2 * Copyright (C) 2007-2008 Andre Noll <maan@systemlinux.org>
4 * Licensed under the GPL v2. For licencing details see COPYING.
7 /** \file osl.c Object storage layer functions. */
8 #include <dirent.h> /* readdir() */
18 * A wrapper for lseek(2).
20 * \param fd The file descriptor whose offset is to be to repositioned.
21 * \param offset A value-result parameter.
22 * \param whence Usual repositioning directive.
24 * Reposition the offset of the file descriptor \a fd to the argument \a offset
25 * according to the directive \a whence. Upon successful return, \a offset
26 * contains the resulting offset location as measured in bytes from the
27 * beginning of the file.
29 * \return Positive on success. Otherwise, the function returns \p -E_LSEEK.
33 int para_lseek(int fd, off_t *offset, int whence)
35 *offset = lseek(fd, *offset, whence);
43 * Wrapper for the write system call.
45 * \param fd The file descriptor to write to.
46 * \param buf The buffer to write.
47 * \param size The length of \a buf in bytes.
49 * This function writes out the given buffer and retries if an interrupt
50 * occurred during the write.
52 * \return On success, the number of bytes written is returned, otherwise, the
53 * function returns \p -E_WRITE.
57 ssize_t para_write(int fd, const void *buf, size_t size)
62 ret = write(fd, buf, size);
63 if ((ret < 0) && (errno == EAGAIN || errno == EINTR))
65 return ret >= 0? ret : -E_WRITE;
70 * Write the whole buffer to a file descriptor.
72 * \param fd The file descriptor to write to.
73 * \param buf The buffer to write.
74 * \param size The length of \a buf in bytes.
76 * This function writes the given buffer and continues on short writes and
77 * when interrupted by a signal.
79 * \return Positive on success, negative on errors. Possible errors: any
80 * errors returned by para_write().
84 ssize_t para_write_all(int fd, const void *buf, size_t size)
86 // DEBUG_LOG("writing %zu bytes\n", size);
89 ssize_t ret = para_write(fd, b, size);
90 // DEBUG_LOG("ret: %zd\n", ret);
99 * Open a file, write the given buffer and close the file.
101 * \param filename Full path to the file to open.
102 * \param buf The buffer to write to the file.
103 * \param size The size of \a buf.
105 * \return Positive on success, negative on errors. Possible errors include:
106 * any errors from para_open() or para_write().
108 * \sa para_open(), para_write().
110 int para_write_file(const char *filename, const void *buf, size_t size)
114 ret = para_open(filename, O_WRONLY | O_CREAT | O_EXCL, 0644);
118 ret = para_write_all(fd, buf, size);
127 static int append_file(const char *filename, char *header, size_t header_size,
128 char *data, size_t data_size, uint32_t *new_pos)
132 // DEBUG_LOG("appending %zu + %zu bytes\n", header_size, data_size);
133 ret = para_open(filename, O_WRONLY | O_CREAT | O_APPEND, 0644);
137 if (header && header_size) {
138 ret = para_write_all(fd, header, header_size);
142 ret = para_write_all(fd, data, data_size);
147 ret = para_lseek(fd, &offset, SEEK_END);
150 // DEBUG_LOG("new file size: " FMT_OFF_T "\n", offset);
160 * Traverse the given directory recursively.
162 * \param dirname The directory to traverse.
163 * \param func The function to call for each entry.
164 * \param private_data Pointer to an arbitrary data structure.
166 * For each regular file under \a dirname, the supplied function \a func is
167 * called. The full path of the regular file and the \a private_data pointer
168 * are passed to \a func. Directories for which the calling process has no
169 * permissions to change to are silently ignored.
173 int for_each_file_in_dir(const char *dirname,
174 int (*func)(const char *, void *), void *private_data)
177 struct dirent *entry;
178 int cwd_fd, ret2, ret = para_opendir(dirname, &dir, &cwd_fd);
181 return ret == -ERRNO_TO_ERROR(EACCES)? 1 : ret;
182 /* scan cwd recursively */
183 while ((entry = readdir(dir))) {
188 if (!strcmp(entry->d_name, "."))
190 if (!strcmp(entry->d_name, ".."))
192 if (lstat(entry->d_name, &s) == -1)
195 if (!S_ISREG(m) && !S_ISDIR(m))
197 tmp = make_message("%s/%s", dirname, entry->d_name);
199 ret = func(tmp, private_data);
206 ret = for_each_file_in_dir(tmp, func, private_data);
214 ret2 = para_fchdir(cwd_fd);
215 if (ret2 < 0 && ret >= 0)
221 static int verify_name(const char *name)
227 if (strchr(name, '/'))
229 if (!strcmp(name, ".."))
231 if (!strcmp(name, "."))
237 * Compare two osl objects pointing to unsigned integers of 32 bit size.
239 * \param obj1 Pointer to the first integer.
240 * \param obj2 Pointer to the second integer.
242 * \return The values required for an osl compare function.
244 * \sa osl_compare_func, osl_hash_compare().
246 int uint32_compare(const struct osl_object *obj1, const struct osl_object *obj2)
248 uint32_t d1 = read_u32((const char *)obj1->data);
249 uint32_t d2 = read_u32((const char *)obj2->data);
259 * Compare two osl objects pointing to hash values.
261 * \param obj1 Pointer to the first hash object.
262 * \param obj2 Pointer to the second hash object.
264 * \return The values required for an osl compare function.
266 * \sa osl_compare_func, uint32_compare().
268 int osl_hash_compare(const struct osl_object *obj1, const struct osl_object *obj2)
270 return hash_compare((HASH_TYPE *)obj1->data, (HASH_TYPE *)obj2->data);
273 static char *disk_storage_dirname(const struct osl_table *t, unsigned col_num,
276 char *dirname, *column_name = column_filename(t, col_num);
278 if (!(t->desc->flags & OSL_LARGE_TABLE))
280 dirname = make_message("%s/%.2s", column_name, ds_name);
285 static char *disk_storage_name_of_object(const struct osl_table *t,
286 const struct osl_object *obj)
288 HASH_TYPE hash[HASH_SIZE];
289 hash_object(obj, hash);
290 return disk_storage_name_of_hash(t, hash);
293 static int disk_storage_name_of_row(const struct osl_table *t,
294 const struct osl_row *row, char **name)
296 struct osl_object obj;
297 int ret = osl_get_object(t, row, t->disk_storage_name_column, &obj);
301 *name = disk_storage_name_of_object(t, &obj);
305 static void column_name_hash(const char *col_name, HASH_TYPE *hash)
307 hash_function(col_name, strlen(col_name), hash);
310 static int init_column_descriptions(struct osl_table *t)
313 const struct osl_column_description *cd;
315 ret = -E_BAD_TABLE_DESC;
316 ret = verify_name(t->desc->name);
320 if (!t->desc->dir && (t->num_disk_storage_columns || t->num_mapped_columns))
322 /* the size of the index header without column descriptions */
323 t->index_header_size = IDX_COLUMN_DESCRIPTIONS;
324 FOR_EACH_COLUMN(i, t->desc, cd) {
325 struct osl_column *col = t->columns + i;
326 if (cd->storage_flags & OSL_RBTREE) {
327 if (!cd->compare_function)
328 return -E_NO_COMPARE_FUNC;
330 if (cd->storage_type == OSL_NO_STORAGE)
332 ret = -E_NO_COLUMN_NAME;
333 if (!cd->name || !cd->name[0])
335 ret = verify_name(cd->name);
338 t->index_header_size += index_column_description_size(cd->name);
339 column_name_hash(cd->name, col->name_hash);
340 ret = -E_DUPLICATE_COL_NAME;
341 for (j = i + 1; j < t->desc->num_columns; j++) {
342 const char *name2 = get_column_description(t->desc,
344 if (cd->name && name2 && !strcmp(cd->name, name2))
354 * Initialize a struct table from given table description.
356 * \param desc The description of the osl table.
357 * \param table_ptr Result is returned here.
359 * This function performs several sanity checks on \p desc and returns if any
360 * of these tests fail. On success, a struct \p osl_table is allocated and
361 * initialized with data derived from \p desc.
363 * \return Positive on success, negative on errors. Possible errors include: \p
364 * E_BAD_TABLE_DESC, \p E_NO_COLUMN_DESC, \p E_NO_COLUMNS, \p
365 * E_BAD_STORAGE_TYPE, \p E_BAD_STORAGE_FLAGS, \p E_BAD_STORAGE_SIZE, \p
366 * E_NO_UNIQUE_RBTREE_COLUMN, \p E_NO_RBTREE_COL.
368 * \sa struct osl_table.
370 int init_table_structure(const struct osl_table_description *desc,
371 struct osl_table **table_ptr)
373 const struct osl_column_description *cd;
374 struct osl_table *t = para_calloc(sizeof(*t));
375 int i, ret = -E_BAD_TABLE_DESC, have_disk_storage_name_column = 0;
379 DEBUG_LOG("creating table structure for '%s' from table "
380 "description\n", desc->name);
381 ret = -E_NO_COLUMN_DESC;
382 if (!desc->column_descriptions)
385 if (!desc->num_columns)
387 t->columns = para_calloc(desc->num_columns * sizeof(struct osl_column));
389 FOR_EACH_COLUMN(i, t->desc, cd) {
390 enum osl_storage_type st = cd->storage_type;
391 enum osl_storage_flags sf = cd->storage_flags;
392 struct osl_column *col = &t->columns[i];
394 ret = -E_BAD_STORAGE_TYPE;
395 if (st != OSL_MAPPED_STORAGE && st != OSL_DISK_STORAGE
396 && st != OSL_NO_STORAGE)
398 ret = -E_BAD_STORAGE_FLAGS;
399 if (st == OSL_DISK_STORAGE && sf & OSL_RBTREE)
401 ret = -E_BAD_STORAGE_SIZE;
402 if (sf & OSL_FIXED_SIZE && !cd->data_size)
405 case OSL_DISK_STORAGE:
406 t->num_disk_storage_columns++;
408 case OSL_MAPPED_STORAGE:
409 t->num_mapped_columns++;
410 col->index_offset = t->row_index_size;
411 t->row_index_size += 8;
414 col->volatile_num = t->num_volatile_columns;
415 t->num_volatile_columns++;
418 if (sf & OSL_RBTREE) {
419 col->rbtree_num = t->num_rbtrees;
421 if ((sf & OSL_UNIQUE) && (st == OSL_MAPPED_STORAGE)) {
422 if (!have_disk_storage_name_column)
423 t->disk_storage_name_column = i;
424 have_disk_storage_name_column = 1;
428 ret = -E_NO_UNIQUE_RBTREE_COLUMN;
429 if (t->num_disk_storage_columns && !have_disk_storage_name_column)
431 ret = -E_NO_RBTREE_COL;
435 DEBUG_LOG("OK. Index entry size: %u\n", t->row_index_size);
436 ret = init_column_descriptions(t);
448 * Read the table description from index header.
450 * \param map The memory mapping of the index file.
451 * \param desc The values found in the index header are returned here.
453 * Read the index header, check for the paraslash magic string and the table version number.
454 * Read all information stored in the index header into \a desc.
456 * \return Positive on success, negative on errors.
458 * \sa struct osl_table_description, osl_create_table.
460 int read_table_desc(struct osl_object *map, struct osl_table_description *desc)
462 char *buf = map->data;
464 uint16_t header_size;
467 struct osl_column_description *cd;
469 if (map->size < MIN_INDEX_HEADER_SIZE(1))
470 return -E_SHORT_TABLE;
471 if (strncmp(buf + IDX_PARA_MAGIC, PARA_MAGIC, strlen(PARA_MAGIC)))
473 version = read_u8(buf + IDX_VERSION);
474 if (version < MIN_TABLE_VERSION || version > MAX_TABLE_VERSION)
475 return -E_VERSION_MISMATCH;
476 desc->flags = read_u8(buf + IDX_TABLE_FLAGS);
477 desc->num_columns = read_u16(buf + IDX_NUM_COLUMNS);
478 DEBUG_LOG("%u columns\n", desc->num_columns);
479 if (!desc->num_columns)
480 return -E_NO_COLUMNS;
481 header_size = read_u16(buf + IDX_HEADER_SIZE);
482 if (map->size < header_size)
484 desc->column_descriptions = para_calloc(desc->num_columns
485 * sizeof(struct osl_column_description));
486 offset = IDX_COLUMN_DESCRIPTIONS;
487 FOR_EACH_COLUMN(i, desc, cd) {
490 ret = -E_SHORT_TABLE;
491 if (map->size < offset + MIN_IDX_COLUMN_DESCRIPTION_SIZE) {
492 ERROR_LOG("map size = %zu < %u = offset + min desc size\n",
493 map->size, offset + MIN_IDX_COLUMN_DESCRIPTION_SIZE);
496 cd->storage_type = read_u16(buf + offset + IDX_CD_STORAGE_TYPE);
497 cd->storage_flags = read_u16(buf + offset +
498 IDX_CD_STORAGE_FLAGS);
499 cd->data_size = read_u32(buf + offset + IDX_CD_DATA_SIZE);
500 null_byte = memchr(buf + offset + IDX_CD_NAME, '\0',
501 map->size - offset - IDX_CD_NAME);
502 ret = -E_INDEX_CORRUPTION;
505 cd->name = para_strdup(buf + offset + IDX_CD_NAME);
506 offset += index_column_description_size(cd->name);
508 if (offset != header_size) {
509 ret = -E_INDEX_CORRUPTION;
510 ERROR_LOG("real header size = %u != %u = stored header size\n",
511 offset, header_size);
516 FOR_EACH_COLUMN(i, desc, cd)
522 * check whether the table description given by \p t->desc matches the on-disk
523 * table structure stored in the index of \a t.
525 static int compare_table_descriptions(struct osl_table *t)
528 struct osl_table_description desc;
529 const struct osl_column_description *cd1, *cd2;
531 /* read the on-disk structure into desc */
532 ret = read_table_desc(&t->index_map, &desc);
535 ret = -E_BAD_TABLE_FLAGS;
536 if (desc.flags != t->desc->flags)
538 ret = -E_BAD_COLUMN_NUM;
539 if (desc.num_columns > t->desc->num_columns)
541 if (desc.num_columns < t->desc->num_columns) {
542 struct osl_column_description *cd;
543 unsigned diff = t->desc->num_columns - desc.num_columns;
544 INFO_LOG("extending table by %u volatile columns\n", diff);
545 desc.column_descriptions = para_realloc(desc.column_descriptions,
546 t->desc->num_columns * sizeof(struct osl_column_description));
547 for (i = desc.num_columns; i < t->desc->num_columns; i++) {
548 cd = get_column_description(&desc, i);
549 cd->storage_type = OSL_NO_STORAGE;
552 desc.num_columns += diff;
554 FOR_EACH_COLUMN(i, t->desc, cd1) {
555 cd2 = get_column_description(&desc, i);
556 ret = -E_BAD_STORAGE_TYPE;
557 if (cd1->storage_type != cd2->storage_type)
559 if (cd1->storage_type == OSL_NO_STORAGE)
561 ret = -E_BAD_STORAGE_FLAGS;
562 if (cd1->storage_flags != cd2->storage_flags) {
563 ERROR_LOG("sf1 = %u != %u = sf2\n",
564 cd1->storage_flags, cd2->storage_flags);
567 ret = -E_BAD_DATA_SIZE;
568 if (cd1->storage_flags & OSL_FIXED_SIZE)
569 if (cd1->data_size != cd2->data_size)
571 ret = -E_BAD_COLUMN_NAME;
572 if (strcmp(cd1->name, cd2->name))
575 DEBUG_LOG("table description of '%s' matches on-disk data, good\n",
579 FOR_EACH_COLUMN(i, &desc, cd1)
581 free(desc.column_descriptions);
585 static int create_table_index(struct osl_table *t)
587 char *buf, *filename;
589 size_t size = t->index_header_size;
590 const struct osl_column_description *cd;
593 INFO_LOG("creating %zu byte index for table %s\n", size,
595 buf = para_calloc(size);
596 sprintf(buf + IDX_PARA_MAGIC, "%s", PARA_MAGIC);
597 write_u8(buf + IDX_TABLE_FLAGS, t->desc->flags);
598 write_u8(buf + IDX_DIRTY_FLAG, 0);
599 write_u8(buf + IDX_VERSION, CURRENT_TABLE_VERSION);
600 write_u16(buf + IDX_NUM_COLUMNS, t->num_mapped_columns + t->num_disk_storage_columns);
601 write_u16(buf + IDX_HEADER_SIZE, t->index_header_size);
602 offset = IDX_COLUMN_DESCRIPTIONS;
603 FOR_EACH_COLUMN(i, t->desc, cd) {
604 /* no need to store info about volatile storage */
605 if (cd->storage_type == OSL_NO_STORAGE)
607 write_u16(buf + offset + IDX_CD_STORAGE_TYPE,
609 write_u16(buf + offset + IDX_CD_STORAGE_FLAGS,
611 if (cd->storage_flags & OSL_FIXED_SIZE)
612 write_u32(buf + offset + IDX_CD_DATA_SIZE,
614 strcpy(buf + offset + IDX_CD_NAME, cd->name);
615 offset += index_column_description_size(cd->name);
617 assert(offset = size);
618 filename = index_filename(t->desc);
619 ret = para_write_file(filename, buf, size);
626 * Create a new osl table.
628 * \param desc Pointer to the table description.
632 int osl_create_table(const struct osl_table_description *desc)
634 const struct osl_column_description *cd;
635 char *table_dir = NULL, *filename;
637 int i, ret = init_table_structure(desc, &t);
641 INFO_LOG("creating %s\n", desc->name);
642 FOR_EACH_COLUMN(i, t->desc, cd) {
643 if (cd->storage_type == OSL_NO_STORAGE)
646 ret = para_mkdir(desc->dir, 0777);
647 if (ret < 0 && !is_errno(-ret, EEXIST))
649 table_dir = make_message("%s/%s", desc->dir,
651 ret = para_mkdir(table_dir, 0777);
655 filename = column_filename(t, i);
656 INFO_LOG("filename: %s\n", filename);
657 if (cd->storage_type == OSL_MAPPED_STORAGE) {
658 ret = para_open(filename, O_RDWR | O_CREAT | O_EXCL,
667 ret = para_mkdir(filename, 0777);
672 if (t->num_mapped_columns) {
673 ret = create_table_index(t);
685 static int table_is_dirty(struct osl_table *t)
687 char *buf = (char *)t->index_map.data + IDX_DIRTY_FLAG;
688 uint8_t dirty = read_u8(buf) & 0x1;
692 static void mark_table_dirty(struct osl_table *t)
694 char *buf = (char *)t->index_map.data + IDX_DIRTY_FLAG;
695 write_u8(buf, read_u8(buf) | 1);
698 static void mark_table_clean(struct osl_table *t)
700 char *buf = (char *)t->index_map.data + IDX_DIRTY_FLAG;
701 write_u8(buf, read_u8(buf) & 0xfe);
704 static void unmap_column(struct osl_table *t, unsigned col_num)
706 struct osl_object map = t->columns[col_num].data_map;
710 ret = para_munmap(map.data, map.size);
716 * Unmap all mapped files of an osl table.
718 * \param t Pointer to a mapped table.
719 * \param flags Options for unmapping.
721 * \return Positive on success, negative on errors.
723 * \sa map_table(), enum osl_close_flags, para_munmap().
725 int unmap_table(struct osl_table *t, enum osl_close_flags flags)
728 const struct osl_column_description *cd;
731 if (!t->num_mapped_columns) /* can this ever happen? */
733 DEBUG_LOG("unmapping table '%s'\n", t->desc->name);
734 if (!t->index_map.data)
735 return -E_NOT_MAPPED;
736 if (flags & OSL_MARK_CLEAN)
738 ret = para_munmap(t->index_map.data, t->index_map.size);
741 t->index_map.data = NULL;
744 FOR_EACH_MAPPED_COLUMN(i, t, cd)
749 static int map_column(struct osl_table *t, unsigned col_num)
752 char *filename = column_filename(t, col_num);
754 if (stat(filename, &statbuf) < 0) {
758 if (!(S_IFREG & statbuf.st_mode)) {
762 ret = mmap_full_file(filename, O_RDWR,
763 &t->columns[col_num].data_map.data,
764 &t->columns[col_num].data_map.size,
771 * Map the index file and all columns of type \p OSL_MAPPED_STORAGE into memory.
773 * \param t Pointer to an initialized table structure.
774 * \param flags Mapping options.
776 * \return Negative return value on errors; on success the number of rows
777 * (including invalid rows) is returned.
779 * \sa unmap_table(), enum map_table_flags, osl_open_table(), mmap(2).
781 int map_table(struct osl_table *t, enum map_table_flags flags)
784 const struct osl_column_description *cd;
785 int i = 0, ret, num_rows = 0;
787 if (!t->num_mapped_columns)
789 if (t->index_map.data)
790 return -E_ALREADY_MAPPED;
791 filename = index_filename(t->desc);
792 DEBUG_LOG("mapping table '%s' (index: %s)\n", t->desc->name, filename);
793 ret = mmap_full_file(filename, flags & MAP_TBL_FL_MAP_RDONLY?
794 O_RDONLY : O_RDWR, &t->index_map.data, &t->index_map.size, NULL);
798 if (flags & MAP_TBL_FL_VERIFY_INDEX) {
799 ret = compare_table_descriptions(t);
804 if (!(flags & MAP_TBL_FL_IGNORE_DIRTY)) {
805 if (table_is_dirty(t)) {
806 ERROR_LOG("%s is dirty\n", t->desc->name);
811 num_rows = table_num_rows(t);
815 FOR_EACH_MAPPED_COLUMN(i, t, cd) {
816 ret = map_column(t, i);
821 err: /* unmap what is already mapped */
822 for (i--; i >= 0; i--) {
823 struct osl_object map = t->columns[i].data_map;
824 para_munmap(map.data, map.size);
827 para_munmap(t->index_map.data, t->index_map.size);
828 t->index_map.data = NULL;
833 * Retrieve a mapped object by row and column number.
835 * \param t Pointer to an open osl table.
836 * \param col_num Number of the mapped column containing the object to retrieve.
837 * \param row_num Number of the row containing the object to retrieve.
838 * \param obj The result is returned here.
840 * It is considered an error if \a col_num does not refer to a column
841 * of storage type \p OSL_MAPPED_STORAGE.
843 * \return Positive on success, negative on errors. Possible errors include:
844 * \p E_BAD_ROW_NUM, \p E_INVALID_OBJECT.
846 * \sa osl_storage_type.
848 int get_mapped_object(const struct osl_table *t, unsigned col_num,
849 uint32_t row_num, struct osl_object *obj)
851 struct osl_column *col = &t->columns[col_num];
857 if (t->num_rows <= row_num)
858 return -E_BAD_ROW_NUM;
859 ret = get_cell_index(t, row_num, col_num, &cell_index);
862 offset = read_u32(cell_index);
863 obj->size = read_u32(cell_index + 4) - 1;
864 header = col->data_map.data + offset;
865 obj->data = header + 1;
866 if (read_u8(header) == 0xff) {
867 ERROR_LOG("col %u, size %zu, offset %u\n", col_num,
869 return -E_INVALID_OBJECT;
874 static int search_rbtree(const struct osl_object *obj,
875 const struct osl_table *t, unsigned col_num,
876 struct rb_node **result, struct rb_node ***rb_link)
878 struct osl_column *col = &t->columns[col_num];
879 struct rb_node **new = &col->rbtree.rb_node, *parent = NULL;
880 const struct osl_column_description *cd =
881 get_column_description(t->desc, col_num);
882 enum osl_storage_type st = cd->storage_type;
884 struct osl_row *this_row = get_row_pointer(*new,
887 struct osl_object this_obj;
889 if (st == OSL_MAPPED_STORAGE) {
890 ret = get_mapped_object(t, col_num, this_row->num,
895 this_obj = this_row->volatile_objects[col->volatile_num];
896 ret = cd->compare_function(obj, &this_obj);
899 *result = get_rb_node_pointer(this_row,
904 new = &((*new)->rb_left);
906 new = &((*new)->rb_right);
912 return -E_RB_KEY_NOT_FOUND;
915 static int insert_rbtree(struct osl_table *t, unsigned col_num,
916 const struct osl_row *row, const struct osl_object *obj)
918 struct rb_node *parent, **rb_link;
921 int ret = search_rbtree(obj, t, col_num, &parent, &rb_link);
924 return -E_RB_KEY_EXISTS;
925 rbtree_num = t->columns[col_num].rbtree_num;
926 n = get_rb_node_pointer(row, rbtree_num);
927 rb_link_node(n, parent, rb_link);
928 rb_insert_color(n, &t->columns[col_num].rbtree);
932 static void remove_rb_node(struct osl_table *t, unsigned col_num,
933 const struct osl_row *row)
935 struct osl_column *col = &t->columns[col_num];
936 const struct osl_column_description *cd =
937 get_column_description(t->desc, col_num);
938 enum osl_storage_flags sf = cd->storage_flags;
939 struct rb_node *victim, *splice_out_node, *tmp;
940 if (!(sf & OSL_RBTREE))
943 * Which node is removed/spliced out actually depends on how many
944 * children the victim node has: If it has no children, it gets
945 * deleted. If it has one child, it gets spliced out. If it has two
946 * children, its successor (which has at most a right child) gets
949 victim = get_rb_node_pointer(row, col->rbtree_num);
950 if (victim->rb_left && victim->rb_right)
951 splice_out_node = rb_next(victim);
953 splice_out_node = victim;
954 /* Go up to the root and decrement the size of each node in the path. */
955 for (tmp = splice_out_node; tmp; tmp = rb_parent(tmp))
957 rb_erase(victim, &col->rbtree);
960 static int add_row_to_rbtrees(struct osl_table *t, uint32_t row_num,
961 struct osl_object *volatile_objs, struct osl_row **row_ptr)
965 struct osl_row *row = allocate_row(t->num_rbtrees);
966 const struct osl_column_description *cd;
969 row->volatile_objects = volatile_objs;
970 FOR_EACH_RBTREE_COLUMN(i, t, cd) {
971 if (cd->storage_type == OSL_MAPPED_STORAGE) {
972 struct osl_object obj;
973 ret = get_mapped_object(t, i, row_num, &obj);
976 ret = insert_rbtree(t, i, row, &obj);
977 } else { /* volatile */
978 const struct osl_object *obj
979 = volatile_objs + t->columns[i].volatile_num;
980 ret = insert_rbtree(t, i, row, obj);
988 err: /* rollback changes, i.e. remove added entries from rbtrees */
990 remove_rb_node(t, i--, row);
995 static void free_volatile_objects(const struct osl_table *t,
996 enum osl_close_flags flags)
1000 struct osl_column *rb_col;
1001 const struct osl_column_description *cd;
1003 if (!t->num_volatile_columns)
1005 /* find the first rbtree column (any will do) */
1006 FOR_EACH_RBTREE_COLUMN(i, t, cd)
1008 rb_col = t->columns + i;
1009 /* walk that rbtree and free all volatile objects */
1010 for (n = rb_first(&rb_col->rbtree); n; n = rb_next(n)) {
1011 struct osl_row *r = get_row_pointer(n, rb_col->rbtree_num);
1012 if (flags & OSL_FREE_VOLATILE)
1013 FOR_EACH_VOLATILE_COLUMN(j, t, cd) {
1014 if (cd->storage_flags & OSL_DONT_FREE)
1016 free(r->volatile_objects[
1017 t->columns[j].volatile_num].data);
1019 // for (j = 0; j < t->num_volatile_columns; j++)
1020 // free(r->volatile_objects[j].data);
1021 free(r->volatile_objects);
1026 * Erase all rbtree nodes and free resources.
1028 * \param t Pointer to an open osl table.
1030 * This function is called by osl_close_table().
1032 void clear_rbtrees(struct osl_table *t)
1034 const struct osl_column_description *cd;
1035 unsigned i, rbtrees_cleared = 0;
1037 FOR_EACH_RBTREE_COLUMN(i, t, cd) {
1038 struct osl_column *col = &t->columns[i];
1041 for (n = rb_first(&col->rbtree); n;) {
1043 rb_erase(n, &col->rbtree);
1044 if (rbtrees_cleared == t->num_rbtrees) {
1045 r = get_row_pointer(n, col->rbtree_num);
1056 * Close an osl table.
1058 * \param t Pointer to the table to be closed.
1059 * \param flags Options for what should be cleaned up.
1061 * If osl_open_table() succeeds, the resulting table pointer must later be
1062 * passed to this function in order to flush all changes to the file system and
1063 * to free the resources that were allocated by osl_open_table().
1065 * \return Positive on success, negative on errors. Possible errors: \p E_BAD_TABLE,
1066 * errors returned by unmap_table().
1068 * \sa osl_open_table(), unmap_table().
1070 int osl_close_table(struct osl_table *t, enum osl_close_flags flags)
1075 return -E_BAD_TABLE;
1076 free_volatile_objects(t, flags);
1078 ret = unmap_table(t, flags);
1080 ERROR_LOG("unmap_table failed: %d\n", ret);
1087 * Find out whether the given row number corresponds to an invalid row.
1089 * \param t Pointer to the osl table.
1090 * \param row_num The number of the row in question.
1092 * By definition, a row is considered invalid if all its index entries
1095 * \return Positive if \a row_num corresponds to an invalid row,
1096 * zero if it corresponds to a valid row, negative on errors.
1098 int row_is_invalid(struct osl_table *t, uint32_t row_num)
1101 int i, ret = get_row_index(t, row_num, &row_index);
1105 for (i = 0; i < t->row_index_size; i++) {
1106 if ((unsigned char)row_index[i] != 0xff)
1109 INFO_LOG("row %d is invalid\n", row_num);
1114 * Invalidate a row of an osl table.
1116 * \param t Pointer to an open osl table.
1117 * \param row_num Number of the row to mark as invalid.
1119 * This function marks each mapped object in the index entry of \a row as
1122 * \return Positive on success, negative on errors.
1124 int mark_row_invalid(struct osl_table *t, uint32_t row_num)
1127 int ret = get_row_index(t, row_num, &row_index);
1131 INFO_LOG("marking row %d as invalid\n", row_num);
1132 memset(row_index, 0xff, t->row_index_size);
1137 * Initialize all rbtrees and compute number of invalid rows.
1139 * \param t The table containing the rbtrees to be initialized.
1141 * \return Positive on success, negative on errors.
1143 int init_rbtrees(struct osl_table *t)
1146 const struct osl_column_description *cd;
1148 /* create rbtrees */
1149 FOR_EACH_RBTREE_COLUMN(i, t, cd)
1150 t->columns[i].rbtree = RB_ROOT;
1151 /* add valid rows to rbtrees */
1152 t->num_invalid_rows = 0;
1153 for (i = 0; i < t->num_rows; i++) {
1154 ret = row_is_invalid(t, i);
1158 t->num_invalid_rows++;
1161 ret = add_row_to_rbtrees(t, i, NULL, NULL);
1169 * Open an osl table.
1171 * Each osl table must be opened before its data can be accessed.
1173 * \param table_desc Describes the table to be opened.
1174 * \param result Contains a pointer to the open table on success.
1176 * The table description given by \a desc should coincide with the
1177 * description used at creation time.
1181 int osl_open_table(const struct osl_table_description *table_desc,
1182 struct osl_table **result)
1185 struct osl_table *t;
1186 const struct osl_column_description *cd;
1188 INFO_LOG("opening table %s\n", table_desc->name);
1189 ret = init_table_structure(table_desc, &t);
1192 FOR_EACH_DISK_STORAGE_COLUMN(i, t, cd) {
1193 /* check if directory exists */
1194 char *dirname = column_filename(t, i);
1195 struct stat statbuf;
1196 ret = stat(dirname, &statbuf);
1199 ret = -ERRNO_TO_ERROR(errno);
1202 ret = -ERRNO_TO_ERROR(ENOTDIR);
1203 if (!S_ISDIR(statbuf.st_mode))
1206 ret = map_table(t, MAP_TBL_FL_VERIFY_INDEX);
1210 DEBUG_LOG("num rows: %d\n", t->num_rows);
1211 ret = init_rbtrees(t);
1213 osl_close_table(t, OSL_MARK_CLEAN); /* ignore further errors */
1224 static int create_disk_storage_object_dir(const struct osl_table *t,
1225 unsigned col_num, const char *ds_name)
1230 if (!(t->desc->flags & OSL_LARGE_TABLE))
1232 dirname = disk_storage_dirname(t, col_num, ds_name);
1233 ret = para_mkdir(dirname, 0777);
1235 if (ret < 0 && !is_errno(-ret, EEXIST))
1240 static int write_disk_storage_file(const struct osl_table *t, unsigned col_num,
1241 const struct osl_object *obj, const char *ds_name)
1246 ret = create_disk_storage_object_dir(t, col_num, ds_name);
1249 filename = disk_storage_path(t, col_num, ds_name);
1250 ret = para_write_file(filename, obj->data, obj->size);
1255 static int append_map_file(const struct osl_table *t, unsigned col_num,
1256 const struct osl_object *obj, uint32_t *new_size)
1258 char *filename = column_filename(t, col_num);
1260 char header = 0; /* zero means valid object */
1262 // DEBUG_LOG("appending %zu + 1 byte\n", obj->size);
1263 ret = append_file(filename, &header, 1, obj->data, obj->size,
1269 static int append_row_index(const struct osl_table *t, char *row_index)
1274 if (!t->num_mapped_columns)
1276 filename = index_filename(t->desc);
1277 ret = append_file(filename, NULL, 0, row_index,
1278 t->row_index_size, NULL);
1284 * A wrapper for truncate(2)
1286 * \param path Name of the regular file to truncate
1287 * \param size Number of bytes to \b shave \b off
1289 * Truncate the regular file named by \a path by \a size bytes.
1291 * \return Positive on success, negative on errors. Possible errors include: \p
1292 * E_STAT, \p E_BAD_SIZE, \p E_TRUNC.
1296 int para_truncate(const char *path, off_t size)
1299 struct stat statbuf;
1302 if (stat(path, &statbuf) < 0)
1305 if (statbuf.st_size < size)
1308 if (truncate(path, statbuf.st_size - size) < 0)
1315 static int truncate_mapped_file(const struct osl_table *t, unsigned col_num,
1318 char *filename = column_filename(t, col_num);
1319 int ret = para_truncate(filename, size);
1324 static int delete_disk_storage_file(const struct osl_table *t, unsigned col_num,
1325 const char *ds_name)
1327 char *dirname, *filename = disk_storage_path(t, col_num, ds_name);
1328 int ret = unlink(filename), err = errno;
1332 return -ERRNO_TO_ERROR(err);
1333 if (!(t->desc->flags & OSL_LARGE_TABLE))
1335 dirname = disk_storage_dirname(t, col_num, ds_name);
1342 * Add a new row to an osl table and retrieve this row.
1344 * \param t Pointer to an open osl table.
1345 * \param objects Array of objects to be added.
1346 * \param row Result pointer.
1348 * The \a objects parameter must point to an array containing one object per
1349 * column. The order of the objects in the array is given by the table
1350 * description of \a table. Several sanity checks are performed during object
1351 * insertion and the function returns without modifying the table if any of
1352 * these tests fail. In fact, it is atomic in the sense that it either
1353 * succeeds or leaves the table unchanged (i.e. either all or none of the
1354 * objects are added to the table).
1356 * It is considered an error if an object is added to a column with associated
1357 * rbtree if this object is equal to an object already contained in that column
1358 * (i.e. the compare function for the column's rbtree returns zero).
1360 * Possible errors include: \p E_RB_KEY_EXISTS, \p E_BAD_DATA_SIZE.
1362 * \return Positive on success, negative on errors.
1364 * \sa struct osl_table_description, osl_compare_func, osl_add_row().
1366 int osl_add_and_get_row(struct osl_table *t, struct osl_object *objects,
1367 struct osl_row **row)
1370 char *ds_name = NULL;
1371 struct rb_node **rb_parents = NULL, ***rb_links = NULL;
1372 char *new_row_index = NULL;
1373 struct osl_object *volatile_objs = NULL;
1374 const struct osl_column_description *cd;
1377 return -E_BAD_TABLE;
1378 rb_parents = para_malloc(t->num_rbtrees * sizeof(struct rn_node*));
1379 rb_links = para_malloc(t->num_rbtrees * sizeof(struct rn_node**));
1380 if (t->num_mapped_columns)
1381 new_row_index = para_malloc(t->row_index_size);
1382 /* pass 1: sanity checks */
1383 // DEBUG_LOG("sanity tests: %p:%p\n", objects[0].data,
1384 // objects[1].data);
1385 FOR_EACH_COLUMN(i, t->desc, cd) {
1386 enum osl_storage_type st = cd->storage_type;
1387 enum osl_storage_flags sf = cd->storage_flags;
1389 // ret = -E_NULL_OBJECT;
1392 if (st == OSL_DISK_STORAGE)
1394 if (sf & OSL_RBTREE) {
1395 unsigned rbtree_num = t->columns[i].rbtree_num;
1396 ret = -E_RB_KEY_EXISTS;
1397 // DEBUG_LOG("checking whether %p exists\n",
1398 // objects[i].data);
1399 if (search_rbtree(objects + i, t, i,
1400 &rb_parents[rbtree_num],
1401 &rb_links[rbtree_num]) > 0)
1404 if (sf & OSL_FIXED_SIZE) {
1405 // DEBUG_LOG("fixed size. need: %zu, have: %d\n",
1406 // objects[i].size, cd->data_size);
1407 ret = -E_BAD_DATA_SIZE;
1408 if (objects[i].size != cd->data_size)
1412 if (t->num_disk_storage_columns)
1413 ds_name = disk_storage_name_of_object(t,
1414 &objects[t->disk_storage_name_column]);
1415 ret = unmap_table(t, OSL_MARK_CLEAN);
1418 // DEBUG_LOG("sanity tests passed%s\n", "");
1419 /* pass 2: create data files, append map data */
1420 FOR_EACH_COLUMN(i, t->desc, cd) {
1421 enum osl_storage_type st = cd->storage_type;
1422 if (st == OSL_NO_STORAGE)
1424 if (st == OSL_MAPPED_STORAGE) {
1426 struct osl_column *col = &t->columns[i];
1427 // DEBUG_LOG("appending object of size %zu\n",
1428 // objects[i].size);
1429 ret = append_map_file(t, i, objects + i, &new_size);
1432 update_cell_index(new_row_index, col, new_size,
1437 ret = write_disk_storage_file(t, i, objects + i, ds_name);
1441 ret = append_row_index(t, new_row_index);
1444 ret = map_table(t, MAP_TBL_FL_VERIFY_INDEX);
1445 if (ret < 0) { /* truncate index and rollback changes */
1446 char *filename = index_filename(t->desc);
1447 para_truncate(filename, t->row_index_size);
1451 /* pass 3: add entry to rbtrees */
1452 if (t->num_volatile_columns) {
1453 volatile_objs = para_calloc(t->num_volatile_columns
1454 * sizeof(struct osl_object));
1455 FOR_EACH_VOLATILE_COLUMN(i, t, cd)
1456 volatile_objs[t->columns[i].volatile_num] = objects[i];
1459 // DEBUG_LOG("adding new entry as row #%d\n", t->num_rows - 1);
1460 ret = add_row_to_rbtrees(t, t->num_rows - 1, volatile_objs, row);
1463 // DEBUG_LOG("added new entry as row #%d\n", t->num_rows - 1);
1466 rollback: /* rollback all changes made, ignore further errors */
1467 for (i--; i >= 0; i--) {
1468 cd = get_column_description(t->desc, i);
1469 enum osl_storage_type st = cd->storage_type;
1470 if (st == OSL_NO_STORAGE)
1473 if (st == OSL_MAPPED_STORAGE)
1474 truncate_mapped_file(t, i, objects[i].size);
1475 else /* disk storage */
1476 delete_disk_storage_file(t, i, ds_name);
1478 /* ignore error and return previous error value */
1479 map_table(t, MAP_TBL_FL_VERIFY_INDEX);
1481 free(new_row_index);
1489 * Add a new row to an osl table.
1491 * \param t Same meaning as osl_add_and_get_row().
1492 * \param objects Same meaning as osl_add_and_get_row().
1494 * \return The return value of the underlying call to osl_add_and_get_row().
1496 * This is equivalent to osl_add_and_get_row(t, objects, NULL).
1498 int osl_add_row(struct osl_table *t, struct osl_object *objects)
1500 return osl_add_and_get_row(t, objects, NULL);
1504 * Retrieve an object identified by row and column
1506 * \param t Pointer to an open osl table.
1507 * \param r Pointer to the row.
1508 * \param col_num The column number.
1509 * \param object The result pointer.
1511 * The column determined by \a col_num must be of type \p OSL_MAPPED_STORAGE
1512 * or \p OSL_NO_STORAGE, i.e. no disk storage objects may be retrieved by this
1515 * \return Positive if object was found, negative on errors. Possible errors
1516 * include: \p E_BAD_TABLE, \p E_BAD_STORAGE_TYPE.
1518 * \sa osl_storage_type, osl_open_disk_object().
1520 int osl_get_object(const struct osl_table *t, const struct osl_row *r,
1521 unsigned col_num, struct osl_object *object)
1523 const struct osl_column_description *cd;
1526 return -E_BAD_TABLE;
1527 cd = get_column_description(t->desc, col_num);
1528 /* col must not be disk storage */
1529 if (cd->storage_type == OSL_DISK_STORAGE)
1530 return -E_BAD_STORAGE_TYPE;
1531 if (cd->storage_type == OSL_MAPPED_STORAGE)
1532 return get_mapped_object(t, col_num, r->num, object);
1534 *object = r->volatile_objects[t->columns[col_num].volatile_num];
1538 static int mark_mapped_object_invalid(const struct osl_table *t,
1539 uint32_t row_num, unsigned col_num)
1541 struct osl_object obj;
1543 int ret = get_mapped_object(t, col_num, row_num, &obj);
1554 * Delete a row from an osl table.
1556 * \param t Pointer to an open osl table.
1557 * \param row Pointer to the row to delete.
1559 * This removes all disk storage objects, removes all rbtree nodes, and frees
1560 * all volatile objects belonging to the given row. For mapped columns, the
1561 * data is merely marked invalid and may be pruned from time to time by
1564 * \return Positive on success, negative on errors. Possible errors include:
1565 * \p E_BAD_TABLE, errors returned by osl_get_object().
1567 int osl_del_row(struct osl_table *t, struct osl_row *row)
1569 struct osl_row *r = row;
1571 const struct osl_column_description *cd;
1574 return -E_BAD_TABLE;
1575 INFO_LOG("deleting row %p\n", row);
1577 if (t->num_disk_storage_columns) {
1579 ret = disk_storage_name_of_row(t, r, &ds_name);
1582 FOR_EACH_DISK_STORAGE_COLUMN(i, t, cd)
1583 delete_disk_storage_file(t, i, ds_name);
1586 FOR_EACH_COLUMN(i, t->desc, cd) {
1587 struct osl_column *col = t->columns + i;
1588 enum osl_storage_type st = cd->storage_type;
1589 remove_rb_node(t, i, r);
1590 if (st == OSL_MAPPED_STORAGE) {
1591 mark_mapped_object_invalid(t, r->num, i);
1594 if (st == OSL_NO_STORAGE && !(cd->storage_flags & OSL_DONT_FREE))
1595 free(r->volatile_objects[col->volatile_num].data);
1597 if (t->num_mapped_columns) {
1598 ret = mark_row_invalid(t, r->num);
1601 t->num_invalid_rows++;
1606 free(r->volatile_objects);
1611 /* test if column has an rbtree */
1612 static int check_rbtree_col(const struct osl_table *t, unsigned col_num,
1613 struct osl_column **col)
1616 return -E_BAD_TABLE;
1617 if (!(get_column_description(t->desc, col_num)->storage_flags & OSL_RBTREE))
1618 return -E_BAD_STORAGE_FLAGS;
1619 *col = t->columns + col_num;
1624 * Get the row that contains the given object.
1626 * \param t Pointer to an open osl table.
1627 * \param col_num The number of the column to be searched.
1628 * \param obj The object to be looked up.
1629 * \param result Points to the row containing \a obj.
1631 * Lookup \a obj in \a t and return the row containing \a obj. The column
1632 * specified by \a col_num must have an associated rbtree.
1634 * \return Positive on success, negative on errors. If an error occurred, \a
1635 * result is set to \p NULL. Possible errors include: \p E_BAD_TABLE, \p
1636 * E_BAD_STORAGE_FLAGS, errors returned by get_mapped_object(), \p
1637 * E_RB_KEY_NOT_FOUND.
1639 * \sa osl_storage_flags
1641 int osl_get_row(const struct osl_table *t, unsigned col_num,
1642 const struct osl_object *obj, struct osl_row **result)
1645 struct rb_node *node;
1646 struct osl_row *row;
1647 struct osl_column *col;
1650 ret = check_rbtree_col(t, col_num, &col);
1653 ret = search_rbtree(obj, t, col_num, &node, NULL);
1656 row = get_row_pointer(node, t->columns[col_num].rbtree_num);
1661 static int rbtree_loop(struct osl_column *col, void *private_data,
1662 osl_rbtree_loop_func *func)
1664 struct rb_node *n, *tmp;
1666 /* this for-loop is safe against removal of an entry */
1667 for (n = rb_first(&col->rbtree), tmp = n? rb_next(n) : NULL;
1669 n = tmp, tmp = tmp? rb_next(tmp) : NULL) {
1670 struct osl_row *r = get_row_pointer(n, col->rbtree_num);
1671 int ret = func(r, private_data);
1678 static int rbtree_loop_reverse(struct osl_column *col, void *private_data,
1679 osl_rbtree_loop_func *func)
1681 struct rb_node *n, *tmp;
1683 /* safe against removal of an entry */
1684 for (n = rb_last(&col->rbtree), tmp = n? rb_prev(n) : NULL;
1686 n = tmp, tmp = tmp? rb_prev(tmp) : NULL) {
1687 struct osl_row *r = get_row_pointer(n, col->rbtree_num);
1688 int ret = func(r, private_data);
1696 * Loop over all nodes in an rbtree.
1698 * \param t Pointer to an open osl table.
1699 * \param col_num The column to use for iterating over the elements.
1700 * \param private_data Pointer that gets passed to \a func.
1701 * \param func The function to be called for each node in the rbtree.
1703 * This function does an in-order walk of the rbtree associated with \a
1704 * col_num. It is an error if the \p OSL_RBTREE flag is not set for this
1705 * column. For each node in the rbtree, the given function \a func is called
1706 * with two pointers as arguments: The first osl_row* argument points to the
1707 * row that contains the object corresponding to the rbtree node currently
1708 * traversed, and the \a private_data pointer is passed verbatim to \a func as the
1709 * second argument. The loop terminates either if \a func returns a negative
1710 * value, or if all nodes of the tree have been visited.
1713 * \return Positive on success, negative on errors. If the termination of the
1714 * loop was caused by \a func returning a negative value, this value is
1717 * \sa osl_storage_flags, osl_rbtree_loop_reverse(), osl_compare_func.
1719 int osl_rbtree_loop(const struct osl_table *t, unsigned col_num,
1720 void *private_data, osl_rbtree_loop_func *func)
1722 struct osl_column *col;
1724 int ret = check_rbtree_col(t, col_num, &col);
1727 return rbtree_loop(col, private_data, func);
1731 * Loop over all nodes in an rbtree in reverse order.
1733 * \param t Identical meaning as in \p osl_rbtree_loop().
1734 * \param col_num Identical meaning as in \p osl_rbtree_loop().
1735 * \param private_data Identical meaning as in \p osl_rbtree_loop().
1736 * \param func Identical meaning as in \p osl_rbtree_loop().
1738 * This function is identical to \p osl_rbtree_loop(), the only difference
1739 * is that the tree is walked in reverse order.
1741 * \return The same return value as \p osl_rbtree_loop().
1743 * \sa osl_rbtree_loop().
1745 int osl_rbtree_loop_reverse(const struct osl_table *t, unsigned col_num,
1746 void *private_data, osl_rbtree_loop_func *func)
1748 struct osl_column *col;
1750 int ret = check_rbtree_col(t, col_num, &col);
1753 return rbtree_loop_reverse(col, private_data, func);
1756 /* TODO: Rollback changes on errors */
1757 static int rename_disk_storage_objects(struct osl_table *t,
1758 struct osl_object *old_obj, struct osl_object *new_obj)
1761 const struct osl_column_description *cd;
1762 char *old_ds_name, *new_ds_name;
1764 if (!t->num_disk_storage_columns)
1765 return 1; /* nothing to do */
1766 if (old_obj->size == new_obj->size && !memcmp(new_obj->data,
1767 old_obj->data, new_obj->size))
1768 return 1; /* object did not change */
1769 old_ds_name = disk_storage_name_of_object(t, old_obj);
1770 new_ds_name = disk_storage_name_of_object(t, new_obj);
1771 FOR_EACH_DISK_STORAGE_COLUMN(i, t, cd) {
1772 char *old_filename, *new_filename;
1773 ret = create_disk_storage_object_dir(t, i, new_ds_name);
1776 old_filename = disk_storage_path(t, i, old_ds_name);
1777 new_filename = disk_storage_path(t, i, new_ds_name);
1778 ret = para_rename(old_filename, new_filename);
1793 * Change an object in an osl table.
1795 * \param t Pointer to an open osl table.
1796 * \param r Pointer to the row containing the object to be updated.
1797 * \param col_num Number of the column containing the object to be updated.
1798 * \param obj Pointer to the replacement object.
1800 * This function gets rid of all references to the old object. This includes
1801 * removal of the rbtree node in case there is an rbtree associated with \a
1802 * col_num. It then inserts \a obj into the table and the rbtree if necessary.
1804 * If the \p OSL_RBTREE flag is set for \a col_num, you \b MUST call this
1805 * function in order to change the contents of an object, even for volatile or
1806 * mapped columns of constant size (which may be updated directly if \p
1807 * OSL_RBTREE is not set). Otherwise the rbtree might become corrupted.
1811 int osl_update_object(struct osl_table *t, const struct osl_row *r,
1812 unsigned col_num, struct osl_object *obj)
1814 struct osl_column *col;
1815 const struct osl_column_description *cd;
1819 return -E_BAD_TABLE;
1820 col = &t->columns[col_num];
1821 cd = get_column_description(t->desc, col_num);
1822 DEBUG_LOG("updating column %u of %s\n", col_num, t->desc->name);
1823 if (cd->storage_flags & OSL_RBTREE) {
1824 if (search_rbtree(obj, t, col_num, NULL, NULL) > 0)
1825 return -E_RB_KEY_EXISTS;
1827 if (cd->storage_flags & OSL_FIXED_SIZE) {
1828 if (obj->size != cd->data_size)
1829 return -E_BAD_DATA_SIZE;
1831 remove_rb_node(t, col_num, r);
1832 if (cd->storage_type == OSL_NO_STORAGE) { /* TODO: If fixed size, reuse object? */
1833 free(r->volatile_objects[col->volatile_num].data);
1834 r->volatile_objects[col->volatile_num] = *obj;
1835 } else if (cd->storage_type == OSL_DISK_STORAGE) {
1837 ret = disk_storage_name_of_row(t, r, &ds_name);
1840 ret = delete_disk_storage_file(t, col_num, ds_name);
1841 if (ret < 0 && !is_errno(-ret, ENOENT)) {
1845 ret = write_disk_storage_file(t, col_num, obj, ds_name);
1849 } else { /* mapped storage */
1850 struct osl_object old_obj;
1851 ret = get_mapped_object(t, col_num, r->num, &old_obj);
1855 * If the updated column is the disk storage name column, the
1856 * disk storage name changes, so we have to rename all disk
1857 * storage objects accordingly.
1859 if (col_num == t->disk_storage_name_column) {
1860 ret = rename_disk_storage_objects(t, &old_obj, obj);
1864 if (cd->storage_flags & OSL_FIXED_SIZE)
1865 memcpy(old_obj.data, obj->data, cd->data_size);
1866 else { /* TODO: if the size doesn't change, use old space */
1867 uint32_t new_data_map_size;
1869 ret = get_row_index(t, r->num, &row_index);
1872 ret = mark_mapped_object_invalid(t, r->num, col_num);
1875 unmap_column(t, col_num);
1876 ret = append_map_file(t, col_num, obj,
1877 &new_data_map_size);
1880 ret = map_column(t, col_num);
1883 update_cell_index(row_index, col, new_data_map_size,
1887 if (cd->storage_flags & OSL_RBTREE) {
1888 ret = insert_rbtree(t, col_num, r, obj);
1896 * Retrieve an object of type \p OSL_DISK_STORAGE by row and column.
1898 * \param t Pointer to an open osl table.
1899 * \param r Pointer to the row containing the object.
1900 * \param col_num The column number.
1901 * \param obj Points to the result upon successful return.
1903 * For columns of type \p OSL_DISK_STORAGE, this function must be used to
1904 * retrieve one of its containing objects. Afterwards, osl_close_disk_object()
1905 * must be called in order to deallocate the resources.
1907 * \return Positive on success, negative on errors. Possible errors include:
1908 * \p E_BAD_TABLE, \p E_BAD_STORAGE_TYPE, errors returned by osl_get_object().
1910 * \sa osl_get_object(), osl_storage_type, osl_close_disk_object().
1912 int osl_open_disk_object(const struct osl_table *t, const struct osl_row *r,
1913 unsigned col_num, struct osl_object *obj)
1915 const struct osl_column_description *cd;
1916 char *ds_name, *filename;
1920 return -E_BAD_TABLE;
1921 cd = get_column_description(t->desc, col_num);
1922 if (cd->storage_type != OSL_DISK_STORAGE)
1923 return -E_BAD_STORAGE_TYPE;
1925 ret = disk_storage_name_of_row(t, r, &ds_name);
1928 filename = disk_storage_path(t, col_num, ds_name);
1930 DEBUG_LOG("filename: %s\n", filename);
1931 ret = mmap_full_file(filename, O_RDONLY, &obj->data, &obj->size, NULL);
1937 * Free resources that were allocated during osl_open_disk_object().
1939 * \param obj Pointer to the object previously returned by open_disk_object().
1941 * \return The return value of the underlying call to para_munmap().
1943 * \sa para_munmap().
1945 int osl_close_disk_object(struct osl_object *obj)
1947 return para_munmap(obj->data, obj->size);
1951 * Get the number of rows of the given table.
1953 * \param t Pointer to an open osl table.
1954 * \param num_rows Result is returned here.
1956 * The number of rows returned via \a num_rows excluding any invalid rows.
1958 * \return Positive on success, \p -E_BAD_TABLE if \a t is \p NULL.
1960 int osl_get_num_rows(const struct osl_table *t, unsigned *num_rows)
1963 return -E_BAD_TABLE;
1964 assert(t->num_rows >= t->num_invalid_rows);
1965 *num_rows = t->num_rows - t->num_invalid_rows;
1970 * Get the rank of a row.
1972 * \param t An open osl table.
1973 * \param r The row to get the rank of.
1974 * \param col_num The number of an rbtree column.
1975 * \param rank Result pointer.
1977 * The rank is, by definition, the position of the row in the linear order
1978 * determined by an in-order tree walk of the rbtree associated with column
1979 * number \a col_num of \a table.
1981 * \return Positive on success, negative on errors.
1983 * \sa osl_get_nth_row().
1985 int osl_get_rank(const struct osl_table *t, struct osl_row *r,
1986 unsigned col_num, unsigned *rank)
1988 struct osl_object obj;
1989 struct osl_column *col;
1990 struct rb_node *node;
1991 int ret = check_rbtree_col(t, col_num, &col);
1995 ret = osl_get_object(t, r, col_num, &obj);
1998 ret = search_rbtree(&obj, t, col_num, &node, NULL);
2001 ret = rb_rank(node, rank);
2008 * Get the row with n-th greatest value.
2010 * \param t Pointer to an open osl table.
2011 * \param col_num The column number.
2012 * \param n The rank of the desired row.
2013 * \param result Row is returned here.
2015 * Retrieve the n-th order statistic with respect to the compare function
2016 * of the rbtree column \a col_num. In other words, get that row with
2017 * \a n th greatest value in column \a col_num. It's an error if
2018 * \a col_num is not a rbtree column, or if \a n is larger than the
2019 * number of rows in the table.
2021 * \return Positive on success, negative on errors. Possible errors:
2022 * \p E_BAD_TABLE, \p E_BAD_STORAGE_FLAGS, \p E_RB_KEY_NOT_FOUND.
2024 * \sa osl_storage_flags, osl_compare_func, osl_get_row(),
2025 * osl_rbtree_last_row(), osl_rbtree_first_row(), osl_get_rank().
2027 int osl_get_nth_row(const struct osl_table *t, unsigned col_num,
2028 unsigned n, struct osl_row **result)
2030 struct osl_column *col;
2031 struct rb_node *node;
2036 return -E_RB_KEY_NOT_FOUND;
2037 ret = osl_get_num_rows(t, &num_rows);
2041 return -E_RB_KEY_NOT_FOUND;
2042 ret = check_rbtree_col(t, col_num, &col);
2045 node = rb_nth(col->rbtree.rb_node, n);
2047 return -E_RB_KEY_NOT_FOUND;
2048 *result = get_row_pointer(node, col->rbtree_num);
2053 * Get the row corresponding to the smallest rbtree node of a column.
2055 * \param t An open rbtree table.
2056 * \param col_num The number of the rbtree column.
2057 * \param result A pointer to the first row is returned here.
2059 * The rbtree node of the smallest object (with respect to the corresponding
2060 * compare function) is selected and the row containing this object is
2061 * returned. It is an error if \a col_num refers to a column without an
2062 * associated rbtree.
2064 * \return Positive on success, negative on errors.
2066 * \sa osl_get_nth_row(), osl_rbtree_last_row().
2068 int osl_rbtree_first_row(const struct osl_table *t, unsigned col_num,
2069 struct osl_row **result)
2071 return osl_get_nth_row(t, col_num, 1, result);
2075 * Get the row corresponding to the greatest rbtree node of a column.
2077 * \param t The same meaning as in \p osl_rbtree_first_row().
2078 * \param col_num The same meaning as in \p osl_rbtree_first_row().
2079 * \param result The same meaning as in \p osl_rbtree_first_row().
2081 * This function works just like osl_rbtree_first_row(), the only difference
2082 * is that the row containing the greatest rather than the smallest object is
2085 * \return Positive on success, negative on errors.
2087 * \sa osl_get_nth_row(), osl_rbtree_first_row().
2089 int osl_rbtree_last_row(const struct osl_table *t, unsigned col_num,
2090 struct osl_row **result)
2093 int ret = osl_get_num_rows(t, &num_rows);
2097 return osl_get_nth_row(t, col_num, num_rows, result);