Get rid of E_OSL_STAT.
[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 "fd.h"
15 #include "list.h"
16 #include "osl_core.h"
17
18 /* Taken from Drepper: How to write shared libraries, Appendix B. */
19 #include <stddef.h>
20 #define MSGSTRFIELD(line) MSGSTRFIELD1(line)
21 #define MSGSTRFIELD1(line) str##line
22 static const union msgstr_t {
23         struct {
24 #define _S(n, s) char MSGSTRFIELD(__LINE__)[sizeof(s)];
25 #include "errtab.h"
26 #undef _S
27         };
28         char str[0];
29 } msgstr = { {
30 #define _S(n, s) s,
31 #include "errtab.h"
32 #undef _S
33 } };
34 static const unsigned int errmsgidx[] = {
35 #define _S(n, s) [n] = offsetof(union msgstr_t, MSGSTRFIELD(__LINE__)),
36 #include "errtab.h"
37 #undef _S
38 };
39
40 __export const char *osl_strerror(int num)
41 {
42         if (IS_SYSTEM_ERROR(num))
43                 return strerror((num) & ((1 << SYSTEM_ERROR_BIT) - 1));
44         return msgstr.str + errmsgidx[num];
45 }
46
47 int loglevel = 0;
48
49 static void __attribute ((constructor)) init_loglevel(void)
50 {
51         char *p = getenv("OSL_LOGLEVEL");
52
53         /* don't log anything if unset */
54         loglevel = p? atoi(p) : EMERG + 1;
55 }
56
57 /**
58  * The log function.
59  *
60  * \param ll Loglevel.
61  * \param fmt Usual format string.
62  *
63  * All XXX_LOG() macros use this function.
64  */
65 __printf_2_3 void __log(int ll, const char* fmt,...)
66 {
67         va_list argp;
68         FILE *outfd;
69         struct tm *tm;
70         time_t t1;
71         char str[255] = "";
72
73         if (ll < loglevel)
74                 return;
75         outfd = stderr;
76         time(&t1);
77         tm = localtime(&t1);
78         strftime(str, sizeof(str), "%b %d %H:%M:%S", tm);
79         fprintf(outfd, "%s ", str);
80         va_start(argp, fmt);
81         vfprintf(outfd, fmt, argp);
82         va_end(argp);
83 }
84
85 /**
86  * A wrapper for lseek(2).
87  *
88  * \param fd The file descriptor whose offset is to be to repositioned.
89  * \param offset A value-result parameter.
90  * \param whence Usual repositioning directive.
91  *
92  * Reposition the offset of the file descriptor \a fd to the argument \a offset
93  * according to the directive \a whence. Upon successful return, \a offset
94  * contains the resulting offset location as measured in bytes from the
95  * beginning of the file.
96  *
97  * \return Positive on success. Otherwise, the function returns \p -E_OSL_LSEEK.
98  *
99  * \sa lseek(2).
100  */
101 static int __lseek(int fd, off_t *offset, int whence)
102 {
103         *offset = lseek(fd, *offset, whence);
104         int ret = -E_OSL_LSEEK;
105         if (*offset == -1)
106                 return ret;
107         return 1;
108 }
109
110 static int append_file(const char *filename, char *header, size_t header_size,
111         char *data, size_t data_size, uint32_t *new_pos)
112 {
113         int ret, fd;
114
115 //      DEBUG_LOG("appending %zu  + %zu bytes\n", header_size, data_size);
116         ret = osl_open(filename, O_WRONLY | O_CREAT | O_APPEND, 0644);
117         if (ret < 0)
118                 return ret;
119         fd = ret;
120         if (header && header_size) {
121                 ret = write_all(fd, header, &header_size);
122                 if (ret < 0)
123                         goto out;
124         }
125         ret = write_all(fd, data, &data_size);
126         if (ret < 0)
127                 goto out;
128         if (new_pos) {
129                 off_t offset = 0;
130                 ret = __lseek(fd, &offset, SEEK_END);
131                 if (ret < 0)
132                         goto out;
133 //              DEBUG_LOG("new file size: " FMT_OFF_T "\n", offset);
134                 *new_pos = offset;
135         }
136         ret = 1;
137 out:
138         close(fd);
139         return ret;
140 }
141
142 static int verify_name(const char *name)
143 {
144         if (!name)
145                 return -E_OSL_BAD_NAME;
146         if (!*name)
147                 return -E_OSL_BAD_NAME;
148         if (strchr(name, '/'))
149                 return -E_OSL_BAD_NAME;
150         if (!strcmp(name, ".."))
151                 return -E_OSL_BAD_NAME;
152         if (!strcmp(name, "."))
153                 return -E_OSL_BAD_NAME;
154         return 1;
155 }
156
157 int osl_hash_compare(const struct osl_object *obj1, const struct osl_object *obj2)
158 {
159         return hash_compare((HASH_TYPE *)obj1->data, (HASH_TYPE *)obj2->data);
160 }
161
162 static char *disk_storage_dirname(const struct osl_table *t, unsigned col_num,
163                 const char *ds_name)
164 {
165         char *dirname, *column_name = column_filename(t, col_num);
166
167         if (!column_name)
168                 return NULL;
169         if (!(t->desc->flags & OSL_LARGE_TABLE))
170                 return column_name;
171         dirname = make_message("%s/%.2s", column_name, ds_name);
172         free(column_name);
173         return dirname;
174 }
175
176 static char *disk_storage_name_of_object(const struct osl_table *t,
177         const struct osl_object *obj)
178 {
179         HASH_TYPE hash[HASH_SIZE];
180         hash_object(obj, hash);
181         return disk_storage_name_of_hash(t, hash);
182 }
183
184 static int disk_storage_name_of_row(const struct osl_table *t,
185                 const struct osl_row *row, char **name)
186 {
187         struct osl_object obj;
188         int ret = osl_get_object(t, row, t->disk_storage_name_column, &obj);
189
190         if (ret < 0)
191                 return ret;
192         *name = disk_storage_name_of_object(t, &obj);
193         if (*name)
194                 return 1;
195         return -ERRNO_TO_ERROR(ENOMEM);
196 }
197
198 static void column_name_hash(const char *col_name, HASH_TYPE *hash)
199 {
200         hash_function(col_name, strlen(col_name), hash);
201 }
202
203 static int init_column_descriptions(struct osl_table *t)
204 {
205         int i, j, ret;
206         const struct osl_column_description *cd;
207
208         ret = -E_OSL_BAD_TABLE_DESC;
209         ret = verify_name(t->desc->name);
210         if (ret < 0)
211                 goto err;
212         ret = -E_OSL_BAD_DB_DIR;
213         if (!t->desc->dir && (t->num_disk_storage_columns || t->num_mapped_columns))
214                 goto err;
215         /* the size of the index header without column descriptions */
216         t->index_header_size = IDX_COLUMN_DESCRIPTIONS;
217         FOR_EACH_COLUMN(i, t->desc, cd) {
218                 struct osl_column *col = t->columns + i;
219                 if (cd->storage_flags & OSL_RBTREE) {
220                         if (!cd->compare_function)
221                                 return -E_OSL_NO_COMPARE_FUNC;
222                 }
223                 if (cd->storage_type == OSL_NO_STORAGE)
224                         continue;
225                 ret = -E_OSL_NO_COLUMN_NAME;
226                 if (!cd->name || !cd->name[0])
227                         goto err;
228                 ret = verify_name(cd->name);
229                 if (ret < 0)
230                         goto err;
231                 t->index_header_size += index_column_description_size(cd->name);
232                 column_name_hash(cd->name, col->name_hash);
233                 ret = -E_OSL_DUPLICATE_COL_NAME;
234                 for (j = i + 1; j < t->desc->num_columns; j++) {
235                         const char *name2 = get_column_description(t->desc,
236                                 j)->name;
237                         if (cd->name && name2 && !strcmp(cd->name, name2))
238                                 goto err;
239                 }
240         }
241         return 1;
242 err:
243         return ret;
244 }
245
246 /**
247  * Initialize a struct table from given table description.
248  *
249  * \param desc The description of the osl table.
250  * \param table_ptr Result is returned here.
251  *
252  * This function performs several sanity checks on \p desc and returns if any
253  * of these tests fail. On success, a struct \p osl_table is allocated and
254  * initialized with data derived from \p desc.
255  *
256  * \return Standard.
257  *
258  * \sa struct osl_table.
259  */
260 int init_table_structure(const struct osl_table_description *desc,
261                 struct osl_table **table_ptr)
262 {
263         const struct osl_column_description *cd;
264         struct osl_table *t = calloc(1, sizeof(*t));
265         int i, ret = -ERRNO_TO_ERROR(ENOMEM), have_disk_storage_name_column = 0;
266
267         if (!t)
268                 return ret;
269         ret = -E_OSL_BAD_TABLE_DESC;
270         if (!desc)
271                 goto err;
272         DEBUG_LOG("creating table structure for '%s' from table "
273                 "description\n", desc->name);
274         ret = -E_OSL_NO_COLUMN_DESC;
275         if (!desc->column_descriptions)
276                 goto err;
277         ret = -E_OSL_NO_COLUMNS;
278         if (!desc->num_columns)
279                 goto err;
280         ret = -ERRNO_TO_ERROR(ENOMEM);
281         t->columns = calloc(desc->num_columns, sizeof(struct osl_column));
282         if (!t->columns)
283                 goto err;
284         t->desc = desc;
285         FOR_EACH_COLUMN(i, t->desc, cd) {
286                 enum osl_storage_type st = cd->storage_type;
287                 enum osl_storage_flags sf = cd->storage_flags;
288                 struct osl_column *col = &t->columns[i];
289
290                 ret = -E_OSL_BAD_STORAGE_TYPE;
291                 if (st != OSL_MAPPED_STORAGE && st != OSL_DISK_STORAGE
292                                 && st != OSL_NO_STORAGE)
293                         goto err;
294                 ret = -E_OSL_BAD_STORAGE_FLAGS;
295                 if (st == OSL_DISK_STORAGE && sf & OSL_RBTREE)
296                         goto err;
297                 ret = -E_OSL_BAD_STORAGE_SIZE;
298                 if (sf & OSL_FIXED_SIZE && !cd->data_size)
299                         goto err;
300                 switch (st) {
301                 case OSL_DISK_STORAGE:
302                         t->num_disk_storage_columns++;
303                         break;
304                 case OSL_MAPPED_STORAGE:
305                         t->num_mapped_columns++;
306                         col->index_offset = t->row_index_size;
307                         t->row_index_size += 8;
308                         break;
309                 case OSL_NO_STORAGE:
310                         col->volatile_num = t->num_volatile_columns;
311                         t->num_volatile_columns++;
312                         break;
313                 }
314                 if (sf & OSL_RBTREE) {
315                         col->rbtree_num = t->num_rbtrees;
316                         t->num_rbtrees++;
317                         if ((sf & OSL_UNIQUE) && (st == OSL_MAPPED_STORAGE)) {
318                                 if (!have_disk_storage_name_column)
319                                         t->disk_storage_name_column = i;
320                                 have_disk_storage_name_column = 1;
321                         }
322                 }
323         }
324         ret = -E_OSL_NO_UNIQUE_RBTREE_COLUMN;
325         if (t->num_disk_storage_columns && !have_disk_storage_name_column)
326                 goto err;
327         ret = -E_OSL_NO_RBTREE_COL;
328         if (!t->num_rbtrees)
329                 goto err;
330         /* success */
331         DEBUG_LOG("OK. Index entry size: %u\n", t->row_index_size);
332         ret = init_column_descriptions(t);
333         if (ret < 0)
334                 goto err;
335         *table_ptr = t;
336         return 1;
337 err:
338         free(t->columns);
339         free(t);
340         return ret;
341 }
342
343 /**
344  * Read the table description from index header.
345  *
346  * \param map The memory mapping of the index file.
347  * \param desc The values found in the index header are returned here.
348  *
349  * Read the index header, check for the paraslash magic string and the table version number.
350  * Read all information stored in the index header into \a desc.
351  *
352  * \return Standard.
353  *
354  * \sa struct osl_table_description, osl_create_table.
355  */
356 int read_table_desc(struct osl_object *map, struct osl_table_description *desc)
357 {
358         char *buf = map->data;
359         uint8_t version;
360         uint16_t header_size;
361         int ret, i;
362         unsigned offset;
363         struct osl_column_description *cd;
364
365         if (map->size < MIN_INDEX_HEADER_SIZE(1))
366                 return -E_OSL_SHORT_TABLE;
367         if (strncmp(buf + IDX_PARA_MAGIC, PARA_MAGIC, strlen(PARA_MAGIC)))
368                 return -E_OSL_NO_MAGIC;
369         version = read_u8(buf + IDX_VERSION);
370         if (version < MIN_TABLE_VERSION || version > MAX_TABLE_VERSION)
371                 return -E_OSL_VERSION_MISMATCH;
372         desc->flags = read_u8(buf + IDX_TABLE_FLAGS);
373         desc->num_columns = read_u16(buf + IDX_NUM_COLUMNS);
374         INFO_LOG("%u columns\n", desc->num_columns);
375         if (!desc->num_columns)
376                 return -E_OSL_NO_COLUMNS;
377         header_size = read_u16(buf + IDX_HEADER_SIZE);
378         if (map->size < header_size)
379                 return -E_OSL_BAD_SIZE;
380         desc->column_descriptions = calloc(desc->num_columns,
381                 sizeof(struct osl_column_description));
382         if (!desc->column_descriptions)
383                 return -ERRNO_TO_ERROR(ENOMEM);
384         offset = IDX_COLUMN_DESCRIPTIONS;
385         FOR_EACH_COLUMN(i, desc, cd) {
386                 char *null_byte;
387
388                 ret = -E_OSL_SHORT_TABLE;
389                 if (map->size < offset + MIN_IDX_COLUMN_DESCRIPTION_SIZE) {
390                         ERROR_LOG("map size = %zu < %u = offset + min desc size\n",
391                                 map->size, offset + MIN_IDX_COLUMN_DESCRIPTION_SIZE);
392                         goto err;
393                 }
394                 cd->storage_type = read_u16(buf + offset + IDX_CD_STORAGE_TYPE);
395                 cd->storage_flags = read_u16(buf + offset +
396                         IDX_CD_STORAGE_FLAGS);
397                 cd->data_size = read_u32(buf + offset + IDX_CD_DATA_SIZE);
398                 null_byte = memchr(buf + offset + IDX_CD_NAME, '\0',
399                         map->size - offset - IDX_CD_NAME);
400                 ret = -E_OSL_INDEX_CORRUPTION;
401                 if (!null_byte)
402                         goto err;
403                 ret = -ERRNO_TO_ERROR(ENOMEM);
404                 cd->name = strdup(buf + offset + IDX_CD_NAME);
405                 if (!cd->name)
406                         goto err;
407                 offset += index_column_description_size(cd->name);
408         }
409         if (offset != header_size) {
410                 ret = -E_OSL_INDEX_CORRUPTION;
411                 ERROR_LOG("real header size = %u != %u = stored header size\n",
412                         offset, header_size);
413                 goto err;
414         }
415         return 1;
416 err:
417         FOR_EACH_COLUMN(i, desc, cd)
418                 free(cd->name);
419         return ret;
420 }
421
422 /*
423  * check whether the table description given by \p t->desc matches the on-disk
424  * table structure stored in the index of \a t.
425  */
426 static int compare_table_descriptions(struct osl_table *t)
427 {
428         int i, ret;
429         struct osl_table_description desc;
430         const struct osl_column_description *cd1, *cd2;
431
432         /* read the on-disk structure into desc */
433         ret = read_table_desc(&t->index_map, &desc);
434         if (ret < 0)
435                 return ret;
436         ret = -E_OSL_BAD_TABLE_FLAGS;
437         if (desc.flags != t->desc->flags)
438                 goto out;
439         ret = -E_OSL_BAD_COLUMN_NUM;
440         if (desc.num_columns > t->desc->num_columns)
441                 goto out;
442         if (desc.num_columns < t->desc->num_columns) {
443                 struct osl_column_description *cd;
444                 unsigned diff = t->desc->num_columns - desc.num_columns;
445                 INFO_LOG("extending table by %u volatile columns\n", diff);
446                 ret = -ERRNO_TO_ERROR(ENOMEM);
447                 desc.column_descriptions = realloc(desc.column_descriptions,
448                         t->desc->num_columns * sizeof(struct osl_column_description));
449                 if (!desc.column_descriptions)
450                         goto out;
451                 for (i = desc.num_columns; i < t->desc->num_columns; i++) {
452                         cd = get_column_description(&desc, i);
453                         cd->storage_type = OSL_NO_STORAGE;
454                         cd->name = NULL;
455                 }
456                 desc.num_columns += diff;
457         }
458         FOR_EACH_COLUMN(i, t->desc, cd1) {
459                 cd2 = get_column_description(&desc, i);
460                 ret = -E_OSL_BAD_STORAGE_TYPE;
461                 if (cd1->storage_type != cd2->storage_type)
462                         goto out;
463                 if (cd1->storage_type == OSL_NO_STORAGE)
464                         continue;
465                 ret = -E_OSL_BAD_STORAGE_FLAGS;
466                 if (cd1->storage_flags != cd2->storage_flags) {
467                         ERROR_LOG("sf1 = %u != %u = sf2\n",
468                                 cd1->storage_flags, cd2->storage_flags);
469                         goto out;
470                 }
471                 ret = -E_OSL_BAD_DATA_SIZE;
472                 if (cd1->storage_flags & OSL_FIXED_SIZE)
473                         if (cd1->data_size != cd2->data_size)
474                                 goto out;
475                 ret = -E_OSL_BAD_COLUMN_NAME;
476                 if (strcmp(cd1->name, cd2->name))
477                         goto out;
478         }
479         INFO_LOG("table description of '%s' matches on-disk data, good\n",
480                 t->desc->name);
481         ret = 1;
482 out:
483         FOR_EACH_COLUMN(i, &desc, cd1)
484                 free(cd1->name);
485         free(desc.column_descriptions);
486         return ret;
487 }
488
489 static int create_table_index(struct osl_table *t)
490 {
491         char *buf, *filename;
492         int i, ret;
493         size_t size = t->index_header_size;
494         const struct osl_column_description *cd;
495         unsigned offset;
496
497         INFO_LOG("creating %zu byte index for table %s\n", size,
498                 t->desc->name);
499         buf = calloc(1, size);
500         if (!buf)
501                 return -ERRNO_TO_ERROR(ENOMEM);
502         sprintf(buf + IDX_PARA_MAGIC, "%s", PARA_MAGIC);
503         write_u8(buf + IDX_TABLE_FLAGS, t->desc->flags);
504         write_u8(buf + IDX_DIRTY_FLAG, 0);
505         write_u8(buf + IDX_VERSION, CURRENT_TABLE_VERSION);
506         write_u16(buf + IDX_NUM_COLUMNS, t->num_mapped_columns + t->num_disk_storage_columns);
507         write_u16(buf + IDX_HEADER_SIZE, t->index_header_size);
508         offset = IDX_COLUMN_DESCRIPTIONS;
509         FOR_EACH_COLUMN(i, t->desc, cd) {
510                 /* no need to store info about volatile storage */
511                 if (cd->storage_type == OSL_NO_STORAGE)
512                         continue;
513                 write_u16(buf + offset + IDX_CD_STORAGE_TYPE,
514                         cd->storage_type);
515                 write_u16(buf + offset + IDX_CD_STORAGE_FLAGS,
516                         cd->storage_flags);
517                 if (cd->storage_flags & OSL_FIXED_SIZE)
518                         write_u32(buf + offset + IDX_CD_DATA_SIZE,
519                                 cd->data_size);
520                 strcpy(buf + offset + IDX_CD_NAME, cd->name);
521                 offset += index_column_description_size(cd->name);
522         }
523         assert(offset = size);
524         filename = index_filename(t->desc);
525         if (filename)
526                 ret = write_file(filename, buf, size);
527         else
528                 ret = -ERRNO_TO_ERROR(ENOMEM);
529         free(buf);
530         free(filename);
531         return ret;
532 }
533
534 __export int osl_create_table(const struct osl_table_description *desc)
535 {
536         const struct osl_column_description *cd;
537         char *table_dir = NULL, *filename;
538         struct osl_table *t;
539         int i, ret = init_table_structure(desc, &t);
540
541         if (ret < 0)
542                 return ret;
543         INFO_LOG("creating %s\n", desc->name);
544         FOR_EACH_COLUMN(i, t->desc, cd) {
545                 if (cd->storage_type == OSL_NO_STORAGE)
546                         continue;
547                 if (!table_dir) {
548                         ret = osl_mkdir(desc->dir, 0777);
549                         if (ret < 0 && !is_errno(-ret, EEXIST))
550                                 goto out;
551                         table_dir = make_message("%s/%s", desc->dir,
552                                 desc->name);
553                         ret = -ERRNO_TO_ERROR(ENOMEM);
554                         if (!table_dir)
555                                 goto out;
556                         ret = osl_mkdir(table_dir, 0777);
557                         if (ret < 0)
558                                 goto out;
559                 }
560                 ret = -ERRNO_TO_ERROR(ENOMEM);
561                 filename = column_filename(t, i);
562                 if (!filename)
563                         goto out;
564                 INFO_LOG("filename: %s\n", filename);
565                 if (cd->storage_type == OSL_MAPPED_STORAGE) {
566                         ret = osl_open(filename, O_RDWR | O_CREAT | O_EXCL,
567                                 0644);
568                         free(filename);
569                         if (ret < 0)
570                                 goto out;
571                         close(ret);
572                         continue;
573                 }
574                 /* DISK STORAGE */
575                 ret = osl_mkdir(filename, 0777);
576                 free(filename);
577                 if (ret < 0)
578                         goto out;
579         }
580         if (t->num_mapped_columns) {
581                 ret = create_table_index(t);
582                 if (ret < 0)
583                         goto out;
584         }
585         ret = 1;
586 out:
587         free(table_dir);
588         free(t->columns);
589         free(t);
590         return ret;
591 }
592
593 static int table_is_dirty(struct osl_table *t)
594 {
595         char *buf = (char *)t->index_map.data + IDX_DIRTY_FLAG;
596         uint8_t dirty = read_u8(buf) & 0x1;
597         return !!dirty;
598 }
599
600 static void mark_table_dirty(struct osl_table *t)
601 {
602         char *buf = (char *)t->index_map.data + IDX_DIRTY_FLAG;
603         write_u8(buf, read_u8(buf) | 1);
604 }
605
606 static void mark_table_clean(struct osl_table *t)
607 {
608         char *buf = (char *)t->index_map.data + IDX_DIRTY_FLAG;
609         write_u8(buf, read_u8(buf) & 0xfe);
610 }
611
612 static void unmap_column(struct osl_table *t, unsigned col_num)
613 {
614         struct osl_object map = t->columns[col_num].data_map;
615         int ret;
616         if (!map.data)
617                 return;
618         ret = osl_munmap(map.data, map.size);
619         assert(ret > 0);
620         map.data = NULL;
621 }
622
623 /**
624  * Unmap all mapped files of an osl table.
625  *
626  * \param t Pointer to a mapped table.
627  * \param flags Options for unmapping.
628  *
629  * \return Positive on success, negative on errors.
630  *
631  * \sa map_table(), enum osl_close_flags, osl_munmap().
632  */
633 int unmap_table(struct osl_table *t, enum osl_close_flags flags)
634 {
635         unsigned i;
636         const struct osl_column_description *cd;
637         int ret;
638
639         if (!t->num_mapped_columns) /* can this ever happen? */
640                 return 1;
641         INFO_LOG("unmapping table '%s'\n", t->desc->name);
642         if (!t->index_map.data)
643                 return -E_OSL_NOT_MAPPED;
644         if (flags & OSL_MARK_CLEAN)
645                 mark_table_clean(t);
646         ret = osl_munmap(t->index_map.data, t->index_map.size);
647         if (ret < 0)
648                 return ret;
649         t->index_map.data = NULL;
650         if (!t->num_rows)
651                 return 1;
652         FOR_EACH_MAPPED_COLUMN(i, t, cd)
653                 unmap_column(t, i);
654         return 1;
655 }
656
657 static int map_column(struct osl_table *t, unsigned col_num)
658 {
659         struct stat statbuf;
660         char *filename = column_filename(t, col_num);
661         int ret;
662
663         if (!filename)
664                 return -ERRNO_TO_ERROR(ENOMEM);
665         ret = osl_stat(filename, &statbuf);
666         if (ret < 0) {
667                 free(filename);
668                 return ret;
669         }
670         if (!(S_IFREG & statbuf.st_mode)) {
671                 free(filename);
672                 return ret;
673         }
674         ret = mmap_full_file(filename, O_RDWR,
675                 &t->columns[col_num].data_map.data,
676                 &t->columns[col_num].data_map.size,
677                 NULL);
678         free(filename);
679         return ret;
680 }
681
682 /**
683  * Map the index file and all columns of type \p OSL_MAPPED_STORAGE into memory.
684  *
685  * \param t Pointer to an initialized table structure.
686  * \param flags Mapping options.
687  *
688  * \return Negative return value on errors; on success the number of rows
689  * (including invalid rows) is returned.
690  *
691  * \sa unmap_table(), enum map_table_flags, osl_open_table(), mmap(2).
692  */
693 int map_table(struct osl_table *t, enum map_table_flags flags)
694 {
695         char *filename;
696         const struct osl_column_description *cd;
697         int i = 0, ret, num_rows = 0;
698
699         if (!t->num_mapped_columns)
700                 return 0;
701         if (t->index_map.data)
702                 return -E_OSL_ALREADY_MAPPED;
703         filename = index_filename(t->desc);
704         if (!filename)
705                 return -ERRNO_TO_ERROR(ENOMEM);
706         INFO_LOG("mapping table '%s' (index: %s)\n", t->desc->name, filename);
707         ret = mmap_full_file(filename, flags & MAP_TBL_FL_MAP_RDONLY?
708                 O_RDONLY : O_RDWR, &t->index_map.data, &t->index_map.size, NULL);
709         free(filename);
710         if (ret < 0)
711                 return ret;
712         if (flags & MAP_TBL_FL_VERIFY_INDEX) {
713                 ret = compare_table_descriptions(t);
714                 if (ret < 0)
715                         goto err;
716         }
717         ret = -E_OSL_BUSY;
718         if (!(flags & MAP_TBL_FL_IGNORE_DIRTY)) {
719                 if (table_is_dirty(t)) {
720                         ERROR_LOG("%s is dirty\n", t->desc->name);
721                         goto err;
722                 }
723         }
724         mark_table_dirty(t);
725         num_rows = table_num_rows(t);
726         if (!num_rows)
727                 return num_rows;
728         /* map data files */
729         FOR_EACH_MAPPED_COLUMN(i, t, cd) {
730                 ret = map_column(t, i);
731                 if (ret < 0)
732                         goto err;
733         }
734         return num_rows;
735 err:    /* unmap what is already mapped */
736         for (i--; i >= 0; i--) {
737                 struct osl_object map = t->columns[i].data_map;
738                 osl_munmap(map.data, map.size);
739                 map.data = NULL;
740         }
741         osl_munmap(t->index_map.data, t->index_map.size);
742         t->index_map.data = NULL;
743         return ret;
744 }
745
746 /**
747  * Retrieve a mapped object by row and column number.
748  *
749  * \param t Pointer to an open osl table.
750  * \param col_num Number of the mapped column containing the object to retrieve.
751  * \param row_num Number of the row containing the object to retrieve.
752  * \param obj The result is returned here.
753  *
754  * It is considered an error if \a col_num does not refer to a column
755  * of storage type \p OSL_MAPPED_STORAGE.
756  *
757  * \return Standard.
758  *
759  * \sa osl_storage_type.
760  */
761 int get_mapped_object(const struct osl_table *t, unsigned col_num,
762         uint32_t row_num, struct osl_object *obj)
763 {
764         struct osl_column *col = &t->columns[col_num];
765         uint32_t offset;
766         char *header;
767         char *cell_index;
768         int ret;
769
770         if (t->num_rows <= row_num)
771                 return -E_OSL_BAD_ROW_NUM;
772         ret = get_cell_index(t, row_num, col_num, &cell_index);
773         if (ret < 0)
774                 return ret;
775         offset = read_u32(cell_index);
776         obj->size = read_u32(cell_index + 4) - 1;
777         header = col->data_map.data + offset;
778         obj->data = header + 1;
779         if (read_u8(header) == 0xff) {
780                 ERROR_LOG("col %u, size %zu, offset %u\n", col_num,
781                         obj->size, offset);
782                 return -E_OSL_INVALID_OBJECT;
783         }
784         return 1;
785 }
786
787 static int search_rbtree(const struct osl_object *obj,
788                 const struct osl_table *t, unsigned col_num,
789                 struct rb_node **result, struct rb_node ***rb_link)
790 {
791         struct osl_column *col = &t->columns[col_num];
792         struct rb_node **new = &col->rbtree.rb_node, *parent = NULL;
793         const struct osl_column_description *cd =
794                 get_column_description(t->desc, col_num);
795         enum osl_storage_type st = cd->storage_type;
796         while (*new) {
797                 struct osl_row *this_row = get_row_pointer(*new,
798                         col->rbtree_num);
799                 int ret;
800                 struct osl_object this_obj;
801                 parent = *new;
802                 if (st == OSL_MAPPED_STORAGE) {
803                         ret = get_mapped_object(t, col_num, this_row->num,
804                                 &this_obj);
805                         if (ret < 0)
806                                 return ret;
807                 } else
808                         this_obj = this_row->volatile_objects[col->volatile_num];
809                 ret = cd->compare_function(obj, &this_obj);
810                 if (!ret) {
811                         if (result)
812                                 *result = get_rb_node_pointer(this_row,
813                                         col->rbtree_num);
814                         return 1;
815                 }
816                 if (ret < 0)
817                         new = &((*new)->rb_left);
818                 else
819                         new = &((*new)->rb_right);
820         }
821         if (result)
822                 *result = parent;
823         if (rb_link)
824                 *rb_link = new;
825         return -E_OSL_RB_KEY_NOT_FOUND;
826 }
827
828 static int insert_rbtree(struct osl_table *t, unsigned col_num,
829         const struct osl_row *row, const struct osl_object *obj)
830 {
831         struct rb_node *parent, **rb_link;
832         unsigned rbtree_num;
833         struct rb_node *n;
834         int ret = search_rbtree(obj, t, col_num, &parent, &rb_link);
835
836         if (ret > 0)
837                 return -E_OSL_RB_KEY_EXISTS;
838         rbtree_num = t->columns[col_num].rbtree_num;
839         n = get_rb_node_pointer(row, rbtree_num);
840         rb_link_node(n, parent, rb_link);
841         rb_insert_color(n, &t->columns[col_num].rbtree);
842         return 1;
843 }
844
845 static void remove_rb_node(struct osl_table *t, unsigned col_num,
846                 const struct osl_row *row)
847 {
848         struct osl_column *col = &t->columns[col_num];
849         const struct osl_column_description *cd =
850                 get_column_description(t->desc, col_num);
851         enum osl_storage_flags sf = cd->storage_flags;
852         struct rb_node *victim, *splice_out_node, *tmp;
853         if (!(sf & OSL_RBTREE))
854                 return;
855         /*
856          * Which node is removed/spliced out actually depends on how many
857          * children the victim node has: If it has no children, it gets
858          * deleted. If it has one child, it gets spliced out. If it has two
859          * children, its successor (which has at most a right child) gets
860          * spliced out.
861          */
862         victim = get_rb_node_pointer(row, col->rbtree_num);
863         if (victim->rb_left && victim->rb_right)
864                 splice_out_node = rb_next(victim);
865         else
866                 splice_out_node = victim;
867         /* Go up to the root and decrement the size of each node in the path. */
868         for (tmp = splice_out_node; tmp; tmp = rb_parent(tmp))
869                 tmp->size--;
870         rb_erase(victim, &col->rbtree);
871 }
872
873 static int add_row_to_rbtrees(struct osl_table *t, uint32_t row_num,
874                 struct osl_object *volatile_objs, struct osl_row **row_ptr)
875 {
876         unsigned i;
877         int ret;
878         struct osl_row *row = allocate_row(t->num_rbtrees);
879         const struct osl_column_description *cd;
880
881         if (!row)
882                 return -ERRNO_TO_ERROR(ENOMEM);
883         row->num = row_num;
884         row->volatile_objects = volatile_objs;
885         FOR_EACH_RBTREE_COLUMN(i, t, cd) {
886                 if (cd->storage_type == OSL_MAPPED_STORAGE) {
887                         struct osl_object obj;
888                         ret = get_mapped_object(t, i, row_num, &obj);
889                         if (ret < 0)
890                                 goto err;
891                         ret = insert_rbtree(t, i, row, &obj);
892                 } else { /* volatile */
893                         const struct osl_object *obj
894                                 = volatile_objs + t->columns[i].volatile_num;
895                         ret = insert_rbtree(t, i, row, obj);
896                 }
897                 if (ret < 0)
898                         goto err;
899         }
900         if (row_ptr)
901                 *row_ptr = row;
902         return 1;
903 err: /* rollback changes, i.e. remove added entries from rbtrees */
904         while (i)
905                 remove_rb_node(t, i--, row);
906         free(row);
907         return ret;
908 }
909
910 static void free_volatile_objects(const struct osl_table *t,
911                 enum osl_close_flags flags)
912 {
913         int i, j;
914         struct rb_node *n;
915         struct osl_column *rb_col;
916         const struct osl_column_description *cd;
917
918         if (!t->num_volatile_columns)
919                 return;
920         /* find the first rbtree column (any will do) */
921         FOR_EACH_RBTREE_COLUMN(i, t, cd)
922                 break;
923         rb_col = t->columns + i;
924         /* walk that rbtree and free all volatile objects */
925         for (n = rb_first(&rb_col->rbtree); n; n = rb_next(n)) {
926                 struct osl_row *r = get_row_pointer(n, rb_col->rbtree_num);
927                 if (flags & OSL_FREE_VOLATILE)
928                         FOR_EACH_VOLATILE_COLUMN(j, t, cd) {
929                                 if (cd->storage_flags & OSL_DONT_FREE)
930                                         continue;
931                                 free(r->volatile_objects[
932                                         t->columns[j].volatile_num].data);
933                         }
934 //                      for (j = 0; j < t->num_volatile_columns; j++)
935 //                              free(r->volatile_objects[j].data);
936                 free(r->volatile_objects);
937         }
938 }
939
940 /**
941  * Erase all rbtree nodes and free resources.
942  *
943  * \param t Pointer to an open osl table.
944  *
945  * This function is called by osl_close_table().
946  */
947 void clear_rbtrees(struct osl_table *t)
948 {
949         const struct osl_column_description *cd;
950         unsigned i, rbtrees_cleared = 0;
951
952         FOR_EACH_RBTREE_COLUMN(i, t, cd) {
953                 struct osl_column *col = &t->columns[i];
954                 struct rb_node *n;
955                 rbtrees_cleared++;
956                 for (n = rb_first(&col->rbtree); n;) {
957                         struct osl_row *r;
958                         rb_erase(n, &col->rbtree);
959                         if (rbtrees_cleared == t->num_rbtrees) {
960                                 r = get_row_pointer(n, col->rbtree_num);
961                                 n = rb_next(n);
962                                 free(r);
963                         } else
964                                 n = rb_next(n);
965                 }
966         }
967
968 }
969
970 __export int osl_close_table(struct osl_table *t, enum osl_close_flags flags)
971 {
972         int ret;
973
974         if (!t)
975                 return -E_OSL_BAD_TABLE;
976         NOTICE_LOG("closing table %s\n", t->desc->name);
977         free_volatile_objects(t, flags);
978         clear_rbtrees(t);
979         ret = unmap_table(t, flags);
980         if (ret < 0)
981                 ERROR_LOG("unmap_table failed: %d\n", ret);
982         free(t->columns);
983         free(t);
984         return ret;
985 }
986
987 /**
988  * Find out whether the given row number corresponds to an invalid row.
989  *
990  * \param t Pointer to the osl table.
991  * \param row_num The number of the row in question.
992  *
993  * By definition, a row is considered invalid if all its index entries
994  * are invalid.
995  *
996  * \return Positive if \a row_num corresponds to an invalid row,
997  * zero if it corresponds to a valid row, negative on errors.
998  */
999 int row_is_invalid(struct osl_table *t, uint32_t row_num)
1000 {
1001         char *row_index;
1002         int i, ret = get_row_index(t, row_num, &row_index);
1003
1004         if (ret < 0)
1005                 return ret;
1006         for (i = 0; i < t->row_index_size; i++) {
1007                 if ((unsigned char)row_index[i] != 0xff)
1008                         return 0;
1009         }
1010         INFO_LOG("row %d is invalid\n", row_num);
1011         return 1;
1012 }
1013
1014 /**
1015  * Invalidate a row of an osl table.
1016  *
1017  * \param t Pointer to an open osl table.
1018  * \param row_num Number of the row to mark as invalid.
1019  *
1020  * This function marks each mapped object in the index entry of \a row as
1021  * invalid.
1022  *
1023  * \return Standard.
1024  */
1025 int mark_row_invalid(struct osl_table *t, uint32_t row_num)
1026 {
1027         char *row_index;
1028         int ret = get_row_index(t, row_num, &row_index);
1029
1030         if (ret < 0)
1031                 return ret;
1032         INFO_LOG("marking row %d as invalid\n", row_num);
1033         memset(row_index, 0xff, t->row_index_size);
1034         return 1;
1035 }
1036
1037 /**
1038  * Initialize all rbtrees and compute number of invalid rows.
1039  *
1040  * \param t The table containing the rbtrees to be initialized.
1041  *
1042  * \return Standard.
1043  */
1044 int init_rbtrees(struct osl_table *t)
1045 {
1046         int i, ret;
1047         const struct osl_column_description *cd;
1048
1049         /* create rbtrees */
1050         FOR_EACH_RBTREE_COLUMN(i, t, cd)
1051                 t->columns[i].rbtree = RB_ROOT;
1052         /* add valid rows to rbtrees */
1053         t->num_invalid_rows = 0;
1054         for (i = 0; i < t->num_rows; i++) {
1055                 ret = row_is_invalid(t, i);
1056                 if (ret < 0)
1057                         return ret;
1058                 if (ret) {
1059                         t->num_invalid_rows++;
1060                         continue;
1061                 }
1062                 ret = add_row_to_rbtrees(t, i, NULL, NULL);
1063                 if (ret < 0)
1064                         return ret;
1065         }
1066         return 1;
1067 }
1068
1069 __export int osl_open_table(const struct osl_table_description *table_desc,
1070                 struct osl_table **result)
1071 {
1072         int i, ret;
1073         struct osl_table *t;
1074         const struct osl_column_description *cd;
1075
1076         NOTICE_LOG("opening table %s\n", table_desc->name);
1077         ret = init_table_structure(table_desc, &t);
1078         if (ret < 0)
1079                 return ret;
1080         FOR_EACH_DISK_STORAGE_COLUMN(i, t, cd) {
1081                 struct stat statbuf;
1082                 char *dirname = column_filename(t, i);
1083
1084                 ret = -ERRNO_TO_ERROR(ENOMEM);
1085                 if (!dirname)
1086                         goto err;
1087                 /* check if directory exists */
1088                 ret = stat(dirname, &statbuf);
1089                 free(dirname);
1090                 if (ret < 0) {
1091                         ret = -ERRNO_TO_ERROR(errno);
1092                         goto err;
1093                 }
1094                 ret = -ERRNO_TO_ERROR(ENOTDIR);
1095                 if (!S_ISDIR(statbuf.st_mode))
1096                         goto err;
1097         }
1098         ret = map_table(t, MAP_TBL_FL_VERIFY_INDEX);
1099         if (ret < 0)
1100                 goto err;
1101         t->num_rows = ret;
1102         DEBUG_LOG("num rows: %d\n", t->num_rows);
1103         ret = init_rbtrees(t);
1104         if (ret < 0) {
1105                 osl_close_table(t, OSL_MARK_CLEAN); /* ignore further errors */
1106                 return ret;
1107         }
1108         *result = t;
1109         return 1;
1110 err:
1111         free(t->columns);
1112         free(t);
1113         return ret;
1114 }
1115
1116 static int create_disk_storage_object_dir(const struct osl_table *t,
1117                 unsigned col_num, const char *ds_name)
1118 {
1119         char *dirname;
1120         int ret;
1121
1122         if (!(t->desc->flags & OSL_LARGE_TABLE))
1123                 return 1;
1124         dirname = disk_storage_dirname(t, col_num, ds_name);
1125         if (!dirname)
1126                 return -ERRNO_TO_ERROR(ENOMEM);
1127         ret = osl_mkdir(dirname, 0777);
1128         free(dirname);
1129         if (ret < 0 && !is_errno(-ret, EEXIST))
1130                 return ret;
1131         return 1;
1132 }
1133
1134 static int write_disk_storage_file(const struct osl_table *t, unsigned col_num,
1135         const struct osl_object *obj, const char *ds_name)
1136 {
1137         int ret;
1138         char *filename;
1139
1140         ret = create_disk_storage_object_dir(t, col_num, ds_name);
1141         if (ret < 0)
1142                 return ret;
1143         filename = disk_storage_path(t, col_num, ds_name);
1144         if (!filename)
1145                 return -ERRNO_TO_ERROR(ENOMEM);
1146         ret = write_file(filename, obj->data, obj->size);
1147         free(filename);
1148         return ret;
1149 }
1150
1151 static int append_map_file(const struct osl_table *t, unsigned col_num,
1152         const struct osl_object *obj, uint32_t *new_size)
1153 {
1154         char *filename = column_filename(t, col_num);
1155         int ret;
1156         char header = 0; /* zero means valid object */
1157
1158         if (!filename)
1159                 return -ERRNO_TO_ERROR(ENOMEM);
1160         ret = append_file(filename, &header, 1, obj->data, obj->size,
1161                 new_size);
1162         free(filename);
1163         return ret;
1164 }
1165
1166 static int append_row_index(const struct osl_table *t, char *row_index)
1167 {
1168         char *filename;
1169         int ret;
1170
1171         if (!t->num_mapped_columns)
1172                 return 1;
1173         filename = index_filename(t->desc);
1174         if (!filename)
1175                 return -ERRNO_TO_ERROR(ENOMEM);
1176         ret = append_file(filename, NULL, 0, row_index,
1177                 t->row_index_size, NULL);
1178         free(filename);
1179         return ret;
1180 }
1181
1182 static int truncate_mapped_file(const struct osl_table *t, unsigned col_num,
1183                 off_t size)
1184 {
1185         int ret;
1186         char *filename = column_filename(t, col_num);
1187
1188         if (!filename)
1189                 return -ERRNO_TO_ERROR(ENOMEM);
1190         ret = truncate_file(filename, size);
1191         free(filename);
1192         return ret;
1193 }
1194
1195 static int delete_disk_storage_file(const struct osl_table *t, unsigned col_num,
1196                 const char *ds_name)
1197 {
1198         char *dirname, *filename = disk_storage_path(t, col_num, ds_name);
1199         int ret, err;
1200
1201         if (!filename)
1202                 return -ERRNO_TO_ERROR(ENOMEM);
1203         ret = unlink(filename);
1204         err = errno;
1205         free(filename);
1206         if (ret < 0)
1207                 return -ERRNO_TO_ERROR(err);
1208         if (!(t->desc->flags & OSL_LARGE_TABLE))
1209                 return 1;
1210         dirname = disk_storage_dirname(t, col_num, ds_name);
1211         if (!dirname)
1212                 return -ERRNO_TO_ERROR(ENOMEM);
1213         rmdir(dirname);
1214         free(dirname);
1215         return 1;
1216 }
1217
1218 __export int osl_add_and_get_row(struct osl_table *t, struct osl_object *objects,
1219                 struct osl_row **row)
1220 {
1221         int i, ret;
1222         char *ds_name = NULL;
1223         struct rb_node **rb_parents = NULL, ***rb_links = NULL;
1224         char *new_row_index = NULL;
1225         struct osl_object *volatile_objs = NULL;
1226         const struct osl_column_description *cd;
1227
1228         if (!t)
1229                 return -E_OSL_BAD_TABLE;
1230         rb_parents = malloc(t->num_rbtrees * sizeof(struct rn_node*));
1231         if (!rb_parents)
1232                 return -ERRNO_TO_ERROR(ENOMEM);
1233         rb_links = malloc(t->num_rbtrees * sizeof(struct rn_node**));
1234         if (!rb_links) {
1235                 free(rb_parents);
1236                 return -ERRNO_TO_ERROR(ENOMEM);
1237         }
1238         if (t->num_mapped_columns) {
1239                 new_row_index = malloc(t->row_index_size);
1240                 if (!new_row_index) {
1241                         free(rb_links);
1242                         free(rb_parents);
1243                         return -ERRNO_TO_ERROR(ENOMEM);
1244                 }
1245         }
1246         /* pass 1: sanity checks */
1247 //      DEBUG_LOG("sanity tests: %p:%p\n", objects[0].data,
1248 //              objects[1].data);
1249         FOR_EACH_COLUMN(i, t->desc, cd) {
1250                 enum osl_storage_type st = cd->storage_type;
1251                 enum osl_storage_flags sf = cd->storage_flags;
1252
1253 //              ret = -E_OSL_NULL_OBJECT;
1254 //              if (!objects[i])
1255 //                      goto out;
1256                 if (st == OSL_DISK_STORAGE)
1257                         continue;
1258                 if (sf & OSL_RBTREE) {
1259                         unsigned rbtree_num = t->columns[i].rbtree_num;
1260                         ret = -E_OSL_RB_KEY_EXISTS;
1261 //                      DEBUG_LOG("checking whether %p exists\n",
1262 //                              objects[i].data);
1263                         if (search_rbtree(objects + i, t, i,
1264                                         &rb_parents[rbtree_num],
1265                                         &rb_links[rbtree_num]) > 0)
1266                                 goto out;
1267                 }
1268                 if (sf & OSL_FIXED_SIZE) {
1269 //                      DEBUG_LOG("fixed size. need: %zu, have: %d\n",
1270 //                              objects[i].size, cd->data_size);
1271                         ret = -E_OSL_BAD_DATA_SIZE;
1272                         if (objects[i].size != cd->data_size)
1273                                 goto out;
1274                 }
1275         }
1276         if (t->num_disk_storage_columns) {
1277                 ds_name = disk_storage_name_of_object(t,
1278                         &objects[t->disk_storage_name_column]);
1279                 ret = -ERRNO_TO_ERROR(ENOMEM);
1280                 if (!ds_name)
1281                         goto out;
1282         }
1283         ret = unmap_table(t, OSL_MARK_CLEAN);
1284         if (ret < 0)
1285                 goto out;
1286 //      DEBUG_LOG("sanity tests passed%s\n", "");
1287         /* pass 2: create data files, append map data */
1288         FOR_EACH_COLUMN(i, t->desc, cd) {
1289                 enum osl_storage_type st = cd->storage_type;
1290                 if (st == OSL_NO_STORAGE)
1291                         continue;
1292                 if (st == OSL_MAPPED_STORAGE) {
1293                         uint32_t new_size;
1294                         struct osl_column *col = &t->columns[i];
1295 //                      DEBUG_LOG("appending object of size %zu\n",
1296 //                              objects[i].size);
1297                         ret = append_map_file(t, i, objects + i, &new_size);
1298                         if (ret < 0)
1299                                 goto rollback;
1300                         update_cell_index(new_row_index, col, new_size,
1301                                 objects[i].size);
1302                         continue;
1303                 }
1304                 /* DISK_STORAGE */
1305                 ret = write_disk_storage_file(t, i, objects + i, ds_name);
1306                 if (ret < 0)
1307                         goto rollback;
1308         }
1309         ret = append_row_index(t, new_row_index);
1310         if (ret < 0)
1311                 goto rollback;
1312         ret = map_table(t, MAP_TBL_FL_VERIFY_INDEX);
1313         if (ret < 0) { /* truncate index and rollback changes */
1314                 char *filename = index_filename(t->desc);
1315                 if (filename)
1316                         truncate_file(filename, t->row_index_size);
1317                 free(filename);
1318                 goto rollback;
1319         }
1320         /* pass 3: add entry to rbtrees */
1321         if (t->num_volatile_columns) {
1322                 ret = -ERRNO_TO_ERROR(ENOMEM);
1323                 volatile_objs = calloc(t->num_volatile_columns,
1324                         sizeof(struct osl_object));
1325                 if (!volatile_objs)
1326                         goto out;
1327                 FOR_EACH_VOLATILE_COLUMN(i, t, cd)
1328                         volatile_objs[t->columns[i].volatile_num] = objects[i];
1329         }
1330         t->num_rows++;
1331 //      DEBUG_LOG("adding new entry as row #%d\n", t->num_rows - 1);
1332         ret = add_row_to_rbtrees(t, t->num_rows - 1, volatile_objs, row);
1333         if (ret < 0)
1334                 goto out;
1335 //      DEBUG_LOG("added new entry as row #%d\n", t->num_rows - 1);
1336         ret = 1;
1337         goto out;
1338 rollback: /* rollback all changes made, ignore further errors */
1339         for (i--; i >= 0; i--) {
1340                 cd = get_column_description(t->desc, i);
1341                 enum osl_storage_type st = cd->storage_type;
1342                 if (st == OSL_NO_STORAGE)
1343                         continue;
1344
1345                 if (st == OSL_MAPPED_STORAGE)
1346                         truncate_mapped_file(t, i, objects[i].size);
1347                 else /* disk storage */
1348                         delete_disk_storage_file(t, i, ds_name);
1349         }
1350         /* ignore error and return previous error value */
1351         map_table(t, MAP_TBL_FL_VERIFY_INDEX);
1352 out:
1353         free(new_row_index);
1354         free(ds_name);
1355         free(rb_parents);
1356         free(rb_links);
1357         return ret;
1358 }
1359
1360 __export int osl_add_row(struct osl_table *t, struct osl_object *objects)
1361 {
1362         return osl_add_and_get_row(t, objects, NULL);
1363 }
1364
1365 __export int osl_get_object(const struct osl_table *t, const struct osl_row *r,
1366         unsigned col_num, struct osl_object *object)
1367 {
1368         const struct osl_column_description *cd;
1369
1370         if (!t)
1371                 return -E_OSL_BAD_TABLE;
1372         cd = get_column_description(t->desc, col_num);
1373         /* col must not be disk storage */
1374         if (cd->storage_type == OSL_DISK_STORAGE)
1375                 return -E_OSL_BAD_STORAGE_TYPE;
1376         if (cd->storage_type == OSL_MAPPED_STORAGE)
1377                 return get_mapped_object(t, col_num, r->num, object);
1378         /* volatile */
1379         *object = r->volatile_objects[t->columns[col_num].volatile_num];
1380         return 1;
1381 }
1382
1383 static int mark_mapped_object_invalid(const struct osl_table *t,
1384                 uint32_t row_num, unsigned col_num)
1385 {
1386         struct osl_object obj;
1387         char *p;
1388         int ret = get_mapped_object(t, col_num, row_num, &obj);
1389
1390         if (ret < 0)
1391                 return ret;
1392         p = obj.data;
1393         p--;
1394         *p = 0xff;
1395         return 1;
1396 }
1397
1398 __export int osl_del_row(struct osl_table *t, struct osl_row *row)
1399 {
1400         struct osl_row *r = row;
1401         int i, ret;
1402         const struct osl_column_description *cd;
1403
1404         if (!t)
1405                 return -E_OSL_BAD_TABLE;
1406         INFO_LOG("deleting row %p\n", row);
1407
1408         if (t->num_disk_storage_columns) {
1409                 char *ds_name;
1410                 ret = disk_storage_name_of_row(t, r, &ds_name);
1411                 if (ret < 0)
1412                         goto out;
1413                 FOR_EACH_DISK_STORAGE_COLUMN(i, t, cd)
1414                         delete_disk_storage_file(t, i, ds_name);
1415                 free(ds_name);
1416         }
1417         FOR_EACH_COLUMN(i, t->desc, cd) {
1418                 struct osl_column *col = t->columns + i;
1419                 enum osl_storage_type st = cd->storage_type;
1420                 remove_rb_node(t, i, r);
1421                 if (st == OSL_MAPPED_STORAGE) {
1422                         mark_mapped_object_invalid(t, r->num, i);
1423                         continue;
1424                 }
1425                 if (st == OSL_NO_STORAGE && !(cd->storage_flags & OSL_DONT_FREE))
1426                         free(r->volatile_objects[col->volatile_num].data);
1427         }
1428         if (t->num_mapped_columns) {
1429                 ret = mark_row_invalid(t, r->num);
1430                 if (ret < 0)
1431                         goto out;
1432                 t->num_invalid_rows++;
1433         } else
1434                 t->num_rows--;
1435         ret = 1;
1436 out:
1437         free(r->volatile_objects);
1438         free(r);
1439         return ret;
1440 }
1441
1442 /* test if column has an rbtree */
1443 static int check_rbtree_col(const struct osl_table *t, unsigned col_num,
1444                 struct osl_column **col)
1445 {
1446         if (!t)
1447                 return -E_OSL_BAD_TABLE;
1448         if (!(get_column_description(t->desc, col_num)->storage_flags & OSL_RBTREE))
1449                 return -E_OSL_BAD_STORAGE_FLAGS;
1450         *col = t->columns + col_num;
1451         return 1;
1452 }
1453
1454 __export int osl_get_row(const struct osl_table *t, unsigned col_num,
1455                 const struct osl_object *obj, struct osl_row **result)
1456 {
1457         int ret;
1458         struct rb_node *node;
1459         struct osl_row *row;
1460         struct osl_column *col;
1461
1462         *result = NULL;
1463         ret = check_rbtree_col(t, col_num, &col);
1464         if (ret < 0)
1465                 return ret;
1466         ret = search_rbtree(obj, t, col_num, &node, NULL);
1467         if (ret < 0)
1468                 return ret;
1469         row = get_row_pointer(node, t->columns[col_num].rbtree_num);
1470         *result = row;
1471         return 1;
1472 }
1473
1474 static int rbtree_loop(struct osl_column *col, void *private_data,
1475                 osl_rbtree_loop_func *func)
1476 {
1477         struct rb_node *n, *tmp;
1478
1479         /* this for-loop is safe against removal of an entry */
1480         for (n = rb_first(&col->rbtree), tmp = n? rb_next(n) : NULL;
1481                         n;
1482                         n = tmp, tmp = tmp? rb_next(tmp) : NULL) {
1483                 struct osl_row *r = get_row_pointer(n, col->rbtree_num);
1484                 if (func(r, private_data) < 0)
1485                         return -E_OSL_LOOP;
1486         }
1487         return 1;
1488 }
1489
1490 static int rbtree_loop_reverse(struct osl_column *col, void *private_data,
1491                 osl_rbtree_loop_func *func)
1492 {
1493         struct rb_node *n, *tmp;
1494
1495         /* safe against removal of an entry */
1496         for (n = rb_last(&col->rbtree), tmp = n? rb_prev(n) : NULL;
1497                         n;
1498                         n = tmp, tmp = tmp? rb_prev(tmp) : NULL) {
1499                 struct osl_row *r = get_row_pointer(n, col->rbtree_num);
1500                 if (func(r, private_data) < 0)
1501                         return -E_OSL_LOOP;
1502         }
1503         return 1;
1504 }
1505
1506 __export int osl_rbtree_loop(const struct osl_table *t, unsigned col_num,
1507         void *private_data, osl_rbtree_loop_func *func)
1508 {
1509         struct osl_column *col;
1510
1511         int ret = check_rbtree_col(t, col_num, &col);
1512         if (ret < 0)
1513                 return ret;
1514         return rbtree_loop(col, private_data, func);
1515 }
1516
1517 __export int osl_rbtree_loop_reverse(const struct osl_table *t, unsigned col_num,
1518         void *private_data, osl_rbtree_loop_func *func)
1519 {
1520         struct osl_column *col;
1521
1522         int ret = check_rbtree_col(t, col_num, &col);
1523         if (ret < 0)
1524                 return ret;
1525         return rbtree_loop_reverse(col, private_data, func);
1526 }
1527
1528 /* TODO: Rollback changes on errors */
1529 static int rename_disk_storage_objects(struct osl_table *t,
1530                 struct osl_object *old_obj, struct osl_object *new_obj)
1531 {
1532         int i, ret;
1533         const struct osl_column_description *cd;
1534         char *old_ds_name, *new_ds_name;
1535
1536         if (!t->num_disk_storage_columns)
1537                 return 1; /* nothing to do */
1538         if (old_obj->size == new_obj->size && !memcmp(new_obj->data,
1539                         old_obj->data, new_obj->size))
1540                 return 1; /* object did not change */
1541         old_ds_name = disk_storage_name_of_object(t, old_obj);
1542         new_ds_name = disk_storage_name_of_object(t, new_obj);
1543         ret = -ERRNO_TO_ERROR(ENOMEM);
1544         if (!old_ds_name || ! new_ds_name)
1545                 goto out;
1546
1547         FOR_EACH_DISK_STORAGE_COLUMN(i, t, cd) {
1548                 char *old_filename, *new_filename;
1549                 ret = create_disk_storage_object_dir(t, i, new_ds_name);
1550                 if (ret < 0)
1551                         goto out;
1552                 old_filename = disk_storage_path(t, i, old_ds_name);
1553                 new_filename = disk_storage_path(t, i, new_ds_name);
1554                 if (!old_filename || !new_filename)
1555                         ret = -ERRNO_TO_ERROR(ENOMEM);
1556                 else
1557                         ret = osl_rename(old_filename, new_filename);
1558                 free(old_filename);
1559                 free(new_filename);
1560                 if (ret < 0)
1561                         goto out;
1562         }
1563         ret = 1;
1564 out:
1565         free(old_ds_name);
1566         free(new_ds_name);
1567         return ret;
1568
1569 }
1570
1571 __export int osl_update_object(struct osl_table *t, const struct osl_row *r,
1572                 unsigned col_num, struct osl_object *obj)
1573 {
1574         struct osl_column *col;
1575         const struct osl_column_description *cd;
1576         int ret;
1577
1578         if (!t)
1579                 return -E_OSL_BAD_TABLE;
1580         col = &t->columns[col_num];
1581         cd = get_column_description(t->desc, col_num);
1582         DEBUG_LOG("updating column %u of %s\n", col_num, t->desc->name);
1583         if (cd->storage_flags & OSL_RBTREE) {
1584                 if (search_rbtree(obj, t, col_num, NULL, NULL) > 0)
1585                         return -E_OSL_RB_KEY_EXISTS;
1586         }
1587         if (cd->storage_flags & OSL_FIXED_SIZE) {
1588                 if (obj->size != cd->data_size)
1589                         return -E_OSL_BAD_DATA_SIZE;
1590         }
1591         remove_rb_node(t, col_num, r);
1592         if (cd->storage_type == OSL_NO_STORAGE) { /* TODO: If fixed size, reuse object? */
1593                 free(r->volatile_objects[col->volatile_num].data);
1594                 r->volatile_objects[col->volatile_num] = *obj;
1595         } else if (cd->storage_type == OSL_DISK_STORAGE) {
1596                 char *ds_name;
1597                 ret = disk_storage_name_of_row(t, r, &ds_name);
1598                 if (ret < 0)
1599                         return ret;
1600                 ret = delete_disk_storage_file(t, col_num, ds_name);
1601                 if (ret < 0 && !is_errno(-ret, ENOENT)) {
1602                         free(ds_name);
1603                         return ret;
1604                 }
1605                 ret = write_disk_storage_file(t, col_num, obj, ds_name);
1606                 free(ds_name);
1607                 if (ret < 0)
1608                         return ret;
1609         } else { /* mapped storage */
1610                 struct osl_object old_obj;
1611                 ret = get_mapped_object(t, col_num, r->num, &old_obj);
1612                 if (ret < 0)
1613                         return ret;
1614                 /*
1615                  * If the updated column is the disk storage name column, the
1616                  * disk storage name changes, so we have to rename all disk
1617                  * storage objects accordingly.
1618                  */
1619                 if (col_num == t->disk_storage_name_column) {
1620                         ret = rename_disk_storage_objects(t, &old_obj, obj);
1621                         if (ret < 0)
1622                                 return ret;
1623                 }
1624                 if (cd->storage_flags & OSL_FIXED_SIZE)
1625                         memcpy(old_obj.data, obj->data, cd->data_size);
1626                 else { /* TODO: if the size doesn't change, use old space */
1627                         uint32_t new_data_map_size;
1628                         char *row_index;
1629                         ret = get_row_index(t, r->num, &row_index);
1630                         if (ret < 0)
1631                                 return ret;
1632                         ret = mark_mapped_object_invalid(t, r->num, col_num);
1633                         if (ret < 0)
1634                                 return ret;
1635                         unmap_column(t, col_num);
1636                         ret = append_map_file(t, col_num, obj,
1637                                 &new_data_map_size);
1638                         if (ret < 0)
1639                                 return ret;
1640                         ret = map_column(t, col_num);
1641                         if (ret < 0)
1642                                 return ret;
1643                         update_cell_index(row_index, col, new_data_map_size,
1644                                 obj->size);
1645                 }
1646         }
1647         if (cd->storage_flags & OSL_RBTREE) {
1648                 ret = insert_rbtree(t, col_num, r, obj);
1649                 if (ret < 0)
1650                         return ret;
1651         }
1652         return 1;
1653 }
1654
1655 __export int osl_open_disk_object(const struct osl_table *t, const struct osl_row *r,
1656                 unsigned col_num, struct osl_object *obj)
1657 {
1658         const struct osl_column_description *cd;
1659         char *ds_name, *filename;
1660         int ret;
1661
1662         if (!t)
1663                 return -E_OSL_BAD_TABLE;
1664         cd = get_column_description(t->desc, col_num);
1665         if (cd->storage_type != OSL_DISK_STORAGE)
1666                 return -E_OSL_BAD_STORAGE_TYPE;
1667
1668         ret = disk_storage_name_of_row(t, r, &ds_name);
1669         if (ret < 0)
1670                 return ret;
1671         filename = disk_storage_path(t, col_num, ds_name);
1672         free(ds_name);
1673         if (!filename)
1674                 return -ERRNO_TO_ERROR(ENOMEM);
1675         DEBUG_LOG("filename: %s\n", filename);
1676         ret = mmap_full_file(filename, O_RDONLY, &obj->data, &obj->size, NULL);
1677         free(filename);
1678         return ret;
1679 }
1680
1681 __export int osl_close_disk_object(struct osl_object *obj)
1682 {
1683         return osl_munmap(obj->data, obj->size);
1684 }
1685
1686 __export int osl_get_num_rows(const struct osl_table *t, unsigned *num_rows)
1687 {
1688         if (!t)
1689                 return -E_OSL_BAD_TABLE;
1690         assert(t->num_rows >= t->num_invalid_rows);
1691         *num_rows = t->num_rows - t->num_invalid_rows;
1692         return 1;
1693 }
1694
1695 __export int osl_get_rank(const struct osl_table *t, struct osl_row *r,
1696                 unsigned col_num, unsigned *rank)
1697 {
1698         struct osl_object obj;
1699         struct osl_column *col;
1700         struct rb_node *node;
1701         int ret = check_rbtree_col(t, col_num, &col);
1702
1703         if (ret < 0)
1704                 return ret;
1705         ret = osl_get_object(t, r, col_num, &obj);
1706         if (ret < 0)
1707                 return ret;
1708         ret = search_rbtree(&obj, t, col_num, &node, NULL);
1709         if (ret < 0)
1710                 return ret;
1711         ret = rb_rank(node, rank);
1712         if (ret < 0)
1713                 return -E_OSL_BAD_ROW;
1714         return 1;
1715 }
1716
1717 __export int osl_get_nth_row(const struct osl_table *t, unsigned col_num,
1718                 unsigned n, struct osl_row **result)
1719 {
1720         struct osl_column *col;
1721         struct rb_node *node;
1722         unsigned num_rows;
1723         int ret;
1724
1725         if (n == 0)
1726                 return -E_OSL_RB_KEY_NOT_FOUND;
1727         ret = osl_get_num_rows(t, &num_rows);
1728         if (ret < 0)
1729                 return ret;
1730         if (n > num_rows)
1731                 return -E_OSL_RB_KEY_NOT_FOUND;
1732         ret = check_rbtree_col(t, col_num, &col);
1733         if (ret < 0)
1734                 return ret;
1735         node = rb_nth(col->rbtree.rb_node, n);
1736         if (!node)
1737                 return -E_OSL_RB_KEY_NOT_FOUND;
1738         *result = get_row_pointer(node, col->rbtree_num);
1739         return 1;
1740 }
1741
1742 __export int osl_rbtree_first_row(const struct osl_table *t, unsigned col_num,
1743                 struct osl_row **result)
1744 {
1745         return osl_get_nth_row(t, col_num, 1, result);
1746 }
1747
1748 __export int osl_rbtree_last_row(const struct osl_table *t, unsigned col_num,
1749                 struct osl_row **result)
1750 {
1751         unsigned num_rows;
1752         int ret = osl_get_num_rows(t, &num_rows);
1753
1754         if (ret < 0)
1755                 return ret;
1756         return osl_get_nth_row(t, col_num, num_rows, result);
1757 }