Merge commit 'meins/master'
[paraslash.git] / osl.c
1 /*
2 * Copyright (C) 2007-2008 Andre Noll <maan@systemlinux.org>
3 *
4 * Licensed under the GPL v2. For licencing details see COPYING.
5 */
6
7 /** \file osl.c Object storage layer functions. */
8 #include <dirent.h> /* readdir() */
9 #include <assert.h>
10
11
12 #include "para.h"
13 #include "error.h"
14 #include "fd.h"
15 #include "osl_core.h"
16 /**
17 * A wrapper for lseek(2).
18 *
19 * \param fd The file descriptor whose offset is to be to repositioned.
20 * \param offset A value-result parameter.
21 * \param whence Usual repositioning directive.
22 *
23 * Reposition the offset of the file descriptor \a fd to the argument \a offset
24 * according to the directive \a whence. Upon successful return, \a offset
25 * contains the resulting offset location as measured in bytes from the
26 * beginning of the file.
27 *
28 * \return Positive on success. Otherwise, the function returns \p -E_LSEEK.
29 *
30 * \sa lseek(2).
31 */
32 int para_lseek(int fd, off_t *offset, int whence)
33 {
34 *offset = lseek(fd, *offset, whence);
35 int ret = -E_LSEEK;
36 if (*offset == -1)
37 return ret;
38 return 1;
39 }
40
41 /**
42 * Wrapper for the write system call.
43 *
44 * \param fd The file descriptor to write to.
45 * \param buf The buffer to write.
46 * \param size The length of \a buf in bytes.
47 *
48 * This function writes out the given buffer and retries if an interrupt
49 * occurred during the write.
50 *
51 * \return On success, the number of bytes written is returned, otherwise, the
52 * function returns \p -E_WRITE.
53 *
54 * \sa write(2).
55 */
56 ssize_t para_write(int fd, const void *buf, size_t size)
57 {
58 ssize_t ret;
59
60 for (;;) {
61 ret = write(fd, buf, size);
62 if ((ret < 0) && (errno == EAGAIN || errno == EINTR))
63 continue;
64 return ret >= 0? ret : -E_WRITE;
65 }
66 }
67
68 /**
69 * Write the whole buffer to a file descriptor.
70 *
71 * \param fd The file descriptor to write to.
72 * \param buf The buffer to write.
73 * \param size The length of \a buf in bytes.
74 *
75 * This function writes the given buffer and continues on short writes and
76 * when interrupted by a signal.
77 *
78 * \return Positive on success, negative on errors. Possible errors: any
79 * errors returned by para_write().
80 *
81 * \sa para_write().
82 */
83 ssize_t para_write_all(int fd, const void *buf, size_t size)
84 {
85 // PARA_DEBUG_LOG("writing %zu bytes\n", size);
86 const char *b = buf;
87 while (size) {
88 ssize_t ret = para_write(fd, b, size);
89 // PARA_DEBUG_LOG("ret: %zd\n", ret);
90 if (ret < 0)
91 return ret;
92 b += ret;
93 size -= ret;
94 }
95 return 1;
96 }
97 /**
98 * Open a file, write the given buffer and close the file.
99 *
100 * \param filename Full path to the file to open.
101 * \param buf The buffer to write to the file.
102 * \param size The size of \a buf.
103 *
104 * \return Positive on success, negative on errors. Possible errors include:
105 * any errors from para_open() or para_write().
106 *
107 * \sa para_open(), para_write().
108 */
109 int para_write_file(const char *filename, const void *buf, size_t size)
110 {
111 int ret, fd;
112
113 ret = para_open(filename, O_WRONLY | O_CREAT | O_EXCL, 0644);
114 if (ret < 0)
115 return ret;
116 fd = ret;
117 ret = para_write_all(fd, buf, size);
118 if (ret < 0)
119 goto out;
120 ret = 1;
121 out:
122 close(fd);
123 return ret;
124 }
125
126 static int append_file(const char *filename, char *header, size_t header_size,
127 char *data, size_t data_size, uint32_t *new_pos)
128 {
129 int ret, fd;
130
131 // PARA_DEBUG_LOG("appending %zu + %zu bytes\n", header_size, data_size);
132 ret = para_open(filename, O_WRONLY | O_CREAT | O_APPEND, 0644);
133 if (ret < 0)
134 return ret;
135 fd = ret;
136 if (header && header_size) {
137 ret = para_write_all(fd, header, header_size);
138 if (ret < 0)
139 goto out;
140 }
141 ret = para_write_all(fd, data, data_size);
142 if (ret < 0)
143 goto out;
144 if (new_pos) {
145 off_t offset = 0;
146 ret = para_lseek(fd, &offset, SEEK_END);
147 if (ret < 0)
148 goto out;
149 // PARA_DEBUG_LOG("new file size: " FMT_OFF_T "\n", offset);
150 *new_pos = offset;
151 }
152 ret = 1;
153 out:
154 close(fd);
155 return ret;
156 }
157
158 /**
159 * Traverse the given directory recursively.
160 *
161 * \param dirname The directory to traverse.
162 * \param func The function to call for each entry.
163 * \param private_data Pointer to an arbitrary data structure.
164 *
165 * For each regular file under \a dirname, the supplied function \a func is
166 * called. The full path of the regular file and the \a private_data pointer
167 * are passed to \a func. Directories for which the calling process has no
168 * permissions to change to are silently ignored.
169 *
170 * \return Standard.
171 */
172 int for_each_file_in_dir(const char *dirname,
173 int (*func)(const char *, void *), void *private_data)
174 {
175 DIR *dir;
176 struct dirent *entry;
177 int cwd_fd, ret2, ret = para_opendir(dirname, &dir, &cwd_fd);
178
179 if (ret < 0)
180 return ret == -ERRNO_TO_PARA_ERROR(EACCES)? 1 : ret;
181 /* scan cwd recursively */
182 while ((entry = readdir(dir))) {
183 mode_t m;
184 char *tmp;
185 struct stat s;
186
187 if (!strcmp(entry->d_name, "."))
188 continue;
189 if (!strcmp(entry->d_name, ".."))
190 continue;
191 if (lstat(entry->d_name, &s) == -1)
192 continue;
193 m = s.st_mode;
194 if (!S_ISREG(m) && !S_ISDIR(m))
195 continue;
196 tmp = make_message("%s/%s", dirname, entry->d_name);
197 if (!S_ISDIR(m)) {
198 ret = func(tmp, private_data);
199 free(tmp);
200 if (ret < 0)
201 goto out;
202 continue;
203 }
204 /* directory */
205 ret = for_each_file_in_dir(tmp, func, private_data);
206 free(tmp);
207 if (ret < 0)
208 goto out;
209 }
210 ret = 1;
211 out:
212 closedir(dir);
213 ret2 = para_fchdir(cwd_fd);
214 if (ret2 < 0 && ret >= 0)
215 ret = ret2;
216 close(cwd_fd);
217 return ret;
218 }
219
220 static int verify_name(const char *name)
221 {
222 if (!name)
223 return -E_BAD_NAME;
224 if (!*name)
225 return -E_BAD_NAME;
226 if (strchr(name, '/'))
227 return -E_BAD_NAME;
228 if (!strcmp(name, ".."))
229 return -E_BAD_NAME;
230 if (!strcmp(name, "."))
231 return -E_BAD_NAME;
232 return 1;
233 }
234
235 /**
236 * Compare two osl objects pointing to unsigned integers of 32 bit size.
237 *
238 * \param obj1 Pointer to the first integer.
239 * \param obj2 Pointer to the second integer.
240 *
241 * \return The values required for an osl compare function.
242 *
243 * \sa osl_compare_func, osl_hash_compare().
244 */
245 int uint32_compare(const struct osl_object *obj1, const struct osl_object *obj2)
246 {
247 uint32_t d1 = read_u32((const char *)obj1->data);
248 uint32_t d2 = read_u32((const char *)obj2->data);
249
250 if (d1 < d2)
251 return 1;
252 if (d1 > d2)
253 return -1;
254 return 0;
255 }
256
257 /**
258 * Compare two osl objects pointing to hash values.
259 *
260 * \param obj1 Pointer to the first hash object.
261 * \param obj2 Pointer to the second hash object.
262 *
263 * \return The values required for an osl compare function.
264 *
265 * \sa osl_compare_func, uint32_compare().
266 */
267 int osl_hash_compare(const struct osl_object *obj1, const struct osl_object *obj2)
268 {
269 return hash_compare((HASH_TYPE *)obj1->data, (HASH_TYPE *)obj2->data);
270 }
271
272 static char *disk_storage_dirname(const struct osl_table *t, unsigned col_num,
273 const char *ds_name)
274 {
275 char *dirname, *column_name = column_filename(t, col_num);
276
277 if (!(t->desc->flags & OSL_LARGE_TABLE))
278 return column_name;
279 dirname = make_message("%s/%.2s", column_name, ds_name);
280 free(column_name);
281 return dirname;
282 }
283
284 static char *disk_storage_name_of_object(const struct osl_table *t,
285 const struct osl_object *obj)
286 {
287 HASH_TYPE hash[HASH_SIZE];
288 hash_object(obj, hash);
289 return disk_storage_name_of_hash(t, hash);
290 }
291
292 static int disk_storage_name_of_row(const struct osl_table *t,
293 const struct osl_row *row, char **name)
294 {
295 struct osl_object obj;
296 int ret = osl_get_object(t, row, t->disk_storage_name_column, &obj);
297
298 if (ret < 0)
299 return ret;
300 *name = disk_storage_name_of_object(t, &obj);
301 return 1;
302 }
303
304 static void column_name_hash(const char *col_name, HASH_TYPE *hash)
305 {
306 hash_function(col_name, strlen(col_name), hash);
307 }
308
309 static int init_column_descriptions(struct osl_table *t)
310 {
311 int i, j, ret;
312 const struct osl_column_description *cd;
313
314 ret = -E_BAD_TABLE_DESC;
315 ret = verify_name(t->desc->name);
316 if (ret < 0)
317 goto err;
318 ret = -E_BAD_DB_DIR;
319 if (!t->desc->dir && (t->num_disk_storage_columns || t->num_mapped_columns))
320 goto err;
321 /* the size of the index header without column descriptions */
322 t->index_header_size = IDX_COLUMN_DESCRIPTIONS;
323 FOR_EACH_COLUMN(i, t->desc, cd) {
324 struct osl_column *col = t->columns + i;
325 if (cd->storage_flags & OSL_RBTREE) {
326 if (!cd->compare_function)
327 return -E_NO_COMPARE_FUNC;
328 }
329 if (cd->storage_type == OSL_NO_STORAGE)
330 continue;
331 ret = -E_NO_COLUMN_NAME;
332 if (!cd->name || !cd->name[0])
333 goto err;
334 ret = verify_name(cd->name);
335 if (ret < 0)
336 goto err;
337 t->index_header_size += index_column_description_size(cd->name);
338 column_name_hash(cd->name, col->name_hash);
339 ret = -E_DUPLICATE_COL_NAME;
340 for (j = i + 1; j < t->desc->num_columns; j++) {
341 const char *name2 = get_column_description(t->desc,
342 j)->name;
343 if (cd->name && name2 && !strcmp(cd->name, name2))
344 goto err;
345 }
346 }
347 return 1;
348 err:
349 return ret;
350 }
351
352 /**
353 * Initialize a struct table from given table description.
354 *
355 * \param desc The description of the osl table.
356 * \param table_ptr Result is returned here.
357 *
358 * This function performs several sanity checks on \p desc and returns if any
359 * of these tests fail. On success, a struct \p osl_table is allocated and
360 * initialized with data derived from \p desc.
361 *
362 * \return Positive on success, negative on errors. Possible errors include: \p
363 * E_BAD_TABLE_DESC, \p E_NO_COLUMN_DESC, \p E_NO_COLUMNS, \p
364 * E_BAD_STORAGE_TYPE, \p E_BAD_STORAGE_FLAGS, \p E_BAD_STORAGE_SIZE, \p
365 * E_NO_UNIQUE_RBTREE_COLUMN, \p E_NO_RBTREE_COL.
366 *
367 * \sa struct osl_table.
368 */
369 int init_table_structure(const struct osl_table_description *desc,
370 struct osl_table **table_ptr)
371 {
372 const struct osl_column_description *cd;
373 struct osl_table *t = para_calloc(sizeof(*t));
374 int i, ret = -E_BAD_TABLE_DESC, have_disk_storage_name_column = 0;
375
376 if (!desc)
377 goto err;
378 PARA_DEBUG_LOG("creating table structure for '%s' from table "
379 "description\n", desc->name);
380 ret = -E_NO_COLUMN_DESC;
381 if (!desc->column_descriptions)
382 goto err;
383 ret = -E_NO_COLUMNS;
384 if (!desc->num_columns)
385 goto err;
386 t->columns = para_calloc(desc->num_columns * sizeof(struct osl_column));
387 t->desc = desc;
388 FOR_EACH_COLUMN(i, t->desc, cd) {
389 enum osl_storage_type st = cd->storage_type;
390 enum osl_storage_flags sf = cd->storage_flags;
391 struct osl_column *col = &t->columns[i];
392
393 ret = -E_BAD_STORAGE_TYPE;
394 if (st != OSL_MAPPED_STORAGE && st != OSL_DISK_STORAGE
395 && st != OSL_NO_STORAGE)
396 goto err;
397 ret = -E_BAD_STORAGE_FLAGS;
398 if (st == OSL_DISK_STORAGE && sf & OSL_RBTREE)
399 goto err;
400 ret = -E_BAD_STORAGE_SIZE;
401 if (sf & OSL_FIXED_SIZE && !cd->data_size)
402 goto err;
403 switch (st) {
404 case OSL_DISK_STORAGE:
405 t->num_disk_storage_columns++;
406 break;
407 case OSL_MAPPED_STORAGE:
408 t->num_mapped_columns++;
409 col->index_offset = t->row_index_size;
410 t->row_index_size += 8;
411 break;
412 case OSL_NO_STORAGE:
413 col->volatile_num = t->num_volatile_columns;
414 t->num_volatile_columns++;
415 break;
416 }
417 if (sf & OSL_RBTREE) {
418 col->rbtree_num = t->num_rbtrees;
419 t->num_rbtrees++;
420 if ((sf & OSL_UNIQUE) && (st == OSL_MAPPED_STORAGE)) {
421 if (!have_disk_storage_name_column)
422 t->disk_storage_name_column = i;
423 have_disk_storage_name_column = 1;
424 }
425 }
426 }
427 ret = -E_NO_UNIQUE_RBTREE_COLUMN;
428 if (t->num_disk_storage_columns && !have_disk_storage_name_column)
429 goto err;
430 ret = -E_NO_RBTREE_COL;
431 if (!t->num_rbtrees)
432 goto err;
433 /* success */
434 PARA_DEBUG_LOG("OK. Index entry size: %u\n", t->row_index_size);
435 ret = init_column_descriptions(t);
436 if (ret < 0)
437 goto err;
438 *table_ptr = t;
439 return 1;
440 err:
441 free(t->columns);
442 free(t);
443 return ret;
444 }
445
446 /**
447 * Read the table description from index header.
448 *
449 * \param map The memory mapping of the index file.
450 * \param desc The values found in the index header are returned here.
451 *
452 * Read the index header, check for the paraslash magic string and the table version number.
453 * Read all information stored in the index header into \a desc.
454 *
455 * \return Positive on success, negative on errors.
456 *
457 * \sa struct osl_table_description, osl_create_table.
458 */
459 int read_table_desc(struct osl_object *map, struct osl_table_description *desc)
460 {
461 char *buf = map->data;
462 uint8_t version;
463 uint16_t header_size;
464 int ret, i;
465 unsigned offset;
466 struct osl_column_description *cd;
467
468 if (map->size < MIN_INDEX_HEADER_SIZE(1))
469 return -E_SHORT_TABLE;
470 if (strncmp(buf + IDX_PARA_MAGIC, PARA_MAGIC, strlen(PARA_MAGIC)))
471 return -E_NO_MAGIC;
472 version = read_u8(buf + IDX_VERSION);
473 if (version < MIN_TABLE_VERSION || version > MAX_TABLE_VERSION)
474 return -E_VERSION_MISMATCH;
475 desc->num_columns = read_u8(buf + IDX_TABLE_FLAGS);
476 desc->flags = read_u8(buf + IDX_TABLE_FLAGS);
477 desc->num_columns = read_u16(buf + IDX_NUM_COLUMNS);
478 PARA_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)
483 return -E_BAD_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) {
488 char *null_byte;
489
490 ret = -E_SHORT_TABLE;
491 if (map->size < offset + MIN_IDX_COLUMN_DESCRIPTION_SIZE) {
492 PARA_ERROR_LOG("map size = %zu < %u = offset + min desc size\n",
493 map->size, offset + MIN_IDX_COLUMN_DESCRIPTION_SIZE);
494 goto err;
495 }
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;
503 if (!null_byte)
504 goto err;
505 cd->name = para_strdup(buf + offset + IDX_CD_NAME);
506 offset += index_column_description_size(cd->name);
507 }
508 if (offset != header_size) {
509 ret = -E_INDEX_CORRUPTION;
510 PARA_ERROR_LOG("real header size = %u != %u = stored header size\n",
511 offset, header_size);
512 goto err;
513 }
514 return 1;
515 err:
516 FOR_EACH_COLUMN(i, desc, cd)
517 free(cd->name);
518 return ret;
519 }
520
521 /*
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.
524 */
525 static int compare_table_descriptions(struct osl_table *t)
526 {
527 int i, ret;
528 struct osl_table_description desc;
529 const struct osl_column_description *cd1, *cd2;
530
531 /* read the on-disk structure into desc */
532 ret = read_table_desc(&t->index_map, &desc);
533 if (ret < 0)
534 return ret;
535 ret = -E_BAD_TABLE_FLAGS;
536 if (desc.flags != t->desc->flags)
537 goto out;
538 ret = -E_BAD_COLUMN_NUM;
539 if (desc.num_columns != t->desc->num_columns)
540 goto out;
541 FOR_EACH_COLUMN(i, t->desc, cd1) {
542 cd2 = get_column_description(&desc, i);
543 ret = -E_BAD_STORAGE_TYPE;
544 if (cd1->storage_type != cd2->storage_type)
545 goto out;
546 ret = -E_BAD_STORAGE_FLAGS;
547 if (cd1->storage_flags != cd2->storage_flags) {
548 PARA_ERROR_LOG("sf1 = %u != %u = sf2\n",
549 cd1->storage_flags, cd2->storage_flags);
550 goto out;
551 }
552 ret = -E_BAD_DATA_SIZE;
553 if (cd1->storage_flags & OSL_FIXED_SIZE)
554 if (cd1->data_size != cd2->data_size)
555 goto out;
556 ret = -E_BAD_COLUMN_NAME;
557 if (strcmp(cd1->name, cd2->name))
558 goto out;
559 }
560 PARA_DEBUG_LOG("table description of '%s' matches on-disk data, good\n",
561 t->desc->name);
562 ret = 1;
563 out:
564 FOR_EACH_COLUMN(i, &desc, cd1)
565 free(cd1->name);
566 free(desc.column_descriptions);
567 return ret;
568 }
569
570 static int create_table_index(struct osl_table *t)
571 {
572 char *buf, *filename;
573 int i, ret;
574 size_t size = t->index_header_size;
575 const struct osl_column_description *cd;
576 unsigned offset;
577
578 PARA_INFO_LOG("creating %zu byte index for table %s\n", size,
579 t->desc->name);
580 buf = para_calloc(size);
581 sprintf(buf + IDX_PARA_MAGIC, "%s", PARA_MAGIC);
582 write_u8(buf + IDX_TABLE_FLAGS, t->desc->flags);
583 write_u8(buf + IDX_DIRTY_FLAG, 0);
584 write_u8(buf + IDX_VERSION, CURRENT_TABLE_VERSION);
585 write_u16(buf + IDX_NUM_COLUMNS, t->desc->num_columns);
586 write_u16(buf + IDX_HEADER_SIZE, t->index_header_size);
587 offset = IDX_COLUMN_DESCRIPTIONS;
588 FOR_EACH_COLUMN(i, t->desc, cd) {
589 write_u16(buf + offset + IDX_CD_STORAGE_TYPE,
590 cd->storage_type);
591 write_u16(buf + offset + IDX_CD_STORAGE_FLAGS,
592 cd->storage_flags);
593 if (cd->storage_flags & OSL_FIXED_SIZE)
594 write_u32(buf + offset + IDX_CD_DATA_SIZE,
595 cd->data_size);
596 strcpy(buf + offset + IDX_CD_NAME, cd->name);
597 offset += index_column_description_size(cd->name);
598 }
599 assert(offset = size);
600 filename = index_filename(t->desc);
601 ret = para_write_file(filename, buf, size);
602 free(buf);
603 free(filename);
604 return ret;
605 }
606
607 /**
608 * Create a new osl table.
609 *
610 * \param desc Pointer to the table description.
611 *
612 * \return Standard.
613 */
614 int osl_create_table(const struct osl_table_description *desc)
615 {
616 const struct osl_column_description *cd;
617 char *table_dir = NULL, *filename;
618 struct osl_table *t;
619 int i, ret = init_table_structure(desc, &t);
620
621 if (ret < 0)
622 return ret;
623 PARA_INFO_LOG("creating %s\n", desc->name);
624 FOR_EACH_COLUMN(i, t->desc, cd) {
625 if (cd->storage_type == OSL_NO_STORAGE)
626 continue;
627 if (!table_dir) {
628 ret = para_mkdir(desc->dir, 0777);
629 if (ret < 0 && !is_errno(-ret, EEXIST))
630 goto out;
631 table_dir = make_message("%s/%s", desc->dir,
632 desc->name);
633 ret = para_mkdir(table_dir, 0777);
634 if (ret < 0)
635 goto out;
636 }
637 filename = column_filename(t, i);
638 PARA_INFO_LOG("filename: %s\n", filename);
639 if (cd->storage_type == OSL_MAPPED_STORAGE) {
640 ret = para_open(filename, O_RDWR | O_CREAT | O_EXCL,
641 0644);
642 free(filename);
643 if (ret < 0)
644 goto out;
645 close(ret);
646 continue;
647 }
648 /* DISK STORAGE */
649 ret = para_mkdir(filename, 0777);
650 free(filename);
651 if (ret < 0)
652 goto out;
653 }
654 if (t->num_mapped_columns) {
655 ret = create_table_index(t);
656 if (ret < 0)
657 goto out;
658 }
659 ret = 1;
660 out:
661 free(table_dir);
662 free(t->columns);
663 free(t);
664 return ret;
665 }
666
667 static int table_is_dirty(struct osl_table *t)
668 {
669 char *buf = (char *)t->index_map.data + IDX_DIRTY_FLAG;
670 uint8_t dirty = read_u8(buf) & 0x1;
671 return !!dirty;
672 }
673
674 static void mark_table_dirty(struct osl_table *t)
675 {
676 char *buf = (char *)t->index_map.data + IDX_DIRTY_FLAG;
677 write_u8(buf, read_u8(buf) | 1);
678 }
679
680 static void mark_table_clean(struct osl_table *t)
681 {
682 char *buf = (char *)t->index_map.data + IDX_DIRTY_FLAG;
683 write_u8(buf, read_u8(buf) & 0xfe);
684 }
685
686 static void unmap_column(struct osl_table *t, unsigned col_num)
687 {
688 struct osl_object map = t->columns[col_num].data_map;
689 int ret;
690 if (!map.data)
691 return;
692 ret = para_munmap(map.data, map.size);
693 assert(ret > 0);
694 map.data = NULL;
695 }
696
697 /**
698 * Unmap all mapped files of an osl table.
699 *
700 * \param t Pointer to a mapped table.
701 * \param flags Options for unmapping.
702 *
703 * \return Positive on success, negative on errors.
704 *
705 * \sa map_table(), enum osl_close_flags, para_munmap().
706 */
707 int unmap_table(struct osl_table *t, enum osl_close_flags flags)
708 {
709 unsigned i;
710 const struct osl_column_description *cd;
711 int ret;
712
713 if (!t->num_mapped_columns) /* can this ever happen? */
714 return 1;
715 PARA_DEBUG_LOG("unmapping table '%s'\n", t->desc->name);
716 if (!t->index_map.data)
717 return -E_NOT_MAPPED;
718 if (flags & OSL_MARK_CLEAN)
719 mark_table_clean(t);
720 ret = para_munmap(t->index_map.data, t->index_map.size);
721 if (ret < 0)
722 return ret;
723 t->index_map.data = NULL;
724 if (!t->num_rows)
725 return 1;
726 FOR_EACH_MAPPED_COLUMN(i, t, cd)
727 unmap_column(t, i);
728 return 1;
729 }
730
731 static int map_column(struct osl_table *t, unsigned col_num)
732 {
733 struct stat statbuf;
734 char *filename = column_filename(t, col_num);
735 int ret = -E_STAT;
736 if (stat(filename, &statbuf) < 0) {
737 free(filename);
738 return ret;
739 }
740 if (!(S_IFREG & statbuf.st_mode)) {
741 free(filename);
742 return ret;
743 }
744 ret = mmap_full_file(filename, O_RDWR,
745 &t->columns[col_num].data_map.data,
746 &t->columns[col_num].data_map.size,
747 NULL);
748 free(filename);
749 return ret;
750 }
751
752 /**
753 * Map the index file and all columns of type \p OSL_MAPPED_STORAGE into memory.
754 *
755 * \param t Pointer to an initialized table structure.
756 * \param flags Mapping options.
757 *
758 * \return Negative return value on errors; on success the number of rows
759 * (including invalid rows) is returned.
760 *
761 * \sa unmap_table(), enum map_table_flags, osl_open_table(), mmap(2).
762 */
763 int map_table(struct osl_table *t, enum map_table_flags flags)
764 {
765 char *filename;
766 const struct osl_column_description *cd;
767 int i = 0, ret, num_rows = 0;
768
769 if (!t->num_mapped_columns)
770 return 0;
771 if (t->index_map.data)
772 return -E_ALREADY_MAPPED;
773 filename = index_filename(t->desc);
774 PARA_DEBUG_LOG("mapping table '%s' (index: %s)\n", t->desc->name, filename);
775 ret = mmap_full_file(filename, flags & MAP_TBL_FL_MAP_RDONLY?
776 O_RDONLY : O_RDWR, &t->index_map.data, &t->index_map.size, NULL);
777 free(filename);
778 if (ret < 0)
779 return ret;
780 if (flags & MAP_TBL_FL_VERIFY_INDEX) {
781 ret = compare_table_descriptions(t);
782 if (ret < 0)
783 goto err;
784 }
785 ret = -E_BUSY;
786 if (!(flags & MAP_TBL_FL_IGNORE_DIRTY)) {
787 if (table_is_dirty(t)) {
788 PARA_ERROR_LOG("%s is dirty\n", t->desc->name);
789 goto err;
790 }
791 }
792 mark_table_dirty(t);
793 num_rows = table_num_rows(t);
794 if (!num_rows)
795 return num_rows;
796 /* map data files */
797 FOR_EACH_MAPPED_COLUMN(i, t, cd) {
798 ret = map_column(t, i);
799 if (ret < 0)
800 goto err;
801 }
802 return num_rows;
803 err: /* unmap what is already mapped */
804 for (i--; i >= 0; i--) {
805 struct osl_object map = t->columns[i].data_map;
806 para_munmap(map.data, map.size);
807 map.data = NULL;
808 }
809 para_munmap(t->index_map.data, t->index_map.size);
810 t->index_map.data = NULL;
811 return ret;
812 }
813
814 /**
815 * Retrieve a mapped object by row and column number.
816 *
817 * \param t Pointer to an open osl table.
818 * \param col_num Number of the mapped column containing the object to retrieve.
819 * \param row_num Number of the row containing the object to retrieve.
820 * \param obj The result is returned here.
821 *
822 * It is considered an error if \a col_num does not refer to a column
823 * of storage type \p OSL_MAPPED_STORAGE.
824 *
825 * \return Positive on success, negative on errors. Possible errors include:
826 * \p E_BAD_ROW_NUM, \p E_INVALID_OBJECT.
827 *
828 * \sa osl_storage_type.
829 */
830 int get_mapped_object(const struct osl_table *t, unsigned col_num,
831 uint32_t row_num, struct osl_object *obj)
832 {
833 struct osl_column *col = &t->columns[col_num];
834 uint32_t offset;
835 char *header;
836 char *cell_index;
837 int ret;
838
839 if (t->num_rows <= row_num)
840 return -E_BAD_ROW_NUM;
841 ret = get_cell_index(t, row_num, col_num, &cell_index);
842 if (ret < 0)
843 return ret;
844 offset = read_u32(cell_index);
845 obj->size = read_u32(cell_index + 4) - 1;
846 header = col->data_map.data + offset;
847 obj->data = header + 1;
848 if (read_u8(header) == 0xff) {
849 PARA_ERROR_LOG("col %u, size %zu, offset %u\n", col_num,
850 obj->size, offset);
851 return -E_INVALID_OBJECT;
852 }
853 return 1;
854 }
855
856 static int search_rbtree(const struct osl_object *obj,
857 const struct osl_table *t, unsigned col_num,
858 struct rb_node **result, struct rb_node ***rb_link)
859 {
860 struct osl_column *col = &t->columns[col_num];
861 struct rb_node **new = &col->rbtree.rb_node, *parent = NULL;
862 const struct osl_column_description *cd =
863 get_column_description(t->desc, col_num);
864 enum osl_storage_type st = cd->storage_type;
865 while (*new) {
866 struct osl_row *this_row = get_row_pointer(*new,
867 col->rbtree_num);
868 int ret;
869 struct osl_object this_obj;
870 parent = *new;
871 if (st == OSL_MAPPED_STORAGE) {
872 ret = get_mapped_object(t, col_num, this_row->num,
873 &this_obj);
874 if (ret < 0)
875 return ret;
876 } else
877 this_obj = this_row->volatile_objects[col->volatile_num];
878 ret = cd->compare_function(obj, &this_obj);
879 if (!ret) {
880 if (result)
881 *result = get_rb_node_pointer(this_row,
882 col->rbtree_num);
883 return 1;
884 }
885 if (ret < 0)
886 new = &((*new)->rb_left);
887 else
888 new = &((*new)->rb_right);
889 }
890 if (result)
891 *result = parent;
892 if (rb_link)
893 *rb_link = new;
894 return -E_RB_KEY_NOT_FOUND;
895 }
896
897 static int insert_rbtree(struct osl_table *t, unsigned col_num,
898 const struct osl_row *row, const struct osl_object *obj)
899 {
900 struct rb_node *parent, **rb_link;
901 unsigned rbtree_num;
902 struct rb_node *n;
903 int ret = search_rbtree(obj, t, col_num, &parent, &rb_link);
904
905 if (ret > 0)
906 return -E_RB_KEY_EXISTS;
907 rbtree_num = t->columns[col_num].rbtree_num;
908 n = get_rb_node_pointer(row, rbtree_num);
909 rb_link_node(n, parent, rb_link);
910 rb_insert_color(n, &t->columns[col_num].rbtree);
911 return 1;
912 }
913
914 static void remove_rb_node(struct osl_table *t, unsigned col_num,
915 const struct osl_row *row)
916 {
917 struct osl_column *col = &t->columns[col_num];
918 const struct osl_column_description *cd =
919 get_column_description(t->desc, col_num);
920 enum osl_storage_flags sf = cd->storage_flags;
921 struct rb_node *victim, *splice_out_node, *tmp;
922 if (!(sf & OSL_RBTREE))
923 return;
924 /*
925 * Which node is removed/spliced out actually depends on how many
926 * children the victim node has: If it has no children, it gets
927 * deleted. If it has one child, it gets spliced out. If it has two
928 * children, its successor (which has at most a right child) gets
929 * spliced out.
930 */
931 victim = get_rb_node_pointer(row, col->rbtree_num);
932 if (victim->rb_left && victim->rb_right)
933 splice_out_node = rb_next(victim);
934 else
935 splice_out_node = victim;
936 /* Go up to the root and decrement the size of each node in the path. */
937 for (tmp = splice_out_node; tmp; tmp = rb_parent(tmp))
938 tmp->size--;
939 rb_erase(victim, &col->rbtree);
940 }
941
942 static int add_row_to_rbtrees(struct osl_table *t, uint32_t row_num,
943 struct osl_object *volatile_objs, struct osl_row **row_ptr)
944 {
945 unsigned i;
946 int ret;
947 struct osl_row *row = allocate_row(t->num_rbtrees);
948 const struct osl_column_description *cd;
949
950 row->num = row_num;
951 row->volatile_objects = volatile_objs;
952 FOR_EACH_RBTREE_COLUMN(i, t, cd) {
953 if (cd->storage_type == OSL_MAPPED_STORAGE) {
954 struct osl_object obj;
955 ret = get_mapped_object(t, i, row_num, &obj);
956 if (ret < 0)
957 goto err;
958 ret = insert_rbtree(t, i, row, &obj);
959 } else { /* volatile */
960 const struct osl_object *obj
961 = volatile_objs + t->columns[i].volatile_num;
962 ret = insert_rbtree(t, i, row, obj);
963 }
964 if (ret < 0)
965 goto err;
966 }
967 if (row_ptr)
968 *row_ptr = row;
969 return 1;
970 err: /* rollback changes, i.e. remove added entries from rbtrees */
971 while (i)
972 remove_rb_node(t, i--, row);
973 free(row);
974 return ret;
975 }
976
977 static void free_volatile_objects(const struct osl_table *t,
978 enum osl_close_flags flags)
979 {
980 int i, j;
981 struct rb_node *n;
982 struct osl_column *rb_col;
983 const struct osl_column_description *cd;
984
985 if (!t->num_volatile_columns)
986 return;
987 /* find the first rbtree column (any will do) */
988 FOR_EACH_RBTREE_COLUMN(i, t, cd)
989 break;
990 rb_col = t->columns + i;
991 /* walk that rbtree and free all volatile objects */
992 for (n = rb_first(&rb_col->rbtree); n; n = rb_next(n)) {
993 struct osl_row *r = get_row_pointer(n, rb_col->rbtree_num);
994 if (flags & OSL_FREE_VOLATILE)
995 FOR_EACH_VOLATILE_COLUMN(j, t, cd) {
996 if (cd->storage_flags & OSL_DONT_FREE)
997 continue;
998 free(r->volatile_objects[
999 t->columns[j].volatile_num].data);
1000 }
1001 // for (j = 0; j < t->num_volatile_columns; j++)
1002 // free(r->volatile_objects[j].data);
1003 free(r->volatile_objects);
1004 }
1005 }
1006
1007 /**
1008 * Erase all rbtree nodes and free resources.
1009 *
1010 * \param t Pointer to an open osl table.
1011 *
1012 * This function is called by osl_close_table().
1013 */
1014 void clear_rbtrees(struct osl_table *t)
1015 {
1016 const struct osl_column_description *cd;
1017 unsigned i, rbtrees_cleared = 0;
1018
1019 FOR_EACH_RBTREE_COLUMN(i, t, cd) {
1020 struct osl_column *col = &t->columns[i];
1021 struct rb_node *n;
1022 rbtrees_cleared++;
1023 for (n = rb_first(&col->rbtree); n;) {
1024 struct osl_row *r;
1025 rb_erase(n, &col->rbtree);
1026 if (rbtrees_cleared == t->num_rbtrees) {
1027 r = get_row_pointer(n, col->rbtree_num);
1028 n = rb_next(n);
1029 free(r);
1030 } else
1031 n = rb_next(n);
1032 }
1033 }
1034
1035 }
1036
1037 /**
1038 * Close an osl table.
1039 *
1040 * \param t Pointer to the table to be closed.
1041 * \param flags Options for what should be cleaned up.
1042 *
1043 * If osl_open_table() succeeds, the resulting table pointer must later be
1044 * passed to this function in order to flush all changes to the file system and
1045 * to free the resources that were allocated by osl_open_table().
1046 *
1047 * \return Positive on success, negative on errors. Possible errors: \p E_BAD_TABLE,
1048 * errors returned by unmap_table().
1049 *
1050 * \sa osl_open_table(), unmap_table().
1051 */
1052 int osl_close_table(struct osl_table *t, enum osl_close_flags flags)
1053 {
1054 int ret;
1055
1056 if (!t)
1057 return -E_BAD_TABLE;
1058 free_volatile_objects(t, flags);
1059 clear_rbtrees(t);
1060 ret = unmap_table(t, flags);
1061 if (ret < 0)
1062 PARA_ERROR_LOG("unmap_table failed: %d\n", ret);
1063 free(t->columns);
1064 free(t);
1065 return ret;
1066 }
1067
1068 /**
1069 * Find out whether the given row number corresponds to an invalid row.
1070 *
1071 * \param t Pointer to the osl table.
1072 * \param row_num The number of the row in question.
1073 *
1074 * By definition, a row is considered invalid if all its index entries
1075 * are invalid.
1076 *
1077 * \return Positive if \a row_num corresponds to an invalid row,
1078 * zero if it corresponds to a valid row, negative on errors.
1079 */
1080 int row_is_invalid(struct osl_table *t, uint32_t row_num)
1081 {
1082 char *row_index;
1083 int i, ret = get_row_index(t, row_num, &row_index);
1084
1085 if (ret < 0)
1086 return ret;
1087 for (i = 0; i < t->row_index_size; i++) {
1088 if ((unsigned char)row_index[i] != 0xff)
1089 return 0;
1090 }
1091 PARA_INFO_LOG("row %d is invalid\n", row_num);
1092 return 1;
1093 }
1094
1095 /**
1096 * Invalidate a row of an osl table.
1097 *
1098 * \param t Pointer to an open osl table.
1099 * \param row_num Number of the row to mark as invalid.
1100 *
1101 * This function marks each mapped object in the index entry of \a row as
1102 * invalid.
1103 *
1104 * \return Positive on success, negative on errors.
1105 */
1106 int mark_row_invalid(struct osl_table *t, uint32_t row_num)
1107 {
1108 char *row_index;
1109 int ret = get_row_index(t, row_num, &row_index);
1110
1111 if (ret < 0)
1112 return ret;
1113 PARA_INFO_LOG("marking row %d as invalid\n", row_num);
1114 memset(row_index, 0xff, t->row_index_size);
1115 return 1;
1116 }
1117
1118 /**
1119 * Initialize all rbtrees and compute number of invalid rows.
1120 *
1121 * \param t The table containing the rbtrees to be initialized.
1122 *
1123 * \return Positive on success, negative on errors.
1124 */
1125 int init_rbtrees(struct osl_table *t)
1126 {
1127 int i, ret;
1128 const struct osl_column_description *cd;
1129
1130 /* create rbtrees */
1131 FOR_EACH_RBTREE_COLUMN(i, t, cd)
1132 t->columns[i].rbtree = RB_ROOT;
1133 /* add valid rows to rbtrees */
1134 t->num_invalid_rows = 0;
1135 for (i = 0; i < t->num_rows; i++) {
1136 ret = row_is_invalid(t, i);
1137 if (ret < 0)
1138 return ret;
1139 if (ret) {
1140 t->num_invalid_rows++;
1141 continue;
1142 }
1143 ret = add_row_to_rbtrees(t, i, NULL, NULL);
1144 if (ret < 0)
1145 return ret;
1146 }
1147 return 1;
1148 }
1149
1150 /**
1151 * Open an osl table.
1152 *
1153 * Each osl table must be opened before its data can be accessed.
1154 *
1155 * \param table_desc Describes the table to be opened.
1156 * \param result Contains a pointer to the open table on success.
1157 *
1158 * The table description given by \a desc should coincide with the
1159 * description used at creation time.
1160 *
1161 * \return Standard.
1162 */
1163 int osl_open_table(const struct osl_table_description *table_desc,
1164 struct osl_table **result)
1165 {
1166 int i, ret;
1167 struct osl_table *t;
1168 const struct osl_column_description *cd;
1169
1170 PARA_INFO_LOG("opening table %s\n", table_desc->name);
1171 ret = init_table_structure(table_desc, &t);
1172 if (ret < 0)
1173 return ret;
1174 FOR_EACH_DISK_STORAGE_COLUMN(i, t, cd) {
1175 /* check if directory exists */
1176 char *dirname = column_filename(t, i);
1177 struct stat statbuf;
1178 ret = stat(dirname, &statbuf);
1179 free(dirname);
1180 if (ret < 0) {
1181 ret = -ERRNO_TO_PARA_ERROR(errno);
1182 goto err;
1183 }
1184 ret = -ERRNO_TO_PARA_ERROR(ENOTDIR);
1185 if (!S_ISDIR(statbuf.st_mode))
1186 goto err;
1187 }
1188 ret = map_table(t, MAP_TBL_FL_VERIFY_INDEX);
1189 if (ret < 0)
1190 goto err;
1191 t->num_rows = ret;
1192 PARA_DEBUG_LOG("num rows: %d\n", t->num_rows);
1193 ret = init_rbtrees(t);
1194 if (ret < 0) {
1195 osl_close_table(t, OSL_MARK_CLEAN); /* ignore further errors */
1196 return ret;
1197 }
1198 *result = t;
1199 return 1;
1200 err:
1201 free(t->columns);
1202 free(t);
1203 return ret;
1204 }
1205
1206 static int create_disk_storage_object_dir(const struct osl_table *t,
1207 unsigned col_num, const char *ds_name)
1208 {
1209 char *dirname;
1210 int ret;
1211
1212 if (!(t->desc->flags & OSL_LARGE_TABLE))
1213 return 1;
1214 dirname = disk_storage_dirname(t, col_num, ds_name);
1215 ret = para_mkdir(dirname, 0777);
1216 free(dirname);
1217 if (ret < 0 && !is_errno(-ret, EEXIST))
1218 return ret;
1219 return 1;
1220 }
1221
1222 static int write_disk_storage_file(const struct osl_table *t, unsigned col_num,
1223 const struct osl_object *obj, const char *ds_name)
1224 {
1225 int ret;
1226 char *filename;
1227
1228 ret = create_disk_storage_object_dir(t, col_num, ds_name);
1229 if (ret < 0)
1230 return ret;
1231 filename = disk_storage_path(t, col_num, ds_name);
1232 ret = para_write_file(filename, obj->data, obj->size);
1233 free(filename);
1234 return ret;
1235 }
1236
1237 static int append_map_file(const struct osl_table *t, unsigned col_num,
1238 const struct osl_object *obj, uint32_t *new_size)
1239 {
1240 char *filename = column_filename(t, col_num);
1241 int ret;
1242 char header = 0; /* zero means valid object */
1243
1244 // PARA_DEBUG_LOG("appending %zu + 1 byte\n", obj->size);
1245 ret = append_file(filename, &header, 1, obj->data, obj->size,
1246 new_size);
1247 free(filename);
1248 return ret;
1249 }
1250
1251 static int append_row_index(const struct osl_table *t, char *row_index)
1252 {
1253 char *filename;
1254 int ret;
1255
1256 if (!t->num_mapped_columns)
1257 return 1;
1258 filename = index_filename(t->desc);
1259 ret = append_file(filename, NULL, 0, row_index,
1260 t->row_index_size, NULL);
1261 free(filename);
1262 return ret;
1263 }
1264
1265 /**
1266 * A wrapper for truncate(2)
1267 *
1268 * \param path Name of the regular file to truncate
1269 * \param size Number of bytes to \b shave \b off
1270 *
1271 * Truncate the regular file named by \a path by \a size bytes.
1272 *
1273 * \return Positive on success, negative on errors. Possible errors include: \p
1274 * E_STAT, \p E_BAD_SIZE, \p E_TRUNC.
1275 *
1276 * \sa truncate(2)
1277 */
1278 int para_truncate(const char *path, off_t size)
1279 {
1280 int ret;
1281 struct stat statbuf;
1282
1283 ret = -E_STAT;
1284 if (stat(path, &statbuf) < 0)
1285 goto out;
1286 ret = -E_BAD_SIZE;
1287 if (statbuf.st_size < size)
1288 goto out;
1289 ret = -E_TRUNC;
1290 if (truncate(path, statbuf.st_size - size) < 0)
1291 goto out;
1292 ret = 1;
1293 out:
1294 return ret;
1295 }
1296
1297 static int truncate_mapped_file(const struct osl_table *t, unsigned col_num,
1298 off_t size)
1299 {
1300 char *filename = column_filename(t, col_num);
1301 int ret = para_truncate(filename, size);
1302 free(filename);
1303 return ret;
1304 }
1305
1306 static int delete_disk_storage_file(const struct osl_table *t, unsigned col_num,
1307 const char *ds_name)
1308 {
1309 char *dirname, *filename = disk_storage_path(t, col_num, ds_name);
1310 int ret = unlink(filename), err = errno;
1311
1312 free(filename);
1313 if (ret < 0)
1314 return -ERRNO_TO_PARA_ERROR(err);
1315 if (!(t->desc->flags & OSL_LARGE_TABLE))
1316 return 1;
1317 dirname = disk_storage_dirname(t, col_num, ds_name);
1318 rmdir(dirname);
1319 free(dirname);
1320 return 1;
1321 }
1322
1323 /**
1324 * Add a new row to an osl table and retrieve this row.
1325 *
1326 * \param t Pointer to an open osl table.
1327 * \param objects Array of objects to be added.
1328 * \param row Result pointer.
1329 *
1330 * The \a objects parameter must point to an array containing one object per
1331 * column. The order of the objects in the array is given by the table
1332 * description of \a table. Several sanity checks are performed during object
1333 * insertion and the function returns without modifying the table if any of
1334 * these tests fail. In fact, it is atomic in the sense that it either
1335 * succeeds or leaves the table unchanged (i.e. either all or none of the
1336 * objects are added to the table).
1337 *
1338 * It is considered an error if an object is added to a column with associated
1339 * rbtree if this object is equal to an object already contained in that column
1340 * (i.e. the compare function for the column's rbtree returns zero).
1341 *
1342 * Possible errors include: \p E_RB_KEY_EXISTS, \p E_BAD_DATA_SIZE.
1343 *
1344 * \return Positive on success, negative on errors.
1345 *
1346 * \sa struct osl_table_description, osl_compare_func, osl_add_row().
1347 */
1348 int osl_add_and_get_row(struct osl_table *t, struct osl_object *objects,
1349 struct osl_row **row)
1350 {
1351 int i, ret;
1352 char *ds_name = NULL;
1353 struct rb_node **rb_parents = NULL, ***rb_links = NULL;
1354 char *new_row_index = NULL;
1355 struct osl_object *volatile_objs = NULL;
1356 const struct osl_column_description *cd;
1357
1358 if (!t)
1359 return -E_BAD_TABLE;
1360 rb_parents = para_malloc(t->num_rbtrees * sizeof(struct rn_node*));
1361 rb_links = para_malloc(t->num_rbtrees * sizeof(struct rn_node**));
1362 if (t->num_mapped_columns)
1363 new_row_index = para_malloc(t->row_index_size);
1364 /* pass 1: sanity checks */
1365 // PARA_DEBUG_LOG("sanity tests: %p:%p\n", objects[0].data,
1366 // objects[1].data);
1367 FOR_EACH_COLUMN(i, t->desc, cd) {
1368 enum osl_storage_type st = cd->storage_type;
1369 enum osl_storage_flags sf = cd->storage_flags;
1370
1371 // ret = -E_NULL_OBJECT;
1372 // if (!objects[i])
1373 // goto out;
1374 if (st == OSL_DISK_STORAGE)
1375 continue;
1376 if (sf & OSL_RBTREE) {
1377 unsigned rbtree_num = t->columns[i].rbtree_num;
1378 ret = -E_RB_KEY_EXISTS;
1379 // PARA_DEBUG_LOG("checking whether %p exists\n",
1380 // objects[i].data);
1381 if (search_rbtree(objects + i, t, i,
1382 &rb_parents[rbtree_num],
1383 &rb_links[rbtree_num]) > 0)
1384 goto out;
1385 }
1386 if (sf & OSL_FIXED_SIZE) {
1387 // PARA_DEBUG_LOG("fixed size. need: %zu, have: %d\n",
1388 // objects[i].size, cd->data_size);
1389 ret = -E_BAD_DATA_SIZE;
1390 if (objects[i].size != cd->data_size)
1391 goto out;
1392 }
1393 }
1394 if (t->num_disk_storage_columns)
1395 ds_name = disk_storage_name_of_object(t,
1396 &objects[t->disk_storage_name_column]);
1397 ret = unmap_table(t, OSL_MARK_CLEAN);
1398 if (ret < 0)
1399 goto out;
1400 // PARA_DEBUG_LOG("sanity tests passed%s\n", "");
1401 /* pass 2: create data files, append map data */
1402 FOR_EACH_COLUMN(i, t->desc, cd) {
1403 enum osl_storage_type st = cd->storage_type;
1404 if (st == OSL_NO_STORAGE)
1405 continue;
1406 if (st == OSL_MAPPED_STORAGE) {
1407 uint32_t new_size;
1408 struct osl_column *col = &t->columns[i];
1409 // PARA_DEBUG_LOG("appending object of size %zu\n",
1410 // objects[i].size);
1411 ret = append_map_file(t, i, objects + i, &new_size);
1412 if (ret < 0)
1413 goto rollback;
1414 update_cell_index(new_row_index, col, new_size,
1415 objects[i].size);
1416 continue;
1417 }
1418 /* DISK_STORAGE */
1419 ret = write_disk_storage_file(t, i, objects + i, ds_name);
1420 if (ret < 0)
1421 goto rollback;
1422 }
1423 ret = append_row_index(t, new_row_index);
1424 if (ret < 0)
1425 goto rollback;
1426 ret = map_table(t, MAP_TBL_FL_VERIFY_INDEX);
1427 if (ret < 0) { /* truncate index and rollback changes */
1428 char *filename = index_filename(t->desc);
1429 para_truncate(filename, t->row_index_size);
1430 free(filename);
1431 goto rollback;
1432 }
1433 /* pass 3: add entry to rbtrees */
1434 if (t->num_volatile_columns) {
1435 volatile_objs = para_calloc(t->num_volatile_columns
1436 * sizeof(struct osl_object));
1437 FOR_EACH_VOLATILE_COLUMN(i, t, cd)
1438 volatile_objs[t->columns[i].volatile_num] = objects[i];
1439 }
1440 t->num_rows++;
1441 // PARA_DEBUG_LOG("adding new entry as row #%d\n", t->num_rows - 1);
1442 ret = add_row_to_rbtrees(t, t->num_rows - 1, volatile_objs, row);
1443 if (ret < 0)
1444 goto out;
1445 // PARA_DEBUG_LOG("added new entry as row #%d\n", t->num_rows - 1);
1446 ret = 1;
1447 goto out;
1448 rollback: /* rollback all changes made, ignore further errors */
1449 for (i--; i >= 0; i--) {
1450 cd = get_column_description(t->desc, i);
1451 enum osl_storage_type st = cd->storage_type;
1452 if (st == OSL_NO_STORAGE)
1453 continue;
1454
1455 if (st == OSL_MAPPED_STORAGE)
1456 truncate_mapped_file(t, i, objects[i].size);
1457 else /* disk storage */
1458 delete_disk_storage_file(t, i, ds_name);
1459 }
1460 /* ignore error and return previous error value */
1461 map_table(t, MAP_TBL_FL_VERIFY_INDEX);
1462 out:
1463 free(new_row_index);
1464 free(ds_name);
1465 free(rb_parents);
1466 free(rb_links);
1467 return ret;
1468 }
1469
1470 /**
1471 * Add a new row to an osl table.
1472 *
1473 * \param t Same meaning as osl_add_and_get_row().
1474 * \param objects Same meaning as osl_add_and_get_row().
1475 *
1476 * \return The return value of the underlying call to osl_add_and_get_row().
1477 *
1478 * This is equivalent to osl_add_and_get_row(t, objects, NULL).
1479 */
1480 int osl_add_row(struct osl_table *t, struct osl_object *objects)
1481 {
1482 return osl_add_and_get_row(t, objects, NULL);
1483 }
1484
1485 /**
1486 * Retrieve an object identified by row and column
1487 *
1488 * \param t Pointer to an open osl table.
1489 * \param r Pointer to the row.
1490 * \param col_num The column number.
1491 * \param object The result pointer.
1492 *
1493 * The column determined by \a col_num must be of type \p OSL_MAPPED_STORAGE
1494 * or \p OSL_NO_STORAGE, i.e. no disk storage objects may be retrieved by this
1495 * function.
1496 *
1497 * \return Positive if object was found, negative on errors. Possible errors
1498 * include: \p E_BAD_TABLE, \p E_BAD_STORAGE_TYPE.
1499 *
1500 * \sa osl_storage_type, osl_open_disk_object().
1501 */
1502 int osl_get_object(const struct osl_table *t, const struct osl_row *r,
1503 unsigned col_num, struct osl_object *object)
1504 {
1505 const struct osl_column_description *cd;
1506
1507 if (!t)
1508 return -E_BAD_TABLE;
1509 cd = get_column_description(t->desc, col_num);
1510 /* col must not be disk storage */
1511 if (cd->storage_type == OSL_DISK_STORAGE)
1512 return -E_BAD_STORAGE_TYPE;
1513 if (cd->storage_type == OSL_MAPPED_STORAGE)
1514 return get_mapped_object(t, col_num, r->num, object);
1515 /* volatile */
1516 *object = r->volatile_objects[t->columns[col_num].volatile_num];
1517 return 1;
1518 }
1519
1520 static int mark_mapped_object_invalid(const struct osl_table *t,
1521 uint32_t row_num, unsigned col_num)
1522 {
1523 struct osl_object obj;
1524 char *p;
1525 int ret = get_mapped_object(t, col_num, row_num, &obj);
1526
1527 if (ret < 0)
1528 return ret;
1529 p = obj.data;
1530 p--;
1531 *p = 0xff;
1532 return 1;
1533 }
1534
1535 /**
1536 * Delete a row from an osl table.
1537 *
1538 * \param t Pointer to an open osl table.
1539 * \param row Pointer to the row to delete.
1540 *
1541 * This removes all disk storage objects, removes all rbtree nodes, and frees
1542 * all volatile objects belonging to the given row. For mapped columns, the
1543 * data is merely marked invalid and may be pruned from time to time by
1544 * para_fsck.
1545 *
1546 * \return Positive on success, negative on errors. Possible errors include:
1547 * \p E_BAD_TABLE, errors returned by osl_get_object().
1548 */
1549 int osl_del_row(struct osl_table *t, struct osl_row *row)
1550 {
1551 struct osl_row *r = row;
1552 int i, ret;
1553 const struct osl_column_description *cd;
1554
1555 if (!t)
1556 return -E_BAD_TABLE;
1557 PARA_INFO_LOG("deleting row %p\n", row);
1558
1559 if (t->num_disk_storage_columns) {
1560 char *ds_name;
1561 ret = disk_storage_name_of_row(t, r, &ds_name);
1562 if (ret < 0)
1563 goto out;
1564 FOR_EACH_DISK_STORAGE_COLUMN(i, t, cd)
1565 delete_disk_storage_file(t, i, ds_name);
1566 free(ds_name);
1567 }
1568 FOR_EACH_COLUMN(i, t->desc, cd) {
1569 struct osl_column *col = t->columns + i;
1570 enum osl_storage_type st = cd->storage_type;
1571 remove_rb_node(t, i, r);
1572 if (st == OSL_MAPPED_STORAGE) {
1573 mark_mapped_object_invalid(t, r->num, i);
1574 continue;
1575 }
1576 if (st == OSL_NO_STORAGE && !(cd->storage_flags & OSL_DONT_FREE))
1577 free(r->volatile_objects[col->volatile_num].data);
1578 }
1579 if (t->num_mapped_columns) {
1580 ret = mark_row_invalid(t, r->num);
1581 if (ret < 0)
1582 goto out;
1583 t->num_invalid_rows++;
1584 } else
1585 t->num_rows--;
1586 ret = 1;
1587 out:
1588 free(r->volatile_objects);
1589 free(r);
1590 return ret;
1591 }
1592
1593 /* test if column has an rbtree */
1594 static int check_rbtree_col(const struct osl_table *t, unsigned col_num,
1595 struct osl_column **col)
1596 {
1597 if (!t)
1598 return -E_BAD_TABLE;
1599 if (!(get_column_description(t->desc, col_num)->storage_flags & OSL_RBTREE))
1600 return -E_BAD_STORAGE_FLAGS;
1601 *col = t->columns + col_num;
1602 return 1;
1603 }
1604
1605 /**
1606 * Get the row that contains the given object.
1607 *
1608 * \param t Pointer to an open osl table.
1609 * \param col_num The number of the column to be searched.
1610 * \param obj The object to be looked up.
1611 * \param result Points to the row containing \a obj.
1612 *
1613 * Lookup \a obj in \a t and return the row containing \a obj. The column
1614 * specified by \a col_num must have an associated rbtree.
1615 *
1616 * \return Positive on success, negative on errors. If an error occurred, \a
1617 * result is set to \p NULL. Possible errors include: \p E_BAD_TABLE, \p
1618 * E_BAD_STORAGE_FLAGS, errors returned by get_mapped_object(), \p
1619 * E_RB_KEY_NOT_FOUND.
1620 *
1621 * \sa osl_storage_flags
1622 */
1623 int osl_get_row(const struct osl_table *t, unsigned col_num,
1624 const struct osl_object *obj, struct osl_row **result)
1625 {
1626 int ret;
1627 struct rb_node *node;
1628 struct osl_row *row;
1629 struct osl_column *col;
1630
1631 *result = NULL;
1632 ret = check_rbtree_col(t, col_num, &col);
1633 if (ret < 0)
1634 return ret;
1635 ret = search_rbtree(obj, t, col_num, &node, NULL);
1636 if (ret < 0)
1637 return ret;
1638 row = get_row_pointer(node, t->columns[col_num].rbtree_num);
1639 *result = row;
1640 return 1;
1641 }
1642
1643 static int rbtree_loop(struct osl_column *col, void *private_data,
1644 osl_rbtree_loop_func *func)
1645 {
1646 struct rb_node *n, *tmp;
1647
1648 /* this for-loop is safe against removal of an entry */
1649 for (n = rb_first(&col->rbtree), tmp = n? rb_next(n) : NULL;
1650 n;
1651 n = tmp, tmp = tmp? rb_next(tmp) : NULL) {
1652 struct osl_row *r = get_row_pointer(n, col->rbtree_num);
1653 int ret = func(r, private_data);
1654 if (ret < 0)
1655 return ret;
1656 }
1657 return 1;
1658 }
1659
1660 static int rbtree_loop_reverse(struct osl_column *col, void *private_data,
1661 osl_rbtree_loop_func *func)
1662 {
1663 struct rb_node *n, *tmp;
1664
1665 /* safe against removal of an entry */
1666 for (n = rb_last(&col->rbtree), tmp = n? rb_prev(n) : NULL;
1667 n;
1668 n = tmp, tmp = tmp? rb_prev(tmp) : NULL) {
1669 struct osl_row *r = get_row_pointer(n, col->rbtree_num);
1670 int ret = func(r, private_data);
1671 if (ret < 0)
1672 return ret;
1673 }
1674 return 1;
1675 }
1676
1677 /**
1678 * Loop over all nodes in an rbtree.
1679 *
1680 * \param t Pointer to an open osl table.
1681 * \param col_num The column to use for iterating over the elements.
1682 * \param private_data Pointer that gets passed to \a func.
1683 * \param func The function to be called for each node in the rbtree.
1684 *
1685 * This function does an in-order walk of the rbtree associated with \a
1686 * col_num. It is an error if the \p OSL_RBTREE flag is not set for this
1687 * column. For each node in the rbtree, the given function \a func is called
1688 * with two pointers as arguments: The first osl_row* argument points to the
1689 * row that contains the object corresponding to the rbtree node currently
1690 * traversed, and the \a private_data pointer is passed verbatim to \a func as the
1691 * second argument. The loop terminates either if \a func returns a negative
1692 * value, or if all nodes of the tree have been visited.
1693 *
1694 *
1695 * \return Positive on success, negative on errors. If the termination of the
1696 * loop was caused by \a func returning a negative value, this value is
1697 * returned.
1698 *
1699 * \sa osl_storage_flags, osl_rbtree_loop_reverse(), osl_compare_func.
1700 */
1701 int osl_rbtree_loop(const struct osl_table *t, unsigned col_num,
1702 void *private_data, osl_rbtree_loop_func *func)
1703 {
1704 struct osl_column *col;
1705
1706 int ret = check_rbtree_col(t, col_num, &col);
1707 if (ret < 0)
1708 return ret;
1709 return rbtree_loop(col, private_data, func);
1710 }
1711
1712 /**
1713 * Loop over all nodes in an rbtree in reverse order.
1714 *
1715 * \param t Identical meaning as in \p osl_rbtree_loop().
1716 * \param col_num Identical meaning as in \p osl_rbtree_loop().
1717 * \param private_data Identical meaning as in \p osl_rbtree_loop().
1718 * \param func Identical meaning as in \p osl_rbtree_loop().
1719 *
1720 * This function is identical to \p osl_rbtree_loop(), the only difference
1721 * is that the tree is walked in reverse order.
1722 *
1723 * \return The same return value as \p osl_rbtree_loop().
1724 *
1725 * \sa osl_rbtree_loop().
1726 */
1727 int osl_rbtree_loop_reverse(const struct osl_table *t, unsigned col_num,
1728 void *private_data, osl_rbtree_loop_func *func)
1729 {
1730 struct osl_column *col;
1731
1732 int ret = check_rbtree_col(t, col_num, &col);
1733 if (ret < 0)
1734 return ret;
1735 return rbtree_loop_reverse(col, private_data, func);
1736 }
1737
1738 /* TODO: Rollback changes on errors */
1739 static int rename_disk_storage_objects(struct osl_table *t,
1740 struct osl_object *old_obj, struct osl_object *new_obj)
1741 {
1742 int i, ret;
1743 const struct osl_column_description *cd;
1744 char *old_ds_name, *new_ds_name;
1745
1746 if (!t->num_disk_storage_columns)
1747 return 1; /* nothing to do */
1748 if (old_obj->size == new_obj->size && !memcmp(new_obj->data,
1749 old_obj->data, new_obj->size))
1750 return 1; /* object did not change */
1751 old_ds_name = disk_storage_name_of_object(t, old_obj);
1752 new_ds_name = disk_storage_name_of_object(t, new_obj);
1753 FOR_EACH_DISK_STORAGE_COLUMN(i, t, cd) {
1754 char *old_filename, *new_filename;
1755 ret = create_disk_storage_object_dir(t, i, new_ds_name);
1756 if (ret < 0)
1757 goto out;
1758 old_filename = disk_storage_path(t, i, old_ds_name);
1759 new_filename = disk_storage_path(t, i, new_ds_name);
1760 ret = para_rename(old_filename, new_filename);
1761 free(old_filename);
1762 free(new_filename);
1763 if (ret < 0)
1764 goto out;
1765 }
1766 ret = 1;
1767 out:
1768 free(old_ds_name);
1769 free(new_ds_name);
1770 return ret;
1771
1772 }
1773
1774 /**
1775 * Change an object in an osl table.
1776 *
1777 * \param t Pointer to an open osl table.
1778 * \param r Pointer to the row containing the object to be updated.
1779 * \param col_num Number of the column containing the object to be updated.
1780 * \param obj Pointer to the replacement object.
1781 *
1782 * This function gets rid of all references to the old object. This includes
1783 * removal of the rbtree node in case there is an rbtree associated with \a
1784 * col_num. It then inserts \a obj into the table and the rbtree if necessary.
1785 *
1786 * If the \p OSL_RBTREE flag is set for \a col_num, you \b MUST call this
1787 * function in order to change the contents of an object, even for volatile or
1788 * mapped columns of constant size (which may be updated directly if \p
1789 * OSL_RBTREE is not set). Otherwise the rbtree might become corrupted.
1790 *
1791 * \return Standard
1792 */
1793 int osl_update_object(struct osl_table *t, const struct osl_row *r,
1794 unsigned col_num, struct osl_object *obj)
1795 {
1796 struct osl_column *col;
1797 const struct osl_column_description *cd;
1798 int ret;
1799
1800 if (!t)
1801 return -E_BAD_TABLE;
1802 col = &t->columns[col_num];
1803 cd = get_column_description(t->desc, col_num);
1804 PARA_DEBUG_LOG("updating column %u of %s\n", col_num, t->desc->name);
1805 if (cd->storage_flags & OSL_RBTREE) {
1806 if (search_rbtree(obj, t, col_num, NULL, NULL) > 0)
1807 return -E_RB_KEY_EXISTS;
1808 }
1809 if (cd->storage_flags & OSL_FIXED_SIZE) {
1810 if (obj->size != cd->data_size)
1811 return -E_BAD_DATA_SIZE;
1812 }
1813 remove_rb_node(t, col_num, r);
1814 if (cd->storage_type == OSL_NO_STORAGE) { /* TODO: If fixed size, reuse object? */
1815 free(r->volatile_objects[col->volatile_num].data);
1816 r->volatile_objects[col->volatile_num] = *obj;
1817 } else if (cd->storage_type == OSL_DISK_STORAGE) {
1818 char *ds_name;
1819 ret = disk_storage_name_of_row(t, r, &ds_name);
1820 if (ret < 0)
1821 return ret;
1822 ret = delete_disk_storage_file(t, col_num, ds_name);
1823 if (ret < 0 && !is_errno(-ret, ENOENT)) {
1824 free(ds_name);
1825 return ret;
1826 }
1827 ret = write_disk_storage_file(t, col_num, obj, ds_name);
1828 free(ds_name);
1829 if (ret < 0)
1830 return ret;
1831 } else { /* mapped storage */
1832 struct osl_object old_obj;
1833 ret = get_mapped_object(t, col_num, r->num, &old_obj);
1834 if (ret < 0)
1835 return ret;
1836 /*
1837 * If the updated column is the disk storage name column, the
1838 * disk storage name changes, so we have to rename all disk
1839 * storage objects accordingly.
1840 */
1841 if (col_num == t->disk_storage_name_column) {
1842 ret = rename_disk_storage_objects(t, &old_obj, obj);
1843 if (ret < 0)
1844 return ret;
1845 }
1846 if (cd->storage_flags & OSL_FIXED_SIZE)
1847 memcpy(old_obj.data, obj->data, cd->data_size);
1848 else { /* TODO: if the size doesn't change, use old space */
1849 uint32_t new_data_map_size;
1850 char *row_index;
1851 ret = get_row_index(t, r->num, &row_index);
1852 if (ret < 0)
1853 return ret;
1854 ret = mark_mapped_object_invalid(t, r->num, col_num);
1855 if (ret < 0)
1856 return ret;
1857 unmap_column(t, col_num);
1858 ret = append_map_file(t, col_num, obj,
1859 &new_data_map_size);
1860 if (ret < 0)
1861 return ret;
1862 ret = map_column(t, col_num);
1863 if (ret < 0)
1864 return ret;
1865 update_cell_index(row_index, col, new_data_map_size,
1866 obj->size);
1867 }
1868 }
1869 if (cd->storage_flags & OSL_RBTREE) {
1870 ret = insert_rbtree(t, col_num, r, obj);
1871 if (ret < 0)
1872 return ret;
1873 }
1874 return 1;
1875 }
1876
1877 /**
1878 * Retrieve an object of type \p OSL_DISK_STORAGE by row and column.
1879 *
1880 * \param t Pointer to an open osl table.
1881 * \param r Pointer to the row containing the object.
1882 * \param col_num The column number.
1883 * \param obj Points to the result upon successful return.
1884 *
1885 * For columns of type \p OSL_DISK_STORAGE, this function must be used to
1886 * retrieve one of its containing objects. Afterwards, osl_close_disk_object()
1887 * must be called in order to deallocate the resources.
1888 *
1889 * \return Positive on success, negative on errors. Possible errors include:
1890 * \p E_BAD_TABLE, \p E_BAD_STORAGE_TYPE, errors returned by osl_get_object().
1891 *
1892 * \sa osl_get_object(), osl_storage_type, osl_close_disk_object().
1893 */
1894 int osl_open_disk_object(const struct osl_table *t, const struct osl_row *r,
1895 unsigned col_num, struct osl_object *obj)
1896 {
1897 const struct osl_column_description *cd;
1898 char *ds_name, *filename;
1899 int ret;
1900
1901 if (!t)
1902 return -E_BAD_TABLE;
1903 cd = get_column_description(t->desc, col_num);
1904 if (cd->storage_type != OSL_DISK_STORAGE)
1905 return -E_BAD_STORAGE_TYPE;
1906
1907 ret = disk_storage_name_of_row(t, r, &ds_name);
1908 if (ret < 0)
1909 return ret;
1910 filename = disk_storage_path(t, col_num, ds_name);
1911 free(ds_name);
1912 PARA_DEBUG_LOG("filename: %s\n", filename);
1913 ret = mmap_full_file(filename, O_RDONLY, &obj->data, &obj->size, NULL);
1914 free(filename);
1915 return ret;
1916 }
1917
1918 /**
1919 * Free resources that were allocated during osl_open_disk_object().
1920 *
1921 * \param obj Pointer to the object previously returned by open_disk_object().
1922 *
1923 * \return The return value of the underlying call to para_munmap().
1924 *
1925 * \sa para_munmap().
1926 */
1927 int osl_close_disk_object(struct osl_object *obj)
1928 {
1929 return para_munmap(obj->data, obj->size);
1930 }
1931
1932 /**
1933 * Get the number of rows of the given table.
1934 *
1935 * \param t Pointer to an open osl table.
1936 * \param num_rows Result is returned here.
1937 *
1938 * The number of rows returned via \a num_rows excluding any invalid rows.
1939 *
1940 * \return Positive on success, \p -E_BAD_TABLE if \a t is \p NULL.
1941 */
1942 int osl_get_num_rows(const struct osl_table *t, unsigned *num_rows)
1943 {
1944 if (!t)
1945 return -E_BAD_TABLE;
1946 assert(t->num_rows >= t->num_invalid_rows);
1947 *num_rows = t->num_rows - t->num_invalid_rows;
1948 return 1;
1949 }
1950
1951 /**
1952 * Get the rank of a row.
1953 *
1954 * \param t An open osl table.
1955 * \param r The row to get the rank of.
1956 * \param col_num The number of an rbtree column.
1957 * \param rank Result pointer.
1958 *
1959 * The rank is, by definition, the position of the row in the linear order
1960 * determined by an in-order tree walk of the rbtree associated with column
1961 * number \a col_num of \a table.
1962 *
1963 * \return Positive on success, negative on errors.
1964 *
1965 * \sa osl_get_nth_row().
1966 */
1967 int osl_get_rank(const struct osl_table *t, struct osl_row *r,
1968 unsigned col_num, unsigned *rank)
1969 {
1970 struct osl_object obj;
1971 struct osl_column *col;
1972 struct rb_node *node;
1973 int ret = check_rbtree_col(t, col_num, &col);
1974
1975 if (ret < 0)
1976 return ret;
1977 ret = osl_get_object(t, r, col_num, &obj);
1978 if (ret < 0)
1979 return ret;
1980 ret = search_rbtree(&obj, t, col_num, &node, NULL);
1981 if (ret < 0)
1982 return ret;
1983 ret = rb_rank(node, rank);
1984 if (ret < 0)
1985 return -E_BAD_ROW;
1986 return 1;
1987 }
1988
1989 /**
1990 * Get the row with n-th greatest value.
1991 *
1992 * \param t Pointer to an open osl table.
1993 * \param col_num The column number.
1994 * \param n The rank of the desired row.
1995 * \param result Row is returned here.
1996 *
1997 * Retrieve the n-th order statistic with respect to the compare function
1998 * of the rbtree column \a col_num. In other words, get that row with
1999 * \a n th greatest value in column \a col_num. It's an error if
2000 * \a col_num is not a rbtree column, or if \a n is larger than the
2001 * number of rows in the table.
2002 *
2003 * \return Positive on success, negative on errors. Possible errors:
2004 * \p E_BAD_TABLE, \p E_BAD_STORAGE_FLAGS, \p E_RB_KEY_NOT_FOUND.
2005 *
2006 * \sa osl_storage_flags, osl_compare_func, osl_get_row(),
2007 * osl_rbtree_last_row(), osl_rbtree_first_row(), osl_get_rank().
2008 */
2009 int osl_get_nth_row(const struct osl_table *t, unsigned col_num,
2010 unsigned n, struct osl_row **result)
2011 {
2012 struct osl_column *col;
2013 struct rb_node *node;
2014 unsigned num_rows;
2015 int ret;
2016
2017 if (n == 0)
2018 return -E_RB_KEY_NOT_FOUND;
2019 ret = osl_get_num_rows(t, &num_rows);
2020 if (ret < 0)
2021 return ret;
2022 if (n > num_rows)
2023 return -E_RB_KEY_NOT_FOUND;
2024 ret = check_rbtree_col(t, col_num, &col);
2025 if (ret < 0)
2026 return ret;
2027 node = rb_nth(col->rbtree.rb_node, n);
2028 if (!node)
2029 return -E_RB_KEY_NOT_FOUND;
2030 *result = get_row_pointer(node, col->rbtree_num);
2031 return 1;
2032 }
2033
2034 /**
2035 * Get the row corresponding to the smallest rbtree node of a column.
2036 *
2037 * \param t An open rbtree table.
2038 * \param col_num The number of the rbtree column.
2039 * \param result A pointer to the first row is returned here.
2040 *
2041 * The rbtree node of the smallest object (with respect to the corresponding
2042 * compare function) is selected and the row containing this object is
2043 * returned. It is an error if \a col_num refers to a column without an
2044 * associated rbtree.
2045 *
2046 * \return Positive on success, negative on errors.
2047 *
2048 * \sa osl_get_nth_row(), osl_rbtree_last_row().
2049 */
2050 int osl_rbtree_first_row(const struct osl_table *t, unsigned col_num,
2051 struct osl_row **result)
2052 {
2053 return osl_get_nth_row(t, col_num, 1, result);
2054 }
2055
2056 /**
2057 * Get the row corresponding to the greatest rbtree node of a column.
2058 *
2059 * \param t The same meaning as in \p osl_rbtree_first_row().
2060 * \param col_num The same meaning as in \p osl_rbtree_first_row().
2061 * \param result The same meaning as in \p osl_rbtree_first_row().
2062 *
2063 * This function works just like osl_rbtree_first_row(), the only difference
2064 * is that the row containing the greatest rather than the smallest object is
2065 * returned.
2066 *
2067 * \return Positive on success, negative on errors.
2068 *
2069 * \sa osl_get_nth_row(), osl_rbtree_first_row().
2070 */
2071 int osl_rbtree_last_row(const struct osl_table *t, unsigned col_num,
2072 struct osl_row **result)
2073 {
2074 unsigned num_rows;
2075 int ret = osl_get_num_rows(t, &num_rows);
2076
2077 if (ret < 0)
2078 return ret;
2079 return osl_get_nth_row(t, col_num, num_rows, result);
2080 }