c92412fbce765444c7a0d713cdeeffca35d824c1
[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 #include "log.h"
12 #include "osl.h"
13 #include "error.h"
14 #include "util.h"
15 #include "osl_core.h"
16
17 /* Taken from Drepper: How to write shared libraries, Appendix B. */
18 #include <stddef.h>
19 #define MSGSTRFIELD(line) MSGSTRFIELD1(line)
20 #define MSGSTRFIELD1(line) str##line
21 static const union msgstr_t {
22 struct {
23 #define OSL_ERROR(n, s) char MSGSTRFIELD(__LINE__)[sizeof(s)];
24 #include "errtab.h"
25 #undef OSL_ERROR
26 };
27 char str[0];
28 } msgstr = { {
29 #define OSL_ERROR(n, s) s,
30 #include "errtab.h"
31 #undef OSL_ERROR
32 } };
33 static const unsigned int errmsgidx[] = {
34 #define OSL_ERROR(n, s) [n] = offsetof(union msgstr_t, MSGSTRFIELD(__LINE__)),
35 #include "errtab.h"
36 #undef OSL_ERROR
37 };
38
39 __export const char *osl_strerror(int num)
40 {
41 if (IS_SYSTEM_ERROR(num))
42 return strerror((num) & ((1 << SYSTEM_ERROR_BIT) - 1));
43 return msgstr.str + errmsgidx[num];
44 }
45
46 int loglevel = 0;
47
48 static void __attribute ((constructor)) init_loglevel(void)
49 {
50 char *p = getenv("OSL_LOGLEVEL");
51
52 /* don't log anything if unset */
53 loglevel = p? atoi(p) : EMERG + 1;
54 }
55
56 /**
57 * The log function.
58 *
59 * \param ll Loglevel.
60 * \param fmt Usual format string.
61 *
62 * All XXX_LOG() macros use this function.
63 */
64 __printf_2_3 void __log(int ll, const char* fmt,...)
65 {
66 va_list argp;
67 FILE *outfd;
68 struct tm *tm;
69 time_t t1;
70 char str[255] = "";
71
72 if (ll < loglevel)
73 return;
74 outfd = stderr;
75 time(&t1);
76 tm = localtime(&t1);
77 strftime(str, sizeof(str), "%b %d %H:%M:%S", tm);
78 fprintf(outfd, "%s ", str);
79 va_start(argp, fmt);
80 vfprintf(outfd, fmt, argp);
81 va_end(argp);
82 }
83
84 /**
85 * A wrapper for lseek(2).
86 *
87 * \param fd The file descriptor whose offset is to be to repositioned.
88 * \param offset A value-result parameter.
89 * \param whence Usual repositioning directive.
90 *
91 * Reposition the offset of the file descriptor \a fd to the argument \a offset
92 * according to the directive \a whence. Upon successful return, \a offset
93 * contains the resulting offset location as measured in bytes from the
94 * beginning of the file.
95 *
96 * \return Positive on success. Otherwise, the function returns \p -E_OSL_LSEEK.
97 *
98 * \sa lseek(2).
99 */
100 static int __lseek(int fd, off_t *offset, int whence)
101 {
102 *offset = lseek(fd, *offset, whence);
103 int ret = -E_OSL_LSEEK;
104 if (*offset == -1)
105 return ret;
106 return 1;
107 }
108
109 static int append_file(const char *filename, char *data, size_t data_size,
110 uint32_t *new_pos)
111 {
112 int ret, fd;
113
114 // DEBUG_LOG("appending %zu + %zu bytes\n", header_size, data_size);
115 ret = osl_open(filename, O_WRONLY | O_CREAT | O_APPEND, 0644);
116 if (ret < 0)
117 return ret;
118 fd = ret;
119 ret = write_all(fd, data, &data_size);
120 if (ret < 0)
121 goto out;
122 if (new_pos) {
123 off_t offset = 0;
124 ret = __lseek(fd, &offset, SEEK_END);
125 if (ret < 0)
126 goto out;
127 // DEBUG_LOG("new file size: " FMT_OFF_T "\n", offset);
128 *new_pos = offset;
129 }
130 ret = 1;
131 out:
132 close(fd);
133 return ret;
134 }
135
136 static int verify_name(const char *name)
137 {
138 if (!name)
139 return -E_OSL_BAD_NAME;
140 if (!*name)
141 return -E_OSL_BAD_NAME;
142 if (strchr(name, '/'))
143 return -E_OSL_BAD_NAME;
144 if (!strcmp(name, ".."))
145 return -E_OSL_BAD_NAME;
146 if (!strcmp(name, "."))
147 return -E_OSL_BAD_NAME;
148 return 1;
149 }
150
151 int osl_hash_compare(const struct osl_object *obj1, const struct osl_object *obj2)
152 {
153 return hash_compare((HASH_TYPE *)obj1->data, (HASH_TYPE *)obj2->data);
154 }
155
156 static char *disk_storage_dirname(const struct osl_table *t, unsigned col_num,
157 const char *ds_name)
158 {
159 char *dirname, *column_name = column_filename(t, col_num);
160
161 if (!column_name)
162 return NULL;
163 if (!(t->desc->flags & OSL_LARGE_TABLE))
164 return column_name;
165 dirname = make_message("%s/%.2s", column_name, ds_name);
166 free(column_name);
167 return dirname;
168 }
169
170 static char *disk_storage_name_of_object(const struct osl_table *t,
171 const struct osl_object *obj)
172 {
173 HASH_TYPE hash[HASH_SIZE];
174 hash_object(obj, hash);
175 return disk_storage_name_of_hash(t, hash);
176 }
177
178 static int disk_storage_name_of_row(const struct osl_table *t,
179 const struct osl_row *row, char **name)
180 {
181 struct osl_object obj;
182 int ret = osl_get_object(t, row, t->disk_storage_name_column, &obj);
183
184 if (ret < 0)
185 return ret;
186 *name = disk_storage_name_of_object(t, &obj);
187 if (*name)
188 return 1;
189 return -ERRNO_TO_ERROR(ENOMEM);
190 }
191
192 static void column_name_hash(const char *col_name, HASH_TYPE *hash)
193 {
194 hash_function(col_name, strlen(col_name), hash);
195 }
196
197 static int init_column_descriptions(struct osl_table *t)
198 {
199 int i, j, ret;
200 const struct osl_column_description *cd;
201
202 ret = -E_OSL_BAD_TABLE_DESC;
203 ret = verify_name(t->desc->name);
204 if (ret < 0)
205 goto err;
206 ret = -E_OSL_BAD_DB_DIR;
207 if (!t->desc->dir && (t->num_disk_storage_columns || t->num_mapped_columns))
208 goto err;
209 /* the size of the index header without column descriptions */
210 t->index_header_size = IDX_COLUMN_DESCRIPTIONS;
211 FOR_EACH_COLUMN(i, t->desc, cd) {
212 struct osl_column *col = t->columns + i;
213 if (cd->storage_flags & OSL_RBTREE) {
214 if (!cd->compare_function)
215 return -E_OSL_NO_COMPARE_FUNC;
216 }
217 if (cd->storage_type == OSL_NO_STORAGE)
218 continue;
219 ret = -E_OSL_NO_COLUMN_NAME;
220 if (!cd->name || !cd->name[0])
221 goto err;
222 ret = verify_name(cd->name);
223 if (ret < 0)
224 goto err;
225 t->index_header_size += index_column_description_size(cd->name);
226 column_name_hash(cd->name, col->name_hash);
227 ret = -E_OSL_DUPLICATE_COL_NAME;
228 for (j = i + 1; j < t->desc->num_columns; j++) {
229 const char *name2 = get_column_description(t->desc,
230 j)->name;
231 if (cd->name && name2 && !strcmp(cd->name, name2))
232 goto err;
233 }
234 }
235 return 1;
236 err:
237 return ret;
238 }
239
240 /**
241 * Initialize a struct table from given table description.
242 *
243 * \param desc The description of the osl table.
244 * \param table_ptr Result is returned here.
245 *
246 * This function performs several sanity checks on \p desc and returns if any
247 * of these tests fail. On success, a struct \p osl_table is allocated and
248 * initialized with data derived from \p desc.
249 *
250 * \return Standard.
251 *
252 * \sa struct osl_table.
253 */
254 int init_table_structure(const struct osl_table_description *desc,
255 struct osl_table **table_ptr)
256 {
257 const struct osl_column_description *cd;
258 struct osl_table *t = calloc(1, sizeof(*t));
259 int i, ret = -ERRNO_TO_ERROR(ENOMEM), have_disk_storage_name_column = 0;
260
261 if (!t)
262 return ret;
263 ret = -E_OSL_BAD_TABLE_DESC;
264 if (!desc)
265 goto err;
266 DEBUG_LOG("creating table structure for '%s' from table "
267 "description\n", desc->name);
268 ret = -E_OSL_NO_COLUMN_DESC;
269 if (!desc->column_descriptions)
270 goto err;
271 ret = -E_OSL_NO_COLUMNS;
272 if (!desc->num_columns)
273 goto err;
274 ret = -ERRNO_TO_ERROR(ENOMEM);
275 t->columns = calloc(desc->num_columns, sizeof(struct osl_column));
276 if (!t->columns)
277 goto err;
278 t->desc = desc;
279 FOR_EACH_COLUMN(i, t->desc, cd) {
280 enum osl_storage_type st = cd->storage_type;
281 enum osl_storage_flags sf = cd->storage_flags;
282 struct osl_column *col = &t->columns[i];
283
284 ret = -E_OSL_BAD_STORAGE_TYPE;
285 if (st != OSL_MAPPED_STORAGE && st != OSL_DISK_STORAGE
286 && st != OSL_NO_STORAGE)
287 goto err;
288 ret = -E_OSL_BAD_STORAGE_FLAGS;
289 if (st == OSL_DISK_STORAGE && sf & OSL_RBTREE)
290 goto err;
291 ret = -E_OSL_BAD_STORAGE_SIZE;
292 if (sf & OSL_FIXED_SIZE && !cd->data_size)
293 goto err;
294 switch (st) {
295 case OSL_DISK_STORAGE:
296 t->num_disk_storage_columns++;
297 break;
298 case OSL_MAPPED_STORAGE:
299 t->num_mapped_columns++;
300 col->index_offset = t->row_index_size;
301 t->row_index_size += 8;
302 break;
303 case OSL_NO_STORAGE:
304 col->volatile_num = t->num_volatile_columns;
305 t->num_volatile_columns++;
306 break;
307 }
308 if (sf & OSL_RBTREE) {
309 col->rbtree_num = t->num_rbtrees;
310 t->num_rbtrees++;
311 if ((sf & OSL_UNIQUE) && (st == OSL_MAPPED_STORAGE)) {
312 if (!have_disk_storage_name_column)
313 t->disk_storage_name_column = i;
314 have_disk_storage_name_column = 1;
315 }
316 }
317 }
318 ret = -E_OSL_NO_UNIQUE_RBTREE_COLUMN;
319 if (t->num_disk_storage_columns && !have_disk_storage_name_column)
320 goto err;
321 ret = -E_OSL_NO_RBTREE_COL;
322 if (!t->num_rbtrees)
323 goto err;
324 /* success */
325 DEBUG_LOG("OK. Index entry size: %u\n", t->row_index_size);
326 ret = init_column_descriptions(t);
327 if (ret < 0)
328 goto err;
329 *table_ptr = t;
330 return 1;
331 err:
332 free(t->columns);
333 free(t);
334 return ret;
335 }
336
337 /**
338 * Read the table description from index header.
339 *
340 * \param map The memory mapping of the index file.
341 * \param desc The values found in the index header are returned here.
342 *
343 * Read the index header, check for the paraslash magic string and the table version number.
344 * Read all information stored in the index header into \a desc.
345 *
346 * \return Standard.
347 *
348 * \sa struct osl_table_description, osl_create_table.
349 */
350 int read_table_desc(struct osl_object *map, struct osl_table_description *desc)
351 {
352 char *buf = map->data;
353 uint8_t version;
354 uint16_t header_size;
355 int ret, i;
356 unsigned offset;
357 struct osl_column_description *cd;
358
359 if (map->size < MIN_INDEX_HEADER_SIZE(1))
360 return -E_OSL_SHORT_TABLE;
361 if (strncmp(buf + IDX_PARA_MAGIC, PARA_MAGIC, strlen(PARA_MAGIC)))
362 return -E_OSL_NO_MAGIC;
363 version = read_u8(buf + IDX_VERSION);
364 if (version < MIN_TABLE_VERSION || version > MAX_TABLE_VERSION)
365 return -E_OSL_VERSION_MISMATCH;
366 desc->flags = read_u8(buf + IDX_TABLE_FLAGS);
367 desc->num_columns = read_u16(buf + IDX_NUM_COLUMNS);
368 INFO_LOG("%u columns\n", desc->num_columns);
369 if (!desc->num_columns)
370 return -E_OSL_NO_COLUMNS;
371 header_size = read_u16(buf + IDX_HEADER_SIZE);
372 if (map->size < header_size)
373 return -E_OSL_BAD_SIZE;
374 desc->column_descriptions = calloc(desc->num_columns,
375 sizeof(struct osl_column_description));
376 if (!desc->column_descriptions)
377 return -ERRNO_TO_ERROR(ENOMEM);
378 offset = IDX_COLUMN_DESCRIPTIONS;
379 FOR_EACH_COLUMN(i, desc, cd) {
380 char *null_byte;
381
382 ret = -E_OSL_SHORT_TABLE;
383 if (map->size < offset + MIN_IDX_COLUMN_DESCRIPTION_SIZE) {
384 ERROR_LOG("map size = %zu < %u = offset + min desc size\n",
385 map->size, offset + MIN_IDX_COLUMN_DESCRIPTION_SIZE);
386 goto err;
387 }
388 cd->storage_type = read_u16(buf + offset + IDX_CD_STORAGE_TYPE);
389 cd->storage_flags = read_u16(buf + offset +
390 IDX_CD_STORAGE_FLAGS);
391 cd->data_size = read_u32(buf + offset + IDX_CD_DATA_SIZE);
392 null_byte = memchr(buf + offset + IDX_CD_NAME, '\0',
393 map->size - offset - IDX_CD_NAME);
394 ret = -E_OSL_INDEX_CORRUPTION;
395 if (!null_byte)
396 goto err;
397 ret = -ERRNO_TO_ERROR(ENOMEM);
398 cd->name = strdup(buf + offset + IDX_CD_NAME);
399 if (!cd->name)
400 goto err;
401 offset += index_column_description_size(cd->name);
402 }
403 if (offset != header_size) {
404 ret = -E_OSL_INDEX_CORRUPTION;
405 ERROR_LOG("real header size = %u != %u = stored header size\n",
406 offset, header_size);
407 goto err;
408 }
409 return 1;
410 err:
411 FOR_EACH_COLUMN(i, desc, cd)
412 free(cd->name);
413 return ret;
414 }
415
416 /*
417 * check whether the table description given by \p t->desc matches the on-disk
418 * table structure stored in the index of \a t.
419 */
420 static int compare_table_descriptions(struct osl_table *t)
421 {
422 int i, ret;
423 struct osl_table_description desc;
424 const struct osl_column_description *cd1, *cd2;
425
426 /* read the on-disk structure into desc */
427 ret = read_table_desc(&t->index_map, &desc);
428 if (ret < 0)
429 return ret;
430 ret = -E_OSL_BAD_TABLE_FLAGS;
431 if (desc.flags != t->desc->flags)
432 goto out;
433 ret = -E_OSL_BAD_COLUMN_NUM;
434 if (desc.num_columns > t->desc->num_columns)
435 goto out;
436 if (desc.num_columns < t->desc->num_columns) {
437 struct osl_column_description *cd;
438 unsigned diff = t->desc->num_columns - desc.num_columns;
439 INFO_LOG("extending table by %u volatile columns\n", diff);
440 ret = -ERRNO_TO_ERROR(ENOMEM);
441 desc.column_descriptions = realloc(desc.column_descriptions,
442 t->desc->num_columns * sizeof(struct osl_column_description));
443 if (!desc.column_descriptions)
444 goto out;
445 for (i = desc.num_columns; i < t->desc->num_columns; i++) {
446 cd = get_column_description(&desc, i);
447 cd->storage_type = OSL_NO_STORAGE;
448 cd->name = NULL;
449 }
450 desc.num_columns += diff;
451 }
452 FOR_EACH_COLUMN(i, t->desc, cd1) {
453 cd2 = get_column_description(&desc, i);
454 ret = -E_OSL_BAD_STORAGE_TYPE;
455 if (cd1->storage_type != cd2->storage_type)
456 goto out;
457 if (cd1->storage_type == OSL_NO_STORAGE)
458 continue;
459 ret = -E_OSL_BAD_STORAGE_FLAGS;
460 if (cd1->storage_flags != cd2->storage_flags) {
461 ERROR_LOG("sf1 = %u != %u = sf2\n",
462 cd1->storage_flags, cd2->storage_flags);
463 goto out;
464 }
465 ret = -E_OSL_BAD_DATA_SIZE;
466 if (cd1->storage_flags & OSL_FIXED_SIZE)
467 if (cd1->data_size != cd2->data_size)
468 goto out;
469 ret = -E_OSL_BAD_COLUMN_NAME;
470 if (strcmp(cd1->name, cd2->name))
471 goto out;
472 }
473 INFO_LOG("table description of '%s' matches on-disk data, good\n",
474 t->desc->name);
475 ret = 1;
476 out:
477 FOR_EACH_COLUMN(i, &desc, cd1)
478 free(cd1->name);
479 free(desc.column_descriptions);
480 return ret;
481 }
482
483 static int create_table_index(struct osl_table *t)
484 {
485 char *buf, *filename;
486 int i, ret;
487 size_t size = t->index_header_size;
488 const struct osl_column_description *cd;
489 unsigned offset;
490
491 INFO_LOG("creating %zu byte index for table %s\n", size,
492 t->desc->name);
493 buf = calloc(1, size);
494 if (!buf)
495 return -ERRNO_TO_ERROR(ENOMEM);
496 sprintf(buf + IDX_PARA_MAGIC, "%s", PARA_MAGIC);
497 write_u8(buf + IDX_TABLE_FLAGS, t->desc->flags);
498 write_u8(buf + IDX_DIRTY_FLAG, 0);
499 write_u8(buf + IDX_VERSION, CURRENT_TABLE_VERSION);
500 write_u16(buf + IDX_NUM_COLUMNS, t->num_mapped_columns + t->num_disk_storage_columns);
501 write_u16(buf + IDX_HEADER_SIZE, t->index_header_size);
502 offset = IDX_COLUMN_DESCRIPTIONS;
503 FOR_EACH_COLUMN(i, t->desc, cd) {
504 /* no need to store info about volatile storage */
505 if (cd->storage_type == OSL_NO_STORAGE)
506 continue;
507 write_u16(buf + offset + IDX_CD_STORAGE_TYPE,
508 cd->storage_type);
509 write_u16(buf + offset + IDX_CD_STORAGE_FLAGS,
510 cd->storage_flags);
511 if (cd->storage_flags & OSL_FIXED_SIZE)
512 write_u32(buf + offset + IDX_CD_DATA_SIZE,
513 cd->data_size);
514 strcpy(buf + offset + IDX_CD_NAME, cd->name);
515 offset += index_column_description_size(cd->name);
516 }
517 assert(offset = size);
518 filename = index_filename(t->desc);
519 if (filename)
520 ret = write_file(filename, buf, size);
521 else
522 ret = -ERRNO_TO_ERROR(ENOMEM);
523 free(buf);
524 free(filename);
525 return ret;
526 }
527
528 __export int osl_create_table(const struct osl_table_description *desc)
529 {
530 const struct osl_column_description *cd;
531 char *table_dir = NULL, *filename;
532 struct osl_table *t;
533 int i, ret = init_table_structure(desc, &t);
534
535 if (ret < 0)
536 return ret;
537 INFO_LOG("creating %s\n", desc->name);
538 FOR_EACH_COLUMN(i, t->desc, cd) {
539 if (cd->storage_type == OSL_NO_STORAGE)
540 continue;
541 if (!table_dir) {
542 ret = osl_mkdir(desc->dir, 0777);
543 if (ret < 0 && !is_errno(-ret, EEXIST))
544 goto out;
545 table_dir = make_message("%s/%s", desc->dir,
546 desc->name);
547 ret = -ERRNO_TO_ERROR(ENOMEM);
548 if (!table_dir)
549 goto out;
550 ret = osl_mkdir(table_dir, 0777);
551 if (ret < 0)
552 goto out;
553 }
554 ret = -ERRNO_TO_ERROR(ENOMEM);
555 filename = column_filename(t, i);
556 if (!filename)
557 goto out;
558 INFO_LOG("filename: %s\n", filename);
559 if (cd->storage_type == OSL_MAPPED_STORAGE) {
560 ret = osl_open(filename, O_RDWR | O_CREAT | O_EXCL,
561 0644);
562 free(filename);
563 if (ret < 0)
564 goto out;
565 close(ret);
566 continue;
567 }
568 /* DISK STORAGE */
569 ret = osl_mkdir(filename, 0777);
570 free(filename);
571 if (ret < 0)
572 goto out;
573 }
574 if (t->num_mapped_columns) {
575 ret = create_table_index(t);
576 if (ret < 0)
577 goto out;
578 }
579 ret = 1;
580 out:
581 free(table_dir);
582 free(t->columns);
583 free(t);
584 return ret;
585 }
586
587 static int table_is_dirty(struct osl_table *t)
588 {
589 char *buf = (char *)t->index_map.data + IDX_DIRTY_FLAG;
590 uint8_t dirty = read_u8(buf) & 0x1;
591 return !!dirty;
592 }
593
594 static void mark_table_dirty(struct osl_table *t)
595 {
596 char *buf = (char *)t->index_map.data + IDX_DIRTY_FLAG;
597 write_u8(buf, read_u8(buf) | 1);
598 }
599
600 static void mark_table_clean(struct osl_table *t)
601 {
602 char *buf = (char *)t->index_map.data + IDX_DIRTY_FLAG;
603 write_u8(buf, read_u8(buf) & 0xfe);
604 }
605
606 static void unmap_column(struct osl_table *t, unsigned col_num)
607 {
608 struct osl_object map = t->columns[col_num].data_map;
609 int ret;
610 if (!map.data)
611 return;
612 ret = osl_munmap(map.data, map.size);
613 assert(ret > 0);
614 map.data = NULL;
615 }
616
617 /**
618 * Unmap all mapped files of an osl table.
619 *
620 * \param t Pointer to a mapped table.
621 * \param flags Options for unmapping.
622 *
623 * \return Positive on success, negative on errors.
624 *
625 * \sa map_table(), enum osl_close_flags, osl_munmap().
626 */
627 int unmap_table(struct osl_table *t, enum osl_close_flags flags)
628 {
629 unsigned i;
630 const struct osl_column_description *cd;
631 int ret;
632
633 if (!t->num_mapped_columns) /* can this ever happen? */
634 return 1;
635 INFO_LOG("unmapping table '%s'\n", t->desc->name);
636 if (!t->index_map.data)
637 return -E_OSL_NOT_MAPPED;
638 if (flags & OSL_MARK_CLEAN)
639 mark_table_clean(t);
640 ret = osl_munmap(t->index_map.data, t->index_map.size);
641 if (ret < 0)
642 return ret;
643 t->index_map.data = NULL;
644 if (!t->num_rows)
645 return 1;
646 FOR_EACH_MAPPED_COLUMN(i, t, cd)
647 unmap_column(t, i);
648 return 1;
649 }
650
651 static int map_column(struct osl_table *t, unsigned col_num)
652 {
653 struct stat statbuf;
654 char *filename = column_filename(t, col_num);
655 int ret;
656
657 if (!filename)
658 return -ERRNO_TO_ERROR(ENOMEM);
659 ret = osl_stat(filename, &statbuf);
660 if (ret < 0) {
661 free(filename);
662 return ret;
663 }
664 if (!(S_IFREG & statbuf.st_mode)) {
665 free(filename);
666 return ret;
667 }
668 ret = mmap_full_file(filename, O_RDWR,
669 &t->columns[col_num].data_map.data,
670 &t->columns[col_num].data_map.size,
671 NULL);
672 free(filename);
673 return ret;
674 }
675
676 /**
677 * Map the index file and all columns of type \p OSL_MAPPED_STORAGE into memory.
678 *
679 * \param t Pointer to an initialized table structure.
680 * \param flags Mapping options.
681 *
682 * \return Negative return value on errors; on success the number of rows
683 * (including invalid rows) is returned.
684 *
685 * \sa unmap_table(), enum map_table_flags, osl_open_table(), mmap(2).
686 */
687 int map_table(struct osl_table *t, enum map_table_flags flags)
688 {
689 char *filename;
690 const struct osl_column_description *cd;
691 int i = 0, ret, num_rows = 0;
692
693 if (!t->num_mapped_columns)
694 return 0;
695 if (t->index_map.data)
696 return -E_OSL_ALREADY_MAPPED;
697 filename = index_filename(t->desc);
698 if (!filename)
699 return -ERRNO_TO_ERROR(ENOMEM);
700 INFO_LOG("mapping table '%s' (index: %s)\n", t->desc->name, filename);
701 ret = mmap_full_file(filename, flags & MAP_TBL_FL_MAP_RDONLY?
702 O_RDONLY : O_RDWR, &t->index_map.data, &t->index_map.size, NULL);
703 free(filename);
704 if (ret < 0)
705 return ret;
706 if (flags & MAP_TBL_FL_VERIFY_INDEX) {
707 ret = compare_table_descriptions(t);
708 if (ret < 0)
709 goto err;
710 }
711 ret = -E_OSL_BUSY;
712 if (!(flags & MAP_TBL_FL_IGNORE_DIRTY)) {
713 if (table_is_dirty(t)) {
714 ERROR_LOG("%s is dirty\n", t->desc->name);
715 goto err;
716 }
717 }
718 mark_table_dirty(t);
719 num_rows = table_num_rows(t);
720 if (!num_rows)
721 return num_rows;
722 /* map data files */
723 FOR_EACH_MAPPED_COLUMN(i, t, cd) {
724 ret = map_column(t, i);
725 if (ret < 0)
726 goto err;
727 }
728 return num_rows;
729 err: /* unmap what is already mapped */
730 for (i--; i >= 0; i--) {
731 struct osl_object map = t->columns[i].data_map;
732 osl_munmap(map.data, map.size);
733 map.data = NULL;
734 }
735 osl_munmap(t->index_map.data, t->index_map.size);
736 t->index_map.data = NULL;
737 return ret;
738 }
739
740 /**
741 * Retrieve a mapped object by row and column number.
742 *
743 * \param t Pointer to an open osl table.
744 * \param col_num Number of the mapped column containing the object to retrieve.
745 * \param row_num Number of the row containing the object to retrieve.
746 * \param obj The result is returned here.
747 *
748 * It is considered an error if \a col_num does not refer to a column
749 * of storage type \p OSL_MAPPED_STORAGE.
750 *
751 * \return Standard.
752 *
753 * \sa osl_storage_type.
754 */
755 int get_mapped_object(const struct osl_table *t, unsigned col_num,
756 uint32_t row_num, struct osl_object *obj)
757 {
758 struct osl_column *col = &t->columns[col_num];
759 uint32_t offset;
760 char *cell_index;
761 int ret;
762
763 if (t->num_rows <= row_num)
764 return -E_OSL_BAD_ROW_NUM;
765 ret = get_cell_index(t, row_num, col_num, &cell_index);
766 if (ret < 0)
767 return ret;
768 offset = read_u32(cell_index);
769 obj->size = read_u32(cell_index + 4);
770 obj->data = col->data_map.data + offset;
771 return 1;
772 }
773
774 /*
775 * It's OK to call this with result = rb_node = NULL. If result is not NULL,
776 * and rb key was not found, result points to the parent node.
777 */
778 static int search_rbtree(const struct osl_object *obj,
779 const struct osl_table *t, unsigned col_num,
780 struct rb_node **result, struct rb_node ***rb_link)
781 {
782 struct osl_column *col = &t->columns[col_num];
783 struct rb_node **new = &col->rbtree.rb_node, *parent = NULL;
784 const struct osl_column_description *cd =
785 get_column_description(t->desc, col_num);
786 enum osl_storage_type st = cd->storage_type;
787 while (*new) {
788 struct osl_row *this_row = get_row_pointer(*new,
789 col->rbtree_num);
790 int ret;
791 struct osl_object this_obj;
792 parent = *new;
793 if (st == OSL_MAPPED_STORAGE) {
794 ret = get_mapped_object(t, col_num, this_row->num,
795 &this_obj);
796 if (ret < 0)
797 return ret;
798 } else
799 this_obj = this_row->volatile_objects[col->volatile_num];
800 ret = cd->compare_function(obj, &this_obj);
801 if (!ret) {
802 if (result)
803 *result = get_rb_node_pointer(this_row,
804 col->rbtree_num);
805 return 1;
806 }
807 if (ret < 0)
808 new = &((*new)->rb_left);
809 else
810 new = &((*new)->rb_right);
811 }
812 if (result)
813 *result = parent;
814 if (rb_link)
815 *rb_link = new;
816 return -E_OSL_RB_KEY_NOT_FOUND;
817 }
818
819 static int insert_rbtree(struct osl_table *t, unsigned col_num,
820 const struct osl_row *row, const struct osl_object *obj)
821 {
822 struct rb_node *parent, **rb_link;
823 unsigned rbtree_num;
824 struct rb_node *n;
825 int ret = search_rbtree(obj, t, col_num, &parent, &rb_link);
826
827 if (ret > 0)
828 return -E_OSL_RB_KEY_EXISTS;
829 rbtree_num = t->columns[col_num].rbtree_num;
830 n = get_rb_node_pointer(row, rbtree_num);
831 rb_link_node(n, parent, rb_link);
832 rb_insert_color(n, &t->columns[col_num].rbtree);
833 return 1;
834 }
835
836 static void remove_rb_node(struct osl_table *t, unsigned col_num,
837 const struct osl_row *row)
838 {
839 struct osl_column *col = &t->columns[col_num];
840 const struct osl_column_description *cd =
841 get_column_description(t->desc, col_num);
842 enum osl_storage_flags sf = cd->storage_flags;
843 struct rb_node *victim, *splice_out_node, *tmp;
844 if (!(sf & OSL_RBTREE))
845 return;
846 /*
847 * Which node is removed/spliced out actually depends on how many
848 * children the victim node has: If it has no children, it gets
849 * deleted. If it has one child, it gets spliced out. If it has two
850 * children, its successor (which has at most a right child) gets
851 * spliced out.
852 */
853 victim = get_rb_node_pointer(row, col->rbtree_num);
854 if (victim->rb_left && victim->rb_right)
855 splice_out_node = rb_next(victim);
856 else
857 splice_out_node = victim;
858 /* Go up to the root and decrement the size of each node in the path. */
859 for (tmp = splice_out_node; tmp; tmp = rb_parent(tmp))
860 tmp->size--;
861 rb_erase(victim, &col->rbtree);
862 }
863
864 static int add_row_to_rbtrees(struct osl_table *t, uint32_t row_num,
865 struct osl_object *volatile_objs, struct osl_row **row_ptr)
866 {
867 unsigned i;
868 int ret;
869 struct osl_row *row = allocate_row(t->num_rbtrees);
870 const struct osl_column_description *cd;
871
872 if (!row)
873 return -ERRNO_TO_ERROR(ENOMEM);
874 row->num = row_num;
875 row->volatile_objects = volatile_objs;
876 FOR_EACH_RBTREE_COLUMN(i, t, cd) {
877 if (cd->storage_type == OSL_MAPPED_STORAGE) {
878 struct osl_object obj;
879 ret = get_mapped_object(t, i, row_num, &obj);
880 if (ret < 0)
881 goto err;
882 ret = insert_rbtree(t, i, row, &obj);
883 } else { /* volatile */
884 const struct osl_object *obj
885 = volatile_objs + t->columns[i].volatile_num;
886 ret = insert_rbtree(t, i, row, obj);
887 }
888 if (ret < 0)
889 goto err;
890 }
891 if (row_ptr)
892 *row_ptr = row;
893 return 1;
894 err: /* rollback changes, i.e. remove added entries from rbtrees */
895 while (i)
896 remove_rb_node(t, i--, row);
897 free(row);
898 return ret;
899 }
900
901 static void free_volatile_objects(const struct osl_table *t,
902 enum osl_close_flags flags)
903 {
904 int i, j;
905 struct rb_node *n;
906 struct osl_column *rb_col;
907 const struct osl_column_description *cd;
908
909 if (!t->num_volatile_columns)
910 return;
911 /* find the first rbtree column (any will do) */
912 FOR_EACH_RBTREE_COLUMN(i, t, cd)
913 break;
914 rb_col = t->columns + i;
915 /* walk that rbtree and free all volatile objects */
916 for (n = rb_first(&rb_col->rbtree); n; n = rb_next(n)) {
917 struct osl_row *r = get_row_pointer(n, rb_col->rbtree_num);
918 if (flags & OSL_FREE_VOLATILE)
919 FOR_EACH_VOLATILE_COLUMN(j, t, cd) {
920 if (cd->storage_flags & OSL_DONT_FREE)
921 continue;
922 free(r->volatile_objects[
923 t->columns[j].volatile_num].data);
924 }
925 // for (j = 0; j < t->num_volatile_columns; j++)
926 // free(r->volatile_objects[j].data);
927 free(r->volatile_objects);
928 }
929 }
930
931 /**
932 * Erase all rbtree nodes and free resources.
933 *
934 * \param t Pointer to an open osl table.
935 *
936 * This function is called by osl_close_table().
937 */
938 void clear_rbtrees(struct osl_table *t)
939 {
940 const struct osl_column_description *cd;
941 unsigned i, rbtrees_cleared = 0;
942
943 FOR_EACH_RBTREE_COLUMN(i, t, cd) {
944 struct osl_column *col = &t->columns[i];
945 struct rb_node *n;
946 rbtrees_cleared++;
947 for (n = rb_first(&col->rbtree); n;) {
948 struct osl_row *r;
949 rb_erase(n, &col->rbtree);
950 if (rbtrees_cleared == t->num_rbtrees) {
951 r = get_row_pointer(n, col->rbtree_num);
952 n = rb_next(n);
953 free(r);
954 } else
955 n = rb_next(n);
956 }
957 }
958
959 }
960
961 __export int osl_close_table(struct osl_table *t, enum osl_close_flags flags)
962 {
963 int ret;
964
965 if (!t)
966 return -E_OSL_BAD_TABLE;
967 NOTICE_LOG("closing table %s\n", t->desc->name);
968 free_volatile_objects(t, flags);
969 clear_rbtrees(t);
970 ret = unmap_table(t, flags);
971 if (ret < 0)
972 ERROR_LOG("unmap_table failed: %d\n", ret);
973 free(t->columns);
974 free(t);
975 return ret;
976 }
977
978 /**
979 * Find out whether the given row number corresponds to an invalid row.
980 *
981 * \param t Pointer to the osl table.
982 * \param row_num The number of the row in question.
983 *
984 * By definition, a row is considered invalid if all its index entries
985 * are invalid.
986 *
987 * \return Positive if \a row_num corresponds to an invalid row,
988 * zero if it corresponds to a valid row, negative on errors.
989 */
990 int row_is_invalid(struct osl_table *t, uint32_t row_num)
991 {
992 char *row_index;
993 int i, ret = get_row_index(t, row_num, &row_index);
994
995 if (ret < 0)
996 return ret;
997 for (i = 0; i < t->row_index_size; i++) {
998 if ((unsigned char)row_index[i] != 0xff)
999 return 0;
1000 }
1001 INFO_LOG("row %d is invalid\n", row_num);
1002 return 1;
1003 }
1004
1005 /**
1006 * Invalidate a row of an osl table.
1007 *
1008 * \param t Pointer to an open osl table.
1009 * \param row_num Number of the row to mark as invalid.
1010 *
1011 * This function marks each mapped object in the index entry of \a row as
1012 * invalid.
1013 *
1014 * \return Standard.
1015 */
1016 int mark_row_invalid(struct osl_table *t, uint32_t row_num)
1017 {
1018 char *row_index;
1019 int ret = get_row_index(t, row_num, &row_index);
1020
1021 if (ret < 0)
1022 return ret;
1023 INFO_LOG("marking row %d as invalid\n", row_num);
1024 memset(row_index, 0xff, t->row_index_size);
1025 return 1;
1026 }
1027
1028 /**
1029 * Initialize all rbtrees and compute number of invalid rows.
1030 *
1031 * \param t The table containing the rbtrees to be initialized.
1032 *
1033 * \return Standard.
1034 */
1035 int init_rbtrees(struct osl_table *t)
1036 {
1037 int i, ret;
1038 const struct osl_column_description *cd;
1039
1040 /* create rbtrees */
1041 FOR_EACH_RBTREE_COLUMN(i, t, cd)
1042 t->columns[i].rbtree = RB_ROOT;
1043 /* add valid rows to rbtrees */
1044 t->num_invalid_rows = 0;
1045 for (i = 0; i < t->num_rows; i++) {
1046 ret = row_is_invalid(t, i);
1047 if (ret < 0)
1048 return ret;
1049 if (ret) {
1050 t->num_invalid_rows++;
1051 continue;
1052 }
1053 ret = add_row_to_rbtrees(t, i, NULL, NULL);
1054 if (ret < 0)
1055 return ret;
1056 }
1057 return 1;
1058 }
1059
1060 __export int osl_open_table(const struct osl_table_description *table_desc,
1061 struct osl_table **result)
1062 {
1063 int i, ret;
1064 struct osl_table *t;
1065 const struct osl_column_description *cd;
1066
1067 NOTICE_LOG("opening table %s\n", table_desc->name);
1068 ret = init_table_structure(table_desc, &t);
1069 if (ret < 0)
1070 return ret;
1071 FOR_EACH_DISK_STORAGE_COLUMN(i, t, cd) {
1072 struct stat statbuf;
1073 char *dirname = column_filename(t, i);
1074
1075 ret = -ERRNO_TO_ERROR(ENOMEM);
1076 if (!dirname)
1077 goto err;
1078 /* check if directory exists */
1079 ret = stat(dirname, &statbuf);
1080 free(dirname);
1081 if (ret < 0) {
1082 ret = -ERRNO_TO_ERROR(errno);
1083 goto err;
1084 }
1085 ret = -ERRNO_TO_ERROR(ENOTDIR);
1086 if (!S_ISDIR(statbuf.st_mode))
1087 goto err;
1088 }
1089 ret = map_table(t, MAP_TBL_FL_VERIFY_INDEX);
1090 if (ret < 0)
1091 goto err;
1092 t->num_rows = ret;
1093 DEBUG_LOG("num rows: %d\n", t->num_rows);
1094 ret = init_rbtrees(t);
1095 if (ret < 0) {
1096 osl_close_table(t, OSL_MARK_CLEAN); /* ignore further errors */
1097 return ret;
1098 }
1099 *result = t;
1100 return 1;
1101 err:
1102 free(t->columns);
1103 free(t);
1104 return ret;
1105 }
1106
1107 static int create_disk_storage_object_dir(const struct osl_table *t,
1108 unsigned col_num, const char *ds_name)
1109 {
1110 char *dirname;
1111 int ret;
1112
1113 if (!(t->desc->flags & OSL_LARGE_TABLE))
1114 return 1;
1115 dirname = disk_storage_dirname(t, col_num, ds_name);
1116 if (!dirname)
1117 return -ERRNO_TO_ERROR(ENOMEM);
1118 ret = osl_mkdir(dirname, 0777);
1119 free(dirname);
1120 if (ret < 0 && !is_errno(-ret, EEXIST))
1121 return ret;
1122 return 1;
1123 }
1124
1125 static int write_disk_storage_file(const struct osl_table *t, unsigned col_num,
1126 const struct osl_object *obj, const char *ds_name)
1127 {
1128 int ret;
1129 char *filename;
1130
1131 ret = create_disk_storage_object_dir(t, col_num, ds_name);
1132 if (ret < 0)
1133 return ret;
1134 filename = disk_storage_path(t, col_num, ds_name);
1135 if (!filename)
1136 return -ERRNO_TO_ERROR(ENOMEM);
1137 ret = write_file(filename, obj->data, obj->size);
1138 free(filename);
1139 return ret;
1140 }
1141
1142 static int append_map_file(const struct osl_table *t, unsigned col_num,
1143 const struct osl_object *obj, uint32_t *new_size)
1144 {
1145 char *filename = column_filename(t, col_num);
1146 int ret;
1147
1148 if (!filename)
1149 return -ERRNO_TO_ERROR(ENOMEM);
1150 ret = append_file(filename, obj->data, obj->size, new_size);
1151 free(filename);
1152 return ret;
1153 }
1154
1155 static int append_row_index(const struct osl_table *t, char *row_index)
1156 {
1157 char *filename;
1158 int ret;
1159
1160 if (!t->num_mapped_columns)
1161 return 1;
1162 filename = index_filename(t->desc);
1163 if (!filename)
1164 return -ERRNO_TO_ERROR(ENOMEM);
1165 ret = append_file(filename, row_index, t->row_index_size, NULL);
1166 free(filename);
1167 return ret;
1168 }
1169
1170 static int truncate_mapped_file(const struct osl_table *t, unsigned col_num,
1171 off_t size)
1172 {
1173 int ret;
1174 char *filename = column_filename(t, col_num);
1175
1176 if (!filename)
1177 return -ERRNO_TO_ERROR(ENOMEM);
1178 ret = truncate_file(filename, size);
1179 free(filename);
1180 return ret;
1181 }
1182
1183 static int delete_disk_storage_file(const struct osl_table *t, unsigned col_num,
1184 const char *ds_name)
1185 {
1186 char *dirname, *filename = disk_storage_path(t, col_num, ds_name);
1187 int ret, err;
1188
1189 if (!filename)
1190 return -ERRNO_TO_ERROR(ENOMEM);
1191 ret = unlink(filename);
1192 err = errno;
1193 free(filename);
1194 if (ret < 0)
1195 return -ERRNO_TO_ERROR(err);
1196 if (!(t->desc->flags & OSL_LARGE_TABLE))
1197 return 1;
1198 dirname = disk_storage_dirname(t, col_num, ds_name);
1199 if (!dirname)
1200 return -ERRNO_TO_ERROR(ENOMEM);
1201 rmdir(dirname);
1202 free(dirname);
1203 return 1;
1204 }
1205
1206 __export int osl_add_and_get_row(struct osl_table *t, struct osl_object *objects,
1207 struct osl_row **row)
1208 {
1209 int i, ret;
1210 char *ds_name = NULL;
1211 struct rb_node **rb_parents = NULL, ***rb_links = NULL;
1212 char *new_row_index = NULL;
1213 struct osl_object *volatile_objs = NULL;
1214 const struct osl_column_description *cd;
1215
1216 if (!t)
1217 return -E_OSL_BAD_TABLE;
1218 rb_parents = malloc(t->num_rbtrees * sizeof(struct rn_node*));
1219 if (!rb_parents)
1220 return -ERRNO_TO_ERROR(ENOMEM);
1221 rb_links = malloc(t->num_rbtrees * sizeof(struct rn_node**));
1222 if (!rb_links) {
1223 free(rb_parents);
1224 return -ERRNO_TO_ERROR(ENOMEM);
1225 }
1226 if (t->num_mapped_columns) {
1227 new_row_index = malloc(t->row_index_size);
1228 if (!new_row_index) {
1229 free(rb_links);
1230 free(rb_parents);
1231 return -ERRNO_TO_ERROR(ENOMEM);
1232 }
1233 }
1234 /* pass 1: sanity checks */
1235 // DEBUG_LOG("sanity tests: %p:%p\n", objects[0].data,
1236 // objects[1].data);
1237 FOR_EACH_COLUMN(i, t->desc, cd) {
1238 enum osl_storage_type st = cd->storage_type;
1239 enum osl_storage_flags sf = cd->storage_flags;
1240
1241 // ret = -E_OSL_NULL_OBJECT;
1242 // if (!objects[i])
1243 // goto out;
1244 if (st == OSL_DISK_STORAGE)
1245 continue;
1246 if (sf & OSL_RBTREE) {
1247 unsigned rbtree_num = t->columns[i].rbtree_num;
1248 ret = -E_OSL_RB_KEY_EXISTS;
1249 // DEBUG_LOG("checking whether %p exists\n",
1250 // objects[i].data);
1251 if (search_rbtree(objects + i, t, i,
1252 &rb_parents[rbtree_num],
1253 &rb_links[rbtree_num]) > 0)
1254 goto out;
1255 }
1256 if (sf & OSL_FIXED_SIZE) {
1257 // DEBUG_LOG("fixed size. need: %zu, have: %d\n",
1258 // objects[i].size, cd->data_size);
1259 ret = -E_OSL_BAD_DATA_SIZE;
1260 if (objects[i].size != cd->data_size)
1261 goto out;
1262 }
1263 }
1264 if (t->num_disk_storage_columns) {
1265 ds_name = disk_storage_name_of_object(t,
1266 &objects[t->disk_storage_name_column]);
1267 ret = -ERRNO_TO_ERROR(ENOMEM);
1268 if (!ds_name)
1269 goto out;
1270 }
1271 ret = unmap_table(t, OSL_MARK_CLEAN);
1272 if (ret < 0)
1273 goto out;
1274 // DEBUG_LOG("sanity tests passed%s\n", "");
1275 /* pass 2: create data files, append map data */
1276 FOR_EACH_COLUMN(i, t->desc, cd) {
1277 enum osl_storage_type st = cd->storage_type;
1278 if (st == OSL_NO_STORAGE)
1279 continue;
1280 if (st == OSL_MAPPED_STORAGE) {
1281 uint32_t new_size;
1282 struct osl_column *col = &t->columns[i];
1283 // DEBUG_LOG("appending object of size %zu\n",
1284 // objects[i].size);
1285 ret = append_map_file(t, i, objects + i, &new_size);
1286 if (ret < 0)
1287 goto rollback;
1288 update_cell_index(new_row_index, col, new_size,
1289 objects[i].size);
1290 continue;
1291 }
1292 /* DISK_STORAGE */
1293 ret = write_disk_storage_file(t, i, objects + i, ds_name);
1294 if (ret < 0)
1295 goto rollback;
1296 }
1297 ret = append_row_index(t, new_row_index);
1298 if (ret < 0)
1299 goto rollback;
1300 ret = map_table(t, MAP_TBL_FL_VERIFY_INDEX);
1301 if (ret < 0) { /* truncate index and rollback changes */
1302 char *filename = index_filename(t->desc);
1303 if (filename)
1304 truncate_file(filename, t->row_index_size);
1305 free(filename);
1306 goto rollback;
1307 }
1308 /* pass 3: add entry to rbtrees */
1309 if (t->num_volatile_columns) {
1310 ret = -ERRNO_TO_ERROR(ENOMEM);
1311 volatile_objs = calloc(t->num_volatile_columns,
1312 sizeof(struct osl_object));
1313 if (!volatile_objs)
1314 goto out;
1315 FOR_EACH_VOLATILE_COLUMN(i, t, cd)
1316 volatile_objs[t->columns[i].volatile_num] = objects[i];
1317 }
1318 t->num_rows++;
1319 // DEBUG_LOG("adding new entry as row #%d\n", t->num_rows - 1);
1320 ret = add_row_to_rbtrees(t, t->num_rows - 1, volatile_objs, row);
1321 if (ret < 0)
1322 goto out;
1323 // DEBUG_LOG("added new entry as row #%d\n", t->num_rows - 1);
1324 ret = 1;
1325 goto out;
1326 rollback: /* rollback all changes made, ignore further errors */
1327 for (i--; i >= 0; i--) {
1328 cd = get_column_description(t->desc, i);
1329 enum osl_storage_type st = cd->storage_type;
1330 if (st == OSL_NO_STORAGE)
1331 continue;
1332
1333 if (st == OSL_MAPPED_STORAGE)
1334 truncate_mapped_file(t, i, objects[i].size);
1335 else /* disk storage */
1336 delete_disk_storage_file(t, i, ds_name);
1337 }
1338 /* ignore error and return previous error value */
1339 map_table(t, MAP_TBL_FL_VERIFY_INDEX);
1340 out:
1341 free(new_row_index);
1342 free(ds_name);
1343 free(rb_parents);
1344 free(rb_links);
1345 return ret;
1346 }
1347
1348 __export int osl_add_row(struct osl_table *t, struct osl_object *objects)
1349 {
1350 return osl_add_and_get_row(t, objects, NULL);
1351 }
1352
1353 __export int osl_get_object(const struct osl_table *t, const struct osl_row *r,
1354 unsigned col_num, struct osl_object *object)
1355 {
1356 const struct osl_column_description *cd;
1357
1358 if (!t)
1359 return -E_OSL_BAD_TABLE;
1360 cd = get_column_description(t->desc, col_num);
1361 /* col must not be disk storage */
1362 if (cd->storage_type == OSL_DISK_STORAGE)
1363 return -E_OSL_BAD_STORAGE_TYPE;
1364 if (cd->storage_type == OSL_MAPPED_STORAGE)
1365 return get_mapped_object(t, col_num, r->num, object);
1366 /* volatile */
1367 *object = r->volatile_objects[t->columns[col_num].volatile_num];
1368 return 1;
1369 }
1370
1371 __export int osl_del_row(struct osl_table *t, struct osl_row *row)
1372 {
1373 struct osl_row *r = row;
1374 int i, ret;
1375 const struct osl_column_description *cd;
1376
1377 if (!t)
1378 return -E_OSL_BAD_TABLE;
1379 INFO_LOG("deleting row %p\n", row);
1380
1381 if (t->num_disk_storage_columns) {
1382 char *ds_name;
1383 ret = disk_storage_name_of_row(t, r, &ds_name);
1384 if (ret < 0)
1385 goto out;
1386 FOR_EACH_DISK_STORAGE_COLUMN(i, t, cd)
1387 delete_disk_storage_file(t, i, ds_name);
1388 free(ds_name);
1389 }
1390 FOR_EACH_COLUMN(i, t->desc, cd) {
1391 struct osl_column *col = t->columns + i;
1392 enum osl_storage_type st = cd->storage_type;
1393 remove_rb_node(t, i, r);
1394 if (st == OSL_MAPPED_STORAGE)
1395 continue;
1396 if (st == OSL_NO_STORAGE && !(cd->storage_flags & OSL_DONT_FREE))
1397 free(r->volatile_objects[col->volatile_num].data);
1398 }
1399 if (t->num_mapped_columns) {
1400 ret = mark_row_invalid(t, r->num);
1401 if (ret < 0)
1402 goto out;
1403 t->num_invalid_rows++;
1404 } else
1405 t->num_rows--;
1406 ret = 1;
1407 out:
1408 free(r->volatile_objects);
1409 free(r);
1410 return ret;
1411 }
1412
1413 /* test if column has an rbtree */
1414 static int check_rbtree_col(const struct osl_table *t, unsigned col_num,
1415 struct osl_column **col)
1416 {
1417 if (!t)
1418 return -E_OSL_BAD_TABLE;
1419 if (!(get_column_description(t->desc, col_num)->storage_flags & OSL_RBTREE))
1420 return -E_OSL_BAD_STORAGE_FLAGS;
1421 *col = t->columns + col_num;
1422 return 1;
1423 }
1424
1425 __export int osl_get_row(const struct osl_table *t, unsigned col_num,
1426 const struct osl_object *obj, struct osl_row **result)
1427 {
1428 int ret;
1429 struct rb_node *node;
1430 struct osl_row *row;
1431 struct osl_column *col;
1432
1433 *result = NULL;
1434 ret = check_rbtree_col(t, col_num, &col);
1435 if (ret < 0)
1436 return ret;
1437 ret = search_rbtree(obj, t, col_num, &node, NULL);
1438 if (ret < 0)
1439 return ret;
1440 row = get_row_pointer(node, t->columns[col_num].rbtree_num);
1441 *result = row;
1442 return 1;
1443 }
1444
1445 static int rbtree_loop(struct osl_column *col, void *private_data,
1446 osl_rbtree_loop_func *func)
1447 {
1448 struct rb_node *n, *tmp;
1449
1450 /* this for-loop is safe against removal of an entry */
1451 for (n = rb_first(&col->rbtree), tmp = n? rb_next(n) : NULL;
1452 n;
1453 n = tmp, tmp = tmp? rb_next(tmp) : NULL) {
1454 struct osl_row *r = get_row_pointer(n, col->rbtree_num);
1455 if (func(r, private_data) < 0)
1456 return -E_OSL_LOOP;
1457 }
1458 return 1;
1459 }
1460
1461 static int rbtree_loop_reverse(struct osl_column *col, void *private_data,
1462 osl_rbtree_loop_func *func)
1463 {
1464 struct rb_node *n, *tmp;
1465
1466 /* safe against removal of an entry */
1467 for (n = rb_last(&col->rbtree), tmp = n? rb_prev(n) : NULL;
1468 n;
1469 n = tmp, tmp = tmp? rb_prev(tmp) : NULL) {
1470 struct osl_row *r = get_row_pointer(n, col->rbtree_num);
1471 if (func(r, private_data) < 0)
1472 return -E_OSL_LOOP;
1473 }
1474 return 1;
1475 }
1476
1477 __export int osl_rbtree_loop(const struct osl_table *t, unsigned col_num,
1478 void *private_data, osl_rbtree_loop_func *func)
1479 {
1480 struct osl_column *col;
1481
1482 int ret = check_rbtree_col(t, col_num, &col);
1483 if (ret < 0)
1484 return ret;
1485 return rbtree_loop(col, private_data, func);
1486 }
1487
1488 __export int osl_rbtree_loop_reverse(const struct osl_table *t, unsigned col_num,
1489 void *private_data, osl_rbtree_loop_func *func)
1490 {
1491 struct osl_column *col;
1492
1493 int ret = check_rbtree_col(t, col_num, &col);
1494 if (ret < 0)
1495 return ret;
1496 return rbtree_loop_reverse(col, private_data, func);
1497 }
1498
1499 /* TODO: Rollback changes on errors */
1500 static int rename_disk_storage_objects(struct osl_table *t,
1501 struct osl_object *old_obj, struct osl_object *new_obj)
1502 {
1503 int i, ret;
1504 const struct osl_column_description *cd;
1505 char *old_ds_name, *new_ds_name;
1506
1507 if (!t->num_disk_storage_columns)
1508 return 1; /* nothing to do */
1509 if (old_obj->size == new_obj->size && !memcmp(new_obj->data,
1510 old_obj->data, new_obj->size))
1511 return 1; /* object did not change */
1512 old_ds_name = disk_storage_name_of_object(t, old_obj);
1513 new_ds_name = disk_storage_name_of_object(t, new_obj);
1514 ret = -ERRNO_TO_ERROR(ENOMEM);
1515 if (!old_ds_name || ! new_ds_name)
1516 goto out;
1517
1518 FOR_EACH_DISK_STORAGE_COLUMN(i, t, cd) {
1519 char *old_filename, *new_filename;
1520 ret = create_disk_storage_object_dir(t, i, new_ds_name);
1521 if (ret < 0)
1522 goto out;
1523 old_filename = disk_storage_path(t, i, old_ds_name);
1524 new_filename = disk_storage_path(t, i, new_ds_name);
1525 if (!old_filename || !new_filename)
1526 ret = -ERRNO_TO_ERROR(ENOMEM);
1527 else
1528 ret = osl_rename(old_filename, new_filename);
1529 free(old_filename);
1530 free(new_filename);
1531 if (ret < 0)
1532 goto out;
1533 }
1534 ret = 1;
1535 out:
1536 free(old_ds_name);
1537 free(new_ds_name);
1538 return ret;
1539
1540 }
1541
1542 __export int osl_update_object(struct osl_table *t, const struct osl_row *r,
1543 unsigned col_num, struct osl_object *obj)
1544 {
1545 struct osl_column *col;
1546 const struct osl_column_description *cd;
1547 int ret;
1548
1549 if (!t)
1550 return -E_OSL_BAD_TABLE;
1551 col = &t->columns[col_num];
1552 cd = get_column_description(t->desc, col_num);
1553 DEBUG_LOG("updating column %u of %s\n", col_num, t->desc->name);
1554 if (cd->storage_flags & OSL_RBTREE) {
1555 if (search_rbtree(obj, t, col_num, NULL, NULL) > 0)
1556 return -E_OSL_RB_KEY_EXISTS;
1557 }
1558 if (cd->storage_flags & OSL_FIXED_SIZE) {
1559 if (obj->size != cd->data_size)
1560 return -E_OSL_BAD_DATA_SIZE;
1561 }
1562 remove_rb_node(t, col_num, r);
1563 if (cd->storage_type == OSL_NO_STORAGE) { /* TODO: If fixed size, reuse object? */
1564 free(r->volatile_objects[col->volatile_num].data);
1565 r->volatile_objects[col->volatile_num] = *obj;
1566 } else if (cd->storage_type == OSL_DISK_STORAGE) {
1567 char *ds_name;
1568 ret = disk_storage_name_of_row(t, r, &ds_name);
1569 if (ret < 0)
1570 return ret;
1571 ret = delete_disk_storage_file(t, col_num, ds_name);
1572 if (ret < 0 && !is_errno(-ret, ENOENT)) {
1573 free(ds_name);
1574 return ret;
1575 }
1576 ret = write_disk_storage_file(t, col_num, obj, ds_name);
1577 free(ds_name);
1578 if (ret < 0)
1579 return ret;
1580 } else { /* mapped storage */
1581 struct osl_object old_obj;
1582 ret = get_mapped_object(t, col_num, r->num, &old_obj);
1583 if (ret < 0)
1584 return ret;
1585 /*
1586 * If the updated column is the disk storage name column, the
1587 * disk storage name changes, so we have to rename all disk
1588 * storage objects accordingly.
1589 */
1590 if (col_num == t->disk_storage_name_column) {
1591 ret = rename_disk_storage_objects(t, &old_obj, obj);
1592 if (ret < 0)
1593 return ret;
1594 }
1595 if (cd->storage_flags & OSL_FIXED_SIZE)
1596 memcpy(old_obj.data, obj->data, cd->data_size);
1597 else { /* TODO: if the size doesn't change, use old space */
1598 uint32_t new_data_map_size;
1599 char *row_index;
1600 ret = get_row_index(t, r->num, &row_index);
1601 if (ret < 0)
1602 return ret;
1603 unmap_column(t, col_num);
1604 ret = append_map_file(t, col_num, obj,
1605 &new_data_map_size);
1606 if (ret < 0)
1607 return ret;
1608 ret = map_column(t, col_num);
1609 if (ret < 0)
1610 return ret;
1611 update_cell_index(row_index, col, new_data_map_size,
1612 obj->size);
1613 }
1614 }
1615 if (cd->storage_flags & OSL_RBTREE) {
1616 ret = insert_rbtree(t, col_num, r, obj);
1617 if (ret < 0)
1618 return ret;
1619 }
1620 return 1;
1621 }
1622
1623 __export int osl_open_disk_object(const struct osl_table *t, const struct osl_row *r,
1624 unsigned col_num, struct osl_object *obj)
1625 {
1626 const struct osl_column_description *cd;
1627 char *ds_name, *filename;
1628 int ret;
1629
1630 if (!t)
1631 return -E_OSL_BAD_TABLE;
1632 cd = get_column_description(t->desc, col_num);
1633 if (cd->storage_type != OSL_DISK_STORAGE)
1634 return -E_OSL_BAD_STORAGE_TYPE;
1635
1636 ret = disk_storage_name_of_row(t, r, &ds_name);
1637 if (ret < 0)
1638 return ret;
1639 filename = disk_storage_path(t, col_num, ds_name);
1640 free(ds_name);
1641 if (!filename)
1642 return -ERRNO_TO_ERROR(ENOMEM);
1643 DEBUG_LOG("filename: %s\n", filename);
1644 ret = mmap_full_file(filename, O_RDONLY, &obj->data, &obj->size, NULL);
1645 free(filename);
1646 return ret;
1647 }
1648
1649 __export int osl_close_disk_object(struct osl_object *obj)
1650 {
1651 return osl_munmap(obj->data, obj->size);
1652 }
1653
1654 __export int osl_get_num_rows(const struct osl_table *t, unsigned *num_rows)
1655 {
1656 if (!t)
1657 return -E_OSL_BAD_TABLE;
1658 assert(t->num_rows >= t->num_invalid_rows);
1659 *num_rows = t->num_rows - t->num_invalid_rows;
1660 return 1;
1661 }
1662
1663 __export int osl_get_rank(const struct osl_table *t, struct osl_row *r,
1664 unsigned col_num, unsigned *rank)
1665 {
1666 struct osl_object obj;
1667 struct osl_column *col;
1668 struct rb_node *node;
1669 int ret = check_rbtree_col(t, col_num, &col);
1670
1671 if (ret < 0)
1672 return ret;
1673 ret = osl_get_object(t, r, col_num, &obj);
1674 if (ret < 0)
1675 return ret;
1676 ret = search_rbtree(&obj, t, col_num, &node, NULL);
1677 if (ret < 0)
1678 return ret;
1679 ret = rb_rank(node, rank);
1680 if (ret < 0)
1681 return -E_OSL_BAD_ROW;
1682 return 1;
1683 }
1684
1685 __export int osl_get_nth_row(const struct osl_table *t, unsigned col_num,
1686 unsigned n, struct osl_row **result)
1687 {
1688 struct osl_column *col;
1689 struct rb_node *node;
1690 unsigned num_rows;
1691 int ret;
1692
1693 if (n == 0)
1694 return -E_OSL_RB_KEY_NOT_FOUND;
1695 ret = osl_get_num_rows(t, &num_rows);
1696 if (ret < 0)
1697 return ret;
1698 if (n > num_rows)
1699 return -E_OSL_RB_KEY_NOT_FOUND;
1700 ret = check_rbtree_col(t, col_num, &col);
1701 if (ret < 0)
1702 return ret;
1703 node = rb_nth(col->rbtree.rb_node, n);
1704 if (!node)
1705 return -E_OSL_RB_KEY_NOT_FOUND;
1706 *result = get_row_pointer(node, col->rbtree_num);
1707 return 1;
1708 }
1709
1710 __export int osl_rbtree_first_row(const struct osl_table *t, unsigned col_num,
1711 struct osl_row **result)
1712 {
1713 return osl_get_nth_row(t, col_num, 1, result);
1714 }
1715
1716 __export int osl_rbtree_last_row(const struct osl_table *t, unsigned col_num,
1717 struct osl_row **result)
1718 {
1719 unsigned num_rows;
1720 int ret = osl_get_num_rows(t, &num_rows);
1721
1722 if (ret < 0)
1723 return ret;
1724 return osl_get_nth_row(t, col_num, num_rows, result);
1725 }