2 * Copyright (C) 2007 Andre Noll <maan@systemlinux.org>
4 * Licensed under the GPL v2. For licencing details see COPYING.
7 /** \file osl.c Object storage layer functions. */
12 #include <dirent.h> /* readdir() */
15 //#define FMT_OFF_T "%li"
19 * A wrapper for lseek(2).
21 * \param fd The filedescriptor whose offset is to be to repositioned.
22 * \param offset A value-result parameter.
23 * \param whence Usual repositioning directive.
25 * Reposition the offset of the file descriptor \a fd to the argument \a offset
26 * according to the directive \a whence. Upon successful return, \a offset
27 * contains the resulting offset location as measured in bytes from the
28 * beginning of the file.
30 * \return Positive on success. Otherwise, the function returns \p -E_LSEEK.
34 int para_lseek(int fd, off_t *offset, int whence)
36 *offset = lseek(fd, *offset, whence);
44 * Waraper for the write system call.
46 * \param fd The file descriptor to write to.
47 * \param buf The buffer to write.
48 * \param size The length of \a buf in bytes.
50 * This function writes out the given bufffer and retries if an interrupt
51 * occured during the write.
53 * \return On success, the number of bytes written is returned, otherwise, the
54 * function returns \p -E_WRITE.
58 ssize_t para_write(int fd, const void *buf, size_t size)
63 ret = write(fd, buf, size);
64 if ((ret < 0) && (errno == EAGAIN || errno == EINTR))
66 return ret >= 0? ret : -E_WRITE;
71 * Write the whole buffer to a file descriptor.
73 * \param fd The file descriptor to write to.
74 * \param buf The buffer to write.
75 * \param size The length of \a buf in bytes.
77 * This function writes the given buffer and continues on short writes and
78 * when interrupted by a signal.
80 * \return Positive on success, negative on errors. Possible errors: any
81 * errors returned by para_write().
85 ssize_t para_write_all(int fd, const void *buf, size_t size)
87 // PARA_DEBUG_LOG("writing %zu bytes\n", size);
90 ssize_t ret = para_write(fd, b, size);
91 // PARA_DEBUG_LOG("ret: %zd\n", ret);
100 * Wrapper for the open(2) system call.
102 * \param path The filename.
103 * \param flags The usual open(2) flags.
104 * \param mode Specifies the permissions to use.
106 * The mode parameter must be specified when O_CREAT is in the flags, and is ignored
109 * \return Positive on success, negative on errors. Possible errors: \p
110 * E_EXIST, \p E_ISDIR, \p E_NOENT, \p E_OSL_PERM.
114 int para_open(const char *path, int flags, mode_t mode)
116 PARA_DEBUG_LOG("opening %s\n", path);
117 int ret = open(path, flags, mode);
135 PARA_ERROR_LOG("failed to open %s: %s\n", path, strerror(errno));
140 * Open a file, write the given buffer and close the file.
142 * \param filename Full path to the file to open.
143 * \param buf The buffer to write to the file.
144 * \param size The size of \a buf.
146 * \return Positive on success, negative on errors. Possible errors include:
147 * any errors from para_open() or para_write().
149 * \sa para_open(), para_write().
151 int para_write_file(const char *filename, const void *buf, size_t size)
155 ret = para_open(filename, O_WRONLY | O_CREAT | O_EXCL, 0644);
159 ret = para_write_all(fd, buf, size);
168 static int append_file(const char *filename, char *header, size_t header_size,
169 char *data, size_t data_size, uint32_t *new_pos)
173 // PARA_DEBUG_LOG("appending %zu + %zu bytes\n", header_size, data_size);
174 ret = para_open(filename, O_WRONLY | O_CREAT | O_APPEND, 0644);
178 if (header && header_size) {
179 ret = para_write_all(fd, header, header_size);
183 ret = para_write_all(fd, data, data_size);
188 ret = para_lseek(fd, &offset, SEEK_END);
191 // PARA_DEBUG_LOG("new file size: " FMT_OFF_T "\n", offset);
201 * Map a file into memory.
203 * \param path Name of the regular file to map.
204 * \param open_mode Either \p O_RDONLY or \p O_RDWR.
205 * \param obj On success, the mapping is returned here.
207 * \return Positive on success, negative on errors. Possible errors include: \p
208 * E_FSTAT, any errors returned by para_open(), \p E_EMPTY, \p E_MMAP.
210 * \sa para_open(), mmap(2).
212 int mmap_full_file(const char *path, int open_mode, struct osl_object *obj)
214 int fd, ret, mmap_prot, mmap_flags;
215 struct stat file_status;
217 if (open_mode == O_RDONLY) {
218 mmap_prot = PROT_READ;
219 mmap_flags = MAP_PRIVATE;
221 mmap_prot = PROT_READ | PROT_WRITE;
222 mmap_flags = MAP_SHARED;
224 ret = para_open(path, open_mode, 0);
229 if (fstat(fd, &file_status) < 0)
231 obj->size = file_status.st_size;
233 PARA_DEBUG_LOG("%s: size %zu\n", path, obj->size);
236 obj->data = mmap(NULL, obj->size, mmap_prot, mmap_flags, fd, 0);
237 if (obj->data == MAP_FAILED) {
249 * Traverse the given directory recursively.
251 * \param dirname The directory to traverse.
252 * \param func The function to call for each entry.
253 * \param private_data Pointer to an arbitrary data structure.
255 * For each regular file in \a dirname, the supplied function \a func is
256 * called. The full path of the regular file and the \a private_data pointer
257 * are passed to \a func.
259 * \return On success, 1 is returned. Otherwise, this function returns a
260 * negative value which indicates the kind of the error.
262 int for_each_file_in_dir(const char *dirname,
263 int (*func)(const char *, const void *), const void *private_data)
266 struct dirent *entry;
268 * Opening the current directory (".") and calling fchdir() to return
269 * is usually faster and more reliable than saving cwd in some buffer
270 * and calling chdir() afterwards (see man 3 getcwd).
272 int cwd_fd = open(".", O_RDONLY);
276 // PARA_DEBUG_LOG("dirname: %s\n", dirname);
278 return -E_OSL_GETCWD;
280 if (chdir(dirname) < 0)
282 ret = -E_OSL_OPENDIR;
286 /* scan cwd recursively */
287 while ((entry = readdir(dir))) {
291 if (!strcmp(entry->d_name, "."))
293 if (!strcmp(entry->d_name, ".."))
296 if (lstat(entry->d_name, &s) == -1)
299 if (!S_ISREG(m) && !S_ISDIR(m))
301 tmp = make_message("%s/%s", dirname, entry->d_name);
303 ret = func(tmp, private_data);
310 ret = for_each_file_in_dir(tmp, func, private_data);
319 if (fchdir(cwd_fd) < 0 && ret >= 0)
325 int para_mkdir(const char *path, mode_t mode)
327 if (!mkdir(path, mode))
333 if (errno == ENOTDIR)
340 static int verify_basename(const char *name)
343 return -E_BAD_BASENAME;
345 return -E_BAD_BASENAME;
346 if (strchr(name, '/'))
347 return -E_BAD_BASENAME;
348 if (!strcmp(name, ".."))
349 return -E_BAD_BASENAME;
350 if (!strcmp(name, "."))
351 return -E_BAD_BASENAME;
356 * Compare two osl objects pointing to unsigned integers of 32 bit size.
358 * \param obj1 Pointer to the first integer.
359 * \param obj2 Pointer to the second integer.
361 * \return The values required for an osl compare function.
363 * \sa osl_compare_func, osl_hash_compare().
365 int uint32_compare(const struct osl_object *obj1, const struct osl_object *obj2)
367 uint32_t d1 = read_u32((const char *)obj1->data);
368 uint32_t d2 = read_u32((const char *)obj2->data);
378 * Compare two osl objects pointing to hash values.
380 * \param obj1 Pointer to the first hash object.
381 * \param obj2 Pointer to the second hash object.
383 * \return The values required for an osl compare function.
385 * \sa osl_compare_func, uint32_compare().
387 int osl_hash_compare(const struct osl_object *obj1, const struct osl_object *obj2)
389 return hash_compare((HASH_TYPE *)obj1->data, (HASH_TYPE *)obj2->data);
392 static char *disk_storage_dirname(const struct osl_table *t, unsigned col_num,
395 char *dirname, *column_name = column_filename(t, col_num);
397 if (!(t->desc->flags & OSL_LARGE_TABLE))
399 dirname = make_message("%s/%.2s", column_name, ds_name);
404 static char *disk_storage_name_of_object(const struct osl_table *t,
405 const struct osl_object *obj)
407 HASH_TYPE hash[HASH_SIZE];
408 hash_object(obj, hash);
409 return disk_storage_name_of_hash(t, hash);
412 static int disk_storage_name_of_row(const struct osl_table *t,
413 const struct osl_row *row, char **name)
415 struct osl_object obj;
416 int ret = osl_get_object(t, row, t->disk_storage_name_column, &obj);
420 *name = disk_storage_name_of_object(t, &obj);
424 static void column_name_hash(const char *col_name, HASH_TYPE *hash)
426 return hash_function(col_name, strlen(col_name), hash);
429 static int init_column_descriptions(struct osl_table *t)
432 const struct osl_column_description *cd;
434 ret = -E_BAD_TABLE_DESC;
435 ret = verify_basename(t->desc->name);
441 /* the size of the index header without column descriptions */
442 t->index_header_size = IDX_COLUMN_DESCRIPTIONS;
443 FOR_EACH_COLUMN(i, t->desc, cd) {
444 struct osl_column *col = t->columns + i;
445 if (cd->storage_flags & OSL_RBTREE) {
446 if (!cd->compare_function)
447 return -E_NO_COMPARE_FUNC;
449 if (cd->storage_type == OSL_NO_STORAGE)
451 ret = -E_NO_COLUMN_NAME;
452 if (!cd->name || !cd->name[0])
454 ret = verify_basename(cd->name);
457 t->index_header_size += index_column_description_size(cd->name);
458 column_name_hash(cd->name, col->name_hash);
459 ret = -E_DUPLICATE_COL_NAME;
460 for (j = i + 1; j < t->desc->num_columns; j++) {
461 const char *name2 = get_column_description(t->desc,
463 if (cd->name && name2 && !strcmp(cd->name, name2))
473 * Initialize a struct table from given table description.
475 * \param desc The description of the osl table.
476 * \param table_ptr Result is returned here.
478 * This function performs several sanity checks on \p desc and returns if any
479 * of these tests fail. On success, a struct \p osl_table is allocated and
480 * initialized with data derived from \p desc.
482 * \return Positive on success, negative on errors. Possible errors include: \p
483 * E_BAD_TABLE_DESC, \p E_NO_COLUMN_DESC, \p E_NO_COLUMNS, \p
484 * E_BAD_STORAGE_TYPE, \p E_BAD_STORAGE_FLAGS, \p E_BAD_STORAGE_SIZE, \p
485 * E_NO_UNIQUE_RBTREE_COLUMN, \p E_NO_RBTREE_COL.
487 * \sa struct osl_table.
489 int init_table_structure(const struct osl_table_description *desc,
490 struct osl_table **table_ptr)
492 const struct osl_column_description *cd;
493 struct osl_table *t = para_calloc(sizeof(*t));
494 int i, ret = -E_BAD_TABLE_DESC, have_disk_storage_name_column = 0;
498 PARA_DEBUG_LOG("creating table structure for '%s' from table "
499 "description\n", desc->name);
500 ret = -E_NO_COLUMN_DESC;
501 if (!desc->column_descriptions)
504 if (!desc->num_columns)
506 t->columns = para_calloc(desc->num_columns * sizeof(struct osl_column));
508 FOR_EACH_COLUMN(i, t->desc, cd) {
509 enum osl_storage_type st = cd->storage_type;
510 enum osl_storage_flags sf = cd->storage_flags;
511 struct osl_column *col = &t->columns[i];
513 ret = -E_BAD_STORAGE_TYPE;
514 if (st != OSL_MAPPED_STORAGE && st != OSL_DISK_STORAGE
515 && st != OSL_NO_STORAGE)
517 ret = -E_BAD_STORAGE_FLAGS;
518 if (st == OSL_DISK_STORAGE && sf & OSL_RBTREE)
520 ret = -E_BAD_STORAGE_SIZE;
521 if (sf & OSL_FIXED_SIZE && !cd->data_size)
524 case OSL_DISK_STORAGE:
525 t->num_disk_storage_columns++;
527 case OSL_MAPPED_STORAGE:
528 t->num_mapped_columns++;
529 col->index_offset = t->row_index_size;
530 t->row_index_size += 8;
533 col->volatile_num = t->num_volatile_columns;
534 t->num_volatile_columns++;
537 if (sf & OSL_RBTREE) {
538 col->rbtree_num = t->num_rbtrees;
540 if ((sf & OSL_UNIQUE) && (st == OSL_MAPPED_STORAGE)) {
541 if (!have_disk_storage_name_column)
542 t->disk_storage_name_column = i;
543 have_disk_storage_name_column = 1;
547 ret = -E_NO_UNIQUE_RBTREE_COLUMN;
548 if (t->num_disk_storage_columns && !have_disk_storage_name_column)
550 ret = -E_NO_RBTREE_COL;
554 PARA_DEBUG_LOG("OK. Index entry size: %u\n", t->row_index_size);
555 ret = init_column_descriptions(t);
567 * Read the table description from index header.
569 * \param map The memory mapping of the index file.
570 * \param desc The values found in the index header are returned here.
572 * Read the index header, check for the paraslash magic string and the table version number.
573 * Read all information stored in the index header into \a desc.
575 * \return Positive on success, negative on errors.
577 * \sa struct osl_table_description, osl_create_table.
579 int read_table_desc(struct osl_object *map, struct osl_table_description *desc)
581 char *buf = map->data;
583 uint16_t header_size;
586 struct osl_column_description *cd;
588 if (map->size < MIN_INDEX_HEADER_SIZE(1))
589 return -E_SHORT_TABLE;
590 if (strncmp(buf + IDX_PARA_MAGIC, PARA_MAGIC, strlen(PARA_MAGIC)))
592 version = read_u8(buf + IDX_VERSION);
593 if (version < MIN_TABLE_VERSION || version > MAX_TABLE_VERSION)
594 return -E_VERSION_MISMATCH;
595 desc->num_columns = read_u8(buf + IDX_TABLE_FLAGS);
596 desc->flags = read_u8(buf + IDX_TABLE_FLAGS);
597 desc->num_columns = read_u16(buf + IDX_NUM_COLUMNS);
598 PARA_DEBUG_LOG("%u columns\n", desc->num_columns);
599 if (!desc->num_columns)
600 return -E_NO_COLUMNS;
601 header_size = read_u16(buf + IDX_HEADER_SIZE);
602 if (map->size < header_size)
604 desc->column_descriptions = para_calloc(desc->num_columns
605 * sizeof(struct osl_column_description));
606 offset = IDX_COLUMN_DESCRIPTIONS;
607 FOR_EACH_COLUMN(i, desc, cd) {
610 ret = -E_SHORT_TABLE;
611 if (map->size < offset + MIN_IDX_COLUMN_DESCRIPTION_SIZE) {
612 PARA_ERROR_LOG("map size = %zu < %u = offset + min desc size\n",
613 map->size, offset + MIN_IDX_COLUMN_DESCRIPTION_SIZE);
616 cd->storage_type = read_u16(buf + offset + IDX_CD_STORAGE_TYPE);
617 cd->storage_flags = read_u16(buf + offset +
618 IDX_CD_STORAGE_FLAGS);
619 cd->data_size = read_u32(buf + offset + IDX_CD_DATA_SIZE);
620 null_byte = memchr(buf + offset + IDX_CD_NAME, '\0',
621 map->size - offset - IDX_CD_NAME);
622 ret = -E_INDEX_CORRUPTION;
625 cd->name = para_strdup(buf + offset + IDX_CD_NAME);
626 offset += index_column_description_size(cd->name);
628 if (offset != header_size) {
629 ret = -E_INDEX_CORRUPTION;
630 PARA_ERROR_LOG("real header size = %u != %u = stored header size\n",
631 offset, header_size);
636 FOR_EACH_COLUMN(i, desc, cd)
642 * check whether the table description given by \p t->desc matches the on-disk
643 * table structure stored in the index of \a t.
645 static int compare_table_descriptions(struct osl_table *t)
648 struct osl_table_description desc;
649 const struct osl_column_description *cd1, *cd2;
651 /* read the on-disk structure into desc */
652 ret = read_table_desc(&t->index_map, &desc);
655 ret = -E_BAD_TABLE_FLAGS;
656 if (desc.flags != t->desc->flags)
658 ret = E_BAD_COLUMN_NUM;
659 if (desc.num_columns != t->desc->num_columns)
661 FOR_EACH_COLUMN(i, t->desc, cd1) {
662 cd2 = get_column_description(&desc, i);
663 ret = -E_BAD_STORAGE_TYPE;
664 if (cd1->storage_type != cd2->storage_type)
666 ret = -E_BAD_STORAGE_FLAGS;
667 if (cd1->storage_flags != cd2->storage_flags) {
668 PARA_ERROR_LOG("sf1 = %u != %u = sf2\n",
669 cd1->storage_flags, cd2->storage_flags);
672 ret = -E_BAD_DATA_SIZE;
673 if (cd1->storage_flags & OSL_FIXED_SIZE)
674 if (cd1->data_size != cd2->data_size)
676 ret = -E_BAD_COLUMN_NAME;
677 if (strcmp(cd1->name, cd2->name))
680 PARA_DEBUG_LOG("table description of '%s' matches on-disk data, good\n",
684 FOR_EACH_COLUMN(i, &desc, cd1)
686 free(desc.column_descriptions);
690 static int create_table_index(struct osl_table *t)
692 char *buf, *filename;
694 size_t size = t->index_header_size;
695 const struct osl_column_description *cd;
698 PARA_INFO_LOG("creating %zu byte index for table %s\n", size,
700 buf = para_calloc(size);
701 sprintf(buf + IDX_PARA_MAGIC, "%s", PARA_MAGIC);
702 write_u8(buf + IDX_TABLE_FLAGS, t->desc->flags);
703 write_u8(buf + IDX_DIRTY_FLAG, 0);
704 write_u8(buf + IDX_VERSION, CURRENT_TABLE_VERSION);
705 write_u16(buf + IDX_NUM_COLUMNS, t->desc->num_columns);
706 write_u16(buf + IDX_HEADER_SIZE, t->index_header_size);
707 offset = IDX_COLUMN_DESCRIPTIONS;
708 FOR_EACH_COLUMN(i, t->desc, cd) {
709 write_u16(buf + offset + IDX_CD_STORAGE_TYPE,
711 write_u16(buf + offset + IDX_CD_STORAGE_FLAGS,
713 if (cd->storage_flags & OSL_FIXED_SIZE)
714 write_u32(buf + offset + IDX_CD_DATA_SIZE,
716 strcpy(buf + offset + IDX_CD_NAME, cd->name);
717 offset += index_column_description_size(cd->name);
719 assert(offset = size);
720 filename = index_filename(t->desc);
721 ret = para_write_file(filename, buf, size);
728 * Create a new osl table.
730 * \param desc Pointer to the table description.
732 * \return Positive on success, negative on errors. Possible errors include: \p
733 * E_BAD_TABLE_DESC, \p E_BAD_DB_DIR, \p E_BAD_BASENAME, \p E_NO_COMPARE_FUNC, \p
734 * E_NO_COLUMN_NAME, \p E_DUPLICATE_COL_NAME, \p E_MKDIR, any errors returned
737 int osl_create_table(const struct osl_table_description *desc)
739 const struct osl_column_description *cd;
740 char *table_dir = NULL, *filename;
742 int i, ret = init_table_structure(desc, &t);
746 PARA_INFO_LOG("creating %s\n", desc->name);
747 FOR_EACH_COLUMN(i, t->desc, cd) {
748 if (cd->storage_type == OSL_NO_STORAGE)
751 ret = para_mkdir(desc->dir, 0777);
752 if (ret < 0 && ret != -E_EXIST)
754 table_dir = make_message("%s/%s", desc->dir,
756 ret = para_mkdir(table_dir, 0777);
760 filename = column_filename(t, i);
761 PARA_INFO_LOG("filename: %s\n", filename);
762 if (cd->storage_type == OSL_MAPPED_STORAGE) {
763 ret = para_open(filename, O_RDWR | O_CREAT | O_EXCL,
772 ret = para_mkdir(filename, 0777);
777 if (t->num_mapped_columns) {
778 ret = create_table_index(t);
790 static int table_is_dirty(struct osl_table *t)
792 char *buf = (char *)t->index_map.data + IDX_DIRTY_FLAG;
793 uint8_t dirty = read_u8(buf) & 0x1;
797 static void mark_table_dirty(struct osl_table *t)
799 char *buf = (char *)t->index_map.data + IDX_DIRTY_FLAG;
800 write_u8(buf, read_u8(buf) | 1);
803 static void mark_table_clean(struct osl_table *t)
805 char *buf = (char *)t->index_map.data + IDX_DIRTY_FLAG;
806 write_u8(buf, read_u8(buf) & 0xfe);
809 static void unmap_column(struct osl_table *t, unsigned col_num)
811 struct osl_object map = t->columns[col_num].data_map;
815 ret = para_munmap(map.data, map.size);
821 * Unmap all mapped files of an osl table.
823 * \param t Pointer to a mapped table.
824 * \param flags Options for unmapping.
826 * \return Positive on success, negative on errors. Possible errors include:
827 * E_NOT_MAPPED, E_MUNMAP.
829 * \sa map_table(), enum osl_close_flags, para_munmap().
831 int unmap_table(struct osl_table *t, enum osl_close_flags flags)
834 const struct osl_column_description *cd;
837 if (!t->num_mapped_columns) /* can this ever happen? */
839 PARA_DEBUG_LOG("unmapping table '%s'\n", t->desc->name);
840 if (!t->index_map.data)
841 return -E_NOT_MAPPED;
842 if (flags & OSL_MARK_CLEAN)
844 ret = para_munmap(t->index_map.data, t->index_map.size);
847 t->index_map.data = NULL;
850 FOR_EACH_MAPPED_COLUMN(i, t, cd)
855 static int map_column(struct osl_table *t, unsigned col_num)
858 char *filename = column_filename(t, col_num);
860 if (stat(filename, &statbuf) < 0) {
864 if (!(S_IFREG & statbuf.st_mode)) {
868 ret = mmap_full_file(filename, O_RDWR,
869 &t->columns[col_num].data_map);
875 * Map the index file and all columns of type \p OSL_MAPPED_STORAGE into memory.
877 * \param t Pointer to an initialized table structure.
878 * \param flags Mapping options.
880 * \return Negative return value on errors; on success the number of rows
881 * (including invalid rows) is returned.
883 * \sa unmap_table(), enum map_table_flags, osl_open_table(), mmap(2).
885 int map_table(struct osl_table *t, enum map_table_flags flags)
888 const struct osl_column_description *cd;
889 int i = 0, ret, num_rows = 0;
891 if (!t->num_mapped_columns)
893 if (t->index_map.data)
894 return -E_ALREADY_MAPPED;
895 filename = index_filename(t->desc);
896 PARA_DEBUG_LOG("mapping table '%s' (index: %s)\n", t->desc->name, filename);
897 ret = mmap_full_file(filename, flags & MAP_TBL_FL_MAP_RDONLY?
898 O_RDONLY : O_RDWR, &t->index_map);
902 if (flags & MAP_TBL_FL_VERIFY_INDEX) {
903 ret = compare_table_descriptions(t);
908 if (!(flags & MAP_TBL_FL_IGNORE_DIRTY)) {
909 if (table_is_dirty(t)) {
910 PARA_ERROR_LOG("%s is dirty\n", t->desc->name);
915 num_rows = table_num_rows(t);
919 FOR_EACH_MAPPED_COLUMN(i, t, cd) {
920 ret = map_column(t, i);
925 err: /* unmap what is already mapped */
926 for (i--; i >= 0; i--) {
927 struct osl_object map = t->columns[i].data_map;
928 para_munmap(map.data, map.size);
931 para_munmap(t->index_map.data, t->index_map.size);
932 t->index_map.data = NULL;
937 * Retrieve a mapped object by row and column number.
939 * \param t Pointer to an open osl table.
940 * \param col_num Number of the mapped column containing the object to retrieve.
941 * \param row_num Number of the row containing the object to retrieve.
942 * \param obj The result is returned here.
944 * It is considered an error if \a col_num does not refer to a column
945 * of storage type \p OSL_MAPPED_STORAGE.
947 * \return Positive on success, negative on errors. Possible errors include:
948 * \p E_BAD_ID, \p E_INVALID_OBJECT.
950 * \sa osl_storage_type.
952 int get_mapped_object(const struct osl_table *t, unsigned col_num,
953 uint32_t row_num, struct osl_object *obj)
955 struct osl_column *col = &t->columns[col_num];
961 if (t->num_rows <= row_num)
963 ret = get_cell_index(t, row_num, col_num, &cell_index);
966 offset = read_u32(cell_index);
967 obj->size = read_u32(cell_index + 4) - 1;
968 header = col->data_map.data + offset;
969 obj->data = header + 1;
970 if (read_u8(header) == 0xff) {
971 PARA_ERROR_LOG("col %u, size %zu, offset %u\n", col_num,
973 return -E_INVALID_OBJECT;
978 static int search_rbtree(const struct osl_object *obj,
979 const struct osl_table *t, unsigned col_num,
980 struct rb_node **result, struct rb_node ***rb_link)
982 struct osl_column *col = &t->columns[col_num];
983 struct rb_node **new = &col->rbtree.rb_node, *parent = NULL;
984 const struct osl_column_description *cd =
985 get_column_description(t->desc, col_num);
986 enum osl_storage_type st = cd->storage_type;
988 struct osl_row *this_row = get_row_pointer(*new,
991 struct osl_object this_obj;
993 if (st == OSL_MAPPED_STORAGE) {
994 ret = get_mapped_object(t, col_num, this_row->id,
999 this_obj = this_row->volatile_objects[col->volatile_num];
1000 ret = cd->compare_function(obj, &this_obj);
1003 *result = get_rb_node_pointer(this_row,
1008 new = &((*new)->rb_left);
1010 new = &((*new)->rb_right);
1016 return -E_RB_KEY_NOT_FOUND;
1019 static int insert_rbtree(struct osl_table *t, unsigned col_num,
1020 const struct osl_row *row, const struct osl_object *obj)
1022 struct rb_node *parent, **rb_link;
1023 unsigned rbtree_num;
1025 int ret = search_rbtree(obj, t, col_num, &parent, &rb_link);
1028 return -E_RB_KEY_EXISTS;
1029 rbtree_num = t->columns[col_num].rbtree_num;
1030 n = get_rb_node_pointer(row, rbtree_num);
1031 rb_link_node(n, parent, rb_link);
1032 rb_insert_color(n, &t->columns[col_num].rbtree);
1036 static void remove_rb_node(struct osl_table *t, unsigned col_num,
1037 const struct osl_row *row)
1039 struct osl_column *col = &t->columns[col_num];
1040 const struct osl_column_description *cd =
1041 get_column_description(t->desc, col_num);
1042 enum osl_storage_flags sf = cd->storage_flags;
1043 struct rb_node *victim, *splice_out_node, *tmp;
1044 if (!(sf & OSL_RBTREE))
1047 * Which node is removed/spliced out actually depends on how many
1048 * children the victim node has: If it has no children, it gets
1049 * deleted. If it has one child, it gets spliced out. If it has two
1050 * children, its successor (which has at most a right child) gets
1053 victim = get_rb_node_pointer(row, col->rbtree_num);
1054 if (victim->rb_left && victim->rb_right)
1055 splice_out_node = rb_next(victim);
1057 splice_out_node = victim;
1058 /* Go up to the root and decrement the size of each node in the path. */
1059 for (tmp = splice_out_node; tmp; tmp = rb_parent(tmp))
1061 rb_erase(victim, &col->rbtree);
1064 static int add_row_to_rbtrees(struct osl_table *t, uint32_t id,
1065 struct osl_object *volatile_objs, struct osl_row **row_ptr)
1069 struct osl_row *row = allocate_row(t->num_rbtrees);
1070 const struct osl_column_description *cd;
1073 row->volatile_objects = volatile_objs;
1074 FOR_EACH_RBTREE_COLUMN(i, t, cd) {
1075 if (cd->storage_type == OSL_MAPPED_STORAGE) {
1076 struct osl_object obj;
1077 ret = get_mapped_object(t, i, id, &obj);
1080 ret = insert_rbtree(t, i, row, &obj);
1081 } else { /* volatile */
1082 const struct osl_object *obj
1083 = volatile_objs + t->columns[i].volatile_num;
1084 ret = insert_rbtree(t, i, row, obj);
1092 err: /* rollback changes, i.e. remove added entries from rbtrees */
1094 remove_rb_node(t, i--, row);
1099 static void free_volatile_objects(const struct osl_table *t,
1100 enum osl_close_flags flags)
1104 struct osl_column *rb_col;
1105 const struct osl_column_description *cd;
1107 if (!t->num_volatile_columns)
1109 /* find the first rbtree column (any will do) */
1110 FOR_EACH_RBTREE_COLUMN(i, t, cd)
1112 rb_col = t->columns + i;
1113 /* walk that rbtree and free all volatile objects */
1114 for (n = rb_first(&rb_col->rbtree); n; n = rb_next(n)) {
1115 struct osl_row *r = get_row_pointer(n, rb_col->rbtree_num);
1116 if (flags & OSL_FREE_VOLATILE)
1117 for (j = 0; j < t->num_volatile_columns; j++)
1118 free(r->volatile_objects[j].data);
1119 free(r->volatile_objects);
1124 * Erase all rbtree nodes and free resources.
1126 * \param t Pointer to an open osl table.
1128 * This function is called by osl_close_table().
1130 void clear_rbtrees(struct osl_table *t)
1132 const struct osl_column_description *cd;
1133 unsigned i, rbtrees_cleared = 0;
1135 FOR_EACH_RBTREE_COLUMN(i, t, cd) {
1136 struct osl_column *col = &t->columns[i];
1139 for (n = rb_first(&col->rbtree); n;) {
1141 rb_erase(n, &col->rbtree);
1142 if (rbtrees_cleared == t->num_rbtrees) {
1143 r = get_row_pointer(n, col->rbtree_num);
1154 * Close an osl table.
1156 * \param t Pointer to the table to be closed.
1157 * \param flags Options for what should be cleaned up.
1159 * If osl_open_table() succeeds, the resulting table pointer must later be
1160 * passed to this function in order to flush all changes to the filesystem and
1161 * to free the resources that were allocated by osl_open_table().
1163 * \return Positive on success, negative on errors. Possible errors: \p E_BAD_TABLE,
1164 * errors returned by unmap_table().
1166 * \sa osl_open_table(), unmap_table().
1168 int osl_close_table(struct osl_table *t, enum osl_close_flags flags)
1173 return -E_BAD_TABLE;
1174 free_volatile_objects(t, flags);
1176 ret = unmap_table(t, flags);
1178 PARA_ERROR_LOG("unmap_table failed: %d\n", ret);
1185 * Find out whether the given row number corresponds to an invalid row.
1187 * \param t Pointer to the osl table.
1188 * \param row_num The number of the row in question.
1190 * By definition, a row is considered invalid if all its index entries
1193 * \return Positive if \a row_num corresponds to an invalid row,
1194 * zero if it corresponds to a valid row, negative on errors.
1196 int row_is_invalid(struct osl_table *t, uint32_t row_num)
1199 int i, ret = get_row_index(t, row_num, &row_index);
1203 for (i = 0; i < t->row_index_size; i++) {
1204 if ((unsigned char)row_index[i] != 0xff)
1207 PARA_INFO_LOG("row %d is invalid\n", row_num);
1212 * Invalidate a row of an osl table.
1214 * \param t Pointer to an open osl table.
1215 * \param row_num Number of the row to mark as invalid.
1217 * This function marks each mapped object in the index entry of \a row as
1220 * \return Positive on success, negative on errors.
1222 int mark_row_invalid(struct osl_table *t, uint32_t row_num)
1225 int ret = get_row_index(t, row_num, &row_index);
1229 PARA_INFO_LOG("marking row %d as invalid\n", row_num);
1230 memset(row_index, 0xff, t->row_index_size);
1235 * Initialize all rbtrees and compute number of invalid rows.
1237 * \param t The table containing the rbtrees to be initialized.
1239 * \return Positive on success, negative on errors.
1241 int init_rbtrees(struct osl_table *t)
1244 const struct osl_column_description *cd;
1246 /* create rbtrees */
1247 FOR_EACH_RBTREE_COLUMN(i, t, cd)
1248 t->columns[i].rbtree = RB_ROOT;
1249 /* add valid rows to rbtrees */
1250 t->num_invalid_rows = 0;
1251 for (i = 0; i < t->num_rows; i++) {
1252 ret = row_is_invalid(t, i);
1256 t->num_invalid_rows++;
1259 ret = add_row_to_rbtrees(t, i, NULL, NULL);
1267 * Open an osl table.
1269 * Each osl table must be opened before its data can be accessed.
1271 * \param table_desc Describes the table to be opened.
1272 * \param result Contains a pointer to the open table on success.
1274 * The table description given by \a desc should coincide with the
1275 * description used at creation time.
1277 * \return Positive on success, negative on errors. Possible errors include:
1278 * errors returned by init_table_structure(), \p E_NOENT, \p E_STAT, \p \p
1279 * E_NOTDIR, \p E_BAD_TABLE_DESC, \p E_BAD_DB_DIR, \p E_NO_COMPARE_FUNC, \p
1280 * E_NO_COLUMN_NAME, errors returned by init_rbtrees().
1282 int osl_open_table(const struct osl_table_description *table_desc,
1283 struct osl_table **result)
1286 struct osl_table *t;
1287 const struct osl_column_description *cd;
1289 PARA_INFO_LOG("opening table %s\n", table_desc->name);
1290 ret = init_table_structure(table_desc, &t);
1293 FOR_EACH_DISK_STORAGE_COLUMN(i, t, cd) {
1294 /* check if directory exists */
1295 char *dirname = column_filename(t, i);
1296 struct stat statbuf;
1297 ret = stat(dirname, &statbuf);
1300 if (errno == ENOENT)
1307 if (!S_ISDIR(statbuf.st_mode))
1310 ret = map_table(t, MAP_TBL_FL_VERIFY_INDEX);
1314 PARA_DEBUG_LOG("num rows: %d\n", t->num_rows);
1315 ret = init_rbtrees(t);
1317 osl_close_table(t, OSL_MARK_CLEAN); /* ignore further errors */
1328 static int create_disk_storage_object_dir(const struct osl_table *t,
1329 unsigned col_num, const char *ds_name)
1334 if (!(t->desc->flags & OSL_LARGE_TABLE))
1336 dirname = disk_storage_dirname(t, col_num, ds_name);
1337 ret = para_mkdir(dirname, 0777);
1339 if (ret < 0 && ret != -E_EXIST)
1344 static int write_disk_storage_file(const struct osl_table *t, unsigned col_num,
1345 const struct osl_object *obj, const char *ds_name)
1350 ret = create_disk_storage_object_dir(t, col_num, ds_name);
1353 filename = disk_storage_path(t, col_num, ds_name);
1354 ret = para_write_file(filename, obj->data, obj->size);
1359 static int append_map_file(const struct osl_table *t, unsigned col_num,
1360 const struct osl_object *obj, uint32_t *new_size)
1362 char *filename = column_filename(t, col_num);
1364 char header = 0; /* zero means valid object */
1366 // PARA_DEBUG_LOG("appending %zu + 1 byte\n", obj->size);
1367 ret = append_file(filename, &header, 1, obj->data, obj->size,
1373 static int append_row_index(const struct osl_table *t, char *row_index)
1378 if (!t->num_mapped_columns)
1380 filename = index_filename(t->desc);
1381 ret = append_file(filename, NULL, 0, row_index,
1382 t->row_index_size, NULL);
1388 * A wrapper for truncate(2)
1390 * \param path Name of the regular file to truncate
1391 * \param size Number of bytes to \b shave \b off
1393 * Truncate the regular file named by \a path by \a size bytes.
1395 * \return Positive on success, negative on errors. Possible errors include: \p
1396 * E_STAT, \p E_BAD_SIZE, \p E_TRUNC.
1400 int para_truncate(const char *path, off_t size)
1403 struct stat statbuf;
1406 if (stat(path, &statbuf) < 0)
1409 if (statbuf.st_size < size)
1412 if (truncate(path, statbuf.st_size - size) < 0)
1419 static int truncate_mapped_file(const struct osl_table *t, unsigned col_num,
1422 char *filename = column_filename(t, col_num);
1423 int ret = para_truncate(filename, size);
1428 static int delete_disk_storage_file(const struct osl_table *t, unsigned col_num,
1429 const char *ds_name)
1431 char *dirname, *filename = disk_storage_path(t, col_num, ds_name);
1432 int ret = unlink(filename);
1434 PARA_DEBUG_LOG("deleted %s\n", filename);
1437 if (errno == ENOENT)
1441 if (!(t->desc->flags & OSL_LARGE_TABLE))
1443 dirname = disk_storage_dirname(t, col_num, ds_name);
1450 * Add a new row to an osl table and retrieve this row.
1452 * \param t Pointer to an open osl table.
1453 * \param objects Array of objects to be added.
1454 * \param row Result pointer.
1456 * The \a objects parameter must point to an array containing one object per
1457 * column. The order of the objects in the array is given by the table
1458 * description of \a table. Several sanity checks are performed during object
1459 * insertion and the function returns without modifying the table if any of
1460 * these tests fail. In fact, it is atomic in the sense that it either
1461 * succeeds or leaves the table unchanged (i.e. either all or none of the
1462 * objects are added to the table).
1464 * It is considered an error if an object is added to a column with associated
1465 * rbtree if this object is equal to an object already contained in that column
1466 * (i.e. the compare function for the column's rbtree returns zero).
1468 * Possible errors include: \p E_RB_KEY_EXISTS, \p E_BAD_DATA_SIZE.
1470 * \return Positive on success, negative on errors.
1472 * \sa struct osl_table_description, osl_compare_func, osl_add_row().
1474 int osl_add_and_get_row(struct osl_table *t, struct osl_object *objects,
1475 struct osl_row **row)
1478 char *ds_name = NULL;
1479 struct rb_node **rb_parents = NULL, ***rb_links = NULL;
1480 char *new_row_index = NULL;
1481 struct osl_object *volatile_objs = NULL;
1482 const struct osl_column_description *cd;
1485 return -E_BAD_TABLE;
1486 rb_parents = para_malloc(t->num_rbtrees * sizeof(struct rn_node*));
1487 rb_links = para_malloc(t->num_rbtrees * sizeof(struct rn_node**));
1488 if (t->num_mapped_columns)
1489 new_row_index = para_malloc(t->row_index_size);
1490 /* pass 1: sanity checks */
1491 // PARA_DEBUG_LOG("sanity tests: %p:%p\n", objects[0].data,
1492 // objects[1].data);
1493 FOR_EACH_COLUMN(i, t->desc, cd) {
1494 enum osl_storage_type st = cd->storage_type;
1495 enum osl_storage_flags sf = cd->storage_flags;
1497 // ret = -E_NULL_OBJECT;
1500 if (st == OSL_DISK_STORAGE)
1502 if (sf & OSL_RBTREE) {
1503 unsigned rbtree_num = t->columns[i].rbtree_num;
1504 ret = -E_RB_KEY_EXISTS;
1505 // PARA_DEBUG_LOG("checking whether %p exists\n",
1506 // objects[i].data);
1507 if (search_rbtree(objects + i, t, i,
1508 &rb_parents[rbtree_num],
1509 &rb_links[rbtree_num]) > 0)
1512 if (sf & OSL_FIXED_SIZE) {
1513 // PARA_DEBUG_LOG("fixed size. need: %zu, have: %d\n",
1514 // objects[i].size, cd->data_size);
1515 ret = -E_BAD_DATA_SIZE;
1516 if (objects[i].size != cd->data_size)
1520 if (t->num_disk_storage_columns)
1521 ds_name = disk_storage_name_of_object(t,
1522 &objects[t->disk_storage_name_column]);
1523 ret = unmap_table(t, OSL_MARK_CLEAN);
1526 // PARA_DEBUG_LOG("sanity tests passed%s\n", "");
1527 /* pass 2: create data files, append map data */
1528 FOR_EACH_COLUMN(i, t->desc, cd) {
1529 enum osl_storage_type st = cd->storage_type;
1530 if (st == OSL_NO_STORAGE)
1532 if (st == OSL_MAPPED_STORAGE) {
1534 struct osl_column *col = &t->columns[i];
1535 // PARA_DEBUG_LOG("appending object of size %zu\n",
1536 // objects[i].size);
1537 ret = append_map_file(t, i, objects + i, &new_size);
1540 update_cell_index(new_row_index, col, new_size,
1545 ret = write_disk_storage_file(t, i, objects + i, ds_name);
1549 ret = append_row_index(t, new_row_index);
1552 ret = map_table(t, MAP_TBL_FL_VERIFY_INDEX);
1553 if (ret < 0) { /* truncate index and rollback changes */
1554 char *filename = index_filename(t->desc);
1555 para_truncate(filename, t->row_index_size);
1559 /* pass 3: add entry to rbtrees */
1560 if (t->num_volatile_columns) {
1561 volatile_objs = para_calloc(t->num_volatile_columns
1562 * sizeof(struct osl_object));
1563 FOR_EACH_VOLATILE_COLUMN(i, t, cd)
1564 volatile_objs[t->columns[i].volatile_num] = objects[i];
1567 // PARA_DEBUG_LOG("adding new entry as row #%d\n", t->num_rows - 1);
1568 ret = add_row_to_rbtrees(t, t->num_rows - 1, volatile_objs, row);
1571 // PARA_DEBUG_LOG("added new entry as row #%d\n", t->num_rows - 1);
1574 rollback: /* rollback all changes made, ignore further errors */
1575 for (i--; i >= 0; i--) {
1576 cd = get_column_description(t->desc, i);
1577 enum osl_storage_type st = cd->storage_type;
1578 if (st == OSL_NO_STORAGE)
1581 if (st == OSL_MAPPED_STORAGE)
1582 truncate_mapped_file(t, i, objects[i].size);
1583 else /* disk storage */
1584 delete_disk_storage_file(t, i, ds_name);
1586 /* ignore error and return previous error value */
1587 map_table(t, MAP_TBL_FL_VERIFY_INDEX);
1589 free(new_row_index);
1597 * Add a new row to an osl table.
1599 * \param t Same meaning as osl_add_and_get_row().
1600 * \param objects Same meaning as osl_add_and_get_row().
1602 * \return The return value of the underlying call to osl_add_and_get_row().
1604 * This is equivalent to osl_add_and_get_row(t, objects, NULL).
1606 int osl_add_row(struct osl_table *t, struct osl_object *objects)
1608 return osl_add_and_get_row(t, objects, NULL);
1612 * Retrieve an object identified by row and column
1614 * \param t Pointer to an open osl table.
1615 * \param r Pointer to the row.
1616 * \param col_num The column number.
1617 * \param object The result pointer.
1619 * The column determined by \a col_num must be of type \p OSL_MAPPED_STORAGE
1620 * or \p OSL_NO_STORAGE, i.e. no disk storage objects may be retrieved by this
1623 * \return Positive if object was found, negative on errors. Possible errors
1624 * include: \p E_BAD_TABLE, \p E_BAD_STORAGE_TYPE.
1626 * \sa osl_storage_type, osl_open_disk_object().
1628 int osl_get_object(const struct osl_table *t, const struct osl_row *r,
1629 unsigned col_num, struct osl_object *object)
1631 const struct osl_column_description *cd;
1634 return -E_BAD_TABLE;
1635 cd = get_column_description(t->desc, col_num);
1636 /* col must not be disk storage */
1637 if (cd->storage_type == OSL_DISK_STORAGE)
1638 return -E_BAD_STORAGE_TYPE;
1639 if (cd->storage_type == OSL_MAPPED_STORAGE)
1640 return get_mapped_object(t, col_num, r->id, object);
1642 *object = r->volatile_objects[t->columns[col_num].volatile_num];
1646 static int mark_mapped_object_invalid(const struct osl_table *t, uint32_t id,
1649 struct osl_object obj;
1651 int ret = get_mapped_object(t, col_num, id, &obj);
1662 * Delete a row from an osl table.
1664 * \param t Pointer to an open osl table.
1665 * \param row Pointer to the row to delete.
1667 * This removes all disk storage objects, removes all rbtree nodes, and frees
1668 * all volatile objects belonging to the given row. For mapped columns, the
1669 * data is merely marked invalid and may be pruned from time to time by
1672 * \return Positive on success, negative on errors. Possible errors include:
1673 * \p E_BAD_TABLE, errors returned by osl_get_object().
1675 int osl_del_row(struct osl_table *t, struct osl_row *row)
1677 struct osl_row *r = row;
1679 const struct osl_column_description *cd;
1682 return -E_BAD_TABLE;
1683 PARA_INFO_LOG("deleting row %p\n", row);
1685 if (t->num_disk_storage_columns) {
1687 ret = disk_storage_name_of_row(t, r, &ds_name);
1690 FOR_EACH_DISK_STORAGE_COLUMN(i, t, cd)
1691 delete_disk_storage_file(t, i, ds_name);
1694 FOR_EACH_COLUMN(i, t->desc, cd) {
1695 struct osl_column *col = t->columns + i;
1696 enum osl_storage_type st = cd->storage_type;
1697 remove_rb_node(t, i, r);
1698 if (st == OSL_MAPPED_STORAGE) {
1699 mark_mapped_object_invalid(t, r->id, i);
1702 if (st == OSL_NO_STORAGE)
1703 free(r->volatile_objects[col->volatile_num].data);
1705 if (t->num_mapped_columns) {
1706 ret = mark_row_invalid(t, r->id);
1709 t->num_invalid_rows++;
1714 free(r->volatile_objects);
1719 /* test if column has an rbtree */
1720 static int check_rbtree_col(const struct osl_table *t, unsigned col_num,
1721 struct osl_column **col)
1724 return -E_BAD_TABLE;
1725 if (!(get_column_description(t->desc, col_num)->storage_flags & OSL_RBTREE))
1726 return -E_BAD_STORAGE_FLAGS;
1727 *col = t->columns + col_num;
1732 * Get the row that contains the given object.
1734 * \param t Pointer to an open osl table.
1735 * \param col_num The number of the column to be searched.
1736 * \param obj The object to be looked up.
1737 * \param result Points to the row containing \a obj.
1739 * Lookup \a obj in \a t and return the row containing \a obj. The column
1740 * specified by \a col_num must have an associated rbtree.
1742 * \return Positive on success, negative on errors. If an error occured, \a
1743 * result is set to \p NULL. Possible errors include: \p E_BAD_TABLE, \p
1744 * E_BAD_STORAGE_FLAGS, errors returned by get_mapped_object(), \p
1745 * E_RB_KEY_NOT_FOUND.
1747 * \sa osl_storage_flags
1749 int osl_get_row(const struct osl_table *t, unsigned col_num,
1750 const struct osl_object *obj, struct osl_row **result)
1753 struct rb_node *node;
1754 struct osl_row *row;
1755 struct osl_column *col;
1758 ret = check_rbtree_col(t, col_num, &col);
1761 ret = search_rbtree(obj, t, col_num, &node, NULL);
1764 row = get_row_pointer(node, t->columns[col_num].rbtree_num);
1769 static int rbtree_loop(struct osl_column *col, void *private_data,
1770 osl_rbtree_loop_func *func)
1774 for (n = rb_first(&col->rbtree); n; n = rb_next(n)) {
1775 struct osl_row *r = get_row_pointer(n, col->rbtree_num);
1776 int ret = func(r, private_data);
1783 static int rbtree_loop_reverse(struct osl_column *col, void *private_data,
1784 osl_rbtree_loop_func *func)
1788 for (n = rb_last(&col->rbtree); n; n = rb_prev(n)) {
1789 struct osl_row *r = get_row_pointer(n, col->rbtree_num);
1790 int ret = func(r, private_data);
1798 * Loop over all nodes in an rbtree.
1800 * \param t Pointer to an open osl table.
1801 * \param col_num The column to use for iterating over the elements.
1802 * \param private_data Pointer that gets passed to \a func.
1803 * \param func The function to be called for each node in the rbtree.
1805 * This function does an in-order walk of the rbtree associated with \a
1806 * col_num. It is an error if the \p OSL_RBTREE flag is not set for this
1807 * column. For each node in the rbtree, the given function \a func is called
1808 * with two \p void* pointers as arguments: The first argument points to the
1809 * row that contains the object corresponding to the rbtree node currently
1810 * traversed, and the \a private_data pointer is passed to \a func as the
1811 * second argument. The loop terminates either if \a func returns a negative
1812 * value, or if all nodes of the tree have been visited.
1815 * \return Positive on success, negative on errors. If the termination of the
1816 * loop was caused by \a func returning a negative value, this value is
1819 * \sa osl_storage_flags, osl_rbtree_loop_reverse(), osl_compare_func.
1821 int osl_rbtree_loop(const struct osl_table *t, unsigned col_num,
1822 void *private_data, osl_rbtree_loop_func *func)
1824 struct osl_column *col;
1826 int ret = check_rbtree_col(t, col_num, &col);
1829 return rbtree_loop(col, private_data, func);
1833 * Loop over all nodes in an rbtree in reverse order.
1835 * \param t Identical meaning as in \p osl_rbtree_loop().
1836 * \param col_num Identical meaning as in \p osl_rbtree_loop().
1837 * \param private_data Identical meaning as in \p osl_rbtree_loop().
1838 * \param func Identical meaning as in \p osl_rbtree_loop().
1840 * This function is identical to \p osl_rbtree_loop(), the only difference
1841 * is that the tree is walked in reverse order.
1843 * \return The same return value as \p osl_rbtree_loop().
1845 * \sa osl_rbtree_loop().
1847 int osl_rbtree_loop_reverse(const struct osl_table *t, unsigned col_num,
1848 void *private_data, osl_rbtree_loop_func *func)
1850 struct osl_column *col;
1852 int ret = check_rbtree_col(t, col_num, &col);
1855 return rbtree_loop_reverse(col, private_data, func);
1858 /* TODO: Rollback changes on errors */
1859 static int rename_disk_storage_objects(struct osl_table *t,
1860 struct osl_object *old_obj, struct osl_object *new_obj)
1863 const struct osl_column_description *cd;
1864 char *old_ds_name, *new_ds_name;
1866 if (!t->num_disk_storage_columns)
1867 return 1; /* nothing to do */
1868 if (old_obj->size == new_obj->size && !memcmp(new_obj->data,
1869 old_obj->data, new_obj->size))
1870 return 1; /* object did not change */
1871 old_ds_name = disk_storage_name_of_object(t, old_obj);
1872 new_ds_name = disk_storage_name_of_object(t, new_obj);
1873 FOR_EACH_DISK_STORAGE_COLUMN(i, t, cd) {
1874 char *old_filename, *new_filename;
1875 ret = create_disk_storage_object_dir(t, i, new_ds_name);
1878 old_filename = disk_storage_path(t, i, old_ds_name);
1879 new_filename = disk_storage_path(t, i, new_ds_name);
1880 ret = para_rename(old_filename, new_filename);
1895 * Change an object in an osl table.
1897 * \param t Pointer to an open osl table.
1898 * \param r Pointer to the row containing the object to be updated.
1899 * \param col_num Number of the column containing the object to be updated.
1900 * \param obj Pointer to the replacement object.
1902 * This function gets rid of all references to the old object. This includes
1903 * removal of the rbtree node in case there is an rbtree associated with \a
1904 * col_num. It then inserts \a obj into the table and the rbtree if neccessary.
1906 * If the \p OSL_RBTREE flag is set for \a col_num, you \b MUST call this
1907 * function in order to change the contents of an object, even for volatile or
1908 * mapped columns of constant size (which may be updated directly if \p
1909 * OSL_RBTREE is not set). Otherwise the rbtree might become corrupted.
1911 * \return Positive on success, negative on errors. Possible errors include: \p
1912 * E_BAD_TABLE, \p E_RB_KEY_EXISTS, \p E_BAD_SIZE, \p E_NOENT, \p E_UNLINK,
1913 * errors returned by para_write_file(), \p E_MKDIR.
1915 int osl_update_object(struct osl_table *t, const struct osl_row *r,
1916 unsigned col_num, struct osl_object *obj)
1918 struct osl_column *col;
1919 const struct osl_column_description *cd;
1923 return -E_BAD_TABLE;
1924 col = &t->columns[col_num];
1925 cd = get_column_description(t->desc, col_num);
1926 PARA_DEBUG_LOG("updating column %u of %s\n", col_num, t->desc->name);
1927 if (cd->storage_flags & OSL_RBTREE) {
1928 if (search_rbtree(obj, t, col_num, NULL, NULL) > 0)
1929 return -E_RB_KEY_EXISTS;
1931 if (cd->storage_flags & OSL_FIXED_SIZE) {
1932 if (obj->size != cd->data_size)
1935 remove_rb_node(t, col_num, r);
1936 if (cd->storage_type == OSL_NO_STORAGE) { /* TODO: If fixed size, reuse object? */
1937 free(r->volatile_objects[col->volatile_num].data);
1938 r->volatile_objects[col->volatile_num] = *obj;
1939 } else if (cd->storage_type == OSL_DISK_STORAGE) {
1941 ret = disk_storage_name_of_row(t, r, &ds_name);
1944 ret = delete_disk_storage_file(t, col_num, ds_name);
1945 if (ret < 0 && ret != -E_NOENT) {
1949 ret = write_disk_storage_file(t, col_num, obj, ds_name);
1953 } else { /* mapped storage */
1954 struct osl_object old_obj;
1955 ret = get_mapped_object(t, col_num, r->id, &old_obj);
1959 * If the updated column is the disk storage name column, the
1960 * disk storage name changes, so we have to rename all disk
1961 * storage objects accordingly.
1963 if (col_num == t->disk_storage_name_column) {
1964 ret = rename_disk_storage_objects(t, &old_obj, obj);
1968 if (cd->storage_flags & OSL_FIXED_SIZE)
1969 memcpy(old_obj.data, obj->data, cd->data_size);
1970 else { /* TODO: if the size doesn't change, use old space */
1971 uint32_t new_data_map_size;
1973 ret = get_row_index(t, r->id, &row_index);
1976 ret = mark_mapped_object_invalid(t, r->id, col_num);
1979 unmap_column(t, col_num);
1980 ret = append_map_file(t, col_num, obj,
1981 &new_data_map_size);
1984 ret = map_column(t, col_num);
1987 update_cell_index(row_index, col, new_data_map_size,
1991 if (cd->storage_flags & OSL_RBTREE) {
1992 ret = insert_rbtree(t, col_num, r, obj);
2000 * Retrieve an object of type \p OSL_DISK_STORAGE by row and column.
2002 * \param t Pointer to an open osl table.
2003 * \param r Pointer to the row containing the object.
2004 * \param col_num The column number.
2005 * \param obj Points to the result upon successful return.
2007 * For columns of type \p OSL_DISK_STORAGE, this function must be used to
2008 * retrieve one of its containing objects. Afterwards, osl_close_disk_object()
2009 * must be called in order to deallocate the resources.
2011 * \return Positive on success, negative on errors. Possible errors include:
2012 * \p E_BAD_TABLE, \p E_BAD_STORAGE_TYPE, errors returned by osl_get_object().
2014 * \sa osl_get_object(), osl_storage_type, osl_close_disk_object().
2016 int osl_open_disk_object(const struct osl_table *t, const struct osl_row *r,
2017 unsigned col_num, struct osl_object *obj)
2019 const struct osl_column_description *cd;
2020 char *ds_name, *filename;
2024 return -E_BAD_TABLE;
2025 cd = get_column_description(t->desc, col_num);
2026 if (cd->storage_type != OSL_DISK_STORAGE)
2027 return -E_BAD_STORAGE_TYPE;
2029 ret = disk_storage_name_of_row(t, r, &ds_name);
2032 filename = disk_storage_path(t, col_num, ds_name);
2034 PARA_DEBUG_LOG("filename: %s\n", filename);
2035 ret = mmap_full_file(filename, O_RDONLY, obj);
2041 * Free resources that were allocated during osl_open_disk_object().
2043 * \param obj Pointer to the object previously returned by open_disk_object().
2045 * \return The return value of the underlying call to para_munmap().
2047 * \sa para_munmap().
2049 int osl_close_disk_object(struct osl_object *obj)
2051 return para_munmap(obj->data, obj->size);
2055 * Get the number of rows of the given table.
2057 * \param t Pointer to an open osl table.
2058 * \param num_rows Result is returned here.
2060 * The number of rows returned via \a num_rows excluding any invalid rows.
2062 * \return Positive on success, \p -E_BAD_TABLE if \a t is \p NULL.
2064 int osl_get_num_rows(const struct osl_table *t, unsigned *num_rows)
2067 return -E_BAD_TABLE;
2068 assert(t->num_rows >= t->num_invalid_rows);
2069 *num_rows = t->num_rows - t->num_invalid_rows;
2074 * Get the rank of a row.
2076 * \param t An open osl table.
2077 * \param r The row to get the rank of.
2078 * \param col_num The number of an rbtree column.
2079 * \param rank Result pointer.
2081 * The rank is, by definition, the position of the row in the linear order
2082 * determined by an inorder tree walk of the rbtree associated with column
2083 * number \a col_num of \a table.
2085 * \return Positive on success, negative on errors.
2087 * \sa osl_get_nth_row().
2089 int osl_get_rank(const struct osl_table *t, struct osl_row *r,
2090 unsigned col_num, unsigned *rank)
2092 struct osl_object obj;
2093 struct osl_column *col;
2094 struct rb_node *node;
2095 int ret = check_rbtree_col(t, col_num, &col);
2099 ret = osl_get_object(t, r, col_num, &obj);
2102 ret = search_rbtree(&obj, t, col_num, &node, NULL);
2105 ret = rb_rank(node, rank);
2112 * Get the row with n-th greatest value.
2114 * \param t Pointer to an open osl table.
2115 * \param col_num The column number.
2116 * \param n The rank of the desired row.
2117 * \param result Row is returned here.
2119 * Retrieve the n-th order statistic with respect to the compare function
2120 * of the rbtree column \a col_num. In other words, get that row with
2121 * \a n th greatest value in column \a col_num. It's an error if
2122 * \a col_num is not a rbtree column, or if \a n is larger than the
2123 * number of rows in the table.
2125 * \return Positive on success, negative on errors. Possible errors:
2126 * \p E_BAD_TABLE, \p E_BAD_STORAGE_FLAGS, \p E_RB_KEY_NOT_FOUND.
2128 * \sa osl_storage_flags, osl_compare_func, osl_get_row(),
2129 * osl_rbtree_last_row(), osl_rbtree_first_row(), osl_get_rank().
2131 int osl_get_nth_row(const struct osl_table *t, unsigned col_num,
2132 unsigned n, struct osl_row **result)
2134 struct osl_column *col;
2135 struct rb_node *node;
2136 int ret = check_rbtree_col(t, col_num, &col);
2140 node = rb_nth(col->rbtree.rb_node, n);
2142 return -E_RB_KEY_NOT_FOUND;
2143 *result = get_row_pointer(node, col->rbtree_num);
2148 * Get the row corresponding to the smallest rbtree node of a column.
2150 * \param t An open rbtree table.
2151 * \param col_num The number of the rbtree column.
2152 * \param result A pointer to the first row is returned here.
2154 * The rbtree node of the smallest object (with respect to the corresponding
2155 * compare function) is selected and the row containing this object is
2156 * returned. It is an error if \a col_num refers to a column without an
2157 * associated rbtree.
2159 * \return Positive on success, negative on errors.
2161 * \sa osl_get_nth_row(), osl_rbtree_last_row().
2163 int osl_rbtree_first_row(const struct osl_table *t, unsigned col_num,
2164 struct osl_row **result)
2166 return osl_get_nth_row(t, col_num, 1, result);
2170 * Get the row corresponding to the greatest rbtree node of a column.
2172 * \param t The same meaning as in \p osl_rbtree_first_row().
2173 * \param col_num The same meaning as in \p osl_rbtree_first_row().
2174 * \param result The same meaning as in \p osl_rbtree_first_row().
2176 * This function works just like osl_rbtree_first_row(), the only difference
2177 * is that the row containing the greatest rather than the smallest object is
2180 * \return Positive on success, negative on errors.
2182 * \sa osl_get_nth_row(), osl_rbtree_first_row().
2184 int osl_rbtree_last_row(const struct osl_table *t, unsigned col_num,
2185 struct osl_row **result)
2188 int ret = osl_get_num_rows(t, &num_rows);
2192 return osl_get_nth_row(t, col_num, num_rows, result);