2 * Copyright (C) 2007-2009 Andre Noll <maan@tuebingen.mpg.de>
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 * The main reason for this rather fancy implementation of strerror() is to
20 * avoid having an array of pointers. This is desirable because initialized
21 * pointer variables increase the startup time of the library due to the
22 * processing of relocations.
25 #define MSGSTRFIELD(line) MSGSTRFIELD1(line)
26 #define MSGSTRFIELD1(line) str##line
27 static const union msgstr_t {
29 #define OSL_ERROR(n, s) char MSGSTRFIELD(__LINE__)[sizeof(s)];
35 #define OSL_ERROR(n, s) s,
39 static const unsigned int errmsgidx[] = {
40 #define OSL_ERROR(n, s) [n] = offsetof(union msgstr_t, MSGSTRFIELD(__LINE__)),
45 __export const char *osl_strerror(int num)
47 return msgstr.str + errmsgidx[num];
52 static void __attribute ((constructor)) init_loglevel(void)
54 char *p = getenv("OSL_LOGLEVEL");
56 /* don't log anything if unset */
57 loglevel = p? atoi(p) : EMERG + 1;
64 * \param fmt Usual format string.
66 * All XXX_LOG() macros use this function.
68 __printf_2_3 void __log(int ll, const char* fmt,...)
79 strftime(str, sizeof(str), "%b %d %H:%M:%S", tm);
80 fprintf(stderr, "%s ", str);
82 vfprintf(stderr, fmt, argp);
87 * A wrapper for lseek(2).
89 * \param fd The file descriptor whose offset is to be to repositioned.
90 * \param offset A value-result parameter.
91 * \param whence Usual repositioning directive.
93 * Reposition the offset of the file descriptor \a fd to the argument \a offset
94 * according to the directive \a whence. Upon successful return, \a offset
95 * contains the resulting offset location as measured in bytes from the
96 * beginning of the file.
98 * \return Positive on success. Otherwise, the function returns \p -E_OSL_LSEEK.
102 static int __lseek(int fd, off_t *offset, int whence)
104 *offset = lseek(fd, *offset, whence);
105 int ret = -E_OSL_LSEEK;
111 static int append_file(const char *filename, char *data, size_t data_size,
116 // DEBUG_LOG("appending %zu + %zu bytes\n", header_size, data_size);
117 ret = osl_open(filename, O_WRONLY | O_CREAT | O_APPEND, 0644);
121 ret = write_all(fd, data, &data_size);
126 ret = __lseek(fd, &offset, SEEK_END);
129 // DEBUG_LOG("new file size: " FMT_OFF_T "\n", offset);
138 static int verify_name(const char *name)
141 return -E_OSL_BAD_NAME;
143 return -E_OSL_BAD_NAME;
144 if (strchr(name, '/'))
145 return -E_OSL_BAD_NAME;
146 if (!strcmp(name, ".."))
147 return -E_OSL_BAD_NAME;
148 if (!strcmp(name, "."))
149 return -E_OSL_BAD_NAME;
153 static char *disk_storage_dirname(const struct osl_table *t, unsigned col_num,
156 char *dirname, *column_name = column_filename(t, col_num);
160 if (!(t->desc->flags & OSL_LARGE_TABLE))
162 dirname = make_message("%s/%.2s", column_name, ds_name);
167 static char *disk_storage_name_of_object(const struct osl_table *t,
168 const struct osl_object *obj)
170 HASH_TYPE hash[HASH_SIZE];
171 hash_object(obj, hash);
172 return disk_storage_name_of_hash(t, hash);
175 static int disk_storage_name_of_row(const struct osl_table *t,
176 const struct osl_row *row, char **name)
178 struct osl_object obj;
179 int ret = osl_get_object(t, row, t->disk_storage_name_column, &obj);
183 *name = disk_storage_name_of_object(t, &obj);
189 static void column_name_hash(const char *col_name, HASH_TYPE *hash)
191 hash_function(col_name, strlen(col_name), hash);
194 static int init_column_descriptions(struct osl_table *t)
197 const struct osl_column_description *cd;
199 ret = verify_name(t->desc->name);
202 ret = -E_OSL_BAD_DB_DIR;
203 if (!t->desc->dir && (t->num_disk_storage_columns || t->num_mapped_columns))
205 /* the size of the index header without column descriptions */
206 t->index_header_size = IDX_COLUMN_DESCRIPTIONS;
207 FOR_EACH_COLUMN(i, t->desc, cd) {
208 struct osl_column *col = t->columns + i;
209 if (cd->storage_flags & OSL_RBTREE) {
210 if (!cd->compare_function)
211 return -E_OSL_NO_COMPARE_FUNC;
213 if (cd->storage_type == OSL_NO_STORAGE)
215 ret = -E_OSL_NO_COLUMN_NAME;
216 if (!cd->name || !cd->name[0])
218 ret = verify_name(cd->name);
221 t->index_header_size += index_column_description_size(cd->name);
222 column_name_hash(cd->name, col->name_hash);
223 ret = -E_OSL_DUPLICATE_COL_NAME;
224 for (j = i + 1; j < t->desc->num_columns; j++) {
225 const char *name2 = get_column_description(t->desc,
227 if (cd->name && name2 && !strcmp(cd->name, name2))
237 * Initialize a struct table from given table description.
239 * \param desc The description of the osl table.
240 * \param table_ptr Result is returned here.
242 * This function performs several sanity checks on \p desc and returns if any
243 * of these tests fail. On success, a struct \p osl_table is allocated and
244 * initialized with data derived from \p desc.
248 * \sa struct osl_table.
250 int init_table_structure(const struct osl_table_description *desc,
251 struct osl_table **table_ptr)
253 const struct osl_column_description *cd;
254 struct osl_table *t = calloc(1, sizeof(*t));
255 int i, ret = -E_OSL_NOMEM, have_disk_storage_name_column = 0;
259 ret = -E_OSL_BAD_TABLE_DESC;
262 DEBUG_LOG("creating table structure for '%s' from table "
263 "description\n", desc->name);
264 ret = -E_OSL_NO_COLUMN_DESC;
265 if (!desc->column_descriptions)
267 ret = -E_OSL_NO_COLUMNS;
268 if (!desc->num_columns)
271 t->columns = calloc(desc->num_columns, sizeof(struct osl_column));
275 FOR_EACH_COLUMN(i, t->desc, cd) {
276 enum osl_storage_type st = cd->storage_type;
277 enum osl_storage_flags sf = cd->storage_flags;
278 struct osl_column *col = &t->columns[i];
280 ret = -E_OSL_BAD_STORAGE_TYPE;
281 if (st != OSL_MAPPED_STORAGE && st != OSL_DISK_STORAGE
282 && st != OSL_NO_STORAGE)
284 ret = -E_OSL_BAD_STORAGE_FLAGS;
285 if (st == OSL_DISK_STORAGE && sf & OSL_RBTREE)
287 if ((sf & OSL_RBTREE) && !(sf & OSL_UNIQUE))
288 WARNING_LOG("invalid storage flags for column %s: "
289 "OSL_RBTREE && !OSL_UNIQUE\n", cd->name);
290 ret = -E_OSL_BAD_STORAGE_SIZE;
291 if (sf & OSL_FIXED_SIZE && !cd->data_size)
294 case OSL_DISK_STORAGE:
295 t->num_disk_storage_columns++;
297 case OSL_MAPPED_STORAGE:
298 t->num_mapped_columns++;
299 col->index_offset = t->row_index_size;
300 t->row_index_size += 8;
303 col->volatile_num = t->num_volatile_columns;
304 t->num_volatile_columns++;
307 if (sf & OSL_RBTREE) {
308 col->rbtree_num = t->num_rbtrees;
310 if ((sf & OSL_UNIQUE) && (st == OSL_MAPPED_STORAGE)) {
311 if (!have_disk_storage_name_column)
312 t->disk_storage_name_column = i;
313 have_disk_storage_name_column = 1;
317 ret = -E_OSL_NO_UNIQUE_RBTREE_COLUMN;
318 if (t->num_disk_storage_columns && !have_disk_storage_name_column)
320 ret = -E_OSL_NO_RBTREE_COL;
324 DEBUG_LOG("OK. Index entry size: %u\n", t->row_index_size);
325 ret = init_column_descriptions(t);
337 * Read the table description from index header.
339 * \param map The memory mapping of the index file.
340 * \param desc The values found in the index header are returned here.
342 * Read the index header, check for the osl magic string and the table version
343 * number. Read all information stored in the index header into \a desc.
347 * \sa struct osl_table_description, osl_create_table.
349 int read_table_desc(struct osl_object *map, struct osl_table_description *desc)
351 char *buf = map->data;
352 uint8_t version, compat_version, create_version;
353 uint16_t header_size;
356 struct osl_column_description *cd;
358 if (map->size < MIN_INDEX_HEADER_SIZE(1))
359 return -E_OSL_SHORT_TABLE;
360 if (strncmp(buf + IDX_OSL_MAGIC, OSL_MAGIC, strlen(OSL_MAGIC)))
361 return -E_OSL_NO_MAGIC;
362 version = read_u8(buf + IDX_VERSION);
364 * The on-disk version consists of two version numbers: the
365 * create_version (low 4 bits) is the CURRENT_TABLE_VERSION version
366 * number of the library that created the table, and compat_version
367 * (high 4 bits) tells us the lowest version of the library that can
368 * still read this table.
370 create_version = version & 0xf;
371 compat_version = version >> 4;
372 INFO_LOG("create_version: %u, compat_version: %u\n", create_version,
374 if (create_version < MIN_TABLE_VERSION /* table too old */
375 || compat_version > CURRENT_TABLE_VERSION) /* libosl too old */
376 return -E_OSL_VERSION_MISMATCH;
377 desc->flags = read_u8(buf + IDX_TABLE_FLAGS);
378 desc->num_columns = read_u16(buf + IDX_NUM_COLUMNS);
379 INFO_LOG("%u columns\n", desc->num_columns);
380 if (!desc->num_columns)
381 return -E_OSL_NO_COLUMNS;
382 header_size = read_u16(buf + IDX_HEADER_SIZE);
383 if (map->size < header_size)
384 return -E_OSL_BAD_SIZE;
385 desc->column_descriptions = calloc(desc->num_columns,
386 sizeof(struct osl_column_description));
387 if (!desc->column_descriptions)
389 offset = IDX_COLUMN_DESCRIPTIONS;
390 FOR_EACH_COLUMN(i, desc, cd) {
393 ret = -E_OSL_SHORT_TABLE;
394 if (map->size < offset + MIN_IDX_COLUMN_DESCRIPTION_SIZE) {
395 ERROR_LOG("map size = %zu < %u = offset + min desc size\n",
396 map->size, offset + MIN_IDX_COLUMN_DESCRIPTION_SIZE);
399 cd->storage_type = read_u16(buf + offset + IDX_CD_STORAGE_TYPE);
400 cd->storage_flags = read_u16(buf + offset +
401 IDX_CD_STORAGE_FLAGS);
402 cd->data_size = read_u32(buf + offset + IDX_CD_DATA_SIZE);
403 null_byte = memchr(buf + offset + IDX_CD_NAME, '\0',
404 map->size - offset - IDX_CD_NAME);
405 ret = -E_OSL_INDEX_CORRUPTION;
409 cd->name = strdup(buf + offset + IDX_CD_NAME);
412 offset += index_column_description_size(cd->name);
414 if (offset != header_size) {
415 ret = -E_OSL_INDEX_CORRUPTION;
416 ERROR_LOG("real header size = %u != %u = stored header size\n",
417 offset, header_size);
422 FOR_EACH_COLUMN(i, desc, cd)
428 * check whether the table description given by \p t->desc matches the on-disk
429 * table structure stored in the index of \a t.
431 static int compare_table_descriptions(struct osl_table *t)
434 struct osl_table_description desc;
435 const struct osl_column_description *cd1, *cd2;
437 /* read the on-disk structure into desc */
438 ret = read_table_desc(&t->index_map, &desc);
441 ret = -E_OSL_BAD_TABLE_FLAGS;
442 if (desc.flags != t->desc->flags)
444 ret = -E_OSL_BAD_COLUMN_NUM;
445 if (desc.num_columns > t->desc->num_columns)
447 if (desc.num_columns < t->desc->num_columns) {
448 struct osl_column_description *cd;
449 unsigned diff = t->desc->num_columns - desc.num_columns;
450 INFO_LOG("extending table by %u volatile columns\n", diff);
452 desc.column_descriptions = realloc(desc.column_descriptions,
453 t->desc->num_columns * sizeof(struct osl_column_description));
454 if (!desc.column_descriptions)
456 for (i = desc.num_columns; i < t->desc->num_columns; i++) {
457 cd = get_column_description(&desc, i);
458 cd->storage_type = OSL_NO_STORAGE;
461 desc.num_columns += diff;
463 FOR_EACH_COLUMN(i, t->desc, cd1) {
464 cd2 = get_column_description(&desc, i);
465 ret = -E_OSL_BAD_STORAGE_TYPE;
466 if (cd1->storage_type != cd2->storage_type)
468 if (cd1->storage_type == OSL_NO_STORAGE)
470 ret = -E_OSL_BAD_STORAGE_FLAGS;
471 if (cd1->storage_flags != cd2->storage_flags) {
472 ERROR_LOG("sf1 = %u != %u = sf2\n",
473 cd1->storage_flags, cd2->storage_flags);
476 ret = -E_OSL_BAD_DATA_SIZE;
477 if (cd1->storage_flags & OSL_FIXED_SIZE)
478 if (cd1->data_size != cd2->data_size)
480 ret = -E_OSL_BAD_COLUMN_NAME;
481 if (strcmp(cd1->name, cd2->name))
484 INFO_LOG("table description of '%s' matches on-disk data, good\n",
488 FOR_EACH_COLUMN(i, &desc, cd1)
490 free(desc.column_descriptions);
494 static int create_table_index(struct osl_table *t)
496 char *buf, *filename;
498 size_t size = t->index_header_size;
499 const struct osl_column_description *cd;
502 INFO_LOG("creating %zu byte index for table %s\n", size,
504 buf = calloc(1, size);
507 sprintf(buf + IDX_OSL_MAGIC, "%s", OSL_MAGIC);
508 write_u8(buf + IDX_TABLE_FLAGS, t->desc->flags);
509 write_u8(buf + IDX_DIRTY_FLAG, 0);
510 write_u8(buf + IDX_VERSION, CURRENT_TABLE_VERSION
511 + (COMPAT_TABLE_VERSION << 4));
512 write_u16(buf + IDX_NUM_COLUMNS, t->num_mapped_columns + t->num_disk_storage_columns);
513 write_u16(buf + IDX_HEADER_SIZE, t->index_header_size);
514 offset = IDX_COLUMN_DESCRIPTIONS;
515 FOR_EACH_COLUMN(i, t->desc, cd) {
516 /* no need to store info about volatile storage */
517 if (cd->storage_type == OSL_NO_STORAGE)
519 write_u16(buf + offset + IDX_CD_STORAGE_TYPE,
521 write_u16(buf + offset + IDX_CD_STORAGE_FLAGS,
523 if (cd->storage_flags & OSL_FIXED_SIZE)
524 write_u32(buf + offset + IDX_CD_DATA_SIZE,
526 strcpy(buf + offset + IDX_CD_NAME, cd->name);
527 offset += index_column_description_size(cd->name);
529 assert(offset == size);
530 filename = index_filename(t->desc);
532 ret = write_file(filename, buf, size);
540 __export int osl_create_table(const struct osl_table_description *desc)
542 const struct osl_column_description *cd;
543 char *table_dir = NULL, *filename;
545 int i, ret = init_table_structure(desc, &t);
549 INFO_LOG("creating %s\n", desc->name);
550 FOR_EACH_COLUMN(i, t->desc, cd) {
551 if (cd->storage_type == OSL_NO_STORAGE)
554 ret = osl_mkdir(desc->dir, 0777);
555 if (ret < 0 && ret != -E_OSL_DIR_EXISTS)
557 table_dir = make_message("%s/%s", desc->dir,
562 ret = osl_mkdir(table_dir, 0777);
567 filename = column_filename(t, i);
570 INFO_LOG("filename: %s\n", filename);
571 if (cd->storage_type == OSL_MAPPED_STORAGE) {
572 ret = osl_open(filename, O_RDWR | O_CREAT | O_EXCL,
581 ret = osl_mkdir(filename, 0777);
586 if (t->num_mapped_columns) {
587 ret = create_table_index(t);
599 static int table_is_dirty(struct osl_table *t)
601 char *buf = (char *)t->index_map.data + IDX_DIRTY_FLAG;
602 uint8_t dirty = read_u8(buf) & 0x1;
606 static void mark_table_dirty(struct osl_table *t)
608 char *buf = (char *)t->index_map.data + IDX_DIRTY_FLAG;
609 write_u8(buf, read_u8(buf) | 1);
612 static void mark_table_clean(struct osl_table *t)
614 char *buf = (char *)t->index_map.data + IDX_DIRTY_FLAG;
615 write_u8(buf, read_u8(buf) & 0xfe);
618 static void unmap_column(struct osl_table *t, unsigned col_num)
620 struct osl_object map = t->columns[col_num].data_map;
624 ret = osl_munmap(map.data, map.size);
630 * Unmap all mapped files of an osl table.
632 * \param t Pointer to a mapped table.
633 * \param flags Options for unmapping.
635 * \return Positive on success, negative on errors.
637 * \sa map_table(), enum osl_close_flags, osl_munmap().
639 int unmap_table(struct osl_table *t, enum osl_close_flags flags)
642 const struct osl_column_description *cd;
645 if (!t->num_mapped_columns) /* can this ever happen? */
647 INFO_LOG("unmapping table '%s'\n", t->desc->name);
648 if (!t->index_map.data)
649 return -E_OSL_NOT_MAPPED;
650 if (flags & OSL_MARK_CLEAN)
652 ret = osl_munmap(t->index_map.data, t->index_map.size);
655 t->index_map.data = NULL;
658 FOR_EACH_MAPPED_COLUMN(i, t, cd)
663 static int map_column(struct osl_table *t, unsigned col_num)
666 char *filename = column_filename(t, col_num);
671 ret = osl_stat(filename, &statbuf);
676 if (!(S_IFREG & statbuf.st_mode)) {
680 ret = mmap_full_file(filename, O_RDWR,
681 &t->columns[col_num].data_map.data,
682 &t->columns[col_num].data_map.size,
689 * Map the index file and all columns of type \p OSL_MAPPED_STORAGE into memory.
691 * \param t Pointer to an initialized table structure.
692 * \param flags Mapping options.
694 * \return Negative return value on errors; on success the number of rows
695 * (including invalid rows) is returned.
697 * \sa unmap_table(), enum map_table_flags, osl_open_table(), mmap(2).
699 int map_table(struct osl_table *t, enum map_table_flags flags)
702 const struct osl_column_description *cd;
703 int i = 0, ret, num_rows = 0;
705 if (!t->num_mapped_columns)
707 if (t->index_map.data)
708 return -E_OSL_ALREADY_MAPPED;
709 filename = index_filename(t->desc);
712 INFO_LOG("mapping table '%s' (index: %s)\n", t->desc->name, filename);
713 ret = mmap_full_file(filename, flags & MAP_TBL_FL_MAP_RDONLY?
714 O_RDONLY : O_RDWR, &t->index_map.data, &t->index_map.size, NULL);
718 if (flags & MAP_TBL_FL_VERIFY_INDEX) {
719 ret = compare_table_descriptions(t);
724 if (!(flags & MAP_TBL_FL_IGNORE_DIRTY)) {
725 if (table_is_dirty(t)) {
726 ERROR_LOG("%s is dirty\n", t->desc->name);
731 num_rows = table_num_rows(t);
735 FOR_EACH_MAPPED_COLUMN(i, t, cd) {
736 ret = map_column(t, i);
741 err: /* unmap what is already mapped */
742 for (i--; i >= 0; i--) {
743 struct osl_object map = t->columns[i].data_map;
744 osl_munmap(map.data, map.size);
747 osl_munmap(t->index_map.data, t->index_map.size);
748 t->index_map.data = NULL;
753 * Retrieve a mapped object by row and column number.
755 * \param t Pointer to an open osl table.
756 * \param col_num Number of the mapped column containing the object to retrieve.
757 * \param row_num Number of the row containing the object to retrieve.
758 * \param obj The result is returned here.
760 * It is considered an error if \a col_num does not refer to a column
761 * of storage type \p OSL_MAPPED_STORAGE.
765 * \sa osl_storage_type.
767 int get_mapped_object(const struct osl_table *t, unsigned col_num,
768 uint32_t row_num, struct osl_object *obj)
770 struct osl_column *col = &t->columns[col_num];
775 if (t->num_rows <= row_num)
776 return -E_OSL_BAD_ROW_NUM;
777 ret = get_cell_index(t, row_num, col_num, &cell_index);
780 offset = read_u32(cell_index);
781 obj->size = read_u32(cell_index + 4);
782 obj->data = col->data_map.data + offset;
787 * It's OK to call this with result = rb_node = NULL. If result is not NULL,
788 * and rb key was not found, result points to the parent node.
790 static int search_rbtree(const struct osl_object *obj,
791 const struct osl_table *t, unsigned col_num,
792 struct rb_node **result, struct rb_node ***rb_link)
794 struct osl_column *col = &t->columns[col_num];
795 struct rb_node **new = &col->rbtree.rb_node, *parent = NULL;
796 const struct osl_column_description *cd =
797 get_column_description(t->desc, col_num);
798 enum osl_storage_type st = cd->storage_type;
800 struct osl_row *this_row = get_row_pointer(*new,
803 struct osl_object this_obj;
805 if (st == OSL_MAPPED_STORAGE) {
806 ret = get_mapped_object(t, col_num, this_row->num,
811 this_obj = this_row->volatile_objects[col->volatile_num];
812 ret = cd->compare_function(obj, &this_obj);
815 *result = get_rb_node_pointer(this_row,
820 new = &((*new)->rb_left);
822 new = &((*new)->rb_right);
828 return -E_OSL_RB_KEY_NOT_FOUND;
831 static int insert_rbtree(struct osl_table *t, unsigned col_num,
832 const struct osl_row *row, const struct osl_object *obj)
834 struct rb_node *parent, **rb_link;
837 int ret = search_rbtree(obj, t, col_num, &parent, &rb_link);
840 return -E_OSL_RB_KEY_EXISTS;
841 rbtree_num = t->columns[col_num].rbtree_num;
842 n = get_rb_node_pointer(row, rbtree_num);
843 rb_link_node(n, parent, rb_link);
844 rb_insert_color(n, &t->columns[col_num].rbtree);
848 static void remove_rb_node(struct osl_table *t, unsigned col_num,
849 const struct osl_row *row)
851 struct osl_column *col = &t->columns[col_num];
852 const struct osl_column_description *cd =
853 get_column_description(t->desc, col_num);
854 enum osl_storage_flags sf = cd->storage_flags;
855 struct rb_node *victim, *splice_out_node, *tmp;
856 if (!(sf & OSL_RBTREE))
859 * Which node is removed/spliced out actually depends on how many
860 * children the victim node has: If it has no children, it gets
861 * deleted. If it has one child, it gets spliced out. If it has two
862 * children, its successor (which has at most a right child) gets
865 victim = get_rb_node_pointer(row, col->rbtree_num);
866 if (victim->rb_left && victim->rb_right)
867 splice_out_node = rb_next(victim);
869 splice_out_node = victim;
870 /* Go up to the root and decrement the size of each node in the path. */
871 for (tmp = splice_out_node; tmp; tmp = rb_parent(tmp))
873 rb_erase(victim, &col->rbtree);
876 static int add_row_to_rbtrees(struct osl_table *t, uint32_t row_num,
877 struct osl_object *volatile_objs, struct osl_row **row_ptr)
881 struct osl_row *row = allocate_row(t->num_rbtrees);
882 const struct osl_column_description *cd;
887 row->volatile_objects = volatile_objs;
888 FOR_EACH_RBTREE_COLUMN(i, t, cd) {
889 if (cd->storage_type == OSL_MAPPED_STORAGE) {
890 struct osl_object obj;
891 ret = get_mapped_object(t, i, row_num, &obj);
894 ret = insert_rbtree(t, i, row, &obj);
895 } else { /* volatile */
896 const struct osl_object *obj
897 = volatile_objs + t->columns[i].volatile_num;
898 ret = insert_rbtree(t, i, row, obj);
906 err: /* rollback changes, i.e. remove added entries from rbtrees */
908 remove_rb_node(t, i--, row);
913 static void free_volatile_objects(const struct osl_table *t,
914 enum osl_close_flags flags)
918 struct osl_column *rb_col;
919 const struct osl_column_description *cd;
921 if (!t->num_volatile_columns)
923 /* find the first rbtree column (any will do) */
924 FOR_EACH_RBTREE_COLUMN(i, t, cd)
926 rb_col = t->columns + i;
927 /* walk that rbtree and free all volatile objects */
928 for (n = rb_first(&rb_col->rbtree); n; n = rb_next(n)) {
929 struct osl_row *r = get_row_pointer(n, rb_col->rbtree_num);
930 if (flags & OSL_FREE_VOLATILE)
931 FOR_EACH_VOLATILE_COLUMN(j, t, cd) {
932 if (cd->storage_flags & OSL_DONT_FREE)
934 free(r->volatile_objects[
935 t->columns[j].volatile_num].data);
937 // for (j = 0; j < t->num_volatile_columns; j++)
938 // free(r->volatile_objects[j].data);
939 free(r->volatile_objects);
944 * Erase all rbtree nodes and free resources.
946 * \param t Pointer to an open osl table.
948 * This function is called by osl_close_table().
950 void clear_rbtrees(struct osl_table *t)
952 const struct osl_column_description *cd;
953 unsigned i, rbtrees_cleared = 0;
955 FOR_EACH_RBTREE_COLUMN(i, t, cd) {
956 struct osl_column *col = &t->columns[i];
959 for (n = rb_first(&col->rbtree); n;) {
961 rb_erase(n, &col->rbtree);
962 if (rbtrees_cleared == t->num_rbtrees) {
963 r = get_row_pointer(n, col->rbtree_num);
973 __export int osl_close_table(struct osl_table *t, enum osl_close_flags flags)
978 return -E_OSL_BAD_TABLE;
979 NOTICE_LOG("closing table %s\n", t->desc->name);
980 free_volatile_objects(t, flags);
982 ret = unmap_table(t, flags);
984 ERROR_LOG("unmap_table failed: %d\n", ret);
991 * Find out whether the given row number corresponds to an invalid row.
993 * \param t Pointer to the osl table.
994 * \param row_num The number of the row in question.
996 * By definition, a row is considered invalid if all its index entries
999 * \return Positive if \a row_num corresponds to an invalid row,
1000 * zero if it corresponds to a valid row, negative on errors.
1002 int row_is_invalid(struct osl_table *t, uint32_t row_num)
1005 int i, ret = get_row_index(t, row_num, &row_index);
1009 for (i = 0; i < t->row_index_size; i++) {
1010 if ((unsigned char)row_index[i] != 0xff)
1013 INFO_LOG("row %d is invalid\n", row_num);
1018 * Invalidate a row of an osl table.
1020 * \param t Pointer to an open osl table.
1021 * \param row_num Number of the row to mark as invalid.
1023 * This function marks each mapped object in the index entry of \a row as
1028 int mark_row_invalid(struct osl_table *t, uint32_t row_num)
1031 int ret = get_row_index(t, row_num, &row_index);
1035 INFO_LOG("marking row %d as invalid\n", row_num);
1036 memset(row_index, 0xff, t->row_index_size);
1041 * Initialize all rbtrees and compute number of invalid rows.
1043 * \param t The table containing the rbtrees to be initialized.
1047 int init_rbtrees(struct osl_table *t)
1050 const struct osl_column_description *cd;
1052 /* create rbtrees */
1053 FOR_EACH_RBTREE_COLUMN(i, t, cd)
1054 t->columns[i].rbtree = RB_ROOT;
1055 /* add valid rows to rbtrees */
1056 t->num_invalid_rows = 0;
1057 for (i = 0; i < t->num_rows; i++) {
1058 struct osl_object *volatile_objs;
1059 ret = row_is_invalid(t, i);
1063 t->num_invalid_rows++;
1066 if (t->num_volatile_columns > 0) {
1067 volatile_objs = calloc(t->num_volatile_columns,
1068 sizeof(struct osl_object));
1070 return -E_OSL_NOMEM;
1072 volatile_objs = NULL;
1073 ret = add_row_to_rbtrees(t, i, volatile_objs, NULL);
1080 __export int osl_open_table(const struct osl_table_description *table_desc,
1081 struct osl_table **result)
1084 struct osl_table *t;
1085 const struct osl_column_description *cd;
1087 NOTICE_LOG("opening table %s\n", table_desc->name);
1088 ret = init_table_structure(table_desc, &t);
1091 FOR_EACH_DISK_STORAGE_COLUMN(i, t, cd) {
1092 struct stat statbuf;
1093 char *dirname = column_filename(t, i);
1098 /* check if directory exists */
1099 ret = osl_stat(dirname, &statbuf);
1103 ret = -E_OSL_NOTDIR;
1104 if (!S_ISDIR(statbuf.st_mode))
1107 ret = map_table(t, MAP_TBL_FL_VERIFY_INDEX);
1111 DEBUG_LOG("num rows: %d\n", t->num_rows);
1112 ret = init_rbtrees(t);
1114 osl_close_table(t, OSL_MARK_CLEAN); /* ignore further errors */
1125 static int create_disk_storage_object_dir(const struct osl_table *t,
1126 unsigned col_num, const char *ds_name)
1131 if (!(t->desc->flags & OSL_LARGE_TABLE))
1133 dirname = disk_storage_dirname(t, col_num, ds_name);
1135 return -E_OSL_NOMEM;
1136 ret = osl_mkdir(dirname, 0777);
1138 if (ret < 0 && ret != -E_OSL_DIR_EXISTS)
1143 static int write_disk_storage_file(const struct osl_table *t, unsigned col_num,
1144 const struct osl_object *obj, const char *ds_name)
1149 ret = create_disk_storage_object_dir(t, col_num, ds_name);
1152 filename = disk_storage_path(t, col_num, ds_name);
1154 return -E_OSL_NOMEM;
1155 ret = write_file(filename, obj->data, obj->size);
1160 static int append_map_file(const struct osl_table *t, unsigned col_num,
1161 const struct osl_object *obj, uint32_t *new_size)
1163 char *filename = column_filename(t, col_num);
1167 return -E_OSL_NOMEM;
1168 ret = append_file(filename, obj->data, obj->size, new_size);
1173 static int append_row_index(const struct osl_table *t, char *row_index)
1178 if (!t->num_mapped_columns)
1180 filename = index_filename(t->desc);
1182 return -E_OSL_NOMEM;
1183 ret = append_file(filename, row_index, t->row_index_size, NULL);
1188 static int truncate_mapped_file(const struct osl_table *t, unsigned col_num,
1192 char *filename = column_filename(t, col_num);
1195 return -E_OSL_NOMEM;
1196 ret = truncate_file(filename, size);
1201 static int delete_disk_storage_file(const struct osl_table *t, unsigned col_num,
1202 const char *ds_name)
1204 char *dirname, *filename = disk_storage_path(t, col_num, ds_name);
1208 return -E_OSL_NOMEM;
1209 if (unlink(filename) < 0)
1210 ret = errno == ENOENT? -E_OSL_NOENT : -E_OSL_UNLINK;
1214 if (!(t->desc->flags & OSL_LARGE_TABLE))
1216 dirname = disk_storage_dirname(t, col_num, ds_name);
1218 return -E_OSL_NOMEM;
1224 __export int osl_add_and_get_row(struct osl_table *t, struct osl_object *objects,
1225 struct osl_row **row)
1228 char *ds_name = NULL;
1229 struct rb_node **rb_parents = NULL, ***rb_links = NULL;
1230 char *new_row_index = NULL;
1231 struct osl_object *volatile_objs = NULL;
1232 const struct osl_column_description *cd;
1235 return -E_OSL_BAD_TABLE;
1236 rb_parents = malloc(t->num_rbtrees * sizeof(struct rn_node*));
1238 return -E_OSL_NOMEM;
1239 rb_links = malloc(t->num_rbtrees * sizeof(struct rn_node**));
1242 return -E_OSL_NOMEM;
1244 if (t->num_mapped_columns) {
1245 new_row_index = malloc(t->row_index_size);
1246 if (!new_row_index) {
1249 return -E_OSL_NOMEM;
1252 /* pass 1: sanity checks */
1253 // DEBUG_LOG("sanity tests: %p:%p\n", objects[0].data,
1254 // objects[1].data);
1255 FOR_EACH_COLUMN(i, t->desc, cd) {
1256 enum osl_storage_type st = cd->storage_type;
1257 enum osl_storage_flags sf = cd->storage_flags;
1259 // ret = -E_OSL_NULL_OBJECT;
1262 if (st == OSL_DISK_STORAGE)
1264 if (sf & OSL_RBTREE) {
1265 unsigned rbtree_num = t->columns[i].rbtree_num;
1266 ret = -E_OSL_RB_KEY_EXISTS;
1267 // DEBUG_LOG("checking whether %p exists\n",
1268 // objects[i].data);
1269 if (search_rbtree(objects + i, t, i,
1270 &rb_parents[rbtree_num],
1271 &rb_links[rbtree_num]) > 0)
1274 if (sf & OSL_FIXED_SIZE) {
1275 // DEBUG_LOG("fixed size. need: %zu, have: %d\n",
1276 // objects[i].size, cd->data_size);
1277 ret = -E_OSL_BAD_DATA_SIZE;
1278 if (objects[i].size != cd->data_size)
1282 if (t->num_disk_storage_columns) {
1283 ds_name = disk_storage_name_of_object(t,
1284 &objects[t->disk_storage_name_column]);
1289 ret = unmap_table(t, OSL_MARK_CLEAN);
1292 // DEBUG_LOG("sanity tests passed%s\n", "");
1293 /* pass 2: create data files, append map data */
1294 FOR_EACH_COLUMN(i, t->desc, cd) {
1295 enum osl_storage_type st = cd->storage_type;
1296 if (st == OSL_NO_STORAGE)
1298 if (st == OSL_MAPPED_STORAGE) {
1300 struct osl_column *col = &t->columns[i];
1301 // DEBUG_LOG("appending object of size %zu\n",
1302 // objects[i].size);
1303 ret = append_map_file(t, i, objects + i, &new_size);
1306 update_cell_index(new_row_index, col, new_size,
1311 ret = write_disk_storage_file(t, i, objects + i, ds_name);
1315 ret = append_row_index(t, new_row_index);
1318 ret = map_table(t, MAP_TBL_FL_VERIFY_INDEX);
1319 if (ret < 0) { /* truncate index and rollback changes */
1320 char *filename = index_filename(t->desc);
1322 truncate_file(filename, t->row_index_size);
1326 /* pass 3: add entry to rbtrees */
1327 if (t->num_volatile_columns) {
1329 volatile_objs = calloc(t->num_volatile_columns,
1330 sizeof(struct osl_object));
1333 FOR_EACH_VOLATILE_COLUMN(i, t, cd)
1334 volatile_objs[t->columns[i].volatile_num] = objects[i];
1337 // DEBUG_LOG("adding new entry as row #%d\n", t->num_rows - 1);
1338 ret = add_row_to_rbtrees(t, t->num_rows - 1, volatile_objs, row);
1341 // DEBUG_LOG("added new entry as row #%d\n", t->num_rows - 1);
1344 rollback: /* rollback all changes made, ignore further errors */
1345 for (i--; i >= 0; i--) {
1346 cd = get_column_description(t->desc, i);
1347 enum osl_storage_type st = cd->storage_type;
1348 if (st == OSL_NO_STORAGE)
1351 if (st == OSL_MAPPED_STORAGE)
1352 truncate_mapped_file(t, i, objects[i].size);
1353 else /* disk storage */
1354 delete_disk_storage_file(t, i, ds_name);
1356 /* ignore error and return previous error value */
1357 map_table(t, MAP_TBL_FL_VERIFY_INDEX);
1359 free(new_row_index);
1366 __export int osl_add_row(struct osl_table *t, struct osl_object *objects)
1368 return osl_add_and_get_row(t, objects, NULL);
1371 __export int osl_get_object(const struct osl_table *t, const struct osl_row *r,
1372 unsigned col_num, struct osl_object *object)
1374 const struct osl_column_description *cd;
1377 return -E_OSL_BAD_TABLE;
1378 cd = get_column_description(t->desc, col_num);
1379 /* col must not be disk storage */
1380 if (cd->storage_type == OSL_DISK_STORAGE)
1381 return -E_OSL_BAD_STORAGE_TYPE;
1382 if (cd->storage_type == OSL_MAPPED_STORAGE)
1383 return get_mapped_object(t, col_num, r->num, object);
1385 *object = r->volatile_objects[t->columns[col_num].volatile_num];
1389 __export int osl_del_row(struct osl_table *t, struct osl_row *row)
1391 struct osl_row *r = row;
1393 const struct osl_column_description *cd;
1396 return -E_OSL_BAD_TABLE;
1397 INFO_LOG("deleting row %p\n", row);
1399 if (t->num_disk_storage_columns) {
1401 ret = disk_storage_name_of_row(t, r, &ds_name);
1404 FOR_EACH_DISK_STORAGE_COLUMN(i, t, cd)
1405 delete_disk_storage_file(t, i, ds_name);
1408 FOR_EACH_COLUMN(i, t->desc, cd) {
1409 struct osl_column *col = t->columns + i;
1410 enum osl_storage_type st = cd->storage_type;
1411 remove_rb_node(t, i, r);
1412 if (st == OSL_MAPPED_STORAGE)
1414 if (st == OSL_NO_STORAGE && !(cd->storage_flags & OSL_DONT_FREE))
1415 free(r->volatile_objects[col->volatile_num].data);
1417 if (t->num_mapped_columns) {
1418 ret = mark_row_invalid(t, r->num);
1421 t->num_invalid_rows++;
1426 free(r->volatile_objects);
1431 /* test if column has an rbtree */
1432 static int check_rbtree_col(const struct osl_table *t, unsigned col_num,
1433 struct osl_column **col)
1436 return -E_OSL_BAD_TABLE;
1437 if (!(get_column_description(t->desc, col_num)->storage_flags & OSL_RBTREE))
1438 return -E_OSL_BAD_STORAGE_FLAGS;
1439 *col = t->columns + col_num;
1443 __export int osl_get_row(const struct osl_table *t, unsigned col_num,
1444 const struct osl_object *obj, struct osl_row **result)
1447 struct rb_node *node;
1448 struct osl_row *row;
1449 struct osl_column *col;
1452 ret = check_rbtree_col(t, col_num, &col);
1455 ret = search_rbtree(obj, t, col_num, &node, NULL);
1458 row = get_row_pointer(node, t->columns[col_num].rbtree_num);
1463 static int rbtree_loop(struct osl_column *col, void *private_data,
1464 osl_rbtree_loop_func *func)
1466 struct rb_node *n, *tmp;
1468 /* this for-loop is safe against removal of an entry */
1469 for (n = rb_first(&col->rbtree), tmp = n? rb_next(n) : NULL;
1471 n = tmp, tmp = tmp? rb_next(tmp) : NULL) {
1472 struct osl_row *r = get_row_pointer(n, col->rbtree_num);
1473 if (func(r, private_data) < 0)
1479 static int rbtree_loop_reverse(struct osl_column *col, void *private_data,
1480 osl_rbtree_loop_func *func)
1482 struct rb_node *n, *tmp;
1484 /* safe against removal of an entry */
1485 for (n = rb_last(&col->rbtree), tmp = n? rb_prev(n) : NULL;
1487 n = tmp, tmp = tmp? rb_prev(tmp) : NULL) {
1488 struct osl_row *r = get_row_pointer(n, col->rbtree_num);
1489 if (func(r, private_data) < 0)
1495 __export int osl_rbtree_loop(const struct osl_table *t, unsigned col_num,
1496 void *private_data, osl_rbtree_loop_func *func)
1498 struct osl_column *col;
1500 int ret = check_rbtree_col(t, col_num, &col);
1503 return rbtree_loop(col, private_data, func);
1506 __export int osl_rbtree_loop_reverse(const struct osl_table *t, unsigned col_num,
1507 void *private_data, osl_rbtree_loop_func *func)
1509 struct osl_column *col;
1511 int ret = check_rbtree_col(t, col_num, &col);
1514 return rbtree_loop_reverse(col, private_data, func);
1517 /* TODO: Rollback changes on errors */
1518 static int rename_disk_storage_objects(struct osl_table *t,
1519 struct osl_object *old_obj, struct osl_object *new_obj)
1522 const struct osl_column_description *cd;
1523 char *old_ds_name, *new_ds_name;
1525 if (!t->num_disk_storage_columns)
1526 return 1; /* nothing to do */
1527 if (old_obj->size == new_obj->size && !memcmp(new_obj->data,
1528 old_obj->data, new_obj->size))
1529 return 1; /* object did not change */
1530 old_ds_name = disk_storage_name_of_object(t, old_obj);
1531 new_ds_name = disk_storage_name_of_object(t, new_obj);
1533 if (!old_ds_name || ! new_ds_name)
1536 FOR_EACH_DISK_STORAGE_COLUMN(i, t, cd) {
1537 char *old_filename, *new_filename;
1538 ret = create_disk_storage_object_dir(t, i, new_ds_name);
1541 old_filename = disk_storage_path(t, i, old_ds_name);
1542 new_filename = disk_storage_path(t, i, new_ds_name);
1543 if (!old_filename || !new_filename)
1546 ret = osl_rename(old_filename, new_filename);
1560 __export int osl_update_object(struct osl_table *t, const struct osl_row *r,
1561 unsigned col_num, struct osl_object *obj)
1563 struct osl_column *col;
1564 const struct osl_column_description *cd;
1568 return -E_OSL_BAD_TABLE;
1569 col = &t->columns[col_num];
1570 cd = get_column_description(t->desc, col_num);
1571 DEBUG_LOG("updating column %u of %s\n", col_num, t->desc->name);
1572 if (cd->storage_flags & OSL_RBTREE) {
1573 if (search_rbtree(obj, t, col_num, NULL, NULL) > 0)
1574 return -E_OSL_RB_KEY_EXISTS;
1576 if (cd->storage_flags & OSL_FIXED_SIZE) {
1577 if (obj->size != cd->data_size)
1578 return -E_OSL_BAD_DATA_SIZE;
1580 remove_rb_node(t, col_num, r);
1581 if (cd->storage_type == OSL_NO_STORAGE) { /* TODO: If fixed size, reuse object? */
1582 if (!(cd->storage_flags & OSL_DONT_FREE))
1583 free(r->volatile_objects[col->volatile_num].data);
1584 r->volatile_objects[col->volatile_num] = *obj;
1585 } else if (cd->storage_type == OSL_DISK_STORAGE) {
1587 ret = disk_storage_name_of_row(t, r, &ds_name);
1590 ret = delete_disk_storage_file(t, col_num, ds_name);
1591 if (ret < 0 && ret != -E_OSL_NOENT) {
1595 ret = write_disk_storage_file(t, col_num, obj, ds_name);
1599 } else { /* mapped storage */
1600 struct osl_object old_obj;
1601 ret = get_mapped_object(t, col_num, r->num, &old_obj);
1605 * If the updated column is the disk storage name column, the
1606 * disk storage name changes, so we have to rename all disk
1607 * storage objects accordingly.
1609 if (col_num == t->disk_storage_name_column) {
1610 ret = rename_disk_storage_objects(t, &old_obj, obj);
1614 if (cd->storage_flags & OSL_FIXED_SIZE)
1615 memcpy(old_obj.data, obj->data, cd->data_size);
1616 else { /* TODO: if the size doesn't change, use old space */
1617 uint32_t new_data_map_size;
1619 ret = get_row_index(t, r->num, &row_index);
1622 unmap_column(t, col_num);
1623 ret = append_map_file(t, col_num, obj,
1624 &new_data_map_size);
1627 ret = map_column(t, col_num);
1630 update_cell_index(row_index, col, new_data_map_size,
1634 if (cd->storage_flags & OSL_RBTREE) {
1635 ret = insert_rbtree(t, col_num, r, obj);
1642 __export int osl_open_disk_object(const struct osl_table *t, const struct osl_row *r,
1643 unsigned col_num, struct osl_object *obj)
1645 const struct osl_column_description *cd;
1646 char *ds_name, *filename;
1650 return -E_OSL_BAD_TABLE;
1651 cd = get_column_description(t->desc, col_num);
1652 if (cd->storage_type != OSL_DISK_STORAGE)
1653 return -E_OSL_BAD_STORAGE_TYPE;
1655 ret = disk_storage_name_of_row(t, r, &ds_name);
1658 filename = disk_storage_path(t, col_num, ds_name);
1661 return -E_OSL_NOMEM;
1662 DEBUG_LOG("filename: %s\n", filename);
1663 ret = mmap_full_file(filename, O_RDONLY, &obj->data, &obj->size, NULL);
1668 __export int osl_close_disk_object(struct osl_object *obj)
1670 return osl_munmap(obj->data, obj->size);
1673 __export int osl_get_num_rows(const struct osl_table *t, unsigned *num_rows)
1676 return -E_OSL_BAD_TABLE;
1677 assert(t->num_rows >= t->num_invalid_rows);
1678 *num_rows = t->num_rows - t->num_invalid_rows;
1682 __export int osl_get_rank(const struct osl_table *t, struct osl_row *r,
1683 unsigned col_num, unsigned *rank)
1685 struct osl_object obj;
1686 struct osl_column *col;
1687 struct rb_node *node;
1688 int ret = check_rbtree_col(t, col_num, &col);
1692 ret = osl_get_object(t, r, col_num, &obj);
1695 ret = search_rbtree(&obj, t, col_num, &node, NULL);
1698 ret = rb_rank(node, rank);
1700 return -E_OSL_BAD_ROW;
1704 __export int osl_get_nth_row(const struct osl_table *t, unsigned col_num,
1705 unsigned n, struct osl_row **result)
1707 struct osl_column *col;
1708 struct rb_node *node;
1714 return -E_OSL_RB_KEY_NOT_FOUND;
1715 ret = osl_get_num_rows(t, &num_rows);
1719 return -E_OSL_RB_KEY_NOT_FOUND;
1720 ret = check_rbtree_col(t, col_num, &col);
1723 node = rb_nth(col->rbtree.rb_node, n);
1725 return -E_OSL_RB_KEY_NOT_FOUND;
1726 *result = get_row_pointer(node, col->rbtree_num);
1730 __export int osl_rbtree_first_row(const struct osl_table *t, unsigned col_num,
1731 struct osl_row **result)
1733 return osl_get_nth_row(t, col_num, 1, result);
1736 __export int osl_rbtree_last_row(const struct osl_table *t, unsigned col_num,
1737 struct osl_row **result)
1740 int ret = osl_get_num_rows(t, &num_rows);
1744 return osl_get_nth_row(t, col_num, num_rows, result);