2 * Copyright (C) 2007-2009 Andre Noll <maan@systemlinux.org>
4 * Licensed under the GPL v2. For licencing details see COPYING.
7 /** \file osl.c Object storage layer functions. */
8 #include <dirent.h> /* readdir() */
16 /* Taken from Drepper: How to write shared libraries, Appendix B. */
18 #define MSGSTRFIELD(line) MSGSTRFIELD1(line)
19 #define MSGSTRFIELD1(line) str##line
20 static const union msgstr_t {
22 #define OSL_ERROR(n, s) char MSGSTRFIELD(__LINE__)[sizeof(s)];
28 #define OSL_ERROR(n, s) s,
32 static const unsigned int errmsgidx[] = {
33 #define OSL_ERROR(n, s) [n] = offsetof(union msgstr_t, MSGSTRFIELD(__LINE__)),
38 __export const char *osl_strerror(int num)
40 return msgstr.str + errmsgidx[num];
45 static void __attribute ((constructor)) init_loglevel(void)
47 char *p = getenv("OSL_LOGLEVEL");
49 /* don't log anything if unset */
50 loglevel = p? atoi(p) : EMERG + 1;
57 * \param fmt Usual format string.
59 * All XXX_LOG() macros use this function.
61 __printf_2_3 void __log(int ll, const char* fmt,...)
74 strftime(str, sizeof(str), "%b %d %H:%M:%S", tm);
75 fprintf(outfd, "%s ", str);
77 vfprintf(outfd, fmt, argp);
82 * A wrapper for lseek(2).
84 * \param fd The file descriptor whose offset is to be to repositioned.
85 * \param offset A value-result parameter.
86 * \param whence Usual repositioning directive.
88 * Reposition the offset of the file descriptor \a fd to the argument \a offset
89 * according to the directive \a whence. Upon successful return, \a offset
90 * contains the resulting offset location as measured in bytes from the
91 * beginning of the file.
93 * \return Positive on success. Otherwise, the function returns \p -E_OSL_LSEEK.
97 static int __lseek(int fd, off_t *offset, int whence)
99 *offset = lseek(fd, *offset, whence);
100 int ret = -E_OSL_LSEEK;
106 static int append_file(const char *filename, char *data, size_t data_size,
111 // DEBUG_LOG("appending %zu + %zu bytes\n", header_size, data_size);
112 ret = osl_open(filename, O_WRONLY | O_CREAT | O_APPEND, 0644);
116 ret = write_all(fd, data, &data_size);
121 ret = __lseek(fd, &offset, SEEK_END);
124 // DEBUG_LOG("new file size: " FMT_OFF_T "\n", offset);
133 static int verify_name(const char *name)
136 return -E_OSL_BAD_NAME;
138 return -E_OSL_BAD_NAME;
139 if (strchr(name, '/'))
140 return -E_OSL_BAD_NAME;
141 if (!strcmp(name, ".."))
142 return -E_OSL_BAD_NAME;
143 if (!strcmp(name, "."))
144 return -E_OSL_BAD_NAME;
148 static char *disk_storage_dirname(const struct osl_table *t, unsigned col_num,
151 char *dirname, *column_name = column_filename(t, col_num);
155 if (!(t->desc->flags & OSL_LARGE_TABLE))
157 dirname = make_message("%s/%.2s", column_name, ds_name);
162 static char *disk_storage_name_of_object(const struct osl_table *t,
163 const struct osl_object *obj)
165 HASH_TYPE hash[HASH_SIZE];
166 hash_object(obj, hash);
167 return disk_storage_name_of_hash(t, hash);
170 static int disk_storage_name_of_row(const struct osl_table *t,
171 const struct osl_row *row, char **name)
173 struct osl_object obj;
174 int ret = osl_get_object(t, row, t->disk_storage_name_column, &obj);
178 *name = disk_storage_name_of_object(t, &obj);
184 static void column_name_hash(const char *col_name, HASH_TYPE *hash)
186 hash_function(col_name, strlen(col_name), hash);
189 static int init_column_descriptions(struct osl_table *t)
192 const struct osl_column_description *cd;
194 ret = verify_name(t->desc->name);
197 ret = -E_OSL_BAD_DB_DIR;
198 if (!t->desc->dir && (t->num_disk_storage_columns || t->num_mapped_columns))
200 /* the size of the index header without column descriptions */
201 t->index_header_size = IDX_COLUMN_DESCRIPTIONS;
202 FOR_EACH_COLUMN(i, t->desc, cd) {
203 struct osl_column *col = t->columns + i;
204 if (cd->storage_flags & OSL_RBTREE) {
205 if (!cd->compare_function)
206 return -E_OSL_NO_COMPARE_FUNC;
208 if (cd->storage_type == OSL_NO_STORAGE)
210 ret = -E_OSL_NO_COLUMN_NAME;
211 if (!cd->name || !cd->name[0])
213 ret = verify_name(cd->name);
216 t->index_header_size += index_column_description_size(cd->name);
217 column_name_hash(cd->name, col->name_hash);
218 ret = -E_OSL_DUPLICATE_COL_NAME;
219 for (j = i + 1; j < t->desc->num_columns; j++) {
220 const char *name2 = get_column_description(t->desc,
222 if (cd->name && name2 && !strcmp(cd->name, name2))
232 * Initialize a struct table from given table description.
234 * \param desc The description of the osl table.
235 * \param table_ptr Result is returned here.
237 * This function performs several sanity checks on \p desc and returns if any
238 * of these tests fail. On success, a struct \p osl_table is allocated and
239 * initialized with data derived from \p desc.
243 * \sa struct osl_table.
245 int init_table_structure(const struct osl_table_description *desc,
246 struct osl_table **table_ptr)
248 const struct osl_column_description *cd;
249 struct osl_table *t = calloc(1, sizeof(*t));
250 int i, ret = -E_OSL_NOMEM, have_disk_storage_name_column = 0;
254 ret = -E_OSL_BAD_TABLE_DESC;
257 DEBUG_LOG("creating table structure for '%s' from table "
258 "description\n", desc->name);
259 ret = -E_OSL_NO_COLUMN_DESC;
260 if (!desc->column_descriptions)
262 ret = -E_OSL_NO_COLUMNS;
263 if (!desc->num_columns)
266 t->columns = calloc(desc->num_columns, sizeof(struct osl_column));
270 FOR_EACH_COLUMN(i, t->desc, cd) {
271 enum osl_storage_type st = cd->storage_type;
272 enum osl_storage_flags sf = cd->storage_flags;
273 struct osl_column *col = &t->columns[i];
275 ret = -E_OSL_BAD_STORAGE_TYPE;
276 if (st != OSL_MAPPED_STORAGE && st != OSL_DISK_STORAGE
277 && st != OSL_NO_STORAGE)
279 ret = -E_OSL_BAD_STORAGE_FLAGS;
280 if (st == OSL_DISK_STORAGE && sf & OSL_RBTREE)
282 ret = -E_OSL_BAD_STORAGE_SIZE;
283 if (sf & OSL_FIXED_SIZE && !cd->data_size)
286 case OSL_DISK_STORAGE:
287 t->num_disk_storage_columns++;
289 case OSL_MAPPED_STORAGE:
290 t->num_mapped_columns++;
291 col->index_offset = t->row_index_size;
292 t->row_index_size += 8;
295 col->volatile_num = t->num_volatile_columns;
296 t->num_volatile_columns++;
299 if (sf & OSL_RBTREE) {
300 col->rbtree_num = t->num_rbtrees;
302 if ((sf & OSL_UNIQUE) && (st == OSL_MAPPED_STORAGE)) {
303 if (!have_disk_storage_name_column)
304 t->disk_storage_name_column = i;
305 have_disk_storage_name_column = 1;
309 ret = -E_OSL_NO_UNIQUE_RBTREE_COLUMN;
310 if (t->num_disk_storage_columns && !have_disk_storage_name_column)
312 ret = -E_OSL_NO_RBTREE_COL;
316 DEBUG_LOG("OK. Index entry size: %u\n", t->row_index_size);
317 ret = init_column_descriptions(t);
329 * Read the table description from index header.
331 * \param map The memory mapping of the index file.
332 * \param desc The values found in the index header are returned here.
334 * Read the index header, check for the osl magic string and the table version
335 * number. Read all information stored in the index header into \a desc.
339 * \sa struct osl_table_description, osl_create_table.
341 int read_table_desc(struct osl_object *map, struct osl_table_description *desc)
343 char *buf = map->data;
345 uint16_t header_size;
348 struct osl_column_description *cd;
350 if (map->size < MIN_INDEX_HEADER_SIZE(1))
351 return -E_OSL_SHORT_TABLE;
352 if (strncmp(buf + IDX_OSL_MAGIC, OSL_MAGIC, strlen(OSL_MAGIC)))
353 return -E_OSL_NO_MAGIC;
354 version = read_u8(buf + IDX_VERSION);
355 if (version < MIN_TABLE_VERSION || version > MAX_TABLE_VERSION)
356 return -E_OSL_VERSION_MISMATCH;
357 desc->flags = read_u8(buf + IDX_TABLE_FLAGS);
358 desc->num_columns = read_u16(buf + IDX_NUM_COLUMNS);
359 INFO_LOG("%u columns\n", desc->num_columns);
360 if (!desc->num_columns)
361 return -E_OSL_NO_COLUMNS;
362 header_size = read_u16(buf + IDX_HEADER_SIZE);
363 if (map->size < header_size)
364 return -E_OSL_BAD_SIZE;
365 desc->column_descriptions = calloc(desc->num_columns,
366 sizeof(struct osl_column_description));
367 if (!desc->column_descriptions)
369 offset = IDX_COLUMN_DESCRIPTIONS;
370 FOR_EACH_COLUMN(i, desc, cd) {
373 ret = -E_OSL_SHORT_TABLE;
374 if (map->size < offset + MIN_IDX_COLUMN_DESCRIPTION_SIZE) {
375 ERROR_LOG("map size = %zu < %u = offset + min desc size\n",
376 map->size, offset + MIN_IDX_COLUMN_DESCRIPTION_SIZE);
379 cd->storage_type = read_u16(buf + offset + IDX_CD_STORAGE_TYPE);
380 cd->storage_flags = read_u16(buf + offset +
381 IDX_CD_STORAGE_FLAGS);
382 cd->data_size = read_u32(buf + offset + IDX_CD_DATA_SIZE);
383 null_byte = memchr(buf + offset + IDX_CD_NAME, '\0',
384 map->size - offset - IDX_CD_NAME);
385 ret = -E_OSL_INDEX_CORRUPTION;
389 cd->name = strdup(buf + offset + IDX_CD_NAME);
392 offset += index_column_description_size(cd->name);
394 if (offset != header_size) {
395 ret = -E_OSL_INDEX_CORRUPTION;
396 ERROR_LOG("real header size = %u != %u = stored header size\n",
397 offset, header_size);
402 FOR_EACH_COLUMN(i, desc, cd)
408 * check whether the table description given by \p t->desc matches the on-disk
409 * table structure stored in the index of \a t.
411 static int compare_table_descriptions(struct osl_table *t)
414 struct osl_table_description desc;
415 const struct osl_column_description *cd1, *cd2;
417 /* read the on-disk structure into desc */
418 ret = read_table_desc(&t->index_map, &desc);
421 ret = -E_OSL_BAD_TABLE_FLAGS;
422 if (desc.flags != t->desc->flags)
424 ret = -E_OSL_BAD_COLUMN_NUM;
425 if (desc.num_columns > t->desc->num_columns)
427 if (desc.num_columns < t->desc->num_columns) {
428 struct osl_column_description *cd;
429 unsigned diff = t->desc->num_columns - desc.num_columns;
430 INFO_LOG("extending table by %u volatile columns\n", diff);
432 desc.column_descriptions = realloc(desc.column_descriptions,
433 t->desc->num_columns * sizeof(struct osl_column_description));
434 if (!desc.column_descriptions)
436 for (i = desc.num_columns; i < t->desc->num_columns; i++) {
437 cd = get_column_description(&desc, i);
438 cd->storage_type = OSL_NO_STORAGE;
441 desc.num_columns += diff;
443 FOR_EACH_COLUMN(i, t->desc, cd1) {
444 cd2 = get_column_description(&desc, i);
445 ret = -E_OSL_BAD_STORAGE_TYPE;
446 if (cd1->storage_type != cd2->storage_type)
448 if (cd1->storage_type == OSL_NO_STORAGE)
450 ret = -E_OSL_BAD_STORAGE_FLAGS;
451 if (cd1->storage_flags != cd2->storage_flags) {
452 ERROR_LOG("sf1 = %u != %u = sf2\n",
453 cd1->storage_flags, cd2->storage_flags);
456 ret = -E_OSL_BAD_DATA_SIZE;
457 if (cd1->storage_flags & OSL_FIXED_SIZE)
458 if (cd1->data_size != cd2->data_size)
460 ret = -E_OSL_BAD_COLUMN_NAME;
461 if (strcmp(cd1->name, cd2->name))
464 INFO_LOG("table description of '%s' matches on-disk data, good\n",
468 FOR_EACH_COLUMN(i, &desc, cd1)
470 free(desc.column_descriptions);
474 static int create_table_index(struct osl_table *t)
476 char *buf, *filename;
478 size_t size = t->index_header_size;
479 const struct osl_column_description *cd;
482 INFO_LOG("creating %zu byte index for table %s\n", size,
484 buf = calloc(1, size);
487 sprintf(buf + IDX_OSL_MAGIC, "%s", OSL_MAGIC);
488 write_u8(buf + IDX_TABLE_FLAGS, t->desc->flags);
489 write_u8(buf + IDX_DIRTY_FLAG, 0);
490 write_u8(buf + IDX_VERSION, CURRENT_TABLE_VERSION);
491 write_u16(buf + IDX_NUM_COLUMNS, t->num_mapped_columns + t->num_disk_storage_columns);
492 write_u16(buf + IDX_HEADER_SIZE, t->index_header_size);
493 offset = IDX_COLUMN_DESCRIPTIONS;
494 FOR_EACH_COLUMN(i, t->desc, cd) {
495 /* no need to store info about volatile storage */
496 if (cd->storage_type == OSL_NO_STORAGE)
498 write_u16(buf + offset + IDX_CD_STORAGE_TYPE,
500 write_u16(buf + offset + IDX_CD_STORAGE_FLAGS,
502 if (cd->storage_flags & OSL_FIXED_SIZE)
503 write_u32(buf + offset + IDX_CD_DATA_SIZE,
505 strcpy(buf + offset + IDX_CD_NAME, cd->name);
506 offset += index_column_description_size(cd->name);
508 assert(offset = size);
509 filename = index_filename(t->desc);
511 ret = write_file(filename, buf, size);
519 __export int osl_create_table(const struct osl_table_description *desc)
521 const struct osl_column_description *cd;
522 char *table_dir = NULL, *filename;
524 int i, ret = init_table_structure(desc, &t);
528 INFO_LOG("creating %s\n", desc->name);
529 FOR_EACH_COLUMN(i, t->desc, cd) {
530 if (cd->storage_type == OSL_NO_STORAGE)
533 ret = osl_mkdir(desc->dir, 0777);
534 if (ret < 0 && ret != -E_OSL_DIR_EXISTS)
536 table_dir = make_message("%s/%s", desc->dir,
541 ret = osl_mkdir(table_dir, 0777);
546 filename = column_filename(t, i);
549 INFO_LOG("filename: %s\n", filename);
550 if (cd->storage_type == OSL_MAPPED_STORAGE) {
551 ret = osl_open(filename, O_RDWR | O_CREAT | O_EXCL,
560 ret = osl_mkdir(filename, 0777);
565 if (t->num_mapped_columns) {
566 ret = create_table_index(t);
578 static int table_is_dirty(struct osl_table *t)
580 char *buf = (char *)t->index_map.data + IDX_DIRTY_FLAG;
581 uint8_t dirty = read_u8(buf) & 0x1;
585 static void mark_table_dirty(struct osl_table *t)
587 char *buf = (char *)t->index_map.data + IDX_DIRTY_FLAG;
588 write_u8(buf, read_u8(buf) | 1);
591 static void mark_table_clean(struct osl_table *t)
593 char *buf = (char *)t->index_map.data + IDX_DIRTY_FLAG;
594 write_u8(buf, read_u8(buf) & 0xfe);
597 static void unmap_column(struct osl_table *t, unsigned col_num)
599 struct osl_object map = t->columns[col_num].data_map;
603 ret = osl_munmap(map.data, map.size);
609 * Unmap all mapped files of an osl table.
611 * \param t Pointer to a mapped table.
612 * \param flags Options for unmapping.
614 * \return Positive on success, negative on errors.
616 * \sa map_table(), enum osl_close_flags, osl_munmap().
618 int unmap_table(struct osl_table *t, enum osl_close_flags flags)
621 const struct osl_column_description *cd;
624 if (!t->num_mapped_columns) /* can this ever happen? */
626 INFO_LOG("unmapping table '%s'\n", t->desc->name);
627 if (!t->index_map.data)
628 return -E_OSL_NOT_MAPPED;
629 if (flags & OSL_MARK_CLEAN)
631 ret = osl_munmap(t->index_map.data, t->index_map.size);
634 t->index_map.data = NULL;
637 FOR_EACH_MAPPED_COLUMN(i, t, cd)
642 static int map_column(struct osl_table *t, unsigned col_num)
645 char *filename = column_filename(t, col_num);
650 ret = osl_stat(filename, &statbuf);
655 if (!(S_IFREG & statbuf.st_mode)) {
659 ret = mmap_full_file(filename, O_RDWR,
660 &t->columns[col_num].data_map.data,
661 &t->columns[col_num].data_map.size,
668 * Map the index file and all columns of type \p OSL_MAPPED_STORAGE into memory.
670 * \param t Pointer to an initialized table structure.
671 * \param flags Mapping options.
673 * \return Negative return value on errors; on success the number of rows
674 * (including invalid rows) is returned.
676 * \sa unmap_table(), enum map_table_flags, osl_open_table(), mmap(2).
678 int map_table(struct osl_table *t, enum map_table_flags flags)
681 const struct osl_column_description *cd;
682 int i = 0, ret, num_rows = 0;
684 if (!t->num_mapped_columns)
686 if (t->index_map.data)
687 return -E_OSL_ALREADY_MAPPED;
688 filename = index_filename(t->desc);
691 INFO_LOG("mapping table '%s' (index: %s)\n", t->desc->name, filename);
692 ret = mmap_full_file(filename, flags & MAP_TBL_FL_MAP_RDONLY?
693 O_RDONLY : O_RDWR, &t->index_map.data, &t->index_map.size, NULL);
697 if (flags & MAP_TBL_FL_VERIFY_INDEX) {
698 ret = compare_table_descriptions(t);
703 if (!(flags & MAP_TBL_FL_IGNORE_DIRTY)) {
704 if (table_is_dirty(t)) {
705 ERROR_LOG("%s is dirty\n", t->desc->name);
710 num_rows = table_num_rows(t);
714 FOR_EACH_MAPPED_COLUMN(i, t, cd) {
715 ret = map_column(t, i);
720 err: /* unmap what is already mapped */
721 for (i--; i >= 0; i--) {
722 struct osl_object map = t->columns[i].data_map;
723 osl_munmap(map.data, map.size);
726 osl_munmap(t->index_map.data, t->index_map.size);
727 t->index_map.data = NULL;
732 * Retrieve a mapped object by row and column number.
734 * \param t Pointer to an open osl table.
735 * \param col_num Number of the mapped column containing the object to retrieve.
736 * \param row_num Number of the row containing the object to retrieve.
737 * \param obj The result is returned here.
739 * It is considered an error if \a col_num does not refer to a column
740 * of storage type \p OSL_MAPPED_STORAGE.
744 * \sa osl_storage_type.
746 int get_mapped_object(const struct osl_table *t, unsigned col_num,
747 uint32_t row_num, struct osl_object *obj)
749 struct osl_column *col = &t->columns[col_num];
754 if (t->num_rows <= row_num)
755 return -E_OSL_BAD_ROW_NUM;
756 ret = get_cell_index(t, row_num, col_num, &cell_index);
759 offset = read_u32(cell_index);
760 obj->size = read_u32(cell_index + 4);
761 obj->data = col->data_map.data + offset;
766 * It's OK to call this with result = rb_node = NULL. If result is not NULL,
767 * and rb key was not found, result points to the parent node.
769 static int search_rbtree(const struct osl_object *obj,
770 const struct osl_table *t, unsigned col_num,
771 struct rb_node **result, struct rb_node ***rb_link)
773 struct osl_column *col = &t->columns[col_num];
774 struct rb_node **new = &col->rbtree.rb_node, *parent = NULL;
775 const struct osl_column_description *cd =
776 get_column_description(t->desc, col_num);
777 enum osl_storage_type st = cd->storage_type;
779 struct osl_row *this_row = get_row_pointer(*new,
782 struct osl_object this_obj;
784 if (st == OSL_MAPPED_STORAGE) {
785 ret = get_mapped_object(t, col_num, this_row->num,
790 this_obj = this_row->volatile_objects[col->volatile_num];
791 ret = cd->compare_function(obj, &this_obj);
794 *result = get_rb_node_pointer(this_row,
799 new = &((*new)->rb_left);
801 new = &((*new)->rb_right);
807 return -E_OSL_RB_KEY_NOT_FOUND;
810 static int insert_rbtree(struct osl_table *t, unsigned col_num,
811 const struct osl_row *row, const struct osl_object *obj)
813 struct rb_node *parent, **rb_link;
816 int ret = search_rbtree(obj, t, col_num, &parent, &rb_link);
819 return -E_OSL_RB_KEY_EXISTS;
820 rbtree_num = t->columns[col_num].rbtree_num;
821 n = get_rb_node_pointer(row, rbtree_num);
822 rb_link_node(n, parent, rb_link);
823 rb_insert_color(n, &t->columns[col_num].rbtree);
827 static void remove_rb_node(struct osl_table *t, unsigned col_num,
828 const struct osl_row *row)
830 struct osl_column *col = &t->columns[col_num];
831 const struct osl_column_description *cd =
832 get_column_description(t->desc, col_num);
833 enum osl_storage_flags sf = cd->storage_flags;
834 struct rb_node *victim, *splice_out_node, *tmp;
835 if (!(sf & OSL_RBTREE))
838 * Which node is removed/spliced out actually depends on how many
839 * children the victim node has: If it has no children, it gets
840 * deleted. If it has one child, it gets spliced out. If it has two
841 * children, its successor (which has at most a right child) gets
844 victim = get_rb_node_pointer(row, col->rbtree_num);
845 if (victim->rb_left && victim->rb_right)
846 splice_out_node = rb_next(victim);
848 splice_out_node = victim;
849 /* Go up to the root and decrement the size of each node in the path. */
850 for (tmp = splice_out_node; tmp; tmp = rb_parent(tmp))
852 rb_erase(victim, &col->rbtree);
855 static int add_row_to_rbtrees(struct osl_table *t, uint32_t row_num,
856 struct osl_object *volatile_objs, struct osl_row **row_ptr)
860 struct osl_row *row = allocate_row(t->num_rbtrees);
861 const struct osl_column_description *cd;
866 row->volatile_objects = volatile_objs;
867 FOR_EACH_RBTREE_COLUMN(i, t, cd) {
868 if (cd->storage_type == OSL_MAPPED_STORAGE) {
869 struct osl_object obj;
870 ret = get_mapped_object(t, i, row_num, &obj);
873 ret = insert_rbtree(t, i, row, &obj);
874 } else { /* volatile */
875 const struct osl_object *obj
876 = volatile_objs + t->columns[i].volatile_num;
877 ret = insert_rbtree(t, i, row, obj);
885 err: /* rollback changes, i.e. remove added entries from rbtrees */
887 remove_rb_node(t, i--, row);
892 static void free_volatile_objects(const struct osl_table *t,
893 enum osl_close_flags flags)
897 struct osl_column *rb_col;
898 const struct osl_column_description *cd;
900 if (!t->num_volatile_columns)
902 /* find the first rbtree column (any will do) */
903 FOR_EACH_RBTREE_COLUMN(i, t, cd)
905 rb_col = t->columns + i;
906 /* walk that rbtree and free all volatile objects */
907 for (n = rb_first(&rb_col->rbtree); n; n = rb_next(n)) {
908 struct osl_row *r = get_row_pointer(n, rb_col->rbtree_num);
909 if (flags & OSL_FREE_VOLATILE)
910 FOR_EACH_VOLATILE_COLUMN(j, t, cd) {
911 if (cd->storage_flags & OSL_DONT_FREE)
913 free(r->volatile_objects[
914 t->columns[j].volatile_num].data);
916 // for (j = 0; j < t->num_volatile_columns; j++)
917 // free(r->volatile_objects[j].data);
918 free(r->volatile_objects);
923 * Erase all rbtree nodes and free resources.
925 * \param t Pointer to an open osl table.
927 * This function is called by osl_close_table().
929 void clear_rbtrees(struct osl_table *t)
931 const struct osl_column_description *cd;
932 unsigned i, rbtrees_cleared = 0;
934 FOR_EACH_RBTREE_COLUMN(i, t, cd) {
935 struct osl_column *col = &t->columns[i];
938 for (n = rb_first(&col->rbtree); n;) {
940 rb_erase(n, &col->rbtree);
941 if (rbtrees_cleared == t->num_rbtrees) {
942 r = get_row_pointer(n, col->rbtree_num);
952 __export int osl_close_table(struct osl_table *t, enum osl_close_flags flags)
957 return -E_OSL_BAD_TABLE;
958 NOTICE_LOG("closing table %s\n", t->desc->name);
959 free_volatile_objects(t, flags);
961 ret = unmap_table(t, flags);
963 ERROR_LOG("unmap_table failed: %d\n", ret);
970 * Find out whether the given row number corresponds to an invalid row.
972 * \param t Pointer to the osl table.
973 * \param row_num The number of the row in question.
975 * By definition, a row is considered invalid if all its index entries
978 * \return Positive if \a row_num corresponds to an invalid row,
979 * zero if it corresponds to a valid row, negative on errors.
981 int row_is_invalid(struct osl_table *t, uint32_t row_num)
984 int i, ret = get_row_index(t, row_num, &row_index);
988 for (i = 0; i < t->row_index_size; i++) {
989 if ((unsigned char)row_index[i] != 0xff)
992 INFO_LOG("row %d is invalid\n", row_num);
997 * Invalidate a row of an osl table.
999 * \param t Pointer to an open osl table.
1000 * \param row_num Number of the row to mark as invalid.
1002 * This function marks each mapped object in the index entry of \a row as
1007 int mark_row_invalid(struct osl_table *t, uint32_t row_num)
1010 int ret = get_row_index(t, row_num, &row_index);
1014 INFO_LOG("marking row %d as invalid\n", row_num);
1015 memset(row_index, 0xff, t->row_index_size);
1020 * Initialize all rbtrees and compute number of invalid rows.
1022 * \param t The table containing the rbtrees to be initialized.
1026 int init_rbtrees(struct osl_table *t)
1029 const struct osl_column_description *cd;
1031 /* create rbtrees */
1032 FOR_EACH_RBTREE_COLUMN(i, t, cd)
1033 t->columns[i].rbtree = RB_ROOT;
1034 /* add valid rows to rbtrees */
1035 t->num_invalid_rows = 0;
1036 for (i = 0; i < t->num_rows; i++) {
1037 ret = row_is_invalid(t, i);
1041 t->num_invalid_rows++;
1044 ret = add_row_to_rbtrees(t, i, NULL, NULL);
1051 __export int osl_open_table(const struct osl_table_description *table_desc,
1052 struct osl_table **result)
1055 struct osl_table *t;
1056 const struct osl_column_description *cd;
1058 NOTICE_LOG("opening table %s\n", table_desc->name);
1059 ret = init_table_structure(table_desc, &t);
1062 FOR_EACH_DISK_STORAGE_COLUMN(i, t, cd) {
1063 struct stat statbuf;
1064 char *dirname = column_filename(t, i);
1069 /* check if directory exists */
1070 ret = osl_stat(dirname, &statbuf);
1074 ret = -E_OSL_NOTDIR;
1075 if (!S_ISDIR(statbuf.st_mode))
1078 ret = map_table(t, MAP_TBL_FL_VERIFY_INDEX);
1082 DEBUG_LOG("num rows: %d\n", t->num_rows);
1083 ret = init_rbtrees(t);
1085 osl_close_table(t, OSL_MARK_CLEAN); /* ignore further errors */
1096 static int create_disk_storage_object_dir(const struct osl_table *t,
1097 unsigned col_num, const char *ds_name)
1102 if (!(t->desc->flags & OSL_LARGE_TABLE))
1104 dirname = disk_storage_dirname(t, col_num, ds_name);
1106 return -E_OSL_NOMEM;
1107 ret = osl_mkdir(dirname, 0777);
1109 if (ret < 0 && ret != -E_OSL_DIR_EXISTS)
1114 static int write_disk_storage_file(const struct osl_table *t, unsigned col_num,
1115 const struct osl_object *obj, const char *ds_name)
1120 ret = create_disk_storage_object_dir(t, col_num, ds_name);
1123 filename = disk_storage_path(t, col_num, ds_name);
1125 return -E_OSL_NOMEM;
1126 ret = write_file(filename, obj->data, obj->size);
1131 static int append_map_file(const struct osl_table *t, unsigned col_num,
1132 const struct osl_object *obj, uint32_t *new_size)
1134 char *filename = column_filename(t, col_num);
1138 return -E_OSL_NOMEM;
1139 ret = append_file(filename, obj->data, obj->size, new_size);
1144 static int append_row_index(const struct osl_table *t, char *row_index)
1149 if (!t->num_mapped_columns)
1151 filename = index_filename(t->desc);
1153 return -E_OSL_NOMEM;
1154 ret = append_file(filename, row_index, t->row_index_size, NULL);
1159 static int truncate_mapped_file(const struct osl_table *t, unsigned col_num,
1163 char *filename = column_filename(t, col_num);
1166 return -E_OSL_NOMEM;
1167 ret = truncate_file(filename, size);
1172 static int delete_disk_storage_file(const struct osl_table *t, unsigned col_num,
1173 const char *ds_name)
1175 char *dirname, *filename = disk_storage_path(t, col_num, ds_name);
1179 return -E_OSL_NOMEM;
1180 if (unlink(filename) < 0)
1181 ret = errno == ENOENT? -E_OSL_NOENT : -E_OSL_UNLINK;
1185 if (!(t->desc->flags & OSL_LARGE_TABLE))
1187 dirname = disk_storage_dirname(t, col_num, ds_name);
1189 return -E_OSL_NOMEM;
1195 __export int osl_add_and_get_row(struct osl_table *t, struct osl_object *objects,
1196 struct osl_row **row)
1199 char *ds_name = NULL;
1200 struct rb_node **rb_parents = NULL, ***rb_links = NULL;
1201 char *new_row_index = NULL;
1202 struct osl_object *volatile_objs = NULL;
1203 const struct osl_column_description *cd;
1206 return -E_OSL_BAD_TABLE;
1207 rb_parents = malloc(t->num_rbtrees * sizeof(struct rn_node*));
1209 return -E_OSL_NOMEM;
1210 rb_links = malloc(t->num_rbtrees * sizeof(struct rn_node**));
1213 return -E_OSL_NOMEM;
1215 if (t->num_mapped_columns) {
1216 new_row_index = malloc(t->row_index_size);
1217 if (!new_row_index) {
1220 return -E_OSL_NOMEM;
1223 /* pass 1: sanity checks */
1224 // DEBUG_LOG("sanity tests: %p:%p\n", objects[0].data,
1225 // objects[1].data);
1226 FOR_EACH_COLUMN(i, t->desc, cd) {
1227 enum osl_storage_type st = cd->storage_type;
1228 enum osl_storage_flags sf = cd->storage_flags;
1230 // ret = -E_OSL_NULL_OBJECT;
1233 if (st == OSL_DISK_STORAGE)
1235 if (sf & OSL_RBTREE) {
1236 unsigned rbtree_num = t->columns[i].rbtree_num;
1237 ret = -E_OSL_RB_KEY_EXISTS;
1238 // DEBUG_LOG("checking whether %p exists\n",
1239 // objects[i].data);
1240 if (search_rbtree(objects + i, t, i,
1241 &rb_parents[rbtree_num],
1242 &rb_links[rbtree_num]) > 0)
1245 if (sf & OSL_FIXED_SIZE) {
1246 // DEBUG_LOG("fixed size. need: %zu, have: %d\n",
1247 // objects[i].size, cd->data_size);
1248 ret = -E_OSL_BAD_DATA_SIZE;
1249 if (objects[i].size != cd->data_size)
1253 if (t->num_disk_storage_columns) {
1254 ds_name = disk_storage_name_of_object(t,
1255 &objects[t->disk_storage_name_column]);
1260 ret = unmap_table(t, OSL_MARK_CLEAN);
1263 // DEBUG_LOG("sanity tests passed%s\n", "");
1264 /* pass 2: create data files, append map data */
1265 FOR_EACH_COLUMN(i, t->desc, cd) {
1266 enum osl_storage_type st = cd->storage_type;
1267 if (st == OSL_NO_STORAGE)
1269 if (st == OSL_MAPPED_STORAGE) {
1271 struct osl_column *col = &t->columns[i];
1272 // DEBUG_LOG("appending object of size %zu\n",
1273 // objects[i].size);
1274 ret = append_map_file(t, i, objects + i, &new_size);
1277 update_cell_index(new_row_index, col, new_size,
1282 ret = write_disk_storage_file(t, i, objects + i, ds_name);
1286 ret = append_row_index(t, new_row_index);
1289 ret = map_table(t, MAP_TBL_FL_VERIFY_INDEX);
1290 if (ret < 0) { /* truncate index and rollback changes */
1291 char *filename = index_filename(t->desc);
1293 truncate_file(filename, t->row_index_size);
1297 /* pass 3: add entry to rbtrees */
1298 if (t->num_volatile_columns) {
1300 volatile_objs = calloc(t->num_volatile_columns,
1301 sizeof(struct osl_object));
1304 FOR_EACH_VOLATILE_COLUMN(i, t, cd)
1305 volatile_objs[t->columns[i].volatile_num] = objects[i];
1308 // DEBUG_LOG("adding new entry as row #%d\n", t->num_rows - 1);
1309 ret = add_row_to_rbtrees(t, t->num_rows - 1, volatile_objs, row);
1312 // DEBUG_LOG("added new entry as row #%d\n", t->num_rows - 1);
1315 rollback: /* rollback all changes made, ignore further errors */
1316 for (i--; i >= 0; i--) {
1317 cd = get_column_description(t->desc, i);
1318 enum osl_storage_type st = cd->storage_type;
1319 if (st == OSL_NO_STORAGE)
1322 if (st == OSL_MAPPED_STORAGE)
1323 truncate_mapped_file(t, i, objects[i].size);
1324 else /* disk storage */
1325 delete_disk_storage_file(t, i, ds_name);
1327 /* ignore error and return previous error value */
1328 map_table(t, MAP_TBL_FL_VERIFY_INDEX);
1330 free(new_row_index);
1337 __export int osl_add_row(struct osl_table *t, struct osl_object *objects)
1339 return osl_add_and_get_row(t, objects, NULL);
1342 __export int osl_get_object(const struct osl_table *t, const struct osl_row *r,
1343 unsigned col_num, struct osl_object *object)
1345 const struct osl_column_description *cd;
1348 return -E_OSL_BAD_TABLE;
1349 cd = get_column_description(t->desc, col_num);
1350 /* col must not be disk storage */
1351 if (cd->storage_type == OSL_DISK_STORAGE)
1352 return -E_OSL_BAD_STORAGE_TYPE;
1353 if (cd->storage_type == OSL_MAPPED_STORAGE)
1354 return get_mapped_object(t, col_num, r->num, object);
1356 *object = r->volatile_objects[t->columns[col_num].volatile_num];
1360 __export int osl_del_row(struct osl_table *t, struct osl_row *row)
1362 struct osl_row *r = row;
1364 const struct osl_column_description *cd;
1367 return -E_OSL_BAD_TABLE;
1368 INFO_LOG("deleting row %p\n", row);
1370 if (t->num_disk_storage_columns) {
1372 ret = disk_storage_name_of_row(t, r, &ds_name);
1375 FOR_EACH_DISK_STORAGE_COLUMN(i, t, cd)
1376 delete_disk_storage_file(t, i, ds_name);
1379 FOR_EACH_COLUMN(i, t->desc, cd) {
1380 struct osl_column *col = t->columns + i;
1381 enum osl_storage_type st = cd->storage_type;
1382 remove_rb_node(t, i, r);
1383 if (st == OSL_MAPPED_STORAGE)
1385 if (st == OSL_NO_STORAGE && !(cd->storage_flags & OSL_DONT_FREE))
1386 free(r->volatile_objects[col->volatile_num].data);
1388 if (t->num_mapped_columns) {
1389 ret = mark_row_invalid(t, r->num);
1392 t->num_invalid_rows++;
1397 free(r->volatile_objects);
1402 /* test if column has an rbtree */
1403 static int check_rbtree_col(const struct osl_table *t, unsigned col_num,
1404 struct osl_column **col)
1407 return -E_OSL_BAD_TABLE;
1408 if (!(get_column_description(t->desc, col_num)->storage_flags & OSL_RBTREE))
1409 return -E_OSL_BAD_STORAGE_FLAGS;
1410 *col = t->columns + col_num;
1414 __export int osl_get_row(const struct osl_table *t, unsigned col_num,
1415 const struct osl_object *obj, struct osl_row **result)
1418 struct rb_node *node;
1419 struct osl_row *row;
1420 struct osl_column *col;
1423 ret = check_rbtree_col(t, col_num, &col);
1426 ret = search_rbtree(obj, t, col_num, &node, NULL);
1429 row = get_row_pointer(node, t->columns[col_num].rbtree_num);
1434 static int rbtree_loop(struct osl_column *col, void *private_data,
1435 osl_rbtree_loop_func *func)
1437 struct rb_node *n, *tmp;
1439 /* this for-loop is safe against removal of an entry */
1440 for (n = rb_first(&col->rbtree), tmp = n? rb_next(n) : NULL;
1442 n = tmp, tmp = tmp? rb_next(tmp) : NULL) {
1443 struct osl_row *r = get_row_pointer(n, col->rbtree_num);
1444 if (func(r, private_data) < 0)
1450 static int rbtree_loop_reverse(struct osl_column *col, void *private_data,
1451 osl_rbtree_loop_func *func)
1453 struct rb_node *n, *tmp;
1455 /* safe against removal of an entry */
1456 for (n = rb_last(&col->rbtree), tmp = n? rb_prev(n) : NULL;
1458 n = tmp, tmp = tmp? rb_prev(tmp) : NULL) {
1459 struct osl_row *r = get_row_pointer(n, col->rbtree_num);
1460 if (func(r, private_data) < 0)
1466 __export int osl_rbtree_loop(const struct osl_table *t, unsigned col_num,
1467 void *private_data, osl_rbtree_loop_func *func)
1469 struct osl_column *col;
1471 int ret = check_rbtree_col(t, col_num, &col);
1474 return rbtree_loop(col, private_data, func);
1477 __export int osl_rbtree_loop_reverse(const struct osl_table *t, unsigned col_num,
1478 void *private_data, osl_rbtree_loop_func *func)
1480 struct osl_column *col;
1482 int ret = check_rbtree_col(t, col_num, &col);
1485 return rbtree_loop_reverse(col, private_data, func);
1488 /* TODO: Rollback changes on errors */
1489 static int rename_disk_storage_objects(struct osl_table *t,
1490 struct osl_object *old_obj, struct osl_object *new_obj)
1493 const struct osl_column_description *cd;
1494 char *old_ds_name, *new_ds_name;
1496 if (!t->num_disk_storage_columns)
1497 return 1; /* nothing to do */
1498 if (old_obj->size == new_obj->size && !memcmp(new_obj->data,
1499 old_obj->data, new_obj->size))
1500 return 1; /* object did not change */
1501 old_ds_name = disk_storage_name_of_object(t, old_obj);
1502 new_ds_name = disk_storage_name_of_object(t, new_obj);
1504 if (!old_ds_name || ! new_ds_name)
1507 FOR_EACH_DISK_STORAGE_COLUMN(i, t, cd) {
1508 char *old_filename, *new_filename;
1509 ret = create_disk_storage_object_dir(t, i, new_ds_name);
1512 old_filename = disk_storage_path(t, i, old_ds_name);
1513 new_filename = disk_storage_path(t, i, new_ds_name);
1514 if (!old_filename || !new_filename)
1517 ret = osl_rename(old_filename, new_filename);
1531 __export int osl_update_object(struct osl_table *t, const struct osl_row *r,
1532 unsigned col_num, struct osl_object *obj)
1534 struct osl_column *col;
1535 const struct osl_column_description *cd;
1539 return -E_OSL_BAD_TABLE;
1540 col = &t->columns[col_num];
1541 cd = get_column_description(t->desc, col_num);
1542 DEBUG_LOG("updating column %u of %s\n", col_num, t->desc->name);
1543 if (cd->storage_flags & OSL_RBTREE) {
1544 if (search_rbtree(obj, t, col_num, NULL, NULL) > 0)
1545 return -E_OSL_RB_KEY_EXISTS;
1547 if (cd->storage_flags & OSL_FIXED_SIZE) {
1548 if (obj->size != cd->data_size)
1549 return -E_OSL_BAD_DATA_SIZE;
1551 remove_rb_node(t, col_num, r);
1552 if (cd->storage_type == OSL_NO_STORAGE) { /* TODO: If fixed size, reuse object? */
1553 free(r->volatile_objects[col->volatile_num].data);
1554 r->volatile_objects[col->volatile_num] = *obj;
1555 } else if (cd->storage_type == OSL_DISK_STORAGE) {
1557 ret = disk_storage_name_of_row(t, r, &ds_name);
1560 ret = delete_disk_storage_file(t, col_num, ds_name);
1561 if (ret < 0 && ret != -E_OSL_NOENT) {
1565 ret = write_disk_storage_file(t, col_num, obj, ds_name);
1569 } else { /* mapped storage */
1570 struct osl_object old_obj;
1571 ret = get_mapped_object(t, col_num, r->num, &old_obj);
1575 * If the updated column is the disk storage name column, the
1576 * disk storage name changes, so we have to rename all disk
1577 * storage objects accordingly.
1579 if (col_num == t->disk_storage_name_column) {
1580 ret = rename_disk_storage_objects(t, &old_obj, obj);
1584 if (cd->storage_flags & OSL_FIXED_SIZE)
1585 memcpy(old_obj.data, obj->data, cd->data_size);
1586 else { /* TODO: if the size doesn't change, use old space */
1587 uint32_t new_data_map_size;
1589 ret = get_row_index(t, r->num, &row_index);
1592 unmap_column(t, col_num);
1593 ret = append_map_file(t, col_num, obj,
1594 &new_data_map_size);
1597 ret = map_column(t, col_num);
1600 update_cell_index(row_index, col, new_data_map_size,
1604 if (cd->storage_flags & OSL_RBTREE) {
1605 ret = insert_rbtree(t, col_num, r, obj);
1612 __export int osl_open_disk_object(const struct osl_table *t, const struct osl_row *r,
1613 unsigned col_num, struct osl_object *obj)
1615 const struct osl_column_description *cd;
1616 char *ds_name, *filename;
1620 return -E_OSL_BAD_TABLE;
1621 cd = get_column_description(t->desc, col_num);
1622 if (cd->storage_type != OSL_DISK_STORAGE)
1623 return -E_OSL_BAD_STORAGE_TYPE;
1625 ret = disk_storage_name_of_row(t, r, &ds_name);
1628 filename = disk_storage_path(t, col_num, ds_name);
1631 return -E_OSL_NOMEM;
1632 DEBUG_LOG("filename: %s\n", filename);
1633 ret = mmap_full_file(filename, O_RDONLY, &obj->data, &obj->size, NULL);
1638 __export int osl_close_disk_object(struct osl_object *obj)
1640 return osl_munmap(obj->data, obj->size);
1643 __export int osl_get_num_rows(const struct osl_table *t, unsigned *num_rows)
1646 return -E_OSL_BAD_TABLE;
1647 assert(t->num_rows >= t->num_invalid_rows);
1648 *num_rows = t->num_rows - t->num_invalid_rows;
1652 __export int osl_get_rank(const struct osl_table *t, struct osl_row *r,
1653 unsigned col_num, unsigned *rank)
1655 struct osl_object obj;
1656 struct osl_column *col;
1657 struct rb_node *node;
1658 int ret = check_rbtree_col(t, col_num, &col);
1662 ret = osl_get_object(t, r, col_num, &obj);
1665 ret = search_rbtree(&obj, t, col_num, &node, NULL);
1668 ret = rb_rank(node, rank);
1670 return -E_OSL_BAD_ROW;
1674 __export int osl_get_nth_row(const struct osl_table *t, unsigned col_num,
1675 unsigned n, struct osl_row **result)
1677 struct osl_column *col;
1678 struct rb_node *node;
1683 return -E_OSL_RB_KEY_NOT_FOUND;
1684 ret = osl_get_num_rows(t, &num_rows);
1688 return -E_OSL_RB_KEY_NOT_FOUND;
1689 ret = check_rbtree_col(t, col_num, &col);
1692 node = rb_nth(col->rbtree.rb_node, n);
1694 return -E_OSL_RB_KEY_NOT_FOUND;
1695 *result = get_row_pointer(node, col->rbtree_num);
1699 __export int osl_rbtree_first_row(const struct osl_table *t, unsigned col_num,
1700 struct osl_row **result)
1702 return osl_get_nth_row(t, col_num, 1, result);
1705 __export int osl_rbtree_last_row(const struct osl_table *t, unsigned col_num,
1706 struct osl_row **result)
1709 int ret = osl_get_num_rows(t, &num_rows);
1713 return osl_get_nth_row(t, col_num, num_rows, result);