X-Git-Url: https://git.sur5r.net/?a=blobdiff_plain;f=libraries%2Fliblmdb%2Fmdb.c;h=fc0340c51e2e33fe22eb62adc3b5b779e13e9588;hb=71c07f0d60be6113e2ea4327a15b3646c19825d0;hp=2ae60436b5bec1848216ee10340d01262e4189b9;hpb=076b773d6bab49e1b467051cc09f2522444a22f3;p=openldap diff --git a/libraries/liblmdb/mdb.c b/libraries/liblmdb/mdb.c index 2ae60436b5..fc0340c51e 100644 --- a/libraries/liblmdb/mdb.c +++ b/libraries/liblmdb/mdb.c @@ -96,7 +96,13 @@ extern int cacheflush(char *addr, int nbytes, int cache); #include #include #include + +#ifdef _MSC_VER +#include +typedef SSIZE_T ssize_t; +#else #include +#endif #if defined(__sun) || defined(ANDROID) /* Most platforms have posix_memalign, older may only have memalign */ @@ -121,7 +127,14 @@ 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 +230,45 @@ 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 __GLIBC__ +#define GLIBC_VER ((__GLIBC__ << 16 )| __GLIBC_MINOR__) +#endif +/** Some platforms define the EOWNERDEAD error code + * even though they don't support Robust Mutexes. + * Compile with -DMDB_USE_ROBUST=0, or use some other + * mechanism like -DMDB_USE_SYSV_SEM instead of + * -DMDB_USE_POSIX_MUTEX. (SysV semaphores are + * also Robust, but some systems don't support them + * either.) + */ +#ifndef MDB_USE_ROBUST +/* Android currently lacks Robust Mutex support. So does glibc < 2.4. */ +# if defined(MDB_USE_POSIX_MUTEX) && (defined(ANDROID) || \ + (defined(__GLIBC__) && GLIBC_VER < 0x020004)) +# define MDB_USE_ROBUST 0 +# else +# define MDB_USE_ROBUST 1 +/* glibc < 2.10 only provided _np API */ +# if defined(__GLIBC__) && GLIBC_VER < 0x02000a +# define PTHREAD_MUTEX_ROBUST PTHREAD_MUTEX_ROBUST_NP +# define pthread_mutexattr_setrobust(attr, flag) pthread_mutexattr_setrobust_np(attr, flag) +# define pthread_mutex_consistent(mutex) pthread_mutex_consistent_np(mutex) +# endif +# endif +#endif /* MDB_USE_ROBUST */ + +#if defined(MDB_OWNERDEAD) && MDB_USE_ROBUST +#define MDB_ROBUST_SUPPORTED 1 +#endif + #ifdef _WIN32 #define MDB_USE_HASH 1 #define MDB_PIDLOCK 0 @@ -224,6 +276,7 @@ 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 +290,9 @@ 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 +317,9 @@ 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 +329,26 @@ 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 +382,19 @@ 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; @@ -335,10 +403,13 @@ mdb_sem_wait(sem_t *sem) * * @note If O_DSYNC is undefined but exists in /usr/include, * preferably set some compiler flag to get the definition. - * Otherwise compile with the less efficient -DMDB_DSYNC=O_SYNC. */ #ifndef MDB_DSYNC +# ifdef O_DSYNC # define MDB_DSYNC O_DSYNC +# else +# define MDB_DSYNC O_SYNC +# endif #endif #endif @@ -643,9 +714,9 @@ 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 +736,7 @@ 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 +747,7 @@ 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)]; @@ -908,6 +979,11 @@ typedef struct MDB_db { #define FREE_DBI 0 /** Handle for the default DB. */ #define MAIN_DBI 1 + /** Number of DBs in metapage (free and main) - also hardcoded elsewhere */ +#define CORE_DBS 2 + + /** Number of meta pages - also hardcoded elsewhere */ +#define NUM_METAS 2 /** Meta page content. * A meta page is the start point for accessing a database snapshot. @@ -921,11 +997,11 @@ typedef struct MDB_meta { uint32_t mm_version; void *mm_address; /**< address for fixed mapping */ size_t mm_mapsize; /**< size of mmap region */ - MDB_db mm_dbs[2]; /**< first is free space, 2nd is main db */ + 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[0].md_pad +#define mm_psize mm_dbs[FREE_DBI].md_pad /** Any persistent environment flags. @ref mdb_env */ -#define mm_flags mm_dbs[0].md_flags +#define mm_flags mm_dbs[FREE_DBI].md_flags pgno_t mm_last_pg; /**< last used page in file */ volatile txnid_t mm_txnid; /**< txnid that committed this page */ } MDB_meta; @@ -960,7 +1036,8 @@ typedef struct MDB_dbx { */ struct MDB_txn { MDB_txn *mt_parent; /**< parent of a nested txn */ - MDB_txn *mt_child; /**< nested txn under this txn */ + /** Nested txn under this txn, set together with flag #MDB_TXN_HAS_CHILD */ + MDB_txn *mt_child; pgno_t mt_next_pgno; /**< next unallocated page */ /** The ID of this transaction. IDs are integers incrementing from 1. * Only committed write transactions increment the ID. If a transaction @@ -1002,13 +1079,15 @@ 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; /** Array of flags for each DB */ unsigned char *mt_dbflags; - /** Number of DB records in use. This number only ever increments; - * we don't decrement it when individual DB handles are closed. + /** Number of DB records in use, or 0 when the txn is finished. + * This number only ever increments until the txn finishes; we + * don't decrement it when individual DB handles are closed. */ MDB_dbi mt_numdbs; @@ -1016,10 +1095,18 @@ struct MDB_txn { * @ingroup internal * @{ */ -#define MDB_TXN_RDONLY 0x01 /**< read-only transaction */ + /** #mdb_txn_begin() flags */ +#define MDB_TXN_BEGIN_FLAGS MDB_RDONLY +#define MDB_TXN_RDONLY MDB_RDONLY /**< read-only transaction */ + /* internal txn flags */ +#define MDB_TXN_WRITEMAP MDB_WRITEMAP /**< copy of #MDB_env flag in writers */ +#define MDB_TXN_FINISHED 0x01 /**< txn is finished or never began */ #define MDB_TXN_ERROR 0x02 /**< txn is unusable after an error */ #define MDB_TXN_DIRTY 0x04 /**< must write, even if dirty list is empty */ #define MDB_TXN_SPILLS 0x08 /**< txn or a parent has spilled pages */ +#define MDB_TXN_HAS_CHILD 0x10 /**< txn has an #MDB_txn.%mt_child */ + /** most operations on the txn are currently illegal */ +#define MDB_TXN_BLOCKED (MDB_TXN_FINISHED|MDB_TXN_ERROR|MDB_TXN_HAS_CHILD) /** @} */ unsigned int mt_flags; /**< @ref mdb_txn */ /** #dirty_list room: Array size - \#dirty pages visible to this txn. @@ -1121,14 +1208,15 @@ struct MDB_env { unsigned int me_psize; /**< DB page size, inited from me_os_psize */ unsigned int me_os_psize; /**< OS page size, from #GET_PAGESIZE */ unsigned int me_maxreaders; /**< size of the reader table */ - unsigned int me_numreaders; /**< max numreaders set by this env */ + /** Max #MDB_txninfo.%mti_numreaders of interest to #mdb_env_close() */ + volatile int me_close_readers; MDB_dbi me_numdbs; /**< number of DBs opened */ MDB_dbi me_maxdbs; /**< size of the DB table */ MDB_PID_T me_pid; /**< process ID of this env */ char *me_path; /**< path to the DB files */ char *me_map; /**< the memory map of the data file */ MDB_txninfo *me_txns; /**< the memory map of the lock file or NULL */ - MDB_meta *me_metas[2]; /**< pointers to the two meta pages */ + MDB_meta *me_metas[NUM_METAS]; /**< pointers to the two meta pages */ void *me_pbuf; /**< scratch area for DUPSORT put() */ MDB_txn *me_txn; /**< current write transaction */ MDB_txn *me_txn0; /**< prealloc'd write transaction */ @@ -1158,11 +1246,13 @@ 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 */ @@ -1185,8 +1275,8 @@ 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) \ @@ -1196,6 +1286,19 @@ static int mdb_page_alloc(MDB_cursor *mc, int num, MDB_page **mp); static int mdb_page_new(MDB_cursor *mc, uint32_t flags, int num, MDB_page **mp); static int mdb_page_touch(MDB_cursor *mc); +#define MDB_END_NAMES {"committed", "empty-commit", "abort", "reset", \ + "reset-tmp", "fail-begin", "fail-beginchild"} +enum { + /* mdb_txn_end operation number, for logging */ + MDB_END_COMMITTED, MDB_END_EMPTY_COMMIT, MDB_END_ABORT, MDB_END_RESET, + MDB_END_RESET_TMP, MDB_END_FAIL_BEGIN, MDB_END_FAIL_BEGINCHILD +}; +#define MDB_END_OPMASK 0x0F /**< mask for #mdb_txn_end() operation number */ +#define MDB_END_UPDATE 0x10 /**< update env state (DBIs) */ +#define MDB_END_FREE 0x20 /**< free txn unless it is #MDB_env.%me_txn0 */ +#define MDB_END_SLOT MDB_NOTLS /**< release any reader slot if #MDB_NOTLS */ +static void mdb_txn_end(MDB_txn *txn, unsigned mode); + static int mdb_page_get(MDB_txn *txn, pgno_t pgno, MDB_page **mp, int *lvl); static int mdb_page_search_root(MDB_cursor *mc, MDB_val *key, int modify); @@ -1212,9 +1315,9 @@ static int mdb_page_split(MDB_cursor *mc, MDB_val *newkey, MDB_val *newdata, pgno_t newpgno, unsigned int nflags); static int mdb_env_read_header(MDB_env *env, MDB_meta *meta); -static int mdb_env_pick_meta(const MDB_env *env); +static MDB_meta *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); @@ -1252,6 +1355,7 @@ 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; @@ -1271,7 +1375,7 @@ static int mdb_sec_inited; #endif /** Return the library version info. */ -char * +char * ESECT mdb_version(int *major, int *minor, int *patch) { if (major) *major = MDB_VERSION_MAJOR; @@ -1286,7 +1390,7 @@ 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", @@ -1299,7 +1403,7 @@ static char *const mdb_errstr[] = { "MDB_MAP_RESIZED: Database contents grew beyond environment mapsize", "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_TXN: Transaction must abort, has a child, or is invalid", "MDB_BAD_VALSIZE: Unsupported size of key/DB name/data, or wrong DUPFIXED size", "MDB_BAD_DBI: The specified DBI handle was closed/changed unexpectedly", }; @@ -1343,7 +1447,7 @@ mdb_strerror(int err) ; } buf[0] = 0; - FormatMessage(FORMAT_MESSAGE_FROM_SYSTEM | + FormatMessageA(FORMAT_MESSAGE_FROM_SYSTEM | FORMAT_MESSAGE_IGNORE_INSERTS, NULL, err, 0, ptr, sizeof(buf), (va_list *)pad); return ptr; @@ -1355,7 +1459,7 @@ mdb_strerror(int err) /** assert(3) variant in cursor context */ #define mdb_cassert(mc, expr) mdb_assert0((mc)->mc_txn->mt_env, expr, #expr) /** assert(3) variant in transaction context */ -#define mdb_tassert(mc, expr) mdb_assert0((txn)->mt_env, expr, #expr) +#define mdb_tassert(txn, expr) mdb_assert0((txn)->mt_env, expr, #expr) /** assert(3) variant in environment context */ #define mdb_eassert(env, expr) mdb_assert0(env, expr, #expr) @@ -1363,7 +1467,7 @@ mdb_strerror(int err) # define mdb_assert0(env, expr, expr_txt) ((expr) ? (void)0 : \ mdb_assert_fail(env, expr_txt, mdb_func_, __FILE__, __LINE__)) -static void +static void ESECT mdb_assert_fail(MDB_env *env, const char *expr_txt, const char *func, const char *file, int line) { @@ -1561,9 +1665,10 @@ static void mdb_audit(MDB_txn *txn) mdb_tassert(txn, rc == MDB_NOTFOUND); } } - if (freecount + count + 2 /* metapages */ != txn->mt_next_pgno) { + if (freecount + count + NUM_METAS != txn->mt_next_pgno) { fprintf(stderr, "audit: %lu freecount: %lu count: %lu total: %lu next_pgno: %lu\n", - txn->mt_txnid, freecount, count+2, freecount+count+2, txn->mt_next_pgno); + txn->mt_txnid, freecount, count+NUM_METAS, + freecount+count+NUM_METAS, txn->mt_next_pgno); } } #endif @@ -1832,7 +1937,7 @@ mdb_page_spill(MDB_cursor *m0, MDB_val *key, MDB_val *data) /* 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) + if (m0->mc_dbi >= CORE_DBS) i += txn->mt_dbs[MAIN_DBI].md_depth; /* For puts, roughly factor in the key+data size */ if (key) @@ -1941,7 +2046,7 @@ mdb_page_dirty(MDB_txn *txn, MDB_page *mp) MDB_ID2 mid; int rc, (*insert)(MDB_ID2L, MDB_ID2 *); - if (txn->mt_env->me_flags & MDB_WRITEMAP) { + if (txn->mt_flags & MDB_TXN_WRITEMAP) { insert = mdb_mid2l_append; } else { insert = mdb_mid2l_insert; @@ -2440,12 +2545,6 @@ mdb_cursors_close(MDB_txn *txn, unsigned merge) } } -#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 @@ -2516,7 +2615,7 @@ mdb_txn_renew0(MDB_txn *txn) if ((flags &= MDB_TXN_RDONLY) != 0) { if (!ti) { - meta = env->me_metas[ mdb_env_pick_meta(env) ]; + meta = mdb_env_pick_meta(env); txn->mt_txnid = meta->mm_txnid; txn->mt_u.reader = NULL; } else { @@ -2528,6 +2627,7 @@ 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); @@ -2536,24 +2636,32 @@ 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; } - ti->mti_readers[i].mr_pid = pid; - ti->mti_readers[i].mr_tid = tid; + r = &ti->mti_readers[i]; + /* Claim the reader slot, carefully since other code + * uses the reader table un-mutexed: First reset the + * slot, next publish it in mti_numreaders. After + * that, it is safe for mdb_env_close() to touch it. + * When it will be closed, we can finally claim it. + */ + r->mr_pid = 0; + r->mr_txnid = (txnid_t)-1; + r->mr_tid = tid; if (i == nr) ti->mti_numreaders = ++nr; - /* Save numreaders for un-mutexed mdb_env_close() */ - env->me_numreaders = nr; - UNLOCK_MUTEX_R(env); + env->me_close_readers = nr; + r->mr_pid = pid; + UNLOCK_MUTEX(rmutex); - 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; @@ -2567,15 +2675,16 @@ mdb_txn_renew0(MDB_txn *txn) txn->mt_u.reader = r; meta = env->me_metas[txn->mt_txnid & 1]; } - txn->mt_dbxs = env->me_dbxs; /* mostly static anyway */ + } 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 { - meta = env->me_metas[ mdb_env_pick_meta(env) ]; + meta = mdb_env_pick_meta(env); txn->mt_txnid = meta->mm_txnid; } txn->mt_txnid++; @@ -2597,7 +2706,7 @@ mdb_txn_renew0(MDB_txn *txn) } /* Copy the DB info and flags */ - memcpy(txn->mt_dbs, meta->mm_dbs, 2 * sizeof(MDB_db)); + memcpy(txn->mt_dbs, meta->mm_dbs, CORE_DBS * sizeof(MDB_db)); /* Moved to here to avoid a data race in read TXNs */ txn->mt_next_pgno = meta->mm_last_pg+1; @@ -2606,23 +2715,24 @@ mdb_txn_renew0(MDB_txn *txn) /* Setup db info */ txn->mt_numdbs = env->me_numdbs; - for (i=2; imt_numdbs; i++) { + for (i=CORE_DBS; 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"); - if (new_notls) { - txn->mt_u.reader->mr_pid = 0; - txn->mt_u.reader = NULL; - } - return MDB_MAP_RESIZED; + if (env->me_flags & MDB_FATAL_ERROR) { + DPUTS("environment had fatal error, must shutdown!"); + rc = MDB_PANIC; + } else if (env->me_maxpg < txn->mt_next_pgno) { + rc = MDB_MAP_RESIZED; + } else { + return MDB_SUCCESS; } - - return MDB_SUCCESS; + mdb_txn_end(txn, new_notls /*0 or MDB_END_SLOT*/ | MDB_END_FAIL_BEGIN); + return rc; } int @@ -2630,14 +2740,9 @@ mdb_txn_renew(MDB_txn *txn) { int rc; - if (!txn || txn->mt_dbxs) /* A reset txn has mt_dbxs==NULL */ + if (!txn || !F_ISSET(txn->mt_flags, MDB_TXN_RDONLY|MDB_TXN_FINISHED)) return EINVAL; - if (txn->mt_env->me_flags & MDB_FATAL_ERROR) { - DPUTS("environment had fatal error, must shutdown!"); - return MDB_PANIC; - } - rc = mdb_txn_renew0(txn); if (rc == MDB_SUCCESS) { DPRINTF(("renew txn %"Z"u%c %p on mdbenv %p, root page %"Z"u", @@ -2652,60 +2757,47 @@ mdb_txn_begin(MDB_env *env, MDB_txn *parent, unsigned int flags, MDB_txn **ret) { MDB_txn *txn; MDB_ntxn *ntxn; - int rc, size, tsize = sizeof(MDB_txn); + int rc, size, tsize; - if (env->me_flags & MDB_FATAL_ERROR) { - DPUTS("environment had fatal error, must shutdown!"); - return MDB_PANIC; - } - if ((env->me_flags & MDB_RDONLY) && !(flags & MDB_RDONLY)) + flags &= MDB_TXN_BEGIN_FLAGS; + flags |= env->me_flags & MDB_WRITEMAP; + + if (env->me_flags & MDB_RDONLY & ~flags) /* write txn in RDONLY env */ return EACCES; + 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|MDB_TXN_ERROR)) || - (env->me_flags & MDB_WRITEMAP)) - { + flags |= parent->mt_flags; + if (flags & (MDB_RDONLY|MDB_WRITEMAP|MDB_TXN_BLOCKED)) { return (parent->mt_flags & MDB_TXN_RDONLY) ? EINVAL : MDB_BAD_TXN; } - tsize = sizeof(MDB_ntxn); - } - size = tsize; - if (!(flags & MDB_RDONLY)) { - if (!parent) { - txn = env->me_txn0; /* just reuse preallocated write txn */ - goto ok; - } - /* child txns use own copy of cursors */ - size += env->me_maxdbs * sizeof(MDB_cursor *); + /* Child txns save MDB_pgstate and use own copy of cursors */ + 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_dbxs = env->me_dbxs; /* static */ txn->mt_dbs = (MDB_db *) ((char *)txn + tsize); - if (flags & MDB_RDONLY) { - txn->mt_flags |= MDB_TXN_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; -ok: 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))) @@ -2719,11 +2811,10 @@ ok: txn->mt_u.dirty_list[0].mid = 0; txn->mt_spill_pgs = NULL; txn->mt_next_pgno = parent->mt_next_pgno; + parent->mt_flags |= MDB_TXN_HAS_CHILD; parent->mt_child = txn; txn->mt_parent = parent; txn->mt_numdbs = parent->mt_numdbs; - txn->mt_flags = parent->mt_flags; - txn->mt_dbxs = parent->mt_dbxs; memcpy(txn->mt_dbs, parent->mt_dbs, txn->mt_numdbs * sizeof(MDB_db)); /* Copy parent's mt_dbflags, but clear DB_NEW */ for (i=0; imt_numdbs; i++) @@ -2742,17 +2833,20 @@ ok: if (!rc) rc = mdb_cursor_shadow(parent, txn); if (rc) - mdb_txn_reset0(txn, "beginchild-fail"); - } else { + mdb_txn_end(txn, MDB_END_FAIL_BEGINCHILD); + } else { /* MDB_RDONLY */ + txn->mt_dbiseqs = env->me_dbiseqs; +renew: rc = mdb_txn_renew0(txn); } if (rc) { if (txn != env->me_txn0) free(txn); } else { + 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, (txn->mt_flags & MDB_TXN_RDONLY) ? 'r' : 'w', + txn->mt_txnid, (flags & MDB_RDONLY) ? 'r' : 'w', (void *) txn, (void *) env, txn->mt_dbs[MAIN_DBI].md_root)); } @@ -2782,7 +2876,7 @@ mdb_dbis_update(MDB_txn *txn, int keep) MDB_env *env = txn->mt_env; unsigned char *tdbflags = txn->mt_dbflags; - for (i = n; --i >= 2;) { + for (i = n; --i >= CORE_DBS;) { if (tdbflags[i] & DB_NEW) { if (keep) { env->me_dbflags[i] = txn->mt_dbs[i].md_flags | MDB_VALID; @@ -2802,39 +2896,52 @@ mdb_dbis_update(MDB_txn *txn, int keep) env->me_numdbs = n; } -/** Common code for #mdb_txn_reset() and #mdb_txn_abort(). +/** End a transaction, except successful commit of a nested transaction. * 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 + * @param[in] txn the transaction handle to end + * @param[in] mode why and how to end the transaction */ static void -mdb_txn_reset0(MDB_txn *txn, const char *act) +mdb_txn_end(MDB_txn *txn, unsigned mode) { MDB_env *env = txn->mt_env; +#if MDB_DEBUG + static const char *const names[] = MDB_END_NAMES; +#endif - /* Close any DBI handles opened in this txn */ - mdb_dbis_update(txn, 0); + /* Export or close DBI handles opened in this txn */ + mdb_dbis_update(txn, mode & MDB_END_UPDATE); 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', + names[mode & MDB_END_OPMASK], + txn->mt_txnid, (txn->mt_flags & MDB_TXN_RDONLY) ? 'r' : 'w', (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_u.reader->mr_txnid = (txnid_t)-1; - if (!(env->me_flags & MDB_NOTLS)) + if (!(env->me_flags & MDB_NOTLS)) { txn->mt_u.reader = NULL; /* txn does not own reader */ + } else if (mode & MDB_END_SLOT) { + txn->mt_u.reader->mr_pid = 0; + txn->mt_u.reader = NULL; + } /* else txn owns the slot until it does MDB_END_SLOT */ } - txn->mt_numdbs = 0; /* close nothing if called again */ - txn->mt_dbxs = NULL; /* mark txn as reset */ - } else { + txn->mt_numdbs = 0; /* prevent further DBI activity */ + txn->mt_flags |= MDB_TXN_FINISHED; + + } else if (!F_ISSET(txn->mt_flags, MDB_TXN_FINISHED)) { pgno_t *pghead = env->me_pghead; - mdb_cursors_close(txn, 0); + if (!(mode & MDB_END_UPDATE)) /* !(already closed cursors) */ + mdb_cursors_close(txn, 0); if (!(env->me_flags & MDB_WRITEMAP)) { mdb_dlist_free(txn); } + txn->mt_numdbs = 0; + txn->mt_flags = MDB_TXN_FINISHED; + if (!txn->mt_parent) { mdb_midl_shrink(&txn->mt_free_pgs); env->me_free_pgs = txn->mt_free_pgs; @@ -2843,11 +2950,14 @@ mdb_txn_reset0(MDB_txn *txn, const char *act) env->me_pglast = 0; env->me_txn = NULL; + mode = 0; /* txn == env->me_txn0, do not free() it */ + /* 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; + txn->mt_parent->mt_flags &= ~MDB_TXN_HAS_CHILD; env->me_pgstate = ((MDB_ntxn *)txn)->mnt_pgstate; mdb_midl_free(txn->mt_free_pgs); mdb_midl_free(txn->mt_spill_pgs); @@ -2856,6 +2966,9 @@ mdb_txn_reset0(MDB_txn *txn, const char *act) mdb_midl_free(pghead); } + + if (mode & MDB_END_FREE) + free(txn); } void @@ -2868,7 +2981,7 @@ mdb_txn_reset(MDB_txn *txn) if (!(txn->mt_flags & MDB_TXN_RDONLY)) return; - mdb_txn_reset0(txn, "reset"); + mdb_txn_end(txn, MDB_END_RESET); } void @@ -2880,13 +2993,7 @@ mdb_txn_abort(MDB_txn *txn) if (txn->mt_child) mdb_txn_abort(txn->mt_child); - mdb_txn_reset0(txn, "abort"); - /* Free reader slot tied to this txn (if MDB_NOTLS && writable FS) */ - if ((txn->mt_flags & MDB_TXN_RDONLY) && txn->mt_u.reader) - txn->mt_u.reader->mr_pid = 0; - - if (txn != txn->mt_env->me_txn0) - free(txn); + mdb_txn_end(txn, MDB_END_ABORT|MDB_END_SLOT|MDB_END_FREE); } /** Save the freelist as of this transaction to the freeDB. @@ -3237,15 +3344,17 @@ int mdb_txn_commit(MDB_txn *txn) { int rc; - unsigned int i; + unsigned int i, end_mode; MDB_env *env; - if (txn == NULL || txn->mt_env == NULL) + if (txn == NULL) return EINVAL; + /* mdb_txn_end() mode for a commit which writes nothing */ + end_mode = MDB_END_EMPTY_COMMIT|MDB_END_UPDATE|MDB_END_SLOT|MDB_END_FREE; + if (txn->mt_child) { rc = mdb_txn_commit(txn->mt_child); - txn->mt_child = NULL; if (rc) goto fail; } @@ -3253,14 +3362,11 @@ mdb_txn_commit(MDB_txn *txn) env = txn->mt_env; if (F_ISSET(txn->mt_flags, MDB_TXN_RDONLY)) { - mdb_dbis_update(txn, 1); - txn->mt_numdbs = 2; /* so txn_abort() doesn't close any new handles */ - mdb_txn_abort(txn); - return MDB_SUCCESS; + goto done; } - if (F_ISSET(txn->mt_flags, MDB_TXN_ERROR)) { - DPUTS("error flag is set, can't commit"); + if (txn->mt_flags & (MDB_TXN_FINISHED|MDB_TXN_ERROR)) { + DPUTS("txn has failed/finished, can't commit"); if (txn->mt_parent) txn->mt_parent->mt_flags |= MDB_TXN_ERROR; rc = MDB_BAD_TXN; @@ -3292,9 +3398,9 @@ mdb_txn_commit(MDB_txn *txn) /* Update parent's DB table. */ memcpy(parent->mt_dbs, txn->mt_dbs, txn->mt_numdbs * sizeof(MDB_db)); 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; imt_numdbs; i++) { + parent->mt_dbflags[FREE_DBI] = txn->mt_dbflags[FREE_DBI]; + parent->mt_dbflags[MAIN_DBI] = txn->mt_dbflags[MAIN_DBI]; + for (i=CORE_DBS; imt_numdbs; i++) { /* preserve parent's DB_NEW status */ x = parent->mt_dbflags[i] & DB_NEW; parent->mt_dbflags[i] = txn->mt_dbflags[i] | x; @@ -3395,14 +3501,14 @@ mdb_txn_commit(MDB_txn *txn) txn->mt_txnid, (void*)txn, (void*)env, txn->mt_dbs[MAIN_DBI].md_root)); /* Update DB root pointers */ - if (txn->mt_numdbs > 2) { + if (txn->mt_numdbs > CORE_DBS) { MDB_cursor mc; MDB_dbi i; MDB_val data; data.mv_size = sizeof(MDB_db); mdb_cursor_init(&mc, txn, MAIN_DBI, NULL); - for (i = 2; i < txn->mt_numdbs; i++) { + for (i = CORE_DBS; i < txn->mt_numdbs; i++) { if (txn->mt_dbflags[i] & DB_DIRTY) { if (TXN_DBI_CHANGED(txn, i)) { rc = MDB_BAD_DBI; @@ -3424,7 +3530,6 @@ mdb_txn_commit(MDB_txn *txn) mdb_midl_free(env->me_pghead); env->me_pghead = NULL; mdb_midl_shrink(&txn->mt_free_pgs); - env->me_free_pgs = txn->mt_free_pgs; #if (MDB_DEBUG) > 2 mdb_audit(txn); @@ -3434,21 +3539,10 @@ mdb_txn_commit(MDB_txn *txn) (rc = mdb_env_sync(env, 0)) || (rc = mdb_env_write_meta(txn))) goto fail; - - /* Free P_LOOSE pages left behind in dirty_list */ - if (!(env->me_flags & MDB_WRITEMAP)) - mdb_dlist_free(txn); + end_mode = MDB_END_COMMITTED|MDB_END_UPDATE; done: - env->me_pglast = 0; - env->me_txn = NULL; - mdb_dbis_update(txn, 1); - - if (env->me_txns) - UNLOCK_MUTEX_W(env); - if (txn != env->me_txn0) - free(txn); - + mdb_txn_end(txn, end_mode); return MDB_SUCCESS; fail: @@ -3475,7 +3569,7 @@ mdb_env_read_header(MDB_env *env, MDB_meta *meta) * Read both meta pages so we can use the latest one. */ - for (i=off=0; i<2; i++, off = meta->mm_psize) { + for (i=off=0; imm_psize) { #ifdef _WIN32 DWORD len; OVERLAPPED ov; @@ -3528,11 +3622,11 @@ mdb_env_init_meta0(MDB_env *env, MDB_meta *meta) meta->mm_version = MDB_DATA_VERSION; meta->mm_mapsize = env->me_mapsize; meta->mm_psize = env->me_psize; - meta->mm_last_pg = 1; + meta->mm_last_pg = NUM_METAS-1; meta->mm_flags = env->me_flags & 0xffff; - meta->mm_flags |= MDB_INTEGERKEY; - meta->mm_dbs[0].md_root = P_INVALID; - meta->mm_dbs[1].md_root = P_INVALID; + meta->mm_flags |= MDB_INTEGERKEY; /* this is mm_dbs[FREE_DBI].md_flags */ + meta->mm_dbs[FREE_DBI].md_root = P_INVALID; + meta->mm_dbs[MAIN_DBI].md_root = P_INVALID; } /** Write the environment parameters of a freshly created DB environment. @@ -3565,7 +3659,7 @@ mdb_env_init_meta(MDB_env *env, MDB_meta *meta) psize = env->me_psize; - p = calloc(2, psize); + p = calloc(NUM_METAS, psize); if (!p) return ENOMEM; @@ -3578,10 +3672,10 @@ mdb_env_init_meta(MDB_env *env, MDB_meta *meta) q->mp_flags = P_META; *(MDB_meta *)METADATA(q) = *meta; - DO_PWRITE(rc, env->me_fd, p, psize * 2, len, 0); + DO_PWRITE(rc, env->me_fd, p, psize * NUM_METAS, len, 0); if (!rc) rc = ErrCode(); - else if ((unsigned) len == psize * 2) + else if ((unsigned) len == psize * NUM_METAS) rc = MDB_SUCCESS; else rc = ENOSPC; @@ -3598,6 +3692,7 @@ mdb_env_write_meta(MDB_txn *txn) { MDB_env *env; MDB_meta meta, metab, *mp; + unsigned flags; size_t mapsize; off_t off; int rc, len, toggle; @@ -3614,30 +3709,33 @@ mdb_env_write_meta(MDB_txn *txn) toggle, txn->mt_dbs[MAIN_DBI].md_root)); env = txn->mt_env; + flags = env->me_flags; mp = env->me_metas[toggle]; mapsize = env->me_metas[toggle ^ 1]->mm_mapsize; /* Persist any increases of mapsize config */ if (mapsize < env->me_mapsize) mapsize = env->me_mapsize; - if (env->me_flags & MDB_WRITEMAP) { + if (flags & MDB_WRITEMAP) { mp->mm_mapsize = mapsize; - mp->mm_dbs[0] = txn->mt_dbs[0]; - mp->mm_dbs[1] = txn->mt_dbs[1]; + mp->mm_dbs[FREE_DBI] = txn->mt_dbs[FREE_DBI]; + mp->mm_dbs[MAIN_DBI] = txn->mt_dbs[MAIN_DBI]; mp->mm_last_pg = txn->mt_next_pgno - 1; +#if (__GNUC__ * 100 + __GNUC_MINOR__ >= 404) && /* TODO: portability */ \ + !(defined(__i386__) || defined(__x86_64__)) + /* LY: issue a memory barrier, if not x86. ITS#7969 */ + __sync_synchronize(); +#endif mp->mm_txnid = txn->mt_txnid; - if (!(env->me_flags & (MDB_NOMETASYNC|MDB_NOSYNC))) { + if (!(flags & (MDB_NOMETASYNC|MDB_NOSYNC))) { unsigned meta_size = env->me_psize; rc = (env->me_flags & MDB_MAPASYNC) ? MS_ASYNC : MS_SYNC; - ptr = env->me_map; - if (toggle) { + ptr = (char *)mp - PAGEHDRSZ; #ifndef _WIN32 /* POSIX msync() requires ptr = start of OS page */ - if (meta_size < env->me_os_psize) - meta_size += meta_size; - else + r2 = (ptr - env->me_map) & (env->me_os_psize - 1); + ptr -= r2; + meta_size += r2; #endif - ptr += meta_size; - } if (MDB_MSYNC(ptr, meta_size, rc)) { rc = ErrCode(); goto fail; @@ -3645,25 +3743,22 @@ mdb_env_write_meta(MDB_txn *txn) } goto done; } - metab.mm_txnid = env->me_metas[toggle]->mm_txnid; - metab.mm_last_pg = env->me_metas[toggle]->mm_last_pg; + metab.mm_txnid = mp->mm_txnid; + metab.mm_last_pg = mp->mm_last_pg; meta.mm_mapsize = mapsize; - meta.mm_dbs[0] = txn->mt_dbs[0]; - meta.mm_dbs[1] = txn->mt_dbs[1]; + meta.mm_dbs[FREE_DBI] = txn->mt_dbs[FREE_DBI]; + meta.mm_dbs[MAIN_DBI] = txn->mt_dbs[MAIN_DBI]; meta.mm_last_pg = txn->mt_next_pgno - 1; meta.mm_txnid = txn->mt_txnid; off = offsetof(MDB_meta, mm_mapsize); ptr = (char *)&meta + off; len = sizeof(MDB_meta) - off; - if (toggle) - off += env->me_psize; - off += PAGEHDRSZ; + off += (char *)mp - env->me_map; /* Write to the SYNC fd */ - mfd = env->me_flags & (MDB_NOSYNC|MDB_NOMETASYNC) ? - env->me_fd : env->me_mfd; + mfd = (flags & (MDB_NOSYNC|MDB_NOMETASYNC)) ? env->me_fd : env->me_mfd; #ifdef _WIN32 { memset(&ov, 0, sizeof(ov)); @@ -3717,12 +3812,13 @@ done: /** Check both meta pages to see which one is newer. * @param[in] env the environment handle - * @return meta toggle (0 or 1). + * @return newest #MDB_meta. */ -static int +static MDB_meta * mdb_env_pick_meta(const MDB_env *env) { - return (env->me_metas[0]->mm_txnid < env->me_metas[1]->mm_txnid); + MDB_meta *const *metas = env->me_metas; + return metas[ metas[0]->mm_txnid < metas[1]->mm_txnid ]; } int ESECT @@ -3735,7 +3831,7 @@ mdb_env_create(MDB_env **env) return ENOMEM; e->me_maxreaders = DEFAULT_READERS; - e->me_maxdbs = e->me_numdbs = 2; + e->me_maxdbs = e->me_numdbs = CORE_DBS; e->me_fd = INVALID_HANDLE_VALUE; e->me_lfd = INVALID_HANDLE_VALUE; e->me_mfd = INVALID_HANDLE_VALUE; @@ -3846,7 +3942,7 @@ mdb_env_set_mapsize(MDB_env *env, size_t size) void *old; if (env->me_txn) return EINVAL; - meta = env->me_metas[mdb_env_pick_meta(env)]; + meta = mdb_env_pick_meta(env); if (!size) size = meta->mm_mapsize; { @@ -3873,7 +3969,7 @@ mdb_env_set_maxdbs(MDB_env *env, MDB_dbi dbs) { if (env->me_map) return EINVAL; - env->me_maxdbs = dbs + 2; /* Named databases + main and free DB */ + env->me_maxdbs = dbs + CORE_DBS; return MDB_SUCCESS; } @@ -3938,6 +4034,7 @@ mdb_env_open2(MDB_env *env) else env->me_pidquery = PROCESS_QUERY_INFORMATION; #endif /* _WIN32 */ + #ifdef BROKEN_FDATASYNC /* ext3/ext4 fdatasync is broken on some older Linux kernels. * https://lkml.org/lkml/2012/9/3/83 @@ -4052,12 +4149,12 @@ mdb_env_open2(MDB_env *env) #if MDB_DEBUG { - int toggle = mdb_env_pick_meta(env); - MDB_db *db = &env->me_metas[toggle]->mm_dbs[MAIN_DBI]; + MDB_meta *meta = mdb_env_pick_meta(env); + MDB_db *db = &meta->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)); + meta->mm_version, env->me_psize)); + DPRINTF(("using meta page %d", (int) (meta->mm_txnid & 1))); DPRINTF(("depth: %u", db->md_depth)); DPRINTF(("entries: %"Z"u", db->md_entries)); DPRINTF(("branch pages: %"Z"u", db->md_branch_pages)); @@ -4144,9 +4241,10 @@ PIMAGE_TLS_CALLBACK mdb_tls_cbp = mdb_tls_callback; static int ESECT mdb_env_share_locks(MDB_env *env, int *excl) { - int rc = 0, toggle = mdb_env_pick_meta(env); + int rc = 0; + MDB_meta *meta = mdb_env_pick_meta(env); - env->me_txns->mti_txnid = env->me_metas[toggle]->mm_txnid; + env->me_txns->mti_txnid = meta->mm_txnid; #ifdef _WIN32 { @@ -4211,8 +4309,8 @@ 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; @@ -4295,7 +4393,7 @@ mdb_hash_val(MDB_val *val, mdb_hash_t hval) */ static const char mdb_a85[]= "0123456789ABCDEFGHIJKLMNOPQRSTUVWXYZabcdefghijklmnopqrstuvwxyz!#$%&()*+-;<=>?@^_`{|}~"; -static void +static void ESECT mdb_pack85(unsigned long l, char *out) { int i; @@ -4306,7 +4404,7 @@ mdb_pack85(unsigned long l, char *out) } } -static void +static void ESECT mdb_hash_enc(MDB_val *val, char *encbuf) { mdb_hash_t h = mdb_hash_val(val, MDB_HASH_INIT); @@ -4342,7 +4440,7 @@ mdb_env_setup_locks(MDB_env *env, char *lpath, int mode, int *excl) off_t size, rsize; #ifdef _WIN32 - env->me_lfd = CreateFile(lpath, GENERIC_READ|GENERIC_WRITE, + env->me_lfd = CreateFileA(lpath, GENERIC_READ|GENERIC_WRITE, FILE_SHARE_READ|FILE_SHARE_WRITE, NULL, OPEN_ALWAYS, FILE_ATTRIBUTE_NORMAL, NULL); #else @@ -4446,9 +4544,9 @@ mdb_env_setup_locks(MDB_env *env, char *lpath, int mode, int *excl) 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); + env->me_rmutex = CreateMutexA(&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); + env->me_wmutex = CreateMutexA(&mdb_all_sa, FALSE, env->me_txns->mti_wmname); if (!env->me_wmutex) goto fail_errno; #elif defined(MDB_USE_POSIX_SEM) struct stat stbuf; @@ -4484,13 +4582,16 @@ 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 */ @@ -4517,9 +4618,9 @@ mdb_env_setup_locks(MDB_env *env, char *lpath, int mode, int *excl) goto fail; } #ifdef _WIN32 - env->me_rmutex = OpenMutex(SYNCHRONIZE, FALSE, env->me_txns->mti_rmname); + env->me_rmutex = OpenMutexA(SYNCHRONIZE, FALSE, env->me_txns->mti_rmname); if (!env->me_rmutex) goto fail_errno; - env->me_wmutex = OpenMutex(SYNCHRONIZE, FALSE, env->me_txns->mti_wmname); + env->me_wmutex = OpenMutexA(SYNCHRONIZE, FALSE, env->me_txns->mti_wmname); if (!env->me_wmutex) goto fail_errno; #elif defined(MDB_USE_POSIX_SEM) env->me_rmutex = sem_open(env->me_txns->mti_rmname, 0); @@ -4547,8 +4648,8 @@ 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" @@ -4604,6 +4705,7 @@ 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))) { @@ -4621,7 +4723,7 @@ mdb_env_open(MDB_env *env, const char *path, unsigned int flags, mdb_mode_t mode len = OPEN_ALWAYS; } mode = FILE_ATTRIBUTE_NORMAL; - env->me_fd = CreateFile(dpath, oflags, FILE_SHARE_READ|FILE_SHARE_WRITE, + env->me_fd = CreateFileA(dpath, oflags, FILE_SHARE_READ|FILE_SHARE_WRITE, NULL, len, mode, NULL); #else if (F_ISSET(flags, MDB_RDONLY)) @@ -4651,7 +4753,7 @@ mdb_env_open(MDB_env *env, const char *path, unsigned int flags, mdb_mode_t mode */ #ifdef _WIN32 len = OPEN_EXISTING; - env->me_mfd = CreateFile(dpath, oflags, + env->me_mfd = CreateFileA(dpath, oflags, FILE_SHARE_READ|FILE_SHARE_WRITE, NULL, len, mode | FILE_FLAG_WRITE_THROUGH, NULL); #else @@ -4669,21 +4771,20 @@ mdb_env_open(MDB_env *env, const char *path, unsigned int flags, mdb_mode_t mode if (rc) goto leave; } - if (!((flags & MDB_RDONLY) || - (env->me_pbuf = calloc(1, env->me_psize)))) - rc = ENOMEM; if (!(flags & MDB_RDONLY)) { MDB_txn *txn; int tsize = sizeof(MDB_txn), size = tsize + env->me_maxdbs * (sizeof(MDB_db)+sizeof(MDB_cursor *)+sizeof(unsigned int)+1); - txn = calloc(1, size); - if (txn) { + if ((env->me_pbuf = calloc(1, env->me_psize)) && + (txn = calloc(1, size))) + { txn->mt_dbs = (MDB_db *)((char *)txn + tsize); txn->mt_cursors = (MDB_cursor **)(txn->mt_dbs + env->me_maxdbs); txn->mt_dbiseqs = (unsigned int *)(txn->mt_cursors + env->me_maxdbs); txn->mt_dbflags = (unsigned char *)(txn->mt_dbiseqs + env->me_maxdbs); txn->mt_env = env; txn->mt_dbxs = env->me_dbxs; + txn->mt_flags = MDB_TXN_FINISHED; env->me_txn0 = txn; } else { rc = ENOMEM; @@ -4710,7 +4811,7 @@ mdb_env_close0(MDB_env *env, int excl) /* Doing this here since me_dbxs may not exist during mdb_env_close */ if (env->me_dbxs) { - for (i = env->me_maxdbs; --i > MAIN_DBI; ) + for (i = env->me_maxdbs; --i >= CORE_DBS; ) free(env->me_dbxs[i].md_name.mv_data); free(env->me_dbxs); } @@ -4747,8 +4848,12 @@ mdb_env_close0(MDB_env *env, int excl) MDB_PID_T pid = env->me_pid; /* Clearing readers is done in this function because * me_txkey with its destructor must be disabled first. + * + * We skip the the reader mutex, so we touch only + * data owned by this process (me_close_readers and + * our readers), and clear each reader atomically. */ - for (i = env->me_numreaders; --i >= 0; ) + for (i = env->me_close_readers; --i >= 0; ) if (env->me_txns->mti_readers[i].mr_pid == pid) env->me_txns->mti_readers[i].mr_pid = 0; #ifdef _WIN32 @@ -4792,7 +4897,6 @@ mdb_env_close0(MDB_env *env, int excl) env->me_flags &= ~(MDB_ENV_ACTIVE|MDB_ENV_TXKEY); } - void ESECT mdb_env_close(MDB_env *env) { @@ -5024,15 +5128,12 @@ 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)); } } @@ -5069,7 +5170,7 @@ mdb_page_get(MDB_txn *txn, pgno_t pgno, MDB_page **ret, int *lvl) MDB_page *p = NULL; int level; - if (!((txn->mt_flags & MDB_TXN_RDONLY) | (env->me_flags & MDB_WRITEMAP))) { + if (! (txn->mt_flags & (MDB_TXN_RDONLY|MDB_TXN_WRITEMAP))) { MDB_txn *tx2 = txn; level = 1; do { @@ -5226,8 +5327,8 @@ mdb_page_search(MDB_cursor *mc, MDB_val *key, int flags) /* Make sure the txn is still viable, then find the root from * 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"); + if (mc->mc_txn->mt_flags & MDB_TXN_BLOCKED) { + DPUTS("transaction may not be used now"); return MDB_BAD_TXN; } else { /* Make sure we're using an up-to-date root */ @@ -5411,10 +5512,10 @@ 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) + if (txn->mt_flags & MDB_TXN_BLOCKED) return MDB_BAD_TXN; mdb_cursor_init(&mc, txn, dbi, &mx); @@ -5935,7 +6036,7 @@ mdb_cursor_get(MDB_cursor *mc, MDB_val *key, MDB_val *data, if (mc == NULL) return EINVAL; - if (mc->mc_txn->mt_flags & MDB_TXN_ERROR) + if (mc->mc_txn->mt_flags & MDB_TXN_BLOCKED) return MDB_BAD_TXN; switch (op) { @@ -6106,7 +6207,7 @@ mdb_cursor_touch(MDB_cursor *mc) { int rc = MDB_SUCCESS; - if (mc->mc_dbi > MAIN_DBI && !(*mc->mc_dbflag & DB_DIRTY)) { + if (mc->mc_dbi >= CORE_DBS && !(*mc->mc_dbflag & DB_DIRTY)) { MDB_cursor mc2; MDB_xcursor mcx; if (TXN_DBI_CHANGED(mc->mc_txn, mc->mc_dbi)) @@ -6134,7 +6235,6 @@ 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; @@ -6166,7 +6266,7 @@ mdb_cursor_put(MDB_cursor *mc, MDB_val *key, MDB_val *data, nospill = flags & MDB_NOSPILL; flags &= ~MDB_NOSPILL; - if (mc->mc_txn->mt_flags & (MDB_TXN_RDONLY|MDB_TXN_ERROR)) + if (mc->mc_txn->mt_flags & (MDB_TXN_RDONLY|MDB_TXN_BLOCKED)) return (mc->mc_txn->mt_flags & MDB_TXN_RDONLY) ? EACCES : MDB_BAD_TXN; if (key->mv_size-1 >= ENV_MAXKEY(env)) @@ -6291,16 +6391,18 @@ fix_parent: * update branch key if there is a parent page */ if (mc->mc_top && !mc->mc_ki[mc->mc_top]) { - unsigned short top = mc->mc_top; + unsigned short dtop = 1; mc->mc_top--; /* slot 0 is always an empty key, find real slot */ - while (mc->mc_top && !mc->mc_ki[mc->mc_top]) + while (mc->mc_top && !mc->mc_ki[mc->mc_top]) { mc->mc_top--; + dtop++; + } if (mc->mc_ki[mc->mc_top]) rc2 = mdb_update_key(mc, key); else rc2 = MDB_SUCCESS; - mc->mc_top = top; + mc->mc_top += dtop; if (rc2) return rc2; } @@ -6659,7 +6761,7 @@ mdb_cursor_del(MDB_cursor *mc, unsigned int flags) MDB_page *mp; int rc; - if (mc->mc_txn->mt_flags & (MDB_TXN_RDONLY|MDB_TXN_ERROR)) + if (mc->mc_txn->mt_flags & (MDB_TXN_RDONLY|MDB_TXN_BLOCKED)) return (mc->mc_txn->mt_flags & MDB_TXN_RDONLY) ? EACCES : MDB_BAD_TXN; if (!(mc->mc_flags & C_INITIALIZED)) @@ -6868,6 +6970,7 @@ mdb_node_add(MDB_cursor *mc, indx_t indx, MDB_node *node; MDB_page *mp = mc->mc_pg[mc->mc_top]; MDB_page *ofp = NULL; /* overflow page */ + void *ndata; DKBUF; mdb_cassert(mc, mp->mp_upper >= mp->mp_lower); @@ -6898,7 +7001,7 @@ mdb_node_add(MDB_cursor *mc, indx_t indx, if (key != NULL) node_size += key->mv_size; if (IS_LEAF(mp)) { - mdb_cassert(mc, data); + mdb_cassert(mc, key && data); if (F_ISSET(flags, F_BIGDATA)) { /* Data already on overflow page. */ node_size += sizeof(pgno_t); @@ -6949,23 +7052,21 @@ update: memcpy(NODEKEY(node), key->mv_data, key->mv_size); if (IS_LEAF(mp)) { - mdb_cassert(mc, key); + ndata = NODEDATA(node); if (ofp == NULL) { if (F_ISSET(flags, F_BIGDATA)) - memcpy(node->mn_data + key->mv_size, data->mv_data, - sizeof(pgno_t)); + memcpy(ndata, data->mv_data, sizeof(pgno_t)); else if (F_ISSET(flags, MDB_RESERVE)) - data->mv_data = node->mn_data + key->mv_size; + data->mv_data = ndata; else - memcpy(node->mn_data + key->mv_size, data->mv_data, - data->mv_size); + memcpy(ndata, data->mv_data, data->mv_size); } else { - memcpy(node->mn_data + key->mv_size, &ofp->mp_pgno, - sizeof(pgno_t)); + memcpy(ndata, &ofp->mp_pgno, sizeof(pgno_t)); + ndata = METADATA(ofp); if (F_ISSET(flags, MDB_RESERVE)) - data->mv_data = METADATA(ofp); + data->mv_data = ndata; else - memcpy(METADATA(ofp), data->mv_data, data->mv_size); + memcpy(ndata, data->mv_data, data->mv_size); } } @@ -7047,45 +7148,38 @@ 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; } @@ -7138,7 +7232,7 @@ mdb_xcursor_init1(MDB_cursor *mc, MDB_node *node) mx->mx_cursor.mc_flags = C_SUB; } else { MDB_page *fp = NODEDATA(node); - mx->mx_db.md_pad = mc->mc_pg[mc->mc_top]->mp_pad; + mx->mx_db.md_pad = 0; mx->mx_db.md_flags = 0; mx->mx_db.md_depth = 1; mx->mx_db.md_branch_pages = 0; @@ -7160,7 +7254,7 @@ 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; @@ -7233,14 +7327,13 @@ 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) + if (txn->mt_flags & MDB_TXN_BLOCKED) return MDB_BAD_TXN; - /* Allow read access to the freelist */ - if (!dbi && !F_ISSET(txn->mt_flags, MDB_TXN_RDONLY)) + if (dbi == FREE_DBI && !F_ISSET(txn->mt_flags, MDB_TXN_RDONLY)) return EINVAL; if (txn->mt_dbs[dbi].md_flags & MDB_DUPSORT) @@ -7265,13 +7358,13 @@ 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) return EINVAL; - if (txn->mt_flags & MDB_TXN_ERROR) + if (txn->mt_flags & MDB_TXN_BLOCKED) return MDB_BAD_TXN; mdb_cursor_init(mc, txn, mc->mc_dbi, mc->mc_xcursor); @@ -7290,7 +7383,7 @@ mdb_cursor_count(MDB_cursor *mc, size_t *countp) if (mc->mc_xcursor == NULL) return MDB_INCOMPATIBLE; - if (mc->mc_txn->mt_flags & MDB_TXN_ERROR) + if (mc->mc_txn->mt_flags & MDB_TXN_BLOCKED) return MDB_BAD_TXN; if (!(mc->mc_flags & C_INITIALIZED)) @@ -7517,32 +7610,48 @@ mdb_node_move(MDB_cursor *csrc, MDB_cursor *cdst) /* Adjust other cursors pointing to mp */ MDB_cursor *m2, *m3; MDB_dbi dbi = csrc->mc_dbi; - MDB_page *mp; - - mp = cdst->mc_pg[csrc->mc_top]; - 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; - else - m3 = m2; - if (m3 == cdst) continue; - if (m3->mc_pg[csrc->mc_top] == mp && m3->mc_ki[csrc->mc_top] >= - cdst->mc_ki[csrc->mc_top]) { - m3->mc_ki[csrc->mc_top]++; + MDB_page *mpd, *mps; + + mps = csrc->mc_pg[csrc->mc_top]; + /* If we're adding on the left, bump others up */ + if (!cdst->mc_ki[csrc->mc_top]) { + mpd = cdst->mc_pg[csrc->mc_top]; + 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; + else + m3 = m2; + if (m3 != cdst && + m3->mc_pg[csrc->mc_top] == mpd && + m3->mc_ki[csrc->mc_top] >= cdst->mc_ki[csrc->mc_top]) { + m3->mc_ki[csrc->mc_top]++; + } + if (m3 !=csrc && + m3->mc_pg[csrc->mc_top] == mps && + m3->mc_ki[csrc->mc_top] == csrc->mc_ki[csrc->mc_top]) { + m3->mc_pg[csrc->mc_top] = cdst->mc_pg[cdst->mc_top]; + m3->mc_ki[csrc->mc_top] = cdst->mc_ki[cdst->mc_top]; + m3->mc_ki[csrc->mc_top-1]++; + } } - } - - mp = csrc->mc_pg[csrc->mc_top]; - 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; - else - m3 = m2; - if (m3 == csrc) continue; - if (m3->mc_pg[csrc->mc_top] == mp && m3->mc_ki[csrc->mc_top] == - csrc->mc_ki[csrc->mc_top]) { - m3->mc_pg[csrc->mc_top] = cdst->mc_pg[cdst->mc_top]; - m3->mc_ki[csrc->mc_top] = cdst->mc_ki[cdst->mc_top]; + } else + /* Adding on the right, bump others down */ + { + 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; + else + m3 = m2; + if (m3 == csrc) continue; + if (m3->mc_pg[csrc->mc_top] == mps) { + if (!m3->mc_ki[csrc->mc_top]) { + m3->mc_pg[csrc->mc_top] = cdst->mc_pg[cdst->mc_top]; + m3->mc_ki[csrc->mc_top] = cdst->mc_ki[cdst->mc_top]; + m3->mc_ki[csrc->mc_top-1]--; + } else { + m3->mc_ki[csrc->mc_top]--; + } + } } } } @@ -7638,6 +7747,9 @@ mdb_page_merge(MDB_cursor *csrc, MDB_cursor *cdst) if ((rc = mdb_page_touch(cdst))) return rc; + /* get dst page again now that we've touched it. */ + pdst = cdst->mc_pg[cdst->mc_top]; + /* Move all nodes from src to dst. */ j = nkeys = NUMKEYS(pdst); @@ -7715,6 +7827,7 @@ mdb_page_merge(MDB_cursor *csrc, MDB_cursor *cdst) /* Adjust other cursors pointing to mp */ MDB_cursor *m2, *m3; MDB_dbi dbi = csrc->mc_dbi; + unsigned int top = csrc->mc_top; for (m2 = csrc->mc_txn->mt_cursors[dbi]; m2; m2=m2->mc_next) { if (csrc->mc_flags & C_SUB) @@ -7723,9 +7836,10 @@ mdb_page_merge(MDB_cursor *csrc, MDB_cursor *cdst) m3 = m2; if (m3 == csrc) continue; if (m3->mc_snum < csrc->mc_snum) continue; - if (m3->mc_pg[csrc->mc_top] == psrc) { - m3->mc_pg[csrc->mc_top] = pdst; - m3->mc_ki[csrc->mc_top] += nkeys; + if (m3->mc_pg[top] == psrc) { + m3->mc_pg[top] = pdst; + m3->mc_ki[top] += nkeys; + m3->mc_ki[top-1] = cdst->mc_ki[top-1]; } } } @@ -7861,9 +7975,9 @@ mdb_rebalance(MDB_cursor *mc) m3 = &m2->mc_xcursor->mx_cursor; else m3 = m2; - if (m3 == mc || m3->mc_snum < mc->mc_snum) continue; + if (m3 == mc) continue; if (m3->mc_pg[0] == mp) { - for (i=0; imc_snum; i++) { + for (i=0; imc_db->md_depth; i++) { m3->mc_pg[i] = m3->mc_pg[i+1]; m3->mc_ki[i] = m3->mc_ki[i+1]; } @@ -7928,7 +8042,8 @@ mdb_rebalance(MDB_cursor *mc) */ if (PAGEFILL(mc->mc_txn->mt_env, mn.mc_pg[mn.mc_top]) >= thresh && NUMKEYS(mn.mc_pg[mn.mc_top]) > minkeys) { rc = mdb_node_move(&mn, mc); - if (mc->mc_ki[mc->mc_top-1]) { + if (!mc->mc_ki[mc->mc_top]) { + /* if we inserted on left, bump position up */ oldki++; } } else { @@ -8037,10 +8152,10 @@ 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)) + if (txn->mt_flags & (MDB_TXN_RDONLY|MDB_TXN_BLOCKED)) return (txn->mt_flags & MDB_TXN_RDONLY) ? EACCES : MDB_BAD_TXN; if (!F_ISSET(txn->mt_dbs[dbi].md_flags, MDB_DUPSORT)) { @@ -8133,14 +8248,22 @@ mdb_page_split(MDB_cursor *mc, MDB_val *newkey, MDB_val *newdata, pgno_t newpgno /* Create a right sibling. */ if ((rc = mdb_page_new(mc, mp->mp_flags, 1, &rp))) return rc; + rp->mp_pad = mp->mp_pad; DPRINTF(("new right sibling: page %"Z"u", rp->mp_pgno)); - if (mc->mc_snum < 2) { + /* Usually when splitting the root page, the cursor + * height is 1. But when called from mdb_update_key, + * the cursor height may be greater because it walks + * up the stack while finding the branch slot to update. + */ + if (mc->mc_top < 1) { if ((rc = mdb_page_new(mc, P_BRANCH, 1, &pp))) goto done; /* shift current top to make room for new parent */ - mc->mc_pg[1] = mc->mc_pg[0]; - mc->mc_ki[1] = mc->mc_ki[0]; + for (i=mc->mc_snum; i>0; i--) { + mc->mc_pg[i] = mc->mc_pg[i-1]; + mc->mc_ki[i] = mc->mc_ki[i-1]; + } mc->mc_pg[0] = pp; mc->mc_ki[0] = 0; mc->mc_db->md_root = pp->mp_pgno; @@ -8156,8 +8279,8 @@ mdb_page_split(MDB_cursor *mc, MDB_val *newkey, MDB_val *newdata, pgno_t newpgno mc->mc_db->md_depth--; goto done; } - mc->mc_snum = 2; - mc->mc_top = 1; + mc->mc_snum++; + mc->mc_top++; ptop = 0; } else { ptop = mc->mc_top-1; @@ -8216,6 +8339,7 @@ mdb_page_split(MDB_cursor *mc, MDB_val *newkey, MDB_val *newdata, pgno_t newpgno rp->mp_upper -= ksize - sizeof(indx_t); mc->mc_ki[mc->mc_top] = x; mc->mc_pg[mc->mc_top] = rp; + mc->mc_ki[ptop]++; } } else { int psize, nsize, k; @@ -8502,12 +8626,15 @@ 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; + if (txn->mt_flags & (MDB_TXN_RDONLY|MDB_TXN_BLOCKED)) + return (txn->mt_flags & MDB_TXN_RDONLY) ? EACCES : MDB_BAD_TXN; + mdb_cursor_init(&mc, txn, dbi, &mx); return mdb_cursor_put(&mc, key, data, flags); } @@ -8812,7 +8939,7 @@ mdb_env_copyfd1(MDB_env *env, HANDLE fd) my.mc_wlen[1] = 0; my.mc_olen[0] = 0; my.mc_olen[1] = 0; - my.mc_next_pgno = 2; + my.mc_next_pgno = NUM_METAS; my.mc_status = 0; my.mc_new = 1; my.mc_toggle = 0; @@ -8825,7 +8952,7 @@ mdb_env_copyfd1(MDB_env *env, HANDLE fd) return rc; mp = (MDB_page *)my.mc_wbuf[0]; - memset(mp, 0, 2*env->me_psize); + memset(mp, 0, NUM_METAS * env->me_psize); mp->mp_pgno = 0; mp->mp_flags = P_META; mm = (MDB_meta *)METADATA(mp); @@ -8848,27 +8975,27 @@ mdb_env_copyfd1(MDB_env *env, HANDLE fd) mdb_cursor_init(&mc, txn, FREE_DBI, NULL); while ((rc = mdb_cursor_get(&mc, &key, &data, MDB_NEXT)) == 0) freecount += *(MDB_ID *)data.mv_data; - freecount += txn->mt_dbs[0].md_branch_pages + - txn->mt_dbs[0].md_leaf_pages + - txn->mt_dbs[0].md_overflow_pages; + freecount += txn->mt_dbs[FREE_DBI].md_branch_pages + + txn->mt_dbs[FREE_DBI].md_leaf_pages + + txn->mt_dbs[FREE_DBI].md_overflow_pages; /* Set metapage 1 */ mm->mm_last_pg = txn->mt_next_pgno - freecount - 1; - mm->mm_dbs[1] = txn->mt_dbs[1]; - if (mm->mm_last_pg > 1) { - mm->mm_dbs[1].md_root = mm->mm_last_pg; + mm->mm_dbs[MAIN_DBI] = txn->mt_dbs[MAIN_DBI]; + if (mm->mm_last_pg > NUM_METAS-1) { + mm->mm_dbs[MAIN_DBI].md_root = mm->mm_last_pg; mm->mm_txnid = 1; } else { - mm->mm_dbs[1].md_root = P_INVALID; + mm->mm_dbs[MAIN_DBI].md_root = P_INVALID; } } - my.mc_wlen[0] = env->me_psize * 2; + my.mc_wlen[0] = env->me_psize * NUM_METAS; my.mc_txn = txn; pthread_mutex_lock(&my.mc_mutex); while(my.mc_new) pthread_cond_wait(&my.mc_cond, &my.mc_mutex); pthread_mutex_unlock(&my.mc_mutex); - rc = mdb_env_cwalk(&my, &txn->mt_dbs[1].md_root, 0); + rc = mdb_env_cwalk(&my, &txn->mt_dbs[MAIN_DBI].md_root, 0); if (rc == MDB_SUCCESS && my.mc_wlen[my.mc_toggle]) rc = mdb_env_cthr_toggle(&my, 1); mdb_env_cthr_toggle(&my, -1); @@ -8896,6 +9023,7 @@ 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; @@ -8917,19 +9045,21 @@ mdb_env_copyfd0(MDB_env *env, HANDLE fd) if (env->me_txns) { /* We must start the actual read txn after blocking writers */ - mdb_txn_reset0(txn, "reset-stage1"); + mdb_txn_end(txn, MDB_END_RESET_TMP); /* 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; } } - wsize = env->me_psize * 2; + wsize = env->me_psize * NUM_METAS; ptr = env->me_map; w2 = wsize; while (w2 > 0) { @@ -8948,8 +9078,8 @@ 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; @@ -9026,7 +9156,7 @@ mdb_env_copy2(MDB_env *env, const char *path, unsigned int flags) * already in the OS cache. */ #ifdef _WIN32 - newfd = CreateFile(lpath, GENERIC_WRITE, 0, NULL, CREATE_NEW, + newfd = CreateFileA(lpath, GENERIC_WRITE, 0, NULL, CREATE_NEW, FILE_FLAG_NO_BUFFERING|FILE_FLAG_WRITE_THROUGH, NULL); #else newfd = open(lpath, O_WRONLY|O_CREAT|O_EXCL, 0666); @@ -9072,7 +9202,7 @@ 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; @@ -9087,7 +9217,7 @@ 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; } @@ -9159,36 +9289,32 @@ mdb_stat0(MDB_env *env, MDB_db *db, MDB_stat *arg) int ESECT mdb_env_stat(MDB_env *env, MDB_stat *arg) { - int toggle; + MDB_meta *meta; if (env == NULL || arg == NULL) return EINVAL; - toggle = mdb_env_pick_meta(env); + meta = mdb_env_pick_meta(env); - return mdb_stat0(env, &env->me_metas[toggle]->mm_dbs[MAIN_DBI], arg); + return mdb_stat0(env, &meta->mm_dbs[MAIN_DBI], arg); } int ESECT mdb_env_info(MDB_env *env, MDB_envinfo *arg) { - int toggle; + MDB_meta *meta; if (env == NULL || arg == NULL) return EINVAL; - toggle = mdb_env_pick_meta(env); - arg->me_mapaddr = env->me_metas[toggle]->mm_address; + meta = mdb_env_pick_meta(env); + arg->me_mapaddr = meta->mm_address; + arg->me_last_pgno = meta->mm_last_pg; + arg->me_last_txnid = meta->mm_txnid; + arg->me_mapsize = env->me_mapsize; arg->me_maxreaders = env->me_maxreaders; - - /* 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; + arg->me_numreaders = env->me_txns ? env->me_txns->mti_numreaders : 0; return MDB_SUCCESS; } @@ -9225,13 +9351,9 @@ 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) + if (txn->mt_flags & MDB_TXN_BLOCKED) return MDB_BAD_TXN; /* main DB? */ @@ -9255,7 +9377,7 @@ int mdb_dbi_open(MDB_txn *txn, const char *name, unsigned int flags, MDB_dbi *db /* Is the DB already open? */ len = strlen(name); - for (i=2; imt_numdbs; i++) { + for (i=CORE_DBS; imt_numdbs; i++) { if (!txn->mt_dbxs[i].md_name.mv_size) { /* Remember this free slot */ if (!unused) unused = i; @@ -9277,7 +9399,7 @@ 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; @@ -9323,12 +9445,13 @@ int mdb_dbi_open(MDB_txn *txn, const char *name, unsigned int flags, MDB_dbi *db return rc; } -int mdb_stat(MDB_txn *txn, MDB_dbi dbi, MDB_stat *arg) +int ESECT +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) + if (txn->mt_flags & MDB_TXN_BLOCKED) return MDB_BAD_TXN; if (txn->mt_dbflags[dbi] & DB_STALE) { @@ -9343,7 +9466,7 @@ int mdb_stat(MDB_txn *txn, MDB_dbi dbi, MDB_stat *arg) void mdb_dbi_close(MDB_env *env, MDB_dbi dbi) { char *ptr; - if (dbi <= MAIN_DBI || dbi >= env->me_maxdbs) + if (dbi < CORE_DBS || dbi >= env->me_maxdbs) return; ptr = env->me_dbxs[dbi].md_name.mv_data; /* If there was no name, this was already closed */ @@ -9359,7 +9482,7 @@ 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; @@ -9382,8 +9505,10 @@ 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); @@ -9457,13 +9582,13 @@ 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); @@ -9478,7 +9603,7 @@ int mdb_drop(MDB_txn *txn, MDB_dbi dbi, int del) goto leave; /* Can't delete the main DB */ - if (del && dbi > MAIN_DBI) { + if (del && dbi >= CORE_DBS) { rc = mdb_del0(txn, MAIN_DBI, &mc->mc_dbx->md_name, NULL, F_SUBDATA); if (!rc) { txn->mt_dbflags[dbi] = DB_STALE; @@ -9505,7 +9630,7 @@ 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; @@ -9514,7 +9639,7 @@ 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; @@ -9523,7 +9648,7 @@ 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; @@ -9532,7 +9657,7 @@ 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; @@ -9626,17 +9751,23 @@ 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 ESECT +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) @@ -9644,22 +9775,32 @@ 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. + */ + meta = mdb_env_pick_meta(env); + env->me_txns->mti_txnid = meta->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 */ /** @} */