osl.c

   1 /*
   2  * Copyright (C) 2007-2009 Andre Noll <maan@tuebingen.mpg.de>
   3  *
   4  * Licensed under the GPL v2. For licencing details see COPYING.
   5  */
   6
   7 /** \file osl.c Object storage layer functions. */
   8 #include <dirent.h> /* readdir() */
   9 #include <assert.h>
  10
  11 #include "log.h"
  12 #include "osl.h"
  13 #include "util.h"
  14 #include "osl_core.h"
  15
  16 /*
  17  * Taken from Drepper: How to write shared libraries, Appendix B.
  18  *
  19  * The main reason for this rather fancy implementation of strerror() is to
  20  * avoid having an array of pointers. This is desirable because initialized
  21  * pointer variables increase the startup time of the library due to the
  22  * processing of relocations.
  23  */
  24 #include <stddef.h>
  25 #define MSGSTRFIELD(line) MSGSTRFIELD1(line)
  26 #define MSGSTRFIELD1(line) str##line
  27 static const union msgstr_t {
  28         struct {
  29 #define OSL_ERROR(n, s) char MSGSTRFIELD(__LINE__)[sizeof(s)];
  30 #include "errtab.h"
  31 #undef OSL_ERROR
  32         };
  33         char str[0];
  34 } msgstr = { {
  35 #define OSL_ERROR(n, s) s,
  36 #include "errtab.h"
  37 #undef OSL_ERROR
  38 } };
  39 static const unsigned int errmsgidx[] = {
  40 #define OSL_ERROR(n, s) [n] = offsetof(union msgstr_t, MSGSTRFIELD(__LINE__)),
  41 #include "errtab.h"
  42 #undef OSL_ERROR
  43 };
  44
  45 __export const char *osl_strerror(int num)
  46 {
  47         return msgstr.str + errmsgidx[num];
  48 }
  49
  50 int loglevel = 0;
  51
  52 static void __attribute ((constructor)) init_loglevel(void)
  53 {
  54         char *p = getenv("OSL_LOGLEVEL");
  55
  56         /* don't log anything if unset */
  57         loglevel = p? atoi(p) : EMERG + 1;
  58 }
  59
  60 /**
  61  * The log function.
  62  *
  63  * \param ll Loglevel.
  64  * \param fmt Usual format string.
  65  *
  66  * All XXX_LOG() macros use this function.
  67  */
  68 __printf_2_3 void __log(int ll, const char* fmt,...)
  69 {
  70         va_list argp;
  71         struct tm *tm;
  72         time_t t1;
  73         char str[255] = "";
  74
  75         if (ll < loglevel)
  76                 return;
  77         time(&t1);
  78         tm = localtime(&t1);
  79         strftime(str, sizeof(str), "%b %d %H:%M:%S", tm);
  80         fprintf(stderr, "%s ", str);
  81         va_start(argp, fmt);
  82         vfprintf(stderr, fmt, argp);
  83         va_end(argp);
  84 }
  85
  86 /**
  87  * A wrapper for lseek(2).
  88  *
  89  * \param fd The file descriptor whose offset is to be to repositioned.
  90  * \param offset A value-result parameter.
  91  * \param whence Usual repositioning directive.
  92  *
  93  * Reposition the offset of the file descriptor \a fd to the argument \a offset
  94  * according to the directive \a whence. Upon successful return, \a offset
  95  * contains the resulting offset location as measured in bytes from the
  96  * beginning of the file.
  97  *
  98  * \return Positive on success. Otherwise, the function returns \p -E_OSL_LSEEK.
  99  *
 100  * \sa lseek(2).
 101  */
 102 static int __lseek(int fd, off_t *offset, int whence)
 103 {
 104         *offset = lseek(fd, *offset, whence);
 105         int ret = -E_OSL_LSEEK;
 106         if (*offset == -1)
 107                 return ret;
 108         return 1;
 109 }
 110
 111 static int append_file(const char *filename, char *data, size_t data_size,
 112                 uint32_t *new_pos)
 113 {
 114         int ret, fd;
 115
 116 //      DEBUG_LOG("appending %zu  + %zu bytes\n", header_size, data_size);
 117         ret = osl_open(filename, O_WRONLY | O_CREAT | O_APPEND, 0644);
 118         if (ret < 0)
 119                 return ret;
 120         fd = ret;
 121         ret = write_all(fd, data, &data_size);
 122         if (ret < 0)
 123                 goto out;
 124         if (new_pos) {
 125                 off_t offset = 0;
 126                 ret = __lseek(fd, &offset, SEEK_END);
 127                 if (ret < 0)
 128                         goto out;
 129 //              DEBUG_LOG("new file size: " FMT_OFF_T "\n", offset);
 130                 *new_pos = offset;
 131         }
 132         ret = 1;
 133 out:
 134         close(fd);
 135         return ret;
 136 }
 137
 138 static int verify_name(const char *name)
 139 {
 140         if (!name)
 141                 return -E_OSL_BAD_NAME;
 142         if (!*name)
 143                 return -E_OSL_BAD_NAME;
 144         if (strchr(name, '/'))
 145                 return -E_OSL_BAD_NAME;
 146         if (!strcmp(name, ".."))
 147                 return -E_OSL_BAD_NAME;
 148         if (!strcmp(name, "."))
 149                 return -E_OSL_BAD_NAME;
 150         return 1;
 151 }
 152
 153 static char *disk_storage_dirname(const struct osl_table *t, unsigned col_num,
 154                 const char *ds_name)
 155 {
 156         char *dirname, *column_name = column_filename(t, col_num);
 157
 158         if (!column_name)
 159                 return NULL;
 160         if (!(t->desc->flags & OSL_LARGE_TABLE))
 161                 return column_name;
 162         dirname = make_message("%s/%.2s", column_name, ds_name);
 163         free(column_name);
 164         return dirname;
 165 }
 166
 167 static char *disk_storage_name_of_object(const struct osl_table *t,
 168         const struct osl_object *obj)
 169 {
 170         HASH_TYPE hash[HASH_SIZE];
 171         hash_object(t, obj, hash);
 172         return disk_storage_name_of_hash(t, hash);
 173 }
 174
 175 static int disk_storage_name_of_row(const struct osl_table *t,
 176                 const struct osl_row *row, char **name)
 177 {
 178         struct osl_object obj;
 179         int ret = osl_get_object(t, row, t->disk_storage_name_column, &obj);
 180
 181         if (ret < 0)
 182                 return ret;
 183         *name = disk_storage_name_of_object(t, &obj);
 184         if (*name)
 185                 return 1;
 186         return -E_OSL_NOMEM;
 187 }
 188
 189 static void column_name_hash(const struct osl_table *t, const char *col_name,
 190                 HASH_TYPE *hash)
 191 {
 192         hash_function(t->version, col_name, strlen(col_name), hash);
 193 }
 194
 195 static int init_column_descriptions(struct osl_table *t)
 196 {
 197         int i, j, ret;
 198         const struct osl_column_description *cd;
 199
 200         ret = verify_name(t->desc->name);
 201         if (ret < 0)
 202                 goto err;
 203         ret = -E_OSL_BAD_DB_DIR;
 204         if (!t->desc->dir && (t->num_disk_storage_columns || t->num_mapped_columns))
 205                 goto err;
 206         /* the size of the index header without column descriptions */
 207         t->index_header_size = IDX_COLUMN_DESCRIPTIONS;
 208         FOR_EACH_COLUMN(i, t->desc, cd) {
 209                 if (cd->storage_flags & OSL_RBTREE) {
 210                         if (!cd->compare_function)
 211                                 return -E_OSL_NO_COMPARE_FUNC;
 212                 }
 213                 if (cd->storage_type == OSL_NO_STORAGE)
 214                         continue;
 215                 ret = -E_OSL_NO_COLUMN_NAME;
 216                 if (!cd->name || !cd->name[0])
 217                         goto err;
 218                 ret = verify_name(cd->name);
 219                 if (ret < 0)
 220                         goto err;
 221                 t->index_header_size += index_column_description_size(cd->name);
 222                 ret = -E_OSL_DUPLICATE_COL_NAME;
 223                 for (j = i + 1; j < t->desc->num_columns; j++) {
 224                         const char *name2 = get_column_description(t->desc,
 225                                 j)->name;
 226                         if (cd->name && name2 && !strcmp(cd->name, name2))
 227                                 goto err;
 228                 }
 229         }
 230         return 1;
 231 err:
 232         return ret;
 233 }
 234
 235 /**
 236  * Initialize a struct table from given table description.
 237  *
 238  * \param desc The description of the osl table.
 239  * \param table_ptr Result is returned here.
 240  *
 241  * This function performs several sanity checks on \p desc and returns if any
 242  * of these tests fail. On success, a struct \p osl_table is allocated and
 243  * initialized with data derived from \p desc.
 244  *
 245  * \return Standard.
 246  *
 247  * \sa struct osl_table.
 248  */
 249 int init_table_structure(const struct osl_table_description *desc,
 250                 struct osl_table **table_ptr)
 251 {
 252         const struct osl_column_description *cd;
 253         struct osl_table *t = calloc(1, sizeof(*t));
 254         int i, ret = -E_OSL_NOMEM, have_disk_storage_name_column = 0;
 255
 256         if (!t)
 257                 return ret;
 258         ret = -E_OSL_BAD_TABLE_DESC;
 259         if (!desc)
 260                 goto err;
 261         DEBUG_LOG("creating table structure for '%s' from table "
 262                 "description\n", desc->name);
 263         ret = -E_OSL_NO_COLUMN_DESC;
 264         if (!desc->column_descriptions)
 265                 goto err;
 266         ret = -E_OSL_NO_COLUMNS;
 267         if (!desc->num_columns)
 268                 goto err;
 269         ret = -E_OSL_NOMEM;
 270         t->columns = calloc(desc->num_columns, sizeof(struct osl_column));
 271         if (!t->columns)
 272                 goto err;
 273         t->desc = desc;
 274         FOR_EACH_COLUMN(i, t->desc, cd) {
 275                 enum osl_storage_type st = cd->storage_type;
 276                 enum osl_storage_flags sf = cd->storage_flags;
 277                 struct osl_column *col = &t->columns[i];
 278
 279                 ret = -E_OSL_BAD_STORAGE_TYPE;
 280                 if (st != OSL_MAPPED_STORAGE && st != OSL_DISK_STORAGE
 281                                 && st != OSL_NO_STORAGE)
 282                         goto err;
 283                 ret = -E_OSL_BAD_STORAGE_FLAGS;
 284                 if (st == OSL_DISK_STORAGE && sf & OSL_RBTREE)
 285                         goto err;
 286                 if ((sf & OSL_RBTREE) && !(sf & OSL_UNIQUE))
 287                         WARNING_LOG("invalid storage flags for column %s: "
 288                                 "OSL_RBTREE && !OSL_UNIQUE\n", cd->name);
 289                 ret = -E_OSL_BAD_STORAGE_SIZE;
 290                 if (sf & OSL_FIXED_SIZE && !cd->data_size)
 291                         goto err;
 292                 switch (st) {
 293                 case OSL_DISK_STORAGE:
 294                         t->num_disk_storage_columns++;
 295                         break;
 296                 case OSL_MAPPED_STORAGE:
 297                         t->num_mapped_columns++;
 298                         col->index_offset = t->row_index_size;
 299                         t->row_index_size += 8;
 300                         break;
 301                 case OSL_NO_STORAGE:
 302                         col->volatile_num = t->num_volatile_columns;
 303                         t->num_volatile_columns++;
 304                         break;
 305                 }
 306                 if (sf & OSL_RBTREE) {
 307                         col->rbtree_num = t->num_rbtrees;
 308                         t->num_rbtrees++;
 309                         if ((sf & OSL_UNIQUE) && (st == OSL_MAPPED_STORAGE)) {
 310                                 if (!have_disk_storage_name_column)
 311                                         t->disk_storage_name_column = i;
 312                                 have_disk_storage_name_column = 1;
 313                         }
 314                 }
 315         }
 316         ret = -E_OSL_NO_UNIQUE_RBTREE_COLUMN;
 317         if (t->num_disk_storage_columns && !have_disk_storage_name_column)
 318                 goto err;
 319         ret = -E_OSL_NO_RBTREE_COL;
 320         if (!t->num_rbtrees)
 321                 goto err;
 322         /* success */
 323         DEBUG_LOG("OK. Index entry size: %u\n", t->row_index_size);
 324         ret = init_column_descriptions(t);
 325         if (ret < 0)
 326                 goto err;
 327         *table_ptr = t;
 328         return 1;
 329 err:
 330         free(t->columns);
 331         free(t);
 332         return ret;
 333 }
 334
 335 /**
 336  * Read the table description from index header.
 337  *
 338  * \param map The memory mapping of the index file.
 339  * \param desc The values found in the index header are returned here.
 340  *
 341  * Read the index header, check for the osl magic string and the table version
 342  * number.  Read all information stored in the index header into \a desc.
 343  *
 344  * \return The on-disk table version number on success, negative on errors.
 345  *
 346  * \sa struct osl_table_description, osl_create_table.
 347  */
 348 int read_table_desc(struct osl_object *map, struct osl_table_description *desc)
 349 {
 350         char *buf = map->data;
 351         uint8_t table_version;
 352         uint16_t header_size;
 353         int ret, i;
 354         unsigned offset;
 355         struct osl_column_description *cd;
 356
 357         if (map->size < MIN_INDEX_HEADER_SIZE(1))
 358                 return -E_OSL_SHORT_TABLE;
 359         if (strncmp(buf + IDX_OSL_MAGIC, OSL_MAGIC, strlen(OSL_MAGIC)))
 360                 return -E_OSL_NO_MAGIC;
 361         table_version = read_u8(buf + IDX_VERSION);
 362         INFO_LOG("osl versions (table/min/current): %u/%u/%u\n",
 363                 table_version, MIN_TABLE_VERSION, CURRENT_TABLE_VERSION);
 364         if (table_version < MIN_TABLE_VERSION /* table too old */
 365                 || table_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 table_version;
 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         t->version = ret;
 432         ret = -E_OSL_BAD_TABLE_FLAGS;
 433         if (desc.flags != t->desc->flags)
 434                 goto out;
 435         ret = -E_OSL_BAD_COLUMN_NUM;
 436         if (desc.num_columns > t->desc->num_columns)
 437                 goto out;
 438         if (desc.num_columns < t->desc->num_columns) {
 439                 struct osl_column_description *cd;
 440                 unsigned diff = t->desc->num_columns - desc.num_columns;
 441                 INFO_LOG("extending table by %u volatile columns\n", diff);
 442                 ret = -E_OSL_NOMEM;
 443                 desc.column_descriptions = realloc(desc.column_descriptions,
 444                         t->desc->num_columns * sizeof(struct osl_column_description));
 445                 if (!desc.column_descriptions)
 446                         goto out;
 447                 for (i = desc.num_columns; i < t->desc->num_columns; i++) {
 448                         cd = get_column_description(&desc, i);
 449                         cd->storage_type = OSL_NO_STORAGE;
 450                         cd->name = NULL;
 451                 }
 452                 desc.num_columns += diff;
 453         }
 454         FOR_EACH_COLUMN(i, t->desc, cd1) {
 455                 cd2 = get_column_description(&desc, i);
 456                 ret = -E_OSL_BAD_STORAGE_TYPE;
 457                 if (cd1->storage_type != cd2->storage_type)
 458                         goto out;
 459                 if (cd1->storage_type == OSL_NO_STORAGE)
 460                         continue;
 461                 ret = -E_OSL_BAD_STORAGE_FLAGS;
 462                 if (cd1->storage_flags != cd2->storage_flags) {
 463                         ERROR_LOG("sf1 = %u != %u = sf2\n",
 464                                 cd1->storage_flags, cd2->storage_flags);
 465                         goto out;
 466                 }
 467                 ret = -E_OSL_BAD_DATA_SIZE;
 468                 if (cd1->storage_flags & OSL_FIXED_SIZE)
 469                         if (cd1->data_size != cd2->data_size)
 470                                 goto out;
 471                 ret = -E_OSL_BAD_COLUMN_NAME;
 472                 if (strcmp(cd1->name, cd2->name))
 473                         goto out;
 474         }
 475         INFO_LOG("table description of '%s' matches on-disk data, good\n",
 476                 t->desc->name);
 477         ret = 1;
 478 out:
 479         FOR_EACH_COLUMN(i, &desc, cd1)
 480                 free(cd1->name);
 481         free(desc.column_descriptions);
 482         return ret;
 483 }
 484
 485 static int create_table_index(struct osl_table *t)
 486 {
 487         char *buf, *filename;
 488         int i, ret;
 489         size_t size = t->index_header_size;
 490         const struct osl_column_description *cd;
 491         unsigned offset;
 492
 493         INFO_LOG("creating %zu byte index for table %s\n", size,
 494                 t->desc->name);
 495         buf = calloc(1, size);
 496         if (!buf)
 497                 return -E_OSL_NOMEM;
 498         sprintf(buf + IDX_OSL_MAGIC, "%s", OSL_MAGIC);
 499         write_u8(buf + IDX_TABLE_FLAGS, t->desc->flags);
 500         write_u8(buf + IDX_DIRTY_FLAG, 0);
 501         write_u8(buf + IDX_VERSION, t->version);
 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         t->version = CURRENT_TABLE_VERSION;
 540         INFO_LOG("creating version %u table %s\n", t->version, desc->name);
 541         FOR_EACH_COLUMN(i, t->desc, cd) {
 542                 if (cd->storage_type == OSL_NO_STORAGE)
 543                         continue;
 544                 if (!table_dir) {
 545                         ret = osl_mkdir(desc->dir, 0777);
 546                         if (ret < 0 && ret != -E_OSL_DIR_EXISTS)
 547                                 goto out;
 548                         table_dir = make_message("%s/%s", desc->dir,
 549                                 desc->name);
 550                         ret = -E_OSL_NOMEM;
 551                         if (!table_dir)
 552                                 goto out;
 553                         ret = osl_mkdir(table_dir, 0777);
 554                         if (ret < 0)
 555                                 goto out;
 556                 }
 557                 column_name_hash(t, cd->name, t->columns[i].name_hash);
 558                 ret = -E_OSL_NOMEM;
 559                 filename = column_filename(t, i);
 560                 if (!filename)
 561                         goto out;
 562                 INFO_LOG("filename: %s\n", filename);
 563                 if (cd->storage_type == OSL_MAPPED_STORAGE) {
 564                         ret = osl_open(filename, O_RDWR | O_CREAT | O_EXCL,
 565                                 0644);
 566                         free(filename);
 567                         if (ret < 0)
 568                                 goto out;
 569                         close(ret);
 570                         continue;
 571                 }
 572                 /* DISK STORAGE */
 573                 ret = osl_mkdir(filename, 0777);
 574                 free(filename);
 575                 if (ret < 0)
 576                         goto out;
 577         }
 578         if (t->num_mapped_columns) {
 579                 ret = create_table_index(t);
 580                 if (ret < 0)
 581                         goto out;
 582         }
 583         ret = 1;
 584 out:
 585         free(table_dir);
 586         free(t->columns);
 587         free(t);
 588         return ret;
 589 }
 590
 591 static int table_is_dirty(struct osl_table *t)
 592 {
 593         char *buf = (char *)t->index_map.data + IDX_DIRTY_FLAG;
 594         uint8_t dirty = read_u8(buf) & 0x1;
 595         return !!dirty;
 596 }
 597
 598 static void mark_table_dirty(struct osl_table *t)
 599 {
 600         char *buf = (char *)t->index_map.data + IDX_DIRTY_FLAG;
 601         write_u8(buf, read_u8(buf) | 1);
 602 }
 603
 604 static void mark_table_clean(struct osl_table *t)
 605 {
 606         char *buf = (char *)t->index_map.data + IDX_DIRTY_FLAG;
 607         write_u8(buf, read_u8(buf) & 0xfe);
 608 }
 609
 610 static void unmap_column(struct osl_table *t, unsigned col_num)
 611 {
 612         struct osl_object map = t->columns[col_num].data_map;
 613         int ret;
 614         if (!map.data)
 615                 return;
 616         ret = osl_munmap(map.data, map.size);
 617         assert(ret > 0);
 618         map.data = NULL;
 619 }
 620
 621 /**
 622  * Unmap all mapped files of an osl table.
 623  *
 624  * \param t Pointer to a mapped table.
 625  * \param flags Options for unmapping.
 626  *
 627  * \return Positive on success, negative on errors.
 628  *
 629  * \sa map_table(), enum osl_close_flags, osl_munmap().
 630  */
 631 int unmap_table(struct osl_table *t, enum osl_close_flags flags)
 632 {
 633         unsigned i;
 634         const struct osl_column_description *cd;
 635         int ret;
 636
 637         if (!t->num_mapped_columns) /* can this ever happen? */
 638                 return 1;
 639         INFO_LOG("unmapping table '%s'\n", t->desc->name);
 640         if (!t->index_map.data)
 641                 return -E_OSL_NOT_MAPPED;
 642         if (flags & OSL_MARK_CLEAN)
 643                 mark_table_clean(t);
 644         ret = osl_munmap(t->index_map.data, t->index_map.size);
 645         if (ret < 0)
 646                 return ret;
 647         t->index_map.data = NULL;
 648         if (!t->num_rows)
 649                 return 1;
 650         FOR_EACH_MAPPED_COLUMN(i, t, cd)
 651                 unmap_column(t, i);
 652         return 1;
 653 }
 654
 655 static int map_column(struct osl_table *t, unsigned col_num)
 656 {
 657         struct stat statbuf;
 658         char *filename = column_filename(t, col_num);
 659         int ret;
 660
 661         if (!filename)
 662                 return -E_OSL_NOMEM;
 663         ret = osl_stat(filename, &statbuf);
 664         if (ret < 0) {
 665                 free(filename);
 666                 return ret;
 667         }
 668         if (!(S_IFREG & statbuf.st_mode)) {
 669                 free(filename);
 670                 return ret;
 671         }
 672         ret = mmap_full_file(filename, O_RDWR,
 673                 &t->columns[col_num].data_map.data,
 674                 &t->columns[col_num].data_map.size,
 675                 NULL);
 676         free(filename);
 677         return ret;
 678 }
 679
 680 /**
 681  * Map the index file and all columns of type \p OSL_MAPPED_STORAGE into memory.
 682  *
 683  * \param t Pointer to an initialized table structure.
 684  * \param flags Mapping options.
 685  *
 686  * \return Negative return value on errors; on success the number of rows
 687  * (including invalid rows) is returned.
 688  *
 689  * \sa unmap_table(), enum map_table_flags, osl_open_table(), mmap(2).
 690  */
 691 int map_table(struct osl_table *t, enum map_table_flags flags)
 692 {
 693         char *filename;
 694         const struct osl_column_description *cd;
 695         int i = 0, ret, num_rows = 0;
 696
 697         if (!t->num_mapped_columns)
 698                 return 0;
 699         if (t->index_map.data)
 700                 return -E_OSL_ALREADY_MAPPED;
 701         filename = index_filename(t->desc);
 702         if (!filename)
 703                 return -E_OSL_NOMEM;
 704         INFO_LOG("mapping table '%s' (index: %s)\n", t->desc->name, filename);
 705         ret = mmap_full_file(filename, flags & MAP_TBL_FL_MAP_RDONLY?
 706                 O_RDONLY : O_RDWR, &t->index_map.data, &t->index_map.size, NULL);
 707         free(filename);
 708         if (ret < 0)
 709                 return ret;
 710         if (flags & MAP_TBL_FL_VERIFY_INDEX) {
 711                 ret = compare_table_descriptions(t);
 712                 if (ret < 0)
 713                         goto err;
 714         }
 715         ret = -E_OSL_BUSY;
 716         if (!(flags & MAP_TBL_FL_IGNORE_DIRTY)) {
 717                 if (table_is_dirty(t)) {
 718                         ERROR_LOG("%s is dirty\n", t->desc->name);
 719                         goto err;
 720                 }
 721         }
 722         mark_table_dirty(t);
 723         num_rows = table_num_rows(t);
 724         /* map data files */
 725         FOR_EACH_COLUMN(i, t->desc, cd) {
 726                 if (cd->storage_type == OSL_NO_STORAGE)
 727                         continue;
 728                 column_name_hash(t, cd->name, t->columns[i].name_hash);
 729                 if (num_rows > 0 && cd->storage_type == OSL_MAPPED_STORAGE) {
 730                         ret = map_column(t, i);
 731                         if (ret < 0)
 732                                 goto err;
 733                 }
 734         }
 735         return num_rows;
 736 err:    /* unmap what is already mapped */
 737         for (i--; i >= 0; i--) {
 738                 struct osl_object map = t->columns[i].data_map;
 739                 osl_munmap(map.data, map.size);
 740                 map.data = NULL;
 741         }
 742         osl_munmap(t->index_map.data, t->index_map.size);
 743         t->index_map.data = NULL;
 744         return ret;
 745 }
 746
 747 /**
 748  * Retrieve a mapped object by row and column number.
 749  *
 750  * \param t Pointer to an open osl table.
 751  * \param col_num Number of the mapped column containing the object to retrieve.
 752  * \param row_num Number of the row containing the object to retrieve.
 753  * \param obj The result is returned here.
 754  *
 755  * It is considered an error if \a col_num does not refer to a column
 756  * of storage type \p OSL_MAPPED_STORAGE.
 757  *
 758  * \return Standard.
 759  *
 760  * \sa osl_storage_type.
 761  */
 762 int get_mapped_object(const struct osl_table *t, unsigned col_num,
 763         uint32_t row_num, struct osl_object *obj)
 764 {
 765         struct osl_column *col = &t->columns[col_num];
 766         uint32_t offset;
 767         char *cell_index;
 768         int ret;
 769
 770         if (t->num_rows <= row_num)
 771                 return -E_OSL_BAD_ROW_NUM;
 772         ret = get_cell_index(t, row_num, col_num, &cell_index);
 773         if (ret < 0)
 774                 return ret;
 775         offset = read_u32(cell_index);
 776         obj->size = read_u32(cell_index + 4);
 777         obj->data = col->data_map.data + offset;
 778         return 1;
 779 }
 780
 781 /*
 782  * It's OK to call this with result = rb_node = NULL.  If result is not NULL,
 783  * and rb key was not found, result points to the parent node.
 784  */
 785 static int search_rbtree(const struct osl_object *obj,
 786                 const struct osl_table *t, unsigned col_num,
 787                 struct rb_node **result, struct rb_node ***rb_link)
 788 {
 789         struct osl_column *col = &t->columns[col_num];
 790         struct rb_node **new = &col->rbtree.rb_node, *parent = NULL;
 791         const struct osl_column_description *cd =
 792                 get_column_description(t->desc, col_num);
 793         enum osl_storage_type st = cd->storage_type;
 794         while (*new) {
 795                 struct osl_row *this_row = get_row_pointer(*new,
 796                         col->rbtree_num);
 797                 int ret;
 798                 struct osl_object this_obj;
 799                 parent = *new;
 800                 if (st == OSL_MAPPED_STORAGE) {
 801                         ret = get_mapped_object(t, col_num, this_row->num,
 802                                 &this_obj);
 803                         if (ret < 0)
 804                                 return ret;
 805                 } else
 806                         this_obj = this_row->volatile_objects[col->volatile_num];
 807                 ret = cd->compare_function(obj, &this_obj);
 808                 if (!ret) {
 809                         if (result)
 810                                 *result = get_rb_node_pointer(this_row,
 811                                         col->rbtree_num);
 812                         return 1;
 813                 }
 814                 if (ret < 0)
 815                         new = &((*new)->rb_left);
 816                 else
 817                         new = &((*new)->rb_right);
 818         }
 819         if (result)
 820                 *result = parent;
 821         if (rb_link)
 822                 *rb_link = new;
 823         return -E_OSL_RB_KEY_NOT_FOUND;
 824 }
 825
 826 static int insert_rbtree(struct osl_table *t, unsigned col_num,
 827         const struct osl_row *row, const struct osl_object *obj)
 828 {
 829         struct rb_node *parent, **rb_link;
 830         unsigned rbtree_num;
 831         struct rb_node *n;
 832         int ret = search_rbtree(obj, t, col_num, &parent, &rb_link);
 833
 834         if (ret > 0)
 835                 return -E_OSL_RB_KEY_EXISTS;
 836         rbtree_num = t->columns[col_num].rbtree_num;
 837         n = get_rb_node_pointer(row, rbtree_num);
 838         rb_link_node(n, parent, rb_link);
 839         rb_insert_color(n, &t->columns[col_num].rbtree);
 840         return 1;
 841 }
 842
 843 static void remove_rb_node(struct osl_table *t, unsigned col_num,
 844                 const struct osl_row *row)
 845 {
 846         struct osl_column *col = &t->columns[col_num];
 847         const struct osl_column_description *cd =
 848                 get_column_description(t->desc, col_num);
 849         enum osl_storage_flags sf = cd->storage_flags;
 850         struct rb_node *victim, *splice_out_node, *tmp;
 851         if (!(sf & OSL_RBTREE))
 852                 return;
 853         /*
 854          * Which node is removed/spliced out actually depends on how many
 855          * children the victim node has: If it has no children, it gets
 856          * deleted. If it has one child, it gets spliced out. If it has two
 857          * children, its successor (which has at most a right child) gets
 858          * spliced out.
 859          */
 860         victim = get_rb_node_pointer(row, col->rbtree_num);
 861         if (victim->rb_left && victim->rb_right)
 862                 splice_out_node = rb_next(victim);
 863         else
 864                 splice_out_node = victim;
 865         /* Go up to the root and decrement the size of each node in the path. */
 866         for (tmp = splice_out_node; tmp; tmp = rb_parent(tmp))
 867                 tmp->size--;
 868         rb_erase(victim, &col->rbtree);
 869 }
 870
 871 static int add_row_to_rbtrees(struct osl_table *t, uint32_t row_num,
 872                 struct osl_object *volatile_objs, struct osl_row **row_ptr)
 873 {
 874         unsigned i;
 875         int ret;
 876         struct osl_row *row = allocate_row(t->num_rbtrees);
 877         const struct osl_column_description *cd;
 878
 879         if (!row)
 880                 return -E_OSL_NOMEM;
 881         row->num = row_num;
 882         row->volatile_objects = volatile_objs;
 883         FOR_EACH_RBTREE_COLUMN(i, t, cd) {
 884                 if (cd->storage_type == OSL_MAPPED_STORAGE) {
 885                         struct osl_object obj;
 886                         ret = get_mapped_object(t, i, row_num, &obj);
 887                         if (ret < 0)
 888                                 goto err;
 889                         ret = insert_rbtree(t, i, row, &obj);
 890                 } else { /* volatile */
 891                         const struct osl_object *obj
 892                                 = volatile_objs + t->columns[i].volatile_num;
 893                         ret = insert_rbtree(t, i, row, obj);
 894                 }
 895                 if (ret < 0)
 896                         goto err;
 897         }
 898         if (row_ptr)
 899                 *row_ptr = row;
 900         return 1;
 901 err: /* rollback changes, i.e. remove added entries from rbtrees */
 902         while (i)
 903                 remove_rb_node(t, i--, row);
 904         free(row);
 905         return ret;
 906 }
 907
 908 static void free_volatile_objects(const struct osl_table *t,
 909                 enum osl_close_flags flags)
 910 {
 911         int i, j;
 912         struct rb_node *n;
 913         struct osl_column *rb_col;
 914         const struct osl_column_description *cd;
 915
 916         if (!t->num_volatile_columns)
 917                 return;
 918         /* find the first rbtree column (any will do) */
 919         FOR_EACH_RBTREE_COLUMN(i, t, cd)
 920                 break;
 921         rb_col = t->columns + i;
 922         /* walk that rbtree and free all volatile objects */
 923         for (n = rb_first(&rb_col->rbtree); n; n = rb_next(n)) {
 924                 struct osl_row *r = get_row_pointer(n, rb_col->rbtree_num);
 925                 if (flags & OSL_FREE_VOLATILE)
 926                         FOR_EACH_VOLATILE_COLUMN(j, t, cd) {
 927                                 if (cd->storage_flags & OSL_DONT_FREE)
 928                                         continue;
 929                                 free(r->volatile_objects[
 930                                         t->columns[j].volatile_num].data);
 931                         }
 932 //                      for (j = 0; j < t->num_volatile_columns; j++)
 933 //                              free(r->volatile_objects[j].data);
 934                 free(r->volatile_objects);
 935         }
 936 }
 937
 938 /**
 939  * Erase all rbtree nodes and free resources.
 940  *
 941  * \param t Pointer to an open osl table.
 942  *
 943  * This function is called by osl_close_table().
 944  */
 945 void clear_rbtrees(struct osl_table *t)
 946 {
 947         const struct osl_column_description *cd;
 948         unsigned i, rbtrees_cleared = 0;
 949
 950         FOR_EACH_RBTREE_COLUMN(i, t, cd) {
 951                 struct osl_column *col = &t->columns[i];
 952                 struct rb_node *n;
 953                 rbtrees_cleared++;
 954                 for (n = rb_first(&col->rbtree); n;) {
 955                         struct osl_row *r;
 956                         rb_erase(n, &col->rbtree);
 957                         if (rbtrees_cleared == t->num_rbtrees) {
 958                                 r = get_row_pointer(n, col->rbtree_num);
 959                                 n = rb_next(n);
 960                                 free(r);
 961                         } else
 962                                 n = rb_next(n);
 963                 }
 964         }
 965
 966 }
 967
 968 __export int osl_close_table(struct osl_table *t, enum osl_close_flags flags)
 969 {
 970         int ret;
 971
 972         if (!t)
 973                 return -E_OSL_BAD_TABLE;
 974         NOTICE_LOG("closing table %s\n", t->desc->name);
 975         free_volatile_objects(t, flags);
 976         clear_rbtrees(t);
 977         ret = unmap_table(t, flags);
 978         if (ret < 0)
 979                 ERROR_LOG("unmap_table failed: %d\n", ret);
 980         free(t->columns);
 981         free(t);
 982         return ret;
 983 }
 984
 985 /**
 986  * Find out whether the given row number corresponds to an invalid row.
 987  *
 988  * \param t Pointer to the osl table.
 989  * \param row_num The number of the row in question.
 990  *
 991  * By definition, a row is considered invalid if all its index entries
 992  * are invalid.
 993  *
 994  * \return Positive if \a row_num corresponds to an invalid row,
 995  * zero if it corresponds to a valid row, negative on errors.
 996  */
 997 int row_is_invalid(struct osl_table *t, uint32_t row_num)
 998 {
 999         char *row_index;
1000         int ret = get_row_index(t, row_num, &row_index);
1001         unsigned n;
1002
1003         if (ret < 0)
1004                 return ret;
1005         for (n = 0; n < t->row_index_size; n++) {
1006                 if ((unsigned char)row_index[n] != 0xff)
1007                         return 0;
1008         }
1009         INFO_LOG("row %d is invalid\n", row_num);
1010         return 1;
1011 }
1012
1013 /**
1014  * Invalidate a row of an osl table.
1015  *
1016  * \param t Pointer to an open osl table.
1017  * \param row_num Number of the row to mark as invalid.
1018  *
1019  * This function marks each mapped object in the index entry of \a row as
1020  * invalid.
1021  *
1022  * \return Standard.
1023  */
1024 int mark_row_invalid(struct osl_table *t, uint32_t row_num)
1025 {
1026         char *row_index;
1027         int ret = get_row_index(t, row_num, &row_index);
1028
1029         if (ret < 0)
1030                 return ret;
1031         INFO_LOG("marking row %d as invalid\n", row_num);
1032         memset(row_index, 0xff, t->row_index_size);
1033         return 1;
1034 }
1035
1036 /**
1037  * Initialize all rbtrees and compute number of invalid rows.
1038  *
1039  * \param t The table containing the rbtrees to be initialized.
1040  *
1041  * \return Standard.
1042  */
1043 int init_rbtrees(struct osl_table *t)
1044 {
1045         int ret;
1046         unsigned n;
1047         const struct osl_column_description *cd;
1048
1049         /* create rbtrees */
1050         FOR_EACH_RBTREE_COLUMN(n, t, cd)
1051                 t->columns[n].rbtree = RB_ROOT;
1052         /* add valid rows to rbtrees */
1053         t->num_invalid_rows = 0;
1054         for (n = 0; n < t->num_rows; n++) {
1055                 struct osl_object *volatile_objs;
1056                 ret = row_is_invalid(t, n);
1057                 if (ret < 0)
1058                         return ret;
1059                 if (ret) {
1060                         t->num_invalid_rows++;
1061                         continue;
1062                 }
1063                 if (t->num_volatile_columns > 0) {
1064                         volatile_objs = calloc(t->num_volatile_columns,
1065                                 sizeof(struct osl_object));
1066                         if (!volatile_objs)
1067                                 return -E_OSL_NOMEM;
1068                 } else
1069                         volatile_objs = NULL;
1070                 ret = add_row_to_rbtrees(t, n, volatile_objs, NULL);
1071                 if (ret < 0)
1072                         return ret;
1073         }
1074         return 1;
1075 }
1076
1077 __export int osl_open_table(const struct osl_table_description *table_desc,
1078                 struct osl_table **result)
1079 {
1080         int ret;
1081         struct osl_table *t;
1082
1083         NOTICE_LOG("opening table %s\n", table_desc->name);
1084         ret = init_table_structure(table_desc, &t);
1085         if (ret < 0)
1086                 return ret;
1087         ret = map_table(t, MAP_TBL_FL_VERIFY_INDEX);
1088         if (ret < 0)
1089                 goto err;
1090         t->num_rows = ret;
1091         DEBUG_LOG("num rows: %d\n", t->num_rows);
1092         ret = init_rbtrees(t);
1093         if (ret < 0) {
1094                 osl_close_table(t, OSL_MARK_CLEAN); /* ignore further errors */
1095                 return ret;
1096         }
1097         *result = t;
1098         return 1;
1099 err:
1100         free(t->columns);
1101         free(t);
1102         return ret;
1103 }
1104
1105 static int create_disk_storage_object_dir(const struct osl_table *t,
1106                 unsigned col_num, const char *ds_name)
1107 {
1108         char *dirname;
1109         int ret;
1110
1111         if (!(t->desc->flags & OSL_LARGE_TABLE))
1112                 return 1;
1113         dirname = disk_storage_dirname(t, col_num, ds_name);
1114         if (!dirname)
1115                 return -E_OSL_NOMEM;
1116         ret = osl_mkdir(dirname, 0777);
1117         free(dirname);
1118         if (ret < 0 && ret != -E_OSL_DIR_EXISTS)
1119                 return ret;
1120         return 1;
1121 }
1122
1123 static int write_disk_storage_file(const struct osl_table *t, unsigned col_num,
1124         const struct osl_object *obj, const char *ds_name)
1125 {
1126         int ret;
1127         char *filename;
1128
1129         ret = create_disk_storage_object_dir(t, col_num, ds_name);
1130         if (ret < 0)
1131                 return ret;
1132         filename = disk_storage_path(t, col_num, ds_name);
1133         if (!filename)
1134                 return -E_OSL_NOMEM;
1135         ret = write_file(filename, obj->data, obj->size);
1136         free(filename);
1137         return ret;
1138 }
1139
1140 static int append_map_file(const struct osl_table *t, unsigned col_num,
1141         const struct osl_object *obj, uint32_t *new_size)
1142 {
1143         char *filename = column_filename(t, col_num);
1144         int ret;
1145
1146         if (!filename)
1147                 return -E_OSL_NOMEM;
1148         ret = append_file(filename, obj->data, obj->size, new_size);
1149         free(filename);
1150         return ret;
1151 }
1152
1153 static int append_row_index(const struct osl_table *t, char *row_index)
1154 {
1155         char *filename;
1156         int ret;
1157
1158         if (!t->num_mapped_columns)
1159                 return 1;
1160         filename = index_filename(t->desc);
1161         if (!filename)
1162                 return -E_OSL_NOMEM;
1163         ret = append_file(filename, row_index, t->row_index_size, NULL);
1164         free(filename);
1165         return ret;
1166 }
1167
1168 static int truncate_mapped_file(const struct osl_table *t, unsigned col_num,
1169                 off_t size)
1170 {
1171         int ret;
1172         char *filename = column_filename(t, col_num);
1173
1174         if (!filename)
1175                 return -E_OSL_NOMEM;
1176         ret = truncate_file(filename, size);
1177         free(filename);
1178         return ret;
1179 }
1180
1181 static int delete_disk_storage_file(const struct osl_table *t, unsigned col_num,
1182                 const char *ds_name)
1183 {
1184         char *dirname, *filename = disk_storage_path(t, col_num, ds_name);
1185         int ret = 1;
1186
1187         if (!filename)
1188                 return -E_OSL_NOMEM;
1189         if (unlink(filename) < 0)
1190                 ret = errno == ENOENT? -E_OSL_NOENT : -E_OSL_UNLINK;
1191         free(filename);
1192         if (ret < 0)
1193                 return ret;
1194         if (!(t->desc->flags & OSL_LARGE_TABLE))
1195                 return 1;
1196         dirname = disk_storage_dirname(t, col_num, ds_name);
1197         if (!dirname)
1198                 return -E_OSL_NOMEM;
1199         rmdir(dirname);
1200         free(dirname);
1201         return 1;
1202 }
1203
1204 __export int osl_add_and_get_row(struct osl_table *t, struct osl_object *objects,
1205                 struct osl_row **row)
1206 {
1207         int i, ret;
1208         char *ds_name = NULL;
1209         struct rb_node **rb_parents = NULL, ***rb_links = NULL;
1210         char *new_row_index = NULL;
1211         struct osl_object *volatile_objs = NULL;
1212         const struct osl_column_description *cd;
1213
1214         if (!t)
1215                 return -E_OSL_BAD_TABLE;
1216         rb_parents = malloc(t->num_rbtrees * sizeof(struct rn_node*));
1217         if (!rb_parents)
1218                 return -E_OSL_NOMEM;
1219         rb_links = malloc(t->num_rbtrees * sizeof(struct rn_node**));
1220         if (!rb_links) {
1221                 free(rb_parents);
1222                 return -E_OSL_NOMEM;
1223         }
1224         if (t->num_mapped_columns) {
1225                 new_row_index = malloc(t->row_index_size);
1226                 if (!new_row_index) {
1227                         free(rb_links);
1228                         free(rb_parents);
1229                         return -E_OSL_NOMEM;
1230                 }
1231         }
1232         /* pass 1: sanity checks */
1233 //      DEBUG_LOG("sanity tests: %p:%p\n", objects[0].data,
1234 //              objects[1].data);
1235         FOR_EACH_COLUMN(i, t->desc, cd) {
1236                 enum osl_storage_type st = cd->storage_type;
1237                 enum osl_storage_flags sf = cd->storage_flags;
1238
1239 //              ret = -E_OSL_NULL_OBJECT;
1240 //              if (!objects[i])
1241 //                      goto out;
1242                 if (st == OSL_DISK_STORAGE)
1243                         continue;
1244                 if (sf & OSL_RBTREE) {
1245                         unsigned rbtree_num = t->columns[i].rbtree_num;
1246                         ret = -E_OSL_RB_KEY_EXISTS;
1247 //                      DEBUG_LOG("checking whether %p exists\n",
1248 //                              objects[i].data);
1249                         if (search_rbtree(objects + i, t, i,
1250                                         &rb_parents[rbtree_num],
1251                                         &rb_links[rbtree_num]) > 0)
1252                                 goto out;
1253                 }
1254                 if (sf & OSL_FIXED_SIZE) {
1255 //                      DEBUG_LOG("fixed size. need: %zu, have: %d\n",
1256 //                              objects[i].size, cd->data_size);
1257                         ret = -E_OSL_BAD_DATA_SIZE;
1258                         if (objects[i].size != cd->data_size)
1259                                 goto out;
1260                 }
1261         }
1262         if (t->num_disk_storage_columns) {
1263                 ds_name = disk_storage_name_of_object(t,
1264                         &objects[t->disk_storage_name_column]);
1265                 ret = -E_OSL_NOMEM;
1266                 if (!ds_name)
1267                         goto out;
1268         }
1269         ret = unmap_table(t, OSL_MARK_CLEAN);
1270         if (ret < 0)
1271                 goto out;
1272 //      DEBUG_LOG("sanity tests passed%s\n", "");
1273         /* pass 2: create data files, append map data */
1274         FOR_EACH_COLUMN(i, t->desc, cd) {
1275                 enum osl_storage_type st = cd->storage_type;
1276                 if (st == OSL_NO_STORAGE)
1277                         continue;
1278                 if (st == OSL_MAPPED_STORAGE) {
1279                         uint32_t new_size;
1280                         struct osl_column *col = &t->columns[i];
1281 //                      DEBUG_LOG("appending object of size %zu\n",
1282 //                              objects[i].size);
1283                         ret = append_map_file(t, i, objects + i, &new_size);
1284                         if (ret < 0)
1285                                 goto rollback;
1286                         update_cell_index(new_row_index, col, new_size,
1287                                 objects[i].size);
1288                         continue;
1289                 }
1290                 /* DISK_STORAGE */
1291                 ret = write_disk_storage_file(t, i, objects + i, ds_name);
1292                 if (ret < 0)
1293                         goto rollback;
1294         }
1295         ret = append_row_index(t, new_row_index);
1296         if (ret < 0)
1297                 goto rollback;
1298         ret = map_table(t, MAP_TBL_FL_VERIFY_INDEX);
1299         if (ret < 0) { /* truncate index and rollback changes */
1300                 char *filename = index_filename(t->desc);
1301                 if (filename)
1302                         truncate_file(filename, t->row_index_size);
1303                 free(filename);
1304                 goto rollback;
1305         }
1306         /* pass 3: add entry to rbtrees */
1307         if (t->num_volatile_columns) {
1308                 ret = -E_OSL_NOMEM;
1309                 volatile_objs = calloc(t->num_volatile_columns,
1310                         sizeof(struct osl_object));
1311                 if (!volatile_objs)
1312                         goto out;
1313                 FOR_EACH_VOLATILE_COLUMN(i, t, cd)
1314                         volatile_objs[t->columns[i].volatile_num] = objects[i];
1315         }
1316         t->num_rows++;
1317 //      DEBUG_LOG("adding new entry as row #%d\n", t->num_rows - 1);
1318         ret = add_row_to_rbtrees(t, t->num_rows - 1, volatile_objs, row);
1319         if (ret < 0)
1320                 goto out;
1321 //      DEBUG_LOG("added new entry as row #%d\n", t->num_rows - 1);
1322         ret = 1;
1323         goto out;
1324 rollback: /* rollback all changes made, ignore further errors */
1325         for (i--; i >= 0; i--) {
1326                 cd = get_column_description(t->desc, i);
1327                 enum osl_storage_type st = cd->storage_type;
1328                 if (st == OSL_NO_STORAGE)
1329                         continue;
1330
1331                 if (st == OSL_MAPPED_STORAGE)
1332                         truncate_mapped_file(t, i, objects[i].size);
1333                 else /* disk storage */
1334                         delete_disk_storage_file(t, i, ds_name);
1335         }
1336         /* ignore error and return previous error value */
1337         map_table(t, MAP_TBL_FL_VERIFY_INDEX);
1338 out:
1339         free(new_row_index);
1340         free(ds_name);
1341         free(rb_parents);
1342         free(rb_links);
1343         return ret;
1344 }
1345
1346 __export int osl_add_row(struct osl_table *t, struct osl_object *objects)
1347 {
1348         return osl_add_and_get_row(t, objects, NULL);
1349 }
1350
1351 __export int osl_get_object(const struct osl_table *t, const struct osl_row *r,
1352         unsigned col_num, struct osl_object *object)
1353 {
1354         const struct osl_column_description *cd;
1355
1356         if (!t)
1357                 return -E_OSL_BAD_TABLE;
1358         cd = get_column_description(t->desc, col_num);
1359         /* col must not be disk storage */
1360         if (cd->storage_type == OSL_DISK_STORAGE)
1361                 return -E_OSL_BAD_STORAGE_TYPE;
1362         if (cd->storage_type == OSL_MAPPED_STORAGE)
1363                 return get_mapped_object(t, col_num, r->num, object);
1364         /* volatile */
1365         *object = r->volatile_objects[t->columns[col_num].volatile_num];
1366         return 1;
1367 }
1368
1369 __export int osl_del_row(struct osl_table *t, struct osl_row *row)
1370 {
1371         struct osl_row *r = row;
1372         int i, ret;
1373         const struct osl_column_description *cd;
1374
1375         if (!t)
1376                 return -E_OSL_BAD_TABLE;
1377         INFO_LOG("deleting row %p\n", row);
1378
1379         if (t->num_disk_storage_columns) {
1380                 char *ds_name;
1381                 ret = disk_storage_name_of_row(t, r, &ds_name);
1382                 if (ret < 0)
1383                         goto out;
1384                 FOR_EACH_DISK_STORAGE_COLUMN(i, t, cd)
1385                         delete_disk_storage_file(t, i, ds_name);
1386                 free(ds_name);
1387         }
1388         FOR_EACH_COLUMN(i, t->desc, cd) {
1389                 struct osl_column *col = t->columns + i;
1390                 enum osl_storage_type st = cd->storage_type;
1391                 remove_rb_node(t, i, r);
1392                 if (st == OSL_MAPPED_STORAGE)
1393                         continue;
1394                 if (st == OSL_NO_STORAGE && !(cd->storage_flags & OSL_DONT_FREE))
1395                         free(r->volatile_objects[col->volatile_num].data);
1396         }
1397         if (t->num_mapped_columns) {
1398                 ret = mark_row_invalid(t, r->num);
1399                 if (ret < 0)
1400                         goto out;
1401                 t->num_invalid_rows++;
1402         } else
1403                 t->num_rows--;
1404         ret = 1;
1405 out:
1406         free(r->volatile_objects);
1407         free(r);
1408         return ret;
1409 }
1410
1411 /* test if column has an rbtree */
1412 static int check_rbtree_col(const struct osl_table *t, unsigned col_num,
1413                 struct osl_column **col)
1414 {
1415         if (!t)
1416                 return -E_OSL_BAD_TABLE;
1417         if (!(get_column_description(t->desc, col_num)->storage_flags & OSL_RBTREE))
1418                 return -E_OSL_BAD_STORAGE_FLAGS;
1419         *col = t->columns + col_num;
1420         return 1;
1421 }
1422
1423 __export int osl_get_row(const struct osl_table *t, unsigned col_num,
1424                 const struct osl_object *obj, struct osl_row **result)
1425 {
1426         int ret;
1427         struct rb_node *node;
1428         struct osl_row *row;
1429         struct osl_column *col;
1430
1431         *result = NULL;
1432         ret = check_rbtree_col(t, col_num, &col);
1433         if (ret < 0)
1434                 return ret;
1435         ret = search_rbtree(obj, t, col_num, &node, NULL);
1436         if (ret < 0)
1437                 return ret;
1438         row = get_row_pointer(node, t->columns[col_num].rbtree_num);
1439         *result = row;
1440         return 1;
1441 }
1442
1443 static int rbtree_loop(struct osl_column *col, void *private_data,
1444                 osl_rbtree_loop_func *func)
1445 {
1446         struct rb_node *n, *tmp;
1447
1448         /* this for-loop is safe against removal of an entry */
1449         for (n = rb_first(&col->rbtree), tmp = n? rb_next(n) : NULL;
1450                         n;
1451                         n = tmp, tmp = tmp? rb_next(tmp) : NULL) {
1452                 struct osl_row *r = get_row_pointer(n, col->rbtree_num);
1453                 if (func(r, private_data) < 0)
1454                         return -E_OSL_LOOP;
1455         }
1456         return 1;
1457 }
1458
1459 static int rbtree_loop_reverse(struct osl_column *col, void *private_data,
1460                 osl_rbtree_loop_func *func)
1461 {
1462         struct rb_node *n, *tmp;
1463
1464         /* safe against removal of an entry */
1465         for (n = rb_last(&col->rbtree), tmp = n? rb_prev(n) : NULL;
1466                         n;
1467                         n = tmp, tmp = tmp? rb_prev(tmp) : NULL) {
1468                 struct osl_row *r = get_row_pointer(n, col->rbtree_num);
1469                 if (func(r, private_data) < 0)
1470                         return -E_OSL_LOOP;
1471         }
1472         return 1;
1473 }
1474
1475 __export int osl_rbtree_loop(const struct osl_table *t, unsigned col_num,
1476         void *private_data, osl_rbtree_loop_func *func)
1477 {
1478         struct osl_column *col;
1479
1480         int ret = check_rbtree_col(t, col_num, &col);
1481         if (ret < 0)
1482                 return ret;
1483         return rbtree_loop(col, private_data, func);
1484 }
1485
1486 __export int osl_rbtree_loop_reverse(const struct osl_table *t, unsigned col_num,
1487         void *private_data, osl_rbtree_loop_func *func)
1488 {
1489         struct osl_column *col;
1490
1491         int ret = check_rbtree_col(t, col_num, &col);
1492         if (ret < 0)
1493                 return ret;
1494         return rbtree_loop_reverse(col, private_data, func);
1495 }
1496
1497 /* TODO: Rollback changes on errors */
1498 static int rename_disk_storage_objects(struct osl_table *t,
1499                 struct osl_object *old_obj, struct osl_object *new_obj)
1500 {
1501         int i, ret;
1502         const struct osl_column_description *cd;
1503         char *old_ds_name, *new_ds_name;
1504
1505         if (!t->num_disk_storage_columns)
1506                 return 1; /* nothing to do */
1507         if (old_obj->size == new_obj->size && !memcmp(new_obj->data,
1508                         old_obj->data, new_obj->size))
1509                 return 1; /* object did not change */
1510         old_ds_name = disk_storage_name_of_object(t, old_obj);
1511         new_ds_name = disk_storage_name_of_object(t, new_obj);
1512         ret = -E_OSL_NOMEM;
1513         if (!old_ds_name || ! new_ds_name)
1514                 goto out;
1515
1516         FOR_EACH_DISK_STORAGE_COLUMN(i, t, cd) {
1517                 char *old_filename, *new_filename;
1518                 ret = create_disk_storage_object_dir(t, i, new_ds_name);
1519                 if (ret < 0)
1520                         goto out;
1521                 old_filename = disk_storage_path(t, i, old_ds_name);
1522                 new_filename = disk_storage_path(t, i, new_ds_name);
1523                 if (!old_filename || !new_filename)
1524                         ret = -E_OSL_NOMEM;
1525                 else
1526                         ret = osl_rename(old_filename, new_filename);
1527                 free(old_filename);
1528                 free(new_filename);
1529                 if (ret < 0)
1530                         goto out;
1531         }
1532         ret = 1;
1533 out:
1534         free(old_ds_name);
1535         free(new_ds_name);
1536         return ret;
1537
1538 }
1539
1540 __export int osl_update_object(struct osl_table *t, const struct osl_row *r,
1541                 unsigned col_num, struct osl_object *obj)
1542 {
1543         struct osl_column *col;
1544         const struct osl_column_description *cd;
1545         int ret;
1546
1547         if (!t)
1548                 return -E_OSL_BAD_TABLE;
1549         col = &t->columns[col_num];
1550         cd = get_column_description(t->desc, col_num);
1551         DEBUG_LOG("updating column %u of %s\n", col_num, t->desc->name);
1552         if (cd->storage_flags & OSL_RBTREE) {
1553                 if (search_rbtree(obj, t, col_num, NULL, NULL) > 0)
1554                         return -E_OSL_RB_KEY_EXISTS;
1555         }
1556         if (cd->storage_flags & OSL_FIXED_SIZE) {
1557                 if (obj->size != cd->data_size)
1558                         return -E_OSL_BAD_DATA_SIZE;
1559         }
1560         remove_rb_node(t, col_num, r);
1561         if (cd->storage_type == OSL_NO_STORAGE) { /* TODO: If fixed size, reuse object? */
1562                 if (!(cd->storage_flags & OSL_DONT_FREE))
1563                         free(r->volatile_objects[col->volatile_num].data);
1564                 r->volatile_objects[col->volatile_num] = *obj;
1565         } else if (cd->storage_type == OSL_DISK_STORAGE) {
1566                 char *ds_name;
1567                 ret = disk_storage_name_of_row(t, r, &ds_name);
1568                 if (ret < 0)
1569                         return ret;
1570                 ret = delete_disk_storage_file(t, col_num, ds_name);
1571                 if (ret < 0 && ret != -E_OSL_NOENT) {
1572                         free(ds_name);
1573                         return ret;
1574                 }
1575                 ret = write_disk_storage_file(t, col_num, obj, ds_name);
1576                 free(ds_name);
1577                 if (ret < 0)
1578                         return ret;
1579         } else { /* mapped storage */
1580                 struct osl_object old_obj;
1581                 ret = get_mapped_object(t, col_num, r->num, &old_obj);
1582                 if (ret < 0)
1583                         return ret;
1584                 /*
1585                  * If the updated column is the disk storage name column, the
1586                  * disk storage name changes, so we have to rename all disk
1587                  * storage objects accordingly.
1588                  */
1589                 if (col_num == t->disk_storage_name_column) {
1590                         ret = rename_disk_storage_objects(t, &old_obj, obj);
1591                         if (ret < 0)
1592                                 return ret;
1593                 }
1594                 if (cd->storage_flags & OSL_FIXED_SIZE)
1595                         memcpy(old_obj.data, obj->data, cd->data_size);
1596                 else { /* TODO: if the size doesn't change, use old space */
1597                         uint32_t new_data_map_size;
1598                         char *row_index;
1599                         ret = get_row_index(t, r->num, &row_index);
1600                         if (ret < 0)
1601                                 return ret;
1602                         unmap_column(t, col_num);
1603                         ret = append_map_file(t, col_num, obj,
1604                                 &new_data_map_size);
1605                         if (ret < 0)
1606                                 return ret;
1607                         ret = map_column(t, col_num);
1608                         if (ret < 0)
1609                                 return ret;
1610                         update_cell_index(row_index, col, new_data_map_size,
1611                                 obj->size);
1612                 }
1613         }
1614         if (cd->storage_flags & OSL_RBTREE) {
1615                 ret = insert_rbtree(t, col_num, r, obj);
1616                 if (ret < 0)
1617                         return ret;
1618         }
1619         return 1;
1620 }
1621
1622 __export int osl_open_disk_object(const struct osl_table *t, const struct osl_row *r,
1623                 unsigned col_num, struct osl_object *obj)
1624 {
1625         const struct osl_column_description *cd;
1626         char *ds_name, *filename;
1627         int ret;
1628
1629         if (!t)
1630                 return -E_OSL_BAD_TABLE;
1631         cd = get_column_description(t->desc, col_num);
1632         if (cd->storage_type != OSL_DISK_STORAGE)
1633                 return -E_OSL_BAD_STORAGE_TYPE;
1634
1635         ret = disk_storage_name_of_row(t, r, &ds_name);
1636         if (ret < 0)
1637                 return ret;
1638         filename = disk_storage_path(t, col_num, ds_name);
1639         free(ds_name);
1640         if (!filename)
1641                 return -E_OSL_NOMEM;
1642         DEBUG_LOG("filename: %s\n", filename);
1643         ret = mmap_full_file(filename, O_RDONLY, &obj->data, &obj->size, NULL);
1644         free(filename);
1645         return ret;
1646 }
1647
1648 __export int osl_close_disk_object(struct osl_object *obj)
1649 {
1650         return osl_munmap(obj->data, obj->size);
1651 }
1652
1653 __export int osl_get_num_rows(const struct osl_table *t, unsigned *num_rows)
1654 {
1655         if (!t)
1656                 return -E_OSL_BAD_TABLE;
1657         assert(t->num_rows >= t->num_invalid_rows);
1658         *num_rows = t->num_rows - t->num_invalid_rows;
1659         return 1;
1660 }
1661
1662 __export int osl_get_rank(const struct osl_table *t, struct osl_row *r,
1663                 unsigned col_num, unsigned *rank)
1664 {
1665         struct osl_object obj;
1666         struct osl_column *col;
1667         struct rb_node *node;
1668         int ret = check_rbtree_col(t, col_num, &col);
1669
1670         if (ret < 0)
1671                 return ret;
1672         ret = osl_get_object(t, r, col_num, &obj);
1673         if (ret < 0)
1674                 return ret;
1675         ret = search_rbtree(&obj, t, col_num, &node, NULL);
1676         if (ret < 0)
1677                 return ret;
1678         ret = rb_rank(node, rank);
1679         if (ret < 0)
1680                 return -E_OSL_BAD_ROW;
1681         return 1;
1682 }
1683
1684 __export int osl_get_nth_row(const struct osl_table *t, unsigned col_num,
1685                 unsigned n, struct osl_row **result)
1686 {
1687         struct osl_column *col;
1688         struct rb_node *node;
1689         unsigned num_rows;
1690         int ret;
1691
1692         *result = NULL;
1693         if (n == 0)
1694                 return -E_OSL_RB_KEY_NOT_FOUND;
1695         ret = osl_get_num_rows(t, &num_rows);
1696         if (ret < 0)
1697                 return ret;
1698         if (n > num_rows)
1699                 return -E_OSL_RB_KEY_NOT_FOUND;
1700         ret = check_rbtree_col(t, col_num, &col);
1701         if (ret < 0)
1702                 return ret;
1703         node = rb_nth(col->rbtree.rb_node, n);
1704         if (!node)
1705                 return -E_OSL_RB_KEY_NOT_FOUND;
1706         *result = get_row_pointer(node, col->rbtree_num);
1707         return 1;
1708 }
1709
1710 __export int osl_rbtree_first_row(const struct osl_table *t, unsigned col_num,
1711                 struct osl_row **result)
1712 {
1713         return osl_get_nth_row(t, col_num, 1, result);
1714 }
1715
1716 __export int osl_rbtree_last_row(const struct osl_table *t, unsigned col_num,
1717                 struct osl_row **result)
1718 {
1719         unsigned num_rows;
1720         int ret = osl_get_num_rows(t, &num_rows);
1721
1722         if (ret < 0)
1723                 return ret;
1724         return osl_get_nth_row(t, col_num, num_rows, result);
1725 }