cc0e641f1c53af37541c97090dc769723658b79b
[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 static int loglevel;
94
95 static void __attribute ((constructor)) init_loglevel(void)
96 {
97 char *p = getenv("OSL_LOGLEVEL");
98
99 /* don't log anything if unset */
100 loglevel = p? atoi(p) : EMERG + 1;
101 }
102
103 /**
104 * The log function.
105 *
106 * \param ll Loglevel.
107 * \param fmt Usual format string.
108 *
109 * All XXX_LOG() macros use this function.
110 */
111 __printf_2_3 void __log(int ll, const char* fmt,...)
112 {
113 va_list argp;
114 FILE *outfd;
115 struct tm *tm;
116 time_t t1;
117 char str[255] = "";
118
119 if (ll < loglevel)
120 return;
121 outfd = stderr;
122 time(&t1);
123 tm = localtime(&t1);
124 strftime(str, sizeof(str), "%b %d %H:%M:%S", tm);
125 fprintf(outfd, "%s ", str);
126 va_start(argp, fmt);
127 vfprintf(outfd, fmt, argp);
128 va_end(argp);
129 }
130
131 /**
132 * A wrapper for lseek(2).
133 *
134 * \param fd The file descriptor whose offset is to be to repositioned.
135 * \param offset A value-result parameter.
136 * \param whence Usual repositioning directive.
137 *
138 * Reposition the offset of the file descriptor \a fd to the argument \a offset
139 * according to the directive \a whence. Upon successful return, \a offset
140 * contains the resulting offset location as measured in bytes from the
141 * beginning of the file.
142 *
143 * \return Positive on success. Otherwise, the function returns \p -E_OSL_LSEEK.
144 *
145 * \sa lseek(2).
146 */
147 static int __lseek(int fd, off_t *offset, int whence)
148 {
149 *offset = lseek(fd, *offset, whence);
150 int ret = -E_OSL_LSEEK;
151 if (*offset == -1)
152 return ret;
153 return 1;
154 }
155
156 /**
157 * Wrapper for the write system call.
158 *
159 * \param fd The file descriptor to write to.
160 * \param buf The buffer to write.
161 * \param size The length of \a buf in bytes.
162 *
163 * This function writes out the given buffer and retries if an interrupt
164 * occurred during the write.
165 *
166 * \return On success, the number of bytes written is returned, otherwise, the
167 * function returns \p -E_OSL_WRITE.
168 *
169 * \sa write(2).
170 */
171 static ssize_t __write(int fd, const void *buf, size_t size)
172 {
173 ssize_t ret;
174
175 for (;;) {
176 ret = write(fd, buf, size);
177 if ((ret < 0) && (errno == EAGAIN || errno == EINTR))
178 continue;
179 return ret >= 0? ret : -E_OSL_WRITE;
180 }
181 }
182
183 /**
184 * Write the whole buffer to a file descriptor.
185 *
186 * \param fd The file descriptor to write to.
187 * \param buf The buffer to write.
188 * \param size The length of \a buf in bytes.
189 *
190 * This function writes the given buffer and continues on short writes and
191 * when interrupted by a signal.
192 *
193 * \return Positive on success, negative on errors. Possible errors: any
194 * errors returned by para_write().
195 *
196 * \sa para_write().
197 */
198 static ssize_t write_all(int fd, const void *buf, size_t size)
199 {
200 // DEBUG_LOG("writing %zu bytes\n", size);
201 const char *b = buf;
202 while (size) {
203 ssize_t ret = __write(fd, b, size);
204 // DEBUG_LOG("ret: %zd\n", ret);
205 if (ret < 0)
206 return ret;
207 b += ret;
208 size -= ret;
209 }
210 return 1;
211 }
212 /**
213 * Open a file, write the given buffer and close the file.
214 *
215 * \param filename Full path to the file to open.
216 * \param buf The buffer to write to the file.
217 * \param size The size of \a buf.
218 *
219 * \return Standard.
220 */
221 static int write_file(const char *filename, const void *buf, size_t size)
222 {
223 int ret, fd;
224
225 ret = osl_open(filename, O_WRONLY | O_CREAT | O_EXCL, 0644);
226 if (ret < 0)
227 return ret;
228 fd = ret;
229 ret = write_all(fd, buf, size);
230 if (ret < 0)
231 goto out;
232 ret = 1;
233 out:
234 close(fd);
235 return ret;
236 }
237
238 static int append_file(const char *filename, char *header, size_t header_size,
239 char *data, size_t data_size, uint32_t *new_pos)
240 {
241 int ret, fd;
242
243 // DEBUG_LOG("appending %zu + %zu bytes\n", header_size, data_size);
244 ret = osl_open(filename, O_WRONLY | O_CREAT | O_APPEND, 0644);
245 if (ret < 0)
246 return ret;
247 fd = ret;
248 if (header && header_size) {
249 ret = write_all(fd, header, header_size);
250 if (ret < 0)
251 goto out;
252 }
253 ret = write_all(fd, data, data_size);
254 if (ret < 0)
255 goto out;
256 if (new_pos) {
257 off_t offset = 0;
258 ret = __lseek(fd, &offset, SEEK_END);
259 if (ret < 0)
260 goto out;
261 // DEBUG_LOG("new file size: " FMT_OFF_T "\n", offset);
262 *new_pos = offset;
263 }
264 ret = 1;
265 out:
266 close(fd);
267 return ret;
268 }
269
270 static int verify_name(const char *name)
271 {
272 if (!name)
273 return -E_OSL_BAD_NAME;
274 if (!*name)
275 return -E_OSL_BAD_NAME;
276 if (strchr(name, '/'))
277 return -E_OSL_BAD_NAME;
278 if (!strcmp(name, ".."))
279 return -E_OSL_BAD_NAME;
280 if (!strcmp(name, "."))
281 return -E_OSL_BAD_NAME;
282 return 1;
283 }
284
285 /**
286 * Compare two osl objects pointing to hash values.
287 *
288 * \param obj1 Pointer to the first hash object.
289 * \param obj2 Pointer to the second hash object.
290 *
291 * \return The values required for an osl compare function.
292 *
293 * \sa osl_compare_func, uint32_compare().
294 */
295 int osl_hash_compare(const struct osl_object *obj1, const struct osl_object *obj2)
296 {
297 return hash_compare((HASH_TYPE *)obj1->data, (HASH_TYPE *)obj2->data);
298 }
299
300 static char *disk_storage_dirname(const struct osl_table *t, unsigned col_num,
301 const char *ds_name)
302 {
303 char *dirname, *column_name = column_filename(t, col_num);
304
305 if (!column_name)
306 return NULL;
307 if (!(t->desc->flags & OSL_LARGE_TABLE))
308 return column_name;
309 dirname = make_message("%s/%.2s", column_name, ds_name);
310 free(column_name);
311 return dirname;
312 }
313
314 static char *disk_storage_name_of_object(const struct osl_table *t,
315 const struct osl_object *obj)
316 {
317 HASH_TYPE hash[HASH_SIZE];
318 hash_object(obj, hash);
319 return disk_storage_name_of_hash(t, hash);
320 }
321
322 static int disk_storage_name_of_row(const struct osl_table *t,
323 const struct osl_row *row, char **name)
324 {
325 struct osl_object obj;
326 int ret = osl_get_object(t, row, t->disk_storage_name_column, &obj);
327
328 if (ret < 0)
329 return ret;
330 *name = disk_storage_name_of_object(t, &obj);
331 if (*name)
332 return 1;
333 return -ERRNO_TO_ERROR(ENOMEM);
334 }
335
336 static void column_name_hash(const char *col_name, HASH_TYPE *hash)
337 {
338 hash_function(col_name, strlen(col_name), hash);
339 }
340
341 static int init_column_descriptions(struct osl_table *t)
342 {
343 int i, j, ret;
344 const struct osl_column_description *cd;
345
346 ret = -E_OSL_BAD_TABLE_DESC;
347 ret = verify_name(t->desc->name);
348 if (ret < 0)
349 goto err;
350 ret = -E_OSL_BAD_DB_DIR;
351 if (!t->desc->dir && (t->num_disk_storage_columns || t->num_mapped_columns))
352 goto err;
353 /* the size of the index header without column descriptions */
354 t->index_header_size = IDX_COLUMN_DESCRIPTIONS;
355 FOR_EACH_COLUMN(i, t->desc, cd) {
356 struct osl_column *col = t->columns + i;
357 if (cd->storage_flags & OSL_RBTREE) {
358 if (!cd->compare_function)
359 return -E_OSL_NO_COMPARE_FUNC;
360 }
361 if (cd->storage_type == OSL_NO_STORAGE)
362 continue;
363 ret = -E_OSL_NO_COLUMN_NAME;
364 if (!cd->name || !cd->name[0])
365 goto err;
366 ret = verify_name(cd->name);
367 if (ret < 0)
368 goto err;
369 t->index_header_size += index_column_description_size(cd->name);
370 column_name_hash(cd->name, col->name_hash);
371 ret = -E_OSL_DUPLICATE_COL_NAME;
372 for (j = i + 1; j < t->desc->num_columns; j++) {
373 const char *name2 = get_column_description(t->desc,
374 j)->name;
375 if (cd->name && name2 && !strcmp(cd->name, name2))
376 goto err;
377 }
378 }
379 return 1;
380 err:
381 return ret;
382 }
383
384 /**
385 * Initialize a struct table from given table description.
386 *
387 * \param desc The description of the osl table.
388 * \param table_ptr Result is returned here.
389 *
390 * This function performs several sanity checks on \p desc and returns if any
391 * of these tests fail. On success, a struct \p osl_table is allocated and
392 * initialized with data derived from \p desc.
393 *
394 * \return Standard.
395 *
396 * \sa struct osl_table.
397 */
398 int init_table_structure(const struct osl_table_description *desc,
399 struct osl_table **table_ptr)
400 {
401 const struct osl_column_description *cd;
402 struct osl_table *t = calloc(1, sizeof(*t));
403 int i, ret = -ERRNO_TO_ERROR(ENOMEM), have_disk_storage_name_column = 0;
404
405 if (!t)
406 return ret;
407 ret = -E_OSL_BAD_TABLE_DESC;
408 if (!desc)
409 goto err;
410 DEBUG_LOG("creating table structure for '%s' from table "
411 "description\n", desc->name);
412 ret = -E_OSL_NO_COLUMN_DESC;
413 if (!desc->column_descriptions)
414 goto err;
415 ret = -E_OSL_NO_COLUMNS;
416 if (!desc->num_columns)
417 goto err;
418 ret = -ERRNO_TO_ERROR(ENOMEM);
419 t->columns = calloc(desc->num_columns, sizeof(struct osl_column));
420 if (!t->columns)
421 goto err;
422 t->desc = desc;
423 FOR_EACH_COLUMN(i, t->desc, cd) {
424 enum osl_storage_type st = cd->storage_type;
425 enum osl_storage_flags sf = cd->storage_flags;
426 struct osl_column *col = &t->columns[i];
427
428 ret = -E_OSL_BAD_STORAGE_TYPE;
429 if (st != OSL_MAPPED_STORAGE && st != OSL_DISK_STORAGE
430 && st != OSL_NO_STORAGE)
431 goto err;
432 ret = -E_OSL_BAD_STORAGE_FLAGS;
433 if (st == OSL_DISK_STORAGE && sf & OSL_RBTREE)
434 goto err;
435 ret = -E_OSL_BAD_STORAGE_SIZE;
436 if (sf & OSL_FIXED_SIZE && !cd->data_size)
437 goto err;
438 switch (st) {
439 case OSL_DISK_STORAGE:
440 t->num_disk_storage_columns++;
441 break;
442 case OSL_MAPPED_STORAGE:
443 t->num_mapped_columns++;
444 col->index_offset = t->row_index_size;
445 t->row_index_size += 8;
446 break;
447 case OSL_NO_STORAGE:
448 col->volatile_num = t->num_volatile_columns;
449 t->num_volatile_columns++;
450 break;
451 }
452 if (sf & OSL_RBTREE) {
453 col->rbtree_num = t->num_rbtrees;
454 t->num_rbtrees++;
455 if ((sf & OSL_UNIQUE) && (st == OSL_MAPPED_STORAGE)) {
456 if (!have_disk_storage_name_column)
457 t->disk_storage_name_column = i;
458 have_disk_storage_name_column = 1;
459 }
460 }
461 }
462 ret = -E_OSL_NO_UNIQUE_RBTREE_COLUMN;
463 if (t->num_disk_storage_columns && !have_disk_storage_name_column)
464 goto err;
465 ret = -E_OSL_NO_RBTREE_COL;
466 if (!t->num_rbtrees)
467 goto err;
468 /* success */
469 DEBUG_LOG("OK. Index entry size: %u\n", t->row_index_size);
470 ret = init_column_descriptions(t);
471 if (ret < 0)
472 goto err;
473 *table_ptr = t;
474 return 1;
475 err:
476 free(t->columns);
477 free(t);
478 return ret;
479 }
480
481 /**
482 * Read the table description from index header.
483 *
484 * \param map The memory mapping of the index file.
485 * \param desc The values found in the index header are returned here.
486 *
487 * Read the index header, check for the paraslash magic string and the table version number.
488 * Read all information stored in the index header into \a desc.
489 *
490 * \return Standard.
491 *
492 * \sa struct osl_table_description, osl_create_table.
493 */
494 int read_table_desc(struct osl_object *map, struct osl_table_description *desc)
495 {
496 char *buf = map->data;
497 uint8_t version;
498 uint16_t header_size;
499 int ret, i;
500 unsigned offset;
501 struct osl_column_description *cd;
502
503 if (map->size < MIN_INDEX_HEADER_SIZE(1))
504 return -E_OSL_SHORT_TABLE;
505 if (strncmp(buf + IDX_PARA_MAGIC, PARA_MAGIC, strlen(PARA_MAGIC)))
506 return -E_OSL_NO_MAGIC;
507 version = read_u8(buf + IDX_VERSION);
508 if (version < MIN_TABLE_VERSION || version > MAX_TABLE_VERSION)
509 return -E_OSL_VERSION_MISMATCH;
510 desc->flags = read_u8(buf + IDX_TABLE_FLAGS);
511 desc->num_columns = read_u16(buf + IDX_NUM_COLUMNS);
512 DEBUG_LOG("%u columns\n", desc->num_columns);
513 if (!desc->num_columns)
514 return -E_OSL_NO_COLUMNS;
515 header_size = read_u16(buf + IDX_HEADER_SIZE);
516 if (map->size < header_size)
517 return -E_OSL_BAD_SIZE;
518 desc->column_descriptions = calloc(desc->num_columns,
519 sizeof(struct osl_column_description));
520 if (!desc->column_descriptions)
521 return -ERRNO_TO_ERROR(ENOMEM);
522 offset = IDX_COLUMN_DESCRIPTIONS;
523 FOR_EACH_COLUMN(i, desc, cd) {
524 char *null_byte;
525
526 ret = -E_OSL_SHORT_TABLE;
527 if (map->size < offset + MIN_IDX_COLUMN_DESCRIPTION_SIZE) {
528 ERROR_LOG("map size = %zu < %u = offset + min desc size\n",
529 map->size, offset + MIN_IDX_COLUMN_DESCRIPTION_SIZE);
530 goto err;
531 }
532 cd->storage_type = read_u16(buf + offset + IDX_CD_STORAGE_TYPE);
533 cd->storage_flags = read_u16(buf + offset +
534 IDX_CD_STORAGE_FLAGS);
535 cd->data_size = read_u32(buf + offset + IDX_CD_DATA_SIZE);
536 null_byte = memchr(buf + offset + IDX_CD_NAME, '\0',
537 map->size - offset - IDX_CD_NAME);
538 ret = -E_OSL_INDEX_CORRUPTION;
539 if (!null_byte)
540 goto err;
541 ret = -ERRNO_TO_ERROR(ENOMEM);
542 cd->name = strdup(buf + offset + IDX_CD_NAME);
543 if (!cd->name)
544 goto err;
545 offset += index_column_description_size(cd->name);
546 }
547 if (offset != header_size) {
548 ret = -E_OSL_INDEX_CORRUPTION;
549 ERROR_LOG("real header size = %u != %u = stored header size\n",
550 offset, header_size);
551 goto err;
552 }
553 return 1;
554 err:
555 FOR_EACH_COLUMN(i, desc, cd)
556 free(cd->name);
557 return ret;
558 }
559
560 /*
561 * check whether the table description given by \p t->desc matches the on-disk
562 * table structure stored in the index of \a t.
563 */
564 static int compare_table_descriptions(struct osl_table *t)
565 {
566 int i, ret;
567 struct osl_table_description desc;
568 const struct osl_column_description *cd1, *cd2;
569
570 /* read the on-disk structure into desc */
571 ret = read_table_desc(&t->index_map, &desc);
572 if (ret < 0)
573 return ret;
574 ret = -E_OSL_BAD_TABLE_FLAGS;
575 if (desc.flags != t->desc->flags)
576 goto out;
577 ret = -E_OSL_BAD_COLUMN_NUM;
578 if (desc.num_columns > t->desc->num_columns)
579 goto out;
580 if (desc.num_columns < t->desc->num_columns) {
581 struct osl_column_description *cd;
582 unsigned diff = t->desc->num_columns - desc.num_columns;
583 INFO_LOG("extending table by %u volatile columns\n", diff);
584 ret = -ERRNO_TO_ERROR(ENOMEM);
585 desc.column_descriptions = realloc(desc.column_descriptions,
586 t->desc->num_columns * sizeof(struct osl_column_description));
587 if (!desc.column_descriptions)
588 goto out;
589 for (i = desc.num_columns; i < t->desc->num_columns; i++) {
590 cd = get_column_description(&desc, i);
591 cd->storage_type = OSL_NO_STORAGE;
592 cd->name = NULL;
593 }
594 desc.num_columns += diff;
595 }
596 FOR_EACH_COLUMN(i, t->desc, cd1) {
597 cd2 = get_column_description(&desc, i);
598 ret = -E_OSL_BAD_STORAGE_TYPE;
599 if (cd1->storage_type != cd2->storage_type)
600 goto out;
601 if (cd1->storage_type == OSL_NO_STORAGE)
602 continue;
603 ret = -E_OSL_BAD_STORAGE_FLAGS;
604 if (cd1->storage_flags != cd2->storage_flags) {
605 ERROR_LOG("sf1 = %u != %u = sf2\n",
606 cd1->storage_flags, cd2->storage_flags);
607 goto out;
608 }
609 ret = -E_OSL_BAD_DATA_SIZE;
610 if (cd1->storage_flags & OSL_FIXED_SIZE)
611 if (cd1->data_size != cd2->data_size)
612 goto out;
613 ret = -E_OSL_BAD_COLUMN_NAME;
614 if (strcmp(cd1->name, cd2->name))
615 goto out;
616 }
617 DEBUG_LOG("table description of '%s' matches on-disk data, good\n",
618 t->desc->name);
619 ret = 1;
620 out:
621 FOR_EACH_COLUMN(i, &desc, cd1)
622 free(cd1->name);
623 free(desc.column_descriptions);
624 return ret;
625 }
626
627 static int create_table_index(struct osl_table *t)
628 {
629 char *buf, *filename;
630 int i, ret;
631 size_t size = t->index_header_size;
632 const struct osl_column_description *cd;
633 unsigned offset;
634
635 INFO_LOG("creating %zu byte index for table %s\n", size,
636 t->desc->name);
637 buf = calloc(1, size);
638 if (!buf)
639 return -ERRNO_TO_ERROR(ENOMEM);
640 sprintf(buf + IDX_PARA_MAGIC, "%s", PARA_MAGIC);
641 write_u8(buf + IDX_TABLE_FLAGS, t->desc->flags);
642 write_u8(buf + IDX_DIRTY_FLAG, 0);
643 write_u8(buf + IDX_VERSION, CURRENT_TABLE_VERSION);
644 write_u16(buf + IDX_NUM_COLUMNS, t->num_mapped_columns + t->num_disk_storage_columns);
645 write_u16(buf + IDX_HEADER_SIZE, t->index_header_size);
646 offset = IDX_COLUMN_DESCRIPTIONS;
647 FOR_EACH_COLUMN(i, t->desc, cd) {
648 /* no need to store info about volatile storage */
649 if (cd->storage_type == OSL_NO_STORAGE)
650 continue;
651 write_u16(buf + offset + IDX_CD_STORAGE_TYPE,
652 cd->storage_type);
653 write_u16(buf + offset + IDX_CD_STORAGE_FLAGS,
654 cd->storage_flags);
655 if (cd->storage_flags & OSL_FIXED_SIZE)
656 write_u32(buf + offset + IDX_CD_DATA_SIZE,
657 cd->data_size);
658 strcpy(buf + offset + IDX_CD_NAME, cd->name);
659 offset += index_column_description_size(cd->name);
660 }
661 assert(offset = size);
662 filename = index_filename(t->desc);
663 if (filename)
664 ret = write_file(filename, buf, size);
665 else
666 ret = -ERRNO_TO_ERROR(ENOMEM);
667 free(buf);
668 free(filename);
669 return ret;
670 }
671
672 /**
673 * Create a new osl table.
674 *
675 * \param desc Pointer to the table description.
676 *
677 * \return Standard.
678 */
679 __export int osl_create_table(const struct osl_table_description *desc)
680 {
681 const struct osl_column_description *cd;
682 char *table_dir = NULL, *filename;
683 struct osl_table *t;
684 int i, ret = init_table_structure(desc, &t);
685
686 if (ret < 0)
687 return ret;
688 INFO_LOG("creating %s\n", desc->name);
689 FOR_EACH_COLUMN(i, t->desc, cd) {
690 if (cd->storage_type == OSL_NO_STORAGE)
691 continue;
692 if (!table_dir) {
693 ret = para_mkdir(desc->dir, 0777);
694 if (ret < 0 && !is_errno(-ret, EEXIST))
695 goto out;
696 table_dir = make_message("%s/%s", desc->dir,
697 desc->name);
698 ret = -ERRNO_TO_ERROR(ENOMEM);
699 if (!table_dir)
700 goto out;
701 ret = para_mkdir(table_dir, 0777);
702 if (ret < 0)
703 goto out;
704 }
705 ret = -ERRNO_TO_ERROR(ENOMEM);
706 filename = column_filename(t, i);
707 if (!filename)
708 goto out;
709 INFO_LOG("filename: %s\n", filename);
710 if (cd->storage_type == OSL_MAPPED_STORAGE) {
711 ret = osl_open(filename, O_RDWR | O_CREAT | O_EXCL,
712 0644);
713 free(filename);
714 if (ret < 0)
715 goto out;
716 close(ret);
717 continue;
718 }
719 /* DISK STORAGE */
720 ret = para_mkdir(filename, 0777);
721 free(filename);
722 if (ret < 0)
723 goto out;
724 }
725 if (t->num_mapped_columns) {
726 ret = create_table_index(t);
727 if (ret < 0)
728 goto out;
729 }
730 ret = 1;
731 out:
732 free(table_dir);
733 free(t->columns);
734 free(t);
735 return ret;
736 }
737
738 static int table_is_dirty(struct osl_table *t)
739 {
740 char *buf = (char *)t->index_map.data + IDX_DIRTY_FLAG;
741 uint8_t dirty = read_u8(buf) & 0x1;
742 return !!dirty;
743 }
744
745 static void mark_table_dirty(struct osl_table *t)
746 {
747 char *buf = (char *)t->index_map.data + IDX_DIRTY_FLAG;
748 write_u8(buf, read_u8(buf) | 1);
749 }
750
751 static void mark_table_clean(struct osl_table *t)
752 {
753 char *buf = (char *)t->index_map.data + IDX_DIRTY_FLAG;
754 write_u8(buf, read_u8(buf) & 0xfe);
755 }
756
757 static void unmap_column(struct osl_table *t, unsigned col_num)
758 {
759 struct osl_object map = t->columns[col_num].data_map;
760 int ret;
761 if (!map.data)
762 return;
763 ret = para_munmap(map.data, map.size);
764 assert(ret > 0);
765 map.data = NULL;
766 }
767
768 /**
769 * Unmap all mapped files of an osl table.
770 *
771 * \param t Pointer to a mapped table.
772 * \param flags Options for unmapping.
773 *
774 * \return Positive on success, negative on errors.
775 *
776 * \sa map_table(), enum osl_close_flags, para_munmap().
777 */
778 int unmap_table(struct osl_table *t, enum osl_close_flags flags)
779 {
780 unsigned i;
781 const struct osl_column_description *cd;
782 int ret;
783
784 if (!t->num_mapped_columns) /* can this ever happen? */
785 return 1;
786 DEBUG_LOG("unmapping table '%s'\n", t->desc->name);
787 if (!t->index_map.data)
788 return -E_OSL_NOT_MAPPED;
789 if (flags & OSL_MARK_CLEAN)
790 mark_table_clean(t);
791 ret = para_munmap(t->index_map.data, t->index_map.size);
792 if (ret < 0)
793 return ret;
794 t->index_map.data = NULL;
795 if (!t->num_rows)
796 return 1;
797 FOR_EACH_MAPPED_COLUMN(i, t, cd)
798 unmap_column(t, i);
799 return 1;
800 }
801
802 static int map_column(struct osl_table *t, unsigned col_num)
803 {
804 struct stat statbuf;
805 char *filename = column_filename(t, col_num);
806 int ret = -E_OSL_STAT;
807
808 if (!filename)
809 return -ERRNO_TO_ERROR(ENOMEM);
810 if (stat(filename, &statbuf) < 0) {
811 free(filename);
812 return ret;
813 }
814 if (!(S_IFREG & statbuf.st_mode)) {
815 free(filename);
816 return ret;
817 }
818 ret = mmap_full_file(filename, O_RDWR,
819 &t->columns[col_num].data_map.data,
820 &t->columns[col_num].data_map.size,
821 NULL);
822 free(filename);
823 return ret;
824 }
825
826 /**
827 * Map the index file and all columns of type \p OSL_MAPPED_STORAGE into memory.
828 *
829 * \param t Pointer to an initialized table structure.
830 * \param flags Mapping options.
831 *
832 * \return Negative return value on errors; on success the number of rows
833 * (including invalid rows) is returned.
834 *
835 * \sa unmap_table(), enum map_table_flags, osl_open_table(), mmap(2).
836 */
837 int map_table(struct osl_table *t, enum map_table_flags flags)
838 {
839 char *filename;
840 const struct osl_column_description *cd;
841 int i = 0, ret, num_rows = 0;
842
843 if (!t->num_mapped_columns)
844 return 0;
845 if (t->index_map.data)
846 return -E_OSL_ALREADY_MAPPED;
847 filename = index_filename(t->desc);
848 if (!filename)
849 return -ERRNO_TO_ERROR(ENOMEM);
850 DEBUG_LOG("mapping table '%s' (index: %s)\n", t->desc->name, filename);
851 ret = mmap_full_file(filename, flags & MAP_TBL_FL_MAP_RDONLY?
852 O_RDONLY : O_RDWR, &t->index_map.data, &t->index_map.size, NULL);
853 free(filename);
854 if (ret < 0)
855 return ret;
856 if (flags & MAP_TBL_FL_VERIFY_INDEX) {
857 ret = compare_table_descriptions(t);
858 if (ret < 0)
859 goto err;
860 }
861 ret = -E_OSL_BUSY;
862 if (!(flags & MAP_TBL_FL_IGNORE_DIRTY)) {
863 if (table_is_dirty(t)) {
864 ERROR_LOG("%s is dirty\n", t->desc->name);
865 goto err;
866 }
867 }
868 mark_table_dirty(t);
869 num_rows = table_num_rows(t);
870 if (!num_rows)
871 return num_rows;
872 /* map data files */
873 FOR_EACH_MAPPED_COLUMN(i, t, cd) {
874 ret = map_column(t, i);
875 if (ret < 0)
876 goto err;
877 }
878 return num_rows;
879 err: /* unmap what is already mapped */
880 for (i--; i >= 0; i--) {
881 struct osl_object map = t->columns[i].data_map;
882 para_munmap(map.data, map.size);
883 map.data = NULL;
884 }
885 para_munmap(t->index_map.data, t->index_map.size);
886 t->index_map.data = NULL;
887 return ret;
888 }
889
890 /**
891 * Retrieve a mapped object by row and column number.
892 *
893 * \param t Pointer to an open osl table.
894 * \param col_num Number of the mapped column containing the object to retrieve.
895 * \param row_num Number of the row containing the object to retrieve.
896 * \param obj The result is returned here.
897 *
898 * It is considered an error if \a col_num does not refer to a column
899 * of storage type \p OSL_MAPPED_STORAGE.
900 *
901 * \return Standard.
902 *
903 * \sa osl_storage_type.
904 */
905 int get_mapped_object(const struct osl_table *t, unsigned col_num,
906 uint32_t row_num, struct osl_object *obj)
907 {
908 struct osl_column *col = &t->columns[col_num];
909 uint32_t offset;
910 char *header;
911 char *cell_index;
912 int ret;
913
914 if (t->num_rows <= row_num)
915 return -E_OSL_BAD_ROW_NUM;
916 ret = get_cell_index(t, row_num, col_num, &cell_index);
917 if (ret < 0)
918 return ret;
919 offset = read_u32(cell_index);
920 obj->size = read_u32(cell_index + 4) - 1;
921 header = col->data_map.data + offset;
922 obj->data = header + 1;
923 if (read_u8(header) == 0xff) {
924 ERROR_LOG("col %u, size %zu, offset %u\n", col_num,
925 obj->size, offset);
926 return -E_OSL_INVALID_OBJECT;
927 }
928 return 1;
929 }
930
931 static int search_rbtree(const struct osl_object *obj,
932 const struct osl_table *t, unsigned col_num,
933 struct rb_node **result, struct rb_node ***rb_link)
934 {
935 struct osl_column *col = &t->columns[col_num];
936 struct rb_node **new = &col->rbtree.rb_node, *parent = NULL;
937 const struct osl_column_description *cd =
938 get_column_description(t->desc, col_num);
939 enum osl_storage_type st = cd->storage_type;
940 while (*new) {
941 struct osl_row *this_row = get_row_pointer(*new,
942 col->rbtree_num);
943 int ret;
944 struct osl_object this_obj;
945 parent = *new;
946 if (st == OSL_MAPPED_STORAGE) {
947 ret = get_mapped_object(t, col_num, this_row->num,
948 &this_obj);
949 if (ret < 0)
950 return ret;
951 } else
952 this_obj = this_row->volatile_objects[col->volatile_num];
953 ret = cd->compare_function(obj, &this_obj);
954 if (!ret) {
955 if (result)
956 *result = get_rb_node_pointer(this_row,
957 col->rbtree_num);
958 return 1;
959 }
960 if (ret < 0)
961 new = &((*new)->rb_left);
962 else
963 new = &((*new)->rb_right);
964 }
965 if (result)
966 *result = parent;
967 if (rb_link)
968 *rb_link = new;
969 return -E_OSL_RB_KEY_NOT_FOUND;
970 }
971
972 static int insert_rbtree(struct osl_table *t, unsigned col_num,
973 const struct osl_row *row, const struct osl_object *obj)
974 {
975 struct rb_node *parent, **rb_link;
976 unsigned rbtree_num;
977 struct rb_node *n;
978 int ret = search_rbtree(obj, t, col_num, &parent, &rb_link);
979
980 if (ret > 0)
981 return -E_OSL_RB_KEY_EXISTS;
982 rbtree_num = t->columns[col_num].rbtree_num;
983 n = get_rb_node_pointer(row, rbtree_num);
984 rb_link_node(n, parent, rb_link);
985 rb_insert_color(n, &t->columns[col_num].rbtree);
986 return 1;
987 }
988
989 static void remove_rb_node(struct osl_table *t, unsigned col_num,
990 const struct osl_row *row)
991 {
992 struct osl_column *col = &t->columns[col_num];
993 const struct osl_column_description *cd =
994 get_column_description(t->desc, col_num);
995 enum osl_storage_flags sf = cd->storage_flags;
996 struct rb_node *victim, *splice_out_node, *tmp;
997 if (!(sf & OSL_RBTREE))
998 return;
999 /*
1000 * Which node is removed/spliced out actually depends on how many
1001 * children the victim node has: If it has no children, it gets
1002 * deleted. If it has one child, it gets spliced out. If it has two
1003 * children, its successor (which has at most a right child) gets
1004 * spliced out.
1005 */
1006 victim = get_rb_node_pointer(row, col->rbtree_num);
1007 if (victim->rb_left && victim->rb_right)
1008 splice_out_node = rb_next(victim);
1009 else
1010 splice_out_node = victim;
1011 /* Go up to the root and decrement the size of each node in the path. */
1012 for (tmp = splice_out_node; tmp; tmp = rb_parent(tmp))
1013 tmp->size--;
1014 rb_erase(victim, &col->rbtree);
1015 }
1016
1017 static int add_row_to_rbtrees(struct osl_table *t, uint32_t row_num,
1018 struct osl_object *volatile_objs, struct osl_row **row_ptr)
1019 {
1020 unsigned i;
1021 int ret;
1022 struct osl_row *row = allocate_row(t->num_rbtrees);
1023 const struct osl_column_description *cd;
1024
1025 if (!row)
1026 return -ERRNO_TO_ERROR(ENOMEM);
1027 row->num = row_num;
1028 row->volatile_objects = volatile_objs;
1029 FOR_EACH_RBTREE_COLUMN(i, t, cd) {
1030 if (cd->storage_type == OSL_MAPPED_STORAGE) {
1031 struct osl_object obj;
1032 ret = get_mapped_object(t, i, row_num, &obj);
1033 if (ret < 0)
1034 goto err;
1035 ret = insert_rbtree(t, i, row, &obj);
1036 } else { /* volatile */
1037 const struct osl_object *obj
1038 = volatile_objs + t->columns[i].volatile_num;
1039 ret = insert_rbtree(t, i, row, obj);
1040 }
1041 if (ret < 0)
1042 goto err;
1043 }
1044 if (row_ptr)
1045 *row_ptr = row;
1046 return 1;
1047 err: /* rollback changes, i.e. remove added entries from rbtrees */
1048 while (i)
1049 remove_rb_node(t, i--, row);
1050 free(row);
1051 return ret;
1052 }
1053
1054 static void free_volatile_objects(const struct osl_table *t,
1055 enum osl_close_flags flags)
1056 {
1057 int i, j;
1058 struct rb_node *n;
1059 struct osl_column *rb_col;
1060 const struct osl_column_description *cd;
1061
1062 if (!t->num_volatile_columns)
1063 return;
1064 /* find the first rbtree column (any will do) */
1065 FOR_EACH_RBTREE_COLUMN(i, t, cd)
1066 break;
1067 rb_col = t->columns + i;
1068 /* walk that rbtree and free all volatile objects */
1069 for (n = rb_first(&rb_col->rbtree); n; n = rb_next(n)) {
1070 struct osl_row *r = get_row_pointer(n, rb_col->rbtree_num);
1071 if (flags & OSL_FREE_VOLATILE)
1072 FOR_EACH_VOLATILE_COLUMN(j, t, cd) {
1073 if (cd->storage_flags & OSL_DONT_FREE)
1074 continue;
1075 free(r->volatile_objects[
1076 t->columns[j].volatile_num].data);
1077 }
1078 // for (j = 0; j < t->num_volatile_columns; j++)
1079 // free(r->volatile_objects[j].data);
1080 free(r->volatile_objects);
1081 }
1082 }
1083
1084 /**
1085 * Erase all rbtree nodes and free resources.
1086 *
1087 * \param t Pointer to an open osl table.
1088 *
1089 * This function is called by osl_close_table().
1090 */
1091 void clear_rbtrees(struct osl_table *t)
1092 {
1093 const struct osl_column_description *cd;
1094 unsigned i, rbtrees_cleared = 0;
1095
1096 FOR_EACH_RBTREE_COLUMN(i, t, cd) {
1097 struct osl_column *col = &t->columns[i];
1098 struct rb_node *n;
1099 rbtrees_cleared++;
1100 for (n = rb_first(&col->rbtree); n;) {
1101 struct osl_row *r;
1102 rb_erase(n, &col->rbtree);
1103 if (rbtrees_cleared == t->num_rbtrees) {
1104 r = get_row_pointer(n, col->rbtree_num);
1105 n = rb_next(n);
1106 free(r);
1107 } else
1108 n = rb_next(n);
1109 }
1110 }
1111
1112 }
1113
1114 /**
1115 * Close an osl table.
1116 *
1117 * \param t Pointer to the table to be closed.
1118 * \param flags Options for what should be cleaned up.
1119 *
1120 * If osl_open_table() succeeds, the resulting table pointer must later be
1121 * passed to this function in order to flush all changes to the file system and
1122 * to free the resources that were allocated by osl_open_table().
1123 *
1124 * \return Standard.
1125 *
1126 * \sa osl_open_table(), unmap_table().
1127 */
1128 __export int osl_close_table(struct osl_table *t, enum osl_close_flags flags)
1129 {
1130 int ret;
1131
1132 if (!t)
1133 return -E_OSL_BAD_TABLE;
1134 free_volatile_objects(t, flags);
1135 clear_rbtrees(t);
1136 ret = unmap_table(t, flags);
1137 if (ret < 0)
1138 ERROR_LOG("unmap_table failed: %d\n", ret);
1139 free(t->columns);
1140 free(t);
1141 return ret;
1142 }
1143
1144 /**
1145 * Find out whether the given row number corresponds to an invalid row.
1146 *
1147 * \param t Pointer to the osl table.
1148 * \param row_num The number of the row in question.
1149 *
1150 * By definition, a row is considered invalid if all its index entries
1151 * are invalid.
1152 *
1153 * \return Positive if \a row_num corresponds to an invalid row,
1154 * zero if it corresponds to a valid row, negative on errors.
1155 */
1156 int row_is_invalid(struct osl_table *t, uint32_t row_num)
1157 {
1158 char *row_index;
1159 int i, ret = get_row_index(t, row_num, &row_index);
1160
1161 if (ret < 0)
1162 return ret;
1163 for (i = 0; i < t->row_index_size; i++) {
1164 if ((unsigned char)row_index[i] != 0xff)
1165 return 0;
1166 }
1167 INFO_LOG("row %d is invalid\n", row_num);
1168 return 1;
1169 }
1170
1171 /**
1172 * Invalidate a row of an osl table.
1173 *
1174 * \param t Pointer to an open osl table.
1175 * \param row_num Number of the row to mark as invalid.
1176 *
1177 * This function marks each mapped object in the index entry of \a row as
1178 * invalid.
1179 *
1180 * \return Standard.
1181 */
1182 int mark_row_invalid(struct osl_table *t, uint32_t row_num)
1183 {
1184 char *row_index;
1185 int ret = get_row_index(t, row_num, &row_index);
1186
1187 if (ret < 0)
1188 return ret;
1189 INFO_LOG("marking row %d as invalid\n", row_num);
1190 memset(row_index, 0xff, t->row_index_size);
1191 return 1;
1192 }
1193
1194 /**
1195 * Initialize all rbtrees and compute number of invalid rows.
1196 *
1197 * \param t The table containing the rbtrees to be initialized.
1198 *
1199 * \return Standard.
1200 */
1201 int init_rbtrees(struct osl_table *t)
1202 {
1203 int i, ret;
1204 const struct osl_column_description *cd;
1205
1206 /* create rbtrees */
1207 FOR_EACH_RBTREE_COLUMN(i, t, cd)
1208 t->columns[i].rbtree = RB_ROOT;
1209 /* add valid rows to rbtrees */
1210 t->num_invalid_rows = 0;
1211 for (i = 0; i < t->num_rows; i++) {
1212 ret = row_is_invalid(t, i);
1213 if (ret < 0)
1214 return ret;
1215 if (ret) {
1216 t->num_invalid_rows++;
1217 continue;
1218 }
1219 ret = add_row_to_rbtrees(t, i, NULL, NULL);
1220 if (ret < 0)
1221 return ret;
1222 }
1223 return 1;
1224 }
1225
1226 /**
1227 * Open an osl table.
1228 *
1229 * Each osl table must be opened before its data can be accessed.
1230 *
1231 * \param table_desc Describes the table to be opened.
1232 * \param result Contains a pointer to the open table on success.
1233 *
1234 * The table description given by \a desc should coincide with the
1235 * description used at creation time.
1236 *
1237 * \return Standard.
1238 */
1239 __export int osl_open_table(const struct osl_table_description *table_desc,
1240 struct osl_table **result)
1241 {
1242 int i, ret;
1243 struct osl_table *t;
1244 const struct osl_column_description *cd;
1245
1246 INFO_LOG("opening table %s\n", table_desc->name);
1247 ret = init_table_structure(table_desc, &t);
1248 if (ret < 0)
1249 return ret;
1250 FOR_EACH_DISK_STORAGE_COLUMN(i, t, cd) {
1251 struct stat statbuf;
1252 char *dirname = column_filename(t, i);
1253
1254 ret = -ERRNO_TO_ERROR(ENOMEM);
1255 if (!dirname)
1256 goto err;
1257 /* check if directory exists */
1258 ret = stat(dirname, &statbuf);
1259 free(dirname);
1260 if (ret < 0) {
1261 ret = -ERRNO_TO_ERROR(errno);
1262 goto err;
1263 }
1264 ret = -ERRNO_TO_ERROR(ENOTDIR);
1265 if (!S_ISDIR(statbuf.st_mode))
1266 goto err;
1267 }
1268 ret = map_table(t, MAP_TBL_FL_VERIFY_INDEX);
1269 if (ret < 0)
1270 goto err;
1271 t->num_rows = ret;
1272 DEBUG_LOG("num rows: %d\n", t->num_rows);
1273 ret = init_rbtrees(t);
1274 if (ret < 0) {
1275 osl_close_table(t, OSL_MARK_CLEAN); /* ignore further errors */
1276 return ret;
1277 }
1278 *result = t;
1279 return 1;
1280 err:
1281 free(t->columns);
1282 free(t);
1283 return ret;
1284 }
1285
1286 static int create_disk_storage_object_dir(const struct osl_table *t,
1287 unsigned col_num, const char *ds_name)
1288 {
1289 char *dirname;
1290 int ret;
1291
1292 if (!(t->desc->flags & OSL_LARGE_TABLE))
1293 return 1;
1294 dirname = disk_storage_dirname(t, col_num, ds_name);
1295 if (!dirname)
1296 return -ERRNO_TO_ERROR(ENOMEM);
1297 ret = para_mkdir(dirname, 0777);
1298 free(dirname);
1299 if (ret < 0 && !is_errno(-ret, EEXIST))
1300 return ret;
1301 return 1;
1302 }
1303
1304 static int write_disk_storage_file(const struct osl_table *t, unsigned col_num,
1305 const struct osl_object *obj, const char *ds_name)
1306 {
1307 int ret;
1308 char *filename;
1309
1310 ret = create_disk_storage_object_dir(t, col_num, ds_name);
1311 if (ret < 0)
1312 return ret;
1313 filename = disk_storage_path(t, col_num, ds_name);
1314 if (!filename)
1315 return -ERRNO_TO_ERROR(ENOMEM);
1316 ret = write_file(filename, obj->data, obj->size);
1317 free(filename);
1318 return ret;
1319 }
1320
1321 static int append_map_file(const struct osl_table *t, unsigned col_num,
1322 const struct osl_object *obj, uint32_t *new_size)
1323 {
1324 char *filename = column_filename(t, col_num);
1325 int ret;
1326 char header = 0; /* zero means valid object */
1327
1328 if (!filename)
1329 return -ERRNO_TO_ERROR(ENOMEM);
1330 ret = append_file(filename, &header, 1, obj->data, obj->size,
1331 new_size);
1332 free(filename);
1333 return ret;
1334 }
1335
1336 static int append_row_index(const struct osl_table *t, char *row_index)
1337 {
1338 char *filename;
1339 int ret;
1340
1341 if (!t->num_mapped_columns)
1342 return 1;
1343 filename = index_filename(t->desc);
1344 if (!filename)
1345 return -ERRNO_TO_ERROR(ENOMEM);
1346 ret = append_file(filename, NULL, 0, row_index,
1347 t->row_index_size, NULL);
1348 free(filename);
1349 return ret;
1350 }
1351
1352 /**
1353 * A wrapper for truncate(2)
1354 *
1355 * \param path Name of the regular file to truncate
1356 * \param size Number of bytes to \b shave \b off
1357 *
1358 * Truncate the regular file named by \a path by \a size bytes.
1359 *
1360 * \return Standard.
1361 *
1362 * \sa truncate(2)
1363 */
1364 int para_truncate(const char *path, off_t size)
1365 {
1366 int ret;
1367 struct stat statbuf;
1368
1369 ret = -E_OSL_STAT;
1370 if (stat(path, &statbuf) < 0)
1371 goto out;
1372 ret = -E_OSL_BAD_SIZE;
1373 if (statbuf.st_size < size)
1374 goto out;
1375 ret = -E_OSL_TRUNC;
1376 if (truncate(path, statbuf.st_size - size) < 0)
1377 goto out;
1378 ret = 1;
1379 out:
1380 return ret;
1381 }
1382
1383 static int truncate_mapped_file(const struct osl_table *t, unsigned col_num,
1384 off_t size)
1385 {
1386 int ret;
1387 char *filename = column_filename(t, col_num);
1388
1389 if (!filename)
1390 return -ERRNO_TO_ERROR(ENOMEM);
1391 ret = para_truncate(filename, size);
1392 free(filename);
1393 return ret;
1394 }
1395
1396 static int delete_disk_storage_file(const struct osl_table *t, unsigned col_num,
1397 const char *ds_name)
1398 {
1399 char *dirname, *filename = disk_storage_path(t, col_num, ds_name);
1400 int ret, err;
1401
1402 if (!filename)
1403 return -ERRNO_TO_ERROR(ENOMEM);
1404 ret = unlink(filename);
1405 err = errno;
1406 free(filename);
1407 if (ret < 0)
1408 return -ERRNO_TO_ERROR(err);
1409 if (!(t->desc->flags & OSL_LARGE_TABLE))
1410 return 1;
1411 dirname = disk_storage_dirname(t, col_num, ds_name);
1412 if (!dirname)
1413 return -ERRNO_TO_ERROR(ENOMEM);
1414 rmdir(dirname);
1415 free(dirname);
1416 return 1;
1417 }
1418
1419 /**
1420 * Add a new row to an osl table and retrieve this row.
1421 *
1422 * \param t Pointer to an open osl table.
1423 * \param objects Array of objects to be added.
1424 * \param row Result pointer.
1425 *
1426 * The \a objects parameter must point to an array containing one object per
1427 * column. The order of the objects in the array is given by the table
1428 * description of \a table. Several sanity checks are performed during object
1429 * insertion and the function returns without modifying the table if any of
1430 * these tests fail. In fact, it is atomic in the sense that it either
1431 * succeeds or leaves the table unchanged (i.e. either all or none of the
1432 * objects are added to the table).
1433 *
1434 * It is considered an error if an object is added to a column with associated
1435 * rbtree if this object is equal to an object already contained in that column
1436 * (i.e. the compare function for the column's rbtree returns zero).
1437 *
1438 * \return Standard.
1439 *
1440 * \sa struct osl_table_description, osl_compare_func, osl_add_row().
1441 */
1442 __export int osl_add_and_get_row(struct osl_table *t, struct osl_object *objects,
1443 struct osl_row **row)
1444 {
1445 int i, ret;
1446 char *ds_name = NULL;
1447 struct rb_node **rb_parents = NULL, ***rb_links = NULL;
1448 char *new_row_index = NULL;
1449 struct osl_object *volatile_objs = NULL;
1450 const struct osl_column_description *cd;
1451
1452 if (!t)
1453 return -E_OSL_BAD_TABLE;
1454 rb_parents = malloc(t->num_rbtrees * sizeof(struct rn_node*));
1455 if (!rb_parents)
1456 return -ERRNO_TO_ERROR(ENOMEM);
1457 rb_links = malloc(t->num_rbtrees * sizeof(struct rn_node**));
1458 if (!rb_links) {
1459 free(rb_parents);
1460 return -ERRNO_TO_ERROR(ENOMEM);
1461 }
1462 if (t->num_mapped_columns) {
1463 new_row_index = malloc(t->row_index_size);
1464 if (!new_row_index) {
1465 free(rb_links);
1466 free(rb_parents);
1467 return -ERRNO_TO_ERROR(ENOMEM);
1468 }
1469 }
1470 /* pass 1: sanity checks */
1471 // DEBUG_LOG("sanity tests: %p:%p\n", objects[0].data,
1472 // objects[1].data);
1473 FOR_EACH_COLUMN(i, t->desc, cd) {
1474 enum osl_storage_type st = cd->storage_type;
1475 enum osl_storage_flags sf = cd->storage_flags;
1476
1477 // ret = -E_OSL_NULL_OBJECT;
1478 // if (!objects[i])
1479 // goto out;
1480 if (st == OSL_DISK_STORAGE)
1481 continue;
1482 if (sf & OSL_RBTREE) {
1483 unsigned rbtree_num = t->columns[i].rbtree_num;
1484 ret = -E_OSL_RB_KEY_EXISTS;
1485 // DEBUG_LOG("checking whether %p exists\n",
1486 // objects[i].data);
1487 if (search_rbtree(objects + i, t, i,
1488 &rb_parents[rbtree_num],
1489 &rb_links[rbtree_num]) > 0)
1490 goto out;
1491 }
1492 if (sf & OSL_FIXED_SIZE) {
1493 // DEBUG_LOG("fixed size. need: %zu, have: %d\n",
1494 // objects[i].size, cd->data_size);
1495 ret = -E_OSL_BAD_DATA_SIZE;
1496 if (objects[i].size != cd->data_size)
1497 goto out;
1498 }
1499 }
1500 if (t->num_disk_storage_columns) {
1501 ds_name = disk_storage_name_of_object(t,
1502 &objects[t->disk_storage_name_column]);
1503 ret = -ERRNO_TO_ERROR(ENOMEM);
1504 if (!ds_name)
1505 goto out;
1506 }
1507 ret = unmap_table(t, OSL_MARK_CLEAN);
1508 if (ret < 0)
1509 goto out;
1510 // DEBUG_LOG("sanity tests passed%s\n", "");
1511 /* pass 2: create data files, append map data */
1512 FOR_EACH_COLUMN(i, t->desc, cd) {
1513 enum osl_storage_type st = cd->storage_type;
1514 if (st == OSL_NO_STORAGE)
1515 continue;
1516 if (st == OSL_MAPPED_STORAGE) {
1517 uint32_t new_size;
1518 struct osl_column *col = &t->columns[i];
1519 // DEBUG_LOG("appending object of size %zu\n",
1520 // objects[i].size);
1521 ret = append_map_file(t, i, objects + i, &new_size);
1522 if (ret < 0)
1523 goto rollback;
1524 update_cell_index(new_row_index, col, new_size,
1525 objects[i].size);
1526 continue;
1527 }
1528 /* DISK_STORAGE */
1529 ret = write_disk_storage_file(t, i, objects + i, ds_name);
1530 if (ret < 0)
1531 goto rollback;
1532 }
1533 ret = append_row_index(t, new_row_index);
1534 if (ret < 0)
1535 goto rollback;
1536 ret = map_table(t, MAP_TBL_FL_VERIFY_INDEX);
1537 if (ret < 0) { /* truncate index and rollback changes */
1538 char *filename = index_filename(t->desc);
1539 if (filename)
1540 para_truncate(filename, t->row_index_size);
1541 free(filename);
1542 goto rollback;
1543 }
1544 /* pass 3: add entry to rbtrees */
1545 if (t->num_volatile_columns) {
1546 ret = -ERRNO_TO_ERROR(ENOMEM);
1547 volatile_objs = calloc(t->num_volatile_columns,
1548 sizeof(struct osl_object));
1549 if (!volatile_objs)
1550 goto out;
1551 FOR_EACH_VOLATILE_COLUMN(i, t, cd)
1552 volatile_objs[t->columns[i].volatile_num] = objects[i];
1553 }
1554 t->num_rows++;
1555 // DEBUG_LOG("adding new entry as row #%d\n", t->num_rows - 1);
1556 ret = add_row_to_rbtrees(t, t->num_rows - 1, volatile_objs, row);
1557 if (ret < 0)
1558 goto out;
1559 // DEBUG_LOG("added new entry as row #%d\n", t->num_rows - 1);
1560 ret = 1;
1561 goto out;
1562 rollback: /* rollback all changes made, ignore further errors */
1563 for (i--; i >= 0; i--) {
1564 cd = get_column_description(t->desc, i);
1565 enum osl_storage_type st = cd->storage_type;
1566 if (st == OSL_NO_STORAGE)
1567 continue;
1568
1569 if (st == OSL_MAPPED_STORAGE)
1570 truncate_mapped_file(t, i, objects[i].size);
1571 else /* disk storage */
1572 delete_disk_storage_file(t, i, ds_name);
1573 }
1574 /* ignore error and return previous error value */
1575 map_table(t, MAP_TBL_FL_VERIFY_INDEX);
1576 out:
1577 free(new_row_index);
1578 free(ds_name);
1579 free(rb_parents);
1580 free(rb_links);
1581 return ret;
1582 }
1583
1584 /**
1585 * Add a new row to an osl table.
1586 *
1587 * \param t Same meaning as osl_add_and_get_row().
1588 * \param objects Same meaning as osl_add_and_get_row().
1589 *
1590 * \return The return value of the underlying call to osl_add_and_get_row().
1591 *
1592 * This is equivalent to osl_add_and_get_row(t, objects, NULL).
1593 */
1594 __export int osl_add_row(struct osl_table *t, struct osl_object *objects)
1595 {
1596 return osl_add_and_get_row(t, objects, NULL);
1597 }
1598
1599 /**
1600 * Retrieve an object identified by row and column
1601 *
1602 * \param t Pointer to an open osl table.
1603 * \param r Pointer to the row.
1604 * \param col_num The column number.
1605 * \param object The result pointer.
1606 *
1607 * The column determined by \a col_num must be of type \p OSL_MAPPED_STORAGE
1608 * or \p OSL_NO_STORAGE, i.e. no disk storage objects may be retrieved by this
1609 * function.
1610 *
1611 * \return Standard.
1612 *
1613 * \sa osl_storage_type, osl_open_disk_object().
1614 */
1615 __export int osl_get_object(const struct osl_table *t, const struct osl_row *r,
1616 unsigned col_num, struct osl_object *object)
1617 {
1618 const struct osl_column_description *cd;
1619
1620 if (!t)
1621 return -E_OSL_BAD_TABLE;
1622 cd = get_column_description(t->desc, col_num);
1623 /* col must not be disk storage */
1624 if (cd->storage_type == OSL_DISK_STORAGE)
1625 return -E_OSL_BAD_STORAGE_TYPE;
1626 if (cd->storage_type == OSL_MAPPED_STORAGE)
1627 return get_mapped_object(t, col_num, r->num, object);
1628 /* volatile */
1629 *object = r->volatile_objects[t->columns[col_num].volatile_num];
1630 return 1;
1631 }
1632
1633 static int mark_mapped_object_invalid(const struct osl_table *t,
1634 uint32_t row_num, unsigned col_num)
1635 {
1636 struct osl_object obj;
1637 char *p;
1638 int ret = get_mapped_object(t, col_num, row_num, &obj);
1639
1640 if (ret < 0)
1641 return ret;
1642 p = obj.data;
1643 p--;
1644 *p = 0xff;
1645 return 1;
1646 }
1647
1648 /**
1649 * Delete a row from an osl table.
1650 *
1651 * \param t Pointer to an open osl table.
1652 * \param row Pointer to the row to delete.
1653 *
1654 * This removes all disk storage objects, removes all rbtree nodes, and frees
1655 * all volatile objects belonging to the given row. For mapped columns, the
1656 * data is merely marked invalid and may be pruned from time to time by
1657 * para_fsck.
1658 *
1659 * \return Standard.
1660 */
1661 __export int osl_del_row(struct osl_table *t, struct osl_row *row)
1662 {
1663 struct osl_row *r = row;
1664 int i, ret;
1665 const struct osl_column_description *cd;
1666
1667 if (!t)
1668 return -E_OSL_BAD_TABLE;
1669 INFO_LOG("deleting row %p\n", row);
1670
1671 if (t->num_disk_storage_columns) {
1672 char *ds_name;
1673 ret = disk_storage_name_of_row(t, r, &ds_name);
1674 if (ret < 0)
1675 goto out;
1676 FOR_EACH_DISK_STORAGE_COLUMN(i, t, cd)
1677 delete_disk_storage_file(t, i, ds_name);
1678 free(ds_name);
1679 }
1680 FOR_EACH_COLUMN(i, t->desc, cd) {
1681 struct osl_column *col = t->columns + i;
1682 enum osl_storage_type st = cd->storage_type;
1683 remove_rb_node(t, i, r);
1684 if (st == OSL_MAPPED_STORAGE) {
1685 mark_mapped_object_invalid(t, r->num, i);
1686 continue;
1687 }
1688 if (st == OSL_NO_STORAGE && !(cd->storage_flags & OSL_DONT_FREE))
1689 free(r->volatile_objects[col->volatile_num].data);
1690 }
1691 if (t->num_mapped_columns) {
1692 ret = mark_row_invalid(t, r->num);
1693 if (ret < 0)
1694 goto out;
1695 t->num_invalid_rows++;
1696 } else
1697 t->num_rows--;
1698 ret = 1;
1699 out:
1700 free(r->volatile_objects);
1701 free(r);
1702 return ret;
1703 }
1704
1705 /* test if column has an rbtree */
1706 static int check_rbtree_col(const struct osl_table *t, unsigned col_num,
1707 struct osl_column **col)
1708 {
1709 if (!t)
1710 return -E_OSL_BAD_TABLE;
1711 if (!(get_column_description(t->desc, col_num)->storage_flags & OSL_RBTREE))
1712 return -E_OSL_BAD_STORAGE_FLAGS;
1713 *col = t->columns + col_num;
1714 return 1;
1715 }
1716
1717 /**
1718 * Get the row that contains the given object.
1719 *
1720 * \param t Pointer to an open osl table.
1721 * \param col_num The number of the column to be searched.
1722 * \param obj The object to be looked up.
1723 * \param result Points to the row containing \a obj.
1724 *
1725 * Lookup \a obj in \a t and return the row containing \a obj. The column
1726 * specified by \a col_num must have an associated rbtree.
1727 *
1728 * \return Standard.
1729 *
1730 * \sa osl_storage_flags
1731 */
1732 __export int osl_get_row(const struct osl_table *t, unsigned col_num,
1733 const struct osl_object *obj, struct osl_row **result)
1734 {
1735 int ret;
1736 struct rb_node *node;
1737 struct osl_row *row;
1738 struct osl_column *col;
1739
1740 *result = NULL;
1741 ret = check_rbtree_col(t, col_num, &col);
1742 if (ret < 0)
1743 return ret;
1744 ret = search_rbtree(obj, t, col_num, &node, NULL);
1745 if (ret < 0)
1746 return ret;
1747 row = get_row_pointer(node, t->columns[col_num].rbtree_num);
1748 *result = row;
1749 return 1;
1750 }
1751
1752 static int rbtree_loop(struct osl_column *col, void *private_data,
1753 osl_rbtree_loop_func *func)
1754 {
1755 struct rb_node *n, *tmp;
1756
1757 /* this for-loop is safe against removal of an entry */
1758 for (n = rb_first(&col->rbtree), tmp = n? rb_next(n) : NULL;
1759 n;
1760 n = tmp, tmp = tmp? rb_next(tmp) : NULL) {
1761 struct osl_row *r = get_row_pointer(n, col->rbtree_num);
1762 if (func(r, private_data) < 0)
1763 return -E_OSL_LOOP;
1764 }
1765 return 1;
1766 }
1767
1768 static int rbtree_loop_reverse(struct osl_column *col, void *private_data,
1769 osl_rbtree_loop_func *func)
1770 {
1771 struct rb_node *n, *tmp;
1772
1773 /* safe against removal of an entry */
1774 for (n = rb_last(&col->rbtree), tmp = n? rb_prev(n) : NULL;
1775 n;
1776 n = tmp, tmp = tmp? rb_prev(tmp) : NULL) {
1777 struct osl_row *r = get_row_pointer(n, col->rbtree_num);
1778 if (func(r, private_data) < 0)
1779 return -E_OSL_LOOP;
1780 }
1781 return 1;
1782 }
1783
1784 /**
1785 * Loop over all nodes in an rbtree.
1786 *
1787 * \param t Pointer to an open osl table.
1788 * \param col_num The column to use for iterating over the elements.
1789 * \param private_data Pointer that gets passed to \a func.
1790 * \param func The function to be called for each node in the rbtree.
1791 *
1792 * This function does an in-order walk of the rbtree associated with \a
1793 * col_num. It is an error if the \p OSL_RBTREE flag is not set for this
1794 * column. For each node in the rbtree, the given function \a func is called
1795 * with two pointers as arguments: The first osl_row* argument points to the
1796 * row that contains the object corresponding to the rbtree node currently
1797 * traversed, and the \a private_data pointer is passed verbatim to \a func as the
1798 * second argument. The loop terminates either if \a func returns a negative
1799 * value, or if all nodes of the tree have been visited.
1800 *
1801 *
1802 * \return Standard. If the termination of the loop was caused by \a func
1803 * returning a negative value, \p -E_OSL_LOOP is returned.
1804 *
1805 * \sa osl_storage_flags, osl_rbtree_loop_reverse(), osl_compare_func.
1806 */
1807 __export int osl_rbtree_loop(const struct osl_table *t, unsigned col_num,
1808 void *private_data, osl_rbtree_loop_func *func)
1809 {
1810 struct osl_column *col;
1811
1812 int ret = check_rbtree_col(t, col_num, &col);
1813 if (ret < 0)
1814 return ret;
1815 return rbtree_loop(col, private_data, func);
1816 }
1817
1818 /**
1819 * Loop over all nodes in an rbtree in reverse order.
1820 *
1821 * \param t Identical meaning as in \p osl_rbtree_loop().
1822 * \param col_num Identical meaning as in \p osl_rbtree_loop().
1823 * \param private_data Identical meaning as in \p osl_rbtree_loop().
1824 * \param func Identical meaning as in \p osl_rbtree_loop().
1825 *
1826 * This function is identical to \p osl_rbtree_loop(), the only difference
1827 * is that the tree is walked in reverse order.
1828 *
1829 * \return The same return value as \p osl_rbtree_loop().
1830 *
1831 * \sa osl_rbtree_loop().
1832 */
1833 __export int osl_rbtree_loop_reverse(const struct osl_table *t, unsigned col_num,
1834 void *private_data, osl_rbtree_loop_func *func)
1835 {
1836 struct osl_column *col;
1837
1838 int ret = check_rbtree_col(t, col_num, &col);
1839 if (ret < 0)
1840 return ret;
1841 return rbtree_loop_reverse(col, private_data, func);
1842 }
1843
1844 /* TODO: Rollback changes on errors */
1845 static int rename_disk_storage_objects(struct osl_table *t,
1846 struct osl_object *old_obj, struct osl_object *new_obj)
1847 {
1848 int i, ret;
1849 const struct osl_column_description *cd;
1850 char *old_ds_name, *new_ds_name;
1851
1852 if (!t->num_disk_storage_columns)
1853 return 1; /* nothing to do */
1854 if (old_obj->size == new_obj->size && !memcmp(new_obj->data,
1855 old_obj->data, new_obj->size))
1856 return 1; /* object did not change */
1857 old_ds_name = disk_storage_name_of_object(t, old_obj);
1858 new_ds_name = disk_storage_name_of_object(t, new_obj);
1859 ret = -ERRNO_TO_ERROR(ENOMEM);
1860 if (!old_ds_name || ! new_ds_name)
1861 goto out;
1862
1863 FOR_EACH_DISK_STORAGE_COLUMN(i, t, cd) {
1864 char *old_filename, *new_filename;
1865 ret = create_disk_storage_object_dir(t, i, new_ds_name);
1866 if (ret < 0)
1867 goto out;
1868 old_filename = disk_storage_path(t, i, old_ds_name);
1869 new_filename = disk_storage_path(t, i, new_ds_name);
1870 if (!old_filename || !new_filename)
1871 ret = -ERRNO_TO_ERROR(ENOMEM);
1872 else
1873 ret = para_rename(old_filename, new_filename);
1874 free(old_filename);
1875 free(new_filename);
1876 if (ret < 0)
1877 goto out;
1878 }
1879 ret = 1;
1880 out:
1881 free(old_ds_name);
1882 free(new_ds_name);
1883 return ret;
1884
1885 }
1886
1887 /**
1888 * Change an object in an osl table.
1889 *
1890 * \param t Pointer to an open osl table.
1891 * \param r Pointer to the row containing the object to be updated.
1892 * \param col_num Number of the column containing the object to be updated.
1893 * \param obj Pointer to the replacement object.
1894 *
1895 * This function gets rid of all references to the old object. This includes
1896 * removal of the rbtree node in case there is an rbtree associated with \a
1897 * col_num. It then inserts \a obj into the table and the rbtree if necessary.
1898 *
1899 * If the \p OSL_RBTREE flag is set for \a col_num, you \b MUST call this
1900 * function in order to change the contents of an object, even for volatile or
1901 * mapped columns of constant size (which may be updated directly if \p
1902 * OSL_RBTREE is not set). Otherwise the rbtree might become corrupted.
1903 *
1904 * \return Standard
1905 */
1906 __export int osl_update_object(struct osl_table *t, const struct osl_row *r,
1907 unsigned col_num, struct osl_object *obj)
1908 {
1909 struct osl_column *col;
1910 const struct osl_column_description *cd;
1911 int ret;
1912
1913 if (!t)
1914 return -E_OSL_BAD_TABLE;
1915 col = &t->columns[col_num];
1916 cd = get_column_description(t->desc, col_num);
1917 DEBUG_LOG("updating column %u of %s\n", col_num, t->desc->name);
1918 if (cd->storage_flags & OSL_RBTREE) {
1919 if (search_rbtree(obj, t, col_num, NULL, NULL) > 0)
1920 return -E_OSL_RB_KEY_EXISTS;
1921 }
1922 if (cd->storage_flags & OSL_FIXED_SIZE) {
1923 if (obj->size != cd->data_size)
1924 return -E_OSL_BAD_DATA_SIZE;
1925 }
1926 remove_rb_node(t, col_num, r);
1927 if (cd->storage_type == OSL_NO_STORAGE) { /* TODO: If fixed size, reuse object? */
1928 free(r->volatile_objects[col->volatile_num].data);
1929 r->volatile_objects[col->volatile_num] = *obj;
1930 } else if (cd->storage_type == OSL_DISK_STORAGE) {
1931 char *ds_name;
1932 ret = disk_storage_name_of_row(t, r, &ds_name);
1933 if (ret < 0)
1934 return ret;
1935 ret = delete_disk_storage_file(t, col_num, ds_name);
1936 if (ret < 0 && !is_errno(-ret, ENOENT)) {
1937 free(ds_name);
1938 return ret;
1939 }
1940 ret = write_disk_storage_file(t, col_num, obj, ds_name);
1941 free(ds_name);
1942 if (ret < 0)
1943 return ret;
1944 } else { /* mapped storage */
1945 struct osl_object old_obj;
1946 ret = get_mapped_object(t, col_num, r->num, &old_obj);
1947 if (ret < 0)
1948 return ret;
1949 /*
1950 * If the updated column is the disk storage name column, the
1951 * disk storage name changes, so we have to rename all disk
1952 * storage objects accordingly.
1953 */
1954 if (col_num == t->disk_storage_name_column) {
1955 ret = rename_disk_storage_objects(t, &old_obj, obj);
1956 if (ret < 0)
1957 return ret;
1958 }
1959 if (cd->storage_flags & OSL_FIXED_SIZE)
1960 memcpy(old_obj.data, obj->data, cd->data_size);
1961 else { /* TODO: if the size doesn't change, use old space */
1962 uint32_t new_data_map_size;
1963 char *row_index;
1964 ret = get_row_index(t, r->num, &row_index);
1965 if (ret < 0)
1966 return ret;
1967 ret = mark_mapped_object_invalid(t, r->num, col_num);
1968 if (ret < 0)
1969 return ret;
1970 unmap_column(t, col_num);
1971 ret = append_map_file(t, col_num, obj,
1972 &new_data_map_size);
1973 if (ret < 0)
1974 return ret;
1975 ret = map_column(t, col_num);
1976 if (ret < 0)
1977 return ret;
1978 update_cell_index(row_index, col, new_data_map_size,
1979 obj->size);
1980 }
1981 }
1982 if (cd->storage_flags & OSL_RBTREE) {
1983 ret = insert_rbtree(t, col_num, r, obj);
1984 if (ret < 0)
1985 return ret;
1986 }
1987 return 1;
1988 }
1989
1990 /**
1991 * Retrieve an object of type \p OSL_DISK_STORAGE by row and column.
1992 *
1993 * \param t Pointer to an open osl table.
1994 * \param r Pointer to the row containing the object.
1995 * \param col_num The column number.
1996 * \param obj Points to the result upon successful return.
1997 *
1998 * For columns of type \p OSL_DISK_STORAGE, this function must be used to
1999 * retrieve one of its containing objects. Afterwards, osl_close_disk_object()
2000 * must be called in order to deallocate the resources.
2001 *
2002 * \return Standard.
2003 *
2004 * \sa osl_get_object(), osl_storage_type, osl_close_disk_object().
2005 */
2006 __export int osl_open_disk_object(const struct osl_table *t, const struct osl_row *r,
2007 unsigned col_num, struct osl_object *obj)
2008 {
2009 const struct osl_column_description *cd;
2010 char *ds_name, *filename;
2011 int ret;
2012
2013 if (!t)
2014 return -E_OSL_BAD_TABLE;
2015 cd = get_column_description(t->desc, col_num);
2016 if (cd->storage_type != OSL_DISK_STORAGE)
2017 return -E_OSL_BAD_STORAGE_TYPE;
2018
2019 ret = disk_storage_name_of_row(t, r, &ds_name);
2020 if (ret < 0)
2021 return ret;
2022 filename = disk_storage_path(t, col_num, ds_name);
2023 free(ds_name);
2024 if (!filename)
2025 return -ERRNO_TO_ERROR(ENOMEM);
2026 DEBUG_LOG("filename: %s\n", filename);
2027 ret = mmap_full_file(filename, O_RDONLY, &obj->data, &obj->size, NULL);
2028 free(filename);
2029 return ret;
2030 }
2031
2032 /**
2033 * Free resources that were allocated during osl_open_disk_object().
2034 *
2035 * \param obj Pointer to the object previously returned by open_disk_object().
2036 *
2037 * \return The return value of the underlying call to para_munmap().
2038 *
2039 * \sa para_munmap().
2040 */
2041 __export int osl_close_disk_object(struct osl_object *obj)
2042 {
2043 return para_munmap(obj->data, obj->size);
2044 }
2045
2046 /**
2047 * Get the number of rows of the given table.
2048 *
2049 * \param t Pointer to an open osl table.
2050 * \param num_rows Result is returned here.
2051 *
2052 * The number of rows returned via \a num_rows excluding any invalid rows.
2053 *
2054 * \return Positive on success, \p -E_OSL_BAD_TABLE if \a t is \p NULL.
2055 */
2056 __export int osl_get_num_rows(const struct osl_table *t, unsigned *num_rows)
2057 {
2058 if (!t)
2059 return -E_OSL_BAD_TABLE;
2060 assert(t->num_rows >= t->num_invalid_rows);
2061 *num_rows = t->num_rows - t->num_invalid_rows;
2062 return 1;
2063 }
2064
2065 /**
2066 * Get the rank of a row.
2067 *
2068 * \param t An open osl table.
2069 * \param r The row to get the rank of.
2070 * \param col_num The number of an rbtree column.
2071 * \param rank Result pointer.
2072 *
2073 * The rank is, by definition, the position of the row in the linear order
2074 * determined by an in-order tree walk of the rbtree associated with column
2075 * number \a col_num of \a table.
2076 *
2077 * \return Standard.
2078 *
2079 * \sa osl_get_nth_row().
2080 */
2081 __export int osl_get_rank(const struct osl_table *t, struct osl_row *r,
2082 unsigned col_num, unsigned *rank)
2083 {
2084 struct osl_object obj;
2085 struct osl_column *col;
2086 struct rb_node *node;
2087 int ret = check_rbtree_col(t, col_num, &col);
2088
2089 if (ret < 0)
2090 return ret;
2091 ret = osl_get_object(t, r, col_num, &obj);
2092 if (ret < 0)
2093 return ret;
2094 ret = search_rbtree(&obj, t, col_num, &node, NULL);
2095 if (ret < 0)
2096 return ret;
2097 ret = rb_rank(node, rank);
2098 if (ret < 0)
2099 return -E_OSL_BAD_ROW;
2100 return 1;
2101 }
2102
2103 /**
2104 * Get the row with n-th greatest value.
2105 *
2106 * \param t Pointer to an open osl table.
2107 * \param col_num The column number.
2108 * \param n The rank of the desired row.
2109 * \param result Row is returned here.
2110 *
2111 * Retrieve the n-th order statistic with respect to the compare function
2112 * of the rbtree column \a col_num. In other words, get that row with
2113 * \a n th greatest value in column \a col_num. It's an error if
2114 * \a col_num is not a rbtree column, or if \a n is larger than the
2115 * number of rows in the table.
2116 *
2117 * \return Standard.
2118 *
2119 * \sa osl_storage_flags, osl_compare_func, osl_get_row(),
2120 * osl_rbtree_last_row(), osl_rbtree_first_row(), osl_get_rank().
2121 */
2122 __export int osl_get_nth_row(const struct osl_table *t, unsigned col_num,
2123 unsigned n, struct osl_row **result)
2124 {
2125 struct osl_column *col;
2126 struct rb_node *node;
2127 unsigned num_rows;
2128 int ret;
2129
2130 if (n == 0)
2131 return -E_OSL_RB_KEY_NOT_FOUND;
2132 ret = osl_get_num_rows(t, &num_rows);
2133 if (ret < 0)
2134 return ret;
2135 if (n > num_rows)
2136 return -E_OSL_RB_KEY_NOT_FOUND;
2137 ret = check_rbtree_col(t, col_num, &col);
2138 if (ret < 0)
2139 return ret;
2140 node = rb_nth(col->rbtree.rb_node, n);
2141 if (!node)
2142 return -E_OSL_RB_KEY_NOT_FOUND;
2143 *result = get_row_pointer(node, col->rbtree_num);
2144 return 1;
2145 }
2146
2147 /**
2148 * Get the row corresponding to the smallest rbtree node of a column.
2149 *
2150 * \param t An open rbtree table.
2151 * \param col_num The number of the rbtree column.
2152 * \param result A pointer to the first row is returned here.
2153 *
2154 * The rbtree node of the smallest object (with respect to the corresponding
2155 * compare function) is selected and the row containing this object is
2156 * returned. It is an error if \a col_num refers to a column without an
2157 * associated rbtree.
2158 *
2159 * \return Standard.
2160 *
2161 * \sa osl_get_nth_row(), osl_rbtree_last_row().
2162 */
2163 __export int osl_rbtree_first_row(const struct osl_table *t, unsigned col_num,
2164 struct osl_row **result)
2165 {
2166 return osl_get_nth_row(t, col_num, 1, result);
2167 }
2168
2169 /**
2170 * Get the row corresponding to the greatest rbtree node of a column.
2171 *
2172 * \param t The same meaning as in \p osl_rbtree_first_row().
2173 * \param col_num The same meaning as in \p osl_rbtree_first_row().
2174 * \param result The same meaning as in \p osl_rbtree_first_row().
2175 *
2176 * This function works just like osl_rbtree_first_row(), the only difference
2177 * is that the row containing the greatest rather than the smallest object is
2178 * returned.
2179 *
2180 * \return Standard.
2181 *
2182 * \sa osl_get_nth_row(), osl_rbtree_first_row().
2183 */
2184 __export int osl_rbtree_last_row(const struct osl_table *t, unsigned col_num,
2185 struct osl_row **result)
2186 {
2187 unsigned num_rows;
2188 int ret = osl_get_num_rows(t, &num_rows);
2189
2190 if (ret < 0)
2191 return ret;
2192 return osl_get_nth_row(t, col_num, num_rows, result);
2193 }