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/mpage.h>
  12 #include <linux/string.h>
  13 #include "sysv.h"
  14
  15 enum {DIRECT = 10, DEPTH = 4};  /* Have triple indirect */
  16
  17 static inline void dirty_indirect(struct buffer_head *bh, struct inode *inode)
  18 {
  19         mark_buffer_dirty_inode(bh, inode);
  20         if (IS_SYNC(inode))
  21                 sync_dirty_buffer(bh);
  22 }
  23
  24 static int block_to_path(struct inode *inode, long block, int offsets[DEPTH])
  25 {
  26         struct super_block *sb = inode->i_sb;
  27         struct sysv_sb_info *sbi = SYSV_SB(sb);
  28         int ptrs_bits = sbi->s_ind_per_block_bits;
  29         unsigned long   indirect_blocks = sbi->s_ind_per_block,
  30                         double_blocks = sbi->s_ind_per_block_2;
  31         int n = 0;
  32
  33         if (block < 0) {
  34                 printk("sysv_block_map: block < 0\n");
  35         } else if (block < DIRECT) {
  36                 offsets[n++] = block;
  37         } else if ( (block -= DIRECT) < indirect_blocks) {
  38                 offsets[n++] = DIRECT;
  39                 offsets[n++] = block;
  40         } else if ((block -= indirect_blocks) < double_blocks) {
  41                 offsets[n++] = DIRECT+1;
  42                 offsets[n++] = block >> ptrs_bits;
  43                 offsets[n++] = block & (indirect_blocks - 1);
  44         } else if (((block -= double_blocks) >> (ptrs_bits * 2)) < indirect_blocks) {
  45                 offsets[n++] = DIRECT+2;
  46                 offsets[n++] = block >> (ptrs_bits * 2);
  47                 offsets[n++] = (block >> ptrs_bits) & (indirect_blocks - 1);
  48                 offsets[n++] = block & (indirect_blocks - 1);
  49         } else {
  50                 /* nothing */;
  51         }
  52         return n;
  53 }
  54
  55 static inline int block_to_cpu(struct sysv_sb_info *sbi, sysv_zone_t nr)
  56 {
  57         return sbi->s_block_base + fs32_to_cpu(sbi, nr);
  58 }
  59
  60 typedef struct {
  61         sysv_zone_t     *p;
  62         sysv_zone_t     key;
  63         struct buffer_head *bh;
  64 } Indirect;
  65
  66 static DEFINE_RWLOCK(pointers_lock);
  67
  68 static inline void add_chain(Indirect *p, struct buffer_head *bh, sysv_zone_t *v)
  69 {
  70         p->key = *(p->p = v);
  71         p->bh = bh;
  72 }
  73
  74 static inline int verify_chain(Indirect *from, Indirect *to)
  75 {
  76         while (from <= to && from->key == *from->p)
  77                 from++;
  78         return (from > to);
  79 }
  80
  81 static inline sysv_zone_t *block_end(struct buffer_head *bh)
  82 {
  83         return (sysv_zone_t*)((char*)bh->b_data + bh->b_size);
  84 }
  85
  86 /*
  87  * Requires read_lock(&pointers_lock) or write_lock(&pointers_lock)
  88  */
  89 static Indirect *get_branch(struct inode *inode,
  90                             int depth,
  91                             int offsets[],
  92                             Indirect chain[],
  93                             int *err)
  94 {
  95         struct super_block *sb = inode->i_sb;
  96         Indirect *p = chain;
  97         struct buffer_head *bh;
  98
  99         *err = 0;
 100         add_chain(chain, NULL, SYSV_I(inode)->i_data + *offsets);
 101         if (!p->key)
 102                 goto no_block;
 103         while (--depth) {
 104                 int block = block_to_cpu(SYSV_SB(sb), p->key);
 105                 bh = sb_bread(sb, block);
 106                 if (!bh)
 107                         goto failure;
 108                 if (!verify_chain(chain, p))
 109                         goto changed;
 110                 add_chain(++p, bh, (sysv_zone_t*)bh->b_data + *++offsets);
 111                 if (!p->key)
 112                         goto no_block;
 113         }
 114         return NULL;
 115
 116 changed:
 117         brelse(bh);
 118         *err = -EAGAIN;
 119         goto no_block;
 120 failure:
 121         *err = -EIO;
 122 no_block:
 123         return p;
 124 }
 125
 126 static int alloc_branch(struct inode *inode,
 127                         int num,
 128                         int *offsets,
 129                         Indirect *branch)
 130 {
 131         int blocksize = inode->i_sb->s_blocksize;
 132         int n = 0;
 133         int i;
 134
 135         branch[0].key = sysv_new_block(inode->i_sb);
 136         if (branch[0].key) for (n = 1; n < num; n++) {
 137                 struct buffer_head *bh;
 138                 int parent;
 139                 /* Allocate the next block */
 140                 branch[n].key = sysv_new_block(inode->i_sb);
 141                 if (!branch[n].key)
 142                         break;
 143                 /*
 144                  * Get buffer_head for parent block, zero it out and set
 145                  * the pointer to new one, then send parent to disk.
 146                  */
 147                 parent = block_to_cpu(SYSV_SB(inode->i_sb), branch[n-1].key);
 148                 bh = sb_getblk(inode->i_sb, parent);
 149                 if (!bh) {
 150                         sysv_free_block(inode->i_sb, branch[n].key);
 151                         break;
 152                 }
 153                 lock_buffer(bh);
 154                 memset(bh->b_data, 0, blocksize);
 155                 branch[n].bh = bh;
 156                 branch[n].p = (sysv_zone_t*) bh->b_data + offsets[n];
 157                 *branch[n].p = branch[n].key;
 158                 set_buffer_uptodate(bh);
 159                 unlock_buffer(bh);
 160                 dirty_indirect(bh, inode);
 161         }
 162         if (n == num)
 163                 return 0;
 164
 165         /* Allocation failed, free what we already allocated */
 166         for (i = 1; i < n; i++)
 167                 bforget(branch[i].bh);
 168         for (i = 0; i < n; i++)
 169                 sysv_free_block(inode->i_sb, branch[i].key);
 170         return -ENOSPC;
 171 }
 172
 173 static inline int splice_branch(struct inode *inode,
 174                                 Indirect chain[],
 175                                 Indirect *where,
 176                                 int num)
 177 {
 178         int i;
 179
 180         /* Verify that place we are splicing to is still there and vacant */
 181         write_lock(&pointers_lock);
 182         if (!verify_chain(chain, where-1) || *where->p)
 183                 goto changed;
 184         *where->p = where->key;
 185         write_unlock(&pointers_lock);
 186
 187         inode_set_ctime_current(inode);
 188
 189         /* had we spliced it onto indirect block? */
 190         if (where->bh)
 191                 dirty_indirect(where->bh, inode);
 192
 193         if (IS_SYNC(inode))
 194                 sysv_sync_inode(inode);
 195         else
 196                 mark_inode_dirty(inode);
 197         return 0;
 198
 199 changed:
 200         write_unlock(&pointers_lock);
 201         for (i = 1; i < num; i++)
 202                 bforget(where[i].bh);
 203         for (i = 0; i < num; i++)
 204                 sysv_free_block(inode->i_sb, where[i].key);
 205         return -EAGAIN;
 206 }
 207
 208 static int get_block(struct inode *inode, sector_t iblock, struct buffer_head *bh_result, int create)
 209 {
 210         int err = -EIO;
 211         int offsets[DEPTH];
 212         Indirect chain[DEPTH];
 213         struct super_block *sb = inode->i_sb;
 214         Indirect *partial;
 215         int left;
 216         int depth = block_to_path(inode, iblock, offsets);
 217
 218         if (depth == 0)
 219                 goto out;
 220
 221 reread:
 222         read_lock(&pointers_lock);
 223         partial = get_branch(inode, depth, offsets, chain, &err);
 224         read_unlock(&pointers_lock);
 225
 226         /* Simplest case - block found, no allocation needed */
 227         if (!partial) {
 228 got_it:
 229                 map_bh(bh_result, sb, block_to_cpu(SYSV_SB(sb),
 230                                         chain[depth-1].key));
 231                 /* Clean up and exit */
 232                 partial = chain+depth-1; /* the whole chain */
 233                 goto cleanup;
 234         }
 235
 236         /* Next simple case - plain lookup or failed read of indirect block */
 237         if (!create || err == -EIO) {
 238 cleanup:
 239                 while (partial > chain) {
 240                         brelse(partial->bh);
 241                         partial--;
 242                 }
 243 out:
 244                 return err;
 245         }
 246
 247         /*
 248          * Indirect block might be removed by truncate while we were
 249          * reading it. Handling of that case (forget what we've got and
 250          * reread) is taken out of the main path.
 251          */
 252         if (err == -EAGAIN)
 253                 goto changed;
 254
 255         left = (chain + depth) - partial;
 256         err = alloc_branch(inode, left, offsets+(partial-chain), partial);
 257         if (err)
 258                 goto cleanup;
 259
 260         if (splice_branch(inode, chain, partial, left) < 0)
 261                 goto changed;
 262
 263         set_buffer_new(bh_result);
 264         goto got_it;
 265
 266 changed:
 267         while (partial > chain) {
 268                 brelse(partial->bh);
 269                 partial--;
 270         }
 271         goto reread;
 272 }
 273
 274 static inline int all_zeroes(sysv_zone_t *p, sysv_zone_t *q)
 275 {
 276         while (p < q)
 277                 if (*p++)
 278                         return 0;
 279         return 1;
 280 }
 281
 282 static Indirect *find_shared(struct inode *inode,
 283                                 int depth,
 284                                 int offsets[],
 285                                 Indirect chain[],
 286                                 sysv_zone_t *top)
 287 {
 288         Indirect *partial, *p;
 289         int k, err;
 290
 291         *top = 0;
 292         for (k = depth; k > 1 && !offsets[k-1]; k--)
 293                 ;
 294
 295         write_lock(&pointers_lock);
 296         partial = get_branch(inode, k, offsets, chain, &err);
 297         if (!partial)
 298                 partial = chain + k-1;
 299         /*
 300          * If the branch acquired continuation since we've looked at it -
 301          * fine, it should all survive and (new) top doesn't belong to us.
 302          */
 303         if (!partial->key && *partial->p) {
 304                 write_unlock(&pointers_lock);
 305                 goto no_top;
 306         }
 307         for (p=partial; p>chain && all_zeroes((sysv_zone_t*)p->bh->b_data,p->p); p--)
 308                 ;
 309         /*
 310          * OK, we've found the last block that must survive. The rest of our
 311          * branch should be detached before unlocking. However, if that rest
 312          * of branch is all ours and does not grow immediately from the inode
 313          * it's easier to cheat and just decrement partial->p.
 314          */
 315         if (p == chain + k - 1 && p > chain) {
 316                 p->p--;
 317         } else {
 318                 *top = *p->p;
 319                 *p->p = 0;
 320         }
 321         write_unlock(&pointers_lock);
 322
 323         while (partial > p) {
 324                 brelse(partial->bh);
 325                 partial--;
 326         }
 327 no_top:
 328         return partial;
 329 }
 330
 331 static inline void free_data(struct inode *inode, sysv_zone_t *p, sysv_zone_t *q)
 332 {
 333         for ( ; p < q ; p++) {
 334                 sysv_zone_t nr = *p;
 335                 if (nr) {
 336                         *p = 0;
 337                         sysv_free_block(inode->i_sb, nr);
 338                         mark_inode_dirty(inode);
 339                 }
 340         }
 341 }
 342
 343 static void free_branches(struct inode *inode, sysv_zone_t *p, sysv_zone_t *q, int depth)
 344 {
 345         struct buffer_head * bh;
 346         struct super_block *sb = inode->i_sb;
 347
 348         if (depth--) {
 349                 for ( ; p < q ; p++) {
 350                         int block;
 351                         sysv_zone_t nr = *p;
 352                         if (!nr)
 353                                 continue;
 354                         *p = 0;
 355                         block = block_to_cpu(SYSV_SB(sb), nr);
 356                         bh = sb_bread(sb, block);
 357                         if (!bh)
 358                                 continue;
 359                         free_branches(inode, (sysv_zone_t*)bh->b_data,
 360                                         block_end(bh), depth);
 361                         bforget(bh);
 362                         sysv_free_block(sb, nr);
 363                         mark_inode_dirty(inode);
 364                 }
 365         } else
 366                 free_data(inode, p, q);
 367 }
 368
 369 void sysv_truncate (struct inode * inode)
 370 {
 371         sysv_zone_t *i_data = SYSV_I(inode)->i_data;
 372         int offsets[DEPTH];
 373         Indirect chain[DEPTH];
 374         Indirect *partial;
 375         sysv_zone_t nr = 0;
 376         int n;
 377         long iblock;
 378         unsigned blocksize;
 379
 380         if (!(S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
 381             S_ISLNK(inode->i_mode)))
 382                 return;
 383
 384         blocksize = inode->i_sb->s_blocksize;
 385         iblock = (inode->i_size + blocksize-1)
 386                                         >> inode->i_sb->s_blocksize_bits;
 387
 388         block_truncate_page(inode->i_mapping, inode->i_size, get_block);
 389
 390         n = block_to_path(inode, iblock, offsets);
 391         if (n == 0)
 392                 return;
 393
 394         if (n == 1) {
 395                 free_data(inode, i_data+offsets[0], i_data + DIRECT);
 396                 goto do_indirects;
 397         }
 398
 399         partial = find_shared(inode, n, offsets, chain, &nr);
 400         /* Kill the top of shared branch (already detached) */
 401         if (nr) {
 402                 if (partial == chain)
 403                         mark_inode_dirty(inode);
 404                 else
 405                         dirty_indirect(partial->bh, inode);
 406                 free_branches(inode, &nr, &nr+1, (chain+n-1) - partial);
 407         }
 408         /* Clear the ends of indirect blocks on the shared branch */
 409         while (partial > chain) {
 410                 free_branches(inode, partial->p + 1, block_end(partial->bh),
 411                                 (chain+n-1) - partial);
 412                 dirty_indirect(partial->bh, inode);
 413                 brelse (partial->bh);
 414                 partial--;
 415         }
 416 do_indirects:
 417         /* Kill the remaining (whole) subtrees (== subtrees deeper than...) */
 418         while (n < DEPTH) {
 419                 nr = i_data[DIRECT + n - 1];
 420                 if (nr) {
 421                         i_data[DIRECT + n - 1] = 0;
 422                         mark_inode_dirty(inode);
 423                         free_branches(inode, &nr, &nr+1, n);
 424                 }
 425                 n++;
 426         }
 427         inode_set_mtime_to_ts(inode, inode_set_ctime_current(inode));
 428         if (IS_SYNC(inode))
 429                 sysv_sync_inode (inode);
 430         else
 431                 mark_inode_dirty(inode);
 432 }
 433
 434 static unsigned sysv_nblocks(struct super_block *s, loff_t size)
 435 {
 436         struct sysv_sb_info *sbi = SYSV_SB(s);
 437         int ptrs_bits = sbi->s_ind_per_block_bits;
 438         unsigned blocks, res, direct = DIRECT, i = DEPTH;
 439         blocks = (size + s->s_blocksize - 1) >> s->s_blocksize_bits;
 440         res = blocks;
 441         while (--i && blocks > direct) {
 442                 blocks = ((blocks - direct - 1) >> ptrs_bits) + 1;
 443                 res += blocks;
 444                 direct = 1;
 445         }
 446         return res;
 447 }
 448
 449 int sysv_getattr(struct mnt_idmap *idmap, const struct path *path,
 450                  struct kstat *stat, u32 request_mask, unsigned int flags)
 451 {
 452         struct super_block *s = path->dentry->d_sb;
 453         generic_fillattr(&nop_mnt_idmap, request_mask, d_inode(path->dentry),
 454                          stat);
 455         stat->blocks = (s->s_blocksize / 512) * sysv_nblocks(s, stat->size);
 456         stat->blksize = s->s_blocksize;
 457         return 0;
 458 }
 459
 460 static int sysv_writepages(struct address_space *mapping,
 461                 struct writeback_control *wbc)
 462 {
 463         return mpage_writepages(mapping, wbc, get_block);
 464 }
 465
 466 static int sysv_read_folio(struct file *file, struct folio *folio)
 467 {
 468         return block_read_full_folio(folio, get_block);
 469 }
 470
 471 int sysv_prepare_chunk(struct page *page, loff_t pos, unsigned len)
 472 {
 473         return __block_write_begin(page, pos, len, get_block);
 474 }
 475
 476 static void sysv_write_failed(struct address_space *mapping, loff_t to)
 477 {
 478         struct inode *inode = mapping->host;
 479
 480         if (to > inode->i_size) {
 481                 truncate_pagecache(inode, inode->i_size);
 482                 sysv_truncate(inode);
 483         }
 484 }
 485
 486 static int sysv_write_begin(struct file *file, struct address_space *mapping,
 487                         loff_t pos, unsigned len,
 488                         struct page **pagep, void **fsdata)
 489 {
 490         int ret;
 491
 492         ret = block_write_begin(mapping, pos, len, pagep, get_block);
 493         if (unlikely(ret))
 494                 sysv_write_failed(mapping, pos + len);
 495
 496         return ret;
 497 }
 498
 499 static sector_t sysv_bmap(struct address_space *mapping, sector_t block)
 500 {
 501         return generic_block_bmap(mapping,block,get_block);
 502 }
 503
 504 const struct address_space_operations sysv_aops = {
 505         .dirty_folio = block_dirty_folio,
 506         .invalidate_folio = block_invalidate_folio,
 507         .read_folio = sysv_read_folio,
 508         .writepages = sysv_writepages,
 509         .write_begin = sysv_write_begin,
 510         .write_end = generic_write_end,
 511         .migrate_folio = buffer_migrate_folio,
 512         .bmap = sysv_bmap
 513 };