* BerkeleyDB API, but much simplified.
*/
/*
- * Copyright 2011-2016 Howard Chu, Symas Corp.
+ * Copyright 2011-2017 Howard Chu, Symas Corp.
* All rights reserved.
*
* Redistribution and use in source and binary forms, with or without
#ifdef _WIN32
#include <malloc.h>
#include <windows.h>
+#include <wchar.h> /* get wcscpy() */
+
/** getpid() returns int; MinGW defines pid_t but MinGW64 typedefs it
* as int64 which is wrong. MSVC doesn't define it at all, so just
* don't use it.
/* Most platforms have posix_memalign, older may only have memalign */
#define HAVE_MEMALIGN 1
#include <malloc.h>
+/* On Solaris, we need the POSIX sigwait function */
+#if defined (__sun)
+# define _POSIX_PTHREAD_SEMANTICS 1
+#endif
#endif
#if !(defined(BYTE_ORDER) || defined(__BYTE_ORDER))
#include <resolv.h> /* defines BYTE_ORDER on HPUX and Solaris */
#endif
-#if defined(__APPLE__) || defined (BSD)
+#if defined(__APPLE__) || defined (BSD) || defined(__FreeBSD_kernel__)
# define MDB_USE_POSIX_SEM 1
# define MDB_FDATASYNC fsync
#elif defined(ANDROID)
#ifndef _WIN32
#include <pthread.h>
+#include <signal.h>
#ifdef MDB_USE_POSIX_SEM
# define MDB_USE_HASH 1
#include <semaphore.h>
# define MDB_USE_ROBUST 0
# else
# define MDB_USE_ROBUST 1
+# endif
+#endif /* !MDB_USE_ROBUST */
+
+#if defined(MDB_USE_POSIX_MUTEX) && (MDB_USE_ROBUST)
/* glibc < 2.12 only provided _np API */
# if (defined(__GLIBC__) && GLIBC_VER < 0x02000c) || \
(defined(PTHREAD_MUTEX_ROBUST_NP) && !defined(PTHREAD_MUTEX_ROBUST))
# define pthread_mutexattr_setrobust(attr, flag) pthread_mutexattr_setrobust_np(attr, flag)
# define pthread_mutex_consistent(mutex) pthread_mutex_consistent_np(mutex)
# endif
-# endif
-#endif /* MDB_USE_ROBUST */
+#endif /* MDB_USE_POSIX_MUTEX && MDB_USE_ROBUST */
-#if defined(MDB_OWNERDEAD) && MDB_USE_ROBUST
+#if defined(MDB_OWNERDEAD) && (MDB_USE_ROBUST)
#define MDB_ROBUST_SUPPORTED 1
#endif
#define pthread_mutex_lock(x) WaitForSingleObject(*x, INFINITE)
#define pthread_cond_signal(x) SetEvent(*x)
#define pthread_cond_wait(cond,mutex) do{SignalObjectAndWait(*mutex, *cond, INFINITE, FALSE); WaitForSingleObject(*mutex, INFINITE);}while(0)
-#define THREAD_CREATE(thr,start,arg) thr=CreateThread(NULL,0,start,arg,0,NULL)
-#define THREAD_FINISH(thr) WaitForSingleObject(thr, INFINITE)
+#define THREAD_CREATE(thr,start,arg) \
+ (((thr) = CreateThread(NULL, 0, start, arg, 0, NULL)) ? 0 : ErrCode())
+#define THREAD_FINISH(thr) \
+ (WaitForSingleObject(thr, INFINITE) ? ErrCode() : 0)
#define LOCK_MUTEX0(mutex) WaitForSingleObject(mutex, INFINITE)
#define UNLOCK_MUTEX(mutex) ReleaseMutex(mutex)
#define mdb_mutex_consistent(mutex) 0
}
#else /* MDB_USE_POSIX_MUTEX: */
- /** Shared mutex/semaphore as it is stored (mdb_mutex_t), and as
- * local variables keep it (mdb_mutexref_t).
+ /** Shared mutex/semaphore as the original is stored.
*
- * When #mdb_mutexref_t is a pointer declaration and #mdb_mutex_t is
- * not, then it is array[size 1] so it can be assigned to a pointer.
- * @{
+ * Not for copies. Instead it can be assigned to an #mdb_mutexref_t.
+ * When mdb_mutexref_t is a pointer and mdb_mutex_t is not, then it
+ * is array[size 1] so it can be assigned to the pointer.
*/
-typedef pthread_mutex_t mdb_mutex_t[1], *mdb_mutexref_t;
- /* @} */
+typedef pthread_mutex_t mdb_mutex_t[1];
+ /** Reference to an #mdb_mutex_t */
+typedef pthread_mutex_t *mdb_mutexref_t;
/** Lock the reader or writer mutex.
* Returns 0 or a code to give #mdb_mutex_failed(), as in #LOCK_MUTEX().
*/
+ (((MDB_PIDLOCK) != 0) << 16)))
/** @} */
-/** Common header for all page types.
- * Overflow records occupy a number of contiguous pages with no
- * headers on any page after the first.
+/** Common header for all page types. The page type depends on #mp_flags.
+ *
+ * #P_BRANCH and #P_LEAF pages have unsorted '#MDB_node's at the end, with
+ * sorted #mp_ptrs[] entries referring to them. Exception: #P_LEAF2 pages
+ * omit mp_ptrs and pack sorted #MDB_DUPFIXED values after the page header.
+ *
+ * #P_OVERFLOW records occupy one or more contiguous pages where only the
+ * first has a page header. They hold the real data of #F_BIGDATA nodes.
+ *
+ * #P_SUBP sub-pages are small leaf "pages" with duplicate data.
+ * A node with flag #F_DUPDATA but not #F_SUBDATA contains a sub-page.
+ * (Duplicate data can also go in sub-databases, which use normal pages.)
+ *
+ * #P_META pages contain #MDB_meta, the start point of an LMDB snapshot.
+ *
+ * Each non-metapage up to #MDB_meta.%mm_last_pg is reachable exactly once
+ * in the snapshot: Either used by a database or listed in a freeDB record.
*/
typedef struct MDB_page {
#define mp_pgno mp_p.p_pgno
pgno_t p_pgno; /**< page number */
struct MDB_page *p_next; /**< for in-memory list of freed pages */
} mp_p;
- uint16_t mp_pad;
+ uint16_t mp_pad; /**< key size if this is a LEAF2 page */
/** @defgroup mdb_page Page Flags
* @ingroup internal
* Flags for the page headers.
/** The number of overflow pages needed to store the given size. */
#define OVPAGES(size, psize) ((PAGEHDRSZ-1 + (size)) / (psize) + 1)
- /** Link in #MDB_txn.%mt_loose_pgs list */
+ /** Link in #MDB_txn.%mt_loose_pgs list.
+ * Kept outside the page header, which is needed when reusing the page.
+ */
#define NEXT_LOOSE_PAGE(p) (*(MDB_page **)((p) + 2))
/** Header for a single key/data pair within a page.
* Used in pages of type #P_BRANCH and #P_LEAF without #P_LEAF2.
* We guarantee 2-byte alignment for 'MDB_node's.
+ *
+ * #mn_lo and #mn_hi are used for data size on leaf nodes, and for child
+ * pgno on branch nodes. On 64 bit platforms, #mn_flags is also used
+ * for pgno. (Branch nodes have no flags). Lo and hi are in host byte
+ * order in case some accesses can be optimized to 32-bit word access.
+ *
+ * Leaf node flags describe node contents. #F_BIGDATA says the node's
+ * data part is the page number of an overflow page with actual data.
+ * #F_DUPDATA and #F_SUBDATA can be combined giving duplicate data in
+ * a sub-page/sub-database, and named databases (just #F_SUBDATA).
*/
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 have no flags).
- * They are in host byte order in case that lets some
- * accesses be optimized into a 32-bit word access.
- */
+ /** part of data size or pgno
+ * @{ */
#if BYTE_ORDER == LITTLE_ENDIAN
- unsigned short mn_lo, mn_hi; /**< part of data size or pgno */
+ unsigned short mn_lo, mn_hi;
#else
unsigned short mn_hi, mn_lo;
#endif
+ /** @} */
/** @defgroup mdb_node Node Flags
* @ingroup internal
* Flags for node headers.
pgno_t md_root; /**< the root page of this tree */
} MDB_db;
- /** mdb_dbi_open flags */
#define MDB_VALID 0x8000 /**< DB handle is valid, for me_dbflags */
#define PERSISTENT_FLAGS (0xffff & ~(MDB_VALID))
+ /** #mdb_dbi_open() flags */
#define VALID_FLAGS (MDB_REVERSEKEY|MDB_DUPSORT|MDB_INTEGERKEY|MDB_DUPFIXED|\
MDB_INTEGERDUP|MDB_REVERSEDUP|MDB_CREATE)
#define mm_psize mm_dbs[FREE_DBI].md_pad
/** Any persistent environment flags. @ref mdb_env */
#define mm_flags mm_dbs[FREE_DBI].md_flags
- pgno_t mm_last_pg; /**< last used page in file */
+ /** Last used page in the datafile.
+ * Actually the file may be shorter if the freeDB lists the final pages.
+ */
+ pgno_t mm_last_pg;
volatile txnid_t mm_txnid; /**< txnid that committed this page */
} MDB_meta;
* in this transaction, linked through #NEXT_LOOSE_PAGE(page).
*/
MDB_page *mt_loose_pgs;
- /* #Number of loose pages (#mt_loose_pgs) */
+ /** Number of loose pages (#mt_loose_pgs) */
int mt_loose_count;
/** The sorted list of dirty pages we temporarily wrote to disk
* because the dirty list was full. page numbers in here are
struct MDB_env {
HANDLE me_fd; /**< The main data file */
HANDLE me_lfd; /**< The lock file */
- HANDLE me_mfd; /**< just for writing the meta pages */
+ HANDLE me_mfd; /**< For writing and syncing the meta pages */
/** Failed to update the meta page. Probably an I/O error. */
#define MDB_FATAL_ERROR 0x80000000U
/** Some fields are initialized. */
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);
+struct MDB_name;
+static int utf8_to_utf16(const char *src, struct MDB_name *dst, int xtra);
#endif
/** Return the library version info. */
pgno, ((MDB_meta *)METADATA(mp))->mm_txnid);
return;
default:
- fprintf(stderr, "Bad page %"Z"u flags 0x%u\n", pgno, mp->mp_flags);
+ fprintf(stderr, "Bad page %"Z"u flags 0x%X\n", pgno, mp->mp_flags);
return;
}
}
}
if (freecount + count + NUM_METAS != txn->mt_next_pgno) {
- fprintf(stderr, "audit: %lu freecount: %lu count: %lu total: %lu next_pgno: %lu\n",
+ fprintf(stderr, "audit: %"Z"u freecount: %"Z"u count: %"Z"u total: %"Z"u next_pgno: %"Z"u\n",
txn->mt_txnid, freecount, count+NUM_METAS,
freecount+count+NUM_METAS, txn->mt_next_pgno);
}
/** Allocate memory for a page.
* Re-use old malloc'd pages first for singletons, otherwise just malloc.
+ * Set #MDB_TXN_ERROR on failure.
*/
static MDB_page *
mdb_page_malloc(MDB_txn *txn, unsigned num)
}
/** Allocate page numbers and memory for writing. Maintain me_pglast,
- * me_pghead and mt_next_pgno.
+ * me_pghead and mt_next_pgno. Set #MDB_TXN_ERROR on failure.
*
* If there are free pages available from older transactions, they
* are re-used first. Otherwise allocate a new page at mt_next_pgno.
np = m2.mc_pg[m2.mc_top];
leaf = NODEPTR(np, m2.mc_ki[m2.mc_top]);
if ((rc = mdb_node_read(&m2, leaf, &data)) != MDB_SUCCESS)
- return rc;
+ goto fail;
idl = (MDB_ID *) data.mv_data;
i = idl[0];
}
/** Touch a page: make it dirty and re-insert into tree with updated pgno.
+ * Set #MDB_TXN_ERROR on failure.
* @param[in] mc cursor pointing to the page to be touched
* @return 0 on success, non-zero on failure.
*/
len = sizeof(MDB_meta) - off;
off += (char *)mp - env->me_map;
- /* Write to the SYNC fd */
+ /* Write to the SYNC fd unless MDB_NOSYNC/MDB_NOMETASYNC.
+ * (me_mfd goes to the same file as me_fd, but writing to it
+ * also syncs to disk. Avoids a separate fdatasync() call.)
+ */
mfd = (flags & (MDB_NOSYNC|MDB_NOMETASYNC)) ? env->me_fd : env->me_mfd;
#ifdef _WIN32
{
return MDB_SUCCESS;
}
+
+#ifdef _WIN32
+typedef wchar_t mdb_nchar_t;
+# define MDB_NAME(str) L##str
+# define mdb_name_cpy wcscpy
+#else
+/** Character type for file names: char on Unix, wchar_t on Windows */
+typedef char mdb_nchar_t;
+# define MDB_NAME(str) str /**< #mdb_nchar_t[] string literal */
+# define mdb_name_cpy strcpy /**< Copy name (#mdb_nchar_t string) */
+#endif
+
+/** Filename - string of #mdb_nchar_t[] */
+typedef struct MDB_name {
+ int mn_len; /**< Length */
+ int mn_alloced; /**< True if #mn_val was malloced */
+ mdb_nchar_t *mn_val; /**< Contents */
+} MDB_name;
+
+/** Filename suffixes [datafile,lockfile][without,with MDB_NOSUBDIR] */
+static const mdb_nchar_t *const mdb_suffixes[2][2] = {
+ { MDB_NAME("/data.mdb"), MDB_NAME("") },
+ { MDB_NAME("/lock.mdb"), MDB_NAME("-lock") }
+};
+
+#define MDB_SUFFLEN 9 /**< Max string length in #mdb_suffixes[] */
+
+/** Set up filename + scratch area for filename suffix, for opening files.
+ * It should be freed with #mdb_fname_destroy().
+ * On Windows, paths are converted from char *UTF-8 to wchar_t *UTF-16.
+ *
+ * @param[in] path Pathname for #mdb_env_open().
+ * @param[in] envflags Whether a subdir and/or lockfile will be used.
+ * @param[out] fname Resulting filename, with room for a suffix if necessary.
+ */
+static int ESECT
+mdb_fname_init(const char *path, unsigned envflags, MDB_name *fname)
+{
+ int no_suffix = F_ISSET(envflags, MDB_NOSUBDIR|MDB_NOLOCK);
+ fname->mn_alloced = 0;
+#ifdef _WIN32
+ return utf8_to_utf16(path, fname, no_suffix ? 0 : MDB_SUFFLEN);
+#else
+ fname->mn_len = strlen(path);
+ if (no_suffix)
+ fname->mn_val = (char *) path;
+ else if ((fname->mn_val = malloc(fname->mn_len + MDB_SUFFLEN+1)) != NULL) {
+ fname->mn_alloced = 1;
+ strcpy(fname->mn_val, path);
+ }
+ else
+ return ENOMEM;
+ return MDB_SUCCESS;
+#endif
+}
+
+/** Destroy \b fname from #mdb_fname_init() */
+#define mdb_fname_destroy(fname) \
+ do { if ((fname).mn_alloced) free((fname).mn_val); } while (0)
+
+#ifdef O_CLOEXEC /* POSIX.1-2008: Set FD_CLOEXEC atomically at open() */
+# define MDB_CLOEXEC O_CLOEXEC
+#else
+# define MDB_CLOEXEC 0
+#endif
+
+/** File type, access mode etc. for #mdb_fopen() */
+enum mdb_fopen_type {
+#ifdef _WIN32
+ MDB_O_RDONLY, MDB_O_RDWR, MDB_O_META, MDB_O_COPY, MDB_O_LOCKS
+#else
+ /* A comment in mdb_fopen() explains some O_* flag choices. */
+ MDB_O_RDONLY= O_RDONLY, /**< for RDONLY me_fd */
+ MDB_O_RDWR = O_RDWR |O_CREAT, /**< for me_fd */
+ MDB_O_META = O_WRONLY|MDB_DSYNC |MDB_CLOEXEC, /**< for me_mfd */
+ MDB_O_COPY = O_WRONLY|O_CREAT|O_EXCL|MDB_CLOEXEC, /**< for #mdb_env_copy() */
+ /** Bitmask for open() flags in enum #mdb_fopen_type. The other bits
+ * distinguish otherwise-equal MDB_O_* constants from each other.
+ */
+ MDB_O_MASK = MDB_O_RDWR|MDB_CLOEXEC | MDB_O_RDONLY|MDB_O_META|MDB_O_COPY,
+ MDB_O_LOCKS = MDB_O_RDWR|MDB_CLOEXEC | ((MDB_O_MASK+1) & ~MDB_O_MASK) /**< for me_lfd */
+#endif
+};
+
+/** Open an LMDB file.
+ * @param[in] env The LMDB environment.
+ * @param[in,out] fname Path from from #mdb_fname_init(). A suffix is
+ * appended if necessary to create the filename, without changing mn_len.
+ * @param[in] which Determines file type, access mode, etc.
+ * @param[in] mode The Unix permissions for the file, if we create it.
+ * @param[out] res Resulting file handle.
+ * @return 0 on success, non-zero on failure.
+ */
+static int ESECT
+mdb_fopen(const MDB_env *env, MDB_name *fname,
+ enum mdb_fopen_type which, mdb_mode_t mode,
+ HANDLE *res)
+{
+ int rc = MDB_SUCCESS;
+ HANDLE fd;
+#ifdef _WIN32
+ DWORD acc, share, disp, attrs;
+#else
+ int flags;
+#endif
+
+ if (fname->mn_alloced) /* modifiable copy */
+ mdb_name_cpy(fname->mn_val + fname->mn_len,
+ mdb_suffixes[which==MDB_O_LOCKS][F_ISSET(env->me_flags, MDB_NOSUBDIR)]);
+
+ /* The directory must already exist. Usually the file need not.
+ * MDB_O_META requires the file because we already created it using
+ * MDB_O_RDWR. MDB_O_COPY must not overwrite an existing file.
+ *
+ * With MDB_O_COPY we do not want the OS to cache the writes, since
+ * the source data is already in the OS cache.
+ *
+ * The lockfile needs FD_CLOEXEC (close file descriptor on exec*())
+ * to avoid the flock() issues noted under Caveats in lmdb.h.
+ * Also set it for other filehandles which the user cannot get at
+ * and close himself, which he may need after fork(). I.e. all but
+ * me_fd, which programs do use via mdb_env_get_fd().
+ */
+
+#ifdef _WIN32
+ acc = GENERIC_READ|GENERIC_WRITE;
+ share = FILE_SHARE_READ|FILE_SHARE_WRITE;
+ disp = OPEN_ALWAYS;
+ attrs = FILE_ATTRIBUTE_NORMAL;
+ switch (which) {
+ case MDB_O_RDONLY: /* read-only datafile */
+ acc = GENERIC_READ;
+ disp = OPEN_EXISTING;
+ break;
+ case MDB_O_META: /* for writing metapages */
+ acc = GENERIC_WRITE;
+ disp = OPEN_EXISTING;
+ attrs = FILE_ATTRIBUTE_NORMAL|FILE_FLAG_WRITE_THROUGH;
+ break;
+ case MDB_O_COPY: /* mdb_env_copy() & co */
+ acc = GENERIC_WRITE;
+ share = 0;
+ disp = CREATE_NEW;
+ attrs = FILE_FLAG_NO_BUFFERING|FILE_FLAG_WRITE_THROUGH;
+ break;
+ default: break; /* silence gcc -Wswitch (not all enum values handled) */
+ }
+ fd = CreateFileW(fname->mn_val, acc, share, NULL, disp, attrs, NULL);
+#else
+ fd = open(fname->mn_val, which & MDB_O_MASK, mode);
+#endif
+
+ if (fd == INVALID_HANDLE_VALUE)
+ rc = ErrCode();
+#ifndef _WIN32
+ else {
+ if (which != MDB_O_RDONLY && which != MDB_O_RDWR) {
+ /* Set CLOEXEC if we could not pass it to open() */
+ if (!MDB_CLOEXEC && (flags = fcntl(fd, F_GETFD)) != -1)
+ (void) fcntl(fd, F_SETFD, flags | FD_CLOEXEC);
+ }
+ if (which == MDB_O_COPY && env->me_psize >= env->me_os_psize) {
+ /* This may require buffer alignment. There is no portable
+ * way to ask how much, so we require OS pagesize alignment.
+ */
+# ifdef F_NOCACHE /* __APPLE__ */
+ (void) fcntl(fd, F_NOCACHE, 1);
+# elif defined O_DIRECT
+ /* open(...O_DIRECT...) would break on filesystems without
+ * O_DIRECT support (ITS#7682). Try to set it here instead.
+ */
+ if ((flags = fcntl(fd, F_GETFL)) != -1)
+ (void) fcntl(fd, F_SETFL, flags | O_DIRECT);
+# endif
+ }
+ }
+#endif /* !_WIN32 */
+
+ *res = fd;
+ return rc;
+}
+
+
#ifdef BROKEN_FDATASYNC
#include <sys/utsname.h>
#include <sys/vfs.h>
{
MDB_reader *reader = ptr;
- reader->mr_pid = 0;
+#ifndef _WIN32
+ if (reader->mr_pid == getpid()) /* catch pthread_exit() in child process */
+#endif
+ /* We omit the mutex, so do this atomically (i.e. skip mr_txnid) */
+ reader->mr_pid = 0;
}
#ifdef _WIN32
/** Open and/or initialize the lock region for the environment.
* @param[in] env The LMDB environment.
- * @param[in] lpath The pathname of the file used for the lock region.
+ * @param[in] fname Filename + scratch area, from #mdb_fname_init().
* @param[in] mode The Unix permissions for the file, if we create it.
* @param[in,out] excl In -1, out lock type: -1 none, 0 shared, 1 exclusive
* @return 0 on success, non-zero on failure.
*/
static int ESECT
-mdb_env_setup_locks(MDB_env *env, char *lpath, int mode, int *excl)
+mdb_env_setup_locks(MDB_env *env, MDB_name *fname, int mode, int *excl)
{
#ifdef _WIN32
# define MDB_ERRCODE_ROFS ERROR_WRITE_PROTECT
#else
# define MDB_ERRCODE_ROFS EROFS
-#ifdef O_CLOEXEC /* Linux: Open file and set FD_CLOEXEC atomically */
-# define MDB_CLOEXEC O_CLOEXEC
-#else
- int fdflags;
-# define MDB_CLOEXEC 0
-#endif
#endif
int rc;
off_t size, rsize;
-#ifdef _WIN32
- 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
- if (env->me_lfd == INVALID_HANDLE_VALUE) {
- rc = ErrCode();
+ rc = mdb_fopen(env, fname, MDB_O_LOCKS, mode, &env->me_lfd);
+ if (rc) {
+ /* Omit lockfile if read-only env on read-only filesystem */
if (rc == MDB_ERRCODE_ROFS && (env->me_flags & MDB_RDONLY)) {
return MDB_SUCCESS;
}
- goto fail_errno;
+ goto fail;
}
-#if ! ((MDB_CLOEXEC) || defined(_WIN32))
- /* Lose record locks when exec*() */
- if ((fdflags = fcntl(env->me_lfd, F_GETFD) | FD_CLOEXEC) >= 0)
- fcntl(env->me_lfd, F_SETFD, fdflags);
-#endif
if (!(env->me_flags & MDB_NOTLS)) {
rc = pthread_key_create(&env->me_txkey, mdb_env_reader_dest);
return rc;
}
- /** 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"
- /** The suffix of the lock file when no subdir is used */
-#define LOCKSUFF "-lock"
/** 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.
int ESECT
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
+ int rc, excl = -1;
+ MDB_name fname;
if (env->me_fd!=INVALID_HANDLE_VALUE || (flags & ~(CHANGEABLE|CHANGELESS)))
return EINVAL;
- len = strlen(path);
- if (flags & MDB_NOSUBDIR) {
- rc = len + sizeof(LOCKSUFF) + len + 1;
- } else {
- rc = len + sizeof(LOCKNAME) + len + sizeof(DATANAME);
- }
- lpath = malloc(rc);
- if (!lpath)
- return ENOMEM;
- if (flags & MDB_NOSUBDIR) {
- dpath = lpath + len + sizeof(LOCKSUFF);
- sprintf(lpath, "%s" LOCKSUFF, path);
- strcpy(dpath, path);
- } else {
- dpath = lpath + len + sizeof(LOCKNAME);
- sprintf(lpath, "%s" LOCKNAME, path);
- sprintf(dpath, "%s" DATANAME, path);
- }
-
- rc = MDB_SUCCESS;
flags |= env->me_flags;
+
+ rc = mdb_fname_init(path, flags, &fname);
+ if (rc)
+ return rc;
+
if (flags & MDB_RDONLY) {
/* silently ignore WRITEMAP when we're only getting read access */
flags &= ~MDB_WRITEMAP;
/* For RDONLY, get lockfile after we know datafile exists */
if (!(flags & (MDB_RDONLY|MDB_NOLOCK))) {
- rc = mdb_env_setup_locks(env, lpath, mode, &excl);
+ rc = mdb_env_setup_locks(env, &fname, mode, &excl);
if (rc)
goto leave;
}
-#ifdef _WIN32
- if (F_ISSET(flags, MDB_RDONLY)) {
- oflags = GENERIC_READ;
- len = OPEN_EXISTING;
- } else {
- oflags = GENERIC_READ|GENERIC_WRITE;
- len = OPEN_ALWAYS;
- }
- mode = FILE_ATTRIBUTE_NORMAL;
- rc = utf8_to_utf16(dpath, -1, &wpath, NULL);
+ rc = mdb_fopen(env, &fname,
+ (flags & MDB_RDONLY) ? MDB_O_RDONLY : MDB_O_RDWR,
+ mode, &env->me_fd);
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;
- else
- oflags = O_RDWR | O_CREAT;
-
- env->me_fd = open(dpath, oflags, mode);
-#endif
- if (env->me_fd == INVALID_HANDLE_VALUE) {
- rc = ErrCode();
- goto leave;
- }
if ((flags & (MDB_RDONLY|MDB_NOLOCK)) == MDB_RDONLY) {
- rc = mdb_env_setup_locks(env, lpath, mode, &excl);
+ rc = mdb_env_setup_locks(env, &fname, mode, &excl);
if (rc)
goto leave;
}
if ((rc = mdb_env_open2(env)) == MDB_SUCCESS) {
- if (flags & (MDB_RDONLY|MDB_WRITEMAP)) {
- env->me_mfd = env->me_fd;
- } else {
+ if (!(flags & (MDB_RDONLY|MDB_WRITEMAP))) {
/* Synchronous fd for meta writes. Needed even with
* MDB_NOSYNC/MDB_NOMETASYNC, in case these get reset.
*/
-#ifdef _WIN32
- len = OPEN_EXISTING;
- rc = utf8_to_utf16(dpath, -1, &wpath, NULL);
+ rc = mdb_fopen(env, &fname, MDB_O_META, mode, &env->me_mfd);
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);
-#endif
- if (env->me_mfd == INVALID_HANDLE_VALUE) {
- rc = ErrCode();
- goto leave;
- }
}
DPRINTF(("opened dbenv %p", (void *) env));
if (excl > 0) {
if (rc) {
mdb_env_close0(env, excl);
}
- free(lpath);
+ mdb_fname_destroy(fname);
return rc;
}
if (env->me_map) {
munmap(env->me_map, env->me_mapsize);
}
- if (env->me_mfd != env->me_fd && env->me_mfd != INVALID_HANDLE_VALUE)
+ if (env->me_mfd != INVALID_HANDLE_VALUE)
(void) close(env->me_mfd);
if (env->me_fd != INVALID_HANDLE_VALUE)
(void) close(env->me_fd);
}
}
-/** Push a page onto the top of the cursor's stack. */
+/** Push a page onto the top of the cursor's stack.
+ * Set #MDB_TXN_ERROR on failure.
+ */
static int
mdb_cursor_push(MDB_cursor *mc, MDB_page *mp)
{
}
/** Find the address of the page corresponding to a given page number.
+ * Set #MDB_TXN_ERROR on failure.
* @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.
if (flags & (MDB_PS_FIRST|MDB_PS_LAST)) {
i = 0;
- if (flags & MDB_PS_LAST)
+ if (flags & MDB_PS_LAST) {
i = NUMKEYS(mp) - 1;
+ /* if already init'd, see if we're already in right place */
+ if (mc->mc_flags & C_INITIALIZED) {
+ if (mc->mc_ki[mc->mc_top] == i) {
+ mc->mc_top = mc->mc_snum++;
+ mp = mc->mc_pg[mc->mc_top];
+ goto ready;
+ }
+ }
+ }
} else {
int exact;
node = mdb_node_search(mc, key, &exact);
if ((rc = mdb_cursor_push(mc, mp)))
return rc;
+ready:
if (flags & MDB_PS_MODIFY) {
if ((rc = mdb_page_touch(mc)) != 0)
return rc;
MDB_node *leaf;
int rc;
- if ((mc->mc_flags & C_EOF) ||
- ((mc->mc_flags & C_DEL) && op == MDB_NEXT_DUP)) {
+ if ((mc->mc_flags & C_DEL && op == MDB_NEXT_DUP))
return MDB_NOTFOUND;
- }
+
if (!(mc->mc_flags & C_INITIALIZED))
return mdb_cursor_first(mc, key, data);
mp = mc->mc_pg[mc->mc_top];
+ if (mc->mc_flags & C_EOF) {
+ if (mc->mc_ki[mc->mc_top] >= NUMKEYS(mp)-1)
+ return MDB_NOTFOUND;
+ mc->mc_flags ^= C_EOF;
+ }
+
if (mc->mc_db->md_flags & MDB_DUPSORT) {
leaf = NODEPTR(mp, mc->mc_ki[mc->mc_top]);
if (F_ISSET(leaf->mn_flags, F_DUPDATA)) {
}
}
rc = 0;
+ mc->mc_flags &= ~C_EOF;
goto set2;
}
}
if (mc->mc_xcursor)
mc->mc_xcursor->mx_cursor.mc_flags &= ~(C_INITIALIZED|C_EOF);
- if (!(mc->mc_flags & C_EOF)) {
-
- if (!(mc->mc_flags & C_INITIALIZED) || mc->mc_top) {
- rc = mdb_page_search(mc, NULL, MDB_PS_LAST);
- if (rc != MDB_SUCCESS)
- return rc;
- }
- mdb_cassert(mc, IS_LEAF(mc->mc_pg[mc->mc_top]));
-
+ if (!(mc->mc_flags & C_INITIALIZED) || mc->mc_top) {
+ rc = mdb_page_search(mc, NULL, MDB_PS_LAST);
+ if (rc != MDB_SUCCESS)
+ return rc;
}
+ mdb_cassert(mc, IS_LEAF(mc->mc_pg[mc->mc_top]));
+
mc->mc_ki[mc->mc_top] = NUMKEYS(mc->mc_pg[mc->mc_top]) - 1;
mc->mc_flags |= C_INITIALIZED|C_EOF;
leaf = NODEPTR(mc->mc_pg[mc->mc_top], mc->mc_ki[mc->mc_top]);
}
/** Allocate and initialize new pages for a database.
+ * Set #MDB_TXN_ERROR on failure.
* @param[in] mc a cursor on the database being added to.
* @param[in] flags flags defining what type of page is being allocated.
* @param[in] num the number of pages to allocate. This is usually 1,
}
/** Add a node to the page pointed to by the cursor.
+ * Set #MDB_TXN_ERROR on failure.
* @param[in] mc The cursor for this operation.
* @param[in] indx The index on the page where the new node should be added.
* @param[in] key The key for the new node.
if (!(mc->mc_flags & C_INITIALIZED))
return EINVAL;
- if (!mc->mc_snum || (mc->mc_flags & C_EOF))
+ if (!mc->mc_snum)
return MDB_NOTFOUND;
+ if (mc->mc_flags & C_EOF) {
+ if (mc->mc_ki[mc->mc_top] >= NUMKEYS(mc->mc_pg[mc->mc_top]))
+ return MDB_NOTFOUND;
+ mc->mc_flags ^= C_EOF;
+ }
+
leaf = NODEPTR(mc->mc_pg[mc->mc_top], mc->mc_ki[mc->mc_top]);
if (!F_ISSET(leaf->mn_flags, F_DUPDATA)) {
*countp = 1;
}
/** Replace the key for a branch node with a new key.
+ * Set #MDB_TXN_ERROR on failure.
* @param[in] mc Cursor pointing to the node to operate on.
* @param[in] key The new key to use.
* @return 0 on success, non-zero on failure.
}
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 this node has dupdata, it may need to be reinited
+ * because its data has moved.
+ * If the xcursor was not initd it must be reinited.
+ * Else if node points to a subDB, nothing is needed.
+ * Else (xcursor was initd, not a subDB) needs mc_pg[0] reset.
*/
- 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
+ if (node->mn_flags & F_DUPDATA) {
+ if (m3->mc_xcursor->mx_cursor.mc_flags & C_INITIALIZED) {
+ if (!(node->mn_flags & F_SUBDATA))
+ m3->mc_xcursor->mx_cursor.mc_pg[0] = NODEDATA(node);
+ } else
mdb_xcursor_init1(m3, node);
}
}
}
/** Split a page and insert a new node.
+ * Set #MDB_TXN_ERROR on failure.
* @param[in,out] mc Cursor pointing to the page and desired insertion index.
* The cursor will be updated to point to the actual page and index where
* the node got inserted after the split.
#ifndef MDB_WBUF
#define MDB_WBUF (1024*1024)
#endif
+#define MDB_EOF 0x10 /**< #mdb_env_copyfd1() is done reading */
- /** State needed for a compacting copy. */
+ /** State needed for a double-buffering compacting copy. */
typedef struct mdb_copy {
+ MDB_env *mc_env;
+ MDB_txn *mc_txn;
pthread_mutex_t mc_mutex;
- pthread_cond_t mc_cond;
+ pthread_cond_t mc_cond; /**< Condition variable for #mc_new */
char *mc_wbuf[2];
char *mc_over[2];
- MDB_env *mc_env;
- MDB_txn *mc_txn;
int mc_wlen[2];
int mc_olen[2];
pgno_t mc_next_pgno;
HANDLE mc_fd;
- int mc_status;
- volatile int mc_new;
- int mc_toggle;
-
+ int mc_toggle; /**< Buffer number in provider */
+ int mc_new; /**< (0-2 buffers to write) | (#MDB_EOF at end) */
+ /** Error code. Never cleared if set. Both threads can set nonzero
+ * to fail the copy. Not mutex-protected, LMDB expects atomic int.
+ */
+ volatile int mc_error;
} mdb_copy;
/** Dedicated writer thread for compacting copy. */
#else
int len;
#define DO_WRITE(rc, fd, ptr, w2, len) len = write(fd, ptr, w2); rc = (len >= 0)
+#ifdef SIGPIPE
+ sigset_t set;
+ sigemptyset(&set);
+ sigaddset(&set, SIGPIPE);
+ if ((rc = pthread_sigmask(SIG_BLOCK, &set, NULL)) != 0)
+ my->mc_error = rc;
+#endif
#endif
pthread_mutex_lock(&my->mc_mutex);
- my->mc_new = 0;
- pthread_cond_signal(&my->mc_cond);
for(;;) {
while (!my->mc_new)
pthread_cond_wait(&my->mc_cond, &my->mc_mutex);
- if (my->mc_new < 0) {
- my->mc_new = 0;
+ if (my->mc_new == 0 + MDB_EOF) /* 0 buffers, just EOF */
break;
- }
- my->mc_new = 0;
wsize = my->mc_wlen[toggle];
ptr = my->mc_wbuf[toggle];
again:
- while (wsize > 0) {
+ rc = MDB_SUCCESS;
+ while (wsize > 0 && !my->mc_error) {
DO_WRITE(rc, my->mc_fd, ptr, wsize, len);
if (!rc) {
rc = ErrCode();
+#if defined(SIGPIPE) && !defined(_WIN32)
+ if (rc == EPIPE) {
+ /* Collect the pending SIGPIPE, otherwise at least OS X
+ * gives it to the process on thread-exit (ITS#8504).
+ */
+ int tmp;
+ sigwait(&set, &tmp);
+ }
+#endif
break;
} else if (len > 0) {
rc = MDB_SUCCESS;
}
}
if (rc) {
- my->mc_status = rc;
- break;
+ my->mc_error = rc;
}
/* If there's an overflow page tail, write it too */
if (my->mc_olen[toggle]) {
}
my->mc_wlen[toggle] = 0;
toggle ^= 1;
+ /* Return the empty buffer to provider */
+ my->mc_new--;
pthread_cond_signal(&my->mc_cond);
}
- pthread_cond_signal(&my->mc_cond);
pthread_mutex_unlock(&my->mc_mutex);
return (THREAD_RET)0;
#undef DO_WRITE
}
- /** Tell the writer thread there's a buffer ready to write */
+ /** Give buffer and/or #MDB_EOF to writer thread, await unused buffer.
+ *
+ * @param[in] my control structure.
+ * @param[in] adjust (1 to hand off 1 buffer) | (MDB_EOF when ending).
+ */
static int ESECT
-mdb_env_cthr_toggle(mdb_copy *my, int st)
+mdb_env_cthr_toggle(mdb_copy *my, int adjust)
{
- int toggle = my->mc_toggle ^ 1;
pthread_mutex_lock(&my->mc_mutex);
- if (my->mc_status) {
- pthread_mutex_unlock(&my->mc_mutex);
- return my->mc_status;
- }
- while (my->mc_new == 1)
- pthread_cond_wait(&my->mc_cond, &my->mc_mutex);
- my->mc_new = st;
- my->mc_toggle = toggle;
+ my->mc_new += adjust;
pthread_cond_signal(&my->mc_cond);
+ while (my->mc_new & 2) /* both buffers in use */
+ pthread_cond_wait(&my->mc_cond, &my->mc_mutex);
pthread_mutex_unlock(&my->mc_mutex);
- return 0;
+
+ my->mc_toggle ^= (adjust & 1);
+ /* Both threads reset mc_wlen, to be safe from threading errors */
+ my->mc_wlen[my->mc_toggle] = 0;
+ return my->mc_error;
}
- /** Depth-first tree traversal for compacting copy. */
+ /** Depth-first tree traversal for compacting copy.
+ * @param[in] my control structure.
+ * @param[in,out] pg database root.
+ * @param[in] flags includes #F_DUPDATA if it is a sorted-duplicate sub-DB.
+ */
static int ESECT
mdb_env_cwalk(mdb_copy *my, pgno_t *pg, int flags)
{
}
memcpy(&pg, NODEDATA(ni), sizeof(pg));
+ memcpy(NODEDATA(ni), &my->mc_next_pgno, sizeof(pgno_t));
rc = mdb_page_get(&mc, pg, &omp, NULL);
if (rc)
goto done;
goto done;
toggle = my->mc_toggle;
}
- memcpy(NODEDATA(ni), &mo->mp_pgno, sizeof(pgno_t));
} else if (ni->mn_flags & F_SUBDATA) {
MDB_db db;
{
MDB_meta *mm;
MDB_page *mp;
- mdb_copy my;
+ mdb_copy my = {0};
MDB_txn *txn = NULL;
pthread_t thr;
- int rc;
+ pgno_t root, new_root;
+ int rc = MDB_SUCCESS;
#ifdef _WIN32
- my.mc_mutex = CreateMutex(NULL, FALSE, NULL);
- my.mc_cond = CreateEvent(NULL, FALSE, FALSE, NULL);
+ if (!(my.mc_mutex = CreateMutex(NULL, FALSE, NULL)) ||
+ !(my.mc_cond = CreateEvent(NULL, FALSE, FALSE, NULL))) {
+ rc = ErrCode();
+ goto done;
+ }
my.mc_wbuf[0] = _aligned_malloc(MDB_WBUF*2, env->me_os_psize);
- if (my.mc_wbuf[0] == NULL)
- return errno;
+ if (my.mc_wbuf[0] == NULL) {
+ /* _aligned_malloc() sets errno, but we use Windows error codes */
+ rc = ERROR_NOT_ENOUGH_MEMORY;
+ goto done;
+ }
#else
- pthread_mutex_init(&my.mc_mutex, NULL);
- pthread_cond_init(&my.mc_cond, NULL);
+ if ((rc = pthread_mutex_init(&my.mc_mutex, NULL)) != 0)
+ return rc;
+ if ((rc = pthread_cond_init(&my.mc_cond, NULL)) != 0)
+ goto done2;
#ifdef HAVE_MEMALIGN
my.mc_wbuf[0] = memalign(env->me_os_psize, MDB_WBUF*2);
- if (my.mc_wbuf[0] == NULL)
- return errno;
+ if (my.mc_wbuf[0] == NULL) {
+ rc = errno;
+ goto done;
+ }
#else
- rc = posix_memalign((void **)&my.mc_wbuf[0], env->me_os_psize, MDB_WBUF*2);
- if (rc)
- return rc;
+ {
+ void *p;
+ if ((rc = posix_memalign(&p, env->me_os_psize, MDB_WBUF*2)) != 0)
+ goto done;
+ my.mc_wbuf[0] = p;
+ }
#endif
#endif
memset(my.mc_wbuf[0], 0, MDB_WBUF*2);
my.mc_wbuf[1] = my.mc_wbuf[0] + MDB_WBUF;
- my.mc_wlen[0] = 0;
- my.mc_wlen[1] = 0;
- my.mc_olen[0] = 0;
- my.mc_olen[1] = 0;
my.mc_next_pgno = NUM_METAS;
- my.mc_status = 0;
- my.mc_new = 1;
- my.mc_toggle = 0;
my.mc_env = env;
my.mc_fd = fd;
- THREAD_CREATE(thr, mdb_env_copythr, &my);
+ rc = THREAD_CREATE(thr, mdb_env_copythr, &my);
+ if (rc)
+ goto done;
rc = mdb_txn_begin(env, NULL, MDB_RDONLY, &txn);
if (rc)
- return rc;
+ goto finish;
mp = (MDB_page *)my.mc_wbuf[0];
memset(mp, 0, NUM_METAS * env->me_psize);
*(MDB_meta *)METADATA(mp) = *mm;
mm = (MDB_meta *)METADATA(mp);
- /* Count the number of free pages, subtract from lastpg to find
- * number of active pages
- */
- {
+ /* Set metapage 1 with current main DB */
+ root = new_root = txn->mt_dbs[MAIN_DBI].md_root;
+ if (root != P_INVALID) {
+ /* Count free pages + freeDB pages. Subtract from last_pg
+ * to find the new last_pg, which also becomes the new root.
+ */
MDB_ID freecount = 0;
MDB_cursor mc;
MDB_val key, data;
mdb_cursor_init(&mc, txn, FREE_DBI, NULL);
while ((rc = mdb_cursor_get(&mc, &key, &data, MDB_NEXT)) == 0)
freecount += *(MDB_ID *)data.mv_data;
+ if (rc != MDB_NOTFOUND)
+ goto finish;
freecount += txn->mt_dbs[FREE_DBI].md_branch_pages +
txn->mt_dbs[FREE_DBI].md_leaf_pages +
txn->mt_dbs[FREE_DBI].md_overflow_pages;
- /* Set metapage 1 */
- mm->mm_last_pg = txn->mt_next_pgno - freecount - 1;
+ new_root = txn->mt_next_pgno - 1 - freecount;
+ mm->mm_last_pg = new_root;
mm->mm_dbs[MAIN_DBI] = txn->mt_dbs[MAIN_DBI];
- if (mm->mm_last_pg > NUM_METAS-1) {
- mm->mm_dbs[MAIN_DBI].md_root = mm->mm_last_pg;
- mm->mm_txnid = 1;
- } else {
- mm->mm_dbs[MAIN_DBI].md_root = P_INVALID;
- }
+ mm->mm_dbs[MAIN_DBI].md_root = new_root;
+ } else {
+ /* When the DB is empty, handle it specially to
+ * fix any breakage like page leaks from ITS#8174.
+ */
+ mm->mm_dbs[MAIN_DBI].md_flags = txn->mt_dbs[MAIN_DBI].md_flags;
}
+ if (root != P_INVALID || mm->mm_dbs[MAIN_DBI].md_flags) {
+ mm->mm_txnid = 1; /* use metapage 1 */
+ }
+
my.mc_wlen[0] = env->me_psize * NUM_METAS;
my.mc_txn = txn;
- pthread_mutex_lock(&my.mc_mutex);
- while(my.mc_new)
- pthread_cond_wait(&my.mc_cond, &my.mc_mutex);
- pthread_mutex_unlock(&my.mc_mutex);
- rc = mdb_env_cwalk(&my, &txn->mt_dbs[MAIN_DBI].md_root, 0);
- if (rc == MDB_SUCCESS && my.mc_wlen[my.mc_toggle])
- rc = mdb_env_cthr_toggle(&my, 1);
- mdb_env_cthr_toggle(&my, -1);
- pthread_mutex_lock(&my.mc_mutex);
- while(my.mc_new)
- pthread_cond_wait(&my.mc_cond, &my.mc_mutex);
- pthread_mutex_unlock(&my.mc_mutex);
- THREAD_FINISH(thr);
+ rc = mdb_env_cwalk(&my, &root, 0);
+ if (rc == MDB_SUCCESS && root != new_root) {
+ rc = MDB_INCOMPATIBLE; /* page leak or corrupt DB */
+ }
+finish:
+ if (rc)
+ my.mc_error = rc;
+ mdb_env_cthr_toggle(&my, 1 | MDB_EOF);
+ rc = THREAD_FINISH(thr);
mdb_txn_abort(txn);
+
+done:
#ifdef _WIN32
- CloseHandle(my.mc_cond);
- CloseHandle(my.mc_mutex);
- _aligned_free(my.mc_wbuf[0]);
+ if (my.mc_wbuf[0]) _aligned_free(my.mc_wbuf[0]);
+ if (my.mc_cond) CloseHandle(my.mc_cond);
+ if (my.mc_mutex) CloseHandle(my.mc_mutex);
#else
+ free(my.mc_wbuf[0]);
pthread_cond_destroy(&my.mc_cond);
+done2:
pthread_mutex_destroy(&my.mc_mutex);
- free(my.mc_wbuf[0]);
#endif
- return rc;
+ return rc ? rc : my.mc_error;
}
/** Copy environment as-is. */
int ESECT
mdb_env_copy2(MDB_env *env, const char *path, unsigned int flags)
{
- int rc, len;
- char *lpath;
+ int rc;
+ MDB_name fname;
HANDLE newfd = INVALID_HANDLE_VALUE;
-#ifdef _WIN32
- wchar_t *wpath;
-#endif
-
- if (env->me_flags & MDB_NOSUBDIR) {
- lpath = (char *)path;
- } else {
- len = strlen(path);
- len += sizeof(DATANAME);
- lpath = malloc(len);
- if (!lpath)
- return ENOMEM;
- sprintf(lpath, "%s" DATANAME, path);
- }
-
- /* The destination path must exist, but the destination file must not.
- * We don't want the OS to cache the writes, since the source data is
- * already in the OS cache.
- */
-#ifdef _WIN32
- 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
- if (newfd == INVALID_HANDLE_VALUE) {
- rc = ErrCode();
- goto leave;
- }
- if (env->me_psize >= env->me_os_psize) {
-#ifdef O_DIRECT
- /* Set O_DIRECT if the file system supports it */
- if ((rc = fcntl(newfd, F_GETFL)) != -1)
- (void) fcntl(newfd, F_SETFL, rc | O_DIRECT);
-#endif
-#ifdef F_NOCACHE /* __APPLE__ */
- rc = fcntl(newfd, F_NOCACHE, 1);
- if (rc) {
- rc = ErrCode();
- goto leave;
- }
-#endif
+ rc = mdb_fname_init(path, env->me_flags | MDB_NOLOCK, &fname);
+ if (rc == MDB_SUCCESS) {
+ rc = mdb_fopen(env, &fname, MDB_O_COPY, 0666, &newfd);
+ mdb_fname_destroy(fname);
}
-
- rc = mdb_env_copyfd2(env, newfd, flags);
-
-leave:
- if (!(env->me_flags & MDB_NOSUBDIR))
- free(lpath);
- if (newfd != INVALID_HANDLE_VALUE)
+ if (rc == MDB_SUCCESS) {
+ rc = mdb_env_copyfd2(env, newfd, flags);
if (close(newfd) < 0 && rc == MDB_SUCCESS)
rc = ErrCode();
-
+ }
return rc;
}
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))) {
- return rc;
+ } else {
+ if (rc != MDB_NOTFOUND || !(flags & MDB_CREATE))
+ return rc;
+ if (F_ISSET(txn->mt_flags, MDB_TXN_RDONLY))
+ return EACCES;
}
/* Done here so we cannot fail after creating a new DB */
memset(&dummy, 0, sizeof(dummy));
dummy.md_root = P_INVALID;
dummy.md_flags = flags & PERSISTENT_FLAGS;
- rc = mdb_cursor_put(&mc, &key, &data, F_SUBDATA);
+ WITH_CURSOR_TRACKING(mc,
+ rc = mdb_cursor_put(&mc, &key, &data, F_SUBDATA));
dbflag |= DB_DIRTY;
}
return env->me_txns ? mdb_reader_check0(env, 0, dead) : MDB_SUCCESS;
}
-/** As #mdb_reader_check(). rlocked = <caller locked the reader mutex>. */
+/** As #mdb_reader_check(). \b rlocked is set if caller locked #me_rmutex. */
static int ESECT
mdb_reader_check0(MDB_env *env, int rlocked, int *dead)
{
return 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;
+/** Convert \b src to new wchar_t[] string with room for \b xtra extra chars */
+static int ESECT
+utf8_to_utf16(const char *src, MDB_name *dst, int xtra)
+{
+ int rc, need = 0;
+ wchar_t *result = NULL;
+ for (;;) { /* malloc result, then fill it in */
+ need = MultiByteToWideChar(CP_UTF8, 0, src, -1, result, need);
+ if (!need) {
+ rc = ErrCode();
+ free(result);
+ return rc;
+ }
+ if (!result) {
+ result = malloc(sizeof(wchar_t) * (need + xtra));
+ if (!result)
+ return ENOMEM;
+ continue;
+ }
+ dst->mn_alloced = 1;
+ dst->mn_len = need - 1;
+ dst->mn_val = result;
+ return MDB_SUCCESS;
+ }
}
#endif /* defined(_WIN32) */
+/** @} */