2 * Copyright (C) 2007-2008 Andre Noll <maan@systemlinux.org>
4 * Licensed under the GPL v2. For licencing details see COPYING.
7 /** \file osl.c Object storage layer functions. */
8 #include <dirent.h> /* readdir() */
18 * A wrapper for lseek(2).
20 * \param fd The file descriptor whose offset is to be to repositioned.
21 * \param offset A value-result parameter.
22 * \param whence Usual repositioning directive.
24 * Reposition the offset of the file descriptor \a fd to the argument \a offset
25 * according to the directive \a whence. Upon successful return, \a offset
26 * contains the resulting offset location as measured in bytes from the
27 * beginning of the file.
29 * \return Positive on success. Otherwise, the function returns \p -E_LSEEK.
33 int para_lseek(int fd, off_t *offset, int whence)
35 *offset = lseek(fd, *offset, whence);
43 * Wrapper for the write system call.
45 * \param fd The file descriptor to write to.
46 * \param buf The buffer to write.
47 * \param size The length of \a buf in bytes.
49 * This function writes out the given buffer and retries if an interrupt
50 * occurred during the write.
52 * \return On success, the number of bytes written is returned, otherwise, the
53 * function returns \p -E_WRITE.
57 ssize_t para_write(int fd, const void *buf, size_t size)
62 ret = write(fd, buf, size);
63 if ((ret < 0) && (errno == EAGAIN || errno == EINTR))
65 return ret >= 0? ret : -E_WRITE;
70 * Write the whole buffer to a file descriptor.
72 * \param fd The file descriptor to write to.
73 * \param buf The buffer to write.
74 * \param size The length of \a buf in bytes.
76 * This function writes the given buffer and continues on short writes and
77 * when interrupted by a signal.
79 * \return Positive on success, negative on errors. Possible errors: any
80 * errors returned by para_write().
84 ssize_t para_write_all(int fd, const void *buf, size_t size)
86 // PARA_DEBUG_LOG("writing %zu bytes\n", size);
89 ssize_t ret = para_write(fd, b, size);
90 // PARA_DEBUG_LOG("ret: %zd\n", ret);
99 * Open a file, write the given buffer and close the file.
101 * \param filename Full path to the file to open.
102 * \param buf The buffer to write to the file.
103 * \param size The size of \a buf.
105 * \return Positive on success, negative on errors. Possible errors include:
106 * any errors from para_open() or para_write().
108 * \sa para_open(), para_write().
110 int para_write_file(const char *filename, const void *buf, size_t size)
114 ret = para_open(filename, O_WRONLY | O_CREAT | O_EXCL, 0644);
118 ret = para_write_all(fd, buf, size);
127 static int append_file(const char *filename, char *header, size_t header_size,
128 char *data, size_t data_size, uint32_t *new_pos)
132 // PARA_DEBUG_LOG("appending %zu + %zu bytes\n", header_size, data_size);
133 ret = para_open(filename, O_WRONLY | O_CREAT | O_APPEND, 0644);
137 if (header && header_size) {
138 ret = para_write_all(fd, header, header_size);
142 ret = para_write_all(fd, data, data_size);
147 ret = para_lseek(fd, &offset, SEEK_END);
150 // PARA_DEBUG_LOG("new file size: " FMT_OFF_T "\n", offset);
160 * Traverse the given directory recursively.
162 * \param dirname The directory to traverse.
163 * \param func The function to call for each entry.
164 * \param private_data Pointer to an arbitrary data structure.
166 * For each regular file under \a dirname, the supplied function \a func is
167 * called. The full path of the regular file and the \a private_data pointer
168 * are passed to \a func. Directories for which the calling process has no
169 * permissions to change to are silently ignored.
173 int for_each_file_in_dir(const char *dirname,
174 int (*func)(const char *, void *), void *private_data)
177 struct dirent *entry;
178 int cwd_fd, ret2, ret = para_opendir(dirname, &dir, &cwd_fd);
181 return ret == -ERRNO_TO_PARA_ERROR(EACCES)? 1 : ret;
182 /* scan cwd recursively */
183 while ((entry = readdir(dir))) {
188 if (!strcmp(entry->d_name, "."))
190 if (!strcmp(entry->d_name, ".."))
192 if (lstat(entry->d_name, &s) == -1)
195 if (!S_ISREG(m) && !S_ISDIR(m))
197 tmp = make_message("%s/%s", dirname, entry->d_name);
199 ret = func(tmp, private_data);
206 ret = for_each_file_in_dir(tmp, func, private_data);
214 ret2 = para_fchdir(cwd_fd);
215 if (ret2 < 0 && ret >= 0)
221 static int verify_name(const char *name)
227 if (strchr(name, '/'))
229 if (!strcmp(name, ".."))
231 if (!strcmp(name, "."))
237 * Compare two osl objects pointing to unsigned integers of 32 bit size.
239 * \param obj1 Pointer to the first integer.
240 * \param obj2 Pointer to the second integer.
242 * \return The values required for an osl compare function.
244 * \sa osl_compare_func, osl_hash_compare().
246 int uint32_compare(const struct osl_object *obj1, const struct osl_object *obj2)
248 uint32_t d1 = read_u32((const char *)obj1->data);
249 uint32_t d2 = read_u32((const char *)obj2->data);
259 * Compare two osl objects pointing to hash values.
261 * \param obj1 Pointer to the first hash object.
262 * \param obj2 Pointer to the second hash object.
264 * \return The values required for an osl compare function.
266 * \sa osl_compare_func, uint32_compare().
268 int osl_hash_compare(const struct osl_object *obj1, const struct osl_object *obj2)
270 return hash_compare((HASH_TYPE *)obj1->data, (HASH_TYPE *)obj2->data);
273 static char *disk_storage_dirname(const struct osl_table *t, unsigned col_num,
276 char *dirname, *column_name = column_filename(t, col_num);
278 if (!(t->desc->flags & OSL_LARGE_TABLE))
280 dirname = make_message("%s/%.2s", column_name, ds_name);
285 static char *disk_storage_name_of_object(const struct osl_table *t,
286 const struct osl_object *obj)
288 HASH_TYPE hash[HASH_SIZE];
289 hash_object(obj, hash);
290 return disk_storage_name_of_hash(t, hash);
293 static int disk_storage_name_of_row(const struct osl_table *t,
294 const struct osl_row *row, char **name)
296 struct osl_object obj;
297 int ret = osl_get_object(t, row, t->disk_storage_name_column, &obj);
301 *name = disk_storage_name_of_object(t, &obj);
305 static void column_name_hash(const char *col_name, HASH_TYPE *hash)
307 hash_function(col_name, strlen(col_name), hash);
310 static int init_column_descriptions(struct osl_table *t)
313 const struct osl_column_description *cd;
315 ret = -E_BAD_TABLE_DESC;
316 ret = verify_name(t->desc->name);
320 if (!t->desc->dir && (t->num_disk_storage_columns || t->num_mapped_columns))
322 /* the size of the index header without column descriptions */
323 t->index_header_size = IDX_COLUMN_DESCRIPTIONS;
324 FOR_EACH_COLUMN(i, t->desc, cd) {
325 struct osl_column *col = t->columns + i;
326 if (cd->storage_flags & OSL_RBTREE) {
327 if (!cd->compare_function)
328 return -E_NO_COMPARE_FUNC;
330 if (cd->storage_type == OSL_NO_STORAGE)
332 ret = -E_NO_COLUMN_NAME;
333 if (!cd->name || !cd->name[0])
335 ret = verify_name(cd->name);
338 t->index_header_size += index_column_description_size(cd->name);
339 column_name_hash(cd->name, col->name_hash);
340 ret = -E_DUPLICATE_COL_NAME;
341 for (j = i + 1; j < t->desc->num_columns; j++) {
342 const char *name2 = get_column_description(t->desc,
344 if (cd->name && name2 && !strcmp(cd->name, name2))
354 * Initialize a struct table from given table description.
356 * \param desc The description of the osl table.
357 * \param table_ptr Result is returned here.
359 * This function performs several sanity checks on \p desc and returns if any
360 * of these tests fail. On success, a struct \p osl_table is allocated and
361 * initialized with data derived from \p desc.
363 * \return Positive on success, negative on errors. Possible errors include: \p
364 * E_BAD_TABLE_DESC, \p E_NO_COLUMN_DESC, \p E_NO_COLUMNS, \p
365 * E_BAD_STORAGE_TYPE, \p E_BAD_STORAGE_FLAGS, \p E_BAD_STORAGE_SIZE, \p
366 * E_NO_UNIQUE_RBTREE_COLUMN, \p E_NO_RBTREE_COL.
368 * \sa struct osl_table.
370 int init_table_structure(const struct osl_table_description *desc,
371 struct osl_table **table_ptr)
373 const struct osl_column_description *cd;
374 struct osl_table *t = para_calloc(sizeof(*t));
375 int i, ret = -E_BAD_TABLE_DESC, have_disk_storage_name_column = 0;
379 PARA_DEBUG_LOG("creating table structure for '%s' from table "
380 "description\n", desc->name);
381 ret = -E_NO_COLUMN_DESC;
382 if (!desc->column_descriptions)
385 if (!desc->num_columns)
387 t->columns = para_calloc(desc->num_columns * sizeof(struct osl_column));
389 FOR_EACH_COLUMN(i, t->desc, cd) {
390 enum osl_storage_type st = cd->storage_type;
391 enum osl_storage_flags sf = cd->storage_flags;
392 struct osl_column *col = &t->columns[i];
394 ret = -E_BAD_STORAGE_TYPE;
395 if (st != OSL_MAPPED_STORAGE && st != OSL_DISK_STORAGE
396 && st != OSL_NO_STORAGE)
398 ret = -E_BAD_STORAGE_FLAGS;
399 if (st == OSL_DISK_STORAGE && sf & OSL_RBTREE)
401 ret = -E_BAD_STORAGE_SIZE;
402 if (sf & OSL_FIXED_SIZE && !cd->data_size)
405 case OSL_DISK_STORAGE:
406 t->num_disk_storage_columns++;
408 case OSL_MAPPED_STORAGE:
409 t->num_mapped_columns++;
410 col->index_offset = t->row_index_size;
411 t->row_index_size += 8;
414 col->volatile_num = t->num_volatile_columns;
415 t->num_volatile_columns++;
418 if (sf & OSL_RBTREE) {
419 col->rbtree_num = t->num_rbtrees;
421 if ((sf & OSL_UNIQUE) && (st == OSL_MAPPED_STORAGE)) {
422 if (!have_disk_storage_name_column)
423 t->disk_storage_name_column = i;
424 have_disk_storage_name_column = 1;
428 ret = -E_NO_UNIQUE_RBTREE_COLUMN;
429 if (t->num_disk_storage_columns && !have_disk_storage_name_column)
431 ret = -E_NO_RBTREE_COL;
435 PARA_DEBUG_LOG("OK. Index entry size: %u\n", t->row_index_size);
436 ret = init_column_descriptions(t);
448 * Read the table description from index header.
450 * \param map The memory mapping of the index file.
451 * \param desc The values found in the index header are returned here.
453 * Read the index header, check for the paraslash magic string and the table version number.
454 * Read all information stored in the index header into \a desc.
456 * \return Positive on success, negative on errors.
458 * \sa struct osl_table_description, osl_create_table.
460 int read_table_desc(struct osl_object *map, struct osl_table_description *desc)
462 char *buf = map->data;
464 uint16_t header_size;
467 struct osl_column_description *cd;
469 if (map->size < MIN_INDEX_HEADER_SIZE(1))
470 return -E_SHORT_TABLE;
471 if (strncmp(buf + IDX_PARA_MAGIC, PARA_MAGIC, strlen(PARA_MAGIC)))
473 version = read_u8(buf + IDX_VERSION);
474 if (version < MIN_TABLE_VERSION || version > MAX_TABLE_VERSION)
475 return -E_VERSION_MISMATCH;
476 desc->num_columns = read_u8(buf + IDX_TABLE_FLAGS);
477 desc->flags = read_u8(buf + IDX_TABLE_FLAGS);
478 desc->num_columns = read_u16(buf + IDX_NUM_COLUMNS);
479 PARA_DEBUG_LOG("%u columns\n", desc->num_columns);
480 if (!desc->num_columns)
481 return -E_NO_COLUMNS;
482 header_size = read_u16(buf + IDX_HEADER_SIZE);
483 if (map->size < header_size)
485 desc->column_descriptions = para_calloc(desc->num_columns
486 * sizeof(struct osl_column_description));
487 offset = IDX_COLUMN_DESCRIPTIONS;
488 FOR_EACH_COLUMN(i, desc, cd) {
491 ret = -E_SHORT_TABLE;
492 if (map->size < offset + MIN_IDX_COLUMN_DESCRIPTION_SIZE) {
493 PARA_ERROR_LOG("map size = %zu < %u = offset + min desc size\n",
494 map->size, offset + MIN_IDX_COLUMN_DESCRIPTION_SIZE);
497 cd->storage_type = read_u16(buf + offset + IDX_CD_STORAGE_TYPE);
498 cd->storage_flags = read_u16(buf + offset +
499 IDX_CD_STORAGE_FLAGS);
500 cd->data_size = read_u32(buf + offset + IDX_CD_DATA_SIZE);
501 null_byte = memchr(buf + offset + IDX_CD_NAME, '\0',
502 map->size - offset - IDX_CD_NAME);
503 ret = -E_INDEX_CORRUPTION;
506 cd->name = para_strdup(buf + offset + IDX_CD_NAME);
507 offset += index_column_description_size(cd->name);
509 if (offset != header_size) {
510 ret = -E_INDEX_CORRUPTION;
511 PARA_ERROR_LOG("real header size = %u != %u = stored header size\n",
512 offset, header_size);
517 FOR_EACH_COLUMN(i, desc, cd)
523 * check whether the table description given by \p t->desc matches the on-disk
524 * table structure stored in the index of \a t.
526 static int compare_table_descriptions(struct osl_table *t)
529 struct osl_table_description desc;
530 const struct osl_column_description *cd1, *cd2;
532 /* read the on-disk structure into desc */
533 ret = read_table_desc(&t->index_map, &desc);
536 ret = -E_BAD_TABLE_FLAGS;
537 if (desc.flags != t->desc->flags)
539 ret = -E_BAD_COLUMN_NUM;
540 if (desc.num_columns != t->desc->num_columns)
542 FOR_EACH_COLUMN(i, t->desc, cd1) {
543 cd2 = get_column_description(&desc, i);
544 ret = -E_BAD_STORAGE_TYPE;
545 if (cd1->storage_type != cd2->storage_type)
547 ret = -E_BAD_STORAGE_FLAGS;
548 if (cd1->storage_flags != cd2->storage_flags) {
549 PARA_ERROR_LOG("sf1 = %u != %u = sf2\n",
550 cd1->storage_flags, cd2->storage_flags);
553 ret = -E_BAD_DATA_SIZE;
554 if (cd1->storage_flags & OSL_FIXED_SIZE)
555 if (cd1->data_size != cd2->data_size)
557 ret = -E_BAD_COLUMN_NAME;
558 if (strcmp(cd1->name, cd2->name))
561 PARA_DEBUG_LOG("table description of '%s' matches on-disk data, good\n",
565 FOR_EACH_COLUMN(i, &desc, cd1)
567 free(desc.column_descriptions);
571 static int create_table_index(struct osl_table *t)
573 char *buf, *filename;
575 size_t size = t->index_header_size;
576 const struct osl_column_description *cd;
579 PARA_INFO_LOG("creating %zu byte index for table %s\n", size,
581 buf = para_calloc(size);
582 sprintf(buf + IDX_PARA_MAGIC, "%s", PARA_MAGIC);
583 write_u8(buf + IDX_TABLE_FLAGS, t->desc->flags);
584 write_u8(buf + IDX_DIRTY_FLAG, 0);
585 write_u8(buf + IDX_VERSION, CURRENT_TABLE_VERSION);
586 write_u16(buf + IDX_NUM_COLUMNS, t->desc->num_columns);
587 write_u16(buf + IDX_HEADER_SIZE, t->index_header_size);
588 offset = IDX_COLUMN_DESCRIPTIONS;
589 FOR_EACH_COLUMN(i, t->desc, cd) {
590 write_u16(buf + offset + IDX_CD_STORAGE_TYPE,
592 write_u16(buf + offset + IDX_CD_STORAGE_FLAGS,
594 if (cd->storage_flags & OSL_FIXED_SIZE)
595 write_u32(buf + offset + IDX_CD_DATA_SIZE,
597 strcpy(buf + offset + IDX_CD_NAME, cd->name);
598 offset += index_column_description_size(cd->name);
600 assert(offset = size);
601 filename = index_filename(t->desc);
602 ret = para_write_file(filename, buf, size);
609 * Create a new osl table.
611 * \param desc Pointer to the table description.
615 int osl_create_table(const struct osl_table_description *desc)
617 const struct osl_column_description *cd;
618 char *table_dir = NULL, *filename;
620 int i, ret = init_table_structure(desc, &t);
624 PARA_INFO_LOG("creating %s\n", desc->name);
625 FOR_EACH_COLUMN(i, t->desc, cd) {
626 if (cd->storage_type == OSL_NO_STORAGE)
629 ret = para_mkdir(desc->dir, 0777);
630 if (ret < 0 && !is_errno(-ret, EEXIST))
632 table_dir = make_message("%s/%s", desc->dir,
634 ret = para_mkdir(table_dir, 0777);
638 filename = column_filename(t, i);
639 PARA_INFO_LOG("filename: %s\n", filename);
640 if (cd->storage_type == OSL_MAPPED_STORAGE) {
641 ret = para_open(filename, O_RDWR | O_CREAT | O_EXCL,
650 ret = para_mkdir(filename, 0777);
655 if (t->num_mapped_columns) {
656 ret = create_table_index(t);
668 static int table_is_dirty(struct osl_table *t)
670 char *buf = (char *)t->index_map.data + IDX_DIRTY_FLAG;
671 uint8_t dirty = read_u8(buf) & 0x1;
675 static void mark_table_dirty(struct osl_table *t)
677 char *buf = (char *)t->index_map.data + IDX_DIRTY_FLAG;
678 write_u8(buf, read_u8(buf) | 1);
681 static void mark_table_clean(struct osl_table *t)
683 char *buf = (char *)t->index_map.data + IDX_DIRTY_FLAG;
684 write_u8(buf, read_u8(buf) & 0xfe);
687 static void unmap_column(struct osl_table *t, unsigned col_num)
689 struct osl_object map = t->columns[col_num].data_map;
693 ret = para_munmap(map.data, map.size);
699 * Unmap all mapped files of an osl table.
701 * \param t Pointer to a mapped table.
702 * \param flags Options for unmapping.
704 * \return Positive on success, negative on errors.
706 * \sa map_table(), enum osl_close_flags, para_munmap().
708 int unmap_table(struct osl_table *t, enum osl_close_flags flags)
711 const struct osl_column_description *cd;
714 if (!t->num_mapped_columns) /* can this ever happen? */
716 PARA_DEBUG_LOG("unmapping table '%s'\n", t->desc->name);
717 if (!t->index_map.data)
718 return -E_NOT_MAPPED;
719 if (flags & OSL_MARK_CLEAN)
721 ret = para_munmap(t->index_map.data, t->index_map.size);
724 t->index_map.data = NULL;
727 FOR_EACH_MAPPED_COLUMN(i, t, cd)
732 static int map_column(struct osl_table *t, unsigned col_num)
735 char *filename = column_filename(t, col_num);
737 if (stat(filename, &statbuf) < 0) {
741 if (!(S_IFREG & statbuf.st_mode)) {
745 ret = mmap_full_file(filename, O_RDWR,
746 &t->columns[col_num].data_map.data,
747 &t->columns[col_num].data_map.size,
754 * Map the index file and all columns of type \p OSL_MAPPED_STORAGE into memory.
756 * \param t Pointer to an initialized table structure.
757 * \param flags Mapping options.
759 * \return Negative return value on errors; on success the number of rows
760 * (including invalid rows) is returned.
762 * \sa unmap_table(), enum map_table_flags, osl_open_table(), mmap(2).
764 int map_table(struct osl_table *t, enum map_table_flags flags)
767 const struct osl_column_description *cd;
768 int i = 0, ret, num_rows = 0;
770 if (!t->num_mapped_columns)
772 if (t->index_map.data)
773 return -E_ALREADY_MAPPED;
774 filename = index_filename(t->desc);
775 PARA_DEBUG_LOG("mapping table '%s' (index: %s)\n", t->desc->name, filename);
776 ret = mmap_full_file(filename, flags & MAP_TBL_FL_MAP_RDONLY?
777 O_RDONLY : O_RDWR, &t->index_map.data, &t->index_map.size, NULL);
781 if (flags & MAP_TBL_FL_VERIFY_INDEX) {
782 ret = compare_table_descriptions(t);
787 if (!(flags & MAP_TBL_FL_IGNORE_DIRTY)) {
788 if (table_is_dirty(t)) {
789 PARA_ERROR_LOG("%s is dirty\n", t->desc->name);
794 num_rows = table_num_rows(t);
798 FOR_EACH_MAPPED_COLUMN(i, t, cd) {
799 ret = map_column(t, i);
804 err: /* unmap what is already mapped */
805 for (i--; i >= 0; i--) {
806 struct osl_object map = t->columns[i].data_map;
807 para_munmap(map.data, map.size);
810 para_munmap(t->index_map.data, t->index_map.size);
811 t->index_map.data = NULL;
816 * Retrieve a mapped object by row and column number.
818 * \param t Pointer to an open osl table.
819 * \param col_num Number of the mapped column containing the object to retrieve.
820 * \param row_num Number of the row containing the object to retrieve.
821 * \param obj The result is returned here.
823 * It is considered an error if \a col_num does not refer to a column
824 * of storage type \p OSL_MAPPED_STORAGE.
826 * \return Positive on success, negative on errors. Possible errors include:
827 * \p E_BAD_ROW_NUM, \p E_INVALID_OBJECT.
829 * \sa osl_storage_type.
831 int get_mapped_object(const struct osl_table *t, unsigned col_num,
832 uint32_t row_num, struct osl_object *obj)
834 struct osl_column *col = &t->columns[col_num];
840 if (t->num_rows <= row_num)
841 return -E_BAD_ROW_NUM;
842 ret = get_cell_index(t, row_num, col_num, &cell_index);
845 offset = read_u32(cell_index);
846 obj->size = read_u32(cell_index + 4) - 1;
847 header = col->data_map.data + offset;
848 obj->data = header + 1;
849 if (read_u8(header) == 0xff) {
850 PARA_ERROR_LOG("col %u, size %zu, offset %u\n", col_num,
852 return -E_INVALID_OBJECT;
857 static int search_rbtree(const struct osl_object *obj,
858 const struct osl_table *t, unsigned col_num,
859 struct rb_node **result, struct rb_node ***rb_link)
861 struct osl_column *col = &t->columns[col_num];
862 struct rb_node **new = &col->rbtree.rb_node, *parent = NULL;
863 const struct osl_column_description *cd =
864 get_column_description(t->desc, col_num);
865 enum osl_storage_type st = cd->storage_type;
867 struct osl_row *this_row = get_row_pointer(*new,
870 struct osl_object this_obj;
872 if (st == OSL_MAPPED_STORAGE) {
873 ret = get_mapped_object(t, col_num, this_row->num,
878 this_obj = this_row->volatile_objects[col->volatile_num];
879 ret = cd->compare_function(obj, &this_obj);
882 *result = get_rb_node_pointer(this_row,
887 new = &((*new)->rb_left);
889 new = &((*new)->rb_right);
895 return -E_RB_KEY_NOT_FOUND;
898 static int insert_rbtree(struct osl_table *t, unsigned col_num,
899 const struct osl_row *row, const struct osl_object *obj)
901 struct rb_node *parent, **rb_link;
904 int ret = search_rbtree(obj, t, col_num, &parent, &rb_link);
907 return -E_RB_KEY_EXISTS;
908 rbtree_num = t->columns[col_num].rbtree_num;
909 n = get_rb_node_pointer(row, rbtree_num);
910 rb_link_node(n, parent, rb_link);
911 rb_insert_color(n, &t->columns[col_num].rbtree);
915 static void remove_rb_node(struct osl_table *t, unsigned col_num,
916 const struct osl_row *row)
918 struct osl_column *col = &t->columns[col_num];
919 const struct osl_column_description *cd =
920 get_column_description(t->desc, col_num);
921 enum osl_storage_flags sf = cd->storage_flags;
922 struct rb_node *victim, *splice_out_node, *tmp;
923 if (!(sf & OSL_RBTREE))
926 * Which node is removed/spliced out actually depends on how many
927 * children the victim node has: If it has no children, it gets
928 * deleted. If it has one child, it gets spliced out. If it has two
929 * children, its successor (which has at most a right child) gets
932 victim = get_rb_node_pointer(row, col->rbtree_num);
933 if (victim->rb_left && victim->rb_right)
934 splice_out_node = rb_next(victim);
936 splice_out_node = victim;
937 /* Go up to the root and decrement the size of each node in the path. */
938 for (tmp = splice_out_node; tmp; tmp = rb_parent(tmp))
940 rb_erase(victim, &col->rbtree);
943 static int add_row_to_rbtrees(struct osl_table *t, uint32_t row_num,
944 struct osl_object *volatile_objs, struct osl_row **row_ptr)
948 struct osl_row *row = allocate_row(t->num_rbtrees);
949 const struct osl_column_description *cd;
952 row->volatile_objects = volatile_objs;
953 FOR_EACH_RBTREE_COLUMN(i, t, cd) {
954 if (cd->storage_type == OSL_MAPPED_STORAGE) {
955 struct osl_object obj;
956 ret = get_mapped_object(t, i, row_num, &obj);
959 ret = insert_rbtree(t, i, row, &obj);
960 } else { /* volatile */
961 const struct osl_object *obj
962 = volatile_objs + t->columns[i].volatile_num;
963 ret = insert_rbtree(t, i, row, obj);
971 err: /* rollback changes, i.e. remove added entries from rbtrees */
973 remove_rb_node(t, i--, row);
978 static void free_volatile_objects(const struct osl_table *t,
979 enum osl_close_flags flags)
983 struct osl_column *rb_col;
984 const struct osl_column_description *cd;
986 if (!t->num_volatile_columns)
988 /* find the first rbtree column (any will do) */
989 FOR_EACH_RBTREE_COLUMN(i, t, cd)
991 rb_col = t->columns + i;
992 /* walk that rbtree and free all volatile objects */
993 for (n = rb_first(&rb_col->rbtree); n; n = rb_next(n)) {
994 struct osl_row *r = get_row_pointer(n, rb_col->rbtree_num);
995 if (flags & OSL_FREE_VOLATILE)
996 FOR_EACH_VOLATILE_COLUMN(j, t, cd) {
997 if (cd->storage_flags & OSL_DONT_FREE)
999 free(r->volatile_objects[
1000 t->columns[j].volatile_num].data);
1002 // for (j = 0; j < t->num_volatile_columns; j++)
1003 // free(r->volatile_objects[j].data);
1004 free(r->volatile_objects);
1009 * Erase all rbtree nodes and free resources.
1011 * \param t Pointer to an open osl table.
1013 * This function is called by osl_close_table().
1015 void clear_rbtrees(struct osl_table *t)
1017 const struct osl_column_description *cd;
1018 unsigned i, rbtrees_cleared = 0;
1020 FOR_EACH_RBTREE_COLUMN(i, t, cd) {
1021 struct osl_column *col = &t->columns[i];
1024 for (n = rb_first(&col->rbtree); n;) {
1026 rb_erase(n, &col->rbtree);
1027 if (rbtrees_cleared == t->num_rbtrees) {
1028 r = get_row_pointer(n, col->rbtree_num);
1039 * Close an osl table.
1041 * \param t Pointer to the table to be closed.
1042 * \param flags Options for what should be cleaned up.
1044 * If osl_open_table() succeeds, the resulting table pointer must later be
1045 * passed to this function in order to flush all changes to the file system and
1046 * to free the resources that were allocated by osl_open_table().
1048 * \return Positive on success, negative on errors. Possible errors: \p E_BAD_TABLE,
1049 * errors returned by unmap_table().
1051 * \sa osl_open_table(), unmap_table().
1053 int osl_close_table(struct osl_table *t, enum osl_close_flags flags)
1058 return -E_BAD_TABLE;
1059 free_volatile_objects(t, flags);
1061 ret = unmap_table(t, flags);
1063 PARA_ERROR_LOG("unmap_table failed: %d\n", ret);
1070 * Find out whether the given row number corresponds to an invalid row.
1072 * \param t Pointer to the osl table.
1073 * \param row_num The number of the row in question.
1075 * By definition, a row is considered invalid if all its index entries
1078 * \return Positive if \a row_num corresponds to an invalid row,
1079 * zero if it corresponds to a valid row, negative on errors.
1081 int row_is_invalid(struct osl_table *t, uint32_t row_num)
1084 int i, ret = get_row_index(t, row_num, &row_index);
1088 for (i = 0; i < t->row_index_size; i++) {
1089 if ((unsigned char)row_index[i] != 0xff)
1092 PARA_INFO_LOG("row %d is invalid\n", row_num);
1097 * Invalidate a row of an osl table.
1099 * \param t Pointer to an open osl table.
1100 * \param row_num Number of the row to mark as invalid.
1102 * This function marks each mapped object in the index entry of \a row as
1105 * \return Positive on success, negative on errors.
1107 int mark_row_invalid(struct osl_table *t, uint32_t row_num)
1110 int ret = get_row_index(t, row_num, &row_index);
1114 PARA_INFO_LOG("marking row %d as invalid\n", row_num);
1115 memset(row_index, 0xff, t->row_index_size);
1120 * Initialize all rbtrees and compute number of invalid rows.
1122 * \param t The table containing the rbtrees to be initialized.
1124 * \return Positive on success, negative on errors.
1126 int init_rbtrees(struct osl_table *t)
1129 const struct osl_column_description *cd;
1131 /* create rbtrees */
1132 FOR_EACH_RBTREE_COLUMN(i, t, cd)
1133 t->columns[i].rbtree = RB_ROOT;
1134 /* add valid rows to rbtrees */
1135 t->num_invalid_rows = 0;
1136 for (i = 0; i < t->num_rows; i++) {
1137 ret = row_is_invalid(t, i);
1141 t->num_invalid_rows++;
1144 ret = add_row_to_rbtrees(t, i, NULL, NULL);
1152 * Open an osl table.
1154 * Each osl table must be opened before its data can be accessed.
1156 * \param table_desc Describes the table to be opened.
1157 * \param result Contains a pointer to the open table on success.
1159 * The table description given by \a desc should coincide with the
1160 * description used at creation time.
1164 int osl_open_table(const struct osl_table_description *table_desc,
1165 struct osl_table **result)
1168 struct osl_table *t;
1169 const struct osl_column_description *cd;
1171 PARA_INFO_LOG("opening table %s\n", table_desc->name);
1172 ret = init_table_structure(table_desc, &t);
1175 FOR_EACH_DISK_STORAGE_COLUMN(i, t, cd) {
1176 /* check if directory exists */
1177 char *dirname = column_filename(t, i);
1178 struct stat statbuf;
1179 ret = stat(dirname, &statbuf);
1182 ret = -ERRNO_TO_PARA_ERROR(errno);
1185 ret = -ERRNO_TO_PARA_ERROR(ENOTDIR);
1186 if (!S_ISDIR(statbuf.st_mode))
1189 ret = map_table(t, MAP_TBL_FL_VERIFY_INDEX);
1193 PARA_DEBUG_LOG("num rows: %d\n", t->num_rows);
1194 ret = init_rbtrees(t);
1196 osl_close_table(t, OSL_MARK_CLEAN); /* ignore further errors */
1207 static int create_disk_storage_object_dir(const struct osl_table *t,
1208 unsigned col_num, const char *ds_name)
1213 if (!(t->desc->flags & OSL_LARGE_TABLE))
1215 dirname = disk_storage_dirname(t, col_num, ds_name);
1216 ret = para_mkdir(dirname, 0777);
1218 if (ret < 0 && !is_errno(-ret, EEXIST))
1223 static int write_disk_storage_file(const struct osl_table *t, unsigned col_num,
1224 const struct osl_object *obj, const char *ds_name)
1229 ret = create_disk_storage_object_dir(t, col_num, ds_name);
1232 filename = disk_storage_path(t, col_num, ds_name);
1233 ret = para_write_file(filename, obj->data, obj->size);
1238 static int append_map_file(const struct osl_table *t, unsigned col_num,
1239 const struct osl_object *obj, uint32_t *new_size)
1241 char *filename = column_filename(t, col_num);
1243 char header = 0; /* zero means valid object */
1245 // PARA_DEBUG_LOG("appending %zu + 1 byte\n", obj->size);
1246 ret = append_file(filename, &header, 1, obj->data, obj->size,
1252 static int append_row_index(const struct osl_table *t, char *row_index)
1257 if (!t->num_mapped_columns)
1259 filename = index_filename(t->desc);
1260 ret = append_file(filename, NULL, 0, row_index,
1261 t->row_index_size, NULL);
1267 * A wrapper for truncate(2)
1269 * \param path Name of the regular file to truncate
1270 * \param size Number of bytes to \b shave \b off
1272 * Truncate the regular file named by \a path by \a size bytes.
1274 * \return Positive on success, negative on errors. Possible errors include: \p
1275 * E_STAT, \p E_BAD_SIZE, \p E_TRUNC.
1279 int para_truncate(const char *path, off_t size)
1282 struct stat statbuf;
1285 if (stat(path, &statbuf) < 0)
1288 if (statbuf.st_size < size)
1291 if (truncate(path, statbuf.st_size - size) < 0)
1298 static int truncate_mapped_file(const struct osl_table *t, unsigned col_num,
1301 char *filename = column_filename(t, col_num);
1302 int ret = para_truncate(filename, size);
1307 static int delete_disk_storage_file(const struct osl_table *t, unsigned col_num,
1308 const char *ds_name)
1310 char *dirname, *filename = disk_storage_path(t, col_num, ds_name);
1311 int ret = unlink(filename), err = errno;
1315 return -ERRNO_TO_PARA_ERROR(err);
1316 if (!(t->desc->flags & OSL_LARGE_TABLE))
1318 dirname = disk_storage_dirname(t, col_num, ds_name);
1325 * Add a new row to an osl table and retrieve this row.
1327 * \param t Pointer to an open osl table.
1328 * \param objects Array of objects to be added.
1329 * \param row Result pointer.
1331 * The \a objects parameter must point to an array containing one object per
1332 * column. The order of the objects in the array is given by the table
1333 * description of \a table. Several sanity checks are performed during object
1334 * insertion and the function returns without modifying the table if any of
1335 * these tests fail. In fact, it is atomic in the sense that it either
1336 * succeeds or leaves the table unchanged (i.e. either all or none of the
1337 * objects are added to the table).
1339 * It is considered an error if an object is added to a column with associated
1340 * rbtree if this object is equal to an object already contained in that column
1341 * (i.e. the compare function for the column's rbtree returns zero).
1343 * Possible errors include: \p E_RB_KEY_EXISTS, \p E_BAD_DATA_SIZE.
1345 * \return Positive on success, negative on errors.
1347 * \sa struct osl_table_description, osl_compare_func, osl_add_row().
1349 int osl_add_and_get_row(struct osl_table *t, struct osl_object *objects,
1350 struct osl_row **row)
1353 char *ds_name = NULL;
1354 struct rb_node **rb_parents = NULL, ***rb_links = NULL;
1355 char *new_row_index = NULL;
1356 struct osl_object *volatile_objs = NULL;
1357 const struct osl_column_description *cd;
1360 return -E_BAD_TABLE;
1361 rb_parents = para_malloc(t->num_rbtrees * sizeof(struct rn_node*));
1362 rb_links = para_malloc(t->num_rbtrees * sizeof(struct rn_node**));
1363 if (t->num_mapped_columns)
1364 new_row_index = para_malloc(t->row_index_size);
1365 /* pass 1: sanity checks */
1366 // PARA_DEBUG_LOG("sanity tests: %p:%p\n", objects[0].data,
1367 // objects[1].data);
1368 FOR_EACH_COLUMN(i, t->desc, cd) {
1369 enum osl_storage_type st = cd->storage_type;
1370 enum osl_storage_flags sf = cd->storage_flags;
1372 // ret = -E_NULL_OBJECT;
1375 if (st == OSL_DISK_STORAGE)
1377 if (sf & OSL_RBTREE) {
1378 unsigned rbtree_num = t->columns[i].rbtree_num;
1379 ret = -E_RB_KEY_EXISTS;
1380 // PARA_DEBUG_LOG("checking whether %p exists\n",
1381 // objects[i].data);
1382 if (search_rbtree(objects + i, t, i,
1383 &rb_parents[rbtree_num],
1384 &rb_links[rbtree_num]) > 0)
1387 if (sf & OSL_FIXED_SIZE) {
1388 // PARA_DEBUG_LOG("fixed size. need: %zu, have: %d\n",
1389 // objects[i].size, cd->data_size);
1390 ret = -E_BAD_DATA_SIZE;
1391 if (objects[i].size != cd->data_size)
1395 if (t->num_disk_storage_columns)
1396 ds_name = disk_storage_name_of_object(t,
1397 &objects[t->disk_storage_name_column]);
1398 ret = unmap_table(t, OSL_MARK_CLEAN);
1401 // PARA_DEBUG_LOG("sanity tests passed%s\n", "");
1402 /* pass 2: create data files, append map data */
1403 FOR_EACH_COLUMN(i, t->desc, cd) {
1404 enum osl_storage_type st = cd->storage_type;
1405 if (st == OSL_NO_STORAGE)
1407 if (st == OSL_MAPPED_STORAGE) {
1409 struct osl_column *col = &t->columns[i];
1410 // PARA_DEBUG_LOG("appending object of size %zu\n",
1411 // objects[i].size);
1412 ret = append_map_file(t, i, objects + i, &new_size);
1415 update_cell_index(new_row_index, col, new_size,
1420 ret = write_disk_storage_file(t, i, objects + i, ds_name);
1424 ret = append_row_index(t, new_row_index);
1427 ret = map_table(t, MAP_TBL_FL_VERIFY_INDEX);
1428 if (ret < 0) { /* truncate index and rollback changes */
1429 char *filename = index_filename(t->desc);
1430 para_truncate(filename, t->row_index_size);
1434 /* pass 3: add entry to rbtrees */
1435 if (t->num_volatile_columns) {
1436 volatile_objs = para_calloc(t->num_volatile_columns
1437 * sizeof(struct osl_object));
1438 FOR_EACH_VOLATILE_COLUMN(i, t, cd)
1439 volatile_objs[t->columns[i].volatile_num] = objects[i];
1442 // PARA_DEBUG_LOG("adding new entry as row #%d\n", t->num_rows - 1);
1443 ret = add_row_to_rbtrees(t, t->num_rows - 1, volatile_objs, row);
1446 // PARA_DEBUG_LOG("added new entry as row #%d\n", t->num_rows - 1);
1449 rollback: /* rollback all changes made, ignore further errors */
1450 for (i--; i >= 0; i--) {
1451 cd = get_column_description(t->desc, i);
1452 enum osl_storage_type st = cd->storage_type;
1453 if (st == OSL_NO_STORAGE)
1456 if (st == OSL_MAPPED_STORAGE)
1457 truncate_mapped_file(t, i, objects[i].size);
1458 else /* disk storage */
1459 delete_disk_storage_file(t, i, ds_name);
1461 /* ignore error and return previous error value */
1462 map_table(t, MAP_TBL_FL_VERIFY_INDEX);
1464 free(new_row_index);
1472 * Add a new row to an osl table.
1474 * \param t Same meaning as osl_add_and_get_row().
1475 * \param objects Same meaning as osl_add_and_get_row().
1477 * \return The return value of the underlying call to osl_add_and_get_row().
1479 * This is equivalent to osl_add_and_get_row(t, objects, NULL).
1481 int osl_add_row(struct osl_table *t, struct osl_object *objects)
1483 return osl_add_and_get_row(t, objects, NULL);
1487 * Retrieve an object identified by row and column
1489 * \param t Pointer to an open osl table.
1490 * \param r Pointer to the row.
1491 * \param col_num The column number.
1492 * \param object The result pointer.
1494 * The column determined by \a col_num must be of type \p OSL_MAPPED_STORAGE
1495 * or \p OSL_NO_STORAGE, i.e. no disk storage objects may be retrieved by this
1498 * \return Positive if object was found, negative on errors. Possible errors
1499 * include: \p E_BAD_TABLE, \p E_BAD_STORAGE_TYPE.
1501 * \sa osl_storage_type, osl_open_disk_object().
1503 int osl_get_object(const struct osl_table *t, const struct osl_row *r,
1504 unsigned col_num, struct osl_object *object)
1506 const struct osl_column_description *cd;
1509 return -E_BAD_TABLE;
1510 cd = get_column_description(t->desc, col_num);
1511 /* col must not be disk storage */
1512 if (cd->storage_type == OSL_DISK_STORAGE)
1513 return -E_BAD_STORAGE_TYPE;
1514 if (cd->storage_type == OSL_MAPPED_STORAGE)
1515 return get_mapped_object(t, col_num, r->num, object);
1517 *object = r->volatile_objects[t->columns[col_num].volatile_num];
1521 static int mark_mapped_object_invalid(const struct osl_table *t,
1522 uint32_t row_num, unsigned col_num)
1524 struct osl_object obj;
1526 int ret = get_mapped_object(t, col_num, row_num, &obj);
1537 * Delete a row from an osl table.
1539 * \param t Pointer to an open osl table.
1540 * \param row Pointer to the row to delete.
1542 * This removes all disk storage objects, removes all rbtree nodes, and frees
1543 * all volatile objects belonging to the given row. For mapped columns, the
1544 * data is merely marked invalid and may be pruned from time to time by
1547 * \return Positive on success, negative on errors. Possible errors include:
1548 * \p E_BAD_TABLE, errors returned by osl_get_object().
1550 int osl_del_row(struct osl_table *t, struct osl_row *row)
1552 struct osl_row *r = row;
1554 const struct osl_column_description *cd;
1557 return -E_BAD_TABLE;
1558 PARA_INFO_LOG("deleting row %p\n", row);
1560 if (t->num_disk_storage_columns) {
1562 ret = disk_storage_name_of_row(t, r, &ds_name);
1565 FOR_EACH_DISK_STORAGE_COLUMN(i, t, cd)
1566 delete_disk_storage_file(t, i, ds_name);
1569 FOR_EACH_COLUMN(i, t->desc, cd) {
1570 struct osl_column *col = t->columns + i;
1571 enum osl_storage_type st = cd->storage_type;
1572 remove_rb_node(t, i, r);
1573 if (st == OSL_MAPPED_STORAGE) {
1574 mark_mapped_object_invalid(t, r->num, i);
1577 if (st == OSL_NO_STORAGE && !(cd->storage_flags & OSL_DONT_FREE))
1578 free(r->volatile_objects[col->volatile_num].data);
1580 if (t->num_mapped_columns) {
1581 ret = mark_row_invalid(t, r->num);
1584 t->num_invalid_rows++;
1589 free(r->volatile_objects);
1594 /* test if column has an rbtree */
1595 static int check_rbtree_col(const struct osl_table *t, unsigned col_num,
1596 struct osl_column **col)
1599 return -E_BAD_TABLE;
1600 if (!(get_column_description(t->desc, col_num)->storage_flags & OSL_RBTREE))
1601 return -E_BAD_STORAGE_FLAGS;
1602 *col = t->columns + col_num;
1607 * Get the row that contains the given object.
1609 * \param t Pointer to an open osl table.
1610 * \param col_num The number of the column to be searched.
1611 * \param obj The object to be looked up.
1612 * \param result Points to the row containing \a obj.
1614 * Lookup \a obj in \a t and return the row containing \a obj. The column
1615 * specified by \a col_num must have an associated rbtree.
1617 * \return Positive on success, negative on errors. If an error occurred, \a
1618 * result is set to \p NULL. Possible errors include: \p E_BAD_TABLE, \p
1619 * E_BAD_STORAGE_FLAGS, errors returned by get_mapped_object(), \p
1620 * E_RB_KEY_NOT_FOUND.
1622 * \sa osl_storage_flags
1624 int osl_get_row(const struct osl_table *t, unsigned col_num,
1625 const struct osl_object *obj, struct osl_row **result)
1628 struct rb_node *node;
1629 struct osl_row *row;
1630 struct osl_column *col;
1633 ret = check_rbtree_col(t, col_num, &col);
1636 ret = search_rbtree(obj, t, col_num, &node, NULL);
1639 row = get_row_pointer(node, t->columns[col_num].rbtree_num);
1644 static int rbtree_loop(struct osl_column *col, void *private_data,
1645 osl_rbtree_loop_func *func)
1647 struct rb_node *n, *tmp;
1649 /* this for-loop is safe against removal of an entry */
1650 for (n = rb_first(&col->rbtree), tmp = n? rb_next(n) : NULL;
1652 n = tmp, tmp = tmp? rb_next(tmp) : NULL) {
1653 struct osl_row *r = get_row_pointer(n, col->rbtree_num);
1654 int ret = func(r, private_data);
1661 static int rbtree_loop_reverse(struct osl_column *col, void *private_data,
1662 osl_rbtree_loop_func *func)
1664 struct rb_node *n, *tmp;
1666 /* safe against removal of an entry */
1667 for (n = rb_last(&col->rbtree), tmp = n? rb_prev(n) : NULL;
1669 n = tmp, tmp = tmp? rb_prev(tmp) : NULL) {
1670 struct osl_row *r = get_row_pointer(n, col->rbtree_num);
1671 int ret = func(r, private_data);
1679 * Loop over all nodes in an rbtree.
1681 * \param t Pointer to an open osl table.
1682 * \param col_num The column to use for iterating over the elements.
1683 * \param private_data Pointer that gets passed to \a func.
1684 * \param func The function to be called for each node in the rbtree.
1686 * This function does an in-order walk of the rbtree associated with \a
1687 * col_num. It is an error if the \p OSL_RBTREE flag is not set for this
1688 * column. For each node in the rbtree, the given function \a func is called
1689 * with two pointers as arguments: The first osl_row* argument points to the
1690 * row that contains the object corresponding to the rbtree node currently
1691 * traversed, and the \a private_data pointer is passed verbatim to \a func as the
1692 * second argument. The loop terminates either if \a func returns a negative
1693 * value, or if all nodes of the tree have been visited.
1696 * \return Positive on success, negative on errors. If the termination of the
1697 * loop was caused by \a func returning a negative value, this value is
1700 * \sa osl_storage_flags, osl_rbtree_loop_reverse(), osl_compare_func.
1702 int osl_rbtree_loop(const struct osl_table *t, unsigned col_num,
1703 void *private_data, osl_rbtree_loop_func *func)
1705 struct osl_column *col;
1707 int ret = check_rbtree_col(t, col_num, &col);
1710 return rbtree_loop(col, private_data, func);
1714 * Loop over all nodes in an rbtree in reverse order.
1716 * \param t Identical meaning as in \p osl_rbtree_loop().
1717 * \param col_num Identical meaning as in \p osl_rbtree_loop().
1718 * \param private_data Identical meaning as in \p osl_rbtree_loop().
1719 * \param func Identical meaning as in \p osl_rbtree_loop().
1721 * This function is identical to \p osl_rbtree_loop(), the only difference
1722 * is that the tree is walked in reverse order.
1724 * \return The same return value as \p osl_rbtree_loop().
1726 * \sa osl_rbtree_loop().
1728 int osl_rbtree_loop_reverse(const struct osl_table *t, unsigned col_num,
1729 void *private_data, osl_rbtree_loop_func *func)
1731 struct osl_column *col;
1733 int ret = check_rbtree_col(t, col_num, &col);
1736 return rbtree_loop_reverse(col, private_data, func);
1739 /* TODO: Rollback changes on errors */
1740 static int rename_disk_storage_objects(struct osl_table *t,
1741 struct osl_object *old_obj, struct osl_object *new_obj)
1744 const struct osl_column_description *cd;
1745 char *old_ds_name, *new_ds_name;
1747 if (!t->num_disk_storage_columns)
1748 return 1; /* nothing to do */
1749 if (old_obj->size == new_obj->size && !memcmp(new_obj->data,
1750 old_obj->data, new_obj->size))
1751 return 1; /* object did not change */
1752 old_ds_name = disk_storage_name_of_object(t, old_obj);
1753 new_ds_name = disk_storage_name_of_object(t, new_obj);
1754 FOR_EACH_DISK_STORAGE_COLUMN(i, t, cd) {
1755 char *old_filename, *new_filename;
1756 ret = create_disk_storage_object_dir(t, i, new_ds_name);
1759 old_filename = disk_storage_path(t, i, old_ds_name);
1760 new_filename = disk_storage_path(t, i, new_ds_name);
1761 ret = para_rename(old_filename, new_filename);
1776 * Change an object in an osl table.
1778 * \param t Pointer to an open osl table.
1779 * \param r Pointer to the row containing the object to be updated.
1780 * \param col_num Number of the column containing the object to be updated.
1781 * \param obj Pointer to the replacement object.
1783 * This function gets rid of all references to the old object. This includes
1784 * removal of the rbtree node in case there is an rbtree associated with \a
1785 * col_num. It then inserts \a obj into the table and the rbtree if necessary.
1787 * If the \p OSL_RBTREE flag is set for \a col_num, you \b MUST call this
1788 * function in order to change the contents of an object, even for volatile or
1789 * mapped columns of constant size (which may be updated directly if \p
1790 * OSL_RBTREE is not set). Otherwise the rbtree might become corrupted.
1794 int osl_update_object(struct osl_table *t, const struct osl_row *r,
1795 unsigned col_num, struct osl_object *obj)
1797 struct osl_column *col;
1798 const struct osl_column_description *cd;
1802 return -E_BAD_TABLE;
1803 col = &t->columns[col_num];
1804 cd = get_column_description(t->desc, col_num);
1805 PARA_DEBUG_LOG("updating column %u of %s\n", col_num, t->desc->name);
1806 if (cd->storage_flags & OSL_RBTREE) {
1807 if (search_rbtree(obj, t, col_num, NULL, NULL) > 0)
1808 return -E_RB_KEY_EXISTS;
1810 if (cd->storage_flags & OSL_FIXED_SIZE) {
1811 if (obj->size != cd->data_size)
1812 return -E_BAD_DATA_SIZE;
1814 remove_rb_node(t, col_num, r);
1815 if (cd->storage_type == OSL_NO_STORAGE) { /* TODO: If fixed size, reuse object? */
1816 free(r->volatile_objects[col->volatile_num].data);
1817 r->volatile_objects[col->volatile_num] = *obj;
1818 } else if (cd->storage_type == OSL_DISK_STORAGE) {
1820 ret = disk_storage_name_of_row(t, r, &ds_name);
1823 ret = delete_disk_storage_file(t, col_num, ds_name);
1824 if (ret < 0 && !is_errno(-ret, ENOENT)) {
1828 ret = write_disk_storage_file(t, col_num, obj, ds_name);
1832 } else { /* mapped storage */
1833 struct osl_object old_obj;
1834 ret = get_mapped_object(t, col_num, r->num, &old_obj);
1838 * If the updated column is the disk storage name column, the
1839 * disk storage name changes, so we have to rename all disk
1840 * storage objects accordingly.
1842 if (col_num == t->disk_storage_name_column) {
1843 ret = rename_disk_storage_objects(t, &old_obj, obj);
1847 if (cd->storage_flags & OSL_FIXED_SIZE)
1848 memcpy(old_obj.data, obj->data, cd->data_size);
1849 else { /* TODO: if the size doesn't change, use old space */
1850 uint32_t new_data_map_size;
1852 ret = get_row_index(t, r->num, &row_index);
1855 ret = mark_mapped_object_invalid(t, r->num, col_num);
1858 unmap_column(t, col_num);
1859 ret = append_map_file(t, col_num, obj,
1860 &new_data_map_size);
1863 ret = map_column(t, col_num);
1866 update_cell_index(row_index, col, new_data_map_size,
1870 if (cd->storage_flags & OSL_RBTREE) {
1871 ret = insert_rbtree(t, col_num, r, obj);
1879 * Retrieve an object of type \p OSL_DISK_STORAGE by row and column.
1881 * \param t Pointer to an open osl table.
1882 * \param r Pointer to the row containing the object.
1883 * \param col_num The column number.
1884 * \param obj Points to the result upon successful return.
1886 * For columns of type \p OSL_DISK_STORAGE, this function must be used to
1887 * retrieve one of its containing objects. Afterwards, osl_close_disk_object()
1888 * must be called in order to deallocate the resources.
1890 * \return Positive on success, negative on errors. Possible errors include:
1891 * \p E_BAD_TABLE, \p E_BAD_STORAGE_TYPE, errors returned by osl_get_object().
1893 * \sa osl_get_object(), osl_storage_type, osl_close_disk_object().
1895 int osl_open_disk_object(const struct osl_table *t, const struct osl_row *r,
1896 unsigned col_num, struct osl_object *obj)
1898 const struct osl_column_description *cd;
1899 char *ds_name, *filename;
1903 return -E_BAD_TABLE;
1904 cd = get_column_description(t->desc, col_num);
1905 if (cd->storage_type != OSL_DISK_STORAGE)
1906 return -E_BAD_STORAGE_TYPE;
1908 ret = disk_storage_name_of_row(t, r, &ds_name);
1911 filename = disk_storage_path(t, col_num, ds_name);
1913 PARA_DEBUG_LOG("filename: %s\n", filename);
1914 ret = mmap_full_file(filename, O_RDONLY, &obj->data, &obj->size, NULL);
1920 * Free resources that were allocated during osl_open_disk_object().
1922 * \param obj Pointer to the object previously returned by open_disk_object().
1924 * \return The return value of the underlying call to para_munmap().
1926 * \sa para_munmap().
1928 int osl_close_disk_object(struct osl_object *obj)
1930 return para_munmap(obj->data, obj->size);
1934 * Get the number of rows of the given table.
1936 * \param t Pointer to an open osl table.
1937 * \param num_rows Result is returned here.
1939 * The number of rows returned via \a num_rows excluding any invalid rows.
1941 * \return Positive on success, \p -E_BAD_TABLE if \a t is \p NULL.
1943 int osl_get_num_rows(const struct osl_table *t, unsigned *num_rows)
1946 return -E_BAD_TABLE;
1947 assert(t->num_rows >= t->num_invalid_rows);
1948 *num_rows = t->num_rows - t->num_invalid_rows;
1953 * Get the rank of a row.
1955 * \param t An open osl table.
1956 * \param r The row to get the rank of.
1957 * \param col_num The number of an rbtree column.
1958 * \param rank Result pointer.
1960 * The rank is, by definition, the position of the row in the linear order
1961 * determined by an in-order tree walk of the rbtree associated with column
1962 * number \a col_num of \a table.
1964 * \return Positive on success, negative on errors.
1966 * \sa osl_get_nth_row().
1968 int osl_get_rank(const struct osl_table *t, struct osl_row *r,
1969 unsigned col_num, unsigned *rank)
1971 struct osl_object obj;
1972 struct osl_column *col;
1973 struct rb_node *node;
1974 int ret = check_rbtree_col(t, col_num, &col);
1978 ret = osl_get_object(t, r, col_num, &obj);
1981 ret = search_rbtree(&obj, t, col_num, &node, NULL);
1984 ret = rb_rank(node, rank);
1991 * Get the row with n-th greatest value.
1993 * \param t Pointer to an open osl table.
1994 * \param col_num The column number.
1995 * \param n The rank of the desired row.
1996 * \param result Row is returned here.
1998 * Retrieve the n-th order statistic with respect to the compare function
1999 * of the rbtree column \a col_num. In other words, get that row with
2000 * \a n th greatest value in column \a col_num. It's an error if
2001 * \a col_num is not a rbtree column, or if \a n is larger than the
2002 * number of rows in the table.
2004 * \return Positive on success, negative on errors. Possible errors:
2005 * \p E_BAD_TABLE, \p E_BAD_STORAGE_FLAGS, \p E_RB_KEY_NOT_FOUND.
2007 * \sa osl_storage_flags, osl_compare_func, osl_get_row(),
2008 * osl_rbtree_last_row(), osl_rbtree_first_row(), osl_get_rank().
2010 int osl_get_nth_row(const struct osl_table *t, unsigned col_num,
2011 unsigned n, struct osl_row **result)
2013 struct osl_column *col;
2014 struct rb_node *node;
2019 return -E_RB_KEY_NOT_FOUND;
2020 ret = osl_get_num_rows(t, &num_rows);
2024 return -E_RB_KEY_NOT_FOUND;
2025 ret = check_rbtree_col(t, col_num, &col);
2028 node = rb_nth(col->rbtree.rb_node, n);
2030 return -E_RB_KEY_NOT_FOUND;
2031 *result = get_row_pointer(node, col->rbtree_num);
2036 * Get the row corresponding to the smallest rbtree node of a column.
2038 * \param t An open rbtree table.
2039 * \param col_num The number of the rbtree column.
2040 * \param result A pointer to the first row is returned here.
2042 * The rbtree node of the smallest object (with respect to the corresponding
2043 * compare function) is selected and the row containing this object is
2044 * returned. It is an error if \a col_num refers to a column without an
2045 * associated rbtree.
2047 * \return Positive on success, negative on errors.
2049 * \sa osl_get_nth_row(), osl_rbtree_last_row().
2051 int osl_rbtree_first_row(const struct osl_table *t, unsigned col_num,
2052 struct osl_row **result)
2054 return osl_get_nth_row(t, col_num, 1, result);
2058 * Get the row corresponding to the greatest rbtree node of a column.
2060 * \param t The same meaning as in \p osl_rbtree_first_row().
2061 * \param col_num The same meaning as in \p osl_rbtree_first_row().
2062 * \param result The same meaning as in \p osl_rbtree_first_row().
2064 * This function works just like osl_rbtree_first_row(), the only difference
2065 * is that the row containing the greatest rather than the smallest object is
2068 * \return Positive on success, negative on errors.
2070 * \sa osl_get_nth_row(), osl_rbtree_first_row().
2072 int osl_rbtree_last_row(const struct osl_table *t, unsigned col_num,
2073 struct osl_row **result)
2076 int ret = osl_get_num_rows(t, &num_rows);
2080 return osl_get_nth_row(t, col_num, num_rows, result);