2 * Copyright (C) 2007-2008 Andre Noll <maan@systemlinux.org>
4 * Licensed under the GPL v2. For licencing details see COPYING.
7 /** \file osl.c Object storage layer functions. */
8 #include <dirent.h> /* readdir() */
18 /* Taken from Drepper: How to write shared libraries, Appendix B. */
20 #define MSGSTRFIELD(line) MSGSTRFIELD1(line)
21 #define MSGSTRFIELD1(line) str##line
22 static const union msgstr_t {
24 #define _S(n, s) char MSGSTRFIELD(__LINE__)[sizeof(s)];
34 static const unsigned int errmsgidx[] = {
35 #define _S(n, s) [n] = offsetof(union msgstr_t, MSGSTRFIELD(__LINE__)),
40 __export const char *osl_strerror(int num)
42 if (IS_SYSTEM_ERROR(num))
43 return strerror((num) & ((1 << SYSTEM_ERROR_BIT) - 1));
44 return msgstr.str + errmsgidx[num];
49 static void __attribute ((constructor)) init_loglevel(void)
51 char *p = getenv("OSL_LOGLEVEL");
53 /* don't log anything if unset */
54 loglevel = p? atoi(p) : EMERG + 1;
61 * \param fmt Usual format string.
63 * All XXX_LOG() macros use this function.
65 __printf_2_3 void __log(int ll, const char* fmt,...)
78 strftime(str, sizeof(str), "%b %d %H:%M:%S", tm);
79 fprintf(outfd, "%s ", str);
81 vfprintf(outfd, fmt, argp);
86 * A wrapper for lseek(2).
88 * \param fd The file descriptor whose offset is to be to repositioned.
89 * \param offset A value-result parameter.
90 * \param whence Usual repositioning directive.
92 * Reposition the offset of the file descriptor \a fd to the argument \a offset
93 * according to the directive \a whence. Upon successful return, \a offset
94 * contains the resulting offset location as measured in bytes from the
95 * beginning of the file.
97 * \return Positive on success. Otherwise, the function returns \p -E_OSL_LSEEK.
101 static int __lseek(int fd, off_t *offset, int whence)
103 *offset = lseek(fd, *offset, whence);
104 int ret = -E_OSL_LSEEK;
110 static int append_file(const char *filename, char *header, size_t header_size,
111 char *data, size_t data_size, uint32_t *new_pos)
115 // DEBUG_LOG("appending %zu + %zu bytes\n", header_size, data_size);
116 ret = osl_open(filename, O_WRONLY | O_CREAT | O_APPEND, 0644);
120 if (header && header_size) {
121 ret = write_all(fd, header, &header_size);
125 ret = write_all(fd, data, &data_size);
130 ret = __lseek(fd, &offset, SEEK_END);
133 // DEBUG_LOG("new file size: " FMT_OFF_T "\n", offset);
142 static int verify_name(const char *name)
145 return -E_OSL_BAD_NAME;
147 return -E_OSL_BAD_NAME;
148 if (strchr(name, '/'))
149 return -E_OSL_BAD_NAME;
150 if (!strcmp(name, ".."))
151 return -E_OSL_BAD_NAME;
152 if (!strcmp(name, "."))
153 return -E_OSL_BAD_NAME;
157 int osl_hash_compare(const struct osl_object *obj1, const struct osl_object *obj2)
159 return hash_compare((HASH_TYPE *)obj1->data, (HASH_TYPE *)obj2->data);
162 static char *disk_storage_dirname(const struct osl_table *t, unsigned col_num,
165 char *dirname, *column_name = column_filename(t, col_num);
169 if (!(t->desc->flags & OSL_LARGE_TABLE))
171 dirname = make_message("%s/%.2s", column_name, ds_name);
176 static char *disk_storage_name_of_object(const struct osl_table *t,
177 const struct osl_object *obj)
179 HASH_TYPE hash[HASH_SIZE];
180 hash_object(obj, hash);
181 return disk_storage_name_of_hash(t, hash);
184 static int disk_storage_name_of_row(const struct osl_table *t,
185 const struct osl_row *row, char **name)
187 struct osl_object obj;
188 int ret = osl_get_object(t, row, t->disk_storage_name_column, &obj);
192 *name = disk_storage_name_of_object(t, &obj);
195 return -ERRNO_TO_ERROR(ENOMEM);
198 static void column_name_hash(const char *col_name, HASH_TYPE *hash)
200 hash_function(col_name, strlen(col_name), hash);
203 static int init_column_descriptions(struct osl_table *t)
206 const struct osl_column_description *cd;
208 ret = -E_OSL_BAD_TABLE_DESC;
209 ret = verify_name(t->desc->name);
212 ret = -E_OSL_BAD_DB_DIR;
213 if (!t->desc->dir && (t->num_disk_storage_columns || t->num_mapped_columns))
215 /* the size of the index header without column descriptions */
216 t->index_header_size = IDX_COLUMN_DESCRIPTIONS;
217 FOR_EACH_COLUMN(i, t->desc, cd) {
218 struct osl_column *col = t->columns + i;
219 if (cd->storage_flags & OSL_RBTREE) {
220 if (!cd->compare_function)
221 return -E_OSL_NO_COMPARE_FUNC;
223 if (cd->storage_type == OSL_NO_STORAGE)
225 ret = -E_OSL_NO_COLUMN_NAME;
226 if (!cd->name || !cd->name[0])
228 ret = verify_name(cd->name);
231 t->index_header_size += index_column_description_size(cd->name);
232 column_name_hash(cd->name, col->name_hash);
233 ret = -E_OSL_DUPLICATE_COL_NAME;
234 for (j = i + 1; j < t->desc->num_columns; j++) {
235 const char *name2 = get_column_description(t->desc,
237 if (cd->name && name2 && !strcmp(cd->name, name2))
247 * Initialize a struct table from given table description.
249 * \param desc The description of the osl table.
250 * \param table_ptr Result is returned here.
252 * This function performs several sanity checks on \p desc and returns if any
253 * of these tests fail. On success, a struct \p osl_table is allocated and
254 * initialized with data derived from \p desc.
258 * \sa struct osl_table.
260 int init_table_structure(const struct osl_table_description *desc,
261 struct osl_table **table_ptr)
263 const struct osl_column_description *cd;
264 struct osl_table *t = calloc(1, sizeof(*t));
265 int i, ret = -ERRNO_TO_ERROR(ENOMEM), have_disk_storage_name_column = 0;
269 ret = -E_OSL_BAD_TABLE_DESC;
272 DEBUG_LOG("creating table structure for '%s' from table "
273 "description\n", desc->name);
274 ret = -E_OSL_NO_COLUMN_DESC;
275 if (!desc->column_descriptions)
277 ret = -E_OSL_NO_COLUMNS;
278 if (!desc->num_columns)
280 ret = -ERRNO_TO_ERROR(ENOMEM);
281 t->columns = calloc(desc->num_columns, sizeof(struct osl_column));
285 FOR_EACH_COLUMN(i, t->desc, cd) {
286 enum osl_storage_type st = cd->storage_type;
287 enum osl_storage_flags sf = cd->storage_flags;
288 struct osl_column *col = &t->columns[i];
290 ret = -E_OSL_BAD_STORAGE_TYPE;
291 if (st != OSL_MAPPED_STORAGE && st != OSL_DISK_STORAGE
292 && st != OSL_NO_STORAGE)
294 ret = -E_OSL_BAD_STORAGE_FLAGS;
295 if (st == OSL_DISK_STORAGE && sf & OSL_RBTREE)
297 ret = -E_OSL_BAD_STORAGE_SIZE;
298 if (sf & OSL_FIXED_SIZE && !cd->data_size)
301 case OSL_DISK_STORAGE:
302 t->num_disk_storage_columns++;
304 case OSL_MAPPED_STORAGE:
305 t->num_mapped_columns++;
306 col->index_offset = t->row_index_size;
307 t->row_index_size += 8;
310 col->volatile_num = t->num_volatile_columns;
311 t->num_volatile_columns++;
314 if (sf & OSL_RBTREE) {
315 col->rbtree_num = t->num_rbtrees;
317 if ((sf & OSL_UNIQUE) && (st == OSL_MAPPED_STORAGE)) {
318 if (!have_disk_storage_name_column)
319 t->disk_storage_name_column = i;
320 have_disk_storage_name_column = 1;
324 ret = -E_OSL_NO_UNIQUE_RBTREE_COLUMN;
325 if (t->num_disk_storage_columns && !have_disk_storage_name_column)
327 ret = -E_OSL_NO_RBTREE_COL;
331 DEBUG_LOG("OK. Index entry size: %u\n", t->row_index_size);
332 ret = init_column_descriptions(t);
344 * Read the table description from index header.
346 * \param map The memory mapping of the index file.
347 * \param desc The values found in the index header are returned here.
349 * Read the index header, check for the paraslash magic string and the table version number.
350 * Read all information stored in the index header into \a desc.
354 * \sa struct osl_table_description, osl_create_table.
356 int read_table_desc(struct osl_object *map, struct osl_table_description *desc)
358 char *buf = map->data;
360 uint16_t header_size;
363 struct osl_column_description *cd;
365 if (map->size < MIN_INDEX_HEADER_SIZE(1))
366 return -E_OSL_SHORT_TABLE;
367 if (strncmp(buf + IDX_PARA_MAGIC, PARA_MAGIC, strlen(PARA_MAGIC)))
368 return -E_OSL_NO_MAGIC;
369 version = read_u8(buf + IDX_VERSION);
370 if (version < MIN_TABLE_VERSION || version > MAX_TABLE_VERSION)
371 return -E_OSL_VERSION_MISMATCH;
372 desc->flags = read_u8(buf + IDX_TABLE_FLAGS);
373 desc->num_columns = read_u16(buf + IDX_NUM_COLUMNS);
374 INFO_LOG("%u columns\n", desc->num_columns);
375 if (!desc->num_columns)
376 return -E_OSL_NO_COLUMNS;
377 header_size = read_u16(buf + IDX_HEADER_SIZE);
378 if (map->size < header_size)
379 return -E_OSL_BAD_SIZE;
380 desc->column_descriptions = calloc(desc->num_columns,
381 sizeof(struct osl_column_description));
382 if (!desc->column_descriptions)
383 return -ERRNO_TO_ERROR(ENOMEM);
384 offset = IDX_COLUMN_DESCRIPTIONS;
385 FOR_EACH_COLUMN(i, desc, cd) {
388 ret = -E_OSL_SHORT_TABLE;
389 if (map->size < offset + MIN_IDX_COLUMN_DESCRIPTION_SIZE) {
390 ERROR_LOG("map size = %zu < %u = offset + min desc size\n",
391 map->size, offset + MIN_IDX_COLUMN_DESCRIPTION_SIZE);
394 cd->storage_type = read_u16(buf + offset + IDX_CD_STORAGE_TYPE);
395 cd->storage_flags = read_u16(buf + offset +
396 IDX_CD_STORAGE_FLAGS);
397 cd->data_size = read_u32(buf + offset + IDX_CD_DATA_SIZE);
398 null_byte = memchr(buf + offset + IDX_CD_NAME, '\0',
399 map->size - offset - IDX_CD_NAME);
400 ret = -E_OSL_INDEX_CORRUPTION;
403 ret = -ERRNO_TO_ERROR(ENOMEM);
404 cd->name = strdup(buf + offset + IDX_CD_NAME);
407 offset += index_column_description_size(cd->name);
409 if (offset != header_size) {
410 ret = -E_OSL_INDEX_CORRUPTION;
411 ERROR_LOG("real header size = %u != %u = stored header size\n",
412 offset, header_size);
417 FOR_EACH_COLUMN(i, desc, cd)
423 * check whether the table description given by \p t->desc matches the on-disk
424 * table structure stored in the index of \a t.
426 static int compare_table_descriptions(struct osl_table *t)
429 struct osl_table_description desc;
430 const struct osl_column_description *cd1, *cd2;
432 /* read the on-disk structure into desc */
433 ret = read_table_desc(&t->index_map, &desc);
436 ret = -E_OSL_BAD_TABLE_FLAGS;
437 if (desc.flags != t->desc->flags)
439 ret = -E_OSL_BAD_COLUMN_NUM;
440 if (desc.num_columns > t->desc->num_columns)
442 if (desc.num_columns < t->desc->num_columns) {
443 struct osl_column_description *cd;
444 unsigned diff = t->desc->num_columns - desc.num_columns;
445 INFO_LOG("extending table by %u volatile columns\n", diff);
446 ret = -ERRNO_TO_ERROR(ENOMEM);
447 desc.column_descriptions = realloc(desc.column_descriptions,
448 t->desc->num_columns * sizeof(struct osl_column_description));
449 if (!desc.column_descriptions)
451 for (i = desc.num_columns; i < t->desc->num_columns; i++) {
452 cd = get_column_description(&desc, i);
453 cd->storage_type = OSL_NO_STORAGE;
456 desc.num_columns += diff;
458 FOR_EACH_COLUMN(i, t->desc, cd1) {
459 cd2 = get_column_description(&desc, i);
460 ret = -E_OSL_BAD_STORAGE_TYPE;
461 if (cd1->storage_type != cd2->storage_type)
463 if (cd1->storage_type == OSL_NO_STORAGE)
465 ret = -E_OSL_BAD_STORAGE_FLAGS;
466 if (cd1->storage_flags != cd2->storage_flags) {
467 ERROR_LOG("sf1 = %u != %u = sf2\n",
468 cd1->storage_flags, cd2->storage_flags);
471 ret = -E_OSL_BAD_DATA_SIZE;
472 if (cd1->storage_flags & OSL_FIXED_SIZE)
473 if (cd1->data_size != cd2->data_size)
475 ret = -E_OSL_BAD_COLUMN_NAME;
476 if (strcmp(cd1->name, cd2->name))
479 INFO_LOG("table description of '%s' matches on-disk data, good\n",
483 FOR_EACH_COLUMN(i, &desc, cd1)
485 free(desc.column_descriptions);
489 static int create_table_index(struct osl_table *t)
491 char *buf, *filename;
493 size_t size = t->index_header_size;
494 const struct osl_column_description *cd;
497 INFO_LOG("creating %zu byte index for table %s\n", size,
499 buf = calloc(1, size);
501 return -ERRNO_TO_ERROR(ENOMEM);
502 sprintf(buf + IDX_PARA_MAGIC, "%s", PARA_MAGIC);
503 write_u8(buf + IDX_TABLE_FLAGS, t->desc->flags);
504 write_u8(buf + IDX_DIRTY_FLAG, 0);
505 write_u8(buf + IDX_VERSION, CURRENT_TABLE_VERSION);
506 write_u16(buf + IDX_NUM_COLUMNS, t->num_mapped_columns + t->num_disk_storage_columns);
507 write_u16(buf + IDX_HEADER_SIZE, t->index_header_size);
508 offset = IDX_COLUMN_DESCRIPTIONS;
509 FOR_EACH_COLUMN(i, t->desc, cd) {
510 /* no need to store info about volatile storage */
511 if (cd->storage_type == OSL_NO_STORAGE)
513 write_u16(buf + offset + IDX_CD_STORAGE_TYPE,
515 write_u16(buf + offset + IDX_CD_STORAGE_FLAGS,
517 if (cd->storage_flags & OSL_FIXED_SIZE)
518 write_u32(buf + offset + IDX_CD_DATA_SIZE,
520 strcpy(buf + offset + IDX_CD_NAME, cd->name);
521 offset += index_column_description_size(cd->name);
523 assert(offset = size);
524 filename = index_filename(t->desc);
526 ret = write_file(filename, buf, size);
528 ret = -ERRNO_TO_ERROR(ENOMEM);
534 __export int osl_create_table(const struct osl_table_description *desc)
536 const struct osl_column_description *cd;
537 char *table_dir = NULL, *filename;
539 int i, ret = init_table_structure(desc, &t);
543 INFO_LOG("creating %s\n", desc->name);
544 FOR_EACH_COLUMN(i, t->desc, cd) {
545 if (cd->storage_type == OSL_NO_STORAGE)
548 ret = osl_mkdir(desc->dir, 0777);
549 if (ret < 0 && !is_errno(-ret, EEXIST))
551 table_dir = make_message("%s/%s", desc->dir,
553 ret = -ERRNO_TO_ERROR(ENOMEM);
556 ret = osl_mkdir(table_dir, 0777);
560 ret = -ERRNO_TO_ERROR(ENOMEM);
561 filename = column_filename(t, i);
564 INFO_LOG("filename: %s\n", filename);
565 if (cd->storage_type == OSL_MAPPED_STORAGE) {
566 ret = osl_open(filename, O_RDWR | O_CREAT | O_EXCL,
575 ret = osl_mkdir(filename, 0777);
580 if (t->num_mapped_columns) {
581 ret = create_table_index(t);
593 static int table_is_dirty(struct osl_table *t)
595 char *buf = (char *)t->index_map.data + IDX_DIRTY_FLAG;
596 uint8_t dirty = read_u8(buf) & 0x1;
600 static void mark_table_dirty(struct osl_table *t)
602 char *buf = (char *)t->index_map.data + IDX_DIRTY_FLAG;
603 write_u8(buf, read_u8(buf) | 1);
606 static void mark_table_clean(struct osl_table *t)
608 char *buf = (char *)t->index_map.data + IDX_DIRTY_FLAG;
609 write_u8(buf, read_u8(buf) & 0xfe);
612 static void unmap_column(struct osl_table *t, unsigned col_num)
614 struct osl_object map = t->columns[col_num].data_map;
618 ret = osl_munmap(map.data, map.size);
624 * Unmap all mapped files of an osl table.
626 * \param t Pointer to a mapped table.
627 * \param flags Options for unmapping.
629 * \return Positive on success, negative on errors.
631 * \sa map_table(), enum osl_close_flags, osl_munmap().
633 int unmap_table(struct osl_table *t, enum osl_close_flags flags)
636 const struct osl_column_description *cd;
639 if (!t->num_mapped_columns) /* can this ever happen? */
641 INFO_LOG("unmapping table '%s'\n", t->desc->name);
642 if (!t->index_map.data)
643 return -E_OSL_NOT_MAPPED;
644 if (flags & OSL_MARK_CLEAN)
646 ret = osl_munmap(t->index_map.data, t->index_map.size);
649 t->index_map.data = NULL;
652 FOR_EACH_MAPPED_COLUMN(i, t, cd)
657 static int map_column(struct osl_table *t, unsigned col_num)
660 char *filename = column_filename(t, col_num);
661 int ret = -E_OSL_STAT;
664 return -ERRNO_TO_ERROR(ENOMEM);
665 if (stat(filename, &statbuf) < 0) {
669 if (!(S_IFREG & statbuf.st_mode)) {
673 ret = mmap_full_file(filename, O_RDWR,
674 &t->columns[col_num].data_map.data,
675 &t->columns[col_num].data_map.size,
682 * Map the index file and all columns of type \p OSL_MAPPED_STORAGE into memory.
684 * \param t Pointer to an initialized table structure.
685 * \param flags Mapping options.
687 * \return Negative return value on errors; on success the number of rows
688 * (including invalid rows) is returned.
690 * \sa unmap_table(), enum map_table_flags, osl_open_table(), mmap(2).
692 int map_table(struct osl_table *t, enum map_table_flags flags)
695 const struct osl_column_description *cd;
696 int i = 0, ret, num_rows = 0;
698 if (!t->num_mapped_columns)
700 if (t->index_map.data)
701 return -E_OSL_ALREADY_MAPPED;
702 filename = index_filename(t->desc);
704 return -ERRNO_TO_ERROR(ENOMEM);
705 INFO_LOG("mapping table '%s' (index: %s)\n", t->desc->name, filename);
706 ret = mmap_full_file(filename, flags & MAP_TBL_FL_MAP_RDONLY?
707 O_RDONLY : O_RDWR, &t->index_map.data, &t->index_map.size, NULL);
711 if (flags & MAP_TBL_FL_VERIFY_INDEX) {
712 ret = compare_table_descriptions(t);
717 if (!(flags & MAP_TBL_FL_IGNORE_DIRTY)) {
718 if (table_is_dirty(t)) {
719 ERROR_LOG("%s is dirty\n", t->desc->name);
724 num_rows = table_num_rows(t);
728 FOR_EACH_MAPPED_COLUMN(i, t, cd) {
729 ret = map_column(t, i);
734 err: /* unmap what is already mapped */
735 for (i--; i >= 0; i--) {
736 struct osl_object map = t->columns[i].data_map;
737 osl_munmap(map.data, map.size);
740 osl_munmap(t->index_map.data, t->index_map.size);
741 t->index_map.data = NULL;
746 * Retrieve a mapped object by row and column number.
748 * \param t Pointer to an open osl table.
749 * \param col_num Number of the mapped column containing the object to retrieve.
750 * \param row_num Number of the row containing the object to retrieve.
751 * \param obj The result is returned here.
753 * It is considered an error if \a col_num does not refer to a column
754 * of storage type \p OSL_MAPPED_STORAGE.
758 * \sa osl_storage_type.
760 int get_mapped_object(const struct osl_table *t, unsigned col_num,
761 uint32_t row_num, struct osl_object *obj)
763 struct osl_column *col = &t->columns[col_num];
769 if (t->num_rows <= row_num)
770 return -E_OSL_BAD_ROW_NUM;
771 ret = get_cell_index(t, row_num, col_num, &cell_index);
774 offset = read_u32(cell_index);
775 obj->size = read_u32(cell_index + 4) - 1;
776 header = col->data_map.data + offset;
777 obj->data = header + 1;
778 if (read_u8(header) == 0xff) {
779 ERROR_LOG("col %u, size %zu, offset %u\n", col_num,
781 return -E_OSL_INVALID_OBJECT;
786 static int search_rbtree(const struct osl_object *obj,
787 const struct osl_table *t, unsigned col_num,
788 struct rb_node **result, struct rb_node ***rb_link)
790 struct osl_column *col = &t->columns[col_num];
791 struct rb_node **new = &col->rbtree.rb_node, *parent = NULL;
792 const struct osl_column_description *cd =
793 get_column_description(t->desc, col_num);
794 enum osl_storage_type st = cd->storage_type;
796 struct osl_row *this_row = get_row_pointer(*new,
799 struct osl_object this_obj;
801 if (st == OSL_MAPPED_STORAGE) {
802 ret = get_mapped_object(t, col_num, this_row->num,
807 this_obj = this_row->volatile_objects[col->volatile_num];
808 ret = cd->compare_function(obj, &this_obj);
811 *result = get_rb_node_pointer(this_row,
816 new = &((*new)->rb_left);
818 new = &((*new)->rb_right);
824 return -E_OSL_RB_KEY_NOT_FOUND;
827 static int insert_rbtree(struct osl_table *t, unsigned col_num,
828 const struct osl_row *row, const struct osl_object *obj)
830 struct rb_node *parent, **rb_link;
833 int ret = search_rbtree(obj, t, col_num, &parent, &rb_link);
836 return -E_OSL_RB_KEY_EXISTS;
837 rbtree_num = t->columns[col_num].rbtree_num;
838 n = get_rb_node_pointer(row, rbtree_num);
839 rb_link_node(n, parent, rb_link);
840 rb_insert_color(n, &t->columns[col_num].rbtree);
844 static void remove_rb_node(struct osl_table *t, unsigned col_num,
845 const struct osl_row *row)
847 struct osl_column *col = &t->columns[col_num];
848 const struct osl_column_description *cd =
849 get_column_description(t->desc, col_num);
850 enum osl_storage_flags sf = cd->storage_flags;
851 struct rb_node *victim, *splice_out_node, *tmp;
852 if (!(sf & OSL_RBTREE))
855 * Which node is removed/spliced out actually depends on how many
856 * children the victim node has: If it has no children, it gets
857 * deleted. If it has one child, it gets spliced out. If it has two
858 * children, its successor (which has at most a right child) gets
861 victim = get_rb_node_pointer(row, col->rbtree_num);
862 if (victim->rb_left && victim->rb_right)
863 splice_out_node = rb_next(victim);
865 splice_out_node = victim;
866 /* Go up to the root and decrement the size of each node in the path. */
867 for (tmp = splice_out_node; tmp; tmp = rb_parent(tmp))
869 rb_erase(victim, &col->rbtree);
872 static int add_row_to_rbtrees(struct osl_table *t, uint32_t row_num,
873 struct osl_object *volatile_objs, struct osl_row **row_ptr)
877 struct osl_row *row = allocate_row(t->num_rbtrees);
878 const struct osl_column_description *cd;
881 return -ERRNO_TO_ERROR(ENOMEM);
883 row->volatile_objects = volatile_objs;
884 FOR_EACH_RBTREE_COLUMN(i, t, cd) {
885 if (cd->storage_type == OSL_MAPPED_STORAGE) {
886 struct osl_object obj;
887 ret = get_mapped_object(t, i, row_num, &obj);
890 ret = insert_rbtree(t, i, row, &obj);
891 } else { /* volatile */
892 const struct osl_object *obj
893 = volatile_objs + t->columns[i].volatile_num;
894 ret = insert_rbtree(t, i, row, obj);
902 err: /* rollback changes, i.e. remove added entries from rbtrees */
904 remove_rb_node(t, i--, row);
909 static void free_volatile_objects(const struct osl_table *t,
910 enum osl_close_flags flags)
914 struct osl_column *rb_col;
915 const struct osl_column_description *cd;
917 if (!t->num_volatile_columns)
919 /* find the first rbtree column (any will do) */
920 FOR_EACH_RBTREE_COLUMN(i, t, cd)
922 rb_col = t->columns + i;
923 /* walk that rbtree and free all volatile objects */
924 for (n = rb_first(&rb_col->rbtree); n; n = rb_next(n)) {
925 struct osl_row *r = get_row_pointer(n, rb_col->rbtree_num);
926 if (flags & OSL_FREE_VOLATILE)
927 FOR_EACH_VOLATILE_COLUMN(j, t, cd) {
928 if (cd->storage_flags & OSL_DONT_FREE)
930 free(r->volatile_objects[
931 t->columns[j].volatile_num].data);
933 // for (j = 0; j < t->num_volatile_columns; j++)
934 // free(r->volatile_objects[j].data);
935 free(r->volatile_objects);
940 * Erase all rbtree nodes and free resources.
942 * \param t Pointer to an open osl table.
944 * This function is called by osl_close_table().
946 void clear_rbtrees(struct osl_table *t)
948 const struct osl_column_description *cd;
949 unsigned i, rbtrees_cleared = 0;
951 FOR_EACH_RBTREE_COLUMN(i, t, cd) {
952 struct osl_column *col = &t->columns[i];
955 for (n = rb_first(&col->rbtree); n;) {
957 rb_erase(n, &col->rbtree);
958 if (rbtrees_cleared == t->num_rbtrees) {
959 r = get_row_pointer(n, col->rbtree_num);
969 __export int osl_close_table(struct osl_table *t, enum osl_close_flags flags)
974 return -E_OSL_BAD_TABLE;
975 NOTICE_LOG("closing table %s\n", t->desc->name);
976 free_volatile_objects(t, flags);
978 ret = unmap_table(t, flags);
980 ERROR_LOG("unmap_table failed: %d\n", ret);
987 * Find out whether the given row number corresponds to an invalid row.
989 * \param t Pointer to the osl table.
990 * \param row_num The number of the row in question.
992 * By definition, a row is considered invalid if all its index entries
995 * \return Positive if \a row_num corresponds to an invalid row,
996 * zero if it corresponds to a valid row, negative on errors.
998 int row_is_invalid(struct osl_table *t, uint32_t row_num)
1001 int i, ret = get_row_index(t, row_num, &row_index);
1005 for (i = 0; i < t->row_index_size; i++) {
1006 if ((unsigned char)row_index[i] != 0xff)
1009 INFO_LOG("row %d is invalid\n", row_num);
1014 * Invalidate a row of an osl table.
1016 * \param t Pointer to an open osl table.
1017 * \param row_num Number of the row to mark as invalid.
1019 * This function marks each mapped object in the index entry of \a row as
1024 int mark_row_invalid(struct osl_table *t, uint32_t row_num)
1027 int ret = get_row_index(t, row_num, &row_index);
1031 INFO_LOG("marking row %d as invalid\n", row_num);
1032 memset(row_index, 0xff, t->row_index_size);
1037 * Initialize all rbtrees and compute number of invalid rows.
1039 * \param t The table containing the rbtrees to be initialized.
1043 int init_rbtrees(struct osl_table *t)
1046 const struct osl_column_description *cd;
1048 /* create rbtrees */
1049 FOR_EACH_RBTREE_COLUMN(i, t, cd)
1050 t->columns[i].rbtree = RB_ROOT;
1051 /* add valid rows to rbtrees */
1052 t->num_invalid_rows = 0;
1053 for (i = 0; i < t->num_rows; i++) {
1054 ret = row_is_invalid(t, i);
1058 t->num_invalid_rows++;
1061 ret = add_row_to_rbtrees(t, i, NULL, NULL);
1068 __export int osl_open_table(const struct osl_table_description *table_desc,
1069 struct osl_table **result)
1072 struct osl_table *t;
1073 const struct osl_column_description *cd;
1075 NOTICE_LOG("opening table %s\n", table_desc->name);
1076 ret = init_table_structure(table_desc, &t);
1079 FOR_EACH_DISK_STORAGE_COLUMN(i, t, cd) {
1080 struct stat statbuf;
1081 char *dirname = column_filename(t, i);
1083 ret = -ERRNO_TO_ERROR(ENOMEM);
1086 /* check if directory exists */
1087 ret = stat(dirname, &statbuf);
1090 ret = -ERRNO_TO_ERROR(errno);
1093 ret = -ERRNO_TO_ERROR(ENOTDIR);
1094 if (!S_ISDIR(statbuf.st_mode))
1097 ret = map_table(t, MAP_TBL_FL_VERIFY_INDEX);
1101 DEBUG_LOG("num rows: %d\n", t->num_rows);
1102 ret = init_rbtrees(t);
1104 osl_close_table(t, OSL_MARK_CLEAN); /* ignore further errors */
1115 static int create_disk_storage_object_dir(const struct osl_table *t,
1116 unsigned col_num, const char *ds_name)
1121 if (!(t->desc->flags & OSL_LARGE_TABLE))
1123 dirname = disk_storage_dirname(t, col_num, ds_name);
1125 return -ERRNO_TO_ERROR(ENOMEM);
1126 ret = osl_mkdir(dirname, 0777);
1128 if (ret < 0 && !is_errno(-ret, EEXIST))
1133 static int write_disk_storage_file(const struct osl_table *t, unsigned col_num,
1134 const struct osl_object *obj, const char *ds_name)
1139 ret = create_disk_storage_object_dir(t, col_num, ds_name);
1142 filename = disk_storage_path(t, col_num, ds_name);
1144 return -ERRNO_TO_ERROR(ENOMEM);
1145 ret = write_file(filename, obj->data, obj->size);
1150 static int append_map_file(const struct osl_table *t, unsigned col_num,
1151 const struct osl_object *obj, uint32_t *new_size)
1153 char *filename = column_filename(t, col_num);
1155 char header = 0; /* zero means valid object */
1158 return -ERRNO_TO_ERROR(ENOMEM);
1159 ret = append_file(filename, &header, 1, obj->data, obj->size,
1165 static int append_row_index(const struct osl_table *t, char *row_index)
1170 if (!t->num_mapped_columns)
1172 filename = index_filename(t->desc);
1174 return -ERRNO_TO_ERROR(ENOMEM);
1175 ret = append_file(filename, NULL, 0, row_index,
1176 t->row_index_size, NULL);
1182 * A wrapper for truncate(2)
1184 * \param path Name of the regular file to truncate
1185 * \param size Number of bytes to \b shave \b off
1187 * Truncate the regular file named by \a path by \a size bytes.
1193 int para_truncate(const char *path, off_t size)
1196 struct stat statbuf;
1199 if (stat(path, &statbuf) < 0)
1201 ret = -E_OSL_BAD_SIZE;
1202 if (statbuf.st_size < size)
1205 if (truncate(path, statbuf.st_size - size) < 0)
1212 static int truncate_mapped_file(const struct osl_table *t, unsigned col_num,
1216 char *filename = column_filename(t, col_num);
1219 return -ERRNO_TO_ERROR(ENOMEM);
1220 ret = para_truncate(filename, size);
1225 static int delete_disk_storage_file(const struct osl_table *t, unsigned col_num,
1226 const char *ds_name)
1228 char *dirname, *filename = disk_storage_path(t, col_num, ds_name);
1232 return -ERRNO_TO_ERROR(ENOMEM);
1233 ret = unlink(filename);
1237 return -ERRNO_TO_ERROR(err);
1238 if (!(t->desc->flags & OSL_LARGE_TABLE))
1240 dirname = disk_storage_dirname(t, col_num, ds_name);
1242 return -ERRNO_TO_ERROR(ENOMEM);
1248 __export int osl_add_and_get_row(struct osl_table *t, struct osl_object *objects,
1249 struct osl_row **row)
1252 char *ds_name = NULL;
1253 struct rb_node **rb_parents = NULL, ***rb_links = NULL;
1254 char *new_row_index = NULL;
1255 struct osl_object *volatile_objs = NULL;
1256 const struct osl_column_description *cd;
1259 return -E_OSL_BAD_TABLE;
1260 rb_parents = malloc(t->num_rbtrees * sizeof(struct rn_node*));
1262 return -ERRNO_TO_ERROR(ENOMEM);
1263 rb_links = malloc(t->num_rbtrees * sizeof(struct rn_node**));
1266 return -ERRNO_TO_ERROR(ENOMEM);
1268 if (t->num_mapped_columns) {
1269 new_row_index = malloc(t->row_index_size);
1270 if (!new_row_index) {
1273 return -ERRNO_TO_ERROR(ENOMEM);
1276 /* pass 1: sanity checks */
1277 // DEBUG_LOG("sanity tests: %p:%p\n", objects[0].data,
1278 // objects[1].data);
1279 FOR_EACH_COLUMN(i, t->desc, cd) {
1280 enum osl_storage_type st = cd->storage_type;
1281 enum osl_storage_flags sf = cd->storage_flags;
1283 // ret = -E_OSL_NULL_OBJECT;
1286 if (st == OSL_DISK_STORAGE)
1288 if (sf & OSL_RBTREE) {
1289 unsigned rbtree_num = t->columns[i].rbtree_num;
1290 ret = -E_OSL_RB_KEY_EXISTS;
1291 // DEBUG_LOG("checking whether %p exists\n",
1292 // objects[i].data);
1293 if (search_rbtree(objects + i, t, i,
1294 &rb_parents[rbtree_num],
1295 &rb_links[rbtree_num]) > 0)
1298 if (sf & OSL_FIXED_SIZE) {
1299 // DEBUG_LOG("fixed size. need: %zu, have: %d\n",
1300 // objects[i].size, cd->data_size);
1301 ret = -E_OSL_BAD_DATA_SIZE;
1302 if (objects[i].size != cd->data_size)
1306 if (t->num_disk_storage_columns) {
1307 ds_name = disk_storage_name_of_object(t,
1308 &objects[t->disk_storage_name_column]);
1309 ret = -ERRNO_TO_ERROR(ENOMEM);
1313 ret = unmap_table(t, OSL_MARK_CLEAN);
1316 // DEBUG_LOG("sanity tests passed%s\n", "");
1317 /* pass 2: create data files, append map data */
1318 FOR_EACH_COLUMN(i, t->desc, cd) {
1319 enum osl_storage_type st = cd->storage_type;
1320 if (st == OSL_NO_STORAGE)
1322 if (st == OSL_MAPPED_STORAGE) {
1324 struct osl_column *col = &t->columns[i];
1325 // DEBUG_LOG("appending object of size %zu\n",
1326 // objects[i].size);
1327 ret = append_map_file(t, i, objects + i, &new_size);
1330 update_cell_index(new_row_index, col, new_size,
1335 ret = write_disk_storage_file(t, i, objects + i, ds_name);
1339 ret = append_row_index(t, new_row_index);
1342 ret = map_table(t, MAP_TBL_FL_VERIFY_INDEX);
1343 if (ret < 0) { /* truncate index and rollback changes */
1344 char *filename = index_filename(t->desc);
1346 para_truncate(filename, t->row_index_size);
1350 /* pass 3: add entry to rbtrees */
1351 if (t->num_volatile_columns) {
1352 ret = -ERRNO_TO_ERROR(ENOMEM);
1353 volatile_objs = calloc(t->num_volatile_columns,
1354 sizeof(struct osl_object));
1357 FOR_EACH_VOLATILE_COLUMN(i, t, cd)
1358 volatile_objs[t->columns[i].volatile_num] = objects[i];
1361 // DEBUG_LOG("adding new entry as row #%d\n", t->num_rows - 1);
1362 ret = add_row_to_rbtrees(t, t->num_rows - 1, volatile_objs, row);
1365 // DEBUG_LOG("added new entry as row #%d\n", t->num_rows - 1);
1368 rollback: /* rollback all changes made, ignore further errors */
1369 for (i--; i >= 0; i--) {
1370 cd = get_column_description(t->desc, i);
1371 enum osl_storage_type st = cd->storage_type;
1372 if (st == OSL_NO_STORAGE)
1375 if (st == OSL_MAPPED_STORAGE)
1376 truncate_mapped_file(t, i, objects[i].size);
1377 else /* disk storage */
1378 delete_disk_storage_file(t, i, ds_name);
1380 /* ignore error and return previous error value */
1381 map_table(t, MAP_TBL_FL_VERIFY_INDEX);
1383 free(new_row_index);
1390 __export int osl_add_row(struct osl_table *t, struct osl_object *objects)
1392 return osl_add_and_get_row(t, objects, NULL);
1395 __export int osl_get_object(const struct osl_table *t, const struct osl_row *r,
1396 unsigned col_num, struct osl_object *object)
1398 const struct osl_column_description *cd;
1401 return -E_OSL_BAD_TABLE;
1402 cd = get_column_description(t->desc, col_num);
1403 /* col must not be disk storage */
1404 if (cd->storage_type == OSL_DISK_STORAGE)
1405 return -E_OSL_BAD_STORAGE_TYPE;
1406 if (cd->storage_type == OSL_MAPPED_STORAGE)
1407 return get_mapped_object(t, col_num, r->num, object);
1409 *object = r->volatile_objects[t->columns[col_num].volatile_num];
1413 static int mark_mapped_object_invalid(const struct osl_table *t,
1414 uint32_t row_num, unsigned col_num)
1416 struct osl_object obj;
1418 int ret = get_mapped_object(t, col_num, row_num, &obj);
1428 __export int osl_del_row(struct osl_table *t, struct osl_row *row)
1430 struct osl_row *r = row;
1432 const struct osl_column_description *cd;
1435 return -E_OSL_BAD_TABLE;
1436 INFO_LOG("deleting row %p\n", row);
1438 if (t->num_disk_storage_columns) {
1440 ret = disk_storage_name_of_row(t, r, &ds_name);
1443 FOR_EACH_DISK_STORAGE_COLUMN(i, t, cd)
1444 delete_disk_storage_file(t, i, ds_name);
1447 FOR_EACH_COLUMN(i, t->desc, cd) {
1448 struct osl_column *col = t->columns + i;
1449 enum osl_storage_type st = cd->storage_type;
1450 remove_rb_node(t, i, r);
1451 if (st == OSL_MAPPED_STORAGE) {
1452 mark_mapped_object_invalid(t, r->num, i);
1455 if (st == OSL_NO_STORAGE && !(cd->storage_flags & OSL_DONT_FREE))
1456 free(r->volatile_objects[col->volatile_num].data);
1458 if (t->num_mapped_columns) {
1459 ret = mark_row_invalid(t, r->num);
1462 t->num_invalid_rows++;
1467 free(r->volatile_objects);
1472 /* test if column has an rbtree */
1473 static int check_rbtree_col(const struct osl_table *t, unsigned col_num,
1474 struct osl_column **col)
1477 return -E_OSL_BAD_TABLE;
1478 if (!(get_column_description(t->desc, col_num)->storage_flags & OSL_RBTREE))
1479 return -E_OSL_BAD_STORAGE_FLAGS;
1480 *col = t->columns + col_num;
1484 __export int osl_get_row(const struct osl_table *t, unsigned col_num,
1485 const struct osl_object *obj, struct osl_row **result)
1488 struct rb_node *node;
1489 struct osl_row *row;
1490 struct osl_column *col;
1493 ret = check_rbtree_col(t, col_num, &col);
1496 ret = search_rbtree(obj, t, col_num, &node, NULL);
1499 row = get_row_pointer(node, t->columns[col_num].rbtree_num);
1504 static int rbtree_loop(struct osl_column *col, void *private_data,
1505 osl_rbtree_loop_func *func)
1507 struct rb_node *n, *tmp;
1509 /* this for-loop is safe against removal of an entry */
1510 for (n = rb_first(&col->rbtree), tmp = n? rb_next(n) : NULL;
1512 n = tmp, tmp = tmp? rb_next(tmp) : NULL) {
1513 struct osl_row *r = get_row_pointer(n, col->rbtree_num);
1514 if (func(r, private_data) < 0)
1520 static int rbtree_loop_reverse(struct osl_column *col, void *private_data,
1521 osl_rbtree_loop_func *func)
1523 struct rb_node *n, *tmp;
1525 /* safe against removal of an entry */
1526 for (n = rb_last(&col->rbtree), tmp = n? rb_prev(n) : NULL;
1528 n = tmp, tmp = tmp? rb_prev(tmp) : NULL) {
1529 struct osl_row *r = get_row_pointer(n, col->rbtree_num);
1530 if (func(r, private_data) < 0)
1536 __export int osl_rbtree_loop(const struct osl_table *t, unsigned col_num,
1537 void *private_data, osl_rbtree_loop_func *func)
1539 struct osl_column *col;
1541 int ret = check_rbtree_col(t, col_num, &col);
1544 return rbtree_loop(col, private_data, func);
1547 __export int osl_rbtree_loop_reverse(const struct osl_table *t, unsigned col_num,
1548 void *private_data, osl_rbtree_loop_func *func)
1550 struct osl_column *col;
1552 int ret = check_rbtree_col(t, col_num, &col);
1555 return rbtree_loop_reverse(col, private_data, func);
1558 /* TODO: Rollback changes on errors */
1559 static int rename_disk_storage_objects(struct osl_table *t,
1560 struct osl_object *old_obj, struct osl_object *new_obj)
1563 const struct osl_column_description *cd;
1564 char *old_ds_name, *new_ds_name;
1566 if (!t->num_disk_storage_columns)
1567 return 1; /* nothing to do */
1568 if (old_obj->size == new_obj->size && !memcmp(new_obj->data,
1569 old_obj->data, new_obj->size))
1570 return 1; /* object did not change */
1571 old_ds_name = disk_storage_name_of_object(t, old_obj);
1572 new_ds_name = disk_storage_name_of_object(t, new_obj);
1573 ret = -ERRNO_TO_ERROR(ENOMEM);
1574 if (!old_ds_name || ! new_ds_name)
1577 FOR_EACH_DISK_STORAGE_COLUMN(i, t, cd) {
1578 char *old_filename, *new_filename;
1579 ret = create_disk_storage_object_dir(t, i, new_ds_name);
1582 old_filename = disk_storage_path(t, i, old_ds_name);
1583 new_filename = disk_storage_path(t, i, new_ds_name);
1584 if (!old_filename || !new_filename)
1585 ret = -ERRNO_TO_ERROR(ENOMEM);
1587 ret = osl_rename(old_filename, new_filename);
1601 __export int osl_update_object(struct osl_table *t, const struct osl_row *r,
1602 unsigned col_num, struct osl_object *obj)
1604 struct osl_column *col;
1605 const struct osl_column_description *cd;
1609 return -E_OSL_BAD_TABLE;
1610 col = &t->columns[col_num];
1611 cd = get_column_description(t->desc, col_num);
1612 DEBUG_LOG("updating column %u of %s\n", col_num, t->desc->name);
1613 if (cd->storage_flags & OSL_RBTREE) {
1614 if (search_rbtree(obj, t, col_num, NULL, NULL) > 0)
1615 return -E_OSL_RB_KEY_EXISTS;
1617 if (cd->storage_flags & OSL_FIXED_SIZE) {
1618 if (obj->size != cd->data_size)
1619 return -E_OSL_BAD_DATA_SIZE;
1621 remove_rb_node(t, col_num, r);
1622 if (cd->storage_type == OSL_NO_STORAGE) { /* TODO: If fixed size, reuse object? */
1623 free(r->volatile_objects[col->volatile_num].data);
1624 r->volatile_objects[col->volatile_num] = *obj;
1625 } else if (cd->storage_type == OSL_DISK_STORAGE) {
1627 ret = disk_storage_name_of_row(t, r, &ds_name);
1630 ret = delete_disk_storage_file(t, col_num, ds_name);
1631 if (ret < 0 && !is_errno(-ret, ENOENT)) {
1635 ret = write_disk_storage_file(t, col_num, obj, ds_name);
1639 } else { /* mapped storage */
1640 struct osl_object old_obj;
1641 ret = get_mapped_object(t, col_num, r->num, &old_obj);
1645 * If the updated column is the disk storage name column, the
1646 * disk storage name changes, so we have to rename all disk
1647 * storage objects accordingly.
1649 if (col_num == t->disk_storage_name_column) {
1650 ret = rename_disk_storage_objects(t, &old_obj, obj);
1654 if (cd->storage_flags & OSL_FIXED_SIZE)
1655 memcpy(old_obj.data, obj->data, cd->data_size);
1656 else { /* TODO: if the size doesn't change, use old space */
1657 uint32_t new_data_map_size;
1659 ret = get_row_index(t, r->num, &row_index);
1662 ret = mark_mapped_object_invalid(t, r->num, col_num);
1665 unmap_column(t, col_num);
1666 ret = append_map_file(t, col_num, obj,
1667 &new_data_map_size);
1670 ret = map_column(t, col_num);
1673 update_cell_index(row_index, col, new_data_map_size,
1677 if (cd->storage_flags & OSL_RBTREE) {
1678 ret = insert_rbtree(t, col_num, r, obj);
1685 __export int osl_open_disk_object(const struct osl_table *t, const struct osl_row *r,
1686 unsigned col_num, struct osl_object *obj)
1688 const struct osl_column_description *cd;
1689 char *ds_name, *filename;
1693 return -E_OSL_BAD_TABLE;
1694 cd = get_column_description(t->desc, col_num);
1695 if (cd->storage_type != OSL_DISK_STORAGE)
1696 return -E_OSL_BAD_STORAGE_TYPE;
1698 ret = disk_storage_name_of_row(t, r, &ds_name);
1701 filename = disk_storage_path(t, col_num, ds_name);
1704 return -ERRNO_TO_ERROR(ENOMEM);
1705 DEBUG_LOG("filename: %s\n", filename);
1706 ret = mmap_full_file(filename, O_RDONLY, &obj->data, &obj->size, NULL);
1711 __export int osl_close_disk_object(struct osl_object *obj)
1713 return osl_munmap(obj->data, obj->size);
1716 __export int osl_get_num_rows(const struct osl_table *t, unsigned *num_rows)
1719 return -E_OSL_BAD_TABLE;
1720 assert(t->num_rows >= t->num_invalid_rows);
1721 *num_rows = t->num_rows - t->num_invalid_rows;
1725 __export int osl_get_rank(const struct osl_table *t, struct osl_row *r,
1726 unsigned col_num, unsigned *rank)
1728 struct osl_object obj;
1729 struct osl_column *col;
1730 struct rb_node *node;
1731 int ret = check_rbtree_col(t, col_num, &col);
1735 ret = osl_get_object(t, r, col_num, &obj);
1738 ret = search_rbtree(&obj, t, col_num, &node, NULL);
1741 ret = rb_rank(node, rank);
1743 return -E_OSL_BAD_ROW;
1747 __export int osl_get_nth_row(const struct osl_table *t, unsigned col_num,
1748 unsigned n, struct osl_row **result)
1750 struct osl_column *col;
1751 struct rb_node *node;
1756 return -E_OSL_RB_KEY_NOT_FOUND;
1757 ret = osl_get_num_rows(t, &num_rows);
1761 return -E_OSL_RB_KEY_NOT_FOUND;
1762 ret = check_rbtree_col(t, col_num, &col);
1765 node = rb_nth(col->rbtree.rb_node, n);
1767 return -E_OSL_RB_KEY_NOT_FOUND;
1768 *result = get_row_pointer(node, col->rbtree_num);
1772 __export int osl_rbtree_first_row(const struct osl_table *t, unsigned col_num,
1773 struct osl_row **result)
1775 return osl_get_nth_row(t, col_num, 1, result);
1778 __export int osl_rbtree_last_row(const struct osl_table *t, unsigned col_num,
1779 struct osl_row **result)
1782 int ret = osl_get_num_rows(t, &num_rows);
1786 return osl_get_nth_row(t, col_num, num_rows, result);