osl.c

   1 /*
   2  * Copyright (C) 2007-2009 Andre Noll <maan@tuebingen.mpg.de>
   3  *
   4  * Licensed under the GPL v2. For licencing details see COPYING.
   5  */
   6
   7 /** \file osl.c Object storage layer functions. */
   8 #include <dirent.h> /* readdir() */
   9 #include <assert.h>
  10
  11 #include "log.h"
  12 #include "osl.h"
  13 #include "util.h"
  14 #include "osl_core.h"
  15
  16 /*
  17  * Taken from Drepper: How to write shared libraries, Appendix B.
  18  *
  19  * The main reason for this rather fancy implementation of strerror() is to
  20  * avoid having an array of pointers. This is desirable because initialized
  21  * pointer variables increase the startup time of the library due to the
  22  * processing of relocations.
  23  */
  24 #include <stddef.h>
  25 #define MSGSTRFIELD(line) MSGSTRFIELD1(line)
  26 #define MSGSTRFIELD1(line) str##line
  27 static const union msgstr_t {
  28         struct {
  29 #define OSL_ERROR(n, s) char MSGSTRFIELD(__LINE__)[sizeof(s)];
  30 #include "errtab.h"
  31 #undef OSL_ERROR
  32         };
  33         char str[0];
  34 } msgstr = { {
  35 #define OSL_ERROR(n, s) s,
  36 #include "errtab.h"
  37 #undef OSL_ERROR
  38 } };
  39 static const unsigned int errmsgidx[] = {
  40 #define OSL_ERROR(n, s) [n] = offsetof(union msgstr_t, MSGSTRFIELD(__LINE__)),
  41 #include "errtab.h"
  42 #undef OSL_ERROR
  43 };
  44
  45 __export const char *osl_strerror(int num)
  46 {
  47         return msgstr.str + errmsgidx[num];
  48 }
  49
  50 int loglevel = 0;
  51
  52 static void __attribute ((constructor)) init_loglevel(void)
  53 {
  54         char *p = getenv("OSL_LOGLEVEL");
  55
  56         /* don't log anything if unset */
  57         loglevel = p? atoi(p) : EMERG + 1;
  58 }
  59
  60 /**
  61  * The log function.
  62  *
  63  * \param ll Loglevel.
  64  * \param fmt Usual format string.
  65  *
  66  * All XXX_LOG() macros use this function.
  67  */
  68 __printf_2_3 void __log(int ll, const char* fmt,...)
  69 {
  70         va_list argp;
  71         struct tm *tm;
  72         time_t t1;
  73         char str[255] = "";
  74
  75         if (ll < loglevel)
  76                 return;
  77         time(&t1);
  78         tm = localtime(&t1);
  79         strftime(str, sizeof(str), "%b %d %H:%M:%S", tm);
  80         fprintf(stderr, "%s ", str);
  81         va_start(argp, fmt);
  82         vfprintf(stderr, fmt, argp);
  83         va_end(argp);
  84 }
  85
  86 /**
  87  * A wrapper for lseek(2).
  88  *
  89  * \param fd The file descriptor whose offset is to be to repositioned.
  90  * \param offset A value-result parameter.
  91  * \param whence Usual repositioning directive.
  92  *
  93  * Reposition the offset of the file descriptor \a fd to the argument \a offset
  94  * according to the directive \a whence. Upon successful return, \a offset
  95  * contains the resulting offset location as measured in bytes from the
  96  * beginning of the file.
  97  *
  98  * \return Positive on success. Otherwise, the function returns \p -E_OSL_LSEEK.
  99  *
 100  * \sa lseek(2).
 101  */
 102 static int __lseek(int fd, off_t *offset, int whence)
 103 {
 104         *offset = lseek(fd, *offset, whence);
 105         int ret = -E_OSL_LSEEK;
 106         if (*offset == -1)
 107                 return ret;
 108         return 1;
 109 }
 110
 111 static int append_file(const char *filename, char *data, size_t data_size,
 112                 uint32_t *new_pos)
 113 {
 114         int ret, fd;
 115
 116 //      DEBUG_LOG("appending %zu  + %zu bytes\n", header_size, data_size);
 117         ret = osl_open(filename, O_WRONLY | O_CREAT | O_APPEND, 0644);
 118         if (ret < 0)
 119                 return ret;
 120         fd = ret;
 121         ret = write_all(fd, data, &data_size);
 122         if (ret < 0)
 123                 goto out;
 124         if (new_pos) {
 125                 off_t offset = 0;
 126                 ret = __lseek(fd, &offset, SEEK_END);
 127                 if (ret < 0)
 128                         goto out;
 129 //              DEBUG_LOG("new file size: " FMT_OFF_T "\n", offset);
 130                 *new_pos = offset;
 131         }
 132         ret = 1;
 133 out:
 134         close(fd);
 135         return ret;
 136 }
 137
 138 static int verify_name(const char *name)
 139 {
 140         if (!name)
 141                 return -E_OSL_BAD_NAME;
 142         if (!*name)
 143                 return -E_OSL_BAD_NAME;
 144         if (strchr(name, '/'))
 145                 return -E_OSL_BAD_NAME;
 146         if (!strcmp(name, ".."))
 147                 return -E_OSL_BAD_NAME;
 148         if (!strcmp(name, "."))
 149                 return -E_OSL_BAD_NAME;
 150         return 1;
 151 }
 152
 153 static char *disk_storage_dirname(const struct osl_table *t, unsigned col_num,
 154                 const char *ds_name)
 155 {
 156         char *dirname, *column_name = column_filename(t, col_num);
 157
 158         if (!column_name)
 159                 return NULL;
 160         if (!(t->desc->flags & OSL_LARGE_TABLE))
 161                 return column_name;
 162         dirname = make_message("%s/%.2s", column_name, ds_name);
 163         free(column_name);
 164         return dirname;
 165 }
 166
 167 static char *disk_storage_name_of_object(const struct osl_table *t,
 168         const struct osl_object *obj)
 169 {
 170         HASH_TYPE hash[HASH_SIZE];
 171         hash_object(obj, hash);
 172         return disk_storage_name_of_hash(t, hash);
 173 }
 174
 175 static int disk_storage_name_of_row(const struct osl_table *t,
 176                 const struct osl_row *row, char **name)
 177 {
 178         struct osl_object obj;
 179         int ret = osl_get_object(t, row, t->disk_storage_name_column, &obj);
 180
 181         if (ret < 0)
 182                 return ret;
 183         *name = disk_storage_name_of_object(t, &obj);
 184         if (*name)
 185                 return 1;
 186         return -E_OSL_NOMEM;
 187 }
 188
 189 static void column_name_hash(const char *col_name, HASH_TYPE *hash)
 190 {
 191         hash_function(col_name, strlen(col_name), hash);
 192 }
 193
 194 static int init_column_descriptions(struct osl_table *t)
 195 {
 196         int i, j, ret;
 197         const struct osl_column_description *cd;
 198
 199         ret = verify_name(t->desc->name);
 200         if (ret < 0)
 201                 goto err;
 202         ret = -E_OSL_BAD_DB_DIR;
 203         if (!t->desc->dir && (t->num_disk_storage_columns || t->num_mapped_columns))
 204                 goto err;
 205         /* the size of the index header without column descriptions */
 206         t->index_header_size = IDX_COLUMN_DESCRIPTIONS;
 207         FOR_EACH_COLUMN(i, t->desc, cd) {
 208                 if (cd->storage_flags & OSL_RBTREE) {
 209                         if (!cd->compare_function)
 210                                 return -E_OSL_NO_COMPARE_FUNC;
 211                 }
 212                 if (cd->storage_type == OSL_NO_STORAGE)
 213                         continue;
 214                 ret = -E_OSL_NO_COLUMN_NAME;
 215                 if (!cd->name || !cd->name[0])
 216                         goto err;
 217                 ret = verify_name(cd->name);
 218                 if (ret < 0)
 219                         goto err;
 220                 t->index_header_size += index_column_description_size(cd->name);
 221                 ret = -E_OSL_DUPLICATE_COL_NAME;
 222                 for (j = i + 1; j < t->desc->num_columns; j++) {
 223                         const char *name2 = get_column_description(t->desc,
 224                                 j)->name;
 225                         if (cd->name && name2 && !strcmp(cd->name, name2))
 226                                 goto err;
 227                 }
 228         }
 229         return 1;
 230 err:
 231         return ret;
 232 }
 233
 234 /**
 235  * Initialize a struct table from given table description.
 236  *
 237  * \param desc The description of the osl table.
 238  * \param table_ptr Result is returned here.
 239  *
 240  * This function performs several sanity checks on \p desc and returns if any
 241  * of these tests fail. On success, a struct \p osl_table is allocated and
 242  * initialized with data derived from \p desc.
 243  *
 244  * \return Standard.
 245  *
 246  * \sa struct osl_table.
 247  */
 248 int init_table_structure(const struct osl_table_description *desc,
 249                 struct osl_table **table_ptr)
 250 {
 251         const struct osl_column_description *cd;
 252         struct osl_table *t = calloc(1, sizeof(*t));
 253         int i, ret = -E_OSL_NOMEM, have_disk_storage_name_column = 0;
 254
 255         if (!t)
 256                 return ret;
 257         ret = -E_OSL_BAD_TABLE_DESC;
 258         if (!desc)
 259                 goto err;
 260         DEBUG_LOG("creating table structure for '%s' from table "
 261                 "description\n", desc->name);
 262         ret = -E_OSL_NO_COLUMN_DESC;
 263         if (!desc->column_descriptions)
 264                 goto err;
 265         ret = -E_OSL_NO_COLUMNS;
 266         if (!desc->num_columns)
 267                 goto err;
 268         ret = -E_OSL_NOMEM;
 269         t->columns = calloc(desc->num_columns, sizeof(struct osl_column));
 270         if (!t->columns)
 271                 goto err;
 272         t->desc = desc;
 273         FOR_EACH_COLUMN(i, t->desc, cd) {
 274                 enum osl_storage_type st = cd->storage_type;
 275                 enum osl_storage_flags sf = cd->storage_flags;
 276                 struct osl_column *col = &t->columns[i];
 277
 278                 ret = -E_OSL_BAD_STORAGE_TYPE;
 279                 if (st != OSL_MAPPED_STORAGE && st != OSL_DISK_STORAGE
 280                                 && st != OSL_NO_STORAGE)
 281                         goto err;
 282                 ret = -E_OSL_BAD_STORAGE_FLAGS;
 283                 if (st == OSL_DISK_STORAGE && sf & OSL_RBTREE)
 284                         goto err;
 285                 if ((sf & OSL_RBTREE) && !(sf & OSL_UNIQUE))
 286                         WARNING_LOG("invalid storage flags for column %s: "
 287                                 "OSL_RBTREE && !OSL_UNIQUE\n", cd->name);
 288                 ret = -E_OSL_BAD_STORAGE_SIZE;
 289                 if (sf & OSL_FIXED_SIZE && !cd->data_size)
 290                         goto err;
 291                 switch (st) {
 292                 case OSL_DISK_STORAGE:
 293                         t->num_disk_storage_columns++;
 294                         break;
 295                 case OSL_MAPPED_STORAGE:
 296                         t->num_mapped_columns++;
 297                         col->index_offset = t->row_index_size;
 298                         t->row_index_size += 8;
 299                         break;
 300                 case OSL_NO_STORAGE:
 301                         col->volatile_num = t->num_volatile_columns;
 302                         t->num_volatile_columns++;
 303                         break;
 304                 }
 305                 if (sf & OSL_RBTREE) {
 306                         col->rbtree_num = t->num_rbtrees;
 307                         t->num_rbtrees++;
 308                         if ((sf & OSL_UNIQUE) && (st == OSL_MAPPED_STORAGE)) {
 309                                 if (!have_disk_storage_name_column)
 310                                         t->disk_storage_name_column = i;
 311                                 have_disk_storage_name_column = 1;
 312                         }
 313                 }
 314         }
 315         ret = -E_OSL_NO_UNIQUE_RBTREE_COLUMN;
 316         if (t->num_disk_storage_columns && !have_disk_storage_name_column)
 317                 goto err;
 318         ret = -E_OSL_NO_RBTREE_COL;
 319         if (!t->num_rbtrees)
 320                 goto err;
 321         /* success */
 322         DEBUG_LOG("OK. Index entry size: %u\n", t->row_index_size);
 323         ret = init_column_descriptions(t);
 324         if (ret < 0)
 325                 goto err;
 326         *table_ptr = t;
 327         return 1;
 328 err:
 329         free(t->columns);
 330         free(t);
 331         return ret;
 332 }
 333
 334 /**
 335  * Read the table description from index header.
 336  *
 337  * \param map The memory mapping of the index file.
 338  * \param desc The values found in the index header are returned here.
 339  *
 340  * Read the index header, check for the osl magic string and the table version
 341  * number.  Read all information stored in the index header into \a desc.
 342  *
 343  * \return The on-disk table version number on success, negative on errors.
 344  *
 345  * \sa struct osl_table_description, osl_create_table.
 346  */
 347 int read_table_desc(struct osl_object *map, struct osl_table_description *desc)
 348 {
 349         char *buf = map->data;
 350         uint8_t table_version;
 351         uint16_t header_size;
 352         int ret, i;
 353         unsigned offset;
 354         struct osl_column_description *cd;
 355
 356         if (map->size < MIN_INDEX_HEADER_SIZE(1))
 357                 return -E_OSL_SHORT_TABLE;
 358         if (strncmp(buf + IDX_OSL_MAGIC, OSL_MAGIC, strlen(OSL_MAGIC)))
 359                 return -E_OSL_NO_MAGIC;
 360         table_version = read_u8(buf + IDX_VERSION);
 361         INFO_LOG("osl versions (table/min/current): %u/%u/%u\n",
 362                 table_version, MIN_TABLE_VERSION, CURRENT_TABLE_VERSION);
 363         if (table_version < MIN_TABLE_VERSION /* table too old */
 364                 || table_version > CURRENT_TABLE_VERSION) /* libosl too old */
 365                 return -E_OSL_VERSION_MISMATCH;
 366         desc->flags = read_u8(buf + IDX_TABLE_FLAGS);
 367         desc->num_columns = read_u16(buf + IDX_NUM_COLUMNS);
 368         INFO_LOG("%u columns\n", desc->num_columns);
 369         if (!desc->num_columns)
 370                 return -E_OSL_NO_COLUMNS;
 371         header_size = read_u16(buf + IDX_HEADER_SIZE);
 372         if (map->size < header_size)
 373                 return -E_OSL_BAD_SIZE;
 374         desc->column_descriptions = calloc(desc->num_columns,
 375                 sizeof(struct osl_column_description));
 376         if (!desc->column_descriptions)
 377                 return -E_OSL_NOMEM;
 378         offset = IDX_COLUMN_DESCRIPTIONS;
 379         FOR_EACH_COLUMN(i, desc, cd) {
 380                 char *null_byte;
 381
 382                 ret = -E_OSL_SHORT_TABLE;
 383                 if (map->size < offset + MIN_IDX_COLUMN_DESCRIPTION_SIZE) {
 384                         ERROR_LOG("map size = %zu < %u = offset + min desc size\n",
 385                                 map->size, offset + MIN_IDX_COLUMN_DESCRIPTION_SIZE);
 386                         goto err;
 387                 }
 388                 cd->storage_type = read_u16(buf + offset + IDX_CD_STORAGE_TYPE);
 389                 cd->storage_flags = read_u16(buf + offset +
 390                         IDX_CD_STORAGE_FLAGS);
 391                 cd->data_size = read_u32(buf + offset + IDX_CD_DATA_SIZE);
 392                 null_byte = memchr(buf + offset + IDX_CD_NAME, '\0',
 393                         map->size - offset - IDX_CD_NAME);
 394                 ret = -E_OSL_INDEX_CORRUPTION;
 395                 if (!null_byte)
 396                         goto err;
 397                 ret = -E_OSL_NOMEM;
 398                 cd->name = strdup(buf + offset + IDX_CD_NAME);
 399                 if (!cd->name)
 400                         goto err;
 401                 offset += index_column_description_size(cd->name);
 402         }
 403         if (offset != header_size) {
 404                 ret = -E_OSL_INDEX_CORRUPTION;
 405                 ERROR_LOG("real header size = %u != %u = stored header size\n",
 406                         offset, header_size);
 407                 goto err;
 408         }
 409         return table_version;
 410 err:
 411         FOR_EACH_COLUMN(i, desc, cd)
 412                 free(cd->name);
 413         return ret;
 414 }
 415
 416 /*
 417  * check whether the table description given by \p t->desc matches the on-disk
 418  * table structure stored in the index of \a t.
 419  */
 420 static int compare_table_descriptions(struct osl_table *t)
 421 {
 422         int i, ret;
 423         struct osl_table_description desc;
 424         const struct osl_column_description *cd1, *cd2;
 425
 426         /* read the on-disk structure into desc */
 427         ret = read_table_desc(&t->index_map, &desc);
 428         if (ret < 0)
 429                 return ret;
 430         t->version = ret;
 431         ret = -E_OSL_BAD_TABLE_FLAGS;
 432         if (desc.flags != t->desc->flags)
 433                 goto out;
 434         ret = -E_OSL_BAD_COLUMN_NUM;
 435         if (desc.num_columns > t->desc->num_columns)
 436                 goto out;
 437         if (desc.num_columns < t->desc->num_columns) {
 438                 struct osl_column_description *cd;
 439                 unsigned diff = t->desc->num_columns - desc.num_columns;
 440                 INFO_LOG("extending table by %u volatile columns\n", diff);
 441                 ret = -E_OSL_NOMEM;
 442                 desc.column_descriptions = realloc(desc.column_descriptions,
 443                         t->desc->num_columns * sizeof(struct osl_column_description));
 444                 if (!desc.column_descriptions)
 445                         goto out;
 446                 for (i = desc.num_columns; i < t->desc->num_columns; i++) {
 447                         cd = get_column_description(&desc, i);
 448                         cd->storage_type = OSL_NO_STORAGE;
 449                         cd->name = NULL;
 450                 }
 451                 desc.num_columns += diff;
 452         }
 453         FOR_EACH_COLUMN(i, t->desc, cd1) {
 454                 cd2 = get_column_description(&desc, i);
 455                 ret = -E_OSL_BAD_STORAGE_TYPE;
 456                 if (cd1->storage_type != cd2->storage_type)
 457                         goto out;
 458                 if (cd1->storage_type == OSL_NO_STORAGE)
 459                         continue;
 460                 ret = -E_OSL_BAD_STORAGE_FLAGS;
 461                 if (cd1->storage_flags != cd2->storage_flags) {
 462                         ERROR_LOG("sf1 = %u != %u = sf2\n",
 463                                 cd1->storage_flags, cd2->storage_flags);
 464                         goto out;
 465                 }
 466                 ret = -E_OSL_BAD_DATA_SIZE;
 467                 if (cd1->storage_flags & OSL_FIXED_SIZE)
 468                         if (cd1->data_size != cd2->data_size)
 469                                 goto out;
 470                 ret = -E_OSL_BAD_COLUMN_NAME;
 471                 if (strcmp(cd1->name, cd2->name))
 472                         goto out;
 473         }
 474         INFO_LOG("table description of '%s' matches on-disk data, good\n",
 475                 t->desc->name);
 476         ret = 1;
 477 out:
 478         FOR_EACH_COLUMN(i, &desc, cd1)
 479                 free(cd1->name);
 480         free(desc.column_descriptions);
 481         return ret;
 482 }
 483
 484 static int create_table_index(struct osl_table *t)
 485 {
 486         char *buf, *filename;
 487         int i, ret;
 488         size_t size = t->index_header_size;
 489         const struct osl_column_description *cd;
 490         unsigned offset;
 491
 492         INFO_LOG("creating %zu byte index for table %s\n", size,
 493                 t->desc->name);
 494         buf = calloc(1, size);
 495         if (!buf)
 496                 return -E_OSL_NOMEM;
 497         sprintf(buf + IDX_OSL_MAGIC, "%s", OSL_MAGIC);
 498         write_u8(buf + IDX_TABLE_FLAGS, t->desc->flags);
 499         write_u8(buf + IDX_DIRTY_FLAG, 0);
 500         write_u8(buf + IDX_VERSION, t->version);
 501         write_u16(buf + IDX_NUM_COLUMNS, t->num_mapped_columns + t->num_disk_storage_columns);
 502         write_u16(buf + IDX_HEADER_SIZE, t->index_header_size);
 503         offset = IDX_COLUMN_DESCRIPTIONS;
 504         FOR_EACH_COLUMN(i, t->desc, cd) {
 505                 /* no need to store info about volatile storage */
 506                 if (cd->storage_type == OSL_NO_STORAGE)
 507                         continue;
 508                 write_u16(buf + offset + IDX_CD_STORAGE_TYPE,
 509                         cd->storage_type);
 510                 write_u16(buf + offset + IDX_CD_STORAGE_FLAGS,
 511                         cd->storage_flags);
 512                 if (cd->storage_flags & OSL_FIXED_SIZE)
 513                         write_u32(buf + offset + IDX_CD_DATA_SIZE,
 514                                 cd->data_size);
 515                 strcpy(buf + offset + IDX_CD_NAME, cd->name);
 516                 offset += index_column_description_size(cd->name);
 517         }
 518         assert(offset == size);
 519         filename = index_filename(t->desc);
 520         if (filename)
 521                 ret = write_file(filename, buf, size);
 522         else
 523                 ret = -E_OSL_NOMEM;
 524         free(buf);
 525         free(filename);
 526         return ret;
 527 }
 528
 529 __export int osl_create_table(const struct osl_table_description *desc)
 530 {
 531         const struct osl_column_description *cd;
 532         char *table_dir = NULL, *filename;
 533         struct osl_table *t;
 534         int i, ret = init_table_structure(desc, &t);
 535
 536         if (ret < 0)
 537                 return ret;
 538         t->version = CURRENT_TABLE_VERSION;
 539         INFO_LOG("creating version %u table %s\n", t->version, desc->name);
 540         FOR_EACH_COLUMN(i, t->desc, cd) {
 541                 if (cd->storage_type == OSL_NO_STORAGE)
 542                         continue;
 543                 if (!table_dir) {
 544                         ret = osl_mkdir(desc->dir, 0777);
 545                         if (ret < 0 && ret != -E_OSL_DIR_EXISTS)
 546                                 goto out;
 547                         table_dir = make_message("%s/%s", desc->dir,
 548                                 desc->name);
 549                         ret = -E_OSL_NOMEM;
 550                         if (!table_dir)
 551                                 goto out;
 552                         ret = osl_mkdir(table_dir, 0777);
 553                         if (ret < 0)
 554                                 goto out;
 555                 }
 556                 column_name_hash(cd->name, t->columns[i].name_hash);
 557                 ret = -E_OSL_NOMEM;
 558                 filename = column_filename(t, i);
 559                 if (!filename)
 560                         goto out;
 561                 INFO_LOG("filename: %s\n", filename);
 562                 if (cd->storage_type == OSL_MAPPED_STORAGE) {
 563                         ret = osl_open(filename, O_RDWR | O_CREAT | O_EXCL,
 564                                 0644);
 565                         free(filename);
 566                         if (ret < 0)
 567                                 goto out;
 568                         close(ret);
 569                         continue;
 570                 }
 571                 /* DISK STORAGE */
 572                 ret = osl_mkdir(filename, 0777);
 573                 free(filename);
 574                 if (ret < 0)
 575                         goto out;
 576         }
 577         if (t->num_mapped_columns) {
 578                 ret = create_table_index(t);
 579                 if (ret < 0)
 580                         goto out;
 581         }
 582         ret = 1;
 583 out:
 584         free(table_dir);
 585         free(t->columns);
 586         free(t);
 587         return ret;
 588 }
 589
 590 static int table_is_dirty(struct osl_table *t)
 591 {
 592         char *buf = (char *)t->index_map.data + IDX_DIRTY_FLAG;
 593         uint8_t dirty = read_u8(buf) & 0x1;
 594         return !!dirty;
 595 }
 596
 597 static void mark_table_dirty(struct osl_table *t)
 598 {
 599         char *buf = (char *)t->index_map.data + IDX_DIRTY_FLAG;
 600         write_u8(buf, read_u8(buf) | 1);
 601 }
 602
 603 static void mark_table_clean(struct osl_table *t)
 604 {
 605         char *buf = (char *)t->index_map.data + IDX_DIRTY_FLAG;
 606         write_u8(buf, read_u8(buf) & 0xfe);
 607 }
 608
 609 static void unmap_column(struct osl_table *t, unsigned col_num)
 610 {
 611         struct osl_object map = t->columns[col_num].data_map;
 612         int ret;
 613         if (!map.data)
 614                 return;
 615         ret = osl_munmap(map.data, map.size);
 616         assert(ret > 0);
 617         map.data = NULL;
 618 }
 619
 620 /**
 621  * Unmap all mapped files of an osl table.
 622  *
 623  * \param t Pointer to a mapped table.
 624  * \param flags Options for unmapping.
 625  *
 626  * \return Positive on success, negative on errors.
 627  *
 628  * \sa map_table(), enum osl_close_flags, osl_munmap().
 629  */
 630 int unmap_table(struct osl_table *t, enum osl_close_flags flags)
 631 {
 632         unsigned i;
 633         const struct osl_column_description *cd;
 634         int ret;
 635
 636         if (!t->num_mapped_columns) /* can this ever happen? */
 637                 return 1;
 638         INFO_LOG("unmapping table '%s'\n", t->desc->name);
 639         if (!t->index_map.data)
 640                 return -E_OSL_NOT_MAPPED;
 641         if (flags & OSL_MARK_CLEAN)
 642                 mark_table_clean(t);
 643         ret = osl_munmap(t->index_map.data, t->index_map.size);
 644         if (ret < 0)
 645                 return ret;
 646         t->index_map.data = NULL;
 647         if (!t->num_rows)
 648                 return 1;
 649         FOR_EACH_MAPPED_COLUMN(i, t, cd)
 650                 unmap_column(t, i);
 651         return 1;
 652 }
 653
 654 static int map_column(struct osl_table *t, unsigned col_num)
 655 {
 656         struct stat statbuf;
 657         char *filename = column_filename(t, col_num);
 658         int ret;
 659
 660         if (!filename)
 661                 return -E_OSL_NOMEM;
 662         ret = osl_stat(filename, &statbuf);
 663         if (ret < 0) {
 664                 free(filename);
 665                 return ret;
 666         }
 667         if (!(S_IFREG & statbuf.st_mode)) {
 668                 free(filename);
 669                 return ret;
 670         }
 671         ret = mmap_full_file(filename, O_RDWR,
 672                 &t->columns[col_num].data_map.data,
 673                 &t->columns[col_num].data_map.size,
 674                 NULL);
 675         free(filename);
 676         return ret;
 677 }
 678
 679 /**
 680  * Map the index file and all columns of type \p OSL_MAPPED_STORAGE into memory.
 681  *
 682  * \param t Pointer to an initialized table structure.
 683  * \param flags Mapping options.
 684  *
 685  * \return Negative return value on errors; on success the number of rows
 686  * (including invalid rows) is returned.
 687  *
 688  * \sa unmap_table(), enum map_table_flags, osl_open_table(), mmap(2).
 689  */
 690 int map_table(struct osl_table *t, enum map_table_flags flags)
 691 {
 692         char *filename;
 693         const struct osl_column_description *cd;
 694         int i = 0, ret, num_rows = 0;
 695
 696         if (!t->num_mapped_columns)
 697                 return 0;
 698         if (t->index_map.data)
 699                 return -E_OSL_ALREADY_MAPPED;
 700         filename = index_filename(t->desc);
 701         if (!filename)
 702                 return -E_OSL_NOMEM;
 703         INFO_LOG("mapping table '%s' (index: %s)\n", t->desc->name, filename);
 704         ret = mmap_full_file(filename, flags & MAP_TBL_FL_MAP_RDONLY?
 705                 O_RDONLY : O_RDWR, &t->index_map.data, &t->index_map.size, NULL);
 706         free(filename);
 707         if (ret < 0)
 708                 return ret;
 709         if (flags & MAP_TBL_FL_VERIFY_INDEX) {
 710                 ret = compare_table_descriptions(t);
 711                 if (ret < 0)
 712                         goto err;
 713         }
 714         ret = -E_OSL_BUSY;
 715         if (!(flags & MAP_TBL_FL_IGNORE_DIRTY)) {
 716                 if (table_is_dirty(t)) {
 717                         ERROR_LOG("%s is dirty\n", t->desc->name);
 718                         goto err;
 719                 }
 720         }
 721         mark_table_dirty(t);
 722         num_rows = table_num_rows(t);
 723         /* map data files */
 724         FOR_EACH_COLUMN(i, t->desc, cd) {
 725                 if (cd->storage_type == OSL_NO_STORAGE)
 726                         continue;
 727                 column_name_hash(cd->name, t->columns[i].name_hash);
 728                 if (num_rows > 0 && cd->storage_type == OSL_MAPPED_STORAGE) {
 729                         ret = map_column(t, i);
 730                         if (ret < 0)
 731                                 goto err;
 732                 }
 733         }
 734         return num_rows;
 735 err:    /* unmap what is already mapped */
 736         for (i--; i >= 0; i--) {
 737                 struct osl_object map = t->columns[i].data_map;
 738                 osl_munmap(map.data, map.size);
 739                 map.data = NULL;
 740         }
 741         osl_munmap(t->index_map.data, t->index_map.size);
 742         t->index_map.data = NULL;
 743         return ret;
 744 }
 745
 746 /**
 747  * Retrieve a mapped object by row and column number.
 748  *
 749  * \param t Pointer to an open osl table.
 750  * \param col_num Number of the mapped column containing the object to retrieve.
 751  * \param row_num Number of the row containing the object to retrieve.
 752  * \param obj The result is returned here.
 753  *
 754  * It is considered an error if \a col_num does not refer to a column
 755  * of storage type \p OSL_MAPPED_STORAGE.
 756  *
 757  * \return Standard.
 758  *
 759  * \sa osl_storage_type.
 760  */
 761 int get_mapped_object(const struct osl_table *t, unsigned col_num,
 762         uint32_t row_num, struct osl_object *obj)
 763 {
 764         struct osl_column *col = &t->columns[col_num];
 765         uint32_t offset;
 766         char *cell_index;
 767         int ret;
 768
 769         if (t->num_rows <= row_num)
 770                 return -E_OSL_BAD_ROW_NUM;
 771         ret = get_cell_index(t, row_num, col_num, &cell_index);
 772         if (ret < 0)
 773                 return ret;
 774         offset = read_u32(cell_index);
 775         obj->size = read_u32(cell_index + 4);
 776         obj->data = col->data_map.data + offset;
 777         return 1;
 778 }
 779
 780 /*
 781  * It's OK to call this with result = rb_node = NULL.  If result is not NULL,
 782  * and rb key was not found, result points to the parent node.
 783  */
 784 static int search_rbtree(const struct osl_object *obj,
 785                 const struct osl_table *t, unsigned col_num,
 786                 struct rb_node **result, struct rb_node ***rb_link)
 787 {
 788         struct osl_column *col = &t->columns[col_num];
 789         struct rb_node **new = &col->rbtree.rb_node, *parent = NULL;
 790         const struct osl_column_description *cd =
 791                 get_column_description(t->desc, col_num);
 792         enum osl_storage_type st = cd->storage_type;
 793         while (*new) {
 794                 struct osl_row *this_row = get_row_pointer(*new,
 795                         col->rbtree_num);
 796                 int ret;
 797                 struct osl_object this_obj;
 798                 parent = *new;
 799                 if (st == OSL_MAPPED_STORAGE) {
 800                         ret = get_mapped_object(t, col_num, this_row->num,
 801                                 &this_obj);
 802                         if (ret < 0)
 803                                 return ret;
 804                 } else
 805                         this_obj = this_row->volatile_objects[col->volatile_num];
 806                 ret = cd->compare_function(obj, &this_obj);
 807                 if (!ret) {
 808                         if (result)
 809                                 *result = get_rb_node_pointer(this_row,
 810                                         col->rbtree_num);
 811                         return 1;
 812                 }
 813                 if (ret < 0)
 814                         new = &((*new)->rb_left);
 815                 else
 816                         new = &((*new)->rb_right);
 817         }
 818         if (result)
 819                 *result = parent;
 820         if (rb_link)
 821                 *rb_link = new;
 822         return -E_OSL_RB_KEY_NOT_FOUND;
 823 }
 824
 825 static int insert_rbtree(struct osl_table *t, unsigned col_num,
 826         const struct osl_row *row, const struct osl_object *obj)
 827 {
 828         struct rb_node *parent, **rb_link;
 829         unsigned rbtree_num;
 830         struct rb_node *n;
 831         int ret = search_rbtree(obj, t, col_num, &parent, &rb_link);
 832
 833         if (ret > 0)
 834                 return -E_OSL_RB_KEY_EXISTS;
 835         rbtree_num = t->columns[col_num].rbtree_num;
 836         n = get_rb_node_pointer(row, rbtree_num);
 837         rb_link_node(n, parent, rb_link);
 838         rb_insert_color(n, &t->columns[col_num].rbtree);
 839         return 1;
 840 }
 841
 842 static void remove_rb_node(struct osl_table *t, unsigned col_num,
 843                 const struct osl_row *row)
 844 {
 845         struct osl_column *col = &t->columns[col_num];
 846         const struct osl_column_description *cd =
 847                 get_column_description(t->desc, col_num);
 848         enum osl_storage_flags sf = cd->storage_flags;
 849         struct rb_node *victim, *splice_out_node, *tmp;
 850         if (!(sf & OSL_RBTREE))
 851                 return;
 852         /*
 853          * Which node is removed/spliced out actually depends on how many
 854          * children the victim node has: If it has no children, it gets
 855          * deleted. If it has one child, it gets spliced out. If it has two
 856          * children, its successor (which has at most a right child) gets
 857          * spliced out.
 858          */
 859         victim = get_rb_node_pointer(row, col->rbtree_num);
 860         if (victim->rb_left && victim->rb_right)
 861                 splice_out_node = rb_next(victim);
 862         else
 863                 splice_out_node = victim;
 864         /* Go up to the root and decrement the size of each node in the path. */
 865         for (tmp = splice_out_node; tmp; tmp = rb_parent(tmp))
 866                 tmp->size--;
 867         rb_erase(victim, &col->rbtree);
 868 }
 869
 870 static int add_row_to_rbtrees(struct osl_table *t, uint32_t row_num,
 871                 struct osl_object *volatile_objs, struct osl_row **row_ptr)
 872 {
 873         unsigned i;
 874         int ret;
 875         struct osl_row *row = allocate_row(t->num_rbtrees);
 876         const struct osl_column_description *cd;
 877
 878         if (!row)
 879                 return -E_OSL_NOMEM;
 880         row->num = row_num;
 881         row->volatile_objects = volatile_objs;
 882         FOR_EACH_RBTREE_COLUMN(i, t, cd) {
 883                 if (cd->storage_type == OSL_MAPPED_STORAGE) {
 884                         struct osl_object obj;
 885                         ret = get_mapped_object(t, i, row_num, &obj);
 886                         if (ret < 0)
 887                                 goto err;
 888                         ret = insert_rbtree(t, i, row, &obj);
 889                 } else { /* volatile */
 890                         const struct osl_object *obj
 891                                 = volatile_objs + t->columns[i].volatile_num;
 892                         ret = insert_rbtree(t, i, row, obj);
 893                 }
 894                 if (ret < 0)
 895                         goto err;
 896         }
 897         if (row_ptr)
 898                 *row_ptr = row;
 899         return 1;
 900 err: /* rollback changes, i.e. remove added entries from rbtrees */
 901         while (i)
 902                 remove_rb_node(t, i--, row);
 903         free(row);
 904         return ret;
 905 }
 906
 907 static void free_volatile_objects(const struct osl_table *t,
 908                 enum osl_close_flags flags)
 909 {
 910         int i, j;
 911         struct rb_node *n;
 912         struct osl_column *rb_col;
 913         const struct osl_column_description *cd;
 914
 915         if (!t->num_volatile_columns)
 916                 return;
 917         /* find the first rbtree column (any will do) */
 918         FOR_EACH_RBTREE_COLUMN(i, t, cd)
 919                 break;
 920         rb_col = t->columns + i;
 921         /* walk that rbtree and free all volatile objects */
 922         for (n = rb_first(&rb_col->rbtree); n; n = rb_next(n)) {
 923                 struct osl_row *r = get_row_pointer(n, rb_col->rbtree_num);
 924                 if (flags & OSL_FREE_VOLATILE)
 925                         FOR_EACH_VOLATILE_COLUMN(j, t, cd) {
 926                                 if (cd->storage_flags & OSL_DONT_FREE)
 927                                         continue;
 928                                 free(r->volatile_objects[
 929                                         t->columns[j].volatile_num].data);
 930                         }
 931 //                      for (j = 0; j < t->num_volatile_columns; j++)
 932 //                              free(r->volatile_objects[j].data);
 933                 free(r->volatile_objects);
 934         }
 935 }
 936
 937 /**
 938  * Erase all rbtree nodes and free resources.
 939  *
 940  * \param t Pointer to an open osl table.
 941  *
 942  * This function is called by osl_close_table().
 943  */
 944 void clear_rbtrees(struct osl_table *t)
 945 {
 946         const struct osl_column_description *cd;
 947         unsigned i, rbtrees_cleared = 0;
 948
 949         FOR_EACH_RBTREE_COLUMN(i, t, cd) {
 950                 struct osl_column *col = &t->columns[i];
 951                 struct rb_node *n;
 952                 rbtrees_cleared++;
 953                 for (n = rb_first(&col->rbtree); n;) {
 954                         struct osl_row *r;
 955                         rb_erase(n, &col->rbtree);
 956                         if (rbtrees_cleared == t->num_rbtrees) {
 957                                 r = get_row_pointer(n, col->rbtree_num);
 958                                 n = rb_next(n);
 959                                 free(r);
 960                         } else
 961                                 n = rb_next(n);
 962                 }
 963         }
 964
 965 }
 966
 967 __export int osl_close_table(struct osl_table *t, enum osl_close_flags flags)
 968 {
 969         int ret;
 970
 971         if (!t)
 972                 return -E_OSL_BAD_TABLE;
 973         NOTICE_LOG("closing table %s\n", t->desc->name);
 974         free_volatile_objects(t, flags);
 975         clear_rbtrees(t);
 976         ret = unmap_table(t, flags);
 977         if (ret < 0)
 978                 ERROR_LOG("unmap_table failed: %d\n", ret);
 979         free(t->columns);
 980         free(t);
 981         return ret;
 982 }
 983
 984 /**
 985  * Find out whether the given row number corresponds to an invalid row.
 986  *
 987  * \param t Pointer to the osl table.
 988  * \param row_num The number of the row in question.
 989  *
 990  * By definition, a row is considered invalid if all its index entries
 991  * are invalid.
 992  *
 993  * \return Positive if \a row_num corresponds to an invalid row,
 994  * zero if it corresponds to a valid row, negative on errors.
 995  */
 996 int row_is_invalid(struct osl_table *t, uint32_t row_num)
 997 {
 998         char *row_index;
 999         int i, ret = get_row_index(t, row_num, &row_index);
1000
1001         if (ret < 0)
1002                 return ret;
1003         for (i = 0; i < t->row_index_size; i++) {
1004                 if ((unsigned char)row_index[i] != 0xff)
1005                         return 0;
1006         }
1007         INFO_LOG("row %d is invalid\n", row_num);
1008         return 1;
1009 }
1010
1011 /**
1012  * Invalidate a row of an osl table.
1013  *
1014  * \param t Pointer to an open osl table.
1015  * \param row_num Number of the row to mark as invalid.
1016  *
1017  * This function marks each mapped object in the index entry of \a row as
1018  * invalid.
1019  *
1020  * \return Standard.
1021  */
1022 int mark_row_invalid(struct osl_table *t, uint32_t row_num)
1023 {
1024         char *row_index;
1025         int ret = get_row_index(t, row_num, &row_index);
1026
1027         if (ret < 0)
1028                 return ret;
1029         INFO_LOG("marking row %d as invalid\n", row_num);
1030         memset(row_index, 0xff, t->row_index_size);
1031         return 1;
1032 }
1033
1034 /**
1035  * Initialize all rbtrees and compute number of invalid rows.
1036  *
1037  * \param t The table containing the rbtrees to be initialized.
1038  *
1039  * \return Standard.
1040  */
1041 int init_rbtrees(struct osl_table *t)
1042 {
1043         int i, ret;
1044         const struct osl_column_description *cd;
1045
1046         /* create rbtrees */
1047         FOR_EACH_RBTREE_COLUMN(i, t, cd)
1048                 t->columns[i].rbtree = RB_ROOT;
1049         /* add valid rows to rbtrees */
1050         t->num_invalid_rows = 0;
1051         for (i = 0; i < t->num_rows; i++) {
1052                 struct osl_object *volatile_objs;
1053                 ret = row_is_invalid(t, i);
1054                 if (ret < 0)
1055                         return ret;
1056                 if (ret) {
1057                         t->num_invalid_rows++;
1058                         continue;
1059                 }
1060                 if (t->num_volatile_columns > 0) {
1061                         volatile_objs = calloc(t->num_volatile_columns,
1062                                 sizeof(struct osl_object));
1063                         if (!volatile_objs)
1064                                 return -E_OSL_NOMEM;
1065                 } else
1066                         volatile_objs = NULL;
1067                 ret = add_row_to_rbtrees(t, i, volatile_objs, NULL);
1068                 if (ret < 0)
1069                         return ret;
1070         }
1071         return 1;
1072 }
1073
1074 __export int osl_open_table(const struct osl_table_description *table_desc,
1075                 struct osl_table **result)
1076 {
1077         int ret;
1078         struct osl_table *t;
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         ret = map_table(t, MAP_TBL_FL_VERIFY_INDEX);
1085         if (ret < 0)
1086                 goto err;
1087         t->num_rows = ret;
1088         DEBUG_LOG("num rows: %d\n", t->num_rows);
1089         ret = init_rbtrees(t);
1090         if (ret < 0) {
1091                 osl_close_table(t, OSL_MARK_CLEAN); /* ignore further errors */
1092                 return ret;
1093         }
1094         *result = t;
1095         return 1;
1096 err:
1097         free(t->columns);
1098         free(t);
1099         return ret;
1100 }
1101
1102 static int create_disk_storage_object_dir(const struct osl_table *t,
1103                 unsigned col_num, const char *ds_name)
1104 {
1105         char *dirname;
1106         int ret;
1107
1108         if (!(t->desc->flags & OSL_LARGE_TABLE))
1109                 return 1;
1110         dirname = disk_storage_dirname(t, col_num, ds_name);
1111         if (!dirname)
1112                 return -E_OSL_NOMEM;
1113         ret = osl_mkdir(dirname, 0777);
1114         free(dirname);
1115         if (ret < 0 && ret != -E_OSL_DIR_EXISTS)
1116                 return ret;
1117         return 1;
1118 }
1119
1120 static int write_disk_storage_file(const struct osl_table *t, unsigned col_num,
1121         const struct osl_object *obj, const char *ds_name)
1122 {
1123         int ret;
1124         char *filename;
1125
1126         ret = create_disk_storage_object_dir(t, col_num, ds_name);
1127         if (ret < 0)
1128                 return ret;
1129         filename = disk_storage_path(t, col_num, ds_name);
1130         if (!filename)
1131                 return -E_OSL_NOMEM;
1132         ret = write_file(filename, obj->data, obj->size);
1133         free(filename);
1134         return ret;
1135 }
1136
1137 static int append_map_file(const struct osl_table *t, unsigned col_num,
1138         const struct osl_object *obj, uint32_t *new_size)
1139 {
1140         char *filename = column_filename(t, col_num);
1141         int ret;
1142
1143         if (!filename)
1144                 return -E_OSL_NOMEM;
1145         ret = append_file(filename, obj->data, obj->size, new_size);
1146         free(filename);
1147         return ret;
1148 }
1149
1150 static int append_row_index(const struct osl_table *t, char *row_index)
1151 {
1152         char *filename;
1153         int ret;
1154
1155         if (!t->num_mapped_columns)
1156                 return 1;
1157         filename = index_filename(t->desc);
1158         if (!filename)
1159                 return -E_OSL_NOMEM;
1160         ret = append_file(filename, row_index, t->row_index_size, NULL);
1161         free(filename);
1162         return ret;
1163 }
1164
1165 static int truncate_mapped_file(const struct osl_table *t, unsigned col_num,
1166                 off_t size)
1167 {
1168         int ret;
1169         char *filename = column_filename(t, col_num);
1170
1171         if (!filename)
1172                 return -E_OSL_NOMEM;
1173         ret = truncate_file(filename, size);
1174         free(filename);
1175         return ret;
1176 }
1177
1178 static int delete_disk_storage_file(const struct osl_table *t, unsigned col_num,
1179                 const char *ds_name)
1180 {
1181         char *dirname, *filename = disk_storage_path(t, col_num, ds_name);
1182         int ret = 1;
1183
1184         if (!filename)
1185                 return -E_OSL_NOMEM;
1186         if (unlink(filename) < 0)
1187                 ret = errno == ENOENT? -E_OSL_NOENT : -E_OSL_UNLINK;
1188         free(filename);
1189         if (ret < 0)
1190                 return ret;
1191         if (!(t->desc->flags & OSL_LARGE_TABLE))
1192                 return 1;
1193         dirname = disk_storage_dirname(t, col_num, ds_name);
1194         if (!dirname)
1195                 return -E_OSL_NOMEM;
1196         rmdir(dirname);
1197         free(dirname);
1198         return 1;
1199 }
1200
1201 __export int osl_add_and_get_row(struct osl_table *t, struct osl_object *objects,
1202                 struct osl_row **row)
1203 {
1204         int i, ret;
1205         char *ds_name = NULL;
1206         struct rb_node **rb_parents = NULL, ***rb_links = NULL;
1207         char *new_row_index = NULL;
1208         struct osl_object *volatile_objs = NULL;
1209         const struct osl_column_description *cd;
1210
1211         if (!t)
1212                 return -E_OSL_BAD_TABLE;
1213         rb_parents = malloc(t->num_rbtrees * sizeof(struct rn_node*));
1214         if (!rb_parents)
1215                 return -E_OSL_NOMEM;
1216         rb_links = malloc(t->num_rbtrees * sizeof(struct rn_node**));
1217         if (!rb_links) {
1218                 free(rb_parents);
1219                 return -E_OSL_NOMEM;
1220         }
1221         if (t->num_mapped_columns) {
1222                 new_row_index = malloc(t->row_index_size);
1223                 if (!new_row_index) {
1224                         free(rb_links);
1225                         free(rb_parents);
1226                         return -E_OSL_NOMEM;
1227                 }
1228         }
1229         /* pass 1: sanity checks */
1230 //      DEBUG_LOG("sanity tests: %p:%p\n", objects[0].data,
1231 //              objects[1].data);
1232         FOR_EACH_COLUMN(i, t->desc, cd) {
1233                 enum osl_storage_type st = cd->storage_type;
1234                 enum osl_storage_flags sf = cd->storage_flags;
1235
1236 //              ret = -E_OSL_NULL_OBJECT;
1237 //              if (!objects[i])
1238 //                      goto out;
1239                 if (st == OSL_DISK_STORAGE)
1240                         continue;
1241                 if (sf & OSL_RBTREE) {
1242                         unsigned rbtree_num = t->columns[i].rbtree_num;
1243                         ret = -E_OSL_RB_KEY_EXISTS;
1244 //                      DEBUG_LOG("checking whether %p exists\n",
1245 //                              objects[i].data);
1246                         if (search_rbtree(objects + i, t, i,
1247                                         &rb_parents[rbtree_num],
1248                                         &rb_links[rbtree_num]) > 0)
1249                                 goto out;
1250                 }
1251                 if (sf & OSL_FIXED_SIZE) {
1252 //                      DEBUG_LOG("fixed size. need: %zu, have: %d\n",
1253 //                              objects[i].size, cd->data_size);
1254                         ret = -E_OSL_BAD_DATA_SIZE;
1255                         if (objects[i].size != cd->data_size)
1256                                 goto out;
1257                 }
1258         }
1259         if (t->num_disk_storage_columns) {
1260                 ds_name = disk_storage_name_of_object(t,
1261                         &objects[t->disk_storage_name_column]);
1262                 ret = -E_OSL_NOMEM;
1263                 if (!ds_name)
1264                         goto out;
1265         }
1266         ret = unmap_table(t, OSL_MARK_CLEAN);
1267         if (ret < 0)
1268                 goto out;
1269 //      DEBUG_LOG("sanity tests passed%s\n", "");
1270         /* pass 2: create data files, append map data */
1271         FOR_EACH_COLUMN(i, t->desc, cd) {
1272                 enum osl_storage_type st = cd->storage_type;
1273                 if (st == OSL_NO_STORAGE)
1274                         continue;
1275                 if (st == OSL_MAPPED_STORAGE) {
1276                         uint32_t new_size;
1277                         struct osl_column *col = &t->columns[i];
1278 //                      DEBUG_LOG("appending object of size %zu\n",
1279 //                              objects[i].size);
1280                         ret = append_map_file(t, i, objects + i, &new_size);
1281                         if (ret < 0)
1282                                 goto rollback;
1283                         update_cell_index(new_row_index, col, new_size,
1284                                 objects[i].size);
1285                         continue;
1286                 }
1287                 /* DISK_STORAGE */
1288                 ret = write_disk_storage_file(t, i, objects + i, ds_name);
1289                 if (ret < 0)
1290                         goto rollback;
1291         }
1292         ret = append_row_index(t, new_row_index);
1293         if (ret < 0)
1294                 goto rollback;
1295         ret = map_table(t, MAP_TBL_FL_VERIFY_INDEX);
1296         if (ret < 0) { /* truncate index and rollback changes */
1297                 char *filename = index_filename(t->desc);
1298                 if (filename)
1299                         truncate_file(filename, t->row_index_size);
1300                 free(filename);
1301                 goto rollback;
1302         }
1303         /* pass 3: add entry to rbtrees */
1304         if (t->num_volatile_columns) {
1305                 ret = -E_OSL_NOMEM;
1306                 volatile_objs = calloc(t->num_volatile_columns,
1307                         sizeof(struct osl_object));
1308                 if (!volatile_objs)
1309                         goto out;
1310                 FOR_EACH_VOLATILE_COLUMN(i, t, cd)
1311                         volatile_objs[t->columns[i].volatile_num] = objects[i];
1312         }
1313         t->num_rows++;
1314 //      DEBUG_LOG("adding new entry as row #%d\n", t->num_rows - 1);
1315         ret = add_row_to_rbtrees(t, t->num_rows - 1, volatile_objs, row);
1316         if (ret < 0)
1317                 goto out;
1318 //      DEBUG_LOG("added new entry as row #%d\n", t->num_rows - 1);
1319         ret = 1;
1320         goto out;
1321 rollback: /* rollback all changes made, ignore further errors */
1322         for (i--; i >= 0; i--) {
1323                 cd = get_column_description(t->desc, i);
1324                 enum osl_storage_type st = cd->storage_type;
1325                 if (st == OSL_NO_STORAGE)
1326                         continue;
1327
1328                 if (st == OSL_MAPPED_STORAGE)
1329                         truncate_mapped_file(t, i, objects[i].size);
1330                 else /* disk storage */
1331                         delete_disk_storage_file(t, i, ds_name);
1332         }
1333         /* ignore error and return previous error value */
1334         map_table(t, MAP_TBL_FL_VERIFY_INDEX);
1335 out:
1336         free(new_row_index);
1337         free(ds_name);
1338         free(rb_parents);
1339         free(rb_links);
1340         return ret;
1341 }
1342
1343 __export int osl_add_row(struct osl_table *t, struct osl_object *objects)
1344 {
1345         return osl_add_and_get_row(t, objects, NULL);
1346 }
1347
1348 __export int osl_get_object(const struct osl_table *t, const struct osl_row *r,
1349         unsigned col_num, struct osl_object *object)
1350 {
1351         const struct osl_column_description *cd;
1352
1353         if (!t)
1354                 return -E_OSL_BAD_TABLE;
1355         cd = get_column_description(t->desc, col_num);
1356         /* col must not be disk storage */
1357         if (cd->storage_type == OSL_DISK_STORAGE)
1358                 return -E_OSL_BAD_STORAGE_TYPE;
1359         if (cd->storage_type == OSL_MAPPED_STORAGE)
1360                 return get_mapped_object(t, col_num, r->num, object);
1361         /* volatile */
1362         *object = r->volatile_objects[t->columns[col_num].volatile_num];
1363         return 1;
1364 }
1365
1366 __export int osl_del_row(struct osl_table *t, struct osl_row *row)
1367 {
1368         struct osl_row *r = row;
1369         int i, ret;
1370         const struct osl_column_description *cd;
1371
1372         if (!t)
1373                 return -E_OSL_BAD_TABLE;
1374         INFO_LOG("deleting row %p\n", row);
1375
1376         if (t->num_disk_storage_columns) {
1377                 char *ds_name;
1378                 ret = disk_storage_name_of_row(t, r, &ds_name);
1379                 if (ret < 0)
1380                         goto out;
1381                 FOR_EACH_DISK_STORAGE_COLUMN(i, t, cd)
1382                         delete_disk_storage_file(t, i, ds_name);
1383                 free(ds_name);
1384         }
1385         FOR_EACH_COLUMN(i, t->desc, cd) {
1386                 struct osl_column *col = t->columns + i;
1387                 enum osl_storage_type st = cd->storage_type;
1388                 remove_rb_node(t, i, r);
1389                 if (st == OSL_MAPPED_STORAGE)
1390                         continue;
1391                 if (st == OSL_NO_STORAGE && !(cd->storage_flags & OSL_DONT_FREE))
1392                         free(r->volatile_objects[col->volatile_num].data);
1393         }
1394         if (t->num_mapped_columns) {
1395                 ret = mark_row_invalid(t, r->num);
1396                 if (ret < 0)
1397                         goto out;
1398                 t->num_invalid_rows++;
1399         } else
1400                 t->num_rows--;
1401         ret = 1;
1402 out:
1403         free(r->volatile_objects);
1404         free(r);
1405         return ret;
1406 }
1407
1408 /* test if column has an rbtree */
1409 static int check_rbtree_col(const struct osl_table *t, unsigned col_num,
1410                 struct osl_column **col)
1411 {
1412         if (!t)
1413                 return -E_OSL_BAD_TABLE;
1414         if (!(get_column_description(t->desc, col_num)->storage_flags & OSL_RBTREE))
1415                 return -E_OSL_BAD_STORAGE_FLAGS;
1416         *col = t->columns + col_num;
1417         return 1;
1418 }
1419
1420 __export int osl_get_row(const struct osl_table *t, unsigned col_num,
1421                 const struct osl_object *obj, struct osl_row **result)
1422 {
1423         int ret;
1424         struct rb_node *node;
1425         struct osl_row *row;
1426         struct osl_column *col;
1427
1428         *result = NULL;
1429         ret = check_rbtree_col(t, col_num, &col);
1430         if (ret < 0)
1431                 return ret;
1432         ret = search_rbtree(obj, t, col_num, &node, NULL);
1433         if (ret < 0)
1434                 return ret;
1435         row = get_row_pointer(node, t->columns[col_num].rbtree_num);
1436         *result = row;
1437         return 1;
1438 }
1439
1440 static int rbtree_loop(struct osl_column *col, void *private_data,
1441                 osl_rbtree_loop_func *func)
1442 {
1443         struct rb_node *n, *tmp;
1444
1445         /* this for-loop is safe against removal of an entry */
1446         for (n = rb_first(&col->rbtree), tmp = n? rb_next(n) : NULL;
1447                         n;
1448                         n = tmp, tmp = tmp? rb_next(tmp) : NULL) {
1449                 struct osl_row *r = get_row_pointer(n, col->rbtree_num);
1450                 if (func(r, private_data) < 0)
1451                         return -E_OSL_LOOP;
1452         }
1453         return 1;
1454 }
1455
1456 static int rbtree_loop_reverse(struct osl_column *col, void *private_data,
1457                 osl_rbtree_loop_func *func)
1458 {
1459         struct rb_node *n, *tmp;
1460
1461         /* safe against removal of an entry */
1462         for (n = rb_last(&col->rbtree), tmp = n? rb_prev(n) : NULL;
1463                         n;
1464                         n = tmp, tmp = tmp? rb_prev(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 __export int osl_rbtree_loop(const struct osl_table *t, unsigned col_num,
1473         void *private_data, osl_rbtree_loop_func *func)
1474 {
1475         struct osl_column *col;
1476
1477         int ret = check_rbtree_col(t, col_num, &col);
1478         if (ret < 0)
1479                 return ret;
1480         return rbtree_loop(col, private_data, func);
1481 }
1482
1483 __export int osl_rbtree_loop_reverse(const struct osl_table *t, unsigned col_num,
1484         void *private_data, osl_rbtree_loop_func *func)
1485 {
1486         struct osl_column *col;
1487
1488         int ret = check_rbtree_col(t, col_num, &col);
1489         if (ret < 0)
1490                 return ret;
1491         return rbtree_loop_reverse(col, private_data, func);
1492 }
1493
1494 /* TODO: Rollback changes on errors */
1495 static int rename_disk_storage_objects(struct osl_table *t,
1496                 struct osl_object *old_obj, struct osl_object *new_obj)
1497 {
1498         int i, ret;
1499         const struct osl_column_description *cd;
1500         char *old_ds_name, *new_ds_name;
1501
1502         if (!t->num_disk_storage_columns)
1503                 return 1; /* nothing to do */
1504         if (old_obj->size == new_obj->size && !memcmp(new_obj->data,
1505                         old_obj->data, new_obj->size))
1506                 return 1; /* object did not change */
1507         old_ds_name = disk_storage_name_of_object(t, old_obj);
1508         new_ds_name = disk_storage_name_of_object(t, new_obj);
1509         ret = -E_OSL_NOMEM;
1510         if (!old_ds_name || ! new_ds_name)
1511                 goto out;
1512
1513         FOR_EACH_DISK_STORAGE_COLUMN(i, t, cd) {
1514                 char *old_filename, *new_filename;
1515                 ret = create_disk_storage_object_dir(t, i, new_ds_name);
1516                 if (ret < 0)
1517                         goto out;
1518                 old_filename = disk_storage_path(t, i, old_ds_name);
1519                 new_filename = disk_storage_path(t, i, new_ds_name);
1520                 if (!old_filename || !new_filename)
1521                         ret = -E_OSL_NOMEM;
1522                 else
1523                         ret = osl_rename(old_filename, new_filename);
1524                 free(old_filename);
1525                 free(new_filename);
1526                 if (ret < 0)
1527                         goto out;
1528         }
1529         ret = 1;
1530 out:
1531         free(old_ds_name);
1532         free(new_ds_name);
1533         return ret;
1534
1535 }
1536
1537 __export int osl_update_object(struct osl_table *t, const struct osl_row *r,
1538                 unsigned col_num, struct osl_object *obj)
1539 {
1540         struct osl_column *col;
1541         const struct osl_column_description *cd;
1542         int ret;
1543
1544         if (!t)
1545                 return -E_OSL_BAD_TABLE;
1546         col = &t->columns[col_num];
1547         cd = get_column_description(t->desc, col_num);
1548         DEBUG_LOG("updating column %u of %s\n", col_num, t->desc->name);
1549         if (cd->storage_flags & OSL_RBTREE) {
1550                 if (search_rbtree(obj, t, col_num, NULL, NULL) > 0)
1551                         return -E_OSL_RB_KEY_EXISTS;
1552         }
1553         if (cd->storage_flags & OSL_FIXED_SIZE) {
1554                 if (obj->size != cd->data_size)
1555                         return -E_OSL_BAD_DATA_SIZE;
1556         }
1557         remove_rb_node(t, col_num, r);
1558         if (cd->storage_type == OSL_NO_STORAGE) { /* TODO: If fixed size, reuse object? */
1559                 if (!(cd->storage_flags & OSL_DONT_FREE))
1560                         free(r->volatile_objects[col->volatile_num].data);
1561                 r->volatile_objects[col->volatile_num] = *obj;
1562         } else if (cd->storage_type == OSL_DISK_STORAGE) {
1563                 char *ds_name;
1564                 ret = disk_storage_name_of_row(t, r, &ds_name);
1565                 if (ret < 0)
1566                         return ret;
1567                 ret = delete_disk_storage_file(t, col_num, ds_name);
1568                 if (ret < 0 && ret != -E_OSL_NOENT) {
1569                         free(ds_name);
1570                         return ret;
1571                 }
1572                 ret = write_disk_storage_file(t, col_num, obj, ds_name);
1573                 free(ds_name);
1574                 if (ret < 0)
1575                         return ret;
1576         } else { /* mapped storage */
1577                 struct osl_object old_obj;
1578                 ret = get_mapped_object(t, col_num, r->num, &old_obj);
1579                 if (ret < 0)
1580                         return ret;
1581                 /*
1582                  * If the updated column is the disk storage name column, the
1583                  * disk storage name changes, so we have to rename all disk
1584                  * storage objects accordingly.
1585                  */
1586                 if (col_num == t->disk_storage_name_column) {
1587                         ret = rename_disk_storage_objects(t, &old_obj, obj);
1588                         if (ret < 0)
1589                                 return ret;
1590                 }
1591                 if (cd->storage_flags & OSL_FIXED_SIZE)
1592                         memcpy(old_obj.data, obj->data, cd->data_size);
1593                 else { /* TODO: if the size doesn't change, use old space */
1594                         uint32_t new_data_map_size;
1595                         char *row_index;
1596                         ret = get_row_index(t, r->num, &row_index);
1597                         if (ret < 0)
1598                                 return ret;
1599                         unmap_column(t, col_num);
1600                         ret = append_map_file(t, col_num, obj,
1601                                 &new_data_map_size);
1602                         if (ret < 0)
1603                                 return ret;
1604                         ret = map_column(t, col_num);
1605                         if (ret < 0)
1606                                 return ret;
1607                         update_cell_index(row_index, col, new_data_map_size,
1608                                 obj->size);
1609                 }
1610         }
1611         if (cd->storage_flags & OSL_RBTREE) {
1612                 ret = insert_rbtree(t, col_num, r, obj);
1613                 if (ret < 0)
1614                         return ret;
1615         }
1616         return 1;
1617 }
1618
1619 __export int osl_open_disk_object(const struct osl_table *t, const struct osl_row *r,
1620                 unsigned col_num, struct osl_object *obj)
1621 {
1622         const struct osl_column_description *cd;
1623         char *ds_name, *filename;
1624         int ret;
1625
1626         if (!t)
1627                 return -E_OSL_BAD_TABLE;
1628         cd = get_column_description(t->desc, col_num);
1629         if (cd->storage_type != OSL_DISK_STORAGE)
1630                 return -E_OSL_BAD_STORAGE_TYPE;
1631
1632         ret = disk_storage_name_of_row(t, r, &ds_name);
1633         if (ret < 0)
1634                 return ret;
1635         filename = disk_storage_path(t, col_num, ds_name);
1636         free(ds_name);
1637         if (!filename)
1638                 return -E_OSL_NOMEM;
1639         DEBUG_LOG("filename: %s\n", filename);
1640         ret = mmap_full_file(filename, O_RDONLY, &obj->data, &obj->size, NULL);
1641         free(filename);
1642         return ret;
1643 }
1644
1645 __export int osl_close_disk_object(struct osl_object *obj)
1646 {
1647         return osl_munmap(obj->data, obj->size);
1648 }
1649
1650 __export int osl_get_num_rows(const struct osl_table *t, unsigned *num_rows)
1651 {
1652         if (!t)
1653                 return -E_OSL_BAD_TABLE;
1654         assert(t->num_rows >= t->num_invalid_rows);
1655         *num_rows = t->num_rows - t->num_invalid_rows;
1656         return 1;
1657 }
1658
1659 __export int osl_get_rank(const struct osl_table *t, struct osl_row *r,
1660                 unsigned col_num, unsigned *rank)
1661 {
1662         struct osl_object obj;
1663         struct osl_column *col;
1664         struct rb_node *node;
1665         int ret = check_rbtree_col(t, col_num, &col);
1666
1667         if (ret < 0)
1668                 return ret;
1669         ret = osl_get_object(t, r, col_num, &obj);
1670         if (ret < 0)
1671                 return ret;
1672         ret = search_rbtree(&obj, t, col_num, &node, NULL);
1673         if (ret < 0)
1674                 return ret;
1675         ret = rb_rank(node, rank);
1676         if (ret < 0)
1677                 return -E_OSL_BAD_ROW;
1678         return 1;
1679 }
1680
1681 __export int osl_get_nth_row(const struct osl_table *t, unsigned col_num,
1682                 unsigned n, struct osl_row **result)
1683 {
1684         struct osl_column *col;
1685         struct rb_node *node;
1686         unsigned num_rows;
1687         int ret;
1688
1689         *result = NULL;
1690         if (n == 0)
1691                 return -E_OSL_RB_KEY_NOT_FOUND;
1692         ret = osl_get_num_rows(t, &num_rows);
1693         if (ret < 0)
1694                 return ret;
1695         if (n > num_rows)
1696                 return -E_OSL_RB_KEY_NOT_FOUND;
1697         ret = check_rbtree_col(t, col_num, &col);
1698         if (ret < 0)
1699                 return ret;
1700         node = rb_nth(col->rbtree.rb_node, n);
1701         if (!node)
1702                 return -E_OSL_RB_KEY_NOT_FOUND;
1703         *result = get_row_pointer(node, col->rbtree_num);
1704         return 1;
1705 }
1706
1707 __export int osl_rbtree_first_row(const struct osl_table *t, unsigned col_num,
1708                 struct osl_row **result)
1709 {
1710         return osl_get_nth_row(t, col_num, 1, result);
1711 }
1712
1713 __export int osl_rbtree_last_row(const struct osl_table *t, unsigned col_num,
1714                 struct osl_row **result)
1715 {
1716         unsigned num_rows;
1717         int ret = osl_get_num_rows(t, &num_rows);
1718
1719         if (ret < 0)
1720                 return ret;
1721         return osl_get_nth_row(t, col_num, num_rows, result);
1722 }