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() */
16 * A wrapper for lseek(2).
18 * \param fd The filedescriptor whose offset is to be to repositioned.
19 * \param offset A value-result parameter.
20 * \param whence Usual repositioning directive.
22 * Reposition the offset of the file descriptor \a fd to the argument \a offset
23 * according to the directive \a whence. Upon successful return, \a offset
24 * contains the resulting offset location as measured in bytes from the
25 * beginning of the file.
27 * \return Positive on success. Otherwise, the function returns \p -E_LSEEK.
31 int para_lseek(int fd, off_t *offset, int whence)
33 *offset = lseek(fd, *offset, whence);
41 * Wrapper for the write system call.
43 * \param fd The file descriptor to write to.
44 * \param buf The buffer to write.
45 * \param size The length of \a buf in bytes.
47 * This function writes out the given bufffer and retries if an interrupt
48 * occured during the write.
50 * \return On success, the number of bytes written is returned, otherwise, the
51 * function returns \p -E_WRITE.
55 ssize_t para_write(int fd, const void *buf, size_t size)
60 ret = write(fd, buf, size);
61 if ((ret < 0) && (errno == EAGAIN || errno == EINTR))
63 return ret >= 0? ret : -E_WRITE;
68 * Write the whole buffer to a file descriptor.
70 * \param fd The file descriptor to write to.
71 * \param buf The buffer to write.
72 * \param size The length of \a buf in bytes.
74 * This function writes the given buffer and continues on short writes and
75 * when interrupted by a signal.
77 * \return Positive on success, negative on errors. Possible errors: any
78 * errors returned by para_write().
82 ssize_t para_write_all(int fd, const void *buf, size_t size)
84 // PARA_DEBUG_LOG("writing %zu bytes\n", size);
87 ssize_t ret = para_write(fd, b, size);
88 // PARA_DEBUG_LOG("ret: %zd\n", ret);
97 * Wrapper for the open(2) system call.
99 * \param path The filename.
100 * \param flags The usual open(2) flags.
101 * \param mode Specifies the permissions to use.
103 * The mode parameter must be specified when O_CREAT is in the flags, and is ignored
106 * \return Positive on success, negative on errors. Possible errors: \p
107 * E_EXIST, \p E_ISDIR, \p E_NOENT, \p E_OSL_PERM.
111 int para_open(const char *path, int flags, mode_t mode)
113 PARA_DEBUG_LOG("opening %s\n", path);
114 int ret = open(path, flags, mode);
132 PARA_ERROR_LOG("failed to open %s: %s\n", path, strerror(errno));
137 * Open a file, write the given buffer and close the file.
139 * \param filename Full path to the file to open.
140 * \param buf The buffer to write to the file.
141 * \param size The size of \a buf.
143 * \return Positive on success, negative on errors. Possible errors include:
144 * any errors from para_open() or para_write().
146 * \sa para_open(), para_write().
148 int para_write_file(const char *filename, const void *buf, size_t size)
152 ret = para_open(filename, O_WRONLY | O_CREAT | O_EXCL, 0644);
156 ret = para_write_all(fd, buf, size);
165 static int append_file(const char *filename, char *header, size_t header_size,
166 char *data, size_t data_size, uint32_t *new_pos)
170 // PARA_DEBUG_LOG("appending %zu + %zu bytes\n", header_size, data_size);
171 ret = para_open(filename, O_WRONLY | O_CREAT | O_APPEND, 0644);
175 if (header && header_size) {
176 ret = para_write_all(fd, header, header_size);
180 ret = para_write_all(fd, data, data_size);
185 ret = para_lseek(fd, &offset, SEEK_END);
188 // PARA_DEBUG_LOG("new file size: " FMT_OFF_T "\n", offset);
198 * Map a file into memory.
200 * \param path Name of the regular file to map.
201 * \param open_mode Either \p O_RDONLY or \p O_RDWR.
202 * \param obj On success, the mapping is returned here.
204 * \return Positive on success, negative on errors. Possible errors include: \p
205 * E_FSTAT, any errors returned by para_open(), \p E_EMPTY, \p E_MMAP.
207 * \sa para_open(), mmap(2).
209 int mmap_full_file(const char *path, int open_mode, struct osl_object *obj)
211 int fd, ret, mmap_prot, mmap_flags;
212 struct stat file_status;
214 if (open_mode == O_RDONLY) {
215 mmap_prot = PROT_READ;
216 mmap_flags = MAP_PRIVATE;
218 mmap_prot = PROT_READ | PROT_WRITE;
219 mmap_flags = MAP_SHARED;
221 ret = para_open(path, open_mode, 0);
226 if (fstat(fd, &file_status) < 0)
228 obj->size = file_status.st_size;
230 PARA_DEBUG_LOG("%s: size %zu\n", path, obj->size);
233 obj->data = mmap(NULL, obj->size, mmap_prot, mmap_flags, fd, 0);
234 if (obj->data == MAP_FAILED) {
246 * Traverse the given directory recursively.
248 * \param dirname The directory to traverse.
249 * \param func The function to call for each entry.
250 * \param private_data Pointer to an arbitrary data structure.
252 * For each regular file in \a dirname, the supplied function \a func is
253 * called. The full path of the regular file and the \a private_data pointer
254 * are passed to \a func.
256 * \return On success, 1 is returned. Otherwise, this function returns a
257 * negative value which indicates the kind of the error.
259 int for_each_file_in_dir(const char *dirname,
260 int (*func)(const char *, const void *), const void *private_data)
263 struct dirent *entry;
265 * Opening the current directory (".") and calling fchdir() to return
266 * is usually faster and more reliable than saving cwd in some buffer
267 * and calling chdir() afterwards (see man 3 getcwd).
269 int cwd_fd = open(".", O_RDONLY);
272 // PARA_DEBUG_LOG("dirname: %s\n", dirname);
274 return -E_OSL_GETCWD;
276 if (chdir(dirname) < 0)
278 ret = -E_OSL_OPENDIR;
282 /* scan cwd recursively */
283 while ((entry = readdir(dir))) {
288 if (!strcmp(entry->d_name, "."))
290 if (!strcmp(entry->d_name, ".."))
292 if (lstat(entry->d_name, &s) == -1)
295 if (!S_ISREG(m) && !S_ISDIR(m))
297 tmp = make_message("%s/%s", dirname, entry->d_name);
299 ret = func(tmp, private_data);
306 ret = for_each_file_in_dir(tmp, func, private_data);
315 if (fchdir(cwd_fd) < 0 && ret >= 0)
321 int para_mkdir(const char *path, mode_t mode)
323 if (!mkdir(path, mode))
329 if (errno == ENOTDIR)
336 static int verify_name(const char *name)
342 if (strchr(name, '/'))
344 if (!strcmp(name, ".."))
346 if (!strcmp(name, "."))
352 * Compare two osl objects pointing to unsigned integers of 32 bit size.
354 * \param obj1 Pointer to the first integer.
355 * \param obj2 Pointer to the second integer.
357 * \return The values required for an osl compare function.
359 * \sa osl_compare_func, osl_hash_compare().
361 int uint32_compare(const struct osl_object *obj1, const struct osl_object *obj2)
363 uint32_t d1 = read_u32((const char *)obj1->data);
364 uint32_t d2 = read_u32((const char *)obj2->data);
374 * Compare two osl objects pointing to hash values.
376 * \param obj1 Pointer to the first hash object.
377 * \param obj2 Pointer to the second hash object.
379 * \return The values required for an osl compare function.
381 * \sa osl_compare_func, uint32_compare().
383 int osl_hash_compare(const struct osl_object *obj1, const struct osl_object *obj2)
385 return hash_compare((HASH_TYPE *)obj1->data, (HASH_TYPE *)obj2->data);
388 static char *disk_storage_dirname(const struct osl_table *t, unsigned col_num,
391 char *dirname, *column_name = column_filename(t, col_num);
393 if (!(t->desc->flags & OSL_LARGE_TABLE))
395 dirname = make_message("%s/%.2s", column_name, ds_name);
400 static char *disk_storage_name_of_object(const struct osl_table *t,
401 const struct osl_object *obj)
403 HASH_TYPE hash[HASH_SIZE];
404 hash_object(obj, hash);
405 return disk_storage_name_of_hash(t, hash);
408 static int disk_storage_name_of_row(const struct osl_table *t,
409 const struct osl_row *row, char **name)
411 struct osl_object obj;
412 int ret = osl_get_object(t, row, t->disk_storage_name_column, &obj);
416 *name = disk_storage_name_of_object(t, &obj);
420 static void column_name_hash(const char *col_name, HASH_TYPE *hash)
422 return hash_function(col_name, strlen(col_name), hash);
425 static int init_column_descriptions(struct osl_table *t)
428 const struct osl_column_description *cd;
430 ret = -E_BAD_TABLE_DESC;
431 ret = verify_name(t->desc->name);
437 /* the size of the index header without column descriptions */
438 t->index_header_size = IDX_COLUMN_DESCRIPTIONS;
439 FOR_EACH_COLUMN(i, t->desc, cd) {
440 struct osl_column *col = t->columns + i;
441 if (cd->storage_flags & OSL_RBTREE) {
442 if (!cd->compare_function)
443 return -E_NO_COMPARE_FUNC;
445 if (cd->storage_type == OSL_NO_STORAGE)
447 ret = -E_NO_COLUMN_NAME;
448 if (!cd->name || !cd->name[0])
450 ret = verify_name(cd->name);
453 t->index_header_size += index_column_description_size(cd->name);
454 column_name_hash(cd->name, col->name_hash);
455 ret = -E_DUPLICATE_COL_NAME;
456 for (j = i + 1; j < t->desc->num_columns; j++) {
457 const char *name2 = get_column_description(t->desc,
459 if (cd->name && name2 && !strcmp(cd->name, name2))
469 * Initialize a struct table from given table description.
471 * \param desc The description of the osl table.
472 * \param table_ptr Result is returned here.
474 * This function performs several sanity checks on \p desc and returns if any
475 * of these tests fail. On success, a struct \p osl_table is allocated and
476 * initialized with data derived from \p desc.
478 * \return Positive on success, negative on errors. Possible errors include: \p
479 * E_BAD_TABLE_DESC, \p E_NO_COLUMN_DESC, \p E_NO_COLUMNS, \p
480 * E_BAD_STORAGE_TYPE, \p E_BAD_STORAGE_FLAGS, \p E_BAD_STORAGE_SIZE, \p
481 * E_NO_UNIQUE_RBTREE_COLUMN, \p E_NO_RBTREE_COL.
483 * \sa struct osl_table.
485 int init_table_structure(const struct osl_table_description *desc,
486 struct osl_table **table_ptr)
488 const struct osl_column_description *cd;
489 struct osl_table *t = para_calloc(sizeof(*t));
490 int i, ret = -E_BAD_TABLE_DESC, have_disk_storage_name_column = 0;
494 PARA_DEBUG_LOG("creating table structure for '%s' from table "
495 "description\n", desc->name);
496 ret = -E_NO_COLUMN_DESC;
497 if (!desc->column_descriptions)
500 if (!desc->num_columns)
502 t->columns = para_calloc(desc->num_columns * sizeof(struct osl_column));
504 FOR_EACH_COLUMN(i, t->desc, cd) {
505 enum osl_storage_type st = cd->storage_type;
506 enum osl_storage_flags sf = cd->storage_flags;
507 struct osl_column *col = &t->columns[i];
509 ret = -E_BAD_STORAGE_TYPE;
510 if (st != OSL_MAPPED_STORAGE && st != OSL_DISK_STORAGE
511 && st != OSL_NO_STORAGE)
513 ret = -E_BAD_STORAGE_FLAGS;
514 if (st == OSL_DISK_STORAGE && sf & OSL_RBTREE)
516 ret = -E_BAD_STORAGE_SIZE;
517 if (sf & OSL_FIXED_SIZE && !cd->data_size)
520 case OSL_DISK_STORAGE:
521 t->num_disk_storage_columns++;
523 case OSL_MAPPED_STORAGE:
524 t->num_mapped_columns++;
525 col->index_offset = t->row_index_size;
526 t->row_index_size += 8;
529 col->volatile_num = t->num_volatile_columns;
530 t->num_volatile_columns++;
533 if (sf & OSL_RBTREE) {
534 col->rbtree_num = t->num_rbtrees;
536 if ((sf & OSL_UNIQUE) && (st == OSL_MAPPED_STORAGE)) {
537 if (!have_disk_storage_name_column)
538 t->disk_storage_name_column = i;
539 have_disk_storage_name_column = 1;
543 ret = -E_NO_UNIQUE_RBTREE_COLUMN;
544 if (t->num_disk_storage_columns && !have_disk_storage_name_column)
546 ret = -E_NO_RBTREE_COL;
550 PARA_DEBUG_LOG("OK. Index entry size: %u\n", t->row_index_size);
551 ret = init_column_descriptions(t);
563 * Read the table description from index header.
565 * \param map The memory mapping of the index file.
566 * \param desc The values found in the index header are returned here.
568 * Read the index header, check for the paraslash magic string and the table version number.
569 * Read all information stored in the index header into \a desc.
571 * \return Positive on success, negative on errors.
573 * \sa struct osl_table_description, osl_create_table.
575 int read_table_desc(struct osl_object *map, struct osl_table_description *desc)
577 char *buf = map->data;
579 uint16_t header_size;
582 struct osl_column_description *cd;
584 if (map->size < MIN_INDEX_HEADER_SIZE(1))
585 return -E_SHORT_TABLE;
586 if (strncmp(buf + IDX_PARA_MAGIC, PARA_MAGIC, strlen(PARA_MAGIC)))
588 version = read_u8(buf + IDX_VERSION);
589 if (version < MIN_TABLE_VERSION || version > MAX_TABLE_VERSION)
590 return -E_VERSION_MISMATCH;
591 desc->num_columns = read_u8(buf + IDX_TABLE_FLAGS);
592 desc->flags = read_u8(buf + IDX_TABLE_FLAGS);
593 desc->num_columns = read_u16(buf + IDX_NUM_COLUMNS);
594 PARA_DEBUG_LOG("%u columns\n", desc->num_columns);
595 if (!desc->num_columns)
596 return -E_NO_COLUMNS;
597 header_size = read_u16(buf + IDX_HEADER_SIZE);
598 if (map->size < header_size)
600 desc->column_descriptions = para_calloc(desc->num_columns
601 * sizeof(struct osl_column_description));
602 offset = IDX_COLUMN_DESCRIPTIONS;
603 FOR_EACH_COLUMN(i, desc, cd) {
606 ret = -E_SHORT_TABLE;
607 if (map->size < offset + MIN_IDX_COLUMN_DESCRIPTION_SIZE) {
608 PARA_ERROR_LOG("map size = %zu < %u = offset + min desc size\n",
609 map->size, offset + MIN_IDX_COLUMN_DESCRIPTION_SIZE);
612 cd->storage_type = read_u16(buf + offset + IDX_CD_STORAGE_TYPE);
613 cd->storage_flags = read_u16(buf + offset +
614 IDX_CD_STORAGE_FLAGS);
615 cd->data_size = read_u32(buf + offset + IDX_CD_DATA_SIZE);
616 null_byte = memchr(buf + offset + IDX_CD_NAME, '\0',
617 map->size - offset - IDX_CD_NAME);
618 ret = -E_INDEX_CORRUPTION;
621 cd->name = para_strdup(buf + offset + IDX_CD_NAME);
622 offset += index_column_description_size(cd->name);
624 if (offset != header_size) {
625 ret = -E_INDEX_CORRUPTION;
626 PARA_ERROR_LOG("real header size = %u != %u = stored header size\n",
627 offset, header_size);
632 FOR_EACH_COLUMN(i, desc, cd)
638 * check whether the table description given by \p t->desc matches the on-disk
639 * table structure stored in the index of \a t.
641 static int compare_table_descriptions(struct osl_table *t)
644 struct osl_table_description desc;
645 const struct osl_column_description *cd1, *cd2;
647 /* read the on-disk structure into desc */
648 ret = read_table_desc(&t->index_map, &desc);
651 ret = -E_BAD_TABLE_FLAGS;
652 if (desc.flags != t->desc->flags)
654 ret = E_BAD_COLUMN_NUM;
655 if (desc.num_columns != t->desc->num_columns)
657 FOR_EACH_COLUMN(i, t->desc, cd1) {
658 cd2 = get_column_description(&desc, i);
659 ret = -E_BAD_STORAGE_TYPE;
660 if (cd1->storage_type != cd2->storage_type)
662 ret = -E_BAD_STORAGE_FLAGS;
663 if (cd1->storage_flags != cd2->storage_flags) {
664 PARA_ERROR_LOG("sf1 = %u != %u = sf2\n",
665 cd1->storage_flags, cd2->storage_flags);
668 ret = -E_BAD_DATA_SIZE;
669 if (cd1->storage_flags & OSL_FIXED_SIZE)
670 if (cd1->data_size != cd2->data_size)
672 ret = -E_BAD_COLUMN_NAME;
673 if (strcmp(cd1->name, cd2->name))
676 PARA_DEBUG_LOG("table description of '%s' matches on-disk data, good\n",
680 FOR_EACH_COLUMN(i, &desc, cd1)
682 free(desc.column_descriptions);
686 static int create_table_index(struct osl_table *t)
688 char *buf, *filename;
690 size_t size = t->index_header_size;
691 const struct osl_column_description *cd;
694 PARA_INFO_LOG("creating %zu byte index for table %s\n", size,
696 buf = para_calloc(size);
697 sprintf(buf + IDX_PARA_MAGIC, "%s", PARA_MAGIC);
698 write_u8(buf + IDX_TABLE_FLAGS, t->desc->flags);
699 write_u8(buf + IDX_DIRTY_FLAG, 0);
700 write_u8(buf + IDX_VERSION, CURRENT_TABLE_VERSION);
701 write_u16(buf + IDX_NUM_COLUMNS, t->desc->num_columns);
702 write_u16(buf + IDX_HEADER_SIZE, t->index_header_size);
703 offset = IDX_COLUMN_DESCRIPTIONS;
704 FOR_EACH_COLUMN(i, t->desc, cd) {
705 write_u16(buf + offset + IDX_CD_STORAGE_TYPE,
707 write_u16(buf + offset + IDX_CD_STORAGE_FLAGS,
709 if (cd->storage_flags & OSL_FIXED_SIZE)
710 write_u32(buf + offset + IDX_CD_DATA_SIZE,
712 strcpy(buf + offset + IDX_CD_NAME, cd->name);
713 offset += index_column_description_size(cd->name);
715 assert(offset = size);
716 filename = index_filename(t->desc);
717 ret = para_write_file(filename, buf, size);
724 * Create a new osl table.
726 * \param desc Pointer to the table description.
728 * \return Positive on success, negative on errors. Possible errors include: \p
729 * E_BAD_TABLE_DESC, \p E_BAD_DB_DIR, \p E_BAD_NAME, \p E_NO_COMPARE_FUNC, \p
730 * E_NO_COLUMN_NAME, \p E_DUPLICATE_COL_NAME, \p E_MKDIR, any errors returned
733 int osl_create_table(const struct osl_table_description *desc)
735 const struct osl_column_description *cd;
736 char *table_dir = NULL, *filename;
738 int i, ret = init_table_structure(desc, &t);
742 PARA_INFO_LOG("creating %s\n", desc->name);
743 FOR_EACH_COLUMN(i, t->desc, cd) {
744 if (cd->storage_type == OSL_NO_STORAGE)
747 ret = para_mkdir(desc->dir, 0777);
748 if (ret < 0 && ret != -E_EXIST)
750 table_dir = make_message("%s/%s", desc->dir,
752 ret = para_mkdir(table_dir, 0777);
756 filename = column_filename(t, i);
757 PARA_INFO_LOG("filename: %s\n", filename);
758 if (cd->storage_type == OSL_MAPPED_STORAGE) {
759 ret = para_open(filename, O_RDWR | O_CREAT | O_EXCL,
768 ret = para_mkdir(filename, 0777);
773 if (t->num_mapped_columns) {
774 ret = create_table_index(t);
786 static int table_is_dirty(struct osl_table *t)
788 char *buf = (char *)t->index_map.data + IDX_DIRTY_FLAG;
789 uint8_t dirty = read_u8(buf) & 0x1;
793 static void mark_table_dirty(struct osl_table *t)
795 char *buf = (char *)t->index_map.data + IDX_DIRTY_FLAG;
796 write_u8(buf, read_u8(buf) | 1);
799 static void mark_table_clean(struct osl_table *t)
801 char *buf = (char *)t->index_map.data + IDX_DIRTY_FLAG;
802 write_u8(buf, read_u8(buf) & 0xfe);
805 static void unmap_column(struct osl_table *t, unsigned col_num)
807 struct osl_object map = t->columns[col_num].data_map;
811 ret = para_munmap(map.data, map.size);
817 * Unmap all mapped files of an osl table.
819 * \param t Pointer to a mapped table.
820 * \param flags Options for unmapping.
822 * \return Positive on success, negative on errors. Possible errors include:
823 * E_NOT_MAPPED, E_MUNMAP.
825 * \sa map_table(), enum osl_close_flags, para_munmap().
827 int unmap_table(struct osl_table *t, enum osl_close_flags flags)
830 const struct osl_column_description *cd;
833 if (!t->num_mapped_columns) /* can this ever happen? */
835 PARA_DEBUG_LOG("unmapping table '%s'\n", t->desc->name);
836 if (!t->index_map.data)
837 return -E_NOT_MAPPED;
838 if (flags & OSL_MARK_CLEAN)
840 ret = para_munmap(t->index_map.data, t->index_map.size);
843 t->index_map.data = NULL;
846 FOR_EACH_MAPPED_COLUMN(i, t, cd)
851 static int map_column(struct osl_table *t, unsigned col_num)
854 char *filename = column_filename(t, col_num);
856 if (stat(filename, &statbuf) < 0) {
860 if (!(S_IFREG & statbuf.st_mode)) {
864 ret = mmap_full_file(filename, O_RDWR,
865 &t->columns[col_num].data_map);
871 * Map the index file and all columns of type \p OSL_MAPPED_STORAGE into memory.
873 * \param t Pointer to an initialized table structure.
874 * \param flags Mapping options.
876 * \return Negative return value on errors; on success the number of rows
877 * (including invalid rows) is returned.
879 * \sa unmap_table(), enum map_table_flags, osl_open_table(), mmap(2).
881 int map_table(struct osl_table *t, enum map_table_flags flags)
884 const struct osl_column_description *cd;
885 int i = 0, ret, num_rows = 0;
887 if (!t->num_mapped_columns)
889 if (t->index_map.data)
890 return -E_ALREADY_MAPPED;
891 filename = index_filename(t->desc);
892 PARA_DEBUG_LOG("mapping table '%s' (index: %s)\n", t->desc->name, filename);
893 ret = mmap_full_file(filename, flags & MAP_TBL_FL_MAP_RDONLY?
894 O_RDONLY : O_RDWR, &t->index_map);
898 if (flags & MAP_TBL_FL_VERIFY_INDEX) {
899 ret = compare_table_descriptions(t);
904 if (!(flags & MAP_TBL_FL_IGNORE_DIRTY)) {
905 if (table_is_dirty(t)) {
906 PARA_ERROR_LOG("%s is dirty\n", t->desc->name);
911 num_rows = table_num_rows(t);
915 FOR_EACH_MAPPED_COLUMN(i, t, cd) {
916 ret = map_column(t, i);
921 err: /* unmap what is already mapped */
922 for (i--; i >= 0; i--) {
923 struct osl_object map = t->columns[i].data_map;
924 para_munmap(map.data, map.size);
927 para_munmap(t->index_map.data, t->index_map.size);
928 t->index_map.data = NULL;
933 * Retrieve a mapped object by row and column number.
935 * \param t Pointer to an open osl table.
936 * \param col_num Number of the mapped column containing the object to retrieve.
937 * \param row_num Number of the row containing the object to retrieve.
938 * \param obj The result is returned here.
940 * It is considered an error if \a col_num does not refer to a column
941 * of storage type \p OSL_MAPPED_STORAGE.
943 * \return Positive on success, negative on errors. Possible errors include:
944 * \p E_BAD_ROW_NUM, \p E_INVALID_OBJECT.
946 * \sa osl_storage_type.
948 int get_mapped_object(const struct osl_table *t, unsigned col_num,
949 uint32_t row_num, struct osl_object *obj)
951 struct osl_column *col = &t->columns[col_num];
957 if (t->num_rows <= row_num)
958 return -E_BAD_ROW_NUM;
959 ret = get_cell_index(t, row_num, col_num, &cell_index);
962 offset = read_u32(cell_index);
963 obj->size = read_u32(cell_index + 4) - 1;
964 header = col->data_map.data + offset;
965 obj->data = header + 1;
966 if (read_u8(header) == 0xff) {
967 PARA_ERROR_LOG("col %u, size %zu, offset %u\n", col_num,
969 return -E_INVALID_OBJECT;
974 static int search_rbtree(const struct osl_object *obj,
975 const struct osl_table *t, unsigned col_num,
976 struct rb_node **result, struct rb_node ***rb_link)
978 struct osl_column *col = &t->columns[col_num];
979 struct rb_node **new = &col->rbtree.rb_node, *parent = NULL;
980 const struct osl_column_description *cd =
981 get_column_description(t->desc, col_num);
982 enum osl_storage_type st = cd->storage_type;
984 struct osl_row *this_row = get_row_pointer(*new,
987 struct osl_object this_obj;
989 if (st == OSL_MAPPED_STORAGE) {
990 ret = get_mapped_object(t, col_num, this_row->num,
995 this_obj = this_row->volatile_objects[col->volatile_num];
996 ret = cd->compare_function(obj, &this_obj);
999 *result = get_rb_node_pointer(this_row,
1004 new = &((*new)->rb_left);
1006 new = &((*new)->rb_right);
1012 return -E_RB_KEY_NOT_FOUND;
1015 static int insert_rbtree(struct osl_table *t, unsigned col_num,
1016 const struct osl_row *row, const struct osl_object *obj)
1018 struct rb_node *parent, **rb_link;
1019 unsigned rbtree_num;
1021 int ret = search_rbtree(obj, t, col_num, &parent, &rb_link);
1024 return -E_RB_KEY_EXISTS;
1025 rbtree_num = t->columns[col_num].rbtree_num;
1026 n = get_rb_node_pointer(row, rbtree_num);
1027 rb_link_node(n, parent, rb_link);
1028 rb_insert_color(n, &t->columns[col_num].rbtree);
1032 static void remove_rb_node(struct osl_table *t, unsigned col_num,
1033 const struct osl_row *row)
1035 struct osl_column *col = &t->columns[col_num];
1036 const struct osl_column_description *cd =
1037 get_column_description(t->desc, col_num);
1038 enum osl_storage_flags sf = cd->storage_flags;
1039 struct rb_node *victim, *splice_out_node, *tmp;
1040 if (!(sf & OSL_RBTREE))
1043 * Which node is removed/spliced out actually depends on how many
1044 * children the victim node has: If it has no children, it gets
1045 * deleted. If it has one child, it gets spliced out. If it has two
1046 * children, its successor (which has at most a right child) gets
1049 victim = get_rb_node_pointer(row, col->rbtree_num);
1050 if (victim->rb_left && victim->rb_right)
1051 splice_out_node = rb_next(victim);
1053 splice_out_node = victim;
1054 /* Go up to the root and decrement the size of each node in the path. */
1055 for (tmp = splice_out_node; tmp; tmp = rb_parent(tmp))
1057 rb_erase(victim, &col->rbtree);
1060 static int add_row_to_rbtrees(struct osl_table *t, uint32_t row_num,
1061 struct osl_object *volatile_objs, struct osl_row **row_ptr)
1065 struct osl_row *row = allocate_row(t->num_rbtrees);
1066 const struct osl_column_description *cd;
1069 row->volatile_objects = volatile_objs;
1070 FOR_EACH_RBTREE_COLUMN(i, t, cd) {
1071 if (cd->storage_type == OSL_MAPPED_STORAGE) {
1072 struct osl_object obj;
1073 ret = get_mapped_object(t, i, row_num, &obj);
1076 ret = insert_rbtree(t, i, row, &obj);
1077 } else { /* volatile */
1078 const struct osl_object *obj
1079 = volatile_objs + t->columns[i].volatile_num;
1080 ret = insert_rbtree(t, i, row, obj);
1088 err: /* rollback changes, i.e. remove added entries from rbtrees */
1090 remove_rb_node(t, i--, row);
1095 static void free_volatile_objects(const struct osl_table *t,
1096 enum osl_close_flags flags)
1100 struct osl_column *rb_col;
1101 const struct osl_column_description *cd;
1103 if (!t->num_volatile_columns)
1105 /* find the first rbtree column (any will do) */
1106 FOR_EACH_RBTREE_COLUMN(i, t, cd)
1108 rb_col = t->columns + i;
1109 /* walk that rbtree and free all volatile objects */
1110 for (n = rb_first(&rb_col->rbtree); n; n = rb_next(n)) {
1111 struct osl_row *r = get_row_pointer(n, rb_col->rbtree_num);
1112 if (flags & OSL_FREE_VOLATILE)
1113 for (j = 0; j < t->num_volatile_columns; j++)
1114 free(r->volatile_objects[j].data);
1115 free(r->volatile_objects);
1120 * Erase all rbtree nodes and free resources.
1122 * \param t Pointer to an open osl table.
1124 * This function is called by osl_close_table().
1126 void clear_rbtrees(struct osl_table *t)
1128 const struct osl_column_description *cd;
1129 unsigned i, rbtrees_cleared = 0;
1131 FOR_EACH_RBTREE_COLUMN(i, t, cd) {
1132 struct osl_column *col = &t->columns[i];
1135 for (n = rb_first(&col->rbtree); n;) {
1137 rb_erase(n, &col->rbtree);
1138 if (rbtrees_cleared == t->num_rbtrees) {
1139 r = get_row_pointer(n, col->rbtree_num);
1150 * Close an osl table.
1152 * \param t Pointer to the table to be closed.
1153 * \param flags Options for what should be cleaned up.
1155 * If osl_open_table() succeeds, the resulting table pointer must later be
1156 * passed to this function in order to flush all changes to the filesystem and
1157 * to free the resources that were allocated by osl_open_table().
1159 * \return Positive on success, negative on errors. Possible errors: \p E_BAD_TABLE,
1160 * errors returned by unmap_table().
1162 * \sa osl_open_table(), unmap_table().
1164 int osl_close_table(struct osl_table *t, enum osl_close_flags flags)
1169 return -E_BAD_TABLE;
1170 free_volatile_objects(t, flags);
1172 ret = unmap_table(t, flags);
1174 PARA_ERROR_LOG("unmap_table failed: %d\n", ret);
1181 * Find out whether the given row number corresponds to an invalid row.
1183 * \param t Pointer to the osl table.
1184 * \param row_num The number of the row in question.
1186 * By definition, a row is considered invalid if all its index entries
1189 * \return Positive if \a row_num corresponds to an invalid row,
1190 * zero if it corresponds to a valid row, negative on errors.
1192 int row_is_invalid(struct osl_table *t, uint32_t row_num)
1195 int i, ret = get_row_index(t, row_num, &row_index);
1199 for (i = 0; i < t->row_index_size; i++) {
1200 if ((unsigned char)row_index[i] != 0xff)
1203 PARA_INFO_LOG("row %d is invalid\n", row_num);
1208 * Invalidate a row of an osl table.
1210 * \param t Pointer to an open osl table.
1211 * \param row_num Number of the row to mark as invalid.
1213 * This function marks each mapped object in the index entry of \a row as
1216 * \return Positive on success, negative on errors.
1218 int mark_row_invalid(struct osl_table *t, uint32_t row_num)
1221 int ret = get_row_index(t, row_num, &row_index);
1225 PARA_INFO_LOG("marking row %d as invalid\n", row_num);
1226 memset(row_index, 0xff, t->row_index_size);
1231 * Initialize all rbtrees and compute number of invalid rows.
1233 * \param t The table containing the rbtrees to be initialized.
1235 * \return Positive on success, negative on errors.
1237 int init_rbtrees(struct osl_table *t)
1240 const struct osl_column_description *cd;
1242 /* create rbtrees */
1243 FOR_EACH_RBTREE_COLUMN(i, t, cd)
1244 t->columns[i].rbtree = RB_ROOT;
1245 /* add valid rows to rbtrees */
1246 t->num_invalid_rows = 0;
1247 for (i = 0; i < t->num_rows; i++) {
1248 ret = row_is_invalid(t, i);
1252 t->num_invalid_rows++;
1255 ret = add_row_to_rbtrees(t, i, NULL, NULL);
1263 * Open an osl table.
1265 * Each osl table must be opened before its data can be accessed.
1267 * \param table_desc Describes the table to be opened.
1268 * \param result Contains a pointer to the open table on success.
1270 * The table description given by \a desc should coincide with the
1271 * description used at creation time.
1273 * \return Positive on success, negative on errors. Possible errors include:
1274 * errors returned by init_table_structure(), \p E_NOENT, \p E_STAT, \p \p
1275 * E_NOTDIR, \p E_BAD_TABLE_DESC, \p E_BAD_DB_DIR, \p E_NO_COMPARE_FUNC, \p
1276 * E_NO_COLUMN_NAME, errors returned by init_rbtrees().
1278 int osl_open_table(const struct osl_table_description *table_desc,
1279 struct osl_table **result)
1282 struct osl_table *t;
1283 const struct osl_column_description *cd;
1285 PARA_INFO_LOG("opening table %s\n", table_desc->name);
1286 ret = init_table_structure(table_desc, &t);
1289 FOR_EACH_DISK_STORAGE_COLUMN(i, t, cd) {
1290 /* check if directory exists */
1291 char *dirname = column_filename(t, i);
1292 struct stat statbuf;
1293 ret = stat(dirname, &statbuf);
1296 if (errno == ENOENT)
1303 if (!S_ISDIR(statbuf.st_mode))
1306 ret = map_table(t, MAP_TBL_FL_VERIFY_INDEX);
1310 PARA_DEBUG_LOG("num rows: %d\n", t->num_rows);
1311 ret = init_rbtrees(t);
1313 osl_close_table(t, OSL_MARK_CLEAN); /* ignore further errors */
1324 static int create_disk_storage_object_dir(const struct osl_table *t,
1325 unsigned col_num, const char *ds_name)
1330 if (!(t->desc->flags & OSL_LARGE_TABLE))
1332 dirname = disk_storage_dirname(t, col_num, ds_name);
1333 ret = para_mkdir(dirname, 0777);
1335 if (ret < 0 && ret != -E_EXIST)
1340 static int write_disk_storage_file(const struct osl_table *t, unsigned col_num,
1341 const struct osl_object *obj, const char *ds_name)
1346 ret = create_disk_storage_object_dir(t, col_num, ds_name);
1349 filename = disk_storage_path(t, col_num, ds_name);
1350 ret = para_write_file(filename, obj->data, obj->size);
1355 static int append_map_file(const struct osl_table *t, unsigned col_num,
1356 const struct osl_object *obj, uint32_t *new_size)
1358 char *filename = column_filename(t, col_num);
1360 char header = 0; /* zero means valid object */
1362 // PARA_DEBUG_LOG("appending %zu + 1 byte\n", obj->size);
1363 ret = append_file(filename, &header, 1, obj->data, obj->size,
1369 static int append_row_index(const struct osl_table *t, char *row_index)
1374 if (!t->num_mapped_columns)
1376 filename = index_filename(t->desc);
1377 ret = append_file(filename, NULL, 0, row_index,
1378 t->row_index_size, NULL);
1384 * A wrapper for truncate(2)
1386 * \param path Name of the regular file to truncate
1387 * \param size Number of bytes to \b shave \b off
1389 * Truncate the regular file named by \a path by \a size bytes.
1391 * \return Positive on success, negative on errors. Possible errors include: \p
1392 * E_STAT, \p E_BAD_SIZE, \p E_TRUNC.
1396 int para_truncate(const char *path, off_t size)
1399 struct stat statbuf;
1402 if (stat(path, &statbuf) < 0)
1405 if (statbuf.st_size < size)
1408 if (truncate(path, statbuf.st_size - size) < 0)
1415 static int truncate_mapped_file(const struct osl_table *t, unsigned col_num,
1418 char *filename = column_filename(t, col_num);
1419 int ret = para_truncate(filename, size);
1424 static int delete_disk_storage_file(const struct osl_table *t, unsigned col_num,
1425 const char *ds_name)
1427 char *dirname, *filename = disk_storage_path(t, col_num, ds_name);
1428 int ret = unlink(filename);
1430 PARA_DEBUG_LOG("deleted %s\n", filename);
1433 if (errno == ENOENT)
1437 if (!(t->desc->flags & OSL_LARGE_TABLE))
1439 dirname = disk_storage_dirname(t, col_num, ds_name);
1446 * Add a new row to an osl table and retrieve this row.
1448 * \param t Pointer to an open osl table.
1449 * \param objects Array of objects to be added.
1450 * \param row Result pointer.
1452 * The \a objects parameter must point to an array containing one object per
1453 * column. The order of the objects in the array is given by the table
1454 * description of \a table. Several sanity checks are performed during object
1455 * insertion and the function returns without modifying the table if any of
1456 * these tests fail. In fact, it is atomic in the sense that it either
1457 * succeeds or leaves the table unchanged (i.e. either all or none of the
1458 * objects are added to the table).
1460 * It is considered an error if an object is added to a column with associated
1461 * rbtree if this object is equal to an object already contained in that column
1462 * (i.e. the compare function for the column's rbtree returns zero).
1464 * Possible errors include: \p E_RB_KEY_EXISTS, \p E_BAD_DATA_SIZE.
1466 * \return Positive on success, negative on errors.
1468 * \sa struct osl_table_description, osl_compare_func, osl_add_row().
1470 int osl_add_and_get_row(struct osl_table *t, struct osl_object *objects,
1471 struct osl_row **row)
1474 char *ds_name = NULL;
1475 struct rb_node **rb_parents = NULL, ***rb_links = NULL;
1476 char *new_row_index = NULL;
1477 struct osl_object *volatile_objs = NULL;
1478 const struct osl_column_description *cd;
1481 return -E_BAD_TABLE;
1482 rb_parents = para_malloc(t->num_rbtrees * sizeof(struct rn_node*));
1483 rb_links = para_malloc(t->num_rbtrees * sizeof(struct rn_node**));
1484 if (t->num_mapped_columns)
1485 new_row_index = para_malloc(t->row_index_size);
1486 /* pass 1: sanity checks */
1487 // PARA_DEBUG_LOG("sanity tests: %p:%p\n", objects[0].data,
1488 // objects[1].data);
1489 FOR_EACH_COLUMN(i, t->desc, cd) {
1490 enum osl_storage_type st = cd->storage_type;
1491 enum osl_storage_flags sf = cd->storage_flags;
1493 // ret = -E_NULL_OBJECT;
1496 if (st == OSL_DISK_STORAGE)
1498 if (sf & OSL_RBTREE) {
1499 unsigned rbtree_num = t->columns[i].rbtree_num;
1500 ret = -E_RB_KEY_EXISTS;
1501 // PARA_DEBUG_LOG("checking whether %p exists\n",
1502 // objects[i].data);
1503 if (search_rbtree(objects + i, t, i,
1504 &rb_parents[rbtree_num],
1505 &rb_links[rbtree_num]) > 0)
1508 if (sf & OSL_FIXED_SIZE) {
1509 // PARA_DEBUG_LOG("fixed size. need: %zu, have: %d\n",
1510 // objects[i].size, cd->data_size);
1511 ret = -E_BAD_DATA_SIZE;
1512 if (objects[i].size != cd->data_size)
1516 if (t->num_disk_storage_columns)
1517 ds_name = disk_storage_name_of_object(t,
1518 &objects[t->disk_storage_name_column]);
1519 ret = unmap_table(t, OSL_MARK_CLEAN);
1522 // PARA_DEBUG_LOG("sanity tests passed%s\n", "");
1523 /* pass 2: create data files, append map data */
1524 FOR_EACH_COLUMN(i, t->desc, cd) {
1525 enum osl_storage_type st = cd->storage_type;
1526 if (st == OSL_NO_STORAGE)
1528 if (st == OSL_MAPPED_STORAGE) {
1530 struct osl_column *col = &t->columns[i];
1531 // PARA_DEBUG_LOG("appending object of size %zu\n",
1532 // objects[i].size);
1533 ret = append_map_file(t, i, objects + i, &new_size);
1536 update_cell_index(new_row_index, col, new_size,
1541 ret = write_disk_storage_file(t, i, objects + i, ds_name);
1545 ret = append_row_index(t, new_row_index);
1548 ret = map_table(t, MAP_TBL_FL_VERIFY_INDEX);
1549 if (ret < 0) { /* truncate index and rollback changes */
1550 char *filename = index_filename(t->desc);
1551 para_truncate(filename, t->row_index_size);
1555 /* pass 3: add entry to rbtrees */
1556 if (t->num_volatile_columns) {
1557 volatile_objs = para_calloc(t->num_volatile_columns
1558 * sizeof(struct osl_object));
1559 FOR_EACH_VOLATILE_COLUMN(i, t, cd)
1560 volatile_objs[t->columns[i].volatile_num] = objects[i];
1563 // PARA_DEBUG_LOG("adding new entry as row #%d\n", t->num_rows - 1);
1564 ret = add_row_to_rbtrees(t, t->num_rows - 1, volatile_objs, row);
1567 // PARA_DEBUG_LOG("added new entry as row #%d\n", t->num_rows - 1);
1570 rollback: /* rollback all changes made, ignore further errors */
1571 for (i--; i >= 0; i--) {
1572 cd = get_column_description(t->desc, i);
1573 enum osl_storage_type st = cd->storage_type;
1574 if (st == OSL_NO_STORAGE)
1577 if (st == OSL_MAPPED_STORAGE)
1578 truncate_mapped_file(t, i, objects[i].size);
1579 else /* disk storage */
1580 delete_disk_storage_file(t, i, ds_name);
1582 /* ignore error and return previous error value */
1583 map_table(t, MAP_TBL_FL_VERIFY_INDEX);
1585 free(new_row_index);
1593 * Add a new row to an osl table.
1595 * \param t Same meaning as osl_add_and_get_row().
1596 * \param objects Same meaning as osl_add_and_get_row().
1598 * \return The return value of the underlying call to osl_add_and_get_row().
1600 * This is equivalent to osl_add_and_get_row(t, objects, NULL).
1602 int osl_add_row(struct osl_table *t, struct osl_object *objects)
1604 return osl_add_and_get_row(t, objects, NULL);
1608 * Retrieve an object identified by row and column
1610 * \param t Pointer to an open osl table.
1611 * \param r Pointer to the row.
1612 * \param col_num The column number.
1613 * \param object The result pointer.
1615 * The column determined by \a col_num must be of type \p OSL_MAPPED_STORAGE
1616 * or \p OSL_NO_STORAGE, i.e. no disk storage objects may be retrieved by this
1619 * \return Positive if object was found, negative on errors. Possible errors
1620 * include: \p E_BAD_TABLE, \p E_BAD_STORAGE_TYPE.
1622 * \sa osl_storage_type, osl_open_disk_object().
1624 int osl_get_object(const struct osl_table *t, const struct osl_row *r,
1625 unsigned col_num, struct osl_object *object)
1627 const struct osl_column_description *cd;
1630 return -E_BAD_TABLE;
1631 cd = get_column_description(t->desc, col_num);
1632 /* col must not be disk storage */
1633 if (cd->storage_type == OSL_DISK_STORAGE)
1634 return -E_BAD_STORAGE_TYPE;
1635 if (cd->storage_type == OSL_MAPPED_STORAGE)
1636 return get_mapped_object(t, col_num, r->num, object);
1638 *object = r->volatile_objects[t->columns[col_num].volatile_num];
1642 static int mark_mapped_object_invalid(const struct osl_table *t,
1643 uint32_t row_num, unsigned col_num)
1645 struct osl_object obj;
1647 int ret = get_mapped_object(t, col_num, row_num, &obj);
1658 * Delete a row from an osl table.
1660 * \param t Pointer to an open osl table.
1661 * \param row Pointer to the row to delete.
1663 * This removes all disk storage objects, removes all rbtree nodes, and frees
1664 * all volatile objects belonging to the given row. For mapped columns, the
1665 * data is merely marked invalid and may be pruned from time to time by
1668 * \return Positive on success, negative on errors. Possible errors include:
1669 * \p E_BAD_TABLE, errors returned by osl_get_object().
1671 int osl_del_row(struct osl_table *t, struct osl_row *row)
1673 struct osl_row *r = row;
1675 const struct osl_column_description *cd;
1678 return -E_BAD_TABLE;
1679 PARA_INFO_LOG("deleting row %p\n", row);
1681 if (t->num_disk_storage_columns) {
1683 ret = disk_storage_name_of_row(t, r, &ds_name);
1686 FOR_EACH_DISK_STORAGE_COLUMN(i, t, cd)
1687 delete_disk_storage_file(t, i, ds_name);
1690 FOR_EACH_COLUMN(i, t->desc, cd) {
1691 struct osl_column *col = t->columns + i;
1692 enum osl_storage_type st = cd->storage_type;
1693 remove_rb_node(t, i, r);
1694 if (st == OSL_MAPPED_STORAGE) {
1695 mark_mapped_object_invalid(t, r->num, i);
1698 if (st == OSL_NO_STORAGE)
1699 free(r->volatile_objects[col->volatile_num].data);
1701 if (t->num_mapped_columns) {
1702 ret = mark_row_invalid(t, r->num);
1705 t->num_invalid_rows++;
1710 free(r->volatile_objects);
1715 /* test if column has an rbtree */
1716 static int check_rbtree_col(const struct osl_table *t, unsigned col_num,
1717 struct osl_column **col)
1720 return -E_BAD_TABLE;
1721 if (!(get_column_description(t->desc, col_num)->storage_flags & OSL_RBTREE))
1722 return -E_BAD_STORAGE_FLAGS;
1723 *col = t->columns + col_num;
1728 * Get the row that contains the given object.
1730 * \param t Pointer to an open osl table.
1731 * \param col_num The number of the column to be searched.
1732 * \param obj The object to be looked up.
1733 * \param result Points to the row containing \a obj.
1735 * Lookup \a obj in \a t and return the row containing \a obj. The column
1736 * specified by \a col_num must have an associated rbtree.
1738 * \return Positive on success, negative on errors. If an error occured, \a
1739 * result is set to \p NULL. Possible errors include: \p E_BAD_TABLE, \p
1740 * E_BAD_STORAGE_FLAGS, errors returned by get_mapped_object(), \p
1741 * E_RB_KEY_NOT_FOUND.
1743 * \sa osl_storage_flags
1745 int osl_get_row(const struct osl_table *t, unsigned col_num,
1746 const struct osl_object *obj, struct osl_row **result)
1749 struct rb_node *node;
1750 struct osl_row *row;
1751 struct osl_column *col;
1754 ret = check_rbtree_col(t, col_num, &col);
1757 ret = search_rbtree(obj, t, col_num, &node, NULL);
1760 row = get_row_pointer(node, t->columns[col_num].rbtree_num);
1765 static int rbtree_loop(struct osl_column *col, void *private_data,
1766 osl_rbtree_loop_func *func)
1770 for (n = rb_first(&col->rbtree); n; n = rb_next(n)) {
1771 struct osl_row *r = get_row_pointer(n, col->rbtree_num);
1772 int ret = func(r, private_data);
1779 static int rbtree_loop_reverse(struct osl_column *col, void *private_data,
1780 osl_rbtree_loop_func *func)
1784 for (n = rb_last(&col->rbtree); n; n = rb_prev(n)) {
1785 struct osl_row *r = get_row_pointer(n, col->rbtree_num);
1786 int ret = func(r, private_data);
1794 * Loop over all nodes in an rbtree.
1796 * \param t Pointer to an open osl table.
1797 * \param col_num The column to use for iterating over the elements.
1798 * \param private_data Pointer that gets passed to \a func.
1799 * \param func The function to be called for each node in the rbtree.
1801 * This function does an in-order walk of the rbtree associated with \a
1802 * col_num. It is an error if the \p OSL_RBTREE flag is not set for this
1803 * column. For each node in the rbtree, the given function \a func is called
1804 * with two \p void* pointers as arguments: The first argument points to the
1805 * row that contains the object corresponding to the rbtree node currently
1806 * traversed, and the \a private_data pointer is passed to \a func as the
1807 * second argument. The loop terminates either if \a func returns a negative
1808 * value, or if all nodes of the tree have been visited.
1811 * \return Positive on success, negative on errors. If the termination of the
1812 * loop was caused by \a func returning a negative value, this value is
1815 * \sa osl_storage_flags, osl_rbtree_loop_reverse(), osl_compare_func.
1817 int osl_rbtree_loop(const struct osl_table *t, unsigned col_num,
1818 void *private_data, osl_rbtree_loop_func *func)
1820 struct osl_column *col;
1822 int ret = check_rbtree_col(t, col_num, &col);
1825 return rbtree_loop(col, private_data, func);
1829 * Loop over all nodes in an rbtree in reverse order.
1831 * \param t Identical meaning as in \p osl_rbtree_loop().
1832 * \param col_num Identical meaning as in \p osl_rbtree_loop().
1833 * \param private_data Identical meaning as in \p osl_rbtree_loop().
1834 * \param func Identical meaning as in \p osl_rbtree_loop().
1836 * This function is identical to \p osl_rbtree_loop(), the only difference
1837 * is that the tree is walked in reverse order.
1839 * \return The same return value as \p osl_rbtree_loop().
1841 * \sa osl_rbtree_loop().
1843 int osl_rbtree_loop_reverse(const struct osl_table *t, unsigned col_num,
1844 void *private_data, osl_rbtree_loop_func *func)
1846 struct osl_column *col;
1848 int ret = check_rbtree_col(t, col_num, &col);
1851 return rbtree_loop_reverse(col, private_data, func);
1854 /* TODO: Rollback changes on errors */
1855 static int rename_disk_storage_objects(struct osl_table *t,
1856 struct osl_object *old_obj, struct osl_object *new_obj)
1859 const struct osl_column_description *cd;
1860 char *old_ds_name, *new_ds_name;
1862 if (!t->num_disk_storage_columns)
1863 return 1; /* nothing to do */
1864 if (old_obj->size == new_obj->size && !memcmp(new_obj->data,
1865 old_obj->data, new_obj->size))
1866 return 1; /* object did not change */
1867 old_ds_name = disk_storage_name_of_object(t, old_obj);
1868 new_ds_name = disk_storage_name_of_object(t, new_obj);
1869 FOR_EACH_DISK_STORAGE_COLUMN(i, t, cd) {
1870 char *old_filename, *new_filename;
1871 ret = create_disk_storage_object_dir(t, i, new_ds_name);
1874 old_filename = disk_storage_path(t, i, old_ds_name);
1875 new_filename = disk_storage_path(t, i, new_ds_name);
1876 ret = para_rename(old_filename, new_filename);
1891 * Change an object in an osl table.
1893 * \param t Pointer to an open osl table.
1894 * \param r Pointer to the row containing the object to be updated.
1895 * \param col_num Number of the column containing the object to be updated.
1896 * \param obj Pointer to the replacement object.
1898 * This function gets rid of all references to the old object. This includes
1899 * removal of the rbtree node in case there is an rbtree associated with \a
1900 * col_num. It then inserts \a obj into the table and the rbtree if neccessary.
1902 * If the \p OSL_RBTREE flag is set for \a col_num, you \b MUST call this
1903 * function in order to change the contents of an object, even for volatile or
1904 * mapped columns of constant size (which may be updated directly if \p
1905 * OSL_RBTREE is not set). Otherwise the rbtree might become corrupted.
1907 * \return Positive on success, negative on errors. Possible errors include: \p
1908 * E_BAD_TABLE, \p E_RB_KEY_EXISTS, \p E_BAD_SIZE, \p E_NOENT, \p E_UNLINK,
1909 * errors returned by para_write_file(), \p E_MKDIR.
1911 int osl_update_object(struct osl_table *t, const struct osl_row *r,
1912 unsigned col_num, struct osl_object *obj)
1914 struct osl_column *col;
1915 const struct osl_column_description *cd;
1919 return -E_BAD_TABLE;
1920 col = &t->columns[col_num];
1921 cd = get_column_description(t->desc, col_num);
1922 PARA_DEBUG_LOG("updating column %u of %s\n", col_num, t->desc->name);
1923 if (cd->storage_flags & OSL_RBTREE) {
1924 if (search_rbtree(obj, t, col_num, NULL, NULL) > 0)
1925 return -E_RB_KEY_EXISTS;
1927 if (cd->storage_flags & OSL_FIXED_SIZE) {
1928 if (obj->size != cd->data_size)
1929 return -E_BAD_DATA_SIZE;
1931 remove_rb_node(t, col_num, r);
1932 if (cd->storage_type == OSL_NO_STORAGE) { /* TODO: If fixed size, reuse object? */
1933 free(r->volatile_objects[col->volatile_num].data);
1934 r->volatile_objects[col->volatile_num] = *obj;
1935 } else if (cd->storage_type == OSL_DISK_STORAGE) {
1937 ret = disk_storage_name_of_row(t, r, &ds_name);
1940 ret = delete_disk_storage_file(t, col_num, ds_name);
1941 if (ret < 0 && ret != -E_NOENT) {
1945 ret = write_disk_storage_file(t, col_num, obj, ds_name);
1949 } else { /* mapped storage */
1950 struct osl_object old_obj;
1951 ret = get_mapped_object(t, col_num, r->num, &old_obj);
1955 * If the updated column is the disk storage name column, the
1956 * disk storage name changes, so we have to rename all disk
1957 * storage objects accordingly.
1959 if (col_num == t->disk_storage_name_column) {
1960 ret = rename_disk_storage_objects(t, &old_obj, obj);
1964 if (cd->storage_flags & OSL_FIXED_SIZE)
1965 memcpy(old_obj.data, obj->data, cd->data_size);
1966 else { /* TODO: if the size doesn't change, use old space */
1967 uint32_t new_data_map_size;
1969 ret = get_row_index(t, r->num, &row_index);
1972 ret = mark_mapped_object_invalid(t, r->num, col_num);
1975 unmap_column(t, col_num);
1976 ret = append_map_file(t, col_num, obj,
1977 &new_data_map_size);
1980 ret = map_column(t, col_num);
1983 update_cell_index(row_index, col, new_data_map_size,
1987 if (cd->storage_flags & OSL_RBTREE) {
1988 ret = insert_rbtree(t, col_num, r, obj);
1996 * Retrieve an object of type \p OSL_DISK_STORAGE by row and column.
1998 * \param t Pointer to an open osl table.
1999 * \param r Pointer to the row containing the object.
2000 * \param col_num The column number.
2001 * \param obj Points to the result upon successful return.
2003 * For columns of type \p OSL_DISK_STORAGE, this function must be used to
2004 * retrieve one of its containing objects. Afterwards, osl_close_disk_object()
2005 * must be called in order to deallocate the resources.
2007 * \return Positive on success, negative on errors. Possible errors include:
2008 * \p E_BAD_TABLE, \p E_BAD_STORAGE_TYPE, errors returned by osl_get_object().
2010 * \sa osl_get_object(), osl_storage_type, osl_close_disk_object().
2012 int osl_open_disk_object(const struct osl_table *t, const struct osl_row *r,
2013 unsigned col_num, struct osl_object *obj)
2015 const struct osl_column_description *cd;
2016 char *ds_name, *filename;
2020 return -E_BAD_TABLE;
2021 cd = get_column_description(t->desc, col_num);
2022 if (cd->storage_type != OSL_DISK_STORAGE)
2023 return -E_BAD_STORAGE_TYPE;
2025 ret = disk_storage_name_of_row(t, r, &ds_name);
2028 filename = disk_storage_path(t, col_num, ds_name);
2030 PARA_DEBUG_LOG("filename: %s\n", filename);
2031 ret = mmap_full_file(filename, O_RDONLY, obj);
2037 * Free resources that were allocated during osl_open_disk_object().
2039 * \param obj Pointer to the object previously returned by open_disk_object().
2041 * \return The return value of the underlying call to para_munmap().
2043 * \sa para_munmap().
2045 int osl_close_disk_object(struct osl_object *obj)
2047 return para_munmap(obj->data, obj->size);
2051 * Get the number of rows of the given table.
2053 * \param t Pointer to an open osl table.
2054 * \param num_rows Result is returned here.
2056 * The number of rows returned via \a num_rows excluding any invalid rows.
2058 * \return Positive on success, \p -E_BAD_TABLE if \a t is \p NULL.
2060 int osl_get_num_rows(const struct osl_table *t, unsigned *num_rows)
2063 return -E_BAD_TABLE;
2064 assert(t->num_rows >= t->num_invalid_rows);
2065 *num_rows = t->num_rows - t->num_invalid_rows;
2070 * Get the rank of a row.
2072 * \param t An open osl table.
2073 * \param r The row to get the rank of.
2074 * \param col_num The number of an rbtree column.
2075 * \param rank Result pointer.
2077 * The rank is, by definition, the position of the row in the linear order
2078 * determined by an inorder tree walk of the rbtree associated with column
2079 * number \a col_num of \a table.
2081 * \return Positive on success, negative on errors.
2083 * \sa osl_get_nth_row().
2085 int osl_get_rank(const struct osl_table *t, struct osl_row *r,
2086 unsigned col_num, unsigned *rank)
2088 struct osl_object obj;
2089 struct osl_column *col;
2090 struct rb_node *node;
2091 int ret = check_rbtree_col(t, col_num, &col);
2095 ret = osl_get_object(t, r, col_num, &obj);
2098 ret = search_rbtree(&obj, t, col_num, &node, NULL);
2101 ret = rb_rank(node, rank);
2108 * Get the row with n-th greatest value.
2110 * \param t Pointer to an open osl table.
2111 * \param col_num The column number.
2112 * \param n The rank of the desired row.
2113 * \param result Row is returned here.
2115 * Retrieve the n-th order statistic with respect to the compare function
2116 * of the rbtree column \a col_num. In other words, get that row with
2117 * \a n th greatest value in column \a col_num. It's an error if
2118 * \a col_num is not a rbtree column, or if \a n is larger than the
2119 * number of rows in the table.
2121 * \return Positive on success, negative on errors. Possible errors:
2122 * \p E_BAD_TABLE, \p E_BAD_STORAGE_FLAGS, \p E_RB_KEY_NOT_FOUND.
2124 * \sa osl_storage_flags, osl_compare_func, osl_get_row(),
2125 * osl_rbtree_last_row(), osl_rbtree_first_row(), osl_get_rank().
2127 int osl_get_nth_row(const struct osl_table *t, unsigned col_num,
2128 unsigned n, struct osl_row **result)
2130 struct osl_column *col;
2131 struct rb_node *node;
2132 int ret = check_rbtree_col(t, col_num, &col);
2136 node = rb_nth(col->rbtree.rb_node, n);
2138 return -E_RB_KEY_NOT_FOUND;
2139 *result = get_row_pointer(node, col->rbtree_num);
2144 * Get the row corresponding to the smallest rbtree node of a column.
2146 * \param t An open rbtree table.
2147 * \param col_num The number of the rbtree column.
2148 * \param result A pointer to the first row is returned here.
2150 * The rbtree node of the smallest object (with respect to the corresponding
2151 * compare function) is selected and the row containing this object is
2152 * returned. It is an error if \a col_num refers to a column without an
2153 * associated rbtree.
2155 * \return Positive on success, negative on errors.
2157 * \sa osl_get_nth_row(), osl_rbtree_last_row().
2159 int osl_rbtree_first_row(const struct osl_table *t, unsigned col_num,
2160 struct osl_row **result)
2162 return osl_get_nth_row(t, col_num, 1, result);
2166 * Get the row corresponding to the greatest rbtree node of a column.
2168 * \param t The same meaning as in \p osl_rbtree_first_row().
2169 * \param col_num The same meaning as in \p osl_rbtree_first_row().
2170 * \param result The same meaning as in \p osl_rbtree_first_row().
2172 * This function works just like osl_rbtree_first_row(), the only difference
2173 * is that the row containing the greatest rather than the smallest object is
2176 * \return Positive on success, negative on errors.
2178 * \sa osl_get_nth_row(), osl_rbtree_first_row().
2180 int osl_rbtree_last_row(const struct osl_table *t, unsigned col_num,
2181 struct osl_row **result)
2184 int ret = osl_get_num_rows(t, &num_rows);
2188 return osl_get_nth_row(t, col_num, num_rows, result);