/**
* Add item as right-sibling of node (cannot be the root node).
*/
void
etree_add_right_sibling(etree_t *tree, void *node, void *item)
{
node_t *n, *i;
etree_check(tree);
g_assert(node != NULL);
g_assert(item != NULL);
g_assert(etree_is_orphan(tree, item));
g_assert(!etree_is_root(tree, node));
n = ptr_add_offset(node, tree->offset);
i = ptr_add_offset(item, tree->offset);
i->parent = n->parent;
i->sibling = n->sibling;
n->sibling = i;
if (NULL == i->sibling && etree_is_extended(tree)) {
nodex_t *px = (nodex_t *) n->parent;
px->last_child = i;
}
tree->count = 0; /* Tree count is now unknown */
}
/**
* Prepend child to parent.
*
* This is always a fast operation.
*/
void
etree_prepend_child(etree_t *tree, void *parent, void *child)
{
node_t *cn, *pn;
etree_check(tree);
g_assert(parent != NULL);
g_assert(child != NULL);
g_assert(etree_is_orphan(tree, child));
cn = ptr_add_offset(child, tree->offset);
pn = ptr_add_offset(parent, tree->offset);
cn->parent = pn;
cn->sibling = pn->child;
pn->child = cn;
if (etree_is_extended(tree)) {
nodex_t *px = (nodex_t *) pn;
if (NULL == px->last_child) {
g_assert(NULL == cn->sibling); /* Parent did not have any child */
px->last_child = cn;
}
}
tree->count = 0; /* Tree count is now unknown */
}
/**
* Add item as left-sibling of node (cannot be the root node).
*
* This is inefficient if the node is not the first child (the head of the
* sibling list) given that the siblings are linked through a one-way list.
*/
void
etree_add_left_sibling(etree_t *tree, void *node, void *item)
{
node_t *n, *i;
etree_check(tree);
g_assert(node != NULL);
g_assert(item != NULL);
g_assert(etree_is_orphan(tree, item));
g_assert(!etree_is_root(tree, node));
n = ptr_add_offset(node, tree->offset);
i = ptr_add_offset(item, tree->offset);
i->parent = n->parent;
i->sibling = n;
if (n == n->parent->child) {
/* node is first child, optimized case */
n->parent->child = i;
} else {
node_t *p;
for (p = n->parent->child; p->sibling != n; p = p->sibling)
/* empty */;
g_assert(p != NULL); /* previous node found */
p->sibling = i;
}
tree->count = 0; /* Tree count is now unknown */
}
/**
* If not already done, initiate bitmap checking by creating all the currently
* defined bitmaps in memory, zeroed, so that we can check that all the pages
* flagged as used are indeed referred to by either a big key or a big value.
*
* @return TRUE if OK.
*/
gboolean
big_check_start(DBM *db)
{
DBMBIG *dbg = db->big;
long i;
if (-1 == dbg->fd && -1 == big_open(dbg))
return FALSE;
if (dbg->bitcheck != NULL)
return TRUE;
/*
* The array of bitmaps is zeroed, with all the bits corresponding to each
* bitmap page (the bit 0) set.
*
* Looping over the big keys and values and marking their blocks set will
* set additional bits in these checking maps, which at the end will be
* compared to the ones on disk.
*/
dbg->bitcheck = halloc0(BIG_BLKSIZE * dbg->bitmaps);
for (i = 0; i < dbg->bitmaps; i++) {
bit_field_t *map = ptr_add_offset(dbg->bitcheck, i * BIG_BLKSIZE);
bit_field_set(map, 0); /* Bit 0 is for the bitmap itself */
}
return TRUE;
}
/**
* Attempt to grow the output buffer.
*
* @param zs the zlib stream object
* @param maxout maximum length of dynamically-allocated buffer (0 = none)
*
* @return TRUE if we can resume processing after the output buffer was
* expanded, FALSE if the buffer cannot be expanded (not dynamically allocated
* or reached the maximum size).
*/
static bool
zlib_stream_grow_output(zlib_stream_t *zs, size_t maxout)
{
if (zs->allocated) {
z_streamp z = zs->z;
/*
* Limit growth if asked to do so.
*
* This is used when inflating, usually, to avoid being given a small
* amount of data that will expand into megabytes...
*/
if (maxout != 0 && zs->outlen >= maxout) {
g_warning("%s(): reached maximum buffer size (%zu bytes): "
"input=%zu, output=%zu",
G_STRFUNC, maxout, zs->inlen, zs->outlen);
return FALSE; /* Cannot continue */
}
zs->outlen += OUT_GROW;
zs->out = hrealloc(zs->out, zs->outlen);
z->next_out = ptr_add_offset(zs->out, zs->outlen - OUT_GROW);
z->avail_out = OUT_GROW;
return TRUE; /* Can process remaining input */
}
g_warning("%s(): under-estimated output buffer size: "
"input=%zu, output=%zu", G_STRFUNC, zs->inlen, zs->outlen);
return FALSE; /* Cannot continue */
}
/**
* Initialize and load linked items into a list.
*
* This routine is meant to allow the creation of an expanded list from
* homogeneous items that happen to be linked into another data structure
* through a single pointer.
*
* It is useful to allow reuse of code that can process such lists, such
* as xslist_sort(), xslist_shuffle(), etc... It is naturally up to the
* caller to then refetch the proper head pointer.
*
* @param list the list into which we are loading items
* @param head first data item part of the linked list (NULL possible)
* @param offset the offset of the expanded link field within items
* @param link_offset the offset of the linking field in the chaining struct
*
* @return the amount of loaded items, as a convenience.
*/
size_t
xslist_load(xslist_t *list, void *head, size_t offset, size_t link_offset)
{
xslink_t *lk, *next;
size_t n;
g_assert(list != NULL);
g_assert(size_is_non_negative(offset));
xslist_init(list, offset, link_offset);
if G_UNLIKELY(NULL == head)
return 0;
lk = ptr_add_offset(head, offset);
list->head = lk;
for (n = 1; NULL != (next = xslist_next(list, lk)); n++, lk = next)
/* empty */;
list->tail = lk;
list->count = n;
safety_assert(xslist_length(list, list->head) == list->count);
return n;
}
/**
* Read data from the pmsg list into supplied buffer. Copied data is
* removed from the list.
*
* @param slist the pmsg list
* @param buf start of buffer where data must be copied
* @param len length of buffer
*
* @return amount of copied bytes.
*/
size_t
pmsg_slist_read(slist_t *slist, void *buf, size_t len)
{
slist_iter_t *iter;
size_t remain = len;
void *p;
g_assert(slist != NULL);
iter = slist_iter_removable_on_head(slist);
p = buf;
while (remain != 0 && slist_iter_has_item(iter)) {
pmsg_t *mb = slist_iter_current(iter);
int n;
n = pmsg_read(mb, p, remain);
remain -= n;
p = ptr_add_offset(p, n);
if (0 == pmsg_size(mb)) { /* Fully copied message */
pmsg_free(mb);
slist_iter_remove(iter); /* Warning: moves to next */
} else {
break; /* No need to continue on partial copy */
}
}
slist_iter_free(&iter);
return len - remain;
}
/**
* Internal recursive matching routine.
*/
static void *
etree_find_depth_internal(const etree_t *tree, node_t *root,
unsigned curdepth, unsigned maxdepth, match_fn_t match, void *data)
{
node_t *n, *next;
void *item;
etree_check(tree);
item = ptr_add_offset(root, -tree->offset);
if ((*match)(item, data))
return item;
if (maxdepth == curdepth)
return NULL;
for (n = root->child; n != NULL; n = next) {
next = n->sibling;
item = etree_find_depth_internal(tree, n,
curdepth + 1, maxdepth, match, data);
if (item != NULL)
return item;
}
return NULL;
}
/**
* Mark blocks in the supplied vector as allocated in the checking bitmap.
*
* @param db the sdbm database
* @param bvec vector where allocated block numbers are stored
* @param bcnt amount of blocks in vector
*/
static void
big_file_mark_used(DBM *db, const void *bvec, int bcnt)
{
DBMBIG *dbg = db->big;
const void *q;
int n;
if (!big_check_start(db))
return;
for (q = bvec, n = bcnt; n > 0; n--) {
size_t bno = peek_be32(q);
bit_field_t *map;
long bmap;
size_t bit;
bmap = bno / BIG_BITCOUNT; /* Bitmap handling this block */
bit = bno & (BIG_BITCOUNT - 1); /* Index within bitmap */
q = const_ptr_add_offset(q, sizeof(guint32));
/*
* It's because of this sanity check that we don't want to consider
* the bitcheck field as a huge continuous map. Also doing that would
* violate the encapsulation: we're not supposed to know how bits are
* allocated in the field.
*/
if (bmap >= dbg->bitmaps)
continue;
map = ptr_add_offset(dbg->bitcheck, bmap * BIG_BLKSIZE);
bit_field_set(map, bit);
}
}
/**
* Randomly swap 1/128 of the array items.
*/
static void
vsort_perturb_sorted_array(void *array, size_t cnt, size_t isize)
{
size_t n;
size_t i;
n = cnt / 128;
for (i = 0; i < n; i++) {
size_t a = random_value(cnt - 1);
size_t b = random_value(cnt - 1);
void *x = ptr_add_offset(array, a * isize);
void *y = ptr_add_offset(array, b * isize);
SWAP(x, y, isize);
}
}
/**
* Make sure vector of block numbers is ordered and points to allocated data,
* but was not already flagged as being used by another key / value.
*
* @param what string describing what is being tested (key or value)
* @param db the sdbm database
* @param bvec vector where allocated block numbers are stored
* @param bcnt amount of blocks in vector
*
* @return TRUE on success.
*/
static gboolean
big_file_check(const char *what, DBM *db, const void *bvec, int bcnt)
{
size_t prev_bno = 0; /* 0 is invalid: it's the first bitmap */
const void *q;
int n;
if (!big_check_start(db))
return TRUE; /* Cannot validate, assume it's OK */
for (q = bvec, n = bcnt; n > 0; n--) {
size_t bno = peek_be32(q);
bit_field_t *map;
long bmap;
size_t bit;
if (!big_block_is_allocated(db, bno)) {
g_warning("sdbm: \"%s\": "
"%s from .pag refers to unallocated block %lu in .dat",
sdbm_name(db), what, (unsigned long) bno);
return FALSE;
}
if (prev_bno != 0 && bno <= prev_bno) {
g_warning("sdbm: \"%s\": "
"%s from .pag lists unordered block list (corrupted file?)",
sdbm_name(db), what);
return FALSE;
}
q = const_ptr_add_offset(q, sizeof(guint32));
prev_bno = bno;
/*
* Make sure block is not used by someone else.
*
* Because we mark blocks as used in big keys and values only after
* we validated both the key and the value for a given pair, we cannot
* detect shared blocks between the key and value of a pair.
*/
bmap = bno / BIG_BITCOUNT; /* Bitmap handling this block */
bit = bno & (BIG_BITCOUNT - 1); /* Index within bitmap */
g_assert(bmap < db->big->bitmaps);
map = ptr_add_offset(db->big->bitcheck, bmap * BIG_BLKSIZE);
if (bit_field_get(map, bit)) {
g_warning("sdbm: \"%s\": "
"%s from .pag refers to already seen block %lu in .dat",
sdbm_name(db), what, (unsigned long) bno);
return FALSE;
}
}
return TRUE;
}
/**
* General tree traversal routine, in depth-first order.
*
* The "enter" function is called when we enter a node, and its returned
* value is monitored: a FALSE indicates that the node should be skipped
* (which includes its children) and that the action callback should not be
* invoked. When missing, it is as if the "enter" callback had returned TRUE.
*
* The "action" callback can be invokded before or after processing the node's
* children, as indicated by the flags. That callback is allowed to free the
* traversed node.
*
* @param tree the tree descriptor
* @param root the item at which traversal starts
* @param flags nodes to visit + when to invoke action callback
* @param curdepth current depth
* @param maxdepth 0 = root node only, 1 = root + its children, etc...
* @param enter (optional) callback when we enter a node
* @param action (mandatory) action on the node
* @param data user-defined argument passed to callbacks
*
* @return amount of nodes visited.
*/
static size_t
etree_traverse_internal(const etree_t *tree, node_t *root,
unsigned flags, unsigned curdepth, unsigned maxdepth,
match_fn_t enter, data_fn_t action, void *data)
{
size_t visited = 0;
void *item;
bool actionable;
node_t *child;
etree_check(tree);
if (curdepth > maxdepth)
return 0;
/*
* Check whether we need to execute action on the node.
*/
child = root->child; /* Save pointer in case "action" frees node */
if ((flags & ETREE_TRAVERSE_NON_LEAVES) && NULL != child)
actionable = TRUE;
else if ((flags & ETREE_TRAVERSE_LEAVES) && NULL == child)
actionable = TRUE;
else
actionable = FALSE;
item = ptr_add_offset(root, -tree->offset);
if (enter != NULL && !(*enter)(item, data))
return 0;
if (actionable && (flags & ETREE_CALL_BEFORE))
(*action)(item, data); /* MUST NOT free node */
/*
* Only visit children when we've not reached the maximum depth.
*/
if (curdepth != maxdepth) {
node_t *n, *next;
for (n = child; n != NULL; n = next) {
next = n->sibling;
visited += etree_traverse_internal(tree, n, flags,
curdepth + 1, maxdepth, enter, action, data);
}
}
if (actionable && (flags & ETREE_CALL_AFTER))
(*action)(item, data); /* Can safely free node */
return visited + 1; /* "+1" for this node */
}
/**
* Append child to parent.
*
* If this is a frequent operation, consider using an extended tree.
*/
void
etree_append_child(etree_t *tree, void *parent, void *child)
{
node_t *cn, *pn;
etree_check(tree);
g_assert(parent != NULL);
g_assert(child != NULL);
g_assert(etree_is_orphan(tree, child));
cn = ptr_add_offset(child, tree->offset);
pn = ptr_add_offset(parent, tree->offset);
if (etree_is_extended(tree)) {
nodex_t *px = ptr_add_offset(parent, tree->offset);
if (px->last_child != NULL) {
node_t *lcn = px->last_child;
g_assert(0 == ptr_cmp(lcn->parent, px));
g_assert(NULL == lcn->sibling);
lcn->sibling = cn;
} else {
g_assert(NULL == px->child);
px->child = cn;
}
px->last_child = cn;
} else {
if (NULL == pn->child) {
pn->child = cn;
} else {
node_t *lcn = etree_node_last_sibling(pn->child);
lcn->sibling = cn;
}
}
cn->parent = pn;
tree->count = 0; /* Tree count is now unknown */
}
/**
* @return pointer to last child of item, NULL if leaf item.
*/
void *
etree_last_child(const etree_t *tree, const void *item)
{
etree_check(tree);
if (etree_is_extended(tree)) {
const nodex_t *n = const_ptr_add_offset(item, tree->offset);
if (NULL == n->last_child)
return NULL;
return ptr_add_offset(n->last_child, -tree->offset);
} else {
const node_t *n = const_ptr_add_offset(item, tree->offset);
node_t *sn = etree_node_last_sibling(n->child);
if (NULL == sn)
return NULL;
return ptr_add_offset(sn, -tree->offset);
}
}
/**
* Fetch value from the hash set.
*
* @param ht the hash table
* @param key the key being looked up
*
* @return found value, or NULL if not found.
*/
void *
hevset_lookup(const hevset_t *ht, const void *key)
{
size_t idx;
void *kptr;
hevset_check(ht);
g_assert(key != NULL);
idx = hash_lookup_key(HASH(ht), key);
if ((size_t) -1 == idx)
return NULL;
kptr = deconstify_pointer(ht->kset.keys[idx]);
return ptr_add_offset(kptr, -ht->offset);
}
/**
* Look up key in the tree, returning the associated key item if found.
*
* @param tree the red-black tree
* @param key pointer to the key structure (NOT a node)
*
* @return found item associated with key, NULL if not found.
*/
void *
erbtree_lookup(const erbtree_t *tree, const void *key)
{
rbnode_t *parent;
bool is_left;
rbnode_t *rn;
erbtree_check(tree);
g_assert(key != NULL);
if (erbtree_is_extended(tree)) {
rn = do_lookup_ext(ERBTREE_E(tree), key, &parent, &is_left);
} else {
rn = do_lookup(tree, key, &parent, &is_left);
}
return NULL == rn ? NULL : ptr_add_offset(rn, -tree->offset);
}
/**
* Load text symbols from the file into supplied table.
*
* @param bc the BFD context pointing to the file
* @param st the symbol table where symbols should be added
*/
static void
bfd_util_load_text(bfd_ctx_t *bc, symbols_t *st)
{
long i;
asymbol* empty;
void *p;
bfd_ctx_check(bc);
g_assert(st != NULL);
if (0 == bc->count)
return;
mutex_lock_fast(&bc->lock);
g_assert(bc->symbols != NULL);
empty = bfd_make_empty_symbol(bc->handle);
symbols_lock(st);
for (
i = 0, p = bc->symbols;
i < bc->count;
i++, p = ptr_add_offset(p, bc->symsize)
) {
asymbol *sym;
symbol_info syminfo;
sym = bfd_minisymbol_to_symbol(bc->handle, bc->dynamic, p, empty);
bfd_get_symbol_info(bc->handle, sym, &syminfo);
if ('T' == syminfo.type || 't' == syminfo.type) {
const char *name = bfd_asymbol_name(sym);
if (name != NULL && name[0] != '.') {
void *addr = ulong_to_pointer(syminfo.value);
symbols_append(st, addr, name);
}
}
}
symbols_unlock(st);
mutex_unlock_fast(&bc->lock);
}
/**
* Fetch key/value from the hash table, returning whether the key exists.
* If it does, the original value pointer is written valptr.
*
* @param ht the hash table
* @param key the key being looked up
* @param valptr if non-NULL, where the original value pointer is written
*
* @return whether key exists in the table.
*/
bool
hikset_lookup_extended(const hikset_t *ht, const void *key, void **valptr)
{
size_t idx;
hikset_check(ht);
idx = hash_lookup_key(HASH(ht), &key);
if ((size_t) -1 == idx)
return FALSE;
if (valptr != NULL) {
void *kptr = deconstify_pointer(ht->kset.keys[idx]);
*valptr = ptr_add_offset(kptr, -ht->offset);
}
return TRUE;
}
/**
* End bitmap allocation checks that have been started by the usage of one
* of the bigkey_mark_used() and bigval_mark_used() routines.
*
* @return the amount of corrections brought to the bitmap, 0 meaning
* everything was consistent.
*/
size_t
big_check_end(DBM *db)
{
DBMBIG *dbg = db->big;
long i;
size_t adjustments = 0;
if (NULL == dbg->bitcheck)
return 0;
for (i = 0; i < dbg->bitmaps; i++) {
if (!fetch_bitbuf(db, i)) {
adjustments += BIG_BITCOUNT; /* Say, everything was wrong */
} else {
guint8 *p = ptr_add_offset(dbg->bitcheck, i * BIG_BLKSIZE);
guint8 *q = dbg->bitbuf;
size_t j;
size_t old_adjustments = adjustments;
for (j = 0; j < BIG_BLKSIZE; j++, p++, q++) {
guint8 mismatch = *p ^ *q;
if (mismatch) {
adjustments += bits_set(mismatch);
*q = *p;
}
}
if (old_adjustments != adjustments) {
size_t adj = adjustments - old_adjustments;
flush_bitbuf(db);
g_warning("sdbm: \"%s\": adjusted %lu bit%s in bitmap #%ld",
sdbm_name(db), (unsigned long) adj, 1 == adj ? "" : "s", i);
}
}
}
HFREE_NULL(dbg->bitcheck);
return adjustments;
}
/**
* Computes the root of the tree, starting from any item.
*
* This disregards the actual root in the etree_t structure passed, which may
* be inaccurate, i.e. a sub-node of the actual tree. The only accurate
* information that etree_t must contain is the offset of the node_t within
* the items.
*
* @param tree the tree descriptor (with possible inaccurate root)
* @param item an item belonging to the tree
*
* @return the root of the tree to which item belongs.
*/
void *
etree_find_root(const etree_t *tree, const void *item)
{
const node_t *n, *p;
void *root;
etree_check(tree);
g_assert(item != NULL);
n = const_ptr_add_offset(item, tree->offset);
for (p = n; p != NULL; p = n->parent)
n = p;
root = ptr_add_offset(deconstify_pointer(n), -tree->offset);
g_assert(etree_is_orphan(tree, root)); /* No parent, no sibling */
return root;
}
/**
* Minimal pseudo-random number generation, combining a simple PRNG with
* past-collected entropy.
*
* @return a 31-bit random number.
*/
static int
entropy_rand31(void)
{
int result;
static size_t offset;
result = rand31();
/*
* Combine with previously generated entropy to create even better
* randomness. That previous entropy is refreshed each time a new
* entropy collection cycle is initiated. We simply loop over the
* five 32-bit words, interpreted in a big-endian way.
*/
result += peek_be32(ptr_add_offset(&entropy_previous, offset));
offset = (offset + 4) % sizeof entropy_previous;
return result & RAND31_MASK;
}
/**
* Fetch value from the hash set.
*
* @param ht the hash table
* @param key the key being looked up
*
* @return found value, or NULL if not found.
*/
void *
hikset_lookup(const hikset_t *ht, const void *key)
{
size_t idx;
void *kptr;
hikset_check(ht);
idx = hash_lookup_key(HASH(ht), &key);
if ((size_t) -1 == idx)
return NULL;
/*
* We stored a pointer to the key in the value structure.
* To get the start of the value, we simply offset that pointer.
*/
kptr = deconstify_pointer(ht->kset.keys[idx]);
return ptr_add_offset(kptr, -ht->offset);
}
/**
* Fetch next entry from iterator.
*
* @param hxi the hash table iterator
* @param vp where value is written, if non-NULL
*
* @return TRUE if a new entry exists, FALSE otherwise.
*/
bool
hikset_iter_next(hikset_iter_t *hxi, void **vp)
{
const hikset_t *hx;
hikset_iter_check(hxi);
hx = hxi->hx;
while (hxi->pos < hx->kset.size && !HASH_IS_REAL(hx->kset.hashes[hxi->pos]))
hxi->pos++;
if (hxi->pos >= hx->kset.size)
return FALSE;
if (vp != NULL) {
void *kptr = deconstify_pointer(hx->kset.keys[hxi->pos]);
*vp = ptr_add_offset(kptr, -hx->offset);
}
hxi->pos++;
return TRUE;
}
/**
* Traverse table, invoking callback for each entry.
*
* @param hx the hash table
* @param fn callback to invoke
* @param data additional callback parameter
*/
void
hikset_foreach(const hikset_t *hx, data_fn_t fn, void *data)
{
unsigned *hp, *end;
size_t i, n;
hikset_check(hx);
end = &hx->kset.hashes[hx->kset.size];
hash_refcnt_inc(HASH(hx)); /* Prevent any key relocation */
for (i = n = 0, hp = hx->kset.hashes; hp != end; i++, hp++) {
if (HASH_IS_REAL(*hp)) {
void *kptr = deconstify_pointer(hx->kset.keys[i]);
(*fn)(ptr_add_offset(kptr, -hx->offset), data);
n++;
}
}
g_assert(n == hx->kset.items);
hash_refcnt_dec(HASH(hx));
}
/**
* Collect entropy by randomly feeding values from array.
*/
static void
entropy_array_data_collect(SHA1Context *ctx,
enum entropy_data data, void *ary, size_t len, size_t elem_size)
{
size_t i;
void *p;
g_assert(ctx != NULL);
g_assert(ary != NULL);
g_assert(size_is_non_negative(len));
g_assert(size_is_positive(elem_size));
entropy_array_shuffle(ary, len, elem_size);
for (i = 0, p = ary; i < len; i++, p = ptr_add_offset(p, elem_size)) {
switch (data) {
case ENTROPY_ULONG:
sha1_feed_ulong(ctx, *(unsigned long *) p);
break;
case ENTROPY_STRING:
sha1_feed_string(ctx, *(char **) p);
break;
case ENTROPY_STAT:
sha1_feed_stat(ctx, *(char **) p);
break;
case ENTROPY_FSTAT:
sha1_feed_fstat(ctx, *(int *) p);
break;
case ENTROPY_DOUBLE:
sha1_feed_double(ctx, *(double *) p);
break;
case ENTROPY_POINTER:
sha1_feed_pointer(ctx, *(void **) p);
break;
}
}
}
/**
* Traverse table, invoking callback for each entry and removing it when
* the callback function returns TRUE.
*
* @param hx the hash table
* @param fn callback to invoke
* @param data additional callback parameter
*
* @return the number of entries removed from the hash table.
*/
size_t
hikset_foreach_remove(hikset_t *hx, data_rm_fn_t fn, void *data)
{
unsigned *hp, *end;
size_t i, n, nr;
hikset_check(hx);
end = &hx->kset.hashes[hx->kset.size];
hash_refcnt_inc(HASH(hx)); /* Prevent any key relocation */
for (i = n = nr = 0, hp = hx->kset.hashes; hp != end; i++, hp++) {
if (HASH_IS_REAL(*hp)) {
void *kptr = deconstify_pointer(hx->kset.keys[i]);
bool r = (*fn)(ptr_add_offset(kptr, -hx->offset), data);
n++;
if (r) {
nr++;
hash_keyset_erect_tombstone(&hx->kset, i);
hx->stamp++;
}
}
}
g_assert(n == hx->kset.items);
g_assert(nr <= hx->kset.items);
hash_refcnt_dec(HASH(hx));
hx->kset.items -= nr;
if (nr != 0)
hash_resize_as_needed(HASH(hx));
return nr;
}
/**
* Detach item and all its sub-tree from a tree, making it the new root
* of a smaller tree.
*/
void
etree_detach(etree_t *tree, void *item)
{
node_t *n;
etree_check(tree);
g_assert(item != NULL);
n = ptr_add_offset(item, tree->offset);
if (NULL == n->parent) {
/* Root node already, cannot have any sibling */
g_assert(NULL == n->sibling);
g_assert(n == tree->root);
tree->root = NULL; /* Root node is now detached */
tree->count = 0;
} else {
node_t *parent = n->parent;
if (n == parent->child) {
if (etree_is_extended(tree)) {
nodex_t *px = (nodex_t *) parent;
if (n == px->last_child) {
g_assert(NULL == n->sibling); /* Last child! */
px->child = px->last_child = NULL;
} else {
g_assert(NULL != n->sibling); /* Not last child */
parent->child = n->sibling;
}
} else {
parent->child = n->sibling;
}
} else {
node_t *cn;
bool found = FALSE;
/*
* Not removing first child of parent, so locate its previous
* sibling in the list of the parent's immediate children.
*/
for (cn = parent->child; cn != NULL; cn = cn->sibling) {
if (cn->sibling == n) {
found = TRUE;
break;
}
}
g_assert(found); /* Must find it or tree is corrupted */
cn->sibling = n->sibling; /* Remove node ``n'' from list */
/*
* Update ``last_child'' in the parent node if we removed the
* last child node.
*/
if (etree_is_extended(tree)) {
nodex_t *px = (nodex_t *) parent;
if (n == px->last_child) {
g_assert(NULL == n->sibling);
px->last_child = cn;
}
}
}
n->sibling = NULL; /* Node is now a root node */
tree->count = 0; /* Count is now unknown */
}
}
/**
* Allocate blocks (consecutive if possible) from the .dat file.
* Block numbers are written back in the specified vector, in sequence.
*
* Blocks are always allocated with increasing block numbers, i.e. the list
* of block numbers returned is guaranteed to be sorted. This will help
* upper layers to quickly determine whether all the blocks are contiguous
* for instance.
*
* The file is extended as necessary to be able to allocate the blocks but
* this is only done when there are no more free blocks available.
*
* @param db the sdbm database
* @param bvec vector where allocated block numbers will be stored
* @param bcnt amount of blocks in vector (amount to allocate)
*
* @return TRUE if we were able to allocate all the requested blocks.
*/
static gboolean
big_file_alloc(DBM *db, void *bvec, int bcnt)
{
DBMBIG *dbg = db->big;
size_t first;
int n;
void *q;
int bmap = 0; /* Initial bitmap from which we allocate */
g_assert(bcnt > 0);
g_return_val_if_fail(NULL == dbg->bitcheck, FALSE);
if (-1 == dbg->fd && -1 == big_open(dbg))
return FALSE;
/*
* First try to allocate all the blocks sequentially.
*/
retry:
first = big_falloc_seq(db, bmap, bcnt);
if (first != 0) {
while (bcnt-- > 0) {
bvec = poke_be32(bvec, first++);
}
goto success;
}
/*
* There are no "bcnt" consecutive free blocks in the file.
*
* Before extending the file, we're going to fill the holes as much
* as possible.
*/
for (first = 0, q = bvec, n = bcnt; n > 0; n--) {
first = big_falloc(db, first + 1);
if (0 == first)
break;
q = poke_be32(q, first);
}
if (0 == n)
goto success; /* Found the requested "bcnt" free blocks */
/*
* Free the incompletely allocated blocks: since we're about to extend
* the file, we'll use consecutive blocks from the new chunk governed
* by the added empty bitmap.
*/
for (q = bvec, n = bcnt - n; n > 0; n--) {
first = peek_be32(q);
big_ffree(db, first);
q = ptr_add_offset(q, sizeof(guint32));
}
/*
* Extend the file by allocating another bitmap.
*/
g_assert(0 == bmap); /* Never retried yet */
if (dbg->bitbuf_dirty && !flush_bitbuf(db))
return FALSE;
memset(dbg->bitbuf, 0, BIG_BLKSIZE);
bit_field_set(dbg->bitbuf, 0); /* First page is the bitmap itself */
dbg->bitbno = dbg->bitmaps * BIG_BITCOUNT;
dbg->bitmaps++;
/*
* Now retry starting to allocate blocks from the newly added bitmap.
*
* This will likely succeed if we're trying to allocate less than 8 MiB
* worth of data (with 1 KiB blocks).
*/
bmap = dbg->bitmaps - 1;
goto retry;
success:
/*
* We successfully allocated blocks from the bitmap.
*
* If the database is not volatile, we need to flush the bitmap to disk
* immediately in case of a crash, to avoid reusing these parts of the file.
*/
if (!db->is_volatile && dbg->bitbuf_dirty && !flush_bitbuf(db)) {
/* Cannot flush -> cannot allocate the blocks: free them */
for (q = bvec, n = bcnt; n > 0; n--) {
first = peek_be32(q);
big_ffree(db, first);
q = ptr_add_offset(q, sizeof(guint32));
}
return FALSE;
}
return TRUE; /* Succeeded */
}
/**
* Look whether semi-reliable UDP header corresponds to valid traffic.
*
* This routine is only used for ambiguous traffic that looks like both
* Gnutella and semi-reliable UDP: we want to make sure we're not mistaking
* a legitimate semi-reliable fragment / ACK for a Gnutella message.
*
* @param utp already classified semi-reliable protocol
* @param s socket which received the message
* @param data received data
* @param len length of data
*
* @return TRUE if message corresponds to valid semi-reliable UDP traffic.
*/
static bool
udp_is_valid_semi_reliable(enum udp_traffic utp, const gnutella_socket_t *s,
const void *data, size_t len)
{
struct ut_header uth;
void *message = NULL;
size_t msglen;
bool valid = TRUE;
/*
* Since we're talking about an ambiguous message, it is highly unlikely
* we'll ever be called with an acknowledgement: they should have been
* ruled out earlier as improbable since ACKs are short message, much
* shorter than a Gnuella header typically.
*
* So we'll only handle fragments for now, assuming ACKs are legitimate.
*/
gnet_stats_inc_general(GNR_UDP_AMBIGUOUS_DEEPER_INSPECTION);
uth.count = udp_reliable_header_get_count(data);
if (0 == uth.count)
return TRUE; /* Acknoweldgments */
uth.part = udp_reliable_header_get_part(data) - 1; /* Zero-based */
uth.flags = udp_reliable_header_get_flags(data);
uth.seqno = udp_reliable_header_get_seqno(data);
/*
* We're going to ask the RX layer about the message: is it a known
* sequence ID for this host?
*
* This works only for messages with more than one fragment, of course,
* but chances are that, for these, we would have possibly already
* received another fragment, not mistaken as a Gnutella message...
*
* This is OK for acknowledged fragments: we're not going to acknowledge
* the unprocessed fragment, but we'll receive other fragments of the
* message, and later on we'll get a retransmission of the unprocessed
* fragment, which this time will be validated since we have already
* partially received the message.
*/
if (uth.count > 1) {
rxdrv_t *rx;
gnet_host_t from;
gnet_host_set(&from, s->addr, s->port);
rx = udp_get_rx_semi_reliable(utp, s->addr, len);
return NULL == rx ? FALSE : ut_valid_message(rx, &uth, &from);
}
/*
* We're facing a single-fragment message.
*
* We can trivially probe it and validate it to see whether it can still
* be interpreted as a valid Gnutella message on its own... If the answer
* is yes, then we can assert we're facing a valid semi-reliable UDP
* message.
*
* For deflated payloads, we already validated that the start of the
* payload is a well-formed zlib header, but we'll attempt deflation anyway
* so we will know for sure whether it's a valid message!
*
* Of course we're doing here work that will have to be redone later when
* processing the message, but this is for proper classification and not
* happening very often: only on a very very small fraction of messages for
* which there is a high level of ambiguity.
*/
g_assert(0 == uth.part); /* First (and only) fragment */
if (uth.flags & UDP_RF_DEFLATED) {
int outlen = settings_max_msg_size();
int ret;
message = xmalloc(outlen);
ret = zlib_inflate_into(
const_ptr_add_offset(data, UDP_RELIABLE_HEADER_SIZE),
len - UDP_RELIABLE_HEADER_SIZE,
message, &outlen);
if (ret != Z_OK) {
valid = FALSE; /* Does not inflate properly */
goto done;
}
msglen = outlen;
} else {
message = ptr_add_offset(
deconstify_pointer(data), UDP_RELIABLE_HEADER_SIZE);
msglen = len - UDP_RELIABLE_HEADER_SIZE;
}
switch (utp) {
case SEMI_RELIABLE_GTA:
/*
* Assume message is valid if the Gnutella size header is consistent
* with the length of the whole message.
*/
{
uint16 size;
switch (gmsg_size_valid(message, &size)) {
case GMSG_VALID:
case GMSG_VALID_MARKED:
break;
case GMSG_VALID_NO_PROCESS: /* Header flags undefined for now */
case GMSG_INVALID:
valid = FALSE;
goto done;
}
valid = (size_t) size + GTA_HEADER_SIZE == msglen;
}
break;
case SEMI_RELIABLE_GND:
valid = TRUE; /* For now */
break;
case GNUTELLA:
case DHT:
case RUDP:
case UNKNOWN:
g_assert_not_reached();
}
done:
if (uth.flags & UDP_RF_DEFLATED)
xfree(message);
return valid;
}
/**
* Insert node in tree.
*
* @return the existing key if the key already existed, NULL if the node
* was properly inserted.
*/
void * G_HOT
erbtree_insert(erbtree_t *tree, rbnode_t *node)
{
rbnode_t *key, *parent;
const void *kbase;
bool is_left;
erbtree_check(tree);
g_assert(node != NULL);
kbase = const_ptr_add_offset(node, -tree->offset);
if (erbtree_is_extended(tree)) {
key = do_lookup_ext(ERBTREE_E(tree), kbase, &parent, &is_left);
} else {
key = do_lookup(tree, kbase, &parent, &is_left);
}
if (key != NULL)
return ptr_add_offset(key, -tree->offset);
g_assert(!is_valid(node)); /* Not yet part of the tree */
node->left = NULL;
node->right = NULL;
set_color(node, RB_RED);
set_parent(node, parent);
tree->count++;
if (parent != NULL) {
if (is_left) {
if (parent == tree->first)
tree->first = node;
} else {
if (parent == tree->last)
tree->last = node;
}
set_child(parent, node, is_left);
} else {
tree->root = node;
tree->first = node;
tree->last = node;
}
/*
* Fixup the modified tree by recoloring nodes and performing
* rotations (2 at most) hence the red-black tree properties are
* preserved.
*/
while (NULL != (parent = get_parent(node)) && is_red(parent)) {
rbnode_t *grandpa = get_parent(parent);
if (parent == grandpa->left) {
rbnode_t *uncle = grandpa->right;
if (uncle != NULL && is_red(uncle)) {
set_color(parent, RB_BLACK);
set_color(uncle, RB_BLACK);
set_color(grandpa, RB_RED);
node = grandpa;
} else {
if (node == parent->right) {
rotate_left(tree, parent);
node = parent;
parent = get_parent(node);
}
set_color(parent, RB_BLACK);
set_color(grandpa, RB_RED);
rotate_right(tree, grandpa);
}
} else {
rbnode_t *uncle = grandpa->left;
if (uncle != NULL && is_red(uncle)) {
set_color(parent, RB_BLACK);
set_color(uncle, RB_BLACK);
set_color(grandpa, RB_RED);
node = grandpa;
} else {
if (node == parent->left) {
rotate_right(tree, parent);
node = parent;
parent = get_parent(node);
}
set_color(parent, RB_BLACK);
set_color(grandpa, RB_RED);
rotate_left(tree, grandpa);
}
}
}
set_color(tree->root, RB_BLACK);
return NULL;
}
