* BerkeleyDB API, but much simplified.
*/
/*
- * Copyright 2011-2016 Howard Chu, Symas Corp.
+ * Copyright 2011-2017 Howard Chu, Symas Corp.
* All rights reserved.
*
* Redistribution and use in source and binary forms, with or without
#include <resolv.h> /* defines BYTE_ORDER on HPUX and Solaris */
#endif
-#if defined(__APPLE__) || defined (BSD)
+#if defined(__APPLE__) || defined (BSD) || defined(__FreeBSD_kernel__)
# if !(defined(MDB_USE_POSIX_MUTEX) || defined(MDB_USE_POSIX_SEM))
# define MDB_USE_SYSV_SEM 1
# endif
#ifndef _WIN32
#include <pthread.h>
+#include <signal.h>
#ifdef MDB_USE_POSIX_SEM
# define MDB_USE_HASH 1
#include <semaphore.h>
#define mdb_mutex_consistent(mutex) 0
#else /* MDB_USE_POSIX_MUTEX: */
- /** Shared mutex/semaphore as it is stored (mdb_mutex_t), and as
- * local variables keep it (mdb_mutexref_t).
+ /** Shared mutex/semaphore as the original is stored.
*
- * An mdb_mutex_t can be assigned to an mdb_mutexref_t. They can
- * be the same, or an array[size 1] and a pointer.
- * @{
+ * Not for copies. Instead it can be assigned to an #mdb_mutexref_t.
+ * When mdb_mutexref_t is a pointer and mdb_mutex_t is not, then it
+ * is array[size 1] so it can be assigned to the pointer.
*/
-typedef pthread_mutex_t mdb_mutex_t[1], *mdb_mutexref_t;
- /* @} */
+typedef pthread_mutex_t mdb_mutex_t[1];
+ /** Reference to an #mdb_mutex_t */
+typedef pthread_mutex_t *mdb_mutexref_t;
/** Lock the reader or writer mutex.
* Returns 0 or a code to give #mdb_mutex_failed(), as in #LOCK_MUTEX().
*/
#define Z MDB_FMT_Z /**< printf/scanf format modifier for size_t */
#define Yu MDB_PRIy(u) /**< printf format for #mdb_size_t */
-#define Yd MDB_PRIy(d) /**< printf format for "signed #mdb_size_t" */
+#define Yd MDB_PRIy(d) /**< printf format for 'signed #mdb_size_t' */
#if defined(_WIN32) || defined(MDB_USE_POSIX_SEM)
#define MNAME_LEN 32
/** The version number for a database's datafile format. */
#define MDB_DATA_VERSION ((MDB_DEVEL) ? 999 : 1)
/** The version number for a database's lockfile format. */
-#define MDB_LOCK_VERSION ((MDB_DEVEL) ? 999 : 1)
+#define MDB_LOCK_VERSION ((MDB_DEVEL) ? 999 : 2)
/** @brief The max size of a key we can write, or 0 for computed max.
*
uint32_t mtb_magic;
/** Format of this lock file. Must be set to #MDB_LOCK_FORMAT. */
uint32_t mtb_format;
+ /** The ID of the last transaction committed to the database.
+ * This is recorded here only for convenience; the value can always
+ * be determined by reading the main database meta pages.
+ */
+ volatile txnid_t mtb_txnid;
+ /** The number of slots that have been used in the reader table.
+ * This always records the maximum count, it is not decremented
+ * when readers release their slots.
+ */
+ volatile unsigned mtb_numreaders;
#if defined(_WIN32) || defined(MDB_USE_POSIX_SEM)
char mtb_rmname[MNAME_LEN];
#elif defined(MDB_USE_SYSV_SEM)
*/
mdb_mutex_t mtb_rmutex;
#endif
- /** The ID of the last transaction committed to the database.
- * This is recorded here only for convenience; the value can always
- * be determined by reading the main database meta pages.
- */
- volatile txnid_t mtb_txnid;
- /** The number of slots that have been used in the reader table.
- * This always records the maximum count, it is not decremented
- * when readers release their slots.
- */
- volatile unsigned mtb_numreaders;
} MDB_txbody;
/** The actual reader table definition. */
/** Header for a single key/data pair within a page.
* Used in pages of type #P_BRANCH and #P_LEAF without #P_LEAF2.
* We guarantee 2-byte alignment for 'MDB_node's.
+ *
+ * #mn_lo and #mn_hi are used for data size on leaf nodes, and for child
+ * pgno on branch nodes. On 64 bit platforms, #mn_flags is also used
+ * for pgno. (Branch nodes have no flags). Lo and hi are in host byte
+ * order in case some accesses can be optimized to 32-bit word access.
+ *
+ * Leaf node flags describe node contents. #F_BIGDATA says the node's
+ * data part is the page number of an overflow page with actual data.
+ * #F_DUPDATA and #F_SUBDATA can be combined giving duplicate data in
+ * a sub-page/sub-database, and named databases (just #F_SUBDATA).
*/
typedef struct MDB_node {
- /** lo and hi are used for data size on leaf nodes and for
- * child pgno on branch nodes. On 64 bit platforms, flags
- * is also used for pgno. (Branch nodes have no flags).
- * They are in host byte order in case that lets some
- * accesses be optimized into a 32-bit word access.
- */
+ /** part of data size or pgno
+ * @{ */
#if BYTE_ORDER == LITTLE_ENDIAN
- unsigned short mn_lo, mn_hi; /**< part of data size or pgno */
+ unsigned short mn_lo, mn_hi;
#else
unsigned short mn_hi, mn_lo;
#endif
+ /** @} */
/** @defgroup mdb_node Node Flags
* @ingroup internal
* Flags for node headers.
#else
void *mm_address; /**< address for fixed mapping */
#endif
- pgno_t mm_mapsize; /**< size of mmap region */
+ mdb_size_t mm_mapsize; /**< size of mmap region */
MDB_db mm_dbs[CORE_DBS]; /**< first is free space, 2nd is main db */
/** The size of pages used in this DB */
#define mm_psize mm_dbs[FREE_DBI].md_pad
* @ingroup internal
* @{
*/
-#define DB_DIRTY 0x01 /**< DB was modified or is DUPSORT data */
+#define DB_DIRTY 0x01 /**< DB was written in this txn */
#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 */
#define DB_USRVALID 0x10 /**< As #DB_VALID, but not set for #FREE_DBI */
+#define DB_DUPDATA 0x20 /**< DB is #MDB_DUPSORT data */
/** @} */
/** In write txns, array of cursors for each DB */
MDB_cursor **mt_cursors;
/** Allocate memory for a page.
* Re-use old malloc'd pages first for singletons, otherwise just malloc.
+ * Set #MDB_TXN_ERROR on failure.
*/
static MDB_page *
mdb_page_malloc(MDB_txn *txn, unsigned num)
mdb_cursor_unref(MDB_cursor *mc)
{
int i;
- if (!mc->mc_snum || !mc->mc_pg[0] || IS_SUBP(mc->mc_pg[0]))
- return;
- for (i=0; i<mc->mc_snum; i++)
- mdb_page_unref(mc->mc_txn, mc->mc_pg[i]);
- if (mc->mc_ovpg) {
- mdb_page_unref(mc->mc_txn, mc->mc_ovpg);
- mc->mc_ovpg = 0;
+ if (mc->mc_txn->mt_rpages[0].mid) {
+ if (!mc->mc_snum || !mc->mc_pg[0] || IS_SUBP(mc->mc_pg[0]))
+ return;
+ for (i=0; i<mc->mc_snum; i++)
+ mdb_page_unref(mc->mc_txn, mc->mc_pg[i]);
+ if (mc->mc_ovpg) {
+ mdb_page_unref(mc->mc_txn, mc->mc_ovpg);
+ mc->mc_ovpg = 0;
+ }
}
mc->mc_snum = mc->mc_top = 0;
mc->mc_pg[0] = NULL;
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;
+ /* Mark pages seen by cursors: First m0, then tracked cursors */
+ for (i = txn->mt_numdbs;; ) {
+ if (mc->mc_flags & C_INITIALIZED) {
for (m3 = mc;; m3 = &mx->mx_cursor) {
mp = NULL;
for (j=0; j<m3->mc_snum; j++) {
break;
}
}
- if (i == 0)
- break;
+ mc = mc->mc_next;
+ for (; !mc || mc == m0; mc = txn->mt_cursors[--i])
+ if (i == 0)
+ goto mark_done;
}
+mark_done:
if (all) {
/* Mark dirty root pages */
for (i=0; i<txn->mt_numdbs; i++) {
}
/** Allocate page numbers and memory for writing. Maintain me_pglast,
- * me_pghead and mt_next_pgno.
+ * me_pghead and mt_next_pgno. Set #MDB_TXN_ERROR on failure.
*
* If there are free pages available from older transactions, they
* are re-used first. Otherwise allocate a new page at mt_next_pgno.
np = m2.mc_pg[m2.mc_top];
leaf = NODEPTR(np, m2.mc_ki[m2.mc_top]);
if ((rc = mdb_node_read(&m2, leaf, &data)) != MDB_SUCCESS)
- return rc;
+ goto fail;
idl = (MDB_ID *) data.mv_data;
i = idl[0];
}
/** Touch a page: make it dirty and re-insert into tree with updated pgno.
+ * Set #MDB_TXN_ERROR on failure.
* @param[in] mc cursor pointing to the page to be touched
* @return 0 on success, non-zero on failure.
*/
}
}
-/** Push a page onto the top of the cursor's stack. */
+/** Push a page onto the top of the cursor's stack.
+ * Set #MDB_TXN_ERROR on failure.
+ */
static int
mdb_cursor_push(MDB_cursor *mc, MDB_page *mp)
{
#endif
/** Find the address of the page corresponding to a given page number.
+ * Set #MDB_TXN_ERROR on failure.
* @param[in] mc the cursor accessing the page.
* @param[in] pgno the page number for the page to retrieve.
* @param[out] ret address of a pointer where the page's address will be stored.
{
#ifdef MDB_VL32
int rc = mdb_rpage_get(txn, pgno, &p);
- if (rc)
+ if (rc) {
+ txn->mt_flags |= MDB_TXN_ERROR;
return rc;
+ }
#else
MDB_env *env = txn->mt_env;
p = (MDB_page *)(env->me_map + env->me_psize * pgno);
if (flags & (MDB_PS_FIRST|MDB_PS_LAST)) {
i = 0;
- if (flags & MDB_PS_LAST)
+ if (flags & MDB_PS_LAST) {
i = NUMKEYS(mp) - 1;
+ /* if already init'd, see if we're already in right place */
+ if (mc->mc_flags & C_INITIALIZED) {
+ if (mc->mc_ki[mc->mc_top] == i) {
+ mc->mc_top = mc->mc_snum++;
+ mp = mc->mc_pg[mc->mc_top];
+ goto ready;
+ }
+ }
+ }
} else {
int exact;
node = mdb_node_search(mc, key, &exact);
if ((rc = mdb_cursor_push(mc, mp)))
return rc;
+ready:
if (flags & MDB_PS_MODIFY) {
if ((rc = mdb_page_touch(mc)) != 0)
return rc;
MDB_node *leaf;
int rc;
- if ((mc->mc_flags & C_EOF) ||
- ((mc->mc_flags & C_DEL) && op == MDB_NEXT_DUP)) {
+ if ((mc->mc_flags & C_DEL && op == MDB_NEXT_DUP))
return MDB_NOTFOUND;
- }
+
if (!(mc->mc_flags & C_INITIALIZED))
return mdb_cursor_first(mc, key, data);
mp = mc->mc_pg[mc->mc_top];
+ if (mc->mc_flags & C_EOF) {
+ if (mc->mc_ki[mc->mc_top] >= NUMKEYS(mp)-1)
+ return MDB_NOTFOUND;
+ mc->mc_flags ^= C_EOF;
+ }
+
if (mc->mc_db->md_flags & MDB_DUPSORT) {
leaf = NODEPTR(mp, mc->mc_ki[mc->mc_top]);
if (F_ISSET(leaf->mn_flags, F_DUPDATA)) {
} else
mc->mc_ki[mc->mc_top]--;
- mc->mc_flags &= ~C_EOF;
-
DPRINTF(("==> cursor points to page %"Yu" with %u keys, key index %u",
mdb_dbg_pgno(mp), NUMKEYS(mp), mc->mc_ki[mc->mc_top]));
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) {
- rc = mdb_page_search(mc, NULL, MDB_PS_LAST);
- if (rc != MDB_SUCCESS)
- return rc;
- }
- mdb_cassert(mc, IS_LEAF(mc->mc_pg[mc->mc_top]));
-
+ if (!(mc->mc_flags & C_INITIALIZED) || mc->mc_top) {
+ rc = mdb_page_search(mc, NULL, MDB_PS_LAST);
+ if (rc != MDB_SUCCESS)
+ return rc;
}
+ mdb_cassert(mc, IS_LEAF(mc->mc_pg[mc->mc_top]));
+
mc->mc_ki[mc->mc_top] = NUMKEYS(mc->mc_pg[mc->mc_top]) - 1;
mc->mc_flags |= C_INITIALIZED|C_EOF;
leaf = NODEPTR(mc->mc_pg[mc->mc_top], mc->mc_ki[mc->mc_top]);
{
int rc = MDB_SUCCESS;
- if (mc->mc_dbi >= CORE_DBS && !(*mc->mc_dbflag & DB_DIRTY)) {
+ if (mc->mc_dbi >= CORE_DBS && !(*mc->mc_dbflag & (DB_DIRTY|DB_DUPDATA))) {
+ /* Touch DB record of named DB */
MDB_cursor mc2;
MDB_xcursor mcx;
if (TXN_DBI_CHANGED(mc->mc_txn, mc->mc_dbi))
}
/** Allocate and initialize new pages for a database.
+ * Set #MDB_TXN_ERROR on failure.
* @param[in] mc a cursor on the database being added to.
* @param[in] flags flags defining what type of page is being allocated.
* @param[in] num the number of pages to allocate. This is usually 1,
}
/** Add a node to the page pointed to by the cursor.
+ * Set #MDB_TXN_ERROR on failure.
* @param[in] mc The cursor for this operation.
* @param[in] indx The index on the page where the new node should be added.
* @param[in] key The key for the new node.
}
DPRINTF(("Sub-db -%u root page %"Yu, mx->mx_cursor.mc_dbi,
mx->mx_db.md_root));
- mx->mx_dbflag = DB_VALID|DB_USRVALID|DB_DIRTY; /* DB_DIRTY guides mdb_cursor_touch */
+ mx->mx_dbflag = DB_VALID|DB_USRVALID|DB_DUPDATA;
if (NEED_CMP_CLONG(mx->mx_dbx.md_cmp, mx->mx_db.md_pad))
mx->mx_dbx.md_cmp = mdb_cmp_clong;
}
mx->mx_cursor.mc_top = 0;
mx->mx_cursor.mc_flags |= C_INITIALIZED;
mx->mx_cursor.mc_ki[0] = 0;
- mx->mx_dbflag = DB_VALID|DB_USRVALID|DB_DIRTY; /* DB_DIRTY guides mdb_cursor_touch */
+ mx->mx_dbflag = DB_VALID|DB_USRVALID|DB_DUPDATA;
#if UINT_MAX < MDB_SIZE_MAX /* matches mdb_xcursor_init1:NEED_CMP_CLONG() */
mx->mx_dbx.md_cmp = src_mx->mx_dbx.md_cmp;
#endif
if (!(mc->mc_flags & C_INITIALIZED))
return EINVAL;
- if (!mc->mc_snum || (mc->mc_flags & C_EOF))
+ if (!mc->mc_snum)
return MDB_NOTFOUND;
+ if (mc->mc_flags & C_EOF) {
+ if (mc->mc_ki[mc->mc_top] >= NUMKEYS(mc->mc_pg[mc->mc_top]))
+ return MDB_NOTFOUND;
+ mc->mc_flags ^= C_EOF;
+ }
+
leaf = NODEPTR(mc->mc_pg[mc->mc_top], mc->mc_ki[mc->mc_top]);
if (!F_ISSET(leaf->mn_flags, F_DUPDATA)) {
*countp = 1;
MDB_CURSOR_UNREF(mc, 0);
}
if (mc && !mc->mc_backup) {
- /* remove from txn, if tracked */
+ /* Remove from txn, if tracked.
+ * A read-only txn (!C_UNTRACK) may have been freed already,
+ * so do not peek inside it. Only write txns track cursors.
+ */
if ((mc->mc_flags & C_UNTRACK) && mc->mc_txn->mt_cursors) {
MDB_cursor **prev = &mc->mc_txn->mt_cursors[mc->mc_dbi];
while (*prev && *prev != mc) prev = &(*prev)->mc_next;
}
/** Replace the key for a branch node with a new key.
+ * Set #MDB_TXN_ERROR on failure.
* @param[in] mc Cursor pointing to the node to operate on.
* @param[in] key The new key to use.
* @return 0 on success, non-zero on failure.
* 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, flags);
}
/** Split a page and insert a new node.
+ * Set #MDB_TXN_ERROR on failure.
* @param[in,out] mc Cursor pointing to the page and desired insertion index.
* The cursor will be updated to point to the actual page and index where
* the node got inserted after the split.
} else {
/* find right page's left sibling */
mc->mc_ki[ptop] = mn.mc_ki[ptop];
- mdb_cursor_sibling(mc, 0);
+ rc = mdb_cursor_sibling(mc, 0);
}
}
} else {
mn.mc_top++;
}
if (rc != MDB_SUCCESS) {
+ if (rc == MDB_NOTFOUND) /* improper mdb_cursor_sibling() result */
+ rc = MDB_PROBLEM;
goto done;
}
if (nflags & MDB_APPEND) {
HANDLE mc_fd;
int mc_toggle; /**< Buffer number in provider */
int mc_new; /**< (0-2 buffers to write) | (#MDB_EOF at end) */
- volatile int mc_error; /**< Error code, never cleared if set */
+ /** Error code. Never cleared if set. Both threads can set nonzero
+ * to fail the copy. Not mutex-protected, LMDB expects atomic int.
+ */
+ volatile int mc_error;
} mdb_copy;
/** Dedicated writer thread for compacting copy. */
#else
int len;
#define DO_WRITE(rc, fd, ptr, w2, len) len = write(fd, ptr, w2); rc = (len >= 0)
+#ifdef SIGPIPE
+ sigset_t set;
+ sigemptyset(&set);
+ sigaddset(&set, SIGPIPE);
+ if ((rc = pthread_sigmask(SIG_BLOCK, &set, NULL)) != 0)
+ my->mc_error = rc;
+#endif
#endif
pthread_mutex_lock(&my->mc_mutex);
DO_WRITE(rc, my->mc_fd, ptr, wsize, len);
if (!rc) {
rc = ErrCode();
+#if defined(SIGPIPE) && !defined(_WIN32)
+ if (rc == EPIPE) {
+ /* Collect the pending SIGPIPE, otherwise at least OS X
+ * gives it to the process on thread-exit (ITS#8504).
+ */
+ int tmp;
+ sigwait(&set, &tmp);
+ }
+#endif
break;
} else if (len > 0) {
rc = MDB_SUCCESS;
return my->mc_error;
}
- /** Depth-first tree traversal for compacting copy. */
+ /** Depth-first tree traversal for compacting copy.
+ * @param[in] my control structure.
+ * @param[in,out] pg database root.
+ * @param[in] flags includes #F_DUPDATA if it is a sorted-duplicate sub-DB.
+ */
static int ESECT
mdb_env_cwalk(mdb_copy *my, pgno_t *pg, int flags)
{
memset(&dummy, 0, sizeof(dummy));
dummy.md_root = P_INVALID;
dummy.md_flags = flags & PERSISTENT_FLAGS;
- rc = mdb_cursor_put(&mc, &key, &data, F_SUBDATA);
+ WITH_CURSOR_TRACKING(mc,
+ rc = mdb_cursor_put(&mc, &key, &data, F_SUBDATA));
dbflag |= DB_DIRTY;
}
return env->me_txns ? mdb_reader_check0(env, 0, dead) : MDB_SUCCESS;
}
-/** As #mdb_reader_check(). rlocked = <caller locked the reader mutex>. */
+/** As #mdb_reader_check(). \b rlocked is set if caller locked #me_rmutex. */
static int ESECT
mdb_reader_check0(MDB_env *env, int rlocked, int *dead)
{
projects / openldap / blobdiff