* @{
*/
/** @defgroup compat Windows Compatibility Macros
+ * A bunch of macros to minimize the amount of platform-specific ifdefs
+ * needed throughout the rest of the code. When the features this library
+ * needs are similar enough to POSIX to be hidden in a one-or-two line
+ * replacement, this macro approach is used.
* @{
*/
#ifdef _WIN32
#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
-#define LOCK_MUTEX_R(env) pthread_mutex_lock(&env->me_txns->mti_mutex)
-#define UNLOCK_MUTEX_R(env) pthread_mutex_unlock(&env->me_txns->mti_mutex)
-#define LOCK_MUTEX_W(env) pthread_mutex_lock(&env->me_txns->mti_wmutex)
-#define UNLOCK_MUTEX_W(env) pthread_mutex_unlock(&env->me_txns->mti_wmutex)
+ /** Lock the reader 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)
+
+ /** 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)
+ /** Unlock the writer mutex.
+ */
+#define UNLOCK_MUTEX_W(env) pthread_mutex_unlock(&(env)->me_txns->mti_wmutex)
+
+ /** Get the error code for the last failed system function.
+ */
#define ErrCode() errno
+
+ /** An abstraction for a file handle.
+ * On POSIX systems file handles are small integers. On Windows
+ * they're opaque pointers.
+ */
#define HANDLE int
-#define INVALID_HANDLE_VALUE -1
-#define GetPageSize(x) (x) = sysconf(_SC_PAGE_SIZE)
+
+ /** A value for an invalid file handle.
+ * Mainly used to initialize file variables and signify that they are
+ * unused.
+ */
+#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 GET_PAGESIZE(x) ((x) = sysconf(_SC_PAGE_SIZE))
#endif
/** @} */
#ifndef _WIN32
-/* Note: If O_DSYNC is undefined but exists in /usr/include,
+/** A flag for opening a file and requesting synchronous data writes.
+ * This is only used when writing a meta page. It's not strictly needed;
+ * we could just do a normal write and then immediately perform a flush.
+ * But if this flag is available it saves us an extra system call.
+ *
+ * @note If O_DSYNC is undefined but exists in /usr/include,
* preferably set some compiler flag to get the definition.
* Otherwise compile with the less efficient -DMDB_DSYNC=O_SYNC.
*/
#endif
#endif
+ /** A page number in the database.
+ * Note that 64 bit page numbers are overkill, since pages themselves
+ * already represent 12-13 bits of addressable memory, and the OS will
+ * always limit applications to a maximum of 63 bits of address space.
+ *
+ * @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;
+/** @defgroup debug Debug Macros
+ * @{
+ */
#ifndef 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
#endif
#if !(__STDC_VERSION__ >= 199901L || defined(__GNUC__))
# define DPRINTF (void) /* Vararg macros may be unsupported */
#elif DEBUG
-# define DPRINTF(fmt, ...) /* Requires 2 or more args */ \
+ /** 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__)
#else
# define DPRINTF(fmt, ...) ((void) 0)
#endif
+ /** Print a debug string.
+ * The string is printed literally, with no format processing.
+ */
#define DPUTS(arg) DPRINTF("%s", arg)
+/** @} */
+ /** 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 #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
+ * Huge pages aren't actually pageable. We rely on the OS
+ * demand-pager to read our data and page it out when memory
+ * pressure from other processes is high. So until OSs have
+ * actual paging support for Huge pages, they're not viable.
+ */
#define 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
+ * maximum size of a data item. Data items larger than this size will
+ * be pushed into overflow pages instead of being stored directly in
+ * the B-tree node. This value used to default to 4. With a page size
+ * of 4096 bytes that meant that any item larger than 1024 bytes would
+ * go into an overflow page. That also meant that on average 2-3KB of
+ * each overflow page was wasted space. The value cannot be lower than
+ * 2 because then there would no longer be a tree structure. With this
+ * value, items larger than 2KB will go into overflow pages, and on
+ * average only 1KB will be wasted.
+ */
#define MDB_MINKEYS 2
+
+ /** A stamp that identifies a file as an MDB file.
+ * There's nothing special about this value other than that it is easily
+ * recognizable, and it will reflect any byte order mismatches.
+ */
#define MDB_MAGIC 0xBEEFC0DE
+
+ /** The version number for a database's file format. */
#define MDB_VERSION 1
+
+ /** 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.
+ */
#define MAXKEYSIZE 511
+
#if DEBUG
-#define KBUF (MAXKEYSIZE*2+1)
-#define DKBUF char kbuf[KBUF]
+ /** A key buffer.
+ * @ingroup debug
+ * This is used for printing a hex dump of a key's contents.
+ */
+#define DKBUF char kbuf[(MAXKEYSIZE*2+1)]
+ /** Display a key in hex.
+ * @ingroup debug
+ * Invoke a function to display a key in hex.
+ */
#define DKEY(x) mdb_dkey(x, kbuf)
#else
-#define DKBUF
+#define DKBUF typedef int dummy_kbuf /* so we can put ';' after */
#define DKEY(x)
#endif
-/* The DB view is always consistent because all writes are wrapped in
- * the wmutex. Finer-grained locks aren't necessary.
+/** @defgroup lazylock Lazy Locking
+ * Macros for locks that are't actually needed.
+ * The DB view is always consistent because all writes are wrapped in
+ * the wmutex. Finer-grained locks aren't necessary.
+ * @{
*/
#ifndef LAZY_LOCKS
+ /** Use lazy locking. I.e., don't lock these accesses at all. */
#define LAZY_LOCKS 1
#endif
#if LAZY_LOCKS
+ /** Grab the reader lock */
#define LAZY_MUTEX_LOCK(x)
+ /** Release the reader lock */
#define LAZY_MUTEX_UNLOCK(x)
+ /** Release the DB table reader/writer lock */
#define LAZY_RWLOCK_UNLOCK(x)
+ /** Grab the DB table write lock */
#define LAZY_RWLOCK_WRLOCK(x)
+ /** Grab the DB table read lock */
#define LAZY_RWLOCK_RDLOCK(x)
+ /** 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)
+ /** Destroy the DB table rwlock */
#define LAZY_RWLOCK_DESTROY(x)
#else
#define LAZY_MUTEX_LOCK(x) pthread_mutex_lock(x)
#define LAZY_RWLOCK_INIT(x,y) pthread_rwlock_init(x,y)
#define LAZY_RWLOCK_DESTROY(x) pthread_rwlock_destroy(x)
#endif
+/** @} */
+ /** An invalid page number.
+ * Mainly used to denote an empty tree.
+ */
#define P_INVALID (~0UL)
+ /** Test if a flag \b f is set in a flag word \b w. */
#define F_ISSET(w, f) (((w) & (f)) == (f))
+ /** Used for offsets within a single page.
+ * Since memory pages are typically 4 or 8KB in size, 12-13 bits,
+ * this is plenty.
+ */
typedef uint16_t indx_t;
-#define DEFAULT_READERS 126
+ /** Default size of memory map.
+ * This is certainly too small for any actual applications. Apps should always set
+ * the size explicitly using #mdb_env_set_mapsize().
+ */
#define DEFAULT_MAPSIZE 1048576
-/* Lock descriptor stuff */
+/** @defgroup readers Reader Lock Table
+ * Readers don't acquire any locks for their data access. Instead, they
+ * simply record their transaction ID in the reader table. The reader
+ * mutex is needed just to find an empty slot in the reader table. The
+ * slot's address is saved in thread-specific data so that subsequent read
+ * transactions started by the same thread need no further locking to proceed.
+ *
+ * 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 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
+ * processor's cache line size. Any slot is only ever used by one thread.
+ * This alignment guarantees that there will be no contention or cache
+ * thrashing as threads update their own slot info, and also eliminates
+ * any need for locking when accessing a slot.
+ *
+ * A writer thread will scan every slot in the table to determine the oldest
+ * outstanding reader transaction. Any freed pages older than this will be
+ * reclaimed by the writer. The writer doesn't use any locks when scanning
+ * this table. This means that there's no guarantee that the writer will
+ * see the most up-to-date reader info, but that's not required for correct
+ * operation - all we need is to know the upper bound on the oldest reader,
+ * we don't care at all about the newest reader. So the only consequence of
+ * reading stale information here is that old pages might hang around a
+ * while longer before being reclaimed. That's actually good anyway, because
+ * the longer we delay reclaiming old pages, the more likely it is that a
+ * string of contiguous pages can be found after coalescing old pages from
+ * many old transactions together.
+ *
+ * @todo We don't actually do such coalescing yet, we grab pages from one
+ * old transaction at a time.
+ * @{
+ */
+ /** Number of slots in the reader table.
+ * This value was chosen somewhat arbitrarily. 126 readers plus a
+ * couple mutexes fit exactly into 8KB on my development machine.
+ * Applications should set the table size using #mdb_env_set_maxreaders().
+ */
+#define DEFAULT_READERS 126
+
+ /** The size of a CPU cache line in bytes. We want our lock structures
+ * aligned to this size to avoid false cache line sharing in the
+ * lock table.
+ * This value works for most CPUs. For Itanium this should be 128.
+ */
#ifndef CACHELINE
-#define CACHELINE 64 /* most CPUs. Itanium uses 128 */
+#define CACHELINE 64
#endif
+ /** The information we store in a single slot of the reader table.
+ * In addition to a transaction ID, we also record the process and
+ * thread ID that owns a slot, so that we can detect stale information,
+ * e.g. threads or processes that went away without cleaning up.
+ * @note We currently don't check for stale records. We simply re-init
+ * the table when we know that we're the only process opening the
+ * lock file.
+ */
typedef struct MDB_rxbody {
+ /** The current Transaction ID when this transaction began.
+ * Multiple readers that start at the same time will probably have the
+ * same ID here. Again, it's not important to exclude them from
+ * anything; all we need to know is which version of the DB they
+ * started from so we can avoid overwriting any data used in that
+ * particular version.
+ */
ULONG 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. */
pthread_t mrb_tid;
} MDB_rxbody;
+ /** The actual reader record, with cacheline padding. */
typedef struct MDB_reader {
union {
MDB_rxbody mrx;
+ /** shorthand for mrb_txnid */
#define mr_txnid mru.mrx.mrb_txnid
#define mr_pid mru.mrx.mrb_pid
#define mr_tid mru.mrx.mrb_tid
- /* cache line alignment */
+ /** cache line alignment */
char pad[(sizeof(MDB_rxbody)+CACHELINE-1) & ~(CACHELINE-1)];
} mru;
} MDB_reader;
+ /** The header for the reader table.
+ * The table resides in a memory-mapped file. (This is a different file
+ * than is used for the main database.)
+ *
+ * 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
+ * MacOSX/Darwin (using named semaphores) since MacOSX doesn't support
+ * process-shared POSIX mutexes.
+ */
typedef struct MDB_txbody {
+ /** Stamp identifying this as an MDB lock 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;
-/* For POSIX the actual mutexes reside in shared memory.
- * On Windows, mutexes are allocated by the kernel; we store
- * the name in shared memory so that other processes can
- * grab them.
- */
#ifdef _WIN32
char mtb_rmname[32];
#else
+ /** Mutex protecting access to this table.
+ * This is the reader lock that #LOCK_MUTEX_R acquires.
+ */
pthread_mutex_t mtb_mutex;
#endif
+ /** The ID of the last transaction committed to the database.
+ * This is recorded here only for convenience; the value can always
+ * be determined by reading the main database meta pages.
+ */
ULONG 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;
+ /** 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.
+ */
uint32_t mtb_me_toggle;
} MDB_txbody;
+ /** The actual reader table definition. */
typedef struct MDB_txninfo {
union {
MDB_txbody mtb;
} mt2;
MDB_reader mti_readers[1];
} MDB_txninfo;
+/** @} */
-/* Common header for all page types. Overflow pages
- * occupy a number of contiguous pages with no
+/** Common header for all page types.
+ * Overflow pages occupy a number of contiguous pages with no
* headers on any page after the first.
*/
-typedef struct MDB_page { /* represents a page of storage */
-#define mp_pgno mp_p.p_pgno
+typedef struct MDB_page {
+#define mp_pgno mp_p.p_pgno
+#define mp_next mp_p.p_next
union padded {
- pgno_t p_pgno; /* page number */
- void * p_align; /* for IL32P64 */
+ pgno_t p_pgno; /**< page number */
+ void * p_next; /**< for in-memory list of freed structs */
} mp_p;
-#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 /* DB with small, fixed size keys and no data */
+#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 mp_lower mp_pb.pb.pb_lower
#define mp_upper mp_pb.pb.pb_upper
#define mp_pages mp_pb.pb_pages
union page_bounds {
struct {
- indx_t pb_lower; /* lower bound of free space */
- indx_t pb_upper; /* upper bound of free space */
+ 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 */
+ uint32_t pb_pages; /**< number of overflow pages */
} mp_pb;
- indx_t mp_ptrs[1]; /* dynamic size */
+ indx_t mp_ptrs[1]; /**< dynamic size */
} MDB_page;
+ /** Size of the page header, excluding dynamic data at the end */
#define PAGEHDRSZ ((unsigned) offsetof(MDB_page, mp_ptrs))
+ /** Address of first usable data byte in a page, after the header */
+#define METADATA(p) ((void *)((char *)(p) + PAGEHDRSZ))
+
+ /** Number of nodes on a page */
#define NUMKEYS(p) (((p)->mp_lower - PAGEHDRSZ) >> 1)
+
+ /** The amount of space remaining in the page */
#define SIZELEFT(p) (indx_t)((p)->mp_upper - (p)->mp_lower)
+
+ /** The percentage of space used in the page, in tenths of a percent. */
#define PAGEFILL(env, p) (1000L * ((env)->me_psize - PAGEHDRSZ - SIZELEFT(p)) / \
((env)->me_psize - PAGEHDRSZ))
+ /** The minimum page fill factor, in tenths of a percent.
+ * Pages emptier than this are candidates for merging.
+ */
+#define FILL_THRESHOLD 250
+
+ /** Test if a page is a leaf page */
#define IS_LEAF(p) F_ISSET((p)->mp_flags, P_LEAF)
+ /** Test if a page is a LEAF2 page */
#define IS_LEAF2(p) F_ISSET((p)->mp_flags, P_LEAF2)
+ /** Test if a page is a branch 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)
+ /** The number of overflow pages needed to store the given size. */
#define OVPAGES(size, psize) ((PAGEHDRSZ-1 + (size)) / (psize) + 1)
-typedef struct MDB_db {
- uint32_t md_pad; /* also ksize for LEAF2 pages */
- uint16_t md_flags;
- uint16_t md_depth;
- ULONG md_branch_pages;
- ULONG md_leaf_pages;
- ULONG md_overflow_pages;
- ULONG md_entries;
- pgno_t md_root;
-} MDB_db;
+ /** Header for a single key/data pair within a page.
+ * We guarantee 2-byte alignment for nodes.
+ */
+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)
+ */
+ unsigned short mn_lo;
+ unsigned short mn_hi; /**< part of dsize or pgno */
+ unsigned short mn_flags; /**< flags for special node types */
+#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 */
+ unsigned short mn_ksize; /**< key size */
+ char mn_data[1]; /**< key and data are appended here */
+} MDB_node;
-#define FREE_DBI 0
-#define MAIN_DBI 1
+ /** Size of the node header, excluding dynamic data at the end */
+#define NODESIZE offsetof(MDB_node, mn_data)
-typedef struct MDB_meta { /* meta (footer) page content */
- uint32_t mm_magic;
- uint32_t mm_version;
- void *mm_address; /* address for fixed mapping */
- size_t mm_mapsize; /* size of mmap region */
- MDB_db mm_dbs[2]; /* first is free space, 2nd is main db */
-#define mm_psize mm_dbs[0].md_pad
-#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 */
-} MDB_meta;
+ /** 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))
-typedef struct MDB_dhead { /* a dirty page */
- MDB_page *md_parent;
- unsigned md_pi; /* parent index */
- int md_num;
-} MDB_dhead;
+ /** 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)
-typedef struct MDB_dpage {
- MDB_dhead h;
- MDB_page p;
-} MDB_dpage;
+ /** Address of node \b i in page \b p */
+#define NODEPTR(p, i) ((MDB_node *)((char *)(p) + (p)->mp_ptrs[i]))
-typedef struct MDB_oldpages {
- struct MDB_oldpages *mo_next;
- ULONG mo_txnid;
- pgno_t mo_pages[1]; /* dynamic */
-} MDB_oldpages;
+ /** Address of the key for the node */
+#define NODEKEY(node) (void *)((node)->mn_data)
-typedef struct MDB_pageparent {
- MDB_page *mp_page;
- MDB_page *mp_parent;
- unsigned mp_pi;
-} MDB_pageparent;
+ /** Address of the data for a node */
+#define NODEDATA(node) (void *)((char *)(node)->mn_data + (node)->mn_ksize)
-static MDB_dpage *mdb_alloc_page(MDB_txn *txn, MDB_dbi dbi, MDB_page *parent, unsigned int parent_idx, int num);
-static int mdb_touch(MDB_txn *txn, MDB_dbi dbi, MDB_pageparent *mp);
+ /** 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))
+ /** 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
-typedef struct MDB_ppage { /* ordered list of pages */
- MDB_page *mp_page;
- unsigned int mp_ki; /* cursor index on page */
-} MDB_ppage;
+ /** Get the size of the data in a leaf node */
+#define NODEDSZ(node) ((node)->mn_lo | ((unsigned)(node)->mn_hi << 16))
+ /** Set the size of the data for a leaf node */
+#define SETDSZ(node,size) do { \
+ (node)->mn_lo = (size) & 0xffff; (node)->mn_hi = (size) >> 16;} while(0)
+ /** The size of a key in a node */
+#define NODEKSZ(node) ((node)->mn_ksize)
-#define CURSOR_TOP(c) (&(c)->mc_stack[(c)->mc_snum-1])
-#define CURSOR_PARENT(c) (&(c)->mc_stack[(c)->mc_snum-2])
-#define CURSOR_STACK 32
+ /** 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.
+ */
+#define LEAF2KEY(p, i, ks) ((char *)(p) + PAGEHDRSZ + ((i)*(ks)))
-struct MDB_xcursor;
+ /** 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); } }
-struct MDB_cursor {
- MDB_txn *mc_txn;
- MDB_ppage mc_stack[CURSOR_STACK]; /* stack of parent pages */
- unsigned int mc_snum; /* number of pushed pages */
- MDB_dbi mc_dbi;
- short mc_initialized; /* 1 if initialized */
- short mc_eof; /* 1 if end is reached */
- struct MDB_xcursor *mc_xcursor;
-};
+ /** 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_root; /**< the root page of this tree */
+} MDB_db;
-#define METADATA(p) ((void *)((char *)(p) + PAGEHDRSZ))
+ /** Handle for the DB used to track free pages. */
+#define FREE_DBI 0
+ /** Handle for the default DB. */
+#define MAIN_DBI 1
-typedef struct MDB_node {
-#define mn_pgno mn_p.np_pgno
-#define mn_dsize mn_p.np_dsize
- union {
- pgno_t np_pgno; /* child page number */
- uint32_t np_dsize; /* leaf data size */
- } mn_p;
- unsigned int mn_flags:4;
- unsigned int mn_ksize:12; /* key size */
-#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 */
- char mn_data[1];
-} MDB_node;
+ /** Meta page content. */
+typedef struct MDB_meta {
+ /** Stamp identifying this as an MDB data file. It must be set
+ * to #MDB_MAGIC. */
+ uint32_t mm_magic;
+ /** Version number of this lock file. Must be set to #MDB_VERSION. */
+ uint32_t mm_version;
+ void *mm_address; /**< address for fixed mapping */
+ size_t mm_mapsize; /**< size of mmap region */
+ MDB_db mm_dbs[2]; /**< first is free space, 2nd is main db */
+ /** The size of pages used in this DB */
+#define mm_psize mm_dbs[0].md_pad
+ /** 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 */
+} MDB_meta;
+ /** 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;
- MDB_cmp_func *md_cmp; /* user compare function */
- MDB_cmp_func *md_dcmp; /* user dupsort function */
- MDB_rel_func *md_rel; /* user relocate function */
- MDB_dbi md_parent;
- unsigned int md_dirty;
+ 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 */
} MDB_dbx;
+ /** A database transaction.
+ * Every operation requires a transaction handle.
+ */
struct MDB_txn {
- pgno_t mt_next_pgno; /* next unallocated page */
+ 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;
- MDB_env *mt_env;
- pgno_t *mt_free_pgs; /* this is an IDL */
+ 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;
union {
- ID2L dirty_list; /* modified pages */
- MDB_reader *reader;
+ ID2L dirty_list; /**< modified pages */
+ MDB_reader *reader; /**< this thread's slot in the reader table */
} mt_u;
- MDB_dbx *mt_dbxs; /* array */
+ /** Array of records for each DB known in the environment. */
+ MDB_dbx *mt_dbxs;
+ /** Array of MDB_db records for each known DB */
MDB_db *mt_dbs;
+ /** 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;
-#define MDB_TXN_RDONLY 0x01 /* read-only transaction */
-#define MDB_TXN_ERROR 0x02 /* an error has occurred */
+#define MDB_TXN_RDONLY 0x01 /**< read-only transaction */
+#define MDB_TXN_ERROR 0x02 /**< an error has occurred */
unsigned int mt_flags;
+ /** Tracks which of the two meta pages was used at the start
+ * of this transaction.
+ */
unsigned int mt_toggle;
};
-/* Context for sorted-dup records */
+/** Enough space for 2^32 nodes with minimum of 2 keys per node. I.e., plenty.
+ * At 4 keys per node, enough for 2^64 nodes, so there's probably no need to
+ * raise this on a 64 bit machine.
+ */
+#define CURSOR_STACK 32
+
+struct MDB_xcursor;
+
+ /** Cursors are used for all DB operations */
+struct MDB_cursor {
+ /** 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;
+ unsigned short mc_snum; /**< number of pushed pages */
+ unsigned short mc_top; /**< index of top page, mc_snum-1 */
+ unsigned int mc_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 */
+ MDB_page *mc_pg[CURSOR_STACK]; /**< stack of pushed pages */
+ indx_t mc_ki[CURSOR_STACK]; /**< stack of page indices */
+};
+
+ /** Context for sorted-dup records.
+ * We could have gone to a fully recursive design, with arbitrarily
+ * deep nesting of sub-databases. But for now we only handle these
+ * levels - main DB, optional sub-DB, sorted-duplicate DB.
+ */
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];
} MDB_xcursor;
+ /** A set of pages freed by an earlier transaction. */
+typedef struct MDB_oldpages {
+ /** Usually we only read one record from the FREEDB at a time, but
+ * in case we read more, this will chain them together.
+ */
+ struct MDB_oldpages *mo_next;
+ /** The ID of the transaction in which these pages were freed. */
+ ULONG mo_txnid;
+ /** An #IDL of the pages */
+ pgno_t mo_pages[1]; /* dynamic */
+} MDB_oldpages;
+
+ /** The database environment. */
struct MDB_env {
- HANDLE me_fd;
- HANDLE me_lfd;
- HANDLE me_mfd; /* just for writing the meta pages */
+ 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_extrapad; /* unused for now */
- unsigned int me_maxreaders;
- unsigned int me_numdbs;
- unsigned int me_maxdbs;
- char *me_path;
- char *me_map;
- MDB_txninfo *me_txns;
- MDB_meta *me_metas[2];
- MDB_txn *me_txn; /* current write transaction */
- size_t me_mapsize;
- off_t me_size; /* current file size */
- pgno_t me_maxpg; /* me_mapsize / me_psize */
- unsigned int me_psize;
- unsigned int me_db_toggle;
- MDB_dbx *me_dbxs; /* array */
- MDB_db *me_dbs[2];
- MDB_oldpages *me_pghead;
- pthread_key_t me_txkey; /* thread-key for readers */
- MDB_dpage *me_dpages;
+ 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 */
+ 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 */
+ MDB_meta *me_metas[2]; /**< pointers to the two meta pages */
+ MDB_txn *me_txn; /**< current write transaction */
+ 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 #GET_PAGESIZE */
+ unsigned int me_db_toggle; /**< which DB table is current */
+ 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];
- ID2 me_dirty_list[MDB_IDL_DB_SIZE];
- LAZY_RWLOCK_DEF(me_dblock);
+ /** 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)
#ifdef _WIN32
HANDLE me_rmutex; /* Windows mutexes don't reside in shared mem */
HANDLE me_wmutex;
#endif
};
+ /** max number of pages to commit in one writev() call */
+#define MDB_COMMIT_PAGES 64
-#define NODESIZE offsetof(MDB_node, mn_data)
-
-#define INDXSIZE(k) (NODESIZE + ((k) == NULL ? 0 : (k)->mv_size))
-#define LEAFSIZE(k, d) (NODESIZE + (k)->mv_size + (d)->mv_size)
-#define NODEPTR(p, i) ((MDB_node *)((char *)(p) + (p)->mp_ptrs[i]))
-#define NODEKEY(node) (void *)((node)->mn_data)
-#define NODEDATA(node) (void *)((char *)(node)->mn_data + (node)->mn_ksize)
-#define NODEPGNO(node) ((node)->mn_pgno)
-#define NODEDSZ(node) ((node)->mn_dsize)
-#define NODEKSZ(node) ((node)->mn_ksize)
-#define LEAF2KEY(p, i, ks) ((char *)(p) + PAGEHDRSZ + ((i)*(ks)))
-
-#define MDB_SET_KEY(node, key) if (key!=NULL) {(key)->mv_size = NODEKSZ(node); (key)->mv_data = NODEKEY(node);}
+static MDB_page *mdb_alloc_page(MDB_cursor *mc, int num);
+static int mdb_touch(MDB_cursor *mc);
-#define MDB_COMMIT_PAGES 64 /* max number of pages to write in one commit */
-
-static int mdb_search_page_root(MDB_txn *txn,
- MDB_dbi dbi, MDB_val *key,
- MDB_cursor *cursor, int modify,
- MDB_pageparent *mpp);
-static int mdb_search_page(MDB_txn *txn,
- MDB_dbi dbi, MDB_val *key,
- MDB_cursor *cursor, int modify,
- MDB_pageparent *mpp);
+static int mdb_search_page_root(MDB_cursor *mc,
+ MDB_val *key, int modify);
+static int mdb_search_page(MDB_cursor *mc,
+ MDB_val *key, int modify);
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_txn *txn, MDB_dbi dbi, MDB_page *mp,
- MDB_val *key, int *exactp, unsigned int *kip);
-static int mdb_add_node(MDB_txn *txn, MDB_dbi dbi, MDB_page *mp,
- indx_t indx, MDB_val *key, MDB_val *data,
- pgno_t pgno, uint8_t flags);
+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_txn *txn, MDB_dbi dbi, unsigned int ki,
- MDB_pageparent *mpp, MDB_node *leaf);
-static int mdb_put0(MDB_txn *txn, MDB_dbi dbi,
- MDB_val *key, MDB_val *data, unsigned int flags);
+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 int mdb_rebalance(MDB_txn *txn, MDB_dbi dbi, MDB_pageparent *mp);
-static int mdb_update_key(MDB_page *mp, indx_t indx, MDB_val *key);
-static int mdb_move_node(MDB_txn *txn, MDB_dbi dbi,
- MDB_pageparent *src, indx_t srcindx,
- MDB_pageparent *dst, indx_t dstindx);
-static int mdb_merge(MDB_txn *txn, MDB_dbi dbi, MDB_pageparent *src,
- MDB_pageparent *dst);
-static int mdb_split(MDB_txn *txn, MDB_dbi dbi, MDB_page **mpp,
- unsigned int *newindxp, MDB_val *newkey,
- MDB_val *newdata, pgno_t newpgno);
-static MDB_dpage *mdb_new_page(MDB_txn *txn, MDB_dbi dbi, uint32_t flags, int num);
-
-static void cursor_pop_page(MDB_cursor *cursor);
-static MDB_ppage *cursor_push_page(MDB_cursor *cursor,
- MDB_page *mp);
-
-static int mdb_sibling(MDB_cursor *cursor, int move_right);
-static int mdb_cursor_next(MDB_cursor *cursor,
- MDB_val *key, MDB_val *data, MDB_cursor_op op);
-static int mdb_cursor_prev(MDB_cursor *cursor,
- MDB_val *key, MDB_val *data, MDB_cursor_op op);
-static int mdb_cursor_set(MDB_cursor *cursor,
- MDB_val *key, MDB_val *data, MDB_cursor_op op, int *exactp);
-static int mdb_cursor_first(MDB_cursor *cursor,
- MDB_val *key, MDB_val *data);
-static int mdb_cursor_last(MDB_cursor *cursor,
- MDB_val *key, MDB_val *data);
-
-static void mdb_xcursor_init0(MDB_cursor *mc);
-static void mdb_xcursor_init1(MDB_txn *txn, MDB_dbi dbi, MDB_xcursor *mx,
- MDB_page *mp, MDB_node *node);
-static void mdb_xcursor_fini(MDB_txn *txn, MDB_dbi dbi, MDB_xcursor *mx);
-
-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 memncmp(const void *s1, size_t n1,
- const void *s2, size_t n2);
-static int memnrcmp(const void *s1, size_t n1,
- const void *s2, size_t n2);
+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 *csrc, MDB_cursor *cdst);
+static int mdb_merge(MDB_cursor *csrc, MDB_cursor *cdst);
+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);
-#ifdef _WIN32
-static SECURITY_DESCRIPTOR mdb_null_sd;
-static SECURITY_ATTRIBUTES mdb_all_sa;
-static int mdb_sec_inited;
-#endif
+static void cursor_pop_page(MDB_cursor *mc);
+static int cursor_push_page(MDB_cursor *mc, MDB_page *mp);
-static int
-memncmp(const void *s1, size_t n1, const void *s2, size_t n2)
-{
- int diff, len_diff = -1;
+static int mdb_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,
+ int *exactp);
+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);
- if (n1 >= n2) {
- len_diff = (n1 > n2);
- n1 = n2;
- }
- diff = memcmp(s1, s2, n1);
- return diff ? diff : len_diff;
-}
+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 int
-memnrcmp(const void *s1, size_t n1, const void *s2, size_t n2)
-{
- const unsigned char *p1, *p2, *p1_lim;
+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);
- if (n2 == 0)
- return n1 != 0;
- if (n1 == 0)
- return -1;
+static void mdb_default_cmp(MDB_txn *txn, MDB_dbi dbi);
- p1 = (const unsigned char *)s1 + n1 - 1;
- p2 = (const unsigned char *)s2 + n2 - 1;
+/** @cond */
+static MDB_cmp_func memncmp, memnrcmp, intcmp, cintcmp;
+/** @endcond */
- for (p1_lim = (n1 <= n2 ? s1 : s2); *p1 == *p2; p1--, p2--) {
- if (p1 == p1_lim)
- return (p1 != s1) ? (p1 != p2) : (p2 != s2) ? -1 : 0;
- }
- return *p1 - *p2;
-}
+#ifdef _WIN32
+static SECURITY_DESCRIPTOR mdb_null_sd;
+static SECURITY_ATTRIBUTES mdb_all_sa;
+static int mdb_sec_inited;
+#endif
+/** Return the library version info. */
char *
-mdb_version(int *maj, int *min, int *pat)
+mdb_version(int *major, int *minor, int *patch)
{
- if (maj) *maj = MDB_VERSION_MAJOR;
- if (min) *min = MDB_VERSION_MINOR;
- if (pat) *pat = MDB_VERSION_PATCH;
+ if (major) *major = MDB_VERSION_MAJOR;
+ if (minor) *minor = MDB_VERSION_MINOR;
+ if (patch) *patch = MDB_VERSION_PATCH;
return MDB_VERSION_STRING;
}
+/** 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",
}
#if DEBUG
-static char *
+/** 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;
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
for (i=0; i<key->mv_size; i++)
ptr += sprintf(ptr, "%02x", *c++);
+#else
+ sprintf(buf, "%.*s", key->mv_size, key->mv_data);
+#endif
return buf;
}
#endif
int
mdb_cmp(MDB_txn *txn, MDB_dbi dbi, const MDB_val *a, const MDB_val *b)
{
- if (txn->mt_dbxs[dbi].md_cmp)
- return txn->mt_dbxs[dbi].md_cmp(a, b);
-
- if (txn->mt_dbs[dbi].md_flags & (MDB_REVERSEKEY
-#if __BYTE_ORDER == __LITTLE_ENDIAN
- |MDB_INTEGERKEY
-#endif
- ))
- return memnrcmp(a->mv_data, a->mv_size, b->mv_data, b->mv_size);
- else
- return memncmp((char *)a->mv_data, a->mv_size, b->mv_data, b->mv_size);
+ return txn->mt_dbxs[dbi].md_cmp(a, b);
}
+/** Compare two data items according to a particular database.
+ * This returns a comparison as if the two items were data items of
+ * a sorted duplicates #MDB_DUPSORT database.
+ * @param[in] txn A transaction handle returned by #mdb_txn_begin()
+ * @param[in] dbi A database handle returned by #mdb_open()
+ * @param[in] a The first item to compare
+ * @param[in] b The second item to compare
+ * @return < 0 if a < b, 0 if a == b, > 0 if a > b
+ */
int
mdb_dcmp(MDB_txn *txn, MDB_dbi dbi, const MDB_val *a, const MDB_val *b)
{
if (txn->mt_dbxs[dbi].md_dcmp)
return txn->mt_dbxs[dbi].md_dcmp(a, b);
-
- if (txn->mt_dbs[dbi].md_flags & (0
-#if __BYTE_ORDER == __LITTLE_ENDIAN
- |MDB_INTEGERDUP
-#endif
- ))
- return memnrcmp(a->mv_data, a->mv_size, b->mv_data, b->mv_size);
else
- return memncmp((char *)a->mv_data, a->mv_size, b->mv_data, b->mv_size);
+ return EINVAL; /* too bad you can't distinguish this from a valid result */
}
-/* Allocate new page(s) for writing */
-static MDB_dpage *
-mdb_alloc_page(MDB_txn *txn, MDB_dbi dbi, MDB_page *parent, unsigned int parent_idx, int num)
+/** 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_dpage *dp;
+ MDB_txn *txn = mc->mc_txn;
+ MDB_page *np;
pgno_t pgno = P_INVALID;
ID2 mid;
if (txn->mt_txnid > 2) {
- if (!txn->mt_env->me_pghead && dbi != FREE_DBI &&
+ if (!txn->mt_env->me_pghead && mc->mc_dbi != FREE_DBI &&
txn->mt_dbs[FREE_DBI].md_root != P_INVALID) {
/* See if there's anything in the free DB */
- MDB_pageparent mpp;
+ MDB_cursor m2;
MDB_node *leaf;
ULONG *kptr, oldest;
- mpp.mp_parent = NULL;
- mpp.mp_pi = 0;
- mdb_search_page(txn, FREE_DBI, NULL, NULL, 0, &mpp);
- leaf = NODEPTR(mpp.mp_page, 0);
+ 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);
{
}
#endif
/* drop this IDL from the DB */
- mpp.mp_parent = NULL;
- mpp.mp_pi = 0;
- mdb_search_page(txn, FREE_DBI, NULL, NULL, 1, &mpp);
- leaf = NODEPTR(mpp.mp_page, 0);
- mdb_del0(txn, FREE_DBI, 0, &mpp, leaf);
+ m2.mc_ki[m2.mc_top] = 0;
+ m2.mc_flags = C_INITIALIZED;
+ mdb_cursor_del(&m2, 0);
}
}
if (txn->mt_env->me_pghead) {
return NULL;
}
if (txn->mt_env->me_dpages && num == 1) {
- dp = txn->mt_env->me_dpages;
- txn->mt_env->me_dpages = (MDB_dpage *)dp->h.md_parent;
+ np = txn->mt_env->me_dpages;
+ txn->mt_env->me_dpages = np->mp_next;
} else {
- if ((dp = malloc(txn->mt_env->me_psize * num + sizeof(MDB_dhead))) == NULL)
+ if ((np = malloc(txn->mt_env->me_psize * num )) == NULL)
return NULL;
}
- dp->h.md_num = num;
- dp->h.md_parent = parent;
- dp->h.md_pi = parent_idx;
if (pgno == P_INVALID) {
- dp->p.mp_pgno = txn->mt_next_pgno;
+ np->mp_pgno = txn->mt_next_pgno;
txn->mt_next_pgno += num;
} else {
- dp->p.mp_pgno = pgno;
+ np->mp_pgno = pgno;
}
- mid.mid = dp->p.mp_pgno;
- mid.mptr = dp;
+ mid.mid = np->mp_pgno;
+ mid.mptr = np;
mdb_mid2l_insert(txn->mt_u.dirty_list, &mid);
- return dp;
+ 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_txn *txn, MDB_dbi dbi, MDB_pageparent *pp)
+mdb_touch(MDB_cursor *mc)
{
- MDB_page *mp = pp->mp_page;
+ MDB_page *mp = mc->mc_pg[mc->mc_top];
pgno_t pgno;
- assert(txn != NULL);
- assert(pp != NULL);
if (!F_ISSET(mp->mp_flags, P_DIRTY)) {
- MDB_dpage *dp;
- if ((dp = mdb_alloc_page(txn, dbi, pp->mp_parent, pp->mp_pi, 1)) == NULL)
+ MDB_page *np;
+ if ((np = mdb_alloc_page(mc, 1)) == NULL)
return ENOMEM;
- DPRINTF("touched db %u page %lu -> %lu", dbi, mp->mp_pgno, dp->p.mp_pgno);
- assert(mp->mp_pgno != dp->p.mp_pgno);
- mdb_midl_insert(txn->mt_free_pgs, mp->mp_pgno);
- pgno = dp->p.mp_pgno;
- memcpy(&dp->p, mp, txn->mt_env->me_psize);
- mp = &dp->p;
+ DPRINTF("touched db %u page %lu -> %lu", 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);
+ 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;
- /* Update the page number to new touched page. */
- if (pp->mp_parent != NULL)
- NODEPGNO(NODEPTR(pp->mp_parent, pp->mp_pi)) = mp->mp_pgno;
- pp->mp_page = mp;
+ mc->mc_pg[mc->mc_top] = mp;
+ /** 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);
}
return 0;
}
static inline void
mdb_txn_reset0(MDB_txn *txn);
+/** 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 inline int
mdb_txn_renew0(MDB_txn *txn)
{
return rc;
}
+/** Common code for #mdb_txn_reset() and #mdb_txn_abort().
+ * @param[in] txn the transaction handle to reset
+ */
static inline void
mdb_txn_reset0(MDB_txn *txn)
{
txn->mt_u.reader->mr_txnid = 0;
} else {
MDB_oldpages *mop;
- MDB_dpage *dp;
+ MDB_page *dp;
unsigned int i;
/* 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 (dp->h.md_num == 1) {
- dp->h.md_parent = (MDB_page *)txn->mt_env->me_dpages;
+ if (!IS_OVERFLOW(dp) || dp->mp_pages == 1) {
+ dp->mp_next = txn->mt_env->me_dpages;
txn->mt_env->me_dpages = dp;
} else {
/* large pages just get freed directly */
env->me_txn = NULL;
for (i=2; i<env->me_numdbs; i++)
env->me_dbxs[i].md_dirty = 0;
+ /* The writer mutex was locked in mdb_txn_begin. */
UNLOCK_MUTEX_W(env);
}
}
unsigned int i;
ssize_t rc;
off_t size;
- MDB_dpage *dp;
+ MDB_page *dp;
MDB_env *env;
pgno_t next;
+ MDB_cursor mc;
assert(txn != NULL);
assert(txn->mt_env != NULL);
DPRINTF("committing txn %lu %p on mdbenv %p, root page %lu",
txn->mt_txnid, txn, (void *)env, txn->mt_dbs[MAIN_DBI].md_root);
+ mc.mc_txn = txn;
+ mc.mc_dbi = FREE_DBI;
+ mc.mc_flags = 0;
+
/* should only be one record now */
if (env->me_pghead) {
- MDB_pageparent mpp;
-
/* make sure first page of freeDB is touched and on freelist */
- mpp.mp_parent = NULL;
- mpp.mp_pi = 0;
- mdb_search_page(txn, FREE_DBI, NULL, NULL, 1, &mpp);
+ mdb_search_page(&mc, NULL, 1);
}
/* save to free list */
if (!MDB_IDL_IS_ZERO(txn->mt_free_pgs)) {
MDB_val key, data;
- MDB_pageparent mpp;
ULONG i;
/* make sure last page of freeDB is touched and on freelist */
key.mv_size = MAXKEYSIZE+1;
key.mv_data = NULL;
- mpp.mp_parent = NULL;
- mpp.mp_pi = 0;
- mdb_search_page(txn, FREE_DBI, &key, NULL, 1, &mpp);
+ mdb_search_page(&mc, &key, 1);
+ mdb_midl_sort(txn->mt_free_pgs);
#if DEBUG > 1
{
unsigned int i;
do {
i = txn->mt_free_pgs[0];
data.mv_size = MDB_IDL_SIZEOF(txn->mt_free_pgs);
- rc = mdb_put0(txn, FREE_DBI, &key, &data, 0);
+ rc = mdb_cursor_put(&mc, &key, &data, 0);
if (rc) {
mdb_txn_abort(txn);
return rc;
key.mv_data = (char *)&mop->mo_txnid;
data.mv_size = MDB_IDL_SIZEOF(mop->mo_pages);
data.mv_data = mop->mo_pages;
- mdb_put0(txn, FREE_DBI, &key, &data, 0);
+ mdb_cursor_put(&mc, &key, &data, 0);
free(env->me_pghead);
env->me_pghead = NULL;
}
MDB_val data;
data.mv_size = sizeof(MDB_db);
+ mc.mc_dbi = MAIN_DBI;
+ mc.mc_flags = 0;
for (i = 2; i < txn->mt_numdbs; i++) {
if (txn->mt_dbxs[i].md_dirty) {
data.mv_data = &txn->mt_dbs[i];
- mdb_put0(txn, MAIN_DBI, &txn->mt_dbxs[i].md_name, &data, 0);
+ mdb_cursor_put(&mc, &txn->mt_dbxs[i].md_name, &data, 0);
}
}
}
OVERLAPPED ov;
memset(&ov, 0, sizeof(ov));
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->p.mp_pgno);
- size = dp->p.mp_pgno * env->me_psize;
+ DPRINTF("committing page %lu", dp->mp_pgno);
+ size = dp->mp_pgno * env->me_psize;
ov.Offset = size & 0xffffffff;
ov.OffsetHigh = size >> 16;
ov.OffsetHigh >>= 16;
/* clear dirty flag */
- dp->p.mp_flags &= ~P_DIRTY;
- rc = WriteFile(env->me_fd, &dp->p, env->me_psize * dp->h.md_num, NULL, &ov);
+ dp->mp_flags &= ~P_DIRTY;
+ wsize = env->me_psize;
+ if (IS_OVERFLOW(dp)) wsize *= dp->mp_pages;
+ rc = WriteFile(env->me_fd, dp, wsize, NULL, &ov);
if (!rc) {
n = ErrCode();
DPRINTF("WriteFile: %d", n);
size = 0;
for (; i<=txn->mt_u.dirty_list[0].mid; i++) {
dp = txn->mt_u.dirty_list[i].mptr;
- if (dp->p.mp_pgno != next) {
+ if (dp->mp_pgno != next) {
if (n) {
DPRINTF("committing %u dirty pages", n);
rc = writev(env->me_fd, iov, n);
n = 0;
size = 0;
}
- lseek(env->me_fd, dp->p.mp_pgno * env->me_psize, SEEK_SET);
- next = dp->p.mp_pgno;
+ lseek(env->me_fd, dp->mp_pgno * env->me_psize, SEEK_SET);
+ next = dp->mp_pgno;
}
- DPRINTF("committing page %lu", dp->p.mp_pgno);
- iov[n].iov_len = env->me_psize * dp->h.md_num;
- iov[n].iov_base = &dp->p;
+ DPRINTF("committing page %lu", 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;
size += iov[n].iov_len;
- next = dp->p.mp_pgno + dp->h.md_num;
+ next = dp->mp_pgno + (IS_OVERFLOW(dp) ? dp->mp_pages : 1);
/* clear dirty flag */
- dp->p.mp_flags &= ~P_DIRTY;
+ dp->mp_flags &= ~P_DIRTY;
if (++n >= MDB_COMMIT_PAGES) {
done = 0;
i++;
*/
for (i=1; i<=txn->mt_u.dirty_list[0].mid; i++) {
dp = txn->mt_u.dirty_list[i].mptr;
- if (dp->h.md_num == 1) {
- dp->h.md_parent = (MDB_page *)txn->mt_env->me_dpages;
+ if (!IS_OVERFLOW(dp) || dp->mp_pages == 1) {
+ dp->mp_next = txn->mt_env->me_dpages;
txn->mt_env->me_dpages = dp;
} else {
free(dp);
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)
{
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)
{
DPUTS("writing new meta page");
- GetPageSize(psize);
+ GET_PAGESIZE(psize);
meta->mm_magic = MDB_MAGIC;
meta->mm_version = MDB_VERSION;
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)
{
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)
{
return MDB_SUCCESS;
}
+/** Further setup required for opening an MDB environment
+ */
static int
mdb_env_open2(MDB_env *env, unsigned int flags)
{
}
+ /** The name of the lock file in the DB environment */
#define LOCKNAME "/lock.mdb"
+ /** The name of the data file in the DB environment */
#define DATANAME "/data.mdb"
int
mdb_env_open(MDB_env *env, const char *path, unsigned int flags, mode_t mode)
void
mdb_env_close(MDB_env *env)
{
- MDB_dpage *dp;
+ MDB_page *dp;
if (env == NULL)
return;
while (env->me_dpages) {
dp = env->me_dpages;
- env->me_dpages = (MDB_dpage *)dp->h.md_parent;
+ env->me_dpages = dp->mp_next;
free(dp);
}
free(env);
}
+/* only for aligned ints */
+static int
+intcmp(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;
+ }
+}
+
+/* ints must always be the same size */
+static int
+cintcmp(const MDB_val *a, const MDB_val *b)
+{
+#if __BYTE_ORDER == __LITTLE_ENDIAN
+ unsigned short *u, *c;
+ int x;
+
+ 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);
+ return x;
+#else
+ return memcmp(a->mv_data, b->mv_data, a->mv_size);
+#endif
+}
+
+static int
+memncmp(const MDB_val *a, const MDB_val *b)
+{
+ int diff, len_diff;
+ unsigned int len;
+
+ len = a->mv_size;
+ len_diff = a->mv_size - b->mv_size;
+ if (len_diff > 0)
+ len = b->mv_size;
+ diff = memcmp(a->mv_data, b->mv_data, len);
+ return diff ? diff : len_diff;
+}
+
+static int
+memnrcmp(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;
+
+ p1 = (const unsigned char *)a->mv_data + a->mv_size - 1;
+ p2 = (const unsigned char *)b->mv_data + b->mv_size - 1;
+
+ 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;
+
+ while (p1 > p1_lim) {
+ diff = *p1 - *p2;
+ if (diff)
+ return diff;
+ p1--;
+ p2--;
+ }
+ return len_diff;
+}
+
/* Search for key within a leaf 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
* If no entry larger or equal to the key is found, returns NULL.
*/
static MDB_node *
-mdb_search_node(MDB_txn *txn, MDB_dbi dbi, MDB_page *mp, MDB_val *key,
- int *exactp, unsigned int *kip)
+mdb_search_node(MDB_cursor *mc, MDB_val *key, int *exactp)
{
- unsigned int i = 0;
+ unsigned int i = 0, nkeys;
int low, high;
int rc = 0;
+ MDB_page *mp = mc->mc_pg[mc->mc_top];
MDB_node *node = NULL;
MDB_val nodekey;
+ MDB_cmp_func *cmp;
DKBUF;
+ nkeys = NUMKEYS(mp);
+
DPRINTF("searching %u keys in %s page %lu",
- NUMKEYS(mp),
- IS_LEAF(mp) ? "leaf" : "branch",
+ nkeys, IS_LEAF(mp) ? "leaf" : "branch",
mp->mp_pgno);
- assert(NUMKEYS(mp) > 0);
-
- memset(&nodekey, 0, sizeof(nodekey));
+ assert(nkeys > 0);
low = IS_LEAF(mp) ? 0 : 1;
- high = NUMKEYS(mp) - 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 */
+ }
while (low <= high) {
i = (low + high) >> 1;
if (IS_LEAF2(mp)) {
- nodekey.mv_size = txn->mt_dbs[dbi].md_pad;
nodekey.mv_data = LEAF2KEY(mp, i, nodekey.mv_size);
} else {
node = NODEPTR(mp, i);
nodekey.mv_data = NODEKEY(node);
}
- rc = mdb_cmp(txn, dbi, key, &nodekey);
+ rc = cmp(key, &nodekey);
+#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);
+#endif
if (rc == 0)
break;
if (rc > 0) { /* Found entry is less than the key. */
i++; /* Skip to get the smallest entry larger than key. */
- if (i >= NUMKEYS(mp))
- /* There is no entry larger or equal to the key. */
- return NULL;
+ if (!IS_LEAF2(mp))
+ node = NODEPTR(mp, i);
}
if (exactp)
*exactp = (rc == 0);
- if (kip) /* Store the key index if requested. */
- *kip = i;
+ /* store the key index */
+ mc->mc_ki[mc->mc_top] = i;
+ if (i >= nkeys)
+ /* There is no entry larger or equal to the key. */
+ return NULL;
/* nodeptr is fake for LEAF2 */
- return IS_LEAF2(mp) ? NODEPTR(mp, 0) : NODEPTR(mp, i);
+ return node;
}
static void
-cursor_pop_page(MDB_cursor *cursor)
+cursor_pop_page(MDB_cursor *mc)
{
- MDB_ppage *top;
+ MDB_page *top;
- if (cursor->mc_snum) {
- top = CURSOR_TOP(cursor);
- cursor->mc_snum--;
+ if (mc->mc_snum) {
+ top = mc->mc_pg[mc->mc_top];
+ mc->mc_snum--;
+ if (mc->mc_snum)
+ mc->mc_top--;
- DPRINTF("popped page %lu off db %u cursor %p", top->mp_page->mp_pgno,
- cursor->mc_dbi, (void *) cursor);
+ DPRINTF("popped page %lu off db %u cursor %p", top->mp_pgno,
+ mc->mc_dbi, (void *) mc);
}
}
-static MDB_ppage *
-cursor_push_page(MDB_cursor *cursor, MDB_page *mp)
+static int
+cursor_push_page(MDB_cursor *mc, MDB_page *mp)
{
- MDB_ppage *ppage;
-
DPRINTF("pushing page %lu on db %u cursor %p", mp->mp_pgno,
- cursor->mc_dbi, (void *) cursor);
+ mc->mc_dbi, (void *) mc);
- assert(cursor->mc_snum < CURSOR_STACK);
+ if (mc->mc_snum >= CURSOR_STACK)
+ return ENOMEM;
+
+ mc->mc_top = mc->mc_snum++;
+ mc->mc_pg[mc->mc_top] = mp;
+ mc->mc_ki[mc->mc_top] = 0;
- ppage = &cursor->mc_stack[cursor->mc_snum++];
- ppage->mp_page = mp;
- ppage->mp_ki = 0;
- return ppage;
+ return MDB_SUCCESS;
}
static int
MDB_page *p = NULL;
if (!F_ISSET(txn->mt_flags, MDB_TXN_RDONLY) && txn->mt_u.dirty_list[0].mid) {
- MDB_dpage *dp;
unsigned x;
x = mdb_mid2l_search(txn->mt_u.dirty_list, pgno);
if (x <= txn->mt_u.dirty_list[0].mid && txn->mt_u.dirty_list[x].mid == pgno) {
- dp = txn->mt_u.dirty_list[x].mptr;
- p = &dp->p;
+ p = txn->mt_u.dirty_list[x].mptr;
}
}
if (!p) {
}
static int
-mdb_search_page_root(MDB_txn *txn, MDB_dbi dbi, MDB_val *key,
- MDB_cursor *cursor, int modify, MDB_pageparent *mpp)
+mdb_search_page_root(MDB_cursor *mc, MDB_val *key, int modify)
{
- MDB_page *mp = mpp->mp_page;
+ MDB_page *mp = mc->mc_pg[mc->mc_top];
DKBUF;
int rc;
- if (cursor && cursor_push_page(cursor, mp) == NULL)
- return ENOMEM;
while (IS_BRANCH(mp)) {
- unsigned int i = 0;
MDB_node *node;
DPRINTF("branch page %lu has %u keys", mp->mp_pgno, NUMKEYS(mp));
DPRINTF("found index 0 to page %lu", NODEPGNO(NODEPTR(mp, 0)));
if (key == NULL) /* Initialize cursor to first page. */
- i = 0;
+ mc->mc_ki[mc->mc_top] = 0;
else if (key->mv_size > MAXKEYSIZE && key->mv_data == NULL) {
/* cursor to last page */
- i = NUMKEYS(mp)-1;
+ mc->mc_ki[mc->mc_top] = NUMKEYS(mp)-1;
} else {
int exact;
- node = mdb_search_node(txn, dbi, mp, key, &exact, &i);
+ node = mdb_search_node(mc, key, &exact);
if (node == NULL)
- i = NUMKEYS(mp) - 1;
+ mc->mc_ki[mc->mc_top] = NUMKEYS(mp) - 1;
else if (!exact) {
- assert(i > 0);
- i--;
+ assert(mc->mc_ki[mc->mc_top] > 0);
+ mc->mc_ki[mc->mc_top]--;
}
}
if (key)
DPRINTF("following index %u for key [%s]",
- i, DKEY(key));
- assert(i < NUMKEYS(mp));
- node = NODEPTR(mp, i);
+ 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]);
- if (cursor)
- CURSOR_TOP(cursor)->mp_ki = i;
-
- mpp->mp_parent = mp;
- if ((rc = mdb_get_page(txn, NODEPGNO(node), &mp)))
+ if ((rc = mdb_get_page(mc->mc_txn, NODEPGNO(node), &mp)))
return rc;
- mpp->mp_pi = i;
- mpp->mp_page = mp;
- if (cursor && cursor_push_page(cursor, mp) == NULL)
- return ENOMEM;
+ if ((rc = cursor_push_page(mc, mp)))
+ return rc;
if (modify) {
- MDB_dhead *dh = ((MDB_dhead *)mp)-1;
- if ((rc = mdb_touch(txn, dbi, mpp)) != 0)
+ if ((rc = mdb_touch(mc)) != 0)
return rc;
- dh = ((MDB_dhead *)mpp->mp_page)-1;
- dh->md_parent = mpp->mp_parent;
- dh->md_pi = mpp->mp_pi;
+ mp = mc->mc_pg[mc->mc_top];
}
-
- mp = mpp->mp_page;
}
if (!IS_LEAF(mp)) {
}
/* Search for the page a given key should be in.
- * Stores a pointer to the found page in *mpp.
+ * Pushes parent pages on the cursor stack.
* If key is NULL, search for the lowest page (used by mdb_cursor_first).
- * If cursor is non-null, pushes parent pages on the cursor stack.
* If modify is true, visited pages are updated with new page numbers.
*/
static int
-mdb_search_page(MDB_txn *txn, MDB_dbi dbi, MDB_val *key,
- MDB_cursor *cursor, int modify, MDB_pageparent *mpp)
+mdb_search_page(MDB_cursor *mc, MDB_val *key, int modify)
{
int rc;
pgno_t root;
- /* Choose which root page to start with. If a transaction is given
- * use the root page from the transaction, otherwise read the last
- * committed root page.
+ /* Make sure the txn is still viable, then find the root from
+ * the txn's db table.
*/
- if (F_ISSET(txn->mt_flags, MDB_TXN_ERROR)) {
+ if (F_ISSET(mc->mc_txn->mt_flags, MDB_TXN_ERROR)) {
DPUTS("transaction has failed, must abort");
return EINVAL;
} else
- root = txn->mt_dbs[dbi].md_root;
+ root = mc->mc_txn->mt_dbs[mc->mc_dbi].md_root;
if (root == P_INVALID) { /* Tree is empty. */
DPUTS("tree is empty");
return MDB_NOTFOUND;
}
- if ((rc = mdb_get_page(txn, root, &mpp->mp_page)))
+ if ((rc = mdb_get_page(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",
- dbi, root, mpp->mp_page->mp_flags);
+ mc->mc_dbi, root, mc->mc_pg[0]->mp_flags);
if (modify) {
/* For sub-databases, update main root first */
- if (dbi > MAIN_DBI && !txn->mt_dbxs[dbi].md_dirty) {
- MDB_pageparent mp2;
- rc = mdb_search_page(txn, MAIN_DBI, &txn->mt_dbxs[dbi].md_name,
- NULL, 1, &mp2);
+ 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;
- txn->mt_dbxs[dbi].md_dirty = 1;
+ mc->mc_txn->mt_dbxs[mc->mc_dbi].md_dirty = 1;
}
- if (!F_ISSET(mpp->mp_page->mp_flags, P_DIRTY)) {
- mpp->mp_parent = NULL;
- mpp->mp_pi = 0;
- if ((rc = mdb_touch(txn, dbi, mpp)))
+ if (!F_ISSET(mc->mc_pg[0]->mp_flags, P_DIRTY)) {
+ if ((rc = mdb_touch(mc)))
return rc;
- txn->mt_dbs[dbi].md_root = mpp->mp_page->mp_pgno;
+ mc->mc_txn->mt_dbs[mc->mc_dbi].md_root = mc->mc_pg[0]->mp_pgno;
}
}
- return mdb_search_page_root(txn, dbi, key, cursor, modify, mpp);
+ return mdb_search_page_root(mc, key, modify);
}
static int
int rc;
if (!F_ISSET(leaf->mn_flags, F_BIGDATA)) {
- data->mv_size = leaf->mn_dsize;
+ data->mv_size = NODEDSZ(leaf);
data->mv_data = NODEDATA(leaf);
return MDB_SUCCESS;
}
/* Read overflow data.
*/
- data->mv_size = leaf->mn_dsize;
+ 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);
{
MDB_cursor mc;
MDB_xcursor mx;
- int exact;
+ int exact = 0;
DKBUF;
assert(key);
mc.mc_txn = txn;
mc.mc_dbi = dbi;
- mc.mc_initialized = 0;
+ mc.mc_flags = 0;
if (txn->mt_dbs[dbi].md_flags & MDB_DUPSORT) {
mc.mc_xcursor = &mx;
mdb_xcursor_init0(&mc);
}
static int
-mdb_sibling(MDB_cursor *cursor, int move_right)
+mdb_sibling(MDB_cursor *mc, int move_right)
{
int rc;
+ unsigned int ptop;
MDB_node *indx;
- MDB_ppage *parent;
MDB_page *mp;
- if (cursor->mc_snum < 2) {
+ if (mc->mc_snum < 2) {
return MDB_NOTFOUND; /* root has no siblings */
}
- parent = CURSOR_PARENT(cursor);
+ ptop = mc->mc_top-1;
DPRINTF("parent page is page %lu, index %u",
- parent->mp_page->mp_pgno, parent->mp_ki);
+ mc->mc_pg[ptop]->mp_pgno, mc->mc_ki[ptop]);
- cursor_pop_page(cursor);
- if (move_right ? (parent->mp_ki + 1 >= NUMKEYS(parent->mp_page))
- : (parent->mp_ki == 0)) {
+ cursor_pop_page(mc);
+ if (move_right ? (mc->mc_ki[ptop] + 1u >= NUMKEYS(mc->mc_pg[ptop]))
+ : (mc->mc_ki[ptop] == 0)) {
DPRINTF("no more keys left, moving to %s sibling",
move_right ? "right" : "left");
- if ((rc = mdb_sibling(cursor, move_right)) != MDB_SUCCESS)
+ if ((rc = mdb_sibling(mc, move_right)) != MDB_SUCCESS)
return rc;
- parent = CURSOR_TOP(cursor);
} else {
if (move_right)
- parent->mp_ki++;
+ mc->mc_ki[ptop]++;
else
- parent->mp_ki--;
+ mc->mc_ki[ptop]--;
DPRINTF("just moving to %s index key %u",
- move_right ? "right" : "left", parent->mp_ki);
+ move_right ? "right" : "left", mc->mc_ki[ptop]);
}
- assert(IS_BRANCH(parent->mp_page));
+ assert(IS_BRANCH(mc->mc_pg[ptop]));
- indx = NODEPTR(parent->mp_page, parent->mp_ki);
- if ((rc = mdb_get_page(cursor->mc_txn, NODEPGNO(indx), &mp)))
+ indx = NODEPTR(mc->mc_pg[ptop], mc->mc_ki[ptop]);
+ if ((rc = mdb_get_page(mc->mc_txn, NODEPGNO(indx), &mp)))
return rc;;
-#if 0
- mp->parent = parent->mp_page;
- mp->parent_index = parent->mp_ki;
-#endif
- cursor_push_page(cursor, mp);
+ cursor_push_page(mc, mp);
return MDB_SUCCESS;
}
static int
-mdb_cursor_next(MDB_cursor *cursor, MDB_val *key, MDB_val *data, MDB_cursor_op op)
+mdb_cursor_next(MDB_cursor *mc, MDB_val *key, MDB_val *data, MDB_cursor_op op)
{
- MDB_ppage *top;
MDB_page *mp;
MDB_node *leaf;
int rc;
- if (cursor->mc_eof) {
+ if (mc->mc_flags & C_EOF) {
return MDB_NOTFOUND;
}
- assert(cursor->mc_initialized);
+ assert(mc->mc_flags & C_INITIALIZED);
- top = CURSOR_TOP(cursor);
- mp = top->mp_page;
+ mp = mc->mc_pg[mc->mc_top];
- if (cursor->mc_txn->mt_dbs[cursor->mc_dbi].md_flags & MDB_DUPSORT) {
- leaf = NODEPTR(mp, top->mp_ki);
+ if (mc->mc_txn->mt_dbs[mc->mc_dbi].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) {
- rc = mdb_cursor_next(&cursor->mc_xcursor->mx_cursor, data, NULL, MDB_NEXT);
+ rc = mdb_cursor_next(&mc->mc_xcursor->mx_cursor, data, NULL, MDB_NEXT);
if (op != MDB_NEXT || rc == MDB_SUCCESS)
return rc;
}
} else {
- cursor->mc_xcursor->mx_cursor.mc_initialized = 0;
+ mc->mc_xcursor->mx_cursor.mc_flags = 0;
if (op == MDB_NEXT_DUP)
return MDB_NOTFOUND;
}
}
- DPRINTF("cursor_next: top page is %lu in cursor %p", mp->mp_pgno, (void *) cursor);
+ DPRINTF("cursor_next: top page is %lu in cursor %p", mp->mp_pgno, (void *) mc);
- if (top->mp_ki + 1 >= NUMKEYS(mp)) {
+ if (mc->mc_ki[mc->mc_top] + 1u >= NUMKEYS(mp)) {
DPUTS("=====> move to next sibling page");
- if (mdb_sibling(cursor, 1) != MDB_SUCCESS) {
- cursor->mc_eof = 1;
+ if (mdb_sibling(mc, 1) != MDB_SUCCESS) {
+ mc->mc_flags |= C_EOF;
return MDB_NOTFOUND;
}
- top = CURSOR_TOP(cursor);
- mp = top->mp_page;
- DPRINTF("next page is %lu, key index %u", mp->mp_pgno, top->mp_ki);
+ mp = mc->mc_pg[mc->mc_top];
+ DPRINTF("next page is %lu, key index %u", mp->mp_pgno, mc->mc_ki[mc->mc_top]);
} else
- top->mp_ki++;
+ mc->mc_ki[mc->mc_top]++;
DPRINTF("==> cursor points to page %lu with %u keys, key index %u",
- mp->mp_pgno, NUMKEYS(mp), top->mp_ki);
+ mp->mp_pgno, NUMKEYS(mp), mc->mc_ki[mc->mc_top]);
if (IS_LEAF2(mp)) {
- key->mv_size = cursor->mc_txn->mt_dbs[cursor->mc_dbi].md_pad;
- key->mv_data = LEAF2KEY(mp, top->mp_ki, key->mv_size);
+ key->mv_size = mc->mc_txn->mt_dbs[mc->mc_dbi].md_pad;
+ key->mv_data = LEAF2KEY(mp, mc->mc_ki[mc->mc_top], key->mv_size);
return MDB_SUCCESS;
}
assert(IS_LEAF(mp));
- leaf = NODEPTR(mp, top->mp_ki);
+ leaf = NODEPTR(mp, mc->mc_ki[mc->mc_top]);
if (F_ISSET(leaf->mn_flags, F_DUPDATA)) {
- mdb_xcursor_init1(cursor->mc_txn, cursor->mc_dbi, cursor->mc_xcursor, mp, leaf);
+ mdb_xcursor_init1(mc, leaf);
}
if (data) {
- if ((rc = mdb_read_data(cursor->mc_txn, leaf, data) != MDB_SUCCESS))
+ if ((rc = mdb_read_data(mc->mc_txn, leaf, data) != MDB_SUCCESS))
return rc;
if (F_ISSET(leaf->mn_flags, F_DUPDATA)) {
- rc = mdb_cursor_first(&cursor->mc_xcursor->mx_cursor, data, NULL);
+ rc = mdb_cursor_first(&mc->mc_xcursor->mx_cursor, data, NULL);
if (rc != MDB_SUCCESS)
return rc;
}
}
static int
-mdb_cursor_prev(MDB_cursor *cursor, MDB_val *key, MDB_val *data, MDB_cursor_op op)
+mdb_cursor_prev(MDB_cursor *mc, MDB_val *key, MDB_val *data, MDB_cursor_op op)
{
- MDB_ppage *top;
MDB_page *mp;
MDB_node *leaf;
int rc;
- assert(cursor->mc_initialized);
+ assert(mc->mc_flags & C_INITIALIZED);
- top = CURSOR_TOP(cursor);
- mp = top->mp_page;
+ mp = mc->mc_pg[mc->mc_top];
- if (cursor->mc_txn->mt_dbs[cursor->mc_dbi].md_flags & MDB_DUPSORT) {
- leaf = NODEPTR(mp, top->mp_ki);
+ if (mc->mc_txn->mt_dbs[mc->mc_dbi].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)) {
- rc = mdb_cursor_prev(&cursor->mc_xcursor->mx_cursor, data, NULL, MDB_PREV);
+ rc = mdb_cursor_prev(&mc->mc_xcursor->mx_cursor, data, NULL, MDB_PREV);
if (op != MDB_PREV || rc == MDB_SUCCESS)
return rc;
} else {
- cursor->mc_xcursor->mx_cursor.mc_initialized = 0;
+ mc->mc_xcursor->mx_cursor.mc_flags = 0;
if (op == MDB_PREV_DUP)
return MDB_NOTFOUND;
}
}
}
- DPRINTF("cursor_prev: top page is %lu in cursor %p", mp->mp_pgno, (void *) cursor);
+ DPRINTF("cursor_prev: top page is %lu in cursor %p", mp->mp_pgno, (void *) mc);
- if (top->mp_ki == 0) {
+ if (mc->mc_ki[mc->mc_top] == 0) {
DPUTS("=====> move to prev sibling page");
- if (mdb_sibling(cursor, 0) != MDB_SUCCESS) {
- cursor->mc_initialized = 0;
+ if (mdb_sibling(mc, 0) != MDB_SUCCESS) {
+ mc->mc_flags &= ~C_INITIALIZED;
return MDB_NOTFOUND;
}
- top = CURSOR_TOP(cursor);
- mp = top->mp_page;
- top->mp_ki = NUMKEYS(mp) - 1;
- DPRINTF("prev page is %lu, key index %u", mp->mp_pgno, top->mp_ki);
+ 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]);
} else
- top->mp_ki--;
+ mc->mc_ki[mc->mc_top]--;
- cursor->mc_eof = 0;
+ mc->mc_flags &= ~C_EOF;
DPRINTF("==> cursor points to page %lu with %u keys, key index %u",
- mp->mp_pgno, NUMKEYS(mp), top->mp_ki);
+ mp->mp_pgno, NUMKEYS(mp), mc->mc_ki[mc->mc_top]);
if (IS_LEAF2(mp)) {
- key->mv_size = cursor->mc_txn->mt_dbs[cursor->mc_dbi].md_pad;
- key->mv_data = LEAF2KEY(mp, top->mp_ki, key->mv_size);
+ key->mv_size = mc->mc_txn->mt_dbs[mc->mc_dbi].md_pad;
+ key->mv_data = LEAF2KEY(mp, mc->mc_ki[mc->mc_top], key->mv_size);
return MDB_SUCCESS;
}
assert(IS_LEAF(mp));
- leaf = NODEPTR(mp, top->mp_ki);
+ leaf = NODEPTR(mp, mc->mc_ki[mc->mc_top]);
if (F_ISSET(leaf->mn_flags, F_DUPDATA)) {
- mdb_xcursor_init1(cursor->mc_txn, cursor->mc_dbi, cursor->mc_xcursor, mp, leaf);
+ mdb_xcursor_init1(mc, leaf);
}
if (data) {
- if ((rc = mdb_read_data(cursor->mc_txn, leaf, data) != MDB_SUCCESS))
+ if ((rc = mdb_read_data(mc->mc_txn, leaf, data) != MDB_SUCCESS))
return rc;
if (F_ISSET(leaf->mn_flags, F_DUPDATA)) {
- rc = mdb_cursor_last(&cursor->mc_xcursor->mx_cursor, data, NULL);
+ rc = mdb_cursor_last(&mc->mc_xcursor->mx_cursor, data, NULL);
if (rc != MDB_SUCCESS)
return rc;
}
}
static int
-mdb_cursor_set(MDB_cursor *cursor, MDB_val *key, MDB_val *data,
+mdb_cursor_set(MDB_cursor *mc, MDB_val *key, MDB_val *data,
MDB_cursor_op op, int *exactp)
{
int rc;
MDB_node *leaf;
- MDB_ppage *top;
- MDB_pageparent mpp;
DKBUF;
- assert(cursor);
+ assert(mc);
assert(key);
assert(key->mv_size > 0);
/* See if we're already on the right page */
- if (cursor->mc_initialized) {
+ if (mc->mc_flags & C_INITIALIZED) {
MDB_val nodekey;
- top = CURSOR_TOP(cursor);
- /* Don't try this for LEAF2 pages. Maybe support that later. */
- if ((top->mp_page->mp_flags & (P_LEAF|P_LEAF2)) == P_LEAF) {
- leaf = NODEPTR(top->mp_page, 0);
+
+ 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);
+ } else {
+ leaf = NODEPTR(mc->mc_pg[mc->mc_top], 0);
MDB_SET_KEY(leaf, &nodekey);
- rc = mdb_cmp(cursor->mc_txn, cursor->mc_dbi, key, &nodekey);
- if (rc >= 0) {
- leaf = NODEPTR(top->mp_page, NUMKEYS(top->mp_page)-1);
- MDB_SET_KEY(leaf, &nodekey);
- rc = mdb_cmp(cursor->mc_txn, cursor->mc_dbi, key, &nodekey);
- if (rc <= 0) {
- /* we're already on the right page */
- mpp.mp_page = top->mp_page;
+ }
+ rc = mc->mc_txn->mt_dbxs[mc->mc_dbi].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:
+ if (exactp)
+ *exactp = 1;
+ leaf = NODEPTR(mc->mc_pg[mc->mc_top], mc->mc_ki[mc->mc_top]);
+ goto set3;
+ }
+ 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);
+ } else {
+ leaf = NODEPTR(mc->mc_pg[mc->mc_top], NUMKEYS(mc->mc_pg[mc->mc_top])-1);
+ MDB_SET_KEY(leaf, &nodekey);
+ }
+ rc = mc->mc_txn->mt_dbxs[mc->mc_dbi].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;
+ goto set1;
+ }
+ if (rc < 0) {
+ /* This is definitely the right page, skip search_page */
rc = 0;
goto set2;
}
}
+ /* If any parents have right-sibs, search.
+ * Otherwise, there's nothing further.
+ */
+ for (i=0; i<mc->mc_top; i++)
+ if (mc->mc_ki[i] <
+ NUMKEYS(mc->mc_pg[i])-1)
+ break;
+ if (i == mc->mc_top) {
+ /* There are no other pages */
+ mc->mc_ki[mc->mc_top] = NUMKEYS(mc->mc_pg[mc->mc_top]);
+ return MDB_NOTFOUND;
+ }
}
}
- cursor->mc_snum = 0;
- rc = mdb_search_page(cursor->mc_txn, cursor->mc_dbi, key, cursor, 0, &mpp);
+ rc = mdb_search_page(mc, key, 0);
if (rc != MDB_SUCCESS)
return rc;
- assert(IS_LEAF(mpp.mp_page));
+ assert(IS_LEAF(mc->mc_pg[mc->mc_top]));
- top = CURSOR_TOP(cursor);
set2:
- leaf = mdb_search_node(cursor->mc_txn, cursor->mc_dbi, mpp.mp_page, key, exactp, &top->mp_ki);
+ leaf = mdb_search_node(mc, key, exactp);
if (exactp != NULL && !*exactp) {
/* MDB_SET specified and not an exact match. */
return MDB_NOTFOUND;
if (leaf == NULL) {
DPUTS("===> inexact leaf not found, goto sibling");
- if ((rc = mdb_sibling(cursor, 1)) != MDB_SUCCESS)
+ if ((rc = mdb_sibling(mc, 1)) != MDB_SUCCESS)
return rc; /* no entries matched */
- top = CURSOR_TOP(cursor);
- top->mp_ki = 0;
- mpp.mp_page = top->mp_page;
- assert(IS_LEAF(mpp.mp_page));
- leaf = NODEPTR(mpp.mp_page, 0);
+ 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);
}
- cursor->mc_initialized = 1;
- cursor->mc_eof = 0;
+set3:
+ mc->mc_flags |= C_INITIALIZED;
+ mc->mc_flags &= ~C_EOF;
- if (IS_LEAF2(mpp.mp_page)) {
- key->mv_size = cursor->mc_txn->mt_dbs[cursor->mc_dbi].md_pad;
- key->mv_data = LEAF2KEY(mpp.mp_page, top->mp_ki, key->mv_size);
+ 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);
return MDB_SUCCESS;
}
if (F_ISSET(leaf->mn_flags, F_DUPDATA)) {
- mdb_xcursor_init1(cursor->mc_txn, cursor->mc_dbi, cursor->mc_xcursor, mpp.mp_page, leaf);
+ mdb_xcursor_init1(mc, leaf);
}
if (data) {
if (F_ISSET(leaf->mn_flags, F_DUPDATA)) {
if (op == MDB_SET || op == MDB_SET_RANGE) {
- rc = mdb_cursor_first(&cursor->mc_xcursor->mx_cursor, data, NULL);
+ rc = mdb_cursor_first(&mc->mc_xcursor->mx_cursor, data, NULL);
} else {
int ex2, *ex2p;
if (op == MDB_GET_BOTH) {
} else {
ex2p = NULL;
}
- rc = mdb_cursor_set(&cursor->mc_xcursor->mx_cursor, data, NULL, MDB_SET_RANGE, ex2p);
+ rc = mdb_cursor_set(&mc->mc_xcursor->mx_cursor, data, NULL, MDB_SET_RANGE, ex2p);
if (rc != MDB_SUCCESS)
return rc;
}
} else if (op == MDB_GET_BOTH || op == MDB_GET_BOTH_RANGE) {
MDB_val d2;
- if ((rc = mdb_read_data(cursor->mc_txn, leaf, &d2)) != MDB_SUCCESS)
+ if ((rc = mdb_read_data(mc->mc_txn, leaf, &d2)) != MDB_SUCCESS)
return rc;
- rc = mdb_dcmp(cursor->mc_txn, cursor->mc_dbi, data, &d2);
+ rc = mc->mc_txn->mt_dbxs[mc->mc_dbi].md_dcmp(data, &d2);
if (rc) {
if (op == MDB_GET_BOTH || rc > 0)
return MDB_NOTFOUND;
}
} else {
- if ((rc = mdb_read_data(cursor->mc_txn, leaf, data)) != MDB_SUCCESS)
+ if ((rc = mdb_read_data(mc->mc_txn, leaf, data)) != MDB_SUCCESS)
return rc;
}
}
}
static int
-mdb_cursor_first(MDB_cursor *cursor, MDB_val *key, MDB_val *data)
+mdb_cursor_first(MDB_cursor *mc, MDB_val *key, MDB_val *data)
{
int rc;
- MDB_pageparent mpp;
MDB_node *leaf;
- cursor->mc_snum = 0;
-
- rc = mdb_search_page(cursor->mc_txn, cursor->mc_dbi, NULL, cursor, 0, &mpp);
+ rc = mdb_search_page(mc, NULL, 0);
if (rc != MDB_SUCCESS)
return rc;
- assert(IS_LEAF(mpp.mp_page));
+ assert(IS_LEAF(mc->mc_pg[mc->mc_top]));
- leaf = NODEPTR(mpp.mp_page, 0);
- cursor->mc_initialized = 1;
- cursor->mc_eof = 0;
+ leaf = NODEPTR(mc->mc_pg[mc->mc_top], 0);
+ mc->mc_flags |= C_INITIALIZED;
+ mc->mc_flags &= ~C_EOF;
- if (IS_LEAF2(mpp.mp_page)) {
- key->mv_size = cursor->mc_txn->mt_dbs[cursor->mc_dbi].md_pad;
- key->mv_data = LEAF2KEY(mpp.mp_page, 0, key->mv_size);
+ 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], 0, key->mv_size);
return MDB_SUCCESS;
}
if (data) {
if (F_ISSET(leaf->mn_flags, F_DUPDATA)) {
- mdb_xcursor_init1(cursor->mc_txn, cursor->mc_dbi, cursor->mc_xcursor, mpp.mp_page, leaf);
- rc = mdb_cursor_first(&cursor->mc_xcursor->mx_cursor, data, NULL);
+ mdb_xcursor_init1(mc, leaf);
+ rc = mdb_cursor_first(&mc->mc_xcursor->mx_cursor, data, NULL);
if (rc)
return rc;
} else {
- if (cursor->mc_xcursor)
- cursor->mc_xcursor->mx_cursor.mc_initialized = 0;
- if ((rc = mdb_read_data(cursor->mc_txn, leaf, data)) != MDB_SUCCESS)
+ if (mc->mc_xcursor)
+ mc->mc_xcursor->mx_cursor.mc_flags = 0;
+ if ((rc = mdb_read_data(mc->mc_txn, leaf, data)) != MDB_SUCCESS)
return rc;
}
}
}
static int
-mdb_cursor_last(MDB_cursor *cursor, MDB_val *key, MDB_val *data)
+mdb_cursor_last(MDB_cursor *mc, MDB_val *key, MDB_val *data)
{
int rc;
- MDB_ppage *top;
- MDB_pageparent mpp;
MDB_node *leaf;
MDB_val lkey;
- cursor->mc_snum = 0;
-
lkey.mv_size = MAXKEYSIZE+1;
lkey.mv_data = NULL;
- rc = mdb_search_page(cursor->mc_txn, cursor->mc_dbi, &lkey, cursor, 0, &mpp);
+ rc = mdb_search_page(mc, &lkey, 0);
if (rc != MDB_SUCCESS)
return rc;
- assert(IS_LEAF(mpp.mp_page));
+ assert(IS_LEAF(mc->mc_pg[mc->mc_top]));
- leaf = NODEPTR(mpp.mp_page, NUMKEYS(mpp.mp_page)-1);
- cursor->mc_initialized = 1;
- cursor->mc_eof = 0;
+ leaf = NODEPTR(mc->mc_pg[mc->mc_top], NUMKEYS(mc->mc_pg[mc->mc_top])-1);
+ mc->mc_flags |= C_INITIALIZED;
+ mc->mc_flags &= ~C_EOF;
- top = CURSOR_TOP(cursor);
- top->mp_ki = NUMKEYS(top->mp_page) - 1;
+ mc->mc_ki[mc->mc_top] = NUMKEYS(mc->mc_pg[mc->mc_top]) - 1;
- if (IS_LEAF2(mpp.mp_page)) {
- key->mv_size = cursor->mc_txn->mt_dbs[cursor->mc_dbi].md_pad;
- key->mv_data = LEAF2KEY(mpp.mp_page, top->mp_ki, key->mv_size);
+ 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);
return MDB_SUCCESS;
}
if (data) {
if (F_ISSET(leaf->mn_flags, F_DUPDATA)) {
- mdb_xcursor_init1(cursor->mc_txn, cursor->mc_dbi, cursor->mc_xcursor, mpp.mp_page, leaf);
- rc = mdb_cursor_last(&cursor->mc_xcursor->mx_cursor, data, NULL);
+ mdb_xcursor_init1(mc, leaf);
+ rc = mdb_cursor_last(&mc->mc_xcursor->mx_cursor, data, NULL);
if (rc)
return rc;
} else {
- if ((rc = mdb_read_data(cursor->mc_txn, leaf, data)) != MDB_SUCCESS)
+ if ((rc = mdb_read_data(mc->mc_txn, leaf, data)) != MDB_SUCCESS)
return rc;
}
}
}
int
-mdb_cursor_get(MDB_cursor *cursor, MDB_val *key, MDB_val *data,
+mdb_cursor_get(MDB_cursor *mc, MDB_val *key, MDB_val *data,
MDB_cursor_op op)
{
int rc;
int exact = 0;
- assert(cursor);
+ assert(mc);
switch (op) {
case MDB_GET_BOTH:
case MDB_GET_BOTH_RANGE:
- if (data == NULL || cursor->mc_xcursor == NULL) {
+ if (data == NULL || mc->mc_xcursor == NULL) {
rc = EINVAL;
break;
}
if (key == NULL || key->mv_size == 0 || key->mv_size > MAXKEYSIZE) {
rc = EINVAL;
} else if (op == MDB_SET_RANGE)
- rc = mdb_cursor_set(cursor, key, data, op, NULL);
+ rc = mdb_cursor_set(mc, key, data, op, NULL);
else
- rc = mdb_cursor_set(cursor, key, data, op, &exact);
+ rc = mdb_cursor_set(mc, key, data, op, &exact);
break;
case MDB_GET_MULTIPLE:
if (data == NULL ||
- !(cursor->mc_txn->mt_dbs[cursor->mc_dbi].md_flags & MDB_DUPFIXED) ||
- !cursor->mc_initialized) {
+ !(mc->mc_txn->mt_dbs[mc->mc_dbi].md_flags & MDB_DUPFIXED) ||
+ !(mc->mc_flags & C_INITIALIZED)) {
rc = EINVAL;
break;
}
rc = MDB_SUCCESS;
- if (!cursor->mc_xcursor->mx_cursor.mc_initialized || cursor->mc_xcursor->mx_cursor.mc_eof)
+ if (!(mc->mc_xcursor->mx_cursor.mc_flags & C_INITIALIZED) ||
+ (mc->mc_xcursor->mx_cursor.mc_flags & C_EOF))
break;
goto fetchm;
case MDB_NEXT_MULTIPLE:
if (data == NULL ||
- !(cursor->mc_txn->mt_dbs[cursor->mc_dbi].md_flags & MDB_DUPFIXED)) {
+ !(mc->mc_txn->mt_dbs[mc->mc_dbi].md_flags & MDB_DUPFIXED)) {
rc = EINVAL;
break;
}
- if (!cursor->mc_initialized)
- rc = mdb_cursor_first(cursor, key, data);
+ if (!(mc->mc_flags & C_INITIALIZED))
+ rc = mdb_cursor_first(mc, key, data);
else
- rc = mdb_cursor_next(cursor, key, data, MDB_NEXT_DUP);
+ rc = mdb_cursor_next(mc, key, data, MDB_NEXT_DUP);
if (rc == MDB_SUCCESS) {
- if (cursor->mc_xcursor->mx_cursor.mc_initialized) {
- MDB_ppage *top;
+ if (mc->mc_xcursor->mx_cursor.mc_flags & C_INITIALIZED) {
+ MDB_cursor *mx;
fetchm:
- top = CURSOR_TOP(&cursor->mc_xcursor->mx_cursor);
- data->mv_size = NUMKEYS(top->mp_page) *
- cursor->mc_xcursor->mx_txn.mt_dbs[cursor->mc_xcursor->mx_cursor.mc_dbi].md_pad;
- data->mv_data = METADATA(top->mp_page);
- top->mp_ki = NUMKEYS(top->mp_page)-1;
+ 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;
+ 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 {
rc = MDB_NOTFOUND;
}
case MDB_NEXT:
case MDB_NEXT_DUP:
case MDB_NEXT_NODUP:
- if (!cursor->mc_initialized)
- rc = mdb_cursor_first(cursor, key, data);
+ if (!(mc->mc_flags & C_INITIALIZED))
+ rc = mdb_cursor_first(mc, key, data);
else
- rc = mdb_cursor_next(cursor, key, data, op);
+ rc = mdb_cursor_next(mc, key, data, op);
break;
case MDB_PREV:
case MDB_PREV_DUP:
case MDB_PREV_NODUP:
- if (!cursor->mc_initialized || cursor->mc_eof)
- rc = mdb_cursor_last(cursor, key, data);
+ if (!(mc->mc_flags & C_INITIALIZED) || (mc->mc_flags & C_EOF))
+ rc = mdb_cursor_last(mc, key, data);
else
- rc = mdb_cursor_prev(cursor, key, data, op);
+ rc = mdb_cursor_prev(mc, key, data, op);
break;
case MDB_FIRST:
- rc = mdb_cursor_first(cursor, key, data);
+ rc = mdb_cursor_first(mc, key, data);
break;
case MDB_FIRST_DUP:
if (data == NULL ||
- !(cursor->mc_txn->mt_dbs[cursor->mc_dbi].md_flags & MDB_DUPSORT) ||
- !cursor->mc_initialized ||
- !cursor->mc_xcursor->mx_cursor.mc_initialized) {
+ !(mc->mc_txn->mt_dbs[mc->mc_dbi].md_flags & MDB_DUPSORT) ||
+ !(mc->mc_flags & C_INITIALIZED) ||
+ !(mc->mc_xcursor->mx_cursor.mc_flags & C_INITIALIZED)) {
rc = EINVAL;
break;
}
- rc = mdb_cursor_first(&cursor->mc_xcursor->mx_cursor, data, NULL);
+ rc = mdb_cursor_first(&mc->mc_xcursor->mx_cursor, data, NULL);
break;
case MDB_LAST:
- rc = mdb_cursor_last(cursor, key, data);
+ rc = mdb_cursor_last(mc, key, data);
break;
case MDB_LAST_DUP:
if (data == NULL ||
- !(cursor->mc_txn->mt_dbs[cursor->mc_dbi].md_flags & MDB_DUPSORT) ||
- !cursor->mc_initialized ||
- !cursor->mc_xcursor->mx_cursor.mc_initialized) {
+ !(mc->mc_txn->mt_dbs[mc->mc_dbi].md_flags & MDB_DUPSORT) ||
+ !(mc->mc_flags & C_INITIALIZED) ||
+ !(mc->mc_xcursor->mx_cursor.mc_flags & C_INITIALIZED)) {
rc = EINVAL;
break;
}
- rc = mdb_cursor_last(&cursor->mc_xcursor->mx_cursor, data, NULL);
+ rc = mdb_cursor_last(&mc->mc_xcursor->mx_cursor, data, NULL);
break;
default:
DPRINTF("unhandled/unimplemented cursor operation %u", op);
}
static int
-mdb_cursor_touch(MDB_cursor *cursor)
+mdb_cursor_touch(MDB_cursor *mc)
{
- MDB_pageparent mpp;
- unsigned int i;
int rc;
- mpp.mp_parent = NULL;
- mpp.mp_pi = 0;
- for(i=0; i<cursor->mc_snum; i++) {
- mpp.mp_page = cursor->mc_stack[i].mp_page;
- if (!F_ISSET(mpp.mp_page->mp_flags, P_DIRTY)) {
- rc = mdb_touch(cursor->mc_txn, cursor->mc_dbi, &mpp);
+ 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;
+ }
+ 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;
- cursor->mc_stack[i].mp_page = mpp.mp_page;
+ if (!mc->mc_top) {
+ mc->mc_txn->mt_dbs[mc->mc_dbi].md_root =
+ mc->mc_pg[mc->mc_top]->mp_pgno;
+ }
}
- mpp.mp_parent = mpp.mp_page;
- mpp.mp_pi = cursor->mc_stack[i].mp_ki;
}
- if (cursor->mc_xcursor)
- return mdb_cursor_touch(&cursor->mc_xcursor->mx_cursor);
+ mc->mc_top = mc->mc_snum-1;
return MDB_SUCCESS;
}
int
-mdb_cursor_put(MDB_cursor *cursor, MDB_val *key, MDB_val *data,
+mdb_cursor_put(MDB_cursor *mc, MDB_val *key, MDB_val *data,
unsigned int flags)
{
- MDB_ppage *top;
MDB_node *leaf;
- int rc, exact;
+ MDB_val xdata, *rdata, dkey;
+ MDB_db dummy;
+ char dbuf[PAGESIZE];
+ int do_sub = 0;
+ size_t nsize;
+ DKBUF;
+ int rc, rc2;
- if (flags == MDB_NODUPDATA) {
- rc = mdb_cursor_set(cursor, key, data, MDB_GET_BOTH, &exact);
- if (rc == 0)
- return MDB_KEYEXIST;
- if (rc != MDB_NOTFOUND)
- return rc;
- } else if (flags != MDB_CURRENT) {
- rc = mdb_cursor_set(cursor, key, NULL, MDB_SET, &exact);
- if (flags == MDB_NOOVERWRITE && rc == 0)
+ 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);
+
+ dkey.mv_size = 0;
+
+ if (flags == MDB_CURRENT) {
+ 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) {
+ 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) {
+ 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_DUPFIXED)
+ np->mp_flags |= P_LEAF2;
+ mc->mc_flags |= C_INITIALIZED;
+ rc = MDB_NOTFOUND;
+ goto top;
+ } else {
+ int exact = 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 != MDB_NOTFOUND)
+ }
+ if (rc && rc != MDB_NOTFOUND)
return rc;
- } else if (!cursor->mc_initialized) {
- return EINVAL;
}
+
/* Cursor is positioned, now make sure all pages are writable */
- rc = mdb_cursor_touch(cursor);
+ rc2 = mdb_cursor_touch(mc);
+ if (rc2) return rc2;
+
+top:
+ /* The key already exists */
+ if (rc == MDB_SUCCESS) {
+ /* there's only a key anyway, so this is a no-op */
+ if (IS_LEAF2(mc->mc_pg[mc->mc_top])) {
+ unsigned int ksize = mc->mc_txn->mt_dbs[mc->mc_dbi].md_pad;
+ if (key->mv_size != ksize)
+ return EINVAL;
+ if (flags == MDB_CURRENT) {
+ char *ptr = LEAF2KEY(mc->mc_pg[mc->mc_top], mc->mc_ki[mc->mc_top], ksize);
+ memcpy(ptr, key->mv_data, ksize);
+ }
+ return MDB_SUCCESS;
+ }
+
+ 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)) {
+ /* Was a single item before, must convert now */
+ if (!F_ISSET(leaf->mn_flags, F_DUPDATA)) {
+ 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))
+ 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;
+ }
+ dummy.md_root = P_INVALID;
+ if (dkey.mv_size == sizeof(MDB_db)) {
+ memcpy(NODEDATA(leaf), &dummy, sizeof(dummy));
+ goto put_sub;
+ }
+ mdb_del_node(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;
+ /* new sub-DB, must fully init xcursor */
+ if (flags == MDB_CURRENT)
+ flags = 0;
+ goto new_sub;
+ }
+ 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);
+ goto done;
+ }
+ mdb_del_node(mc->mc_pg[mc->mc_top], mc->mc_ki[mc->mc_top], 0);
+ } else {
+ DPRINTF("inserting key at index %i", mc->mc_ki[mc->mc_top]);
+ }
+
+ rdata = data;
+
+new_sub:
+ 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);
+ } else {
+ /* There is room already in this leaf page. */
+ rc = mdb_add_node(mc, mc->mc_ki[mc->mc_top], key, rdata, 0, 0);
+ }
+
+ 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]);
+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;
+ /* 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, 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;
+}
+
+int
+mdb_cursor_del(MDB_cursor *mc, unsigned int flags)
+{
+ MDB_node *leaf;
+ int rc;
+
+ if (F_ISSET(mc->mc_txn->mt_flags, MDB_TXN_RDONLY))
+ return EACCES;
+
+ if (!mc->mc_flags & C_INITIALIZED)
+ return EINVAL;
+
+ rc = mdb_cursor_touch(mc);
if (rc) return rc;
- top = CURSOR_TOP(cursor);
+ 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);
+ 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--;
+ 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; i<NUMKEYS(mx->mc_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;
+ }
+ }
+ /* 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);
}
/* Allocate a page and initialize it
*/
-static MDB_dpage *
-mdb_new_page(MDB_txn *txn, MDB_dbi dbi, uint32_t flags, int num)
+static MDB_page *
+mdb_new_page(MDB_cursor *mc, uint32_t flags, int num)
{
- MDB_dpage *dp;
+ MDB_page *np;
- if ((dp = mdb_alloc_page(txn, dbi, NULL, 0, num)) == NULL)
+ if ((np = mdb_alloc_page(mc, num)) == NULL)
return NULL;
DPRINTF("allocated new mpage %lu, page size %u",
- dp->p.mp_pgno, txn->mt_env->me_psize);
- dp->p.mp_flags = flags | P_DIRTY;
- dp->p.mp_lower = PAGEHDRSZ;
- dp->p.mp_upper = txn->mt_env->me_psize;
+ 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(&dp->p))
- txn->mt_dbs[dbi].md_branch_pages++;
- else if (IS_LEAF(&dp->p))
- txn->mt_dbs[dbi].md_leaf_pages++;
- else if (IS_OVERFLOW(&dp->p)) {
- txn->mt_dbs[dbi].md_overflow_pages += num;
- dp->p.mp_pages = num;
+ if (IS_BRANCH(np))
+ mc->mc_txn->mt_dbs[mc->mc_dbi].md_branch_pages++;
+ else if (IS_LEAF(np))
+ mc->mc_txn->mt_dbs[mc->mc_dbi].md_leaf_pages++;
+ else if (IS_OVERFLOW(np)) {
+ mc->mc_txn->mt_dbs[mc->mc_dbi].md_overflow_pages += num;
+ np->mp_pages = num;
}
- return dp;
+ return np;
}
static size_t
/* put on overflow page */
sz -= data->mv_size - sizeof(pgno_t);
}
+ sz += sz & 1;
return sz + sizeof(indx_t);
}
}
static int
-mdb_add_node(MDB_txn *txn, MDB_dbi dbi, MDB_page *mp, indx_t indx,
+mdb_add_node(MDB_cursor *mc, indx_t indx,
MDB_val *key, MDB_val *data, pgno_t pgno, uint8_t flags)
{
unsigned int i;
size_t node_size = NODESIZE;
indx_t ofs;
MDB_node *node;
- MDB_dpage *ofp = NULL; /* overflow page */
+ MDB_page *mp = mc->mc_pg[mc->mc_top];
+ MDB_page *ofp = NULL; /* overflow page */
DKBUF;
assert(mp->mp_upper >= mp->mp_lower);
if (IS_LEAF2(mp)) {
/* Move higher keys up one slot. */
- int ksize = txn->mt_dbs[dbi].md_pad, dif;
+ int ksize = mc->mc_txn->mt_dbs[mc->mc_dbi].md_pad, dif;
char *ptr = LEAF2KEY(mp, indx, ksize);
dif = NUMKEYS(mp) - indx;
if (dif > 0)
if (F_ISSET(flags, F_BIGDATA)) {
/* Data already on overflow page. */
node_size += sizeof(pgno_t);
- } else if (data->mv_size >= txn->mt_env->me_psize / MDB_MINKEYS) {
- int ovpages = OVPAGES(data->mv_size, txn->mt_env->me_psize);
+ } else if (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);
node_size += sizeof(pgno_t);
- if ((ofp = mdb_new_page(txn, dbi, P_OVERFLOW, ovpages)) == NULL)
+ if ((ofp = mdb_new_page(mc, P_OVERFLOW, ovpages)) == NULL)
return ENOMEM;
- DPRINTF("allocated overflow page %lu", ofp->p.mp_pgno);
+ DPRINTF("allocated overflow page %lu", ofp->mp_pgno);
flags |= F_BIGDATA;
} else {
node_size += data->mv_size;
}
}
+ node_size += node_size & 1;
if (node_size + sizeof(indx_t) > SIZELEFT(mp)) {
DPRINTF("not enough room in page %lu, got %u ptrs",
node->mn_ksize = (key == NULL) ? 0 : key->mv_size;
node->mn_flags = flags;
if (IS_LEAF(mp))
- node->mn_dsize = data->mv_size;
+ SETDSZ(node,data->mv_size);
else
- NODEPGNO(node) = pgno;
+ SETPGNO(node,pgno);
if (key)
memcpy(NODEKEY(node), key->mv_data, key->mv_size);
memcpy(node->mn_data + key->mv_size, data->mv_data,
data->mv_size);
} else {
- memcpy(node->mn_data + key->mv_size, &ofp->p.mp_pgno,
+ memcpy(node->mn_data + key->mv_size, &ofp->mp_pgno,
sizeof(pgno_t));
- memcpy(METADATA(&ofp->p), data->mv_data, data->mv_size);
+ memcpy(METADATA(ofp), data->mv_data, data->mv_size);
}
}
else
sz += NODEDSZ(node);
}
+ sz += sz & 1;
ptr = mp->mp_ptrs[indx];
numkeys = NUMKEYS(mp);
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_snum = 0;
+ mx->mx_cursor.mc_xcursor = NULL;
mx->mx_cursor.mc_txn = &mx->mx_txn;
mx->mx_cursor.mc_dbi = dbn+1;
}
static void
-mdb_xcursor_init1(MDB_txn *txn, MDB_dbi dbi, MDB_xcursor *mx, MDB_page *mp, MDB_node *node)
+mdb_xcursor_init1(MDB_cursor *mc, MDB_node *node)
{
MDB_db *db = NODEDATA(node);
+ MDB_xcursor *mx = mc->mc_xcursor;
MDB_dbi dbn;
- mx->mx_dbs[0] = txn->mt_dbs[0];
- mx->mx_dbs[1] = txn->mt_dbs[1];
- if (dbi > 1) {
- mx->mx_dbs[2] = txn->mt_dbs[dbi];
+ 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, dbi, db->md_root);
+ 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(mp->mp_flags, P_DIRTY))
+ 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 = txn->mt_next_pgno;
- mx->mx_txn.mt_u = txn->mt_u;
- mx->mx_cursor.mc_initialized = 0;
- mx->mx_cursor.mc_eof = 0;
+ mx->mx_txn.mt_next_pgno = mc->mc_txn->mt_next_pgno;
+ mx->mx_cursor.mc_snum = 0;
+ mx->mx_cursor.mc_flags = 0;
}
static void
-mdb_xcursor_fini(MDB_txn *txn, MDB_dbi dbi, MDB_xcursor *mx)
+mdb_xcursor_init2(MDB_cursor *mc)
{
- txn->mt_next_pgno = mx->mx_txn.mt_next_pgno;
- txn->mt_u = mx->mx_txn.mt_u;
- txn->mt_dbs[0] = mx->mx_dbs[0];
- txn->mt_dbs[1] = mx->mx_dbs[1];
- txn->mt_dbxs[0].md_dirty = mx->mx_dbxs[0].md_dirty;
- txn->mt_dbxs[1].md_dirty = mx->mx_dbxs[1].md_dirty;
- if (dbi > 1) {
- txn->mt_dbs[dbi] = mx->mx_dbs[2];
- txn->mt_dbxs[dbi].md_dirty = mx->mx_dbxs[2].md_dirty;
+ 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,
+ mx->mx_dbs[dbn].md_root);
+ mx->mx_txn.mt_next_pgno = mc->mc_txn->mt_next_pgno;
+}
+
+static void
+mdb_xcursor_fini(MDB_cursor *mc)
+{
+ 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;
}
}
int
mdb_cursor_open(MDB_txn *txn, MDB_dbi dbi, MDB_cursor **ret)
{
- MDB_cursor *cursor;
+ MDB_cursor *mc;
size_t size = sizeof(MDB_cursor);
if (txn == NULL || ret == NULL || !dbi || dbi >= txn->mt_numdbs)
if (txn->mt_dbs[dbi].md_flags & MDB_DUPSORT)
size += sizeof(MDB_xcursor);
- if ((cursor = calloc(1, size)) != NULL) {
- cursor->mc_dbi = dbi;
- cursor->mc_txn = txn;
+ if ((mc = calloc(1, size)) != NULL) {
+ mc->mc_dbi = dbi;
+ mc->mc_txn = txn;
if (txn->mt_dbs[dbi].md_flags & MDB_DUPSORT) {
- MDB_xcursor *mx = (MDB_xcursor *)(cursor + 1);
- cursor->mc_xcursor = mx;
- mdb_xcursor_init0(cursor);
+ MDB_xcursor *mx = (MDB_xcursor *)(mc + 1);
+ mc->mc_xcursor = mx;
+ mdb_xcursor_init0(mc);
}
} else {
return ENOMEM;
}
- *ret = cursor;
+ *ret = mc;
return MDB_SUCCESS;
}
int
mdb_cursor_count(MDB_cursor *mc, unsigned long *countp)
{
- MDB_ppage *top;
MDB_node *leaf;
if (mc == NULL || countp == NULL)
if (!(mc->mc_txn->mt_dbs[mc->mc_dbi].md_flags & MDB_DUPSORT))
return EINVAL;
- top = CURSOR_TOP(mc);
- leaf = NODEPTR(top->mp_page, top->mp_ki);
+ leaf = NODEPTR(mc->mc_pg[mc->mc_top], mc->mc_ki[mc->mc_top]);
if (!F_ISSET(leaf->mn_flags, F_DUPDATA)) {
*countp = 1;
} else {
- if (!mc->mc_xcursor->mx_cursor.mc_initialized)
+ 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;
}
void
-mdb_cursor_close(MDB_cursor *cursor)
+mdb_cursor_close(MDB_cursor *mc)
{
- if (cursor != NULL) {
- free(cursor);
+ if (mc != NULL) {
+ free(mc);
}
}
return MDB_SUCCESS;
}
-/* Move a node from src to dst.
+/* Move a node from csrc to cdst.
*/
static int
-mdb_move_node(MDB_txn *txn, MDB_dbi dbi, MDB_pageparent *src, indx_t srcindx,
- MDB_pageparent *dst, indx_t dstindx)
+mdb_move_node(MDB_cursor *csrc, MDB_cursor *cdst)
{
int rc;
MDB_node *srcnode;
DKBUF;
/* Mark src and dst as dirty. */
- if ((rc = mdb_touch(txn, dbi, src)) ||
- (rc = mdb_touch(txn, dbi, dst)))
+ if ((rc = mdb_touch(csrc)) ||
+ (rc = mdb_touch(cdst)))
return rc;;
- if (IS_LEAF2(src->mp_page)) {
- srcnode = NODEPTR(src->mp_page, 0); /* fake */
- key.mv_size = txn->mt_dbs[dbi].md_pad;
- key.mv_data = LEAF2KEY(src->mp_page, srcindx, key.mv_size);
+ 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_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;
} else {
- srcnode = NODEPTR(src->mp_page, srcindx);
+ if (csrc->mc_ki[csrc->mc_top] == 0 && IS_BRANCH(csrc->mc_pg[csrc->mc_top])) {
+ unsigned int snum = csrc->mc_snum;
+ /* must find the lowest key below src */
+ mdb_search_page_root(csrc, NULL, 0);
+ srcnode = NODEPTR(csrc->mc_pg[csrc->mc_top], 0);
+ 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);
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",
- IS_LEAF(src->mp_page) ? "leaf" : "branch",
- srcindx,
+ IS_LEAF(csrc->mc_pg[csrc->mc_top]) ? "leaf" : "branch",
+ csrc->mc_ki[csrc->mc_top],
DKEY(&key),
- src->mp_page->mp_pgno,
- dstindx, dst->mp_page->mp_pgno);
+ csrc->mc_pg[csrc->mc_top]->mp_pgno,
+ cdst->mc_ki[cdst->mc_top], cdst->mc_pg[cdst->mc_top]->mp_pgno);
/* Add the node to the destination page.
*/
- rc = mdb_add_node(txn, dbi, dst->mp_page, dstindx, &key, &data, NODEPGNO(srcnode),
+ rc = mdb_add_node(cdst, cdst->mc_ki[cdst->mc_top], &key, &data, NODEPGNO(srcnode),
srcnode->mn_flags);
if (rc != MDB_SUCCESS)
return rc;
/* Delete the node from the source page.
*/
- mdb_del_node(src->mp_page, srcindx, key.mv_size);
-
- /* The key value just changed due to del_node, find it again.
- */
- if (!IS_LEAF2(src->mp_page)) {
- srcnode = NODEPTR(src->mp_page, srcindx);
- key.mv_data = NODEKEY(srcnode);
- }
+ mdb_del_node(csrc->mc_pg[csrc->mc_top], csrc->mc_ki[csrc->mc_top], key.mv_size);
/* Update the parent separators.
*/
- if (srcindx == 0) {
- if (src->mp_pi != 0) {
+ 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);
+ } 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);
+ }
DPRINTF("update separator for source page %lu to [%s]",
- src->mp_page->mp_pgno, DKEY(&key));
- if ((rc = mdb_update_key(src->mp_parent, src->mp_pi,
+ 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)
return rc;
}
- if (IS_BRANCH(src->mp_page)) {
+ if (IS_BRANCH(csrc->mc_pg[csrc->mc_top])) {
MDB_val nullkey;
nullkey.mv_size = 0;
- assert(mdb_update_key(src->mp_page, 0, &nullkey) == MDB_SUCCESS);
+ assert(mdb_update_key(csrc->mc_pg[csrc->mc_top], 0, &nullkey) == MDB_SUCCESS);
}
}
- if (dstindx == 0) {
- if (dst->mp_pi != 0) {
+ if (cdst->mc_ki[cdst->mc_top] == 0) {
+ if (cdst->mc_ki[cdst->mc_top-1] != 0) {
+ if (IS_LEAF2(csrc->mc_pg[csrc->mc_top])) {
+ key.mv_data = LEAF2KEY(cdst->mc_pg[cdst->mc_top], 0, key.mv_size);
+ } else {
+ srcnode = NODEPTR(cdst->mc_pg[cdst->mc_top], 0);
+ key.mv_size = NODEKSZ(srcnode);
+ key.mv_data = NODEKEY(srcnode);
+ }
DPRINTF("update separator for destination page %lu to [%s]",
- dst->mp_page->mp_pgno, DKEY(&key));
- if ((rc = mdb_update_key(dst->mp_parent, dst->mp_pi,
+ 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)
return rc;
}
- if (IS_BRANCH(dst->mp_page)) {
+ if (IS_BRANCH(cdst->mc_pg[cdst->mc_top])) {
MDB_val nullkey;
nullkey.mv_size = 0;
- assert(mdb_update_key(dst->mp_page, 0, &nullkey) == MDB_SUCCESS);
+ assert(mdb_update_key(cdst->mc_pg[cdst->mc_top], 0, &nullkey) == MDB_SUCCESS);
}
}
}
static int
-mdb_merge(MDB_txn *txn, MDB_dbi dbi, MDB_pageparent *src, MDB_pageparent *dst)
+mdb_merge(MDB_cursor *csrc, MDB_cursor *cdst)
{
int rc;
- indx_t i;
+ indx_t i, j;
MDB_node *srcnode;
MDB_val key, data;
- MDB_pageparent mpp;
- MDB_dhead *dh;
- DPRINTF("merging page %lu and %lu", src->mp_page->mp_pgno, dst->mp_page->mp_pgno);
+ DPRINTF("merging page %lu into %lu", csrc->mc_pg[csrc->mc_top]->mp_pgno, cdst->mc_pg[cdst->mc_top]->mp_pgno);
- assert(txn != NULL);
- assert(src->mp_parent); /* can't merge root page */
- assert(dst->mp_parent);
+ assert(csrc->mc_snum > 1); /* can't merge root page */
+ assert(cdst->mc_snum > 1);
- /* Mark src and dst as dirty. */
- if ((rc = mdb_touch(txn, dbi, src)) ||
- (rc = mdb_touch(txn, dbi, dst)))
+ /* Mark dst as dirty. */
+ if ((rc = mdb_touch(cdst)))
return rc;
/* Move all nodes from src to dst.
*/
- if (IS_LEAF2(src->mp_page)) {
- key.mv_size = txn->mt_dbs[dbi].md_pad;
- key.mv_data = METADATA(src->mp_page);
- for (i = 0; i < NUMKEYS(src->mp_page); i++) {
- rc = mdb_add_node(txn, dbi, dst->mp_page, NUMKEYS(dst->mp_page), &key,
- NULL, 0, 0);
+ j = 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_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);
if (rc != MDB_SUCCESS)
return rc;
key.mv_data = (char *)key.mv_data + key.mv_size;
}
} else {
- for (i = 0; i < NUMKEYS(src->mp_page); i++) {
- srcnode = NODEPTR(src->mp_page, i);
+ for (i = 0; i < NUMKEYS(csrc->mc_pg[csrc->mc_top]); i++, j++) {
+ srcnode = NODEPTR(csrc->mc_pg[csrc->mc_top], i);
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(txn, dbi, dst->mp_page, NUMKEYS(dst->mp_page), &key,
- &data, NODEPGNO(srcnode), srcnode->mn_flags);
+ rc = mdb_add_node(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)",
- dst->mp_page->mp_pgno, NUMKEYS(dst->mp_page), (float)PAGEFILL(txn->mt_env, dst->mp_page) / 10);
+ 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.
+ /* Unlink the src page from parent and add to free list.
*/
- mdb_del_node(src->mp_parent, src->mp_pi, 0);
- if (src->mp_pi == 0) {
+ mdb_del_node(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(src->mp_parent, 0, &key)) != MDB_SUCCESS)
+ if ((rc = mdb_update_key(csrc->mc_pg[csrc->mc_top-1], 0, &key)) != MDB_SUCCESS)
return rc;
}
- if (IS_LEAF(src->mp_page))
- txn->mt_dbs[dbi].md_leaf_pages--;
+ 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--;
else
- txn->mt_dbs[dbi].md_branch_pages--;
+ csrc->mc_txn->mt_dbs[csrc->mc_dbi].md_branch_pages--;
+ cursor_pop_page(csrc);
- mpp.mp_page = src->mp_parent;
- dh = (MDB_dhead *)src->mp_parent;
- dh--;
- mpp.mp_parent = dh->md_parent;
- mpp.mp_pi = dh->md_pi;
-
- return mdb_rebalance(txn, dbi, &mpp);
+ return mdb_rebalance(csrc);
}
-#define FILL_THRESHOLD 250
+static void
+mdb_cursor_copy(const MDB_cursor *csrc, MDB_cursor *cdst)
+{
+ unsigned int i;
+
+ cdst->mc_txn = csrc->mc_txn;
+ cdst->mc_dbi = csrc->mc_dbi;
+ cdst->mc_snum = csrc->mc_snum;
+ cdst->mc_top = csrc->mc_top;
+ cdst->mc_flags = csrc->mc_flags;
+
+ for (i=0; i<csrc->mc_snum; i++) {
+ cdst->mc_pg[i] = csrc->mc_pg[i];
+ cdst->mc_ki[i] = csrc->mc_ki[i];
+ }
+}
static int
-mdb_rebalance(MDB_txn *txn, MDB_dbi dbi, MDB_pageparent *mpp)
+mdb_rebalance(MDB_cursor *mc)
{
MDB_node *node;
MDB_page *root;
- MDB_pageparent npp;
- indx_t si = 0, di = 0;
int rc;
-
- assert(txn != NULL);
- assert(mpp != NULL);
+ unsigned int ptop;
+ MDB_cursor mn;
DPRINTF("rebalancing %s page %lu (has %u keys, %.1f%% full)",
- IS_LEAF(mpp->mp_page) ? "leaf" : "branch",
- mpp->mp_page->mp_pgno, NUMKEYS(mpp->mp_page), (float)PAGEFILL(txn->mt_env, mpp->mp_page) / 10);
+ 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);
- if (PAGEFILL(txn->mt_env, mpp->mp_page) >= FILL_THRESHOLD) {
+ 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",
- mpp->mp_page->mp_pgno);
+ mc->mc_pg[mc->mc_top]->mp_pgno);
return MDB_SUCCESS;
}
- if (mpp->mp_parent == NULL) {
- if (NUMKEYS(mpp->mp_page) == 0) {
+ if (mc->mc_snum < 2) {
+ if (NUMKEYS(mc->mc_pg[mc->mc_top]) == 0) {
DPUTS("tree is completely empty");
- txn->mt_dbs[dbi].md_root = P_INVALID;
- txn->mt_dbs[dbi].md_depth = 0;
- txn->mt_dbs[dbi].md_leaf_pages = 0;
- } else if (IS_BRANCH(mpp->mp_page) && NUMKEYS(mpp->mp_page) == 1) {
+ 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) {
DPUTS("collapsing root page!");
- txn->mt_dbs[dbi].md_root = NODEPGNO(NODEPTR(mpp->mp_page, 0));
- if ((rc = mdb_get_page(txn, txn->mt_dbs[dbi].md_root, &root)))
+ 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)))
return rc;
- txn->mt_dbs[dbi].md_depth--;
- txn->mt_dbs[dbi].md_branch_pages--;
+ mc->mc_txn->mt_dbs[mc->mc_dbi].md_depth--;
+ mc->mc_txn->mt_dbs[mc->mc_dbi].md_branch_pages--;
} else
DPUTS("root page doesn't need rebalancing");
return MDB_SUCCESS;
/* The parent (branch page) must have at least 2 pointers,
* otherwise the tree is invalid.
*/
- assert(NUMKEYS(mpp->mp_parent) > 1);
+ ptop = mc->mc_top-1;
+ assert(NUMKEYS(mc->mc_pg[ptop]) > 1);
/* Leaf page fill factor is below the threshold.
* Try to move keys from left or right neighbor, or
/* Find neighbors.
*/
- if (mpp->mp_pi == 0) {
+ mdb_cursor_copy(mc, &mn);
+ mn.mc_xcursor = NULL;
+
+ if (mc->mc_ki[ptop] == 0) {
/* We're the leftmost leaf in our parent.
*/
DPUTS("reading right neighbor");
- node = NODEPTR(mpp->mp_parent, mpp->mp_pi + 1);
- if ((rc = mdb_get_page(txn, NODEPGNO(node), &npp.mp_page)))
+ 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])))
return rc;
- npp.mp_pi = mpp->mp_pi + 1;
- si = 0;
- di = NUMKEYS(mpp->mp_page);
+ mn.mc_ki[mn.mc_top] = 0;
+ mc->mc_ki[mc->mc_top] = NUMKEYS(mc->mc_pg[mc->mc_top]);
} else {
/* There is at least one neighbor to the left.
*/
DPUTS("reading left neighbor");
- node = NODEPTR(mpp->mp_parent, mpp->mp_pi - 1);
- if ((rc = mdb_get_page(txn, NODEPGNO(node), &npp.mp_page)))
+ 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])))
return rc;
- npp.mp_pi = mpp->mp_pi - 1;
- si = NUMKEYS(npp.mp_page) - 1;
- di = 0;
+ mn.mc_ki[mn.mc_top] = NUMKEYS(mn.mc_pg[mn.mc_top]) - 1;
+ mc->mc_ki[mc->mc_top] = 0;
}
- npp.mp_parent = mpp->mp_parent;
DPRINTF("found neighbor page %lu (%u keys, %.1f%% full)",
- npp.mp_page->mp_pgno, NUMKEYS(npp.mp_page), (float)PAGEFILL(txn->mt_env, npp.mp_page) / 10);
+ 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
* keys, move one key from it.
*
* Otherwise we should try to merge them.
*/
- if (PAGEFILL(txn->mt_env, npp.mp_page) >= FILL_THRESHOLD && NUMKEYS(npp.mp_page) >= 2)
- return mdb_move_node(txn, dbi, &npp, si, mpp, di);
+ 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);
else { /* FIXME: if (has_enough_room()) */
- if (mpp->mp_pi == 0)
- return mdb_merge(txn, dbi, &npp, mpp);
+ if (mc->mc_ki[ptop] == 0)
+ return mdb_merge(&mn, mc);
else
- return mdb_merge(txn, dbi, mpp, &npp);
+ return mdb_merge(mc, &mn);
}
}
static int
-mdb_del0(MDB_txn *txn, MDB_dbi dbi, unsigned int ki, MDB_pageparent *mpp, MDB_node *leaf)
+mdb_del0(MDB_cursor *mc, MDB_node *leaf)
{
int rc;
/* add overflow pages to free list */
- if (!IS_LEAF2(mpp->mp_page) && F_ISSET(leaf->mn_flags, F_BIGDATA)) {
+ if (!IS_LEAF2(mc->mc_pg[mc->mc_top]) && F_ISSET(leaf->mn_flags, F_BIGDATA)) {
int i, ovpages;
pgno_t pg;
memcpy(&pg, NODEDATA(leaf), sizeof(pg));
- ovpages = OVPAGES(NODEDSZ(leaf), txn->mt_env->me_psize);
+ ovpages = OVPAGES(NODEDSZ(leaf), mc->mc_txn->mt_env->me_psize);
for (i=0; i<ovpages; i++) {
DPRINTF("freed ov page %lu", pg);
- mdb_midl_insert(txn->mt_free_pgs, pg);
+ mdb_midl_append(mc->mc_txn->mt_free_pgs, pg);
pg++;
}
}
- mdb_del_node(mpp->mp_page, ki, txn->mt_dbs[dbi].md_pad);
- txn->mt_dbs[dbi].md_entries--;
- rc = mdb_rebalance(txn, dbi, mpp);
+ 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--;
+ rc = mdb_rebalance(mc);
if (rc != MDB_SUCCESS)
- txn->mt_flags |= MDB_TXN_ERROR;
+ mc->mc_txn->mt_flags |= MDB_TXN_ERROR;
return rc;
}
mdb_del(MDB_txn *txn, MDB_dbi dbi,
MDB_val *key, MDB_val *data)
{
+ MDB_cursor mc;
+ MDB_xcursor mx;
+ MDB_cursor_op op;
+ MDB_val rdata, *xdata;
int rc, exact;
- unsigned int ki;
- MDB_node *leaf;
- MDB_pageparent mpp;
DKBUF;
assert(key != NULL);
return EINVAL;
}
- mpp.mp_parent = NULL;
- mpp.mp_pi = 0;
- if ((rc = mdb_search_page(txn, dbi, key, NULL, 1, &mpp)) != MDB_SUCCESS)
- return rc;
-
- leaf = mdb_search_node(txn, dbi, mpp.mp_page, key, &exact, &ki);
- if (leaf == NULL || !exact) {
- return MDB_NOTFOUND;
- }
-
- if (!IS_LEAF2(mpp.mp_page) && F_ISSET(leaf->mn_flags, F_DUPDATA)) {
- MDB_cursor mc;
- MDB_xcursor mx;
- MDB_pageparent mp2;
-
- mc.mc_txn = txn;
- mc.mc_dbi = dbi;
- mc.mc_initialized = 0;
+ 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);
- mdb_xcursor_init1(txn, dbi, &mx, mpp.mp_page, leaf);
- if (data) {
- rc = mdb_del(&mx.mx_txn, mx.mx_cursor.mc_dbi, data, NULL);
- mdb_xcursor_fini(txn, dbi, &mx);
- /* If sub-DB still has entries, we're done */
- if (mx.mx_txn.mt_dbs[mx.mx_cursor.mc_dbi].md_root != P_INVALID) {
- memcpy(NODEDATA(leaf), &mx.mx_txn.mt_dbs[mx.mx_cursor.mc_dbi],
- sizeof(MDB_db));
- txn->mt_dbs[dbi].md_entries--;
- return rc;
- }
- /* otherwise fall thru and delete the sub-DB */
- } else {
- /* add all the child DB's pages to the free list */
- rc = mdb_search_page(&mx.mx_txn, mx.mx_cursor.mc_dbi,
- NULL, &mx.mx_cursor, 0, &mp2);
- if (rc == MDB_SUCCESS) {
- MDB_ppage *top, *parent;
- MDB_node *ni;
- unsigned int i;
-
- cursor_pop_page(&mx.mx_cursor);
- if (mx.mx_cursor.mc_snum) {
- top = CURSOR_TOP(&mx.mx_cursor);
- while (mx.mx_cursor.mc_snum > 1) {
- parent = CURSOR_PARENT(&mx.mx_cursor);
- for (i=0; i<NUMKEYS(top->mp_page); i++) {
- ni = NODEPTR(top->mp_page, i);
- mdb_midl_insert(txn->mt_free_pgs, NODEPGNO(ni));
- }
- parent->mp_ki++;
- if (parent->mp_ki >= NUMKEYS(parent->mp_page)) {
- cursor_pop_page(&mx.mx_cursor);
- top = parent;
- } else {
- ni = NODEPTR(parent->mp_page, parent->mp_ki);
- rc = mdb_get_page(&mx.mx_txn, NODEPGNO(ni), &top->mp_page);
- }
- }
- }
- mdb_midl_insert(txn->mt_free_pgs, mx.mx_txn.mt_dbs[mx.mx_cursor.mc_dbi].md_root);
- }
- }
+ } else {
+ mc.mc_xcursor = NULL;
}
- return mdb_del0(txn, dbi, ki, &mpp, leaf);
+ exact = 0;
+ if (data) {
+ op = MDB_GET_BOTH;
+ rdata = *data;
+ xdata = &rdata;
+ } else {
+ op = MDB_SET;
+ xdata = NULL;
+ }
+ rc = mdb_cursor_set(&mc, key, xdata, op, &exact);
+ if (rc == 0)
+ rc = mdb_cursor_del(&mc, data ? 0 : MDB_NODUPDATA);
+ return rc;
}
-/* Split page <*mpp>, and insert <key,(data|newpgno)> in either left or
- * right sibling, at index <*newindxp> (as if unsplit). Updates *mpp and
- * *newindxp with the actual values after split, ie if *mpp and *newindxp
+/* Split page <mc->top>, and insert <key,(data|newpgno)> in either left or
+ * right sibling, at index <mc->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.
*/
static int
-mdb_split(MDB_txn *txn, MDB_dbi dbi, MDB_page **mpp, unsigned int *newindxp,
- MDB_val *newkey, MDB_val *newdata, pgno_t newpgno)
+mdb_split(MDB_cursor *mc, MDB_val *newkey, MDB_val *newdata, pgno_t newpgno)
{
uint8_t flags;
int rc = MDB_SUCCESS, ins_new = 0;
unsigned int i, j, split_indx, nkeys, pmax;
MDB_node *node;
MDB_val sepkey, rkey, rdata;
- MDB_page *copy, *cptr;
- MDB_dpage *mdp, *rdp, *pdp;
- MDB_dhead *dh;
+ MDB_page *copy;
+ MDB_page *mp, *rp, *pp;
+ unsigned int ptop;
+ MDB_cursor mn;
DKBUF;
- assert(txn != NULL);
-
- dh = ((MDB_dhead *)*mpp) - 1;
- mdp = (MDB_dpage *)dh;
- newindx = *newindxp;
+ 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",
- IS_LEAF(&mdp->p) ? "leaf" : "branch", mdp->p.mp_pgno,
- DKEY(newkey), *newindxp);
+ IS_LEAF(mp) ? "leaf" : "branch", mp->mp_pgno,
+ DKEY(newkey), mc->mc_ki[mc->mc_top]);
- if (mdp->h.md_parent == NULL) {
- if ((pdp = mdb_new_page(txn, dbi, P_BRANCH, 1)) == NULL)
+ if (mc->mc_snum < 2) {
+ if ((pp = mdb_new_page(mc, P_BRANCH, 1)) == NULL)
return ENOMEM;
- mdp->h.md_pi = 0;
- mdp->h.md_parent = &pdp->p;
- txn->mt_dbs[dbi].md_root = pdp->p.mp_pgno;
- DPRINTF("root split! new root = %lu", pdp->p.mp_pgno);
- txn->mt_dbs[dbi].md_depth++;
+ /* 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++;
/* Add left (implicit) pointer. */
- if ((rc = mdb_add_node(txn, dbi, &pdp->p, 0, NULL, NULL,
- mdp->p.mp_pgno, 0)) != MDB_SUCCESS)
+ if ((rc = mdb_add_node(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--;
return rc;
+ }
+ mc->mc_snum = 2;
+ mc->mc_top = 1;
+ ptop = 0;
} else {
- DPRINTF("parent branch page is %lu", mdp->h.md_parent->mp_pgno);
+ ptop = mc->mc_top-1;
+ DPRINTF("parent branch page is %lu", mc->mc_pg[ptop]->mp_pgno);
}
/* Create a right sibling. */
- if ((rdp = mdb_new_page(txn, dbi, mdp->p.mp_flags, 1)) == NULL)
+ if ((rp = mdb_new_page(mc, mp->mp_flags, 1)) == NULL)
return ENOMEM;
- rdp->h.md_parent = mdp->h.md_parent;
- rdp->h.md_pi = mdp->h.md_pi + 1;
- DPRINTF("new right sibling: page %lu", rdp->p.mp_pgno);
+ 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);
- nkeys = NUMKEYS(&mdp->p);
+ nkeys = NUMKEYS(mp);
split_indx = nkeys / 2 + 1;
- if (IS_LEAF2(&rdp->p)) {
+ if (IS_LEAF2(rp)) {
char *split, *ins;
int x;
unsigned int lsize, rsize, ksize;
/* Move half of the keys to the right sibling */
copy = NULL;
- x = *newindxp - split_indx;
- ksize = txn->mt_dbs[dbi].md_pad;
- split = LEAF2KEY(&mdp->p, split_indx, ksize);
+ x = mc->mc_ki[mc->mc_top] - split_indx;
+ ksize = mc->mc_txn->mt_dbs[mc->mc_dbi].md_pad;
+ split = LEAF2KEY(mp, split_indx, ksize);
rsize = (nkeys - split_indx) * ksize;
lsize = (nkeys - split_indx) * sizeof(indx_t);
- mdp->p.mp_lower -= lsize;
- rdp->p.mp_lower += lsize;
- mdp->p.mp_upper += rsize - lsize;
- rdp->p.mp_upper -= rsize - lsize;
+ mp->mp_lower -= lsize;
+ rp->mp_lower += lsize;
+ mp->mp_upper += rsize - lsize;
+ rp->mp_upper -= rsize - lsize;
sepkey.mv_size = ksize;
if (newindx == split_indx) {
sepkey.mv_data = newkey->mv_data;
sepkey.mv_data = split;
}
if (x<0) {
- ins = LEAF2KEY(&mdp->p, *newindxp, ksize);
- memcpy(&rdp->p.mp_ptrs, split, rsize);
- sepkey.mv_data = &rdp->p.mp_ptrs;
- memmove(ins+ksize, ins, (split_indx - *newindxp) * ksize);
+ ins = LEAF2KEY(mp, mc->mc_ki[mc->mc_top], ksize);
+ memcpy(rp->mp_ptrs, split, rsize);
+ sepkey.mv_data = rp->mp_ptrs;
+ memmove(ins+ksize, ins, (split_indx - mc->mc_ki[mc->mc_top]) * ksize);
memcpy(ins, newkey->mv_data, ksize);
- mdp->p.mp_lower += sizeof(indx_t);
- mdp->p.mp_upper -= ksize - sizeof(indx_t);
+ mp->mp_lower += sizeof(indx_t);
+ mp->mp_upper -= ksize - sizeof(indx_t);
} else {
if (x)
- memcpy(&rdp->p.mp_ptrs, split, x * ksize);
- ins = LEAF2KEY(&rdp->p, x, ksize);
+ memcpy(rp->mp_ptrs, split, x * ksize);
+ ins = LEAF2KEY(rp, x, ksize);
memcpy(ins, newkey->mv_data, ksize);
memcpy(ins+ksize, split + x * ksize, rsize - x * ksize);
- rdp->p.mp_lower += sizeof(indx_t);
- rdp->p.mp_upper -= ksize - sizeof(indx_t);
- *newindxp = x;
- *mpp = &rdp->p;
+ rp->mp_lower += sizeof(indx_t);
+ rp->mp_upper -= ksize - sizeof(indx_t);
+ mc->mc_ki[mc->mc_top] = x;
+ mc->mc_pg[mc->mc_top] = rp;
}
goto newsep;
}
/* For leaf pages, check the split point based on what
* fits where, since otherwise add_node can fail.
*/
- if (IS_LEAF(&mdp->p)) {
+ if (IS_LEAF(mp)) {
unsigned int psize, nsize;
/* Maximum free space in an empty page */
- pmax = txn->mt_env->me_psize - PAGEHDRSZ;
- nsize = mdb_leaf_size(txn->mt_env, newkey, newdata);
+ pmax = mc->mc_txn->mt_env->me_psize - PAGEHDRSZ;
+ nsize = mdb_leaf_size(mc->mc_txn->mt_env, newkey, newdata);
if (newindx < split_indx) {
psize = nsize;
for (i=0; i<split_indx; i++) {
- node = NODEPTR(&mdp->p, i);
+ node = NODEPTR(mp, i);
psize += NODESIZE + NODEKSZ(node) + sizeof(indx_t);
if (F_ISSET(node->mn_flags, F_BIGDATA))
psize += sizeof(pgno_t);
else
psize += NODEDSZ(node);
+ psize += psize & 1;
if (psize > pmax) {
split_indx = i;
break;
} else {
psize = nsize;
for (i=nkeys-1; i>=split_indx; i--) {
- node = NODEPTR(&mdp->p, i);
+ node = NODEPTR(mp, i);
psize += NODESIZE + NODEKSZ(node) + sizeof(indx_t);
if (F_ISSET(node->mn_flags, F_BIGDATA))
psize += sizeof(pgno_t);
else
psize += NODEDSZ(node);
+ psize += psize & 1;
if (psize > pmax) {
split_indx = i+1;
break;
/* First find the separating key between the split pages.
*/
- memset(&sepkey, 0, sizeof(sepkey));
if (newindx == split_indx) {
sepkey.mv_size = newkey->mv_size;
sepkey.mv_data = newkey->mv_data;
} else {
- node = NODEPTR(&mdp->p, split_indx);
+ node = NODEPTR(mp, split_indx);
sepkey.mv_size = node->mn_ksize;
sepkey.mv_data = NODEKEY(node);
}
/* Copy separator key to the parent.
*/
- if (SIZELEFT(rdp->h.md_parent) < mdb_branch_size(txn->mt_env, &sepkey)) {
- rc = mdb_split(txn, dbi, &rdp->h.md_parent, &rdp->h.md_pi,
- &sepkey, NULL, rdp->p.mp_pgno);
+ 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);
/* Right page might now have changed parent.
* Check if left page also changed parent.
*/
- if (rdp->h.md_parent != mdp->h.md_parent &&
- mdp->h.md_pi >= NUMKEYS(mdp->h.md_parent)) {
- mdp->h.md_parent = rdp->h.md_parent;
- mdp->h.md_pi = rdp->h.md_pi - 1;
+ if (mn.mc_pg[ptop] != mc->mc_pg[ptop] &&
+ mc->mc_ki[ptop] >= NUMKEYS(mc->mc_pg[ptop])) {
+ mc->mc_pg[ptop] = mn.mc_pg[ptop];
+ mc->mc_ki[ptop] = mn.mc_ki[ptop] - 1;
}
} else {
- rc = mdb_add_node(txn, dbi, rdp->h.md_parent, rdp->h.md_pi,
- &sepkey, NULL, rdp->p.mp_pgno, 0);
+ mn.mc_top--;
+ rc = mdb_add_node(&mn, mn.mc_ki[ptop], &sepkey, NULL, rp->mp_pgno, 0);
+ mn.mc_top++;
}
- if (IS_LEAF2(&rdp->p)) {
+ if (IS_LEAF2(rp)) {
return rc;
}
if (rc != MDB_SUCCESS) {
}
/* Move half of the keys to the right sibling. */
- if ((copy = malloc(txn->mt_env->me_psize)) == NULL)
- return ENOMEM;
- copy->mp_pgno = mdp->p.mp_pgno;
- copy->mp_flags = mdp->p.mp_flags;
+ /* 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->mp_pgno = mp->mp_pgno;
+ copy->mp_flags = mp->mp_flags;
copy->mp_lower = PAGEHDRSZ;
- copy->mp_upper = txn->mt_env->me_psize;
- cptr = copy;
+ copy->mp_upper = mc->mc_txn->mt_env->me_psize;
+ mc->mc_pg[mc->mc_top] = copy;
for (i = j = 0; i <= nkeys; j++) {
if (i == split_indx) {
/* Insert in right sibling. */
/* Reset insert index for right sibling. */
j = (i == newindx && ins_new);
- cptr = &rdp->p;
+ mc->mc_pg[mc->mc_top] = rp;
}
if (i == newindx && !ins_new) {
/* Insert the original entry that caused the split. */
rkey.mv_data = newkey->mv_data;
rkey.mv_size = newkey->mv_size;
- if (IS_LEAF(&mdp->p)) {
+ if (IS_LEAF(mp)) {
rdata.mv_data = newdata->mv_data;
rdata.mv_size = newdata->mv_size;
} else
ins_new = 1;
/* Update page and index for the new key. */
- *newindxp = j;
- if (cptr == &rdp->p)
- *mpp = cptr;
+ mc->mc_ki[mc->mc_top] = j;
} else if (i == nkeys) {
break;
} else {
- node = NODEPTR(&mdp->p, i);
+ node = NODEPTR(mp, i);
rkey.mv_data = NODEKEY(node);
rkey.mv_size = node->mn_ksize;
- if (IS_LEAF(&mdp->p)) {
+ if (IS_LEAF(mp)) {
rdata.mv_data = NODEDATA(node);
- rdata.mv_size = node->mn_dsize;
+ rdata.mv_size = NODEDSZ(node);
} else
pgno = NODEPGNO(node);
flags = node->mn_flags;
i++;
}
- if (!IS_LEAF(&mdp->p) && j == 0) {
+ if (!IS_LEAF(mp) && j == 0) {
/* First branch index doesn't need key data. */
rkey.mv_size = 0;
}
- rc = mdb_add_node(txn, dbi, cptr, j, &rkey, &rdata, pgno, flags);
+ rc = mdb_add_node(mc, j, &rkey, &rdata, pgno, flags);
}
- nkeys = NUMKEYS(copy);
- for (i=0; i<nkeys; i++)
- mdp->p.mp_ptrs[i] = copy->mp_ptrs[i];
- mdp->p.mp_lower = copy->mp_lower;
- mdp->p.mp_upper = copy->mp_upper;
- memcpy(NODEPTR(&mdp->p, nkeys-1), NODEPTR(copy, nkeys-1),
- txn->mt_env->me_psize - copy->mp_upper);
- free(copy);
- return rc;
-}
-
-static int
-mdb_put0(MDB_txn *txn, MDB_dbi dbi,
- MDB_val *key, MDB_val *data, unsigned int flags)
-{
- int rc = MDB_SUCCESS, exact;
- unsigned int ki;
- MDB_node *leaf;
- MDB_pageparent mpp;
- MDB_val xdata, *rdata, dkey;
- MDB_db dummy;
- char dbuf[PAGESIZE];
- int do_sub = 0;
- size_t nsize;
- DKBUF;
-
- DPRINTF("==> put db %u key [%s], size %zu, data size %zu",
- dbi, DKEY(key), key->mv_size, data->mv_size);
+ /* reset back to original page */
+ if (newindx < split_indx)
+ mc->mc_pg[mc->mc_top] = mp;
- dkey.mv_size = 0;
- mpp.mp_parent = NULL;
- mpp.mp_pi = 0;
- rc = mdb_search_page(txn, dbi, key, NULL, 1, &mpp);
- if (rc == MDB_SUCCESS) {
- leaf = mdb_search_node(txn, dbi, mpp.mp_page, key, &exact, &ki);
- if (leaf && exact) {
- if (flags == MDB_NOOVERWRITE) {
- DPRINTF("duplicate key [%s]", DKEY(key));
- return MDB_KEYEXIST;
- }
- /* there's only a key anyway, so this is a no-op */
- if (IS_LEAF2(mpp.mp_page))
- return MDB_SUCCESS;
-
- if (F_ISSET(txn->mt_dbs[dbi].md_flags, MDB_DUPSORT)) {
- /* Was a single item before, must convert now */
- if (!F_ISSET(leaf->mn_flags, F_DUPDATA)) {
- dkey.mv_size = NODEDSZ(leaf);
- dkey.mv_data = dbuf;
- memcpy(dbuf, NODEDATA(leaf), dkey.mv_size);
- /* data matches, ignore it */
- if (!mdb_dcmp(txn, dbi, data, &dkey))
- return (flags == MDB_NODUPDATA) ? MDB_KEYEXIST : MDB_SUCCESS;
- memset(&dummy, 0, sizeof(dummy));
- if (txn->mt_dbs[dbi].md_flags & MDB_DUPFIXED) {
- dummy.md_pad = data->mv_size;
- dummy.md_flags = MDB_DUPFIXED;
- if (txn->mt_dbs[dbi].md_flags & MDB_INTEGERDUP)
- dummy.md_flags |= MDB_INTEGERKEY;
- }
- dummy.md_root = P_INVALID;
- if (dkey.mv_size == sizeof(MDB_db)) {
- memcpy(NODEDATA(leaf), &dummy, sizeof(dummy));
- goto put_sub;
- }
- mdb_del_node(mpp.mp_page, ki, 0);
- do_sub = 1;
- rdata = &xdata;
- xdata.mv_size = sizeof(MDB_db);
- xdata.mv_data = &dummy;
- goto new_sub;
- }
- 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);
- goto done;
- }
- mdb_del_node(mpp.mp_page, ki, 0);
- }
- if (leaf == NULL) { /* append if not found */
- ki = NUMKEYS(mpp.mp_page);
- DPRINTF("appending key at index %i", ki);
- }
- } else if (rc == MDB_NOTFOUND) {
- MDB_dpage *dp;
- /* new file, just write a root leaf page */
- DPUTS("allocating new root leaf page");
- if ((dp = mdb_new_page(txn, dbi, P_LEAF, 1)) == NULL) {
- return ENOMEM;
- }
- mpp.mp_page = &dp->p;
- txn->mt_dbs[dbi].md_root = mpp.mp_page->mp_pgno;
- txn->mt_dbs[dbi].md_depth++;
- txn->mt_dbxs[dbi].md_dirty = 1;
- if ((txn->mt_dbs[dbi].md_flags & (MDB_DUPSORT|MDB_DUPFIXED)) == MDB_DUPFIXED)
- mpp.mp_page->mp_flags |= P_LEAF2;
- ki = 0;
- }
- else
- goto done;
-
- assert(IS_LEAF(mpp.mp_page));
- DPRINTF("there are %u keys, should insert new key at index %i",
- NUMKEYS(mpp.mp_page), ki);
-
- rdata = data;
-
-new_sub:
- nsize = IS_LEAF2(mpp.mp_page) ? key->mv_size : mdb_leaf_size(txn->mt_env, key, rdata);
- if (SIZELEFT(mpp.mp_page) < nsize) {
- rc = mdb_split(txn, dbi, &mpp.mp_page, &ki, key, rdata, P_INVALID);
- } else {
- /* There is room already in this leaf page. */
- rc = mdb_add_node(txn, dbi, mpp.mp_page, ki, key, rdata, 0, 0);
- }
-
- if (rc != MDB_SUCCESS)
- txn->mt_flags |= MDB_TXN_ERROR;
- else {
- /* Remember if we just added a subdatabase */
- if (flags & F_SUBDATA) {
- leaf = NODEPTR(mpp.mp_page, ki);
- 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) {
- MDB_cursor mc;
- MDB_xcursor mx;
-
- leaf = NODEPTR(mpp.mp_page, ki);
-put_sub:
- mc.mc_txn = txn;
- mc.mc_dbi = dbi;
- mc.mc_initialized = 0;
- mc.mc_xcursor = &mx;
- mdb_xcursor_init0(&mc);
- mdb_xcursor_init1(txn, dbi, &mx, mpp.mp_page, leaf);
- xdata.mv_size = 0;
- xdata.mv_data = "";
- if (flags == MDB_NODUPDATA)
- flags = MDB_NOOVERWRITE;
- /* converted, write the original data first */
- if (dkey.mv_size) {
- rc = mdb_put0(&mx.mx_txn, mx.mx_cursor.mc_dbi, &dkey, &xdata, flags);
- if (rc) return rc;
- leaf->mn_flags |= F_DUPDATA;
- }
- rc = mdb_put0(&mx.mx_txn, mx.mx_cursor.mc_dbi, data, &xdata, flags);
- mdb_xcursor_fini(txn, dbi, &mx);
- memcpy(NODEDATA(leaf), &mx.mx_txn.mt_dbs[mx.mx_cursor.mc_dbi],
- sizeof(MDB_db));
- }
- txn->mt_dbs[dbi].md_entries++;
- }
-
-done:
+ nkeys = NUMKEYS(copy);
+ for (i=0; i<nkeys; i++)
+ mp->mp_ptrs[i] = copy->mp_ptrs[i];
+ mp->mp_lower = copy->mp_lower;
+ mp->mp_upper = copy->mp_upper;
+ memcpy(NODEPTR(mp, nkeys-1), NODEPTR(copy, nkeys-1),
+ mc->mc_txn->mt_env->me_psize - copy->mp_upper);
+
+ /* return tmp page to freelist */
+ copy->mp_next = mc->mc_txn->mt_env->me_dpages;
+ mc->mc_txn->mt_env->me_dpages = copy;
return rc;
}
{
MDB_cursor mc;
MDB_xcursor mx;
- int rc;
assert(key != NULL);
assert(data != NULL);
mc.mc_txn = txn;
mc.mc_dbi = dbi;
mc.mc_snum = 0;
- mc.mc_initialized = 0;
- mc.mc_eof = 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;
}
- rc = mdb_cursor_put(&mc, key, data, flags);
- if (mc.mc_xcursor) {
- MDB_ppage *top;
- MDB_node *leaf;
- mdb_xcursor_fini(txn, dbi, &mx);
- top = CURSOR_TOP(&mc);
- leaf = NODEPTR(top->mp_page, top->mp_ki);
- memcpy(NODEDATA(leaf), &mx.mx_txn.mt_dbs[mx.mx_cursor.mc_dbi],
- sizeof(MDB_db));
- }
- return rc;
+ return mdb_cursor_put(&mc, key, data, flags);
}
int
mdb_env_set_flags(MDB_env *env, unsigned int flag, int onoff)
{
+ /** 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)
if ((flag & CHANGEABLE) != flag)
return EINVAL;
return mdb_stat0(env, &env->me_metas[toggle]->mm_dbs[MAIN_DBI], arg);
}
+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;
+ else if (txn->mt_dbs[dbi].md_flags & MDB_INTEGERKEY)
+ txn->mt_dbxs[dbi].md_cmp = cintcmp;
+ else
+ txn->mt_dbxs[dbi].md_cmp = memncmp;
+
+ 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;
+ else
+ txn->mt_dbxs[dbi].md_dcmp = cintcmp;
+ } else if (txn->mt_dbs[dbi].md_flags & MDB_REVERSEDUP) {
+ txn->mt_dbxs[dbi].md_dcmp = memnrcmp;
+ } else {
+ txn->mt_dbxs[dbi].md_dcmp = memncmp;
+ }
+ } else {
+ txn->mt_dbxs[dbi].md_dcmp = NULL;
+ }
+}
+
int mdb_open(MDB_txn *txn, const char *name, unsigned int flags, MDB_dbi *dbi)
{
MDB_val key, data;
int rc, dirty = 0;
size_t len;
+ if (txn->mt_dbxs[FREE_DBI].md_cmp == NULL) {
+ mdb_default_cmp(txn, FREE_DBI);
+ }
+
/* main DB? */
if (!name) {
*dbi = MAIN_DBI;
if (flags & (MDB_DUPSORT|MDB_REVERSEKEY|MDB_INTEGERKEY))
txn->mt_dbs[MAIN_DBI].md_flags |= (flags & (MDB_DUPSORT|MDB_REVERSEKEY|MDB_INTEGERKEY));
+ mdb_default_cmp(txn, MAIN_DBI);
return MDB_SUCCESS;
}
+ if (txn->mt_dbxs[MAIN_DBI].md_cmp == NULL) {
+ mdb_default_cmp(txn, MAIN_DBI);
+ }
+
/* Is the DB already open? */
len = strlen(name);
for (i=2; i<txn->mt_numdbs; i++) {
/* Create if requested */
if (rc == MDB_NOTFOUND && (flags & MDB_CREATE)) {
+ MDB_cursor mc;
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;
- rc = mdb_put0(txn, MAIN_DBI, &key, &data, F_SUBDATA);
+ mc.mc_txn = txn;
+ mc.mc_dbi = MAIN_DBI;
+ mc.mc_flags = 0;
+ rc = mdb_cursor_put(&mc, &key, &data, F_SUBDATA);
dirty = 1;
}
if (rc == MDB_SUCCESS) {
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_cmp = NULL;
- txn->mt_dbxs[txn->mt_numdbs].md_dcmp = NULL;
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;
*dbi = txn->mt_numdbs;
txn->mt_env->me_dbs[0][txn->mt_numdbs] = txn->mt_dbs[txn->mt_numdbs];
txn->mt_env->me_dbs[1][txn->mt_numdbs] = txn->mt_dbs[txn->mt_numdbs];
+ mdb_default_cmp(txn, txn->mt_numdbs);
txn->mt_numdbs++;
}