*/
#include <sys/types.h>
#include <sys/stat.h>
-#include <sys/queue.h>
#include <sys/param.h>
#include <sys/uio.h>
#include <sys/mman.h>
#include <fcntl.h>
#include <assert.h>
-#include <err.h>
#include <errno.h>
-#include <fcntl.h>
#include <stddef.h>
#include <stdint.h>
#include <stdio.h>
#include <time.h>
#include <unistd.h>
#include <pthread.h>
-#include <endian.h>
#include "mdb.h"
#define ULONG unsigned long
typedef ULONG pgno_t;
-#include "idl.h"
+#include "midl.h"
+
+/* Note: If O_DSYNC is undefined but exists in /usr/include,
+ * preferably set some compiler flag to get the definition.
+ * Otherwise compile with the less efficient -DMDB_DSYNC=O_SYNC.
+ */
+#ifndef MDB_DSYNC
+# define MDB_DSYNC O_DSYNC
+#endif
#ifndef DEBUG
#define DEBUG 1
#endif
-#if DEBUG && defined(__GNUC__)
-# define DPRINTF(fmt, ...) \
- fprintf(stderr, "%s:%d: " fmt "\n", __func__, __LINE__, ##__VA_ARGS__)
+#if !(__STDC_VERSION__ >= 199901L || defined(__GNUC__))
+# define DPRINTF (void) /* Vararg macros may be unsupported */
+#elif DEBUG
+# define DPRINTF(fmt, ...) /* Requires 2 or more args */ \
+ fprintf(stderr, "%s:%d: " fmt "\n", __func__, __LINE__, __VA_ARGS__)
#else
-# define DPRINTF(...) ((void) 0)
+# define DPRINTF(fmt, ...) ((void) 0)
#endif
+#define DPUTS(arg) DPRINTF("%s", arg)
#define PAGESIZE 4096
#define MDB_MINKEYS 4
#define DEFAULT_MAPSIZE 1048576
/* Lock descriptor stuff */
-#define RXBODY \
- ULONG mr_txnid; \
- pid_t mr_pid; \
- pthread_t mr_tid
-typedef struct MDB_rxbody {
- RXBODY;
-} MDB_rxbody;
-
#ifndef CACHELINE
#define CACHELINE 64 /* most CPUs. Itanium uses 128 */
#endif
+typedef struct MDB_rxbody {
+ ULONG mrb_txnid;
+ pid_t mrb_pid;
+ pthread_t mrb_tid;
+} MDB_rxbody;
+
typedef struct MDB_reader {
- RXBODY;
- /* cache line alignment */
- char pad[CACHELINE-sizeof(MDB_rxbody)];
+ union {
+ MDB_rxbody mrx;
+#define mr_txnid mru.mrx.mrb_txnid
+#define mr_pid mru.mrx.mrb_pid
+#define mr_tid mru.mrx.mrb_tid
+ /* cache line alignment */
+ char pad[(sizeof(MDB_rxbody)+CACHELINE-1) & ~(CACHELINE-1)];
+ } mru;
} MDB_reader;
-#define TXBODY \
- uint32_t mt_magic; \
- uint32_t mt_version; \
- pthread_mutex_t mt_mutex; \
- ULONG mt_txnid; \
- uint32_t mt_numreaders
typedef struct MDB_txbody {
- TXBODY;
+ uint32_t mtb_magic;
+ uint32_t mtb_version;
+ pthread_mutex_t mtb_mutex;
+ ULONG mtb_txnid;
+ uint32_t mtb_numreaders;
+ uint32_t mtb_me_toggle;
} MDB_txbody;
typedef struct MDB_txninfo {
- TXBODY;
- char pad[CACHELINE-sizeof(MDB_txbody)];
- pthread_mutex_t mt_wmutex;
- char pad2[CACHELINE-sizeof(pthread_mutex_t)];
- MDB_reader mt_readers[1];
+ union {
+ MDB_txbody mtb;
+#define mti_magic mt1.mtb.mtb_magic
+#define mti_version mt1.mtb.mtb_version
+#define mti_mutex mt1.mtb.mtb_mutex
+#define mti_txnid mt1.mtb.mtb_txnid
+#define mti_numreaders mt1.mtb.mtb_numreaders
+#define mti_me_toggle mt1.mtb.mtb_me_toggle
+ char pad[(sizeof(MDB_txbody)+CACHELINE-1) & ~(CACHELINE-1)];
+ } mt1;
+ union {
+ pthread_mutex_t mt2_wmutex;
+#define mti_wmutex mt2.mt2_wmutex
+ char pad[(sizeof(pthread_mutex_t)+CACHELINE-1) & ~(CACHELINE-1)];
+ } mt2;
+ MDB_reader mti_readers[1];
} MDB_txninfo;
/* Common header for all page types. Overflow pages
#define mp_pgno mp_p.p_pgno
union padded {
pgno_t p_pgno; /* page number */
- void * p_pad;
+ void * p_align; /* for IL32P64 */
} mp_p;
#define P_BRANCH 0x01 /* branch page */
#define P_LEAF 0x02 /* leaf page */
#define P_OVERFLOW 0x04 /* overflow page */
#define P_META 0x08 /* meta page */
#define P_DIRTY 0x10 /* dirty page */
+#define P_LEAF2 0x20 /* DB with small, fixed size keys and no data */
uint32_t mp_flags;
#define mp_lower mp_pb.pb.pb_lower
#define mp_upper mp_pb.pb.pb_upper
indx_t pb_upper; /* upper bound of free space */
} pb;
uint32_t pb_pages; /* number of overflow pages */
+ struct {
+ indx_t pb_ksize; /* on a LEAF2 page */
+ indx_t pb_numkeys;
+ } pb2;
} mp_pb;
indx_t mp_ptrs[1]; /* dynamic size */
} MDB_page;
#define IS_BRANCH(p) F_ISSET((p)->mp_flags, P_BRANCH)
#define IS_OVERFLOW(p) F_ISSET((p)->mp_flags, P_OVERFLOW)
-#define OVPAGES(size, psize) (PAGEHDRSZ + size + psize - 1) / psize;
+#define OVPAGES(size, psize) ((PAGEHDRSZ-1 + (size)) / (psize) + 1)
typedef struct MDB_db {
uint32_t md_pad;
} MDB_meta;
typedef struct MDB_dhead { /* a dirty page */
- STAILQ_ENTRY(MDB_dpage) md_next; /* queue of dirty pages */
MDB_page *md_parent;
unsigned md_pi; /* parent index */
int md_num;
MDB_page p;
} MDB_dpage;
-STAILQ_HEAD(dirty_queue, MDB_dpage); /* FIXME: use a sorted data structure */
-
typedef struct MDB_oldpages {
struct MDB_oldpages *mo_next;
ULONG mo_txnid;
static int mdb_touch(MDB_txn *txn, MDB_pageparent *mp);
typedef struct MDB_ppage { /* ordered list of pages */
- SLIST_ENTRY(MDB_ppage) mp_entry;
MDB_page *mp_page;
unsigned int mp_ki; /* cursor index on page */
} MDB_ppage;
-SLIST_HEAD(page_stack, MDB_ppage);
-#define CURSOR_EMPTY(c) SLIST_EMPTY(&(c)->mc_stack)
-#define CURSOR_TOP(c) SLIST_FIRST(&(c)->mc_stack)
-#define CURSOR_POP(c) SLIST_REMOVE_HEAD(&(c)->mc_stack, mp_entry)
-#define CURSOR_PUSH(c,p) SLIST_INSERT_HEAD(&(c)->mc_stack, p, mp_entry)
+#define CURSOR_TOP(c) (&(c)->mc_stack[(c)->mc_snum-1])
+#define CURSOR_PARENT(c) (&(c)->mc_stack[(c)->mc_snum-2])
struct MDB_xcursor;
struct MDB_cursor {
MDB_txn *mc_txn;
- struct page_stack mc_stack; /* stack of parent pages */
+ MDB_ppage mc_stack[32]; /* stack of parent pages */
+ unsigned int mc_snum; /* number of pushed pages */
MDB_dbi mc_dbi;
short mc_initialized; /* 1 if initialized */
short mc_eof; /* 1 if end is reached */
struct MDB_xcursor *mc_xcursor;
};
-#define METAHASHLEN offsetof(MDB_meta, mm_hash)
-#define METADATA(p) ((void *)((char *)p + PAGEHDRSZ))
+#define METADATA(p) ((void *)((char *)(p) + PAGEHDRSZ))
typedef struct MDB_node {
#define mn_pgno mn_p.np_pgno
unsigned int mn_flags:4;
unsigned int mn_ksize:12; /* key size */
#define F_BIGDATA 0x01 /* data put on overflow page */
+#define F_SUBDATA 0x02 /* data is a sub-database */
+#define F_DUPDATA 0x04 /* data has duplicates */
char mn_data[1];
} MDB_node;
-
typedef struct MDB_dbx {
MDB_val md_name;
MDB_cmp_func *md_cmp; /* user compare function */
MDB_env *mt_env;
pgno_t *mt_free_pgs; /* this is an IDL */
union {
- struct dirty_queue *dirty_queue; /* modified pages */
+ MIDL2 *dirty_list; /* modified pages */
MDB_reader *reader;
} mt_u;
MDB_dbx *mt_dbxs; /* array */
MDB_db *mt_dbs;
unsigned int mt_numdbs;
-#define MDB_TXN_RDONLY 0x01 /* read-only transaction */
-#define MDB_TXN_ERROR 0x02 /* an error has occurred */
+#define MDB_TXN_RDONLY 0x01 /* read-only transaction */
+#define MDB_TXN_ERROR 0x02 /* an error has occurred */
#define MDB_TXN_METOGGLE 0x04 /* used meta page 1 */
unsigned int mt_flags;
};
struct MDB_env {
int me_fd;
int me_lfd;
- uint32_t me_flags;
+ int me_mfd; /* just for writing the meta pages */
+#define MDB_FATAL_ERROR 0x80000000U
+ uint32_t me_flags;
+ uint32_t me_extrapad; /* unused for now */
unsigned int me_maxreaders;
unsigned int me_numdbs;
unsigned int me_maxdbs;
MDB_txn *me_txn; /* current write transaction */
size_t me_mapsize;
off_t me_size; /* current file size */
+ pgno_t me_maxpg; /* me_mapsize / me_psize */
unsigned int me_psize;
- int me_db_toggle;
+ unsigned int me_db_toggle;
MDB_dbx *me_dbxs; /* array */
MDB_db *me_dbs[2];
MDB_oldpages *me_pghead;
pthread_key_t me_txkey; /* thread-key for readers */
+ MDB_dpage *me_dpages;
pgno_t me_free_pgs[MDB_IDL_UM_SIZE];
+ MIDL2 me_dirty_list[MDB_IDL_DB_SIZE];
};
#define NODESIZE offsetof(MDB_node, mn_data)
#define NODEDSZ(node) ((node)->mn_dsize)
#define MDB_COMMIT_PAGES 64 /* max number of pages to write in one commit */
-#define MDB_MAXCACHE_DEF 1024 /* max number of pages to keep in cache */
static int mdb_search_page_root(MDB_txn *txn,
MDB_dbi dbi, MDB_val *key,
MDB_cursor *cursor, int modify,
MDB_pageparent *mpp);
-static int mdbenv_read_header(MDB_env *env, MDB_meta *meta);
-static int mdbenv_read_meta(MDB_env *env, int *which);
-static int mdbenv_write_meta(MDB_txn *txn);
+static int mdb_env_read_header(MDB_env *env, MDB_meta *meta);
+static int mdb_env_read_meta(MDB_env *env, int *which);
+static int mdb_env_write_meta(MDB_txn *txn);
static MDB_page *mdb_get_page(MDB_txn *txn, pgno_t pgno);
static MDB_node *mdb_search_node(MDB_txn *txn, MDB_dbi dbi, MDB_page *mp,
static void mdb_del_node(MDB_page *mp, indx_t indx);
static int mdb_del0(MDB_txn *txn, MDB_dbi dbi, unsigned int ki,
MDB_pageparent *mpp, MDB_node *leaf);
+static int mdb_put0(MDB_txn *txn, MDB_dbi dbi,
+ MDB_val *key, MDB_val *data, unsigned int flags);
static int mdb_read_data(MDB_txn *txn, MDB_node *leaf, MDB_val *data);
static int mdb_rebalance(MDB_txn *txn, MDB_dbi dbi, MDB_pageparent *mp);
static int mdb_set_key(MDB_node *node, MDB_val *key);
static int mdb_sibling(MDB_cursor *cursor, int move_right);
static int mdb_cursor_next(MDB_cursor *cursor,
- MDB_val *key, MDB_val *data);
+ MDB_val *key, MDB_val *data, MDB_cursor_op op);
static int mdb_cursor_prev(MDB_cursor *cursor,
- MDB_val *key, MDB_val *data);
+ MDB_val *key, MDB_val *data, MDB_cursor_op op);
static int mdb_cursor_set(MDB_cursor *cursor,
- MDB_val *key, MDB_val *data, int *exactp);
+ MDB_val *key, MDB_val *data, MDB_cursor_op op, int *exactp);
static int mdb_cursor_first(MDB_cursor *cursor,
MDB_val *key, MDB_val *data);
static int mdb_cursor_last(MDB_cursor *cursor,
MDB_val *key, MDB_val *data);
+static void mdb_xcursor_init0(MDB_txn *txn, MDB_dbi dbi, MDB_xcursor *mx);
+static void mdb_xcursor_init1(MDB_txn *txn, MDB_dbi dbi, MDB_xcursor *mx, MDB_node *node);
+static void mdb_xcursor_fini(MDB_txn *txn, MDB_dbi dbi, MDB_xcursor *mx);
+
static size_t mdb_leaf_size(MDB_env *env, MDB_val *key,
MDB_val *data);
static size_t mdb_branch_size(MDB_env *env, MDB_val *key);
return *p1 - *p2;
}
+char *
+mdb_version(int *maj, int *min, int *pat)
+{
+ *maj = MDB_VERSION_MAJOR;
+ *min = MDB_VERSION_MINOR;
+ *pat = MDB_VERSION_PATCH;
+ return MDB_VERSION_STRING;
+}
+
+static char *const errstr[] = {
+ "MDB_KEYEXIST: Key/data pair already exists",
+ "MDB_NOTFOUND: No matching key/data pair found",
+ "MDB_PAGE_NOTFOUND: Requested page not found",
+ "MDB_CORRUPTED: Located page was wrong type",
+ "MDB_PANIC: Update of meta page failed",
+ "MDB_VERSION_MISMATCH: Database environment version mismatch"
+};
+
+char *
+mdb_strerror(int err)
+{
+ if (!err)
+ return ("Successful return: 0");
+
+ if (err >= MDB_KEYEXIST && err <= MDB_VERSION_MISMATCH)
+ return errstr[err - MDB_KEYEXIST];
+
+ return strerror(err);
+}
+
int
mdb_cmp(MDB_txn *txn, MDB_dbi dbi, const MDB_val *a, const MDB_val *b)
{
- return txn->mt_dbxs[dbi].md_cmp(a, b);
+ if (txn->mt_dbxs[dbi].md_cmp)
+ return txn->mt_dbxs[dbi].md_cmp(a, b);
+
+ if (txn->mt_dbs[dbi].md_flags & (MDB_REVERSEKEY
+#if __BYTE_ORDER == __LITTLE_ENDIAN
+ |MDB_INTEGERKEY
+#endif
+ ))
+ return memnrcmp(a->mv_data, a->mv_size, b->mv_data, b->mv_size);
+ else
+ return memncmp((char *)a->mv_data, a->mv_size, b->mv_data, b->mv_size);
}
-static int
-_mdb_cmp(MDB_txn *txn, MDB_dbi dbi, const MDB_val *key1, const MDB_val *key2)
+int
+mdb_dcmp(MDB_txn *txn, MDB_dbi dbi, const MDB_val *a, const MDB_val *b)
{
- if (F_ISSET(txn->mt_dbs[dbi].md_flags, MDB_REVERSEKEY))
- return memnrcmp(key1->mv_data, key1->mv_size, key2->mv_data, key2->mv_size);
- else
- return memncmp((char *)key1->mv_data, key1->mv_size, key2->mv_data, key2->mv_size);
+ if (txn->mt_dbxs[dbi].md_dcmp)
+ return txn->mt_dbxs[dbi].md_dcmp(a, b);
+
+ return memncmp((char *)a->mv_data, a->mv_size, b->mv_data, b->mv_size);
}
/* Allocate new page(s) for writing */
{
MDB_dpage *dp;
pgno_t pgno = P_INVALID;
- ULONG oldest = txn->mt_txnid - 2;
+ ULONG oldest;
+ MIDL2 mid;
+
+ if (txn->mt_txnid > 2) {
+ oldest = txn->mt_txnid - 2;
if (!txn->mt_env->me_pghead && txn->mt_dbs[FREE_DBI].md_root != P_INVALID) {
/* See if there's anything in the free DB */
MDB_pageparent mpp;
}
if (txn->mt_env->me_pghead) {
unsigned int i;
- for (i=0; i<txn->mt_env->me_txns->mt_numreaders; i++) {
- ULONG mr = txn->mt_env->me_txns->mt_readers[i].mr_txnid;
+ for (i=0; i<txn->mt_env->me_txns->mti_numreaders; i++) {
+ ULONG mr = txn->mt_env->me_txns->mti_readers[i].mr_txnid;
if (!mr) continue;
if (mr < oldest)
- oldest = txn->mt_env->me_txns->mt_readers[i].mr_txnid;
+ oldest = txn->mt_env->me_txns->mti_readers[i].mr_txnid;
}
if (oldest > txn->mt_env->me_pghead->mo_txnid) {
MDB_oldpages *mop = txn->mt_env->me_pghead;
}
}
}
+ }
- if ((dp = malloc(txn->mt_env->me_psize * num + sizeof(MDB_dhead))) == NULL)
- return NULL;
+ if (pgno == P_INVALID) {
+ /* DB size is maxed out */
+ if (txn->mt_next_pgno + num >= txn->mt_env->me_maxpg)
+ return NULL;
+ }
+ if (txn->mt_env->me_dpages && num == 1) {
+ dp = txn->mt_env->me_dpages;
+ txn->mt_env->me_dpages = (MDB_dpage *)dp->h.md_parent;
+ } else {
+ if ((dp = malloc(txn->mt_env->me_psize * num + sizeof(MDB_dhead))) == NULL)
+ return NULL;
+ }
dp->h.md_num = num;
dp->h.md_parent = parent;
dp->h.md_pi = parent_idx;
- STAILQ_INSERT_TAIL(txn->mt_u.dirty_queue, dp, h.md_next);
if (pgno == P_INVALID) {
dp->p.mp_pgno = txn->mt_next_pgno;
txn->mt_next_pgno += num;
} else {
dp->p.mp_pgno = pgno;
}
+ mid.mid = dp->p.mp_pgno;
+ mid.mptr = dp;
+ mdb_midl2_insert(txn->mt_u.dirty_list, &mid);
return dp;
}
if ((dp = mdb_alloc_page(txn, pp->mp_parent, pp->mp_pi, 1)) == NULL)
return ENOMEM;
DPRINTF("touched page %lu -> %lu", mp->mp_pgno, dp->p.mp_pgno);
- mdb_idl_insert(txn->mt_free_pgs, mp->mp_pgno);
+ mdb_midl_insert(txn->mt_free_pgs, mp->mp_pgno);
pgno = dp->p.mp_pgno;
memcpy(&dp->p, mp, txn->mt_env->me_psize);
mp = &dp->p;
}
int
-mdbenv_sync(MDB_env *env)
+mdb_env_sync(MDB_env *env, int force)
{
int rc = 0;
- if (!F_ISSET(env->me_flags, MDB_NOSYNC)) {
- if (fsync(env->me_fd))
+ if (force || !F_ISSET(env->me_flags, MDB_NOSYNC)) {
+ if (fdatasync(env->me_fd))
rc = errno;
}
return rc;
MDB_txn *txn;
int rc, toggle;
- if ((txn = calloc(1, sizeof(*txn))) == NULL) {
+ if (env->me_flags & MDB_FATAL_ERROR) {
+ DPUTS("mdb_txn_begin: environment had fatal error, must shutdown!");
+ return MDB_PANIC;
+ }
+ if ((txn = calloc(1, sizeof(MDB_txn))) == NULL) {
DPRINTF("calloc: %s", strerror(errno));
return ENOMEM;
}
if (rdonly) {
txn->mt_flags |= MDB_TXN_RDONLY;
} else {
- txn->mt_u.dirty_queue = calloc(1, sizeof(*txn->mt_u.dirty_queue));
- if (txn->mt_u.dirty_queue == NULL) {
- free(txn);
- return ENOMEM;
- }
- STAILQ_INIT(txn->mt_u.dirty_queue);
-
- pthread_mutex_lock(&env->me_txns->mt_wmutex);
- env->me_txns->mt_txnid++;
+ txn->mt_u.dirty_list = env->me_dirty_list;
+ txn->mt_u.dirty_list[0].mid = 0;
txn->mt_free_pgs = env->me_free_pgs;
txn->mt_free_pgs[0] = 0;
+
+ pthread_mutex_lock(&env->me_txns->mti_wmutex);
+ env->me_txns->mti_txnid++;
}
- txn->mt_txnid = env->me_txns->mt_txnid;
+ txn->mt_txnid = env->me_txns->mti_txnid;
if (rdonly) {
MDB_reader *r = pthread_getspecific(env->me_txkey);
if (!r) {
unsigned int i;
- pthread_mutex_lock(&env->me_txns->mt_mutex);
- for (i=0; i<env->me_txns->mt_numreaders; i++)
- if (env->me_txns->mt_readers[i].mr_pid == 0)
+ pthread_mutex_lock(&env->me_txns->mti_mutex);
+ for (i=0; i<env->me_txns->mti_numreaders; i++)
+ if (env->me_txns->mti_readers[i].mr_pid == 0)
break;
if (i == env->me_maxreaders) {
pthread_mutex_unlock(&env->me_txns->mti_mutex);
return ENOSPC;
}
- env->me_txns->mt_readers[i].mr_pid = getpid();
- env->me_txns->mt_readers[i].mr_tid = pthread_self();
- r = &env->me_txns->mt_readers[i];
+ env->me_txns->mti_readers[i].mr_pid = getpid();
+ env->me_txns->mti_readers[i].mr_tid = pthread_self();
+ r = &env->me_txns->mti_readers[i];
pthread_setspecific(env->me_txkey, r);
- if (i >= env->me_txns->mt_numreaders)
- env->me_txns->mt_numreaders = i+1;
- pthread_mutex_unlock(&env->me_txns->mt_mutex);
+ if (i >= env->me_txns->mti_numreaders)
+ env->me_txns->mti_numreaders = i+1;
+ pthread_mutex_unlock(&env->me_txns->mti_mutex);
}
r->mr_txnid = txn->mt_txnid;
txn->mt_u.reader = r;
txn->mt_env = env;
- if ((rc = mdbenv_read_meta(env, &toggle)) != MDB_SUCCESS) {
+ toggle = env->me_txns->mti_me_toggle;
+ if ((rc = mdb_env_read_meta(env, &toggle)) != MDB_SUCCESS) {
mdb_txn_abort(txn);
return rc;
}
void
mdb_txn_abort(MDB_txn *txn)
{
- MDB_dpage *dp;
MDB_env *env;
if (txn == NULL)
txn->mt_u.reader->mr_txnid = 0;
} else {
MDB_oldpages *mop;
+ MDB_dpage *dp;
unsigned int i;
- /* Discard all dirty pages. */
- while (!STAILQ_EMPTY(txn->mt_u.dirty_queue)) {
- dp = STAILQ_FIRST(txn->mt_u.dirty_queue);
- STAILQ_REMOVE_HEAD(txn->mt_u.dirty_queue, h.md_next);
- free(dp);
+ /* return all dirty pages to dpage list */
+ for (i=1; i<=txn->mt_u.dirty_list[0].mid; i++) {
+ dp = txn->mt_u.dirty_list[i].mptr;
+ if (dp->h.md_num == 1) {
+ dp->h.md_parent = (MDB_page *)txn->mt_env->me_dpages;
+ txn->mt_env->me_dpages = dp;
+ } else {
+ /* large pages just get freed directly */
+ free(dp);
+ }
}
- free(txn->mt_u.dirty_queue);
while ((mop = txn->mt_env->me_pghead)) {
txn->mt_env->me_pghead = mop->mo_next;
}
env->me_txn = NULL;
- env->me_txns->mt_txnid--;
+ env->me_txns->mti_txnid--;
for (i=2; i<env->me_numdbs; i++)
env->me_dbxs[i].md_dirty = 0;
- pthread_mutex_unlock(&env->me_txns->mt_wmutex);
+ pthread_mutex_unlock(&env->me_txns->mti_wmutex);
}
free(txn);
env = txn->mt_env;
if (F_ISSET(txn->mt_flags, MDB_TXN_RDONLY)) {
- DPRINTF("attempt to commit read-only transaction");
mdb_txn_abort(txn);
- return EPERM;
+ return MDB_SUCCESS;
}
if (txn != env->me_txn) {
- DPRINTF("attempt to commit unknown transaction");
+ DPUTS("attempt to commit unknown transaction");
mdb_txn_abort(txn);
return EINVAL;
}
if (F_ISSET(txn->mt_flags, MDB_TXN_ERROR)) {
- DPRINTF("error flag is set, can't commit");
+ DPUTS("error flag is set, can't commit");
mdb_txn_abort(txn);
return EINVAL;
}
- if (STAILQ_EMPTY(txn->mt_u.dirty_queue))
+ if (!txn->mt_u.dirty_list[0].mid)
goto done;
DPRINTF("committing transaction %lu on mdbenv %p, root page %lu",
key.mv_data = (char *)&mop->mo_txnid;
data.mv_size = MDB_IDL_SIZEOF(mop->mo_pages);
data.mv_data = mop->mo_pages;
- mdb_put(txn, FREE_DBI, &key, &data, 0);
+ mdb_put0(txn, FREE_DBI, &key, &data, 0);
free(env->me_pghead);
env->me_pghead = NULL;
}
key.mv_data = (char *)&txn->mt_txnid;
data.mv_size = MDB_IDL_SIZEOF(txn->mt_free_pgs);
data.mv_data = txn->mt_free_pgs;
- mdb_put(txn, FREE_DBI, &key, &data, 0);
+ mdb_put0(txn, FREE_DBI, &key, &data, 0);
}
/* Update DB root pointers. Their pages have already been
for (i = 2; i < txn->mt_numdbs; i++) {
if (txn->mt_dbxs[i].md_dirty) {
data.mv_data = &txn->mt_dbs[i];
- mdb_put(txn, i, &txn->mt_dbxs[i].md_name, &data, 0);
+ mdb_put0(txn, MAIN_DBI, &txn->mt_dbxs[i].md_name, &data, 0);
}
}
}
/* Commit up to MDB_COMMIT_PAGES dirty pages to disk until done.
*/
next = 0;
+ i = 1;
do {
n = 0;
done = 1;
size = 0;
- STAILQ_FOREACH(dp, txn->mt_u.dirty_queue, h.md_next) {
+ for (; i<=txn->mt_u.dirty_list[0].mid; i++) {
+ dp = txn->mt_u.dirty_list[i].mptr;
if (dp->p.mp_pgno != next) {
if (n) {
DPRINTF("committing %u dirty pages", n);
if (rc != size) {
n = errno;
if (rc > 0)
- DPRINTF("short write, filesystem full?");
+ DPUTS("short write, filesystem full?");
else
DPRINTF("writev: %s", strerror(errno));
mdb_txn_abort(txn);
if (rc != size) {
n = errno;
if (rc > 0)
- DPRINTF("short write, filesystem full?");
+ DPUTS("short write, filesystem full?");
else
DPRINTF("writev: %s", strerror(errno));
mdb_txn_abort(txn);
/* Drop the dirty pages.
*/
- while (!STAILQ_EMPTY(txn->mt_u.dirty_queue)) {
- dp = STAILQ_FIRST(txn->mt_u.dirty_queue);
- STAILQ_REMOVE_HEAD(txn->mt_u.dirty_queue, h.md_next);
- free(dp);
+ for (i=1; i<=txn->mt_u.dirty_list[0].mid; i++) {
+ dp = txn->mt_u.dirty_list[i].mptr;
+ if (dp->h.md_num == 1) {
+ dp->h.md_parent = (MDB_page *)txn->mt_env->me_dpages;
+ txn->mt_env->me_dpages = dp;
+ } else {
+ free(dp);
+ }
+ txn->mt_u.dirty_list[i].mid = 0;
}
- if ((n = mdbenv_sync(env)) != 0 ||
- (n = mdbenv_write_meta(txn)) != MDB_SUCCESS ||
- (n = mdbenv_sync(env)) != 0) {
+ if ((n = mdb_env_sync(env, 0)) != 0 ||
+ (n = mdb_env_write_meta(txn)) != MDB_SUCCESS) {
mdb_txn_abort(txn);
return n;
}
- env->me_txn = NULL;
+done:
+ env->me_txn = NULL;
/* update the DB tables */
{
int toggle = !env->me_db_toggle;
free(txn->mt_dbs);
}
- pthread_mutex_unlock(&env->me_txns->mt_wmutex);
- free(txn->mt_u.dirty_queue);
+ pthread_mutex_unlock(&env->me_txns->mti_wmutex);
free(txn);
- txn = NULL;
-
-done:
- mdb_txn_abort(txn);
return MDB_SUCCESS;
}
static int
-mdbenv_read_header(MDB_env *env, MDB_meta *meta)
+mdb_env_read_header(MDB_env *env, MDB_meta *meta)
{
char page[PAGESIZE];
MDB_page *p;
m = METADATA(p);
if (m->mm_magic != MDB_MAGIC) {
- DPRINTF("meta has invalid magic");
+ DPUTS("meta has invalid magic");
return EINVAL;
}
if (m->mm_version != MDB_VERSION) {
DPRINTF("database is version %u, expected version %u",
m->mm_version, MDB_VERSION);
- return EINVAL;
+ return MDB_VERSION_MISMATCH;
}
memcpy(meta, m, sizeof(*m));
}
static int
-mdbenv_init_meta(MDB_env *env, MDB_meta *meta)
+mdb_env_init_meta(MDB_env *env, MDB_meta *meta)
{
MDB_page *p, *q;
MDB_meta *m;
int rc;
unsigned int psize;
- DPRINTF("writing new meta page");
+ DPUTS("writing new meta page");
psize = sysconf(_SC_PAGE_SIZE);
meta->mm_magic = MDB_MAGIC;
meta->mm_psize = psize;
meta->mm_last_pg = 1;
meta->mm_flags = env->me_flags & 0xffff;
-#if __BYTE_ORDER == __LITTLE_ENDIAN
- /* freeDB keys are pgno_t's, must compare in int order */
- meta->mm_flags |= MDB_REVERSEKEY;
-#endif
+ meta->mm_flags |= MDB_INTEGERKEY;
meta->mm_dbs[0].md_root = P_INVALID;
meta->mm_dbs[1].md_root = P_INVALID;
}
static int
-mdbenv_write_meta(MDB_txn *txn)
+mdb_env_write_meta(MDB_txn *txn)
{
MDB_env *env;
- MDB_meta meta;
+ MDB_meta meta, metab;
off_t off;
- int rc, len;
+ int rc, len, toggle;
char *ptr;
assert(txn != NULL);
assert(txn->mt_env != NULL);
- DPRINTF("writing meta page for root page %lu", txn->mt_dbs[MAIN_DBI].md_root);
+ toggle = !F_ISSET(txn->mt_flags, MDB_TXN_METOGGLE);
+ DPRINTF("writing meta page %d for root page %lu",
+ toggle, txn->mt_dbs[MAIN_DBI].md_root);
env = txn->mt_env;
+ metab.mm_txnid = env->me_metas[toggle]->mm_txnid;
+ metab.mm_last_pg = env->me_metas[toggle]->mm_last_pg;
+
ptr = (char *)&meta;
off = offsetof(MDB_meta, mm_dbs[0].md_depth);
len = sizeof(MDB_meta) - off;
meta.mm_last_pg = txn->mt_next_pgno - 1;
meta.mm_txnid = txn->mt_txnid;
- if (!F_ISSET(txn->mt_flags, MDB_TXN_METOGGLE))
+ if (toggle)
off += env->me_psize;
off += PAGEHDRSZ;
- lseek(env->me_fd, off, SEEK_SET);
- rc = write(env->me_fd, ptr, len);
+ /* Write to the SYNC fd */
+ rc = pwrite(env->me_mfd, ptr, len, off);
if (rc != len) {
- DPRINTF("write failed, disk error?");
+ 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.
+ * Use the non-SYNC fd; we know it will fail anyway.
+ */
+ meta.mm_last_pg = metab.mm_last_pg;
+ meta.mm_txnid = metab.mm_txnid;
+ rc = pwrite(env->me_fd, ptr, len, off);
+ env->me_flags |= MDB_FATAL_ERROR;
return errno;
}
+ txn->mt_env->me_txns->mti_me_toggle = toggle;
return MDB_SUCCESS;
}
static int
-mdbenv_read_meta(MDB_env *env, int *which)
+mdb_env_read_meta(MDB_env *env, int *which)
{
int toggle = 0;
assert(env != NULL);
- if (env->me_metas[0]->mm_txnid < env->me_metas[1]->mm_txnid)
+ if (which)
+ toggle = *which;
+ else if (env->me_metas[0]->mm_txnid < env->me_metas[1]->mm_txnid)
toggle = 1;
if (env->me_meta != env->me_metas[toggle])
env->me_meta = env->me_metas[toggle];
- if (which)
- *which = toggle;
DPRINTF("Using meta page %d", toggle);
}
int
-mdbenv_create(MDB_env **env)
+mdb_env_create(MDB_env **env)
{
MDB_env *e;
- e = calloc(1, sizeof(*e));
+ e = calloc(1, sizeof(MDB_env));
if (!e) return ENOMEM;
e->me_maxreaders = DEFAULT_READERS;
e->me_maxdbs = 2;
e->me_fd = -1;
e->me_lfd = -1;
+ e->me_mfd = -1;
*env = e;
return MDB_SUCCESS;
}
int
-mdbenv_set_mapsize(MDB_env *env, size_t size)
+mdb_env_set_mapsize(MDB_env *env, size_t size)
{
if (env->me_map)
return EINVAL;
}
int
-mdbenv_set_maxdbs(MDB_env *env, int dbs)
+mdb_env_set_maxdbs(MDB_env *env, int dbs)
{
- if (env->me_map)
- return EINVAL;
env->me_maxdbs = dbs;
return MDB_SUCCESS;
}
int
-mdbenv_set_maxreaders(MDB_env *env, int readers)
+mdb_env_set_maxreaders(MDB_env *env, int readers)
{
env->me_maxreaders = readers;
return MDB_SUCCESS;
}
int
-mdbenv_get_maxreaders(MDB_env *env, int *readers)
+mdb_env_get_maxreaders(MDB_env *env, int *readers)
{
if (!env || !readers)
return EINVAL;
return MDB_SUCCESS;
}
-int
-mdbenv_open2(MDB_env *env, unsigned int flags)
+static int
+mdb_env_open2(MDB_env *env, unsigned int flags)
{
int i, newenv = 0;
MDB_meta meta;
memset(&meta, 0, sizeof(meta));
- if ((i = mdbenv_read_header(env, &meta)) != 0) {
+ if ((i = mdb_env_read_header(env, &meta)) != 0) {
if (i != ENOENT)
return i;
- DPRINTF("new mdbenv");
+ DPUTS("new mdbenv");
newenv = 1;
}
meta.mm_mapsize = env->me_mapsize;
if (flags & MDB_FIXEDMAP)
meta.mm_address = env->me_map;
- i = mdbenv_init_meta(env, &meta);
+ i = mdb_env_init_meta(env, &meta);
if (i != MDB_SUCCESS) {
munmap(env->me_map, env->me_mapsize);
return i;
}
env->me_psize = meta.mm_psize;
- p = (MDB_page *)(MDB_page *)(MDB_page *)(MDB_page *)(MDB_page *)(MDB_page *)(MDB_page *)(MDB_page *)(MDB_page *)env->me_map;
+ env->me_maxpg = env->me_mapsize / env->me_psize;
+
+ p = (MDB_page *)env->me_map;
env->me_metas[0] = METADATA(p);
env->me_metas[1] = (MDB_meta *)((char *)env->me_metas[0] + meta.mm_psize);
- if ((i = mdbenv_read_meta(env, NULL)) != 0)
+ if ((i = mdb_env_read_meta(env, NULL)) != 0)
return i;
DPRINTF("opened database version %u, pagesize %u",
}
static void
-mdbenv_reader_dest(void *ptr)
+mdb_env_reader_dest(void *ptr)
{
MDB_reader *reader = ptr;
/* downgrade the exclusive lock on the region back to shared */
static void
-mdbenv_share_locks(MDB_env *env)
+mdb_env_share_locks(MDB_env *env)
{
struct flock lock_info;
- env->me_txns->mt_txnid = env->me_meta->mm_txnid;
+ env->me_txns->mti_txnid = env->me_meta->mm_txnid;
+ if (env->me_metas[0]->mm_txnid < env->me_metas[1]->mm_txnid)
+ env->me_txns->mti_me_toggle = 1;
memset((void *)&lock_info, 0, sizeof(lock_info));
lock_info.l_type = F_RDLCK;
}
static int
-mdbenv_setup_locks(MDB_env *env, char *lpath, int mode, int *excl)
+mdb_env_setup_locks(MDB_env *env, char *lpath, int mode, int *excl)
{
int rc;
off_t size, rsize;
pthread_mutexattr_t mattr;
pthread_mutexattr_init(&mattr);
- pthread_mutexattr_setpshared(&mattr, PTHREAD_PROCESS_SHARED);
- pthread_mutex_init(&env->me_txns->mt_mutex, &mattr);
- pthread_mutex_init(&env->me_txns->mt_wmutex, &mattr);
- env->me_txns->mt_version = MDB_VERSION;
- env->me_txns->mt_magic = MDB_MAGIC;
- env->me_txns->mt_txnid = 0;
- env->me_txns->mt_numreaders = 0;
+ rc = pthread_mutexattr_setpshared(&mattr, PTHREAD_PROCESS_SHARED);
+ if (rc) {
+ goto fail;
+ }
+ pthread_mutex_init(&env->me_txns->mti_mutex, &mattr);
+ pthread_mutex_init(&env->me_txns->mti_wmutex, &mattr);
+ env->me_txns->mti_version = MDB_VERSION;
+ env->me_txns->mti_magic = MDB_MAGIC;
+ env->me_txns->mti_txnid = 0;
+ env->me_txns->mti_numreaders = 0;
+ env->me_txns->mti_me_toggle = 0;
} else {
- if (env->me_txns->mt_magic != MDB_MAGIC) {
- DPRINTF("lock region has invalid magic");
- errno = EINVAL;
+ if (env->me_txns->mti_magic != MDB_MAGIC) {
+ DPUTS("lock region has invalid magic");
+ rc = EINVAL;
+ goto fail;
}
- if (env->me_txns->mt_version != MDB_VERSION) {
+ if (env->me_txns->mti_version != MDB_VERSION) {
DPRINTF("lock region is version %u, expected version %u",
- env->me_txns->mt_version, MDB_VERSION);
- errno = EINVAL;
+ env->me_txns->mti_version, MDB_VERSION);
+ rc = MDB_VERSION_MISMATCH;
+ goto fail;
}
if (errno != EACCES && errno != EAGAIN) {
rc = errno;
fail:
close(env->me_lfd);
+ env->me_lfd = -1;
return rc;
}
+#define LOCKNAME "/lock.mdb"
+#define DATANAME "/data.mdb"
int
-mdbenv_open(MDB_env *env, const char *path, unsigned int flags, mode_t mode)
+mdb_env_open(MDB_env *env, const char *path, unsigned int flags, mode_t mode)
{
int oflags, rc, len, excl;
char *lpath, *dpath;
len = strlen(path);
- lpath = malloc(len + sizeof("/lock.mdb") + len + sizeof("/data.db"));
+ lpath = malloc(len + sizeof(LOCKNAME) + len + sizeof(DATANAME));
if (!lpath)
return ENOMEM;
- dpath = lpath + len + sizeof("/lock.mdb");
- sprintf(lpath, "%s/lock.mdb", path);
- sprintf(dpath, "%s/data.mdb", path);
+ dpath = lpath + len + sizeof(LOCKNAME);
+ sprintf(lpath, "%s" LOCKNAME, path);
+ sprintf(dpath, "%s" DATANAME, path);
- rc = mdbenv_setup_locks(env, lpath, mode, &excl);
+ rc = mdb_env_setup_locks(env, lpath, mode, &excl);
if (rc)
goto leave;
else
oflags = O_RDWR | O_CREAT;
- if ((env->me_fd = open(dpath, oflags, mode)) == -1)
- return errno;
+ if ((env->me_fd = open(dpath, oflags, mode)) == -1) {
+ rc = errno;
+ goto leave;
+ }
+
+ if ((rc = mdb_env_open2(env, flags)) == MDB_SUCCESS) {
+ /* synchronous fd for meta writes */
+ if (!(flags & (MDB_RDONLY|MDB_NOSYNC)))
+ oflags |= MDB_DSYNC;
+ if ((env->me_mfd = open(dpath, oflags, mode)) == -1) {
+ rc = errno;
+ goto leave;
+ }
- if ((rc = mdbenv_open2(env, flags)) != MDB_SUCCESS) {
- close(env->me_fd);
- env->me_fd = -1;
- } else {
env->me_path = strdup(path);
DPRINTF("opened dbenv %p", (void *) env);
- pthread_key_create(&env->me_txkey, mdbenv_reader_dest);
+ pthread_key_create(&env->me_txkey, mdb_env_reader_dest);
if (excl)
- mdbenv_share_locks(env);
+ mdb_env_share_locks(env);
env->me_dbxs = calloc(env->me_maxdbs, sizeof(MDB_dbx));
env->me_dbs[0] = calloc(env->me_maxdbs, sizeof(MDB_db));
env->me_dbs[1] = calloc(env->me_maxdbs, sizeof(MDB_db));
}
leave:
+ if (rc) {
+ if (env->me_fd >= 0) {
+ close(env->me_fd);
+ env->me_fd = -1;
+ }
+ if (env->me_lfd >= 0) {
+ close(env->me_lfd);
+ env->me_lfd = -1;
+ }
+ }
free(lpath);
return rc;
}
void
-mdbenv_close(MDB_env *env)
+mdb_env_close(MDB_env *env)
{
+ MDB_dpage *dp;
+
if (env == NULL)
return;
+ while (env->me_dpages) {
+ dp = env->me_dpages;
+ env->me_dpages = (MDB_dpage *)dp->h.md_parent;
+ free(dp);
+ }
+
+ free(env->me_dbs[1]);
+ free(env->me_dbs[0]);
free(env->me_dbxs);
free(env->me_path);
+ pthread_key_delete(env->me_txkey);
+
if (env->me_map) {
munmap(env->me_map, env->me_mapsize);
}
+ close(env->me_mfd);
close(env->me_fd);
if (env->me_txns) {
+ pid_t pid = getpid();
size_t size = (env->me_maxreaders-1) * sizeof(MDB_reader) + sizeof(MDB_txninfo);
+ int i;
+ for (i=0; i<env->me_txns->mti_numreaders; i++)
+ if (env->me_txns->mti_readers[i].mr_pid == pid)
+ env->me_txns->mti_readers[i].mr_pid = 0;
munmap(env->me_txns, size);
}
close(env->me_lfd);
nodekey.mv_size = node->mn_ksize;
nodekey.mv_data = NODEKEY(node);
- if (txn->mt_dbxs[dbi].md_cmp)
- rc = txn->mt_dbxs[dbi].md_cmp(key, &nodekey);
- else
- rc = _mdb_cmp(txn, dbi, key, &nodekey);
+ rc = mdb_cmp(txn, dbi, key, &nodekey);
if (IS_LEAF(mp))
DPRINTF("found leaf index %u [%.*s], rc = %i",
{
MDB_ppage *top;
- top = CURSOR_TOP(cursor);
- CURSOR_POP(cursor);
-
- DPRINTF("popped page %lu off cursor %p", top->mp_page->mp_pgno, (void *) cursor);
+ if (cursor->mc_snum) {
+ top = CURSOR_TOP(cursor);
+ cursor->mc_snum--;
- free(top);
+ DPRINTF("popped page %lu off cursor %p", top->mp_page->mp_pgno, (void *) cursor);
+ }
}
static MDB_ppage *
DPRINTF("pushing page %lu on cursor %p", mp->mp_pgno, (void *) cursor);
- if ((ppage = calloc(1, sizeof(*ppage))) == NULL)
- return NULL;
+ ppage = &cursor->mc_stack[cursor->mc_snum++];
ppage->mp_page = mp;
- CURSOR_PUSH(cursor, ppage);
+ ppage->mp_ki = 0;
return ppage;
}
MDB_page *p = NULL;
int found = 0;
- if (!F_ISSET(txn->mt_flags, MDB_TXN_RDONLY) && !STAILQ_EMPTY(txn->mt_u.dirty_queue)) {
+ if (!F_ISSET(txn->mt_flags, MDB_TXN_RDONLY) && txn->mt_u.dirty_list[0].mid) {
MDB_dpage *dp;
- STAILQ_FOREACH(dp, txn->mt_u.dirty_queue, h.md_next) {
- if (dp->p.mp_pgno == pgno) {
- p = &dp->p;
- found = 1;
- break;
- }
+ MIDL2 id;
+ unsigned x;
+ id.mid = pgno;
+ x = mdb_midl2_search(txn->mt_u.dirty_list, &id);
+ if (x <= txn->mt_u.dirty_list[0].mid && txn->mt_u.dirty_list[x].mid == pgno) {
+ dp = txn->mt_u.dirty_list[x].mptr;
+ p = &dp->p;
+ found = 1;
}
}
if (!found) {
+ if (pgno > txn->mt_env->me_meta->mm_last_pg)
+ return NULL;
p = (MDB_page *)(txn->mt_env->me_map + txn->mt_env->me_psize * pgno);
}
return p;
int rc;
if (cursor && cursor_push_page(cursor, mp) == NULL)
- return MDB_FAIL;
+ return ENOMEM;
while (IS_BRANCH(mp)) {
unsigned int i = 0;
mpp->mp_parent = mp;
if ((mp = mdb_get_page(txn, NODEPGNO(node))) == NULL)
- return MDB_FAIL;
+ return MDB_PAGE_NOTFOUND;
mpp->mp_pi = i;
mpp->mp_page = mp;
if (cursor && cursor_push_page(cursor, mp) == NULL)
- return MDB_FAIL;
+ return ENOMEM;
if (modify) {
MDB_dhead *dh = ((MDB_dhead *)mp)-1;
if (!IS_LEAF(mp)) {
DPRINTF("internal error, index points to a %02X page!?",
mp->mp_flags);
- return MDB_FAIL;
+ return MDB_CORRUPTED;
}
DPRINTF("found leaf page %lu for key %.*s", mp->mp_pgno,
* committed root page.
*/
if (F_ISSET(txn->mt_flags, MDB_TXN_ERROR)) {
- DPRINTF("transaction has failed, must abort");
+ DPUTS("transaction has failed, must abort");
return EINVAL;
} else
root = txn->mt_dbs[dbi].md_root;
if (root == P_INVALID) { /* Tree is empty. */
- DPRINTF("tree is empty");
- return ENOENT;
+ DPUTS("tree is empty");
+ return MDB_NOTFOUND;
}
if ((mpp->mp_page = mdb_get_page(txn, root)) == NULL)
- return MDB_FAIL;
+ return MDB_PAGE_NOTFOUND;
DPRINTF("root page has flags 0x%X", mpp->mp_page->mp_flags);
/* For sub-databases, update main root first */
if (dbi > MAIN_DBI && !txn->mt_dbxs[dbi].md_dirty) {
MDB_pageparent mp2;
- rc = mdb_search_page(txn, 0, &txn->mt_dbxs[dbi].md_name,
+ rc = mdb_search_page(txn, MAIN_DBI, &txn->mt_dbxs[dbi].md_name,
NULL, 1, &mp2);
if (rc)
return rc;
memcpy(&pgno, NODEDATA(leaf), sizeof(pgno));
if ((omp = mdb_get_page(txn, pgno)) == NULL) {
DPRINTF("read overflow page %lu failed", pgno);
- return MDB_FAIL;
+ return MDB_PAGE_NOTFOUND;
}
- data->mv_data = omp;
+ data->mv_data = METADATA(omp);
return MDB_SUCCESS;
}
assert(data);
DPRINTF("===> get key [%.*s]", (int)key->mv_size, (char *)key->mv_data);
+ if (txn == NULL || !dbi || dbi >= txn->mt_numdbs)
+ return EINVAL;
+
if (key->mv_size == 0 || key->mv_size > MAXKEYSIZE) {
return EINVAL;
}
return rc;
leaf = mdb_search_node(txn, dbi, mpp.mp_page, key, &exact, NULL);
- if (leaf && exact)
+ if (leaf && exact) {
+ /* Return first duplicate data item */
+ if (F_ISSET(leaf->mn_flags, F_DUPDATA)) {
+ MDB_xcursor mx;
+
+ mdb_xcursor_init0(txn, dbi, &mx);
+ mdb_xcursor_init1(txn, dbi, &mx, leaf);
+ rc = mdb_search_page(&mx.mx_txn, mx.mx_cursor.mc_dbi, NULL, NULL, 0, &mpp);
+ if (rc != MDB_SUCCESS)
+ return rc;
+ leaf = NODEPTR(mpp.mp_page, 0);
+ }
rc = mdb_read_data(txn, leaf, data);
- else {
- rc = ENOENT;
+ } else {
+ rc = MDB_NOTFOUND;
}
return rc;
{
int rc;
MDB_node *indx;
- MDB_ppage *parent, *top;
+ MDB_ppage *parent;
MDB_page *mp;
- top = CURSOR_TOP(cursor);
- if ((parent = SLIST_NEXT(top, mp_entry)) == NULL) {
- return ENOENT; /* root has no siblings */
+ if (cursor->mc_snum < 2) {
+ return MDB_NOTFOUND; /* root has no siblings */
}
+ parent = CURSOR_PARENT(cursor);
DPRINTF("parent page is page %lu, index %u",
parent->mp_page->mp_pgno, parent->mp_ki);
indx = NODEPTR(parent->mp_page, parent->mp_ki);
if ((mp = mdb_get_page(cursor->mc_txn, indx->mn_pgno)) == NULL)
- return MDB_FAIL;
+ return MDB_PAGE_NOTFOUND;
#if 0
mp->parent = parent->mp_page;
mp->parent_index = parent->mp_ki;
}
static int
-mdb_cursor_next(MDB_cursor *cursor, MDB_val *key, MDB_val *data)
+mdb_cursor_next(MDB_cursor *cursor, MDB_val *key, MDB_val *data, MDB_cursor_op op)
{
MDB_ppage *top;
MDB_page *mp;
MDB_node *leaf;
+ int rc;
if (cursor->mc_eof) {
- return ENOENT;
+ return MDB_NOTFOUND;
}
assert(cursor->mc_initialized);
top = CURSOR_TOP(cursor);
mp = top->mp_page;
+ if (cursor->mc_txn->mt_dbs[cursor->mc_dbi].md_flags & MDB_DUPSORT) {
+ leaf = NODEPTR(mp, top->mp_ki);
+ if ((op == MDB_NEXT || op == MDB_NEXT_DUP) && F_ISSET(leaf->mn_flags, F_DUPDATA)) {
+ rc = mdb_cursor_next(&cursor->mc_xcursor->mx_cursor, data, NULL, MDB_NEXT);
+ if (op != MDB_NEXT || rc == MDB_SUCCESS)
+ return rc;
+ }
+ }
+
DPRINTF("cursor_next: top page is %lu in cursor %p", mp->mp_pgno, (void *) cursor);
if (top->mp_ki + 1 >= NUMKEYS(mp)) {
- DPRINTF("=====> move to next sibling page");
+ DPUTS("=====> move to next sibling page");
if (mdb_sibling(cursor, 1) != MDB_SUCCESS) {
cursor->mc_eof = 1;
- return ENOENT;
+ return MDB_NOTFOUND;
}
top = CURSOR_TOP(cursor);
mp = top->mp_page;
assert(IS_LEAF(mp));
leaf = NODEPTR(mp, top->mp_ki);
- if (data && mdb_read_data(cursor->mc_txn, leaf, data) != MDB_SUCCESS)
- return MDB_FAIL;
+ if (F_ISSET(leaf->mn_flags, F_DUPDATA)) {
+ mdb_xcursor_init1(cursor->mc_txn, cursor->mc_dbi, cursor->mc_xcursor, leaf);
+ }
+ if (data) {
+ if ((rc = mdb_read_data(cursor->mc_txn, leaf, data) != MDB_SUCCESS))
+ return rc;
+
+ if (F_ISSET(leaf->mn_flags, F_DUPDATA)) {
+ rc = mdb_cursor_first(&cursor->mc_xcursor->mx_cursor, data, NULL);
+ if (rc != MDB_SUCCESS)
+ return rc;
+ }
+ }
return mdb_set_key(leaf, key);
}
static int
-mdb_cursor_prev(MDB_cursor *cursor, MDB_val *key, MDB_val *data)
+mdb_cursor_prev(MDB_cursor *cursor, MDB_val *key, MDB_val *data, MDB_cursor_op op)
{
MDB_ppage *top;
MDB_page *mp;
MDB_node *leaf;
+ int rc;
assert(cursor->mc_initialized);
top = CURSOR_TOP(cursor);
mp = top->mp_page;
+ if (cursor->mc_txn->mt_dbs[cursor->mc_dbi].md_flags & MDB_DUPSORT) {
+ leaf = NODEPTR(mp, top->mp_ki);
+ if ((op == MDB_PREV || op == MDB_PREV_DUP) && F_ISSET(leaf->mn_flags, F_DUPDATA)) {
+ rc = mdb_cursor_prev(&cursor->mc_xcursor->mx_cursor, data, NULL, MDB_PREV);
+ if (op != MDB_PREV || rc == MDB_SUCCESS)
+ return rc;
+ }
+ }
+
DPRINTF("cursor_prev: top page is %lu in cursor %p", mp->mp_pgno, (void *) cursor);
if (top->mp_ki == 0) {
- DPRINTF("=====> move to prev sibling page");
+ DPUTS("=====> move to prev sibling page");
if (mdb_sibling(cursor, 0) != MDB_SUCCESS) {
- return ENOENT;
+ return MDB_NOTFOUND;
}
top = CURSOR_TOP(cursor);
mp = top->mp_page;
assert(IS_LEAF(mp));
leaf = NODEPTR(mp, top->mp_ki);
- if (data && mdb_read_data(cursor->mc_txn, leaf, data) != MDB_SUCCESS)
- return MDB_FAIL;
+ if (F_ISSET(leaf->mn_flags, F_DUPDATA)) {
+ mdb_xcursor_init1(cursor->mc_txn, cursor->mc_dbi, cursor->mc_xcursor, leaf);
+ }
+ if (data) {
+ if ((rc = mdb_read_data(cursor->mc_txn, leaf, data) != MDB_SUCCESS))
+ return rc;
+
+ if (F_ISSET(leaf->mn_flags, F_DUPDATA)) {
+ rc = mdb_cursor_last(&cursor->mc_xcursor->mx_cursor, data, NULL);
+ if (rc != MDB_SUCCESS)
+ return rc;
+ }
+ }
return mdb_set_key(leaf, key);
}
static int
mdb_cursor_set(MDB_cursor *cursor, MDB_val *key, MDB_val *data,
- int *exactp)
+ MDB_cursor_op op, int *exactp)
{
int rc;
MDB_node *leaf;
assert(key);
assert(key->mv_size > 0);
+ cursor->mc_snum = 0;
+
rc = mdb_search_page(cursor->mc_txn, cursor->mc_dbi, key, cursor, 0, &mpp);
if (rc != MDB_SUCCESS)
return rc;
leaf = mdb_search_node(cursor->mc_txn, cursor->mc_dbi, mpp.mp_page, key, exactp, &top->mp_ki);
if (exactp != NULL && !*exactp) {
/* MDB_SET specified and not an exact match. */
- return ENOENT;
+ return MDB_NOTFOUND;
}
if (leaf == NULL) {
- DPRINTF("===> inexact leaf not found, goto sibling");
+ DPUTS("===> inexact leaf not found, goto sibling");
if ((rc = mdb_sibling(cursor, 1)) != MDB_SUCCESS)
return rc; /* no entries matched */
top = CURSOR_TOP(cursor);
cursor->mc_initialized = 1;
cursor->mc_eof = 0;
- if (data && (rc = mdb_read_data(cursor->mc_txn, leaf, data)) != MDB_SUCCESS)
- return rc;
+ if (F_ISSET(leaf->mn_flags, F_DUPDATA)) {
+ mdb_xcursor_init1(cursor->mc_txn, cursor->mc_dbi, cursor->mc_xcursor, leaf);
+ }
+ if (data) {
+ if (F_ISSET(leaf->mn_flags, F_DUPDATA)) {
+ if (op == MDB_SET || op == MDB_SET_RANGE) {
+ rc = mdb_cursor_first(&cursor->mc_xcursor->mx_cursor, data, NULL);
+ } else {
+ int ex2, *ex2p;
+ MDB_cursor_op op2;
+ if (op == MDB_GET_BOTH) {
+ ex2p = &ex2;
+ op2 = MDB_SET;
+ } else {
+ ex2p = NULL;
+ op2 = MDB_SET_RANGE;
+ }
+ rc = mdb_cursor_set(&cursor->mc_xcursor->mx_cursor, data, NULL, op2, ex2p);
+ if (rc != MDB_SUCCESS)
+ return rc;
+ }
+ } else if (op == MDB_GET_BOTH || op == MDB_GET_BOTH_RANGE) {
+ MDB_val d2;
+ if ((rc = mdb_read_data(cursor->mc_txn, leaf, &d2)) != MDB_SUCCESS)
+ return rc;
+ rc = mdb_dcmp(cursor->mc_txn, cursor->mc_dbi, data, &d2);
+ if (rc) {
+ if (op == MDB_GET_BOTH || rc > 0)
+ return MDB_NOTFOUND;
+ }
+
+ } else {
+ if ((rc = mdb_read_data(cursor->mc_txn, leaf, data)) != MDB_SUCCESS)
+ return rc;
+ }
+ }
rc = mdb_set_key(leaf, key);
if (rc == MDB_SUCCESS) {
MDB_pageparent mpp;
MDB_node *leaf;
+ cursor->mc_snum = 0;
+
rc = mdb_search_page(cursor->mc_txn, cursor->mc_dbi, NULL, cursor, 0, &mpp);
if (rc != MDB_SUCCESS)
return rc;
cursor->mc_initialized = 1;
cursor->mc_eof = 0;
- if (data && (rc = mdb_read_data(cursor->mc_txn, leaf, data)) != MDB_SUCCESS)
- return rc;
-
+ if (data) {
+ if (F_ISSET(leaf->mn_flags, F_DUPDATA)) {
+ mdb_xcursor_init1(cursor->mc_txn, cursor->mc_dbi, cursor->mc_xcursor, leaf);
+ rc = mdb_cursor_first(&cursor->mc_xcursor->mx_cursor, data, NULL);
+ if (rc)
+ return rc;
+ } else {
+ if ((rc = mdb_read_data(cursor->mc_txn, leaf, data)) != MDB_SUCCESS)
+ return rc;
+ }
+ }
return mdb_set_key(leaf, key);
}
MDB_node *leaf;
MDB_val lkey;
+ cursor->mc_snum = 0;
+
lkey.mv_size = MAXKEYSIZE+1;
lkey.mv_data = NULL;
top = CURSOR_TOP(cursor);
top->mp_ki = NUMKEYS(top->mp_page) - 1;
- if (data && (rc = mdb_read_data(cursor->mc_txn, leaf, data)) != MDB_SUCCESS)
- return rc;
+ if (data) {
+ if (F_ISSET(leaf->mn_flags, F_DUPDATA)) {
+ mdb_xcursor_init1(cursor->mc_txn, cursor->mc_dbi, cursor->mc_xcursor, leaf);
+ rc = mdb_cursor_last(&cursor->mc_xcursor->mx_cursor, data, NULL);
+ if (rc)
+ return rc;
+ } else {
+ if ((rc = mdb_read_data(cursor->mc_txn, leaf, data)) != MDB_SUCCESS)
+ return rc;
+ }
+ }
return mdb_set_key(leaf, key);
}
assert(cursor);
switch (op) {
+ case MDB_GET_BOTH:
+ case MDB_GET_BOTH_RANGE:
+ if (data == NULL || cursor->mc_xcursor == NULL) {
+ rc = EINVAL;
+ break;
+ }
+ /* FALLTHRU */
case MDB_SET:
case MDB_SET_RANGE:
- while (CURSOR_TOP(cursor) != NULL)
- cursor_pop_page(cursor);
if (key == NULL || key->mv_size == 0 || key->mv_size > MAXKEYSIZE) {
rc = EINVAL;
- } else if (op == MDB_SET)
- rc = mdb_cursor_set(cursor, key, data, &exact);
+ } else if (op == MDB_SET_RANGE)
+ rc = mdb_cursor_set(cursor, key, data, op, NULL);
else
- rc = mdb_cursor_set(cursor, key, data, NULL);
+ rc = mdb_cursor_set(cursor, key, data, op, &exact);
break;
case MDB_NEXT:
+ case MDB_NEXT_DUP:
+ case MDB_NEXT_NODUP:
if (!cursor->mc_initialized)
rc = mdb_cursor_first(cursor, key, data);
else
- rc = mdb_cursor_next(cursor, key, data);
+ rc = mdb_cursor_next(cursor, key, data, op);
break;
case MDB_PREV:
- if (!cursor->mc_initialized || cursor->mc_eof) {
- while (CURSOR_TOP(cursor) != NULL)
- cursor_pop_page(cursor);
+ case MDB_PREV_DUP:
+ case MDB_PREV_NODUP:
+ if (!cursor->mc_initialized || cursor->mc_eof)
rc = mdb_cursor_last(cursor, key, data);
- } else
- rc = mdb_cursor_prev(cursor, key, data);
+ else
+ rc = mdb_cursor_prev(cursor, key, data, op);
break;
case MDB_FIRST:
- while (CURSOR_TOP(cursor) != NULL)
- cursor_pop_page(cursor);
rc = mdb_cursor_first(cursor, key, data);
break;
case MDB_LAST:
- while (CURSOR_TOP(cursor) != NULL)
- cursor_pop_page(cursor);
rc = mdb_cursor_last(cursor, key, data);
break;
default:
data->mv_size);
node_size += sizeof(pgno_t);
if ((ofp = mdb_new_page(txn, dbi, P_OVERFLOW, ovpages)) == NULL)
- return MDB_FAIL;
+ return ENOMEM;
DPRINTF("allocated overflow page %lu", ofp->p.mp_pgno);
flags |= F_BIGDATA;
} else {
{
MDB_dbi dbn;
- mx->mx_cursor.mc_txn = &mx->mx_txn;
mx->mx_txn = *txn;
mx->mx_txn.mt_dbxs = mx->mx_dbxs;
mx->mx_txn.mt_dbs = mx->mx_dbs;
mx->mx_dbxs[dbn+1].md_rel = mx->mx_dbxs[dbn].md_rel;
mx->mx_dbxs[dbn+1].md_dirty = 0;
mx->mx_txn.mt_numdbs = dbn+2;
+
+ mx->mx_cursor.mc_snum = 0;
+ mx->mx_cursor.mc_txn = &mx->mx_txn;
+ mx->mx_cursor.mc_dbi = dbn+1;
}
static void
-mdb_xcursor_init1(MDB_txn *txn, MDB_dbi dbi, MDB_xcursor *mx, MDB_db *db)
+mdb_xcursor_init1(MDB_txn *txn, MDB_dbi dbi, MDB_xcursor *mx, MDB_node *node)
{
+ MDB_db *db = NODEDATA(node);
+ MDB_dbi dbn;
mx->mx_dbs[0] = txn->mt_dbs[0];
mx->mx_dbs[1] = txn->mt_dbs[1];
if (dbi > 1) {
mx->mx_dbs[2] = txn->mt_dbs[dbi];
- mx->mx_dbs[3] = *db;
+ dbn = 3;
} else {
- mx->mx_dbs[2] = *db;
+ dbn = 2;
}
+ mx->mx_dbs[dbn] = *db;
+ mx->mx_dbxs[dbn].md_name.mv_data = NODEKEY(node);
+ mx->mx_dbxs[dbn].md_name.mv_size = node->mn_ksize;
+ mx->mx_txn.mt_next_pgno = txn->mt_next_pgno;
+ mx->mx_txn.mt_oldest = txn->mt_oldest;
+ mx->mx_txn.mt_u = txn->mt_u;
}
static void
mdb_xcursor_fini(MDB_txn *txn, MDB_dbi dbi, MDB_xcursor *mx)
{
+ txn->mt_next_pgno = mx->mx_txn.mt_next_pgno;
+ txn->mt_oldest = mx->mx_txn.mt_oldest;
+ txn->mt_u = mx->mx_txn.mt_u;
txn->mt_dbs[0] = mx->mx_dbs[0];
txn->mt_dbs[1] = mx->mx_dbs[1];
txn->mt_dbxs[0].md_dirty = mx->mx_dbxs[0].md_dirty;
txn->mt_dbxs[1].md_dirty = mx->mx_dbxs[1].md_dirty;
if (dbi > 1) {
txn->mt_dbs[dbi] = mx->mx_dbs[2];
- txn->mt_dbxs[2].md_dirty = mx->mx_dbxs[2].md_dirty;
+ txn->mt_dbxs[dbi].md_dirty = mx->mx_dbxs[2].md_dirty;
}
}
size += sizeof(MDB_xcursor);
if ((cursor = calloc(1, size)) != NULL) {
- SLIST_INIT(&cursor->mc_stack);
cursor->mc_dbi = dbi;
cursor->mc_txn = txn;
if (txn->mt_dbs[dbi].md_flags & MDB_DUPSORT) {
return MDB_SUCCESS;
}
+/* Return the count of duplicate data items for the current key */
+int
+mdb_cursor_count(MDB_cursor *mc, unsigned long *countp)
+{
+ MDB_ppage *top;
+ MDB_node *leaf;
+
+ if (mc == NULL || countp == NULL)
+ return EINVAL;
+
+ if (!(mc->mc_txn->mt_dbs[mc->mc_dbi].md_flags & MDB_DUPSORT))
+ return EINVAL;
+
+ top = CURSOR_TOP(mc);
+ leaf = NODEPTR(top->mp_page, top->mp_ki);
+ if (!F_ISSET(leaf->mn_flags, F_DUPDATA)) {
+ *countp = 1;
+ } else {
+ if (!mc->mc_xcursor->mx_cursor.mc_initialized)
+ return EINVAL;
+
+ *countp = mc->mc_xcursor->mx_txn.mt_dbs[mc->mc_xcursor->mx_cursor.mc_dbi].md_entries;
+ }
+ return MDB_SUCCESS;
+}
+
void
mdb_cursor_close(MDB_cursor *cursor)
{
if (cursor != NULL) {
- while (!CURSOR_EMPTY(cursor))
- cursor_pop_page(cursor);
-
-/* btree_close(cursor->bt); */
free(cursor);
}
}
if (mpp->mp_parent == NULL) {
if (NUMKEYS(mpp->mp_page) == 0) {
- DPRINTF("tree is completely empty");
+ DPUTS("tree is completely empty");
txn->mt_dbs[dbi].md_root = P_INVALID;
txn->mt_dbs[dbi].md_depth--;
txn->mt_dbs[dbi].md_leaf_pages--;
} else if (IS_BRANCH(mpp->mp_page) && NUMKEYS(mpp->mp_page) == 1) {
- DPRINTF("collapsing root page!");
+ DPUTS("collapsing root page!");
txn->mt_dbs[dbi].md_root = NODEPGNO(NODEPTR(mpp->mp_page, 0));
if ((root = mdb_get_page(txn, txn->mt_dbs[dbi].md_root)) == NULL)
- return MDB_FAIL;
+ return MDB_PAGE_NOTFOUND;
txn->mt_dbs[dbi].md_depth--;
txn->mt_dbs[dbi].md_branch_pages--;
} else
- DPRINTF("root page doesn't need rebalancing");
+ DPUTS("root page doesn't need rebalancing");
return MDB_SUCCESS;
}
if (mpp->mp_pi == 0) {
/* We're the leftmost leaf in our parent.
*/
- DPRINTF("reading right neighbor");
+ DPUTS("reading right neighbor");
node = NODEPTR(mpp->mp_parent, mpp->mp_pi + 1);
if ((npp.mp_page = mdb_get_page(txn, NODEPGNO(node))) == NULL)
- return MDB_FAIL;
+ return MDB_PAGE_NOTFOUND;
npp.mp_pi = mpp->mp_pi + 1;
si = 0;
di = NUMKEYS(mpp->mp_page);
} else {
/* There is at least one neighbor to the left.
*/
- DPRINTF("reading left neighbor");
+ DPUTS("reading left neighbor");
node = NODEPTR(mpp->mp_parent, mpp->mp_pi - 1);
if ((npp.mp_page = mdb_get_page(txn, NODEPGNO(node))) == NULL)
- return MDB_FAIL;
+ return MDB_PAGE_NOTFOUND;
npp.mp_pi = mpp->mp_pi - 1;
si = NUMKEYS(npp.mp_page) - 1;
di = 0;
ovpages = OVPAGES(NODEDSZ(leaf), txn->mt_env->me_psize);
for (i=0; i<ovpages; i++) {
DPRINTF("freed ov page %lu", pg);
- mdb_idl_insert(txn->mt_free_pgs, pg);
+ mdb_midl_insert(txn->mt_free_pgs, pg);
pg++;
}
}
int
mdb_del(MDB_txn *txn, MDB_dbi dbi,
- MDB_val *key, MDB_val *data)
+ MDB_val *key, MDB_val *data,
+ unsigned int flags)
{
int rc, exact;
unsigned int ki;
assert(key != NULL);
- if (txn == NULL || dbi >= txn->mt_numdbs)
+ if (txn == NULL || !dbi || dbi >= txn->mt_numdbs)
return EINVAL;
if (F_ISSET(txn->mt_flags, MDB_TXN_RDONLY)) {
leaf = mdb_search_node(txn, dbi, mpp.mp_page, key, &exact, &ki);
if (leaf == NULL || !exact) {
- return ENOENT;
+ return MDB_NOTFOUND;
}
- if (data && (rc = mdb_read_data(txn, leaf, data)) != MDB_SUCCESS)
- return rc;
-
- if (F_ISSET(txn->mt_dbs[dbi].md_flags, MDB_DUPSORT)) {
- /* add all the child DB's pages to the free list */
- MDB_cursor mc;
+ if (F_ISSET(leaf->mn_flags, F_DUPDATA)) {
MDB_xcursor mx;
MDB_pageparent mp2;
mdb_xcursor_init0(txn, dbi, &mx);
- mdb_xcursor_init1(txn, dbi, &mx, NODEDATA(leaf));
- SLIST_INIT(&mc.mc_stack);
- mc.mc_dbi = mx.mx_txn.mt_numdbs-1;
- mc.mc_txn = &mx.mx_txn;
- rc = mdb_search_page(&mx.mx_txn, mx.mx_txn.mt_numdbs - 1, NULL, &mc, 0, &mp2);
- if (rc == MDB_SUCCESS) {
- MDB_ppage *top, *parent;
- MDB_node *ni;
- unsigned int i;
+ mdb_xcursor_init1(txn, dbi, &mx, leaf);
+ if (flags == MDB_DEL_DUP) {
+ rc = mdb_del(&mx.mx_txn, mx.mx_cursor.mc_dbi, data, NULL, 0);
+ mdb_xcursor_fini(txn, dbi, &mx);
+ if (rc != MDB_SUCCESS)
+ return rc;
+ /* If sub-DB still has entries, we're done */
+ if (mx.mx_txn.mt_dbs[mx.mx_cursor.mc_dbi].md_root != P_INVALID) {
+ memcpy(NODEDATA(leaf), &mx.mx_txn.mt_dbs[mx.mx_cursor.mc_dbi],
+ sizeof(MDB_db));
+ txn->mt_dbs[dbi].md_entries--;
+ return rc;
+ }
+ /* otherwise fall thru and delete the sub-DB */
+ } else {
+ /* add all the child DB's pages to the free list */
+ rc = mdb_search_page(&mx.mx_txn, mx.mx_cursor.mc_dbi,
+ NULL, &mx.mx_cursor, 0, &mp2);
+ if (rc == MDB_SUCCESS) {
+ MDB_ppage *top, *parent;
+ MDB_node *ni;
+ unsigned int i;
- cursor_pop_page(&mc);
- top = CURSOR_TOP(&mc);
- parent = SLIST_NEXT(top, mp_entry);
- do {
- for (i=0; i<NUMKEYS(top->mp_page); i++) {
- ni = NODEPTR(top->mp_page, i);
- mdb_idl_insert(txn->mt_free_pgs, ni->mn_pgno);
- }
- if (parent) {
- parent->mp_ki++;
- if (parent->mp_ki >= NUMKEYS(parent->mp_page)) {
- cursor_pop_page(&mc);
- top = CURSOR_TOP(&mc);
- parent = SLIST_NEXT(top, mp_entry);
- } else {
- ni = NODEPTR(parent->mp_page, parent->mp_ki);
- top = mdb_get_page(mc.mc_txn, ni->mn_pgno);
+ cursor_pop_page(&mx.mx_cursor);
+ if (mx.mx_cursor.mc_snum) {
+ top = CURSOR_TOP(&mx.mx_cursor);
+ while (mx.mx_cursor.mc_snum > 1) {
+ parent = CURSOR_PARENT(&mx.mx_cursor);
+ for (i=0; i<NUMKEYS(top->mp_page); i++) {
+ ni = NODEPTR(top->mp_page, i);
+ mdb_midl_insert(txn->mt_free_pgs, ni->mn_pgno);
+ }
+ parent->mp_ki++;
+ if (parent->mp_ki >= NUMKEYS(parent->mp_page)) {
+ cursor_pop_page(&mx.mx_cursor);
+ top = parent;
+ } else {
+ ni = NODEPTR(parent->mp_page, parent->mp_ki);
+ top->mp_page = mdb_get_page(&mx.mx_txn, ni->mn_pgno);
+ }
}
}
+ mdb_midl_insert(txn->mt_free_pgs, mx.mx_txn.mt_dbs[mx.mx_cursor.mc_dbi].md_root);
}
- mdb_idl_insert(txn->mt_free_pgs, mx.mx_txn.mt_dbs[mc.mc_dbi].md_root);
}
}
+ if (data && (rc = mdb_read_data(txn, leaf, data)) != MDB_SUCCESS)
+ return rc;
+
return mdb_del0(txn, dbi, ki, &mpp, leaf);
}
if (mdp->h.md_parent == NULL) {
if ((pdp = mdb_new_page(txn, dbi, P_BRANCH, 1)) == NULL)
- return MDB_FAIL;
+ return ENOMEM;
mdp->h.md_pi = 0;
mdp->h.md_parent = &pdp->p;
txn->mt_dbs[dbi].md_root = pdp->p.mp_pgno;
txn->mt_dbs[dbi].md_depth++;
/* Add left (implicit) pointer. */
- if (mdb_add_node(txn, dbi, &pdp->p, 0, NULL, NULL,
- mdp->p.mp_pgno, 0) != MDB_SUCCESS)
- return MDB_FAIL;
+ if ((rc = mdb_add_node(txn, dbi, &pdp->p, 0, NULL, NULL,
+ mdp->p.mp_pgno, 0)) != MDB_SUCCESS)
+ return rc;
} else {
DPRINTF("parent branch page is %lu", mdp->h.md_parent->mp_pgno);
}
/* Create a right sibling. */
if ((rdp = mdb_new_page(txn, dbi, mdp->p.mp_flags, 1)) == NULL)
- return MDB_FAIL;
+ return ENOMEM;
rdp->h.md_parent = mdp->h.md_parent;
rdp->h.md_pi = mdp->h.md_pi + 1;
DPRINTF("new right sibling: page %lu", rdp->p.mp_pgno);
/* Move half of the keys to the right sibling. */
if ((copy = malloc(txn->mt_env->me_psize)) == NULL)
- return MDB_FAIL;
+ return ENOMEM;
memcpy(copy, &mdp->p, txn->mt_env->me_psize);
memset(&mdp->p.mp_ptrs, 0, txn->mt_env->me_psize - PAGEHDRSZ);
mdp->p.mp_lower = PAGEHDRSZ;
}
if (rc != MDB_SUCCESS) {
free(copy);
- return MDB_FAIL;
+ return rc;
}
for (i = j = 0; i <= NUMKEYS(copy); j++) {
return rc;
}
-int
-mdb_put(MDB_txn *txn, MDB_dbi dbi,
+static int
+mdb_put0(MDB_txn *txn, MDB_dbi dbi,
MDB_val *key, MDB_val *data, unsigned int flags)
{
int rc = MDB_SUCCESS, exact;
unsigned int ki;
MDB_node *leaf;
MDB_pageparent mpp;
- MDB_val xdata, *rdata;
+ MDB_val xdata, *rdata, dkey;
MDB_db dummy;
-
- assert(key != NULL);
- assert(data != NULL);
-
- if (txn == NULL)
- return EINVAL;
-
- if (F_ISSET(txn->mt_flags, MDB_TXN_RDONLY)) {
- return EINVAL;
- }
-
- if (txn->mt_env->me_txn != txn) {
- return EINVAL;
- }
-
- if (key->mv_size == 0 || key->mv_size > MAXKEYSIZE) {
- return EINVAL;
- }
+ char dbuf[PAGESIZE];
+ int do_sub = 0;
DPRINTF("==> put key %.*s, size %zu, data size %zu",
(int)key->mv_size, (char *)key->mv_data, key->mv_size, data->mv_size);
+ dkey.mv_size = 0;
mpp.mp_parent = NULL;
mpp.mp_pi = 0;
rc = mdb_search_page(txn, dbi, key, NULL, 1, &mpp);
if (rc == MDB_SUCCESS) {
leaf = mdb_search_node(txn, dbi, mpp.mp_page, key, &exact, &ki);
if (leaf && exact) {
- if (F_ISSET(txn->mt_dbs[dbi].md_flags, MDB_DUPSORT)) {
- goto put_sub;
- }
- if (F_ISSET(flags, MDB_NOOVERWRITE)) {
+ if (flags == MDB_NOOVERWRITE) {
DPRINTF("duplicate key %.*s",
(int)key->mv_size, (char *)key->mv_data);
- return EEXIST;
+ return MDB_KEYEXIST;
+ }
+ if (F_ISSET(txn->mt_dbs[dbi].md_flags, MDB_DUPSORT)) {
+ /* Was a single item before, must convert now */
+ if (!F_ISSET(leaf->mn_flags, F_DUPDATA)) {
+ dkey.mv_size = NODEDSZ(leaf);
+ memcpy(dbuf, NODEDATA(leaf), dkey.mv_size);
+ memset(&dummy, 0, sizeof(dummy));
+ dummy.md_root = P_INVALID;
+ if (dkey.mv_size == sizeof(MDB_db)) {
+ memcpy(NODEDATA(leaf), &dummy, sizeof(dummy));
+ goto put_sub;
+ }
+ mdb_del_node(mpp.mp_page, ki);
+ do_sub = 1;
+ rdata = &xdata;
+ xdata.mv_size = sizeof(MDB_db);
+ xdata.mv_data = &dummy;
+ goto new_sub;
+ }
+ goto put_sub;
}
/* same size, just replace it */
if (NODEDSZ(leaf) == data->mv_size) {
ki = NUMKEYS(mpp.mp_page);
DPRINTF("appending key at index %i", ki);
}
- } else if (rc == ENOENT) {
+ } else if (rc == MDB_NOTFOUND) {
MDB_dpage *dp;
/* new file, just write a root leaf page */
- DPRINTF("allocating new root leaf page");
+ DPUTS("allocating new root leaf page");
if ((dp = mdb_new_page(txn, dbi, P_LEAF, 1)) == NULL) {
return ENOMEM;
}
mpp.mp_page = &dp->p;
txn->mt_dbs[dbi].md_root = mpp.mp_page->mp_pgno;
txn->mt_dbs[dbi].md_depth++;
+ txn->mt_dbxs[dbi].md_dirty = 1;
ki = 0;
}
else
DPRINTF("there are %u keys, should insert new key at index %i",
NUMKEYS(mpp.mp_page), ki);
- /* For sorted dups, the data item at this level is a DB record
- * for a child DB; the actual data elements are stored as keys
- * in the child DB.
- */
- if (F_ISSET(txn->mt_dbs[dbi].md_flags, MDB_DUPSORT)) {
- rdata = &xdata;
- xdata.mv_size = sizeof(MDB_db);
- xdata.mv_data = &dummy;
- memset(&dummy, 0, sizeof(dummy));
- dummy.md_root = P_INVALID;
- } else {
- rdata = data;
- }
+ rdata = data;
- if (SIZELEFT(mpp.mp_page) < mdb_leaf_size(txn->mt_env, key, data)) {
- rc = mdb_split(txn, dbi, &mpp.mp_page, &ki, key, data, P_INVALID);
+new_sub:
+ if (SIZELEFT(mpp.mp_page) < mdb_leaf_size(txn->mt_env, key, rdata)) {
+ rc = mdb_split(txn, dbi, &mpp.mp_page, &ki, key, rdata, P_INVALID);
} else {
/* There is room already in this leaf page. */
- rc = mdb_add_node(txn, dbi, mpp.mp_page, ki, key, data, 0, 0);
+ rc = mdb_add_node(txn, dbi, mpp.mp_page, ki, key, rdata, 0, 0);
}
if (rc != MDB_SUCCESS)
txn->mt_flags |= MDB_TXN_ERROR;
else {
- txn->mt_dbs[dbi].md_entries++;
+ /* Remember if we just added a subdatabase */
+ if (flags & F_SUBDATA) {
+ leaf = NODEPTR(mpp.mp_page, ki);
+ leaf->mn_flags |= F_SUBDATA;
+ }
+
/* Now store the actual data in the child DB. Note that we're
* storing the user data in the keys field, so there are strict
* size limits on dupdata. The actual data fields of the child
* DB are all zero size.
*/
- if (F_ISSET(txn->mt_dbs[dbi].md_flags, MDB_DUPSORT)) {
+ if (do_sub) {
MDB_xcursor mx;
leaf = NODEPTR(mpp.mp_page, ki);
put_sub:
mdb_xcursor_init0(txn, dbi, &mx);
- mdb_xcursor_init1(txn, dbi, &mx, NODEDATA(leaf));
+ mdb_xcursor_init1(txn, dbi, &mx, leaf);
xdata.mv_size = 0;
xdata.mv_data = "";
- rc = mdb_put(&mx.mx_txn, mx.mx_txn.mt_numdbs-1, data, &xdata, flags);
+ if (flags == MDB_NODUPDATA)
+ flags = MDB_NOOVERWRITE;
+ /* converted, write the original data first */
+ if (dkey.mv_size) {
+ dkey.mv_data = dbuf;
+ rc = mdb_put0(&mx.mx_txn, mx.mx_cursor.mc_dbi, &dkey, &xdata, flags);
+ if (rc) return rc;
+ leaf->mn_flags |= F_DUPDATA;
+ }
+ rc = mdb_put0(&mx.mx_txn, mx.mx_cursor.mc_dbi, data, &xdata, flags);
mdb_xcursor_fini(txn, dbi, &mx);
+ memcpy(NODEDATA(leaf), &mx.mx_txn.mt_dbs[mx.mx_cursor.mc_dbi],
+ sizeof(MDB_db));
}
+ txn->mt_dbs[dbi].md_entries++;
}
done:
}
int
-mdbenv_get_flags(MDB_env *env, unsigned int *arg)
+mdb_put(MDB_txn *txn, MDB_dbi dbi,
+ MDB_val *key, MDB_val *data, unsigned int flags)
+{
+ assert(key != NULL);
+ assert(data != NULL);
+
+ if (txn == NULL || !dbi || dbi >= txn->mt_numdbs)
+ return EINVAL;
+
+ if (F_ISSET(txn->mt_flags, MDB_TXN_RDONLY)) {
+ return EINVAL;
+ }
+
+ if (key->mv_size == 0 || key->mv_size > MAXKEYSIZE) {
+ return EINVAL;
+ }
+
+ if ((flags & (MDB_NOOVERWRITE|MDB_NODUPDATA)) != flags)
+ return EINVAL;
+
+ return mdb_put0(txn, dbi, key, data, flags);
+}
+
+int
+mdb_env_set_flags(MDB_env *env, unsigned int flag, int onoff)
+{
+#define CHANGEABLE (MDB_NOSYNC)
+ if ((flag & CHANGEABLE) != flag)
+ return EINVAL;
+ if (onoff)
+ env->me_flags |= flag;
+ else
+ env->me_flags &= ~flag;
+ return MDB_SUCCESS;
+}
+
+int
+mdb_env_get_flags(MDB_env *env, unsigned int *arg)
{
if (!env || !arg)
return EINVAL;
}
int
-mdbenv_get_path(MDB_env *env, const char **arg)
+mdb_env_get_path(MDB_env *env, const char **arg)
{
if (!env || !arg)
return EINVAL;
return MDB_SUCCESS;
}
+static int
+mdb_stat0(MDB_env *env, MDB_db *db, MDB_stat *arg)
+{
+ arg->ms_psize = env->me_psize;
+ arg->ms_depth = db->md_depth;
+ arg->ms_branch_pages = db->md_branch_pages;
+ arg->ms_leaf_pages = db->md_leaf_pages;
+ arg->ms_overflow_pages = db->md_overflow_pages;
+ arg->ms_entries = db->md_entries;
+
+ return MDB_SUCCESS;
+}
int
-mdbenv_stat(MDB_env *env, MDB_stat *arg)
+mdb_env_stat(MDB_env *env, MDB_stat *arg)
{
if (env == NULL || arg == NULL)
return EINVAL;
- arg->ms_psize = env->me_psize;
- arg->ms_depth = env->me_meta->mm_dbs[MAIN_DBI].md_depth;
- arg->ms_branch_pages = env->me_meta->mm_dbs[MAIN_DBI].md_branch_pages;
- arg->ms_leaf_pages = env->me_meta->mm_dbs[MAIN_DBI].md_leaf_pages;
- arg->ms_overflow_pages = env->me_meta->mm_dbs[MAIN_DBI].md_overflow_pages;
- arg->ms_entries = env->me_meta->mm_dbs[MAIN_DBI].md_entries;
-
- return MDB_SUCCESS;
+ return mdb_stat0(env, &env->me_meta->mm_dbs[MAIN_DBI], arg);
}
int mdb_open(MDB_txn *txn, const char *name, unsigned int flags, MDB_dbi *dbi)
{
MDB_val key, data;
MDB_dbi i;
- int rc;
+ int rc, dirty = 0;
size_t len;
/* main DB? */
if (!name) {
*dbi = MAIN_DBI;
+ if (flags & (MDB_DUPSORT|MDB_REVERSEKEY|MDB_INTEGERKEY))
+ txn->mt_dbs[MAIN_DBI].md_flags |= (flags & (MDB_DUPSORT|MDB_REVERSEKEY|MDB_INTEGERKEY));
return MDB_SUCCESS;
}
rc = mdb_get(txn, MAIN_DBI, &key, &data);
/* Create if requested */
- if (rc == ENOENT && (flags & MDB_CREATE)) {
+ if (rc == MDB_NOTFOUND && (flags & MDB_CREATE)) {
MDB_db dummy;
data.mv_size = sizeof(MDB_db);
data.mv_data = &dummy;
memset(&dummy, 0, sizeof(dummy));
dummy.md_root = P_INVALID;
dummy.md_flags = flags & 0xffff;
- rc = mdb_put(txn, 0, &key, &data, 0);
+ rc = mdb_put0(txn, MAIN_DBI, &key, &data, F_SUBDATA);
+ dirty = 1;
}
/* OK, got info, add to table */
txn->mt_dbxs[txn->mt_numdbs].md_dcmp = NULL;
txn->mt_dbxs[txn->mt_numdbs].md_rel = NULL;
txn->mt_dbxs[txn->mt_numdbs].md_parent = MAIN_DBI;
- txn->mt_dbxs[txn->mt_numdbs].md_dirty = 0;
+ txn->mt_dbxs[txn->mt_numdbs].md_dirty = dirty;
memcpy(&txn->mt_dbs[txn->mt_numdbs], data.mv_data, sizeof(MDB_db));
*dbi = txn->mt_numdbs;
+ txn->mt_env->me_dbs[0][txn->mt_numdbs] = txn->mt_dbs[txn->mt_numdbs];
+ txn->mt_env->me_dbs[1][txn->mt_numdbs] = txn->mt_dbs[txn->mt_numdbs];
txn->mt_numdbs++;
}
int mdb_stat(MDB_txn *txn, MDB_dbi dbi, MDB_stat *arg)
{
- if (txn == NULL || arg == NULL)
+ if (txn == NULL || arg == NULL || dbi >= txn->mt_numdbs)
return EINVAL;
- arg->ms_psize = txn->mt_env->me_psize;
- arg->ms_depth = txn->mt_dbs[dbi].md_depth;
- arg->ms_branch_pages = txn->mt_dbs[dbi].md_branch_pages;
- arg->ms_leaf_pages = txn->mt_dbs[dbi].md_leaf_pages;
- arg->ms_overflow_pages = txn->mt_dbs[dbi].md_overflow_pages;
- arg->ms_entries = txn->mt_dbs[dbi].md_entries;
-
- return MDB_SUCCESS;
+ return mdb_stat0(txn->mt_env, &txn->mt_dbs[dbi], arg);
}
void mdb_close(MDB_txn *txn, MDB_dbi dbi)
txn->mt_dbxs[dbi].md_name.mv_size = 0;
free(ptr);
}
+
+int mdb_set_compare(MDB_txn *txn, MDB_dbi dbi, MDB_cmp_func *cmp)
+{
+ if (txn == NULL || !dbi || dbi >= txn->mt_numdbs)
+ return EINVAL;
+
+ txn->mt_dbxs[dbi].md_cmp = cmp;
+ return MDB_SUCCESS;
+}
+
+int mdb_set_dupsort(MDB_txn *txn, MDB_dbi dbi, MDB_cmp_func *cmp)
+{
+ if (txn == NULL || !dbi || dbi >= txn->mt_numdbs)
+ return EINVAL;
+
+ txn->mt_dbxs[dbi].md_dcmp = cmp;
+ return MDB_SUCCESS;
+}
+
+int mdb_set_relfunc(MDB_txn *txn, MDB_dbi dbi, MDB_rel_func *rel)
+{
+ if (txn == NULL || !dbi || dbi >= txn->mt_numdbs)
+ return EINVAL;
+
+ txn->mt_dbxs[dbi].md_rel = rel;
+ return MDB_SUCCESS;
+}
projects / openldap / blobdiff