Fix osl_del_row().
[paraslash.git] / osl.c
1 /*
2 * Copyright (C) 2007 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 "list.h"
16 #include "osl_core.h"
17 /**
18 * A wrapper for lseek(2).
19 *
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.
23 *
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.
28 *
29 * \return Positive on success. Otherwise, the function returns \p -E_LSEEK.
30 *
31 * \sa lseek(2).
32 */
33 int para_lseek(int fd, off_t *offset, int whence)
34 {
35 *offset = lseek(fd, *offset, whence);
36 int ret = -E_LSEEK;
37 if (*offset == -1)
38 return ret;
39 return 1;
40 }
41
42 /**
43 * Wrapper for the write system call.
44 *
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.
48 *
49 * This function writes out the given buffer and retries if an interrupt
50 * occurred during the write.
51 *
52 * \return On success, the number of bytes written is returned, otherwise, the
53 * function returns \p -E_WRITE.
54 *
55 * \sa write(2).
56 */
57 ssize_t para_write(int fd, const void *buf, size_t size)
58 {
59 ssize_t ret;
60
61 for (;;) {
62 ret = write(fd, buf, size);
63 if ((ret < 0) && (errno == EAGAIN || errno == EINTR))
64 continue;
65 return ret >= 0? ret : -E_WRITE;
66 }
67 }
68
69 /**
70 * Write the whole buffer to a file descriptor.
71 *
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.
75 *
76 * This function writes the given buffer and continues on short writes and
77 * when interrupted by a signal.
78 *
79 * \return Positive on success, negative on errors. Possible errors: any
80 * errors returned by para_write().
81 *
82 * \sa para_write().
83 */
84 ssize_t para_write_all(int fd, const void *buf, size_t size)
85 {
86 // PARA_DEBUG_LOG("writing %zu bytes\n", size);
87 const char *b = buf;
88 while (size) {
89 ssize_t ret = para_write(fd, b, size);
90 // PARA_DEBUG_LOG("ret: %zd\n", ret);
91 if (ret < 0)
92 return ret;
93 b += ret;
94 size -= ret;
95 }
96 return 1;
97 }
98 /**
99 * Open a file, write the given buffer and close the file.
100 *
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.
104 *
105 * \return Positive on success, negative on errors. Possible errors include:
106 * any errors from para_open() or para_write().
107 *
108 * \sa para_open(), para_write().
109 */
110 int para_write_file(const char *filename, const void *buf, size_t size)
111 {
112 int ret, fd;
113
114 ret = para_open(filename, O_WRONLY | O_CREAT | O_EXCL, 0644);
115 if (ret < 0)
116 return ret;
117 fd = ret;
118 ret = para_write_all(fd, buf, size);
119 if (ret < 0)
120 goto out;
121 ret = 1;
122 out:
123 close(fd);
124 return ret;
125 }
126
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)
129 {
130 int ret, fd;
131
132 // PARA_DEBUG_LOG("appending %zu + %zu bytes\n", header_size, data_size);
133 ret = para_open(filename, O_WRONLY | O_CREAT | O_APPEND, 0644);
134 if (ret < 0)
135 return ret;
136 fd = ret;
137 if (header && header_size) {
138 ret = para_write_all(fd, header, header_size);
139 if (ret < 0)
140 goto out;
141 }
142 ret = para_write_all(fd, data, data_size);
143 if (ret < 0)
144 goto out;
145 if (new_pos) {
146 off_t offset = 0;
147 ret = para_lseek(fd, &offset, SEEK_END);
148 if (ret < 0)
149 goto out;
150 // PARA_DEBUG_LOG("new file size: " FMT_OFF_T "\n", offset);
151 *new_pos = offset;
152 }
153 ret = 1;
154 out:
155 close(fd);
156 return ret;
157 }
158
159 /**
160 * Traverse the given directory recursively.
161 *
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.
165 *
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.
170 *
171 * \return On success, 1 is returned. Otherwise, this function returns a
172 * negative value which indicates the kind of the error.
173 */
174 int for_each_file_in_dir(const char *dirname,
175 int (*func)(const char *, const void *), const void *private_data)
176 {
177 DIR *dir;
178 struct dirent *entry;
179 int cwd_fd, ret2, ret = para_opendir(dirname, &dir, &cwd_fd);
180
181 if (ret < 0)
182 return ret == -E_CHDIR_PERM? 1 : ret;
183 /* scan cwd recursively */
184 while ((entry = readdir(dir))) {
185 mode_t m;
186 char *tmp;
187 struct stat s;
188
189 if (!strcmp(entry->d_name, "."))
190 continue;
191 if (!strcmp(entry->d_name, ".."))
192 continue;
193 if (lstat(entry->d_name, &s) == -1)
194 continue;
195 m = s.st_mode;
196 if (!S_ISREG(m) && !S_ISDIR(m))
197 continue;
198 tmp = make_message("%s/%s", dirname, entry->d_name);
199 if (!S_ISDIR(m)) {
200 ret = func(tmp, private_data);
201 free(tmp);
202 if (ret < 0)
203 goto out;
204 continue;
205 }
206 /* directory */
207 ret = for_each_file_in_dir(tmp, func, private_data);
208 free(tmp);
209 if (ret < 0)
210 goto out;
211 }
212 ret = 1;
213 out:
214 closedir(dir);
215 ret2 = para_fchdir(cwd_fd);
216 if (ret2 < 0 && ret >= 0)
217 ret = ret2;
218 close(cwd_fd);
219 return ret;
220 }
221
222 static int verify_name(const char *name)
223 {
224 if (!name)
225 return -E_BAD_NAME;
226 if (!*name)
227 return -E_BAD_NAME;
228 if (strchr(name, '/'))
229 return -E_BAD_NAME;
230 if (!strcmp(name, ".."))
231 return -E_BAD_NAME;
232 if (!strcmp(name, "."))
233 return -E_BAD_NAME;
234 return 1;
235 }
236
237 /**
238 * Compare two osl objects pointing to unsigned integers of 32 bit size.
239 *
240 * \param obj1 Pointer to the first integer.
241 * \param obj2 Pointer to the second integer.
242 *
243 * \return The values required for an osl compare function.
244 *
245 * \sa osl_compare_func, osl_hash_compare().
246 */
247 int uint32_compare(const struct osl_object *obj1, const struct osl_object *obj2)
248 {
249 uint32_t d1 = read_u32((const char *)obj1->data);
250 uint32_t d2 = read_u32((const char *)obj2->data);
251
252 if (d1 < d2)
253 return 1;
254 if (d1 > d2)
255 return -1;
256 return 0;
257 }
258
259 /**
260 * Compare two osl objects pointing to hash values.
261 *
262 * \param obj1 Pointer to the first hash object.
263 * \param obj2 Pointer to the second hash object.
264 *
265 * \return The values required for an osl compare function.
266 *
267 * \sa osl_compare_func, uint32_compare().
268 */
269 int osl_hash_compare(const struct osl_object *obj1, const struct osl_object *obj2)
270 {
271 return hash_compare((HASH_TYPE *)obj1->data, (HASH_TYPE *)obj2->data);
272 }
273
274 static char *disk_storage_dirname(const struct osl_table *t, unsigned col_num,
275 const char *ds_name)
276 {
277 char *dirname, *column_name = column_filename(t, col_num);
278
279 if (!(t->desc->flags & OSL_LARGE_TABLE))
280 return column_name;
281 dirname = make_message("%s/%.2s", column_name, ds_name);
282 free(column_name);
283 return dirname;
284 }
285
286 static char *disk_storage_name_of_object(const struct osl_table *t,
287 const struct osl_object *obj)
288 {
289 HASH_TYPE hash[HASH_SIZE];
290 hash_object(obj, hash);
291 return disk_storage_name_of_hash(t, hash);
292 }
293
294 static int disk_storage_name_of_row(const struct osl_table *t,
295 const struct osl_row *row, char **name)
296 {
297 struct osl_object obj;
298 int ret = osl_get_object(t, row, t->disk_storage_name_column, &obj);
299
300 if (ret < 0)
301 return ret;
302 *name = disk_storage_name_of_object(t, &obj);
303 return 1;
304 }
305
306 static void column_name_hash(const char *col_name, HASH_TYPE *hash)
307 {
308 return hash_function(col_name, strlen(col_name), hash);
309 }
310
311 static int init_column_descriptions(struct osl_table *t)
312 {
313 int i, j, ret;
314 const struct osl_column_description *cd;
315
316 ret = -E_BAD_TABLE_DESC;
317 ret = verify_name(t->desc->name);
318 if (ret < 0)
319 goto err;
320 ret = -E_BAD_DB_DIR;
321 if (!t->desc->dir && (t->num_disk_storage_columns || t->num_mapped_columns))
322 goto err;
323 /* the size of the index header without column descriptions */
324 t->index_header_size = IDX_COLUMN_DESCRIPTIONS;
325 FOR_EACH_COLUMN(i, t->desc, cd) {
326 struct osl_column *col = t->columns + i;
327 if (cd->storage_flags & OSL_RBTREE) {
328 if (!cd->compare_function)
329 return -E_NO_COMPARE_FUNC;
330 }
331 if (cd->storage_type == OSL_NO_STORAGE)
332 continue;
333 ret = -E_NO_COLUMN_NAME;
334 if (!cd->name || !cd->name[0])
335 goto err;
336 ret = verify_name(cd->name);
337 if (ret < 0)
338 goto err;
339 t->index_header_size += index_column_description_size(cd->name);
340 column_name_hash(cd->name, col->name_hash);
341 ret = -E_DUPLICATE_COL_NAME;
342 for (j = i + 1; j < t->desc->num_columns; j++) {
343 const char *name2 = get_column_description(t->desc,
344 j)->name;
345 if (cd->name && name2 && !strcmp(cd->name, name2))
346 goto err;
347 }
348 }
349 return 1;
350 err:
351 return ret;
352 }
353
354 /**
355 * Initialize a struct table from given table description.
356 *
357 * \param desc The description of the osl table.
358 * \param table_ptr Result is returned here.
359 *
360 * This function performs several sanity checks on \p desc and returns if any
361 * of these tests fail. On success, a struct \p osl_table is allocated and
362 * initialized with data derived from \p desc.
363 *
364 * \return Positive on success, negative on errors. Possible errors include: \p
365 * E_BAD_TABLE_DESC, \p E_NO_COLUMN_DESC, \p E_NO_COLUMNS, \p
366 * E_BAD_STORAGE_TYPE, \p E_BAD_STORAGE_FLAGS, \p E_BAD_STORAGE_SIZE, \p
367 * E_NO_UNIQUE_RBTREE_COLUMN, \p E_NO_RBTREE_COL.
368 *
369 * \sa struct osl_table.
370 */
371 int init_table_structure(const struct osl_table_description *desc,
372 struct osl_table **table_ptr)
373 {
374 const struct osl_column_description *cd;
375 struct osl_table *t = para_calloc(sizeof(*t));
376 int i, ret = -E_BAD_TABLE_DESC, have_disk_storage_name_column = 0;
377
378 if (!desc)
379 goto err;
380 PARA_DEBUG_LOG("creating table structure for '%s' from table "
381 "description\n", desc->name);
382 ret = -E_NO_COLUMN_DESC;
383 if (!desc->column_descriptions)
384 goto err;
385 ret = -E_NO_COLUMNS;
386 if (!desc->num_columns)
387 goto err;
388 t->columns = para_calloc(desc->num_columns * sizeof(struct osl_column));
389 t->desc = desc;
390 FOR_EACH_COLUMN(i, t->desc, cd) {
391 enum osl_storage_type st = cd->storage_type;
392 enum osl_storage_flags sf = cd->storage_flags;
393 struct osl_column *col = &t->columns[i];
394
395 ret = -E_BAD_STORAGE_TYPE;
396 if (st != OSL_MAPPED_STORAGE && st != OSL_DISK_STORAGE
397 && st != OSL_NO_STORAGE)
398 goto err;
399 ret = -E_BAD_STORAGE_FLAGS;
400 if (st == OSL_DISK_STORAGE && sf & OSL_RBTREE)
401 goto err;
402 ret = -E_BAD_STORAGE_SIZE;
403 if (sf & OSL_FIXED_SIZE && !cd->data_size)
404 goto err;
405 switch (st) {
406 case OSL_DISK_STORAGE:
407 t->num_disk_storage_columns++;
408 break;
409 case OSL_MAPPED_STORAGE:
410 t->num_mapped_columns++;
411 col->index_offset = t->row_index_size;
412 t->row_index_size += 8;
413 break;
414 case OSL_NO_STORAGE:
415 col->volatile_num = t->num_volatile_columns;
416 t->num_volatile_columns++;
417 break;
418 }
419 if (sf & OSL_RBTREE) {
420 col->rbtree_num = t->num_rbtrees;
421 t->num_rbtrees++;
422 if ((sf & OSL_UNIQUE) && (st == OSL_MAPPED_STORAGE)) {
423 if (!have_disk_storage_name_column)
424 t->disk_storage_name_column = i;
425 have_disk_storage_name_column = 1;
426 }
427 }
428 }
429 ret = -E_NO_UNIQUE_RBTREE_COLUMN;
430 if (t->num_disk_storage_columns && !have_disk_storage_name_column)
431 goto err;
432 ret = -E_NO_RBTREE_COL;
433 if (!t->num_rbtrees)
434 goto err;
435 /* success */
436 PARA_DEBUG_LOG("OK. Index entry size: %u\n", t->row_index_size);
437 ret = init_column_descriptions(t);
438 if (ret < 0)
439 goto err;
440 *table_ptr = t;
441 return 1;
442 err:
443 free(t->columns);
444 free(t);
445 return ret;
446 }
447
448 /**
449 * Read the table description from index header.
450 *
451 * \param map The memory mapping of the index file.
452 * \param desc The values found in the index header are returned here.
453 *
454 * Read the index header, check for the paraslash magic string and the table version number.
455 * Read all information stored in the index header into \a desc.
456 *
457 * \return Positive on success, negative on errors.
458 *
459 * \sa struct osl_table_description, osl_create_table.
460 */
461 int read_table_desc(struct osl_object *map, struct osl_table_description *desc)
462 {
463 char *buf = map->data;
464 uint8_t version;
465 uint16_t header_size;
466 int ret, i;
467 unsigned offset;
468 struct osl_column_description *cd;
469
470 if (map->size < MIN_INDEX_HEADER_SIZE(1))
471 return -E_SHORT_TABLE;
472 if (strncmp(buf + IDX_PARA_MAGIC, PARA_MAGIC, strlen(PARA_MAGIC)))
473 return -E_NO_MAGIC;
474 version = read_u8(buf + IDX_VERSION);
475 if (version < MIN_TABLE_VERSION || version > MAX_TABLE_VERSION)
476 return -E_VERSION_MISMATCH;
477 desc->num_columns = read_u8(buf + IDX_TABLE_FLAGS);
478 desc->flags = read_u8(buf + IDX_TABLE_FLAGS);
479 desc->num_columns = read_u16(buf + IDX_NUM_COLUMNS);
480 PARA_DEBUG_LOG("%u columns\n", desc->num_columns);
481 if (!desc->num_columns)
482 return -E_NO_COLUMNS;
483 header_size = read_u16(buf + IDX_HEADER_SIZE);
484 if (map->size < header_size)
485 return -E_BAD_SIZE;
486 desc->column_descriptions = para_calloc(desc->num_columns
487 * sizeof(struct osl_column_description));
488 offset = IDX_COLUMN_DESCRIPTIONS;
489 FOR_EACH_COLUMN(i, desc, cd) {
490 char *null_byte;
491
492 ret = -E_SHORT_TABLE;
493 if (map->size < offset + MIN_IDX_COLUMN_DESCRIPTION_SIZE) {
494 PARA_ERROR_LOG("map size = %zu < %u = offset + min desc size\n",
495 map->size, offset + MIN_IDX_COLUMN_DESCRIPTION_SIZE);
496 goto err;
497 }
498 cd->storage_type = read_u16(buf + offset + IDX_CD_STORAGE_TYPE);
499 cd->storage_flags = read_u16(buf + offset +
500 IDX_CD_STORAGE_FLAGS);
501 cd->data_size = read_u32(buf + offset + IDX_CD_DATA_SIZE);
502 null_byte = memchr(buf + offset + IDX_CD_NAME, '\0',
503 map->size - offset - IDX_CD_NAME);
504 ret = -E_INDEX_CORRUPTION;
505 if (!null_byte)
506 goto err;
507 cd->name = para_strdup(buf + offset + IDX_CD_NAME);
508 offset += index_column_description_size(cd->name);
509 }
510 if (offset != header_size) {
511 ret = -E_INDEX_CORRUPTION;
512 PARA_ERROR_LOG("real header size = %u != %u = stored header size\n",
513 offset, header_size);
514 goto err;
515 }
516 return 1;
517 err:
518 FOR_EACH_COLUMN(i, desc, cd)
519 free(cd->name);
520 return ret;
521 }
522
523 /*
524 * check whether the table description given by \p t->desc matches the on-disk
525 * table structure stored in the index of \a t.
526 */
527 static int compare_table_descriptions(struct osl_table *t)
528 {
529 int i, ret;
530 struct osl_table_description desc;
531 const struct osl_column_description *cd1, *cd2;
532
533 /* read the on-disk structure into desc */
534 ret = read_table_desc(&t->index_map, &desc);
535 if (ret < 0)
536 return ret;
537 ret = -E_BAD_TABLE_FLAGS;
538 if (desc.flags != t->desc->flags)
539 goto out;
540 ret = -E_BAD_COLUMN_NUM;
541 if (desc.num_columns != t->desc->num_columns)
542 goto out;
543 FOR_EACH_COLUMN(i, t->desc, cd1) {
544 cd2 = get_column_description(&desc, i);
545 ret = -E_BAD_STORAGE_TYPE;
546 if (cd1->storage_type != cd2->storage_type)
547 goto out;
548 ret = -E_BAD_STORAGE_FLAGS;
549 if (cd1->storage_flags != cd2->storage_flags) {
550 PARA_ERROR_LOG("sf1 = %u != %u = sf2\n",
551 cd1->storage_flags, cd2->storage_flags);
552 goto out;
553 }
554 ret = -E_BAD_DATA_SIZE;
555 if (cd1->storage_flags & OSL_FIXED_SIZE)
556 if (cd1->data_size != cd2->data_size)
557 goto out;
558 ret = -E_BAD_COLUMN_NAME;
559 if (strcmp(cd1->name, cd2->name))
560 goto out;
561 }
562 PARA_DEBUG_LOG("table description of '%s' matches on-disk data, good\n",
563 t->desc->name);
564 ret = 1;
565 out:
566 FOR_EACH_COLUMN(i, &desc, cd1)
567 free(cd1->name);
568 free(desc.column_descriptions);
569 return ret;
570 }
571
572 static int create_table_index(struct osl_table *t)
573 {
574 char *buf, *filename;
575 int i, ret;
576 size_t size = t->index_header_size;
577 const struct osl_column_description *cd;
578 unsigned offset;
579
580 PARA_INFO_LOG("creating %zu byte index for table %s\n", size,
581 t->desc->name);
582 buf = para_calloc(size);
583 sprintf(buf + IDX_PARA_MAGIC, "%s", PARA_MAGIC);
584 write_u8(buf + IDX_TABLE_FLAGS, t->desc->flags);
585 write_u8(buf + IDX_DIRTY_FLAG, 0);
586 write_u8(buf + IDX_VERSION, CURRENT_TABLE_VERSION);
587 write_u16(buf + IDX_NUM_COLUMNS, t->desc->num_columns);
588 write_u16(buf + IDX_HEADER_SIZE, t->index_header_size);
589 offset = IDX_COLUMN_DESCRIPTIONS;
590 FOR_EACH_COLUMN(i, t->desc, cd) {
591 write_u16(buf + offset + IDX_CD_STORAGE_TYPE,
592 cd->storage_type);
593 write_u16(buf + offset + IDX_CD_STORAGE_FLAGS,
594 cd->storage_flags);
595 if (cd->storage_flags & OSL_FIXED_SIZE)
596 write_u32(buf + offset + IDX_CD_DATA_SIZE,
597 cd->data_size);
598 strcpy(buf + offset + IDX_CD_NAME, cd->name);
599 offset += index_column_description_size(cd->name);
600 }
601 assert(offset = size);
602 filename = index_filename(t->desc);
603 ret = para_write_file(filename, buf, size);
604 free(buf);
605 free(filename);
606 return ret;
607 }
608
609 /**
610 * Create a new osl table.
611 *
612 * \param desc Pointer to the table description.
613 *
614 * \return Standard.
615 */
616 int osl_create_table(const struct osl_table_description *desc)
617 {
618 const struct osl_column_description *cd;
619 char *table_dir = NULL, *filename;
620 struct osl_table *t;
621 int i, ret = init_table_structure(desc, &t);
622
623 if (ret < 0)
624 return ret;
625 PARA_INFO_LOG("creating %s\n", desc->name);
626 FOR_EACH_COLUMN(i, t->desc, cd) {
627 if (cd->storage_type == OSL_NO_STORAGE)
628 continue;
629 if (!table_dir) {
630 ret = para_mkdir(desc->dir, 0777);
631 if (ret < 0 && !is_errno(-ret, EEXIST))
632 goto out;
633 table_dir = make_message("%s/%s", desc->dir,
634 desc->name);
635 ret = para_mkdir(table_dir, 0777);
636 if (ret < 0)
637 goto out;
638 }
639 filename = column_filename(t, i);
640 PARA_INFO_LOG("filename: %s\n", filename);
641 if (cd->storage_type == OSL_MAPPED_STORAGE) {
642 ret = para_open(filename, O_RDWR | O_CREAT | O_EXCL,
643 0644);
644 free(filename);
645 if (ret < 0)
646 goto out;
647 close(ret);
648 continue;
649 }
650 /* DISK STORAGE */
651 ret = para_mkdir(filename, 0777);
652 free(filename);
653 if (ret < 0)
654 goto out;
655 }
656 if (t->num_mapped_columns) {
657 ret = create_table_index(t);
658 if (ret < 0)
659 goto out;
660 }
661 ret = 1;
662 out:
663 free(table_dir);
664 free(t->columns);
665 free(t);
666 return ret;
667 }
668
669 static int table_is_dirty(struct osl_table *t)
670 {
671 char *buf = (char *)t->index_map.data + IDX_DIRTY_FLAG;
672 uint8_t dirty = read_u8(buf) & 0x1;
673 return !!dirty;
674 }
675
676 static void mark_table_dirty(struct osl_table *t)
677 {
678 char *buf = (char *)t->index_map.data + IDX_DIRTY_FLAG;
679 write_u8(buf, read_u8(buf) | 1);
680 }
681
682 static void mark_table_clean(struct osl_table *t)
683 {
684 char *buf = (char *)t->index_map.data + IDX_DIRTY_FLAG;
685 write_u8(buf, read_u8(buf) & 0xfe);
686 }
687
688 static void unmap_column(struct osl_table *t, unsigned col_num)
689 {
690 struct osl_object map = t->columns[col_num].data_map;
691 int ret;
692 if (!map.data)
693 return;
694 ret = para_munmap(map.data, map.size);
695 assert(ret > 0);
696 map.data = NULL;
697 }
698
699 /**
700 * Unmap all mapped files of an osl table.
701 *
702 * \param t Pointer to a mapped table.
703 * \param flags Options for unmapping.
704 *
705 * \return Positive on success, negative on errors.
706 *
707 * \sa map_table(), enum osl_close_flags, para_munmap().
708 */
709 int unmap_table(struct osl_table *t, enum osl_close_flags flags)
710 {
711 unsigned i;
712 const struct osl_column_description *cd;
713 int ret;
714
715 if (!t->num_mapped_columns) /* can this ever happen? */
716 return 1;
717 PARA_DEBUG_LOG("unmapping table '%s'\n", t->desc->name);
718 if (!t->index_map.data)
719 return -E_NOT_MAPPED;
720 if (flags & OSL_MARK_CLEAN)
721 mark_table_clean(t);
722 ret = para_munmap(t->index_map.data, t->index_map.size);
723 if (ret < 0)
724 return ret;
725 t->index_map.data = NULL;
726 if (!t->num_rows)
727 return 1;
728 FOR_EACH_MAPPED_COLUMN(i, t, cd)
729 unmap_column(t, i);
730 return 1;
731 }
732
733 static int map_column(struct osl_table *t, unsigned col_num)
734 {
735 struct stat statbuf;
736 char *filename = column_filename(t, col_num);
737 int ret = -E_STAT;
738 if (stat(filename, &statbuf) < 0) {
739 free(filename);
740 return ret;
741 }
742 if (!(S_IFREG & statbuf.st_mode)) {
743 free(filename);
744 return ret;
745 }
746 ret = mmap_full_file(filename, O_RDWR,
747 &t->columns[col_num].data_map.data,
748 &t->columns[col_num].data_map.size,
749 NULL);
750 free(filename);
751 return ret;
752 }
753
754 /**
755 * Map the index file and all columns of type \p OSL_MAPPED_STORAGE into memory.
756 *
757 * \param t Pointer to an initialized table structure.
758 * \param flags Mapping options.
759 *
760 * \return Negative return value on errors; on success the number of rows
761 * (including invalid rows) is returned.
762 *
763 * \sa unmap_table(), enum map_table_flags, osl_open_table(), mmap(2).
764 */
765 int map_table(struct osl_table *t, enum map_table_flags flags)
766 {
767 char *filename;
768 const struct osl_column_description *cd;
769 int i = 0, ret, num_rows = 0;
770
771 if (!t->num_mapped_columns)
772 return 0;
773 if (t->index_map.data)
774 return -E_ALREADY_MAPPED;
775 filename = index_filename(t->desc);
776 PARA_DEBUG_LOG("mapping table '%s' (index: %s)\n", t->desc->name, filename);
777 ret = mmap_full_file(filename, flags & MAP_TBL_FL_MAP_RDONLY?
778 O_RDONLY : O_RDWR, &t->index_map.data, &t->index_map.size, NULL);
779 free(filename);
780 if (ret < 0)
781 return ret;
782 if (flags & MAP_TBL_FL_VERIFY_INDEX) {
783 ret = compare_table_descriptions(t);
784 if (ret < 0)
785 goto err;
786 }
787 ret = -E_BUSY;
788 if (!(flags & MAP_TBL_FL_IGNORE_DIRTY)) {
789 if (table_is_dirty(t)) {
790 PARA_ERROR_LOG("%s is dirty\n", t->desc->name);
791 goto err;
792 }
793 }
794 mark_table_dirty(t);
795 num_rows = table_num_rows(t);
796 if (!num_rows)
797 return num_rows;
798 /* map data files */
799 FOR_EACH_MAPPED_COLUMN(i, t, cd) {
800 ret = map_column(t, i);
801 if (ret < 0)
802 goto err;
803 }
804 return num_rows;
805 err: /* unmap what is already mapped */
806 for (i--; i >= 0; i--) {
807 struct osl_object map = t->columns[i].data_map;
808 para_munmap(map.data, map.size);
809 map.data = NULL;
810 }
811 para_munmap(t->index_map.data, t->index_map.size);
812 t->index_map.data = NULL;
813 return ret;
814 }
815
816 /**
817 * Retrieve a mapped object by row and column number.
818 *
819 * \param t Pointer to an open osl table.
820 * \param col_num Number of the mapped column containing the object to retrieve.
821 * \param row_num Number of the row containing the object to retrieve.
822 * \param obj The result is returned here.
823 *
824 * It is considered an error if \a col_num does not refer to a column
825 * of storage type \p OSL_MAPPED_STORAGE.
826 *
827 * \return Positive on success, negative on errors. Possible errors include:
828 * \p E_BAD_ROW_NUM, \p E_INVALID_OBJECT.
829 *
830 * \sa osl_storage_type.
831 */
832 int get_mapped_object(const struct osl_table *t, unsigned col_num,
833 uint32_t row_num, struct osl_object *obj)
834 {
835 struct osl_column *col = &t->columns[col_num];
836 uint32_t offset;
837 char *header;
838 char *cell_index;
839 int ret;
840
841 if (t->num_rows <= row_num)
842 return -E_BAD_ROW_NUM;
843 ret = get_cell_index(t, row_num, col_num, &cell_index);
844 if (ret < 0)
845 return ret;
846 offset = read_u32(cell_index);
847 obj->size = read_u32(cell_index + 4) - 1;
848 header = col->data_map.data + offset;
849 obj->data = header + 1;
850 if (read_u8(header) == 0xff) {
851 PARA_ERROR_LOG("col %u, size %zu, offset %u\n", col_num,
852 obj->size, offset);
853 return -E_INVALID_OBJECT;
854 }
855 return 1;
856 }
857
858 static int search_rbtree(const struct osl_object *obj,
859 const struct osl_table *t, unsigned col_num,
860 struct rb_node **result, struct rb_node ***rb_link)
861 {
862 struct osl_column *col = &t->columns[col_num];
863 struct rb_node **new = &col->rbtree.rb_node, *parent = NULL;
864 const struct osl_column_description *cd =
865 get_column_description(t->desc, col_num);
866 enum osl_storage_type st = cd->storage_type;
867 while (*new) {
868 struct osl_row *this_row = get_row_pointer(*new,
869 col->rbtree_num);
870 int ret;
871 struct osl_object this_obj;
872 parent = *new;
873 if (st == OSL_MAPPED_STORAGE) {
874 ret = get_mapped_object(t, col_num, this_row->num,
875 &this_obj);
876 if (ret < 0)
877 return ret;
878 } else
879 this_obj = this_row->volatile_objects[col->volatile_num];
880 ret = cd->compare_function(obj, &this_obj);
881 if (!ret) {
882 if (result)
883 *result = get_rb_node_pointer(this_row,
884 col->rbtree_num);
885 return 1;
886 }
887 if (ret < 0)
888 new = &((*new)->rb_left);
889 else
890 new = &((*new)->rb_right);
891 }
892 if (result)
893 *result = parent;
894 if (rb_link)
895 *rb_link = new;
896 return -E_RB_KEY_NOT_FOUND;
897 }
898
899 static int insert_rbtree(struct osl_table *t, unsigned col_num,
900 const struct osl_row *row, const struct osl_object *obj)
901 {
902 struct rb_node *parent, **rb_link;
903 unsigned rbtree_num;
904 struct rb_node *n;
905 int ret = search_rbtree(obj, t, col_num, &parent, &rb_link);
906
907 if (ret > 0)
908 return -E_RB_KEY_EXISTS;
909 rbtree_num = t->columns[col_num].rbtree_num;
910 n = get_rb_node_pointer(row, rbtree_num);
911 rb_link_node(n, parent, rb_link);
912 rb_insert_color(n, &t->columns[col_num].rbtree);
913 return 1;
914 }
915
916 static void remove_rb_node(struct osl_table *t, unsigned col_num,
917 const struct osl_row *row)
918 {
919 struct osl_column *col = &t->columns[col_num];
920 const struct osl_column_description *cd =
921 get_column_description(t->desc, col_num);
922 enum osl_storage_flags sf = cd->storage_flags;
923 struct rb_node *victim, *splice_out_node, *tmp;
924 if (!(sf & OSL_RBTREE))
925 return;
926 /*
927 * Which node is removed/spliced out actually depends on how many
928 * children the victim node has: If it has no children, it gets
929 * deleted. If it has one child, it gets spliced out. If it has two
930 * children, its successor (which has at most a right child) gets
931 * spliced out.
932 */
933 victim = get_rb_node_pointer(row, col->rbtree_num);
934 if (victim->rb_left && victim->rb_right)
935 splice_out_node = rb_next(victim);
936 else
937 splice_out_node = victim;
938 /* Go up to the root and decrement the size of each node in the path. */
939 for (tmp = splice_out_node; tmp; tmp = rb_parent(tmp))
940 tmp->size--;
941 rb_erase(victim, &col->rbtree);
942 }
943
944 static int add_row_to_rbtrees(struct osl_table *t, uint32_t row_num,
945 struct osl_object *volatile_objs, struct osl_row **row_ptr)
946 {
947 unsigned i;
948 int ret;
949 struct osl_row *row = allocate_row(t->num_rbtrees);
950 const struct osl_column_description *cd;
951
952 row->num = row_num;
953 row->volatile_objects = volatile_objs;
954 FOR_EACH_RBTREE_COLUMN(i, t, cd) {
955 if (cd->storage_type == OSL_MAPPED_STORAGE) {
956 struct osl_object obj;
957 ret = get_mapped_object(t, i, row_num, &obj);
958 if (ret < 0)
959 goto err;
960 ret = insert_rbtree(t, i, row, &obj);
961 } else { /* volatile */
962 const struct osl_object *obj
963 = volatile_objs + t->columns[i].volatile_num;
964 ret = insert_rbtree(t, i, row, obj);
965 }
966 if (ret < 0)
967 goto err;
968 }
969 if (row_ptr)
970 *row_ptr = row;
971 return 1;
972 err: /* rollback changes, i.e. remove added entries from rbtrees */
973 while (i)
974 remove_rb_node(t, i--, row);
975 free(row);
976 return ret;
977 }
978
979 static void free_volatile_objects(const struct osl_table *t,
980 enum osl_close_flags flags)
981 {
982 int i, j;
983 struct rb_node *n;
984 struct osl_column *rb_col;
985 const struct osl_column_description *cd;
986
987 if (!t->num_volatile_columns)
988 return;
989 /* find the first rbtree column (any will do) */
990 FOR_EACH_RBTREE_COLUMN(i, t, cd)
991 break;
992 rb_col = t->columns + i;
993 /* walk that rbtree and free all volatile objects */
994 for (n = rb_first(&rb_col->rbtree); n; n = rb_next(n)) {
995 struct osl_row *r = get_row_pointer(n, rb_col->rbtree_num);
996 if (flags & OSL_FREE_VOLATILE)
997 FOR_EACH_VOLATILE_COLUMN(j, t, cd) {
998 if (cd->storage_flags & OSL_DONT_FREE)
999 continue;
1000 free(r->volatile_objects[
1001 t->columns[j].volatile_num].data);
1002 }
1003 // for (j = 0; j < t->num_volatile_columns; j++)
1004 // free(r->volatile_objects[j].data);
1005 free(r->volatile_objects);
1006 }
1007 }
1008
1009 /**
1010 * Erase all rbtree nodes and free resources.
1011 *
1012 * \param t Pointer to an open osl table.
1013 *
1014 * This function is called by osl_close_table().
1015 */
1016 void clear_rbtrees(struct osl_table *t)
1017 {
1018 const struct osl_column_description *cd;
1019 unsigned i, rbtrees_cleared = 0;
1020
1021 FOR_EACH_RBTREE_COLUMN(i, t, cd) {
1022 struct osl_column *col = &t->columns[i];
1023 struct rb_node *n;
1024 rbtrees_cleared++;
1025 for (n = rb_first(&col->rbtree); n;) {
1026 struct osl_row *r;
1027 rb_erase(n, &col->rbtree);
1028 if (rbtrees_cleared == t->num_rbtrees) {
1029 r = get_row_pointer(n, col->rbtree_num);
1030 n = rb_next(n);
1031 free(r);
1032 } else
1033 n = rb_next(n);
1034 }
1035 }
1036
1037 }
1038
1039 /**
1040 * Close an osl table.
1041 *
1042 * \param t Pointer to the table to be closed.
1043 * \param flags Options for what should be cleaned up.
1044 *
1045 * If osl_open_table() succeeds, the resulting table pointer must later be
1046 * passed to this function in order to flush all changes to the file system and
1047 * to free the resources that were allocated by osl_open_table().
1048 *
1049 * \return Positive on success, negative on errors. Possible errors: \p E_BAD_TABLE,
1050 * errors returned by unmap_table().
1051 *
1052 * \sa osl_open_table(), unmap_table().
1053 */
1054 int osl_close_table(struct osl_table *t, enum osl_close_flags flags)
1055 {
1056 int ret;
1057
1058 if (!t)
1059 return -E_BAD_TABLE;
1060 free_volatile_objects(t, flags);
1061 clear_rbtrees(t);
1062 ret = unmap_table(t, flags);
1063 if (ret < 0)
1064 PARA_ERROR_LOG("unmap_table failed: %d\n", ret);
1065 free(t->columns);
1066 free(t);
1067 return ret;
1068 }
1069
1070 /**
1071 * Find out whether the given row number corresponds to an invalid row.
1072 *
1073 * \param t Pointer to the osl table.
1074 * \param row_num The number of the row in question.
1075 *
1076 * By definition, a row is considered invalid if all its index entries
1077 * are invalid.
1078 *
1079 * \return Positive if \a row_num corresponds to an invalid row,
1080 * zero if it corresponds to a valid row, negative on errors.
1081 */
1082 int row_is_invalid(struct osl_table *t, uint32_t row_num)
1083 {
1084 char *row_index;
1085 int i, ret = get_row_index(t, row_num, &row_index);
1086
1087 if (ret < 0)
1088 return ret;
1089 for (i = 0; i < t->row_index_size; i++) {
1090 if ((unsigned char)row_index[i] != 0xff)
1091 return 0;
1092 }
1093 PARA_INFO_LOG("row %d is invalid\n", row_num);
1094 return 1;
1095 }
1096
1097 /**
1098 * Invalidate a row of an osl table.
1099 *
1100 * \param t Pointer to an open osl table.
1101 * \param row_num Number of the row to mark as invalid.
1102 *
1103 * This function marks each mapped object in the index entry of \a row as
1104 * invalid.
1105 *
1106 * \return Positive on success, negative on errors.
1107 */
1108 int mark_row_invalid(struct osl_table *t, uint32_t row_num)
1109 {
1110 char *row_index;
1111 int ret = get_row_index(t, row_num, &row_index);
1112
1113 if (ret < 0)
1114 return ret;
1115 PARA_INFO_LOG("marking row %d as invalid\n", row_num);
1116 memset(row_index, 0xff, t->row_index_size);
1117 return 1;
1118 }
1119
1120 /**
1121 * Initialize all rbtrees and compute number of invalid rows.
1122 *
1123 * \param t The table containing the rbtrees to be initialized.
1124 *
1125 * \return Positive on success, negative on errors.
1126 */
1127 int init_rbtrees(struct osl_table *t)
1128 {
1129 int i, ret;
1130 const struct osl_column_description *cd;
1131
1132 /* create rbtrees */
1133 FOR_EACH_RBTREE_COLUMN(i, t, cd)
1134 t->columns[i].rbtree = RB_ROOT;
1135 /* add valid rows to rbtrees */
1136 t->num_invalid_rows = 0;
1137 for (i = 0; i < t->num_rows; i++) {
1138 ret = row_is_invalid(t, i);
1139 if (ret < 0)
1140 return ret;
1141 if (ret) {
1142 t->num_invalid_rows++;
1143 continue;
1144 }
1145 ret = add_row_to_rbtrees(t, i, NULL, NULL);
1146 if (ret < 0)
1147 return ret;
1148 }
1149 return 1;
1150 }
1151
1152 /**
1153 * Open an osl table.
1154 *
1155 * Each osl table must be opened before its data can be accessed.
1156 *
1157 * \param table_desc Describes the table to be opened.
1158 * \param result Contains a pointer to the open table on success.
1159 *
1160 * The table description given by \a desc should coincide with the
1161 * description used at creation time.
1162 *
1163 * \return Standard.
1164 */
1165 int osl_open_table(const struct osl_table_description *table_desc,
1166 struct osl_table **result)
1167 {
1168 int i, ret;
1169 struct osl_table *t;
1170 const struct osl_column_description *cd;
1171
1172 PARA_INFO_LOG("opening table %s\n", table_desc->name);
1173 ret = init_table_structure(table_desc, &t);
1174 if (ret < 0)
1175 return ret;
1176 FOR_EACH_DISK_STORAGE_COLUMN(i, t, cd) {
1177 /* check if directory exists */
1178 char *dirname = column_filename(t, i);
1179 struct stat statbuf;
1180 ret = stat(dirname, &statbuf);
1181 free(dirname);
1182 if (ret < 0) {
1183 ret = -ERRNO_TO_PARA_ERROR(errno);
1184 goto err;
1185 }
1186 ret = -ERRNO_TO_PARA_ERROR(ENOTDIR);
1187 if (!S_ISDIR(statbuf.st_mode))
1188 goto err;
1189 }
1190 ret = map_table(t, MAP_TBL_FL_VERIFY_INDEX);
1191 if (ret < 0)
1192 goto err;
1193 t->num_rows = ret;
1194 PARA_DEBUG_LOG("num rows: %d\n", t->num_rows);
1195 ret = init_rbtrees(t);
1196 if (ret < 0) {
1197 osl_close_table(t, OSL_MARK_CLEAN); /* ignore further errors */
1198 return ret;
1199 }
1200 *result = t;
1201 return 1;
1202 err:
1203 free(t->columns);
1204 free(t);
1205 return ret;
1206 }
1207
1208 static int create_disk_storage_object_dir(const struct osl_table *t,
1209 unsigned col_num, const char *ds_name)
1210 {
1211 char *dirname;
1212 int ret;
1213
1214 if (!(t->desc->flags & OSL_LARGE_TABLE))
1215 return 1;
1216 dirname = disk_storage_dirname(t, col_num, ds_name);
1217 ret = para_mkdir(dirname, 0777);
1218 free(dirname);
1219 if (ret < 0 && !is_errno(-ret, EEXIST))
1220 return ret;
1221 return 1;
1222 }
1223
1224 static int write_disk_storage_file(const struct osl_table *t, unsigned col_num,
1225 const struct osl_object *obj, const char *ds_name)
1226 {
1227 int ret;
1228 char *filename;
1229
1230 ret = create_disk_storage_object_dir(t, col_num, ds_name);
1231 if (ret < 0)
1232 return ret;
1233 filename = disk_storage_path(t, col_num, ds_name);
1234 ret = para_write_file(filename, obj->data, obj->size);
1235 free(filename);
1236 return ret;
1237 }
1238
1239 static int append_map_file(const struct osl_table *t, unsigned col_num,
1240 const struct osl_object *obj, uint32_t *new_size)
1241 {
1242 char *filename = column_filename(t, col_num);
1243 int ret;
1244 char header = 0; /* zero means valid object */
1245
1246 // PARA_DEBUG_LOG("appending %zu + 1 byte\n", obj->size);
1247 ret = append_file(filename, &header, 1, obj->data, obj->size,
1248 new_size);
1249 free(filename);
1250 return ret;
1251 }
1252
1253 static int append_row_index(const struct osl_table *t, char *row_index)
1254 {
1255 char *filename;
1256 int ret;
1257
1258 if (!t->num_mapped_columns)
1259 return 1;
1260 filename = index_filename(t->desc);
1261 ret = append_file(filename, NULL, 0, row_index,
1262 t->row_index_size, NULL);
1263 free(filename);
1264 return ret;
1265 }
1266
1267 /**
1268 * A wrapper for truncate(2)
1269 *
1270 * \param path Name of the regular file to truncate
1271 * \param size Number of bytes to \b shave \b off
1272 *
1273 * Truncate the regular file named by \a path by \a size bytes.
1274 *
1275 * \return Positive on success, negative on errors. Possible errors include: \p
1276 * E_STAT, \p E_BAD_SIZE, \p E_TRUNC.
1277 *
1278 * \sa truncate(2)
1279 */
1280 int para_truncate(const char *path, off_t size)
1281 {
1282 int ret;
1283 struct stat statbuf;
1284
1285 ret = -E_STAT;
1286 if (stat(path, &statbuf) < 0)
1287 goto out;
1288 ret = -E_BAD_SIZE;
1289 if (statbuf.st_size < size)
1290 goto out;
1291 ret = -E_TRUNC;
1292 if (truncate(path, statbuf.st_size - size) < 0)
1293 goto out;
1294 ret = 1;
1295 out:
1296 return ret;
1297 }
1298
1299 static int truncate_mapped_file(const struct osl_table *t, unsigned col_num,
1300 off_t size)
1301 {
1302 char *filename = column_filename(t, col_num);
1303 int ret = para_truncate(filename, size);
1304 free(filename);
1305 return ret;
1306 }
1307
1308 static int delete_disk_storage_file(const struct osl_table *t, unsigned col_num,
1309 const char *ds_name)
1310 {
1311 char *dirname, *filename = disk_storage_path(t, col_num, ds_name);
1312 int ret = unlink(filename), err = errno;
1313
1314 free(filename);
1315 if (ret < 0)
1316 return -ERRNO_TO_PARA_ERROR(err);
1317 if (!(t->desc->flags & OSL_LARGE_TABLE))
1318 return 1;
1319 dirname = disk_storage_dirname(t, col_num, ds_name);
1320 rmdir(dirname);
1321 free(dirname);
1322 return 1;
1323 }
1324
1325 /**
1326 * Add a new row to an osl table and retrieve this row.
1327 *
1328 * \param t Pointer to an open osl table.
1329 * \param objects Array of objects to be added.
1330 * \param row Result pointer.
1331 *
1332 * The \a objects parameter must point to an array containing one object per
1333 * column. The order of the objects in the array is given by the table
1334 * description of \a table. Several sanity checks are performed during object
1335 * insertion and the function returns without modifying the table if any of
1336 * these tests fail. In fact, it is atomic in the sense that it either
1337 * succeeds or leaves the table unchanged (i.e. either all or none of the
1338 * objects are added to the table).
1339 *
1340 * It is considered an error if an object is added to a column with associated
1341 * rbtree if this object is equal to an object already contained in that column
1342 * (i.e. the compare function for the column's rbtree returns zero).
1343 *
1344 * Possible errors include: \p E_RB_KEY_EXISTS, \p E_BAD_DATA_SIZE.
1345 *
1346 * \return Positive on success, negative on errors.
1347 *
1348 * \sa struct osl_table_description, osl_compare_func, osl_add_row().
1349 */
1350 int osl_add_and_get_row(struct osl_table *t, struct osl_object *objects,
1351 struct osl_row **row)
1352 {
1353 int i, ret;
1354 char *ds_name = NULL;
1355 struct rb_node **rb_parents = NULL, ***rb_links = NULL;
1356 char *new_row_index = NULL;
1357 struct osl_object *volatile_objs = NULL;
1358 const struct osl_column_description *cd;
1359
1360 if (!t)
1361 return -E_BAD_TABLE;
1362 rb_parents = para_malloc(t->num_rbtrees * sizeof(struct rn_node*));
1363 rb_links = para_malloc(t->num_rbtrees * sizeof(struct rn_node**));
1364 if (t->num_mapped_columns)
1365 new_row_index = para_malloc(t->row_index_size);
1366 /* pass 1: sanity checks */
1367 // PARA_DEBUG_LOG("sanity tests: %p:%p\n", objects[0].data,
1368 // objects[1].data);
1369 FOR_EACH_COLUMN(i, t->desc, cd) {
1370 enum osl_storage_type st = cd->storage_type;
1371 enum osl_storage_flags sf = cd->storage_flags;
1372
1373 // ret = -E_NULL_OBJECT;
1374 // if (!objects[i])
1375 // goto out;
1376 if (st == OSL_DISK_STORAGE)
1377 continue;
1378 if (sf & OSL_RBTREE) {
1379 unsigned rbtree_num = t->columns[i].rbtree_num;
1380 ret = -E_RB_KEY_EXISTS;
1381 // PARA_DEBUG_LOG("checking whether %p exists\n",
1382 // objects[i].data);
1383 if (search_rbtree(objects + i, t, i,
1384 &rb_parents[rbtree_num],
1385 &rb_links[rbtree_num]) > 0)
1386 goto out;
1387 }
1388 if (sf & OSL_FIXED_SIZE) {
1389 // PARA_DEBUG_LOG("fixed size. need: %zu, have: %d\n",
1390 // objects[i].size, cd->data_size);
1391 ret = -E_BAD_DATA_SIZE;
1392 if (objects[i].size != cd->data_size)
1393 goto out;
1394 }
1395 }
1396 if (t->num_disk_storage_columns)
1397 ds_name = disk_storage_name_of_object(t,
1398 &objects[t->disk_storage_name_column]);
1399 ret = unmap_table(t, OSL_MARK_CLEAN);
1400 if (ret < 0)
1401 goto out;
1402 // PARA_DEBUG_LOG("sanity tests passed%s\n", "");
1403 /* pass 2: create data files, append map data */
1404 FOR_EACH_COLUMN(i, t->desc, cd) {
1405 enum osl_storage_type st = cd->storage_type;
1406 if (st == OSL_NO_STORAGE)
1407 continue;
1408 if (st == OSL_MAPPED_STORAGE) {
1409 uint32_t new_size;
1410 struct osl_column *col = &t->columns[i];
1411 // PARA_DEBUG_LOG("appending object of size %zu\n",
1412 // objects[i].size);
1413 ret = append_map_file(t, i, objects + i, &new_size);
1414 if (ret < 0)
1415 goto rollback;
1416 update_cell_index(new_row_index, col, new_size,
1417 objects[i].size);
1418 continue;
1419 }
1420 /* DISK_STORAGE */
1421 ret = write_disk_storage_file(t, i, objects + i, ds_name);
1422 if (ret < 0)
1423 goto rollback;
1424 }
1425 ret = append_row_index(t, new_row_index);
1426 if (ret < 0)
1427 goto rollback;
1428 ret = map_table(t, MAP_TBL_FL_VERIFY_INDEX);
1429 if (ret < 0) { /* truncate index and rollback changes */
1430 char *filename = index_filename(t->desc);
1431 para_truncate(filename, t->row_index_size);
1432 free(filename);
1433 goto rollback;
1434 }
1435 /* pass 3: add entry to rbtrees */
1436 if (t->num_volatile_columns) {
1437 volatile_objs = para_calloc(t->num_volatile_columns
1438 * sizeof(struct osl_object));
1439 FOR_EACH_VOLATILE_COLUMN(i, t, cd)
1440 volatile_objs[t->columns[i].volatile_num] = objects[i];
1441 }
1442 t->num_rows++;
1443 // PARA_DEBUG_LOG("adding new entry as row #%d\n", t->num_rows - 1);
1444 ret = add_row_to_rbtrees(t, t->num_rows - 1, volatile_objs, row);
1445 if (ret < 0)
1446 goto out;
1447 // PARA_DEBUG_LOG("added new entry as row #%d\n", t->num_rows - 1);
1448 ret = 1;
1449 goto out;
1450 rollback: /* rollback all changes made, ignore further errors */
1451 for (i--; i >= 0; i--) {
1452 cd = get_column_description(t->desc, i);
1453 enum osl_storage_type st = cd->storage_type;
1454 if (st == OSL_NO_STORAGE)
1455 continue;
1456
1457 if (st == OSL_MAPPED_STORAGE)
1458 truncate_mapped_file(t, i, objects[i].size);
1459 else /* disk storage */
1460 delete_disk_storage_file(t, i, ds_name);
1461 }
1462 /* ignore error and return previous error value */
1463 map_table(t, MAP_TBL_FL_VERIFY_INDEX);
1464 out:
1465 free(new_row_index);
1466 free(ds_name);
1467 free(rb_parents);
1468 free(rb_links);
1469 return ret;
1470 }
1471
1472 /**
1473 * Add a new row to an osl table.
1474 *
1475 * \param t Same meaning as osl_add_and_get_row().
1476 * \param objects Same meaning as osl_add_and_get_row().
1477 *
1478 * \return The return value of the underlying call to osl_add_and_get_row().
1479 *
1480 * This is equivalent to osl_add_and_get_row(t, objects, NULL).
1481 */
1482 int osl_add_row(struct osl_table *t, struct osl_object *objects)
1483 {
1484 return osl_add_and_get_row(t, objects, NULL);
1485 }
1486
1487 /**
1488 * Retrieve an object identified by row and column
1489 *
1490 * \param t Pointer to an open osl table.
1491 * \param r Pointer to the row.
1492 * \param col_num The column number.
1493 * \param object The result pointer.
1494 *
1495 * The column determined by \a col_num must be of type \p OSL_MAPPED_STORAGE
1496 * or \p OSL_NO_STORAGE, i.e. no disk storage objects may be retrieved by this
1497 * function.
1498 *
1499 * \return Positive if object was found, negative on errors. Possible errors
1500 * include: \p E_BAD_TABLE, \p E_BAD_STORAGE_TYPE.
1501 *
1502 * \sa osl_storage_type, osl_open_disk_object().
1503 */
1504 int osl_get_object(const struct osl_table *t, const struct osl_row *r,
1505 unsigned col_num, struct osl_object *object)
1506 {
1507 const struct osl_column_description *cd;
1508
1509 if (!t)
1510 return -E_BAD_TABLE;
1511 cd = get_column_description(t->desc, col_num);
1512 /* col must not be disk storage */
1513 if (cd->storage_type == OSL_DISK_STORAGE)
1514 return -E_BAD_STORAGE_TYPE;
1515 if (cd->storage_type == OSL_MAPPED_STORAGE)
1516 return get_mapped_object(t, col_num, r->num, object);
1517 /* volatile */
1518 *object = r->volatile_objects[t->columns[col_num].volatile_num];
1519 return 1;
1520 }
1521
1522 static int mark_mapped_object_invalid(const struct osl_table *t,
1523 uint32_t row_num, unsigned col_num)
1524 {
1525 struct osl_object obj;
1526 char *p;
1527 int ret = get_mapped_object(t, col_num, row_num, &obj);
1528
1529 if (ret < 0)
1530 return ret;
1531 p = obj.data;
1532 p--;
1533 *p = 0xff;
1534 return 1;
1535 }
1536
1537 /**
1538 * Delete a row from an osl table.
1539 *
1540 * \param t Pointer to an open osl table.
1541 * \param row Pointer to the row to delete.
1542 *
1543 * This removes all disk storage objects, removes all rbtree nodes, and frees
1544 * all volatile objects belonging to the given row. For mapped columns, the
1545 * data is merely marked invalid and may be pruned from time to time by
1546 * para_fsck.
1547 *
1548 * \return Positive on success, negative on errors. Possible errors include:
1549 * \p E_BAD_TABLE, errors returned by osl_get_object().
1550 */
1551 int osl_del_row(struct osl_table *t, struct osl_row *row)
1552 {
1553 struct osl_row *r = row;
1554 int i, ret;
1555 const struct osl_column_description *cd;
1556
1557 if (!t)
1558 return -E_BAD_TABLE;
1559 PARA_INFO_LOG("deleting row %p\n", row);
1560
1561 if (t->num_disk_storage_columns) {
1562 char *ds_name;
1563 ret = disk_storage_name_of_row(t, r, &ds_name);
1564 if (ret < 0)
1565 goto out;
1566 FOR_EACH_DISK_STORAGE_COLUMN(i, t, cd)
1567 delete_disk_storage_file(t, i, ds_name);
1568 free(ds_name);
1569 }
1570 FOR_EACH_COLUMN(i, t->desc, cd) {
1571 struct osl_column *col = t->columns + i;
1572 enum osl_storage_type st = cd->storage_type;
1573 remove_rb_node(t, i, r);
1574 if (st == OSL_MAPPED_STORAGE) {
1575 mark_mapped_object_invalid(t, r->num, i);
1576 continue;
1577 }
1578 if (st == OSL_NO_STORAGE && !(cd->storage_flags & OSL_DONT_FREE))
1579 free(r->volatile_objects[col->volatile_num].data);
1580 }
1581 if (t->num_mapped_columns) {
1582 ret = mark_row_invalid(t, r->num);
1583 if (ret < 0)
1584 goto out;
1585 t->num_invalid_rows++;
1586 } else
1587 t->num_rows--;
1588 ret = 1;
1589 out:
1590 free(r->volatile_objects);
1591 free(r);
1592 return ret;
1593 }
1594
1595 /* test if column has an rbtree */
1596 static int check_rbtree_col(const struct osl_table *t, unsigned col_num,
1597 struct osl_column **col)
1598 {
1599 if (!t)
1600 return -E_BAD_TABLE;
1601 if (!(get_column_description(t->desc, col_num)->storage_flags & OSL_RBTREE))
1602 return -E_BAD_STORAGE_FLAGS;
1603 *col = t->columns + col_num;
1604 return 1;
1605 }
1606
1607 /**
1608 * Get the row that contains the given object.
1609 *
1610 * \param t Pointer to an open osl table.
1611 * \param col_num The number of the column to be searched.
1612 * \param obj The object to be looked up.
1613 * \param result Points to the row containing \a obj.
1614 *
1615 * Lookup \a obj in \a t and return the row containing \a obj. The column
1616 * specified by \a col_num must have an associated rbtree.
1617 *
1618 * \return Positive on success, negative on errors. If an error occurred, \a
1619 * result is set to \p NULL. Possible errors include: \p E_BAD_TABLE, \p
1620 * E_BAD_STORAGE_FLAGS, errors returned by get_mapped_object(), \p
1621 * E_RB_KEY_NOT_FOUND.
1622 *
1623 * \sa osl_storage_flags
1624 */
1625 int osl_get_row(const struct osl_table *t, unsigned col_num,
1626 const struct osl_object *obj, struct osl_row **result)
1627 {
1628 int ret;
1629 struct rb_node *node;
1630 struct osl_row *row;
1631 struct osl_column *col;
1632
1633 *result = NULL;
1634 ret = check_rbtree_col(t, col_num, &col);
1635 if (ret < 0)
1636 return ret;
1637 ret = search_rbtree(obj, t, col_num, &node, NULL);
1638 if (ret < 0)
1639 return ret;
1640 row = get_row_pointer(node, t->columns[col_num].rbtree_num);
1641 *result = row;
1642 return 1;
1643 }
1644
1645 static int rbtree_loop(struct osl_column *col, void *private_data,
1646 osl_rbtree_loop_func *func)
1647 {
1648 struct rb_node *n, *tmp;
1649
1650 /* this for-loop is safe against removal of an entry */
1651 for (n = rb_first(&col->rbtree), tmp = n? rb_next(n) : NULL;
1652 n;
1653 n = tmp, tmp = tmp? rb_next(tmp) : NULL) {
1654 struct osl_row *r = get_row_pointer(n, col->rbtree_num);
1655 int ret = func(r, private_data);
1656 if (ret < 0)
1657 return ret;
1658 }
1659 return 1;
1660 }
1661
1662 static int rbtree_loop_reverse(struct osl_column *col, void *private_data,
1663 osl_rbtree_loop_func *func)
1664 {
1665 struct rb_node *n, *tmp;
1666
1667 /* safe against removal of an entry */
1668 for (n = rb_last(&col->rbtree), tmp = n? rb_prev(n) : NULL;
1669 n;
1670 n = tmp, tmp = tmp? rb_prev(tmp) : NULL) {
1671 struct osl_row *r = get_row_pointer(n, col->rbtree_num);
1672 int ret = func(r, private_data);
1673 if (ret < 0)
1674 return ret;
1675 }
1676 return 1;
1677 }
1678
1679 /**
1680 * Loop over all nodes in an rbtree.
1681 *
1682 * \param t Pointer to an open osl table.
1683 * \param col_num The column to use for iterating over the elements.
1684 * \param private_data Pointer that gets passed to \a func.
1685 * \param func The function to be called for each node in the rbtree.
1686 *
1687 * This function does an in-order walk of the rbtree associated with \a
1688 * col_num. It is an error if the \p OSL_RBTREE flag is not set for this
1689 * column. For each node in the rbtree, the given function \a func is called
1690 * with two pointers as arguments: The first osl_row* argument points to the
1691 * row that contains the object corresponding to the rbtree node currently
1692 * traversed, and the \a private_data pointer is passed verbatim to \a func as the
1693 * second argument. The loop terminates either if \a func returns a negative
1694 * value, or if all nodes of the tree have been visited.
1695 *
1696 *
1697 * \return Positive on success, negative on errors. If the termination of the
1698 * loop was caused by \a func returning a negative value, this value is
1699 * returned.
1700 *
1701 * \sa osl_storage_flags, osl_rbtree_loop_reverse(), osl_compare_func.
1702 */
1703 int osl_rbtree_loop(const struct osl_table *t, unsigned col_num,
1704 void *private_data, osl_rbtree_loop_func *func)
1705 {
1706 struct osl_column *col;
1707
1708 int ret = check_rbtree_col(t, col_num, &col);
1709 if (ret < 0)
1710 return ret;
1711 return rbtree_loop(col, private_data, func);
1712 }
1713
1714 /**
1715 * Loop over all nodes in an rbtree in reverse order.
1716 *
1717 * \param t Identical meaning as in \p osl_rbtree_loop().
1718 * \param col_num Identical meaning as in \p osl_rbtree_loop().
1719 * \param private_data Identical meaning as in \p osl_rbtree_loop().
1720 * \param func Identical meaning as in \p osl_rbtree_loop().
1721 *
1722 * This function is identical to \p osl_rbtree_loop(), the only difference
1723 * is that the tree is walked in reverse order.
1724 *
1725 * \return The same return value as \p osl_rbtree_loop().
1726 *
1727 * \sa osl_rbtree_loop().
1728 */
1729 int osl_rbtree_loop_reverse(const struct osl_table *t, unsigned col_num,
1730 void *private_data, osl_rbtree_loop_func *func)
1731 {
1732 struct osl_column *col;
1733
1734 int ret = check_rbtree_col(t, col_num, &col);
1735 if (ret < 0)
1736 return ret;
1737 return rbtree_loop_reverse(col, private_data, func);
1738 }
1739
1740 /* TODO: Rollback changes on errors */
1741 static int rename_disk_storage_objects(struct osl_table *t,
1742 struct osl_object *old_obj, struct osl_object *new_obj)
1743 {
1744 int i, ret;
1745 const struct osl_column_description *cd;
1746 char *old_ds_name, *new_ds_name;
1747
1748 if (!t->num_disk_storage_columns)
1749 return 1; /* nothing to do */
1750 if (old_obj->size == new_obj->size && !memcmp(new_obj->data,
1751 old_obj->data, new_obj->size))
1752 return 1; /* object did not change */
1753 old_ds_name = disk_storage_name_of_object(t, old_obj);
1754 new_ds_name = disk_storage_name_of_object(t, new_obj);
1755 FOR_EACH_DISK_STORAGE_COLUMN(i, t, cd) {
1756 char *old_filename, *new_filename;
1757 ret = create_disk_storage_object_dir(t, i, new_ds_name);
1758 if (ret < 0)
1759 goto out;
1760 old_filename = disk_storage_path(t, i, old_ds_name);
1761 new_filename = disk_storage_path(t, i, new_ds_name);
1762 ret = para_rename(old_filename, new_filename);
1763 free(old_filename);
1764 free(new_filename);
1765 if (ret < 0)
1766 goto out;
1767 }
1768 ret = 1;
1769 out:
1770 free(old_ds_name);
1771 free(new_ds_name);
1772 return ret;
1773
1774 }
1775
1776 /**
1777 * Change an object in an osl table.
1778 *
1779 * \param t Pointer to an open osl table.
1780 * \param r Pointer to the row containing the object to be updated.
1781 * \param col_num Number of the column containing the object to be updated.
1782 * \param obj Pointer to the replacement object.
1783 *
1784 * This function gets rid of all references to the old object. This includes
1785 * removal of the rbtree node in case there is an rbtree associated with \a
1786 * col_num. It then inserts \a obj into the table and the rbtree if necessary.
1787 *
1788 * If the \p OSL_RBTREE flag is set for \a col_num, you \b MUST call this
1789 * function in order to change the contents of an object, even for volatile or
1790 * mapped columns of constant size (which may be updated directly if \p
1791 * OSL_RBTREE is not set). Otherwise the rbtree might become corrupted.
1792 *
1793 * \return Standard
1794 */
1795 int osl_update_object(struct osl_table *t, const struct osl_row *r,
1796 unsigned col_num, struct osl_object *obj)
1797 {
1798 struct osl_column *col;
1799 const struct osl_column_description *cd;
1800 int ret;
1801
1802 if (!t)
1803 return -E_BAD_TABLE;
1804 col = &t->columns[col_num];
1805 cd = get_column_description(t->desc, col_num);
1806 PARA_DEBUG_LOG("updating column %u of %s\n", col_num, t->desc->name);
1807 if (cd->storage_flags & OSL_RBTREE) {
1808 if (search_rbtree(obj, t, col_num, NULL, NULL) > 0)
1809 return -E_RB_KEY_EXISTS;
1810 }
1811 if (cd->storage_flags & OSL_FIXED_SIZE) {
1812 if (obj->size != cd->data_size)
1813 return -E_BAD_DATA_SIZE;
1814 }
1815 remove_rb_node(t, col_num, r);
1816 if (cd->storage_type == OSL_NO_STORAGE) { /* TODO: If fixed size, reuse object? */
1817 free(r->volatile_objects[col->volatile_num].data);
1818 r->volatile_objects[col->volatile_num] = *obj;
1819 } else if (cd->storage_type == OSL_DISK_STORAGE) {
1820 char *ds_name;
1821 ret = disk_storage_name_of_row(t, r, &ds_name);
1822 if (ret < 0)
1823 return ret;
1824 ret = delete_disk_storage_file(t, col_num, ds_name);
1825 if (ret < 0 && !is_errno(-ret, ENOENT)) {
1826 free(ds_name);
1827 return ret;
1828 }
1829 ret = write_disk_storage_file(t, col_num, obj, ds_name);
1830 free(ds_name);
1831 if (ret < 0)
1832 return ret;
1833 } else { /* mapped storage */
1834 struct osl_object old_obj;
1835 ret = get_mapped_object(t, col_num, r->num, &old_obj);
1836 if (ret < 0)
1837 return ret;
1838 /*
1839 * If the updated column is the disk storage name column, the
1840 * disk storage name changes, so we have to rename all disk
1841 * storage objects accordingly.
1842 */
1843 if (col_num == t->disk_storage_name_column) {
1844 ret = rename_disk_storage_objects(t, &old_obj, obj);
1845 if (ret < 0)
1846 return ret;
1847 }
1848 if (cd->storage_flags & OSL_FIXED_SIZE)
1849 memcpy(old_obj.data, obj->data, cd->data_size);
1850 else { /* TODO: if the size doesn't change, use old space */
1851 uint32_t new_data_map_size;
1852 char *row_index;
1853 ret = get_row_index(t, r->num, &row_index);
1854 if (ret < 0)
1855 return ret;
1856 ret = mark_mapped_object_invalid(t, r->num, col_num);
1857 if (ret < 0)
1858 return ret;
1859 unmap_column(t, col_num);
1860 ret = append_map_file(t, col_num, obj,
1861 &new_data_map_size);
1862 if (ret < 0)
1863 return ret;
1864 ret = map_column(t, col_num);
1865 if (ret < 0)
1866 return ret;
1867 update_cell_index(row_index, col, new_data_map_size,
1868 obj->size);
1869 }
1870 }
1871 if (cd->storage_flags & OSL_RBTREE) {
1872 ret = insert_rbtree(t, col_num, r, obj);
1873 if (ret < 0)
1874 return ret;
1875 }
1876 return 1;
1877 }
1878
1879 /**
1880 * Retrieve an object of type \p OSL_DISK_STORAGE by row and column.
1881 *
1882 * \param t Pointer to an open osl table.
1883 * \param r Pointer to the row containing the object.
1884 * \param col_num The column number.
1885 * \param obj Points to the result upon successful return.
1886 *
1887 * For columns of type \p OSL_DISK_STORAGE, this function must be used to
1888 * retrieve one of its containing objects. Afterwards, osl_close_disk_object()
1889 * must be called in order to deallocate the resources.
1890 *
1891 * \return Positive on success, negative on errors. Possible errors include:
1892 * \p E_BAD_TABLE, \p E_BAD_STORAGE_TYPE, errors returned by osl_get_object().
1893 *
1894 * \sa osl_get_object(), osl_storage_type, osl_close_disk_object().
1895 */
1896 int osl_open_disk_object(const struct osl_table *t, const struct osl_row *r,
1897 unsigned col_num, struct osl_object *obj)
1898 {
1899 const struct osl_column_description *cd;
1900 char *ds_name, *filename;
1901 int ret;
1902
1903 if (!t)
1904 return -E_BAD_TABLE;
1905 cd = get_column_description(t->desc, col_num);
1906 if (cd->storage_type != OSL_DISK_STORAGE)
1907 return -E_BAD_STORAGE_TYPE;
1908
1909 ret = disk_storage_name_of_row(t, r, &ds_name);
1910 if (ret < 0)
1911 return ret;
1912 filename = disk_storage_path(t, col_num, ds_name);
1913 free(ds_name);
1914 PARA_DEBUG_LOG("filename: %s\n", filename);
1915 ret = mmap_full_file(filename, O_RDONLY, &obj->data, &obj->size, NULL);
1916 free(filename);
1917 return ret;
1918 }
1919
1920 /**
1921 * Free resources that were allocated during osl_open_disk_object().
1922 *
1923 * \param obj Pointer to the object previously returned by open_disk_object().
1924 *
1925 * \return The return value of the underlying call to para_munmap().
1926 *
1927 * \sa para_munmap().
1928 */
1929 int osl_close_disk_object(struct osl_object *obj)
1930 {
1931 return para_munmap(obj->data, obj->size);
1932 }
1933
1934 /**
1935 * Get the number of rows of the given table.
1936 *
1937 * \param t Pointer to an open osl table.
1938 * \param num_rows Result is returned here.
1939 *
1940 * The number of rows returned via \a num_rows excluding any invalid rows.
1941 *
1942 * \return Positive on success, \p -E_BAD_TABLE if \a t is \p NULL.
1943 */
1944 int osl_get_num_rows(const struct osl_table *t, unsigned *num_rows)
1945 {
1946 if (!t)
1947 return -E_BAD_TABLE;
1948 assert(t->num_rows >= t->num_invalid_rows);
1949 *num_rows = t->num_rows - t->num_invalid_rows;
1950 return 1;
1951 }
1952
1953 /**
1954 * Get the rank of a row.
1955 *
1956 * \param t An open osl table.
1957 * \param r The row to get the rank of.
1958 * \param col_num The number of an rbtree column.
1959 * \param rank Result pointer.
1960 *
1961 * The rank is, by definition, the position of the row in the linear order
1962 * determined by an in-order tree walk of the rbtree associated with column
1963 * number \a col_num of \a table.
1964 *
1965 * \return Positive on success, negative on errors.
1966 *
1967 * \sa osl_get_nth_row().
1968 */
1969 int osl_get_rank(const struct osl_table *t, struct osl_row *r,
1970 unsigned col_num, unsigned *rank)
1971 {
1972 struct osl_object obj;
1973 struct osl_column *col;
1974 struct rb_node *node;
1975 int ret = check_rbtree_col(t, col_num, &col);
1976
1977 if (ret < 0)
1978 return ret;
1979 ret = osl_get_object(t, r, col_num, &obj);
1980 if (ret < 0)
1981 return ret;
1982 ret = search_rbtree(&obj, t, col_num, &node, NULL);
1983 if (ret < 0)
1984 return ret;
1985 ret = rb_rank(node, rank);
1986 if (ret < 0)
1987 return -E_BAD_ROW;
1988 return 1;
1989 }
1990
1991 /**
1992 * Get the row with n-th greatest value.
1993 *
1994 * \param t Pointer to an open osl table.
1995 * \param col_num The column number.
1996 * \param n The rank of the desired row.
1997 * \param result Row is returned here.
1998 *
1999 * Retrieve the n-th order statistic with respect to the compare function
2000 * of the rbtree column \a col_num. In other words, get that row with
2001 * \a n th greatest value in column \a col_num. It's an error if
2002 * \a col_num is not a rbtree column, or if \a n is larger than the
2003 * number of rows in the table.
2004 *
2005 * \return Positive on success, negative on errors. Possible errors:
2006 * \p E_BAD_TABLE, \p E_BAD_STORAGE_FLAGS, \p E_RB_KEY_NOT_FOUND.
2007 *
2008 * \sa osl_storage_flags, osl_compare_func, osl_get_row(),
2009 * osl_rbtree_last_row(), osl_rbtree_first_row(), osl_get_rank().
2010 */
2011 int osl_get_nth_row(const struct osl_table *t, unsigned col_num,
2012 unsigned n, struct osl_row **result)
2013 {
2014 struct osl_column *col;
2015 struct rb_node *node;
2016 int ret = check_rbtree_col(t, col_num, &col);
2017
2018 if (ret < 0)
2019 return ret;
2020 node = rb_nth(col->rbtree.rb_node, n);
2021 if (!node)
2022 return -E_RB_KEY_NOT_FOUND;
2023 *result = get_row_pointer(node, col->rbtree_num);
2024 return 1;
2025 }
2026
2027 /**
2028 * Get the row corresponding to the smallest rbtree node of a column.
2029 *
2030 * \param t An open rbtree table.
2031 * \param col_num The number of the rbtree column.
2032 * \param result A pointer to the first row is returned here.
2033 *
2034 * The rbtree node of the smallest object (with respect to the corresponding
2035 * compare function) is selected and the row containing this object is
2036 * returned. It is an error if \a col_num refers to a column without an
2037 * associated rbtree.
2038 *
2039 * \return Positive on success, negative on errors.
2040 *
2041 * \sa osl_get_nth_row(), osl_rbtree_last_row().
2042 */
2043 int osl_rbtree_first_row(const struct osl_table *t, unsigned col_num,
2044 struct osl_row **result)
2045 {
2046 return osl_get_nth_row(t, col_num, 1, result);
2047 }
2048
2049 /**
2050 * Get the row corresponding to the greatest rbtree node of a column.
2051 *
2052 * \param t The same meaning as in \p osl_rbtree_first_row().
2053 * \param col_num The same meaning as in \p osl_rbtree_first_row().
2054 * \param result The same meaning as in \p osl_rbtree_first_row().
2055 *
2056 * This function works just like osl_rbtree_first_row(), the only difference
2057 * is that the row containing the greatest rather than the smallest object is
2058 * returned.
2059 *
2060 * \return Positive on success, negative on errors.
2061 *
2062 * \sa osl_get_nth_row(), osl_rbtree_first_row().
2063 */
2064 int osl_rbtree_last_row(const struct osl_table *t, unsigned col_num,
2065 struct osl_row **result)
2066 {
2067 unsigned num_rows;
2068 int ret = osl_get_num_rows(t, &num_rows);
2069
2070 if (ret < 0)
2071 return ret;
2072 return osl_get_nth_row(t, col_num, num_rows, result);
2073 }