osl.c

   1 /*
   2  * Copyright (C) 2007-2009 Andre Noll <maan@tuebingen.mpg.de>
   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 /*
  17  * Taken from Drepper: How to write shared libraries, Appendix B.
  18  *
  19  * The main reason for this rather fancy implementation of strerror() is to
  20  * avoid having an array of pointers. This is desirable because initialized
  21  * pointer variables increase the startup time of the library due to the
  22  * processing of relocations.
  23  */
  24 #include <stddef.h>
  25 #define MSGSTRFIELD(line) MSGSTRFIELD1(line)
  26 #define MSGSTRFIELD1(line) str##line
  27 static const union msgstr_t {
  28         struct {
  29 #define OSL_ERROR(n, s) char MSGSTRFIELD(__LINE__)[sizeof(s)];
  30 #include "errtab.h"
  31 #undef OSL_ERROR
  32         };
  33         char str[0];
  34 } msgstr = { {
  35 #define OSL_ERROR(n, s) s,
  36 #include "errtab.h"
  37 #undef OSL_ERROR
  38 } };
  39 static const unsigned int errmsgidx[] = {
  40 #define OSL_ERROR(n, s) [n] = offsetof(union msgstr_t, MSGSTRFIELD(__LINE__)),
  41 #include "errtab.h"
  42 #undef OSL_ERROR
  43 };
  44
  45 __export const char *osl_strerror(int num)
  46 {
  47         return msgstr.str + errmsgidx[num];
  48 }
  49
  50 int loglevel = 0;
  51
  52 static void __attribute ((constructor)) init_loglevel(void)
  53 {
  54         char *p = getenv("OSL_LOGLEVEL");
  55
  56         /* don't log anything if unset */
  57         loglevel = p? atoi(p) : EMERG + 1;
  58 }
  59
  60 /**
  61  * The log function.
  62  *
  63  * \param ll Loglevel.
  64  * \param fmt Usual format string.
  65  *
  66  * All XXX_LOG() macros use this function.
  67  */
  68 __printf_2_3 void __log(int ll, const char* fmt,...)
  69 {
  70         va_list argp;
  71         struct tm *tm;
  72         time_t t1;
  73         char str[255] = "";
  74
  75         if (ll < loglevel)
  76                 return;
  77         time(&t1);
  78         tm = localtime(&t1);
  79         strftime(str, sizeof(str), "%b %d %H:%M:%S", tm);
  80         fprintf(stderr, "%s ", str);
  81         va_start(argp, fmt);
  82         vfprintf(stderr, fmt, argp);
  83         va_end(argp);
  84 }
  85
  86 /**
  87  * A wrapper for lseek(2).
  88  *
  89  * \param fd The file descriptor whose offset is to be to repositioned.
  90  * \param offset A value-result parameter.
  91  * \param whence Usual repositioning directive.
  92  *
  93  * Reposition the offset of the file descriptor \a fd to the argument \a offset
  94  * according to the directive \a whence. Upon successful return, \a offset
  95  * contains the resulting offset location as measured in bytes from the
  96  * beginning of the file.
  97  *
  98  * \return Positive on success. Otherwise, the function returns \p -E_OSL_LSEEK.
  99  *
 100  * \sa lseek(2).
 101  */
 102 static int __lseek(int fd, off_t *offset, int whence)
 103 {
 104         *offset = lseek(fd, *offset, whence);
 105         int ret = -E_OSL_LSEEK;
 106         if (*offset == -1)
 107                 return ret;
 108         return 1;
 109 }
 110
 111 static int append_file(const char *filename, char *data, size_t data_size,
 112                 uint32_t *new_pos)
 113 {
 114         int ret, fd;
 115
 116 //      DEBUG_LOG("appending %zu  + %zu bytes\n", header_size, data_size);
 117         ret = osl_open(filename, O_WRONLY | O_CREAT | O_APPEND, 0644);
 118         if (ret < 0)
 119                 return ret;
 120         fd = ret;
 121         ret = write_all(fd, data, &data_size);
 122         if (ret < 0)
 123                 goto out;
 124         if (new_pos) {
 125                 off_t offset = 0;
 126                 ret = __lseek(fd, &offset, SEEK_END);
 127                 if (ret < 0)
 128                         goto out;
 129 //              DEBUG_LOG("new file size: " FMT_OFF_T "\n", offset);
 130                 *new_pos = offset;
 131         }
 132         ret = 1;
 133 out:
 134         close(fd);
 135         return ret;
 136 }
 137
 138 static int verify_name(const char *name)
 139 {
 140         if (!name)
 141                 return -E_OSL_BAD_NAME;
 142         if (!*name)
 143                 return -E_OSL_BAD_NAME;
 144         if (strchr(name, '/'))
 145                 return -E_OSL_BAD_NAME;
 146         if (!strcmp(name, ".."))
 147                 return -E_OSL_BAD_NAME;
 148         if (!strcmp(name, "."))
 149                 return -E_OSL_BAD_NAME;
 150         return 1;
 151 }
 152
 153 static char *disk_storage_dirname(const struct osl_table *t, unsigned col_num,
 154                 const char *ds_name)
 155 {
 156         char *dirname, *column_name = column_filename(t, col_num);
 157
 158         if (!column_name)
 159                 return NULL;
 160         if (!(t->desc->flags & OSL_LARGE_TABLE))
 161                 return column_name;
 162         dirname = make_message("%s/%.2s", column_name, ds_name);
 163         free(column_name);
 164         return dirname;
 165 }
 166
 167 static char *disk_storage_name_of_object(const struct osl_table *t,
 168         const struct osl_object *obj)
 169 {
 170         HASH_TYPE hash[HASH_SIZE];
 171         hash_object(obj, hash);
 172         return disk_storage_name_of_hash(t, hash);
 173 }
 174
 175 static int disk_storage_name_of_row(const struct osl_table *t,
 176                 const struct osl_row *row, char **name)
 177 {
 178         struct osl_object obj;
 179         int ret = osl_get_object(t, row, t->disk_storage_name_column, &obj);
 180
 181         if (ret < 0)
 182                 return ret;
 183         *name = disk_storage_name_of_object(t, &obj);
 184         if (*name)
 185                 return 1;
 186         return -E_OSL_NOMEM;
 187 }
 188
 189 static void column_name_hash(const char *col_name, HASH_TYPE *hash)
 190 {
 191         hash_function(col_name, strlen(col_name), hash);
 192 }
 193
 194 static int init_column_descriptions(struct osl_table *t)
 195 {
 196         int i, j, ret;
 197         const struct osl_column_description *cd;
 198
 199         ret = verify_name(t->desc->name);
 200         if (ret < 0)
 201                 goto err;
 202         ret = -E_OSL_BAD_DB_DIR;
 203         if (!t->desc->dir && (t->num_disk_storage_columns || t->num_mapped_columns))
 204                 goto err;
 205         /* the size of the index header without column descriptions */
 206         t->index_header_size = IDX_COLUMN_DESCRIPTIONS;
 207         FOR_EACH_COLUMN(i, t->desc, cd) {
 208                 struct osl_column *col = t->columns + i;
 209                 if (cd->storage_flags & OSL_RBTREE) {
 210                         if (!cd->compare_function)
 211                                 return -E_OSL_NO_COMPARE_FUNC;
 212                 }
 213                 if (cd->storage_type == OSL_NO_STORAGE)
 214                         continue;
 215                 ret = -E_OSL_NO_COLUMN_NAME;
 216                 if (!cd->name || !cd->name[0])
 217                         goto err;
 218                 ret = verify_name(cd->name);
 219                 if (ret < 0)
 220                         goto err;
 221                 t->index_header_size += index_column_description_size(cd->name);
 222                 column_name_hash(cd->name, col->name_hash);
 223                 ret = -E_OSL_DUPLICATE_COL_NAME;
 224                 for (j = i + 1; j < t->desc->num_columns; j++) {
 225                         const char *name2 = get_column_description(t->desc,
 226                                 j)->name;
 227                         if (cd->name && name2 && !strcmp(cd->name, name2))
 228                                 goto err;
 229                 }
 230         }
 231         return 1;
 232 err:
 233         return ret;
 234 }
 235
 236 /**
 237  * Initialize a struct table from given table description.
 238  *
 239  * \param desc The description of the osl table.
 240  * \param table_ptr Result is returned here.
 241  *
 242  * This function performs several sanity checks on \p desc and returns if any
 243  * of these tests fail. On success, a struct \p osl_table is allocated and
 244  * initialized with data derived from \p desc.
 245  *
 246  * \return Standard.
 247  *
 248  * \sa struct osl_table.
 249  */
 250 int init_table_structure(const struct osl_table_description *desc,
 251                 struct osl_table **table_ptr)
 252 {
 253         const struct osl_column_description *cd;
 254         struct osl_table *t = calloc(1, sizeof(*t));
 255         int i, ret = -E_OSL_NOMEM, have_disk_storage_name_column = 0;
 256
 257         if (!t)
 258                 return ret;
 259         ret = -E_OSL_BAD_TABLE_DESC;
 260         if (!desc)
 261                 goto err;
 262         DEBUG_LOG("creating table structure for '%s' from table "
 263                 "description\n", desc->name);
 264         ret = -E_OSL_NO_COLUMN_DESC;
 265         if (!desc->column_descriptions)
 266                 goto err;
 267         ret = -E_OSL_NO_COLUMNS;
 268         if (!desc->num_columns)
 269                 goto err;
 270         ret = -E_OSL_NOMEM;
 271         t->columns = calloc(desc->num_columns, sizeof(struct osl_column));
 272         if (!t->columns)
 273                 goto err;
 274         t->desc = desc;
 275         FOR_EACH_COLUMN(i, t->desc, cd) {
 276                 enum osl_storage_type st = cd->storage_type;
 277                 enum osl_storage_flags sf = cd->storage_flags;
 278                 struct osl_column *col = &t->columns[i];
 279
 280                 ret = -E_OSL_BAD_STORAGE_TYPE;
 281                 if (st != OSL_MAPPED_STORAGE && st != OSL_DISK_STORAGE
 282                                 && st != OSL_NO_STORAGE)
 283                         goto err;
 284                 ret = -E_OSL_BAD_STORAGE_FLAGS;
 285                 if (st == OSL_DISK_STORAGE && sf & OSL_RBTREE)
 286                         goto err;
 287                 if ((sf & OSL_RBTREE) && !(sf & OSL_UNIQUE))
 288                         WARNING_LOG("invalid storage flags for column %s: "
 289                                 "OSL_RBTREE && !OSL_UNIQUE\n", cd->name);
 290                 ret = -E_OSL_BAD_STORAGE_SIZE;
 291                 if (sf & OSL_FIXED_SIZE && !cd->data_size)
 292                         goto err;
 293                 switch (st) {
 294                 case OSL_DISK_STORAGE:
 295                         t->num_disk_storage_columns++;
 296                         break;
 297                 case OSL_MAPPED_STORAGE:
 298                         t->num_mapped_columns++;
 299                         col->index_offset = t->row_index_size;
 300                         t->row_index_size += 8;
 301                         break;
 302                 case OSL_NO_STORAGE:
 303                         col->volatile_num = t->num_volatile_columns;
 304                         t->num_volatile_columns++;
 305                         break;
 306                 }
 307                 if (sf & OSL_RBTREE) {
 308                         col->rbtree_num = t->num_rbtrees;
 309                         t->num_rbtrees++;
 310                         if ((sf & OSL_UNIQUE) && (st == OSL_MAPPED_STORAGE)) {
 311                                 if (!have_disk_storage_name_column)
 312                                         t->disk_storage_name_column = i;
 313                                 have_disk_storage_name_column = 1;
 314                         }
 315                 }
 316         }
 317         ret = -E_OSL_NO_UNIQUE_RBTREE_COLUMN;
 318         if (t->num_disk_storage_columns && !have_disk_storage_name_column)
 319                 goto err;
 320         ret = -E_OSL_NO_RBTREE_COL;
 321         if (!t->num_rbtrees)
 322                 goto err;
 323         /* success */
 324         DEBUG_LOG("OK. Index entry size: %u\n", t->row_index_size);
 325         ret = init_column_descriptions(t);
 326         if (ret < 0)
 327                 goto err;
 328         *table_ptr = t;
 329         return 1;
 330 err:
 331         free(t->columns);
 332         free(t);
 333         return ret;
 334 }
 335
 336 /**
 337  * Read the table description from index header.
 338  *
 339  * \param map The memory mapping of the index file.
 340  * \param desc The values found in the index header are returned here.
 341  *
 342  * Read the index header, check for the osl magic string and the table version
 343  * number.  Read all information stored in the index header into \a desc.
 344  *
 345  * \return Standard.
 346  *
 347  * \sa struct osl_table_description, osl_create_table.
 348  */
 349 int read_table_desc(struct osl_object *map, struct osl_table_description *desc)
 350 {
 351         char *buf = map->data;
 352         uint8_t version, compat_version, create_version;
 353         uint16_t header_size;
 354         int ret, i;
 355         unsigned offset;
 356         struct osl_column_description *cd;
 357
 358         if (map->size < MIN_INDEX_HEADER_SIZE(1))
 359                 return -E_OSL_SHORT_TABLE;
 360         if (strncmp(buf + IDX_OSL_MAGIC, OSL_MAGIC, strlen(OSL_MAGIC)))
 361                 return -E_OSL_NO_MAGIC;
 362         version = read_u8(buf + IDX_VERSION);
 363         /*
 364          * The on-disk version consists of two version numbers: the
 365          * create_version (low 4 bits) is the CURRENT_TABLE_VERSION version
 366          * number of the library that created the table, and compat_version
 367          * (high 4 bits) tells us the lowest version of the library that can
 368          * still read this table.
 369          */
 370         create_version = version & 0xf;
 371         compat_version = version >> 4;
 372         INFO_LOG("create_version: %u, compat_version: %u\n", create_version,
 373                 compat_version);
 374         if (create_version < MIN_TABLE_VERSION /* table too old */
 375                 || compat_version > CURRENT_TABLE_VERSION) /* libosl too old */
 376                 return -E_OSL_VERSION_MISMATCH;
 377         desc->flags = read_u8(buf + IDX_TABLE_FLAGS);
 378         desc->num_columns = read_u16(buf + IDX_NUM_COLUMNS);
 379         INFO_LOG("%u columns\n", desc->num_columns);
 380         if (!desc->num_columns)
 381                 return -E_OSL_NO_COLUMNS;
 382         header_size = read_u16(buf + IDX_HEADER_SIZE);
 383         if (map->size < header_size)
 384                 return -E_OSL_BAD_SIZE;
 385         desc->column_descriptions = calloc(desc->num_columns,
 386                 sizeof(struct osl_column_description));
 387         if (!desc->column_descriptions)
 388                 return -E_OSL_NOMEM;
 389         offset = IDX_COLUMN_DESCRIPTIONS;
 390         FOR_EACH_COLUMN(i, desc, cd) {
 391                 char *null_byte;
 392
 393                 ret = -E_OSL_SHORT_TABLE;
 394                 if (map->size < offset + MIN_IDX_COLUMN_DESCRIPTION_SIZE) {
 395                         ERROR_LOG("map size = %zu < %u = offset + min desc size\n",
 396                                 map->size, offset + MIN_IDX_COLUMN_DESCRIPTION_SIZE);
 397                         goto err;
 398                 }
 399                 cd->storage_type = read_u16(buf + offset + IDX_CD_STORAGE_TYPE);
 400                 cd->storage_flags = read_u16(buf + offset +
 401                         IDX_CD_STORAGE_FLAGS);
 402                 cd->data_size = read_u32(buf + offset + IDX_CD_DATA_SIZE);
 403                 null_byte = memchr(buf + offset + IDX_CD_NAME, '\0',
 404                         map->size - offset - IDX_CD_NAME);
 405                 ret = -E_OSL_INDEX_CORRUPTION;
 406                 if (!null_byte)
 407                         goto err;
 408                 ret = -E_OSL_NOMEM;
 409                 cd->name = strdup(buf + offset + IDX_CD_NAME);
 410                 if (!cd->name)
 411                         goto err;
 412                 offset += index_column_description_size(cd->name);
 413         }
 414         if (offset != header_size) {
 415                 ret = -E_OSL_INDEX_CORRUPTION;
 416                 ERROR_LOG("real header size = %u != %u = stored header size\n",
 417                         offset, header_size);
 418                 goto err;
 419         }
 420         return 1;
 421 err:
 422         FOR_EACH_COLUMN(i, desc, cd)
 423                 free(cd->name);
 424         return ret;
 425 }
 426
 427 /*
 428  * check whether the table description given by \p t->desc matches the on-disk
 429  * table structure stored in the index of \a t.
 430  */
 431 static int compare_table_descriptions(struct osl_table *t)
 432 {
 433         int i, ret;
 434         struct osl_table_description desc;
 435         const struct osl_column_description *cd1, *cd2;
 436
 437         /* read the on-disk structure into desc */
 438         ret = read_table_desc(&t->index_map, &desc);
 439         if (ret < 0)
 440                 return ret;
 441         ret = -E_OSL_BAD_TABLE_FLAGS;
 442         if (desc.flags != t->desc->flags)
 443                 goto out;
 444         ret = -E_OSL_BAD_COLUMN_NUM;
 445         if (desc.num_columns > t->desc->num_columns)
 446                 goto out;
 447         if (desc.num_columns < t->desc->num_columns) {
 448                 struct osl_column_description *cd;
 449                 unsigned diff = t->desc->num_columns - desc.num_columns;
 450                 INFO_LOG("extending table by %u volatile columns\n", diff);
 451                 ret = -E_OSL_NOMEM;
 452                 desc.column_descriptions = realloc(desc.column_descriptions,
 453                         t->desc->num_columns * sizeof(struct osl_column_description));
 454                 if (!desc.column_descriptions)
 455                         goto out;
 456                 for (i = desc.num_columns; i < t->desc->num_columns; i++) {
 457                         cd = get_column_description(&desc, i);
 458                         cd->storage_type = OSL_NO_STORAGE;
 459                         cd->name = NULL;
 460                 }
 461                 desc.num_columns += diff;
 462         }
 463         FOR_EACH_COLUMN(i, t->desc, cd1) {
 464                 cd2 = get_column_description(&desc, i);
 465                 ret = -E_OSL_BAD_STORAGE_TYPE;
 466                 if (cd1->storage_type != cd2->storage_type)
 467                         goto out;
 468                 if (cd1->storage_type == OSL_NO_STORAGE)
 469                         continue;
 470                 ret = -E_OSL_BAD_STORAGE_FLAGS;
 471                 if (cd1->storage_flags != cd2->storage_flags) {
 472                         ERROR_LOG("sf1 = %u != %u = sf2\n",
 473                                 cd1->storage_flags, cd2->storage_flags);
 474                         goto out;
 475                 }
 476                 ret = -E_OSL_BAD_DATA_SIZE;
 477                 if (cd1->storage_flags & OSL_FIXED_SIZE)
 478                         if (cd1->data_size != cd2->data_size)
 479                                 goto out;
 480                 ret = -E_OSL_BAD_COLUMN_NAME;
 481                 if (strcmp(cd1->name, cd2->name))
 482                         goto out;
 483         }
 484         INFO_LOG("table description of '%s' matches on-disk data, good\n",
 485                 t->desc->name);
 486         ret = 1;
 487 out:
 488         FOR_EACH_COLUMN(i, &desc, cd1)
 489                 free(cd1->name);
 490         free(desc.column_descriptions);
 491         return ret;
 492 }
 493
 494 static int create_table_index(struct osl_table *t)
 495 {
 496         char *buf, *filename;
 497         int i, ret;
 498         size_t size = t->index_header_size;
 499         const struct osl_column_description *cd;
 500         unsigned offset;
 501
 502         INFO_LOG("creating %zu byte index for table %s\n", size,
 503                 t->desc->name);
 504         buf = calloc(1, size);
 505         if (!buf)
 506                 return -E_OSL_NOMEM;
 507         sprintf(buf + IDX_OSL_MAGIC, "%s", OSL_MAGIC);
 508         write_u8(buf + IDX_TABLE_FLAGS, t->desc->flags);
 509         write_u8(buf + IDX_DIRTY_FLAG, 0);
 510         write_u8(buf + IDX_VERSION, CURRENT_TABLE_VERSION
 511                 + (COMPAT_TABLE_VERSION << 4));
 512         write_u16(buf + IDX_NUM_COLUMNS, t->num_mapped_columns + t->num_disk_storage_columns);
 513         write_u16(buf + IDX_HEADER_SIZE, t->index_header_size);
 514         offset = IDX_COLUMN_DESCRIPTIONS;
 515         FOR_EACH_COLUMN(i, t->desc, cd) {
 516                 /* no need to store info about volatile storage */
 517                 if (cd->storage_type == OSL_NO_STORAGE)
 518                         continue;
 519                 write_u16(buf + offset + IDX_CD_STORAGE_TYPE,
 520                         cd->storage_type);
 521                 write_u16(buf + offset + IDX_CD_STORAGE_FLAGS,
 522                         cd->storage_flags);
 523                 if (cd->storage_flags & OSL_FIXED_SIZE)
 524                         write_u32(buf + offset + IDX_CD_DATA_SIZE,
 525                                 cd->data_size);
 526                 strcpy(buf + offset + IDX_CD_NAME, cd->name);
 527                 offset += index_column_description_size(cd->name);
 528         }
 529         assert(offset == size);
 530         filename = index_filename(t->desc);
 531         if (filename)
 532                 ret = write_file(filename, buf, size);
 533         else
 534                 ret = -E_OSL_NOMEM;
 535         free(buf);
 536         free(filename);
 537         return ret;
 538 }
 539
 540 __export int osl_create_table(const struct osl_table_description *desc)
 541 {
 542         const struct osl_column_description *cd;
 543         char *table_dir = NULL, *filename;
 544         struct osl_table *t;
 545         int i, ret = init_table_structure(desc, &t);
 546
 547         if (ret < 0)
 548                 return ret;
 549         INFO_LOG("creating %s\n", desc->name);
 550         FOR_EACH_COLUMN(i, t->desc, cd) {
 551                 if (cd->storage_type == OSL_NO_STORAGE)
 552                         continue;
 553                 if (!table_dir) {
 554                         ret = osl_mkdir(desc->dir, 0777);
 555                         if (ret < 0 && ret != -E_OSL_DIR_EXISTS)
 556                                 goto out;
 557                         table_dir = make_message("%s/%s", desc->dir,
 558                                 desc->name);
 559                         ret = -E_OSL_NOMEM;
 560                         if (!table_dir)
 561                                 goto out;
 562                         ret = osl_mkdir(table_dir, 0777);
 563                         if (ret < 0)
 564                                 goto out;
 565                 }
 566                 ret = -E_OSL_NOMEM;
 567                 filename = column_filename(t, i);
 568                 if (!filename)
 569                         goto out;
 570                 INFO_LOG("filename: %s\n", filename);
 571                 if (cd->storage_type == OSL_MAPPED_STORAGE) {
 572                         ret = osl_open(filename, O_RDWR | O_CREAT | O_EXCL,
 573                                 0644);
 574                         free(filename);
 575                         if (ret < 0)
 576                                 goto out;
 577                         close(ret);
 578                         continue;
 579                 }
 580                 /* DISK STORAGE */
 581                 ret = osl_mkdir(filename, 0777);
 582                 free(filename);
 583                 if (ret < 0)
 584                         goto out;
 585         }
 586         if (t->num_mapped_columns) {
 587                 ret = create_table_index(t);
 588                 if (ret < 0)
 589                         goto out;
 590         }
 591         ret = 1;
 592 out:
 593         free(table_dir);
 594         free(t->columns);
 595         free(t);
 596         return ret;
 597 }
 598
 599 static int table_is_dirty(struct osl_table *t)
 600 {
 601         char *buf = (char *)t->index_map.data + IDX_DIRTY_FLAG;
 602         uint8_t dirty = read_u8(buf) & 0x1;
 603         return !!dirty;
 604 }
 605
 606 static void mark_table_dirty(struct osl_table *t)
 607 {
 608         char *buf = (char *)t->index_map.data + IDX_DIRTY_FLAG;
 609         write_u8(buf, read_u8(buf) | 1);
 610 }
 611
 612 static void mark_table_clean(struct osl_table *t)
 613 {
 614         char *buf = (char *)t->index_map.data + IDX_DIRTY_FLAG;
 615         write_u8(buf, read_u8(buf) & 0xfe);
 616 }
 617
 618 static void unmap_column(struct osl_table *t, unsigned col_num)
 619 {
 620         struct osl_object map = t->columns[col_num].data_map;
 621         int ret;
 622         if (!map.data)
 623                 return;
 624         ret = osl_munmap(map.data, map.size);
 625         assert(ret > 0);
 626         map.data = NULL;
 627 }
 628
 629 /**
 630  * Unmap all mapped files of an osl table.
 631  *
 632  * \param t Pointer to a mapped table.
 633  * \param flags Options for unmapping.
 634  *
 635  * \return Positive on success, negative on errors.
 636  *
 637  * \sa map_table(), enum osl_close_flags, osl_munmap().
 638  */
 639 int unmap_table(struct osl_table *t, enum osl_close_flags flags)
 640 {
 641         unsigned i;
 642         const struct osl_column_description *cd;
 643         int ret;
 644
 645         if (!t->num_mapped_columns) /* can this ever happen? */
 646                 return 1;
 647         INFO_LOG("unmapping table '%s'\n", t->desc->name);
 648         if (!t->index_map.data)
 649                 return -E_OSL_NOT_MAPPED;
 650         if (flags & OSL_MARK_CLEAN)
 651                 mark_table_clean(t);
 652         ret = osl_munmap(t->index_map.data, t->index_map.size);
 653         if (ret < 0)
 654                 return ret;
 655         t->index_map.data = NULL;
 656         if (!t->num_rows)
 657                 return 1;
 658         FOR_EACH_MAPPED_COLUMN(i, t, cd)
 659                 unmap_column(t, i);
 660         return 1;
 661 }
 662
 663 static int map_column(struct osl_table *t, unsigned col_num)
 664 {
 665         struct stat statbuf;
 666         char *filename = column_filename(t, col_num);
 667         int ret;
 668
 669         if (!filename)
 670                 return -E_OSL_NOMEM;
 671         ret = osl_stat(filename, &statbuf);
 672         if (ret < 0) {
 673                 free(filename);
 674                 return ret;
 675         }
 676         if (!(S_IFREG & statbuf.st_mode)) {
 677                 free(filename);
 678                 return ret;
 679         }
 680         ret = mmap_full_file(filename, O_RDWR,
 681                 &t->columns[col_num].data_map.data,
 682                 &t->columns[col_num].data_map.size,
 683                 NULL);
 684         free(filename);
 685         return ret;
 686 }
 687
 688 /**
 689  * Map the index file and all columns of type \p OSL_MAPPED_STORAGE into memory.
 690  *
 691  * \param t Pointer to an initialized table structure.
 692  * \param flags Mapping options.
 693  *
 694  * \return Negative return value on errors; on success the number of rows
 695  * (including invalid rows) is returned.
 696  *
 697  * \sa unmap_table(), enum map_table_flags, osl_open_table(), mmap(2).
 698  */
 699 int map_table(struct osl_table *t, enum map_table_flags flags)
 700 {
 701         char *filename;
 702         const struct osl_column_description *cd;
 703         int i = 0, ret, num_rows = 0;
 704
 705         if (!t->num_mapped_columns)
 706                 return 0;
 707         if (t->index_map.data)
 708                 return -E_OSL_ALREADY_MAPPED;
 709         filename = index_filename(t->desc);
 710         if (!filename)
 711                 return -E_OSL_NOMEM;
 712         INFO_LOG("mapping table '%s' (index: %s)\n", t->desc->name, filename);
 713         ret = mmap_full_file(filename, flags & MAP_TBL_FL_MAP_RDONLY?
 714                 O_RDONLY : O_RDWR, &t->index_map.data, &t->index_map.size, NULL);
 715         free(filename);
 716         if (ret < 0)
 717                 return ret;
 718         if (flags & MAP_TBL_FL_VERIFY_INDEX) {
 719                 ret = compare_table_descriptions(t);
 720                 if (ret < 0)
 721                         goto err;
 722         }
 723         ret = -E_OSL_BUSY;
 724         if (!(flags & MAP_TBL_FL_IGNORE_DIRTY)) {
 725                 if (table_is_dirty(t)) {
 726                         ERROR_LOG("%s is dirty\n", t->desc->name);
 727                         goto err;
 728                 }
 729         }
 730         mark_table_dirty(t);
 731         num_rows = table_num_rows(t);
 732         if (!num_rows)
 733                 return num_rows;
 734         /* map data files */
 735         FOR_EACH_MAPPED_COLUMN(i, t, cd) {
 736                 ret = map_column(t, i);
 737                 if (ret < 0)
 738                         goto err;
 739         }
 740         return num_rows;
 741 err:    /* unmap what is already mapped */
 742         for (i--; i >= 0; i--) {
 743                 struct osl_object map = t->columns[i].data_map;
 744                 osl_munmap(map.data, map.size);
 745                 map.data = NULL;
 746         }
 747         osl_munmap(t->index_map.data, t->index_map.size);
 748         t->index_map.data = NULL;
 749         return ret;
 750 }
 751
 752 /**
 753  * Retrieve a mapped object by row and column number.
 754  *
 755  * \param t Pointer to an open osl table.
 756  * \param col_num Number of the mapped column containing the object to retrieve.
 757  * \param row_num Number of the row containing the object to retrieve.
 758  * \param obj The result is returned here.
 759  *
 760  * It is considered an error if \a col_num does not refer to a column
 761  * of storage type \p OSL_MAPPED_STORAGE.
 762  *
 763  * \return Standard.
 764  *
 765  * \sa osl_storage_type.
 766  */
 767 int get_mapped_object(const struct osl_table *t, unsigned col_num,
 768         uint32_t row_num, struct osl_object *obj)
 769 {
 770         struct osl_column *col = &t->columns[col_num];
 771         uint32_t offset;
 772         char *cell_index;
 773         int ret;
 774
 775         if (t->num_rows <= row_num)
 776                 return -E_OSL_BAD_ROW_NUM;
 777         ret = get_cell_index(t, row_num, col_num, &cell_index);
 778         if (ret < 0)
 779                 return ret;
 780         offset = read_u32(cell_index);
 781         obj->size = read_u32(cell_index + 4);
 782         obj->data = col->data_map.data + offset;
 783         return 1;
 784 }
 785
 786 /*
 787  * It's OK to call this with result = rb_node = NULL.  If result is not NULL,
 788  * and rb key was not found, result points to the parent node.
 789  */
 790 static int search_rbtree(const struct osl_object *obj,
 791                 const struct osl_table *t, unsigned col_num,
 792                 struct rb_node **result, struct rb_node ***rb_link)
 793 {
 794         struct osl_column *col = &t->columns[col_num];
 795         struct rb_node **new = &col->rbtree.rb_node, *parent = NULL;
 796         const struct osl_column_description *cd =
 797                 get_column_description(t->desc, col_num);
 798         enum osl_storage_type st = cd->storage_type;
 799         while (*new) {
 800                 struct osl_row *this_row = get_row_pointer(*new,
 801                         col->rbtree_num);
 802                 int ret;
 803                 struct osl_object this_obj;
 804                 parent = *new;
 805                 if (st == OSL_MAPPED_STORAGE) {
 806                         ret = get_mapped_object(t, col_num, this_row->num,
 807                                 &this_obj);
 808                         if (ret < 0)
 809                                 return ret;
 810                 } else
 811                         this_obj = this_row->volatile_objects[col->volatile_num];
 812                 ret = cd->compare_function(obj, &this_obj);
 813                 if (!ret) {
 814                         if (result)
 815                                 *result = get_rb_node_pointer(this_row,
 816                                         col->rbtree_num);
 817                         return 1;
 818                 }
 819                 if (ret < 0)
 820                         new = &((*new)->rb_left);
 821                 else
 822                         new = &((*new)->rb_right);
 823         }
 824         if (result)
 825                 *result = parent;
 826         if (rb_link)
 827                 *rb_link = new;
 828         return -E_OSL_RB_KEY_NOT_FOUND;
 829 }
 830
 831 static int insert_rbtree(struct osl_table *t, unsigned col_num,
 832         const struct osl_row *row, const struct osl_object *obj)
 833 {
 834         struct rb_node *parent, **rb_link;
 835         unsigned rbtree_num;
 836         struct rb_node *n;
 837         int ret = search_rbtree(obj, t, col_num, &parent, &rb_link);
 838
 839         if (ret > 0)
 840                 return -E_OSL_RB_KEY_EXISTS;
 841         rbtree_num = t->columns[col_num].rbtree_num;
 842         n = get_rb_node_pointer(row, rbtree_num);
 843         rb_link_node(n, parent, rb_link);
 844         rb_insert_color(n, &t->columns[col_num].rbtree);
 845         return 1;
 846 }
 847
 848 static void remove_rb_node(struct osl_table *t, unsigned col_num,
 849                 const struct osl_row *row)
 850 {
 851         struct osl_column *col = &t->columns[col_num];
 852         const struct osl_column_description *cd =
 853                 get_column_description(t->desc, col_num);
 854         enum osl_storage_flags sf = cd->storage_flags;
 855         struct rb_node *victim, *splice_out_node, *tmp;
 856         if (!(sf & OSL_RBTREE))
 857                 return;
 858         /*
 859          * Which node is removed/spliced out actually depends on how many
 860          * children the victim node has: If it has no children, it gets
 861          * deleted. If it has one child, it gets spliced out. If it has two
 862          * children, its successor (which has at most a right child) gets
 863          * spliced out.
 864          */
 865         victim = get_rb_node_pointer(row, col->rbtree_num);
 866         if (victim->rb_left && victim->rb_right)
 867                 splice_out_node = rb_next(victim);
 868         else
 869                 splice_out_node = victim;
 870         /* Go up to the root and decrement the size of each node in the path. */
 871         for (tmp = splice_out_node; tmp; tmp = rb_parent(tmp))
 872                 tmp->size--;
 873         rb_erase(victim, &col->rbtree);
 874 }
 875
 876 static int add_row_to_rbtrees(struct osl_table *t, uint32_t row_num,
 877                 struct osl_object *volatile_objs, struct osl_row **row_ptr)
 878 {
 879         unsigned i;
 880         int ret;
 881         struct osl_row *row = allocate_row(t->num_rbtrees);
 882         const struct osl_column_description *cd;
 883
 884         if (!row)
 885                 return -E_OSL_NOMEM;
 886         row->num = row_num;
 887         row->volatile_objects = volatile_objs;
 888         FOR_EACH_RBTREE_COLUMN(i, t, cd) {
 889                 if (cd->storage_type == OSL_MAPPED_STORAGE) {
 890                         struct osl_object obj;
 891                         ret = get_mapped_object(t, i, row_num, &obj);
 892                         if (ret < 0)
 893                                 goto err;
 894                         ret = insert_rbtree(t, i, row, &obj);
 895                 } else { /* volatile */
 896                         const struct osl_object *obj
 897                                 = volatile_objs + t->columns[i].volatile_num;
 898                         ret = insert_rbtree(t, i, row, obj);
 899                 }
 900                 if (ret < 0)
 901                         goto err;
 902         }
 903         if (row_ptr)
 904                 *row_ptr = row;
 905         return 1;
 906 err: /* rollback changes, i.e. remove added entries from rbtrees */
 907         while (i)
 908                 remove_rb_node(t, i--, row);
 909         free(row);
 910         return ret;
 911 }
 912
 913 static void free_volatile_objects(const struct osl_table *t,
 914                 enum osl_close_flags flags)
 915 {
 916         int i, j;
 917         struct rb_node *n;
 918         struct osl_column *rb_col;
 919         const struct osl_column_description *cd;
 920
 921         if (!t->num_volatile_columns)
 922                 return;
 923         /* find the first rbtree column (any will do) */
 924         FOR_EACH_RBTREE_COLUMN(i, t, cd)
 925                 break;
 926         rb_col = t->columns + i;
 927         /* walk that rbtree and free all volatile objects */
 928         for (n = rb_first(&rb_col->rbtree); n; n = rb_next(n)) {
 929                 struct osl_row *r = get_row_pointer(n, rb_col->rbtree_num);
 930                 if (flags & OSL_FREE_VOLATILE)
 931                         FOR_EACH_VOLATILE_COLUMN(j, t, cd) {
 932                                 if (cd->storage_flags & OSL_DONT_FREE)
 933                                         continue;
 934                                 free(r->volatile_objects[
 935                                         t->columns[j].volatile_num].data);
 936                         }
 937 //                      for (j = 0; j < t->num_volatile_columns; j++)
 938 //                              free(r->volatile_objects[j].data);
 939                 free(r->volatile_objects);
 940         }
 941 }
 942
 943 /**
 944  * Erase all rbtree nodes and free resources.
 945  *
 946  * \param t Pointer to an open osl table.
 947  *
 948  * This function is called by osl_close_table().
 949  */
 950 void clear_rbtrees(struct osl_table *t)
 951 {
 952         const struct osl_column_description *cd;
 953         unsigned i, rbtrees_cleared = 0;
 954
 955         FOR_EACH_RBTREE_COLUMN(i, t, cd) {
 956                 struct osl_column *col = &t->columns[i];
 957                 struct rb_node *n;
 958                 rbtrees_cleared++;
 959                 for (n = rb_first(&col->rbtree); n;) {
 960                         struct osl_row *r;
 961                         rb_erase(n, &col->rbtree);
 962                         if (rbtrees_cleared == t->num_rbtrees) {
 963                                 r = get_row_pointer(n, col->rbtree_num);
 964                                 n = rb_next(n);
 965                                 free(r);
 966                         } else
 967                                 n = rb_next(n);
 968                 }
 969         }
 970
 971 }
 972
 973 __export int osl_close_table(struct osl_table *t, enum osl_close_flags flags)
 974 {
 975         int ret;
 976
 977         if (!t)
 978                 return -E_OSL_BAD_TABLE;
 979         NOTICE_LOG("closing table %s\n", t->desc->name);
 980         free_volatile_objects(t, flags);
 981         clear_rbtrees(t);
 982         ret = unmap_table(t, flags);
 983         if (ret < 0)
 984                 ERROR_LOG("unmap_table failed: %d\n", ret);
 985         free(t->columns);
 986         free(t);
 987         return ret;
 988 }
 989
 990 /**
 991  * Find out whether the given row number corresponds to an invalid row.
 992  *
 993  * \param t Pointer to the osl table.
 994  * \param row_num The number of the row in question.
 995  *
 996  * By definition, a row is considered invalid if all its index entries
 997  * are invalid.
 998  *
 999  * \return Positive if \a row_num corresponds to an invalid row,
1000  * zero if it corresponds to a valid row, negative on errors.
1001  */
1002 int row_is_invalid(struct osl_table *t, uint32_t row_num)
1003 {
1004         char *row_index;
1005         int i, ret = get_row_index(t, row_num, &row_index);
1006
1007         if (ret < 0)
1008                 return ret;
1009         for (i = 0; i < t->row_index_size; i++) {
1010                 if ((unsigned char)row_index[i] != 0xff)
1011                         return 0;
1012         }
1013         INFO_LOG("row %d is invalid\n", row_num);
1014         return 1;
1015 }
1016
1017 /**
1018  * Invalidate a row of an osl table.
1019  *
1020  * \param t Pointer to an open osl table.
1021  * \param row_num Number of the row to mark as invalid.
1022  *
1023  * This function marks each mapped object in the index entry of \a row as
1024  * invalid.
1025  *
1026  * \return Standard.
1027  */
1028 int mark_row_invalid(struct osl_table *t, uint32_t row_num)
1029 {
1030         char *row_index;
1031         int ret = get_row_index(t, row_num, &row_index);
1032
1033         if (ret < 0)
1034                 return ret;
1035         INFO_LOG("marking row %d as invalid\n", row_num);
1036         memset(row_index, 0xff, t->row_index_size);
1037         return 1;
1038 }
1039
1040 /**
1041  * Initialize all rbtrees and compute number of invalid rows.
1042  *
1043  * \param t The table containing the rbtrees to be initialized.
1044  *
1045  * \return Standard.
1046  */
1047 int init_rbtrees(struct osl_table *t)
1048 {
1049         int i, ret;
1050         const struct osl_column_description *cd;
1051
1052         /* create rbtrees */
1053         FOR_EACH_RBTREE_COLUMN(i, t, cd)
1054                 t->columns[i].rbtree = RB_ROOT;
1055         /* add valid rows to rbtrees */
1056         t->num_invalid_rows = 0;
1057         for (i = 0; i < t->num_rows; i++) {
1058                 struct osl_object *volatile_objs;
1059                 ret = row_is_invalid(t, i);
1060                 if (ret < 0)
1061                         return ret;
1062                 if (ret) {
1063                         t->num_invalid_rows++;
1064                         continue;
1065                 }
1066                 if (t->num_volatile_columns > 0) {
1067                         volatile_objs = calloc(t->num_volatile_columns,
1068                                 sizeof(struct osl_object));
1069                         if (!volatile_objs)
1070                                 return -E_OSL_NOMEM;
1071                 } else
1072                         volatile_objs = NULL;
1073                 ret = add_row_to_rbtrees(t, i, volatile_objs, NULL);
1074                 if (ret < 0)
1075                         return ret;
1076         }
1077         return 1;
1078 }
1079
1080 __export int osl_open_table(const struct osl_table_description *table_desc,
1081                 struct osl_table **result)
1082 {
1083         int i, ret;
1084         struct osl_table *t;
1085         const struct osl_column_description *cd;
1086
1087         NOTICE_LOG("opening table %s\n", table_desc->name);
1088         ret = init_table_structure(table_desc, &t);
1089         if (ret < 0)
1090                 return ret;
1091         FOR_EACH_DISK_STORAGE_COLUMN(i, t, cd) {
1092                 struct stat statbuf;
1093                 char *dirname = column_filename(t, i);
1094
1095                 ret = -E_OSL_NOMEM;
1096                 if (!dirname)
1097                         goto err;
1098                 /* check if directory exists */
1099                 ret = osl_stat(dirname, &statbuf);
1100                 free(dirname);
1101                 if (ret < 0)
1102                         goto err;
1103                 ret = -E_OSL_NOTDIR;
1104                 if (!S_ISDIR(statbuf.st_mode))
1105                         goto err;
1106         }
1107         ret = map_table(t, MAP_TBL_FL_VERIFY_INDEX);
1108         if (ret < 0)
1109                 goto err;
1110         t->num_rows = ret;
1111         DEBUG_LOG("num rows: %d\n", t->num_rows);
1112         ret = init_rbtrees(t);
1113         if (ret < 0) {
1114                 osl_close_table(t, OSL_MARK_CLEAN); /* ignore further errors */
1115                 return ret;
1116         }
1117         *result = t;
1118         return 1;
1119 err:
1120         free(t->columns);
1121         free(t);
1122         return ret;
1123 }
1124
1125 static int create_disk_storage_object_dir(const struct osl_table *t,
1126                 unsigned col_num, const char *ds_name)
1127 {
1128         char *dirname;
1129         int ret;
1130
1131         if (!(t->desc->flags & OSL_LARGE_TABLE))
1132                 return 1;
1133         dirname = disk_storage_dirname(t, col_num, ds_name);
1134         if (!dirname)
1135                 return -E_OSL_NOMEM;
1136         ret = osl_mkdir(dirname, 0777);
1137         free(dirname);
1138         if (ret < 0 && ret != -E_OSL_DIR_EXISTS)
1139                 return ret;
1140         return 1;
1141 }
1142
1143 static int write_disk_storage_file(const struct osl_table *t, unsigned col_num,
1144         const struct osl_object *obj, const char *ds_name)
1145 {
1146         int ret;
1147         char *filename;
1148
1149         ret = create_disk_storage_object_dir(t, col_num, ds_name);
1150         if (ret < 0)
1151                 return ret;
1152         filename = disk_storage_path(t, col_num, ds_name);
1153         if (!filename)
1154                 return -E_OSL_NOMEM;
1155         ret = write_file(filename, obj->data, obj->size);
1156         free(filename);
1157         return ret;
1158 }
1159
1160 static int append_map_file(const struct osl_table *t, unsigned col_num,
1161         const struct osl_object *obj, uint32_t *new_size)
1162 {
1163         char *filename = column_filename(t, col_num);
1164         int ret;
1165
1166         if (!filename)
1167                 return -E_OSL_NOMEM;
1168         ret = append_file(filename, obj->data, obj->size, new_size);
1169         free(filename);
1170         return ret;
1171 }
1172
1173 static int append_row_index(const struct osl_table *t, char *row_index)
1174 {
1175         char *filename;
1176         int ret;
1177
1178         if (!t->num_mapped_columns)
1179                 return 1;
1180         filename = index_filename(t->desc);
1181         if (!filename)
1182                 return -E_OSL_NOMEM;
1183         ret = append_file(filename, row_index, t->row_index_size, NULL);
1184         free(filename);
1185         return ret;
1186 }
1187
1188 static int truncate_mapped_file(const struct osl_table *t, unsigned col_num,
1189                 off_t size)
1190 {
1191         int ret;
1192         char *filename = column_filename(t, col_num);
1193
1194         if (!filename)
1195                 return -E_OSL_NOMEM;
1196         ret = truncate_file(filename, size);
1197         free(filename);
1198         return ret;
1199 }
1200
1201 static int delete_disk_storage_file(const struct osl_table *t, unsigned col_num,
1202                 const char *ds_name)
1203 {
1204         char *dirname, *filename = disk_storage_path(t, col_num, ds_name);
1205         int ret = 1;
1206
1207         if (!filename)
1208                 return -E_OSL_NOMEM;
1209         if (unlink(filename) < 0)
1210                 ret = errno == ENOENT? -E_OSL_NOENT : -E_OSL_UNLINK;
1211         free(filename);
1212         if (ret < 0)
1213                 return ret;
1214         if (!(t->desc->flags & OSL_LARGE_TABLE))
1215                 return 1;
1216         dirname = disk_storage_dirname(t, col_num, ds_name);
1217         if (!dirname)
1218                 return -E_OSL_NOMEM;
1219         rmdir(dirname);
1220         free(dirname);
1221         return 1;
1222 }
1223
1224 __export int osl_add_and_get_row(struct osl_table *t, struct osl_object *objects,
1225                 struct osl_row **row)
1226 {
1227         int i, ret;
1228         char *ds_name = NULL;
1229         struct rb_node **rb_parents = NULL, ***rb_links = NULL;
1230         char *new_row_index = NULL;
1231         struct osl_object *volatile_objs = NULL;
1232         const struct osl_column_description *cd;
1233
1234         if (!t)
1235                 return -E_OSL_BAD_TABLE;
1236         rb_parents = malloc(t->num_rbtrees * sizeof(struct rn_node*));
1237         if (!rb_parents)
1238                 return -E_OSL_NOMEM;
1239         rb_links = malloc(t->num_rbtrees * sizeof(struct rn_node**));
1240         if (!rb_links) {
1241                 free(rb_parents);
1242                 return -E_OSL_NOMEM;
1243         }
1244         if (t->num_mapped_columns) {
1245                 new_row_index = malloc(t->row_index_size);
1246                 if (!new_row_index) {
1247                         free(rb_links);
1248                         free(rb_parents);
1249                         return -E_OSL_NOMEM;
1250                 }
1251         }
1252         /* pass 1: sanity checks */
1253 //      DEBUG_LOG("sanity tests: %p:%p\n", objects[0].data,
1254 //              objects[1].data);
1255         FOR_EACH_COLUMN(i, t->desc, cd) {
1256                 enum osl_storage_type st = cd->storage_type;
1257                 enum osl_storage_flags sf = cd->storage_flags;
1258
1259 //              ret = -E_OSL_NULL_OBJECT;
1260 //              if (!objects[i])
1261 //                      goto out;
1262                 if (st == OSL_DISK_STORAGE)
1263                         continue;
1264                 if (sf & OSL_RBTREE) {
1265                         unsigned rbtree_num = t->columns[i].rbtree_num;
1266                         ret = -E_OSL_RB_KEY_EXISTS;
1267 //                      DEBUG_LOG("checking whether %p exists\n",
1268 //                              objects[i].data);
1269                         if (search_rbtree(objects + i, t, i,
1270                                         &rb_parents[rbtree_num],
1271                                         &rb_links[rbtree_num]) > 0)
1272                                 goto out;
1273                 }
1274                 if (sf & OSL_FIXED_SIZE) {
1275 //                      DEBUG_LOG("fixed size. need: %zu, have: %d\n",
1276 //                              objects[i].size, cd->data_size);
1277                         ret = -E_OSL_BAD_DATA_SIZE;
1278                         if (objects[i].size != cd->data_size)
1279                                 goto out;
1280                 }
1281         }
1282         if (t->num_disk_storage_columns) {
1283                 ds_name = disk_storage_name_of_object(t,
1284                         &objects[t->disk_storage_name_column]);
1285                 ret = -E_OSL_NOMEM;
1286                 if (!ds_name)
1287                         goto out;
1288         }
1289         ret = unmap_table(t, OSL_MARK_CLEAN);
1290         if (ret < 0)
1291                 goto out;
1292 //      DEBUG_LOG("sanity tests passed%s\n", "");
1293         /* pass 2: create data files, append map data */
1294         FOR_EACH_COLUMN(i, t->desc, cd) {
1295                 enum osl_storage_type st = cd->storage_type;
1296                 if (st == OSL_NO_STORAGE)
1297                         continue;
1298                 if (st == OSL_MAPPED_STORAGE) {
1299                         uint32_t new_size;
1300                         struct osl_column *col = &t->columns[i];
1301 //                      DEBUG_LOG("appending object of size %zu\n",
1302 //                              objects[i].size);
1303                         ret = append_map_file(t, i, objects + i, &new_size);
1304                         if (ret < 0)
1305                                 goto rollback;
1306                         update_cell_index(new_row_index, col, new_size,
1307                                 objects[i].size);
1308                         continue;
1309                 }
1310                 /* DISK_STORAGE */
1311                 ret = write_disk_storage_file(t, i, objects + i, ds_name);
1312                 if (ret < 0)
1313                         goto rollback;
1314         }
1315         ret = append_row_index(t, new_row_index);
1316         if (ret < 0)
1317                 goto rollback;
1318         ret = map_table(t, MAP_TBL_FL_VERIFY_INDEX);
1319         if (ret < 0) { /* truncate index and rollback changes */
1320                 char *filename = index_filename(t->desc);
1321                 if (filename)
1322                         truncate_file(filename, t->row_index_size);
1323                 free(filename);
1324                 goto rollback;
1325         }
1326         /* pass 3: add entry to rbtrees */
1327         if (t->num_volatile_columns) {
1328                 ret = -E_OSL_NOMEM;
1329                 volatile_objs = calloc(t->num_volatile_columns,
1330                         sizeof(struct osl_object));
1331                 if (!volatile_objs)
1332                         goto out;
1333                 FOR_EACH_VOLATILE_COLUMN(i, t, cd)
1334                         volatile_objs[t->columns[i].volatile_num] = objects[i];
1335         }
1336         t->num_rows++;
1337 //      DEBUG_LOG("adding new entry as row #%d\n", t->num_rows - 1);
1338         ret = add_row_to_rbtrees(t, t->num_rows - 1, volatile_objs, row);
1339         if (ret < 0)
1340                 goto out;
1341 //      DEBUG_LOG("added new entry as row #%d\n", t->num_rows - 1);
1342         ret = 1;
1343         goto out;
1344 rollback: /* rollback all changes made, ignore further errors */
1345         for (i--; i >= 0; i--) {
1346                 cd = get_column_description(t->desc, i);
1347                 enum osl_storage_type st = cd->storage_type;
1348                 if (st == OSL_NO_STORAGE)
1349                         continue;
1350
1351                 if (st == OSL_MAPPED_STORAGE)
1352                         truncate_mapped_file(t, i, objects[i].size);
1353                 else /* disk storage */
1354                         delete_disk_storage_file(t, i, ds_name);
1355         }
1356         /* ignore error and return previous error value */
1357         map_table(t, MAP_TBL_FL_VERIFY_INDEX);
1358 out:
1359         free(new_row_index);
1360         free(ds_name);
1361         free(rb_parents);
1362         free(rb_links);
1363         return ret;
1364 }
1365
1366 __export int osl_add_row(struct osl_table *t, struct osl_object *objects)
1367 {
1368         return osl_add_and_get_row(t, objects, NULL);
1369 }
1370
1371 __export int osl_get_object(const struct osl_table *t, const struct osl_row *r,
1372         unsigned col_num, struct osl_object *object)
1373 {
1374         const struct osl_column_description *cd;
1375
1376         if (!t)
1377                 return -E_OSL_BAD_TABLE;
1378         cd = get_column_description(t->desc, col_num);
1379         /* col must not be disk storage */
1380         if (cd->storage_type == OSL_DISK_STORAGE)
1381                 return -E_OSL_BAD_STORAGE_TYPE;
1382         if (cd->storage_type == OSL_MAPPED_STORAGE)
1383                 return get_mapped_object(t, col_num, r->num, object);
1384         /* volatile */
1385         *object = r->volatile_objects[t->columns[col_num].volatile_num];
1386         return 1;
1387 }
1388
1389 __export int osl_del_row(struct osl_table *t, struct osl_row *row)
1390 {
1391         struct osl_row *r = row;
1392         int i, ret;
1393         const struct osl_column_description *cd;
1394
1395         if (!t)
1396                 return -E_OSL_BAD_TABLE;
1397         INFO_LOG("deleting row %p\n", row);
1398
1399         if (t->num_disk_storage_columns) {
1400                 char *ds_name;
1401                 ret = disk_storage_name_of_row(t, r, &ds_name);
1402                 if (ret < 0)
1403                         goto out;
1404                 FOR_EACH_DISK_STORAGE_COLUMN(i, t, cd)
1405                         delete_disk_storage_file(t, i, ds_name);
1406                 free(ds_name);
1407         }
1408         FOR_EACH_COLUMN(i, t->desc, cd) {
1409                 struct osl_column *col = t->columns + i;
1410                 enum osl_storage_type st = cd->storage_type;
1411                 remove_rb_node(t, i, r);
1412                 if (st == OSL_MAPPED_STORAGE)
1413                         continue;
1414                 if (st == OSL_NO_STORAGE && !(cd->storage_flags & OSL_DONT_FREE))
1415                         free(r->volatile_objects[col->volatile_num].data);
1416         }
1417         if (t->num_mapped_columns) {
1418                 ret = mark_row_invalid(t, r->num);
1419                 if (ret < 0)
1420                         goto out;
1421                 t->num_invalid_rows++;
1422         } else
1423                 t->num_rows--;
1424         ret = 1;
1425 out:
1426         free(r->volatile_objects);
1427         free(r);
1428         return ret;
1429 }
1430
1431 /* test if column has an rbtree */
1432 static int check_rbtree_col(const struct osl_table *t, unsigned col_num,
1433                 struct osl_column **col)
1434 {
1435         if (!t)
1436                 return -E_OSL_BAD_TABLE;
1437         if (!(get_column_description(t->desc, col_num)->storage_flags & OSL_RBTREE))
1438                 return -E_OSL_BAD_STORAGE_FLAGS;
1439         *col = t->columns + col_num;
1440         return 1;
1441 }
1442
1443 __export int osl_get_row(const struct osl_table *t, unsigned col_num,
1444                 const struct osl_object *obj, struct osl_row **result)
1445 {
1446         int ret;
1447         struct rb_node *node;
1448         struct osl_row *row;
1449         struct osl_column *col;
1450
1451         *result = NULL;
1452         ret = check_rbtree_col(t, col_num, &col);
1453         if (ret < 0)
1454                 return ret;
1455         ret = search_rbtree(obj, t, col_num, &node, NULL);
1456         if (ret < 0)
1457                 return ret;
1458         row = get_row_pointer(node, t->columns[col_num].rbtree_num);
1459         *result = row;
1460         return 1;
1461 }
1462
1463 static int rbtree_loop(struct osl_column *col, void *private_data,
1464                 osl_rbtree_loop_func *func)
1465 {
1466         struct rb_node *n, *tmp;
1467
1468         /* this for-loop is safe against removal of an entry */
1469         for (n = rb_first(&col->rbtree), tmp = n? rb_next(n) : NULL;
1470                         n;
1471                         n = tmp, tmp = tmp? rb_next(tmp) : NULL) {
1472                 struct osl_row *r = get_row_pointer(n, col->rbtree_num);
1473                 if (func(r, private_data) < 0)
1474                         return -E_OSL_LOOP;
1475         }
1476         return 1;
1477 }
1478
1479 static int rbtree_loop_reverse(struct osl_column *col, void *private_data,
1480                 osl_rbtree_loop_func *func)
1481 {
1482         struct rb_node *n, *tmp;
1483
1484         /* safe against removal of an entry */
1485         for (n = rb_last(&col->rbtree), tmp = n? rb_prev(n) : NULL;
1486                         n;
1487                         n = tmp, tmp = tmp? rb_prev(tmp) : NULL) {
1488                 struct osl_row *r = get_row_pointer(n, col->rbtree_num);
1489                 if (func(r, private_data) < 0)
1490                         return -E_OSL_LOOP;
1491         }
1492         return 1;
1493 }
1494
1495 __export int osl_rbtree_loop(const struct osl_table *t, unsigned col_num,
1496         void *private_data, osl_rbtree_loop_func *func)
1497 {
1498         struct osl_column *col;
1499
1500         int ret = check_rbtree_col(t, col_num, &col);
1501         if (ret < 0)
1502                 return ret;
1503         return rbtree_loop(col, private_data, func);
1504 }
1505
1506 __export int osl_rbtree_loop_reverse(const struct osl_table *t, unsigned col_num,
1507         void *private_data, osl_rbtree_loop_func *func)
1508 {
1509         struct osl_column *col;
1510
1511         int ret = check_rbtree_col(t, col_num, &col);
1512         if (ret < 0)
1513                 return ret;
1514         return rbtree_loop_reverse(col, private_data, func);
1515 }
1516
1517 /* TODO: Rollback changes on errors */
1518 static int rename_disk_storage_objects(struct osl_table *t,
1519                 struct osl_object *old_obj, struct osl_object *new_obj)
1520 {
1521         int i, ret;
1522         const struct osl_column_description *cd;
1523         char *old_ds_name, *new_ds_name;
1524
1525         if (!t->num_disk_storage_columns)
1526                 return 1; /* nothing to do */
1527         if (old_obj->size == new_obj->size && !memcmp(new_obj->data,
1528                         old_obj->data, new_obj->size))
1529                 return 1; /* object did not change */
1530         old_ds_name = disk_storage_name_of_object(t, old_obj);
1531         new_ds_name = disk_storage_name_of_object(t, new_obj);
1532         ret = -E_OSL_NOMEM;
1533         if (!old_ds_name || ! new_ds_name)
1534                 goto out;
1535
1536         FOR_EACH_DISK_STORAGE_COLUMN(i, t, cd) {
1537                 char *old_filename, *new_filename;
1538                 ret = create_disk_storage_object_dir(t, i, new_ds_name);
1539                 if (ret < 0)
1540                         goto out;
1541                 old_filename = disk_storage_path(t, i, old_ds_name);
1542                 new_filename = disk_storage_path(t, i, new_ds_name);
1543                 if (!old_filename || !new_filename)
1544                         ret = -E_OSL_NOMEM;
1545                 else
1546                         ret = osl_rename(old_filename, new_filename);
1547                 free(old_filename);
1548                 free(new_filename);
1549                 if (ret < 0)
1550                         goto out;
1551         }
1552         ret = 1;
1553 out:
1554         free(old_ds_name);
1555         free(new_ds_name);
1556         return ret;
1557
1558 }
1559
1560 __export int osl_update_object(struct osl_table *t, const struct osl_row *r,
1561                 unsigned col_num, struct osl_object *obj)
1562 {
1563         struct osl_column *col;
1564         const struct osl_column_description *cd;
1565         int ret;
1566
1567         if (!t)
1568                 return -E_OSL_BAD_TABLE;
1569         col = &t->columns[col_num];
1570         cd = get_column_description(t->desc, col_num);
1571         DEBUG_LOG("updating column %u of %s\n", col_num, t->desc->name);
1572         if (cd->storage_flags & OSL_RBTREE) {
1573                 if (search_rbtree(obj, t, col_num, NULL, NULL) > 0)
1574                         return -E_OSL_RB_KEY_EXISTS;
1575         }
1576         if (cd->storage_flags & OSL_FIXED_SIZE) {
1577                 if (obj->size != cd->data_size)
1578                         return -E_OSL_BAD_DATA_SIZE;
1579         }
1580         remove_rb_node(t, col_num, r);
1581         if (cd->storage_type == OSL_NO_STORAGE) { /* TODO: If fixed size, reuse object? */
1582                 if (!(cd->storage_flags & OSL_DONT_FREE))
1583                         free(r->volatile_objects[col->volatile_num].data);
1584                 r->volatile_objects[col->volatile_num] = *obj;
1585         } else if (cd->storage_type == OSL_DISK_STORAGE) {
1586                 char *ds_name;
1587                 ret = disk_storage_name_of_row(t, r, &ds_name);
1588                 if (ret < 0)
1589                         return ret;
1590                 ret = delete_disk_storage_file(t, col_num, ds_name);
1591                 if (ret < 0 && ret != -E_OSL_NOENT) {
1592                         free(ds_name);
1593                         return ret;
1594                 }
1595                 ret = write_disk_storage_file(t, col_num, obj, ds_name);
1596                 free(ds_name);
1597                 if (ret < 0)
1598                         return ret;
1599         } else { /* mapped storage */
1600                 struct osl_object old_obj;
1601                 ret = get_mapped_object(t, col_num, r->num, &old_obj);
1602                 if (ret < 0)
1603                         return ret;
1604                 /*
1605                  * If the updated column is the disk storage name column, the
1606                  * disk storage name changes, so we have to rename all disk
1607                  * storage objects accordingly.
1608                  */
1609                 if (col_num == t->disk_storage_name_column) {
1610                         ret = rename_disk_storage_objects(t, &old_obj, obj);
1611                         if (ret < 0)
1612                                 return ret;
1613                 }
1614                 if (cd->storage_flags & OSL_FIXED_SIZE)
1615                         memcpy(old_obj.data, obj->data, cd->data_size);
1616                 else { /* TODO: if the size doesn't change, use old space */
1617                         uint32_t new_data_map_size;
1618                         char *row_index;
1619                         ret = get_row_index(t, r->num, &row_index);
1620                         if (ret < 0)
1621                                 return ret;
1622                         unmap_column(t, col_num);
1623                         ret = append_map_file(t, col_num, obj,
1624                                 &new_data_map_size);
1625                         if (ret < 0)
1626                                 return ret;
1627                         ret = map_column(t, col_num);
1628                         if (ret < 0)
1629                                 return ret;
1630                         update_cell_index(row_index, col, new_data_map_size,
1631                                 obj->size);
1632                 }
1633         }
1634         if (cd->storage_flags & OSL_RBTREE) {
1635                 ret = insert_rbtree(t, col_num, r, obj);
1636                 if (ret < 0)
1637                         return ret;
1638         }
1639         return 1;
1640 }
1641
1642 __export int osl_open_disk_object(const struct osl_table *t, const struct osl_row *r,
1643                 unsigned col_num, struct osl_object *obj)
1644 {
1645         const struct osl_column_description *cd;
1646         char *ds_name, *filename;
1647         int ret;
1648
1649         if (!t)
1650                 return -E_OSL_BAD_TABLE;
1651         cd = get_column_description(t->desc, col_num);
1652         if (cd->storage_type != OSL_DISK_STORAGE)
1653                 return -E_OSL_BAD_STORAGE_TYPE;
1654
1655         ret = disk_storage_name_of_row(t, r, &ds_name);
1656         if (ret < 0)
1657                 return ret;
1658         filename = disk_storage_path(t, col_num, ds_name);
1659         free(ds_name);
1660         if (!filename)
1661                 return -E_OSL_NOMEM;
1662         DEBUG_LOG("filename: %s\n", filename);
1663         ret = mmap_full_file(filename, O_RDONLY, &obj->data, &obj->size, NULL);
1664         free(filename);
1665         return ret;
1666 }
1667
1668 __export int osl_close_disk_object(struct osl_object *obj)
1669 {
1670         return osl_munmap(obj->data, obj->size);
1671 }
1672
1673 __export int osl_get_num_rows(const struct osl_table *t, unsigned *num_rows)
1674 {
1675         if (!t)
1676                 return -E_OSL_BAD_TABLE;
1677         assert(t->num_rows >= t->num_invalid_rows);
1678         *num_rows = t->num_rows - t->num_invalid_rows;
1679         return 1;
1680 }
1681
1682 __export int osl_get_rank(const struct osl_table *t, struct osl_row *r,
1683                 unsigned col_num, unsigned *rank)
1684 {
1685         struct osl_object obj;
1686         struct osl_column *col;
1687         struct rb_node *node;
1688         int ret = check_rbtree_col(t, col_num, &col);
1689
1690         if (ret < 0)
1691                 return ret;
1692         ret = osl_get_object(t, r, col_num, &obj);
1693         if (ret < 0)
1694                 return ret;
1695         ret = search_rbtree(&obj, t, col_num, &node, NULL);
1696         if (ret < 0)
1697                 return ret;
1698         ret = rb_rank(node, rank);
1699         if (ret < 0)
1700                 return -E_OSL_BAD_ROW;
1701         return 1;
1702 }
1703
1704 __export int osl_get_nth_row(const struct osl_table *t, unsigned col_num,
1705                 unsigned n, struct osl_row **result)
1706 {
1707         struct osl_column *col;
1708         struct rb_node *node;
1709         unsigned num_rows;
1710         int ret;
1711
1712         *result = NULL;
1713         if (n == 0)
1714                 return -E_OSL_RB_KEY_NOT_FOUND;
1715         ret = osl_get_num_rows(t, &num_rows);
1716         if (ret < 0)
1717                 return ret;
1718         if (n > num_rows)
1719                 return -E_OSL_RB_KEY_NOT_FOUND;
1720         ret = check_rbtree_col(t, col_num, &col);
1721         if (ret < 0)
1722                 return ret;
1723         node = rb_nth(col->rbtree.rb_node, n);
1724         if (!node)
1725                 return -E_OSL_RB_KEY_NOT_FOUND;
1726         *result = get_row_pointer(node, col->rbtree_num);
1727         return 1;
1728 }
1729
1730 __export int osl_rbtree_first_row(const struct osl_table *t, unsigned col_num,
1731                 struct osl_row **result)
1732 {
1733         return osl_get_nth_row(t, col_num, 1, result);
1734 }
1735
1736 __export int osl_rbtree_last_row(const struct osl_table *t, unsigned col_num,
1737                 struct osl_row **result)
1738 {
1739         unsigned num_rows;
1740         int ret = osl_get_num_rows(t, &num_rows);
1741
1742         if (ret < 0)
1743                 return ret;
1744         return osl_get_nth_row(t, col_num, num_rows, result);
1745 }