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 table_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 table_version = read_u8(buf + IDX_VERSION);
363 INFO_LOG("osl versions (table/min/current): %u/%u/%u\n",
364 table_version, MIN_TABLE_VERSION, CURRENT_TABLE_VERSION);
365 if (table_version < MIN_TABLE_VERSION /* table too old */
366 || table_version > CURRENT_TABLE_VERSION) /* libosl too old */
367 return -E_OSL_VERSION_MISMATCH;
368 desc->flags = read_u8(buf + IDX_TABLE_FLAGS);
369 desc->num_columns = read_u16(buf + IDX_NUM_COLUMNS);
370 INFO_LOG("%u columns\n", desc->num_columns);
371 if (!desc->num_columns)
372 return -E_OSL_NO_COLUMNS;
373 header_size = read_u16(buf + IDX_HEADER_SIZE);
374 if (map->size < header_size)
375 return -E_OSL_BAD_SIZE;
376 desc->column_descriptions = calloc(desc->num_columns,
377 sizeof(struct osl_column_description));
378 if (!desc->column_descriptions)
380 offset = IDX_COLUMN_DESCRIPTIONS;
381 FOR_EACH_COLUMN(i, desc, cd) {
384 ret = -E_OSL_SHORT_TABLE;
385 if (map->size < offset + MIN_IDX_COLUMN_DESCRIPTION_SIZE) {
386 ERROR_LOG("map size = %zu < %u = offset + min desc size\n",
387 map->size, offset + MIN_IDX_COLUMN_DESCRIPTION_SIZE);
390 cd->storage_type = read_u16(buf + offset + IDX_CD_STORAGE_TYPE);
391 cd->storage_flags = read_u16(buf + offset +
392 IDX_CD_STORAGE_FLAGS);
393 cd->data_size = read_u32(buf + offset + IDX_CD_DATA_SIZE);
394 null_byte = memchr(buf + offset + IDX_CD_NAME, '\0',
395 map->size - offset - IDX_CD_NAME);
396 ret = -E_OSL_INDEX_CORRUPTION;
400 cd->name = strdup(buf + offset + IDX_CD_NAME);
403 offset += index_column_description_size(cd->name);
405 if (offset != header_size) {
406 ret = -E_OSL_INDEX_CORRUPTION;
407 ERROR_LOG("real header size = %u != %u = stored header size\n",
408 offset, header_size);
413 FOR_EACH_COLUMN(i, desc, cd)
419 * check whether the table description given by \p t->desc matches the on-disk
420 * table structure stored in the index of \a t.
422 static int compare_table_descriptions(struct osl_table *t)
425 struct osl_table_description desc;
426 const struct osl_column_description *cd1, *cd2;
428 /* read the on-disk structure into desc */
429 ret = read_table_desc(&t->index_map, &desc);
432 ret = -E_OSL_BAD_TABLE_FLAGS;
433 if (desc.flags != t->desc->flags)
435 ret = -E_OSL_BAD_COLUMN_NUM;
436 if (desc.num_columns > t->desc->num_columns)
438 if (desc.num_columns < t->desc->num_columns) {
439 struct osl_column_description *cd;
440 unsigned diff = t->desc->num_columns - desc.num_columns;
441 INFO_LOG("extending table by %u volatile columns\n", diff);
443 desc.column_descriptions = realloc(desc.column_descriptions,
444 t->desc->num_columns * sizeof(struct osl_column_description));
445 if (!desc.column_descriptions)
447 for (i = desc.num_columns; i < t->desc->num_columns; i++) {
448 cd = get_column_description(&desc, i);
449 cd->storage_type = OSL_NO_STORAGE;
452 desc.num_columns += diff;
454 FOR_EACH_COLUMN(i, t->desc, cd1) {
455 cd2 = get_column_description(&desc, i);
456 ret = -E_OSL_BAD_STORAGE_TYPE;
457 if (cd1->storage_type != cd2->storage_type)
459 if (cd1->storage_type == OSL_NO_STORAGE)
461 ret = -E_OSL_BAD_STORAGE_FLAGS;
462 if (cd1->storage_flags != cd2->storage_flags) {
463 ERROR_LOG("sf1 = %u != %u = sf2\n",
464 cd1->storage_flags, cd2->storage_flags);
467 ret = -E_OSL_BAD_DATA_SIZE;
468 if (cd1->storage_flags & OSL_FIXED_SIZE)
469 if (cd1->data_size != cd2->data_size)
471 ret = -E_OSL_BAD_COLUMN_NAME;
472 if (strcmp(cd1->name, cd2->name))
475 INFO_LOG("table description of '%s' matches on-disk data, good\n",
479 FOR_EACH_COLUMN(i, &desc, cd1)
481 free(desc.column_descriptions);
485 static int create_table_index(struct osl_table *t)
487 char *buf, *filename;
489 size_t size = t->index_header_size;
490 const struct osl_column_description *cd;
493 INFO_LOG("creating %zu byte index for table %s\n", size,
495 buf = calloc(1, size);
498 sprintf(buf + IDX_OSL_MAGIC, "%s", OSL_MAGIC);
499 write_u8(buf + IDX_TABLE_FLAGS, t->desc->flags);
500 write_u8(buf + IDX_DIRTY_FLAG, 0);
501 write_u8(buf + IDX_VERSION, CURRENT_TABLE_VERSION);
502 write_u16(buf + IDX_NUM_COLUMNS, t->num_mapped_columns + t->num_disk_storage_columns);
503 write_u16(buf + IDX_HEADER_SIZE, t->index_header_size);
504 offset = IDX_COLUMN_DESCRIPTIONS;
505 FOR_EACH_COLUMN(i, t->desc, cd) {
506 /* no need to store info about volatile storage */
507 if (cd->storage_type == OSL_NO_STORAGE)
509 write_u16(buf + offset + IDX_CD_STORAGE_TYPE,
511 write_u16(buf + offset + IDX_CD_STORAGE_FLAGS,
513 if (cd->storage_flags & OSL_FIXED_SIZE)
514 write_u32(buf + offset + IDX_CD_DATA_SIZE,
516 strcpy(buf + offset + IDX_CD_NAME, cd->name);
517 offset += index_column_description_size(cd->name);
519 assert(offset == size);
520 filename = index_filename(t->desc);
522 ret = write_file(filename, buf, size);
530 __export int osl_create_table(const struct osl_table_description *desc)
532 const struct osl_column_description *cd;
533 char *table_dir = NULL, *filename;
535 int i, ret = init_table_structure(desc, &t);
539 INFO_LOG("creating %s\n", desc->name);
540 FOR_EACH_COLUMN(i, t->desc, cd) {
541 if (cd->storage_type == OSL_NO_STORAGE)
544 ret = osl_mkdir(desc->dir, 0777);
545 if (ret < 0 && ret != -E_OSL_DIR_EXISTS)
547 table_dir = make_message("%s/%s", desc->dir,
552 ret = osl_mkdir(table_dir, 0777);
557 filename = column_filename(t, i);
560 INFO_LOG("filename: %s\n", filename);
561 if (cd->storage_type == OSL_MAPPED_STORAGE) {
562 ret = osl_open(filename, O_RDWR | O_CREAT | O_EXCL,
571 ret = osl_mkdir(filename, 0777);
576 if (t->num_mapped_columns) {
577 ret = create_table_index(t);
589 static int table_is_dirty(struct osl_table *t)
591 char *buf = (char *)t->index_map.data + IDX_DIRTY_FLAG;
592 uint8_t dirty = read_u8(buf) & 0x1;
596 static void mark_table_dirty(struct osl_table *t)
598 char *buf = (char *)t->index_map.data + IDX_DIRTY_FLAG;
599 write_u8(buf, read_u8(buf) | 1);
602 static void mark_table_clean(struct osl_table *t)
604 char *buf = (char *)t->index_map.data + IDX_DIRTY_FLAG;
605 write_u8(buf, read_u8(buf) & 0xfe);
608 static void unmap_column(struct osl_table *t, unsigned col_num)
610 struct osl_object map = t->columns[col_num].data_map;
614 ret = osl_munmap(map.data, map.size);
620 * Unmap all mapped files of an osl table.
622 * \param t Pointer to a mapped table.
623 * \param flags Options for unmapping.
625 * \return Positive on success, negative on errors.
627 * \sa map_table(), enum osl_close_flags, osl_munmap().
629 int unmap_table(struct osl_table *t, enum osl_close_flags flags)
632 const struct osl_column_description *cd;
635 if (!t->num_mapped_columns) /* can this ever happen? */
637 INFO_LOG("unmapping table '%s'\n", t->desc->name);
638 if (!t->index_map.data)
639 return -E_OSL_NOT_MAPPED;
640 if (flags & OSL_MARK_CLEAN)
642 ret = osl_munmap(t->index_map.data, t->index_map.size);
645 t->index_map.data = NULL;
648 FOR_EACH_MAPPED_COLUMN(i, t, cd)
653 static int map_column(struct osl_table *t, unsigned col_num)
656 char *filename = column_filename(t, col_num);
661 ret = osl_stat(filename, &statbuf);
666 if (!(S_IFREG & statbuf.st_mode)) {
670 ret = mmap_full_file(filename, O_RDWR,
671 &t->columns[col_num].data_map.data,
672 &t->columns[col_num].data_map.size,
679 * Map the index file and all columns of type \p OSL_MAPPED_STORAGE into memory.
681 * \param t Pointer to an initialized table structure.
682 * \param flags Mapping options.
684 * \return Negative return value on errors; on success the number of rows
685 * (including invalid rows) is returned.
687 * \sa unmap_table(), enum map_table_flags, osl_open_table(), mmap(2).
689 int map_table(struct osl_table *t, enum map_table_flags flags)
692 const struct osl_column_description *cd;
693 int i = 0, ret, num_rows = 0;
695 if (!t->num_mapped_columns)
697 if (t->index_map.data)
698 return -E_OSL_ALREADY_MAPPED;
699 filename = index_filename(t->desc);
702 INFO_LOG("mapping table '%s' (index: %s)\n", t->desc->name, filename);
703 ret = mmap_full_file(filename, flags & MAP_TBL_FL_MAP_RDONLY?
704 O_RDONLY : O_RDWR, &t->index_map.data, &t->index_map.size, NULL);
708 if (flags & MAP_TBL_FL_VERIFY_INDEX) {
709 ret = compare_table_descriptions(t);
714 if (!(flags & MAP_TBL_FL_IGNORE_DIRTY)) {
715 if (table_is_dirty(t)) {
716 ERROR_LOG("%s is dirty\n", t->desc->name);
721 num_rows = table_num_rows(t);
725 FOR_EACH_MAPPED_COLUMN(i, t, cd) {
726 ret = map_column(t, i);
731 err: /* unmap what is already mapped */
732 for (i--; i >= 0; i--) {
733 struct osl_object map = t->columns[i].data_map;
734 osl_munmap(map.data, map.size);
737 osl_munmap(t->index_map.data, t->index_map.size);
738 t->index_map.data = NULL;
743 * Retrieve a mapped object by row and column number.
745 * \param t Pointer to an open osl table.
746 * \param col_num Number of the mapped column containing the object to retrieve.
747 * \param row_num Number of the row containing the object to retrieve.
748 * \param obj The result is returned here.
750 * It is considered an error if \a col_num does not refer to a column
751 * of storage type \p OSL_MAPPED_STORAGE.
755 * \sa osl_storage_type.
757 int get_mapped_object(const struct osl_table *t, unsigned col_num,
758 uint32_t row_num, struct osl_object *obj)
760 struct osl_column *col = &t->columns[col_num];
765 if (t->num_rows <= row_num)
766 return -E_OSL_BAD_ROW_NUM;
767 ret = get_cell_index(t, row_num, col_num, &cell_index);
770 offset = read_u32(cell_index);
771 obj->size = read_u32(cell_index + 4);
772 obj->data = col->data_map.data + offset;
777 * It's OK to call this with result = rb_node = NULL. If result is not NULL,
778 * and rb key was not found, result points to the parent node.
780 static int search_rbtree(const struct osl_object *obj,
781 const struct osl_table *t, unsigned col_num,
782 struct rb_node **result, struct rb_node ***rb_link)
784 struct osl_column *col = &t->columns[col_num];
785 struct rb_node **new = &col->rbtree.rb_node, *parent = NULL;
786 const struct osl_column_description *cd =
787 get_column_description(t->desc, col_num);
788 enum osl_storage_type st = cd->storage_type;
790 struct osl_row *this_row = get_row_pointer(*new,
793 struct osl_object this_obj;
795 if (st == OSL_MAPPED_STORAGE) {
796 ret = get_mapped_object(t, col_num, this_row->num,
801 this_obj = this_row->volatile_objects[col->volatile_num];
802 ret = cd->compare_function(obj, &this_obj);
805 *result = get_rb_node_pointer(this_row,
810 new = &((*new)->rb_left);
812 new = &((*new)->rb_right);
818 return -E_OSL_RB_KEY_NOT_FOUND;
821 static int insert_rbtree(struct osl_table *t, unsigned col_num,
822 const struct osl_row *row, const struct osl_object *obj)
824 struct rb_node *parent, **rb_link;
827 int ret = search_rbtree(obj, t, col_num, &parent, &rb_link);
830 return -E_OSL_RB_KEY_EXISTS;
831 rbtree_num = t->columns[col_num].rbtree_num;
832 n = get_rb_node_pointer(row, rbtree_num);
833 rb_link_node(n, parent, rb_link);
834 rb_insert_color(n, &t->columns[col_num].rbtree);
838 static void remove_rb_node(struct osl_table *t, unsigned col_num,
839 const struct osl_row *row)
841 struct osl_column *col = &t->columns[col_num];
842 const struct osl_column_description *cd =
843 get_column_description(t->desc, col_num);
844 enum osl_storage_flags sf = cd->storage_flags;
845 struct rb_node *victim, *splice_out_node, *tmp;
846 if (!(sf & OSL_RBTREE))
849 * Which node is removed/spliced out actually depends on how many
850 * children the victim node has: If it has no children, it gets
851 * deleted. If it has one child, it gets spliced out. If it has two
852 * children, its successor (which has at most a right child) gets
855 victim = get_rb_node_pointer(row, col->rbtree_num);
856 if (victim->rb_left && victim->rb_right)
857 splice_out_node = rb_next(victim);
859 splice_out_node = victim;
860 /* Go up to the root and decrement the size of each node in the path. */
861 for (tmp = splice_out_node; tmp; tmp = rb_parent(tmp))
863 rb_erase(victim, &col->rbtree);
866 static int add_row_to_rbtrees(struct osl_table *t, uint32_t row_num,
867 struct osl_object *volatile_objs, struct osl_row **row_ptr)
871 struct osl_row *row = allocate_row(t->num_rbtrees);
872 const struct osl_column_description *cd;
877 row->volatile_objects = volatile_objs;
878 FOR_EACH_RBTREE_COLUMN(i, t, cd) {
879 if (cd->storage_type == OSL_MAPPED_STORAGE) {
880 struct osl_object obj;
881 ret = get_mapped_object(t, i, row_num, &obj);
884 ret = insert_rbtree(t, i, row, &obj);
885 } else { /* volatile */
886 const struct osl_object *obj
887 = volatile_objs + t->columns[i].volatile_num;
888 ret = insert_rbtree(t, i, row, obj);
896 err: /* rollback changes, i.e. remove added entries from rbtrees */
898 remove_rb_node(t, i--, row);
903 static void free_volatile_objects(const struct osl_table *t,
904 enum osl_close_flags flags)
908 struct osl_column *rb_col;
909 const struct osl_column_description *cd;
911 if (!t->num_volatile_columns)
913 /* find the first rbtree column (any will do) */
914 FOR_EACH_RBTREE_COLUMN(i, t, cd)
916 rb_col = t->columns + i;
917 /* walk that rbtree and free all volatile objects */
918 for (n = rb_first(&rb_col->rbtree); n; n = rb_next(n)) {
919 struct osl_row *r = get_row_pointer(n, rb_col->rbtree_num);
920 if (flags & OSL_FREE_VOLATILE)
921 FOR_EACH_VOLATILE_COLUMN(j, t, cd) {
922 if (cd->storage_flags & OSL_DONT_FREE)
924 free(r->volatile_objects[
925 t->columns[j].volatile_num].data);
927 // for (j = 0; j < t->num_volatile_columns; j++)
928 // free(r->volatile_objects[j].data);
929 free(r->volatile_objects);
934 * Erase all rbtree nodes and free resources.
936 * \param t Pointer to an open osl table.
938 * This function is called by osl_close_table().
940 void clear_rbtrees(struct osl_table *t)
942 const struct osl_column_description *cd;
943 unsigned i, rbtrees_cleared = 0;
945 FOR_EACH_RBTREE_COLUMN(i, t, cd) {
946 struct osl_column *col = &t->columns[i];
949 for (n = rb_first(&col->rbtree); n;) {
951 rb_erase(n, &col->rbtree);
952 if (rbtrees_cleared == t->num_rbtrees) {
953 r = get_row_pointer(n, col->rbtree_num);
963 __export int osl_close_table(struct osl_table *t, enum osl_close_flags flags)
968 return -E_OSL_BAD_TABLE;
969 NOTICE_LOG("closing table %s\n", t->desc->name);
970 free_volatile_objects(t, flags);
972 ret = unmap_table(t, flags);
974 ERROR_LOG("unmap_table failed: %d\n", ret);
981 * Find out whether the given row number corresponds to an invalid row.
983 * \param t Pointer to the osl table.
984 * \param row_num The number of the row in question.
986 * By definition, a row is considered invalid if all its index entries
989 * \return Positive if \a row_num corresponds to an invalid row,
990 * zero if it corresponds to a valid row, negative on errors.
992 int row_is_invalid(struct osl_table *t, uint32_t row_num)
995 int i, ret = get_row_index(t, row_num, &row_index);
999 for (i = 0; i < t->row_index_size; i++) {
1000 if ((unsigned char)row_index[i] != 0xff)
1003 INFO_LOG("row %d is invalid\n", row_num);
1008 * Invalidate a row of an osl table.
1010 * \param t Pointer to an open osl table.
1011 * \param row_num Number of the row to mark as invalid.
1013 * This function marks each mapped object in the index entry of \a row as
1018 int mark_row_invalid(struct osl_table *t, uint32_t row_num)
1021 int ret = get_row_index(t, row_num, &row_index);
1025 INFO_LOG("marking row %d as invalid\n", row_num);
1026 memset(row_index, 0xff, t->row_index_size);
1031 * Initialize all rbtrees and compute number of invalid rows.
1033 * \param t The table containing the rbtrees to be initialized.
1037 int init_rbtrees(struct osl_table *t)
1040 const struct osl_column_description *cd;
1042 /* create rbtrees */
1043 FOR_EACH_RBTREE_COLUMN(i, t, cd)
1044 t->columns[i].rbtree = RB_ROOT;
1045 /* add valid rows to rbtrees */
1046 t->num_invalid_rows = 0;
1047 for (i = 0; i < t->num_rows; i++) {
1048 struct osl_object *volatile_objs;
1049 ret = row_is_invalid(t, i);
1053 t->num_invalid_rows++;
1056 if (t->num_volatile_columns > 0) {
1057 volatile_objs = calloc(t->num_volatile_columns,
1058 sizeof(struct osl_object));
1060 return -E_OSL_NOMEM;
1062 volatile_objs = NULL;
1063 ret = add_row_to_rbtrees(t, i, volatile_objs, NULL);
1070 __export int osl_open_table(const struct osl_table_description *table_desc,
1071 struct osl_table **result)
1074 struct osl_table *t;
1076 NOTICE_LOG("opening table %s\n", table_desc->name);
1077 ret = init_table_structure(table_desc, &t);
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 -E_OSL_NOMEM;
1109 ret = osl_mkdir(dirname, 0777);
1111 if (ret < 0 && ret != -E_OSL_DIR_EXISTS)
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 -E_OSL_NOMEM;
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 -E_OSL_NOMEM;
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 -E_OSL_NOMEM;
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 -E_OSL_NOMEM;
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 -E_OSL_NOMEM;
1182 if (unlink(filename) < 0)
1183 ret = errno == ENOENT? -E_OSL_NOENT : -E_OSL_UNLINK;
1187 if (!(t->desc->flags & OSL_LARGE_TABLE))
1189 dirname = disk_storage_dirname(t, col_num, ds_name);
1191 return -E_OSL_NOMEM;
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 -E_OSL_NOMEM;
1212 rb_links = malloc(t->num_rbtrees * sizeof(struct rn_node**));
1215 return -E_OSL_NOMEM;
1217 if (t->num_mapped_columns) {
1218 new_row_index = malloc(t->row_index_size);
1219 if (!new_row_index) {
1222 return -E_OSL_NOMEM;
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]);
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) {
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);
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)
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 if (!(cd->storage_flags & OSL_DONT_FREE))
1556 free(r->volatile_objects[col->volatile_num].data);
1557 r->volatile_objects[col->volatile_num] = *obj;
1558 } else if (cd->storage_type == OSL_DISK_STORAGE) {
1560 ret = disk_storage_name_of_row(t, r, &ds_name);
1563 ret = delete_disk_storage_file(t, col_num, ds_name);
1564 if (ret < 0 && ret != -E_OSL_NOENT) {
1568 ret = write_disk_storage_file(t, col_num, obj, ds_name);
1572 } else { /* mapped storage */
1573 struct osl_object old_obj;
1574 ret = get_mapped_object(t, col_num, r->num, &old_obj);
1578 * If the updated column is the disk storage name column, the
1579 * disk storage name changes, so we have to rename all disk
1580 * storage objects accordingly.
1582 if (col_num == t->disk_storage_name_column) {
1583 ret = rename_disk_storage_objects(t, &old_obj, obj);
1587 if (cd->storage_flags & OSL_FIXED_SIZE)
1588 memcpy(old_obj.data, obj->data, cd->data_size);
1589 else { /* TODO: if the size doesn't change, use old space */
1590 uint32_t new_data_map_size;
1592 ret = get_row_index(t, r->num, &row_index);
1595 unmap_column(t, col_num);
1596 ret = append_map_file(t, col_num, obj,
1597 &new_data_map_size);
1600 ret = map_column(t, col_num);
1603 update_cell_index(row_index, col, new_data_map_size,
1607 if (cd->storage_flags & OSL_RBTREE) {
1608 ret = insert_rbtree(t, col_num, r, obj);
1615 __export int osl_open_disk_object(const struct osl_table *t, const struct osl_row *r,
1616 unsigned col_num, struct osl_object *obj)
1618 const struct osl_column_description *cd;
1619 char *ds_name, *filename;
1623 return -E_OSL_BAD_TABLE;
1624 cd = get_column_description(t->desc, col_num);
1625 if (cd->storage_type != OSL_DISK_STORAGE)
1626 return -E_OSL_BAD_STORAGE_TYPE;
1628 ret = disk_storage_name_of_row(t, r, &ds_name);
1631 filename = disk_storage_path(t, col_num, ds_name);
1634 return -E_OSL_NOMEM;
1635 DEBUG_LOG("filename: %s\n", filename);
1636 ret = mmap_full_file(filename, O_RDONLY, &obj->data, &obj->size, NULL);
1641 __export int osl_close_disk_object(struct osl_object *obj)
1643 return osl_munmap(obj->data, obj->size);
1646 __export int osl_get_num_rows(const struct osl_table *t, unsigned *num_rows)
1649 return -E_OSL_BAD_TABLE;
1650 assert(t->num_rows >= t->num_invalid_rows);
1651 *num_rows = t->num_rows - t->num_invalid_rows;
1655 __export int osl_get_rank(const struct osl_table *t, struct osl_row *r,
1656 unsigned col_num, unsigned *rank)
1658 struct osl_object obj;
1659 struct osl_column *col;
1660 struct rb_node *node;
1661 int ret = check_rbtree_col(t, col_num, &col);
1665 ret = osl_get_object(t, r, col_num, &obj);
1668 ret = search_rbtree(&obj, t, col_num, &node, NULL);
1671 ret = rb_rank(node, rank);
1673 return -E_OSL_BAD_ROW;
1677 __export int osl_get_nth_row(const struct osl_table *t, unsigned col_num,
1678 unsigned n, struct osl_row **result)
1680 struct osl_column *col;
1681 struct rb_node *node;
1687 return -E_OSL_RB_KEY_NOT_FOUND;
1688 ret = osl_get_num_rows(t, &num_rows);
1692 return -E_OSL_RB_KEY_NOT_FOUND;
1693 ret = check_rbtree_col(t, col_num, &col);
1696 node = rb_nth(col->rbtree.rb_node, n);
1698 return -E_OSL_RB_KEY_NOT_FOUND;
1699 *result = get_row_pointer(node, col->rbtree_num);
1703 __export int osl_rbtree_first_row(const struct osl_table *t, unsigned col_num,
1704 struct osl_row **result)
1706 return osl_get_nth_row(t, col_num, 1, result);
1709 __export int osl_rbtree_last_row(const struct osl_table *t, unsigned col_num,
1710 struct osl_row **result)
1713 int ret = osl_get_num_rows(t, &num_rows);
1717 return osl_get_nth_row(t, col_num, num_rows, result);