/** @file mdb.c
- * @brief memory-mapped database library
+ * @brief Lightning memory-mapped database library
*
* A Btree-based database management library modeled loosely on the
* BerkeleyDB API, but much simplified.
*/
/*
- * Copyright 2011-2013 Howard Chu, Symas Corp.
+ * Copyright 2011-2014 Howard Chu, Symas Corp.
* All rights reserved.
*
* Redistribution and use in source and binary forms, with or without
#ifndef _GNU_SOURCE
#define _GNU_SOURCE 1
#endif
-#include <sys/types.h>
-#include <sys/stat.h>
#ifdef _WIN32
+#include <malloc.h>
#include <windows.h>
/** getpid() returns int; MinGW defines pid_t but MinGW64 typedefs it
* as int64 which is wrong. MSVC doesn't define it at all, so just
* don't use it.
*/
#define MDB_PID_T int
+#define MDB_THR_T DWORD
+#include <sys/types.h>
+#include <sys/stat.h>
#ifdef __GNUC__
# include <sys/param.h>
#else
# endif
#endif
#else
+#include <sys/types.h>
+#include <sys/stat.h>
#define MDB_PID_T pid_t
+#define MDB_THR_T pthread_t
#include <sys/param.h>
#include <sys/uio.h>
#include <sys/mman.h>
# error "Two's complement, reasonably sized integer types, please"
#endif
-/** @defgroup internal MDB Internals
+/** @defgroup internal LMDB Internals
* @{
*/
/** @defgroup compat Compatibility Macros
#ifdef _WIN32
#define MDB_USE_HASH 1
#define MDB_PIDLOCK 0
-#define pthread_t DWORD
+#define THREAD_RET DWORD
+#define pthread_t HANDLE
#define pthread_mutex_t HANDLE
+#define pthread_cond_t HANDLE
#define pthread_key_t DWORD
#define pthread_self() GetCurrentThreadId()
#define pthread_key_create(x,y) \
#define pthread_setspecific(x,y) (TlsSetValue(x,y) ? 0 : ErrCode())
#define pthread_mutex_unlock(x) ReleaseMutex(x)
#define pthread_mutex_lock(x) WaitForSingleObject(x, INFINITE)
+#define pthread_cond_signal(x) SetEvent(*x)
+#define pthread_cond_wait(cond,mutex) SignalObjectAndWait(*mutex, *cond, INFINITE, FALSE); WaitForSingleObject(*mutex, INFINITE)
+#define THREAD_CREATE(thr,start,arg) thr=CreateThread(NULL,0,start,arg,0,NULL)
+#define THREAD_FINISH(thr) WaitForSingleObject(thr, INFINITE)
#define LOCK_MUTEX_R(env) pthread_mutex_lock((env)->me_rmutex)
#define UNLOCK_MUTEX_R(env) pthread_mutex_unlock((env)->me_rmutex)
#define LOCK_MUTEX_W(env) pthread_mutex_lock((env)->me_wmutex)
#endif
#define Z "I"
#else
-
+#define THREAD_RET void *
+#define THREAD_CREATE(thr,start,arg) pthread_create(&thr,NULL,start,arg)
+#define THREAD_FINISH(thr) pthread_join(thr,NULL)
#define Z "z" /**< printf format modifier for size_t */
/** For MDB_LOCK_FORMAT: True if readers take a pid lock in the lockfile */
*/
#define MDB_MINKEYS 2
- /** A stamp that identifies a file as an MDB file.
+ /** A stamp that identifies a file as an LMDB file.
* There's nothing special about this value other than that it is easily
* recognizable, and it will reflect any byte order mismatches.
*/
/** The process ID of the process owning this reader txn. */
MDB_PID_T mrb_pid;
/** The thread ID of the thread owning this txn. */
- pthread_t mrb_tid;
+ MDB_THR_T mrb_tid;
} MDB_rxbody;
/** The actual reader record, with cacheline padding. */
* unlikely. If a collision occurs, the results are unpredictable.
*/
typedef struct MDB_txbody {
- /** Stamp identifying this as an MDB file. It must be set
+ /** Stamp identifying this as an LMDB file. It must be set
* to #MDB_MAGIC. */
uint32_t mtb_magic;
/** Format of this lock file. Must be set to #MDB_LOCK_FORMAT. */
#define P_DIRTY 0x10 /**< dirty page, also set for #P_SUBP pages */
#define P_LEAF2 0x20 /**< for #MDB_DUPFIXED records */
#define P_SUBP 0x40 /**< for #MDB_DUPSORT sub-pages */
+#define P_LOOSE 0x4000 /**< page was dirtied then freed, can be reused */
#define P_KEEP 0x8000 /**< leave this page alone during spill */
/** @} */
uint16_t mp_flags; /**< @ref mdb_page */
/** The number of overflow pages needed to store the given size. */
#define OVPAGES(size, psize) ((PAGEHDRSZ-1 + (size)) / (psize) + 1)
+ /** Link in #MDB_txn.%mt_loose_pages list */
+#define NEXT_LOOSE_PAGE(p) (*(MDB_page **)METADATA(p))
+
/** Header for a single key/data pair within a page.
* Used in pages of type #P_BRANCH and #P_LEAF without #P_LEAF2.
* We guarantee 2-byte alignment for 'MDB_node's.
* Pages 0-1 are meta pages. Transaction N writes meta page #(N % 2).
*/
typedef struct MDB_meta {
- /** Stamp identifying this as an MDB file. It must be set
+ /** Stamp identifying this as an LMDB file. It must be set
* to #MDB_MAGIC. */
uint32_t mm_magic;
/** Version number of this lock file. Must be set to #MDB_DATA_VERSION. */
/** The list of pages that became unused during this transaction.
*/
MDB_IDL mt_free_pgs;
+ /** The list of loose pages that became unused and may be reused
+ * in this transaction, linked through #NEXT_LOOSE_PAGE(page).
+ */
+ MDB_page *mt_loose_pgs;
/** The sorted list of dirty pages we temporarily wrote to disk
* because the dirty list was full. page numbers in here are
* shifted left by 1, deleted slots have the LSB set.
* @{
*/
#define MDB_TXN_RDONLY 0x01 /**< read-only transaction */
-#define MDB_TXN_ERROR 0x02 /**< an error has occurred */
+#define MDB_TXN_ERROR 0x02 /**< txn is unusable after an error */
#define MDB_TXN_DIRTY 0x04 /**< must write, even if dirty list is empty */
#define MDB_TXN_SPILLS 0x08 /**< txn or a parent has spilled pages */
/** @} */
return MDB_VERSION_STRING;
}
-/** Table of descriptions for MDB @ref errors */
+/** Table of descriptions for LMDB @ref errors */
static char *const mdb_errstr[] = {
"MDB_KEYEXIST: Key/data pair already exists",
"MDB_NOTFOUND: No matching key/data pair found",
"MDB_CORRUPTED: Located page was wrong type",
"MDB_PANIC: Update of meta page failed",
"MDB_VERSION_MISMATCH: Database environment version mismatch",
- "MDB_INVALID: File is not an MDB file",
+ "MDB_INVALID: File is not an LMDB file",
"MDB_MAP_FULL: Environment mapsize limit reached",
"MDB_DBS_FULL: Environment maxdbs limit reached",
"MDB_READERS_FULL: Environment maxreaders limit reached",
"MDB_INCOMPATIBLE: Operation and DB incompatible, or DB flags changed",
"MDB_BAD_RSLOT: Invalid reuse of reader locktable slot",
"MDB_BAD_TXN: Transaction cannot recover - it must be aborted",
- "MDB_BAD_VALSIZE: Too big key/data, key is empty, or wrong DUPFIXED size",
+ "MDB_BAD_VALSIZE: Unsupported size of key/DB name/data, or wrong DUPFIXED size",
};
char *
mdb_cursor_init(&mc, txn, FREE_DBI, NULL);
while ((rc = mdb_cursor_get(&mc, &key, &data, MDB_NEXT)) == 0)
freecount += *(MDB_ID *)data.mv_data;
+ mdb_tassert(txn, rc == MDB_NOTFOUND);
count = 0;
for (i = 0; i<txn->mt_numdbs; i++) {
txn->mt_dbs[i].md_leaf_pages +
txn->mt_dbs[i].md_overflow_pages;
if (txn->mt_dbs[i].md_flags & MDB_DUPSORT) {
- mdb_page_search(&mc, NULL, MDB_PS_FIRST);
- do {
+ rc = mdb_page_search(&mc, NULL, MDB_PS_FIRST);
+ for (; rc == MDB_SUCCESS; rc = mdb_cursor_sibling(&mc, 1)) {
unsigned j;
MDB_page *mp;
mp = mc.mc_pg[mc.mc_top];
}
}
}
- while (mdb_cursor_sibling(&mc, 1) == 0);
+ mdb_tassert(txn, rc == MDB_NOTFOUND);
}
}
if (freecount + count + 2 /* metapages */ != txn->mt_next_pgno) {
}
return ret;
}
-
/** Free a single page.
* Saves single pages to a list, for future reuse.
* (This is not used for multi-page overflow pages.)
dl[0].mid = 0;
}
+/** Loosen or free a single page.
+ * Saves single pages to a list for future reuse
+ * in this same txn. It has been pulled from the freeDB
+ * and already resides on the dirty list, but has been
+ * deleted. Use these pages first before pulling again
+ * from the freeDB.
+ *
+ * If the page wasn't dirtied in this txn, just add it
+ * to this txn's free list.
+ */
+static int
+mdb_page_loose(MDB_cursor *mc, MDB_page *mp)
+{
+ int loose = 0;
+ pgno_t pgno = mp->mp_pgno;
+
+ if ((mp->mp_flags & P_DIRTY) && mc->mc_dbi != FREE_DBI) {
+ if (mc->mc_txn->mt_parent) {
+ MDB_ID2 *dl = mc->mc_txn->mt_u.dirty_list;
+ /* If txn has a parent, make sure the page is in our
+ * dirty list.
+ */
+ if (dl[0].mid) {
+ unsigned x = mdb_mid2l_search(dl, pgno);
+ if (x <= dl[0].mid && dl[x].mid == pgno) {
+ if (mp != dl[x].mptr) { /* bad cursor? */
+ mc->mc_flags &= ~(C_INITIALIZED|C_EOF);
+ mc->mc_txn->mt_flags |= MDB_TXN_ERROR;
+ return MDB_CORRUPTED;
+ }
+ /* ok, it's ours */
+ loose = 1;
+ }
+ }
+ } else {
+ /* no parent txn, so it's just ours */
+ loose = 1;
+ }
+ }
+ if (loose) {
+ NEXT_LOOSE_PAGE(mp) = mc->mc_txn->mt_loose_pgs;
+ mc->mc_txn->mt_loose_pgs = mp;
+ mp->mp_flags |= P_LOOSE;
+ } else {
+ int rc = mdb_midl_append(&mc->mc_txn->mt_free_pgs, pgno);
+ if (rc)
+ return rc;
+ }
+
+ return MDB_SUCCESS;
+}
+
/** Set or clear P_KEEP in dirty, non-overflow, non-sub pages watched by txn.
* @param[in] mc A cursor handle for the current operation.
* @param[in] pflags Flags of the pages to update:
break;
}
+ /* Loose pages shouldn't be spilled */
+ for (dp = txn->mt_loose_pgs; dp; dp = NEXT_LOOSE_PAGE(dp)) {
+ if ((dp->mp_flags & Mask) == pflags)
+ dp->mp_flags ^= P_KEEP;
+ }
+
if (all) {
/* Mark dirty root pages */
for (i=0; i<txn->mt_numdbs; i++) {
#else
enum { Paranoid = 0, Max_retries = INT_MAX /*infinite*/ };
#endif
- int rc, retry = Max_retries;
+ int rc, retry = num * 20;
MDB_txn *txn = mc->mc_txn;
MDB_env *env = txn->mt_env;
pgno_t pgno, *mop = env->me_pghead;
MDB_cursor_op op;
MDB_cursor m2;
+ /* If there are any loose pages, just use them */
+ if (num == 1 && txn->mt_loose_pgs) {
+ np = txn->mt_loose_pgs;
+ txn->mt_loose_pgs = NEXT_LOOSE_PAGE(np);
+ *mp = np;
+ return MDB_SUCCESS;
+ }
+
*mp = NULL;
/* If our dirty list is already full, we can't do anything */
if (mop[i-n2] == pgno+n2)
goto search_done;
} while (--i > n2);
- if (Max_retries < INT_MAX && --retry < 0)
+ if (--retry < 0)
break;
}
return MDB_BAD_RSLOT;
} else {
MDB_PID_T pid = env->me_pid;
- pthread_t tid = pthread_self();
+ MDB_THR_T tid = pthread_self();
if (!env->me_live_reader) {
rc = mdb_reader_pid(env, Pidset, pid);
return rc;
}
+ /* Dispose of loose pages. Usually they will have all
+ * been used up by the time we get here.
+ */
+ if (txn->mt_loose_pgs) {
+ MDB_page *mp = txn->mt_loose_pgs;
+ /* Just return them to freeDB */
+ if (env->me_pghead) {
+ int i, j;
+ mop = env->me_pghead;
+ for (; mp; mp = NEXT_LOOSE_PAGE(mp)) {
+ pgno_t pg = mp->mp_pgno;
+ j = mop[0] + 1;
+ for (i = mop[0]; i && mop[i] < pg; i--)
+ mop[j--] = mop[i];
+ mop[j] = pg;
+ mop[0] += 1;
+ }
+ } else {
+ /* Oh well, they were wasted. Put on freelist */
+ for (; mp; mp = NEXT_LOOSE_PAGE(mp)) {
+ mdb_midl_append(&txn->mt_free_pgs, mp->mp_pgno);
+ }
+ }
+ txn->mt_loose_pgs = NULL;
+ }
+
/* MDB_RESERVE cancels meminit in ovpage malloc (when no WRITEMAP) */
clean_limit = (env->me_flags & (MDB_NOMEMINIT|MDB_WRITEMAP))
? SSIZE_MAX : maxfree_1pg;
mop += mop_len;
rc = mdb_cursor_first(&mc, &key, &data);
for (; !rc; rc = mdb_cursor_next(&mc, &key, &data, MDB_NEXT)) {
- unsigned flags = MDB_CURRENT;
txnid_t id = *(txnid_t *)key.mv_data;
ssize_t len = (ssize_t)(data.mv_size / sizeof(MDB_ID)) - 1;
MDB_ID save;
mdb_tassert(txn, len >= 0 && id <= env->me_pglast);
+ key.mv_data = &id;
if (len > mop_len) {
len = mop_len;
data.mv_size = (len + 1) * sizeof(MDB_ID);
- /* Drop MDB_CURRENT when changing the data size */
- key.mv_data = &id;
- flags = 0;
}
data.mv_data = mop -= len;
save = mop[0];
mop[0] = len;
- rc = mdb_cursor_put(&mc, &key, &data, flags);
+ rc = mdb_cursor_put(&mc, &key, &data, MDB_CURRENT);
mop[0] = save;
if (rc || !(mop_len -= len))
break;
return 0;
}
+static void
+mdb_env_init_meta0(MDB_env *env, MDB_meta *meta)
+{
+ meta->mm_magic = MDB_MAGIC;
+ meta->mm_version = MDB_DATA_VERSION;
+ meta->mm_mapsize = env->me_mapsize;
+ meta->mm_psize = env->me_psize;
+ meta->mm_last_pg = 1;
+ meta->mm_flags = env->me_flags & 0xffff;
+ meta->mm_flags |= MDB_INTEGERKEY;
+ meta->mm_dbs[0].md_root = P_INVALID;
+ meta->mm_dbs[1].md_root = P_INVALID;
+}
+
/** Write the environment parameters of a freshly created DB environment.
* @param[in] env the environment handle
* @param[out] meta address of where to store the meta information
psize = env->me_psize;
- meta->mm_magic = MDB_MAGIC;
- meta->mm_version = MDB_DATA_VERSION;
- meta->mm_mapsize = env->me_mapsize;
- meta->mm_psize = psize;
- meta->mm_last_pg = 1;
- meta->mm_flags = env->me_flags & 0xffff;
- meta->mm_flags |= MDB_INTEGERKEY;
- meta->mm_dbs[0].md_root = P_INVALID;
- meta->mm_dbs[1].md_root = P_INVALID;
+ mdb_env_init_meta0(env, meta);
p = calloc(2, psize);
p->mp_pgno = 0;
int rc;
HANDLE mh;
LONG sizelo, sizehi;
- sizelo = env->me_mapsize & 0xffffffff;
- sizehi = env->me_mapsize >> 16 >> 16; /* only needed on Win64 */
+ size_t msize;
+
+ if (flags & MDB_RDONLY) {
+ msize = 0;
+ sizelo = 0;
+ sizehi = 0;
+ } else {
+ msize = env->me_mapsize;
+ sizelo = msize & 0xffffffff;
+ sizehi = msize >> 16 >> 16; /* only needed on Win64 */
+ }
/* Windows won't create mappings for zero length files.
* Just allocate the maxsize right now.
return ErrCode();
env->me_map = MapViewOfFileEx(mh, flags & MDB_WRITEMAP ?
FILE_MAP_WRITE : FILE_MAP_READ,
- 0, 0, env->me_mapsize, addr);
+ 0, 0, msize, addr);
rc = env->me_map ? 0 : ErrCode();
CloseHandle(mh);
if (rc)
return MDB_SUCCESS;
}
-/** Further setup required for opening an MDB environment
+/** Further setup required for opening an LMDB environment
*/
static int
mdb_env_open2(MDB_env *env)
case DLL_THREAD_DETACH:
for (i=0; i<mdb_tls_nkeys; i++) {
MDB_reader *r = pthread_getspecific(mdb_tls_keys[i]);
- mdb_env_reader_dest(r);
+ if (r) {
+ mdb_env_reader_dest(r);
+ }
}
break;
case DLL_PROCESS_DETACH: break;
#endif
/** Open and/or initialize the lock region for the environment.
- * @param[in] env The MDB environment.
+ * @param[in] env The LMDB environment.
* @param[in] lpath The pathname of the file used for the lock region.
* @param[in] mode The Unix permissions for the file, if we create it.
* @param[out] excl Resulting file lock type: -1 none, 0 shared, 1 exclusive
env->me_flags &= ~(MDB_ENV_ACTIVE|MDB_ENV_TXKEY);
}
-int
-mdb_env_copyfd(MDB_env *env, HANDLE fd)
-{
- MDB_txn *txn = NULL;
- int rc;
- size_t wsize;
- char *ptr;
-#ifdef _WIN32
- DWORD len, w2;
-#define DO_WRITE(rc, fd, ptr, w2, len) rc = WriteFile(fd, ptr, w2, &len, NULL)
-#else
- ssize_t len;
- size_t w2;
-#define DO_WRITE(rc, fd, ptr, w2, len) len = write(fd, ptr, w2); rc = (len >= 0)
-#endif
-
- /* Do the lock/unlock of the reader mutex before starting the
- * write txn. Otherwise other read txns could block writers.
- */
- rc = mdb_txn_begin(env, NULL, MDB_RDONLY, &txn);
- if (rc)
- return rc;
-
- if (env->me_txns) {
- /* We must start the actual read txn after blocking writers */
- mdb_txn_reset0(txn, "reset-stage1");
-
- /* Temporarily block writers until we snapshot the meta pages */
- LOCK_MUTEX_W(env);
-
- rc = mdb_txn_renew0(txn);
- if (rc) {
- UNLOCK_MUTEX_W(env);
- goto leave;
- }
- }
-
- wsize = env->me_psize * 2;
- ptr = env->me_map;
- w2 = wsize;
- while (w2 > 0) {
- DO_WRITE(rc, fd, ptr, w2, len);
- if (!rc) {
- rc = ErrCode();
- break;
- } else if (len > 0) {
- rc = MDB_SUCCESS;
- ptr += len;
- w2 -= len;
- continue;
- } else {
- /* Non-blocking or async handles are not supported */
- rc = EIO;
- break;
- }
- }
- if (env->me_txns)
- UNLOCK_MUTEX_W(env);
-
- if (rc)
- goto leave;
-
- wsize = txn->mt_next_pgno * env->me_psize - wsize;
- while (wsize > 0) {
- if (wsize > MAX_WRITE)
- w2 = MAX_WRITE;
- else
- w2 = wsize;
- DO_WRITE(rc, fd, ptr, w2, len);
- if (!rc) {
- rc = ErrCode();
- break;
- } else if (len > 0) {
- rc = MDB_SUCCESS;
- ptr += len;
- wsize -= len;
- continue;
- } else {
- rc = EIO;
- break;
- }
- }
-
-leave:
- mdb_txn_abort(txn);
- return rc;
-}
-
-int
-mdb_env_copy(MDB_env *env, const char *path)
-{
- int rc, len;
- char *lpath;
- HANDLE newfd = INVALID_HANDLE_VALUE;
-
- if (env->me_flags & MDB_NOSUBDIR) {
- lpath = (char *)path;
- } else {
- len = strlen(path);
- len += sizeof(DATANAME);
- lpath = malloc(len);
- if (!lpath)
- return ENOMEM;
- sprintf(lpath, "%s" DATANAME, path);
- }
-
- /* The destination path must exist, but the destination file must not.
- * We don't want the OS to cache the writes, since the source data is
- * already in the OS cache.
- */
-#ifdef _WIN32
- newfd = CreateFile(lpath, GENERIC_WRITE, 0, NULL, CREATE_NEW,
- FILE_FLAG_NO_BUFFERING|FILE_FLAG_WRITE_THROUGH, NULL);
-#else
- newfd = open(lpath, O_WRONLY|O_CREAT|O_EXCL, 0666);
-#endif
- if (newfd == INVALID_HANDLE_VALUE) {
- rc = ErrCode();
- goto leave;
- }
-
-#ifdef O_DIRECT
- /* Set O_DIRECT if the file system supports it */
- if ((rc = fcntl(newfd, F_GETFL)) != -1)
- (void) fcntl(newfd, F_SETFL, rc | O_DIRECT);
-#endif
-#ifdef F_NOCACHE /* __APPLE__ */
- rc = fcntl(newfd, F_NOCACHE, 1);
- if (rc) {
- rc = ErrCode();
- goto leave;
- }
-#endif
-
- rc = mdb_env_copyfd(env, newfd);
-
-leave:
- if (!(env->me_flags & MDB_NOSUBDIR))
- free(lpath);
- if (newfd != INVALID_HANDLE_VALUE)
- if (close(newfd) < 0 && rc == MDB_SUCCESS)
- rc = ErrCode();
-
- return rc;
-}
void
mdb_env_close(MDB_env *env)
} while(!x && u > (unsigned short *)a->mv_data);
return x;
#else
- return memcmp(a->mv_data, b->mv_data, a->mv_size);
+ unsigned short *u, *c, *end;
+ int x;
+
+ end = (unsigned short *) ((char *) a->mv_data + a->mv_size);
+ u = (unsigned short *)a->mv_data;
+ c = (unsigned short *)b->mv_data;
+ do {
+ x = *u++ - *c++;
+ } while(!x && u < end);
+ return x;
#endif
}
+/** Compare two items pointing at size_t's of unknown alignment. */
+#ifdef MISALIGNED_OK
+# define mdb_cmp_clong mdb_cmp_long
+#else
+# define mdb_cmp_clong mdb_cmp_cint
+#endif
+
/** Compare two items lexically */
static int
mdb_cmp_memn(const MDB_val *a, const MDB_val *b)
if (!mc->mc_top) {
/* There are no other pages */
mc->mc_ki[mc->mc_top] = 0;
- if (op == MDB_SET_RANGE) {
+ if (op == MDB_SET_RANGE && !exactp) {
rc = 0;
goto set1;
} else
mc->mc_flags &= ~C_EOF;
if (IS_LEAF2(mp)) {
- key->mv_size = mc->mc_db->md_pad;
- key->mv_data = LEAF2KEY(mp, mc->mc_ki[mc->mc_top], key->mv_size);
+ if (op == MDB_SET_RANGE || op == MDB_SET_KEY) {
+ key->mv_size = mc->mc_db->md_pad;
+ key->mv_data = LEAF2KEY(mp, mc->mc_ki[mc->mc_top], key->mv_size);
+ }
return MDB_SUCCESS;
}
uint16_t fp_flags;
MDB_val xdata, *rdata, dkey, olddata;
MDB_db dummy;
- int do_sub = 0, insert;
+ int do_sub = 0, insert_key, insert_data;
unsigned int mcount = 0, dcount = 0, nospill;
size_t nsize;
int rc, rc2;
unsigned int nflags;
DKBUF;
- if (mc == NULL)
+ if (mc == NULL || key == NULL)
return EINVAL;
env = mc->mc_txn->mt_env;
if (mc->mc_txn->mt_flags & (MDB_TXN_RDONLY|MDB_TXN_ERROR))
return (mc->mc_txn->mt_flags & MDB_TXN_RDONLY) ? EACCES : MDB_BAD_TXN;
- if (flags != MDB_CURRENT) {
- if (key == NULL)
- return EINVAL;
- if (key->mv_size-1 >= ENV_MAXKEY(env))
- return MDB_BAD_VALSIZE;
- } else {
- /* Ignore key except in sub-cursor, where key holds the data */
- if (!(mc->mc_flags & C_SUB))
- key = NULL;
- }
+ if (key->mv_size-1 >= ENV_MAXKEY(env))
+ return MDB_BAD_VALSIZE;
#if SIZE_MAX > MAXDATASIZE
if (data->mv_size > ((mc->mc_db->md_flags & MDB_DUPSORT) ? ENV_MAXKEY(env) : MAXDATASIZE))
return rc2;
}
- insert = rc;
- if (insert) {
+ insert_key = insert_data = rc;
+ if (insert_key) {
/* The key does not exist */
DPRINTF(("inserting key at index %i", mc->mc_ki[mc->mc_top]));
if ((mc->mc_db->md_flags & MDB_DUPSORT) &&
} else {
/* there's only a key anyway, so this is a no-op */
if (IS_LEAF2(mc->mc_pg[mc->mc_top])) {
+ char *ptr;
unsigned int ksize = mc->mc_db->md_pad;
if (key->mv_size != ksize)
return MDB_BAD_VALSIZE;
- if (flags == MDB_CURRENT) {
- char *ptr = LEAF2KEY(mc->mc_pg[mc->mc_top], mc->mc_ki[mc->mc_top], ksize);
- memcpy(ptr, key->mv_data, ksize);
- }
+ ptr = LEAF2KEY(mc->mc_pg[mc->mc_top], mc->mc_ki[mc->mc_top], ksize);
+ memcpy(ptr, key->mv_data, ksize);
return MDB_SUCCESS;
}
#if UINT_MAX < SIZE_MAX
if (mc->mc_dbx->md_dcmp == mdb_cmp_int && olddata.mv_size == sizeof(size_t))
-#ifdef MISALIGNED_OK
- mc->mc_dbx->md_dcmp = mdb_cmp_long;
-#else
- mc->mc_dbx->md_dcmp = mdb_cmp_cint;
+ mc->mc_dbx->md_dcmp = mdb_cmp_clong;
#endif
-#endif
- /* if data matches, skip it */
+ /* does data match? */
if (!mc->mc_dbx->md_dcmp(data, &olddata)) {
if (flags & MDB_NODUPDATA)
- rc = MDB_KEYEXIST;
- else if (flags & MDB_MULTIPLE)
- goto next_mult;
- else
- rc = MDB_SUCCESS;
- return rc;
+ return MDB_KEYEXIST;
+ /* overwrite it */
+ goto current;
}
/* Back up original data item */
rdata = &xdata;
flags |= F_DUPDATA;
do_sub = 1;
- if (!insert)
+ if (!insert_key)
mdb_node_del(mc, 0);
goto new_sub;
}
return ENOMEM;
id2.mid = pg;
id2.mptr = np;
- rc = mdb_mid2l_insert(mc->mc_txn->mt_u.dirty_list, &id2);
- mdb_cassert(mc, rc == 0);
+ rc2 = mdb_mid2l_insert(mc->mc_txn->mt_u.dirty_list, &id2);
+ mdb_cassert(mc, rc2 == 0);
if (!(flags & MDB_RESERVE)) {
/* Copy end of page, adjusting alignment so
* compiler may copy words instead of bytes.
data->mv_data = METADATA(omp);
else
memcpy(METADATA(omp), data->mv_data, data->mv_size);
- goto done;
+ return MDB_SUCCESS;
}
}
if ((rc2 = mdb_ovpage_free(mc, omp)) != MDB_SUCCESS)
memcpy(olddata.mv_data, data->mv_data, data->mv_size);
else
memcpy(NODEKEY(leaf), key->mv_data, key->mv_size);
- goto done;
+ return MDB_SUCCESS;
}
mdb_node_del(mc, 0);
- mc->mc_db->md_entries--;
}
rdata = data;
nsize = IS_LEAF2(mc->mc_pg[mc->mc_top]) ? key->mv_size : mdb_leaf_size(env, key, rdata);
if (SIZELEFT(mc->mc_pg[mc->mc_top]) < nsize) {
if (( flags & (F_DUPDATA|F_SUBDATA)) == F_DUPDATA )
- nflags &= ~MDB_APPEND;
- if (!insert)
+ nflags &= ~MDB_APPEND; /* sub-page may need room to grow */
+ if (!insert_key)
nflags |= MDB_SPLIT_REPLACE;
rc = mdb_page_split(mc, key, rdata, P_INVALID, nflags);
} else {
/* There is room already in this leaf page. */
rc = mdb_node_add(mc, mc->mc_ki[mc->mc_top], key, rdata, 0, nflags);
- if (rc == 0 && !do_sub && insert) {
+ if (rc == 0 && insert_key) {
/* Adjust other cursors pointing to mp */
MDB_cursor *m2, *m3;
MDB_dbi dbi = mc->mc_dbi;
}
}
- if (rc != MDB_SUCCESS)
- mc->mc_txn->mt_flags |= MDB_TXN_ERROR;
- else {
+ if (rc == MDB_SUCCESS) {
/* 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
*/
if (do_sub) {
int xflags;
+ size_t ecount;
put_sub:
xdata.mv_size = 0;
xdata.mv_data = "";
if (dkey.mv_size) {
rc = mdb_cursor_put(&mc->mc_xcursor->mx_cursor, &dkey, &xdata, xflags);
if (rc)
- return rc;
+ goto bad_sub;
{
/* Adjust other cursors pointing to mp */
MDB_cursor *m2;
/* we've done our job */
dkey.mv_size = 0;
}
+ ecount = mc->mc_xcursor->mx_db.md_entries;
if (flags & MDB_APPENDDUP)
xflags |= MDB_APPEND;
rc = mdb_cursor_put(&mc->mc_xcursor->mx_cursor, data, &xdata, xflags);
void *db = NODEDATA(leaf);
memcpy(db, &mc->mc_xcursor->mx_db, sizeof(MDB_db));
}
+ insert_data = mc->mc_xcursor->mx_db.md_entries - ecount;
}
- /* sub-writes might have failed so check rc again.
- * Don't increment count if we just replaced an existing item.
- */
- if (!rc && !(flags & MDB_CURRENT))
+ /* Increment count unless we just replaced an existing item. */
+ if (insert_data)
mc->mc_db->md_entries++;
+ if (insert_key) {
+ /* Invalidate txn if we created an empty sub-DB */
+ if (rc)
+ goto bad_sub;
+ /* If we succeeded and the key didn't exist before,
+ * make sure the cursor is marked valid.
+ */
+ mc->mc_flags |= C_INITIALIZED;
+ }
if (flags & MDB_MULTIPLE) {
if (!rc) {
-next_mult:
mcount++;
/* let caller know how many succeeded, if any */
data[1].mv_size = mcount;
if (mcount < dcount) {
data[0].mv_data = (char *)data[0].mv_data + data[0].mv_size;
+ insert_key = insert_data = 0;
goto more;
}
}
}
+ return rc;
+bad_sub:
+ if (rc == MDB_KEYEXIST) /* should not happen, we deleted that item */
+ rc = MDB_CORRUPTED;
}
-done:
- /* If we succeeded and the key didn't exist before, make sure
- * the cursor is marked valid.
- */
- if (!rc && insert)
- mc->mc_flags |= C_INITIALIZED;
+ mc->mc_txn->mt_flags |= MDB_TXN_ERROR;
return rc;
}
leaf = NODEPTR(mp, mc->mc_ki[mc->mc_top]);
if (F_ISSET(leaf->mn_flags, F_DUPDATA)) {
- if (!(flags & MDB_NODUPDATA)) {
+ if (flags & MDB_NODUPDATA) {
+ /* mdb_cursor_del0() will subtract the final entry */
+ mc->mc_db->md_entries -= mc->mc_xcursor->mx_db.md_entries - 1;
+ } else {
if (!F_ISSET(leaf->mn_flags, F_SUBDATA)) {
mc->mc_xcursor->mx_cursor.mc_pg[0] = NODEDATA(leaf);
}
rc = mdb_cursor_del(&mc->mc_xcursor->mx_cursor, MDB_NOSPILL);
+ if (rc)
+ return rc;
/* If sub-DB still has entries, we're done */
if (mc->mc_xcursor->mx_db.md_entries) {
if (leaf->mn_flags & F_SUBDATA) {
if (leaf->mn_flags & F_SUBDATA) {
/* add all the child DB's pages to the free list */
rc = mdb_drop0(&mc->mc_xcursor->mx_cursor, 0);
- if (rc == MDB_SUCCESS) {
- mc->mc_db->md_entries -=
- mc->mc_xcursor->mx_db.md_entries;
- }
+ if (rc)
+ goto fail;
}
}
memcpy(&pg, NODEDATA(leaf), sizeof(pg));
if ((rc = mdb_page_get(mc->mc_txn, pg, &omp, NULL)) ||
(rc = mdb_ovpage_free(mc, omp)))
- return rc;
+ goto fail;
}
del_key:
return mdb_cursor_del0(mc);
+
+fail:
+ mc->mc_txn->mt_flags |= MDB_TXN_ERROR;
+ return rc;
}
/** Allocate and initialize new pages for a database.
mx->mx_dbflag = DB_VALID|DB_DIRTY; /* DB_DIRTY guides mdb_cursor_touch */
#if UINT_MAX < SIZE_MAX
if (mx->mx_dbx.md_cmp == mdb_cmp_int && mx->mx_db.md_pad == sizeof(size_t))
-#ifdef MISALIGNED_OK
- mx->mx_dbx.md_cmp = mdb_cmp_long;
-#else
- mx->mx_dbx.md_cmp = mdb_cmp_cint;
-#endif
+ mx->mx_dbx.md_cmp = mdb_cmp_clong;
#endif
}
unsigned int snum = csrc->mc_snum;
MDB_node *s2;
/* must find the lowest key below src */
- mdb_page_search_lowest(csrc);
+ rc = mdb_page_search_lowest(csrc);
+ if (rc)
+ return rc;
if (IS_LEAF2(csrc->mc_pg[csrc->mc_top])) {
key.mv_size = csrc->mc_db->md_pad;
key.mv_data = LEAF2KEY(csrc->mc_pg[csrc->mc_top], 0, key.mv_size);
MDB_node *s2;
MDB_val bkey;
/* must find the lowest key below dst */
- mdb_page_search_lowest(cdst);
- if (IS_LEAF2(cdst->mc_pg[cdst->mc_top])) {
- bkey.mv_size = cdst->mc_db->md_pad;
- bkey.mv_data = LEAF2KEY(cdst->mc_pg[cdst->mc_top], 0, bkey.mv_size);
+ mdb_cursor_copy(cdst, &mn);
+ rc = mdb_page_search_lowest(&mn);
+ if (rc)
+ return rc;
+ if (IS_LEAF2(mn.mc_pg[mn.mc_top])) {
+ bkey.mv_size = mn.mc_db->md_pad;
+ bkey.mv_data = LEAF2KEY(mn.mc_pg[mn.mc_top], 0, bkey.mv_size);
} else {
- s2 = NODEPTR(cdst->mc_pg[cdst->mc_top], 0);
+ s2 = NODEPTR(mn.mc_pg[mn.mc_top], 0);
bkey.mv_size = NODEKSZ(s2);
bkey.mv_data = NODEKEY(s2);
}
- cdst->mc_snum = snum--;
- cdst->mc_top = snum;
- mdb_cursor_copy(cdst, &mn);
+ mn.mc_snum = snum--;
+ mn.mc_top = snum;
mn.mc_ki[snum] = 0;
rc = mdb_update_key(&mn, &bkey);
if (rc)
static int
mdb_page_merge(MDB_cursor *csrc, MDB_cursor *cdst)
{
- int rc;
- indx_t i, j;
- MDB_node *srcnode;
+ MDB_page *psrc, *pdst;
+ MDB_node *srcnode;
MDB_val key, data;
- unsigned nkeys;
+ unsigned nkeys;
+ int rc;
+ indx_t i, j;
+
+ psrc = csrc->mc_pg[csrc->mc_top];
+ pdst = cdst->mc_pg[cdst->mc_top];
- DPRINTF(("merging page %"Z"u into %"Z"u", csrc->mc_pg[csrc->mc_top]->mp_pgno,
- cdst->mc_pg[cdst->mc_top]->mp_pgno));
+ DPRINTF(("merging page %"Z"u into %"Z"u", psrc->mp_pgno, pdst->mp_pgno));
mdb_cassert(csrc, csrc->mc_snum > 1); /* can't merge root page */
mdb_cassert(csrc, cdst->mc_snum > 1);
/* Move all nodes from src to dst.
*/
- j = nkeys = NUMKEYS(cdst->mc_pg[cdst->mc_top]);
- if (IS_LEAF2(csrc->mc_pg[csrc->mc_top])) {
+ j = nkeys = NUMKEYS(pdst);
+ if (IS_LEAF2(psrc)) {
key.mv_size = csrc->mc_db->md_pad;
- key.mv_data = METADATA(csrc->mc_pg[csrc->mc_top]);
- for (i = 0; i < NUMKEYS(csrc->mc_pg[csrc->mc_top]); i++, j++) {
+ key.mv_data = METADATA(psrc);
+ for (i = 0; i < NUMKEYS(psrc); i++, j++) {
rc = mdb_node_add(cdst, j, &key, NULL, 0, 0);
if (rc != MDB_SUCCESS)
return rc;
key.mv_data = (char *)key.mv_data + key.mv_size;
}
} else {
- for (i = 0; i < NUMKEYS(csrc->mc_pg[csrc->mc_top]); i++, j++) {
- srcnode = NODEPTR(csrc->mc_pg[csrc->mc_top], i);
- if (i == 0 && IS_BRANCH(csrc->mc_pg[csrc->mc_top])) {
- unsigned int snum = csrc->mc_snum;
+ for (i = 0; i < NUMKEYS(psrc); i++, j++) {
+ srcnode = NODEPTR(psrc, i);
+ if (i == 0 && IS_BRANCH(psrc)) {
+ MDB_cursor mn;
MDB_node *s2;
+ mdb_cursor_copy(csrc, &mn);
/* must find the lowest key below src */
- mdb_page_search_lowest(csrc);
- if (IS_LEAF2(csrc->mc_pg[csrc->mc_top])) {
- key.mv_size = csrc->mc_db->md_pad;
- key.mv_data = LEAF2KEY(csrc->mc_pg[csrc->mc_top], 0, key.mv_size);
+ rc = mdb_page_search_lowest(&mn);
+ if (rc)
+ return rc;
+ if (IS_LEAF2(mn.mc_pg[mn.mc_top])) {
+ key.mv_size = mn.mc_db->md_pad;
+ key.mv_data = LEAF2KEY(mn.mc_pg[mn.mc_top], 0, key.mv_size);
} else {
- s2 = NODEPTR(csrc->mc_pg[csrc->mc_top], 0);
+ s2 = NODEPTR(mn.mc_pg[mn.mc_top], 0);
key.mv_size = NODEKSZ(s2);
key.mv_data = NODEKEY(s2);
}
- csrc->mc_snum = snum--;
- csrc->mc_top = snum;
} else {
key.mv_size = srcnode->mn_ksize;
key.mv_data = NODEKEY(srcnode);
}
DPRINTF(("dst page %"Z"u now has %u keys (%.1f%% filled)",
- cdst->mc_pg[cdst->mc_top]->mp_pgno, NUMKEYS(cdst->mc_pg[cdst->mc_top]),
- (float)PAGEFILL(cdst->mc_txn->mt_env, cdst->mc_pg[cdst->mc_top]) / 10));
+ pdst->mp_pgno, NUMKEYS(pdst),
+ (float)PAGEFILL(cdst->mc_txn->mt_env, pdst) / 10));
/* Unlink the src page from parent and add to free list.
*/
}
csrc->mc_top++;
- rc = mdb_midl_append(&csrc->mc_txn->mt_free_pgs,
- csrc->mc_pg[csrc->mc_top]->mp_pgno);
+ psrc = csrc->mc_pg[csrc->mc_top];
+ /* If not operating on FreeDB, allow this page to be reused
+ * in this txn. Otherwise just add to free list.
+ */
+ rc = mdb_page_loose(csrc, psrc);
if (rc)
return rc;
- if (IS_LEAF(csrc->mc_pg[csrc->mc_top]))
+ if (IS_LEAF(psrc))
csrc->mc_db->md_leaf_pages--;
else
csrc->mc_db->md_branch_pages--;
/* Adjust other cursors pointing to mp */
MDB_cursor *m2, *m3;
MDB_dbi dbi = csrc->mc_dbi;
- MDB_page *mp = cdst->mc_pg[cdst->mc_top];
for (m2 = csrc->mc_txn->mt_cursors[dbi]; m2; m2=m2->mc_next) {
if (csrc->mc_flags & C_SUB)
m3 = m2;
if (m3 == csrc) continue;
if (m3->mc_snum < csrc->mc_snum) continue;
- if (m3->mc_pg[csrc->mc_top] == csrc->mc_pg[csrc->mc_top]) {
- m3->mc_pg[csrc->mc_top] = mp;
+ if (m3->mc_pg[csrc->mc_top] == psrc) {
+ m3->mc_pg[csrc->mc_top] = pdst;
m3->mc_ki[csrc->mc_top] += nkeys;
}
}
}
- mdb_cursor_pop(csrc);
-
- return mdb_rebalance(csrc);
+ {
+ unsigned int snum = cdst->mc_snum;
+ uint16_t depth = cdst->mc_db->md_depth;
+ mdb_cursor_pop(cdst);
+ rc = mdb_rebalance(cdst);
+ /* Did the tree shrink? */
+ if (depth > cdst->mc_db->md_depth)
+ snum--;
+ cdst->mc_snum = snum;
+ cdst->mc_top = snum-1;
+ }
+ return rc;
}
/** Copy the contents of a cursor.
int rc;
unsigned int ptop, minkeys;
MDB_cursor mn;
+ indx_t oldki;
minkeys = 1 + (IS_BRANCH(mc->mc_pg[mc->mc_top]));
DPRINTF(("rebalancing %s page %"Z"u (has %u keys, %.1f%% full)",
}
}
} else if (IS_BRANCH(mp) && NUMKEYS(mp) == 1) {
+ int i;
DPUTS("collapsing root page!");
rc = mdb_midl_append(&mc->mc_txn->mt_free_pgs, mp->mp_pgno);
if (rc)
mc->mc_db->md_depth--;
mc->mc_db->md_branch_pages--;
mc->mc_ki[0] = mc->mc_ki[1];
+ for (i = 1; i<mc->mc_db->md_depth; i++) {
+ mc->mc_pg[i] = mc->mc_pg[i+1];
+ mc->mc_ki[i] = mc->mc_ki[i+1];
+ }
{
/* Adjust other cursors pointing to mp */
MDB_cursor *m2, *m3;
m3 = m2;
if (m3 == mc || m3->mc_snum < mc->mc_snum) continue;
if (m3->mc_pg[0] == mp) {
- int i;
m3->mc_snum--;
m3->mc_top--;
for (i=0; i<m3->mc_snum; i++) {
mdb_cursor_copy(mc, &mn);
mn.mc_xcursor = NULL;
+ oldki = mc->mc_ki[mc->mc_top];
if (mc->mc_ki[ptop] == 0) {
/* We're the leftmost leaf in our parent.
*/
* (A branch page must never have less than 2 keys.)
*/
minkeys = 1 + (IS_BRANCH(mn.mc_pg[mn.mc_top]));
- if (PAGEFILL(mc->mc_txn->mt_env, mn.mc_pg[mn.mc_top]) >= FILL_THRESHOLD && NUMKEYS(mn.mc_pg[mn.mc_top]) > minkeys)
- return mdb_node_move(&mn, mc);
- else {
- if (mc->mc_ki[ptop] == 0)
+ if (PAGEFILL(mc->mc_txn->mt_env, mn.mc_pg[mn.mc_top]) >= FILL_THRESHOLD && NUMKEYS(mn.mc_pg[mn.mc_top]) > minkeys) {
+ rc = mdb_node_move(&mn, mc);
+ if (mc->mc_ki[ptop]) {
+ oldki++;
+ }
+ } else {
+ if (mc->mc_ki[ptop] == 0) {
rc = mdb_page_merge(&mn, mc);
- else {
+ } else {
+ oldki += NUMKEYS(mn.mc_pg[mn.mc_top]);
mn.mc_ki[mn.mc_top] += mc->mc_ki[mn.mc_top] + 1;
rc = mdb_page_merge(mc, &mn);
mdb_cursor_copy(&mn, mc);
}
- mc->mc_flags &= ~(C_INITIALIZED|C_EOF);
+ mc->mc_flags &= ~C_EOF;
}
+ mc->mc_ki[mc->mc_top] = oldki;
return rc;
}
mdb_node_del(mc, mc->mc_db->md_pad);
mc->mc_db->md_entries--;
rc = mdb_rebalance(mc);
- if (rc != MDB_SUCCESS)
- mc->mc_txn->mt_flags |= MDB_TXN_ERROR;
- else {
+
+ if (rc == MDB_SUCCESS) {
MDB_cursor *m2, *m3;
MDB_dbi dbi = mc->mc_dbi;
nkeys = NUMKEYS(mp);
/* if mc points past last node in page, find next sibling */
- if (mc->mc_ki[mc->mc_top] >= nkeys)
- mdb_cursor_sibling(mc, 1);
+ if (mc->mc_ki[mc->mc_top] >= nkeys) {
+ rc = mdb_cursor_sibling(mc, 1);
+ if (rc == MDB_NOTFOUND)
+ rc = MDB_SUCCESS;
+ }
/* Adjust other cursors pointing to mp */
- for (m2 = mc->mc_txn->mt_cursors[dbi]; m2; m2=m2->mc_next) {
+ for (m2 = mc->mc_txn->mt_cursors[dbi]; !rc && m2; m2=m2->mc_next) {
m3 = (mc->mc_flags & C_SUB) ? &m2->mc_xcursor->mx_cursor : m2;
if (! (m2->mc_flags & m3->mc_flags & C_INITIALIZED))
continue;
if (m3->mc_ki[mc->mc_top] > ki)
m3->mc_ki[mc->mc_top]--;
}
- if (m3->mc_ki[mc->mc_top] >= nkeys)
- mdb_cursor_sibling(m3, 1);
+ if (m3->mc_ki[mc->mc_top] >= nkeys) {
+ rc = mdb_cursor_sibling(m3, 1);
+ if (rc == MDB_NOTFOUND)
+ rc = MDB_SUCCESS;
+ }
}
}
mc->mc_flags |= C_DEL;
}
+ if (rc)
+ mc->mc_txn->mt_flags |= MDB_TXN_ERROR;
return rc;
}
if (mc->mc_snum < 2) {
if ((rc = mdb_page_new(mc, P_BRANCH, 1, &pp)))
- return rc;
+ goto done;
/* shift current top to make room for new parent */
mc->mc_pg[1] = mc->mc_pg[0];
mc->mc_ki[1] = mc->mc_ki[0];
mc->mc_ki[0] = mc->mc_ki[1];
mc->mc_db->md_root = mp->mp_pgno;
mc->mc_db->md_depth--;
- return rc;
+ goto done;
}
mc->mc_snum = 2;
mc->mc_top = 1;
int x;
unsigned int lsize, rsize, ksize;
/* Move half of the keys to the right sibling */
- copy = NULL;
x = mc->mc_ki[mc->mc_top] - split_indx;
ksize = mc->mc_db->md_pad;
split = LEAF2KEY(mp, split_indx, ksize);
/* grab a page to hold a temporary copy */
copy = mdb_page_malloc(mc->mc_txn, 1);
- if (copy == NULL)
- return ENOMEM;
+ if (copy == NULL) {
+ rc = ENOMEM;
+ goto done;
+ }
copy->mp_pgno = mp->mp_pgno;
copy->mp_flags = mp->mp_flags;
copy->mp_lower = PAGEHDRSZ;
mn.mc_top--;
did_split = 1;
rc = mdb_page_split(&mn, &sepkey, NULL, rp->mp_pgno, 0);
+ if (rc)
+ goto done;
/* root split? */
if (mn.mc_snum == mc->mc_snum) {
}
mc->mc_flags ^= C_SPLITTING;
if (rc != MDB_SUCCESS) {
- return rc;
+ goto done;
}
if (nflags & MDB_APPEND) {
mc->mc_pg[mc->mc_top] = rp;
mc->mc_ki[mc->mc_top] = 0;
rc = mdb_node_add(mc, 0, newkey, newdata, newpgno, nflags);
if (rc)
- return rc;
+ goto done;
for (i=0; i<mc->mc_top; i++)
mc->mc_ki[i] = mn.mc_ki[i];
} else if (!IS_LEAF2(mp)) {
}
rc = mdb_node_add(mc, j, &rkey, rdata, pgno, flags);
- if (rc) {
- /* return tmp page to freelist */
- mdb_page_free(env, copy);
- return rc;
- }
+ if (rc)
+ goto done;
if (i == nkeys) {
i = 0;
j = 0;
}
}
}
- /* return tmp page to freelist */
- mdb_page_free(env, copy);
}
{
}
}
DPRINTF(("mp left: %d, rp left: %d", SIZELEFT(mp), SIZELEFT(rp)));
+
+done:
+ if (copy) /* tmp page */
+ mdb_page_free(env, copy);
+ if (rc)
+ mc->mc_txn->mt_flags |= MDB_TXN_ERROR;
return rc;
}
return mdb_cursor_put(&mc, key, data, flags);
}
+#ifndef MDB_WBUF
+#define MDB_WBUF (1024*1024)
+#endif
+
+ /** State needed for a compacting copy. */
+typedef struct mdb_copy {
+ pthread_mutex_t mc_mutex;
+ pthread_cond_t mc_cond;
+ char *mc_wbuf[2];
+ char *mc_over[2];
+ MDB_env *mc_env;
+ MDB_txn *mc_txn;
+ int mc_wlen[2];
+ int mc_olen[2];
+ pgno_t mc_next_pgno;
+ HANDLE mc_fd;
+ int mc_status;
+ volatile int mc_new;
+ int mc_toggle;
+
+} mdb_copy;
+
+ /** Dedicated writer thread for compacting copy. */
+static THREAD_RET
+mdb_env_copythr(void *arg)
+{
+ mdb_copy *my = arg;
+ char *ptr;
+ int toggle = 0, wsize, rc;
+#ifdef _WIN32
+ DWORD len;
+#define DO_WRITE(rc, fd, ptr, w2, len) rc = WriteFile(fd, ptr, w2, &len, NULL)
+#else
+ int len;
+#define DO_WRITE(rc, fd, ptr, w2, len) len = write(fd, ptr, w2); rc = (len >= 0)
+#endif
+
+ pthread_mutex_lock(&my->mc_mutex);
+ my->mc_new = 0;
+ pthread_cond_signal(&my->mc_cond);
+ for(;;) {
+ while (!my->mc_new)
+ pthread_cond_wait(&my->mc_cond, &my->mc_mutex);
+ if (my->mc_new < 0) {
+ my->mc_new = 0;
+ break;
+ }
+ my->mc_new = 0;
+ wsize = my->mc_wlen[toggle];
+ ptr = my->mc_wbuf[toggle];
+again:
+ while (wsize > 0) {
+ DO_WRITE(rc, my->mc_fd, ptr, wsize, len);
+ if (!rc) {
+ rc = ErrCode();
+ break;
+ } else if (len > 0) {
+ rc = MDB_SUCCESS;
+ ptr += len;
+ wsize -= len;
+ continue;
+ } else {
+ rc = EIO;
+ break;
+ }
+ }
+ if (rc) {
+ my->mc_status = rc;
+ break;
+ }
+ /* If there's an overflow page tail, write it too */
+ if (my->mc_olen[toggle]) {
+ wsize = my->mc_olen[toggle];
+ ptr = my->mc_over[toggle];
+ my->mc_olen[toggle] = 0;
+ goto again;
+ }
+ my->mc_wlen[toggle] = 0;
+ toggle ^= 1;
+ pthread_cond_signal(&my->mc_cond);
+ }
+ pthread_cond_signal(&my->mc_cond);
+ pthread_mutex_unlock(&my->mc_mutex);
+ return (THREAD_RET)0;
+#undef DO_WRITE
+}
+
+ /** Tell the writer thread there's a buffer ready to write */
+static int
+mdb_env_cthr_toggle(mdb_copy *my, int st)
+{
+ int toggle = my->mc_toggle ^ 1;
+ pthread_mutex_lock(&my->mc_mutex);
+ if (my->mc_status) {
+ pthread_mutex_unlock(&my->mc_mutex);
+ return my->mc_status;
+ }
+ while (my->mc_new == 1)
+ pthread_cond_wait(&my->mc_cond, &my->mc_mutex);
+ my->mc_new = st;
+ my->mc_toggle = toggle;
+ pthread_cond_signal(&my->mc_cond);
+ pthread_mutex_unlock(&my->mc_mutex);
+ return 0;
+}
+
+ /** Depth-first tree traversal for compacting copy. */
+static int
+mdb_env_cwalk(mdb_copy *my, pgno_t *pg, int flags)
+{
+ MDB_cursor mc;
+ MDB_txn *txn = my->mc_txn;
+ MDB_node *ni;
+ MDB_page *mo, *mp, *leaf;
+ char *buf, *ptr;
+ int rc, toggle;
+ unsigned int i;
+
+ /* Empty DB, nothing to do */
+ if (*pg == P_INVALID)
+ return MDB_SUCCESS;
+
+ mc.mc_snum = 1;
+ mc.mc_top = 0;
+ mc.mc_txn = txn;
+
+ rc = mdb_page_get(my->mc_txn, *pg, &mc.mc_pg[0], NULL);
+ if (rc)
+ return rc;
+ rc = mdb_page_search_root(&mc, NULL, MDB_PS_FIRST);
+ if (rc)
+ return rc;
+
+ /* Make cursor pages writable */
+ buf = ptr = malloc(my->mc_env->me_psize * mc.mc_snum);
+ if (buf == NULL)
+ return ENOMEM;
+
+ for (i=0; i<mc.mc_top; i++) {
+ mdb_page_copy((MDB_page *)ptr, mc.mc_pg[i], my->mc_env->me_psize);
+ mc.mc_pg[i] = (MDB_page *)ptr;
+ ptr += my->mc_env->me_psize;
+ }
+
+ /* This is writable space for a leaf page. Usually not needed. */
+ leaf = (MDB_page *)ptr;
+
+ toggle = my->mc_toggle;
+ while (mc.mc_snum > 0) {
+ unsigned n;
+ mp = mc.mc_pg[mc.mc_top];
+ n = NUMKEYS(mp);
+
+ if (IS_LEAF(mp)) {
+ if (!IS_LEAF2(mp) && !(flags & F_DUPDATA)) {
+ for (i=0; i<n; i++) {
+ ni = NODEPTR(mp, i);
+ if (ni->mn_flags & F_BIGDATA) {
+ MDB_page *omp;
+ pgno_t pg;
+
+ /* Need writable leaf */
+ if (mp != leaf) {
+ mc.mc_pg[mc.mc_top] = leaf;
+ mdb_page_copy(leaf, mp, my->mc_env->me_psize);
+ mp = leaf;
+ ni = NODEPTR(mp, i);
+ }
+
+ memcpy(&pg, NODEDATA(ni), sizeof(pg));
+ rc = mdb_page_get(txn, pg, &omp, NULL);
+ if (rc)
+ goto done;
+ if (my->mc_wlen[toggle] >= MDB_WBUF) {
+ rc = mdb_env_cthr_toggle(my, 1);
+ if (rc)
+ goto done;
+ toggle = my->mc_toggle;
+ }
+ mo = (MDB_page *)(my->mc_wbuf[toggle] + my->mc_wlen[toggle]);
+ memcpy(mo, omp, my->mc_env->me_psize);
+ mo->mp_pgno = my->mc_next_pgno;
+ my->mc_next_pgno += omp->mp_pages;
+ my->mc_wlen[toggle] += my->mc_env->me_psize;
+ if (omp->mp_pages > 1) {
+ my->mc_olen[toggle] = my->mc_env->me_psize * (omp->mp_pages - 1);
+ my->mc_over[toggle] = (char *)omp + my->mc_env->me_psize;
+ rc = mdb_env_cthr_toggle(my, 1);
+ if (rc)
+ goto done;
+ toggle = my->mc_toggle;
+ }
+ memcpy(NODEDATA(ni), &mo->mp_pgno, sizeof(pgno_t));
+ } else if (ni->mn_flags & F_SUBDATA) {
+ MDB_db db;
+
+ /* Need writable leaf */
+ if (mp != leaf) {
+ mc.mc_pg[mc.mc_top] = leaf;
+ mdb_page_copy(leaf, mp, my->mc_env->me_psize);
+ mp = leaf;
+ ni = NODEPTR(mp, i);
+ }
+
+ memcpy(&db, NODEDATA(ni), sizeof(db));
+ my->mc_toggle = toggle;
+ rc = mdb_env_cwalk(my, &db.md_root, ni->mn_flags & F_DUPDATA);
+ if (rc)
+ goto done;
+ toggle = my->mc_toggle;
+ memcpy(NODEDATA(ni), &db, sizeof(db));
+ }
+ }
+ }
+ } else {
+ mc.mc_ki[mc.mc_top]++;
+ if (mc.mc_ki[mc.mc_top] < n) {
+ pgno_t pg;
+again:
+ ni = NODEPTR(mp, mc.mc_ki[mc.mc_top]);
+ pg = NODEPGNO(ni);
+ rc = mdb_page_get(txn, pg, &mp, NULL);
+ if (rc)
+ goto done;
+ mc.mc_top++;
+ mc.mc_snum++;
+ mc.mc_ki[mc.mc_top] = 0;
+ if (IS_BRANCH(mp)) {
+ /* Whenever we advance to a sibling branch page,
+ * we must proceed all the way down to its first leaf.
+ */
+ mdb_page_copy(mc.mc_pg[mc.mc_top], mp, my->mc_env->me_psize);
+ goto again;
+ } else
+ mc.mc_pg[mc.mc_top] = mp;
+ continue;
+ }
+ }
+ if (my->mc_wlen[toggle] >= MDB_WBUF) {
+ rc = mdb_env_cthr_toggle(my, 1);
+ if (rc)
+ goto done;
+ toggle = my->mc_toggle;
+ }
+ mo = (MDB_page *)(my->mc_wbuf[toggle] + my->mc_wlen[toggle]);
+ mdb_page_copy(mo, mp, my->mc_env->me_psize);
+ mo->mp_pgno = my->mc_next_pgno++;
+ my->mc_wlen[toggle] += my->mc_env->me_psize;
+ if (mc.mc_top) {
+ /* Update parent if there is one */
+ ni = NODEPTR(mc.mc_pg[mc.mc_top-1], mc.mc_ki[mc.mc_top-1]);
+ SETPGNO(ni, mo->mp_pgno);
+ mdb_cursor_pop(&mc);
+ } else {
+ /* Otherwise we're done */
+ *pg = mo->mp_pgno;
+ break;
+ }
+ }
+done:
+ free(buf);
+ return rc;
+}
+
+ /** Copy environment with compaction. */
+static int
+mdb_env_copyfd1(MDB_env *env, HANDLE fd)
+{
+ MDB_meta *mm;
+ MDB_page *mp;
+ mdb_copy my;
+ MDB_txn *txn = NULL;
+ pthread_t thr;
+ int rc;
+
+#ifdef _WIN32
+ my.mc_mutex = CreateMutex(NULL, FALSE, NULL);
+ my.mc_cond = CreateEvent(NULL, FALSE, FALSE, NULL);
+ my.mc_wbuf[0] = _aligned_malloc(MDB_WBUF*2, env->me_psize);
+ if (my.mc_wbuf[0] == NULL)
+ return errno;
+#else
+ pthread_mutex_init(&my.mc_mutex, NULL);
+ pthread_cond_init(&my.mc_cond, NULL);
+ rc = posix_memalign((void **)&my.mc_wbuf[0], env->me_psize, MDB_WBUF*2);
+ if (rc)
+ return rc;
+#endif
+ my.mc_wbuf[1] = my.mc_wbuf[0] + MDB_WBUF;
+ my.mc_wlen[0] = 0;
+ my.mc_wlen[1] = 0;
+ my.mc_olen[0] = 0;
+ my.mc_olen[1] = 0;
+ my.mc_next_pgno = 2;
+ my.mc_status = 0;
+ my.mc_new = 1;
+ my.mc_toggle = 0;
+ my.mc_env = env;
+ my.mc_fd = fd;
+
+ /* Do the lock/unlock of the reader mutex before starting the
+ * write txn. Otherwise other read txns could block writers.
+ */
+ rc = mdb_txn_begin(env, NULL, MDB_RDONLY, &txn);
+ if (rc)
+ return rc;
+
+ if (env->me_txns) {
+ /* We must start the actual read txn after blocking writers */
+ mdb_txn_reset0(txn, "reset-stage1");
+
+ /* Temporarily block writers until we snapshot the meta pages */
+ LOCK_MUTEX_W(env);
+
+ rc = mdb_txn_renew0(txn);
+ if (rc) {
+ UNLOCK_MUTEX_W(env);
+ goto leave;
+ }
+ }
+
+ THREAD_CREATE(thr, mdb_env_copythr, &my);
+ mp = (MDB_page *)my.mc_wbuf[0];
+ memset(mp, 0, 2*env->me_psize);
+ mp->mp_pgno = 0;
+ mp->mp_flags = P_META;
+ mm = (MDB_meta *)METADATA(mp);
+ mdb_env_init_meta0(env, mm);
+ mm->mm_address = env->me_metas[0]->mm_address;
+
+ mp = (MDB_page *)(my.mc_wbuf[0] + env->me_psize);
+ mp->mp_pgno = 1;
+ mp->mp_flags = P_META;
+ *(MDB_meta *)METADATA(mp) = *mm;
+ mm = (MDB_meta *)METADATA(mp);
+
+ /* Count the number of free pages, subtract from lastpg to find
+ * number of active pages
+ */
+ {
+ MDB_ID freecount = 0;
+ MDB_cursor mc;
+ MDB_val key, data;
+ mdb_cursor_init(&mc, txn, FREE_DBI, NULL);
+ while ((rc = mdb_cursor_get(&mc, &key, &data, MDB_NEXT)) == 0)
+ freecount += *(MDB_ID *)data.mv_data;
+ freecount += txn->mt_dbs[0].md_branch_pages +
+ txn->mt_dbs[0].md_leaf_pages +
+ txn->mt_dbs[0].md_overflow_pages;
+
+ /* Set metapage 1 */
+ mm->mm_last_pg = txn->mt_next_pgno - freecount - 1;
+ mm->mm_dbs[1] = txn->mt_dbs[1];
+ mm->mm_dbs[1].md_root = mm->mm_last_pg;
+ mm->mm_txnid = 1;
+ }
+ my.mc_wlen[0] = env->me_psize * 2;
+ my.mc_txn = txn;
+ pthread_mutex_lock(&my.mc_mutex);
+ while(my.mc_new)
+ pthread_cond_wait(&my.mc_cond, &my.mc_mutex);
+ pthread_mutex_unlock(&my.mc_mutex);
+ rc = mdb_env_cwalk(&my, &txn->mt_dbs[1].md_root, 0);
+ if (rc == MDB_SUCCESS && my.mc_wlen[my.mc_toggle])
+ rc = mdb_env_cthr_toggle(&my, 1);
+ mdb_env_cthr_toggle(&my, -1);
+ pthread_mutex_lock(&my.mc_mutex);
+ while(my.mc_new)
+ pthread_cond_wait(&my.mc_cond, &my.mc_mutex);
+ pthread_mutex_unlock(&my.mc_mutex);
+ THREAD_FINISH(thr);
+leave:
+ mdb_txn_abort(txn);
+#ifdef _WIN32
+ CloseHandle(my.mc_cond);
+ CloseHandle(my.mc_mutex);
+ _aligned_free(my.mc_wbuf[0]);
+#else
+ pthread_cond_destroy(&my.mc_cond);
+ pthread_mutex_destroy(&my.mc_mutex);
+ free(my.mc_wbuf[0]);
+#endif
+ return rc;
+}
+
+ /** Copy environment as-is. */
+static int
+mdb_env_copyfd0(MDB_env *env, HANDLE fd)
+{
+ MDB_txn *txn = NULL;
+ int rc;
+ size_t wsize;
+ char *ptr;
+#ifdef _WIN32
+ DWORD len, w2;
+#define DO_WRITE(rc, fd, ptr, w2, len) rc = WriteFile(fd, ptr, w2, &len, NULL)
+#else
+ ssize_t len;
+ size_t w2;
+#define DO_WRITE(rc, fd, ptr, w2, len) len = write(fd, ptr, w2); rc = (len >= 0)
+#endif
+
+ /* Do the lock/unlock of the reader mutex before starting the
+ * write txn. Otherwise other read txns could block writers.
+ */
+ rc = mdb_txn_begin(env, NULL, MDB_RDONLY, &txn);
+ if (rc)
+ return rc;
+
+ if (env->me_txns) {
+ /* We must start the actual read txn after blocking writers */
+ mdb_txn_reset0(txn, "reset-stage1");
+
+ /* Temporarily block writers until we snapshot the meta pages */
+ LOCK_MUTEX_W(env);
+
+ rc = mdb_txn_renew0(txn);
+ if (rc) {
+ UNLOCK_MUTEX_W(env);
+ goto leave;
+ }
+ }
+
+ wsize = env->me_psize * 2;
+ ptr = env->me_map;
+ w2 = wsize;
+ while (w2 > 0) {
+ DO_WRITE(rc, fd, ptr, w2, len);
+ if (!rc) {
+ rc = ErrCode();
+ break;
+ } else if (len > 0) {
+ rc = MDB_SUCCESS;
+ ptr += len;
+ w2 -= len;
+ continue;
+ } else {
+ /* Non-blocking or async handles are not supported */
+ rc = EIO;
+ break;
+ }
+ }
+ if (env->me_txns)
+ UNLOCK_MUTEX_W(env);
+
+ if (rc)
+ goto leave;
+
+ w2 = txn->mt_next_pgno * env->me_psize;
+#ifdef WIN32
+ {
+ LARGE_INTEGER fsize;
+ GetFileSizeEx(env->me_fd, &fsize);
+ if (w2 > fsize.QuadPart)
+ w2 = fsize.QuadPart;
+ }
+#else
+ {
+ struct stat st;
+ fstat(env->me_fd, &st);
+ if (w2 > (size_t)st.st_size)
+ w2 = st.st_size;
+ }
+#endif
+ wsize = w2 - wsize;
+ while (wsize > 0) {
+ if (wsize > MAX_WRITE)
+ w2 = MAX_WRITE;
+ else
+ w2 = wsize;
+ DO_WRITE(rc, fd, ptr, w2, len);
+ if (!rc) {
+ rc = ErrCode();
+ break;
+ } else if (len > 0) {
+ rc = MDB_SUCCESS;
+ ptr += len;
+ wsize -= len;
+ continue;
+ } else {
+ rc = EIO;
+ break;
+ }
+ }
+
+leave:
+ mdb_txn_abort(txn);
+ return rc;
+}
+
+int
+mdb_env_copyfd2(MDB_env *env, HANDLE fd, unsigned int flags)
+{
+ if (flags & MDB_CP_COMPACT)
+ return mdb_env_copyfd1(env, fd);
+ else
+ return mdb_env_copyfd0(env, fd);
+}
+
+int
+mdb_env_copyfd(MDB_env *env, HANDLE fd)
+{
+ return mdb_env_copyfd2(env, fd, 0);
+}
+
+int
+mdb_env_copy2(MDB_env *env, const char *path, unsigned int flags)
+{
+ int rc, len;
+ char *lpath;
+ HANDLE newfd = INVALID_HANDLE_VALUE;
+
+ if (env->me_flags & MDB_NOSUBDIR) {
+ lpath = (char *)path;
+ } else {
+ len = strlen(path);
+ len += sizeof(DATANAME);
+ lpath = malloc(len);
+ if (!lpath)
+ return ENOMEM;
+ sprintf(lpath, "%s" DATANAME, path);
+ }
+
+ /* The destination path must exist, but the destination file must not.
+ * We don't want the OS to cache the writes, since the source data is
+ * already in the OS cache.
+ */
+#ifdef _WIN32
+ newfd = CreateFile(lpath, GENERIC_WRITE, 0, NULL, CREATE_NEW,
+ FILE_FLAG_NO_BUFFERING|FILE_FLAG_WRITE_THROUGH, NULL);
+#else
+ newfd = open(lpath, O_WRONLY|O_CREAT|O_EXCL, 0666);
+#endif
+ if (newfd == INVALID_HANDLE_VALUE) {
+ rc = ErrCode();
+ goto leave;
+ }
+
+#ifdef O_DIRECT
+ /* Set O_DIRECT if the file system supports it */
+ if ((rc = fcntl(newfd, F_GETFL)) != -1)
+ (void) fcntl(newfd, F_SETFL, rc | O_DIRECT);
+#endif
+#ifdef F_NOCACHE /* __APPLE__ */
+ rc = fcntl(newfd, F_NOCACHE, 1);
+ if (rc) {
+ rc = ErrCode();
+ goto leave;
+ }
+#endif
+
+ rc = mdb_env_copyfd2(env, newfd, flags);
+
+leave:
+ if (!(env->me_flags & MDB_NOSUBDIR))
+ free(lpath);
+ if (newfd != INVALID_HANDLE_VALUE)
+ if (close(newfd) < 0 && rc == MDB_SUCCESS)
+ rc = ErrCode();
+
+ return rc;
+}
+
+int
+mdb_env_copy(MDB_env *env, const char *path)
+{
+ return mdb_env_copy2(env, path, 0);
+}
+
int
mdb_env_set_flags(MDB_env *env, unsigned int flag, int onoff)
{
return EINVAL;
toggle = mdb_env_pick_meta(env);
- arg->me_mapaddr = (env->me_flags & MDB_FIXEDMAP) ? env->me_map : 0;
+ arg->me_mapaddr = env->me_metas[toggle]->mm_address;
arg->me_mapsize = env->me_mapsize;
arg->me_maxreaders = env->me_maxreaders;
memcpy(&pg, NODEDATA(ni), sizeof(pg));
rc = mdb_page_get(txn, pg, &omp, NULL);
if (rc != 0)
- return rc;
+ goto done;
mdb_cassert(mc, IS_OVERFLOW(omp));
rc = mdb_midl_append_range(&txn->mt_free_pgs,
pg, omp->mp_pages);
if (rc)
- return rc;
+ goto done;
} else if (subs && (ni->mn_flags & F_SUBDATA)) {
mdb_xcursor_init1(mc, ni);
rc = mdb_drop0(&mc->mc_xcursor->mx_cursor, 0);
if (rc)
- return rc;
+ goto done;
}
}
} else {
if ((rc = mdb_midl_need(&txn->mt_free_pgs, n)) != 0)
- return rc;
+ goto done;
for (i=0; i<n; i++) {
pgno_t pg;
ni = NODEPTR(mp, i);
mc->mc_ki[mc->mc_top] = i;
rc = mdb_cursor_sibling(mc, 1);
if (rc) {
+ if (rc != MDB_NOTFOUND)
+ goto done;
/* no more siblings, go back to beginning
* of previous level.
*/
}
/* free it */
rc = mdb_midl_append(&txn->mt_free_pgs, mc->mc_db->md_root);
+done:
+ if (rc)
+ txn->mt_flags |= MDB_TXN_ERROR;
} else if (rc == MDB_NOTFOUND) {
rc = MDB_SUCCESS;
}
if (!rc) {
txn->mt_dbflags[dbi] = DB_STALE;
mdb_dbi_close(txn->mt_env, dbi);
+ } else {
+ txn->mt_flags |= MDB_TXN_ERROR;
}
} else {
/* reset the DB record, mark it dirty */
projects / openldap / blobdiff