ham_status_t
blob_duplicate_get(ham_env_t *env, ham_offset_t table_id,
ham_size_t position, dupe_entry_t *entry)
{
ham_status_t st;
dupe_table_t *table;
ham_page_t *page=0;
st = __get_duplicate_table(&table, &page, env, table_id);
ham_assert(st ? table == NULL : 1, (0));
ham_assert(st ? page == NULL : 1, (0));
if (!table)
return st ? st : HAM_INTERNAL_ERROR;
if (position>=dupe_table_get_count(table))
{
if (!(env_get_rt_flags(env)&HAM_IN_MEMORY_DB))
if (!page)
allocator_free(env_get_allocator(env), table);
return HAM_KEY_NOT_FOUND;
}
memcpy(entry, dupe_table_get_entry(table, position), sizeof(*entry));
if (!(env_get_rt_flags(env)&HAM_IN_MEMORY_DB))
{
if (!page)
allocator_free(env_get_allocator(env), table);
}
return (0);
}
ham_status_t
blob_get_datasize(ham_db_t *db, ham_offset_t blobid, ham_offset_t *size)
{
ham_status_t st;
ham_page_t *page;
blob_t hdr;
/*
* in-memory-database: the blobid is actually a pointer to the memory
* buffer, in which the blob is stored
*/
if (env_get_rt_flags(db_get_env(db))&HAM_IN_MEMORY_DB) {
blob_t *hdr=(blob_t *)U64_TO_PTR(blobid);
*size=blob_get_size(hdr);
return (0);
}
ham_assert(blobid%DB_CHUNKSIZE==0, ("blobid is %llu", blobid));
/* read the blob header */
st=__read_chunk(db_get_env(db), 0, &page, blobid,
(ham_u8_t *)&hdr, sizeof(hdr));
if (st)
return (st);
ham_assert(blob_get_alloc_size(&hdr)%DB_CHUNKSIZE==0, (0));
if (blob_get_self(&hdr)!=blobid)
return (HAM_BLOB_NOT_FOUND);
*size=blob_get_size(&hdr);
return (0);
}
static ham_status_t
__get_duplicate_table(dupe_table_t **table_ref, ham_page_t **page, ham_env_t *env, ham_u64_t table_id)
{
ham_status_t st;
blob_t hdr;
ham_page_t *hdrpage=0;
dupe_table_t *table;
*page = 0;
if (env_get_rt_flags(env)&HAM_IN_MEMORY_DB) {
ham_u8_t *p=(ham_u8_t *)U64_TO_PTR(table_id);
*table_ref = (dupe_table_t *)(p+sizeof(hdr));
return HAM_SUCCESS;
}
*table_ref = 0;
/*
* load the blob header
*/
st=__read_chunk(env, 0, &hdrpage, table_id, (ham_u8_t *)&hdr, sizeof(hdr));
if (st) {
return st;
}
/*
* if the whole table is in a page (and not split between several
* pages), just return a pointer directly in the page
*/
if (page_get_self(hdrpage)+env_get_usable_pagesize(env) >=
table_id+blob_get_size(&hdr))
{
ham_u8_t *p=page_get_raw_payload(hdrpage);
/* yes, table is in the page */
*page=hdrpage;
*table_ref = (dupe_table_t *)
&p[table_id-page_get_self(hdrpage)+sizeof(hdr)];
return HAM_SUCCESS;
}
/*
* otherwise allocate memory for the table
*/
table=allocator_alloc(env_get_allocator(env), (ham_size_t)blob_get_size(&hdr));
if (!table) {
return HAM_OUT_OF_MEMORY;
}
/*
* then read the rest of the blob
*/
st=__read_chunk(env, hdrpage, 0, table_id+sizeof(hdr),
(ham_u8_t *)table, (ham_size_t)blob_get_size(&hdr));
if (st) {
return st;
}
*table_ref = table;
return HAM_SUCCESS;
}
static ham_status_t
_remote_fun_open(ham_env_t *env, const char *filename, ham_u32_t flags,
const ham_parameter_t *param)
{
ham_status_t st;
proto_wrapper_t *request, *reply;
CURL *handle=curl_easy_init();
request=proto_init_connect_request(filename);
st=_perform_request(env, handle, request, &reply);
proto_delete(request);
if (st) {
curl_easy_cleanup(handle);
if (reply)
proto_delete(reply);
return (st);
}
ham_assert(reply!=0, (""));
ham_assert(proto_has_connect_reply(reply), (""));
st=proto_connect_reply_get_status(reply);
if (st==0) {
env_set_curl(env, handle);
env_set_rt_flags(env,
env_get_rt_flags(env)
|proto_connect_reply_get_env_flags(reply));
}
proto_delete(reply);
return (st);
}
ham_status_t
env_initialize_remote(ham_env_t *env)
{
#if HAM_ENABLE_REMOTE
env->_fun_create =_remote_fun_create;
env->_fun_open =_remote_fun_open;
env->_fun_rename_db =_remote_fun_rename_db;
env->_fun_erase_db =_remote_fun_erase_db;
env->_fun_get_database_names =_remote_fun_get_database_names;
env->_fun_get_parameters =_remote_fun_env_get_parameters;
env->_fun_flush =_remote_fun_env_flush;
env->_fun_create_db =_remote_fun_create_db;
env->_fun_open_db =_remote_fun_open_db;
env->_fun_close =_remote_fun_env_close;
env->_fun_txn_begin =_remote_fun_txn_begin;
env->_fun_txn_commit =_remote_fun_txn_commit;
env->_fun_txn_abort =_remote_fun_txn_abort;
env_set_rt_flags(env, env_get_rt_flags(env)|DB_IS_REMOTE);
#else
return (HAM_NOT_IMPLEMENTED);
#endif
return (0);
}
void
extkey_cache_destroy(extkey_cache_t *cache)
{
ham_size_t i;
extkey_t *e, *n;
ham_db_t *db=extkey_cache_get_db(cache);
ham_env_t *env = db_get_env(db);
/*
* make sure that all entries are empty
*/
for (i=0; i<extkey_cache_get_bucketsize(cache); i++) {
e=extkey_cache_get_bucket(cache, i);
while (e) {
#if HAM_DEBUG
/*
* make sure that the extkey-cache is empty - only for in-memory
* databases and DEBUG builds.
*/
if (env_get_rt_flags(env)&HAM_IN_MEMORY_DB)
ham_assert(!"extkey-cache is not empty!", (0));
#endif
n=extkey_get_next(e);
allocator_free(env_get_allocator(env), e);
e=n;
}
}
allocator_free(env_get_allocator(env), cache);
}
/**
* Remove all extended keys for the given @a page from the
* extended key cache.
*/
static ham_status_t
my_fun_free_page_extkeys(ham_btree_t *be, ham_page_t *page, ham_u32_t flags)
{
ham_db_t *db=be_get_db(be);
ham_assert(page_get_owner(page) == db, (0));
ham_assert(0 == (flags & ~DB_MOVE_TO_FREELIST), (0));
/*
* if this page has a header, and it's either a B-Tree root page or
* a B-Tree index page: remove all extended keys from the cache,
* and/or free their blobs
*/
if (page_get_pers(page)
&& (!(page_get_npers_flags(page)&PAGE_NPERS_NO_HEADER))
&& (page_get_type(page)==PAGE_TYPE_B_ROOT
|| page_get_type(page)==PAGE_TYPE_B_INDEX))
{
ham_size_t i;
ham_offset_t blobid;
int_key_t *bte;
btree_node_t *node=ham_page_get_btree_node(page);
extkey_cache_t *c;
ham_assert(db, ("Must be set as page owner when this is a Btree page"));
ham_assert(db=page_get_owner(page), (""));
c=db_get_extkey_cache(db);
for (i=0; i<btree_node_get_count(node); i++)
{
bte=btree_node_get_key(db, node, i);
if (key_get_flags(bte)&KEY_IS_EXTENDED)
{
blobid=key_get_extended_rid(db, bte);
if (env_get_rt_flags(db_get_env(db))&HAM_IN_MEMORY_DB)
{
/* delete the blobid to prevent that it's freed twice */
*(ham_offset_t *)(key_get_key(bte)+
(db_get_keysize(db)-sizeof(ham_offset_t)))=0;
}
//(void)key_erase_record(db, bte, 0, BLOB_FREE_ALL_DUPES);
if (c)
(void)extkey_cache_remove(c, blobid);
}
}
}
return (HAM_SUCCESS);
}
ham_status_t
blob_free(ham_env_t *env, ham_db_t *db, ham_offset_t blobid, ham_u32_t flags)
{
ham_status_t st;
blob_t hdr;
/*
* in-memory-database: the blobid is actually a pointer to the memory
* buffer, in which the blob is stored
*/
if (env_get_rt_flags(env)&HAM_IN_MEMORY_DB) {
allocator_free(env_get_allocator(env), (void *)U64_TO_PTR(blobid));
return (0);
}
ham_assert(blobid%DB_CHUNKSIZE==0, (0));
/*
* fetch the blob header
*/
st=__read_chunk(env, 0, 0, blobid, (ham_u8_t *)&hdr, sizeof(hdr));
if (st)
return (st);
ham_assert(blob_get_alloc_size(&hdr)%DB_CHUNKSIZE==0, (0));
/*
* sanity check
*/
ham_assert(blob_get_self(&hdr)==blobid,
("invalid blobid %llu != %llu", blob_get_self(&hdr), blobid));
if (blob_get_self(&hdr)!=blobid)
return (HAM_BLOB_NOT_FOUND);
/*
* move the blob to the freelist
*/
st = freel_mark_free(env, db, blobid,
(ham_size_t)blob_get_alloc_size(&hdr), HAM_FALSE);
ham_assert(!st, ("unexpected error, at least not covered in the old code"));
return st;
}
void
db_update_global_stats_erase_query(ham_db_t *db, ham_size_t key_size)
{
ham_env_t *env = db_get_env(db);
if (!(env_get_rt_flags(env)&HAM_IN_MEMORY_DB))
{
ham_runtime_statistics_globdata_t *globalstats = env_get_global_perf_data(env);
ham_runtime_statistics_opdbdata_t *opstats = db_get_op_perf_data(db, HAM_OPERATION_STATS_ERASE);
#ifdef HAM_DEBUG
ham_u16_t bucket = ham_bitcount2bucket_index(key_size / DB_CHUNKSIZE);
ham_assert(bucket < HAM_FREELIST_SLOT_SPREAD, (0));
//ham_assert(device_get_freelist_cache(dev), (0));
#endif
globalstats->erase_query_count++;
opstats->query_count++;
}
}
ham_status_t
blob_overwrite(ham_env_t *env, ham_db_t *db, ham_offset_t old_blobid,
ham_record_t *record, ham_u32_t flags, ham_offset_t *new_blobid)
{
ham_status_t st;
ham_size_t alloc_size;
blob_t old_hdr;
blob_t new_hdr;
ham_page_t *page;
/*
* PARTIAL WRITE
*
* if offset+partial_size equals the full record size, then we won't
* have any gaps. In this case we just write the full record and ignore
* the partial parameters.
*/
if (flags&HAM_PARTIAL) {
if (record->partial_offset==0
&& record->partial_offset+record->partial_size==record->size)
flags&=~HAM_PARTIAL;
}
/*
* inmemory-databases: free the old blob,
* allocate a new blob (but if both sizes are equal, just overwrite
* the data)
*/
if (env_get_rt_flags(env)&HAM_IN_MEMORY_DB)
{
blob_t *nhdr, *phdr=(blob_t *)U64_TO_PTR(old_blobid);
if (blob_get_size(phdr)==record->size) {
ham_u8_t *p=(ham_u8_t *)phdr;
if (flags&HAM_PARTIAL) {
memmove(p+sizeof(blob_t)+record->partial_offset,
record->data, record->partial_size);
}
else {
memmove(p+sizeof(blob_t), record->data, record->size);
}
*new_blobid=(ham_offset_t)PTR_TO_U64(phdr);
}
else {
st=blob_allocate(env, db, record, flags, new_blobid);
if (st)
return (st);
nhdr=(blob_t *)U64_TO_PTR(*new_blobid);
blob_set_flags(nhdr, blob_get_flags(phdr));
allocator_free(env_get_allocator(env), phdr);
}
return (HAM_SUCCESS);
}
ham_assert(old_blobid%DB_CHUNKSIZE==0, (0));
/*
* blobs are CHUNKSIZE-allocated
*/
alloc_size=sizeof(blob_t)+record->size;
alloc_size += DB_CHUNKSIZE - 1;
alloc_size -= alloc_size % DB_CHUNKSIZE;
/*
* first, read the blob header; if the new blob fits into the
* old blob, we overwrite the old blob (and add the remaining
* space to the freelist, if there is any)
*/
st=__read_chunk(env, 0, &page, old_blobid, (ham_u8_t *)&old_hdr,
sizeof(old_hdr));
if (st)
return (st);
ham_assert(blob_get_alloc_size(&old_hdr)%DB_CHUNKSIZE==0, (0));
/*
* sanity check
*/
ham_verify(blob_get_self(&old_hdr)==old_blobid,
("invalid blobid %llu != %llu", blob_get_self(&old_hdr),
old_blobid));
if (blob_get_self(&old_hdr)!=old_blobid)
return (HAM_BLOB_NOT_FOUND);
/*
* now compare the sizes; does the new data fit in the old allocated
* space?
*/
if (alloc_size<=blob_get_alloc_size(&old_hdr))
{
ham_u8_t *chunk_data[2];
ham_size_t chunk_size[2];
/*
* setup the new blob header
*/
blob_set_self(&new_hdr, blob_get_self(&old_hdr));
blob_set_size(&new_hdr, record->size);
blob_set_flags(&new_hdr, blob_get_flags(&old_hdr));
if (blob_get_alloc_size(&old_hdr)-alloc_size>SMALLEST_CHUNK_SIZE)
blob_set_alloc_size(&new_hdr, alloc_size);
else
blob_set_alloc_size(&new_hdr, blob_get_alloc_size(&old_hdr));
/*
* PARTIAL WRITE
*
* if we have a gap at the beginning, then we have to write the
* blob header and the blob data in two steps; otherwise we can
* write both immediately
*/
if ((flags&HAM_PARTIAL) && (record->partial_offset)) {
chunk_data[0]=(ham_u8_t *)&new_hdr;
chunk_size[0]=sizeof(new_hdr);
st=__write_chunks(env, page, blob_get_self(&new_hdr), HAM_FALSE,
HAM_FALSE, chunk_data, chunk_size, 1);
if (st)
return (st);
chunk_data[0]=record->data;
chunk_size[0]=record->partial_size;
st=__write_chunks(env, page,
blob_get_self(&new_hdr)+sizeof(new_hdr)
+record->partial_offset,
HAM_FALSE, HAM_FALSE, chunk_data, chunk_size, 1);
if (st)
return (st);
}
else {
chunk_data[0]=(ham_u8_t *)&new_hdr;
chunk_size[0]=sizeof(new_hdr);
chunk_data[1]=record->data;
chunk_size[1]=(flags&HAM_PARTIAL)
? record->partial_size
: record->size;
st=__write_chunks(env, page, blob_get_self(&new_hdr), HAM_FALSE,
HAM_FALSE, chunk_data, chunk_size, 2);
if (st)
return (st);
}
/*
* move remaining data to the freelist
*/
if (blob_get_alloc_size(&old_hdr)!=blob_get_alloc_size(&new_hdr)) {
(void)freel_mark_free(env, db,
blob_get_self(&new_hdr)+blob_get_alloc_size(&new_hdr),
(ham_size_t)(blob_get_alloc_size(&old_hdr)-
blob_get_alloc_size(&new_hdr)), HAM_FALSE);
}
/*
* the old rid is the new rid
*/
*new_blobid=blob_get_self(&new_hdr);
return (HAM_SUCCESS);
}
else {
/*
* when the new data is larger, allocate a fresh space for it
* and discard the old;
'overwrite' has become (delete + insert) now.
*/
st=blob_allocate(env, db, record, flags, new_blobid);
if (st)
return (st);
(void)freel_mark_free(env, db, old_blobid,
(ham_size_t)blob_get_alloc_size(&old_hdr), HAM_FALSE);
}
return (HAM_SUCCESS);
}
ham_status_t
blob_read(ham_db_t *db, ham_offset_t blobid,
ham_record_t *record, ham_u32_t flags)
{
ham_status_t st;
ham_page_t *page;
blob_t hdr;
ham_size_t blobsize=0;
/*
* in-memory-database: the blobid is actually a pointer to the memory
* buffer, in which the blob is stored
*/
if (env_get_rt_flags(db_get_env(db))&HAM_IN_MEMORY_DB) {
blob_t *hdr=(blob_t *)U64_TO_PTR(blobid);
ham_u8_t *data=(ham_u8_t *)(U64_TO_PTR(blobid))+sizeof(blob_t);
/* when the database is closing, the header is already deleted */
if (!hdr) {
record->size = 0;
return (0);
}
blobsize = (ham_size_t)blob_get_size(hdr);
if (flags&HAM_PARTIAL) {
if (record->partial_offset>blobsize) {
ham_trace(("partial offset is greater than the total "
"record size"));
return (HAM_INV_PARAMETER);
}
if (record->partial_offset+record->partial_size>blobsize)
blobsize=blobsize-record->partial_offset;
else
blobsize=record->partial_size;
}
if (!blobsize) {
/* empty blob? */
record->data = 0;
record->size = 0;
}
else {
ham_u8_t *d=data;
if (flags&HAM_PARTIAL)
d+=record->partial_offset;
if ((flags&HAM_DIRECT_ACCESS)
&& !(record->flags&HAM_RECORD_USER_ALLOC)) {
record->size=blobsize;
record->data=d;
}
else {
/* resize buffer, if necessary */
if (!(record->flags & HAM_RECORD_USER_ALLOC)) {
st=db_resize_record_allocdata(db, blobsize);
if (st)
return (st);
record->data = db_get_record_allocdata(db);
}
/* and copy the data */
memcpy(record->data, d, blobsize);
record->size = blobsize;
}
}
return (0);
}
ham_assert(blobid%DB_CHUNKSIZE==0, ("blobid is %llu", blobid));
/*
* first step: read the blob header
*/
st=__read_chunk(db_get_env(db), 0, &page, blobid,
(ham_u8_t *)&hdr, sizeof(hdr));
if (st)
return (st);
ham_assert(blob_get_alloc_size(&hdr)%DB_CHUNKSIZE==0, (0));
/*
* sanity check
*/
if (blob_get_self(&hdr)!=blobid)
return (HAM_BLOB_NOT_FOUND);
blobsize = (ham_size_t)blob_get_size(&hdr);
if (flags&HAM_PARTIAL) {
if (record->partial_offset>blobsize) {
ham_trace(("partial offset+size is greater than the total "
"record size"));
return (HAM_INV_PARAMETER);
}
if (record->partial_offset+record->partial_size>blobsize)
blobsize=blobsize-record->partial_offset;
else
blobsize=record->partial_size;
}
/*
* empty blob?
*/
if (!blobsize) {
record->data = 0;
record->size = 0;
return (0);
}
/*
* second step: resize the blob buffer
*/
if (!(record->flags & HAM_RECORD_USER_ALLOC)) {
st=db_resize_record_allocdata(db, blobsize);
if (st)
return (st);
record->data = db_get_record_allocdata(db);
}
/*
* third step: read the blob data
*/
st=__read_chunk(db_get_env(db), page, 0,
blobid+sizeof(blob_t)+(flags&HAM_PARTIAL
? record->partial_offset
: 0),
record->data, blobsize);
if (st)
return (st);
record->size = blobsize;
return (0);
}
/**
* Allocate space in storage for and write the content references by 'data'
* (and length 'size') to storage.
*
* Conditions will apply whether the data is written through cache or direct
* to device.
*
* The content is, of course, prefixed by a BLOB header.
*
* Partial writes are handled in this function.
*/
ham_status_t
blob_allocate(ham_env_t *env, ham_db_t *db, ham_record_t *record,
ham_u32_t flags, ham_offset_t *blobid)
{
ham_status_t st;
ham_page_t *page=0;
ham_offset_t addr;
blob_t hdr;
ham_u8_t *chunk_data[2];
ham_size_t alloc_size;
ham_size_t chunk_size[2];
ham_device_t *device=env_get_device(env);
ham_bool_t freshly_created = HAM_FALSE;
*blobid=0;
/*
* PARTIAL WRITE
*
* if offset+partial_size equals the full record size, then we won't
* have any gaps. In this case we just write the full record and ignore
* the partial parameters.
*/
if (flags&HAM_PARTIAL) {
if (record->partial_offset==0
&& record->partial_offset+record->partial_size==record->size)
flags&=~HAM_PARTIAL;
}
/*
* in-memory-database: the blobid is actually a pointer to the memory
* buffer, in which the blob (with the blob-header) is stored
*/
if (env_get_rt_flags(env)&HAM_IN_MEMORY_DB) {
blob_t *hdr;
ham_u8_t *p=(ham_u8_t *)allocator_alloc(env_get_allocator(env),
record->size+sizeof(blob_t));
if (!p) {
return HAM_OUT_OF_MEMORY;
}
/* initialize the header */
hdr=(blob_t *)p;
memset(hdr, 0, sizeof(*hdr));
blob_set_self(hdr, (ham_offset_t)PTR_TO_U64(p));
blob_set_alloc_size(hdr, record->size+sizeof(blob_t));
blob_set_size(hdr, record->size);
/* do we have gaps? if yes, fill them with zeroes */
if (flags&HAM_PARTIAL) {
ham_u8_t *s=p+sizeof(blob_t);
if (record->partial_offset)
memset(s, 0, record->partial_offset);
memcpy(s+record->partial_offset,
record->data, record->partial_size);
if (record->partial_offset+record->partial_size<record->size)
memset(s+record->partial_offset+record->partial_size, 0,
record->size-(record->partial_offset+record->partial_size));
}
else {
memcpy(p+sizeof(blob_t), record->data, record->size);
}
*blobid=(ham_offset_t)PTR_TO_U64(p);
return (0);
}
memset(&hdr, 0, sizeof(hdr));
/*
* blobs are CHUNKSIZE-allocated
*/
alloc_size=sizeof(blob_t)+record->size;
alloc_size += DB_CHUNKSIZE - 1;
alloc_size -= alloc_size % DB_CHUNKSIZE;
/*
* check if we have space in the freelist
*/
st = freel_alloc_area(&addr, env, db, alloc_size);
if (!addr)
{
if (st)
return st;
/*
* if the blob is small AND if logging is disabled: load the page
* through the cache
*/
if (__blob_from_cache(env, alloc_size)) {
st = db_alloc_page(&page, db, PAGE_TYPE_BLOB,
PAGE_IGNORE_FREELIST);
ham_assert(st ? page == NULL : 1, (0));
ham_assert(!st ? page != NULL : 1, (0));
if (st)
return st;
/* blob pages don't have a page header */
page_set_npers_flags(page,
page_get_npers_flags(page)|PAGE_NPERS_NO_HEADER);
addr=page_get_self(page);
/* move the remaining space to the freelist */
(void)freel_mark_free(env, db, addr+alloc_size,
env_get_pagesize(env)-alloc_size, HAM_FALSE);
blob_set_alloc_size(&hdr, alloc_size);
}
else {
/*
* otherwise use direct IO to allocate the space
*/
ham_size_t aligned=alloc_size;
aligned += env_get_pagesize(env) - 1;
aligned -= aligned % env_get_pagesize(env);
st=device->alloc(device, aligned, &addr);
if (st)
return (st);
/* if aligned!=size, and the remaining chunk is large enough:
* move it to the freelist */
{
ham_size_t diff=aligned-alloc_size;
if (diff > SMALLEST_CHUNK_SIZE) {
(void)freel_mark_free(env, db, addr+alloc_size,
diff, HAM_FALSE);
blob_set_alloc_size(&hdr, aligned-diff);
}
else {
blob_set_alloc_size(&hdr, aligned);
}
}
freshly_created = HAM_TRUE;
}
ham_assert(HAM_SUCCESS == freel_check_area_is_allocated(env, db,
addr, alloc_size), (0));
}
else {
ham_assert(!st, (0));
blob_set_alloc_size(&hdr, alloc_size);
}
blob_set_size(&hdr, record->size);
blob_set_self(&hdr, addr);
/*
* PARTIAL WRITE
*
* are there gaps at the beginning? If yes, then we'll fill with zeros
*/
if ((flags&HAM_PARTIAL) && (record->partial_offset)) {
ham_u8_t *ptr;
ham_size_t gapsize=record->partial_offset;
ptr=allocator_calloc(env_get_allocator(env),
gapsize > env_get_pagesize(env)
? env_get_pagesize(env)
: gapsize);
if (!ptr)
return (HAM_OUT_OF_MEMORY);
/*
* first: write the header
*/
chunk_data[0]=(ham_u8_t *)&hdr;
chunk_size[0]=sizeof(hdr);
st=__write_chunks(env, page, addr, HAM_TRUE, freshly_created,
chunk_data, chunk_size, 1);
if (st)
return (st);
addr+=sizeof(hdr);
/* now fill the gap; if the gap is bigger than a pagesize we'll
* split the gap into smaller chunks
*/
while (gapsize>=env_get_pagesize(env)) {
chunk_data[0]=ptr;
chunk_size[0]=env_get_pagesize(env);
st=__write_chunks(env, page, addr, HAM_TRUE,
freshly_created, chunk_data, chunk_size, 1);
if (st)
break;
gapsize-=env_get_pagesize(env);
addr+=env_get_pagesize(env);
}
/* fill the remaining gap */
if (gapsize) {
chunk_data[0]=ptr;
chunk_size[0]=gapsize;
st=__write_chunks(env, page, addr, HAM_TRUE, freshly_created,
chunk_data, chunk_size, 1);
if (st)
return (st);
addr+=gapsize;
}
allocator_free(env_get_allocator(env), ptr);
/* now write the "real" data */
chunk_data[0]=(ham_u8_t *)record->data;
chunk_size[0]=record->partial_size;
st=__write_chunks(env, page, addr, HAM_TRUE, freshly_created,
chunk_data, chunk_size, 1);
if (st)
return (st);
addr+=record->partial_size;
}
else {
/*
* not writing partially: write header and data, then we're done
*/
chunk_data[0]=(ham_u8_t *)&hdr;
chunk_size[0]=sizeof(hdr);
chunk_data[1]=(ham_u8_t *)record->data;
chunk_size[1]=(flags&HAM_PARTIAL)
? record->partial_size
: record->size;
st=__write_chunks(env, page, addr, HAM_TRUE, freshly_created,
chunk_data, chunk_size, 2);
if (st)
return (st);
addr+=sizeof(hdr)+
((flags&HAM_PARTIAL) ? record->partial_size : record->size);
}
/*
* store the blobid; it will be returned to the caller
*/
*blobid=blob_get_self(&hdr);
/*
* PARTIAL WRITES:
*
* if we have gaps at the end of the blob: just append more chunks to
* fill these gaps. Since they can be pretty large we split them into
* smaller chunks if necessary.
*/
if (flags&HAM_PARTIAL) {
if (record->partial_offset+record->partial_size < record->size) {
ham_u8_t *ptr;
ham_size_t gapsize=record->size
- (record->partial_offset+record->partial_size);
/* now fill the gap; if the gap is bigger than a pagesize we'll
* split the gap into smaller chunks
*
* we split this loop in two - the outer loop will allocate the
* memory buffer, thus saving some allocations
*/
while (gapsize>env_get_pagesize(env)) {
ham_u8_t *ptr=allocator_calloc(env_get_allocator(env),
env_get_pagesize(env));
if (!ptr)
return (HAM_OUT_OF_MEMORY);
while (gapsize>env_get_pagesize(env)) {
chunk_data[0]=ptr;
chunk_size[0]=env_get_pagesize(env);
st=__write_chunks(env, page, addr, HAM_TRUE,
freshly_created, chunk_data, chunk_size, 1);
if (st)
break;
gapsize-=env_get_pagesize(env);
addr+=env_get_pagesize(env);
}
allocator_free(env_get_allocator(env), ptr);
if (st)
return (st);
}
/* now write the remainder, which is less than a pagesize */
ham_assert(gapsize<env_get_pagesize(env), (""));
chunk_size[0]=gapsize;
ptr=chunk_data[0]=allocator_calloc(env_get_allocator(env), gapsize);
if (!ptr)
return (HAM_OUT_OF_MEMORY);
st=__write_chunks(env, page, addr, HAM_TRUE, freshly_created,
chunk_data, chunk_size, 1);
allocator_free(env_get_allocator(env), ptr);
if (st)
return (st);
}
}
return (0);
}
ham_status_t
blob_duplicate_insert(ham_db_t *db, ham_offset_t table_id,
ham_record_t *record, ham_size_t position, ham_u32_t flags,
dupe_entry_t *entries, ham_size_t num_entries,
ham_offset_t *rid, ham_size_t *new_position)
{
ham_status_t st=0;
dupe_table_t *table=0;
ham_bool_t alloc_table=0;
ham_bool_t resize=0;
ham_page_t *page=0;
ham_env_t *env=db_get_env(db);
/*
* create a new duplicate table if none existed, and insert
* the first entry
*/
if (!table_id) {
ham_assert(num_entries==2, (""));
/* allocates space for 8 (!) entries */
table=allocator_calloc(env_get_allocator(env),
sizeof(dupe_table_t)+7*sizeof(dupe_entry_t));
if (!table)
return HAM_OUT_OF_MEMORY;
dupe_table_set_capacity(table, 8);
dupe_table_set_count(table, 1);
memcpy(dupe_table_get_entry(table, 0), &entries[0],
sizeof(entries[0]));
/* skip the first entry */
entries++;
num_entries--;
alloc_table=1;
}
else {
/*
* otherwise load the existing table
*/
st=__get_duplicate_table(&table, &page, env, table_id);
ham_assert(st ? table == NULL : 1, (0));
ham_assert(st ? page == NULL : 1, (0));
if (!table)
return st ? st : HAM_INTERNAL_ERROR;
if (!page && !(env_get_rt_flags(env)&HAM_IN_MEMORY_DB))
alloc_table=1;
}
if (page)
if ((st=ham_log_add_page_before(page)))
return (st);
ham_assert(num_entries==1, (""));
/*
* resize the table, if necessary
*/
if (!(flags & HAM_OVERWRITE)
&& dupe_table_get_count(table)+1>=dupe_table_get_capacity(table)) {
dupe_table_t *old=table;
ham_size_t new_cap=dupe_table_get_capacity(table);
if (new_cap < 3*8)
new_cap += 8;
else
new_cap += new_cap/3;
table=allocator_calloc(env_get_allocator(env), sizeof(dupe_table_t)+
(new_cap-1)*sizeof(dupe_entry_t));
if (!table)
return (HAM_OUT_OF_MEMORY);
dupe_table_set_capacity(table, new_cap);
dupe_table_set_count(table, dupe_table_get_count(old));
memcpy(dupe_table_get_entry(table, 0), dupe_table_get_entry(old, 0),
dupe_table_get_count(old)*sizeof(dupe_entry_t));
if (alloc_table)
allocator_free(env_get_allocator(env), old);
alloc_table=1;
resize=1;
}
/*
* insert sorted, unsorted or overwrite the entry at the requested position
*/
if (flags&HAM_OVERWRITE) {
dupe_entry_t *e=dupe_table_get_entry(table, position);
if (!(dupe_entry_get_flags(e)&(KEY_BLOB_SIZE_SMALL
|KEY_BLOB_SIZE_TINY
|KEY_BLOB_SIZE_EMPTY))) {
(void)blob_free(env, db, dupe_entry_get_rid(e), 0);
}
memcpy(dupe_table_get_entry(table, position),
&entries[0], sizeof(entries[0]));
}
else {
if (db_get_rt_flags(db)&HAM_SORT_DUPLICATES) {
if (page)
page_add_ref(page);
position=__get_sorted_position(db, table, record, flags);
if (page)
page_release_ref(page);
if (position<0)
return ((ham_status_t)position);
}
else if (flags&HAM_DUPLICATE_INSERT_BEFORE) {
/* do nothing, insert at the current position */
}
else if (flags&HAM_DUPLICATE_INSERT_AFTER) {
position++;
if (position > dupe_table_get_count(table))
position=dupe_table_get_count(table);
}
else if (flags&HAM_DUPLICATE_INSERT_FIRST) {
position=0;
}
else if (flags&HAM_DUPLICATE_INSERT_LAST) {
position=dupe_table_get_count(table);
}
else {
position=dupe_table_get_count(table);
}
if (position != dupe_table_get_count(table)) {
memmove(dupe_table_get_entry(table, position+1),
dupe_table_get_entry(table, position),
sizeof(entries[0])*(dupe_table_get_count(table)-position));
}
memcpy(dupe_table_get_entry(table, position),
&entries[0], sizeof(entries[0]));
dupe_table_set_count(table, dupe_table_get_count(table)+1);
}
/*
* write the table back to disk and return the blobid of the table
*/
if ((table_id && !page) || resize) {
ham_record_t rec={0};
rec.data=(ham_u8_t *)table;
rec.size=sizeof(dupe_table_t)
+(dupe_table_get_capacity(table)-1)*sizeof(dupe_entry_t);
st=blob_overwrite(env, db, table_id, &rec, 0, rid);
}
else if (!table_id) {
ham_record_t rec={0};
rec.data=(ham_u8_t *)table;
rec.size=sizeof(dupe_table_t)
+(dupe_table_get_capacity(table)-1)*sizeof(dupe_entry_t);
st=blob_allocate(env, db, &rec, 0, rid);
}
else if (table_id && page) {
page_set_dirty(page, env);
}
else {
ham_assert(!"shouldn't be here", (0));
}
if (alloc_table)
allocator_free(env_get_allocator(env), table);
if (new_position)
*new_position=position;
return (st);
}
static ham_status_t
__insert_cursor(ham_btree_t *be, ham_key_t *key, ham_record_t *record,
ham_bt_cursor_t *cursor, insert_hints_t *hints)
{
ham_status_t st;
ham_page_t *root;
ham_db_t *db=be_get_db(be);
ham_env_t *env = db_get_env(db);
insert_scratchpad_t scratchpad;
ham_assert(hints->force_append == HAM_FALSE, (0));
ham_assert(hints->force_prepend == HAM_FALSE, (0));
/*
* initialize the scratchpad
*/
memset(&scratchpad, 0, sizeof(scratchpad));
scratchpad.be=be;
scratchpad.record=record;
scratchpad.cursor=cursor;
/*
* get the root-page...
*/
ham_assert(btree_get_rootpage(be)!=0, ("btree has no root page"));
st=db_fetch_page(&root, db, btree_get_rootpage(be), 0);
ham_assert(st ? root == NULL : 1, (0));
if (st)
return st;
/*
* ... and start the recursion
*/
st=__insert_recursive(root, key, 0, &scratchpad, hints);
/*
* if the root page was split, we have to create a new
* root page.
*/
if (st==SPLIT) {
ham_page_t *newroot;
btree_node_t *node;
/*
* the root-page will be changed...
*/
st=ham_log_add_page_before(root);
if (st)
return (st);
/*
* allocate a new root page
*/
st=db_alloc_page(&newroot, db, PAGE_TYPE_B_ROOT, 0);
ham_assert(st ? newroot == NULL : 1, (0));
if (st)
return (st);
ham_assert(page_get_owner(newroot), (""));
/* clear the node header */
memset(page_get_payload(newroot), 0, sizeof(btree_node_t));
stats_page_is_nuked(db, root, HAM_TRUE);
/*
* insert the pivot element and the ptr_left
*/
node=ham_page_get_btree_node(newroot);
btree_node_set_ptr_left(node, btree_get_rootpage(be));
st=__insert_nosplit(newroot, &scratchpad.key,
scratchpad.rid, scratchpad.record, scratchpad.cursor,
hints);
ham_assert(!(scratchpad.key.flags & HAM_KEY_USER_ALLOC), (0));
scratchpad.cursor=0; /* don't overwrite cursor if __insert_nosplit
is called again */
if (st) {
ham_assert(!(scratchpad.key.flags & HAM_KEY_USER_ALLOC), (0));
if (scratchpad.key.data)
allocator_free(env_get_allocator(env), scratchpad.key.data);
return (st);
}
/*
* set the new root page
*
* !!
* do NOT delete the old root page - it's still in use!
*
* also don't forget to flush the backend - otherwise the header
* page of the database will not contain the updated information.
* The backend is flushed when the database is closed, but if
* recovery is enabled then the flush here is critical.
*/
btree_set_rootpage(be, page_get_self(newroot));
be_set_dirty(be, HAM_TRUE);
be->_fun_flush(be);
/*
* As we re-purpose a page, we will reset its pagecounter
* as well to signal its first use as the new type assigned
* here.
*/
if (env_get_cache(env) && (page_get_type(root)!=PAGE_TYPE_B_INDEX))
cache_update_page_access_counter(root, env_get_cache(env), 0);
page_set_type(root, PAGE_TYPE_B_INDEX);
page_set_dirty(root, env);
page_set_dirty(newroot, env);
/* the root page was modified (btree_set_rootpage) - make sure that
* it's logged */
if (env_get_rt_flags(env)&HAM_ENABLE_RECOVERY) {
st=txn_add_page(env_get_txn(env), env_get_header_page(env),
HAM_TRUE);
if (st)
return (st);
}
}
/*
* release the scratchpad-memory and return to caller
*/
ham_assert(!(scratchpad.key.flags & HAM_KEY_USER_ALLOC), (0));
if (scratchpad.key.data)
allocator_free(env_get_allocator(env), scratchpad.key.data);
return (st);
}
