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 <linux/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 unsigned long lcn; 14 /* compression extent information gathered */ 15 u8 type, headtype; 16 u16 clusterofs; 17 u16 delta[2]; 18 erofs_blk_t pblk, compressedblks; 19 erofs_off_t nextpackoff; 20 bool partialref; 21 }; 22 23 static int z_erofs_load_full_lcluster(struct z_erofs_maprecorder *m, 24 unsigned long lcn) 25 { 26 struct inode *const inode = m->inode; 27 struct erofs_inode *const vi = EROFS_I(inode); 28 const erofs_off_t pos = Z_EROFS_FULL_INDEX_START(erofs_iloc(inode) + 29 vi->inode_isize + vi->xattr_isize) + 30 lcn * sizeof(struct z_erofs_lcluster_index); 31 struct z_erofs_lcluster_index *di; 32 unsigned int advise; 33 34 di = erofs_read_metabuf(&m->map->buf, inode->i_sb, pos, true); 35 if (IS_ERR(di)) 36 return PTR_ERR(di); 37 m->lcn = lcn; 38 m->nextpackoff = pos + sizeof(struct z_erofs_lcluster_index); 39 40 advise = le16_to_cpu(di->di_advise); 41 m->type = advise & Z_EROFS_LI_LCLUSTER_TYPE_MASK; 42 if (m->type == Z_EROFS_LCLUSTER_TYPE_NONHEAD) { 43 m->clusterofs = 1 << vi->z_lclusterbits; 44 m->delta[0] = le16_to_cpu(di->di_u.delta[0]); 45 if (m->delta[0] & Z_EROFS_LI_D0_CBLKCNT) { 46 if (!(vi->z_advise & (Z_EROFS_ADVISE_BIG_PCLUSTER_1 | 47 Z_EROFS_ADVISE_BIG_PCLUSTER_2))) { 48 DBG_BUGON(1); 49 return -EFSCORRUPTED; 50 } 51 m->compressedblks = m->delta[0] & ~Z_EROFS_LI_D0_CBLKCNT; 52 m->delta[0] = 1; 53 } 54 m->delta[1] = le16_to_cpu(di->di_u.delta[1]); 55 } else { 56 m->partialref = !!(advise & Z_EROFS_LI_PARTIAL_REF); 57 m->clusterofs = le16_to_cpu(di->di_clusterofs); 58 if (m->clusterofs >= 1 << vi->z_lclusterbits) { 59 DBG_BUGON(1); 60 return -EFSCORRUPTED; 61 } 62 m->pblk = le32_to_cpu(di->di_u.blkaddr); 63 } 64 return 0; 65 } 66 67 static unsigned int decode_compactedbits(unsigned int lobits, 68 u8 *in, unsigned int pos, u8 *type) 69 { 70 const unsigned int v = get_unaligned_le32(in + pos / 8) >> (pos & 7); 71 const unsigned int lo = v & ((1 << lobits) - 1); 72 73 *type = (v >> lobits) & 3; 74 return lo; 75 } 76 77 static int get_compacted_la_distance(unsigned int lobits, 78 unsigned int encodebits, 79 unsigned int vcnt, u8 *in, int i) 80 { 81 unsigned int lo, d1 = 0; 82 u8 type; 83 84 DBG_BUGON(i >= vcnt); 85 86 do { 87 lo = decode_compactedbits(lobits, in, encodebits * i, &type); 88 89 if (type != Z_EROFS_LCLUSTER_TYPE_NONHEAD) 90 return d1; 91 ++d1; 92 } while (++i < vcnt); 93 94 /* vcnt - 1 (Z_EROFS_LCLUSTER_TYPE_NONHEAD) item */ 95 if (!(lo & Z_EROFS_LI_D0_CBLKCNT)) 96 d1 += lo - 1; 97 return d1; 98 } 99 100 static int z_erofs_load_compact_lcluster(struct z_erofs_maprecorder *m, 101 unsigned long lcn, bool lookahead) 102 { 103 struct inode *const inode = m->inode; 104 struct erofs_inode *const vi = EROFS_I(inode); 105 const erofs_off_t ebase = Z_EROFS_MAP_HEADER_END(erofs_iloc(inode) + 106 vi->inode_isize + vi->xattr_isize); 107 const unsigned int lclusterbits = vi->z_lclusterbits; 108 const unsigned int totalidx = erofs_iblks(inode); 109 unsigned int compacted_4b_initial, compacted_2b, amortizedshift; 110 unsigned int vcnt, lo, lobits, encodebits, nblk, bytes; 111 bool big_pcluster = vi->z_advise & Z_EROFS_ADVISE_BIG_PCLUSTER_1; 112 erofs_off_t pos; 113 u8 *in, type; 114 int i; 115 116 if (lcn >= totalidx || lclusterbits > 14) 117 return -EINVAL; 118 119 m->lcn = lcn; 120 /* used to align to 32-byte (compacted_2b) alignment */ 121 compacted_4b_initial = ((32 - ebase % 32) / 4) & 7; 122 compacted_2b = 0; 123 if ((vi->z_advise & Z_EROFS_ADVISE_COMPACTED_2B) && 124 compacted_4b_initial < totalidx) 125 compacted_2b = rounddown(totalidx - compacted_4b_initial, 16); 126 127 pos = ebase; 128 amortizedshift = 2; /* compact_4b */ 129 if (lcn >= compacted_4b_initial) { 130 pos += compacted_4b_initial * 4; 131 lcn -= compacted_4b_initial; 132 if (lcn < compacted_2b) { 133 amortizedshift = 1; 134 } else { 135 pos += compacted_2b * 2; 136 lcn -= compacted_2b; 137 } 138 } 139 pos += lcn * (1 << amortizedshift); 140 141 /* figure out the lcluster count in this pack */ 142 if (1 << amortizedshift == 4 && lclusterbits <= 14) 143 vcnt = 2; 144 else if (1 << amortizedshift == 2 && lclusterbits <= 12) 145 vcnt = 16; 146 else 147 return -EOPNOTSUPP; 148 149 in = erofs_read_metabuf(&m->map->buf, m->inode->i_sb, pos, true); 150 if (IS_ERR(in)) 151 return PTR_ERR(in); 152 153 /* it doesn't equal to round_up(..) */ 154 m->nextpackoff = round_down(pos, vcnt << amortizedshift) + 155 (vcnt << amortizedshift); 156 lobits = max(lclusterbits, ilog2(Z_EROFS_LI_D0_CBLKCNT) + 1U); 157 encodebits = ((vcnt << amortizedshift) - sizeof(__le32)) * 8 / vcnt; 158 bytes = pos & ((vcnt << amortizedshift) - 1); 159 in -= bytes; 160 i = bytes >> amortizedshift; 161 162 lo = decode_compactedbits(lobits, in, encodebits * i, &type); 163 m->type = type; 164 if (type == Z_EROFS_LCLUSTER_TYPE_NONHEAD) { 165 m->clusterofs = 1 << lclusterbits; 166 167 /* figure out lookahead_distance: delta[1] if needed */ 168 if (lookahead) 169 m->delta[1] = get_compacted_la_distance(lobits, 170 encodebits, vcnt, in, i); 171 if (lo & Z_EROFS_LI_D0_CBLKCNT) { 172 if (!big_pcluster) { 173 DBG_BUGON(1); 174 return -EFSCORRUPTED; 175 } 176 m->compressedblks = lo & ~Z_EROFS_LI_D0_CBLKCNT; 177 m->delta[0] = 1; 178 return 0; 179 } else if (i + 1 != (int)vcnt) { 180 m->delta[0] = lo; 181 return 0; 182 } 183 /* 184 * since the last lcluster in the pack is special, 185 * of which lo saves delta[1] rather than delta[0]. 186 * Hence, get delta[0] by the previous lcluster indirectly. 187 */ 188 lo = decode_compactedbits(lobits, in, 189 encodebits * (i - 1), &type); 190 if (type != Z_EROFS_LCLUSTER_TYPE_NONHEAD) 191 lo = 0; 192 else if (lo & Z_EROFS_LI_D0_CBLKCNT) 193 lo = 1; 194 m->delta[0] = lo + 1; 195 return 0; 196 } 197 m->clusterofs = lo; 198 m->delta[0] = 0; 199 /* figout out blkaddr (pblk) for HEAD lclusters */ 200 if (!big_pcluster) { 201 nblk = 1; 202 while (i > 0) { 203 --i; 204 lo = decode_compactedbits(lobits, in, 205 encodebits * i, &type); 206 if (type == Z_EROFS_LCLUSTER_TYPE_NONHEAD) 207 i -= lo; 208 209 if (i >= 0) 210 ++nblk; 211 } 212 } else { 213 nblk = 0; 214 while (i > 0) { 215 --i; 216 lo = decode_compactedbits(lobits, in, 217 encodebits * i, &type); 218 if (type == Z_EROFS_LCLUSTER_TYPE_NONHEAD) { 219 if (lo & Z_EROFS_LI_D0_CBLKCNT) { 220 --i; 221 nblk += lo & ~Z_EROFS_LI_D0_CBLKCNT; 222 continue; 223 } 224 /* bigpcluster shouldn't have plain d0 == 1 */ 225 if (lo <= 1) { 226 DBG_BUGON(1); 227 return -EFSCORRUPTED; 228 } 229 i -= lo - 2; 230 continue; 231 } 232 ++nblk; 233 } 234 } 235 in += (vcnt << amortizedshift) - sizeof(__le32); 236 m->pblk = le32_to_cpu(*(__le32 *)in) + nblk; 237 return 0; 238 } 239 240 static int z_erofs_load_lcluster_from_disk(struct z_erofs_maprecorder *m, 241 unsigned int lcn, bool lookahead) 242 { 243 switch (EROFS_I(m->inode)->datalayout) { 244 case EROFS_INODE_COMPRESSED_FULL: 245 return z_erofs_load_full_lcluster(m, lcn); 246 case EROFS_INODE_COMPRESSED_COMPACT: 247 return z_erofs_load_compact_lcluster(m, lcn, lookahead); 248 default: 249 return -EINVAL; 250 } 251 } 252 253 static int z_erofs_extent_lookback(struct z_erofs_maprecorder *m, 254 unsigned int lookback_distance) 255 { 256 struct super_block *sb = m->inode->i_sb; 257 struct erofs_inode *const vi = EROFS_I(m->inode); 258 const unsigned int lclusterbits = vi->z_lclusterbits; 259 260 while (m->lcn >= lookback_distance) { 261 unsigned long lcn = m->lcn - lookback_distance; 262 int err; 263 264 err = z_erofs_load_lcluster_from_disk(m, lcn, false); 265 if (err) 266 return err; 267 268 if (m->type >= Z_EROFS_LCLUSTER_TYPE_MAX) { 269 erofs_err(sb, "unknown type %u @ lcn %lu of nid %llu", 270 m->type, lcn, vi->nid); 271 DBG_BUGON(1); 272 return -EOPNOTSUPP; 273 } else if (m->type == Z_EROFS_LCLUSTER_TYPE_NONHEAD) { 274 lookback_distance = m->delta[0]; 275 if (!lookback_distance) 276 break; 277 continue; 278 } else { 279 m->headtype = m->type; 280 m->map->m_la = (lcn << lclusterbits) | m->clusterofs; 281 return 0; 282 } 283 } 284 erofs_err(sb, "bogus lookback distance %u @ lcn %lu of nid %llu", 285 lookback_distance, m->lcn, vi->nid); 286 DBG_BUGON(1); 287 return -EFSCORRUPTED; 288 } 289 290 static int z_erofs_get_extent_compressedlen(struct z_erofs_maprecorder *m, 291 unsigned int initial_lcn) 292 { 293 struct inode *inode = m->inode; 294 struct super_block *sb = inode->i_sb; 295 struct erofs_inode *vi = EROFS_I(inode); 296 bool bigpcl1 = vi->z_advise & Z_EROFS_ADVISE_BIG_PCLUSTER_1; 297 bool bigpcl2 = vi->z_advise & Z_EROFS_ADVISE_BIG_PCLUSTER_2; 298 unsigned long lcn = m->lcn + 1; 299 int err; 300 301 DBG_BUGON(m->type == Z_EROFS_LCLUSTER_TYPE_NONHEAD); 302 DBG_BUGON(m->type != m->headtype); 303 304 if ((m->headtype == Z_EROFS_LCLUSTER_TYPE_HEAD1 && !bigpcl1) || 305 ((m->headtype == Z_EROFS_LCLUSTER_TYPE_PLAIN || 306 m->headtype == Z_EROFS_LCLUSTER_TYPE_HEAD2) && !bigpcl2) || 307 (lcn << vi->z_lclusterbits) >= inode->i_size) 308 m->compressedblks = 1; 309 310 if (m->compressedblks) 311 goto out; 312 313 err = z_erofs_load_lcluster_from_disk(m, lcn, false); 314 if (err) 315 return err; 316 317 /* 318 * If the 1st NONHEAD lcluster has already been handled initially w/o 319 * valid compressedblks, which means at least it mustn't be CBLKCNT, or 320 * an internal implemenatation error is detected. 321 * 322 * The following code can also handle it properly anyway, but let's 323 * BUG_ON in the debugging mode only for developers to notice that. 324 */ 325 DBG_BUGON(lcn == initial_lcn && 326 m->type == Z_EROFS_LCLUSTER_TYPE_NONHEAD); 327 328 if (m->type == Z_EROFS_LCLUSTER_TYPE_NONHEAD) { 329 if (m->delta[0] != 1) { 330 erofs_err(sb, "bogus CBLKCNT @ lcn %lu of nid %llu", lcn, vi->nid); 331 DBG_BUGON(1); 332 return -EFSCORRUPTED; 333 } 334 if (m->compressedblks) 335 goto out; 336 } else if (m->type < Z_EROFS_LCLUSTER_TYPE_MAX) { 337 /* 338 * if the 1st NONHEAD lcluster is actually PLAIN or HEAD type 339 * rather than CBLKCNT, it's a 1 block-sized pcluster. 340 */ 341 m->compressedblks = 1; 342 goto out; 343 } 344 erofs_err(sb, "cannot found CBLKCNT @ lcn %lu of nid %llu", lcn, vi->nid); 345 DBG_BUGON(1); 346 return -EFSCORRUPTED; 347 out: 348 m->map->m_plen = erofs_pos(sb, m->compressedblks); 349 return 0; 350 } 351 352 static int z_erofs_get_extent_decompressedlen(struct z_erofs_maprecorder *m) 353 { 354 struct inode *inode = m->inode; 355 struct erofs_inode *vi = EROFS_I(inode); 356 struct erofs_map_blocks *map = m->map; 357 unsigned int lclusterbits = vi->z_lclusterbits; 358 u64 lcn = m->lcn, headlcn = map->m_la >> lclusterbits; 359 int err; 360 361 while (1) { 362 /* handle the last EOF pcluster (no next HEAD lcluster) */ 363 if ((lcn << lclusterbits) >= inode->i_size) { 364 map->m_llen = inode->i_size - map->m_la; 365 return 0; 366 } 367 368 err = z_erofs_load_lcluster_from_disk(m, lcn, true); 369 if (err) 370 return err; 371 372 if (m->type == Z_EROFS_LCLUSTER_TYPE_NONHEAD) { 373 /* work around invalid d1 generated by pre-1.0 mkfs */ 374 if (unlikely(!m->delta[1])) { 375 m->delta[1] = 1; 376 DBG_BUGON(1); 377 } 378 } else if (m->type < Z_EROFS_LCLUSTER_TYPE_MAX) { 379 if (lcn != headlcn) 380 break; /* ends at the next HEAD lcluster */ 381 m->delta[1] = 1; 382 } else { 383 erofs_err(inode->i_sb, "unknown type %u @ lcn %llu of nid %llu", 384 m->type, lcn, vi->nid); 385 DBG_BUGON(1); 386 return -EOPNOTSUPP; 387 } 388 lcn += m->delta[1]; 389 } 390 map->m_llen = (lcn << lclusterbits) + m->clusterofs - map->m_la; 391 return 0; 392 } 393 394 static int z_erofs_map_blocks_fo(struct inode *inode, 395 struct erofs_map_blocks *map, int flags) 396 { 397 struct erofs_inode *vi = EROFS_I(inode); 398 struct super_block *sb = inode->i_sb; 399 bool fragment = vi->z_advise & Z_EROFS_ADVISE_FRAGMENT_PCLUSTER; 400 bool ztailpacking = vi->z_idata_size; 401 unsigned int lclusterbits = vi->z_lclusterbits; 402 struct z_erofs_maprecorder m = { 403 .inode = inode, 404 .map = map, 405 }; 406 int err = 0; 407 unsigned int endoff, afmt; 408 unsigned long initial_lcn; 409 unsigned long long ofs, end; 410 411 ofs = flags & EROFS_GET_BLOCKS_FINDTAIL ? inode->i_size - 1 : map->m_la; 412 if (fragment && !(flags & EROFS_GET_BLOCKS_FINDTAIL) && 413 !vi->z_tailextent_headlcn) { 414 map->m_la = 0; 415 map->m_llen = inode->i_size; 416 map->m_flags = EROFS_MAP_FRAGMENT; 417 return 0; 418 } 419 initial_lcn = ofs >> lclusterbits; 420 endoff = ofs & ((1 << lclusterbits) - 1); 421 422 err = z_erofs_load_lcluster_from_disk(&m, initial_lcn, false); 423 if (err) 424 goto unmap_out; 425 426 if ((flags & EROFS_GET_BLOCKS_FINDTAIL) && ztailpacking) 427 vi->z_fragmentoff = m.nextpackoff; 428 map->m_flags = EROFS_MAP_MAPPED | EROFS_MAP_ENCODED; 429 end = (m.lcn + 1ULL) << lclusterbits; 430 431 switch (m.type) { 432 case Z_EROFS_LCLUSTER_TYPE_PLAIN: 433 case Z_EROFS_LCLUSTER_TYPE_HEAD1: 434 case Z_EROFS_LCLUSTER_TYPE_HEAD2: 435 if (endoff >= m.clusterofs) { 436 m.headtype = m.type; 437 map->m_la = (m.lcn << lclusterbits) | m.clusterofs; 438 /* 439 * For ztailpacking files, in order to inline data more 440 * effectively, special EOF lclusters are now supported 441 * which can have three parts at most. 442 */ 443 if (ztailpacking && end > inode->i_size) 444 end = inode->i_size; 445 break; 446 } 447 /* m.lcn should be >= 1 if endoff < m.clusterofs */ 448 if (!m.lcn) { 449 erofs_err(sb, "invalid logical cluster 0 at nid %llu", 450 vi->nid); 451 err = -EFSCORRUPTED; 452 goto unmap_out; 453 } 454 end = (m.lcn << lclusterbits) | m.clusterofs; 455 map->m_flags |= EROFS_MAP_FULL_MAPPED; 456 m.delta[0] = 1; 457 fallthrough; 458 case Z_EROFS_LCLUSTER_TYPE_NONHEAD: 459 /* get the corresponding first chunk */ 460 err = z_erofs_extent_lookback(&m, m.delta[0]); 461 if (err) 462 goto unmap_out; 463 break; 464 default: 465 erofs_err(sb, "unknown type %u @ offset %llu of nid %llu", 466 m.type, ofs, vi->nid); 467 err = -EOPNOTSUPP; 468 goto unmap_out; 469 } 470 if (m.partialref) 471 map->m_flags |= EROFS_MAP_PARTIAL_REF; 472 map->m_llen = end - map->m_la; 473 474 if (flags & EROFS_GET_BLOCKS_FINDTAIL) { 475 vi->z_tailextent_headlcn = m.lcn; 476 /* for non-compact indexes, fragmentoff is 64 bits */ 477 if (fragment && vi->datalayout == EROFS_INODE_COMPRESSED_FULL) 478 vi->z_fragmentoff |= (u64)m.pblk << 32; 479 } 480 if (ztailpacking && m.lcn == vi->z_tailextent_headlcn) { 481 map->m_flags |= EROFS_MAP_META; 482 map->m_pa = vi->z_fragmentoff; 483 map->m_plen = vi->z_idata_size; 484 if (erofs_blkoff(sb, map->m_pa) + map->m_plen > sb->s_blocksize) { 485 erofs_err(sb, "invalid tail-packing pclustersize %llu", 486 map->m_plen); 487 err = -EFSCORRUPTED; 488 goto unmap_out; 489 } 490 } else if (fragment && m.lcn == vi->z_tailextent_headlcn) { 491 map->m_flags = EROFS_MAP_FRAGMENT; 492 } else { 493 map->m_pa = erofs_pos(sb, m.pblk); 494 err = z_erofs_get_extent_compressedlen(&m, initial_lcn); 495 if (err) 496 goto unmap_out; 497 } 498 499 if (m.headtype == Z_EROFS_LCLUSTER_TYPE_PLAIN) { 500 if (map->m_llen > map->m_plen) { 501 DBG_BUGON(1); 502 err = -EFSCORRUPTED; 503 goto unmap_out; 504 } 505 afmt = vi->z_advise & Z_EROFS_ADVISE_INTERLACED_PCLUSTER ? 506 Z_EROFS_COMPRESSION_INTERLACED : 507 Z_EROFS_COMPRESSION_SHIFTED; 508 } else { 509 afmt = m.headtype == Z_EROFS_LCLUSTER_TYPE_HEAD2 ? 510 vi->z_algorithmtype[1] : vi->z_algorithmtype[0]; 511 if (!(EROFS_I_SB(inode)->available_compr_algs & (1 << afmt))) { 512 erofs_err(sb, "inconsistent algorithmtype %u for nid %llu", 513 afmt, vi->nid); 514 err = -EFSCORRUPTED; 515 goto unmap_out; 516 } 517 } 518 map->m_algorithmformat = afmt; 519 520 if ((flags & EROFS_GET_BLOCKS_FIEMAP) || 521 ((flags & EROFS_GET_BLOCKS_READMORE) && 522 (map->m_algorithmformat == Z_EROFS_COMPRESSION_LZMA || 523 map->m_algorithmformat == Z_EROFS_COMPRESSION_DEFLATE || 524 map->m_algorithmformat == Z_EROFS_COMPRESSION_ZSTD) && 525 map->m_llen >= i_blocksize(inode))) { 526 err = z_erofs_get_extent_decompressedlen(&m); 527 if (!err) 528 map->m_flags |= EROFS_MAP_FULL_MAPPED; 529 } 530 531 unmap_out: 532 erofs_unmap_metabuf(&m.map->buf); 533 return err; 534 } 535 536 static int z_erofs_map_blocks_ext(struct inode *inode, 537 struct erofs_map_blocks *map, int flags) 538 { 539 struct erofs_inode *vi = EROFS_I(inode); 540 struct super_block *sb = inode->i_sb; 541 bool interlaced = vi->z_advise & Z_EROFS_ADVISE_INTERLACED_PCLUSTER; 542 unsigned int recsz = z_erofs_extent_recsize(vi->z_advise); 543 erofs_off_t pos = round_up(Z_EROFS_MAP_HEADER_END(erofs_iloc(inode) + 544 vi->inode_isize + vi->xattr_isize), recsz); 545 erofs_off_t lend = inode->i_size; 546 erofs_off_t l, r, mid, pa, la, lstart; 547 struct z_erofs_extent *ext; 548 unsigned int fmt; 549 bool last; 550 551 map->m_flags = 0; 552 if (recsz <= offsetof(struct z_erofs_extent, pstart_hi)) { 553 if (recsz <= offsetof(struct z_erofs_extent, pstart_lo)) { 554 ext = erofs_read_metabuf(&map->buf, sb, pos, true); 555 if (IS_ERR(ext)) 556 return PTR_ERR(ext); 557 pa = le64_to_cpu(*(__le64 *)ext); 558 pos += sizeof(__le64); 559 lstart = 0; 560 } else { 561 lstart = round_down(map->m_la, 1 << vi->z_lclusterbits); 562 pos += (lstart >> vi->z_lclusterbits) * recsz; 563 pa = EROFS_NULL_ADDR; 564 } 565 566 for (; lstart <= map->m_la; lstart += 1 << vi->z_lclusterbits) { 567 ext = erofs_read_metabuf(&map->buf, sb, pos, true); 568 if (IS_ERR(ext)) 569 return PTR_ERR(ext); 570 map->m_plen = le32_to_cpu(ext->plen); 571 if (pa != EROFS_NULL_ADDR) { 572 map->m_pa = pa; 573 pa += map->m_plen & Z_EROFS_EXTENT_PLEN_MASK; 574 } else { 575 map->m_pa = le32_to_cpu(ext->pstart_lo); 576 } 577 pos += recsz; 578 } 579 last = (lstart >= round_up(lend, 1 << vi->z_lclusterbits)); 580 lend = min(lstart, lend); 581 lstart -= 1 << vi->z_lclusterbits; 582 } else { 583 lstart = lend; 584 for (l = 0, r = vi->z_extents; l < r; ) { 585 mid = l + (r - l) / 2; 586 ext = erofs_read_metabuf(&map->buf, sb, 587 pos + mid * recsz, true); 588 if (IS_ERR(ext)) 589 return PTR_ERR(ext); 590 591 la = le32_to_cpu(ext->lstart_lo); 592 pa = le32_to_cpu(ext->pstart_lo) | 593 (u64)le32_to_cpu(ext->pstart_hi) << 32; 594 if (recsz > offsetof(struct z_erofs_extent, lstart_hi)) 595 la |= (u64)le32_to_cpu(ext->lstart_hi) << 32; 596 597 if (la > map->m_la) { 598 r = mid; 599 if (la > lend) { 600 DBG_BUGON(1); 601 return -EFSCORRUPTED; 602 } 603 lend = la; 604 } else { 605 l = mid + 1; 606 if (map->m_la == la) 607 r = min(l + 1, r); 608 lstart = la; 609 map->m_plen = le32_to_cpu(ext->plen); 610 map->m_pa = pa; 611 } 612 } 613 last = (l >= vi->z_extents); 614 } 615 616 if (lstart < lend) { 617 map->m_la = lstart; 618 if (last && (vi->z_advise & Z_EROFS_ADVISE_FRAGMENT_PCLUSTER)) { 619 map->m_flags = EROFS_MAP_FRAGMENT; 620 vi->z_fragmentoff = map->m_plen; 621 if (recsz > offsetof(struct z_erofs_extent, pstart_lo)) 622 vi->z_fragmentoff |= map->m_pa << 32; 623 } else if (map->m_plen) { 624 map->m_flags |= EROFS_MAP_MAPPED | 625 EROFS_MAP_FULL_MAPPED | EROFS_MAP_ENCODED; 626 fmt = map->m_plen >> Z_EROFS_EXTENT_PLEN_FMT_BIT; 627 if (fmt) 628 map->m_algorithmformat = fmt - 1; 629 else if (interlaced && !erofs_blkoff(sb, map->m_pa)) 630 map->m_algorithmformat = 631 Z_EROFS_COMPRESSION_INTERLACED; 632 else 633 map->m_algorithmformat = 634 Z_EROFS_COMPRESSION_SHIFTED; 635 if (map->m_plen & Z_EROFS_EXTENT_PLEN_PARTIAL) 636 map->m_flags |= EROFS_MAP_PARTIAL_REF; 637 map->m_plen &= Z_EROFS_EXTENT_PLEN_MASK; 638 } 639 } 640 map->m_llen = lend - map->m_la; 641 return 0; 642 } 643 644 static int z_erofs_fill_inode_lazy(struct inode *inode) 645 { 646 struct erofs_inode *const vi = EROFS_I(inode); 647 struct super_block *const sb = inode->i_sb; 648 int err, headnr; 649 erofs_off_t pos; 650 struct erofs_buf buf = __EROFS_BUF_INITIALIZER; 651 struct z_erofs_map_header *h; 652 653 if (test_bit(EROFS_I_Z_INITED_BIT, &vi->flags)) { 654 /* 655 * paired with smp_mb() at the end of the function to ensure 656 * fields will only be observed after the bit is set. 657 */ 658 smp_mb(); 659 return 0; 660 } 661 662 if (wait_on_bit_lock(&vi->flags, EROFS_I_BL_Z_BIT, TASK_KILLABLE)) 663 return -ERESTARTSYS; 664 665 err = 0; 666 if (test_bit(EROFS_I_Z_INITED_BIT, &vi->flags)) 667 goto out_unlock; 668 669 pos = ALIGN(erofs_iloc(inode) + vi->inode_isize + vi->xattr_isize, 8); 670 h = erofs_read_metabuf(&buf, sb, pos, true); 671 if (IS_ERR(h)) { 672 err = PTR_ERR(h); 673 goto out_unlock; 674 } 675 676 /* 677 * if the highest bit of the 8-byte map header is set, the whole file 678 * is stored in the packed inode. The rest bits keeps z_fragmentoff. 679 */ 680 if (h->h_clusterbits >> Z_EROFS_FRAGMENT_INODE_BIT) { 681 vi->z_advise = Z_EROFS_ADVISE_FRAGMENT_PCLUSTER; 682 vi->z_fragmentoff = le64_to_cpu(*(__le64 *)h) ^ (1ULL << 63); 683 vi->z_tailextent_headlcn = 0; 684 goto done; 685 } 686 vi->z_advise = le16_to_cpu(h->h_advise); 687 vi->z_lclusterbits = sb->s_blocksize_bits + (h->h_clusterbits & 15); 688 if (vi->datalayout == EROFS_INODE_COMPRESSED_FULL && 689 (vi->z_advise & Z_EROFS_ADVISE_EXTENTS)) { 690 vi->z_extents = le32_to_cpu(h->h_extents_lo) | 691 ((u64)le16_to_cpu(h->h_extents_hi) << 32); 692 goto done; 693 } 694 695 vi->z_algorithmtype[0] = h->h_algorithmtype & 15; 696 vi->z_algorithmtype[1] = h->h_algorithmtype >> 4; 697 if (vi->z_advise & Z_EROFS_ADVISE_FRAGMENT_PCLUSTER) 698 vi->z_fragmentoff = le32_to_cpu(h->h_fragmentoff); 699 else if (vi->z_advise & Z_EROFS_ADVISE_INLINE_PCLUSTER) 700 vi->z_idata_size = le16_to_cpu(h->h_idata_size); 701 702 headnr = 0; 703 if (vi->z_algorithmtype[0] >= Z_EROFS_COMPRESSION_MAX || 704 vi->z_algorithmtype[++headnr] >= Z_EROFS_COMPRESSION_MAX) { 705 erofs_err(sb, "unknown HEAD%u format %u for nid %llu, please upgrade kernel", 706 headnr + 1, vi->z_algorithmtype[headnr], vi->nid); 707 err = -EOPNOTSUPP; 708 goto out_put_metabuf; 709 } 710 711 if (!erofs_sb_has_big_pcluster(EROFS_SB(sb)) && 712 vi->z_advise & (Z_EROFS_ADVISE_BIG_PCLUSTER_1 | 713 Z_EROFS_ADVISE_BIG_PCLUSTER_2)) { 714 erofs_err(sb, "per-inode big pcluster without sb feature for nid %llu", 715 vi->nid); 716 err = -EFSCORRUPTED; 717 goto out_put_metabuf; 718 } 719 if (vi->datalayout == EROFS_INODE_COMPRESSED_COMPACT && 720 !(vi->z_advise & Z_EROFS_ADVISE_BIG_PCLUSTER_1) ^ 721 !(vi->z_advise & Z_EROFS_ADVISE_BIG_PCLUSTER_2)) { 722 erofs_err(sb, "big pcluster head1/2 of compact indexes should be consistent for nid %llu", 723 vi->nid); 724 err = -EFSCORRUPTED; 725 goto out_put_metabuf; 726 } 727 728 if (vi->z_idata_size || 729 (vi->z_advise & Z_EROFS_ADVISE_FRAGMENT_PCLUSTER)) { 730 struct erofs_map_blocks map = { 731 .buf = __EROFS_BUF_INITIALIZER 732 }; 733 734 err = z_erofs_map_blocks_fo(inode, &map, 735 EROFS_GET_BLOCKS_FINDTAIL); 736 erofs_put_metabuf(&map.buf); 737 if (err < 0) 738 goto out_put_metabuf; 739 } 740 done: 741 /* paired with smp_mb() at the beginning of the function */ 742 smp_mb(); 743 set_bit(EROFS_I_Z_INITED_BIT, &vi->flags); 744 out_put_metabuf: 745 erofs_put_metabuf(&buf); 746 out_unlock: 747 clear_and_wake_up_bit(EROFS_I_BL_Z_BIT, &vi->flags); 748 return err; 749 } 750 751 int z_erofs_map_blocks_iter(struct inode *inode, struct erofs_map_blocks *map, 752 int flags) 753 { 754 struct erofs_inode *const vi = EROFS_I(inode); 755 int err = 0; 756 757 trace_erofs_map_blocks_enter(inode, map, flags); 758 if (map->m_la >= inode->i_size) { /* post-EOF unmapped extent */ 759 map->m_llen = map->m_la + 1 - inode->i_size; 760 map->m_la = inode->i_size; 761 map->m_flags = 0; 762 } else { 763 err = z_erofs_fill_inode_lazy(inode); 764 if (!err) { 765 if (vi->datalayout == EROFS_INODE_COMPRESSED_FULL && 766 (vi->z_advise & Z_EROFS_ADVISE_EXTENTS)) 767 err = z_erofs_map_blocks_ext(inode, map, flags); 768 else 769 err = z_erofs_map_blocks_fo(inode, map, flags); 770 } 771 if (!err && (map->m_flags & EROFS_MAP_ENCODED) && 772 unlikely(map->m_plen > Z_EROFS_PCLUSTER_MAX_SIZE || 773 map->m_llen > Z_EROFS_PCLUSTER_MAX_DSIZE)) 774 err = -EOPNOTSUPP; 775 if (err) 776 map->m_llen = 0; 777 } 778 trace_erofs_map_blocks_exit(inode, map, flags, err); 779 return err; 780 } 781 782 static int z_erofs_iomap_begin_report(struct inode *inode, loff_t offset, 783 loff_t length, unsigned int flags, 784 struct iomap *iomap, struct iomap *srcmap) 785 { 786 int ret; 787 struct erofs_map_blocks map = { .m_la = offset }; 788 789 ret = z_erofs_map_blocks_iter(inode, &map, EROFS_GET_BLOCKS_FIEMAP); 790 erofs_put_metabuf(&map.buf); 791 if (ret < 0) 792 return ret; 793 794 iomap->bdev = inode->i_sb->s_bdev; 795 iomap->offset = map.m_la; 796 iomap->length = map.m_llen; 797 if (map.m_flags & EROFS_MAP_MAPPED) { 798 iomap->type = IOMAP_MAPPED; 799 iomap->addr = map.m_flags & __EROFS_MAP_FRAGMENT ? 800 IOMAP_NULL_ADDR : map.m_pa; 801 } else { 802 iomap->type = IOMAP_HOLE; 803 iomap->addr = IOMAP_NULL_ADDR; 804 /* 805 * No strict rule on how to describe extents for post EOF, yet 806 * we need to do like below. Otherwise, iomap itself will get 807 * into an endless loop on post EOF. 808 * 809 * Calculate the effective offset by subtracting extent start 810 * (map.m_la) from the requested offset, and add it to length. 811 * (NB: offset >= map.m_la always) 812 */ 813 if (iomap->offset >= inode->i_size) 814 iomap->length = length + offset - map.m_la; 815 } 816 iomap->flags = 0; 817 return 0; 818 } 819 820 const struct iomap_ops z_erofs_iomap_report_ops = { 821 .iomap_begin = z_erofs_iomap_begin_report, 822 }; 823