diff --git a/libraries/liblmdb/lmdb.h b/libraries/liblmdb/lmdb.h
index 5cf8bd75ec0c2d1220eb3641eb9f150705fe5ff3..280b43795c4793c1526f42d926fb58d98c8ff835 100644
--- a/libraries/liblmdb/lmdb.h
+++ b/libraries/liblmdb/lmdb.h
@@ -66,6 +66,7 @@
*
* - A thread can only use one transaction at a time, plus any child
* transactions. Each transaction belongs to one thread. See below.
+ * The #MDB_NOTLS flag changes this for read-only transactions.
*
* - Use an MDB_env* in the process which opened it, without fork()ing.
*
@@ -249,6 +250,8 @@ typedef void (MDB_rel_func)(MDB_val *item, void *oldptr, void *newptr, void *rel
#define MDB_WRITEMAP 0x80000
/** use asynchronous msync when MDB_WRITEMAP is used */
#define MDB_MAPASYNC 0x100000
+ /** tie reader locktable slots to #MDB_txn objects instead of to threads */
+#define MDB_NOTLS 0x200000
/** @} */
/** @defgroup mdb_dbi_open Database Flags
@@ -370,7 +373,9 @@ typedef enum MDB_cursor_op {
#define MDB_MAP_RESIZED (-30785)
/** Database flags changed or would change */
#define MDB_INCOMPATIBLE (-30784)
-#define MDB_LAST_ERRCODE MDB_INCOMPATIBLE
+ /** Invalid reuse of reader locktable slot */
+#define MDB_BAD_RSLOT (-30783)
+#define MDB_LAST_ERRCODE MDB_BAD_RSLOT
/** @} */
/** @brief Statistics for a database in the environment */
@@ -390,8 +395,8 @@ typedef struct MDB_envinfo {
size_t me_mapsize; /**< Size of the data memory map */
size_t me_last_pgno; /**< ID of the last used page */
size_t me_last_txnid; /**< ID of the last committed transaction */
- unsigned int me_maxreaders; /**< maximum number of threads for the environment */
- unsigned int me_numreaders; /**< maximum number of threads used in the environment */
+ unsigned int me_maxreaders; /**< max reader slots in the environment */
+ unsigned int me_numreaders; /**< max reader slots used in the environment */
} MDB_envinfo;
/** @brief Return the mdb library version information.
@@ -490,6 +495,15 @@ int mdb_env_create(MDB_env **env);
* database or lose the last transactions. Calling #mdb_env_sync()
* ensures on-disk database integrity until next commit.
* This flag may be changed at any time using #mdb_env_set_flags().
+ * <li>#MDB_NOTLS
+ * Don't use Thread-Local Storage. Tie reader locktable slots to
+ * #MDB_txn objects instead of to threads. I.e. #mdb_txn_reset() keeps
+ * the slot reseved for the #MDB_txn object. A thread may use parallel
+ * read-only transactions. A read-only transaction may span threads if
+ * the user synchronizes its use. Applications that multiplex many
+ * user threads over individual OS threads need this option. Such an
+ * application must also serialize the write transactions in an OS
+ * thread, since MDB's write locking is unaware of the user threads.
* </ul>
* @param[in] mode The UNIX permissions to set on created files. This parameter
* is ignored on Windows.
@@ -624,13 +638,17 @@ int mdb_env_get_path(MDB_env *env, const char **path);
*/
int mdb_env_set_mapsize(MDB_env *env, size_t size);
- /** @brief Set the maximum number of threads for the environment.
+ /** @brief Set the maximum number of threads/reader slots for the environment.
*
* This defines the number of slots in the lock table that is used to track readers in the
* the environment. The default is 126.
+ * Starting a read-only transaction normally ties a lock table slot to the
+ * current thread until the environment closes or the thread exits. If
+ * MDB_NOTLS is in use, #mdb_txn_begin() instead ties the slot to the
+ * MDB_txn object until it or the #MDB_env object is destroyed.
* This function may only be called after #mdb_env_create() and before #mdb_env_open().
* @param[in] env An environment handle returned by #mdb_env_create()
- * @param[in] readers The maximum number of threads
+ * @param[in] readers The maximum number of reader lock table slots
* @return A non-zero error value on failure and 0 on success. Some possible
* errors are:
* <ul>
@@ -639,7 +657,7 @@ int mdb_env_set_mapsize(MDB_env *env, size_t size);
*/
int mdb_env_set_maxreaders(MDB_env *env, unsigned int readers);
- /** @brief Get the maximum number of threads for the environment.
+ /** @brief Get the maximum number of threads/reader slots for the environment.
*
* @param[in] env An environment handle returned by #mdb_env_create()
* @param[out] readers Address of an integer to store the number of readers
@@ -670,8 +688,9 @@ int mdb_env_set_maxdbs(MDB_env *env, MDB_dbi dbs);
/** @brief Create a transaction for use with the environment.
*
* The transaction handle may be discarded using #mdb_txn_abort() or #mdb_txn_commit().
- * @note Transactions may not span threads; a transaction must only be used by a
- * single thread. Also, a thread may only have a single transaction.
+ * @note A transaction and its cursors must only be used by a single
+ * thread, and a thread may only have a single transaction at a time.
+ * If #MDB_NOTLS is in use, this does not apply to read-only transactions.
* @note Cursors may not span transactions; each cursor must be opened and closed
* within a single transaction.
* @param[in] env An environment handle returned by #mdb_env_create()
@@ -692,7 +711,7 @@ int mdb_env_set_maxdbs(MDB_env *env, MDB_dbi dbs);
* errors are:
* <ul>
* <li>#MDB_PANIC - a fatal error occurred earlier and the environment
-- * must be shut down.
+ * must be shut down.
* <li>#MDB_MAP_RESIZED - another process wrote data beyond this MDB_env's
* mapsize and the environment must be shut down.
* <li>#MDB_READERS_FULL - a read-only transaction was requested and
@@ -728,11 +747,13 @@ void mdb_txn_abort(MDB_txn *txn);
/** @brief Reset a read-only transaction.
*
- * This releases the current reader lock but doesn't free the
- * transaction handle, allowing it to be used again later by #mdb_txn_renew().
- * It otherwise has the same effect as #mdb_txn_abort() but saves some memory
- * allocation/deallocation overhead if a thread is going to start a new
- * read-only transaction again soon.
+ * Abort the transaction like #mdb_txn_abort(), but keep the transaction
+ * handle. #mdb_txn_renew() may reuse the handle. This saves allocation
+ * overhead if the process will start a new read-only transaction soon,
+ * and also locking overhead if #MDB_NOTLS is in use. The reader table
+ * lock is released, but the table slot stays tied to its thread or
+ * #MDB_txn. Use mdb_txn_abort() to discard a reset handle, and to free
+ * its lock table slot if MDB_NOTLS is in use.
* All cursors opened within the transaction must be closed before the transaction
* is reset.
* Reader locks generally don't interfere with writers, but they keep old
@@ -1043,8 +1064,8 @@ int mdb_del(MDB_txn *txn, MDB_dbi dbi, MDB_val *key, MDB_val *data);
/** @brief Create a cursor handle.
*
- * Cursors are associated with a specific transaction and database and
- * may not span threads.
+ * A cursor is associated with a specific transaction and database.
+ * It must be closed before its transaction ends.
* @param[in] txn A transaction handle returned by #mdb_txn_begin()
* @param[in] dbi A database handle returned by #mdb_dbi_open()
* @param[out] cursor Address where the new #MDB_cursor handle will be stored
@@ -1065,8 +1086,9 @@ void mdb_cursor_close(MDB_cursor *cursor);
/** @brief Renew a cursor handle.
*
- * Cursors are associated with a specific transaction and database and
- * may not span threads. Cursors that are only used in read-only
+ * A cursor is associated with a specific transaction and database.
+ * It must be closed before its transaction ends.
+ * Cursors that are only used in read-only
* transactions may be re-used, to avoid unnecessary malloc/free overhead.
* The cursor may be associated with a new read-only transaction, and
* referencing the same database handle as it was created with.
diff --git a/libraries/liblmdb/mdb.c b/libraries/liblmdb/mdb.c
index 8a2875cf089e8f4262a5cc31ce7005c7d414f1da..09fc3e29e8c9a85cb4850f5c172e800d14838904 100644
--- a/libraries/liblmdb/mdb.c
+++ b/libraries/liblmdb/mdb.c
@@ -406,6 +406,8 @@ typedef uint16_t indx_t;
* slot's address is saved in thread-specific data so that subsequent read
* transactions started by the same thread need no further locking to proceed.
*
+ * 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.
*
* Since the database uses multi-version concurrency control, readers don't
@@ -746,7 +748,8 @@ typedef struct MDB_db {
} MDB_db;
/** mdb_dbi_open flags */
-#define PERSISTENT_FLAGS 0x7fff
+#define MDB_VALID 0x8000 /**< DB handle is valid, for me_dbflags */
+#define PERSISTENT_FLAGS (0xffff & ~(MDB_VALID))
#define VALID_FLAGS (MDB_REVERSEKEY|MDB_DUPSORT|MDB_INTEGERKEY|MDB_DUPFIXED|\
MDB_INTEGERDUP|MDB_REVERSEDUP|MDB_CREATE)
@@ -830,8 +833,7 @@ struct MDB_txn {
#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_VALID 0x08 /**< DB handle is valid */
-#define MDB_VALID 0x8000 /**< DB handle is valid, for me_dbflags */
+#define DB_VALID 0x08 /**< DB handle is valid, see also #MDB_VALID */
/** @} */
/** In write txns, array of cursors for each DB */
MDB_cursor **mt_cursors;
@@ -934,10 +936,10 @@ struct MDB_env {
HANDLE me_mfd; /**< just for writing the meta pages */
/** Failed to update the meta page. Probably an I/O error. */
#define MDB_FATAL_ERROR 0x80000000U
- /** Read-only Filesystem. Allow read access, no locking. */
-#define MDB_ROFS 0x40000000U
/** Some fields are initialized. */
#define MDB_ENV_ACTIVE 0x20000000U
+ /** me_txkey is set */
+#define MDB_ENV_TXKEY 0x10000000U
uint32_t me_flags; /**< @ref mdb_env */
unsigned int me_psize; /**< size of a page, from #GET_PAGESIZE */
unsigned int me_maxreaders; /**< size of the reader table */
@@ -963,8 +965,8 @@ struct MDB_env {
MDB_page *me_dpages; /**< list of malloc'd blocks for re-use */
/** IDL of pages that became unused in a write txn */
MDB_IDL me_free_pgs;
- /** ID2L of pages that were written during a write txn */
- MDB_ID2 me_dirty_list[MDB_IDL_UM_SIZE];
+ /** ID2L of pages written during a write txn. Length MDB_IDL_UM_SIZE. */
+ MDB_ID2L me_dirty_list;
/** Max number of freelist items that can fit in a single overflow page */
unsigned int me_maxfree_1pg;
/** Max size of a node on a page */
@@ -1083,6 +1085,7 @@ static char *const mdb_errstr[] = {
"MDB_PAGE_FULL: Internal error - page has no more space",
"MDB_MAP_RESIZED: Database contents grew beyond environment mapsize",
"MDB_INCOMPATIBLE: Database flags changed or would change",
+ "MDB_BAD_RSLOT: Invalid reuse of reader locktable slot",
};
char *
@@ -1769,9 +1772,7 @@ mdb_txn_reset0(MDB_txn *txn);
/** Common code for #mdb_txn_begin() and #mdb_txn_renew().
* @param[in] txn the transaction handle to initialize
- * @return 0 on success, non-zero on failure. This can only
- * fail for read-only transactions, and then only if the
- * reader table is full.
+ * @return 0 on success, non-zero on failure.
*/
static int
mdb_txn_renew0(MDB_txn *txn)
@@ -1786,13 +1787,17 @@ mdb_txn_renew0(MDB_txn *txn)
txn->mt_dbxs = env->me_dbxs; /* mostly static anyway */
if (txn->mt_flags & MDB_TXN_RDONLY) {
- if (env->me_flags & MDB_ROFS) {
+ if (!env->me_txns) {
i = mdb_env_pick_meta(env);
txn->mt_txnid = env->me_metas[i]->mm_txnid;
txn->mt_u.reader = NULL;
} else {
- MDB_reader *r = pthread_getspecific(env->me_txkey);
- if (!r) {
+ MDB_reader *r = (env->me_flags & MDB_NOTLS) ? txn->mt_u.reader :
+ pthread_getspecific(env->me_txkey);
+ if (r) {
+ if (r->mr_pid != env->me_pid || r->mr_txnid != (txnid_t)-1)
+ return MDB_BAD_RSLOT;
+ } else {
pid_t pid = env->me_pid;
pthread_t tid = pthread_self();
@@ -1812,7 +1817,8 @@ mdb_txn_renew0(MDB_txn *txn)
env->me_numreaders = env->me_txns->mti_numreaders;
UNLOCK_MUTEX_R(env);
r = &env->me_txns->mti_readers[i];
- if ((rc = pthread_setspecific(env->me_txkey, r)) != 0) {
+ if (!(env->me_flags & MDB_NOTLS) &&
+ (rc = pthread_setspecific(env->me_txkey, r)) != 0) {
env->me_txns->mti_readers[i].mr_pid = 0;
return rc;
}
@@ -1844,7 +1850,8 @@ mdb_txn_renew0(MDB_txn *txn)
/* Copy the DB info and flags */
memcpy(txn->mt_dbs, env->me_metas[txn->mt_toggle]->mm_dbs, 2 * sizeof(MDB_db));
for (i=2; i<txn->mt_numdbs; i++) {
- txn->mt_dbs[i].md_flags = x = env->me_dbflags[i];
+ x = env->me_dbflags[i];
+ txn->mt_dbs[i].md_flags = x & PERSISTENT_FLAGS;
txn->mt_dbflags[i] = (x & MDB_VALID) ? DB_VALID|DB_STALE : 0;
}
txn->mt_dbflags[0] = txn->mt_dbflags[1] = DB_VALID;
@@ -1862,7 +1869,7 @@ mdb_txn_renew(MDB_txn *txn)
{
int rc;
- if (! (txn && (txn->mt_flags & MDB_TXN_RDONLY)))
+ if (!txn || txn->mt_numdbs || !(txn->mt_flags & MDB_TXN_RDONLY))
return EINVAL;
if (txn->mt_env->me_flags & MDB_FATAL_ERROR) {
@@ -1979,6 +1986,7 @@ mdb_txn_begin(MDB_env *env, MDB_txn *parent, unsigned int flags, MDB_txn **ret)
}
/** Common code for #mdb_txn_reset() and #mdb_txn_abort().
+ * May be called twice for readonly txns: First reset it, then abort.
* @param[in] txn the transaction handle to reset
*/
static void
@@ -1998,8 +2006,12 @@ mdb_txn_reset0(MDB_txn *txn)
}
if (F_ISSET(txn->mt_flags, MDB_TXN_RDONLY)) {
- if (!(env->me_flags & MDB_ROFS))
+ if (txn->mt_u.reader) {
txn->mt_u.reader->mr_txnid = (txnid_t)-1;
+ if (!(env->me_flags & MDB_NOTLS))
+ txn->mt_u.reader = NULL; /* txn does not own reader */
+ }
+ txn->mt_numdbs = 0; /* mark txn as reset, do not close DBs again */
} else {
MDB_page *dp;
@@ -2061,6 +2073,10 @@ mdb_txn_reset(MDB_txn *txn)
txn->mt_txnid, (txn->mt_flags & MDB_TXN_RDONLY) ? 'r' : 'w',
(void *) txn, (void *)txn->mt_env, txn->mt_dbs[MAIN_DBI].md_root);
+ /* This call is only valid for read-only txns */
+ if (!(txn->mt_flags & MDB_TXN_RDONLY))
+ return;
+
mdb_txn_reset0(txn);
}
@@ -2078,6 +2094,10 @@ mdb_txn_abort(MDB_txn *txn)
mdb_txn_abort(txn->mt_child);
mdb_txn_reset0(txn);
+ /* Free reader slot tied to this txn (if MDB_NOTLS && writable FS) */
+ if ((txn->mt_flags & MDB_TXN_RDONLY) && txn->mt_u.reader)
+ txn->mt_u.reader->mr_pid = 0;
+
free(txn);
}
@@ -2769,13 +2789,8 @@ mdb_env_create(MDB_env **env)
if (!e)
return ENOMEM;
- e->me_free_pgs = mdb_midl_alloc();
- if (!e->me_free_pgs) {
- free(e);
- return ENOMEM;
- }
e->me_maxreaders = DEFAULT_READERS;
- e->me_maxdbs = 2;
+ e->me_maxdbs = e->me_numdbs = 2;
e->me_fd = INVALID_HANDLE_VALUE;
e->me_lfd = INVALID_HANDLE_VALUE;
e->me_mfd = INVALID_HANDLE_VALUE;
@@ -3192,65 +3207,69 @@ mdb_hash_hex(MDB_val *val, char *hexbuf)
* @param[in] lpath The pathname of the file used for the lock region.
* @param[in] mode The Unix permissions for the file, if we create it.
* @param[out] excl Resulting file lock type: -1 none, 0 shared, 1 exclusive
+ * @param[in,out] excl In -1, out lock type: -1 none, 0 shared, 1 exclusive
* @return 0 on success, non-zero on failure.
*/
static int
mdb_env_setup_locks(MDB_env *env, char *lpath, int mode, int *excl)
{
+#ifdef _WIN32
+# define MDB_ERRCODE_ROFS ERROR_WRITE_PROTECT
+#else
+# define MDB_ERRCODE_ROFS EROFS
+#ifdef O_CLOEXEC /* Linux: Open file and set FD_CLOEXEC atomically */
+# define MDB_CLOEXEC O_CLOEXEC
+#else
+ int fdflags;
+# define MDB_CLOEXEC 0
+#endif
+#endif
int rc;
off_t size, rsize;
- *excl = -1;
-
#ifdef _WIN32
- if ((env->me_lfd = CreateFile(lpath, GENERIC_READ|GENERIC_WRITE,
+ env->me_lfd = CreateFile(lpath, GENERIC_READ|GENERIC_WRITE,
FILE_SHARE_READ|FILE_SHARE_WRITE, NULL, OPEN_ALWAYS,
- FILE_ATTRIBUTE_NORMAL, NULL)) == INVALID_HANDLE_VALUE) {
+ FILE_ATTRIBUTE_NORMAL, NULL);
+#else
+ env->me_lfd = open(lpath, O_RDWR|O_CREAT|MDB_CLOEXEC, mode);
+#endif
+ if (env->me_lfd == INVALID_HANDLE_VALUE) {
rc = ErrCode();
- if (rc == ERROR_WRITE_PROTECT && (env->me_flags & MDB_RDONLY)) {
- env->me_flags |= MDB_ROFS;
+ if (rc == MDB_ERRCODE_ROFS && (env->me_flags & MDB_RDONLY)) {
return MDB_SUCCESS;
}
goto fail_errno;
}
- /* Try to get exclusive lock. If we succeed, then
- * nobody is using the lock region and we should initialize it.
- */
- if ((rc = mdb_env_excl_lock(env, excl))) goto fail;
- size = GetFileSize(env->me_lfd, NULL);
-
-#else
-#if !(O_CLOEXEC)
- {
- int fdflags;
- if ((env->me_lfd = open(lpath, O_RDWR|O_CREAT, mode)) == -1) {
- rc = ErrCode();
- if (rc == EROFS && (env->me_flags & MDB_RDONLY)) {
- env->me_flags |= MDB_ROFS;
- return MDB_SUCCESS;
- }
- goto fail_errno;
- }
- /* Lose record locks when exec*() */
- if ((fdflags = fcntl(env->me_lfd, F_GETFD) | FD_CLOEXEC) >= 0)
+#if ! ((MDB_CLOEXEC) || defined(_WIN32))
+ /* Lose record locks when exec*() */
+ if ((fdflags = fcntl(env->me_lfd, F_GETFD) | FD_CLOEXEC) >= 0)
fcntl(env->me_lfd, F_SETFD, fdflags);
- }
-#else /* O_CLOEXEC on Linux: Open file and set FD_CLOEXEC atomically */
- if ((env->me_lfd = open(lpath, O_RDWR|O_CREAT|O_CLOEXEC, mode)) == -1) {
- rc = ErrCode();
- if (rc == EROFS && (env->me_flags & MDB_RDONLY)) {
- env->me_flags |= MDB_ROFS;
- return MDB_SUCCESS;
+#endif
+
+ if (!(env->me_flags & MDB_NOTLS)) {
+ rc = pthread_key_create(&env->me_txkey, mdb_env_reader_dest);
+ if (rc)
+ goto fail;
+ env->me_flags |= MDB_ENV_TXKEY;
+#ifdef _WIN32
+ /* Windows TLS callbacks need help finding their TLS info. */
+ if (mdb_tls_nkeys >= MAX_TLS_KEYS) {
+ rc = MDB_TLS_FULL;
+ goto fail;
}
- goto fail_errno;
- }
+ mdb_tls_keys[mdb_tls_nkeys++] = env->me_txkey;
#endif
+ }
/* Try to get exclusive lock. If we succeed, then
* nobody is using the lock region and we should initialize it.
*/
if ((rc = mdb_env_excl_lock(env, excl))) goto fail;
+#ifdef _WIN32
+ size = GetFileSize(env->me_lfd, NULL);
+#else
size = lseek(env->me_lfd, 0, SEEK_END);
#endif
rsize = (env->me_maxreaders-1) * sizeof(MDB_reader) + sizeof(MDB_txninfo);
@@ -3406,12 +3425,12 @@ fail:
* 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)
+#define CHANGELESS (MDB_FIXEDMAP|MDB_NOSUBDIR|MDB_RDONLY|MDB_WRITEMAP|MDB_NOTLS)
int
mdb_env_open(MDB_env *env, const char *path, unsigned int flags, mdb_mode_t mode)
{
- int oflags, rc, len, excl;
+ int oflags, rc, len, excl = -1;
char *lpath, *dpath;
if (env->me_fd!=INVALID_HANDLE_VALUE || (flags & ~(CHANGEABLE|CHANGELESS)))
@@ -3436,11 +3455,27 @@ mdb_env_open(MDB_env *env, const char *path, unsigned int flags, mdb_mode_t mode
sprintf(dpath, "%s" DATANAME, path);
}
+ rc = MDB_SUCCESS;
flags |= env->me_flags;
- /* silently ignore WRITEMAP if we're only getting read access */
- if (F_ISSET(flags, MDB_RDONLY|MDB_WRITEMAP))
- flags ^= MDB_WRITEMAP;
+ if (flags & MDB_RDONLY) {
+ /* silently ignore WRITEMAP when we're only getting read access */
+ flags &= ~MDB_WRITEMAP;
+ } else {
+ if (!((env->me_free_pgs = mdb_midl_alloc()) &&
+ (env->me_dirty_list = calloc(MDB_IDL_UM_SIZE, sizeof(MDB_ID2)))))
+ rc = ENOMEM;
+ }
env->me_flags = flags |= MDB_ENV_ACTIVE;
+ if (rc)
+ goto leave;
+
+ env->me_path = strdup(path);
+ env->me_dbxs = calloc(env->me_maxdbs, sizeof(MDB_dbx));
+ env->me_dbflags = calloc(env->me_maxdbs, sizeof(uint16_t));
+ if (!(env->me_dbxs && env->me_path && env->me_dbflags)) {
+ rc = ENOMEM;
+ goto leave;
+ }
rc = mdb_env_setup_locks(env, lpath, mode, &excl);
if (rc)
@@ -3490,29 +3525,9 @@ mdb_env_open(MDB_env *env, const char *path, unsigned int flags, mdb_mode_t mode
}
}
DPRINTF("opened dbenv %p", (void *) env);
- rc = pthread_key_create(&env->me_txkey, mdb_env_reader_dest);
- if (rc)
- goto leave;
- env->me_numdbs = 2; /* this notes that me_txkey was set */
-#ifdef _WIN32
- /* Windows TLS callbacks need help finding their TLS info. */
- if (mdb_tls_nkeys < MAX_TLS_KEYS)
- mdb_tls_keys[mdb_tls_nkeys++] = env->me_txkey;
- else {
- rc = MDB_TLS_FULL;
- goto leave;
- }
-#endif
if (excl > 0) {
rc = mdb_env_share_locks(env, &excl);
- if (rc)
- goto leave;
}
- env->me_dbxs = calloc(env->me_maxdbs, sizeof(MDB_dbx));
- env->me_dbflags = calloc(env->me_maxdbs, sizeof(uint16_t));
- env->me_path = strdup(path);
- if (!env->me_dbxs || !env->me_dbflags || !env->me_path)
- rc = ENOMEM;
}
leave:
@@ -3535,8 +3550,11 @@ mdb_env_close0(MDB_env *env, int excl)
free(env->me_dbflags);
free(env->me_dbxs);
free(env->me_path);
+ free(env->me_dirty_list);
+ if (env->me_free_pgs)
+ mdb_midl_free(env->me_free_pgs);
- if (env->me_numdbs) {
+ if (env->me_flags & MDB_ENV_TXKEY) {
pthread_key_delete(env->me_txkey);
#ifdef _WIN32
/* Delete our key from the global list */
@@ -3602,7 +3620,7 @@ mdb_env_close0(MDB_env *env, int excl)
close(env->me_lfd);
}
- env->me_flags &= ~MDB_ENV_ACTIVE;
+ env->me_flags &= ~(MDB_ENV_ACTIVE|MDB_ENV_TXKEY);
}
int
@@ -3661,7 +3679,7 @@ mdb_env_copy(MDB_env *env, const char *path)
if (rc)
goto leave;
- if (!(env->me_flags & MDB_ROFS)) {
+ if (env->me_txns) {
/* We must start the actual read txn after blocking writers */
mdb_txn_reset0(txn);
@@ -3686,7 +3704,7 @@ mdb_env_copy(MDB_env *env, const char *path)
rc = write(newfd, env->me_map, wsize);
rc = (rc == (int)wsize) ? MDB_SUCCESS : ErrCode();
#endif
- if (! (env->me_flags & MDB_ROFS))
+ if (env->me_txns)
UNLOCK_MUTEX_W(env);
if (rc)
@@ -3752,7 +3770,6 @@ mdb_env_close(MDB_env *env)
}
mdb_env_close0(env, 0);
- mdb_midl_free(env->me_free_pgs);
free(env);
}
@@ -4111,6 +4128,28 @@ mdb_page_search_root(MDB_cursor *mc, MDB_val *key, int modify)
return MDB_SUCCESS;
}
+/** Search for the lowest key under the current branch page.
+ * This just bypasses a NUMKEYS check in the current page
+ * before calling mdb_page_search_root(), because the callers
+ * are all in situations where the current page is known to
+ * be underfilled.
+ */
+static int
+mdb_page_search_lowest(MDB_cursor *mc)
+{
+ MDB_page *mp = mc->mc_pg[mc->mc_top];
+ MDB_node *node = NODEPTR(mp, 0);
+ int rc;
+
+ if ((rc = mdb_page_get(mc->mc_txn, NODEPGNO(node), &mp)))
+ return rc;
+
+ mc->mc_ki[mc->mc_top] = 0;
+ if ((rc = mdb_cursor_push(mc, mp)))
+ return rc;
+ return mdb_page_search_root(mc, NULL, 0);
+}
+
/** 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
@@ -4688,19 +4727,19 @@ mdb_cursor_last(MDB_cursor *mc, MDB_val *key, MDB_val *data)
if (!(mc->mc_flags & C_EOF)) {
- if (!(mc->mc_flags & C_INITIALIZED) || mc->mc_top) {
- MDB_val lkey;
+ 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);
- if (rc != MDB_SUCCESS)
- return rc;
- }
- assert(IS_LEAF(mc->mc_pg[mc->mc_top]));
+ lkey.mv_size = MDB_MAXKEYSIZE+1;
+ lkey.mv_data = NULL;
+ rc = mdb_page_search(mc, &lkey, 0);
+ if (rc != MDB_SUCCESS)
+ return rc;
+ }
+ assert(IS_LEAF(mc->mc_pg[mc->mc_top]));
- mc->mc_ki[mc->mc_top] = NUMKEYS(mc->mc_pg[mc->mc_top]) - 1;
}
+ mc->mc_ki[mc->mc_top] = NUMKEYS(mc->mc_pg[mc->mc_top]) - 1;
mc->mc_flags |= C_INITIALIZED|C_EOF;
leaf = NODEPTR(mc->mc_pg[mc->mc_top], mc->mc_ki[mc->mc_top]);
@@ -5823,13 +5862,19 @@ mdb_cursor_open(MDB_txn *txn, MDB_dbi dbi, MDB_cursor **ret)
int
mdb_cursor_renew(MDB_txn *txn, MDB_cursor *mc)
{
+ unsigned flags;
+
if (txn == NULL || mc == NULL || mc->mc_dbi >= txn->mt_numdbs)
return EINVAL;
if (txn->mt_cursors)
return EINVAL;
+ flags = mc->mc_flags;
+
mdb_cursor_init(mc, txn, mc->mc_dbi, mc->mc_xcursor);
+
+ mc->mc_flags |= (flags & C_ALLOCD);
return MDB_SUCCESS;
}
@@ -6001,7 +6046,7 @@ mdb_node_move(MDB_cursor *csrc, MDB_cursor *cdst)
unsigned int snum = csrc->mc_snum;
MDB_node *s2;
/* must find the lowest key below src */
- mdb_page_search_root(csrc, NULL, 0);
+ mdb_page_search_lowest(csrc);
if (IS_LEAF2(csrc->mc_pg[csrc->mc_top])) {
key.mv_size = csrc->mc_db->md_pad;
key.mv_data = LEAF2KEY(csrc->mc_pg[csrc->mc_top], 0, key.mv_size);
@@ -6024,7 +6069,7 @@ mdb_node_move(MDB_cursor *csrc, MDB_cursor *cdst)
MDB_node *s2;
MDB_val bkey;
/* must find the lowest key below dst */
- mdb_page_search_root(cdst, NULL, 0);
+ mdb_page_search_lowest(cdst);
if (IS_LEAF2(cdst->mc_pg[cdst->mc_top])) {
bkey.mv_size = cdst->mc_db->md_pad;
bkey.mv_data = LEAF2KEY(cdst->mc_pg[cdst->mc_top], 0, bkey.mv_size);
@@ -6188,7 +6233,7 @@ mdb_page_merge(MDB_cursor *csrc, MDB_cursor *cdst)
unsigned int snum = csrc->mc_snum;
MDB_node *s2;
/* must find the lowest key below src */
- mdb_page_search_root(csrc, NULL, 0);
+ mdb_page_search_lowest(csrc);
if (IS_LEAF2(csrc->mc_pg[csrc->mc_top])) {
key.mv_size = csrc->mc_db->md_pad;
key.mv_data = LEAF2KEY(csrc->mc_pg[csrc->mc_top], 0, key.mv_size);
@@ -6295,6 +6340,7 @@ mdb_rebalance(MDB_cursor *mc)
unsigned int ptop, minkeys;
MDB_cursor mn;
+ minkeys = 1 + (IS_BRANCH(mc->mc_pg[mc->mc_top]));
#if MDB_DEBUG
{
pgno_t pgno;
@@ -6305,7 +6351,8 @@ mdb_rebalance(MDB_cursor *mc)
}
#endif
- 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]) >= FILL_THRESHOLD &&
+ NUMKEYS(mc->mc_pg[mc->mc_top]) >= minkeys) {
#if MDB_DEBUG
pgno_t pgno;
COPY_PGNO(pgno, mc->mc_pg[mc->mc_top]->mp_pgno);
@@ -6317,6 +6364,10 @@ mdb_rebalance(MDB_cursor *mc)
if (mc->mc_snum < 2) {
MDB_page *mp = mc->mc_pg[0];
+ if (IS_SUBP(mp)) {
+ DPUTS("Can't rebalance a subpage, ignoring");
+ return MDB_SUCCESS;
+ }
if (NUMKEYS(mp) == 0) {
DPUTS("tree is completely empty");
mc->mc_db->md_root = P_INVALID;