2 * Copyright (C) 2007 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 * Map a file into memory.
162 * \param path Name of the regular file to map.
163 * \param open_mode Either \p O_RDONLY or \p O_RDWR.
164 * \param obj On success, the mapping is returned here.
166 * \return Positive on success, negative on errors. Possible errors include: \p
167 * E_FSTAT, any errors returned by para_open(), \p E_EMPTY, \p E_MMAP.
169 * \sa para_open(), mmap(2).
171 int mmap_full_file(const char *path, int open_mode, struct osl_object *obj)
173 int fd, ret, mmap_prot, mmap_flags;
174 struct stat file_status;
176 if (open_mode == O_RDONLY) {
177 mmap_prot = PROT_READ;
178 mmap_flags = MAP_PRIVATE;
180 mmap_prot = PROT_READ | PROT_WRITE;
181 mmap_flags = MAP_SHARED;
183 ret = para_open(path, open_mode, 0);
188 if (fstat(fd, &file_status) < 0)
190 obj->size = file_status.st_size;
192 PARA_DEBUG_LOG("%s: size %zu\n", path, obj->size);
195 obj->data = mmap(NULL, obj->size, mmap_prot, mmap_flags, fd, 0);
196 if (obj->data == MAP_FAILED) {
208 * Traverse the given directory recursively.
210 * \param dirname The directory to traverse.
211 * \param func The function to call for each entry.
212 * \param private_data Pointer to an arbitrary data structure.
214 * For each regular file under \a dirname, the supplied function \a func is
215 * called. The full path of the regular file and the \a private_data pointer
216 * are passed to \a func. Directories for which the calling process has no
217 * permissions to change to are silently ignored.
219 * \return On success, 1 is returned. Otherwise, this function returns a
220 * negative value which indicates the kind of the error.
222 int for_each_file_in_dir(const char *dirname,
223 int (*func)(const char *, const void *), const void *private_data)
226 struct dirent *entry;
227 int cwd_fd, ret2, ret = para_opendir(dirname, &dir, &cwd_fd);
230 return ret == -E_CHDIR_PERM? 1 : ret;
231 /* scan cwd recursively */
232 while ((entry = readdir(dir))) {
237 if (!strcmp(entry->d_name, "."))
239 if (!strcmp(entry->d_name, ".."))
241 if (lstat(entry->d_name, &s) == -1)
244 if (!S_ISREG(m) && !S_ISDIR(m))
246 tmp = make_message("%s/%s", dirname, entry->d_name);
248 ret = func(tmp, private_data);
255 ret = for_each_file_in_dir(tmp, func, private_data);
263 ret2 = para_fchdir(cwd_fd);
264 if (ret2 < 0 && ret >= 0)
270 static int verify_name(const char *name)
276 if (strchr(name, '/'))
278 if (!strcmp(name, ".."))
280 if (!strcmp(name, "."))
286 * Compare two osl objects pointing to unsigned integers of 32 bit size.
288 * \param obj1 Pointer to the first integer.
289 * \param obj2 Pointer to the second integer.
291 * \return The values required for an osl compare function.
293 * \sa osl_compare_func, osl_hash_compare().
295 int uint32_compare(const struct osl_object *obj1, const struct osl_object *obj2)
297 uint32_t d1 = read_u32((const char *)obj1->data);
298 uint32_t d2 = read_u32((const char *)obj2->data);
308 * Compare two osl objects pointing to hash values.
310 * \param obj1 Pointer to the first hash object.
311 * \param obj2 Pointer to the second hash object.
313 * \return The values required for an osl compare function.
315 * \sa osl_compare_func, uint32_compare().
317 int osl_hash_compare(const struct osl_object *obj1, const struct osl_object *obj2)
319 return hash_compare((HASH_TYPE *)obj1->data, (HASH_TYPE *)obj2->data);
322 static char *disk_storage_dirname(const struct osl_table *t, unsigned col_num,
325 char *dirname, *column_name = column_filename(t, col_num);
327 if (!(t->desc->flags & OSL_LARGE_TABLE))
329 dirname = make_message("%s/%.2s", column_name, ds_name);
334 static char *disk_storage_name_of_object(const struct osl_table *t,
335 const struct osl_object *obj)
337 HASH_TYPE hash[HASH_SIZE];
338 hash_object(obj, hash);
339 return disk_storage_name_of_hash(t, hash);
342 static int disk_storage_name_of_row(const struct osl_table *t,
343 const struct osl_row *row, char **name)
345 struct osl_object obj;
346 int ret = osl_get_object(t, row, t->disk_storage_name_column, &obj);
350 *name = disk_storage_name_of_object(t, &obj);
354 static void column_name_hash(const char *col_name, HASH_TYPE *hash)
356 return hash_function(col_name, strlen(col_name), hash);
359 static int init_column_descriptions(struct osl_table *t)
362 const struct osl_column_description *cd;
364 ret = -E_BAD_TABLE_DESC;
365 ret = verify_name(t->desc->name);
371 /* the size of the index header without column descriptions */
372 t->index_header_size = IDX_COLUMN_DESCRIPTIONS;
373 FOR_EACH_COLUMN(i, t->desc, cd) {
374 struct osl_column *col = t->columns + i;
375 if (cd->storage_flags & OSL_RBTREE) {
376 if (!cd->compare_function)
377 return -E_NO_COMPARE_FUNC;
379 if (cd->storage_type == OSL_NO_STORAGE)
381 ret = -E_NO_COLUMN_NAME;
382 if (!cd->name || !cd->name[0])
384 ret = verify_name(cd->name);
387 t->index_header_size += index_column_description_size(cd->name);
388 column_name_hash(cd->name, col->name_hash);
389 ret = -E_DUPLICATE_COL_NAME;
390 for (j = i + 1; j < t->desc->num_columns; j++) {
391 const char *name2 = get_column_description(t->desc,
393 if (cd->name && name2 && !strcmp(cd->name, name2))
403 * Initialize a struct table from given table description.
405 * \param desc The description of the osl table.
406 * \param table_ptr Result is returned here.
408 * This function performs several sanity checks on \p desc and returns if any
409 * of these tests fail. On success, a struct \p osl_table is allocated and
410 * initialized with data derived from \p desc.
412 * \return Positive on success, negative on errors. Possible errors include: \p
413 * E_BAD_TABLE_DESC, \p E_NO_COLUMN_DESC, \p E_NO_COLUMNS, \p
414 * E_BAD_STORAGE_TYPE, \p E_BAD_STORAGE_FLAGS, \p E_BAD_STORAGE_SIZE, \p
415 * E_NO_UNIQUE_RBTREE_COLUMN, \p E_NO_RBTREE_COL.
417 * \sa struct osl_table.
419 int init_table_structure(const struct osl_table_description *desc,
420 struct osl_table **table_ptr)
422 const struct osl_column_description *cd;
423 struct osl_table *t = para_calloc(sizeof(*t));
424 int i, ret = -E_BAD_TABLE_DESC, have_disk_storage_name_column = 0;
428 PARA_DEBUG_LOG("creating table structure for '%s' from table "
429 "description\n", desc->name);
430 ret = -E_NO_COLUMN_DESC;
431 if (!desc->column_descriptions)
434 if (!desc->num_columns)
436 t->columns = para_calloc(desc->num_columns * sizeof(struct osl_column));
438 FOR_EACH_COLUMN(i, t->desc, cd) {
439 enum osl_storage_type st = cd->storage_type;
440 enum osl_storage_flags sf = cd->storage_flags;
441 struct osl_column *col = &t->columns[i];
443 ret = -E_BAD_STORAGE_TYPE;
444 if (st != OSL_MAPPED_STORAGE && st != OSL_DISK_STORAGE
445 && st != OSL_NO_STORAGE)
447 ret = -E_BAD_STORAGE_FLAGS;
448 if (st == OSL_DISK_STORAGE && sf & OSL_RBTREE)
450 ret = -E_BAD_STORAGE_SIZE;
451 if (sf & OSL_FIXED_SIZE && !cd->data_size)
454 case OSL_DISK_STORAGE:
455 t->num_disk_storage_columns++;
457 case OSL_MAPPED_STORAGE:
458 t->num_mapped_columns++;
459 col->index_offset = t->row_index_size;
460 t->row_index_size += 8;
463 col->volatile_num = t->num_volatile_columns;
464 t->num_volatile_columns++;
467 if (sf & OSL_RBTREE) {
468 col->rbtree_num = t->num_rbtrees;
470 if ((sf & OSL_UNIQUE) && (st == OSL_MAPPED_STORAGE)) {
471 if (!have_disk_storage_name_column)
472 t->disk_storage_name_column = i;
473 have_disk_storage_name_column = 1;
477 ret = -E_NO_UNIQUE_RBTREE_COLUMN;
478 if (t->num_disk_storage_columns && !have_disk_storage_name_column)
480 ret = -E_NO_RBTREE_COL;
484 PARA_DEBUG_LOG("OK. Index entry size: %u\n", t->row_index_size);
485 ret = init_column_descriptions(t);
497 * Read the table description from index header.
499 * \param map The memory mapping of the index file.
500 * \param desc The values found in the index header are returned here.
502 * Read the index header, check for the paraslash magic string and the table version number.
503 * Read all information stored in the index header into \a desc.
505 * \return Positive on success, negative on errors.
507 * \sa struct osl_table_description, osl_create_table.
509 int read_table_desc(struct osl_object *map, struct osl_table_description *desc)
511 char *buf = map->data;
513 uint16_t header_size;
516 struct osl_column_description *cd;
518 if (map->size < MIN_INDEX_HEADER_SIZE(1))
519 return -E_SHORT_TABLE;
520 if (strncmp(buf + IDX_PARA_MAGIC, PARA_MAGIC, strlen(PARA_MAGIC)))
522 version = read_u8(buf + IDX_VERSION);
523 if (version < MIN_TABLE_VERSION || version > MAX_TABLE_VERSION)
524 return -E_VERSION_MISMATCH;
525 desc->num_columns = read_u8(buf + IDX_TABLE_FLAGS);
526 desc->flags = read_u8(buf + IDX_TABLE_FLAGS);
527 desc->num_columns = read_u16(buf + IDX_NUM_COLUMNS);
528 PARA_DEBUG_LOG("%u columns\n", desc->num_columns);
529 if (!desc->num_columns)
530 return -E_NO_COLUMNS;
531 header_size = read_u16(buf + IDX_HEADER_SIZE);
532 if (map->size < header_size)
534 desc->column_descriptions = para_calloc(desc->num_columns
535 * sizeof(struct osl_column_description));
536 offset = IDX_COLUMN_DESCRIPTIONS;
537 FOR_EACH_COLUMN(i, desc, cd) {
540 ret = -E_SHORT_TABLE;
541 if (map->size < offset + MIN_IDX_COLUMN_DESCRIPTION_SIZE) {
542 PARA_ERROR_LOG("map size = %zu < %u = offset + min desc size\n",
543 map->size, offset + MIN_IDX_COLUMN_DESCRIPTION_SIZE);
546 cd->storage_type = read_u16(buf + offset + IDX_CD_STORAGE_TYPE);
547 cd->storage_flags = read_u16(buf + offset +
548 IDX_CD_STORAGE_FLAGS);
549 cd->data_size = read_u32(buf + offset + IDX_CD_DATA_SIZE);
550 null_byte = memchr(buf + offset + IDX_CD_NAME, '\0',
551 map->size - offset - IDX_CD_NAME);
552 ret = -E_INDEX_CORRUPTION;
555 cd->name = para_strdup(buf + offset + IDX_CD_NAME);
556 offset += index_column_description_size(cd->name);
558 if (offset != header_size) {
559 ret = -E_INDEX_CORRUPTION;
560 PARA_ERROR_LOG("real header size = %u != %u = stored header size\n",
561 offset, header_size);
566 FOR_EACH_COLUMN(i, desc, cd)
572 * check whether the table description given by \p t->desc matches the on-disk
573 * table structure stored in the index of \a t.
575 static int compare_table_descriptions(struct osl_table *t)
578 struct osl_table_description desc;
579 const struct osl_column_description *cd1, *cd2;
581 /* read the on-disk structure into desc */
582 ret = read_table_desc(&t->index_map, &desc);
585 ret = -E_BAD_TABLE_FLAGS;
586 if (desc.flags != t->desc->flags)
588 ret = -E_BAD_COLUMN_NUM;
589 if (desc.num_columns != t->desc->num_columns)
591 FOR_EACH_COLUMN(i, t->desc, cd1) {
592 cd2 = get_column_description(&desc, i);
593 ret = -E_BAD_STORAGE_TYPE;
594 if (cd1->storage_type != cd2->storage_type)
596 ret = -E_BAD_STORAGE_FLAGS;
597 if (cd1->storage_flags != cd2->storage_flags) {
598 PARA_ERROR_LOG("sf1 = %u != %u = sf2\n",
599 cd1->storage_flags, cd2->storage_flags);
602 ret = -E_BAD_DATA_SIZE;
603 if (cd1->storage_flags & OSL_FIXED_SIZE)
604 if (cd1->data_size != cd2->data_size)
606 ret = -E_BAD_COLUMN_NAME;
607 if (strcmp(cd1->name, cd2->name))
610 PARA_DEBUG_LOG("table description of '%s' matches on-disk data, good\n",
614 FOR_EACH_COLUMN(i, &desc, cd1)
616 free(desc.column_descriptions);
620 static int create_table_index(struct osl_table *t)
622 char *buf, *filename;
624 size_t size = t->index_header_size;
625 const struct osl_column_description *cd;
628 PARA_INFO_LOG("creating %zu byte index for table %s\n", size,
630 buf = para_calloc(size);
631 sprintf(buf + IDX_PARA_MAGIC, "%s", PARA_MAGIC);
632 write_u8(buf + IDX_TABLE_FLAGS, t->desc->flags);
633 write_u8(buf + IDX_DIRTY_FLAG, 0);
634 write_u8(buf + IDX_VERSION, CURRENT_TABLE_VERSION);
635 write_u16(buf + IDX_NUM_COLUMNS, t->desc->num_columns);
636 write_u16(buf + IDX_HEADER_SIZE, t->index_header_size);
637 offset = IDX_COLUMN_DESCRIPTIONS;
638 FOR_EACH_COLUMN(i, t->desc, cd) {
639 write_u16(buf + offset + IDX_CD_STORAGE_TYPE,
641 write_u16(buf + offset + IDX_CD_STORAGE_FLAGS,
643 if (cd->storage_flags & OSL_FIXED_SIZE)
644 write_u32(buf + offset + IDX_CD_DATA_SIZE,
646 strcpy(buf + offset + IDX_CD_NAME, cd->name);
647 offset += index_column_description_size(cd->name);
649 assert(offset = size);
650 filename = index_filename(t->desc);
651 ret = para_write_file(filename, buf, size);
658 * Create a new osl table.
660 * \param desc Pointer to the table description.
662 * \return Positive on success, negative on errors. Possible errors include: \p
663 * E_BAD_TABLE_DESC, \p E_BAD_DB_DIR, \p E_BAD_NAME, \p E_NO_COMPARE_FUNC, \p
664 * E_NO_COLUMN_NAME, \p E_DUPLICATE_COL_NAME, \p E_MKDIR, any errors returned
667 int osl_create_table(const struct osl_table_description *desc)
669 const struct osl_column_description *cd;
670 char *table_dir = NULL, *filename;
672 int i, ret = init_table_structure(desc, &t);
676 PARA_INFO_LOG("creating %s\n", desc->name);
677 FOR_EACH_COLUMN(i, t->desc, cd) {
678 if (cd->storage_type == OSL_NO_STORAGE)
681 ret = para_mkdir(desc->dir, 0777);
682 if (ret < 0 && !is_errno(-ret, EEXIST))
684 table_dir = make_message("%s/%s", desc->dir,
686 ret = para_mkdir(table_dir, 0777);
690 filename = column_filename(t, i);
691 PARA_INFO_LOG("filename: %s\n", filename);
692 if (cd->storage_type == OSL_MAPPED_STORAGE) {
693 ret = para_open(filename, O_RDWR | O_CREAT | O_EXCL,
702 ret = para_mkdir(filename, 0777);
707 if (t->num_mapped_columns) {
708 ret = create_table_index(t);
720 static int table_is_dirty(struct osl_table *t)
722 char *buf = (char *)t->index_map.data + IDX_DIRTY_FLAG;
723 uint8_t dirty = read_u8(buf) & 0x1;
727 static void mark_table_dirty(struct osl_table *t)
729 char *buf = (char *)t->index_map.data + IDX_DIRTY_FLAG;
730 write_u8(buf, read_u8(buf) | 1);
733 static void mark_table_clean(struct osl_table *t)
735 char *buf = (char *)t->index_map.data + IDX_DIRTY_FLAG;
736 write_u8(buf, read_u8(buf) & 0xfe);
739 static void unmap_column(struct osl_table *t, unsigned col_num)
741 struct osl_object map = t->columns[col_num].data_map;
745 ret = para_munmap(map.data, map.size);
751 * Unmap all mapped files of an osl table.
753 * \param t Pointer to a mapped table.
754 * \param flags Options for unmapping.
756 * \return Positive on success, negative on errors.
758 * \sa map_table(), enum osl_close_flags, para_munmap().
760 int unmap_table(struct osl_table *t, enum osl_close_flags flags)
763 const struct osl_column_description *cd;
766 if (!t->num_mapped_columns) /* can this ever happen? */
768 PARA_DEBUG_LOG("unmapping table '%s'\n", t->desc->name);
769 if (!t->index_map.data)
770 return -E_NOT_MAPPED;
771 if (flags & OSL_MARK_CLEAN)
773 ret = para_munmap(t->index_map.data, t->index_map.size);
776 t->index_map.data = NULL;
779 FOR_EACH_MAPPED_COLUMN(i, t, cd)
784 static int map_column(struct osl_table *t, unsigned col_num)
787 char *filename = column_filename(t, col_num);
789 if (stat(filename, &statbuf) < 0) {
793 if (!(S_IFREG & statbuf.st_mode)) {
797 ret = mmap_full_file(filename, O_RDWR,
798 &t->columns[col_num].data_map);
804 * Map the index file and all columns of type \p OSL_MAPPED_STORAGE into memory.
806 * \param t Pointer to an initialized table structure.
807 * \param flags Mapping options.
809 * \return Negative return value on errors; on success the number of rows
810 * (including invalid rows) is returned.
812 * \sa unmap_table(), enum map_table_flags, osl_open_table(), mmap(2).
814 int map_table(struct osl_table *t, enum map_table_flags flags)
817 const struct osl_column_description *cd;
818 int i = 0, ret, num_rows = 0;
820 if (!t->num_mapped_columns)
822 if (t->index_map.data)
823 return -E_ALREADY_MAPPED;
824 filename = index_filename(t->desc);
825 PARA_DEBUG_LOG("mapping table '%s' (index: %s)\n", t->desc->name, filename);
826 ret = mmap_full_file(filename, flags & MAP_TBL_FL_MAP_RDONLY?
827 O_RDONLY : O_RDWR, &t->index_map);
831 if (flags & MAP_TBL_FL_VERIFY_INDEX) {
832 ret = compare_table_descriptions(t);
837 if (!(flags & MAP_TBL_FL_IGNORE_DIRTY)) {
838 if (table_is_dirty(t)) {
839 PARA_ERROR_LOG("%s is dirty\n", t->desc->name);
844 num_rows = table_num_rows(t);
848 FOR_EACH_MAPPED_COLUMN(i, t, cd) {
849 ret = map_column(t, i);
854 err: /* unmap what is already mapped */
855 for (i--; i >= 0; i--) {
856 struct osl_object map = t->columns[i].data_map;
857 para_munmap(map.data, map.size);
860 para_munmap(t->index_map.data, t->index_map.size);
861 t->index_map.data = NULL;
866 * Retrieve a mapped object by row and column number.
868 * \param t Pointer to an open osl table.
869 * \param col_num Number of the mapped column containing the object to retrieve.
870 * \param row_num Number of the row containing the object to retrieve.
871 * \param obj The result is returned here.
873 * It is considered an error if \a col_num does not refer to a column
874 * of storage type \p OSL_MAPPED_STORAGE.
876 * \return Positive on success, negative on errors. Possible errors include:
877 * \p E_BAD_ROW_NUM, \p E_INVALID_OBJECT.
879 * \sa osl_storage_type.
881 int get_mapped_object(const struct osl_table *t, unsigned col_num,
882 uint32_t row_num, struct osl_object *obj)
884 struct osl_column *col = &t->columns[col_num];
890 if (t->num_rows <= row_num)
891 return -E_BAD_ROW_NUM;
892 ret = get_cell_index(t, row_num, col_num, &cell_index);
895 offset = read_u32(cell_index);
896 obj->size = read_u32(cell_index + 4) - 1;
897 header = col->data_map.data + offset;
898 obj->data = header + 1;
899 if (read_u8(header) == 0xff) {
900 PARA_ERROR_LOG("col %u, size %zu, offset %u\n", col_num,
902 return -E_INVALID_OBJECT;
907 static int search_rbtree(const struct osl_object *obj,
908 const struct osl_table *t, unsigned col_num,
909 struct rb_node **result, struct rb_node ***rb_link)
911 struct osl_column *col = &t->columns[col_num];
912 struct rb_node **new = &col->rbtree.rb_node, *parent = NULL;
913 const struct osl_column_description *cd =
914 get_column_description(t->desc, col_num);
915 enum osl_storage_type st = cd->storage_type;
917 struct osl_row *this_row = get_row_pointer(*new,
920 struct osl_object this_obj;
922 if (st == OSL_MAPPED_STORAGE) {
923 ret = get_mapped_object(t, col_num, this_row->num,
928 this_obj = this_row->volatile_objects[col->volatile_num];
929 ret = cd->compare_function(obj, &this_obj);
932 *result = get_rb_node_pointer(this_row,
937 new = &((*new)->rb_left);
939 new = &((*new)->rb_right);
945 return -E_RB_KEY_NOT_FOUND;
948 static int insert_rbtree(struct osl_table *t, unsigned col_num,
949 const struct osl_row *row, const struct osl_object *obj)
951 struct rb_node *parent, **rb_link;
954 int ret = search_rbtree(obj, t, col_num, &parent, &rb_link);
957 return -E_RB_KEY_EXISTS;
958 rbtree_num = t->columns[col_num].rbtree_num;
959 n = get_rb_node_pointer(row, rbtree_num);
960 rb_link_node(n, parent, rb_link);
961 rb_insert_color(n, &t->columns[col_num].rbtree);
965 static void remove_rb_node(struct osl_table *t, unsigned col_num,
966 const struct osl_row *row)
968 struct osl_column *col = &t->columns[col_num];
969 const struct osl_column_description *cd =
970 get_column_description(t->desc, col_num);
971 enum osl_storage_flags sf = cd->storage_flags;
972 struct rb_node *victim, *splice_out_node, *tmp;
973 if (!(sf & OSL_RBTREE))
976 * Which node is removed/spliced out actually depends on how many
977 * children the victim node has: If it has no children, it gets
978 * deleted. If it has one child, it gets spliced out. If it has two
979 * children, its successor (which has at most a right child) gets
982 victim = get_rb_node_pointer(row, col->rbtree_num);
983 if (victim->rb_left && victim->rb_right)
984 splice_out_node = rb_next(victim);
986 splice_out_node = victim;
987 /* Go up to the root and decrement the size of each node in the path. */
988 for (tmp = splice_out_node; tmp; tmp = rb_parent(tmp))
990 rb_erase(victim, &col->rbtree);
993 static int add_row_to_rbtrees(struct osl_table *t, uint32_t row_num,
994 struct osl_object *volatile_objs, struct osl_row **row_ptr)
998 struct osl_row *row = allocate_row(t->num_rbtrees);
999 const struct osl_column_description *cd;
1002 row->volatile_objects = volatile_objs;
1003 FOR_EACH_RBTREE_COLUMN(i, t, cd) {
1004 if (cd->storage_type == OSL_MAPPED_STORAGE) {
1005 struct osl_object obj;
1006 ret = get_mapped_object(t, i, row_num, &obj);
1009 ret = insert_rbtree(t, i, row, &obj);
1010 } else { /* volatile */
1011 const struct osl_object *obj
1012 = volatile_objs + t->columns[i].volatile_num;
1013 ret = insert_rbtree(t, i, row, obj);
1021 err: /* rollback changes, i.e. remove added entries from rbtrees */
1023 remove_rb_node(t, i--, row);
1028 static void free_volatile_objects(const struct osl_table *t,
1029 enum osl_close_flags flags)
1033 struct osl_column *rb_col;
1034 const struct osl_column_description *cd;
1036 if (!t->num_volatile_columns)
1038 /* find the first rbtree column (any will do) */
1039 FOR_EACH_RBTREE_COLUMN(i, t, cd)
1041 rb_col = t->columns + i;
1042 /* walk that rbtree and free all volatile objects */
1043 for (n = rb_first(&rb_col->rbtree); n; n = rb_next(n)) {
1044 struct osl_row *r = get_row_pointer(n, rb_col->rbtree_num);
1045 if (flags & OSL_FREE_VOLATILE)
1046 for (j = 0; j < t->num_volatile_columns; j++)
1047 free(r->volatile_objects[j].data);
1048 free(r->volatile_objects);
1053 * Erase all rbtree nodes and free resources.
1055 * \param t Pointer to an open osl table.
1057 * This function is called by osl_close_table().
1059 void clear_rbtrees(struct osl_table *t)
1061 const struct osl_column_description *cd;
1062 unsigned i, rbtrees_cleared = 0;
1064 FOR_EACH_RBTREE_COLUMN(i, t, cd) {
1065 struct osl_column *col = &t->columns[i];
1068 for (n = rb_first(&col->rbtree); n;) {
1070 rb_erase(n, &col->rbtree);
1071 if (rbtrees_cleared == t->num_rbtrees) {
1072 r = get_row_pointer(n, col->rbtree_num);
1083 * Close an osl table.
1085 * \param t Pointer to the table to be closed.
1086 * \param flags Options for what should be cleaned up.
1088 * If osl_open_table() succeeds, the resulting table pointer must later be
1089 * passed to this function in order to flush all changes to the file system and
1090 * to free the resources that were allocated by osl_open_table().
1092 * \return Positive on success, negative on errors. Possible errors: \p E_BAD_TABLE,
1093 * errors returned by unmap_table().
1095 * \sa osl_open_table(), unmap_table().
1097 int osl_close_table(struct osl_table *t, enum osl_close_flags flags)
1102 return -E_BAD_TABLE;
1103 free_volatile_objects(t, flags);
1105 ret = unmap_table(t, flags);
1107 PARA_ERROR_LOG("unmap_table failed: %d\n", ret);
1114 * Find out whether the given row number corresponds to an invalid row.
1116 * \param t Pointer to the osl table.
1117 * \param row_num The number of the row in question.
1119 * By definition, a row is considered invalid if all its index entries
1122 * \return Positive if \a row_num corresponds to an invalid row,
1123 * zero if it corresponds to a valid row, negative on errors.
1125 int row_is_invalid(struct osl_table *t, uint32_t row_num)
1128 int i, ret = get_row_index(t, row_num, &row_index);
1132 for (i = 0; i < t->row_index_size; i++) {
1133 if ((unsigned char)row_index[i] != 0xff)
1136 PARA_INFO_LOG("row %d is invalid\n", row_num);
1141 * Invalidate a row of an osl table.
1143 * \param t Pointer to an open osl table.
1144 * \param row_num Number of the row to mark as invalid.
1146 * This function marks each mapped object in the index entry of \a row as
1149 * \return Positive on success, negative on errors.
1151 int mark_row_invalid(struct osl_table *t, uint32_t row_num)
1154 int ret = get_row_index(t, row_num, &row_index);
1158 PARA_INFO_LOG("marking row %d as invalid\n", row_num);
1159 memset(row_index, 0xff, t->row_index_size);
1164 * Initialize all rbtrees and compute number of invalid rows.
1166 * \param t The table containing the rbtrees to be initialized.
1168 * \return Positive on success, negative on errors.
1170 int init_rbtrees(struct osl_table *t)
1173 const struct osl_column_description *cd;
1175 /* create rbtrees */
1176 FOR_EACH_RBTREE_COLUMN(i, t, cd)
1177 t->columns[i].rbtree = RB_ROOT;
1178 /* add valid rows to rbtrees */
1179 t->num_invalid_rows = 0;
1180 for (i = 0; i < t->num_rows; i++) {
1181 ret = row_is_invalid(t, i);
1185 t->num_invalid_rows++;
1188 ret = add_row_to_rbtrees(t, i, NULL, NULL);
1196 * Open an osl table.
1198 * Each osl table must be opened before its data can be accessed.
1200 * \param table_desc Describes the table to be opened.
1201 * \param result Contains a pointer to the open table on success.
1203 * The table description given by \a desc should coincide with the
1204 * description used at creation time.
1206 * \return Positive on success, negative on errors. Possible errors include:
1207 * errors returned by init_table_structure(), \p E_NOENT, \p E_STAT, \p \p
1208 * E_NOTDIR, \p E_BAD_TABLE_DESC, \p E_BAD_DB_DIR, \p E_NO_COMPARE_FUNC, \p
1209 * E_NO_COLUMN_NAME, errors returned by init_rbtrees().
1211 int osl_open_table(const struct osl_table_description *table_desc,
1212 struct osl_table **result)
1215 struct osl_table *t;
1216 const struct osl_column_description *cd;
1218 PARA_INFO_LOG("opening table %s\n", table_desc->name);
1219 ret = init_table_structure(table_desc, &t);
1222 FOR_EACH_DISK_STORAGE_COLUMN(i, t, cd) {
1223 /* check if directory exists */
1224 char *dirname = column_filename(t, i);
1225 struct stat statbuf;
1226 ret = stat(dirname, &statbuf);
1229 ret = -ERRNO_TO_PARA_ERROR(errno);
1232 ret = -ERRNO_TO_PARA_ERROR(ENOTDIR);
1233 if (!S_ISDIR(statbuf.st_mode))
1236 ret = map_table(t, MAP_TBL_FL_VERIFY_INDEX);
1240 PARA_DEBUG_LOG("num rows: %d\n", t->num_rows);
1241 ret = init_rbtrees(t);
1243 osl_close_table(t, OSL_MARK_CLEAN); /* ignore further errors */
1254 static int create_disk_storage_object_dir(const struct osl_table *t,
1255 unsigned col_num, const char *ds_name)
1260 if (!(t->desc->flags & OSL_LARGE_TABLE))
1262 dirname = disk_storage_dirname(t, col_num, ds_name);
1263 ret = para_mkdir(dirname, 0777);
1265 if (ret < 0 && !is_errno(-ret, EEXIST))
1270 static int write_disk_storage_file(const struct osl_table *t, unsigned col_num,
1271 const struct osl_object *obj, const char *ds_name)
1276 ret = create_disk_storage_object_dir(t, col_num, ds_name);
1279 filename = disk_storage_path(t, col_num, ds_name);
1280 ret = para_write_file(filename, obj->data, obj->size);
1285 static int append_map_file(const struct osl_table *t, unsigned col_num,
1286 const struct osl_object *obj, uint32_t *new_size)
1288 char *filename = column_filename(t, col_num);
1290 char header = 0; /* zero means valid object */
1292 // PARA_DEBUG_LOG("appending %zu + 1 byte\n", obj->size);
1293 ret = append_file(filename, &header, 1, obj->data, obj->size,
1299 static int append_row_index(const struct osl_table *t, char *row_index)
1304 if (!t->num_mapped_columns)
1306 filename = index_filename(t->desc);
1307 ret = append_file(filename, NULL, 0, row_index,
1308 t->row_index_size, NULL);
1314 * A wrapper for truncate(2)
1316 * \param path Name of the regular file to truncate
1317 * \param size Number of bytes to \b shave \b off
1319 * Truncate the regular file named by \a path by \a size bytes.
1321 * \return Positive on success, negative on errors. Possible errors include: \p
1322 * E_STAT, \p E_BAD_SIZE, \p E_TRUNC.
1326 int para_truncate(const char *path, off_t size)
1329 struct stat statbuf;
1332 if (stat(path, &statbuf) < 0)
1335 if (statbuf.st_size < size)
1338 if (truncate(path, statbuf.st_size - size) < 0)
1345 static int truncate_mapped_file(const struct osl_table *t, unsigned col_num,
1348 char *filename = column_filename(t, col_num);
1349 int ret = para_truncate(filename, size);
1354 static int delete_disk_storage_file(const struct osl_table *t, unsigned col_num,
1355 const char *ds_name)
1357 char *dirname, *filename = disk_storage_path(t, col_num, ds_name);
1358 int ret = unlink(filename), err = errno;
1362 return -ERRNO_TO_PARA_ERROR(err);
1363 if (!(t->desc->flags & OSL_LARGE_TABLE))
1365 dirname = disk_storage_dirname(t, col_num, ds_name);
1372 * Add a new row to an osl table and retrieve this row.
1374 * \param t Pointer to an open osl table.
1375 * \param objects Array of objects to be added.
1376 * \param row Result pointer.
1378 * The \a objects parameter must point to an array containing one object per
1379 * column. The order of the objects in the array is given by the table
1380 * description of \a table. Several sanity checks are performed during object
1381 * insertion and the function returns without modifying the table if any of
1382 * these tests fail. In fact, it is atomic in the sense that it either
1383 * succeeds or leaves the table unchanged (i.e. either all or none of the
1384 * objects are added to the table).
1386 * It is considered an error if an object is added to a column with associated
1387 * rbtree if this object is equal to an object already contained in that column
1388 * (i.e. the compare function for the column's rbtree returns zero).
1390 * Possible errors include: \p E_RB_KEY_EXISTS, \p E_BAD_DATA_SIZE.
1392 * \return Positive on success, negative on errors.
1394 * \sa struct osl_table_description, osl_compare_func, osl_add_row().
1396 int osl_add_and_get_row(struct osl_table *t, struct osl_object *objects,
1397 struct osl_row **row)
1400 char *ds_name = NULL;
1401 struct rb_node **rb_parents = NULL, ***rb_links = NULL;
1402 char *new_row_index = NULL;
1403 struct osl_object *volatile_objs = NULL;
1404 const struct osl_column_description *cd;
1407 return -E_BAD_TABLE;
1408 rb_parents = para_malloc(t->num_rbtrees * sizeof(struct rn_node*));
1409 rb_links = para_malloc(t->num_rbtrees * sizeof(struct rn_node**));
1410 if (t->num_mapped_columns)
1411 new_row_index = para_malloc(t->row_index_size);
1412 /* pass 1: sanity checks */
1413 // PARA_DEBUG_LOG("sanity tests: %p:%p\n", objects[0].data,
1414 // objects[1].data);
1415 FOR_EACH_COLUMN(i, t->desc, cd) {
1416 enum osl_storage_type st = cd->storage_type;
1417 enum osl_storage_flags sf = cd->storage_flags;
1419 // ret = -E_NULL_OBJECT;
1422 if (st == OSL_DISK_STORAGE)
1424 if (sf & OSL_RBTREE) {
1425 unsigned rbtree_num = t->columns[i].rbtree_num;
1426 ret = -E_RB_KEY_EXISTS;
1427 // PARA_DEBUG_LOG("checking whether %p exists\n",
1428 // objects[i].data);
1429 if (search_rbtree(objects + i, t, i,
1430 &rb_parents[rbtree_num],
1431 &rb_links[rbtree_num]) > 0)
1434 if (sf & OSL_FIXED_SIZE) {
1435 // PARA_DEBUG_LOG("fixed size. need: %zu, have: %d\n",
1436 // objects[i].size, cd->data_size);
1437 ret = -E_BAD_DATA_SIZE;
1438 if (objects[i].size != cd->data_size)
1442 if (t->num_disk_storage_columns)
1443 ds_name = disk_storage_name_of_object(t,
1444 &objects[t->disk_storage_name_column]);
1445 ret = unmap_table(t, OSL_MARK_CLEAN);
1448 // PARA_DEBUG_LOG("sanity tests passed%s\n", "");
1449 /* pass 2: create data files, append map data */
1450 FOR_EACH_COLUMN(i, t->desc, cd) {
1451 enum osl_storage_type st = cd->storage_type;
1452 if (st == OSL_NO_STORAGE)
1454 if (st == OSL_MAPPED_STORAGE) {
1456 struct osl_column *col = &t->columns[i];
1457 // PARA_DEBUG_LOG("appending object of size %zu\n",
1458 // objects[i].size);
1459 ret = append_map_file(t, i, objects + i, &new_size);
1462 update_cell_index(new_row_index, col, new_size,
1467 ret = write_disk_storage_file(t, i, objects + i, ds_name);
1471 ret = append_row_index(t, new_row_index);
1474 ret = map_table(t, MAP_TBL_FL_VERIFY_INDEX);
1475 if (ret < 0) { /* truncate index and rollback changes */
1476 char *filename = index_filename(t->desc);
1477 para_truncate(filename, t->row_index_size);
1481 /* pass 3: add entry to rbtrees */
1482 if (t->num_volatile_columns) {
1483 volatile_objs = para_calloc(t->num_volatile_columns
1484 * sizeof(struct osl_object));
1485 FOR_EACH_VOLATILE_COLUMN(i, t, cd)
1486 volatile_objs[t->columns[i].volatile_num] = objects[i];
1489 // PARA_DEBUG_LOG("adding new entry as row #%d\n", t->num_rows - 1);
1490 ret = add_row_to_rbtrees(t, t->num_rows - 1, volatile_objs, row);
1493 // PARA_DEBUG_LOG("added new entry as row #%d\n", t->num_rows - 1);
1496 rollback: /* rollback all changes made, ignore further errors */
1497 for (i--; i >= 0; i--) {
1498 cd = get_column_description(t->desc, i);
1499 enum osl_storage_type st = cd->storage_type;
1500 if (st == OSL_NO_STORAGE)
1503 if (st == OSL_MAPPED_STORAGE)
1504 truncate_mapped_file(t, i, objects[i].size);
1505 else /* disk storage */
1506 delete_disk_storage_file(t, i, ds_name);
1508 /* ignore error and return previous error value */
1509 map_table(t, MAP_TBL_FL_VERIFY_INDEX);
1511 free(new_row_index);
1519 * Add a new row to an osl table.
1521 * \param t Same meaning as osl_add_and_get_row().
1522 * \param objects Same meaning as osl_add_and_get_row().
1524 * \return The return value of the underlying call to osl_add_and_get_row().
1526 * This is equivalent to osl_add_and_get_row(t, objects, NULL).
1528 int osl_add_row(struct osl_table *t, struct osl_object *objects)
1530 return osl_add_and_get_row(t, objects, NULL);
1534 * Retrieve an object identified by row and column
1536 * \param t Pointer to an open osl table.
1537 * \param r Pointer to the row.
1538 * \param col_num The column number.
1539 * \param object The result pointer.
1541 * The column determined by \a col_num must be of type \p OSL_MAPPED_STORAGE
1542 * or \p OSL_NO_STORAGE, i.e. no disk storage objects may be retrieved by this
1545 * \return Positive if object was found, negative on errors. Possible errors
1546 * include: \p E_BAD_TABLE, \p E_BAD_STORAGE_TYPE.
1548 * \sa osl_storage_type, osl_open_disk_object().
1550 int osl_get_object(const struct osl_table *t, const struct osl_row *r,
1551 unsigned col_num, struct osl_object *object)
1553 const struct osl_column_description *cd;
1556 return -E_BAD_TABLE;
1557 cd = get_column_description(t->desc, col_num);
1558 /* col must not be disk storage */
1559 if (cd->storage_type == OSL_DISK_STORAGE)
1560 return -E_BAD_STORAGE_TYPE;
1561 if (cd->storage_type == OSL_MAPPED_STORAGE)
1562 return get_mapped_object(t, col_num, r->num, object);
1564 *object = r->volatile_objects[t->columns[col_num].volatile_num];
1568 static int mark_mapped_object_invalid(const struct osl_table *t,
1569 uint32_t row_num, unsigned col_num)
1571 struct osl_object obj;
1573 int ret = get_mapped_object(t, col_num, row_num, &obj);
1584 * Delete a row from an osl table.
1586 * \param t Pointer to an open osl table.
1587 * \param row Pointer to the row to delete.
1589 * This removes all disk storage objects, removes all rbtree nodes, and frees
1590 * all volatile objects belonging to the given row. For mapped columns, the
1591 * data is merely marked invalid and may be pruned from time to time by
1594 * \return Positive on success, negative on errors. Possible errors include:
1595 * \p E_BAD_TABLE, errors returned by osl_get_object().
1597 int osl_del_row(struct osl_table *t, struct osl_row *row)
1599 struct osl_row *r = row;
1601 const struct osl_column_description *cd;
1604 return -E_BAD_TABLE;
1605 PARA_INFO_LOG("deleting row %p\n", row);
1607 if (t->num_disk_storage_columns) {
1609 ret = disk_storage_name_of_row(t, r, &ds_name);
1612 FOR_EACH_DISK_STORAGE_COLUMN(i, t, cd)
1613 delete_disk_storage_file(t, i, ds_name);
1616 FOR_EACH_COLUMN(i, t->desc, cd) {
1617 struct osl_column *col = t->columns + i;
1618 enum osl_storage_type st = cd->storage_type;
1619 remove_rb_node(t, i, r);
1620 if (st == OSL_MAPPED_STORAGE) {
1621 mark_mapped_object_invalid(t, r->num, i);
1624 if (st == OSL_NO_STORAGE)
1625 free(r->volatile_objects[col->volatile_num].data);
1627 if (t->num_mapped_columns) {
1628 ret = mark_row_invalid(t, r->num);
1631 t->num_invalid_rows++;
1636 free(r->volatile_objects);
1641 /* test if column has an rbtree */
1642 static int check_rbtree_col(const struct osl_table *t, unsigned col_num,
1643 struct osl_column **col)
1646 return -E_BAD_TABLE;
1647 if (!(get_column_description(t->desc, col_num)->storage_flags & OSL_RBTREE))
1648 return -E_BAD_STORAGE_FLAGS;
1649 *col = t->columns + col_num;
1654 * Get the row that contains the given object.
1656 * \param t Pointer to an open osl table.
1657 * \param col_num The number of the column to be searched.
1658 * \param obj The object to be looked up.
1659 * \param result Points to the row containing \a obj.
1661 * Lookup \a obj in \a t and return the row containing \a obj. The column
1662 * specified by \a col_num must have an associated rbtree.
1664 * \return Positive on success, negative on errors. If an error occurred, \a
1665 * result is set to \p NULL. Possible errors include: \p E_BAD_TABLE, \p
1666 * E_BAD_STORAGE_FLAGS, errors returned by get_mapped_object(), \p
1667 * E_RB_KEY_NOT_FOUND.
1669 * \sa osl_storage_flags
1671 int osl_get_row(const struct osl_table *t, unsigned col_num,
1672 const struct osl_object *obj, struct osl_row **result)
1675 struct rb_node *node;
1676 struct osl_row *row;
1677 struct osl_column *col;
1680 ret = check_rbtree_col(t, col_num, &col);
1683 ret = search_rbtree(obj, t, col_num, &node, NULL);
1686 row = get_row_pointer(node, t->columns[col_num].rbtree_num);
1691 static int rbtree_loop(struct osl_column *col, void *private_data,
1692 osl_rbtree_loop_func *func)
1694 struct rb_node *n, *tmp;
1696 /* this for-loop is safe against removal of an entry */
1697 for (n = rb_first(&col->rbtree), tmp = n? rb_next(n) : NULL;
1699 n = tmp, tmp = tmp? rb_next(tmp) : NULL) {
1700 struct osl_row *r = get_row_pointer(n, col->rbtree_num);
1701 int ret = func(r, private_data);
1708 static int rbtree_loop_reverse(struct osl_column *col, void *private_data,
1709 osl_rbtree_loop_func *func)
1711 struct rb_node *n, *tmp;
1713 /* safe against removal of an entry */
1714 for (n = rb_last(&col->rbtree), tmp = n? rb_prev(n) : NULL;
1716 n = tmp, tmp = tmp? rb_prev(tmp) : NULL) {
1717 struct osl_row *r = get_row_pointer(n, col->rbtree_num);
1718 int ret = func(r, private_data);
1726 * Loop over all nodes in an rbtree.
1728 * \param t Pointer to an open osl table.
1729 * \param col_num The column to use for iterating over the elements.
1730 * \param private_data Pointer that gets passed to \a func.
1731 * \param func The function to be called for each node in the rbtree.
1733 * This function does an in-order walk of the rbtree associated with \a
1734 * col_num. It is an error if the \p OSL_RBTREE flag is not set for this
1735 * column. For each node in the rbtree, the given function \a func is called
1736 * with two pointers as arguments: The first osl_row* argument points to the
1737 * row that contains the object corresponding to the rbtree node currently
1738 * traversed, and the \a private_data pointer is passed verbatim to \a func as the
1739 * second argument. The loop terminates either if \a func returns a negative
1740 * value, or if all nodes of the tree have been visited.
1743 * \return Positive on success, negative on errors. If the termination of the
1744 * loop was caused by \a func returning a negative value, this value is
1747 * \sa osl_storage_flags, osl_rbtree_loop_reverse(), osl_compare_func.
1749 int osl_rbtree_loop(const struct osl_table *t, unsigned col_num,
1750 void *private_data, osl_rbtree_loop_func *func)
1752 struct osl_column *col;
1754 int ret = check_rbtree_col(t, col_num, &col);
1757 return rbtree_loop(col, private_data, func);
1761 * Loop over all nodes in an rbtree in reverse order.
1763 * \param t Identical meaning as in \p osl_rbtree_loop().
1764 * \param col_num Identical meaning as in \p osl_rbtree_loop().
1765 * \param private_data Identical meaning as in \p osl_rbtree_loop().
1766 * \param func Identical meaning as in \p osl_rbtree_loop().
1768 * This function is identical to \p osl_rbtree_loop(), the only difference
1769 * is that the tree is walked in reverse order.
1771 * \return The same return value as \p osl_rbtree_loop().
1773 * \sa osl_rbtree_loop().
1775 int osl_rbtree_loop_reverse(const struct osl_table *t, unsigned col_num,
1776 void *private_data, osl_rbtree_loop_func *func)
1778 struct osl_column *col;
1780 int ret = check_rbtree_col(t, col_num, &col);
1783 return rbtree_loop_reverse(col, private_data, func);
1786 /* TODO: Rollback changes on errors */
1787 static int rename_disk_storage_objects(struct osl_table *t,
1788 struct osl_object *old_obj, struct osl_object *new_obj)
1791 const struct osl_column_description *cd;
1792 char *old_ds_name, *new_ds_name;
1794 if (!t->num_disk_storage_columns)
1795 return 1; /* nothing to do */
1796 if (old_obj->size == new_obj->size && !memcmp(new_obj->data,
1797 old_obj->data, new_obj->size))
1798 return 1; /* object did not change */
1799 old_ds_name = disk_storage_name_of_object(t, old_obj);
1800 new_ds_name = disk_storage_name_of_object(t, new_obj);
1801 FOR_EACH_DISK_STORAGE_COLUMN(i, t, cd) {
1802 char *old_filename, *new_filename;
1803 ret = create_disk_storage_object_dir(t, i, new_ds_name);
1806 old_filename = disk_storage_path(t, i, old_ds_name);
1807 new_filename = disk_storage_path(t, i, new_ds_name);
1808 ret = para_rename(old_filename, new_filename);
1823 * Change an object in an osl table.
1825 * \param t Pointer to an open osl table.
1826 * \param r Pointer to the row containing the object to be updated.
1827 * \param col_num Number of the column containing the object to be updated.
1828 * \param obj Pointer to the replacement object.
1830 * This function gets rid of all references to the old object. This includes
1831 * removal of the rbtree node in case there is an rbtree associated with \a
1832 * col_num. It then inserts \a obj into the table and the rbtree if necessary.
1834 * If the \p OSL_RBTREE flag is set for \a col_num, you \b MUST call this
1835 * function in order to change the contents of an object, even for volatile or
1836 * mapped columns of constant size (which may be updated directly if \p
1837 * OSL_RBTREE is not set). Otherwise the rbtree might become corrupted.
1839 * \return Positive on success, negative on errors. Possible errors include: \p
1840 * E_BAD_TABLE, \p E_RB_KEY_EXISTS, \p E_BAD_SIZE, \p E_NOENT, \p E_UNLINK,
1841 * errors returned by para_write_file(), \p E_MKDIR.
1843 int osl_update_object(struct osl_table *t, const struct osl_row *r,
1844 unsigned col_num, struct osl_object *obj)
1846 struct osl_column *col;
1847 const struct osl_column_description *cd;
1851 return -E_BAD_TABLE;
1852 col = &t->columns[col_num];
1853 cd = get_column_description(t->desc, col_num);
1854 PARA_DEBUG_LOG("updating column %u of %s\n", col_num, t->desc->name);
1855 if (cd->storage_flags & OSL_RBTREE) {
1856 if (search_rbtree(obj, t, col_num, NULL, NULL) > 0)
1857 return -E_RB_KEY_EXISTS;
1859 if (cd->storage_flags & OSL_FIXED_SIZE) {
1860 if (obj->size != cd->data_size)
1861 return -E_BAD_DATA_SIZE;
1863 remove_rb_node(t, col_num, r);
1864 if (cd->storage_type == OSL_NO_STORAGE) { /* TODO: If fixed size, reuse object? */
1865 free(r->volatile_objects[col->volatile_num].data);
1866 r->volatile_objects[col->volatile_num] = *obj;
1867 } else if (cd->storage_type == OSL_DISK_STORAGE) {
1869 ret = disk_storage_name_of_row(t, r, &ds_name);
1872 ret = delete_disk_storage_file(t, col_num, ds_name);
1873 if (ret < 0 && !is_errno(-ret, ENOENT)) {
1877 ret = write_disk_storage_file(t, col_num, obj, ds_name);
1881 } else { /* mapped storage */
1882 struct osl_object old_obj;
1883 ret = get_mapped_object(t, col_num, r->num, &old_obj);
1887 * If the updated column is the disk storage name column, the
1888 * disk storage name changes, so we have to rename all disk
1889 * storage objects accordingly.
1891 if (col_num == t->disk_storage_name_column) {
1892 ret = rename_disk_storage_objects(t, &old_obj, obj);
1896 if (cd->storage_flags & OSL_FIXED_SIZE)
1897 memcpy(old_obj.data, obj->data, cd->data_size);
1898 else { /* TODO: if the size doesn't change, use old space */
1899 uint32_t new_data_map_size;
1901 ret = get_row_index(t, r->num, &row_index);
1904 ret = mark_mapped_object_invalid(t, r->num, col_num);
1907 unmap_column(t, col_num);
1908 ret = append_map_file(t, col_num, obj,
1909 &new_data_map_size);
1912 ret = map_column(t, col_num);
1915 update_cell_index(row_index, col, new_data_map_size,
1919 if (cd->storage_flags & OSL_RBTREE) {
1920 ret = insert_rbtree(t, col_num, r, obj);
1928 * Retrieve an object of type \p OSL_DISK_STORAGE by row and column.
1930 * \param t Pointer to an open osl table.
1931 * \param r Pointer to the row containing the object.
1932 * \param col_num The column number.
1933 * \param obj Points to the result upon successful return.
1935 * For columns of type \p OSL_DISK_STORAGE, this function must be used to
1936 * retrieve one of its containing objects. Afterwards, osl_close_disk_object()
1937 * must be called in order to deallocate the resources.
1939 * \return Positive on success, negative on errors. Possible errors include:
1940 * \p E_BAD_TABLE, \p E_BAD_STORAGE_TYPE, errors returned by osl_get_object().
1942 * \sa osl_get_object(), osl_storage_type, osl_close_disk_object().
1944 int osl_open_disk_object(const struct osl_table *t, const struct osl_row *r,
1945 unsigned col_num, struct osl_object *obj)
1947 const struct osl_column_description *cd;
1948 char *ds_name, *filename;
1952 return -E_BAD_TABLE;
1953 cd = get_column_description(t->desc, col_num);
1954 if (cd->storage_type != OSL_DISK_STORAGE)
1955 return -E_BAD_STORAGE_TYPE;
1957 ret = disk_storage_name_of_row(t, r, &ds_name);
1960 filename = disk_storage_path(t, col_num, ds_name);
1962 PARA_DEBUG_LOG("filename: %s\n", filename);
1963 ret = mmap_full_file(filename, O_RDONLY, obj);
1969 * Free resources that were allocated during osl_open_disk_object().
1971 * \param obj Pointer to the object previously returned by open_disk_object().
1973 * \return The return value of the underlying call to para_munmap().
1975 * \sa para_munmap().
1977 int osl_close_disk_object(struct osl_object *obj)
1979 return para_munmap(obj->data, obj->size);
1983 * Get the number of rows of the given table.
1985 * \param t Pointer to an open osl table.
1986 * \param num_rows Result is returned here.
1988 * The number of rows returned via \a num_rows excluding any invalid rows.
1990 * \return Positive on success, \p -E_BAD_TABLE if \a t is \p NULL.
1992 int osl_get_num_rows(const struct osl_table *t, unsigned *num_rows)
1995 return -E_BAD_TABLE;
1996 assert(t->num_rows >= t->num_invalid_rows);
1997 *num_rows = t->num_rows - t->num_invalid_rows;
2002 * Get the rank of a row.
2004 * \param t An open osl table.
2005 * \param r The row to get the rank of.
2006 * \param col_num The number of an rbtree column.
2007 * \param rank Result pointer.
2009 * The rank is, by definition, the position of the row in the linear order
2010 * determined by an inorder tree walk of the rbtree associated with column
2011 * number \a col_num of \a table.
2013 * \return Positive on success, negative on errors.
2015 * \sa osl_get_nth_row().
2017 int osl_get_rank(const struct osl_table *t, struct osl_row *r,
2018 unsigned col_num, unsigned *rank)
2020 struct osl_object obj;
2021 struct osl_column *col;
2022 struct rb_node *node;
2023 int ret = check_rbtree_col(t, col_num, &col);
2027 ret = osl_get_object(t, r, col_num, &obj);
2030 ret = search_rbtree(&obj, t, col_num, &node, NULL);
2033 ret = rb_rank(node, rank);
2040 * Get the row with n-th greatest value.
2042 * \param t Pointer to an open osl table.
2043 * \param col_num The column number.
2044 * \param n The rank of the desired row.
2045 * \param result Row is returned here.
2047 * Retrieve the n-th order statistic with respect to the compare function
2048 * of the rbtree column \a col_num. In other words, get that row with
2049 * \a n th greatest value in column \a col_num. It's an error if
2050 * \a col_num is not a rbtree column, or if \a n is larger than the
2051 * number of rows in the table.
2053 * \return Positive on success, negative on errors. Possible errors:
2054 * \p E_BAD_TABLE, \p E_BAD_STORAGE_FLAGS, \p E_RB_KEY_NOT_FOUND.
2056 * \sa osl_storage_flags, osl_compare_func, osl_get_row(),
2057 * osl_rbtree_last_row(), osl_rbtree_first_row(), osl_get_rank().
2059 int osl_get_nth_row(const struct osl_table *t, unsigned col_num,
2060 unsigned n, struct osl_row **result)
2062 struct osl_column *col;
2063 struct rb_node *node;
2064 int ret = check_rbtree_col(t, col_num, &col);
2068 node = rb_nth(col->rbtree.rb_node, n);
2070 return -E_RB_KEY_NOT_FOUND;
2071 *result = get_row_pointer(node, col->rbtree_num);
2076 * Get the row corresponding to the smallest rbtree node of a column.
2078 * \param t An open rbtree table.
2079 * \param col_num The number of the rbtree column.
2080 * \param result A pointer to the first row is returned here.
2082 * The rbtree node of the smallest object (with respect to the corresponding
2083 * compare function) is selected and the row containing this object is
2084 * returned. It is an error if \a col_num refers to a column without an
2085 * associated rbtree.
2087 * \return Positive on success, negative on errors.
2089 * \sa osl_get_nth_row(), osl_rbtree_last_row().
2091 int osl_rbtree_first_row(const struct osl_table *t, unsigned col_num,
2092 struct osl_row **result)
2094 return osl_get_nth_row(t, col_num, 1, result);
2098 * Get the row corresponding to the greatest rbtree node of a column.
2100 * \param t The same meaning as in \p osl_rbtree_first_row().
2101 * \param col_num The same meaning as in \p osl_rbtree_first_row().
2102 * \param result The same meaning as in \p osl_rbtree_first_row().
2104 * This function works just like osl_rbtree_first_row(), the only difference
2105 * is that the row containing the greatest rather than the smallest object is
2108 * \return Positive on success, negative on errors.
2110 * \sa osl_get_nth_row(), osl_rbtree_first_row().
2112 int osl_rbtree_last_row(const struct osl_table *t, unsigned col_num,
2113 struct osl_row **result)
2116 int ret = osl_get_num_rows(t, &num_rows);
2120 return osl_get_nth_row(t, col_num, num_rows, result);