fs/erofs/zmap.c

   1 // SPDX-License-Identifier: GPL-2.0-only
   2 /*
   3  * Copyright (C) 2018-2019 HUAWEI, Inc.
   4  *             https://www.huawei.com/
   5  */
   6 #include "internal.h"
   7 #include <asm/unaligned.h>
   8 #include <trace/events/erofs.h>
   9
  10 struct z_erofs_maprecorder {
  11         struct inode *inode;
  12         struct erofs_map_blocks *map;
  13         void *kaddr;
  14
  15         unsigned long lcn;
  16         /* compression extent information gathered */
  17         u8  type, headtype;
  18         u16 clusterofs;
  19         u16 delta[2];
  20         erofs_blk_t pblk, compressedblks;
  21         erofs_off_t nextpackoff;
  22         bool partialref;
  23 };
  24
  25 static int z_erofs_load_full_lcluster(struct z_erofs_maprecorder *m,
  26                                       unsigned long lcn)
  27 {
  28         struct inode *const inode = m->inode;
  29         struct erofs_inode *const vi = EROFS_I(inode);
  30         const erofs_off_t pos = Z_EROFS_FULL_INDEX_ALIGN(erofs_iloc(inode) +
  31                         vi->inode_isize + vi->xattr_isize) +
  32                         lcn * sizeof(struct z_erofs_lcluster_index);
  33         struct z_erofs_lcluster_index *di;
  34         unsigned int advise, type;
  35
  36         m->kaddr = erofs_read_metabuf(&m->map->buf, inode->i_sb,
  37                                       erofs_blknr(inode->i_sb, pos), EROFS_KMAP);
  38         if (IS_ERR(m->kaddr))
  39                 return PTR_ERR(m->kaddr);
  40
  41         m->nextpackoff = pos + sizeof(struct z_erofs_lcluster_index);
  42         m->lcn = lcn;
  43         di = m->kaddr + erofs_blkoff(inode->i_sb, pos);
  44
  45         advise = le16_to_cpu(di->di_advise);
  46         type = (advise >> Z_EROFS_LI_LCLUSTER_TYPE_BIT) &
  47                 ((1 << Z_EROFS_LI_LCLUSTER_TYPE_BITS) - 1);
  48         switch (type) {
  49         case Z_EROFS_LCLUSTER_TYPE_NONHEAD:
  50                 m->clusterofs = 1 << vi->z_logical_clusterbits;
  51                 m->delta[0] = le16_to_cpu(di->di_u.delta[0]);
  52                 if (m->delta[0] & Z_EROFS_LI_D0_CBLKCNT) {
  53                         if (!(vi->z_advise & (Z_EROFS_ADVISE_BIG_PCLUSTER_1 |
  54                                         Z_EROFS_ADVISE_BIG_PCLUSTER_2))) {
  55                                 DBG_BUGON(1);
  56                                 return -EFSCORRUPTED;
  57                         }
  58                         m->compressedblks = m->delta[0] &
  59                                 ~Z_EROFS_LI_D0_CBLKCNT;
  60                         m->delta[0] = 1;
  61                 }
  62                 m->delta[1] = le16_to_cpu(di->di_u.delta[1]);
  63                 break;
  64         case Z_EROFS_LCLUSTER_TYPE_PLAIN:
  65         case Z_EROFS_LCLUSTER_TYPE_HEAD1:
  66         case Z_EROFS_LCLUSTER_TYPE_HEAD2:
  67                 if (advise & Z_EROFS_LI_PARTIAL_REF)
  68                         m->partialref = true;
  69                 m->clusterofs = le16_to_cpu(di->di_clusterofs);
  70                 if (m->clusterofs >= 1 << vi->z_logical_clusterbits) {
  71                         DBG_BUGON(1);
  72                         return -EFSCORRUPTED;
  73                 }
  74                 m->pblk = le32_to_cpu(di->di_u.blkaddr);
  75                 break;
  76         default:
  77                 DBG_BUGON(1);
  78                 return -EOPNOTSUPP;
  79         }
  80         m->type = type;
  81         return 0;
  82 }
  83
  84 static unsigned int decode_compactedbits(unsigned int lobits,
  85                                          unsigned int lomask,
  86                                          u8 *in, unsigned int pos, u8 *type)
  87 {
  88         const unsigned int v = get_unaligned_le32(in + pos / 8) >> (pos & 7);
  89         const unsigned int lo = v & lomask;
  90
  91         *type = (v >> lobits) & 3;
  92         return lo;
  93 }
  94
  95 static int get_compacted_la_distance(unsigned int lclusterbits,
  96                                      unsigned int encodebits,
  97                                      unsigned int vcnt, u8 *in, int i)
  98 {
  99         const unsigned int lomask = (1 << lclusterbits) - 1;
 100         unsigned int lo, d1 = 0;
 101         u8 type;
 102
 103         DBG_BUGON(i >= vcnt);
 104
 105         do {
 106                 lo = decode_compactedbits(lclusterbits, lomask,
 107                                           in, encodebits * i, &type);
 108
 109                 if (type != Z_EROFS_LCLUSTER_TYPE_NONHEAD)
 110                         return d1;
 111                 ++d1;
 112         } while (++i < vcnt);
 113
 114         /* vcnt - 1 (Z_EROFS_LCLUSTER_TYPE_NONHEAD) item */
 115         if (!(lo & Z_EROFS_LI_D0_CBLKCNT))
 116                 d1 += lo - 1;
 117         return d1;
 118 }
 119
 120 static int unpack_compacted_index(struct z_erofs_maprecorder *m,
 121                                   unsigned int amortizedshift,
 122                                   erofs_off_t pos, bool lookahead)
 123 {
 124         struct erofs_inode *const vi = EROFS_I(m->inode);
 125         const unsigned int lclusterbits = vi->z_logical_clusterbits;
 126         const unsigned int lomask = (1 << lclusterbits) - 1;
 127         unsigned int vcnt, base, lo, encodebits, nblk, eofs;
 128         int i;
 129         u8 *in, type;
 130         bool big_pcluster;
 131
 132         if (1 << amortizedshift == 4 && lclusterbits <= 14)
 133                 vcnt = 2;
 134         else if (1 << amortizedshift == 2 && lclusterbits == 12)
 135                 vcnt = 16;
 136         else
 137                 return -EOPNOTSUPP;
 138
 139         /* it doesn't equal to round_up(..) */
 140         m->nextpackoff = round_down(pos, vcnt << amortizedshift) +
 141                          (vcnt << amortizedshift);
 142         big_pcluster = vi->z_advise & Z_EROFS_ADVISE_BIG_PCLUSTER_1;
 143         encodebits = ((vcnt << amortizedshift) - sizeof(__le32)) * 8 / vcnt;
 144         eofs = erofs_blkoff(m->inode->i_sb, pos);
 145         base = round_down(eofs, vcnt << amortizedshift);
 146         in = m->kaddr + base;
 147
 148         i = (eofs - base) >> amortizedshift;
 149
 150         lo = decode_compactedbits(lclusterbits, lomask,
 151                                   in, encodebits * i, &type);
 152         m->type = type;
 153         if (type == Z_EROFS_LCLUSTER_TYPE_NONHEAD) {
 154                 m->clusterofs = 1 << lclusterbits;
 155
 156                 /* figure out lookahead_distance: delta[1] if needed */
 157                 if (lookahead)
 158                         m->delta[1] = get_compacted_la_distance(lclusterbits,
 159                                                 encodebits, vcnt, in, i);
 160                 if (lo & Z_EROFS_LI_D0_CBLKCNT) {
 161                         if (!big_pcluster) {
 162                                 DBG_BUGON(1);
 163                                 return -EFSCORRUPTED;
 164                         }
 165                         m->compressedblks = lo & ~Z_EROFS_LI_D0_CBLKCNT;
 166                         m->delta[0] = 1;
 167                         return 0;
 168                 } else if (i + 1 != (int)vcnt) {
 169                         m->delta[0] = lo;
 170                         return 0;
 171                 }
 172                 /*
 173                  * since the last lcluster in the pack is special,
 174                  * of which lo saves delta[1] rather than delta[0].
 175                  * Hence, get delta[0] by the previous lcluster indirectly.
 176                  */
 177                 lo = decode_compactedbits(lclusterbits, lomask,
 178                                           in, encodebits * (i - 1), &type);
 179                 if (type != Z_EROFS_LCLUSTER_TYPE_NONHEAD)
 180                         lo = 0;
 181                 else if (lo & Z_EROFS_LI_D0_CBLKCNT)
 182                         lo = 1;
 183                 m->delta[0] = lo + 1;
 184                 return 0;
 185         }
 186         m->clusterofs = lo;
 187         m->delta[0] = 0;
 188         /* figout out blkaddr (pblk) for HEAD lclusters */
 189         if (!big_pcluster) {
 190                 nblk = 1;
 191                 while (i > 0) {
 192                         --i;
 193                         lo = decode_compactedbits(lclusterbits, lomask,
 194                                                   in, encodebits * i, &type);
 195                         if (type == Z_EROFS_LCLUSTER_TYPE_NONHEAD)
 196                                 i -= lo;
 197
 198                         if (i >= 0)
 199                                 ++nblk;
 200                 }
 201         } else {
 202                 nblk = 0;
 203                 while (i > 0) {
 204                         --i;
 205                         lo = decode_compactedbits(lclusterbits, lomask,
 206                                                   in, encodebits * i, &type);
 207                         if (type == Z_EROFS_LCLUSTER_TYPE_NONHEAD) {
 208                                 if (lo & Z_EROFS_LI_D0_CBLKCNT) {
 209                                         --i;
 210                                         nblk += lo & ~Z_EROFS_LI_D0_CBLKCNT;
 211                                         continue;
 212                                 }
 213                                 /* bigpcluster shouldn't have plain d0 == 1 */
 214                                 if (lo <= 1) {
 215                                         DBG_BUGON(1);
 216                                         return -EFSCORRUPTED;
 217                                 }
 218                                 i -= lo - 2;
 219                                 continue;
 220                         }
 221                         ++nblk;
 222                 }
 223         }
 224         in += (vcnt << amortizedshift) - sizeof(__le32);
 225         m->pblk = le32_to_cpu(*(__le32 *)in) + nblk;
 226         return 0;
 227 }
 228
 229 static int z_erofs_load_compact_lcluster(struct z_erofs_maprecorder *m,
 230                                          unsigned long lcn, bool lookahead)
 231 {
 232         struct inode *const inode = m->inode;
 233         struct erofs_inode *const vi = EROFS_I(inode);
 234         const erofs_off_t ebase = sizeof(struct z_erofs_map_header) +
 235                 ALIGN(erofs_iloc(inode) + vi->inode_isize + vi->xattr_isize, 8);
 236         unsigned int totalidx = erofs_iblks(inode);
 237         unsigned int compacted_4b_initial, compacted_2b;
 238         unsigned int amortizedshift;
 239         erofs_off_t pos;
 240
 241         if (lcn >= totalidx)
 242                 return -EINVAL;
 243
 244         m->lcn = lcn;
 245         /* used to align to 32-byte (compacted_2b) alignment */
 246         compacted_4b_initial = (32 - ebase % 32) / 4;
 247         if (compacted_4b_initial == 32 / 4)
 248                 compacted_4b_initial = 0;
 249
 250         if ((vi->z_advise & Z_EROFS_ADVISE_COMPACTED_2B) &&
 251             compacted_4b_initial < totalidx)
 252                 compacted_2b = rounddown(totalidx - compacted_4b_initial, 16);
 253         else
 254                 compacted_2b = 0;
 255
 256         pos = ebase;
 257         if (lcn < compacted_4b_initial) {
 258                 amortizedshift = 2;
 259                 goto out;
 260         }
 261         pos += compacted_4b_initial * 4;
 262         lcn -= compacted_4b_initial;
 263
 264         if (lcn < compacted_2b) {
 265                 amortizedshift = 1;
 266                 goto out;
 267         }
 268         pos += compacted_2b * 2;
 269         lcn -= compacted_2b;
 270         amortizedshift = 2;
 271 out:
 272         pos += lcn * (1 << amortizedshift);
 273         m->kaddr = erofs_read_metabuf(&m->map->buf, inode->i_sb,
 274                                       erofs_blknr(inode->i_sb, pos), EROFS_KMAP);
 275         if (IS_ERR(m->kaddr))
 276                 return PTR_ERR(m->kaddr);
 277         return unpack_compacted_index(m, amortizedshift, pos, lookahead);
 278 }
 279
 280 static int z_erofs_load_lcluster_from_disk(struct z_erofs_maprecorder *m,
 281                                            unsigned int lcn, bool lookahead)
 282 {
 283         switch (EROFS_I(m->inode)->datalayout) {
 284         case EROFS_INODE_COMPRESSED_FULL:
 285                 return z_erofs_load_full_lcluster(m, lcn);
 286         case EROFS_INODE_COMPRESSED_COMPACT:
 287                 return z_erofs_load_compact_lcluster(m, lcn, lookahead);
 288         default:
 289                 return -EINVAL;
 290         }
 291 }
 292
 293 static int z_erofs_extent_lookback(struct z_erofs_maprecorder *m,
 294                                    unsigned int lookback_distance)
 295 {
 296         struct super_block *sb = m->inode->i_sb;
 297         struct erofs_inode *const vi = EROFS_I(m->inode);
 298         const unsigned int lclusterbits = vi->z_logical_clusterbits;
 299
 300         while (m->lcn >= lookback_distance) {
 301                 unsigned long lcn = m->lcn - lookback_distance;
 302                 int err;
 303
 304                 err = z_erofs_load_lcluster_from_disk(m, lcn, false);
 305                 if (err)
 306                         return err;
 307
 308                 switch (m->type) {
 309                 case Z_EROFS_LCLUSTER_TYPE_NONHEAD:
 310                         lookback_distance = m->delta[0];
 311                         if (!lookback_distance)
 312                                 goto err_bogus;
 313                         continue;
 314                 case Z_EROFS_LCLUSTER_TYPE_PLAIN:
 315                 case Z_EROFS_LCLUSTER_TYPE_HEAD1:
 316                 case Z_EROFS_LCLUSTER_TYPE_HEAD2:
 317                         m->headtype = m->type;
 318                         m->map->m_la = (lcn << lclusterbits) | m->clusterofs;
 319                         return 0;
 320                 default:
 321                         erofs_err(sb, "unknown type %u @ lcn %lu of nid %llu",
 322                                   m->type, lcn, vi->nid);
 323                         DBG_BUGON(1);
 324                         return -EOPNOTSUPP;
 325                 }
 326         }
 327 err_bogus:
 328         erofs_err(sb, "bogus lookback distance %u @ lcn %lu of nid %llu",
 329                   lookback_distance, m->lcn, vi->nid);
 330         DBG_BUGON(1);
 331         return -EFSCORRUPTED;
 332 }
 333
 334 static int z_erofs_get_extent_compressedlen(struct z_erofs_maprecorder *m,
 335                                             unsigned int initial_lcn)
 336 {
 337         struct super_block *sb = m->inode->i_sb;
 338         struct erofs_inode *const vi = EROFS_I(m->inode);
 339         struct erofs_map_blocks *const map = m->map;
 340         const unsigned int lclusterbits = vi->z_logical_clusterbits;
 341         unsigned long lcn;
 342         int err;
 343
 344         DBG_BUGON(m->type != Z_EROFS_LCLUSTER_TYPE_PLAIN &&
 345                   m->type != Z_EROFS_LCLUSTER_TYPE_HEAD1 &&
 346                   m->type != Z_EROFS_LCLUSTER_TYPE_HEAD2);
 347         DBG_BUGON(m->type != m->headtype);
 348
 349         if (m->headtype == Z_EROFS_LCLUSTER_TYPE_PLAIN ||
 350             ((m->headtype == Z_EROFS_LCLUSTER_TYPE_HEAD1) &&
 351              !(vi->z_advise & Z_EROFS_ADVISE_BIG_PCLUSTER_1)) ||
 352             ((m->headtype == Z_EROFS_LCLUSTER_TYPE_HEAD2) &&
 353              !(vi->z_advise & Z_EROFS_ADVISE_BIG_PCLUSTER_2))) {
 354                 map->m_plen = 1ULL << lclusterbits;
 355                 return 0;
 356         }
 357         lcn = m->lcn + 1;
 358         if (m->compressedblks)
 359                 goto out;
 360
 361         err = z_erofs_load_lcluster_from_disk(m, lcn, false);
 362         if (err)
 363                 return err;
 364
 365         /*
 366          * If the 1st NONHEAD lcluster has already been handled initially w/o
 367          * valid compressedblks, which means at least it mustn't be CBLKCNT, or
 368          * an internal implemenatation error is detected.
 369          *
 370          * The following code can also handle it properly anyway, but let's
 371          * BUG_ON in the debugging mode only for developers to notice that.
 372          */
 373         DBG_BUGON(lcn == initial_lcn &&
 374                   m->type == Z_EROFS_LCLUSTER_TYPE_NONHEAD);
 375
 376         switch (m->type) {
 377         case Z_EROFS_LCLUSTER_TYPE_PLAIN:
 378         case Z_EROFS_LCLUSTER_TYPE_HEAD1:
 379         case Z_EROFS_LCLUSTER_TYPE_HEAD2:
 380                 /*
 381                  * if the 1st NONHEAD lcluster is actually PLAIN or HEAD type
 382                  * rather than CBLKCNT, it's a 1 lcluster-sized pcluster.
 383                  */
 384                 m->compressedblks = 1 << (lclusterbits - sb->s_blocksize_bits);
 385                 break;
 386         case Z_EROFS_LCLUSTER_TYPE_NONHEAD:
 387                 if (m->delta[0] != 1)
 388                         goto err_bonus_cblkcnt;
 389                 if (m->compressedblks)
 390                         break;
 391                 fallthrough;
 392         default:
 393                 erofs_err(sb, "cannot found CBLKCNT @ lcn %lu of nid %llu", lcn,
 394                           vi->nid);
 395                 DBG_BUGON(1);
 396                 return -EFSCORRUPTED;
 397         }
 398 out:
 399         map->m_plen = erofs_pos(sb, m->compressedblks);
 400         return 0;
 401 err_bonus_cblkcnt:
 402         erofs_err(sb, "bogus CBLKCNT @ lcn %lu of nid %llu", lcn, vi->nid);
 403         DBG_BUGON(1);
 404         return -EFSCORRUPTED;
 405 }
 406
 407 static int z_erofs_get_extent_decompressedlen(struct z_erofs_maprecorder *m)
 408 {
 409         struct inode *inode = m->inode;
 410         struct erofs_inode *vi = EROFS_I(inode);
 411         struct erofs_map_blocks *map = m->map;
 412         unsigned int lclusterbits = vi->z_logical_clusterbits;
 413         u64 lcn = m->lcn, headlcn = map->m_la >> lclusterbits;
 414         int err;
 415
 416         do {
 417                 /* handle the last EOF pcluster (no next HEAD lcluster) */
 418                 if ((lcn << lclusterbits) >= inode->i_size) {
 419                         map->m_llen = inode->i_size - map->m_la;
 420                         return 0;
 421                 }
 422
 423                 err = z_erofs_load_lcluster_from_disk(m, lcn, true);
 424                 if (err)
 425                         return err;
 426
 427                 if (m->type == Z_EROFS_LCLUSTER_TYPE_NONHEAD) {
 428                         DBG_BUGON(!m->delta[1] &&
 429                                   m->clusterofs != 1 << lclusterbits);
 430                 } else if (m->type == Z_EROFS_LCLUSTER_TYPE_PLAIN ||
 431                            m->type == Z_EROFS_LCLUSTER_TYPE_HEAD1 ||
 432                            m->type == Z_EROFS_LCLUSTER_TYPE_HEAD2) {
 433                         /* go on until the next HEAD lcluster */
 434                         if (lcn != headlcn)
 435                                 break;
 436                         m->delta[1] = 1;
 437                 } else {
 438                         erofs_err(inode->i_sb, "unknown type %u @ lcn %llu of nid %llu",
 439                                   m->type, lcn, vi->nid);
 440                         DBG_BUGON(1);
 441                         return -EOPNOTSUPP;
 442                 }
 443                 lcn += m->delta[1];
 444         } while (m->delta[1]);
 445
 446         map->m_llen = (lcn << lclusterbits) + m->clusterofs - map->m_la;
 447         return 0;
 448 }
 449
 450 static int z_erofs_do_map_blocks(struct inode *inode,
 451                                  struct erofs_map_blocks *map, int flags)
 452 {
 453         struct erofs_inode *const vi = EROFS_I(inode);
 454         bool ztailpacking = vi->z_advise & Z_EROFS_ADVISE_INLINE_PCLUSTER;
 455         bool fragment = vi->z_advise & Z_EROFS_ADVISE_FRAGMENT_PCLUSTER;
 456         struct z_erofs_maprecorder m = {
 457                 .inode = inode,
 458                 .map = map,
 459         };
 460         int err = 0;
 461         unsigned int lclusterbits, endoff;
 462         unsigned long initial_lcn;
 463         unsigned long long ofs, end;
 464
 465         lclusterbits = vi->z_logical_clusterbits;
 466         ofs = flags & EROFS_GET_BLOCKS_FINDTAIL ? inode->i_size - 1 : map->m_la;
 467         initial_lcn = ofs >> lclusterbits;
 468         endoff = ofs & ((1 << lclusterbits) - 1);
 469
 470         err = z_erofs_load_lcluster_from_disk(&m, initial_lcn, false);
 471         if (err)
 472                 goto unmap_out;
 473
 474         if (ztailpacking && (flags & EROFS_GET_BLOCKS_FINDTAIL))
 475                 vi->z_idataoff = m.nextpackoff;
 476
 477         map->m_flags = EROFS_MAP_MAPPED | EROFS_MAP_ENCODED;
 478         end = (m.lcn + 1ULL) << lclusterbits;
 479
 480         switch (m.type) {
 481         case Z_EROFS_LCLUSTER_TYPE_PLAIN:
 482         case Z_EROFS_LCLUSTER_TYPE_HEAD1:
 483         case Z_EROFS_LCLUSTER_TYPE_HEAD2:
 484                 if (endoff >= m.clusterofs) {
 485                         m.headtype = m.type;
 486                         map->m_la = (m.lcn << lclusterbits) | m.clusterofs;
 487                         /*
 488                          * For ztailpacking files, in order to inline data more
 489                          * effectively, special EOF lclusters are now supported
 490                          * which can have three parts at most.
 491                          */
 492                         if (ztailpacking && end > inode->i_size)
 493                                 end = inode->i_size;
 494                         break;
 495                 }
 496                 /* m.lcn should be >= 1 if endoff < m.clusterofs */
 497                 if (!m.lcn) {
 498                         erofs_err(inode->i_sb,
 499                                   "invalid logical cluster 0 at nid %llu",
 500                                   vi->nid);
 501                         err = -EFSCORRUPTED;
 502                         goto unmap_out;
 503                 }
 504                 end = (m.lcn << lclusterbits) | m.clusterofs;
 505                 map->m_flags |= EROFS_MAP_FULL_MAPPED;
 506                 m.delta[0] = 1;
 507                 fallthrough;
 508         case Z_EROFS_LCLUSTER_TYPE_NONHEAD:
 509                 /* get the corresponding first chunk */
 510                 err = z_erofs_extent_lookback(&m, m.delta[0]);
 511                 if (err)
 512                         goto unmap_out;
 513                 break;
 514         default:
 515                 erofs_err(inode->i_sb,
 516                           "unknown type %u @ offset %llu of nid %llu",
 517                           m.type, ofs, vi->nid);
 518                 err = -EOPNOTSUPP;
 519                 goto unmap_out;
 520         }
 521         if (m.partialref)
 522                 map->m_flags |= EROFS_MAP_PARTIAL_REF;
 523         map->m_llen = end - map->m_la;
 524
 525         if (flags & EROFS_GET_BLOCKS_FINDTAIL) {
 526                 vi->z_tailextent_headlcn = m.lcn;
 527                 /* for non-compact indexes, fragmentoff is 64 bits */
 528                 if (fragment && vi->datalayout == EROFS_INODE_COMPRESSED_FULL)
 529                         vi->z_fragmentoff |= (u64)m.pblk << 32;
 530         }
 531         if (ztailpacking && m.lcn == vi->z_tailextent_headlcn) {
 532                 map->m_flags |= EROFS_MAP_META;
 533                 map->m_pa = vi->z_idataoff;
 534                 map->m_plen = vi->z_idata_size;
 535         } else if (fragment && m.lcn == vi->z_tailextent_headlcn) {
 536                 map->m_flags |= EROFS_MAP_FRAGMENT;
 537         } else {
 538                 map->m_pa = erofs_pos(inode->i_sb, m.pblk);
 539                 err = z_erofs_get_extent_compressedlen(&m, initial_lcn);
 540                 if (err)
 541                         goto unmap_out;
 542         }
 543
 544         if (m.headtype == Z_EROFS_LCLUSTER_TYPE_PLAIN) {
 545                 if (map->m_llen > map->m_plen) {
 546                         DBG_BUGON(1);
 547                         err = -EFSCORRUPTED;
 548                         goto unmap_out;
 549                 }
 550                 if (vi->z_advise & Z_EROFS_ADVISE_INTERLACED_PCLUSTER)
 551                         map->m_algorithmformat =
 552                                 Z_EROFS_COMPRESSION_INTERLACED;
 553                 else
 554                         map->m_algorithmformat =
 555                                 Z_EROFS_COMPRESSION_SHIFTED;
 556         } else if (m.headtype == Z_EROFS_LCLUSTER_TYPE_HEAD2) {
 557                 map->m_algorithmformat = vi->z_algorithmtype[1];
 558         } else {
 559                 map->m_algorithmformat = vi->z_algorithmtype[0];
 560         }
 561
 562         if ((flags & EROFS_GET_BLOCKS_FIEMAP) ||
 563             ((flags & EROFS_GET_BLOCKS_READMORE) &&
 564              map->m_algorithmformat == Z_EROFS_COMPRESSION_LZMA &&
 565              map->m_llen >= i_blocksize(inode))) {
 566                 err = z_erofs_get_extent_decompressedlen(&m);
 567                 if (!err)
 568                         map->m_flags |= EROFS_MAP_FULL_MAPPED;
 569         }
 570
 571 unmap_out:
 572         erofs_unmap_metabuf(&m.map->buf);
 573         return err;
 574 }
 575
 576 static int z_erofs_fill_inode_lazy(struct inode *inode)
 577 {
 578         struct erofs_inode *const vi = EROFS_I(inode);
 579         struct super_block *const sb = inode->i_sb;
 580         int err, headnr;
 581         erofs_off_t pos;
 582         struct erofs_buf buf = __EROFS_BUF_INITIALIZER;
 583         void *kaddr;
 584         struct z_erofs_map_header *h;
 585
 586         if (test_bit(EROFS_I_Z_INITED_BIT, &vi->flags)) {
 587                 /*
 588                  * paired with smp_mb() at the end of the function to ensure
 589                  * fields will only be observed after the bit is set.
 590                  */
 591                 smp_mb();
 592                 return 0;
 593         }
 594
 595         if (wait_on_bit_lock(&vi->flags, EROFS_I_BL_Z_BIT, TASK_KILLABLE))
 596                 return -ERESTARTSYS;
 597
 598         err = 0;
 599         if (test_bit(EROFS_I_Z_INITED_BIT, &vi->flags))
 600                 goto out_unlock;
 601
 602         pos = ALIGN(erofs_iloc(inode) + vi->inode_isize + vi->xattr_isize, 8);
 603         kaddr = erofs_read_metabuf(&buf, sb, erofs_blknr(sb, pos), EROFS_KMAP);
 604         if (IS_ERR(kaddr)) {
 605                 err = PTR_ERR(kaddr);
 606                 goto out_unlock;
 607         }
 608
 609         h = kaddr + erofs_blkoff(sb, pos);
 610         /*
 611          * if the highest bit of the 8-byte map header is set, the whole file
 612          * is stored in the packed inode. The rest bits keeps z_fragmentoff.
 613          */
 614         if (h->h_clusterbits >> Z_EROFS_FRAGMENT_INODE_BIT) {
 615                 vi->z_advise = Z_EROFS_ADVISE_FRAGMENT_PCLUSTER;
 616                 vi->z_fragmentoff = le64_to_cpu(*(__le64 *)h) ^ (1ULL << 63);
 617                 vi->z_tailextent_headlcn = 0;
 618                 goto done;
 619         }
 620         vi->z_advise = le16_to_cpu(h->h_advise);
 621         vi->z_algorithmtype[0] = h->h_algorithmtype & 15;
 622         vi->z_algorithmtype[1] = h->h_algorithmtype >> 4;
 623
 624         headnr = 0;
 625         if (vi->z_algorithmtype[0] >= Z_EROFS_COMPRESSION_MAX ||
 626             vi->z_algorithmtype[++headnr] >= Z_EROFS_COMPRESSION_MAX) {
 627                 erofs_err(sb, "unknown HEAD%u format %u for nid %llu, please upgrade kernel",
 628                           headnr + 1, vi->z_algorithmtype[headnr], vi->nid);
 629                 err = -EOPNOTSUPP;
 630                 goto out_put_metabuf;
 631         }
 632
 633         vi->z_logical_clusterbits = sb->s_blocksize_bits + (h->h_clusterbits & 7);
 634         if (!erofs_sb_has_big_pcluster(EROFS_SB(sb)) &&
 635             vi->z_advise & (Z_EROFS_ADVISE_BIG_PCLUSTER_1 |
 636                             Z_EROFS_ADVISE_BIG_PCLUSTER_2)) {
 637                 erofs_err(sb, "per-inode big pcluster without sb feature for nid %llu",
 638                           vi->nid);
 639                 err = -EFSCORRUPTED;
 640                 goto out_put_metabuf;
 641         }
 642         if (vi->datalayout == EROFS_INODE_COMPRESSED_COMPACT &&
 643             !(vi->z_advise & Z_EROFS_ADVISE_BIG_PCLUSTER_1) ^
 644             !(vi->z_advise & Z_EROFS_ADVISE_BIG_PCLUSTER_2)) {
 645                 erofs_err(sb, "big pcluster head1/2 of compact indexes should be consistent for nid %llu",
 646                           vi->nid);
 647                 err = -EFSCORRUPTED;
 648                 goto out_put_metabuf;
 649         }
 650
 651         if (vi->z_advise & Z_EROFS_ADVISE_INLINE_PCLUSTER) {
 652                 struct erofs_map_blocks map = {
 653                         .buf = __EROFS_BUF_INITIALIZER
 654                 };
 655
 656                 vi->z_idata_size = le16_to_cpu(h->h_idata_size);
 657                 err = z_erofs_do_map_blocks(inode, &map,
 658                                             EROFS_GET_BLOCKS_FINDTAIL);
 659                 erofs_put_metabuf(&map.buf);
 660
 661                 if (!map.m_plen ||
 662                     erofs_blkoff(sb, map.m_pa) + map.m_plen > sb->s_blocksize) {
 663                         erofs_err(sb, "invalid tail-packing pclustersize %llu",
 664                                   map.m_plen);
 665                         err = -EFSCORRUPTED;
 666                 }
 667                 if (err < 0)
 668                         goto out_put_metabuf;
 669         }
 670
 671         if (vi->z_advise & Z_EROFS_ADVISE_FRAGMENT_PCLUSTER &&
 672             !(h->h_clusterbits >> Z_EROFS_FRAGMENT_INODE_BIT)) {
 673                 struct erofs_map_blocks map = {
 674                         .buf = __EROFS_BUF_INITIALIZER
 675                 };
 676
 677                 vi->z_fragmentoff = le32_to_cpu(h->h_fragmentoff);
 678                 err = z_erofs_do_map_blocks(inode, &map,
 679                                             EROFS_GET_BLOCKS_FINDTAIL);
 680                 erofs_put_metabuf(&map.buf);
 681                 if (err < 0)
 682                         goto out_put_metabuf;
 683         }
 684 done:
 685         /* paired with smp_mb() at the beginning of the function */
 686         smp_mb();
 687         set_bit(EROFS_I_Z_INITED_BIT, &vi->flags);
 688 out_put_metabuf:
 689         erofs_put_metabuf(&buf);
 690 out_unlock:
 691         clear_and_wake_up_bit(EROFS_I_BL_Z_BIT, &vi->flags);
 692         return err;
 693 }
 694
 695 int z_erofs_map_blocks_iter(struct inode *inode, struct erofs_map_blocks *map,
 696                             int flags)
 697 {
 698         struct erofs_inode *const vi = EROFS_I(inode);
 699         int err = 0;
 700
 701         trace_z_erofs_map_blocks_iter_enter(inode, map, flags);
 702
 703         /* when trying to read beyond EOF, leave it unmapped */
 704         if (map->m_la >= inode->i_size) {
 705                 map->m_llen = map->m_la + 1 - inode->i_size;
 706                 map->m_la = inode->i_size;
 707                 map->m_flags = 0;
 708                 goto out;
 709         }
 710
 711         err = z_erofs_fill_inode_lazy(inode);
 712         if (err)
 713                 goto out;
 714
 715         if ((vi->z_advise & Z_EROFS_ADVISE_FRAGMENT_PCLUSTER) &&
 716             !vi->z_tailextent_headlcn) {
 717                 map->m_la = 0;
 718                 map->m_llen = inode->i_size;
 719                 map->m_flags = EROFS_MAP_MAPPED | EROFS_MAP_FULL_MAPPED |
 720                                 EROFS_MAP_FRAGMENT;
 721                 goto out;
 722         }
 723
 724         err = z_erofs_do_map_blocks(inode, map, flags);
 725 out:
 726         trace_z_erofs_map_blocks_iter_exit(inode, map, flags, err);
 727         return err;
 728 }
 729
 730 static int z_erofs_iomap_begin_report(struct inode *inode, loff_t offset,
 731                                 loff_t length, unsigned int flags,
 732                                 struct iomap *iomap, struct iomap *srcmap)
 733 {
 734         int ret;
 735         struct erofs_map_blocks map = { .m_la = offset };
 736
 737         ret = z_erofs_map_blocks_iter(inode, &map, EROFS_GET_BLOCKS_FIEMAP);
 738         erofs_put_metabuf(&map.buf);
 739         if (ret < 0)
 740                 return ret;
 741
 742         iomap->bdev = inode->i_sb->s_bdev;
 743         iomap->offset = map.m_la;
 744         iomap->length = map.m_llen;
 745         if (map.m_flags & EROFS_MAP_MAPPED) {
 746                 iomap->type = IOMAP_MAPPED;
 747                 iomap->addr = map.m_flags & EROFS_MAP_FRAGMENT ?
 748                               IOMAP_NULL_ADDR : map.m_pa;
 749         } else {
 750                 iomap->type = IOMAP_HOLE;
 751                 iomap->addr = IOMAP_NULL_ADDR;
 752                 /*
 753                  * No strict rule on how to describe extents for post EOF, yet
 754                  * we need to do like below. Otherwise, iomap itself will get
 755                  * into an endless loop on post EOF.
 756                  *
 757                  * Calculate the effective offset by subtracting extent start
 758                  * (map.m_la) from the requested offset, and add it to length.
 759                  * (NB: offset >= map.m_la always)
 760                  */
 761                 if (iomap->offset >= inode->i_size)
 762                         iomap->length = length + offset - map.m_la;
 763         }
 764         iomap->flags = 0;
 765         return 0;
 766 }
 767
 768 const struct iomap_ops z_erofs_iomap_report_ops = {
 769         .iomap_begin = z_erofs_iomap_begin_report,
 770 };