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 static int z_erofs_do_map_blocks(struct inode *inode,
  11                                  struct erofs_map_blocks *map,
  12                                  int flags);
  13
  14 int z_erofs_fill_inode(struct inode *inode)
  15 {
  16         struct erofs_inode *const vi = EROFS_I(inode);
  17         struct erofs_sb_info *sbi = EROFS_SB(inode->i_sb);
  18
  19         if (!erofs_sb_has_big_pcluster(sbi) &&
  20             !erofs_sb_has_ztailpacking(sbi) &&
  21             vi->datalayout == EROFS_INODE_FLAT_COMPRESSION_LEGACY) {
  22                 vi->z_advise = 0;
  23                 vi->z_algorithmtype[0] = 0;
  24                 vi->z_algorithmtype[1] = 0;
  25                 vi->z_logical_clusterbits = LOG_BLOCK_SIZE;
  26                 set_bit(EROFS_I_Z_INITED_BIT, &vi->flags);
  27         }
  28         inode->i_mapping->a_ops = &z_erofs_aops;
  29         return 0;
  30 }
  31
  32 static int z_erofs_fill_inode_lazy(struct inode *inode)
  33 {
  34         struct erofs_inode *const vi = EROFS_I(inode);
  35         struct super_block *const sb = inode->i_sb;
  36         int err, headnr;
  37         erofs_off_t pos;
  38         struct erofs_buf buf = __EROFS_BUF_INITIALIZER;
  39         void *kaddr;
  40         struct z_erofs_map_header *h;
  41
  42         if (test_bit(EROFS_I_Z_INITED_BIT, &vi->flags)) {
  43                 /*
  44                  * paired with smp_mb() at the end of the function to ensure
  45                  * fields will only be observed after the bit is set.
  46                  */
  47                 smp_mb();
  48                 return 0;
  49         }
  50
  51         if (wait_on_bit_lock(&vi->flags, EROFS_I_BL_Z_BIT, TASK_KILLABLE))
  52                 return -ERESTARTSYS;
  53
  54         err = 0;
  55         if (test_bit(EROFS_I_Z_INITED_BIT, &vi->flags))
  56                 goto out_unlock;
  57
  58         DBG_BUGON(!erofs_sb_has_big_pcluster(EROFS_SB(sb)) &&
  59                   !erofs_sb_has_ztailpacking(EROFS_SB(sb)) &&
  60                   vi->datalayout == EROFS_INODE_FLAT_COMPRESSION_LEGACY);
  61
  62         pos = ALIGN(iloc(EROFS_SB(sb), vi->nid) + vi->inode_isize +
  63                     vi->xattr_isize, 8);
  64         kaddr = erofs_read_metabuf(&buf, sb, erofs_blknr(pos),
  65                                    EROFS_KMAP_ATOMIC);
  66         if (IS_ERR(kaddr)) {
  67                 err = PTR_ERR(kaddr);
  68                 goto out_unlock;
  69         }
  70
  71         h = kaddr + erofs_blkoff(pos);
  72         vi->z_advise = le16_to_cpu(h->h_advise);
  73         vi->z_algorithmtype[0] = h->h_algorithmtype & 15;
  74         vi->z_algorithmtype[1] = h->h_algorithmtype >> 4;
  75
  76         headnr = 0;
  77         if (vi->z_algorithmtype[0] >= Z_EROFS_COMPRESSION_MAX ||
  78             vi->z_algorithmtype[++headnr] >= Z_EROFS_COMPRESSION_MAX) {
  79                 erofs_err(sb, "unknown HEAD%u format %u for nid %llu, please upgrade kernel",
  80                           headnr + 1, vi->z_algorithmtype[headnr], vi->nid);
  81                 err = -EOPNOTSUPP;
  82                 goto unmap_done;
  83         }
  84
  85         vi->z_logical_clusterbits = LOG_BLOCK_SIZE + (h->h_clusterbits & 7);
  86         if (!erofs_sb_has_big_pcluster(EROFS_SB(sb)) &&
  87             vi->z_advise & (Z_EROFS_ADVISE_BIG_PCLUSTER_1 |
  88                             Z_EROFS_ADVISE_BIG_PCLUSTER_2)) {
  89                 erofs_err(sb, "per-inode big pcluster without sb feature for nid %llu",
  90                           vi->nid);
  91                 err = -EFSCORRUPTED;
  92                 goto unmap_done;
  93         }
  94         if (vi->datalayout == EROFS_INODE_FLAT_COMPRESSION &&
  95             !(vi->z_advise & Z_EROFS_ADVISE_BIG_PCLUSTER_1) ^
  96             !(vi->z_advise & Z_EROFS_ADVISE_BIG_PCLUSTER_2)) {
  97                 erofs_err(sb, "big pcluster head1/2 of compact indexes should be consistent for nid %llu",
  98                           vi->nid);
  99                 err = -EFSCORRUPTED;
 100                 goto unmap_done;
 101         }
 102 unmap_done:
 103         erofs_put_metabuf(&buf);
 104         if (err)
 105                 goto out_unlock;
 106
 107         if (vi->z_advise & Z_EROFS_ADVISE_INLINE_PCLUSTER) {
 108                 struct erofs_map_blocks map = {
 109                         .buf = __EROFS_BUF_INITIALIZER
 110                 };
 111
 112                 vi->z_idata_size = le16_to_cpu(h->h_idata_size);
 113                 err = z_erofs_do_map_blocks(inode, &map,
 114                                             EROFS_GET_BLOCKS_FINDTAIL);
 115                 erofs_put_metabuf(&map.buf);
 116
 117                 if (!map.m_plen ||
 118                     erofs_blkoff(map.m_pa) + map.m_plen > EROFS_BLKSIZ) {
 119                         erofs_err(sb, "invalid tail-packing pclustersize %llu",
 120                                   map.m_plen);
 121                         err = -EFSCORRUPTED;
 122                 }
 123                 if (err < 0)
 124                         goto out_unlock;
 125         }
 126         /* paired with smp_mb() at the beginning of the function */
 127         smp_mb();
 128         set_bit(EROFS_I_Z_INITED_BIT, &vi->flags);
 129 out_unlock:
 130         clear_and_wake_up_bit(EROFS_I_BL_Z_BIT, &vi->flags);
 131         return err;
 132 }
 133
 134 struct z_erofs_maprecorder {
 135         struct inode *inode;
 136         struct erofs_map_blocks *map;
 137         void *kaddr;
 138
 139         unsigned long lcn;
 140         /* compression extent information gathered */
 141         u8  type, headtype;
 142         u16 clusterofs;
 143         u16 delta[2];
 144         erofs_blk_t pblk, compressedlcs;
 145         erofs_off_t nextpackoff;
 146 };
 147
 148 static int z_erofs_reload_indexes(struct z_erofs_maprecorder *m,
 149                                   erofs_blk_t eblk)
 150 {
 151         struct super_block *const sb = m->inode->i_sb;
 152
 153         m->kaddr = erofs_read_metabuf(&m->map->buf, sb, eblk,
 154                                       EROFS_KMAP_ATOMIC);
 155         if (IS_ERR(m->kaddr))
 156                 return PTR_ERR(m->kaddr);
 157         return 0;
 158 }
 159
 160 static int legacy_load_cluster_from_disk(struct z_erofs_maprecorder *m,
 161                                          unsigned long lcn)
 162 {
 163         struct inode *const inode = m->inode;
 164         struct erofs_inode *const vi = EROFS_I(inode);
 165         const erofs_off_t ibase = iloc(EROFS_I_SB(inode), vi->nid);
 166         const erofs_off_t pos =
 167                 Z_EROFS_VLE_LEGACY_INDEX_ALIGN(ibase + vi->inode_isize +
 168                                                vi->xattr_isize) +
 169                 lcn * sizeof(struct z_erofs_vle_decompressed_index);
 170         struct z_erofs_vle_decompressed_index *di;
 171         unsigned int advise, type;
 172         int err;
 173
 174         err = z_erofs_reload_indexes(m, erofs_blknr(pos));
 175         if (err)
 176                 return err;
 177
 178         m->nextpackoff = pos + sizeof(struct z_erofs_vle_decompressed_index);
 179         m->lcn = lcn;
 180         di = m->kaddr + erofs_blkoff(pos);
 181
 182         advise = le16_to_cpu(di->di_advise);
 183         type = (advise >> Z_EROFS_VLE_DI_CLUSTER_TYPE_BIT) &
 184                 ((1 << Z_EROFS_VLE_DI_CLUSTER_TYPE_BITS) - 1);
 185         switch (type) {
 186         case Z_EROFS_VLE_CLUSTER_TYPE_NONHEAD:
 187                 m->clusterofs = 1 << vi->z_logical_clusterbits;
 188                 m->delta[0] = le16_to_cpu(di->di_u.delta[0]);
 189                 if (m->delta[0] & Z_EROFS_VLE_DI_D0_CBLKCNT) {
 190                         if (!(vi->z_advise & (Z_EROFS_ADVISE_BIG_PCLUSTER_1 |
 191                                         Z_EROFS_ADVISE_BIG_PCLUSTER_2))) {
 192                                 DBG_BUGON(1);
 193                                 return -EFSCORRUPTED;
 194                         }
 195                         m->compressedlcs = m->delta[0] &
 196                                 ~Z_EROFS_VLE_DI_D0_CBLKCNT;
 197                         m->delta[0] = 1;
 198                 }
 199                 m->delta[1] = le16_to_cpu(di->di_u.delta[1]);
 200                 break;
 201         case Z_EROFS_VLE_CLUSTER_TYPE_PLAIN:
 202         case Z_EROFS_VLE_CLUSTER_TYPE_HEAD1:
 203         case Z_EROFS_VLE_CLUSTER_TYPE_HEAD2:
 204                 m->clusterofs = le16_to_cpu(di->di_clusterofs);
 205                 m->pblk = le32_to_cpu(di->di_u.blkaddr);
 206                 break;
 207         default:
 208                 DBG_BUGON(1);
 209                 return -EOPNOTSUPP;
 210         }
 211         m->type = type;
 212         return 0;
 213 }
 214
 215 static unsigned int decode_compactedbits(unsigned int lobits,
 216                                          unsigned int lomask,
 217                                          u8 *in, unsigned int pos, u8 *type)
 218 {
 219         const unsigned int v = get_unaligned_le32(in + pos / 8) >> (pos & 7);
 220         const unsigned int lo = v & lomask;
 221
 222         *type = (v >> lobits) & 3;
 223         return lo;
 224 }
 225
 226 static int get_compacted_la_distance(unsigned int lclusterbits,
 227                                      unsigned int encodebits,
 228                                      unsigned int vcnt, u8 *in, int i)
 229 {
 230         const unsigned int lomask = (1 << lclusterbits) - 1;
 231         unsigned int lo, d1 = 0;
 232         u8 type;
 233
 234         DBG_BUGON(i >= vcnt);
 235
 236         do {
 237                 lo = decode_compactedbits(lclusterbits, lomask,
 238                                           in, encodebits * i, &type);
 239
 240                 if (type != Z_EROFS_VLE_CLUSTER_TYPE_NONHEAD)
 241                         return d1;
 242                 ++d1;
 243         } while (++i < vcnt);
 244
 245         /* vcnt - 1 (Z_EROFS_VLE_CLUSTER_TYPE_NONHEAD) item */
 246         if (!(lo & Z_EROFS_VLE_DI_D0_CBLKCNT))
 247                 d1 += lo - 1;
 248         return d1;
 249 }
 250
 251 static int unpack_compacted_index(struct z_erofs_maprecorder *m,
 252                                   unsigned int amortizedshift,
 253                                   erofs_off_t pos, bool lookahead)
 254 {
 255         struct erofs_inode *const vi = EROFS_I(m->inode);
 256         const unsigned int lclusterbits = vi->z_logical_clusterbits;
 257         const unsigned int lomask = (1 << lclusterbits) - 1;
 258         unsigned int vcnt, base, lo, encodebits, nblk, eofs;
 259         int i;
 260         u8 *in, type;
 261         bool big_pcluster;
 262
 263         if (1 << amortizedshift == 4)
 264                 vcnt = 2;
 265         else if (1 << amortizedshift == 2 && lclusterbits == 12)
 266                 vcnt = 16;
 267         else
 268                 return -EOPNOTSUPP;
 269
 270         /* it doesn't equal to round_up(..) */
 271         m->nextpackoff = round_down(pos, vcnt << amortizedshift) +
 272                          (vcnt << amortizedshift);
 273         big_pcluster = vi->z_advise & Z_EROFS_ADVISE_BIG_PCLUSTER_1;
 274         encodebits = ((vcnt << amortizedshift) - sizeof(__le32)) * 8 / vcnt;
 275         eofs = erofs_blkoff(pos);
 276         base = round_down(eofs, vcnt << amortizedshift);
 277         in = m->kaddr + base;
 278
 279         i = (eofs - base) >> amortizedshift;
 280
 281         lo = decode_compactedbits(lclusterbits, lomask,
 282                                   in, encodebits * i, &type);
 283         m->type = type;
 284         if (type == Z_EROFS_VLE_CLUSTER_TYPE_NONHEAD) {
 285                 m->clusterofs = 1 << lclusterbits;
 286
 287                 /* figure out lookahead_distance: delta[1] if needed */
 288                 if (lookahead)
 289                         m->delta[1] = get_compacted_la_distance(lclusterbits,
 290                                                 encodebits, vcnt, in, i);
 291                 if (lo & Z_EROFS_VLE_DI_D0_CBLKCNT) {
 292                         if (!big_pcluster) {
 293                                 DBG_BUGON(1);
 294                                 return -EFSCORRUPTED;
 295                         }
 296                         m->compressedlcs = lo & ~Z_EROFS_VLE_DI_D0_CBLKCNT;
 297                         m->delta[0] = 1;
 298                         return 0;
 299                 } else if (i + 1 != (int)vcnt) {
 300                         m->delta[0] = lo;
 301                         return 0;
 302                 }
 303                 /*
 304                  * since the last lcluster in the pack is special,
 305                  * of which lo saves delta[1] rather than delta[0].
 306                  * Hence, get delta[0] by the previous lcluster indirectly.
 307                  */
 308                 lo = decode_compactedbits(lclusterbits, lomask,
 309                                           in, encodebits * (i - 1), &type);
 310                 if (type != Z_EROFS_VLE_CLUSTER_TYPE_NONHEAD)
 311                         lo = 0;
 312                 else if (lo & Z_EROFS_VLE_DI_D0_CBLKCNT)
 313                         lo = 1;
 314                 m->delta[0] = lo + 1;
 315                 return 0;
 316         }
 317         m->clusterofs = lo;
 318         m->delta[0] = 0;
 319         /* figout out blkaddr (pblk) for HEAD lclusters */
 320         if (!big_pcluster) {
 321                 nblk = 1;
 322                 while (i > 0) {
 323                         --i;
 324                         lo = decode_compactedbits(lclusterbits, lomask,
 325                                                   in, encodebits * i, &type);
 326                         if (type == Z_EROFS_VLE_CLUSTER_TYPE_NONHEAD)
 327                                 i -= lo;
 328
 329                         if (i >= 0)
 330                                 ++nblk;
 331                 }
 332         } else {
 333                 nblk = 0;
 334                 while (i > 0) {
 335                         --i;
 336                         lo = decode_compactedbits(lclusterbits, lomask,
 337                                                   in, encodebits * i, &type);
 338                         if (type == Z_EROFS_VLE_CLUSTER_TYPE_NONHEAD) {
 339                                 if (lo & Z_EROFS_VLE_DI_D0_CBLKCNT) {
 340                                         --i;
 341                                         nblk += lo & ~Z_EROFS_VLE_DI_D0_CBLKCNT;
 342                                         continue;
 343                                 }
 344                                 /* bigpcluster shouldn't have plain d0 == 1 */
 345                                 if (lo <= 1) {
 346                                         DBG_BUGON(1);
 347                                         return -EFSCORRUPTED;
 348                                 }
 349                                 i -= lo - 2;
 350                                 continue;
 351                         }
 352                         ++nblk;
 353                 }
 354         }
 355         in += (vcnt << amortizedshift) - sizeof(__le32);
 356         m->pblk = le32_to_cpu(*(__le32 *)in) + nblk;
 357         return 0;
 358 }
 359
 360 static int compacted_load_cluster_from_disk(struct z_erofs_maprecorder *m,
 361                                             unsigned long lcn, bool lookahead)
 362 {
 363         struct inode *const inode = m->inode;
 364         struct erofs_inode *const vi = EROFS_I(inode);
 365         const unsigned int lclusterbits = vi->z_logical_clusterbits;
 366         const erofs_off_t ebase = ALIGN(iloc(EROFS_I_SB(inode), vi->nid) +
 367                                         vi->inode_isize + vi->xattr_isize, 8) +
 368                 sizeof(struct z_erofs_map_header);
 369         const unsigned int totalidx = DIV_ROUND_UP(inode->i_size, EROFS_BLKSIZ);
 370         unsigned int compacted_4b_initial, compacted_2b;
 371         unsigned int amortizedshift;
 372         erofs_off_t pos;
 373         int err;
 374
 375         if (lclusterbits != 12)
 376                 return -EOPNOTSUPP;
 377
 378         if (lcn >= totalidx)
 379                 return -EINVAL;
 380
 381         m->lcn = lcn;
 382         /* used to align to 32-byte (compacted_2b) alignment */
 383         compacted_4b_initial = (32 - ebase % 32) / 4;
 384         if (compacted_4b_initial == 32 / 4)
 385                 compacted_4b_initial = 0;
 386
 387         if ((vi->z_advise & Z_EROFS_ADVISE_COMPACTED_2B) &&
 388             compacted_4b_initial < totalidx)
 389                 compacted_2b = rounddown(totalidx - compacted_4b_initial, 16);
 390         else
 391                 compacted_2b = 0;
 392
 393         pos = ebase;
 394         if (lcn < compacted_4b_initial) {
 395                 amortizedshift = 2;
 396                 goto out;
 397         }
 398         pos += compacted_4b_initial * 4;
 399         lcn -= compacted_4b_initial;
 400
 401         if (lcn < compacted_2b) {
 402                 amortizedshift = 1;
 403                 goto out;
 404         }
 405         pos += compacted_2b * 2;
 406         lcn -= compacted_2b;
 407         amortizedshift = 2;
 408 out:
 409         pos += lcn * (1 << amortizedshift);
 410         err = z_erofs_reload_indexes(m, erofs_blknr(pos));
 411         if (err)
 412                 return err;
 413         return unpack_compacted_index(m, amortizedshift, pos, lookahead);
 414 }
 415
 416 static int z_erofs_load_cluster_from_disk(struct z_erofs_maprecorder *m,
 417                                           unsigned int lcn, bool lookahead)
 418 {
 419         const unsigned int datamode = EROFS_I(m->inode)->datalayout;
 420
 421         if (datamode == EROFS_INODE_FLAT_COMPRESSION_LEGACY)
 422                 return legacy_load_cluster_from_disk(m, lcn);
 423
 424         if (datamode == EROFS_INODE_FLAT_COMPRESSION)
 425                 return compacted_load_cluster_from_disk(m, lcn, lookahead);
 426
 427         return -EINVAL;
 428 }
 429
 430 static int z_erofs_extent_lookback(struct z_erofs_maprecorder *m,
 431                                    unsigned int lookback_distance)
 432 {
 433         struct erofs_inode *const vi = EROFS_I(m->inode);
 434         struct erofs_map_blocks *const map = m->map;
 435         const unsigned int lclusterbits = vi->z_logical_clusterbits;
 436         unsigned long lcn = m->lcn;
 437         int err;
 438
 439         if (lcn < lookback_distance) {
 440                 erofs_err(m->inode->i_sb,
 441                           "bogus lookback distance @ nid %llu", vi->nid);
 442                 DBG_BUGON(1);
 443                 return -EFSCORRUPTED;
 444         }
 445
 446         /* load extent head logical cluster if needed */
 447         lcn -= lookback_distance;
 448         err = z_erofs_load_cluster_from_disk(m, lcn, false);
 449         if (err)
 450                 return err;
 451
 452         switch (m->type) {
 453         case Z_EROFS_VLE_CLUSTER_TYPE_NONHEAD:
 454                 if (!m->delta[0]) {
 455                         erofs_err(m->inode->i_sb,
 456                                   "invalid lookback distance 0 @ nid %llu",
 457                                   vi->nid);
 458                         DBG_BUGON(1);
 459                         return -EFSCORRUPTED;
 460                 }
 461                 return z_erofs_extent_lookback(m, m->delta[0]);
 462         case Z_EROFS_VLE_CLUSTER_TYPE_PLAIN:
 463         case Z_EROFS_VLE_CLUSTER_TYPE_HEAD1:
 464         case Z_EROFS_VLE_CLUSTER_TYPE_HEAD2:
 465                 m->headtype = m->type;
 466                 map->m_la = (lcn << lclusterbits) | m->clusterofs;
 467                 break;
 468         default:
 469                 erofs_err(m->inode->i_sb,
 470                           "unknown type %u @ lcn %lu of nid %llu",
 471                           m->type, lcn, vi->nid);
 472                 DBG_BUGON(1);
 473                 return -EOPNOTSUPP;
 474         }
 475         return 0;
 476 }
 477
 478 static int z_erofs_get_extent_compressedlen(struct z_erofs_maprecorder *m,
 479                                             unsigned int initial_lcn)
 480 {
 481         struct erofs_inode *const vi = EROFS_I(m->inode);
 482         struct erofs_map_blocks *const map = m->map;
 483         const unsigned int lclusterbits = vi->z_logical_clusterbits;
 484         unsigned long lcn;
 485         int err;
 486
 487         DBG_BUGON(m->type != Z_EROFS_VLE_CLUSTER_TYPE_PLAIN &&
 488                   m->type != Z_EROFS_VLE_CLUSTER_TYPE_HEAD1 &&
 489                   m->type != Z_EROFS_VLE_CLUSTER_TYPE_HEAD2);
 490         DBG_BUGON(m->type != m->headtype);
 491
 492         if (m->headtype == Z_EROFS_VLE_CLUSTER_TYPE_PLAIN ||
 493             ((m->headtype == Z_EROFS_VLE_CLUSTER_TYPE_HEAD1) &&
 494              !(vi->z_advise & Z_EROFS_ADVISE_BIG_PCLUSTER_1)) ||
 495             ((m->headtype == Z_EROFS_VLE_CLUSTER_TYPE_HEAD2) &&
 496              !(vi->z_advise & Z_EROFS_ADVISE_BIG_PCLUSTER_2))) {
 497                 map->m_plen = 1 << lclusterbits;
 498                 return 0;
 499         }
 500         lcn = m->lcn + 1;
 501         if (m->compressedlcs)
 502                 goto out;
 503
 504         err = z_erofs_load_cluster_from_disk(m, lcn, false);
 505         if (err)
 506                 return err;
 507
 508         /*
 509          * If the 1st NONHEAD lcluster has already been handled initially w/o
 510          * valid compressedlcs, which means at least it mustn't be CBLKCNT, or
 511          * an internal implemenatation error is detected.
 512          *
 513          * The following code can also handle it properly anyway, but let's
 514          * BUG_ON in the debugging mode only for developers to notice that.
 515          */
 516         DBG_BUGON(lcn == initial_lcn &&
 517                   m->type == Z_EROFS_VLE_CLUSTER_TYPE_NONHEAD);
 518
 519         switch (m->type) {
 520         case Z_EROFS_VLE_CLUSTER_TYPE_PLAIN:
 521         case Z_EROFS_VLE_CLUSTER_TYPE_HEAD1:
 522         case Z_EROFS_VLE_CLUSTER_TYPE_HEAD2:
 523                 /*
 524                  * if the 1st NONHEAD lcluster is actually PLAIN or HEAD type
 525                  * rather than CBLKCNT, it's a 1 lcluster-sized pcluster.
 526                  */
 527                 m->compressedlcs = 1;
 528                 break;
 529         case Z_EROFS_VLE_CLUSTER_TYPE_NONHEAD:
 530                 if (m->delta[0] != 1)
 531                         goto err_bonus_cblkcnt;
 532                 if (m->compressedlcs)
 533                         break;
 534                 fallthrough;
 535         default:
 536                 erofs_err(m->inode->i_sb,
 537                           "cannot found CBLKCNT @ lcn %lu of nid %llu",
 538                           lcn, vi->nid);
 539                 DBG_BUGON(1);
 540                 return -EFSCORRUPTED;
 541         }
 542 out:
 543         map->m_plen = m->compressedlcs << lclusterbits;
 544         return 0;
 545 err_bonus_cblkcnt:
 546         erofs_err(m->inode->i_sb,
 547                   "bogus CBLKCNT @ lcn %lu of nid %llu",
 548                   lcn, vi->nid);
 549         DBG_BUGON(1);
 550         return -EFSCORRUPTED;
 551 }
 552
 553 static int z_erofs_get_extent_decompressedlen(struct z_erofs_maprecorder *m)
 554 {
 555         struct inode *inode = m->inode;
 556         struct erofs_inode *vi = EROFS_I(inode);
 557         struct erofs_map_blocks *map = m->map;
 558         unsigned int lclusterbits = vi->z_logical_clusterbits;
 559         u64 lcn = m->lcn, headlcn = map->m_la >> lclusterbits;
 560         int err;
 561
 562         do {
 563                 /* handle the last EOF pcluster (no next HEAD lcluster) */
 564                 if ((lcn << lclusterbits) >= inode->i_size) {
 565                         map->m_llen = inode->i_size - map->m_la;
 566                         return 0;
 567                 }
 568
 569                 err = z_erofs_load_cluster_from_disk(m, lcn, true);
 570                 if (err)
 571                         return err;
 572
 573                 if (m->type == Z_EROFS_VLE_CLUSTER_TYPE_NONHEAD) {
 574                         DBG_BUGON(!m->delta[1] &&
 575                                   m->clusterofs != 1 << lclusterbits);
 576                 } else if (m->type == Z_EROFS_VLE_CLUSTER_TYPE_PLAIN ||
 577                            m->type == Z_EROFS_VLE_CLUSTER_TYPE_HEAD1 ||
 578                            m->type == Z_EROFS_VLE_CLUSTER_TYPE_HEAD2) {
 579                         /* go on until the next HEAD lcluster */
 580                         if (lcn != headlcn)
 581                                 break;
 582                         m->delta[1] = 1;
 583                 } else {
 584                         erofs_err(inode->i_sb, "unknown type %u @ lcn %llu of nid %llu",
 585                                   m->type, lcn, vi->nid);
 586                         DBG_BUGON(1);
 587                         return -EOPNOTSUPP;
 588                 }
 589                 lcn += m->delta[1];
 590         } while (m->delta[1]);
 591
 592         map->m_llen = (lcn << lclusterbits) + m->clusterofs - map->m_la;
 593         return 0;
 594 }
 595
 596 static int z_erofs_do_map_blocks(struct inode *inode,
 597                                  struct erofs_map_blocks *map,
 598                                  int flags)
 599 {
 600         struct erofs_inode *const vi = EROFS_I(inode);
 601         bool ztailpacking = vi->z_advise & Z_EROFS_ADVISE_INLINE_PCLUSTER;
 602         struct z_erofs_maprecorder m = {
 603                 .inode = inode,
 604                 .map = map,
 605         };
 606         int err = 0;
 607         unsigned int lclusterbits, endoff;
 608         unsigned long initial_lcn;
 609         unsigned long long ofs, end;
 610
 611         lclusterbits = vi->z_logical_clusterbits;
 612         ofs = flags & EROFS_GET_BLOCKS_FINDTAIL ? inode->i_size - 1 : map->m_la;
 613         initial_lcn = ofs >> lclusterbits;
 614         endoff = ofs & ((1 << lclusterbits) - 1);
 615
 616         err = z_erofs_load_cluster_from_disk(&m, initial_lcn, false);
 617         if (err)
 618                 goto unmap_out;
 619
 620         if (ztailpacking && (flags & EROFS_GET_BLOCKS_FINDTAIL))
 621                 vi->z_idataoff = m.nextpackoff;
 622
 623         map->m_flags = EROFS_MAP_MAPPED | EROFS_MAP_ENCODED;
 624         end = (m.lcn + 1ULL) << lclusterbits;
 625
 626         switch (m.type) {
 627         case Z_EROFS_VLE_CLUSTER_TYPE_PLAIN:
 628         case Z_EROFS_VLE_CLUSTER_TYPE_HEAD1:
 629         case Z_EROFS_VLE_CLUSTER_TYPE_HEAD2:
 630                 if (endoff >= m.clusterofs) {
 631                         m.headtype = m.type;
 632                         map->m_la = (m.lcn << lclusterbits) | m.clusterofs;
 633                         break;
 634                 }
 635                 /* m.lcn should be >= 1 if endoff < m.clusterofs */
 636                 if (!m.lcn) {
 637                         erofs_err(inode->i_sb,
 638                                   "invalid logical cluster 0 at nid %llu",
 639                                   vi->nid);
 640                         err = -EFSCORRUPTED;
 641                         goto unmap_out;
 642                 }
 643                 end = (m.lcn << lclusterbits) | m.clusterofs;
 644                 map->m_flags |= EROFS_MAP_FULL_MAPPED;
 645                 m.delta[0] = 1;
 646                 fallthrough;
 647         case Z_EROFS_VLE_CLUSTER_TYPE_NONHEAD:
 648                 /* get the corresponding first chunk */
 649                 err = z_erofs_extent_lookback(&m, m.delta[0]);
 650                 if (err)
 651                         goto unmap_out;
 652                 break;
 653         default:
 654                 erofs_err(inode->i_sb,
 655                           "unknown type %u @ offset %llu of nid %llu",
 656                           m.type, ofs, vi->nid);
 657                 err = -EOPNOTSUPP;
 658                 goto unmap_out;
 659         }
 660
 661         map->m_llen = end - map->m_la;
 662
 663         if (flags & EROFS_GET_BLOCKS_FINDTAIL)
 664                 vi->z_tailextent_headlcn = m.lcn;
 665         if (ztailpacking && m.lcn == vi->z_tailextent_headlcn) {
 666                 map->m_flags |= EROFS_MAP_META;
 667                 map->m_pa = vi->z_idataoff;
 668                 map->m_plen = vi->z_idata_size;
 669         } else {
 670                 map->m_pa = blknr_to_addr(m.pblk);
 671                 err = z_erofs_get_extent_compressedlen(&m, initial_lcn);
 672                 if (err)
 673                         goto out;
 674         }
 675
 676         if (m.headtype == Z_EROFS_VLE_CLUSTER_TYPE_PLAIN)
 677                 map->m_algorithmformat = Z_EROFS_COMPRESSION_SHIFTED;
 678         else if (m.headtype == Z_EROFS_VLE_CLUSTER_TYPE_HEAD2)
 679                 map->m_algorithmformat = vi->z_algorithmtype[1];
 680         else
 681                 map->m_algorithmformat = vi->z_algorithmtype[0];
 682
 683         if ((flags & EROFS_GET_BLOCKS_FIEMAP) ||
 684             ((flags & EROFS_GET_BLOCKS_READMORE) &&
 685              map->m_algorithmformat == Z_EROFS_COMPRESSION_LZMA &&
 686              map->m_llen >= EROFS_BLKSIZ)) {
 687                 err = z_erofs_get_extent_decompressedlen(&m);
 688                 if (!err)
 689                         map->m_flags |= EROFS_MAP_FULL_MAPPED;
 690         }
 691 unmap_out:
 692         erofs_unmap_metabuf(&m.map->buf);
 693
 694 out:
 695         erofs_dbg("%s, m_la %llu m_pa %llu m_llen %llu m_plen %llu m_flags 0%o",
 696                   __func__, map->m_la, map->m_pa,
 697                   map->m_llen, map->m_plen, map->m_flags);
 698
 699         return err;
 700 }
 701
 702 int z_erofs_map_blocks_iter(struct inode *inode,
 703                             struct erofs_map_blocks *map,
 704                             int flags)
 705 {
 706         int err = 0;
 707
 708         trace_z_erofs_map_blocks_iter_enter(inode, map, flags);
 709
 710         /* when trying to read beyond EOF, leave it unmapped */
 711         if (map->m_la >= inode->i_size) {
 712                 map->m_llen = map->m_la + 1 - inode->i_size;
 713                 map->m_la = inode->i_size;
 714                 map->m_flags = 0;
 715                 goto out;
 716         }
 717
 718         err = z_erofs_fill_inode_lazy(inode);
 719         if (err)
 720                 goto out;
 721
 722         err = z_erofs_do_map_blocks(inode, map, flags);
 723 out:
 724         trace_z_erofs_map_blocks_iter_exit(inode, map, flags, err);
 725
 726         /* aggressively BUG_ON iff CONFIG_EROFS_FS_DEBUG is on */
 727         DBG_BUGON(err < 0 && err != -ENOMEM);
 728         return err;
 729 }
 730
 731 static int z_erofs_iomap_begin_report(struct inode *inode, loff_t offset,
 732                                 loff_t length, unsigned int flags,
 733                                 struct iomap *iomap, struct iomap *srcmap)
 734 {
 735         int ret;
 736         struct erofs_map_blocks map = { .m_la = offset };
 737
 738         ret = z_erofs_map_blocks_iter(inode, &map, EROFS_GET_BLOCKS_FIEMAP);
 739         erofs_put_metabuf(&map.buf);
 740         if (ret < 0)
 741                 return ret;
 742
 743         iomap->bdev = inode->i_sb->s_bdev;
 744         iomap->offset = map.m_la;
 745         iomap->length = map.m_llen;
 746         if (map.m_flags & EROFS_MAP_MAPPED) {
 747                 iomap->type = IOMAP_MAPPED;
 748                 iomap->addr = map.m_pa;
 749         } else {
 750                 iomap->type = IOMAP_HOLE;
 751                 iomap->addr = IOMAP_NULL_ADDR;
 752                 /*
 753                  * No strict rule how to describe extents for post EOF, yet
 754                  * we need do like below. Otherwise, iomap itself will get
 755                  * into an endless loop on post EOF.
 756                  */
 757                 if (iomap->offset >= inode->i_size)
 758                         iomap->length = length + map.m_la - offset;
 759         }
 760         iomap->flags = 0;
 761         return 0;
 762 }
 763
 764 const struct iomap_ops z_erofs_iomap_report_ops = {
 765         .iomap_begin = z_erofs_iomap_begin_report,
 766 };