X-Git-Url: http://git.tuebingen.mpg.de/?p=osl.git;a=blobdiff_plain;f=osl.c;h=9480325ea8ab3c81977bfd33c892e11f3b524eb9;hp=0c43801c4d7c27deecfa985d1788008822efe424;hb=HEAD;hpb=f20ee31e76615776f4b56d57a3a4fc1919de2914

diff --git a/osl.c b/osl.c
index 0c43801..9480325 100644
--- a/osl.c
+++ b/osl.c
@@ -1,5 +1,5 @@
 /*
- * Copyright (C) 2007-2008 Andre Noll <maan@systemlinux.org>
+ * Copyright (C) 2007-2009 Andre Noll <maan@tuebingen.mpg.de>
  *
  * Licensed under the GPL v2. For licencing details see COPYING.
  */
@@ -8,90 +8,46 @@
 #include <dirent.h> /* readdir() */
 #include <assert.h>
 
-
 #include "log.h"
 #include "osl.h"
-#include "error.h"
-#include "fd.h"
-#include "list.h"
+#include "util.h"
 #include "osl_core.h"
 
-/**
- * Allocate a sufficiently large string and print into it.
- *
- * \param fmt A usual format string.
- *
- * Produce output according to \p fmt. No artificial bound on the length of the
- * resulting string is imposed.
- *
- * \return This function either returns a pointer to a string that must be
- * freed by the caller or \p NULL if memory allocation failed.
+/*
+ * Taken from Drepper: How to write shared libraries, Appendix B.
  *
- * \sa printf(3).
+ * The main reason for this rather fancy implementation of strerror() is to
+ * avoid having an array of pointers. This is desirable because initialized
+ * pointer variables increase the startup time of the library due to the
+ * processing of relocations.
  */
-static __must_check __printf_1_2 __malloc char *make_message(const char *fmt, ...)
-{
-	int n;
-	size_t size = 100;
-	char *p = malloc(size);
-
-	if (!p)
-		return NULL;
-	while (1) {
-		char *q;
-		va_list ap;
-		/* Try to print in the allocated space. */
-		va_start(ap, fmt);
-		n = vsnprintf(p, size, fmt, ap);
-		va_end(ap);
-		/* If that worked, return the string. */
-		if (n > -1 && n < size)
-			break;
-		/* Else try again with more space. */
-		if (n > -1) /* glibc 2.1 */
-			size = n + 1; /* precisely what is needed */
-		else /* glibc 2.0 */
-			size *= 2; /* twice the old size */
-		q = realloc(p, size);
-		if (!q) {
-			free(p);
-			return NULL;
-		}
-		p = q;
-	}
-	return p;
-}
-
-/* Taken from Drepper: How to write shared libraries, Appendix B. */
 #include <stddef.h>
 #define MSGSTRFIELD(line) MSGSTRFIELD1(line)
 #define MSGSTRFIELD1(line) str##line
 static const union msgstr_t {
 	struct {
-#define _S(n, s) char MSGSTRFIELD(__LINE__)[sizeof(s)];
+#define OSL_ERROR(n, s) char MSGSTRFIELD(__LINE__)[sizeof(s)];
 #include "errtab.h"
-#undef _S
+#undef OSL_ERROR
 	};
 	char str[0];
 } msgstr = { {
-#define _S(n, s) s,
+#define OSL_ERROR(n, s) s,
 #include "errtab.h"
-#undef _S
+#undef OSL_ERROR
 } };
 static const unsigned int errmsgidx[] = {
-#define _S(n, s) [n] = offsetof(union msgstr_t, MSGSTRFIELD(__LINE__)),
+#define OSL_ERROR(n, s) [n] = offsetof(union msgstr_t, MSGSTRFIELD(__LINE__)),
 #include "errtab.h"
-#undef _S
+#undef OSL_ERROR
 };
 
 __export const char *osl_strerror(int num)
 {
-	if (IS_SYSTEM_ERROR(num))
-		return strerror((num) & ((1 << SYSTEM_ERROR_BIT) - 1));
 	return msgstr.str + errmsgidx[num];
 }
 
-static int loglevel;
+int loglevel = 0;
 
 static void __attribute ((constructor)) init_loglevel(void)
 {
@@ -112,20 +68,18 @@ static void __attribute ((constructor)) init_loglevel(void)
 __printf_2_3 void __log(int ll, const char* fmt,...)
 {
 	va_list argp;
-	FILE *outfd;
 	struct tm *tm;
 	time_t t1;
 	char str[255] = "";
 
 	if (ll < loglevel)
 		return;
-	outfd = stderr;
 	time(&t1);
 	tm = localtime(&t1);
 	strftime(str, sizeof(str), "%b %d %H:%M:%S", tm);
-	fprintf(outfd, "%s ", str);
+	fprintf(stderr, "%s ", str);
 	va_start(argp, fmt);
-	vfprintf(outfd, fmt, argp);
+	vfprintf(stderr, fmt, argp);
 	va_end(argp);
 }
 
@@ -154,90 +108,8 @@ static int __lseek(int fd, off_t *offset, int whence)
 	return 1;
 }
 
-/**
- * Wrapper for the write system call.
- *
- * \param fd The file descriptor to write to.
- * \param buf The buffer to write.
- * \param size The length of \a buf in bytes.
- *
- * This function writes out the given buffer and retries if an interrupt
- * occurred during the write.
- *
- * \return On success, the number of bytes written is returned, otherwise, the
- * function returns \p -E_OSL_WRITE.
- *
- * \sa write(2).
- */
-static ssize_t __write(int fd, const void *buf, size_t size)
-{
-	ssize_t ret;
-
-	for (;;) {
-		ret = write(fd, buf, size);
-		if ((ret < 0) && (errno == EAGAIN || errno == EINTR))
-			continue;
-		return ret >= 0? ret : -E_OSL_WRITE;
-	}
-}
-
-/**
- * Write the whole buffer to a file descriptor.
- *
- * \param fd The file descriptor to write to.
- * \param buf The buffer to write.
- * \param size The length of \a buf in bytes.
- *
- * This function writes the given buffer and continues on short writes and
- * when interrupted by a signal.
- *
- * \return Positive on success, negative on errors. Possible errors: any
- * errors returned by para_write().
- *
- * \sa para_write().
- */
-static ssize_t write_all(int fd, const void *buf, size_t size)
-{
-//	DEBUG_LOG("writing %zu bytes\n", size);
-	const char *b = buf;
-	while (size) {
-		ssize_t ret = __write(fd, b, size);
-//		DEBUG_LOG("ret: %zd\n", ret);
-		if (ret < 0)
-			return ret;
-		b += ret;
-		size -= ret;
-	}
-	return 1;
-}
-/**
- * Open a file, write the given buffer and close the file.
- *
- * \param filename Full path to the file to open.
- * \param buf The buffer to write to the file.
- * \param size The size of \a buf.
- *
- * \return Standard.
- */
-static int write_file(const char *filename, const void *buf, size_t size)
-{
-	int ret, fd;
-
-	ret = osl_open(filename, O_WRONLY | O_CREAT | O_EXCL, 0644);
-	if (ret < 0)
-		return ret;
-	fd = ret;
-	ret = write_all(fd, buf, size);
-	if (ret < 0)
-		goto out;
-	ret = 1;
-out:
-	close(fd);
-	return ret;
-}
-
-static int append_file(const char *filename, char *header, size_t header_size,
-	char *data, size_t data_size, uint32_t *new_pos)
+static int append_file(const char *filename, char *data, size_t data_size,
+		uint32_t *new_pos)
 {
 	int ret, fd;
 
@@ -246,12 +118,7 @@ static int append_file(const char *filename, char *header, size_t header_size,
 	if (ret < 0)
 		return ret;
 	fd = ret;
-	if (header && header_size) {
-		ret = write_all(fd, header, header_size);
-		if (ret < 0)
-			goto out;
-	}
-	ret = write_all(fd, data, data_size);
+	ret = write_all(fd, data, &data_size);
 	if (ret < 0)
 		goto out;
 	if (new_pos) {
@@ -283,11 +150,6 @@ static int verify_name(const char *name)
 	return 1;
 }
 
-int osl_hash_compare(const struct osl_object *obj1, const struct osl_object *obj2)
-{
-	return hash_compare((HASH_TYPE *)obj1->data, (HASH_TYPE *)obj2->data);
-}
-
 static char *disk_storage_dirname(const struct osl_table *t, unsigned col_num,
 		const char *ds_name)
 {
@@ -306,7 +168,7 @@ static char *disk_storage_name_of_object(const struct osl_table *t,
 	const struct osl_object *obj)
 {
 	HASH_TYPE hash[HASH_SIZE];
-	hash_object(obj, hash);
+	hash_object(t, obj, hash);
 	return disk_storage_name_of_hash(t, hash);
 }
 
@@ -321,12 +183,13 @@ static int disk_storage_name_of_row(const struct osl_table *t,
 	*name = disk_storage_name_of_object(t, &obj);
 	if (*name)
 		return 1;
-	return -ERRNO_TO_ERROR(ENOMEM);
+	return -E_OSL_NOMEM;
 }
 
-static void column_name_hash(const char *col_name, HASH_TYPE *hash)
+static void column_name_hash(const struct osl_table *t, const char *col_name,
+		HASH_TYPE *hash)
 {
-	hash_function(col_name, strlen(col_name), hash);
+	hash_function(t->version, col_name, strlen(col_name), hash);
 }
 
 static int init_column_descriptions(struct osl_table *t)
@@ -334,7 +197,6 @@ static int init_column_descriptions(struct osl_table *t)
 	int i, j, ret;
 	const struct osl_column_description *cd;
 
-	ret = -E_OSL_BAD_TABLE_DESC;
 	ret = verify_name(t->desc->name);
 	if (ret < 0)
 		goto err;
@@ -344,7 +206,6 @@ static int init_column_descriptions(struct osl_table *t)
 	/* the size of the index header without column descriptions */
 	t->index_header_size = IDX_COLUMN_DESCRIPTIONS;
 	FOR_EACH_COLUMN(i, t->desc, cd) {
-		struct osl_column *col = t->columns + i;
 		if (cd->storage_flags & OSL_RBTREE) {
 			if (!cd->compare_function)
 				return -E_OSL_NO_COMPARE_FUNC;
@@ -358,7 +219,6 @@ static int init_column_descriptions(struct osl_table *t)
 		if (ret < 0)
 			goto err;
 		t->index_header_size += index_column_description_size(cd->name);
-		column_name_hash(cd->name, col->name_hash);
 		ret = -E_OSL_DUPLICATE_COL_NAME;
 		for (j = i + 1; j < t->desc->num_columns; j++) {
 			const char *name2 = get_column_description(t->desc,
@@ -391,7 +251,7 @@ int init_table_structure(const struct osl_table_description *desc,
 {
 	const struct osl_column_description *cd;
 	struct osl_table *t = calloc(1, sizeof(*t));
-	int i, ret = -ERRNO_TO_ERROR(ENOMEM), have_disk_storage_name_column = 0;
+	int i, ret = -E_OSL_NOMEM, have_disk_storage_name_column = 0;
 
 	if (!t)
 		return ret;
@@ -406,7 +266,7 @@ int init_table_structure(const struct osl_table_description *desc,
 	ret = -E_OSL_NO_COLUMNS;
 	if (!desc->num_columns)
 		goto err;
-	ret = -ERRNO_TO_ERROR(ENOMEM);
+	ret = -E_OSL_NOMEM;
 	t->columns = calloc(desc->num_columns, sizeof(struct osl_column));
 	if (!t->columns)
 		goto err;
@@ -423,6 +283,9 @@ int init_table_structure(const struct osl_table_description *desc,
 		ret = -E_OSL_BAD_STORAGE_FLAGS;
 		if (st == OSL_DISK_STORAGE && sf & OSL_RBTREE)
 			goto err;
+		if ((sf & OSL_RBTREE) && !(sf & OSL_UNIQUE))
+			WARNING_LOG("invalid storage flags for column %s: "
+				"OSL_RBTREE && !OSL_UNIQUE\n", cd->name);
 		ret = -E_OSL_BAD_STORAGE_SIZE;
 		if (sf & OSL_FIXED_SIZE && !cd->data_size)
 			goto err;
@@ -475,17 +338,17 @@ err:
  * \param map The memory mapping of the index file.
  * \param desc The values found in the index header are returned here.
  *
- * Read the index header, check for the paraslash magic string and the table version number.
- * Read all information stored in the index header into \a desc.
+ * Read the index header, check for the osl magic string and the table version
+ * number.  Read all information stored in the index header into \a desc.
  *
- * \return Standard.
+ * \return The on-disk table version number on success, negative on errors.
  *
  * \sa struct osl_table_description, osl_create_table.
  */
 int read_table_desc(struct osl_object *map, struct osl_table_description *desc)
 {
 	char *buf = map->data;
-	uint8_t version;
+	uint8_t table_version;
 	uint16_t header_size;
 	int ret, i;
 	unsigned offset;
@@ -493,10 +356,13 @@ int read_table_desc(struct osl_object *map, struct osl_table_description *desc)
 
 	if (map->size < MIN_INDEX_HEADER_SIZE(1))
 		return -E_OSL_SHORT_TABLE;
-	if (strncmp(buf + IDX_PARA_MAGIC, PARA_MAGIC, strlen(PARA_MAGIC)))
+	if (strncmp(buf + IDX_OSL_MAGIC, OSL_MAGIC, strlen(OSL_MAGIC)))
 		return -E_OSL_NO_MAGIC;
-	version = read_u8(buf + IDX_VERSION);
-	if (version < MIN_TABLE_VERSION || version > MAX_TABLE_VERSION)
+	table_version = read_u8(buf + IDX_VERSION);
+	INFO_LOG("osl versions (table/min/current): %u/%u/%u\n",
+		table_version, MIN_TABLE_VERSION, CURRENT_TABLE_VERSION);
+	if (table_version < MIN_TABLE_VERSION /* table too old */
+		|| table_version > CURRENT_TABLE_VERSION) /* libosl too old */
 		return -E_OSL_VERSION_MISMATCH;
 	desc->flags = read_u8(buf + IDX_TABLE_FLAGS);
 	desc->num_columns = read_u16(buf + IDX_NUM_COLUMNS);
@@ -509,7 +375,7 @@ int read_table_desc(struct osl_object *map, struct osl_table_description *desc)
 	desc->column_descriptions = calloc(desc->num_columns,
 		sizeof(struct osl_column_description));
 	if (!desc->column_descriptions)
-		return -ERRNO_TO_ERROR(ENOMEM);
+		return -E_OSL_NOMEM;
 	offset = IDX_COLUMN_DESCRIPTIONS;
 	FOR_EACH_COLUMN(i, desc, cd) {
 		char *null_byte;
@@ -529,7 +395,7 @@ int read_table_desc(struct osl_object *map, struct osl_table_description *desc)
 		ret = -E_OSL_INDEX_CORRUPTION;
 		if (!null_byte)
 			goto err;
-		ret = -ERRNO_TO_ERROR(ENOMEM);
+		ret = -E_OSL_NOMEM;
 		cd->name = strdup(buf + offset + IDX_CD_NAME);
 		if (!cd->name)
 			goto err;
@@ -541,7 +407,7 @@ int read_table_desc(struct osl_object *map, struct osl_table_description *desc)
 			offset, header_size);
 		goto err;
 	}
-	return 1;
+	return table_version;
 err:
 	FOR_EACH_COLUMN(i, desc, cd)
 		free(cd->name);
@@ -562,6 +428,7 @@ static int compare_table_descriptions(struct osl_table *t)
 	ret = read_table_desc(&t->index_map, &desc);
 	if (ret < 0)
 		return ret;
+	t->version = ret;
 	ret = -E_OSL_BAD_TABLE_FLAGS;
 	if (desc.flags != t->desc->flags)
 		goto out;
@@ -572,7 +439,7 @@ static int compare_table_descriptions(struct osl_table *t)
 		struct osl_column_description *cd;
 		unsigned diff = t->desc->num_columns - desc.num_columns;
 		INFO_LOG("extending table by %u volatile columns\n", diff);
-		ret = -ERRNO_TO_ERROR(ENOMEM);
+		ret = -E_OSL_NOMEM;
 		desc.column_descriptions = realloc(desc.column_descriptions,
 			t->desc->num_columns * sizeof(struct osl_column_description));
 		if (!desc.column_descriptions)
@@ -627,11 +494,11 @@ static int create_table_index(struct osl_table *t)
 		t->desc->name);
 	buf = calloc(1, size);
 	if (!buf)
-		return -ERRNO_TO_ERROR(ENOMEM);
-	sprintf(buf + IDX_PARA_MAGIC, "%s", PARA_MAGIC);
+		return -E_OSL_NOMEM;
+	sprintf(buf + IDX_OSL_MAGIC, "%s", OSL_MAGIC);
 	write_u8(buf + IDX_TABLE_FLAGS, t->desc->flags);
 	write_u8(buf + IDX_DIRTY_FLAG, 0);
-	write_u8(buf + IDX_VERSION, CURRENT_TABLE_VERSION);
+	write_u8(buf + IDX_VERSION, t->version);
 	write_u16(buf + IDX_NUM_COLUMNS, t->num_mapped_columns + t->num_disk_storage_columns);
 	write_u16(buf + IDX_HEADER_SIZE, t->index_header_size);
 	offset = IDX_COLUMN_DESCRIPTIONS;
@@ -649,12 +516,12 @@ static int create_table_index(struct osl_table *t)
 		strcpy(buf + offset + IDX_CD_NAME, cd->name);
 		offset += index_column_description_size(cd->name);
 	}
-	assert(offset = size);
+	assert(offset == size);
 	filename = index_filename(t->desc);
 	if (filename)
 		ret = write_file(filename, buf, size);
 	else
-		ret = -ERRNO_TO_ERROR(ENOMEM);
+		ret = -E_OSL_NOMEM;
 	free(buf);
 	free(filename);
 	return ret;
@@ -669,24 +536,26 @@ __export int osl_create_table(const struct osl_table_description *desc)
 
 	if (ret < 0)
 		return ret;
-	INFO_LOG("creating %s\n", desc->name);
+	t->version = CURRENT_TABLE_VERSION;
+	INFO_LOG("creating version %u table %s\n", t->version, desc->name);
 	FOR_EACH_COLUMN(i, t->desc, cd) {
 		if (cd->storage_type == OSL_NO_STORAGE)
 			continue;
 		if (!table_dir) {
 			ret = osl_mkdir(desc->dir, 0777);
-			if (ret < 0 && !is_errno(-ret, EEXIST))
+			if (ret < 0 && ret != -E_OSL_DIR_EXISTS)
 				goto out;
 			table_dir = make_message("%s/%s", desc->dir,
 				desc->name);
-			ret = -ERRNO_TO_ERROR(ENOMEM);
+			ret = -E_OSL_NOMEM;
 			if (!table_dir)
 				goto out;
 			ret = osl_mkdir(table_dir, 0777);
 			if (ret < 0)
 				goto out;
 		}
-		ret = -ERRNO_TO_ERROR(ENOMEM);
+		column_name_hash(t, cd->name, t->columns[i].name_hash);
+		ret = -E_OSL_NOMEM;
 		filename = column_filename(t, i);
 		if (!filename)
 			goto out;
@@ -744,7 +613,7 @@ static void unmap_column(struct osl_table *t, unsigned col_num)
 	int ret;
 	if (!map.data)
 		return;
-	ret = para_munmap(map.data, map.size);
+	ret = osl_munmap(map.data, map.size);
 	assert(ret > 0);
 	map.data = NULL;
 }
@@ -757,7 +626,7 @@ static void unmap_column(struct osl_table *t, unsigned col_num)
  *
  * \return Positive on success, negative on errors.
  *
- * \sa map_table(), enum osl_close_flags, para_munmap().
+ * \sa map_table(), enum osl_close_flags, osl_munmap().
  */
 int unmap_table(struct osl_table *t, enum osl_close_flags flags)
 {
@@ -772,7 +641,7 @@ int unmap_table(struct osl_table *t, enum osl_close_flags flags)
 		return -E_OSL_NOT_MAPPED;
 	if (flags & OSL_MARK_CLEAN)
 		mark_table_clean(t);
-	ret = para_munmap(t->index_map.data, t->index_map.size);
+	ret = osl_munmap(t->index_map.data, t->index_map.size);
 	if (ret < 0)
 		return ret;
 	t->index_map.data = NULL;
@@ -787,11 +656,12 @@ static int map_column(struct osl_table *t, unsigned col_num)
 {
 	struct stat statbuf;
 	char *filename = column_filename(t, col_num);
-	int ret = -E_OSL_STAT;
+	int ret;
 
 	if (!filename)
-		return -ERRNO_TO_ERROR(ENOMEM);
-	if (stat(filename, &statbuf) < 0) {
+		return -E_OSL_NOMEM;
+	ret = osl_stat(filename, &statbuf);
+	if (ret < 0) {
 		free(filename);
 		return ret;
 	}
@@ -830,7 +700,7 @@ int map_table(struct osl_table *t, enum map_table_flags flags)
 		return -E_OSL_ALREADY_MAPPED;
 	filename = index_filename(t->desc);
 	if (!filename)
-		return -ERRNO_TO_ERROR(ENOMEM);
+		return -E_OSL_NOMEM;
 	INFO_LOG("mapping table '%s' (index: %s)\n", t->desc->name, filename);
 	ret = mmap_full_file(filename, flags & MAP_TBL_FL_MAP_RDONLY?
 		O_RDONLY : O_RDWR, &t->index_map.data, &t->index_map.size, NULL);
@@ -851,22 +721,25 @@ int map_table(struct osl_table *t, enum map_table_flags flags)
 	}
 	mark_table_dirty(t);
 	num_rows = table_num_rows(t);
-	if (!num_rows)
-		return num_rows;
 	/* map data files */
-	FOR_EACH_MAPPED_COLUMN(i, t, cd) {
-		ret = map_column(t, i);
-		if (ret < 0)
-			goto err;
+	FOR_EACH_COLUMN(i, t->desc, cd) {
+		if (cd->storage_type == OSL_NO_STORAGE)
+			continue;
+		column_name_hash(t, cd->name, t->columns[i].name_hash);
+		if (num_rows > 0 && cd->storage_type == OSL_MAPPED_STORAGE) {
+			ret = map_column(t, i);
+			if (ret < 0)
+				goto err;
+		}
 	}
 	return num_rows;
 err:	/* unmap what is already mapped */
 	for (i--; i >= 0; i--) {
 		struct osl_object map = t->columns[i].data_map;
-		para_munmap(map.data, map.size);
+		osl_munmap(map.data, map.size);
 		map.data = NULL;
 	}
-	para_munmap(t->index_map.data, t->index_map.size);
+	osl_munmap(t->index_map.data, t->index_map.size);
 	t->index_map.data = NULL;
 	return ret;
 }
@@ -891,7 +764,6 @@ int get_mapped_object(const struct osl_table *t, unsigned col_num,
 {
 	struct osl_column *col = &t->columns[col_num];
 	uint32_t offset;
-	char *header;
 	char *cell_index;
 	int ret;
 
@@ -901,17 +773,15 @@ int get_mapped_object(const struct osl_table *t, unsigned col_num,
 	if (ret < 0)
 		return ret;
 	offset = read_u32(cell_index);
-	obj->size = read_u32(cell_index + 4) - 1;
-	header = col->data_map.data + offset;
-	obj->data = header + 1;
-	if (read_u8(header) == 0xff) {
-		ERROR_LOG("col %u, size %zu, offset %u\n", col_num,
-			obj->size, offset);
-		return -E_OSL_INVALID_OBJECT;
-	}
+	obj->size = read_u32(cell_index + 4);
+	obj->data = col->data_map.data + offset;
 	return 1;
 }
 
+/*
+ * It's OK to call this with result = rb_node = NULL.  If result is not NULL,
+ * and rb key was not found, result points to the parent node.
+ */
 static int search_rbtree(const struct osl_object *obj,
 		const struct osl_table *t, unsigned col_num,
 		struct rb_node **result, struct rb_node ***rb_link)
@@ -1007,7 +877,7 @@ static int add_row_to_rbtrees(struct osl_table *t, uint32_t row_num,
 	const struct osl_column_description *cd;
 
 	if (!row)
-		return -ERRNO_TO_ERROR(ENOMEM);
+		return -E_OSL_NOMEM;
 	row->num = row_num;
 	row->volatile_objects = volatile_objs;
 	FOR_EACH_RBTREE_COLUMN(i, t, cd) {
@@ -1127,12 +997,13 @@ __export int osl_close_table(struct osl_table *t, enum osl_close_flags flags)
 int row_is_invalid(struct osl_table *t, uint32_t row_num)
 {
 	char *row_index;
-	int i, ret = get_row_index(t, row_num, &row_index);
+	int ret = get_row_index(t, row_num, &row_index);
+	unsigned n;
 
 	if (ret < 0)
 		return ret;
-	for (i = 0; i < t->row_index_size; i++) {
-		if ((unsigned char)row_index[i] != 0xff)
+	for (n = 0; n < t->row_index_size; n++) {
+		if ((unsigned char)row_index[n] != 0xff)
 			return 0;
 	}
 	INFO_LOG("row %d is invalid\n", row_num);
@@ -1171,23 +1042,32 @@ int mark_row_invalid(struct osl_table *t, uint32_t row_num)
  */
 int init_rbtrees(struct osl_table *t)
 {
-	int i, ret;
+	int ret;
+	unsigned n;
 	const struct osl_column_description *cd;
 
 	/* create rbtrees */
-	FOR_EACH_RBTREE_COLUMN(i, t, cd)
-		t->columns[i].rbtree = RB_ROOT;
+	FOR_EACH_RBTREE_COLUMN(n, t, cd)
+		t->columns[n].rbtree = RB_ROOT;
 	/* add valid rows to rbtrees */
 	t->num_invalid_rows = 0;
-	for (i = 0; i < t->num_rows; i++) {
-		ret = row_is_invalid(t, i);
+	for (n = 0; n < t->num_rows; n++) {
+		struct osl_object *volatile_objs;
+		ret = row_is_invalid(t, n);
 		if (ret < 0)
 			return ret;
 		if (ret) {
 			t->num_invalid_rows++;
 			continue;
 		}
-		ret = add_row_to_rbtrees(t, i, NULL, NULL);
+		if (t->num_volatile_columns > 0) {
+			volatile_objs = calloc(t->num_volatile_columns,
+				sizeof(struct osl_object));
+			if (!volatile_objs)
+				return -E_OSL_NOMEM;
+		} else
+			volatile_objs = NULL;
+		ret = add_row_to_rbtrees(t, n, volatile_objs, NULL);
 		if (ret < 0)
 			return ret;
 	}
@@ -1197,32 +1077,13 @@ int init_rbtrees(struct osl_table *t)
 __export int osl_open_table(const struct osl_table_description *table_desc,
 		struct osl_table **result)
 {
-	int i, ret;
+	int ret;
 	struct osl_table *t;
-	const struct osl_column_description *cd;
 
 	NOTICE_LOG("opening table %s\n", table_desc->name);
 	ret = init_table_structure(table_desc, &t);
 	if (ret < 0)
 		return ret;
-	FOR_EACH_DISK_STORAGE_COLUMN(i, t, cd) {
-		struct stat statbuf;
-		char *dirname = column_filename(t, i);
-
-		ret = -ERRNO_TO_ERROR(ENOMEM);
-		if (!dirname)
-			goto err;
-		/* check if directory exists */
-		ret = stat(dirname, &statbuf);
-		free(dirname);
-		if (ret < 0) {
-			ret = -ERRNO_TO_ERROR(errno);
-			goto err;
-		}
-		ret = -ERRNO_TO_ERROR(ENOTDIR);
-		if (!S_ISDIR(statbuf.st_mode))
-			goto err;
-	}
 	ret = map_table(t, MAP_TBL_FL_VERIFY_INDEX);
 	if (ret < 0)
 		goto err;
@@ -1251,10 +1112,10 @@ static int create_disk_storage_object_dir(const struct osl_table *t,
 		return 1;
 	dirname = disk_storage_dirname(t, col_num, ds_name);
 	if (!dirname)
-		return -ERRNO_TO_ERROR(ENOMEM);
+		return -E_OSL_NOMEM;
 	ret = osl_mkdir(dirname, 0777);
 	free(dirname);
-	if (ret < 0 && !is_errno(-ret, EEXIST))
+	if (ret < 0 && ret != -E_OSL_DIR_EXISTS)
 		return ret;
 	return 1;
 }
@@ -1270,7 +1131,7 @@ static int write_disk_storage_file(const struct osl_table *t, unsigned col_num,
 		return ret;
 	filename = disk_storage_path(t, col_num, ds_name);
 	if (!filename)
-		return -ERRNO_TO_ERROR(ENOMEM);
+		return -E_OSL_NOMEM;
 	ret = write_file(filename, obj->data, obj->size);
 	free(filename);
 	return ret;
@@ -1281,12 +1142,10 @@ static int append_map_file(const struct osl_table *t, unsigned col_num,
 {
 	char *filename = column_filename(t, col_num);
 	int ret;
-	char header = 0; /* zero means valid object */
 
 	if (!filename)
-		return -ERRNO_TO_ERROR(ENOMEM);
-	ret = append_file(filename, &header, 1, obj->data, obj->size,
-		new_size);
+		return -E_OSL_NOMEM;
+	ret = append_file(filename, obj->data, obj->size, new_size);
 	free(filename);
 	return ret;
 }
@@ -1300,44 +1159,12 @@ static int append_row_index(const struct osl_table *t, char *row_index)
 		return 1;
 	filename = index_filename(t->desc);
 	if (!filename)
-		return -ERRNO_TO_ERROR(ENOMEM);
-	ret = append_file(filename, NULL, 0, row_index,
-		t->row_index_size, NULL);
+		return -E_OSL_NOMEM;
+	ret = append_file(filename, row_index, t->row_index_size, NULL);
 	free(filename);
 	return ret;
 }
 
-/**
- * A wrapper for truncate(2)
- *
- * \param path Name of the regular file to truncate
- * \param size Number of bytes to \b shave \b off
- *
- * Truncate the regular file named by \a path by \a size bytes.
- *
- * \return Standard.
- *
- * \sa truncate(2)
- */
-int para_truncate(const char *path, off_t size)
-{
-	int ret;
-	struct stat statbuf;
-
-	ret = -E_OSL_STAT;
-	if (stat(path, &statbuf) < 0)
-		goto out;
-	ret = -E_OSL_BAD_SIZE;
-	if (statbuf.st_size < size)
-		goto out;
-	ret = -E_OSL_TRUNC;
-	if (truncate(path, statbuf.st_size - size) < 0)
-		goto out;
-	ret = 1;
-out:
-	return ret;
-}
-
 static int truncate_mapped_file(const struct osl_table *t, unsigned col_num,
 		off_t size)
 {
@@ -1345,8 +1172,8 @@ static int truncate_mapped_file(const struct osl_table *t, unsigned col_num,
 	char *filename = column_filename(t, col_num);
 
 	if (!filename)
-		return -ERRNO_TO_ERROR(ENOMEM);
-	ret = para_truncate(filename, size);
+		return -E_OSL_NOMEM;
+	ret = truncate_file(filename, size);
 	free(filename);
 	return ret;
 }
@@ -1355,20 +1182,20 @@ static int delete_disk_storage_file(const struct osl_table *t, unsigned col_num,
 		const char *ds_name)
 {
 	char *dirname, *filename = disk_storage_path(t, col_num, ds_name);
-	int ret, err;
+	int ret = 1;
 
 	if (!filename)
-		return -ERRNO_TO_ERROR(ENOMEM);
-	ret = unlink(filename);
-	err = errno;
+		return -E_OSL_NOMEM;
+	if (unlink(filename) < 0)
+		ret = errno == ENOENT? -E_OSL_NOENT : -E_OSL_UNLINK;
 	free(filename);
 	if (ret < 0)
-		return -ERRNO_TO_ERROR(err);
+		return ret;
 	if (!(t->desc->flags & OSL_LARGE_TABLE))
 		return 1;
 	dirname = disk_storage_dirname(t, col_num, ds_name);
 	if (!dirname)
-		return -ERRNO_TO_ERROR(ENOMEM);
+		return -E_OSL_NOMEM;
 	rmdir(dirname);
 	free(dirname);
 	return 1;
@@ -1388,18 +1215,18 @@ __export int osl_add_and_get_row(struct osl_table *t, struct osl_object *objects
 		return -E_OSL_BAD_TABLE;
 	rb_parents = malloc(t->num_rbtrees * sizeof(struct rn_node*));
 	if (!rb_parents)
-		return -ERRNO_TO_ERROR(ENOMEM);
+		return -E_OSL_NOMEM;
 	rb_links = malloc(t->num_rbtrees * sizeof(struct rn_node**));
 	if (!rb_links) {
 		free(rb_parents);
-		return -ERRNO_TO_ERROR(ENOMEM);
+		return -E_OSL_NOMEM;
 	}
 	if (t->num_mapped_columns) {
 		new_row_index = malloc(t->row_index_size);
 		if (!new_row_index) {
 			free(rb_links);
 			free(rb_parents);
-			return -ERRNO_TO_ERROR(ENOMEM);
+			return -E_OSL_NOMEM;
 		}
 	}
 	/* pass 1: sanity checks */
@@ -1435,7 +1262,7 @@ __export int osl_add_and_get_row(struct osl_table *t, struct osl_object *objects
 	if (t->num_disk_storage_columns) {
 		ds_name = disk_storage_name_of_object(t,
 			&objects[t->disk_storage_name_column]);
-		ret = -ERRNO_TO_ERROR(ENOMEM);
+		ret = -E_OSL_NOMEM;
 		if (!ds_name)
 			goto out;
 	}
@@ -1472,13 +1299,13 @@ __export int osl_add_and_get_row(struct osl_table *t, struct osl_object *objects
 	if (ret < 0) { /* truncate index and rollback changes */
 		char *filename = index_filename(t->desc);
 		if (filename)
-			para_truncate(filename, t->row_index_size);
+			truncate_file(filename, t->row_index_size);
 		free(filename);
 		goto rollback;
 	}
 	/* pass 3: add entry to rbtrees */
 	if (t->num_volatile_columns) {
-		ret = -ERRNO_TO_ERROR(ENOMEM);
+		ret = -E_OSL_NOMEM;
 		volatile_objs = calloc(t->num_volatile_columns,
 			sizeof(struct osl_object));
 		if (!volatile_objs)
@@ -1539,21 +1366,6 @@ __export int osl_get_object(const struct osl_table *t, const struct osl_row *r,
 	return 1;
 }
 
-static int mark_mapped_object_invalid(const struct osl_table *t,
-		uint32_t row_num, unsigned col_num)
-{
-	struct osl_object obj;
-	char *p;
-	int ret = get_mapped_object(t, col_num, row_num, &obj);
-
-	if (ret < 0)
-		return ret;
-	p = obj.data;
-	p--;
-	*p = 0xff;
-	return 1;
-}
-
 __export int osl_del_row(struct osl_table *t, struct osl_row *row)
 {
 	struct osl_row *r = row;
@@ -1577,10 +1389,8 @@ __export int osl_del_row(struct osl_table *t, struct osl_row *row)
 		struct osl_column *col = t->columns + i;
 		enum osl_storage_type st = cd->storage_type;
 		remove_rb_node(t, i, r);
-		if (st == OSL_MAPPED_STORAGE) {
-			mark_mapped_object_invalid(t, r->num, i);
+		if (st == OSL_MAPPED_STORAGE)
 			continue;
-		}
 		if (st == OSL_NO_STORAGE && !(cd->storage_flags & OSL_DONT_FREE))
 			free(r->volatile_objects[col->volatile_num].data);
 	}
@@ -1699,7 +1509,7 @@ static int rename_disk_storage_objects(struct osl_table *t,
 		return 1; /* object did not change */
 	old_ds_name = disk_storage_name_of_object(t, old_obj);
 	new_ds_name = disk_storage_name_of_object(t, new_obj);
-	ret = -ERRNO_TO_ERROR(ENOMEM);
+	ret = -E_OSL_NOMEM;
 	if (!old_ds_name || ! new_ds_name)
 		goto out;
 
@@ -1711,7 +1521,7 @@ static int rename_disk_storage_objects(struct osl_table *t,
 		old_filename = disk_storage_path(t, i, old_ds_name);
 		new_filename = disk_storage_path(t, i, new_ds_name);
 		if (!old_filename || !new_filename)
-			ret = -ERRNO_TO_ERROR(ENOMEM);
+			ret = -E_OSL_NOMEM;
 		else
 			ret = osl_rename(old_filename, new_filename);
 		free(old_filename);
@@ -1749,7 +1559,8 @@ __export int osl_update_object(struct osl_table *t, const struct osl_row *r,
 	}
 	remove_rb_node(t, col_num, r);
 	if (cd->storage_type == OSL_NO_STORAGE) { /* TODO: If fixed size, reuse object? */
-		free(r->volatile_objects[col->volatile_num].data);
+		if (!(cd->storage_flags & OSL_DONT_FREE))
+			free(r->volatile_objects[col->volatile_num].data);
 		r->volatile_objects[col->volatile_num] = *obj;
 	} else if (cd->storage_type == OSL_DISK_STORAGE) {
 		char *ds_name;
@@ -1757,7 +1568,7 @@ __export int osl_update_object(struct osl_table *t, const struct osl_row *r,
 		if (ret < 0)
 			return ret;
 		ret = delete_disk_storage_file(t, col_num, ds_name);
-		if (ret < 0 && !is_errno(-ret, ENOENT)) {
+		if (ret < 0 && ret != -E_OSL_NOENT) {
 			free(ds_name);
 			return ret;
 		}
@@ -1786,9 +1597,6 @@ __export int osl_update_object(struct osl_table *t, const struct osl_row *r,
 			uint32_t new_data_map_size;
 			char *row_index;
 			ret = get_row_index(t, r->num, &row_index);
-			if (ret < 0)
-				return ret;
-			ret = mark_mapped_object_invalid(t, r->num, col_num);
 			if (ret < 0)
 				return ret;
 			unmap_column(t, col_num);
@@ -1830,7 +1638,7 @@ __export int osl_open_disk_object(const struct osl_table *t, const struct osl_ro
 	filename = disk_storage_path(t, col_num, ds_name);
 	free(ds_name);
 	if (!filename)
-		return -ERRNO_TO_ERROR(ENOMEM);
+		return -E_OSL_NOMEM;
 	DEBUG_LOG("filename: %s\n", filename);
 	ret = mmap_full_file(filename, O_RDONLY, &obj->data, &obj->size, NULL);
 	free(filename);
@@ -1839,7 +1647,7 @@ __export int osl_open_disk_object(const struct osl_table *t, const struct osl_ro
 
 __export int osl_close_disk_object(struct osl_object *obj)
 {
-	return para_munmap(obj->data, obj->size);
+	return osl_munmap(obj->data, obj->size);
 }
 
 __export int osl_get_num_rows(const struct osl_table *t, unsigned *num_rows)
@@ -1881,6 +1689,7 @@ __export int osl_get_nth_row(const struct osl_table *t, unsigned col_num,
 	unsigned num_rows;
 	int ret;
 
+	*result = NULL;
 	if (n == 0)
 		return -E_OSL_RB_KEY_NOT_FOUND;
 	ret = osl_get_num_rows(t, &num_rows);