2 * Copyright (C) 2007-2008 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() */
17 /* Taken from Drepper: How to write shared libraries, Appendix B. */
19 #define MSGSTRFIELD(line) MSGSTRFIELD1(line)
20 #define MSGSTRFIELD1(line) str##line
21 static const union msgstr_t {
23 #define OSL_ERROR(n, s) char MSGSTRFIELD(__LINE__)[sizeof(s)];
29 #define OSL_ERROR(n, s) s,
33 static const unsigned int errmsgidx[] = {
34 #define OSL_ERROR(n, s) [n] = offsetof(union msgstr_t, MSGSTRFIELD(__LINE__)),
39 __export const char *osl_strerror(int num)
41 if (IS_SYSTEM_ERROR(num))
42 return strerror((num) & ((1 << SYSTEM_ERROR_BIT) - 1));
43 return msgstr.str + errmsgidx[num];
48 static void __attribute ((constructor)) init_loglevel(void)
50 char *p = getenv("OSL_LOGLEVEL");
52 /* don't log anything if unset */
53 loglevel = p? atoi(p) : EMERG + 1;
60 * \param fmt Usual format string.
62 * All XXX_LOG() macros use this function.
64 __printf_2_3 void __log(int ll, const char* fmt,...)
77 strftime(str, sizeof(str), "%b %d %H:%M:%S", tm);
78 fprintf(outfd, "%s ", str);
80 vfprintf(outfd, fmt, argp);
85 * A wrapper for lseek(2).
87 * \param fd The file descriptor whose offset is to be to repositioned.
88 * \param offset A value-result parameter.
89 * \param whence Usual repositioning directive.
91 * Reposition the offset of the file descriptor \a fd to the argument \a offset
92 * according to the directive \a whence. Upon successful return, \a offset
93 * contains the resulting offset location as measured in bytes from the
94 * beginning of the file.
96 * \return Positive on success. Otherwise, the function returns \p -E_OSL_LSEEK.
100 static int __lseek(int fd, off_t *offset, int whence)
102 *offset = lseek(fd, *offset, whence);
103 int ret = -E_OSL_LSEEK;
109 static int append_file(const char *filename, char *data, size_t data_size,
114 // DEBUG_LOG("appending %zu + %zu bytes\n", header_size, data_size);
115 ret = osl_open(filename, O_WRONLY | O_CREAT | O_APPEND, 0644);
119 ret = write_all(fd, data, &data_size);
124 ret = __lseek(fd, &offset, SEEK_END);
127 // DEBUG_LOG("new file size: " FMT_OFF_T "\n", offset);
136 static int verify_name(const char *name)
139 return -E_OSL_BAD_NAME;
141 return -E_OSL_BAD_NAME;
142 if (strchr(name, '/'))
143 return -E_OSL_BAD_NAME;
144 if (!strcmp(name, ".."))
145 return -E_OSL_BAD_NAME;
146 if (!strcmp(name, "."))
147 return -E_OSL_BAD_NAME;
151 static char *disk_storage_dirname(const struct osl_table *t, unsigned col_num,
154 char *dirname, *column_name = column_filename(t, col_num);
158 if (!(t->desc->flags & OSL_LARGE_TABLE))
160 dirname = make_message("%s/%.2s", column_name, ds_name);
165 static char *disk_storage_name_of_object(const struct osl_table *t,
166 const struct osl_object *obj)
168 HASH_TYPE hash[HASH_SIZE];
169 hash_object(obj, hash);
170 return disk_storage_name_of_hash(t, hash);
173 static int disk_storage_name_of_row(const struct osl_table *t,
174 const struct osl_row *row, char **name)
176 struct osl_object obj;
177 int ret = osl_get_object(t, row, t->disk_storage_name_column, &obj);
181 *name = disk_storage_name_of_object(t, &obj);
184 return -ERRNO_TO_ERROR(ENOMEM);
187 static void column_name_hash(const char *col_name, HASH_TYPE *hash)
189 hash_function(col_name, strlen(col_name), hash);
192 static int init_column_descriptions(struct osl_table *t)
195 const struct osl_column_description *cd;
197 ret = -E_OSL_BAD_TABLE_DESC;
198 ret = verify_name(t->desc->name);
201 ret = -E_OSL_BAD_DB_DIR;
202 if (!t->desc->dir && (t->num_disk_storage_columns || t->num_mapped_columns))
204 /* the size of the index header without column descriptions */
205 t->index_header_size = IDX_COLUMN_DESCRIPTIONS;
206 FOR_EACH_COLUMN(i, t->desc, cd) {
207 struct osl_column *col = t->columns + i;
208 if (cd->storage_flags & OSL_RBTREE) {
209 if (!cd->compare_function)
210 return -E_OSL_NO_COMPARE_FUNC;
212 if (cd->storage_type == OSL_NO_STORAGE)
214 ret = -E_OSL_NO_COLUMN_NAME;
215 if (!cd->name || !cd->name[0])
217 ret = verify_name(cd->name);
220 t->index_header_size += index_column_description_size(cd->name);
221 column_name_hash(cd->name, col->name_hash);
222 ret = -E_OSL_DUPLICATE_COL_NAME;
223 for (j = i + 1; j < t->desc->num_columns; j++) {
224 const char *name2 = get_column_description(t->desc,
226 if (cd->name && name2 && !strcmp(cd->name, name2))
236 * Initialize a struct table from given table description.
238 * \param desc The description of the osl table.
239 * \param table_ptr Result is returned here.
241 * This function performs several sanity checks on \p desc and returns if any
242 * of these tests fail. On success, a struct \p osl_table is allocated and
243 * initialized with data derived from \p desc.
247 * \sa struct osl_table.
249 int init_table_structure(const struct osl_table_description *desc,
250 struct osl_table **table_ptr)
252 const struct osl_column_description *cd;
253 struct osl_table *t = calloc(1, sizeof(*t));
254 int i, ret = -ERRNO_TO_ERROR(ENOMEM), have_disk_storage_name_column = 0;
258 ret = -E_OSL_BAD_TABLE_DESC;
261 DEBUG_LOG("creating table structure for '%s' from table "
262 "description\n", desc->name);
263 ret = -E_OSL_NO_COLUMN_DESC;
264 if (!desc->column_descriptions)
266 ret = -E_OSL_NO_COLUMNS;
267 if (!desc->num_columns)
269 ret = -ERRNO_TO_ERROR(ENOMEM);
270 t->columns = calloc(desc->num_columns, sizeof(struct osl_column));
274 FOR_EACH_COLUMN(i, t->desc, cd) {
275 enum osl_storage_type st = cd->storage_type;
276 enum osl_storage_flags sf = cd->storage_flags;
277 struct osl_column *col = &t->columns[i];
279 ret = -E_OSL_BAD_STORAGE_TYPE;
280 if (st != OSL_MAPPED_STORAGE && st != OSL_DISK_STORAGE
281 && st != OSL_NO_STORAGE)
283 ret = -E_OSL_BAD_STORAGE_FLAGS;
284 if (st == OSL_DISK_STORAGE && sf & OSL_RBTREE)
286 ret = -E_OSL_BAD_STORAGE_SIZE;
287 if (sf & OSL_FIXED_SIZE && !cd->data_size)
290 case OSL_DISK_STORAGE:
291 t->num_disk_storage_columns++;
293 case OSL_MAPPED_STORAGE:
294 t->num_mapped_columns++;
295 col->index_offset = t->row_index_size;
296 t->row_index_size += 8;
299 col->volatile_num = t->num_volatile_columns;
300 t->num_volatile_columns++;
303 if (sf & OSL_RBTREE) {
304 col->rbtree_num = t->num_rbtrees;
306 if ((sf & OSL_UNIQUE) && (st == OSL_MAPPED_STORAGE)) {
307 if (!have_disk_storage_name_column)
308 t->disk_storage_name_column = i;
309 have_disk_storage_name_column = 1;
313 ret = -E_OSL_NO_UNIQUE_RBTREE_COLUMN;
314 if (t->num_disk_storage_columns && !have_disk_storage_name_column)
316 ret = -E_OSL_NO_RBTREE_COL;
320 DEBUG_LOG("OK. Index entry size: %u\n", t->row_index_size);
321 ret = init_column_descriptions(t);
333 * Read the table description from index header.
335 * \param map The memory mapping of the index file.
336 * \param desc The values found in the index header are returned here.
338 * Read the index header, check for the paraslash magic string and the table version number.
339 * Read all information stored in the index header into \a desc.
343 * \sa struct osl_table_description, osl_create_table.
345 int read_table_desc(struct osl_object *map, struct osl_table_description *desc)
347 char *buf = map->data;
349 uint16_t header_size;
352 struct osl_column_description *cd;
354 if (map->size < MIN_INDEX_HEADER_SIZE(1))
355 return -E_OSL_SHORT_TABLE;
356 if (strncmp(buf + IDX_PARA_MAGIC, PARA_MAGIC, strlen(PARA_MAGIC)))
357 return -E_OSL_NO_MAGIC;
358 version = read_u8(buf + IDX_VERSION);
359 if (version < MIN_TABLE_VERSION || version > MAX_TABLE_VERSION)
360 return -E_OSL_VERSION_MISMATCH;
361 desc->flags = read_u8(buf + IDX_TABLE_FLAGS);
362 desc->num_columns = read_u16(buf + IDX_NUM_COLUMNS);
363 INFO_LOG("%u columns\n", desc->num_columns);
364 if (!desc->num_columns)
365 return -E_OSL_NO_COLUMNS;
366 header_size = read_u16(buf + IDX_HEADER_SIZE);
367 if (map->size < header_size)
368 return -E_OSL_BAD_SIZE;
369 desc->column_descriptions = calloc(desc->num_columns,
370 sizeof(struct osl_column_description));
371 if (!desc->column_descriptions)
372 return -ERRNO_TO_ERROR(ENOMEM);
373 offset = IDX_COLUMN_DESCRIPTIONS;
374 FOR_EACH_COLUMN(i, desc, cd) {
377 ret = -E_OSL_SHORT_TABLE;
378 if (map->size < offset + MIN_IDX_COLUMN_DESCRIPTION_SIZE) {
379 ERROR_LOG("map size = %zu < %u = offset + min desc size\n",
380 map->size, offset + MIN_IDX_COLUMN_DESCRIPTION_SIZE);
383 cd->storage_type = read_u16(buf + offset + IDX_CD_STORAGE_TYPE);
384 cd->storage_flags = read_u16(buf + offset +
385 IDX_CD_STORAGE_FLAGS);
386 cd->data_size = read_u32(buf + offset + IDX_CD_DATA_SIZE);
387 null_byte = memchr(buf + offset + IDX_CD_NAME, '\0',
388 map->size - offset - IDX_CD_NAME);
389 ret = -E_OSL_INDEX_CORRUPTION;
392 ret = -ERRNO_TO_ERROR(ENOMEM);
393 cd->name = strdup(buf + offset + IDX_CD_NAME);
396 offset += index_column_description_size(cd->name);
398 if (offset != header_size) {
399 ret = -E_OSL_INDEX_CORRUPTION;
400 ERROR_LOG("real header size = %u != %u = stored header size\n",
401 offset, header_size);
406 FOR_EACH_COLUMN(i, desc, cd)
412 * check whether the table description given by \p t->desc matches the on-disk
413 * table structure stored in the index of \a t.
415 static int compare_table_descriptions(struct osl_table *t)
418 struct osl_table_description desc;
419 const struct osl_column_description *cd1, *cd2;
421 /* read the on-disk structure into desc */
422 ret = read_table_desc(&t->index_map, &desc);
425 ret = -E_OSL_BAD_TABLE_FLAGS;
426 if (desc.flags != t->desc->flags)
428 ret = -E_OSL_BAD_COLUMN_NUM;
429 if (desc.num_columns > t->desc->num_columns)
431 if (desc.num_columns < t->desc->num_columns) {
432 struct osl_column_description *cd;
433 unsigned diff = t->desc->num_columns - desc.num_columns;
434 INFO_LOG("extending table by %u volatile columns\n", diff);
435 ret = -ERRNO_TO_ERROR(ENOMEM);
436 desc.column_descriptions = realloc(desc.column_descriptions,
437 t->desc->num_columns * sizeof(struct osl_column_description));
438 if (!desc.column_descriptions)
440 for (i = desc.num_columns; i < t->desc->num_columns; i++) {
441 cd = get_column_description(&desc, i);
442 cd->storage_type = OSL_NO_STORAGE;
445 desc.num_columns += diff;
447 FOR_EACH_COLUMN(i, t->desc, cd1) {
448 cd2 = get_column_description(&desc, i);
449 ret = -E_OSL_BAD_STORAGE_TYPE;
450 if (cd1->storage_type != cd2->storage_type)
452 if (cd1->storage_type == OSL_NO_STORAGE)
454 ret = -E_OSL_BAD_STORAGE_FLAGS;
455 if (cd1->storage_flags != cd2->storage_flags) {
456 ERROR_LOG("sf1 = %u != %u = sf2\n",
457 cd1->storage_flags, cd2->storage_flags);
460 ret = -E_OSL_BAD_DATA_SIZE;
461 if (cd1->storage_flags & OSL_FIXED_SIZE)
462 if (cd1->data_size != cd2->data_size)
464 ret = -E_OSL_BAD_COLUMN_NAME;
465 if (strcmp(cd1->name, cd2->name))
468 INFO_LOG("table description of '%s' matches on-disk data, good\n",
472 FOR_EACH_COLUMN(i, &desc, cd1)
474 free(desc.column_descriptions);
478 static int create_table_index(struct osl_table *t)
480 char *buf, *filename;
482 size_t size = t->index_header_size;
483 const struct osl_column_description *cd;
486 INFO_LOG("creating %zu byte index for table %s\n", size,
488 buf = calloc(1, size);
490 return -ERRNO_TO_ERROR(ENOMEM);
491 sprintf(buf + IDX_PARA_MAGIC, "%s", PARA_MAGIC);
492 write_u8(buf + IDX_TABLE_FLAGS, t->desc->flags);
493 write_u8(buf + IDX_DIRTY_FLAG, 0);
494 write_u8(buf + IDX_VERSION, CURRENT_TABLE_VERSION);
495 write_u16(buf + IDX_NUM_COLUMNS, t->num_mapped_columns + t->num_disk_storage_columns);
496 write_u16(buf + IDX_HEADER_SIZE, t->index_header_size);
497 offset = IDX_COLUMN_DESCRIPTIONS;
498 FOR_EACH_COLUMN(i, t->desc, cd) {
499 /* no need to store info about volatile storage */
500 if (cd->storage_type == OSL_NO_STORAGE)
502 write_u16(buf + offset + IDX_CD_STORAGE_TYPE,
504 write_u16(buf + offset + IDX_CD_STORAGE_FLAGS,
506 if (cd->storage_flags & OSL_FIXED_SIZE)
507 write_u32(buf + offset + IDX_CD_DATA_SIZE,
509 strcpy(buf + offset + IDX_CD_NAME, cd->name);
510 offset += index_column_description_size(cd->name);
512 assert(offset = size);
513 filename = index_filename(t->desc);
515 ret = write_file(filename, buf, size);
517 ret = -ERRNO_TO_ERROR(ENOMEM);
523 __export int osl_create_table(const struct osl_table_description *desc)
525 const struct osl_column_description *cd;
526 char *table_dir = NULL, *filename;
528 int i, ret = init_table_structure(desc, &t);
532 INFO_LOG("creating %s\n", desc->name);
533 FOR_EACH_COLUMN(i, t->desc, cd) {
534 if (cd->storage_type == OSL_NO_STORAGE)
537 ret = osl_mkdir(desc->dir, 0777);
538 if (ret < 0 && !is_errno(-ret, EEXIST))
540 table_dir = make_message("%s/%s", desc->dir,
542 ret = -ERRNO_TO_ERROR(ENOMEM);
545 ret = osl_mkdir(table_dir, 0777);
549 ret = -ERRNO_TO_ERROR(ENOMEM);
550 filename = column_filename(t, i);
553 INFO_LOG("filename: %s\n", filename);
554 if (cd->storage_type == OSL_MAPPED_STORAGE) {
555 ret = osl_open(filename, O_RDWR | O_CREAT | O_EXCL,
564 ret = osl_mkdir(filename, 0777);
569 if (t->num_mapped_columns) {
570 ret = create_table_index(t);
582 static int table_is_dirty(struct osl_table *t)
584 char *buf = (char *)t->index_map.data + IDX_DIRTY_FLAG;
585 uint8_t dirty = read_u8(buf) & 0x1;
589 static void mark_table_dirty(struct osl_table *t)
591 char *buf = (char *)t->index_map.data + IDX_DIRTY_FLAG;
592 write_u8(buf, read_u8(buf) | 1);
595 static void mark_table_clean(struct osl_table *t)
597 char *buf = (char *)t->index_map.data + IDX_DIRTY_FLAG;
598 write_u8(buf, read_u8(buf) & 0xfe);
601 static void unmap_column(struct osl_table *t, unsigned col_num)
603 struct osl_object map = t->columns[col_num].data_map;
607 ret = osl_munmap(map.data, map.size);
613 * Unmap all mapped files of an osl table.
615 * \param t Pointer to a mapped table.
616 * \param flags Options for unmapping.
618 * \return Positive on success, negative on errors.
620 * \sa map_table(), enum osl_close_flags, osl_munmap().
622 int unmap_table(struct osl_table *t, enum osl_close_flags flags)
625 const struct osl_column_description *cd;
628 if (!t->num_mapped_columns) /* can this ever happen? */
630 INFO_LOG("unmapping table '%s'\n", t->desc->name);
631 if (!t->index_map.data)
632 return -E_OSL_NOT_MAPPED;
633 if (flags & OSL_MARK_CLEAN)
635 ret = osl_munmap(t->index_map.data, t->index_map.size);
638 t->index_map.data = NULL;
641 FOR_EACH_MAPPED_COLUMN(i, t, cd)
646 static int map_column(struct osl_table *t, unsigned col_num)
649 char *filename = column_filename(t, col_num);
653 return -ERRNO_TO_ERROR(ENOMEM);
654 ret = osl_stat(filename, &statbuf);
659 if (!(S_IFREG & statbuf.st_mode)) {
663 ret = mmap_full_file(filename, O_RDWR,
664 &t->columns[col_num].data_map.data,
665 &t->columns[col_num].data_map.size,
672 * Map the index file and all columns of type \p OSL_MAPPED_STORAGE into memory.
674 * \param t Pointer to an initialized table structure.
675 * \param flags Mapping options.
677 * \return Negative return value on errors; on success the number of rows
678 * (including invalid rows) is returned.
680 * \sa unmap_table(), enum map_table_flags, osl_open_table(), mmap(2).
682 int map_table(struct osl_table *t, enum map_table_flags flags)
685 const struct osl_column_description *cd;
686 int i = 0, ret, num_rows = 0;
688 if (!t->num_mapped_columns)
690 if (t->index_map.data)
691 return -E_OSL_ALREADY_MAPPED;
692 filename = index_filename(t->desc);
694 return -ERRNO_TO_ERROR(ENOMEM);
695 INFO_LOG("mapping table '%s' (index: %s)\n", t->desc->name, filename);
696 ret = mmap_full_file(filename, flags & MAP_TBL_FL_MAP_RDONLY?
697 O_RDONLY : O_RDWR, &t->index_map.data, &t->index_map.size, NULL);
701 if (flags & MAP_TBL_FL_VERIFY_INDEX) {
702 ret = compare_table_descriptions(t);
707 if (!(flags & MAP_TBL_FL_IGNORE_DIRTY)) {
708 if (table_is_dirty(t)) {
709 ERROR_LOG("%s is dirty\n", t->desc->name);
714 num_rows = table_num_rows(t);
718 FOR_EACH_MAPPED_COLUMN(i, t, cd) {
719 ret = map_column(t, i);
724 err: /* unmap what is already mapped */
725 for (i--; i >= 0; i--) {
726 struct osl_object map = t->columns[i].data_map;
727 osl_munmap(map.data, map.size);
730 osl_munmap(t->index_map.data, t->index_map.size);
731 t->index_map.data = NULL;
736 * Retrieve a mapped object by row and column number.
738 * \param t Pointer to an open osl table.
739 * \param col_num Number of the mapped column containing the object to retrieve.
740 * \param row_num Number of the row containing the object to retrieve.
741 * \param obj The result is returned here.
743 * It is considered an error if \a col_num does not refer to a column
744 * of storage type \p OSL_MAPPED_STORAGE.
748 * \sa osl_storage_type.
750 int get_mapped_object(const struct osl_table *t, unsigned col_num,
751 uint32_t row_num, struct osl_object *obj)
753 struct osl_column *col = &t->columns[col_num];
758 if (t->num_rows <= row_num)
759 return -E_OSL_BAD_ROW_NUM;
760 ret = get_cell_index(t, row_num, col_num, &cell_index);
763 offset = read_u32(cell_index);
764 obj->size = read_u32(cell_index + 4);
765 obj->data = col->data_map.data + offset;
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;
864 return -ERRNO_TO_ERROR(ENOMEM);
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);
1066 ret = -ERRNO_TO_ERROR(ENOMEM);
1069 /* check if directory exists */
1070 ret = stat(dirname, &statbuf);
1073 ret = -ERRNO_TO_ERROR(errno);
1076 ret = -ERRNO_TO_ERROR(ENOTDIR);
1077 if (!S_ISDIR(statbuf.st_mode))
1080 ret = map_table(t, MAP_TBL_FL_VERIFY_INDEX);
1084 DEBUG_LOG("num rows: %d\n", t->num_rows);
1085 ret = init_rbtrees(t);
1087 osl_close_table(t, OSL_MARK_CLEAN); /* ignore further errors */
1098 static int create_disk_storage_object_dir(const struct osl_table *t,
1099 unsigned col_num, const char *ds_name)
1104 if (!(t->desc->flags & OSL_LARGE_TABLE))
1106 dirname = disk_storage_dirname(t, col_num, ds_name);
1108 return -ERRNO_TO_ERROR(ENOMEM);
1109 ret = osl_mkdir(dirname, 0777);
1111 if (ret < 0 && !is_errno(-ret, EEXIST))
1116 static int write_disk_storage_file(const struct osl_table *t, unsigned col_num,
1117 const struct osl_object *obj, const char *ds_name)
1122 ret = create_disk_storage_object_dir(t, col_num, ds_name);
1125 filename = disk_storage_path(t, col_num, ds_name);
1127 return -ERRNO_TO_ERROR(ENOMEM);
1128 ret = write_file(filename, obj->data, obj->size);
1133 static int append_map_file(const struct osl_table *t, unsigned col_num,
1134 const struct osl_object *obj, uint32_t *new_size)
1136 char *filename = column_filename(t, col_num);
1140 return -ERRNO_TO_ERROR(ENOMEM);
1141 ret = append_file(filename, obj->data, obj->size, new_size);
1146 static int append_row_index(const struct osl_table *t, char *row_index)
1151 if (!t->num_mapped_columns)
1153 filename = index_filename(t->desc);
1155 return -ERRNO_TO_ERROR(ENOMEM);
1156 ret = append_file(filename, row_index, t->row_index_size, NULL);
1161 static int truncate_mapped_file(const struct osl_table *t, unsigned col_num,
1165 char *filename = column_filename(t, col_num);
1168 return -ERRNO_TO_ERROR(ENOMEM);
1169 ret = truncate_file(filename, size);
1174 static int delete_disk_storage_file(const struct osl_table *t, unsigned col_num,
1175 const char *ds_name)
1177 char *dirname, *filename = disk_storage_path(t, col_num, ds_name);
1181 return -ERRNO_TO_ERROR(ENOMEM);
1182 ret = unlink(filename);
1186 return -ERRNO_TO_ERROR(err);
1187 if (!(t->desc->flags & OSL_LARGE_TABLE))
1189 dirname = disk_storage_dirname(t, col_num, ds_name);
1191 return -ERRNO_TO_ERROR(ENOMEM);
1197 __export int osl_add_and_get_row(struct osl_table *t, struct osl_object *objects,
1198 struct osl_row **row)
1201 char *ds_name = NULL;
1202 struct rb_node **rb_parents = NULL, ***rb_links = NULL;
1203 char *new_row_index = NULL;
1204 struct osl_object *volatile_objs = NULL;
1205 const struct osl_column_description *cd;
1208 return -E_OSL_BAD_TABLE;
1209 rb_parents = malloc(t->num_rbtrees * sizeof(struct rn_node*));
1211 return -ERRNO_TO_ERROR(ENOMEM);
1212 rb_links = malloc(t->num_rbtrees * sizeof(struct rn_node**));
1215 return -ERRNO_TO_ERROR(ENOMEM);
1217 if (t->num_mapped_columns) {
1218 new_row_index = malloc(t->row_index_size);
1219 if (!new_row_index) {
1222 return -ERRNO_TO_ERROR(ENOMEM);
1225 /* pass 1: sanity checks */
1226 // DEBUG_LOG("sanity tests: %p:%p\n", objects[0].data,
1227 // objects[1].data);
1228 FOR_EACH_COLUMN(i, t->desc, cd) {
1229 enum osl_storage_type st = cd->storage_type;
1230 enum osl_storage_flags sf = cd->storage_flags;
1232 // ret = -E_OSL_NULL_OBJECT;
1235 if (st == OSL_DISK_STORAGE)
1237 if (sf & OSL_RBTREE) {
1238 unsigned rbtree_num = t->columns[i].rbtree_num;
1239 ret = -E_OSL_RB_KEY_EXISTS;
1240 // DEBUG_LOG("checking whether %p exists\n",
1241 // objects[i].data);
1242 if (search_rbtree(objects + i, t, i,
1243 &rb_parents[rbtree_num],
1244 &rb_links[rbtree_num]) > 0)
1247 if (sf & OSL_FIXED_SIZE) {
1248 // DEBUG_LOG("fixed size. need: %zu, have: %d\n",
1249 // objects[i].size, cd->data_size);
1250 ret = -E_OSL_BAD_DATA_SIZE;
1251 if (objects[i].size != cd->data_size)
1255 if (t->num_disk_storage_columns) {
1256 ds_name = disk_storage_name_of_object(t,
1257 &objects[t->disk_storage_name_column]);
1258 ret = -ERRNO_TO_ERROR(ENOMEM);
1262 ret = unmap_table(t, OSL_MARK_CLEAN);
1265 // DEBUG_LOG("sanity tests passed%s\n", "");
1266 /* pass 2: create data files, append map data */
1267 FOR_EACH_COLUMN(i, t->desc, cd) {
1268 enum osl_storage_type st = cd->storage_type;
1269 if (st == OSL_NO_STORAGE)
1271 if (st == OSL_MAPPED_STORAGE) {
1273 struct osl_column *col = &t->columns[i];
1274 // DEBUG_LOG("appending object of size %zu\n",
1275 // objects[i].size);
1276 ret = append_map_file(t, i, objects + i, &new_size);
1279 update_cell_index(new_row_index, col, new_size,
1284 ret = write_disk_storage_file(t, i, objects + i, ds_name);
1288 ret = append_row_index(t, new_row_index);
1291 ret = map_table(t, MAP_TBL_FL_VERIFY_INDEX);
1292 if (ret < 0) { /* truncate index and rollback changes */
1293 char *filename = index_filename(t->desc);
1295 truncate_file(filename, t->row_index_size);
1299 /* pass 3: add entry to rbtrees */
1300 if (t->num_volatile_columns) {
1301 ret = -ERRNO_TO_ERROR(ENOMEM);
1302 volatile_objs = calloc(t->num_volatile_columns,
1303 sizeof(struct osl_object));
1306 FOR_EACH_VOLATILE_COLUMN(i, t, cd)
1307 volatile_objs[t->columns[i].volatile_num] = objects[i];
1310 // DEBUG_LOG("adding new entry as row #%d\n", t->num_rows - 1);
1311 ret = add_row_to_rbtrees(t, t->num_rows - 1, volatile_objs, row);
1314 // DEBUG_LOG("added new entry as row #%d\n", t->num_rows - 1);
1317 rollback: /* rollback all changes made, ignore further errors */
1318 for (i--; i >= 0; i--) {
1319 cd = get_column_description(t->desc, i);
1320 enum osl_storage_type st = cd->storage_type;
1321 if (st == OSL_NO_STORAGE)
1324 if (st == OSL_MAPPED_STORAGE)
1325 truncate_mapped_file(t, i, objects[i].size);
1326 else /* disk storage */
1327 delete_disk_storage_file(t, i, ds_name);
1329 /* ignore error and return previous error value */
1330 map_table(t, MAP_TBL_FL_VERIFY_INDEX);
1332 free(new_row_index);
1339 __export int osl_add_row(struct osl_table *t, struct osl_object *objects)
1341 return osl_add_and_get_row(t, objects, NULL);
1344 __export int osl_get_object(const struct osl_table *t, const struct osl_row *r,
1345 unsigned col_num, struct osl_object *object)
1347 const struct osl_column_description *cd;
1350 return -E_OSL_BAD_TABLE;
1351 cd = get_column_description(t->desc, col_num);
1352 /* col must not be disk storage */
1353 if (cd->storage_type == OSL_DISK_STORAGE)
1354 return -E_OSL_BAD_STORAGE_TYPE;
1355 if (cd->storage_type == OSL_MAPPED_STORAGE)
1356 return get_mapped_object(t, col_num, r->num, object);
1358 *object = r->volatile_objects[t->columns[col_num].volatile_num];
1362 __export int osl_del_row(struct osl_table *t, struct osl_row *row)
1364 struct osl_row *r = row;
1366 const struct osl_column_description *cd;
1369 return -E_OSL_BAD_TABLE;
1370 INFO_LOG("deleting row %p\n", row);
1372 if (t->num_disk_storage_columns) {
1374 ret = disk_storage_name_of_row(t, r, &ds_name);
1377 FOR_EACH_DISK_STORAGE_COLUMN(i, t, cd)
1378 delete_disk_storage_file(t, i, ds_name);
1381 FOR_EACH_COLUMN(i, t->desc, cd) {
1382 struct osl_column *col = t->columns + i;
1383 enum osl_storage_type st = cd->storage_type;
1384 remove_rb_node(t, i, r);
1385 if (st == OSL_MAPPED_STORAGE)
1387 if (st == OSL_NO_STORAGE && !(cd->storage_flags & OSL_DONT_FREE))
1388 free(r->volatile_objects[col->volatile_num].data);
1390 if (t->num_mapped_columns) {
1391 ret = mark_row_invalid(t, r->num);
1394 t->num_invalid_rows++;
1399 free(r->volatile_objects);
1404 /* test if column has an rbtree */
1405 static int check_rbtree_col(const struct osl_table *t, unsigned col_num,
1406 struct osl_column **col)
1409 return -E_OSL_BAD_TABLE;
1410 if (!(get_column_description(t->desc, col_num)->storage_flags & OSL_RBTREE))
1411 return -E_OSL_BAD_STORAGE_FLAGS;
1412 *col = t->columns + col_num;
1416 __export int osl_get_row(const struct osl_table *t, unsigned col_num,
1417 const struct osl_object *obj, struct osl_row **result)
1420 struct rb_node *node;
1421 struct osl_row *row;
1422 struct osl_column *col;
1425 ret = check_rbtree_col(t, col_num, &col);
1428 ret = search_rbtree(obj, t, col_num, &node, NULL);
1431 row = get_row_pointer(node, t->columns[col_num].rbtree_num);
1436 static int rbtree_loop(struct osl_column *col, void *private_data,
1437 osl_rbtree_loop_func *func)
1439 struct rb_node *n, *tmp;
1441 /* this for-loop is safe against removal of an entry */
1442 for (n = rb_first(&col->rbtree), tmp = n? rb_next(n) : NULL;
1444 n = tmp, tmp = tmp? rb_next(tmp) : NULL) {
1445 struct osl_row *r = get_row_pointer(n, col->rbtree_num);
1446 if (func(r, private_data) < 0)
1452 static int rbtree_loop_reverse(struct osl_column *col, void *private_data,
1453 osl_rbtree_loop_func *func)
1455 struct rb_node *n, *tmp;
1457 /* safe against removal of an entry */
1458 for (n = rb_last(&col->rbtree), tmp = n? rb_prev(n) : NULL;
1460 n = tmp, tmp = tmp? rb_prev(tmp) : NULL) {
1461 struct osl_row *r = get_row_pointer(n, col->rbtree_num);
1462 if (func(r, private_data) < 0)
1468 __export int osl_rbtree_loop(const struct osl_table *t, unsigned col_num,
1469 void *private_data, osl_rbtree_loop_func *func)
1471 struct osl_column *col;
1473 int ret = check_rbtree_col(t, col_num, &col);
1476 return rbtree_loop(col, private_data, func);
1479 __export int osl_rbtree_loop_reverse(const struct osl_table *t, unsigned col_num,
1480 void *private_data, osl_rbtree_loop_func *func)
1482 struct osl_column *col;
1484 int ret = check_rbtree_col(t, col_num, &col);
1487 return rbtree_loop_reverse(col, private_data, func);
1490 /* TODO: Rollback changes on errors */
1491 static int rename_disk_storage_objects(struct osl_table *t,
1492 struct osl_object *old_obj, struct osl_object *new_obj)
1495 const struct osl_column_description *cd;
1496 char *old_ds_name, *new_ds_name;
1498 if (!t->num_disk_storage_columns)
1499 return 1; /* nothing to do */
1500 if (old_obj->size == new_obj->size && !memcmp(new_obj->data,
1501 old_obj->data, new_obj->size))
1502 return 1; /* object did not change */
1503 old_ds_name = disk_storage_name_of_object(t, old_obj);
1504 new_ds_name = disk_storage_name_of_object(t, new_obj);
1505 ret = -ERRNO_TO_ERROR(ENOMEM);
1506 if (!old_ds_name || ! new_ds_name)
1509 FOR_EACH_DISK_STORAGE_COLUMN(i, t, cd) {
1510 char *old_filename, *new_filename;
1511 ret = create_disk_storage_object_dir(t, i, new_ds_name);
1514 old_filename = disk_storage_path(t, i, old_ds_name);
1515 new_filename = disk_storage_path(t, i, new_ds_name);
1516 if (!old_filename || !new_filename)
1517 ret = -ERRNO_TO_ERROR(ENOMEM);
1519 ret = osl_rename(old_filename, new_filename);
1533 __export int osl_update_object(struct osl_table *t, const struct osl_row *r,
1534 unsigned col_num, struct osl_object *obj)
1536 struct osl_column *col;
1537 const struct osl_column_description *cd;
1541 return -E_OSL_BAD_TABLE;
1542 col = &t->columns[col_num];
1543 cd = get_column_description(t->desc, col_num);
1544 DEBUG_LOG("updating column %u of %s\n", col_num, t->desc->name);
1545 if (cd->storage_flags & OSL_RBTREE) {
1546 if (search_rbtree(obj, t, col_num, NULL, NULL) > 0)
1547 return -E_OSL_RB_KEY_EXISTS;
1549 if (cd->storage_flags & OSL_FIXED_SIZE) {
1550 if (obj->size != cd->data_size)
1551 return -E_OSL_BAD_DATA_SIZE;
1553 remove_rb_node(t, col_num, r);
1554 if (cd->storage_type == OSL_NO_STORAGE) { /* TODO: If fixed size, reuse object? */
1555 free(r->volatile_objects[col->volatile_num].data);
1556 r->volatile_objects[col->volatile_num] = *obj;
1557 } else if (cd->storage_type == OSL_DISK_STORAGE) {
1559 ret = disk_storage_name_of_row(t, r, &ds_name);
1562 ret = delete_disk_storage_file(t, col_num, ds_name);
1563 if (ret < 0 && !is_errno(-ret, ENOENT)) {
1567 ret = write_disk_storage_file(t, col_num, obj, ds_name);
1571 } else { /* mapped storage */
1572 struct osl_object old_obj;
1573 ret = get_mapped_object(t, col_num, r->num, &old_obj);
1577 * If the updated column is the disk storage name column, the
1578 * disk storage name changes, so we have to rename all disk
1579 * storage objects accordingly.
1581 if (col_num == t->disk_storage_name_column) {
1582 ret = rename_disk_storage_objects(t, &old_obj, obj);
1586 if (cd->storage_flags & OSL_FIXED_SIZE)
1587 memcpy(old_obj.data, obj->data, cd->data_size);
1588 else { /* TODO: if the size doesn't change, use old space */
1589 uint32_t new_data_map_size;
1591 ret = get_row_index(t, r->num, &row_index);
1594 unmap_column(t, col_num);
1595 ret = append_map_file(t, col_num, obj,
1596 &new_data_map_size);
1599 ret = map_column(t, col_num);
1602 update_cell_index(row_index, col, new_data_map_size,
1606 if (cd->storage_flags & OSL_RBTREE) {
1607 ret = insert_rbtree(t, col_num, r, obj);
1614 __export int osl_open_disk_object(const struct osl_table *t, const struct osl_row *r,
1615 unsigned col_num, struct osl_object *obj)
1617 const struct osl_column_description *cd;
1618 char *ds_name, *filename;
1622 return -E_OSL_BAD_TABLE;
1623 cd = get_column_description(t->desc, col_num);
1624 if (cd->storage_type != OSL_DISK_STORAGE)
1625 return -E_OSL_BAD_STORAGE_TYPE;
1627 ret = disk_storage_name_of_row(t, r, &ds_name);
1630 filename = disk_storage_path(t, col_num, ds_name);
1633 return -ERRNO_TO_ERROR(ENOMEM);
1634 DEBUG_LOG("filename: %s\n", filename);
1635 ret = mmap_full_file(filename, O_RDONLY, &obj->data, &obj->size, NULL);
1640 __export int osl_close_disk_object(struct osl_object *obj)
1642 return osl_munmap(obj->data, obj->size);
1645 __export int osl_get_num_rows(const struct osl_table *t, unsigned *num_rows)
1648 return -E_OSL_BAD_TABLE;
1649 assert(t->num_rows >= t->num_invalid_rows);
1650 *num_rows = t->num_rows - t->num_invalid_rows;
1654 __export int osl_get_rank(const struct osl_table *t, struct osl_row *r,
1655 unsigned col_num, unsigned *rank)
1657 struct osl_object obj;
1658 struct osl_column *col;
1659 struct rb_node *node;
1660 int ret = check_rbtree_col(t, col_num, &col);
1664 ret = osl_get_object(t, r, col_num, &obj);
1667 ret = search_rbtree(&obj, t, col_num, &node, NULL);
1670 ret = rb_rank(node, rank);
1672 return -E_OSL_BAD_ROW;
1676 __export int osl_get_nth_row(const struct osl_table *t, unsigned col_num,
1677 unsigned n, struct osl_row **result)
1679 struct osl_column *col;
1680 struct rb_node *node;
1685 return -E_OSL_RB_KEY_NOT_FOUND;
1686 ret = osl_get_num_rows(t, &num_rows);
1690 return -E_OSL_RB_KEY_NOT_FOUND;
1691 ret = check_rbtree_col(t, col_num, &col);
1694 node = rb_nth(col->rbtree.rb_node, n);
1696 return -E_OSL_RB_KEY_NOT_FOUND;
1697 *result = get_row_pointer(node, col->rbtree_num);
1701 __export int osl_rbtree_first_row(const struct osl_table *t, unsigned col_num,
1702 struct osl_row **result)
1704 return osl_get_nth_row(t, col_num, 1, result);
1707 __export int osl_rbtree_last_row(const struct osl_table *t, unsigned col_num,
1708 struct osl_row **result)
1711 int ret = osl_get_num_rows(t, &num_rows);
1715 return osl_get_nth_row(t, col_num, num_rows, result);