#define ESECT
#endif
+#ifdef _MSC_VER
+#define CALL_CONV WINAPI
+#else
+#define CALL_CONV
+#endif
+
/** @defgroup internal LMDB Internals
* @{
*/
/** The version number for a database's lockfile format. */
#define MDB_LOCK_VERSION 1
- /** @brief The max size of a key we can write, or 0 for dynamic max.
+ /** @brief The max size of a key we can write, or 0 for computed max.
*
- * Define this as 0 to compute the max from the page size. 511
- * is default for backwards compat: liblmdb <= 0.9.10 can break
- * when modifying a DB with keys/dupsort data bigger than its max.
- * #MDB_DEVEL sets the default to 0.
+ * This macro should normally be left alone or set to 0.
+ * Note that a database with big keys or dupsort data cannot be
+ * reliably modified by a liblmdb which uses a smaller max.
+ * The default is 511 for backwards compat, or 0 when #MDB_DEVEL.
+ *
+ * Other values are allowed, for backwards compat. However:
+ * A value bigger than the computed max can break if you do not
+ * know what you are doing, and liblmdb <= 0.9.10 can break when
+ * modifying a DB with keys/dupsort data bigger than its max.
*
* Data items in an #MDB_DUPSORT database are also limited to
* this size, since they're actually keys of a sub-DB. Keys and
* @ingroup internal
* @{
*/
-#define MDB_TXN_RDONLY 0x01 /**< read-only transaction */
+ /** #mdb_txn_begin() flags */
+#define MDB_TXN_BEGIN_FLAGS MDB_RDONLY
+#define MDB_TXN_RDONLY MDB_RDONLY /**< read-only transaction */
+ /* internal txn flags */
+#define MDB_TXN_WRITEMAP MDB_WRITEMAP /**< copy of #MDB_env flag in writers */
#define MDB_TXN_ERROR 0x02 /**< txn is unusable after an error */
#define MDB_TXN_DIRTY 0x04 /**< must write, even if dirty list is empty */
#define MDB_TXN_SPILLS 0x08 /**< txn or a parent has spilled pages */
unsigned int me_psize; /**< DB page size, inited from me_os_psize */
unsigned int me_os_psize; /**< OS page size, from #GET_PAGESIZE */
unsigned int me_maxreaders; /**< size of the reader table */
- unsigned int me_numreaders; /**< max numreaders set by this env */
+ /** Max #MDB_txninfo.%mti_numreaders of interest to #mdb_env_close() */
+ volatile int me_close_readers;
MDB_dbi me_numdbs; /**< number of DBs opened */
MDB_dbi me_maxdbs; /**< size of the DB table */
MDB_PID_T me_pid; /**< process ID of this env */
static void mdb_cursor_init(MDB_cursor *mc, MDB_txn *txn, MDB_dbi dbi, MDB_xcursor *mx);
static void mdb_xcursor_init0(MDB_cursor *mc);
static void mdb_xcursor_init1(MDB_cursor *mc, MDB_node *node);
+static void mdb_xcursor_init2(MDB_cursor *mc, MDB_xcursor *src_mx, int force);
static int mdb_drop0(MDB_cursor *mc, int subs);
static void mdb_default_cmp(MDB_txn *txn, MDB_dbi dbi);
static MDB_cmp_func mdb_cmp_memn, mdb_cmp_memnr, mdb_cmp_int, mdb_cmp_cint, mdb_cmp_long;
/** @endcond */
+/** Compare two items pointing at size_t's of unknown alignment. */
+#ifdef MISALIGNED_OK
+# define mdb_cmp_clong mdb_cmp_long
+#else
+# define mdb_cmp_clong mdb_cmp_cint
+#endif
+
#ifdef _WIN32
static SECURITY_DESCRIPTOR mdb_null_sd;
static SECURITY_ATTRIBUTES mdb_all_sa;
int
mdb_dcmp(MDB_txn *txn, MDB_dbi dbi, const MDB_val *a, const MDB_val *b)
{
- return txn->mt_dbxs[dbi].md_dcmp(a, b);
+ MDB_cmp_func *dcmp = txn->mt_dbxs[dbi].md_dcmp;
+#if UINT_MAX < SIZE_MAX
+ if (dcmp == mdb_cmp_int && a->mv_size == sizeof(size_t))
+ dcmp = mdb_cmp_clong;
+#endif
+ return dcmp(a, b);
}
/** Allocate memory for a page.
MDB_ID2 mid;
int rc, (*insert)(MDB_ID2L, MDB_ID2 *);
- if (txn->mt_env->me_flags & MDB_WRITEMAP) {
+ if (txn->mt_flags & MDB_TXN_WRITEMAP) {
insert = mdb_mid2l_append;
} else {
insert = mdb_mid2l_insert;
MDB_env *env = txn->mt_env;
MDB_txninfo *ti = env->me_txns;
MDB_meta *meta;
- unsigned int i, nr;
+ unsigned int i, nr, flags = txn->mt_flags;
uint16_t x;
int rc, new_notls = 0;
- if (txn->mt_flags & MDB_TXN_RDONLY) {
- txn->mt_flags = MDB_TXN_RDONLY;
- /* Setup db info */
- txn->mt_numdbs = env->me_numdbs;
- txn->mt_dbxs = env->me_dbxs; /* mostly static anyway */
+ if ((flags &= MDB_TXN_RDONLY) != 0) {
if (!ti) {
meta = env->me_metas[ mdb_env_pick_meta(env) ];
txn->mt_txnid = meta->mm_txnid;
UNLOCK_MUTEX_R(env);
return MDB_READERS_FULL;
}
- ti->mti_readers[i].mr_pid = pid;
- ti->mti_readers[i].mr_tid = tid;
+ r = &ti->mti_readers[i];
+ /* Claim the reader slot, carefully since other code
+ * uses the reader table un-mutexed: First reset the
+ * slot, next publish it in mti_numreaders. After
+ * that, it is safe for mdb_env_close() to touch it.
+ * When it will be closed, we can finally claim it.
+ */
+ r->mr_pid = 0;
+ r->mr_txnid = (txnid_t)-1;
+ r->mr_tid = tid;
if (i == nr)
ti->mti_numreaders = ++nr;
- /* Save numreaders for un-mutexed mdb_env_close() */
- env->me_numreaders = nr;
+ env->me_close_readers = nr;
+ r->mr_pid = pid;
UNLOCK_MUTEX_R(env);
- 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;
txn->mt_u.reader = r;
meta = env->me_metas[txn->mt_txnid & 1];
}
+ txn->mt_dbxs = env->me_dbxs; /* mostly static anyway */
} else {
+ /* Not yet touching txn == env->me_txn0, it may be active */
if (ti) {
LOCK_MUTEX_W(env);
meta = env->me_metas[ mdb_env_pick_meta(env) ];
txn->mt_txnid = meta->mm_txnid;
}
- /* Setup db info */
- txn->mt_numdbs = env->me_numdbs;
txn->mt_txnid++;
#if MDB_DEBUG
if (txn->mt_txnid == mdb_debug_start)
mdb_debug = 1;
#endif
- txn->mt_flags = 0;
txn->mt_child = NULL;
txn->mt_loose_pgs = NULL;
txn->mt_loose_count = 0;
/* Moved to here to avoid a data race in read TXNs */
txn->mt_next_pgno = meta->mm_last_pg+1;
+ txn->mt_flags = flags;
+
+ /* Setup db info */
+ txn->mt_numdbs = env->me_numdbs;
for (i=2; i<txn->mt_numdbs; i++) {
x = env->me_dbflags[i];
txn->mt_dbs[i].md_flags = x & PERSISTENT_FLAGS;
{
MDB_txn *txn;
MDB_ntxn *ntxn;
- int rc, size, tsize = sizeof(MDB_txn);
+ int rc, size, tsize;
+
+ flags &= MDB_TXN_BEGIN_FLAGS;
+ flags |= env->me_flags & MDB_WRITEMAP;
if (env->me_flags & MDB_FATAL_ERROR) {
DPUTS("environment had fatal error, must shutdown!");
return MDB_PANIC;
}
- if ((env->me_flags & MDB_RDONLY) && !(flags & MDB_RDONLY))
+ if (env->me_flags & MDB_RDONLY & ~flags) /* write txn in RDONLY env */
return EACCES;
+
+ size = tsize = sizeof(MDB_txn);
if (parent) {
/* Nested transactions: Max 1 child, write txns only, no writemap */
+ flags |= parent->mt_flags;
if (parent->mt_child ||
- (flags & MDB_RDONLY) ||
- (parent->mt_flags & (MDB_TXN_RDONLY|MDB_TXN_ERROR)) ||
- (env->me_flags & MDB_WRITEMAP))
+ (flags & (MDB_RDONLY|MDB_WRITEMAP|MDB_TXN_ERROR)))
{
return (parent->mt_flags & MDB_TXN_RDONLY) ? EINVAL : MDB_BAD_TXN;
}
- tsize = sizeof(MDB_ntxn);
- }
- size = tsize;
- if (!(flags & MDB_RDONLY)) {
- if (!parent) {
- txn = env->me_txn0; /* just reuse preallocated write txn */
- goto ok;
- }
- /* child txns use own copy of cursors */
+ /* Child txns save MDB_pgstate and use own copy of cursors */
+ size = tsize = sizeof(MDB_ntxn);
size += env->me_maxdbs * sizeof(MDB_cursor *);
+ } else if (!(flags & MDB_RDONLY)) {
+ /* Reuse preallocated write txn. However, do not touch it until
+ * mdb_txn_renew0() succeeds, since it currently may be active.
+ */
+ txn = env->me_txn0;
+ goto renew;
}
size += env->me_maxdbs * (sizeof(MDB_db)+1);
}
txn->mt_dbs = (MDB_db *) ((char *)txn + tsize);
if (flags & MDB_RDONLY) {
- txn->mt_flags |= MDB_TXN_RDONLY;
txn->mt_dbflags = (unsigned char *)(txn->mt_dbs + env->me_maxdbs);
txn->mt_dbiseqs = env->me_dbiseqs;
} else {
txn->mt_dbflags = (unsigned char *)(txn->mt_dbiseqs + env->me_maxdbs);
}
}
+ txn->mt_flags = flags;
txn->mt_env = env;
-ok:
if (parent) {
unsigned int i;
txn->mt_u.dirty_list = malloc(sizeof(MDB_ID2)*MDB_IDL_UM_SIZE);
parent->mt_child = txn;
txn->mt_parent = parent;
txn->mt_numdbs = parent->mt_numdbs;
- txn->mt_flags = parent->mt_flags;
txn->mt_dbxs = parent->mt_dbxs;
memcpy(txn->mt_dbs, parent->mt_dbs, txn->mt_numdbs * sizeof(MDB_db));
/* Copy parent's mt_dbflags, but clear DB_NEW */
if (rc)
mdb_txn_reset0(txn, "beginchild-fail");
} else {
+renew:
rc = mdb_txn_renew0(txn);
}
if (rc) {
if (txn != env->me_txn0)
free(txn);
} else {
+ txn->mt_flags |= flags; /* for txn==me_txn0, no effect otherwise */
*ret = txn;
DPRINTF(("begin txn %"Z"u%c %p on mdbenv %p, root page %"Z"u",
- txn->mt_txnid, (txn->mt_flags & MDB_TXN_RDONLY) ? 'r' : 'w',
+ txn->mt_txnid, (flags & MDB_RDONLY) ? 'r' : 'w',
(void *) txn, (void *) env, txn->mt_dbs[MAIN_DBI].md_root));
}
return txn->mt_env;
}
+size_t
+mdb_txn_id(MDB_txn *txn)
+{
+ if(!txn) return 0;
+ return txn->mt_txnid;
+}
+
/** Export or close DBI handles opened in this txn. */
static void
mdb_dbis_update(MDB_txn *txn, int keep)
}
if (!txn->mt_parent) {
- if (mdb_midl_shrink(&txn->mt_free_pgs))
- env->me_free_pgs = txn->mt_free_pgs;
+ mdb_midl_shrink(&txn->mt_free_pgs);
+ env->me_free_pgs = txn->mt_free_pgs;
/* me_pgstate: */
env->me_pghead = NULL;
env->me_pglast = 0;
goto fail;
}
data.mv_data = &txn->mt_dbs[i];
- rc = mdb_cursor_put(&mc, &txn->mt_dbxs[i].md_name, &data, 0);
+ rc = mdb_cursor_put(&mc, &txn->mt_dbxs[i].md_name, &data,
+ F_SUBDATA);
if (rc)
goto fail;
}
mdb_midl_free(env->me_pghead);
env->me_pghead = NULL;
- if (mdb_midl_shrink(&txn->mt_free_pgs))
- env->me_free_pgs = txn->mt_free_pgs;
+ mdb_midl_shrink(&txn->mt_free_pgs);
+ env->me_free_pgs = txn->mt_free_pgs;
#if (MDB_DEBUG) > 2
mdb_audit(txn);
return 0;
}
+/** Fill in most of the zeroed #MDB_meta for an empty database environment */
static void ESECT
mdb_env_init_meta0(MDB_env *env, MDB_meta *meta)
{
/** Write the environment parameters of a freshly created DB environment.
* @param[in] env the environment handle
- * @param[out] meta address of where to store the meta information
+ * @param[in] meta the #MDB_meta to write
* @return 0 on success, non-zero on failure.
*/
static int ESECT
psize = env->me_psize;
- mdb_env_init_meta0(env, meta);
-
p = calloc(2, psize);
+ if (!p)
+ return ENOMEM;
+
p->mp_pgno = 0;
p->mp_flags = P_META;
*(MDB_meta *)METADATA(p) = *meta;
{
MDB_env *env;
MDB_meta meta, metab, *mp;
+ unsigned flags;
size_t mapsize;
off_t off;
int rc, len, toggle;
toggle, txn->mt_dbs[MAIN_DBI].md_root));
env = txn->mt_env;
+ flags = env->me_flags;
mp = env->me_metas[toggle];
mapsize = env->me_metas[toggle ^ 1]->mm_mapsize;
/* Persist any increases of mapsize config */
if (mapsize < env->me_mapsize)
mapsize = env->me_mapsize;
- if (env->me_flags & MDB_WRITEMAP) {
+ if (flags & MDB_WRITEMAP) {
mp->mm_mapsize = mapsize;
mp->mm_dbs[0] = txn->mt_dbs[0];
mp->mm_dbs[1] = txn->mt_dbs[1];
mp->mm_last_pg = txn->mt_next_pgno - 1;
mp->mm_txnid = txn->mt_txnid;
- if (!(env->me_flags & (MDB_NOMETASYNC|MDB_NOSYNC))) {
+ if (!(flags & (MDB_NOMETASYNC|MDB_NOSYNC))) {
unsigned meta_size = env->me_psize;
rc = (env->me_flags & MDB_MAPASYNC) ? MS_ASYNC : MS_SYNC;
ptr = env->me_map;
off += PAGEHDRSZ;
/* Write to the SYNC fd */
- mfd = env->me_flags & (MDB_NOSYNC|MDB_NOMETASYNC) ?
- env->me_fd : env->me_mfd;
-retry_write:
+ mfd = (flags & (MDB_NOSYNC|MDB_NOMETASYNC)) ? env->me_fd : env->me_mfd;
#ifdef _WIN32
{
memset(&ov, 0, sizeof(ov));
rc = -1;
}
#else
+retry_write:
rc = pwrite(mfd, ptr, len, off);
#endif
if (rc != len) {
rc = rc < 0 ? ErrCode() : EIO;
+#ifndef _WIN32
if (rc == EINTR)
goto retry_write;
+#endif
DPUTS("write failed, disk error?");
/* On a failure, the pagecache still contains the new data.
* Write some old data back, to prevent it from being used.
*/
if (env->me_map) {
int rc;
+ MDB_meta *meta;
void *old;
if (env->me_txn)
return EINVAL;
+ meta = env->me_metas[mdb_env_pick_meta(env)];
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;
+ size = meta->mm_mapsize;
+ {
+ /* Silently round up to minimum if the size is too small */
+ size_t minsize = (meta->mm_last_pg + 1) * env->me_psize;
if (size < minsize)
size = minsize;
}
}
#endif
- memset(&meta, 0, sizeof(meta));
-
if ((i = mdb_env_read_header(env, &meta)) != 0) {
if (i != ENOENT)
return i;
env->me_psize = env->me_os_psize;
if (env->me_psize > MAX_PAGESIZE)
env->me_psize = MAX_PAGESIZE;
+ memset(&meta, 0, sizeof(meta));
+ mdb_env_init_meta0(env, &meta);
+ meta.mm_mapsize = DEFAULT_MAPSIZE;
} else {
env->me_psize = meta.mm_psize;
}
/* Was a mapsize configured? */
if (!env->me_mapsize) {
- /* If this is a new environment, take the default,
- * else use the size recorded in the existing env.
- */
- env->me_mapsize = newenv ? DEFAULT_MAPSIZE : meta.mm_mapsize;
- } else if (env->me_mapsize < meta.mm_mapsize) {
- /* If the configured size is smaller, make sure it's
- * still big enough. Silently round up to minimum if not.
+ env->me_mapsize = meta.mm_mapsize;
+ }
+ {
+ /* Make sure mapsize >= committed data size. Even when using
+ * mm_mapsize, which could be broken in old files (ITS#7789).
*/
size_t minsize = (meta.mm_last_pg + 1) * meta.mm_psize;
if (env->me_mapsize < minsize)
env->me_mapsize = minsize;
}
+ meta.mm_mapsize = env->me_mapsize;
+
+ if (newenv && !(flags & MDB_FIXEDMAP)) {
+ /* mdb_env_map() may grow the datafile. Write the metapages
+ * first, so the file will be valid if initialization fails.
+ * Except with FIXEDMAP, since we do not yet know mm_address.
+ * We could fill in mm_address later, but then a different
+ * program might end up doing that - one with a memory layout
+ * and map address which does not suit the main program.
+ */
+ rc = mdb_env_init_meta(env, &meta);
+ if (rc)
+ return rc;
+ newenv = 0;
+ }
rc = mdb_env_map(env, (flags & MDB_FIXEDMAP) ? meta.mm_address : NULL);
if (rc)
if (rc)
goto leave;
}
- if (!((flags & MDB_RDONLY) ||
- (env->me_pbuf = calloc(1, env->me_psize))))
- rc = ENOMEM;
if (!(flags & MDB_RDONLY)) {
MDB_txn *txn;
int tsize = sizeof(MDB_txn), size = tsize + env->me_maxdbs *
(sizeof(MDB_db)+sizeof(MDB_cursor *)+sizeof(unsigned int)+1);
- txn = calloc(1, size);
- if (txn) {
+ if ((env->me_pbuf = calloc(1, env->me_psize)) &&
+ (txn = calloc(1, size)))
+ {
txn->mt_dbs = (MDB_db *)((char *)txn + tsize);
txn->mt_cursors = (MDB_cursor **)(txn->mt_dbs + env->me_maxdbs);
txn->mt_dbiseqs = (unsigned int *)(txn->mt_cursors + env->me_maxdbs);
MDB_PID_T pid = env->me_pid;
/* Clearing readers is done in this function because
* me_txkey with its destructor must be disabled first.
+ *
+ * We skip the the reader mutex, so we touch only
+ * data owned by this process (me_close_readers and
+ * our readers), and clear each reader atomically.
*/
- for (i = env->me_numreaders; --i >= 0; )
+ for (i = env->me_close_readers; --i >= 0; )
if (env->me_txns->mti_readers[i].mr_pid == pid)
env->me_txns->mti_readers[i].mr_pid = 0;
#ifdef _WIN32
*(size_t *)a->mv_data > *(size_t *)b->mv_data;
}
-/** Compare two items pointing at aligned unsigned int's */
+/** Compare two items pointing at aligned unsigned int's.
+ *
+ * This is also set as #MDB_INTEGERDUP|#MDB_DUPFIXED's #MDB_dbx.%md_dcmp,
+ * but #mdb_cmp_clong() is called instead if the data type is size_t.
+ */
static int
mdb_cmp_int(const MDB_val *a, const MDB_val *b)
{
#endif
}
-/** Compare two items pointing at size_t's of unknown alignment. */
-#ifdef MISALIGNED_OK
-# define mdb_cmp_clong mdb_cmp_long
-#else
-# define mdb_cmp_clong mdb_cmp_cint
-#endif
-
/** Compare two items lexically */
static int
mdb_cmp_memn(const MDB_val *a, const MDB_val *b)
MDB_page *p = NULL;
int level;
- if (!((txn->mt_flags & MDB_TXN_RDONLY) | (env->me_flags & MDB_WRITEMAP))) {
+ if (! (txn->mt_flags & (MDB_TXN_RDONLY|MDB_TXN_WRITEMAP))) {
MDB_txn *tx2 = txn;
level = 1;
do {
&mc->mc_dbx->md_name, &exact);
if (!exact)
return MDB_NOTFOUND;
+ if ((leaf->mn_flags & (F_DUPDATA|F_SUBDATA)) != F_SUBDATA)
+ return MDB_INCOMPATIBLE; /* not a named DB */
rc = mdb_node_read(mc->mc_txn, leaf, &data);
if (rc)
return rc;
if (leaf == NULL) {
DPUTS("===> inexact leaf not found, goto sibling");
- if ((rc = mdb_cursor_sibling(mc, 1)) != MDB_SUCCESS)
+ if ((rc = mdb_cursor_sibling(mc, 1)) != MDB_SUCCESS) {
+ mc->mc_flags |= C_EOF;
return rc; /* no entries matched */
+ }
mp = mc->mc_pg[mc->mc_top];
mdb_cassert(mc, IS_LEAF(mp));
leaf = NODEPTR(mp, 0);
return rc;
}
} else if (op == MDB_GET_BOTH || op == MDB_GET_BOTH_RANGE) {
- MDB_val d2;
- if ((rc = mdb_node_read(mc->mc_txn, leaf, &d2)) != MDB_SUCCESS)
+ MDB_val olddata;
+ MDB_cmp_func *dcmp;
+ if ((rc = mdb_node_read(mc->mc_txn, leaf, &olddata)) != MDB_SUCCESS)
return rc;
- rc = mc->mc_dbx->md_dcmp(data, &d2);
+ dcmp = mc->mc_dbx->md_dcmp;
+#if UINT_MAX < SIZE_MAX
+ if (dcmp == mdb_cmp_int && olddata.mv_size == sizeof(size_t))
+ dcmp = mdb_cmp_clong;
+#endif
+ rc = dcmp(data, &olddata);
if (rc) {
if (op == MDB_GET_BOTH || rc > 0)
return MDB_NOTFOUND;
rc = 0;
- *data = d2;
+ *data = olddata;
}
} else {
enum { MDB_NO_ROOT = MDB_LAST_ERRCODE+10 }; /* internal code */
MDB_env *env;
MDB_node *leaf = NULL;
- MDB_page *fp, *mp;
+ MDB_page *fp, *mp, *sub_root = NULL;
uint16_t fp_flags;
MDB_val xdata, *rdata, dkey, olddata;
MDB_db dummy;
/* Was a single item before, must convert now */
if (!F_ISSET(leaf->mn_flags, F_DUPDATA)) {
+ MDB_cmp_func *dcmp;
/* Just overwrite the current item */
if (flags == MDB_CURRENT)
goto current;
-
+ dcmp = mc->mc_dbx->md_dcmp;
#if UINT_MAX < SIZE_MAX
- if (mc->mc_dbx->md_dcmp == mdb_cmp_int && olddata.mv_size == sizeof(size_t))
- mc->mc_dbx->md_dcmp = mdb_cmp_clong;
+ if (dcmp == mdb_cmp_int && olddata.mv_size == sizeof(size_t))
+ dcmp = mdb_cmp_clong;
#endif
/* does data match? */
- if (!mc->mc_dbx->md_dcmp(data, &olddata)) {
+ if (!dcmp(data, &olddata)) {
if (flags & MDB_NODUPDATA)
return MDB_KEYEXIST;
/* overwrite it */
offset = env->me_psize - olddata.mv_size;
flags |= F_DUPDATA|F_SUBDATA;
dummy.md_root = mp->mp_pgno;
+ sub_root = mp;
}
if (mp != fp) {
mp->mp_flags = fp_flags | P_DIRTY;
goto new_sub;
}
current:
+ /* LMDB passes F_SUBDATA in 'flags' to write a DB record */
+ if ((leaf->mn_flags ^ flags) & F_SUBDATA)
+ return MDB_INCOMPATIBLE;
/* overflow page overwrites need special handling */
if (F_ISSET(leaf->mn_flags, F_BIGDATA)) {
MDB_page *omp;
* DB are all zero size.
*/
if (do_sub) {
- int xflags;
+ int xflags, new_dupdata;
size_t ecount;
put_sub:
xdata.mv_size = 0;
xflags = (flags & MDB_NODUPDATA) ?
MDB_NOOVERWRITE|MDB_NOSPILL : MDB_NOSPILL;
}
+ if (sub_root)
+ mc->mc_xcursor->mx_cursor.mc_pg[0] = sub_root;
+ new_dupdata = (int)dkey.mv_size;
/* converted, write the original data first */
if (dkey.mv_size) {
rc = mdb_cursor_put(&mc->mc_xcursor->mx_cursor, &dkey, &xdata, xflags);
if (rc)
goto bad_sub;
- {
- /* Adjust other cursors pointing to mp */
- MDB_cursor *m2;
- unsigned i = mc->mc_top;
- MDB_page *mp = mc->mc_pg[i];
-
- 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_flags & C_INITIALIZED)) continue;
- if (m2->mc_pg[i] == mp && m2->mc_ki[i] == mc->mc_ki[i]) {
- mdb_xcursor_init1(m2, leaf);
+ /* we've done our job */
+ dkey.mv_size = 0;
+ }
+ if (!(leaf->mn_flags & F_SUBDATA) || sub_root) {
+ /* Adjust other cursors pointing to mp */
+ MDB_cursor *m2;
+ MDB_xcursor *mx = mc->mc_xcursor;
+ unsigned i = mc->mc_top;
+ MDB_page *mp = mc->mc_pg[i];
+
+ 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_flags & C_INITIALIZED)) continue;
+ if (m2->mc_pg[i] == mp) {
+ if (m2->mc_ki[i] == mc->mc_ki[i]) {
+ mdb_xcursor_init2(m2, mx, new_dupdata);
+ } else if (!insert_key) {
+ MDB_node *n2 = NODEPTR(mp, m2->mc_ki[i]);
+ if (!(n2->mn_flags & F_SUBDATA))
+ m2->mc_xcursor->mx_cursor.mc_pg[0] = NODEDATA(n2);
}
}
}
- /* we've done our job */
- dkey.mv_size = 0;
}
ecount = mc->mc_xcursor->mx_db.md_entries;
if (flags & MDB_APPENDDUP)
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 */
+ /* fix other sub-DB cursors pointed at fake pages on this 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] == mp &&
- m2->mc_ki[mc->mc_top] == mc->mc_ki[mc->mc_top])
- m2->mc_xcursor->mx_cursor.mc_pg[0] = NODEDATA(leaf);
+ if (!(m2->mc_flags & C_INITIALIZED)) continue;
+ if (m2->mc_pg[mc->mc_top] == mp) {
+ if (m2->mc_ki[mc->mc_top] == mc->mc_ki[mc->mc_top]) {
+ m2->mc_xcursor->mx_cursor.mc_pg[0] = NODEDATA(leaf);
+ } else {
+ MDB_node *n2 = NODEPTR(mp, m2->mc_ki[mc->mc_top]);
+ if (!(n2->mn_flags & F_SUBDATA))
+ m2->mc_xcursor->mx_cursor.mc_pg[0] = NODEDATA(n2);
+ }
+ }
}
}
mc->mc_db->md_entries--;
goto fail;
}
}
+ /* LMDB passes F_SUBDATA in 'flags' to delete a DB record */
+ else if ((leaf->mn_flags ^ flags) & F_SUBDATA) {
+ rc = MDB_INCOMPATIBLE;
+ goto fail;
+ }
/* add overflow pages to free list */
if (F_ISSET(leaf->mn_flags, F_BIGDATA)) {
mx->mx_cursor.mc_flags = C_SUB;
} else {
MDB_page *fp = NODEDATA(node);
- mx->mx_db.md_pad = mc->mc_pg[mc->mc_top]->mp_pad;
+ mx->mx_db.md_pad = 0;
mx->mx_db.md_flags = 0;
mx->mx_db.md_depth = 1;
mx->mx_db.md_branch_pages = 0;
#endif
}
+
+/** Fixup a sorted-dups cursor due to underlying update.
+ * Sets up some fields that depend on the data from the main cursor.
+ * Almost the same as init1, but skips initialization steps if the
+ * xcursor had already been used.
+ * @param[in] mc The main cursor whose sorted-dups cursor is to be fixed up.
+ * @param[in] src_mx The xcursor of an up-to-date cursor.
+ * @param[in] new_dupdata True if converting from a non-#F_DUPDATA item.
+ */
+static void
+mdb_xcursor_init2(MDB_cursor *mc, MDB_xcursor *src_mx, int new_dupdata)
+{
+ MDB_xcursor *mx = mc->mc_xcursor;
+
+ if (new_dupdata) {
+ mx->mx_cursor.mc_snum = 1;
+ mx->mx_cursor.mc_top = 0;
+ mx->mx_cursor.mc_flags |= C_INITIALIZED;
+ mx->mx_cursor.mc_ki[0] = 0;
+ mx->mx_dbflag = DB_VALID|DB_USRVALID|DB_DIRTY; /* DB_DIRTY guides mdb_cursor_touch */
+#if UINT_MAX < SIZE_MAX
+ mx->mx_dbx.md_cmp = src_mx->mx_dbx.md_cmp;
+#endif
+ } else if (!(mx->mx_cursor.mc_flags & C_INITIALIZED)) {
+ return;
+ }
+ mx->mx_db = src_mx->mx_db;
+ mx->mx_cursor.mc_pg[0] = src_mx->mx_cursor.mc_pg[0];
+ DPRINTF(("Sub-db -%u root page %"Z"u", mx->mx_cursor.mc_dbi,
+ mx->mx_db.md_root));
+}
+
/** Initialize a cursor for a given transaction and database. */
static void
mdb_cursor_init(MDB_cursor *mc, MDB_txn *txn, MDB_dbi dbi, MDB_xcursor *mx)
/* Adjust other cursors pointing to mp */
MDB_cursor *m2, *m3;
MDB_dbi dbi = csrc->mc_dbi;
- MDB_page *mp = csrc->mc_pg[csrc->mc_top];
+ MDB_page *mp;
+ mp = cdst->mc_pg[csrc->mc_top];
+ 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;
+ else
+ m3 = m2;
+ if (m3 == cdst) continue;
+ if (m3->mc_pg[csrc->mc_top] == mp && m3->mc_ki[csrc->mc_top] >=
+ cdst->mc_ki[csrc->mc_top]) {
+ m3->mc_ki[csrc->mc_top]++;
+ }
+ }
+
+ mp = csrc->mc_pg[csrc->mc_top];
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;
uint16_t depth = cdst->mc_db->md_depth;
mdb_cursor_pop(cdst);
rc = mdb_rebalance(cdst);
- /* Did the tree shrink? */
- if (depth > cdst->mc_db->md_depth)
- snum--;
+ /* Did the tree height change? */
+ if (depth != cdst->mc_db->md_depth)
+ snum += cdst->mc_db->md_depth - depth;
cdst->mc_snum = snum;
cdst->mc_top = snum-1;
}
{
MDB_node *node;
int rc;
- unsigned int ptop, minkeys;
+ unsigned int ptop, minkeys, thresh;
MDB_cursor mn;
indx_t oldki;
- minkeys = 1 + (IS_BRANCH(mc->mc_pg[mc->mc_top]));
+ if (IS_BRANCH(mc->mc_pg[mc->mc_top])) {
+ minkeys = 2;
+ thresh = 1;
+ } else {
+ minkeys = 1;
+ thresh = FILL_THRESHOLD;
+ }
DPRINTF(("rebalancing %s page %"Z"u (has %u keys, %.1f%% full)",
IS_LEAF(mc->mc_pg[mc->mc_top]) ? "leaf" : "branch",
mdb_dbg_pgno(mc->mc_pg[mc->mc_top]), NUMKEYS(mc->mc_pg[mc->mc_top]),
(float)PAGEFILL(mc->mc_txn->mt_env, mc->mc_pg[mc->mc_top]) / 10));
- if (PAGEFILL(mc->mc_txn->mt_env, mc->mc_pg[mc->mc_top]) >= FILL_THRESHOLD &&
+ if (PAGEFILL(mc->mc_txn->mt_env, mc->mc_pg[mc->mc_top]) >= thresh &&
NUMKEYS(mc->mc_pg[mc->mc_top]) >= minkeys) {
DPRINTF(("no need to rebalance page %"Z"u, above fill threshold",
mdb_dbg_pgno(mc->mc_pg[mc->mc_top])));
* move one key from it. Otherwise we should try to merge them.
* (A branch page must never have less than 2 keys.)
*/
- minkeys = 1 + (IS_BRANCH(mn.mc_pg[mn.mc_top]));
- if (PAGEFILL(mc->mc_txn->mt_env, mn.mc_pg[mn.mc_top]) >= FILL_THRESHOLD && NUMKEYS(mn.mc_pg[mn.mc_top]) > minkeys) {
+ if (PAGEFILL(mc->mc_txn->mt_env, mn.mc_pg[mn.mc_top]) >= thresh && NUMKEYS(mn.mc_pg[mn.mc_top]) > minkeys) {
rc = mdb_node_move(&mn, mc);
- if (mc->mc_ki[ptop]) {
+ if (mc->mc_ki[mc->mc_top-1]) {
oldki++;
}
} else {
MDB_page *mp;
indx_t ki;
unsigned int nkeys;
+ MDB_cursor *m2, *m3;
+ MDB_dbi dbi = mc->mc_dbi;
ki = mc->mc_ki[mc->mc_top];
+ mp = mc->mc_pg[mc->mc_top];
mdb_node_del(mc, mc->mc_db->md_pad);
mc->mc_db->md_entries--;
+ {
+ /* Adjust other cursors pointing to mp */
+ for (m2 = mc->mc_txn->mt_cursors[dbi]; m2; m2=m2->mc_next) {
+ m3 = (mc->mc_flags & C_SUB) ? &m2->mc_xcursor->mx_cursor : m2;
+ if (! (m2->mc_flags & m3->mc_flags & C_INITIALIZED))
+ continue;
+ if (m3 == mc || m3->mc_snum < mc->mc_snum)
+ continue;
+ if (m3->mc_pg[mc->mc_top] == mp) {
+ if (m3->mc_ki[mc->mc_top] >= ki) {
+ m3->mc_flags |= C_DEL;
+ if (m3->mc_ki[mc->mc_top] > ki)
+ m3->mc_ki[mc->mc_top]--;
+ else if (mc->mc_db->md_flags & MDB_DUPSORT)
+ m3->mc_xcursor->mx_cursor.mc_flags |= C_EOF;
+ }
+ }
+ }
+ }
rc = mdb_rebalance(mc);
if (rc == MDB_SUCCESS) {
- MDB_cursor *m2, *m3;
- MDB_dbi dbi = mc->mc_dbi;
-
/* DB is totally empty now, just bail out.
* Other cursors adjustments were already done
* by mdb_rebalance and aren't needed here.
mp = mc->mc_pg[mc->mc_top];
nkeys = NUMKEYS(mp);
- /* if mc points past last node in page, find next sibling */
- if (mc->mc_ki[mc->mc_top] >= nkeys) {
- rc = mdb_cursor_sibling(mc, 1);
- if (rc == MDB_NOTFOUND) {
- mc->mc_flags |= C_EOF;
- rc = MDB_SUCCESS;
- }
- }
-
/* Adjust other cursors pointing to mp */
for (m2 = mc->mc_txn->mt_cursors[dbi]; !rc && m2; m2=m2->mc_next) {
m3 = (mc->mc_flags & C_SUB) ? &m2->mc_xcursor->mx_cursor : m2;
if (! (m2->mc_flags & m3->mc_flags & C_INITIALIZED))
continue;
- if (m3 == mc || m3->mc_snum < mc->mc_snum)
+ if (m3->mc_snum < mc->mc_snum)
continue;
if (m3->mc_pg[mc->mc_top] == mp) {
- if (m3->mc_ki[mc->mc_top] >= ki) {
- m3->mc_flags |= C_DEL;
- if (m3->mc_ki[mc->mc_top] > ki)
- m3->mc_ki[mc->mc_top]--;
- else if (mc->mc_db->md_flags & MDB_DUPSORT)
- m3->mc_xcursor->mx_cursor.mc_flags |= C_EOF;
- }
+ /* if m3 points past last node in page, find next sibling */
if (m3->mc_ki[mc->mc_top] >= nkeys) {
rc = mdb_cursor_sibling(m3, 1);
if (rc == MDB_NOTFOUND) {
/* Create a right sibling. */
if ((rc = mdb_page_new(mc, mp->mp_flags, 1, &rp)))
return rc;
+ rp->mp_pad = mp->mp_pad;
DPRINTF(("new right sibling: page %"Z"u", rp->mp_pgno));
if (mc->mc_snum < 2) {
mc->mc_ki[0] = 0;
mc->mc_db->md_root = pp->mp_pgno;
DPRINTF(("root split! new root = %"Z"u", pp->mp_pgno));
- mc->mc_db->md_depth++;
- new_root = 1;
+ new_root = mc->mc_db->md_depth++;
/* Add left (implicit) pointer. */
if ((rc = mdb_node_add(mc, 0, NULL, NULL, mp->mp_pgno, 0)) != MDB_SUCCESS) {
psize = 0;
if (newindx <= split_indx || newindx >= nkeys) {
i = 0; j = 1;
- k = newindx >= nkeys ? nkeys : split_indx+2;
+ k = newindx >= nkeys ? nkeys : split_indx+1+IS_LEAF(mp);
} else {
i = nkeys; j = -1;
k = split_indx-1;
if (new_root) {
int k;
/* root split */
- for (k=m3->mc_top; k>=0; k--) {
+ for (k=new_root; k>=0; k--) {
m3->mc_ki[k+1] = m3->mc_ki[k];
m3->mc_pg[k+1] = m3->mc_pg[k];
}
} mdb_copy;
/** Dedicated writer thread for compacting copy. */
-static THREAD_RET ESECT
+static THREAD_RET ESECT CALL_CONV
mdb_env_copythr(void *arg)
{
mdb_copy *my = arg;
arg->me_mapaddr = env->me_metas[toggle]->mm_address;
arg->me_mapsize = env->me_mapsize;
arg->me_maxreaders = env->me_maxreaders;
-
- /* me_numreaders may be zero if this process never used any readers. Use
- * the shared numreader count if it exists.
- */
- arg->me_numreaders = env->me_txns ? env->me_txns->mti_numreaders : env->me_numreaders;
+ arg->me_numreaders = env->me_txns ? env->me_txns->mti_numreaders : 0;
arg->me_last_pgno = env->me_metas[toggle]->mm_last_pg;
arg->me_last_txnid = env->me_metas[toggle]->mm_txnid;
if (rc == MDB_SUCCESS) {
/* make sure this is actually a DB */
MDB_node *node = NODEPTR(mc.mc_pg[mc.mc_top], mc.mc_ki[mc.mc_top]);
- if (!(node->mn_flags & F_SUBDATA))
+ if ((node->mn_flags & (F_DUPDATA|F_SUBDATA)) != F_SUBDATA)
return MDB_INCOMPATIBLE;
} else if (rc == MDB_NOTFOUND && (flags & MDB_CREATE)) {
/* Create if requested */
/* Can't delete the main DB */
if (del && dbi > MAIN_DBI) {
- rc = mdb_del0(txn, MAIN_DBI, &mc->mc_dbx->md_name, NULL, 0);
+ rc = mdb_del0(txn, MAIN_DBI, &mc->mc_dbx->md_name, NULL, F_SUBDATA);
if (!rc) {
txn->mt_dbflags[dbi] = DB_STALE;
mdb_dbi_close(txn->mt_env, dbi);