assert(cursor->path[level].buffer != FREE_BUFFER);
assert(cursor->path[level].next != FREE_NEXT);
#endif
blockput(cursor->path[level].buffer);
cursor->path[level].buffer = buffer;
cursor->path[level].next = next;
}
static loff_t unatom_dict_read(struct inode *atable, atom_t atom)
{
struct buffer_head *buffer;
unsigned offset;
buffer = blockread_unatom(atable, atom, &offset);
if (!buffer)
return -EIO;
__be64 *unatom_dict = bufdata(buffer);
loff_t where = be64_to_cpu(unatom_dict[offset]);
blockput(buffer);
return where;
}
static int
ext2sync(Fsys *fsys)
{
int i;
Group g;
Block *b;
Super super;
Ext2 *fs;
Disk *disk;
fs = fsys->priv;
disk = fs->disk;
if((b = diskread(disk, SBSIZE, SBOFF)) == nil)
return -1;
parsesuper(&super, b->data);
blockput(b);
if(checksuper(&super) < 0)
return -1;
fs->blocksize = MINBLOCKSIZE<<super.logblocksize;
fs->nblock = super.nblock;
fs->ngroup = (super.nblock+super.blockspergroup-1)
/ super.blockspergroup;
fs->inospergroup = super.inospergroup;
fs->blockspergroup = super.blockspergroup;
if(super.revlevel >= 1)
fs->inosize = super.inosize;
else
fs->inosize = 128;
fs->inosperblock = fs->blocksize / fs->inosize;
if(fs->blocksize == SBOFF)
fs->groupaddr = 2;
else
fs->groupaddr = 1;
fs->descperblock = fs->blocksize / GroupSize;
fs->firstblock = super.firstdatablock;
fsys->blocksize = fs->blocksize;
fsys->nblock = fs->nblock;
if(debug) fprint(2, "ext2 %d %d-byte blocks, first data block %d, %d groups of %d\n",
fs->nblock, fs->blocksize, fs->firstblock, fs->ngroup, fs->blockspergroup);
if(0){
for(i=0; i<fs->ngroup; i++)
if(ext2group(fs, i, &g) >= 0)
fprint(2, "grp %d: bitblock=%d\n", i, g.bitblock);
}
return 0;
}
static int
ext2group(Ext2 *fs, u32int i, Group *g)
{
Block *b;
u64int addr;
if(i >= fs->ngroup)
return -1;
addr = fs->groupaddr + i/fs->descperblock;
b = diskread(fs->disk, fs->blocksize, addr*fs->blocksize);
if(b == nil)
return -1;
parsegroup(g, b->data+i%fs->descperblock*GroupSize);
blockput(b);
return 0;
}
static void level_replace_blockput(struct cursor *cursor, int level,
struct buffer_head *buffer,
struct index_entry *next)
{
if(DEBUG_MODE_K==1)
{
printf("\t\t\t\t%25s[K] %25s %4d #in\n",__FILE__,__func__,__LINE__);
}
#ifdef CURSOR_DEBUG
assert(buffer);
assert(level <= cursor->level);
assert(cursor->path[level].buffer != FREE_BUFFER);
assert(cursor->path[level].next != FREE_NEXT);
#endif
blockput(cursor->path[level].buffer);
cursor->path[level].buffer = buffer;
cursor->path[level].next = next;
}
static loff_t unatom_dict_read(struct inode *atable, atom_t atom)
{
if(DEBUG_MODE_K==1)
{
printf("\t\t\t\t%25s[K] %25s %4d #in\n",__FILE__,__func__,__LINE__);
}
struct buffer_head *buffer;
unsigned offset;
buffer = blockread_unatom(atable, atom, &offset);
if (!buffer)
return -EIO;
__be64 *unatom_dict = bufdata(buffer);
loff_t where = be64_to_cpu(unatom_dict[offset]);
blockput(buffer);
return where;
}
/* Find atom of name */
static int find_atom(struct inode *atable, const char *name, unsigned len,
atom_t *atom)
{
struct sb *sb = tux_sb(atable->i_sb);
struct buffer_head *buffer;
struct tux3_dirent *entry;
entry = tux_find_entry(atable, name, len, &buffer, sb->atomdictsize);
if (IS_ERR(entry)) {
int err = PTR_ERR(entry);
if (err == -ENOENT)
return -ENODATA;
return err;
}
*atom = entry_atom(entry);
blockput(buffer);
return 0;
}
static void add_maps(struct inode *inode, block_t index,
struct block_segment *seg, int nr_segs)
{
unsigned delta = tux3_get_current_delta();
for (int i = 0; i < nr_segs; i++) {
struct block_segment *s = &seg[i];
for (unsigned j = 0; j < s->count; j++) {
struct buffer_head *buf;
buf = blockget(inode->map, index + j);
buf = blockdirty(buf, delta);
memset(buf->data, 0, inode->i_sb->blocksize);
*(block_t *)buf->data = s->block + j;
mark_buffer_dirty_non(buf);
blockput(buf);
}
index += s->count;
}
}
static void check_maps(struct inode *inode, block_t index,
struct block_segment *seg, int nr_segs)
{
for (int i = 0; i < nr_segs; i++) {
struct block_segment *s = &seg[i];
for (unsigned j = 0; j < s->count; j++) {
struct buffer_head *buf;
buf = peekblk(inode->map, index + j);
if (s->state == BLOCK_SEG_HOLE)
test_assert(buf == NULL);
else {
block_t blk = *(block_t *)buf->data;
test_assert(blk == s->block + j);
blockput(buf);
}
}
index += s->count;
}
}
void
cmpproc(void *dummy)
{
uchar *data;
Block *db;
u32int bno, bsize;
uchar score[VtScoreSize];
uchar score1[VtScoreSize];
USED(dummy);
if(incremental)
vtfilelock(vscores, VtOREAD);
bsize = fsys->blocksize;
while((db = qread(qcmp, &bno)) != nil){
data = db->data;
sha1(data, vtzerotruncate(VtDataType, data, bsize), score, nil);
if(incremental){
if(vtfileblockscore(vscores, bno, score1) < 0)
sysfatal("cmpproc vtfileblockscore %d: %r", bno);
}else{
if(Bseek(&bscores, (vlong)bno*VtScoreSize, 0) < 0)
sysfatal("cmpproc Bseek: %r");
if(Bread(&bscores, score1, VtScoreSize) != VtScoreSize)
sysfatal("cmpproc Bread: %r");
}
if(memcmp(score, score1, VtScoreSize) != 0){
nchange++;
if(verbose)
print("# block %ud: old %V new %V\n", bno, score1, score);
qwrite(qventi, db, bno);
}else
blockput(db);
}
qclose(qventi);
if(incremental)
vtfileunlock(vscores);
if(statustime)
print("# %T cmp proc exiting\n");
runlock(&endlk);
}
/* Find atom of name */
static int find_atom(struct inode *atable, const char *name, unsigned len,
atom_t *atom)
{
if(DEBUG_MODE_K==1)
{
printf("\t\t\t\t%25s[K] %25s %4d #in\n",__FILE__,__func__,__LINE__);
}
struct sb *sb = tux_sb(atable->i_sb);
struct buffer_head *buffer;
tux_dirent *entry;
entry = tux_find_entry(atable, name, len, &buffer, sb->atomdictsize);
if (IS_ERR(entry)) {
int err = PTR_ERR(entry);
if (err == -ENOENT)
return -ENODATA;
return err;
}
*atom = entry_atom(entry);
blockput(buffer);
return 0;
}
int alloc_empty_btree(struct btree *btree)
{
if(DEBUG_MODE_K==1)
{
printf("\t\t\t\t%25s[K] %25s %4d #in\n",__FILE__,__func__,__LINE__);
}
struct sb *sb = btree->sb;
struct buffer_head *rootbuf = new_node(btree);
if (IS_ERR(rootbuf))
goto error;
struct buffer_head *leafbuf = new_leaf(btree);
if (IS_ERR(leafbuf))
goto error_leafbuf;
assert(!has_root(btree));
struct bnode *rootnode = bufdata(rootbuf);
block_t rootblock = bufindex(rootbuf);
block_t leafblock = bufindex(leafbuf);
trace("root at %Lx", rootblock);
trace("leaf at %Lx", leafblock);
bnode_init_root(rootnode, 1, leafblock, 0, 0);
log_bnode_root(sb, rootblock, 1, leafblock, 0, 0);
log_balloc(sb, leafblock, 1);
mark_buffer_unify_non(rootbuf);
blockput(rootbuf);
mark_buffer_dirty_non(leafbuf);
blockput(leafbuf);
btree->root = (struct root){ .block = rootblock, .depth = 1 };
tux3_mark_btree_dirty(btree);
return 0;
error_leafbuf:
(btree->ops->bfree)(sb, bufindex(rootbuf), 1);
blockput(rootbuf);
rootbuf = leafbuf;
error:
return PTR_ERR(rootbuf);
}
/* FIXME: right? and this should be done by btree_chop()? */
int free_empty_btree(struct btree *btree)
{
if(DEBUG_MODE_K==1)
{
printf("\t\t\t\t%25s[K] %25s %4d #in\n",__FILE__,__func__,__LINE__);
}
struct btree_ops *ops = btree->ops;
if (!has_root(btree))
return 0;
assert(btree->root.depth == 1);
struct sb *sb = btree->sb;
struct buffer_head *rootbuf = vol_bread(sb, btree->root.block);
if (!rootbuf)
return -EIO;
assert(bnode_sniff(bufdata(rootbuf)));
/* Make btree has no root */
btree->root = no_root;
tux3_mark_btree_dirty(btree);
struct bnode *rootnode = bufdata(rootbuf);
assert(bcount(rootnode) == 1);
block_t leaf = be64_to_cpu(rootnode->entries[0].block);
struct buffer_head *leafbuf = vol_find_get_block(sb, leaf);
if (leafbuf && !leaf_need_redirect(sb, leafbuf)) {
/*
* This is redirected leaf. So, in here, we can just
* cancel leaf_redirect by bfree(), instead of
* defered_bfree().
*/
bfree(sb, leaf, 1);
log_leaf_free(sb, leaf);
assert(ops->leaf_can_free(btree, bufdata(leafbuf)));
blockput_free(sb, leafbuf);
} else {
defer_bfree(&sb->defree, leaf, 1);
log_bfree(sb, leaf, 1);
if (leafbuf) {
assert(ops->leaf_can_free(btree, bufdata(leafbuf)));
blockput(leafbuf);
}
}
if (!bnode_need_redirect(sb, rootbuf)) {
/*
* This is redirected bnode. So, in here, we can just
* cancel bnode_redirect by bfree(), instead of
* defered_bfree().
*/
bfree(sb, bufindex(rootbuf), 1);
log_bnode_free(sb, bufindex(rootbuf));
blockput_free_unify(sb, rootbuf);
} else {
defer_bfree(&sb->deunify, bufindex(rootbuf), 1);
log_bfree_on_unify(sb, bufindex(rootbuf), 1);
blockput(rootbuf);
}
return 0;
}
int replay_bnode_redirect(struct replay *rp, block_t oldblock, block_t newblock)
{
if(DEBUG_MODE_K==1)
{
printf("\t\t\t\t%25s[K] %25s %4d #in\n",__FILE__,__func__,__LINE__);
}
struct sb *sb = rp->sb;
struct buffer_head *newbuf, *oldbuf;
int err = 0;
newbuf = vol_getblk(sb, newblock);
if (!newbuf) {
err = -ENOMEM; /* FIXME: error code */
goto error;
}
oldbuf = vol_bread(sb, oldblock);
if (!oldbuf) {
err = -EIO; /* FIXME: error code */
goto error_put_newbuf;
}
assert(bnode_sniff(bufdata(oldbuf)));
memcpy(bufdata(newbuf), bufdata(oldbuf), bufsize(newbuf));
mark_buffer_unify_atomic(newbuf);
blockput(oldbuf);
error_put_newbuf:
blockput(newbuf);
error:
return err;
}
int replay_bnode_root(struct replay *rp, block_t root, unsigned count,
block_t left, block_t right, tuxkey_t rkey)
{
if(DEBUG_MODE_K==1)
{
printf("\t\t\t\t%25s[K] %25s %4d #in\n",__FILE__,__func__,__LINE__);
}
struct sb *sb = rp->sb;
struct buffer_head *rootbuf;
rootbuf = vol_getblk(sb, root);
if (!rootbuf)
return -ENOMEM;
bnode_buffer_init(rootbuf);
bnode_init_root(bufdata(rootbuf), count, left, right, rkey);
mark_buffer_unify_atomic(rootbuf);
blockput(rootbuf);
return 0;
}
/*
* Before this replay, replay should already dirty the buffer of src.
* (e.g. by redirect)
*/
int replay_bnode_split(struct replay *rp, block_t src, unsigned pos,
block_t dst)
{
if(DEBUG_MODE_K==1)
{
printf("\t\t\t\t%25s[K] %25s %4d #in\n",__FILE__,__func__,__LINE__);
}
struct sb *sb = rp->sb;
struct buffer_head *srcbuf, *dstbuf;
int err = 0;
srcbuf = vol_getblk(sb, src);
if (!srcbuf) {
err = -ENOMEM; /* FIXME: error code */
goto error;
}
dstbuf = vol_getblk(sb, dst);
if (!dstbuf) {
err = -ENOMEM; /* FIXME: error code */
goto error_put_srcbuf;
}
bnode_buffer_init(dstbuf);
bnode_split(bufdata(srcbuf), pos, bufdata(dstbuf));
mark_buffer_unify_non(srcbuf);
mark_buffer_unify_atomic(dstbuf);
blockput(dstbuf);
error_put_srcbuf:
blockput(srcbuf);
error:
return err;
}
/*
* Before this replay, replay should already dirty the buffer of bnodeblock.
* (e.g. by redirect)
*/
static int replay_bnode_change(struct sb *sb, block_t bnodeblock,
u64 val1, u64 val2,
void (*change)(struct bnode *, u64, u64))
{
if(DEBUG_MODE_K==1)
{
printf("\t\t\t\t%25s[K] %25s %4d #in\n",__FILE__,__func__,__LINE__);
}
struct buffer_head *bnodebuf;
bnodebuf = vol_getblk(sb, bnodeblock);
if (!bnodebuf)
return -ENOMEM; /* FIXME: error code */
struct bnode *bnode = bufdata(bnodebuf);
change(bnode, val1, val2);
mark_buffer_unify_non(bnodebuf);
blockput(bnodebuf);
return 0;
}
static void add_func(struct bnode *bnode, u64 child, u64 key)
{
if(DEBUG_MODE_K==1)
{
printf("\t\t\t\t%25s[K] %25s %4d #in\n",__FILE__,__func__,__LINE__);
}
struct index_entry *entry = bnode_lookup(bnode, key) + 1;
bnode_add_index(bnode, entry, child, key);
}
int replay_bnode_add(struct replay *rp, block_t parent, block_t child,
tuxkey_t key)
{
if(DEBUG_MODE_K==1)
{
printf("\t\t\t\t%25s[K] %25s %4d #in\n",__FILE__,__func__,__LINE__);
}
return replay_bnode_change(rp->sb, parent, child, key, add_func);
}
static void update_func(struct bnode *bnode, u64 child, u64 key)
{
if(DEBUG_MODE_K==1)
{
printf("\t\t\t\t%25s[K] %25s %4d #in\n",__FILE__,__func__,__LINE__);
}
struct index_entry *entry = bnode_lookup(bnode, key);
assert(be64_to_cpu(entry->key) == key);
entry->block = cpu_to_be64(child);
}
int replay_bnode_update(struct replay *rp, block_t parent, block_t child,
tuxkey_t key)
{
if(DEBUG_MODE_K==1)
{
printf("\t\t\t\t%25s[K] %25s %4d #in\n",__FILE__,__func__,__LINE__);
}
return replay_bnode_change(rp->sb, parent, child, key, update_func);
}
int replay_bnode_merge(struct replay *rp, block_t src, block_t dst)
{
if(DEBUG_MODE_K==1)
{
printf("\t\t\t\t%25s[K] %25s %4d #in\n",__FILE__,__func__,__LINE__);
}
struct sb *sb = rp->sb;
struct buffer_head *srcbuf, *dstbuf;
int err = 0, ret;
srcbuf = vol_getblk(sb, src);
if (!srcbuf) {
err = -ENOMEM; /* FIXME: error code */
goto error;
}
dstbuf = vol_getblk(sb, dst);
if (!dstbuf) {
err = -ENOMEM; /* FIXME: error code */
goto error_put_srcbuf;
}
ret = bnode_merge_nodes(sb, bufdata(dstbuf), bufdata(srcbuf));
assert(ret == 1);
mark_buffer_unify_non(dstbuf);
mark_buffer_unify_non(srcbuf);
blockput(dstbuf);
error_put_srcbuf:
blockput(srcbuf);
error:
return err;
}
static void del_func(struct bnode *bnode, u64 key, u64 count)
{
if(DEBUG_MODE_K==1)
{
printf("\t\t\t\t%25s[K] %25s %4d #in\n",__FILE__,__func__,__LINE__);
}
struct index_entry *entry = bnode_lookup(bnode, key);
assert(be64_to_cpu(entry->key) == key);
bnode_remove_index(bnode, entry, count);
}
int replay_bnode_del(struct replay *rp, block_t bnode, tuxkey_t key,
unsigned count)
{
if(DEBUG_MODE_K==1)
{
printf("\t\t\t\t%25s[K] %25s %4d #in\n",__FILE__,__func__,__LINE__);
}
return replay_bnode_change(rp->sb, bnode, key, count, del_func);
}
static void adjust_func(struct bnode *bnode, u64 from, u64 to)
{
if(DEBUG_MODE_K==1)
{
printf("\t\t\t\t%25s[K] %25s %4d #in\n",__FILE__,__func__,__LINE__);
}
struct index_entry *entry = bnode_lookup(bnode, from);
assert(be64_to_cpu(entry->key) == from);
entry->key = cpu_to_be64(to);
}
int replay_bnode_adjust(struct replay *rp, block_t bnode, tuxkey_t from,
tuxkey_t to)
{
if(DEBUG_MODE_K==1)
{
printf("\t\t\t\t%25s[K] %25s %4d #in\n",__FILE__,__func__,__LINE__);
}
return replay_bnode_change(rp->sb, bnode, from, to, adjust_func);
}
/*
* Insert new leaf to next cursor position.
* keep == 1: keep current cursor position.
* keep == 0, set cursor position to new leaf.
*/
static int insert_leaf(struct cursor *cursor, tuxkey_t childkey, struct buffer_head *leafbuf, int keep)
{
if(DEBUG_MODE_K==1)
{
printf("\t\t\t\t%25s[K] %25s %4d #in\n",__FILE__,__func__,__LINE__);
}
struct btree *btree = cursor->btree;
struct sb *sb = btree->sb;
int level = btree->root.depth;
block_t childblock = bufindex(leafbuf);
if (keep)
blockput(leafbuf);
else {
cursor_pop_blockput(cursor);
cursor_push(cursor, leafbuf, NULL);
}
while (level--) {
struct path_level *at = &cursor->path[level];
struct buffer_head *parentbuf = at->buffer;
struct bnode *parent = bufdata(parentbuf);
/* insert and exit if not full */
if (bcount(parent) < btree->sb->entries_per_node) {
bnode_add_index(parent, at->next, childblock, childkey);
if (!keep)
at->next++;
log_bnode_add(sb, bufindex(parentbuf), childblock, childkey);
mark_buffer_unify_non(parentbuf);
cursor_check(cursor);
return 0;
}
/* split a full index node */
struct buffer_head *newbuf = new_node(btree);
if (IS_ERR(newbuf))
return PTR_ERR(newbuf);
struct bnode *newnode = bufdata(newbuf);
unsigned half = bcount(parent) / 2;
u64 newkey = be64_to_cpu(parent->entries[half].key);
bnode_split(parent, half, newnode);
log_bnode_split(sb, bufindex(parentbuf), half, bufindex(newbuf));
/* if the cursor is in the new node, use that as the parent */
int child_is_left = at->next <= parent->entries + half;
if (!child_is_left) {
struct index_entry *newnext;
mark_buffer_unify_non(parentbuf);
newnext = newnode->entries + (at->next - &parent->entries[half]);
get_bh(newbuf);
level_replace_blockput(cursor, level, newbuf, newnext);
parentbuf = newbuf;
parent = newnode;
} else
mark_buffer_unify_non(newbuf);
bnode_add_index(parent, at->next, childblock, childkey);
if (!keep)
at->next++;
log_bnode_add(sb, bufindex(parentbuf), childblock, childkey);
mark_buffer_unify_non(parentbuf);
childkey = newkey;
childblock = bufindex(newbuf);
blockput(newbuf);
/*
* If child is in left bnode, we should keep the
* cursor position to child, otherwise adjust cursor
* to new bnode.
*/
keep = child_is_left;
}
/* Make new root bnode */
trace("add tree level");
struct buffer_head *newbuf = new_node(btree);
if (IS_ERR(newbuf))
return PTR_ERR(newbuf);
struct bnode *newroot = bufdata(newbuf);
block_t newrootblock = bufindex(newbuf);
block_t oldrootblock = btree->root.block;
int left_node = bufindex(cursor->path[0].buffer) != childblock;
bnode_init_root(newroot, 2, oldrootblock, childblock, childkey);
cursor_root_add(cursor, newbuf, newroot->entries + 1 + !left_node);
log_bnode_root(sb, newrootblock, 2, oldrootblock, childblock, childkey);
/* Change btree to point the new root */
btree->root.block = newrootblock;
btree->root.depth++;
mark_buffer_unify_non(newbuf);
tux3_mark_btree_dirty(btree);
cursor_check(cursor);
return 0;
}
/*
* This is range deletion. So, instead of adjusting balance of the
* space on sibling nodes for each change, this just removes the range
* and merges from right to left even if it is not same parent.
*
* +--------------- (A, B, C)--------------------+
* | | |
* +-- (AA, AB, AC) -+ +- (BA, BB, BC) -+ + (CA, CB, CC) +
* | | | | | | | | |
* (AAA,AAB)(ABA,ABB)(ACA,ACB) (BAA,BAB)(BBA)(BCA,BCB) (CAA)(CBA,CBB)(CCA)
*
* [less : A, AA, AAA, AAB, AB, ABA, ABB, AC, ACA, ACB, B, BA ... : greater]
*
* If we merged from cousin (or re-distributed), we may have to update
* the index until common parent. (e.g. removed (ACB), then merged
* from (BAA,BAB) to (ACA), we have to adjust B in root node to BB)
*
* See, adjust_parent_sep().
*
* FIXME: no re-distribute. so, we don't guarantee above than 50%
* space efficiency. And if range is end of key (truncate() case), we
* don't need to merge, and adjust_parent_sep().
*
* FIXME2: we may want to split chop work for each step. instead of
* blocking for a long time.
*/
int btree_chop(struct btree *btree, tuxkey_t start, u64 len)
{
if(DEBUG_MODE_K==1)
{
printf("\t\t\t\t%25s[K] %25s %4d #in\n",__FILE__,__func__,__LINE__);
}
struct sb *sb = btree->sb;
struct btree_ops *ops = btree->ops;
struct buffer_head **prev, *leafprev = NULL;
struct chopped_index_info *cii;
struct cursor *cursor;
tuxkey_t limit;
int ret, done = 0;
if (!has_root(btree))
return 0;
/* Chop all range if len >= TUXKEY_LIMIT */
limit = (len >= TUXKEY_LIMIT) ? TUXKEY_LIMIT : start + len;
prev = malloc(sizeof(*prev) * btree->root.depth);
if (prev == NULL)
return -ENOMEM;
memset(prev, 0, sizeof(*prev) * btree->root.depth);
cii = malloc(sizeof(*cii) * btree->root.depth);
if (cii == NULL) {
ret = -ENOMEM;
goto error_cii;
}
memset(cii, 0, sizeof(*cii) * btree->root.depth);
cursor = alloc_cursor(btree, 0);
if (!cursor) {
ret = -ENOMEM;
goto error_alloc_cursor;
}
down_write(&btree->lock);
ret = btree_probe(cursor, start);
if (ret)
goto error_btree_probe;
/* Walk leaves */
while (1) {
struct buffer_head *leafbuf;
tuxkey_t this_key;
/*
* FIXME: If leaf was merged and freed later, we don't
* need to redirect leaf and leaf_chop()
*/
if ((ret = cursor_redirect(cursor)))
goto out;
leafbuf = cursor_pop(cursor);
/* Adjust start and len for this leaf */
this_key = cursor_level_this_key(cursor);
if (start < this_key) {
if (limit < TUXKEY_LIMIT)
len -= this_key - start;
start = this_key;
}
ret = ops->leaf_chop(btree, start, len, bufdata(leafbuf));
if (ret) {
if (ret < 0) {
blockput(leafbuf);
goto out;
}
mark_buffer_dirty_non(leafbuf);
}
/* Try to merge this leaf with prev */
if (leafprev) {
if (try_leaf_merge(btree, leafprev, leafbuf)) {
trace(">>> can merge leaf %p into leaf %p", leafbuf, leafprev);
remove_index(cursor, cii);
mark_buffer_dirty_non(leafprev);
blockput_free(sb, leafbuf);
goto keep_prev_leaf;
}
blockput(leafprev);
}
leafprev = leafbuf;
keep_prev_leaf:
if (cursor_level_next_key(cursor) >= limit)
done = 1;
/* Pop and try to merge finished nodes */
while (done || cursor_level_finished(cursor)) {
struct buffer_head *buf;
int level = cursor->level;
struct chopped_index_info *ciil = &cii[level];
/* Get merge src buffer, and go parent level */
buf = cursor_pop(cursor);
/*
* Logging chopped indexes
* FIXME: If node is freed later (e.g. merged),
* we dont't need to log this
*/
if (ciil->count) {
log_bnode_del(sb, bufindex(buf), ciil->start,
ciil->count);
}
memset(ciil, 0, sizeof(*ciil));
/* Try to merge node with prev */
if (prev[level]) {
assert(level);
if (try_bnode_merge(sb, prev[level], buf)) {
trace(">>> can merge node %p into node %p", buf, prev[level]);
remove_index(cursor, cii);
mark_buffer_unify_non(prev[level]);
blockput_free_unify(sb, buf);
goto keep_prev_node;
}
blockput(prev[level]);
}
prev[level] = buf;
keep_prev_node:
if (!level)
goto chop_root;
}
/* Push back down to leaf level */
do {
ret = cursor_advance_down(cursor);
if (ret < 0)
goto out;
} while (ret);
}
chop_root:
/* Remove depth if possible */
while (btree->root.depth > 1 && bcount(bufdata(prev[0])) == 1) {
trace("drop btree level");
btree->root.block = bufindex(prev[1]);
btree->root.depth--;
tux3_mark_btree_dirty(btree);
/*
* We know prev[0] is redirected and dirty. So, in
* here, we can just cancel bnode_redirect by bfree(),
* instead of defered_bfree()
* FIXME: we can optimize freeing bnode without
* bnode_redirect, and if we did, this is not true.
*/
bfree(sb, bufindex(prev[0]), 1);
log_bnode_free(sb, bufindex(prev[0]));
blockput_free_unify(sb, prev[0]);
vecmove(prev, prev + 1, btree->root.depth);
}
ret = 0;
out:
if (leafprev)
blockput(leafprev);
for (int i = 0; i < btree->root.depth; i++) {
if (prev[i])
blockput(prev[i]);
}
release_cursor(cursor);
error_btree_probe:
up_write(&btree->lock);
free_cursor(cursor);
error_alloc_cursor:
free(cii);
error_cii:
free(prev);
return ret;
}
static Block*
ext2blockread(Fsys *fsys, u64int vbno)
{
Block *bitb;
Group g;
uchar *bits;
u32int bno, boff, bitblock;
u64int bitpos;
Ext2 *fs;
fs = fsys->priv;
if(vbno >= fs->nblock)
return nil;
bno = vbno;
if(bno != vbno)
return nil;
/*
if(bno < fs->firstblock)
return diskread(fs->disk, fs->blocksize, (u64int)bno*fs->blocksize);
*/
if(bno < fs->firstblock)
return nil;
bno -= fs->firstblock;
if(ext2group(fs, bno/fs->blockspergroup, &g) < 0){
if(debug)
fprint(2, "loading group: %r...");
return nil;
}
/*
if(debug)
fprint(2, "ext2 group %d: bitblock=%ud inodebitblock=%ud inodeaddr=%ud freeblocks=%ud freeinodes=%ud useddirs=%ud\n",
(int)(bno/fs->blockspergroup),
g.bitblock,
g.inodebitblock,
g.inodeaddr,
g.freeblockscount,
g.freeinodescount,
g.useddirscount);
if(debug)
fprint(2, "group %d bitblock=%d...", bno/fs->blockspergroup, g.bitblock);
*/
bitblock = g.bitblock;
bitpos = (u64int)bitblock*fs->blocksize;
if((bitb = diskread(fs->disk, fs->blocksize, bitpos)) == nil){
if(debug)
fprint(2, "loading bitblock: %r...");
return nil;
}
bits = bitb->data;
boff = bno%fs->blockspergroup;
if((bits[boff>>3] & (1<<(boff&7))) == 0){
if(debug)
fprint(2, "block %d not allocated in group %d: bitblock %d/%lld bits[%d] = %#x\n",
boff, bno/fs->blockspergroup,
(int)bitblock,
bitpos,
boff>>3,
bits[boff>>3]);
blockput(bitb);
return nil;
}
blockput(bitb);
bno += fs->firstblock;
return diskread(fs->disk, fs->blocksize, (u64int)bno*fs->blocksize);
}