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() */
20 * Allocate a sufficiently large string and print into it.
22 * \param fmt A usual format string.
24 * Produce output according to \p fmt. No artificial bound on the length of the
25 * resulting string is imposed.
27 * \return This function either returns a pointer to a string that must be
28 * freed by the caller or \p NULL if memory allocation failed.
32 static __must_check __printf_1_2 __malloc char *make_message(const char *fmt, ...)
36 char *p = malloc(size);
43 /* Try to print in the allocated space. */
45 n = vsnprintf(p, size, fmt, ap);
47 /* If that worked, return the string. */
48 if (n > -1 && n < size)
50 /* Else try again with more space. */
51 if (n > -1) /* glibc 2.1 */
52 size = n + 1; /* precisely what is needed */
54 size *= 2; /* twice the old size */
64 /* Taken from Drepper: How to write shared libraries, Appendix B. */
66 #define MSGSTRFIELD(line) MSGSTRFIELD1(line)
67 #define MSGSTRFIELD1(line) str##line
68 static const union msgstr_t {
70 #define _S(n, s) char MSGSTRFIELD(__LINE__)[sizeof(s)];
80 static const unsigned int errmsgidx[] = {
81 #define _S(n, s) [n] = offsetof(union msgstr_t, MSGSTRFIELD(__LINE__)),
85 __export const char *osl_strerror(int num)
87 if (IS_SYSTEM_ERROR(num))
88 return strerror((num) & ((1 << SYSTEM_ERROR_BIT) - 1));
89 return msgstr.str + errmsgidx[num];
96 * \param fmt Usual format string.
98 * All XXX_LOG() macros use this function.
100 __printf_2_3 void __log(int ll, const char* fmt,...)
113 strftime(str, sizeof(str), "%b %d %H:%M:%S", tm);
114 fprintf(outfd, "%s ", str);
116 vfprintf(outfd, fmt, argp);
121 * A wrapper for lseek(2).
123 * \param fd The file descriptor whose offset is to be to repositioned.
124 * \param offset A value-result parameter.
125 * \param whence Usual repositioning directive.
127 * Reposition the offset of the file descriptor \a fd to the argument \a offset
128 * according to the directive \a whence. Upon successful return, \a offset
129 * contains the resulting offset location as measured in bytes from the
130 * beginning of the file.
132 * \return Positive on success. Otherwise, the function returns \p -E_OSL_LSEEK.
136 static int __lseek(int fd, off_t *offset, int whence)
138 *offset = lseek(fd, *offset, whence);
139 int ret = -E_OSL_LSEEK;
146 * Wrapper for the write system call.
148 * \param fd The file descriptor to write to.
149 * \param buf The buffer to write.
150 * \param size The length of \a buf in bytes.
152 * This function writes out the given buffer and retries if an interrupt
153 * occurred during the write.
155 * \return On success, the number of bytes written is returned, otherwise, the
156 * function returns \p -E_OSL_WRITE.
160 static ssize_t __write(int fd, const void *buf, size_t size)
165 ret = write(fd, buf, size);
166 if ((ret < 0) && (errno == EAGAIN || errno == EINTR))
168 return ret >= 0? ret : -E_OSL_WRITE;
173 * Write the whole buffer to a file descriptor.
175 * \param fd The file descriptor to write to.
176 * \param buf The buffer to write.
177 * \param size The length of \a buf in bytes.
179 * This function writes the given buffer and continues on short writes and
180 * when interrupted by a signal.
182 * \return Positive on success, negative on errors. Possible errors: any
183 * errors returned by para_write().
187 static ssize_t write_all(int fd, const void *buf, size_t size)
189 // DEBUG_LOG("writing %zu bytes\n", size);
192 ssize_t ret = __write(fd, b, size);
193 // DEBUG_LOG("ret: %zd\n", ret);
202 * Open a file, write the given buffer and close the file.
204 * \param filename Full path to the file to open.
205 * \param buf The buffer to write to the file.
206 * \param size The size of \a buf.
210 static int write_file(const char *filename, const void *buf, size_t size)
214 ret = osl_open(filename, O_WRONLY | O_CREAT | O_EXCL, 0644);
218 ret = write_all(fd, buf, size);
227 static int append_file(const char *filename, char *header, size_t header_size,
228 char *data, size_t data_size, uint32_t *new_pos)
232 // DEBUG_LOG("appending %zu + %zu bytes\n", header_size, data_size);
233 ret = osl_open(filename, O_WRONLY | O_CREAT | O_APPEND, 0644);
237 if (header && header_size) {
238 ret = write_all(fd, header, header_size);
242 ret = write_all(fd, data, data_size);
247 ret = __lseek(fd, &offset, SEEK_END);
250 // DEBUG_LOG("new file size: " FMT_OFF_T "\n", offset);
259 static int verify_name(const char *name)
262 return -E_OSL_BAD_NAME;
264 return -E_OSL_BAD_NAME;
265 if (strchr(name, '/'))
266 return -E_OSL_BAD_NAME;
267 if (!strcmp(name, ".."))
268 return -E_OSL_BAD_NAME;
269 if (!strcmp(name, "."))
270 return -E_OSL_BAD_NAME;
275 * Compare two osl objects pointing to unsigned integers of 32 bit size.
277 * \param obj1 Pointer to the first integer.
278 * \param obj2 Pointer to the second integer.
280 * \return The values required for an osl compare function.
282 * \sa osl_compare_func, osl_hash_compare().
284 int uint32_compare(const struct osl_object *obj1, const struct osl_object *obj2)
286 uint32_t d1 = read_u32((const char *)obj1->data);
287 uint32_t d2 = read_u32((const char *)obj2->data);
297 * Compare two osl objects pointing to hash values.
299 * \param obj1 Pointer to the first hash object.
300 * \param obj2 Pointer to the second hash object.
302 * \return The values required for an osl compare function.
304 * \sa osl_compare_func, uint32_compare().
306 int osl_hash_compare(const struct osl_object *obj1, const struct osl_object *obj2)
308 return hash_compare((HASH_TYPE *)obj1->data, (HASH_TYPE *)obj2->data);
311 static char *disk_storage_dirname(const struct osl_table *t, unsigned col_num,
314 char *dirname, *column_name = column_filename(t, col_num);
318 if (!(t->desc->flags & OSL_LARGE_TABLE))
320 dirname = make_message("%s/%.2s", column_name, ds_name);
325 static char *disk_storage_name_of_object(const struct osl_table *t,
326 const struct osl_object *obj)
328 HASH_TYPE hash[HASH_SIZE];
329 hash_object(obj, hash);
330 return disk_storage_name_of_hash(t, hash);
333 static int disk_storage_name_of_row(const struct osl_table *t,
334 const struct osl_row *row, char **name)
336 struct osl_object obj;
337 int ret = osl_get_object(t, row, t->disk_storage_name_column, &obj);
341 *name = disk_storage_name_of_object(t, &obj);
344 return -ERRNO_TO_ERROR(ENOMEM);
347 static void column_name_hash(const char *col_name, HASH_TYPE *hash)
349 hash_function(col_name, strlen(col_name), hash);
352 static int init_column_descriptions(struct osl_table *t)
355 const struct osl_column_description *cd;
357 ret = -E_OSL_BAD_TABLE_DESC;
358 ret = verify_name(t->desc->name);
361 ret = -E_OSL_BAD_DB_DIR;
362 if (!t->desc->dir && (t->num_disk_storage_columns || t->num_mapped_columns))
364 /* the size of the index header without column descriptions */
365 t->index_header_size = IDX_COLUMN_DESCRIPTIONS;
366 FOR_EACH_COLUMN(i, t->desc, cd) {
367 struct osl_column *col = t->columns + i;
368 if (cd->storage_flags & OSL_RBTREE) {
369 if (!cd->compare_function)
370 return -E_OSL_NO_COMPARE_FUNC;
372 if (cd->storage_type == OSL_NO_STORAGE)
374 ret = -E_OSL_NO_COLUMN_NAME;
375 if (!cd->name || !cd->name[0])
377 ret = verify_name(cd->name);
380 t->index_header_size += index_column_description_size(cd->name);
381 column_name_hash(cd->name, col->name_hash);
382 ret = -E_OSL_DUPLICATE_COL_NAME;
383 for (j = i + 1; j < t->desc->num_columns; j++) {
384 const char *name2 = get_column_description(t->desc,
386 if (cd->name && name2 && !strcmp(cd->name, name2))
396 * Initialize a struct table from given table description.
398 * \param desc The description of the osl table.
399 * \param table_ptr Result is returned here.
401 * This function performs several sanity checks on \p desc and returns if any
402 * of these tests fail. On success, a struct \p osl_table is allocated and
403 * initialized with data derived from \p desc.
407 * \sa struct osl_table.
409 int init_table_structure(const struct osl_table_description *desc,
410 struct osl_table **table_ptr)
412 const struct osl_column_description *cd;
413 struct osl_table *t = calloc(1, sizeof(*t));
414 int i, ret = -ERRNO_TO_ERROR(ENOMEM), have_disk_storage_name_column = 0;
418 ret = -E_OSL_BAD_TABLE_DESC;
421 DEBUG_LOG("creating table structure for '%s' from table "
422 "description\n", desc->name);
423 ret = -E_OSL_NO_COLUMN_DESC;
424 if (!desc->column_descriptions)
426 ret = -E_OSL_NO_COLUMNS;
427 if (!desc->num_columns)
429 ret = -ERRNO_TO_ERROR(ENOMEM);
430 t->columns = calloc(desc->num_columns, sizeof(struct osl_column));
434 FOR_EACH_COLUMN(i, t->desc, cd) {
435 enum osl_storage_type st = cd->storage_type;
436 enum osl_storage_flags sf = cd->storage_flags;
437 struct osl_column *col = &t->columns[i];
439 ret = -E_OSL_BAD_STORAGE_TYPE;
440 if (st != OSL_MAPPED_STORAGE && st != OSL_DISK_STORAGE
441 && st != OSL_NO_STORAGE)
443 ret = -E_OSL_BAD_STORAGE_FLAGS;
444 if (st == OSL_DISK_STORAGE && sf & OSL_RBTREE)
446 ret = -E_OSL_BAD_STORAGE_SIZE;
447 if (sf & OSL_FIXED_SIZE && !cd->data_size)
450 case OSL_DISK_STORAGE:
451 t->num_disk_storage_columns++;
453 case OSL_MAPPED_STORAGE:
454 t->num_mapped_columns++;
455 col->index_offset = t->row_index_size;
456 t->row_index_size += 8;
459 col->volatile_num = t->num_volatile_columns;
460 t->num_volatile_columns++;
463 if (sf & OSL_RBTREE) {
464 col->rbtree_num = t->num_rbtrees;
466 if ((sf & OSL_UNIQUE) && (st == OSL_MAPPED_STORAGE)) {
467 if (!have_disk_storage_name_column)
468 t->disk_storage_name_column = i;
469 have_disk_storage_name_column = 1;
473 ret = -E_OSL_NO_UNIQUE_RBTREE_COLUMN;
474 if (t->num_disk_storage_columns && !have_disk_storage_name_column)
476 ret = -E_OSL_NO_RBTREE_COL;
480 DEBUG_LOG("OK. Index entry size: %u\n", t->row_index_size);
481 ret = init_column_descriptions(t);
493 * Read the table description from index header.
495 * \param map The memory mapping of the index file.
496 * \param desc The values found in the index header are returned here.
498 * Read the index header, check for the paraslash magic string and the table version number.
499 * Read all information stored in the index header into \a desc.
503 * \sa struct osl_table_description, osl_create_table.
505 int read_table_desc(struct osl_object *map, struct osl_table_description *desc)
507 char *buf = map->data;
509 uint16_t header_size;
512 struct osl_column_description *cd;
514 if (map->size < MIN_INDEX_HEADER_SIZE(1))
515 return -E_OSL_SHORT_TABLE;
516 if (strncmp(buf + IDX_PARA_MAGIC, PARA_MAGIC, strlen(PARA_MAGIC)))
517 return -E_OSL_NO_MAGIC;
518 version = read_u8(buf + IDX_VERSION);
519 if (version < MIN_TABLE_VERSION || version > MAX_TABLE_VERSION)
520 return -E_OSL_VERSION_MISMATCH;
521 desc->flags = read_u8(buf + IDX_TABLE_FLAGS);
522 desc->num_columns = read_u16(buf + IDX_NUM_COLUMNS);
523 DEBUG_LOG("%u columns\n", desc->num_columns);
524 if (!desc->num_columns)
525 return -E_OSL_NO_COLUMNS;
526 header_size = read_u16(buf + IDX_HEADER_SIZE);
527 if (map->size < header_size)
528 return -E_OSL_BAD_SIZE;
529 desc->column_descriptions = calloc(desc->num_columns,
530 sizeof(struct osl_column_description));
531 if (!desc->column_descriptions)
532 return -ERRNO_TO_ERROR(ENOMEM);
533 offset = IDX_COLUMN_DESCRIPTIONS;
534 FOR_EACH_COLUMN(i, desc, cd) {
537 ret = -E_OSL_SHORT_TABLE;
538 if (map->size < offset + MIN_IDX_COLUMN_DESCRIPTION_SIZE) {
539 ERROR_LOG("map size = %zu < %u = offset + min desc size\n",
540 map->size, offset + MIN_IDX_COLUMN_DESCRIPTION_SIZE);
543 cd->storage_type = read_u16(buf + offset + IDX_CD_STORAGE_TYPE);
544 cd->storage_flags = read_u16(buf + offset +
545 IDX_CD_STORAGE_FLAGS);
546 cd->data_size = read_u32(buf + offset + IDX_CD_DATA_SIZE);
547 null_byte = memchr(buf + offset + IDX_CD_NAME, '\0',
548 map->size - offset - IDX_CD_NAME);
549 ret = -E_OSL_INDEX_CORRUPTION;
552 ret = -ERRNO_TO_ERROR(ENOMEM);
553 cd->name = strdup(buf + offset + IDX_CD_NAME);
556 offset += index_column_description_size(cd->name);
558 if (offset != header_size) {
559 ret = -E_OSL_INDEX_CORRUPTION;
560 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_OSL_BAD_TABLE_FLAGS;
586 if (desc.flags != t->desc->flags)
588 ret = -E_OSL_BAD_COLUMN_NUM;
589 if (desc.num_columns > t->desc->num_columns)
591 if (desc.num_columns < t->desc->num_columns) {
592 struct osl_column_description *cd;
593 unsigned diff = t->desc->num_columns - desc.num_columns;
594 INFO_LOG("extending table by %u volatile columns\n", diff);
595 ret = -ERRNO_TO_ERROR(ENOMEM);
596 desc.column_descriptions = realloc(desc.column_descriptions,
597 t->desc->num_columns * sizeof(struct osl_column_description));
598 if (!desc.column_descriptions)
600 for (i = desc.num_columns; i < t->desc->num_columns; i++) {
601 cd = get_column_description(&desc, i);
602 cd->storage_type = OSL_NO_STORAGE;
605 desc.num_columns += diff;
607 FOR_EACH_COLUMN(i, t->desc, cd1) {
608 cd2 = get_column_description(&desc, i);
609 ret = -E_OSL_BAD_STORAGE_TYPE;
610 if (cd1->storage_type != cd2->storage_type)
612 if (cd1->storage_type == OSL_NO_STORAGE)
614 ret = -E_OSL_BAD_STORAGE_FLAGS;
615 if (cd1->storage_flags != cd2->storage_flags) {
616 ERROR_LOG("sf1 = %u != %u = sf2\n",
617 cd1->storage_flags, cd2->storage_flags);
620 ret = -E_OSL_BAD_DATA_SIZE;
621 if (cd1->storage_flags & OSL_FIXED_SIZE)
622 if (cd1->data_size != cd2->data_size)
624 ret = -E_OSL_BAD_COLUMN_NAME;
625 if (strcmp(cd1->name, cd2->name))
628 DEBUG_LOG("table description of '%s' matches on-disk data, good\n",
632 FOR_EACH_COLUMN(i, &desc, cd1)
634 free(desc.column_descriptions);
638 static int create_table_index(struct osl_table *t)
640 char *buf, *filename;
642 size_t size = t->index_header_size;
643 const struct osl_column_description *cd;
646 INFO_LOG("creating %zu byte index for table %s\n", size,
648 buf = calloc(1, size);
650 return -ERRNO_TO_ERROR(ENOMEM);
651 sprintf(buf + IDX_PARA_MAGIC, "%s", PARA_MAGIC);
652 write_u8(buf + IDX_TABLE_FLAGS, t->desc->flags);
653 write_u8(buf + IDX_DIRTY_FLAG, 0);
654 write_u8(buf + IDX_VERSION, CURRENT_TABLE_VERSION);
655 write_u16(buf + IDX_NUM_COLUMNS, t->num_mapped_columns + t->num_disk_storage_columns);
656 write_u16(buf + IDX_HEADER_SIZE, t->index_header_size);
657 offset = IDX_COLUMN_DESCRIPTIONS;
658 FOR_EACH_COLUMN(i, t->desc, cd) {
659 /* no need to store info about volatile storage */
660 if (cd->storage_type == OSL_NO_STORAGE)
662 write_u16(buf + offset + IDX_CD_STORAGE_TYPE,
664 write_u16(buf + offset + IDX_CD_STORAGE_FLAGS,
666 if (cd->storage_flags & OSL_FIXED_SIZE)
667 write_u32(buf + offset + IDX_CD_DATA_SIZE,
669 strcpy(buf + offset + IDX_CD_NAME, cd->name);
670 offset += index_column_description_size(cd->name);
672 assert(offset = size);
673 filename = index_filename(t->desc);
675 ret = write_file(filename, buf, size);
677 ret = -ERRNO_TO_ERROR(ENOMEM);
684 * Create a new osl table.
686 * \param desc Pointer to the table description.
690 __export int osl_create_table(const struct osl_table_description *desc)
692 const struct osl_column_description *cd;
693 char *table_dir = NULL, *filename;
695 int i, ret = init_table_structure(desc, &t);
699 INFO_LOG("creating %s\n", desc->name);
700 FOR_EACH_COLUMN(i, t->desc, cd) {
701 if (cd->storage_type == OSL_NO_STORAGE)
704 ret = para_mkdir(desc->dir, 0777);
705 if (ret < 0 && !is_errno(-ret, EEXIST))
707 table_dir = make_message("%s/%s", desc->dir,
709 ret = -ERRNO_TO_ERROR(ENOMEM);
712 ret = para_mkdir(table_dir, 0777);
716 ret = -ERRNO_TO_ERROR(ENOMEM);
717 filename = column_filename(t, i);
720 INFO_LOG("filename: %s\n", filename);
721 if (cd->storage_type == OSL_MAPPED_STORAGE) {
722 ret = osl_open(filename, O_RDWR | O_CREAT | O_EXCL,
731 ret = para_mkdir(filename, 0777);
736 if (t->num_mapped_columns) {
737 ret = create_table_index(t);
749 static int table_is_dirty(struct osl_table *t)
751 char *buf = (char *)t->index_map.data + IDX_DIRTY_FLAG;
752 uint8_t dirty = read_u8(buf) & 0x1;
756 static void mark_table_dirty(struct osl_table *t)
758 char *buf = (char *)t->index_map.data + IDX_DIRTY_FLAG;
759 write_u8(buf, read_u8(buf) | 1);
762 static void mark_table_clean(struct osl_table *t)
764 char *buf = (char *)t->index_map.data + IDX_DIRTY_FLAG;
765 write_u8(buf, read_u8(buf) & 0xfe);
768 static void unmap_column(struct osl_table *t, unsigned col_num)
770 struct osl_object map = t->columns[col_num].data_map;
774 ret = para_munmap(map.data, map.size);
780 * Unmap all mapped files of an osl table.
782 * \param t Pointer to a mapped table.
783 * \param flags Options for unmapping.
785 * \return Positive on success, negative on errors.
787 * \sa map_table(), enum osl_close_flags, para_munmap().
789 int unmap_table(struct osl_table *t, enum osl_close_flags flags)
792 const struct osl_column_description *cd;
795 if (!t->num_mapped_columns) /* can this ever happen? */
797 DEBUG_LOG("unmapping table '%s'\n", t->desc->name);
798 if (!t->index_map.data)
799 return -E_OSL_NOT_MAPPED;
800 if (flags & OSL_MARK_CLEAN)
802 ret = para_munmap(t->index_map.data, t->index_map.size);
805 t->index_map.data = NULL;
808 FOR_EACH_MAPPED_COLUMN(i, t, cd)
813 static int map_column(struct osl_table *t, unsigned col_num)
816 char *filename = column_filename(t, col_num);
817 int ret = -E_OSL_STAT;
820 return -ERRNO_TO_ERROR(ENOMEM);
821 if (stat(filename, &statbuf) < 0) {
825 if (!(S_IFREG & statbuf.st_mode)) {
829 ret = mmap_full_file(filename, O_RDWR,
830 &t->columns[col_num].data_map.data,
831 &t->columns[col_num].data_map.size,
838 * Map the index file and all columns of type \p OSL_MAPPED_STORAGE into memory.
840 * \param t Pointer to an initialized table structure.
841 * \param flags Mapping options.
843 * \return Negative return value on errors; on success the number of rows
844 * (including invalid rows) is returned.
846 * \sa unmap_table(), enum map_table_flags, osl_open_table(), mmap(2).
848 int map_table(struct osl_table *t, enum map_table_flags flags)
851 const struct osl_column_description *cd;
852 int i = 0, ret, num_rows = 0;
854 if (!t->num_mapped_columns)
856 if (t->index_map.data)
857 return -E_OSL_ALREADY_MAPPED;
858 filename = index_filename(t->desc);
860 return -ERRNO_TO_ERROR(ENOMEM);
861 DEBUG_LOG("mapping table '%s' (index: %s)\n", t->desc->name, filename);
862 ret = mmap_full_file(filename, flags & MAP_TBL_FL_MAP_RDONLY?
863 O_RDONLY : O_RDWR, &t->index_map.data, &t->index_map.size, NULL);
867 if (flags & MAP_TBL_FL_VERIFY_INDEX) {
868 ret = compare_table_descriptions(t);
873 if (!(flags & MAP_TBL_FL_IGNORE_DIRTY)) {
874 if (table_is_dirty(t)) {
875 ERROR_LOG("%s is dirty\n", t->desc->name);
880 num_rows = table_num_rows(t);
884 FOR_EACH_MAPPED_COLUMN(i, t, cd) {
885 ret = map_column(t, i);
890 err: /* unmap what is already mapped */
891 for (i--; i >= 0; i--) {
892 struct osl_object map = t->columns[i].data_map;
893 para_munmap(map.data, map.size);
896 para_munmap(t->index_map.data, t->index_map.size);
897 t->index_map.data = NULL;
902 * Retrieve a mapped object by row and column number.
904 * \param t Pointer to an open osl table.
905 * \param col_num Number of the mapped column containing the object to retrieve.
906 * \param row_num Number of the row containing the object to retrieve.
907 * \param obj The result is returned here.
909 * It is considered an error if \a col_num does not refer to a column
910 * of storage type \p OSL_MAPPED_STORAGE.
914 * \sa osl_storage_type.
916 int get_mapped_object(const struct osl_table *t, unsigned col_num,
917 uint32_t row_num, struct osl_object *obj)
919 struct osl_column *col = &t->columns[col_num];
925 if (t->num_rows <= row_num)
926 return -E_OSL_BAD_ROW_NUM;
927 ret = get_cell_index(t, row_num, col_num, &cell_index);
930 offset = read_u32(cell_index);
931 obj->size = read_u32(cell_index + 4) - 1;
932 header = col->data_map.data + offset;
933 obj->data = header + 1;
934 if (read_u8(header) == 0xff) {
935 ERROR_LOG("col %u, size %zu, offset %u\n", col_num,
937 return -E_OSL_INVALID_OBJECT;
942 static int search_rbtree(const struct osl_object *obj,
943 const struct osl_table *t, unsigned col_num,
944 struct rb_node **result, struct rb_node ***rb_link)
946 struct osl_column *col = &t->columns[col_num];
947 struct rb_node **new = &col->rbtree.rb_node, *parent = NULL;
948 const struct osl_column_description *cd =
949 get_column_description(t->desc, col_num);
950 enum osl_storage_type st = cd->storage_type;
952 struct osl_row *this_row = get_row_pointer(*new,
955 struct osl_object this_obj;
957 if (st == OSL_MAPPED_STORAGE) {
958 ret = get_mapped_object(t, col_num, this_row->num,
963 this_obj = this_row->volatile_objects[col->volatile_num];
964 ret = cd->compare_function(obj, &this_obj);
967 *result = get_rb_node_pointer(this_row,
972 new = &((*new)->rb_left);
974 new = &((*new)->rb_right);
980 return -E_OSL_RB_KEY_NOT_FOUND;
983 static int insert_rbtree(struct osl_table *t, unsigned col_num,
984 const struct osl_row *row, const struct osl_object *obj)
986 struct rb_node *parent, **rb_link;
989 int ret = search_rbtree(obj, t, col_num, &parent, &rb_link);
992 return -E_OSL_RB_KEY_EXISTS;
993 rbtree_num = t->columns[col_num].rbtree_num;
994 n = get_rb_node_pointer(row, rbtree_num);
995 rb_link_node(n, parent, rb_link);
996 rb_insert_color(n, &t->columns[col_num].rbtree);
1000 static void remove_rb_node(struct osl_table *t, unsigned col_num,
1001 const struct osl_row *row)
1003 struct osl_column *col = &t->columns[col_num];
1004 const struct osl_column_description *cd =
1005 get_column_description(t->desc, col_num);
1006 enum osl_storage_flags sf = cd->storage_flags;
1007 struct rb_node *victim, *splice_out_node, *tmp;
1008 if (!(sf & OSL_RBTREE))
1011 * Which node is removed/spliced out actually depends on how many
1012 * children the victim node has: If it has no children, it gets
1013 * deleted. If it has one child, it gets spliced out. If it has two
1014 * children, its successor (which has at most a right child) gets
1017 victim = get_rb_node_pointer(row, col->rbtree_num);
1018 if (victim->rb_left && victim->rb_right)
1019 splice_out_node = rb_next(victim);
1021 splice_out_node = victim;
1022 /* Go up to the root and decrement the size of each node in the path. */
1023 for (tmp = splice_out_node; tmp; tmp = rb_parent(tmp))
1025 rb_erase(victim, &col->rbtree);
1028 static int add_row_to_rbtrees(struct osl_table *t, uint32_t row_num,
1029 struct osl_object *volatile_objs, struct osl_row **row_ptr)
1033 struct osl_row *row = allocate_row(t->num_rbtrees);
1034 const struct osl_column_description *cd;
1037 return -ERRNO_TO_ERROR(ENOMEM);
1039 row->volatile_objects = volatile_objs;
1040 FOR_EACH_RBTREE_COLUMN(i, t, cd) {
1041 if (cd->storage_type == OSL_MAPPED_STORAGE) {
1042 struct osl_object obj;
1043 ret = get_mapped_object(t, i, row_num, &obj);
1046 ret = insert_rbtree(t, i, row, &obj);
1047 } else { /* volatile */
1048 const struct osl_object *obj
1049 = volatile_objs + t->columns[i].volatile_num;
1050 ret = insert_rbtree(t, i, row, obj);
1058 err: /* rollback changes, i.e. remove added entries from rbtrees */
1060 remove_rb_node(t, i--, row);
1065 static void free_volatile_objects(const struct osl_table *t,
1066 enum osl_close_flags flags)
1070 struct osl_column *rb_col;
1071 const struct osl_column_description *cd;
1073 if (!t->num_volatile_columns)
1075 /* find the first rbtree column (any will do) */
1076 FOR_EACH_RBTREE_COLUMN(i, t, cd)
1078 rb_col = t->columns + i;
1079 /* walk that rbtree and free all volatile objects */
1080 for (n = rb_first(&rb_col->rbtree); n; n = rb_next(n)) {
1081 struct osl_row *r = get_row_pointer(n, rb_col->rbtree_num);
1082 if (flags & OSL_FREE_VOLATILE)
1083 FOR_EACH_VOLATILE_COLUMN(j, t, cd) {
1084 if (cd->storage_flags & OSL_DONT_FREE)
1086 free(r->volatile_objects[
1087 t->columns[j].volatile_num].data);
1089 // for (j = 0; j < t->num_volatile_columns; j++)
1090 // free(r->volatile_objects[j].data);
1091 free(r->volatile_objects);
1096 * Erase all rbtree nodes and free resources.
1098 * \param t Pointer to an open osl table.
1100 * This function is called by osl_close_table().
1102 void clear_rbtrees(struct osl_table *t)
1104 const struct osl_column_description *cd;
1105 unsigned i, rbtrees_cleared = 0;
1107 FOR_EACH_RBTREE_COLUMN(i, t, cd) {
1108 struct osl_column *col = &t->columns[i];
1111 for (n = rb_first(&col->rbtree); n;) {
1113 rb_erase(n, &col->rbtree);
1114 if (rbtrees_cleared == t->num_rbtrees) {
1115 r = get_row_pointer(n, col->rbtree_num);
1126 * Close an osl table.
1128 * \param t Pointer to the table to be closed.
1129 * \param flags Options for what should be cleaned up.
1131 * If osl_open_table() succeeds, the resulting table pointer must later be
1132 * passed to this function in order to flush all changes to the file system and
1133 * to free the resources that were allocated by osl_open_table().
1137 * \sa osl_open_table(), unmap_table().
1139 __export int osl_close_table(struct osl_table *t, enum osl_close_flags flags)
1144 return -E_OSL_BAD_TABLE;
1145 free_volatile_objects(t, flags);
1147 ret = unmap_table(t, flags);
1149 ERROR_LOG("unmap_table failed: %d\n", ret);
1156 * Find out whether the given row number corresponds to an invalid row.
1158 * \param t Pointer to the osl table.
1159 * \param row_num The number of the row in question.
1161 * By definition, a row is considered invalid if all its index entries
1164 * \return Positive if \a row_num corresponds to an invalid row,
1165 * zero if it corresponds to a valid row, negative on errors.
1167 int row_is_invalid(struct osl_table *t, uint32_t row_num)
1170 int i, ret = get_row_index(t, row_num, &row_index);
1174 for (i = 0; i < t->row_index_size; i++) {
1175 if ((unsigned char)row_index[i] != 0xff)
1178 INFO_LOG("row %d is invalid\n", row_num);
1183 * Invalidate a row of an osl table.
1185 * \param t Pointer to an open osl table.
1186 * \param row_num Number of the row to mark as invalid.
1188 * This function marks each mapped object in the index entry of \a row as
1193 int mark_row_invalid(struct osl_table *t, uint32_t row_num)
1196 int ret = get_row_index(t, row_num, &row_index);
1200 INFO_LOG("marking row %d as invalid\n", row_num);
1201 memset(row_index, 0xff, t->row_index_size);
1206 * Initialize all rbtrees and compute number of invalid rows.
1208 * \param t The table containing the rbtrees to be initialized.
1212 int init_rbtrees(struct osl_table *t)
1215 const struct osl_column_description *cd;
1217 /* create rbtrees */
1218 FOR_EACH_RBTREE_COLUMN(i, t, cd)
1219 t->columns[i].rbtree = RB_ROOT;
1220 /* add valid rows to rbtrees */
1221 t->num_invalid_rows = 0;
1222 for (i = 0; i < t->num_rows; i++) {
1223 ret = row_is_invalid(t, i);
1227 t->num_invalid_rows++;
1230 ret = add_row_to_rbtrees(t, i, NULL, NULL);
1238 * Open an osl table.
1240 * Each osl table must be opened before its data can be accessed.
1242 * \param table_desc Describes the table to be opened.
1243 * \param result Contains a pointer to the open table on success.
1245 * The table description given by \a desc should coincide with the
1246 * description used at creation time.
1250 __export int osl_open_table(const struct osl_table_description *table_desc,
1251 struct osl_table **result)
1254 struct osl_table *t;
1255 const struct osl_column_description *cd;
1257 INFO_LOG("opening table %s\n", table_desc->name);
1258 ret = init_table_structure(table_desc, &t);
1261 FOR_EACH_DISK_STORAGE_COLUMN(i, t, cd) {
1262 struct stat statbuf;
1263 char *dirname = column_filename(t, i);
1265 ret = -ERRNO_TO_ERROR(ENOMEM);
1268 /* check if directory exists */
1269 ret = stat(dirname, &statbuf);
1272 ret = -ERRNO_TO_ERROR(errno);
1275 ret = -ERRNO_TO_ERROR(ENOTDIR);
1276 if (!S_ISDIR(statbuf.st_mode))
1279 ret = map_table(t, MAP_TBL_FL_VERIFY_INDEX);
1283 DEBUG_LOG("num rows: %d\n", t->num_rows);
1284 ret = init_rbtrees(t);
1286 osl_close_table(t, OSL_MARK_CLEAN); /* ignore further errors */
1297 static int create_disk_storage_object_dir(const struct osl_table *t,
1298 unsigned col_num, const char *ds_name)
1303 if (!(t->desc->flags & OSL_LARGE_TABLE))
1305 dirname = disk_storage_dirname(t, col_num, ds_name);
1307 return -ERRNO_TO_ERROR(ENOMEM);
1308 ret = para_mkdir(dirname, 0777);
1310 if (ret < 0 && !is_errno(-ret, EEXIST))
1315 static int write_disk_storage_file(const struct osl_table *t, unsigned col_num,
1316 const struct osl_object *obj, const char *ds_name)
1321 ret = create_disk_storage_object_dir(t, col_num, ds_name);
1324 filename = disk_storage_path(t, col_num, ds_name);
1326 return -ERRNO_TO_ERROR(ENOMEM);
1327 ret = write_file(filename, obj->data, obj->size);
1332 static int append_map_file(const struct osl_table *t, unsigned col_num,
1333 const struct osl_object *obj, uint32_t *new_size)
1335 char *filename = column_filename(t, col_num);
1337 char header = 0; /* zero means valid object */
1340 return -ERRNO_TO_ERROR(ENOMEM);
1341 ret = append_file(filename, &header, 1, obj->data, obj->size,
1347 static int append_row_index(const struct osl_table *t, char *row_index)
1352 if (!t->num_mapped_columns)
1354 filename = index_filename(t->desc);
1356 return -ERRNO_TO_ERROR(ENOMEM);
1357 ret = append_file(filename, NULL, 0, row_index,
1358 t->row_index_size, NULL);
1364 * A wrapper for truncate(2)
1366 * \param path Name of the regular file to truncate
1367 * \param size Number of bytes to \b shave \b off
1369 * Truncate the regular file named by \a path by \a size bytes.
1375 int para_truncate(const char *path, off_t size)
1378 struct stat statbuf;
1381 if (stat(path, &statbuf) < 0)
1383 ret = -E_OSL_BAD_SIZE;
1384 if (statbuf.st_size < size)
1387 if (truncate(path, statbuf.st_size - size) < 0)
1394 static int truncate_mapped_file(const struct osl_table *t, unsigned col_num,
1398 char *filename = column_filename(t, col_num);
1401 return -ERRNO_TO_ERROR(ENOMEM);
1402 ret = para_truncate(filename, size);
1407 static int delete_disk_storage_file(const struct osl_table *t, unsigned col_num,
1408 const char *ds_name)
1410 char *dirname, *filename = disk_storage_path(t, col_num, ds_name);
1414 return -ERRNO_TO_ERROR(ENOMEM);
1415 ret = unlink(filename);
1419 return -ERRNO_TO_ERROR(err);
1420 if (!(t->desc->flags & OSL_LARGE_TABLE))
1422 dirname = disk_storage_dirname(t, col_num, ds_name);
1424 return -ERRNO_TO_ERROR(ENOMEM);
1431 * Add a new row to an osl table and retrieve this row.
1433 * \param t Pointer to an open osl table.
1434 * \param objects Array of objects to be added.
1435 * \param row Result pointer.
1437 * The \a objects parameter must point to an array containing one object per
1438 * column. The order of the objects in the array is given by the table
1439 * description of \a table. Several sanity checks are performed during object
1440 * insertion and the function returns without modifying the table if any of
1441 * these tests fail. In fact, it is atomic in the sense that it either
1442 * succeeds or leaves the table unchanged (i.e. either all or none of the
1443 * objects are added to the table).
1445 * It is considered an error if an object is added to a column with associated
1446 * rbtree if this object is equal to an object already contained in that column
1447 * (i.e. the compare function for the column's rbtree returns zero).
1451 * \sa struct osl_table_description, osl_compare_func, osl_add_row().
1453 __export int osl_add_and_get_row(struct osl_table *t, struct osl_object *objects,
1454 struct osl_row **row)
1457 char *ds_name = NULL;
1458 struct rb_node **rb_parents = NULL, ***rb_links = NULL;
1459 char *new_row_index = NULL;
1460 struct osl_object *volatile_objs = NULL;
1461 const struct osl_column_description *cd;
1464 return -E_OSL_BAD_TABLE;
1465 rb_parents = malloc(t->num_rbtrees * sizeof(struct rn_node*));
1467 return -ERRNO_TO_ERROR(ENOMEM);
1468 rb_links = malloc(t->num_rbtrees * sizeof(struct rn_node**));
1471 return -ERRNO_TO_ERROR(ENOMEM);
1473 if (t->num_mapped_columns) {
1474 new_row_index = malloc(t->row_index_size);
1475 if (!new_row_index) {
1478 return -ERRNO_TO_ERROR(ENOMEM);
1481 /* pass 1: sanity checks */
1482 // DEBUG_LOG("sanity tests: %p:%p\n", objects[0].data,
1483 // objects[1].data);
1484 FOR_EACH_COLUMN(i, t->desc, cd) {
1485 enum osl_storage_type st = cd->storage_type;
1486 enum osl_storage_flags sf = cd->storage_flags;
1488 // ret = -E_OSL_NULL_OBJECT;
1491 if (st == OSL_DISK_STORAGE)
1493 if (sf & OSL_RBTREE) {
1494 unsigned rbtree_num = t->columns[i].rbtree_num;
1495 ret = -E_OSL_RB_KEY_EXISTS;
1496 // DEBUG_LOG("checking whether %p exists\n",
1497 // objects[i].data);
1498 if (search_rbtree(objects + i, t, i,
1499 &rb_parents[rbtree_num],
1500 &rb_links[rbtree_num]) > 0)
1503 if (sf & OSL_FIXED_SIZE) {
1504 // DEBUG_LOG("fixed size. need: %zu, have: %d\n",
1505 // objects[i].size, cd->data_size);
1506 ret = -E_OSL_BAD_DATA_SIZE;
1507 if (objects[i].size != cd->data_size)
1511 if (t->num_disk_storage_columns) {
1512 ds_name = disk_storage_name_of_object(t,
1513 &objects[t->disk_storage_name_column]);
1514 ret = -ERRNO_TO_ERROR(ENOMEM);
1518 ret = unmap_table(t, OSL_MARK_CLEAN);
1521 // DEBUG_LOG("sanity tests passed%s\n", "");
1522 /* pass 2: create data files, append map data */
1523 FOR_EACH_COLUMN(i, t->desc, cd) {
1524 enum osl_storage_type st = cd->storage_type;
1525 if (st == OSL_NO_STORAGE)
1527 if (st == OSL_MAPPED_STORAGE) {
1529 struct osl_column *col = &t->columns[i];
1530 // DEBUG_LOG("appending object of size %zu\n",
1531 // objects[i].size);
1532 ret = append_map_file(t, i, objects + i, &new_size);
1535 update_cell_index(new_row_index, col, new_size,
1540 ret = write_disk_storage_file(t, i, objects + i, ds_name);
1544 ret = append_row_index(t, new_row_index);
1547 ret = map_table(t, MAP_TBL_FL_VERIFY_INDEX);
1548 if (ret < 0) { /* truncate index and rollback changes */
1549 char *filename = index_filename(t->desc);
1551 para_truncate(filename, t->row_index_size);
1555 /* pass 3: add entry to rbtrees */
1556 if (t->num_volatile_columns) {
1557 ret = -ERRNO_TO_ERROR(ENOMEM);
1558 volatile_objs = calloc(t->num_volatile_columns,
1559 sizeof(struct osl_object));
1562 FOR_EACH_VOLATILE_COLUMN(i, t, cd)
1563 volatile_objs[t->columns[i].volatile_num] = objects[i];
1566 // DEBUG_LOG("adding new entry as row #%d\n", t->num_rows - 1);
1567 ret = add_row_to_rbtrees(t, t->num_rows - 1, volatile_objs, row);
1570 // DEBUG_LOG("added new entry as row #%d\n", t->num_rows - 1);
1573 rollback: /* rollback all changes made, ignore further errors */
1574 for (i--; i >= 0; i--) {
1575 cd = get_column_description(t->desc, i);
1576 enum osl_storage_type st = cd->storage_type;
1577 if (st == OSL_NO_STORAGE)
1580 if (st == OSL_MAPPED_STORAGE)
1581 truncate_mapped_file(t, i, objects[i].size);
1582 else /* disk storage */
1583 delete_disk_storage_file(t, i, ds_name);
1585 /* ignore error and return previous error value */
1586 map_table(t, MAP_TBL_FL_VERIFY_INDEX);
1588 free(new_row_index);
1596 * Add a new row to an osl table.
1598 * \param t Same meaning as osl_add_and_get_row().
1599 * \param objects Same meaning as osl_add_and_get_row().
1601 * \return The return value of the underlying call to osl_add_and_get_row().
1603 * This is equivalent to osl_add_and_get_row(t, objects, NULL).
1605 __export int osl_add_row(struct osl_table *t, struct osl_object *objects)
1607 return osl_add_and_get_row(t, objects, NULL);
1611 * Retrieve an object identified by row and column
1613 * \param t Pointer to an open osl table.
1614 * \param r Pointer to the row.
1615 * \param col_num The column number.
1616 * \param object The result pointer.
1618 * The column determined by \a col_num must be of type \p OSL_MAPPED_STORAGE
1619 * or \p OSL_NO_STORAGE, i.e. no disk storage objects may be retrieved by this
1624 * \sa osl_storage_type, osl_open_disk_object().
1626 __export int osl_get_object(const struct osl_table *t, const struct osl_row *r,
1627 unsigned col_num, struct osl_object *object)
1629 const struct osl_column_description *cd;
1632 return -E_OSL_BAD_TABLE;
1633 cd = get_column_description(t->desc, col_num);
1634 /* col must not be disk storage */
1635 if (cd->storage_type == OSL_DISK_STORAGE)
1636 return -E_OSL_BAD_STORAGE_TYPE;
1637 if (cd->storage_type == OSL_MAPPED_STORAGE)
1638 return get_mapped_object(t, col_num, r->num, object);
1640 *object = r->volatile_objects[t->columns[col_num].volatile_num];
1644 static int mark_mapped_object_invalid(const struct osl_table *t,
1645 uint32_t row_num, unsigned col_num)
1647 struct osl_object obj;
1649 int ret = get_mapped_object(t, col_num, row_num, &obj);
1660 * Delete a row from an osl table.
1662 * \param t Pointer to an open osl table.
1663 * \param row Pointer to the row to delete.
1665 * This removes all disk storage objects, removes all rbtree nodes, and frees
1666 * all volatile objects belonging to the given row. For mapped columns, the
1667 * data is merely marked invalid and may be pruned from time to time by
1672 __export int osl_del_row(struct osl_table *t, struct osl_row *row)
1674 struct osl_row *r = row;
1676 const struct osl_column_description *cd;
1679 return -E_OSL_BAD_TABLE;
1680 INFO_LOG("deleting row %p\n", row);
1682 if (t->num_disk_storage_columns) {
1684 ret = disk_storage_name_of_row(t, r, &ds_name);
1687 FOR_EACH_DISK_STORAGE_COLUMN(i, t, cd)
1688 delete_disk_storage_file(t, i, ds_name);
1691 FOR_EACH_COLUMN(i, t->desc, cd) {
1692 struct osl_column *col = t->columns + i;
1693 enum osl_storage_type st = cd->storage_type;
1694 remove_rb_node(t, i, r);
1695 if (st == OSL_MAPPED_STORAGE) {
1696 mark_mapped_object_invalid(t, r->num, i);
1699 if (st == OSL_NO_STORAGE && !(cd->storage_flags & OSL_DONT_FREE))
1700 free(r->volatile_objects[col->volatile_num].data);
1702 if (t->num_mapped_columns) {
1703 ret = mark_row_invalid(t, r->num);
1706 t->num_invalid_rows++;
1711 free(r->volatile_objects);
1716 /* test if column has an rbtree */
1717 static int check_rbtree_col(const struct osl_table *t, unsigned col_num,
1718 struct osl_column **col)
1721 return -E_OSL_BAD_TABLE;
1722 if (!(get_column_description(t->desc, col_num)->storage_flags & OSL_RBTREE))
1723 return -E_OSL_BAD_STORAGE_FLAGS;
1724 *col = t->columns + col_num;
1729 * Get the row that contains the given object.
1731 * \param t Pointer to an open osl table.
1732 * \param col_num The number of the column to be searched.
1733 * \param obj The object to be looked up.
1734 * \param result Points to the row containing \a obj.
1736 * Lookup \a obj in \a t and return the row containing \a obj. The column
1737 * specified by \a col_num must have an associated rbtree.
1741 * \sa osl_storage_flags
1743 __export int osl_get_row(const struct osl_table *t, unsigned col_num,
1744 const struct osl_object *obj, struct osl_row **result)
1747 struct rb_node *node;
1748 struct osl_row *row;
1749 struct osl_column *col;
1752 ret = check_rbtree_col(t, col_num, &col);
1755 ret = search_rbtree(obj, t, col_num, &node, NULL);
1758 row = get_row_pointer(node, t->columns[col_num].rbtree_num);
1763 static int rbtree_loop(struct osl_column *col, void *private_data,
1764 osl_rbtree_loop_func *func)
1766 struct rb_node *n, *tmp;
1768 /* this for-loop is safe against removal of an entry */
1769 for (n = rb_first(&col->rbtree), tmp = n? rb_next(n) : NULL;
1771 n = tmp, tmp = tmp? rb_next(tmp) : NULL) {
1772 struct osl_row *r = get_row_pointer(n, col->rbtree_num);
1773 int ret = func(r, private_data);
1780 static int rbtree_loop_reverse(struct osl_column *col, void *private_data,
1781 osl_rbtree_loop_func *func)
1783 struct rb_node *n, *tmp;
1785 /* safe against removal of an entry */
1786 for (n = rb_last(&col->rbtree), tmp = n? rb_prev(n) : NULL;
1788 n = tmp, tmp = tmp? rb_prev(tmp) : NULL) {
1789 struct osl_row *r = get_row_pointer(n, col->rbtree_num);
1790 int ret = func(r, private_data);
1798 * Loop over all nodes in an rbtree.
1800 * \param t Pointer to an open osl table.
1801 * \param col_num The column to use for iterating over the elements.
1802 * \param private_data Pointer that gets passed to \a func.
1803 * \param func The function to be called for each node in the rbtree.
1805 * This function does an in-order walk of the rbtree associated with \a
1806 * col_num. It is an error if the \p OSL_RBTREE flag is not set for this
1807 * column. For each node in the rbtree, the given function \a func is called
1808 * with two pointers as arguments: The first osl_row* argument points to the
1809 * row that contains the object corresponding to the rbtree node currently
1810 * traversed, and the \a private_data pointer is passed verbatim to \a func as the
1811 * second argument. The loop terminates either if \a func returns a negative
1812 * value, or if all nodes of the tree have been visited.
1815 * \return Standard. If the termination of the loop was caused by \a func
1816 * returning a negative value, this value is returned.
1818 * \sa osl_storage_flags, osl_rbtree_loop_reverse(), osl_compare_func.
1820 __export int osl_rbtree_loop(const struct osl_table *t, unsigned col_num,
1821 void *private_data, osl_rbtree_loop_func *func)
1823 struct osl_column *col;
1825 int ret = check_rbtree_col(t, col_num, &col);
1828 return rbtree_loop(col, private_data, func);
1832 * Loop over all nodes in an rbtree in reverse order.
1834 * \param t Identical meaning as in \p osl_rbtree_loop().
1835 * \param col_num Identical meaning as in \p osl_rbtree_loop().
1836 * \param private_data Identical meaning as in \p osl_rbtree_loop().
1837 * \param func Identical meaning as in \p osl_rbtree_loop().
1839 * This function is identical to \p osl_rbtree_loop(), the only difference
1840 * is that the tree is walked in reverse order.
1842 * \return The same return value as \p osl_rbtree_loop().
1844 * \sa osl_rbtree_loop().
1846 __export int osl_rbtree_loop_reverse(const struct osl_table *t, unsigned col_num,
1847 void *private_data, osl_rbtree_loop_func *func)
1849 struct osl_column *col;
1851 int ret = check_rbtree_col(t, col_num, &col);
1854 return rbtree_loop_reverse(col, private_data, func);
1857 /* TODO: Rollback changes on errors */
1858 static int rename_disk_storage_objects(struct osl_table *t,
1859 struct osl_object *old_obj, struct osl_object *new_obj)
1862 const struct osl_column_description *cd;
1863 char *old_ds_name, *new_ds_name;
1865 if (!t->num_disk_storage_columns)
1866 return 1; /* nothing to do */
1867 if (old_obj->size == new_obj->size && !memcmp(new_obj->data,
1868 old_obj->data, new_obj->size))
1869 return 1; /* object did not change */
1870 old_ds_name = disk_storage_name_of_object(t, old_obj);
1871 new_ds_name = disk_storage_name_of_object(t, new_obj);
1872 ret = -ERRNO_TO_ERROR(ENOMEM);
1873 if (!old_ds_name || ! new_ds_name)
1876 FOR_EACH_DISK_STORAGE_COLUMN(i, t, cd) {
1877 char *old_filename, *new_filename;
1878 ret = create_disk_storage_object_dir(t, i, new_ds_name);
1881 old_filename = disk_storage_path(t, i, old_ds_name);
1882 new_filename = disk_storage_path(t, i, new_ds_name);
1883 if (!old_filename || !new_filename)
1884 ret = -ERRNO_TO_ERROR(ENOMEM);
1886 ret = para_rename(old_filename, new_filename);
1901 * Change an object in an osl table.
1903 * \param t Pointer to an open osl table.
1904 * \param r Pointer to the row containing the object to be updated.
1905 * \param col_num Number of the column containing the object to be updated.
1906 * \param obj Pointer to the replacement object.
1908 * This function gets rid of all references to the old object. This includes
1909 * removal of the rbtree node in case there is an rbtree associated with \a
1910 * col_num. It then inserts \a obj into the table and the rbtree if necessary.
1912 * If the \p OSL_RBTREE flag is set for \a col_num, you \b MUST call this
1913 * function in order to change the contents of an object, even for volatile or
1914 * mapped columns of constant size (which may be updated directly if \p
1915 * OSL_RBTREE is not set). Otherwise the rbtree might become corrupted.
1919 __export int osl_update_object(struct osl_table *t, const struct osl_row *r,
1920 unsigned col_num, struct osl_object *obj)
1922 struct osl_column *col;
1923 const struct osl_column_description *cd;
1927 return -E_OSL_BAD_TABLE;
1928 col = &t->columns[col_num];
1929 cd = get_column_description(t->desc, col_num);
1930 DEBUG_LOG("updating column %u of %s\n", col_num, t->desc->name);
1931 if (cd->storage_flags & OSL_RBTREE) {
1932 if (search_rbtree(obj, t, col_num, NULL, NULL) > 0)
1933 return -E_OSL_RB_KEY_EXISTS;
1935 if (cd->storage_flags & OSL_FIXED_SIZE) {
1936 if (obj->size != cd->data_size)
1937 return -E_OSL_BAD_DATA_SIZE;
1939 remove_rb_node(t, col_num, r);
1940 if (cd->storage_type == OSL_NO_STORAGE) { /* TODO: If fixed size, reuse object? */
1941 free(r->volatile_objects[col->volatile_num].data);
1942 r->volatile_objects[col->volatile_num] = *obj;
1943 } else if (cd->storage_type == OSL_DISK_STORAGE) {
1945 ret = disk_storage_name_of_row(t, r, &ds_name);
1948 ret = delete_disk_storage_file(t, col_num, ds_name);
1949 if (ret < 0 && !is_errno(-ret, ENOENT)) {
1953 ret = write_disk_storage_file(t, col_num, obj, ds_name);
1957 } else { /* mapped storage */
1958 struct osl_object old_obj;
1959 ret = get_mapped_object(t, col_num, r->num, &old_obj);
1963 * If the updated column is the disk storage name column, the
1964 * disk storage name changes, so we have to rename all disk
1965 * storage objects accordingly.
1967 if (col_num == t->disk_storage_name_column) {
1968 ret = rename_disk_storage_objects(t, &old_obj, obj);
1972 if (cd->storage_flags & OSL_FIXED_SIZE)
1973 memcpy(old_obj.data, obj->data, cd->data_size);
1974 else { /* TODO: if the size doesn't change, use old space */
1975 uint32_t new_data_map_size;
1977 ret = get_row_index(t, r->num, &row_index);
1980 ret = mark_mapped_object_invalid(t, r->num, col_num);
1983 unmap_column(t, col_num);
1984 ret = append_map_file(t, col_num, obj,
1985 &new_data_map_size);
1988 ret = map_column(t, col_num);
1991 update_cell_index(row_index, col, new_data_map_size,
1995 if (cd->storage_flags & OSL_RBTREE) {
1996 ret = insert_rbtree(t, col_num, r, obj);
2004 * Retrieve an object of type \p OSL_DISK_STORAGE by row and column.
2006 * \param t Pointer to an open osl table.
2007 * \param r Pointer to the row containing the object.
2008 * \param col_num The column number.
2009 * \param obj Points to the result upon successful return.
2011 * For columns of type \p OSL_DISK_STORAGE, this function must be used to
2012 * retrieve one of its containing objects. Afterwards, osl_close_disk_object()
2013 * must be called in order to deallocate the resources.
2017 * \sa osl_get_object(), osl_storage_type, osl_close_disk_object().
2019 __export int osl_open_disk_object(const struct osl_table *t, const struct osl_row *r,
2020 unsigned col_num, struct osl_object *obj)
2022 const struct osl_column_description *cd;
2023 char *ds_name, *filename;
2027 return -E_OSL_BAD_TABLE;
2028 cd = get_column_description(t->desc, col_num);
2029 if (cd->storage_type != OSL_DISK_STORAGE)
2030 return -E_OSL_BAD_STORAGE_TYPE;
2032 ret = disk_storage_name_of_row(t, r, &ds_name);
2035 filename = disk_storage_path(t, col_num, ds_name);
2038 return -ERRNO_TO_ERROR(ENOMEM);
2039 DEBUG_LOG("filename: %s\n", filename);
2040 ret = mmap_full_file(filename, O_RDONLY, &obj->data, &obj->size, NULL);
2046 * Free resources that were allocated during osl_open_disk_object().
2048 * \param obj Pointer to the object previously returned by open_disk_object().
2050 * \return The return value of the underlying call to para_munmap().
2052 * \sa para_munmap().
2054 __export int osl_close_disk_object(struct osl_object *obj)
2056 return para_munmap(obj->data, obj->size);
2060 * Get the number of rows of the given table.
2062 * \param t Pointer to an open osl table.
2063 * \param num_rows Result is returned here.
2065 * The number of rows returned via \a num_rows excluding any invalid rows.
2067 * \return Positive on success, \p -E_OSL_BAD_TABLE if \a t is \p NULL.
2069 __export int osl_get_num_rows(const struct osl_table *t, unsigned *num_rows)
2072 return -E_OSL_BAD_TABLE;
2073 assert(t->num_rows >= t->num_invalid_rows);
2074 *num_rows = t->num_rows - t->num_invalid_rows;
2079 * Get the rank of a row.
2081 * \param t An open osl table.
2082 * \param r The row to get the rank of.
2083 * \param col_num The number of an rbtree column.
2084 * \param rank Result pointer.
2086 * The rank is, by definition, the position of the row in the linear order
2087 * determined by an in-order tree walk of the rbtree associated with column
2088 * number \a col_num of \a table.
2092 * \sa osl_get_nth_row().
2094 __export int osl_get_rank(const struct osl_table *t, struct osl_row *r,
2095 unsigned col_num, unsigned *rank)
2097 struct osl_object obj;
2098 struct osl_column *col;
2099 struct rb_node *node;
2100 int ret = check_rbtree_col(t, col_num, &col);
2104 ret = osl_get_object(t, r, col_num, &obj);
2107 ret = search_rbtree(&obj, t, col_num, &node, NULL);
2110 ret = rb_rank(node, rank);
2112 return -E_OSL_BAD_ROW;
2117 * Get the row with n-th greatest value.
2119 * \param t Pointer to an open osl table.
2120 * \param col_num The column number.
2121 * \param n The rank of the desired row.
2122 * \param result Row is returned here.
2124 * Retrieve the n-th order statistic with respect to the compare function
2125 * of the rbtree column \a col_num. In other words, get that row with
2126 * \a n th greatest value in column \a col_num. It's an error if
2127 * \a col_num is not a rbtree column, or if \a n is larger than the
2128 * number of rows in the table.
2132 * \sa osl_storage_flags, osl_compare_func, osl_get_row(),
2133 * osl_rbtree_last_row(), osl_rbtree_first_row(), osl_get_rank().
2135 __export int osl_get_nth_row(const struct osl_table *t, unsigned col_num,
2136 unsigned n, struct osl_row **result)
2138 struct osl_column *col;
2139 struct rb_node *node;
2144 return -E_OSL_RB_KEY_NOT_FOUND;
2145 ret = osl_get_num_rows(t, &num_rows);
2149 return -E_OSL_RB_KEY_NOT_FOUND;
2150 ret = check_rbtree_col(t, col_num, &col);
2153 node = rb_nth(col->rbtree.rb_node, n);
2155 return -E_OSL_RB_KEY_NOT_FOUND;
2156 *result = get_row_pointer(node, col->rbtree_num);
2161 * Get the row corresponding to the smallest rbtree node of a column.
2163 * \param t An open rbtree table.
2164 * \param col_num The number of the rbtree column.
2165 * \param result A pointer to the first row is returned here.
2167 * The rbtree node of the smallest object (with respect to the corresponding
2168 * compare function) is selected and the row containing this object is
2169 * returned. It is an error if \a col_num refers to a column without an
2170 * associated rbtree.
2174 * \sa osl_get_nth_row(), osl_rbtree_last_row().
2176 __export int osl_rbtree_first_row(const struct osl_table *t, unsigned col_num,
2177 struct osl_row **result)
2179 return osl_get_nth_row(t, col_num, 1, result);
2183 * Get the row corresponding to the greatest rbtree node of a column.
2185 * \param t The same meaning as in \p osl_rbtree_first_row().
2186 * \param col_num The same meaning as in \p osl_rbtree_first_row().
2187 * \param result The same meaning as in \p osl_rbtree_first_row().
2189 * This function works just like osl_rbtree_first_row(), the only difference
2190 * is that the row containing the greatest rather than the smallest object is
2195 * \sa osl_get_nth_row(), osl_rbtree_first_row().
2197 __export int osl_rbtree_last_row(const struct osl_table *t, unsigned col_num,
2198 struct osl_row **result)
2201 int ret = osl_get_num_rows(t, &num_rows);
2205 return osl_get_nth_row(t, col_num, num_rows, result);