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() */
19 * Allocate a sufficiently large string and print into it.
21 * \param fmt A usual format string.
23 * Produce output according to \p fmt. No artificial bound on the length of the
24 * resulting string is imposed.
26 * \return This function either returns a pointer to a string that must be
27 * freed by the caller or \p NULL if memory allocation failed.
31 static __must_check __printf_1_2 __malloc char *make_message(const char *fmt, ...)
35 char *p = malloc(size);
42 /* Try to print in the allocated space. */
44 n = vsnprintf(p, size, fmt, ap);
46 /* If that worked, return the string. */
47 if (n > -1 && n < size)
49 /* Else try again with more space. */
50 if (n > -1) /* glibc 2.1 */
51 size = n + 1; /* precisely what is needed */
53 size *= 2; /* twice the old size */
67 * \param fmt Usual format string.
69 * All XXX_LOG() macros use this function.
71 __printf_2_3 void __log(int ll, const char* fmt,...)
84 strftime(str, sizeof(str), "%b %d %H:%M:%S", tm);
85 fprintf(outfd, "%s ", str);
87 vfprintf(outfd, fmt, argp);
92 * A wrapper for lseek(2).
94 * \param fd The file descriptor whose offset is to be to repositioned.
95 * \param offset A value-result parameter.
96 * \param whence Usual repositioning directive.
98 * Reposition the offset of the file descriptor \a fd to the argument \a offset
99 * according to the directive \a whence. Upon successful return, \a offset
100 * contains the resulting offset location as measured in bytes from the
101 * beginning of the file.
103 * \return Positive on success. Otherwise, the function returns \p -E_OSL_LSEEK.
107 static int __lseek(int fd, off_t *offset, int whence)
109 *offset = lseek(fd, *offset, whence);
110 int ret = -E_OSL_LSEEK;
117 * Wrapper for the write system call.
119 * \param fd The file descriptor to write to.
120 * \param buf The buffer to write.
121 * \param size The length of \a buf in bytes.
123 * This function writes out the given buffer and retries if an interrupt
124 * occurred during the write.
126 * \return On success, the number of bytes written is returned, otherwise, the
127 * function returns \p -E_OSL_WRITE.
131 static ssize_t __write(int fd, const void *buf, size_t size)
136 ret = write(fd, buf, size);
137 if ((ret < 0) && (errno == EAGAIN || errno == EINTR))
139 return ret >= 0? ret : -E_OSL_WRITE;
144 * Write the whole buffer to a file descriptor.
146 * \param fd The file descriptor to write to.
147 * \param buf The buffer to write.
148 * \param size The length of \a buf in bytes.
150 * This function writes the given buffer and continues on short writes and
151 * when interrupted by a signal.
153 * \return Positive on success, negative on errors. Possible errors: any
154 * errors returned by para_write().
158 static ssize_t write_all(int fd, const void *buf, size_t size)
160 // DEBUG_LOG("writing %zu bytes\n", size);
163 ssize_t ret = __write(fd, b, size);
164 // DEBUG_LOG("ret: %zd\n", ret);
173 * Open a file, write the given buffer and close the file.
175 * \param filename Full path to the file to open.
176 * \param buf The buffer to write to the file.
177 * \param size The size of \a buf.
181 static int write_file(const char *filename, const void *buf, size_t size)
185 ret = osl_open(filename, O_WRONLY | O_CREAT | O_EXCL, 0644);
189 ret = write_all(fd, buf, size);
198 static int append_file(const char *filename, char *header, size_t header_size,
199 char *data, size_t data_size, uint32_t *new_pos)
203 // DEBUG_LOG("appending %zu + %zu bytes\n", header_size, data_size);
204 ret = osl_open(filename, O_WRONLY | O_CREAT | O_APPEND, 0644);
208 if (header && header_size) {
209 ret = write_all(fd, header, header_size);
213 ret = write_all(fd, data, data_size);
218 ret = __lseek(fd, &offset, SEEK_END);
221 // DEBUG_LOG("new file size: " FMT_OFF_T "\n", offset);
230 static int verify_name(const char *name)
233 return -E_OSL_BAD_NAME;
235 return -E_OSL_BAD_NAME;
236 if (strchr(name, '/'))
237 return -E_OSL_BAD_NAME;
238 if (!strcmp(name, ".."))
239 return -E_OSL_BAD_NAME;
240 if (!strcmp(name, "."))
241 return -E_OSL_BAD_NAME;
246 * Compare two osl objects pointing to unsigned integers of 32 bit size.
248 * \param obj1 Pointer to the first integer.
249 * \param obj2 Pointer to the second integer.
251 * \return The values required for an osl compare function.
253 * \sa osl_compare_func, osl_hash_compare().
255 int uint32_compare(const struct osl_object *obj1, const struct osl_object *obj2)
257 uint32_t d1 = read_u32((const char *)obj1->data);
258 uint32_t d2 = read_u32((const char *)obj2->data);
268 * Compare two osl objects pointing to hash values.
270 * \param obj1 Pointer to the first hash object.
271 * \param obj2 Pointer to the second hash object.
273 * \return The values required for an osl compare function.
275 * \sa osl_compare_func, uint32_compare().
277 int osl_hash_compare(const struct osl_object *obj1, const struct osl_object *obj2)
279 return hash_compare((HASH_TYPE *)obj1->data, (HASH_TYPE *)obj2->data);
282 static char *disk_storage_dirname(const struct osl_table *t, unsigned col_num,
285 char *dirname, *column_name = column_filename(t, col_num);
289 if (!(t->desc->flags & OSL_LARGE_TABLE))
291 dirname = make_message("%s/%.2s", column_name, ds_name);
296 static char *disk_storage_name_of_object(const struct osl_table *t,
297 const struct osl_object *obj)
299 HASH_TYPE hash[HASH_SIZE];
300 hash_object(obj, hash);
301 return disk_storage_name_of_hash(t, hash);
304 static int disk_storage_name_of_row(const struct osl_table *t,
305 const struct osl_row *row, char **name)
307 struct osl_object obj;
308 int ret = osl_get_object(t, row, t->disk_storage_name_column, &obj);
312 *name = disk_storage_name_of_object(t, &obj);
315 return -ERRNO_TO_ERROR(ENOMEM);
318 static void column_name_hash(const char *col_name, HASH_TYPE *hash)
320 hash_function(col_name, strlen(col_name), hash);
323 static int init_column_descriptions(struct osl_table *t)
326 const struct osl_column_description *cd;
328 ret = -E_OSL_BAD_TABLE_DESC;
329 ret = verify_name(t->desc->name);
332 ret = -E_OSL_BAD_DB_DIR;
333 if (!t->desc->dir && (t->num_disk_storage_columns || t->num_mapped_columns))
335 /* the size of the index header without column descriptions */
336 t->index_header_size = IDX_COLUMN_DESCRIPTIONS;
337 FOR_EACH_COLUMN(i, t->desc, cd) {
338 struct osl_column *col = t->columns + i;
339 if (cd->storage_flags & OSL_RBTREE) {
340 if (!cd->compare_function)
341 return -E_OSL_NO_COMPARE_FUNC;
343 if (cd->storage_type == OSL_NO_STORAGE)
345 ret = -E_OSL_NO_COLUMN_NAME;
346 if (!cd->name || !cd->name[0])
348 ret = verify_name(cd->name);
351 t->index_header_size += index_column_description_size(cd->name);
352 column_name_hash(cd->name, col->name_hash);
353 ret = -E_OSL_DUPLICATE_COL_NAME;
354 for (j = i + 1; j < t->desc->num_columns; j++) {
355 const char *name2 = get_column_description(t->desc,
357 if (cd->name && name2 && !strcmp(cd->name, name2))
367 * Initialize a struct table from given table description.
369 * \param desc The description of the osl table.
370 * \param table_ptr Result is returned here.
372 * This function performs several sanity checks on \p desc and returns if any
373 * of these tests fail. On success, a struct \p osl_table is allocated and
374 * initialized with data derived from \p desc.
378 * \sa struct osl_table.
380 int init_table_structure(const struct osl_table_description *desc,
381 struct osl_table **table_ptr)
383 const struct osl_column_description *cd;
384 struct osl_table *t = calloc(1, sizeof(*t));
385 int i, ret = -ERRNO_TO_ERROR(ENOMEM), have_disk_storage_name_column = 0;
389 ret = -E_OSL_BAD_TABLE_DESC;
392 DEBUG_LOG("creating table structure for '%s' from table "
393 "description\n", desc->name);
394 ret = -E_OSL_NO_COLUMN_DESC;
395 if (!desc->column_descriptions)
397 ret = -E_OSL_NO_COLUMNS;
398 if (!desc->num_columns)
400 ret = -ERRNO_TO_ERROR(ENOMEM);
401 t->columns = calloc(desc->num_columns, sizeof(struct osl_column));
405 FOR_EACH_COLUMN(i, t->desc, cd) {
406 enum osl_storage_type st = cd->storage_type;
407 enum osl_storage_flags sf = cd->storage_flags;
408 struct osl_column *col = &t->columns[i];
410 ret = -E_OSL_BAD_STORAGE_TYPE;
411 if (st != OSL_MAPPED_STORAGE && st != OSL_DISK_STORAGE
412 && st != OSL_NO_STORAGE)
414 ret = -E_OSL_BAD_STORAGE_FLAGS;
415 if (st == OSL_DISK_STORAGE && sf & OSL_RBTREE)
417 ret = -E_OSL_BAD_STORAGE_SIZE;
418 if (sf & OSL_FIXED_SIZE && !cd->data_size)
421 case OSL_DISK_STORAGE:
422 t->num_disk_storage_columns++;
424 case OSL_MAPPED_STORAGE:
425 t->num_mapped_columns++;
426 col->index_offset = t->row_index_size;
427 t->row_index_size += 8;
430 col->volatile_num = t->num_volatile_columns;
431 t->num_volatile_columns++;
434 if (sf & OSL_RBTREE) {
435 col->rbtree_num = t->num_rbtrees;
437 if ((sf & OSL_UNIQUE) && (st == OSL_MAPPED_STORAGE)) {
438 if (!have_disk_storage_name_column)
439 t->disk_storage_name_column = i;
440 have_disk_storage_name_column = 1;
444 ret = -E_OSL_NO_UNIQUE_RBTREE_COLUMN;
445 if (t->num_disk_storage_columns && !have_disk_storage_name_column)
447 ret = -E_OSL_NO_RBTREE_COL;
451 DEBUG_LOG("OK. Index entry size: %u\n", t->row_index_size);
452 ret = init_column_descriptions(t);
464 * Read the table description from index header.
466 * \param map The memory mapping of the index file.
467 * \param desc The values found in the index header are returned here.
469 * Read the index header, check for the paraslash magic string and the table version number.
470 * Read all information stored in the index header into \a desc.
474 * \sa struct osl_table_description, osl_create_table.
476 int read_table_desc(struct osl_object *map, struct osl_table_description *desc)
478 char *buf = map->data;
480 uint16_t header_size;
483 struct osl_column_description *cd;
485 if (map->size < MIN_INDEX_HEADER_SIZE(1))
486 return -E_OSL_SHORT_TABLE;
487 if (strncmp(buf + IDX_PARA_MAGIC, PARA_MAGIC, strlen(PARA_MAGIC)))
488 return -E_OSL_NO_MAGIC;
489 version = read_u8(buf + IDX_VERSION);
490 if (version < MIN_TABLE_VERSION || version > MAX_TABLE_VERSION)
491 return -E_OSL_VERSION_MISMATCH;
492 desc->flags = read_u8(buf + IDX_TABLE_FLAGS);
493 desc->num_columns = read_u16(buf + IDX_NUM_COLUMNS);
494 DEBUG_LOG("%u columns\n", desc->num_columns);
495 if (!desc->num_columns)
496 return -E_OSL_NO_COLUMNS;
497 header_size = read_u16(buf + IDX_HEADER_SIZE);
498 if (map->size < header_size)
499 return -E_OSL_BAD_SIZE;
500 desc->column_descriptions = calloc(desc->num_columns,
501 sizeof(struct osl_column_description));
502 if (!desc->column_descriptions)
503 return -ERRNO_TO_ERROR(ENOMEM);
504 offset = IDX_COLUMN_DESCRIPTIONS;
505 FOR_EACH_COLUMN(i, desc, cd) {
508 ret = -E_OSL_SHORT_TABLE;
509 if (map->size < offset + MIN_IDX_COLUMN_DESCRIPTION_SIZE) {
510 ERROR_LOG("map size = %zu < %u = offset + min desc size\n",
511 map->size, offset + MIN_IDX_COLUMN_DESCRIPTION_SIZE);
514 cd->storage_type = read_u16(buf + offset + IDX_CD_STORAGE_TYPE);
515 cd->storage_flags = read_u16(buf + offset +
516 IDX_CD_STORAGE_FLAGS);
517 cd->data_size = read_u32(buf + offset + IDX_CD_DATA_SIZE);
518 null_byte = memchr(buf + offset + IDX_CD_NAME, '\0',
519 map->size - offset - IDX_CD_NAME);
520 ret = -E_OSL_INDEX_CORRUPTION;
523 ret = -ERRNO_TO_ERROR(ENOMEM);
524 cd->name = strdup(buf + offset + IDX_CD_NAME);
527 offset += index_column_description_size(cd->name);
529 if (offset != header_size) {
530 ret = -E_OSL_INDEX_CORRUPTION;
531 ERROR_LOG("real header size = %u != %u = stored header size\n",
532 offset, header_size);
537 FOR_EACH_COLUMN(i, desc, cd)
543 * check whether the table description given by \p t->desc matches the on-disk
544 * table structure stored in the index of \a t.
546 static int compare_table_descriptions(struct osl_table *t)
549 struct osl_table_description desc;
550 const struct osl_column_description *cd1, *cd2;
552 /* read the on-disk structure into desc */
553 ret = read_table_desc(&t->index_map, &desc);
556 ret = -E_OSL_BAD_TABLE_FLAGS;
557 if (desc.flags != t->desc->flags)
559 ret = -E_OSL_BAD_COLUMN_NUM;
560 if (desc.num_columns > t->desc->num_columns)
562 if (desc.num_columns < t->desc->num_columns) {
563 struct osl_column_description *cd;
564 unsigned diff = t->desc->num_columns - desc.num_columns;
565 INFO_LOG("extending table by %u volatile columns\n", diff);
566 ret = -ERRNO_TO_ERROR(ENOMEM);
567 desc.column_descriptions = realloc(desc.column_descriptions,
568 t->desc->num_columns * sizeof(struct osl_column_description));
569 if (!desc.column_descriptions)
571 for (i = desc.num_columns; i < t->desc->num_columns; i++) {
572 cd = get_column_description(&desc, i);
573 cd->storage_type = OSL_NO_STORAGE;
576 desc.num_columns += diff;
578 FOR_EACH_COLUMN(i, t->desc, cd1) {
579 cd2 = get_column_description(&desc, i);
580 ret = -E_OSL_BAD_STORAGE_TYPE;
581 if (cd1->storage_type != cd2->storage_type)
583 if (cd1->storage_type == OSL_NO_STORAGE)
585 ret = -E_OSL_BAD_STORAGE_FLAGS;
586 if (cd1->storage_flags != cd2->storage_flags) {
587 ERROR_LOG("sf1 = %u != %u = sf2\n",
588 cd1->storage_flags, cd2->storage_flags);
591 ret = -E_OSL_BAD_DATA_SIZE;
592 if (cd1->storage_flags & OSL_FIXED_SIZE)
593 if (cd1->data_size != cd2->data_size)
595 ret = -E_OSL_BAD_COLUMN_NAME;
596 if (strcmp(cd1->name, cd2->name))
599 DEBUG_LOG("table description of '%s' matches on-disk data, good\n",
603 FOR_EACH_COLUMN(i, &desc, cd1)
605 free(desc.column_descriptions);
609 static int create_table_index(struct osl_table *t)
611 char *buf, *filename;
613 size_t size = t->index_header_size;
614 const struct osl_column_description *cd;
617 INFO_LOG("creating %zu byte index for table %s\n", size,
619 buf = calloc(1, size);
621 return -ERRNO_TO_ERROR(ENOMEM);
622 sprintf(buf + IDX_PARA_MAGIC, "%s", PARA_MAGIC);
623 write_u8(buf + IDX_TABLE_FLAGS, t->desc->flags);
624 write_u8(buf + IDX_DIRTY_FLAG, 0);
625 write_u8(buf + IDX_VERSION, CURRENT_TABLE_VERSION);
626 write_u16(buf + IDX_NUM_COLUMNS, t->num_mapped_columns + t->num_disk_storage_columns);
627 write_u16(buf + IDX_HEADER_SIZE, t->index_header_size);
628 offset = IDX_COLUMN_DESCRIPTIONS;
629 FOR_EACH_COLUMN(i, t->desc, cd) {
630 /* no need to store info about volatile storage */
631 if (cd->storage_type == OSL_NO_STORAGE)
633 write_u16(buf + offset + IDX_CD_STORAGE_TYPE,
635 write_u16(buf + offset + IDX_CD_STORAGE_FLAGS,
637 if (cd->storage_flags & OSL_FIXED_SIZE)
638 write_u32(buf + offset + IDX_CD_DATA_SIZE,
640 strcpy(buf + offset + IDX_CD_NAME, cd->name);
641 offset += index_column_description_size(cd->name);
643 assert(offset = size);
644 filename = index_filename(t->desc);
646 ret = write_file(filename, buf, size);
648 ret = -ERRNO_TO_ERROR(ENOMEM);
655 * Create a new osl table.
657 * \param desc Pointer to the table description.
661 __export int osl_create_table(const struct osl_table_description *desc)
663 const struct osl_column_description *cd;
664 char *table_dir = NULL, *filename;
666 int i, ret = init_table_structure(desc, &t);
670 INFO_LOG("creating %s\n", desc->name);
671 FOR_EACH_COLUMN(i, t->desc, cd) {
672 if (cd->storage_type == OSL_NO_STORAGE)
675 ret = para_mkdir(desc->dir, 0777);
676 if (ret < 0 && !is_errno(-ret, EEXIST))
678 table_dir = make_message("%s/%s", desc->dir,
680 ret = -ERRNO_TO_ERROR(ENOMEM);
683 ret = para_mkdir(table_dir, 0777);
687 ret = -ERRNO_TO_ERROR(ENOMEM);
688 filename = column_filename(t, i);
691 INFO_LOG("filename: %s\n", filename);
692 if (cd->storage_type == OSL_MAPPED_STORAGE) {
693 ret = osl_open(filename, O_RDWR | O_CREAT | O_EXCL,
702 ret = para_mkdir(filename, 0777);
707 if (t->num_mapped_columns) {
708 ret = create_table_index(t);
720 static int table_is_dirty(struct osl_table *t)
722 char *buf = (char *)t->index_map.data + IDX_DIRTY_FLAG;
723 uint8_t dirty = read_u8(buf) & 0x1;
727 static void mark_table_dirty(struct osl_table *t)
729 char *buf = (char *)t->index_map.data + IDX_DIRTY_FLAG;
730 write_u8(buf, read_u8(buf) | 1);
733 static void mark_table_clean(struct osl_table *t)
735 char *buf = (char *)t->index_map.data + IDX_DIRTY_FLAG;
736 write_u8(buf, read_u8(buf) & 0xfe);
739 static void unmap_column(struct osl_table *t, unsigned col_num)
741 struct osl_object map = t->columns[col_num].data_map;
745 ret = para_munmap(map.data, map.size);
751 * Unmap all mapped files of an osl table.
753 * \param t Pointer to a mapped table.
754 * \param flags Options for unmapping.
756 * \return Positive on success, negative on errors.
758 * \sa map_table(), enum osl_close_flags, para_munmap().
760 int unmap_table(struct osl_table *t, enum osl_close_flags flags)
763 const struct osl_column_description *cd;
766 if (!t->num_mapped_columns) /* can this ever happen? */
768 DEBUG_LOG("unmapping table '%s'\n", t->desc->name);
769 if (!t->index_map.data)
770 return -E_OSL_NOT_MAPPED;
771 if (flags & OSL_MARK_CLEAN)
773 ret = para_munmap(t->index_map.data, t->index_map.size);
776 t->index_map.data = NULL;
779 FOR_EACH_MAPPED_COLUMN(i, t, cd)
784 static int map_column(struct osl_table *t, unsigned col_num)
787 char *filename = column_filename(t, col_num);
788 int ret = -E_OSL_STAT;
791 return -ERRNO_TO_ERROR(ENOMEM);
792 if (stat(filename, &statbuf) < 0) {
796 if (!(S_IFREG & statbuf.st_mode)) {
800 ret = mmap_full_file(filename, O_RDWR,
801 &t->columns[col_num].data_map.data,
802 &t->columns[col_num].data_map.size,
809 * Map the index file and all columns of type \p OSL_MAPPED_STORAGE into memory.
811 * \param t Pointer to an initialized table structure.
812 * \param flags Mapping options.
814 * \return Negative return value on errors; on success the number of rows
815 * (including invalid rows) is returned.
817 * \sa unmap_table(), enum map_table_flags, osl_open_table(), mmap(2).
819 int map_table(struct osl_table *t, enum map_table_flags flags)
822 const struct osl_column_description *cd;
823 int i = 0, ret, num_rows = 0;
825 if (!t->num_mapped_columns)
827 if (t->index_map.data)
828 return -E_OSL_ALREADY_MAPPED;
829 filename = index_filename(t->desc);
831 return -ERRNO_TO_ERROR(ENOMEM);
832 DEBUG_LOG("mapping table '%s' (index: %s)\n", t->desc->name, filename);
833 ret = mmap_full_file(filename, flags & MAP_TBL_FL_MAP_RDONLY?
834 O_RDONLY : O_RDWR, &t->index_map.data, &t->index_map.size, NULL);
838 if (flags & MAP_TBL_FL_VERIFY_INDEX) {
839 ret = compare_table_descriptions(t);
844 if (!(flags & MAP_TBL_FL_IGNORE_DIRTY)) {
845 if (table_is_dirty(t)) {
846 ERROR_LOG("%s is dirty\n", t->desc->name);
851 num_rows = table_num_rows(t);
855 FOR_EACH_MAPPED_COLUMN(i, t, cd) {
856 ret = map_column(t, i);
861 err: /* unmap what is already mapped */
862 for (i--; i >= 0; i--) {
863 struct osl_object map = t->columns[i].data_map;
864 para_munmap(map.data, map.size);
867 para_munmap(t->index_map.data, t->index_map.size);
868 t->index_map.data = NULL;
873 * Retrieve a mapped object by row and column number.
875 * \param t Pointer to an open osl table.
876 * \param col_num Number of the mapped column containing the object to retrieve.
877 * \param row_num Number of the row containing the object to retrieve.
878 * \param obj The result is returned here.
880 * It is considered an error if \a col_num does not refer to a column
881 * of storage type \p OSL_MAPPED_STORAGE.
885 * \sa osl_storage_type.
887 int get_mapped_object(const struct osl_table *t, unsigned col_num,
888 uint32_t row_num, struct osl_object *obj)
890 struct osl_column *col = &t->columns[col_num];
896 if (t->num_rows <= row_num)
897 return -E_OSL_BAD_ROW_NUM;
898 ret = get_cell_index(t, row_num, col_num, &cell_index);
901 offset = read_u32(cell_index);
902 obj->size = read_u32(cell_index + 4) - 1;
903 header = col->data_map.data + offset;
904 obj->data = header + 1;
905 if (read_u8(header) == 0xff) {
906 ERROR_LOG("col %u, size %zu, offset %u\n", col_num,
908 return -E_OSL_INVALID_OBJECT;
913 static int search_rbtree(const struct osl_object *obj,
914 const struct osl_table *t, unsigned col_num,
915 struct rb_node **result, struct rb_node ***rb_link)
917 struct osl_column *col = &t->columns[col_num];
918 struct rb_node **new = &col->rbtree.rb_node, *parent = NULL;
919 const struct osl_column_description *cd =
920 get_column_description(t->desc, col_num);
921 enum osl_storage_type st = cd->storage_type;
923 struct osl_row *this_row = get_row_pointer(*new,
926 struct osl_object this_obj;
928 if (st == OSL_MAPPED_STORAGE) {
929 ret = get_mapped_object(t, col_num, this_row->num,
934 this_obj = this_row->volatile_objects[col->volatile_num];
935 ret = cd->compare_function(obj, &this_obj);
938 *result = get_rb_node_pointer(this_row,
943 new = &((*new)->rb_left);
945 new = &((*new)->rb_right);
951 return -E_OSL_RB_KEY_NOT_FOUND;
954 static int insert_rbtree(struct osl_table *t, unsigned col_num,
955 const struct osl_row *row, const struct osl_object *obj)
957 struct rb_node *parent, **rb_link;
960 int ret = search_rbtree(obj, t, col_num, &parent, &rb_link);
963 return -E_OSL_RB_KEY_EXISTS;
964 rbtree_num = t->columns[col_num].rbtree_num;
965 n = get_rb_node_pointer(row, rbtree_num);
966 rb_link_node(n, parent, rb_link);
967 rb_insert_color(n, &t->columns[col_num].rbtree);
971 static void remove_rb_node(struct osl_table *t, unsigned col_num,
972 const struct osl_row *row)
974 struct osl_column *col = &t->columns[col_num];
975 const struct osl_column_description *cd =
976 get_column_description(t->desc, col_num);
977 enum osl_storage_flags sf = cd->storage_flags;
978 struct rb_node *victim, *splice_out_node, *tmp;
979 if (!(sf & OSL_RBTREE))
982 * Which node is removed/spliced out actually depends on how many
983 * children the victim node has: If it has no children, it gets
984 * deleted. If it has one child, it gets spliced out. If it has two
985 * children, its successor (which has at most a right child) gets
988 victim = get_rb_node_pointer(row, col->rbtree_num);
989 if (victim->rb_left && victim->rb_right)
990 splice_out_node = rb_next(victim);
992 splice_out_node = victim;
993 /* Go up to the root and decrement the size of each node in the path. */
994 for (tmp = splice_out_node; tmp; tmp = rb_parent(tmp))
996 rb_erase(victim, &col->rbtree);
999 static int add_row_to_rbtrees(struct osl_table *t, uint32_t row_num,
1000 struct osl_object *volatile_objs, struct osl_row **row_ptr)
1004 struct osl_row *row = allocate_row(t->num_rbtrees);
1005 const struct osl_column_description *cd;
1008 return -ERRNO_TO_ERROR(ENOMEM);
1010 row->volatile_objects = volatile_objs;
1011 FOR_EACH_RBTREE_COLUMN(i, t, cd) {
1012 if (cd->storage_type == OSL_MAPPED_STORAGE) {
1013 struct osl_object obj;
1014 ret = get_mapped_object(t, i, row_num, &obj);
1017 ret = insert_rbtree(t, i, row, &obj);
1018 } else { /* volatile */
1019 const struct osl_object *obj
1020 = volatile_objs + t->columns[i].volatile_num;
1021 ret = insert_rbtree(t, i, row, obj);
1029 err: /* rollback changes, i.e. remove added entries from rbtrees */
1031 remove_rb_node(t, i--, row);
1036 static void free_volatile_objects(const struct osl_table *t,
1037 enum osl_close_flags flags)
1041 struct osl_column *rb_col;
1042 const struct osl_column_description *cd;
1044 if (!t->num_volatile_columns)
1046 /* find the first rbtree column (any will do) */
1047 FOR_EACH_RBTREE_COLUMN(i, t, cd)
1049 rb_col = t->columns + i;
1050 /* walk that rbtree and free all volatile objects */
1051 for (n = rb_first(&rb_col->rbtree); n; n = rb_next(n)) {
1052 struct osl_row *r = get_row_pointer(n, rb_col->rbtree_num);
1053 if (flags & OSL_FREE_VOLATILE)
1054 FOR_EACH_VOLATILE_COLUMN(j, t, cd) {
1055 if (cd->storage_flags & OSL_DONT_FREE)
1057 free(r->volatile_objects[
1058 t->columns[j].volatile_num].data);
1060 // for (j = 0; j < t->num_volatile_columns; j++)
1061 // free(r->volatile_objects[j].data);
1062 free(r->volatile_objects);
1067 * Erase all rbtree nodes and free resources.
1069 * \param t Pointer to an open osl table.
1071 * This function is called by osl_close_table().
1073 void clear_rbtrees(struct osl_table *t)
1075 const struct osl_column_description *cd;
1076 unsigned i, rbtrees_cleared = 0;
1078 FOR_EACH_RBTREE_COLUMN(i, t, cd) {
1079 struct osl_column *col = &t->columns[i];
1082 for (n = rb_first(&col->rbtree); n;) {
1084 rb_erase(n, &col->rbtree);
1085 if (rbtrees_cleared == t->num_rbtrees) {
1086 r = get_row_pointer(n, col->rbtree_num);
1097 * Close an osl table.
1099 * \param t Pointer to the table to be closed.
1100 * \param flags Options for what should be cleaned up.
1102 * If osl_open_table() succeeds, the resulting table pointer must later be
1103 * passed to this function in order to flush all changes to the file system and
1104 * to free the resources that were allocated by osl_open_table().
1108 * \sa osl_open_table(), unmap_table().
1110 __export int osl_close_table(struct osl_table *t, enum osl_close_flags flags)
1115 return -E_OSL_BAD_TABLE;
1116 free_volatile_objects(t, flags);
1118 ret = unmap_table(t, flags);
1120 ERROR_LOG("unmap_table failed: %d\n", ret);
1127 * Find out whether the given row number corresponds to an invalid row.
1129 * \param t Pointer to the osl table.
1130 * \param row_num The number of the row in question.
1132 * By definition, a row is considered invalid if all its index entries
1135 * \return Positive if \a row_num corresponds to an invalid row,
1136 * zero if it corresponds to a valid row, negative on errors.
1138 int row_is_invalid(struct osl_table *t, uint32_t row_num)
1141 int i, ret = get_row_index(t, row_num, &row_index);
1145 for (i = 0; i < t->row_index_size; i++) {
1146 if ((unsigned char)row_index[i] != 0xff)
1149 INFO_LOG("row %d is invalid\n", row_num);
1154 * Invalidate a row of an osl table.
1156 * \param t Pointer to an open osl table.
1157 * \param row_num Number of the row to mark as invalid.
1159 * This function marks each mapped object in the index entry of \a row as
1164 int mark_row_invalid(struct osl_table *t, uint32_t row_num)
1167 int ret = get_row_index(t, row_num, &row_index);
1171 INFO_LOG("marking row %d as invalid\n", row_num);
1172 memset(row_index, 0xff, t->row_index_size);
1177 * Initialize all rbtrees and compute number of invalid rows.
1179 * \param t The table containing the rbtrees to be initialized.
1183 int init_rbtrees(struct osl_table *t)
1186 const struct osl_column_description *cd;
1188 /* create rbtrees */
1189 FOR_EACH_RBTREE_COLUMN(i, t, cd)
1190 t->columns[i].rbtree = RB_ROOT;
1191 /* add valid rows to rbtrees */
1192 t->num_invalid_rows = 0;
1193 for (i = 0; i < t->num_rows; i++) {
1194 ret = row_is_invalid(t, i);
1198 t->num_invalid_rows++;
1201 ret = add_row_to_rbtrees(t, i, NULL, NULL);
1209 * Open an osl table.
1211 * Each osl table must be opened before its data can be accessed.
1213 * \param table_desc Describes the table to be opened.
1214 * \param result Contains a pointer to the open table on success.
1216 * The table description given by \a desc should coincide with the
1217 * description used at creation time.
1221 __export int osl_open_table(const struct osl_table_description *table_desc,
1222 struct osl_table **result)
1225 struct osl_table *t;
1226 const struct osl_column_description *cd;
1228 INFO_LOG("opening table %s\n", table_desc->name);
1229 ret = init_table_structure(table_desc, &t);
1232 FOR_EACH_DISK_STORAGE_COLUMN(i, t, cd) {
1233 struct stat statbuf;
1234 char *dirname = column_filename(t, i);
1236 ret = -ERRNO_TO_ERROR(ENOMEM);
1239 /* check if directory exists */
1240 ret = stat(dirname, &statbuf);
1243 ret = -ERRNO_TO_ERROR(errno);
1246 ret = -ERRNO_TO_ERROR(ENOTDIR);
1247 if (!S_ISDIR(statbuf.st_mode))
1250 ret = map_table(t, MAP_TBL_FL_VERIFY_INDEX);
1254 DEBUG_LOG("num rows: %d\n", t->num_rows);
1255 ret = init_rbtrees(t);
1257 osl_close_table(t, OSL_MARK_CLEAN); /* ignore further errors */
1268 static int create_disk_storage_object_dir(const struct osl_table *t,
1269 unsigned col_num, const char *ds_name)
1274 if (!(t->desc->flags & OSL_LARGE_TABLE))
1276 dirname = disk_storage_dirname(t, col_num, ds_name);
1278 return -ERRNO_TO_ERROR(ENOMEM);
1279 ret = para_mkdir(dirname, 0777);
1281 if (ret < 0 && !is_errno(-ret, EEXIST))
1286 static int write_disk_storage_file(const struct osl_table *t, unsigned col_num,
1287 const struct osl_object *obj, const char *ds_name)
1292 ret = create_disk_storage_object_dir(t, col_num, ds_name);
1295 filename = disk_storage_path(t, col_num, ds_name);
1297 return -ERRNO_TO_ERROR(ENOMEM);
1298 ret = write_file(filename, obj->data, obj->size);
1303 static int append_map_file(const struct osl_table *t, unsigned col_num,
1304 const struct osl_object *obj, uint32_t *new_size)
1306 char *filename = column_filename(t, col_num);
1308 char header = 0; /* zero means valid object */
1311 return -ERRNO_TO_ERROR(ENOMEM);
1312 ret = append_file(filename, &header, 1, obj->data, obj->size,
1318 static int append_row_index(const struct osl_table *t, char *row_index)
1323 if (!t->num_mapped_columns)
1325 filename = index_filename(t->desc);
1327 return -ERRNO_TO_ERROR(ENOMEM);
1328 ret = append_file(filename, NULL, 0, row_index,
1329 t->row_index_size, NULL);
1335 * A wrapper for truncate(2)
1337 * \param path Name of the regular file to truncate
1338 * \param size Number of bytes to \b shave \b off
1340 * Truncate the regular file named by \a path by \a size bytes.
1346 int para_truncate(const char *path, off_t size)
1349 struct stat statbuf;
1352 if (stat(path, &statbuf) < 0)
1354 ret = -E_OSL_BAD_SIZE;
1355 if (statbuf.st_size < size)
1358 if (truncate(path, statbuf.st_size - size) < 0)
1365 static int truncate_mapped_file(const struct osl_table *t, unsigned col_num,
1369 char *filename = column_filename(t, col_num);
1372 return -ERRNO_TO_ERROR(ENOMEM);
1373 ret = para_truncate(filename, size);
1378 static int delete_disk_storage_file(const struct osl_table *t, unsigned col_num,
1379 const char *ds_name)
1381 char *dirname, *filename = disk_storage_path(t, col_num, ds_name);
1385 return -ERRNO_TO_ERROR(ENOMEM);
1386 ret = unlink(filename);
1390 return -ERRNO_TO_ERROR(err);
1391 if (!(t->desc->flags & OSL_LARGE_TABLE))
1393 dirname = disk_storage_dirname(t, col_num, ds_name);
1395 return -ERRNO_TO_ERROR(ENOMEM);
1402 * Add a new row to an osl table and retrieve this row.
1404 * \param t Pointer to an open osl table.
1405 * \param objects Array of objects to be added.
1406 * \param row Result pointer.
1408 * The \a objects parameter must point to an array containing one object per
1409 * column. The order of the objects in the array is given by the table
1410 * description of \a table. Several sanity checks are performed during object
1411 * insertion and the function returns without modifying the table if any of
1412 * these tests fail. In fact, it is atomic in the sense that it either
1413 * succeeds or leaves the table unchanged (i.e. either all or none of the
1414 * objects are added to the table).
1416 * It is considered an error if an object is added to a column with associated
1417 * rbtree if this object is equal to an object already contained in that column
1418 * (i.e. the compare function for the column's rbtree returns zero).
1422 * \sa struct osl_table_description, osl_compare_func, osl_add_row().
1424 __export int osl_add_and_get_row(struct osl_table *t, struct osl_object *objects,
1425 struct osl_row **row)
1428 char *ds_name = NULL;
1429 struct rb_node **rb_parents = NULL, ***rb_links = NULL;
1430 char *new_row_index = NULL;
1431 struct osl_object *volatile_objs = NULL;
1432 const struct osl_column_description *cd;
1435 return -E_OSL_BAD_TABLE;
1436 rb_parents = malloc(t->num_rbtrees * sizeof(struct rn_node*));
1438 return -ERRNO_TO_ERROR(ENOMEM);
1439 rb_links = malloc(t->num_rbtrees * sizeof(struct rn_node**));
1442 return -ERRNO_TO_ERROR(ENOMEM);
1444 if (t->num_mapped_columns) {
1445 new_row_index = malloc(t->row_index_size);
1446 if (!new_row_index) {
1449 return -ERRNO_TO_ERROR(ENOMEM);
1452 /* pass 1: sanity checks */
1453 // DEBUG_LOG("sanity tests: %p:%p\n", objects[0].data,
1454 // objects[1].data);
1455 FOR_EACH_COLUMN(i, t->desc, cd) {
1456 enum osl_storage_type st = cd->storage_type;
1457 enum osl_storage_flags sf = cd->storage_flags;
1459 // ret = -E_OSL_NULL_OBJECT;
1462 if (st == OSL_DISK_STORAGE)
1464 if (sf & OSL_RBTREE) {
1465 unsigned rbtree_num = t->columns[i].rbtree_num;
1466 ret = -E_OSL_RB_KEY_EXISTS;
1467 // DEBUG_LOG("checking whether %p exists\n",
1468 // objects[i].data);
1469 if (search_rbtree(objects + i, t, i,
1470 &rb_parents[rbtree_num],
1471 &rb_links[rbtree_num]) > 0)
1474 if (sf & OSL_FIXED_SIZE) {
1475 // DEBUG_LOG("fixed size. need: %zu, have: %d\n",
1476 // objects[i].size, cd->data_size);
1477 ret = -E_OSL_BAD_DATA_SIZE;
1478 if (objects[i].size != cd->data_size)
1482 if (t->num_disk_storage_columns) {
1483 ds_name = disk_storage_name_of_object(t,
1484 &objects[t->disk_storage_name_column]);
1485 ret = -ERRNO_TO_ERROR(ENOMEM);
1489 ret = unmap_table(t, OSL_MARK_CLEAN);
1492 // DEBUG_LOG("sanity tests passed%s\n", "");
1493 /* pass 2: create data files, append map data */
1494 FOR_EACH_COLUMN(i, t->desc, cd) {
1495 enum osl_storage_type st = cd->storage_type;
1496 if (st == OSL_NO_STORAGE)
1498 if (st == OSL_MAPPED_STORAGE) {
1500 struct osl_column *col = &t->columns[i];
1501 // DEBUG_LOG("appending object of size %zu\n",
1502 // objects[i].size);
1503 ret = append_map_file(t, i, objects + i, &new_size);
1506 update_cell_index(new_row_index, col, new_size,
1511 ret = write_disk_storage_file(t, i, objects + i, ds_name);
1515 ret = append_row_index(t, new_row_index);
1518 ret = map_table(t, MAP_TBL_FL_VERIFY_INDEX);
1519 if (ret < 0) { /* truncate index and rollback changes */
1520 char *filename = index_filename(t->desc);
1522 para_truncate(filename, t->row_index_size);
1526 /* pass 3: add entry to rbtrees */
1527 if (t->num_volatile_columns) {
1528 ret = -ERRNO_TO_ERROR(ENOMEM);
1529 volatile_objs = calloc(t->num_volatile_columns,
1530 sizeof(struct osl_object));
1533 FOR_EACH_VOLATILE_COLUMN(i, t, cd)
1534 volatile_objs[t->columns[i].volatile_num] = objects[i];
1537 // DEBUG_LOG("adding new entry as row #%d\n", t->num_rows - 1);
1538 ret = add_row_to_rbtrees(t, t->num_rows - 1, volatile_objs, row);
1541 // DEBUG_LOG("added new entry as row #%d\n", t->num_rows - 1);
1544 rollback: /* rollback all changes made, ignore further errors */
1545 for (i--; i >= 0; i--) {
1546 cd = get_column_description(t->desc, i);
1547 enum osl_storage_type st = cd->storage_type;
1548 if (st == OSL_NO_STORAGE)
1551 if (st == OSL_MAPPED_STORAGE)
1552 truncate_mapped_file(t, i, objects[i].size);
1553 else /* disk storage */
1554 delete_disk_storage_file(t, i, ds_name);
1556 /* ignore error and return previous error value */
1557 map_table(t, MAP_TBL_FL_VERIFY_INDEX);
1559 free(new_row_index);
1567 * Add a new row to an osl table.
1569 * \param t Same meaning as osl_add_and_get_row().
1570 * \param objects Same meaning as osl_add_and_get_row().
1572 * \return The return value of the underlying call to osl_add_and_get_row().
1574 * This is equivalent to osl_add_and_get_row(t, objects, NULL).
1576 __export int osl_add_row(struct osl_table *t, struct osl_object *objects)
1578 return osl_add_and_get_row(t, objects, NULL);
1582 * Retrieve an object identified by row and column
1584 * \param t Pointer to an open osl table.
1585 * \param r Pointer to the row.
1586 * \param col_num The column number.
1587 * \param object The result pointer.
1589 * The column determined by \a col_num must be of type \p OSL_MAPPED_STORAGE
1590 * or \p OSL_NO_STORAGE, i.e. no disk storage objects may be retrieved by this
1595 * \sa osl_storage_type, osl_open_disk_object().
1597 __export int osl_get_object(const struct osl_table *t, const struct osl_row *r,
1598 unsigned col_num, struct osl_object *object)
1600 const struct osl_column_description *cd;
1603 return -E_OSL_BAD_TABLE;
1604 cd = get_column_description(t->desc, col_num);
1605 /* col must not be disk storage */
1606 if (cd->storage_type == OSL_DISK_STORAGE)
1607 return -E_OSL_BAD_STORAGE_TYPE;
1608 if (cd->storage_type == OSL_MAPPED_STORAGE)
1609 return get_mapped_object(t, col_num, r->num, object);
1611 *object = r->volatile_objects[t->columns[col_num].volatile_num];
1615 static int mark_mapped_object_invalid(const struct osl_table *t,
1616 uint32_t row_num, unsigned col_num)
1618 struct osl_object obj;
1620 int ret = get_mapped_object(t, col_num, row_num, &obj);
1631 * Delete a row from an osl table.
1633 * \param t Pointer to an open osl table.
1634 * \param row Pointer to the row to delete.
1636 * This removes all disk storage objects, removes all rbtree nodes, and frees
1637 * all volatile objects belonging to the given row. For mapped columns, the
1638 * data is merely marked invalid and may be pruned from time to time by
1643 __export int osl_del_row(struct osl_table *t, struct osl_row *row)
1645 struct osl_row *r = row;
1647 const struct osl_column_description *cd;
1650 return -E_OSL_BAD_TABLE;
1651 INFO_LOG("deleting row %p\n", row);
1653 if (t->num_disk_storage_columns) {
1655 ret = disk_storage_name_of_row(t, r, &ds_name);
1658 FOR_EACH_DISK_STORAGE_COLUMN(i, t, cd)
1659 delete_disk_storage_file(t, i, ds_name);
1662 FOR_EACH_COLUMN(i, t->desc, cd) {
1663 struct osl_column *col = t->columns + i;
1664 enum osl_storage_type st = cd->storage_type;
1665 remove_rb_node(t, i, r);
1666 if (st == OSL_MAPPED_STORAGE) {
1667 mark_mapped_object_invalid(t, r->num, i);
1670 if (st == OSL_NO_STORAGE && !(cd->storage_flags & OSL_DONT_FREE))
1671 free(r->volatile_objects[col->volatile_num].data);
1673 if (t->num_mapped_columns) {
1674 ret = mark_row_invalid(t, r->num);
1677 t->num_invalid_rows++;
1682 free(r->volatile_objects);
1687 /* test if column has an rbtree */
1688 static int check_rbtree_col(const struct osl_table *t, unsigned col_num,
1689 struct osl_column **col)
1692 return -E_OSL_BAD_TABLE;
1693 if (!(get_column_description(t->desc, col_num)->storage_flags & OSL_RBTREE))
1694 return -E_OSL_BAD_STORAGE_FLAGS;
1695 *col = t->columns + col_num;
1700 * Get the row that contains the given object.
1702 * \param t Pointer to an open osl table.
1703 * \param col_num The number of the column to be searched.
1704 * \param obj The object to be looked up.
1705 * \param result Points to the row containing \a obj.
1707 * Lookup \a obj in \a t and return the row containing \a obj. The column
1708 * specified by \a col_num must have an associated rbtree.
1712 * \sa osl_storage_flags
1714 __export int osl_get_row(const struct osl_table *t, unsigned col_num,
1715 const struct osl_object *obj, struct osl_row **result)
1718 struct rb_node *node;
1719 struct osl_row *row;
1720 struct osl_column *col;
1723 ret = check_rbtree_col(t, col_num, &col);
1726 ret = search_rbtree(obj, t, col_num, &node, NULL);
1729 row = get_row_pointer(node, t->columns[col_num].rbtree_num);
1734 static int rbtree_loop(struct osl_column *col, void *private_data,
1735 osl_rbtree_loop_func *func)
1737 struct rb_node *n, *tmp;
1739 /* this for-loop is safe against removal of an entry */
1740 for (n = rb_first(&col->rbtree), tmp = n? rb_next(n) : NULL;
1742 n = tmp, tmp = tmp? rb_next(tmp) : NULL) {
1743 struct osl_row *r = get_row_pointer(n, col->rbtree_num);
1744 int ret = func(r, private_data);
1751 static int rbtree_loop_reverse(struct osl_column *col, void *private_data,
1752 osl_rbtree_loop_func *func)
1754 struct rb_node *n, *tmp;
1756 /* safe against removal of an entry */
1757 for (n = rb_last(&col->rbtree), tmp = n? rb_prev(n) : NULL;
1759 n = tmp, tmp = tmp? rb_prev(tmp) : NULL) {
1760 struct osl_row *r = get_row_pointer(n, col->rbtree_num);
1761 int ret = func(r, private_data);
1769 * Loop over all nodes in an rbtree.
1771 * \param t Pointer to an open osl table.
1772 * \param col_num The column to use for iterating over the elements.
1773 * \param private_data Pointer that gets passed to \a func.
1774 * \param func The function to be called for each node in the rbtree.
1776 * This function does an in-order walk of the rbtree associated with \a
1777 * col_num. It is an error if the \p OSL_RBTREE flag is not set for this
1778 * column. For each node in the rbtree, the given function \a func is called
1779 * with two pointers as arguments: The first osl_row* argument points to the
1780 * row that contains the object corresponding to the rbtree node currently
1781 * traversed, and the \a private_data pointer is passed verbatim to \a func as the
1782 * second argument. The loop terminates either if \a func returns a negative
1783 * value, or if all nodes of the tree have been visited.
1786 * \return Standard. If the termination of the loop was caused by \a func
1787 * returning a negative value, this value is returned.
1789 * \sa osl_storage_flags, osl_rbtree_loop_reverse(), osl_compare_func.
1791 __export int osl_rbtree_loop(const struct osl_table *t, unsigned col_num,
1792 void *private_data, osl_rbtree_loop_func *func)
1794 struct osl_column *col;
1796 int ret = check_rbtree_col(t, col_num, &col);
1799 return rbtree_loop(col, private_data, func);
1803 * Loop over all nodes in an rbtree in reverse order.
1805 * \param t Identical meaning as in \p osl_rbtree_loop().
1806 * \param col_num Identical meaning as in \p osl_rbtree_loop().
1807 * \param private_data Identical meaning as in \p osl_rbtree_loop().
1808 * \param func Identical meaning as in \p osl_rbtree_loop().
1810 * This function is identical to \p osl_rbtree_loop(), the only difference
1811 * is that the tree is walked in reverse order.
1813 * \return The same return value as \p osl_rbtree_loop().
1815 * \sa osl_rbtree_loop().
1817 __export int osl_rbtree_loop_reverse(const struct osl_table *t, unsigned col_num,
1818 void *private_data, osl_rbtree_loop_func *func)
1820 struct osl_column *col;
1822 int ret = check_rbtree_col(t, col_num, &col);
1825 return rbtree_loop_reverse(col, private_data, func);
1828 /* TODO: Rollback changes on errors */
1829 static int rename_disk_storage_objects(struct osl_table *t,
1830 struct osl_object *old_obj, struct osl_object *new_obj)
1833 const struct osl_column_description *cd;
1834 char *old_ds_name, *new_ds_name;
1836 if (!t->num_disk_storage_columns)
1837 return 1; /* nothing to do */
1838 if (old_obj->size == new_obj->size && !memcmp(new_obj->data,
1839 old_obj->data, new_obj->size))
1840 return 1; /* object did not change */
1841 old_ds_name = disk_storage_name_of_object(t, old_obj);
1842 new_ds_name = disk_storage_name_of_object(t, new_obj);
1843 ret = -ERRNO_TO_ERROR(ENOMEM);
1844 if (!old_ds_name || ! new_ds_name)
1847 FOR_EACH_DISK_STORAGE_COLUMN(i, t, cd) {
1848 char *old_filename, *new_filename;
1849 ret = create_disk_storage_object_dir(t, i, new_ds_name);
1852 old_filename = disk_storage_path(t, i, old_ds_name);
1853 new_filename = disk_storage_path(t, i, new_ds_name);
1854 if (!old_filename || !new_filename)
1855 ret = -ERRNO_TO_ERROR(ENOMEM);
1857 ret = para_rename(old_filename, new_filename);
1872 * Change an object in an osl table.
1874 * \param t Pointer to an open osl table.
1875 * \param r Pointer to the row containing the object to be updated.
1876 * \param col_num Number of the column containing the object to be updated.
1877 * \param obj Pointer to the replacement object.
1879 * This function gets rid of all references to the old object. This includes
1880 * removal of the rbtree node in case there is an rbtree associated with \a
1881 * col_num. It then inserts \a obj into the table and the rbtree if necessary.
1883 * If the \p OSL_RBTREE flag is set for \a col_num, you \b MUST call this
1884 * function in order to change the contents of an object, even for volatile or
1885 * mapped columns of constant size (which may be updated directly if \p
1886 * OSL_RBTREE is not set). Otherwise the rbtree might become corrupted.
1890 __export int osl_update_object(struct osl_table *t, const struct osl_row *r,
1891 unsigned col_num, struct osl_object *obj)
1893 struct osl_column *col;
1894 const struct osl_column_description *cd;
1898 return -E_OSL_BAD_TABLE;
1899 col = &t->columns[col_num];
1900 cd = get_column_description(t->desc, col_num);
1901 DEBUG_LOG("updating column %u of %s\n", col_num, t->desc->name);
1902 if (cd->storage_flags & OSL_RBTREE) {
1903 if (search_rbtree(obj, t, col_num, NULL, NULL) > 0)
1904 return -E_OSL_RB_KEY_EXISTS;
1906 if (cd->storage_flags & OSL_FIXED_SIZE) {
1907 if (obj->size != cd->data_size)
1908 return -E_OSL_BAD_DATA_SIZE;
1910 remove_rb_node(t, col_num, r);
1911 if (cd->storage_type == OSL_NO_STORAGE) { /* TODO: If fixed size, reuse object? */
1912 free(r->volatile_objects[col->volatile_num].data);
1913 r->volatile_objects[col->volatile_num] = *obj;
1914 } else if (cd->storage_type == OSL_DISK_STORAGE) {
1916 ret = disk_storage_name_of_row(t, r, &ds_name);
1919 ret = delete_disk_storage_file(t, col_num, ds_name);
1920 if (ret < 0 && !is_errno(-ret, ENOENT)) {
1924 ret = write_disk_storage_file(t, col_num, obj, ds_name);
1928 } else { /* mapped storage */
1929 struct osl_object old_obj;
1930 ret = get_mapped_object(t, col_num, r->num, &old_obj);
1934 * If the updated column is the disk storage name column, the
1935 * disk storage name changes, so we have to rename all disk
1936 * storage objects accordingly.
1938 if (col_num == t->disk_storage_name_column) {
1939 ret = rename_disk_storage_objects(t, &old_obj, obj);
1943 if (cd->storage_flags & OSL_FIXED_SIZE)
1944 memcpy(old_obj.data, obj->data, cd->data_size);
1945 else { /* TODO: if the size doesn't change, use old space */
1946 uint32_t new_data_map_size;
1948 ret = get_row_index(t, r->num, &row_index);
1951 ret = mark_mapped_object_invalid(t, r->num, col_num);
1954 unmap_column(t, col_num);
1955 ret = append_map_file(t, col_num, obj,
1956 &new_data_map_size);
1959 ret = map_column(t, col_num);
1962 update_cell_index(row_index, col, new_data_map_size,
1966 if (cd->storage_flags & OSL_RBTREE) {
1967 ret = insert_rbtree(t, col_num, r, obj);
1975 * Retrieve an object of type \p OSL_DISK_STORAGE by row and column.
1977 * \param t Pointer to an open osl table.
1978 * \param r Pointer to the row containing the object.
1979 * \param col_num The column number.
1980 * \param obj Points to the result upon successful return.
1982 * For columns of type \p OSL_DISK_STORAGE, this function must be used to
1983 * retrieve one of its containing objects. Afterwards, osl_close_disk_object()
1984 * must be called in order to deallocate the resources.
1988 * \sa osl_get_object(), osl_storage_type, osl_close_disk_object().
1990 __export int osl_open_disk_object(const struct osl_table *t, const struct osl_row *r,
1991 unsigned col_num, struct osl_object *obj)
1993 const struct osl_column_description *cd;
1994 char *ds_name, *filename;
1998 return -E_OSL_BAD_TABLE;
1999 cd = get_column_description(t->desc, col_num);
2000 if (cd->storage_type != OSL_DISK_STORAGE)
2001 return -E_OSL_BAD_STORAGE_TYPE;
2003 ret = disk_storage_name_of_row(t, r, &ds_name);
2006 filename = disk_storage_path(t, col_num, ds_name);
2009 return -ERRNO_TO_ERROR(ENOMEM);
2010 DEBUG_LOG("filename: %s\n", filename);
2011 ret = mmap_full_file(filename, O_RDONLY, &obj->data, &obj->size, NULL);
2017 * Free resources that were allocated during osl_open_disk_object().
2019 * \param obj Pointer to the object previously returned by open_disk_object().
2021 * \return The return value of the underlying call to para_munmap().
2023 * \sa para_munmap().
2025 __export int osl_close_disk_object(struct osl_object *obj)
2027 return para_munmap(obj->data, obj->size);
2031 * Get the number of rows of the given table.
2033 * \param t Pointer to an open osl table.
2034 * \param num_rows Result is returned here.
2036 * The number of rows returned via \a num_rows excluding any invalid rows.
2038 * \return Positive on success, \p -E_OSL_BAD_TABLE if \a t is \p NULL.
2040 __export int osl_get_num_rows(const struct osl_table *t, unsigned *num_rows)
2043 return -E_OSL_BAD_TABLE;
2044 assert(t->num_rows >= t->num_invalid_rows);
2045 *num_rows = t->num_rows - t->num_invalid_rows;
2050 * Get the rank of a row.
2052 * \param t An open osl table.
2053 * \param r The row to get the rank of.
2054 * \param col_num The number of an rbtree column.
2055 * \param rank Result pointer.
2057 * The rank is, by definition, the position of the row in the linear order
2058 * determined by an in-order tree walk of the rbtree associated with column
2059 * number \a col_num of \a table.
2063 * \sa osl_get_nth_row().
2065 __export int osl_get_rank(const struct osl_table *t, struct osl_row *r,
2066 unsigned col_num, unsigned *rank)
2068 struct osl_object obj;
2069 struct osl_column *col;
2070 struct rb_node *node;
2071 int ret = check_rbtree_col(t, col_num, &col);
2075 ret = osl_get_object(t, r, col_num, &obj);
2078 ret = search_rbtree(&obj, t, col_num, &node, NULL);
2081 ret = rb_rank(node, rank);
2083 return -E_OSL_BAD_ROW;
2088 * Get the row with n-th greatest value.
2090 * \param t Pointer to an open osl table.
2091 * \param col_num The column number.
2092 * \param n The rank of the desired row.
2093 * \param result Row is returned here.
2095 * Retrieve the n-th order statistic with respect to the compare function
2096 * of the rbtree column \a col_num. In other words, get that row with
2097 * \a n th greatest value in column \a col_num. It's an error if
2098 * \a col_num is not a rbtree column, or if \a n is larger than the
2099 * number of rows in the table.
2103 * \sa osl_storage_flags, osl_compare_func, osl_get_row(),
2104 * osl_rbtree_last_row(), osl_rbtree_first_row(), osl_get_rank().
2106 __export int osl_get_nth_row(const struct osl_table *t, unsigned col_num,
2107 unsigned n, struct osl_row **result)
2109 struct osl_column *col;
2110 struct rb_node *node;
2115 return -E_OSL_RB_KEY_NOT_FOUND;
2116 ret = osl_get_num_rows(t, &num_rows);
2120 return -E_OSL_RB_KEY_NOT_FOUND;
2121 ret = check_rbtree_col(t, col_num, &col);
2124 node = rb_nth(col->rbtree.rb_node, n);
2126 return -E_OSL_RB_KEY_NOT_FOUND;
2127 *result = get_row_pointer(node, col->rbtree_num);
2132 * Get the row corresponding to the smallest rbtree node of a column.
2134 * \param t An open rbtree table.
2135 * \param col_num The number of the rbtree column.
2136 * \param result A pointer to the first row is returned here.
2138 * The rbtree node of the smallest object (with respect to the corresponding
2139 * compare function) is selected and the row containing this object is
2140 * returned. It is an error if \a col_num refers to a column without an
2141 * associated rbtree.
2145 * \sa osl_get_nth_row(), osl_rbtree_last_row().
2147 __export int osl_rbtree_first_row(const struct osl_table *t, unsigned col_num,
2148 struct osl_row **result)
2150 return osl_get_nth_row(t, col_num, 1, result);
2154 * Get the row corresponding to the greatest rbtree node of a column.
2156 * \param t The same meaning as in \p osl_rbtree_first_row().
2157 * \param col_num The same meaning as in \p osl_rbtree_first_row().
2158 * \param result The same meaning as in \p osl_rbtree_first_row().
2160 * This function works just like osl_rbtree_first_row(), the only difference
2161 * is that the row containing the greatest rather than the smallest object is
2166 * \sa osl_get_nth_row(), osl_rbtree_first_row().
2168 __export int osl_rbtree_last_row(const struct osl_table *t, unsigned col_num,
2169 struct osl_row **result)
2172 int ret = osl_get_num_rows(t, &num_rows);
2176 return osl_get_nth_row(t, col_num, num_rows, result);