fs/logfs/segment.c

   1 /*
   2  * fs/logfs/segment.c   - Handling the Object Store
   3  *
   4  * As should be obvious for Linux kernel code, license is GPLv2
   5  *
   6  * Copyright (c) 2005-2008 Joern Engel <joern@logfs.org>
   7  *
   8  * Object store or ostore makes up the complete device with exception of
   9  * the superblock and journal areas.  Apart from its own metadata it stores
  10  * three kinds of objects: inodes, dentries and blocks, both data and indirect.
  11  */
  12 #include "logfs.h"
  13
  14 static int logfs_mark_segment_bad(struct super_block *sb, u32 segno)
  15 {
  16         struct logfs_super *super = logfs_super(sb);
  17         struct btree_head32 *head = &super->s_reserved_segments;
  18         int err;
  19
  20         err = btree_insert32(head, segno, (void *)1, GFP_NOFS);
  21         if (err)
  22                 return err;
  23         logfs_super(sb)->s_bad_segments++;
  24         /* FIXME: write to journal */
  25         return 0;
  26 }
  27
  28 int logfs_erase_segment(struct super_block *sb, u32 segno, int ensure_erase)
  29 {
  30         struct logfs_super *super = logfs_super(sb);
  31
  32         super->s_gec++;
  33
  34         return super->s_devops->erase(sb, (u64)segno << super->s_segshift,
  35                         super->s_segsize, ensure_erase);
  36 }
  37
  38 static s64 logfs_get_free_bytes(struct logfs_area *area, size_t bytes)
  39 {
  40         s32 ofs;
  41
  42         logfs_open_area(area, bytes);
  43
  44         ofs = area->a_used_bytes;
  45         area->a_used_bytes += bytes;
  46         BUG_ON(area->a_used_bytes >= logfs_super(area->a_sb)->s_segsize);
  47
  48         return dev_ofs(area->a_sb, area->a_segno, ofs);
  49 }
  50
  51 static struct page *get_mapping_page(struct super_block *sb, pgoff_t index,
  52                 int use_filler)
  53 {
  54         struct logfs_super *super = logfs_super(sb);
  55         struct address_space *mapping = super->s_mapping_inode->i_mapping;
  56         filler_t *filler = super->s_devops->readpage;
  57         struct page *page;
  58
  59         BUG_ON(mapping_gfp_mask(mapping) & __GFP_FS);
  60         if (use_filler)
  61                 page = read_cache_page(mapping, index, filler, sb);
  62         else {
  63                 page = find_or_create_page(mapping, index, GFP_NOFS);
  64                 unlock_page(page);
  65         }
  66         return page;
  67 }
  68
  69 void __logfs_buf_write(struct logfs_area *area, u64 ofs, void *buf, size_t len,
  70                 int use_filler)
  71 {
  72         pgoff_t index = ofs >> PAGE_SHIFT;
  73         struct page *page;
  74         long offset = ofs & (PAGE_SIZE-1);
  75         long copylen;
  76
  77         /* Only logfs_wbuf_recover may use len==0 */
  78         BUG_ON(!len && !use_filler);
  79         do {
  80                 copylen = min((ulong)len, PAGE_SIZE - offset);
  81
  82                 page = get_mapping_page(area->a_sb, index, use_filler);
  83                 SetPageUptodate(page);
  84                 BUG_ON(!page); /* FIXME: reserve a pool */
  85                 memcpy(page_address(page) + offset, buf, copylen);
  86                 SetPagePrivate(page);
  87                 page_cache_release(page);
  88
  89                 buf += copylen;
  90                 len -= copylen;
  91                 offset = 0;
  92                 index++;
  93         } while (len);
  94 }
  95
  96 static void pad_partial_page(struct logfs_area *area)
  97 {
  98         struct super_block *sb = area->a_sb;
  99         struct page *page;
 100         u64 ofs = dev_ofs(sb, area->a_segno, area->a_used_bytes);
 101         pgoff_t index = ofs >> PAGE_SHIFT;
 102         long offset = ofs & (PAGE_SIZE-1);
 103         u32 len = PAGE_SIZE - offset;
 104
 105         if (len % PAGE_SIZE) {
 106                 page = get_mapping_page(sb, index, 0);
 107                 BUG_ON(!page); /* FIXME: reserve a pool */
 108                 memset(page_address(page) + offset, 0xff, len);
 109                 SetPagePrivate(page);
 110                 page_cache_release(page);
 111         }
 112 }
 113
 114 static void pad_full_pages(struct logfs_area *area)
 115 {
 116         struct super_block *sb = area->a_sb;
 117         struct logfs_super *super = logfs_super(sb);
 118         u64 ofs = dev_ofs(sb, area->a_segno, area->a_used_bytes);
 119         u32 len = super->s_segsize - area->a_used_bytes;
 120         pgoff_t index = PAGE_CACHE_ALIGN(ofs) >> PAGE_CACHE_SHIFT;
 121         pgoff_t no_indizes = len >> PAGE_CACHE_SHIFT;
 122         struct page *page;
 123
 124         while (no_indizes) {
 125                 page = get_mapping_page(sb, index, 0);
 126                 BUG_ON(!page); /* FIXME: reserve a pool */
 127                 SetPageUptodate(page);
 128                 memset(page_address(page), 0xff, PAGE_CACHE_SIZE);
 129                 SetPagePrivate(page);
 130                 page_cache_release(page);
 131                 index++;
 132                 no_indizes--;
 133         }
 134 }
 135
 136 /*
 137  * bdev_writeseg will write full pages.  Memset the tail to prevent data leaks.
 138  * Also make sure we allocate (and memset) all pages for final writeout.
 139  */
 140 static void pad_wbuf(struct logfs_area *area, int final)
 141 {
 142         pad_partial_page(area);
 143         if (final)
 144                 pad_full_pages(area);
 145 }
 146
 147 /*
 148  * We have to be careful with the alias tree.  Since lookup is done by bix,
 149  * it needs to be normalized, so 14, 15, 16, etc. all match when dealing with
 150  * indirect blocks.  So always use it through accessor functions.
 151  */
 152 static void *alias_tree_lookup(struct super_block *sb, u64 ino, u64 bix,
 153                 level_t level)
 154 {
 155         struct btree_head128 *head = &logfs_super(sb)->s_object_alias_tree;
 156         pgoff_t index = logfs_pack_index(bix, level);
 157
 158         return btree_lookup128(head, ino, index);
 159 }
 160
 161 static int alias_tree_insert(struct super_block *sb, u64 ino, u64 bix,
 162                 level_t level, void *val)
 163 {
 164         struct btree_head128 *head = &logfs_super(sb)->s_object_alias_tree;
 165         pgoff_t index = logfs_pack_index(bix, level);
 166
 167         return btree_insert128(head, ino, index, val, GFP_NOFS);
 168 }
 169
 170 static int btree_write_alias(struct super_block *sb, struct logfs_block *block,
 171                 write_alias_t *write_one_alias)
 172 {
 173         struct object_alias_item *item;
 174         int err;
 175
 176         list_for_each_entry(item, &block->item_list, list) {
 177                 err = write_alias_journal(sb, block->ino, block->bix,
 178                                 block->level, item->child_no, item->val);
 179                 if (err)
 180                         return err;
 181         }
 182         return 0;
 183 }
 184
 185 static struct logfs_block_ops btree_block_ops = {
 186         .write_block    = btree_write_block,
 187         .free_block     = __free_block,
 188         .write_alias    = btree_write_alias,
 189 };
 190
 191 int logfs_load_object_aliases(struct super_block *sb,
 192                 struct logfs_obj_alias *oa, int count)
 193 {
 194         struct logfs_super *super = logfs_super(sb);
 195         struct logfs_block *block;
 196         struct object_alias_item *item;
 197         u64 ino, bix;
 198         level_t level;
 199         int i, err;
 200
 201         super->s_flags |= LOGFS_SB_FLAG_OBJ_ALIAS;
 202         count /= sizeof(*oa);
 203         for (i = 0; i < count; i++) {
 204                 item = mempool_alloc(super->s_alias_pool, GFP_NOFS);
 205                 if (!item)
 206                         return -ENOMEM;
 207                 memset(item, 0, sizeof(*item));
 208
 209                 super->s_no_object_aliases++;
 210                 item->val = oa[i].val;
 211                 item->child_no = be16_to_cpu(oa[i].child_no);
 212
 213                 ino = be64_to_cpu(oa[i].ino);
 214                 bix = be64_to_cpu(oa[i].bix);
 215                 level = LEVEL(oa[i].level);
 216
 217                 log_aliases("logfs_load_object_aliases(%llx, %llx, %x, %x) %llx\n",
 218                                 ino, bix, level, item->child_no,
 219                                 be64_to_cpu(item->val));
 220                 block = alias_tree_lookup(sb, ino, bix, level);
 221                 if (!block) {
 222                         block = __alloc_block(sb, ino, bix, level);
 223                         block->ops = &btree_block_ops;
 224                         err = alias_tree_insert(sb, ino, bix, level, block);
 225                         BUG_ON(err); /* mempool empty */
 226                 }
 227                 if (test_and_set_bit(item->child_no, block->alias_map)) {
 228                         printk(KERN_ERR"LogFS: Alias collision detected\n");
 229                         return -EIO;
 230                 }
 231                 list_move_tail(&block->alias_list, &super->s_object_alias);
 232                 list_add(&item->list, &block->item_list);
 233         }
 234         return 0;
 235 }
 236
 237 static void kill_alias(void *_block, unsigned long ignore0,
 238                 u64 ignore1, u64 ignore2, size_t ignore3)
 239 {
 240         struct logfs_block *block = _block;
 241         struct super_block *sb = block->sb;
 242         struct logfs_super *super = logfs_super(sb);
 243         struct object_alias_item *item;
 244
 245         while (!list_empty(&block->item_list)) {
 246                 item = list_entry(block->item_list.next, typeof(*item), list);
 247                 list_del(&item->list);
 248                 mempool_free(item, super->s_alias_pool);
 249         }
 250         block->ops->free_block(sb, block);
 251 }
 252
 253 static int obj_type(struct inode *inode, level_t level)
 254 {
 255         if (level == 0) {
 256                 if (S_ISDIR(inode->i_mode))
 257                         return OBJ_DENTRY;
 258                 if (inode->i_ino == LOGFS_INO_MASTER)
 259                         return OBJ_INODE;
 260         }
 261         return OBJ_BLOCK;
 262 }
 263
 264 static int obj_len(struct super_block *sb, int obj_type)
 265 {
 266         switch (obj_type) {
 267         case OBJ_DENTRY:
 268                 return sizeof(struct logfs_disk_dentry);
 269         case OBJ_INODE:
 270                 return sizeof(struct logfs_disk_inode);
 271         case OBJ_BLOCK:
 272                 return sb->s_blocksize;
 273         default:
 274                 BUG();
 275         }
 276 }
 277
 278 static int __logfs_segment_write(struct inode *inode, void *buf,
 279                 struct logfs_shadow *shadow, int type, int len, int compr)
 280 {
 281         struct logfs_area *area;
 282         struct super_block *sb = inode->i_sb;
 283         s64 ofs;
 284         struct logfs_object_header h;
 285         int acc_len;
 286
 287         if (shadow->gc_level == 0)
 288                 acc_len = len;
 289         else
 290                 acc_len = obj_len(sb, type);
 291
 292         area = get_area(sb, shadow->gc_level);
 293         ofs = logfs_get_free_bytes(area, len + LOGFS_OBJECT_HEADERSIZE);
 294         LOGFS_BUG_ON(ofs <= 0, sb);
 295         /*
 296          * Order is important.  logfs_get_free_bytes(), by modifying the
 297          * segment file, may modify the content of the very page we're about
 298          * to write now.  Which is fine, as long as the calculated crc and
 299          * written data still match.  So do the modifications _before_
 300          * calculating the crc.
 301          */
 302
 303         h.len   = cpu_to_be16(len);
 304         h.type  = type;
 305         h.compr = compr;
 306         h.ino   = cpu_to_be64(inode->i_ino);
 307         h.bix   = cpu_to_be64(shadow->bix);
 308         h.crc   = logfs_crc32(&h, sizeof(h) - 4, 4);
 309         h.data_crc = logfs_crc32(buf, len, 0);
 310
 311         logfs_buf_write(area, ofs, &h, sizeof(h));
 312         logfs_buf_write(area, ofs + LOGFS_OBJECT_HEADERSIZE, buf, len);
 313
 314         shadow->new_ofs = ofs;
 315         shadow->new_len = acc_len + LOGFS_OBJECT_HEADERSIZE;
 316
 317         return 0;
 318 }
 319
 320 static s64 logfs_segment_write_compress(struct inode *inode, void *buf,
 321                 struct logfs_shadow *shadow, int type, int len)
 322 {
 323         struct super_block *sb = inode->i_sb;
 324         void *compressor_buf = logfs_super(sb)->s_compressed_je;
 325         ssize_t compr_len;
 326         int ret;
 327
 328         mutex_lock(&logfs_super(sb)->s_journal_mutex);
 329         compr_len = logfs_compress(buf, compressor_buf, len, len);
 330
 331         if (compr_len >= 0) {
 332                 ret = __logfs_segment_write(inode, compressor_buf, shadow,
 333                                 type, compr_len, COMPR_ZLIB);
 334         } else {
 335                 ret = __logfs_segment_write(inode, buf, shadow, type, len,
 336                                 COMPR_NONE);
 337         }
 338         mutex_unlock(&logfs_super(sb)->s_journal_mutex);
 339         return ret;
 340 }
 341
 342 /**
 343  * logfs_segment_write - write data block to object store
 344  * @inode:              inode containing data
 345  *
 346  * Returns an errno or zero.
 347  */
 348 int logfs_segment_write(struct inode *inode, struct page *page,
 349                 struct logfs_shadow *shadow)
 350 {
 351         struct super_block *sb = inode->i_sb;
 352         struct logfs_super *super = logfs_super(sb);
 353         int do_compress, type, len;
 354         int ret;
 355         void *buf;
 356
 357         super->s_flags |= LOGFS_SB_FLAG_DIRTY;
 358         BUG_ON(super->s_flags & LOGFS_SB_FLAG_SHUTDOWN);
 359         do_compress = logfs_inode(inode)->li_flags & LOGFS_IF_COMPRESSED;
 360         if (shadow->gc_level != 0) {
 361                 /* temporarily disable compression for indirect blocks */
 362                 do_compress = 0;
 363         }
 364
 365         type = obj_type(inode, shrink_level(shadow->gc_level));
 366         len = obj_len(sb, type);
 367         buf = kmap(page);
 368         if (do_compress)
 369                 ret = logfs_segment_write_compress(inode, buf, shadow, type,
 370                                 len);
 371         else
 372                 ret = __logfs_segment_write(inode, buf, shadow, type, len,
 373                                 COMPR_NONE);
 374         kunmap(page);
 375
 376         log_segment("logfs_segment_write(%llx, %llx, %x) %llx->%llx %x->%x\n",
 377                         shadow->ino, shadow->bix, shadow->gc_level,
 378                         shadow->old_ofs, shadow->new_ofs,
 379                         shadow->old_len, shadow->new_len);
 380         /* this BUG_ON did catch a locking bug.  useful */
 381         BUG_ON(!(shadow->new_ofs & (super->s_segsize - 1)));
 382         return ret;
 383 }
 384
 385 int wbuf_read(struct super_block *sb, u64 ofs, size_t len, void *buf)
 386 {
 387         pgoff_t index = ofs >> PAGE_SHIFT;
 388         struct page *page;
 389         long offset = ofs & (PAGE_SIZE-1);
 390         long copylen;
 391
 392         while (len) {
 393                 copylen = min((ulong)len, PAGE_SIZE - offset);
 394
 395                 page = get_mapping_page(sb, index, 1);
 396                 if (IS_ERR(page))
 397                         return PTR_ERR(page);
 398                 memcpy(buf, page_address(page) + offset, copylen);
 399                 page_cache_release(page);
 400
 401                 buf += copylen;
 402                 len -= copylen;
 403                 offset = 0;
 404                 index++;
 405         }
 406         return 0;
 407 }
 408
 409 /*
 410  * The "position" of indirect blocks is ambiguous.  It can be the position
 411  * of any data block somewhere behind this indirect block.  So we need to
 412  * normalize the positions through logfs_block_mask() before comparing.
 413  */
 414 static int check_pos(struct super_block *sb, u64 pos1, u64 pos2, level_t level)
 415 {
 416         return  (pos1 & logfs_block_mask(sb, level)) !=
 417                 (pos2 & logfs_block_mask(sb, level));
 418 }
 419
 420 #if 0
 421 static int read_seg_header(struct super_block *sb, u64 ofs,
 422                 struct logfs_segment_header *sh)
 423 {
 424         __be32 crc;
 425         int err;
 426
 427         err = wbuf_read(sb, ofs, sizeof(*sh), sh);
 428         if (err)
 429                 return err;
 430         crc = logfs_crc32(sh, sizeof(*sh), 4);
 431         if (crc != sh->crc) {
 432                 printk(KERN_ERR"LOGFS: header crc error at %llx: expected %x, "
 433                                 "got %x\n", ofs, be32_to_cpu(sh->crc),
 434                                 be32_to_cpu(crc));
 435                 return -EIO;
 436         }
 437         return 0;
 438 }
 439 #endif
 440
 441 static int read_obj_header(struct super_block *sb, u64 ofs,
 442                 struct logfs_object_header *oh)
 443 {
 444         __be32 crc;
 445         int err;
 446
 447         err = wbuf_read(sb, ofs, sizeof(*oh), oh);
 448         if (err)
 449                 return err;
 450         crc = logfs_crc32(oh, sizeof(*oh) - 4, 4);
 451         if (crc != oh->crc) {
 452                 printk(KERN_ERR"LOGFS: header crc error at %llx: expected %x, "
 453                                 "got %x\n", ofs, be32_to_cpu(oh->crc),
 454                                 be32_to_cpu(crc));
 455                 return -EIO;
 456         }
 457         return 0;
 458 }
 459
 460 static void move_btree_to_page(struct inode *inode, struct page *page,
 461                 __be64 *data)
 462 {
 463         struct super_block *sb = inode->i_sb;
 464         struct logfs_super *super = logfs_super(sb);
 465         struct btree_head128 *head = &super->s_object_alias_tree;
 466         struct logfs_block *block;
 467         struct object_alias_item *item, *next;
 468
 469         if (!(super->s_flags & LOGFS_SB_FLAG_OBJ_ALIAS))
 470                 return;
 471
 472         block = btree_remove128(head, inode->i_ino, page->index);
 473         if (!block)
 474                 return;
 475
 476         log_blockmove("move_btree_to_page(%llx, %llx, %x)\n",
 477                         block->ino, block->bix, block->level);
 478         list_for_each_entry_safe(item, next, &block->item_list, list) {
 479                 data[item->child_no] = item->val;
 480                 list_del(&item->list);
 481                 mempool_free(item, super->s_alias_pool);
 482         }
 483         block->page = page;
 484         SetPagePrivate(page);
 485         page->private = (unsigned long)block;
 486         block->ops = &indirect_block_ops;
 487         initialize_block_counters(page, block, data, 0);
 488 }
 489
 490 /*
 491  * This silences a false, yet annoying gcc warning.  I hate it when my editor
 492  * jumps into bitops.h each time I recompile this file.
 493  * TODO: Complain to gcc folks about this and upgrade compiler.
 494  */
 495 static unsigned long fnb(const unsigned long *addr,
 496                 unsigned long size, unsigned long offset)
 497 {
 498         return find_next_bit(addr, size, offset);
 499 }
 500
 501 void move_page_to_btree(struct page *page)
 502 {
 503         struct logfs_block *block = logfs_block(page);
 504         struct super_block *sb = block->sb;
 505         struct logfs_super *super = logfs_super(sb);
 506         struct object_alias_item *item;
 507         unsigned long pos;
 508         __be64 *child;
 509         int err;
 510
 511         if (super->s_flags & LOGFS_SB_FLAG_SHUTDOWN) {
 512                 block->ops->free_block(sb, block);
 513                 return;
 514         }
 515         log_blockmove("move_page_to_btree(%llx, %llx, %x)\n",
 516                         block->ino, block->bix, block->level);
 517         super->s_flags |= LOGFS_SB_FLAG_OBJ_ALIAS;
 518
 519         for (pos = 0; ; pos++) {
 520                 pos = fnb(block->alias_map, LOGFS_BLOCK_FACTOR, pos);
 521                 if (pos >= LOGFS_BLOCK_FACTOR)
 522                         break;
 523
 524                 item = mempool_alloc(super->s_alias_pool, GFP_NOFS);
 525                 BUG_ON(!item); /* mempool empty */
 526                 memset(item, 0, sizeof(*item));
 527
 528                 child = kmap_atomic(page, KM_USER0);
 529                 item->val = child[pos];
 530                 kunmap_atomic(child, KM_USER0);
 531                 item->child_no = pos;
 532                 list_add(&item->list, &block->item_list);
 533         }
 534         block->page = NULL;
 535         ClearPagePrivate(page);
 536         page->private = 0;
 537         block->ops = &btree_block_ops;
 538         err = alias_tree_insert(block->sb, block->ino, block->bix, block->level,
 539                         block);
 540         BUG_ON(err); /* mempool empty */
 541         ClearPageUptodate(page);
 542 }
 543
 544 static int __logfs_segment_read(struct inode *inode, void *buf,
 545                 u64 ofs, u64 bix, level_t level)
 546 {
 547         struct super_block *sb = inode->i_sb;
 548         void *compressor_buf = logfs_super(sb)->s_compressed_je;
 549         struct logfs_object_header oh;
 550         __be32 crc;
 551         u16 len;
 552         int err, block_len;
 553
 554         block_len = obj_len(sb, obj_type(inode, level));
 555         err = read_obj_header(sb, ofs, &oh);
 556         if (err)
 557                 goto out_err;
 558
 559         err = -EIO;
 560         if (be64_to_cpu(oh.ino) != inode->i_ino
 561                         || check_pos(sb, be64_to_cpu(oh.bix), bix, level)) {
 562                 printk(KERN_ERR"LOGFS: (ino, bix) don't match at %llx: "
 563                                 "expected (%lx, %llx), got (%llx, %llx)\n",
 564                                 ofs, inode->i_ino, bix,
 565                                 be64_to_cpu(oh.ino), be64_to_cpu(oh.bix));
 566                 goto out_err;
 567         }
 568
 569         len = be16_to_cpu(oh.len);
 570
 571         switch (oh.compr) {
 572         case COMPR_NONE:
 573                 err = wbuf_read(sb, ofs + LOGFS_OBJECT_HEADERSIZE, len, buf);
 574                 if (err)
 575                         goto out_err;
 576                 crc = logfs_crc32(buf, len, 0);
 577                 if (crc != oh.data_crc) {
 578                         printk(KERN_ERR"LOGFS: uncompressed data crc error at "
 579                                         "%llx: expected %x, got %x\n", ofs,
 580                                         be32_to_cpu(oh.data_crc),
 581                                         be32_to_cpu(crc));
 582                         goto out_err;
 583                 }
 584                 break;
 585         case COMPR_ZLIB:
 586                 mutex_lock(&logfs_super(sb)->s_journal_mutex);
 587                 err = wbuf_read(sb, ofs + LOGFS_OBJECT_HEADERSIZE, len,
 588                                 compressor_buf);
 589                 if (err) {
 590                         mutex_unlock(&logfs_super(sb)->s_journal_mutex);
 591                         goto out_err;
 592                 }
 593                 crc = logfs_crc32(compressor_buf, len, 0);
 594                 if (crc != oh.data_crc) {
 595                         printk(KERN_ERR"LOGFS: compressed data crc error at "
 596                                         "%llx: expected %x, got %x\n", ofs,
 597                                         be32_to_cpu(oh.data_crc),
 598                                         be32_to_cpu(crc));
 599                         mutex_unlock(&logfs_super(sb)->s_journal_mutex);
 600                         goto out_err;
 601                 }
 602                 err = logfs_uncompress(compressor_buf, buf, len, block_len);
 603                 mutex_unlock(&logfs_super(sb)->s_journal_mutex);
 604                 if (err) {
 605                         printk(KERN_ERR"LOGFS: uncompress error at %llx\n", ofs);
 606                         goto out_err;
 607                 }
 608                 break;
 609         default:
 610                 LOGFS_BUG(sb);
 611                 err = -EIO;
 612                 goto out_err;
 613         }
 614         return 0;
 615
 616 out_err:
 617         logfs_set_ro(sb);
 618         printk(KERN_ERR"LOGFS: device is read-only now\n");
 619         LOGFS_BUG(sb);
 620         return err;
 621 }
 622
 623 /**
 624  * logfs_segment_read - read data block from object store
 625  * @inode:              inode containing data
 626  * @buf:                data buffer
 627  * @ofs:                physical data offset
 628  * @bix:                block index
 629  * @level:              block level
 630  *
 631  * Returns 0 on success or a negative errno.
 632  */
 633 int logfs_segment_read(struct inode *inode, struct page *page,
 634                 u64 ofs, u64 bix, level_t level)
 635 {
 636         int err;
 637         void *buf;
 638
 639         if (PageUptodate(page))
 640                 return 0;
 641
 642         ofs &= ~LOGFS_FULLY_POPULATED;
 643
 644         buf = kmap(page);
 645         err = __logfs_segment_read(inode, buf, ofs, bix, level);
 646         if (!err) {
 647                 move_btree_to_page(inode, page, buf);
 648                 SetPageUptodate(page);
 649         }
 650         kunmap(page);
 651         log_segment("logfs_segment_read(%lx, %llx, %x) %llx (%d)\n",
 652                         inode->i_ino, bix, level, ofs, err);
 653         return err;
 654 }
 655
 656 int logfs_segment_delete(struct inode *inode, struct logfs_shadow *shadow)
 657 {
 658         struct super_block *sb = inode->i_sb;
 659         struct logfs_super *super = logfs_super(sb);
 660         struct logfs_object_header h;
 661         u16 len;
 662         int err;
 663
 664         super->s_flags |= LOGFS_SB_FLAG_DIRTY;
 665         BUG_ON(super->s_flags & LOGFS_SB_FLAG_SHUTDOWN);
 666         BUG_ON(shadow->old_ofs & LOGFS_FULLY_POPULATED);
 667         if (!shadow->old_ofs)
 668                 return 0;
 669
 670         log_segment("logfs_segment_delete(%llx, %llx, %x) %llx->%llx %x->%x\n",
 671                         shadow->ino, shadow->bix, shadow->gc_level,
 672                         shadow->old_ofs, shadow->new_ofs,
 673                         shadow->old_len, shadow->new_len);
 674         err = read_obj_header(sb, shadow->old_ofs, &h);
 675         LOGFS_BUG_ON(err, sb);
 676         LOGFS_BUG_ON(be64_to_cpu(h.ino) != inode->i_ino, sb);
 677         LOGFS_BUG_ON(check_pos(sb, shadow->bix, be64_to_cpu(h.bix),
 678                                 shrink_level(shadow->gc_level)), sb);
 679
 680         if (shadow->gc_level == 0)
 681                 len = be16_to_cpu(h.len);
 682         else
 683                 len = obj_len(sb, h.type);
 684         shadow->old_len = len + sizeof(h);
 685         return 0;
 686 }
 687
 688 void freeseg(struct super_block *sb, u32 segno)
 689 {
 690         struct logfs_super *super = logfs_super(sb);
 691         struct address_space *mapping = super->s_mapping_inode->i_mapping;
 692         struct page *page;
 693         u64 ofs, start, end;
 694
 695         start = dev_ofs(sb, segno, 0);
 696         end = dev_ofs(sb, segno + 1, 0);
 697         for (ofs = start; ofs < end; ofs += PAGE_SIZE) {
 698                 page = find_get_page(mapping, ofs >> PAGE_SHIFT);
 699                 if (!page)
 700                         continue;
 701                 ClearPagePrivate(page);
 702                 page_cache_release(page);
 703         }
 704 }
 705
 706 int logfs_open_area(struct logfs_area *area, size_t bytes)
 707 {
 708         struct super_block *sb = area->a_sb;
 709         struct logfs_super *super = logfs_super(sb);
 710         int err, closed = 0;
 711
 712         if (area->a_is_open && area->a_used_bytes + bytes <= super->s_segsize)
 713                 return 0;
 714
 715         if (area->a_is_open) {
 716                 u64 ofs = dev_ofs(sb, area->a_segno, area->a_written_bytes);
 717                 u32 len = super->s_segsize - area->a_written_bytes;
 718
 719                 log_gc("logfs_close_area(%x)\n", area->a_segno);
 720                 pad_wbuf(area, 1);
 721                 super->s_devops->writeseg(area->a_sb, ofs, len);
 722                 freeseg(sb, area->a_segno);
 723                 closed = 1;
 724         }
 725
 726         area->a_used_bytes = 0;
 727         area->a_written_bytes = 0;
 728 again:
 729         area->a_ops->get_free_segment(area);
 730         area->a_ops->get_erase_count(area);
 731
 732         log_gc("logfs_open_area(%x, %x)\n", area->a_segno, area->a_level);
 733         err = area->a_ops->erase_segment(area);
 734         if (err) {
 735                 printk(KERN_WARNING "LogFS: Error erasing segment %x\n",
 736                                 area->a_segno);
 737                 logfs_mark_segment_bad(sb, area->a_segno);
 738                 goto again;
 739         }
 740         area->a_is_open = 1;
 741         return closed;
 742 }
 743
 744 void logfs_sync_area(struct logfs_area *area)
 745 {
 746         struct super_block *sb = area->a_sb;
 747         struct logfs_super *super = logfs_super(sb);
 748         u64 ofs = dev_ofs(sb, area->a_segno, area->a_written_bytes);
 749         u32 len = (area->a_used_bytes - area->a_written_bytes);
 750
 751         if (super->s_writesize)
 752                 len &= ~(super->s_writesize - 1);
 753         if (len == 0)
 754                 return;
 755         pad_wbuf(area, 0);
 756         super->s_devops->writeseg(sb, ofs, len);
 757         area->a_written_bytes += len;
 758 }
 759
 760 void logfs_sync_segments(struct super_block *sb)
 761 {
 762         struct logfs_super *super = logfs_super(sb);
 763         int i;
 764
 765         for_each_area(i)
 766                 logfs_sync_area(super->s_area[i]);
 767 }
 768
 769 /*
 770  * Pick a free segment to be used for this area.  Effectively takes a
 771  * candidate from the free list (not really a candidate anymore).
 772  */
 773 static void ostore_get_free_segment(struct logfs_area *area)
 774 {
 775         struct super_block *sb = area->a_sb;
 776         struct logfs_super *super = logfs_super(sb);
 777
 778         if (super->s_free_list.count == 0) {
 779                 printk(KERN_ERR"LOGFS: ran out of free segments\n");
 780                 LOGFS_BUG(sb);
 781         }
 782
 783         area->a_segno = get_best_cand(sb, &super->s_free_list, NULL);
 784 }
 785
 786 static void ostore_get_erase_count(struct logfs_area *area)
 787 {
 788         struct logfs_segment_entry se;
 789         u32 ec_level;
 790
 791         logfs_get_segment_entry(area->a_sb, area->a_segno, &se);
 792         BUG_ON(se.ec_level == cpu_to_be32(BADSEG) ||
 793                         se.valid == cpu_to_be32(RESERVED));
 794
 795         ec_level = be32_to_cpu(se.ec_level);
 796         area->a_erase_count = (ec_level >> 4) + 1;
 797 }
 798
 799 static int ostore_erase_segment(struct logfs_area *area)
 800 {
 801         struct super_block *sb = area->a_sb;
 802         struct logfs_segment_header sh;
 803         u64 ofs;
 804         int err;
 805
 806         err = logfs_erase_segment(sb, area->a_segno, 0);
 807         if (err)
 808                 return err;
 809
 810         sh.pad = 0;
 811         sh.type = SEG_OSTORE;
 812         sh.level = (__force u8)area->a_level;
 813         sh.segno = cpu_to_be32(area->a_segno);
 814         sh.ec = cpu_to_be32(area->a_erase_count);
 815         sh.gec = cpu_to_be64(logfs_super(sb)->s_gec);
 816         sh.crc = logfs_crc32(&sh, sizeof(sh), 4);
 817
 818         logfs_set_segment_erased(sb, area->a_segno, area->a_erase_count,
 819                         area->a_level);
 820
 821         ofs = dev_ofs(sb, area->a_segno, 0);
 822         area->a_used_bytes = sizeof(sh);
 823         logfs_buf_write(area, ofs, &sh, sizeof(sh));
 824         return 0;
 825 }
 826
 827 static const struct logfs_area_ops ostore_area_ops = {
 828         .get_free_segment       = ostore_get_free_segment,
 829         .get_erase_count        = ostore_get_erase_count,
 830         .erase_segment          = ostore_erase_segment,
 831 };
 832
 833 static void free_area(struct logfs_area *area)
 834 {
 835         if (area)
 836                 freeseg(area->a_sb, area->a_segno);
 837         kfree(area);
 838 }
 839
 840 static struct logfs_area *alloc_area(struct super_block *sb)
 841 {
 842         struct logfs_area *area;
 843
 844         area = kzalloc(sizeof(*area), GFP_KERNEL);
 845         if (!area)
 846                 return NULL;
 847
 848         area->a_sb = sb;
 849         return area;
 850 }
 851
 852 static void map_invalidatepage(struct page *page, unsigned long l)
 853 {
 854         BUG();
 855 }
 856
 857 static int map_releasepage(struct page *page, gfp_t g)
 858 {
 859         /* Don't release these pages */
 860         return 0;
 861 }
 862
 863 static const struct address_space_operations mapping_aops = {
 864         .invalidatepage = map_invalidatepage,
 865         .releasepage    = map_releasepage,
 866         .set_page_dirty = __set_page_dirty_nobuffers,
 867 };
 868
 869 int logfs_init_mapping(struct super_block *sb)
 870 {
 871         struct logfs_super *super = logfs_super(sb);
 872         struct address_space *mapping;
 873         struct inode *inode;
 874
 875         inode = logfs_new_meta_inode(sb, LOGFS_INO_MAPPING);
 876         if (IS_ERR(inode))
 877                 return PTR_ERR(inode);
 878         super->s_mapping_inode = inode;
 879         mapping = inode->i_mapping;
 880         mapping->a_ops = &mapping_aops;
 881         /* Would it be possible to use __GFP_HIGHMEM as well? */
 882         mapping_set_gfp_mask(mapping, GFP_NOFS);
 883         return 0;
 884 }
 885
 886 int logfs_init_areas(struct super_block *sb)
 887 {
 888         struct logfs_super *super = logfs_super(sb);
 889         int i = -1;
 890
 891         super->s_alias_pool = mempool_create_kmalloc_pool(600,
 892                         sizeof(struct object_alias_item));
 893         if (!super->s_alias_pool)
 894                 return -ENOMEM;
 895
 896         super->s_journal_area = alloc_area(sb);
 897         if (!super->s_journal_area)
 898                 goto err;
 899
 900         for_each_area(i) {
 901                 super->s_area[i] = alloc_area(sb);
 902                 if (!super->s_area[i])
 903                         goto err;
 904                 super->s_area[i]->a_level = GC_LEVEL(i);
 905                 super->s_area[i]->a_ops = &ostore_area_ops;
 906         }
 907         btree_init_mempool128(&super->s_object_alias_tree,
 908                         super->s_btree_pool);
 909         return 0;
 910
 911 err:
 912         for (i--; i >= 0; i--)
 913                 free_area(super->s_area[i]);
 914         free_area(super->s_journal_area);
 915         logfs_mempool_destroy(super->s_alias_pool);
 916         return -ENOMEM;
 917 }
 918
 919 void logfs_cleanup_areas(struct super_block *sb)
 920 {
 921         struct logfs_super *super = logfs_super(sb);
 922         int i;
 923
 924         btree_grim_visitor128(&super->s_object_alias_tree, 0, kill_alias);
 925         for_each_area(i)
 926                 free_area(super->s_area[i]);
 927         free_area(super->s_journal_area);
 928         destroy_meta_inode(super->s_mapping_inode);
 929 }