* BerkeleyDB API, but much simplified.
*/
/*
- * Copyright 2011-2014 Howard Chu, Symas Corp.
+ * Copyright 2011-2015 Howard Chu, Symas Corp.
* All rights reserved.
*
* Redistribution and use in source and binary forms, with or without
#define CACHEFLUSH(addr, bytes, cache)
#endif
+#if defined(__linux) && !defined(MDB_FDATASYNC_WORKS)
+/** fdatasync is broken on ext3/ext4fs on older kernels, see
+ * description in #mdb_env_open2 comments. You can safely
+ * define MDB_FDATASYNC_WORKS if this code will only be run
+ * on kernels 3.6 and newer.
+ */
+#define BROKEN_FDATASYNC
+#endif
#include <errno.h>
#include <limits.h>
/** 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.
+ *
+ * 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.
*
- * 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.
+ * 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
* started from so we can avoid overwriting any data used in that
* particular version.
*/
- txnid_t mrb_txnid;
+ volatile txnid_t mrb_txnid;
/** The process ID of the process owning this reader txn. */
- MDB_PID_T mrb_pid;
+ volatile MDB_PID_T mrb_pid;
/** The thread ID of the thread owning this txn. */
- MDB_THR_T mrb_tid;
+ volatile MDB_THR_T mrb_tid;
} MDB_rxbody;
/** The actual reader record, with cacheline padding. */
* This is recorded here only for convenience; the value can always
* be determined by reading the main database meta pages.
*/
- txnid_t mtb_txnid;
+ volatile txnid_t mtb_txnid;
/** The number of slots that have been used in the reader table.
* This always records the maximum count, it is not decremented
* when readers release their slots.
*/
- unsigned mtb_numreaders;
+ volatile unsigned mtb_numreaders;
} MDB_txbody;
/** The actual reader table definition. */
/** Stamp identifying this as an LMDB file. It must be set
* to #MDB_MAGIC. */
uint32_t mm_magic;
- /** Version number of this lock file. Must be set to #MDB_DATA_VERSION. */
+ /** Version number of this file. Must be set to #MDB_DATA_VERSION. */
uint32_t mm_version;
void *mm_address; /**< address for fixed mapping */
size_t mm_mapsize; /**< size of mmap region */
/** Any persistent environment flags. @ref mdb_env */
#define mm_flags mm_dbs[0].md_flags
pgno_t mm_last_pg; /**< last used page in file */
- txnid_t mm_txnid; /**< txnid that committed this page */
+ volatile txnid_t mm_txnid; /**< txnid that committed this page */
} MDB_meta;
/** Buffer for a stack-allocated meta page.
#define MDB_ENV_ACTIVE 0x20000000U
/** me_txkey is set */
#define MDB_ENV_TXKEY 0x10000000U
+ /** fdatasync is unreliable */
+#define MDB_FSYNCONLY 0x08000000U
uint32_t me_flags; /**< @ref mdb_env */
unsigned int me_psize; /**< DB page size, inited from me_os_psize */
unsigned int me_os_psize; /**< OS page size, from #GET_PAGESIZE */
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;
buf[0] = 0;
FormatMessage(FORMAT_MESSAGE_FROM_SYSTEM |
FORMAT_MESSAGE_IGNORE_INSERTS,
- NULL, err, 0, ptr, sizeof(buf), pad);
+ NULL, err, 0, ptr, sizeof(buf), (va_list *)pad);
return ptr;
#else
return strerror(err);
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_env_sync(MDB_env *env, int force)
{
int rc = 0;
+ if (env->me_flags & MDB_RDONLY)
+ return EACCES;
if (force || !F_ISSET(env->me_flags, MDB_NOSYNC)) {
if (env->me_flags & MDB_WRITEMAP) {
int flags = ((env->me_flags & MDB_MAPASYNC) && !force)
rc = ErrCode();
#endif
} else {
+#ifdef BROKEN_FDATASYNC
+ if (env->me_flags & MDB_FSYNCONLY) {
+ if (fsync(env->me_fd))
+ rc = ErrCode();
+ } else
+#endif
if (MDB_FDATASYNC(env->me_fd))
rc = ErrCode();
}
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) {
- /* 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;
return rc;
}
}
- txn->mt_txnid = r->mr_txnid = ti->mti_txnid;
+ do /* LY: Retry on a race, ITS#7970. */
+ r->mr_txnid = ti->mti_txnid;
+ while(r->mr_txnid != ti->mti_txnid);
+ txn->mt_txnid = r->mr_txnid;
txn->mt_u.reader = r;
meta = env->me_metas[txn->mt_txnid & 1];
}
+ txn->mt_dbxs = env->me_dbxs; /* mostly static anyway */
} else {
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;
}
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;
/* Write up to MDB_COMMIT_PAGES dirty pages at a time. */
if (pos!=next_pos || n==MDB_COMMIT_PAGES || wsize+size>MAX_WRITE) {
if (n) {
+retry_write:
/* Write previous page(s) */
#ifdef MDB_USE_PWRITEV
wres = pwritev(env->me_fd, iov, n, wpos);
if (n == 1) {
wres = pwrite(env->me_fd, iov[0].iov_base, wsize, wpos);
} else {
+retry_seek:
if (lseek(env->me_fd, wpos, SEEK_SET) == -1) {
rc = ErrCode();
+ if (rc == EINTR)
+ goto retry_seek;
DPRINTF(("lseek: %s", strerror(rc)));
return rc;
}
if (wres != wsize) {
if (wres < 0) {
rc = ErrCode();
+ if (rc == EINTR)
+ goto retry_write;
DPRINTF(("Write error: %s", strerror(rc)));
} else {
rc = EIO; /* TODO: Use which error code? */
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);
int len;
#define DO_PWRITE(rc, fd, ptr, size, len, pos) do { \
len = pwrite(fd, ptr, size, pos); \
- rc = (len >= 0); } while(0)
+ if (len == -1 && ErrCode() == EINTR) continue; \
+ rc = (len >= 0); break; } while(1)
#endif
DPUTS("writing new meta page");
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;
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.
static int ESECT
mdb_fsize(HANDLE fd, size_t *size)
{
-#ifdef WIN32
+#ifdef _WIN32
LARGE_INTEGER fsize;
if (!GetFileSizeEx(fd, &fsize))
return MDB_SUCCESS;
}
+#ifdef BROKEN_FDATASYNC
+#include <sys/utsname.h>
+#include <sys/vfs.h>
+#endif
+
/** Further setup required for opening an LMDB environment
*/
static int ESECT
else
env->me_pidquery = PROCESS_QUERY_INFORMATION;
#endif /* _WIN32 */
+#ifdef BROKEN_FDATASYNC
+ /* ext3/ext4 fdatasync is broken on some older Linux kernels.
+ * https://lkml.org/lkml/2012/9/3/83
+ * Kernels after 3.6-rc6 are known good.
+ * https://lkml.org/lkml/2012/9/10/556
+ * See if the DB is on ext3/ext4, then check for new enough kernel
+ * Kernels 2.6.32.60, 2.6.34.15, 3.2.30, and 3.5.4 are also known
+ * to be patched.
+ */
+ {
+ struct statfs st;
+ fstatfs(env->me_fd, &st);
+ while (st.f_type == 0xEF53) {
+ struct utsname uts;
+ int i;
+ uname(&uts);
+ if (uts.release[0] < '3') {
+ if (!strncmp(uts.release, "2.6.32.", 7)) {
+ i = atoi(uts.release+7);
+ if (i >= 60)
+ break; /* 2.6.32.60 and newer is OK */
+ } else if (!strncmp(uts.release, "2.6.34.", 7)) {
+ i = atoi(uts.release+7);
+ if (i >= 15)
+ break; /* 2.6.34.15 and newer is OK */
+ }
+ } else if (uts.release[0] == '3') {
+ i = atoi(uts.release+2);
+ if (i > 5)
+ break; /* 3.6 and newer is OK */
+ if (i == 5) {
+ i = atoi(uts.release+4);
+ if (i >= 4)
+ break; /* 3.5.4 and newer is OK */
+ } else if (i == 2) {
+ i = atoi(uts.release+4);
+ if (i >= 30)
+ break; /* 3.2.30 and newer is OK */
+ }
+ } else { /* 4.x and newer is OK */
+ break;
+ }
+ env->me_flags |= MDB_FSYNCONLY;
+ break;
+ }
+ }
+#endif
memset(&meta, 0, sizeof(meta));
extern const PIMAGE_TLS_CALLBACK mdb_tls_cbp;
const PIMAGE_TLS_CALLBACK mdb_tls_cbp = mdb_tls_callback;
#pragma const_seg()
-#else /* WIN32 */
+#else /* _WIN32 */
#pragma comment(linker, "/INCLUDE:__tls_used")
#pragma comment(linker, "/INCLUDE:_mdb_tls_cbp")
#pragma data_seg(".CRT$XLB")
return rc;
}
-/** Try to get exlusive lock, otherwise shared.
+/** Try to get exclusive lock, otherwise shared.
* Maintain *excl = -1: no/unknown lock, 0: shared, 1: exclusive.
*/
static int ESECT
* @param[in] env The LMDB environment.
* @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.
*/
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);
+ (sizeof(MDB_db)+sizeof(MDB_cursor *)+sizeof(unsigned int)+1);
txn = calloc(1, size);
if (txn) {
txn->mt_dbs = (MDB_db *)((char *)txn + tsize);
return;
/* Doing this here since me_dbxs may not exist during mdb_env_close */
- for (i = env->me_maxdbs; --i > MAIN_DBI; )
- free(env->me_dbxs[i].md_name.mv_data);
+ if (env->me_dbxs) {
+ for (i = env->me_maxdbs; --i > MAIN_DBI; )
+ free(env->me_dbxs[i].md_name.mv_data);
+ free(env->me_dbxs);
+ }
free(env->me_pbuf);
free(env->me_dbiseqs);
free(env->me_dbflags);
- free(env->me_dbxs);
free(env->me_path);
free(env->me_dirty_list);
free(env->me_txn0);
*(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)
&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 && m2->mc_ki[i] == mc->mc_ki[i]) {
+ mdb_xcursor_init2(m2, mx, new_dupdata);
+ }
+ }
+ }
ecount = mc->mc_xcursor->mx_db.md_entries;
if (flags & MDB_APPENDDUP)
xflags |= MDB_APPEND;
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)) {
#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)
mc->mc_snum = 0;
mc->mc_top = 0;
mc->mc_pg[0] = 0;
+ mc->mc_ki[0] = 0;
mc->mc_flags = 0;
if (txn->mt_dbs[dbi].md_flags & MDB_DUPSORT) {
mdb_tassert(txn, mx != NULL);
/* 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;
cdst->mc_ki[cdst->mc_top] = 0;
rc = mdb_update_key(cdst, &nullkey);
cdst->mc_ki[cdst->mc_top] = ix;
- mdb_cassert(csrc, rc == MDB_SUCCESS);
+ mdb_cassert(cdst, rc == MDB_SUCCESS);
}
}
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])));
m3 = m2;
if (m3 == mc || m3->mc_snum < mc->mc_snum) continue;
if (m3->mc_pg[0] == mp) {
- 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];
}
+ m3->mc_snum--;
+ m3->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 {
if (mc->mc_ki[ptop] == 0) {
rc = mdb_page_merge(&mn, mc);
} else {
+ MDB_cursor dummy;
oldki += NUMKEYS(mn.mc_pg[mn.mc_top]);
mn.mc_ki[mn.mc_top] += mc->mc_ki[mn.mc_top] + 1;
+ /* We want mdb_rebalance to find mn when doing fixups */
+ if (mc->mc_flags & C_SUB) {
+ dummy.mc_next = mc->mc_txn->mt_cursors[mc->mc_dbi];
+ mc->mc_txn->mt_cursors[mc->mc_dbi] = &dummy;
+ dummy.mc_xcursor = (MDB_xcursor *)&mn;
+ } else {
+ mn.mc_next = mc->mc_txn->mt_cursors[mc->mc_dbi];
+ mc->mc_txn->mt_cursors[mc->mc_dbi] = &mn;
+ }
rc = mdb_page_merge(mc, &mn);
+ if (mc->mc_flags & C_SUB)
+ mc->mc_txn->mt_cursors[mc->mc_dbi] = dummy.mc_next;
+ else
+ mc->mc_txn->mt_cursors[mc->mc_dbi] = mn.mc_next;
mdb_cursor_copy(&mn, mc);
}
mc->mc_flags &= ~C_EOF;
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--;
- rc = mdb_rebalance(mc);
-
- if (rc == MDB_SUCCESS) {
- MDB_cursor *m2, *m3;
- MDB_dbi dbi = mc->mc_dbi;
-
- 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) {
+ 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;
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) {
+ /* DB is totally empty now, just bail out.
+ * Other cursors adjustments were already done
+ * by mdb_rebalance and aren't needed here.
+ */
+ if (!mc->mc_snum)
+ return rc;
+
+ mp = mc->mc_pg[mc->mc_top];
+ nkeys = NUMKEYS(mp);
+
+ /* 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_snum < mc->mc_snum)
+ continue;
+ if (m3->mc_pg[mc->mc_top] == mp) {
+ /* 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) {
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];
}
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);
projects / openldap / blobdiff