fs/sysv/itree.c

   1 // SPDX-License-Identifier: GPL-2.0
   2 /*
   3  *  linux/fs/sysv/itree.c
   4  *
   5  *  Handling of indirect blocks' trees.
   6  *  AV, Sep--Dec 2000
   7  */
   8
   9 #include <linux/buffer_head.h>
  10 #include <linux/mount.h>
  11 #include <linux/string.h>
  12 #include "sysv.h"
  13
  14 enum {DIRECT = 10, DEPTH = 4};  /* Have triple indirect */
  15
  16 static inline void dirty_indirect(struct buffer_head *bh, struct inode *inode)
  17 {
  18         mark_buffer_dirty_inode(bh, inode);
  19         if (IS_SYNC(inode))
  20                 sync_dirty_buffer(bh);
  21 }
  22
  23 static int block_to_path(struct inode *inode, long block, int offsets[DEPTH])
  24 {
  25         struct super_block *sb = inode->i_sb;
  26         struct sysv_sb_info *sbi = SYSV_SB(sb);
  27         int ptrs_bits = sbi->s_ind_per_block_bits;
  28         unsigned long   indirect_blocks = sbi->s_ind_per_block,
  29                         double_blocks = sbi->s_ind_per_block_2;
  30         int n = 0;
  31
  32         if (block < 0) {
  33                 printk("sysv_block_map: block < 0\n");
  34         } else if (block < DIRECT) {
  35                 offsets[n++] = block;
  36         } else if ( (block -= DIRECT) < indirect_blocks) {
  37                 offsets[n++] = DIRECT;
  38                 offsets[n++] = block;
  39         } else if ((block -= indirect_blocks) < double_blocks) {
  40                 offsets[n++] = DIRECT+1;
  41                 offsets[n++] = block >> ptrs_bits;
  42                 offsets[n++] = block & (indirect_blocks - 1);
  43         } else if (((block -= double_blocks) >> (ptrs_bits * 2)) < indirect_blocks) {
  44                 offsets[n++] = DIRECT+2;
  45                 offsets[n++] = block >> (ptrs_bits * 2);
  46                 offsets[n++] = (block >> ptrs_bits) & (indirect_blocks - 1);
  47                 offsets[n++] = block & (indirect_blocks - 1);
  48         } else {
  49                 /* nothing */;
  50         }
  51         return n;
  52 }
  53
  54 static inline int block_to_cpu(struct sysv_sb_info *sbi, sysv_zone_t nr)
  55 {
  56         return sbi->s_block_base + fs32_to_cpu(sbi, nr);
  57 }
  58
  59 typedef struct {
  60         sysv_zone_t     *p;
  61         sysv_zone_t     key;
  62         struct buffer_head *bh;
  63 } Indirect;
  64
  65 static DEFINE_RWLOCK(pointers_lock);
  66
  67 static inline void add_chain(Indirect *p, struct buffer_head *bh, sysv_zone_t *v)
  68 {
  69         p->key = *(p->p = v);
  70         p->bh = bh;
  71 }
  72
  73 static inline int verify_chain(Indirect *from, Indirect *to)
  74 {
  75         while (from <= to && from->key == *from->p)
  76                 from++;
  77         return (from > to);
  78 }
  79
  80 static inline sysv_zone_t *block_end(struct buffer_head *bh)
  81 {
  82         return (sysv_zone_t*)((char*)bh->b_data + bh->b_size);
  83 }
  84
  85 /*
  86  * Requires read_lock(&pointers_lock) or write_lock(&pointers_lock)
  87  */
  88 static Indirect *get_branch(struct inode *inode,
  89                             int depth,
  90                             int offsets[],
  91                             Indirect chain[],
  92                             int *err)
  93 {
  94         struct super_block *sb = inode->i_sb;
  95         Indirect *p = chain;
  96         struct buffer_head *bh;
  97
  98         *err = 0;
  99         add_chain(chain, NULL, SYSV_I(inode)->i_data + *offsets);
 100         if (!p->key)
 101                 goto no_block;
 102         while (--depth) {
 103                 int block = block_to_cpu(SYSV_SB(sb), p->key);
 104                 bh = sb_bread(sb, block);
 105                 if (!bh)
 106                         goto failure;
 107                 if (!verify_chain(chain, p))
 108                         goto changed;
 109                 add_chain(++p, bh, (sysv_zone_t*)bh->b_data + *++offsets);
 110                 if (!p->key)
 111                         goto no_block;
 112         }
 113         return NULL;
 114
 115 changed:
 116         brelse(bh);
 117         *err = -EAGAIN;
 118         goto no_block;
 119 failure:
 120         *err = -EIO;
 121 no_block:
 122         return p;
 123 }
 124
 125 static int alloc_branch(struct inode *inode,
 126                         int num,
 127                         int *offsets,
 128                         Indirect *branch)
 129 {
 130         int blocksize = inode->i_sb->s_blocksize;
 131         int n = 0;
 132         int i;
 133
 134         branch[0].key = sysv_new_block(inode->i_sb);
 135         if (branch[0].key) for (n = 1; n < num; n++) {
 136                 struct buffer_head *bh;
 137                 int parent;
 138                 /* Allocate the next block */
 139                 branch[n].key = sysv_new_block(inode->i_sb);
 140                 if (!branch[n].key)
 141                         break;
 142                 /*
 143                  * Get buffer_head for parent block, zero it out and set
 144                  * the pointer to new one, then send parent to disk.
 145                  */
 146                 parent = block_to_cpu(SYSV_SB(inode->i_sb), branch[n-1].key);
 147                 bh = sb_getblk(inode->i_sb, parent);
 148                 if (!bh) {
 149                         sysv_free_block(inode->i_sb, branch[n].key);
 150                         break;
 151                 }
 152                 lock_buffer(bh);
 153                 memset(bh->b_data, 0, blocksize);
 154                 branch[n].bh = bh;
 155                 branch[n].p = (sysv_zone_t*) bh->b_data + offsets[n];
 156                 *branch[n].p = branch[n].key;
 157                 set_buffer_uptodate(bh);
 158                 unlock_buffer(bh);
 159                 dirty_indirect(bh, inode);
 160         }
 161         if (n == num)
 162                 return 0;
 163
 164         /* Allocation failed, free what we already allocated */
 165         for (i = 1; i < n; i++)
 166                 bforget(branch[i].bh);
 167         for (i = 0; i < n; i++)
 168                 sysv_free_block(inode->i_sb, branch[i].key);
 169         return -ENOSPC;
 170 }
 171
 172 static inline int splice_branch(struct inode *inode,
 173                                 Indirect chain[],
 174                                 Indirect *where,
 175                                 int num)
 176 {
 177         int i;
 178
 179         /* Verify that place we are splicing to is still there and vacant */
 180         write_lock(&pointers_lock);
 181         if (!verify_chain(chain, where-1) || *where->p)
 182                 goto changed;
 183         *where->p = where->key;
 184         write_unlock(&pointers_lock);
 185
 186         inode->i_ctime = current_time(inode);
 187
 188         /* had we spliced it onto indirect block? */
 189         if (where->bh)
 190                 dirty_indirect(where->bh, inode);
 191
 192         if (IS_SYNC(inode))
 193                 sysv_sync_inode(inode);
 194         else
 195                 mark_inode_dirty(inode);
 196         return 0;
 197
 198 changed:
 199         write_unlock(&pointers_lock);
 200         for (i = 1; i < num; i++)
 201                 bforget(where[i].bh);
 202         for (i = 0; i < num; i++)
 203                 sysv_free_block(inode->i_sb, where[i].key);
 204         return -EAGAIN;
 205 }
 206
 207 static int get_block(struct inode *inode, sector_t iblock, struct buffer_head *bh_result, int create)
 208 {
 209         int err = -EIO;
 210         int offsets[DEPTH];
 211         Indirect chain[DEPTH];
 212         struct super_block *sb = inode->i_sb;
 213         Indirect *partial;
 214         int left;
 215         int depth = block_to_path(inode, iblock, offsets);
 216
 217         if (depth == 0)
 218                 goto out;
 219
 220 reread:
 221         read_lock(&pointers_lock);
 222         partial = get_branch(inode, depth, offsets, chain, &err);
 223         read_unlock(&pointers_lock);
 224
 225         /* Simplest case - block found, no allocation needed */
 226         if (!partial) {
 227 got_it:
 228                 map_bh(bh_result, sb, block_to_cpu(SYSV_SB(sb),
 229                                         chain[depth-1].key));
 230                 /* Clean up and exit */
 231                 partial = chain+depth-1; /* the whole chain */
 232                 goto cleanup;
 233         }
 234
 235         /* Next simple case - plain lookup or failed read of indirect block */
 236         if (!create || err == -EIO) {
 237 cleanup:
 238                 while (partial > chain) {
 239                         brelse(partial->bh);
 240                         partial--;
 241                 }
 242 out:
 243                 return err;
 244         }
 245
 246         /*
 247          * Indirect block might be removed by truncate while we were
 248          * reading it. Handling of that case (forget what we've got and
 249          * reread) is taken out of the main path.
 250          */
 251         if (err == -EAGAIN)
 252                 goto changed;
 253
 254         left = (chain + depth) - partial;
 255         err = alloc_branch(inode, left, offsets+(partial-chain), partial);
 256         if (err)
 257                 goto cleanup;
 258
 259         if (splice_branch(inode, chain, partial, left) < 0)
 260                 goto changed;
 261
 262         set_buffer_new(bh_result);
 263         goto got_it;
 264
 265 changed:
 266         while (partial > chain) {
 267                 brelse(partial->bh);
 268                 partial--;
 269         }
 270         goto reread;
 271 }
 272
 273 static inline int all_zeroes(sysv_zone_t *p, sysv_zone_t *q)
 274 {
 275         while (p < q)
 276                 if (*p++)
 277                         return 0;
 278         return 1;
 279 }
 280
 281 static Indirect *find_shared(struct inode *inode,
 282                                 int depth,
 283                                 int offsets[],
 284                                 Indirect chain[],
 285                                 sysv_zone_t *top)
 286 {
 287         Indirect *partial, *p;
 288         int k, err;
 289
 290         *top = 0;
 291         for (k = depth; k > 1 && !offsets[k-1]; k--)
 292                 ;
 293
 294         write_lock(&pointers_lock);
 295         partial = get_branch(inode, k, offsets, chain, &err);
 296         if (!partial)
 297                 partial = chain + k-1;
 298         /*
 299          * If the branch acquired continuation since we've looked at it -
 300          * fine, it should all survive and (new) top doesn't belong to us.
 301          */
 302         if (!partial->key && *partial->p) {
 303                 write_unlock(&pointers_lock);
 304                 goto no_top;
 305         }
 306         for (p=partial; p>chain && all_zeroes((sysv_zone_t*)p->bh->b_data,p->p); p--)
 307                 ;
 308         /*
 309          * OK, we've found the last block that must survive. The rest of our
 310          * branch should be detached before unlocking. However, if that rest
 311          * of branch is all ours and does not grow immediately from the inode
 312          * it's easier to cheat and just decrement partial->p.
 313          */
 314         if (p == chain + k - 1 && p > chain) {
 315                 p->p--;
 316         } else {
 317                 *top = *p->p;
 318                 *p->p = 0;
 319         }
 320         write_unlock(&pointers_lock);
 321
 322         while (partial > p) {
 323                 brelse(partial->bh);
 324                 partial--;
 325         }
 326 no_top:
 327         return partial;
 328 }
 329
 330 static inline void free_data(struct inode *inode, sysv_zone_t *p, sysv_zone_t *q)
 331 {
 332         for ( ; p < q ; p++) {
 333                 sysv_zone_t nr = *p;
 334                 if (nr) {
 335                         *p = 0;
 336                         sysv_free_block(inode->i_sb, nr);
 337                         mark_inode_dirty(inode);
 338                 }
 339         }
 340 }
 341
 342 static void free_branches(struct inode *inode, sysv_zone_t *p, sysv_zone_t *q, int depth)
 343 {
 344         struct buffer_head * bh;
 345         struct super_block *sb = inode->i_sb;
 346
 347         if (depth--) {
 348                 for ( ; p < q ; p++) {
 349                         int block;
 350                         sysv_zone_t nr = *p;
 351                         if (!nr)
 352                                 continue;
 353                         *p = 0;
 354                         block = block_to_cpu(SYSV_SB(sb), nr);
 355                         bh = sb_bread(sb, block);
 356                         if (!bh)
 357                                 continue;
 358                         free_branches(inode, (sysv_zone_t*)bh->b_data,
 359                                         block_end(bh), depth);
 360                         bforget(bh);
 361                         sysv_free_block(sb, nr);
 362                         mark_inode_dirty(inode);
 363                 }
 364         } else
 365                 free_data(inode, p, q);
 366 }
 367
 368 void sysv_truncate (struct inode * inode)
 369 {
 370         sysv_zone_t *i_data = SYSV_I(inode)->i_data;
 371         int offsets[DEPTH];
 372         Indirect chain[DEPTH];
 373         Indirect *partial;
 374         sysv_zone_t nr = 0;
 375         int n;
 376         long iblock;
 377         unsigned blocksize;
 378
 379         if (!(S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
 380             S_ISLNK(inode->i_mode)))
 381                 return;
 382
 383         blocksize = inode->i_sb->s_blocksize;
 384         iblock = (inode->i_size + blocksize-1)
 385                                         >> inode->i_sb->s_blocksize_bits;
 386
 387         block_truncate_page(inode->i_mapping, inode->i_size, get_block);
 388
 389         n = block_to_path(inode, iblock, offsets);
 390         if (n == 0)
 391                 return;
 392
 393         if (n == 1) {
 394                 free_data(inode, i_data+offsets[0], i_data + DIRECT);
 395                 goto do_indirects;
 396         }
 397
 398         partial = find_shared(inode, n, offsets, chain, &nr);
 399         /* Kill the top of shared branch (already detached) */
 400         if (nr) {
 401                 if (partial == chain)
 402                         mark_inode_dirty(inode);
 403                 else
 404                         dirty_indirect(partial->bh, inode);
 405                 free_branches(inode, &nr, &nr+1, (chain+n-1) - partial);
 406         }
 407         /* Clear the ends of indirect blocks on the shared branch */
 408         while (partial > chain) {
 409                 free_branches(inode, partial->p + 1, block_end(partial->bh),
 410                                 (chain+n-1) - partial);
 411                 dirty_indirect(partial->bh, inode);
 412                 brelse (partial->bh);
 413                 partial--;
 414         }
 415 do_indirects:
 416         /* Kill the remaining (whole) subtrees (== subtrees deeper than...) */
 417         while (n < DEPTH) {
 418                 nr = i_data[DIRECT + n - 1];
 419                 if (nr) {
 420                         i_data[DIRECT + n - 1] = 0;
 421                         mark_inode_dirty(inode);
 422                         free_branches(inode, &nr, &nr+1, n);
 423                 }
 424                 n++;
 425         }
 426         inode->i_mtime = inode->i_ctime = current_time(inode);
 427         if (IS_SYNC(inode))
 428                 sysv_sync_inode (inode);
 429         else
 430                 mark_inode_dirty(inode);
 431 }
 432
 433 static unsigned sysv_nblocks(struct super_block *s, loff_t size)
 434 {
 435         struct sysv_sb_info *sbi = SYSV_SB(s);
 436         int ptrs_bits = sbi->s_ind_per_block_bits;
 437         unsigned blocks, res, direct = DIRECT, i = DEPTH;
 438         blocks = (size + s->s_blocksize - 1) >> s->s_blocksize_bits;
 439         res = blocks;
 440         while (--i && blocks > direct) {
 441                 blocks = ((blocks - direct - 1) >> ptrs_bits) + 1;
 442                 res += blocks;
 443                 direct = 1;
 444         }
 445         return res;
 446 }
 447
 448 int sysv_getattr(struct mnt_idmap *idmap, const struct path *path,
 449                  struct kstat *stat, u32 request_mask, unsigned int flags)
 450 {
 451         struct super_block *s = path->dentry->d_sb;
 452         generic_fillattr(&nop_mnt_idmap, d_inode(path->dentry), stat);
 453         stat->blocks = (s->s_blocksize / 512) * sysv_nblocks(s, stat->size);
 454         stat->blksize = s->s_blocksize;
 455         return 0;
 456 }
 457
 458 static int sysv_writepage(struct page *page, struct writeback_control *wbc)
 459 {
 460         return block_write_full_page(page,get_block,wbc);
 461 }
 462
 463 static int sysv_read_folio(struct file *file, struct folio *folio)
 464 {
 465         return block_read_full_folio(folio, get_block);
 466 }
 467
 468 int sysv_prepare_chunk(struct page *page, loff_t pos, unsigned len)
 469 {
 470         return __block_write_begin(page, pos, len, get_block);
 471 }
 472
 473 static void sysv_write_failed(struct address_space *mapping, loff_t to)
 474 {
 475         struct inode *inode = mapping->host;
 476
 477         if (to > inode->i_size) {
 478                 truncate_pagecache(inode, inode->i_size);
 479                 sysv_truncate(inode);
 480         }
 481 }
 482
 483 static int sysv_write_begin(struct file *file, struct address_space *mapping,
 484                         loff_t pos, unsigned len,
 485                         struct page **pagep, void **fsdata)
 486 {
 487         int ret;
 488
 489         ret = block_write_begin(mapping, pos, len, pagep, get_block);
 490         if (unlikely(ret))
 491                 sysv_write_failed(mapping, pos + len);
 492
 493         return ret;
 494 }
 495
 496 static sector_t sysv_bmap(struct address_space *mapping, sector_t block)
 497 {
 498         return generic_block_bmap(mapping,block,get_block);
 499 }
 500
 501 const struct address_space_operations sysv_aops = {
 502         .dirty_folio = block_dirty_folio,
 503         .invalidate_folio = block_invalidate_folio,
 504         .read_folio = sysv_read_folio,
 505         .writepage = sysv_writepage,
 506         .write_begin = sysv_write_begin,
 507         .write_end = generic_write_end,
 508         .bmap = sysv_bmap
 509 };