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