Fix a design flaw in osl_rbtree_loop() and osl_rbtree_loop_reverse().
[osl.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 "log.h"
13 #include "osl.h"
14 #include "error.h"
15 #include "fd.h"
16 #include "list.h"
17 #include "osl_core.h"
18
19 /**
20 * Allocate a sufficiently large string and print into it.
21 *
22 * \param fmt A usual format string.
23 *
24 * Produce output according to \p fmt. No artificial bound on the length of the
25 * resulting string is imposed.
26 *
27 * \return This function either returns a pointer to a string that must be
28 * freed by the caller or \p NULL if memory allocation failed.
29 *
30 * \sa printf(3).
31 */
32 static __must_check __printf_1_2 __malloc char *make_message(const char *fmt, ...)
33 {
34 int n;
35 size_t size = 100;
36 char *p = malloc(size);
37
38 if (!p)
39 return NULL;
40 while (1) {
41 char *q;
42 va_list ap;
43 /* Try to print in the allocated space. */
44 va_start(ap, fmt);
45 n = vsnprintf(p, size, fmt, ap);
46 va_end(ap);
47 /* If that worked, return the string. */
48 if (n > -1 && n < size)
49 break;
50 /* Else try again with more space. */
51 if (n > -1) /* glibc 2.1 */
52 size = n + 1; /* precisely what is needed */
53 else /* glibc 2.0 */
54 size *= 2; /* twice the old size */
55 q = realloc(p, size);
56 if (!q) {
57 free(p);
58 return NULL;
59 }
60 }
61 return p;
62 }
63
64 /* Taken from Drepper: How to write shared libraries, Appendix B. */
65 #include <stddef.h>
66 #define MSGSTRFIELD(line) MSGSTRFIELD1(line)
67 #define MSGSTRFIELD1(line) str##line
68 static const union msgstr_t {
69 struct {
70 #define _S(n, s) char MSGSTRFIELD(__LINE__)[sizeof(s)];
71 #include "errtab.h"
72 #undef _S
73 };
74 char str[0];
75 } msgstr = { {
76 #define _S(n, s) s,
77 #include "errtab.h"
78 #undef _S
79 } };
80 static const unsigned int errmsgidx[] = {
81 #define _S(n, s) [n] = offsetof(union msgstr_t, MSGSTRFIELD(__LINE__)),
82 #include "errtab.h"
83 #undef _S
84 };
85
86 __export const char *osl_strerror(int num)
87 {
88 if (IS_SYSTEM_ERROR(num))
89 return strerror((num) & ((1 << SYSTEM_ERROR_BIT) - 1));
90 return msgstr.str + errmsgidx[num];
91 }
92
93 /**
94 * The log function.
95 *
96 * \param ll Loglevel.
97 * \param fmt Usual format string.
98 *
99 * All XXX_LOG() macros use this function.
100 */
101 __printf_2_3 void __log(int ll, const char* fmt,...)
102 {
103 va_list argp;
104 FILE *outfd;
105 struct tm *tm;
106 time_t t1;
107 char str[255] = "";
108
109 if (ll < 2)
110 return;
111 outfd = stderr;
112 time(&t1);
113 tm = localtime(&t1);
114 strftime(str, sizeof(str), "%b %d %H:%M:%S", tm);
115 fprintf(outfd, "%s ", str);
116 va_start(argp, fmt);
117 vfprintf(outfd, fmt, argp);
118 va_end(argp);
119 }
120
121 /**
122 * A wrapper for lseek(2).
123 *
124 * \param fd The file descriptor whose offset is to be to repositioned.
125 * \param offset A value-result parameter.
126 * \param whence Usual repositioning directive.
127 *
128 * Reposition the offset of the file descriptor \a fd to the argument \a offset
129 * according to the directive \a whence. Upon successful return, \a offset
130 * contains the resulting offset location as measured in bytes from the
131 * beginning of the file.
132 *
133 * \return Positive on success. Otherwise, the function returns \p -E_OSL_LSEEK.
134 *
135 * \sa lseek(2).
136 */
137 static int __lseek(int fd, off_t *offset, int whence)
138 {
139 *offset = lseek(fd, *offset, whence);
140 int ret = -E_OSL_LSEEK;
141 if (*offset == -1)
142 return ret;
143 return 1;
144 }
145
146 /**
147 * Wrapper for the write system call.
148 *
149 * \param fd The file descriptor to write to.
150 * \param buf The buffer to write.
151 * \param size The length of \a buf in bytes.
152 *
153 * This function writes out the given buffer and retries if an interrupt
154 * occurred during the write.
155 *
156 * \return On success, the number of bytes written is returned, otherwise, the
157 * function returns \p -E_OSL_WRITE.
158 *
159 * \sa write(2).
160 */
161 static ssize_t __write(int fd, const void *buf, size_t size)
162 {
163 ssize_t ret;
164
165 for (;;) {
166 ret = write(fd, buf, size);
167 if ((ret < 0) && (errno == EAGAIN || errno == EINTR))
168 continue;
169 return ret >= 0? ret : -E_OSL_WRITE;
170 }
171 }
172
173 /**
174 * Write the whole buffer to a file descriptor.
175 *
176 * \param fd The file descriptor to write to.
177 * \param buf The buffer to write.
178 * \param size The length of \a buf in bytes.
179 *
180 * This function writes the given buffer and continues on short writes and
181 * when interrupted by a signal.
182 *
183 * \return Positive on success, negative on errors. Possible errors: any
184 * errors returned by para_write().
185 *
186 * \sa para_write().
187 */
188 static ssize_t write_all(int fd, const void *buf, size_t size)
189 {
190 // DEBUG_LOG("writing %zu bytes\n", size);
191 const char *b = buf;
192 while (size) {
193 ssize_t ret = __write(fd, b, size);
194 // DEBUG_LOG("ret: %zd\n", ret);
195 if (ret < 0)
196 return ret;
197 b += ret;
198 size -= ret;
199 }
200 return 1;
201 }
202 /**
203 * Open a file, write the given buffer and close the file.
204 *
205 * \param filename Full path to the file to open.
206 * \param buf The buffer to write to the file.
207 * \param size The size of \a buf.
208 *
209 * \return Standard.
210 */
211 static int write_file(const char *filename, const void *buf, size_t size)
212 {
213 int ret, fd;
214
215 ret = osl_open(filename, O_WRONLY | O_CREAT | O_EXCL, 0644);
216 if (ret < 0)
217 return ret;
218 fd = ret;
219 ret = write_all(fd, buf, size);
220 if (ret < 0)
221 goto out;
222 ret = 1;
223 out:
224 close(fd);
225 return ret;
226 }
227
228 static int append_file(const char *filename, char *header, size_t header_size,
229 char *data, size_t data_size, uint32_t *new_pos)
230 {
231 int ret, fd;
232
233 // DEBUG_LOG("appending %zu + %zu bytes\n", header_size, data_size);
234 ret = osl_open(filename, O_WRONLY | O_CREAT | O_APPEND, 0644);
235 if (ret < 0)
236 return ret;
237 fd = ret;
238 if (header && header_size) {
239 ret = write_all(fd, header, header_size);
240 if (ret < 0)
241 goto out;
242 }
243 ret = write_all(fd, data, data_size);
244 if (ret < 0)
245 goto out;
246 if (new_pos) {
247 off_t offset = 0;
248 ret = __lseek(fd, &offset, SEEK_END);
249 if (ret < 0)
250 goto out;
251 // DEBUG_LOG("new file size: " FMT_OFF_T "\n", offset);
252 *new_pos = offset;
253 }
254 ret = 1;
255 out:
256 close(fd);
257 return ret;
258 }
259
260 static int verify_name(const char *name)
261 {
262 if (!name)
263 return -E_OSL_BAD_NAME;
264 if (!*name)
265 return -E_OSL_BAD_NAME;
266 if (strchr(name, '/'))
267 return -E_OSL_BAD_NAME;
268 if (!strcmp(name, ".."))
269 return -E_OSL_BAD_NAME;
270 if (!strcmp(name, "."))
271 return -E_OSL_BAD_NAME;
272 return 1;
273 }
274
275 /**
276 * Compare two osl objects pointing to hash values.
277 *
278 * \param obj1 Pointer to the first hash object.
279 * \param obj2 Pointer to the second hash object.
280 *
281 * \return The values required for an osl compare function.
282 *
283 * \sa osl_compare_func, uint32_compare().
284 */
285 int osl_hash_compare(const struct osl_object *obj1, const struct osl_object *obj2)
286 {
287 return hash_compare((HASH_TYPE *)obj1->data, (HASH_TYPE *)obj2->data);
288 }
289
290 static char *disk_storage_dirname(const struct osl_table *t, unsigned col_num,
291 const char *ds_name)
292 {
293 char *dirname, *column_name = column_filename(t, col_num);
294
295 if (!column_name)
296 return NULL;
297 if (!(t->desc->flags & OSL_LARGE_TABLE))
298 return column_name;
299 dirname = make_message("%s/%.2s", column_name, ds_name);
300 free(column_name);
301 return dirname;
302 }
303
304 static char *disk_storage_name_of_object(const struct osl_table *t,
305 const struct osl_object *obj)
306 {
307 HASH_TYPE hash[HASH_SIZE];
308 hash_object(obj, hash);
309 return disk_storage_name_of_hash(t, hash);
310 }
311
312 static int disk_storage_name_of_row(const struct osl_table *t,
313 const struct osl_row *row, char **name)
314 {
315 struct osl_object obj;
316 int ret = osl_get_object(t, row, t->disk_storage_name_column, &obj);
317
318 if (ret < 0)
319 return ret;
320 *name = disk_storage_name_of_object(t, &obj);
321 if (*name)
322 return 1;
323 return -ERRNO_TO_ERROR(ENOMEM);
324 }
325
326 static void column_name_hash(const char *col_name, HASH_TYPE *hash)
327 {
328 hash_function(col_name, strlen(col_name), hash);
329 }
330
331 static int init_column_descriptions(struct osl_table *t)
332 {
333 int i, j, ret;
334 const struct osl_column_description *cd;
335
336 ret = -E_OSL_BAD_TABLE_DESC;
337 ret = verify_name(t->desc->name);
338 if (ret < 0)
339 goto err;
340 ret = -E_OSL_BAD_DB_DIR;
341 if (!t->desc->dir && (t->num_disk_storage_columns || t->num_mapped_columns))
342 goto err;
343 /* the size of the index header without column descriptions */
344 t->index_header_size = IDX_COLUMN_DESCRIPTIONS;
345 FOR_EACH_COLUMN(i, t->desc, cd) {
346 struct osl_column *col = t->columns + i;
347 if (cd->storage_flags & OSL_RBTREE) {
348 if (!cd->compare_function)
349 return -E_OSL_NO_COMPARE_FUNC;
350 }
351 if (cd->storage_type == OSL_NO_STORAGE)
352 continue;
353 ret = -E_OSL_NO_COLUMN_NAME;
354 if (!cd->name || !cd->name[0])
355 goto err;
356 ret = verify_name(cd->name);
357 if (ret < 0)
358 goto err;
359 t->index_header_size += index_column_description_size(cd->name);
360 column_name_hash(cd->name, col->name_hash);
361 ret = -E_OSL_DUPLICATE_COL_NAME;
362 for (j = i + 1; j < t->desc->num_columns; j++) {
363 const char *name2 = get_column_description(t->desc,
364 j)->name;
365 if (cd->name && name2 && !strcmp(cd->name, name2))
366 goto err;
367 }
368 }
369 return 1;
370 err:
371 return ret;
372 }
373
374 /**
375 * Initialize a struct table from given table description.
376 *
377 * \param desc The description of the osl table.
378 * \param table_ptr Result is returned here.
379 *
380 * This function performs several sanity checks on \p desc and returns if any
381 * of these tests fail. On success, a struct \p osl_table is allocated and
382 * initialized with data derived from \p desc.
383 *
384 * \return Standard.
385 *
386 * \sa struct osl_table.
387 */
388 int init_table_structure(const struct osl_table_description *desc,
389 struct osl_table **table_ptr)
390 {
391 const struct osl_column_description *cd;
392 struct osl_table *t = calloc(1, sizeof(*t));
393 int i, ret = -ERRNO_TO_ERROR(ENOMEM), have_disk_storage_name_column = 0;
394
395 if (!t)
396 return ret;
397 ret = -E_OSL_BAD_TABLE_DESC;
398 if (!desc)
399 goto err;
400 DEBUG_LOG("creating table structure for '%s' from table "
401 "description\n", desc->name);
402 ret = -E_OSL_NO_COLUMN_DESC;
403 if (!desc->column_descriptions)
404 goto err;
405 ret = -E_OSL_NO_COLUMNS;
406 if (!desc->num_columns)
407 goto err;
408 ret = -ERRNO_TO_ERROR(ENOMEM);
409 t->columns = calloc(desc->num_columns, sizeof(struct osl_column));
410 if (!t->columns)
411 goto err;
412 t->desc = desc;
413 FOR_EACH_COLUMN(i, t->desc, cd) {
414 enum osl_storage_type st = cd->storage_type;
415 enum osl_storage_flags sf = cd->storage_flags;
416 struct osl_column *col = &t->columns[i];
417
418 ret = -E_OSL_BAD_STORAGE_TYPE;
419 if (st != OSL_MAPPED_STORAGE && st != OSL_DISK_STORAGE
420 && st != OSL_NO_STORAGE)
421 goto err;
422 ret = -E_OSL_BAD_STORAGE_FLAGS;
423 if (st == OSL_DISK_STORAGE && sf & OSL_RBTREE)
424 goto err;
425 ret = -E_OSL_BAD_STORAGE_SIZE;
426 if (sf & OSL_FIXED_SIZE && !cd->data_size)
427 goto err;
428 switch (st) {
429 case OSL_DISK_STORAGE:
430 t->num_disk_storage_columns++;
431 break;
432 case OSL_MAPPED_STORAGE:
433 t->num_mapped_columns++;
434 col->index_offset = t->row_index_size;
435 t->row_index_size += 8;
436 break;
437 case OSL_NO_STORAGE:
438 col->volatile_num = t->num_volatile_columns;
439 t->num_volatile_columns++;
440 break;
441 }
442 if (sf & OSL_RBTREE) {
443 col->rbtree_num = t->num_rbtrees;
444 t->num_rbtrees++;
445 if ((sf & OSL_UNIQUE) && (st == OSL_MAPPED_STORAGE)) {
446 if (!have_disk_storage_name_column)
447 t->disk_storage_name_column = i;
448 have_disk_storage_name_column = 1;
449 }
450 }
451 }
452 ret = -E_OSL_NO_UNIQUE_RBTREE_COLUMN;
453 if (t->num_disk_storage_columns && !have_disk_storage_name_column)
454 goto err;
455 ret = -E_OSL_NO_RBTREE_COL;
456 if (!t->num_rbtrees)
457 goto err;
458 /* success */
459 DEBUG_LOG("OK. Index entry size: %u\n", t->row_index_size);
460 ret = init_column_descriptions(t);
461 if (ret < 0)
462 goto err;
463 *table_ptr = t;
464 return 1;
465 err:
466 free(t->columns);
467 free(t);
468 return ret;
469 }
470
471 /**
472 * Read the table description from index header.
473 *
474 * \param map The memory mapping of the index file.
475 * \param desc The values found in the index header are returned here.
476 *
477 * Read the index header, check for the paraslash magic string and the table version number.
478 * Read all information stored in the index header into \a desc.
479 *
480 * \return Standard.
481 *
482 * \sa struct osl_table_description, osl_create_table.
483 */
484 int read_table_desc(struct osl_object *map, struct osl_table_description *desc)
485 {
486 char *buf = map->data;
487 uint8_t version;
488 uint16_t header_size;
489 int ret, i;
490 unsigned offset;
491 struct osl_column_description *cd;
492
493 if (map->size < MIN_INDEX_HEADER_SIZE(1))
494 return -E_OSL_SHORT_TABLE;
495 if (strncmp(buf + IDX_PARA_MAGIC, PARA_MAGIC, strlen(PARA_MAGIC)))
496 return -E_OSL_NO_MAGIC;
497 version = read_u8(buf + IDX_VERSION);
498 if (version < MIN_TABLE_VERSION || version > MAX_TABLE_VERSION)
499 return -E_OSL_VERSION_MISMATCH;
500 desc->flags = read_u8(buf + IDX_TABLE_FLAGS);
501 desc->num_columns = read_u16(buf + IDX_NUM_COLUMNS);
502 DEBUG_LOG("%u columns\n", desc->num_columns);
503 if (!desc->num_columns)
504 return -E_OSL_NO_COLUMNS;
505 header_size = read_u16(buf + IDX_HEADER_SIZE);
506 if (map->size < header_size)
507 return -E_OSL_BAD_SIZE;
508 desc->column_descriptions = calloc(desc->num_columns,
509 sizeof(struct osl_column_description));
510 if (!desc->column_descriptions)
511 return -ERRNO_TO_ERROR(ENOMEM);
512 offset = IDX_COLUMN_DESCRIPTIONS;
513 FOR_EACH_COLUMN(i, desc, cd) {
514 char *null_byte;
515
516 ret = -E_OSL_SHORT_TABLE;
517 if (map->size < offset + MIN_IDX_COLUMN_DESCRIPTION_SIZE) {
518 ERROR_LOG("map size = %zu < %u = offset + min desc size\n",
519 map->size, offset + MIN_IDX_COLUMN_DESCRIPTION_SIZE);
520 goto err;
521 }
522 cd->storage_type = read_u16(buf + offset + IDX_CD_STORAGE_TYPE);
523 cd->storage_flags = read_u16(buf + offset +
524 IDX_CD_STORAGE_FLAGS);
525 cd->data_size = read_u32(buf + offset + IDX_CD_DATA_SIZE);
526 null_byte = memchr(buf + offset + IDX_CD_NAME, '\0',
527 map->size - offset - IDX_CD_NAME);
528 ret = -E_OSL_INDEX_CORRUPTION;
529 if (!null_byte)
530 goto err;
531 ret = -ERRNO_TO_ERROR(ENOMEM);
532 cd->name = strdup(buf + offset + IDX_CD_NAME);
533 if (!cd->name)
534 goto err;
535 offset += index_column_description_size(cd->name);
536 }
537 if (offset != header_size) {
538 ret = -E_OSL_INDEX_CORRUPTION;
539 ERROR_LOG("real header size = %u != %u = stored header size\n",
540 offset, header_size);
541 goto err;
542 }
543 return 1;
544 err:
545 FOR_EACH_COLUMN(i, desc, cd)
546 free(cd->name);
547 return ret;
548 }
549
550 /*
551 * check whether the table description given by \p t->desc matches the on-disk
552 * table structure stored in the index of \a t.
553 */
554 static int compare_table_descriptions(struct osl_table *t)
555 {
556 int i, ret;
557 struct osl_table_description desc;
558 const struct osl_column_description *cd1, *cd2;
559
560 /* read the on-disk structure into desc */
561 ret = read_table_desc(&t->index_map, &desc);
562 if (ret < 0)
563 return ret;
564 ret = -E_OSL_BAD_TABLE_FLAGS;
565 if (desc.flags != t->desc->flags)
566 goto out;
567 ret = -E_OSL_BAD_COLUMN_NUM;
568 if (desc.num_columns > t->desc->num_columns)
569 goto out;
570 if (desc.num_columns < t->desc->num_columns) {
571 struct osl_column_description *cd;
572 unsigned diff = t->desc->num_columns - desc.num_columns;
573 INFO_LOG("extending table by %u volatile columns\n", diff);
574 ret = -ERRNO_TO_ERROR(ENOMEM);
575 desc.column_descriptions = realloc(desc.column_descriptions,
576 t->desc->num_columns * sizeof(struct osl_column_description));
577 if (!desc.column_descriptions)
578 goto out;
579 for (i = desc.num_columns; i < t->desc->num_columns; i++) {
580 cd = get_column_description(&desc, i);
581 cd->storage_type = OSL_NO_STORAGE;
582 cd->name = NULL;
583 }
584 desc.num_columns += diff;
585 }
586 FOR_EACH_COLUMN(i, t->desc, cd1) {
587 cd2 = get_column_description(&desc, i);
588 ret = -E_OSL_BAD_STORAGE_TYPE;
589 if (cd1->storage_type != cd2->storage_type)
590 goto out;
591 if (cd1->storage_type == OSL_NO_STORAGE)
592 continue;
593 ret = -E_OSL_BAD_STORAGE_FLAGS;
594 if (cd1->storage_flags != cd2->storage_flags) {
595 ERROR_LOG("sf1 = %u != %u = sf2\n",
596 cd1->storage_flags, cd2->storage_flags);
597 goto out;
598 }
599 ret = -E_OSL_BAD_DATA_SIZE;
600 if (cd1->storage_flags & OSL_FIXED_SIZE)
601 if (cd1->data_size != cd2->data_size)
602 goto out;
603 ret = -E_OSL_BAD_COLUMN_NAME;
604 if (strcmp(cd1->name, cd2->name))
605 goto out;
606 }
607 DEBUG_LOG("table description of '%s' matches on-disk data, good\n",
608 t->desc->name);
609 ret = 1;
610 out:
611 FOR_EACH_COLUMN(i, &desc, cd1)
612 free(cd1->name);
613 free(desc.column_descriptions);
614 return ret;
615 }
616
617 static int create_table_index(struct osl_table *t)
618 {
619 char *buf, *filename;
620 int i, ret;
621 size_t size = t->index_header_size;
622 const struct osl_column_description *cd;
623 unsigned offset;
624
625 INFO_LOG("creating %zu byte index for table %s\n", size,
626 t->desc->name);
627 buf = calloc(1, size);
628 if (!buf)
629 return -ERRNO_TO_ERROR(ENOMEM);
630 sprintf(buf + IDX_PARA_MAGIC, "%s", PARA_MAGIC);
631 write_u8(buf + IDX_TABLE_FLAGS, t->desc->flags);
632 write_u8(buf + IDX_DIRTY_FLAG, 0);
633 write_u8(buf + IDX_VERSION, CURRENT_TABLE_VERSION);
634 write_u16(buf + IDX_NUM_COLUMNS, t->num_mapped_columns + t->num_disk_storage_columns);
635 write_u16(buf + IDX_HEADER_SIZE, t->index_header_size);
636 offset = IDX_COLUMN_DESCRIPTIONS;
637 FOR_EACH_COLUMN(i, t->desc, cd) {
638 /* no need to store info about volatile storage */
639 if (cd->storage_type == OSL_NO_STORAGE)
640 continue;
641 write_u16(buf + offset + IDX_CD_STORAGE_TYPE,
642 cd->storage_type);
643 write_u16(buf + offset + IDX_CD_STORAGE_FLAGS,
644 cd->storage_flags);
645 if (cd->storage_flags & OSL_FIXED_SIZE)
646 write_u32(buf + offset + IDX_CD_DATA_SIZE,
647 cd->data_size);
648 strcpy(buf + offset + IDX_CD_NAME, cd->name);
649 offset += index_column_description_size(cd->name);
650 }
651 assert(offset = size);
652 filename = index_filename(t->desc);
653 if (filename)
654 ret = write_file(filename, buf, size);
655 else
656 ret = -ERRNO_TO_ERROR(ENOMEM);
657 free(buf);
658 free(filename);
659 return ret;
660 }
661
662 /**
663 * Create a new osl table.
664 *
665 * \param desc Pointer to the table description.
666 *
667 * \return Standard.
668 */
669 __export int osl_create_table(const struct osl_table_description *desc)
670 {
671 const struct osl_column_description *cd;
672 char *table_dir = NULL, *filename;
673 struct osl_table *t;
674 int i, ret = init_table_structure(desc, &t);
675
676 if (ret < 0)
677 return ret;
678 INFO_LOG("creating %s\n", desc->name);
679 FOR_EACH_COLUMN(i, t->desc, cd) {
680 if (cd->storage_type == OSL_NO_STORAGE)
681 continue;
682 if (!table_dir) {
683 ret = para_mkdir(desc->dir, 0777);
684 if (ret < 0 && !is_errno(-ret, EEXIST))
685 goto out;
686 table_dir = make_message("%s/%s", desc->dir,
687 desc->name);
688 ret = -ERRNO_TO_ERROR(ENOMEM);
689 if (!table_dir)
690 goto out;
691 ret = para_mkdir(table_dir, 0777);
692 if (ret < 0)
693 goto out;
694 }
695 ret = -ERRNO_TO_ERROR(ENOMEM);
696 filename = column_filename(t, i);
697 if (!filename)
698 goto out;
699 INFO_LOG("filename: %s\n", filename);
700 if (cd->storage_type == OSL_MAPPED_STORAGE) {
701 ret = osl_open(filename, O_RDWR | O_CREAT | O_EXCL,
702 0644);
703 free(filename);
704 if (ret < 0)
705 goto out;
706 close(ret);
707 continue;
708 }
709 /* DISK STORAGE */
710 ret = para_mkdir(filename, 0777);
711 free(filename);
712 if (ret < 0)
713 goto out;
714 }
715 if (t->num_mapped_columns) {
716 ret = create_table_index(t);
717 if (ret < 0)
718 goto out;
719 }
720 ret = 1;
721 out:
722 free(table_dir);
723 free(t->columns);
724 free(t);
725 return ret;
726 }
727
728 static int table_is_dirty(struct osl_table *t)
729 {
730 char *buf = (char *)t->index_map.data + IDX_DIRTY_FLAG;
731 uint8_t dirty = read_u8(buf) & 0x1;
732 return !!dirty;
733 }
734
735 static void mark_table_dirty(struct osl_table *t)
736 {
737 char *buf = (char *)t->index_map.data + IDX_DIRTY_FLAG;
738 write_u8(buf, read_u8(buf) | 1);
739 }
740
741 static void mark_table_clean(struct osl_table *t)
742 {
743 char *buf = (char *)t->index_map.data + IDX_DIRTY_FLAG;
744 write_u8(buf, read_u8(buf) & 0xfe);
745 }
746
747 static void unmap_column(struct osl_table *t, unsigned col_num)
748 {
749 struct osl_object map = t->columns[col_num].data_map;
750 int ret;
751 if (!map.data)
752 return;
753 ret = para_munmap(map.data, map.size);
754 assert(ret > 0);
755 map.data = NULL;
756 }
757
758 /**
759 * Unmap all mapped files of an osl table.
760 *
761 * \param t Pointer to a mapped table.
762 * \param flags Options for unmapping.
763 *
764 * \return Positive on success, negative on errors.
765 *
766 * \sa map_table(), enum osl_close_flags, para_munmap().
767 */
768 int unmap_table(struct osl_table *t, enum osl_close_flags flags)
769 {
770 unsigned i;
771 const struct osl_column_description *cd;
772 int ret;
773
774 if (!t->num_mapped_columns) /* can this ever happen? */
775 return 1;
776 DEBUG_LOG("unmapping table '%s'\n", t->desc->name);
777 if (!t->index_map.data)
778 return -E_OSL_NOT_MAPPED;
779 if (flags & OSL_MARK_CLEAN)
780 mark_table_clean(t);
781 ret = para_munmap(t->index_map.data, t->index_map.size);
782 if (ret < 0)
783 return ret;
784 t->index_map.data = NULL;
785 if (!t->num_rows)
786 return 1;
787 FOR_EACH_MAPPED_COLUMN(i, t, cd)
788 unmap_column(t, i);
789 return 1;
790 }
791
792 static int map_column(struct osl_table *t, unsigned col_num)
793 {
794 struct stat statbuf;
795 char *filename = column_filename(t, col_num);
796 int ret = -E_OSL_STAT;
797
798 if (!filename)
799 return -ERRNO_TO_ERROR(ENOMEM);
800 if (stat(filename, &statbuf) < 0) {
801 free(filename);
802 return ret;
803 }
804 if (!(S_IFREG & statbuf.st_mode)) {
805 free(filename);
806 return ret;
807 }
808 ret = mmap_full_file(filename, O_RDWR,
809 &t->columns[col_num].data_map.data,
810 &t->columns[col_num].data_map.size,
811 NULL);
812 free(filename);
813 return ret;
814 }
815
816 /**
817 * Map the index file and all columns of type \p OSL_MAPPED_STORAGE into memory.
818 *
819 * \param t Pointer to an initialized table structure.
820 * \param flags Mapping options.
821 *
822 * \return Negative return value on errors; on success the number of rows
823 * (including invalid rows) is returned.
824 *
825 * \sa unmap_table(), enum map_table_flags, osl_open_table(), mmap(2).
826 */
827 int map_table(struct osl_table *t, enum map_table_flags flags)
828 {
829 char *filename;
830 const struct osl_column_description *cd;
831 int i = 0, ret, num_rows = 0;
832
833 if (!t->num_mapped_columns)
834 return 0;
835 if (t->index_map.data)
836 return -E_OSL_ALREADY_MAPPED;
837 filename = index_filename(t->desc);
838 if (!filename)
839 return -ERRNO_TO_ERROR(ENOMEM);
840 DEBUG_LOG("mapping table '%s' (index: %s)\n", t->desc->name, filename);
841 ret = mmap_full_file(filename, flags & MAP_TBL_FL_MAP_RDONLY?
842 O_RDONLY : O_RDWR, &t->index_map.data, &t->index_map.size, NULL);
843 free(filename);
844 if (ret < 0)
845 return ret;
846 if (flags & MAP_TBL_FL_VERIFY_INDEX) {
847 ret = compare_table_descriptions(t);
848 if (ret < 0)
849 goto err;
850 }
851 ret = -E_OSL_BUSY;
852 if (!(flags & MAP_TBL_FL_IGNORE_DIRTY)) {
853 if (table_is_dirty(t)) {
854 ERROR_LOG("%s is dirty\n", t->desc->name);
855 goto err;
856 }
857 }
858 mark_table_dirty(t);
859 num_rows = table_num_rows(t);
860 if (!num_rows)
861 return num_rows;
862 /* map data files */
863 FOR_EACH_MAPPED_COLUMN(i, t, cd) {
864 ret = map_column(t, i);
865 if (ret < 0)
866 goto err;
867 }
868 return num_rows;
869 err: /* unmap what is already mapped */
870 for (i--; i >= 0; i--) {
871 struct osl_object map = t->columns[i].data_map;
872 para_munmap(map.data, map.size);
873 map.data = NULL;
874 }
875 para_munmap(t->index_map.data, t->index_map.size);
876 t->index_map.data = NULL;
877 return ret;
878 }
879
880 /**
881 * Retrieve a mapped object by row and column number.
882 *
883 * \param t Pointer to an open osl table.
884 * \param col_num Number of the mapped column containing the object to retrieve.
885 * \param row_num Number of the row containing the object to retrieve.
886 * \param obj The result is returned here.
887 *
888 * It is considered an error if \a col_num does not refer to a column
889 * of storage type \p OSL_MAPPED_STORAGE.
890 *
891 * \return Standard.
892 *
893 * \sa osl_storage_type.
894 */
895 int get_mapped_object(const struct osl_table *t, unsigned col_num,
896 uint32_t row_num, struct osl_object *obj)
897 {
898 struct osl_column *col = &t->columns[col_num];
899 uint32_t offset;
900 char *header;
901 char *cell_index;
902 int ret;
903
904 if (t->num_rows <= row_num)
905 return -E_OSL_BAD_ROW_NUM;
906 ret = get_cell_index(t, row_num, col_num, &cell_index);
907 if (ret < 0)
908 return ret;
909 offset = read_u32(cell_index);
910 obj->size = read_u32(cell_index + 4) - 1;
911 header = col->data_map.data + offset;
912 obj->data = header + 1;
913 if (read_u8(header) == 0xff) {
914 ERROR_LOG("col %u, size %zu, offset %u\n", col_num,
915 obj->size, offset);
916 return -E_OSL_INVALID_OBJECT;
917 }
918 return 1;
919 }
920
921 static int search_rbtree(const struct osl_object *obj,
922 const struct osl_table *t, unsigned col_num,
923 struct rb_node **result, struct rb_node ***rb_link)
924 {
925 struct osl_column *col = &t->columns[col_num];
926 struct rb_node **new = &col->rbtree.rb_node, *parent = NULL;
927 const struct osl_column_description *cd =
928 get_column_description(t->desc, col_num);
929 enum osl_storage_type st = cd->storage_type;
930 while (*new) {
931 struct osl_row *this_row = get_row_pointer(*new,
932 col->rbtree_num);
933 int ret;
934 struct osl_object this_obj;
935 parent = *new;
936 if (st == OSL_MAPPED_STORAGE) {
937 ret = get_mapped_object(t, col_num, this_row->num,
938 &this_obj);
939 if (ret < 0)
940 return ret;
941 } else
942 this_obj = this_row->volatile_objects[col->volatile_num];
943 ret = cd->compare_function(obj, &this_obj);
944 if (!ret) {
945 if (result)
946 *result = get_rb_node_pointer(this_row,
947 col->rbtree_num);
948 return 1;
949 }
950 if (ret < 0)
951 new = &((*new)->rb_left);
952 else
953 new = &((*new)->rb_right);
954 }
955 if (result)
956 *result = parent;
957 if (rb_link)
958 *rb_link = new;
959 return -E_OSL_RB_KEY_NOT_FOUND;
960 }
961
962 static int insert_rbtree(struct osl_table *t, unsigned col_num,
963 const struct osl_row *row, const struct osl_object *obj)
964 {
965 struct rb_node *parent, **rb_link;
966 unsigned rbtree_num;
967 struct rb_node *n;
968 int ret = search_rbtree(obj, t, col_num, &parent, &rb_link);
969
970 if (ret > 0)
971 return -E_OSL_RB_KEY_EXISTS;
972 rbtree_num = t->columns[col_num].rbtree_num;
973 n = get_rb_node_pointer(row, rbtree_num);
974 rb_link_node(n, parent, rb_link);
975 rb_insert_color(n, &t->columns[col_num].rbtree);
976 return 1;
977 }
978
979 static void remove_rb_node(struct osl_table *t, unsigned col_num,
980 const struct osl_row *row)
981 {
982 struct osl_column *col = &t->columns[col_num];
983 const struct osl_column_description *cd =
984 get_column_description(t->desc, col_num);
985 enum osl_storage_flags sf = cd->storage_flags;
986 struct rb_node *victim, *splice_out_node, *tmp;
987 if (!(sf & OSL_RBTREE))
988 return;
989 /*
990 * Which node is removed/spliced out actually depends on how many
991 * children the victim node has: If it has no children, it gets
992 * deleted. If it has one child, it gets spliced out. If it has two
993 * children, its successor (which has at most a right child) gets
994 * spliced out.
995 */
996 victim = get_rb_node_pointer(row, col->rbtree_num);
997 if (victim->rb_left && victim->rb_right)
998 splice_out_node = rb_next(victim);
999 else
1000 splice_out_node = victim;
1001 /* Go up to the root and decrement the size of each node in the path. */
1002 for (tmp = splice_out_node; tmp; tmp = rb_parent(tmp))
1003 tmp->size--;
1004 rb_erase(victim, &col->rbtree);
1005 }
1006
1007 static int add_row_to_rbtrees(struct osl_table *t, uint32_t row_num,
1008 struct osl_object *volatile_objs, struct osl_row **row_ptr)
1009 {
1010 unsigned i;
1011 int ret;
1012 struct osl_row *row = allocate_row(t->num_rbtrees);
1013 const struct osl_column_description *cd;
1014
1015 if (!row)
1016 return -ERRNO_TO_ERROR(ENOMEM);
1017 row->num = row_num;
1018 row->volatile_objects = volatile_objs;
1019 FOR_EACH_RBTREE_COLUMN(i, t, cd) {
1020 if (cd->storage_type == OSL_MAPPED_STORAGE) {
1021 struct osl_object obj;
1022 ret = get_mapped_object(t, i, row_num, &obj);
1023 if (ret < 0)
1024 goto err;
1025 ret = insert_rbtree(t, i, row, &obj);
1026 } else { /* volatile */
1027 const struct osl_object *obj
1028 = volatile_objs + t->columns[i].volatile_num;
1029 ret = insert_rbtree(t, i, row, obj);
1030 }
1031 if (ret < 0)
1032 goto err;
1033 }
1034 if (row_ptr)
1035 *row_ptr = row;
1036 return 1;
1037 err: /* rollback changes, i.e. remove added entries from rbtrees */
1038 while (i)
1039 remove_rb_node(t, i--, row);
1040 free(row);
1041 return ret;
1042 }
1043
1044 static void free_volatile_objects(const struct osl_table *t,
1045 enum osl_close_flags flags)
1046 {
1047 int i, j;
1048 struct rb_node *n;
1049 struct osl_column *rb_col;
1050 const struct osl_column_description *cd;
1051
1052 if (!t->num_volatile_columns)
1053 return;
1054 /* find the first rbtree column (any will do) */
1055 FOR_EACH_RBTREE_COLUMN(i, t, cd)
1056 break;
1057 rb_col = t->columns + i;
1058 /* walk that rbtree and free all volatile objects */
1059 for (n = rb_first(&rb_col->rbtree); n; n = rb_next(n)) {
1060 struct osl_row *r = get_row_pointer(n, rb_col->rbtree_num);
1061 if (flags & OSL_FREE_VOLATILE)
1062 FOR_EACH_VOLATILE_COLUMN(j, t, cd) {
1063 if (cd->storage_flags & OSL_DONT_FREE)
1064 continue;
1065 free(r->volatile_objects[
1066 t->columns[j].volatile_num].data);
1067 }
1068 // for (j = 0; j < t->num_volatile_columns; j++)
1069 // free(r->volatile_objects[j].data);
1070 free(r->volatile_objects);
1071 }
1072 }
1073
1074 /**
1075 * Erase all rbtree nodes and free resources.
1076 *
1077 * \param t Pointer to an open osl table.
1078 *
1079 * This function is called by osl_close_table().
1080 */
1081 void clear_rbtrees(struct osl_table *t)
1082 {
1083 const struct osl_column_description *cd;
1084 unsigned i, rbtrees_cleared = 0;
1085
1086 FOR_EACH_RBTREE_COLUMN(i, t, cd) {
1087 struct osl_column *col = &t->columns[i];
1088 struct rb_node *n;
1089 rbtrees_cleared++;
1090 for (n = rb_first(&col->rbtree); n;) {
1091 struct osl_row *r;
1092 rb_erase(n, &col->rbtree);
1093 if (rbtrees_cleared == t->num_rbtrees) {
1094 r = get_row_pointer(n, col->rbtree_num);
1095 n = rb_next(n);
1096 free(r);
1097 } else
1098 n = rb_next(n);
1099 }
1100 }
1101
1102 }
1103
1104 /**
1105 * Close an osl table.
1106 *
1107 * \param t Pointer to the table to be closed.
1108 * \param flags Options for what should be cleaned up.
1109 *
1110 * If osl_open_table() succeeds, the resulting table pointer must later be
1111 * passed to this function in order to flush all changes to the file system and
1112 * to free the resources that were allocated by osl_open_table().
1113 *
1114 * \return Standard.
1115 *
1116 * \sa osl_open_table(), unmap_table().
1117 */
1118 __export int osl_close_table(struct osl_table *t, enum osl_close_flags flags)
1119 {
1120 int ret;
1121
1122 if (!t)
1123 return -E_OSL_BAD_TABLE;
1124 free_volatile_objects(t, flags);
1125 clear_rbtrees(t);
1126 ret = unmap_table(t, flags);
1127 if (ret < 0)
1128 ERROR_LOG("unmap_table failed: %d\n", ret);
1129 free(t->columns);
1130 free(t);
1131 return ret;
1132 }
1133
1134 /**
1135 * Find out whether the given row number corresponds to an invalid row.
1136 *
1137 * \param t Pointer to the osl table.
1138 * \param row_num The number of the row in question.
1139 *
1140 * By definition, a row is considered invalid if all its index entries
1141 * are invalid.
1142 *
1143 * \return Positive if \a row_num corresponds to an invalid row,
1144 * zero if it corresponds to a valid row, negative on errors.
1145 */
1146 int row_is_invalid(struct osl_table *t, uint32_t row_num)
1147 {
1148 char *row_index;
1149 int i, ret = get_row_index(t, row_num, &row_index);
1150
1151 if (ret < 0)
1152 return ret;
1153 for (i = 0; i < t->row_index_size; i++) {
1154 if ((unsigned char)row_index[i] != 0xff)
1155 return 0;
1156 }
1157 INFO_LOG("row %d is invalid\n", row_num);
1158 return 1;
1159 }
1160
1161 /**
1162 * Invalidate a row of an osl table.
1163 *
1164 * \param t Pointer to an open osl table.
1165 * \param row_num Number of the row to mark as invalid.
1166 *
1167 * This function marks each mapped object in the index entry of \a row as
1168 * invalid.
1169 *
1170 * \return Standard.
1171 */
1172 int mark_row_invalid(struct osl_table *t, uint32_t row_num)
1173 {
1174 char *row_index;
1175 int ret = get_row_index(t, row_num, &row_index);
1176
1177 if (ret < 0)
1178 return ret;
1179 INFO_LOG("marking row %d as invalid\n", row_num);
1180 memset(row_index, 0xff, t->row_index_size);
1181 return 1;
1182 }
1183
1184 /**
1185 * Initialize all rbtrees and compute number of invalid rows.
1186 *
1187 * \param t The table containing the rbtrees to be initialized.
1188 *
1189 * \return Standard.
1190 */
1191 int init_rbtrees(struct osl_table *t)
1192 {
1193 int i, ret;
1194 const struct osl_column_description *cd;
1195
1196 /* create rbtrees */
1197 FOR_EACH_RBTREE_COLUMN(i, t, cd)
1198 t->columns[i].rbtree = RB_ROOT;
1199 /* add valid rows to rbtrees */
1200 t->num_invalid_rows = 0;
1201 for (i = 0; i < t->num_rows; i++) {
1202 ret = row_is_invalid(t, i);
1203 if (ret < 0)
1204 return ret;
1205 if (ret) {
1206 t->num_invalid_rows++;
1207 continue;
1208 }
1209 ret = add_row_to_rbtrees(t, i, NULL, NULL);
1210 if (ret < 0)
1211 return ret;
1212 }
1213 return 1;
1214 }
1215
1216 /**
1217 * Open an osl table.
1218 *
1219 * Each osl table must be opened before its data can be accessed.
1220 *
1221 * \param table_desc Describes the table to be opened.
1222 * \param result Contains a pointer to the open table on success.
1223 *
1224 * The table description given by \a desc should coincide with the
1225 * description used at creation time.
1226 *
1227 * \return Standard.
1228 */
1229 __export int osl_open_table(const struct osl_table_description *table_desc,
1230 struct osl_table **result)
1231 {
1232 int i, ret;
1233 struct osl_table *t;
1234 const struct osl_column_description *cd;
1235
1236 INFO_LOG("opening table %s\n", table_desc->name);
1237 ret = init_table_structure(table_desc, &t);
1238 if (ret < 0)
1239 return ret;
1240 FOR_EACH_DISK_STORAGE_COLUMN(i, t, cd) {
1241 struct stat statbuf;
1242 char *dirname = column_filename(t, i);
1243
1244 ret = -ERRNO_TO_ERROR(ENOMEM);
1245 if (!dirname)
1246 goto err;
1247 /* check if directory exists */
1248 ret = stat(dirname, &statbuf);
1249 free(dirname);
1250 if (ret < 0) {
1251 ret = -ERRNO_TO_ERROR(errno);
1252 goto err;
1253 }
1254 ret = -ERRNO_TO_ERROR(ENOTDIR);
1255 if (!S_ISDIR(statbuf.st_mode))
1256 goto err;
1257 }
1258 ret = map_table(t, MAP_TBL_FL_VERIFY_INDEX);
1259 if (ret < 0)
1260 goto err;
1261 t->num_rows = ret;
1262 DEBUG_LOG("num rows: %d\n", t->num_rows);
1263 ret = init_rbtrees(t);
1264 if (ret < 0) {
1265 osl_close_table(t, OSL_MARK_CLEAN); /* ignore further errors */
1266 return ret;
1267 }
1268 *result = t;
1269 return 1;
1270 err:
1271 free(t->columns);
1272 free(t);
1273 return ret;
1274 }
1275
1276 static int create_disk_storage_object_dir(const struct osl_table *t,
1277 unsigned col_num, const char *ds_name)
1278 {
1279 char *dirname;
1280 int ret;
1281
1282 if (!(t->desc->flags & OSL_LARGE_TABLE))
1283 return 1;
1284 dirname = disk_storage_dirname(t, col_num, ds_name);
1285 if (!dirname)
1286 return -ERRNO_TO_ERROR(ENOMEM);
1287 ret = para_mkdir(dirname, 0777);
1288 free(dirname);
1289 if (ret < 0 && !is_errno(-ret, EEXIST))
1290 return ret;
1291 return 1;
1292 }
1293
1294 static int write_disk_storage_file(const struct osl_table *t, unsigned col_num,
1295 const struct osl_object *obj, const char *ds_name)
1296 {
1297 int ret;
1298 char *filename;
1299
1300 ret = create_disk_storage_object_dir(t, col_num, ds_name);
1301 if (ret < 0)
1302 return ret;
1303 filename = disk_storage_path(t, col_num, ds_name);
1304 if (!filename)
1305 return -ERRNO_TO_ERROR(ENOMEM);
1306 ret = write_file(filename, obj->data, obj->size);
1307 free(filename);
1308 return ret;
1309 }
1310
1311 static int append_map_file(const struct osl_table *t, unsigned col_num,
1312 const struct osl_object *obj, uint32_t *new_size)
1313 {
1314 char *filename = column_filename(t, col_num);
1315 int ret;
1316 char header = 0; /* zero means valid object */
1317
1318 if (!filename)
1319 return -ERRNO_TO_ERROR(ENOMEM);
1320 ret = append_file(filename, &header, 1, obj->data, obj->size,
1321 new_size);
1322 free(filename);
1323 return ret;
1324 }
1325
1326 static int append_row_index(const struct osl_table *t, char *row_index)
1327 {
1328 char *filename;
1329 int ret;
1330
1331 if (!t->num_mapped_columns)
1332 return 1;
1333 filename = index_filename(t->desc);
1334 if (!filename)
1335 return -ERRNO_TO_ERROR(ENOMEM);
1336 ret = append_file(filename, NULL, 0, row_index,
1337 t->row_index_size, NULL);
1338 free(filename);
1339 return ret;
1340 }
1341
1342 /**
1343 * A wrapper for truncate(2)
1344 *
1345 * \param path Name of the regular file to truncate
1346 * \param size Number of bytes to \b shave \b off
1347 *
1348 * Truncate the regular file named by \a path by \a size bytes.
1349 *
1350 * \return Standard.
1351 *
1352 * \sa truncate(2)
1353 */
1354 int para_truncate(const char *path, off_t size)
1355 {
1356 int ret;
1357 struct stat statbuf;
1358
1359 ret = -E_OSL_STAT;
1360 if (stat(path, &statbuf) < 0)
1361 goto out;
1362 ret = -E_OSL_BAD_SIZE;
1363 if (statbuf.st_size < size)
1364 goto out;
1365 ret = -E_OSL_TRUNC;
1366 if (truncate(path, statbuf.st_size - size) < 0)
1367 goto out;
1368 ret = 1;
1369 out:
1370 return ret;
1371 }
1372
1373 static int truncate_mapped_file(const struct osl_table *t, unsigned col_num,
1374 off_t size)
1375 {
1376 int ret;
1377 char *filename = column_filename(t, col_num);
1378
1379 if (!filename)
1380 return -ERRNO_TO_ERROR(ENOMEM);
1381 ret = para_truncate(filename, size);
1382 free(filename);
1383 return ret;
1384 }
1385
1386 static int delete_disk_storage_file(const struct osl_table *t, unsigned col_num,
1387 const char *ds_name)
1388 {
1389 char *dirname, *filename = disk_storage_path(t, col_num, ds_name);
1390 int ret, err;
1391
1392 if (!filename)
1393 return -ERRNO_TO_ERROR(ENOMEM);
1394 ret = unlink(filename);
1395 err = errno;
1396 free(filename);
1397 if (ret < 0)
1398 return -ERRNO_TO_ERROR(err);
1399 if (!(t->desc->flags & OSL_LARGE_TABLE))
1400 return 1;
1401 dirname = disk_storage_dirname(t, col_num, ds_name);
1402 if (!dirname)
1403 return -ERRNO_TO_ERROR(ENOMEM);
1404 rmdir(dirname);
1405 free(dirname);
1406 return 1;
1407 }
1408
1409 /**
1410 * Add a new row to an osl table and retrieve this row.
1411 *
1412 * \param t Pointer to an open osl table.
1413 * \param objects Array of objects to be added.
1414 * \param row Result pointer.
1415 *
1416 * The \a objects parameter must point to an array containing one object per
1417 * column. The order of the objects in the array is given by the table
1418 * description of \a table. Several sanity checks are performed during object
1419 * insertion and the function returns without modifying the table if any of
1420 * these tests fail. In fact, it is atomic in the sense that it either
1421 * succeeds or leaves the table unchanged (i.e. either all or none of the
1422 * objects are added to the table).
1423 *
1424 * It is considered an error if an object is added to a column with associated
1425 * rbtree if this object is equal to an object already contained in that column
1426 * (i.e. the compare function for the column's rbtree returns zero).
1427 *
1428 * \return Standard.
1429 *
1430 * \sa struct osl_table_description, osl_compare_func, osl_add_row().
1431 */
1432 __export int osl_add_and_get_row(struct osl_table *t, struct osl_object *objects,
1433 struct osl_row **row)
1434 {
1435 int i, ret;
1436 char *ds_name = NULL;
1437 struct rb_node **rb_parents = NULL, ***rb_links = NULL;
1438 char *new_row_index = NULL;
1439 struct osl_object *volatile_objs = NULL;
1440 const struct osl_column_description *cd;
1441
1442 if (!t)
1443 return -E_OSL_BAD_TABLE;
1444 rb_parents = malloc(t->num_rbtrees * sizeof(struct rn_node*));
1445 if (!rb_parents)
1446 return -ERRNO_TO_ERROR(ENOMEM);
1447 rb_links = malloc(t->num_rbtrees * sizeof(struct rn_node**));
1448 if (!rb_links) {
1449 free(rb_parents);
1450 return -ERRNO_TO_ERROR(ENOMEM);
1451 }
1452 if (t->num_mapped_columns) {
1453 new_row_index = malloc(t->row_index_size);
1454 if (!new_row_index) {
1455 free(rb_links);
1456 free(rb_parents);
1457 return -ERRNO_TO_ERROR(ENOMEM);
1458 }
1459 }
1460 /* pass 1: sanity checks */
1461 // DEBUG_LOG("sanity tests: %p:%p\n", objects[0].data,
1462 // objects[1].data);
1463 FOR_EACH_COLUMN(i, t->desc, cd) {
1464 enum osl_storage_type st = cd->storage_type;
1465 enum osl_storage_flags sf = cd->storage_flags;
1466
1467 // ret = -E_OSL_NULL_OBJECT;
1468 // if (!objects[i])
1469 // goto out;
1470 if (st == OSL_DISK_STORAGE)
1471 continue;
1472 if (sf & OSL_RBTREE) {
1473 unsigned rbtree_num = t->columns[i].rbtree_num;
1474 ret = -E_OSL_RB_KEY_EXISTS;
1475 // DEBUG_LOG("checking whether %p exists\n",
1476 // objects[i].data);
1477 if (search_rbtree(objects + i, t, i,
1478 &rb_parents[rbtree_num],
1479 &rb_links[rbtree_num]) > 0)
1480 goto out;
1481 }
1482 if (sf & OSL_FIXED_SIZE) {
1483 // DEBUG_LOG("fixed size. need: %zu, have: %d\n",
1484 // objects[i].size, cd->data_size);
1485 ret = -E_OSL_BAD_DATA_SIZE;
1486 if (objects[i].size != cd->data_size)
1487 goto out;
1488 }
1489 }
1490 if (t->num_disk_storage_columns) {
1491 ds_name = disk_storage_name_of_object(t,
1492 &objects[t->disk_storage_name_column]);
1493 ret = -ERRNO_TO_ERROR(ENOMEM);
1494 if (!ds_name)
1495 goto out;
1496 }
1497 ret = unmap_table(t, OSL_MARK_CLEAN);
1498 if (ret < 0)
1499 goto out;
1500 // DEBUG_LOG("sanity tests passed%s\n", "");
1501 /* pass 2: create data files, append map data */
1502 FOR_EACH_COLUMN(i, t->desc, cd) {
1503 enum osl_storage_type st = cd->storage_type;
1504 if (st == OSL_NO_STORAGE)
1505 continue;
1506 if (st == OSL_MAPPED_STORAGE) {
1507 uint32_t new_size;
1508 struct osl_column *col = &t->columns[i];
1509 // DEBUG_LOG("appending object of size %zu\n",
1510 // objects[i].size);
1511 ret = append_map_file(t, i, objects + i, &new_size);
1512 if (ret < 0)
1513 goto rollback;
1514 update_cell_index(new_row_index, col, new_size,
1515 objects[i].size);
1516 continue;
1517 }
1518 /* DISK_STORAGE */
1519 ret = write_disk_storage_file(t, i, objects + i, ds_name);
1520 if (ret < 0)
1521 goto rollback;
1522 }
1523 ret = append_row_index(t, new_row_index);
1524 if (ret < 0)
1525 goto rollback;
1526 ret = map_table(t, MAP_TBL_FL_VERIFY_INDEX);
1527 if (ret < 0) { /* truncate index and rollback changes */
1528 char *filename = index_filename(t->desc);
1529 if (filename)
1530 para_truncate(filename, t->row_index_size);
1531 free(filename);
1532 goto rollback;
1533 }
1534 /* pass 3: add entry to rbtrees */
1535 if (t->num_volatile_columns) {
1536 ret = -ERRNO_TO_ERROR(ENOMEM);
1537 volatile_objs = calloc(t->num_volatile_columns,
1538 sizeof(struct osl_object));
1539 if (!volatile_objs)
1540 goto out;
1541 FOR_EACH_VOLATILE_COLUMN(i, t, cd)
1542 volatile_objs[t->columns[i].volatile_num] = objects[i];
1543 }
1544 t->num_rows++;
1545 // DEBUG_LOG("adding new entry as row #%d\n", t->num_rows - 1);
1546 ret = add_row_to_rbtrees(t, t->num_rows - 1, volatile_objs, row);
1547 if (ret < 0)
1548 goto out;
1549 // DEBUG_LOG("added new entry as row #%d\n", t->num_rows - 1);
1550 ret = 1;
1551 goto out;
1552 rollback: /* rollback all changes made, ignore further errors */
1553 for (i--; i >= 0; i--) {
1554 cd = get_column_description(t->desc, i);
1555 enum osl_storage_type st = cd->storage_type;
1556 if (st == OSL_NO_STORAGE)
1557 continue;
1558
1559 if (st == OSL_MAPPED_STORAGE)
1560 truncate_mapped_file(t, i, objects[i].size);
1561 else /* disk storage */
1562 delete_disk_storage_file(t, i, ds_name);
1563 }
1564 /* ignore error and return previous error value */
1565 map_table(t, MAP_TBL_FL_VERIFY_INDEX);
1566 out:
1567 free(new_row_index);
1568 free(ds_name);
1569 free(rb_parents);
1570 free(rb_links);
1571 return ret;
1572 }
1573
1574 /**
1575 * Add a new row to an osl table.
1576 *
1577 * \param t Same meaning as osl_add_and_get_row().
1578 * \param objects Same meaning as osl_add_and_get_row().
1579 *
1580 * \return The return value of the underlying call to osl_add_and_get_row().
1581 *
1582 * This is equivalent to osl_add_and_get_row(t, objects, NULL).
1583 */
1584 __export int osl_add_row(struct osl_table *t, struct osl_object *objects)
1585 {
1586 return osl_add_and_get_row(t, objects, NULL);
1587 }
1588
1589 /**
1590 * Retrieve an object identified by row and column
1591 *
1592 * \param t Pointer to an open osl table.
1593 * \param r Pointer to the row.
1594 * \param col_num The column number.
1595 * \param object The result pointer.
1596 *
1597 * The column determined by \a col_num must be of type \p OSL_MAPPED_STORAGE
1598 * or \p OSL_NO_STORAGE, i.e. no disk storage objects may be retrieved by this
1599 * function.
1600 *
1601 * \return Standard.
1602 *
1603 * \sa osl_storage_type, osl_open_disk_object().
1604 */
1605 __export int osl_get_object(const struct osl_table *t, const struct osl_row *r,
1606 unsigned col_num, struct osl_object *object)
1607 {
1608 const struct osl_column_description *cd;
1609
1610 if (!t)
1611 return -E_OSL_BAD_TABLE;
1612 cd = get_column_description(t->desc, col_num);
1613 /* col must not be disk storage */
1614 if (cd->storage_type == OSL_DISK_STORAGE)
1615 return -E_OSL_BAD_STORAGE_TYPE;
1616 if (cd->storage_type == OSL_MAPPED_STORAGE)
1617 return get_mapped_object(t, col_num, r->num, object);
1618 /* volatile */
1619 *object = r->volatile_objects[t->columns[col_num].volatile_num];
1620 return 1;
1621 }
1622
1623 static int mark_mapped_object_invalid(const struct osl_table *t,
1624 uint32_t row_num, unsigned col_num)
1625 {
1626 struct osl_object obj;
1627 char *p;
1628 int ret = get_mapped_object(t, col_num, row_num, &obj);
1629
1630 if (ret < 0)
1631 return ret;
1632 p = obj.data;
1633 p--;
1634 *p = 0xff;
1635 return 1;
1636 }
1637
1638 /**
1639 * Delete a row from an osl table.
1640 *
1641 * \param t Pointer to an open osl table.
1642 * \param row Pointer to the row to delete.
1643 *
1644 * This removes all disk storage objects, removes all rbtree nodes, and frees
1645 * all volatile objects belonging to the given row. For mapped columns, the
1646 * data is merely marked invalid and may be pruned from time to time by
1647 * para_fsck.
1648 *
1649 * \return Standard.
1650 */
1651 __export int osl_del_row(struct osl_table *t, struct osl_row *row)
1652 {
1653 struct osl_row *r = row;
1654 int i, ret;
1655 const struct osl_column_description *cd;
1656
1657 if (!t)
1658 return -E_OSL_BAD_TABLE;
1659 INFO_LOG("deleting row %p\n", row);
1660
1661 if (t->num_disk_storage_columns) {
1662 char *ds_name;
1663 ret = disk_storage_name_of_row(t, r, &ds_name);
1664 if (ret < 0)
1665 goto out;
1666 FOR_EACH_DISK_STORAGE_COLUMN(i, t, cd)
1667 delete_disk_storage_file(t, i, ds_name);
1668 free(ds_name);
1669 }
1670 FOR_EACH_COLUMN(i, t->desc, cd) {
1671 struct osl_column *col = t->columns + i;
1672 enum osl_storage_type st = cd->storage_type;
1673 remove_rb_node(t, i, r);
1674 if (st == OSL_MAPPED_STORAGE) {
1675 mark_mapped_object_invalid(t, r->num, i);
1676 continue;
1677 }
1678 if (st == OSL_NO_STORAGE && !(cd->storage_flags & OSL_DONT_FREE))
1679 free(r->volatile_objects[col->volatile_num].data);
1680 }
1681 if (t->num_mapped_columns) {
1682 ret = mark_row_invalid(t, r->num);
1683 if (ret < 0)
1684 goto out;
1685 t->num_invalid_rows++;
1686 } else
1687 t->num_rows--;
1688 ret = 1;
1689 out:
1690 free(r->volatile_objects);
1691 free(r);
1692 return ret;
1693 }
1694
1695 /* test if column has an rbtree */
1696 static int check_rbtree_col(const struct osl_table *t, unsigned col_num,
1697 struct osl_column **col)
1698 {
1699 if (!t)
1700 return -E_OSL_BAD_TABLE;
1701 if (!(get_column_description(t->desc, col_num)->storage_flags & OSL_RBTREE))
1702 return -E_OSL_BAD_STORAGE_FLAGS;
1703 *col = t->columns + col_num;
1704 return 1;
1705 }
1706
1707 /**
1708 * Get the row that contains the given object.
1709 *
1710 * \param t Pointer to an open osl table.
1711 * \param col_num The number of the column to be searched.
1712 * \param obj The object to be looked up.
1713 * \param result Points to the row containing \a obj.
1714 *
1715 * Lookup \a obj in \a t and return the row containing \a obj. The column
1716 * specified by \a col_num must have an associated rbtree.
1717 *
1718 * \return Standard.
1719 *
1720 * \sa osl_storage_flags
1721 */
1722 __export int osl_get_row(const struct osl_table *t, unsigned col_num,
1723 const struct osl_object *obj, struct osl_row **result)
1724 {
1725 int ret;
1726 struct rb_node *node;
1727 struct osl_row *row;
1728 struct osl_column *col;
1729
1730 *result = NULL;
1731 ret = check_rbtree_col(t, col_num, &col);
1732 if (ret < 0)
1733 return ret;
1734 ret = search_rbtree(obj, t, col_num, &node, NULL);
1735 if (ret < 0)
1736 return ret;
1737 row = get_row_pointer(node, t->columns[col_num].rbtree_num);
1738 *result = row;
1739 return 1;
1740 }
1741
1742 static int rbtree_loop(struct osl_column *col, void *private_data,
1743 osl_rbtree_loop_func *func)
1744 {
1745 struct rb_node *n, *tmp;
1746
1747 /* this for-loop is safe against removal of an entry */
1748 for (n = rb_first(&col->rbtree), tmp = n? rb_next(n) : NULL;
1749 n;
1750 n = tmp, tmp = tmp? rb_next(tmp) : NULL) {
1751 struct osl_row *r = get_row_pointer(n, col->rbtree_num);
1752 if (func(r, private_data) < 0)
1753 return -E_OSL_LOOP;
1754 }
1755 return 1;
1756 }
1757
1758 static int rbtree_loop_reverse(struct osl_column *col, void *private_data,
1759 osl_rbtree_loop_func *func)
1760 {
1761 struct rb_node *n, *tmp;
1762
1763 /* safe against removal of an entry */
1764 for (n = rb_last(&col->rbtree), tmp = n? rb_prev(n) : NULL;
1765 n;
1766 n = tmp, tmp = tmp? rb_prev(tmp) : NULL) {
1767 struct osl_row *r = get_row_pointer(n, col->rbtree_num);
1768 if (func(r, private_data) < 0)
1769 return -E_OSL_LOOP;
1770 }
1771 return 1;
1772 }
1773
1774 /**
1775 * Loop over all nodes in an rbtree.
1776 *
1777 * \param t Pointer to an open osl table.
1778 * \param col_num The column to use for iterating over the elements.
1779 * \param private_data Pointer that gets passed to \a func.
1780 * \param func The function to be called for each node in the rbtree.
1781 *
1782 * This function does an in-order walk of the rbtree associated with \a
1783 * col_num. It is an error if the \p OSL_RBTREE flag is not set for this
1784 * column. For each node in the rbtree, the given function \a func is called
1785 * with two pointers as arguments: The first osl_row* argument points to the
1786 * row that contains the object corresponding to the rbtree node currently
1787 * traversed, and the \a private_data pointer is passed verbatim to \a func as the
1788 * second argument. The loop terminates either if \a func returns a negative
1789 * value, or if all nodes of the tree have been visited.
1790 *
1791 *
1792 * \return Standard. If the termination of the loop was caused by \a func
1793 * returning a negative value, \p -E_OSL_LOOP is returned.
1794 *
1795 * \sa osl_storage_flags, osl_rbtree_loop_reverse(), osl_compare_func.
1796 */
1797 __export int osl_rbtree_loop(const struct osl_table *t, unsigned col_num,
1798 void *private_data, osl_rbtree_loop_func *func)
1799 {
1800 struct osl_column *col;
1801
1802 int ret = check_rbtree_col(t, col_num, &col);
1803 if (ret < 0)
1804 return ret;
1805 return rbtree_loop(col, private_data, func);
1806 }
1807
1808 /**
1809 * Loop over all nodes in an rbtree in reverse order.
1810 *
1811 * \param t Identical meaning as in \p osl_rbtree_loop().
1812 * \param col_num Identical meaning as in \p osl_rbtree_loop().
1813 * \param private_data Identical meaning as in \p osl_rbtree_loop().
1814 * \param func Identical meaning as in \p osl_rbtree_loop().
1815 *
1816 * This function is identical to \p osl_rbtree_loop(), the only difference
1817 * is that the tree is walked in reverse order.
1818 *
1819 * \return The same return value as \p osl_rbtree_loop().
1820 *
1821 * \sa osl_rbtree_loop().
1822 */
1823 __export int osl_rbtree_loop_reverse(const struct osl_table *t, unsigned col_num,
1824 void *private_data, osl_rbtree_loop_func *func)
1825 {
1826 struct osl_column *col;
1827
1828 int ret = check_rbtree_col(t, col_num, &col);
1829 if (ret < 0)
1830 return ret;
1831 return rbtree_loop_reverse(col, private_data, func);
1832 }
1833
1834 /* TODO: Rollback changes on errors */
1835 static int rename_disk_storage_objects(struct osl_table *t,
1836 struct osl_object *old_obj, struct osl_object *new_obj)
1837 {
1838 int i, ret;
1839 const struct osl_column_description *cd;
1840 char *old_ds_name, *new_ds_name;
1841
1842 if (!t->num_disk_storage_columns)
1843 return 1; /* nothing to do */
1844 if (old_obj->size == new_obj->size && !memcmp(new_obj->data,
1845 old_obj->data, new_obj->size))
1846 return 1; /* object did not change */
1847 old_ds_name = disk_storage_name_of_object(t, old_obj);
1848 new_ds_name = disk_storage_name_of_object(t, new_obj);
1849 ret = -ERRNO_TO_ERROR(ENOMEM);
1850 if (!old_ds_name || ! new_ds_name)
1851 goto out;
1852
1853 FOR_EACH_DISK_STORAGE_COLUMN(i, t, cd) {
1854 char *old_filename, *new_filename;
1855 ret = create_disk_storage_object_dir(t, i, new_ds_name);
1856 if (ret < 0)
1857 goto out;
1858 old_filename = disk_storage_path(t, i, old_ds_name);
1859 new_filename = disk_storage_path(t, i, new_ds_name);
1860 if (!old_filename || !new_filename)
1861 ret = -ERRNO_TO_ERROR(ENOMEM);
1862 else
1863 ret = para_rename(old_filename, new_filename);
1864 free(old_filename);
1865 free(new_filename);
1866 if (ret < 0)
1867 goto out;
1868 }
1869 ret = 1;
1870 out:
1871 free(old_ds_name);
1872 free(new_ds_name);
1873 return ret;
1874
1875 }
1876
1877 /**
1878 * Change an object in an osl table.
1879 *
1880 * \param t Pointer to an open osl table.
1881 * \param r Pointer to the row containing the object to be updated.
1882 * \param col_num Number of the column containing the object to be updated.
1883 * \param obj Pointer to the replacement object.
1884 *
1885 * This function gets rid of all references to the old object. This includes
1886 * removal of the rbtree node in case there is an rbtree associated with \a
1887 * col_num. It then inserts \a obj into the table and the rbtree if necessary.
1888 *
1889 * If the \p OSL_RBTREE flag is set for \a col_num, you \b MUST call this
1890 * function in order to change the contents of an object, even for volatile or
1891 * mapped columns of constant size (which may be updated directly if \p
1892 * OSL_RBTREE is not set). Otherwise the rbtree might become corrupted.
1893 *
1894 * \return Standard
1895 */
1896 __export int osl_update_object(struct osl_table *t, const struct osl_row *r,
1897 unsigned col_num, struct osl_object *obj)
1898 {
1899 struct osl_column *col;
1900 const struct osl_column_description *cd;
1901 int ret;
1902
1903 if (!t)
1904 return -E_OSL_BAD_TABLE;
1905 col = &t->columns[col_num];
1906 cd = get_column_description(t->desc, col_num);
1907 DEBUG_LOG("updating column %u of %s\n", col_num, t->desc->name);
1908 if (cd->storage_flags & OSL_RBTREE) {
1909 if (search_rbtree(obj, t, col_num, NULL, NULL) > 0)
1910 return -E_OSL_RB_KEY_EXISTS;
1911 }
1912 if (cd->storage_flags & OSL_FIXED_SIZE) {
1913 if (obj->size != cd->data_size)
1914 return -E_OSL_BAD_DATA_SIZE;
1915 }
1916 remove_rb_node(t, col_num, r);
1917 if (cd->storage_type == OSL_NO_STORAGE) { /* TODO: If fixed size, reuse object? */
1918 free(r->volatile_objects[col->volatile_num].data);
1919 r->volatile_objects[col->volatile_num] = *obj;
1920 } else if (cd->storage_type == OSL_DISK_STORAGE) {
1921 char *ds_name;
1922 ret = disk_storage_name_of_row(t, r, &ds_name);
1923 if (ret < 0)
1924 return ret;
1925 ret = delete_disk_storage_file(t, col_num, ds_name);
1926 if (ret < 0 && !is_errno(-ret, ENOENT)) {
1927 free(ds_name);
1928 return ret;
1929 }
1930 ret = write_disk_storage_file(t, col_num, obj, ds_name);
1931 free(ds_name);
1932 if (ret < 0)
1933 return ret;
1934 } else { /* mapped storage */
1935 struct osl_object old_obj;
1936 ret = get_mapped_object(t, col_num, r->num, &old_obj);
1937 if (ret < 0)
1938 return ret;
1939 /*
1940 * If the updated column is the disk storage name column, the
1941 * disk storage name changes, so we have to rename all disk
1942 * storage objects accordingly.
1943 */
1944 if (col_num == t->disk_storage_name_column) {
1945 ret = rename_disk_storage_objects(t, &old_obj, obj);
1946 if (ret < 0)
1947 return ret;
1948 }
1949 if (cd->storage_flags & OSL_FIXED_SIZE)
1950 memcpy(old_obj.data, obj->data, cd->data_size);
1951 else { /* TODO: if the size doesn't change, use old space */
1952 uint32_t new_data_map_size;
1953 char *row_index;
1954 ret = get_row_index(t, r->num, &row_index);
1955 if (ret < 0)
1956 return ret;
1957 ret = mark_mapped_object_invalid(t, r->num, col_num);
1958 if (ret < 0)
1959 return ret;
1960 unmap_column(t, col_num);
1961 ret = append_map_file(t, col_num, obj,
1962 &new_data_map_size);
1963 if (ret < 0)
1964 return ret;
1965 ret = map_column(t, col_num);
1966 if (ret < 0)
1967 return ret;
1968 update_cell_index(row_index, col, new_data_map_size,
1969 obj->size);
1970 }
1971 }
1972 if (cd->storage_flags & OSL_RBTREE) {
1973 ret = insert_rbtree(t, col_num, r, obj);
1974 if (ret < 0)
1975 return ret;
1976 }
1977 return 1;
1978 }
1979
1980 /**
1981 * Retrieve an object of type \p OSL_DISK_STORAGE by row and column.
1982 *
1983 * \param t Pointer to an open osl table.
1984 * \param r Pointer to the row containing the object.
1985 * \param col_num The column number.
1986 * \param obj Points to the result upon successful return.
1987 *
1988 * For columns of type \p OSL_DISK_STORAGE, this function must be used to
1989 * retrieve one of its containing objects. Afterwards, osl_close_disk_object()
1990 * must be called in order to deallocate the resources.
1991 *
1992 * \return Standard.
1993 *
1994 * \sa osl_get_object(), osl_storage_type, osl_close_disk_object().
1995 */
1996 __export int osl_open_disk_object(const struct osl_table *t, const struct osl_row *r,
1997 unsigned col_num, struct osl_object *obj)
1998 {
1999 const struct osl_column_description *cd;
2000 char *ds_name, *filename;
2001 int ret;
2002
2003 if (!t)
2004 return -E_OSL_BAD_TABLE;
2005 cd = get_column_description(t->desc, col_num);
2006 if (cd->storage_type != OSL_DISK_STORAGE)
2007 return -E_OSL_BAD_STORAGE_TYPE;
2008
2009 ret = disk_storage_name_of_row(t, r, &ds_name);
2010 if (ret < 0)
2011 return ret;
2012 filename = disk_storage_path(t, col_num, ds_name);
2013 free(ds_name);
2014 if (!filename)
2015 return -ERRNO_TO_ERROR(ENOMEM);
2016 DEBUG_LOG("filename: %s\n", filename);
2017 ret = mmap_full_file(filename, O_RDONLY, &obj->data, &obj->size, NULL);
2018 free(filename);
2019 return ret;
2020 }
2021
2022 /**
2023 * Free resources that were allocated during osl_open_disk_object().
2024 *
2025 * \param obj Pointer to the object previously returned by open_disk_object().
2026 *
2027 * \return The return value of the underlying call to para_munmap().
2028 *
2029 * \sa para_munmap().
2030 */
2031 __export int osl_close_disk_object(struct osl_object *obj)
2032 {
2033 return para_munmap(obj->data, obj->size);
2034 }
2035
2036 /**
2037 * Get the number of rows of the given table.
2038 *
2039 * \param t Pointer to an open osl table.
2040 * \param num_rows Result is returned here.
2041 *
2042 * The number of rows returned via \a num_rows excluding any invalid rows.
2043 *
2044 * \return Positive on success, \p -E_OSL_BAD_TABLE if \a t is \p NULL.
2045 */
2046 __export int osl_get_num_rows(const struct osl_table *t, unsigned *num_rows)
2047 {
2048 if (!t)
2049 return -E_OSL_BAD_TABLE;
2050 assert(t->num_rows >= t->num_invalid_rows);
2051 *num_rows = t->num_rows - t->num_invalid_rows;
2052 return 1;
2053 }
2054
2055 /**
2056 * Get the rank of a row.
2057 *
2058 * \param t An open osl table.
2059 * \param r The row to get the rank of.
2060 * \param col_num The number of an rbtree column.
2061 * \param rank Result pointer.
2062 *
2063 * The rank is, by definition, the position of the row in the linear order
2064 * determined by an in-order tree walk of the rbtree associated with column
2065 * number \a col_num of \a table.
2066 *
2067 * \return Standard.
2068 *
2069 * \sa osl_get_nth_row().
2070 */
2071 __export int osl_get_rank(const struct osl_table *t, struct osl_row *r,
2072 unsigned col_num, unsigned *rank)
2073 {
2074 struct osl_object obj;
2075 struct osl_column *col;
2076 struct rb_node *node;
2077 int ret = check_rbtree_col(t, col_num, &col);
2078
2079 if (ret < 0)
2080 return ret;
2081 ret = osl_get_object(t, r, col_num, &obj);
2082 if (ret < 0)
2083 return ret;
2084 ret = search_rbtree(&obj, t, col_num, &node, NULL);
2085 if (ret < 0)
2086 return ret;
2087 ret = rb_rank(node, rank);
2088 if (ret < 0)
2089 return -E_OSL_BAD_ROW;
2090 return 1;
2091 }
2092
2093 /**
2094 * Get the row with n-th greatest value.
2095 *
2096 * \param t Pointer to an open osl table.
2097 * \param col_num The column number.
2098 * \param n The rank of the desired row.
2099 * \param result Row is returned here.
2100 *
2101 * Retrieve the n-th order statistic with respect to the compare function
2102 * of the rbtree column \a col_num. In other words, get that row with
2103 * \a n th greatest value in column \a col_num. It's an error if
2104 * \a col_num is not a rbtree column, or if \a n is larger than the
2105 * number of rows in the table.
2106 *
2107 * \return Standard.
2108 *
2109 * \sa osl_storage_flags, osl_compare_func, osl_get_row(),
2110 * osl_rbtree_last_row(), osl_rbtree_first_row(), osl_get_rank().
2111 */
2112 __export int osl_get_nth_row(const struct osl_table *t, unsigned col_num,
2113 unsigned n, struct osl_row **result)
2114 {
2115 struct osl_column *col;
2116 struct rb_node *node;
2117 unsigned num_rows;
2118 int ret;
2119
2120 if (n == 0)
2121 return -E_OSL_RB_KEY_NOT_FOUND;
2122 ret = osl_get_num_rows(t, &num_rows);
2123 if (ret < 0)
2124 return ret;
2125 if (n > num_rows)
2126 return -E_OSL_RB_KEY_NOT_FOUND;
2127 ret = check_rbtree_col(t, col_num, &col);
2128 if (ret < 0)
2129 return ret;
2130 node = rb_nth(col->rbtree.rb_node, n);
2131 if (!node)
2132 return -E_OSL_RB_KEY_NOT_FOUND;
2133 *result = get_row_pointer(node, col->rbtree_num);
2134 return 1;
2135 }
2136
2137 /**
2138 * Get the row corresponding to the smallest rbtree node of a column.
2139 *
2140 * \param t An open rbtree table.
2141 * \param col_num The number of the rbtree column.
2142 * \param result A pointer to the first row is returned here.
2143 *
2144 * The rbtree node of the smallest object (with respect to the corresponding
2145 * compare function) is selected and the row containing this object is
2146 * returned. It is an error if \a col_num refers to a column without an
2147 * associated rbtree.
2148 *
2149 * \return Standard.
2150 *
2151 * \sa osl_get_nth_row(), osl_rbtree_last_row().
2152 */
2153 __export int osl_rbtree_first_row(const struct osl_table *t, unsigned col_num,
2154 struct osl_row **result)
2155 {
2156 return osl_get_nth_row(t, col_num, 1, result);
2157 }
2158
2159 /**
2160 * Get the row corresponding to the greatest rbtree node of a column.
2161 *
2162 * \param t The same meaning as in \p osl_rbtree_first_row().
2163 * \param col_num The same meaning as in \p osl_rbtree_first_row().
2164 * \param result The same meaning as in \p osl_rbtree_first_row().
2165 *
2166 * This function works just like osl_rbtree_first_row(), the only difference
2167 * is that the row containing the greatest rather than the smallest object is
2168 * returned.
2169 *
2170 * \return Standard.
2171 *
2172 * \sa osl_get_nth_row(), osl_rbtree_first_row().
2173 */
2174 __export int osl_rbtree_last_row(const struct osl_table *t, unsigned col_num,
2175 struct osl_row **result)
2176 {
2177 unsigned num_rows;
2178 int ret = osl_get_num_rows(t, &num_rows);
2179
2180 if (ret < 0)
2181 return ret;
2182 return osl_get_nth_row(t, col_num, num_rows, result);
2183 }