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_LSEEK.
107 static int __lseek(int fd, off_t *offset, int whence)
109 *offset = lseek(fd, *offset, whence);
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_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_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)
236 if (strchr(name, '/'))
238 if (!strcmp(name, ".."))
240 if (!strcmp(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_BAD_TABLE_DESC;
329 ret = verify_name(t->desc->name);
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_NO_COMPARE_FUNC;
343 if (cd->storage_type == OSL_NO_STORAGE)
345 ret = -E_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_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.
376 * \return Positive on success, negative on errors. Possible errors include: \p
377 * E_BAD_TABLE_DESC, \p E_NO_COLUMN_DESC, \p E_NO_COLUMNS, \p
378 * E_BAD_STORAGE_TYPE, \p E_BAD_STORAGE_FLAGS, \p E_BAD_STORAGE_SIZE, \p
379 * E_NO_UNIQUE_RBTREE_COLUMN, \p E_NO_RBTREE_COL.
381 * \sa struct osl_table.
383 int init_table_structure(const struct osl_table_description *desc,
384 struct osl_table **table_ptr)
386 const struct osl_column_description *cd;
387 struct osl_table *t = calloc(1, sizeof(*t));
388 int i, ret = -ERRNO_TO_ERROR(ENOMEM), have_disk_storage_name_column = 0;
392 ret = -E_BAD_TABLE_DESC;
395 DEBUG_LOG("creating table structure for '%s' from table "
396 "description\n", desc->name);
397 ret = -E_NO_COLUMN_DESC;
398 if (!desc->column_descriptions)
401 if (!desc->num_columns)
403 ret = -ERRNO_TO_ERROR(ENOMEM);
404 t->columns = calloc(desc->num_columns, sizeof(struct osl_column));
408 FOR_EACH_COLUMN(i, t->desc, cd) {
409 enum osl_storage_type st = cd->storage_type;
410 enum osl_storage_flags sf = cd->storage_flags;
411 struct osl_column *col = &t->columns[i];
413 ret = -E_BAD_STORAGE_TYPE;
414 if (st != OSL_MAPPED_STORAGE && st != OSL_DISK_STORAGE
415 && st != OSL_NO_STORAGE)
417 ret = -E_BAD_STORAGE_FLAGS;
418 if (st == OSL_DISK_STORAGE && sf & OSL_RBTREE)
420 ret = -E_BAD_STORAGE_SIZE;
421 if (sf & OSL_FIXED_SIZE && !cd->data_size)
424 case OSL_DISK_STORAGE:
425 t->num_disk_storage_columns++;
427 case OSL_MAPPED_STORAGE:
428 t->num_mapped_columns++;
429 col->index_offset = t->row_index_size;
430 t->row_index_size += 8;
433 col->volatile_num = t->num_volatile_columns;
434 t->num_volatile_columns++;
437 if (sf & OSL_RBTREE) {
438 col->rbtree_num = t->num_rbtrees;
440 if ((sf & OSL_UNIQUE) && (st == OSL_MAPPED_STORAGE)) {
441 if (!have_disk_storage_name_column)
442 t->disk_storage_name_column = i;
443 have_disk_storage_name_column = 1;
447 ret = -E_NO_UNIQUE_RBTREE_COLUMN;
448 if (t->num_disk_storage_columns && !have_disk_storage_name_column)
450 ret = -E_NO_RBTREE_COL;
454 DEBUG_LOG("OK. Index entry size: %u\n", t->row_index_size);
455 ret = init_column_descriptions(t);
467 * Read the table description from index header.
469 * \param map The memory mapping of the index file.
470 * \param desc The values found in the index header are returned here.
472 * Read the index header, check for the paraslash magic string and the table version number.
473 * Read all information stored in the index header into \a desc.
475 * \return Positive on success, negative on errors.
477 * \sa struct osl_table_description, osl_create_table.
479 int read_table_desc(struct osl_object *map, struct osl_table_description *desc)
481 char *buf = map->data;
483 uint16_t header_size;
486 struct osl_column_description *cd;
488 if (map->size < MIN_INDEX_HEADER_SIZE(1))
489 return -E_SHORT_TABLE;
490 if (strncmp(buf + IDX_PARA_MAGIC, PARA_MAGIC, strlen(PARA_MAGIC)))
492 version = read_u8(buf + IDX_VERSION);
493 if (version < MIN_TABLE_VERSION || version > MAX_TABLE_VERSION)
494 return -E_VERSION_MISMATCH;
495 desc->flags = read_u8(buf + IDX_TABLE_FLAGS);
496 desc->num_columns = read_u16(buf + IDX_NUM_COLUMNS);
497 DEBUG_LOG("%u columns\n", desc->num_columns);
498 if (!desc->num_columns)
499 return -E_NO_COLUMNS;
500 header_size = read_u16(buf + IDX_HEADER_SIZE);
501 if (map->size < header_size)
503 desc->column_descriptions = calloc(desc->num_columns,
504 sizeof(struct osl_column_description));
505 if (!desc->column_descriptions)
506 return -ERRNO_TO_ERROR(ENOMEM);
507 offset = IDX_COLUMN_DESCRIPTIONS;
508 FOR_EACH_COLUMN(i, desc, cd) {
511 ret = -E_SHORT_TABLE;
512 if (map->size < offset + MIN_IDX_COLUMN_DESCRIPTION_SIZE) {
513 ERROR_LOG("map size = %zu < %u = offset + min desc size\n",
514 map->size, offset + MIN_IDX_COLUMN_DESCRIPTION_SIZE);
517 cd->storage_type = read_u16(buf + offset + IDX_CD_STORAGE_TYPE);
518 cd->storage_flags = read_u16(buf + offset +
519 IDX_CD_STORAGE_FLAGS);
520 cd->data_size = read_u32(buf + offset + IDX_CD_DATA_SIZE);
521 null_byte = memchr(buf + offset + IDX_CD_NAME, '\0',
522 map->size - offset - IDX_CD_NAME);
523 ret = -E_INDEX_CORRUPTION;
526 ret = -ERRNO_TO_ERROR(ENOMEM);
527 cd->name = strdup(buf + offset + IDX_CD_NAME);
530 offset += index_column_description_size(cd->name);
532 if (offset != header_size) {
533 ret = -E_INDEX_CORRUPTION;
534 ERROR_LOG("real header size = %u != %u = stored header size\n",
535 offset, header_size);
540 FOR_EACH_COLUMN(i, desc, cd)
546 * check whether the table description given by \p t->desc matches the on-disk
547 * table structure stored in the index of \a t.
549 static int compare_table_descriptions(struct osl_table *t)
552 struct osl_table_description desc;
553 const struct osl_column_description *cd1, *cd2;
555 /* read the on-disk structure into desc */
556 ret = read_table_desc(&t->index_map, &desc);
559 ret = -E_BAD_TABLE_FLAGS;
560 if (desc.flags != t->desc->flags)
562 ret = -E_BAD_COLUMN_NUM;
563 if (desc.num_columns > t->desc->num_columns)
565 if (desc.num_columns < t->desc->num_columns) {
566 struct osl_column_description *cd;
567 unsigned diff = t->desc->num_columns - desc.num_columns;
568 INFO_LOG("extending table by %u volatile columns\n", diff);
569 ret = -ERRNO_TO_ERROR(ENOMEM);
570 desc.column_descriptions = realloc(desc.column_descriptions,
571 t->desc->num_columns * sizeof(struct osl_column_description));
572 if (!desc.column_descriptions)
574 for (i = desc.num_columns; i < t->desc->num_columns; i++) {
575 cd = get_column_description(&desc, i);
576 cd->storage_type = OSL_NO_STORAGE;
579 desc.num_columns += diff;
581 FOR_EACH_COLUMN(i, t->desc, cd1) {
582 cd2 = get_column_description(&desc, i);
583 ret = -E_BAD_STORAGE_TYPE;
584 if (cd1->storage_type != cd2->storage_type)
586 if (cd1->storage_type == OSL_NO_STORAGE)
588 ret = -E_BAD_STORAGE_FLAGS;
589 if (cd1->storage_flags != cd2->storage_flags) {
590 ERROR_LOG("sf1 = %u != %u = sf2\n",
591 cd1->storage_flags, cd2->storage_flags);
594 ret = -E_BAD_DATA_SIZE;
595 if (cd1->storage_flags & OSL_FIXED_SIZE)
596 if (cd1->data_size != cd2->data_size)
598 ret = -E_BAD_COLUMN_NAME;
599 if (strcmp(cd1->name, cd2->name))
602 DEBUG_LOG("table description of '%s' matches on-disk data, good\n",
606 FOR_EACH_COLUMN(i, &desc, cd1)
608 free(desc.column_descriptions);
612 static int create_table_index(struct osl_table *t)
614 char *buf, *filename;
616 size_t size = t->index_header_size;
617 const struct osl_column_description *cd;
620 INFO_LOG("creating %zu byte index for table %s\n", size,
622 buf = calloc(1, size);
624 return -ERRNO_TO_ERROR(ENOMEM);
625 sprintf(buf + IDX_PARA_MAGIC, "%s", PARA_MAGIC);
626 write_u8(buf + IDX_TABLE_FLAGS, t->desc->flags);
627 write_u8(buf + IDX_DIRTY_FLAG, 0);
628 write_u8(buf + IDX_VERSION, CURRENT_TABLE_VERSION);
629 write_u16(buf + IDX_NUM_COLUMNS, t->num_mapped_columns + t->num_disk_storage_columns);
630 write_u16(buf + IDX_HEADER_SIZE, t->index_header_size);
631 offset = IDX_COLUMN_DESCRIPTIONS;
632 FOR_EACH_COLUMN(i, t->desc, cd) {
633 /* no need to store info about volatile storage */
634 if (cd->storage_type == OSL_NO_STORAGE)
636 write_u16(buf + offset + IDX_CD_STORAGE_TYPE,
638 write_u16(buf + offset + IDX_CD_STORAGE_FLAGS,
640 if (cd->storage_flags & OSL_FIXED_SIZE)
641 write_u32(buf + offset + IDX_CD_DATA_SIZE,
643 strcpy(buf + offset + IDX_CD_NAME, cd->name);
644 offset += index_column_description_size(cd->name);
646 assert(offset = size);
647 filename = index_filename(t->desc);
649 ret = write_file(filename, buf, size);
651 ret = -ERRNO_TO_ERROR(ENOMEM);
658 * Create a new osl table.
660 * \param desc Pointer to the table description.
664 int osl_create_table(const struct osl_table_description *desc)
666 const struct osl_column_description *cd;
667 char *table_dir = NULL, *filename;
669 int i, ret = init_table_structure(desc, &t);
673 INFO_LOG("creating %s\n", desc->name);
674 FOR_EACH_COLUMN(i, t->desc, cd) {
675 if (cd->storage_type == OSL_NO_STORAGE)
678 ret = para_mkdir(desc->dir, 0777);
679 if (ret < 0 && !is_errno(-ret, EEXIST))
681 table_dir = make_message("%s/%s", desc->dir,
683 ret = -ERRNO_TO_ERROR(ENOMEM);
686 ret = para_mkdir(table_dir, 0777);
690 ret = -ERRNO_TO_ERROR(ENOMEM);
691 filename = column_filename(t, i);
694 INFO_LOG("filename: %s\n", filename);
695 if (cd->storage_type == OSL_MAPPED_STORAGE) {
696 ret = osl_open(filename, O_RDWR | O_CREAT | O_EXCL,
705 ret = para_mkdir(filename, 0777);
710 if (t->num_mapped_columns) {
711 ret = create_table_index(t);
723 static int table_is_dirty(struct osl_table *t)
725 char *buf = (char *)t->index_map.data + IDX_DIRTY_FLAG;
726 uint8_t dirty = read_u8(buf) & 0x1;
730 static void mark_table_dirty(struct osl_table *t)
732 char *buf = (char *)t->index_map.data + IDX_DIRTY_FLAG;
733 write_u8(buf, read_u8(buf) | 1);
736 static void mark_table_clean(struct osl_table *t)
738 char *buf = (char *)t->index_map.data + IDX_DIRTY_FLAG;
739 write_u8(buf, read_u8(buf) & 0xfe);
742 static void unmap_column(struct osl_table *t, unsigned col_num)
744 struct osl_object map = t->columns[col_num].data_map;
748 ret = para_munmap(map.data, map.size);
754 * Unmap all mapped files of an osl table.
756 * \param t Pointer to a mapped table.
757 * \param flags Options for unmapping.
759 * \return Positive on success, negative on errors.
761 * \sa map_table(), enum osl_close_flags, para_munmap().
763 int unmap_table(struct osl_table *t, enum osl_close_flags flags)
766 const struct osl_column_description *cd;
769 if (!t->num_mapped_columns) /* can this ever happen? */
771 DEBUG_LOG("unmapping table '%s'\n", t->desc->name);
772 if (!t->index_map.data)
773 return -E_NOT_MAPPED;
774 if (flags & OSL_MARK_CLEAN)
776 ret = para_munmap(t->index_map.data, t->index_map.size);
779 t->index_map.data = NULL;
782 FOR_EACH_MAPPED_COLUMN(i, t, cd)
787 static int map_column(struct osl_table *t, unsigned col_num)
790 char *filename = column_filename(t, col_num);
794 return -ERRNO_TO_ERROR(ENOMEM);
795 if (stat(filename, &statbuf) < 0) {
799 if (!(S_IFREG & statbuf.st_mode)) {
803 ret = mmap_full_file(filename, O_RDWR,
804 &t->columns[col_num].data_map.data,
805 &t->columns[col_num].data_map.size,
812 * Map the index file and all columns of type \p OSL_MAPPED_STORAGE into memory.
814 * \param t Pointer to an initialized table structure.
815 * \param flags Mapping options.
817 * \return Negative return value on errors; on success the number of rows
818 * (including invalid rows) is returned.
820 * \sa unmap_table(), enum map_table_flags, osl_open_table(), mmap(2).
822 int map_table(struct osl_table *t, enum map_table_flags flags)
825 const struct osl_column_description *cd;
826 int i = 0, ret, num_rows = 0;
828 if (!t->num_mapped_columns)
830 if (t->index_map.data)
831 return -E_ALREADY_MAPPED;
832 filename = index_filename(t->desc);
834 return -ERRNO_TO_ERROR(ENOMEM);
835 DEBUG_LOG("mapping table '%s' (index: %s)\n", t->desc->name, filename);
836 ret = mmap_full_file(filename, flags & MAP_TBL_FL_MAP_RDONLY?
837 O_RDONLY : O_RDWR, &t->index_map.data, &t->index_map.size, NULL);
841 if (flags & MAP_TBL_FL_VERIFY_INDEX) {
842 ret = compare_table_descriptions(t);
847 if (!(flags & MAP_TBL_FL_IGNORE_DIRTY)) {
848 if (table_is_dirty(t)) {
849 ERROR_LOG("%s is dirty\n", t->desc->name);
854 num_rows = table_num_rows(t);
858 FOR_EACH_MAPPED_COLUMN(i, t, cd) {
859 ret = map_column(t, i);
864 err: /* unmap what is already mapped */
865 for (i--; i >= 0; i--) {
866 struct osl_object map = t->columns[i].data_map;
867 para_munmap(map.data, map.size);
870 para_munmap(t->index_map.data, t->index_map.size);
871 t->index_map.data = NULL;
876 * Retrieve a mapped object by row and column number.
878 * \param t Pointer to an open osl table.
879 * \param col_num Number of the mapped column containing the object to retrieve.
880 * \param row_num Number of the row containing the object to retrieve.
881 * \param obj The result is returned here.
883 * It is considered an error if \a col_num does not refer to a column
884 * of storage type \p OSL_MAPPED_STORAGE.
886 * \return Positive on success, negative on errors. Possible errors include:
887 * \p E_BAD_ROW_NUM, \p E_INVALID_OBJECT.
889 * \sa osl_storage_type.
891 int get_mapped_object(const struct osl_table *t, unsigned col_num,
892 uint32_t row_num, struct osl_object *obj)
894 struct osl_column *col = &t->columns[col_num];
900 if (t->num_rows <= row_num)
901 return -E_BAD_ROW_NUM;
902 ret = get_cell_index(t, row_num, col_num, &cell_index);
905 offset = read_u32(cell_index);
906 obj->size = read_u32(cell_index + 4) - 1;
907 header = col->data_map.data + offset;
908 obj->data = header + 1;
909 if (read_u8(header) == 0xff) {
910 ERROR_LOG("col %u, size %zu, offset %u\n", col_num,
912 return -E_INVALID_OBJECT;
917 static int search_rbtree(const struct osl_object *obj,
918 const struct osl_table *t, unsigned col_num,
919 struct rb_node **result, struct rb_node ***rb_link)
921 struct osl_column *col = &t->columns[col_num];
922 struct rb_node **new = &col->rbtree.rb_node, *parent = NULL;
923 const struct osl_column_description *cd =
924 get_column_description(t->desc, col_num);
925 enum osl_storage_type st = cd->storage_type;
927 struct osl_row *this_row = get_row_pointer(*new,
930 struct osl_object this_obj;
932 if (st == OSL_MAPPED_STORAGE) {
933 ret = get_mapped_object(t, col_num, this_row->num,
938 this_obj = this_row->volatile_objects[col->volatile_num];
939 ret = cd->compare_function(obj, &this_obj);
942 *result = get_rb_node_pointer(this_row,
947 new = &((*new)->rb_left);
949 new = &((*new)->rb_right);
955 return -E_RB_KEY_NOT_FOUND;
958 static int insert_rbtree(struct osl_table *t, unsigned col_num,
959 const struct osl_row *row, const struct osl_object *obj)
961 struct rb_node *parent, **rb_link;
964 int ret = search_rbtree(obj, t, col_num, &parent, &rb_link);
967 return -E_RB_KEY_EXISTS;
968 rbtree_num = t->columns[col_num].rbtree_num;
969 n = get_rb_node_pointer(row, rbtree_num);
970 rb_link_node(n, parent, rb_link);
971 rb_insert_color(n, &t->columns[col_num].rbtree);
975 static void remove_rb_node(struct osl_table *t, unsigned col_num,
976 const struct osl_row *row)
978 struct osl_column *col = &t->columns[col_num];
979 const struct osl_column_description *cd =
980 get_column_description(t->desc, col_num);
981 enum osl_storage_flags sf = cd->storage_flags;
982 struct rb_node *victim, *splice_out_node, *tmp;
983 if (!(sf & OSL_RBTREE))
986 * Which node is removed/spliced out actually depends on how many
987 * children the victim node has: If it has no children, it gets
988 * deleted. If it has one child, it gets spliced out. If it has two
989 * children, its successor (which has at most a right child) gets
992 victim = get_rb_node_pointer(row, col->rbtree_num);
993 if (victim->rb_left && victim->rb_right)
994 splice_out_node = rb_next(victim);
996 splice_out_node = victim;
997 /* Go up to the root and decrement the size of each node in the path. */
998 for (tmp = splice_out_node; tmp; tmp = rb_parent(tmp))
1000 rb_erase(victim, &col->rbtree);
1003 static int add_row_to_rbtrees(struct osl_table *t, uint32_t row_num,
1004 struct osl_object *volatile_objs, struct osl_row **row_ptr)
1008 struct osl_row *row = allocate_row(t->num_rbtrees);
1009 const struct osl_column_description *cd;
1012 return -ERRNO_TO_ERROR(ENOMEM);
1014 row->volatile_objects = volatile_objs;
1015 FOR_EACH_RBTREE_COLUMN(i, t, cd) {
1016 if (cd->storage_type == OSL_MAPPED_STORAGE) {
1017 struct osl_object obj;
1018 ret = get_mapped_object(t, i, row_num, &obj);
1021 ret = insert_rbtree(t, i, row, &obj);
1022 } else { /* volatile */
1023 const struct osl_object *obj
1024 = volatile_objs + t->columns[i].volatile_num;
1025 ret = insert_rbtree(t, i, row, obj);
1033 err: /* rollback changes, i.e. remove added entries from rbtrees */
1035 remove_rb_node(t, i--, row);
1040 static void free_volatile_objects(const struct osl_table *t,
1041 enum osl_close_flags flags)
1045 struct osl_column *rb_col;
1046 const struct osl_column_description *cd;
1048 if (!t->num_volatile_columns)
1050 /* find the first rbtree column (any will do) */
1051 FOR_EACH_RBTREE_COLUMN(i, t, cd)
1053 rb_col = t->columns + i;
1054 /* walk that rbtree and free all volatile objects */
1055 for (n = rb_first(&rb_col->rbtree); n; n = rb_next(n)) {
1056 struct osl_row *r = get_row_pointer(n, rb_col->rbtree_num);
1057 if (flags & OSL_FREE_VOLATILE)
1058 FOR_EACH_VOLATILE_COLUMN(j, t, cd) {
1059 if (cd->storage_flags & OSL_DONT_FREE)
1061 free(r->volatile_objects[
1062 t->columns[j].volatile_num].data);
1064 // for (j = 0; j < t->num_volatile_columns; j++)
1065 // free(r->volatile_objects[j].data);
1066 free(r->volatile_objects);
1071 * Erase all rbtree nodes and free resources.
1073 * \param t Pointer to an open osl table.
1075 * This function is called by osl_close_table().
1077 void clear_rbtrees(struct osl_table *t)
1079 const struct osl_column_description *cd;
1080 unsigned i, rbtrees_cleared = 0;
1082 FOR_EACH_RBTREE_COLUMN(i, t, cd) {
1083 struct osl_column *col = &t->columns[i];
1086 for (n = rb_first(&col->rbtree); n;) {
1088 rb_erase(n, &col->rbtree);
1089 if (rbtrees_cleared == t->num_rbtrees) {
1090 r = get_row_pointer(n, col->rbtree_num);
1101 * Close an osl table.
1103 * \param t Pointer to the table to be closed.
1104 * \param flags Options for what should be cleaned up.
1106 * If osl_open_table() succeeds, the resulting table pointer must later be
1107 * passed to this function in order to flush all changes to the file system and
1108 * to free the resources that were allocated by osl_open_table().
1110 * \return Positive on success, negative on errors. Possible errors: \p E_BAD_TABLE,
1111 * errors returned by unmap_table().
1113 * \sa osl_open_table(), unmap_table().
1115 int osl_close_table(struct osl_table *t, enum osl_close_flags flags)
1120 return -E_BAD_TABLE;
1121 free_volatile_objects(t, flags);
1123 ret = unmap_table(t, flags);
1125 ERROR_LOG("unmap_table failed: %d\n", ret);
1132 * Find out whether the given row number corresponds to an invalid row.
1134 * \param t Pointer to the osl table.
1135 * \param row_num The number of the row in question.
1137 * By definition, a row is considered invalid if all its index entries
1140 * \return Positive if \a row_num corresponds to an invalid row,
1141 * zero if it corresponds to a valid row, negative on errors.
1143 int row_is_invalid(struct osl_table *t, uint32_t row_num)
1146 int i, ret = get_row_index(t, row_num, &row_index);
1150 for (i = 0; i < t->row_index_size; i++) {
1151 if ((unsigned char)row_index[i] != 0xff)
1154 INFO_LOG("row %d is invalid\n", row_num);
1159 * Invalidate a row of an osl table.
1161 * \param t Pointer to an open osl table.
1162 * \param row_num Number of the row to mark as invalid.
1164 * This function marks each mapped object in the index entry of \a row as
1167 * \return Positive on success, negative on errors.
1169 int mark_row_invalid(struct osl_table *t, uint32_t row_num)
1172 int ret = get_row_index(t, row_num, &row_index);
1176 INFO_LOG("marking row %d as invalid\n", row_num);
1177 memset(row_index, 0xff, t->row_index_size);
1182 * Initialize all rbtrees and compute number of invalid rows.
1184 * \param t The table containing the rbtrees to be initialized.
1186 * \return Positive on success, negative on errors.
1188 int init_rbtrees(struct osl_table *t)
1191 const struct osl_column_description *cd;
1193 /* create rbtrees */
1194 FOR_EACH_RBTREE_COLUMN(i, t, cd)
1195 t->columns[i].rbtree = RB_ROOT;
1196 /* add valid rows to rbtrees */
1197 t->num_invalid_rows = 0;
1198 for (i = 0; i < t->num_rows; i++) {
1199 ret = row_is_invalid(t, i);
1203 t->num_invalid_rows++;
1206 ret = add_row_to_rbtrees(t, i, NULL, NULL);
1214 * Open an osl table.
1216 * Each osl table must be opened before its data can be accessed.
1218 * \param table_desc Describes the table to be opened.
1219 * \param result Contains a pointer to the open table on success.
1221 * The table description given by \a desc should coincide with the
1222 * description used at creation time.
1226 int osl_open_table(const struct osl_table_description *table_desc,
1227 struct osl_table **result)
1230 struct osl_table *t;
1231 const struct osl_column_description *cd;
1233 INFO_LOG("opening table %s\n", table_desc->name);
1234 ret = init_table_structure(table_desc, &t);
1237 FOR_EACH_DISK_STORAGE_COLUMN(i, t, cd) {
1238 struct stat statbuf;
1239 char *dirname = column_filename(t, i);
1241 ret = -ERRNO_TO_ERROR(ENOMEM);
1244 /* check if directory exists */
1245 ret = stat(dirname, &statbuf);
1248 ret = -ERRNO_TO_ERROR(errno);
1251 ret = -ERRNO_TO_ERROR(ENOTDIR);
1252 if (!S_ISDIR(statbuf.st_mode))
1255 ret = map_table(t, MAP_TBL_FL_VERIFY_INDEX);
1259 DEBUG_LOG("num rows: %d\n", t->num_rows);
1260 ret = init_rbtrees(t);
1262 osl_close_table(t, OSL_MARK_CLEAN); /* ignore further errors */
1273 static int create_disk_storage_object_dir(const struct osl_table *t,
1274 unsigned col_num, const char *ds_name)
1279 if (!(t->desc->flags & OSL_LARGE_TABLE))
1281 dirname = disk_storage_dirname(t, col_num, ds_name);
1283 return -ERRNO_TO_ERROR(ENOMEM);
1284 ret = para_mkdir(dirname, 0777);
1286 if (ret < 0 && !is_errno(-ret, EEXIST))
1291 static int write_disk_storage_file(const struct osl_table *t, unsigned col_num,
1292 const struct osl_object *obj, const char *ds_name)
1297 ret = create_disk_storage_object_dir(t, col_num, ds_name);
1300 filename = disk_storage_path(t, col_num, ds_name);
1302 return -ERRNO_TO_ERROR(ENOMEM);
1303 ret = write_file(filename, obj->data, obj->size);
1308 static int append_map_file(const struct osl_table *t, unsigned col_num,
1309 const struct osl_object *obj, uint32_t *new_size)
1311 char *filename = column_filename(t, col_num);
1313 char header = 0; /* zero means valid object */
1316 return -ERRNO_TO_ERROR(ENOMEM);
1317 ret = append_file(filename, &header, 1, obj->data, obj->size,
1323 static int append_row_index(const struct osl_table *t, char *row_index)
1328 if (!t->num_mapped_columns)
1330 filename = index_filename(t->desc);
1332 return -ERRNO_TO_ERROR(ENOMEM);
1333 ret = append_file(filename, NULL, 0, row_index,
1334 t->row_index_size, NULL);
1340 * A wrapper for truncate(2)
1342 * \param path Name of the regular file to truncate
1343 * \param size Number of bytes to \b shave \b off
1345 * Truncate the regular file named by \a path by \a size bytes.
1347 * \return Positive on success, negative on errors. Possible errors include: \p
1348 * E_STAT, \p E_BAD_SIZE, \p E_TRUNC.
1352 int para_truncate(const char *path, off_t size)
1355 struct stat statbuf;
1358 if (stat(path, &statbuf) < 0)
1361 if (statbuf.st_size < size)
1364 if (truncate(path, statbuf.st_size - size) < 0)
1371 static int truncate_mapped_file(const struct osl_table *t, unsigned col_num,
1375 char *filename = column_filename(t, col_num);
1378 return -ERRNO_TO_ERROR(ENOMEM);
1379 ret = para_truncate(filename, size);
1384 static int delete_disk_storage_file(const struct osl_table *t, unsigned col_num,
1385 const char *ds_name)
1387 char *dirname, *filename = disk_storage_path(t, col_num, ds_name);
1391 return -ERRNO_TO_ERROR(ENOMEM);
1392 ret = unlink(filename);
1396 return -ERRNO_TO_ERROR(err);
1397 if (!(t->desc->flags & OSL_LARGE_TABLE))
1399 dirname = disk_storage_dirname(t, col_num, ds_name);
1401 return -ERRNO_TO_ERROR(ENOMEM);
1408 * Add a new row to an osl table and retrieve this row.
1410 * \param t Pointer to an open osl table.
1411 * \param objects Array of objects to be added.
1412 * \param row Result pointer.
1414 * The \a objects parameter must point to an array containing one object per
1415 * column. The order of the objects in the array is given by the table
1416 * description of \a table. Several sanity checks are performed during object
1417 * insertion and the function returns without modifying the table if any of
1418 * these tests fail. In fact, it is atomic in the sense that it either
1419 * succeeds or leaves the table unchanged (i.e. either all or none of the
1420 * objects are added to the table).
1422 * It is considered an error if an object is added to a column with associated
1423 * rbtree if this object is equal to an object already contained in that column
1424 * (i.e. the compare function for the column's rbtree returns zero).
1426 * Possible errors include: \p E_RB_KEY_EXISTS, \p E_BAD_DATA_SIZE.
1428 * \return Positive on success, negative on errors.
1430 * \sa struct osl_table_description, osl_compare_func, osl_add_row().
1432 int osl_add_and_get_row(struct osl_table *t, struct osl_object *objects,
1433 struct osl_row **row)
1436 char *ds_name = NULL;
1437 struct rb_node **rb_parents = NULL, ***rb_links = NULL;
1438 char *new_row_index = NULL;
1439 struct osl_object *volatile_objs = NULL;
1440 const struct osl_column_description *cd;
1443 return -E_BAD_TABLE;
1444 rb_parents = malloc(t->num_rbtrees * sizeof(struct rn_node*));
1446 return -ERRNO_TO_ERROR(ENOMEM);
1447 rb_links = malloc(t->num_rbtrees * sizeof(struct rn_node**));
1450 return -ERRNO_TO_ERROR(ENOMEM);
1452 if (t->num_mapped_columns) {
1453 new_row_index = malloc(t->row_index_size);
1454 if (!new_row_index) {
1457 return -ERRNO_TO_ERROR(ENOMEM);
1460 /* pass 1: sanity checks */
1461 // DEBUG_LOG("sanity tests: %p:%p\n", objects[0].data,
1462 // objects[1].data);
1463 FOR_EACH_COLUMN(i, t->desc, cd) {
1464 enum osl_storage_type st = cd->storage_type;
1465 enum osl_storage_flags sf = cd->storage_flags;
1467 // ret = -E_NULL_OBJECT;
1470 if (st == OSL_DISK_STORAGE)
1472 if (sf & OSL_RBTREE) {
1473 unsigned rbtree_num = t->columns[i].rbtree_num;
1474 ret = -E_RB_KEY_EXISTS;
1475 // DEBUG_LOG("checking whether %p exists\n",
1476 // objects[i].data);
1477 if (search_rbtree(objects + i, t, i,
1478 &rb_parents[rbtree_num],
1479 &rb_links[rbtree_num]) > 0)
1482 if (sf & OSL_FIXED_SIZE) {
1483 // DEBUG_LOG("fixed size. need: %zu, have: %d\n",
1484 // objects[i].size, cd->data_size);
1485 ret = -E_BAD_DATA_SIZE;
1486 if (objects[i].size != cd->data_size)
1490 if (t->num_disk_storage_columns) {
1491 ds_name = disk_storage_name_of_object(t,
1492 &objects[t->disk_storage_name_column]);
1493 ret = -ERRNO_TO_ERROR(ENOMEM);
1497 ret = unmap_table(t, OSL_MARK_CLEAN);
1500 // DEBUG_LOG("sanity tests passed%s\n", "");
1501 /* pass 2: create data files, append map data */
1502 FOR_EACH_COLUMN(i, t->desc, cd) {
1503 enum osl_storage_type st = cd->storage_type;
1504 if (st == OSL_NO_STORAGE)
1506 if (st == OSL_MAPPED_STORAGE) {
1508 struct osl_column *col = &t->columns[i];
1509 // DEBUG_LOG("appending object of size %zu\n",
1510 // objects[i].size);
1511 ret = append_map_file(t, i, objects + i, &new_size);
1514 update_cell_index(new_row_index, col, new_size,
1519 ret = write_disk_storage_file(t, i, objects + i, ds_name);
1523 ret = append_row_index(t, new_row_index);
1526 ret = map_table(t, MAP_TBL_FL_VERIFY_INDEX);
1527 if (ret < 0) { /* truncate index and rollback changes */
1528 char *filename = index_filename(t->desc);
1530 para_truncate(filename, t->row_index_size);
1534 /* pass 3: add entry to rbtrees */
1535 if (t->num_volatile_columns) {
1536 ret = -ERRNO_TO_ERROR(ENOMEM);
1537 volatile_objs = calloc(t->num_volatile_columns,
1538 sizeof(struct osl_object));
1541 FOR_EACH_VOLATILE_COLUMN(i, t, cd)
1542 volatile_objs[t->columns[i].volatile_num] = objects[i];
1545 // DEBUG_LOG("adding new entry as row #%d\n", t->num_rows - 1);
1546 ret = add_row_to_rbtrees(t, t->num_rows - 1, volatile_objs, row);
1549 // DEBUG_LOG("added new entry as row #%d\n", t->num_rows - 1);
1552 rollback: /* rollback all changes made, ignore further errors */
1553 for (i--; i >= 0; i--) {
1554 cd = get_column_description(t->desc, i);
1555 enum osl_storage_type st = cd->storage_type;
1556 if (st == OSL_NO_STORAGE)
1559 if (st == OSL_MAPPED_STORAGE)
1560 truncate_mapped_file(t, i, objects[i].size);
1561 else /* disk storage */
1562 delete_disk_storage_file(t, i, ds_name);
1564 /* ignore error and return previous error value */
1565 map_table(t, MAP_TBL_FL_VERIFY_INDEX);
1567 free(new_row_index);
1575 * Add a new row to an osl table.
1577 * \param t Same meaning as osl_add_and_get_row().
1578 * \param objects Same meaning as osl_add_and_get_row().
1580 * \return The return value of the underlying call to osl_add_and_get_row().
1582 * This is equivalent to osl_add_and_get_row(t, objects, NULL).
1584 int osl_add_row(struct osl_table *t, struct osl_object *objects)
1586 return osl_add_and_get_row(t, objects, NULL);
1590 * Retrieve an object identified by row and column
1592 * \param t Pointer to an open osl table.
1593 * \param r Pointer to the row.
1594 * \param col_num The column number.
1595 * \param object The result pointer.
1597 * The column determined by \a col_num must be of type \p OSL_MAPPED_STORAGE
1598 * or \p OSL_NO_STORAGE, i.e. no disk storage objects may be retrieved by this
1601 * \return Positive if object was found, negative on errors. Possible errors
1602 * include: \p E_BAD_TABLE, \p E_BAD_STORAGE_TYPE.
1604 * \sa osl_storage_type, osl_open_disk_object().
1606 int osl_get_object(const struct osl_table *t, const struct osl_row *r,
1607 unsigned col_num, struct osl_object *object)
1609 const struct osl_column_description *cd;
1612 return -E_BAD_TABLE;
1613 cd = get_column_description(t->desc, col_num);
1614 /* col must not be disk storage */
1615 if (cd->storage_type == OSL_DISK_STORAGE)
1616 return -E_BAD_STORAGE_TYPE;
1617 if (cd->storage_type == OSL_MAPPED_STORAGE)
1618 return get_mapped_object(t, col_num, r->num, object);
1620 *object = r->volatile_objects[t->columns[col_num].volatile_num];
1624 static int mark_mapped_object_invalid(const struct osl_table *t,
1625 uint32_t row_num, unsigned col_num)
1627 struct osl_object obj;
1629 int ret = get_mapped_object(t, col_num, row_num, &obj);
1640 * Delete a row from an osl table.
1642 * \param t Pointer to an open osl table.
1643 * \param row Pointer to the row to delete.
1645 * This removes all disk storage objects, removes all rbtree nodes, and frees
1646 * all volatile objects belonging to the given row. For mapped columns, the
1647 * data is merely marked invalid and may be pruned from time to time by
1650 * \return Positive on success, negative on errors. Possible errors include:
1651 * \p E_BAD_TABLE, errors returned by osl_get_object().
1653 int osl_del_row(struct osl_table *t, struct osl_row *row)
1655 struct osl_row *r = row;
1657 const struct osl_column_description *cd;
1660 return -E_BAD_TABLE;
1661 INFO_LOG("deleting row %p\n", row);
1663 if (t->num_disk_storage_columns) {
1665 ret = disk_storage_name_of_row(t, r, &ds_name);
1668 FOR_EACH_DISK_STORAGE_COLUMN(i, t, cd)
1669 delete_disk_storage_file(t, i, ds_name);
1672 FOR_EACH_COLUMN(i, t->desc, cd) {
1673 struct osl_column *col = t->columns + i;
1674 enum osl_storage_type st = cd->storage_type;
1675 remove_rb_node(t, i, r);
1676 if (st == OSL_MAPPED_STORAGE) {
1677 mark_mapped_object_invalid(t, r->num, i);
1680 if (st == OSL_NO_STORAGE && !(cd->storage_flags & OSL_DONT_FREE))
1681 free(r->volatile_objects[col->volatile_num].data);
1683 if (t->num_mapped_columns) {
1684 ret = mark_row_invalid(t, r->num);
1687 t->num_invalid_rows++;
1692 free(r->volatile_objects);
1697 /* test if column has an rbtree */
1698 static int check_rbtree_col(const struct osl_table *t, unsigned col_num,
1699 struct osl_column **col)
1702 return -E_BAD_TABLE;
1703 if (!(get_column_description(t->desc, col_num)->storage_flags & OSL_RBTREE))
1704 return -E_BAD_STORAGE_FLAGS;
1705 *col = t->columns + col_num;
1710 * Get the row that contains the given object.
1712 * \param t Pointer to an open osl table.
1713 * \param col_num The number of the column to be searched.
1714 * \param obj The object to be looked up.
1715 * \param result Points to the row containing \a obj.
1717 * Lookup \a obj in \a t and return the row containing \a obj. The column
1718 * specified by \a col_num must have an associated rbtree.
1720 * \return Positive on success, negative on errors. If an error occurred, \a
1721 * result is set to \p NULL. Possible errors include: \p E_BAD_TABLE, \p
1722 * E_BAD_STORAGE_FLAGS, errors returned by get_mapped_object(), \p
1723 * E_RB_KEY_NOT_FOUND.
1725 * \sa osl_storage_flags
1727 int osl_get_row(const struct osl_table *t, unsigned col_num,
1728 const struct osl_object *obj, struct osl_row **result)
1731 struct rb_node *node;
1732 struct osl_row *row;
1733 struct osl_column *col;
1736 ret = check_rbtree_col(t, col_num, &col);
1739 ret = search_rbtree(obj, t, col_num, &node, NULL);
1742 row = get_row_pointer(node, t->columns[col_num].rbtree_num);
1747 static int rbtree_loop(struct osl_column *col, void *private_data,
1748 osl_rbtree_loop_func *func)
1750 struct rb_node *n, *tmp;
1752 /* this for-loop is safe against removal of an entry */
1753 for (n = rb_first(&col->rbtree), tmp = n? rb_next(n) : NULL;
1755 n = tmp, tmp = tmp? rb_next(tmp) : NULL) {
1756 struct osl_row *r = get_row_pointer(n, col->rbtree_num);
1757 int ret = func(r, private_data);
1764 static int rbtree_loop_reverse(struct osl_column *col, void *private_data,
1765 osl_rbtree_loop_func *func)
1767 struct rb_node *n, *tmp;
1769 /* safe against removal of an entry */
1770 for (n = rb_last(&col->rbtree), tmp = n? rb_prev(n) : NULL;
1772 n = tmp, tmp = tmp? rb_prev(tmp) : NULL) {
1773 struct osl_row *r = get_row_pointer(n, col->rbtree_num);
1774 int ret = func(r, private_data);
1782 * Loop over all nodes in an rbtree.
1784 * \param t Pointer to an open osl table.
1785 * \param col_num The column to use for iterating over the elements.
1786 * \param private_data Pointer that gets passed to \a func.
1787 * \param func The function to be called for each node in the rbtree.
1789 * This function does an in-order walk of the rbtree associated with \a
1790 * col_num. It is an error if the \p OSL_RBTREE flag is not set for this
1791 * column. For each node in the rbtree, the given function \a func is called
1792 * with two pointers as arguments: The first osl_row* argument points to the
1793 * row that contains the object corresponding to the rbtree node currently
1794 * traversed, and the \a private_data pointer is passed verbatim to \a func as the
1795 * second argument. The loop terminates either if \a func returns a negative
1796 * value, or if all nodes of the tree have been visited.
1799 * \return Positive on success, negative on errors. If the termination of the
1800 * loop was caused by \a func returning a negative value, this value is
1803 * \sa osl_storage_flags, osl_rbtree_loop_reverse(), osl_compare_func.
1805 int osl_rbtree_loop(const struct osl_table *t, unsigned col_num,
1806 void *private_data, osl_rbtree_loop_func *func)
1808 struct osl_column *col;
1810 int ret = check_rbtree_col(t, col_num, &col);
1813 return rbtree_loop(col, private_data, func);
1817 * Loop over all nodes in an rbtree in reverse order.
1819 * \param t Identical meaning as in \p osl_rbtree_loop().
1820 * \param col_num Identical meaning as in \p osl_rbtree_loop().
1821 * \param private_data Identical meaning as in \p osl_rbtree_loop().
1822 * \param func Identical meaning as in \p osl_rbtree_loop().
1824 * This function is identical to \p osl_rbtree_loop(), the only difference
1825 * is that the tree is walked in reverse order.
1827 * \return The same return value as \p osl_rbtree_loop().
1829 * \sa osl_rbtree_loop().
1831 int osl_rbtree_loop_reverse(const struct osl_table *t, unsigned col_num,
1832 void *private_data, osl_rbtree_loop_func *func)
1834 struct osl_column *col;
1836 int ret = check_rbtree_col(t, col_num, &col);
1839 return rbtree_loop_reverse(col, private_data, func);
1842 /* TODO: Rollback changes on errors */
1843 static int rename_disk_storage_objects(struct osl_table *t,
1844 struct osl_object *old_obj, struct osl_object *new_obj)
1847 const struct osl_column_description *cd;
1848 char *old_ds_name, *new_ds_name;
1850 if (!t->num_disk_storage_columns)
1851 return 1; /* nothing to do */
1852 if (old_obj->size == new_obj->size && !memcmp(new_obj->data,
1853 old_obj->data, new_obj->size))
1854 return 1; /* object did not change */
1855 old_ds_name = disk_storage_name_of_object(t, old_obj);
1856 new_ds_name = disk_storage_name_of_object(t, new_obj);
1857 ret = -ERRNO_TO_ERROR(ENOMEM);
1858 if (!old_ds_name || ! new_ds_name)
1861 FOR_EACH_DISK_STORAGE_COLUMN(i, t, cd) {
1862 char *old_filename, *new_filename;
1863 ret = create_disk_storage_object_dir(t, i, new_ds_name);
1866 old_filename = disk_storage_path(t, i, old_ds_name);
1867 new_filename = disk_storage_path(t, i, new_ds_name);
1868 if (!old_filename || !new_filename)
1869 ret = -ERRNO_TO_ERROR(ENOMEM);
1871 ret = para_rename(old_filename, new_filename);
1886 * Change an object in an osl table.
1888 * \param t Pointer to an open osl table.
1889 * \param r Pointer to the row containing the object to be updated.
1890 * \param col_num Number of the column containing the object to be updated.
1891 * \param obj Pointer to the replacement object.
1893 * This function gets rid of all references to the old object. This includes
1894 * removal of the rbtree node in case there is an rbtree associated with \a
1895 * col_num. It then inserts \a obj into the table and the rbtree if necessary.
1897 * If the \p OSL_RBTREE flag is set for \a col_num, you \b MUST call this
1898 * function in order to change the contents of an object, even for volatile or
1899 * mapped columns of constant size (which may be updated directly if \p
1900 * OSL_RBTREE is not set). Otherwise the rbtree might become corrupted.
1904 int osl_update_object(struct osl_table *t, const struct osl_row *r,
1905 unsigned col_num, struct osl_object *obj)
1907 struct osl_column *col;
1908 const struct osl_column_description *cd;
1912 return -E_BAD_TABLE;
1913 col = &t->columns[col_num];
1914 cd = get_column_description(t->desc, col_num);
1915 DEBUG_LOG("updating column %u of %s\n", col_num, t->desc->name);
1916 if (cd->storage_flags & OSL_RBTREE) {
1917 if (search_rbtree(obj, t, col_num, NULL, NULL) > 0)
1918 return -E_RB_KEY_EXISTS;
1920 if (cd->storage_flags & OSL_FIXED_SIZE) {
1921 if (obj->size != cd->data_size)
1922 return -E_BAD_DATA_SIZE;
1924 remove_rb_node(t, col_num, r);
1925 if (cd->storage_type == OSL_NO_STORAGE) { /* TODO: If fixed size, reuse object? */
1926 free(r->volatile_objects[col->volatile_num].data);
1927 r->volatile_objects[col->volatile_num] = *obj;
1928 } else if (cd->storage_type == OSL_DISK_STORAGE) {
1930 ret = disk_storage_name_of_row(t, r, &ds_name);
1933 ret = delete_disk_storage_file(t, col_num, ds_name);
1934 if (ret < 0 && !is_errno(-ret, ENOENT)) {
1938 ret = write_disk_storage_file(t, col_num, obj, ds_name);
1942 } else { /* mapped storage */
1943 struct osl_object old_obj;
1944 ret = get_mapped_object(t, col_num, r->num, &old_obj);
1948 * If the updated column is the disk storage name column, the
1949 * disk storage name changes, so we have to rename all disk
1950 * storage objects accordingly.
1952 if (col_num == t->disk_storage_name_column) {
1953 ret = rename_disk_storage_objects(t, &old_obj, obj);
1957 if (cd->storage_flags & OSL_FIXED_SIZE)
1958 memcpy(old_obj.data, obj->data, cd->data_size);
1959 else { /* TODO: if the size doesn't change, use old space */
1960 uint32_t new_data_map_size;
1962 ret = get_row_index(t, r->num, &row_index);
1965 ret = mark_mapped_object_invalid(t, r->num, col_num);
1968 unmap_column(t, col_num);
1969 ret = append_map_file(t, col_num, obj,
1970 &new_data_map_size);
1973 ret = map_column(t, col_num);
1976 update_cell_index(row_index, col, new_data_map_size,
1980 if (cd->storage_flags & OSL_RBTREE) {
1981 ret = insert_rbtree(t, col_num, r, obj);
1989 * Retrieve an object of type \p OSL_DISK_STORAGE by row and column.
1991 * \param t Pointer to an open osl table.
1992 * \param r Pointer to the row containing the object.
1993 * \param col_num The column number.
1994 * \param obj Points to the result upon successful return.
1996 * For columns of type \p OSL_DISK_STORAGE, this function must be used to
1997 * retrieve one of its containing objects. Afterwards, osl_close_disk_object()
1998 * must be called in order to deallocate the resources.
2000 * \return Positive on success, negative on errors. Possible errors include:
2001 * \p E_BAD_TABLE, \p E_BAD_STORAGE_TYPE, errors returned by osl_get_object().
2003 * \sa osl_get_object(), osl_storage_type, osl_close_disk_object().
2005 int osl_open_disk_object(const struct osl_table *t, const struct osl_row *r,
2006 unsigned col_num, struct osl_object *obj)
2008 const struct osl_column_description *cd;
2009 char *ds_name, *filename;
2013 return -E_BAD_TABLE;
2014 cd = get_column_description(t->desc, col_num);
2015 if (cd->storage_type != OSL_DISK_STORAGE)
2016 return -E_BAD_STORAGE_TYPE;
2018 ret = disk_storage_name_of_row(t, r, &ds_name);
2021 filename = disk_storage_path(t, col_num, ds_name);
2024 return -ERRNO_TO_ERROR(ENOMEM);
2025 DEBUG_LOG("filename: %s\n", filename);
2026 ret = mmap_full_file(filename, O_RDONLY, &obj->data, &obj->size, NULL);
2032 * Free resources that were allocated during osl_open_disk_object().
2034 * \param obj Pointer to the object previously returned by open_disk_object().
2036 * \return The return value of the underlying call to para_munmap().
2038 * \sa para_munmap().
2040 int osl_close_disk_object(struct osl_object *obj)
2042 return para_munmap(obj->data, obj->size);
2046 * Get the number of rows of the given table.
2048 * \param t Pointer to an open osl table.
2049 * \param num_rows Result is returned here.
2051 * The number of rows returned via \a num_rows excluding any invalid rows.
2053 * \return Positive on success, \p -E_BAD_TABLE if \a t is \p NULL.
2055 int osl_get_num_rows(const struct osl_table *t, unsigned *num_rows)
2058 return -E_BAD_TABLE;
2059 assert(t->num_rows >= t->num_invalid_rows);
2060 *num_rows = t->num_rows - t->num_invalid_rows;
2065 * Get the rank of a row.
2067 * \param t An open osl table.
2068 * \param r The row to get the rank of.
2069 * \param col_num The number of an rbtree column.
2070 * \param rank Result pointer.
2072 * The rank is, by definition, the position of the row in the linear order
2073 * determined by an in-order tree walk of the rbtree associated with column
2074 * number \a col_num of \a table.
2076 * \return Positive on success, negative on errors.
2078 * \sa osl_get_nth_row().
2080 int osl_get_rank(const struct osl_table *t, struct osl_row *r,
2081 unsigned col_num, unsigned *rank)
2083 struct osl_object obj;
2084 struct osl_column *col;
2085 struct rb_node *node;
2086 int ret = check_rbtree_col(t, col_num, &col);
2090 ret = osl_get_object(t, r, col_num, &obj);
2093 ret = search_rbtree(&obj, t, col_num, &node, NULL);
2096 ret = rb_rank(node, rank);
2103 * Get the row with n-th greatest value.
2105 * \param t Pointer to an open osl table.
2106 * \param col_num The column number.
2107 * \param n The rank of the desired row.
2108 * \param result Row is returned here.
2110 * Retrieve the n-th order statistic with respect to the compare function
2111 * of the rbtree column \a col_num. In other words, get that row with
2112 * \a n th greatest value in column \a col_num. It's an error if
2113 * \a col_num is not a rbtree column, or if \a n is larger than the
2114 * number of rows in the table.
2116 * \return Positive on success, negative on errors. Possible errors:
2117 * \p E_BAD_TABLE, \p E_BAD_STORAGE_FLAGS, \p E_RB_KEY_NOT_FOUND.
2119 * \sa osl_storage_flags, osl_compare_func, osl_get_row(),
2120 * osl_rbtree_last_row(), osl_rbtree_first_row(), osl_get_rank().
2122 int osl_get_nth_row(const struct osl_table *t, unsigned col_num,
2123 unsigned n, struct osl_row **result)
2125 struct osl_column *col;
2126 struct rb_node *node;
2131 return -E_RB_KEY_NOT_FOUND;
2132 ret = osl_get_num_rows(t, &num_rows);
2136 return -E_RB_KEY_NOT_FOUND;
2137 ret = check_rbtree_col(t, col_num, &col);
2140 node = rb_nth(col->rbtree.rb_node, n);
2142 return -E_RB_KEY_NOT_FOUND;
2143 *result = get_row_pointer(node, col->rbtree_num);
2148 * Get the row corresponding to the smallest rbtree node of a column.
2150 * \param t An open rbtree table.
2151 * \param col_num The number of the rbtree column.
2152 * \param result A pointer to the first row is returned here.
2154 * The rbtree node of the smallest object (with respect to the corresponding
2155 * compare function) is selected and the row containing this object is
2156 * returned. It is an error if \a col_num refers to a column without an
2157 * associated rbtree.
2159 * \return Positive on success, negative on errors.
2161 * \sa osl_get_nth_row(), osl_rbtree_last_row().
2163 int osl_rbtree_first_row(const struct osl_table *t, unsigned col_num,
2164 struct osl_row **result)
2166 return osl_get_nth_row(t, col_num, 1, result);
2170 * Get the row corresponding to the greatest rbtree node of a column.
2172 * \param t The same meaning as in \p osl_rbtree_first_row().
2173 * \param col_num The same meaning as in \p osl_rbtree_first_row().
2174 * \param result The same meaning as in \p osl_rbtree_first_row().
2176 * This function works just like osl_rbtree_first_row(), the only difference
2177 * is that the row containing the greatest rather than the smallest object is
2180 * \return Positive on success, negative on errors.
2182 * \sa osl_get_nth_row(), osl_rbtree_first_row().
2184 int osl_rbtree_last_row(const struct osl_table *t, unsigned col_num,
2185 struct osl_row **result)
2188 int ret = osl_get_num_rows(t, &num_rows);
2192 return osl_get_nth_row(t, col_num, num_rows, result);