2 * Copyright (C) 2007-2009 Andre Noll <maan@systemlinux.org>
4 * Licensed under the GPL v2. For licencing details see COPYING.
7 /** \file osl.c Object storage layer functions. */
8 #include <dirent.h> /* readdir() */
16 /* Taken from Drepper: How to write shared libraries, Appendix B. */
18 #define MSGSTRFIELD(line) MSGSTRFIELD1(line)
19 #define MSGSTRFIELD1(line) str##line
20 static const union msgstr_t {
22 #define OSL_ERROR(n, s) char MSGSTRFIELD(__LINE__)[sizeof(s)];
28 #define OSL_ERROR(n, s) s,
32 static const unsigned int errmsgidx[] = {
33 #define OSL_ERROR(n, s) [n] = offsetof(union msgstr_t, MSGSTRFIELD(__LINE__)),
38 __export const char *osl_strerror(int num)
40 return msgstr.str + errmsgidx[num];
45 static void __attribute ((constructor)) init_loglevel(void)
47 char *p = getenv("OSL_LOGLEVEL");
49 /* don't log anything if unset */
50 loglevel = p? atoi(p) : EMERG + 1;
57 * \param fmt Usual format string.
59 * All XXX_LOG() macros use this function.
61 __printf_2_3 void __log(int ll, const char* fmt,...)
74 strftime(str, sizeof(str), "%b %d %H:%M:%S", tm);
75 fprintf(outfd, "%s ", str);
77 vfprintf(outfd, fmt, argp);
82 * A wrapper for lseek(2).
84 * \param fd The file descriptor whose offset is to be to repositioned.
85 * \param offset A value-result parameter.
86 * \param whence Usual repositioning directive.
88 * Reposition the offset of the file descriptor \a fd to the argument \a offset
89 * according to the directive \a whence. Upon successful return, \a offset
90 * contains the resulting offset location as measured in bytes from the
91 * beginning of the file.
93 * \return Positive on success. Otherwise, the function returns \p -E_OSL_LSEEK.
97 static int __lseek(int fd, off_t *offset, int whence)
99 *offset = lseek(fd, *offset, whence);
100 int ret = -E_OSL_LSEEK;
106 static int append_file(const char *filename, char *data, size_t data_size,
111 // DEBUG_LOG("appending %zu + %zu bytes\n", header_size, data_size);
112 ret = osl_open(filename, O_WRONLY | O_CREAT | O_APPEND, 0644);
116 ret = write_all(fd, data, &data_size);
121 ret = __lseek(fd, &offset, SEEK_END);
124 // DEBUG_LOG("new file size: " FMT_OFF_T "\n", offset);
133 static int verify_name(const char *name)
136 return -E_OSL_BAD_NAME;
138 return -E_OSL_BAD_NAME;
139 if (strchr(name, '/'))
140 return -E_OSL_BAD_NAME;
141 if (!strcmp(name, ".."))
142 return -E_OSL_BAD_NAME;
143 if (!strcmp(name, "."))
144 return -E_OSL_BAD_NAME;
148 static char *disk_storage_dirname(const struct osl_table *t, unsigned col_num,
151 char *dirname, *column_name = column_filename(t, col_num);
155 if (!(t->desc->flags & OSL_LARGE_TABLE))
157 dirname = make_message("%s/%.2s", column_name, ds_name);
162 static char *disk_storage_name_of_object(const struct osl_table *t,
163 const struct osl_object *obj)
165 HASH_TYPE hash[HASH_SIZE];
166 hash_object(obj, hash);
167 return disk_storage_name_of_hash(t, hash);
170 static int disk_storage_name_of_row(const struct osl_table *t,
171 const struct osl_row *row, char **name)
173 struct osl_object obj;
174 int ret = osl_get_object(t, row, t->disk_storage_name_column, &obj);
178 *name = disk_storage_name_of_object(t, &obj);
184 static void column_name_hash(const char *col_name, HASH_TYPE *hash)
186 hash_function(col_name, strlen(col_name), hash);
189 static int init_column_descriptions(struct osl_table *t)
192 const struct osl_column_description *cd;
194 ret = verify_name(t->desc->name);
197 ret = -E_OSL_BAD_DB_DIR;
198 if (!t->desc->dir && (t->num_disk_storage_columns || t->num_mapped_columns))
200 /* the size of the index header without column descriptions */
201 t->index_header_size = IDX_COLUMN_DESCRIPTIONS;
202 FOR_EACH_COLUMN(i, t->desc, cd) {
203 struct osl_column *col = t->columns + i;
204 if (cd->storage_flags & OSL_RBTREE) {
205 if (!cd->compare_function)
206 return -E_OSL_NO_COMPARE_FUNC;
208 if (cd->storage_type == OSL_NO_STORAGE)
210 ret = -E_OSL_NO_COLUMN_NAME;
211 if (!cd->name || !cd->name[0])
213 ret = verify_name(cd->name);
216 t->index_header_size += index_column_description_size(cd->name);
217 column_name_hash(cd->name, col->name_hash);
218 ret = -E_OSL_DUPLICATE_COL_NAME;
219 for (j = i + 1; j < t->desc->num_columns; j++) {
220 const char *name2 = get_column_description(t->desc,
222 if (cd->name && name2 && !strcmp(cd->name, name2))
232 * Initialize a struct table from given table description.
234 * \param desc The description of the osl table.
235 * \param table_ptr Result is returned here.
237 * This function performs several sanity checks on \p desc and returns if any
238 * of these tests fail. On success, a struct \p osl_table is allocated and
239 * initialized with data derived from \p desc.
243 * \sa struct osl_table.
245 int init_table_structure(const struct osl_table_description *desc,
246 struct osl_table **table_ptr)
248 const struct osl_column_description *cd;
249 struct osl_table *t = calloc(1, sizeof(*t));
250 int i, ret = -E_OSL_NOMEM, have_disk_storage_name_column = 0;
254 ret = -E_OSL_BAD_TABLE_DESC;
257 DEBUG_LOG("creating table structure for '%s' from table "
258 "description\n", desc->name);
259 ret = -E_OSL_NO_COLUMN_DESC;
260 if (!desc->column_descriptions)
262 ret = -E_OSL_NO_COLUMNS;
263 if (!desc->num_columns)
266 t->columns = calloc(desc->num_columns, sizeof(struct osl_column));
270 FOR_EACH_COLUMN(i, t->desc, cd) {
271 enum osl_storage_type st = cd->storage_type;
272 enum osl_storage_flags sf = cd->storage_flags;
273 struct osl_column *col = &t->columns[i];
275 ret = -E_OSL_BAD_STORAGE_TYPE;
276 if (st != OSL_MAPPED_STORAGE && st != OSL_DISK_STORAGE
277 && st != OSL_NO_STORAGE)
279 ret = -E_OSL_BAD_STORAGE_FLAGS;
280 if (st == OSL_DISK_STORAGE && sf & OSL_RBTREE)
282 ret = -E_OSL_BAD_STORAGE_SIZE;
283 if (sf & OSL_FIXED_SIZE && !cd->data_size)
286 case OSL_DISK_STORAGE:
287 t->num_disk_storage_columns++;
289 case OSL_MAPPED_STORAGE:
290 t->num_mapped_columns++;
291 col->index_offset = t->row_index_size;
292 t->row_index_size += 8;
295 col->volatile_num = t->num_volatile_columns;
296 t->num_volatile_columns++;
299 if (sf & OSL_RBTREE) {
300 col->rbtree_num = t->num_rbtrees;
302 if ((sf & OSL_UNIQUE) && (st == OSL_MAPPED_STORAGE)) {
303 if (!have_disk_storage_name_column)
304 t->disk_storage_name_column = i;
305 have_disk_storage_name_column = 1;
309 ret = -E_OSL_NO_UNIQUE_RBTREE_COLUMN;
310 if (t->num_disk_storage_columns && !have_disk_storage_name_column)
312 ret = -E_OSL_NO_RBTREE_COL;
316 DEBUG_LOG("OK. Index entry size: %u\n", t->row_index_size);
317 ret = init_column_descriptions(t);
329 * Read the table description from index header.
331 * \param map The memory mapping of the index file.
332 * \param desc The values found in the index header are returned here.
334 * Read the index header, check for the osl magic string and the table version
335 * number. Read all information stored in the index header into \a desc.
339 * \sa struct osl_table_description, osl_create_table.
341 int read_table_desc(struct osl_object *map, struct osl_table_description *desc)
343 char *buf = map->data;
344 uint8_t version, compat_version, create_version;
345 uint16_t header_size;
348 struct osl_column_description *cd;
350 if (map->size < MIN_INDEX_HEADER_SIZE(1))
351 return -E_OSL_SHORT_TABLE;
352 if (strncmp(buf + IDX_OSL_MAGIC, OSL_MAGIC, strlen(OSL_MAGIC)))
353 return -E_OSL_NO_MAGIC;
354 version = read_u8(buf + IDX_VERSION);
356 * The on-disk version consists of two version numbers: the
357 * create_version (low 4 bits) is the CURRENT_TABLE_VERSION version
358 * number of the library that created the table, and compat_version
359 * (high 4 bits) tells us the lowest version of the library that can
360 * still read this table.
362 create_version = version & 0xf;
363 compat_version = version >> 4;
364 INFO_LOG("create_version: %u, compat_version: %u\n", create_version,
366 if (create_version < MIN_TABLE_VERSION /* table too old */
367 || compat_version > CURRENT_TABLE_VERSION) /* libosl too old */
368 return -E_OSL_VERSION_MISMATCH;
369 desc->flags = read_u8(buf + IDX_TABLE_FLAGS);
370 desc->num_columns = read_u16(buf + IDX_NUM_COLUMNS);
371 INFO_LOG("%u columns\n", desc->num_columns);
372 if (!desc->num_columns)
373 return -E_OSL_NO_COLUMNS;
374 header_size = read_u16(buf + IDX_HEADER_SIZE);
375 if (map->size < header_size)
376 return -E_OSL_BAD_SIZE;
377 desc->column_descriptions = calloc(desc->num_columns,
378 sizeof(struct osl_column_description));
379 if (!desc->column_descriptions)
381 offset = IDX_COLUMN_DESCRIPTIONS;
382 FOR_EACH_COLUMN(i, desc, cd) {
385 ret = -E_OSL_SHORT_TABLE;
386 if (map->size < offset + MIN_IDX_COLUMN_DESCRIPTION_SIZE) {
387 ERROR_LOG("map size = %zu < %u = offset + min desc size\n",
388 map->size, offset + MIN_IDX_COLUMN_DESCRIPTION_SIZE);
391 cd->storage_type = read_u16(buf + offset + IDX_CD_STORAGE_TYPE);
392 cd->storage_flags = read_u16(buf + offset +
393 IDX_CD_STORAGE_FLAGS);
394 cd->data_size = read_u32(buf + offset + IDX_CD_DATA_SIZE);
395 null_byte = memchr(buf + offset + IDX_CD_NAME, '\0',
396 map->size - offset - IDX_CD_NAME);
397 ret = -E_OSL_INDEX_CORRUPTION;
401 cd->name = strdup(buf + offset + IDX_CD_NAME);
404 offset += index_column_description_size(cd->name);
406 if (offset != header_size) {
407 ret = -E_OSL_INDEX_CORRUPTION;
408 ERROR_LOG("real header size = %u != %u = stored header size\n",
409 offset, header_size);
414 FOR_EACH_COLUMN(i, desc, cd)
420 * check whether the table description given by \p t->desc matches the on-disk
421 * table structure stored in the index of \a t.
423 static int compare_table_descriptions(struct osl_table *t)
426 struct osl_table_description desc;
427 const struct osl_column_description *cd1, *cd2;
429 /* read the on-disk structure into desc */
430 ret = read_table_desc(&t->index_map, &desc);
433 ret = -E_OSL_BAD_TABLE_FLAGS;
434 if (desc.flags != t->desc->flags)
436 ret = -E_OSL_BAD_COLUMN_NUM;
437 if (desc.num_columns > t->desc->num_columns)
439 if (desc.num_columns < t->desc->num_columns) {
440 struct osl_column_description *cd;
441 unsigned diff = t->desc->num_columns - desc.num_columns;
442 INFO_LOG("extending table by %u volatile columns\n", diff);
444 desc.column_descriptions = realloc(desc.column_descriptions,
445 t->desc->num_columns * sizeof(struct osl_column_description));
446 if (!desc.column_descriptions)
448 for (i = desc.num_columns; i < t->desc->num_columns; i++) {
449 cd = get_column_description(&desc, i);
450 cd->storage_type = OSL_NO_STORAGE;
453 desc.num_columns += diff;
455 FOR_EACH_COLUMN(i, t->desc, cd1) {
456 cd2 = get_column_description(&desc, i);
457 ret = -E_OSL_BAD_STORAGE_TYPE;
458 if (cd1->storage_type != cd2->storage_type)
460 if (cd1->storage_type == OSL_NO_STORAGE)
462 ret = -E_OSL_BAD_STORAGE_FLAGS;
463 if (cd1->storage_flags != cd2->storage_flags) {
464 ERROR_LOG("sf1 = %u != %u = sf2\n",
465 cd1->storage_flags, cd2->storage_flags);
468 ret = -E_OSL_BAD_DATA_SIZE;
469 if (cd1->storage_flags & OSL_FIXED_SIZE)
470 if (cd1->data_size != cd2->data_size)
472 ret = -E_OSL_BAD_COLUMN_NAME;
473 if (strcmp(cd1->name, cd2->name))
476 INFO_LOG("table description of '%s' matches on-disk data, good\n",
480 FOR_EACH_COLUMN(i, &desc, cd1)
482 free(desc.column_descriptions);
486 static int create_table_index(struct osl_table *t)
488 char *buf, *filename;
490 size_t size = t->index_header_size;
491 const struct osl_column_description *cd;
494 INFO_LOG("creating %zu byte index for table %s\n", size,
496 buf = calloc(1, size);
499 sprintf(buf + IDX_OSL_MAGIC, "%s", OSL_MAGIC);
500 write_u8(buf + IDX_TABLE_FLAGS, t->desc->flags);
501 write_u8(buf + IDX_DIRTY_FLAG, 0);
502 write_u8(buf + IDX_VERSION, CURRENT_TABLE_VERSION
503 + (COMPAT_TABLE_VERSION << 4));
504 write_u16(buf + IDX_NUM_COLUMNS, t->num_mapped_columns + t->num_disk_storage_columns);
505 write_u16(buf + IDX_HEADER_SIZE, t->index_header_size);
506 offset = IDX_COLUMN_DESCRIPTIONS;
507 FOR_EACH_COLUMN(i, t->desc, cd) {
508 /* no need to store info about volatile storage */
509 if (cd->storage_type == OSL_NO_STORAGE)
511 write_u16(buf + offset + IDX_CD_STORAGE_TYPE,
513 write_u16(buf + offset + IDX_CD_STORAGE_FLAGS,
515 if (cd->storage_flags & OSL_FIXED_SIZE)
516 write_u32(buf + offset + IDX_CD_DATA_SIZE,
518 strcpy(buf + offset + IDX_CD_NAME, cd->name);
519 offset += index_column_description_size(cd->name);
521 assert(offset = size);
522 filename = index_filename(t->desc);
524 ret = write_file(filename, buf, size);
532 __export int osl_create_table(const struct osl_table_description *desc)
534 const struct osl_column_description *cd;
535 char *table_dir = NULL, *filename;
537 int i, ret = init_table_structure(desc, &t);
541 INFO_LOG("creating %s\n", desc->name);
542 FOR_EACH_COLUMN(i, t->desc, cd) {
543 if (cd->storage_type == OSL_NO_STORAGE)
546 ret = osl_mkdir(desc->dir, 0777);
547 if (ret < 0 && ret != -E_OSL_DIR_EXISTS)
549 table_dir = make_message("%s/%s", desc->dir,
554 ret = osl_mkdir(table_dir, 0777);
559 filename = column_filename(t, i);
562 INFO_LOG("filename: %s\n", filename);
563 if (cd->storage_type == OSL_MAPPED_STORAGE) {
564 ret = osl_open(filename, O_RDWR | O_CREAT | O_EXCL,
573 ret = osl_mkdir(filename, 0777);
578 if (t->num_mapped_columns) {
579 ret = create_table_index(t);
591 static int table_is_dirty(struct osl_table *t)
593 char *buf = (char *)t->index_map.data + IDX_DIRTY_FLAG;
594 uint8_t dirty = read_u8(buf) & 0x1;
598 static void mark_table_dirty(struct osl_table *t)
600 char *buf = (char *)t->index_map.data + IDX_DIRTY_FLAG;
601 write_u8(buf, read_u8(buf) | 1);
604 static void mark_table_clean(struct osl_table *t)
606 char *buf = (char *)t->index_map.data + IDX_DIRTY_FLAG;
607 write_u8(buf, read_u8(buf) & 0xfe);
610 static void unmap_column(struct osl_table *t, unsigned col_num)
612 struct osl_object map = t->columns[col_num].data_map;
616 ret = osl_munmap(map.data, map.size);
622 * Unmap all mapped files of an osl table.
624 * \param t Pointer to a mapped table.
625 * \param flags Options for unmapping.
627 * \return Positive on success, negative on errors.
629 * \sa map_table(), enum osl_close_flags, osl_munmap().
631 int unmap_table(struct osl_table *t, enum osl_close_flags flags)
634 const struct osl_column_description *cd;
637 if (!t->num_mapped_columns) /* can this ever happen? */
639 INFO_LOG("unmapping table '%s'\n", t->desc->name);
640 if (!t->index_map.data)
641 return -E_OSL_NOT_MAPPED;
642 if (flags & OSL_MARK_CLEAN)
644 ret = osl_munmap(t->index_map.data, t->index_map.size);
647 t->index_map.data = NULL;
650 FOR_EACH_MAPPED_COLUMN(i, t, cd)
655 static int map_column(struct osl_table *t, unsigned col_num)
658 char *filename = column_filename(t, col_num);
663 ret = osl_stat(filename, &statbuf);
668 if (!(S_IFREG & statbuf.st_mode)) {
672 ret = mmap_full_file(filename, O_RDWR,
673 &t->columns[col_num].data_map.data,
674 &t->columns[col_num].data_map.size,
681 * Map the index file and all columns of type \p OSL_MAPPED_STORAGE into memory.
683 * \param t Pointer to an initialized table structure.
684 * \param flags Mapping options.
686 * \return Negative return value on errors; on success the number of rows
687 * (including invalid rows) is returned.
689 * \sa unmap_table(), enum map_table_flags, osl_open_table(), mmap(2).
691 int map_table(struct osl_table *t, enum map_table_flags flags)
694 const struct osl_column_description *cd;
695 int i = 0, ret, num_rows = 0;
697 if (!t->num_mapped_columns)
699 if (t->index_map.data)
700 return -E_OSL_ALREADY_MAPPED;
701 filename = index_filename(t->desc);
704 INFO_LOG("mapping table '%s' (index: %s)\n", t->desc->name, filename);
705 ret = mmap_full_file(filename, flags & MAP_TBL_FL_MAP_RDONLY?
706 O_RDONLY : O_RDWR, &t->index_map.data, &t->index_map.size, NULL);
710 if (flags & MAP_TBL_FL_VERIFY_INDEX) {
711 ret = compare_table_descriptions(t);
716 if (!(flags & MAP_TBL_FL_IGNORE_DIRTY)) {
717 if (table_is_dirty(t)) {
718 ERROR_LOG("%s is dirty\n", t->desc->name);
723 num_rows = table_num_rows(t);
727 FOR_EACH_MAPPED_COLUMN(i, t, cd) {
728 ret = map_column(t, i);
733 err: /* unmap what is already mapped */
734 for (i--; i >= 0; i--) {
735 struct osl_object map = t->columns[i].data_map;
736 osl_munmap(map.data, map.size);
739 osl_munmap(t->index_map.data, t->index_map.size);
740 t->index_map.data = NULL;
745 * Retrieve a mapped object by row and column number.
747 * \param t Pointer to an open osl table.
748 * \param col_num Number of the mapped column containing the object to retrieve.
749 * \param row_num Number of the row containing the object to retrieve.
750 * \param obj The result is returned here.
752 * It is considered an error if \a col_num does not refer to a column
753 * of storage type \p OSL_MAPPED_STORAGE.
757 * \sa osl_storage_type.
759 int get_mapped_object(const struct osl_table *t, unsigned col_num,
760 uint32_t row_num, struct osl_object *obj)
762 struct osl_column *col = &t->columns[col_num];
767 if (t->num_rows <= row_num)
768 return -E_OSL_BAD_ROW_NUM;
769 ret = get_cell_index(t, row_num, col_num, &cell_index);
772 offset = read_u32(cell_index);
773 obj->size = read_u32(cell_index + 4);
774 obj->data = col->data_map.data + offset;
779 * It's OK to call this with result = rb_node = NULL. If result is not NULL,
780 * and rb key was not found, result points to the parent node.
782 static int search_rbtree(const struct osl_object *obj,
783 const struct osl_table *t, unsigned col_num,
784 struct rb_node **result, struct rb_node ***rb_link)
786 struct osl_column *col = &t->columns[col_num];
787 struct rb_node **new = &col->rbtree.rb_node, *parent = NULL;
788 const struct osl_column_description *cd =
789 get_column_description(t->desc, col_num);
790 enum osl_storage_type st = cd->storage_type;
792 struct osl_row *this_row = get_row_pointer(*new,
795 struct osl_object this_obj;
797 if (st == OSL_MAPPED_STORAGE) {
798 ret = get_mapped_object(t, col_num, this_row->num,
803 this_obj = this_row->volatile_objects[col->volatile_num];
804 ret = cd->compare_function(obj, &this_obj);
807 *result = get_rb_node_pointer(this_row,
812 new = &((*new)->rb_left);
814 new = &((*new)->rb_right);
820 return -E_OSL_RB_KEY_NOT_FOUND;
823 static int insert_rbtree(struct osl_table *t, unsigned col_num,
824 const struct osl_row *row, const struct osl_object *obj)
826 struct rb_node *parent, **rb_link;
829 int ret = search_rbtree(obj, t, col_num, &parent, &rb_link);
832 return -E_OSL_RB_KEY_EXISTS;
833 rbtree_num = t->columns[col_num].rbtree_num;
834 n = get_rb_node_pointer(row, rbtree_num);
835 rb_link_node(n, parent, rb_link);
836 rb_insert_color(n, &t->columns[col_num].rbtree);
840 static void remove_rb_node(struct osl_table *t, unsigned col_num,
841 const struct osl_row *row)
843 struct osl_column *col = &t->columns[col_num];
844 const struct osl_column_description *cd =
845 get_column_description(t->desc, col_num);
846 enum osl_storage_flags sf = cd->storage_flags;
847 struct rb_node *victim, *splice_out_node, *tmp;
848 if (!(sf & OSL_RBTREE))
851 * Which node is removed/spliced out actually depends on how many
852 * children the victim node has: If it has no children, it gets
853 * deleted. If it has one child, it gets spliced out. If it has two
854 * children, its successor (which has at most a right child) gets
857 victim = get_rb_node_pointer(row, col->rbtree_num);
858 if (victim->rb_left && victim->rb_right)
859 splice_out_node = rb_next(victim);
861 splice_out_node = victim;
862 /* Go up to the root and decrement the size of each node in the path. */
863 for (tmp = splice_out_node; tmp; tmp = rb_parent(tmp))
865 rb_erase(victim, &col->rbtree);
868 static int add_row_to_rbtrees(struct osl_table *t, uint32_t row_num,
869 struct osl_object *volatile_objs, struct osl_row **row_ptr)
873 struct osl_row *row = allocate_row(t->num_rbtrees);
874 const struct osl_column_description *cd;
879 row->volatile_objects = volatile_objs;
880 FOR_EACH_RBTREE_COLUMN(i, t, cd) {
881 if (cd->storage_type == OSL_MAPPED_STORAGE) {
882 struct osl_object obj;
883 ret = get_mapped_object(t, i, row_num, &obj);
886 ret = insert_rbtree(t, i, row, &obj);
887 } else { /* volatile */
888 const struct osl_object *obj
889 = volatile_objs + t->columns[i].volatile_num;
890 ret = insert_rbtree(t, i, row, obj);
898 err: /* rollback changes, i.e. remove added entries from rbtrees */
900 remove_rb_node(t, i--, row);
905 static void free_volatile_objects(const struct osl_table *t,
906 enum osl_close_flags flags)
910 struct osl_column *rb_col;
911 const struct osl_column_description *cd;
913 if (!t->num_volatile_columns)
915 /* find the first rbtree column (any will do) */
916 FOR_EACH_RBTREE_COLUMN(i, t, cd)
918 rb_col = t->columns + i;
919 /* walk that rbtree and free all volatile objects */
920 for (n = rb_first(&rb_col->rbtree); n; n = rb_next(n)) {
921 struct osl_row *r = get_row_pointer(n, rb_col->rbtree_num);
922 if (flags & OSL_FREE_VOLATILE)
923 FOR_EACH_VOLATILE_COLUMN(j, t, cd) {
924 if (cd->storage_flags & OSL_DONT_FREE)
926 free(r->volatile_objects[
927 t->columns[j].volatile_num].data);
929 // for (j = 0; j < t->num_volatile_columns; j++)
930 // free(r->volatile_objects[j].data);
931 free(r->volatile_objects);
936 * Erase all rbtree nodes and free resources.
938 * \param t Pointer to an open osl table.
940 * This function is called by osl_close_table().
942 void clear_rbtrees(struct osl_table *t)
944 const struct osl_column_description *cd;
945 unsigned i, rbtrees_cleared = 0;
947 FOR_EACH_RBTREE_COLUMN(i, t, cd) {
948 struct osl_column *col = &t->columns[i];
951 for (n = rb_first(&col->rbtree); n;) {
953 rb_erase(n, &col->rbtree);
954 if (rbtrees_cleared == t->num_rbtrees) {
955 r = get_row_pointer(n, col->rbtree_num);
965 __export int osl_close_table(struct osl_table *t, enum osl_close_flags flags)
970 return -E_OSL_BAD_TABLE;
971 NOTICE_LOG("closing table %s\n", t->desc->name);
972 free_volatile_objects(t, flags);
974 ret = unmap_table(t, flags);
976 ERROR_LOG("unmap_table failed: %d\n", ret);
983 * Find out whether the given row number corresponds to an invalid row.
985 * \param t Pointer to the osl table.
986 * \param row_num The number of the row in question.
988 * By definition, a row is considered invalid if all its index entries
991 * \return Positive if \a row_num corresponds to an invalid row,
992 * zero if it corresponds to a valid row, negative on errors.
994 int row_is_invalid(struct osl_table *t, uint32_t row_num)
997 int i, ret = get_row_index(t, row_num, &row_index);
1001 for (i = 0; i < t->row_index_size; i++) {
1002 if ((unsigned char)row_index[i] != 0xff)
1005 INFO_LOG("row %d is invalid\n", row_num);
1010 * Invalidate a row of an osl table.
1012 * \param t Pointer to an open osl table.
1013 * \param row_num Number of the row to mark as invalid.
1015 * This function marks each mapped object in the index entry of \a row as
1020 int mark_row_invalid(struct osl_table *t, uint32_t row_num)
1023 int ret = get_row_index(t, row_num, &row_index);
1027 INFO_LOG("marking row %d as invalid\n", row_num);
1028 memset(row_index, 0xff, t->row_index_size);
1033 * Initialize all rbtrees and compute number of invalid rows.
1035 * \param t The table containing the rbtrees to be initialized.
1039 int init_rbtrees(struct osl_table *t)
1042 const struct osl_column_description *cd;
1044 /* create rbtrees */
1045 FOR_EACH_RBTREE_COLUMN(i, t, cd)
1046 t->columns[i].rbtree = RB_ROOT;
1047 /* add valid rows to rbtrees */
1048 t->num_invalid_rows = 0;
1049 for (i = 0; i < t->num_rows; i++) {
1050 ret = row_is_invalid(t, i);
1054 t->num_invalid_rows++;
1057 ret = add_row_to_rbtrees(t, i, NULL, NULL);
1064 __export int osl_open_table(const struct osl_table_description *table_desc,
1065 struct osl_table **result)
1068 struct osl_table *t;
1069 const struct osl_column_description *cd;
1071 NOTICE_LOG("opening table %s\n", table_desc->name);
1072 ret = init_table_structure(table_desc, &t);
1075 FOR_EACH_DISK_STORAGE_COLUMN(i, t, cd) {
1076 struct stat statbuf;
1077 char *dirname = column_filename(t, i);
1082 /* check if directory exists */
1083 ret = osl_stat(dirname, &statbuf);
1087 ret = -E_OSL_NOTDIR;
1088 if (!S_ISDIR(statbuf.st_mode))
1091 ret = map_table(t, MAP_TBL_FL_VERIFY_INDEX);
1095 DEBUG_LOG("num rows: %d\n", t->num_rows);
1096 ret = init_rbtrees(t);
1098 osl_close_table(t, OSL_MARK_CLEAN); /* ignore further errors */
1109 static int create_disk_storage_object_dir(const struct osl_table *t,
1110 unsigned col_num, const char *ds_name)
1115 if (!(t->desc->flags & OSL_LARGE_TABLE))
1117 dirname = disk_storage_dirname(t, col_num, ds_name);
1119 return -E_OSL_NOMEM;
1120 ret = osl_mkdir(dirname, 0777);
1122 if (ret < 0 && ret != -E_OSL_DIR_EXISTS)
1127 static int write_disk_storage_file(const struct osl_table *t, unsigned col_num,
1128 const struct osl_object *obj, const char *ds_name)
1133 ret = create_disk_storage_object_dir(t, col_num, ds_name);
1136 filename = disk_storage_path(t, col_num, ds_name);
1138 return -E_OSL_NOMEM;
1139 ret = write_file(filename, obj->data, obj->size);
1144 static int append_map_file(const struct osl_table *t, unsigned col_num,
1145 const struct osl_object *obj, uint32_t *new_size)
1147 char *filename = column_filename(t, col_num);
1151 return -E_OSL_NOMEM;
1152 ret = append_file(filename, obj->data, obj->size, new_size);
1157 static int append_row_index(const struct osl_table *t, char *row_index)
1162 if (!t->num_mapped_columns)
1164 filename = index_filename(t->desc);
1166 return -E_OSL_NOMEM;
1167 ret = append_file(filename, row_index, t->row_index_size, NULL);
1172 static int truncate_mapped_file(const struct osl_table *t, unsigned col_num,
1176 char *filename = column_filename(t, col_num);
1179 return -E_OSL_NOMEM;
1180 ret = truncate_file(filename, size);
1185 static int delete_disk_storage_file(const struct osl_table *t, unsigned col_num,
1186 const char *ds_name)
1188 char *dirname, *filename = disk_storage_path(t, col_num, ds_name);
1192 return -E_OSL_NOMEM;
1193 if (unlink(filename) < 0)
1194 ret = errno == ENOENT? -E_OSL_NOENT : -E_OSL_UNLINK;
1198 if (!(t->desc->flags & OSL_LARGE_TABLE))
1200 dirname = disk_storage_dirname(t, col_num, ds_name);
1202 return -E_OSL_NOMEM;
1208 __export int osl_add_and_get_row(struct osl_table *t, struct osl_object *objects,
1209 struct osl_row **row)
1212 char *ds_name = NULL;
1213 struct rb_node **rb_parents = NULL, ***rb_links = NULL;
1214 char *new_row_index = NULL;
1215 struct osl_object *volatile_objs = NULL;
1216 const struct osl_column_description *cd;
1219 return -E_OSL_BAD_TABLE;
1220 rb_parents = malloc(t->num_rbtrees * sizeof(struct rn_node*));
1222 return -E_OSL_NOMEM;
1223 rb_links = malloc(t->num_rbtrees * sizeof(struct rn_node**));
1226 return -E_OSL_NOMEM;
1228 if (t->num_mapped_columns) {
1229 new_row_index = malloc(t->row_index_size);
1230 if (!new_row_index) {
1233 return -E_OSL_NOMEM;
1236 /* pass 1: sanity checks */
1237 // DEBUG_LOG("sanity tests: %p:%p\n", objects[0].data,
1238 // objects[1].data);
1239 FOR_EACH_COLUMN(i, t->desc, cd) {
1240 enum osl_storage_type st = cd->storage_type;
1241 enum osl_storage_flags sf = cd->storage_flags;
1243 // ret = -E_OSL_NULL_OBJECT;
1246 if (st == OSL_DISK_STORAGE)
1248 if (sf & OSL_RBTREE) {
1249 unsigned rbtree_num = t->columns[i].rbtree_num;
1250 ret = -E_OSL_RB_KEY_EXISTS;
1251 // DEBUG_LOG("checking whether %p exists\n",
1252 // objects[i].data);
1253 if (search_rbtree(objects + i, t, i,
1254 &rb_parents[rbtree_num],
1255 &rb_links[rbtree_num]) > 0)
1258 if (sf & OSL_FIXED_SIZE) {
1259 // DEBUG_LOG("fixed size. need: %zu, have: %d\n",
1260 // objects[i].size, cd->data_size);
1261 ret = -E_OSL_BAD_DATA_SIZE;
1262 if (objects[i].size != cd->data_size)
1266 if (t->num_disk_storage_columns) {
1267 ds_name = disk_storage_name_of_object(t,
1268 &objects[t->disk_storage_name_column]);
1273 ret = unmap_table(t, OSL_MARK_CLEAN);
1276 // DEBUG_LOG("sanity tests passed%s\n", "");
1277 /* pass 2: create data files, append map data */
1278 FOR_EACH_COLUMN(i, t->desc, cd) {
1279 enum osl_storage_type st = cd->storage_type;
1280 if (st == OSL_NO_STORAGE)
1282 if (st == OSL_MAPPED_STORAGE) {
1284 struct osl_column *col = &t->columns[i];
1285 // DEBUG_LOG("appending object of size %zu\n",
1286 // objects[i].size);
1287 ret = append_map_file(t, i, objects + i, &new_size);
1290 update_cell_index(new_row_index, col, new_size,
1295 ret = write_disk_storage_file(t, i, objects + i, ds_name);
1299 ret = append_row_index(t, new_row_index);
1302 ret = map_table(t, MAP_TBL_FL_VERIFY_INDEX);
1303 if (ret < 0) { /* truncate index and rollback changes */
1304 char *filename = index_filename(t->desc);
1306 truncate_file(filename, t->row_index_size);
1310 /* pass 3: add entry to rbtrees */
1311 if (t->num_volatile_columns) {
1313 volatile_objs = calloc(t->num_volatile_columns,
1314 sizeof(struct osl_object));
1317 FOR_EACH_VOLATILE_COLUMN(i, t, cd)
1318 volatile_objs[t->columns[i].volatile_num] = objects[i];
1321 // DEBUG_LOG("adding new entry as row #%d\n", t->num_rows - 1);
1322 ret = add_row_to_rbtrees(t, t->num_rows - 1, volatile_objs, row);
1325 // DEBUG_LOG("added new entry as row #%d\n", t->num_rows - 1);
1328 rollback: /* rollback all changes made, ignore further errors */
1329 for (i--; i >= 0; i--) {
1330 cd = get_column_description(t->desc, i);
1331 enum osl_storage_type st = cd->storage_type;
1332 if (st == OSL_NO_STORAGE)
1335 if (st == OSL_MAPPED_STORAGE)
1336 truncate_mapped_file(t, i, objects[i].size);
1337 else /* disk storage */
1338 delete_disk_storage_file(t, i, ds_name);
1340 /* ignore error and return previous error value */
1341 map_table(t, MAP_TBL_FL_VERIFY_INDEX);
1343 free(new_row_index);
1350 __export int osl_add_row(struct osl_table *t, struct osl_object *objects)
1352 return osl_add_and_get_row(t, objects, NULL);
1355 __export int osl_get_object(const struct osl_table *t, const struct osl_row *r,
1356 unsigned col_num, struct osl_object *object)
1358 const struct osl_column_description *cd;
1361 return -E_OSL_BAD_TABLE;
1362 cd = get_column_description(t->desc, col_num);
1363 /* col must not be disk storage */
1364 if (cd->storage_type == OSL_DISK_STORAGE)
1365 return -E_OSL_BAD_STORAGE_TYPE;
1366 if (cd->storage_type == OSL_MAPPED_STORAGE)
1367 return get_mapped_object(t, col_num, r->num, object);
1369 *object = r->volatile_objects[t->columns[col_num].volatile_num];
1373 __export int osl_del_row(struct osl_table *t, struct osl_row *row)
1375 struct osl_row *r = row;
1377 const struct osl_column_description *cd;
1380 return -E_OSL_BAD_TABLE;
1381 INFO_LOG("deleting row %p\n", row);
1383 if (t->num_disk_storage_columns) {
1385 ret = disk_storage_name_of_row(t, r, &ds_name);
1388 FOR_EACH_DISK_STORAGE_COLUMN(i, t, cd)
1389 delete_disk_storage_file(t, i, ds_name);
1392 FOR_EACH_COLUMN(i, t->desc, cd) {
1393 struct osl_column *col = t->columns + i;
1394 enum osl_storage_type st = cd->storage_type;
1395 remove_rb_node(t, i, r);
1396 if (st == OSL_MAPPED_STORAGE)
1398 if (st == OSL_NO_STORAGE && !(cd->storage_flags & OSL_DONT_FREE))
1399 free(r->volatile_objects[col->volatile_num].data);
1401 if (t->num_mapped_columns) {
1402 ret = mark_row_invalid(t, r->num);
1405 t->num_invalid_rows++;
1410 free(r->volatile_objects);
1415 /* test if column has an rbtree */
1416 static int check_rbtree_col(const struct osl_table *t, unsigned col_num,
1417 struct osl_column **col)
1420 return -E_OSL_BAD_TABLE;
1421 if (!(get_column_description(t->desc, col_num)->storage_flags & OSL_RBTREE))
1422 return -E_OSL_BAD_STORAGE_FLAGS;
1423 *col = t->columns + col_num;
1427 __export int osl_get_row(const struct osl_table *t, unsigned col_num,
1428 const struct osl_object *obj, struct osl_row **result)
1431 struct rb_node *node;
1432 struct osl_row *row;
1433 struct osl_column *col;
1436 ret = check_rbtree_col(t, col_num, &col);
1439 ret = search_rbtree(obj, t, col_num, &node, NULL);
1442 row = get_row_pointer(node, t->columns[col_num].rbtree_num);
1447 static int rbtree_loop(struct osl_column *col, void *private_data,
1448 osl_rbtree_loop_func *func)
1450 struct rb_node *n, *tmp;
1452 /* this for-loop is safe against removal of an entry */
1453 for (n = rb_first(&col->rbtree), tmp = n? rb_next(n) : NULL;
1455 n = tmp, tmp = tmp? rb_next(tmp) : NULL) {
1456 struct osl_row *r = get_row_pointer(n, col->rbtree_num);
1457 if (func(r, private_data) < 0)
1463 static int rbtree_loop_reverse(struct osl_column *col, void *private_data,
1464 osl_rbtree_loop_func *func)
1466 struct rb_node *n, *tmp;
1468 /* safe against removal of an entry */
1469 for (n = rb_last(&col->rbtree), tmp = n? rb_prev(n) : NULL;
1471 n = tmp, tmp = tmp? rb_prev(tmp) : NULL) {
1472 struct osl_row *r = get_row_pointer(n, col->rbtree_num);
1473 if (func(r, private_data) < 0)
1479 __export int osl_rbtree_loop(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(col, private_data, func);
1490 __export int osl_rbtree_loop_reverse(const struct osl_table *t, unsigned col_num,
1491 void *private_data, osl_rbtree_loop_func *func)
1493 struct osl_column *col;
1495 int ret = check_rbtree_col(t, col_num, &col);
1498 return rbtree_loop_reverse(col, private_data, func);
1501 /* TODO: Rollback changes on errors */
1502 static int rename_disk_storage_objects(struct osl_table *t,
1503 struct osl_object *old_obj, struct osl_object *new_obj)
1506 const struct osl_column_description *cd;
1507 char *old_ds_name, *new_ds_name;
1509 if (!t->num_disk_storage_columns)
1510 return 1; /* nothing to do */
1511 if (old_obj->size == new_obj->size && !memcmp(new_obj->data,
1512 old_obj->data, new_obj->size))
1513 return 1; /* object did not change */
1514 old_ds_name = disk_storage_name_of_object(t, old_obj);
1515 new_ds_name = disk_storage_name_of_object(t, new_obj);
1517 if (!old_ds_name || ! new_ds_name)
1520 FOR_EACH_DISK_STORAGE_COLUMN(i, t, cd) {
1521 char *old_filename, *new_filename;
1522 ret = create_disk_storage_object_dir(t, i, new_ds_name);
1525 old_filename = disk_storage_path(t, i, old_ds_name);
1526 new_filename = disk_storage_path(t, i, new_ds_name);
1527 if (!old_filename || !new_filename)
1530 ret = osl_rename(old_filename, new_filename);
1544 __export int osl_update_object(struct osl_table *t, const struct osl_row *r,
1545 unsigned col_num, struct osl_object *obj)
1547 struct osl_column *col;
1548 const struct osl_column_description *cd;
1552 return -E_OSL_BAD_TABLE;
1553 col = &t->columns[col_num];
1554 cd = get_column_description(t->desc, col_num);
1555 DEBUG_LOG("updating column %u of %s\n", col_num, t->desc->name);
1556 if (cd->storage_flags & OSL_RBTREE) {
1557 if (search_rbtree(obj, t, col_num, NULL, NULL) > 0)
1558 return -E_OSL_RB_KEY_EXISTS;
1560 if (cd->storage_flags & OSL_FIXED_SIZE) {
1561 if (obj->size != cd->data_size)
1562 return -E_OSL_BAD_DATA_SIZE;
1564 remove_rb_node(t, col_num, r);
1565 if (cd->storage_type == OSL_NO_STORAGE) { /* TODO: If fixed size, reuse object? */
1566 free(r->volatile_objects[col->volatile_num].data);
1567 r->volatile_objects[col->volatile_num] = *obj;
1568 } else if (cd->storage_type == OSL_DISK_STORAGE) {
1570 ret = disk_storage_name_of_row(t, r, &ds_name);
1573 ret = delete_disk_storage_file(t, col_num, ds_name);
1574 if (ret < 0 && ret != -E_OSL_NOENT) {
1578 ret = write_disk_storage_file(t, col_num, obj, ds_name);
1582 } else { /* mapped storage */
1583 struct osl_object old_obj;
1584 ret = get_mapped_object(t, col_num, r->num, &old_obj);
1588 * If the updated column is the disk storage name column, the
1589 * disk storage name changes, so we have to rename all disk
1590 * storage objects accordingly.
1592 if (col_num == t->disk_storage_name_column) {
1593 ret = rename_disk_storage_objects(t, &old_obj, obj);
1597 if (cd->storage_flags & OSL_FIXED_SIZE)
1598 memcpy(old_obj.data, obj->data, cd->data_size);
1599 else { /* TODO: if the size doesn't change, use old space */
1600 uint32_t new_data_map_size;
1602 ret = get_row_index(t, r->num, &row_index);
1605 unmap_column(t, col_num);
1606 ret = append_map_file(t, col_num, obj,
1607 &new_data_map_size);
1610 ret = map_column(t, col_num);
1613 update_cell_index(row_index, col, new_data_map_size,
1617 if (cd->storage_flags & OSL_RBTREE) {
1618 ret = insert_rbtree(t, col_num, r, obj);
1625 __export int osl_open_disk_object(const struct osl_table *t, const struct osl_row *r,
1626 unsigned col_num, struct osl_object *obj)
1628 const struct osl_column_description *cd;
1629 char *ds_name, *filename;
1633 return -E_OSL_BAD_TABLE;
1634 cd = get_column_description(t->desc, col_num);
1635 if (cd->storage_type != OSL_DISK_STORAGE)
1636 return -E_OSL_BAD_STORAGE_TYPE;
1638 ret = disk_storage_name_of_row(t, r, &ds_name);
1641 filename = disk_storage_path(t, col_num, ds_name);
1644 return -E_OSL_NOMEM;
1645 DEBUG_LOG("filename: %s\n", filename);
1646 ret = mmap_full_file(filename, O_RDONLY, &obj->data, &obj->size, NULL);
1651 __export int osl_close_disk_object(struct osl_object *obj)
1653 return osl_munmap(obj->data, obj->size);
1656 __export int osl_get_num_rows(const struct osl_table *t, unsigned *num_rows)
1659 return -E_OSL_BAD_TABLE;
1660 assert(t->num_rows >= t->num_invalid_rows);
1661 *num_rows = t->num_rows - t->num_invalid_rows;
1665 __export int osl_get_rank(const struct osl_table *t, struct osl_row *r,
1666 unsigned col_num, unsigned *rank)
1668 struct osl_object obj;
1669 struct osl_column *col;
1670 struct rb_node *node;
1671 int ret = check_rbtree_col(t, col_num, &col);
1675 ret = osl_get_object(t, r, col_num, &obj);
1678 ret = search_rbtree(&obj, t, col_num, &node, NULL);
1681 ret = rb_rank(node, rank);
1683 return -E_OSL_BAD_ROW;
1687 __export int osl_get_nth_row(const struct osl_table *t, unsigned col_num,
1688 unsigned n, struct osl_row **result)
1690 struct osl_column *col;
1691 struct rb_node *node;
1696 return -E_OSL_RB_KEY_NOT_FOUND;
1697 ret = osl_get_num_rows(t, &num_rows);
1701 return -E_OSL_RB_KEY_NOT_FOUND;
1702 ret = check_rbtree_col(t, col_num, &col);
1705 node = rb_nth(col->rbtree.rb_node, n);
1707 return -E_OSL_RB_KEY_NOT_FOUND;
1708 *result = get_row_pointer(node, col->rbtree_num);
1712 __export int osl_rbtree_first_row(const struct osl_table *t, unsigned col_num,
1713 struct osl_row **result)
1715 return osl_get_nth_row(t, col_num, 1, result);
1718 __export int osl_rbtree_last_row(const struct osl_table *t, unsigned col_num,
1719 struct osl_row **result)
1722 int ret = osl_get_num_rows(t, &num_rows);
1726 return osl_get_nth_row(t, col_num, num_rows, result);