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