#ifdef HAVE_SYS_FILE_H
#include <sys/file.h>
#endif
+#include <fcntl.h>
#include <assert.h>
#include <err.h>
#include <string.h>
#include <time.h>
#include <unistd.h>
-
-#include <openssl/sha.h>
+#include <pthread.h>
+#include <endian.h>
#include "mdb.h"
-#define DEBUG
+#define ULONG unsigned long
+typedef ULONG pgno_t;
-#ifdef DEBUG
-# define DPRINTF(...) do { fprintf(stderr, "%s:%d: ", __func__, __LINE__); \
- fprintf(stderr, __VA_ARGS__); \
- fprintf(stderr, "\n"); } while(0)
+#include "idl.h"
+
+#ifndef DEBUG
+#define DEBUG 1
+#endif
+
+#if DEBUG && defined(__GNUC__)
+# define DPRINTF(fmt, ...) \
+ fprintf(stderr, "%s:%d: " fmt "\n", __func__, __LINE__, ##__VA_ARGS__)
#else
-# define DPRINTF(...)
+# define DPRINTF(...) ((void) 0)
#endif
#define PAGESIZE 4096
#define MDB_VERSION 1
#define MAXKEYSIZE 255
-#define P_INVALID (~0L)
+#define P_INVALID (~0UL)
#define F_ISSET(w, f) (((w) & (f)) == (f))
-typedef ulong pgno_t;
typedef uint16_t indx_t;
+#define DEFAULT_READERS 126
+#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_reader {
+ RXBODY;
+ /* cache line alignment */
+ char pad[CACHELINE-sizeof(MDB_rxbody)];
+} 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;
+} 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];
+} MDB_txninfo;
+
/* Common header for all page types. Overflow pages
* occupy a number of contiguous pages with no
* headers on any page after the first.
*/
typedef struct MDB_page { /* represents a page of storage */
- pgno_t mp_pgno; /* page number */
+#define mp_pgno mp_p.p_pgno
+ union padded {
+ pgno_t p_pgno; /* page number */
+ void * p_pad;
+ } 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_HEAD 0x10 /* header page */
-#define P_DIRTY 0x20 /* dirty page */
+#define P_DIRTY 0x10 /* dirty page */
uint32_t mp_flags;
#define mp_lower mp_pb.pb.pb_lower
#define mp_upper mp_pb.pb.pb_upper
indx_t mp_ptrs[1]; /* dynamic size */
} MDB_page;
-#define PAGEHDRSZ offsetof(MDB_page, mp_ptrs)
+#define PAGEHDRSZ ((unsigned) offsetof(MDB_page, mp_ptrs))
#define NUMKEYS(p) (((p)->mp_lower - PAGEHDRSZ) >> 1)
#define SIZELEFT(p) (indx_t)((p)->mp_upper - (p)->mp_lower)
-#define PAGEFILL(env, p) (1000 * ((env)->me_head.mh_psize - PAGEHDRSZ - SIZELEFT(p)) / \
- ((env)->me_head.mh_psize - PAGEHDRSZ))
+#define PAGEFILL(env, p) (1000L * ((env)->me_psize - PAGEHDRSZ - SIZELEFT(p)) / \
+ ((env)->me_psize - PAGEHDRSZ))
#define IS_LEAF(p) F_ISSET((p)->mp_flags, P_LEAF)
#define IS_BRANCH(p) F_ISSET((p)->mp_flags, P_BRANCH)
#define IS_OVERFLOW(p) F_ISSET((p)->mp_flags, P_OVERFLOW)
-typedef struct MDB_head { /* header page content */
- uint32_t mh_magic;
- uint32_t mh_version;
- uint32_t mh_flags;
- uint32_t mh_psize; /* page size */
- void *mh_address; /* address for fixed mapping */
- size_t mh_mapsize; /* size of mmap region */
-} MDB_head;
+#define OVPAGES(size, psize) (PAGEHDRSZ + size + psize - 1) / psize;
-typedef struct MDB_rootstat {
- pgno_t mr_root; /* page number of root page */
- MDB_stat mr_stat;
-} MDB_rootstat;
+typedef struct MDB_db {
+ uint32_t md_pad;
+ uint16_t md_flags;
+ uint16_t md_depth;
+ ULONG md_branch_pages;
+ ULONG md_leaf_pages;
+ ULONG md_overflow_pages;
+ ULONG md_entries;
+ pgno_t md_root;
+} MDB_db;
+
+#define FREE_DBI 0
+#define MAIN_DBI 1
typedef struct MDB_meta { /* meta (footer) page content */
- pgno_t mm_root; /* page number of root page */
- MDB_stat mm_stat;
- pgno_t mm_prev_meta; /* previous meta page number */
-#define MDB_TOMBSTONE 0x01 /* file is replaced */
- uint32_t mm_flags;
-#define mm_revisions mm_stat.ms_revisions
-#define mm_created_at mm_stat.ms_created_at
- uint8_t mm_hash[SHA_DIGEST_LENGTH];
+ uint32_t mm_magic;
+ uint32_t mm_version;
+ void *mm_address; /* address for fixed mapping */
+ size_t mm_mapsize; /* size of mmap region */
+ MDB_db mm_dbs[2]; /* first is free space, 2nd is main db */
+#define mm_psize mm_dbs[0].md_pad
+#define mm_flags mm_dbs[0].md_flags
+ pgno_t mm_last_pg; /* last used page in file */
+ ULONG mm_txnid; /* txnid that committed this page */
} MDB_meta;
typedef struct MDB_dhead { /* a dirty page */
- SIMPLEQ_ENTRY(MDB_dpage) md_next; /* queue of dirty pages */
+ STAILQ_ENTRY(MDB_dpage) md_next; /* queue of dirty pages */
MDB_page *md_parent;
- int md_pi; /* parent index */
+ unsigned md_pi; /* parent index */
int md_num;
} MDB_dhead;
MDB_page p;
} MDB_dpage;
-SIMPLEQ_HEAD(dirty_queue, MDB_dpage);
+STAILQ_HEAD(dirty_queue, MDB_dpage);
+
+typedef struct MDB_oldpages {
+ struct MDB_oldpages *mo_next;
+ ULONG mo_txnid;
+ pgno_t mo_pages[1]; /* dynamic */
+} MDB_oldpages;
typedef struct MDB_pageparent {
MDB_page *mp_page;
MDB_page *mp_parent;
- int mp_pi;
+ unsigned mp_pi;
} MDB_pageparent;
-static MDB_dpage *mdb_newpage(MDB_txn *txn, MDB_page *parent, int parent_idx, int num);
+static MDB_dpage *mdb_alloc_page(MDB_txn *txn, MDB_page *parent, unsigned int parent_idx, int num);
static int mdb_touch(MDB_txn *txn, MDB_pageparent *mp);
typedef struct MDB_ppage { /* ordered list of pages */
#define CURSOR_PUSH(c,p) SLIST_INSERT_HEAD(&(c)->mc_stack, p, mp_entry)
struct MDB_cursor {
- MDB_db *mc_db;
MDB_txn *mc_txn;
struct page_stack mc_stack; /* stack of parent pages */
+ MDB_dbi mc_dbi;
short mc_initialized; /* 1 if initialized */
short mc_eof; /* 1 if end is reached */
};
pgno_t np_pgno; /* child page number */
uint32_t np_dsize; /* leaf data size */
} mn_p;
- uint16_t mn_ksize; /* key size */
+ unsigned int mn_flags:4;
+ unsigned int mn_ksize:12; /* key size */
#define F_BIGDATA 0x01 /* data put on overflow page */
- uint8_t mn_flags;
char mn_data[1];
} MDB_node;
+
+typedef struct MDB_dbx {
+ MDB_val md_name;
+ MDB_cmp_func *md_cmp; /* user compare function */
+ MDB_cmp_func *md_dcmp; /* user dupsort function */
+ MDB_rel_func *md_rel; /* user relocate function */
+} MDB_dbx;
+
struct MDB_txn {
- pgno_t mt_root; /* current / new root page */
pgno_t mt_next_pgno; /* next unallocated page */
- pgno_t mt_first_pgno;
+ ULONG mt_txnid;
+ ULONG mt_oldest;
MDB_env *mt_env;
- struct dirty_queue *mt_dirty_queue; /* modified pages */
+ pgno_t *mt_free_pgs; /* this is an IDL */
+ union {
+ struct dirty_queue *dirty_queue; /* 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_METOGGLE 0x04 /* used meta page 1 */
unsigned int mt_flags;
};
-/* Must be same as MDB_db, minus md_root/md_stat */
-typedef struct MDB_db0 {
- unsigned int md_flags;
- MDB_cmp_func *md_cmp; /* user compare function */
- MDB_rel_func *md_rel; /* user relocate function */
- MDB_db *md_parent; /* parent tree */
- MDB_env *md_env;
-} MDB_db0;
-
-struct MDB_db {
- unsigned int md_flags;
- MDB_cmp_func *md_cmp; /* user compare function */
- MDB_rel_func *md_rel; /* user relocate function */
- MDB_db *md_parent; /* parent tree */
- MDB_env *md_env;
- pgno_t md_root; /* page number of root page */
- MDB_stat md_stat;
-};
-
struct MDB_env {
int me_fd;
-#define MDB_FIXPADDING 0x01 /* internal */
+ int me_lfd;
uint32_t me_flags;
+ unsigned int me_maxreaders;
+ unsigned int me_numdbs;
+ unsigned int me_maxdbs;
char *me_path;
- char *me_map;
- MDB_head me_head;
- MDB_db0 me_db; /* first DB, overlaps with meta */
- MDB_meta me_meta;
- MDB_txn *me_txn; /* current write transaction */
+ char *me_map;
+ MDB_txninfo *me_txns;
+ MDB_meta *me_metas[2];
+ MDB_meta *me_meta;
+ MDB_txn *me_txn; /* current write transaction */
size_t me_mapsize;
off_t me_size; /* current file size */
+ unsigned int me_psize;
+ 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 */
};
#define NODESIZE offsetof(MDB_node, mn_data)
#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_db *db,
- MDB_val *key,
+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 mdb_search_page(MDB_db *db,
- MDB_txn *txn, MDB_val *key,
+static int mdb_search_page(MDB_txn *txn,
+ MDB_dbi dbi, MDB_val *key,
MDB_cursor *cursor, int modify,
MDB_pageparent *mpp);
-static int mdbenv_write_header(MDB_env *env);
-static int mdbenv_read_header(MDB_env *env);
-static int mdb_check_meta_page(MDB_page *p);
-static int mdbenv_read_meta(MDB_env *env, pgno_t *p_next);
-static int mdbenv_write_meta(MDB_env *env, pgno_t root,
- unsigned int flags);
+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 MDB_page *mdbenv_get_page(MDB_env *env, pgno_t pgno);
-static MDB_node *mdb_search_node(MDB_db *db, MDB_page *mp,
+static MDB_node *mdb_search_node(MDB_txn *txn, MDB_dbi dbi, MDB_page *mp,
MDB_val *key, int *exactp, unsigned int *kip);
-static int mdb_add_node(MDB_db *bt, MDB_page *mp,
+static int mdb_add_node(MDB_txn *txn, MDB_dbi dbi, MDB_page *mp,
indx_t indx, MDB_val *key, MDB_val *data,
pgno_t pgno, uint8_t flags);
-static void mdb_del_node(MDB_db *bt, MDB_page *mp,
- indx_t indx);
-static int mdb_read_data(MDB_db *bt, MDB_page *mp,
- MDB_node *leaf, MDB_val *data);
-
-static int mdb_rebalance(MDB_db *bt, MDB_pageparent *mp);
-static int mdb_update_key(MDB_db *bt, MDB_page *mp,
- indx_t indx, MDB_val *key);
-static int mdb_move_node(MDB_db *bt,
+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_read_data(MDB_env *env, MDB_node *leaf, MDB_val *data);
+
+static int mdb_rebalance(MDB_txn *txn, MDB_dbi dbi, MDB_pageparent *mp);
+static int mdb_update_key(MDB_page *mp, indx_t indx, MDB_val *key);
+static int mdb_move_node(MDB_txn *txn, MDB_dbi dbi,
MDB_pageparent *src, indx_t srcindx,
MDB_pageparent *dst, indx_t dstindx);
-static int mdb_merge(MDB_db *bt, MDB_pageparent *src,
+static int mdb_merge(MDB_txn *txn, MDB_dbi dbi, MDB_pageparent *src,
MDB_pageparent *dst);
-static int mdb_split(MDB_db *bt, MDB_page **mpp,
+static int mdb_split(MDB_txn *txn, MDB_dbi dbi, MDB_page **mpp,
unsigned int *newindxp, MDB_val *newkey,
MDB_val *newdata, pgno_t newpgno);
-static MDB_dpage *mdbenv_new_page(MDB_env *env, uint32_t flags, int num);
+static MDB_dpage *mdb_new_page(MDB_txn *txn, MDB_dbi dbi, uint32_t flags, int num);
static void cursor_pop_page(MDB_cursor *cursor);
static MDB_ppage *cursor_push_page(MDB_cursor *cursor,
MDB_page *mp);
-static int mdb_set_key(MDB_db *bt, MDB_page *mp,
- MDB_node *node, MDB_val *key);
+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, 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 size_t mdb_leaf_size(MDB_db *bt, MDB_val *key,
+static size_t mdb_leaf_size(MDB_env *env, MDB_val *key,
MDB_val *data);
-static size_t mdb_branch_size(MDB_db *bt, MDB_val *key);
-
-static pgno_t mdbenv_compact_tree(MDB_env *env, pgno_t pgno,
- MDB_env *envc);
+static size_t mdb_branch_size(MDB_env *env, MDB_val *key);
static int memncmp(const void *s1, size_t n1,
const void *s2, size_t n2);
static int
memncmp(const void *s1, size_t n1, const void *s2, size_t n2)
{
- if (n1 < n2) {
- if (memcmp(s1, s2, n1) == 0)
- return -1;
- }
- else if (n1 > n2) {
- if (memcmp(s1, s2, n2) == 0)
- return 1;
+ int diff, len_diff = -1;
+
+ if (n1 >= n2) {
+ len_diff = (n1 > n2);
+ n1 = n2;
}
- return memcmp(s1, s2, n1);
+ diff = memcmp(s1, s2, n1);
+ return diff ? diff : len_diff;
}
static int
memnrcmp(const void *s1, size_t n1, const void *s2, size_t n2)
{
- const unsigned char *p1;
- const unsigned char *p2;
-
- if (n1 == 0)
- return n2 == 0 ? 0 : -1;
+ const unsigned char *p1, *p2, *p1_lim;
if (n2 == 0)
- return n1 == 0 ? 0 : 1;
+ return n1 != 0;
+ if (n1 == 0)
+ return -1;
p1 = (const unsigned char *)s1 + n1 - 1;
p2 = (const unsigned char *)s2 + n2 - 1;
- while (*p1 == *p2) {
- if (p1 == s1)
- return (p2 == s2) ? 0 : -1;
- if (p2 == s2)
- return (p1 == p2) ? 0 : 1;
- p1--;
- p2--;
+ for (p1_lim = (n1 <= n2 ? s1 : s2); *p1 == *p2; p1--, p2--) {
+ if (p1 == p1_lim)
+ return (p1 != s1) ? (p1 != p2) : (p2 != s2) ? -1 : 0;
}
return *p1 - *p2;
}
int
-mdb_cmp(MDB_db *db, const MDB_val *a, const MDB_val *b)
+mdb_cmp(MDB_txn *txn, MDB_dbi dbi, const MDB_val *a, const MDB_val *b)
{
- return db->md_cmp(a, b);
+ return txn->mt_dbxs[dbi].md_cmp(a, b);
}
static int
-_mdb_cmp(MDB_db *db, const MDB_val *key1, const MDB_val *key2)
+_mdb_cmp(MDB_txn *txn, MDB_dbi dbi, const MDB_val *key1, const MDB_val *key2)
{
- if (F_ISSET(db->md_flags, MDB_REVERSEKEY))
+ 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);
/* Allocate new page(s) for writing */
static MDB_dpage *
-mdb_newpage(MDB_txn *txn, MDB_page *parent, int parent_idx, int num)
+mdb_alloc_page(MDB_txn *txn, MDB_page *parent, unsigned int parent_idx, int num)
{
MDB_dpage *dp;
+ pgno_t pgno = P_INVALID;
+ ULONG 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;
+ MDB_node *leaf;
+ ULONG *kptr;
+
+ mpp.mp_parent = NULL;
+ mpp.mp_pi = 0;
+ mdb_search_page(txn, FREE_DBI, NULL, NULL, 0, &mpp);
+ leaf = NODEPTR(mpp.mp_page, 0);
+ kptr = (ULONG *)NODEKEY(leaf);
+
+ /* It's potentially usable, unless there are still
+ * older readers outstanding. Grab it.
+ */
+ if (oldest > *kptr) {
+ MDB_oldpages *mop;
+ MDB_val data;
+ pgno_t *idl;
+
+ mdb_read_data(txn->mt_env, leaf, &data);
+ idl = (ULONG *)data.mv_data;
+ mop = malloc(sizeof(MDB_oldpages) + MDB_IDL_SIZEOF(idl) - sizeof(pgno_t));
+ mop->mo_next = txn->mt_env->me_pghead;
+ mop->mo_txnid = *kptr;
+ txn->mt_env->me_pghead = mop;
+ memcpy(mop->mo_pages, idl, MDB_IDL_SIZEOF(idl));
+
+#if DEBUG > 1
+ {
+ unsigned int i;
+ DPRINTF("IDL read txn %lu root %lu num %lu",
+ mop->mo_txnid, txn->mt_dbs[FREE_DBI].md_root, idl[0]);
+ for (i=0; i<idl[0]; i++) {
+ DPRINTF("IDL %lu", idl[i+1]);
+ }
+ }
+#endif
+ /* drop this IDL from the DB */
+ mpp.mp_parent = NULL;
+ mpp.mp_pi = 0;
+ mdb_search_page(txn, FREE_DBI, NULL, NULL, 1, &mpp);
+ leaf = NODEPTR(mpp.mp_page, 0);
+ mdb_del0(txn, FREE_DBI, 0, &mpp, leaf);
+ }
+ }
+ 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;
+ if (!mr) continue;
+ if (mr < oldest)
+ oldest = txn->mt_env->me_txns->mt_readers[i].mr_txnid;
+ }
+ if (oldest > txn->mt_env->me_pghead->mo_txnid) {
+ MDB_oldpages *mop = txn->mt_env->me_pghead;
+ txn->mt_oldest = oldest;
+ if (num > 1) {
+ /* FIXME: For now, always use fresh pages. We
+ * really ought to search the free list for a
+ * contiguous range.
+ */
+ ;
+ } else {
+ /* peel pages off tail, so we only have to truncate the list */
+ pgno = MDB_IDL_LAST(mop->mo_pages);
+ if (MDB_IDL_IS_RANGE(mop->mo_pages)) {
+ mop->mo_pages[2]++;
+ if (mop->mo_pages[2] > mop->mo_pages[1])
+ mop->mo_pages[0] = 0;
+ } else {
+ mop->mo_pages[0]--;
+ }
+ if (MDB_IDL_IS_ZERO(mop->mo_pages)) {
+ txn->mt_env->me_pghead = mop->mo_next;
+ free(mop);
+ }
+ }
+ }
+ }
- if ((dp = malloc(txn->mt_env->me_head.mh_psize * num + sizeof(MDB_dhead))) == NULL)
+ 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;
- SIMPLEQ_INSERT_TAIL(txn->mt_dirty_queue, dp, h.md_next);
- dp->p.mp_pgno = txn->mt_next_pgno;
- txn->mt_next_pgno += num;
+ 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;
+ }
return dp;
}
static int
mdb_touch(MDB_txn *txn, MDB_pageparent *pp)
{
- int rc;
MDB_page *mp = pp->mp_page;
pgno_t pgno;
assert(txn != NULL);
if (!F_ISSET(mp->mp_flags, P_DIRTY)) {
MDB_dpage *dp;
- DPRINTF("touching page %lu -> %lu", mp->mp_pgno, txn->mt_next_pgno);
- if ((dp = mdb_newpage(txn, pp->mp_parent, pp->mp_pi, 1)) == NULL)
+ 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);
pgno = dp->p.mp_pgno;
- bcopy(mp, &dp->p, txn->mt_env->me_head.mh_psize);
+ memcpy(&dp->p, mp, txn->mt_env->me_psize);
mp = &dp->p;
mp->mp_pgno = pgno;
mp->mp_flags |= P_DIRTY;
mdb_txn_begin(MDB_env *env, int rdonly, MDB_txn **ret)
{
MDB_txn *txn;
- int rc;
-
- if (!rdonly && env->me_txn != NULL) {
- DPRINTF("write transaction already begun");
- return EBUSY;
- }
+ int rc, toggle;
if ((txn = calloc(1, sizeof(*txn))) == NULL) {
DPRINTF("calloc: %s", strerror(errno));
if (rdonly) {
txn->mt_flags |= MDB_TXN_RDONLY;
} else {
- txn->mt_dirty_queue = calloc(1, sizeof(*txn->mt_dirty_queue));
- if (txn->mt_dirty_queue == NULL) {
+ txn->mt_u.dirty_queue = calloc(1, sizeof(*txn->mt_u.dirty_queue));
+ if (txn->mt_u.dirty_queue == NULL) {
free(txn);
return ENOMEM;
}
- SIMPLEQ_INIT(txn->mt_dirty_queue);
+ STAILQ_INIT(txn->mt_u.dirty_queue);
-#if 0
- DPRINTF("taking write lock on txn %p", txn);
- if (flock(bt->fd, LOCK_EX | LOCK_NB) != 0) {
- DPRINTF("flock: %s", strerror(errno));
- errno = EBUSY;
- free(txn->dirty_queue);
+ pthread_mutex_lock(&env->me_txns->mt_wmutex);
+ env->me_txns->mt_txnid++;
+ txn->mt_free_pgs = malloc(MDB_IDL_UM_SIZEOF);
+ if (txn->mt_free_pgs == NULL) {
+ free(txn->mt_u.dirty_queue);
free(txn);
- return NULL;
+ return ENOMEM;
}
-#endif
+ txn->mt_free_pgs[0] = 0;
+ }
+
+ txn->mt_txnid = env->me_txns->mt_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_maxreaders; i++) {
+ if (env->me_txns->mt_readers[i].mr_pid == 0) {
+ 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];
+ pthread_setspecific(env->me_txkey, r);
+ if (i >= env->me_txns->mt_numreaders)
+ env->me_txns->mt_numreaders = i+1;
+ break;
+ }
+ }
+ pthread_mutex_unlock(&env->me_txns->mt_mutex);
+ if (i == env->me_maxreaders) {
+ return ENOSPC;
+ }
+ }
+ r->mr_txnid = txn->mt_txnid;
+ txn->mt_u.reader = r;
+ } else {
env->me_txn = txn;
}
txn->mt_env = env;
- if ((rc = mdbenv_read_meta(env, &txn->mt_next_pgno)) != MDB_SUCCESS) {
+ if ((rc = mdbenv_read_meta(env, &toggle)) != MDB_SUCCESS) {
mdb_txn_abort(txn);
return rc;
}
- txn->mt_first_pgno = txn->mt_next_pgno;
- txn->mt_root = env->me_meta.mm_root;
- DPRINTF("begin transaction on mdbenv %p, root page %lu", env, txn->mt_root);
+ /* Copy the DB arrays */
+ txn->mt_numdbs = env->me_numdbs;
+ txn->mt_dbxs = env->me_dbxs; /* mostly static anyway */
+ txn->mt_dbs = malloc(env->me_maxdbs * sizeof(MDB_db));
+ memcpy(txn->mt_dbs, env->me_meta->mm_dbs, 2 * sizeof(MDB_db));
+ if (txn->mt_numdbs > 2)
+ memcpy(txn->mt_dbs+2, env->me_dbs[env->me_db_toggle]+2,
+ (txn->mt_numdbs - 2) * sizeof(MDB_db));
+
+ if (!rdonly) {
+ if (toggle)
+ txn->mt_flags |= MDB_TXN_METOGGLE;
+ txn->mt_next_pgno = env->me_meta->mm_last_pg+1;
+ }
+
+ DPRINTF("begin transaction %lu on mdbenv %p, root page %lu",
+ txn->mt_txnid, (void *) env, txn->mt_dbs[MAIN_DBI].md_root);
*ret = txn;
return MDB_SUCCESS;
return;
env = txn->mt_env;
- DPRINTF("abort transaction on mdbenv %p, root page %lu", env, txn->mt_root);
+ DPRINTF("abort transaction %lu on mdbenv %p, root page %lu",
+ txn->mt_txnid, (void *) env, txn->mt_dbs[MAIN_DBI].md_root);
- if (!F_ISSET(txn->mt_flags, MDB_TXN_RDONLY)) {
- /* Discard all dirty pages.
- */
- while (!SIMPLEQ_EMPTY(txn->mt_dirty_queue)) {
- dp = SIMPLEQ_FIRST(txn->mt_dirty_queue);
- SIMPLEQ_REMOVE_HEAD(txn->mt_dirty_queue, h.md_next);
+ free(txn->mt_dbs);
+
+ if (F_ISSET(txn->mt_flags, MDB_TXN_RDONLY)) {
+ txn->mt_u.reader->mr_txnid = 0;
+ } else {
+ MDB_oldpages *mop;
+ /* 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);
}
+ free(txn->mt_free_pgs);
+ free(txn->mt_u.dirty_queue);
-#if 0
- DPRINTF("releasing write lock on txn %p", txn);
- txn->bt->txn = NULL;
- if (flock(txn->bt->fd, LOCK_UN) != 0) {
- DPRINTF("failed to unlock fd %d: %s",
- txn->bt->fd, strerror(errno));
+ while ((mop = txn->mt_env->me_pghead)) {
+ txn->mt_env->me_pghead = mop->mo_next;
+ free(mop);
}
-#endif
- free(txn->mt_dirty_queue);
- }
- if (txn == env->me_txn)
env->me_txn = NULL;
+ env->me_txns->mt_txnid--;
+ pthread_mutex_unlock(&env->me_txns->mt_wmutex);
+ }
+
free(txn);
}
mdb_txn_commit(MDB_txn *txn)
{
int n, done;
+ unsigned int i;
ssize_t rc;
off_t size;
MDB_dpage *dp;
MDB_env *env;
+ pgno_t next;
struct iovec iov[MDB_COMMIT_PAGES];
assert(txn != NULL);
return EINVAL;
}
- if (SIMPLEQ_EMPTY(txn->mt_dirty_queue))
+ if (STAILQ_EMPTY(txn->mt_u.dirty_queue))
goto done;
- if (F_ISSET(env->me_flags, MDB_FIXPADDING)) {
- size = lseek(env->me_fd, 0, SEEK_END);
- size += env->me_head.mh_psize - (size % env->me_head.mh_psize);
- DPRINTF("extending to multiple of page size: %lu", size);
- if (ftruncate(env->me_fd, size) != 0) {
- n = errno;
- DPRINTF("ftruncate: %s", strerror(errno));
- mdb_txn_abort(txn);
- return n;
+ DPRINTF("committing transaction %lu on mdbenv %p, root page %lu",
+ txn->mt_txnid, (void *) env, txn->mt_dbs[MAIN_DBI].md_root);
+
+ /* should only be one record now */
+ if (env->me_pghead) {
+ MDB_val key, data;
+ MDB_oldpages *mop;
+
+ mop = env->me_pghead;
+ key.mv_size = sizeof(pgno_t);
+ 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);
+ free(env->me_pghead);
+ env->me_pghead = NULL;
+ }
+ /* save to free list */
+ if (!MDB_IDL_IS_ZERO(txn->mt_free_pgs)) {
+ MDB_val key, data;
+ MDB_pageparent mpp;
+
+ /* make sure last page of freeDB is touched and on freelist */
+ key.mv_size = MAXKEYSIZE+1;
+ key.mv_data = NULL;
+ mpp.mp_parent = NULL;
+ mpp.mp_pi = 0;
+ mdb_search_page(txn, FREE_DBI, &key, NULL, 1, &mpp);
+
+#if DEBUG > 1
+ {
+ unsigned int i;
+ ULONG *idl = txn->mt_free_pgs;
+ DPRINTF("IDL write txn %lu root %lu num %lu",
+ txn->mt_txnid, txn->mt_dbs[FREE_DBI].md_root, idl[0]);
+ for (i=0; i<idl[0]; i++) {
+ DPRINTF("IDL %lu", idl[i+1]);
+ }
}
- env->me_flags ^= MDB_FIXPADDING;
+#endif
+ /* write to last page of freeDB */
+ key.mv_size = sizeof(pgno_t);
+ 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);
}
- DPRINTF("committing transaction on mdbenv %p, root page %lu",
- env, txn->mt_root);
+ /* Update DB root pointers. Their pages have already been
+ * touched so this is all in-place and cannot fail.
+ */
+ {
+ MDB_val data;
+ data.mv_size = sizeof(MDB_db);
+
+ for (i = 2; i < env->me_numdbs; i++) {
+ if (env->me_dbs[env->me_db_toggle][i].md_root != txn->mt_dbs[i].md_root) {
+ data.mv_data = &txn->mt_dbs[i];
+ mdb_put(txn, i, &txn->mt_dbxs[i].md_name, &data, 0);
+ }
+ }
+ for (i = env->me_numdbs; i < txn->mt_numdbs; i++) {
+ data.mv_data = &txn->mt_dbs[i];
+ mdb_put(txn, i, &txn->mt_dbxs[i].md_name, &data, 0);
+ }
+ }
/* Commit up to MDB_COMMIT_PAGES dirty pages to disk until done.
*/
+ next = 0;
do {
n = 0;
done = 1;
- SIMPLEQ_FOREACH(dp, txn->mt_dirty_queue, h.md_next) {
+ size = 0;
+ STAILQ_FOREACH(dp, txn->mt_u.dirty_queue, h.md_next) {
+ if (dp->p.mp_pgno != next) {
+ if (n) {
+ DPRINTF("committing %u dirty pages", n);
+ rc = writev(env->me_fd, iov, n);
+ if (rc != size) {
+ n = errno;
+ if (rc > 0)
+ DPRINTF("short write, filesystem full?");
+ else
+ DPRINTF("writev: %s", strerror(errno));
+ mdb_txn_abort(txn);
+ return n;
+ }
+ n = 0;
+ size = 0;
+ }
+ lseek(env->me_fd, dp->p.mp_pgno * env->me_psize, SEEK_SET);
+ next = dp->p.mp_pgno;
+ }
DPRINTF("committing page %lu", dp->p.mp_pgno);
- iov[n].iov_len = env->me_head.mh_psize * dp->h.md_num;
+ iov[n].iov_len = env->me_psize * dp->h.md_num;
iov[n].iov_base = &dp->p;
+ size += iov[n].iov_len;
+ next = dp->p.mp_pgno + dp->h.md_num;
/* clear dirty flag */
dp->p.mp_flags &= ~P_DIRTY;
if (++n >= MDB_COMMIT_PAGES) {
DPRINTF("committing %u dirty pages", n);
rc = writev(env->me_fd, iov, n);
- if (rc != (ssize_t)env->me_head.mh_psize*n) {
+ if (rc != size) {
n = errno;
if (rc > 0)
DPRINTF("short write, filesystem full?");
return n;
}
- /* Drop the dirty pages.
- */
- while (!SIMPLEQ_EMPTY(txn->mt_dirty_queue)) {
- dp = SIMPLEQ_FIRST(txn->mt_dirty_queue);
- SIMPLEQ_REMOVE_HEAD(txn->mt_dirty_queue, h.md_next);
- free(dp);
- if (--n == 0)
- break;
- }
} while (!done);
+ /* 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);
+ }
+
if ((n = mdbenv_sync(env)) != 0 ||
- (n = mdbenv_write_meta(env, txn->mt_root, 0)) != MDB_SUCCESS ||
+ (n = mdbenv_write_meta(txn)) != MDB_SUCCESS ||
(n = mdbenv_sync(env)) != 0) {
mdb_txn_abort(txn);
return n;
}
- txn = NULL;
env->me_txn = NULL;
-done:
- mdb_txn_abort(txn);
-
- return MDB_SUCCESS;
-}
+ /* update the DB tables */
+ {
+ int toggle = !env->me_db_toggle;
-static int
-mdbenv_write_header(MDB_env *env)
-{
- struct stat sb;
- MDB_head *h;
- MDB_page *p;
- ssize_t rc;
- unsigned int psize;
-
- DPRINTF("writing header page");
- assert(env != NULL);
-
- psize = sysconf(_SC_PAGE_SIZE);
-
- if ((p = calloc(1, psize)) == NULL)
- return ENOMEM;
- p->mp_flags = P_HEAD;
+ for (i = 0; i < env->me_numdbs; i++) {
+ if (env->me_dbs[toggle][i].md_root != txn->mt_dbs[i].md_root)
+ env->me_dbs[toggle][i] = txn->mt_dbs[i];
+ }
+ for (i = env->me_numdbs; i < txn->mt_numdbs; i++) {
+ env->me_dbxs[i] = txn->mt_dbxs[i];
+ env->me_dbs[toggle][i] = txn->mt_dbs[i];
+ }
+ env->me_db_toggle = toggle;
+ env->me_numdbs = txn->mt_numdbs;
- env->me_head.mh_psize = psize;
- env->me_head.mh_flags = env->me_flags & 0xffff;
+ free(txn->mt_dbs);
+ }
- h = METADATA(p);
- bcopy(&env->me_head, h, sizeof(*h));
+ pthread_mutex_unlock(&env->me_txns->mt_wmutex);
+ free(txn->mt_free_pgs);
+ free(txn->mt_u.dirty_queue);
+ free(txn);
+ txn = NULL;
- rc = write(env->me_fd, p, env->me_head.mh_psize);
- free(p);
- if (rc != (ssize_t)env->me_head.mh_psize) {
- int err = errno;
- if (rc > 0)
- DPRINTF("short write, filesystem full?");
- return err;
- }
+done:
+ mdb_txn_abort(txn);
return MDB_SUCCESS;
}
static int
-mdbenv_read_header(MDB_env *env)
+mdbenv_read_header(MDB_env *env, MDB_meta *meta)
{
char page[PAGESIZE];
MDB_page *p;
- MDB_head *h;
+ MDB_meta *m;
int rc;
assert(env != NULL);
p = (MDB_page *)page;
- if (!F_ISSET(p->mp_flags, P_HEAD)) {
- DPRINTF("page %lu not a header page", p->mp_pgno);
+ if (!F_ISSET(p->mp_flags, P_META)) {
+ DPRINTF("page %lu not a meta page", p->mp_pgno);
return EINVAL;
}
- h = METADATA(p);
- if (h->mh_magic != MDB_MAGIC) {
- DPRINTF("header has invalid magic");
+ m = METADATA(p);
+ if (m->mm_magic != MDB_MAGIC) {
+ DPRINTF("meta has invalid magic");
return EINVAL;
}
- if (h->mh_version != MDB_VERSION) {
+ if (m->mm_version != MDB_VERSION) {
DPRINTF("database is version %u, expected version %u",
- h->mh_version, MDB_VERSION);
+ m->mm_version, MDB_VERSION);
return EINVAL;
}
- bcopy(h, &env->me_head, sizeof(*h));
+ memcpy(meta, m, sizeof(*m));
return 0;
}
static int
-mdbenv_write_meta(MDB_env *env, pgno_t root, unsigned int flags)
+mdbenv_init_meta(MDB_env *env, MDB_meta *meta)
{
- MDB_dpage *dp;
- MDB_meta *meta;
- ssize_t rc;
-
- DPRINTF("writing meta page for root page %lu", root);
-
- assert(env != NULL);
- assert(env->me_txn != NULL);
-
- if ((dp = mdbenv_new_page(env, P_META, 1)) == NULL)
- return ENOMEM;
-
- env->me_meta.mm_prev_meta = env->me_meta.mm_root;
- env->me_meta.mm_root = root;
- env->me_meta.mm_flags = flags;
- env->me_meta.mm_created_at = time(0);
- env->me_meta.mm_revisions++;
- SHA1((unsigned char *)&env->me_meta, METAHASHLEN, env->me_meta.mm_hash);
-
- /* Copy the meta data changes to the new meta page. */
- meta = METADATA(&dp->p);
- bcopy(&env->me_meta, meta, sizeof(*meta));
-
- rc = write(env->me_fd, &dp->p, env->me_head.mh_psize);
- SIMPLEQ_REMOVE_HEAD(env->me_txn->mt_dirty_queue, h.md_next);
- free(dp);
- if (rc != (ssize_t)env->me_head.mh_psize) {
- int err = errno;
- if (rc > 0)
- DPRINTF("short write, filesystem full?");
- return err;
- }
+ MDB_page *p, *q;
+ MDB_meta *m;
+ int rc;
+ unsigned int psize;
- if ((env->me_size = lseek(env->me_fd, 0, SEEK_END)) == -1) {
- DPRINTF("failed to update file size: %s", strerror(errno));
- env->me_size = 0;
- }
+ DPRINTF("writing new meta page");
+ psize = sysconf(_SC_PAGE_SIZE);
- return MDB_SUCCESS;
-}
+ meta->mm_magic = MDB_MAGIC;
+ meta->mm_version = MDB_VERSION;
+ 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_dbs[0].md_root = P_INVALID;
+ meta->mm_dbs[1].md_root = P_INVALID;
-/* Returns true if page p is a valid meta page, false otherwise.
- */
-static int
-mdb_check_meta_page(MDB_page *p)
-{
- MDB_meta *m;
- unsigned char hash[SHA_DIGEST_LENGTH];
+ p = calloc(2, psize);
+ p->mp_pgno = 0;
+ p->mp_flags = P_META;
m = METADATA(p);
- if (!F_ISSET(p->mp_flags, P_META)) {
- DPRINTF("page %lu not a meta page", p->mp_pgno);
- return EINVAL;
- }
+ memcpy(m, meta, sizeof(*meta));
- if (m->mm_root >= p->mp_pgno && m->mm_root != P_INVALID) {
- DPRINTF("page %lu points to an invalid root page", p->mp_pgno);
- return EINVAL;
- }
+ q = (MDB_page *)((char *)p + psize);
- SHA1((unsigned char *)m, METAHASHLEN, hash);
- if (bcmp(hash, m->mm_hash, SHA_DIGEST_LENGTH) != 0) {
- DPRINTF("page %lu has an invalid digest", p->mp_pgno);
- return EINVAL;
- }
+ q->mp_pgno = 1;
+ q->mp_flags = P_META;
- return 0;
+ m = METADATA(q);
+ memcpy(m, meta, sizeof(*meta));
+
+ rc = write(env->me_fd, p, psize * 2);
+ free(p);
+ return (rc == (int)psize * 2) ? MDB_SUCCESS : errno;
}
static int
-mdbenv_read_meta(MDB_env *env, pgno_t *p_next)
+mdbenv_write_meta(MDB_txn *txn)
{
- MDB_page *mp;
- MDB_meta *meta;
- pgno_t meta_pgno, next_pgno;
- off_t size;
+ MDB_env *env;
+ MDB_meta meta;
+ off_t off;
+ int rc, len;
+ char *ptr;
- assert(env != NULL);
+ assert(txn != NULL);
+ assert(txn->mt_env != NULL);
- if ((size = lseek(env->me_fd, 0, SEEK_END)) == -1)
- goto fail;
+ DPRINTF("writing meta page for root page %lu", txn->mt_dbs[MAIN_DBI].md_root);
- DPRINTF("mdbenv_read_meta: size = %lu", size);
+ env = txn->mt_env;
- if (size < env->me_size) {
- DPRINTF("file has shrunk!");
- errno = EIO;
- goto fail;
- }
+ ptr = (char *)&meta;
+ off = offsetof(MDB_meta, mm_dbs[0].md_depth);
+ len = sizeof(MDB_meta) - off;
- if (size == env->me_head.mh_psize) { /* there is only the header */
- if (p_next != NULL)
- *p_next = 1;
- env->me_meta.mm_stat.ms_psize = env->me_head.mh_psize;
- return MDB_SUCCESS; /* new file */
- }
+ ptr += off;
+ meta.mm_dbs[0] = txn->mt_dbs[0];
+ meta.mm_dbs[1] = txn->mt_dbs[1];
+ meta.mm_last_pg = txn->mt_next_pgno - 1;
+ meta.mm_txnid = txn->mt_txnid;
- next_pgno = size / env->me_head.mh_psize;
- if (next_pgno == 0) {
- DPRINTF("corrupt file");
- errno = EIO;
- goto fail;
+ if (!F_ISSET(txn->mt_flags, MDB_TXN_METOGGLE))
+ off += env->me_psize;
+ off += PAGEHDRSZ;
+
+ lseek(env->me_fd, off, SEEK_SET);
+ rc = write(env->me_fd, ptr, len);
+ if (rc != len) {
+ DPRINTF("write failed, disk error?");
+ return errno;
}
- meta_pgno = next_pgno - 1;
+ return MDB_SUCCESS;
+}
- if (size % env->me_head.mh_psize != 0) {
- DPRINTF("filesize not a multiple of the page size!");
- env->me_flags |= MDB_FIXPADDING;
- next_pgno++;
- }
+static int
+mdbenv_read_meta(MDB_env *env, int *which)
+{
+ int toggle = 0;
- if (p_next != NULL)
- *p_next = next_pgno;
+ assert(env != NULL);
- if (size == env->me_size) {
- DPRINTF("size unchanged, keeping current meta page");
- if (F_ISSET(env->me_meta.mm_flags, MDB_TOMBSTONE)) {
- DPRINTF("file is dead");
- errno = ESTALE;
- return MDB_FAIL;
- } else
- return MDB_SUCCESS;
- } else {
- env->me_size = size;
- }
+ if (env->me_metas[0]->mm_txnid < env->me_metas[1]->mm_txnid)
+ toggle = 1;
- while (meta_pgno > 0) {
- if ((mp = mdbenv_get_page(env, meta_pgno)) == NULL)
- break;
- if (!mdb_check_meta_page(mp)) {
- meta = METADATA(mp);
- DPRINTF("flags = 0x%x", meta->mm_flags);
- if (F_ISSET(meta->mm_flags, MDB_TOMBSTONE)) {
- DPRINTF("file is dead");
- errno = ESTALE;
- return MDB_FAIL;
- } else {
- /* Make copy of last meta page. */
- bcopy(meta, &env->me_meta, sizeof(env->me_meta));
- return MDB_SUCCESS;
- }
- }
- --meta_pgno; /* scan backwards to first valid meta page */
- }
+ if (env->me_meta != env->me_metas[toggle])
+ env->me_meta = env->me_metas[toggle];
+ if (which)
+ *which = toggle;
- errno = EIO;
-fail:
- if (p_next != NULL)
- *p_next = P_INVALID;
- return MDB_FAIL;
+ DPRINTF("Using meta page %d", toggle);
+
+ return MDB_SUCCESS;
}
int
-mdbenv_create(MDB_env **env, size_t size)
+mdbenv_create(MDB_env **env)
{
MDB_env *e;
e = calloc(1, sizeof(*e));
if (!e) return ENOMEM;
- e->me_head.mh_magic = MDB_MAGIC;
- e->me_head.mh_version = MDB_VERSION;
- e->me_mapsize = e->me_head.mh_mapsize = size;
- e->me_db.md_env = e;
+ e->me_maxreaders = DEFAULT_READERS;
+ e->me_maxdbs = 2;
+ e->me_fd = -1;
+ e->me_lfd = -1;
*env = e;
return MDB_SUCCESS;
}
+int
+mdbenv_set_mapsize(MDB_env *env, size_t size)
+{
+ if (env->me_map)
+ return EINVAL;
+ env->me_mapsize = size;
+ return MDB_SUCCESS;
+}
+
+int
+mdbenv_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)
+{
+ env->me_maxreaders = readers;
+ return MDB_SUCCESS;
+}
+
+int
+mdbenv_get_maxreaders(MDB_env *env, int *readers)
+{
+ if (!env || !readers)
+ return EINVAL;
+ *readers = env->me_maxreaders;
+ return MDB_SUCCESS;
+}
+
int
mdbenv_open2(MDB_env *env, unsigned int flags)
{
int i, newenv = 0;
-
- i = fcntl(env->me_fd, F_GETFL, 0);
- if (fcntl(env->me_fd, F_SETFL, i | O_APPEND) == -1)
- return errno;
+ MDB_meta meta;
+ MDB_page *p;
env->me_flags = flags;
- env->me_flags &= ~MDB_FIXPADDING;
- env->me_meta.mm_root = P_INVALID;
- if ((i = mdbenv_read_header(env)) != 0) {
+ memset(&meta, 0, sizeof(meta));
+
+ if ((i = mdbenv_read_header(env, &meta)) != 0) {
if (i != ENOENT)
return i;
DPRINTF("new mdbenv");
newenv = 1;
}
- if (!env->me_mapsize)
- env->me_mapsize = env->me_head.mh_mapsize;
+ if (!env->me_mapsize) {
+ env->me_mapsize = newenv ? DEFAULT_MAPSIZE : meta.mm_mapsize;
+ }
i = MAP_SHARED;
- if (env->me_head.mh_address && (flags & MDB_FIXEDMAP))
+ if (meta.mm_address && (flags & MDB_FIXEDMAP))
i |= MAP_FIXED;
- env->me_map = mmap(env->me_head.mh_address, env->me_mapsize, PROT_READ, i,
+ env->me_map = mmap(meta.mm_address, env->me_mapsize, PROT_READ, i,
env->me_fd, 0);
if (env->me_map == MAP_FAILED)
return errno;
if (newenv) {
- env->me_head.mh_mapsize = env->me_mapsize;
+ meta.mm_mapsize = env->me_mapsize;
if (flags & MDB_FIXEDMAP)
- env->me_head.mh_address = env->me_map;
- i = mdbenv_write_header(env);
+ meta.mm_address = env->me_map;
+ i = mdbenv_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_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)
return i;
DPRINTF("opened database version %u, pagesize %u",
- env->me_head.mh_version, env->me_head.mh_psize);
- DPRINTF("timestamp: %s", ctime(&env->me_meta.mm_created_at));
- DPRINTF("depth: %u", env->me_meta.mm_stat.ms_depth);
- DPRINTF("entries: %lu", env->me_meta.mm_stat.ms_entries);
- DPRINTF("revisions: %lu", env->me_meta.mm_stat.ms_revisions);
- DPRINTF("branch pages: %lu", env->me_meta.mm_stat.ms_branch_pages);
- DPRINTF("leaf pages: %lu", env->me_meta.mm_stat.ms_leaf_pages);
- DPRINTF("overflow pages: %lu", env->me_meta.mm_stat.ms_overflow_pages);
- DPRINTF("root: %lu", env->me_meta.mm_root);
- DPRINTF("previous me_meta.me_page: %lu", env->me_meta.mm_prev_meta);
+ env->me_meta->mm_version, env->me_psize);
+ DPRINTF("depth: %u", env->me_meta->mm_dbs[MAIN_DBI].md_depth);
+ DPRINTF("entries: %lu", env->me_meta->mm_dbs[MAIN_DBI].md_entries);
+ DPRINTF("branch pages: %lu", env->me_meta->mm_dbs[MAIN_DBI].md_branch_pages);
+ DPRINTF("leaf pages: %lu", env->me_meta->mm_dbs[MAIN_DBI].md_leaf_pages);
+ DPRINTF("overflow pages: %lu", env->me_meta->mm_dbs[MAIN_DBI].md_overflow_pages);
+ DPRINTF("root: %lu", env->me_meta->mm_dbs[MAIN_DBI].md_root);
return MDB_SUCCESS;
}
+static void
+mdbenv_reader_dest(void *ptr)
+{
+ MDB_reader *reader = ptr;
+
+ reader->mr_txnid = 0;
+ reader->mr_pid = 0;
+ reader->mr_tid = 0;
+}
+
+/* downgrade the exclusive lock on the region back to shared */
+static void
+mdbenv_share_locks(MDB_env *env)
+{
+ struct flock lock_info;
+
+ env->me_txns->mt_txnid = env->me_meta->mm_txnid;
+
+ memset((void *)&lock_info, 0, sizeof(lock_info));
+ lock_info.l_type = F_RDLCK;
+ lock_info.l_whence = SEEK_SET;
+ lock_info.l_start = 0;
+ lock_info.l_len = 1;
+ fcntl(env->me_lfd, F_SETLK, &lock_info);
+}
+
+static int
+mdbenv_setup_locks(MDB_env *env, char *lpath, int mode, int *excl)
+{
+ int rc;
+ off_t size, rsize;
+ struct flock lock_info;
+
+ *excl = 0;
+
+ if ((env->me_lfd = open(lpath, O_RDWR|O_CREAT, mode)) == -1) {
+ rc = errno;
+ return rc;
+ }
+ /* Try to get exclusive lock. If we succeed, then
+ * nobody is using the lock region and we should initialize it.
+ */
+ memset((void *)&lock_info, 0, sizeof(lock_info));
+ lock_info.l_type = F_WRLCK;
+ lock_info.l_whence = SEEK_SET;
+ lock_info.l_start = 0;
+ lock_info.l_len = 1;
+ rc = fcntl(env->me_lfd, F_SETLK, &lock_info);
+ if (rc == 0) {
+ *excl = 1;
+ } else {
+ lock_info.l_type = F_RDLCK;
+ rc = fcntl(env->me_lfd, F_SETLK, &lock_info);
+ if (rc) {
+ rc = errno;
+ goto fail;
+ }
+ }
+ size = lseek(env->me_lfd, 0, SEEK_END);
+ rsize = (env->me_maxreaders-1) * sizeof(MDB_reader) + sizeof(MDB_txninfo);
+ if (size < rsize && *excl) {
+ if (ftruncate(env->me_lfd, rsize) != 0) {
+ rc = errno;
+ goto fail;
+ }
+ } else {
+ rsize = size;
+ size = rsize - sizeof(MDB_txninfo);
+ env->me_maxreaders = size/sizeof(MDB_reader) + 1;
+ }
+ env->me_txns = mmap(0, rsize, PROT_READ|PROT_WRITE, MAP_SHARED,
+ env->me_lfd, 0);
+ if (env->me_txns == MAP_FAILED) {
+ rc = errno;
+ goto fail;
+ }
+ if (*excl) {
+ 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;
+
+ } else {
+ if (env->me_txns->mt_magic != MDB_MAGIC) {
+ DPRINTF("lock region has invalid magic");
+ errno = EINVAL;
+ }
+ if (env->me_txns->mt_version != MDB_VERSION) {
+ DPRINTF("lock region is version %u, expected version %u",
+ env->me_txns->mt_version, MDB_VERSION);
+ errno = EINVAL;
+ }
+ if (errno != EACCES && errno != EAGAIN) {
+ rc = errno;
+ goto fail;
+ }
+ }
+ return MDB_SUCCESS;
+
+fail:
+ close(env->me_lfd);
+ return rc;
+
+}
+
int
mdbenv_open(MDB_env *env, const char *path, unsigned int flags, mode_t mode)
{
- int oflags, rc;
+ int oflags, rc, len, excl;
+ char *lpath, *dpath;
+
+ len = strlen(path);
+ lpath = malloc(len + sizeof("/lock.mdb") + len + sizeof("/data.db"));
+ if (!lpath)
+ return ENOMEM;
+ dpath = lpath + len + sizeof("/lock.mdb");
+ sprintf(lpath, "%s/lock.mdb", path);
+ sprintf(dpath, "%s/data.mdb", path);
+
+ rc = mdbenv_setup_locks(env, lpath, mode, &excl);
+ if (rc)
+ goto leave;
if (F_ISSET(flags, MDB_RDONLY))
oflags = O_RDONLY;
else
- oflags = O_RDWR | O_CREAT | O_APPEND;
+ oflags = O_RDWR | O_CREAT;
- if ((env->me_fd = open(path, oflags, mode)) == -1)
+ if ((env->me_fd = open(dpath, oflags, mode)) == -1)
return errno;
if ((rc = mdbenv_open2(env, flags)) != MDB_SUCCESS) {
env->me_fd = -1;
} else {
env->me_path = strdup(path);
- DPRINTF("opened dbenv %p", env);
- }
-
+ DPRINTF("opened dbenv %p", (void *) env);
+ pthread_key_create(&env->me_txkey, mdbenv_reader_dest);
+ if (excl)
+ mdbenv_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));
+ env->me_numdbs = 2;
+ }
+
+leave:
+ free(lpath);
return rc;
}
if (env == NULL)
return;
+ free(env->me_dbxs);
free(env->me_path);
if (env->me_map) {
munmap(env->me_map, env->me_mapsize);
}
close(env->me_fd);
+ if (env->me_txns) {
+ size_t size = (env->me_maxreaders-1) * sizeof(MDB_reader) + sizeof(MDB_txninfo);
+ munmap(env->me_txns, size);
+ }
+ close(env->me_lfd);
+ free(env);
}
/* Search for key within a leaf page, using binary search.
* If no entry larger or equal to the key is found, returns NULL.
*/
static MDB_node *
-mdb_search_node(MDB_db *bt, MDB_page *mp, MDB_val *key,
+mdb_search_node(MDB_txn *txn, MDB_dbi dbi, MDB_page *mp, MDB_val *key,
int *exactp, unsigned int *kip)
{
unsigned int i = 0;
MDB_node *node;
MDB_val nodekey;
- DPRINTF("searching %lu keys in %s page %lu",
+ DPRINTF("searching %u keys in %s page %lu",
NUMKEYS(mp),
IS_LEAF(mp) ? "leaf" : "branch",
mp->mp_pgno);
assert(NUMKEYS(mp) > 0);
- bzero(&nodekey, sizeof(nodekey));
+ memset(&nodekey, 0, sizeof(nodekey));
low = IS_LEAF(mp) ? 0 : 1;
high = NUMKEYS(mp) - 1;
nodekey.mv_size = node->mn_ksize;
nodekey.mv_data = NODEKEY(node);
- if (bt->md_cmp)
- rc = bt->md_cmp(key, &nodekey);
+ if (txn->mt_dbxs[dbi].md_cmp)
+ rc = txn->mt_dbxs[dbi].md_cmp(key, &nodekey);
else
- rc = _mdb_cmp(bt, key, &nodekey);
+ rc = _mdb_cmp(txn, dbi, key, &nodekey);
if (IS_LEAF(mp))
DPRINTF("found leaf index %u [%.*s], rc = %i",
top = CURSOR_TOP(cursor);
CURSOR_POP(cursor);
- DPRINTF("popped page %lu off cursor %p", top->mp_page->mp_pgno, cursor);
+ DPRINTF("popped page %lu off cursor %p", top->mp_page->mp_pgno, (void *) cursor);
free(top);
}
{
MDB_ppage *ppage;
- DPRINTF("pushing page %lu on cursor %p", mp->mp_pgno, cursor);
+ DPRINTF("pushing page %lu on cursor %p", mp->mp_pgno, (void *) cursor);
if ((ppage = calloc(1, sizeof(*ppage))) == NULL)
return NULL;
{
MDB_page *p = NULL;
MDB_txn *txn = env->me_txn;
+ int found = 0;
- if (txn && pgno >= txn->mt_first_pgno &&
- !SIMPLEQ_EMPTY(txn->mt_dirty_queue)) {
+ if (txn && !STAILQ_EMPTY(txn->mt_u.dirty_queue)) {
MDB_dpage *dp;
- SIMPLEQ_FOREACH(dp, txn->mt_dirty_queue, h.md_next) {
+ STAILQ_FOREACH(dp, txn->mt_u.dirty_queue, h.md_next) {
if (dp->p.mp_pgno == pgno) {
p = &dp->p;
+ found = 1;
break;
}
}
- } else {
- p = (MDB_page *)(env->me_map + env->me_head.mh_psize * pgno);
+ }
+ if (!found) {
+ p = (MDB_page *)(env->me_map + env->me_psize * pgno);
}
return p;
}
static int
-mdb_search_page_root(MDB_db *bt, MDB_val *key,
+mdb_search_page_root(MDB_txn *txn, MDB_dbi dbi, MDB_val *key,
MDB_cursor *cursor, int modify, MDB_pageparent *mpp)
{
MDB_page *mp = mpp->mp_page;
unsigned int i = 0;
MDB_node *node;
- DPRINTF("branch page %lu has %lu keys", mp->mp_pgno, NUMKEYS(mp));
+ DPRINTF("branch page %lu has %u keys", mp->mp_pgno, NUMKEYS(mp));
assert(NUMKEYS(mp) > 1);
DPRINTF("found index 0 to page %lu", NODEPGNO(NODEPTR(mp, 0)));
if (key == NULL) /* Initialize cursor to first page. */
i = 0;
- else {
+ else if (key->mv_size > MAXKEYSIZE && key->mv_data == NULL) {
+ /* cursor to last page */
+ i = NUMKEYS(mp)-1;
+ } else {
int exact;
- node = mdb_search_node(bt, mp, key, &exact, &i);
+ node = mdb_search_node(txn, dbi, mp, key, &exact, &i);
if (node == NULL)
i = NUMKEYS(mp) - 1;
else if (!exact) {
if (key)
DPRINTF("following index %u for key %.*s",
i, (int)key->mv_size, (char *)key->mv_data);
- assert(i >= 0 && i < NUMKEYS(mp));
+ assert(i < NUMKEYS(mp));
node = NODEPTR(mp, i);
if (cursor)
CURSOR_TOP(cursor)->mp_ki = i;
mpp->mp_parent = mp;
- if ((mp = mdbenv_get_page(bt->md_env, NODEPGNO(node))) == NULL)
+ if ((mp = mdbenv_get_page(txn->mt_env, NODEPGNO(node))) == NULL)
return MDB_FAIL;
mpp->mp_pi = i;
mpp->mp_page = mp;
if (modify) {
MDB_dhead *dh = ((MDB_dhead *)mp)-1;
- if (rc = mdb_touch(bt->md_env->me_txn, mpp))
+ if ((rc = mdb_touch(txn, mpp)) != 0)
return rc;
dh = ((MDB_dhead *)mpp->mp_page)-1;
dh->md_parent = mpp->mp_parent;
* If modify is true, visited pages are updated with new page numbers.
*/
static int
-mdb_search_page(MDB_db *db, MDB_txn *txn, MDB_val *key,
+mdb_search_page(MDB_txn *txn, MDB_dbi dbi, MDB_val *key,
MDB_cursor *cursor, int modify, MDB_pageparent *mpp)
{
int rc;
pgno_t root;
- /* Can't modify pages outside a transaction. */
- if (txn == NULL && modify) {
- return EINVAL;
- }
-
/* Choose which root page to start with. If a transaction is given
* use the root page from the transaction, otherwise read the last
* committed root page.
*/
- if (txn == NULL) {
- if ((rc = mdbenv_read_meta(db->md_env, NULL)) != MDB_SUCCESS)
- return rc;
- root = db->md_env->me_meta.mm_root;
- } else if (F_ISSET(txn->mt_flags, MDB_TXN_ERROR)) {
+ if (F_ISSET(txn->mt_flags, MDB_TXN_ERROR)) {
DPRINTF("transaction has failed, must abort");
return EINVAL;
} else
- root = txn->mt_root;
+ root = txn->mt_dbs[dbi].md_root;
if (root == P_INVALID) { /* Tree is empty. */
DPRINTF("tree is empty");
return ENOENT;
}
- if ((mpp->mp_page = mdbenv_get_page(db->md_env, root)) == NULL)
+ if ((mpp->mp_page = mdbenv_get_page(txn->mt_env, root)) == NULL)
return MDB_FAIL;
DPRINTF("root page has flags 0x%X", mpp->mp_page->mp_flags);
- if (modify && !F_ISSET(mpp->mp_page->mp_flags, P_DIRTY)) {
- mpp->mp_parent = NULL;
- mpp->mp_pi = 0;
- if ((rc = mdb_touch(txn, mpp)))
- return rc;
- txn->mt_root = mpp->mp_page->mp_pgno;
+ if (modify) {
+ /* For sub-databases, update main root first */
+ if (dbi > MAIN_DBI && txn->mt_dbs[dbi].md_root ==
+ txn->mt_env->me_dbs[txn->mt_env->me_db_toggle][dbi].md_root) {
+ MDB_pageparent mp2;
+ rc = mdb_search_page(txn, 0, &txn->mt_dbxs[dbi].md_name,
+ NULL, 1, &mp2);
+ if (rc)
+ return rc;
+ }
+ if (!F_ISSET(mpp->mp_page->mp_flags, P_DIRTY)) {
+ mpp->mp_parent = NULL;
+ mpp->mp_pi = 0;
+ if ((rc = mdb_touch(txn, mpp)))
+ return rc;
+ txn->mt_dbs[dbi].md_root = mpp->mp_page->mp_pgno;
+ }
}
- return mdb_search_page_root(db, key, cursor, modify, mpp);
+ return mdb_search_page_root(txn, dbi, key, cursor, modify, mpp);
}
static int
-mdb_read_data(MDB_db *db, MDB_page *mp, MDB_node *leaf,
- MDB_val *data)
+mdb_read_data(MDB_env *env, MDB_node *leaf, MDB_val *data)
{
MDB_page *omp; /* overflow mpage */
- size_t psz;
- size_t max;
- size_t sz = 0;
pgno_t pgno;
- bzero(data, sizeof(*data));
- max = db->md_env->me_head.mh_psize - PAGEHDRSZ;
-
if (!F_ISSET(leaf->mn_flags, F_BIGDATA)) {
data->mv_size = leaf->mn_dsize;
data->mv_data = NODEDATA(leaf);
/* Read overflow data.
*/
data->mv_size = leaf->mn_dsize;
- bcopy(NODEDATA(leaf), &pgno, sizeof(pgno));
- if ((omp = mdbenv_get_page(db->md_env, pgno)) == NULL) {
+ memcpy(&pgno, NODEDATA(leaf), sizeof(pgno));
+ if ((omp = mdbenv_get_page(env, pgno)) == NULL) {
DPRINTF("read overflow page %lu failed", pgno);
return MDB_FAIL;
}
}
int
-mdb_get(MDB_db *db, MDB_txn *txn,
+mdb_get(MDB_txn *txn, MDB_dbi dbi,
MDB_val *key, MDB_val *data)
{
int rc, exact;
assert(data);
DPRINTF("===> get key [%.*s]", (int)key->mv_size, (char *)key->mv_data);
- if (db == NULL)
- return EINVAL;
-
- if (txn != NULL && db->md_env != txn->mt_env) {
- return EINVAL;
- }
-
if (key->mv_size == 0 || key->mv_size > MAXKEYSIZE) {
return EINVAL;
}
- if ((rc = mdb_search_page(db, txn, key, NULL, 0, &mpp)) != MDB_SUCCESS)
+ if ((rc = mdb_search_page(txn, dbi, key, NULL, 0, &mpp)) != MDB_SUCCESS)
return rc;
- leaf = mdb_search_node(db, mpp.mp_page, key, &exact, NULL);
+ leaf = mdb_search_node(txn, dbi, mpp.mp_page, key, &exact, NULL);
if (leaf && exact)
- rc = mdb_read_data(db, mpp.mp_page, leaf, data);
+ rc = mdb_read_data(txn->mt_env, leaf, data);
else {
rc = ENOENT;
}
assert(IS_BRANCH(parent->mp_page));
indx = NODEPTR(parent->mp_page, parent->mp_ki);
- if ((mp = mdbenv_get_page(cursor->mc_db->md_env, indx->mn_pgno)) == NULL)
+ if ((mp = mdbenv_get_page(cursor->mc_txn->mt_env, indx->mn_pgno)) == NULL)
return MDB_FAIL;
#if 0
mp->parent = parent->mp_page;
}
static int
-mdb_set_key(MDB_db *bt, MDB_page *mp, MDB_node *node,
- MDB_val *key)
+mdb_set_key(MDB_node *node, MDB_val *key)
{
if (key == NULL)
return 0;
top = CURSOR_TOP(cursor);
mp = top->mp_page;
- DPRINTF("cursor_next: top page is %lu in cursor %p", mp->mp_pgno, cursor);
+ 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");
} else
top->mp_ki++;
- DPRINTF("==> cursor points to page %lu with %lu keys, key index %u",
+ DPRINTF("==> cursor points to page %lu with %u keys, key index %u",
mp->mp_pgno, NUMKEYS(mp), top->mp_ki);
assert(IS_LEAF(mp));
leaf = NODEPTR(mp, top->mp_ki);
- if (data && mdb_read_data(cursor->mc_db, mp, leaf, data) != MDB_SUCCESS)
+ if (data && mdb_read_data(cursor->mc_txn->mt_env, leaf, data) != MDB_SUCCESS)
return MDB_FAIL;
- return mdb_set_key(cursor->mc_db, mp, leaf, key);
+ return mdb_set_key(leaf, key);
}
static int
assert(key);
assert(key->mv_size > 0);
- rc = mdb_search_page(cursor->mc_db, cursor->mc_txn, key, cursor, 0, &mpp);
+ rc = mdb_search_page(cursor->mc_txn, cursor->mc_dbi, key, cursor, 0, &mpp);
if (rc != MDB_SUCCESS)
return rc;
assert(IS_LEAF(mpp.mp_page));
top = CURSOR_TOP(cursor);
- leaf = mdb_search_node(cursor->mc_db, mpp.mp_page, key, exactp, &top->mp_ki);
+ leaf = mdb_search_node(cursor->mc_txn, cursor->mc_dbi, mpp.mp_page, key, exactp, &top->mp_ki);
if (exactp != NULL && !*exactp) {
/* MDB_CURSOR_EXACT specified and not an exact match. */
return ENOENT;
cursor->mc_initialized = 1;
cursor->mc_eof = 0;
- if (data && (rc = mdb_read_data(cursor->mc_db, mpp.mp_page, leaf, data)) != MDB_SUCCESS)
+ if (data && (rc = mdb_read_data(cursor->mc_txn->mt_env, leaf, data)) != MDB_SUCCESS)
return rc;
- rc = mdb_set_key(cursor->mc_db, mpp.mp_page, leaf, key);
+ rc = mdb_set_key(leaf, key);
if (rc == MDB_SUCCESS) {
DPRINTF("==> cursor placed on key %.*s",
(int)key->mv_size, (char *)key->mv_data);
MDB_pageparent mpp;
MDB_node *leaf;
- rc = mdb_search_page(cursor->mc_db, cursor->mc_txn, NULL, cursor, 0, &mpp);
+ rc = mdb_search_page(cursor->mc_txn, cursor->mc_dbi, NULL, cursor, 0, &mpp);
if (rc != MDB_SUCCESS)
return rc;
assert(IS_LEAF(mpp.mp_page));
cursor->mc_initialized = 1;
cursor->mc_eof = 0;
- if (data && (rc = mdb_read_data(cursor->mc_db, mpp.mp_page, leaf, data)) != MDB_SUCCESS)
+ if (data && (rc = mdb_read_data(cursor->mc_txn->mt_env, leaf, data)) != MDB_SUCCESS)
+ return rc;
+
+ return mdb_set_key(leaf, key);
+}
+
+static int
+mdb_cursor_last(MDB_cursor *cursor, MDB_val *key, MDB_val *data)
+{
+ int rc;
+ MDB_pageparent mpp;
+ MDB_node *leaf;
+ MDB_val lkey;
+
+ lkey.mv_size = MAXKEYSIZE+1;
+ lkey.mv_data = NULL;
+
+ rc = mdb_search_page(cursor->mc_txn, cursor->mc_dbi, &lkey, cursor, 0, &mpp);
+ if (rc != MDB_SUCCESS)
+ return rc;
+ assert(IS_LEAF(mpp.mp_page));
+
+ leaf = NODEPTR(mpp.mp_page, NUMKEYS(mpp.mp_page)-1);
+ cursor->mc_initialized = 1;
+ cursor->mc_eof = 1;
+
+ if (data && (rc = mdb_read_data(cursor->mc_txn->mt_env, leaf, data)) != MDB_SUCCESS)
return rc;
- return mdb_set_key(cursor->mc_db, mpp.mp_page, leaf, key);
+ return mdb_set_key(leaf, key);
}
int
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:
DPRINTF("unhandled/unimplemented cursor operation %u", op);
rc = EINVAL;
/* Allocate a page and initialize it
*/
static MDB_dpage *
-mdbenv_new_page(MDB_env *env, uint32_t flags, int num)
+mdb_new_page(MDB_txn *txn, MDB_dbi dbi, uint32_t flags, int num)
{
MDB_dpage *dp;
- assert(env != NULL);
- assert(env->me_txn != NULL);
-
- DPRINTF("allocating new mpage %lu, page size %u",
- env->me_txn->mt_next_pgno, env->me_head.mh_psize);
- if ((dp = mdb_newpage(env->me_txn, NULL, 0, num)) == NULL)
+ if ((dp = mdb_alloc_page(txn, NULL, 0, num)) == NULL)
return NULL;
+ DPRINTF("allocated new mpage %lu, page size %u",
+ dp->p.mp_pgno, txn->mt_env->me_psize);
dp->p.mp_flags = flags | P_DIRTY;
dp->p.mp_lower = PAGEHDRSZ;
- dp->p.mp_upper = env->me_head.mh_psize;
+ dp->p.mp_upper = txn->mt_env->me_psize;
if (IS_BRANCH(&dp->p))
- env->me_meta.mm_stat.ms_branch_pages++;
+ txn->mt_dbs[dbi].md_branch_pages++;
else if (IS_LEAF(&dp->p))
- env->me_meta.mm_stat.ms_leaf_pages++;
+ txn->mt_dbs[dbi].md_leaf_pages++;
else if (IS_OVERFLOW(&dp->p)) {
- env->me_meta.mm_stat.ms_overflow_pages += num;
+ txn->mt_dbs[dbi].md_overflow_pages += num;
dp->p.mp_pages = num;
}
}
static size_t
-mdb_leaf_size(MDB_db *db, MDB_val *key, MDB_val *data)
+mdb_leaf_size(MDB_env *env, MDB_val *key, MDB_val *data)
{
size_t sz;
sz = LEAFSIZE(key, data);
- if (data->mv_size >= db->md_env->me_head.mh_psize / MDB_MINKEYS) {
+ if (data->mv_size >= env->me_psize / MDB_MINKEYS) {
/* put on overflow page */
sz -= data->mv_size - sizeof(pgno_t);
}
}
static size_t
-mdb_branch_size(MDB_db *db, MDB_val *key)
+mdb_branch_size(MDB_env *env, MDB_val *key)
{
size_t sz;
sz = INDXSIZE(key);
- if (sz >= db->md_env->me_head.mh_psize / MDB_MINKEYS) {
+ if (sz >= env->me_psize / MDB_MINKEYS) {
/* put on overflow page */
/* not implemented */
/* sz -= key->size - sizeof(pgno_t); */
}
static int
-mdb_add_node(MDB_db *db, MDB_page *mp, indx_t indx,
+mdb_add_node(MDB_txn *txn, MDB_dbi dbi, MDB_page *mp, indx_t indx,
MDB_val *key, MDB_val *data, pgno_t pgno, uint8_t flags)
{
unsigned int i;
if (IS_LEAF(mp)) {
assert(data);
- node_size += data->mv_size;
if (F_ISSET(flags, F_BIGDATA)) {
/* Data already on overflow page. */
- node_size -= data->mv_size - sizeof(pgno_t);
- } else if (data->mv_size >= db->md_env->me_head.mh_psize / MDB_MINKEYS) {
- int ovpages = PAGEHDRSZ + data->mv_size + db->md_env->me_head.mh_psize - 1;
- ovpages /= db->md_env->me_head.mh_psize;
+ node_size += sizeof(pgno_t);
+ } else if (data->mv_size >= txn->mt_env->me_psize / MDB_MINKEYS) {
+ int ovpages = OVPAGES(data->mv_size, txn->mt_env->me_psize);
/* Put data on overflow page. */
DPRINTF("data size is %zu, put on overflow page",
data->mv_size);
- node_size -= data->mv_size - sizeof(pgno_t);
- if ((ofp = mdbenv_new_page(db->md_env, P_OVERFLOW, ovpages)) == NULL)
+ node_size += sizeof(pgno_t);
+ if ((ofp = mdb_new_page(txn, dbi, P_OVERFLOW, ovpages)) == NULL)
return MDB_FAIL;
DPRINTF("allocated overflow page %lu", ofp->p.mp_pgno);
flags |= F_BIGDATA;
+ } else {
+ node_size += data->mv_size;
}
}
if (node_size + sizeof(indx_t) > SIZELEFT(mp)) {
- DPRINTF("not enough room in page %lu, got %lu ptrs",
+ DPRINTF("not enough room in page %lu, got %u ptrs",
mp->mp_pgno, NUMKEYS(mp));
DPRINTF("upper - lower = %u - %u = %u", mp->mp_upper, mp->mp_lower,
mp->mp_upper - mp->mp_lower);
node->mn_pgno = pgno;
if (key)
- bcopy(key->mv_data, NODEKEY(node), key->mv_size);
+ memcpy(NODEKEY(node), key->mv_data, key->mv_size);
if (IS_LEAF(mp)) {
assert(key);
if (ofp == NULL) {
if (F_ISSET(flags, F_BIGDATA))
- bcopy(data->mv_data, node->mn_data + key->mv_size,
+ memcpy(node->mn_data + key->mv_size, data->mv_data,
sizeof(pgno_t));
else
- bcopy(data->mv_data, node->mn_data + key->mv_size,
+ memcpy(node->mn_data + key->mv_size, data->mv_data,
data->mv_size);
} else {
- bcopy(&ofp->p.mp_pgno, node->mn_data + key->mv_size,
+ memcpy(node->mn_data + key->mv_size, &ofp->p.mp_pgno,
sizeof(pgno_t));
- bcopy(data->mv_data, METADATA(&ofp->p), data->mv_size);
+ memcpy(METADATA(&ofp->p), data->mv_data, data->mv_size);
}
}
}
static void
-mdb_del_node(MDB_db *db, MDB_page *mp, indx_t indx)
+mdb_del_node(MDB_page *mp, indx_t indx)
{
unsigned int sz;
indx_t i, j, numkeys, ptr;
}
base = (char *)mp + mp->mp_upper;
- bcopy(base, base + sz, ptr - mp->mp_upper);
+ memmove(base + sz, base, ptr - mp->mp_upper);
mp->mp_lower -= sizeof(indx_t);
mp->mp_upper += sz;
}
int
-mdb_cursor_open(MDB_db *db, MDB_txn *txn, MDB_cursor **ret)
+mdb_cursor_open(MDB_txn *txn, MDB_dbi dbi, MDB_cursor **ret)
{
MDB_cursor *cursor;
- if (db == NULL || ret == NULL)
+ if (txn == NULL || ret == NULL)
return EINVAL;
- if (txn != NULL && db->md_env != txn->mt_env) {
- return EINVAL;
- }
-
if ((cursor = calloc(1, sizeof(*cursor))) != NULL) {
SLIST_INIT(&cursor->mc_stack);
- cursor->mc_db = db;
+ cursor->mc_dbi = dbi;
cursor->mc_txn = txn;
}
}
static int
-mdb_update_key(MDB_db *db, MDB_page *mp, indx_t indx,
- MDB_val *key)
+mdb_update_key(MDB_page *mp, indx_t indx, MDB_val *key)
{
indx_t ptr, i, numkeys;
int delta;
base = (char *)mp + mp->mp_upper;
len = ptr - mp->mp_upper + NODESIZE;
- bcopy(base, base - delta, len);
+ memmove(base - delta, base, len);
mp->mp_upper -= delta;
node = NODEPTR(mp, indx);
node->mn_ksize = key->mv_size;
}
- bcopy(key->mv_data, NODEKEY(node), key->mv_size);
+ memcpy(NODEKEY(node), key->mv_data, key->mv_size);
return MDB_SUCCESS;
}
/* Move a node from src to dst.
*/
static int
-mdb_move_node(MDB_db *bt, MDB_pageparent *src, indx_t srcindx,
+mdb_move_node(MDB_txn *txn, MDB_dbi dbi, MDB_pageparent *src, indx_t srcindx,
MDB_pageparent *dst, indx_t dstindx)
{
int rc;
dstindx, dst->mp_page->mp_pgno);
/* Mark src and dst as dirty. */
- if ((rc = mdb_touch(bt->md_env->me_txn, src)) ||
- (rc = mdb_touch(bt->md_env->me_txn, dst)))
+ if ((rc = mdb_touch(txn, src)) ||
+ (rc = mdb_touch(txn, dst)))
return rc;;
/* Add the node to the destination page.
key.mv_data = NODEKEY(srcnode);
data.mv_size = NODEDSZ(srcnode);
data.mv_data = NODEDATA(srcnode);
- rc = mdb_add_node(bt, dst->mp_page, dstindx, &key, &data, NODEPGNO(srcnode),
+ rc = mdb_add_node(txn, dbi, dst->mp_page, dstindx, &key, &data, NODEPGNO(srcnode),
srcnode->mn_flags);
if (rc != MDB_SUCCESS)
return rc;
/* Delete the node from the source page.
*/
- mdb_del_node(bt, src->mp_page, srcindx);
+ mdb_del_node(src->mp_page, srcindx);
/* Update the parent separators.
*/
if (srcindx == 0 && src->mp_pi != 0) {
DPRINTF("update separator for source page %lu to [%.*s]",
src->mp_page->mp_pgno, (int)key.mv_size, (char *)key.mv_data);
- if ((rc = mdb_update_key(bt, src->mp_parent, src->mp_pi,
+ if ((rc = mdb_update_key(src->mp_parent, src->mp_pi,
&key)) != MDB_SUCCESS)
return rc;
}
if (srcindx == 0 && IS_BRANCH(src->mp_page)) {
MDB_val nullkey;
nullkey.mv_size = 0;
- assert(mdb_update_key(bt, src->mp_page, 0, &nullkey) == MDB_SUCCESS);
+ assert(mdb_update_key(src->mp_page, 0, &nullkey) == MDB_SUCCESS);
}
if (dstindx == 0 && dst->mp_pi != 0) {
DPRINTF("update separator for destination page %lu to [%.*s]",
dst->mp_page->mp_pgno, (int)key.mv_size, (char *)key.mv_data);
- if ((rc = mdb_update_key(bt, dst->mp_parent, dst->mp_pi,
+ if ((rc = mdb_update_key(dst->mp_parent, dst->mp_pi,
&key)) != MDB_SUCCESS)
return rc;
}
if (dstindx == 0 && IS_BRANCH(dst->mp_page)) {
MDB_val nullkey;
nullkey.mv_size = 0;
- assert(mdb_update_key(bt, dst->mp_page, 0, &nullkey) == MDB_SUCCESS);
+ assert(mdb_update_key(dst->mp_page, 0, &nullkey) == MDB_SUCCESS);
}
return MDB_SUCCESS;
}
static int
-mdb_merge(MDB_db *bt, MDB_pageparent *src, MDB_pageparent *dst)
+mdb_merge(MDB_txn *txn, MDB_dbi dbi, MDB_pageparent *src, MDB_pageparent *dst)
{
int rc;
indx_t i;
DPRINTF("merging page %lu and %lu", src->mp_page->mp_pgno, dst->mp_page->mp_pgno);
+ assert(txn != NULL);
assert(src->mp_parent); /* can't merge root page */
assert(dst->mp_parent);
- assert(bt->md_env->me_txn != NULL);
/* Mark src and dst as dirty. */
- if ((rc = mdb_touch(bt->md_env->me_txn, src)) ||
- (rc = mdb_touch(bt->md_env->me_txn, dst)))
+ if ((rc = mdb_touch(txn, src)) ||
+ (rc = mdb_touch(txn, dst)))
return rc;
/* Move all nodes from src to dst.
key.mv_data = NODEKEY(srcnode);
data.mv_size = NODEDSZ(srcnode);
data.mv_data = NODEDATA(srcnode);
- rc = mdb_add_node(bt, dst->mp_page, NUMKEYS(dst->mp_page), &key,
+ rc = mdb_add_node(txn, dbi, dst->mp_page, NUMKEYS(dst->mp_page), &key,
&data, NODEPGNO(srcnode), srcnode->mn_flags);
if (rc != MDB_SUCCESS)
return rc;
}
- DPRINTF("dst page %lu now has %lu keys (%.1f%% filled)",
- dst->mp_page->mp_pgno, NUMKEYS(dst->mp_page), (float)PAGEFILL(bt->md_env, dst->mp_page) / 10);
+ DPRINTF("dst page %lu now has %u keys (%.1f%% filled)",
+ dst->mp_page->mp_pgno, NUMKEYS(dst->mp_page), (float)PAGEFILL(txn->mt_env, dst->mp_page) / 10);
/* Unlink the src page from parent.
*/
- mdb_del_node(bt, src->mp_parent, src->mp_pi);
+ mdb_del_node(src->mp_parent, src->mp_pi);
if (src->mp_pi == 0) {
key.mv_size = 0;
- if ((rc = mdb_update_key(bt, src->mp_parent, 0, &key)) != MDB_SUCCESS)
+ if ((rc = mdb_update_key(src->mp_parent, 0, &key)) != MDB_SUCCESS)
return rc;
}
if (IS_LEAF(src->mp_page))
- bt->md_env->me_meta.mm_stat.ms_leaf_pages--;
+ txn->mt_dbs[dbi].md_leaf_pages--;
else
- bt->md_env->me_meta.mm_stat.ms_branch_pages--;
+ txn->mt_dbs[dbi].md_branch_pages--;
mpp.mp_page = src->mp_parent;
dh = (MDB_dhead *)src->mp_parent;
mpp.mp_parent = dh->md_parent;
mpp.mp_pi = dh->md_pi;
- return mdb_rebalance(bt, &mpp);
+ return mdb_rebalance(txn, dbi, &mpp);
}
#define FILL_THRESHOLD 250
static int
-mdb_rebalance(MDB_db *db, MDB_pageparent *mpp)
+mdb_rebalance(MDB_txn *txn, MDB_dbi dbi, MDB_pageparent *mpp)
{
MDB_node *node;
MDB_page *root;
MDB_pageparent npp;
indx_t si = 0, di = 0;
- assert(db != NULL);
- assert(db->md_env->me_txn != NULL);
+ assert(txn != NULL);
assert(mpp != NULL);
- DPRINTF("rebalancing %s page %lu (has %lu keys, %.1f%% full)",
+ DPRINTF("rebalancing %s page %lu (has %u keys, %.1f%% full)",
IS_LEAF(mpp->mp_page) ? "leaf" : "branch",
- mpp->mp_page->mp_pgno, NUMKEYS(mpp->mp_page), (float)PAGEFILL(db->md_env, mpp->mp_page) / 10);
+ mpp->mp_page->mp_pgno, NUMKEYS(mpp->mp_page), (float)PAGEFILL(txn->mt_env, mpp->mp_page) / 10);
- if (PAGEFILL(db->md_env, mpp->mp_page) >= FILL_THRESHOLD) {
+ if (PAGEFILL(txn->mt_env, mpp->mp_page) >= FILL_THRESHOLD) {
DPRINTF("no need to rebalance page %lu, above fill threshold",
mpp->mp_page->mp_pgno);
return MDB_SUCCESS;
if (mpp->mp_parent == NULL) {
if (NUMKEYS(mpp->mp_page) == 0) {
DPRINTF("tree is completely empty");
- db->md_env->me_txn->mt_root = P_INVALID;
- db->md_env->me_meta.mm_stat.ms_depth--;
- db->md_env->me_meta.mm_stat.ms_leaf_pages--;
+ 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!");
- db->md_env->me_txn->mt_root = NODEPGNO(NODEPTR(mpp->mp_page, 0));
- if ((root = mdbenv_get_page(db->md_env, db->md_env->me_txn->mt_root)) == NULL)
+ txn->mt_dbs[dbi].md_root = NODEPGNO(NODEPTR(mpp->mp_page, 0));
+ if ((root = mdbenv_get_page(txn->mt_env, txn->mt_dbs[dbi].md_root)) == NULL)
return MDB_FAIL;
- db->md_env->me_meta.mm_stat.ms_depth--;
- db->md_env->me_meta.mm_stat.ms_branch_pages--;
+ txn->mt_dbs[dbi].md_depth--;
+ txn->mt_dbs[dbi].md_branch_pages--;
} else
DPRINTF("root page doesn't need rebalancing");
return MDB_SUCCESS;
*/
DPRINTF("reading right neighbor");
node = NODEPTR(mpp->mp_parent, mpp->mp_pi + 1);
- if ((npp.mp_page = mdbenv_get_page(db->md_env, NODEPGNO(node))) == NULL)
+ if ((npp.mp_page = mdbenv_get_page(txn->mt_env, NODEPGNO(node))) == NULL)
return MDB_FAIL;
npp.mp_pi = mpp->mp_pi + 1;
si = 0;
*/
DPRINTF("reading left neighbor");
node = NODEPTR(mpp->mp_parent, mpp->mp_pi - 1);
- if ((npp.mp_page = mdbenv_get_page(db->md_env, NODEPGNO(node))) == NULL)
+ if ((npp.mp_page = mdbenv_get_page(txn->mt_env, NODEPGNO(node))) == NULL)
return MDB_FAIL;
npp.mp_pi = mpp->mp_pi - 1;
si = NUMKEYS(npp.mp_page) - 1;
}
npp.mp_parent = mpp->mp_parent;
- DPRINTF("found neighbor page %lu (%lu keys, %.1f%% full)",
- npp.mp_page->mp_pgno, NUMKEYS(npp.mp_page), (float)PAGEFILL(db->md_env, npp.mp_page) / 10);
+ DPRINTF("found neighbor page %lu (%u keys, %.1f%% full)",
+ npp.mp_page->mp_pgno, NUMKEYS(npp.mp_page), (float)PAGEFILL(txn->mt_env, npp.mp_page) / 10);
/* If the neighbor page is above threshold and has at least two
* keys, move one key from it.
*
* Otherwise we should try to merge them.
*/
- if (PAGEFILL(db->md_env, npp.mp_page) >= FILL_THRESHOLD && NUMKEYS(npp.mp_page) >= 2)
- return mdb_move_node(db, &npp, si, mpp, di);
+ if (PAGEFILL(txn->mt_env, npp.mp_page) >= FILL_THRESHOLD && NUMKEYS(npp.mp_page) >= 2)
+ return mdb_move_node(txn, dbi, &npp, si, mpp, di);
else { /* FIXME: if (has_enough_room()) */
if (mpp->mp_pi == 0)
- return mdb_merge(db, &npp, mpp);
+ return mdb_merge(txn, dbi, &npp, mpp);
else
- return mdb_merge(db, mpp, &npp);
+ return mdb_merge(txn, dbi, mpp, &npp);
+ }
+}
+
+static int
+mdb_del0(MDB_txn *txn, MDB_dbi dbi, unsigned int ki, MDB_pageparent *mpp, MDB_node *leaf)
+{
+ int rc;
+
+ /* add overflow pages to free list */
+ if (F_ISSET(leaf->mn_flags, F_BIGDATA)) {
+ int i, ovpages;
+ pgno_t pg;
+
+ memcpy(&pg, NODEDATA(leaf), sizeof(pg));
+ 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);
+ pg++;
+ }
}
+ mdb_del_node(mpp->mp_page, ki);
+ txn->mt_dbs[dbi].md_entries--;
+ rc = mdb_rebalance(txn, dbi, mpp);
+ if (rc != MDB_SUCCESS)
+ txn->mt_flags |= MDB_TXN_ERROR;
+
+ return rc;
}
int
-mdb_del(MDB_db *bt, MDB_txn *txn,
+mdb_del(MDB_txn *txn, MDB_dbi dbi,
MDB_val *key, MDB_val *data)
{
int rc, exact;
assert(key != NULL);
- if (bt == NULL || txn == NULL)
+ if (txn == NULL || dbi >= txn->mt_numdbs)
return EINVAL;
if (F_ISSET(txn->mt_flags, MDB_TXN_RDONLY)) {
return EINVAL;
}
- if (bt->md_env->me_txn != txn) {
- return EINVAL;
- }
-
if (key->mv_size == 0 || key->mv_size > MAXKEYSIZE) {
return EINVAL;
}
- if ((rc = mdb_search_page(bt, txn, key, NULL, 1, &mpp)) != MDB_SUCCESS)
+ mpp.mp_parent = NULL;
+ mpp.mp_pi = 0;
+ if ((rc = mdb_search_page(txn, dbi, key, NULL, 1, &mpp)) != MDB_SUCCESS)
return rc;
- leaf = mdb_search_node(bt, mpp.mp_page, key, &exact, &ki);
+ leaf = mdb_search_node(txn, dbi, mpp.mp_page, key, &exact, &ki);
if (leaf == NULL || !exact) {
return ENOENT;
}
- if (data && (rc = mdb_read_data(bt, NULL, leaf, data)) != MDB_SUCCESS)
+ if (data && (rc = mdb_read_data(txn->mt_env, leaf, data)) != MDB_SUCCESS)
return rc;
- mdb_del_node(bt, mpp.mp_page, ki);
- bt->md_env->me_meta.mm_stat.ms_entries--;
- rc = mdb_rebalance(bt, &mpp);
- if (rc != MDB_SUCCESS)
- txn->mt_flags |= MDB_TXN_ERROR;
-
- return rc;
+ return mdb_del0(txn, dbi, ki, &mpp, leaf);
}
/* Split page <*mpp>, and insert <key,(data|newpgno)> in either left or
* refer to a node in the new right sibling page.
*/
static int
-mdb_split(MDB_db *bt, MDB_page **mpp, unsigned int *newindxp,
+mdb_split(MDB_txn *txn, MDB_dbi dbi, MDB_page **mpp, unsigned int *newindxp,
MDB_val *newkey, MDB_val *newdata, pgno_t newpgno)
{
uint8_t flags;
pgno_t pgno = 0;
unsigned int i, j, split_indx;
MDB_node *node;
- MDB_page *pright, *p;
MDB_val sepkey, rkey, rdata;
MDB_page *copy;
MDB_dpage *mdp, *rdp, *pdp;
MDB_dhead *dh;
- assert(bt != NULL);
- assert(bt->md_env != NULL);
+ assert(txn != NULL);
dh = ((MDB_dhead *)*mpp) - 1;
mdp = (MDB_dpage *)dh;
(int)newkey->mv_size, (char *)newkey->mv_data, *newindxp);
if (mdp->h.md_parent == NULL) {
- if ((pdp = mdbenv_new_page(bt->md_env, P_BRANCH, 1)) == NULL)
+ if ((pdp = mdb_new_page(txn, dbi, P_BRANCH, 1)) == NULL)
return MDB_FAIL;
mdp->h.md_pi = 0;
mdp->h.md_parent = &pdp->p;
- bt->md_env->me_txn->mt_root = pdp->p.mp_pgno;
+ txn->mt_dbs[dbi].md_root = pdp->p.mp_pgno;
DPRINTF("root split! new root = %lu", pdp->p.mp_pgno);
- bt->md_env->me_meta.mm_stat.ms_depth++;
+ txn->mt_dbs[dbi].md_depth++;
/* Add left (implicit) pointer. */
- if (mdb_add_node(bt, &pdp->p, 0, NULL, NULL,
+ if (mdb_add_node(txn, dbi, &pdp->p, 0, NULL, NULL,
mdp->p.mp_pgno, 0) != MDB_SUCCESS)
return MDB_FAIL;
} else {
}
/* Create a right sibling. */
- if ((rdp = mdbenv_new_page(bt->md_env, mdp->p.mp_flags, 1)) == NULL)
+ if ((rdp = mdb_new_page(txn, dbi, mdp->p.mp_flags, 1)) == NULL)
return MDB_FAIL;
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(bt->md_env->me_head.mh_psize)) == NULL)
+ if ((copy = malloc(txn->mt_env->me_psize)) == NULL)
return MDB_FAIL;
- bcopy(&mdp->p, copy, bt->md_env->me_head.mh_psize);
- bzero(&mdp->p.mp_ptrs, bt->md_env->me_head.mh_psize - PAGEHDRSZ);
+ 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;
- mdp->p.mp_upper = bt->md_env->me_head.mh_psize;
+ mdp->p.mp_upper = txn->mt_env->me_psize;
split_indx = NUMKEYS(copy) / 2 + 1;
/* First find the separating key between the split pages.
*/
- bzero(&sepkey, sizeof(sepkey));
+ memset(&sepkey, 0, sizeof(sepkey));
if (newindx == split_indx) {
sepkey.mv_size = newkey->mv_size;
sepkey.mv_data = newkey->mv_data;
/* Copy separator key to the parent.
*/
- if (SIZELEFT(rdp->h.md_parent) < mdb_branch_size(bt, &sepkey)) {
- rc = mdb_split(bt, &rdp->h.md_parent, &rdp->h.md_pi,
+ if (SIZELEFT(rdp->h.md_parent) < mdb_branch_size(txn->mt_env, &sepkey)) {
+ rc = mdb_split(txn, dbi, &rdp->h.md_parent, &rdp->h.md_pi,
&sepkey, NULL, rdp->p.mp_pgno);
/* Right page might now have changed parent.
mdp->h.md_pi = rdp->h.md_pi - 1;
}
} else {
- rc = mdb_add_node(bt, rdp->h.md_parent, rdp->h.md_pi,
+ rc = mdb_add_node(txn, dbi, rdp->h.md_parent, rdp->h.md_pi,
&sepkey, NULL, rdp->p.mp_pgno, 0);
}
if (rc != MDB_SUCCESS) {
rkey.mv_size = 0;
}
- rc = mdb_add_node(bt, &pdp->p, j, &rkey, &rdata, pgno,flags);
+ rc = mdb_add_node(txn, dbi, &pdp->p, j, &rkey, &rdata, pgno,flags);
}
free(copy);
}
int
-mdb_put(MDB_db *bt, MDB_txn *txn,
+mdb_put(MDB_txn *txn, MDB_dbi dbi,
MDB_val *key, MDB_val *data, unsigned int flags)
{
int rc = MDB_SUCCESS, exact;
assert(key != NULL);
assert(data != NULL);
- if (bt == NULL || txn == NULL)
+ if (txn == NULL)
return EINVAL;
if (F_ISSET(txn->mt_flags, MDB_TXN_RDONLY)) {
return EINVAL;
}
- if (bt->md_env->me_txn != txn) {
+ if (txn->mt_env->me_txn != txn) {
return EINVAL;
}
DPRINTF("==> put key %.*s, size %zu, data size %zu",
(int)key->mv_size, (char *)key->mv_data, key->mv_size, data->mv_size);
- rc = mdb_search_page(bt, txn, key, NULL, 1, &mpp);
+ 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(bt, mpp.mp_page, key, &exact, &ki);
+ leaf = mdb_search_node(txn, dbi, mpp.mp_page, key, &exact, &ki);
if (leaf && exact) {
if (F_ISSET(flags, MDB_NOOVERWRITE)) {
DPRINTF("duplicate key %.*s",
(int)key->mv_size, (char *)key->mv_data);
return EEXIST;
}
- mdb_del_node(bt, mpp.mp_page, ki);
+ /* same size, just replace it */
+ if (NODEDSZ(leaf) == data->mv_size) {
+ memcpy(NODEDATA(leaf), data->mv_data, data->mv_size);
+ goto done;
+ }
+ mdb_del_node(mpp.mp_page, ki);
}
if (leaf == NULL) { /* append if not found */
ki = NUMKEYS(mpp.mp_page);
MDB_dpage *dp;
/* new file, just write a root leaf page */
DPRINTF("allocating new root leaf page");
- if ((dp = mdbenv_new_page(bt->md_env, P_LEAF, 1)) == NULL) {
+ if ((dp = mdb_new_page(txn, dbi, P_LEAF, 1)) == NULL) {
return ENOMEM;
}
mpp.mp_page = &dp->p;
- txn->mt_root = mpp.mp_page->mp_pgno;
- bt->md_env->me_meta.mm_stat.ms_depth++;
+ txn->mt_dbs[dbi].md_root = mpp.mp_page->mp_pgno;
+ txn->mt_dbs[dbi].md_depth++;
ki = 0;
}
else
goto done;
assert(IS_LEAF(mpp.mp_page));
- DPRINTF("there are %lu keys, should insert new key at index %i",
+ DPRINTF("there are %u keys, should insert new key at index %i",
NUMKEYS(mpp.mp_page), ki);
- if (SIZELEFT(mpp.mp_page) < mdb_leaf_size(bt, key, data)) {
- rc = mdb_split(bt, &mpp.mp_page, &ki, key, data, P_INVALID);
+ 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);
} else {
/* There is room already in this leaf page. */
- rc = mdb_add_node(bt, mpp.mp_page, ki, key, data, 0, 0);
+ rc = mdb_add_node(txn, dbi, mpp.mp_page, ki, key, data, 0, 0);
}
if (rc != MDB_SUCCESS)
txn->mt_flags |= MDB_TXN_ERROR;
else
- bt->md_env->me_meta.mm_stat.ms_entries++;
+ txn->mt_dbs[dbi].md_entries++;
done:
return rc;
}
-static pgno_t
-mdbenv_compact_tree(MDB_env *env, pgno_t pgno, MDB_env *envc)
-{
- ssize_t rc;
- indx_t i;
- pgno_t *pnext, next;
- MDB_node *node;
- MDB_page *p;
- MDB_page *mp;
-
- /* Get the page and make a copy of it.
- */
- if ((mp = mdbenv_get_page(env, pgno)) == NULL)
- return P_INVALID;
- if ((p = malloc(env->me_head.mh_psize)) == NULL)
- return P_INVALID;
- bcopy(mp, p, env->me_head.mh_psize);
-
- /* Go through all nodes in the (copied) page and update the
- * page pointers.
- */
- if (F_ISSET(p->mp_flags, P_BRANCH)) {
- for (i = 0; i < NUMKEYS(p); i++) {
- node = NODEPTR(p, i);
- node->mn_pgno = mdbenv_compact_tree(env, node->mn_pgno, envc);
- if (node->mn_pgno == P_INVALID) {
- free(p);
- return P_INVALID;
- }
- }
- } else if (F_ISSET(p->mp_flags, P_LEAF)) {
- for (i = 0; i < NUMKEYS(p); i++) {
- node = NODEPTR(p, i);
- if (F_ISSET(node->mn_flags, F_BIGDATA)) {
- bcopy(NODEDATA(node), &next, sizeof(next));
- next = mdbenv_compact_tree(env, next, envc);
- if (next == P_INVALID) {
- free(p);
- return P_INVALID;
- }
- bcopy(&next, NODEDATA(node), sizeof(next));
- }
- }
- } else if (F_ISSET(p->mp_flags, P_OVERFLOW)) {
- ; /* handled below */
- } else
- assert(0);
-
- pgno = p->mp_pgno = envc->me_txn->mt_next_pgno++;
- rc = write(envc->me_fd, p, env->me_head.mh_psize);
- free(p);
- if (rc != (ssize_t)env->me_head.mh_psize)
- return P_INVALID;
-
- if (F_ISSET(mp->mp_flags, P_OVERFLOW)) {
- if (mp->mp_pages > 1) {
- size_t len = (mp->mp_pages-1) * env->me_head.mh_psize;
- envc->me_txn->mt_next_pgno += mp->mp_pages-1;
- rc = write(envc->me_fd, mp+1, len);
- if (rc != len)
- return P_INVALID;
- }
- }
- return pgno;
-}
-
-int
-mdbenv_compact(MDB_env *env)
-{
- char *compact_path = NULL;
- MDB_env *envc;
- MDB_txn *txn, *txnc = NULL;
- int fd, rc;
- pgno_t root;
-
- assert(env != NULL);
-
- if (env->me_path == NULL) {
- return EINVAL;
- }
-
- DPRINTF("compacting mdbenv %p with path %s", env, env->me_path);
-
- rc = mdb_txn_begin(env, 0, &txn);
- if (rc) return rc;
-
- asprintf(&compact_path, "%s.compact.XXXXXX", env->me_path);
- fd = mkstemp(compact_path);
- if (fd == -1) {
- rc = errno;
- free(compact_path);
- mdb_txn_abort(txn);
- return rc;
- }
-
- rc = mdbenv_create(&envc, env->me_mapsize);
- if (rc) goto failed;
-
- envc->me_fd = fd;
- rc = mdbenv_open2(envc, env->me_flags);
- if (rc) goto failed;
-
- bcopy(&env->me_meta, &envc->me_meta, sizeof(env->me_meta));
- envc->me_meta.mm_stat.ms_revisions = 0;
-
- rc = mdb_txn_begin(envc, 0, &txnc);
- if (rc) goto failed;
-
- if (env->me_meta.mm_root != P_INVALID) {
- root = mdbenv_compact_tree(env, env->me_meta.mm_root, envc);
- if (root == P_INVALID)
- goto failed;
- if ((rc = mdbenv_write_meta(envc, root, 0)) != MDB_SUCCESS)
- goto failed;
- }
-
- fsync(fd);
-
- DPRINTF("renaming %s to %s", compact_path, env->me_path);
- if (rename(compact_path, env->me_path) != 0) {
- rc = errno;
- goto failed;
- }
-
- /* Write a "tombstone" meta page so other processes can pick up
- * the change and re-open the file.
- */
- if (mdbenv_write_meta(env, P_INVALID, MDB_TOMBSTONE) != MDB_SUCCESS)
- goto failed;
-
- mdb_txn_abort(txn);
- mdb_txn_abort(txnc);
- free(compact_path);
- mdbenv_close(envc);
- return 0;
-
-failed:
- mdb_txn_abort(txn);
- mdb_txn_abort(txnc);
- unlink(compact_path);
- free(compact_path);
- mdbenv_close(envc);
- return rc;
-}
-
int
mdbenv_get_flags(MDB_env *env, unsigned int *arg)
{
if (!env || !arg)
return EINVAL;
- *arg = (env->me_flags & ~MDB_FIXPADDING);
+ *arg = env->me_flags;
return MDB_SUCCESS;
}
}
int
-mdbenv_stat(MDB_env *env, MDB_stat **arg)
+mdbenv_stat(MDB_env *env, MDB_stat *arg)
{
if (env == NULL || arg == NULL)
return EINVAL;
- *arg = &env->me_meta.mm_stat;
+ 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;
}
-int mdb_open(MDB_env *env, MDB_txn *txn, const char *name, unsigned int flags, MDB_db **db)
+int mdb_open(MDB_txn *txn, const char *name, unsigned int flags, MDB_dbi *dbi)
{
+ MDB_val key, data;
+ MDB_dbi i;
+ int rc;
+ size_t len;
+
+ /* main DB? */
if (!name) {
- *db = (MDB_db *)&env->me_db;
+ *dbi = MAIN_DBI;
return MDB_SUCCESS;
}
- return EINVAL;
+
+ /* Is the DB already open? */
+ len = strlen(name);
+ for (i=2; i<txn->mt_numdbs; i++) {
+ if (len == txn->mt_dbxs[i].md_name.mv_size &&
+ !strncmp(name, txn->mt_dbxs[i].md_name.mv_data, len)) {
+ *dbi = i;
+ return MDB_SUCCESS;
+ }
+ }
+
+ if (txn->mt_numdbs >= txn->mt_env->me_maxdbs - 1)
+ return ENFILE;
+
+ /* Find the DB info */
+ key.mv_size = len;
+ key.mv_data = (void *)name;
+ rc = mdb_get(txn, MAIN_DBI, &key, &data);
+
+ /* Create if requested */
+ if (rc == ENOENT && (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);
+ }
+
+ /* OK, got info, add to table */
+ if (rc == MDB_SUCCESS) {
+ txn->mt_dbxs[txn->mt_numdbs].md_name.mv_data = strdup(name);
+ txn->mt_dbxs[txn->mt_numdbs].md_name.mv_size = len;
+ txn->mt_dbxs[txn->mt_numdbs].md_cmp = NULL;
+ txn->mt_dbxs[txn->mt_numdbs].md_dcmp = NULL;
+ txn->mt_dbxs[txn->mt_numdbs].md_rel = NULL;
+ memcpy(&txn->mt_dbs[txn->mt_numdbs], data.mv_data, sizeof(MDB_db));
+ *dbi = txn->mt_numdbs;
+ txn->mt_numdbs++;
+ }
+
+ return rc;
}
-int mdb_stat(MDB_db *db, MDB_stat **arg)
+int mdb_stat(MDB_txn *txn, MDB_dbi dbi, MDB_stat *arg)
{
- if (db == NULL || arg == NULL)
+ if (txn == NULL || arg == NULL)
return EINVAL;
- *arg = &db->md_stat;
+ 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;
}
-void mdb_close(MDB_db *db)
+void mdb_close(MDB_txn *txn, MDB_dbi dbi)
{
+ char *ptr;
+ if (dbi <= MAIN_DBI || dbi >= txn->mt_numdbs)
+ return;
+ ptr = txn->mt_dbxs[dbi].md_name.mv_data;
+ txn->mt_dbxs[dbi].md_name.mv_data = NULL;
+ txn->mt_dbxs[dbi].md_name.mv_size = 0;
+ free(ptr);
}
projects / openldap / blobdiff