#ifndef _WIN32
#include <pthread.h>
#ifdef MDB_USE_POSIX_SEM
+# define MDB_USE_HASH 1
#include <semaphore.h>
#endif
#endif
* @{
*/
#ifdef _WIN32
+#define MDB_USE_HASH 1
#define MDB_PIDLOCK 0
#define pthread_t DWORD
#define pthread_mutex_t HANDLE
* The string is printed literally, with no format processing.
*/
#define DPUTS(arg) DPRINTF(("%s", arg))
+ /** Debuging output value of a cursor DBI: Negative in a sub-cursor. */
+#define DDBI(mc) \
+ (((mc)->mc_flags & C_SUB) ? -(int)(mc)->mc_dbi : (int)(mc)->mc_dbi)
/** @} */
/** A default memory page size.
*
* If #MDB_NOTLS is set, the slot address is not saved in thread-specific data.
*
- * No reader table is used if the database is on a read-only filesystem.
+ * No reader table is used if the database is on a read-only filesystem, or
+ * if #MDB_NOLOCK is set.
*
* Since the database uses multi-version concurrency control, readers don't
* actually need any locking. This table is used to keep track of which
* @ingroup internal
* @{
*/
-#define DB_DIRTY 0x01 /**< DB was written in this txn */
-#define DB_STALE 0x02 /**< DB record is older than txnID */
-#define DB_NEW 0x04 /**< DB handle opened in this txn */
+#define DB_DIRTY 0x01 /**< DB was modified or is DUPSORT data */
+#define DB_STALE 0x02 /**< Named-DB record is older than txnID */
+#define DB_NEW 0x04 /**< Named-DB handle opened in this txn */
#define DB_VALID 0x08 /**< DB handle is valid, see also #MDB_VALID */
/** @} */
/** In write txns, array of cursors for each DB */
#define MDB_TXN_SPILLS 0x08 /**< txn or a parent has spilled pages */
/** @} */
unsigned int mt_flags; /**< @ref mdb_txn */
- /** dirty_list maxsize - # of allocated pages allowed, including in parent txns */
- unsigned int mt_dirty_room;
- /** Tracks which of the two meta pages was used at the start
- * of this transaction.
+ /** dirty_list room: Array size - #dirty pages visible to this txn.
+ * Includes ancestor txns' dirty pages not hidden by other txns'
+ * dirty/spilled pages. Thus commit(nested txn) has room to merge
+ * dirty_list into mt_parent after freeing hidden mt_parent pages.
*/
- unsigned int mt_toggle;
+ unsigned int mt_dirty_room;
};
/** Enough space for 2^32 nodes with minimum of 2 keys per node. I.e., plenty.
MDB_val *key, int modify);
#define MDB_PS_MODIFY 1
#define MDB_PS_ROOTONLY 2
+#define MDB_PS_FIRST 4
+#define MDB_PS_LAST 8
static int mdb_page_search(MDB_cursor *mc,
MDB_val *key, int flags);
static int mdb_page_merge(MDB_cursor *csrc, MDB_cursor *cdst);
txn->mt_dbs[i].md_leaf_pages +
txn->mt_dbs[i].md_overflow_pages;
if (txn->mt_dbs[i].md_flags & MDB_DUPSORT) {
- mdb_page_search(&mc, NULL, 0);
+ mdb_page_search(&mc, NULL, MDB_PS_FIRST);
do {
unsigned j;
MDB_page *mp;
MDB_cursor *m3;
MDB_xcursor *mx;
MDB_page *dp, *mp;
+ MDB_node *leaf;
unsigned i, j;
int rc = MDB_SUCCESS, level;
mc = NULL; /* will find mc in mt_cursors */
for (i = txn->mt_numdbs;; mc = txn->mt_cursors[--i]) {
for (; mc; mc=mc->mc_next) {
- for (m3 = mc; m3->mc_flags & C_INITIALIZED; m3 = &mx->mx_cursor) {
+ if (!(mc->mc_flags & C_INITIALIZED))
+ continue;
+ for (m3 = mc;; m3 = &mx->mx_cursor) {
+ mp = NULL;
for (j=0; j<m3->mc_snum; j++) {
mp = m3->mc_pg[j];
if ((mp->mp_flags & Mask) == pflags)
mp->mp_flags ^= P_KEEP;
}
mx = m3->mc_xcursor;
- if (mx == NULL)
+ /* Proceed to mx if it is at a sub-database */
+ if (! (mx && (mx->mx_cursor.mc_flags & C_INITIALIZED)))
+ break;
+ if (! (mp && (mp->mp_flags & P_LEAF)))
+ break;
+ leaf = NODEPTR(mp, m3->mc_ki[j-1]);
+ if (!(leaf->mn_flags & F_SUBDATA))
break;
}
}
rc = mdb_pages_xkeep(m0, P_DIRTY|P_KEEP, i);
done:
- if (rc == 0) {
- if (txn->mt_parent) {
- txn->mt_dirty_room = txn->mt_parent->mt_dirty_room - dl[0].mid;
- /* dirty pages that are dirty in an ancestor don't
- * count against this txn's dirty_room.
- */
- for (i=1; i<=dl[0].mid; i++) {
- pgno_t pgno = dl[i].mid;
- MDB_txn *tx2;
- for (tx2 = txn->mt_parent; tx2; tx2 = tx2->mt_parent) {
- j = mdb_mid2l_search(tx2->mt_u.dirty_list, pgno);
- if (j <= tx2->mt_u.dirty_list[0].mid &&
- tx2->mt_u.dirty_list[j].mid == pgno) {
- txn->mt_dirty_room++;
- break;
- }
- }
- }
- } else {
- txn->mt_dirty_room = MDB_IDL_UM_MAX - dl[0].mid;
- }
- txn->mt_flags |= MDB_TXN_SPILLS;
- } else {
- txn->mt_flags |= MDB_TXN_ERROR;
- }
+ txn->mt_flags |= rc ? MDB_TXN_ERROR : MDB_TXN_SPILLS;
return rc;
}
{
int i;
txnid_t mr, oldest = txn->mt_txnid - 1;
- MDB_reader *r = txn->mt_env->me_txns->mti_readers;
- for (i = txn->mt_env->me_txns->mti_numreaders; --i >= 0; ) {
- if (r[i].mr_pid) {
- mr = r[i].mr_txnid;
- if (oldest > mr)
- oldest = mr;
+ if (txn->mt_env->me_txns) {
+ MDB_reader *r = txn->mt_env->me_txns->mti_readers;
+ for (i = txn->mt_env->me_txns->mti_numreaders; --i >= 0; ) {
+ if (r[i].mr_pid) {
+ mr = r[i].mr_txnid;
+ if (oldest > mr)
+ oldest = mr;
+ }
}
}
return oldest;
/** Pull a page off the txn's spill list, if present.
* If a page being referenced was spilled to disk in this txn, bring
* it back and make it dirty/writable again.
- * @param[in] tx0 the transaction handle.
+ * @param[in] txn the transaction handle.
* @param[in] mp the page being referenced.
* @param[out] ret the writable page, if any. ret is unchanged if
* mp wasn't spilled.
*/
static int
-mdb_page_unspill(MDB_txn *tx0, MDB_page *mp, MDB_page **ret)
+mdb_page_unspill(MDB_txn *txn, MDB_page *mp, MDB_page **ret)
{
- MDB_env *env = tx0->mt_env;
- MDB_txn *txn;
+ MDB_env *env = txn->mt_env;
+ const MDB_txn *tx2;
unsigned x;
pgno_t pgno = mp->mp_pgno, pn = pgno << 1;
- for (txn = tx0; txn; txn=txn->mt_parent) {
- if (!txn->mt_spill_pgs)
+ for (tx2 = txn; tx2; tx2=tx2->mt_parent) {
+ if (!tx2->mt_spill_pgs)
continue;
- x = mdb_midl_search(txn->mt_spill_pgs, pn);
- if (x <= txn->mt_spill_pgs[0] && txn->mt_spill_pgs[x] == pn) {
+ x = mdb_midl_search(tx2->mt_spill_pgs, pn);
+ if (x <= tx2->mt_spill_pgs[0] && tx2->mt_spill_pgs[x] == pn) {
MDB_page *np;
int num;
+ if (txn->mt_dirty_room == 0)
+ return MDB_TXN_FULL;
if (IS_OVERFLOW(mp))
num = mp->mp_pages;
else
else
mdb_page_copy(np, mp, env->me_psize);
}
- if (txn == tx0) {
+ if (tx2 == txn) {
/* If in current txn, this page is no longer spilled.
* If it happens to be the last page, truncate the spill list.
* Otherwise mark it as deleted by setting the LSB.
* page remains spilled until child commits
*/
- if (txn->mt_parent) {
- MDB_txn *tx2;
- /* If this page is also in a parent's dirty list, then
- * it's already accounted in dirty_room, and we need to
- * cancel out the decrement that mdb_page_dirty does.
- */
- for (tx2 = txn->mt_parent; tx2; tx2 = tx2->mt_parent) {
- x = mdb_mid2l_search(tx2->mt_u.dirty_list, pgno);
- if (x <= tx2->mt_u.dirty_list[0].mid &&
- tx2->mt_u.dirty_list[x].mid == pgno) {
- tx0->mt_dirty_room++;
- break;
- }
- }
- }
- mdb_page_dirty(tx0, np);
+ mdb_page_dirty(txn, np);
np->mp_flags |= P_DIRTY;
*ret = np;
break;
MDB_page *mp = mc->mc_pg[mc->mc_top], *np;
MDB_txn *txn = mc->mc_txn;
MDB_cursor *m2, *m3;
- MDB_dbi dbi;
pgno_t pgno;
int rc;
(rc = mdb_page_alloc(mc, 1, &np)))
return rc;
pgno = np->mp_pgno;
- DPRINTF(("touched db %u page %"Z"u -> %"Z"u", mc->mc_dbi,mp->mp_pgno,pgno));
+ DPRINTF(("touched db %d page %"Z"u -> %"Z"u", DDBI(mc),
+ mp->mp_pgno, pgno));
assert(mp->mp_pgno != pgno);
mdb_midl_xappend(txn->mt_free_pgs, mp->mp_pgno);
/* Update the parent page, if any, to point to the new page */
done:
/* Adjust cursors pointing to mp */
mc->mc_pg[mc->mc_top] = np;
- dbi = mc->mc_dbi;
+ m2 = txn->mt_cursors[mc->mc_dbi];
if (mc->mc_flags & C_SUB) {
- dbi--;
- for (m2 = txn->mt_cursors[dbi]; m2; m2=m2->mc_next) {
+ for (; m2; m2=m2->mc_next) {
m3 = &m2->mc_xcursor->mx_cursor;
if (m3->mc_snum < mc->mc_snum) continue;
if (m3->mc_pg[mc->mc_top] == mp)
m3->mc_pg[mc->mc_top] = np;
}
} else {
- for (m2 = txn->mt_cursors[dbi]; m2; m2=m2->mc_next) {
+ for (; m2; m2=m2->mc_next) {
if (m2->mc_snum < mc->mc_snum) continue;
if (m2->mc_pg[mc->mc_top] == mp) {
m2->mc_pg[mc->mc_top] = np;
mdb_txn_renew0(MDB_txn *txn)
{
MDB_env *env = txn->mt_env;
- unsigned int i;
+ MDB_txninfo *ti = env->me_txns;
+ MDB_meta *meta;
+ unsigned int i, nr;
uint16_t x;
int rc, new_notls = 0;
txn->mt_dbxs = env->me_dbxs; /* mostly static anyway */
if (txn->mt_flags & MDB_TXN_RDONLY) {
- if (!env->me_txns) {
- i = mdb_env_pick_meta(env);
- txn->mt_txnid = env->me_metas[i]->mm_txnid;
+ if (!ti) {
+ meta = env->me_metas[ mdb_env_pick_meta(env) ];
+ txn->mt_txnid = meta->mm_txnid;
txn->mt_u.reader = NULL;
} else {
MDB_reader *r = (env->me_flags & MDB_NOTLS) ? txn->mt_u.reader :
}
LOCK_MUTEX_R(env);
- for (i=0; i<env->me_txns->mti_numreaders; i++)
- if (env->me_txns->mti_readers[i].mr_pid == 0)
+ nr = ti->mti_numreaders;
+ for (i=0; i<nr; i++)
+ if (ti->mti_readers[i].mr_pid == 0)
break;
if (i == env->me_maxreaders) {
UNLOCK_MUTEX_R(env);
return MDB_READERS_FULL;
}
- env->me_txns->mti_readers[i].mr_pid = pid;
- env->me_txns->mti_readers[i].mr_tid = tid;
- if (i >= env->me_txns->mti_numreaders)
- env->me_txns->mti_numreaders = i+1;
+ ti->mti_readers[i].mr_pid = pid;
+ ti->mti_readers[i].mr_tid = tid;
+ if (i == nr)
+ ti->mti_numreaders = ++nr;
/* Save numreaders for un-mutexed mdb_env_close() */
- env->me_numreaders = env->me_txns->mti_numreaders;
+ env->me_numreaders = nr;
UNLOCK_MUTEX_R(env);
- r = &env->me_txns->mti_readers[i];
+
+ r = &ti->mti_readers[i];
new_notls = (env->me_flags & MDB_NOTLS);
if (!new_notls && (rc=pthread_setspecific(env->me_txkey, r))) {
r->mr_pid = 0;
return rc;
}
}
- txn->mt_txnid = r->mr_txnid = env->me_txns->mti_txnid;
+ txn->mt_txnid = r->mr_txnid = ti->mti_txnid;
txn->mt_u.reader = r;
+ meta = env->me_metas[txn->mt_txnid & 1];
}
- txn->mt_toggle = txn->mt_txnid & 1;
} else {
- LOCK_MUTEX_W(env);
+ if (ti) {
+ LOCK_MUTEX_W(env);
- txn->mt_txnid = env->me_txns->mti_txnid;
- txn->mt_toggle = txn->mt_txnid & 1;
+ txn->mt_txnid = ti->mti_txnid;
+ meta = env->me_metas[txn->mt_txnid & 1];
+ } else {
+ meta = env->me_metas[ mdb_env_pick_meta(env) ];
+ txn->mt_txnid = meta->mm_txnid;
+ }
txn->mt_txnid++;
#if MDB_DEBUG
if (txn->mt_txnid == mdb_debug_start)
}
/* Copy the DB info and flags */
- memcpy(txn->mt_dbs, env->me_metas[txn->mt_toggle]->mm_dbs, 2 * sizeof(MDB_db));
+ memcpy(txn->mt_dbs, meta->mm_dbs, 2 * sizeof(MDB_db));
/* Moved to here to avoid a data race in read TXNs */
- txn->mt_next_pgno = env->me_metas[txn->mt_toggle]->mm_last_pg+1;
+ txn->mt_next_pgno = meta->mm_last_pg+1;
for (i=2; i<txn->mt_numdbs; i++) {
x = env->me_dbflags[i];
return ENOMEM;
}
txn->mt_txnid = parent->mt_txnid;
- txn->mt_toggle = parent->mt_toggle;
txn->mt_dirty_room = parent->mt_dirty_room;
txn->mt_u.dirty_list[0].mid = 0;
txn->mt_spill_pgs = NULL;
env->me_txn = NULL;
/* The writer mutex was locked in mdb_txn_begin. */
- UNLOCK_MUTEX_W(env);
+ if (env->me_txns)
+ UNLOCK_MUTEX_W(env);
}
}
if (env->me_pghead) {
/* Make sure first page of freeDB is touched and on freelist */
- rc = mdb_page_search(&mc, NULL, MDB_PS_MODIFY);
+ rc = mdb_page_search(&mc, NULL, MDB_PS_FIRST|MDB_PS_MODIFY);
if (rc && rc != MDB_NOTFOUND)
return rc;
}
if (freecnt < txn->mt_free_pgs[0]) {
if (!freecnt) {
/* Make sure last page of freeDB is touched and on freelist */
- key.mv_size = MDB_MAXKEYSIZE+1;
- key.mv_data = NULL;
- rc = mdb_page_search(&mc, &key, MDB_PS_MODIFY);
+ rc = mdb_page_search(&mc, NULL, MDB_PS_LAST|MDB_PS_MODIFY);
if (rc && rc != MDB_NOTFOUND)
return rc;
}
}
dp->mp_flags &= ~P_DIRTY;
}
- dl[0].mid = j;
- return MDB_SUCCESS;
+ goto done;
}
/* Write the pages */
}
mdb_dpage_free(env, dp);
}
- dl[0].mid = j;
+done:
+ i--;
+ txn->mt_dirty_room += i - j;
+ dl[0].mid = j;
return MDB_SUCCESS;
}
if (txn->mt_parent) {
MDB_txn *parent = txn->mt_parent;
- unsigned x, y, len;
MDB_ID2L dst, src;
+ MDB_IDL pspill;
+ unsigned x, y, len, ps_len;
/* Append our free list to parent's */
rc = mdb_midl_append_list(&parent->mt_free_pgs, txn->mt_free_pgs);
if (rc)
goto fail;
mdb_midl_free(txn->mt_free_pgs);
+ /* Failures after this must either undo the changes
+ * to the parent or set MDB_TXN_ERROR in the parent.
+ */
parent->mt_next_pgno = txn->mt_next_pgno;
parent->mt_flags = txn->mt_flags;
dst = parent->mt_u.dirty_list;
src = txn->mt_u.dirty_list;
/* Remove anything in our dirty list from parent's spill list */
- if (parent->mt_spill_pgs) {
- x = parent->mt_spill_pgs[0];
- len = x;
- /* zero out our dirty pages in parent spill list */
- for (i=1; i<=src[0].mid; i++) {
+ if ((pspill = parent->mt_spill_pgs) && (ps_len = pspill[0])) {
+ x = y = ps_len;
+ pspill[0] = (pgno_t)-1;
+ /* Mark our dirty pages as deleted in parent spill list */
+ for (i=0, len=src[0].mid; ++i <= len; ) {
MDB_ID pn = src[i].mid << 1;
- if (pn < parent->mt_spill_pgs[x])
- continue;
- if (pn > parent->mt_spill_pgs[x]) {
- if (x <= 1)
- break;
+ while (pn > pspill[x])
x--;
- continue;
+ if (pn == pspill[x]) {
+ pspill[x] = 1;
+ y = --x;
}
- parent->mt_spill_pgs[x] = 0;
- len--;
- }
- /* OK, we had a few hits, squash zeros from the spill list */
- if (len < parent->mt_spill_pgs[0]) {
- x=1;
- for (y=1; y<=parent->mt_spill_pgs[0]; y++) {
- if (parent->mt_spill_pgs[y]) {
- if (y != x) {
- parent->mt_spill_pgs[x] = parent->mt_spill_pgs[y];
- }
- x++;
- }
- }
- parent->mt_spill_pgs[0] = len;
}
+ /* Squash deleted pagenums if we deleted any */
+ for (x=y; ++x <= ps_len; )
+ if (!(pspill[x] & 1))
+ pspill[++y] = pspill[x];
+ pspill[0] = y;
}
+
/* Find len = length of merging our dirty list with parent's */
x = dst[0].mid;
dst[0].mid = 0; /* simplify loops */
parent->mt_dirty_room = txn->mt_dirty_room;
if (txn->mt_spill_pgs) {
if (parent->mt_spill_pgs) {
- mdb_midl_append_list(&parent->mt_spill_pgs, txn->mt_spill_pgs);
+ /* TODO: Prevent failure here, so parent does not fail */
+ rc = mdb_midl_append_list(&parent->mt_spill_pgs, txn->mt_spill_pgs);
+ if (rc)
+ parent->mt_flags |= MDB_TXN_ERROR;
mdb_midl_free(txn->mt_spill_pgs);
mdb_midl_sort(parent->mt_spill_pgs);
} else {
parent->mt_child = NULL;
mdb_midl_free(((MDB_ntxn *)txn)->mnt_pgstate.mf_pghead);
free(txn);
- return MDB_SUCCESS;
+ return rc;
}
if (txn != env->me_txn) {
env->me_txn = NULL;
mdb_dbis_update(txn, 1);
- UNLOCK_MUTEX_W(env);
+ if (env->me_txns)
+ UNLOCK_MUTEX_W(env);
free(txn);
return MDB_SUCCESS;
assert(txn != NULL);
assert(txn->mt_env != NULL);
- toggle = !txn->mt_toggle;
+ toggle = txn->mt_txnid & 1;
DPRINTF(("writing meta page %d for root page %"Z"u",
toggle, txn->mt_dbs[MAIN_DBI].md_root));
* readers will get consistent data regardless of how fresh or
* how stale their view of these values is.
*/
- env->me_txns->mti_txnid = txn->mt_txnid;
+ if (env->me_txns)
+ env->me_txns->mti_txnid = txn->mt_txnid;
return MDB_SUCCESS;
}
int prot = PROT_READ;
if (flags & MDB_WRITEMAP) {
prot |= PROT_WRITE;
- if (newsize && ftruncate(env->me_fd, env->me_mapsize) < 0)
+ if (ftruncate(env->me_fd, env->me_mapsize) < 0)
return ErrCode();
}
env->me_map = mmap(addr, env->me_mapsize, prot, MAP_SHARED,
env->me_map = NULL;
return ErrCode();
}
- /* Turn off readahead. It's harmful when the DB is larger than RAM. */
+
+ if (flags & MDB_NORDAHEAD) {
+ /* Turn off readahead. It's harmful when the DB is larger than RAM. */
#ifdef MADV_RANDOM
- madvise(env->me_map, env->me_mapsize, MADV_RANDOM);
+ madvise(env->me_map, env->me_mapsize, MADV_RANDOM);
#else
#ifdef POSIX_MADV_RANDOM
- posix_madvise(env->me_map, env->me_mapsize, POSIX_MADV_RANDOM);
+ posix_madvise(env->me_map, env->me_mapsize, POSIX_MADV_RANDOM);
#endif /* POSIX_MADV_RANDOM */
#endif /* MADV_RANDOM */
+ }
#endif /* _WIN32 */
/* Can happen because the address argument to mmap() is just a
return EINVAL;
if (!size)
size = env->me_metas[mdb_env_pick_meta(env)]->mm_mapsize;
+ else if (size < env->me_mapsize) {
+ /* If the configured size is smaller, make sure it's
+ * still big enough. Silently round up to minimum if not.
+ */
+ size_t minsize = (env->me_metas[mdb_env_pick_meta(env)]->mm_last_pg + 1) * env->me_psize;
+ if (size < minsize)
+ size = minsize;
+ }
munmap(env->me_map, env->me_mapsize);
env->me_mapsize = size;
old = (env->me_flags & MDB_FIXEDMAP) ? env->me_map : NULL;
return rc;
}
-#if defined(_WIN32) || defined(MDB_USE_POSIX_SEM)
+#ifdef MDB_USE_HASH
/*
* hash_64 - 64 bit Fowler/Noll/Vo-0 FNV-1a hash code
*
* environment and re-opening it with the new flags.
*/
#define CHANGEABLE (MDB_NOSYNC|MDB_NOMETASYNC|MDB_MAPASYNC)
-#define CHANGELESS (MDB_FIXEDMAP|MDB_NOSUBDIR|MDB_RDONLY|MDB_WRITEMAP|MDB_NOTLS)
+#define CHANGELESS (MDB_FIXEDMAP|MDB_NOSUBDIR|MDB_RDONLY|MDB_WRITEMAP|MDB_NOTLS|MDB_NOLOCK|MDB_NORDAHEAD)
int
mdb_env_open(MDB_env *env, const char *path, unsigned int flags, mdb_mode_t mode)
}
/* For RDONLY, get lockfile after we know datafile exists */
- if (!F_ISSET(flags, MDB_RDONLY)) {
+ if (!(flags & (MDB_RDONLY|MDB_NOLOCK))) {
rc = mdb_env_setup_locks(env, lpath, mode, &excl);
if (rc)
goto leave;
goto leave;
}
- if (F_ISSET(flags, MDB_RDONLY)) {
+ if ((flags & (MDB_RDONLY|MDB_NOLOCK)) == MDB_RDONLY) {
rc = mdb_env_setup_locks(env, lpath, mode, &excl);
if (rc)
goto leave;
if (mc->mc_snum)
mc->mc_top--;
- DPRINTF(("popped page %"Z"u off db %u cursor %p", top->mp_pgno,
- mc->mc_dbi, (void *) mc));
+ DPRINTF(("popped page %"Z"u off db %d cursor %p", top->mp_pgno,
+ DDBI(mc), (void *) mc));
}
}
static int
mdb_cursor_push(MDB_cursor *mc, MDB_page *mp)
{
- DPRINTF(("pushing page %"Z"u on db %u cursor %p", mp->mp_pgno,
- mc->mc_dbi, (void *) mc));
+ DPRINTF(("pushing page %"Z"u on db %d cursor %p", mp->mp_pgno,
+ DDBI(mc), (void *) mc));
if (mc->mc_snum >= CURSOR_STACK) {
assert(mc->mc_snum < CURSOR_STACK);
return MDB_SUCCESS;
}
-/** Search for the page a given key should be in.
- * Pushes parent pages on the cursor stack. This function continues a
- * search on a cursor that has already been initialized. (Usually by
- * #mdb_page_search() but also by #mdb_node_move().)
- * @param[in,out] mc the cursor for this operation.
- * @param[in] key the key to search for. If NULL, search for the lowest
- * page. (This is used by #mdb_cursor_first().)
- * @param[in] modify If true, visited pages are updated with new page numbers.
- * @return 0 on success, non-zero on failure.
+/** Finish #mdb_page_search() / #mdb_page_search_lowest().
+ * The cursor is at the root page, set up the rest of it.
*/
static int
-mdb_page_search_root(MDB_cursor *mc, MDB_val *key, int modify)
+mdb_page_search_root(MDB_cursor *mc, MDB_val *key, int flags)
{
MDB_page *mp = mc->mc_pg[mc->mc_top];
int rc;
assert(NUMKEYS(mp) > 1);
DPRINTF(("found index 0 to page %"Z"u", NODEPGNO(NODEPTR(mp, 0))));
- if (key == NULL) /* Initialize cursor to first page. */
+ if (flags & (MDB_PS_FIRST|MDB_PS_LAST)) {
i = 0;
- else if (key->mv_size > MDB_MAXKEYSIZE && key->mv_data == NULL) {
- /* cursor to last page */
- i = NUMKEYS(mp)-1;
+ if (flags & MDB_PS_LAST)
+ i = NUMKEYS(mp) - 1;
} else {
int exact;
node = mdb_node_search(mc, key, &exact);
i--;
}
}
+ DPRINTF(("following index %u for key [%s]", i, DKEY(key)));
}
- if (key)
- DPRINTF(("following index %u for key [%s]", i, DKEY(key)));
assert(i < NUMKEYS(mp));
node = NODEPTR(mp, i);
if ((rc = mdb_cursor_push(mc, mp)))
return rc;
- if (modify) {
+ if (flags & MDB_PS_MODIFY) {
if ((rc = mdb_page_touch(mc)) != 0)
return rc;
mp = mc->mc_pg[mc->mc_top];
}
DPRINTF(("found leaf page %"Z"u for key [%s]", mp->mp_pgno,
- key ? DKEY(key) : NULL));
+ key ? DKEY(key) : "null"));
mc->mc_flags |= C_INITIALIZED;
mc->mc_flags &= ~C_EOF;
mc->mc_ki[mc->mc_top] = 0;
if ((rc = mdb_cursor_push(mc, mp)))
return rc;
- return mdb_page_search_root(mc, NULL, 0);
+ return mdb_page_search_root(mc, NULL, MDB_PS_FIRST);
}
/** Search for the page a given key should be in.
- * Pushes parent pages on the cursor stack. This function just sets up
- * the search; it finds the root page for \b mc's database and sets this
- * as the root of the cursor's stack. Then #mdb_page_search_root() is
- * called to complete the search.
+ * Push it and its parent pages on the cursor stack.
* @param[in,out] mc the cursor for this operation.
- * @param[in] key the key to search for. If NULL, search for the lowest
- * page. (This is used by #mdb_cursor_first().)
- * @param[in] flags If MDB_PS_MODIFY set, visited pages are updated with new page numbers.
+ * @param[in] key the key to search for, or NULL for first/last page.
+ * @param[in] flags If MDB_PS_MODIFY is set, visited pages in the DB
+ * are touched (updated with new page numbers).
+ * If MDB_PS_FIRST or MDB_PS_LAST is set, find first or last leaf.
+ * This is used by #mdb_cursor_first() and #mdb_cursor_last().
* If MDB_PS_ROOTONLY set, just fetch root node, no further lookups.
* @return 0 on success, non-zero on failure.
*/
pgno_t root;
/* Make sure the txn is still viable, then find the root from
- * the txn's db table.
+ * the txn's db table and set it as the root of the cursor's stack.
*/
if (F_ISSET(mc->mc_txn->mt_flags, MDB_TXN_ERROR)) {
DPUTS("transaction has failed, must abort");
return MDB_BAD_TXN;
} else {
/* Make sure we're using an up-to-date root */
- if (mc->mc_dbi > MAIN_DBI) {
- if ((*mc->mc_dbflag & DB_STALE) ||
- ((flags & MDB_PS_MODIFY) && !(*mc->mc_dbflag & DB_DIRTY))) {
+ if (*mc->mc_dbflag & DB_STALE) {
MDB_cursor mc2;
- unsigned char dbflag = 0;
mdb_cursor_init(&mc2, mc->mc_txn, MAIN_DBI, NULL);
- rc = mdb_page_search(&mc2, &mc->mc_dbx->md_name, flags & MDB_PS_MODIFY);
+ rc = mdb_page_search(&mc2, &mc->mc_dbx->md_name, 0);
if (rc)
return rc;
- if (*mc->mc_dbflag & DB_STALE) {
+ {
MDB_val data;
int exact = 0;
uint16_t flags;
return MDB_INCOMPATIBLE;
memcpy(mc->mc_db, data.mv_data, sizeof(MDB_db));
}
- if (flags & MDB_PS_MODIFY)
- dbflag = DB_DIRTY;
*mc->mc_dbflag &= ~DB_STALE;
- *mc->mc_dbflag |= dbflag;
- }
}
root = mc->mc_db->md_root;
mc->mc_snum = 1;
mc->mc_top = 0;
- DPRINTF(("db %u root page %"Z"u has flags 0x%X",
- mc->mc_dbi, root, mc->mc_pg[0]->mp_flags));
+ DPRINTF(("db %d root page %"Z"u has flags 0x%X",
+ DDBI(mc), root, mc->mc_pg[0]->mp_flags));
if (flags & MDB_PS_MODIFY) {
if ((rc = mdb_page_touch(mc)))
if (txn->mt_flags & MDB_TXN_ERROR)
return MDB_BAD_TXN;
- if (key->mv_size == 0 || key->mv_size > MDB_MAXKEYSIZE) {
+ if (key->mv_size > MDB_MAXKEYSIZE) {
return MDB_BAD_VALSIZE;
}
assert(IS_BRANCH(mc->mc_pg[mc->mc_top]));
indx = NODEPTR(mc->mc_pg[mc->mc_top], mc->mc_ki[mc->mc_top]);
- if ((rc = mdb_page_get(mc->mc_txn, NODEPGNO(indx), &mp, NULL) != 0))
+ if ((rc = mdb_page_get(mc->mc_txn, NODEPGNO(indx), &mp, NULL)) != 0) {
+ /* mc will be inconsistent if caller does mc_snum++ as above */
+ mc->mc_flags &= ~(C_INITIALIZED|C_EOF);
return rc;
+ }
mdb_cursor_push(mc, mp);
if (!move_right)
assert(mc);
assert(key);
- assert(key->mv_size > 0);
+ if (key->mv_size == 0)
+ return MDB_BAD_VALSIZE;
if (mc->mc_xcursor)
mc->mc_xcursor->mx_cursor.mc_flags &= ~(C_INITIALIZED|C_EOF);
mc->mc_xcursor->mx_cursor.mc_flags &= ~(C_INITIALIZED|C_EOF);
if (!(mc->mc_flags & C_INITIALIZED) || mc->mc_top) {
- rc = mdb_page_search(mc, NULL, 0);
+ rc = mdb_page_search(mc, NULL, MDB_PS_FIRST);
if (rc != MDB_SUCCESS)
return rc;
}
if (!(mc->mc_flags & C_EOF)) {
if (!(mc->mc_flags & C_INITIALIZED) || mc->mc_top) {
- MDB_val lkey;
-
- lkey.mv_size = MDB_MAXKEYSIZE+1;
- lkey.mv_data = NULL;
- rc = mdb_page_search(mc, &lkey, 0);
+ rc = mdb_page_search(mc, NULL, MDB_PS_LAST);
if (rc != MDB_SUCCESS)
return rc;
}
rc = EINVAL;
} else {
MDB_page *mp = mc->mc_pg[mc->mc_top];
- if (!NUMKEYS(mp)) {
- mc->mc_ki[mc->mc_top] = 0;
+ int nkeys = NUMKEYS(mp);
+ if (!nkeys || mc->mc_ki[mc->mc_top] >= nkeys) {
+ mc->mc_ki[mc->mc_top] = nkeys;
rc = MDB_NOTFOUND;
break;
}
case MDB_SET_RANGE:
if (key == NULL) {
rc = EINVAL;
- } else if (key->mv_size == 0 || key->mv_size > MDB_MAXKEYSIZE) {
+ } else if (key->mv_size > MDB_MAXKEYSIZE) {
rc = MDB_BAD_VALSIZE;
} else if (op == MDB_SET_RANGE)
rc = mdb_cursor_set(mc, key, data, op, NULL);
return rc;
}
-/** Touch all the pages in the cursor stack.
+/** Touch all the pages in the cursor stack. Set mc_top.
* Makes sure all the pages are writable, before attempting a write operation.
* @param[in] mc The cursor to operate on.
*/
static int
mdb_cursor_touch(MDB_cursor *mc)
{
- int rc;
+ int rc = MDB_SUCCESS;
if (mc->mc_dbi > MAIN_DBI && !(*mc->mc_dbflag & DB_DIRTY)) {
MDB_cursor mc2;
return rc;
*mc->mc_dbflag |= DB_DIRTY;
}
- for (mc->mc_top = 0; mc->mc_top < mc->mc_snum; mc->mc_top++) {
- rc = mdb_page_touch(mc);
- if (rc)
- return rc;
+ mc->mc_top = 0;
+ if (mc->mc_snum) {
+ do {
+ rc = mdb_page_touch(mc);
+ } while (!rc && ++(mc->mc_top) < mc->mc_snum);
+ mc->mc_top = mc->mc_snum-1;
}
- mc->mc_top = mc->mc_snum-1;
- return MDB_SUCCESS;
+ return rc;
}
/** Do not spill pages to disk if txn is getting full, may fail instead */
return MDB_BAD_VALSIZE;
#endif
- DPRINTF(("==> put db %u key [%s], size %"Z"u, data size %"Z"u",
- mc->mc_dbi, DKEY(key), key ? key->mv_size:0, data->mv_size));
+ DPRINTF(("==> put db %d key [%s], size %"Z"u, data size %"Z"u",
+ DDBI(mc), DKEY(key), key ? key->mv_size : 0, data->mv_size));
dkey.mv_size = 0;
} else if (mc->mc_db->md_root == P_INVALID) {
/* new database, cursor has nothing to point to */
mc->mc_snum = 0;
+ mc->mc_top = 0;
mc->mc_flags &= ~C_INITIALIZED;
rc = MDB_NO_ROOT;
} else {
unsigned i = mc->mc_top;
MDB_page *mp = mc->mc_pg[i];
- if (mc->mc_flags & C_SUB)
- dbi--;
-
for (m2 = mc->mc_txn->mt_cursors[dbi]; m2; m2=m2->mc_next) {
if (mc->mc_flags & C_SUB)
m3 = &m2->mc_xcursor->mx_cursor;
mdb_cursor_del(MDB_cursor *mc, unsigned int flags)
{
MDB_node *leaf;
+ MDB_page *mp;
int rc;
if (mc->mc_txn->mt_flags & (MDB_TXN_RDONLY|MDB_TXN_ERROR))
if (!(mc->mc_flags & C_INITIALIZED))
return EINVAL;
+ if (mc->mc_ki[mc->mc_top] >= NUMKEYS(mc->mc_pg[mc->mc_top]))
+ return MDB_NOTFOUND;
+
if (!(flags & MDB_NOSPILL) && (rc = mdb_page_spill(mc, NULL, NULL)))
return rc;
- flags &= ~MDB_NOSPILL; /* TODO: Or change (flags != MDB_NODUPDATA) to ~(flags & MDB_NODUPDATA), not looking at the logic of that code just now */
rc = mdb_cursor_touch(mc);
if (rc)
return rc;
- leaf = NODEPTR(mc->mc_pg[mc->mc_top], mc->mc_ki[mc->mc_top]);
+ mp = mc->mc_pg[mc->mc_top];
+ leaf = NODEPTR(mp, mc->mc_ki[mc->mc_top]);
- if (!IS_LEAF2(mc->mc_pg[mc->mc_top]) && F_ISSET(leaf->mn_flags, F_DUPDATA)) {
+ if (!IS_LEAF2(mp) && F_ISSET(leaf->mn_flags, F_DUPDATA)) {
if (!(flags & MDB_NODUPDATA)) {
if (!F_ISSET(leaf->mn_flags, F_SUBDATA)) {
mc->mc_xcursor->mx_cursor.mc_pg[0] = NODEDATA(leaf);
} else {
MDB_cursor *m2;
/* shrink fake page */
- mdb_node_shrink(mc->mc_pg[mc->mc_top], mc->mc_ki[mc->mc_top]);
- leaf = NODEPTR(mc->mc_pg[mc->mc_top], mc->mc_ki[mc->mc_top]);
+ mdb_node_shrink(mp, mc->mc_ki[mc->mc_top]);
+ leaf = NODEPTR(mp, mc->mc_ki[mc->mc_top]);
mc->mc_xcursor->mx_cursor.mc_pg[0] = NODEDATA(leaf);
/* fix other sub-DB cursors pointed at this fake page */
for (m2 = mc->mc_txn->mt_cursors[mc->mc_dbi]; m2; m2=m2->mc_next) {
if (m2 == mc || m2->mc_snum < mc->mc_snum) continue;
- if (m2->mc_pg[mc->mc_top] == mc->mc_pg[mc->mc_top] &&
+ if (m2->mc_pg[mc->mc_top] == mp &&
m2->mc_ki[mc->mc_top] == mc->mc_ki[mc->mc_top])
m2->mc_xcursor->mx_cursor.mc_pg[0] = NODEDATA(leaf);
}
{
unsigned int i;
size_t node_size = NODESIZE;
+ ssize_t room;
indx_t ofs;
MDB_node *node;
MDB_page *mp = mc->mc_pg[mc->mc_top];
IS_LEAF(mp) ? "leaf" : "branch",
IS_SUBP(mp) ? "sub-" : "",
mp->mp_pgno, indx, data ? data->mv_size : 0,
- key ? key->mv_size : 0, key ? DKEY(key) : NULL));
+ key ? key->mv_size : 0, key ? DKEY(key) : "null"));
if (IS_LEAF2(mp)) {
/* Move higher keys up one slot. */
return MDB_SUCCESS;
}
+ room = (ssize_t)SIZELEFT(mp) - (ssize_t)sizeof(indx_t);
if (key != NULL)
node_size += key->mv_size;
-
if (IS_LEAF(mp)) {
assert(data);
if (F_ISSET(flags, F_BIGDATA)) {
/* Put data on overflow page. */
DPRINTF(("data size is %"Z"u, node would be %"Z"u, put data on overflow page",
data->mv_size, node_size+data->mv_size));
- node_size += sizeof(pgno_t);
+ node_size += sizeof(pgno_t) + (node_size & 1);
+ if ((ssize_t)node_size > room)
+ goto full;
if ((rc = mdb_page_new(mc, P_OVERFLOW, ovpages, &ofp)))
return rc;
DPRINTF(("allocated overflow page %"Z"u", ofp->mp_pgno));
flags |= F_BIGDATA;
+ goto update;
} else {
node_size += data->mv_size;
}
}
node_size += node_size & 1;
+ if ((ssize_t)node_size > room)
+ goto full;
- if (node_size + sizeof(indx_t) > SIZELEFT(mp)) {
- DPRINTF(("not enough room in page %"Z"u, got %u ptrs",
- mp->mp_pgno, NUMKEYS(mp)));
- DPRINTF(("upper - lower = %u - %u = %u", mp->mp_upper, mp->mp_lower,
- mp->mp_upper - mp->mp_lower));
- DPRINTF(("node size = %"Z"u", node_size));
- return MDB_PAGE_FULL;
- }
-
+update:
/* Move higher pointers up one slot. */
for (i = NUMKEYS(mp); i > indx; i--)
mp->mp_ptrs[i] = mp->mp_ptrs[i - 1];
}
return MDB_SUCCESS;
+
+full:
+ DPRINTF(("not enough room in page %"Z"u, got %u ptrs",
+ mp->mp_pgno, NUMKEYS(mp)));
+ DPRINTF(("upper-lower = %u - %u = %"Z"d", mp->mp_upper,mp->mp_lower,room));
+ DPRINTF(("node size = %"Z"u", node_size));
+ return MDB_PAGE_FULL;
}
/** Delete the specified node from a page.
mx->mx_cursor.mc_txn = mc->mc_txn;
mx->mx_cursor.mc_db = &mx->mx_db;
mx->mx_cursor.mc_dbx = &mx->mx_dbx;
- mx->mx_cursor.mc_dbi = mc->mc_dbi+1;
+ mx->mx_cursor.mc_dbi = mc->mc_dbi;
mx->mx_cursor.mc_dbflag = &mx->mx_dbflag;
mx->mx_cursor.mc_snum = 0;
mx->mx_cursor.mc_top = 0;
mx->mx_cursor.mc_flags = C_SUB;
+ mx->mx_dbx.md_name.mv_size = 0;
+ mx->mx_dbx.md_name.mv_data = NULL;
mx->mx_dbx.md_cmp = mc->mc_dbx->md_dcmp;
mx->mx_dbx.md_dcmp = NULL;
mx->mx_dbx.md_rel = mc->mc_dbx->md_rel;
memcpy(&mx->mx_db, NODEDATA(node), sizeof(MDB_db));
mx->mx_cursor.mc_pg[0] = 0;
mx->mx_cursor.mc_snum = 0;
+ mx->mx_cursor.mc_top = 0;
mx->mx_cursor.mc_flags = C_SUB;
} else {
MDB_page *fp = NODEDATA(node);
mx->mx_db.md_entries = NUMKEYS(fp);
COPY_PGNO(mx->mx_db.md_root, fp->mp_pgno);
mx->mx_cursor.mc_snum = 1;
- mx->mx_cursor.mc_flags = C_INITIALIZED|C_SUB;
mx->mx_cursor.mc_top = 0;
+ mx->mx_cursor.mc_flags = C_INITIALIZED|C_SUB;
mx->mx_cursor.mc_pg[0] = fp;
mx->mx_cursor.mc_ki[0] = 0;
if (mc->mc_db->md_flags & MDB_DUPFIXED) {
mx->mx_db.md_flags |= MDB_INTEGERKEY;
}
}
- DPRINTF(("Sub-db %u for db %u root page %"Z"u", mx->mx_cursor.mc_dbi, mc->mc_dbi,
+ DPRINTF(("Sub-db -%u root page %"Z"u", mx->mx_cursor.mc_dbi,
mx->mx_db.md_root));
- mx->mx_dbflag = DB_VALID | (F_ISSET(mc->mc_pg[mc->mc_top]->mp_flags, P_DIRTY) ?
- DB_DIRTY : 0);
- mx->mx_dbx.md_name.mv_data = NODEKEY(node);
- mx->mx_dbx.md_name.mv_size = node->mn_ksize;
+ mx->mx_dbflag = DB_VALID|DB_DIRTY; /* DB_DIRTY guides mdb_cursor_touch */
#if UINT_MAX < SIZE_MAX
if (mx->mx_dbx.md_cmp == mdb_cmp_int && mx->mx_db.md_pad == sizeof(size_t))
#ifdef MISALIGNED_OK
MDB_dbi dbi = csrc->mc_dbi;
MDB_page *mp = csrc->mc_pg[csrc->mc_top];
- if (csrc->mc_flags & C_SUB)
- dbi--;
-
for (m2 = csrc->mc_txn->mt_cursors[dbi]; m2; m2=m2->mc_next) {
if (csrc->mc_flags & C_SUB)
m3 = &m2->mc_xcursor->mx_cursor;
MDB_dbi dbi = csrc->mc_dbi;
MDB_page *mp = cdst->mc_pg[cdst->mc_top];
- if (csrc->mc_flags & C_SUB)
- dbi--;
-
for (m2 = csrc->mc_txn->mt_cursors[dbi]; m2; m2=m2->mc_next) {
if (csrc->mc_flags & C_SUB)
m3 = &m2->mc_xcursor->mx_cursor;
/* Adjust cursors pointing to mp */
mc->mc_snum = 0;
mc->mc_top = 0;
+ mc->mc_flags &= ~C_INITIALIZED;
{
MDB_cursor *m2, *m3;
MDB_dbi dbi = mc->mc_dbi;
- if (mc->mc_flags & C_SUB)
- dbi--;
-
for (m2 = mc->mc_txn->mt_cursors[dbi]; m2; m2=m2->mc_next) {
if (mc->mc_flags & C_SUB)
m3 = &m2->mc_xcursor->mx_cursor;
if (m3->mc_pg[0] == mp) {
m3->mc_snum = 0;
m3->mc_top = 0;
+ m3->mc_flags &= ~C_INITIALIZED;
}
}
}
MDB_cursor *m2, *m3;
MDB_dbi dbi = mc->mc_dbi;
- if (mc->mc_flags & C_SUB)
- dbi--;
-
for (m2 = mc->mc_txn->mt_cursors[dbi]; m2; m2=m2->mc_next) {
if (mc->mc_flags & C_SUB)
m3 = &m2->mc_xcursor->mx_cursor;
m3 = m2;
if (m3 == mc || m3->mc_snum < mc->mc_snum) continue;
if (m3->mc_pg[0] == mp) {
- m3->mc_pg[0] = mc->mc_pg[0];
- m3->mc_snum = 1;
- m3->mc_top = 0;
- m3->mc_ki[0] = m3->mc_ki[1];
+ int i;
+ m3->mc_snum--;
+ m3->mc_top--;
+ for (i=0; i<m3->mc_snum; i++) {
+ m3->mc_pg[i] = m3->mc_pg[i+1];
+ m3->mc_ki[i] = m3->mc_ki[i+1];
+ }
}
}
}
/* Adjust other cursors pointing to mp */
for (m2 = mc->mc_txn->mt_cursors[dbi]; m2; m2=m2->mc_next) {
- if (m2 == mc)
+ if (m2 == mc || m2->mc_snum < mc->mc_snum)
continue;
if (!(m2->mc_flags & C_INITIALIZED))
continue;
if (txn->mt_flags & (MDB_TXN_RDONLY|MDB_TXN_ERROR))
return (txn->mt_flags & MDB_TXN_RDONLY) ? EACCES : MDB_BAD_TXN;
- if (key->mv_size == 0 || key->mv_size > MDB_MAXKEYSIZE) {
+ if (key->mv_size > MDB_MAXKEYSIZE) {
return MDB_BAD_VALSIZE;
}
unsigned int nflags)
{
unsigned int flags;
- int rc = MDB_SUCCESS, ins_new = 0, new_root = 0, newpos = 1, did_split = 0;
+ int rc = MDB_SUCCESS, new_root = 0, did_split = 0;
indx_t newindx;
pgno_t pgno = 0;
- unsigned int i, j, split_indx, nkeys, pmax;
+ int i, j, split_indx, nkeys, pmax;
+ MDB_env *env = mc->mc_txn->mt_env;
MDB_node *node;
MDB_val sepkey, rkey, xdata, *rdata = &xdata;
- MDB_page *copy;
+ MDB_page *copy = NULL;
MDB_page *mp, *rp, *pp;
- unsigned int ptop;
+ int ptop;
MDB_cursor mn;
DKBUF;
mp = mc->mc_pg[mc->mc_top];
newindx = mc->mc_ki[mc->mc_top];
+ nkeys = NUMKEYS(mp);
- DPRINTF(("-----> splitting %s page %"Z"u and adding [%s] at index %i",
+ DPRINTF(("-----> splitting %s page %"Z"u and adding [%s] at index %i/%i",
IS_LEAF(mp) ? "leaf" : "branch", mp->mp_pgno,
- DKEY(newkey), mc->mc_ki[mc->mc_top]));
+ DKEY(newkey), mc->mc_ki[mc->mc_top], nkeys));
/* Create a right sibling. */
if ((rc = mdb_page_new(mc, mp->mp_flags, 1, &rp)))
sepkey = *newkey;
split_indx = newindx;
nkeys = 0;
- goto newsep;
- }
+ } else {
- nkeys = NUMKEYS(mp);
- split_indx = nkeys / 2;
- if (newindx < split_indx)
- newpos = 0;
-
- if (IS_LEAF2(rp)) {
- char *split, *ins;
- int x;
- unsigned int lsize, rsize, ksize;
- /* Move half of the keys to the right sibling */
- copy = NULL;
- x = mc->mc_ki[mc->mc_top] - split_indx;
- ksize = mc->mc_db->md_pad;
- split = LEAF2KEY(mp, split_indx, ksize);
- rsize = (nkeys - split_indx) * ksize;
- lsize = (nkeys - split_indx) * sizeof(indx_t);
- mp->mp_lower -= lsize;
- rp->mp_lower += lsize;
- mp->mp_upper += rsize - lsize;
- rp->mp_upper -= rsize - lsize;
- sepkey.mv_size = ksize;
- if (newindx == split_indx) {
- sepkey.mv_data = newkey->mv_data;
- } else {
- sepkey.mv_data = split;
- }
- if (x<0) {
- ins = LEAF2KEY(mp, mc->mc_ki[mc->mc_top], ksize);
- memcpy(rp->mp_ptrs, split, rsize);
- sepkey.mv_data = rp->mp_ptrs;
- memmove(ins+ksize, ins, (split_indx - mc->mc_ki[mc->mc_top]) * ksize);
- memcpy(ins, newkey->mv_data, ksize);
- mp->mp_lower += sizeof(indx_t);
- mp->mp_upper -= ksize - sizeof(indx_t);
+ split_indx = (nkeys+1) / 2;
+
+ if (IS_LEAF2(rp)) {
+ char *split, *ins;
+ int x;
+ unsigned int lsize, rsize, ksize;
+ /* Move half of the keys to the right sibling */
+ copy = NULL;
+ x = mc->mc_ki[mc->mc_top] - split_indx;
+ ksize = mc->mc_db->md_pad;
+ split = LEAF2KEY(mp, split_indx, ksize);
+ rsize = (nkeys - split_indx) * ksize;
+ lsize = (nkeys - split_indx) * sizeof(indx_t);
+ mp->mp_lower -= lsize;
+ rp->mp_lower += lsize;
+ mp->mp_upper += rsize - lsize;
+ rp->mp_upper -= rsize - lsize;
+ sepkey.mv_size = ksize;
+ if (newindx == split_indx) {
+ sepkey.mv_data = newkey->mv_data;
+ } else {
+ sepkey.mv_data = split;
+ }
+ if (x<0) {
+ ins = LEAF2KEY(mp, mc->mc_ki[mc->mc_top], ksize);
+ memcpy(rp->mp_ptrs, split, rsize);
+ sepkey.mv_data = rp->mp_ptrs;
+ memmove(ins+ksize, ins, (split_indx - mc->mc_ki[mc->mc_top]) * ksize);
+ memcpy(ins, newkey->mv_data, ksize);
+ mp->mp_lower += sizeof(indx_t);
+ mp->mp_upper -= ksize - sizeof(indx_t);
+ } else {
+ if (x)
+ memcpy(rp->mp_ptrs, split, x * ksize);
+ ins = LEAF2KEY(rp, x, ksize);
+ memcpy(ins, newkey->mv_data, ksize);
+ memcpy(ins+ksize, split + x * ksize, rsize - x * ksize);
+ rp->mp_lower += sizeof(indx_t);
+ rp->mp_upper -= ksize - sizeof(indx_t);
+ mc->mc_ki[mc->mc_top] = x;
+ mc->mc_pg[mc->mc_top] = rp;
+ }
} else {
- if (x)
- memcpy(rp->mp_ptrs, split, x * ksize);
- ins = LEAF2KEY(rp, x, ksize);
- memcpy(ins, newkey->mv_data, ksize);
- memcpy(ins+ksize, split + x * ksize, rsize - x * ksize);
- rp->mp_lower += sizeof(indx_t);
- rp->mp_upper -= ksize - sizeof(indx_t);
- mc->mc_ki[mc->mc_top] = x;
- mc->mc_pg[mc->mc_top] = rp;
- }
- goto newsep;
- }
+ int psize, nsize, k;
+ /* Maximum free space in an empty page */
+ pmax = env->me_psize - PAGEHDRSZ;
+ if (IS_LEAF(mp))
+ nsize = mdb_leaf_size(env, newkey, newdata);
+ else
+ nsize = mdb_branch_size(env, newkey);
+ nsize += nsize & 1;
- /* For leaf pages, check the split point based on what
- * fits where, since otherwise mdb_node_add can fail.
- *
- * This check is only needed when the data items are
- * relatively large, such that being off by one will
- * make the difference between success or failure.
- *
- * It's also relevant if a page happens to be laid out
- * such that one half of its nodes are all "small" and
- * the other half of its nodes are "large." If the new
- * item is also "large" and falls on the half with
- * "large" nodes, it also may not fit.
- */
- if (IS_LEAF(mp)) {
- unsigned int psize, nsize;
- /* Maximum free space in an empty page */
- pmax = mc->mc_txn->mt_env->me_psize - PAGEHDRSZ;
- nsize = mdb_leaf_size(mc->mc_txn->mt_env, newkey, newdata);
- if ((nkeys < 20) || (nsize > pmax/16)) {
- if (newindx <= split_indx) {
- psize = nsize;
- newpos = 0;
- for (i=0; i<split_indx; i++) {
- node = NODEPTR(mp, i);
- psize += NODESIZE + NODEKSZ(node) + sizeof(indx_t);
- if (F_ISSET(node->mn_flags, F_BIGDATA))
- psize += sizeof(pgno_t);
- else
- psize += NODEDSZ(node);
- psize += psize & 1;
- if (psize > pmax) {
- if (i <= newindx) {
- split_indx = newindx;
- if (i < newindx)
- newpos = 1;
+ /* grab a page to hold a temporary copy */
+ copy = mdb_page_malloc(mc->mc_txn, 1);
+ if (copy == NULL)
+ return ENOMEM;
+ copy->mp_pgno = mp->mp_pgno;
+ copy->mp_flags = mp->mp_flags;
+ copy->mp_lower = PAGEHDRSZ;
+ copy->mp_upper = env->me_psize;
+
+ /* prepare to insert */
+ for (i=0, j=0; i<nkeys; i++) {
+ if (i == newindx) {
+ copy->mp_ptrs[j++] = 0;
+ }
+ copy->mp_ptrs[j++] = mp->mp_ptrs[i];
+ }
+
+ /* When items are relatively large the split point needs
+ * to be checked, because being off-by-one will make the
+ * difference between success or failure in mdb_node_add.
+ *
+ * It's also relevant if a page happens to be laid out
+ * such that one half of its nodes are all "small" and
+ * the other half of its nodes are "large." If the new
+ * item is also "large" and falls on the half with
+ * "large" nodes, it also may not fit.
+ *
+ * As a final tweak, if the new item goes on the last
+ * spot on the page (and thus, onto the new page), bias
+ * the split so the new page is emptier than the old page.
+ * This yields better packing during sequential inserts.
+ */
+ if (nkeys < 20 || nsize > pmax/16 || newindx >= nkeys) {
+ /* Find split point */
+ psize = 0;
+ if (newindx <= split_indx || newindx >= nkeys) {
+ i = 0; j = 1;
+ k = newindx >= nkeys ? nkeys : split_indx+2;
+ } else {
+ i = nkeys; j = -1;
+ k = split_indx-1;
+ }
+ for (; i!=k; i+=j) {
+ if (i == newindx) {
+ psize += nsize;
+ node = NULL;
+ } else {
+ node = (MDB_node *)((char *)mp + copy->mp_ptrs[i]);
+ psize += NODESIZE + NODEKSZ(node) + sizeof(indx_t);
+ if (IS_LEAF(mp)) {
+ if (F_ISSET(node->mn_flags, F_BIGDATA))
+ psize += sizeof(pgno_t);
+ else
+ psize += NODEDSZ(node);
}
- else
- split_indx = i;
- break;
+ psize += psize & 1;
}
- }
- } else {
- psize = nsize;
- for (i=nkeys-1; i>=split_indx; i--) {
- node = NODEPTR(mp, i);
- psize += NODESIZE + NODEKSZ(node) + sizeof(indx_t);
- if (F_ISSET(node->mn_flags, F_BIGDATA))
- psize += sizeof(pgno_t);
- else
- psize += NODEDSZ(node);
- psize += psize & 1;
- if (psize > pmax) {
- if (i >= newindx) {
- split_indx = newindx;
- newpos = 0;
- } else
- split_indx = i+1;
+ if (psize > pmax || i == k-j) {
+ split_indx = i + (j<0);
break;
}
}
}
+ if (split_indx == newindx) {
+ sepkey.mv_size = newkey->mv_size;
+ sepkey.mv_data = newkey->mv_data;
+ } else {
+ node = (MDB_node *)((char *)mp + copy->mp_ptrs[split_indx]);
+ sepkey.mv_size = node->mn_ksize;
+ sepkey.mv_data = NODEKEY(node);
+ }
}
}
- /* First find the separating key between the split pages.
- * The case where newindx == split_indx is ambiguous; the
- * new item could go to the new page or stay on the original
- * page. If newpos == 1 it goes to the new page.
- */
- if (newindx == split_indx && newpos) {
- sepkey.mv_size = newkey->mv_size;
- sepkey.mv_data = newkey->mv_data;
- } else {
- node = NODEPTR(mp, split_indx);
- sepkey.mv_size = node->mn_ksize;
- sepkey.mv_data = NODEKEY(node);
- }
-
-newsep:
- DPRINTF(("separator is [%s]", DKEY(&sepkey)));
+ DPRINTF(("separator is %d [%s]", split_indx, DKEY(&sepkey)));
/* Copy separator key to the parent.
*/
- if (SIZELEFT(mn.mc_pg[ptop]) < mdb_branch_size(mc->mc_txn->mt_env, &sepkey)) {
+ if (SIZELEFT(mn.mc_pg[ptop]) < mdb_branch_size(env, &sepkey)) {
mn.mc_snum--;
mn.mc_top--;
did_split = 1;
return rc;
for (i=0; i<mc->mc_top; i++)
mc->mc_ki[i] = mn.mc_ki[i];
- goto done;
- }
- if (IS_LEAF2(rp)) {
- goto done;
- }
-
- /* Move half of the keys to the right sibling. */
+ } else if (!IS_LEAF2(mp)) {
+ /* Move nodes */
+ mc->mc_pg[mc->mc_top] = rp;
+ i = split_indx;
+ j = 0;
+ do {
+ if (i == newindx) {
+ rkey.mv_data = newkey->mv_data;
+ rkey.mv_size = newkey->mv_size;
+ if (IS_LEAF(mp)) {
+ rdata = newdata;
+ } else
+ pgno = newpgno;
+ flags = nflags;
+ /* Update index for the new key. */
+ mc->mc_ki[mc->mc_top] = j;
+ } else {
+ node = (MDB_node *)((char *)mp + copy->mp_ptrs[i]);
+ rkey.mv_data = NODEKEY(node);
+ rkey.mv_size = node->mn_ksize;
+ if (IS_LEAF(mp)) {
+ xdata.mv_data = NODEDATA(node);
+ xdata.mv_size = NODEDSZ(node);
+ rdata = &xdata;
+ } else
+ pgno = NODEPGNO(node);
+ flags = node->mn_flags;
+ }
- /* grab a page to hold a temporary copy */
- copy = mdb_page_malloc(mc->mc_txn, 1);
- if (copy == NULL)
- return ENOMEM;
+ if (!IS_LEAF(mp) && j == 0) {
+ /* First branch index doesn't need key data. */
+ rkey.mv_size = 0;
+ }
- copy->mp_pgno = mp->mp_pgno;
- copy->mp_flags = mp->mp_flags;
- copy->mp_lower = PAGEHDRSZ;
- copy->mp_upper = mc->mc_txn->mt_env->me_psize;
- mc->mc_pg[mc->mc_top] = copy;
- for (i = j = 0; i <= nkeys; j++) {
- if (i == split_indx) {
- /* Insert in right sibling. */
- /* Reset insert index for right sibling. */
- if (i != newindx || (newpos ^ ins_new)) {
+ rc = mdb_node_add(mc, j, &rkey, rdata, pgno, flags);
+ if (rc) {
+ /* return tmp page to freelist */
+ mdb_page_free(env, copy);
+ return rc;
+ }
+ if (i == nkeys) {
+ i = 0;
j = 0;
- mc->mc_pg[mc->mc_top] = rp;
+ mc->mc_pg[mc->mc_top] = copy;
+ } else {
+ i++;
+ j++;
+ }
+ } while (i != split_indx);
+
+ nkeys = NUMKEYS(copy);
+ for (i=0; i<nkeys; i++)
+ mp->mp_ptrs[i] = copy->mp_ptrs[i];
+ mp->mp_lower = copy->mp_lower;
+ mp->mp_upper = copy->mp_upper;
+ memcpy(NODEPTR(mp, nkeys-1), NODEPTR(copy, nkeys-1),
+ env->me_psize - copy->mp_upper);
+
+ /* reset back to original page */
+ if (newindx < split_indx) {
+ mc->mc_pg[mc->mc_top] = mp;
+ if (nflags & MDB_RESERVE) {
+ node = NODEPTR(mp, mc->mc_ki[mc->mc_top]);
+ if (!(node->mn_flags & F_BIGDATA))
+ newdata->mv_data = NODEDATA(node);
}
- }
-
- if (i == newindx && !ins_new) {
- /* Insert the original entry that caused the split. */
- rkey.mv_data = newkey->mv_data;
- rkey.mv_size = newkey->mv_size;
- if (IS_LEAF(mp)) {
- rdata = newdata;
- } else
- pgno = newpgno;
- flags = nflags;
-
- ins_new = 1;
-
- /* Update index for the new key. */
- mc->mc_ki[mc->mc_top] = j;
- } else if (i == nkeys) {
- break;
} else {
- node = NODEPTR(mp, i);
- rkey.mv_data = NODEKEY(node);
- rkey.mv_size = node->mn_ksize;
- if (IS_LEAF(mp)) {
- xdata.mv_data = NODEDATA(node);
- xdata.mv_size = NODEDSZ(node);
- rdata = &xdata;
- } else
- pgno = NODEPGNO(node);
- flags = node->mn_flags;
-
- i++;
- }
-
- if (!IS_LEAF(mp) && j == 0) {
- /* First branch index doesn't need key data. */
- rkey.mv_size = 0;
- }
-
- rc = mdb_node_add(mc, j, &rkey, rdata, pgno, flags);
- if (rc) break;
- }
-
- nkeys = NUMKEYS(copy);
- for (i=0; i<nkeys; i++)
- mp->mp_ptrs[i] = copy->mp_ptrs[i];
- mp->mp_lower = copy->mp_lower;
- mp->mp_upper = copy->mp_upper;
- memcpy(NODEPTR(mp, nkeys-1), NODEPTR(copy, nkeys-1),
- mc->mc_txn->mt_env->me_psize - copy->mp_upper);
-
- /* reset back to original page */
- if (newindx < split_indx || (!newpos && newindx == split_indx)) {
- mc->mc_pg[mc->mc_top] = mp;
- if (nflags & MDB_RESERVE) {
- node = NODEPTR(mp, mc->mc_ki[mc->mc_top]);
- if (!(node->mn_flags & F_BIGDATA))
- newdata->mv_data = NODEDATA(node);
- }
- } else {
- mc->mc_ki[ptop]++;
- /* Make sure mc_ki is still valid.
- */
- if (mn.mc_pg[ptop] != mc->mc_pg[ptop] &&
- mc->mc_ki[ptop] >= NUMKEYS(mc->mc_pg[ptop])) {
- for (i=0; i<ptop; i++) {
- mc->mc_pg[i] = mn.mc_pg[i];
- mc->mc_ki[i] = mn.mc_ki[i];
+ mc->mc_pg[mc->mc_top] = rp;
+ mc->mc_ki[ptop]++;
+ /* Make sure mc_ki is still valid.
+ */
+ if (mn.mc_pg[ptop] != mc->mc_pg[ptop] &&
+ mc->mc_ki[ptop] >= NUMKEYS(mc->mc_pg[ptop])) {
+ for (i=0; i<ptop; i++) {
+ mc->mc_pg[i] = mn.mc_pg[i];
+ mc->mc_ki[i] = mn.mc_ki[i];
+ }
+ mc->mc_pg[ptop] = mn.mc_pg[ptop];
+ mc->mc_ki[ptop] = mn.mc_ki[ptop] - 1;
}
- mc->mc_pg[ptop] = mn.mc_pg[ptop];
- mc->mc_ki[ptop] = mn.mc_ki[ptop] - 1;
}
+ /* return tmp page to freelist */
+ mdb_page_free(env, copy);
}
- /* return tmp page to freelist */
- mdb_page_free(mc->mc_txn->mt_env, copy);
-done:
{
/* Adjust other cursors pointing to mp */
MDB_cursor *m2, *m3;
MDB_dbi dbi = mc->mc_dbi;
int fixup = NUMKEYS(mp);
- if (mc->mc_flags & C_SUB)
- dbi--;
-
for (m2 = mc->mc_txn->mt_cursors[dbi]; m2; m2=m2->mc_next) {
if (mc->mc_flags & C_SUB)
m3 = &m2->mc_xcursor->mx_cursor;
}
}
}
+ DPRINTF(("mp left: %d, rp left: %d", SIZELEFT(mp), SIZELEFT(rp)));
return rc;
}
if (txn == NULL || !dbi || dbi >= txn->mt_numdbs || !(txn->mt_dbflags[dbi] & DB_VALID))
return EINVAL;
- if (txn->mt_flags & (MDB_TXN_RDONLY|MDB_TXN_ERROR))
- return (txn->mt_flags & MDB_TXN_RDONLY) ? EACCES : MDB_BAD_TXN;
-
- if (key->mv_size == 0 || key->mv_size > MDB_MAXKEYSIZE) {
- return MDB_BAD_VALSIZE;
- }
-
if ((flags & (MDB_NOOVERWRITE|MDB_NODUPDATA|MDB_RESERVE|MDB_APPEND|MDB_APPENDDUP)) != flags)
return EINVAL;
return MDB_SUCCESS;
}
+int
+mdb_env_get_fd(MDB_env *env, mdb_filehandle_t *arg)
+{
+ if (!env || !arg)
+ return EINVAL;
+
+ *arg = env->me_fd;
+ return MDB_SUCCESS;
+}
+
/** Common code for #mdb_stat() and #mdb_env_stat().
* @param[in] env the environment to operate in.
* @param[in] db the #MDB_db record containing the stats to return.
{
int rc;
- rc = mdb_page_search(mc, NULL, 0);
+ rc = mdb_page_search(mc, NULL, MDB_PS_FIRST);
if (rc == MDB_SUCCESS) {
MDB_txn *txn = mc->mc_txn;
MDB_node *ni;
return -1;
}
}
-
+
if( val > 0 ) {
++cursor;
}
if (!mdb_reader_pid(env, Pidcheck, pid)) {
for (j=i; j<rdrs; j++)
if (mr[j].mr_pid == pid) {
+ DPRINTF(("clear stale reader pid %u txn %"Z"d",
+ (unsigned) pid, mr[j].mr_txnid));
mr[j].mr_pid = 0;
count++;
}