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