* @{
*/
#ifdef _WIN32
+#define MDB_PIDLOCK 0
#define pthread_t DWORD
#define pthread_mutex_t HANDLE
#define pthread_key_t DWORD
#define GET_PAGESIZE(x) {SYSTEM_INFO si; GetSystemInfo(&si); (x) = si.dwPageSize;}
#define close(fd) (CloseHandle(fd) ? 0 : -1)
#define munmap(ptr,len) UnmapViewOfFile(ptr)
+#ifdef PROCESS_QUERY_LIMITED_INFORMATION
+#define MDB_PROCESS_QUERY_LIMITED_INFORMATION PROCESS_QUERY_LIMITED_INFORMATION
#else
+#define MDB_PROCESS_QUERY_LIMITED_INFORMATION 0x1000
+#endif
+#define Z "I"
+#else
+
+#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_PIDLOCK 1
#ifdef MDB_USE_POSIX_SEM
* @{
*/
#ifndef MDB_DEBUG
- /** Enable debug output.
+ /** Enable debug output. Needs variable argument macros (a C99 feature).
* Set this to 1 for copious tracing. Set to 2 to add dumps of all IDLs
* read from and written to the database (used for free space management).
*/
#define MDB_DEBUG 0
#endif
-#if !(__STDC_VERSION__ >= 199901L || defined(__GNUC__))
-# undef MDB_DEBUG
-# define MDB_DEBUG 0
-# define DPRINTF (void) /* Vararg macros may be unsupported */
-#elif MDB_DEBUG
+#if MDB_DEBUG
static int mdb_debug;
static txnid_t mdb_debug_start;
- /** Print a debug message with printf formatting. */
-# define DPRINTF(fmt, ...) /**< Requires 2 or more args */ \
- ((void) ((mdb_debug) && \
- fprintf(stderr, "%s:%d " fmt "\n", __func__, __LINE__, __VA_ARGS__)))
+ /** Print a debug message with printf formatting.
+ * Requires double parenthesis around 2 or more args.
+ */
+# define DPRINTF(args) ((void) ((mdb_debug) && DPRINTF0 args))
+# define DPRINTF0(fmt, ...) \
+ fprintf(stderr, "%s:%d " fmt "\n", __func__, __LINE__, __VA_ARGS__)
#else
-# define DPRINTF(fmt, ...) ((void) 0)
-# define MDB_DEBUG_SKIP
+# define DPRINTF(args) ((void) 0)
#endif
/** Print a debug string.
* The string is printed literally, with no format processing.
*/
-#define DPUTS(arg) DPRINTF("%s", arg)
+#define DPUTS(arg) DPRINTF(("%s", arg))
+ /** Debuging output value of a cursor DBI: Negative in a sub-cursor. */
+#define DDBI(mc) \
+ (((mc)->mc_flags & C_SUB) ? -(int)(mc)->mc_dbi : (int)(mc)->mc_dbi)
/** @} */
/** A default memory page size.
*/
#define MDB_MAGIC 0xBEEFC0DE
- /** The version number for a database's file format. */
-#define MDB_VERSION 1
+ /** The version number for a database's datafile format. */
+#define MDB_DATA_VERSION 1
+ /** The version number for a database's lockfile format. */
+#define MDB_LOCK_VERSION 1
/** @brief The maximum size of a key in the database.
*
*/
#define DKEY(x) mdb_dkey(x, kbuf)
#else
-#define DKBUF typedef int dummy_kbuf /* so we can put ';' after */
+#define DKBUF
#define DKEY(x) 0
#endif
*
* If #MDB_NOTLS is set, the slot address is not saved in thread-specific data.
*
- * No reader table is used if the database is on a read-only filesystem.
+ * No reader table is used if the database is on a read-only filesystem, or
+ * if #MDB_NOLOCK is set.
*
* Since the database uses multi-version concurrency control, readers don't
* actually need any locking. This table is used to keep track of which
/** Stamp identifying this as an MDB file. It must be set
* to #MDB_MAGIC. */
uint32_t mtb_magic;
- /** Version number of this lock file. Must be set to #MDB_VERSION. */
- uint32_t mtb_version;
+ /** Format of this lock file. Must be set to #MDB_LOCK_FORMAT. */
+ uint32_t mtb_format;
#if defined(_WIN32) || defined(MDB_USE_POSIX_SEM)
char mtb_rmname[MNAME_LEN];
#else
union {
MDB_txbody mtb;
#define mti_magic mt1.mtb.mtb_magic
-#define mti_version mt1.mtb.mtb_version
+#define mti_format mt1.mtb.mtb_format
#define mti_mutex mt1.mtb.mtb_mutex
#define mti_rmname mt1.mtb.mtb_rmname
#define mti_txnid mt1.mtb.mtb_txnid
} mt2;
MDB_reader mti_readers[1];
} MDB_txninfo;
+
+ /** Lockfile format signature: version, features and field layout */
+#define MDB_LOCK_FORMAT \
+ ((uint32_t) \
+ ((MDB_LOCK_VERSION) \
+ /* Flags which describe functionality */ \
+ + (((MDB_PIDLOCK) != 0) << 16)))
/** @} */
/** Common header for all page types.
#define P_LEAF 0x02 /**< leaf page */
#define P_OVERFLOW 0x04 /**< overflow page */
#define P_META 0x08 /**< meta page */
-#define P_DIRTY 0x10 /**< dirty page */
+#define P_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_KEEP 0x8000 /**< leave this page alone during spill */
/** @} */
uint16_t mp_flags; /**< @ref mdb_page */
#define mp_lower mp_pb.pb.pb_lower
*/
#define LEAF2KEY(p, i, ks) ((char *)(p) + PAGEHDRSZ + ((i)*(ks)))
- /** Set the \b node's key into \b key, if requested. */
-#define MDB_GET_KEY(node, key) { if ((key) != NULL) { \
- (key)->mv_size = NODEKSZ(node); (key)->mv_data = NODEKEY(node); } }
+ /** Set the \b node's key into \b keyptr, if requested. */
+#define MDB_GET_KEY(node, keyptr) { if ((keyptr) != NULL) { \
+ (keyptr)->mv_size = NODEKSZ(node); (keyptr)->mv_data = NODEKEY(node); } }
+
+ /** Set the \b node's key into \b key. */
+#define MDB_GET_KEY2(node, key) { key.mv_size = NODEKSZ(node); key.mv_data = NODEKEY(node); }
/** Information about a single database in the environment. */
typedef struct MDB_db {
/** Handle for the default DB. */
#define MAIN_DBI 1
- /** Meta page content. */
+ /** Meta page content.
+ * A meta page is the start point for accessing a database snapshot.
+ * 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
* to #MDB_MAGIC. */
uint32_t mm_magic;
- /** Version number of this lock file. Must be set to #MDB_VERSION. */
+ /** Version number of this lock file. Must be set to #MDB_DATA_VERSION. */
uint32_t mm_version;
void *mm_address; /**< address for fixed mapping */
size_t mm_mapsize; /**< size of mmap region */
/** The list of pages that became unused during this transaction.
*/
MDB_IDL mt_free_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.
+ */
+ MDB_IDL mt_spill_pgs;
union {
- MDB_ID2L dirty_list; /**< for write txns: modified pages */
- MDB_reader *reader; /**< this thread's reader table slot or NULL */
+ /** For write txns: Modified pages. Sorted when not MDB_WRITEMAP. */
+ MDB_ID2L dirty_list;
+ /** For read txns: This thread/txn's reader table slot, or NULL. */
+ MDB_reader *reader;
} mt_u;
/** Array of records for each DB known in the environment. */
MDB_dbx *mt_dbxs;
* @ingroup internal
* @{
*/
-#define DB_DIRTY 0x01 /**< DB was written in this txn */
-#define DB_STALE 0x02 /**< DB record is older than txnID */
-#define DB_NEW 0x04 /**< DB handle opened in this txn */
+#define DB_DIRTY 0x01 /**< DB was modified or is DUPSORT data */
+#define DB_STALE 0x02 /**< Named-DB record is older than txnID */
+#define DB_NEW 0x04 /**< Named-DB handle opened in this txn */
#define DB_VALID 0x08 /**< DB handle is valid, see also #MDB_VALID */
/** @} */
/** In write txns, array of cursors for each DB */
#define MDB_TXN_RDONLY 0x01 /**< read-only transaction */
#define MDB_TXN_ERROR 0x02 /**< an error has occurred */
#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 */
/** @} */
unsigned int mt_flags; /**< @ref mdb_txn */
- /** dirty_list maxsize - # of allocated pages allowed, including in parent txns */
- unsigned int mt_dirty_room;
- /** Tracks which of the two meta pages was used at the start
- * of this transaction.
+ /** dirty_list room: Array size - #dirty pages visible to this txn.
+ * Includes ancestor txns' dirty pages not hidden by other txns'
+ * dirty/spilled pages. Thus commit(nested txn) has room to merge
+ * dirty_list into mt_parent after freeing hidden mt_parent pages.
*/
- unsigned int mt_toggle;
+ unsigned int mt_dirty_room;
};
/** Enough space for 2^32 nodes with minimum of 2 keys per node. I.e., plenty.
struct MDB_xcursor;
- /** Cursors are used for all DB operations */
+ /** Cursors are used for all DB operations.
+ * A cursor holds a path of (page pointer, key index) from the DB
+ * root to a position in the DB, plus other state. #MDB_DUPSORT
+ * cursors include an xcursor to the current data item. Write txns
+ * track their cursors and keep them up to date when data moves.
+ * Exception: An xcursor's pointer to a #P_SUBP page can be stale.
+ * (A node with #F_DUPDATA but no #F_SUBDATA contains a subpage).
+ */
struct MDB_cursor {
/** Next cursor on this DB in this txn */
MDB_cursor *mc_next;
#define C_INITIALIZED 0x01 /**< cursor has been initialized and is valid */
#define C_EOF 0x02 /**< No more data */
#define C_SUB 0x04 /**< Cursor is a sub-cursor */
+#define C_DEL 0x08 /**< last op was a cursor_del */
#define C_SPLITTING 0x20 /**< Cursor is in page_split */
#define C_UNTRACK 0x40 /**< Un-track cursor when closing */
/** @} */
#define MDB_ENV_ACTIVE 0x20000000U
/** me_txkey is set */
#define MDB_ENV_TXKEY 0x10000000U
+ /** Have liveness lock in reader table */
+#define MDB_LIVE_READER 0x08000000U
uint32_t me_flags; /**< @ref mdb_env */
unsigned int me_psize; /**< size of a page, from #GET_PAGESIZE */
unsigned int me_maxreaders; /**< size of the reader table */
/** Max size of a node on a page */
unsigned int me_nodemax;
#ifdef _WIN32
+ int me_pidquery; /**< Used in OpenProcess */
HANDLE me_rmutex; /* Windows mutexes don't reside in shared mem */
HANDLE me_wmutex;
#elif defined(MDB_USE_POSIX_SEM)
/** Nested transaction */
typedef struct MDB_ntxn {
- MDB_txn mnt_txn; /* the transaction */
- MDB_pgstate mnt_pgstate; /* parent transaction's saved freestate */
+ MDB_txn mnt_txn; /**< the transaction */
+ MDB_pgstate mnt_pgstate; /**< parent transaction's saved freestate */
} MDB_ntxn;
/** max number of pages to commit in one writev() call */
MDB_val *key, int modify);
#define MDB_PS_MODIFY 1
#define MDB_PS_ROOTONLY 2
+#define MDB_PS_FIRST 4
+#define MDB_PS_LAST 8
static int mdb_page_search(MDB_cursor *mc,
MDB_val *key, int flags);
static int mdb_page_merge(MDB_cursor *csrc, MDB_cursor *cdst);
"MDB_CURSOR_FULL: Internal error - cursor stack limit reached",
"MDB_PAGE_FULL: Internal error - page has no more space",
"MDB_MAP_RESIZED: Database contents grew beyond environment mapsize",
- "MDB_INCOMPATIBLE: Database flags changed or would change",
+ "MDB_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",
};
char *
DKBUF;
nkeys = NUMKEYS(mp);
- fprintf(stderr, "Page %zu numkeys %d\n", mp->mp_pgno, nkeys);
+ fprintf(stderr, "Page %"Z"u numkeys %d\n", mp->mp_pgno, nkeys);
for (i=0; i<nkeys; i++) {
node = NODEPTR(mp, i);
key.mv_size = node->mn_ksize;
key.mv_data = node->mn_data;
nsize = NODESIZE + NODEKSZ(node) + sizeof(indx_t);
if (IS_BRANCH(mp)) {
- fprintf(stderr, "key %d: page %zu, %s\n", i, NODEPGNO(node),
+ fprintf(stderr, "key %d: page %"Z"u, %s\n", i, NODEPGNO(node),
DKEY(&key));
} else {
if (F_ISSET(node->mn_flags, F_BIGDATA))
}
#endif
-#if MDB_DEBUG > 2
+#if (MDB_DEBUG) > 2
/** Count all the pages in each DB and in the freelist
* and make sure it matches the actual number of pages
* being used.
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, 0);
+ mdb_page_search(&mc, NULL, MDB_PS_FIRST);
do {
unsigned j;
MDB_page *mp;
return txn->mt_dbxs[dbi].md_dcmp(a, b);
}
-/** Allocate a page.
+/** Allocate memory for a page.
* Re-use old malloc'd pages first for singletons, otherwise just malloc.
*/
static MDB_page *
env->me_dpages = mp;
}
-/* Free a dirty page */
+/** Free a dirty page */
static void
mdb_dpage_free(MDB_env *env, MDB_page *dp)
{
}
}
-/* Return all dirty pages to dpage list */
+/** Return all dirty pages to dpage list */
static void
mdb_dlist_free(MDB_txn *txn)
{
dl[0].mid = 0;
}
+/** 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:
+ * P_DIRTY to set P_KEEP, P_DIRTY|P_KEEP to clear it.
+ * @param[in] all No shortcuts. Needed except after a full #mdb_page_flush().
+ * @return 0 on success, non-zero on failure.
+ */
+static int
+mdb_pages_xkeep(MDB_cursor *mc, unsigned pflags, int all)
+{
+ enum { Mask = P_SUBP|P_DIRTY|P_KEEP };
+ MDB_txn *txn = mc->mc_txn;
+ MDB_cursor *m3;
+ MDB_xcursor *mx;
+ MDB_page *dp, *mp;
+ MDB_node *leaf;
+ unsigned i, j;
+ int rc = MDB_SUCCESS, level;
+
+ /* Mark pages seen by cursors */
+ if (mc->mc_flags & C_UNTRACK)
+ mc = NULL; /* will find mc in mt_cursors */
+ for (i = txn->mt_numdbs;; mc = txn->mt_cursors[--i]) {
+ for (; mc; mc=mc->mc_next) {
+ if (!(mc->mc_flags & C_INITIALIZED))
+ continue;
+ for (m3 = mc;; m3 = &mx->mx_cursor) {
+ mp = NULL;
+ for (j=0; j<m3->mc_snum; j++) {
+ mp = m3->mc_pg[j];
+ if ((mp->mp_flags & Mask) == pflags)
+ mp->mp_flags ^= P_KEEP;
+ }
+ mx = m3->mc_xcursor;
+ /* Proceed to mx if it is at a sub-database */
+ if (! (mx && (mx->mx_cursor.mc_flags & C_INITIALIZED)))
+ break;
+ if (! (mp && (mp->mp_flags & P_LEAF)))
+ break;
+ leaf = NODEPTR(mp, m3->mc_ki[j-1]);
+ if (!(leaf->mn_flags & F_SUBDATA))
+ break;
+ }
+ }
+ if (i == 0)
+ break;
+ }
+
+ if (all) {
+ /* Mark dirty root pages */
+ for (i=0; i<txn->mt_numdbs; i++) {
+ if (txn->mt_dbflags[i] & DB_DIRTY) {
+ pgno_t pgno = txn->mt_dbs[i].md_root;
+ if (pgno == P_INVALID)
+ continue;
+ if ((rc = mdb_page_get(txn, pgno, &dp, &level)) != MDB_SUCCESS)
+ break;
+ if ((dp->mp_flags & Mask) == pflags && level <= 1)
+ dp->mp_flags ^= P_KEEP;
+ }
+ }
+ }
+
+ return rc;
+}
+
+static int mdb_page_flush(MDB_txn *txn, int keep);
+
+/** Spill pages from the dirty list back to disk.
+ * This is intended to prevent running into #MDB_TXN_FULL situations,
+ * but note that they may still occur in a few cases:
+ * 1) our estimate of the txn size could be too small. Currently this
+ * seems unlikely, except with a large number of #MDB_MULTIPLE items.
+ * 2) child txns may run out of space if their parents dirtied a
+ * lot of pages and never spilled them. TODO: we probably should do
+ * a preemptive spill during #mdb_txn_begin() of a child txn, if
+ * the parent's dirty_room is below a given threshold.
+ *
+ * Otherwise, if not using nested txns, it is expected that apps will
+ * not run into #MDB_TXN_FULL any more. The pages are flushed to disk
+ * the same way as for a txn commit, e.g. their P_DIRTY flag is cleared.
+ * If the txn never references them again, they can be left alone.
+ * If the txn only reads them, they can be used without any fuss.
+ * If the txn writes them again, they can be dirtied immediately without
+ * going thru all of the work of #mdb_page_touch(). Such references are
+ * handled by #mdb_page_unspill().
+ *
+ * Also note, we never spill DB root pages, nor pages of active cursors,
+ * because we'll need these back again soon anyway. And in nested txns,
+ * we can't spill a page in a child txn if it was already spilled in a
+ * parent txn. That would alter the parent txns' data even though
+ * the child hasn't committed yet, and we'd have no way to undo it if
+ * the child aborted.
+ *
+ * @param[in] m0 cursor A cursor handle identifying the transaction and
+ * database for which we are checking space.
+ * @param[in] key For a put operation, the key being stored.
+ * @param[in] data For a put operation, the data being stored.
+ * @return 0 on success, non-zero on failure.
+ */
+static int
+mdb_page_spill(MDB_cursor *m0, MDB_val *key, MDB_val *data)
+{
+ MDB_txn *txn = m0->mc_txn;
+ MDB_page *dp;
+ MDB_ID2L dl = txn->mt_u.dirty_list;
+ unsigned int i, j, need;
+ int rc;
+
+ if (m0->mc_flags & C_SUB)
+ return MDB_SUCCESS;
+
+ /* Estimate how much space this op will take */
+ i = m0->mc_db->md_depth;
+ /* Named DBs also dirty the main DB */
+ if (m0->mc_dbi > MAIN_DBI)
+ i += txn->mt_dbs[MAIN_DBI].md_depth;
+ /* For puts, roughly factor in the key+data size */
+ if (key)
+ i += (LEAFSIZE(key, data) + txn->mt_env->me_psize) / txn->mt_env->me_psize;
+ i += i; /* double it for good measure */
+ need = i;
+
+ if (txn->mt_dirty_room > i)
+ return MDB_SUCCESS;
+
+ if (!txn->mt_spill_pgs) {
+ txn->mt_spill_pgs = mdb_midl_alloc(MDB_IDL_UM_MAX);
+ if (!txn->mt_spill_pgs)
+ return ENOMEM;
+ } else {
+ /* purge deleted slots */
+ MDB_IDL sl = txn->mt_spill_pgs;
+ unsigned int num = sl[0];
+ j=0;
+ for (i=1; i<=num; i++) {
+ if (!(sl[i] & 1))
+ sl[++j] = sl[i];
+ }
+ sl[0] = j;
+ }
+
+ /* Preserve pages which may soon be dirtied again */
+ if ((rc = mdb_pages_xkeep(m0, P_DIRTY, 1)) != MDB_SUCCESS)
+ goto done;
+
+ /* Less aggressive spill - we originally spilled the entire dirty list,
+ * with a few exceptions for cursor pages and DB root pages. But this
+ * turns out to be a lot of wasted effort because in a large txn many
+ * of those pages will need to be used again. So now we spill only 1/8th
+ * of the dirty pages. Testing revealed this to be a good tradeoff,
+ * better than 1/2, 1/4, or 1/10.
+ */
+ if (need < MDB_IDL_UM_MAX / 8)
+ need = MDB_IDL_UM_MAX / 8;
+
+ /* Save the page IDs of all the pages we're flushing */
+ /* flush from the tail forward, this saves a lot of shifting later on. */
+ for (i=dl[0].mid; i && need; i--) {
+ MDB_ID pn = dl[i].mid << 1;
+ dp = dl[i].mptr;
+ if (dp->mp_flags & P_KEEP)
+ continue;
+ /* Can't spill twice, make sure it's not already in a parent's
+ * spill list.
+ */
+ if (txn->mt_parent) {
+ MDB_txn *tx2;
+ for (tx2 = txn->mt_parent; tx2; tx2 = tx2->mt_parent) {
+ if (tx2->mt_spill_pgs) {
+ j = mdb_midl_search(tx2->mt_spill_pgs, pn);
+ if (j <= tx2->mt_spill_pgs[0] && tx2->mt_spill_pgs[j] == pn) {
+ dp->mp_flags |= P_KEEP;
+ break;
+ }
+ }
+ }
+ if (tx2)
+ continue;
+ }
+ if ((rc = mdb_midl_append(&txn->mt_spill_pgs, pn)))
+ goto done;
+ need--;
+ }
+ mdb_midl_sort(txn->mt_spill_pgs);
+
+ /* Flush the spilled part of dirty list */
+ if ((rc = mdb_page_flush(txn, i)) != MDB_SUCCESS)
+ goto done;
+
+ /* Reset any dirty pages we kept that page_flush didn't see */
+ rc = mdb_pages_xkeep(m0, P_DIRTY|P_KEEP, i);
+
+done:
+ txn->mt_flags |= rc ? MDB_TXN_ERROR : MDB_TXN_SPILLS;
+ return rc;
+}
+
/** Find oldest txnid still referenced. Expects txn->mt_txnid > 0. */
static txnid_t
mdb_find_oldest(MDB_txn *txn)
{
int i;
txnid_t mr, oldest = txn->mt_txnid - 1;
- MDB_reader *r = txn->mt_env->me_txns->mti_readers;
- for (i = txn->mt_env->me_txns->mti_numreaders; --i >= 0; ) {
- if (r[i].mr_pid) {
- mr = r[i].mr_txnid;
- if (oldest > mr)
- oldest = mr;
+ if (txn->mt_env->me_txns) {
+ MDB_reader *r = txn->mt_env->me_txns->mti_readers;
+ for (i = txn->mt_env->me_txns->mti_numreaders; --i >= 0; ) {
+ if (r[i].mr_pid) {
+ mr = r[i].mr_txnid;
+ if (oldest > mr)
+ oldest = mr;
+ }
}
}
return oldest;
}
-/** Allocate pages for writing.
+/** Add a page to the txn's dirty list */
+static void
+mdb_page_dirty(MDB_txn *txn, MDB_page *mp)
+{
+ MDB_ID2 mid;
+ int (*insert)(MDB_ID2L, MDB_ID2 *);
+
+ if (txn->mt_env->me_flags & MDB_WRITEMAP) {
+ insert = mdb_mid2l_append;
+ } else {
+ insert = mdb_mid2l_insert;
+ }
+ mid.mid = mp->mp_pgno;
+ mid.mptr = mp;
+ insert(txn->mt_u.dirty_list, &mid);
+ txn->mt_dirty_room--;
+}
+
+/** Allocate page numbers and memory for writing. Maintain me_pglast,
+ * me_pghead and mt_next_pgno.
+ *
* If there are free pages available from older transactions, they
- * will be re-used first. Otherwise a new page will be allocated.
+ * are re-used first. Otherwise allocate a new page at mt_next_pgno.
+ * Do not modify the freedB, just merge freeDB records into me_pghead[]
+ * and move me_pglast to say which records were consumed. Only this
+ * function can create me_pghead and move me_pglast/mt_next_pgno.
* @param[in] mc cursor A cursor handle identifying the transaction and
* database for which we are allocating.
* @param[in] num the number of pages to allocate.
pgno_t pgno, *mop = env->me_pghead;
unsigned i, j, k, mop_len = mop ? mop[0] : 0;
MDB_page *np;
- MDB_ID2 mid;
txnid_t oldest = 0, last;
MDB_cursor_op op;
MDB_cursor m2;
- int (*insert)(MDB_ID2L, MDB_ID2 *);
*mp = NULL;
mdb_cursor_init(&m2, txn, FREE_DBI, NULL);
if (last) {
op = MDB_SET_RANGE;
- key.mv_data = &last; /* will loop up last+1 */
+ key.mv_data = &last; /* will look up last+1 */
key.mv_size = sizeof(last);
}
if (Paranoid && mc->mc_dbi == FREE_DBI)
mop = env->me_pghead;
}
env->me_pglast = last;
-#if MDB_DEBUG > 1
- DPRINTF("IDL read txn %zu root %zu num %u",
- last, txn->mt_dbs[FREE_DBI].md_root, i);
+#if (MDB_DEBUG) > 1
+ DPRINTF(("IDL read txn %"Z"u root %"Z"u num %u",
+ last, txn->mt_dbs[FREE_DBI].md_root, i));
for (k = i; k; k--)
- DPRINTF("IDL %zu", idl[k]);
+ DPRINTF(("IDL %"Z"u", idl[k]));
#endif
/* Merge in descending sorted order */
j = mop_len;
search_done:
if (env->me_flags & MDB_WRITEMAP) {
np = (MDB_page *)(env->me_map + env->me_psize * pgno);
- insert = mdb_mid2l_append;
} else {
if (!(np = mdb_page_malloc(txn, num)))
return ENOMEM;
- insert = mdb_mid2l_insert;
}
if (i) {
mop[0] = mop_len -= num;
} else {
txn->mt_next_pgno = pgno + num;
}
- mid.mid = np->mp_pgno = pgno;
- mid.mptr = np;
- insert(txn->mt_u.dirty_list, &mid);
- txn->mt_dirty_room--;
+ np->mp_pgno = pgno;
+ mdb_page_dirty(txn, np);
*mp = np;
return MDB_SUCCESS;
}
}
+/** Pull a page off the txn's spill list, if present.
+ * If a page being referenced was spilled to disk in this txn, bring
+ * it back and make it dirty/writable again.
+ * @param[in] txn the transaction handle.
+ * @param[in] mp the page being referenced.
+ * @param[out] ret the writable page, if any. ret is unchanged if
+ * mp wasn't spilled.
+ */
+static int
+mdb_page_unspill(MDB_txn *txn, MDB_page *mp, MDB_page **ret)
+{
+ MDB_env *env = txn->mt_env;
+ const MDB_txn *tx2;
+ unsigned x;
+ pgno_t pgno = mp->mp_pgno, pn = pgno << 1;
+
+ for (tx2 = txn; tx2; tx2=tx2->mt_parent) {
+ if (!tx2->mt_spill_pgs)
+ continue;
+ x = mdb_midl_search(tx2->mt_spill_pgs, pn);
+ if (x <= tx2->mt_spill_pgs[0] && tx2->mt_spill_pgs[x] == pn) {
+ MDB_page *np;
+ int num;
+ if (txn->mt_dirty_room == 0)
+ return MDB_TXN_FULL;
+ if (IS_OVERFLOW(mp))
+ num = mp->mp_pages;
+ else
+ num = 1;
+ if (env->me_flags & MDB_WRITEMAP) {
+ np = mp;
+ } else {
+ np = mdb_page_malloc(txn, num);
+ if (!np)
+ return ENOMEM;
+ if (num > 1)
+ memcpy(np, mp, num * env->me_psize);
+ else
+ mdb_page_copy(np, mp, env->me_psize);
+ }
+ if (tx2 == txn) {
+ /* If in current txn, this page is no longer spilled.
+ * If it happens to be the last page, truncate the spill list.
+ * Otherwise mark it as deleted by setting the LSB.
+ */
+ if (x == txn->mt_spill_pgs[0])
+ txn->mt_spill_pgs[0]--;
+ else
+ txn->mt_spill_pgs[x] |= 1;
+ } /* otherwise, if belonging to a parent txn, the
+ * page remains spilled until child commits
+ */
+
+ mdb_page_dirty(txn, np);
+ np->mp_flags |= P_DIRTY;
+ *ret = np;
+ break;
+ }
+ }
+ return MDB_SUCCESS;
+}
+
/** Touch a page: make it dirty and re-insert into tree with updated pgno.
* @param[in] mc cursor pointing to the page to be touched
* @return 0 on success, non-zero on failure.
MDB_page *mp = mc->mc_pg[mc->mc_top], *np;
MDB_txn *txn = mc->mc_txn;
MDB_cursor *m2, *m3;
- MDB_dbi dbi;
pgno_t pgno;
int rc;
if (!F_ISSET(mp->mp_flags, P_DIRTY)) {
+ if (txn->mt_flags & MDB_TXN_SPILLS) {
+ np = NULL;
+ rc = mdb_page_unspill(txn, mp, &np);
+ if (rc)
+ return rc;
+ if (np)
+ goto done;
+ }
if ((rc = mdb_midl_need(&txn->mt_free_pgs, 1)) ||
(rc = mdb_page_alloc(mc, 1, &np)))
return rc;
pgno = np->mp_pgno;
- DPRINTF("touched db %u page %zu -> %zu", mc->mc_dbi,mp->mp_pgno,pgno);
+ DPRINTF(("touched db %d page %"Z"u -> %"Z"u", DDBI(mc),
+ mp->mp_pgno, pgno));
assert(mp->mp_pgno != pgno);
mdb_midl_xappend(txn->mt_free_pgs, mp->mp_pgno);
/* Update the parent page, if any, to point to the new page */
np->mp_pgno = pgno;
np->mp_flags |= P_DIRTY;
+done:
/* Adjust cursors pointing to mp */
mc->mc_pg[mc->mc_top] = np;
- dbi = mc->mc_dbi;
+ m2 = txn->mt_cursors[mc->mc_dbi];
if (mc->mc_flags & C_SUB) {
- dbi--;
- for (m2 = txn->mt_cursors[dbi]; m2; m2=m2->mc_next) {
+ for (; m2; m2=m2->mc_next) {
m3 = &m2->mc_xcursor->mx_cursor;
if (m3->mc_snum < mc->mc_snum) continue;
if (m3->mc_pg[mc->mc_top] == mp)
m3->mc_pg[mc->mc_top] = np;
}
} else {
- for (m2 = txn->mt_cursors[dbi]; m2; m2=m2->mc_next) {
+ for (; m2; m2=m2->mc_next) {
if (m2->mc_snum < mc->mc_snum) continue;
if (m2->mc_pg[mc->mc_top] == mp) {
m2->mc_pg[mc->mc_top] = np;
}
mc = bk;
}
+ /* Only malloced cursors are permanently tracked. */
free(mc);
}
cursors[i] = NULL;
}
}
-#ifdef MDB_DEBUG_SKIP
+#if !(MDB_DEBUG)
#define mdb_txn_reset0(txn, act) mdb_txn_reset0(txn)
#endif
static void
mdb_txn_reset0(MDB_txn *txn, const char *act);
+#if !(MDB_PIDLOCK) /* Currently the same as defined(_WIN32) */
+enum Pidlock_op {
+ Pidset, Pidcheck
+};
+#else
+enum Pidlock_op {
+ Pidset = F_SETLK, Pidcheck = F_GETLK
+};
+#endif
+
+/** Set or check a pid lock. Set returns 0 on success.
+ * Check returns 0 if the process is certainly dead, nonzero if it may
+ * be alive (the lock exists or an error happened so we do not know).
+ *
+ * On Windows Pidset is a no-op, we merely check for the existence
+ * of the process with the given pid. On POSIX we use a single byte
+ * lock on the lockfile, set at an offset equal to the pid.
+ */
+static int
+mdb_reader_pid(MDB_env *env, enum Pidlock_op op, pid_t pid)
+{
+#if !(MDB_PIDLOCK) /* Currently the same as defined(_WIN32) */
+ int ret = 0;
+ HANDLE h;
+ if (op == Pidcheck) {
+ h = OpenProcess(env->me_pidquery, FALSE, pid);
+ /* No documented "no such process" code, but other program use this: */
+ if (!h)
+ return ErrCode() != ERROR_INVALID_PARAMETER;
+ /* A process exists until all handles to it close. Has it exited? */
+ ret = WaitForSingleObject(h, 0) != 0;
+ CloseHandle(h);
+ }
+ return ret;
+#else
+ for (;;) {
+ int rc;
+ struct flock lock_info;
+ memset(&lock_info, 0, sizeof(lock_info));
+ lock_info.l_type = F_WRLCK;
+ lock_info.l_whence = SEEK_SET;
+ lock_info.l_start = pid;
+ lock_info.l_len = 1;
+ if ((rc = fcntl(env->me_lfd, op, &lock_info)) == 0) {
+ if (op == F_GETLK && lock_info.l_type != F_UNLCK)
+ rc = -1;
+ } else if ((rc = ErrCode()) == EINTR) {
+ continue;
+ }
+ return rc;
+ }
+#endif
+}
+
/** Common code for #mdb_txn_begin() and #mdb_txn_renew().
* @param[in] txn the transaction handle to initialize
* @return 0 on success, non-zero on failure.
mdb_txn_renew0(MDB_txn *txn)
{
MDB_env *env = txn->mt_env;
- unsigned int i;
+ MDB_txninfo *ti = env->me_txns;
+ MDB_meta *meta;
+ unsigned int i, nr;
uint16_t x;
int rc, new_notls = 0;
txn->mt_dbxs = env->me_dbxs; /* mostly static anyway */
if (txn->mt_flags & MDB_TXN_RDONLY) {
- if (!env->me_txns) {
- i = mdb_env_pick_meta(env);
- txn->mt_txnid = env->me_metas[i]->mm_txnid;
+ if (!ti) {
+ meta = env->me_metas[ mdb_env_pick_meta(env) ];
+ txn->mt_txnid = meta->mm_txnid;
txn->mt_u.reader = NULL;
} else {
MDB_reader *r = (env->me_flags & MDB_NOTLS) ? txn->mt_u.reader :
pid_t pid = env->me_pid;
pthread_t tid = pthread_self();
+ if (!(env->me_flags & MDB_LIVE_READER)) {
+ rc = mdb_reader_pid(env, Pidset, pid);
+ if (rc) {
+ UNLOCK_MUTEX_R(env);
+ return rc;
+ }
+ env->me_flags |= MDB_LIVE_READER;
+ }
+
LOCK_MUTEX_R(env);
- for (i=0; i<env->me_txns->mti_numreaders; i++)
- if (env->me_txns->mti_readers[i].mr_pid == 0)
+ nr = ti->mti_numreaders;
+ for (i=0; i<nr; i++)
+ if (ti->mti_readers[i].mr_pid == 0)
break;
if (i == env->me_maxreaders) {
UNLOCK_MUTEX_R(env);
return MDB_READERS_FULL;
}
- env->me_txns->mti_readers[i].mr_pid = pid;
- env->me_txns->mti_readers[i].mr_tid = tid;
- if (i >= env->me_txns->mti_numreaders)
- env->me_txns->mti_numreaders = i+1;
+ ti->mti_readers[i].mr_pid = pid;
+ ti->mti_readers[i].mr_tid = tid;
+ if (i == nr)
+ ti->mti_numreaders = ++nr;
/* Save numreaders for un-mutexed mdb_env_close() */
- env->me_numreaders = env->me_txns->mti_numreaders;
+ env->me_numreaders = nr;
UNLOCK_MUTEX_R(env);
- r = &env->me_txns->mti_readers[i];
+
+ r = &ti->mti_readers[i];
new_notls = (env->me_flags & MDB_NOTLS);
if (!new_notls && (rc=pthread_setspecific(env->me_txkey, r))) {
r->mr_pid = 0;
return rc;
}
}
- txn->mt_txnid = r->mr_txnid = env->me_txns->mti_txnid;
+ txn->mt_txnid = r->mr_txnid = ti->mti_txnid;
txn->mt_u.reader = r;
+ meta = env->me_metas[txn->mt_txnid & 1];
}
- txn->mt_toggle = txn->mt_txnid & 1;
} else {
- LOCK_MUTEX_W(env);
+ if (ti) {
+ LOCK_MUTEX_W(env);
- txn->mt_txnid = env->me_txns->mti_txnid;
- txn->mt_toggle = txn->mt_txnid & 1;
+ txn->mt_txnid = ti->mti_txnid;
+ meta = env->me_metas[txn->mt_txnid & 1];
+ } else {
+ meta = env->me_metas[ mdb_env_pick_meta(env) ];
+ txn->mt_txnid = meta->mm_txnid;
+ }
txn->mt_txnid++;
#if MDB_DEBUG
if (txn->mt_txnid == mdb_debug_start)
txn->mt_u.dirty_list[0].mid = 0;
txn->mt_free_pgs = env->me_free_pgs;
txn->mt_free_pgs[0] = 0;
+ txn->mt_spill_pgs = NULL;
env->me_txn = txn;
}
/* Copy the DB info and flags */
- memcpy(txn->mt_dbs, env->me_metas[txn->mt_toggle]->mm_dbs, 2 * sizeof(MDB_db));
+ memcpy(txn->mt_dbs, meta->mm_dbs, 2 * sizeof(MDB_db));
/* Moved to here to avoid a data race in read TXNs */
- txn->mt_next_pgno = env->me_metas[txn->mt_toggle]->mm_last_pg+1;
+ txn->mt_next_pgno = meta->mm_last_pg+1;
for (i=2; i<txn->mt_numdbs; i++) {
x = env->me_dbflags[i];
rc = mdb_txn_renew0(txn);
if (rc == MDB_SUCCESS) {
- DPRINTF("renew txn %zu%c %p on mdbenv %p, root page %zu",
+ DPRINTF(("renew txn %"Z"u%c %p on mdbenv %p, root page %"Z"u",
txn->mt_txnid, (txn->mt_flags & MDB_TXN_RDONLY) ? 'r' : 'w',
- (void *)txn, (void *)txn->mt_env, txn->mt_dbs[MAIN_DBI].md_root);
+ (void *)txn, (void *)txn->mt_env, txn->mt_dbs[MAIN_DBI].md_root));
}
return rc;
}
if (parent) {
/* Nested transactions: Max 1 child, write txns only, no writemap */
if (parent->mt_child ||
- (flags & MDB_RDONLY) || (parent->mt_flags & MDB_TXN_RDONLY) ||
+ (flags & MDB_RDONLY) ||
+ (parent->mt_flags & (MDB_TXN_RDONLY|MDB_TXN_ERROR)) ||
(env->me_flags & MDB_WRITEMAP))
{
- return EINVAL;
+ return (parent->mt_flags & MDB_TXN_RDONLY) ? EINVAL : MDB_BAD_TXN;
}
tsize = sizeof(MDB_ntxn);
}
size += env->me_maxdbs * sizeof(MDB_cursor *);
if ((txn = calloc(1, size)) == NULL) {
- DPRINTF("calloc: %s", strerror(ErrCode()));
+ DPRINTF(("calloc: %s", strerror(ErrCode())));
return ENOMEM;
}
txn->mt_dbs = (MDB_db *) ((char *)txn + tsize);
return ENOMEM;
}
txn->mt_txnid = parent->mt_txnid;
- txn->mt_toggle = parent->mt_toggle;
txn->mt_dirty_room = parent->mt_dirty_room;
txn->mt_u.dirty_list[0].mid = 0;
+ txn->mt_spill_pgs = NULL;
txn->mt_next_pgno = parent->mt_next_pgno;
parent->mt_child = txn;
txn->mt_parent = parent;
free(txn);
else {
*ret = txn;
- DPRINTF("begin txn %zu%c %p on mdbenv %p, root page %zu",
+ DPRINTF(("begin txn %"Z"u%c %p on mdbenv %p, root page %"Z"u",
txn->mt_txnid, (txn->mt_flags & MDB_TXN_RDONLY) ? 'r' : 'w',
- (void *) txn, (void *) env, txn->mt_dbs[MAIN_DBI].md_root);
+ (void *) txn, (void *) env, txn->mt_dbs[MAIN_DBI].md_root));
}
return rc;
}
+MDB_env *
+mdb_txn_env(MDB_txn *txn)
+{
+ if(!txn) return NULL;
+ return txn->mt_env;
+}
+
/** Export or close DBI handles opened in this txn. */
static void
mdb_dbis_update(MDB_txn *txn, int keep)
/** Common code for #mdb_txn_reset() and #mdb_txn_abort().
* May be called twice for readonly txns: First reset it, then abort.
* @param[in] txn the transaction handle to reset
+ * @param[in] act why the transaction is being reset
*/
static void
mdb_txn_reset0(MDB_txn *txn, const char *act)
/* Close any DBI handles opened in this txn */
mdb_dbis_update(txn, 0);
- DPRINTF("%s txn %zu%c %p on mdbenv %p, root page %zu",
+ DPRINTF(("%s txn %"Z"u%c %p on mdbenv %p, root page %"Z"u",
act, txn->mt_txnid, (txn->mt_flags & MDB_TXN_RDONLY) ? 'r' : 'w',
- (void *) txn, (void *)env, txn->mt_dbs[MAIN_DBI].md_root);
+ (void *) txn, (void *)env, txn->mt_dbs[MAIN_DBI].md_root));
if (F_ISSET(txn->mt_flags, MDB_TXN_RDONLY)) {
if (txn->mt_u.reader) {
txn->mt_parent->mt_child = NULL;
env->me_pgstate = ((MDB_ntxn *)txn)->mnt_pgstate;
mdb_midl_free(txn->mt_free_pgs);
+ mdb_midl_free(txn->mt_spill_pgs);
free(txn->mt_u.dirty_list);
return;
}
env->me_txn = NULL;
/* The writer mutex was locked in mdb_txn_begin. */
- UNLOCK_MUTEX_W(env);
+ if (env->me_txns)
+ UNLOCK_MUTEX_W(env);
}
}
if (env->me_pghead) {
/* Make sure first page of freeDB is touched and on freelist */
- rc = mdb_page_search(&mc, NULL, MDB_PS_MODIFY);
+ rc = mdb_page_search(&mc, NULL, MDB_PS_FIRST|MDB_PS_MODIFY);
if (rc && rc != MDB_NOTFOUND)
return rc;
}
if (freecnt < txn->mt_free_pgs[0]) {
if (!freecnt) {
/* Make sure last page of freeDB is touched and on freelist */
- key.mv_size = MDB_MAXKEYSIZE+1;
- key.mv_data = NULL;
- rc = mdb_page_search(&mc, &key, MDB_PS_MODIFY);
+ rc = mdb_page_search(&mc, NULL, MDB_PS_LAST|MDB_PS_MODIFY);
if (rc && rc != MDB_NOTFOUND)
return rc;
}
} while (freecnt < free_pgs[0]);
mdb_midl_sort(free_pgs);
memcpy(data.mv_data, free_pgs, data.mv_size);
-#if MDB_DEBUG > 1
+#if (MDB_DEBUG) > 1
{
unsigned int i = free_pgs[0];
- DPRINTF("IDL write txn %zu root %zu num %u",
- txn->mt_txnid, txn->mt_dbs[FREE_DBI].md_root, i);
+ DPRINTF(("IDL write txn %"Z"u root %"Z"u num %u",
+ txn->mt_txnid, txn->mt_dbs[FREE_DBI].md_root, i));
for (; i; i--)
- DPRINTF("IDL %zu", free_pgs[i]);
+ DPRINTF(("IDL %"Z"u", free_pgs[i]));
}
#endif
continue;
total_room += head_room;
}
- /* Fill in the reserved, touched me_pghead records. Avoid write ops
- * so they cannot rearrange anything, just read the destinations.
- */
+ /* Fill in the reserved me_pghead records */
rc = MDB_SUCCESS;
if (mop_len) {
MDB_val key, data;
- mop += mop_len + 1;
+ mop += mop_len;
rc = mdb_cursor_first(&mc, &key, &data);
for (; !rc; rc = mdb_cursor_next(&mc, &key, &data, MDB_NEXT)) {
- MDB_IDL dest = data.mv_data;
+ 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;
- assert(len >= 0 && *(txnid_t*)key.mv_data <= env->me_pglast);
- if (len > mop_len)
+ assert(len >= 0 && id <= env->me_pglast);
+ key.mv_data = &id;
+ if (len > mop_len) {
len = mop_len;
- *dest++ = len;
- memcpy(dest, mop -= len, len * sizeof(MDB_ID));
- if (! (mop_len -= len))
+ data.mv_size = (len + 1) * sizeof(MDB_ID);
+ flags = 0;
+ }
+ data.mv_data = mop -= len;
+ save = mop[0];
+ mop[0] = len;
+ rc = mdb_cursor_put(&mc, &key, &data, flags);
+ mop[0] = save;
+ if (rc || !(mop_len -= len))
break;
}
}
return rc;
}
-/** Flush dirty pages to the map, after clearing their dirty flag.
+/** Flush (some) dirty pages to the map, after clearing their dirty flag.
+ * @param[in] txn the transaction that's being committed
+ * @param[in] keep number of initial pages in dirty_list to keep dirty.
+ * @return 0 on success, non-zero on failure.
*/
static int
-mdb_page_flush(MDB_txn *txn)
+mdb_page_flush(MDB_txn *txn, int keep)
{
MDB_env *env = txn->mt_env;
MDB_ID2L dl = txn->mt_u.dirty_list;
- unsigned psize = env->me_psize;
+ unsigned psize = env->me_psize, j;
int i, pagecount = dl[0].mid, rc;
size_t size = 0, pos = 0;
pgno_t pgno = 0;
int n = 0;
#endif
+ j = i = keep;
+
if (env->me_flags & MDB_WRITEMAP) {
/* Clear dirty flags */
- for (i = pagecount; i; i--) {
+ while (++i <= pagecount) {
dp = dl[i].mptr;
+ /* Don't flush this page yet */
+ if (dp->mp_flags & P_KEEP) {
+ dp->mp_flags ^= P_KEEP;
+ dl[++j] = dl[i];
+ continue;
+ }
dp->mp_flags &= ~P_DIRTY;
}
- dl[0].mid = 0;
- return MDB_SUCCESS;
+ goto done;
}
/* Write the pages */
- for (i = 1;; i++) {
- if (i <= pagecount) {
+ for (;;) {
+ if (++i <= pagecount) {
dp = dl[i].mptr;
+ /* Don't flush this page yet */
+ if (dp->mp_flags & P_KEEP) {
+ dp->mp_flags ^= P_KEEP;
+ dl[i].mid = 0;
+ continue;
+ }
pgno = dl[i].mid;
/* clear dirty flag */
dp->mp_flags &= ~P_DIRTY;
* the write offset, to at least save the overhead of a Seek
* system call.
*/
- DPRINTF("committing page %zu", pgno);
+ DPRINTF(("committing page %"Z"u", pgno));
memset(&ov, 0, sizeof(ov));
ov.Offset = pos & 0xffffffff;
ov.OffsetHigh = pos >> 16 >> 16;
if (!WriteFile(env->me_fd, dp, size, NULL, &ov)) {
rc = ErrCode();
- DPRINTF("WriteFile: %d", rc);
+ DPRINTF(("WriteFile: %d", rc));
return rc;
}
#else
} else {
if (lseek(env->me_fd, wpos, SEEK_SET) == -1) {
rc = ErrCode();
- DPRINTF("lseek: %s", strerror(rc));
+ DPRINTF(("lseek: %s", strerror(rc)));
return rc;
}
wres = writev(env->me_fd, iov, n);
if (wres != wsize) {
if (wres < 0) {
rc = ErrCode();
- DPRINTF("Write error: %s", strerror(rc));
+ DPRINTF(("Write error: %s", strerror(rc)));
} else {
rc = EIO; /* TODO: Use which error code? */
DPUTS("short write, filesystem full?");
wpos = pos;
wsize = 0;
}
- DPRINTF("committing page %zu", pgno);
+ DPRINTF(("committing page %"Z"u", pgno));
next_pos = pos + size;
iov[n].iov_len = size;
iov[n].iov_base = (char *)dp;
#endif /* _WIN32 */
}
- mdb_dlist_free(txn);
+ for (i = keep; ++i <= pagecount; ) {
+ dp = dl[i].mptr;
+ /* This is a page we skipped above */
+ if (!dl[i].mid) {
+ dl[++j] = dl[i];
+ dl[j].mid = dp->mp_pgno;
+ continue;
+ }
+ mdb_dpage_free(env, dp);
+ }
+done:
+ i--;
+ txn->mt_dirty_room += i - j;
+ dl[0].mid = j;
return MDB_SUCCESS;
}
DPUTS("error flag is set, can't commit");
if (txn->mt_parent)
txn->mt_parent->mt_flags |= MDB_TXN_ERROR;
- rc = EINVAL;
+ rc = MDB_BAD_TXN;
goto fail;
}
if (txn->mt_parent) {
MDB_txn *parent = txn->mt_parent;
- unsigned x, y, len;
MDB_ID2L dst, src;
+ MDB_IDL pspill;
+ unsigned x, y, len, ps_len;
/* Append our free list to parent's */
rc = mdb_midl_append_list(&parent->mt_free_pgs, txn->mt_free_pgs);
if (rc)
goto fail;
mdb_midl_free(txn->mt_free_pgs);
+ /* Failures after this must either undo the changes
+ * to the parent or set MDB_TXN_ERROR in the parent.
+ */
parent->mt_next_pgno = txn->mt_next_pgno;
parent->mt_flags = txn->mt_flags;
/* Update parent's DB table. */
memcpy(parent->mt_dbs, txn->mt_dbs, txn->mt_numdbs * sizeof(MDB_db));
- txn->mt_parent->mt_numdbs = txn->mt_numdbs;
- txn->mt_parent->mt_dbflags[0] = txn->mt_dbflags[0];
- txn->mt_parent->mt_dbflags[1] = txn->mt_dbflags[1];
+ parent->mt_numdbs = txn->mt_numdbs;
+ parent->mt_dbflags[0] = txn->mt_dbflags[0];
+ parent->mt_dbflags[1] = txn->mt_dbflags[1];
for (i=2; i<txn->mt_numdbs; i++) {
/* preserve parent's DB_NEW status */
- x = txn->mt_parent->mt_dbflags[i] & DB_NEW;
- txn->mt_parent->mt_dbflags[i] = txn->mt_dbflags[i] | x;
+ x = parent->mt_dbflags[i] & DB_NEW;
+ parent->mt_dbflags[i] = txn->mt_dbflags[i] | x;
}
- dst = txn->mt_parent->mt_u.dirty_list;
+ dst = parent->mt_u.dirty_list;
src = txn->mt_u.dirty_list;
+ /* Remove anything in our dirty list from parent's spill list */
+ if ((pspill = parent->mt_spill_pgs) && (ps_len = pspill[0])) {
+ x = y = ps_len;
+ pspill[0] = (pgno_t)-1;
+ /* Mark our dirty pages as deleted in parent spill list */
+ for (i=0, len=src[0].mid; ++i <= len; ) {
+ MDB_ID pn = src[i].mid << 1;
+ while (pn > pspill[x])
+ x--;
+ if (pn == pspill[x]) {
+ pspill[x] = 1;
+ y = --x;
+ }
+ }
+ /* Squash deleted pagenums if we deleted any */
+ for (x=y; ++x <= ps_len; )
+ if (!(pspill[x] & 1))
+ pspill[++y] = pspill[x];
+ pspill[0] = y;
+ }
+
/* Find len = length of merging our dirty list with parent's */
x = dst[0].mid;
dst[0].mid = 0; /* simplify loops */
dst[0].mid = len;
free(txn->mt_u.dirty_list);
parent->mt_dirty_room = txn->mt_dirty_room;
+ if (txn->mt_spill_pgs) {
+ if (parent->mt_spill_pgs) {
+ /* TODO: Prevent failure here, so parent does not fail */
+ rc = mdb_midl_append_list(&parent->mt_spill_pgs, txn->mt_spill_pgs);
+ if (rc)
+ parent->mt_flags |= MDB_TXN_ERROR;
+ mdb_midl_free(txn->mt_spill_pgs);
+ mdb_midl_sort(parent->mt_spill_pgs);
+ } else {
+ parent->mt_spill_pgs = txn->mt_spill_pgs;
+ }
+ }
- txn->mt_parent->mt_child = NULL;
+ parent->mt_child = NULL;
mdb_midl_free(((MDB_ntxn *)txn)->mnt_pgstate.mf_pghead);
free(txn);
- return MDB_SUCCESS;
+ return rc;
}
if (txn != env->me_txn) {
mdb_cursors_close(txn, 0);
- if (!txn->mt_u.dirty_list[0].mid && !(txn->mt_flags & MDB_TXN_DIRTY))
+ if (!txn->mt_u.dirty_list[0].mid &&
+ !(txn->mt_flags & (MDB_TXN_DIRTY|MDB_TXN_SPILLS)))
goto done;
- DPRINTF("committing txn %zu %p on mdbenv %p, root page %zu",
- txn->mt_txnid, (void *)txn, (void *)env, txn->mt_dbs[MAIN_DBI].md_root);
+ DPRINTF(("committing txn %"Z"u %p on mdbenv %p, root page %"Z"u",
+ txn->mt_txnid, (void*)txn, (void*)env, txn->mt_dbs[MAIN_DBI].md_root));
/* Update DB root pointers */
if (txn->mt_numdbs > 2) {
if (mdb_midl_shrink(&txn->mt_free_pgs))
env->me_free_pgs = txn->mt_free_pgs;
-#if MDB_DEBUG > 2
+#if (MDB_DEBUG) > 2
mdb_audit(txn);
#endif
- if ((rc = mdb_page_flush(txn)) ||
+ if ((rc = mdb_page_flush(txn, 0)) ||
(rc = mdb_env_sync(env, 0)) ||
(rc = mdb_env_write_meta(txn)))
goto fail;
env->me_txn = NULL;
mdb_dbis_update(txn, 1);
- UNLOCK_MUTEX_W(env);
+ if (env->me_txns)
+ UNLOCK_MUTEX_W(env);
free(txn);
return MDB_SUCCESS;
memset(&ov, 0, sizeof(ov));
ov.Offset = off;
rc = ReadFile(env->me_fd,&pbuf,MDB_PAGESIZE,&len,&ov) ? (int)len : -1;
+ if (rc == -1 && ErrCode() == ERROR_HANDLE_EOF)
+ rc = 0;
#else
rc = pread(env->me_fd, &pbuf, MDB_PAGESIZE, off);
#endif
if (rc == 0 && off == 0)
return ENOENT;
rc = rc < 0 ? (int) ErrCode() : MDB_INVALID;
- DPRINTF("read: %s", mdb_strerror(rc));
+ DPRINTF(("read: %s", mdb_strerror(rc)));
return rc;
}
p = (MDB_page *)&pbuf;
if (!F_ISSET(p->mp_flags, P_META)) {
- DPRINTF("page %zu not a meta page", p->mp_pgno);
+ DPRINTF(("page %"Z"u not a meta page", p->mp_pgno));
return MDB_INVALID;
}
return MDB_INVALID;
}
- if (m->mm_version != MDB_VERSION) {
- DPRINTF("database is version %u, expected version %u",
- m->mm_version, MDB_VERSION);
+ if (m->mm_version != MDB_DATA_VERSION) {
+ DPRINTF(("database is version %u, expected version %u",
+ m->mm_version, MDB_DATA_VERSION));
return MDB_VERSION_MISMATCH;
}
MDB_page *p, *q;
int rc;
unsigned int psize;
+#ifdef _WIN32
+ DWORD len;
+ OVERLAPPED ov;
+ memset(&ov, 0, sizeof(ov));
+#define DO_PWRITE(rc, fd, ptr, size, len, pos) do { \
+ ov.Offset = pos; \
+ rc = WriteFile(fd, ptr, size, &len, &ov); } while(0)
+#else
+ int len;
+#define DO_PWRITE(rc, fd, ptr, size, len, pos) do { \
+ len = pwrite(fd, ptr, size, pos); \
+ rc = (len >= 0); } while(0)
+#endif
DPUTS("writing new meta page");
- GET_PAGESIZE(psize);
+ psize = env->me_psize;
meta->mm_magic = MDB_MAGIC;
- meta->mm_version = MDB_VERSION;
+ meta->mm_version = MDB_DATA_VERSION;
meta->mm_mapsize = env->me_mapsize;
meta->mm_psize = psize;
meta->mm_last_pg = 1;
q->mp_flags = P_META;
*(MDB_meta *)METADATA(q) = *meta;
-#ifdef _WIN32
- {
- DWORD len;
- OVERLAPPED ov;
- memset(&ov, 0, sizeof(ov));
- rc = WriteFile(env->me_fd, p, psize * 2, &len, &ov);
- rc = rc ? (len == psize * 2 ? MDB_SUCCESS : EIO) : ErrCode();
- }
-#else
- rc = pwrite(env->me_fd, p, psize * 2, 0);
- rc = (rc == (int)psize * 2) ? MDB_SUCCESS : rc < 0 ? ErrCode() : EIO;
-#endif
+ DO_PWRITE(rc, env->me_fd, p, psize * 2, len, 0);
+ if (!rc)
+ rc = ErrCode();
+ else if ((unsigned) len == psize * 2)
+ rc = MDB_SUCCESS;
+ else
+ rc = ENOSPC;
free(p);
return rc;
}
assert(txn != NULL);
assert(txn->mt_env != NULL);
- toggle = !txn->mt_toggle;
- DPRINTF("writing meta page %d for root page %zu",
- toggle, txn->mt_dbs[MAIN_DBI].md_root);
+ toggle = txn->mt_txnid & 1;
+ DPRINTF(("writing meta page %d for root page %"Z"u",
+ toggle, txn->mt_dbs[MAIN_DBI].md_root));
env = txn->mt_env;
mp = env->me_metas[toggle];
WriteFile(env->me_fd, ptr, len, NULL, &ov);
#else
r2 = pwrite(env->me_fd, ptr, len, off);
+ (void)r2; /* Silence warnings. We don't care about pwrite's return value */
#endif
fail:
env->me_flags |= MDB_FATAL_ERROR;
* readers will get consistent data regardless of how fresh or
* how stale their view of these values is.
*/
- env->me_txns->mti_txnid = txn->mt_txnid;
+ if (env->me_txns)
+ env->me_txns->mti_txnid = txn->mt_txnid;
return MDB_SUCCESS;
}
return MDB_SUCCESS;
}
+static int
+mdb_env_map(MDB_env *env, void *addr, int newsize)
+{
+ MDB_page *p;
+ unsigned int flags = env->me_flags;
+#ifdef _WIN32
+ int rc;
+ HANDLE mh;
+ LONG sizelo, sizehi;
+ sizelo = env->me_mapsize & 0xffffffff;
+ sizehi = env->me_mapsize >> 16 >> 16; /* only needed on Win64 */
+
+ /* Windows won't create mappings for zero length files.
+ * Just allocate the maxsize right now.
+ */
+ if (newsize) {
+ if (SetFilePointer(env->me_fd, sizelo, &sizehi, 0) != (DWORD)sizelo
+ || !SetEndOfFile(env->me_fd)
+ || SetFilePointer(env->me_fd, 0, NULL, 0) != 0)
+ return ErrCode();
+ }
+ mh = CreateFileMapping(env->me_fd, NULL, flags & MDB_WRITEMAP ?
+ PAGE_READWRITE : PAGE_READONLY,
+ sizehi, sizelo, NULL);
+ if (!mh)
+ return ErrCode();
+ env->me_map = MapViewOfFileEx(mh, flags & MDB_WRITEMAP ?
+ FILE_MAP_WRITE : FILE_MAP_READ,
+ 0, 0, env->me_mapsize, addr);
+ rc = env->me_map ? 0 : ErrCode();
+ CloseHandle(mh);
+ if (rc)
+ return rc;
+#else
+ int prot = PROT_READ;
+ if (flags & MDB_WRITEMAP) {
+ prot |= PROT_WRITE;
+ if (ftruncate(env->me_fd, env->me_mapsize) < 0)
+ return ErrCode();
+ }
+ env->me_map = mmap(addr, env->me_mapsize, prot, MAP_SHARED,
+ env->me_fd, 0);
+ if (env->me_map == MAP_FAILED) {
+ env->me_map = NULL;
+ return ErrCode();
+ }
+ /* Turn off readahead. It's harmful when the DB is larger than RAM. */
+#ifdef MADV_RANDOM
+ madvise(env->me_map, env->me_mapsize, MADV_RANDOM);
+#else
+#ifdef POSIX_MADV_RANDOM
+ posix_madvise(env->me_map, env->me_mapsize, POSIX_MADV_RANDOM);
+#endif /* POSIX_MADV_RANDOM */
+#endif /* MADV_RANDOM */
+#endif /* _WIN32 */
+
+ /* Can happen because the address argument to mmap() is just a
+ * hint. mmap() can pick another, e.g. if the range is in use.
+ * The MAP_FIXED flag would prevent that, but then mmap could
+ * instead unmap existing pages to make room for the new map.
+ */
+ if (addr && env->me_map != addr)
+ return EBUSY; /* TODO: Make a new MDB_* error code? */
+
+ p = (MDB_page *)env->me_map;
+ env->me_metas[0] = METADATA(p);
+ env->me_metas[1] = (MDB_meta *)((char *)env->me_metas[0] + env->me_psize);
+
+ return MDB_SUCCESS;
+}
+
int
mdb_env_set_mapsize(MDB_env *env, size_t size)
{
- if (env->me_map)
- return EINVAL;
+ /* If env is already open, caller is responsible for making
+ * sure there are no active txns.
+ */
+ if (env->me_map) {
+ int rc;
+ void *old;
+ if (env->me_txn)
+ return EINVAL;
+ if (!size)
+ size = env->me_metas[mdb_env_pick_meta(env)]->mm_mapsize;
+ else if (size < env->me_mapsize) {
+ /* If the configured size is smaller, make sure it's
+ * still big enough. Silently round up to minimum if not.
+ */
+ size_t minsize = (env->me_metas[mdb_env_pick_meta(env)]->mm_last_pg + 1) * env->me_psize;
+ if (size < minsize)
+ size = minsize;
+ }
+ munmap(env->me_map, env->me_mapsize);
+ env->me_mapsize = size;
+ old = (env->me_flags & MDB_FIXEDMAP) ? env->me_map : NULL;
+ rc = mdb_env_map(env, old, 1);
+ if (rc)
+ return rc;
+ }
env->me_mapsize = size;
if (env->me_psize)
env->me_maxpg = env->me_mapsize / env->me_psize;
mdb_env_open2(MDB_env *env)
{
unsigned int flags = env->me_flags;
- int i, newenv = 0;
+ int i, newenv = 0, rc;
MDB_meta meta;
- MDB_page *p;
-#ifndef _WIN32
- int prot;
-#endif
+
+#ifdef _WIN32
+ /* See if we should use QueryLimited */
+ rc = GetVersion();
+ if ((rc & 0xff) > 5)
+ env->me_pidquery = MDB_PROCESS_QUERY_LIMITED_INFORMATION;
+ else
+ env->me_pidquery = PROCESS_QUERY_INFORMATION;
+#endif /* _WIN32 */
memset(&meta, 0, sizeof(meta));
return i;
DPUTS("new mdbenv");
newenv = 1;
+ GET_PAGESIZE(env->me_psize);
+ } else {
+ env->me_psize = meta.mm_psize;
}
/* Was a mapsize configured? */
env->me_mapsize = minsize;
}
-#ifdef _WIN32
- {
- int rc;
- HANDLE mh;
- LONG sizelo, sizehi;
- sizelo = env->me_mapsize & 0xffffffff;
- sizehi = env->me_mapsize >> 16 >> 16; /* only needed on Win64 */
- /* Windows won't create mappings for zero length files.
- * Just allocate the maxsize right now.
- */
- if (newenv) {
- if (SetFilePointer(env->me_fd, sizelo, &sizehi, 0) != (DWORD)sizelo
- || !SetEndOfFile(env->me_fd)
- || SetFilePointer(env->me_fd, 0, NULL, 0) != 0)
- return ErrCode();
- }
- mh = CreateFileMapping(env->me_fd, NULL, flags & MDB_WRITEMAP ?
- PAGE_READWRITE : PAGE_READONLY,
- sizehi, sizelo, NULL);
- if (!mh)
- return ErrCode();
- env->me_map = MapViewOfFileEx(mh, flags & MDB_WRITEMAP ?
- FILE_MAP_WRITE : FILE_MAP_READ,
- 0, 0, env->me_mapsize, meta.mm_address);
- rc = env->me_map ? 0 : ErrCode();
- CloseHandle(mh);
- if (rc)
- return rc;
- }
-#else
- i = MAP_SHARED;
- prot = PROT_READ;
- if (flags & MDB_WRITEMAP) {
- prot |= PROT_WRITE;
- if (ftruncate(env->me_fd, env->me_mapsize) < 0)
- return ErrCode();
- }
- env->me_map = mmap(meta.mm_address, env->me_mapsize, prot, i,
- env->me_fd, 0);
- if (env->me_map == MAP_FAILED) {
- env->me_map = NULL;
- return ErrCode();
- }
- /* Turn off readahead. It's harmful when the DB is larger than RAM. */
-#ifdef MADV_RANDOM
- madvise(env->me_map, env->me_mapsize, MADV_RANDOM);
-#else
-#ifdef POSIX_MADV_RANDOM
- posix_madvise(env->me_map, env->me_mapsize, POSIX_MADV_RANDOM);
-#endif /* POSIX_MADV_RANDOM */
-#endif /* MADV_RANDOM */
-#endif /* _WIN32 */
+ rc = mdb_env_map(env, meta.mm_address, newenv);
+ if (rc)
+ return rc;
if (newenv) {
if (flags & MDB_FIXEDMAP)
if (i != MDB_SUCCESS) {
return i;
}
- } else if (meta.mm_address && env->me_map != meta.mm_address) {
- /* Can happen because the address argument to mmap() is just a
- * hint. mmap() can pick another, e.g. if the range is in use.
- * The MAP_FIXED flag would prevent that, but then mmap could
- * instead unmap existing pages to make room for the new map.
- */
- return EBUSY; /* TODO: Make a new MDB_* error code? */
}
- env->me_psize = meta.mm_psize;
env->me_maxfree_1pg = (env->me_psize - PAGEHDRSZ) / sizeof(pgno_t) - 1;
env->me_nodemax = (env->me_psize - PAGEHDRSZ) / MDB_MINKEYS;
env->me_maxpg = env->me_mapsize / env->me_psize;
-
- p = (MDB_page *)env->me_map;
- env->me_metas[0] = METADATA(p);
- env->me_metas[1] = (MDB_meta *)((char *)env->me_metas[0] + meta.mm_psize);
-
#if MDB_DEBUG
{
int toggle = mdb_env_pick_meta(env);
MDB_db *db = &env->me_metas[toggle]->mm_dbs[MAIN_DBI];
- DPRINTF("opened database version %u, pagesize %u",
- env->me_metas[0]->mm_version, env->me_psize);
- DPRINTF("using meta page %d", toggle);
- DPRINTF("depth: %u", db->md_depth);
- DPRINTF("entries: %zu", db->md_entries);
- DPRINTF("branch pages: %zu", db->md_branch_pages);
- DPRINTF("leaf pages: %zu", db->md_leaf_pages);
- DPRINTF("overflow pages: %zu", db->md_overflow_pages);
- DPRINTF("root: %zu", db->md_root);
+ DPRINTF(("opened database version %u, pagesize %u",
+ env->me_metas[0]->mm_version, env->me_psize));
+ DPRINTF(("using meta page %d", toggle));
+ DPRINTF(("depth: %u", db->md_depth));
+ DPRINTF(("entries: %"Z"u", db->md_entries));
+ DPRINTF(("branch pages: %"Z"u", db->md_branch_pages));
+ DPRINTF(("leaf pages: %"Z"u", db->md_leaf_pages));
+ DPRINTF(("overflow pages: %"Z"u", db->md_overflow_pages));
+ DPRINTF(("root: %"Z"u", db->md_root));
}
#endif
#define MDB_HASH_INIT ((mdb_hash_t)0xcbf29ce484222325ULL)
/** perform a 64 bit Fowler/Noll/Vo FNV-1a hash on a buffer
- * @param[in] str string to hash
+ * @param[in] val value to hash
* @param[in] hval initial value for hash
* @return 64 bit hash
*
return hval;
}
-/** Hash the string and output the hash in hex.
+/** Hash the string and output the encoded hash.
+ * This uses modified RFC1924 Ascii85 encoding to accommodate systems with
+ * very short name limits. We don't care about the encoding being reversible,
+ * we just want to preserve as many bits of the input as possible in a
+ * small printable string.
* @param[in] str string to hash
- * @param[out] hexbuf an array of 17 chars to hold the hash
+ * @param[out] encbuf an array of 11 chars to hold the hash
*/
+static const char mdb_a85[]= "0123456789ABCDEFGHIJKLMNOPQRSTUVWXYZabcdefghijklmnopqrstuvwxyz!#$%&()*+-;<=>?@^_`{|}~";
+
static void
-mdb_hash_hex(MDB_val *val, char *hexbuf)
+mdb_pack85(unsigned long l, char *out)
{
int i;
- mdb_hash_t h = mdb_hash_val(val, MDB_HASH_INIT);
- for (i=0; i<8; i++) {
- hexbuf += sprintf(hexbuf, "%02x", (unsigned int)h & 0xff);
- h >>= 8;
+
+ for (i=0; i<5; i++) {
+ *out++ = mdb_a85[l % 85];
+ l /= 85;
}
}
+
+static void
+mdb_hash_enc(MDB_val *val, char *encbuf)
+{
+ mdb_hash_t h = mdb_hash_val(val, MDB_HASH_INIT);
+
+ mdb_pack85(h, encbuf);
+ mdb_pack85(h>>32, encbuf+5);
+ encbuf[10] = '\0';
+}
#endif
/** Open and/or initialize the lock region for the environment.
DWORD nlow;
} idbuf;
MDB_val val;
- char hexbuf[17];
+ char encbuf[11];
if (!mdb_sec_inited) {
InitializeSecurityDescriptor(&mdb_null_sd,
idbuf.nlow = stbuf.nFileIndexLow;
val.mv_data = &idbuf;
val.mv_size = sizeof(idbuf);
- mdb_hash_hex(&val, hexbuf);
- sprintf(env->me_txns->mti_rmname, "Global\\MDBr%s", hexbuf);
- sprintf(env->me_txns->mti_wmname, "Global\\MDBw%s", hexbuf);
+ mdb_hash_enc(&val, encbuf);
+ sprintf(env->me_txns->mti_rmname, "Global\\MDBr%s", encbuf);
+ sprintf(env->me_txns->mti_wmname, "Global\\MDBw%s", encbuf);
env->me_rmutex = CreateMutex(&mdb_all_sa, FALSE, env->me_txns->mti_rmname);
if (!env->me_rmutex) goto fail_errno;
env->me_wmutex = CreateMutex(&mdb_all_sa, FALSE, env->me_txns->mti_wmname);
ino_t ino;
} idbuf;
MDB_val val;
- char hexbuf[17];
+ char encbuf[11];
+#if defined(__NetBSD__)
+#define MDB_SHORT_SEMNAMES 1 /* limited to 14 chars */
+#endif
if (fstat(env->me_lfd, &stbuf)) goto fail_errno;
idbuf.dev = stbuf.st_dev;
idbuf.ino = stbuf.st_ino;
val.mv_data = &idbuf;
val.mv_size = sizeof(idbuf);
- mdb_hash_hex(&val, hexbuf);
- sprintf(env->me_txns->mti_rmname, "/MDBr%s", hexbuf);
- sprintf(env->me_txns->mti_wmname, "/MDBw%s", hexbuf);
+ mdb_hash_enc(&val, encbuf);
+#ifdef MDB_SHORT_SEMNAMES
+ encbuf[9] = '\0'; /* drop name from 15 chars to 14 chars */
+#endif
+ sprintf(env->me_txns->mti_rmname, "/MDBr%s", encbuf);
+ sprintf(env->me_txns->mti_wmname, "/MDBw%s", encbuf);
/* Clean up after a previous run, if needed: Try to
* remove both semaphores before doing anything else.
*/
pthread_mutexattr_destroy(&mattr);
#endif /* _WIN32 || MDB_USE_POSIX_SEM */
- env->me_txns->mti_version = MDB_VERSION;
env->me_txns->mti_magic = MDB_MAGIC;
+ env->me_txns->mti_format = MDB_LOCK_FORMAT;
env->me_txns->mti_txnid = 0;
env->me_txns->mti_numreaders = 0;
rc = MDB_INVALID;
goto fail;
}
- if (env->me_txns->mti_version != MDB_VERSION) {
- DPRINTF("lock region is version %u, expected version %u",
- env->me_txns->mti_version, MDB_VERSION);
+ if (env->me_txns->mti_format != MDB_LOCK_FORMAT) {
+ DPRINTF(("lock region has format+version 0x%x, expected 0x%x",
+ env->me_txns->mti_format, MDB_LOCK_FORMAT));
rc = MDB_VERSION_MISMATCH;
goto fail;
}
* environment and re-opening it with the new flags.
*/
#define CHANGEABLE (MDB_NOSYNC|MDB_NOMETASYNC|MDB_MAPASYNC)
-#define CHANGELESS (MDB_FIXEDMAP|MDB_NOSUBDIR|MDB_RDONLY|MDB_WRITEMAP|MDB_NOTLS)
+#define CHANGELESS (MDB_FIXEDMAP|MDB_NOSUBDIR|MDB_RDONLY|MDB_WRITEMAP|MDB_NOTLS|MDB_NOLOCK)
int
mdb_env_open(MDB_env *env, const char *path, unsigned int flags, mdb_mode_t mode)
goto leave;
}
- rc = mdb_env_setup_locks(env, lpath, mode, &excl);
- if (rc)
- goto leave;
-
+ /* For RDONLY, get lockfile after we know datafile exists */
+ if (!(flags & (MDB_RDONLY|MDB_NOLOCK))) {
+ rc = mdb_env_setup_locks(env, lpath, mode, &excl);
+ if (rc)
+ goto leave;
+ }
+
#ifdef _WIN32
if (F_ISSET(flags, MDB_RDONLY)) {
oflags = GENERIC_READ;
goto leave;
}
+ if ((flags & (MDB_RDONLY|MDB_NOLOCK)) == MDB_RDONLY) {
+ rc = mdb_env_setup_locks(env, lpath, mode, &excl);
+ if (rc)
+ goto leave;
+ }
+
if ((rc = mdb_env_open2(env)) == MDB_SUCCESS) {
if (flags & (MDB_RDONLY|MDB_WRITEMAP)) {
env->me_mfd = env->me_fd;
* MDB_NOSYNC/MDB_NOMETASYNC, in case these get reset.
*/
#ifdef _WIN32
+ len = OPEN_EXISTING;
env->me_mfd = CreateFile(dpath, oflags,
FILE_SHARE_READ|FILE_SHARE_WRITE, NULL, len,
mode | FILE_FLAG_WRITE_THROUGH, NULL);
#else
+ oflags &= ~O_CREAT;
env->me_mfd = open(dpath, oflags | MDB_DSYNC, mode);
#endif
if (env->me_mfd == INVALID_HANDLE_VALUE) {
goto leave;
}
}
- DPRINTF("opened dbenv %p", (void *) env);
+ DPRINTF(("opened dbenv %p", (void *) env));
if (excl > 0) {
rc = mdb_env_share_locks(env, &excl);
}
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.
}
wsize = env->me_psize * 2;
-#ifdef _WIN32
- {
- DWORD len;
- rc = WriteFile(fd, env->me_map, wsize, &len, NULL);
- rc = rc ? (len == wsize ? MDB_SUCCESS : EIO) : ErrCode();
+ 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;
+ }
}
-#else
- rc = write(fd, env->me_map, wsize);
- rc = rc == (int)wsize ? MDB_SUCCESS : rc < 0 ? ErrCode() : EIO;
-#endif
if (env->me_txns)
UNLOCK_MUTEX_W(env);
if (rc)
goto leave;
- ptr = env->me_map + wsize;
wsize = txn->mt_next_pgno * env->me_psize - wsize;
-#ifdef _WIN32
- while (wsize > 0) {
- DWORD len, w2;
- if (wsize > MAX_WRITE)
- w2 = MAX_WRITE;
- else
- w2 = wsize;
- rc = WriteFile(fd, ptr, w2, &len, NULL);
- rc = rc ? (len == w2 ? MDB_SUCCESS : EIO) : ErrCode();
- if (rc) break;
- wsize -= w2;
- ptr += w2;
- }
-#else
while (wsize > 0) {
- size_t w2;
- ssize_t wres;
if (wsize > MAX_WRITE)
w2 = MAX_WRITE;
else
w2 = wsize;
- wres = write(fd, ptr, w2);
- rc = wres == (ssize_t)w2 ? MDB_SUCCESS : wres < 0 ? ErrCode() : EIO;
- if (rc) break;
- wsize -= wres;
- ptr += wres;
+ 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;
+ }
}
-#endif
leave:
mdb_txn_abort(txn);
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
-#ifdef O_DIRECT
- |O_DIRECT
-#endif
- , 0666);
+ 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) {
*(size_t *)a->mv_data > *(size_t *)b->mv_data;
}
-/** Compare two items pointing at aligned int's */
+/** Compare two items pointing at aligned unsigned int's */
static int
mdb_cmp_int(const MDB_val *a, const MDB_val *b)
{
*(unsigned int *)a->mv_data > *(unsigned int *)b->mv_data;
}
-/** Compare two items pointing at ints of unknown alignment.
+/** Compare two items pointing at unsigned ints of unknown alignment.
* Nodes and keys are guaranteed to be 2-byte aligned.
*/
static int
{
pgno_t pgno;
COPY_PGNO(pgno, mp->mp_pgno);
- DPRINTF("searching %u keys in %s %spage %zu",
+ DPRINTF(("searching %u keys in %s %spage %"Z"u",
nkeys, IS_LEAF(mp) ? "leaf" : "branch", IS_SUBP(mp) ? "sub-" : "",
- pgno);
+ pgno));
}
#endif
i = (low + high) >> 1;
nodekey.mv_data = LEAF2KEY(mp, i, nodekey.mv_size);
rc = cmp(key, &nodekey);
- DPRINTF("found leaf index %u [%s], rc = %i",
- i, DKEY(&nodekey), rc);
+ DPRINTF(("found leaf index %u [%s], rc = %i",
+ i, DKEY(&nodekey), rc));
if (rc == 0)
break;
if (rc > 0)
rc = cmp(key, &nodekey);
#if MDB_DEBUG
if (IS_LEAF(mp))
- DPRINTF("found leaf index %u [%s], rc = %i",
- i, DKEY(&nodekey), rc);
+ DPRINTF(("found leaf index %u [%s], rc = %i",
+ i, DKEY(&nodekey), rc));
else
- DPRINTF("found branch index %u [%s -> %zu], rc = %i",
- i, DKEY(&nodekey), NODEPGNO(node), rc);
+ DPRINTF(("found branch index %u [%s -> %"Z"u], rc = %i",
+ i, DKEY(&nodekey), NODEPGNO(node), rc));
#endif
if (rc == 0)
break;
mdb_cursor_pop(MDB_cursor *mc)
{
if (mc->mc_snum) {
-#ifndef MDB_DEBUG_SKIP
+#if MDB_DEBUG
MDB_page *top = mc->mc_pg[mc->mc_top];
#endif
mc->mc_snum--;
if (mc->mc_snum)
mc->mc_top--;
- DPRINTF("popped page %zu off db %u cursor %p", top->mp_pgno,
- mc->mc_dbi, (void *) mc);
+ DPRINTF(("popped page %"Z"u off db %d cursor %p", top->mp_pgno,
+ DDBI(mc), (void *) mc));
}
}
static int
mdb_cursor_push(MDB_cursor *mc, MDB_page *mp)
{
- DPRINTF("pushing page %zu on db %u cursor %p", mp->mp_pgno,
- mc->mc_dbi, (void *) mc);
+ DPRINTF(("pushing page %"Z"u on db %d cursor %p", mp->mp_pgno,
+ DDBI(mc), (void *) mc));
if (mc->mc_snum >= CURSOR_STACK) {
assert(mc->mc_snum < CURSOR_STACK);
static int
mdb_page_get(MDB_txn *txn, pgno_t pgno, MDB_page **ret, int *lvl)
{
+ MDB_env *env = txn->mt_env;
MDB_page *p = NULL;
int level;
- if (!((txn->mt_flags & MDB_TXN_RDONLY) |
- (txn->mt_env->me_flags & MDB_WRITEMAP)))
- {
+ if (!((txn->mt_flags & MDB_TXN_RDONLY) | (env->me_flags & MDB_WRITEMAP))) {
MDB_txn *tx2 = txn;
level = 1;
do {
MDB_ID2L dl = tx2->mt_u.dirty_list;
+ unsigned x;
+ /* Spilled pages were dirtied in this txn and flushed
+ * because the dirty list got full. Bring this page
+ * back in from the map (but don't unspill it here,
+ * leave that unless page_touch happens again).
+ */
+ if (tx2->mt_spill_pgs) {
+ MDB_ID pn = pgno << 1;
+ x = mdb_midl_search(tx2->mt_spill_pgs, pn);
+ if (x <= tx2->mt_spill_pgs[0] && tx2->mt_spill_pgs[x] == pn) {
+ p = (MDB_page *)(env->me_map + env->me_psize * pgno);
+ goto done;
+ }
+ }
if (dl[0].mid) {
unsigned x = mdb_mid2l_search(dl, pgno);
if (x <= dl[0].mid && dl[x].mid == pgno) {
if (pgno < txn->mt_next_pgno) {
level = 0;
- p = (MDB_page *)(txn->mt_env->me_map + txn->mt_env->me_psize * pgno);
+ p = (MDB_page *)(env->me_map + env->me_psize * pgno);
} else {
- DPRINTF("page %zu not found", pgno);
+ DPRINTF(("page %"Z"u not found", pgno));
assert(p != NULL);
return MDB_PAGE_NOTFOUND;
}
return MDB_SUCCESS;
}
-/** Search for the page a given key should be in.
- * Pushes parent pages on the cursor stack. This function continues a
- * search on a cursor that has already been initialized. (Usually by
- * #mdb_page_search() but also by #mdb_node_move().)
- * @param[in,out] mc the cursor for this operation.
- * @param[in] key the key to search for. If NULL, search for the lowest
- * page. (This is used by #mdb_cursor_first().)
- * @param[in] modify If true, visited pages are updated with new page numbers.
- * @return 0 on success, non-zero on failure.
+/** Finish #mdb_page_search() / #mdb_page_search_lowest().
+ * The cursor is at the root page, set up the rest of it.
*/
static int
-mdb_page_search_root(MDB_cursor *mc, MDB_val *key, int modify)
+mdb_page_search_root(MDB_cursor *mc, MDB_val *key, int flags)
{
MDB_page *mp = mc->mc_pg[mc->mc_top];
- DKBUF;
int rc;
-
+ DKBUF;
while (IS_BRANCH(mp)) {
MDB_node *node;
indx_t i;
- DPRINTF("branch page %zu has %u keys", mp->mp_pgno, NUMKEYS(mp));
+ DPRINTF(("branch page %"Z"u has %u keys", mp->mp_pgno, NUMKEYS(mp)));
assert(NUMKEYS(mp) > 1);
- DPRINTF("found index 0 to page %zu", NODEPGNO(NODEPTR(mp, 0)));
+ DPRINTF(("found index 0 to page %"Z"u", NODEPGNO(NODEPTR(mp, 0))));
- if (key == NULL) /* Initialize cursor to first page. */
+ if (flags & (MDB_PS_FIRST|MDB_PS_LAST)) {
i = 0;
- else if (key->mv_size > MDB_MAXKEYSIZE && key->mv_data == NULL) {
- /* cursor to last page */
- i = NUMKEYS(mp)-1;
+ if (flags & MDB_PS_LAST)
+ i = NUMKEYS(mp) - 1;
} else {
int exact;
node = mdb_node_search(mc, key, &exact);
i--;
}
}
+ DPRINTF(("following index %u for key [%s]", i, DKEY(key)));
}
- if (key)
- DPRINTF("following index %u for key [%s]",
- i, DKEY(key));
assert(i < NUMKEYS(mp));
node = NODEPTR(mp, i);
if ((rc = mdb_cursor_push(mc, mp)))
return rc;
- if (modify) {
+ if (flags & MDB_PS_MODIFY) {
if ((rc = mdb_page_touch(mc)) != 0)
return rc;
mp = mc->mc_pg[mc->mc_top];
}
if (!IS_LEAF(mp)) {
- DPRINTF("internal error, index points to a %02X page!?",
- mp->mp_flags);
+ DPRINTF(("internal error, index points to a %02X page!?",
+ mp->mp_flags));
return MDB_CORRUPTED;
}
- DPRINTF("found leaf page %zu for key [%s]", mp->mp_pgno,
- key ? DKEY(key) : NULL);
+ DPRINTF(("found leaf page %"Z"u for key [%s]", mp->mp_pgno,
+ key ? DKEY(key) : "null"));
+ mc->mc_flags |= C_INITIALIZED;
+ mc->mc_flags &= ~C_EOF;
return MDB_SUCCESS;
}
mc->mc_ki[mc->mc_top] = 0;
if ((rc = mdb_cursor_push(mc, mp)))
return rc;
- return mdb_page_search_root(mc, NULL, 0);
+ return mdb_page_search_root(mc, NULL, MDB_PS_FIRST);
}
/** Search for the page a given key should be in.
- * Pushes parent pages on the cursor stack. This function just sets up
- * the search; it finds the root page for \b mc's database and sets this
- * as the root of the cursor's stack. Then #mdb_page_search_root() is
- * called to complete the search.
+ * Push it and its parent pages on the cursor stack.
* @param[in,out] mc the cursor for this operation.
- * @param[in] key the key to search for. If NULL, search for the lowest
- * page. (This is used by #mdb_cursor_first().)
- * @param[in] flags If MDB_PS_MODIFY set, visited pages are updated with new page numbers.
+ * @param[in] key the key to search for, or NULL for first/last page.
+ * @param[in] flags If MDB_PS_MODIFY is set, visited pages in the DB
+ * are touched (updated with new page numbers).
+ * If MDB_PS_FIRST or MDB_PS_LAST is set, find first or last leaf.
+ * This is used by #mdb_cursor_first() and #mdb_cursor_last().
* If MDB_PS_ROOTONLY set, just fetch root node, no further lookups.
* @return 0 on success, non-zero on failure.
*/
pgno_t root;
/* Make sure the txn is still viable, then find the root from
- * the txn's db table.
+ * the txn's db table and set it as the root of the cursor's stack.
*/
if (F_ISSET(mc->mc_txn->mt_flags, MDB_TXN_ERROR)) {
DPUTS("transaction has failed, must abort");
- return EINVAL;
+ return MDB_BAD_TXN;
} else {
/* Make sure we're using an up-to-date root */
- if (mc->mc_dbi > MAIN_DBI) {
- if ((*mc->mc_dbflag & DB_STALE) ||
- ((flags & MDB_PS_MODIFY) && !(*mc->mc_dbflag & DB_DIRTY))) {
+ if (*mc->mc_dbflag & DB_STALE) {
MDB_cursor mc2;
- unsigned char dbflag = 0;
mdb_cursor_init(&mc2, mc->mc_txn, MAIN_DBI, NULL);
- rc = mdb_page_search(&mc2, &mc->mc_dbx->md_name, flags & MDB_PS_MODIFY);
+ rc = mdb_page_search(&mc2, &mc->mc_dbx->md_name, 0);
if (rc)
return rc;
- if (*mc->mc_dbflag & DB_STALE) {
+ {
MDB_val data;
int exact = 0;
uint16_t flags;
return MDB_INCOMPATIBLE;
memcpy(mc->mc_db, data.mv_data, sizeof(MDB_db));
}
- if (flags & MDB_PS_MODIFY)
- dbflag = DB_DIRTY;
*mc->mc_dbflag &= ~DB_STALE;
- *mc->mc_dbflag |= dbflag;
- }
}
root = mc->mc_db->md_root;
mc->mc_snum = 1;
mc->mc_top = 0;
- DPRINTF("db %u root page %zu has flags 0x%X",
- mc->mc_dbi, root, mc->mc_pg[0]->mp_flags);
+ DPRINTF(("db %d root page %"Z"u has flags 0x%X",
+ DDBI(mc), root, mc->mc_pg[0]->mp_flags));
if (flags & MDB_PS_MODIFY) {
if ((rc = mdb_page_touch(mc)))
{
MDB_txn *txn = mc->mc_txn;
pgno_t pg = mp->mp_pgno;
- unsigned i, ovpages = mp->mp_pages;
+ unsigned x = 0, ovpages = mp->mp_pages;
MDB_env *env = txn->mt_env;
+ MDB_IDL sl = txn->mt_spill_pgs;
+ MDB_ID pn = pg << 1;
int rc;
- DPRINTF("free ov page %zu (%d)", pg, ovpages);
- /* If the page is dirty we just acquired it, so we should
- * give it back to our current free list, if any.
- * Not currently supported in nested txns.
+ DPRINTF(("free ov page %"Z"u (%d)", pg, ovpages));
+ /* If the page is dirty or on the spill list we just acquired it,
+ * so we should give it back to our current free list, if any.
* Otherwise put it onto the list of pages we freed in this txn.
+ *
+ * Won't create me_pghead: me_pglast must be inited along with it.
+ * Unsupported in nested txns: They would need to hide the page
+ * range in ancestor txns' dirty and spilled lists.
*/
- if ((mp->mp_flags & P_DIRTY) && !txn->mt_parent && env->me_pghead) {
- unsigned j, x;
+ if (env->me_pghead &&
+ !txn->mt_parent &&
+ ((mp->mp_flags & P_DIRTY) ||
+ (sl && (x = mdb_midl_search(sl, pn)) <= sl[0] && sl[x] == pn)))
+ {
+ unsigned i, j;
pgno_t *mop;
MDB_ID2 *dl, ix, iy;
rc = mdb_midl_need(&env->me_pghead, ovpages);
if (rc)
return rc;
+ if (!(mp->mp_flags & P_DIRTY)) {
+ /* This page is no longer spilled */
+ if (x == sl[0])
+ sl[0]--;
+ else
+ sl[x] |= 1;
+ goto release;
+ }
/* Remove from dirty list */
dl = txn->mt_u.dirty_list;
x = dl[0].mid--;
}
if (!(env->me_flags & MDB_WRITEMAP))
mdb_dpage_free(env, mp);
+release:
/* Insert in me_pghead */
mop = env->me_pghead;
j = mop[0] + ovpages;
data->mv_size = NODEDSZ(leaf);
memcpy(&pgno, NODEDATA(leaf), sizeof(pgno));
if ((rc = mdb_page_get(txn, pgno, &omp, NULL)) != 0) {
- DPRINTF("read overflow page %zu failed", pgno);
+ DPRINTF(("read overflow page %"Z"u failed", pgno));
return rc;
}
data->mv_data = METADATA(omp);
assert(key);
assert(data);
- DPRINTF("===> get db %u key [%s]", dbi, DKEY(key));
+ DPRINTF(("===> get db %u key [%s]", dbi, DKEY(key)));
if (txn == NULL || !dbi || dbi >= txn->mt_numdbs || !(txn->mt_dbflags[dbi] & DB_VALID))
return EINVAL;
- if (key->mv_size == 0 || key->mv_size > MDB_MAXKEYSIZE) {
- return EINVAL;
+ if (txn->mt_flags & MDB_TXN_ERROR)
+ return MDB_BAD_TXN;
+
+ if (key->mv_size > MDB_MAXKEYSIZE) {
+ return MDB_BAD_VALSIZE;
}
mdb_cursor_init(&mc, txn, dbi, &mx);
}
mdb_cursor_pop(mc);
- DPRINTF("parent page is page %zu, index %u",
- mc->mc_pg[mc->mc_top]->mp_pgno, mc->mc_ki[mc->mc_top]);
+ DPRINTF(("parent page is page %"Z"u, index %u",
+ mc->mc_pg[mc->mc_top]->mp_pgno, mc->mc_ki[mc->mc_top]));
if (move_right ? (mc->mc_ki[mc->mc_top] + 1u >= NUMKEYS(mc->mc_pg[mc->mc_top]))
: (mc->mc_ki[mc->mc_top] == 0)) {
- DPRINTF("no more keys left, moving to %s sibling",
- move_right ? "right" : "left");
+ DPRINTF(("no more keys left, moving to %s sibling",
+ move_right ? "right" : "left"));
if ((rc = mdb_cursor_sibling(mc, move_right)) != MDB_SUCCESS) {
/* undo cursor_pop before returning */
mc->mc_top++;
mc->mc_ki[mc->mc_top]++;
else
mc->mc_ki[mc->mc_top]--;
- DPRINTF("just moving to %s index key %u",
- move_right ? "right" : "left", mc->mc_ki[mc->mc_top]);
+ DPRINTF(("just moving to %s index key %u",
+ move_right ? "right" : "left", mc->mc_ki[mc->mc_top]));
}
assert(IS_BRANCH(mc->mc_pg[mc->mc_top]));
indx = NODEPTR(mc->mc_pg[mc->mc_top], mc->mc_ki[mc->mc_top]);
- if ((rc = mdb_page_get(mc->mc_txn, NODEPGNO(indx), &mp, NULL) != 0))
+ if ((rc = mdb_page_get(mc->mc_txn, NODEPGNO(indx), &mp, NULL)) != 0) {
+ /* mc will be inconsistent if caller does mc_snum++ as above */
+ mc->mc_flags &= ~(C_INITIALIZED|C_EOF);
return rc;
+ }
mdb_cursor_push(mc, mp);
if (!move_right)
if (F_ISSET(leaf->mn_flags, F_DUPDATA)) {
if (op == MDB_NEXT || op == MDB_NEXT_DUP) {
rc = mdb_cursor_next(&mc->mc_xcursor->mx_cursor, data, NULL, MDB_NEXT);
- if (op != MDB_NEXT || rc != MDB_NOTFOUND)
+ if (op != MDB_NEXT || rc != MDB_NOTFOUND) {
+ if (rc == MDB_SUCCESS)
+ MDB_GET_KEY(leaf, key);
return rc;
+ }
}
} else {
mc->mc_xcursor->mx_cursor.mc_flags &= ~(C_INITIALIZED|C_EOF);
}
}
- DPRINTF("cursor_next: top page is %zu in cursor %p", mp->mp_pgno, (void *) mc);
+ DPRINTF(("cursor_next: top page is %"Z"u in cursor %p", mp->mp_pgno, (void *) mc));
+ if (mc->mc_flags & C_DEL)
+ goto skip;
if (mc->mc_ki[mc->mc_top] + 1u >= NUMKEYS(mp)) {
DPUTS("=====> move to next sibling page");
return rc;
}
mp = mc->mc_pg[mc->mc_top];
- DPRINTF("next page is %zu, key index %u", mp->mp_pgno, mc->mc_ki[mc->mc_top]);
+ DPRINTF(("next page is %"Z"u, key index %u", mp->mp_pgno, mc->mc_ki[mc->mc_top]));
} else
mc->mc_ki[mc->mc_top]++;
- DPRINTF("==> cursor points to page %zu with %u keys, key index %u",
- mp->mp_pgno, NUMKEYS(mp), mc->mc_ki[mc->mc_top]);
+skip:
+ DPRINTF(("==> cursor points to page %"Z"u with %u keys, key index %u",
+ mp->mp_pgno, NUMKEYS(mp), mc->mc_ki[mc->mc_top]));
if (IS_LEAF2(mp)) {
key->mv_size = mc->mc_db->md_pad;
if (mc->mc_db->md_flags & MDB_DUPSORT) {
leaf = NODEPTR(mp, mc->mc_ki[mc->mc_top]);
- if (op == MDB_PREV || op == MDB_PREV_DUP) {
- if (F_ISSET(leaf->mn_flags, F_DUPDATA)) {
+ if (F_ISSET(leaf->mn_flags, F_DUPDATA)) {
+ if (op == MDB_PREV || op == MDB_PREV_DUP) {
rc = mdb_cursor_prev(&mc->mc_xcursor->mx_cursor, data, NULL, MDB_PREV);
- if (op != MDB_PREV || rc != MDB_NOTFOUND)
+ if (op != MDB_PREV || rc != MDB_NOTFOUND) {
+ if (rc == MDB_SUCCESS)
+ MDB_GET_KEY(leaf, key);
return rc;
+ }
} else {
mc->mc_xcursor->mx_cursor.mc_flags &= ~(C_INITIALIZED|C_EOF);
if (op == MDB_PREV_DUP)
}
}
- DPRINTF("cursor_prev: top page is %zu in cursor %p", mp->mp_pgno, (void *) mc);
+ DPRINTF(("cursor_prev: top page is %"Z"u in cursor %p", mp->mp_pgno, (void *) mc));
if (mc->mc_ki[mc->mc_top] == 0) {
DPUTS("=====> move to prev sibling page");
}
mp = mc->mc_pg[mc->mc_top];
mc->mc_ki[mc->mc_top] = NUMKEYS(mp) - 1;
- DPRINTF("prev page is %zu, key index %u", mp->mp_pgno, mc->mc_ki[mc->mc_top]);
+ DPRINTF(("prev page is %"Z"u, key index %u", mp->mp_pgno, mc->mc_ki[mc->mc_top]));
} else
mc->mc_ki[mc->mc_top]--;
mc->mc_flags &= ~C_EOF;
- DPRINTF("==> cursor points to page %zu with %u keys, key index %u",
- mp->mp_pgno, NUMKEYS(mp), mc->mc_ki[mc->mc_top]);
+ DPRINTF(("==> cursor points to page %"Z"u with %u keys, key index %u",
+ mp->mp_pgno, NUMKEYS(mp), mc->mc_ki[mc->mc_top]));
if (IS_LEAF2(mp)) {
key->mv_size = mc->mc_db->md_pad;
assert(mc);
assert(key);
- assert(key->mv_size > 0);
+ if (key->mv_size == 0)
+ return MDB_BAD_VALSIZE;
+
+ if (mc->mc_xcursor)
+ mc->mc_xcursor->mx_cursor.mc_flags &= ~(C_INITIALIZED|C_EOF);
/* See if we're already on the right page */
if (mc->mc_flags & C_INITIALIZED) {
nodekey.mv_data = LEAF2KEY(mp, 0, nodekey.mv_size);
} else {
leaf = NODEPTR(mp, 0);
- MDB_GET_KEY(leaf, &nodekey);
+ MDB_GET_KEY2(leaf, nodekey);
}
rc = mc->mc_dbx->md_cmp(key, &nodekey);
if (rc == 0) {
nkeys-1, nodekey.mv_size);
} else {
leaf = NODEPTR(mp, nkeys-1);
- MDB_GET_KEY(leaf, &nodekey);
+ MDB_GET_KEY2(leaf, nodekey);
}
rc = mc->mc_dbx->md_cmp(key, &nodekey);
if (rc == 0) {
mc->mc_ki[mc->mc_top], nodekey.mv_size);
} else {
leaf = NODEPTR(mp, mc->mc_ki[mc->mc_top]);
- MDB_GET_KEY(leaf, &nodekey);
+ MDB_GET_KEY2(leaf, nodekey);
}
rc = mc->mc_dbx->md_cmp(key, &nodekey);
if (rc == 0) {
if (!mc->mc_top) {
/* There are no other pages */
mc->mc_ki[mc->mc_top] = 0;
- return MDB_NOTFOUND;
+ if (op == MDB_SET_RANGE) {
+ rc = 0;
+ goto set1;
+ } else
+ return MDB_NOTFOUND;
}
}
if (rc) {
if (op == MDB_GET_BOTH || rc > 0)
return MDB_NOTFOUND;
+ rc = 0;
}
} else {
/* The key already matches in all other cases */
if (op == MDB_SET_RANGE || op == MDB_SET_KEY)
MDB_GET_KEY(leaf, key);
- DPRINTF("==> cursor placed on key [%s]", DKEY(key));
+ DPRINTF(("==> cursor placed on key [%s]", DKEY(key)));
return rc;
}
int rc;
MDB_node *leaf;
+ if (mc->mc_xcursor)
+ mc->mc_xcursor->mx_cursor.mc_flags &= ~(C_INITIALIZED|C_EOF);
+
if (!(mc->mc_flags & C_INITIALIZED) || mc->mc_top) {
- rc = mdb_page_search(mc, NULL, 0);
+ rc = mdb_page_search(mc, NULL, MDB_PS_FIRST);
if (rc != MDB_SUCCESS)
return rc;
}
if (rc)
return rc;
} else {
- if (mc->mc_xcursor)
- mc->mc_xcursor->mx_cursor.mc_flags &= ~(C_INITIALIZED|C_EOF);
if ((rc = mdb_node_read(mc->mc_txn, leaf, data)) != MDB_SUCCESS)
return rc;
}
int rc;
MDB_node *leaf;
+ if (mc->mc_xcursor)
+ mc->mc_xcursor->mx_cursor.mc_flags &= ~(C_INITIALIZED|C_EOF);
+
if (!(mc->mc_flags & C_EOF)) {
if (!(mc->mc_flags & C_INITIALIZED) || mc->mc_top) {
- MDB_val lkey;
-
- lkey.mv_size = MDB_MAXKEYSIZE+1;
- lkey.mv_data = NULL;
- rc = mdb_page_search(mc, &lkey, 0);
+ rc = mdb_page_search(mc, NULL, MDB_PS_LAST);
if (rc != MDB_SUCCESS)
return rc;
}
if (rc)
return rc;
} else {
- if (mc->mc_xcursor)
- mc->mc_xcursor->mx_cursor.mc_flags &= ~(C_INITIALIZED|C_EOF);
if ((rc = mdb_node_read(mc->mc_txn, leaf, data)) != MDB_SUCCESS)
return rc;
}
{
int rc;
int exact = 0;
+ int (*mfunc)(MDB_cursor *mc, MDB_val *key, MDB_val *data);
assert(mc);
+ if (mc->mc_txn->mt_flags & MDB_TXN_ERROR)
+ return MDB_BAD_TXN;
+
switch (op) {
case MDB_GET_CURRENT:
if (!(mc->mc_flags & C_INITIALIZED)) {
MDB_GET_KEY(leaf, key);
if (data) {
if (F_ISSET(leaf->mn_flags, F_DUPDATA)) {
+ if (mc->mc_flags & C_DEL)
+ mdb_xcursor_init1(mc, leaf);
rc = mdb_cursor_get(&mc->mc_xcursor->mx_cursor, data, NULL, MDB_GET_CURRENT);
} else {
rc = mdb_node_read(mc->mc_txn, leaf, data);
break;
case MDB_GET_BOTH:
case MDB_GET_BOTH_RANGE:
- if (data == NULL || mc->mc_xcursor == NULL) {
+ if (data == NULL) {
rc = EINVAL;
break;
}
+ if (mc->mc_xcursor == NULL) {
+ rc = MDB_INCOMPATIBLE;
+ break;
+ }
/* FALLTHRU */
case MDB_SET:
case MDB_SET_KEY:
case MDB_SET_RANGE:
- if (key == NULL || key->mv_size == 0 || key->mv_size > MDB_MAXKEYSIZE) {
+ if (key == NULL) {
rc = EINVAL;
+ } else if (key->mv_size > MDB_MAXKEYSIZE) {
+ rc = MDB_BAD_VALSIZE;
} else if (op == MDB_SET_RANGE)
rc = mdb_cursor_set(mc, key, data, op, NULL);
else
rc = mdb_cursor_set(mc, key, data, op, &exact);
break;
case MDB_GET_MULTIPLE:
- if (data == NULL ||
- !(mc->mc_db->md_flags & MDB_DUPFIXED) ||
- !(mc->mc_flags & C_INITIALIZED)) {
+ if (data == NULL || !(mc->mc_flags & C_INITIALIZED)) {
rc = EINVAL;
break;
}
+ if (!(mc->mc_db->md_flags & MDB_DUPFIXED)) {
+ rc = MDB_INCOMPATIBLE;
+ break;
+ }
rc = MDB_SUCCESS;
if (!(mc->mc_xcursor->mx_cursor.mc_flags & C_INITIALIZED) ||
(mc->mc_xcursor->mx_cursor.mc_flags & C_EOF))
break;
goto fetchm;
case MDB_NEXT_MULTIPLE:
- if (data == NULL ||
- !(mc->mc_db->md_flags & MDB_DUPFIXED)) {
+ if (data == NULL) {
rc = EINVAL;
break;
}
+ if (!(mc->mc_db->md_flags & MDB_DUPFIXED)) {
+ rc = MDB_INCOMPATIBLE;
+ break;
+ }
if (!(mc->mc_flags & C_INITIALIZED))
rc = mdb_cursor_first(mc, key, data);
else
rc = mdb_cursor_first(mc, key, data);
break;
case MDB_FIRST_DUP:
- if (data == NULL ||
- !(mc->mc_db->md_flags & MDB_DUPSORT) ||
- !(mc->mc_flags & C_INITIALIZED) ||
- !(mc->mc_xcursor->mx_cursor.mc_flags & C_INITIALIZED)) {
+ mfunc = mdb_cursor_first;
+ mmove:
+ if (data == NULL || !(mc->mc_flags & C_INITIALIZED)) {
+ rc = EINVAL;
+ break;
+ }
+ if (mc->mc_xcursor == NULL) {
+ rc = MDB_INCOMPATIBLE;
+ break;
+ }
+ if (!(mc->mc_xcursor->mx_cursor.mc_flags & C_INITIALIZED)) {
rc = EINVAL;
break;
}
- rc = mdb_cursor_first(&mc->mc_xcursor->mx_cursor, data, NULL);
+ rc = mfunc(&mc->mc_xcursor->mx_cursor, data, NULL);
break;
case MDB_LAST:
rc = mdb_cursor_last(mc, key, data);
break;
case MDB_LAST_DUP:
- if (data == NULL ||
- !(mc->mc_db->md_flags & MDB_DUPSORT) ||
- !(mc->mc_flags & C_INITIALIZED) ||
- !(mc->mc_xcursor->mx_cursor.mc_flags & C_INITIALIZED)) {
- rc = EINVAL;
- break;
- }
- rc = mdb_cursor_last(&mc->mc_xcursor->mx_cursor, data, NULL);
- break;
+ mfunc = mdb_cursor_last;
+ goto mmove;
default:
- DPRINTF("unhandled/unimplemented cursor operation %u", op);
+ DPRINTF(("unhandled/unimplemented cursor operation %u", op));
rc = EINVAL;
break;
}
+ if (mc->mc_flags & C_DEL)
+ mc->mc_flags ^= C_DEL;
+
return rc;
}
-/** Touch all the pages in the cursor stack.
+/** Touch all the pages in the cursor stack. Set mc_top.
* Makes sure all the pages are writable, before attempting a write operation.
* @param[in] mc The cursor to operate on.
*/
static int
mdb_cursor_touch(MDB_cursor *mc)
{
- int rc;
+ int rc = MDB_SUCCESS;
if (mc->mc_dbi > MAIN_DBI && !(*mc->mc_dbflag & DB_DIRTY)) {
MDB_cursor mc2;
return rc;
*mc->mc_dbflag |= DB_DIRTY;
}
- for (mc->mc_top = 0; mc->mc_top < mc->mc_snum; mc->mc_top++) {
- rc = mdb_page_touch(mc);
- if (rc)
- return rc;
+ mc->mc_top = 0;
+ if (mc->mc_snum) {
+ do {
+ rc = mdb_page_touch(mc);
+ } while (!rc && ++(mc->mc_top) < mc->mc_snum);
+ mc->mc_top = mc->mc_snum-1;
}
- mc->mc_top = mc->mc_snum-1;
- return MDB_SUCCESS;
+ return rc;
}
+/** Do not spill pages to disk if txn is getting full, may fail instead */
+#define MDB_NOSPILL 0x8000
+
int
mdb_cursor_put(MDB_cursor *mc, MDB_val *key, MDB_val *data,
unsigned int flags)
{
+ enum { MDB_NO_ROOT = MDB_LAST_ERRCODE+10 }; /* internal code */
MDB_node *leaf = NULL;
MDB_val xdata, *rdata, dkey;
MDB_page *fp;
MDB_db dummy;
int do_sub = 0, insert = 0;
- unsigned int mcount = 0, dcount = 0;
+ unsigned int mcount = 0, dcount = 0, nospill;
size_t nsize;
int rc, rc2;
MDB_pagebuf pbuf;
dcount = data[1].mv_size;
data[1].mv_size = 0;
if (!F_ISSET(mc->mc_db->md_flags, MDB_DUPFIXED))
- return EINVAL;
+ return MDB_INCOMPATIBLE;
}
- if (F_ISSET(mc->mc_txn->mt_flags, MDB_TXN_RDONLY))
- return EACCES;
+ nospill = flags & MDB_NOSPILL;
+ flags &= ~MDB_NOSPILL;
+
+ 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 && (key->mv_size == 0 || key->mv_size > MDB_MAXKEYSIZE))
- return EINVAL;
+ return MDB_BAD_VALSIZE;
if (F_ISSET(mc->mc_db->md_flags, MDB_DUPSORT) && data->mv_size > MDB_MAXKEYSIZE)
- return EINVAL;
+ return MDB_BAD_VALSIZE;
#if SIZE_MAX > MAXDATASIZE
if (data->mv_size > MAXDATASIZE)
- return EINVAL;
+ return MDB_BAD_VALSIZE;
#endif
- DPRINTF("==> put db %u key [%s], size %zu, data size %zu",
- mc->mc_dbi, DKEY(key), key ? key->mv_size:0, data->mv_size);
+ DPRINTF(("==> put db %d key [%s], size %"Z"u, data size %"Z"u",
+ DDBI(mc), DKEY(key), key ? key->mv_size : 0, data->mv_size));
dkey.mv_size = 0;
return EINVAL;
rc = MDB_SUCCESS;
} else if (mc->mc_db->md_root == P_INVALID) {
- MDB_page *np;
- /* new database, write a root leaf page */
- DPUTS("allocating new root leaf page");
- if ((rc = mdb_page_new(mc, P_LEAF, 1, &np))) {
- return rc;
- }
+ /* new database, cursor has nothing to point to */
mc->mc_snum = 0;
- mdb_cursor_push(mc, np);
- mc->mc_db->md_root = np->mp_pgno;
- mc->mc_db->md_depth++;
- *mc->mc_dbflag |= DB_DIRTY;
- if ((mc->mc_db->md_flags & (MDB_DUPSORT|MDB_DUPFIXED))
- == MDB_DUPFIXED)
- np->mp_flags |= P_LEAF2;
- mc->mc_flags |= C_INITIALIZED;
- rc = MDB_NOTFOUND;
- goto top;
+ mc->mc_top = 0;
+ mc->mc_flags &= ~C_INITIALIZED;
+ rc = MDB_NO_ROOT;
} else {
int exact = 0;
MDB_val d2;
}
}
} else {
- rc = mdb_cursor_set(mc, key, &d2, MDB_SET, &exact);
+ rc = mdb_cursor_set(mc, key, &d2, MDB_SET, &exact);
}
if ((flags & MDB_NOOVERWRITE) && rc == 0) {
- DPRINTF("duplicate key [%s]", DKEY(key));
+ DPRINTF(("duplicate key [%s]", DKEY(key)));
*data = d2;
return MDB_KEYEXIST;
}
return rc;
}
- /* Cursor is positioned, now make sure all pages are writable */
- rc2 = mdb_cursor_touch(mc);
- if (rc2)
- return rc2;
+ if (mc->mc_flags & C_DEL)
+ mc->mc_flags ^= C_DEL;
+
+ /* Cursor is positioned, check for room in the dirty list */
+ if (!nospill) {
+ if (flags & MDB_MULTIPLE) {
+ rdata = &xdata;
+ xdata.mv_size = data->mv_size * dcount;
+ } else {
+ rdata = data;
+ }
+ if ((rc2 = mdb_page_spill(mc, key, rdata)))
+ return rc2;
+ }
+
+ if (rc == MDB_NO_ROOT) {
+ MDB_page *np;
+ /* new database, write a root leaf page */
+ DPUTS("allocating new root leaf page");
+ if ((rc2 = mdb_page_new(mc, P_LEAF, 1, &np))) {
+ return rc2;
+ }
+ mdb_cursor_push(mc, np);
+ mc->mc_db->md_root = np->mp_pgno;
+ mc->mc_db->md_depth++;
+ *mc->mc_dbflag |= DB_DIRTY;
+ if ((mc->mc_db->md_flags & (MDB_DUPSORT|MDB_DUPFIXED))
+ == MDB_DUPFIXED)
+ np->mp_flags |= P_LEAF2;
+ mc->mc_flags |= C_INITIALIZED;
+ } else {
+ /* make sure all cursor pages are writable */
+ rc2 = mdb_cursor_touch(mc);
+ if (rc2)
+ return rc2;
+ }
-top:
/* The key already exists */
if (rc == MDB_SUCCESS) {
/* there's only a key anyway, so this is a no-op */
if (IS_LEAF2(mc->mc_pg[mc->mc_top])) {
unsigned int ksize = mc->mc_db->md_pad;
if (key->mv_size != ksize)
- return EINVAL;
+ 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);
mc->mc_dbx->md_dcmp = mdb_cmp_cint;
#endif
#endif
- /* if data matches, ignore it */
- if (!mc->mc_dbx->md_dcmp(data, &dkey))
- return (flags == MDB_NODUPDATA) ? MDB_KEYEXIST : MDB_SUCCESS;
+ /* if data matches, skip it */
+ if (!mc->mc_dbx->md_dcmp(data, &dkey)) {
+ if (flags & MDB_NODUPDATA)
+ rc = MDB_KEYEXIST;
+ else if (flags & MDB_MULTIPLE)
+ goto next_mult;
+ else
+ rc = MDB_SUCCESS;
+ return rc;
+ }
/* create a fake page for the dup items */
memcpy(dbuf, dkey.mv_data, dkey.mv_size);
return rc2;
ovpages = omp->mp_pages;
- /* Is the ov page writable and large enough? */
- if ((omp->mp_flags & P_DIRTY) && ovpages >= dpages) {
+ /* Is the ov page large enough? */
+ if (ovpages >= dpages) {
+ if (!(omp->mp_flags & P_DIRTY) &&
+ (level || (mc->mc_txn->mt_env->me_flags & MDB_WRITEMAP)))
+ {
+ rc = mdb_page_unspill(mc->mc_txn, omp, &omp);
+ if (rc)
+ return rc;
+ level = 0; /* dirty in this txn or clean */
+ }
+ /* Is it dirty? */
+ if (omp->mp_flags & P_DIRTY) {
/* yes, overwrite it. Note in this case we don't
* bother to try shrinking the page if the new data
* is smaller than the overflow threshold.
else
memcpy(METADATA(omp), data->mv_data, data->mv_size);
goto done;
- } else {
- if ((rc2 = mdb_ovpage_free(mc, omp)) != MDB_SUCCESS)
- return rc2;
+ }
}
+ if ((rc2 = mdb_ovpage_free(mc, omp)) != MDB_SUCCESS)
+ return rc2;
} else if (NODEDSZ(leaf) == data->mv_size) {
/* same size, just replace it. Note that we could
* also reuse this node if the new data is smaller,
mdb_node_del(mc->mc_pg[mc->mc_top], mc->mc_ki[mc->mc_top], 0);
mc->mc_db->md_entries--;
} else {
- DPRINTF("inserting key at index %i", mc->mc_ki[mc->mc_top]);
+ DPRINTF(("inserting key at index %i", mc->mc_ki[mc->mc_top]));
insert = 1;
}
unsigned i = mc->mc_top;
MDB_page *mp = mc->mc_pg[i];
- if (mc->mc_flags & C_SUB)
- dbi--;
-
for (m2 = mc->mc_txn->mt_cursors[dbi]; m2; m2=m2->mc_next) {
if (mc->mc_flags & C_SUB)
m3 = &m2->mc_xcursor->mx_cursor;
xdata.mv_data = "";
leaf = NODEPTR(mc->mc_pg[mc->mc_top], mc->mc_ki[mc->mc_top]);
if (flags & MDB_CURRENT) {
- xflags = MDB_CURRENT;
+ xflags = MDB_CURRENT|MDB_NOSPILL;
} else {
mdb_xcursor_init1(mc, leaf);
- xflags = (flags & MDB_NODUPDATA) ? MDB_NOOVERWRITE : 0;
+ xflags = (flags & MDB_NODUPDATA) ?
+ MDB_NOOVERWRITE|MDB_NOSPILL : MDB_NOSPILL;
}
/* converted, write the original data first */
if (dkey.mv_size) {
for (m2 = mc->mc_txn->mt_cursors[mc->mc_dbi]; m2; m2=m2->mc_next) {
if (m2 == mc || m2->mc_snum < mc->mc_snum) continue;
+ if (!(m2->mc_flags & C_INITIALIZED)) continue;
if (m2->mc_pg[i] == mp && m2->mc_ki[i] == mc->mc_ki[i]) {
mdb_xcursor_init1(m2, leaf);
}
mc->mc_db->md_entries++;
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;
leaf = NODEPTR(mc->mc_pg[mc->mc_top], mc->mc_ki[mc->mc_top]);
goto more;
}
}
- /* let caller know how many succeeded, if any */
- data[1].mv_size = mcount;
}
}
done:
MDB_node *leaf;
int rc;
- if (F_ISSET(mc->mc_txn->mt_flags, MDB_TXN_RDONLY))
- return EACCES;
+ 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 (!(mc->mc_flags & C_INITIALIZED))
return EINVAL;
+ if (!(flags & MDB_NOSPILL) && (rc = mdb_page_spill(mc, NULL, NULL)))
+ return rc;
+
rc = mdb_cursor_touch(mc);
if (rc)
return rc;
leaf = NODEPTR(mc->mc_pg[mc->mc_top], mc->mc_ki[mc->mc_top]);
if (!IS_LEAF2(mc->mc_pg[mc->mc_top]) && F_ISSET(leaf->mn_flags, F_DUPDATA)) {
- if (flags != MDB_NODUPDATA) {
+ if (!(flags & MDB_NODUPDATA)) {
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, 0);
+ rc = mdb_cursor_del(&mc->mc_xcursor->mx_cursor, MDB_NOSPILL);
/* If sub-DB still has entries, we're done */
if (mc->mc_xcursor->mx_db.md_entries) {
if (leaf->mn_flags & F_SUBDATA) {
}
}
mc->mc_db->md_entries--;
+ mc->mc_flags |= C_DEL;
return rc;
}
/* otherwise fall thru and delete the sub-DB */
if ((rc = mdb_page_alloc(mc, num, &np)))
return rc;
- DPRINTF("allocated new mpage %zu, page size %u",
- np->mp_pgno, mc->mc_txn->mt_env->me_psize);
+ DPRINTF(("allocated new mpage %"Z"u, page size %u",
+ np->mp_pgno, mc->mc_txn->mt_env->me_psize));
np->mp_flags = flags | P_DIRTY;
np->mp_lower = PAGEHDRSZ;
np->mp_upper = mc->mc_txn->mt_env->me_psize;
{
unsigned int i;
size_t node_size = NODESIZE;
+ ssize_t room;
indx_t ofs;
MDB_node *node;
MDB_page *mp = mc->mc_pg[mc->mc_top];
assert(mp->mp_upper >= mp->mp_lower);
- DPRINTF("add to %s %spage %zu index %i, data size %zu key size %zu [%s]",
+ DPRINTF(("add to %s %spage %"Z"u index %i, data size %"Z"u key size %"Z"u [%s]",
IS_LEAF(mp) ? "leaf" : "branch",
IS_SUBP(mp) ? "sub-" : "",
mp->mp_pgno, indx, data ? data->mv_size : 0,
- key ? key->mv_size : 0, key ? DKEY(key) : NULL);
+ key ? key->mv_size : 0, key ? DKEY(key) : "null"));
if (IS_LEAF2(mp)) {
/* Move higher keys up one slot. */
return MDB_SUCCESS;
}
+ room = (ssize_t)SIZELEFT(mp) - (ssize_t)sizeof(indx_t);
if (key != NULL)
node_size += key->mv_size;
-
if (IS_LEAF(mp)) {
assert(data);
if (F_ISSET(flags, F_BIGDATA)) {
int ovpages = OVPAGES(data->mv_size, mc->mc_txn->mt_env->me_psize);
int rc;
/* Put data on overflow page. */
- DPRINTF("data size is %zu, node would be %zu, put data on overflow page",
- data->mv_size, node_size+data->mv_size);
- node_size += sizeof(pgno_t);
+ DPRINTF(("data size is %"Z"u, node would be %"Z"u, put data on overflow page",
+ data->mv_size, node_size+data->mv_size));
+ node_size += sizeof(pgno_t) + (node_size & 1);
+ if ((ssize_t)node_size > room)
+ goto full;
if ((rc = mdb_page_new(mc, P_OVERFLOW, ovpages, &ofp)))
return rc;
- DPRINTF("allocated overflow page %zu", ofp->mp_pgno);
+ DPRINTF(("allocated overflow page %"Z"u", ofp->mp_pgno));
flags |= F_BIGDATA;
+ goto update;
} else {
node_size += data->mv_size;
}
}
node_size += node_size & 1;
+ if ((ssize_t)node_size > room)
+ goto full;
- if (node_size + sizeof(indx_t) > SIZELEFT(mp)) {
- DPRINTF("not enough room in page %zu, 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);
- DPRINTF("node size = %zu", node_size);
- return MDB_PAGE_FULL;
- }
-
+update:
/* Move higher pointers up one slot. */
for (i = NUMKEYS(mp); i > indx; i--)
mp->mp_ptrs[i] = mp->mp_ptrs[i - 1];
}
return MDB_SUCCESS;
+
+full:
+ DPRINTF(("not enough room in page %"Z"u, got %u ptrs",
+ mp->mp_pgno, NUMKEYS(mp)));
+ DPRINTF(("upper-lower = %u - %u = %"Z"d", mp->mp_upper,mp->mp_lower,room));
+ DPRINTF(("node size = %"Z"u", node_size));
+ return MDB_PAGE_FULL;
}
/** Delete the specified node from a page.
{
pgno_t pgno;
COPY_PGNO(pgno, mp->mp_pgno);
- DPRINTF("delete node %u on %s page %zu", indx,
- IS_LEAF(mp) ? "leaf" : "branch", pgno);
+ DPRINTF(("delete node %u on %s page %"Z"u", indx,
+ IS_LEAF(mp) ? "leaf" : "branch", pgno));
}
#endif
assert(indx < NUMKEYS(mp));
mx->mx_cursor.mc_txn = mc->mc_txn;
mx->mx_cursor.mc_db = &mx->mx_db;
mx->mx_cursor.mc_dbx = &mx->mx_dbx;
- mx->mx_cursor.mc_dbi = mc->mc_dbi+1;
+ mx->mx_cursor.mc_dbi = mc->mc_dbi;
mx->mx_cursor.mc_dbflag = &mx->mx_dbflag;
mx->mx_cursor.mc_snum = 0;
mx->mx_cursor.mc_top = 0;
mx->mx_cursor.mc_flags = C_SUB;
+ mx->mx_dbx.md_name.mv_size = 0;
+ mx->mx_dbx.md_name.mv_data = NULL;
mx->mx_dbx.md_cmp = mc->mc_dbx->md_dcmp;
mx->mx_dbx.md_dcmp = NULL;
mx->mx_dbx.md_rel = mc->mc_dbx->md_rel;
memcpy(&mx->mx_db, NODEDATA(node), sizeof(MDB_db));
mx->mx_cursor.mc_pg[0] = 0;
mx->mx_cursor.mc_snum = 0;
+ mx->mx_cursor.mc_top = 0;
mx->mx_cursor.mc_flags = C_SUB;
} else {
MDB_page *fp = NODEDATA(node);
mx->mx_db.md_entries = NUMKEYS(fp);
COPY_PGNO(mx->mx_db.md_root, fp->mp_pgno);
mx->mx_cursor.mc_snum = 1;
- mx->mx_cursor.mc_flags = C_INITIALIZED|C_SUB;
mx->mx_cursor.mc_top = 0;
+ mx->mx_cursor.mc_flags = C_INITIALIZED|C_SUB;
mx->mx_cursor.mc_pg[0] = fp;
mx->mx_cursor.mc_ki[0] = 0;
if (mc->mc_db->md_flags & MDB_DUPFIXED) {
mx->mx_db.md_flags |= MDB_INTEGERKEY;
}
}
- DPRINTF("Sub-db %u for db %u root page %zu", mx->mx_cursor.mc_dbi, mc->mc_dbi,
- mx->mx_db.md_root);
- mx->mx_dbflag = DB_VALID | (F_ISSET(mc->mc_pg[mc->mc_top]->mp_flags, P_DIRTY) ?
- DB_DIRTY : 0);
- mx->mx_dbx.md_name.mv_data = NODEKEY(node);
- mx->mx_dbx.md_name.mv_size = node->mn_ksize;
+ DPRINTF(("Sub-db -%u root page %"Z"u", mx->mx_cursor.mc_dbi,
+ mx->mx_db.md_root));
+ 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
static void
mdb_cursor_init(MDB_cursor *mc, MDB_txn *txn, MDB_dbi dbi, MDB_xcursor *mx)
{
+ mc->mc_next = NULL;
mc->mc_backup = NULL;
mc->mc_dbi = dbi;
mc->mc_txn = txn;
if (txn == NULL || ret == NULL || dbi >= txn->mt_numdbs || !(txn->mt_dbflags[dbi] & DB_VALID))
return EINVAL;
+ if (txn->mt_flags & MDB_TXN_ERROR)
+ return MDB_BAD_TXN;
+
/* Allow read access to the freelist */
if (!dbi && !F_ISSET(txn->mt_flags, MDB_TXN_RDONLY))
return EINVAL;
if (mc == NULL || countp == NULL)
return EINVAL;
- if (!(mc->mc_db->md_flags & MDB_DUPSORT))
- return EINVAL;
+ if (mc->mc_xcursor == NULL)
+ return MDB_INCOMPATIBLE;
leaf = NODEPTR(mc->mc_pg[mc->mc_top], mc->mc_ki[mc->mc_top]);
if (!F_ISSET(leaf->mn_flags, F_DUPDATA)) {
char kbuf2[(MDB_MAXKEYSIZE*2+1)];
k2.mv_data = NODEKEY(node);
k2.mv_size = node->mn_ksize;
- DPRINTF("update key %u (ofs %u) [%s] to [%s] on page %zu",
+ DPRINTF(("update key %u (ofs %u) [%s] to [%s] on page %"Z"u",
indx, ptr,
mdb_dkey(&k2, kbuf2),
DKEY(key),
- mp->mp_pgno);
+ mp->mp_pgno));
}
#endif
if (delta > 0 && SIZELEFT(mp) < delta) {
pgno_t pgno;
/* not enough space left, do a delete and split */
- DPRINTF("Not enough room, delta = %d, splitting...", delta);
+ DPRINTF(("Not enough room, delta = %d, splitting...", delta));
pgno = NODEPGNO(node);
mdb_node_del(mc->mc_pg[mc->mc_top], mc->mc_ki[mc->mc_top], 0);
return mdb_page_split(mc, key, NULL, pgno, MDB_SPLIT_REPLACE);
return rc;
}
- DPRINTF("moving %s node %u [%s] on page %zu to node %u on page %zu",
+ DPRINTF(("moving %s node %u [%s] on page %"Z"u to node %u on page %"Z"u",
IS_LEAF(csrc->mc_pg[csrc->mc_top]) ? "leaf" : "branch",
csrc->mc_ki[csrc->mc_top],
DKEY(&key),
csrc->mc_pg[csrc->mc_top]->mp_pgno,
- cdst->mc_ki[cdst->mc_top], cdst->mc_pg[cdst->mc_top]->mp_pgno);
+ cdst->mc_ki[cdst->mc_top], cdst->mc_pg[cdst->mc_top]->mp_pgno));
/* Add the node to the destination page.
*/
MDB_dbi dbi = csrc->mc_dbi;
MDB_page *mp = csrc->mc_pg[csrc->mc_top];
- if (csrc->mc_flags & C_SUB)
- dbi--;
-
for (m2 = csrc->mc_txn->mt_cursors[dbi]; m2; m2=m2->mc_next) {
if (csrc->mc_flags & C_SUB)
m3 = &m2->mc_xcursor->mx_cursor;
key.mv_size = NODEKSZ(srcnode);
key.mv_data = NODEKEY(srcnode);
}
- DPRINTF("update separator for source page %zu to [%s]",
- csrc->mc_pg[csrc->mc_top]->mp_pgno, DKEY(&key));
+ DPRINTF(("update separator for source page %"Z"u to [%s]",
+ csrc->mc_pg[csrc->mc_top]->mp_pgno, DKEY(&key)));
mdb_cursor_copy(csrc, &mn);
mn.mc_snum--;
mn.mc_top--;
key.mv_size = NODEKSZ(srcnode);
key.mv_data = NODEKEY(srcnode);
}
- DPRINTF("update separator for destination page %zu to [%s]",
- cdst->mc_pg[cdst->mc_top]->mp_pgno, DKEY(&key));
+ DPRINTF(("update separator for destination page %"Z"u to [%s]",
+ cdst->mc_pg[cdst->mc_top]->mp_pgno, DKEY(&key)));
mdb_cursor_copy(cdst, &mn);
mn.mc_snum--;
mn.mc_top--;
MDB_val key, data;
unsigned nkeys;
- DPRINTF("merging page %zu into %zu", csrc->mc_pg[csrc->mc_top]->mp_pgno,
- cdst->mc_pg[cdst->mc_top]->mp_pgno);
+ DPRINTF(("merging page %"Z"u into %"Z"u", csrc->mc_pg[csrc->mc_top]->mp_pgno,
+ cdst->mc_pg[cdst->mc_top]->mp_pgno));
assert(csrc->mc_snum > 1); /* can't merge root page */
assert(cdst->mc_snum > 1);
}
}
- DPRINTF("dst page %zu 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);
+ 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));
/* Unlink the src page from parent and add to free list.
*/
MDB_dbi dbi = csrc->mc_dbi;
MDB_page *mp = cdst->mc_pg[cdst->mc_top];
- if (csrc->mc_flags & C_SUB)
- dbi--;
-
for (m2 = csrc->mc_txn->mt_cursors[dbi]; m2; m2=m2->mc_next) {
if (csrc->mc_flags & C_SUB)
m3 = &m2->mc_xcursor->mx_cursor;
{
pgno_t pgno;
COPY_PGNO(pgno, mc->mc_pg[mc->mc_top]->mp_pgno);
- DPRINTF("rebalancing %s page %zu (has %u keys, %.1f%% full)",
+ DPRINTF(("rebalancing %s page %"Z"u (has %u keys, %.1f%% full)",
IS_LEAF(mc->mc_pg[mc->mc_top]) ? "leaf" : "branch",
- pgno, NUMKEYS(mc->mc_pg[mc->mc_top]), (float)PAGEFILL(mc->mc_txn->mt_env, mc->mc_pg[mc->mc_top]) / 10);
+ pgno, NUMKEYS(mc->mc_pg[mc->mc_top]),
+ (float)PAGEFILL(mc->mc_txn->mt_env, mc->mc_pg[mc->mc_top]) / 10));
}
#endif
#if MDB_DEBUG
pgno_t pgno;
COPY_PGNO(pgno, mc->mc_pg[mc->mc_top]->mp_pgno);
- DPRINTF("no need to rebalance page %zu, above fill threshold",
- pgno);
+ DPRINTF(("no need to rebalance page %"Z"u, above fill threshold",
+ pgno));
#endif
return MDB_SUCCESS;
}
/* Adjust cursors pointing to mp */
mc->mc_snum = 0;
mc->mc_top = 0;
+ mc->mc_flags &= ~C_INITIALIZED;
{
MDB_cursor *m2, *m3;
MDB_dbi dbi = mc->mc_dbi;
- if (mc->mc_flags & C_SUB)
- dbi--;
-
for (m2 = mc->mc_txn->mt_cursors[dbi]; m2; m2=m2->mc_next) {
if (mc->mc_flags & C_SUB)
m3 = &m2->mc_xcursor->mx_cursor;
if (m3->mc_pg[0] == mp) {
m3->mc_snum = 0;
m3->mc_top = 0;
+ m3->mc_flags &= ~C_INITIALIZED;
}
}
}
MDB_cursor *m2, *m3;
MDB_dbi dbi = mc->mc_dbi;
- if (mc->mc_flags & C_SUB)
- dbi--;
-
for (m2 = mc->mc_txn->mt_cursors[dbi]; m2; m2=m2->mc_next) {
if (mc->mc_flags & C_SUB)
m3 = &m2->mc_xcursor->mx_cursor;
mc->mc_ki[mc->mc_top] = 0;
}
- DPRINTF("found neighbor page %zu (%u keys, %.1f%% full)",
- mn.mc_pg[mn.mc_top]->mp_pgno, NUMKEYS(mn.mc_pg[mn.mc_top]), (float)PAGEFILL(mc->mc_txn->mt_env, mn.mc_pg[mn.mc_top]) / 10);
+ DPRINTF(("found neighbor page %"Z"u (%u keys, %.1f%% full)",
+ mn.mc_pg[mn.mc_top]->mp_pgno, NUMKEYS(mn.mc_pg[mn.mc_top]),
+ (float)PAGEFILL(mc->mc_txn->mt_env, mn.mc_pg[mn.mc_top]) / 10));
/* If the neighbor page is above threshold and has enough keys,
* move one key from it. Otherwise we should try to merge them.
else {
if (mc->mc_ki[ptop] == 0)
rc = mdb_page_merge(&mn, mc);
- else
+ else {
+ 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);
}
return rc;
int rc;
MDB_page *mp;
indx_t ki;
+ unsigned int nkeys;
mp = mc->mc_pg[mc->mc_top];
ki = mc->mc_ki[mc->mc_top];
rc = mdb_rebalance(mc);
if (rc != MDB_SUCCESS)
mc->mc_txn->mt_flags |= MDB_TXN_ERROR;
- /* if mc points past last node in page, invalidate */
- else if (mc->mc_ki[mc->mc_top] >= NUMKEYS(mc->mc_pg[mc->mc_top]))
- mc->mc_flags &= ~(C_INITIALIZED|C_EOF);
-
- {
- /* Adjust other cursors pointing to mp */
+ else {
MDB_cursor *m2;
- unsigned int nkeys;
MDB_dbi dbi = mc->mc_dbi;
mp = mc->mc_pg[mc->mc_top];
nkeys = NUMKEYS(mp);
+
+ /* if mc points past last node in page, find next sibling */
+ if (mc->mc_ki[mc->mc_top] >= nkeys)
+ mdb_cursor_sibling(mc, 1);
+
+ /* Adjust other cursors pointing to mp */
for (m2 = mc->mc_txn->mt_cursors[dbi]; m2; m2=m2->mc_next) {
if (m2 == mc)
continue;
if (!(m2->mc_flags & C_INITIALIZED))
continue;
if (m2->mc_pg[mc->mc_top] == mp) {
- if (m2->mc_ki[mc->mc_top] > ki)
- m2->mc_ki[mc->mc_top]--;
+ if (m2->mc_ki[mc->mc_top] >= ki) {
+ m2->mc_flags |= C_DEL;
+ if (m2->mc_ki[mc->mc_top] > ki)
+ m2->mc_ki[mc->mc_top]--;
+ }
if (m2->mc_ki[mc->mc_top] >= nkeys)
- m2->mc_flags &= ~(C_INITIALIZED|C_EOF);
+ mdb_cursor_sibling(m2, 1);
}
}
+ mc->mc_flags |= C_DEL;
}
return rc;
assert(key != NULL);
- DPRINTF("====> delete db %u key [%s]", dbi, DKEY(key));
+ DPRINTF(("====> delete db %u key [%s]", dbi, DKEY(key)));
if (txn == NULL || !dbi || dbi >= txn->mt_numdbs || !(txn->mt_dbflags[dbi] & DB_VALID))
return EINVAL;
- if (F_ISSET(txn->mt_flags, MDB_TXN_RDONLY)) {
- return EACCES;
- }
+ if (txn->mt_flags & (MDB_TXN_RDONLY|MDB_TXN_ERROR))
+ return (txn->mt_flags & MDB_TXN_RDONLY) ? EACCES : MDB_BAD_TXN;
- if (key->mv_size == 0 || key->mv_size > MDB_MAXKEYSIZE) {
- return EINVAL;
+ if (key->mv_size > MDB_MAXKEYSIZE) {
+ return MDB_BAD_VALSIZE;
}
mdb_cursor_init(&mc, txn, dbi, &mx);
exact = 0;
+ if (!F_ISSET(txn->mt_dbs[dbi].md_flags, MDB_DUPSORT)) {
+ /* must ignore any data */
+ data = NULL;
+ }
if (data) {
op = MDB_GET_BOTH;
rdata = *data;
* run out of space, triggering a split. We need this
* cursor to be consistent until the end of the rebalance.
*/
+ mc.mc_flags |= C_UNTRACK;
mc.mc_next = txn->mt_cursors[dbi];
txn->mt_cursors[dbi] = &mc;
rc = mdb_cursor_del(&mc, data ? 0 : MDB_NODUPDATA);
mp = mc->mc_pg[mc->mc_top];
newindx = mc->mc_ki[mc->mc_top];
- DPRINTF("-----> splitting %s page %zu and adding [%s] at index %i",
+ DPRINTF(("-----> splitting %s page %"Z"u and adding [%s] at index %i",
IS_LEAF(mp) ? "leaf" : "branch", mp->mp_pgno,
- DKEY(newkey), mc->mc_ki[mc->mc_top]);
+ DKEY(newkey), mc->mc_ki[mc->mc_top]));
/* Create a right sibling. */
if ((rc = mdb_page_new(mc, mp->mp_flags, 1, &rp)))
return rc;
- DPRINTF("new right sibling: page %zu", rp->mp_pgno);
+ DPRINTF(("new right sibling: page %"Z"u", rp->mp_pgno));
if (mc->mc_snum < 2) {
if ((rc = mdb_page_new(mc, P_BRANCH, 1, &pp)))
mc->mc_pg[0] = pp;
mc->mc_ki[0] = 0;
mc->mc_db->md_root = pp->mp_pgno;
- DPRINTF("root split! new root = %zu", pp->mp_pgno);
+ DPRINTF(("root split! new root = %"Z"u", pp->mp_pgno));
mc->mc_db->md_depth++;
new_root = 1;
ptop = 0;
} else {
ptop = mc->mc_top-1;
- DPRINTF("parent branch page is %zu", mc->mc_pg[ptop]->mp_pgno);
+ DPRINTF(("parent branch page is %"Z"u", mc->mc_pg[ptop]->mp_pgno));
}
mc->mc_flags |= C_SPLITTING;
}
newsep:
- DPRINTF("separator is [%s]", DKEY(&sepkey));
+ DPRINTF(("separator is [%s]", DKEY(&sepkey)));
/* Copy separator key to the parent.
*/
MDB_dbi dbi = mc->mc_dbi;
int fixup = NUMKEYS(mp);
- if (mc->mc_flags & C_SUB)
- dbi--;
-
for (m2 = mc->mc_txn->mt_cursors[dbi]; m2; m2=m2->mc_next) {
if (mc->mc_flags & C_SUB)
m3 = &m2->mc_xcursor->mx_cursor;
m3 = m2;
if (m3 == mc)
continue;
- if (!(m3->mc_flags & C_INITIALIZED))
+ if (!(m2->mc_flags & m3->mc_flags & C_INITIALIZED))
continue;
if (m3->mc_flags & C_SPLITTING)
continue;
m3->mc_snum++;
m3->mc_top++;
}
- if (m3->mc_pg[mc->mc_top] == mp) {
+ if (m3->mc_top >= mc->mc_top && m3->mc_pg[mc->mc_top] == mp) {
if (m3->mc_ki[mc->mc_top] >= newindx && !(nflags & MDB_SPLIT_REPLACE))
m3->mc_ki[mc->mc_top]++;
if (m3->mc_ki[mc->mc_top] >= fixup) {
m3->mc_ki[mc->mc_top] -= fixup;
m3->mc_ki[ptop] = mn.mc_ki[ptop];
}
- } else if (!did_split && m3->mc_pg[ptop] == mc->mc_pg[ptop] &&
+ } else if (!did_split && m3->mc_top >= ptop && m3->mc_pg[ptop] == mc->mc_pg[ptop] &&
m3->mc_ki[ptop] >= mc->mc_ki[ptop]) {
m3->mc_ki[ptop]++;
}
if (txn == NULL || !dbi || dbi >= txn->mt_numdbs || !(txn->mt_dbflags[dbi] & DB_VALID))
return EINVAL;
- if (F_ISSET(txn->mt_flags, MDB_TXN_RDONLY)) {
- return EACCES;
- }
-
- if (key->mv_size == 0 || key->mv_size > MDB_MAXKEYSIZE) {
- return EINVAL;
- }
-
if ((flags & (MDB_NOOVERWRITE|MDB_NODUPDATA|MDB_RESERVE|MDB_APPEND|MDB_APPENDDUP)) != flags)
return EINVAL;
arg->me_mapaddr = (env->me_flags & MDB_FIXEDMAP) ? env->me_map : 0;
arg->me_mapsize = env->me_mapsize;
arg->me_maxreaders = env->me_maxreaders;
- arg->me_numreaders = env->me_numreaders;
+
+ /* me_numreaders may be zero if this process never used any readers. Use
+ * the shared numreader count if it exists.
+ */
+ arg->me_numreaders = env->me_txns ? env->me_txns->mti_numreaders : env->me_numreaders;
+
arg->me_last_pgno = env->me_metas[toggle]->mm_last_pg;
arg->me_last_txnid = env->me_metas[toggle]->mm_txnid;
return MDB_SUCCESS;
if ((flags & VALID_FLAGS) != flags)
return EINVAL;
+ if (txn->mt_flags & MDB_TXN_ERROR)
+ return MDB_BAD_TXN;
/* main DB? */
if (!name) {
/* make sure this is actually a DB */
MDB_node *node = NODEPTR(mc.mc_pg[mc.mc_top], mc.mc_ki[mc.mc_top]);
if (!(node->mn_flags & F_SUBDATA))
- return EINVAL;
+ return MDB_INCOMPATIBLE;
} else if (rc == MDB_NOTFOUND && (flags & MDB_CREATE)) {
/* Create if requested */
MDB_db dummy;
txn->mt_dbflags[slot] = dbflag;
memcpy(&txn->mt_dbs[slot], data.mv_data, sizeof(MDB_db));
*dbi = slot;
- txn->mt_env->me_dbflags[slot] = txn->mt_dbs[slot].md_flags;
mdb_default_cmp(txn, slot);
if (!unused) {
txn->mt_numdbs++;
free(ptr);
}
+int mdb_dbi_flags(MDB_txn *txn, MDB_dbi dbi, unsigned int *flags)
+{
+ /* We could return the flags for the FREE_DBI too but what's the point? */
+ if (txn == NULL || dbi < MAIN_DBI || dbi >= txn->mt_numdbs)
+ return EINVAL;
+ *flags = txn->mt_dbs[dbi].md_flags & PERSISTENT_FLAGS;
+ return MDB_SUCCESS;
+}
+
/** Add all the DB's pages to the free list.
* @param[in] mc Cursor on the DB to free.
* @param[in] subs non-Zero to check for sub-DBs in this DB.
{
int rc;
- rc = mdb_page_search(mc, NULL, 0);
+ rc = mdb_page_search(mc, NULL, MDB_PS_FIRST);
if (rc == MDB_SUCCESS) {
MDB_txn *txn = mc->mc_txn;
MDB_node *ni;
return MDB_SUCCESS;
}
+int mdb_env_get_maxkeysize(MDB_env *env)
+{
+ return MDB_MAXKEYSIZE;
+}
+
+int mdb_reader_list(MDB_env *env, MDB_msg_func *func, void *ctx)
+{
+ unsigned int i, rdrs;
+ MDB_reader *mr;
+ char buf[64];
+ int first = 1;
+
+ if (!env || !func)
+ return -1;
+ if (!env->me_txns) {
+ return func("(no reader locks)\n", ctx);
+ }
+ rdrs = env->me_txns->mti_numreaders;
+ mr = env->me_txns->mti_readers;
+ for (i=0; i<rdrs; i++) {
+ if (mr[i].mr_pid) {
+ size_t tid;
+ int rc;
+ tid = mr[i].mr_tid;
+ if (mr[i].mr_txnid == (txnid_t)-1) {
+ sprintf(buf, "%10d %"Z"x -\n", mr[i].mr_pid, tid);
+ } else {
+ sprintf(buf, "%10d %"Z"x %"Z"u\n", mr[i].mr_pid, tid, mr[i].mr_txnid);
+ }
+ if (first) {
+ first = 0;
+ func(" pid thread txnid\n", ctx);
+ }
+ rc = func(buf, ctx);
+ if (rc < 0)
+ return rc;
+ }
+ }
+ if (first) {
+ func("(no active readers)\n", ctx);
+ }
+ return 0;
+}
+
+/** Insert pid into list if not already present.
+ * return -1 if already present.
+ */
+static int mdb_pid_insert(pid_t *ids, pid_t pid)
+{
+ /* binary search of pid in list */
+ unsigned base = 0;
+ unsigned cursor = 1;
+ int val = 0;
+ unsigned n = ids[0];
+
+ while( 0 < n ) {
+ unsigned pivot = n >> 1;
+ cursor = base + pivot + 1;
+ val = pid - ids[cursor];
+
+ if( val < 0 ) {
+ n = pivot;
+
+ } else if ( val > 0 ) {
+ base = cursor;
+ n -= pivot + 1;
+
+ } else {
+ /* found, so it's a duplicate */
+ return -1;
+ }
+ }
+
+ if( val > 0 ) {
+ ++cursor;
+ }
+ ids[0]++;
+ for (n = ids[0]; n > cursor; n--)
+ ids[n] = ids[n-1];
+ ids[n] = pid;
+ return 0;
+}
+
+int mdb_reader_check(MDB_env *env, int *dead)
+{
+ unsigned int i, j, rdrs;
+ MDB_reader *mr;
+ pid_t *pids, pid;
+ int count = 0;
+
+ if (!env)
+ return EINVAL;
+ if (dead)
+ *dead = 0;
+ if (!env->me_txns)
+ return MDB_SUCCESS;
+ rdrs = env->me_txns->mti_numreaders;
+ pids = malloc((rdrs+1) * sizeof(pid_t));
+ if (!pids)
+ return ENOMEM;
+ pids[0] = 0;
+ mr = env->me_txns->mti_readers;
+ j = 0;
+ for (i=0; i<rdrs; i++) {
+ if (mr[i].mr_pid && mr[i].mr_pid != env->me_pid) {
+ pid = mr[i].mr_pid;
+ if (mdb_pid_insert(pids, pid) == 0) {
+ if (!mdb_reader_pid(env, Pidcheck, pid)) {
+ LOCK_MUTEX_R(env);
+ /* Recheck, a new process may have reused pid */
+ if (!mdb_reader_pid(env, Pidcheck, pid)) {
+ for (j=i; j<rdrs; j++)
+ if (mr[j].mr_pid == pid) {
+ DPRINTF(("clear stale reader pid %u txn %"Z"d",
+ (unsigned) pid, mr[j].mr_txnid));
+ mr[j].mr_pid = 0;
+ count++;
+ }
+ }
+ UNLOCK_MUTEX_R(env);
+ }
+ }
+ }
+ }
+ free(pids);
+ if (dead)
+ *dead = count;
+ return MDB_SUCCESS;
+}
/** @} */
projects / openldap / blobdiff