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() */
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,...)
72 strftime(str, sizeof(str), "%b %d %H:%M:%S", tm);
73 fprintf(stderr, "%s ", str);
75 vfprintf(stderr, fmt, argp);
80 * A wrapper for lseek(2).
82 * \param fd The file descriptor whose offset is to be to repositioned.
83 * \param offset A value-result parameter.
84 * \param whence Usual repositioning directive.
86 * Reposition the offset of the file descriptor \a fd to the argument \a offset
87 * according to the directive \a whence. Upon successful return, \a offset
88 * contains the resulting offset location as measured in bytes from the
89 * beginning of the file.
91 * \return Positive on success. Otherwise, the function returns \p -E_OSL_LSEEK.
95 static int __lseek(int fd, off_t *offset, int whence)
97 *offset = lseek(fd, *offset, whence);
98 int ret = -E_OSL_LSEEK;
104 static int append_file(const char *filename, char *data, size_t data_size,
109 // DEBUG_LOG("appending %zu + %zu bytes\n", header_size, data_size);
110 ret = osl_open(filename, O_WRONLY | O_CREAT | O_APPEND, 0644);
114 ret = write_all(fd, data, &data_size);
119 ret = __lseek(fd, &offset, SEEK_END);
122 // DEBUG_LOG("new file size: " FMT_OFF_T "\n", offset);
131 static int verify_name(const char *name)
134 return -E_OSL_BAD_NAME;
136 return -E_OSL_BAD_NAME;
137 if (strchr(name, '/'))
138 return -E_OSL_BAD_NAME;
139 if (!strcmp(name, ".."))
140 return -E_OSL_BAD_NAME;
141 if (!strcmp(name, "."))
142 return -E_OSL_BAD_NAME;
146 static char *disk_storage_dirname(const struct osl_table *t, unsigned col_num,
149 char *dirname, *column_name = column_filename(t, col_num);
153 if (!(t->desc->flags & OSL_LARGE_TABLE))
155 dirname = make_message("%s/%.2s", column_name, ds_name);
160 static char *disk_storage_name_of_object(const struct osl_table *t,
161 const struct osl_object *obj)
163 HASH_TYPE hash[HASH_SIZE];
164 hash_object(obj, hash);
165 return disk_storage_name_of_hash(t, hash);
168 static int disk_storage_name_of_row(const struct osl_table *t,
169 const struct osl_row *row, char **name)
171 struct osl_object obj;
172 int ret = osl_get_object(t, row, t->disk_storage_name_column, &obj);
176 *name = disk_storage_name_of_object(t, &obj);
182 static void column_name_hash(const char *col_name, HASH_TYPE *hash)
184 hash_function(col_name, strlen(col_name), hash);
187 static int init_column_descriptions(struct osl_table *t)
190 const struct osl_column_description *cd;
192 ret = verify_name(t->desc->name);
195 ret = -E_OSL_BAD_DB_DIR;
196 if (!t->desc->dir && (t->num_disk_storage_columns || t->num_mapped_columns))
198 /* the size of the index header without column descriptions */
199 t->index_header_size = IDX_COLUMN_DESCRIPTIONS;
200 FOR_EACH_COLUMN(i, t->desc, cd) {
201 struct osl_column *col = t->columns + i;
202 if (cd->storage_flags & OSL_RBTREE) {
203 if (!cd->compare_function)
204 return -E_OSL_NO_COMPARE_FUNC;
206 if (cd->storage_type == OSL_NO_STORAGE)
208 ret = -E_OSL_NO_COLUMN_NAME;
209 if (!cd->name || !cd->name[0])
211 ret = verify_name(cd->name);
214 t->index_header_size += index_column_description_size(cd->name);
215 column_name_hash(cd->name, col->name_hash);
216 ret = -E_OSL_DUPLICATE_COL_NAME;
217 for (j = i + 1; j < t->desc->num_columns; j++) {
218 const char *name2 = get_column_description(t->desc,
220 if (cd->name && name2 && !strcmp(cd->name, name2))
230 * Initialize a struct table from given table description.
232 * \param desc The description of the osl table.
233 * \param table_ptr Result is returned here.
235 * This function performs several sanity checks on \p desc and returns if any
236 * of these tests fail. On success, a struct \p osl_table is allocated and
237 * initialized with data derived from \p desc.
241 * \sa struct osl_table.
243 int init_table_structure(const struct osl_table_description *desc,
244 struct osl_table **table_ptr)
246 const struct osl_column_description *cd;
247 struct osl_table *t = calloc(1, sizeof(*t));
248 int i, ret = -E_OSL_NOMEM, have_disk_storage_name_column = 0;
252 ret = -E_OSL_BAD_TABLE_DESC;
255 DEBUG_LOG("creating table structure for '%s' from table "
256 "description\n", desc->name);
257 ret = -E_OSL_NO_COLUMN_DESC;
258 if (!desc->column_descriptions)
260 ret = -E_OSL_NO_COLUMNS;
261 if (!desc->num_columns)
264 t->columns = calloc(desc->num_columns, sizeof(struct osl_column));
268 FOR_EACH_COLUMN(i, t->desc, cd) {
269 enum osl_storage_type st = cd->storage_type;
270 enum osl_storage_flags sf = cd->storage_flags;
271 struct osl_column *col = &t->columns[i];
273 ret = -E_OSL_BAD_STORAGE_TYPE;
274 if (st != OSL_MAPPED_STORAGE && st != OSL_DISK_STORAGE
275 && st != OSL_NO_STORAGE)
277 ret = -E_OSL_BAD_STORAGE_FLAGS;
278 if (st == OSL_DISK_STORAGE && sf & OSL_RBTREE)
280 ret = -E_OSL_BAD_STORAGE_SIZE;
281 if (sf & OSL_FIXED_SIZE && !cd->data_size)
284 case OSL_DISK_STORAGE:
285 t->num_disk_storage_columns++;
287 case OSL_MAPPED_STORAGE:
288 t->num_mapped_columns++;
289 col->index_offset = t->row_index_size;
290 t->row_index_size += 8;
293 col->volatile_num = t->num_volatile_columns;
294 t->num_volatile_columns++;
297 if (sf & OSL_RBTREE) {
298 col->rbtree_num = t->num_rbtrees;
300 if ((sf & OSL_UNIQUE) && (st == OSL_MAPPED_STORAGE)) {
301 if (!have_disk_storage_name_column)
302 t->disk_storage_name_column = i;
303 have_disk_storage_name_column = 1;
307 ret = -E_OSL_NO_UNIQUE_RBTREE_COLUMN;
308 if (t->num_disk_storage_columns && !have_disk_storage_name_column)
310 ret = -E_OSL_NO_RBTREE_COL;
314 DEBUG_LOG("OK. Index entry size: %u\n", t->row_index_size);
315 ret = init_column_descriptions(t);
327 * Read the table description from index header.
329 * \param map The memory mapping of the index file.
330 * \param desc The values found in the index header are returned here.
332 * Read the index header, check for the osl magic string and the table version
333 * number. Read all information stored in the index header into \a desc.
337 * \sa struct osl_table_description, osl_create_table.
339 int read_table_desc(struct osl_object *map, struct osl_table_description *desc)
341 char *buf = map->data;
342 uint8_t version, compat_version, create_version;
343 uint16_t header_size;
346 struct osl_column_description *cd;
348 if (map->size < MIN_INDEX_HEADER_SIZE(1))
349 return -E_OSL_SHORT_TABLE;
350 if (strncmp(buf + IDX_OSL_MAGIC, OSL_MAGIC, strlen(OSL_MAGIC)))
351 return -E_OSL_NO_MAGIC;
352 version = read_u8(buf + IDX_VERSION);
354 * The on-disk version consists of two version numbers: the
355 * create_version (low 4 bits) is the CURRENT_TABLE_VERSION version
356 * number of the library that created the table, and compat_version
357 * (high 4 bits) tells us the lowest version of the library that can
358 * still read this table.
360 create_version = version & 0xf;
361 compat_version = version >> 4;
362 INFO_LOG("create_version: %u, compat_version: %u\n", create_version,
364 if (create_version < MIN_TABLE_VERSION /* table too old */
365 || compat_version > CURRENT_TABLE_VERSION) /* libosl too old */
366 return -E_OSL_VERSION_MISMATCH;
367 desc->flags = read_u8(buf + IDX_TABLE_FLAGS);
368 desc->num_columns = read_u16(buf + IDX_NUM_COLUMNS);
369 INFO_LOG("%u columns\n", desc->num_columns);
370 if (!desc->num_columns)
371 return -E_OSL_NO_COLUMNS;
372 header_size = read_u16(buf + IDX_HEADER_SIZE);
373 if (map->size < header_size)
374 return -E_OSL_BAD_SIZE;
375 desc->column_descriptions = calloc(desc->num_columns,
376 sizeof(struct osl_column_description));
377 if (!desc->column_descriptions)
379 offset = IDX_COLUMN_DESCRIPTIONS;
380 FOR_EACH_COLUMN(i, desc, cd) {
383 ret = -E_OSL_SHORT_TABLE;
384 if (map->size < offset + MIN_IDX_COLUMN_DESCRIPTION_SIZE) {
385 ERROR_LOG("map size = %zu < %u = offset + min desc size\n",
386 map->size, offset + MIN_IDX_COLUMN_DESCRIPTION_SIZE);
389 cd->storage_type = read_u16(buf + offset + IDX_CD_STORAGE_TYPE);
390 cd->storage_flags = read_u16(buf + offset +
391 IDX_CD_STORAGE_FLAGS);
392 cd->data_size = read_u32(buf + offset + IDX_CD_DATA_SIZE);
393 null_byte = memchr(buf + offset + IDX_CD_NAME, '\0',
394 map->size - offset - IDX_CD_NAME);
395 ret = -E_OSL_INDEX_CORRUPTION;
399 cd->name = strdup(buf + offset + IDX_CD_NAME);
402 offset += index_column_description_size(cd->name);
404 if (offset != header_size) {
405 ret = -E_OSL_INDEX_CORRUPTION;
406 ERROR_LOG("real header size = %u != %u = stored header size\n",
407 offset, header_size);
412 FOR_EACH_COLUMN(i, desc, cd)
418 * check whether the table description given by \p t->desc matches the on-disk
419 * table structure stored in the index of \a t.
421 static int compare_table_descriptions(struct osl_table *t)
424 struct osl_table_description desc;
425 const struct osl_column_description *cd1, *cd2;
427 /* read the on-disk structure into desc */
428 ret = read_table_desc(&t->index_map, &desc);
431 ret = -E_OSL_BAD_TABLE_FLAGS;
432 if (desc.flags != t->desc->flags)
434 ret = -E_OSL_BAD_COLUMN_NUM;
435 if (desc.num_columns > t->desc->num_columns)
437 if (desc.num_columns < t->desc->num_columns) {
438 struct osl_column_description *cd;
439 unsigned diff = t->desc->num_columns - desc.num_columns;
440 INFO_LOG("extending table by %u volatile columns\n", diff);
442 desc.column_descriptions = realloc(desc.column_descriptions,
443 t->desc->num_columns * sizeof(struct osl_column_description));
444 if (!desc.column_descriptions)
446 for (i = desc.num_columns; i < t->desc->num_columns; i++) {
447 cd = get_column_description(&desc, i);
448 cd->storage_type = OSL_NO_STORAGE;
451 desc.num_columns += diff;
453 FOR_EACH_COLUMN(i, t->desc, cd1) {
454 cd2 = get_column_description(&desc, i);
455 ret = -E_OSL_BAD_STORAGE_TYPE;
456 if (cd1->storage_type != cd2->storage_type)
458 if (cd1->storage_type == OSL_NO_STORAGE)
460 ret = -E_OSL_BAD_STORAGE_FLAGS;
461 if (cd1->storage_flags != cd2->storage_flags) {
462 ERROR_LOG("sf1 = %u != %u = sf2\n",
463 cd1->storage_flags, cd2->storage_flags);
466 ret = -E_OSL_BAD_DATA_SIZE;
467 if (cd1->storage_flags & OSL_FIXED_SIZE)
468 if (cd1->data_size != cd2->data_size)
470 ret = -E_OSL_BAD_COLUMN_NAME;
471 if (strcmp(cd1->name, cd2->name))
474 INFO_LOG("table description of '%s' matches on-disk data, good\n",
478 FOR_EACH_COLUMN(i, &desc, cd1)
480 free(desc.column_descriptions);
484 static int create_table_index(struct osl_table *t)
486 char *buf, *filename;
488 size_t size = t->index_header_size;
489 const struct osl_column_description *cd;
492 INFO_LOG("creating %zu byte index for table %s\n", size,
494 buf = calloc(1, size);
497 sprintf(buf + IDX_OSL_MAGIC, "%s", OSL_MAGIC);
498 write_u8(buf + IDX_TABLE_FLAGS, t->desc->flags);
499 write_u8(buf + IDX_DIRTY_FLAG, 0);
500 write_u8(buf + IDX_VERSION, CURRENT_TABLE_VERSION
501 + (COMPAT_TABLE_VERSION << 4));
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;
1075 const struct osl_column_description *cd;
1077 NOTICE_LOG("opening table %s\n", table_desc->name);
1078 ret = init_table_structure(table_desc, &t);
1081 FOR_EACH_DISK_STORAGE_COLUMN(i, t, cd) {
1082 struct stat statbuf;
1083 char *dirname = column_filename(t, i);
1088 /* check if directory exists */
1089 ret = osl_stat(dirname, &statbuf);
1093 ret = -E_OSL_NOTDIR;
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 -E_OSL_NOMEM;
1126 ret = osl_mkdir(dirname, 0777);
1128 if (ret < 0 && ret != -E_OSL_DIR_EXISTS)
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 -E_OSL_NOMEM;
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);
1157 return -E_OSL_NOMEM;
1158 ret = append_file(filename, obj->data, obj->size, new_size);
1163 static int append_row_index(const struct osl_table *t, char *row_index)
1168 if (!t->num_mapped_columns)
1170 filename = index_filename(t->desc);
1172 return -E_OSL_NOMEM;
1173 ret = append_file(filename, row_index, t->row_index_size, NULL);
1178 static int truncate_mapped_file(const struct osl_table *t, unsigned col_num,
1182 char *filename = column_filename(t, col_num);
1185 return -E_OSL_NOMEM;
1186 ret = truncate_file(filename, size);
1191 static int delete_disk_storage_file(const struct osl_table *t, unsigned col_num,
1192 const char *ds_name)
1194 char *dirname, *filename = disk_storage_path(t, col_num, ds_name);
1198 return -E_OSL_NOMEM;
1199 if (unlink(filename) < 0)
1200 ret = errno == ENOENT? -E_OSL_NOENT : -E_OSL_UNLINK;
1204 if (!(t->desc->flags & OSL_LARGE_TABLE))
1206 dirname = disk_storage_dirname(t, col_num, ds_name);
1208 return -E_OSL_NOMEM;
1214 __export int osl_add_and_get_row(struct osl_table *t, struct osl_object *objects,
1215 struct osl_row **row)
1218 char *ds_name = NULL;
1219 struct rb_node **rb_parents = NULL, ***rb_links = NULL;
1220 char *new_row_index = NULL;
1221 struct osl_object *volatile_objs = NULL;
1222 const struct osl_column_description *cd;
1225 return -E_OSL_BAD_TABLE;
1226 rb_parents = malloc(t->num_rbtrees * sizeof(struct rn_node*));
1228 return -E_OSL_NOMEM;
1229 rb_links = malloc(t->num_rbtrees * sizeof(struct rn_node**));
1232 return -E_OSL_NOMEM;
1234 if (t->num_mapped_columns) {
1235 new_row_index = malloc(t->row_index_size);
1236 if (!new_row_index) {
1239 return -E_OSL_NOMEM;
1242 /* pass 1: sanity checks */
1243 // DEBUG_LOG("sanity tests: %p:%p\n", objects[0].data,
1244 // objects[1].data);
1245 FOR_EACH_COLUMN(i, t->desc, cd) {
1246 enum osl_storage_type st = cd->storage_type;
1247 enum osl_storage_flags sf = cd->storage_flags;
1249 // ret = -E_OSL_NULL_OBJECT;
1252 if (st == OSL_DISK_STORAGE)
1254 if (sf & OSL_RBTREE) {
1255 unsigned rbtree_num = t->columns[i].rbtree_num;
1256 ret = -E_OSL_RB_KEY_EXISTS;
1257 // DEBUG_LOG("checking whether %p exists\n",
1258 // objects[i].data);
1259 if (search_rbtree(objects + i, t, i,
1260 &rb_parents[rbtree_num],
1261 &rb_links[rbtree_num]) > 0)
1264 if (sf & OSL_FIXED_SIZE) {
1265 // DEBUG_LOG("fixed size. need: %zu, have: %d\n",
1266 // objects[i].size, cd->data_size);
1267 ret = -E_OSL_BAD_DATA_SIZE;
1268 if (objects[i].size != cd->data_size)
1272 if (t->num_disk_storage_columns) {
1273 ds_name = disk_storage_name_of_object(t,
1274 &objects[t->disk_storage_name_column]);
1279 ret = unmap_table(t, OSL_MARK_CLEAN);
1282 // DEBUG_LOG("sanity tests passed%s\n", "");
1283 /* pass 2: create data files, append map data */
1284 FOR_EACH_COLUMN(i, t->desc, cd) {
1285 enum osl_storage_type st = cd->storage_type;
1286 if (st == OSL_NO_STORAGE)
1288 if (st == OSL_MAPPED_STORAGE) {
1290 struct osl_column *col = &t->columns[i];
1291 // DEBUG_LOG("appending object of size %zu\n",
1292 // objects[i].size);
1293 ret = append_map_file(t, i, objects + i, &new_size);
1296 update_cell_index(new_row_index, col, new_size,
1301 ret = write_disk_storage_file(t, i, objects + i, ds_name);
1305 ret = append_row_index(t, new_row_index);
1308 ret = map_table(t, MAP_TBL_FL_VERIFY_INDEX);
1309 if (ret < 0) { /* truncate index and rollback changes */
1310 char *filename = index_filename(t->desc);
1312 truncate_file(filename, t->row_index_size);
1316 /* pass 3: add entry to rbtrees */
1317 if (t->num_volatile_columns) {
1319 volatile_objs = calloc(t->num_volatile_columns,
1320 sizeof(struct osl_object));
1323 FOR_EACH_VOLATILE_COLUMN(i, t, cd)
1324 volatile_objs[t->columns[i].volatile_num] = objects[i];
1327 // DEBUG_LOG("adding new entry as row #%d\n", t->num_rows - 1);
1328 ret = add_row_to_rbtrees(t, t->num_rows - 1, volatile_objs, row);
1331 // DEBUG_LOG("added new entry as row #%d\n", t->num_rows - 1);
1334 rollback: /* rollback all changes made, ignore further errors */
1335 for (i--; i >= 0; i--) {
1336 cd = get_column_description(t->desc, i);
1337 enum osl_storage_type st = cd->storage_type;
1338 if (st == OSL_NO_STORAGE)
1341 if (st == OSL_MAPPED_STORAGE)
1342 truncate_mapped_file(t, i, objects[i].size);
1343 else /* disk storage */
1344 delete_disk_storage_file(t, i, ds_name);
1346 /* ignore error and return previous error value */
1347 map_table(t, MAP_TBL_FL_VERIFY_INDEX);
1349 free(new_row_index);
1356 __export int osl_add_row(struct osl_table *t, struct osl_object *objects)
1358 return osl_add_and_get_row(t, objects, NULL);
1361 __export int osl_get_object(const struct osl_table *t, const struct osl_row *r,
1362 unsigned col_num, struct osl_object *object)
1364 const struct osl_column_description *cd;
1367 return -E_OSL_BAD_TABLE;
1368 cd = get_column_description(t->desc, col_num);
1369 /* col must not be disk storage */
1370 if (cd->storage_type == OSL_DISK_STORAGE)
1371 return -E_OSL_BAD_STORAGE_TYPE;
1372 if (cd->storage_type == OSL_MAPPED_STORAGE)
1373 return get_mapped_object(t, col_num, r->num, object);
1375 *object = r->volatile_objects[t->columns[col_num].volatile_num];
1379 __export int osl_del_row(struct osl_table *t, struct osl_row *row)
1381 struct osl_row *r = row;
1383 const struct osl_column_description *cd;
1386 return -E_OSL_BAD_TABLE;
1387 INFO_LOG("deleting row %p\n", row);
1389 if (t->num_disk_storage_columns) {
1391 ret = disk_storage_name_of_row(t, r, &ds_name);
1394 FOR_EACH_DISK_STORAGE_COLUMN(i, t, cd)
1395 delete_disk_storage_file(t, i, ds_name);
1398 FOR_EACH_COLUMN(i, t->desc, cd) {
1399 struct osl_column *col = t->columns + i;
1400 enum osl_storage_type st = cd->storage_type;
1401 remove_rb_node(t, i, r);
1402 if (st == OSL_MAPPED_STORAGE)
1404 if (st == OSL_NO_STORAGE && !(cd->storage_flags & OSL_DONT_FREE))
1405 free(r->volatile_objects[col->volatile_num].data);
1407 if (t->num_mapped_columns) {
1408 ret = mark_row_invalid(t, r->num);
1411 t->num_invalid_rows++;
1416 free(r->volatile_objects);
1421 /* test if column has an rbtree */
1422 static int check_rbtree_col(const struct osl_table *t, unsigned col_num,
1423 struct osl_column **col)
1426 return -E_OSL_BAD_TABLE;
1427 if (!(get_column_description(t->desc, col_num)->storage_flags & OSL_RBTREE))
1428 return -E_OSL_BAD_STORAGE_FLAGS;
1429 *col = t->columns + col_num;
1433 __export int osl_get_row(const struct osl_table *t, unsigned col_num,
1434 const struct osl_object *obj, struct osl_row **result)
1437 struct rb_node *node;
1438 struct osl_row *row;
1439 struct osl_column *col;
1442 ret = check_rbtree_col(t, col_num, &col);
1445 ret = search_rbtree(obj, t, col_num, &node, NULL);
1448 row = get_row_pointer(node, t->columns[col_num].rbtree_num);
1453 static int rbtree_loop(struct osl_column *col, void *private_data,
1454 osl_rbtree_loop_func *func)
1456 struct rb_node *n, *tmp;
1458 /* this for-loop is safe against removal of an entry */
1459 for (n = rb_first(&col->rbtree), tmp = n? rb_next(n) : NULL;
1461 n = tmp, tmp = tmp? rb_next(tmp) : NULL) {
1462 struct osl_row *r = get_row_pointer(n, col->rbtree_num);
1463 if (func(r, private_data) < 0)
1469 static int rbtree_loop_reverse(struct osl_column *col, void *private_data,
1470 osl_rbtree_loop_func *func)
1472 struct rb_node *n, *tmp;
1474 /* safe against removal of an entry */
1475 for (n = rb_last(&col->rbtree), tmp = n? rb_prev(n) : NULL;
1477 n = tmp, tmp = tmp? rb_prev(tmp) : NULL) {
1478 struct osl_row *r = get_row_pointer(n, col->rbtree_num);
1479 if (func(r, private_data) < 0)
1485 __export int osl_rbtree_loop(const struct osl_table *t, unsigned col_num,
1486 void *private_data, osl_rbtree_loop_func *func)
1488 struct osl_column *col;
1490 int ret = check_rbtree_col(t, col_num, &col);
1493 return rbtree_loop(col, private_data, func);
1496 __export int osl_rbtree_loop_reverse(const struct osl_table *t, unsigned col_num,
1497 void *private_data, osl_rbtree_loop_func *func)
1499 struct osl_column *col;
1501 int ret = check_rbtree_col(t, col_num, &col);
1504 return rbtree_loop_reverse(col, private_data, func);
1507 /* TODO: Rollback changes on errors */
1508 static int rename_disk_storage_objects(struct osl_table *t,
1509 struct osl_object *old_obj, struct osl_object *new_obj)
1512 const struct osl_column_description *cd;
1513 char *old_ds_name, *new_ds_name;
1515 if (!t->num_disk_storage_columns)
1516 return 1; /* nothing to do */
1517 if (old_obj->size == new_obj->size && !memcmp(new_obj->data,
1518 old_obj->data, new_obj->size))
1519 return 1; /* object did not change */
1520 old_ds_name = disk_storage_name_of_object(t, old_obj);
1521 new_ds_name = disk_storage_name_of_object(t, new_obj);
1523 if (!old_ds_name || ! new_ds_name)
1526 FOR_EACH_DISK_STORAGE_COLUMN(i, t, cd) {
1527 char *old_filename, *new_filename;
1528 ret = create_disk_storage_object_dir(t, i, new_ds_name);
1531 old_filename = disk_storage_path(t, i, old_ds_name);
1532 new_filename = disk_storage_path(t, i, new_ds_name);
1533 if (!old_filename || !new_filename)
1536 ret = osl_rename(old_filename, new_filename);
1550 __export int osl_update_object(struct osl_table *t, const struct osl_row *r,
1551 unsigned col_num, struct osl_object *obj)
1553 struct osl_column *col;
1554 const struct osl_column_description *cd;
1558 return -E_OSL_BAD_TABLE;
1559 col = &t->columns[col_num];
1560 cd = get_column_description(t->desc, col_num);
1561 DEBUG_LOG("updating column %u of %s\n", col_num, t->desc->name);
1562 if (cd->storage_flags & OSL_RBTREE) {
1563 if (search_rbtree(obj, t, col_num, NULL, NULL) > 0)
1564 return -E_OSL_RB_KEY_EXISTS;
1566 if (cd->storage_flags & OSL_FIXED_SIZE) {
1567 if (obj->size != cd->data_size)
1568 return -E_OSL_BAD_DATA_SIZE;
1570 remove_rb_node(t, col_num, r);
1571 if (cd->storage_type == OSL_NO_STORAGE) { /* TODO: If fixed size, reuse object? */
1572 free(r->volatile_objects[col->volatile_num].data);
1573 r->volatile_objects[col->volatile_num] = *obj;
1574 } else if (cd->storage_type == OSL_DISK_STORAGE) {
1576 ret = disk_storage_name_of_row(t, r, &ds_name);
1579 ret = delete_disk_storage_file(t, col_num, ds_name);
1580 if (ret < 0 && ret != -E_OSL_NOENT) {
1584 ret = write_disk_storage_file(t, col_num, obj, ds_name);
1588 } else { /* mapped storage */
1589 struct osl_object old_obj;
1590 ret = get_mapped_object(t, col_num, r->num, &old_obj);
1594 * If the updated column is the disk storage name column, the
1595 * disk storage name changes, so we have to rename all disk
1596 * storage objects accordingly.
1598 if (col_num == t->disk_storage_name_column) {
1599 ret = rename_disk_storage_objects(t, &old_obj, obj);
1603 if (cd->storage_flags & OSL_FIXED_SIZE)
1604 memcpy(old_obj.data, obj->data, cd->data_size);
1605 else { /* TODO: if the size doesn't change, use old space */
1606 uint32_t new_data_map_size;
1608 ret = get_row_index(t, r->num, &row_index);
1611 unmap_column(t, col_num);
1612 ret = append_map_file(t, col_num, obj,
1613 &new_data_map_size);
1616 ret = map_column(t, col_num);
1619 update_cell_index(row_index, col, new_data_map_size,
1623 if (cd->storage_flags & OSL_RBTREE) {
1624 ret = insert_rbtree(t, col_num, r, obj);
1631 __export int osl_open_disk_object(const struct osl_table *t, const struct osl_row *r,
1632 unsigned col_num, struct osl_object *obj)
1634 const struct osl_column_description *cd;
1635 char *ds_name, *filename;
1639 return -E_OSL_BAD_TABLE;
1640 cd = get_column_description(t->desc, col_num);
1641 if (cd->storage_type != OSL_DISK_STORAGE)
1642 return -E_OSL_BAD_STORAGE_TYPE;
1644 ret = disk_storage_name_of_row(t, r, &ds_name);
1647 filename = disk_storage_path(t, col_num, ds_name);
1650 return -E_OSL_NOMEM;
1651 DEBUG_LOG("filename: %s\n", filename);
1652 ret = mmap_full_file(filename, O_RDONLY, &obj->data, &obj->size, NULL);
1657 __export int osl_close_disk_object(struct osl_object *obj)
1659 return osl_munmap(obj->data, obj->size);
1662 __export int osl_get_num_rows(const struct osl_table *t, unsigned *num_rows)
1665 return -E_OSL_BAD_TABLE;
1666 assert(t->num_rows >= t->num_invalid_rows);
1667 *num_rows = t->num_rows - t->num_invalid_rows;
1671 __export int osl_get_rank(const struct osl_table *t, struct osl_row *r,
1672 unsigned col_num, unsigned *rank)
1674 struct osl_object obj;
1675 struct osl_column *col;
1676 struct rb_node *node;
1677 int ret = check_rbtree_col(t, col_num, &col);
1681 ret = osl_get_object(t, r, col_num, &obj);
1684 ret = search_rbtree(&obj, t, col_num, &node, NULL);
1687 ret = rb_rank(node, rank);
1689 return -E_OSL_BAD_ROW;
1693 __export int osl_get_nth_row(const struct osl_table *t, unsigned col_num,
1694 unsigned n, struct osl_row **result)
1696 struct osl_column *col;
1697 struct rb_node *node;
1702 return -E_OSL_RB_KEY_NOT_FOUND;
1703 ret = osl_get_num_rows(t, &num_rows);
1707 return -E_OSL_RB_KEY_NOT_FOUND;
1708 ret = check_rbtree_col(t, col_num, &col);
1711 node = rb_nth(col->rbtree.rb_node, n);
1713 return -E_OSL_RB_KEY_NOT_FOUND;
1714 *result = get_row_pointer(node, col->rbtree_num);
1718 __export int osl_rbtree_first_row(const struct osl_table *t, unsigned col_num,
1719 struct osl_row **result)
1721 return osl_get_nth_row(t, col_num, 1, result);
1724 __export int osl_rbtree_last_row(const struct osl_table *t, unsigned col_num,
1725 struct osl_row **result)
1728 int ret = osl_get_num_rows(t, &num_rows);
1732 return osl_get_nth_row(t, col_num, num_rows, result);