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 static char *disk_storage_dirname(const struct osl_table *t, unsigned col_num,
 152                 const char *ds_name)
 153 {
 154         char *dirname, *column_name = column_filename(t, col_num);
 155
 156         if (!column_name)
 157                 return NULL;
 158         if (!(t->desc->flags & OSL_LARGE_TABLE))
 159                 return column_name;
 160         dirname = make_message("%s/%.2s", column_name, ds_name);
 161         free(column_name);
 162         return dirname;
 163 }
 164
 165 static char *disk_storage_name_of_object(const struct osl_table *t,
 166         const struct osl_object *obj)
 167 {
 168         HASH_TYPE hash[HASH_SIZE];
 169         hash_object(obj, hash);
 170         return disk_storage_name_of_hash(t, hash);
 171 }
 172
 173 static int disk_storage_name_of_row(const struct osl_table *t,
 174                 const struct osl_row *row, char **name)
 175 {
 176         struct osl_object obj;
 177         int ret = osl_get_object(t, row, t->disk_storage_name_column, &obj);
 178
 179         if (ret < 0)
 180                 return ret;
 181         *name = disk_storage_name_of_object(t, &obj);
 182         if (*name)
 183                 return 1;
 184         return -ERRNO_TO_ERROR(ENOMEM);
 185 }
 186
 187 static void column_name_hash(const char *col_name, HASH_TYPE *hash)
 188 {
 189         hash_function(col_name, strlen(col_name), hash);
 190 }
 191
 192 static int init_column_descriptions(struct osl_table *t)
 193 {
 194         int i, j, ret;
 195         const struct osl_column_description *cd;
 196
 197         ret = -E_OSL_BAD_TABLE_DESC;
 198         ret = verify_name(t->desc->name);
 199         if (ret < 0)
 200                 goto err;
 201         ret = -E_OSL_BAD_DB_DIR;
 202         if (!t->desc->dir && (t->num_disk_storage_columns || t->num_mapped_columns))
 203                 goto err;
 204         /* the size of the index header without column descriptions */
 205         t->index_header_size = IDX_COLUMN_DESCRIPTIONS;
 206         FOR_EACH_COLUMN(i, t->desc, cd) {
 207                 struct osl_column *col = t->columns + i;
 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                 column_name_hash(cd->name, col->name_hash);
 222                 ret = -E_OSL_DUPLICATE_COL_NAME;
 223                 for (j = i + 1; j < t->desc->num_columns; j++) {
 224                         const char *name2 = get_column_description(t->desc,
 225                                 j)->name;
 226                         if (cd->name && name2 && !strcmp(cd->name, name2))
 227                                 goto err;
 228                 }
 229         }
 230         return 1;
 231 err:
 232         return ret;
 233 }
 234
 235 /**
 236  * Initialize a struct table from given table description.
 237  *
 238  * \param desc The description of the osl table.
 239  * \param table_ptr Result is returned here.
 240  *
 241  * This function performs several sanity checks on \p desc and returns if any
 242  * of these tests fail. On success, a struct \p osl_table is allocated and
 243  * initialized with data derived from \p desc.
 244  *
 245  * \return Standard.
 246  *
 247  * \sa struct osl_table.
 248  */
 249 int init_table_structure(const struct osl_table_description *desc,
 250                 struct osl_table **table_ptr)
 251 {
 252         const struct osl_column_description *cd;
 253         struct osl_table *t = calloc(1, sizeof(*t));
 254         int i, ret = -ERRNO_TO_ERROR(ENOMEM), have_disk_storage_name_column = 0;
 255
 256         if (!t)
 257                 return ret;
 258         ret = -E_OSL_BAD_TABLE_DESC;
 259         if (!desc)
 260                 goto err;
 261         DEBUG_LOG("creating table structure for '%s' from table "
 262                 "description\n", desc->name);
 263         ret = -E_OSL_NO_COLUMN_DESC;
 264         if (!desc->column_descriptions)
 265                 goto err;
 266         ret = -E_OSL_NO_COLUMNS;
 267         if (!desc->num_columns)
 268                 goto err;
 269         ret = -ERRNO_TO_ERROR(ENOMEM);
 270         t->columns = calloc(desc->num_columns, sizeof(struct osl_column));
 271         if (!t->columns)
 272                 goto err;
 273         t->desc = desc;
 274         FOR_EACH_COLUMN(i, t->desc, cd) {
 275                 enum osl_storage_type st = cd->storage_type;
 276                 enum osl_storage_flags sf = cd->storage_flags;
 277                 struct osl_column *col = &t->columns[i];
 278
 279                 ret = -E_OSL_BAD_STORAGE_TYPE;
 280                 if (st != OSL_MAPPED_STORAGE && st != OSL_DISK_STORAGE
 281                                 && st != OSL_NO_STORAGE)
 282                         goto err;
 283                 ret = -E_OSL_BAD_STORAGE_FLAGS;
 284                 if (st == OSL_DISK_STORAGE && sf & OSL_RBTREE)
 285                         goto err;
 286                 ret = -E_OSL_BAD_STORAGE_SIZE;
 287                 if (sf & OSL_FIXED_SIZE && !cd->data_size)
 288                         goto err;
 289                 switch (st) {
 290                 case OSL_DISK_STORAGE:
 291                         t->num_disk_storage_columns++;
 292                         break;
 293                 case OSL_MAPPED_STORAGE:
 294                         t->num_mapped_columns++;
 295                         col->index_offset = t->row_index_size;
 296                         t->row_index_size += 8;
 297                         break;
 298                 case OSL_NO_STORAGE:
 299                         col->volatile_num = t->num_volatile_columns;
 300                         t->num_volatile_columns++;
 301                         break;
 302                 }
 303                 if (sf & OSL_RBTREE) {
 304                         col->rbtree_num = t->num_rbtrees;
 305                         t->num_rbtrees++;
 306                         if ((sf & OSL_UNIQUE) && (st == OSL_MAPPED_STORAGE)) {
 307                                 if (!have_disk_storage_name_column)
 308                                         t->disk_storage_name_column = i;
 309                                 have_disk_storage_name_column = 1;
 310                         }
 311                 }
 312         }
 313         ret = -E_OSL_NO_UNIQUE_RBTREE_COLUMN;
 314         if (t->num_disk_storage_columns && !have_disk_storage_name_column)
 315                 goto err;
 316         ret = -E_OSL_NO_RBTREE_COL;
 317         if (!t->num_rbtrees)
 318                 goto err;
 319         /* success */
 320         DEBUG_LOG("OK. Index entry size: %u\n", t->row_index_size);
 321         ret = init_column_descriptions(t);
 322         if (ret < 0)
 323                 goto err;
 324         *table_ptr = t;
 325         return 1;
 326 err:
 327         free(t->columns);
 328         free(t);
 329         return ret;
 330 }
 331
 332 /**
 333  * Read the table description from index header.
 334  *
 335  * \param map The memory mapping of the index file.
 336  * \param desc The values found in the index header are returned here.
 337  *
 338  * Read the index header, check for the paraslash magic string and the table version number.
 339  * Read all information stored in the index header into \a desc.
 340  *
 341  * \return Standard.
 342  *
 343  * \sa struct osl_table_description, osl_create_table.
 344  */
 345 int read_table_desc(struct osl_object *map, struct osl_table_description *desc)
 346 {
 347         char *buf = map->data;
 348         uint8_t version;
 349         uint16_t header_size;
 350         int ret, i;
 351         unsigned offset;
 352         struct osl_column_description *cd;
 353
 354         if (map->size < MIN_INDEX_HEADER_SIZE(1))
 355                 return -E_OSL_SHORT_TABLE;
 356         if (strncmp(buf + IDX_PARA_MAGIC, PARA_MAGIC, strlen(PARA_MAGIC)))
 357                 return -E_OSL_NO_MAGIC;
 358         version = read_u8(buf + IDX_VERSION);
 359         if (version < MIN_TABLE_VERSION || version > MAX_TABLE_VERSION)
 360                 return -E_OSL_VERSION_MISMATCH;
 361         desc->flags = read_u8(buf + IDX_TABLE_FLAGS);
 362         desc->num_columns = read_u16(buf + IDX_NUM_COLUMNS);
 363         INFO_LOG("%u columns\n", desc->num_columns);
 364         if (!desc->num_columns)
 365                 return -E_OSL_NO_COLUMNS;
 366         header_size = read_u16(buf + IDX_HEADER_SIZE);
 367         if (map->size < header_size)
 368                 return -E_OSL_BAD_SIZE;
 369         desc->column_descriptions = calloc(desc->num_columns,
 370                 sizeof(struct osl_column_description));
 371         if (!desc->column_descriptions)
 372                 return -ERRNO_TO_ERROR(ENOMEM);
 373         offset = IDX_COLUMN_DESCRIPTIONS;
 374         FOR_EACH_COLUMN(i, desc, cd) {
 375                 char *null_byte;
 376
 377                 ret = -E_OSL_SHORT_TABLE;
 378                 if (map->size < offset + MIN_IDX_COLUMN_DESCRIPTION_SIZE) {
 379                         ERROR_LOG("map size = %zu < %u = offset + min desc size\n",
 380                                 map->size, offset + MIN_IDX_COLUMN_DESCRIPTION_SIZE);
 381                         goto err;
 382                 }
 383                 cd->storage_type = read_u16(buf + offset + IDX_CD_STORAGE_TYPE);
 384                 cd->storage_flags = read_u16(buf + offset +
 385                         IDX_CD_STORAGE_FLAGS);
 386                 cd->data_size = read_u32(buf + offset + IDX_CD_DATA_SIZE);
 387                 null_byte = memchr(buf + offset + IDX_CD_NAME, '\0',
 388                         map->size - offset - IDX_CD_NAME);
 389                 ret = -E_OSL_INDEX_CORRUPTION;
 390                 if (!null_byte)
 391                         goto err;
 392                 ret = -ERRNO_TO_ERROR(ENOMEM);
 393                 cd->name = strdup(buf + offset + IDX_CD_NAME);
 394                 if (!cd->name)
 395                         goto err;
 396                 offset += index_column_description_size(cd->name);
 397         }
 398         if (offset != header_size) {
 399                 ret = -E_OSL_INDEX_CORRUPTION;
 400                 ERROR_LOG("real header size = %u != %u = stored header size\n",
 401                         offset, header_size);
 402                 goto err;
 403         }
 404         return 1;
 405 err:
 406         FOR_EACH_COLUMN(i, desc, cd)
 407                 free(cd->name);
 408         return ret;
 409 }
 410
 411 /*
 412  * check whether the table description given by \p t->desc matches the on-disk
 413  * table structure stored in the index of \a t.
 414  */
 415 static int compare_table_descriptions(struct osl_table *t)
 416 {
 417         int i, ret;
 418         struct osl_table_description desc;
 419         const struct osl_column_description *cd1, *cd2;
 420
 421         /* read the on-disk structure into desc */
 422         ret = read_table_desc(&t->index_map, &desc);
 423         if (ret < 0)
 424                 return ret;
 425         ret = -E_OSL_BAD_TABLE_FLAGS;
 426         if (desc.flags != t->desc->flags)
 427                 goto out;
 428         ret = -E_OSL_BAD_COLUMN_NUM;
 429         if (desc.num_columns > t->desc->num_columns)
 430                 goto out;
 431         if (desc.num_columns < t->desc->num_columns) {
 432                 struct osl_column_description *cd;
 433                 unsigned diff = t->desc->num_columns - desc.num_columns;
 434                 INFO_LOG("extending table by %u volatile columns\n", diff);
 435                 ret = -ERRNO_TO_ERROR(ENOMEM);
 436                 desc.column_descriptions = realloc(desc.column_descriptions,
 437                         t->desc->num_columns * sizeof(struct osl_column_description));
 438                 if (!desc.column_descriptions)
 439                         goto out;
 440                 for (i = desc.num_columns; i < t->desc->num_columns; i++) {
 441                         cd = get_column_description(&desc, i);
 442                         cd->storage_type = OSL_NO_STORAGE;
 443                         cd->name = NULL;
 444                 }
 445                 desc.num_columns += diff;
 446         }
 447         FOR_EACH_COLUMN(i, t->desc, cd1) {
 448                 cd2 = get_column_description(&desc, i);
 449                 ret = -E_OSL_BAD_STORAGE_TYPE;
 450                 if (cd1->storage_type != cd2->storage_type)
 451                         goto out;
 452                 if (cd1->storage_type == OSL_NO_STORAGE)
 453                         continue;
 454                 ret = -E_OSL_BAD_STORAGE_FLAGS;
 455                 if (cd1->storage_flags != cd2->storage_flags) {
 456                         ERROR_LOG("sf1 = %u != %u = sf2\n",
 457                                 cd1->storage_flags, cd2->storage_flags);
 458                         goto out;
 459                 }
 460                 ret = -E_OSL_BAD_DATA_SIZE;
 461                 if (cd1->storage_flags & OSL_FIXED_SIZE)
 462                         if (cd1->data_size != cd2->data_size)
 463                                 goto out;
 464                 ret = -E_OSL_BAD_COLUMN_NAME;
 465                 if (strcmp(cd1->name, cd2->name))
 466                         goto out;
 467         }
 468         INFO_LOG("table description of '%s' matches on-disk data, good\n",
 469                 t->desc->name);
 470         ret = 1;
 471 out:
 472         FOR_EACH_COLUMN(i, &desc, cd1)
 473                 free(cd1->name);
 474         free(desc.column_descriptions);
 475         return ret;
 476 }
 477
 478 static int create_table_index(struct osl_table *t)
 479 {
 480         char *buf, *filename;
 481         int i, ret;
 482         size_t size = t->index_header_size;
 483         const struct osl_column_description *cd;
 484         unsigned offset;
 485
 486         INFO_LOG("creating %zu byte index for table %s\n", size,
 487                 t->desc->name);
 488         buf = calloc(1, size);
 489         if (!buf)
 490                 return -ERRNO_TO_ERROR(ENOMEM);
 491         sprintf(buf + IDX_PARA_MAGIC, "%s", PARA_MAGIC);
 492         write_u8(buf + IDX_TABLE_FLAGS, t->desc->flags);
 493         write_u8(buf + IDX_DIRTY_FLAG, 0);
 494         write_u8(buf + IDX_VERSION, CURRENT_TABLE_VERSION);
 495         write_u16(buf + IDX_NUM_COLUMNS, t->num_mapped_columns + t->num_disk_storage_columns);
 496         write_u16(buf + IDX_HEADER_SIZE, t->index_header_size);
 497         offset = IDX_COLUMN_DESCRIPTIONS;
 498         FOR_EACH_COLUMN(i, t->desc, cd) {
 499                 /* no need to store info about volatile storage */
 500                 if (cd->storage_type == OSL_NO_STORAGE)
 501                         continue;
 502                 write_u16(buf + offset + IDX_CD_STORAGE_TYPE,
 503                         cd->storage_type);
 504                 write_u16(buf + offset + IDX_CD_STORAGE_FLAGS,
 505                         cd->storage_flags);
 506                 if (cd->storage_flags & OSL_FIXED_SIZE)
 507                         write_u32(buf + offset + IDX_CD_DATA_SIZE,
 508                                 cd->data_size);
 509                 strcpy(buf + offset + IDX_CD_NAME, cd->name);
 510                 offset += index_column_description_size(cd->name);
 511         }
 512         assert(offset = size);
 513         filename = index_filename(t->desc);
 514         if (filename)
 515                 ret = write_file(filename, buf, size);
 516         else
 517                 ret = -ERRNO_TO_ERROR(ENOMEM);
 518         free(buf);
 519         free(filename);
 520         return ret;
 521 }
 522
 523 __export int osl_create_table(const struct osl_table_description *desc)
 524 {
 525         const struct osl_column_description *cd;
 526         char *table_dir = NULL, *filename;
 527         struct osl_table *t;
 528         int i, ret = init_table_structure(desc, &t);
 529
 530         if (ret < 0)
 531                 return ret;
 532         INFO_LOG("creating %s\n", desc->name);
 533         FOR_EACH_COLUMN(i, t->desc, cd) {
 534                 if (cd->storage_type == OSL_NO_STORAGE)
 535                         continue;
 536                 if (!table_dir) {
 537                         ret = osl_mkdir(desc->dir, 0777);
 538                         if (ret < 0 && !is_errno(-ret, EEXIST))
 539                                 goto out;
 540                         table_dir = make_message("%s/%s", desc->dir,
 541                                 desc->name);
 542                         ret = -ERRNO_TO_ERROR(ENOMEM);
 543                         if (!table_dir)
 544                                 goto out;
 545                         ret = osl_mkdir(table_dir, 0777);
 546                         if (ret < 0)
 547                                 goto out;
 548                 }
 549                 ret = -ERRNO_TO_ERROR(ENOMEM);
 550                 filename = column_filename(t, i);
 551                 if (!filename)
 552                         goto out;
 553                 INFO_LOG("filename: %s\n", filename);
 554                 if (cd->storage_type == OSL_MAPPED_STORAGE) {
 555                         ret = osl_open(filename, O_RDWR | O_CREAT | O_EXCL,
 556                                 0644);
 557                         free(filename);
 558                         if (ret < 0)
 559                                 goto out;
 560                         close(ret);
 561                         continue;
 562                 }
 563                 /* DISK STORAGE */
 564                 ret = osl_mkdir(filename, 0777);
 565                 free(filename);
 566                 if (ret < 0)
 567                         goto out;
 568         }
 569         if (t->num_mapped_columns) {
 570                 ret = create_table_index(t);
 571                 if (ret < 0)
 572                         goto out;
 573         }
 574         ret = 1;
 575 out:
 576         free(table_dir);
 577         free(t->columns);
 578         free(t);
 579         return ret;
 580 }
 581
 582 static int table_is_dirty(struct osl_table *t)
 583 {
 584         char *buf = (char *)t->index_map.data + IDX_DIRTY_FLAG;
 585         uint8_t dirty = read_u8(buf) & 0x1;
 586         return !!dirty;
 587 }
 588
 589 static void mark_table_dirty(struct osl_table *t)
 590 {
 591         char *buf = (char *)t->index_map.data + IDX_DIRTY_FLAG;
 592         write_u8(buf, read_u8(buf) | 1);
 593 }
 594
 595 static void mark_table_clean(struct osl_table *t)
 596 {
 597         char *buf = (char *)t->index_map.data + IDX_DIRTY_FLAG;
 598         write_u8(buf, read_u8(buf) & 0xfe);
 599 }
 600
 601 static void unmap_column(struct osl_table *t, unsigned col_num)
 602 {
 603         struct osl_object map = t->columns[col_num].data_map;
 604         int ret;
 605         if (!map.data)
 606                 return;
 607         ret = osl_munmap(map.data, map.size);
 608         assert(ret > 0);
 609         map.data = NULL;
 610 }
 611
 612 /**
 613  * Unmap all mapped files of an osl table.
 614  *
 615  * \param t Pointer to a mapped table.
 616  * \param flags Options for unmapping.
 617  *
 618  * \return Positive on success, negative on errors.
 619  *
 620  * \sa map_table(), enum osl_close_flags, osl_munmap().
 621  */
 622 int unmap_table(struct osl_table *t, enum osl_close_flags flags)
 623 {
 624         unsigned i;
 625         const struct osl_column_description *cd;
 626         int ret;
 627
 628         if (!t->num_mapped_columns) /* can this ever happen? */
 629                 return 1;
 630         INFO_LOG("unmapping table '%s'\n", t->desc->name);
 631         if (!t->index_map.data)
 632                 return -E_OSL_NOT_MAPPED;
 633         if (flags & OSL_MARK_CLEAN)
 634                 mark_table_clean(t);
 635         ret = osl_munmap(t->index_map.data, t->index_map.size);
 636         if (ret < 0)
 637                 return ret;
 638         t->index_map.data = NULL;
 639         if (!t->num_rows)
 640                 return 1;
 641         FOR_EACH_MAPPED_COLUMN(i, t, cd)
 642                 unmap_column(t, i);
 643         return 1;
 644 }
 645
 646 static int map_column(struct osl_table *t, unsigned col_num)
 647 {
 648         struct stat statbuf;
 649         char *filename = column_filename(t, col_num);
 650         int ret;
 651
 652         if (!filename)
 653                 return -ERRNO_TO_ERROR(ENOMEM);
 654         ret = osl_stat(filename, &statbuf);
 655         if (ret < 0) {
 656                 free(filename);
 657                 return ret;
 658         }
 659         if (!(S_IFREG & statbuf.st_mode)) {
 660                 free(filename);
 661                 return ret;
 662         }
 663         ret = mmap_full_file(filename, O_RDWR,
 664                 &t->columns[col_num].data_map.data,
 665                 &t->columns[col_num].data_map.size,
 666                 NULL);
 667         free(filename);
 668         return ret;
 669 }
 670
 671 /**
 672  * Map the index file and all columns of type \p OSL_MAPPED_STORAGE into memory.
 673  *
 674  * \param t Pointer to an initialized table structure.
 675  * \param flags Mapping options.
 676  *
 677  * \return Negative return value on errors; on success the number of rows
 678  * (including invalid rows) is returned.
 679  *
 680  * \sa unmap_table(), enum map_table_flags, osl_open_table(), mmap(2).
 681  */
 682 int map_table(struct osl_table *t, enum map_table_flags flags)
 683 {
 684         char *filename;
 685         const struct osl_column_description *cd;
 686         int i = 0, ret, num_rows = 0;
 687
 688         if (!t->num_mapped_columns)
 689                 return 0;
 690         if (t->index_map.data)
 691                 return -E_OSL_ALREADY_MAPPED;
 692         filename = index_filename(t->desc);
 693         if (!filename)
 694                 return -ERRNO_TO_ERROR(ENOMEM);
 695         INFO_LOG("mapping table '%s' (index: %s)\n", t->desc->name, filename);
 696         ret = mmap_full_file(filename, flags & MAP_TBL_FL_MAP_RDONLY?
 697                 O_RDONLY : O_RDWR, &t->index_map.data, &t->index_map.size, NULL);
 698         free(filename);
 699         if (ret < 0)
 700                 return ret;
 701         if (flags & MAP_TBL_FL_VERIFY_INDEX) {
 702                 ret = compare_table_descriptions(t);
 703                 if (ret < 0)
 704                         goto err;
 705         }
 706         ret = -E_OSL_BUSY;
 707         if (!(flags & MAP_TBL_FL_IGNORE_DIRTY)) {
 708                 if (table_is_dirty(t)) {
 709                         ERROR_LOG("%s is dirty\n", t->desc->name);
 710                         goto err;
 711                 }
 712         }
 713         mark_table_dirty(t);
 714         num_rows = table_num_rows(t);
 715         if (!num_rows)
 716                 return num_rows;
 717         /* map data files */
 718         FOR_EACH_MAPPED_COLUMN(i, t, cd) {
 719                 ret = map_column(t, i);
 720                 if (ret < 0)
 721                         goto err;
 722         }
 723         return num_rows;
 724 err:    /* unmap what is already mapped */
 725         for (i--; i >= 0; i--) {
 726                 struct osl_object map = t->columns[i].data_map;
 727                 osl_munmap(map.data, map.size);
 728                 map.data = NULL;
 729         }
 730         osl_munmap(t->index_map.data, t->index_map.size);
 731         t->index_map.data = NULL;
 732         return ret;
 733 }
 734
 735 /**
 736  * Retrieve a mapped object by row and column number.
 737  *
 738  * \param t Pointer to an open osl table.
 739  * \param col_num Number of the mapped column containing the object to retrieve.
 740  * \param row_num Number of the row containing the object to retrieve.
 741  * \param obj The result is returned here.
 742  *
 743  * It is considered an error if \a col_num does not refer to a column
 744  * of storage type \p OSL_MAPPED_STORAGE.
 745  *
 746  * \return Standard.
 747  *
 748  * \sa osl_storage_type.
 749  */
 750 int get_mapped_object(const struct osl_table *t, unsigned col_num,
 751         uint32_t row_num, struct osl_object *obj)
 752 {
 753         struct osl_column *col = &t->columns[col_num];
 754         uint32_t offset;
 755         char *cell_index;
 756         int ret;
 757
 758         if (t->num_rows <= row_num)
 759                 return -E_OSL_BAD_ROW_NUM;
 760         ret = get_cell_index(t, row_num, col_num, &cell_index);
 761         if (ret < 0)
 762                 return ret;
 763         offset = read_u32(cell_index);
 764         obj->size = read_u32(cell_index + 4);
 765         obj->data = col->data_map.data + offset;
 766         return 1;
 767 }
 768
 769 /*
 770  * It's OK to call this with result = rb_node = NULL.  If result is not NULL,
 771  * and rb key was not found, result points to the parent node.
 772  */
 773 static int search_rbtree(const struct osl_object *obj,
 774                 const struct osl_table *t, unsigned col_num,
 775                 struct rb_node **result, struct rb_node ***rb_link)
 776 {
 777         struct osl_column *col = &t->columns[col_num];
 778         struct rb_node **new = &col->rbtree.rb_node, *parent = NULL;
 779         const struct osl_column_description *cd =
 780                 get_column_description(t->desc, col_num);
 781         enum osl_storage_type st = cd->storage_type;
 782         while (*new) {
 783                 struct osl_row *this_row = get_row_pointer(*new,
 784                         col->rbtree_num);
 785                 int ret;
 786                 struct osl_object this_obj;
 787                 parent = *new;
 788                 if (st == OSL_MAPPED_STORAGE) {
 789                         ret = get_mapped_object(t, col_num, this_row->num,
 790                                 &this_obj);
 791                         if (ret < 0)
 792                                 return ret;
 793                 } else
 794                         this_obj = this_row->volatile_objects[col->volatile_num];
 795                 ret = cd->compare_function(obj, &this_obj);
 796                 if (!ret) {
 797                         if (result)
 798                                 *result = get_rb_node_pointer(this_row,
 799                                         col->rbtree_num);
 800                         return 1;
 801                 }
 802                 if (ret < 0)
 803                         new = &((*new)->rb_left);
 804                 else
 805                         new = &((*new)->rb_right);
 806         }
 807         if (result)
 808                 *result = parent;
 809         if (rb_link)
 810                 *rb_link = new;
 811         return -E_OSL_RB_KEY_NOT_FOUND;
 812 }
 813
 814 static int insert_rbtree(struct osl_table *t, unsigned col_num,
 815         const struct osl_row *row, const struct osl_object *obj)
 816 {
 817         struct rb_node *parent, **rb_link;
 818         unsigned rbtree_num;
 819         struct rb_node *n;
 820         int ret = search_rbtree(obj, t, col_num, &parent, &rb_link);
 821
 822         if (ret > 0)
 823                 return -E_OSL_RB_KEY_EXISTS;
 824         rbtree_num = t->columns[col_num].rbtree_num;
 825         n = get_rb_node_pointer(row, rbtree_num);
 826         rb_link_node(n, parent, rb_link);
 827         rb_insert_color(n, &t->columns[col_num].rbtree);
 828         return 1;
 829 }
 830
 831 static void remove_rb_node(struct osl_table *t, unsigned col_num,
 832                 const struct osl_row *row)
 833 {
 834         struct osl_column *col = &t->columns[col_num];
 835         const struct osl_column_description *cd =
 836                 get_column_description(t->desc, col_num);
 837         enum osl_storage_flags sf = cd->storage_flags;
 838         struct rb_node *victim, *splice_out_node, *tmp;
 839         if (!(sf & OSL_RBTREE))
 840                 return;
 841         /*
 842          * Which node is removed/spliced out actually depends on how many
 843          * children the victim node has: If it has no children, it gets
 844          * deleted. If it has one child, it gets spliced out. If it has two
 845          * children, its successor (which has at most a right child) gets
 846          * spliced out.
 847          */
 848         victim = get_rb_node_pointer(row, col->rbtree_num);
 849         if (victim->rb_left && victim->rb_right)
 850                 splice_out_node = rb_next(victim);
 851         else
 852                 splice_out_node = victim;
 853         /* Go up to the root and decrement the size of each node in the path. */
 854         for (tmp = splice_out_node; tmp; tmp = rb_parent(tmp))
 855                 tmp->size--;
 856         rb_erase(victim, &col->rbtree);
 857 }
 858
 859 static int add_row_to_rbtrees(struct osl_table *t, uint32_t row_num,
 860                 struct osl_object *volatile_objs, struct osl_row **row_ptr)
 861 {
 862         unsigned i;
 863         int ret;
 864         struct osl_row *row = allocate_row(t->num_rbtrees);
 865         const struct osl_column_description *cd;
 866
 867         if (!row)
 868                 return -ERRNO_TO_ERROR(ENOMEM);
 869         row->num = row_num;
 870         row->volatile_objects = volatile_objs;
 871         FOR_EACH_RBTREE_COLUMN(i, t, cd) {
 872                 if (cd->storage_type == OSL_MAPPED_STORAGE) {
 873                         struct osl_object obj;
 874                         ret = get_mapped_object(t, i, row_num, &obj);
 875                         if (ret < 0)
 876                                 goto err;
 877                         ret = insert_rbtree(t, i, row, &obj);
 878                 } else { /* volatile */
 879                         const struct osl_object *obj
 880                                 = volatile_objs + t->columns[i].volatile_num;
 881                         ret = insert_rbtree(t, i, row, obj);
 882                 }
 883                 if (ret < 0)
 884                         goto err;
 885         }
 886         if (row_ptr)
 887                 *row_ptr = row;
 888         return 1;
 889 err: /* rollback changes, i.e. remove added entries from rbtrees */
 890         while (i)
 891                 remove_rb_node(t, i--, row);
 892         free(row);
 893         return ret;
 894 }
 895
 896 static void free_volatile_objects(const struct osl_table *t,
 897                 enum osl_close_flags flags)
 898 {
 899         int i, j;
 900         struct rb_node *n;
 901         struct osl_column *rb_col;
 902         const struct osl_column_description *cd;
 903
 904         if (!t->num_volatile_columns)
 905                 return;
 906         /* find the first rbtree column (any will do) */
 907         FOR_EACH_RBTREE_COLUMN(i, t, cd)
 908                 break;
 909         rb_col = t->columns + i;
 910         /* walk that rbtree and free all volatile objects */
 911         for (n = rb_first(&rb_col->rbtree); n; n = rb_next(n)) {
 912                 struct osl_row *r = get_row_pointer(n, rb_col->rbtree_num);
 913                 if (flags & OSL_FREE_VOLATILE)
 914                         FOR_EACH_VOLATILE_COLUMN(j, t, cd) {
 915                                 if (cd->storage_flags & OSL_DONT_FREE)
 916                                         continue;
 917                                 free(r->volatile_objects[
 918                                         t->columns[j].volatile_num].data);
 919                         }
 920 //                      for (j = 0; j < t->num_volatile_columns; j++)
 921 //                              free(r->volatile_objects[j].data);
 922                 free(r->volatile_objects);
 923         }
 924 }
 925
 926 /**
 927  * Erase all rbtree nodes and free resources.
 928  *
 929  * \param t Pointer to an open osl table.
 930  *
 931  * This function is called by osl_close_table().
 932  */
 933 void clear_rbtrees(struct osl_table *t)
 934 {
 935         const struct osl_column_description *cd;
 936         unsigned i, rbtrees_cleared = 0;
 937
 938         FOR_EACH_RBTREE_COLUMN(i, t, cd) {
 939                 struct osl_column *col = &t->columns[i];
 940                 struct rb_node *n;
 941                 rbtrees_cleared++;
 942                 for (n = rb_first(&col->rbtree); n;) {
 943                         struct osl_row *r;
 944                         rb_erase(n, &col->rbtree);
 945                         if (rbtrees_cleared == t->num_rbtrees) {
 946                                 r = get_row_pointer(n, col->rbtree_num);
 947                                 n = rb_next(n);
 948                                 free(r);
 949                         } else
 950                                 n = rb_next(n);
 951                 }
 952         }
 953
 954 }
 955
 956 __export int osl_close_table(struct osl_table *t, enum osl_close_flags flags)
 957 {
 958         int ret;
 959
 960         if (!t)
 961                 return -E_OSL_BAD_TABLE;
 962         NOTICE_LOG("closing table %s\n", t->desc->name);
 963         free_volatile_objects(t, flags);
 964         clear_rbtrees(t);
 965         ret = unmap_table(t, flags);
 966         if (ret < 0)
 967                 ERROR_LOG("unmap_table failed: %d\n", ret);
 968         free(t->columns);
 969         free(t);
 970         return ret;
 971 }
 972
 973 /**
 974  * Find out whether the given row number corresponds to an invalid row.
 975  *
 976  * \param t Pointer to the osl table.
 977  * \param row_num The number of the row in question.
 978  *
 979  * By definition, a row is considered invalid if all its index entries
 980  * are invalid.
 981  *
 982  * \return Positive if \a row_num corresponds to an invalid row,
 983  * zero if it corresponds to a valid row, negative on errors.
 984  */
 985 int row_is_invalid(struct osl_table *t, uint32_t row_num)
 986 {
 987         char *row_index;
 988         int i, ret = get_row_index(t, row_num, &row_index);
 989
 990         if (ret < 0)
 991                 return ret;
 992         for (i = 0; i < t->row_index_size; i++) {
 993                 if ((unsigned char)row_index[i] != 0xff)
 994                         return 0;
 995         }
 996         INFO_LOG("row %d is invalid\n", row_num);
 997         return 1;
 998 }
 999
1000 /**
1001  * Invalidate a row of an osl table.
1002  *
1003  * \param t Pointer to an open osl table.
1004  * \param row_num Number of the row to mark as invalid.
1005  *
1006  * This function marks each mapped object in the index entry of \a row as
1007  * invalid.
1008  *
1009  * \return Standard.
1010  */
1011 int mark_row_invalid(struct osl_table *t, uint32_t row_num)
1012 {
1013         char *row_index;
1014         int ret = get_row_index(t, row_num, &row_index);
1015
1016         if (ret < 0)
1017                 return ret;
1018         INFO_LOG("marking row %d as invalid\n", row_num);
1019         memset(row_index, 0xff, t->row_index_size);
1020         return 1;
1021 }
1022
1023 /**
1024  * Initialize all rbtrees and compute number of invalid rows.
1025  *
1026  * \param t The table containing the rbtrees to be initialized.
1027  *
1028  * \return Standard.
1029  */
1030 int init_rbtrees(struct osl_table *t)
1031 {
1032         int i, ret;
1033         const struct osl_column_description *cd;
1034
1035         /* create rbtrees */
1036         FOR_EACH_RBTREE_COLUMN(i, t, cd)
1037                 t->columns[i].rbtree = RB_ROOT;
1038         /* add valid rows to rbtrees */
1039         t->num_invalid_rows = 0;
1040         for (i = 0; i < t->num_rows; i++) {
1041                 ret = row_is_invalid(t, i);
1042                 if (ret < 0)
1043                         return ret;
1044                 if (ret) {
1045                         t->num_invalid_rows++;
1046                         continue;
1047                 }
1048                 ret = add_row_to_rbtrees(t, i, NULL, NULL);
1049                 if (ret < 0)
1050                         return ret;
1051         }
1052         return 1;
1053 }
1054
1055 __export int osl_open_table(const struct osl_table_description *table_desc,
1056                 struct osl_table **result)
1057 {
1058         int i, ret;
1059         struct osl_table *t;
1060         const struct osl_column_description *cd;
1061
1062         NOTICE_LOG("opening table %s\n", table_desc->name);
1063         ret = init_table_structure(table_desc, &t);
1064         if (ret < 0)
1065                 return ret;
1066         FOR_EACH_DISK_STORAGE_COLUMN(i, t, cd) {
1067                 struct stat statbuf;
1068                 char *dirname = column_filename(t, i);
1069
1070                 ret = -ERRNO_TO_ERROR(ENOMEM);
1071                 if (!dirname)
1072                         goto err;
1073                 /* check if directory exists */
1074                 ret = stat(dirname, &statbuf);
1075                 free(dirname);
1076                 if (ret < 0) {
1077                         ret = -ERRNO_TO_ERROR(errno);
1078                         goto err;
1079                 }
1080                 ret = -ERRNO_TO_ERROR(ENOTDIR);
1081                 if (!S_ISDIR(statbuf.st_mode))
1082                         goto err;
1083         }
1084         ret = map_table(t, MAP_TBL_FL_VERIFY_INDEX);
1085         if (ret < 0)
1086                 goto err;
1087         t->num_rows = ret;
1088         DEBUG_LOG("num rows: %d\n", t->num_rows);
1089         ret = init_rbtrees(t);
1090         if (ret < 0) {
1091                 osl_close_table(t, OSL_MARK_CLEAN); /* ignore further errors */
1092                 return ret;
1093         }
1094         *result = t;
1095         return 1;
1096 err:
1097         free(t->columns);
1098         free(t);
1099         return ret;
1100 }
1101
1102 static int create_disk_storage_object_dir(const struct osl_table *t,
1103                 unsigned col_num, const char *ds_name)
1104 {
1105         char *dirname;
1106         int ret;
1107
1108         if (!(t->desc->flags & OSL_LARGE_TABLE))
1109                 return 1;
1110         dirname = disk_storage_dirname(t, col_num, ds_name);
1111         if (!dirname)
1112                 return -ERRNO_TO_ERROR(ENOMEM);
1113         ret = osl_mkdir(dirname, 0777);
1114         free(dirname);
1115         if (ret < 0 && !is_errno(-ret, EEXIST))
1116                 return ret;
1117         return 1;
1118 }
1119
1120 static int write_disk_storage_file(const struct osl_table *t, unsigned col_num,
1121         const struct osl_object *obj, const char *ds_name)
1122 {
1123         int ret;
1124         char *filename;
1125
1126         ret = create_disk_storage_object_dir(t, col_num, ds_name);
1127         if (ret < 0)
1128                 return ret;
1129         filename = disk_storage_path(t, col_num, ds_name);
1130         if (!filename)
1131                 return -ERRNO_TO_ERROR(ENOMEM);
1132         ret = write_file(filename, obj->data, obj->size);
1133         free(filename);
1134         return ret;
1135 }
1136
1137 static int append_map_file(const struct osl_table *t, unsigned col_num,
1138         const struct osl_object *obj, uint32_t *new_size)
1139 {
1140         char *filename = column_filename(t, col_num);
1141         int ret;
1142
1143         if (!filename)
1144                 return -ERRNO_TO_ERROR(ENOMEM);
1145         ret = append_file(filename, obj->data, obj->size, new_size);
1146         free(filename);
1147         return ret;
1148 }
1149
1150 static int append_row_index(const struct osl_table *t, char *row_index)
1151 {
1152         char *filename;
1153         int ret;
1154
1155         if (!t->num_mapped_columns)
1156                 return 1;
1157         filename = index_filename(t->desc);
1158         if (!filename)
1159                 return -ERRNO_TO_ERROR(ENOMEM);
1160         ret = append_file(filename, row_index, t->row_index_size, NULL);
1161         free(filename);
1162         return ret;
1163 }
1164
1165 static int truncate_mapped_file(const struct osl_table *t, unsigned col_num,
1166                 off_t size)
1167 {
1168         int ret;
1169         char *filename = column_filename(t, col_num);
1170
1171         if (!filename)
1172                 return -ERRNO_TO_ERROR(ENOMEM);
1173         ret = truncate_file(filename, size);
1174         free(filename);
1175         return ret;
1176 }
1177
1178 static int delete_disk_storage_file(const struct osl_table *t, unsigned col_num,
1179                 const char *ds_name)
1180 {
1181         char *dirname, *filename = disk_storage_path(t, col_num, ds_name);
1182         int ret, err;
1183
1184         if (!filename)
1185                 return -ERRNO_TO_ERROR(ENOMEM);
1186         ret = unlink(filename);
1187         err = errno;
1188         free(filename);
1189         if (ret < 0)
1190                 return -ERRNO_TO_ERROR(err);
1191         if (!(t->desc->flags & OSL_LARGE_TABLE))
1192                 return 1;
1193         dirname = disk_storage_dirname(t, col_num, ds_name);
1194         if (!dirname)
1195                 return -ERRNO_TO_ERROR(ENOMEM);
1196         rmdir(dirname);
1197         free(dirname);
1198         return 1;
1199 }
1200
1201 __export int osl_add_and_get_row(struct osl_table *t, struct osl_object *objects,
1202                 struct osl_row **row)
1203 {
1204         int i, ret;
1205         char *ds_name = NULL;
1206         struct rb_node **rb_parents = NULL, ***rb_links = NULL;
1207         char *new_row_index = NULL;
1208         struct osl_object *volatile_objs = NULL;
1209         const struct osl_column_description *cd;
1210
1211         if (!t)
1212                 return -E_OSL_BAD_TABLE;
1213         rb_parents = malloc(t->num_rbtrees * sizeof(struct rn_node*));
1214         if (!rb_parents)
1215                 return -ERRNO_TO_ERROR(ENOMEM);
1216         rb_links = malloc(t->num_rbtrees * sizeof(struct rn_node**));
1217         if (!rb_links) {
1218                 free(rb_parents);
1219                 return -ERRNO_TO_ERROR(ENOMEM);
1220         }
1221         if (t->num_mapped_columns) {
1222                 new_row_index = malloc(t->row_index_size);
1223                 if (!new_row_index) {
1224                         free(rb_links);
1225                         free(rb_parents);
1226                         return -ERRNO_TO_ERROR(ENOMEM);
1227                 }
1228         }
1229         /* pass 1: sanity checks */
1230 //      DEBUG_LOG("sanity tests: %p:%p\n", objects[0].data,
1231 //              objects[1].data);
1232         FOR_EACH_COLUMN(i, t->desc, cd) {
1233                 enum osl_storage_type st = cd->storage_type;
1234                 enum osl_storage_flags sf = cd->storage_flags;
1235
1236 //              ret = -E_OSL_NULL_OBJECT;
1237 //              if (!objects[i])
1238 //                      goto out;
1239                 if (st == OSL_DISK_STORAGE)
1240                         continue;
1241                 if (sf & OSL_RBTREE) {
1242                         unsigned rbtree_num = t->columns[i].rbtree_num;
1243                         ret = -E_OSL_RB_KEY_EXISTS;
1244 //                      DEBUG_LOG("checking whether %p exists\n",
1245 //                              objects[i].data);
1246                         if (search_rbtree(objects + i, t, i,
1247                                         &rb_parents[rbtree_num],
1248                                         &rb_links[rbtree_num]) > 0)
1249                                 goto out;
1250                 }
1251                 if (sf & OSL_FIXED_SIZE) {
1252 //                      DEBUG_LOG("fixed size. need: %zu, have: %d\n",
1253 //                              objects[i].size, cd->data_size);
1254                         ret = -E_OSL_BAD_DATA_SIZE;
1255                         if (objects[i].size != cd->data_size)
1256                                 goto out;
1257                 }
1258         }
1259         if (t->num_disk_storage_columns) {
1260                 ds_name = disk_storage_name_of_object(t,
1261                         &objects[t->disk_storage_name_column]);
1262                 ret = -ERRNO_TO_ERROR(ENOMEM);
1263                 if (!ds_name)
1264                         goto out;
1265         }
1266         ret = unmap_table(t, OSL_MARK_CLEAN);
1267         if (ret < 0)
1268                 goto out;
1269 //      DEBUG_LOG("sanity tests passed%s\n", "");
1270         /* pass 2: create data files, append map data */
1271         FOR_EACH_COLUMN(i, t->desc, cd) {
1272                 enum osl_storage_type st = cd->storage_type;
1273                 if (st == OSL_NO_STORAGE)
1274                         continue;
1275                 if (st == OSL_MAPPED_STORAGE) {
1276                         uint32_t new_size;
1277                         struct osl_column *col = &t->columns[i];
1278 //                      DEBUG_LOG("appending object of size %zu\n",
1279 //                              objects[i].size);
1280                         ret = append_map_file(t, i, objects + i, &new_size);
1281                         if (ret < 0)
1282                                 goto rollback;
1283                         update_cell_index(new_row_index, col, new_size,
1284                                 objects[i].size);
1285                         continue;
1286                 }
1287                 /* DISK_STORAGE */
1288                 ret = write_disk_storage_file(t, i, objects + i, ds_name);
1289                 if (ret < 0)
1290                         goto rollback;
1291         }
1292         ret = append_row_index(t, new_row_index);
1293         if (ret < 0)
1294                 goto rollback;
1295         ret = map_table(t, MAP_TBL_FL_VERIFY_INDEX);
1296         if (ret < 0) { /* truncate index and rollback changes */
1297                 char *filename = index_filename(t->desc);
1298                 if (filename)
1299                         truncate_file(filename, t->row_index_size);
1300                 free(filename);
1301                 goto rollback;
1302         }
1303         /* pass 3: add entry to rbtrees */
1304         if (t->num_volatile_columns) {
1305                 ret = -ERRNO_TO_ERROR(ENOMEM);
1306                 volatile_objs = calloc(t->num_volatile_columns,
1307                         sizeof(struct osl_object));
1308                 if (!volatile_objs)
1309                         goto out;
1310                 FOR_EACH_VOLATILE_COLUMN(i, t, cd)
1311                         volatile_objs[t->columns[i].volatile_num] = objects[i];
1312         }
1313         t->num_rows++;
1314 //      DEBUG_LOG("adding new entry as row #%d\n", t->num_rows - 1);
1315         ret = add_row_to_rbtrees(t, t->num_rows - 1, volatile_objs, row);
1316         if (ret < 0)
1317                 goto out;
1318 //      DEBUG_LOG("added new entry as row #%d\n", t->num_rows - 1);
1319         ret = 1;
1320         goto out;
1321 rollback: /* rollback all changes made, ignore further errors */
1322         for (i--; i >= 0; i--) {
1323                 cd = get_column_description(t->desc, i);
1324                 enum osl_storage_type st = cd->storage_type;
1325                 if (st == OSL_NO_STORAGE)
1326                         continue;
1327
1328                 if (st == OSL_MAPPED_STORAGE)
1329                         truncate_mapped_file(t, i, objects[i].size);
1330                 else /* disk storage */
1331                         delete_disk_storage_file(t, i, ds_name);
1332         }
1333         /* ignore error and return previous error value */
1334         map_table(t, MAP_TBL_FL_VERIFY_INDEX);
1335 out:
1336         free(new_row_index);
1337         free(ds_name);
1338         free(rb_parents);
1339         free(rb_links);
1340         return ret;
1341 }
1342
1343 __export int osl_add_row(struct osl_table *t, struct osl_object *objects)
1344 {
1345         return osl_add_and_get_row(t, objects, NULL);
1346 }
1347
1348 __export int osl_get_object(const struct osl_table *t, const struct osl_row *r,
1349         unsigned col_num, struct osl_object *object)
1350 {
1351         const struct osl_column_description *cd;
1352
1353         if (!t)
1354                 return -E_OSL_BAD_TABLE;
1355         cd = get_column_description(t->desc, col_num);
1356         /* col must not be disk storage */
1357         if (cd->storage_type == OSL_DISK_STORAGE)
1358                 return -E_OSL_BAD_STORAGE_TYPE;
1359         if (cd->storage_type == OSL_MAPPED_STORAGE)
1360                 return get_mapped_object(t, col_num, r->num, object);
1361         /* volatile */
1362         *object = r->volatile_objects[t->columns[col_num].volatile_num];
1363         return 1;
1364 }
1365
1366 __export int osl_del_row(struct osl_table *t, struct osl_row *row)
1367 {
1368         struct osl_row *r = row;
1369         int i, ret;
1370         const struct osl_column_description *cd;
1371
1372         if (!t)
1373                 return -E_OSL_BAD_TABLE;
1374         INFO_LOG("deleting row %p\n", row);
1375
1376         if (t->num_disk_storage_columns) {
1377                 char *ds_name;
1378                 ret = disk_storage_name_of_row(t, r, &ds_name);
1379                 if (ret < 0)
1380                         goto out;
1381                 FOR_EACH_DISK_STORAGE_COLUMN(i, t, cd)
1382                         delete_disk_storage_file(t, i, ds_name);
1383                 free(ds_name);
1384         }
1385         FOR_EACH_COLUMN(i, t->desc, cd) {
1386                 struct osl_column *col = t->columns + i;
1387                 enum osl_storage_type st = cd->storage_type;
1388                 remove_rb_node(t, i, r);
1389                 if (st == OSL_MAPPED_STORAGE)
1390                         continue;
1391                 if (st == OSL_NO_STORAGE && !(cd->storage_flags & OSL_DONT_FREE))
1392                         free(r->volatile_objects[col->volatile_num].data);
1393         }
1394         if (t->num_mapped_columns) {
1395                 ret = mark_row_invalid(t, r->num);
1396                 if (ret < 0)
1397                         goto out;
1398                 t->num_invalid_rows++;
1399         } else
1400                 t->num_rows--;
1401         ret = 1;
1402 out:
1403         free(r->volatile_objects);
1404         free(r);
1405         return ret;
1406 }
1407
1408 /* test if column has an rbtree */
1409 static int check_rbtree_col(const struct osl_table *t, unsigned col_num,
1410                 struct osl_column **col)
1411 {
1412         if (!t)
1413                 return -E_OSL_BAD_TABLE;
1414         if (!(get_column_description(t->desc, col_num)->storage_flags & OSL_RBTREE))
1415                 return -E_OSL_BAD_STORAGE_FLAGS;
1416         *col = t->columns + col_num;
1417         return 1;
1418 }
1419
1420 __export int osl_get_row(const struct osl_table *t, unsigned col_num,
1421                 const struct osl_object *obj, struct osl_row **result)
1422 {
1423         int ret;
1424         struct rb_node *node;
1425         struct osl_row *row;
1426         struct osl_column *col;
1427
1428         *result = NULL;
1429         ret = check_rbtree_col(t, col_num, &col);
1430         if (ret < 0)
1431                 return ret;
1432         ret = search_rbtree(obj, t, col_num, &node, NULL);
1433         if (ret < 0)
1434                 return ret;
1435         row = get_row_pointer(node, t->columns[col_num].rbtree_num);
1436         *result = row;
1437         return 1;
1438 }
1439
1440 static int rbtree_loop(struct osl_column *col, void *private_data,
1441                 osl_rbtree_loop_func *func)
1442 {
1443         struct rb_node *n, *tmp;
1444
1445         /* this for-loop is safe against removal of an entry */
1446         for (n = rb_first(&col->rbtree), tmp = n? rb_next(n) : NULL;
1447                         n;
1448                         n = tmp, tmp = tmp? rb_next(tmp) : NULL) {
1449                 struct osl_row *r = get_row_pointer(n, col->rbtree_num);
1450                 if (func(r, private_data) < 0)
1451                         return -E_OSL_LOOP;
1452         }
1453         return 1;
1454 }
1455
1456 static int rbtree_loop_reverse(struct osl_column *col, void *private_data,
1457                 osl_rbtree_loop_func *func)
1458 {
1459         struct rb_node *n, *tmp;
1460
1461         /* safe against removal of an entry */
1462         for (n = rb_last(&col->rbtree), tmp = n? rb_prev(n) : NULL;
1463                         n;
1464                         n = tmp, tmp = tmp? rb_prev(tmp) : NULL) {
1465                 struct osl_row *r = get_row_pointer(n, col->rbtree_num);
1466                 if (func(r, private_data) < 0)
1467                         return -E_OSL_LOOP;
1468         }
1469         return 1;
1470 }
1471
1472 __export int osl_rbtree_loop(const struct osl_table *t, unsigned col_num,
1473         void *private_data, osl_rbtree_loop_func *func)
1474 {
1475         struct osl_column *col;
1476
1477         int ret = check_rbtree_col(t, col_num, &col);
1478         if (ret < 0)
1479                 return ret;
1480         return rbtree_loop(col, private_data, func);
1481 }
1482
1483 __export int osl_rbtree_loop_reverse(const struct osl_table *t, unsigned col_num,
1484         void *private_data, osl_rbtree_loop_func *func)
1485 {
1486         struct osl_column *col;
1487
1488         int ret = check_rbtree_col(t, col_num, &col);
1489         if (ret < 0)
1490                 return ret;
1491         return rbtree_loop_reverse(col, private_data, func);
1492 }
1493
1494 /* TODO: Rollback changes on errors */
1495 static int rename_disk_storage_objects(struct osl_table *t,
1496                 struct osl_object *old_obj, struct osl_object *new_obj)
1497 {
1498         int i, ret;
1499         const struct osl_column_description *cd;
1500         char *old_ds_name, *new_ds_name;
1501
1502         if (!t->num_disk_storage_columns)
1503                 return 1; /* nothing to do */
1504         if (old_obj->size == new_obj->size && !memcmp(new_obj->data,
1505                         old_obj->data, new_obj->size))
1506                 return 1; /* object did not change */
1507         old_ds_name = disk_storage_name_of_object(t, old_obj);
1508         new_ds_name = disk_storage_name_of_object(t, new_obj);
1509         ret = -ERRNO_TO_ERROR(ENOMEM);
1510         if (!old_ds_name || ! new_ds_name)
1511                 goto out;
1512
1513         FOR_EACH_DISK_STORAGE_COLUMN(i, t, cd) {
1514                 char *old_filename, *new_filename;
1515                 ret = create_disk_storage_object_dir(t, i, new_ds_name);
1516                 if (ret < 0)
1517                         goto out;
1518                 old_filename = disk_storage_path(t, i, old_ds_name);
1519                 new_filename = disk_storage_path(t, i, new_ds_name);
1520                 if (!old_filename || !new_filename)
1521                         ret = -ERRNO_TO_ERROR(ENOMEM);
1522                 else
1523                         ret = osl_rename(old_filename, new_filename);
1524                 free(old_filename);
1525                 free(new_filename);
1526                 if (ret < 0)
1527                         goto out;
1528         }
1529         ret = 1;
1530 out:
1531         free(old_ds_name);
1532         free(new_ds_name);
1533         return ret;
1534
1535 }
1536
1537 __export int osl_update_object(struct osl_table *t, const struct osl_row *r,
1538                 unsigned col_num, struct osl_object *obj)
1539 {
1540         struct osl_column *col;
1541         const struct osl_column_description *cd;
1542         int ret;
1543
1544         if (!t)
1545                 return -E_OSL_BAD_TABLE;
1546         col = &t->columns[col_num];
1547         cd = get_column_description(t->desc, col_num);
1548         DEBUG_LOG("updating column %u of %s\n", col_num, t->desc->name);
1549         if (cd->storage_flags & OSL_RBTREE) {
1550                 if (search_rbtree(obj, t, col_num, NULL, NULL) > 0)
1551                         return -E_OSL_RB_KEY_EXISTS;
1552         }
1553         if (cd->storage_flags & OSL_FIXED_SIZE) {
1554                 if (obj->size != cd->data_size)
1555                         return -E_OSL_BAD_DATA_SIZE;
1556         }
1557         remove_rb_node(t, col_num, r);
1558         if (cd->storage_type == OSL_NO_STORAGE) { /* TODO: If fixed size, reuse object? */
1559                 free(r->volatile_objects[col->volatile_num].data);
1560                 r->volatile_objects[col->volatile_num] = *obj;
1561         } else if (cd->storage_type == OSL_DISK_STORAGE) {
1562                 char *ds_name;
1563                 ret = disk_storage_name_of_row(t, r, &ds_name);
1564                 if (ret < 0)
1565                         return ret;
1566                 ret = delete_disk_storage_file(t, col_num, ds_name);
1567                 if (ret < 0 && !is_errno(-ret, ENOENT)) {
1568                         free(ds_name);
1569                         return ret;
1570                 }
1571                 ret = write_disk_storage_file(t, col_num, obj, ds_name);
1572                 free(ds_name);
1573                 if (ret < 0)
1574                         return ret;
1575         } else { /* mapped storage */
1576                 struct osl_object old_obj;
1577                 ret = get_mapped_object(t, col_num, r->num, &old_obj);
1578                 if (ret < 0)
1579                         return ret;
1580                 /*
1581                  * If the updated column is the disk storage name column, the
1582                  * disk storage name changes, so we have to rename all disk
1583                  * storage objects accordingly.
1584                  */
1585                 if (col_num == t->disk_storage_name_column) {
1586                         ret = rename_disk_storage_objects(t, &old_obj, obj);
1587                         if (ret < 0)
1588                                 return ret;
1589                 }
1590                 if (cd->storage_flags & OSL_FIXED_SIZE)
1591                         memcpy(old_obj.data, obj->data, cd->data_size);
1592                 else { /* TODO: if the size doesn't change, use old space */
1593                         uint32_t new_data_map_size;
1594                         char *row_index;
1595                         ret = get_row_index(t, r->num, &row_index);
1596                         if (ret < 0)
1597                                 return ret;
1598                         unmap_column(t, col_num);
1599                         ret = append_map_file(t, col_num, obj,
1600                                 &new_data_map_size);
1601                         if (ret < 0)
1602                                 return ret;
1603                         ret = map_column(t, col_num);
1604                         if (ret < 0)
1605                                 return ret;
1606                         update_cell_index(row_index, col, new_data_map_size,
1607                                 obj->size);
1608                 }
1609         }
1610         if (cd->storage_flags & OSL_RBTREE) {
1611                 ret = insert_rbtree(t, col_num, r, obj);
1612                 if (ret < 0)
1613                         return ret;
1614         }
1615         return 1;
1616 }
1617
1618 __export int osl_open_disk_object(const struct osl_table *t, const struct osl_row *r,
1619                 unsigned col_num, struct osl_object *obj)
1620 {
1621         const struct osl_column_description *cd;
1622         char *ds_name, *filename;
1623         int ret;
1624
1625         if (!t)
1626                 return -E_OSL_BAD_TABLE;
1627         cd = get_column_description(t->desc, col_num);
1628         if (cd->storage_type != OSL_DISK_STORAGE)
1629                 return -E_OSL_BAD_STORAGE_TYPE;
1630
1631         ret = disk_storage_name_of_row(t, r, &ds_name);
1632         if (ret < 0)
1633                 return ret;
1634         filename = disk_storage_path(t, col_num, ds_name);
1635         free(ds_name);
1636         if (!filename)
1637                 return -ERRNO_TO_ERROR(ENOMEM);
1638         DEBUG_LOG("filename: %s\n", filename);
1639         ret = mmap_full_file(filename, O_RDONLY, &obj->data, &obj->size, NULL);
1640         free(filename);
1641         return ret;
1642 }
1643
1644 __export int osl_close_disk_object(struct osl_object *obj)
1645 {
1646         return osl_munmap(obj->data, obj->size);
1647 }
1648
1649 __export int osl_get_num_rows(const struct osl_table *t, unsigned *num_rows)
1650 {
1651         if (!t)
1652                 return -E_OSL_BAD_TABLE;
1653         assert(t->num_rows >= t->num_invalid_rows);
1654         *num_rows = t->num_rows - t->num_invalid_rows;
1655         return 1;
1656 }
1657
1658 __export int osl_get_rank(const struct osl_table *t, struct osl_row *r,
1659                 unsigned col_num, unsigned *rank)
1660 {
1661         struct osl_object obj;
1662         struct osl_column *col;
1663         struct rb_node *node;
1664         int ret = check_rbtree_col(t, col_num, &col);
1665
1666         if (ret < 0)
1667                 return ret;
1668         ret = osl_get_object(t, r, col_num, &obj);
1669         if (ret < 0)
1670                 return ret;
1671         ret = search_rbtree(&obj, t, col_num, &node, NULL);
1672         if (ret < 0)
1673                 return ret;
1674         ret = rb_rank(node, rank);
1675         if (ret < 0)
1676                 return -E_OSL_BAD_ROW;
1677         return 1;
1678 }
1679
1680 __export int osl_get_nth_row(const struct osl_table *t, unsigned col_num,
1681                 unsigned n, struct osl_row **result)
1682 {
1683         struct osl_column *col;
1684         struct rb_node *node;
1685         unsigned num_rows;
1686         int ret;
1687
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 }