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];
97 * \param fmt Usual format string.
99 * All XXX_LOG() macros use this function.
101 __printf_2_3 void __log(int ll, const char* fmt,...)
114 strftime(str, sizeof(str), "%b %d %H:%M:%S", tm);
115 fprintf(outfd, "%s ", str);
117 vfprintf(outfd, fmt, argp);
122 * A wrapper for lseek(2).
124 * \param fd The file descriptor whose offset is to be to repositioned.
125 * \param offset A value-result parameter.
126 * \param whence Usual repositioning directive.
128 * Reposition the offset of the file descriptor \a fd to the argument \a offset
129 * according to the directive \a whence. Upon successful return, \a offset
130 * contains the resulting offset location as measured in bytes from the
131 * beginning of the file.
133 * \return Positive on success. Otherwise, the function returns \p -E_OSL_LSEEK.
137 static int __lseek(int fd, off_t *offset, int whence)
139 *offset = lseek(fd, *offset, whence);
140 int ret = -E_OSL_LSEEK;
147 * Wrapper for the write system call.
149 * \param fd The file descriptor to write to.
150 * \param buf The buffer to write.
151 * \param size The length of \a buf in bytes.
153 * This function writes out the given buffer and retries if an interrupt
154 * occurred during the write.
156 * \return On success, the number of bytes written is returned, otherwise, the
157 * function returns \p -E_OSL_WRITE.
161 static ssize_t __write(int fd, const void *buf, size_t size)
166 ret = write(fd, buf, size);
167 if ((ret < 0) && (errno == EAGAIN || errno == EINTR))
169 return ret >= 0? ret : -E_OSL_WRITE;
174 * Write the whole buffer to a file descriptor.
176 * \param fd The file descriptor to write to.
177 * \param buf The buffer to write.
178 * \param size The length of \a buf in bytes.
180 * This function writes the given buffer and continues on short writes and
181 * when interrupted by a signal.
183 * \return Positive on success, negative on errors. Possible errors: any
184 * errors returned by para_write().
188 static ssize_t write_all(int fd, const void *buf, size_t size)
190 // DEBUG_LOG("writing %zu bytes\n", size);
193 ssize_t ret = __write(fd, b, size);
194 // DEBUG_LOG("ret: %zd\n", ret);
203 * Open a file, write the given buffer and close the file.
205 * \param filename Full path to the file to open.
206 * \param buf The buffer to write to the file.
207 * \param size The size of \a buf.
211 static int write_file(const char *filename, const void *buf, size_t size)
215 ret = osl_open(filename, O_WRONLY | O_CREAT | O_EXCL, 0644);
219 ret = write_all(fd, buf, size);
228 static int append_file(const char *filename, char *header, size_t header_size,
229 char *data, size_t data_size, uint32_t *new_pos)
233 // DEBUG_LOG("appending %zu + %zu bytes\n", header_size, data_size);
234 ret = osl_open(filename, O_WRONLY | O_CREAT | O_APPEND, 0644);
238 if (header && header_size) {
239 ret = write_all(fd, header, header_size);
243 ret = write_all(fd, data, data_size);
248 ret = __lseek(fd, &offset, SEEK_END);
251 // DEBUG_LOG("new file size: " FMT_OFF_T "\n", offset);
260 static int verify_name(const char *name)
263 return -E_OSL_BAD_NAME;
265 return -E_OSL_BAD_NAME;
266 if (strchr(name, '/'))
267 return -E_OSL_BAD_NAME;
268 if (!strcmp(name, ".."))
269 return -E_OSL_BAD_NAME;
270 if (!strcmp(name, "."))
271 return -E_OSL_BAD_NAME;
276 * Compare two osl objects pointing to hash values.
278 * \param obj1 Pointer to the first hash object.
279 * \param obj2 Pointer to the second hash object.
281 * \return The values required for an osl compare function.
283 * \sa osl_compare_func, uint32_compare().
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 DEBUG_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 DEBUG_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);
663 * Create a new osl table.
665 * \param desc Pointer to the table description.
669 __export int osl_create_table(const struct osl_table_description *desc)
671 const struct osl_column_description *cd;
672 char *table_dir = NULL, *filename;
674 int i, ret = init_table_structure(desc, &t);
678 INFO_LOG("creating %s\n", desc->name);
679 FOR_EACH_COLUMN(i, t->desc, cd) {
680 if (cd->storage_type == OSL_NO_STORAGE)
683 ret = para_mkdir(desc->dir, 0777);
684 if (ret < 0 && !is_errno(-ret, EEXIST))
686 table_dir = make_message("%s/%s", desc->dir,
688 ret = -ERRNO_TO_ERROR(ENOMEM);
691 ret = para_mkdir(table_dir, 0777);
695 ret = -ERRNO_TO_ERROR(ENOMEM);
696 filename = column_filename(t, i);
699 INFO_LOG("filename: %s\n", filename);
700 if (cd->storage_type == OSL_MAPPED_STORAGE) {
701 ret = osl_open(filename, O_RDWR | O_CREAT | O_EXCL,
710 ret = para_mkdir(filename, 0777);
715 if (t->num_mapped_columns) {
716 ret = create_table_index(t);
728 static int table_is_dirty(struct osl_table *t)
730 char *buf = (char *)t->index_map.data + IDX_DIRTY_FLAG;
731 uint8_t dirty = read_u8(buf) & 0x1;
735 static void mark_table_dirty(struct osl_table *t)
737 char *buf = (char *)t->index_map.data + IDX_DIRTY_FLAG;
738 write_u8(buf, read_u8(buf) | 1);
741 static void mark_table_clean(struct osl_table *t)
743 char *buf = (char *)t->index_map.data + IDX_DIRTY_FLAG;
744 write_u8(buf, read_u8(buf) & 0xfe);
747 static void unmap_column(struct osl_table *t, unsigned col_num)
749 struct osl_object map = t->columns[col_num].data_map;
753 ret = para_munmap(map.data, map.size);
759 * Unmap all mapped files of an osl table.
761 * \param t Pointer to a mapped table.
762 * \param flags Options for unmapping.
764 * \return Positive on success, negative on errors.
766 * \sa map_table(), enum osl_close_flags, para_munmap().
768 int unmap_table(struct osl_table *t, enum osl_close_flags flags)
771 const struct osl_column_description *cd;
774 if (!t->num_mapped_columns) /* can this ever happen? */
776 DEBUG_LOG("unmapping table '%s'\n", t->desc->name);
777 if (!t->index_map.data)
778 return -E_OSL_NOT_MAPPED;
779 if (flags & OSL_MARK_CLEAN)
781 ret = para_munmap(t->index_map.data, t->index_map.size);
784 t->index_map.data = NULL;
787 FOR_EACH_MAPPED_COLUMN(i, t, cd)
792 static int map_column(struct osl_table *t, unsigned col_num)
795 char *filename = column_filename(t, col_num);
796 int ret = -E_OSL_STAT;
799 return -ERRNO_TO_ERROR(ENOMEM);
800 if (stat(filename, &statbuf) < 0) {
804 if (!(S_IFREG & statbuf.st_mode)) {
808 ret = mmap_full_file(filename, O_RDWR,
809 &t->columns[col_num].data_map.data,
810 &t->columns[col_num].data_map.size,
817 * Map the index file and all columns of type \p OSL_MAPPED_STORAGE into memory.
819 * \param t Pointer to an initialized table structure.
820 * \param flags Mapping options.
822 * \return Negative return value on errors; on success the number of rows
823 * (including invalid rows) is returned.
825 * \sa unmap_table(), enum map_table_flags, osl_open_table(), mmap(2).
827 int map_table(struct osl_table *t, enum map_table_flags flags)
830 const struct osl_column_description *cd;
831 int i = 0, ret, num_rows = 0;
833 if (!t->num_mapped_columns)
835 if (t->index_map.data)
836 return -E_OSL_ALREADY_MAPPED;
837 filename = index_filename(t->desc);
839 return -ERRNO_TO_ERROR(ENOMEM);
840 DEBUG_LOG("mapping table '%s' (index: %s)\n", t->desc->name, filename);
841 ret = mmap_full_file(filename, flags & MAP_TBL_FL_MAP_RDONLY?
842 O_RDONLY : O_RDWR, &t->index_map.data, &t->index_map.size, NULL);
846 if (flags & MAP_TBL_FL_VERIFY_INDEX) {
847 ret = compare_table_descriptions(t);
852 if (!(flags & MAP_TBL_FL_IGNORE_DIRTY)) {
853 if (table_is_dirty(t)) {
854 ERROR_LOG("%s is dirty\n", t->desc->name);
859 num_rows = table_num_rows(t);
863 FOR_EACH_MAPPED_COLUMN(i, t, cd) {
864 ret = map_column(t, i);
869 err: /* unmap what is already mapped */
870 for (i--; i >= 0; i--) {
871 struct osl_object map = t->columns[i].data_map;
872 para_munmap(map.data, map.size);
875 para_munmap(t->index_map.data, t->index_map.size);
876 t->index_map.data = NULL;
881 * Retrieve a mapped object by row and column number.
883 * \param t Pointer to an open osl table.
884 * \param col_num Number of the mapped column containing the object to retrieve.
885 * \param row_num Number of the row containing the object to retrieve.
886 * \param obj The result is returned here.
888 * It is considered an error if \a col_num does not refer to a column
889 * of storage type \p OSL_MAPPED_STORAGE.
893 * \sa osl_storage_type.
895 int get_mapped_object(const struct osl_table *t, unsigned col_num,
896 uint32_t row_num, struct osl_object *obj)
898 struct osl_column *col = &t->columns[col_num];
904 if (t->num_rows <= row_num)
905 return -E_OSL_BAD_ROW_NUM;
906 ret = get_cell_index(t, row_num, col_num, &cell_index);
909 offset = read_u32(cell_index);
910 obj->size = read_u32(cell_index + 4) - 1;
911 header = col->data_map.data + offset;
912 obj->data = header + 1;
913 if (read_u8(header) == 0xff) {
914 ERROR_LOG("col %u, size %zu, offset %u\n", col_num,
916 return -E_OSL_INVALID_OBJECT;
921 static int search_rbtree(const struct osl_object *obj,
922 const struct osl_table *t, unsigned col_num,
923 struct rb_node **result, struct rb_node ***rb_link)
925 struct osl_column *col = &t->columns[col_num];
926 struct rb_node **new = &col->rbtree.rb_node, *parent = NULL;
927 const struct osl_column_description *cd =
928 get_column_description(t->desc, col_num);
929 enum osl_storage_type st = cd->storage_type;
931 struct osl_row *this_row = get_row_pointer(*new,
934 struct osl_object this_obj;
936 if (st == OSL_MAPPED_STORAGE) {
937 ret = get_mapped_object(t, col_num, this_row->num,
942 this_obj = this_row->volatile_objects[col->volatile_num];
943 ret = cd->compare_function(obj, &this_obj);
946 *result = get_rb_node_pointer(this_row,
951 new = &((*new)->rb_left);
953 new = &((*new)->rb_right);
959 return -E_OSL_RB_KEY_NOT_FOUND;
962 static int insert_rbtree(struct osl_table *t, unsigned col_num,
963 const struct osl_row *row, const struct osl_object *obj)
965 struct rb_node *parent, **rb_link;
968 int ret = search_rbtree(obj, t, col_num, &parent, &rb_link);
971 return -E_OSL_RB_KEY_EXISTS;
972 rbtree_num = t->columns[col_num].rbtree_num;
973 n = get_rb_node_pointer(row, rbtree_num);
974 rb_link_node(n, parent, rb_link);
975 rb_insert_color(n, &t->columns[col_num].rbtree);
979 static void remove_rb_node(struct osl_table *t, unsigned col_num,
980 const struct osl_row *row)
982 struct osl_column *col = &t->columns[col_num];
983 const struct osl_column_description *cd =
984 get_column_description(t->desc, col_num);
985 enum osl_storage_flags sf = cd->storage_flags;
986 struct rb_node *victim, *splice_out_node, *tmp;
987 if (!(sf & OSL_RBTREE))
990 * Which node is removed/spliced out actually depends on how many
991 * children the victim node has: If it has no children, it gets
992 * deleted. If it has one child, it gets spliced out. If it has two
993 * children, its successor (which has at most a right child) gets
996 victim = get_rb_node_pointer(row, col->rbtree_num);
997 if (victim->rb_left && victim->rb_right)
998 splice_out_node = rb_next(victim);
1000 splice_out_node = victim;
1001 /* Go up to the root and decrement the size of each node in the path. */
1002 for (tmp = splice_out_node; tmp; tmp = rb_parent(tmp))
1004 rb_erase(victim, &col->rbtree);
1007 static int add_row_to_rbtrees(struct osl_table *t, uint32_t row_num,
1008 struct osl_object *volatile_objs, struct osl_row **row_ptr)
1012 struct osl_row *row = allocate_row(t->num_rbtrees);
1013 const struct osl_column_description *cd;
1016 return -ERRNO_TO_ERROR(ENOMEM);
1018 row->volatile_objects = volatile_objs;
1019 FOR_EACH_RBTREE_COLUMN(i, t, cd) {
1020 if (cd->storage_type == OSL_MAPPED_STORAGE) {
1021 struct osl_object obj;
1022 ret = get_mapped_object(t, i, row_num, &obj);
1025 ret = insert_rbtree(t, i, row, &obj);
1026 } else { /* volatile */
1027 const struct osl_object *obj
1028 = volatile_objs + t->columns[i].volatile_num;
1029 ret = insert_rbtree(t, i, row, obj);
1037 err: /* rollback changes, i.e. remove added entries from rbtrees */
1039 remove_rb_node(t, i--, row);
1044 static void free_volatile_objects(const struct osl_table *t,
1045 enum osl_close_flags flags)
1049 struct osl_column *rb_col;
1050 const struct osl_column_description *cd;
1052 if (!t->num_volatile_columns)
1054 /* find the first rbtree column (any will do) */
1055 FOR_EACH_RBTREE_COLUMN(i, t, cd)
1057 rb_col = t->columns + i;
1058 /* walk that rbtree and free all volatile objects */
1059 for (n = rb_first(&rb_col->rbtree); n; n = rb_next(n)) {
1060 struct osl_row *r = get_row_pointer(n, rb_col->rbtree_num);
1061 if (flags & OSL_FREE_VOLATILE)
1062 FOR_EACH_VOLATILE_COLUMN(j, t, cd) {
1063 if (cd->storage_flags & OSL_DONT_FREE)
1065 free(r->volatile_objects[
1066 t->columns[j].volatile_num].data);
1068 // for (j = 0; j < t->num_volatile_columns; j++)
1069 // free(r->volatile_objects[j].data);
1070 free(r->volatile_objects);
1075 * Erase all rbtree nodes and free resources.
1077 * \param t Pointer to an open osl table.
1079 * This function is called by osl_close_table().
1081 void clear_rbtrees(struct osl_table *t)
1083 const struct osl_column_description *cd;
1084 unsigned i, rbtrees_cleared = 0;
1086 FOR_EACH_RBTREE_COLUMN(i, t, cd) {
1087 struct osl_column *col = &t->columns[i];
1090 for (n = rb_first(&col->rbtree); n;) {
1092 rb_erase(n, &col->rbtree);
1093 if (rbtrees_cleared == t->num_rbtrees) {
1094 r = get_row_pointer(n, col->rbtree_num);
1105 * Close an osl table.
1107 * \param t Pointer to the table to be closed.
1108 * \param flags Options for what should be cleaned up.
1110 * If osl_open_table() succeeds, the resulting table pointer must later be
1111 * passed to this function in order to flush all changes to the file system and
1112 * to free the resources that were allocated by osl_open_table().
1116 * \sa osl_open_table(), unmap_table().
1118 __export int osl_close_table(struct osl_table *t, enum osl_close_flags flags)
1123 return -E_OSL_BAD_TABLE;
1124 free_volatile_objects(t, flags);
1126 ret = unmap_table(t, flags);
1128 ERROR_LOG("unmap_table failed: %d\n", ret);
1135 * Find out whether the given row number corresponds to an invalid row.
1137 * \param t Pointer to the osl table.
1138 * \param row_num The number of the row in question.
1140 * By definition, a row is considered invalid if all its index entries
1143 * \return Positive if \a row_num corresponds to an invalid row,
1144 * zero if it corresponds to a valid row, negative on errors.
1146 int row_is_invalid(struct osl_table *t, uint32_t row_num)
1149 int i, ret = get_row_index(t, row_num, &row_index);
1153 for (i = 0; i < t->row_index_size; i++) {
1154 if ((unsigned char)row_index[i] != 0xff)
1157 INFO_LOG("row %d is invalid\n", row_num);
1162 * Invalidate a row of an osl table.
1164 * \param t Pointer to an open osl table.
1165 * \param row_num Number of the row to mark as invalid.
1167 * This function marks each mapped object in the index entry of \a row as
1172 int mark_row_invalid(struct osl_table *t, uint32_t row_num)
1175 int ret = get_row_index(t, row_num, &row_index);
1179 INFO_LOG("marking row %d as invalid\n", row_num);
1180 memset(row_index, 0xff, t->row_index_size);
1185 * Initialize all rbtrees and compute number of invalid rows.
1187 * \param t The table containing the rbtrees to be initialized.
1191 int init_rbtrees(struct osl_table *t)
1194 const struct osl_column_description *cd;
1196 /* create rbtrees */
1197 FOR_EACH_RBTREE_COLUMN(i, t, cd)
1198 t->columns[i].rbtree = RB_ROOT;
1199 /* add valid rows to rbtrees */
1200 t->num_invalid_rows = 0;
1201 for (i = 0; i < t->num_rows; i++) {
1202 ret = row_is_invalid(t, i);
1206 t->num_invalid_rows++;
1209 ret = add_row_to_rbtrees(t, i, NULL, NULL);
1217 * Open an osl table.
1219 * Each osl table must be opened before its data can be accessed.
1221 * \param table_desc Describes the table to be opened.
1222 * \param result Contains a pointer to the open table on success.
1224 * The table description given by \a desc should coincide with the
1225 * description used at creation time.
1229 __export int osl_open_table(const struct osl_table_description *table_desc,
1230 struct osl_table **result)
1233 struct osl_table *t;
1234 const struct osl_column_description *cd;
1236 INFO_LOG("opening table %s\n", table_desc->name);
1237 ret = init_table_structure(table_desc, &t);
1240 FOR_EACH_DISK_STORAGE_COLUMN(i, t, cd) {
1241 struct stat statbuf;
1242 char *dirname = column_filename(t, i);
1244 ret = -ERRNO_TO_ERROR(ENOMEM);
1247 /* check if directory exists */
1248 ret = stat(dirname, &statbuf);
1251 ret = -ERRNO_TO_ERROR(errno);
1254 ret = -ERRNO_TO_ERROR(ENOTDIR);
1255 if (!S_ISDIR(statbuf.st_mode))
1258 ret = map_table(t, MAP_TBL_FL_VERIFY_INDEX);
1262 DEBUG_LOG("num rows: %d\n", t->num_rows);
1263 ret = init_rbtrees(t);
1265 osl_close_table(t, OSL_MARK_CLEAN); /* ignore further errors */
1276 static int create_disk_storage_object_dir(const struct osl_table *t,
1277 unsigned col_num, const char *ds_name)
1282 if (!(t->desc->flags & OSL_LARGE_TABLE))
1284 dirname = disk_storage_dirname(t, col_num, ds_name);
1286 return -ERRNO_TO_ERROR(ENOMEM);
1287 ret = para_mkdir(dirname, 0777);
1289 if (ret < 0 && !is_errno(-ret, EEXIST))
1294 static int write_disk_storage_file(const struct osl_table *t, unsigned col_num,
1295 const struct osl_object *obj, const char *ds_name)
1300 ret = create_disk_storage_object_dir(t, col_num, ds_name);
1303 filename = disk_storage_path(t, col_num, ds_name);
1305 return -ERRNO_TO_ERROR(ENOMEM);
1306 ret = write_file(filename, obj->data, obj->size);
1311 static int append_map_file(const struct osl_table *t, unsigned col_num,
1312 const struct osl_object *obj, uint32_t *new_size)
1314 char *filename = column_filename(t, col_num);
1316 char header = 0; /* zero means valid object */
1319 return -ERRNO_TO_ERROR(ENOMEM);
1320 ret = append_file(filename, &header, 1, obj->data, obj->size,
1326 static int append_row_index(const struct osl_table *t, char *row_index)
1331 if (!t->num_mapped_columns)
1333 filename = index_filename(t->desc);
1335 return -ERRNO_TO_ERROR(ENOMEM);
1336 ret = append_file(filename, NULL, 0, row_index,
1337 t->row_index_size, NULL);
1343 * A wrapper for truncate(2)
1345 * \param path Name of the regular file to truncate
1346 * \param size Number of bytes to \b shave \b off
1348 * Truncate the regular file named by \a path by \a size bytes.
1354 int para_truncate(const char *path, off_t size)
1357 struct stat statbuf;
1360 if (stat(path, &statbuf) < 0)
1362 ret = -E_OSL_BAD_SIZE;
1363 if (statbuf.st_size < size)
1366 if (truncate(path, statbuf.st_size - size) < 0)
1373 static int truncate_mapped_file(const struct osl_table *t, unsigned col_num,
1377 char *filename = column_filename(t, col_num);
1380 return -ERRNO_TO_ERROR(ENOMEM);
1381 ret = para_truncate(filename, size);
1386 static int delete_disk_storage_file(const struct osl_table *t, unsigned col_num,
1387 const char *ds_name)
1389 char *dirname, *filename = disk_storage_path(t, col_num, ds_name);
1393 return -ERRNO_TO_ERROR(ENOMEM);
1394 ret = unlink(filename);
1398 return -ERRNO_TO_ERROR(err);
1399 if (!(t->desc->flags & OSL_LARGE_TABLE))
1401 dirname = disk_storage_dirname(t, col_num, ds_name);
1403 return -ERRNO_TO_ERROR(ENOMEM);
1410 * Add a new row to an osl table and retrieve this row.
1412 * \param t Pointer to an open osl table.
1413 * \param objects Array of objects to be added.
1414 * \param row Result pointer.
1416 * The \a objects parameter must point to an array containing one object per
1417 * column. The order of the objects in the array is given by the table
1418 * description of \a table. Several sanity checks are performed during object
1419 * insertion and the function returns without modifying the table if any of
1420 * these tests fail. In fact, it is atomic in the sense that it either
1421 * succeeds or leaves the table unchanged (i.e. either all or none of the
1422 * objects are added to the table).
1424 * It is considered an error if an object is added to a column with associated
1425 * rbtree if this object is equal to an object already contained in that column
1426 * (i.e. the compare function for the column's rbtree returns zero).
1430 * \sa struct osl_table_description, osl_compare_func, osl_add_row().
1432 __export int osl_add_and_get_row(struct osl_table *t, struct osl_object *objects,
1433 struct osl_row **row)
1436 char *ds_name = NULL;
1437 struct rb_node **rb_parents = NULL, ***rb_links = NULL;
1438 char *new_row_index = NULL;
1439 struct osl_object *volatile_objs = NULL;
1440 const struct osl_column_description *cd;
1443 return -E_OSL_BAD_TABLE;
1444 rb_parents = malloc(t->num_rbtrees * sizeof(struct rn_node*));
1446 return -ERRNO_TO_ERROR(ENOMEM);
1447 rb_links = malloc(t->num_rbtrees * sizeof(struct rn_node**));
1450 return -ERRNO_TO_ERROR(ENOMEM);
1452 if (t->num_mapped_columns) {
1453 new_row_index = malloc(t->row_index_size);
1454 if (!new_row_index) {
1457 return -ERRNO_TO_ERROR(ENOMEM);
1460 /* pass 1: sanity checks */
1461 // DEBUG_LOG("sanity tests: %p:%p\n", objects[0].data,
1462 // objects[1].data);
1463 FOR_EACH_COLUMN(i, t->desc, cd) {
1464 enum osl_storage_type st = cd->storage_type;
1465 enum osl_storage_flags sf = cd->storage_flags;
1467 // ret = -E_OSL_NULL_OBJECT;
1470 if (st == OSL_DISK_STORAGE)
1472 if (sf & OSL_RBTREE) {
1473 unsigned rbtree_num = t->columns[i].rbtree_num;
1474 ret = -E_OSL_RB_KEY_EXISTS;
1475 // DEBUG_LOG("checking whether %p exists\n",
1476 // objects[i].data);
1477 if (search_rbtree(objects + i, t, i,
1478 &rb_parents[rbtree_num],
1479 &rb_links[rbtree_num]) > 0)
1482 if (sf & OSL_FIXED_SIZE) {
1483 // DEBUG_LOG("fixed size. need: %zu, have: %d\n",
1484 // objects[i].size, cd->data_size);
1485 ret = -E_OSL_BAD_DATA_SIZE;
1486 if (objects[i].size != cd->data_size)
1490 if (t->num_disk_storage_columns) {
1491 ds_name = disk_storage_name_of_object(t,
1492 &objects[t->disk_storage_name_column]);
1493 ret = -ERRNO_TO_ERROR(ENOMEM);
1497 ret = unmap_table(t, OSL_MARK_CLEAN);
1500 // DEBUG_LOG("sanity tests passed%s\n", "");
1501 /* pass 2: create data files, append map data */
1502 FOR_EACH_COLUMN(i, t->desc, cd) {
1503 enum osl_storage_type st = cd->storage_type;
1504 if (st == OSL_NO_STORAGE)
1506 if (st == OSL_MAPPED_STORAGE) {
1508 struct osl_column *col = &t->columns[i];
1509 // DEBUG_LOG("appending object of size %zu\n",
1510 // objects[i].size);
1511 ret = append_map_file(t, i, objects + i, &new_size);
1514 update_cell_index(new_row_index, col, new_size,
1519 ret = write_disk_storage_file(t, i, objects + i, ds_name);
1523 ret = append_row_index(t, new_row_index);
1526 ret = map_table(t, MAP_TBL_FL_VERIFY_INDEX);
1527 if (ret < 0) { /* truncate index and rollback changes */
1528 char *filename = index_filename(t->desc);
1530 para_truncate(filename, t->row_index_size);
1534 /* pass 3: add entry to rbtrees */
1535 if (t->num_volatile_columns) {
1536 ret = -ERRNO_TO_ERROR(ENOMEM);
1537 volatile_objs = calloc(t->num_volatile_columns,
1538 sizeof(struct osl_object));
1541 FOR_EACH_VOLATILE_COLUMN(i, t, cd)
1542 volatile_objs[t->columns[i].volatile_num] = objects[i];
1545 // DEBUG_LOG("adding new entry as row #%d\n", t->num_rows - 1);
1546 ret = add_row_to_rbtrees(t, t->num_rows - 1, volatile_objs, row);
1549 // DEBUG_LOG("added new entry as row #%d\n", t->num_rows - 1);
1552 rollback: /* rollback all changes made, ignore further errors */
1553 for (i--; i >= 0; i--) {
1554 cd = get_column_description(t->desc, i);
1555 enum osl_storage_type st = cd->storage_type;
1556 if (st == OSL_NO_STORAGE)
1559 if (st == OSL_MAPPED_STORAGE)
1560 truncate_mapped_file(t, i, objects[i].size);
1561 else /* disk storage */
1562 delete_disk_storage_file(t, i, ds_name);
1564 /* ignore error and return previous error value */
1565 map_table(t, MAP_TBL_FL_VERIFY_INDEX);
1567 free(new_row_index);
1575 * Add a new row to an osl table.
1577 * \param t Same meaning as osl_add_and_get_row().
1578 * \param objects Same meaning as osl_add_and_get_row().
1580 * \return The return value of the underlying call to osl_add_and_get_row().
1582 * This is equivalent to osl_add_and_get_row(t, objects, NULL).
1584 __export int osl_add_row(struct osl_table *t, struct osl_object *objects)
1586 return osl_add_and_get_row(t, objects, NULL);
1590 * Retrieve an object identified by row and column
1592 * \param t Pointer to an open osl table.
1593 * \param r Pointer to the row.
1594 * \param col_num The column number.
1595 * \param object The result pointer.
1597 * The column determined by \a col_num must be of type \p OSL_MAPPED_STORAGE
1598 * or \p OSL_NO_STORAGE, i.e. no disk storage objects may be retrieved by this
1603 * \sa osl_storage_type, osl_open_disk_object().
1605 __export int osl_get_object(const struct osl_table *t, const struct osl_row *r,
1606 unsigned col_num, struct osl_object *object)
1608 const struct osl_column_description *cd;
1611 return -E_OSL_BAD_TABLE;
1612 cd = get_column_description(t->desc, col_num);
1613 /* col must not be disk storage */
1614 if (cd->storage_type == OSL_DISK_STORAGE)
1615 return -E_OSL_BAD_STORAGE_TYPE;
1616 if (cd->storage_type == OSL_MAPPED_STORAGE)
1617 return get_mapped_object(t, col_num, r->num, object);
1619 *object = r->volatile_objects[t->columns[col_num].volatile_num];
1623 static int mark_mapped_object_invalid(const struct osl_table *t,
1624 uint32_t row_num, unsigned col_num)
1626 struct osl_object obj;
1628 int ret = get_mapped_object(t, col_num, row_num, &obj);
1639 * Delete a row from an osl table.
1641 * \param t Pointer to an open osl table.
1642 * \param row Pointer to the row to delete.
1644 * This removes all disk storage objects, removes all rbtree nodes, and frees
1645 * all volatile objects belonging to the given row. For mapped columns, the
1646 * data is merely marked invalid and may be pruned from time to time by
1651 __export int osl_del_row(struct osl_table *t, struct osl_row *row)
1653 struct osl_row *r = row;
1655 const struct osl_column_description *cd;
1658 return -E_OSL_BAD_TABLE;
1659 INFO_LOG("deleting row %p\n", row);
1661 if (t->num_disk_storage_columns) {
1663 ret = disk_storage_name_of_row(t, r, &ds_name);
1666 FOR_EACH_DISK_STORAGE_COLUMN(i, t, cd)
1667 delete_disk_storage_file(t, i, ds_name);
1670 FOR_EACH_COLUMN(i, t->desc, cd) {
1671 struct osl_column *col = t->columns + i;
1672 enum osl_storage_type st = cd->storage_type;
1673 remove_rb_node(t, i, r);
1674 if (st == OSL_MAPPED_STORAGE) {
1675 mark_mapped_object_invalid(t, r->num, i);
1678 if (st == OSL_NO_STORAGE && !(cd->storage_flags & OSL_DONT_FREE))
1679 free(r->volatile_objects[col->volatile_num].data);
1681 if (t->num_mapped_columns) {
1682 ret = mark_row_invalid(t, r->num);
1685 t->num_invalid_rows++;
1690 free(r->volatile_objects);
1695 /* test if column has an rbtree */
1696 static int check_rbtree_col(const struct osl_table *t, unsigned col_num,
1697 struct osl_column **col)
1700 return -E_OSL_BAD_TABLE;
1701 if (!(get_column_description(t->desc, col_num)->storage_flags & OSL_RBTREE))
1702 return -E_OSL_BAD_STORAGE_FLAGS;
1703 *col = t->columns + col_num;
1708 * Get the row that contains the given object.
1710 * \param t Pointer to an open osl table.
1711 * \param col_num The number of the column to be searched.
1712 * \param obj The object to be looked up.
1713 * \param result Points to the row containing \a obj.
1715 * Lookup \a obj in \a t and return the row containing \a obj. The column
1716 * specified by \a col_num must have an associated rbtree.
1720 * \sa osl_storage_flags
1722 __export int osl_get_row(const struct osl_table *t, unsigned col_num,
1723 const struct osl_object *obj, struct osl_row **result)
1726 struct rb_node *node;
1727 struct osl_row *row;
1728 struct osl_column *col;
1731 ret = check_rbtree_col(t, col_num, &col);
1734 ret = search_rbtree(obj, t, col_num, &node, NULL);
1737 row = get_row_pointer(node, t->columns[col_num].rbtree_num);
1742 static int rbtree_loop(struct osl_column *col, void *private_data,
1743 osl_rbtree_loop_func *func)
1745 struct rb_node *n, *tmp;
1747 /* this for-loop is safe against removal of an entry */
1748 for (n = rb_first(&col->rbtree), tmp = n? rb_next(n) : NULL;
1750 n = tmp, tmp = tmp? rb_next(tmp) : NULL) {
1751 struct osl_row *r = get_row_pointer(n, col->rbtree_num);
1752 int ret = func(r, private_data);
1759 static int rbtree_loop_reverse(struct osl_column *col, void *private_data,
1760 osl_rbtree_loop_func *func)
1762 struct rb_node *n, *tmp;
1764 /* safe against removal of an entry */
1765 for (n = rb_last(&col->rbtree), tmp = n? rb_prev(n) : NULL;
1767 n = tmp, tmp = tmp? rb_prev(tmp) : NULL) {
1768 struct osl_row *r = get_row_pointer(n, col->rbtree_num);
1769 int ret = func(r, private_data);
1777 * Loop over all nodes in an rbtree.
1779 * \param t Pointer to an open osl table.
1780 * \param col_num The column to use for iterating over the elements.
1781 * \param private_data Pointer that gets passed to \a func.
1782 * \param func The function to be called for each node in the rbtree.
1784 * This function does an in-order walk of the rbtree associated with \a
1785 * col_num. It is an error if the \p OSL_RBTREE flag is not set for this
1786 * column. For each node in the rbtree, the given function \a func is called
1787 * with two pointers as arguments: The first osl_row* argument points to the
1788 * row that contains the object corresponding to the rbtree node currently
1789 * traversed, and the \a private_data pointer is passed verbatim to \a func as the
1790 * second argument. The loop terminates either if \a func returns a negative
1791 * value, or if all nodes of the tree have been visited.
1794 * \return Standard. If the termination of the loop was caused by \a func
1795 * returning a negative value, this value is returned.
1797 * \sa osl_storage_flags, osl_rbtree_loop_reverse(), osl_compare_func.
1799 __export int osl_rbtree_loop(const struct osl_table *t, unsigned col_num,
1800 void *private_data, osl_rbtree_loop_func *func)
1802 struct osl_column *col;
1804 int ret = check_rbtree_col(t, col_num, &col);
1807 return rbtree_loop(col, private_data, func);
1811 * Loop over all nodes in an rbtree in reverse order.
1813 * \param t Identical meaning as in \p osl_rbtree_loop().
1814 * \param col_num Identical meaning as in \p osl_rbtree_loop().
1815 * \param private_data Identical meaning as in \p osl_rbtree_loop().
1816 * \param func Identical meaning as in \p osl_rbtree_loop().
1818 * This function is identical to \p osl_rbtree_loop(), the only difference
1819 * is that the tree is walked in reverse order.
1821 * \return The same return value as \p osl_rbtree_loop().
1823 * \sa osl_rbtree_loop().
1825 __export int osl_rbtree_loop_reverse(const struct osl_table *t, unsigned col_num,
1826 void *private_data, osl_rbtree_loop_func *func)
1828 struct osl_column *col;
1830 int ret = check_rbtree_col(t, col_num, &col);
1833 return rbtree_loop_reverse(col, private_data, func);
1836 /* TODO: Rollback changes on errors */
1837 static int rename_disk_storage_objects(struct osl_table *t,
1838 struct osl_object *old_obj, struct osl_object *new_obj)
1841 const struct osl_column_description *cd;
1842 char *old_ds_name, *new_ds_name;
1844 if (!t->num_disk_storage_columns)
1845 return 1; /* nothing to do */
1846 if (old_obj->size == new_obj->size && !memcmp(new_obj->data,
1847 old_obj->data, new_obj->size))
1848 return 1; /* object did not change */
1849 old_ds_name = disk_storage_name_of_object(t, old_obj);
1850 new_ds_name = disk_storage_name_of_object(t, new_obj);
1851 ret = -ERRNO_TO_ERROR(ENOMEM);
1852 if (!old_ds_name || ! new_ds_name)
1855 FOR_EACH_DISK_STORAGE_COLUMN(i, t, cd) {
1856 char *old_filename, *new_filename;
1857 ret = create_disk_storage_object_dir(t, i, new_ds_name);
1860 old_filename = disk_storage_path(t, i, old_ds_name);
1861 new_filename = disk_storage_path(t, i, new_ds_name);
1862 if (!old_filename || !new_filename)
1863 ret = -ERRNO_TO_ERROR(ENOMEM);
1865 ret = para_rename(old_filename, new_filename);
1880 * Change an object in an osl table.
1882 * \param t Pointer to an open osl table.
1883 * \param r Pointer to the row containing the object to be updated.
1884 * \param col_num Number of the column containing the object to be updated.
1885 * \param obj Pointer to the replacement object.
1887 * This function gets rid of all references to the old object. This includes
1888 * removal of the rbtree node in case there is an rbtree associated with \a
1889 * col_num. It then inserts \a obj into the table and the rbtree if necessary.
1891 * If the \p OSL_RBTREE flag is set for \a col_num, you \b MUST call this
1892 * function in order to change the contents of an object, even for volatile or
1893 * mapped columns of constant size (which may be updated directly if \p
1894 * OSL_RBTREE is not set). Otherwise the rbtree might become corrupted.
1898 __export int osl_update_object(struct osl_table *t, const struct osl_row *r,
1899 unsigned col_num, struct osl_object *obj)
1901 struct osl_column *col;
1902 const struct osl_column_description *cd;
1906 return -E_OSL_BAD_TABLE;
1907 col = &t->columns[col_num];
1908 cd = get_column_description(t->desc, col_num);
1909 DEBUG_LOG("updating column %u of %s\n", col_num, t->desc->name);
1910 if (cd->storage_flags & OSL_RBTREE) {
1911 if (search_rbtree(obj, t, col_num, NULL, NULL) > 0)
1912 return -E_OSL_RB_KEY_EXISTS;
1914 if (cd->storage_flags & OSL_FIXED_SIZE) {
1915 if (obj->size != cd->data_size)
1916 return -E_OSL_BAD_DATA_SIZE;
1918 remove_rb_node(t, col_num, r);
1919 if (cd->storage_type == OSL_NO_STORAGE) { /* TODO: If fixed size, reuse object? */
1920 free(r->volatile_objects[col->volatile_num].data);
1921 r->volatile_objects[col->volatile_num] = *obj;
1922 } else if (cd->storage_type == OSL_DISK_STORAGE) {
1924 ret = disk_storage_name_of_row(t, r, &ds_name);
1927 ret = delete_disk_storage_file(t, col_num, ds_name);
1928 if (ret < 0 && !is_errno(-ret, ENOENT)) {
1932 ret = write_disk_storage_file(t, col_num, obj, ds_name);
1936 } else { /* mapped storage */
1937 struct osl_object old_obj;
1938 ret = get_mapped_object(t, col_num, r->num, &old_obj);
1942 * If the updated column is the disk storage name column, the
1943 * disk storage name changes, so we have to rename all disk
1944 * storage objects accordingly.
1946 if (col_num == t->disk_storage_name_column) {
1947 ret = rename_disk_storage_objects(t, &old_obj, obj);
1951 if (cd->storage_flags & OSL_FIXED_SIZE)
1952 memcpy(old_obj.data, obj->data, cd->data_size);
1953 else { /* TODO: if the size doesn't change, use old space */
1954 uint32_t new_data_map_size;
1956 ret = get_row_index(t, r->num, &row_index);
1959 ret = mark_mapped_object_invalid(t, r->num, col_num);
1962 unmap_column(t, col_num);
1963 ret = append_map_file(t, col_num, obj,
1964 &new_data_map_size);
1967 ret = map_column(t, col_num);
1970 update_cell_index(row_index, col, new_data_map_size,
1974 if (cd->storage_flags & OSL_RBTREE) {
1975 ret = insert_rbtree(t, col_num, r, obj);
1983 * Retrieve an object of type \p OSL_DISK_STORAGE by row and column.
1985 * \param t Pointer to an open osl table.
1986 * \param r Pointer to the row containing the object.
1987 * \param col_num The column number.
1988 * \param obj Points to the result upon successful return.
1990 * For columns of type \p OSL_DISK_STORAGE, this function must be used to
1991 * retrieve one of its containing objects. Afterwards, osl_close_disk_object()
1992 * must be called in order to deallocate the resources.
1996 * \sa osl_get_object(), osl_storage_type, osl_close_disk_object().
1998 __export int osl_open_disk_object(const struct osl_table *t, const struct osl_row *r,
1999 unsigned col_num, struct osl_object *obj)
2001 const struct osl_column_description *cd;
2002 char *ds_name, *filename;
2006 return -E_OSL_BAD_TABLE;
2007 cd = get_column_description(t->desc, col_num);
2008 if (cd->storage_type != OSL_DISK_STORAGE)
2009 return -E_OSL_BAD_STORAGE_TYPE;
2011 ret = disk_storage_name_of_row(t, r, &ds_name);
2014 filename = disk_storage_path(t, col_num, ds_name);
2017 return -ERRNO_TO_ERROR(ENOMEM);
2018 DEBUG_LOG("filename: %s\n", filename);
2019 ret = mmap_full_file(filename, O_RDONLY, &obj->data, &obj->size, NULL);
2025 * Free resources that were allocated during osl_open_disk_object().
2027 * \param obj Pointer to the object previously returned by open_disk_object().
2029 * \return The return value of the underlying call to para_munmap().
2031 * \sa para_munmap().
2033 __export int osl_close_disk_object(struct osl_object *obj)
2035 return para_munmap(obj->data, obj->size);
2039 * Get the number of rows of the given table.
2041 * \param t Pointer to an open osl table.
2042 * \param num_rows Result is returned here.
2044 * The number of rows returned via \a num_rows excluding any invalid rows.
2046 * \return Positive on success, \p -E_OSL_BAD_TABLE if \a t is \p NULL.
2048 __export int osl_get_num_rows(const struct osl_table *t, unsigned *num_rows)
2051 return -E_OSL_BAD_TABLE;
2052 assert(t->num_rows >= t->num_invalid_rows);
2053 *num_rows = t->num_rows - t->num_invalid_rows;
2058 * Get the rank of a row.
2060 * \param t An open osl table.
2061 * \param r The row to get the rank of.
2062 * \param col_num The number of an rbtree column.
2063 * \param rank Result pointer.
2065 * The rank is, by definition, the position of the row in the linear order
2066 * determined by an in-order tree walk of the rbtree associated with column
2067 * number \a col_num of \a table.
2071 * \sa osl_get_nth_row().
2073 __export int osl_get_rank(const struct osl_table *t, struct osl_row *r,
2074 unsigned col_num, unsigned *rank)
2076 struct osl_object obj;
2077 struct osl_column *col;
2078 struct rb_node *node;
2079 int ret = check_rbtree_col(t, col_num, &col);
2083 ret = osl_get_object(t, r, col_num, &obj);
2086 ret = search_rbtree(&obj, t, col_num, &node, NULL);
2089 ret = rb_rank(node, rank);
2091 return -E_OSL_BAD_ROW;
2096 * Get the row with n-th greatest value.
2098 * \param t Pointer to an open osl table.
2099 * \param col_num The column number.
2100 * \param n The rank of the desired row.
2101 * \param result Row is returned here.
2103 * Retrieve the n-th order statistic with respect to the compare function
2104 * of the rbtree column \a col_num. In other words, get that row with
2105 * \a n th greatest value in column \a col_num. It's an error if
2106 * \a col_num is not a rbtree column, or if \a n is larger than the
2107 * number of rows in the table.
2111 * \sa osl_storage_flags, osl_compare_func, osl_get_row(),
2112 * osl_rbtree_last_row(), osl_rbtree_first_row(), osl_get_rank().
2114 __export int osl_get_nth_row(const struct osl_table *t, unsigned col_num,
2115 unsigned n, struct osl_row **result)
2117 struct osl_column *col;
2118 struct rb_node *node;
2123 return -E_OSL_RB_KEY_NOT_FOUND;
2124 ret = osl_get_num_rows(t, &num_rows);
2128 return -E_OSL_RB_KEY_NOT_FOUND;
2129 ret = check_rbtree_col(t, col_num, &col);
2132 node = rb_nth(col->rbtree.rb_node, n);
2134 return -E_OSL_RB_KEY_NOT_FOUND;
2135 *result = get_row_pointer(node, col->rbtree_num);
2140 * Get the row corresponding to the smallest rbtree node of a column.
2142 * \param t An open rbtree table.
2143 * \param col_num The number of the rbtree column.
2144 * \param result A pointer to the first row is returned here.
2146 * The rbtree node of the smallest object (with respect to the corresponding
2147 * compare function) is selected and the row containing this object is
2148 * returned. It is an error if \a col_num refers to a column without an
2149 * associated rbtree.
2153 * \sa osl_get_nth_row(), osl_rbtree_last_row().
2155 __export int osl_rbtree_first_row(const struct osl_table *t, unsigned col_num,
2156 struct osl_row **result)
2158 return osl_get_nth_row(t, col_num, 1, result);
2162 * Get the row corresponding to the greatest rbtree node of a column.
2164 * \param t The same meaning as in \p osl_rbtree_first_row().
2165 * \param col_num The same meaning as in \p osl_rbtree_first_row().
2166 * \param result The same meaning as in \p osl_rbtree_first_row().
2168 * This function works just like osl_rbtree_first_row(), the only difference
2169 * is that the row containing the greatest rather than the smallest object is
2174 * \sa osl_get_nth_row(), osl_rbtree_first_row().
2176 __export int osl_rbtree_last_row(const struct osl_table *t, unsigned col_num,
2177 struct osl_row **result)
2180 int ret = osl_get_num_rows(t, &num_rows);
2184 return osl_get_nth_row(t, col_num, num_rows, result);