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__)),
86 __export const char *osl_strerror(int num)
88 if (IS_SYSTEM_ERROR(num))
89 return strerror((num) & ((1 << SYSTEM_ERROR_BIT) - 1));
90 return msgstr.str + errmsgidx[num];
95 static void __attribute ((constructor)) init_loglevel(void)
97 char *p = getenv("OSL_LOGLEVEL");
99 /* don't log anything if unset */
100 loglevel = p? atoi(p) : EMERG + 1;
106 * \param ll Loglevel.
107 * \param fmt Usual format string.
109 * All XXX_LOG() macros use this function.
111 __printf_2_3 void __log(int ll, const char* fmt,...)
124 strftime(str, sizeof(str), "%b %d %H:%M:%S", tm);
125 fprintf(outfd, "%s ", str);
127 vfprintf(outfd, fmt, argp);
132 * A wrapper for lseek(2).
134 * \param fd The file descriptor whose offset is to be to repositioned.
135 * \param offset A value-result parameter.
136 * \param whence Usual repositioning directive.
138 * Reposition the offset of the file descriptor \a fd to the argument \a offset
139 * according to the directive \a whence. Upon successful return, \a offset
140 * contains the resulting offset location as measured in bytes from the
141 * beginning of the file.
143 * \return Positive on success. Otherwise, the function returns \p -E_OSL_LSEEK.
147 static int __lseek(int fd, off_t *offset, int whence)
149 *offset = lseek(fd, *offset, whence);
150 int ret = -E_OSL_LSEEK;
157 * Wrapper for the write system call.
159 * \param fd The file descriptor to write to.
160 * \param buf The buffer to write.
161 * \param size The length of \a buf in bytes.
163 * This function writes out the given buffer and retries if an interrupt
164 * occurred during the write.
166 * \return On success, the number of bytes written is returned, otherwise, the
167 * function returns \p -E_OSL_WRITE.
171 static ssize_t __write(int fd, const void *buf, size_t size)
176 ret = write(fd, buf, size);
177 if ((ret < 0) && (errno == EAGAIN || errno == EINTR))
179 return ret >= 0? ret : -E_OSL_WRITE;
184 * Write the whole buffer to a file descriptor.
186 * \param fd The file descriptor to write to.
187 * \param buf The buffer to write.
188 * \param size The length of \a buf in bytes.
190 * This function writes the given buffer and continues on short writes and
191 * when interrupted by a signal.
193 * \return Positive on success, negative on errors. Possible errors: any
194 * errors returned by para_write().
198 static ssize_t write_all(int fd, const void *buf, size_t size)
200 // DEBUG_LOG("writing %zu bytes\n", size);
203 ssize_t ret = __write(fd, b, size);
204 // DEBUG_LOG("ret: %zd\n", ret);
213 * Open a file, write the given buffer and close the file.
215 * \param filename Full path to the file to open.
216 * \param buf The buffer to write to the file.
217 * \param size The size of \a buf.
221 static int write_file(const char *filename, const void *buf, size_t size)
225 ret = osl_open(filename, O_WRONLY | O_CREAT | O_EXCL, 0644);
229 ret = write_all(fd, buf, size);
238 static int append_file(const char *filename, char *header, size_t header_size,
239 char *data, size_t data_size, uint32_t *new_pos)
243 // DEBUG_LOG("appending %zu + %zu bytes\n", header_size, data_size);
244 ret = osl_open(filename, O_WRONLY | O_CREAT | O_APPEND, 0644);
248 if (header && header_size) {
249 ret = write_all(fd, header, header_size);
253 ret = write_all(fd, data, data_size);
258 ret = __lseek(fd, &offset, SEEK_END);
261 // DEBUG_LOG("new file size: " FMT_OFF_T "\n", offset);
270 static int verify_name(const char *name)
273 return -E_OSL_BAD_NAME;
275 return -E_OSL_BAD_NAME;
276 if (strchr(name, '/'))
277 return -E_OSL_BAD_NAME;
278 if (!strcmp(name, ".."))
279 return -E_OSL_BAD_NAME;
280 if (!strcmp(name, "."))
281 return -E_OSL_BAD_NAME;
285 int osl_hash_compare(const struct osl_object *obj1, const struct osl_object *obj2)
287 return hash_compare((HASH_TYPE *)obj1->data, (HASH_TYPE *)obj2->data);
290 static char *disk_storage_dirname(const struct osl_table *t, unsigned col_num,
293 char *dirname, *column_name = column_filename(t, col_num);
297 if (!(t->desc->flags & OSL_LARGE_TABLE))
299 dirname = make_message("%s/%.2s", column_name, ds_name);
304 static char *disk_storage_name_of_object(const struct osl_table *t,
305 const struct osl_object *obj)
307 HASH_TYPE hash[HASH_SIZE];
308 hash_object(obj, hash);
309 return disk_storage_name_of_hash(t, hash);
312 static int disk_storage_name_of_row(const struct osl_table *t,
313 const struct osl_row *row, char **name)
315 struct osl_object obj;
316 int ret = osl_get_object(t, row, t->disk_storage_name_column, &obj);
320 *name = disk_storage_name_of_object(t, &obj);
323 return -ERRNO_TO_ERROR(ENOMEM);
326 static void column_name_hash(const char *col_name, HASH_TYPE *hash)
328 hash_function(col_name, strlen(col_name), hash);
331 static int init_column_descriptions(struct osl_table *t)
334 const struct osl_column_description *cd;
336 ret = -E_OSL_BAD_TABLE_DESC;
337 ret = verify_name(t->desc->name);
340 ret = -E_OSL_BAD_DB_DIR;
341 if (!t->desc->dir && (t->num_disk_storage_columns || t->num_mapped_columns))
343 /* the size of the index header without column descriptions */
344 t->index_header_size = IDX_COLUMN_DESCRIPTIONS;
345 FOR_EACH_COLUMN(i, t->desc, cd) {
346 struct osl_column *col = t->columns + i;
347 if (cd->storage_flags & OSL_RBTREE) {
348 if (!cd->compare_function)
349 return -E_OSL_NO_COMPARE_FUNC;
351 if (cd->storage_type == OSL_NO_STORAGE)
353 ret = -E_OSL_NO_COLUMN_NAME;
354 if (!cd->name || !cd->name[0])
356 ret = verify_name(cd->name);
359 t->index_header_size += index_column_description_size(cd->name);
360 column_name_hash(cd->name, col->name_hash);
361 ret = -E_OSL_DUPLICATE_COL_NAME;
362 for (j = i + 1; j < t->desc->num_columns; j++) {
363 const char *name2 = get_column_description(t->desc,
365 if (cd->name && name2 && !strcmp(cd->name, name2))
375 * Initialize a struct table from given table description.
377 * \param desc The description of the osl table.
378 * \param table_ptr Result is returned here.
380 * This function performs several sanity checks on \p desc and returns if any
381 * of these tests fail. On success, a struct \p osl_table is allocated and
382 * initialized with data derived from \p desc.
386 * \sa struct osl_table.
388 int init_table_structure(const struct osl_table_description *desc,
389 struct osl_table **table_ptr)
391 const struct osl_column_description *cd;
392 struct osl_table *t = calloc(1, sizeof(*t));
393 int i, ret = -ERRNO_TO_ERROR(ENOMEM), have_disk_storage_name_column = 0;
397 ret = -E_OSL_BAD_TABLE_DESC;
400 DEBUG_LOG("creating table structure for '%s' from table "
401 "description\n", desc->name);
402 ret = -E_OSL_NO_COLUMN_DESC;
403 if (!desc->column_descriptions)
405 ret = -E_OSL_NO_COLUMNS;
406 if (!desc->num_columns)
408 ret = -ERRNO_TO_ERROR(ENOMEM);
409 t->columns = calloc(desc->num_columns, sizeof(struct osl_column));
413 FOR_EACH_COLUMN(i, t->desc, cd) {
414 enum osl_storage_type st = cd->storage_type;
415 enum osl_storage_flags sf = cd->storage_flags;
416 struct osl_column *col = &t->columns[i];
418 ret = -E_OSL_BAD_STORAGE_TYPE;
419 if (st != OSL_MAPPED_STORAGE && st != OSL_DISK_STORAGE
420 && st != OSL_NO_STORAGE)
422 ret = -E_OSL_BAD_STORAGE_FLAGS;
423 if (st == OSL_DISK_STORAGE && sf & OSL_RBTREE)
425 ret = -E_OSL_BAD_STORAGE_SIZE;
426 if (sf & OSL_FIXED_SIZE && !cd->data_size)
429 case OSL_DISK_STORAGE:
430 t->num_disk_storage_columns++;
432 case OSL_MAPPED_STORAGE:
433 t->num_mapped_columns++;
434 col->index_offset = t->row_index_size;
435 t->row_index_size += 8;
438 col->volatile_num = t->num_volatile_columns;
439 t->num_volatile_columns++;
442 if (sf & OSL_RBTREE) {
443 col->rbtree_num = t->num_rbtrees;
445 if ((sf & OSL_UNIQUE) && (st == OSL_MAPPED_STORAGE)) {
446 if (!have_disk_storage_name_column)
447 t->disk_storage_name_column = i;
448 have_disk_storage_name_column = 1;
452 ret = -E_OSL_NO_UNIQUE_RBTREE_COLUMN;
453 if (t->num_disk_storage_columns && !have_disk_storage_name_column)
455 ret = -E_OSL_NO_RBTREE_COL;
459 DEBUG_LOG("OK. Index entry size: %u\n", t->row_index_size);
460 ret = init_column_descriptions(t);
472 * Read the table description from index header.
474 * \param map The memory mapping of the index file.
475 * \param desc The values found in the index header are returned here.
477 * Read the index header, check for the paraslash magic string and the table version number.
478 * Read all information stored in the index header into \a desc.
482 * \sa struct osl_table_description, osl_create_table.
484 int read_table_desc(struct osl_object *map, struct osl_table_description *desc)
486 char *buf = map->data;
488 uint16_t header_size;
491 struct osl_column_description *cd;
493 if (map->size < MIN_INDEX_HEADER_SIZE(1))
494 return -E_OSL_SHORT_TABLE;
495 if (strncmp(buf + IDX_PARA_MAGIC, PARA_MAGIC, strlen(PARA_MAGIC)))
496 return -E_OSL_NO_MAGIC;
497 version = read_u8(buf + IDX_VERSION);
498 if (version < MIN_TABLE_VERSION || version > MAX_TABLE_VERSION)
499 return -E_OSL_VERSION_MISMATCH;
500 desc->flags = read_u8(buf + IDX_TABLE_FLAGS);
501 desc->num_columns = read_u16(buf + IDX_NUM_COLUMNS);
502 INFO_LOG("%u columns\n", desc->num_columns);
503 if (!desc->num_columns)
504 return -E_OSL_NO_COLUMNS;
505 header_size = read_u16(buf + IDX_HEADER_SIZE);
506 if (map->size < header_size)
507 return -E_OSL_BAD_SIZE;
508 desc->column_descriptions = calloc(desc->num_columns,
509 sizeof(struct osl_column_description));
510 if (!desc->column_descriptions)
511 return -ERRNO_TO_ERROR(ENOMEM);
512 offset = IDX_COLUMN_DESCRIPTIONS;
513 FOR_EACH_COLUMN(i, desc, cd) {
516 ret = -E_OSL_SHORT_TABLE;
517 if (map->size < offset + MIN_IDX_COLUMN_DESCRIPTION_SIZE) {
518 ERROR_LOG("map size = %zu < %u = offset + min desc size\n",
519 map->size, offset + MIN_IDX_COLUMN_DESCRIPTION_SIZE);
522 cd->storage_type = read_u16(buf + offset + IDX_CD_STORAGE_TYPE);
523 cd->storage_flags = read_u16(buf + offset +
524 IDX_CD_STORAGE_FLAGS);
525 cd->data_size = read_u32(buf + offset + IDX_CD_DATA_SIZE);
526 null_byte = memchr(buf + offset + IDX_CD_NAME, '\0',
527 map->size - offset - IDX_CD_NAME);
528 ret = -E_OSL_INDEX_CORRUPTION;
531 ret = -ERRNO_TO_ERROR(ENOMEM);
532 cd->name = strdup(buf + offset + IDX_CD_NAME);
535 offset += index_column_description_size(cd->name);
537 if (offset != header_size) {
538 ret = -E_OSL_INDEX_CORRUPTION;
539 ERROR_LOG("real header size = %u != %u = stored header size\n",
540 offset, header_size);
545 FOR_EACH_COLUMN(i, desc, cd)
551 * check whether the table description given by \p t->desc matches the on-disk
552 * table structure stored in the index of \a t.
554 static int compare_table_descriptions(struct osl_table *t)
557 struct osl_table_description desc;
558 const struct osl_column_description *cd1, *cd2;
560 /* read the on-disk structure into desc */
561 ret = read_table_desc(&t->index_map, &desc);
564 ret = -E_OSL_BAD_TABLE_FLAGS;
565 if (desc.flags != t->desc->flags)
567 ret = -E_OSL_BAD_COLUMN_NUM;
568 if (desc.num_columns > t->desc->num_columns)
570 if (desc.num_columns < t->desc->num_columns) {
571 struct osl_column_description *cd;
572 unsigned diff = t->desc->num_columns - desc.num_columns;
573 INFO_LOG("extending table by %u volatile columns\n", diff);
574 ret = -ERRNO_TO_ERROR(ENOMEM);
575 desc.column_descriptions = realloc(desc.column_descriptions,
576 t->desc->num_columns * sizeof(struct osl_column_description));
577 if (!desc.column_descriptions)
579 for (i = desc.num_columns; i < t->desc->num_columns; i++) {
580 cd = get_column_description(&desc, i);
581 cd->storage_type = OSL_NO_STORAGE;
584 desc.num_columns += diff;
586 FOR_EACH_COLUMN(i, t->desc, cd1) {
587 cd2 = get_column_description(&desc, i);
588 ret = -E_OSL_BAD_STORAGE_TYPE;
589 if (cd1->storage_type != cd2->storage_type)
591 if (cd1->storage_type == OSL_NO_STORAGE)
593 ret = -E_OSL_BAD_STORAGE_FLAGS;
594 if (cd1->storage_flags != cd2->storage_flags) {
595 ERROR_LOG("sf1 = %u != %u = sf2\n",
596 cd1->storage_flags, cd2->storage_flags);
599 ret = -E_OSL_BAD_DATA_SIZE;
600 if (cd1->storage_flags & OSL_FIXED_SIZE)
601 if (cd1->data_size != cd2->data_size)
603 ret = -E_OSL_BAD_COLUMN_NAME;
604 if (strcmp(cd1->name, cd2->name))
607 INFO_LOG("table description of '%s' matches on-disk data, good\n",
611 FOR_EACH_COLUMN(i, &desc, cd1)
613 free(desc.column_descriptions);
617 static int create_table_index(struct osl_table *t)
619 char *buf, *filename;
621 size_t size = t->index_header_size;
622 const struct osl_column_description *cd;
625 INFO_LOG("creating %zu byte index for table %s\n", size,
627 buf = calloc(1, size);
629 return -ERRNO_TO_ERROR(ENOMEM);
630 sprintf(buf + IDX_PARA_MAGIC, "%s", PARA_MAGIC);
631 write_u8(buf + IDX_TABLE_FLAGS, t->desc->flags);
632 write_u8(buf + IDX_DIRTY_FLAG, 0);
633 write_u8(buf + IDX_VERSION, CURRENT_TABLE_VERSION);
634 write_u16(buf + IDX_NUM_COLUMNS, t->num_mapped_columns + t->num_disk_storage_columns);
635 write_u16(buf + IDX_HEADER_SIZE, t->index_header_size);
636 offset = IDX_COLUMN_DESCRIPTIONS;
637 FOR_EACH_COLUMN(i, t->desc, cd) {
638 /* no need to store info about volatile storage */
639 if (cd->storage_type == OSL_NO_STORAGE)
641 write_u16(buf + offset + IDX_CD_STORAGE_TYPE,
643 write_u16(buf + offset + IDX_CD_STORAGE_FLAGS,
645 if (cd->storage_flags & OSL_FIXED_SIZE)
646 write_u32(buf + offset + IDX_CD_DATA_SIZE,
648 strcpy(buf + offset + IDX_CD_NAME, cd->name);
649 offset += index_column_description_size(cd->name);
651 assert(offset = size);
652 filename = index_filename(t->desc);
654 ret = write_file(filename, buf, size);
656 ret = -ERRNO_TO_ERROR(ENOMEM);
662 __export int osl_create_table(const struct osl_table_description *desc)
664 const struct osl_column_description *cd;
665 char *table_dir = NULL, *filename;
667 int i, ret = init_table_structure(desc, &t);
671 INFO_LOG("creating %s\n", desc->name);
672 FOR_EACH_COLUMN(i, t->desc, cd) {
673 if (cd->storage_type == OSL_NO_STORAGE)
676 ret = para_mkdir(desc->dir, 0777);
677 if (ret < 0 && !is_errno(-ret, EEXIST))
679 table_dir = make_message("%s/%s", desc->dir,
681 ret = -ERRNO_TO_ERROR(ENOMEM);
684 ret = para_mkdir(table_dir, 0777);
688 ret = -ERRNO_TO_ERROR(ENOMEM);
689 filename = column_filename(t, i);
692 INFO_LOG("filename: %s\n", filename);
693 if (cd->storage_type == OSL_MAPPED_STORAGE) {
694 ret = osl_open(filename, O_RDWR | O_CREAT | O_EXCL,
703 ret = para_mkdir(filename, 0777);
708 if (t->num_mapped_columns) {
709 ret = create_table_index(t);
721 static int table_is_dirty(struct osl_table *t)
723 char *buf = (char *)t->index_map.data + IDX_DIRTY_FLAG;
724 uint8_t dirty = read_u8(buf) & 0x1;
728 static void mark_table_dirty(struct osl_table *t)
730 char *buf = (char *)t->index_map.data + IDX_DIRTY_FLAG;
731 write_u8(buf, read_u8(buf) | 1);
734 static void mark_table_clean(struct osl_table *t)
736 char *buf = (char *)t->index_map.data + IDX_DIRTY_FLAG;
737 write_u8(buf, read_u8(buf) & 0xfe);
740 static void unmap_column(struct osl_table *t, unsigned col_num)
742 struct osl_object map = t->columns[col_num].data_map;
746 ret = para_munmap(map.data, map.size);
752 * Unmap all mapped files of an osl table.
754 * \param t Pointer to a mapped table.
755 * \param flags Options for unmapping.
757 * \return Positive on success, negative on errors.
759 * \sa map_table(), enum osl_close_flags, para_munmap().
761 int unmap_table(struct osl_table *t, enum osl_close_flags flags)
764 const struct osl_column_description *cd;
767 if (!t->num_mapped_columns) /* can this ever happen? */
769 INFO_LOG("unmapping table '%s'\n", t->desc->name);
770 if (!t->index_map.data)
771 return -E_OSL_NOT_MAPPED;
772 if (flags & OSL_MARK_CLEAN)
774 ret = para_munmap(t->index_map.data, t->index_map.size);
777 t->index_map.data = NULL;
780 FOR_EACH_MAPPED_COLUMN(i, t, cd)
785 static int map_column(struct osl_table *t, unsigned col_num)
788 char *filename = column_filename(t, col_num);
789 int ret = -E_OSL_STAT;
792 return -ERRNO_TO_ERROR(ENOMEM);
793 if (stat(filename, &statbuf) < 0) {
797 if (!(S_IFREG & statbuf.st_mode)) {
801 ret = mmap_full_file(filename, O_RDWR,
802 &t->columns[col_num].data_map.data,
803 &t->columns[col_num].data_map.size,
810 * Map the index file and all columns of type \p OSL_MAPPED_STORAGE into memory.
812 * \param t Pointer to an initialized table structure.
813 * \param flags Mapping options.
815 * \return Negative return value on errors; on success the number of rows
816 * (including invalid rows) is returned.
818 * \sa unmap_table(), enum map_table_flags, osl_open_table(), mmap(2).
820 int map_table(struct osl_table *t, enum map_table_flags flags)
823 const struct osl_column_description *cd;
824 int i = 0, ret, num_rows = 0;
826 if (!t->num_mapped_columns)
828 if (t->index_map.data)
829 return -E_OSL_ALREADY_MAPPED;
830 filename = index_filename(t->desc);
832 return -ERRNO_TO_ERROR(ENOMEM);
833 INFO_LOG("mapping table '%s' (index: %s)\n", t->desc->name, filename);
834 ret = mmap_full_file(filename, flags & MAP_TBL_FL_MAP_RDONLY?
835 O_RDONLY : O_RDWR, &t->index_map.data, &t->index_map.size, NULL);
839 if (flags & MAP_TBL_FL_VERIFY_INDEX) {
840 ret = compare_table_descriptions(t);
845 if (!(flags & MAP_TBL_FL_IGNORE_DIRTY)) {
846 if (table_is_dirty(t)) {
847 ERROR_LOG("%s is dirty\n", t->desc->name);
852 num_rows = table_num_rows(t);
856 FOR_EACH_MAPPED_COLUMN(i, t, cd) {
857 ret = map_column(t, i);
862 err: /* unmap what is already mapped */
863 for (i--; i >= 0; i--) {
864 struct osl_object map = t->columns[i].data_map;
865 para_munmap(map.data, map.size);
868 para_munmap(t->index_map.data, t->index_map.size);
869 t->index_map.data = NULL;
874 * Retrieve a mapped object by row and column number.
876 * \param t Pointer to an open osl table.
877 * \param col_num Number of the mapped column containing the object to retrieve.
878 * \param row_num Number of the row containing the object to retrieve.
879 * \param obj The result is returned here.
881 * It is considered an error if \a col_num does not refer to a column
882 * of storage type \p OSL_MAPPED_STORAGE.
886 * \sa osl_storage_type.
888 int get_mapped_object(const struct osl_table *t, unsigned col_num,
889 uint32_t row_num, struct osl_object *obj)
891 struct osl_column *col = &t->columns[col_num];
897 if (t->num_rows <= row_num)
898 return -E_OSL_BAD_ROW_NUM;
899 ret = get_cell_index(t, row_num, col_num, &cell_index);
902 offset = read_u32(cell_index);
903 obj->size = read_u32(cell_index + 4) - 1;
904 header = col->data_map.data + offset;
905 obj->data = header + 1;
906 if (read_u8(header) == 0xff) {
907 ERROR_LOG("col %u, size %zu, offset %u\n", col_num,
909 return -E_OSL_INVALID_OBJECT;
914 static int search_rbtree(const struct osl_object *obj,
915 const struct osl_table *t, unsigned col_num,
916 struct rb_node **result, struct rb_node ***rb_link)
918 struct osl_column *col = &t->columns[col_num];
919 struct rb_node **new = &col->rbtree.rb_node, *parent = NULL;
920 const struct osl_column_description *cd =
921 get_column_description(t->desc, col_num);
922 enum osl_storage_type st = cd->storage_type;
924 struct osl_row *this_row = get_row_pointer(*new,
927 struct osl_object this_obj;
929 if (st == OSL_MAPPED_STORAGE) {
930 ret = get_mapped_object(t, col_num, this_row->num,
935 this_obj = this_row->volatile_objects[col->volatile_num];
936 ret = cd->compare_function(obj, &this_obj);
939 *result = get_rb_node_pointer(this_row,
944 new = &((*new)->rb_left);
946 new = &((*new)->rb_right);
952 return -E_OSL_RB_KEY_NOT_FOUND;
955 static int insert_rbtree(struct osl_table *t, unsigned col_num,
956 const struct osl_row *row, const struct osl_object *obj)
958 struct rb_node *parent, **rb_link;
961 int ret = search_rbtree(obj, t, col_num, &parent, &rb_link);
964 return -E_OSL_RB_KEY_EXISTS;
965 rbtree_num = t->columns[col_num].rbtree_num;
966 n = get_rb_node_pointer(row, rbtree_num);
967 rb_link_node(n, parent, rb_link);
968 rb_insert_color(n, &t->columns[col_num].rbtree);
972 static void remove_rb_node(struct osl_table *t, unsigned col_num,
973 const struct osl_row *row)
975 struct osl_column *col = &t->columns[col_num];
976 const struct osl_column_description *cd =
977 get_column_description(t->desc, col_num);
978 enum osl_storage_flags sf = cd->storage_flags;
979 struct rb_node *victim, *splice_out_node, *tmp;
980 if (!(sf & OSL_RBTREE))
983 * Which node is removed/spliced out actually depends on how many
984 * children the victim node has: If it has no children, it gets
985 * deleted. If it has one child, it gets spliced out. If it has two
986 * children, its successor (which has at most a right child) gets
989 victim = get_rb_node_pointer(row, col->rbtree_num);
990 if (victim->rb_left && victim->rb_right)
991 splice_out_node = rb_next(victim);
993 splice_out_node = victim;
994 /* Go up to the root and decrement the size of each node in the path. */
995 for (tmp = splice_out_node; tmp; tmp = rb_parent(tmp))
997 rb_erase(victim, &col->rbtree);
1000 static int add_row_to_rbtrees(struct osl_table *t, uint32_t row_num,
1001 struct osl_object *volatile_objs, struct osl_row **row_ptr)
1005 struct osl_row *row = allocate_row(t->num_rbtrees);
1006 const struct osl_column_description *cd;
1009 return -ERRNO_TO_ERROR(ENOMEM);
1011 row->volatile_objects = volatile_objs;
1012 FOR_EACH_RBTREE_COLUMN(i, t, cd) {
1013 if (cd->storage_type == OSL_MAPPED_STORAGE) {
1014 struct osl_object obj;
1015 ret = get_mapped_object(t, i, row_num, &obj);
1018 ret = insert_rbtree(t, i, row, &obj);
1019 } else { /* volatile */
1020 const struct osl_object *obj
1021 = volatile_objs + t->columns[i].volatile_num;
1022 ret = insert_rbtree(t, i, row, obj);
1030 err: /* rollback changes, i.e. remove added entries from rbtrees */
1032 remove_rb_node(t, i--, row);
1037 static void free_volatile_objects(const struct osl_table *t,
1038 enum osl_close_flags flags)
1042 struct osl_column *rb_col;
1043 const struct osl_column_description *cd;
1045 if (!t->num_volatile_columns)
1047 /* find the first rbtree column (any will do) */
1048 FOR_EACH_RBTREE_COLUMN(i, t, cd)
1050 rb_col = t->columns + i;
1051 /* walk that rbtree and free all volatile objects */
1052 for (n = rb_first(&rb_col->rbtree); n; n = rb_next(n)) {
1053 struct osl_row *r = get_row_pointer(n, rb_col->rbtree_num);
1054 if (flags & OSL_FREE_VOLATILE)
1055 FOR_EACH_VOLATILE_COLUMN(j, t, cd) {
1056 if (cd->storage_flags & OSL_DONT_FREE)
1058 free(r->volatile_objects[
1059 t->columns[j].volatile_num].data);
1061 // for (j = 0; j < t->num_volatile_columns; j++)
1062 // free(r->volatile_objects[j].data);
1063 free(r->volatile_objects);
1068 * Erase all rbtree nodes and free resources.
1070 * \param t Pointer to an open osl table.
1072 * This function is called by osl_close_table().
1074 void clear_rbtrees(struct osl_table *t)
1076 const struct osl_column_description *cd;
1077 unsigned i, rbtrees_cleared = 0;
1079 FOR_EACH_RBTREE_COLUMN(i, t, cd) {
1080 struct osl_column *col = &t->columns[i];
1083 for (n = rb_first(&col->rbtree); n;) {
1085 rb_erase(n, &col->rbtree);
1086 if (rbtrees_cleared == t->num_rbtrees) {
1087 r = get_row_pointer(n, col->rbtree_num);
1097 __export int osl_close_table(struct osl_table *t, enum osl_close_flags flags)
1102 return -E_OSL_BAD_TABLE;
1103 NOTICE_LOG("closing table %s\n", t->desc->name);
1104 free_volatile_objects(t, flags);
1106 ret = unmap_table(t, flags);
1108 ERROR_LOG("unmap_table failed: %d\n", ret);
1115 * Find out whether the given row number corresponds to an invalid row.
1117 * \param t Pointer to the osl table.
1118 * \param row_num The number of the row in question.
1120 * By definition, a row is considered invalid if all its index entries
1123 * \return Positive if \a row_num corresponds to an invalid row,
1124 * zero if it corresponds to a valid row, negative on errors.
1126 int row_is_invalid(struct osl_table *t, uint32_t row_num)
1129 int i, ret = get_row_index(t, row_num, &row_index);
1133 for (i = 0; i < t->row_index_size; i++) {
1134 if ((unsigned char)row_index[i] != 0xff)
1137 INFO_LOG("row %d is invalid\n", row_num);
1142 * Invalidate a row of an osl table.
1144 * \param t Pointer to an open osl table.
1145 * \param row_num Number of the row to mark as invalid.
1147 * This function marks each mapped object in the index entry of \a row as
1152 int mark_row_invalid(struct osl_table *t, uint32_t row_num)
1155 int ret = get_row_index(t, row_num, &row_index);
1159 INFO_LOG("marking row %d as invalid\n", row_num);
1160 memset(row_index, 0xff, t->row_index_size);
1165 * Initialize all rbtrees and compute number of invalid rows.
1167 * \param t The table containing the rbtrees to be initialized.
1171 int init_rbtrees(struct osl_table *t)
1174 const struct osl_column_description *cd;
1176 /* create rbtrees */
1177 FOR_EACH_RBTREE_COLUMN(i, t, cd)
1178 t->columns[i].rbtree = RB_ROOT;
1179 /* add valid rows to rbtrees */
1180 t->num_invalid_rows = 0;
1181 for (i = 0; i < t->num_rows; i++) {
1182 ret = row_is_invalid(t, i);
1186 t->num_invalid_rows++;
1189 ret = add_row_to_rbtrees(t, i, NULL, NULL);
1196 __export int osl_open_table(const struct osl_table_description *table_desc,
1197 struct osl_table **result)
1200 struct osl_table *t;
1201 const struct osl_column_description *cd;
1203 NOTICE_LOG("opening table %s\n", table_desc->name);
1204 ret = init_table_structure(table_desc, &t);
1207 FOR_EACH_DISK_STORAGE_COLUMN(i, t, cd) {
1208 struct stat statbuf;
1209 char *dirname = column_filename(t, i);
1211 ret = -ERRNO_TO_ERROR(ENOMEM);
1214 /* check if directory exists */
1215 ret = stat(dirname, &statbuf);
1218 ret = -ERRNO_TO_ERROR(errno);
1221 ret = -ERRNO_TO_ERROR(ENOTDIR);
1222 if (!S_ISDIR(statbuf.st_mode))
1225 ret = map_table(t, MAP_TBL_FL_VERIFY_INDEX);
1229 DEBUG_LOG("num rows: %d\n", t->num_rows);
1230 ret = init_rbtrees(t);
1232 osl_close_table(t, OSL_MARK_CLEAN); /* ignore further errors */
1243 static int create_disk_storage_object_dir(const struct osl_table *t,
1244 unsigned col_num, const char *ds_name)
1249 if (!(t->desc->flags & OSL_LARGE_TABLE))
1251 dirname = disk_storage_dirname(t, col_num, ds_name);
1253 return -ERRNO_TO_ERROR(ENOMEM);
1254 ret = para_mkdir(dirname, 0777);
1256 if (ret < 0 && !is_errno(-ret, EEXIST))
1261 static int write_disk_storage_file(const struct osl_table *t, unsigned col_num,
1262 const struct osl_object *obj, const char *ds_name)
1267 ret = create_disk_storage_object_dir(t, col_num, ds_name);
1270 filename = disk_storage_path(t, col_num, ds_name);
1272 return -ERRNO_TO_ERROR(ENOMEM);
1273 ret = write_file(filename, obj->data, obj->size);
1278 static int append_map_file(const struct osl_table *t, unsigned col_num,
1279 const struct osl_object *obj, uint32_t *new_size)
1281 char *filename = column_filename(t, col_num);
1283 char header = 0; /* zero means valid object */
1286 return -ERRNO_TO_ERROR(ENOMEM);
1287 ret = append_file(filename, &header, 1, obj->data, obj->size,
1293 static int append_row_index(const struct osl_table *t, char *row_index)
1298 if (!t->num_mapped_columns)
1300 filename = index_filename(t->desc);
1302 return -ERRNO_TO_ERROR(ENOMEM);
1303 ret = append_file(filename, NULL, 0, row_index,
1304 t->row_index_size, NULL);
1310 * A wrapper for truncate(2)
1312 * \param path Name of the regular file to truncate
1313 * \param size Number of bytes to \b shave \b off
1315 * Truncate the regular file named by \a path by \a size bytes.
1321 int para_truncate(const char *path, off_t size)
1324 struct stat statbuf;
1327 if (stat(path, &statbuf) < 0)
1329 ret = -E_OSL_BAD_SIZE;
1330 if (statbuf.st_size < size)
1333 if (truncate(path, statbuf.st_size - size) < 0)
1340 static int truncate_mapped_file(const struct osl_table *t, unsigned col_num,
1344 char *filename = column_filename(t, col_num);
1347 return -ERRNO_TO_ERROR(ENOMEM);
1348 ret = para_truncate(filename, size);
1353 static int delete_disk_storage_file(const struct osl_table *t, unsigned col_num,
1354 const char *ds_name)
1356 char *dirname, *filename = disk_storage_path(t, col_num, ds_name);
1360 return -ERRNO_TO_ERROR(ENOMEM);
1361 ret = unlink(filename);
1365 return -ERRNO_TO_ERROR(err);
1366 if (!(t->desc->flags & OSL_LARGE_TABLE))
1368 dirname = disk_storage_dirname(t, col_num, ds_name);
1370 return -ERRNO_TO_ERROR(ENOMEM);
1376 __export int osl_add_and_get_row(struct osl_table *t, struct osl_object *objects,
1377 struct osl_row **row)
1380 char *ds_name = NULL;
1381 struct rb_node **rb_parents = NULL, ***rb_links = NULL;
1382 char *new_row_index = NULL;
1383 struct osl_object *volatile_objs = NULL;
1384 const struct osl_column_description *cd;
1387 return -E_OSL_BAD_TABLE;
1388 rb_parents = malloc(t->num_rbtrees * sizeof(struct rn_node*));
1390 return -ERRNO_TO_ERROR(ENOMEM);
1391 rb_links = malloc(t->num_rbtrees * sizeof(struct rn_node**));
1394 return -ERRNO_TO_ERROR(ENOMEM);
1396 if (t->num_mapped_columns) {
1397 new_row_index = malloc(t->row_index_size);
1398 if (!new_row_index) {
1401 return -ERRNO_TO_ERROR(ENOMEM);
1404 /* pass 1: sanity checks */
1405 // DEBUG_LOG("sanity tests: %p:%p\n", objects[0].data,
1406 // objects[1].data);
1407 FOR_EACH_COLUMN(i, t->desc, cd) {
1408 enum osl_storage_type st = cd->storage_type;
1409 enum osl_storage_flags sf = cd->storage_flags;
1411 // ret = -E_OSL_NULL_OBJECT;
1414 if (st == OSL_DISK_STORAGE)
1416 if (sf & OSL_RBTREE) {
1417 unsigned rbtree_num = t->columns[i].rbtree_num;
1418 ret = -E_OSL_RB_KEY_EXISTS;
1419 // DEBUG_LOG("checking whether %p exists\n",
1420 // objects[i].data);
1421 if (search_rbtree(objects + i, t, i,
1422 &rb_parents[rbtree_num],
1423 &rb_links[rbtree_num]) > 0)
1426 if (sf & OSL_FIXED_SIZE) {
1427 // DEBUG_LOG("fixed size. need: %zu, have: %d\n",
1428 // objects[i].size, cd->data_size);
1429 ret = -E_OSL_BAD_DATA_SIZE;
1430 if (objects[i].size != cd->data_size)
1434 if (t->num_disk_storage_columns) {
1435 ds_name = disk_storage_name_of_object(t,
1436 &objects[t->disk_storage_name_column]);
1437 ret = -ERRNO_TO_ERROR(ENOMEM);
1441 ret = unmap_table(t, OSL_MARK_CLEAN);
1444 // DEBUG_LOG("sanity tests passed%s\n", "");
1445 /* pass 2: create data files, append map data */
1446 FOR_EACH_COLUMN(i, t->desc, cd) {
1447 enum osl_storage_type st = cd->storage_type;
1448 if (st == OSL_NO_STORAGE)
1450 if (st == OSL_MAPPED_STORAGE) {
1452 struct osl_column *col = &t->columns[i];
1453 // DEBUG_LOG("appending object of size %zu\n",
1454 // objects[i].size);
1455 ret = append_map_file(t, i, objects + i, &new_size);
1458 update_cell_index(new_row_index, col, new_size,
1463 ret = write_disk_storage_file(t, i, objects + i, ds_name);
1467 ret = append_row_index(t, new_row_index);
1470 ret = map_table(t, MAP_TBL_FL_VERIFY_INDEX);
1471 if (ret < 0) { /* truncate index and rollback changes */
1472 char *filename = index_filename(t->desc);
1474 para_truncate(filename, t->row_index_size);
1478 /* pass 3: add entry to rbtrees */
1479 if (t->num_volatile_columns) {
1480 ret = -ERRNO_TO_ERROR(ENOMEM);
1481 volatile_objs = calloc(t->num_volatile_columns,
1482 sizeof(struct osl_object));
1485 FOR_EACH_VOLATILE_COLUMN(i, t, cd)
1486 volatile_objs[t->columns[i].volatile_num] = objects[i];
1489 // 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 // 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);
1518 __export int osl_add_row(struct osl_table *t, struct osl_object *objects)
1520 return osl_add_and_get_row(t, objects, NULL);
1523 __export int osl_get_object(const struct osl_table *t, const struct osl_row *r,
1524 unsigned col_num, struct osl_object *object)
1526 const struct osl_column_description *cd;
1529 return -E_OSL_BAD_TABLE;
1530 cd = get_column_description(t->desc, col_num);
1531 /* col must not be disk storage */
1532 if (cd->storage_type == OSL_DISK_STORAGE)
1533 return -E_OSL_BAD_STORAGE_TYPE;
1534 if (cd->storage_type == OSL_MAPPED_STORAGE)
1535 return get_mapped_object(t, col_num, r->num, object);
1537 *object = r->volatile_objects[t->columns[col_num].volatile_num];
1541 static int mark_mapped_object_invalid(const struct osl_table *t,
1542 uint32_t row_num, unsigned col_num)
1544 struct osl_object obj;
1546 int ret = get_mapped_object(t, col_num, row_num, &obj);
1556 __export int osl_del_row(struct osl_table *t, struct osl_row *row)
1558 struct osl_row *r = row;
1560 const struct osl_column_description *cd;
1563 return -E_OSL_BAD_TABLE;
1564 INFO_LOG("deleting row %p\n", row);
1566 if (t->num_disk_storage_columns) {
1568 ret = disk_storage_name_of_row(t, r, &ds_name);
1571 FOR_EACH_DISK_STORAGE_COLUMN(i, t, cd)
1572 delete_disk_storage_file(t, i, ds_name);
1575 FOR_EACH_COLUMN(i, t->desc, cd) {
1576 struct osl_column *col = t->columns + i;
1577 enum osl_storage_type st = cd->storage_type;
1578 remove_rb_node(t, i, r);
1579 if (st == OSL_MAPPED_STORAGE) {
1580 mark_mapped_object_invalid(t, r->num, i);
1583 if (st == OSL_NO_STORAGE && !(cd->storage_flags & OSL_DONT_FREE))
1584 free(r->volatile_objects[col->volatile_num].data);
1586 if (t->num_mapped_columns) {
1587 ret = mark_row_invalid(t, r->num);
1590 t->num_invalid_rows++;
1595 free(r->volatile_objects);
1600 /* test if column has an rbtree */
1601 static int check_rbtree_col(const struct osl_table *t, unsigned col_num,
1602 struct osl_column **col)
1605 return -E_OSL_BAD_TABLE;
1606 if (!(get_column_description(t->desc, col_num)->storage_flags & OSL_RBTREE))
1607 return -E_OSL_BAD_STORAGE_FLAGS;
1608 *col = t->columns + col_num;
1612 __export int osl_get_row(const struct osl_table *t, unsigned col_num,
1613 const struct osl_object *obj, struct osl_row **result)
1616 struct rb_node *node;
1617 struct osl_row *row;
1618 struct osl_column *col;
1621 ret = check_rbtree_col(t, col_num, &col);
1624 ret = search_rbtree(obj, t, col_num, &node, NULL);
1627 row = get_row_pointer(node, t->columns[col_num].rbtree_num);
1632 static int rbtree_loop(struct osl_column *col, void *private_data,
1633 osl_rbtree_loop_func *func)
1635 struct rb_node *n, *tmp;
1637 /* this for-loop is safe against removal of an entry */
1638 for (n = rb_first(&col->rbtree), tmp = n? rb_next(n) : NULL;
1640 n = tmp, tmp = tmp? rb_next(tmp) : NULL) {
1641 struct osl_row *r = get_row_pointer(n, col->rbtree_num);
1642 if (func(r, private_data) < 0)
1648 static int rbtree_loop_reverse(struct osl_column *col, void *private_data,
1649 osl_rbtree_loop_func *func)
1651 struct rb_node *n, *tmp;
1653 /* safe against removal of an entry */
1654 for (n = rb_last(&col->rbtree), tmp = n? rb_prev(n) : NULL;
1656 n = tmp, tmp = tmp? rb_prev(tmp) : NULL) {
1657 struct osl_row *r = get_row_pointer(n, col->rbtree_num);
1658 if (func(r, private_data) < 0)
1664 __export int osl_rbtree_loop(const struct osl_table *t, unsigned col_num,
1665 void *private_data, osl_rbtree_loop_func *func)
1667 struct osl_column *col;
1669 int ret = check_rbtree_col(t, col_num, &col);
1672 return rbtree_loop(col, private_data, func);
1675 __export int osl_rbtree_loop_reverse(const struct osl_table *t, unsigned col_num,
1676 void *private_data, osl_rbtree_loop_func *func)
1678 struct osl_column *col;
1680 int ret = check_rbtree_col(t, col_num, &col);
1683 return rbtree_loop_reverse(col, private_data, func);
1686 /* TODO: Rollback changes on errors */
1687 static int rename_disk_storage_objects(struct osl_table *t,
1688 struct osl_object *old_obj, struct osl_object *new_obj)
1691 const struct osl_column_description *cd;
1692 char *old_ds_name, *new_ds_name;
1694 if (!t->num_disk_storage_columns)
1695 return 1; /* nothing to do */
1696 if (old_obj->size == new_obj->size && !memcmp(new_obj->data,
1697 old_obj->data, new_obj->size))
1698 return 1; /* object did not change */
1699 old_ds_name = disk_storage_name_of_object(t, old_obj);
1700 new_ds_name = disk_storage_name_of_object(t, new_obj);
1701 ret = -ERRNO_TO_ERROR(ENOMEM);
1702 if (!old_ds_name || ! new_ds_name)
1705 FOR_EACH_DISK_STORAGE_COLUMN(i, t, cd) {
1706 char *old_filename, *new_filename;
1707 ret = create_disk_storage_object_dir(t, i, new_ds_name);
1710 old_filename = disk_storage_path(t, i, old_ds_name);
1711 new_filename = disk_storage_path(t, i, new_ds_name);
1712 if (!old_filename || !new_filename)
1713 ret = -ERRNO_TO_ERROR(ENOMEM);
1715 ret = para_rename(old_filename, new_filename);
1729 __export int osl_update_object(struct osl_table *t, const struct osl_row *r,
1730 unsigned col_num, struct osl_object *obj)
1732 struct osl_column *col;
1733 const struct osl_column_description *cd;
1737 return -E_OSL_BAD_TABLE;
1738 col = &t->columns[col_num];
1739 cd = get_column_description(t->desc, col_num);
1740 DEBUG_LOG("updating column %u of %s\n", col_num, t->desc->name);
1741 if (cd->storage_flags & OSL_RBTREE) {
1742 if (search_rbtree(obj, t, col_num, NULL, NULL) > 0)
1743 return -E_OSL_RB_KEY_EXISTS;
1745 if (cd->storage_flags & OSL_FIXED_SIZE) {
1746 if (obj->size != cd->data_size)
1747 return -E_OSL_BAD_DATA_SIZE;
1749 remove_rb_node(t, col_num, r);
1750 if (cd->storage_type == OSL_NO_STORAGE) { /* TODO: If fixed size, reuse object? */
1751 free(r->volatile_objects[col->volatile_num].data);
1752 r->volatile_objects[col->volatile_num] = *obj;
1753 } else if (cd->storage_type == OSL_DISK_STORAGE) {
1755 ret = disk_storage_name_of_row(t, r, &ds_name);
1758 ret = delete_disk_storage_file(t, col_num, ds_name);
1759 if (ret < 0 && !is_errno(-ret, ENOENT)) {
1763 ret = write_disk_storage_file(t, col_num, obj, ds_name);
1767 } else { /* mapped storage */
1768 struct osl_object old_obj;
1769 ret = get_mapped_object(t, col_num, r->num, &old_obj);
1773 * If the updated column is the disk storage name column, the
1774 * disk storage name changes, so we have to rename all disk
1775 * storage objects accordingly.
1777 if (col_num == t->disk_storage_name_column) {
1778 ret = rename_disk_storage_objects(t, &old_obj, obj);
1782 if (cd->storage_flags & OSL_FIXED_SIZE)
1783 memcpy(old_obj.data, obj->data, cd->data_size);
1784 else { /* TODO: if the size doesn't change, use old space */
1785 uint32_t new_data_map_size;
1787 ret = get_row_index(t, r->num, &row_index);
1790 ret = mark_mapped_object_invalid(t, r->num, col_num);
1793 unmap_column(t, col_num);
1794 ret = append_map_file(t, col_num, obj,
1795 &new_data_map_size);
1798 ret = map_column(t, col_num);
1801 update_cell_index(row_index, col, new_data_map_size,
1805 if (cd->storage_flags & OSL_RBTREE) {
1806 ret = insert_rbtree(t, col_num, r, obj);
1813 __export int osl_open_disk_object(const struct osl_table *t, const struct osl_row *r,
1814 unsigned col_num, struct osl_object *obj)
1816 const struct osl_column_description *cd;
1817 char *ds_name, *filename;
1821 return -E_OSL_BAD_TABLE;
1822 cd = get_column_description(t->desc, col_num);
1823 if (cd->storage_type != OSL_DISK_STORAGE)
1824 return -E_OSL_BAD_STORAGE_TYPE;
1826 ret = disk_storage_name_of_row(t, r, &ds_name);
1829 filename = disk_storage_path(t, col_num, ds_name);
1832 return -ERRNO_TO_ERROR(ENOMEM);
1833 DEBUG_LOG("filename: %s\n", filename);
1834 ret = mmap_full_file(filename, O_RDONLY, &obj->data, &obj->size, NULL);
1839 __export int osl_close_disk_object(struct osl_object *obj)
1841 return para_munmap(obj->data, obj->size);
1844 __export int osl_get_num_rows(const struct osl_table *t, unsigned *num_rows)
1847 return -E_OSL_BAD_TABLE;
1848 assert(t->num_rows >= t->num_invalid_rows);
1849 *num_rows = t->num_rows - t->num_invalid_rows;
1853 __export int osl_get_rank(const struct osl_table *t, struct osl_row *r,
1854 unsigned col_num, unsigned *rank)
1856 struct osl_object obj;
1857 struct osl_column *col;
1858 struct rb_node *node;
1859 int ret = check_rbtree_col(t, col_num, &col);
1863 ret = osl_get_object(t, r, col_num, &obj);
1866 ret = search_rbtree(&obj, t, col_num, &node, NULL);
1869 ret = rb_rank(node, rank);
1871 return -E_OSL_BAD_ROW;
1875 __export int osl_get_nth_row(const struct osl_table *t, unsigned col_num,
1876 unsigned n, struct osl_row **result)
1878 struct osl_column *col;
1879 struct rb_node *node;
1884 return -E_OSL_RB_KEY_NOT_FOUND;
1885 ret = osl_get_num_rows(t, &num_rows);
1889 return -E_OSL_RB_KEY_NOT_FOUND;
1890 ret = check_rbtree_col(t, col_num, &col);
1893 node = rb_nth(col->rbtree.rb_node, n);
1895 return -E_OSL_RB_KEY_NOT_FOUND;
1896 *result = get_row_pointer(node, col->rbtree_num);
1900 __export int osl_rbtree_first_row(const struct osl_table *t, unsigned col_num,
1901 struct osl_row **result)
1903 return osl_get_nth_row(t, col_num, 1, result);
1906 __export int osl_rbtree_last_row(const struct osl_table *t, unsigned col_num,
1907 struct osl_row **result)
1910 int ret = osl_get_num_rows(t, &num_rows);
1914 return osl_get_nth_row(t, col_num, num_rows, result);