X-Git-Url: https://git.sur5r.net/?a=blobdiff_plain;f=libraries%2Fliblmdb%2Fmdb.c;h=7d34276566a27158c7b5632db72c36f83f104c98;hb=f24d7d2c83670848194f89c8455538394d2e09f5;hp=def326a672a93ace2cfa2dc2f2da170e76e6bd5f;hpb=2a839d5dfab19b23fe90c1680a5e29bf4a6e2f09;p=openldap diff --git a/libraries/liblmdb/mdb.c b/libraries/liblmdb/mdb.c index def326a672..7d34276566 100644 --- a/libraries/liblmdb/mdb.c +++ b/libraries/liblmdb/mdb.c @@ -5,7 +5,7 @@ * BerkeleyDB API, but much simplified. */ /* - * Copyright 2011-2015 Howard Chu, Symas Corp. + * Copyright 2011-2016 Howard Chu, Symas Corp. * All rights reserved. * * Redistribution and use in source and binary forms, with or without @@ -35,6 +35,9 @@ #ifndef _GNU_SOURCE #define _GNU_SOURCE 1 #endif +#if defined(__WIN64__) +#define _FILE_OFFSET_BITS 64 +#endif #ifdef _WIN32 #include #include @@ -198,7 +201,7 @@ typedef SSIZE_T ssize_t; #define ESECT #endif -#ifdef _MSC_VER +#ifdef _WIN32 #define CALL_CONV WINAPI #else #define CALL_CONV @@ -244,10 +247,9 @@ typedef SSIZE_T ssize_t; /** Some platforms define the EOWNERDEAD error code * even though they don't support Robust Mutexes. * Compile with -DMDB_USE_ROBUST=0, or use some other - * mechanism like -DMDB_USE_SYSV_SEM instead of - * -DMDB_USE_POSIX_MUTEX. (SysV semaphores are - * also Robust, but some systems don't support them - * either.) + * mechanism like -DMDB_USE_POSIX_SEM instead of + * -DMDB_USE_POSIX_MUTEX. + * (Posix semaphores are not robust.) */ #ifndef MDB_USE_ROBUST /* Android currently lacks Robust Mutex support. So does glibc < 2.4. */ @@ -256,8 +258,9 @@ typedef SSIZE_T ssize_t; # define MDB_USE_ROBUST 0 # else # define MDB_USE_ROBUST 1 -/* glibc < 2.10 only provided _np API */ -# if defined(__GLIBC__) && GLIBC_VER < 0x02000a +/* glibc < 2.12 only provided _np API */ +# if (defined(__GLIBC__) && GLIBC_VER < 0x02000c) || \ + (defined(PTHREAD_MUTEX_ROBUST_NP) && !defined(PTHREAD_MUTEX_ROBUST)) # define PTHREAD_MUTEX_ROBUST PTHREAD_MUTEX_ROBUST_NP # define pthread_mutexattr_setrobust(attr, flag) pthread_mutexattr_setrobust_np(attr, flag) # define pthread_mutex_consistent(mutex) pthread_mutex_consistent_np(mutex) @@ -1075,11 +1078,12 @@ struct MDB_txn { * @ingroup internal * @{ */ -#define DB_DIRTY 0x01 /**< DB was modified or is DUPSORT data */ +#define DB_DIRTY 0x01 /**< DB was written in this txn */ #define DB_STALE 0x02 /**< Named-DB record is older than txnID */ #define DB_NEW 0x04 /**< Named-DB handle opened in this txn */ #define DB_VALID 0x08 /**< DB handle is valid, see also #MDB_VALID */ #define DB_USRVALID 0x10 /**< As #DB_VALID, but not set for #FREE_DBI */ +#define DB_DUPDATA 0x20 /**< DB is #MDB_DUPSORT data */ /** @} */ /** In write txns, array of cursors for each DB */ MDB_cursor **mt_cursors; @@ -1161,7 +1165,6 @@ struct MDB_cursor { #define C_EOF 0x02 /**< No more data */ #define C_SUB 0x04 /**< Cursor is a sub-cursor */ #define C_DEL 0x08 /**< last op was a cursor_del */ -#define C_SPLITTING 0x20 /**< Cursor is in page_split */ #define C_UNTRACK 0x40 /**< Un-track cursor when closing */ /** @} */ unsigned int mc_flags; /**< @ref mdb_cursor */ @@ -1185,6 +1188,21 @@ typedef struct MDB_xcursor { unsigned char mx_dbflag; } MDB_xcursor; + /** Check if there is an inited xcursor, so #XCURSOR_REFRESH() is proper */ +#define XCURSOR_INITED(mc) \ + ((mc)->mc_xcursor && ((mc)->mc_xcursor->mx_cursor.mc_flags & C_INITIALIZED)) + + /** Update sub-page pointer, if any, in \b mc->mc_xcursor. Needed + * when the node which contains the sub-page may have moved. Called + * with \b mp = mc->mc_pg[mc->mc_top], \b ki = mc->mc_ki[mc->mc_top]. + */ +#define XCURSOR_REFRESH(mc, mp, ki) do { \ + MDB_page *xr_pg = (mp); \ + MDB_node *xr_node = NODEPTR(xr_pg, ki); \ + if ((xr_node->mn_flags & (F_DUPDATA|F_SUBDATA)) == F_DUPDATA) \ + (mc)->mc_xcursor->mx_cursor.mc_pg[0] = NODEDATA(xr_node); \ +} while (0) + /** State of FreeDB old pages, stored in the MDB_env */ typedef struct MDB_pgstate { pgno_t *mf_pghead; /**< Reclaimed freeDB pages, or NULL before use */ @@ -1272,7 +1290,7 @@ typedef struct MDB_ntxn { #endif /** max bytes to write in one call */ -#define MAX_WRITE (0x80000000U >> (sizeof(ssize_t) == 4)) +#define MAX_WRITE (0x40000000U >> (sizeof(ssize_t) == 4)) /** Check \b txn and \b dbi arguments to a function */ #define TXN_DBI_EXIST(txn, dbi, validity) \ @@ -1299,7 +1317,7 @@ enum { #define MDB_END_SLOT MDB_NOTLS /**< release any reader slot if #MDB_NOTLS */ static void mdb_txn_end(MDB_txn *txn, unsigned mode); -static int mdb_page_get(MDB_txn *txn, pgno_t pgno, MDB_page **mp, int *lvl); +static int mdb_page_get(MDB_cursor *mc, pgno_t pgno, MDB_page **mp, int *lvl); static int mdb_page_search_root(MDB_cursor *mc, MDB_val *key, int modify); #define MDB_PS_MODIFY 1 @@ -1327,8 +1345,8 @@ static int mdb_node_add(MDB_cursor *mc, indx_t indx, MDB_val *key, MDB_val *data, pgno_t pgno, unsigned int flags); static void mdb_node_del(MDB_cursor *mc, int ksize); static void mdb_node_shrink(MDB_page *mp, indx_t indx); -static int mdb_node_move(MDB_cursor *csrc, MDB_cursor *cdst); -static int mdb_node_read(MDB_txn *txn, MDB_node *leaf, MDB_val *data); +static int mdb_node_move(MDB_cursor *csrc, MDB_cursor *cdst, int fromleft); +static int mdb_node_read(MDB_cursor *mc, MDB_node *leaf, MDB_val *data); static size_t mdb_leaf_size(MDB_env *env, MDB_val *key, MDB_val *data); static size_t mdb_branch_size(MDB_env *env, MDB_val *key); @@ -1372,6 +1390,8 @@ static MDB_cmp_func mdb_cmp_memn, mdb_cmp_memnr, mdb_cmp_int, mdb_cmp_cint, mdb_ static SECURITY_DESCRIPTOR mdb_null_sd; static SECURITY_ATTRIBUTES mdb_all_sa; static int mdb_sec_inited; + +static int utf8_to_utf16(const char *src, int srcsize, wchar_t **dst, int *dstsize); #endif /** Return the library version info. */ @@ -1416,8 +1436,9 @@ mdb_strerror(int err) * This works as long as no function between the call to mdb_strerror * and the actual use of the message uses more than 4K of stack. */ - char pad[4096]; - char buf[1024], *ptr = buf; +#define MSGSIZE 1024 +#define PADSIZE 4096 + char buf[MSGSIZE+PADSIZE], *ptr = buf; #endif int i; if (!err) @@ -1449,7 +1470,7 @@ mdb_strerror(int err) buf[0] = 0; FormatMessageA(FORMAT_MESSAGE_FROM_SYSTEM | FORMAT_MESSAGE_IGNORE_INSERTS, - NULL, err, 0, ptr, sizeof(buf), (va_list *)pad); + NULL, err, 0, ptr, MSGSIZE, (va_list *)buf+MSGSIZE); return ptr; #else return strerror(err); @@ -1603,7 +1624,7 @@ mdb_cursor_chk(MDB_cursor *mc) MDB_node *node; MDB_page *mp; - if (!mc->mc_snum && !(mc->mc_flags & C_INITIALIZED)) return; + if (!mc->mc_snum || !(mc->mc_flags & C_INITIALIZED)) return; for (i=0; imc_top; i++) { mp = mc->mc_pg[i]; node = NODEPTR(mp, mc->mc_ki[i]); @@ -1612,6 +1633,13 @@ mdb_cursor_chk(MDB_cursor *mc) } if (mc->mc_ki[i] >= NUMKEYS(mc->mc_pg[i])) printf("ack!\n"); + if (XCURSOR_INITED(mc)) { + node = NODEPTR(mc->mc_pg[mc->mc_top], mc->mc_ki[mc->mc_top]); + if (((node->mn_flags & (F_DUPDATA|F_SUBDATA)) == F_DUPDATA) && + mc->mc_xcursor->mx_cursor.mc_pg[0] != NODEDATA(node)) { + printf("blah!\n"); + } + } } #endif @@ -1834,7 +1862,7 @@ mdb_pages_xkeep(MDB_cursor *mc, unsigned pflags, int all) { enum { Mask = P_SUBP|P_DIRTY|P_LOOSE|P_KEEP }; MDB_txn *txn = mc->mc_txn; - MDB_cursor *m3; + MDB_cursor *m3, *m0 = mc; MDB_xcursor *mx; MDB_page *dp, *mp; MDB_node *leaf; @@ -1877,7 +1905,7 @@ mdb_pages_xkeep(MDB_cursor *mc, unsigned pflags, int all) pgno_t pgno = txn->mt_dbs[i].md_root; if (pgno == P_INVALID) continue; - if ((rc = mdb_page_get(txn, pgno, &dp, &level)) != MDB_SUCCESS) + if ((rc = mdb_page_get(m0, pgno, &dp, &level)) != MDB_SUCCESS) break; if ((dp->mp_flags & Mask) == pflags && level <= 1) dp->mp_flags ^= P_KEEP; @@ -2181,7 +2209,7 @@ mdb_page_alloc(MDB_cursor *mc, int num, MDB_page **mp) } 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) + if ((rc = mdb_node_read(&m2, leaf, &data)) != MDB_SUCCESS) return rc; idl = (MDB_ID *) data.mv_data; @@ -2417,16 +2445,11 @@ done: } else { for (; m2; m2=m2->mc_next) { if (m2->mc_snum < mc->mc_snum) continue; + if (m2 == mc) continue; if (m2->mc_pg[mc->mc_top] == mp) { m2->mc_pg[mc->mc_top] = np; - if ((mc->mc_db->md_flags & MDB_DUPSORT) && - IS_LEAF(np) && - m2->mc_ki[mc->mc_top] == mc->mc_ki[mc->mc_top]) - { - MDB_node *leaf = NODEPTR(np, mc->mc_ki[mc->mc_top]); - if (!(leaf->mn_flags & F_SUBDATA)) - m2->mc_xcursor->mx_cursor.mc_pg[0] = NODEDATA(leaf); - } + if (XCURSOR_INITED(m2) && IS_LEAF(np)) + XCURSOR_REFRESH(m2, np, m2->mc_ki[mc->mc_top]); } } } @@ -2488,14 +2511,15 @@ mdb_cursor_shadow(MDB_txn *src, MDB_txn *dst) *bk = *mc; mc->mc_backup = bk; mc->mc_db = &dst->mt_dbs[i]; - /* Kill pointers into src - and dst to reduce abuse: The - * user may not use mc until dst ends. Otherwise we'd... + /* Kill pointers into src to reduce abuse: The + * user may not use mc until dst ends. But we need a valid + * txn pointer here for cursor fixups to keep working. */ - mc->mc_txn = NULL; /* ...set this to dst */ - mc->mc_dbflag = NULL; /* ...and &dst->mt_dbflags[i] */ + mc->mc_txn = dst; + mc->mc_dbflag = &dst->mt_dbflags[i]; if ((mx = mc->mc_xcursor) != NULL) { *(MDB_xcursor *)(bk+1) = *mx; - mx->mx_cursor.mc_txn = NULL; /* ...and dst. */ + mx->mx_cursor.mc_txn = dst; } mc->mc_next = dst->mt_cursors[i]; dst->mt_cursors[i] = mc; @@ -3429,6 +3453,25 @@ mdb_txn_commit(MDB_txn *txn) pspill[0] = y; } + /* Remove anything in our spill list from parent's dirty list */ + if (txn->mt_spill_pgs && txn->mt_spill_pgs[0]) { + for (i=1; i<=txn->mt_spill_pgs[0]; i++) { + MDB_ID pn = txn->mt_spill_pgs[i]; + if (pn & 1) + continue; /* deleted spillpg */ + pn >>= 1; + y = mdb_mid2l_search(dst, pn); + if (y <= dst[0].mid && dst[y].mid == pn) { + free(dst[y].mptr); + while (y < dst[0].mid) { + dst[y] = dst[y+1]; + y++; + } + dst[0].mid--; + } + } + } + /* Find len = length of merging our dirty list with parent's */ x = dst[0].mid; dst[0].mid = 0; /* simplify loops */ @@ -3474,7 +3517,7 @@ mdb_txn_commit(MDB_txn *txn) } /* Append our loose page list to parent's */ - for (lp = &parent->mt_loose_pgs; *lp; lp = &NEXT_LOOSE_PAGE(lp)) + for (lp = &parent->mt_loose_pgs; *lp; lp = &NEXT_LOOSE_PAGE(*lp)) ; *lp = txn->mt_loose_pgs; parent->mt_loose_count += txn->mt_loose_count; @@ -4440,9 +4483,14 @@ mdb_env_setup_locks(MDB_env *env, char *lpath, int mode, int *excl) off_t size, rsize; #ifdef _WIN32 - env->me_lfd = CreateFileA(lpath, GENERIC_READ|GENERIC_WRITE, + wchar_t *wlpath; + rc = utf8_to_utf16(lpath, -1, &wlpath, NULL); + if (rc) + return rc; + env->me_lfd = CreateFileW(wlpath, GENERIC_READ|GENERIC_WRITE, FILE_SHARE_READ|FILE_SHARE_WRITE, NULL, OPEN_ALWAYS, FILE_ATTRIBUTE_NORMAL, NULL); + free(wlpath); #else env->me_lfd = open(lpath, O_RDWR|O_CREAT|MDB_CLOEXEC, mode); #endif @@ -4585,15 +4633,25 @@ mdb_env_setup_locks(MDB_env *env, char *lpath, int mode, int *excl) #else /* MDB_USE_POSIX_MUTEX: */ pthread_mutexattr_t mattr; - if ((rc = pthread_mutexattr_init(&mattr)) - || (rc = pthread_mutexattr_setpshared(&mattr, PTHREAD_PROCESS_SHARED)) + /* Solaris needs this before initing a robust mutex. Otherwise + * it may skip the init and return EBUSY "seems someone already + * inited" or EINVAL "it was inited differently". + */ + memset(env->me_txns->mti_rmutex, 0, sizeof(*env->me_txns->mti_rmutex)); + memset(env->me_txns->mti_wmutex, 0, sizeof(*env->me_txns->mti_wmutex)); + + if ((rc = pthread_mutexattr_init(&mattr))) + goto fail; + + rc = pthread_mutexattr_setpshared(&mattr, PTHREAD_PROCESS_SHARED); #ifdef MDB_ROBUST_SUPPORTED - || (rc = pthread_mutexattr_setrobust(&mattr, PTHREAD_MUTEX_ROBUST)) + if (!rc) rc = pthread_mutexattr_setrobust(&mattr, PTHREAD_MUTEX_ROBUST); #endif - || (rc = pthread_mutex_init(env->me_txns->mti_rmutex, &mattr)) - || (rc = pthread_mutex_init(env->me_txns->mti_wmutex, &mattr))) - goto fail; + if (!rc) rc = pthread_mutex_init(env->me_txns->mti_rmutex, &mattr); + if (!rc) rc = pthread_mutex_init(env->me_txns->mti_wmutex, &mattr); pthread_mutexattr_destroy(&mattr); + if (rc) + goto fail; #endif /* _WIN32 || MDB_USE_POSIX_SEM */ env->me_txns->mti_magic = MDB_MAGIC; @@ -4660,6 +4718,9 @@ mdb_env_open(MDB_env *env, const char *path, unsigned int flags, mdb_mode_t mode { int oflags, rc, len, excl = -1; char *lpath, *dpath; +#ifdef _WIN32 + wchar_t *wpath; +#endif if (env->me_fd!=INVALID_HANDLE_VALUE || (flags & ~(CHANGEABLE|CHANGELESS))) return EINVAL; @@ -4723,8 +4784,12 @@ mdb_env_open(MDB_env *env, const char *path, unsigned int flags, mdb_mode_t mode len = OPEN_ALWAYS; } mode = FILE_ATTRIBUTE_NORMAL; - env->me_fd = CreateFileA(dpath, oflags, FILE_SHARE_READ|FILE_SHARE_WRITE, + rc = utf8_to_utf16(dpath, -1, &wpath, NULL); + if (rc) + goto leave; + env->me_fd = CreateFileW(wpath, oflags, FILE_SHARE_READ|FILE_SHARE_WRITE, NULL, len, mode, NULL); + free(wpath); #else if (F_ISSET(flags, MDB_RDONLY)) oflags = O_RDONLY; @@ -4753,9 +4818,13 @@ mdb_env_open(MDB_env *env, const char *path, unsigned int flags, mdb_mode_t mode */ #ifdef _WIN32 len = OPEN_EXISTING; - env->me_mfd = CreateFileA(dpath, oflags, + rc = utf8_to_utf16(dpath, -1, &wpath, NULL); + if (rc) + goto leave; + env->me_mfd = CreateFileW(wpath, oflags, FILE_SHARE_READ|FILE_SHARE_WRITE, NULL, len, mode | FILE_FLAG_WRITE_THROUGH, NULL); + free(wpath); #else oflags &= ~O_CREAT; env->me_mfd = open(dpath, oflags | MDB_DSYNC, mode); @@ -5132,8 +5201,11 @@ mdb_cursor_pop(MDB_cursor *mc) mc->mc_pg[mc->mc_top]->mp_pgno, DDBI(mc), (void *) mc)); mc->mc_snum--; - if (mc->mc_snum) + if (mc->mc_snum) { mc->mc_top--; + } else { + mc->mc_flags &= ~C_INITIALIZED; + } } } @@ -5157,15 +5229,16 @@ mdb_cursor_push(MDB_cursor *mc, MDB_page *mp) } /** Find the address of the page corresponding to a given page number. - * @param[in] txn the transaction for this access. + * @param[in] mc the cursor accessing the page. * @param[in] pgno the page number for the page to retrieve. * @param[out] ret address of a pointer where the page's address will be stored. * @param[out] lvl dirty_list inheritance level of found page. 1=current txn, 0=mapped page. * @return 0 on success, non-zero on failure. */ static int -mdb_page_get(MDB_txn *txn, pgno_t pgno, MDB_page **ret, int *lvl) +mdb_page_get(MDB_cursor *mc, pgno_t pgno, MDB_page **ret, int *lvl) { + MDB_txn *txn = mc->mc_txn; MDB_env *env = txn->mt_env; MDB_page *p = NULL; int level; @@ -5231,7 +5304,11 @@ mdb_page_search_root(MDB_cursor *mc, MDB_val *key, int flags) indx_t i; DPRINTF(("branch page %"Z"u has %u keys", mp->mp_pgno, NUMKEYS(mp))); - mdb_cassert(mc, NUMKEYS(mp) > 1); + /* Don't assert on branch pages in the FreeDB. We can get here + * while in the process of rebalancing a FreeDB branch page; we must + * let that proceed. ITS#8336 + */ + mdb_cassert(mc, !mc->mc_dbi || NUMKEYS(mp) > 1); DPRINTF(("found index 0 to page %"Z"u", NODEPGNO(NODEPTR(mp, 0)))); if (flags & (MDB_PS_FIRST|MDB_PS_LAST)) { @@ -5256,7 +5333,7 @@ mdb_page_search_root(MDB_cursor *mc, MDB_val *key, int flags) mdb_cassert(mc, i < NUMKEYS(mp)); node = NODEPTR(mp, i); - if ((rc = mdb_page_get(mc->mc_txn, NODEPGNO(node), &mp, NULL)) != 0) + if ((rc = mdb_page_get(mc, NODEPGNO(node), &mp, NULL)) != 0) return rc; mc->mc_ki[mc->mc_top] = i; @@ -5298,7 +5375,7 @@ mdb_page_search_lowest(MDB_cursor *mc) MDB_node *node = NODEPTR(mp, 0); int rc; - if ((rc = mdb_page_get(mc->mc_txn, NODEPGNO(node), &mp, NULL)) != 0) + if ((rc = mdb_page_get(mc, NODEPGNO(node), &mp, NULL)) != 0) return rc; mc->mc_ki[mc->mc_top] = 0; @@ -5350,7 +5427,7 @@ mdb_page_search(MDB_cursor *mc, MDB_val *key, int flags) return MDB_NOTFOUND; if ((leaf->mn_flags & (F_DUPDATA|F_SUBDATA)) != F_SUBDATA) return MDB_INCOMPATIBLE; /* not a named DB */ - rc = mdb_node_read(mc->mc_txn, leaf, &data); + rc = mdb_node_read(&mc2, leaf, &data); if (rc) return rc; memcpy(&flags, ((char *) data.mv_data + offsetof(MDB_db, md_flags)), @@ -5374,7 +5451,7 @@ mdb_page_search(MDB_cursor *mc, MDB_val *key, int flags) mdb_cassert(mc, root > 1); if (!mc->mc_pg[0] || mc->mc_pg[0]->mp_pgno != root) - if ((rc = mdb_page_get(mc->mc_txn, root, &mc->mc_pg[0], NULL)) != 0) + if ((rc = mdb_page_get(mc, root, &mc->mc_pg[0], NULL)) != 0) return rc; mc->mc_snum = 1; @@ -5449,6 +5526,7 @@ mdb_ovpage_free(MDB_cursor *mc, MDB_page *mp) return MDB_CORRUPTED; } } + txn->mt_dirty_room++; if (!(env->me_flags & MDB_WRITEMAP)) mdb_dpage_free(env, mp); release: @@ -5470,13 +5548,13 @@ release: } /** Return the data associated with a given node. - * @param[in] txn The transaction for this operation. + * @param[in] mc The cursor for this operation. * @param[in] leaf The node being read. * @param[out] data Updated to point to the node's data. * @return 0 on success, non-zero on failure. */ static int -mdb_node_read(MDB_txn *txn, MDB_node *leaf, MDB_val *data) +mdb_node_read(MDB_cursor *mc, MDB_node *leaf, MDB_val *data) { MDB_page *omp; /* overflow page */ pgno_t pgno; @@ -5492,7 +5570,7 @@ mdb_node_read(MDB_txn *txn, MDB_node *leaf, MDB_val *data) */ data->mv_size = NODEDSZ(leaf); memcpy(&pgno, NODEDATA(leaf), sizeof(pgno)); - if ((rc = mdb_page_get(txn, pgno, &omp, NULL)) != 0) { + if ((rc = mdb_page_get(mc, pgno, &omp, NULL)) != 0) { DPRINTF(("read overflow page %"Z"u failed", pgno)); return rc; } @@ -5566,7 +5644,7 @@ mdb_cursor_sibling(MDB_cursor *mc, int move_right) mdb_cassert(mc, IS_BRANCH(mc->mc_pg[mc->mc_top])); indx = NODEPTR(mc->mc_pg[mc->mc_top], mc->mc_ki[mc->mc_top]); - if ((rc = mdb_page_get(mc->mc_txn, NODEPGNO(indx), &mp, NULL)) != 0) { + if ((rc = mdb_page_get(mc, NODEPGNO(indx), &mp, NULL)) != 0) { /* mc will be inconsistent if caller does mc_snum++ as above */ mc->mc_flags &= ~(C_INITIALIZED|C_EOF); return rc; @@ -5587,11 +5665,12 @@ mdb_cursor_next(MDB_cursor *mc, MDB_val *key, MDB_val *data, MDB_cursor_op op) MDB_node *leaf; int rc; - if (mc->mc_flags & C_EOF) { + if ((mc->mc_flags & C_EOF) || + ((mc->mc_flags & C_DEL) && op == MDB_NEXT_DUP)) { return MDB_NOTFOUND; } - - mdb_cassert(mc, mc->mc_flags & C_INITIALIZED); + if (!(mc->mc_flags & C_INITIALIZED)) + return mdb_cursor_first(mc, key, data); mp = mc->mc_pg[mc->mc_top]; @@ -5615,8 +5694,10 @@ mdb_cursor_next(MDB_cursor *mc, MDB_val *key, MDB_val *data, MDB_cursor_op op) DPRINTF(("cursor_next: top page is %"Z"u in cursor %p", mdb_dbg_pgno(mp), (void *) mc)); - if (mc->mc_flags & C_DEL) + if (mc->mc_flags & C_DEL) { + mc->mc_flags ^= C_DEL; goto skip; + } if (mc->mc_ki[mc->mc_top] + 1u >= NUMKEYS(mp)) { DPUTS("=====> move to next sibling page"); @@ -5646,7 +5727,7 @@ skip: mdb_xcursor_init1(mc, leaf); } if (data) { - if ((rc = mdb_node_read(mc->mc_txn, leaf, data)) != MDB_SUCCESS) + if ((rc = mdb_node_read(mc, leaf, data)) != MDB_SUCCESS) return rc; if (F_ISSET(leaf->mn_flags, F_DUPDATA)) { @@ -5668,7 +5749,12 @@ mdb_cursor_prev(MDB_cursor *mc, MDB_val *key, MDB_val *data, MDB_cursor_op op) MDB_node *leaf; int rc; - mdb_cassert(mc, mc->mc_flags & C_INITIALIZED); + if (!(mc->mc_flags & C_INITIALIZED)) { + rc = mdb_cursor_last(mc, key, data); + if (rc) + return rc; + mc->mc_ki[mc->mc_top]++; + } mp = mc->mc_pg[mc->mc_top]; @@ -5695,6 +5781,8 @@ mdb_cursor_prev(MDB_cursor *mc, MDB_val *key, MDB_val *data, MDB_cursor_op op) DPRINTF(("cursor_prev: top page is %"Z"u in cursor %p", mdb_dbg_pgno(mp), (void *) mc)); + mc->mc_flags &= ~(C_EOF|C_DEL); + if (mc->mc_ki[mc->mc_top] == 0) { DPUTS("=====> move to prev sibling page"); if ((rc = mdb_cursor_sibling(mc, 0)) != MDB_SUCCESS) { @@ -5706,8 +5794,6 @@ mdb_cursor_prev(MDB_cursor *mc, MDB_val *key, MDB_val *data, MDB_cursor_op op) } else mc->mc_ki[mc->mc_top]--; - mc->mc_flags &= ~C_EOF; - DPRINTF(("==> cursor points to page %"Z"u with %u keys, key index %u", mdb_dbg_pgno(mp), NUMKEYS(mp), mc->mc_ki[mc->mc_top])); @@ -5724,7 +5810,7 @@ mdb_cursor_prev(MDB_cursor *mc, MDB_val *key, MDB_val *data, MDB_cursor_op op) mdb_xcursor_init1(mc, leaf); } if (data) { - if ((rc = mdb_node_read(mc->mc_txn, leaf, data)) != MDB_SUCCESS) + if ((rc = mdb_node_read(mc, leaf, data)) != MDB_SUCCESS) return rc; if (F_ISSET(leaf->mn_flags, F_DUPDATA)) { @@ -5843,6 +5929,8 @@ mdb_cursor_set(MDB_cursor *mc, MDB_val *key, MDB_val *data, } else return MDB_NOTFOUND; } + } else { + mc->mc_pg[0] = 0; } rc = mdb_page_search(mc, key, 0); @@ -5904,7 +5992,7 @@ set1: } else if (op == MDB_GET_BOTH || op == MDB_GET_BOTH_RANGE) { MDB_val olddata; MDB_cmp_func *dcmp; - if ((rc = mdb_node_read(mc->mc_txn, leaf, &olddata)) != MDB_SUCCESS) + if ((rc = mdb_node_read(mc, leaf, &olddata)) != MDB_SUCCESS) return rc; dcmp = mc->mc_dbx->md_dcmp; #if UINT_MAX < SIZE_MAX @@ -5916,13 +6004,13 @@ set1: if (op == MDB_GET_BOTH || rc > 0) return MDB_NOTFOUND; rc = 0; - *data = olddata; } + *data = olddata; } else { if (mc->mc_xcursor) mc->mc_xcursor->mx_cursor.mc_flags &= ~(C_INITIALIZED|C_EOF); - if ((rc = mdb_node_read(mc->mc_txn, leaf, data)) != MDB_SUCCESS) + if ((rc = mdb_node_read(mc, leaf, data)) != MDB_SUCCESS) return rc; } } @@ -5971,7 +6059,7 @@ mdb_cursor_first(MDB_cursor *mc, MDB_val *key, MDB_val *data) if (rc) return rc; } else { - if ((rc = mdb_node_read(mc->mc_txn, leaf, data)) != MDB_SUCCESS) + if ((rc = mdb_node_read(mc, leaf, data)) != MDB_SUCCESS) return rc; } } @@ -6016,7 +6104,7 @@ mdb_cursor_last(MDB_cursor *mc, MDB_val *key, MDB_val *data) if (rc) return rc; } else { - if ((rc = mdb_node_read(mc->mc_txn, leaf, data)) != MDB_SUCCESS) + if ((rc = mdb_node_read(mc, leaf, data)) != MDB_SUCCESS) return rc; } } @@ -6060,11 +6148,9 @@ mdb_cursor_get(MDB_cursor *mc, MDB_val *key, MDB_val *data, MDB_GET_KEY(leaf, key); if (data) { if (F_ISSET(leaf->mn_flags, F_DUPDATA)) { - if (mc->mc_flags & C_DEL) - mdb_xcursor_init1(mc, leaf); rc = mdb_cursor_get(&mc->mc_xcursor->mx_cursor, data, NULL, MDB_GET_CURRENT); } else { - rc = mdb_node_read(mc->mc_txn, leaf, data); + rc = mdb_node_read(mc, leaf, data); } } } @@ -6114,10 +6200,7 @@ mdb_cursor_get(MDB_cursor *mc, MDB_val *key, MDB_val *data, rc = MDB_INCOMPATIBLE; break; } - if (!(mc->mc_flags & C_INITIALIZED)) - rc = mdb_cursor_first(mc, key, data); - else - rc = mdb_cursor_next(mc, key, data, MDB_NEXT_DUP); + rc = mdb_cursor_next(mc, key, data, MDB_NEXT_DUP); if (rc == MDB_SUCCESS) { if (mc->mc_xcursor->mx_cursor.mc_flags & C_INITIALIZED) { MDB_cursor *mx; @@ -6132,24 +6215,38 @@ fetchm: } } break; + case MDB_PREV_MULTIPLE: + if (data == NULL) { + rc = EINVAL; + break; + } + if (!(mc->mc_db->md_flags & MDB_DUPFIXED)) { + rc = MDB_INCOMPATIBLE; + break; + } + if (!(mc->mc_flags & C_INITIALIZED)) + rc = mdb_cursor_last(mc, key, data); + else + rc = MDB_SUCCESS; + if (rc == MDB_SUCCESS) { + MDB_cursor *mx = &mc->mc_xcursor->mx_cursor; + if (mx->mc_flags & C_INITIALIZED) { + rc = mdb_cursor_sibling(mx, 0); + if (rc == MDB_SUCCESS) + goto fetchm; + } else { + rc = MDB_NOTFOUND; + } + } + break; case MDB_NEXT: case MDB_NEXT_DUP: case MDB_NEXT_NODUP: - if (!(mc->mc_flags & C_INITIALIZED)) - rc = mdb_cursor_first(mc, key, data); - else - rc = mdb_cursor_next(mc, key, data, op); + rc = mdb_cursor_next(mc, key, data, op); break; case MDB_PREV: case MDB_PREV_DUP: case MDB_PREV_NODUP: - if (!(mc->mc_flags & C_INITIALIZED)) { - rc = mdb_cursor_last(mc, key, data); - if (rc) - break; - mc->mc_flags |= C_INITIALIZED; - mc->mc_ki[mc->mc_top]++; - } rc = mdb_cursor_prev(mc, key, data, op); break; case MDB_FIRST: @@ -6170,7 +6267,7 @@ fetchm: MDB_node *leaf = NODEPTR(mc->mc_pg[mc->mc_top], mc->mc_ki[mc->mc_top]); if (!F_ISSET(leaf->mn_flags, F_DUPDATA)) { MDB_GET_KEY(leaf, key); - rc = mdb_node_read(mc->mc_txn, leaf, data); + rc = mdb_node_read(mc, leaf, data); break; } } @@ -6207,7 +6304,8 @@ mdb_cursor_touch(MDB_cursor *mc) { int rc = MDB_SUCCESS; - if (mc->mc_dbi >= CORE_DBS && !(*mc->mc_dbflag & DB_DIRTY)) { + if (mc->mc_dbi >= CORE_DBS && !(*mc->mc_dbflag & (DB_DIRTY|DB_DUPDATA))) { + /* Touch DB record of named DB */ MDB_cursor mc2; MDB_xcursor mcx; if (TXN_DBI_CHANGED(mc->mc_txn, mc->mc_dbi)) @@ -6391,16 +6489,18 @@ fix_parent: * update branch key if there is a parent page */ if (mc->mc_top && !mc->mc_ki[mc->mc_top]) { - unsigned short top = mc->mc_top; + unsigned short dtop = 1; mc->mc_top--; /* slot 0 is always an empty key, find real slot */ - while (mc->mc_top && !mc->mc_ki[mc->mc_top]) + while (mc->mc_top && !mc->mc_ki[mc->mc_top]) { mc->mc_top--; + dtop++; + } if (mc->mc_ki[mc->mc_top]) rc2 = mdb_update_key(mc, key); else rc2 = MDB_SUCCESS; - mc->mc_top = top; + mc->mc_top += dtop; if (rc2) return rc2; } @@ -6436,7 +6536,7 @@ more: #endif /* does data match? */ if (!dcmp(data, &olddata)) { - if (flags & MDB_NODUPDATA) + if (flags & (MDB_NODUPDATA|MDB_APPENDDUP)) return MDB_KEYEXIST; /* overwrite it */ goto current; @@ -6552,7 +6652,7 @@ current: 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) + if ((rc2 = mdb_page_get(mc, pg, &omp, &level)) != 0) return rc2; ovpages = omp->mp_pages; @@ -6581,10 +6681,16 @@ current: return ENOMEM; id2.mid = pg; id2.mptr = np; + /* Note - this page is already counted in parent's dirty_room */ rc2 = mdb_mid2l_insert(mc->mc_txn->mt_u.dirty_list, &id2); mdb_cassert(mc, rc2 == 0); + /* Currently we make the page look as with put() in the + * parent txn, in case the user peeks at MDB_RESERVEd + * or unused parts. Some users treat ovpages specially. + */ if (!(flags & MDB_RESERVE)) { - /* Copy end of page, adjusting alignment so + /* Skip the part where LMDB will put *data. + * Copy end of page, adjusting alignment so * compiler may copy words instead of bytes. */ off = (PAGEHDRSZ + data->mv_size) & -sizeof(size_t); @@ -6637,7 +6743,7 @@ new_sub: } 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 && insert_key) { + if (rc == 0) { /* Adjust other cursors pointing to mp */ MDB_cursor *m2, *m3; MDB_dbi dbi = mc->mc_dbi; @@ -6649,10 +6755,12 @@ new_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]) { + if (m3 == mc || m3->mc_snum < mc->mc_snum || m3->mc_pg[i] != mp) continue; + if (m3->mc_ki[i] >= mc->mc_ki[i] && insert_key) { m3->mc_ki[i]++; } + if (XCURSOR_INITED(m3)) + XCURSOR_REFRESH(m3, mp, m3->mc_ki[i]); } } } @@ -6694,6 +6802,7 @@ put_sub: MDB_xcursor *mx = mc->mc_xcursor; unsigned i = mc->mc_top; MDB_page *mp = mc->mc_pg[i]; + int nkeys = NUMKEYS(mp); 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; @@ -6701,10 +6810,8 @@ put_sub: if (m2->mc_pg[i] == mp) { if (m2->mc_ki[i] == mc->mc_ki[i]) { mdb_xcursor_init2(m2, mx, new_dupdata); - } else if (!insert_key) { - MDB_node *n2 = NODEPTR(mp, m2->mc_ki[i]); - if (!(n2->mn_flags & F_SUBDATA)) - m2->mc_xcursor->mx_cursor.mc_pg[0] = NODEDATA(n2); + } else if (!insert_key && m2->mc_ki[i] < nkeys) { + XCURSOR_REFRESH(m2, mp, m2->mc_ki[i]); } } } @@ -6784,6 +6891,7 @@ mdb_cursor_del(MDB_cursor *mc, unsigned int flags) if (flags & MDB_NODUPDATA) { /* mdb_cursor_del0() will subtract the final entry */ mc->mc_db->md_entries -= mc->mc_xcursor->mx_db.md_entries - 1; + mc->mc_xcursor->mx_cursor.mc_flags &= ~C_INITIALIZED; } else { if (!F_ISSET(leaf->mn_flags, F_SUBDATA)) { mc->mc_xcursor->mx_cursor.mc_pg[0] = NODEDATA(leaf); @@ -6808,19 +6916,19 @@ mdb_cursor_del(MDB_cursor *mc, unsigned int flags) if (m2 == mc || m2->mc_snum < mc->mc_snum) continue; if (!(m2->mc_flags & C_INITIALIZED)) continue; if (m2->mc_pg[mc->mc_top] == mp) { - if (m2->mc_ki[mc->mc_top] == mc->mc_ki[mc->mc_top]) { - m2->mc_xcursor->mx_cursor.mc_pg[0] = NODEDATA(leaf); - } else { - MDB_node *n2 = NODEPTR(mp, m2->mc_ki[mc->mc_top]); - if (!(n2->mn_flags & F_SUBDATA)) - m2->mc_xcursor->mx_cursor.mc_pg[0] = NODEDATA(n2); + MDB_node *n2 = leaf; + if (m2->mc_ki[mc->mc_top] != mc->mc_ki[mc->mc_top]) { + n2 = NODEPTR(mp, m2->mc_ki[mc->mc_top]); + if (n2->mn_flags & F_SUBDATA) continue; } + m2->mc_xcursor->mx_cursor.mc_pg[0] = NODEDATA(n2); } } } mc->mc_db->md_entries--; - mc->mc_flags |= C_DEL; return rc; + } else { + mc->mc_xcursor->mx_cursor.mc_flags &= ~C_INITIALIZED; } /* otherwise fall thru and delete the sub-DB */ } @@ -6844,7 +6952,7 @@ mdb_cursor_del(MDB_cursor *mc, unsigned int flags) pgno_t pg; memcpy(&pg, NODEDATA(leaf), sizeof(pg)); - if ((rc = mdb_page_get(mc->mc_txn, pg, &omp, NULL)) || + if ((rc = mdb_page_get(mc, pg, &omp, NULL)) || (rc = mdb_ovpage_free(mc, omp))) goto fail; } @@ -7252,7 +7360,7 @@ mdb_xcursor_init1(MDB_cursor *mc, MDB_node *node) } DPRINTF(("Sub-db -%u root page %"Z"u", mx->mx_cursor.mc_dbi, mx->mx_db.md_root)); - mx->mx_dbflag = DB_VALID|DB_USRVALID|DB_DIRTY; /* DB_DIRTY guides mdb_cursor_touch */ + mx->mx_dbflag = DB_VALID|DB_USRVALID|DB_DUPDATA; #if UINT_MAX < SIZE_MAX if (mx->mx_dbx.md_cmp == mdb_cmp_int && mx->mx_db.md_pad == sizeof(size_t)) mx->mx_dbx.md_cmp = mdb_cmp_clong; @@ -7278,7 +7386,7 @@ mdb_xcursor_init2(MDB_cursor *mc, MDB_xcursor *src_mx, int new_dupdata) mx->mx_cursor.mc_top = 0; mx->mx_cursor.mc_flags |= C_INITIALIZED; mx->mx_cursor.mc_ki[0] = 0; - mx->mx_dbflag = DB_VALID|DB_USRVALID|DB_DIRTY; /* DB_DIRTY guides mdb_cursor_touch */ + mx->mx_dbflag = DB_VALID|DB_USRVALID|DB_DUPDATA; #if UINT_MAX < SIZE_MAX mx->mx_dbx.md_cmp = src_mx->mx_dbx.md_cmp; #endif @@ -7507,10 +7615,26 @@ mdb_update_key(MDB_cursor *mc, MDB_val *key) static void mdb_cursor_copy(const MDB_cursor *csrc, MDB_cursor *cdst); +/** Perform \b act while tracking temporary cursor \b mn */ +#define WITH_CURSOR_TRACKING(mn, act) do { \ + MDB_cursor dummy, *tracked, **tp = &(mn).mc_txn->mt_cursors[mn.mc_dbi]; \ + if ((mn).mc_flags & C_SUB) { \ + dummy.mc_flags = C_INITIALIZED; \ + dummy.mc_xcursor = (MDB_xcursor *)&(mn); \ + tracked = &dummy; \ + } else { \ + tracked = &(mn); \ + } \ + tracked->mc_next = *tp; \ + *tp = tracked; \ + { act; } \ + *tp = tracked->mc_next; \ +} while (0) + /** Move a node from csrc to cdst. */ static int -mdb_node_move(MDB_cursor *csrc, MDB_cursor *cdst) +mdb_node_move(MDB_cursor *csrc, MDB_cursor *cdst, int fromleft) { MDB_node *srcnode; MDB_val key, data; @@ -7562,6 +7686,7 @@ mdb_node_move(MDB_cursor *csrc, MDB_cursor *cdst) data.mv_size = NODEDSZ(srcnode); data.mv_data = NODEDATA(srcnode); } + mn.mc_xcursor = NULL; if (IS_BRANCH(cdst->mc_pg[cdst->mc_top]) && cdst->mc_ki[cdst->mc_top] == 0) { unsigned int snum = cdst->mc_snum; MDB_node *s2; @@ -7612,13 +7737,15 @@ mdb_node_move(MDB_cursor *csrc, MDB_cursor *cdst) mps = csrc->mc_pg[csrc->mc_top]; /* If we're adding on the left, bump others up */ - if (!cdst->mc_ki[csrc->mc_top]) { + if (fromleft) { mpd = cdst->mc_pg[csrc->mc_top]; for (m2 = csrc->mc_txn->mt_cursors[dbi]; m2; m2=m2->mc_next) { if (csrc->mc_flags & C_SUB) m3 = &m2->mc_xcursor->mx_cursor; else m3 = m2; + if (!(m3->mc_flags & C_INITIALIZED) || m3->mc_top < csrc->mc_top) + continue; if (m3 != cdst && m3->mc_pg[csrc->mc_top] == mpd && m3->mc_ki[csrc->mc_top] >= cdst->mc_ki[csrc->mc_top]) { @@ -7631,6 +7758,8 @@ mdb_node_move(MDB_cursor *csrc, MDB_cursor *cdst) m3->mc_ki[csrc->mc_top] = cdst->mc_ki[cdst->mc_top]; m3->mc_ki[csrc->mc_top-1]++; } + if (XCURSOR_INITED(m3) && IS_LEAF(mps)) + XCURSOR_REFRESH(m3, m3->mc_pg[csrc->mc_top], m3->mc_ki[csrc->mc_top]); } } else /* Adding on the right, bump others down */ @@ -7641,6 +7770,8 @@ mdb_node_move(MDB_cursor *csrc, MDB_cursor *cdst) else m3 = m2; if (m3 == csrc) continue; + if (!(m3->mc_flags & C_INITIALIZED) || m3->mc_top < csrc->mc_top) + continue; if (m3->mc_pg[csrc->mc_top] == mps) { if (!m3->mc_ki[csrc->mc_top]) { m3->mc_pg[csrc->mc_top] = cdst->mc_pg[cdst->mc_top]; @@ -7649,6 +7780,8 @@ mdb_node_move(MDB_cursor *csrc, MDB_cursor *cdst) } else { m3->mc_ki[csrc->mc_top]--; } + if (XCURSOR_INITED(m3) && IS_LEAF(mps)) + XCURSOR_REFRESH(m3, m3->mc_pg[csrc->mc_top], m3->mc_ki[csrc->mc_top]); } } } @@ -7670,7 +7803,10 @@ mdb_node_move(MDB_cursor *csrc, MDB_cursor *cdst) mdb_cursor_copy(csrc, &mn); mn.mc_snum--; mn.mc_top--; - if ((rc = mdb_update_key(&mn, &key)) != MDB_SUCCESS) + /* We want mdb_rebalance to find mn when doing fixups */ + WITH_CURSOR_TRACKING(mn, + rc = mdb_update_key(&mn, &key)); + if (rc) return rc; } if (IS_BRANCH(csrc->mc_pg[csrc->mc_top])) { @@ -7698,7 +7834,10 @@ mdb_node_move(MDB_cursor *csrc, MDB_cursor *cdst) mdb_cursor_copy(cdst, &mn); mn.mc_snum--; mn.mc_top--; - if ((rc = mdb_update_key(&mn, &key)) != MDB_SUCCESS) + /* We want mdb_rebalance to find mn when doing fixups */ + WITH_CURSOR_TRACKING(mn, + rc = mdb_update_key(&mn, &key)); + if (rc) return rc; } if (IS_BRANCH(cdst->mc_pg[cdst->mc_top])) { @@ -7767,6 +7906,7 @@ mdb_page_merge(MDB_cursor *csrc, MDB_cursor *cdst) MDB_cursor mn; MDB_node *s2; mdb_cursor_copy(csrc, &mn); + mn.mc_xcursor = NULL; /* must find the lowest key below src */ rc = mdb_page_search_lowest(&mn); if (rc) @@ -7838,7 +7978,12 @@ mdb_page_merge(MDB_cursor *csrc, MDB_cursor *cdst) m3->mc_pg[top] = pdst; m3->mc_ki[top] += nkeys; m3->mc_ki[top-1] = cdst->mc_ki[top-1]; + } else if (m3->mc_pg[top-1] == csrc->mc_pg[top-1] && + m3->mc_ki[top-1] > csrc->mc_ki[top-1]) { + m3->mc_ki[top-1]--; } + if (XCURSOR_INITED(m3) && IS_LEAF(psrc)) + XCURSOR_REFRESH(m3, m3->mc_pg[top], m3->mc_ki[top]); } } { @@ -7887,7 +8032,7 @@ static int mdb_rebalance(MDB_cursor *mc) { MDB_node *node; - int rc; + int rc, fromleft; unsigned int ptop, minkeys, thresh; MDB_cursor mn; indx_t oldki; @@ -7938,7 +8083,8 @@ mdb_rebalance(MDB_cursor *mc) m3 = &m2->mc_xcursor->mx_cursor; else m3 = m2; - if (m3->mc_snum < mc->mc_snum) continue; + if (!(m3->mc_flags & C_INITIALIZED) || (m3->mc_snum < mc->mc_snum)) + continue; if (m3->mc_pg[0] == mp) { m3->mc_snum = 0; m3->mc_top = 0; @@ -7953,7 +8099,7 @@ mdb_rebalance(MDB_cursor *mc) if (rc) return rc; mc->mc_db->md_root = NODEPGNO(NODEPTR(mp, 0)); - rc = mdb_page_get(mc->mc_txn,mc->mc_db->md_root,&mc->mc_pg[0],NULL); + rc = mdb_page_get(mc, mc->mc_db->md_root, &mc->mc_pg[0], NULL); if (rc) return rc; mc->mc_db->md_depth--; @@ -7974,6 +8120,8 @@ mdb_rebalance(MDB_cursor *mc) else m3 = m2; if (m3 == mc) continue; + if (!(m3->mc_flags & C_INITIALIZED)) + continue; if (m3->mc_pg[0] == mp) { for (i=0; imc_db->md_depth; i++) { m3->mc_pg[i] = m3->mc_pg[i+1]; @@ -8012,22 +8160,24 @@ mdb_rebalance(MDB_cursor *mc) DPUTS("reading right neighbor"); mn.mc_ki[ptop]++; node = NODEPTR(mc->mc_pg[ptop], mn.mc_ki[ptop]); - rc = mdb_page_get(mc->mc_txn,NODEPGNO(node),&mn.mc_pg[mn.mc_top],NULL); + rc = mdb_page_get(mc, NODEPGNO(node), &mn.mc_pg[mn.mc_top], NULL); if (rc) return rc; mn.mc_ki[mn.mc_top] = 0; mc->mc_ki[mc->mc_top] = NUMKEYS(mc->mc_pg[mc->mc_top]); + fromleft = 0; } else { /* There is at least one neighbor to the left. */ DPUTS("reading left neighbor"); mn.mc_ki[ptop]--; node = NODEPTR(mc->mc_pg[ptop], mn.mc_ki[ptop]); - rc = mdb_page_get(mc->mc_txn,NODEPGNO(node),&mn.mc_pg[mn.mc_top],NULL); + rc = mdb_page_get(mc, NODEPGNO(node), &mn.mc_pg[mn.mc_top], NULL); if (rc) return rc; mn.mc_ki[mn.mc_top] = NUMKEYS(mn.mc_pg[mn.mc_top]) - 1; mc->mc_ki[mc->mc_top] = 0; + fromleft = 1; } DPRINTF(("found neighbor page %"Z"u (%u keys, %.1f%% full)", @@ -8039,32 +8189,20 @@ mdb_rebalance(MDB_cursor *mc) * (A branch page must never have less than 2 keys.) */ if (PAGEFILL(mc->mc_txn->mt_env, mn.mc_pg[mn.mc_top]) >= thresh && NUMKEYS(mn.mc_pg[mn.mc_top]) > minkeys) { - rc = mdb_node_move(&mn, mc); - if (!mc->mc_ki[mc->mc_top]) { + rc = mdb_node_move(&mn, mc, fromleft); + if (fromleft) { /* if we inserted on left, bump position up */ oldki++; } } else { - if (mc->mc_ki[ptop] == 0) { + if (!fromleft) { rc = mdb_page_merge(&mn, mc); } else { - MDB_cursor dummy; oldki += NUMKEYS(mn.mc_pg[mn.mc_top]); mn.mc_ki[mn.mc_top] += mc->mc_ki[mn.mc_top] + 1; /* We want mdb_rebalance to find mn when doing fixups */ - if (mc->mc_flags & C_SUB) { - dummy.mc_next = mc->mc_txn->mt_cursors[mc->mc_dbi]; - mc->mc_txn->mt_cursors[mc->mc_dbi] = &dummy; - dummy.mc_xcursor = (MDB_xcursor *)&mn; - } else { - mn.mc_next = mc->mc_txn->mt_cursors[mc->mc_dbi]; - mc->mc_txn->mt_cursors[mc->mc_dbi] = &mn; - } - rc = mdb_page_merge(mc, &mn); - if (mc->mc_flags & C_SUB) - mc->mc_txn->mt_cursors[mc->mc_dbi] = dummy.mc_next; - else - mc->mc_txn->mt_cursors[mc->mc_dbi] = mn.mc_next; + WITH_CURSOR_TRACKING(mn, + rc = mdb_page_merge(mc, &mn)); mdb_cursor_copy(&mn, mc); } mc->mc_flags &= ~C_EOF; @@ -8097,13 +8235,18 @@ mdb_cursor_del0(MDB_cursor *mc) if (m3 == mc || m3->mc_snum < mc->mc_snum) continue; if (m3->mc_pg[mc->mc_top] == mp) { - if (m3->mc_ki[mc->mc_top] >= ki) { + if (m3->mc_ki[mc->mc_top] == ki) { m3->mc_flags |= C_DEL; - if (m3->mc_ki[mc->mc_top] > ki) - m3->mc_ki[mc->mc_top]--; - else if (mc->mc_db->md_flags & MDB_DUPSORT) - m3->mc_xcursor->mx_cursor.mc_flags |= C_EOF; + if (mc->mc_db->md_flags & MDB_DUPSORT) { + /* Sub-cursor referred into dataset which is gone */ + m3->mc_xcursor->mx_cursor.mc_flags &= ~(C_INITIALIZED|C_EOF); + } + continue; + } else if (m3->mc_ki[mc->mc_top] > ki) { + m3->mc_ki[mc->mc_top]--; } + if (XCURSOR_INITED(m3)) + XCURSOR_REFRESH(m3, m3->mc_pg[mc->mc_top], m3->mc_ki[mc->mc_top]); } } } @@ -8129,11 +8272,27 @@ mdb_cursor_del0(MDB_cursor *mc) continue; if (m3->mc_pg[mc->mc_top] == mp) { /* if m3 points past last node in page, find next sibling */ - if (m3->mc_ki[mc->mc_top] >= nkeys) { - rc = mdb_cursor_sibling(m3, 1); - if (rc == MDB_NOTFOUND) { - m3->mc_flags |= C_EOF; - rc = MDB_SUCCESS; + if (m3->mc_ki[mc->mc_top] >= mc->mc_ki[mc->mc_top]) { + if (m3->mc_ki[mc->mc_top] >= nkeys) { + rc = mdb_cursor_sibling(m3, 1); + if (rc == MDB_NOTFOUND) { + m3->mc_flags |= C_EOF; + rc = MDB_SUCCESS; + continue; + } + } + if (mc->mc_db->md_flags & MDB_DUPSORT) { + MDB_node *node = NODEPTR(m3->mc_pg[m3->mc_top], m3->mc_ki[m3->mc_top]); + /* If this node is a fake page, it needs to be reinited + * because its data has moved. But just reset mc_pg[0] + * if the xcursor is already live. + */ + if ((node->mn_flags & (F_DUPDATA|F_SUBDATA)) == F_DUPDATA) { + if (m3->mc_xcursor->mx_cursor.mc_flags & C_INITIALIZED) + m3->mc_xcursor->mx_cursor.mc_pg[0] = NODEDATA(node); + else + mdb_xcursor_init1(m3, node); + } } } } @@ -8249,12 +8408,19 @@ mdb_page_split(MDB_cursor *mc, MDB_val *newkey, MDB_val *newdata, pgno_t newpgno rp->mp_pad = mp->mp_pad; DPRINTF(("new right sibling: page %"Z"u", rp->mp_pgno)); - if (mc->mc_snum < 2) { + /* Usually when splitting the root page, the cursor + * height is 1. But when called from mdb_update_key, + * the cursor height may be greater because it walks + * up the stack while finding the branch slot to update. + */ + if (mc->mc_top < 1) { if ((rc = mdb_page_new(mc, P_BRANCH, 1, &pp))) goto done; /* shift current top to make room for new parent */ - mc->mc_pg[1] = mc->mc_pg[0]; - mc->mc_ki[1] = mc->mc_ki[0]; + for (i=mc->mc_snum; i>0; i--) { + mc->mc_pg[i] = mc->mc_pg[i-1]; + mc->mc_ki[i] = mc->mc_ki[i-1]; + } mc->mc_pg[0] = pp; mc->mc_ki[0] = 0; mc->mc_db->md_root = pp->mp_pgno; @@ -8270,16 +8436,16 @@ mdb_page_split(MDB_cursor *mc, MDB_val *newkey, MDB_val *newdata, pgno_t newpgno mc->mc_db->md_depth--; goto done; } - mc->mc_snum = 2; - mc->mc_top = 1; + mc->mc_snum++; + mc->mc_top++; ptop = 0; } else { ptop = mc->mc_top-1; DPRINTF(("parent branch page is %"Z"u", mc->mc_pg[ptop]->mp_pgno)); } - mc->mc_flags |= C_SPLITTING; mdb_cursor_copy(mc, &mn); + mn.mc_xcursor = NULL; mn.mc_pg[mn.mc_top] = rp; mn.mc_ki[ptop] = mc->mc_ki[ptop]+1; @@ -8329,8 +8495,6 @@ mdb_page_split(MDB_cursor *mc, MDB_val *newkey, MDB_val *newdata, pgno_t newpgno 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; - mc->mc_ki[ptop]++; } } else { int psize, nsize, k; @@ -8423,21 +8587,18 @@ mdb_page_split(MDB_cursor *mc, MDB_val *newkey, MDB_val *newdata, pgno_t newpgno /* Copy separator key to the parent. */ if (SIZELEFT(mn.mc_pg[ptop]) < mdb_branch_size(env, &sepkey)) { + int snum = mc->mc_snum; mn.mc_snum--; mn.mc_top--; did_split = 1; - rc = mdb_page_split(&mn, &sepkey, NULL, rp->mp_pgno, 0); + /* We want other splits to find mn when doing fixups */ + WITH_CURSOR_TRACKING(mn, + rc = mdb_page_split(&mn, &sepkey, NULL, rp->mp_pgno, 0)); if (rc) goto done; /* root split? */ - if (mn.mc_snum == mc->mc_snum) { - mc->mc_pg[mc->mc_snum] = mc->mc_pg[mc->mc_top]; - mc->mc_ki[mc->mc_snum] = mc->mc_ki[mc->mc_top]; - mc->mc_pg[mc->mc_top] = mc->mc_pg[ptop]; - mc->mc_ki[mc->mc_top] = mc->mc_ki[ptop]; - mc->mc_snum++; - mc->mc_top++; + if (mc->mc_snum > snum) { ptop++; } /* Right page might now have changed parent. @@ -8463,7 +8624,6 @@ mdb_page_split(MDB_cursor *mc, MDB_val *newkey, MDB_val *newdata, pgno_t newpgno rc = mdb_node_add(&mn, mn.mc_ki[ptop], &sepkey, NULL, rp->mp_pgno, 0); mn.mc_top++; } - mc->mc_flags ^= C_SPLITTING; if (rc != MDB_SUCCESS) { goto done; } @@ -8533,11 +8693,6 @@ mdb_page_split(MDB_cursor *mc, MDB_val *newkey, MDB_val *newdata, pgno_t newpgno /* reset back to original page */ if (newindx < split_indx) { mc->mc_pg[mc->mc_top] = mp; - if (nflags & MDB_RESERVE) { - node = NODEPTR(mp, mc->mc_ki[mc->mc_top]); - if (!(node->mn_flags & F_BIGDATA)) - newdata->mv_data = NODEDATA(node); - } } else { mc->mc_pg[mc->mc_top] = rp; mc->mc_ki[ptop]++; @@ -8551,13 +8706,32 @@ mdb_page_split(MDB_cursor *mc, MDB_val *newkey, MDB_val *newdata, pgno_t newpgno } } } + if (nflags & MDB_RESERVE) { + node = NODEPTR(mc->mc_pg[mc->mc_top], mc->mc_ki[mc->mc_top]); + if (!(node->mn_flags & F_BIGDATA)) + newdata->mv_data = NODEDATA(node); + } + } else { + if (newindx >= split_indx) { + mc->mc_pg[mc->mc_top] = rp; + mc->mc_ki[ptop]++; + /* Make sure mc_ki is still valid. + */ + if (mn.mc_pg[ptop] != mc->mc_pg[ptop] && + mc->mc_ki[ptop] >= NUMKEYS(mc->mc_pg[ptop])) { + for (i=0; i<=ptop; i++) { + mc->mc_pg[i] = mn.mc_pg[i]; + mc->mc_ki[i] = mn.mc_ki[i]; + } + } + } } { /* Adjust other cursors pointing to mp */ MDB_cursor *m2, *m3; MDB_dbi dbi = mc->mc_dbi; - int fixup = NUMKEYS(mp); + nkeys = NUMKEYS(mp); for (m2 = mc->mc_txn->mt_cursors[dbi]; m2; m2=m2->mc_next) { if (mc->mc_flags & C_SUB) @@ -8568,16 +8742,17 @@ mdb_page_split(MDB_cursor *mc, MDB_val *newkey, MDB_val *newdata, pgno_t newpgno continue; if (!(m2->mc_flags & m3->mc_flags & C_INITIALIZED)) continue; - if (m3->mc_flags & C_SPLITTING) - continue; if (new_root) { int k; + /* sub cursors may be on different DB */ + if (m3->mc_pg[0] != mp) + continue; /* root split */ for (k=new_root; k>=0; k--) { m3->mc_ki[k+1] = m3->mc_ki[k]; m3->mc_pg[k+1] = m3->mc_pg[k]; } - if (m3->mc_ki[0] >= split_indx) { + if (m3->mc_ki[0] >= nkeys) { m3->mc_ki[0] = 1; } else { m3->mc_ki[0] = 0; @@ -8589,15 +8764,20 @@ mdb_page_split(MDB_cursor *mc, MDB_val *newkey, MDB_val *newdata, pgno_t newpgno if (m3->mc_top >= mc->mc_top && m3->mc_pg[mc->mc_top] == mp) { if (m3->mc_ki[mc->mc_top] >= newindx && !(nflags & MDB_SPLIT_REPLACE)) m3->mc_ki[mc->mc_top]++; - if (m3->mc_ki[mc->mc_top] >= fixup) { + if (m3->mc_ki[mc->mc_top] >= nkeys) { m3->mc_pg[mc->mc_top] = rp; - m3->mc_ki[mc->mc_top] -= fixup; - m3->mc_ki[ptop] = mn.mc_ki[ptop]; + m3->mc_ki[mc->mc_top] -= nkeys; + for (i=0; imc_top; i++) { + m3->mc_ki[i] = mn.mc_ki[i]; + m3->mc_pg[i] = mn.mc_pg[i]; + } } } else if (!did_split && m3->mc_top >= ptop && m3->mc_pg[ptop] == mc->mc_pg[ptop] && m3->mc_ki[ptop] >= mc->mc_ki[ptop]) { m3->mc_ki[ptop]++; } + if (XCURSOR_INITED(m3) && IS_LEAF(mp)) + XCURSOR_REFRESH(m3, m3->mc_pg[mc->mc_top], m3->mc_ki[mc->mc_top]); } } DPRINTF(("mp left: %d, rp left: %d", SIZELEFT(mp), SIZELEFT(rp))); @@ -8616,6 +8796,7 @@ mdb_put(MDB_txn *txn, MDB_dbi dbi, { MDB_cursor mc; MDB_xcursor mx; + int rc; if (!key || !data || !TXN_DBI_EXIST(txn, dbi, DB_USRVALID)) return EINVAL; @@ -8627,7 +8808,11 @@ mdb_put(MDB_txn *txn, MDB_dbi dbi, return (txn->mt_flags & MDB_TXN_RDONLY) ? EACCES : MDB_BAD_TXN; mdb_cursor_init(&mc, txn, dbi, &mx); - return mdb_cursor_put(&mc, key, data, flags); + mc.mc_next = txn->mt_cursors[dbi]; + txn->mt_cursors[dbi] = &mc; + rc = mdb_cursor_put(&mc, key, data, flags); + txn->mt_cursors[dbi] = mc.mc_next; + return rc; } #ifndef MDB_WBUF @@ -8740,8 +8925,7 @@ mdb_env_cthr_toggle(mdb_copy *my, int st) static int ESECT mdb_env_cwalk(mdb_copy *my, pgno_t *pg, int flags) { - MDB_cursor mc; - MDB_txn *txn = my->mc_txn; + MDB_cursor mc = {0}; MDB_node *ni; MDB_page *mo, *mp, *leaf; char *buf, *ptr; @@ -8753,10 +8937,9 @@ mdb_env_cwalk(mdb_copy *my, pgno_t *pg, int flags) return MDB_SUCCESS; mc.mc_snum = 1; - mc.mc_top = 0; - mc.mc_txn = txn; + mc.mc_txn = my->mc_txn; - rc = mdb_page_get(my->mc_txn, *pg, &mc.mc_pg[0], NULL); + rc = mdb_page_get(&mc, *pg, &mc.mc_pg[0], NULL); if (rc) return rc; rc = mdb_page_search_root(&mc, NULL, MDB_PS_FIRST); @@ -8800,7 +8983,7 @@ mdb_env_cwalk(mdb_copy *my, pgno_t *pg, int flags) } memcpy(&pg, NODEDATA(ni), sizeof(pg)); - rc = mdb_page_get(txn, pg, &omp, NULL); + rc = mdb_page_get(&mc, pg, &omp, NULL); if (rc) goto done; if (my->mc_wlen[toggle] >= MDB_WBUF) { @@ -8851,7 +9034,7 @@ mdb_env_cwalk(mdb_copy *my, pgno_t *pg, int flags) again: ni = NODEPTR(mp, mc.mc_ki[mc.mc_top]); pg = NODEPGNO(ni); - rc = mdb_page_get(txn, pg, &mp, NULL); + rc = mdb_page_get(&mc, pg, &mp, NULL); if (rc) goto done; mc.mc_top++; @@ -9016,7 +9199,7 @@ mdb_env_copyfd0(MDB_env *env, HANDLE fd) MDB_txn *txn = NULL; mdb_mutexref_t wmutex = NULL; int rc; - size_t wsize; + size_t wsize, w3; char *ptr; #ifdef _WIN32 DWORD len, w2; @@ -9075,15 +9258,15 @@ mdb_env_copyfd0(MDB_env *env, HANDLE fd) if (rc) goto leave; - w2 = txn->mt_next_pgno * env->me_psize; + w3 = txn->mt_next_pgno * env->me_psize; { size_t fsize = 0; if ((rc = mdb_fsize(env->me_fd, &fsize))) goto leave; - if (w2 > fsize) - w2 = fsize; + if (w3 > fsize) + w3 = fsize; } - wsize = w2 - wsize; + wsize = w3 - wsize; while (wsize > 0) { if (wsize > MAX_WRITE) w2 = MAX_WRITE; @@ -9130,6 +9313,9 @@ mdb_env_copy2(MDB_env *env, const char *path, unsigned int flags) int rc, len; char *lpath; HANDLE newfd = INVALID_HANDLE_VALUE; +#ifdef _WIN32 + wchar_t *wpath; +#endif if (env->me_flags & MDB_NOSUBDIR) { lpath = (char *)path; @@ -9147,8 +9333,12 @@ mdb_env_copy2(MDB_env *env, const char *path, unsigned int flags) * already in the OS cache. */ #ifdef _WIN32 - newfd = CreateFileA(lpath, GENERIC_WRITE, 0, NULL, CREATE_NEW, + rc = utf8_to_utf16(lpath, -1, &wpath, NULL); + if (rc) + goto leave; + newfd = CreateFileW(wpath, GENERIC_WRITE, 0, NULL, CREATE_NEW, FILE_FLAG_NO_BUFFERING|FILE_FLAG_WRITE_THROUGH, NULL); + free(wpath); #else newfd = open(lpath, O_WRONLY|O_CREAT|O_EXCL, 0666); #endif @@ -9340,6 +9530,7 @@ int mdb_dbi_open(MDB_txn *txn, const char *name, unsigned int flags, MDB_dbi *db MDB_db dummy; int rc, dbflag, exact; unsigned int unused = 0, seq; + char *namedup; size_t len; if (flags & ~VALID_FLAGS) @@ -9401,8 +9592,16 @@ int mdb_dbi_open(MDB_txn *txn, const char *name, unsigned int flags, MDB_dbi *db MDB_node *node = NODEPTR(mc.mc_pg[mc.mc_top], mc.mc_ki[mc.mc_top]); if ((node->mn_flags & (F_DUPDATA|F_SUBDATA)) != F_SUBDATA) return MDB_INCOMPATIBLE; - } else if (rc == MDB_NOTFOUND && (flags & MDB_CREATE)) { - /* Create if requested */ + } else if (! (rc == MDB_NOTFOUND && (flags & MDB_CREATE))) { + return rc; + } + + /* Done here so we cannot fail after creating a new DB */ + if ((namedup = strdup(name)) == NULL) + return ENOMEM; + + if (rc) { + /* MDB_NOTFOUND and MDB_CREATE: Create new DB */ data.mv_size = sizeof(MDB_db); data.mv_data = &dummy; memset(&dummy, 0, sizeof(dummy)); @@ -9412,10 +9611,12 @@ int mdb_dbi_open(MDB_txn *txn, const char *name, unsigned int flags, MDB_dbi *db dbflag |= DB_DIRTY; } - /* OK, got info, add to table */ - if (rc == MDB_SUCCESS) { + if (rc) { + free(namedup); + } else { + /* Got info, register DBI in this txn */ unsigned int slot = unused ? unused : txn->mt_numdbs; - txn->mt_dbxs[slot].md_name.mv_data = strdup(name); + txn->mt_dbxs[slot].md_name.mv_data = namedup; txn->mt_dbxs[slot].md_name.mv_size = len; txn->mt_dbxs[slot].md_rel = NULL; txn->mt_dbflags[slot] = dbflag; @@ -9498,8 +9699,11 @@ mdb_drop0(MDB_cursor *mc, int subs) /* DUPSORT sub-DBs have no ovpages/DBs. Omit scanning leaves. * This also avoids any P_LEAF2 pages, which have no nodes. + * Also if the DB doesn't have sub-DBs and has no overflow + * pages, omit scanning leaves. */ - if (mc->mc_flags & C_SUB) + if ((mc->mc_flags & C_SUB) || + (!subs && !mc->mc_db->md_overflow_pages)) mdb_cursor_pop(mc); mdb_cursor_copy(mc, &mx); @@ -9513,7 +9717,7 @@ mdb_drop0(MDB_cursor *mc, int subs) MDB_page *omp; pgno_t pg; memcpy(&pg, NODEDATA(ni), sizeof(pg)); - rc = mdb_page_get(txn, pg, &omp, NULL); + rc = mdb_page_get(mc, pg, &omp, NULL); if (rc != 0) goto done; mdb_cassert(mc, IS_OVERFLOW(omp)); @@ -9521,6 +9725,9 @@ mdb_drop0(MDB_cursor *mc, int subs) pg, omp->mp_pages); if (rc) goto done; + mc->mc_db->md_overflow_pages -= omp->mp_pages; + if (!mc->mc_db->md_overflow_pages && !subs) + break; } else if (subs && (ni->mn_flags & F_SUBDATA)) { mdb_xcursor_init1(mc, ni); rc = mdb_drop0(&mc->mc_xcursor->mx_cursor, 0); @@ -9528,6 +9735,8 @@ mdb_drop0(MDB_cursor *mc, int subs) goto done; } } + if (!subs && !mc->mc_db->md_overflow_pages) + goto pop; } else { if ((rc = mdb_midl_need(&txn->mt_free_pgs, n)) != 0) goto done; @@ -9549,6 +9758,7 @@ mdb_drop0(MDB_cursor *mc, int subs) /* no more siblings, go back to beginning * of previous level. */ +pop: mdb_cursor_pop(mc); mc->mc_ki[0] = 0; for (i=1; imc_snum; i++) { @@ -9565,6 +9775,7 @@ done: } else if (rc == MDB_NOTFOUND) { rc = MDB_SUCCESS; } + mc->mc_flags &= ~C_INITIALIZED; return rc; } @@ -9853,3 +10064,24 @@ mdb_mutex_failed(MDB_env *env, mdb_mutexref_t mutex, int rc) } #endif /* MDB_ROBUST_SUPPORTED */ /** @} */ + +#if defined(_WIN32) +static int utf8_to_utf16(const char *src, int srcsize, wchar_t **dst, int *dstsize) +{ + int need; + wchar_t *result; + need = MultiByteToWideChar(CP_UTF8, 0, src, srcsize, NULL, 0); + if (need == 0xFFFD) + return EILSEQ; + if (need == 0) + return EINVAL; + result = malloc(sizeof(wchar_t) * need); + if (!result) + return ENOMEM; + MultiByteToWideChar(CP_UTF8, 0, src, srcsize, result, need); + if (dstsize) + *dstsize = need; + *dst = result; + return 0; +} +#endif /* defined(_WIN32) */