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