fs/sysv/itree.c

   1 /*
   2  *  linux/fs/sysv/itree.c
   3  *
   4  *  Handling of indirect blocks' trees.
   5  *  AV, Sep--Dec 2000
   6  */
   7
   8 #include <linux/buffer_head.h>
   9 #include <linux/mount.h>
  10 #include <linux/string.h>
  11 #include "sysv.h"
  12
  13 enum {DIRECT = 10, DEPTH = 4};  /* Have triple indirect */
  14
  15 static inline void dirty_indirect(struct buffer_head *bh, struct inode *inode)
  16 {
  17         mark_buffer_dirty_inode(bh, inode);
  18         if (IS_SYNC(inode))
  19                 sync_dirty_buffer(bh);
  20 }
  21
  22 static int block_to_path(struct inode *inode, long block, int offsets[DEPTH])
  23 {
  24         struct super_block *sb = inode->i_sb;
  25         struct sysv_sb_info *sbi = SYSV_SB(sb);
  26         int ptrs_bits = sbi->s_ind_per_block_bits;
  27         unsigned long   indirect_blocks = sbi->s_ind_per_block,
  28                         double_blocks = sbi->s_ind_per_block_2;
  29         int n = 0;
  30
  31         if (block < 0) {
  32                 printk("sysv_block_map: block < 0\n");
  33         } else if (block < DIRECT) {
  34                 offsets[n++] = block;
  35         } else if ( (block -= DIRECT) < indirect_blocks) {
  36                 offsets[n++] = DIRECT;
  37                 offsets[n++] = block;
  38         } else if ((block -= indirect_blocks) < double_blocks) {
  39                 offsets[n++] = DIRECT+1;
  40                 offsets[n++] = block >> ptrs_bits;
  41                 offsets[n++] = block & (indirect_blocks - 1);
  42         } else if (((block -= double_blocks) >> (ptrs_bits * 2)) < indirect_blocks) {
  43                 offsets[n++] = DIRECT+2;
  44                 offsets[n++] = block >> (ptrs_bits * 2);
  45                 offsets[n++] = (block >> ptrs_bits) & (indirect_blocks - 1);
  46                 offsets[n++] = block & (indirect_blocks - 1);
  47         } else {
  48                 /* nothing */;
  49         }
  50         return n;
  51 }
  52
  53 static inline int block_to_cpu(struct sysv_sb_info *sbi, u32 nr)
  54 {
  55         return sbi->s_block_base + fs32_to_cpu(sbi, nr);
  56 }
  57
  58 typedef struct {
  59         u32     *p;
  60         u32     key;
  61         struct buffer_head *bh;
  62 } Indirect;
  63
  64 static rwlock_t pointers_lock = RW_LOCK_UNLOCKED;
  65
  66 static inline void add_chain(Indirect *p, struct buffer_head *bh, u32 *v)
  67 {
  68         p->key = *(p->p = v);
  69         p->bh = bh;
  70 }
  71
  72 static inline int verify_chain(Indirect *from, Indirect *to)
  73 {
  74         while (from <= to && from->key == *from->p)
  75                 from++;
  76         return (from > to);
  77 }
  78
  79 static inline u32 *block_end(struct buffer_head *bh)
  80 {
  81         return (u32*)((char*)bh->b_data + bh->b_size);
  82 }
  83
  84 static Indirect *get_branch(struct inode *inode,
  85                             int depth,
  86                             int offsets[],
  87                             Indirect chain[],
  88                             int *err)
  89 {
  90         struct super_block *sb = inode->i_sb;
  91         Indirect *p = chain;
  92         struct buffer_head *bh;
  93
  94         *err = 0;
  95         add_chain(chain, NULL, SYSV_I(inode)->i_data + *offsets);
  96         if (!p->key)
  97                 goto no_block;
  98         while (--depth) {
  99                 int block = block_to_cpu(SYSV_SB(sb), p->key);
 100                 bh = sb_bread(sb, block);
 101                 if (!bh)
 102                         goto failure;
 103                 read_lock(&pointers_lock);
 104                 if (!verify_chain(chain, p))
 105                         goto changed;
 106                 add_chain(++p, bh, (u32*)bh->b_data + *++offsets);
 107                 read_unlock(&pointers_lock);
 108                 if (!p->key)
 109                         goto no_block;
 110         }
 111         return NULL;
 112
 113 changed:
 114         read_unlock(&pointers_lock);
 115         brelse(bh);
 116         *err = -EAGAIN;
 117         goto no_block;
 118 failure:
 119         *err = -EIO;
 120 no_block:
 121         return p;
 122 }
 123
 124 static int alloc_branch(struct inode *inode,
 125                         int num,
 126                         int *offsets,
 127                         Indirect *branch)
 128 {
 129         int blocksize = inode->i_sb->s_blocksize;
 130         int n = 0;
 131         int i;
 132
 133         branch[0].key = sysv_new_block(inode->i_sb);
 134         if (branch[0].key) for (n = 1; n < num; n++) {
 135                 struct buffer_head *bh;
 136                 int parent;
 137                 /* Allocate the next block */
 138                 branch[n].key = sysv_new_block(inode->i_sb);
 139                 if (!branch[n].key)
 140                         break;
 141                 /*
 142                  * Get buffer_head for parent block, zero it out and set
 143                  * the pointer to new one, then send parent to disk.
 144                  */
 145                 parent = block_to_cpu(SYSV_SB(inode->i_sb), branch[n-1].key);
 146                 bh = sb_getblk(inode->i_sb, parent);
 147                 lock_buffer(bh);
 148                 memset(bh->b_data, 0, blocksize);
 149                 branch[n].bh = bh;
 150                 branch[n].p = (u32*) bh->b_data + offsets[n];
 151                 *branch[n].p = branch[n].key;
 152                 set_buffer_uptodate(bh);
 153                 unlock_buffer(bh);
 154                 dirty_indirect(bh, inode);
 155         }
 156         if (n == num)
 157                 return 0;
 158
 159         /* Allocation failed, free what we already allocated */
 160         for (i = 1; i < n; i++)
 161                 bforget(branch[i].bh);
 162         for (i = 0; i < n; i++)
 163                 sysv_free_block(inode->i_sb, branch[i].key);
 164         return -ENOSPC;
 165 }
 166
 167 static inline int splice_branch(struct inode *inode,
 168                                 Indirect chain[],
 169                                 Indirect *where,
 170                                 int num)
 171 {
 172         int i;
 173
 174         /* Verify that place we are splicing to is still there and vacant */
 175         write_lock(&pointers_lock);
 176         if (!verify_chain(chain, where-1) || *where->p)
 177                 goto changed;
 178         *where->p = where->key;
 179         write_unlock(&pointers_lock);
 180
 181         inode->i_ctime = CURRENT_TIME;
 182
 183         /* had we spliced it onto indirect block? */
 184         if (where->bh)
 185                 dirty_indirect(where->bh, inode);
 186
 187         if (IS_SYNC(inode))
 188                 sysv_sync_inode(inode);
 189         else
 190                 mark_inode_dirty(inode);
 191         return 0;
 192
 193 changed:
 194         write_unlock(&pointers_lock);
 195         for (i = 1; i < num; i++)
 196                 bforget(where[i].bh);
 197         for (i = 0; i < num; i++)
 198                 sysv_free_block(inode->i_sb, where[i].key);
 199         return -EAGAIN;
 200 }
 201
 202 static int get_block(struct inode *inode, sector_t iblock, struct buffer_head *bh_result, int create)
 203 {
 204         int err = -EIO;
 205         int offsets[DEPTH];
 206         Indirect chain[DEPTH];
 207         struct super_block *sb = inode->i_sb;
 208         Indirect *partial;
 209         int left;
 210         int depth = block_to_path(inode, iblock, offsets);
 211
 212         if (depth == 0)
 213                 goto out;
 214
 215 reread:
 216         partial = get_branch(inode, depth, offsets, chain, &err);
 217
 218         /* Simplest case - block found, no allocation needed */
 219         if (!partial) {
 220 got_it:
 221                 map_bh(bh_result, sb, block_to_cpu(SYSV_SB(sb),
 222                                         chain[depth-1].key));
 223                 /* Clean up and exit */
 224                 partial = chain+depth-1; /* the whole chain */
 225                 goto cleanup;
 226         }
 227
 228         /* Next simple case - plain lookup or failed read of indirect block */
 229         if (!create || err == -EIO) {
 230 cleanup:
 231                 while (partial > chain) {
 232                         brelse(partial->bh);
 233                         partial--;
 234                 }
 235 out:
 236                 return err;
 237         }
 238
 239         /*
 240          * Indirect block might be removed by truncate while we were
 241          * reading it. Handling of that case (forget what we've got and
 242          * reread) is taken out of the main path.
 243          */
 244         if (err == -EAGAIN)
 245                 goto changed;
 246
 247         left = (chain + depth) - partial;
 248         err = alloc_branch(inode, left, offsets+(partial-chain), partial);
 249         if (err)
 250                 goto cleanup;
 251
 252         if (splice_branch(inode, chain, partial, left) < 0)
 253                 goto changed;
 254
 255         set_buffer_new(bh_result);
 256         goto got_it;
 257
 258 changed:
 259         while (partial > chain) {
 260                 brelse(partial->bh);
 261                 partial--;
 262         }
 263         goto reread;
 264 }
 265
 266 static inline int all_zeroes(u32 *p, u32 *q)
 267 {
 268         while (p < q)
 269                 if (*p++)
 270                         return 0;
 271         return 1;
 272 }
 273
 274 static Indirect *find_shared(struct inode *inode,
 275                                 int depth,
 276                                 int offsets[],
 277                                 Indirect chain[],
 278                                 u32 *top)
 279 {
 280         Indirect *partial, *p;
 281         int k, err;
 282
 283         *top = 0;
 284         for (k = depth; k > 1 && !offsets[k-1]; k--)
 285                 ;
 286
 287         write_lock(&pointers_lock);
 288         partial = get_branch(inode, k, offsets, chain, &err);
 289         if (!partial)
 290                 partial = chain + k-1;
 291         /*
 292          * If the branch acquired continuation since we've looked at it -
 293          * fine, it should all survive and (new) top doesn't belong to us.
 294          */
 295         if (!partial->key && *partial->p) {
 296                 write_unlock(&pointers_lock);
 297                 goto no_top;
 298         }
 299         for (p=partial; p>chain && all_zeroes((u32*)p->bh->b_data,p->p); p--)
 300                 ;
 301         /*
 302          * OK, we've found the last block that must survive. The rest of our
 303          * branch should be detached before unlocking. However, if that rest
 304          * of branch is all ours and does not grow immediately from the inode
 305          * it's easier to cheat and just decrement partial->p.
 306          */
 307         if (p == chain + k - 1 && p > chain) {
 308                 p->p--;
 309         } else {
 310                 *top = *p->p;
 311                 *p->p = 0;
 312         }
 313         write_unlock(&pointers_lock);
 314
 315         while (partial > p) {
 316                 brelse(partial->bh);
 317                 partial--;
 318         }
 319 no_top:
 320         return partial;
 321 }
 322
 323 static inline void free_data(struct inode *inode, u32 *p, u32 *q)
 324 {
 325         for ( ; p < q ; p++) {
 326                 u32 nr = *p;
 327                 if (nr) {
 328                         *p = 0;
 329                         sysv_free_block(inode->i_sb, nr);
 330                         mark_inode_dirty(inode);
 331                 }
 332         }
 333 }
 334
 335 static void free_branches(struct inode *inode, u32 *p, u32 *q, int depth)
 336 {
 337         struct buffer_head * bh;
 338         struct super_block *sb = inode->i_sb;
 339
 340         if (depth--) {
 341                 for ( ; p < q ; p++) {
 342                         int block;
 343                         u32 nr = *p;
 344                         if (!nr)
 345                                 continue;
 346                         *p = 0;
 347                         block = block_to_cpu(SYSV_SB(sb), nr);
 348                         bh = sb_bread(sb, block);
 349                         if (!bh)
 350                                 continue;
 351                         free_branches(inode, (u32*)bh->b_data,
 352                                         block_end(bh), depth);
 353                         bforget(bh);
 354                         sysv_free_block(sb, nr);
 355                         mark_inode_dirty(inode);
 356                 }
 357         } else
 358                 free_data(inode, p, q);
 359 }
 360
 361 void sysv_truncate (struct inode * inode)
 362 {
 363         u32 *i_data = SYSV_I(inode)->i_data;
 364         int offsets[DEPTH];
 365         Indirect chain[DEPTH];
 366         Indirect *partial;
 367         int nr = 0;
 368         int n;
 369         long iblock;
 370         unsigned blocksize;
 371
 372         if (!(S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
 373             S_ISLNK(inode->i_mode)))
 374                 return;
 375
 376         blocksize = inode->i_sb->s_blocksize;
 377         iblock = (inode->i_size + blocksize-1)
 378                                         >> inode->i_sb->s_blocksize_bits;
 379
 380         block_truncate_page(inode->i_mapping, inode->i_size, get_block);
 381
 382         n = block_to_path(inode, iblock, offsets);
 383         if (n == 0)
 384                 return;
 385
 386         if (n == 1) {
 387                 free_data(inode, i_data+offsets[0], i_data + DIRECT);
 388                 goto do_indirects;
 389         }
 390
 391         partial = find_shared(inode, n, offsets, chain, &nr);
 392         /* Kill the top of shared branch (already detached) */
 393         if (nr) {
 394                 if (partial == chain)
 395                         mark_inode_dirty(inode);
 396                 else
 397                         dirty_indirect(partial->bh, inode);
 398                 free_branches(inode, &nr, &nr+1, (chain+n-1) - partial);
 399         }
 400         /* Clear the ends of indirect blocks on the shared branch */
 401         while (partial > chain) {
 402                 free_branches(inode, partial->p + 1, block_end(partial->bh),
 403                                 (chain+n-1) - partial);
 404                 dirty_indirect(partial->bh, inode);
 405                 brelse (partial->bh);
 406                 partial--;
 407         }
 408 do_indirects:
 409         /* Kill the remaining (whole) subtrees (== subtrees deeper than...) */
 410         while (n < DEPTH) {
 411                 nr = i_data[DIRECT + n - 1];
 412                 if (nr) {
 413                         i_data[DIRECT + n - 1] = 0;
 414                         mark_inode_dirty(inode);
 415                         free_branches(inode, &nr, &nr+1, n);
 416                 }
 417                 n++;
 418         }
 419         inode->i_mtime = inode->i_ctime = CURRENT_TIME;
 420         if (IS_SYNC(inode))
 421                 sysv_sync_inode (inode);
 422         else
 423                 mark_inode_dirty(inode);
 424 }
 425
 426 static unsigned sysv_nblocks(struct super_block *s, loff_t size)
 427 {
 428         struct sysv_sb_info *sbi = SYSV_SB(s);
 429         int ptrs_bits = sbi->s_ind_per_block_bits;
 430         unsigned blocks, res, direct = DIRECT, i = DEPTH;
 431         blocks = (size + s->s_blocksize - 1) >> s->s_blocksize_bits;
 432         res = blocks;
 433         while (--i && blocks > direct) {
 434                 blocks = ((blocks - direct - 1) >> ptrs_bits) + 1;
 435                 res += blocks;
 436                 direct = 1;
 437         }
 438         return blocks;
 439 }
 440
 441 int sysv_getattr(struct vfsmount *mnt, struct dentry *dentry, struct kstat *stat)
 442 {
 443         struct super_block *s = mnt->mnt_sb;
 444         generic_fillattr(dentry->d_inode, stat);
 445         stat->blocks = (s->s_blocksize / 512) * sysv_nblocks(s, stat->size);
 446         stat->blksize = s->s_blocksize;
 447         return 0;
 448 }
 449
 450 static int sysv_writepage(struct page *page, struct writeback_control *wbc)
 451 {
 452         return block_write_full_page(page,get_block,wbc);
 453 }
 454 static int sysv_readpage(struct file *file, struct page *page)
 455 {
 456         return block_read_full_page(page,get_block);
 457 }
 458 static int sysv_prepare_write(struct file *file, struct page *page, unsigned from, unsigned to)
 459 {
 460         return block_prepare_write(page,from,to,get_block);
 461 }
 462 static sector_t sysv_bmap(struct address_space *mapping, sector_t block)
 463 {
 464         return generic_block_bmap(mapping,block,get_block);
 465 }
 466 struct address_space_operations sysv_aops = {
 467         .readpage = sysv_readpage,
 468         .writepage = sysv_writepage,
 469         .sync_page = block_sync_page,
 470         .prepare_write = sysv_prepare_write,
 471         .commit_write = generic_commit_write,
 472         .bmap = sysv_bmap
 473 };