X-Git-Url: http://git.tuebingen.mpg.de/?p=adu.git;a=blobdiff_plain;f=adu.c;h=54af136ad789314b88392658077fa88c926d766f;hp=0e32f8564079b4d5eac0c9302f5e95a248d64062;hb=694171759afb90bfdc552676ff6d04697c93af00;hpb=a4906ffe6dce718374052de10b9def84433c84df

diff --git a/adu.c b/adu.c
index 0e32f85..54af136 100644
--- a/adu.c
+++ b/adu.c
@@ -2,18 +2,129 @@
 #include <dirent.h> /* readdir() */
 
 #include "gcc-compat.h"
-#include "osl.h"
+#include "cmdline.h"
 #include "fd.h"
-#include "hash.h"
 #include "string.h"
 #include "error.h"
+#include "portable_io.h"
 
 DEFINE_ERRLIST;
+int osl_errno;
+
+/** Command line and config file options. */
+static struct gengetopt_args_info conf;
+
+enum uid_info_flags {
+	/** whether this slot of the hash table is used. */
+	UI_FL_SLOT_USED = 1,
+	/** whether this uid should be taken into account. */
+	UI_FL_ADMISSIBLE = 2,
+};
+
+struct user_info {
+	uint32_t uid;
+	uint32_t flags;
+	struct osl_table *table;
+	uint64_t files;
+	uint64_t bytes;
+	uint64_t dirs;
+	struct osl_table_description *desc;
+};
+
+/** The decimal representation of an uint64_t never exceeds that size. */
+#define FORMATED_VALUE_SIZE 25
+
+#define FOR_EACH_USER(ui) for (ui = uid_hash_table; ui && ui < uid_hash_table \
+		+ uid_hash_table_size; ui++)
+
+
+/**
+ * Contains info for each user that owns at least one regular file.
+ *
+ * Even users that are not taken into account because of the --uid
+ * option occupy a slot in this hash table. This allows to find out
+ * quicky whether a uid is admissible. And yes, this has to be fast.
+ */
+static struct user_info *uid_hash_table;
+
+/* these get filled in by the select command. */
+static char count_unit_buf[4] = "( )", size_unit_buf[4] = "( )";
+
+static inline int ui_used(struct user_info *ui)
+{
+	return ui->flags & UI_FL_SLOT_USED;
+}
+
+static inline int ui_admissible(struct user_info *ui)
+{
+	return ui->flags & UI_FL_ADMISSIBLE;
+}
+
+struct uid_range {
+	uint32_t low;
+	uint32_t high;
+};
+
+static struct uid_range *admissible_uids;
+
+static inline int check_uid_arg(const char *arg, uint32_t *uid)
+{
+	const uint32_t max = ~0U;
+	/*
+	 * we need an 64-bit int for string -> uid conversion because strtoll()
+	 * returns a signed value.
+	 */
+	int64_t val;
+	int ret = para_atoi64(arg, &val);
+
+	if (ret < 0)
+		return ret;
+	if (val < 0 || val > max)
+		return -ERRNO_TO_ERROR(EINVAL);
+	*uid = val;
+	return 1;
+}
+
+static int parse_uid_range(const char *orig_arg, struct uid_range *ur)
+{
+	int ret;
+	char *arg = para_strdup(orig_arg), *p = strchr(arg, '-');
+
+	if (!p || p == arg) { /* -42 or 42 */
+		ret = check_uid_arg(p? p + 1 : arg, &ur->high);
+		if (ret < 0)
+			goto out;
+		ur->low = p? 0 : ur->high;
+		ret = 1;
+		goto out;
+	}
+	/* 42- or 42-4711 */
+	*p = '\0';
+	p++;
+	ret = check_uid_arg(arg, &ur->low);
+	if (ret < 0)
+		goto out;
+	ur->high = ~0U;
+	if (*p) { /* 42-4711 */
+		ret = check_uid_arg(p, &ur->high);
+		if (ret < 0)
+			goto out;
+	}
+	if (ur->low > ur->high)
+		ret = -ERRNO_TO_ERROR(EINVAL);
+out:
+	if (ret < 0)
+		ERROR_LOG("bad uid option: %s\n", orig_arg);
+	else
+		INFO_LOG("admissible uid range: %u - %u\n", ur->low,
+			ur->high);
+	free(arg);
+	return ret;
+}
 
 /** evaluates to 1 if x < y, to -1 if x > y and to 0 if x == y */
 #define NUM_COMPARE(x, y) ((int)((x) < (y)) - (int)((x) > (y)))
 
-
 /**
  * The log function.
  *
@@ -30,7 +141,7 @@ __printf_2_3 void __log(int ll, const char* fmt,...)
 	time_t t1;
 	char str[255] = "";
 
-	if (ll < 4)
+	if (ll < conf.loglevel_arg)
 		return;
 	outfd = stderr;
 	time(&t1);
@@ -65,25 +176,26 @@ static int size_compare(const struct osl_object *obj1, const struct osl_object *
 }
 
 /**
- * Compare two osl objects of string type.
+ * Compare two osl objects pointing to unsigned integers of 64 bit size.
  *
- * \param obj1 Pointer to the first object.
- * \param obj2 Pointer to the second object.
- *
- * In any case, only \p MIN(obj1->size, obj2->size) characters of each string
- * are taken into account.
+ * \param obj1 Pointer to the first integer.
+ * \param obj2 Pointer to the second integer.
  *
- * \return It returns an integer less than, equal to, or greater than zero if
- * \a obj1 is found, respectively, to be less than, to match, or be greater than
- * obj2.
+ * \return The values required for an osl compare function.
  *
- * \sa strcmp(3), strncmp(3), osl_compare_func.
+ * \sa osl_compare_func, osl_hash_compare().
  */
-int string_compare(const struct osl_object *obj1, const struct osl_object *obj2)
+static int uint64_compare(const struct osl_object *obj1,
+		const struct osl_object *obj2)
 {
-	const char *str1 = (const char *)obj1->data;
-	const char *str2 = (const char *)obj2->data;
-	return strncmp(str1, str2, MIN(obj1->size, obj2->size));
+	uint64_t d1 = read_u64((const char *)obj1->data);
+	uint64_t d2 = read_u64((const char *)obj2->data);
+
+	if (d1 < d2)
+		return 1;
+	if (d1 > d2)
+		return -1;
+	return 0;
 }
 
 /** The columns of the directory table. */
@@ -92,6 +204,8 @@ enum dir_table_columns {
 	DT_NAME,
 	/** The dir count number. */
 	DT_NUM,
+	/** The number of the parent directory. */
+	DT_PARENT_NUM,
 	/** The number of bytes of all regular files. */
 	DT_BYTES,
 	/** The number of all regular files. */
@@ -103,16 +217,22 @@ enum dir_table_columns {
 static struct osl_column_description dir_table_cols[] = {
 	[DT_NAME] = {
 		.storage_type = OSL_MAPPED_STORAGE,
-		.storage_flags = OSL_RBTREE | OSL_UNIQUE,
+		.storage_flags = 0,
 		.name = "dir",
-		.compare_function = string_compare,
 	},
 	[DT_NUM] = {
 		.storage_type = OSL_MAPPED_STORAGE,
 		.storage_flags = OSL_RBTREE | OSL_FIXED_SIZE | OSL_UNIQUE,
 		.name = "num",
-		.compare_function = uint32_compare,
-		.data_size = sizeof(uint32_t)
+		.compare_function = uint64_compare,
+		.data_size = sizeof(uint64_t)
+	},
+	[DT_PARENT_NUM] = {
+		.storage_type = OSL_MAPPED_STORAGE,
+		.storage_flags = OSL_RBTREE | OSL_FIXED_SIZE | OSL_UNIQUE,
+		.name = "parent_num",
+		.compare_function = size_compare,
+		.data_size = sizeof(uint64_t)
 	},
 	[DT_BYTES] = {
 		.storage_type = OSL_MAPPED_STORAGE,
@@ -135,59 +255,6 @@ static struct osl_table_description dir_table_desc = {
 	.num_columns = NUM_DT_COLUMNS,
 	.flags = 0,
 	.column_descriptions = dir_table_cols,
-	.dir = "/tmp/adu"
-};
-
-/** The columns of the id table. */
-enum id_table_columns {
-	/** The user id. */
-	IDT_UID,
-	/** The number of bytes of all regular files owned by this id. */
-	IDT_BYTES,
-	/** The number of regular files owned by this id. */
-	IDT_FILES,
-	/** The user table for this uid. */
-	IDT_TABLE,
-	/** Number of columns in this table. */
-	NUM_IDT_COLUMNS
-};
-
-static struct osl_column_description id_table_cols[] = {
-	[IDT_UID] = {
-		.storage_type = OSL_MAPPED_STORAGE,
-		.storage_flags = OSL_RBTREE | OSL_FIXED_SIZE | OSL_UNIQUE,
-		.name = "uid",
-		.compare_function = uint32_compare,
-		.data_size = sizeof(uint32_t)
-	},
-	[IDT_BYTES] = {
-		.storage_type = OSL_MAPPED_STORAGE,
-		.storage_flags = OSL_RBTREE | OSL_FIXED_SIZE,
-		.compare_function = size_compare,
-		.name = "num_bytes",
-		.data_size = sizeof(uint64_t)
-	},
-	[IDT_FILES] = {
-		.storage_type = OSL_MAPPED_STORAGE,
-		.storage_flags = OSL_RBTREE | OSL_FIXED_SIZE,
-		.compare_function = size_compare,
-		.name = "num_filess",
-		.data_size = sizeof(uint64_t)
-	},
-	[IDT_TABLE] = {
-		.storage_type = OSL_NO_STORAGE,
-		.storage_flags = OSL_FIXED_SIZE | OSL_UNIQUE,
-		.name = "user_table",
-		.data_size = sizeof(void *)
-	}
-};
-
-static struct osl_table_description id_table_desc = {
-	.name = "id_table",
-	.num_columns = NUM_IDT_COLUMNS,
-	.flags = 0,
-	.column_descriptions = id_table_cols,
-	.dir = "/tmp/adu"
 };
 
 /** The columns of the id table. */
@@ -207,17 +274,17 @@ static struct osl_column_description user_table_cols[] = {
 		.storage_type = OSL_MAPPED_STORAGE,
 		.storage_flags = OSL_RBTREE | OSL_FIXED_SIZE | OSL_UNIQUE,
 		.name = "dir_num",
-		.compare_function = uint32_compare,
-		.data_size = sizeof(uint32_t)
+		.compare_function = uint64_compare,
+		.data_size = sizeof(uint64_t)
 	},
-	[IDT_BYTES] = {
+	[UT_BYTES] = {
 		.storage_type = OSL_MAPPED_STORAGE,
 		.storage_flags = OSL_RBTREE | OSL_FIXED_SIZE,
 		.compare_function = size_compare,
 		.name = "num_bytes",
 		.data_size = sizeof(uint64_t)
 	},
-	[IDT_FILES] = {
+	[UT_FILES] = {
 		.storage_type = OSL_MAPPED_STORAGE,
 		.storage_flags = OSL_RBTREE | OSL_FIXED_SIZE,
 		.compare_function = size_compare,
@@ -226,124 +293,251 @@ static struct osl_column_description user_table_cols[] = {
 	},
 };
 
-static struct osl_table_description user_table_desc = {
-	.num_columns = NUM_UT_COLUMNS,
-	.flags = 0,
-	.column_descriptions = user_table_cols,
-	.dir = "/tmp/adu"
-};
 static struct osl_table *dir_table;
-static struct osl_table *id_table;
 
-static int create_tables(void)
-{
-	int ret = osl_create_table(&dir_table_desc);
-	if (ret < 0)
-		return ret;
-	ret = osl_create_table(&id_table_desc);
-	if (ret < 0)
-		return ret;
-	return 1;
-}
-
-int add_directory(char *dirname, uint32_t dir_num, uint64_t *dir_size,
-		uint64_t *dir_files)
+static int add_directory(char *dirname, uint64_t *dir_num, uint64_t *parent_dir_num,
+		uint64_t *dir_size, uint64_t *dir_files)
 {
 	struct osl_object dir_objects[NUM_DT_COLUMNS];
 
-	INFO_LOG("adding #%u: %s\n", dir_num, dirname);
+	INFO_LOG("adding #%llu: %s\n", (long long unsigned)*dir_num, dirname);
 	dir_objects[DT_NAME].data = dirname;
 	dir_objects[DT_NAME].size = strlen(dirname) + 1;
-	dir_objects[DT_NUM].data = &dir_num;
-	dir_objects[DT_NUM].size = sizeof(dir_num);
+	dir_objects[DT_NUM].data = dir_num;
+	dir_objects[DT_NUM].size = sizeof(*dir_num);
+	dir_objects[DT_PARENT_NUM].data = parent_dir_num;
+	dir_objects[DT_PARENT_NUM].size = sizeof(*parent_dir_num);
 	dir_objects[DT_BYTES].data = dir_size;
 	dir_objects[DT_BYTES].size = sizeof(*dir_size);
 	dir_objects[DT_FILES].data = dir_files;
 	dir_objects[DT_FILES].size = sizeof(*dir_files);
-
-	return osl_add_row(dir_table, dir_objects);
+	return osl(osl_add_row(dir_table, dir_objects));
 }
 
-int create_and_open_user_table(uint32_t uid, struct osl_table **t)
+static uint32_t num_uids;
+
+static int open_user_table(struct user_info *ui, int create)
 {
 	int ret;
-	struct osl_table_description *desc = para_malloc(sizeof(*desc));
-
-	desc->num_columns = NUM_UT_COLUMNS;
-	desc->flags = 0;
-	desc->column_descriptions = user_table_cols;
-	desc->dir = para_strdup("/tmp/adu");
-	desc->name = make_message("%u", uid);
-	INFO_LOG("................................. %u\n", uid);
-//	user_table_desc.name = make_message("%u", uid);
-	ret = osl_create_table(desc);
+
+	ui->desc = para_malloc(sizeof(*ui->desc));
+	ui->desc->num_columns = NUM_UT_COLUMNS;
+	ui->desc->flags = 0;
+	ui->desc->column_descriptions = user_table_cols;
+	ui->desc->dir = para_strdup(conf.database_dir_arg);
+	ui->desc->name = make_message("%u", (unsigned)ui->uid);
+	INFO_LOG(".............................uid #%u: %u\n",
+		(unsigned)num_uids, (unsigned)ui->uid);
+	if (create) {
+		ret = osl(osl_create_table(ui->desc));
+		if (ret < 0)
+			goto err;
+		num_uids++;
+	}
+	ret = osl(osl_open_table(ui->desc, &ui->table));
 	if (ret < 0)
-		return ret;
-	return osl_open_table(desc, t);
+		goto err;
+	return 1;
+err:
+	free((char *)ui->desc->name);
+	free((char *)ui->desc->dir);
+	free(ui->desc);
+	ui->desc->name = NULL;
+	ui->desc->dir = NULL;
+	ui->desc = NULL;
+	ui->table = NULL;
+	ui->flags = 0;
+	return ret;
 }
 
-static int insert_id_row(uint32_t uid, struct osl_table *t, struct osl_row **row)
-{
-	struct osl_object objects[NUM_IDT_COLUMNS];
-	uint64_t num = 0;
+#define uid_hash_bits 8
+static uint32_t uid_hash_table_size = 1 << uid_hash_bits;
+#define PRIME1 0x811c9dc5
+#define PRIME2 0x01000193
 
-	struct osl_table **table_ptr = para_malloc(sizeof(*table_ptr));
-	*table_ptr = t;
+static void create_hash_table(void)
+{
+	uid_hash_table = para_calloc(uid_hash_table_size
+		* sizeof(struct user_info));
+}
 
-	INFO_LOG("§§§§§§§§§§§§§§§§§§§§§ uid: %d, t: %p\n", uid, t);
-	objects[IDT_UID].data = &uid;
-	objects[IDT_UID].size = sizeof(uid);
-	objects[IDT_BYTES].data = &num;
-	objects[IDT_BYTES].size = sizeof(num);
-	objects[IDT_FILES].data = &num;
-	objects[IDT_FILES].size = sizeof(num);
-	objects[IDT_TABLE].data = table_ptr;
-	objects[IDT_TABLE].size = sizeof(*table_ptr);
-	return osl_add_and_get_row(id_table, objects, row);
+static void free_hash_table(void)
+{
+	free(uid_hash_table);
+	uid_hash_table = NULL;
 }
 
-static int get_user_table(struct osl_row *row, struct osl_table **t)
+static int create_tables(void)
 {
-	struct osl_object obj;
+	int ret;
 
-	int ret = osl_get_object(id_table, row, IDT_TABLE, &obj);
+	dir_table_desc.dir = para_strdup(conf.database_dir_arg);
+	ret = osl(osl_create_table(&dir_table_desc));
 	if (ret < 0)
 		return ret;
-	*t = *(struct osl_table **)obj.data;
-	INFO_LOG("^^^^^^^^^^^^^^^^^^ t: %p\n", *t);
+	create_hash_table();
 	return 1;
 }
 
-static int add_id_bytes(struct osl_row *row, uint64_t *add)
+static void close_dir_table(void)
 {
-	uint64_t num;
-	struct osl_object obj1, obj2 = {.data = &num, .size = sizeof(num)};
+	int ret;
 
-	/* update number of bytes */
-	int ret = osl_get_object(id_table, row, IDT_BYTES, &obj1);
-	if (ret < 0)
-		return ret;
-	num = *(uint64_t *)obj1.data + *add;
-	ret = osl_update_object(id_table, row, IDT_BYTES, &obj2);
+	if (!dir_table)
+		return;
+	ret = osl(osl_close_table(dir_table, OSL_MARK_CLEAN));
 	if (ret < 0)
-		return ret;
-	/* increment number of files */
-	ret = osl_get_object(id_table, row, IDT_FILES, &obj1);
+		ERROR_LOG("failed to close dir table: %s\n", adu_strerror(-ret));
+	free((char *)dir_table_desc.dir);
+	dir_table = NULL;
+}
+
+static void close_user_table(struct user_info *ui)
+{
+	int ret;
+
+	if (!ui || !ui_used(ui) || !ui_admissible(ui))
+		return;
+	ret = osl(osl_close_table(ui->table, OSL_MARK_CLEAN));
 	if (ret < 0)
-		return ret;
-	num = *(uint64_t *)obj1.data + 1;
-	return osl_update_object(id_table, row, IDT_FILES, &obj2);
+		ERROR_LOG("failed to close user table %u: %s\n",
+			(unsigned) ui->uid, adu_strerror(-ret));
+	free((char *)ui->desc->name);
+	ui->desc->name = NULL;
+	free((char *)ui->desc->dir);
+	ui->desc->dir = NULL;
+	free(ui->desc);
+	ui->desc = NULL;
+	ui->table = NULL;
+	ui->flags = 0;
+}
+
+static void close_user_tables(void)
+{
+	struct user_info *ui;
+
+	FOR_EACH_USER(ui)
+		close_user_table(ui);
 }
 
-static int update_user_row(struct osl_table *t, uint32_t dir_num,
+static void close_all_tables(void)
+{
+	close_dir_table();
+	close_user_tables();
+	free_hash_table();
+}
+
+static int signum;
+
+static void signal_handler(int s)
+{
+	signum = s;
+}
+
+static void check_signals(void)
+{
+	if (likely(!signum))
+		return;
+	EMERG_LOG("caught signal %d\n", signum);
+	close_all_tables();
+	exit(EXIT_FAILURE);
+}
+
+static int init_signals(void)
+{
+	if (signal(SIGINT, &signal_handler) == SIG_ERR)
+		return -E_SIGNAL_SIG_ERR;
+	if (signal(SIGTERM, &signal_handler) == SIG_ERR)
+		return -E_SIGNAL_SIG_ERR;
+	return 1;
+}
+
+/*
+ * We use a hash table of size s=2^uid_hash_bits to map the uids into the
+ * interval [0..s]. Hash collisions are treated by open addressing, i.e.
+ * unused slots in the table are used to store different uids that hash to the
+ * same slot.
+ *
+ * If a hash collision occurs, different slots are successively probed in order
+ * to find an unused slot for the new uid. Probing is implemented via a second
+ * hash function that maps the uid to h=(uid * PRIME2) | 1, which is always an
+ * odd number.
+ *
+ * An odd number is sufficient to make sure each entry of the hash table gets
+ * probed for probe_num between 0 and s-1 because s is a power of two, hence
+ * the second hash value has never a common divisor with the hash table size.
+ * IOW: h is invertible in the ring [0..s].
+ */
+static uint32_t double_hash(uint32_t uid, uint32_t probe_num)
+{
+	return (uid * PRIME1 + ((uid * PRIME2) | 1) * probe_num)
+		% uid_hash_table_size;
+}
+
+enum search_uid_flags {
+	OPEN_USER_TABLE = 1,
+	CREATE_USER_TABLE = 2,
+};
+
+static int uid_is_admissible(uint32_t uid)
+{
+	int i;
+
+	for (i = 0; i < conf.uid_given; i++) {
+		struct uid_range *ur = admissible_uids + i;
+
+		if (ur->low <= uid && ur->high >= uid)
+			break;
+	}
+	i = !conf.uid_given || i < conf.uid_given;
+	DEBUG_LOG("uid %u is %sadmissible\n", (unsigned)uid,
+		i? "" : "not ");
+	return i;
+}
+
+static int search_uid(uint32_t uid, enum search_uid_flags flags,
+		struct user_info **ui_ptr)
+{
+	uint32_t p;
+
+	for (p = 0; p < uid_hash_table_size; p++) {
+		struct user_info *ui = uid_hash_table + double_hash(uid, p);
+
+		if (!ui_used(ui)) {
+			int ret;
+			if (!flags)
+				return -E_BAD_UID;
+			ui->uid = uid;
+			ui->flags |= UI_FL_SLOT_USED;
+			if (!uid_is_admissible(uid))
+				return 0;
+			ui->flags |= UI_FL_ADMISSIBLE;
+			ret = open_user_table(ui, flags & CREATE_USER_TABLE);
+			if (ret < 0)
+				return ret;
+
+			if (ui_ptr)
+				*ui_ptr = ui;
+			return 1;
+		}
+		if (ui->uid != uid)
+			continue;
+		if (ui_ptr)
+			*ui_ptr = ui;
+		return 0;
+	}
+	return flags? -E_HASH_TABLE_OVERFLOW : -E_BAD_UID;
+}
+
+static int update_user_row(struct osl_table *t, uint64_t dir_num,
 		uint64_t *add)
 {
 	struct osl_row *row;
 	struct osl_object obj = {.data = &dir_num, .size = sizeof(dir_num)};
 
-	int ret = osl_get_row(t, UT_DIR_NUM, &obj, &row);
+	int ret = osl(osl_get_row(t, UT_DIR_NUM, &obj, &row));
 
-	if (ret < 0 && ret != -E_RB_KEY_NOT_FOUND)
+	if (ret == -E_OSL && osl_errno != E_OSL_RB_KEY_NOT_FOUND)
 		return ret;
 	if (ret < 0) { /* this is the first file we add */
 		struct osl_object objects[NUM_UT_COLUMNS];
@@ -356,39 +550,42 @@ static int update_user_row(struct osl_table *t, uint32_t dir_num,
 		objects[UT_FILES].data = &num_files;
 		objects[UT_FILES].size = sizeof(num_files);
 		INFO_LOG("######################### ret: %d\n", ret);
-		ret = osl_add_row(t, objects);
+		ret = osl(osl_add_row(t, objects));
 		INFO_LOG("######################### ret: %d\n", ret);
 		return ret;
 	} else { /* add size and increment file count */
 		uint64_t num;
 		struct osl_object obj1, obj2 = {.data = &num, .size = sizeof(num)};
 
-		ret = osl_get_object(t, row, UT_BYTES, &obj1);
+		ret = osl(osl_get_object(t, row, UT_BYTES, &obj1));
 		if (ret < 0)
 			return ret;
 		num = *(uint64_t *)obj1.data + *add;
-		ret = osl_update_object(t, row, UT_BYTES, &obj2);
+		ret = osl(osl_update_object(t, row, UT_BYTES, &obj2));
 		if (ret < 0)
 			return ret;
-		ret = osl_get_object(t, row, UT_FILES, &obj1);
+		ret = osl(osl_get_object(t, row, UT_FILES, &obj1));
 		if (ret < 0)
 			return ret;
 		num = *(uint64_t *)obj1.data + 1;
-		return osl_update_object(t, row, UT_FILES, &obj2);
+		return osl(osl_update_object(t, row, UT_FILES, &obj2));
 	}
 }
 
-static uint32_t dir_num;
+static uint64_t num_dirs;
+static uint64_t num_files;
+static uint64_t num_bytes;
 
-int scan_dir(char *dirname)
+int scan_dir(char *dirname, uint64_t *parent_dir_num)
 {
 	DIR *dir;
 	struct dirent *entry;
 	int ret, cwd_fd, ret2;
 	uint64_t dir_size = 0, dir_files = 0;
-	struct osl_object obj;
+	uint64_t this_dir_num = ++num_dirs;
 
-	INFO_LOG("----------------- %s\n", dirname);
+	check_signals();
+	DEBUG_LOG("----------------- %llu: %s\n", (long long unsigned)num_dirs, dirname);
 	ret = para_opendir(dirname, &dir, &cwd_fd);
 	if (ret < 0) {
 		if (ret != -ERRNO_TO_ERROR(EACCES))
@@ -398,26 +595,25 @@ int scan_dir(char *dirname)
 	}
 	while ((entry = readdir(dir))) {
 		mode_t m;
-		char *tmp;
 		struct stat s;
 		uint32_t uid;
 		uint64_t size;
-		struct osl_row *id_row;
-		struct osl_table *user_table;
+		struct user_info *ui;
 
 		if (!strcmp(entry->d_name, "."))
 			continue;
 		if (!strcmp(entry->d_name, ".."))
 			continue;
-		if (lstat(entry->d_name, &s) == -1)
+		if (lstat(entry->d_name, &s) == -1) {
+			WARNING_LOG("lstat error for %s/%s\n", dirname,
+				entry->d_name);
 			continue;
+		}
 		m = s.st_mode;
 		if (!S_ISREG(m) && !S_ISDIR(m))
 			continue;
-		tmp = make_message("%s/%s", dirname, entry->d_name);
 		if (S_ISDIR(m)) {
-			ret = scan_dir(tmp);
-			free(tmp);
+			ret = scan_dir(entry->d_name, &this_dir_num);
 			if (ret < 0)
 				goto out;
 			continue;
@@ -425,36 +621,21 @@ int scan_dir(char *dirname)
 		/* regular file */
 		size = s.st_size;
 		dir_size += size;
+		num_bytes += size;
 		dir_files++;
+		num_files++;
 		uid = s.st_uid;
-		INFO_LOG("++++++++++++++++++++++++++ %s, uid: %u\n", entry->d_name, uid);
-		obj.data = &uid;
-		obj.size = sizeof(uid);
-		ret = osl_get_row(id_table, IDT_UID, &obj, &id_row);
-		if (ret < 0 && ret != -E_RB_KEY_NOT_FOUND)
-			goto out;
-		if (ret < 0) {
-			ret = create_and_open_user_table(uid, &user_table);
-			if (ret < 0)
-				goto out;
-			ret = insert_id_row(uid, user_table, &id_row);
-			if (ret < 0)
-				goto out;
-		} else {
-			ret = get_user_table(id_row, &user_table);
-			if (ret < 0)
-				goto out;
-		}
-		ret = add_id_bytes(id_row, &size);
+		ret = search_uid(uid, CREATE_USER_TABLE | OPEN_USER_TABLE, &ui);
 		if (ret < 0)
 			goto out;
-		INFO_LOG("user_table: %p\n", user_table);
-		ret = update_user_row(user_table, dir_num, &size);
-		INFO_LOG("update_user  ret: %d\n", ret);
+		ui->bytes += size;
+		ui->files++;
+		ret = update_user_row(ui->table, this_dir_num, &size);
 		if (ret < 0)
 			goto out;
 	}
-	ret = add_directory(dirname, dir_num++, &dir_size, &dir_files);
+	ret = add_directory(dirname, &this_dir_num, parent_dir_num,
+			&dir_size, &dir_files);
 out:
 	closedir(dir);
 	ret2 = para_fchdir(cwd_fd);
@@ -464,182 +645,611 @@ out:
 	return ret;
 }
 
-static int get_dir_name(struct osl_row *row, char **name)
+static int get_dir_name_by_number(uint64_t *dirnum, char **name)
 {
-	struct osl_object obj;
-	int ret = osl_get_object(dir_table, row, DT_NAME, &obj);
+	char *result = NULL, *tmp;
+	struct osl_row *row;
+	uint64_t val = *dirnum;
+	struct osl_object obj = {.data = &val, .size = sizeof(val)};
+	int ret;
 
+again:
+	ret = osl(osl_get_row(dir_table, DT_NUM, &obj, &row));
 	if (ret < 0)
-		return ret;
-	*name = obj.data;
-	return 1;
+		goto out;
+	ret = osl(osl_get_object(dir_table, row, DT_NAME, &obj));
+	if (ret < 0)
+		goto out;
+	if (result) {
+		tmp = make_message("%s/%s", (char *)obj.data, result);
+		free(result);
+		result = tmp;
+	} else
+		result = para_strdup((char *)obj.data);
+	ret = osl(osl_get_object(dir_table, row, DT_PARENT_NUM, &obj));
+	if (ret < 0)
+		goto out;
+	val = *(uint64_t *)obj.data;
+	if (val)
+		goto again;
+out:
+	if (ret < 0) {
+		free(result);
+		*name = NULL;
+	} else
+		*name = result;
+	return ret;
 }
 
-static int print_dirname_and_size(struct osl_row *row, void *data)
+static int get_dir_name_of_row(struct osl_row *dir_table_row, char **name)
 {
-	unsigned *count = data;
 	struct osl_object obj;
-	char *name;
 	int ret;
+	char *this_dir, *prefix = NULL;
 
-	if ((*count)++ > 100)
-		return -E_LOOP_COMPLETE;
-	ret = get_dir_name(row, &name);
+	*name = NULL;
+	ret = osl(osl_get_object(dir_table, dir_table_row, DT_NAME, &obj));
 	if (ret < 0)
 		return ret;
-	ret = osl_get_object(dir_table, row, DT_BYTES, &obj);
+	this_dir = para_strdup((char *)obj.data);
+	ret = osl(osl_get_object(dir_table, dir_table_row, DT_PARENT_NUM, &obj));
 	if (ret < 0)
-		return ret;
-	printf("%s\t%llu\n", name, *(long long unsigned *)obj.data);
-	return 1;
+		goto out;
+	if (!*(uint64_t *)obj.data) {
+		*name = this_dir;
+		return 1;
+	}
+	ret = get_dir_name_by_number((uint64_t *)obj.data, &prefix);
+	if (ret < 0)
+		goto out;
+	*name = make_message("%s/%s", prefix, this_dir);
+	free(prefix);
+	ret = 1;
+out:
+	free(this_dir);
+	return ret;
 }
 
-static int print_dirname_and_file_count(struct osl_row *row, void *data)
+const uint64_t size_unit_divisors[] = {
+	[size_unit_arg_b] = 1ULL,
+	[size_unit_arg_k] = 1024ULL,
+	[size_unit_arg_m] = 1024ULL * 1024ULL,
+	[size_unit_arg_g] = 1024ULL * 1024ULL * 1024ULL,
+	[size_unit_arg_t] = 1024ULL * 1024ULL * 1024ULL * 1024ULL,
+};
+
+const uint64_t count_unit_divisors[] = {
+
+	[count_unit_arg_n] = 1ULL,
+	[count_unit_arg_k] = 1000ULL,
+	[count_unit_arg_m] = 1000ULL * 1000ULL,
+	[count_unit_arg_g] = 1000ULL * 1000ULL * 1000ULL,
+	[count_unit_arg_t] = 1000ULL * 1000ULL * 1000ULL * 1000ULL,
+};
+
+const char size_unit_abbrevs[] = " BKMGT";
+const char count_unit_abbrevs[] = "  kmgt";
+
+static enum enum_size_unit format_size_value(enum enum_size_unit unit,
+		uint64_t value, int print_unit, char *result)
 {
-	unsigned *count = data;
-	struct osl_object obj;
-	char *name;
+	enum enum_size_unit u = unit;
+	char unit_buf[2] = "\0\0";
+
+	if (unit == size_unit_arg_h) /* human readable */
+		for (u = size_unit_arg_b; u < size_unit_arg_t &&
+				value > size_unit_divisors[u + 1]; u++)
+			; /* nothing */
+	if (print_unit)
+		unit_buf[0] = size_unit_abbrevs[u];
+	sprintf(result, "%llu%s",
+		(long long unsigned)value / size_unit_divisors[u], unit_buf);
+	return u;
+}
+
+static enum enum_count_unit format_count_value(enum enum_count_unit unit,
+		uint64_t value, int print_unit, char *result)
+{
+	enum enum_count_unit u = unit;
+	char unit_buf[2] = "\0\0";
+
+	if (unit == count_unit_arg_h) /* human readable */
+		for (u = count_unit_arg_n; u < count_unit_arg_t &&
+				value > count_unit_divisors[u + 1]; u++)
+			; /* nothing */
+	if (print_unit)
+		unit_buf[0] = count_unit_abbrevs[u];
+	sprintf(result, "%llu%s",
+		(long long unsigned)value / count_unit_divisors[u], unit_buf);
+	return u;
+}
+
+enum global_stats_flags {
+	GSF_PRINT_DIRNAME = 1,
+	GSF_PRINT_BYTES = 2,
+	GSF_PRINT_FILES = 4,
+	GSF_COMPUTE_SUMMARY = 8,
+};
+
+struct global_stats_info {
+	uint32_t count;
 	int ret;
+	int osl_errno;
+	enum global_stats_flags flags;
+};
 
-	if ((*count)++ > 100)
-		return -E_LOOP_COMPLETE;
-	ret = get_dir_name(row, &name);
-	if (ret < 0)
-		return ret;
-	ret = osl_get_object(dir_table, row, DT_FILES, &obj);
-	if (ret < 0)
-		return ret;
-	printf("%s\t%llu\n", name, *(long long unsigned *)obj.data);
+static int global_stats_loop_function(struct osl_row *row, void *data)
+{
+	struct global_stats_info *gsi = data;
+	struct osl_object obj;
+	char *dirname, formated_value[FORMATED_VALUE_SIZE];
+	int ret, summary = gsi->flags & GSF_COMPUTE_SUMMARY;
+
+	check_signals();
+	if (!gsi->count && !summary) {
+		ret = -E_LOOP_COMPLETE;
+		goto err;
+	}
+	if (summary || (gsi->count && (gsi->flags & GSF_PRINT_FILES))) {
+		uint64_t files;
+		ret = osl(osl_get_object(dir_table, row, DT_FILES, &obj));
+		if (ret < 0)
+			goto err;
+		files = *(uint64_t *)obj.data;
+		if (gsi->count && (gsi->flags & GSF_PRINT_FILES)) {
+			format_count_value(conf.count_unit_arg, files,
+				conf.count_unit_arg == count_unit_arg_h,
+				formated_value);
+			printf("\t%s%s", formated_value,
+				(gsi->flags & (GSF_PRINT_BYTES | GSF_PRINT_DIRNAME))?
+				"\t" : "\n");
+		}
+		if (summary)
+			num_files += files;
+	}
+	if (summary || (gsi->count && (gsi->flags & GSF_PRINT_BYTES))) {
+		uint64_t bytes;
+		ret = osl(osl_get_object(dir_table, row, DT_BYTES, &obj));
+		if (ret < 0)
+			goto err;
+		bytes = *(uint64_t *)obj.data;
+		if (gsi->count && (gsi->flags & GSF_PRINT_BYTES)) {
+			format_size_value(conf.size_unit_arg, bytes,
+				conf.size_unit_arg == size_unit_arg_h,
+				formated_value);
+			printf("%s%s%s",
+				(gsi->flags & GSF_PRINT_FILES)? "" : "\t",
+				formated_value,
+				(gsi->flags & GSF_PRINT_DIRNAME)? "\t" : "\n"
+			);
+		}
+		if (summary) {
+			num_bytes += bytes;
+			num_dirs++;
+		}
+	}
+	if (gsi->count && (gsi->flags & GSF_PRINT_DIRNAME)) {
+		ret = get_dir_name_of_row(row, &dirname);
+		if (ret < 0)
+			goto err;
+		printf("%s%s\n",
+			(gsi->flags & (GSF_PRINT_BYTES | GSF_PRINT_FILES))? "" : "\t",
+			dirname);
+		free(dirname);
+	}
+	if (gsi->count > 0)
+		gsi->count--;
 	return 1;
+err:
+	gsi->ret = ret;
+	gsi->osl_errno = (ret == -E_OSL)? osl_errno : 0;
+	return -1;
 }
 
-static int print_id_stats(struct osl_row *row, __a_unused void *data)
+static void print_id_stats(void)
 {
+	struct user_info *ui;
+
+	printf("User summary "
+		"(uid/dirs%s/files%s/size%s):\n",
+		count_unit_buf, count_unit_buf, size_unit_buf);
+	FOR_EACH_USER(ui) {
+		char formated_dir_count[FORMATED_VALUE_SIZE],
+			formated_file_count[FORMATED_VALUE_SIZE],
+			formated_bytes[FORMATED_VALUE_SIZE ];
+		if (!ui_used(ui) || !ui_admissible(ui))
+			continue;
+		format_count_value(conf.count_unit_arg, ui->dirs,
+			conf.count_unit_arg == count_unit_arg_h,
+			formated_dir_count);
+		format_count_value(conf.count_unit_arg, ui->files,
+			conf.count_unit_arg == count_unit_arg_h,
+			formated_file_count);
+		format_size_value(conf.size_unit_arg, ui->bytes,
+			conf.size_unit_arg == size_unit_arg_h,
+			formated_bytes);
+		printf("\t%u\t%s\t%s\t%s\n", (unsigned)ui->uid,
+			formated_dir_count,
+			formated_file_count,
+			formated_bytes
+		);
+	}
+}
+
+enum user_stats_flags {
+	USF_PRINT_DIRNAME = 1,
+	USF_PRINT_BYTES = 2,
+	USF_PRINT_FILES = 4,
+	USF_COMPUTE_SUMMARY = 8,
+};
+
+struct user_stats_info {
+	uint32_t count;
+	enum user_stats_flags flags;
+	int ret;
+	int osl_errno;
+	struct user_info *ui;
+};
+
+static int user_stats_loop_function(struct osl_row *row, void *data)
+{
+	struct user_stats_info *usi = data;
 	struct osl_object obj;
-	uint32_t uid;
-	uint64_t bytes, files;
-	int ret = osl_get_object(id_table, row, IDT_UID, &obj);
+	int ret, summary = usi->flags & GSF_COMPUTE_SUMMARY;
+	char formated_value[FORMATED_VALUE_SIZE];
 
-	if (ret < 0)
-		return ret;
-	uid = *(uint32_t *)obj.data;
-	ret = osl_get_object(id_table, row, IDT_BYTES, &obj);
-	if (ret < 0)
+	check_signals();
+	if (!usi->count && !summary) {
+		ret = -E_LOOP_COMPLETE;
+		goto err;
+	}
+	if (summary || (usi->count && (usi->flags & USF_PRINT_FILES))) {
+		uint64_t files;
+		ret = osl(osl_get_object(usi->ui->table, row, UT_FILES, &obj));
+		if (ret < 0)
+			goto err;
+		files = *(uint64_t *)obj.data;
+		if (usi->count && (usi->flags & USF_PRINT_FILES)) {
+			format_count_value(conf.count_unit_arg, files,
+				conf.count_unit_arg == count_unit_arg_h,
+				formated_value);
+			printf("\t%s%s", formated_value,
+				(usi->flags & (USF_PRINT_BYTES | USF_PRINT_DIRNAME))?
+					"\t" : "\n"
+			);
+		}
+		if (summary)
+			usi->ui->files += files;
+	}
+	if (summary || (usi->count && (usi->flags & USF_PRINT_BYTES))) {
+		uint64_t bytes;
+		ret = osl(osl_get_object(usi->ui->table, row, UT_BYTES, &obj));
+		if (ret < 0)
+			goto err;
+		bytes = *(uint64_t *)obj.data;
+		if (usi->count && (usi->flags & USF_PRINT_BYTES)) {
+			format_size_value(conf.size_unit_arg, bytes,
+				conf.size_unit_arg == size_unit_arg_h,
+				formated_value);
+			printf("%s%s%s",
+				(usi->flags & USF_PRINT_FILES)? "" : "\t",
+				formated_value,
+				usi->flags & USF_PRINT_DIRNAME?  "\t" : "\n"
+			);
+		}
+		if (summary) {
+			usi->ui->bytes += bytes;
+			usi->ui->dirs++;
+		}
+
+	}
+	if (usi->count && (usi->flags & USF_PRINT_DIRNAME)) {
+		char *dirname;
+		ret = osl(osl_get_object(usi->ui->table, row, UT_DIR_NUM, &obj));
+		if (ret < 0)
+			goto err;
+		ret = get_dir_name_by_number((uint64_t *)obj.data, &dirname);
+		if (ret < 0)
+			goto err;
+		printf("%s%s\n",
+			(usi->flags & (USF_PRINT_BYTES | USF_PRINT_FILES))? "" : "\t",
+			dirname);
+		free(dirname);
+	}
+	if (usi->count > 0)
+		usi->count--;
+	return 1;
+err:
+	usi->ret = ret;
+	usi->osl_errno = (ret == -E_OSL)? osl_errno : 0;
+	return -1;
+}
+
+static int check_loop_return(int ret, int loop_ret, int loop_osl_errno)
+{
+	if (ret >= 0)
 		return ret;
-	bytes = *(uint64_t *)obj.data;
-	ret = osl_get_object(id_table, row, IDT_FILES, &obj);
-	if (ret < 0)
+	assert(ret == -E_OSL);
+	if (osl_errno != E_OSL_LOOP)
+		/* error not caused by loop function returning negative. */
 		return ret;
-	files = *(uint64_t *)obj.data;
+	assert(loop_ret < 0);
+	if (loop_ret == -E_LOOP_COMPLETE) /* no error */
+		return 1;
+	if (loop_ret == -E_OSL) { /* osl error in loop function */
+		assert(loop_osl_errno);
+		osl_errno = loop_osl_errno;
+	}
+	return loop_ret;
+}
+
+static int adu_loop_reverse(struct osl_table *t, unsigned col_num, void *private_data,
+		osl_rbtree_loop_func *func, int *loop_ret, int *loop_osl_errno)
+{
+	int ret = osl(osl_rbtree_loop_reverse(t, col_num, private_data, func));
+	return check_loop_return(ret, *loop_ret, *loop_osl_errno);
+}
+
+static int print_user_stats(void)
+{
+	struct user_info *ui;
+	int ret;
 
-	printf("%u\t%llu%llu\n", (unsigned)uid, (long long unsigned)files,
-		(long long unsigned)bytes);
+	FOR_EACH_USER(ui) {
+		struct user_stats_info usi = {
+			.count = conf.limit_arg,
+			.ui = ui
+		};
+		if (!ui_used(ui) || !ui_admissible(ui))
+			continue;
+		usi.flags = USF_PRINT_DIRNAME | USF_PRINT_BYTES | USF_COMPUTE_SUMMARY;
+		printf("uid %u, by size%s:\n",
+			(unsigned) ui->uid, size_unit_buf);
+		ret = adu_loop_reverse(ui->table, UT_BYTES, &usi, user_stats_loop_function,
+			&usi.ret, &usi.osl_errno);
+		if (ret < 0)
+			return ret;
+		printf("\nuid %u, by file count%s:\n",
+			(unsigned) ui->uid, count_unit_buf);
+		usi.count = conf.limit_arg,
+		usi.flags = USF_PRINT_DIRNAME | USF_PRINT_FILES;
+		ret = adu_loop_reverse(ui->table, UT_FILES, &usi, user_stats_loop_function,
+			&usi.ret, &usi.osl_errno);
+		if (ret < 0)
+			return ret;
+		printf("\n");
+	}
 	return 1;
 }
 
-struct id_dir_stat_info {
-	unsigned count;
-	struct osl_table *user_table;
-};
+static void print_global_summary(void)
+{
+	char d[FORMATED_VALUE_SIZE], f[FORMATED_VALUE_SIZE],
+		s[FORMATED_VALUE_SIZE];
+	enum enum_count_unit ud, uf;
+	enum enum_size_unit us;
+
+	ud = format_count_value(conf.count_unit_arg, num_dirs, 0, d);
+	uf = format_count_value(conf.count_unit_arg, num_files, 0, f);
+	us = format_size_value(conf.size_unit_arg, num_bytes, 0, s);
+
+	printf("Global summary "
+		"(dirs(%c)/files(%c)/size(%c))\n"
+		"\t%s\t%s\t%s\n\n",
+		count_unit_abbrevs[ud],
+		count_unit_abbrevs[uf],
+		size_unit_abbrevs[us],
+		d, f, s
+	);
+
+}
 
-static int print_big_dir(struct osl_row *row, void *data)
+static int print_statistics(void)
 {
-	struct id_dir_stat_info *info = data;
-	info->count++;
 	int ret;
-	struct osl_row *dir_row;
-	char *dirname;
-	uint64_t bytes;
-	struct osl_object obj;
+	struct global_stats_info gsi = {
+		.count = conf.limit_arg,
+		.flags = GSF_PRINT_DIRNAME | GSF_PRINT_BYTES | GSF_COMPUTE_SUMMARY
+	};
 
-	if (info->count > 10)
-		return -E_LOOP_COMPLETE;
-	ret = osl_get_object(info->user_table, row, UT_BYTES, &obj);
-	if (ret < 0)
-		return ret;
-	bytes = *(uint64_t *)obj.data;
-	ret = osl_get_object(info->user_table, row, UT_DIR_NUM, &obj);
+	printf("By size%s:\n",
+		size_unit_buf);
+	ret = adu_loop_reverse(dir_table, DT_BYTES, &gsi,
+		global_stats_loop_function, &gsi.ret, &gsi.osl_errno);
 	if (ret < 0)
 		return ret;
-	ret = osl_get_row(dir_table, DT_NUM, &obj, &dir_row);
-	if (ret < 0)
-		return ret;
-	ret = osl_get_object(dir_table, dir_row, DT_NAME, &obj);
+	printf("\n");
+
+	gsi.count = conf.limit_arg;
+	gsi.flags = GSF_PRINT_DIRNAME | GSF_PRINT_FILES;
+	printf("By file count%s:\n",
+		count_unit_buf);
+	ret = adu_loop_reverse(dir_table, DT_FILES, &gsi,
+		global_stats_loop_function, &gsi.ret, &gsi.osl_errno);
 	if (ret < 0)
 		return ret;
-	dirname = obj.data;
-	printf("%s: %llu\n", dirname, (long long unsigned)bytes);
+	printf("\n");
+	print_global_summary();
+	print_user_stats();
+	print_id_stats();
 	return 1;
 }
 
-static int print_id_dir_stats(struct osl_row *row, __a_unused void *data)
+static char *get_uid_list_name(void)
 {
-	struct osl_object obj;
-	uint32_t uid;
-	int ret = osl_get_object(id_table, row, IDT_UID, &obj);
-	struct id_dir_stat_info info = {.count = 0};
+	return make_message("%s/uid_list", conf.database_dir_arg);
+}
+
+static int write_uid_list(void)
+{
+	char *buf, *filename;
+	uint32_t count = 0;
+	struct user_info *ui;
+	size_t size = num_uids * sizeof(uint32_t);
+	int ret;
+
+	if (!num_uids)
+		return 0;
+	buf = para_malloc(size);
+	FOR_EACH_USER(ui) {
+		if (!ui_used(ui) || !ui_admissible(ui))
+			continue;
+		DEBUG_LOG("saving uid %u\n", (unsigned) ui->uid);
+		write_u32(buf + count++ * sizeof(uint32_t), ui->uid);
+	}
+	filename = get_uid_list_name();
+	ret = para_write_file(filename, buf, size);
+	free(filename);
+	free(buf);
+	return ret;
+}
+
+static int open_dir_table(void)
+{
+	if (!dir_table_desc.dir) /* we did not create the table */
+		dir_table_desc.dir = para_strdup(conf.database_dir_arg);
+	return osl(osl_open_table(&dir_table_desc, &dir_table));
+}
+static int com_create()
+{
+	uint64_t zero = 0ULL;
+	int ret = create_tables();
 
 	if (ret < 0)
 		return ret;
-	uid = *(uint32_t *)obj.data;
-
-	ret = osl_get_object(id_table, row, IDT_TABLE, &obj);
+	check_signals();
+	ret = open_dir_table();
 	if (ret < 0)
 		return ret;
-	info.user_table = *(struct osl_table **)obj.data;
+	check_signals();
+	ret = scan_dir(conf.base_dir_arg, &zero);
+	if (ret < 0)
+		goto out;
+	ret = write_uid_list();
+out:
+	close_all_tables();
+	return ret;
+}
 
-	printf("************************* Big dirs owned by uid %u\n", (unsigned) uid);
-	osl_rbtree_loop_reverse(info.user_table, IDT_BYTES, &info, print_big_dir);
-	return 1;
+static int read_uid_file(void)
+{
+	size_t size;
+	uint32_t n;
+	char *filename = get_uid_list_name(), *map;
+	int ret = mmap_full_file(filename, O_RDONLY, (void **)&map, &size, NULL);
+
+	if (ret < 0) {
+		INFO_LOG("failed to map %s\n", filename);
+		free(filename);
+		return ret;
+	}
+	num_uids = size / 4;
+	INFO_LOG("found %u uids in %s\n", (unsigned)num_uids, filename);
+	free(filename);
+	/* hash table size should be a power of two and larger than the number of uids */
+	uid_hash_table_size = 4;
+	while (uid_hash_table_size < num_uids)
+		uid_hash_table_size *= 2;
+	create_hash_table();
+	for (n = 0; n < num_uids; n++) {
+		uint32_t uid = read_u32(map + n * sizeof(uid));
+		ret = search_uid(uid, OPEN_USER_TABLE, NULL);
+		if (ret < 0)
+			goto out;
+	}
+out:
+	para_munmap(map, size);
+	return ret;
 }
 
-static int print_statistics(void)
+static int com_select(void)
 {
-	unsigned count = 0;
 	int ret;
 
-	printf("************************* Biggest dirs\n");
-	ret = osl_rbtree_loop_reverse(dir_table, DT_BYTES, &count, print_dirname_and_size);
-	if (ret < 0 && ret != -E_LOOP_COMPLETE)
-		return ret;
-	count = 0;
-	printf("************************* dirs containing many files\n");
-	ret = osl_rbtree_loop_reverse(dir_table, DT_FILES, &count, print_dirname_and_file_count);
-	if (ret < 0 && ret != -E_LOOP_COMPLETE)
-		return ret;
+	if (conf.count_unit_arg != count_unit_arg_h)
+		count_unit_buf[1] = count_unit_abbrevs[conf.count_unit_arg];
+	else
+		count_unit_buf[0] = '\0';
+	if (conf.size_unit_arg != size_unit_arg_h)
+		size_unit_buf[1] = size_unit_abbrevs[conf.size_unit_arg];
+	else
+		size_unit_buf[0] = '\0';
 
-	printf("************************* dirs stats by owner\n");
-	ret = osl_rbtree_loop(id_table, IDT_BYTES, NULL, print_id_stats);
+	ret = open_dir_table();
 	if (ret < 0)
 		return ret;
-
-	return osl_rbtree_loop(id_table, IDT_BYTES, NULL, print_id_dir_stats);
+	check_signals();
+	ret = read_uid_file();
+	if (ret < 0)
+		return ret;
+	check_signals();
+	ret = print_statistics();
+	close_all_tables();
+	return ret;
 }
 
+static int check_args(void)
+{
+	int i, ret;
+
+	/* remove trailing slashes from base-dir arg */
+	if (conf.base_dir_given) {
+		size_t len = strlen(conf.base_dir_arg);
+		for (;;) {
+			if (!len) /* empty string */
+				return -ERRNO_TO_ERROR(EINVAL);
+			if (!--len) /* length 1 is always OK */
+				break;
+			if (conf.base_dir_arg[len] != '/')
+				break; /* no trailing slash, also OK */
+			conf.base_dir_arg[len] = '\0';
+		}
+	}
+	if (!conf.uid_given)
+		return 0;
+	admissible_uids = para_malloc(conf.uid_given * sizeof(*admissible_uids));
+	for (i = 0; i < conf.uid_given; i++) {
+		ret = parse_uid_range(conf.uid_arg[i], admissible_uids + i);
+		if (ret < 0)
+			goto err;
+	}
+	return 1;
+err:
+	free(admissible_uids);
+	admissible_uids = NULL;
+	return ret;
+}
 
 int main(int argc, char **argv)
 {
-	int ret = create_tables();
-	if (ret < 0)
-		goto out;
-	ret = osl_open_table(&dir_table_desc, &dir_table);
+	int ret;
+	struct cmdline_parser_params params = {
+		.override = 0,
+		.initialize = 1,
+		.check_required = 0,
+		.check_ambiguity = 0,
+		.print_errors = 1
+	};
+
+	cmdline_parser_ext(argc, argv, &conf, &params); /* aborts on errors */
+	ret = init_signals();
 	if (ret < 0)
 		goto out;
-	ret = osl_open_table(&id_table_desc, &id_table);
+	ret = check_args();
 	if (ret < 0)
 		goto out;
 	ret = -E_SYNTAX;
-	if (argc != 2)
-		goto out;
-	ret = scan_dir(argv[1]);
+	if (conf.select_given)
+		ret = com_select();
+	else
+		ret = com_create();
 	if (ret < 0)
 		goto out;
-	print_statistics();
 out:
+	free(admissible_uids);
 	if (ret < 0) {
-		ERROR_LOG("%s\n", error_txt(-ret));
+		ERROR_LOG("%s\n", adu_strerror(-ret));
 		return -EXIT_FAILURE;
 	}
 	return EXIT_SUCCESS;
 }
-