From bce3e667dff9d6ab51f4e37bc79ef961e36bfb31 Mon Sep 17 00:00:00 2001 From: Ben Johnson Date: Tue, 21 Jan 2014 14:37:55 -0700 Subject: [PATCH] Intermediate commit. --- README.md | 34 ++ branch_node.go | 79 ---- bucket.go | 1 - cursor.go | 48 +-- db.go | 10 +- leaf_node.go => node.go | 16 +- page.go | 13 +- rwcursor.go | 861 ++++++++++++++++++++-------------------- rwtransaction.go | 2 - 9 files changed, 505 insertions(+), 559 deletions(-) rename leaf_node.go => node.go (57%) diff --git a/README.md b/README.md index e26dc46..0309695 100644 --- a/README.md +++ b/README.md @@ -1,4 +1,38 @@ bolt ==== +## Overview + A low-level key/value database for Go. + + +## API + +### DB + +### Creating a database + +``` +db := DB() +err := db.Open("/path/to/db", 0666) +... +err := db.Close() +``` + +### Creating a bucket + + +* Cursor + + +``` +DB +Bucket +Transaction / RWTransaction +Cursor / RWCursor + +page +meta +branchNode +leafNode +``` diff --git a/branch_node.go b/branch_node.go index 58ab5a5..a5bd071 100644 --- a/branch_node.go +++ b/branch_node.go @@ -10,86 +10,7 @@ const ( dupNode = 0x04 ) -// branchNode represents a node on a branch page. -type branchNode struct { - pgno uint32 - flags uint16 - keySize uint16 - data uintptr // Pointer to the beginning of the data. -} - // key returns a byte slice that of the key data. func (n *branchNode) key() []byte { return (*[MaxKeySize]byte)(unsafe.Pointer(&n.data))[:n.keySize] } - -func (n *branchNode) size() int { - return 0 // TODO: offsetof(MDB_node, mn_data) -} - -// TODO: #define INDXSIZE(k) (NODESIZE + ((k) == NULL ? 0 : (k)->mv_size)) -// TODO: #define LEAFSIZE(k, d) (NODESIZE + (k)->mv_size + (d)->mv_size) -// TODO: #define NODEPTR(p, i) ((MDB_node *)((char *)(p) + (p)->mp_ptrs[i])) -// TODO: #define NODEKEY(node) (void *)((node)->mn_data) -// TODO: #define NODEDATA(node) (void *)((char *)(node)->mn_data + (node)->mn_ksize) -// TODO: #define NODEPGNO(node) ((node)->mn_lo | ((pgno_t) (node)->mn_hi << 16) | (PGNO_TOPWORD ? ((pgno_t) (node)->mn_flags << PGNO_TOPWORD) : 0)) -// TODO: #define SETPGNO(node,pgno) do { (node)->mn_lo = (pgno) & 0xffff; (node)->mn_hi = (pgno) >> 16; if (PGNO_TOPWORD) (node)->mn_flags = (pgno) >> PGNO_TOPWORD; } while(0) -// TODO: #define NODEDSZ(node) ((node)->mn_lo | ((unsigned)(node)->mn_hi << 16)) -// TODO: #define SETDSZ(node,size) do { (node)->mn_lo = (size) & 0xffff; (node)->mn_hi = (size) >> 16;} while(0) -// TODO: #define NODEKSZ(node) ((node)->mn_ksize) - -// TODO: #define LEAF2KEY(p, i, ks) ((char *)(p) + PAGEHDRSZ + ((i)*(ks))) - -// TODO: #define MDB_GET_KEY(node, keyptr) { if ((keyptr) != NULL) { (keyptr)->mv_size = NODEKSZ(node); (keyptr)->mv_data = NODEKEY(node); } } -// TODO: #define MDB_GET_KEY2(node, key) { key.mv_size = NODEKSZ(node); key.mv_data = NODEKEY(node); } - -// Compact the main page after deleting a node on a subpage. -// @param[in] mp The main page to operate on. -// @param[in] indx The index of the subpage on the main page. -func (n *node) shrink(index int) { - /* - MDB_node *node; - MDB_page *sp, *xp; - char *base; - int nsize, delta; - indx_t i, numkeys, ptr; - - node = NODEPTR(mp, indx); - sp = (MDB_page *)NODEDATA(node); - delta = SIZELEFT(sp); - xp = (MDB_page *)((char *)sp + delta); - - // shift subpage upward - if (IS_LEAF2(sp)) { - nsize = NUMKEYS(sp) * sp->mp_pad; - if (nsize & 1) - return; // do not make the node uneven-sized - memmove(METADATA(xp), METADATA(sp), nsize); - } else { - int i; - numkeys = NUMKEYS(sp); - for (i=numkeys-1; i>=0; i--) - xp->mp_ptrs[i] = sp->mp_ptrs[i] - delta; - } - xp->mp_upper = sp->mp_lower; - xp->mp_lower = sp->mp_lower; - xp->mp_flags = sp->mp_flags; - xp->mp_pad = sp->mp_pad; - COPY_PGNO(xp->mp_pgno, mp->mp_pgno); - - nsize = NODEDSZ(node) - delta; - SETDSZ(node, nsize); - - // shift lower nodes upward - ptr = mp->mp_ptrs[indx]; - numkeys = NUMKEYS(mp); - for (i = 0; i < numkeys; i++) { - if (mp->mp_ptrs[i] <= ptr) - mp->mp_ptrs[i] += delta; - } - - base = (char *)mp + mp->mp_upper; - memmove(base + delta, base, ptr - mp->mp_upper + NODESIZE + NODEKSZ(node)); - mp->mp_upper += delta; - */ -} diff --git a/bucket.go b/bucket.go index f42db47..ece92b5 100644 --- a/bucket.go +++ b/bucket.go @@ -29,4 +29,3 @@ type bucket struct { entries uint64 root pgno } - diff --git a/cursor.go b/cursor.go index d63cf94..13a56cf 100644 --- a/cursor.go +++ b/cursor.go @@ -5,31 +5,19 @@ package bolt const ( c_initialized = 0x01 /**< cursor has been initialized and is valid */ c_eof = 0x02 /**< No more data */ - c_sub = 0x04 /**< Cursor is a sub-cursor */ c_del = 0x08 /**< last op was a cursor_del */ c_splitting = 0x20 /**< Cursor is in page_split */ c_untrack = 0x40 /**< Un-track cursor when closing */ ) -// TODO: #define MDB_NOSPILL 0x8000 /** Do not spill pages to disk if txn is getting full, may fail instead */ - /* type Cursor interface { First() error - FirstDup() error - Get() ([]byte, []byte, error) - GetRange() ([]byte, []byte, error) - Current() ([]byte, []byte, error) - Last() - LastDup() + Last() error Next() ([]byte, []byte, error) - NextDup() ([]byte, []byte, error) - NextNoDup() ([]byte, []byte, error) Prev() ([]byte, []byte, error) - PrevDup() ([]byte, []byte, error) - PrevNoDup() ([]byte, []byte, error) - Set() ([]byte, []byte, error) - SetRange() ([]byte, []byte, error) + Current() ([]byte, []byte, error) + Get([]byte) ([]byte, error) } */ @@ -37,13 +25,11 @@ type Cursor struct { flags int next *Cursor backup *Cursor - subcursor *Cursor transaction *Transaction bucket *Bucket - subbucket *Bucket top int pages []*page - indices []int /* the index of the node for the page at the same level */ + indices []int /* the index of the node for the page at the same level */ } // , data []byte, op int @@ -53,12 +39,6 @@ func (c *Cursor) Get(key []byte) ([]byte, error) { int exact = 0; int (*mfunc)(MDB_cursor *mc, MDB_val *key, MDB_val *data); - if (mc == NULL) - return EINVAL; - - if (mc->mc_txn->mt_flags & MDB_TXN_ERROR) - return MDB_BAD_TXN; - switch (op) { case MDB_GET_CURRENT: if (!(mc->mc_flags & C_INITIALIZED)) { @@ -222,14 +202,14 @@ func (c *Cursor) page(key []byte, flags int) (*page, error) { if (flags & ps_first) != 0 { index = 0 } else if (flags & ps_last) != 0 { - index = indx(p.numkeys()) - 1; + index = indx(p.numkeys()) - 1 } else { - node, i, exact := p.find(key, c.transaction.db.pageSize); + node, i, exact := p.find(key, c.transaction.db.pageSize) if exact { c.indices[c.top] = i } if node == nil { - index = indx(p.numkeys()) - 1; + index = indx(p.numkeys()) - 1 } else { index = indx(c.indices[c.top]) if !exact { @@ -255,7 +235,7 @@ func (c *Cursor) page(key []byte, flags int) (*page, error) { } // If we ended up with a non-leaf page by the end then something is wrong. - if p.flags & p_leaf == 0 { + if p.flags&p_leaf == 0 { return nil, CorruptedError } @@ -267,7 +247,7 @@ func (c *Cursor) page(key []byte, flags int) (*page, error) { // pop moves the last page off the cursor's page stack. func (c *Cursor) pop() { - top := len(c.pages)-1 + top := len(c.pages) - 1 c.pages = c.pages[0:c.top] c.indices = c.indices[0:c.top] } @@ -279,8 +259,8 @@ func (c *Cursor) push(p *page) { c.top = len(c.pages) - 1 } -// page retrieves the last page on the page stack. -func (c *Cursor) page() *page { +// currentPage retrieves the last page on the page stack. +func (c *Cursor) currentPage() *page { top := len(c.pages) if top > 0 { return c.pages[top] @@ -306,7 +286,6 @@ func (c *Cursor) currentLeafNode() *node { return nil } - // // // // // // @@ -642,9 +621,6 @@ func (c *Cursor) page_touch() int { return 0 } - - - // Search for the lowest key under the current branch page. // This just bypasses a NUMKEYS check in the current page // before calling mdb_page_search_root(), because the callers @@ -1262,7 +1238,7 @@ func (c *Cursor) touch() error { } return rc; } -*/ + */ return nil } diff --git a/db.go b/db.go index 480d059..d7510c2 100644 --- a/db.go +++ b/db.go @@ -46,11 +46,11 @@ type DB struct { size int /**< current file size */ pbuf []byte transaction *RWTransaction /**< current write transaction */ - maxPageNumber int /**< me_mapsize / me_psize */ - dpages []*page /**< list of malloc'd blocks for re-use */ - freePages []int /** IDL of pages that became unused in a write txn */ - dirtyPages []int /** ID2L of pages written during a write txn. Length MDB_IDL_UM_SIZE. */ - maxFreeOnePage int /** Max number of freelist items that can fit in a single overflow page */ + maxPageNumber int /**< me_mapsize / me_psize */ + dpages []*page /**< list of malloc'd blocks for re-use */ + freePages []int /** IDL of pages that became unused in a write txn */ + dirtyPages []int /** ID2L of pages written during a write txn. Length MDB_IDL_UM_SIZE. */ + maxFreeOnePage int /** Max number of freelist items that can fit in a single overflow page */ maxPageDataSize int maxNodeSize int /** Max size of a node on a page */ maxKeySize int /**< max size of a key */ diff --git a/leaf_node.go b/node.go similarity index 57% rename from leaf_node.go rename to node.go index 80b5cf7..7a64459 100644 --- a/leaf_node.go +++ b/node.go @@ -4,14 +4,26 @@ import ( "unsafe" ) +// node represents a node on a page. +type node struct { + flags uint16 + keySize uint16 +} + // leafNode represents a node on a leaf page. type leafNode struct { - flags uint16 - keySize uint16 + node dataSize uint32 data uintptr // Pointer to the beginning of the data. } +// branchNode represents a node on a branch page. +type branchNode struct { + node + pgno uint32 + data uintptr // Pointer to the beginning of the data. +} + // key returns a byte slice that of the key data. func (n *leafNode) key() []byte { return (*[MaxKeySize]byte)(unsafe.Pointer(&n.data))[:n.keySize] diff --git a/page.go b/page.go index ecbaedd..8ecf920 100644 --- a/page.go +++ b/page.go @@ -20,8 +20,7 @@ const ( p_leaf = 0x02 p_overflow = 0x04 p_meta = 0x08 - p_dirty = 0x10 /**< dirty page, also set for #P_SUBP pages */ - p_sub = 0x40 + p_dirty = 0x10 /**< dirty page, also set for #P_SUBP pages */ p_keep = 0x8000 /**< leave this page alone during spill */ p_invalid = ^pgno(0) @@ -91,13 +90,13 @@ func (p *page) init(pageSize int) { // branchNode retrieves the branch node at the given index within the page. func (p *page) branchNode(index indx) *branchNode { b := (*[maxPageSize]byte)(unsafe.Pointer(&p.ptr)) - return (*branchNode)(unsafe.Pointer(&b[index * indx(unsafe.Sizeof(index))])) + return (*branchNode)(unsafe.Pointer(&b[index*indx(unsafe.Sizeof(index))])) } // leafNode retrieves the leaf node at the given index within the page. func (p *page) leafNode(index indx) *leafNode { b := (*[maxPageSize]byte)(unsafe.Pointer(&p.ptr)) - return (*leafNode)(unsafe.Pointer(&b[index * indx(unsafe.Sizeof(index))])) + return (*leafNode)(unsafe.Pointer(&b[index*indx(unsafe.Sizeof(index))])) } // numkeys returns the number of nodes in the page. @@ -117,20 +116,20 @@ func (p *page) find(key []byte, pageSize int) (*node, int, bool) { var node *node nkeys := p.numkeys() - low, high := 1, nkeys - 1 + low, high := 1, nkeys-1 if (p.flags & p_leaf) != 0 { low = 0 } // Perform a binary search to find the correct node. var i, rc int - for ; low <= high; { + for low <= high { i = (low + high) / 2 node = p.node(indx(i)) rc = bytes.Compare(key, node.key()) if rc == 0 { - break; + break } else if rc > 0 { low = i + 1 } else { diff --git a/rwcursor.go b/rwcursor.go index 553989f..cd5168b 100644 --- a/rwcursor.go +++ b/rwcursor.go @@ -1,11 +1,18 @@ package bolt +/* +type RWCursor interface { + Put([]byte, []byte) (error) + Delete([]byte) (error) +} +*/ + // RWCursor represents a cursor that can read and write data for a bucket. type RWCursor struct { Cursor transaction *RWTransaction - reclaimed []pgno /**< Reclaimed freeDB pages, or NULL before use (was me_pghead) */ - last txnid /**< ID of last used record, or 0 if len(reclaimed) == 0 */ + reclaimed []pgno /**< Reclaimed freeDB pages, or NULL before use (was me_pghead) */ + last txnid /**< ID of last used record, or 0 if len(reclaimed) == 0 */ } func (c *RWCursor) Put(key []byte, value []byte) error { @@ -61,339 +68,339 @@ func (c *RWCursor) Put(key []byte, value []byte) error { // If key does not exist the if exists { node := c.currentNode() - + } /* - insert = rc; - if (insert) { - // The key does not exist - DPRINTF(("inserting key at index %i", mc->mc_ki[mc->mc_top])); - if ((mc->mc_db->md_flags & MDB_DUPSORT) && - LEAFSIZE(key, data) > env->me_nodemax) - { - // Too big for a node, insert in sub-DB - fp_flags = P_LEAF|P_DIRTY; - fp = env->me_pbuf; - fp->mp_pad = data->mv_size; // used if MDB_DUPFIXED - fp->mp_lower = fp->mp_upper = olddata.mv_size = PAGEHDRSZ; - goto prep_subDB; - } - } else { - -more: - leaf = NODEPTR(mc->mc_pg[mc->mc_top], mc->mc_ki[mc->mc_top]); - olddata.mv_size = NODEDSZ(leaf); - olddata.mv_data = NODEDATA(leaf); - - // DB has dups? - if (F_ISSET(mc->mc_db->md_flags, MDB_DUPSORT)) { - // Prepare (sub-)page/sub-DB to accept the new item, - // if needed. fp: old sub-page or a header faking - // it. mp: new (sub-)page. offset: growth in page - // size. xdata: node data with new page or DB. - ssize_t i, offset = 0; - mp = fp = xdata.mv_data = env->me_pbuf; - mp->mp_pgno = mc->mc_pg[mc->mc_top]->mp_pgno; - - // Was a single item before, must convert now - if (!F_ISSET(leaf->mn_flags, F_DUPDATA)) { - // Just overwrite the current item - if (flags == MDB_CURRENT) - goto current; - -#if UINT_MAX < SIZE_MAX - if (mc->mc_dbx->md_dcmp == mdb_cmp_int && olddata.mv_size == sizeof(size_t)) -#ifdef MISALIGNED_OK - mc->mc_dbx->md_dcmp = mdb_cmp_long; -#else - mc->mc_dbx->md_dcmp = mdb_cmp_cint; -#endif -#endif - // if data matches, skip it - if (!mc->mc_dbx->md_dcmp(data, &olddata)) { - if (flags & MDB_NODUPDATA) - rc = MDB_KEYEXIST; - else if (flags & MDB_MULTIPLE) - goto next_mult; - else - rc = MDB_SUCCESS; - return rc; - } - - // Back up original data item - dkey.mv_size = olddata.mv_size; - dkey.mv_data = memcpy(fp+1, olddata.mv_data, olddata.mv_size); - - // Make sub-page header for the dup items, with dummy body - fp->mp_flags = P_LEAF|P_DIRTY|P_SUBP; - fp->mp_lower = PAGEHDRSZ; - xdata.mv_size = PAGEHDRSZ + dkey.mv_size + data->mv_size; - if (mc->mc_db->md_flags & MDB_DUPFIXED) { - fp->mp_flags |= P_LEAF2; - fp->mp_pad = data->mv_size; - xdata.mv_size += 2 * data->mv_size; // leave space for 2 more - } else { - xdata.mv_size += 2 * (sizeof(indx_t) + NODESIZE) + - (dkey.mv_size & 1) + (data->mv_size & 1); - } - fp->mp_upper = xdata.mv_size; - olddata.mv_size = fp->mp_upper; // pretend olddata is fp - } else if (leaf->mn_flags & F_SUBDATA) { - // Data is on sub-DB, just store it - flags |= F_DUPDATA|F_SUBDATA; - goto put_sub; - } else { - // Data is on sub-page - fp = olddata.mv_data; - switch (flags) { - default: - i = -(ssize_t)SIZELEFT(fp); - if (!(mc->mc_db->md_flags & MDB_DUPFIXED)) { - offset = i += (ssize_t) EVEN( - sizeof(indx_t) + NODESIZE + data->mv_size); - } else { - i += offset = fp->mp_pad; - offset *= 4; // space for 4 more - } - if (i > 0) - break; - // FALLTHRU: Sub-page is big enough - case MDB_CURRENT: - fp->mp_flags |= P_DIRTY; - COPY_PGNO(fp->mp_pgno, mp->mp_pgno); - mc->mc_xcursor->mx_cursor.mc_pg[0] = fp; - flags |= F_DUPDATA; - goto put_sub; - } - xdata.mv_size = olddata.mv_size + offset; + insert = rc; + if (insert) { + // The key does not exist + DPRINTF(("inserting key at index %i", mc->mc_ki[mc->mc_top])); + if ((mc->mc_db->md_flags & MDB_DUPSORT) && + LEAFSIZE(key, data) > env->me_nodemax) + { + // Too big for a node, insert in sub-DB + fp_flags = P_LEAF|P_DIRTY; + fp = env->me_pbuf; + fp->mp_pad = data->mv_size; // used if MDB_DUPFIXED + fp->mp_lower = fp->mp_upper = olddata.mv_size = PAGEHDRSZ; + goto prep_subDB; } + } else { - fp_flags = fp->mp_flags; - if (NODESIZE + NODEKSZ(leaf) + xdata.mv_size > env->me_nodemax) { - // Too big for a sub-page, convert to sub-DB - fp_flags &= ~P_SUBP; -prep_subDB: - dummy.md_pad = 0; - dummy.md_flags = 0; - dummy.md_depth = 1; - dummy.md_branch_pages = 0; - dummy.md_leaf_pages = 1; - dummy.md_overflow_pages = 0; - dummy.md_entries = NUMKEYS(fp); - xdata.mv_size = sizeof(MDB_db); - xdata.mv_data = &dummy; - if ((rc = mdb_page_alloc(mc, 1, &mp))) - return rc; - offset = env->me_psize - olddata.mv_size; - flags |= F_DUPDATA|F_SUBDATA; - dummy.md_root = mp->mp_pgno; - } - if (mp != fp) { - mp->mp_flags = fp_flags | P_DIRTY; - mp->mp_pad = fp->mp_pad; - mp->mp_lower = fp->mp_lower; - mp->mp_upper = fp->mp_upper + offset; - if (fp_flags & P_LEAF2) { - memcpy(METADATA(mp), METADATA(fp), NUMKEYS(fp) * fp->mp_pad); - } else { - memcpy((char *)mp + mp->mp_upper, (char *)fp + fp->mp_upper, - olddata.mv_size - fp->mp_upper); - for (i = NUMKEYS(fp); --i >= 0; ) - mp->mp_ptrs[i] = fp->mp_ptrs[i] + offset; - } - } - - rdata = &xdata; - flags |= F_DUPDATA; - do_sub = 1; - if (!insert) - mdb_node_del(mc, 0); - goto new_sub; - } -current: - // overflow page overwrites need special handling - if (F_ISSET(leaf->mn_flags, F_BIGDATA)) { - MDB_page *omp; - pgno_t pg; - int level, ovpages, dpages = OVPAGES(data->mv_size, env->me_psize); - - memcpy(&pg, olddata.mv_data, sizeof(pg)); - if ((rc2 = mdb_page_get(mc->mc_txn, pg, &omp, &level)) != 0) - return rc2; - ovpages = omp->mp_pages; - - // Is the ov page large enough? - if (ovpages >= dpages) { - if (!(omp->mp_flags & P_DIRTY) && - (level || (env->me_flags & MDB_WRITEMAP))) - { - rc = mdb_page_unspill(mc->mc_txn, omp, &omp); - if (rc) - return rc; - level = 0; // dirty in this txn or clean - } - // Is it dirty? - if (omp->mp_flags & P_DIRTY) { - // yes, overwrite it. Note in this case we don't - // bother to try shrinking the page if the new data - // is smaller than the overflow threshold. - if (level > 1) { - // It is writable only in a parent txn - size_t sz = (size_t) env->me_psize * ovpages, off; - MDB_page *np = mdb_page_malloc(mc->mc_txn, ovpages); - MDB_ID2 id2; - if (!np) - return ENOMEM; - id2.mid = pg; - id2.mptr = np; - rc = mdb_mid2l_insert(mc->mc_txn->mt_u.dirty_list, &id2); - mdb_cassert(mc, rc == 0); - if (!(flags & MDB_RESERVE)) { - // Copy end of page, adjusting alignment so - // compiler may copy words instead of bytes. - off = (PAGEHDRSZ + data->mv_size) & -sizeof(size_t); - memcpy((size_t *)((char *)np + off), - (size_t *)((char *)omp + off), sz - off); - sz = PAGEHDRSZ; - } - memcpy(np, omp, sz); // Copy beginning of page - omp = np; - } - SETDSZ(leaf, data->mv_size); - if (F_ISSET(flags, MDB_RESERVE)) - data->mv_data = METADATA(omp); - else - memcpy(METADATA(omp), data->mv_data, data->mv_size); - goto done; - } - } - if ((rc2 = mdb_ovpage_free(mc, omp)) != MDB_SUCCESS) - return rc2; - } else if (data->mv_size == olddata.mv_size) { - // same size, just replace it. Note that we could - // also reuse this node if the new data is smaller, - // but instead we opt to shrink the node in that case. - if (F_ISSET(flags, MDB_RESERVE)) - data->mv_data = olddata.mv_data; - else if (data->mv_size) - memcpy(olddata.mv_data, data->mv_data, data->mv_size); - else - memcpy(NODEKEY(leaf), key->mv_data, key->mv_size); - goto done; - } - mdb_node_del(mc, 0); - mc->mc_db->md_entries--; - } - - rdata = data; - -new_sub: - nflags = flags & NODE_ADD_FLAGS; - nsize = IS_LEAF2(mc->mc_pg[mc->mc_top]) ? key->mv_size : mdb_leaf_size(env, key, rdata); - if (SIZELEFT(mc->mc_pg[mc->mc_top]) < nsize) { - if (( flags & (F_DUPDATA|F_SUBDATA)) == F_DUPDATA ) - nflags &= ~MDB_APPEND; - if (!insert) - nflags |= MDB_SPLIT_REPLACE; - rc = mdb_page_split(mc, key, rdata, P_INVALID, nflags); - } else { - // There is room already in this leaf page. - rc = mdb_node_add(mc, mc->mc_ki[mc->mc_top], key, rdata, 0, nflags); - if (rc == 0 && !do_sub && insert) { - // Adjust other cursors pointing to mp - MDB_cursor *m2, *m3; - MDB_dbi dbi = mc->mc_dbi; - unsigned i = mc->mc_top; - MDB_page *mp = mc->mc_pg[i]; - - for (m2 = mc->mc_txn->mt_cursors[dbi]; m2; m2=m2->mc_next) { - if (mc->mc_flags & C_SUB) - m3 = &m2->mc_xcursor->mx_cursor; - else - m3 = m2; - if (m3 == mc || m3->mc_snum < mc->mc_snum) continue; - if (m3->mc_pg[i] == mp && m3->mc_ki[i] >= mc->mc_ki[i]) { - m3->mc_ki[i]++; - } - } - } - } - - if (rc != MDB_SUCCESS) - mc->mc_txn->mt_flags |= MDB_TXN_ERROR; - else { - // 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) { - int xflags; -put_sub: - xdata.mv_size = 0; - xdata.mv_data = ""; + more: leaf = NODEPTR(mc->mc_pg[mc->mc_top], mc->mc_ki[mc->mc_top]); - if (flags & MDB_CURRENT) { - xflags = MDB_CURRENT|MDB_NOSPILL; - } else { - mdb_xcursor_init1(mc, leaf); - xflags = (flags & MDB_NODUPDATA) ? - MDB_NOOVERWRITE|MDB_NOSPILL : MDB_NOSPILL; - } - // converted, write the original data first - if (dkey.mv_size) { - rc = mdb_cursor_put(&mc->mc_xcursor->mx_cursor, &dkey, &xdata, xflags); - if (rc) - return rc; - { - // Adjust other cursors pointing to mp - MDB_cursor *m2; - unsigned i = mc->mc_top; - MDB_page *mp = mc->mc_pg[i]; + olddata.mv_size = NODEDSZ(leaf); + olddata.mv_data = NODEDATA(leaf); - for (m2 = mc->mc_txn->mt_cursors[mc->mc_dbi]; m2; m2=m2->mc_next) { - if (m2 == mc || m2->mc_snum < mc->mc_snum) continue; - if (!(m2->mc_flags & C_INITIALIZED)) continue; - if (m2->mc_pg[i] == mp && m2->mc_ki[i] == mc->mc_ki[i]) { - mdb_xcursor_init1(m2, leaf); + // DB has dups? + if (F_ISSET(mc->mc_db->md_flags, MDB_DUPSORT)) { + // Prepare (sub-)page/sub-DB to accept the new item, + // if needed. fp: old sub-page or a header faking + // it. mp: new (sub-)page. offset: growth in page + // size. xdata: node data with new page or DB. + ssize_t i, offset = 0; + mp = fp = xdata.mv_data = env->me_pbuf; + mp->mp_pgno = mc->mc_pg[mc->mc_top]->mp_pgno; + + // Was a single item before, must convert now + if (!F_ISSET(leaf->mn_flags, F_DUPDATA)) { + // Just overwrite the current item + if (flags == MDB_CURRENT) + goto current; + + #if UINT_MAX < SIZE_MAX + if (mc->mc_dbx->md_dcmp == mdb_cmp_int && olddata.mv_size == sizeof(size_t)) + #ifdef MISALIGNED_OK + mc->mc_dbx->md_dcmp = mdb_cmp_long; + #else + mc->mc_dbx->md_dcmp = mdb_cmp_cint; + #endif + #endif + // if data matches, skip it + if (!mc->mc_dbx->md_dcmp(data, &olddata)) { + if (flags & MDB_NODUPDATA) + rc = MDB_KEYEXIST; + else if (flags & MDB_MULTIPLE) + goto next_mult; + else + rc = MDB_SUCCESS; + return rc; + } + + // Back up original data item + dkey.mv_size = olddata.mv_size; + dkey.mv_data = memcpy(fp+1, olddata.mv_data, olddata.mv_size); + + // Make sub-page header for the dup items, with dummy body + fp->mp_flags = P_LEAF|P_DIRTY|P_SUBP; + fp->mp_lower = PAGEHDRSZ; + xdata.mv_size = PAGEHDRSZ + dkey.mv_size + data->mv_size; + if (mc->mc_db->md_flags & MDB_DUPFIXED) { + fp->mp_flags |= P_LEAF2; + fp->mp_pad = data->mv_size; + xdata.mv_size += 2 * data->mv_size; // leave space for 2 more + } else { + xdata.mv_size += 2 * (sizeof(indx_t) + NODESIZE) + + (dkey.mv_size & 1) + (data->mv_size & 1); + } + fp->mp_upper = xdata.mv_size; + olddata.mv_size = fp->mp_upper; // pretend olddata is fp + } else if (leaf->mn_flags & F_SUBDATA) { + // Data is on sub-DB, just store it + flags |= F_DUPDATA|F_SUBDATA; + goto put_sub; + } else { + // Data is on sub-page + fp = olddata.mv_data; + switch (flags) { + default: + i = -(ssize_t)SIZELEFT(fp); + if (!(mc->mc_db->md_flags & MDB_DUPFIXED)) { + offset = i += (ssize_t) EVEN( + sizeof(indx_t) + NODESIZE + data->mv_size); + } else { + i += offset = fp->mp_pad; + offset *= 4; // space for 4 more + } + if (i > 0) + break; + // FALLTHRU: Sub-page is big enough + case MDB_CURRENT: + fp->mp_flags |= P_DIRTY; + COPY_PGNO(fp->mp_pgno, mp->mp_pgno); + mc->mc_xcursor->mx_cursor.mc_pg[0] = fp; + flags |= F_DUPDATA; + goto put_sub; + } + xdata.mv_size = olddata.mv_size + offset; + } + + fp_flags = fp->mp_flags; + if (NODESIZE + NODEKSZ(leaf) + xdata.mv_size > env->me_nodemax) { + // Too big for a sub-page, convert to sub-DB + fp_flags &= ~P_SUBP; + prep_subDB: + dummy.md_pad = 0; + dummy.md_flags = 0; + dummy.md_depth = 1; + dummy.md_branch_pages = 0; + dummy.md_leaf_pages = 1; + dummy.md_overflow_pages = 0; + dummy.md_entries = NUMKEYS(fp); + xdata.mv_size = sizeof(MDB_db); + xdata.mv_data = &dummy; + if ((rc = mdb_page_alloc(mc, 1, &mp))) + return rc; + offset = env->me_psize - olddata.mv_size; + flags |= F_DUPDATA|F_SUBDATA; + dummy.md_root = mp->mp_pgno; + } + if (mp != fp) { + mp->mp_flags = fp_flags | P_DIRTY; + mp->mp_pad = fp->mp_pad; + mp->mp_lower = fp->mp_lower; + mp->mp_upper = fp->mp_upper + offset; + if (fp_flags & P_LEAF2) { + memcpy(METADATA(mp), METADATA(fp), NUMKEYS(fp) * fp->mp_pad); + } else { + memcpy((char *)mp + mp->mp_upper, (char *)fp + fp->mp_upper, + olddata.mv_size - fp->mp_upper); + for (i = NUMKEYS(fp); --i >= 0; ) + mp->mp_ptrs[i] = fp->mp_ptrs[i] + offset; + } + } + + rdata = &xdata; + flags |= F_DUPDATA; + do_sub = 1; + if (!insert) + mdb_node_del(mc, 0); + goto new_sub; + } + current: + // overflow page overwrites need special handling + if (F_ISSET(leaf->mn_flags, F_BIGDATA)) { + MDB_page *omp; + pgno_t pg; + int level, ovpages, dpages = OVPAGES(data->mv_size, env->me_psize); + + memcpy(&pg, olddata.mv_data, sizeof(pg)); + if ((rc2 = mdb_page_get(mc->mc_txn, pg, &omp, &level)) != 0) + return rc2; + ovpages = omp->mp_pages; + + // Is the ov page large enough? + if (ovpages >= dpages) { + if (!(omp->mp_flags & P_DIRTY) && + (level || (env->me_flags & MDB_WRITEMAP))) + { + rc = mdb_page_unspill(mc->mc_txn, omp, &omp); + if (rc) + return rc; + level = 0; // dirty in this txn or clean + } + // Is it dirty? + if (omp->mp_flags & P_DIRTY) { + // yes, overwrite it. Note in this case we don't + // bother to try shrinking the page if the new data + // is smaller than the overflow threshold. + if (level > 1) { + // It is writable only in a parent txn + size_t sz = (size_t) env->me_psize * ovpages, off; + MDB_page *np = mdb_page_malloc(mc->mc_txn, ovpages); + MDB_ID2 id2; + if (!np) + return ENOMEM; + id2.mid = pg; + id2.mptr = np; + rc = mdb_mid2l_insert(mc->mc_txn->mt_u.dirty_list, &id2); + mdb_cassert(mc, rc == 0); + if (!(flags & MDB_RESERVE)) { + // Copy end of page, adjusting alignment so + // compiler may copy words instead of bytes. + off = (PAGEHDRSZ + data->mv_size) & -sizeof(size_t); + memcpy((size_t *)((char *)np + off), + (size_t *)((char *)omp + off), sz - off); + sz = PAGEHDRSZ; + } + memcpy(np, omp, sz); // Copy beginning of page + omp = np; + } + SETDSZ(leaf, data->mv_size); + if (F_ISSET(flags, MDB_RESERVE)) + data->mv_data = METADATA(omp); + else + memcpy(METADATA(omp), data->mv_data, data->mv_size); + goto done; + } + } + if ((rc2 = mdb_ovpage_free(mc, omp)) != MDB_SUCCESS) + return rc2; + } else if (data->mv_size == olddata.mv_size) { + // same size, just replace it. Note that we could + // also reuse this node if the new data is smaller, + // but instead we opt to shrink the node in that case. + if (F_ISSET(flags, MDB_RESERVE)) + data->mv_data = olddata.mv_data; + else if (data->mv_size) + memcpy(olddata.mv_data, data->mv_data, data->mv_size); + else + memcpy(NODEKEY(leaf), key->mv_data, key->mv_size); + goto done; + } + mdb_node_del(mc, 0); + mc->mc_db->md_entries--; + } + + rdata = data; + + new_sub: + nflags = flags & NODE_ADD_FLAGS; + nsize = IS_LEAF2(mc->mc_pg[mc->mc_top]) ? key->mv_size : mdb_leaf_size(env, key, rdata); + if (SIZELEFT(mc->mc_pg[mc->mc_top]) < nsize) { + if (( flags & (F_DUPDATA|F_SUBDATA)) == F_DUPDATA ) + nflags &= ~MDB_APPEND; + if (!insert) + nflags |= MDB_SPLIT_REPLACE; + rc = mdb_page_split(mc, key, rdata, P_INVALID, nflags); + } else { + // There is room already in this leaf page. + rc = mdb_node_add(mc, mc->mc_ki[mc->mc_top], key, rdata, 0, nflags); + if (rc == 0 && !do_sub && insert) { + // Adjust other cursors pointing to mp + MDB_cursor *m2, *m3; + MDB_dbi dbi = mc->mc_dbi; + unsigned i = mc->mc_top; + MDB_page *mp = mc->mc_pg[i]; + + for (m2 = mc->mc_txn->mt_cursors[dbi]; m2; m2=m2->mc_next) { + if (mc->mc_flags & C_SUB) + m3 = &m2->mc_xcursor->mx_cursor; + else + m3 = m2; + if (m3 == mc || m3->mc_snum < mc->mc_snum) continue; + if (m3->mc_pg[i] == mp && m3->mc_ki[i] >= mc->mc_ki[i]) { + m3->mc_ki[i]++; + } + } + } + } + + if (rc != MDB_SUCCESS) + mc->mc_txn->mt_flags |= MDB_TXN_ERROR; + else { + // 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) { + int xflags; + put_sub: + xdata.mv_size = 0; + xdata.mv_data = ""; + leaf = NODEPTR(mc->mc_pg[mc->mc_top], mc->mc_ki[mc->mc_top]); + if (flags & MDB_CURRENT) { + xflags = MDB_CURRENT|MDB_NOSPILL; + } else { + mdb_xcursor_init1(mc, leaf); + xflags = (flags & MDB_NODUPDATA) ? + MDB_NOOVERWRITE|MDB_NOSPILL : MDB_NOSPILL; + } + // converted, write the original data first + if (dkey.mv_size) { + rc = mdb_cursor_put(&mc->mc_xcursor->mx_cursor, &dkey, &xdata, xflags); + if (rc) + return rc; + { + // Adjust other cursors pointing to mp + MDB_cursor *m2; + unsigned i = mc->mc_top; + MDB_page *mp = mc->mc_pg[i]; + + for (m2 = mc->mc_txn->mt_cursors[mc->mc_dbi]; m2; m2=m2->mc_next) { + if (m2 == mc || m2->mc_snum < mc->mc_snum) continue; + if (!(m2->mc_flags & C_INITIALIZED)) continue; + if (m2->mc_pg[i] == mp && m2->mc_ki[i] == mc->mc_ki[i]) { + mdb_xcursor_init1(m2, leaf); + } } } + // we've done our job + dkey.mv_size = 0; + } + if (flags & MDB_APPENDDUP) + xflags |= MDB_APPEND; + rc = mdb_cursor_put(&mc->mc_xcursor->mx_cursor, data, &xdata, xflags); + if (flags & F_SUBDATA) { + void *db = NODEDATA(leaf); + memcpy(db, &mc->mc_xcursor->mx_db, sizeof(MDB_db)); } - // we've done our job - dkey.mv_size = 0; } - if (flags & MDB_APPENDDUP) - xflags |= MDB_APPEND; - rc = mdb_cursor_put(&mc->mc_xcursor->mx_cursor, data, &xdata, xflags); - if (flags & F_SUBDATA) { - void *db = NODEDATA(leaf); - memcpy(db, &mc->mc_xcursor->mx_db, sizeof(MDB_db)); - } - } - // sub-writes might have failed so check rc again. - // Don't increment count if we just replaced an existing item. - if (!rc && !(flags & MDB_CURRENT)) - mc->mc_db->md_entries++; - if (flags & MDB_MULTIPLE) { - if (!rc) { -next_mult: - mcount++; - // let caller know how many succeeded, if any - data[1].mv_size = mcount; - if (mcount < dcount) { - data[0].mv_data = (char *)data[0].mv_data + data[0].mv_size; - goto more; + // sub-writes might have failed so check rc again. + // Don't increment count if we just replaced an existing item. + if (!rc && !(flags & MDB_CURRENT)) + mc->mc_db->md_entries++; + if (flags & MDB_MULTIPLE) { + if (!rc) { + next_mult: + mcount++; + // let caller know how many succeeded, if any + data[1].mv_size = mcount; + if (mcount < dcount) { + data[0].mv_data = (char *)data[0].mv_data + data[0].mv_size; + goto more; + } } } } - } -done: - // If we succeeded and the key didn't exist before, make sure - // the cursor is marked valid. - if (!rc && insert) - mc->mc_flags |= C_INITIALIZED; - return rc; + done: + // If we succeeded and the key didn't exist before, make sure + // the cursor is marked valid. + if (!rc && insert) + mc->mc_flags |= C_INITIALIZED; + return rc; */ return nil } @@ -476,126 +483,126 @@ func (c *RWCursor) allocatePage(count int) (*page, error) { // } /* - int rc, retry = INT_MAX; - MDB_txn *txn = mc->mc_txn; - MDB_env *env = txn->mt_env; - pgno_t pgno, *mop = env->me_pghead; - unsigned i, j, k, mop_len = mop ? mop[0] : 0, n2 = num-1; - MDB_page *np; - txnid_t oldest = 0, last; - MDB_cursor_op op; - MDB_cursor m2; + int rc, retry = INT_MAX; + MDB_txn *txn = mc->mc_txn; + MDB_env *env = txn->mt_env; + pgno_t pgno, *mop = env->me_pghead; + unsigned i, j, k, mop_len = mop ? mop[0] : 0, n2 = num-1; + MDB_page *np; + txnid_t oldest = 0, last; + MDB_cursor_op op; + MDB_cursor m2; - *mp = NULL; + *mp = NULL; - for (op = MDB_FIRST;; op = MDB_NEXT) { - MDB_val key, data; - MDB_node *leaf; - pgno_t *idl, old_id, new_id; + for (op = MDB_FIRST;; op = MDB_NEXT) { + MDB_val key, data; + MDB_node *leaf; + pgno_t *idl, old_id, new_id; - // Seek a big enough contiguous page range. Prefer - // pages at the tail, just truncating the list. - if (mop_len > n2) { - i = mop_len; - do { - pgno = mop[i]; - if (mop[i-n2] == pgno+n2) - goto search_done; - } while (--i > n2); - if (Max_retries < INT_MAX && --retry < 0) - break; - } - - if (op == MDB_FIRST) { // 1st iteration - // Prepare to fetch more and coalesce - oldest = mdb_find_oldest(txn); - last = env->me_pglast; - mdb_cursor_init(&m2, txn, FREE_DBI, NULL); - if (last) { - op = MDB_SET_RANGE; - key.mv_data = &last; // will look up last+1 - key.mv_size = sizeof(last); + // Seek a big enough contiguous page range. Prefer + // pages at the tail, just truncating the list. + if (mop_len > n2) { + i = mop_len; + do { + pgno = mop[i]; + if (mop[i-n2] == pgno+n2) + goto search_done; + } while (--i > n2); + if (Max_retries < INT_MAX && --retry < 0) + break; } - } - last++; - // Do not fetch more if the record will be too recent - if (oldest <= last) - break; - rc = mdb_cursor_get(&m2, &key, NULL, op); - if (rc) { - if (rc == MDB_NOTFOUND) + if (op == MDB_FIRST) { // 1st iteration + // Prepare to fetch more and coalesce + oldest = mdb_find_oldest(txn); + last = env->me_pglast; + mdb_cursor_init(&m2, txn, FREE_DBI, NULL); + if (last) { + op = MDB_SET_RANGE; + key.mv_data = &last; // will look up last+1 + key.mv_size = sizeof(last); + } + } + + last++; + // Do not fetch more if the record will be too recent + if (oldest <= last) break; - goto fail; - } - last = *(txnid_t*)key.mv_data; - if (oldest <= last) - break; - np = m2.mc_pg[m2.mc_top]; - leaf = NODEPTR(np, m2.mc_ki[m2.mc_top]); - if ((rc = mdb_node_read(txn, leaf, &data)) != MDB_SUCCESS) - return rc; - - idl = (MDB_ID *) data.mv_data; - i = idl[0]; - if (!mop) { - if (!(env->me_pghead = mop = mdb_midl_alloc(i))) { - rc = ENOMEM; + rc = mdb_cursor_get(&m2, &key, NULL, op); + if (rc) { + if (rc == MDB_NOTFOUND) + break; goto fail; } + last = *(txnid_t*)key.mv_data; + if (oldest <= last) + break; + np = m2.mc_pg[m2.mc_top]; + leaf = NODEPTR(np, m2.mc_ki[m2.mc_top]); + if ((rc = mdb_node_read(txn, leaf, &data)) != MDB_SUCCESS) + return rc; + + idl = (MDB_ID *) data.mv_data; + i = idl[0]; + if (!mop) { + if (!(env->me_pghead = mop = mdb_midl_alloc(i))) { + rc = ENOMEM; + goto fail; + } + } else { + if ((rc = mdb_midl_need(&env->me_pghead, i)) != 0) + goto fail; + mop = env->me_pghead; + } + env->me_pglast = last; + + // Merge in descending sorted order + j = mop_len; + k = mop_len += i; + mop[0] = (pgno_t)-1; + old_id = mop[j]; + while (i) { + new_id = idl[i--]; + for (; old_id < new_id; old_id = mop[--j]) + mop[k--] = old_id; + mop[k--] = new_id; + } + mop[0] = mop_len; + } + + // Use new pages from the map when nothing suitable in the freeDB + i = 0; + pgno = txn->mt_next_pgno; + if (pgno + num >= env->me_maxpg) { + DPUTS("DB size maxed out"); + rc = MDB_MAP_FULL; + goto fail; + } + + search_done: + if (!(np = mdb_page_malloc(txn, num))) { + rc = ENOMEM; + goto fail; + } + if (i) { + mop[0] = mop_len -= num; + // Move any stragglers down + for (j = i-num; j < mop_len; ) + mop[++j] = mop[++i]; } else { - if ((rc = mdb_midl_need(&env->me_pghead, i)) != 0) - goto fail; - mop = env->me_pghead; + txn->mt_next_pgno = pgno + num; } - env->me_pglast = last; + np->mp_pgno = pgno; + mdb_page_dirty(txn, np); + *mp = np; - // Merge in descending sorted order - j = mop_len; - k = mop_len += i; - mop[0] = (pgno_t)-1; - old_id = mop[j]; - while (i) { - new_id = idl[i--]; - for (; old_id < new_id; old_id = mop[--j]) - mop[k--] = old_id; - mop[k--] = new_id; - } - mop[0] = mop_len; - } + return MDB_SUCCESS; - // Use new pages from the map when nothing suitable in the freeDB - i = 0; - pgno = txn->mt_next_pgno; - if (pgno + num >= env->me_maxpg) { - DPUTS("DB size maxed out"); - rc = MDB_MAP_FULL; - goto fail; - } - - search_done: - if (!(np = mdb_page_malloc(txn, num))) { - rc = ENOMEM; - goto fail; - } - if (i) { - mop[0] = mop_len -= num; - // Move any stragglers down - for (j = i-num; j < mop_len; ) - mop[++j] = mop[++i]; - } else { - txn->mt_next_pgno = pgno + num; - } - np->mp_pgno = pgno; - mdb_page_dirty(txn, np); - *mp = np; - - return MDB_SUCCESS; - - fail: - txn->mt_flags |= MDB_TXN_ERROR; - return rc; + fail: + txn->mt_flags |= MDB_TXN_ERROR; + return rc; */ return nil } diff --git a/rwtransaction.go b/rwtransaction.go index 4cb7b4f..d0fe440 100644 --- a/rwtransaction.go +++ b/rwtransaction.go @@ -4,6 +4,4 @@ package bolt // Only one read/write transaction can be active for a DB at a time. type RWTransaction struct { Transaction - pagestate pagestate } -