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