osl.c

   1 /*
   2  * Copyright (C) 2007-2008 Andre Noll <maan@systemlinux.org>
   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 "error.h"
  14 #include "util.h"
  15 #include "osl_core.h"
  16
  17 /* Taken from Drepper: How to write shared libraries, Appendix B. */
  18 #include <stddef.h>
  19 #define MSGSTRFIELD(line) MSGSTRFIELD1(line)
  20 #define MSGSTRFIELD1(line) str##line
  21 static const union msgstr_t {
  22         struct {
  23 #define OSL_ERROR(n, s) char MSGSTRFIELD(__LINE__)[sizeof(s)];
  24 #include "errtab.h"
  25 #undef OSL_ERROR
  26         };
  27         char str[0];
  28 } msgstr = { {
  29 #define OSL_ERROR(n, s) s,
  30 #include "errtab.h"
  31 #undef OSL_ERROR
  32 } };
  33 static const unsigned int errmsgidx[] = {
  34 #define OSL_ERROR(n, s) [n] = offsetof(union msgstr_t, MSGSTRFIELD(__LINE__)),
  35 #include "errtab.h"
  36 #undef OSL_ERROR
  37 };
  38
  39 __export const char *osl_strerror(int num)
  40 {
  41         if (IS_SYSTEM_ERROR(num))
  42                 return strerror((num) & ((1 << SYSTEM_ERROR_BIT) - 1));
  43         return msgstr.str + errmsgidx[num];
  44 }
  45
  46 int loglevel = 0;
  47
  48 static void __attribute ((constructor)) init_loglevel(void)
  49 {
  50         char *p = getenv("OSL_LOGLEVEL");
  51
  52         /* don't log anything if unset */
  53         loglevel = p? atoi(p) : EMERG + 1;
  54 }
  55
  56 /**
  57  * The log function.
  58  *
  59  * \param ll Loglevel.
  60  * \param fmt Usual format string.
  61  *
  62  * All XXX_LOG() macros use this function.
  63  */
  64 __printf_2_3 void __log(int ll, const char* fmt,...)
  65 {
  66         va_list argp;
  67         FILE *outfd;
  68         struct tm *tm;
  69         time_t t1;
  70         char str[255] = "";
  71
  72         if (ll < loglevel)
  73                 return;
  74         outfd = stderr;
  75         time(&t1);
  76         tm = localtime(&t1);
  77         strftime(str, sizeof(str), "%b %d %H:%M:%S", tm);
  78         fprintf(outfd, "%s ", str);
  79         va_start(argp, fmt);
  80         vfprintf(outfd, fmt, argp);
  81         va_end(argp);
  82 }
  83
  84 /**
  85  * A wrapper for lseek(2).
  86  *
  87  * \param fd The file descriptor whose offset is to be to repositioned.
  88  * \param offset A value-result parameter.
  89  * \param whence Usual repositioning directive.
  90  *
  91  * Reposition the offset of the file descriptor \a fd to the argument \a offset
  92  * according to the directive \a whence. Upon successful return, \a offset
  93  * contains the resulting offset location as measured in bytes from the
  94  * beginning of the file.
  95  *
  96  * \return Positive on success. Otherwise, the function returns \p -E_OSL_LSEEK.
  97  *
  98  * \sa lseek(2).
  99  */
 100 static int __lseek(int fd, off_t *offset, int whence)
 101 {
 102         *offset = lseek(fd, *offset, whence);
 103         int ret = -E_OSL_LSEEK;
 104         if (*offset == -1)
 105                 return ret;
 106         return 1;
 107 }
 108
 109 static int append_file(const char *filename, char *data, size_t data_size,
 110                 uint32_t *new_pos)
 111 {
 112         int ret, fd;
 113
 114 //      DEBUG_LOG("appending %zu  + %zu bytes\n", header_size, data_size);
 115         ret = osl_open(filename, O_WRONLY | O_CREAT | O_APPEND, 0644);
 116         if (ret < 0)
 117                 return ret;
 118         fd = ret;
 119         ret = write_all(fd, data, &data_size);
 120         if (ret < 0)
 121                 goto out;
 122         if (new_pos) {
 123                 off_t offset = 0;
 124                 ret = __lseek(fd, &offset, SEEK_END);
 125                 if (ret < 0)
 126                         goto out;
 127 //              DEBUG_LOG("new file size: " FMT_OFF_T "\n", offset);
 128                 *new_pos = offset;
 129         }
 130         ret = 1;
 131 out:
 132         close(fd);
 133         return ret;
 134 }
 135
 136 static int verify_name(const char *name)
 137 {
 138         if (!name)
 139                 return -E_OSL_BAD_NAME;
 140         if (!*name)
 141                 return -E_OSL_BAD_NAME;
 142         if (strchr(name, '/'))
 143                 return -E_OSL_BAD_NAME;
 144         if (!strcmp(name, ".."))
 145                 return -E_OSL_BAD_NAME;
 146         if (!strcmp(name, "."))
 147                 return -E_OSL_BAD_NAME;
 148         return 1;
 149 }
 150
 151 int osl_hash_compare(const struct osl_object *obj1, const struct osl_object *obj2)
 152 {
 153         return hash_compare((HASH_TYPE *)obj1->data, (HASH_TYPE *)obj2->data);
 154 }
 155
 156 static char *disk_storage_dirname(const struct osl_table *t, unsigned col_num,
 157                 const char *ds_name)
 158 {
 159         char *dirname, *column_name = column_filename(t, col_num);
 160
 161         if (!column_name)
 162                 return NULL;
 163         if (!(t->desc->flags & OSL_LARGE_TABLE))
 164                 return column_name;
 165         dirname = make_message("%s/%.2s", column_name, ds_name);
 166         free(column_name);
 167         return dirname;
 168 }
 169
 170 static char *disk_storage_name_of_object(const struct osl_table *t,
 171         const struct osl_object *obj)
 172 {
 173         HASH_TYPE hash[HASH_SIZE];
 174         hash_object(obj, hash);
 175         return disk_storage_name_of_hash(t, hash);
 176 }
 177
 178 static int disk_storage_name_of_row(const struct osl_table *t,
 179                 const struct osl_row *row, char **name)
 180 {
 181         struct osl_object obj;
 182         int ret = osl_get_object(t, row, t->disk_storage_name_column, &obj);
 183
 184         if (ret < 0)
 185                 return ret;
 186         *name = disk_storage_name_of_object(t, &obj);
 187         if (*name)
 188                 return 1;
 189         return -ERRNO_TO_ERROR(ENOMEM);
 190 }
 191
 192 static void column_name_hash(const char *col_name, HASH_TYPE *hash)
 193 {
 194         hash_function(col_name, strlen(col_name), hash);
 195 }
 196
 197 static int init_column_descriptions(struct osl_table *t)
 198 {
 199         int i, j, ret;
 200         const struct osl_column_description *cd;
 201
 202         ret = -E_OSL_BAD_TABLE_DESC;
 203         ret = verify_name(t->desc->name);
 204         if (ret < 0)
 205                 goto err;
 206         ret = -E_OSL_BAD_DB_DIR;
 207         if (!t->desc->dir && (t->num_disk_storage_columns || t->num_mapped_columns))
 208                 goto err;
 209         /* the size of the index header without column descriptions */
 210         t->index_header_size = IDX_COLUMN_DESCRIPTIONS;
 211         FOR_EACH_COLUMN(i, t->desc, cd) {
 212                 struct osl_column *col = t->columns + i;
 213                 if (cd->storage_flags & OSL_RBTREE) {
 214                         if (!cd->compare_function)
 215                                 return -E_OSL_NO_COMPARE_FUNC;
 216                 }
 217                 if (cd->storage_type == OSL_NO_STORAGE)
 218                         continue;
 219                 ret = -E_OSL_NO_COLUMN_NAME;
 220                 if (!cd->name || !cd->name[0])
 221                         goto err;
 222                 ret = verify_name(cd->name);
 223                 if (ret < 0)
 224                         goto err;
 225                 t->index_header_size += index_column_description_size(cd->name);
 226                 column_name_hash(cd->name, col->name_hash);
 227                 ret = -E_OSL_DUPLICATE_COL_NAME;
 228                 for (j = i + 1; j < t->desc->num_columns; j++) {
 229                         const char *name2 = get_column_description(t->desc,
 230                                 j)->name;
 231                         if (cd->name && name2 && !strcmp(cd->name, name2))
 232                                 goto err;
 233                 }
 234         }
 235         return 1;
 236 err:
 237         return ret;
 238 }
 239
 240 /**
 241  * Initialize a struct table from given table description.
 242  *
 243  * \param desc The description of the osl table.
 244  * \param table_ptr Result is returned here.
 245  *
 246  * This function performs several sanity checks on \p desc and returns if any
 247  * of these tests fail. On success, a struct \p osl_table is allocated and
 248  * initialized with data derived from \p desc.
 249  *
 250  * \return Standard.
 251  *
 252  * \sa struct osl_table.
 253  */
 254 int init_table_structure(const struct osl_table_description *desc,
 255                 struct osl_table **table_ptr)
 256 {
 257         const struct osl_column_description *cd;
 258         struct osl_table *t = calloc(1, sizeof(*t));
 259         int i, ret = -ERRNO_TO_ERROR(ENOMEM), have_disk_storage_name_column = 0;
 260
 261         if (!t)
 262                 return ret;
 263         ret = -E_OSL_BAD_TABLE_DESC;
 264         if (!desc)
 265                 goto err;
 266         DEBUG_LOG("creating table structure for '%s' from table "
 267                 "description\n", desc->name);
 268         ret = -E_OSL_NO_COLUMN_DESC;
 269         if (!desc->column_descriptions)
 270                 goto err;
 271         ret = -E_OSL_NO_COLUMNS;
 272         if (!desc->num_columns)
 273                 goto err;
 274         ret = -ERRNO_TO_ERROR(ENOMEM);
 275         t->columns = calloc(desc->num_columns, sizeof(struct osl_column));
 276         if (!t->columns)
 277                 goto err;
 278         t->desc = desc;
 279         FOR_EACH_COLUMN(i, t->desc, cd) {
 280                 enum osl_storage_type st = cd->storage_type;
 281                 enum osl_storage_flags sf = cd->storage_flags;
 282                 struct osl_column *col = &t->columns[i];
 283
 284                 ret = -E_OSL_BAD_STORAGE_TYPE;
 285                 if (st != OSL_MAPPED_STORAGE && st != OSL_DISK_STORAGE
 286                                 && st != OSL_NO_STORAGE)
 287                         goto err;
 288                 ret = -E_OSL_BAD_STORAGE_FLAGS;
 289                 if (st == OSL_DISK_STORAGE && sf & OSL_RBTREE)
 290                         goto err;
 291                 ret = -E_OSL_BAD_STORAGE_SIZE;
 292                 if (sf & OSL_FIXED_SIZE && !cd->data_size)
 293                         goto err;
 294                 switch (st) {
 295                 case OSL_DISK_STORAGE:
 296                         t->num_disk_storage_columns++;
 297                         break;
 298                 case OSL_MAPPED_STORAGE:
 299                         t->num_mapped_columns++;
 300                         col->index_offset = t->row_index_size;
 301                         t->row_index_size += 8;
 302                         break;
 303                 case OSL_NO_STORAGE:
 304                         col->volatile_num = t->num_volatile_columns;
 305                         t->num_volatile_columns++;
 306                         break;
 307                 }
 308                 if (sf & OSL_RBTREE) {
 309                         col->rbtree_num = t->num_rbtrees;
 310                         t->num_rbtrees++;
 311                         if ((sf & OSL_UNIQUE) && (st == OSL_MAPPED_STORAGE)) {
 312                                 if (!have_disk_storage_name_column)
 313                                         t->disk_storage_name_column = i;
 314                                 have_disk_storage_name_column = 1;
 315                         }
 316                 }
 317         }
 318         ret = -E_OSL_NO_UNIQUE_RBTREE_COLUMN;
 319         if (t->num_disk_storage_columns && !have_disk_storage_name_column)
 320                 goto err;
 321         ret = -E_OSL_NO_RBTREE_COL;
 322         if (!t->num_rbtrees)
 323                 goto err;
 324         /* success */
 325         DEBUG_LOG("OK. Index entry size: %u\n", t->row_index_size);
 326         ret = init_column_descriptions(t);
 327         if (ret < 0)
 328                 goto err;
 329         *table_ptr = t;
 330         return 1;
 331 err:
 332         free(t->columns);
 333         free(t);
 334         return ret;
 335 }
 336
 337 /**
 338  * Read the table description from index header.
 339  *
 340  * \param map The memory mapping of the index file.
 341  * \param desc The values found in the index header are returned here.
 342  *
 343  * Read the index header, check for the paraslash magic string and the table version number.
 344  * Read all information stored in the index header into \a desc.
 345  *
 346  * \return Standard.
 347  *
 348  * \sa struct osl_table_description, osl_create_table.
 349  */
 350 int read_table_desc(struct osl_object *map, struct osl_table_description *desc)
 351 {
 352         char *buf = map->data;
 353         uint8_t version;
 354         uint16_t header_size;
 355         int ret, i;
 356         unsigned offset;
 357         struct osl_column_description *cd;
 358
 359         if (map->size < MIN_INDEX_HEADER_SIZE(1))
 360                 return -E_OSL_SHORT_TABLE;
 361         if (strncmp(buf + IDX_PARA_MAGIC, PARA_MAGIC, strlen(PARA_MAGIC)))
 362                 return -E_OSL_NO_MAGIC;
 363         version = read_u8(buf + IDX_VERSION);
 364         if (version < MIN_TABLE_VERSION || version > MAX_TABLE_VERSION)
 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 -ERRNO_TO_ERROR(ENOMEM);
 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 = -ERRNO_TO_ERROR(ENOMEM);
 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 = -ERRNO_TO_ERROR(ENOMEM);
 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 -ERRNO_TO_ERROR(ENOMEM);
 496         sprintf(buf + IDX_PARA_MAGIC, "%s", PARA_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 = -ERRNO_TO_ERROR(ENOMEM);
 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 && !is_errno(-ret, EEXIST))
 544                                 goto out;
 545                         table_dir = make_message("%s/%s", desc->dir,
 546                                 desc->name);
 547                         ret = -ERRNO_TO_ERROR(ENOMEM);
 548                         if (!table_dir)
 549                                 goto out;
 550                         ret = osl_mkdir(table_dir, 0777);
 551                         if (ret < 0)
 552                                 goto out;
 553                 }
 554                 ret = -ERRNO_TO_ERROR(ENOMEM);
 555                 filename = column_filename(t, i);
 556                 if (!filename)
 557                         goto out;
 558                 INFO_LOG("filename: %s\n", filename);
 559                 if (cd->storage_type == OSL_MAPPED_STORAGE) {
 560                         ret = osl_open(filename, O_RDWR | O_CREAT | O_EXCL,
 561                                 0644);
 562                         free(filename);
 563                         if (ret < 0)
 564                                 goto out;
 565                         close(ret);
 566                         continue;
 567                 }
 568                 /* DISK STORAGE */
 569                 ret = osl_mkdir(filename, 0777);
 570                 free(filename);
 571                 if (ret < 0)
 572                         goto out;
 573         }
 574         if (t->num_mapped_columns) {
 575                 ret = create_table_index(t);
 576                 if (ret < 0)
 577                         goto out;
 578         }
 579         ret = 1;
 580 out:
 581         free(table_dir);
 582         free(t->columns);
 583         free(t);
 584         return ret;
 585 }
 586
 587 static int table_is_dirty(struct osl_table *t)
 588 {
 589         char *buf = (char *)t->index_map.data + IDX_DIRTY_FLAG;
 590         uint8_t dirty = read_u8(buf) & 0x1;
 591         return !!dirty;
 592 }
 593
 594 static void mark_table_dirty(struct osl_table *t)
 595 {
 596         char *buf = (char *)t->index_map.data + IDX_DIRTY_FLAG;
 597         write_u8(buf, read_u8(buf) | 1);
 598 }
 599
 600 static void mark_table_clean(struct osl_table *t)
 601 {
 602         char *buf = (char *)t->index_map.data + IDX_DIRTY_FLAG;
 603         write_u8(buf, read_u8(buf) & 0xfe);
 604 }
 605
 606 static void unmap_column(struct osl_table *t, unsigned col_num)
 607 {
 608         struct osl_object map = t->columns[col_num].data_map;
 609         int ret;
 610         if (!map.data)
 611                 return;
 612         ret = osl_munmap(map.data, map.size);
 613         assert(ret > 0);
 614         map.data = NULL;
 615 }
 616
 617 /**
 618  * Unmap all mapped files of an osl table.
 619  *
 620  * \param t Pointer to a mapped table.
 621  * \param flags Options for unmapping.
 622  *
 623  * \return Positive on success, negative on errors.
 624  *
 625  * \sa map_table(), enum osl_close_flags, osl_munmap().
 626  */
 627 int unmap_table(struct osl_table *t, enum osl_close_flags flags)
 628 {
 629         unsigned i;
 630         const struct osl_column_description *cd;
 631         int ret;
 632
 633         if (!t->num_mapped_columns) /* can this ever happen? */
 634                 return 1;
 635         INFO_LOG("unmapping table '%s'\n", t->desc->name);
 636         if (!t->index_map.data)
 637                 return -E_OSL_NOT_MAPPED;
 638         if (flags & OSL_MARK_CLEAN)
 639                 mark_table_clean(t);
 640         ret = osl_munmap(t->index_map.data, t->index_map.size);
 641         if (ret < 0)
 642                 return ret;
 643         t->index_map.data = NULL;
 644         if (!t->num_rows)
 645                 return 1;
 646         FOR_EACH_MAPPED_COLUMN(i, t, cd)
 647                 unmap_column(t, i);
 648         return 1;
 649 }
 650
 651 static int map_column(struct osl_table *t, unsigned col_num)
 652 {
 653         struct stat statbuf;
 654         char *filename = column_filename(t, col_num);
 655         int ret;
 656
 657         if (!filename)
 658                 return -ERRNO_TO_ERROR(ENOMEM);
 659         ret = osl_stat(filename, &statbuf);
 660         if (ret < 0) {
 661                 free(filename);
 662                 return ret;
 663         }
 664         if (!(S_IFREG & statbuf.st_mode)) {
 665                 free(filename);
 666                 return ret;
 667         }
 668         ret = mmap_full_file(filename, O_RDWR,
 669                 &t->columns[col_num].data_map.data,
 670                 &t->columns[col_num].data_map.size,
 671                 NULL);
 672         free(filename);
 673         return ret;
 674 }
 675
 676 /**
 677  * Map the index file and all columns of type \p OSL_MAPPED_STORAGE into memory.
 678  *
 679  * \param t Pointer to an initialized table structure.
 680  * \param flags Mapping options.
 681  *
 682  * \return Negative return value on errors; on success the number of rows
 683  * (including invalid rows) is returned.
 684  *
 685  * \sa unmap_table(), enum map_table_flags, osl_open_table(), mmap(2).
 686  */
 687 int map_table(struct osl_table *t, enum map_table_flags flags)
 688 {
 689         char *filename;
 690         const struct osl_column_description *cd;
 691         int i = 0, ret, num_rows = 0;
 692
 693         if (!t->num_mapped_columns)
 694                 return 0;
 695         if (t->index_map.data)
 696                 return -E_OSL_ALREADY_MAPPED;
 697         filename = index_filename(t->desc);
 698         if (!filename)
 699                 return -ERRNO_TO_ERROR(ENOMEM);
 700         INFO_LOG("mapping table '%s' (index: %s)\n", t->desc->name, filename);
 701         ret = mmap_full_file(filename, flags & MAP_TBL_FL_MAP_RDONLY?
 702                 O_RDONLY : O_RDWR, &t->index_map.data, &t->index_map.size, NULL);
 703         free(filename);
 704         if (ret < 0)
 705                 return ret;
 706         if (flags & MAP_TBL_FL_VERIFY_INDEX) {
 707                 ret = compare_table_descriptions(t);
 708                 if (ret < 0)
 709                         goto err;
 710         }
 711         ret = -E_OSL_BUSY;
 712         if (!(flags & MAP_TBL_FL_IGNORE_DIRTY)) {
 713                 if (table_is_dirty(t)) {
 714                         ERROR_LOG("%s is dirty\n", t->desc->name);
 715                         goto err;
 716                 }
 717         }
 718         mark_table_dirty(t);
 719         num_rows = table_num_rows(t);
 720         if (!num_rows)
 721                 return num_rows;
 722         /* map data files */
 723         FOR_EACH_MAPPED_COLUMN(i, t, cd) {
 724                 ret = map_column(t, i);
 725                 if (ret < 0)
 726                         goto err;
 727         }
 728         return num_rows;
 729 err:    /* unmap what is already mapped */
 730         for (i--; i >= 0; i--) {
 731                 struct osl_object map = t->columns[i].data_map;
 732                 osl_munmap(map.data, map.size);
 733                 map.data = NULL;
 734         }
 735         osl_munmap(t->index_map.data, t->index_map.size);
 736         t->index_map.data = NULL;
 737         return ret;
 738 }
 739
 740 /**
 741  * Retrieve a mapped object by row and column number.
 742  *
 743  * \param t Pointer to an open osl table.
 744  * \param col_num Number of the mapped column containing the object to retrieve.
 745  * \param row_num Number of the row containing the object to retrieve.
 746  * \param obj The result is returned here.
 747  *
 748  * It is considered an error if \a col_num does not refer to a column
 749  * of storage type \p OSL_MAPPED_STORAGE.
 750  *
 751  * \return Standard.
 752  *
 753  * \sa osl_storage_type.
 754  */
 755 int get_mapped_object(const struct osl_table *t, unsigned col_num,
 756         uint32_t row_num, struct osl_object *obj)
 757 {
 758         struct osl_column *col = &t->columns[col_num];
 759         uint32_t offset;
 760         char *cell_index;
 761         int ret;
 762
 763         if (t->num_rows <= row_num)
 764                 return -E_OSL_BAD_ROW_NUM;
 765         ret = get_cell_index(t, row_num, col_num, &cell_index);
 766         if (ret < 0)
 767                 return ret;
 768         offset = read_u32(cell_index);
 769         obj->size = read_u32(cell_index + 4);
 770         obj->data = col->data_map.data + offset;
 771         return 1;
 772 }
 773
 774 static int search_rbtree(const struct osl_object *obj,
 775                 const struct osl_table *t, unsigned col_num,
 776                 struct rb_node **result, struct rb_node ***rb_link)
 777 {
 778         struct osl_column *col = &t->columns[col_num];
 779         struct rb_node **new = &col->rbtree.rb_node, *parent = NULL;
 780         const struct osl_column_description *cd =
 781                 get_column_description(t->desc, col_num);
 782         enum osl_storage_type st = cd->storage_type;
 783         while (*new) {
 784                 struct osl_row *this_row = get_row_pointer(*new,
 785                         col->rbtree_num);
 786                 int ret;
 787                 struct osl_object this_obj;
 788                 parent = *new;
 789                 if (st == OSL_MAPPED_STORAGE) {
 790                         ret = get_mapped_object(t, col_num, this_row->num,
 791                                 &this_obj);
 792                         if (ret < 0)
 793                                 return ret;
 794                 } else
 795                         this_obj = this_row->volatile_objects[col->volatile_num];
 796                 ret = cd->compare_function(obj, &this_obj);
 797                 if (!ret) {
 798                         if (result)
 799                                 *result = get_rb_node_pointer(this_row,
 800                                         col->rbtree_num);
 801                         return 1;
 802                 }
 803                 if (ret < 0)
 804                         new = &((*new)->rb_left);
 805                 else
 806                         new = &((*new)->rb_right);
 807         }
 808         if (result)
 809                 *result = parent;
 810         if (rb_link)
 811                 *rb_link = new;
 812         return -E_OSL_RB_KEY_NOT_FOUND;
 813 }
 814
 815 static int insert_rbtree(struct osl_table *t, unsigned col_num,
 816         const struct osl_row *row, const struct osl_object *obj)
 817 {
 818         struct rb_node *parent, **rb_link;
 819         unsigned rbtree_num;
 820         struct rb_node *n;
 821         int ret = search_rbtree(obj, t, col_num, &parent, &rb_link);
 822
 823         if (ret > 0)
 824                 return -E_OSL_RB_KEY_EXISTS;
 825         rbtree_num = t->columns[col_num].rbtree_num;
 826         n = get_rb_node_pointer(row, rbtree_num);
 827         rb_link_node(n, parent, rb_link);
 828         rb_insert_color(n, &t->columns[col_num].rbtree);
 829         return 1;
 830 }
 831
 832 static void remove_rb_node(struct osl_table *t, unsigned col_num,
 833                 const struct osl_row *row)
 834 {
 835         struct osl_column *col = &t->columns[col_num];
 836         const struct osl_column_description *cd =
 837                 get_column_description(t->desc, col_num);
 838         enum osl_storage_flags sf = cd->storage_flags;
 839         struct rb_node *victim, *splice_out_node, *tmp;
 840         if (!(sf & OSL_RBTREE))
 841                 return;
 842         /*
 843          * Which node is removed/spliced out actually depends on how many
 844          * children the victim node has: If it has no children, it gets
 845          * deleted. If it has one child, it gets spliced out. If it has two
 846          * children, its successor (which has at most a right child) gets
 847          * spliced out.
 848          */
 849         victim = get_rb_node_pointer(row, col->rbtree_num);
 850         if (victim->rb_left && victim->rb_right)
 851                 splice_out_node = rb_next(victim);
 852         else
 853                 splice_out_node = victim;
 854         /* Go up to the root and decrement the size of each node in the path. */
 855         for (tmp = splice_out_node; tmp; tmp = rb_parent(tmp))
 856                 tmp->size--;
 857         rb_erase(victim, &col->rbtree);
 858 }
 859
 860 static int add_row_to_rbtrees(struct osl_table *t, uint32_t row_num,
 861                 struct osl_object *volatile_objs, struct osl_row **row_ptr)
 862 {
 863         unsigned i;
 864         int ret;
 865         struct osl_row *row = allocate_row(t->num_rbtrees);
 866         const struct osl_column_description *cd;
 867
 868         if (!row)
 869                 return -ERRNO_TO_ERROR(ENOMEM);
 870         row->num = row_num;
 871         row->volatile_objects = volatile_objs;
 872         FOR_EACH_RBTREE_COLUMN(i, t, cd) {
 873                 if (cd->storage_type == OSL_MAPPED_STORAGE) {
 874                         struct osl_object obj;
 875                         ret = get_mapped_object(t, i, row_num, &obj);
 876                         if (ret < 0)
 877                                 goto err;
 878                         ret = insert_rbtree(t, i, row, &obj);
 879                 } else { /* volatile */
 880                         const struct osl_object *obj
 881                                 = volatile_objs + t->columns[i].volatile_num;
 882                         ret = insert_rbtree(t, i, row, obj);
 883                 }
 884                 if (ret < 0)
 885                         goto err;
 886         }
 887         if (row_ptr)
 888                 *row_ptr = row;
 889         return 1;
 890 err: /* rollback changes, i.e. remove added entries from rbtrees */
 891         while (i)
 892                 remove_rb_node(t, i--, row);
 893         free(row);
 894         return ret;
 895 }
 896
 897 static void free_volatile_objects(const struct osl_table *t,
 898                 enum osl_close_flags flags)
 899 {
 900         int i, j;
 901         struct rb_node *n;
 902         struct osl_column *rb_col;
 903         const struct osl_column_description *cd;
 904
 905         if (!t->num_volatile_columns)
 906                 return;
 907         /* find the first rbtree column (any will do) */
 908         FOR_EACH_RBTREE_COLUMN(i, t, cd)
 909                 break;
 910         rb_col = t->columns + i;
 911         /* walk that rbtree and free all volatile objects */
 912         for (n = rb_first(&rb_col->rbtree); n; n = rb_next(n)) {
 913                 struct osl_row *r = get_row_pointer(n, rb_col->rbtree_num);
 914                 if (flags & OSL_FREE_VOLATILE)
 915                         FOR_EACH_VOLATILE_COLUMN(j, t, cd) {
 916                                 if (cd->storage_flags & OSL_DONT_FREE)
 917                                         continue;
 918                                 free(r->volatile_objects[
 919                                         t->columns[j].volatile_num].data);
 920                         }
 921 //                      for (j = 0; j < t->num_volatile_columns; j++)
 922 //                              free(r->volatile_objects[j].data);
 923                 free(r->volatile_objects);
 924         }
 925 }
 926
 927 /**
 928  * Erase all rbtree nodes and free resources.
 929  *
 930  * \param t Pointer to an open osl table.
 931  *
 932  * This function is called by osl_close_table().
 933  */
 934 void clear_rbtrees(struct osl_table *t)
 935 {
 936         const struct osl_column_description *cd;
 937         unsigned i, rbtrees_cleared = 0;
 938
 939         FOR_EACH_RBTREE_COLUMN(i, t, cd) {
 940                 struct osl_column *col = &t->columns[i];
 941                 struct rb_node *n;
 942                 rbtrees_cleared++;
 943                 for (n = rb_first(&col->rbtree); n;) {
 944                         struct osl_row *r;
 945                         rb_erase(n, &col->rbtree);
 946                         if (rbtrees_cleared == t->num_rbtrees) {
 947                                 r = get_row_pointer(n, col->rbtree_num);
 948                                 n = rb_next(n);
 949                                 free(r);
 950                         } else
 951                                 n = rb_next(n);
 952                 }
 953         }
 954
 955 }
 956
 957 __export int osl_close_table(struct osl_table *t, enum osl_close_flags flags)
 958 {
 959         int ret;
 960
 961         if (!t)
 962                 return -E_OSL_BAD_TABLE;
 963         NOTICE_LOG("closing table %s\n", t->desc->name);
 964         free_volatile_objects(t, flags);
 965         clear_rbtrees(t);
 966         ret = unmap_table(t, flags);
 967         if (ret < 0)
 968                 ERROR_LOG("unmap_table failed: %d\n", ret);
 969         free(t->columns);
 970         free(t);
 971         return ret;
 972 }
 973
 974 /**
 975  * Find out whether the given row number corresponds to an invalid row.
 976  *
 977  * \param t Pointer to the osl table.
 978  * \param row_num The number of the row in question.
 979  *
 980  * By definition, a row is considered invalid if all its index entries
 981  * are invalid.
 982  *
 983  * \return Positive if \a row_num corresponds to an invalid row,
 984  * zero if it corresponds to a valid row, negative on errors.
 985  */
 986 int row_is_invalid(struct osl_table *t, uint32_t row_num)
 987 {
 988         char *row_index;
 989         int i, ret = get_row_index(t, row_num, &row_index);
 990
 991         if (ret < 0)
 992                 return ret;
 993         for (i = 0; i < t->row_index_size; i++) {
 994                 if ((unsigned char)row_index[i] != 0xff)
 995                         return 0;
 996         }
 997         INFO_LOG("row %d is invalid\n", row_num);
 998         return 1;
 999 }
1000
1001 /**
1002  * Invalidate a row of an osl table.
1003  *
1004  * \param t Pointer to an open osl table.
1005  * \param row_num Number of the row to mark as invalid.
1006  *
1007  * This function marks each mapped object in the index entry of \a row as
1008  * invalid.
1009  *
1010  * \return Standard.
1011  */
1012 int mark_row_invalid(struct osl_table *t, uint32_t row_num)
1013 {
1014         char *row_index;
1015         int ret = get_row_index(t, row_num, &row_index);
1016
1017         if (ret < 0)
1018                 return ret;
1019         INFO_LOG("marking row %d as invalid\n", row_num);
1020         memset(row_index, 0xff, t->row_index_size);
1021         return 1;
1022 }
1023
1024 /**
1025  * Initialize all rbtrees and compute number of invalid rows.
1026  *
1027  * \param t The table containing the rbtrees to be initialized.
1028  *
1029  * \return Standard.
1030  */
1031 int init_rbtrees(struct osl_table *t)
1032 {
1033         int i, ret;
1034         const struct osl_column_description *cd;
1035
1036         /* create rbtrees */
1037         FOR_EACH_RBTREE_COLUMN(i, t, cd)
1038                 t->columns[i].rbtree = RB_ROOT;
1039         /* add valid rows to rbtrees */
1040         t->num_invalid_rows = 0;
1041         for (i = 0; i < t->num_rows; i++) {
1042                 ret = row_is_invalid(t, i);
1043                 if (ret < 0)
1044                         return ret;
1045                 if (ret) {
1046                         t->num_invalid_rows++;
1047                         continue;
1048                 }
1049                 ret = add_row_to_rbtrees(t, i, NULL, NULL);
1050                 if (ret < 0)
1051                         return ret;
1052         }
1053         return 1;
1054 }
1055
1056 __export int osl_open_table(const struct osl_table_description *table_desc,
1057                 struct osl_table **result)
1058 {
1059         int i, ret;
1060         struct osl_table *t;
1061         const struct osl_column_description *cd;
1062
1063         NOTICE_LOG("opening table %s\n", table_desc->name);
1064         ret = init_table_structure(table_desc, &t);
1065         if (ret < 0)
1066                 return ret;
1067         FOR_EACH_DISK_STORAGE_COLUMN(i, t, cd) {
1068                 struct stat statbuf;
1069                 char *dirname = column_filename(t, i);
1070
1071                 ret = -ERRNO_TO_ERROR(ENOMEM);
1072                 if (!dirname)
1073                         goto err;
1074                 /* check if directory exists */
1075                 ret = stat(dirname, &statbuf);
1076                 free(dirname);
1077                 if (ret < 0) {
1078                         ret = -ERRNO_TO_ERROR(errno);
1079                         goto err;
1080                 }
1081                 ret = -ERRNO_TO_ERROR(ENOTDIR);
1082                 if (!S_ISDIR(statbuf.st_mode))
1083                         goto err;
1084         }
1085         ret = map_table(t, MAP_TBL_FL_VERIFY_INDEX);
1086         if (ret < 0)
1087                 goto err;
1088         t->num_rows = ret;
1089         DEBUG_LOG("num rows: %d\n", t->num_rows);
1090         ret = init_rbtrees(t);
1091         if (ret < 0) {
1092                 osl_close_table(t, OSL_MARK_CLEAN); /* ignore further errors */
1093                 return ret;
1094         }
1095         *result = t;
1096         return 1;
1097 err:
1098         free(t->columns);
1099         free(t);
1100         return ret;
1101 }
1102
1103 static int create_disk_storage_object_dir(const struct osl_table *t,
1104                 unsigned col_num, const char *ds_name)
1105 {
1106         char *dirname;
1107         int ret;
1108
1109         if (!(t->desc->flags & OSL_LARGE_TABLE))
1110                 return 1;
1111         dirname = disk_storage_dirname(t, col_num, ds_name);
1112         if (!dirname)
1113                 return -ERRNO_TO_ERROR(ENOMEM);
1114         ret = osl_mkdir(dirname, 0777);
1115         free(dirname);
1116         if (ret < 0 && !is_errno(-ret, EEXIST))
1117                 return ret;
1118         return 1;
1119 }
1120
1121 static int write_disk_storage_file(const struct osl_table *t, unsigned col_num,
1122         const struct osl_object *obj, const char *ds_name)
1123 {
1124         int ret;
1125         char *filename;
1126
1127         ret = create_disk_storage_object_dir(t, col_num, ds_name);
1128         if (ret < 0)
1129                 return ret;
1130         filename = disk_storage_path(t, col_num, ds_name);
1131         if (!filename)
1132                 return -ERRNO_TO_ERROR(ENOMEM);
1133         ret = write_file(filename, obj->data, obj->size);
1134         free(filename);
1135         return ret;
1136 }
1137
1138 static int append_map_file(const struct osl_table *t, unsigned col_num,
1139         const struct osl_object *obj, uint32_t *new_size)
1140 {
1141         char *filename = column_filename(t, col_num);
1142         int ret;
1143
1144         if (!filename)
1145                 return -ERRNO_TO_ERROR(ENOMEM);
1146         ret = append_file(filename, obj->data, obj->size, new_size);
1147         free(filename);
1148         return ret;
1149 }
1150
1151 static int append_row_index(const struct osl_table *t, char *row_index)
1152 {
1153         char *filename;
1154         int ret;
1155
1156         if (!t->num_mapped_columns)
1157                 return 1;
1158         filename = index_filename(t->desc);
1159         if (!filename)
1160                 return -ERRNO_TO_ERROR(ENOMEM);
1161         ret = append_file(filename, row_index, t->row_index_size, NULL);
1162         free(filename);
1163         return ret;
1164 }
1165
1166 static int truncate_mapped_file(const struct osl_table *t, unsigned col_num,
1167                 off_t size)
1168 {
1169         int ret;
1170         char *filename = column_filename(t, col_num);
1171
1172         if (!filename)
1173                 return -ERRNO_TO_ERROR(ENOMEM);
1174         ret = truncate_file(filename, size);
1175         free(filename);
1176         return ret;
1177 }
1178
1179 static int delete_disk_storage_file(const struct osl_table *t, unsigned col_num,
1180                 const char *ds_name)
1181 {
1182         char *dirname, *filename = disk_storage_path(t, col_num, ds_name);
1183         int ret, err;
1184
1185         if (!filename)
1186                 return -ERRNO_TO_ERROR(ENOMEM);
1187         ret = unlink(filename);
1188         err = errno;
1189         free(filename);
1190         if (ret < 0)
1191                 return -ERRNO_TO_ERROR(err);
1192         if (!(t->desc->flags & OSL_LARGE_TABLE))
1193                 return 1;
1194         dirname = disk_storage_dirname(t, col_num, ds_name);
1195         if (!dirname)
1196                 return -ERRNO_TO_ERROR(ENOMEM);
1197         rmdir(dirname);
1198         free(dirname);
1199         return 1;
1200 }
1201
1202 __export int osl_add_and_get_row(struct osl_table *t, struct osl_object *objects,
1203                 struct osl_row **row)
1204 {
1205         int i, ret;
1206         char *ds_name = NULL;
1207         struct rb_node **rb_parents = NULL, ***rb_links = NULL;
1208         char *new_row_index = NULL;
1209         struct osl_object *volatile_objs = NULL;
1210         const struct osl_column_description *cd;
1211
1212         if (!t)
1213                 return -E_OSL_BAD_TABLE;
1214         rb_parents = malloc(t->num_rbtrees * sizeof(struct rn_node*));
1215         if (!rb_parents)
1216                 return -ERRNO_TO_ERROR(ENOMEM);
1217         rb_links = malloc(t->num_rbtrees * sizeof(struct rn_node**));
1218         if (!rb_links) {
1219                 free(rb_parents);
1220                 return -ERRNO_TO_ERROR(ENOMEM);
1221         }
1222         if (t->num_mapped_columns) {
1223                 new_row_index = malloc(t->row_index_size);
1224                 if (!new_row_index) {
1225                         free(rb_links);
1226                         free(rb_parents);
1227                         return -ERRNO_TO_ERROR(ENOMEM);
1228                 }
1229         }
1230         /* pass 1: sanity checks */
1231 //      DEBUG_LOG("sanity tests: %p:%p\n", objects[0].data,
1232 //              objects[1].data);
1233         FOR_EACH_COLUMN(i, t->desc, cd) {
1234                 enum osl_storage_type st = cd->storage_type;
1235                 enum osl_storage_flags sf = cd->storage_flags;
1236
1237 //              ret = -E_OSL_NULL_OBJECT;
1238 //              if (!objects[i])
1239 //                      goto out;
1240                 if (st == OSL_DISK_STORAGE)
1241                         continue;
1242                 if (sf & OSL_RBTREE) {
1243                         unsigned rbtree_num = t->columns[i].rbtree_num;
1244                         ret = -E_OSL_RB_KEY_EXISTS;
1245 //                      DEBUG_LOG("checking whether %p exists\n",
1246 //                              objects[i].data);
1247                         if (search_rbtree(objects + i, t, i,
1248                                         &rb_parents[rbtree_num],
1249                                         &rb_links[rbtree_num]) > 0)
1250                                 goto out;
1251                 }
1252                 if (sf & OSL_FIXED_SIZE) {
1253 //                      DEBUG_LOG("fixed size. need: %zu, have: %d\n",
1254 //                              objects[i].size, cd->data_size);
1255                         ret = -E_OSL_BAD_DATA_SIZE;
1256                         if (objects[i].size != cd->data_size)
1257                                 goto out;
1258                 }
1259         }
1260         if (t->num_disk_storage_columns) {
1261                 ds_name = disk_storage_name_of_object(t,
1262                         &objects[t->disk_storage_name_column]);
1263                 ret = -ERRNO_TO_ERROR(ENOMEM);
1264                 if (!ds_name)
1265                         goto out;
1266         }
1267         ret = unmap_table(t, OSL_MARK_CLEAN);
1268         if (ret < 0)
1269                 goto out;
1270 //      DEBUG_LOG("sanity tests passed%s\n", "");
1271         /* pass 2: create data files, append map data */
1272         FOR_EACH_COLUMN(i, t->desc, cd) {
1273                 enum osl_storage_type st = cd->storage_type;
1274                 if (st == OSL_NO_STORAGE)
1275                         continue;
1276                 if (st == OSL_MAPPED_STORAGE) {
1277                         uint32_t new_size;
1278                         struct osl_column *col = &t->columns[i];
1279 //                      DEBUG_LOG("appending object of size %zu\n",
1280 //                              objects[i].size);
1281                         ret = append_map_file(t, i, objects + i, &new_size);
1282                         if (ret < 0)
1283                                 goto rollback;
1284                         update_cell_index(new_row_index, col, new_size,
1285                                 objects[i].size);
1286                         continue;
1287                 }
1288                 /* DISK_STORAGE */
1289                 ret = write_disk_storage_file(t, i, objects + i, ds_name);
1290                 if (ret < 0)
1291                         goto rollback;
1292         }
1293         ret = append_row_index(t, new_row_index);
1294         if (ret < 0)
1295                 goto rollback;
1296         ret = map_table(t, MAP_TBL_FL_VERIFY_INDEX);
1297         if (ret < 0) { /* truncate index and rollback changes */
1298                 char *filename = index_filename(t->desc);
1299                 if (filename)
1300                         truncate_file(filename, t->row_index_size);
1301                 free(filename);
1302                 goto rollback;
1303         }
1304         /* pass 3: add entry to rbtrees */
1305         if (t->num_volatile_columns) {
1306                 ret = -ERRNO_TO_ERROR(ENOMEM);
1307                 volatile_objs = calloc(t->num_volatile_columns,
1308                         sizeof(struct osl_object));
1309                 if (!volatile_objs)
1310                         goto out;
1311                 FOR_EACH_VOLATILE_COLUMN(i, t, cd)
1312                         volatile_objs[t->columns[i].volatile_num] = objects[i];
1313         }
1314         t->num_rows++;
1315 //      DEBUG_LOG("adding new entry as row #%d\n", t->num_rows - 1);
1316         ret = add_row_to_rbtrees(t, t->num_rows - 1, volatile_objs, row);
1317         if (ret < 0)
1318                 goto out;
1319 //      DEBUG_LOG("added new entry as row #%d\n", t->num_rows - 1);
1320         ret = 1;
1321         goto out;
1322 rollback: /* rollback all changes made, ignore further errors */
1323         for (i--; i >= 0; i--) {
1324                 cd = get_column_description(t->desc, i);
1325                 enum osl_storage_type st = cd->storage_type;
1326                 if (st == OSL_NO_STORAGE)
1327                         continue;
1328
1329                 if (st == OSL_MAPPED_STORAGE)
1330                         truncate_mapped_file(t, i, objects[i].size);
1331                 else /* disk storage */
1332                         delete_disk_storage_file(t, i, ds_name);
1333         }
1334         /* ignore error and return previous error value */
1335         map_table(t, MAP_TBL_FL_VERIFY_INDEX);
1336 out:
1337         free(new_row_index);
1338         free(ds_name);
1339         free(rb_parents);
1340         free(rb_links);
1341         return ret;
1342 }
1343
1344 __export int osl_add_row(struct osl_table *t, struct osl_object *objects)
1345 {
1346         return osl_add_and_get_row(t, objects, NULL);
1347 }
1348
1349 __export int osl_get_object(const struct osl_table *t, const struct osl_row *r,
1350         unsigned col_num, struct osl_object *object)
1351 {
1352         const struct osl_column_description *cd;
1353
1354         if (!t)
1355                 return -E_OSL_BAD_TABLE;
1356         cd = get_column_description(t->desc, col_num);
1357         /* col must not be disk storage */
1358         if (cd->storage_type == OSL_DISK_STORAGE)
1359                 return -E_OSL_BAD_STORAGE_TYPE;
1360         if (cd->storage_type == OSL_MAPPED_STORAGE)
1361                 return get_mapped_object(t, col_num, r->num, object);
1362         /* volatile */
1363         *object = r->volatile_objects[t->columns[col_num].volatile_num];
1364         return 1;
1365 }
1366
1367 __export int osl_del_row(struct osl_table *t, struct osl_row *row)
1368 {
1369         struct osl_row *r = row;
1370         int i, ret;
1371         const struct osl_column_description *cd;
1372
1373         if (!t)
1374                 return -E_OSL_BAD_TABLE;
1375         INFO_LOG("deleting row %p\n", row);
1376
1377         if (t->num_disk_storage_columns) {
1378                 char *ds_name;
1379                 ret = disk_storage_name_of_row(t, r, &ds_name);
1380                 if (ret < 0)
1381                         goto out;
1382                 FOR_EACH_DISK_STORAGE_COLUMN(i, t, cd)
1383                         delete_disk_storage_file(t, i, ds_name);
1384                 free(ds_name);
1385         }
1386         FOR_EACH_COLUMN(i, t->desc, cd) {
1387                 struct osl_column *col = t->columns + i;
1388                 enum osl_storage_type st = cd->storage_type;
1389                 remove_rb_node(t, i, r);
1390                 if (st == OSL_MAPPED_STORAGE)
1391                         continue;
1392                 if (st == OSL_NO_STORAGE && !(cd->storage_flags & OSL_DONT_FREE))
1393                         free(r->volatile_objects[col->volatile_num].data);
1394         }
1395         if (t->num_mapped_columns) {
1396                 ret = mark_row_invalid(t, r->num);
1397                 if (ret < 0)
1398                         goto out;
1399                 t->num_invalid_rows++;
1400         } else
1401                 t->num_rows--;
1402         ret = 1;
1403 out:
1404         free(r->volatile_objects);
1405         free(r);
1406         return ret;
1407 }
1408
1409 /* test if column has an rbtree */
1410 static int check_rbtree_col(const struct osl_table *t, unsigned col_num,
1411                 struct osl_column **col)
1412 {
1413         if (!t)
1414                 return -E_OSL_BAD_TABLE;
1415         if (!(get_column_description(t->desc, col_num)->storage_flags & OSL_RBTREE))
1416                 return -E_OSL_BAD_STORAGE_FLAGS;
1417         *col = t->columns + col_num;
1418         return 1;
1419 }
1420
1421 __export int osl_get_row(const struct osl_table *t, unsigned col_num,
1422                 const struct osl_object *obj, struct osl_row **result)
1423 {
1424         int ret;
1425         struct rb_node *node;
1426         struct osl_row *row;
1427         struct osl_column *col;
1428
1429         *result = NULL;
1430         ret = check_rbtree_col(t, col_num, &col);
1431         if (ret < 0)
1432                 return ret;
1433         ret = search_rbtree(obj, t, col_num, &node, NULL);
1434         if (ret < 0)
1435                 return ret;
1436         row = get_row_pointer(node, t->columns[col_num].rbtree_num);
1437         *result = row;
1438         return 1;
1439 }
1440
1441 static int rbtree_loop(struct osl_column *col, void *private_data,
1442                 osl_rbtree_loop_func *func)
1443 {
1444         struct rb_node *n, *tmp;
1445
1446         /* this for-loop is safe against removal of an entry */
1447         for (n = rb_first(&col->rbtree), tmp = n? rb_next(n) : NULL;
1448                         n;
1449                         n = tmp, tmp = tmp? rb_next(tmp) : NULL) {
1450                 struct osl_row *r = get_row_pointer(n, col->rbtree_num);
1451                 if (func(r, private_data) < 0)
1452                         return -E_OSL_LOOP;
1453         }
1454         return 1;
1455 }
1456
1457 static int rbtree_loop_reverse(struct osl_column *col, void *private_data,
1458                 osl_rbtree_loop_func *func)
1459 {
1460         struct rb_node *n, *tmp;
1461
1462         /* safe against removal of an entry */
1463         for (n = rb_last(&col->rbtree), tmp = n? rb_prev(n) : NULL;
1464                         n;
1465                         n = tmp, tmp = tmp? rb_prev(tmp) : NULL) {
1466                 struct osl_row *r = get_row_pointer(n, col->rbtree_num);
1467                 if (func(r, private_data) < 0)
1468                         return -E_OSL_LOOP;
1469         }
1470         return 1;
1471 }
1472
1473 __export int osl_rbtree_loop(const struct osl_table *t, unsigned col_num,
1474         void *private_data, osl_rbtree_loop_func *func)
1475 {
1476         struct osl_column *col;
1477
1478         int ret = check_rbtree_col(t, col_num, &col);
1479         if (ret < 0)
1480                 return ret;
1481         return rbtree_loop(col, private_data, func);
1482 }
1483
1484 __export int osl_rbtree_loop_reverse(const struct osl_table *t, unsigned col_num,
1485         void *private_data, osl_rbtree_loop_func *func)
1486 {
1487         struct osl_column *col;
1488
1489         int ret = check_rbtree_col(t, col_num, &col);
1490         if (ret < 0)
1491                 return ret;
1492         return rbtree_loop_reverse(col, private_data, func);
1493 }
1494
1495 /* TODO: Rollback changes on errors */
1496 static int rename_disk_storage_objects(struct osl_table *t,
1497                 struct osl_object *old_obj, struct osl_object *new_obj)
1498 {
1499         int i, ret;
1500         const struct osl_column_description *cd;
1501         char *old_ds_name, *new_ds_name;
1502
1503         if (!t->num_disk_storage_columns)
1504                 return 1; /* nothing to do */
1505         if (old_obj->size == new_obj->size && !memcmp(new_obj->data,
1506                         old_obj->data, new_obj->size))
1507                 return 1; /* object did not change */
1508         old_ds_name = disk_storage_name_of_object(t, old_obj);
1509         new_ds_name = disk_storage_name_of_object(t, new_obj);
1510         ret = -ERRNO_TO_ERROR(ENOMEM);
1511         if (!old_ds_name || ! new_ds_name)
1512                 goto out;
1513
1514         FOR_EACH_DISK_STORAGE_COLUMN(i, t, cd) {
1515                 char *old_filename, *new_filename;
1516                 ret = create_disk_storage_object_dir(t, i, new_ds_name);
1517                 if (ret < 0)
1518                         goto out;
1519                 old_filename = disk_storage_path(t, i, old_ds_name);
1520                 new_filename = disk_storage_path(t, i, new_ds_name);
1521                 if (!old_filename || !new_filename)
1522                         ret = -ERRNO_TO_ERROR(ENOMEM);
1523                 else
1524                         ret = osl_rename(old_filename, new_filename);
1525                 free(old_filename);
1526                 free(new_filename);
1527                 if (ret < 0)
1528                         goto out;
1529         }
1530         ret = 1;
1531 out:
1532         free(old_ds_name);
1533         free(new_ds_name);
1534         return ret;
1535
1536 }
1537
1538 __export int osl_update_object(struct osl_table *t, const struct osl_row *r,
1539                 unsigned col_num, struct osl_object *obj)
1540 {
1541         struct osl_column *col;
1542         const struct osl_column_description *cd;
1543         int ret;
1544
1545         if (!t)
1546                 return -E_OSL_BAD_TABLE;
1547         col = &t->columns[col_num];
1548         cd = get_column_description(t->desc, col_num);
1549         DEBUG_LOG("updating column %u of %s\n", col_num, t->desc->name);
1550         if (cd->storage_flags & OSL_RBTREE) {
1551                 if (search_rbtree(obj, t, col_num, NULL, NULL) > 0)
1552                         return -E_OSL_RB_KEY_EXISTS;
1553         }
1554         if (cd->storage_flags & OSL_FIXED_SIZE) {
1555                 if (obj->size != cd->data_size)
1556                         return -E_OSL_BAD_DATA_SIZE;
1557         }
1558         remove_rb_node(t, col_num, r);
1559         if (cd->storage_type == OSL_NO_STORAGE) { /* TODO: If fixed size, reuse object? */
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 && !is_errno(-ret, ENOENT)) {
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 -ERRNO_TO_ERROR(ENOMEM);
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         if (n == 0)
1690                 return -E_OSL_RB_KEY_NOT_FOUND;
1691         ret = osl_get_num_rows(t, &num_rows);
1692         if (ret < 0)
1693                 return ret;
1694         if (n > num_rows)
1695                 return -E_OSL_RB_KEY_NOT_FOUND;
1696         ret = check_rbtree_col(t, col_num, &col);
1697         if (ret < 0)
1698                 return ret;
1699         node = rb_nth(col->rbtree.rb_node, n);
1700         if (!node)
1701                 return -E_OSL_RB_KEY_NOT_FOUND;
1702         *result = get_row_pointer(node, col->rbtree_num);
1703         return 1;
1704 }
1705
1706 __export int osl_rbtree_first_row(const struct osl_table *t, unsigned col_num,
1707                 struct osl_row **result)
1708 {
1709         return osl_get_nth_row(t, col_num, 1, result);
1710 }
1711
1712 __export int osl_rbtree_last_row(const struct osl_table *t, unsigned col_num,
1713                 struct osl_row **result)
1714 {
1715         unsigned num_rows;
1716         int ret = osl_get_num_rows(t, &num_rows);
1717
1718         if (ret < 0)
1719                 return ret;
1720         return osl_get_nth_row(t, col_num, num_rows, result);
1721 }