X-Git-Url: https://git.sur5r.net/?a=blobdiff_plain;f=libraries%2Fliblmdb%2Fmdb.c;h=2a3fb1a0dfb763614676d1b17c301f469bf644db;hb=110fba2ef1fcc5d08a78ed5ee72134828ce70b1a;hp=440966b9ef978f1cc18bf8e5fbb6346e64397253;hpb=f25c4f0db9416248f260e8b24043226aecfa7122;p=openldap diff --git a/libraries/liblmdb/mdb.c b/libraries/liblmdb/mdb.c index 440966b9ef..2a3fb1a0df 100644 --- a/libraries/liblmdb/mdb.c +++ b/libraries/liblmdb/mdb.c @@ -121,7 +121,14 @@ extern int cacheflush(char *addr, int nbytes, int cache); #ifdef MDB_USE_POSIX_SEM # define MDB_USE_HASH 1 #include +#else +#define MDB_USE_POSIX_MUTEX 1 +#endif #endif + +#if defined(_WIN32) + defined(MDB_USE_POSIX_SEM) \ + + defined(MDB_USE_POSIX_MUTEX) != 1 +# error "Ambiguous shared-lock implementation" #endif #ifdef USE_VALGRIND @@ -217,6 +224,18 @@ extern int cacheflush(char *addr, int nbytes, int cache); # define mdb_func_ "" #endif +/* Internal error codes, not exposed outside liblmdb */ +#define MDB_NO_ROOT (MDB_LAST_ERRCODE + 10) +#ifdef _WIN32 +#define MDB_OWNERDEAD ((int) WAIT_ABANDONED) +#elif defined(MDB_USE_POSIX_MUTEX) && defined(EOWNERDEAD) +#define MDB_OWNERDEAD EOWNERDEAD /**< #LOCK_MUTEX0() result if dead owner */ +#endif + +#ifdef MDB_OWNERDEAD +#define MDB_ROBUST_SUPPORTED 1 +#endif + #ifdef _WIN32 #define MDB_USE_HASH 1 #define MDB_PIDLOCK 0 @@ -224,6 +243,7 @@ extern int cacheflush(char *addr, int nbytes, int cache); #define pthread_t HANDLE #define pthread_mutex_t HANDLE #define pthread_cond_t HANDLE +typedef HANDLE mdb_mutex_t, mdb_mutexref_t; #define pthread_key_t DWORD #define pthread_self() GetCurrentThreadId() #define pthread_key_create(x,y) \ @@ -237,10 +257,9 @@ extern int cacheflush(char *addr, int nbytes, int cache); #define pthread_cond_wait(cond,mutex) do{SignalObjectAndWait(*mutex, *cond, INFINITE, FALSE); WaitForSingleObject(*mutex, INFINITE);}while(0) #define THREAD_CREATE(thr,start,arg) thr=CreateThread(NULL,0,start,arg,0,NULL) #define THREAD_FINISH(thr) WaitForSingleObject(thr, INFINITE) -#define LOCK_MUTEX_R(env) pthread_mutex_lock(&(env)->me_rmutex) -#define UNLOCK_MUTEX_R(env) pthread_mutex_unlock(&(env)->me_rmutex) -#define LOCK_MUTEX_W(env) pthread_mutex_lock(&(env)->me_wmutex) -#define UNLOCK_MUTEX_W(env) pthread_mutex_unlock(&(env)->me_wmutex) +#define LOCK_MUTEX0(mutex) WaitForSingleObject(mutex, INFINITE) +#define UNLOCK_MUTEX(mutex) ReleaseMutex(mutex) +#define mdb_mutex_consistent(mutex) 0 #define getpid() GetCurrentProcessId() #define MDB_FDATASYNC(fd) (!FlushFileBuffers(fd)) #define MDB_MSYNC(addr,len,flags) (!FlushViewOfFile(addr,len)) @@ -265,10 +284,9 @@ extern int cacheflush(char *addr, int nbytes, int cache); #ifdef MDB_USE_POSIX_SEM -#define LOCK_MUTEX_R(env) mdb_sem_wait((env)->me_rmutex) -#define UNLOCK_MUTEX_R(env) sem_post((env)->me_rmutex) -#define LOCK_MUTEX_W(env) mdb_sem_wait((env)->me_wmutex) -#define UNLOCK_MUTEX_W(env) sem_post((env)->me_wmutex) +typedef sem_t *mdb_mutex_t, *mdb_mutexref_t; +#define LOCK_MUTEX0(mutex) mdb_sem_wait(mutex) +#define UNLOCK_MUTEX(mutex) sem_post(mutex) static int mdb_sem_wait(sem_t *sem) @@ -278,22 +296,26 @@ mdb_sem_wait(sem_t *sem) return rc; } -#else - /** Lock the reader mutex. +#else /* MDB_USE_POSIX_MUTEX: */ + /** Shared mutex/semaphore as it is stored (mdb_mutex_t), and as + * local variables keep it (mdb_mutexref_t). + * + * When #mdb_mutexref_t is a pointer declaration and #mdb_mutex_t is + * not, then it is array[size 1] so it can be assigned to a pointer. + * @{ */ -#define LOCK_MUTEX_R(env) pthread_mutex_lock(&(env)->me_txns->mti_mutex) - /** Unlock the reader mutex. +typedef pthread_mutex_t mdb_mutex_t[1], *mdb_mutexref_t; + /* @} */ + /** Lock the reader or writer mutex. + * Returns 0 or a code to give #mdb_mutex_failed(), as in #LOCK_MUTEX(). */ -#define UNLOCK_MUTEX_R(env) pthread_mutex_unlock(&(env)->me_txns->mti_mutex) - - /** Lock the writer mutex. - * Only a single write transaction is allowed at a time. Other writers - * will block waiting for this mutex. +#define LOCK_MUTEX0(mutex) pthread_mutex_lock(mutex) + /** Unlock the reader or writer mutex. */ -#define LOCK_MUTEX_W(env) pthread_mutex_lock(&(env)->me_txns->mti_wmutex) - /** Unlock the writer mutex. +#define UNLOCK_MUTEX(mutex) pthread_mutex_unlock(mutex) + /** Mark mutex-protected data as repaired, after death of previous owner. */ -#define UNLOCK_MUTEX_W(env) pthread_mutex_unlock(&(env)->me_txns->mti_wmutex) +#define mdb_mutex_consistent(mutex) pthread_mutex_consistent(mutex) #endif /* MDB_USE_POSIX_SEM */ /** Get the error code for the last failed system function. @@ -327,6 +349,19 @@ mdb_sem_wait(sem_t *sem) /** @} */ +#ifdef MDB_ROBUST_SUPPORTED + /** Lock mutex, handle any error, set rc = result. + * Return 0 on success, nonzero (not rc) on error. + */ +#define LOCK_MUTEX(rc, env, mutex) \ + (((rc) = LOCK_MUTEX0(mutex)) && \ + ((rc) = mdb_mutex_failed(env, mutex, rc))) +static int mdb_mutex_failed(MDB_env *env, mdb_mutexref_t mutex, int rc); +#else +#define LOCK_MUTEX(rc, env, mutex) ((rc) = LOCK_MUTEX0(mutex)) +#define mdb_mutex_failed(env, mutex, rc) (rc) +#endif + #ifndef _WIN32 /** A flag for opening a file and requesting synchronous data writes. * This is only used when writing a meta page. It's not strictly needed; @@ -643,9 +678,9 @@ typedef struct MDB_txbody { char mtb_rmname[MNAME_LEN]; #else /** Mutex protecting access to this table. - * This is the reader lock that #LOCK_MUTEX_R acquires. + * This is the reader table lock used with LOCK_MUTEX(). */ - pthread_mutex_t mtb_mutex; + 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 @@ -665,7 +700,7 @@ typedef struct MDB_txninfo { MDB_txbody mtb; #define mti_magic mt1.mtb.mtb_magic #define mti_format mt1.mtb.mtb_format -#define mti_mutex mt1.mtb.mtb_mutex +#define mti_rmutex mt1.mtb.mtb_rmutex #define mti_rmname mt1.mtb.mtb_rmname #define mti_txnid mt1.mtb.mtb_txnid #define mti_numreaders mt1.mtb.mtb_numreaders @@ -676,7 +711,7 @@ typedef struct MDB_txninfo { char mt2_wmname[MNAME_LEN]; #define mti_wmname mt2.mt2_wmname #else - pthread_mutex_t mt2_wmutex; + mdb_mutex_t mt2_wmutex; #define mti_wmutex mt2.mt2_wmutex #endif char pad[(MNAME_LEN+CACHELINE-1) & ~(CACHELINE-1)]; @@ -1002,6 +1037,7 @@ struct MDB_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 */ /** @} */ /** In write txns, array of cursors for each DB */ MDB_cursor **mt_cursors; @@ -1163,11 +1199,13 @@ struct MDB_env { int me_live_reader; /**< have liveness lock in reader table */ #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) - sem_t *me_rmutex; /* Shared mutexes are not supported */ - sem_t *me_wmutex; +#endif +#ifdef MDB_USE_POSIX_MUTEX /* Posix mutexes reside in shared mem */ +# define me_rmutex me_txns->mti_rmutex /**< Shared reader lock */ +# define me_wmutex me_txns->mti_wmutex /**< Shared writer lock */ +#else + mdb_mutex_t me_rmutex; + mdb_mutex_t me_wmutex; #endif void *me_userctx; /**< User-settable context */ MDB_assert_func *me_assert_func; /**< Callback for assertion failures */ @@ -1190,8 +1228,8 @@ typedef struct MDB_ntxn { #define MAX_WRITE (0x80000000U >> (sizeof(ssize_t) == 4)) /** Check \b txn and \b dbi arguments to a function */ -#define TXN_DBI_EXIST(txn, dbi) \ - ((txn) && (dbi) < (txn)->mt_numdbs && ((txn)->mt_dbflags[dbi] & DB_VALID)) +#define TXN_DBI_EXIST(txn, dbi, validity) \ + ((txn) && (dbi)<(txn)->mt_numdbs && ((txn)->mt_dbflags[dbi] & (validity))) /** Check for misused \b dbi handles */ #define TXN_DBI_CHANGED(txn, dbi) \ @@ -1219,7 +1257,7 @@ static int mdb_page_split(MDB_cursor *mc, MDB_val *newkey, MDB_val *newdata, static int mdb_env_read_header(MDB_env *env, MDB_meta *meta); static int mdb_env_pick_meta(const MDB_env *env); static int mdb_env_write_meta(MDB_txn *txn); -#if !(defined(_WIN32) || defined(MDB_USE_POSIX_SEM)) /* Drop unused excl arg */ +#ifdef MDB_USE_POSIX_MUTEX /* Drop unused excl arg */ # define mdb_env_close0(env, excl) mdb_env_close1(env) #endif static void mdb_env_close0(MDB_env *env, int excl); @@ -1257,6 +1295,7 @@ static void mdb_xcursor_init2(MDB_cursor *mc, MDB_xcursor *src_mx, int force); static int mdb_drop0(MDB_cursor *mc, int subs); static void mdb_default_cmp(MDB_txn *txn, MDB_dbi dbi); +static int mdb_reader_check0(MDB_env *env, int rlocked, int *dead); /** @cond */ static MDB_cmp_func mdb_cmp_memn, mdb_cmp_memnr, mdb_cmp_int, mdb_cmp_cint, mdb_cmp_long; @@ -1291,7 +1330,7 @@ static char *const mdb_errstr[] = { "MDB_NOTFOUND: No matching key/data pair found", "MDB_PAGE_NOTFOUND: Requested page not found", "MDB_CORRUPTED: Located page was wrong type", - "MDB_PANIC: Update of meta page failed", + "MDB_PANIC: Update of meta page failed or environment had fatal error", "MDB_VERSION_MISMATCH: Database environment version mismatch", "MDB_INVALID: File is not an LMDB file", "MDB_MAP_FULL: Environment mapsize limit reached", @@ -2533,6 +2572,7 @@ mdb_txn_renew0(MDB_txn *txn) } else { MDB_PID_T pid = env->me_pid; MDB_THR_T tid = pthread_self(); + mdb_mutexref_t rmutex = env->me_rmutex; if (!env->me_live_reader) { rc = mdb_reader_pid(env, Pidset, pid); @@ -2541,13 +2581,14 @@ mdb_txn_renew0(MDB_txn *txn) env->me_live_reader = 1; } - LOCK_MUTEX_R(env); + if (LOCK_MUTEX(rc, env, rmutex)) + return rc; nr = ti->mti_numreaders; for (i=0; imti_readers[i].mr_pid == 0) break; if (i == env->me_maxreaders) { - UNLOCK_MUTEX_R(env); + UNLOCK_MUTEX(rmutex); return MDB_READERS_FULL; } r = &ti->mti_readers[i]; @@ -2564,7 +2605,7 @@ mdb_txn_renew0(MDB_txn *txn) ti->mti_numreaders = ++nr; env->me_close_readers = nr; r->mr_pid = pid; - UNLOCK_MUTEX_R(env); + UNLOCK_MUTEX(rmutex); new_notls = (env->me_flags & MDB_NOTLS); if (!new_notls && (rc=pthread_setspecific(env->me_txkey, r))) { @@ -2583,8 +2624,8 @@ mdb_txn_renew0(MDB_txn *txn) } else { /* Not yet touching txn == env->me_txn0, it may be active */ if (ti) { - LOCK_MUTEX_W(env); - + if (LOCK_MUTEX(rc, env, env->me_wmutex)) + return rc; txn->mt_txnid = ti->mti_txnid; meta = env->me_metas[txn->mt_txnid & 1]; } else { @@ -2622,9 +2663,10 @@ mdb_txn_renew0(MDB_txn *txn) for (i=2; imt_numdbs; i++) { x = env->me_dbflags[i]; txn->mt_dbs[i].md_flags = x & PERSISTENT_FLAGS; - txn->mt_dbflags[i] = (x & MDB_VALID) ? DB_VALID|DB_STALE : 0; + txn->mt_dbflags[i] = (x & MDB_VALID) ? DB_VALID|DB_USRVALID|DB_STALE : 0; } - txn->mt_dbflags[0] = txn->mt_dbflags[1] = DB_VALID; + txn->mt_dbflags[MAIN_DBI] = DB_VALID|DB_USRVALID; + txn->mt_dbflags[FREE_DBI] = DB_VALID; if (env->me_maxpg < txn->mt_next_pgno) { mdb_txn_reset0(txn, "renew0-mapfail"); @@ -2677,7 +2719,6 @@ mdb_txn_begin(MDB_env *env, MDB_txn *parent, unsigned int flags, MDB_txn **ret) if (env->me_flags & MDB_RDONLY & ~flags) /* write txn in RDONLY env */ return EACCES; - size = tsize = sizeof(MDB_txn); if (parent) { /* Nested transactions: Max 1 child, write txns only, no writemap */ flags |= parent->mt_flags; @@ -2687,40 +2728,31 @@ mdb_txn_begin(MDB_env *env, MDB_txn *parent, unsigned int flags, MDB_txn **ret) return (parent->mt_flags & MDB_TXN_RDONLY) ? EINVAL : MDB_BAD_TXN; } /* Child txns save MDB_pgstate and use own copy of cursors */ - size = tsize = sizeof(MDB_ntxn); - size += env->me_maxdbs * sizeof(MDB_cursor *); - } else if (!(flags & MDB_RDONLY)) { + size = env->me_maxdbs * (sizeof(MDB_db)+sizeof(MDB_cursor *)+1); + size += tsize = sizeof(MDB_ntxn); + } else if (flags & MDB_RDONLY) { + size = env->me_maxdbs * (sizeof(MDB_db)+1); + size += tsize = sizeof(MDB_txn); + } else { /* Reuse preallocated write txn. However, do not touch it until * mdb_txn_renew0() succeeds, since it currently may be active. */ txn = env->me_txn0; goto renew; } - size += env->me_maxdbs * (sizeof(MDB_db)+1); - if ((txn = calloc(1, size)) == NULL) { DPRINTF(("calloc: %s", strerror(errno))); return ENOMEM; } txn->mt_dbs = (MDB_db *) ((char *)txn + tsize); - if (flags & MDB_RDONLY) { - txn->mt_dbflags = (unsigned char *)(txn->mt_dbs + env->me_maxdbs); - txn->mt_dbiseqs = env->me_dbiseqs; - } else { - txn->mt_cursors = (MDB_cursor **)(txn->mt_dbs + env->me_maxdbs); - if (parent) { - txn->mt_dbiseqs = parent->mt_dbiseqs; - txn->mt_dbflags = (unsigned char *)(txn->mt_cursors + env->me_maxdbs); - } else { - txn->mt_dbiseqs = (unsigned int *)(txn->mt_cursors + env->me_maxdbs); - txn->mt_dbflags = (unsigned char *)(txn->mt_dbiseqs + env->me_maxdbs); - } - } + txn->mt_dbflags = (unsigned char *)txn + size - env->me_maxdbs; txn->mt_flags = flags; txn->mt_env = env; if (parent) { unsigned int i; + txn->mt_cursors = (MDB_cursor **)(txn->mt_dbs + env->me_maxdbs); + txn->mt_dbiseqs = parent->mt_dbiseqs; txn->mt_u.dirty_list = malloc(sizeof(MDB_ID2)*MDB_IDL_UM_SIZE); if (!txn->mt_u.dirty_list || !(txn->mt_free_pgs = mdb_midl_alloc(MDB_IDL_UM_MAX))) @@ -2757,7 +2789,8 @@ mdb_txn_begin(MDB_env *env, MDB_txn *parent, unsigned int flags, MDB_txn **ret) rc = mdb_cursor_shadow(parent, txn); if (rc) mdb_txn_reset0(txn, "beginchild-fail"); - } else { + } else { /* MDB_RDONLY */ + txn->mt_dbiseqs = env->me_dbiseqs; renew: rc = mdb_txn_renew0(txn); } @@ -2765,7 +2798,7 @@ renew: if (txn != env->me_txn0) free(txn); } else { - txn->mt_flags |= flags; /* for txn==me_txn0, no effect otherwise */ + txn->mt_flags |= flags; /* could not change txn=me_txn0 earlier */ *ret = txn; DPRINTF(("begin txn %"Z"u%c %p on mdbenv %p, root page %"Z"u", txn->mt_txnid, (flags & MDB_RDONLY) ? 'r' : 'w', @@ -2861,7 +2894,7 @@ mdb_txn_reset0(MDB_txn *txn, const char *act) env->me_txn = NULL; /* The writer mutex was locked in mdb_txn_begin. */ if (env->me_txns) - UNLOCK_MUTEX_W(env); + UNLOCK_MUTEX(env->me_wmutex); } else { txn->mt_parent->mt_child = NULL; env->me_pgstate = ((MDB_ntxn *)txn)->mnt_pgstate; @@ -3256,12 +3289,11 @@ mdb_txn_commit(MDB_txn *txn) unsigned int i; MDB_env *env; - if (txn == NULL || txn->mt_env == NULL) + if (txn == NULL) return EINVAL; if (txn->mt_child) { rc = mdb_txn_commit(txn->mt_child); - txn->mt_child = NULL; if (rc) goto fail; } @@ -3461,7 +3493,7 @@ done: mdb_dbis_update(txn, 1); if (env->me_txns) - UNLOCK_MUTEX_W(env); + UNLOCK_MUTEX(env->me_wmutex); if (txn != env->me_txn0) free(txn); @@ -4234,8 +4266,8 @@ mdb_env_excl_lock(MDB_env *env, int *excl) if (!rc) { *excl = 1; } else -# ifdef MDB_USE_POSIX_SEM - if (*excl < 0) /* always true when !MDB_USE_POSIX_SEM */ +# ifndef MDB_USE_POSIX_MUTEX + if (*excl < 0) /* always true when MDB_USE_POSIX_MUTEX */ # endif { lock_info.l_type = F_RDLCK; @@ -4507,13 +4539,16 @@ mdb_env_setup_locks(MDB_env *env, char *lpath, int mode, int *excl) env->me_wmutex = sem_open(env->me_txns->mti_wmname, O_CREAT|O_EXCL, mode, 1); if (env->me_wmutex == SEM_FAILED) goto fail_errno; -#else /* MDB_USE_POSIX_SEM */ +#else /* MDB_USE_POSIX_MUTEX: */ pthread_mutexattr_t mattr; if ((rc = pthread_mutexattr_init(&mattr)) || (rc = pthread_mutexattr_setpshared(&mattr, PTHREAD_PROCESS_SHARED)) - || (rc = pthread_mutex_init(&env->me_txns->mti_mutex, &mattr)) - || (rc = pthread_mutex_init(&env->me_txns->mti_wmutex, &mattr))) +#ifdef MDB_ROBUST_SUPPORTED + || (rc = pthread_mutexattr_setrobust(&mattr, PTHREAD_MUTEX_ROBUST)) +#endif + || (rc = pthread_mutex_init(env->me_txns->mti_rmutex, &mattr)) + || (rc = pthread_mutex_init(env->me_txns->mti_wmutex, &mattr))) goto fail; pthread_mutexattr_destroy(&mattr); #endif /* _WIN32 || MDB_USE_POSIX_SEM */ @@ -4570,8 +4605,8 @@ fail: * environment and re-opening it with the new flags. */ #define CHANGEABLE (MDB_NOSYNC|MDB_NOMETASYNC|MDB_MAPASYNC|MDB_NOMEMINIT) -#define CHANGELESS (MDB_FIXEDMAP|MDB_NOSUBDIR|MDB_RDONLY|MDB_WRITEMAP| \ - MDB_NOTLS|MDB_NOLOCK|MDB_NORDAHEAD) +#define CHANGELESS (MDB_FIXEDMAP|MDB_NOSUBDIR|MDB_RDONLY| \ + MDB_WRITEMAP|MDB_NOTLS|MDB_NOLOCK|MDB_NORDAHEAD) #if VALID_FLAGS & PERSISTENT_FLAGS & (CHANGEABLE|CHANGELESS) # error "Persistent DB flags & env flags overlap, but both go in mm_flags" @@ -4627,6 +4662,7 @@ mdb_env_open(MDB_env *env, const char *path, unsigned int flags, mdb_mode_t mode rc = ENOMEM; goto leave; } + env->me_dbxs[FREE_DBI].md_cmp = mdb_cmp_long; /* aligned MDB_INTEGERKEY */ /* For RDONLY, get lockfile after we know datafile exists */ if (!(flags & (MDB_RDONLY|MDB_NOLOCK))) { @@ -5048,15 +5084,12 @@ static void mdb_cursor_pop(MDB_cursor *mc) { if (mc->mc_snum) { -#if MDB_DEBUG - MDB_page *top = mc->mc_pg[mc->mc_top]; -#endif + DPRINTF(("popping page %"Z"u off db %d cursor %p", + mc->mc_pg[mc->mc_top]->mp_pgno, DDBI(mc), (void *) mc)); + mc->mc_snum--; if (mc->mc_snum) mc->mc_top--; - - DPRINTF(("popped page %"Z"u off db %d cursor %p", top->mp_pgno, - DDBI(mc), (void *) mc)); } } @@ -5435,7 +5468,7 @@ mdb_get(MDB_txn *txn, MDB_dbi dbi, DPRINTF(("===> get db %u key [%s]", dbi, DKEY(key))); - if (!key || !data || dbi == FREE_DBI || !TXN_DBI_EXIST(txn, dbi)) + if (!key || !data || !TXN_DBI_EXIST(txn, dbi, DB_USRVALID)) return EINVAL; if (txn->mt_flags & MDB_TXN_ERROR) @@ -6158,7 +6191,6 @@ 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_env *env; MDB_node *leaf = NULL; MDB_page *fp, *mp, *sub_root = NULL; @@ -7071,45 +7103,38 @@ mdb_node_shrink(MDB_page *mp, indx_t indx) MDB_node *node; MDB_page *sp, *xp; char *base; - int nsize, delta; - indx_t i, numkeys, ptr; + indx_t delta, nsize, len, ptr; + int i; node = NODEPTR(mp, indx); sp = (MDB_page *)NODEDATA(node); delta = SIZELEFT(sp); - xp = (MDB_page *)((char *)sp + delta); + nsize = NODEDSZ(node) - delta; - /* shift subpage upward */ + /* Prepare to shift upward, set len = length(subpage part to shift) */ if (IS_LEAF2(sp)) { - nsize = NUMKEYS(sp) * sp->mp_pad; + len = nsize; if (nsize & 1) return; /* do not make the node uneven-sized */ - memmove(METADATA(xp), METADATA(sp), nsize); } else { - int i; - numkeys = NUMKEYS(sp); - for (i=numkeys-1; i>=0; i--) + xp = (MDB_page *)((char *)sp + delta); /* destination subpage */ + for (i = NUMKEYS(sp); --i >= 0; ) xp->mp_ptrs[i] = sp->mp_ptrs[i] - delta; + len = PAGEHDRSZ; } - xp->mp_upper = sp->mp_lower; - xp->mp_lower = sp->mp_lower; - xp->mp_flags = sp->mp_flags; - xp->mp_pad = sp->mp_pad; - COPY_PGNO(xp->mp_pgno, mp->mp_pgno); - - nsize = NODEDSZ(node) - delta; + sp->mp_upper = sp->mp_lower; + COPY_PGNO(sp->mp_pgno, mp->mp_pgno); SETDSZ(node, nsize); - /* shift lower nodes upward */ + /* Shift upward */ + base = (char *)mp + mp->mp_upper + PAGEBASE; + memmove(base + delta, base, (char *)sp + len - base); + ptr = mp->mp_ptrs[indx]; - numkeys = NUMKEYS(mp); - for (i = 0; i < numkeys; i++) { + for (i = NUMKEYS(mp); --i >= 0; ) { if (mp->mp_ptrs[i] <= ptr) mp->mp_ptrs[i] += delta; } - - base = (char *)mp + mp->mp_upper + PAGEBASE; - memmove(base + delta, base, ptr - mp->mp_upper + NODESIZE + NODEKSZ(node)); mp->mp_upper += delta; } @@ -7184,7 +7209,7 @@ mdb_xcursor_init1(MDB_cursor *mc, MDB_node *node) } 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 */ + mx->mx_dbflag = DB_VALID|DB_USRVALID|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)) mx->mx_dbx.md_cmp = mdb_cmp_clong; @@ -7257,7 +7282,7 @@ mdb_cursor_open(MDB_txn *txn, MDB_dbi dbi, MDB_cursor **ret) MDB_cursor *mc; size_t size = sizeof(MDB_cursor); - if (!ret || !TXN_DBI_EXIST(txn, dbi)) + if (!ret || !TXN_DBI_EXIST(txn, dbi, DB_VALID)) return EINVAL; if (txn->mt_flags & MDB_TXN_ERROR) @@ -7289,7 +7314,7 @@ mdb_cursor_open(MDB_txn *txn, MDB_dbi dbi, MDB_cursor **ret) int mdb_cursor_renew(MDB_txn *txn, MDB_cursor *mc) { - if (!mc || !TXN_DBI_EXIST(txn, mc->mc_dbi)) + if (!mc || !TXN_DBI_EXIST(txn, mc->mc_dbi, DB_VALID)) return EINVAL; if ((mc->mc_flags & C_UNTRACK) || txn->mt_cursors) @@ -8061,7 +8086,7 @@ int mdb_del(MDB_txn *txn, MDB_dbi dbi, MDB_val *key, MDB_val *data) { - if (!key || dbi == FREE_DBI || !TXN_DBI_EXIST(txn, dbi)) + if (!key || !TXN_DBI_EXIST(txn, dbi, DB_USRVALID)) return EINVAL; if (txn->mt_flags & (MDB_TXN_RDONLY|MDB_TXN_ERROR)) @@ -8527,10 +8552,10 @@ mdb_put(MDB_txn *txn, MDB_dbi dbi, MDB_cursor mc; MDB_xcursor mx; - if (!key || !data || dbi == FREE_DBI || !TXN_DBI_EXIST(txn, dbi)) + if (!key || !data || !TXN_DBI_EXIST(txn, dbi, DB_USRVALID)) return EINVAL; - if ((flags & (MDB_NOOVERWRITE|MDB_NODUPDATA|MDB_RESERVE|MDB_APPEND|MDB_APPENDDUP)) != flags) + if (flags & ~(MDB_NOOVERWRITE|MDB_NODUPDATA|MDB_RESERVE|MDB_APPEND|MDB_APPENDDUP)) return EINVAL; mdb_cursor_init(&mc, txn, dbi, &mx); @@ -8921,6 +8946,7 @@ static int ESECT mdb_env_copyfd0(MDB_env *env, HANDLE fd) { MDB_txn *txn = NULL; + mdb_mutexref_t wmutex = NULL; int rc; size_t wsize; char *ptr; @@ -8945,11 +8971,13 @@ mdb_env_copyfd0(MDB_env *env, HANDLE fd) mdb_txn_reset0(txn, "reset-stage1"); /* Temporarily block writers until we snapshot the meta pages */ - LOCK_MUTEX_W(env); + wmutex = env->me_wmutex; + if (LOCK_MUTEX(rc, env, wmutex)) + goto leave; rc = mdb_txn_renew0(txn); if (rc) { - UNLOCK_MUTEX_W(env); + UNLOCK_MUTEX(wmutex); goto leave; } } @@ -8973,8 +9001,8 @@ mdb_env_copyfd0(MDB_env *env, HANDLE fd) break; } } - if (env->me_txns) - UNLOCK_MUTEX_W(env); + if (wmutex) + UNLOCK_MUTEX(wmutex); if (rc) goto leave; @@ -9097,7 +9125,7 @@ mdb_env_copy(MDB_env *env, const char *path) int ESECT mdb_env_set_flags(MDB_env *env, unsigned int flag, int onoff) { - if ((flag & CHANGEABLE) != flag) + if (flag & ~CHANGEABLE) return EINVAL; if (onoff) env->me_flags |= flag; @@ -9112,7 +9140,7 @@ mdb_env_get_flags(MDB_env *env, unsigned int *arg) if (!env || !arg) return EINVAL; - *arg = env->me_flags; + *arg = env->me_flags & (CHANGEABLE|CHANGELESS); return MDB_SUCCESS; } @@ -9246,11 +9274,7 @@ int mdb_dbi_open(MDB_txn *txn, const char *name, unsigned int flags, MDB_dbi *db unsigned int unused = 0, seq; size_t len; - if (txn->mt_dbxs[FREE_DBI].md_cmp == NULL) { - mdb_default_cmp(txn, FREE_DBI); - } - - if ((flags & VALID_FLAGS) != flags) + if (flags & ~VALID_FLAGS) return EINVAL; if (txn->mt_flags & MDB_TXN_ERROR) return MDB_BAD_TXN; @@ -9298,7 +9322,7 @@ int mdb_dbi_open(MDB_txn *txn, const char *name, unsigned int flags, MDB_dbi *db return (flags & MDB_CREATE) ? MDB_INCOMPATIBLE : MDB_NOTFOUND; /* Find the DB info */ - dbflag = DB_NEW|DB_VALID; + dbflag = DB_NEW|DB_VALID|DB_USRVALID; exact = 0; key.mv_size = len; key.mv_data = (void *)name; @@ -9346,7 +9370,7 @@ int mdb_dbi_open(MDB_txn *txn, const char *name, unsigned int flags, MDB_dbi *db int mdb_stat(MDB_txn *txn, MDB_dbi dbi, MDB_stat *arg) { - if (!arg || !TXN_DBI_EXIST(txn, dbi)) + if (!arg || !TXN_DBI_EXIST(txn, dbi, DB_VALID)) return EINVAL; if (txn->mt_flags & MDB_TXN_ERROR) @@ -9380,7 +9404,7 @@ void mdb_dbi_close(MDB_env *env, MDB_dbi dbi) 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 (dbi == FREE_DBI || !TXN_DBI_EXIST(txn, dbi)) + if (!TXN_DBI_EXIST(txn, dbi, DB_USRVALID)) return EINVAL; *flags = txn->mt_dbs[dbi].md_flags & PERSISTENT_FLAGS; return MDB_SUCCESS; @@ -9403,8 +9427,10 @@ mdb_drop0(MDB_cursor *mc, int subs) MDB_cursor mx; unsigned int i; - /* LEAF2 pages have no nodes, cannot have sub-DBs */ - if (IS_LEAF2(mc->mc_pg[mc->mc_top])) + /* DUPSORT sub-DBs have no ovpages/DBs. Omit scanning leaves. + * This also avoids any P_LEAF2 pages, which have no nodes. + */ + if (mc->mc_flags & C_SUB) mdb_cursor_pop(mc); mdb_cursor_copy(mc, &mx); @@ -9478,13 +9504,13 @@ int mdb_drop(MDB_txn *txn, MDB_dbi dbi, int del) MDB_cursor *mc, *m2; int rc; - if ((unsigned)del > 1 || dbi == FREE_DBI || !TXN_DBI_EXIST(txn, dbi)) + if ((unsigned)del > 1 || !TXN_DBI_EXIST(txn, dbi, DB_USRVALID)) return EINVAL; if (F_ISSET(txn->mt_flags, MDB_TXN_RDONLY)) return EACCES; - if (dbi > MAIN_DBI && TXN_DBI_CHANGED(txn, dbi)) + if (TXN_DBI_CHANGED(txn, dbi)) return MDB_BAD_DBI; rc = mdb_cursor_open(txn, dbi, &mc); @@ -9526,7 +9552,7 @@ leave: int mdb_set_compare(MDB_txn *txn, MDB_dbi dbi, MDB_cmp_func *cmp) { - if (dbi == FREE_DBI || !TXN_DBI_EXIST(txn, dbi)) + if (!TXN_DBI_EXIST(txn, dbi, DB_USRVALID)) return EINVAL; txn->mt_dbxs[dbi].md_cmp = cmp; @@ -9535,7 +9561,7 @@ int mdb_set_compare(MDB_txn *txn, MDB_dbi dbi, MDB_cmp_func *cmp) int mdb_set_dupsort(MDB_txn *txn, MDB_dbi dbi, MDB_cmp_func *cmp) { - if (dbi == FREE_DBI || !TXN_DBI_EXIST(txn, dbi)) + if (!TXN_DBI_EXIST(txn, dbi, DB_USRVALID)) return EINVAL; txn->mt_dbxs[dbi].md_dcmp = cmp; @@ -9544,7 +9570,7 @@ int mdb_set_dupsort(MDB_txn *txn, MDB_dbi dbi, MDB_cmp_func *cmp) int mdb_set_relfunc(MDB_txn *txn, MDB_dbi dbi, MDB_rel_func *rel) { - if (dbi == FREE_DBI || !TXN_DBI_EXIST(txn, dbi)) + if (!TXN_DBI_EXIST(txn, dbi, DB_USRVALID)) return EINVAL; txn->mt_dbxs[dbi].md_rel = rel; @@ -9553,7 +9579,7 @@ int mdb_set_relfunc(MDB_txn *txn, MDB_dbi dbi, MDB_rel_func *rel) int mdb_set_relctx(MDB_txn *txn, MDB_dbi dbi, void *ctx) { - if (dbi == FREE_DBI || !TXN_DBI_EXIST(txn, dbi)) + if (!TXN_DBI_EXIST(txn, dbi, DB_USRVALID)) return EINVAL; txn->mt_dbxs[dbi].md_relctx = ctx; @@ -9647,17 +9673,22 @@ mdb_pid_insert(MDB_PID_T *ids, MDB_PID_T pid) int ESECT mdb_reader_check(MDB_env *env, int *dead) { - unsigned int i, j, rdrs; - MDB_reader *mr; - MDB_PID_T *pids, pid; - int count = 0; - if (!env) return EINVAL; if (dead) *dead = 0; - if (!env->me_txns) - return MDB_SUCCESS; + return env->me_txns ? mdb_reader_check0(env, 0, dead) : MDB_SUCCESS; +} + +/** As #mdb_reader_check(). rlocked = . */ +static int mdb_reader_check0(MDB_env *env, int rlocked, int *dead) +{ + mdb_mutexref_t rmutex = rlocked ? NULL : env->me_rmutex; + unsigned int i, j, rdrs; + MDB_reader *mr; + MDB_PID_T *pids, pid; + int rc = MDB_SUCCESS, count = 0; + rdrs = env->me_txns->mti_numreaders; pids = malloc((rdrs+1) * sizeof(MDB_PID_T)); if (!pids) @@ -9665,22 +9696,32 @@ mdb_reader_check(MDB_env *env, int *dead) pids[0] = 0; mr = env->me_txns->mti_readers; for (i=0; ime_pid) { - pid = mr[i].mr_pid; + pid = mr[i].mr_pid; + if (pid && pid != env->me_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; jme_rmutex); + if (!rlocked) { + /* Keep mti_txnid updated, otherwise next writer can + * overwrite data which latest meta page refers to. + */ + toggle = mdb_env_pick_meta(env); + env->me_txns->mti_txnid = env->me_metas[toggle]->mm_txnid; + /* env is hosed if the dead thread was ours */ + if (env->me_txn) { + env->me_flags |= MDB_FATAL_ERROR; + env->me_txn = NULL; + rc = MDB_PANIC; + } + } + DPRINTF(("%cmutex owner died, %s", (rlocked ? 'r' : 'w'), + (rc ? "this process' env is hosed" : "recovering"))); + rc2 = mdb_reader_check0(env, rlocked, NULL); + if (rc2 == 0) + rc2 = mdb_mutex_consistent(mutex); + if (rc || (rc = rc2)) { + DPRINTF(("LOCK_MUTEX recovery failed, %s", mdb_strerror(rc))); + UNLOCK_MUTEX(mutex); + } + } else { +#ifdef _WIN32 + rc = ErrCode(); +#endif + DPRINTF(("LOCK_MUTEX failed, %s", mdb_strerror(rc))); + } + + return rc; } +#endif /* MDB_ROBUST_SUPPORTED */ /** @} */