X-Git-Url: https://git.sur5r.net/?a=blobdiff_plain;f=libraries%2Flibmdb%2Fmdb.c;h=eddeb5e8993c0f46c39bf66d6ac17d5c5ddbaad5;hb=75b97f03e4137325b1141f8fa6f4034ef03b47a0;hp=b913d7d7d1702f8c0c915f895aace4b0d8d1af9c;hpb=bed3123b828a83d0a3f07e87a70fad1251e84ab8;p=openldap diff --git a/libraries/libmdb/mdb.c b/libraries/libmdb/mdb.c index b913d7d7d1..eddeb5e899 100644 --- a/libraries/libmdb/mdb.c +++ b/libraries/libmdb/mdb.c @@ -5,7 +5,7 @@ * BerkeleyDB API, but much simplified. */ /* - * Copyright 2011 Howard Chu, Symas Corp. + * Copyright 2011-2012 Howard Chu, Symas Corp. * All rights reserved. * * Redistribution and use in source and binary forms, with or without @@ -48,21 +48,76 @@ #include #include +#include #include -#include +#include #include #include #include #include #include +#if !(defined(BYTE_ORDER) || defined(__BYTE_ORDER)) +#include /* defines BYTE_ORDER on HPUX and Solaris */ +#endif + #ifndef _WIN32 #include +#ifdef __APPLE__ +#include +#endif +#endif + +#ifdef USE_VALGRIND +#include +#define VGMEMP_CREATE(h,r,z) VALGRIND_CREATE_MEMPOOL(h,r,z) +#define VGMEMP_ALLOC(h,a,s) VALGRIND_MEMPOOL_ALLOC(h,a,s) +#define VGMEMP_FREE(h,a) VALGRIND_MEMPOOL_FREE(h,a) +#define VGMEMP_DESTROY(h) VALGRIND_DESTROY_MEMPOOL(h) +#define VGMEMP_DEFINED(a,s) VALGRIND_MAKE_MEM_DEFINED(a,s) +#else +#define VGMEMP_CREATE(h,r,z) +#define VGMEMP_ALLOC(h,a,s) +#define VGMEMP_FREE(h,a) +#define VGMEMP_DESTROY(h) +#define VGMEMP_DEFINED(a,s) +#endif + +#ifndef BYTE_ORDER +# if (defined(_LITTLE_ENDIAN) || defined(_BIG_ENDIAN)) && !(defined(_LITTLE_ENDIAN) && defined(_BIG_ENDIAN)) +/* Solaris just defines one or the other */ +# define LITTLE_ENDIAN 1234 +# define BIG_ENDIAN 4321 +# ifdef _LITTLE_ENDIAN +# define BYTE_ORDER LITTLE_ENDIAN +# else +# define BYTE_ORDER BIG_ENDIAN +# endif +# else +# define BYTE_ORDER __BYTE_ORDER +# endif +#endif + +#ifndef LITTLE_ENDIAN +#define LITTLE_ENDIAN __LITTLE_ENDIAN +#endif +#ifndef BIG_ENDIAN +#define BIG_ENDIAN __BIG_ENDIAN +#endif + +#if defined(__i386) || defined(__x86_64) +#define MISALIGNED_OK 1 #endif #include "mdb.h" #include "midl.h" +#if (BYTE_ORDER == LITTLE_ENDIAN) == (BYTE_ORDER == BIG_ENDIAN) +# error "Unknown or unsupported endianness (BYTE_ORDER)" +#elif (-6 & 5) || CHAR_BIT != 8 || UINT_MAX < 0xffffffff || ULONG_MAX % 0xFFFF +# error "Two's complement, reasonably sized integer types, please" +#endif + /** @defgroup internal MDB Internals * @{ */ @@ -78,38 +133,49 @@ #define pthread_mutex_t HANDLE #define pthread_key_t DWORD #define pthread_self() GetCurrentThreadId() -#define pthread_key_create(x,y) *(x) = TlsAlloc() +#define pthread_key_create(x,y) (*(x) = TlsAlloc()) #define pthread_key_delete(x) TlsFree(x) #define pthread_getspecific(x) TlsGetValue(x) #define pthread_setspecific(x,y) TlsSetValue(x,y) #define pthread_mutex_unlock(x) ReleaseMutex(x) #define pthread_mutex_lock(x) WaitForSingleObject(x, 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_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 getpid() GetCurrentProcessId() -#define fdatasync(fd) !FlushFileBuffers(fd) +#define fdatasync(fd) (!FlushFileBuffers(fd)) #define ErrCode() GetLastError() -#define GetPageSize(x) {SYSTEM_INFO si; GetSystemInfo(&si); (x) = si.dwPageSize;} +#define GET_PAGESIZE(x) {SYSTEM_INFO si; GetSystemInfo(&si); (x) = si.dwPageSize;} #define close(fd) CloseHandle(fd) #define munmap(ptr,len) UnmapViewOfFile(ptr) #else +#ifdef __APPLE__ +#define LOCK_MUTEX_R(env) sem_wait((env)->me_rmutex) +#define UNLOCK_MUTEX_R(env) sem_post((env)->me_rmutex) +#define LOCK_MUTEX_W(env) sem_wait((env)->me_wmutex) +#define UNLOCK_MUTEX_W(env) sem_post((env)->me_wmutex) +#define fdatasync(fd) fsync(fd) +#else +#ifdef ANDROID +#define fdatasync(fd) fsync(fd) +#endif /** Lock the reader mutex. */ -#define LOCK_MUTEX_R(env) pthread_mutex_lock(&env->me_txns->mti_mutex) +#define LOCK_MUTEX_R(env) pthread_mutex_lock(&(env)->me_txns->mti_mutex) /** Unlock the reader mutex. */ -#define UNLOCK_MUTEX_R(env) pthread_mutex_unlock(&env->me_txns->mti_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_MUTEX_W(env) pthread_mutex_lock(&env->me_txns->mti_wmutex) +#define LOCK_MUTEX_W(env) pthread_mutex_lock(&(env)->me_txns->mti_wmutex) /** Unlock the writer mutex. */ -#define UNLOCK_MUTEX_W(env) pthread_mutex_unlock(&env->me_txns->mti_wmutex) +#define UNLOCK_MUTEX_W(env) pthread_mutex_unlock(&(env)->me_txns->mti_wmutex) +#endif /* __APPLE__ */ /** Get the error code for the last failed system function. */ @@ -125,13 +191,19 @@ * Mainly used to initialize file variables and signify that they are * unused. */ -#define INVALID_HANDLE_VALUE -1 +#define INVALID_HANDLE_VALUE (-1) /** Get the size of a memory page for the system. * This is the basic size that the platform's memory manager uses, and is * fundamental to the use of memory-mapped files. */ -#define GetPageSize(x) (x) = sysconf(_SC_PAGE_SIZE) +#define GET_PAGESIZE(x) ((x) = sysconf(_SC_PAGE_SIZE)) +#endif + +#if defined(_WIN32) || defined(__APPLE__) +#define MNAME_LEN 32 +#else +#define MNAME_LEN (sizeof(pthread_mutex_t)) #endif /** @} */ @@ -149,6 +221,13 @@ #ifndef MDB_DSYNC # define MDB_DSYNC O_DSYNC #endif +#endif + +/** Function for flushing the data of a file. Define this to fsync + * if fdatasync() is not supported. + */ +#ifndef MDB_FDATASYNC +# define MDB_FDATASYNC fdatasync #endif /** A page number in the database. @@ -159,25 +238,30 @@ * @note In the #MDB_node structure, we only store 48 bits of this value, * which thus limits us to only 60 bits of addressable data. */ -typedef ULONG pgno_t; +typedef ID pgno_t; + + /** A transaction ID. + * See struct MDB_txn.mt_txnid for details. + */ +typedef ID txnid_t; /** @defgroup debug Debug Macros * @{ */ -#ifndef DEBUG +#ifndef MDB_DEBUG /** Enable debug output. * Set this to 1 for copious tracing. Set to 2 to add dumps of all IDLs * read from and written to the database (used for free space management). */ -#define DEBUG 0 +#define MDB_DEBUG 0 #endif #if !(__STDC_VERSION__ >= 199901L || defined(__GNUC__)) # define DPRINTF (void) /* Vararg macros may be unsupported */ -#elif DEBUG +#elif MDB_DEBUG /** Print a debug message with printf formatting. */ # define DPRINTF(fmt, ...) /**< Requires 2 or more args */ \ - fprintf(stderr, "%s:%d:(%p) " fmt "\n", __func__, __LINE__, pthread_self(), __VA_ARGS__) + fprintf(stderr, "%s:%d " fmt "\n", __func__, __LINE__, __VA_ARGS__) #else # define DPRINTF(fmt, ...) ((void) 0) #endif @@ -190,7 +274,7 @@ typedef ULONG pgno_t; /** A default memory page size. * The actual size is platform-dependent, but we use this for * boot-strapping. We probably should not be using this any more. - * The #GetPageSize() macro is used to get the actual size. + * The #GET_PAGESIZE() macro is used to get the actual size. * * Note that we don't currently support Huge pages. On Linux, * regular data files cannot use Huge pages, and in general @@ -199,7 +283,7 @@ typedef ULONG pgno_t; * pressure from other processes is high. So until OSs have * actual paging support for Huge pages, they're not viable. */ -#define PAGESIZE 4096 +#define MDB_PAGESIZE 4096 /** The minimum number of keys required in a database page. * Setting this to a larger value will place a smaller bound on the @@ -227,11 +311,11 @@ typedef ULONG pgno_t; /** The maximum size of a key in the database. * While data items have essentially unbounded size, we require that * keys all fit onto a regular page. This limit could be raised a bit - * further if needed; to something just under #PAGESIZE / #MDB_MINKEYS. + * further if needed; to something just under #MDB_PAGESIZE / #MDB_MINKEYS. */ #define MAXKEYSIZE 511 -#if DEBUG +#if MDB_DEBUG /** A key buffer. * @ingroup debug * This is used for printing a hex dump of a key's contents. @@ -243,12 +327,12 @@ typedef ULONG pgno_t; */ #define DKEY(x) mdb_dkey(x, kbuf) #else -#define DKBUF -#define DKEY(x) +#define DKBUF typedef int dummy_kbuf /* so we can put ';' after */ +#define DKEY(x) 0 #endif /** @defgroup lazylock Lazy Locking - * Macros for locks that are't actually needed. + * Macros for locks that aren't actually needed. * The DB view is always consistent because all writes are wrapped in * the wmutex. Finer-grained locks aren't necessary. * @{ @@ -268,7 +352,7 @@ typedef ULONG pgno_t; #define LAZY_RWLOCK_WRLOCK(x) /** Grab the DB table read lock */ #define LAZY_RWLOCK_RDLOCK(x) - /** Declare the DB table rwlock */ + /** Declare the DB table rwlock. Should not be followed by ';'. */ #define LAZY_RWLOCK_DEF(x) /** Initialize the DB table rwlock */ #define LAZY_RWLOCK_INIT(x,y) @@ -280,7 +364,7 @@ typedef ULONG pgno_t; #define LAZY_RWLOCK_UNLOCK(x) pthread_rwlock_unlock(x) #define LAZY_RWLOCK_WRLOCK(x) pthread_rwlock_wrlock(x) #define LAZY_RWLOCK_RDLOCK(x) pthread_rwlock_rdlock(x) -#define LAZY_RWLOCK_DEF(x) pthread_rwlock_t x +#define LAZY_RWLOCK_DEF(x) pthread_rwlock_t x; #define LAZY_RWLOCK_INIT(x,y) pthread_rwlock_init(x,y) #define LAZY_RWLOCK_DESTROY(x) pthread_rwlock_destroy(x) #endif @@ -316,8 +400,8 @@ typedef uint16_t indx_t; * Since the database uses multi-version concurrency control, readers don't * actually need any locking. This table is used to keep track of which * readers are using data from which old transactions, so that we'll know - * when a particular old transaction is no longer in use, Old transactions - * that have freed any data pages can then have their freed pages reclaimed + * when a particular old transaction is no longer in use. Old transactions + * that have discarded any data pages can then have those pages reclaimed * for use by a later write transaction. * * The lock table is constructed such that reader slots are aligned with the @@ -375,7 +459,7 @@ typedef struct MDB_rxbody { * started from so we can avoid overwriting any data used in that * particular version. */ - ULONG mrb_txnid; + txnid_t mrb_txnid; /** The process ID of the process owning this reader txn. */ pid_t mrb_pid; /** The thread ID of the thread owning this txn. */ @@ -402,18 +486,21 @@ typedef struct MDB_reader { * For POSIX the actual mutexes reside in the shared memory of this * mapped file. On Windows, mutexes are named objects allocated by the * kernel; we store the mutex names in this mapped file so that other - * processes can grab them. This same approach will also be used on + * processes can grab them. This same approach is also used on * MacOSX/Darwin (using named semaphores) since MacOSX doesn't support - * process-shared POSIX mutexes. + * process-shared POSIX mutexes. For these cases where a named object + * is used, the object name is derived from a 64 bit FNV hash of the + * environment pathname. As such, naming collisions are extremely + * unlikely. If a collision occurs, the results are unpredictable. */ typedef struct MDB_txbody { - /** Stamp identifying this as an MDB lock file. It must be set + /** Stamp identifying this as an MDB file. It must be set * to #MDB_MAGIC. */ uint32_t mtb_magic; /** Version number of this lock file. Must be set to #MDB_VERSION. */ uint32_t mtb_version; -#ifdef _WIN32 - char mtb_rmname[32]; +#if defined(_WIN32) || defined(__APPLE__) + char mtb_rmname[MNAME_LEN]; #else /** Mutex protecting access to this table. * This is the reader lock that #LOCK_MUTEX_R acquires. @@ -424,12 +511,12 @@ typedef struct MDB_txbody { * This is recorded here only for convenience; the value can always * be determined by reading the main database meta pages. */ - ULONG mtb_txnid; + txnid_t mtb_txnid; /** The number of slots that have been used in the reader table. * This always records the maximum count, it is not decremented * when readers release their slots. */ - uint32_t mtb_numreaders; + unsigned mtb_numreaders; /** The ID of the most recent meta page in the database. * This is recorded here only for convenience; the value can always * be determined by reading the main database meta pages. @@ -451,42 +538,55 @@ typedef struct MDB_txninfo { char pad[(sizeof(MDB_txbody)+CACHELINE-1) & ~(CACHELINE-1)]; } mt1; union { -#ifdef _WIN32 - char mt2_wmname[32]; +#if defined(_WIN32) || defined(__APPLE__) + char mt2_wmname[MNAME_LEN]; #define mti_wmname mt2.mt2_wmname #else pthread_mutex_t mt2_wmutex; #define mti_wmutex mt2.mt2_wmutex #endif - char pad[(sizeof(pthread_mutex_t)+CACHELINE-1) & ~(CACHELINE-1)]; + char pad[(MNAME_LEN+CACHELINE-1) & ~(CACHELINE-1)]; } mt2; MDB_reader mti_readers[1]; } MDB_txninfo; /** @} */ /** Common header for all page types. - * Overflow pages occupy a number of contiguous pages with no + * Overflow records occupy a number of contiguous pages with no * headers on any page after the first. */ typedef struct MDB_page { +#define mp_pgno mp_p.p_pgno +#define mp_next mp_p.p_next union { - pgno_t mp_pgno; /**< page number */ - void * mp_next; /**< for in-memory list of freed structs */ - }; + pgno_t p_pgno; /**< page number */ + void * p_next; /**< for in-memory list of freed structs */ + } mp_p; + uint16_t mp_pad; +/** @defgroup mdb_page Page Flags + * @ingroup internal + * Flags for the page headers. + * @{ + */ #define P_BRANCH 0x01 /**< branch page */ #define P_LEAF 0x02 /**< leaf page */ #define P_OVERFLOW 0x04 /**< overflow page */ #define P_META 0x08 /**< meta page */ #define P_DIRTY 0x10 /**< dirty page */ #define P_LEAF2 0x20 /**< for #MDB_DUPFIXED records */ - uint32_t mp_flags; +#define P_SUBP 0x40 /**< for #MDB_DUPSORT sub-pages */ +/** @} */ + uint16_t mp_flags; /**< @ref mdb_page */ +#define mp_lower mp_pb.pb.pb_lower +#define mp_upper mp_pb.pb.pb_upper +#define mp_pages mp_pb.pb_pages union { struct { - indx_t mp_lower; /**< lower bound of free space */ - indx_t mp_upper; /**< upper bound of free space */ - }; - uint32_t mp_pages; /**< number of overflow pages */ - }; + indx_t pb_lower; /**< lower bound of free space */ + indx_t pb_upper; /**< upper bound of free space */ + } pb; + uint32_t pb_pages; /**< number of overflow pages */ + } mp_pb; indx_t mp_ptrs[1]; /**< dynamic size */ } MDB_page; @@ -518,6 +618,8 @@ typedef struct MDB_page { #define IS_BRANCH(p) F_ISSET((p)->mp_flags, P_BRANCH) /** Test if a page is an overflow page */ #define IS_OVERFLOW(p) F_ISSET((p)->mp_flags, P_OVERFLOW) + /** Test if a page is a sub page */ +#define IS_SUBP(p) F_ISSET((p)->mp_flags, P_SUBP) /** The number of overflow pages needed to store the given size. */ #define OVPAGES(size, psize) ((PAGEHDRSZ-1 + (size)) / (psize) + 1) @@ -528,14 +630,27 @@ typedef struct MDB_page { typedef struct MDB_node { /** lo and hi are used for data size on leaf nodes and for * child pgno on branch nodes. On 64 bit platforms, flags - * is also used for pgno. (branch nodes ignore flags) + * is also used for pgno. (Branch nodes have no flags). + * They are in host byte order in case that lets some + * accesses be optimized into a 32-bit word access. */ - unsigned short mn_lo; - unsigned short mn_hi; /**< part of dsize or pgno */ - unsigned short mn_flags; /**< flags for special node types */ +#define mn_lo mn_offset[BYTE_ORDER!=LITTLE_ENDIAN] +#define mn_hi mn_offset[BYTE_ORDER==LITTLE_ENDIAN] /**< part of dsize or pgno */ + unsigned short mn_offset[2]; /**< storage for #mn_lo and #mn_hi */ +/** @defgroup mdb_node Node Flags + * @ingroup internal + * Flags for node headers. + * @{ + */ #define F_BIGDATA 0x01 /**< data put on overflow page */ #define F_SUBDATA 0x02 /**< data is a sub-database */ #define F_DUPDATA 0x04 /**< data has duplicates */ + +/** valid flags for #mdb_node_add() */ +#define NODE_ADD_FLAGS (F_DUPDATA|F_SUBDATA|MDB_RESERVE|MDB_APPEND) + +/** @} */ + unsigned short mn_flags; /**< @ref mdb_node */ unsigned short mn_ksize; /**< key size */ char mn_data[1]; /**< key and data are appended here */ } MDB_node; @@ -543,17 +658,20 @@ typedef struct MDB_node { /** Size of the node header, excluding dynamic data at the end */ #define NODESIZE offsetof(MDB_node, mn_data) - /** Size of a node in a branch page. + /** Bit position of top word in page number, for shifting mn_flags */ +#define PGNO_TOPWORD ((pgno_t)-1 > 0xffffffffu ? 32 : 0) + + /** Size of a node in a branch page with a given key. * This is just the node header plus the key, there is no data. */ #define INDXSIZE(k) (NODESIZE + ((k) == NULL ? 0 : (k)->mv_size)) - /** Size of a node in a leaf page. + /** Size of a node in a leaf page with a given key and data. * This is node header plus key plus data size. */ #define LEAFSIZE(k, d) (NODESIZE + (k)->mv_size + (d)->mv_size) - /** Address of node \i in page \p */ + /** Address of node \b i in page \b p */ #define NODEPTR(p, i) ((MDB_node *)((char *)(p) + (p)->mp_ptrs[i])) /** Address of the key for the node */ @@ -563,18 +681,13 @@ typedef struct MDB_node { #define NODEDATA(node) (void *)((char *)(node)->mn_data + (node)->mn_ksize) /** Get the page number pointed to by a branch node */ -#if LONG_MAX == 0x7fffffff -#define NODEPGNO(node) ((node)->mn_lo | ((node)->mn_hi << 16)) - /** Set the page number in a branch node */ -#define SETPGNO(node,pgno) do { \ - (node)->mn_lo = (pgno) & 0xffff; (node)->mn_hi = (pgno) >> 16;} while(0) -#else -#define NODEPGNO(node) ((node)->mn_lo | ((node)->mn_hi << 16) | ((unsigned long)(node)->mn_flags << 32)) +#define NODEPGNO(node) \ + ((node)->mn_lo | ((pgno_t) (node)->mn_hi << 16) | \ + (PGNO_TOPWORD ? ((pgno_t) (node)->mn_flags << PGNO_TOPWORD) : 0)) /** Set the page number in a branch node */ #define SETPGNO(node,pgno) do { \ (node)->mn_lo = (pgno) & 0xffff; (node)->mn_hi = (pgno) >> 16; \ - (node)->mn_flags = (pgno) >> 32; } while(0) -#endif + if (PGNO_TOPWORD) (node)->mn_flags = (pgno) >> PGNO_TOPWORD; } while(0) /** Get the size of the data in a leaf node */ #define NODEDSZ(node) ((node)->mn_lo | ((unsigned)(node)->mn_hi << 16)) @@ -584,6 +697,30 @@ typedef struct MDB_node { /** The size of a key in a node */ #define NODEKSZ(node) ((node)->mn_ksize) + /** Copy a page number from src to dst */ +#ifdef MISALIGNED_OK +#define COPY_PGNO(dst,src) dst = src +#else +#if SIZE_MAX > 4294967295UL +#define COPY_PGNO(dst,src) do { \ + unsigned short *s, *d; \ + s = (unsigned short *)&(src); \ + d = (unsigned short *)&(dst); \ + *d++ = *s++; \ + *d++ = *s++; \ + *d++ = *s++; \ + *d = *s; \ +} while (0) +#else +#define COPY_PGNO(dst,src) do { \ + unsigned short *s, *d; \ + s = (unsigned short *)&(src); \ + d = (unsigned short *)&(dst); \ + *d++ = *s++; \ + *d = *s; \ +} while (0) +#endif +#endif /** The address of a key in a LEAF2 page. * LEAF2 pages are used for #MDB_DUPFIXED sorted-duplicate sub-DBs. * There are no node headers, keys are stored contiguously. @@ -591,17 +728,18 @@ typedef struct MDB_node { #define LEAF2KEY(p, i, ks) ((char *)(p) + PAGEHDRSZ + ((i)*(ks))) /** Set the \b node's key into \b key, if requested. */ -#define MDB_SET_KEY(node, key) if (key!=NULL) {(key)->mv_size = NODEKSZ(node); (key)->mv_data = NODEKEY(node);} +#define MDB_SET_KEY(node, key) { if ((key) != NULL) { \ + (key)->mv_size = NODEKSZ(node); (key)->mv_data = NODEKEY(node); } } /** Information about a single database in the environment. */ typedef struct MDB_db { uint32_t md_pad; /**< also ksize for LEAF2 pages */ uint16_t md_flags; /**< @ref mdb_open */ uint16_t md_depth; /**< depth of this tree */ - ULONG md_branch_pages; /**< number of internal pages */ - ULONG md_leaf_pages; /**< number of leaf pages */ - ULONG md_overflow_pages; /**< number of overflow pages */ - ULONG md_entries; /**< number of data items */ + pgno_t md_branch_pages; /**< number of internal pages */ + pgno_t md_leaf_pages; /**< number of leaf pages */ + pgno_t md_overflow_pages; /**< number of overflow pages */ + size_t md_entries; /**< number of data items */ pgno_t md_root; /**< the root page of this tree */ } MDB_db; @@ -612,7 +750,7 @@ typedef struct MDB_db { /** Meta page content. */ typedef struct MDB_meta { - /** Stamp identifying this as an MDB data file. It must be set + /** Stamp identifying this as an MDB file. It must be set * to #MDB_MAGIC. */ uint32_t mm_magic; /** Version number of this lock file. Must be set to #MDB_VERSION. */ @@ -625,40 +763,51 @@ typedef struct MDB_meta { /** Any persistent environment flags. @ref mdb_env */ #define mm_flags mm_dbs[0].md_flags pgno_t mm_last_pg; /**< last used page in file */ - ULONG mm_txnid; /**< txnid that committed this page */ + txnid_t mm_txnid; /**< txnid that committed this page */ } MDB_meta; + /** Buffer for a stack-allocated dirty page. + * The members define size and alignment, and silence type + * aliasing warnings. They are not used directly; that could + * mean incorrectly using several union members in parallel. + */ +typedef union MDB_pagebuf { + char mb_raw[MDB_PAGESIZE]; + MDB_page mb_page; + struct { + char mm_pad[PAGEHDRSZ]; + MDB_meta mm_meta; + } mb_metabuf; +} MDB_pagebuf; + /** Auxiliary DB info. * The information here is mostly static/read-only. There is * only a single copy of this record in the environment. - * The \b md_dirty flag is not read-only, but only a write - * transaction can ever update it, and only write transactions - * need to worry about it. */ typedef struct MDB_dbx { MDB_val md_name; /**< name of the database */ MDB_cmp_func *md_cmp; /**< function for comparing keys */ MDB_cmp_func *md_dcmp; /**< function for comparing data items */ MDB_rel_func *md_rel; /**< user relocate function */ - MDB_dbi md_parent; /**< parent DB of a sub-DB */ - unsigned int md_dirty; /**< TRUE if DB was written in this txn */ + void *md_relctx; /**< user-provided context for md_rel */ } MDB_dbx; /** A database transaction. * Every operation requires a transaction handle. */ struct MDB_txn { + MDB_txn *mt_parent; /**< parent of a nested txn */ + MDB_txn *mt_child; /**< nested txn under this txn */ 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 * aborts, the ID may be re-used by the next writer. */ - ULONG mt_txnid; + txnid_t mt_txnid; MDB_env *mt_env; /**< the DB environment */ /** The list of pages that became unused during this transaction. - * This is an #IDL. */ - pgno_t *mt_free_pgs; + IDL mt_free_pgs; union { ID2L dirty_list; /**< modified pages */ MDB_reader *reader; /**< this thread's slot in the reader table */ @@ -667,14 +816,30 @@ struct MDB_txn { MDB_dbx *mt_dbxs; /** Array of MDB_db records for each known DB */ MDB_db *mt_dbs; +/** @defgroup mt_dbflag Transaction DB Flags + * @ingroup internal + * @{ + */ +#define DB_DIRTY 0x01 /**< DB was written in this txn */ +#define DB_STALE 0x02 /**< DB record is older than txnID */ +/** @} */ + /** 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. */ - unsigned int mt_numdbs; + MDB_dbi mt_numdbs; +/** @defgroup mdb_txn Transaction Flags + * @ingroup internal + * @{ + */ #define MDB_TXN_RDONLY 0x01 /**< read-only transaction */ #define MDB_TXN_ERROR 0x02 /**< an error has occurred */ - unsigned int mt_flags; +/** @} */ + unsigned int mt_flags; /**< @ref mdb_txn */ /** Tracks which of the two meta pages was used at the start * of this transaction. */ @@ -691,18 +856,36 @@ struct MDB_xcursor; /** Cursors are used for all DB operations */ struct MDB_cursor { + /** Next cursor on this DB in this txn */ + MDB_cursor *mc_next; + /** Original cursor if this is a shadow */ + MDB_cursor *mc_orig; /** Context used for databases with #MDB_DUPSORT, otherwise NULL */ struct MDB_xcursor *mc_xcursor; /** The transaction that owns this cursor */ MDB_txn *mc_txn; /** The database handle this cursor operates on */ MDB_dbi mc_dbi; + /** The database record for this cursor */ + MDB_db *mc_db; + /** The database auxiliary record for this cursor */ + MDB_dbx *mc_dbx; + /** The @ref mt_dbflag for this database */ + unsigned char *mc_dbflag; unsigned short mc_snum; /**< number of pushed pages */ - unsigned short mc_top; /**< index of top page, mc_snum-1 */ - unsigned int mc_flags; + unsigned short mc_top; /**< index of top page, normally mc_snum-1 */ +/** @defgroup mdb_cursor Cursor Flags + * @ingroup internal + * Cursor state flags. + * @{ + */ #define C_INITIALIZED 0x01 /**< cursor has been initialized and is valid */ #define C_EOF 0x02 /**< No more data */ -#define C_XDIRTY 0x04 /**< @deprecated mc_xcursor needs to be flushed */ +#define C_SUB 0x04 /**< Cursor is a sub-cursor */ +#define C_SHADOW 0x08 /**< Cursor is a dup from a parent txn */ +#define C_ALLOCD 0x10 /**< Cursor was malloc'd */ +/** @} */ + unsigned int mc_flags; /**< @ref mdb_cursor */ MDB_page *mc_pg[CURSOR_STACK]; /**< stack of pushed pages */ indx_t mc_ki[CURSOR_STACK]; /**< stack of page indices */ }; @@ -715,20 +898,12 @@ struct MDB_cursor { typedef struct MDB_xcursor { /** A sub-cursor for traversing the Dup DB */ MDB_cursor mx_cursor; - /** A fake transaction struct for pointing to our own table - * of DB info. - */ - MDB_txn mx_txn; - /** Our private DB information tables. Slots 0 and 1 are always - * copies of the corresponding slots in the main transaction. These - * hold the FREEDB and MAINDB, respectively. If the main cursor is - * on a sub-database, that will be copied to slot 2, and the duplicate - * database info will be in slot 3. If the main cursor is on the MAINDB - * then the duplicate DB info will be in slot 2 and slot 3 will be unused. - */ - MDB_dbx mx_dbxs[4]; - /** MDB_db table */ - MDB_db mx_dbs[4]; + /** The database record for this Dup DB */ + MDB_db mx_db; + /** The auxiliary DB record for this Dup DB */ + MDB_dbx mx_dbx; + /** The @ref mt_dbflag for this Dup DB */ + unsigned char mx_dbflag; } MDB_xcursor; /** A set of pages freed by an earlier transaction. */ @@ -738,7 +913,7 @@ typedef struct MDB_oldpages { */ struct MDB_oldpages *mo_next; /** The ID of the transaction in which these pages were freed. */ - ULONG mo_txnid; + txnid_t mo_txnid; /** An #IDL of the pages */ pgno_t mo_pages[1]; /* dynamic */ } MDB_oldpages; @@ -748,12 +923,13 @@ struct MDB_env { HANDLE me_fd; /**< The main data file */ HANDLE me_lfd; /**< The lock file */ HANDLE me_mfd; /**< just for writing the meta pages */ + /** Failed to update the meta page. Probably an I/O error. */ #define MDB_FATAL_ERROR 0x80000000U - uint32_t me_flags; + uint32_t me_flags; /**< @ref mdb_env */ uint32_t me_extrapad; /**< unused for now */ unsigned int me_maxreaders; /**< size of the reader table */ - unsigned int me_numdbs; /**< number of DBs opened */ - unsigned int me_maxdbs; /**< size of the DB table */ + MDB_dbi me_numdbs; /**< number of DBs opened */ + MDB_dbi me_maxdbs; /**< size of the DB table */ 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 */ @@ -762,59 +938,73 @@ struct MDB_env { size_t me_mapsize; /**< size of the data memory map */ off_t me_size; /**< current file size */ pgno_t me_maxpg; /**< me_mapsize / me_psize */ - unsigned int me_psize; /**< size of a page, from #GetPageSize */ + unsigned int me_psize; /**< size of a page, from #GET_PAGESIZE */ unsigned int me_db_toggle; /**< which DB table is current */ + txnid_t me_wtxnid; /**< ID of last txn we committed */ + txnid_t me_pgfirst; /**< ID of first old page record we used */ + txnid_t me_pglast; /**< ID of last old page record we used */ MDB_dbx *me_dbxs; /**< array of static DB info */ MDB_db *me_dbs[2]; /**< two arrays of MDB_db info */ MDB_oldpages *me_pghead; /**< list of old page records */ pthread_key_t me_txkey; /**< thread-key for readers */ MDB_page *me_dpages; /**< list of malloc'd blocks for re-use */ /** IDL of pages that became unused in a write txn */ - pgno_t me_free_pgs[MDB_IDL_UM_SIZE]; + IDL me_free_pgs; /** ID2L of pages that were written during a write txn */ ID2 me_dirty_list[MDB_IDL_UM_SIZE]; /** rwlock for the DB tables, if #LAZY_LOCKS is false */ - LAZY_RWLOCK_DEF(me_dblock); + LAZY_RWLOCK_DEF(me_dblock) #ifdef _WIN32 HANDLE me_rmutex; /* Windows mutexes don't reside in shared mem */ HANDLE me_wmutex; #endif +#ifdef __APPLE__ + sem_t *me_rmutex; /* Apple doesn't support shared mutexes */ + sem_t *me_wmutex; +#endif }; /** max number of pages to commit in one writev() call */ #define MDB_COMMIT_PAGES 64 +#if defined(IOV_MAX) && IOV_MAX < MDB_COMMIT_PAGES +#undef MDB_COMMIT_PAGES +#define MDB_COMMIT_PAGES IOV_MAX +#endif -static MDB_page *mdb_alloc_page(MDB_cursor *mc, int num); -static int mdb_touch(MDB_cursor *mc); +static MDB_page *mdb_page_alloc(MDB_cursor *mc, int num); +static MDB_page *mdb_page_new(MDB_cursor *mc, uint32_t flags, int num); +static int mdb_page_touch(MDB_cursor *mc); -static int mdb_search_page_root(MDB_cursor *mc, +static int mdb_page_get(MDB_txn *txn, pgno_t pgno, MDB_page **mp); +static int mdb_page_search_root(MDB_cursor *mc, MDB_val *key, int modify); -static int mdb_search_page(MDB_cursor *mc, +static int mdb_page_search(MDB_cursor *mc, MDB_val *key, int modify); +static int mdb_page_merge(MDB_cursor *csrc, MDB_cursor *cdst); +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_read_meta(MDB_env *env, int *which); static int mdb_env_write_meta(MDB_txn *txn); -static int mdb_get_page(MDB_txn *txn, pgno_t pgno, MDB_page **mp); -static MDB_node *mdb_search_node(MDB_cursor *mc, MDB_val *key, int *exactp); -static int mdb_add_node(MDB_cursor *mc, indx_t indx, - MDB_val *key, MDB_val *data, pgno_t pgno, uint8_t flags); -static void mdb_del_node(MDB_page *mp, indx_t indx, int ksize); -static int mdb_del0(MDB_cursor *mc, MDB_node *leaf); -static int mdb_read_data(MDB_txn *txn, MDB_node *leaf, MDB_val *data); +static MDB_node *mdb_node_search(MDB_cursor *mc, MDB_val *key, int *exactp); +static int mdb_node_add(MDB_cursor *mc, indx_t indx, + MDB_val *key, MDB_val *data, pgno_t pgno, unsigned int flags); +static void mdb_node_del(MDB_page *mp, indx_t indx, int ksize); +static void mdb_node_shrink(MDB_page *mp, indx_t indx); +static int mdb_node_move(MDB_cursor *csrc, MDB_cursor *cdst); +static int mdb_node_read(MDB_txn *txn, MDB_node *leaf, MDB_val *data); +static size_t mdb_leaf_size(MDB_env *env, MDB_val *key, MDB_val *data); +static size_t mdb_branch_size(MDB_env *env, MDB_val *key); static int mdb_rebalance(MDB_cursor *mc); static int mdb_update_key(MDB_page *mp, indx_t indx, MDB_val *key); -static int mdb_move_node(MDB_cursor *csrcrc, MDB_cursor *cdstst); -static int mdb_merge(MDB_cursor *csrcrc, MDB_cursor *cdstst); -static int mdb_split(MDB_cursor *mc, MDB_val *newkey, MDB_val *newdata, - pgno_t newpgno); -static MDB_page *mdb_new_page(MDB_cursor *mc, uint32_t flags, int num); -static void cursor_pop_page(MDB_cursor *mc); -static int cursor_push_page(MDB_cursor *mc, MDB_page *mp); +static void mdb_cursor_pop(MDB_cursor *mc); +static int mdb_cursor_push(MDB_cursor *mc, MDB_page *mp); -static int mdb_sibling(MDB_cursor *mc, int move_right); +static int mdb_cursor_del0(MDB_cursor *mc, MDB_node *leaf); +static int mdb_cursor_sibling(MDB_cursor *mc, int move_right); static int mdb_cursor_next(MDB_cursor *mc, MDB_val *key, MDB_val *data, MDB_cursor_op op); static int mdb_cursor_prev(MDB_cursor *mc, MDB_val *key, MDB_val *data, MDB_cursor_op op); static int mdb_cursor_set(MDB_cursor *mc, MDB_val *key, MDB_val *data, MDB_cursor_op op, @@ -822,18 +1012,15 @@ static int mdb_cursor_set(MDB_cursor *mc, MDB_val *key, MDB_val *data, MDB_curso static int mdb_cursor_first(MDB_cursor *mc, MDB_val *key, MDB_val *data); static int mdb_cursor_last(MDB_cursor *mc, MDB_val *key, MDB_val *data); +static void mdb_cursor_init(MDB_cursor *mc, MDB_txn *txn, MDB_dbi dbi, MDB_xcursor *mx); static void mdb_xcursor_init0(MDB_cursor *mc); static void mdb_xcursor_init1(MDB_cursor *mc, MDB_node *node); -static void mdb_xcursor_init2(MDB_cursor *mc); -static void mdb_xcursor_fini(MDB_cursor *mc); - -static size_t mdb_leaf_size(MDB_env *env, MDB_val *key, MDB_val *data); -static size_t mdb_branch_size(MDB_env *env, MDB_val *key); +static int mdb_drop0(MDB_cursor *mc, int subs); static void mdb_default_cmp(MDB_txn *txn, MDB_dbi dbi); /** @cond */ -static MDB_cmp_func memncmp, memnrcmp, intcmp, cintcmp; +static MDB_cmp_func mdb_cmp_memn, mdb_cmp_memnr, mdb_cmp_int, mdb_cmp_cint, mdb_cmp_long; /** @endcond */ #ifdef _WIN32 @@ -842,6 +1029,7 @@ static SECURITY_ATTRIBUTES mdb_all_sa; static int mdb_sec_inited; #endif +/** Return the library version info. */ char * mdb_version(int *major, int *minor, int *patch) { @@ -851,7 +1039,7 @@ mdb_version(int *major, int *minor, int *patch) return MDB_VERSION_STRING; } - /** Table of descriptions for MDB @ref error codes */ +/** Table of descriptions for MDB @ref errors */ static char *const mdb_errstr[] = { "MDB_KEYEXIST: Key/data pair already exists", "MDB_NOTFOUND: No matching key/data pair found", @@ -873,8 +1061,13 @@ mdb_strerror(int err) return strerror(err); } -#if DEBUG -static char * +#if MDB_DEBUG +/** Display a key in hexadecimal and return the address of the result. + * @param[in] key the key to display + * @param[in] buf the buffer to write into. Should always be #DKBUF. + * @return The key in hexadecimal form. + */ +char * mdb_dkey(MDB_val *key, char *buf) { char *ptr = buf; @@ -882,7 +1075,11 @@ mdb_dkey(MDB_val *key, char *buf) unsigned int i; if (key->mv_size > MAXKEYSIZE) return "MAXKEYSIZE"; + /* may want to make this a dynamic check: if the key is mostly + * printable characters, print it as-is instead of converting to hex. + */ #if 1 + buf[0] = '\0'; for (i=0; imv_size; i++) ptr += sprintf(ptr, "%02x", *c++); #else @@ -890,6 +1087,74 @@ mdb_dkey(MDB_val *key, char *buf) #endif return buf; } + +/** Display all the keys in the page. */ +static void +mdb_page_keys(MDB_page *mp) +{ + MDB_node *node; + unsigned int i, nkeys; + MDB_val key; + DKBUF; + + nkeys = NUMKEYS(mp); + DPRINTF("numkeys %d", nkeys); + for (i=0; imn_ksize; + key.mv_data = node->mn_data; + DPRINTF("key %d: %s", i, DKEY(&key)); + } +} +#endif + +#if MDB_DEBUG > 2 +/** Count all the pages in each DB and in the freelist + * and make sure it matches the actual number of pages + * being used. + */ +static void mdb_audit(MDB_txn *txn) +{ + MDB_cursor mc; + MDB_val key, data; + int rc, i; + ID freecount, count; + + freecount = 0; + mdb_cursor_init(&mc, txn, FREE_DBI, NULL); + while ((rc = mdb_cursor_get(&mc, &key, &data, MDB_NEXT)) == 0) + freecount += *(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; + + count = 0; + for (i = 0; imt_numdbs; i++) { + count += txn->mt_dbs[i].md_branch_pages + + txn->mt_dbs[i].md_leaf_pages + + txn->mt_dbs[i].md_overflow_pages; + if (txn->mt_dbs[i].md_flags & MDB_DUPSORT) { + MDB_xcursor mx; + mdb_cursor_init(&mc, txn, i, &mx); + mdb_page_search(&mc, NULL, 0); + do { + int j; + MDB_page *mp; + mp = mc.mc_pg[mc.mc_top]; + for (j=0; jmn_flags & F_SUBDATA) { + MDB_db db; + memcpy(&db, NODEDATA(leaf), sizeof(db)); + count += db.md_branch_pages + db.md_leaf_pages + + db.md_overflow_pages; + } + } + } + while (mdb_cursor_sibling(&mc, 1) == 0); + } + } + assert(freecount + count + 2 >= txn->mt_next_pgno - 1); +} #endif int @@ -907,9 +1172,34 @@ mdb_dcmp(MDB_txn *txn, MDB_dbi dbi, const MDB_val *a, const MDB_val *b) return EINVAL; /* too bad you can't distinguish this from a valid result */ } -/* Allocate new page(s) for writing */ +/** Allocate a single page. + * Re-use old malloc'd pages first, otherwise just malloc. + */ +static MDB_page * +mdb_page_malloc(MDB_cursor *mc) { + MDB_page *ret; + size_t sz = mc->mc_txn->mt_env->me_psize; + if ((ret = mc->mc_txn->mt_env->me_dpages) != NULL) { + VGMEMP_ALLOC(mc->mc_txn->mt_env, ret, sz); + VGMEMP_DEFINED(ret, sizeof(ret->mp_next)); + mc->mc_txn->mt_env->me_dpages = ret->mp_next; + } else if ((ret = malloc(sz)) != NULL) { + VGMEMP_ALLOC(mc->mc_txn->mt_env, ret, sz); + } + return ret; +} + +/** Allocate pages for writing. + * If there are free pages available from older transactions, they + * will be re-used first. Otherwise a new page will be allocated. + * @param[in] mc cursor A cursor handle identifying the transaction and + * database for which we are allocating. + * @param[in] num the number of pages to allocate. + * @return Address of the allocated page(s). Requests for multiple pages + * will always be satisfied by a single contiguous chunk of memory. + */ static MDB_page * -mdb_alloc_page(MDB_cursor *mc, int num) +mdb_page_alloc(MDB_cursor *mc, int num) { MDB_txn *txn = mc->mc_txn; MDB_page *np; @@ -918,62 +1208,81 @@ mdb_alloc_page(MDB_cursor *mc, int num) if (txn->mt_txnid > 2) { - if (!txn->mt_env->me_pghead && mc->mc_dbi != FREE_DBI && + if (!txn->mt_env->me_pghead && txn->mt_dbs[FREE_DBI].md_root != P_INVALID) { /* See if there's anything in the free DB */ MDB_cursor m2; MDB_node *leaf; - ULONG *kptr, oldest; - - m2.mc_txn = txn; - m2.mc_dbi = FREE_DBI; - m2.mc_snum = 0; - m2.mc_flags = 0; - mdb_search_page(&m2, NULL, 0); - leaf = NODEPTR(m2.mc_pg[m2.mc_top], 0); - kptr = (ULONG *)NODEKEY(leaf); + MDB_val data; + txnid_t *kptr, oldest, last; + + mdb_cursor_init(&m2, txn, FREE_DBI, NULL); + if (!txn->mt_env->me_pgfirst) { + mdb_page_search(&m2, NULL, 0); + leaf = NODEPTR(m2.mc_pg[m2.mc_top], 0); + kptr = (txnid_t *)NODEKEY(leaf); + last = *kptr; + } else { + MDB_val key; + int rc, exact; +again: + exact = 0; + last = txn->mt_env->me_pglast + 1; + leaf = NULL; + key.mv_data = &last; + key.mv_size = sizeof(last); + rc = mdb_cursor_set(&m2, &key, &data, MDB_SET, &exact); + if (rc) + goto none; + last = *(txnid_t *)key.mv_data; + } { unsigned int i; oldest = txn->mt_txnid - 1; for (i=0; imt_env->me_txns->mti_numreaders; i++) { - ULONG mr = txn->mt_env->me_txns->mti_readers[i].mr_txnid; + txnid_t mr = txn->mt_env->me_txns->mti_readers[i].mr_txnid; if (mr && mr < oldest) oldest = mr; } } - if (oldest > *kptr) { + if (oldest > last) { /* It's usable, grab it. */ MDB_oldpages *mop; - MDB_val data; pgno_t *idl; - mdb_read_data(txn, leaf, &data); - idl = (ULONG *)data.mv_data; + if (!txn->mt_env->me_pgfirst) { + mdb_node_read(txn, leaf, &data); + } + txn->mt_env->me_pglast = last; + if (!txn->mt_env->me_pgfirst) + txn->mt_env->me_pgfirst = last; + idl = (ID *) data.mv_data; + /* We might have a zero-length IDL due to freelist growth + * during a prior commit + */ + if (!idl[0]) goto again; mop = malloc(sizeof(MDB_oldpages) + MDB_IDL_SIZEOF(idl) - sizeof(pgno_t)); mop->mo_next = txn->mt_env->me_pghead; - mop->mo_txnid = *kptr; + mop->mo_txnid = last; txn->mt_env->me_pghead = mop; memcpy(mop->mo_pages, idl, MDB_IDL_SIZEOF(idl)); -#if DEBUG > 1 +#if MDB_DEBUG > 1 { unsigned int i; - DPRINTF("IDL read txn %lu root %lu num %lu", + DPRINTF("IDL read txn %zu root %zu num %zu", mop->mo_txnid, txn->mt_dbs[FREE_DBI].md_root, idl[0]); for (i=0; imt_env->me_pghead) { MDB_oldpages *mop = txn->mt_env->me_pghead; if (num > 1) { @@ -1002,15 +1311,21 @@ mdb_alloc_page(MDB_cursor *mc, int num) if (pgno == P_INVALID) { /* DB size is maxed out */ - if (txn->mt_next_pgno + num >= txn->mt_env->me_maxpg) + if (txn->mt_next_pgno + num >= txn->mt_env->me_maxpg) { + DPUTS("DB size maxed out"); return NULL; + } } if (txn->mt_env->me_dpages && num == 1) { np = txn->mt_env->me_dpages; + VGMEMP_ALLOC(txn->mt_env, np, txn->mt_env->me_psize); + VGMEMP_DEFINED(np, sizeof(np->mp_next)); txn->mt_env->me_dpages = np->mp_next; } else { - if ((np = malloc(txn->mt_env->me_psize * num )) == NULL) + size_t sz = txn->mt_env->me_psize * num; + if ((np = malloc(sz)) == NULL) return NULL; + VGMEMP_ALLOC(txn->mt_env, np, sz); } if (pgno == P_INVALID) { np->mp_pgno = txn->mt_next_pgno; @@ -1025,31 +1340,86 @@ mdb_alloc_page(MDB_cursor *mc, int num) return np; } -/* Touch a page: make it dirty and re-insert into tree with updated pgno. +/** Touch a page: make it dirty and re-insert into tree with updated pgno. + * @param[in] mc cursor pointing to the page to be touched + * @return 0 on success, non-zero on failure. */ static int -mdb_touch(MDB_cursor *mc) +mdb_page_touch(MDB_cursor *mc) { MDB_page *mp = mc->mc_pg[mc->mc_top]; pgno_t pgno; if (!F_ISSET(mp->mp_flags, P_DIRTY)) { MDB_page *np; - if ((np = mdb_alloc_page(mc, 1)) == NULL) + if ((np = mdb_page_alloc(mc, 1)) == NULL) return ENOMEM; - DPRINTF("touched db %u page %lu -> %lu", mc->mc_dbi, mp->mp_pgno, np->mp_pgno); + DPRINTF("touched db %u page %zu -> %zu", mc->mc_dbi, mp->mp_pgno, np->mp_pgno); assert(mp->mp_pgno != np->mp_pgno); - mdb_midl_append(mc->mc_txn->mt_free_pgs, mp->mp_pgno); + mdb_midl_append(&mc->mc_txn->mt_free_pgs, mp->mp_pgno); pgno = np->mp_pgno; memcpy(np, mp, mc->mc_txn->mt_env->me_psize); mp = np; mp->mp_pgno = pgno; mp->mp_flags |= P_DIRTY; +finish: + /* Adjust other cursors pointing to mp */ + if (mc->mc_flags & C_SUB) { + MDB_cursor *m2, *m3; + MDB_dbi dbi = mc->mc_dbi-1; + + for (m2 = mc->mc_txn->mt_cursors[dbi]; m2; m2=m2->mc_next) { + if (m2 == mc) continue; + m3 = &m2->mc_xcursor->mx_cursor; + if (m3->mc_snum < mc->mc_snum) continue; + if (m3->mc_pg[mc->mc_top] == mc->mc_pg[mc->mc_top]) { + m3->mc_pg[mc->mc_top] = mp; + } + } + } else { + MDB_cursor *m2; + + for (m2 = mc->mc_txn->mt_cursors[mc->mc_dbi]; m2; m2=m2->mc_next) { + if (m2 == mc || m2->mc_snum < mc->mc_snum) continue; + if (m2->mc_pg[mc->mc_top] == mc->mc_pg[mc->mc_top]) { + m2->mc_pg[mc->mc_top] = mp; + } + } + } mc->mc_pg[mc->mc_top] = mp; - /* Update the page number to new touched page. */ + /** If this page has a parent, update the parent to point to + * this new page. + */ if (mc->mc_top) SETPGNO(NODEPTR(mc->mc_pg[mc->mc_top-1], mc->mc_ki[mc->mc_top-1]), mp->mp_pgno); + else + mc->mc_db->md_root = mp->mp_pgno; + } else if (mc->mc_txn->mt_parent) { + MDB_page *np; + ID2 mid; + /* If txn has a parent, make sure the page is in our + * dirty list. + */ + if (mc->mc_txn->mt_u.dirty_list[0].mid) { + unsigned x = mdb_mid2l_search(mc->mc_txn->mt_u.dirty_list, mp->mp_pgno); + if (x <= mc->mc_txn->mt_u.dirty_list[0].mid && + mc->mc_txn->mt_u.dirty_list[x].mid == mp->mp_pgno) { + if (mc->mc_txn->mt_u.dirty_list[x].mptr != mp) { + mp = mc->mc_txn->mt_u.dirty_list[x].mptr; + mc->mc_pg[mc->mc_top] = mp; + } + return 0; + } + } + /* No - copy it */ + np = mdb_page_malloc(mc); + memcpy(np, mp, mc->mc_txn->mt_env->me_psize); + mid.mid = np->mp_pgno; + mid.mptr = np; + mdb_mid2l_insert(mc->mc_txn->mt_u.dirty_list, &mid); + mp = np; + goto finish; } return 0; } @@ -1059,19 +1429,113 @@ mdb_env_sync(MDB_env *env, int force) { int rc = 0; if (force || !F_ISSET(env->me_flags, MDB_NOSYNC)) { - if (fdatasync(env->me_fd)) + if (MDB_FDATASYNC(env->me_fd)) rc = ErrCode(); } return rc; } -static inline void +/** Make shadow copies of all of parent txn's cursors */ +static int +mdb_cursor_shadow(MDB_txn *src, MDB_txn *dst) +{ + MDB_cursor *mc, *m2; + unsigned int i, j, size; + + for (i=0;imt_numdbs; i++) { + if (src->mt_cursors[i]) { + size = sizeof(MDB_cursor); + if (src->mt_cursors[i]->mc_xcursor) + size += sizeof(MDB_xcursor); + for (m2 = src->mt_cursors[i]; m2; m2=m2->mc_next) { + mc = malloc(size); + if (!mc) + return ENOMEM; + mc->mc_orig = m2; + mc->mc_txn = dst; + mc->mc_dbi = i; + mc->mc_db = &dst->mt_dbs[i]; + mc->mc_dbx = m2->mc_dbx; + mc->mc_dbflag = &dst->mt_dbflags[i]; + mc->mc_snum = m2->mc_snum; + mc->mc_top = m2->mc_top; + mc->mc_flags = m2->mc_flags | C_SHADOW; + for (j=0; jmc_snum; j++) { + mc->mc_pg[j] = m2->mc_pg[j]; + mc->mc_ki[j] = m2->mc_ki[j]; + } + if (m2->mc_xcursor) { + MDB_xcursor *mx, *mx2; + mx = (MDB_xcursor *)(mc+1); + mc->mc_xcursor = mx; + mx2 = m2->mc_xcursor; + mx->mx_db = mx2->mx_db; + mx->mx_dbx = mx2->mx_dbx; + mx->mx_dbflag = mx2->mx_dbflag; + mx->mx_cursor.mc_txn = dst; + mx->mx_cursor.mc_dbi = mx2->mx_cursor.mc_dbi; + mx->mx_cursor.mc_db = &mx->mx_db; + mx->mx_cursor.mc_dbx = &mx->mx_dbx; + mx->mx_cursor.mc_dbflag = &mx->mx_dbflag; + mx->mx_cursor.mc_snum = mx2->mx_cursor.mc_snum; + mx->mx_cursor.mc_top = mx2->mx_cursor.mc_top; + mx->mx_cursor.mc_flags = mx2->mx_cursor.mc_flags | C_SHADOW; + for (j=0; jmx_cursor.mc_snum; j++) { + mx->mx_cursor.mc_pg[j] = mx2->mx_cursor.mc_pg[j]; + mx->mx_cursor.mc_ki[j] = mx2->mx_cursor.mc_ki[j]; + } + } else { + mc->mc_xcursor = NULL; + } + mc->mc_next = dst->mt_cursors[i]; + dst->mt_cursors[i] = mc; + } + } + } + return MDB_SUCCESS; +} + +/** Merge shadow cursors back into parent's */ +static void +mdb_cursor_merge(MDB_txn *txn) +{ + MDB_dbi i; + for (i=0; imt_numdbs; i++) { + if (txn->mt_cursors[i]) { + MDB_cursor *mc; + while ((mc = txn->mt_cursors[i])) { + txn->mt_cursors[i] = mc->mc_next; + if (mc->mc_flags & C_SHADOW) { + MDB_cursor *m2 = mc->mc_orig; + unsigned int j; + m2->mc_snum = mc->mc_snum; + m2->mc_top = mc->mc_top; + for (j=0; jmc_snum; j++) { + m2->mc_pg[j] = mc->mc_pg[j]; + m2->mc_ki[j] = mc->mc_ki[j]; + } + } + if (mc->mc_flags & C_ALLOCD) + free(mc); + } + } + } +} + +static void mdb_txn_reset0(MDB_txn *txn); -static inline int +/** Common code for #mdb_txn_begin() and #mdb_txn_renew(). + * @param[in] txn the transaction handle to initialize + * @return 0 on success, non-zero on failure. This can only + * fail for read-only transactions, and then only if the + * reader table is full. + */ +static int mdb_txn_renew0(MDB_txn *txn) { MDB_env *env = txn->mt_env; + char mt_dbflag = 0; if (txn->mt_flags & MDB_TXN_RDONLY) { MDB_reader *r = pthread_getspecific(env->me_txkey); @@ -1096,14 +1560,22 @@ mdb_txn_renew0(MDB_txn *txn) r = &env->me_txns->mti_readers[i]; pthread_setspecific(env->me_txkey, r); } - txn->mt_txnid = env->me_txns->mti_txnid; txn->mt_toggle = env->me_txns->mti_me_toggle; + txn->mt_txnid = env->me_txns->mti_txnid; + /* This happens if a different process was the + * last writer to the DB. + */ + if (env->me_wtxnid < txn->mt_txnid) + mt_dbflag = DB_STALE; r->mr_txnid = txn->mt_txnid; txn->mt_u.reader = r; } else { LOCK_MUTEX_W(env); - txn->mt_txnid = env->me_txns->mti_txnid+1; + txn->mt_txnid = env->me_txns->mti_txnid; + if (env->me_wtxnid < txn->mt_txnid) + mt_dbflag = DB_STALE; + txn->mt_txnid++; txn->mt_toggle = env->me_txns->mti_me_toggle; txn->mt_u.dirty_list = env->me_dirty_list; txn->mt_u.dirty_list[0].mid = 0; @@ -1123,6 +1595,8 @@ mdb_txn_renew0(MDB_txn *txn) (txn->mt_numdbs - 2) * sizeof(MDB_db)); LAZY_RWLOCK_UNLOCK(&env->me_dblock); + memset(txn->mt_dbflags, mt_dbflag, env->me_numdbs); + return MDB_SUCCESS; } @@ -1141,47 +1615,91 @@ mdb_txn_renew(MDB_txn *txn) rc = mdb_txn_renew0(txn); if (rc == MDB_SUCCESS) { - DPRINTF("renew txn %lu%c %p on mdbenv %p, root page %lu", - txn->mt_txnid, (txn->mt_flags & MDB_TXN_RDONLY) ? 'r' : 'w', txn, - (void *)txn->mt_env, txn->mt_dbs[MAIN_DBI].md_root); + DPRINTF("renew txn %zu%c %p on mdbenv %p, root page %zu", + txn->mt_txnid, (txn->mt_flags & MDB_TXN_RDONLY) ? 'r' : 'w', + (void *)txn, (void *)txn->mt_env, txn->mt_dbs[MAIN_DBI].md_root); } return rc; } int -mdb_txn_begin(MDB_env *env, unsigned int flags, MDB_txn **ret) +mdb_txn_begin(MDB_env *env, MDB_txn *parent, unsigned int flags, MDB_txn **ret) { MDB_txn *txn; - int rc; + int rc, size; if (env->me_flags & MDB_FATAL_ERROR) { DPUTS("environment had fatal error, must shutdown!"); return MDB_PANIC; } - if ((txn = calloc(1, sizeof(MDB_txn) + env->me_maxdbs * sizeof(MDB_db))) == NULL) { + if (parent) { + /* parent already has an active child txn */ + if (parent->mt_child) { + return EINVAL; + } + } + size = sizeof(MDB_txn) + env->me_maxdbs * (sizeof(MDB_db)+1); + if (!(flags & MDB_RDONLY)) + size += env->me_maxdbs * sizeof(MDB_cursor *); + + if ((txn = calloc(1, size)) == NULL) { DPRINTF("calloc: %s", strerror(ErrCode())); return ENOMEM; } txn->mt_dbs = (MDB_db *)(txn+1); if (flags & MDB_RDONLY) { txn->mt_flags |= MDB_TXN_RDONLY; + txn->mt_dbflags = (unsigned char *)(txn->mt_dbs + env->me_maxdbs); + } else { + txn->mt_cursors = (MDB_cursor **)(txn->mt_dbs + env->me_maxdbs); + txn->mt_dbflags = (unsigned char *)(txn->mt_cursors + env->me_maxdbs); } txn->mt_env = env; - rc = mdb_txn_renew0(txn); + if (parent) { + txn->mt_free_pgs = mdb_midl_alloc(); + if (!txn->mt_free_pgs) { + free(txn); + return ENOMEM; + } + txn->mt_u.dirty_list = malloc(sizeof(ID2)*MDB_IDL_UM_SIZE); + if (!txn->mt_u.dirty_list) { + free(txn->mt_free_pgs); + free(txn); + return ENOMEM; + } + txn->mt_txnid = parent->mt_txnid; + txn->mt_toggle = parent->mt_toggle; + txn->mt_u.dirty_list[0].mid = 0; + txn->mt_free_pgs[0] = 0; + txn->mt_next_pgno = parent->mt_next_pgno; + parent->mt_child = txn; + txn->mt_parent = parent; + txn->mt_numdbs = parent->mt_numdbs; + txn->mt_dbxs = parent->mt_dbxs; + memcpy(txn->mt_dbs, parent->mt_dbs, txn->mt_numdbs * sizeof(MDB_db)); + memcpy(txn->mt_dbflags, parent->mt_dbflags, txn->mt_numdbs); + mdb_cursor_shadow(parent, txn); + rc = 0; + } else { + rc = mdb_txn_renew0(txn); + } if (rc) free(txn); else { *ret = txn; - DPRINTF("begin txn %lu%c %p on mdbenv %p, root page %lu", - txn->mt_txnid, (txn->mt_flags & MDB_TXN_RDONLY) ? 'r' : 'w', txn, - (void *) env, txn->mt_dbs[MAIN_DBI].md_root); + DPRINTF("begin txn %zu%c %p on mdbenv %p, root page %zu", + txn->mt_txnid, (txn->mt_flags & MDB_TXN_RDONLY) ? 'r' : 'w', + (void *) txn, (void *) env, txn->mt_dbs[MAIN_DBI].md_root); } return rc; } -static inline void +/** Common code for #mdb_txn_reset() and #mdb_txn_abort(). + * @param[in] txn the transaction handle to reset + */ +static void mdb_txn_reset0(MDB_txn *txn) { MDB_env *env = txn->mt_env; @@ -1193,26 +1711,51 @@ mdb_txn_reset0(MDB_txn *txn) MDB_page *dp; unsigned int i; + /* close(free) all cursors */ + for (i=0; imt_numdbs; i++) { + if (txn->mt_cursors[i]) { + MDB_cursor *mc; + while ((mc = txn->mt_cursors[i])) { + txn->mt_cursors[i] = mc->mc_next; + if (mc->mc_flags & C_ALLOCD) + free(mc); + } + } + } + /* return all dirty pages to dpage list */ for (i=1; i<=txn->mt_u.dirty_list[0].mid; i++) { dp = txn->mt_u.dirty_list[i].mptr; if (!IS_OVERFLOW(dp) || dp->mp_pages == 1) { dp->mp_next = txn->mt_env->me_dpages; + VGMEMP_FREE(txn->mt_env, dp); txn->mt_env->me_dpages = dp; } else { /* large pages just get freed directly */ + VGMEMP_FREE(txn->mt_env, dp); free(dp); } } + if (txn->mt_parent) { + txn->mt_parent->mt_child = NULL; + free(txn->mt_free_pgs); + free(txn->mt_u.dirty_list); + return; + } else { + if (mdb_midl_shrink(&txn->mt_free_pgs)) + env->me_free_pgs = txn->mt_free_pgs; + } + while ((mop = txn->mt_env->me_pghead)) { txn->mt_env->me_pghead = mop->mo_next; free(mop); } + txn->mt_env->me_pgfirst = 0; + txn->mt_env->me_pglast = 0; env->me_txn = NULL; - for (i=2; ime_numdbs; i++) - env->me_dbxs[i].md_dirty = 0; + /* The writer mutex was locked in mdb_txn_begin. */ UNLOCK_MUTEX_W(env); } } @@ -1223,9 +1766,9 @@ mdb_txn_reset(MDB_txn *txn) if (txn == NULL) return; - DPRINTF("reset txn %lu%c %p on mdbenv %p, root page %lu", - txn->mt_txnid, (txn->mt_flags & MDB_TXN_RDONLY) ? 'r' : 'w', txn, - (void *)txn->mt_env, txn->mt_dbs[MAIN_DBI].md_root); + DPRINTF("reset txn %zu%c %p on mdbenv %p, root page %zu", + txn->mt_txnid, (txn->mt_flags & MDB_TXN_RDONLY) ? 'r' : 'w', + (void *) txn, (void *)txn->mt_env, txn->mt_dbs[MAIN_DBI].md_root); mdb_txn_reset0(txn); } @@ -1236,9 +1779,12 @@ mdb_txn_abort(MDB_txn *txn) if (txn == NULL) return; - DPRINTF("abort txn %lu%c %p on mdbenv %p, root page %lu", - txn->mt_txnid, (txn->mt_flags & MDB_TXN_RDONLY) ? 'r' : 'w', txn, - (void *)txn->mt_env, txn->mt_dbs[MAIN_DBI].md_root); + DPRINTF("abort txn %zu%c %p on mdbenv %p, root page %zu", + txn->mt_txnid, (txn->mt_flags & MDB_TXN_RDONLY) ? 'r' : 'w', + (void *)txn, (void *)txn->mt_env, txn->mt_dbs[MAIN_DBI].md_root); + + if (txn->mt_child) + mdb_txn_abort(txn->mt_child); mdb_txn_reset0(txn); free(txn); @@ -1253,27 +1799,101 @@ mdb_txn_commit(MDB_txn *txn) off_t size; MDB_page *dp; MDB_env *env; - pgno_t next; + pgno_t next, freecnt; MDB_cursor mc; assert(txn != NULL); assert(txn->mt_env != NULL); + if (txn->mt_child) { + mdb_txn_commit(txn->mt_child); + txn->mt_child = NULL; + } + env = txn->mt_env; if (F_ISSET(txn->mt_flags, MDB_TXN_RDONLY)) { + if (txn->mt_numdbs > env->me_numdbs) { + /* update the DB tables */ + int toggle = !env->me_db_toggle; + MDB_db *ip, *jp; + MDB_dbi i; + + ip = &env->me_dbs[toggle][env->me_numdbs]; + jp = &txn->mt_dbs[env->me_numdbs]; + LAZY_RWLOCK_WRLOCK(&env->me_dblock); + for (i = env->me_numdbs; i < txn->mt_numdbs; i++) { + *ip++ = *jp++; + } + + env->me_db_toggle = toggle; + env->me_numdbs = txn->mt_numdbs; + LAZY_RWLOCK_UNLOCK(&env->me_dblock); + } mdb_txn_abort(txn); return MDB_SUCCESS; } - if (txn != env->me_txn) { - DPUTS("attempt to commit unknown transaction"); + if (F_ISSET(txn->mt_flags, MDB_TXN_ERROR)) { + DPUTS("error flag is set, can't commit"); + if (txn->mt_parent) + txn->mt_parent->mt_flags |= MDB_TXN_ERROR; mdb_txn_abort(txn); return EINVAL; } - if (F_ISSET(txn->mt_flags, MDB_TXN_ERROR)) { - DPUTS("error flag is set, can't commit"); + /* Merge (and close) our cursors with parent's */ + mdb_cursor_merge(txn); + + if (txn->mt_parent) { + MDB_db *ip, *jp; + MDB_dbi i; + unsigned x, y; + ID2L dst, src; + + /* Update parent's DB table */ + ip = &txn->mt_parent->mt_dbs[2]; + jp = &txn->mt_dbs[2]; + for (i = 2; i < txn->mt_numdbs; i++) { + if (ip->md_root != jp->md_root) + *ip = *jp; + ip++; jp++; + } + txn->mt_parent->mt_numdbs = txn->mt_numdbs; + + /* Append our free list to parent's */ + mdb_midl_append_list(&txn->mt_parent->mt_free_pgs, + txn->mt_free_pgs); + mdb_midl_free(txn->mt_free_pgs); + + /* Merge our dirty list with parent's */ + dst = txn->mt_parent->mt_u.dirty_list; + src = txn->mt_u.dirty_list; + x = mdb_mid2l_search(dst, src[1].mid); + for (y=1; y<=src[0].mid; y++) { + while (x <= dst[0].mid && dst[x].mid != src[y].mid) x++; + if (x > dst[0].mid) + break; + free(dst[x].mptr); + dst[x].mptr = src[y].mptr; + } + x = dst[0].mid; + for (; y<=src[0].mid; y++) { + if (++x >= MDB_IDL_UM_MAX) { + mdb_txn_abort(txn); + return ENOMEM; + } + dst[x] = src[y]; + } + dst[0].mid = x; + free(txn->mt_u.dirty_list); + txn->mt_parent->mt_child = NULL; + free(txn); + return MDB_SUCCESS; + } + + if (txn != env->me_txn) { + DPUTS("attempt to commit unknown transaction"); mdb_txn_abort(txn); return EINVAL; } @@ -1281,89 +1901,141 @@ mdb_txn_commit(MDB_txn *txn) if (!txn->mt_u.dirty_list[0].mid) goto done; - DPRINTF("committing txn %lu %p on mdbenv %p, root page %lu", - txn->mt_txnid, txn, (void *)env, txn->mt_dbs[MAIN_DBI].md_root); + DPRINTF("committing txn %zu %p on mdbenv %p, root page %zu", + txn->mt_txnid, (void *)txn, (void *)env, txn->mt_dbs[MAIN_DBI].md_root); - mc.mc_txn = txn; - mc.mc_dbi = FREE_DBI; - mc.mc_flags = 0; + mdb_cursor_init(&mc, txn, FREE_DBI, NULL); /* should only be one record now */ if (env->me_pghead) { /* make sure first page of freeDB is touched and on freelist */ - mdb_search_page(&mc, NULL, 1); + mdb_page_search(&mc, NULL, 1); } + + /* Delete IDLs we used from the free list */ + if (env->me_pgfirst) { + txnid_t cur; + MDB_val key; + int exact = 0; + + key.mv_size = sizeof(cur); + for (cur = env->me_pgfirst; cur <= env->me_pglast; cur++) { + key.mv_data = &cur; + + mdb_cursor_set(&mc, &key, NULL, MDB_SET, &exact); + mdb_cursor_del(&mc, 0); + } + env->me_pgfirst = 0; + env->me_pglast = 0; + } + /* save to free list */ +free2: + freecnt = txn->mt_free_pgs[0]; if (!MDB_IDL_IS_ZERO(txn->mt_free_pgs)) { MDB_val key, data; - ULONG i; /* make sure last page of freeDB is touched and on freelist */ key.mv_size = MAXKEYSIZE+1; key.mv_data = NULL; - mdb_search_page(&mc, &key, 1); + mdb_page_search(&mc, &key, 1); mdb_midl_sort(txn->mt_free_pgs); -#if DEBUG > 1 +#if MDB_DEBUG > 1 { unsigned int i; - ULONG *idl = txn->mt_free_pgs; - DPRINTF("IDL write txn %lu root %lu num %lu", + ID *idl = txn->mt_free_pgs; + DPRINTF("IDL write txn %zu root %zu num %zu", txn->mt_txnid, txn->mt_dbs[FREE_DBI].md_root, idl[0]); for (i=0; imt_txnid; + key.mv_data = &txn->mt_txnid; data.mv_data = txn->mt_free_pgs; /* The free list can still grow during this call, * despite the pre-emptive touches above. So check * and make sure the entire thing got written. */ do { - i = txn->mt_free_pgs[0]; + freecnt = txn->mt_free_pgs[0]; data.mv_size = MDB_IDL_SIZEOF(txn->mt_free_pgs); rc = mdb_cursor_put(&mc, &key, &data, 0); if (rc) { mdb_txn_abort(txn); return rc; } - } while (i != txn->mt_free_pgs[0]); + } while (freecnt != txn->mt_free_pgs[0]); } /* should only be one record now */ +again: if (env->me_pghead) { MDB_val key, data; MDB_oldpages *mop; + pgno_t orig; + txnid_t id; mop = env->me_pghead; - key.mv_size = sizeof(pgno_t); - key.mv_data = (char *)&mop->mo_txnid; + id = mop->mo_txnid; + key.mv_size = sizeof(id); + key.mv_data = &id; data.mv_size = MDB_IDL_SIZEOF(mop->mo_pages); data.mv_data = mop->mo_pages; + orig = mop->mo_pages[0]; + /* These steps may grow the freelist again + * due to freed overflow pages... + */ mdb_cursor_put(&mc, &key, &data, 0); - free(env->me_pghead); - env->me_pghead = NULL; + if (mop == env->me_pghead && env->me_pghead->mo_txnid == id) { + /* could have been used again here */ + if (mop->mo_pages[0] != orig) { + data.mv_size = MDB_IDL_SIZEOF(mop->mo_pages); + data.mv_data = mop->mo_pages; + id = mop->mo_txnid; + mdb_cursor_put(&mc, &key, &data, 0); + } + env->me_pghead = NULL; + free(mop); + } else { + /* was completely used up */ + mdb_cursor_del(&mc, 0); + if (env->me_pghead) + goto again; + } + env->me_pgfirst = 0; + env->me_pglast = 0; + } + /* Check for growth of freelist again */ + if (freecnt != txn->mt_free_pgs[0]) + goto free2; + + if (!MDB_IDL_IS_ZERO(txn->mt_free_pgs)) { + if (mdb_midl_shrink(&txn->mt_free_pgs)) + env->me_free_pgs = txn->mt_free_pgs; } /* Update DB root pointers. Their pages have already been * touched so this is all in-place and cannot fail. */ { + MDB_dbi i; MDB_val data; data.mv_size = sizeof(MDB_db); - mc.mc_dbi = MAIN_DBI; - mc.mc_flags = 0; + mdb_cursor_init(&mc, txn, MAIN_DBI, NULL); for (i = 2; i < txn->mt_numdbs; i++) { - if (txn->mt_dbxs[i].md_dirty) { + if (txn->mt_dbflags[i] & DB_DIRTY) { data.mv_data = &txn->mt_dbs[i]; mdb_cursor_put(&mc, &txn->mt_dbxs[i].md_name, &data, 0); } } } +#if MDB_DEBUG > 2 + mdb_audit(txn); +#endif /* Commit up to MDB_COMMIT_PAGES dirty pages to disk until done. */ @@ -1383,7 +2055,7 @@ mdb_txn_commit(MDB_txn *txn) for (; i<=txn->mt_u.dirty_list[0].mid; i++) { size_t wsize; dp = txn->mt_u.dirty_list[i].mptr; - DPRINTF("committing page %lu", dp->mp_pgno); + DPRINTF("committing page %zu", dp->mp_pgno); size = dp->mp_pgno * env->me_psize; ov.Offset = size & 0xffffffff; ov.OffsetHigh = size >> 16; @@ -1400,7 +2072,7 @@ mdb_txn_commit(MDB_txn *txn) return n; } } - done = 1;; + done = 1; #else struct iovec iov[MDB_COMMIT_PAGES]; n = 0; @@ -1410,7 +2082,6 @@ mdb_txn_commit(MDB_txn *txn) dp = txn->mt_u.dirty_list[i].mptr; if (dp->mp_pgno != next) { if (n) { - DPRINTF("committing %u dirty pages", n); rc = writev(env->me_fd, iov, n); if (rc != size) { n = ErrCode(); @@ -1427,10 +2098,10 @@ mdb_txn_commit(MDB_txn *txn) lseek(env->me_fd, dp->mp_pgno * env->me_psize, SEEK_SET); next = dp->mp_pgno; } - DPRINTF("committing page %lu", dp->mp_pgno); + DPRINTF("committing page %zu", dp->mp_pgno); iov[n].iov_len = env->me_psize; if (IS_OVERFLOW(dp)) iov[n].iov_len *= dp->mp_pages; - iov[n].iov_base = dp; + iov[n].iov_base = (char *)dp; size += iov[n].iov_len; next = dp->mp_pgno + (IS_OVERFLOW(dp) ? dp->mp_pages : 1); /* clear dirty flag */ @@ -1445,7 +2116,6 @@ mdb_txn_commit(MDB_txn *txn) if (n == 0) break; - DPRINTF("committing %u dirty pages", n); rc = writev(env->me_fd, iov, n); if (rc != size) { n = ErrCode(); @@ -1465,8 +2135,10 @@ mdb_txn_commit(MDB_txn *txn) dp = txn->mt_u.dirty_list[i].mptr; if (!IS_OVERFLOW(dp) || dp->mp_pages == 1) { dp->mp_next = txn->mt_env->me_dpages; + VGMEMP_FREE(txn->mt_env, dp); txn->mt_env->me_dpages = dp; } else { + VGMEMP_FREE(txn->mt_env, dp); free(dp); } txn->mt_u.dirty_list[i].mid = 0; @@ -1478,6 +2150,7 @@ mdb_txn_commit(MDB_txn *txn) mdb_txn_abort(txn); return n; } + env->me_wtxnid = txn->mt_txnid; done: env->me_txn = NULL; @@ -1485,6 +2158,7 @@ done: { int toggle = !env->me_db_toggle; MDB_db *ip, *jp; + MDB_dbi i; ip = &env->me_dbs[toggle][2]; jp = &txn->mt_dbs[2]; @@ -1495,10 +2169,6 @@ done: ip++; jp++; } - for (i = 2; i < txn->mt_numdbs; i++) { - if (txn->mt_dbxs[i].md_dirty) - txn->mt_dbxs[i].md_dirty = 0; - } env->me_db_toggle = toggle; env->me_numdbs = txn->mt_numdbs; LAZY_RWLOCK_UNLOCK(&env->me_dblock); @@ -1510,10 +2180,16 @@ done: return MDB_SUCCESS; } +/** Read the environment parameters of a DB environment before + * mapping it into memory. + * @param[in] env the environment handle + * @param[out] meta address of where to store the meta information + * @return 0 on success, non-zero on failure. + */ static int mdb_env_read_header(MDB_env *env, MDB_meta *meta) { - char page[PAGESIZE]; + MDB_pagebuf pbuf; MDB_page *p; MDB_meta *m; int rc, err; @@ -1522,14 +2198,14 @@ mdb_env_read_header(MDB_env *env, MDB_meta *meta) */ #ifdef _WIN32 - if (!ReadFile(env->me_fd, page, PAGESIZE, (DWORD *)&rc, NULL) || rc == 0) + if (!ReadFile(env->me_fd, &pbuf, MDB_PAGESIZE, (DWORD *)&rc, NULL) || rc == 0) #else - if ((rc = read(env->me_fd, page, PAGESIZE)) == 0) + if ((rc = read(env->me_fd, &pbuf, MDB_PAGESIZE)) == 0) #endif { return ENOENT; } - else if (rc != PAGESIZE) { + else if (rc != MDB_PAGESIZE) { err = ErrCode(); if (rc > 0) err = EINVAL; @@ -1537,10 +2213,10 @@ mdb_env_read_header(MDB_env *env, MDB_meta *meta) return err; } - p = (MDB_page *)page; + p = (MDB_page *)&pbuf; if (!F_ISSET(p->mp_flags, P_META)) { - DPRINTF("page %lu not a meta page", p->mp_pgno); + DPRINTF("page %zu not a meta page", p->mp_pgno); return EINVAL; } @@ -1560,6 +2236,11 @@ mdb_env_read_header(MDB_env *env, MDB_meta *meta) return 0; } +/** Write the environment parameters of a freshly created DB environment. + * @param[in] env the environment handle + * @param[out] meta address of where to store the meta information + * @return 0 on success, non-zero on failure. + */ static int mdb_env_init_meta(MDB_env *env, MDB_meta *meta) { @@ -1570,7 +2251,7 @@ mdb_env_init_meta(MDB_env *env, MDB_meta *meta) DPUTS("writing new meta page"); - GetPageSize(psize); + GET_PAGESIZE(psize); meta->mm_magic = MDB_MAGIC; meta->mm_version = MDB_VERSION; @@ -1610,6 +2291,10 @@ mdb_env_init_meta(MDB_env *env, MDB_meta *meta) return rc; } +/** Update the environment info to commit a transaction. + * @param[in] txn the transaction that's being committed + * @return 0 on success, non-zero on failure. + */ static int mdb_env_write_meta(MDB_txn *txn) { @@ -1626,7 +2311,7 @@ mdb_env_write_meta(MDB_txn *txn) assert(txn->mt_env != NULL); toggle = !txn->mt_toggle; - DPRINTF("writing meta page %d for root page %lu", + DPRINTF("writing meta page %d for root page %zu", toggle, txn->mt_dbs[MAIN_DBI].md_root); env = txn->mt_env; @@ -1690,6 +2375,11 @@ mdb_env_write_meta(MDB_txn *txn) return MDB_SUCCESS; } +/** Check both meta pages to see which one is newer. + * @param[in] env the environment handle + * @param[out] which address of where to store the meta toggle ID + * @return 0 on success, non-zero on failure. + */ static int mdb_env_read_meta(MDB_env *env, int *which) { @@ -1712,13 +2402,20 @@ mdb_env_create(MDB_env **env) MDB_env *e; e = calloc(1, sizeof(MDB_env)); - if (!e) return ENOMEM; + if (!e) + return ENOMEM; + e->me_free_pgs = mdb_midl_alloc(); + if (!e->me_free_pgs) { + free(e); + return ENOMEM; + } e->me_maxreaders = DEFAULT_READERS; e->me_maxdbs = 2; e->me_fd = INVALID_HANDLE_VALUE; e->me_lfd = INVALID_HANDLE_VALUE; e->me_mfd = INVALID_HANDLE_VALUE; + VGMEMP_CREATE(e,0,0); *env = e; return MDB_SUCCESS; } @@ -1729,11 +2426,13 @@ mdb_env_set_mapsize(MDB_env *env, size_t size) if (env->me_map) return EINVAL; env->me_mapsize = size; + if (env->me_psize) + env->me_maxpg = env->me_mapsize / env->me_psize; return MDB_SUCCESS; } int -mdb_env_set_maxdbs(MDB_env *env, int dbs) +mdb_env_set_maxdbs(MDB_env *env, MDB_dbi dbs) { if (env->me_map) return EINVAL; @@ -1742,16 +2441,16 @@ mdb_env_set_maxdbs(MDB_env *env, int dbs) } int -mdb_env_set_maxreaders(MDB_env *env, int readers) +mdb_env_set_maxreaders(MDB_env *env, unsigned int readers) { - if (env->me_map) + if (env->me_map || readers < 1) return EINVAL; env->me_maxreaders = readers; return MDB_SUCCESS; } int -mdb_env_get_maxreaders(MDB_env *env, int *readers) +mdb_env_get_maxreaders(MDB_env *env, unsigned int *readers) { if (!env || !readers) return EINVAL; @@ -1759,6 +2458,8 @@ mdb_env_get_maxreaders(MDB_env *env, int *readers) return MDB_SUCCESS; } +/** Further setup required for opening an MDB environment + */ static int mdb_env_open2(MDB_env *env, unsigned int flags) { @@ -1813,8 +2514,10 @@ mdb_env_open2(MDB_env *env, unsigned int flags) i |= MAP_FIXED; env->me_map = mmap(meta.mm_address, env->me_mapsize, PROT_READ, i, env->me_fd, 0); - if (env->me_map == MAP_FAILED) + if (env->me_map == MAP_FAILED) { + env->me_map = NULL; return ErrCode(); + } #endif if (newenv) { @@ -1841,17 +2544,22 @@ mdb_env_open2(MDB_env *env, unsigned int flags) DPRINTF("opened database version %u, pagesize %u", env->me_metas[toggle]->mm_version, env->me_psize); DPRINTF("depth: %u", env->me_metas[toggle]->mm_dbs[MAIN_DBI].md_depth); - DPRINTF("entries: %lu", env->me_metas[toggle]->mm_dbs[MAIN_DBI].md_entries); - DPRINTF("branch pages: %lu", env->me_metas[toggle]->mm_dbs[MAIN_DBI].md_branch_pages); - DPRINTF("leaf pages: %lu", env->me_metas[toggle]->mm_dbs[MAIN_DBI].md_leaf_pages); - DPRINTF("overflow pages: %lu", env->me_metas[toggle]->mm_dbs[MAIN_DBI].md_overflow_pages); - DPRINTF("root: %lu", env->me_metas[toggle]->mm_dbs[MAIN_DBI].md_root); + DPRINTF("entries: %zu", env->me_metas[toggle]->mm_dbs[MAIN_DBI].md_entries); + DPRINTF("branch pages: %zu", env->me_metas[toggle]->mm_dbs[MAIN_DBI].md_branch_pages); + DPRINTF("leaf pages: %zu", env->me_metas[toggle]->mm_dbs[MAIN_DBI].md_leaf_pages); + DPRINTF("overflow pages: %zu", env->me_metas[toggle]->mm_dbs[MAIN_DBI].md_overflow_pages); + DPRINTF("root: %zu", env->me_metas[toggle]->mm_dbs[MAIN_DBI].md_root); return MDB_SUCCESS; } #ifndef _WIN32 -/* Windows doesn't support destructor callbacks for thread-specific storage */ +/** Release a reader thread's slot in the reader lock table. + * This function is called automatically when a thread exits. + * Windows doesn't support destructor callbacks for thread-specific storage, + * so this function is not compiled there. + * @param[in] ptr This points to the slot in the reader lock table. + */ static void mdb_env_reader_dest(void *ptr) { @@ -1863,7 +2571,7 @@ mdb_env_reader_dest(void *ptr) } #endif -/* downgrade the exclusive lock on the region back to shared */ +/** Downgrade the exclusive lock on the region back to shared */ static void mdb_env_share_locks(MDB_env *env) { @@ -1897,21 +2605,102 @@ mdb_env_share_locks(MDB_env *env) } #endif } +#if defined(_WIN32) || defined(__APPLE__) +/* + * hash_64 - 64 bit Fowler/Noll/Vo-0 FNV-1a hash code + * + * @(#) $Revision: 5.1 $ + * @(#) $Id: hash_64a.c,v 5.1 2009/06/30 09:01:38 chongo Exp $ + * @(#) $Source: /usr/local/src/cmd/fnv/RCS/hash_64a.c,v $ + * + * http://www.isthe.com/chongo/tech/comp/fnv/index.html + * + *** + * + * Please do not copyright this code. This code is in the public domain. + * + * LANDON CURT NOLL DISCLAIMS ALL WARRANTIES WITH REGARD TO THIS SOFTWARE, + * INCLUDING ALL IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS. IN NO + * EVENT SHALL LANDON CURT NOLL BE LIABLE FOR ANY SPECIAL, INDIRECT OR + * CONSEQUENTIAL DAMAGES OR ANY DAMAGES WHATSOEVER RESULTING FROM LOSS OF + * USE, DATA OR PROFITS, WHETHER IN AN ACTION OF CONTRACT, NEGLIGENCE OR + * OTHER TORTIOUS ACTION, ARISING OUT OF OR IN CONNECTION WITH THE USE OR + * PERFORMANCE OF THIS SOFTWARE. + * + * By: + * chongo /\oo/\ + * http://www.isthe.com/chongo/ + * + * Share and Enjoy! :-) + */ -static int -mdb_env_setup_locks(MDB_env *env, char *lpath, int mode, int *excl) -{ - int rc; - off_t size, rsize; +typedef unsigned long long mdb_hash_t; +#define MDB_HASH_INIT ((mdb_hash_t)0xcbf29ce484222325ULL) - *excl = 0; +/** perform a 64 bit Fowler/Noll/Vo FNV-1a hash on a buffer + * @param[in] str string to hash + * @param[in] hval initial value for hash + * @return 64 bit hash + * + * NOTE: To use the recommended 64 bit FNV-1a hash, use MDB_HASH_INIT as the + * hval arg on the first call. + */ +static mdb_hash_t +mdb_hash_str(char *str, mdb_hash_t hval) +{ + unsigned char *s = (unsigned char *)str; /* unsigned string */ + /* + * FNV-1a hash each octet of the string + */ + while (*s) { + /* xor the bottom with the current octet */ + hval ^= (mdb_hash_t)*s++; -#ifdef _WIN32 - if ((env->me_lfd = CreateFile(lpath, GENERIC_READ|GENERIC_WRITE, - FILE_SHARE_READ|FILE_SHARE_WRITE, NULL, OPEN_ALWAYS, - FILE_ATTRIBUTE_NORMAL, NULL)) == INVALID_HANDLE_VALUE) { - rc = ErrCode(); - return rc; + /* multiply by the 64 bit FNV magic prime mod 2^64 */ + hval += (hval << 1) + (hval << 4) + (hval << 5) + + (hval << 7) + (hval << 8) + (hval << 40); + } + /* return our new hash value */ + return hval; +} + +/** Hash the string and output the hash in hex. + * @param[in] str string to hash + * @param[out] hexbuf an array of 17 chars to hold the hash + */ +static void +mdb_hash_hex(char *str, char *hexbuf) +{ + int i; + mdb_hash_t h = mdb_hash_str(str, MDB_HASH_INIT); + for (i=0; i<8; i++) { + hexbuf += sprintf(hexbuf, "%02x", (unsigned int)h & 0xff); + h >>= 8; + } +} +#endif + +/** Open and/or initialize the lock region for the environment. + * @param[in] env The MDB environment. + * @param[in] lpath The pathname of the file used for the lock region. + * @param[in] mode The Unix permissions for the file, if we create it. + * @param[out] excl Set to true if we got an exclusive lock on the region. + * @return 0 on success, non-zero on failure. + */ +static int +mdb_env_setup_locks(MDB_env *env, char *lpath, int mode, int *excl) +{ + int rc; + off_t size, rsize; + + *excl = 0; + +#ifdef _WIN32 + if ((env->me_lfd = CreateFile(lpath, GENERIC_READ|GENERIC_WRITE, + FILE_SHARE_READ|FILE_SHARE_WRITE, NULL, OPEN_ALWAYS, + FILE_ATTRIBUTE_NORMAL, NULL)) == INVALID_HANDLE_VALUE) { + rc = ErrCode(); + return rc; } /* Try to get exclusive lock. If we succeed, then * nobody is using the lock region and we should initialize it. @@ -1977,8 +2766,8 @@ mdb_env_setup_locks(MDB_env *env, char *lpath, int mode, int *excl) size = rsize - sizeof(MDB_txninfo); env->me_maxreaders = size/sizeof(MDB_reader) + 1; } -#ifdef _WIN32 { +#ifdef _WIN32 HANDLE mh; mh = CreateFileMapping(env->me_lfd, NULL, PAGE_READWRITE, 0, 0, NULL); @@ -1992,18 +2781,20 @@ mdb_env_setup_locks(MDB_env *env, char *lpath, int mode, int *excl) rc = ErrCode(); goto fail; } - } #else - env->me_txns = mmap(0, rsize, PROT_READ|PROT_WRITE, MAP_SHARED, - env->me_lfd, 0); - if (env->me_txns == MAP_FAILED) { - rc = ErrCode(); - goto fail; - } + void *m = mmap(NULL, rsize, PROT_READ|PROT_WRITE, MAP_SHARED, + env->me_lfd, 0); + if (m == MAP_FAILED) { + env->me_txns = NULL; + rc = ErrCode(); + goto fail; + } + env->me_txns = m; #endif + } if (*excl) { #ifdef _WIN32 - char *ptr; + char hexbuf[17]; if (!mdb_sec_inited) { InitializeSecurityDescriptor(&mdb_null_sd, SECURITY_DESCRIPTOR_REVISION); @@ -2013,28 +2804,46 @@ mdb_env_setup_locks(MDB_env *env, char *lpath, int mode, int *excl) mdb_all_sa.lpSecurityDescriptor = &mdb_null_sd; mdb_sec_inited = 1; } - /* FIXME: only using up to 20 characters of the env path here, - * probably not enough to assure uniqueness... - */ - sprintf(env->me_txns->mti_rmname, "Global\\MDBr%.20s", lpath); - ptr = env->me_txns->mti_rmname + sizeof("Global\\MDBr"); - while ((ptr = strchr(ptr, '\\'))) - *ptr++ = '/'; + mdb_hash_hex(lpath, hexbuf); + sprintf(env->me_txns->mti_rmname, "Global\\MDBr%s", hexbuf); env->me_rmutex = CreateMutex(&mdb_all_sa, FALSE, env->me_txns->mti_rmname); if (!env->me_rmutex) { rc = ErrCode(); goto fail; } - sprintf(env->me_txns->mti_rmname, "Global\\MDBw%.20s", lpath); - ptr = env->me_txns->mti_rmname + sizeof("Global\\MDBw"); - while ((ptr = strchr(ptr, '\\'))) - *ptr++ = '/'; - env->me_wmutex = CreateMutex(&mdb_all_sa, FALSE, env->me_txns->mti_rmname); + sprintf(env->me_txns->mti_wmname, "Global\\MDBw%s", hexbuf); + env->me_wmutex = CreateMutex(&mdb_all_sa, FALSE, env->me_txns->mti_wmname); if (!env->me_wmutex) { rc = ErrCode(); goto fail; } -#else +#else /* _WIN32 */ +#ifdef __APPLE__ + char hexbuf[17]; + mdb_hash_hex(lpath, hexbuf); + sprintf(env->me_txns->mti_rmname, "MDBr%s", hexbuf); + if (sem_unlink(env->me_txns->mti_rmname)) { + rc = ErrCode(); + if (rc != ENOENT && rc != EINVAL) + goto fail; + } + env->me_rmutex = sem_open(env->me_txns->mti_rmname, O_CREAT, mode, 1); + if (!env->me_rmutex) { + rc = ErrCode(); + goto fail; + } + sprintf(env->me_txns->mti_wmname, "MDBw%s", hexbuf); + if (sem_unlink(env->me_txns->mti_wmname)) { + rc = ErrCode(); + if (rc != ENOENT && rc != EINVAL) + goto fail; + } + env->me_wmutex = sem_open(env->me_txns->mti_wmname, O_CREAT, mode, 1); + if (!env->me_wmutex) { + rc = ErrCode(); + goto fail; + } +#else /* __APPLE__ */ pthread_mutexattr_t mattr; pthread_mutexattr_init(&mattr); @@ -2044,7 +2853,8 @@ mdb_env_setup_locks(MDB_env *env, char *lpath, int mode, int *excl) } pthread_mutex_init(&env->me_txns->mti_mutex, &mattr); pthread_mutex_init(&env->me_txns->mti_wmutex, &mattr); -#endif +#endif /* __APPLE__ */ +#endif /* _WIN32 */ env->me_txns->mti_version = MDB_VERSION; env->me_txns->mti_magic = MDB_MAGIC; env->me_txns->mti_txnid = 0; @@ -2078,6 +2888,18 @@ mdb_env_setup_locks(MDB_env *env, char *lpath, int mode, int *excl) rc = ErrCode(); goto fail; } +#endif +#ifdef __APPLE__ + env->me_rmutex = sem_open(env->me_txns->mti_rmname, 0); + if (!env->me_rmutex) { + rc = ErrCode(); + goto fail; + } + env->me_wmutex = sem_open(env->me_txns->mti_wmname, 0); + if (!env->me_wmutex) { + rc = ErrCode(); + goto fail; + } #endif } return MDB_SUCCESS; @@ -2093,6 +2915,9 @@ fail: #define LOCKNAME "/lock.mdb" /** The name of the data file in the DB environment */ #define DATANAME "/data.mdb" + /** The suffix of the lock file when no subdir is used */ +#define LOCKSUFF "-lock" + int mdb_env_open(MDB_env *env, const char *path, unsigned int flags, mode_t mode) { @@ -2100,12 +2925,23 @@ mdb_env_open(MDB_env *env, const char *path, unsigned int flags, mode_t mode) char *lpath, *dpath; len = strlen(path); - lpath = malloc(len + sizeof(LOCKNAME) + len + sizeof(DATANAME)); + if (flags & MDB_NOSUBDIR) { + rc = len + sizeof(LOCKSUFF) + len + 1; + } else { + rc = len + sizeof(LOCKNAME) + len + sizeof(DATANAME); + } + lpath = malloc(rc); if (!lpath) return ENOMEM; - dpath = lpath + len + sizeof(LOCKNAME); - sprintf(lpath, "%s" LOCKNAME, path); - sprintf(dpath, "%s" DATANAME, path); + if (flags & MDB_NOSUBDIR) { + dpath = lpath + len + sizeof(LOCKSUFF); + sprintf(lpath, "%s" LOCKSUFF, path); + strcpy(dpath, path); + } else { + dpath = lpath + len + sizeof(LOCKNAME); + sprintf(lpath, "%s" LOCKNAME, path); + sprintf(dpath, "%s" DATANAME, path); + } rc = mdb_env_setup_locks(env, lpath, mode, &excl); if (rc) @@ -2190,8 +3026,10 @@ mdb_env_close(MDB_env *env) if (env == NULL) return; + VGMEMP_DESTROY(env); while (env->me_dpages) { dp = env->me_dpages; + VGMEMP_DEFINED(&dp->mp_next, sizeof(dp->mp_next)); env->me_dpages = dp->mp_next; free(dp); } @@ -2215,40 +3053,41 @@ mdb_env_close(MDB_env *env) for (i=0; ime_txns->mti_numreaders; i++) if (env->me_txns->mti_readers[i].mr_pid == pid) env->me_txns->mti_readers[i].mr_pid = 0; - munmap(env->me_txns, (env->me_maxreaders-1)*sizeof(MDB_reader)+sizeof(MDB_txninfo)); + munmap((void *)env->me_txns, (env->me_maxreaders-1)*sizeof(MDB_reader)+sizeof(MDB_txninfo)); } close(env->me_lfd); + mdb_midl_free(env->me_free_pgs); free(env); } -/* only for aligned ints */ +/** Compare two items pointing at aligned size_t's */ static int -intcmp(const MDB_val *a, const MDB_val *b) +mdb_cmp_long(const MDB_val *a, const MDB_val *b) { - if (a->mv_size == sizeof(long)) - { - unsigned long *la, *lb; - la = a->mv_data; - lb = b->mv_data; - return *la - *lb; - } else { - unsigned int *ia, *ib; - ia = a->mv_data; - ib = b->mv_data; - return *ia - *ib; - } + return (*(size_t *)a->mv_data < *(size_t *)b->mv_data) ? -1 : + *(size_t *)a->mv_data > *(size_t *)b->mv_data; } -/* ints must always be the same size */ +/** Compare two items pointing at aligned int's */ static int -cintcmp(const MDB_val *a, const MDB_val *b) +mdb_cmp_int(const MDB_val *a, const MDB_val *b) { -#if __BYTE_ORDER == __LITTLE_ENDIAN + return (*(unsigned int *)a->mv_data < *(unsigned int *)b->mv_data) ? -1 : + *(unsigned int *)a->mv_data > *(unsigned int *)b->mv_data; +} + +/** Compare two items pointing at ints of unknown alignment. + * Nodes and keys are guaranteed to be 2-byte aligned. + */ +static int +mdb_cmp_cint(const MDB_val *a, const MDB_val *b) +{ +#if BYTE_ORDER == LITTLE_ENDIAN unsigned short *u, *c; int x; - u = a->mv_data + a->mv_size; - c = b->mv_data + a->mv_size; + u = (unsigned short *) ((char *) a->mv_data + a->mv_size); + c = (unsigned short *) ((char *) b->mv_data + a->mv_size); do { x = *--u - *--c; } while(!x && u > (unsigned short *)a->mv_data); @@ -2258,59 +3097,60 @@ cintcmp(const MDB_val *a, const MDB_val *b) #endif } +/** Compare two items lexically */ static int -memncmp(const MDB_val *a, const MDB_val *b) +mdb_cmp_memn(const MDB_val *a, const MDB_val *b) { - int diff, len_diff; + int diff; + ssize_t len_diff; unsigned int len; len = a->mv_size; - len_diff = a->mv_size - b->mv_size; - if (len_diff > 0) + len_diff = (ssize_t) a->mv_size - (ssize_t) b->mv_size; + if (len_diff > 0) { len = b->mv_size; + len_diff = 1; + } + diff = memcmp(a->mv_data, b->mv_data, len); - return diff ? diff : len_diff; + return diff ? diff : len_diff<0 ? -1 : len_diff; } +/** Compare two items in reverse byte order */ static int -memnrcmp(const MDB_val *a, const MDB_val *b) +mdb_cmp_memnr(const MDB_val *a, const MDB_val *b) { const unsigned char *p1, *p2, *p1_lim; - int diff, len_diff; - - if (b->mv_size == 0) - return a->mv_size != 0; - if (a->mv_size == 0) - return -1; + ssize_t len_diff; + int diff; - p1 = (const unsigned char *)a->mv_data + a->mv_size - 1; - p2 = (const unsigned char *)b->mv_data + b->mv_size - 1; + p1_lim = (const unsigned char *)a->mv_data; + p1 = (const unsigned char *)a->mv_data + a->mv_size; + p2 = (const unsigned char *)b->mv_data + b->mv_size; - len_diff = a->mv_size - b->mv_size; - if (len_diff < 0) - p1_lim = p1 - a->mv_size; - else - p1_lim = p1 - b->mv_size; + len_diff = (ssize_t) a->mv_size - (ssize_t) b->mv_size; + if (len_diff > 0) { + p1_lim += len_diff; + len_diff = 1; + } while (p1 > p1_lim) { - diff = *p1 - *p2; + diff = *--p1 - *--p2; if (diff) return diff; - p1--; - p2--; } - return len_diff; + return len_diff<0 ? -1 : len_diff; } -/* Search for key within a leaf page, using binary search. +/** Search for key within a page, using binary search. * Returns the smallest entry larger or equal to the key. * If exactp is non-null, stores whether the found entry was an exact match * in *exactp (1 or 0). - * If kip is non-null, stores the index of the found entry in *kip. + * Updates the cursor index with the index of the found entry. * If no entry larger or equal to the key is found, returns NULL. */ static MDB_node * -mdb_search_node(MDB_cursor *mc, MDB_val *key, int *exactp) +mdb_node_search(MDB_cursor *mc, MDB_val *key, int *exactp) { unsigned int i = 0, nkeys; int low, high; @@ -2323,48 +3163,72 @@ mdb_search_node(MDB_cursor *mc, MDB_val *key, int *exactp) nkeys = NUMKEYS(mp); - DPRINTF("searching %u keys in %s page %lu", - nkeys, IS_LEAF(mp) ? "leaf" : "branch", - mp->mp_pgno); +#if MDB_DEBUG + { + pgno_t pgno; + COPY_PGNO(pgno, mp->mp_pgno); + DPRINTF("searching %u keys in %s %spage %zu", + nkeys, IS_LEAF(mp) ? "leaf" : "branch", IS_SUBP(mp) ? "sub-" : "", + pgno); + } +#endif assert(nkeys > 0); low = IS_LEAF(mp) ? 0 : 1; high = nkeys - 1; - cmp = mc->mc_txn->mt_dbxs[mc->mc_dbi].md_cmp; - if (IS_LEAF2(mp)) { - nodekey.mv_size = mc->mc_txn->mt_dbs[mc->mc_dbi].md_pad; - node = NODEPTR(mp, 0); /* fake */ + cmp = mc->mc_dbx->md_cmp; + + /* Branch pages have no data, so if using integer keys, + * alignment is guaranteed. Use faster mdb_cmp_int. + */ + if (cmp == mdb_cmp_cint && IS_BRANCH(mp)) { + if (NODEPTR(mp, 1)->mn_ksize == sizeof(size_t)) + cmp = mdb_cmp_long; + else + cmp = mdb_cmp_int; } - while (low <= high) { - i = (low + high) >> 1; - if (IS_LEAF2(mp)) { + if (IS_LEAF2(mp)) { + nodekey.mv_size = mc->mc_db->md_pad; + node = NODEPTR(mp, 0); /* fake */ + while (low <= high) { + i = (low + high) >> 1; nodekey.mv_data = LEAF2KEY(mp, i, nodekey.mv_size); - } else { - node = NODEPTR(mp, i); - - nodekey.mv_size = node->mn_ksize; - nodekey.mv_data = NODEKEY(node); + rc = cmp(key, &nodekey); + DPRINTF("found leaf index %u [%s], rc = %i", + i, DKEY(&nodekey), rc); + if (rc == 0) + break; + if (rc > 0) + low = i + 1; + else + high = i - 1; } + } else { + while (low <= high) { + i = (low + high) >> 1; - rc = cmp(key, &nodekey); + node = NODEPTR(mp, i); + nodekey.mv_size = NODEKSZ(node); + nodekey.mv_data = NODEKEY(node); -#if DEBUG - if (IS_LEAF(mp)) - DPRINTF("found leaf index %u [%s], rc = %i", - i, DKEY(&nodekey), rc); - else - DPRINTF("found branch index %u [%s -> %lu], rc = %i", - i, DKEY(&nodekey), NODEPGNO(node), rc); + rc = cmp(key, &nodekey); +#if MDB_DEBUG + if (IS_LEAF(mp)) + DPRINTF("found leaf index %u [%s], rc = %i", + i, DKEY(&nodekey), rc); + else + DPRINTF("found branch index %u [%s -> %zu], rc = %i", + i, DKEY(&nodekey), NODEPGNO(node), rc); #endif - - if (rc == 0) - break; - if (rc > 0) - low = i + 1; - else - high = i - 1; + if (rc == 0) + break; + if (rc > 0) + low = i + 1; + else + high = i - 1; + } } if (rc > 0) { /* Found entry is less than the key. */ @@ -2384,8 +3248,23 @@ mdb_search_node(MDB_cursor *mc, MDB_val *key, int *exactp) return node; } +#if 0 +static void +mdb_cursor_adjust(MDB_cursor *mc, func) +{ + MDB_cursor *m2; + + for (m2 = mc->mc_txn->mt_cursors[mc->mc_dbi]; m2; m2=m2->mc_next) { + if (m2->mc_pg[m2->mc_top] == mc->mc_pg[mc->mc_top]) { + func(mc, m2); + } + } +} +#endif + +/** Pop a page off the top of the cursor's stack. */ static void -cursor_pop_page(MDB_cursor *mc) +mdb_cursor_pop(MDB_cursor *mc) { MDB_page *top; @@ -2395,19 +3274,22 @@ cursor_pop_page(MDB_cursor *mc) if (mc->mc_snum) mc->mc_top--; - DPRINTF("popped page %lu off db %u cursor %p", top->mp_pgno, + DPRINTF("popped page %zu off db %u cursor %p", top->mp_pgno, mc->mc_dbi, (void *) mc); } } +/** Push a page onto the top of the cursor's stack. */ static int -cursor_push_page(MDB_cursor *mc, MDB_page *mp) +mdb_cursor_push(MDB_cursor *mc, MDB_page *mp) { - DPRINTF("pushing page %lu on db %u cursor %p", mp->mp_pgno, + DPRINTF("pushing page %zu on db %u cursor %p", mp->mp_pgno, mc->mc_dbi, (void *) mc); - if (mc->mc_snum >= CURSOR_STACK) + if (mc->mc_snum >= CURSOR_STACK) { + assert(mc->mc_snum < CURSOR_STACK); return ENOMEM; + } mc->mc_top = mc->mc_snum++; mc->mc_pg[mc->mc_top] = mp; @@ -2416,8 +3298,14 @@ cursor_push_page(MDB_cursor *mc, MDB_page *mp) return MDB_SUCCESS; } +/** Find the address of the page corresponding to a given page number. + * @param[in] txn the transaction for this access. + * @param[in] pgno the page number for the page to retrieve. + * @param[out] ret address of a pointer where the page's address will be stored. + * @return 0 on success, non-zero on failure. + */ static int -mdb_get_page(MDB_txn *txn, pgno_t pgno, MDB_page **ret) +mdb_page_get(MDB_txn *txn, pgno_t pgno, MDB_page **ret) { MDB_page *p = NULL; @@ -2434,14 +3322,24 @@ mdb_get_page(MDB_txn *txn, pgno_t pgno, MDB_page **ret) } *ret = p; if (!p) { - DPRINTF("page %lu not found", pgno); + DPRINTF("page %zu not found", pgno); assert(p != NULL); } return (p != NULL) ? MDB_SUCCESS : MDB_PAGE_NOTFOUND; } +/** Search for the page a given key should be in. + * Pushes parent pages on the cursor stack. This function continues a + * search on a cursor that has already been initialized. (Usually by + * #mdb_page_search() but also by #mdb_node_move().) + * @param[in,out] mc the cursor for this operation. + * @param[in] key the key to search for. If NULL, search for the lowest + * page. (This is used by #mdb_cursor_first().) + * @param[in] modify If true, visited pages are updated with new page numbers. + * @return 0 on success, non-zero on failure. + */ static int -mdb_search_page_root(MDB_cursor *mc, MDB_val *key, int modify) +mdb_page_search_root(MDB_cursor *mc, MDB_val *key, int modify) { MDB_page *mp = mc->mc_pg[mc->mc_top]; DKBUF; @@ -2450,41 +3348,46 @@ mdb_search_page_root(MDB_cursor *mc, MDB_val *key, int modify) while (IS_BRANCH(mp)) { MDB_node *node; + indx_t i; - DPRINTF("branch page %lu has %u keys", mp->mp_pgno, NUMKEYS(mp)); + DPRINTF("branch page %zu has %u keys", mp->mp_pgno, NUMKEYS(mp)); assert(NUMKEYS(mp) > 1); - DPRINTF("found index 0 to page %lu", NODEPGNO(NODEPTR(mp, 0))); + DPRINTF("found index 0 to page %zu", NODEPGNO(NODEPTR(mp, 0))); if (key == NULL) /* Initialize cursor to first page. */ - mc->mc_ki[mc->mc_top] = 0; + i = 0; else if (key->mv_size > MAXKEYSIZE && key->mv_data == NULL) { /* cursor to last page */ - mc->mc_ki[mc->mc_top] = NUMKEYS(mp)-1; + i = NUMKEYS(mp)-1; } else { int exact; - node = mdb_search_node(mc, key, &exact); + node = mdb_node_search(mc, key, &exact); if (node == NULL) - mc->mc_ki[mc->mc_top] = NUMKEYS(mp) - 1; - else if (!exact) { - assert(mc->mc_ki[mc->mc_top] > 0); - mc->mc_ki[mc->mc_top]--; + i = NUMKEYS(mp) - 1; + else { + i = mc->mc_ki[mc->mc_top]; + if (!exact) { + assert(i > 0); + i--; + } } } if (key) DPRINTF("following index %u for key [%s]", - mc->mc_ki[mc->mc_top], DKEY(key)); - assert(mc->mc_ki[mc->mc_top] < NUMKEYS(mp)); - node = NODEPTR(mp, mc->mc_ki[mc->mc_top]); + i, DKEY(key)); + assert(i < NUMKEYS(mp)); + node = NODEPTR(mp, i); - if ((rc = mdb_get_page(mc->mc_txn, NODEPGNO(node), &mp))) + if ((rc = mdb_page_get(mc->mc_txn, NODEPGNO(node), &mp))) return rc; - if ((rc = cursor_push_page(mc, mp))) + mc->mc_ki[mc->mc_top] = i; + if ((rc = mdb_cursor_push(mc, mp))) return rc; if (modify) { - if ((rc = mdb_touch(mc)) != 0) + if ((rc = mdb_page_touch(mc)) != 0) return rc; mp = mc->mc_pg[mc->mc_top]; } @@ -2496,19 +3399,25 @@ mdb_search_page_root(MDB_cursor *mc, MDB_val *key, int modify) return MDB_CORRUPTED; } - DPRINTF("found leaf page %lu for key [%s]", mp->mp_pgno, + DPRINTF("found leaf page %zu for key [%s]", mp->mp_pgno, key ? DKEY(key) : NULL); return MDB_SUCCESS; } -/* Search for the page a given key should be in. - * Pushes parent pages on the cursor stack. - * If key is NULL, search for the lowest page (used by mdb_cursor_first). - * If modify is true, visited pages are updated with new page numbers. +/** Search for the page a given key should be in. + * Pushes parent pages on the cursor stack. This function just sets up + * the search; it finds the root page for \b mc's database and sets this + * as the root of the cursor's stack. Then #mdb_page_search_root() is + * called to complete the search. + * @param[in,out] mc the cursor for this operation. + * @param[in] key the key to search for. If NULL, search for the lowest + * page. (This is used by #mdb_cursor_first().) + * @param[in] modify If true, visited pages are updated with new page numbers. + * @return 0 on success, non-zero on failure. */ static int -mdb_search_page(MDB_cursor *mc, MDB_val *key, int modify) +mdb_page_search(MDB_cursor *mc, MDB_val *key, int modify) { int rc; pgno_t root; @@ -2519,48 +3428,68 @@ mdb_search_page(MDB_cursor *mc, MDB_val *key, int modify) if (F_ISSET(mc->mc_txn->mt_flags, MDB_TXN_ERROR)) { DPUTS("transaction has failed, must abort"); return EINVAL; - } else - root = mc->mc_txn->mt_dbs[mc->mc_dbi].md_root; + } else { + /* Make sure we're using an up-to-date root */ + if (mc->mc_dbi > MAIN_DBI) { + if ((*mc->mc_dbflag & DB_STALE) || + (modify && !(*mc->mc_dbflag & DB_DIRTY))) { + MDB_cursor mc2; + unsigned char dbflag = 0; + mdb_cursor_init(&mc2, mc->mc_txn, MAIN_DBI, NULL); + rc = mdb_page_search(&mc2, &mc->mc_dbx->md_name, modify); + if (rc) + return rc; + if (*mc->mc_dbflag & DB_STALE) { + MDB_val data; + int exact = 0; + MDB_node *leaf = mdb_node_search(&mc2, + &mc->mc_dbx->md_name, &exact); + if (!exact) + return MDB_NOTFOUND; + mdb_node_read(mc->mc_txn, leaf, &data); + memcpy(mc->mc_db, data.mv_data, sizeof(MDB_db)); + } + if (modify) + dbflag = DB_DIRTY; + *mc->mc_dbflag = dbflag; + } + } + root = mc->mc_db->md_root; - if (root == P_INVALID) { /* Tree is empty. */ - DPUTS("tree is empty"); - return MDB_NOTFOUND; + if (root == P_INVALID) { /* Tree is empty. */ + DPUTS("tree is empty"); + return MDB_NOTFOUND; + } } - if ((rc = mdb_get_page(mc->mc_txn, root, &mc->mc_pg[0]))) + assert(root > 1); + if ((rc = mdb_page_get(mc->mc_txn, root, &mc->mc_pg[0]))) return rc; mc->mc_snum = 1; mc->mc_top = 0; - DPRINTF("db %u root page %lu has flags 0x%X", + DPRINTF("db %u root page %zu has flags 0x%X", mc->mc_dbi, root, mc->mc_pg[0]->mp_flags); if (modify) { - /* For sub-databases, update main root first */ - if (mc->mc_dbi > MAIN_DBI && !mc->mc_txn->mt_dbxs[mc->mc_dbi].md_dirty) { - MDB_cursor mc2; - mc2.mc_txn = mc->mc_txn; - mc2.mc_dbi = MAIN_DBI; - rc = mdb_search_page(&mc2, &mc->mc_txn->mt_dbxs[mc->mc_dbi].md_name, 1); - if (rc) - return rc; - mc->mc_txn->mt_dbxs[mc->mc_dbi].md_dirty = 1; - } - if (!F_ISSET(mc->mc_pg[0]->mp_flags, P_DIRTY)) { - if ((rc = mdb_touch(mc))) - return rc; - mc->mc_txn->mt_dbs[mc->mc_dbi].md_root = mc->mc_pg[0]->mp_pgno; - } + if ((rc = mdb_page_touch(mc))) + return rc; } - return mdb_search_page_root(mc, key, modify); + return mdb_page_search_root(mc, key, modify); } +/** Return the data associated with a given node. + * @param[in] txn The transaction for this operation. + * @param[in] leaf The node being read. + * @param[out] data Updated to point to the node's data. + * @return 0 on success, non-zero on failure. + */ static int -mdb_read_data(MDB_txn *txn, MDB_node *leaf, MDB_val *data) +mdb_node_read(MDB_txn *txn, MDB_node *leaf, MDB_val *data) { - MDB_page *omp; /* overflow mpage */ + MDB_page *omp; /* overflow page */ pgno_t pgno; int rc; @@ -2574,8 +3503,8 @@ mdb_read_data(MDB_txn *txn, MDB_node *leaf, MDB_val *data) */ data->mv_size = NODEDSZ(leaf); memcpy(&pgno, NODEDATA(leaf), sizeof(pgno)); - if ((rc = mdb_get_page(txn, pgno, &omp))) { - DPRINTF("read overflow page %lu failed", pgno); + if ((rc = mdb_page_get(txn, pgno, &omp))) { + DPRINTF("read overflow page %zu failed", pgno); return rc; } data->mv_data = METADATA(omp); @@ -2603,60 +3532,59 @@ mdb_get(MDB_txn *txn, MDB_dbi dbi, return EINVAL; } - mc.mc_txn = txn; - mc.mc_dbi = dbi; - mc.mc_flags = 0; - if (txn->mt_dbs[dbi].md_flags & MDB_DUPSORT) { - mc.mc_xcursor = &mx; - mdb_xcursor_init0(&mc); - } else { - mc.mc_xcursor = NULL; - } + mdb_cursor_init(&mc, txn, dbi, &mx); return mdb_cursor_set(&mc, key, data, MDB_SET, &exact); } +/** Find a sibling for a page. + * Replaces the page at the top of the cursor's stack with the + * specified sibling, if one exists. + * @param[in] mc The cursor for this operation. + * @param[in] move_right Non-zero if the right sibling is requested, + * otherwise the left sibling. + * @return 0 on success, non-zero on failure. + */ static int -mdb_sibling(MDB_cursor *mc, int move_right) +mdb_cursor_sibling(MDB_cursor *mc, int move_right) { int rc; - unsigned int ptop; MDB_node *indx; MDB_page *mp; if (mc->mc_snum < 2) { return MDB_NOTFOUND; /* root has no siblings */ } - ptop = mc->mc_top-1; - DPRINTF("parent page is page %lu, index %u", - mc->mc_pg[ptop]->mp_pgno, mc->mc_ki[ptop]); + mdb_cursor_pop(mc); + DPRINTF("parent page is page %zu, index %u", + mc->mc_pg[mc->mc_top]->mp_pgno, mc->mc_ki[mc->mc_top]); - cursor_pop_page(mc); - if (move_right ? (mc->mc_ki[ptop] + 1u >= NUMKEYS(mc->mc_pg[ptop])) - : (mc->mc_ki[ptop] == 0)) { + if (move_right ? (mc->mc_ki[mc->mc_top] + 1u >= NUMKEYS(mc->mc_pg[mc->mc_top])) + : (mc->mc_ki[mc->mc_top] == 0)) { DPRINTF("no more keys left, moving to %s sibling", move_right ? "right" : "left"); - if ((rc = mdb_sibling(mc, move_right)) != MDB_SUCCESS) + if ((rc = mdb_cursor_sibling(mc, move_right)) != MDB_SUCCESS) return rc; } else { if (move_right) - mc->mc_ki[ptop]++; + mc->mc_ki[mc->mc_top]++; else - mc->mc_ki[ptop]--; + mc->mc_ki[mc->mc_top]--; DPRINTF("just moving to %s index key %u", - move_right ? "right" : "left", mc->mc_ki[ptop]); + move_right ? "right" : "left", mc->mc_ki[mc->mc_top]); } - assert(IS_BRANCH(mc->mc_pg[ptop])); + assert(IS_BRANCH(mc->mc_pg[mc->mc_top])); - indx = NODEPTR(mc->mc_pg[ptop], mc->mc_ki[ptop]); - if ((rc = mdb_get_page(mc->mc_txn, NODEPGNO(indx), &mp))) + indx = NODEPTR(mc->mc_pg[mc->mc_top], mc->mc_ki[mc->mc_top]); + if ((rc = mdb_page_get(mc->mc_txn, NODEPGNO(indx), &mp))) return rc;; - cursor_push_page(mc, mp); + mdb_cursor_push(mc, mp); return MDB_SUCCESS; } +/** Move the cursor to the next data item. */ static int mdb_cursor_next(MDB_cursor *mc, MDB_val *key, MDB_val *data, MDB_cursor_op op) { @@ -2672,7 +3600,7 @@ mdb_cursor_next(MDB_cursor *mc, MDB_val *key, MDB_val *data, MDB_cursor_op op) mp = mc->mc_pg[mc->mc_top]; - if (mc->mc_txn->mt_dbs[mc->mc_dbi].md_flags & MDB_DUPSORT) { + if (mc->mc_db->md_flags & MDB_DUPSORT) { leaf = NODEPTR(mp, mc->mc_ki[mc->mc_top]); if (F_ISSET(leaf->mn_flags, F_DUPDATA)) { if (op == MDB_NEXT || op == MDB_NEXT_DUP) { @@ -2681,30 +3609,31 @@ mdb_cursor_next(MDB_cursor *mc, MDB_val *key, MDB_val *data, MDB_cursor_op op) return rc; } } else { - mc->mc_xcursor->mx_cursor.mc_flags = 0; + mc->mc_xcursor->mx_cursor.mc_flags &= ~C_INITIALIZED; if (op == MDB_NEXT_DUP) return MDB_NOTFOUND; } } - DPRINTF("cursor_next: top page is %lu in cursor %p", mp->mp_pgno, (void *) mc); + DPRINTF("cursor_next: top page is %zu in cursor %p", mp->mp_pgno, (void *) mc); if (mc->mc_ki[mc->mc_top] + 1u >= NUMKEYS(mp)) { DPUTS("=====> move to next sibling page"); - if (mdb_sibling(mc, 1) != MDB_SUCCESS) { + if (mdb_cursor_sibling(mc, 1) != MDB_SUCCESS) { mc->mc_flags |= C_EOF; + mc->mc_flags &= ~C_INITIALIZED; return MDB_NOTFOUND; } mp = mc->mc_pg[mc->mc_top]; - DPRINTF("next page is %lu, key index %u", mp->mp_pgno, mc->mc_ki[mc->mc_top]); + DPRINTF("next page is %zu, key index %u", mp->mp_pgno, mc->mc_ki[mc->mc_top]); } else mc->mc_ki[mc->mc_top]++; - DPRINTF("==> cursor points to page %lu with %u keys, key index %u", + DPRINTF("==> cursor points to page %zu with %u keys, key index %u", mp->mp_pgno, NUMKEYS(mp), mc->mc_ki[mc->mc_top]); if (IS_LEAF2(mp)) { - key->mv_size = mc->mc_txn->mt_dbs[mc->mc_dbi].md_pad; + key->mv_size = mc->mc_db->md_pad; key->mv_data = LEAF2KEY(mp, mc->mc_ki[mc->mc_top], key->mv_size); return MDB_SUCCESS; } @@ -2716,7 +3645,7 @@ mdb_cursor_next(MDB_cursor *mc, MDB_val *key, MDB_val *data, MDB_cursor_op op) mdb_xcursor_init1(mc, leaf); } if (data) { - if ((rc = mdb_read_data(mc->mc_txn, leaf, data) != MDB_SUCCESS)) + if ((rc = mdb_node_read(mc->mc_txn, leaf, data) != MDB_SUCCESS)) return rc; if (F_ISSET(leaf->mn_flags, F_DUPDATA)) { @@ -2730,6 +3659,7 @@ mdb_cursor_next(MDB_cursor *mc, MDB_val *key, MDB_val *data, MDB_cursor_op op) return MDB_SUCCESS; } +/** Move the cursor to the previous data item. */ static int mdb_cursor_prev(MDB_cursor *mc, MDB_val *key, MDB_val *data, MDB_cursor_op op) { @@ -2741,7 +3671,7 @@ mdb_cursor_prev(MDB_cursor *mc, MDB_val *key, MDB_val *data, MDB_cursor_op op) mp = mc->mc_pg[mc->mc_top]; - if (mc->mc_txn->mt_dbs[mc->mc_dbi].md_flags & MDB_DUPSORT) { + if (mc->mc_db->md_flags & MDB_DUPSORT) { leaf = NODEPTR(mp, mc->mc_ki[mc->mc_top]); if (op == MDB_PREV || op == MDB_PREV_DUP) { if (F_ISSET(leaf->mn_flags, F_DUPDATA)) { @@ -2749,34 +3679,34 @@ mdb_cursor_prev(MDB_cursor *mc, MDB_val *key, MDB_val *data, MDB_cursor_op op) if (op != MDB_PREV || rc == MDB_SUCCESS) return rc; } else { - mc->mc_xcursor->mx_cursor.mc_flags = 0; + mc->mc_xcursor->mx_cursor.mc_flags &= ~C_INITIALIZED; if (op == MDB_PREV_DUP) return MDB_NOTFOUND; } } } - DPRINTF("cursor_prev: top page is %lu in cursor %p", mp->mp_pgno, (void *) mc); + DPRINTF("cursor_prev: top page is %zu in cursor %p", mp->mp_pgno, (void *) mc); if (mc->mc_ki[mc->mc_top] == 0) { DPUTS("=====> move to prev sibling page"); - if (mdb_sibling(mc, 0) != MDB_SUCCESS) { + if (mdb_cursor_sibling(mc, 0) != MDB_SUCCESS) { mc->mc_flags &= ~C_INITIALIZED; return MDB_NOTFOUND; } mp = mc->mc_pg[mc->mc_top]; mc->mc_ki[mc->mc_top] = NUMKEYS(mp) - 1; - DPRINTF("prev page is %lu, key index %u", mp->mp_pgno, mc->mc_ki[mc->mc_top]); + DPRINTF("prev page is %zu, key index %u", mp->mp_pgno, mc->mc_ki[mc->mc_top]); } else mc->mc_ki[mc->mc_top]--; mc->mc_flags &= ~C_EOF; - DPRINTF("==> cursor points to page %lu with %u keys, key index %u", + DPRINTF("==> cursor points to page %zu with %u keys, key index %u", mp->mp_pgno, NUMKEYS(mp), mc->mc_ki[mc->mc_top]); if (IS_LEAF2(mp)) { - key->mv_size = mc->mc_txn->mt_dbs[mc->mc_dbi].md_pad; + key->mv_size = mc->mc_db->md_pad; key->mv_data = LEAF2KEY(mp, mc->mc_ki[mc->mc_top], key->mv_size); return MDB_SUCCESS; } @@ -2788,7 +3718,7 @@ mdb_cursor_prev(MDB_cursor *mc, MDB_val *key, MDB_val *data, MDB_cursor_op op) mdb_xcursor_init1(mc, leaf); } if (data) { - if ((rc = mdb_read_data(mc->mc_txn, leaf, data) != MDB_SUCCESS)) + if ((rc = mdb_node_read(mc->mc_txn, leaf, data) != MDB_SUCCESS)) return rc; if (F_ISSET(leaf->mn_flags, F_DUPDATA)) { @@ -2802,11 +3732,13 @@ mdb_cursor_prev(MDB_cursor *mc, MDB_val *key, MDB_val *data, MDB_cursor_op op) return MDB_SUCCESS; } +/** Set the cursor on a specific data item. */ static int mdb_cursor_set(MDB_cursor *mc, MDB_val *key, MDB_val *data, MDB_cursor_op op, int *exactp) { int rc; + MDB_page *mp; MDB_node *leaf; DKBUF; @@ -2818,40 +3750,47 @@ mdb_cursor_set(MDB_cursor *mc, MDB_val *key, MDB_val *data, if (mc->mc_flags & C_INITIALIZED) { MDB_val nodekey; - if (mc->mc_pg[mc->mc_top]->mp_flags & P_LEAF2) { - nodekey.mv_size = mc->mc_txn->mt_dbs[mc->mc_dbi].md_pad; - nodekey.mv_data = LEAF2KEY(mc->mc_pg[mc->mc_top], 0, nodekey.mv_size); + mp = mc->mc_pg[mc->mc_top]; + if (!NUMKEYS(mp)) { + mc->mc_ki[mc->mc_top] = 0; + return MDB_NOTFOUND; + } + if (mp->mp_flags & P_LEAF2) { + nodekey.mv_size = mc->mc_db->md_pad; + nodekey.mv_data = LEAF2KEY(mp, 0, nodekey.mv_size); } else { - leaf = NODEPTR(mc->mc_pg[mc->mc_top], 0); + leaf = NODEPTR(mp, 0); MDB_SET_KEY(leaf, &nodekey); } - rc = mc->mc_txn->mt_dbxs[mc->mc_dbi].md_cmp(key, &nodekey); + rc = mc->mc_dbx->md_cmp(key, &nodekey); if (rc == 0) { /* Probably happens rarely, but first node on the page * was the one we wanted. */ mc->mc_ki[mc->mc_top] = 0; -set1: + leaf = NODEPTR(mp, 0); if (exactp) *exactp = 1; - rc = 0; - leaf = NODEPTR(mc->mc_pg[mc->mc_top], mc->mc_ki[mc->mc_top]); - goto set3; + goto set1; } if (rc > 0) { unsigned int i; - if (NUMKEYS(mc->mc_pg[mc->mc_top]) > 1) { - if (mc->mc_pg[mc->mc_top]->mp_flags & P_LEAF2) { - nodekey.mv_data = LEAF2KEY(mc->mc_pg[mc->mc_top], - NUMKEYS(mc->mc_pg[mc->mc_top])-1, nodekey.mv_size); + unsigned int nkeys = NUMKEYS(mp); + if (nkeys > 1) { + if (mp->mp_flags & P_LEAF2) { + nodekey.mv_data = LEAF2KEY(mp, + nkeys-1, nodekey.mv_size); } else { - leaf = NODEPTR(mc->mc_pg[mc->mc_top], NUMKEYS(mc->mc_pg[mc->mc_top])-1); + leaf = NODEPTR(mp, nkeys-1); MDB_SET_KEY(leaf, &nodekey); } - rc = mc->mc_txn->mt_dbxs[mc->mc_dbi].md_cmp(key, &nodekey); + rc = mc->mc_dbx->md_cmp(key, &nodekey); if (rc == 0) { /* last node was the one we wanted */ - mc->mc_ki[mc->mc_top] = NUMKEYS(mc->mc_pg[mc->mc_top])-1; + mc->mc_ki[mc->mc_top] = nkeys-1; + leaf = NODEPTR(mp, nkeys-1); + if (exactp) + *exactp = 1; goto set1; } if (rc < 0) { @@ -2869,20 +3808,26 @@ set1: break; if (i == mc->mc_top) { /* There are no other pages */ - mc->mc_ki[mc->mc_top] = NUMKEYS(mc->mc_pg[mc->mc_top]); + mc->mc_ki[mc->mc_top] = nkeys; return MDB_NOTFOUND; } } + if (!mc->mc_top) { + /* There are no other pages */ + mc->mc_ki[mc->mc_top] = 0; + return MDB_NOTFOUND; + } } - rc = mdb_search_page(mc, key, 0); + rc = mdb_page_search(mc, key, 0); if (rc != MDB_SUCCESS) return rc; - assert(IS_LEAF(mc->mc_pg[mc->mc_top])); + mp = mc->mc_pg[mc->mc_top]; + assert(IS_LEAF(mp)); set2: - leaf = mdb_search_node(mc, key, exactp); + leaf = mdb_node_search(mc, key, exactp); if (exactp != NULL && !*exactp) { /* MDB_SET specified and not an exact match. */ return MDB_NOTFOUND; @@ -2890,20 +3835,20 @@ set2: if (leaf == NULL) { DPUTS("===> inexact leaf not found, goto sibling"); - if ((rc = mdb_sibling(mc, 1)) != MDB_SUCCESS) + if ((rc = mdb_cursor_sibling(mc, 1)) != MDB_SUCCESS) return rc; /* no entries matched */ - mc->mc_ki[mc->mc_top] = 0; - assert(IS_LEAF(mc->mc_pg[mc->mc_top])); - leaf = NODEPTR(mc->mc_pg[mc->mc_top], 0); + mp = mc->mc_pg[mc->mc_top]; + assert(IS_LEAF(mp)); + leaf = NODEPTR(mp, 0); } -set3: +set1: mc->mc_flags |= C_INITIALIZED; mc->mc_flags &= ~C_EOF; - if (IS_LEAF2(mc->mc_pg[mc->mc_top])) { - key->mv_size = mc->mc_txn->mt_dbs[mc->mc_dbi].md_pad; - key->mv_data = LEAF2KEY(mc->mc_pg[mc->mc_top], mc->mc_ki[mc->mc_top], key->mv_size); + if (IS_LEAF2(mp)) { + key->mv_size = mc->mc_db->md_pad; + key->mv_data = LEAF2KEY(mp, mc->mc_ki[mc->mc_top], key->mv_size); return MDB_SUCCESS; } @@ -2928,16 +3873,18 @@ set3: } } else if (op == MDB_GET_BOTH || op == MDB_GET_BOTH_RANGE) { MDB_val d2; - if ((rc = mdb_read_data(mc->mc_txn, leaf, &d2)) != MDB_SUCCESS) + if ((rc = mdb_node_read(mc->mc_txn, leaf, &d2)) != MDB_SUCCESS) return rc; - rc = mc->mc_txn->mt_dbxs[mc->mc_dbi].md_dcmp(data, &d2); + rc = mc->mc_dbx->md_dcmp(data, &d2); if (rc) { if (op == MDB_GET_BOTH || rc > 0) return MDB_NOTFOUND; } } else { - if ((rc = mdb_read_data(mc->mc_txn, leaf, data)) != MDB_SUCCESS) + if (mc->mc_xcursor) + mc->mc_xcursor->mx_cursor.mc_flags &= ~C_INITIALIZED; + if ((rc = mdb_node_read(mc->mc_txn, leaf, data)) != MDB_SUCCESS) return rc; } } @@ -2950,23 +3897,28 @@ set3: return rc; } +/** Move the cursor to the first item in the database. */ static int mdb_cursor_first(MDB_cursor *mc, MDB_val *key, MDB_val *data) { int rc; MDB_node *leaf; - rc = mdb_search_page(mc, NULL, 0); - if (rc != MDB_SUCCESS) - return rc; + if (!(mc->mc_flags & C_INITIALIZED) || mc->mc_top) { + rc = mdb_page_search(mc, NULL, 0); + if (rc != MDB_SUCCESS) + return rc; + } assert(IS_LEAF(mc->mc_pg[mc->mc_top])); leaf = NODEPTR(mc->mc_pg[mc->mc_top], 0); mc->mc_flags |= C_INITIALIZED; mc->mc_flags &= ~C_EOF; + mc->mc_ki[mc->mc_top] = 0; + if (IS_LEAF2(mc->mc_pg[mc->mc_top])) { - key->mv_size = mc->mc_txn->mt_dbs[mc->mc_dbi].md_pad; + key->mv_size = mc->mc_db->md_pad; key->mv_data = LEAF2KEY(mc->mc_pg[mc->mc_top], 0, key->mv_size); return MDB_SUCCESS; } @@ -2979,8 +3931,8 @@ mdb_cursor_first(MDB_cursor *mc, MDB_val *key, MDB_val *data) return rc; } else { if (mc->mc_xcursor) - mc->mc_xcursor->mx_cursor.mc_flags = 0; - if ((rc = mdb_read_data(mc->mc_txn, leaf, data)) != MDB_SUCCESS) + mc->mc_xcursor->mx_cursor.mc_flags &= ~C_INITIALIZED; + if ((rc = mdb_node_read(mc->mc_txn, leaf, data)) != MDB_SUCCESS) return rc; } } @@ -2988,6 +3940,7 @@ mdb_cursor_first(MDB_cursor *mc, MDB_val *key, MDB_val *data) return MDB_SUCCESS; } +/** Move the cursor to the last item in the database. */ static int mdb_cursor_last(MDB_cursor *mc, MDB_val *key, MDB_val *data) { @@ -2998,9 +3951,11 @@ mdb_cursor_last(MDB_cursor *mc, MDB_val *key, MDB_val *data) lkey.mv_size = MAXKEYSIZE+1; lkey.mv_data = NULL; - rc = mdb_search_page(mc, &lkey, 0); - if (rc != MDB_SUCCESS) - return rc; + if (!(mc->mc_flags & C_INITIALIZED) || mc->mc_top) { + rc = mdb_page_search(mc, &lkey, 0); + if (rc != MDB_SUCCESS) + return rc; + } assert(IS_LEAF(mc->mc_pg[mc->mc_top])); leaf = NODEPTR(mc->mc_pg[mc->mc_top], NUMKEYS(mc->mc_pg[mc->mc_top])-1); @@ -3010,7 +3965,7 @@ mdb_cursor_last(MDB_cursor *mc, MDB_val *key, MDB_val *data) mc->mc_ki[mc->mc_top] = NUMKEYS(mc->mc_pg[mc->mc_top]) - 1; if (IS_LEAF2(mc->mc_pg[mc->mc_top])) { - key->mv_size = mc->mc_txn->mt_dbs[mc->mc_dbi].md_pad; + key->mv_size = mc->mc_db->md_pad; key->mv_data = LEAF2KEY(mc->mc_pg[mc->mc_top], mc->mc_ki[mc->mc_top], key->mv_size); return MDB_SUCCESS; } @@ -3022,7 +3977,9 @@ mdb_cursor_last(MDB_cursor *mc, MDB_val *key, MDB_val *data) if (rc) return rc; } else { - if ((rc = mdb_read_data(mc->mc_txn, leaf, data)) != MDB_SUCCESS) + if (mc->mc_xcursor) + mc->mc_xcursor->mx_cursor.mc_flags &= ~C_INITIALIZED; + if ((rc = mdb_node_read(mc->mc_txn, leaf, data)) != MDB_SUCCESS) return rc; } } @@ -3059,7 +4016,7 @@ mdb_cursor_get(MDB_cursor *mc, MDB_val *key, MDB_val *data, break; case MDB_GET_MULTIPLE: if (data == NULL || - !(mc->mc_txn->mt_dbs[mc->mc_dbi].md_flags & MDB_DUPFIXED) || + !(mc->mc_db->md_flags & MDB_DUPFIXED) || !(mc->mc_flags & C_INITIALIZED)) { rc = EINVAL; break; @@ -3071,7 +4028,7 @@ mdb_cursor_get(MDB_cursor *mc, MDB_val *key, MDB_val *data, goto fetchm; case MDB_NEXT_MULTIPLE: if (data == NULL || - !(mc->mc_txn->mt_dbs[mc->mc_dbi].md_flags & MDB_DUPFIXED)) { + !(mc->mc_db->md_flags & MDB_DUPFIXED)) { rc = EINVAL; break; } @@ -3085,7 +4042,7 @@ mdb_cursor_get(MDB_cursor *mc, MDB_val *key, MDB_val *data, fetchm: mx = &mc->mc_xcursor->mx_cursor; data->mv_size = NUMKEYS(mx->mc_pg[mx->mc_top]) * - mx->mc_txn->mt_dbs[mx->mc_dbi].md_pad; + mx->mc_db->md_pad; data->mv_data = METADATA(mx->mc_pg[mx->mc_top]); mx->mc_ki[mx->mc_top] = NUMKEYS(mx->mc_pg[mx->mc_top])-1; } else { @@ -3114,7 +4071,7 @@ fetchm: break; case MDB_FIRST_DUP: if (data == NULL || - !(mc->mc_txn->mt_dbs[mc->mc_dbi].md_flags & MDB_DUPSORT) || + !(mc->mc_db->md_flags & MDB_DUPSORT) || !(mc->mc_flags & C_INITIALIZED) || !(mc->mc_xcursor->mx_cursor.mc_flags & C_INITIALIZED)) { rc = EINVAL; @@ -3127,7 +4084,7 @@ fetchm: break; case MDB_LAST_DUP: if (data == NULL || - !(mc->mc_txn->mt_dbs[mc->mc_dbi].md_flags & MDB_DUPSORT) || + !(mc->mc_db->md_flags & MDB_DUPSORT) || !(mc->mc_flags & C_INITIALIZED) || !(mc->mc_xcursor->mx_cursor.mc_flags & C_INITIALIZED)) { rc = EINVAL; @@ -3144,28 +4101,27 @@ fetchm: return rc; } +/** Touch all the pages in the cursor stack. + * Makes sure all the pages are writable, before attempting a write operation. + * @param[in] mc The cursor to operate on. + */ static int mdb_cursor_touch(MDB_cursor *mc) { int rc; - if (mc->mc_dbi > MAIN_DBI && !mc->mc_txn->mt_dbxs[mc->mc_dbi].md_dirty) { + if (mc->mc_dbi > MAIN_DBI && !(*mc->mc_dbflag & DB_DIRTY)) { MDB_cursor mc2; - mc2.mc_txn = mc->mc_txn; - mc2.mc_dbi = MAIN_DBI; - rc = mdb_search_page(&mc2, &mc->mc_txn->mt_dbxs[mc->mc_dbi].md_name, 1); - if (rc) return rc; - mc->mc_txn->mt_dbxs[mc->mc_dbi].md_dirty = 1; + mdb_cursor_init(&mc2, mc->mc_txn, MAIN_DBI, NULL); + rc = mdb_page_search(&mc2, &mc->mc_dbx->md_name, 1); + if (rc) + return rc; + *mc->mc_dbflag = DB_DIRTY; } for (mc->mc_top = 0; mc->mc_top < mc->mc_snum; mc->mc_top++) { - if (!F_ISSET(mc->mc_pg[mc->mc_top]->mp_flags, P_DIRTY)) { - rc = mdb_touch(mc); - if (rc) return rc; - if (!mc->mc_top) { - mc->mc_txn->mt_dbs[mc->mc_dbi].md_root = - mc->mc_pg[mc->mc_top]->mp_pgno; - } - } + rc = mdb_page_touch(mc); + if (rc) + return rc; } mc->mc_top = mc->mc_snum-1; return MDB_SUCCESS; @@ -3175,20 +4131,24 @@ int mdb_cursor_put(MDB_cursor *mc, MDB_val *key, MDB_val *data, unsigned int flags) { - MDB_node *leaf; + MDB_node *leaf = NULL; MDB_val xdata, *rdata, dkey; + MDB_page *fp; MDB_db dummy; - char dbuf[PAGESIZE]; int do_sub = 0; + unsigned int mcount = 0; size_t nsize; - DKBUF; int rc, rc2; + MDB_pagebuf pbuf; + char dbuf[MAXKEYSIZE+1]; + unsigned int nflags; + DKBUF; if (F_ISSET(mc->mc_txn->mt_flags, MDB_TXN_RDONLY)) return EACCES; DPRINTF("==> put db %u key [%s], size %zu, data size %zu", - mc->mc_dbi, DKEY(key), key->mv_size, data->mv_size); + mc->mc_dbi, DKEY(key), key ? key->mv_size:0, data->mv_size); dkey.mv_size = 0; @@ -3196,19 +4156,19 @@ mdb_cursor_put(MDB_cursor *mc, MDB_val *key, MDB_val *data, if (!(mc->mc_flags & C_INITIALIZED)) return EINVAL; rc = MDB_SUCCESS; - } else if (mc->mc_txn->mt_dbs[mc->mc_dbi].md_root == P_INVALID) { + } else if (mc->mc_db->md_root == P_INVALID) { MDB_page *np; /* new database, write a root leaf page */ DPUTS("allocating new root leaf page"); - if ((np = mdb_new_page(mc, P_LEAF, 1)) == NULL) { + if ((np = mdb_page_new(mc, P_LEAF, 1)) == NULL) { return ENOMEM; } mc->mc_snum = 0; - cursor_push_page(mc, np); - mc->mc_txn->mt_dbs[mc->mc_dbi].md_root = np->mp_pgno; - mc->mc_txn->mt_dbs[mc->mc_dbi].md_depth++; - mc->mc_txn->mt_dbxs[mc->mc_dbi].md_dirty = 1; - if ((mc->mc_txn->mt_dbs[mc->mc_dbi].md_flags & (MDB_DUPSORT|MDB_DUPFIXED)) + mdb_cursor_push(mc, np); + mc->mc_db->md_root = np->mp_pgno; + mc->mc_db->md_depth++; + *mc->mc_dbflag = DB_DIRTY; + if ((mc->mc_db->md_flags & (MDB_DUPSORT|MDB_DUPFIXED)) == MDB_DUPFIXED) np->mp_flags |= P_LEAF2; mc->mc_flags |= C_INITIALIZED; @@ -3216,9 +4176,11 @@ mdb_cursor_put(MDB_cursor *mc, MDB_val *key, MDB_val *data, goto top; } else { int exact = 0; - rc = mdb_cursor_set(mc, key, NULL, MDB_SET, &exact); - if (flags == MDB_NOOVERWRITE && rc == 0) { + MDB_val d2; + rc = mdb_cursor_set(mc, key, &d2, MDB_SET, &exact); + if ((flags & MDB_NOOVERWRITE) && rc == 0) { DPRINTF("duplicate key [%s]", DKEY(key)); + *data = d2; return MDB_KEYEXIST; } if (rc && rc != MDB_NOTFOUND) @@ -3227,14 +4189,15 @@ mdb_cursor_put(MDB_cursor *mc, MDB_val *key, MDB_val *data, /* Cursor is positioned, now make sure all pages are writable */ rc2 = mdb_cursor_touch(mc); - if (rc2) return rc2; + if (rc2) + return rc2; top: /* The key already exists */ if (rc == MDB_SUCCESS) { /* there's only a key anyway, so this is a no-op */ if (IS_LEAF2(mc->mc_pg[mc->mc_top])) { - unsigned int ksize = mc->mc_txn->mt_dbs[mc->mc_dbi].md_pad; + unsigned int ksize = mc->mc_db->md_pad; if (key->mv_size != ksize) return EINVAL; if (flags == MDB_CURRENT) { @@ -3247,43 +4210,170 @@ top: leaf = NODEPTR(mc->mc_pg[mc->mc_top], mc->mc_ki[mc->mc_top]); /* DB has dups? */ - if (F_ISSET(mc->mc_txn->mt_dbs[mc->mc_dbi].md_flags, MDB_DUPSORT)) { + if (F_ISSET(mc->mc_db->md_flags, MDB_DUPSORT)) { /* Was a single item before, must convert now */ +more: if (!F_ISSET(leaf->mn_flags, F_DUPDATA)) { + /* Just overwrite the current item */ + if (flags == MDB_CURRENT) + goto current; + dkey.mv_size = NODEDSZ(leaf); - dkey.mv_data = dbuf; - memcpy(dbuf, NODEDATA(leaf), dkey.mv_size); - /* data matches, ignore it */ - if (!mdb_dcmp(mc->mc_txn, mc->mc_dbi, data, &dkey)) + dkey.mv_data = NODEDATA(leaf); +#if UINT_MAX < SIZE_MAX + if (mc->mc_dbx->md_dcmp == mdb_cmp_int && dkey.mv_size == sizeof(size_t)) +#ifdef MISALIGNED_OK + mc->mc_dbx->md_dcmp = mdb_cmp_long; +#else + mc->mc_dbx->md_dcmp = mdb_cmp_cint; +#endif +#endif + /* if data matches, ignore it */ + if (!mc->mc_dbx->md_dcmp(data, &dkey)) return (flags == MDB_NODUPDATA) ? MDB_KEYEXIST : MDB_SUCCESS; - memset(&dummy, 0, sizeof(dummy)); - if (mc->mc_txn->mt_dbs[mc->mc_dbi].md_flags & MDB_DUPFIXED) { - dummy.md_pad = data->mv_size; - dummy.md_flags = MDB_DUPFIXED; - if (mc->mc_txn->mt_dbs[mc->mc_dbi].md_flags & MDB_INTEGERDUP) - dummy.md_flags |= MDB_INTEGERKEY; + + /* create a fake page for the dup items */ + memcpy(dbuf, dkey.mv_data, dkey.mv_size); + dkey.mv_data = dbuf; + fp = (MDB_page *)&pbuf; + fp->mp_pgno = mc->mc_pg[mc->mc_top]->mp_pgno; + fp->mp_flags = P_LEAF|P_DIRTY|P_SUBP; + fp->mp_lower = PAGEHDRSZ; + fp->mp_upper = PAGEHDRSZ + dkey.mv_size + data->mv_size; + if (mc->mc_db->md_flags & MDB_DUPFIXED) { + fp->mp_flags |= P_LEAF2; + fp->mp_pad = data->mv_size; + } else { + fp->mp_upper += 2 * sizeof(indx_t) + 2 * NODESIZE + + (dkey.mv_size & 1) + (data->mv_size & 1); } - dummy.md_root = P_INVALID; - if (dkey.mv_size == sizeof(MDB_db)) { - memcpy(NODEDATA(leaf), &dummy, sizeof(dummy)); + mdb_node_del(mc->mc_pg[mc->mc_top], mc->mc_ki[mc->mc_top], 0); + do_sub = 1; + rdata = &xdata; + xdata.mv_size = fp->mp_upper; + xdata.mv_data = fp; + flags |= F_DUPDATA; + goto new_sub; + } + if (!F_ISSET(leaf->mn_flags, F_SUBDATA)) { + /* See if we need to convert from fake page to subDB */ + MDB_page *mp; + unsigned int offset; + unsigned int i; + + fp = NODEDATA(leaf); + if (flags == MDB_CURRENT) { + fp->mp_flags |= P_DIRTY; + COPY_PGNO(fp->mp_pgno, mc->mc_pg[mc->mc_top]->mp_pgno); + mc->mc_xcursor->mx_cursor.mc_pg[0] = fp; + flags |= F_DUPDATA; goto put_sub; } - mdb_del_node(mc->mc_pg[mc->mc_top], mc->mc_ki[mc->mc_top], 0); + if (mc->mc_db->md_flags & MDB_DUPFIXED) { + offset = fp->mp_pad; + } else { + offset = NODESIZE + sizeof(indx_t) + data->mv_size; + } + offset += offset & 1; + if (NODESIZE + sizeof(indx_t) + NODEKSZ(leaf) + NODEDSZ(leaf) + + offset >= (mc->mc_txn->mt_env->me_psize - PAGEHDRSZ) / + MDB_MINKEYS) { + /* yes, convert it */ + dummy.md_flags = 0; + if (mc->mc_db->md_flags & MDB_DUPFIXED) { + dummy.md_pad = fp->mp_pad; + dummy.md_flags = MDB_DUPFIXED; + if (mc->mc_db->md_flags & MDB_INTEGERDUP) + dummy.md_flags |= MDB_INTEGERKEY; + } + dummy.md_depth = 1; + dummy.md_branch_pages = 0; + dummy.md_leaf_pages = 1; + dummy.md_overflow_pages = 0; + dummy.md_entries = NUMKEYS(fp); + rdata = &xdata; + xdata.mv_size = sizeof(MDB_db); + xdata.mv_data = &dummy; + mp = mdb_page_alloc(mc, 1); + if (!mp) + return ENOMEM; + offset = mc->mc_txn->mt_env->me_psize - NODEDSZ(leaf); + flags |= F_DUPDATA|F_SUBDATA; + dummy.md_root = mp->mp_pgno; + } else { + /* no, just grow it */ + rdata = &xdata; + xdata.mv_size = NODEDSZ(leaf) + offset; + xdata.mv_data = &pbuf; + mp = (MDB_page *)&pbuf; + mp->mp_pgno = mc->mc_pg[mc->mc_top]->mp_pgno; + flags |= F_DUPDATA; + } + mp->mp_flags = fp->mp_flags | P_DIRTY; + mp->mp_pad = fp->mp_pad; + mp->mp_lower = fp->mp_lower; + mp->mp_upper = fp->mp_upper + offset; + if (IS_LEAF2(fp)) { + memcpy(METADATA(mp), METADATA(fp), NUMKEYS(fp) * fp->mp_pad); + } else { + nsize = NODEDSZ(leaf) - fp->mp_upper; + memcpy((char *)mp + mp->mp_upper, (char *)fp + fp->mp_upper, nsize); + for (i=0; imp_ptrs[i] = fp->mp_ptrs[i] + offset; + } + mdb_node_del(mc->mc_pg[mc->mc_top], mc->mc_ki[mc->mc_top], 0); do_sub = 1; - rdata = &xdata; - xdata.mv_size = sizeof(MDB_db); - xdata.mv_data = &dummy; goto new_sub; } + /* data is on sub-DB, just store it */ + flags |= F_DUPDATA|F_SUBDATA; goto put_sub; } - /* same size, just replace it */ - if (!F_ISSET(leaf->mn_flags, F_BIGDATA) && - NODEDSZ(leaf) == data->mv_size) { - memcpy(NODEDATA(leaf), data->mv_data, data->mv_size); +current: + /* overflow page overwrites need special handling */ + if (F_ISSET(leaf->mn_flags, F_BIGDATA)) { + MDB_page *omp; + pgno_t pg; + int ovpages, dpages; + + ovpages = OVPAGES(NODEDSZ(leaf), mc->mc_txn->mt_env->me_psize); + dpages = OVPAGES(data->mv_size, mc->mc_txn->mt_env->me_psize); + memcpy(&pg, NODEDATA(leaf), sizeof(pg)); + mdb_page_get(mc->mc_txn, pg, &omp); + /* Is the ov page writable and large enough? */ + if ((omp->mp_flags & P_DIRTY) && ovpages >= dpages) { + /* yes, overwrite it. Note in this case we don't + * bother to try shrinking the node if the new data + * is smaller than the overflow threshold. + */ + if (F_ISSET(flags, MDB_RESERVE)) + data->mv_data = METADATA(omp); + else + memcpy(METADATA(omp), data->mv_data, data->mv_size); + goto done; + } else { + /* no, free ovpages */ + int i; + mc->mc_db->md_overflow_pages -= ovpages; + for (i=0; imc_txn->mt_free_pgs, pg); + pg++; + } + } + } else if (NODEDSZ(leaf) == data->mv_size) { + /* same size, just replace it. Note that we could + * also reuse this node if the new data is smaller, + * but instead we opt to shrink the node in that case. + */ + if (F_ISSET(flags, MDB_RESERVE)) + data->mv_data = NODEDATA(leaf); + else + memcpy(NODEDATA(leaf), data->mv_data, data->mv_size); goto done; } - mdb_del_node(mc->mc_pg[mc->mc_top], mc->mc_ki[mc->mc_top], 0); + mdb_node_del(mc->mc_pg[mc->mc_top], mc->mc_ki[mc->mc_top], 0); + mc->mc_db->md_entries--; } else { DPRINTF("inserting key at index %i", mc->mc_ki[mc->mc_top]); } @@ -3291,52 +4381,97 @@ top: rdata = data; new_sub: + nflags = flags & NODE_ADD_FLAGS; nsize = IS_LEAF2(mc->mc_pg[mc->mc_top]) ? key->mv_size : mdb_leaf_size(mc->mc_txn->mt_env, key, rdata); if (SIZELEFT(mc->mc_pg[mc->mc_top]) < nsize) { - rc = mdb_split(mc, key, rdata, P_INVALID); + if (( flags & (F_DUPDATA|F_SUBDATA)) == F_DUPDATA ) + nflags &= ~MDB_APPEND; + rc = mdb_page_split(mc, key, rdata, P_INVALID, nflags); } else { /* There is room already in this leaf page. */ - rc = mdb_add_node(mc, mc->mc_ki[mc->mc_top], key, rdata, 0, 0); + rc = mdb_node_add(mc, mc->mc_ki[mc->mc_top], key, rdata, 0, nflags); + if (rc == 0 && !do_sub) { + /* Adjust other cursors pointing to mp */ + MDB_cursor *m2, *m3; + MDB_dbi dbi = mc->mc_dbi; + unsigned i = mc->mc_top; + MDB_page *mp = mc->mc_pg[i]; + + if (mc->mc_flags & C_SUB) + dbi--; + + for (m2 = mc->mc_txn->mt_cursors[dbi]; m2; m2=m2->mc_next) { + if (mc->mc_flags & C_SUB) + m3 = &m2->mc_xcursor->mx_cursor; + else + m3 = m2; + if (m3 == mc || m3->mc_snum < mc->mc_snum) continue; + if (m3->mc_pg[i] == mp && m3->mc_ki[i] >= mc->mc_ki[i]) { + m3->mc_ki[i]++; + } + } + } } if (rc != MDB_SUCCESS) mc->mc_txn->mt_flags |= MDB_TXN_ERROR; else { - /* Remember if we just added a subdatabase */ - if (flags & F_SUBDATA) { - leaf = NODEPTR(mc->mc_pg[mc->mc_top], mc->mc_ki[mc->mc_top]); - leaf->mn_flags |= F_SUBDATA; - } - /* Now store the actual data in the child DB. Note that we're * storing the user data in the keys field, so there are strict * size limits on dupdata. The actual data fields of the child * DB are all zero size. */ if (do_sub) { - leaf = NODEPTR(mc->mc_pg[mc->mc_top], mc->mc_ki[mc->mc_top]); + int xflags; put_sub: - if (flags == MDB_CURRENT) - mdb_xcursor_init2(mc); - else - mdb_xcursor_init1(mc, leaf); xdata.mv_size = 0; xdata.mv_data = ""; - if (flags == MDB_NODUPDATA) - flags = MDB_NOOVERWRITE; + leaf = NODEPTR(mc->mc_pg[mc->mc_top], mc->mc_ki[mc->mc_top]); + if (flags & MDB_CURRENT) { + xflags = MDB_CURRENT; + } else { + mdb_xcursor_init1(mc, leaf); + xflags = (flags & MDB_NODUPDATA) ? MDB_NOOVERWRITE : 0; + } /* converted, write the original data first */ if (dkey.mv_size) { - rc = mdb_cursor_put(&mc->mc_xcursor->mx_cursor, &dkey, &xdata, flags); - if (rc) return rc; - leaf->mn_flags |= F_DUPDATA; + rc = mdb_cursor_put(&mc->mc_xcursor->mx_cursor, &dkey, &xdata, xflags); + if (rc) + return rc; + { + /* Adjust other cursors pointing to mp */ + MDB_cursor *m2; + unsigned i = mc->mc_top; + MDB_page *mp = mc->mc_pg[i]; + + for (m2 = mc->mc_txn->mt_cursors[mc->mc_dbi]; m2; m2=m2->mc_next) { + if (m2 == mc || m2->mc_snum < mc->mc_snum) continue; + if (m2->mc_pg[i] == mp && m2->mc_ki[i] == mc->mc_ki[i]) { + mdb_xcursor_init1(m2, leaf); + } + } + } + } + xflags |= (flags & MDB_APPEND); + rc = mdb_cursor_put(&mc->mc_xcursor->mx_cursor, data, &xdata, xflags); + if (flags & F_SUBDATA) { + void *db = NODEDATA(leaf); + memcpy(db, &mc->mc_xcursor->mx_db, sizeof(MDB_db)); + } + } + /* sub-writes might have failed so check rc again. + * Don't increment count if we just replaced an existing item. + */ + if (!rc && !(flags & MDB_CURRENT)) + mc->mc_db->md_entries++; + if (flags & MDB_MULTIPLE) { + mcount++; + if (mcount < data[1].mv_size) { + data[0].mv_data = (char *)data[0].mv_data + data[0].mv_size; + leaf = NODEPTR(mc->mc_pg[mc->mc_top], mc->mc_ki[mc->mc_top]); + goto more; } - rc = mdb_cursor_put(&mc->mc_xcursor->mx_cursor, data, &xdata, flags); - mdb_xcursor_fini(mc); - memcpy(NODEDATA(leaf), - &mc->mc_xcursor->mx_txn.mt_dbs[mc->mc_xcursor->mx_cursor.mc_dbi], - sizeof(MDB_db)); } - mc->mc_txn->mt_dbs[mc->mc_dbi].md_entries++; } done: return rc; @@ -3355,95 +4490,96 @@ mdb_cursor_del(MDB_cursor *mc, unsigned int flags) return EINVAL; rc = mdb_cursor_touch(mc); - if (rc) return rc; + if (rc) + return rc; leaf = NODEPTR(mc->mc_pg[mc->mc_top], mc->mc_ki[mc->mc_top]); if (!IS_LEAF2(mc->mc_pg[mc->mc_top]) && F_ISSET(leaf->mn_flags, F_DUPDATA)) { if (flags != MDB_NODUPDATA) { - mdb_xcursor_init2(mc); + if (!F_ISSET(leaf->mn_flags, F_SUBDATA)) { + mc->mc_xcursor->mx_cursor.mc_pg[0] = NODEDATA(leaf); + } rc = mdb_cursor_del(&mc->mc_xcursor->mx_cursor, 0); - mdb_xcursor_fini(mc); /* If sub-DB still has entries, we're done */ - if (mc->mc_xcursor->mx_txn.mt_dbs[mc->mc_xcursor->mx_cursor.mc_dbi].md_root - != P_INVALID) { - memcpy(NODEDATA(leaf), - &mc->mc_xcursor->mx_txn.mt_dbs[mc->mc_xcursor->mx_cursor.mc_dbi], - sizeof(MDB_db)); - mc->mc_txn->mt_dbs[mc->mc_dbi].md_entries--; + if (mc->mc_xcursor->mx_db.md_entries) { + if (leaf->mn_flags & F_SUBDATA) { + /* update subDB info */ + void *db = NODEDATA(leaf); + memcpy(db, &mc->mc_xcursor->mx_db, sizeof(MDB_db)); + } else { + /* shrink fake page */ + mdb_node_shrink(mc->mc_pg[mc->mc_top], mc->mc_ki[mc->mc_top]); + } + mc->mc_db->md_entries--; return rc; } /* otherwise fall thru and delete the sub-DB */ } - /* add all the child DB's pages to the free list */ - rc = mdb_search_page(&mc->mc_xcursor->mx_cursor, NULL, 0); - if (rc == MDB_SUCCESS) { - MDB_node *ni; - MDB_cursor *mx; - unsigned int i; - - mx = &mc->mc_xcursor->mx_cursor; - mc->mc_txn->mt_dbs[mc->mc_dbi].md_entries -= - mx->mc_txn->mt_dbs[mx->mc_dbi].md_entries; - - cursor_pop_page(mx); - if (mx->mc_snum) { - while (mx->mc_snum > 1) { - for (i=0; imc_pg[mx->mc_top]); i++) { - pgno_t pg; - ni = NODEPTR(mx->mc_pg[mx->mc_top], i); - pg = NODEPGNO(ni); - /* free it */ - mdb_midl_append(mc->mc_txn->mt_free_pgs, pg); - } - rc = mdb_sibling(mx, 1); - if (rc) break; - } + if (leaf->mn_flags & F_SUBDATA) { + /* add all the child DB's pages to the free list */ + rc = mdb_drop0(&mc->mc_xcursor->mx_cursor, 0); + if (rc == MDB_SUCCESS) { + mc->mc_db->md_entries -= + mc->mc_xcursor->mx_db.md_entries; } - /* free it */ - mdb_midl_append(mc->mc_txn->mt_free_pgs, - mx->mc_txn->mt_dbs[mx->mc_dbi].md_root); } } - return mdb_del0(mc, leaf); + return mdb_cursor_del0(mc, leaf); } -/* Allocate a page and initialize it +/** Allocate and initialize new pages for a database. + * @param[in] mc a cursor on the database being added to. + * @param[in] flags flags defining what type of page is being allocated. + * @param[in] num the number of pages to allocate. This is usually 1, + * unless allocating overflow pages for a large record. + * @return Address of a page, or NULL on failure. */ static MDB_page * -mdb_new_page(MDB_cursor *mc, uint32_t flags, int num) +mdb_page_new(MDB_cursor *mc, uint32_t flags, int num) { MDB_page *np; - if ((np = mdb_alloc_page(mc, num)) == NULL) + if ((np = mdb_page_alloc(mc, num)) == NULL) return NULL; - DPRINTF("allocated new mpage %lu, page size %u", + DPRINTF("allocated new mpage %zu, page size %u", np->mp_pgno, mc->mc_txn->mt_env->me_psize); np->mp_flags = flags | P_DIRTY; np->mp_lower = PAGEHDRSZ; np->mp_upper = mc->mc_txn->mt_env->me_psize; if (IS_BRANCH(np)) - mc->mc_txn->mt_dbs[mc->mc_dbi].md_branch_pages++; + mc->mc_db->md_branch_pages++; else if (IS_LEAF(np)) - mc->mc_txn->mt_dbs[mc->mc_dbi].md_leaf_pages++; + mc->mc_db->md_leaf_pages++; else if (IS_OVERFLOW(np)) { - mc->mc_txn->mt_dbs[mc->mc_dbi].md_overflow_pages += num; + mc->mc_db->md_overflow_pages += num; np->mp_pages = num; } return np; } +/** Calculate the size of a leaf node. + * The size depends on the environment's page size; if a data item + * is too large it will be put onto an overflow page and the node + * size will only include the key and not the data. Sizes are always + * rounded up to an even number of bytes, to guarantee 2-byte alignment + * of the #MDB_node headers. + * @param[in] env The environment handle. + * @param[in] key The key for the node. + * @param[in] data The data for the node. + * @return The number of bytes needed to store the node. + */ static size_t mdb_leaf_size(MDB_env *env, MDB_val *key, MDB_val *data) { size_t sz; sz = LEAFSIZE(key, data); - if (data->mv_size >= env->me_psize / MDB_MINKEYS) { + if (sz >= env->me_psize / MDB_MINKEYS) { /* put on overflow page */ sz -= data->mv_size - sizeof(pgno_t); } @@ -3452,6 +4588,16 @@ mdb_leaf_size(MDB_env *env, MDB_val *key, MDB_val *data) return sz + sizeof(indx_t); } +/** Calculate the size of a branch node. + * The size should depend on the environment's page size but since + * we currently don't support spilling large keys onto overflow + * pages, it's simply the size of the #MDB_node header plus the + * size of the key. Sizes are always rounded up to an even number + * of bytes, to guarantee 2-byte alignment of the #MDB_node headers. + * @param[in] env The environment handle. + * @param[in] key The key for the node. + * @return The number of bytes needed to store the node. + */ static size_t mdb_branch_size(MDB_env *env, MDB_val *key) { @@ -3467,9 +4613,24 @@ mdb_branch_size(MDB_env *env, MDB_val *key) return sz + sizeof(indx_t); } +/** Add a node to the page pointed to by the cursor. + * @param[in] mc The cursor for this operation. + * @param[in] indx The index on the page where the new node should be added. + * @param[in] key The key for the new node. + * @param[in] data The data for the new node, if any. + * @param[in] pgno The page number, if adding a branch node. + * @param[in] flags Flags for the node. + * @return 0 on success, non-zero on failure. Possible errors are: + *
    + *
  • ENOMEM - failed to allocate overflow pages for the node. + *
  • ENOSPC - there is insufficient room in the page. This error + * should never happen since all callers already calculate the + * page's free space before calling this function. + *
+ */ static int -mdb_add_node(MDB_cursor *mc, indx_t indx, - MDB_val *key, MDB_val *data, pgno_t pgno, uint8_t flags) +mdb_node_add(MDB_cursor *mc, indx_t indx, + MDB_val *key, MDB_val *data, pgno_t pgno, unsigned int flags) { unsigned int i; size_t node_size = NODESIZE; @@ -3481,14 +4642,15 @@ mdb_add_node(MDB_cursor *mc, indx_t indx, assert(mp->mp_upper >= mp->mp_lower); - DPRINTF("add to %s page %lu index %i, data size %zu key size %zu [%s]", + DPRINTF("add to %s %spage %zu index %i, data size %zu key size %zu [%s]", IS_LEAF(mp) ? "leaf" : "branch", + IS_SUBP(mp) ? "sub-" : "", mp->mp_pgno, indx, data ? data->mv_size : 0, key ? key->mv_size : 0, key ? DKEY(key) : NULL); if (IS_LEAF2(mp)) { /* Move higher keys up one slot. */ - int ksize = mc->mc_txn->mt_dbs[mc->mc_dbi].md_pad, dif; + int ksize = mc->mc_db->md_pad, dif; char *ptr = LEAF2KEY(mp, indx, ksize); dif = NUMKEYS(mp) - indx; if (dif > 0) @@ -3510,15 +4672,15 @@ mdb_add_node(MDB_cursor *mc, indx_t indx, if (F_ISSET(flags, F_BIGDATA)) { /* Data already on overflow page. */ node_size += sizeof(pgno_t); - } else if (data->mv_size >= mc->mc_txn->mt_env->me_psize / MDB_MINKEYS) { + } else if (node_size + data->mv_size >= mc->mc_txn->mt_env->me_psize / MDB_MINKEYS) { int ovpages = OVPAGES(data->mv_size, mc->mc_txn->mt_env->me_psize); /* Put data on overflow page. */ - DPRINTF("data size is %zu, put on overflow page", - data->mv_size); + DPRINTF("data size is %zu, node would be %zu, put data on overflow page", + data->mv_size, node_size+data->mv_size); node_size += sizeof(pgno_t); - if ((ofp = mdb_new_page(mc, P_OVERFLOW, ovpages)) == NULL) + if ((ofp = mdb_page_new(mc, P_OVERFLOW, ovpages)) == NULL) return ENOMEM; - DPRINTF("allocated overflow page %lu", ofp->mp_pgno); + DPRINTF("allocated overflow page %zu", ofp->mp_pgno); flags |= F_BIGDATA; } else { node_size += data->mv_size; @@ -3527,7 +4689,7 @@ mdb_add_node(MDB_cursor *mc, indx_t indx, node_size += node_size & 1; if (node_size + sizeof(indx_t) > SIZELEFT(mp)) { - DPRINTF("not enough room in page %lu, got %u ptrs", + DPRINTF("not enough room in page %zu, got %u ptrs", mp->mp_pgno, NUMKEYS(mp)); DPRINTF("upper - lower = %u - %u = %u", mp->mp_upper, mp->mp_lower, mp->mp_upper - mp->mp_lower); @@ -3564,29 +4726,46 @@ mdb_add_node(MDB_cursor *mc, indx_t indx, if (F_ISSET(flags, F_BIGDATA)) memcpy(node->mn_data + key->mv_size, data->mv_data, sizeof(pgno_t)); + else if (F_ISSET(flags, MDB_RESERVE)) + data->mv_data = node->mn_data + key->mv_size; else memcpy(node->mn_data + key->mv_size, data->mv_data, data->mv_size); } else { memcpy(node->mn_data + key->mv_size, &ofp->mp_pgno, sizeof(pgno_t)); - memcpy(METADATA(ofp), data->mv_data, data->mv_size); + if (F_ISSET(flags, MDB_RESERVE)) + data->mv_data = METADATA(ofp); + else + memcpy(METADATA(ofp), data->mv_data, data->mv_size); } } return MDB_SUCCESS; } +/** Delete the specified node from a page. + * @param[in] mp The page to operate on. + * @param[in] indx The index of the node to delete. + * @param[in] ksize The size of a node. Only used if the page is + * part of a #MDB_DUPFIXED database. + */ static void -mdb_del_node(MDB_page *mp, indx_t indx, int ksize) +mdb_node_del(MDB_page *mp, indx_t indx, int ksize) { unsigned int sz; indx_t i, j, numkeys, ptr; MDB_node *node; char *base; - DPRINTF("delete node %u on %s page %lu", indx, - IS_LEAF(mp) ? "leaf" : "branch", mp->mp_pgno); +#if MDB_DEBUG + { + pgno_t pgno; + COPY_PGNO(pgno, mp->mp_pgno); + DPRINTF("delete node %u on %s page %zu", indx, + IS_LEAF(mp) ? "leaf" : "branch", pgno); + } +#endif assert(indx < NUMKEYS(mp)); if (IS_LEAF2(mp)) { @@ -3627,90 +4806,158 @@ mdb_del_node(MDB_page *mp, indx_t indx, int ksize) mp->mp_upper += sz; } +/** Compact the main page after deleting a node on a subpage. + * @param[in] mp The main page to operate on. + * @param[in] indx The index of the subpage on the main page. + */ static void -mdb_xcursor_init0(MDB_cursor *mc) +mdb_node_shrink(MDB_page *mp, indx_t indx) { - MDB_xcursor *mx = mc->mc_xcursor; - MDB_dbi dbn; - - mx->mx_txn = *mc->mc_txn; - mx->mx_txn.mt_dbxs = mx->mx_dbxs; - mx->mx_txn.mt_dbs = mx->mx_dbs; - mx->mx_dbxs[0] = mc->mc_txn->mt_dbxs[0]; - mx->mx_dbxs[1] = mc->mc_txn->mt_dbxs[1]; - if (mc->mc_dbi > 1) { - mx->mx_dbxs[2] = mc->mc_txn->mt_dbxs[mc->mc_dbi]; - dbn = 2; + MDB_node *node; + MDB_page *sp, *xp; + char *base; + int osize, nsize; + int delta; + indx_t i, numkeys, ptr; + + node = NODEPTR(mp, indx); + sp = (MDB_page *)NODEDATA(node); + osize = NODEDSZ(node); + + delta = sp->mp_upper - sp->mp_lower; + SETDSZ(node, osize - delta); + xp = (MDB_page *)((char *)sp + delta); + + /* shift subpage upward */ + if (IS_LEAF2(sp)) { + nsize = NUMKEYS(sp) * sp->mp_pad; + memmove(METADATA(xp), METADATA(sp), nsize); } else { - dbn = 1; + int i; + nsize = osize - sp->mp_upper; + numkeys = NUMKEYS(sp); + for (i=numkeys-1; i>=0; i--) + xp->mp_ptrs[i] = sp->mp_ptrs[i] - delta; + } + 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); + + /* shift lower nodes upward */ + ptr = mp->mp_ptrs[indx]; + numkeys = NUMKEYS(mp); + for (i = 0; i < numkeys; i++) { + if (mp->mp_ptrs[i] <= ptr) + mp->mp_ptrs[i] += delta; } - mx->mx_dbxs[dbn+1].md_parent = dbn; - mx->mx_dbxs[dbn+1].md_cmp = mx->mx_dbxs[dbn].md_dcmp; - mx->mx_dbxs[dbn+1].md_rel = mx->mx_dbxs[dbn].md_rel; - mx->mx_dbxs[dbn+1].md_dirty = 0; - mx->mx_txn.mt_numdbs = dbn+2; - mx->mx_txn.mt_u = mc->mc_txn->mt_u; - mx->mx_cursor.mc_xcursor = NULL; - mx->mx_cursor.mc_txn = &mx->mx_txn; - mx->mx_cursor.mc_dbi = dbn+1; + base = (char *)mp + mp->mp_upper; + memmove(base + delta, base, ptr - mp->mp_upper + NODESIZE + NODEKSZ(node)); + mp->mp_upper += delta; } +/** Initial setup of a sorted-dups cursor. + * Sorted duplicates are implemented as a sub-database for the given key. + * The duplicate data items are actually keys of the sub-database. + * Operations on the duplicate data items are performed using a sub-cursor + * initialized when the sub-database is first accessed. This function does + * the preliminary setup of the sub-cursor, filling in the fields that + * depend only on the parent DB. + * @param[in] mc The main cursor whose sorted-dups cursor is to be initialized. + */ static void -mdb_xcursor_init1(MDB_cursor *mc, MDB_node *node) +mdb_xcursor_init0(MDB_cursor *mc) { - MDB_db *db = NODEDATA(node); MDB_xcursor *mx = mc->mc_xcursor; - MDB_dbi dbn; - mx->mx_dbs[0] = mc->mc_txn->mt_dbs[0]; - mx->mx_dbs[1] = mc->mc_txn->mt_dbs[1]; - if (mc->mc_dbi > 1) { - mx->mx_dbs[2] = mc->mc_txn->mt_dbs[mc->mc_dbi]; - mx->mx_dbxs[2].md_dirty = mc->mc_txn->mt_dbxs[mc->mc_dbi].md_dirty; - dbn = 3; - } else { - dbn = 2; - } - DPRINTF("Sub-db %u for db %u root page %lu", dbn, mc->mc_dbi, db->md_root); - mx->mx_dbs[dbn] = *db; - if (F_ISSET(mc->mc_pg[mc->mc_top]->mp_flags, P_DIRTY)) - mx->mx_dbxs[dbn].md_dirty = 1; - mx->mx_dbxs[dbn].md_name.mv_data = NODEKEY(node); - mx->mx_dbxs[dbn].md_name.mv_size = node->mn_ksize; - mx->mx_txn.mt_next_pgno = mc->mc_txn->mt_next_pgno; + + mx->mx_cursor.mc_xcursor = NULL; + mx->mx_cursor.mc_txn = mc->mc_txn; + mx->mx_cursor.mc_db = &mx->mx_db; + mx->mx_cursor.mc_dbx = &mx->mx_dbx; + mx->mx_cursor.mc_dbi = mc->mc_dbi+1; + mx->mx_cursor.mc_dbflag = &mx->mx_dbflag; mx->mx_cursor.mc_snum = 0; - mx->mx_cursor.mc_flags = 0; + mx->mx_cursor.mc_top = 0; + mx->mx_cursor.mc_flags = C_SUB; + mx->mx_dbx.md_cmp = mc->mc_dbx->md_dcmp; + mx->mx_dbx.md_dcmp = NULL; + mx->mx_dbx.md_rel = mc->mc_dbx->md_rel; } +/** Final setup of a sorted-dups cursor. + * Sets up the fields that depend on the data from the main cursor. + * @param[in] mc The main cursor whose sorted-dups cursor is to be initialized. + * @param[in] node The data containing the #MDB_db record for the + * sorted-dup database. + */ static void -mdb_xcursor_init2(MDB_cursor *mc) +mdb_xcursor_init1(MDB_cursor *mc, MDB_node *node) { MDB_xcursor *mx = mc->mc_xcursor; - MDB_dbi dbn; - mx->mx_dbs[0] = mc->mc_txn->mt_dbs[0]; - mx->mx_dbs[1] = mc->mc_txn->mt_dbs[1]; - if (mc->mc_dbi > 1) { - mx->mx_dbs[2] = mc->mc_txn->mt_dbs[mc->mc_dbi]; - mx->mx_dbxs[2].md_dirty = mc->mc_txn->mt_dbxs[mc->mc_dbi].md_dirty; - dbn = 3; + + if (node->mn_flags & F_SUBDATA) { + memcpy(&mx->mx_db, NODEDATA(node), sizeof(MDB_db)); + mx->mx_cursor.mc_snum = 0; + mx->mx_cursor.mc_flags = C_SUB; } else { - dbn = 2; + MDB_page *fp = NODEDATA(node); + mx->mx_db.md_pad = mc->mc_pg[mc->mc_top]->mp_pad; + mx->mx_db.md_flags = 0; + mx->mx_db.md_depth = 1; + mx->mx_db.md_branch_pages = 0; + mx->mx_db.md_leaf_pages = 1; + mx->mx_db.md_overflow_pages = 0; + mx->mx_db.md_entries = NUMKEYS(fp); + COPY_PGNO(mx->mx_db.md_root, fp->mp_pgno); + mx->mx_cursor.mc_snum = 1; + mx->mx_cursor.mc_flags = C_INITIALIZED|C_SUB; + mx->mx_cursor.mc_top = 0; + mx->mx_cursor.mc_pg[0] = fp; + mx->mx_cursor.mc_ki[0] = 0; + if (mc->mc_db->md_flags & MDB_DUPFIXED) { + mx->mx_db.md_flags = MDB_DUPFIXED; + mx->mx_db.md_pad = fp->mp_pad; + if (mc->mc_db->md_flags & MDB_INTEGERDUP) + mx->mx_db.md_flags |= MDB_INTEGERKEY; + } } - DPRINTF("Sub-db %u for db %u root page %lu", dbn, mc->mc_dbi, - mx->mx_dbs[dbn].md_root); - mx->mx_txn.mt_next_pgno = mc->mc_txn->mt_next_pgno; + DPRINTF("Sub-db %u for db %u root page %zu", mx->mx_cursor.mc_dbi, mc->mc_dbi, + mx->mx_db.md_root); + mx->mx_dbflag = (F_ISSET(mc->mc_pg[mc->mc_top]->mp_flags, P_DIRTY)) ? + DB_DIRTY : 0; + mx->mx_dbx.md_name.mv_data = NODEKEY(node); + mx->mx_dbx.md_name.mv_size = node->mn_ksize; +#if UINT_MAX < SIZE_MAX + if (mx->mx_dbx.md_cmp == mdb_cmp_int && mx->mx_db.md_pad == sizeof(size_t)) +#ifdef MISALIGNED_OK + mx->mx_dbx.md_cmp = mdb_cmp_long; +#else + mx->mx_dbx.md_cmp = mdb_cmp_cint; +#endif +#endif } +/** Initialize a cursor for a given transaction and database. */ static void -mdb_xcursor_fini(MDB_cursor *mc) +mdb_cursor_init(MDB_cursor *mc, MDB_txn *txn, MDB_dbi dbi, MDB_xcursor *mx) { - MDB_xcursor *mx = mc->mc_xcursor; - mc->mc_txn->mt_next_pgno = mx->mx_txn.mt_next_pgno; - mc->mc_txn->mt_dbs[0] = mx->mx_dbs[0]; - mc->mc_txn->mt_dbs[1] = mx->mx_dbs[1]; - if (mc->mc_dbi > 1) { - mc->mc_txn->mt_dbs[mc->mc_dbi] = mx->mx_dbs[2]; - mc->mc_txn->mt_dbxs[mc->mc_dbi].md_dirty = mx->mx_dbxs[2].md_dirty; + mc->mc_orig = NULL; + mc->mc_dbi = dbi; + mc->mc_txn = txn; + mc->mc_db = &txn->mt_dbs[dbi]; + mc->mc_dbx = &txn->mt_dbxs[dbi]; + mc->mc_dbflag = &txn->mt_dbflags[dbi]; + mc->mc_snum = 0; + mc->mc_top = 0; + mc->mc_flags = 0; + if (txn->mt_dbs[dbi].md_flags & MDB_DUPSORT) { + assert(mx != NULL); + mc->mc_xcursor = mx; + mdb_xcursor_init0(mc); + } else { + mc->mc_xcursor = NULL; } } @@ -3718,22 +4965,29 @@ int mdb_cursor_open(MDB_txn *txn, MDB_dbi dbi, MDB_cursor **ret) { MDB_cursor *mc; + MDB_xcursor *mx = NULL; size_t size = sizeof(MDB_cursor); - if (txn == NULL || ret == NULL || !dbi || dbi >= txn->mt_numdbs) + if (txn == NULL || ret == NULL || dbi >= txn->mt_numdbs) + return EINVAL; + + /* Allow read access to the freelist */ + if (!dbi && !F_ISSET(txn->mt_flags, MDB_TXN_RDONLY)) return EINVAL; if (txn->mt_dbs[dbi].md_flags & MDB_DUPSORT) size += sizeof(MDB_xcursor); - if ((mc = calloc(1, size)) != NULL) { - mc->mc_dbi = dbi; - mc->mc_txn = txn; + if ((mc = malloc(size)) != NULL) { if (txn->mt_dbs[dbi].md_flags & MDB_DUPSORT) { - MDB_xcursor *mx = (MDB_xcursor *)(mc + 1); - mc->mc_xcursor = mx; - mdb_xcursor_init0(mc); + mx = (MDB_xcursor *)(mc + 1); + } + mdb_cursor_init(mc, txn, dbi, mx); + if (txn->mt_cursors) { + mc->mc_next = txn->mt_cursors[dbi]; + txn->mt_cursors[dbi] = mc; } + mc->mc_flags |= C_ALLOCD; } else { return ENOMEM; } @@ -3745,14 +4999,14 @@ mdb_cursor_open(MDB_txn *txn, MDB_dbi dbi, MDB_cursor **ret) /* Return the count of duplicate data items for the current key */ int -mdb_cursor_count(MDB_cursor *mc, unsigned long *countp) +mdb_cursor_count(MDB_cursor *mc, size_t *countp) { MDB_node *leaf; if (mc == NULL || countp == NULL) return EINVAL; - if (!(mc->mc_txn->mt_dbs[mc->mc_dbi].md_flags & MDB_DUPSORT)) + if (!(mc->mc_db->md_flags & MDB_DUPSORT)) return EINVAL; leaf = NODEPTR(mc->mc_pg[mc->mc_top], mc->mc_ki[mc->mc_top]); @@ -3762,7 +5016,7 @@ mdb_cursor_count(MDB_cursor *mc, unsigned long *countp) if (!(mc->mc_xcursor->mx_cursor.mc_flags & C_INITIALIZED)) return EINVAL; - *countp = mc->mc_xcursor->mx_txn.mt_dbs[mc->mc_xcursor->mx_cursor.mc_dbi].md_entries; + *countp = mc->mc_xcursor->mx_db.md_entries; } return MDB_SUCCESS; } @@ -3771,29 +5025,70 @@ void mdb_cursor_close(MDB_cursor *mc) { if (mc != NULL) { - free(mc); + /* remove from txn, if tracked */ + if (mc->mc_txn->mt_cursors) { + MDB_cursor **prev = &mc->mc_txn->mt_cursors[mc->mc_dbi]; + while (*prev && *prev != mc) prev = &(*prev)->mc_next; + if (*prev == mc) + *prev = mc->mc_next; + } + if (mc->mc_flags & C_ALLOCD) + free(mc); } } +MDB_txn * +mdb_cursor_txn(MDB_cursor *mc) +{ + if (!mc) return NULL; + return mc->mc_txn; +} + +MDB_dbi +mdb_cursor_dbi(MDB_cursor *mc) +{ + if (!mc) return 0; + return mc->mc_dbi; +} + +/** Replace the key for a node with a new key. + * @param[in] mp The page containing the node to operate on. + * @param[in] indx The index of the node to operate on. + * @param[in] key The new key to use. + * @return 0 on success, non-zero on failure. + */ static int mdb_update_key(MDB_page *mp, indx_t indx, MDB_val *key) { - indx_t ptr, i, numkeys; - int delta; - size_t len; MDB_node *node; char *base; + size_t len; + int delta, delta0; + indx_t ptr, i, numkeys; DKBUF; node = NODEPTR(mp, indx); ptr = mp->mp_ptrs[indx]; - DPRINTF("update key %u (ofs %u) [%.*s] to [%s] on page %lu", - indx, ptr, - (int)node->mn_ksize, (char *)NODEKEY(node), - DKEY(key), - mp->mp_pgno); +#if MDB_DEBUG + { + MDB_val k2; + char kbuf2[(MAXKEYSIZE*2+1)]; + k2.mv_data = NODEKEY(node); + k2.mv_size = node->mn_ksize; + DPRINTF("update key %u (ofs %u) [%s] to [%s] on page %zu", + indx, ptr, + mdb_dkey(&k2, kbuf2), + DKEY(key), + mp->mp_pgno); + } +#endif - delta = key->mv_size - node->mn_ksize; + delta0 = delta = key->mv_size - node->mn_ksize; + + /* Must be 2-byte aligned. If new key is + * shorter by 1, the shift will be skipped. + */ + delta += (delta & 1); if (delta) { if (delta > 0 && SIZELEFT(mp) < delta) { DPRINTF("OUCH! Not enough room, delta = %d", delta); @@ -3812,52 +5107,81 @@ mdb_update_key(MDB_page *mp, indx_t indx, MDB_val *key) mp->mp_upper -= delta; node = NODEPTR(mp, indx); - node->mn_ksize = key->mv_size; } - memcpy(NODEKEY(node), key->mv_data, key->mv_size); + /* But even if no shift was needed, update ksize */ + if (delta0) + node->mn_ksize = key->mv_size; + + if (key->mv_size) + memcpy(NODEKEY(node), key->mv_data, key->mv_size); return MDB_SUCCESS; } -/* Move a node from csrc to cdst. +/** Move a node from csrc to cdst. */ static int -mdb_move_node(MDB_cursor *csrc, MDB_cursor *cdst) +mdb_node_move(MDB_cursor *csrc, MDB_cursor *cdst) { int rc; MDB_node *srcnode; MDB_val key, data; + pgno_t srcpg; + unsigned short flags; + DKBUF; /* Mark src and dst as dirty. */ - if ((rc = mdb_touch(csrc)) || - (rc = mdb_touch(cdst))) - return rc;; + if ((rc = mdb_page_touch(csrc)) || + (rc = mdb_page_touch(cdst))) + return rc; if (IS_LEAF2(csrc->mc_pg[csrc->mc_top])) { srcnode = NODEPTR(csrc->mc_pg[csrc->mc_top], 0); /* fake */ - key.mv_size = csrc->mc_txn->mt_dbs[csrc->mc_dbi].md_pad; + key.mv_size = csrc->mc_db->md_pad; key.mv_data = LEAF2KEY(csrc->mc_pg[csrc->mc_top], csrc->mc_ki[csrc->mc_top], key.mv_size); data.mv_size = 0; data.mv_data = NULL; + srcpg = 0; + flags = 0; } else { + srcnode = NODEPTR(csrc->mc_pg[csrc->mc_top], csrc->mc_ki[csrc->mc_top]); + assert(!((long)srcnode&1)); + srcpg = NODEPGNO(srcnode); + flags = srcnode->mn_flags; if (csrc->mc_ki[csrc->mc_top] == 0 && IS_BRANCH(csrc->mc_pg[csrc->mc_top])) { unsigned int snum = csrc->mc_snum; + MDB_node *s2; /* must find the lowest key below src */ - mdb_search_page_root(csrc, NULL, 0); - srcnode = NODEPTR(csrc->mc_pg[csrc->mc_top], 0); + mdb_page_search_root(csrc, NULL, 0); + s2 = NODEPTR(csrc->mc_pg[csrc->mc_top], 0); + key.mv_size = NODEKSZ(s2); + key.mv_data = NODEKEY(s2); csrc->mc_snum = snum--; csrc->mc_top = snum; } else { - srcnode = NODEPTR(csrc->mc_pg[csrc->mc_top], csrc->mc_ki[csrc->mc_top]); + key.mv_size = NODEKSZ(srcnode); + key.mv_data = NODEKEY(srcnode); } - key.mv_size = NODEKSZ(srcnode); - key.mv_data = NODEKEY(srcnode); data.mv_size = NODEDSZ(srcnode); data.mv_data = NODEDATA(srcnode); } - DPRINTF("moving %s node %u [%s] on page %lu to node %u on page %lu", + if (IS_BRANCH(cdst->mc_pg[cdst->mc_top]) && cdst->mc_ki[cdst->mc_top] == 0) { + unsigned int snum = cdst->mc_snum; + MDB_node *s2; + MDB_val bkey; + /* must find the lowest key below dst */ + mdb_page_search_root(cdst, NULL, 0); + s2 = NODEPTR(cdst->mc_pg[cdst->mc_top], 0); + bkey.mv_size = NODEKSZ(s2); + bkey.mv_data = NODEKEY(s2); + cdst->mc_snum = snum--; + cdst->mc_top = snum; + rc = mdb_update_key(cdst->mc_pg[cdst->mc_top], 0, &bkey); + } + + DPRINTF("moving %s node %u [%s] on page %zu to node %u on page %zu", IS_LEAF(csrc->mc_pg[csrc->mc_top]) ? "leaf" : "branch", csrc->mc_ki[csrc->mc_top], DKEY(&key), @@ -3866,27 +5190,49 @@ mdb_move_node(MDB_cursor *csrc, MDB_cursor *cdst) /* Add the node to the destination page. */ - rc = mdb_add_node(cdst, cdst->mc_ki[cdst->mc_top], &key, &data, NODEPGNO(srcnode), - srcnode->mn_flags); + rc = mdb_node_add(cdst, cdst->mc_ki[cdst->mc_top], &key, &data, srcpg, flags); if (rc != MDB_SUCCESS) return rc; /* Delete the node from the source page. */ - mdb_del_node(csrc->mc_pg[csrc->mc_top], csrc->mc_ki[csrc->mc_top], key.mv_size); + mdb_node_del(csrc->mc_pg[csrc->mc_top], csrc->mc_ki[csrc->mc_top], key.mv_size); + + { + /* Adjust other cursors pointing to mp */ + MDB_cursor *m2, *m3; + MDB_dbi dbi = csrc->mc_dbi; + MDB_page *mp = csrc->mc_pg[csrc->mc_top]; + + if (csrc->mc_flags & C_SUB) + dbi--; + + for (m2 = csrc->mc_txn->mt_cursors[dbi]; m2; m2=m2->mc_next) { + if (m2 == csrc) continue; + if (csrc->mc_flags & C_SUB) + m3 = &m2->mc_xcursor->mx_cursor; + else + m3 = m2; + 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]; + } + } + } /* Update the parent separators. */ if (csrc->mc_ki[csrc->mc_top] == 0) { if (csrc->mc_ki[csrc->mc_top-1] != 0) { if (IS_LEAF2(csrc->mc_pg[csrc->mc_top])) { - key.mv_data = LEAF2KEY(csrc->mc_pg[csrc->mc_top], csrc->mc_ki[csrc->mc_top], key.mv_size); + key.mv_data = LEAF2KEY(csrc->mc_pg[csrc->mc_top], 0, key.mv_size); } else { - srcnode = NODEPTR(csrc->mc_pg[csrc->mc_top], csrc->mc_ki[csrc->mc_top]); + srcnode = NODEPTR(csrc->mc_pg[csrc->mc_top], 0); key.mv_size = NODEKSZ(srcnode); key.mv_data = NODEKEY(srcnode); } - DPRINTF("update separator for source page %lu to [%s]", + DPRINTF("update separator for source page %zu to [%s]", csrc->mc_pg[csrc->mc_top]->mp_pgno, DKEY(&key)); if ((rc = mdb_update_key(csrc->mc_pg[csrc->mc_top-1], csrc->mc_ki[csrc->mc_top-1], &key)) != MDB_SUCCESS) @@ -3895,7 +5241,8 @@ mdb_move_node(MDB_cursor *csrc, MDB_cursor *cdst) if (IS_BRANCH(csrc->mc_pg[csrc->mc_top])) { MDB_val nullkey; nullkey.mv_size = 0; - assert(mdb_update_key(csrc->mc_pg[csrc->mc_top], 0, &nullkey) == MDB_SUCCESS); + rc = mdb_update_key(csrc->mc_pg[csrc->mc_top], 0, &nullkey); + assert(rc == MDB_SUCCESS); } } @@ -3908,7 +5255,7 @@ mdb_move_node(MDB_cursor *csrc, MDB_cursor *cdst) key.mv_size = NODEKSZ(srcnode); key.mv_data = NODEKEY(srcnode); } - DPRINTF("update separator for destination page %lu to [%s]", + DPRINTF("update separator for destination page %zu to [%s]", cdst->mc_pg[cdst->mc_top]->mp_pgno, DKEY(&key)); if ((rc = mdb_update_key(cdst->mc_pg[cdst->mc_top-1], cdst->mc_ki[cdst->mc_top-1], &key)) != MDB_SUCCESS) @@ -3917,38 +5264,48 @@ mdb_move_node(MDB_cursor *csrc, MDB_cursor *cdst) if (IS_BRANCH(cdst->mc_pg[cdst->mc_top])) { MDB_val nullkey; nullkey.mv_size = 0; - assert(mdb_update_key(cdst->mc_pg[cdst->mc_top], 0, &nullkey) == MDB_SUCCESS); + rc = mdb_update_key(cdst->mc_pg[cdst->mc_top], 0, &nullkey); + assert(rc == MDB_SUCCESS); } } return MDB_SUCCESS; } +/** Merge one page into another. + * The nodes from the page pointed to by \b csrc will + * be copied to the page pointed to by \b cdst and then + * the \b csrc page will be freed. + * @param[in] csrc Cursor pointing to the source page. + * @param[in] cdst Cursor pointing to the destination page. + */ static int -mdb_merge(MDB_cursor *csrc, MDB_cursor *cdst) +mdb_page_merge(MDB_cursor *csrc, MDB_cursor *cdst) { int rc; indx_t i, j; MDB_node *srcnode; MDB_val key, data; + unsigned nkeys; - DPRINTF("merging page %lu into %lu", csrc->mc_pg[csrc->mc_top]->mp_pgno, cdst->mc_pg[cdst->mc_top]->mp_pgno); + DPRINTF("merging page %zu into %zu", csrc->mc_pg[csrc->mc_top]->mp_pgno, + cdst->mc_pg[cdst->mc_top]->mp_pgno); assert(csrc->mc_snum > 1); /* can't merge root page */ assert(cdst->mc_snum > 1); /* Mark dst as dirty. */ - if ((rc = mdb_touch(cdst))) + if ((rc = mdb_page_touch(cdst))) return rc; /* Move all nodes from src to dst. */ - j = NUMKEYS(cdst->mc_pg[cdst->mc_top]); + j = nkeys = NUMKEYS(cdst->mc_pg[cdst->mc_top]); if (IS_LEAF2(csrc->mc_pg[csrc->mc_top])) { - key.mv_size = csrc->mc_txn->mt_dbs[csrc->mc_dbi].md_pad; + key.mv_size = csrc->mc_db->md_pad; key.mv_data = METADATA(csrc->mc_pg[csrc->mc_top]); for (i = 0; i < NUMKEYS(csrc->mc_pg[csrc->mc_top]); i++, j++) { - rc = mdb_add_node(cdst, j, &key, NULL, 0, 0); + rc = mdb_node_add(cdst, j, &key, NULL, 0, 0); if (rc != MDB_SUCCESS) return rc; key.mv_data = (char *)key.mv_data + key.mv_size; @@ -3956,39 +5313,77 @@ mdb_merge(MDB_cursor *csrc, MDB_cursor *cdst) } else { for (i = 0; i < NUMKEYS(csrc->mc_pg[csrc->mc_top]); i++, j++) { srcnode = NODEPTR(csrc->mc_pg[csrc->mc_top], i); + if (i == 0 && IS_BRANCH(csrc->mc_pg[csrc->mc_top])) { + unsigned int snum = csrc->mc_snum; + MDB_node *s2; + /* must find the lowest key below src */ + mdb_page_search_root(csrc, NULL, 0); + s2 = NODEPTR(csrc->mc_pg[csrc->mc_top], 0); + key.mv_size = NODEKSZ(s2); + key.mv_data = NODEKEY(s2); + csrc->mc_snum = snum--; + csrc->mc_top = snum; + } else { + key.mv_size = srcnode->mn_ksize; + key.mv_data = NODEKEY(srcnode); + } - key.mv_size = srcnode->mn_ksize; - key.mv_data = NODEKEY(srcnode); data.mv_size = NODEDSZ(srcnode); data.mv_data = NODEDATA(srcnode); - rc = mdb_add_node(cdst, j, &key, &data, NODEPGNO(srcnode), srcnode->mn_flags); + rc = mdb_node_add(cdst, j, &key, &data, NODEPGNO(srcnode), srcnode->mn_flags); if (rc != MDB_SUCCESS) return rc; } } - DPRINTF("dst page %lu now has %u keys (%.1f%% filled)", + DPRINTF("dst page %zu now has %u keys (%.1f%% filled)", cdst->mc_pg[cdst->mc_top]->mp_pgno, NUMKEYS(cdst->mc_pg[cdst->mc_top]), (float)PAGEFILL(cdst->mc_txn->mt_env, cdst->mc_pg[cdst->mc_top]) / 10); /* Unlink the src page from parent and add to free list. */ - mdb_del_node(csrc->mc_pg[csrc->mc_top-1], csrc->mc_ki[csrc->mc_top-1], 0); + mdb_node_del(csrc->mc_pg[csrc->mc_top-1], csrc->mc_ki[csrc->mc_top-1], 0); if (csrc->mc_ki[csrc->mc_top-1] == 0) { key.mv_size = 0; if ((rc = mdb_update_key(csrc->mc_pg[csrc->mc_top-1], 0, &key)) != MDB_SUCCESS) return rc; } - mdb_midl_append(csrc->mc_txn->mt_free_pgs, csrc->mc_pg[csrc->mc_top]->mp_pgno); + mdb_midl_append(&csrc->mc_txn->mt_free_pgs, csrc->mc_pg[csrc->mc_top]->mp_pgno); if (IS_LEAF(csrc->mc_pg[csrc->mc_top])) - csrc->mc_txn->mt_dbs[csrc->mc_dbi].md_leaf_pages--; + csrc->mc_db->md_leaf_pages--; else - csrc->mc_txn->mt_dbs[csrc->mc_dbi].md_branch_pages--; - cursor_pop_page(csrc); + csrc->mc_db->md_branch_pages--; + { + /* Adjust other cursors pointing to mp */ + MDB_cursor *m2, *m3; + MDB_dbi dbi = csrc->mc_dbi; + MDB_page *mp = cdst->mc_pg[cdst->mc_top]; + + if (csrc->mc_flags & C_SUB) + dbi--; + + for (m2 = csrc->mc_txn->mt_cursors[dbi]; m2; m2=m2->mc_next) { + if (csrc->mc_flags & C_SUB) + m3 = &m2->mc_xcursor->mx_cursor; + else + m3 = m2; + if (m3 == csrc) continue; + if (m3->mc_snum < csrc->mc_snum) continue; + if (m3->mc_pg[csrc->mc_top] == csrc->mc_pg[csrc->mc_top]) { + m3->mc_pg[csrc->mc_top] = mp; + m3->mc_ki[csrc->mc_top] += nkeys; + } + } + } + mdb_cursor_pop(csrc); return mdb_rebalance(csrc); } +/** Copy the contents of a cursor. + * @param[in] csrc The cursor to copy from. + * @param[out] cdst The cursor to copy to. + */ static void mdb_cursor_copy(const MDB_cursor *csrc, MDB_cursor *cdst) { @@ -3996,6 +5391,8 @@ mdb_cursor_copy(const MDB_cursor *csrc, MDB_cursor *cdst) cdst->mc_txn = csrc->mc_txn; cdst->mc_dbi = csrc->mc_dbi; + cdst->mc_db = csrc->mc_db; + cdst->mc_dbx = csrc->mc_dbx; cdst->mc_snum = csrc->mc_snum; cdst->mc_top = csrc->mc_top; cdst->mc_flags = csrc->mc_flags; @@ -4006,40 +5403,99 @@ mdb_cursor_copy(const MDB_cursor *csrc, MDB_cursor *cdst) } } +/** Rebalance the tree after a delete operation. + * @param[in] mc Cursor pointing to the page where rebalancing + * should begin. + * @return 0 on success, non-zero on failure. + */ static int mdb_rebalance(MDB_cursor *mc) { MDB_node *node; - MDB_page *root; int rc; unsigned int ptop; MDB_cursor mn; - DPRINTF("rebalancing %s page %lu (has %u keys, %.1f%% full)", +#if MDB_DEBUG + { + pgno_t pgno; + COPY_PGNO(pgno, mc->mc_pg[mc->mc_top]->mp_pgno); + DPRINTF("rebalancing %s page %zu (has %u keys, %.1f%% full)", IS_LEAF(mc->mc_pg[mc->mc_top]) ? "leaf" : "branch", - mc->mc_pg[mc->mc_top]->mp_pgno, NUMKEYS(mc->mc_pg[mc->mc_top]), (float)PAGEFILL(mc->mc_txn->mt_env, mc->mc_pg[mc->mc_top]) / 10); + pgno, NUMKEYS(mc->mc_pg[mc->mc_top]), (float)PAGEFILL(mc->mc_txn->mt_env, mc->mc_pg[mc->mc_top]) / 10); + } +#endif if (PAGEFILL(mc->mc_txn->mt_env, mc->mc_pg[mc->mc_top]) >= FILL_THRESHOLD) { - DPRINTF("no need to rebalance page %lu, above fill threshold", - mc->mc_pg[mc->mc_top]->mp_pgno); +#if MDB_DEBUG + pgno_t pgno; + COPY_PGNO(pgno, mc->mc_pg[mc->mc_top]->mp_pgno); + DPRINTF("no need to rebalance page %zu, above fill threshold", + pgno); +#endif return MDB_SUCCESS; } if (mc->mc_snum < 2) { - if (NUMKEYS(mc->mc_pg[mc->mc_top]) == 0) { + MDB_page *mp = mc->mc_pg[0]; + if (NUMKEYS(mp) == 0) { DPUTS("tree is completely empty"); - mc->mc_txn->mt_dbs[mc->mc_dbi].md_root = P_INVALID; - mc->mc_txn->mt_dbs[mc->mc_dbi].md_depth = 0; - mc->mc_txn->mt_dbs[mc->mc_dbi].md_leaf_pages = 0; - mdb_midl_append(mc->mc_txn->mt_free_pgs, mc->mc_pg[mc->mc_top]->mp_pgno); - } else if (IS_BRANCH(mc->mc_pg[mc->mc_top]) && NUMKEYS(mc->mc_pg[mc->mc_top]) == 1) { + mc->mc_db->md_root = P_INVALID; + mc->mc_db->md_depth = 0; + mc->mc_db->md_leaf_pages = 0; + mdb_midl_append(&mc->mc_txn->mt_free_pgs, mp->mp_pgno); + mc->mc_snum = 0; + mc->mc_top = 0; + { + /* Adjust other cursors pointing to mp */ + MDB_cursor *m2, *m3; + MDB_dbi dbi = mc->mc_dbi; + + if (mc->mc_flags & C_SUB) + dbi--; + + for (m2 = mc->mc_txn->mt_cursors[dbi]; m2; m2=m2->mc_next) { + if (m2 == mc) continue; + if (mc->mc_flags & C_SUB) + m3 = &m2->mc_xcursor->mx_cursor; + else + m3 = m2; + if (m3->mc_snum < mc->mc_snum) continue; + if (m3->mc_pg[0] == mp) { + m3->mc_snum = 0; + m3->mc_top = 0; + } + } + } + } else if (IS_BRANCH(mp) && NUMKEYS(mp) == 1) { DPUTS("collapsing root page!"); - mdb_midl_append(mc->mc_txn->mt_free_pgs, mc->mc_pg[mc->mc_top]->mp_pgno); - mc->mc_txn->mt_dbs[mc->mc_dbi].md_root = NODEPGNO(NODEPTR(mc->mc_pg[mc->mc_top], 0)); - if ((rc = mdb_get_page(mc->mc_txn, mc->mc_txn->mt_dbs[mc->mc_dbi].md_root, &root))) + mdb_midl_append(&mc->mc_txn->mt_free_pgs, mp->mp_pgno); + mc->mc_db->md_root = NODEPGNO(NODEPTR(mp, 0)); + if ((rc = mdb_page_get(mc->mc_txn, mc->mc_db->md_root, + &mc->mc_pg[0]))) return rc; - mc->mc_txn->mt_dbs[mc->mc_dbi].md_depth--; - mc->mc_txn->mt_dbs[mc->mc_dbi].md_branch_pages--; + mc->mc_db->md_depth--; + mc->mc_db->md_branch_pages--; + { + /* Adjust other cursors pointing to mp */ + MDB_cursor *m2, *m3; + MDB_dbi dbi = mc->mc_dbi; + + if (mc->mc_flags & C_SUB) + dbi--; + + for (m2 = mc->mc_txn->mt_cursors[dbi]; m2; m2=m2->mc_next) { + if (m2 == mc) continue; + if (mc->mc_flags & C_SUB) + m3 = &m2->mc_xcursor->mx_cursor; + else + m3 = m2; + if (m3->mc_snum < mc->mc_snum) continue; + if (m3->mc_pg[0] == mp) { + m3->mc_pg[0] = mc->mc_pg[0]; + } + } + } } else DPUTS("root page doesn't need rebalancing"); return MDB_SUCCESS; @@ -4067,7 +5523,7 @@ mdb_rebalance(MDB_cursor *mc) DPUTS("reading right neighbor"); mn.mc_ki[ptop]++; node = NODEPTR(mc->mc_pg[ptop], mn.mc_ki[ptop]); - if ((rc = mdb_get_page(mc->mc_txn, NODEPGNO(node), &mn.mc_pg[mn.mc_top]))) + if ((rc = mdb_page_get(mc->mc_txn, NODEPGNO(node), &mn.mc_pg[mn.mc_top]))) return rc; mn.mc_ki[mn.mc_top] = 0; mc->mc_ki[mc->mc_top] = NUMKEYS(mc->mc_pg[mc->mc_top]); @@ -4077,13 +5533,13 @@ mdb_rebalance(MDB_cursor *mc) DPUTS("reading left neighbor"); mn.mc_ki[ptop]--; node = NODEPTR(mc->mc_pg[ptop], mn.mc_ki[ptop]); - if ((rc = mdb_get_page(mc->mc_txn, NODEPGNO(node), &mn.mc_pg[mn.mc_top]))) + if ((rc = mdb_page_get(mc->mc_txn, NODEPGNO(node), &mn.mc_pg[mn.mc_top]))) return rc; mn.mc_ki[mn.mc_top] = NUMKEYS(mn.mc_pg[mn.mc_top]) - 1; mc->mc_ki[mc->mc_top] = 0; } - DPRINTF("found neighbor page %lu (%u keys, %.1f%% full)", + DPRINTF("found neighbor page %zu (%u keys, %.1f%% full)", mn.mc_pg[mn.mc_top]->mp_pgno, NUMKEYS(mn.mc_pg[mn.mc_top]), (float)PAGEFILL(mc->mc_txn->mt_env, mn.mc_pg[mn.mc_top]) / 10); /* If the neighbor page is above threshold and has at least two @@ -4092,17 +5548,19 @@ mdb_rebalance(MDB_cursor *mc) * Otherwise we should try to merge them. */ if (PAGEFILL(mc->mc_txn->mt_env, mn.mc_pg[mn.mc_top]) >= FILL_THRESHOLD && NUMKEYS(mn.mc_pg[mn.mc_top]) >= 2) - return mdb_move_node(&mn, mc); + return mdb_node_move(&mn, mc); else { /* FIXME: if (has_enough_room()) */ + mc->mc_flags &= ~C_INITIALIZED; if (mc->mc_ki[ptop] == 0) - return mdb_merge(&mn, mc); + return mdb_page_merge(&mn, mc); else - return mdb_merge(mc, &mn); + return mdb_page_merge(mc, &mn); } } +/** Complete a delete operation started by #mdb_cursor_del(). */ static int -mdb_del0(MDB_cursor *mc, MDB_node *leaf) +mdb_cursor_del0(MDB_cursor *mc, MDB_node *leaf) { int rc; @@ -4113,14 +5571,15 @@ mdb_del0(MDB_cursor *mc, MDB_node *leaf) memcpy(&pg, NODEDATA(leaf), sizeof(pg)); ovpages = OVPAGES(NODEDSZ(leaf), mc->mc_txn->mt_env->me_psize); + mc->mc_db->md_overflow_pages -= ovpages; for (i=0; imc_txn->mt_free_pgs, pg); + DPRINTF("freed ov page %zu", pg); + mdb_midl_append(&mc->mc_txn->mt_free_pgs, pg); pg++; } } - mdb_del_node(mc->mc_pg[mc->mc_top], mc->mc_ki[mc->mc_top], mc->mc_txn->mt_dbs[mc->mc_dbi].md_pad); - mc->mc_txn->mt_dbs[mc->mc_dbi].md_entries--; + mdb_node_del(mc->mc_pg[mc->mc_top], mc->mc_ki[mc->mc_top], mc->mc_db->md_pad); + mc->mc_db->md_entries--; rc = mdb_rebalance(mc); if (rc != MDB_SUCCESS) mc->mc_txn->mt_flags |= MDB_TXN_ERROR; @@ -4154,15 +5613,7 @@ mdb_del(MDB_txn *txn, MDB_dbi dbi, return EINVAL; } - mc.mc_txn = txn; - mc.mc_dbi = dbi; - mc.mc_flags = 0; - if (txn->mt_dbs[dbi].md_flags & MDB_DUPSORT) { - mc.mc_xcursor = &mx; - mdb_xcursor_init0(&mc); - } else { - mc.mc_xcursor = NULL; - } + mdb_cursor_init(&mc, txn, dbi, &mx); exact = 0; if (data) { @@ -4179,21 +5630,26 @@ mdb_del(MDB_txn *txn, MDB_dbi dbi, return rc; } -/* Split page top>, and insert in either left or - * right sibling, at index ki> (as if unsplit). Updates mc->top and - * mc->ki with the actual values after split, ie if mc->top and mc->ki - * refer to a node in the new right sibling page. +/** Split a page and insert a new node. + * @param[in,out] mc Cursor pointing to the page and desired insertion index. + * The cursor will be updated to point to the actual page and index where + * the node got inserted after the split. + * @param[in] newkey The key for the newly inserted node. + * @param[in] newdata The data for the newly inserted node. + * @param[in] newpgno The page number, if the new node is a branch node. + * @return 0 on success, non-zero on failure. */ static int -mdb_split(MDB_cursor *mc, MDB_val *newkey, MDB_val *newdata, pgno_t newpgno) +mdb_page_split(MDB_cursor *mc, MDB_val *newkey, MDB_val *newdata, pgno_t newpgno, + unsigned int nflags) { - uint8_t flags; - int rc = MDB_SUCCESS, ins_new = 0; + unsigned int flags; + int rc = MDB_SUCCESS, ins_new = 0, new_root = 0, newpos = 1; indx_t newindx; pgno_t pgno = 0; unsigned int i, j, split_indx, nkeys, pmax; MDB_node *node; - MDB_val sepkey, rkey, rdata; + MDB_val sepkey, rkey, xdata, *rdata = &xdata; MDB_page *copy; MDB_page *mp, *rp, *pp; unsigned int ptop; @@ -4203,29 +5659,35 @@ mdb_split(MDB_cursor *mc, MDB_val *newkey, MDB_val *newdata, pgno_t newpgno) mp = mc->mc_pg[mc->mc_top]; newindx = mc->mc_ki[mc->mc_top]; - DPRINTF("-----> splitting %s page %lu and adding [%s] at index %i", + DPRINTF("-----> splitting %s page %zu and adding [%s] at index %i", IS_LEAF(mp) ? "leaf" : "branch", mp->mp_pgno, DKEY(newkey), mc->mc_ki[mc->mc_top]); + /* Create a right sibling. */ + if ((rp = mdb_page_new(mc, mp->mp_flags, 1)) == NULL) + return ENOMEM; + DPRINTF("new right sibling: page %zu", rp->mp_pgno); + if (mc->mc_snum < 2) { - if ((pp = mdb_new_page(mc, P_BRANCH, 1)) == NULL) + if ((pp = mdb_page_new(mc, P_BRANCH, 1)) == NULL) return ENOMEM; /* shift current top to make room for new parent */ mc->mc_pg[1] = mc->mc_pg[0]; mc->mc_ki[1] = mc->mc_ki[0]; mc->mc_pg[0] = pp; mc->mc_ki[0] = 0; - mc->mc_txn->mt_dbs[mc->mc_dbi].md_root = pp->mp_pgno; - DPRINTF("root split! new root = %lu", pp->mp_pgno); - mc->mc_txn->mt_dbs[mc->mc_dbi].md_depth++; + mc->mc_db->md_root = pp->mp_pgno; + DPRINTF("root split! new root = %zu", pp->mp_pgno); + mc->mc_db->md_depth++; + new_root = 1; /* Add left (implicit) pointer. */ - if ((rc = mdb_add_node(mc, 0, NULL, NULL, mp->mp_pgno, 0)) != MDB_SUCCESS) { + if ((rc = mdb_node_add(mc, 0, NULL, NULL, mp->mp_pgno, 0)) != MDB_SUCCESS) { /* undo the pre-push */ mc->mc_pg[0] = mc->mc_pg[1]; mc->mc_ki[0] = mc->mc_ki[1]; - mc->mc_txn->mt_dbs[mc->mc_dbi].md_root = mp->mp_pgno; - mc->mc_txn->mt_dbs[mc->mc_dbi].md_depth--; + mc->mc_db->md_root = mp->mp_pgno; + mc->mc_db->md_depth--; return rc; } mc->mc_snum = 2; @@ -4233,19 +5695,23 @@ mdb_split(MDB_cursor *mc, MDB_val *newkey, MDB_val *newdata, pgno_t newpgno) ptop = 0; } else { ptop = mc->mc_top-1; - DPRINTF("parent branch page is %lu", mc->mc_pg[ptop]->mp_pgno); + DPRINTF("parent branch page is %zu", mc->mc_pg[ptop]->mp_pgno); } - /* Create a right sibling. */ - if ((rp = mdb_new_page(mc, mp->mp_flags, 1)) == NULL) - return ENOMEM; mdb_cursor_copy(mc, &mn); mn.mc_pg[mn.mc_top] = rp; mn.mc_ki[ptop] = mc->mc_ki[ptop]+1; - DPRINTF("new right sibling: page %lu", rp->mp_pgno); + + if (nflags & MDB_APPEND) { + mn.mc_ki[mn.mc_top] = 0; + sepkey = *newkey; + nkeys = 0; + split_indx = 0; + goto newsep; + } nkeys = NUMKEYS(mp); - split_indx = nkeys / 2 + 1; + split_indx = (nkeys + 1) / 2; if (IS_LEAF2(rp)) { char *split, *ins; @@ -4254,7 +5720,7 @@ mdb_split(MDB_cursor *mc, MDB_val *newkey, MDB_val *newdata, pgno_t newpgno) /* Move half of the keys to the right sibling */ copy = NULL; x = mc->mc_ki[mc->mc_top] - split_indx; - ksize = mc->mc_txn->mt_dbs[mc->mc_dbi].md_pad; + ksize = mc->mc_db->md_pad; split = LEAF2KEY(mp, split_indx, ksize); rsize = (nkeys - split_indx) * ksize; lsize = (nkeys - split_indx) * sizeof(indx_t); @@ -4291,15 +5757,22 @@ mdb_split(MDB_cursor *mc, MDB_val *newkey, MDB_val *newdata, pgno_t newpgno) } /* For leaf pages, check the split point based on what - * fits where, since otherwise add_node can fail. + * fits where, since otherwise mdb_node_add can fail. + * + * This check is only needed when the data items are + * relatively large, such that being off by one will + * make the difference between success or failure. + * When the size of the data items is much smaller than + * one-half of a page, this check is irrelevant. */ - if (IS_LEAF(mp)) { + if (IS_LEAF(mp) && nkeys < 16) { unsigned int psize, nsize; /* Maximum free space in an empty page */ pmax = mc->mc_txn->mt_env->me_psize - PAGEHDRSZ; nsize = mdb_leaf_size(mc->mc_txn->mt_env, newkey, newdata); - if (newindx < split_indx) { + if (newindx <= split_indx) { psize = nsize; + newpos = 0; for (i=0; i pmax) { - split_indx = i; + if (i == split_indx - 1 && newindx == split_indx) + newpos = 1; + else + split_indx = i; break; } } @@ -4332,8 +5808,11 @@ mdb_split(MDB_cursor *mc, MDB_val *newkey, MDB_val *newdata, pgno_t newpgno) } /* First find the separating key between the split pages. + * The case where newindx == split_indx is ambiguous; the + * new item could go to the new page or stay on the original + * page. If newpos == 1 it goes to the new page. */ - if (newindx == split_indx) { + if (newindx == split_indx && newpos) { sepkey.mv_size = newkey->mv_size; sepkey.mv_data = newkey->mv_data; } else { @@ -4350,7 +5829,7 @@ newsep: if (SIZELEFT(mn.mc_pg[ptop]) < mdb_branch_size(mc->mc_txn->mt_env, &sepkey)) { mn.mc_snum--; mn.mc_top--; - rc = mdb_split(&mn, &sepkey, NULL, rp->mp_pgno); + rc = mdb_page_split(&mn, &sepkey, NULL, rp->mp_pgno, 0); /* Right page might now have changed parent. * Check if left page also changed parent. @@ -4362,26 +5841,30 @@ newsep: } } else { mn.mc_top--; - rc = mdb_add_node(&mn, mn.mc_ki[ptop], &sepkey, NULL, rp->mp_pgno, 0); + rc = mdb_node_add(&mn, mn.mc_ki[ptop], &sepkey, NULL, rp->mp_pgno, 0); mn.mc_top++; } - if (IS_LEAF2(rp)) { - return rc; - } if (rc != MDB_SUCCESS) { return rc; } + if (nflags & MDB_APPEND) { + mc->mc_pg[mc->mc_top] = rp; + mc->mc_ki[mc->mc_top] = 0; + rc = mdb_node_add(mc, 0, newkey, newdata, newpgno, nflags); + if (rc) + return rc; + goto done; + } + if (IS_LEAF2(rp)) { + goto done; + } /* Move half of the keys to the right sibling. */ /* grab a page to hold a temporary copy */ - if (mc->mc_txn->mt_env->me_dpages) { - copy = mc->mc_txn->mt_env->me_dpages; - mc->mc_txn->mt_env->me_dpages = copy->mp_next; - } else { - if ((copy = malloc(mc->mc_txn->mt_env->me_psize)) == NULL) - return ENOMEM; - } + copy = mdb_page_malloc(mc); + if (copy == NULL) + return ENOMEM; copy->mp_pgno = mp->mp_pgno; copy->mp_flags = mp->mp_flags; @@ -4392,8 +5875,10 @@ newsep: if (i == split_indx) { /* Insert in right sibling. */ /* Reset insert index for right sibling. */ - j = (i == newindx && ins_new); - mc->mc_pg[mc->mc_top] = rp; + if (i != newindx || (newpos ^ ins_new)) { + j = 0; + mc->mc_pg[mc->mc_top] = rp; + } } if (i == newindx && !ins_new) { @@ -4401,15 +5886,14 @@ newsep: rkey.mv_data = newkey->mv_data; rkey.mv_size = newkey->mv_size; if (IS_LEAF(mp)) { - rdata.mv_data = newdata->mv_data; - rdata.mv_size = newdata->mv_size; + rdata = newdata; } else pgno = newpgno; - flags = 0; + flags = nflags; ins_new = 1; - /* Update page and index for the new key. */ + /* Update index for the new key. */ mc->mc_ki[mc->mc_top] = j; } else if (i == nkeys) { break; @@ -4418,8 +5902,9 @@ newsep: rkey.mv_data = NODEKEY(node); rkey.mv_size = node->mn_ksize; if (IS_LEAF(mp)) { - rdata.mv_data = NODEDATA(node); - rdata.mv_size = NODEDSZ(node); + xdata.mv_data = NODEDATA(node); + xdata.mv_size = NODEDSZ(node); + rdata = &xdata; } else pgno = NODEPGNO(node); flags = node->mn_flags; @@ -4432,13 +5917,9 @@ newsep: rkey.mv_size = 0; } - rc = mdb_add_node(mc, j, &rkey, &rdata, pgno, flags); + rc = mdb_node_add(mc, j, &rkey, rdata, pgno, flags); } - /* reset back to original page */ - if (newindx < split_indx) - mc->mc_pg[mc->mc_top] = mp; - nkeys = NUMKEYS(copy); for (i=0; imp_ptrs[i] = copy->mp_ptrs[i]; @@ -4447,9 +5928,57 @@ newsep: memcpy(NODEPTR(mp, nkeys-1), NODEPTR(copy, nkeys-1), mc->mc_txn->mt_env->me_psize - copy->mp_upper); + /* reset back to original page */ + if (newindx < split_indx || (!newpos && newindx == split_indx)) { + mc->mc_pg[mc->mc_top] = mp; + if (nflags & MDB_RESERVE) { + node = NODEPTR(mp, mc->mc_ki[mc->mc_top]); + if (!(node->mn_flags & F_BIGDATA)) + newdata->mv_data = NODEDATA(node); + } + } + /* return tmp page to freelist */ copy->mp_next = mc->mc_txn->mt_env->me_dpages; + VGMEMP_FREE(mc->mc_txn->mt_env, copy); mc->mc_txn->mt_env->me_dpages = copy; +done: + { + /* Adjust other cursors pointing to mp */ + MDB_cursor *m2, *m3; + MDB_dbi dbi = mc->mc_dbi; + + if (mc->mc_flags & C_SUB) + dbi--; + + for (m2 = mc->mc_txn->mt_cursors[dbi]; m2; m2=m2->mc_next) { + if (m2 == mc) continue; + if (mc->mc_flags & C_SUB) + m3 = &m2->mc_xcursor->mx_cursor; + else + m3 = m2; + if (!(m3->mc_flags & C_INITIALIZED)) + continue; + if (new_root) { + int k; + /* root split */ + for (k=m3->mc_top; k>=0; k--) { + m3->mc_ki[k+1] = m3->mc_ki[k]; + m3->mc_pg[k+1] = m3->mc_pg[k]; + } + m3->mc_ki[0] = mc->mc_ki[0]; + m3->mc_pg[0] = mc->mc_pg[0]; + m3->mc_snum++; + m3->mc_top++; + } + if (m3->mc_pg[mc->mc_top] == mp) { + if (m3->mc_ki[m3->mc_top] >= split_indx) { + m3->mc_pg[m3->mc_top] = rp; + m3->mc_ki[m3->mc_top] -= split_indx; + } + } + } + } return rc; } @@ -4474,26 +6003,21 @@ mdb_put(MDB_txn *txn, MDB_dbi dbi, return EINVAL; } - if ((flags & (MDB_NOOVERWRITE|MDB_NODUPDATA)) != flags) + if ((flags & (MDB_NOOVERWRITE|MDB_NODUPDATA|MDB_RESERVE|MDB_APPEND)) != flags) return EINVAL; - mc.mc_txn = txn; - mc.mc_dbi = dbi; - mc.mc_snum = 0; - mc.mc_flags = 0; - if (txn->mt_dbs[dbi].md_flags & MDB_DUPSORT) { - mc.mc_xcursor = &mx; - mdb_xcursor_init0(&mc); - } else { - mc.mc_xcursor = NULL; - } + mdb_cursor_init(&mc, txn, dbi, &mx); return mdb_cursor_put(&mc, key, data, flags); } +/** Only a subset of the @ref mdb_env flags can be changed + * at runtime. Changing other flags requires closing the environment + * and re-opening it with the new flags. + */ +#define CHANGEABLE (MDB_NOSYNC) int mdb_env_set_flags(MDB_env *env, unsigned int flag, int onoff) { -#define CHANGEABLE (MDB_NOSYNC) if ((flag & CHANGEABLE) != flag) return EINVAL; if (onoff) @@ -4523,6 +6047,12 @@ mdb_env_get_path(MDB_env *env, const char **arg) return MDB_SUCCESS; } +/** Common code for #mdb_stat() and #mdb_env_stat(). + * @param[in] env the environment to operate in. + * @param[in] db the #MDB_db record containing the stats to return. + * @param[out] arg the address of an #MDB_stat structure to receive the stats. + * @return 0, this function always succeeds. + */ static int mdb_stat0(MDB_env *env, MDB_db *db, MDB_stat *arg) { @@ -4548,26 +6078,33 @@ mdb_env_stat(MDB_env *env, MDB_stat *arg) return mdb_stat0(env, &env->me_metas[toggle]->mm_dbs[MAIN_DBI], arg); } +/** Set the default comparison functions for a database. + * Called immediately after a database is opened to set the defaults. + * The user can then override them with #mdb_set_compare() or + * #mdb_set_dupsort(). + * @param[in] txn A transaction handle returned by #mdb_txn_begin() + * @param[in] dbi A database handle returned by #mdb_open() + */ static void mdb_default_cmp(MDB_txn *txn, MDB_dbi dbi) { if (txn->mt_dbs[dbi].md_flags & MDB_REVERSEKEY) - txn->mt_dbxs[dbi].md_cmp = memnrcmp; + txn->mt_dbxs[dbi].md_cmp = mdb_cmp_memnr; else if (txn->mt_dbs[dbi].md_flags & MDB_INTEGERKEY) - txn->mt_dbxs[dbi].md_cmp = cintcmp; + txn->mt_dbxs[dbi].md_cmp = mdb_cmp_cint; else - txn->mt_dbxs[dbi].md_cmp = memncmp; + txn->mt_dbxs[dbi].md_cmp = mdb_cmp_memn; if (txn->mt_dbs[dbi].md_flags & MDB_DUPSORT) { if (txn->mt_dbs[dbi].md_flags & MDB_INTEGERDUP) { if (txn->mt_dbs[dbi].md_flags & MDB_DUPFIXED) - txn->mt_dbxs[dbi].md_dcmp = intcmp; + txn->mt_dbxs[dbi].md_dcmp = mdb_cmp_int; else - txn->mt_dbxs[dbi].md_dcmp = cintcmp; + txn->mt_dbxs[dbi].md_dcmp = mdb_cmp_cint; } else if (txn->mt_dbs[dbi].md_flags & MDB_REVERSEDUP) { - txn->mt_dbxs[dbi].md_dcmp = memnrcmp; + txn->mt_dbxs[dbi].md_dcmp = mdb_cmp_memnr; } else { - txn->mt_dbxs[dbi].md_dcmp = memncmp; + txn->mt_dbxs[dbi].md_dcmp = mdb_cmp_memn; } } else { txn->mt_dbxs[dbi].md_dcmp = NULL; @@ -4578,7 +6115,8 @@ int mdb_open(MDB_txn *txn, const char *name, unsigned int flags, MDB_dbi *dbi) { MDB_val key, data; MDB_dbi i; - int rc, dirty = 0; + MDB_cursor mc; + int rc, dbflag, exact; size_t len; if (txn->mt_dbxs[FREE_DBI].md_cmp == NULL) { @@ -4612,24 +6150,27 @@ int mdb_open(MDB_txn *txn, const char *name, unsigned int flags, MDB_dbi *dbi) return ENFILE; /* Find the DB info */ + dbflag = 0; + exact = 0; key.mv_size = len; key.mv_data = (void *)name; - rc = mdb_get(txn, MAIN_DBI, &key, &data); - - /* Create if requested */ - if (rc == MDB_NOTFOUND && (flags & MDB_CREATE)) { - MDB_cursor mc; + mdb_cursor_init(&mc, txn, MAIN_DBI, NULL); + rc = mdb_cursor_set(&mc, &key, &data, MDB_SET, &exact); + if (rc == MDB_SUCCESS) { + /* make sure this is actually a DB */ + MDB_node *node = NODEPTR(mc.mc_pg[mc.mc_top], mc.mc_ki[mc.mc_top]); + if (!(node->mn_flags & F_SUBDATA)) + return EINVAL; + } else if (rc == MDB_NOTFOUND && (flags & MDB_CREATE)) { + /* Create if requested */ MDB_db dummy; data.mv_size = sizeof(MDB_db); data.mv_data = &dummy; memset(&dummy, 0, sizeof(dummy)); dummy.md_root = P_INVALID; dummy.md_flags = flags & 0xffff; - mc.mc_txn = txn; - mc.mc_dbi = MAIN_DBI; - mc.mc_flags = 0; rc = mdb_cursor_put(&mc, &key, &data, F_SUBDATA); - dirty = 1; + dbflag = DB_DIRTY; } /* OK, got info, add to table */ @@ -4637,8 +6178,7 @@ int mdb_open(MDB_txn *txn, const char *name, unsigned int flags, MDB_dbi *dbi) txn->mt_dbxs[txn->mt_numdbs].md_name.mv_data = strdup(name); txn->mt_dbxs[txn->mt_numdbs].md_name.mv_size = len; txn->mt_dbxs[txn->mt_numdbs].md_rel = NULL; - txn->mt_dbxs[txn->mt_numdbs].md_parent = MAIN_DBI; - txn->mt_dbxs[txn->mt_numdbs].md_dirty = dirty; + txn->mt_dbflags[txn->mt_numdbs] = dbflag; memcpy(&txn->mt_dbs[txn->mt_numdbs], data.mv_data, sizeof(MDB_db)); *dbi = txn->mt_numdbs; txn->mt_env->me_dbs[0][txn->mt_numdbs] = txn->mt_dbs[txn->mt_numdbs]; @@ -4658,17 +6198,115 @@ int mdb_stat(MDB_txn *txn, MDB_dbi dbi, MDB_stat *arg) return mdb_stat0(txn->mt_env, &txn->mt_dbs[dbi], arg); } -void mdb_close(MDB_txn *txn, MDB_dbi dbi) +void mdb_close(MDB_env *env, MDB_dbi dbi) { char *ptr; - if (dbi <= MAIN_DBI || dbi >= txn->mt_numdbs) + if (dbi <= MAIN_DBI || dbi >= env->me_numdbs) return; - ptr = txn->mt_dbxs[dbi].md_name.mv_data; - txn->mt_dbxs[dbi].md_name.mv_data = NULL; - txn->mt_dbxs[dbi].md_name.mv_size = 0; + ptr = env->me_dbxs[dbi].md_name.mv_data; + env->me_dbxs[dbi].md_name.mv_data = NULL; + env->me_dbxs[dbi].md_name.mv_size = 0; free(ptr); } +/** Add all the DB's pages to the free list. + * @param[in] mc Cursor on the DB to free. + * @param[in] subs non-Zero to check for sub-DBs in this DB. + * @return 0 on success, non-zero on failure. + */ +static int +mdb_drop0(MDB_cursor *mc, int subs) +{ + int rc; + + rc = mdb_page_search(mc, NULL, 0); + if (rc == MDB_SUCCESS) { + MDB_node *ni; + MDB_cursor mx; + unsigned int i; + + /* LEAF2 pages have no nodes, cannot have sub-DBs */ + if (!subs || IS_LEAF2(mc->mc_pg[mc->mc_top])) + mdb_cursor_pop(mc); + + mdb_cursor_copy(mc, &mx); + while (mc->mc_snum > 0) { + if (IS_LEAF(mc->mc_pg[mc->mc_top])) { + for (i=0; imc_pg[mc->mc_top]); i++) { + ni = NODEPTR(mc->mc_pg[mc->mc_top], i); + if (ni->mn_flags & F_SUBDATA) { + mdb_xcursor_init1(mc, ni); + rc = mdb_drop0(&mc->mc_xcursor->mx_cursor, 0); + if (rc) + return rc; + } + } + } else { + for (i=0; imc_pg[mc->mc_top]); i++) { + pgno_t pg; + ni = NODEPTR(mc->mc_pg[mc->mc_top], i); + pg = NODEPGNO(ni); + /* free it */ + mdb_midl_append(&mc->mc_txn->mt_free_pgs, pg); + } + } + if (!mc->mc_top) + break; + rc = mdb_cursor_sibling(mc, 1); + if (rc) { + /* no more siblings, go back to beginning + * of previous level. (stack was already popped + * by mdb_cursor_sibling) + */ + for (i=1; imc_top; i++) + mc->mc_pg[i] = mx.mc_pg[i]; + } + } + /* free it */ + mdb_midl_append(&mc->mc_txn->mt_free_pgs, + mc->mc_db->md_root); + } + return 0; +} + +int mdb_drop(MDB_txn *txn, MDB_dbi dbi, int del) +{ + MDB_cursor *mc; + int rc; + + if (!txn || !dbi || dbi >= txn->mt_numdbs) + return EINVAL; + + if (F_ISSET(txn->mt_flags, MDB_TXN_RDONLY)) + return EACCES; + + rc = mdb_cursor_open(txn, dbi, &mc); + if (rc) + return rc; + + rc = mdb_drop0(mc, mc->mc_db->md_flags & MDB_DUPSORT); + if (rc) + goto leave; + + /* Can't delete the main DB */ + if (del && dbi > MAIN_DBI) { + rc = mdb_del(txn, MAIN_DBI, &mc->mc_dbx->md_name, NULL); + if (!rc) + mdb_close(txn->mt_env, dbi); + } else { + txn->mt_dbflags[dbi] |= DB_DIRTY; + txn->mt_dbs[dbi].md_depth = 0; + txn->mt_dbs[dbi].md_branch_pages = 0; + txn->mt_dbs[dbi].md_leaf_pages = 0; + txn->mt_dbs[dbi].md_overflow_pages = 0; + txn->mt_dbs[dbi].md_entries = 0; + txn->mt_dbs[dbi].md_root = P_INVALID; + } +leave: + mdb_cursor_close(mc); + return rc; +} + int mdb_set_compare(MDB_txn *txn, MDB_dbi dbi, MDB_cmp_func *cmp) { if (txn == NULL || !dbi || dbi >= txn->mt_numdbs) @@ -4696,4 +6334,13 @@ int mdb_set_relfunc(MDB_txn *txn, MDB_dbi dbi, MDB_rel_func *rel) return MDB_SUCCESS; } +int mdb_set_relctx(MDB_txn *txn, MDB_dbi dbi, void *ctx) +{ + if (txn == NULL || !dbi || dbi >= txn->mt_numdbs) + return EINVAL; + + txn->mt_dbxs[dbi].md_relctx = ctx; + return MDB_SUCCESS; +} + /** @} */