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 static Indirect *get_branch(struct inode *inode,
  86                             int depth,
  87                             int offsets[],
  88                             Indirect chain[],
  89                             int *err)
  90 {
  91         struct super_block *sb = inode->i_sb;
  92         Indirect *p = chain;
  93         struct buffer_head *bh;
  94
  95         *err = 0;
  96         add_chain(chain, NULL, SYSV_I(inode)->i_data + *offsets);
  97         if (!p->key)
  98                 goto no_block;
  99         while (--depth) {
 100                 int block = block_to_cpu(SYSV_SB(sb), p->key);
 101                 bh = sb_bread(sb, block);
 102                 if (!bh)
 103                         goto failure;
 104                 read_lock(&pointers_lock);
 105                 if (!verify_chain(chain, p))
 106                         goto changed;
 107                 add_chain(++p, bh, (sysv_zone_t*)bh->b_data + *++offsets);
 108                 read_unlock(&pointers_lock);
 109                 if (!p->key)
 110                         goto no_block;
 111         }
 112         return NULL;
 113
 114 changed:
 115         read_unlock(&pointers_lock);
 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         partial = get_branch(inode, depth, offsets, chain, &err);
 222
 223         /* Simplest case - block found, no allocation needed */
 224         if (!partial) {
 225 got_it:
 226                 map_bh(bh_result, sb, block_to_cpu(SYSV_SB(sb),
 227                                         chain[depth-1].key));
 228                 /* Clean up and exit */
 229                 partial = chain+depth-1; /* the whole chain */
 230                 goto cleanup;
 231         }
 232
 233         /* Next simple case - plain lookup or failed read of indirect block */
 234         if (!create || err == -EIO) {
 235 cleanup:
 236                 while (partial > chain) {
 237                         brelse(partial->bh);
 238                         partial--;
 239                 }
 240 out:
 241                 return err;
 242         }
 243
 244         /*
 245          * Indirect block might be removed by truncate while we were
 246          * reading it. Handling of that case (forget what we've got and
 247          * reread) is taken out of the main path.
 248          */
 249         if (err == -EAGAIN)
 250                 goto changed;
 251
 252         left = (chain + depth) - partial;
 253         err = alloc_branch(inode, left, offsets+(partial-chain), partial);
 254         if (err)
 255                 goto cleanup;
 256
 257         if (splice_branch(inode, chain, partial, left) < 0)
 258                 goto changed;
 259
 260         set_buffer_new(bh_result);
 261         goto got_it;
 262
 263 changed:
 264         while (partial > chain) {
 265                 brelse(partial->bh);
 266                 partial--;
 267         }
 268         goto reread;
 269 }
 270
 271 static inline int all_zeroes(sysv_zone_t *p, sysv_zone_t *q)
 272 {
 273         while (p < q)
 274                 if (*p++)
 275                         return 0;
 276         return 1;
 277 }
 278
 279 static Indirect *find_shared(struct inode *inode,
 280                                 int depth,
 281                                 int offsets[],
 282                                 Indirect chain[],
 283                                 sysv_zone_t *top)
 284 {
 285         Indirect *partial, *p;
 286         int k, err;
 287
 288         *top = 0;
 289         for (k = depth; k > 1 && !offsets[k-1]; k--)
 290                 ;
 291         partial = get_branch(inode, k, offsets, chain, &err);
 292
 293         write_lock(&pointers_lock);
 294         if (!partial)
 295                 partial = chain + k-1;
 296         /*
 297          * If the branch acquired continuation since we've looked at it -
 298          * fine, it should all survive and (new) top doesn't belong to us.
 299          */
 300         if (!partial->key && *partial->p) {
 301                 write_unlock(&pointers_lock);
 302                 goto no_top;
 303         }
 304         for (p=partial; p>chain && all_zeroes((sysv_zone_t*)p->bh->b_data,p->p); p--)
 305                 ;
 306         /*
 307          * OK, we've found the last block that must survive. The rest of our
 308          * branch should be detached before unlocking. However, if that rest
 309          * of branch is all ours and does not grow immediately from the inode
 310          * it's easier to cheat and just decrement partial->p.
 311          */
 312         if (p == chain + k - 1 && p > chain) {
 313                 p->p--;
 314         } else {
 315                 *top = *p->p;
 316                 *p->p = 0;
 317         }
 318         write_unlock(&pointers_lock);
 319
 320         while (partial > p) {
 321                 brelse(partial->bh);
 322                 partial--;
 323         }
 324 no_top:
 325         return partial;
 326 }
 327
 328 static inline void free_data(struct inode *inode, sysv_zone_t *p, sysv_zone_t *q)
 329 {
 330         for ( ; p < q ; p++) {
 331                 sysv_zone_t nr = *p;
 332                 if (nr) {
 333                         *p = 0;
 334                         sysv_free_block(inode->i_sb, nr);
 335                         mark_inode_dirty(inode);
 336                 }
 337         }
 338 }
 339
 340 static void free_branches(struct inode *inode, sysv_zone_t *p, sysv_zone_t *q, int depth)
 341 {
 342         struct buffer_head * bh;
 343         struct super_block *sb = inode->i_sb;
 344
 345         if (depth--) {
 346                 for ( ; p < q ; p++) {
 347                         int block;
 348                         sysv_zone_t nr = *p;
 349                         if (!nr)
 350                                 continue;
 351                         *p = 0;
 352                         block = block_to_cpu(SYSV_SB(sb), nr);
 353                         bh = sb_bread(sb, block);
 354                         if (!bh)
 355                                 continue;
 356                         free_branches(inode, (sysv_zone_t*)bh->b_data,
 357                                         block_end(bh), depth);
 358                         bforget(bh);
 359                         sysv_free_block(sb, nr);
 360                         mark_inode_dirty(inode);
 361                 }
 362         } else
 363                 free_data(inode, p, q);
 364 }
 365
 366 void sysv_truncate (struct inode * inode)
 367 {
 368         sysv_zone_t *i_data = SYSV_I(inode)->i_data;
 369         int offsets[DEPTH];
 370         Indirect chain[DEPTH];
 371         Indirect *partial;
 372         sysv_zone_t nr = 0;
 373         int n;
 374         long iblock;
 375         unsigned blocksize;
 376
 377         if (!(S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
 378             S_ISLNK(inode->i_mode)))
 379                 return;
 380
 381         blocksize = inode->i_sb->s_blocksize;
 382         iblock = (inode->i_size + blocksize-1)
 383                                         >> inode->i_sb->s_blocksize_bits;
 384
 385         block_truncate_page(inode->i_mapping, inode->i_size, get_block);
 386
 387         n = block_to_path(inode, iblock, offsets);
 388         if (n == 0)
 389                 return;
 390
 391         if (n == 1) {
 392                 free_data(inode, i_data+offsets[0], i_data + DIRECT);
 393                 goto do_indirects;
 394         }
 395
 396         partial = find_shared(inode, n, offsets, chain, &nr);
 397         /* Kill the top of shared branch (already detached) */
 398         if (nr) {
 399                 if (partial == chain)
 400                         mark_inode_dirty(inode);
 401                 else
 402                         dirty_indirect(partial->bh, inode);
 403                 free_branches(inode, &nr, &nr+1, (chain+n-1) - partial);
 404         }
 405         /* Clear the ends of indirect blocks on the shared branch */
 406         while (partial > chain) {
 407                 free_branches(inode, partial->p + 1, block_end(partial->bh),
 408                                 (chain+n-1) - partial);
 409                 dirty_indirect(partial->bh, inode);
 410                 brelse (partial->bh);
 411                 partial--;
 412         }
 413 do_indirects:
 414         /* Kill the remaining (whole) subtrees (== subtrees deeper than...) */
 415         while (n < DEPTH) {
 416                 nr = i_data[DIRECT + n - 1];
 417                 if (nr) {
 418                         i_data[DIRECT + n - 1] = 0;
 419                         mark_inode_dirty(inode);
 420                         free_branches(inode, &nr, &nr+1, n);
 421                 }
 422                 n++;
 423         }
 424         inode->i_mtime = inode->i_ctime = current_time(inode);
 425         if (IS_SYNC(inode))
 426                 sysv_sync_inode (inode);
 427         else
 428                 mark_inode_dirty(inode);
 429 }
 430
 431 static unsigned sysv_nblocks(struct super_block *s, loff_t size)
 432 {
 433         struct sysv_sb_info *sbi = SYSV_SB(s);
 434         int ptrs_bits = sbi->s_ind_per_block_bits;
 435         unsigned blocks, res, direct = DIRECT, i = DEPTH;
 436         blocks = (size + s->s_blocksize - 1) >> s->s_blocksize_bits;
 437         res = blocks;
 438         while (--i && blocks > direct) {
 439                 blocks = ((blocks - direct - 1) >> ptrs_bits) + 1;
 440                 res += blocks;
 441                 direct = 1;
 442         }
 443         return res;
 444 }
 445
 446 int sysv_getattr(struct user_namespace *mnt_userns, const struct path *path,
 447                  struct kstat *stat, u32 request_mask, unsigned int flags)
 448 {
 449         struct super_block *s = path->dentry->d_sb;
 450         generic_fillattr(&init_user_ns, d_inode(path->dentry), stat);
 451         stat->blocks = (s->s_blocksize / 512) * sysv_nblocks(s, stat->size);
 452         stat->blksize = s->s_blocksize;
 453         return 0;
 454 }
 455
 456 static int sysv_writepage(struct page *page, struct writeback_control *wbc)
 457 {
 458         return block_write_full_page(page,get_block,wbc);
 459 }
 460
 461 static int sysv_read_folio(struct file *file, struct folio *folio)
 462 {
 463         return block_read_full_folio(folio, get_block);
 464 }
 465
 466 int sysv_prepare_chunk(struct page *page, loff_t pos, unsigned len)
 467 {
 468         return __block_write_begin(page, pos, len, get_block);
 469 }
 470
 471 static void sysv_write_failed(struct address_space *mapping, loff_t to)
 472 {
 473         struct inode *inode = mapping->host;
 474
 475         if (to > inode->i_size) {
 476                 truncate_pagecache(inode, inode->i_size);
 477                 sysv_truncate(inode);
 478         }
 479 }
 480
 481 static int sysv_write_begin(struct file *file, struct address_space *mapping,
 482                         loff_t pos, unsigned len,
 483                         struct page **pagep, void **fsdata)
 484 {
 485         int ret;
 486
 487         ret = block_write_begin(mapping, pos, len, pagep, get_block);
 488         if (unlikely(ret))
 489                 sysv_write_failed(mapping, pos + len);
 490
 491         return ret;
 492 }
 493
 494 static sector_t sysv_bmap(struct address_space *mapping, sector_t block)
 495 {
 496         return generic_block_bmap(mapping,block,get_block);
 497 }
 498
 499 const struct address_space_operations sysv_aops = {
 500         .dirty_folio = block_dirty_folio,
 501         .invalidate_folio = block_invalidate_folio,
 502         .read_folio = sysv_read_folio,
 503         .writepage = sysv_writepage,
 504         .write_begin = sysv_write_begin,
 505         .write_end = generic_write_end,
 506         .bmap = sysv_bmap
 507 };