1 // SPDX-License-Identifier: GPL-2.0-only 2 /* 3 * Copyright (C) 2017-2018 HUAWEI, Inc. 4 * https://www.huawei.com/ 5 * Copyright (C) 2021, Alibaba Cloud 6 */ 7 #include "internal.h" 8 #include <linux/filelock.h> 9 #include <linux/sched/mm.h> 10 #include <trace/events/erofs.h> 11 12 void erofs_unmap_metabuf(struct erofs_buf *buf) 13 { 14 if (!buf->base) 15 return; 16 kunmap_local(buf->base); 17 buf->base = NULL; 18 } 19 20 void erofs_put_metabuf(struct erofs_buf *buf) 21 { 22 if (!buf->page) 23 return; 24 erofs_unmap_metabuf(buf); 25 folio_put(page_folio(buf->page)); 26 buf->page = NULL; 27 } 28 29 void *erofs_bread(struct erofs_buf *buf, erofs_off_t offset, bool need_kmap) 30 { 31 pgoff_t index = (buf->off + offset) >> PAGE_SHIFT; 32 struct folio *folio = NULL; 33 loff_t fpos; 34 int err; 35 36 /* 37 * Metadata access for file-backed mounts reuses page cache of backing 38 * fs inodes (only folio data will be needed) to prevent double caching. 39 * However, the data access range must be verified here in advance. 40 */ 41 if (buf->file) { 42 fpos = index << PAGE_SHIFT; 43 err = rw_verify_area(READ, buf->file, &fpos, PAGE_SIZE); 44 if (err < 0) 45 return ERR_PTR(err); 46 } 47 48 if (buf->page) { 49 folio = page_folio(buf->page); 50 if (folio_file_page(folio, index) != buf->page) 51 erofs_unmap_metabuf(buf); 52 } 53 if (!folio || !folio_contains(folio, index)) { 54 erofs_put_metabuf(buf); 55 folio = read_mapping_folio(buf->mapping, index, buf->file); 56 if (IS_ERR(folio)) 57 return folio; 58 } 59 buf->page = folio_file_page(folio, index); 60 if (!need_kmap) 61 return NULL; 62 if (!buf->base) 63 buf->base = kmap_local_page(buf->page); 64 return buf->base + (offset & ~PAGE_MASK); 65 } 66 67 int erofs_init_metabuf(struct erofs_buf *buf, struct super_block *sb, 68 bool in_metabox) 69 { 70 struct erofs_sb_info *sbi = EROFS_SB(sb); 71 72 buf->file = NULL; 73 if (in_metabox) { 74 if (unlikely(!sbi->metabox_inode)) 75 return -EFSCORRUPTED; 76 buf->mapping = sbi->metabox_inode->i_mapping; 77 return 0; 78 } 79 buf->off = sbi->dif0.fsoff; 80 if (erofs_is_fileio_mode(sbi)) { 81 buf->file = sbi->dif0.file; /* some fs like FUSE needs it */ 82 buf->mapping = buf->file->f_mapping; 83 } else if (erofs_is_fscache_mode(sb)) 84 buf->mapping = sbi->dif0.fscache->inode->i_mapping; 85 else 86 buf->mapping = sb->s_bdev->bd_mapping; 87 return 0; 88 } 89 90 void *erofs_read_metabuf(struct erofs_buf *buf, struct super_block *sb, 91 erofs_off_t offset, bool in_metabox) 92 { 93 int err; 94 95 err = erofs_init_metabuf(buf, sb, in_metabox); 96 if (err) 97 return ERR_PTR(err); 98 return erofs_bread(buf, offset, true); 99 } 100 101 int erofs_map_blocks(struct inode *inode, struct erofs_map_blocks *map) 102 { 103 struct erofs_buf buf = __EROFS_BUF_INITIALIZER; 104 struct super_block *sb = inode->i_sb; 105 unsigned int unit, blksz = sb->s_blocksize; 106 struct erofs_inode *vi = EROFS_I(inode); 107 struct erofs_inode_chunk_index *idx; 108 erofs_blk_t startblk, addrmask; 109 bool tailpacking; 110 erofs_off_t pos; 111 u64 chunknr; 112 int err = 0; 113 114 trace_erofs_map_blocks_enter(inode, map, 0); 115 map->m_deviceid = 0; 116 map->m_flags = 0; 117 if (map->m_la >= inode->i_size) 118 goto out; 119 120 if (vi->datalayout != EROFS_INODE_CHUNK_BASED) { 121 tailpacking = (vi->datalayout == EROFS_INODE_FLAT_INLINE); 122 if (!tailpacking && vi->startblk == EROFS_NULL_ADDR) 123 goto out; 124 pos = erofs_pos(sb, erofs_iblks(inode) - tailpacking); 125 126 map->m_flags = EROFS_MAP_MAPPED; 127 if (map->m_la < pos) { 128 map->m_pa = erofs_pos(sb, vi->startblk) + map->m_la; 129 map->m_llen = pos - map->m_la; 130 } else { 131 map->m_pa = erofs_iloc(inode) + vi->inode_isize + 132 vi->xattr_isize + erofs_blkoff(sb, map->m_la); 133 map->m_llen = inode->i_size - map->m_la; 134 map->m_flags |= EROFS_MAP_META; 135 } 136 goto out; 137 } 138 139 if (vi->chunkformat & EROFS_CHUNK_FORMAT_INDEXES) 140 unit = sizeof(*idx); /* chunk index */ 141 else 142 unit = EROFS_BLOCK_MAP_ENTRY_SIZE; /* block map */ 143 144 chunknr = map->m_la >> vi->chunkbits; 145 pos = ALIGN(erofs_iloc(inode) + vi->inode_isize + 146 vi->xattr_isize, unit) + unit * chunknr; 147 148 idx = erofs_read_metabuf(&buf, sb, pos, erofs_inode_in_metabox(inode)); 149 if (IS_ERR(idx)) { 150 err = PTR_ERR(idx); 151 goto out; 152 } 153 map->m_la = chunknr << vi->chunkbits; 154 map->m_llen = min_t(erofs_off_t, 1UL << vi->chunkbits, 155 round_up(inode->i_size - map->m_la, blksz)); 156 if (vi->chunkformat & EROFS_CHUNK_FORMAT_INDEXES) { 157 addrmask = (vi->chunkformat & EROFS_CHUNK_FORMAT_48BIT) ? 158 BIT_ULL(48) - 1 : BIT_ULL(32) - 1; 159 startblk = (((u64)le16_to_cpu(idx->startblk_hi) << 32) | 160 le32_to_cpu(idx->startblk_lo)) & addrmask; 161 if ((startblk ^ EROFS_NULL_ADDR) & addrmask) { 162 map->m_deviceid = le16_to_cpu(idx->device_id) & 163 EROFS_SB(sb)->device_id_mask; 164 map->m_pa = erofs_pos(sb, startblk); 165 map->m_flags = EROFS_MAP_MAPPED; 166 } 167 } else { 168 startblk = le32_to_cpu(*(__le32 *)idx); 169 if (startblk != (u32)EROFS_NULL_ADDR) { 170 map->m_pa = erofs_pos(sb, startblk); 171 map->m_flags = EROFS_MAP_MAPPED; 172 } 173 } 174 erofs_put_metabuf(&buf); 175 out: 176 if (!err) { 177 map->m_plen = map->m_llen; 178 /* inline data should be located in the same meta block */ 179 if ((map->m_flags & EROFS_MAP_META) && 180 erofs_blkoff(sb, map->m_pa) + map->m_plen > blksz) { 181 erofs_err(sb, "inline data across blocks @ nid %llu", vi->nid); 182 DBG_BUGON(1); 183 return -EFSCORRUPTED; 184 } 185 } 186 trace_erofs_map_blocks_exit(inode, map, 0, err); 187 return err; 188 } 189 190 static void erofs_fill_from_devinfo(struct erofs_map_dev *map, 191 struct super_block *sb, struct erofs_device_info *dif) 192 { 193 map->m_sb = sb; 194 map->m_dif = dif; 195 map->m_bdev = NULL; 196 if (dif->file && S_ISBLK(file_inode(dif->file)->i_mode)) 197 map->m_bdev = file_bdev(dif->file); 198 } 199 200 int erofs_map_dev(struct super_block *sb, struct erofs_map_dev *map) 201 { 202 struct erofs_dev_context *devs = EROFS_SB(sb)->devs; 203 struct erofs_device_info *dif; 204 erofs_off_t startoff; 205 int id; 206 207 erofs_fill_from_devinfo(map, sb, &EROFS_SB(sb)->dif0); 208 map->m_bdev = sb->s_bdev; /* use s_bdev for the primary device */ 209 if (map->m_deviceid) { 210 down_read(&devs->rwsem); 211 dif = idr_find(&devs->tree, map->m_deviceid - 1); 212 if (!dif) { 213 up_read(&devs->rwsem); 214 return -ENODEV; 215 } 216 if (devs->flatdev) { 217 map->m_pa += erofs_pos(sb, dif->uniaddr); 218 up_read(&devs->rwsem); 219 return 0; 220 } 221 erofs_fill_from_devinfo(map, sb, dif); 222 up_read(&devs->rwsem); 223 } else if (devs->extra_devices && !devs->flatdev) { 224 down_read(&devs->rwsem); 225 idr_for_each_entry(&devs->tree, dif, id) { 226 if (!dif->uniaddr) 227 continue; 228 229 startoff = erofs_pos(sb, dif->uniaddr); 230 if (map->m_pa >= startoff && 231 map->m_pa < startoff + erofs_pos(sb, dif->blocks)) { 232 map->m_pa -= startoff; 233 erofs_fill_from_devinfo(map, sb, dif); 234 break; 235 } 236 } 237 up_read(&devs->rwsem); 238 } 239 return 0; 240 } 241 242 /* 243 * bit 30: I/O error occurred on this folio 244 * bit 29: CPU has dirty data in D-cache (needs aliasing handling); 245 * bit 0 - 29: remaining parts to complete this folio 246 */ 247 #define EROFS_ONLINEFOLIO_EIO 30 248 #define EROFS_ONLINEFOLIO_DIRTY 29 249 250 void erofs_onlinefolio_init(struct folio *folio) 251 { 252 union { 253 atomic_t o; 254 void *v; 255 } u = { .o = ATOMIC_INIT(1) }; 256 257 folio->private = u.v; /* valid only if file-backed folio is locked */ 258 } 259 260 void erofs_onlinefolio_split(struct folio *folio) 261 { 262 atomic_inc((atomic_t *)&folio->private); 263 } 264 265 void erofs_onlinefolio_end(struct folio *folio, int err, bool dirty) 266 { 267 int orig, v; 268 269 do { 270 orig = atomic_read((atomic_t *)&folio->private); 271 DBG_BUGON(orig <= 0); 272 v = dirty << EROFS_ONLINEFOLIO_DIRTY; 273 v |= (orig - 1) | (!!err << EROFS_ONLINEFOLIO_EIO); 274 } while (atomic_cmpxchg((atomic_t *)&folio->private, orig, v) != orig); 275 276 if (v & (BIT(EROFS_ONLINEFOLIO_DIRTY) - 1)) 277 return; 278 folio->private = 0; 279 if (v & BIT(EROFS_ONLINEFOLIO_DIRTY)) 280 flush_dcache_folio(folio); 281 folio_end_read(folio, !(v & BIT(EROFS_ONLINEFOLIO_EIO))); 282 } 283 284 struct erofs_iomap_iter_ctx { 285 struct page *page; 286 void *base; 287 struct inode *realinode; 288 }; 289 290 static int erofs_iomap_begin(struct inode *inode, loff_t offset, loff_t length, 291 unsigned int flags, struct iomap *iomap, struct iomap *srcmap) 292 { 293 struct iomap_iter *iter = container_of(iomap, struct iomap_iter, iomap); 294 struct erofs_iomap_iter_ctx *ctx = iter->private; 295 struct inode *realinode = ctx ? ctx->realinode : inode; 296 struct super_block *sb = realinode->i_sb; 297 struct erofs_map_blocks map; 298 struct erofs_map_dev mdev; 299 int ret; 300 301 map.m_la = offset; 302 map.m_llen = length; 303 ret = erofs_map_blocks(realinode, &map); 304 if (ret < 0) 305 return ret; 306 307 iomap->offset = map.m_la; 308 iomap->length = map.m_llen; 309 iomap->flags = 0; 310 iomap->addr = IOMAP_NULL_ADDR; 311 if (!(map.m_flags & EROFS_MAP_MAPPED)) { 312 iomap->type = IOMAP_HOLE; 313 return 0; 314 } 315 316 if (!(map.m_flags & EROFS_MAP_META) || !erofs_inode_in_metabox(realinode)) { 317 mdev = (struct erofs_map_dev) { 318 .m_deviceid = map.m_deviceid, 319 .m_pa = map.m_pa, 320 }; 321 ret = erofs_map_dev(sb, &mdev); 322 if (ret) 323 return ret; 324 325 if (flags & IOMAP_DAX) 326 iomap->dax_dev = mdev.m_dif->dax_dev; 327 else 328 iomap->bdev = mdev.m_bdev; 329 iomap->addr = mdev.m_dif->fsoff + mdev.m_pa; 330 if (flags & IOMAP_DAX) 331 iomap->addr += mdev.m_dif->dax_part_off; 332 } 333 334 if (map.m_flags & EROFS_MAP_META) { 335 iomap->type = IOMAP_INLINE; 336 /* read context should read the inlined data */ 337 if (ctx) { 338 struct erofs_buf buf = __EROFS_BUF_INITIALIZER; 339 void *ptr; 340 341 ptr = erofs_read_metabuf(&buf, sb, map.m_pa, 342 erofs_inode_in_metabox(realinode)); 343 if (IS_ERR(ptr)) 344 return PTR_ERR(ptr); 345 iomap->inline_data = ptr; 346 ctx->page = buf.page; 347 ctx->base = buf.base; 348 } 349 } else { 350 iomap->type = IOMAP_MAPPED; 351 } 352 return 0; 353 } 354 355 static int erofs_iomap_end(struct inode *inode, loff_t pos, loff_t length, 356 ssize_t written, unsigned int flags, struct iomap *iomap) 357 { 358 struct iomap_iter *iter = container_of(iomap, struct iomap_iter, iomap); 359 struct erofs_iomap_iter_ctx *ctx = iter->private; 360 361 if (ctx && ctx->base) { 362 struct erofs_buf buf = { 363 .page = ctx->page, 364 .base = ctx->base, 365 }; 366 367 DBG_BUGON(iomap->type != IOMAP_INLINE); 368 erofs_put_metabuf(&buf); 369 ctx->base = NULL; 370 } 371 return written; 372 } 373 374 static const struct iomap_ops erofs_iomap_ops = { 375 .iomap_begin = erofs_iomap_begin, 376 .iomap_end = erofs_iomap_end, 377 }; 378 379 int erofs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo, 380 u64 start, u64 len) 381 { 382 if (erofs_inode_is_data_compressed(EROFS_I(inode)->datalayout)) { 383 if (!IS_ENABLED(CONFIG_EROFS_FS_ZIP)) 384 return -EOPNOTSUPP; 385 return iomap_fiemap(inode, fieinfo, start, len, 386 &z_erofs_iomap_report_ops); 387 } 388 return iomap_fiemap(inode, fieinfo, start, len, &erofs_iomap_ops); 389 } 390 391 /* 392 * since we dont have write or truncate flows, so no inode 393 * locking needs to be held at the moment. 394 */ 395 static int erofs_read_folio(struct file *file, struct folio *folio) 396 { 397 struct iomap_read_folio_ctx read_ctx = { 398 .ops = &iomap_bio_read_ops, 399 .cur_folio = folio, 400 }; 401 bool need_iput; 402 struct erofs_iomap_iter_ctx iter_ctx = { 403 .realinode = erofs_real_inode(folio_inode(folio), &need_iput), 404 }; 405 406 trace_erofs_read_folio(iter_ctx.realinode, folio, true); 407 iomap_read_folio(&erofs_iomap_ops, &read_ctx, &iter_ctx); 408 if (need_iput) 409 iput(iter_ctx.realinode); 410 return 0; 411 } 412 413 static void erofs_readahead(struct readahead_control *rac) 414 { 415 struct iomap_read_folio_ctx read_ctx = { 416 .ops = &iomap_bio_read_ops, 417 .rac = rac, 418 }; 419 bool need_iput; 420 struct erofs_iomap_iter_ctx iter_ctx = { 421 .realinode = erofs_real_inode(rac->mapping->host, &need_iput), 422 }; 423 424 trace_erofs_readahead(iter_ctx.realinode, readahead_index(rac), 425 readahead_count(rac), true); 426 iomap_readahead(&erofs_iomap_ops, &read_ctx, &iter_ctx); 427 if (need_iput) 428 iput(iter_ctx.realinode); 429 } 430 431 static sector_t erofs_bmap(struct address_space *mapping, sector_t block) 432 { 433 return iomap_bmap(mapping, block, &erofs_iomap_ops); 434 } 435 436 static ssize_t erofs_file_read_iter(struct kiocb *iocb, struct iov_iter *to) 437 { 438 struct inode *inode = file_inode(iocb->ki_filp); 439 440 /* no need taking (shared) inode lock since it's a ro filesystem */ 441 if (!iov_iter_count(to)) 442 return 0; 443 444 if (IS_ENABLED(CONFIG_FS_DAX) && IS_DAX(inode)) 445 return dax_iomap_rw(iocb, to, &erofs_iomap_ops); 446 447 if ((iocb->ki_flags & IOCB_DIRECT) && inode->i_sb->s_bdev) { 448 struct erofs_iomap_iter_ctx iter_ctx = { 449 .realinode = inode, 450 }; 451 452 return iomap_dio_rw(iocb, to, &erofs_iomap_ops, 453 NULL, 0, &iter_ctx, 0); 454 } 455 return filemap_read(iocb, to, 0); 456 } 457 458 /* for uncompressed (aligned) files and raw access for other files */ 459 const struct address_space_operations erofs_aops = { 460 .read_folio = erofs_read_folio, 461 .readahead = erofs_readahead, 462 .bmap = erofs_bmap, 463 .direct_IO = noop_direct_IO, 464 .release_folio = iomap_release_folio, 465 .invalidate_folio = iomap_invalidate_folio, 466 }; 467 468 #ifdef CONFIG_FS_DAX 469 static vm_fault_t erofs_dax_huge_fault(struct vm_fault *vmf, 470 unsigned int order) 471 { 472 return dax_iomap_fault(vmf, order, NULL, NULL, &erofs_iomap_ops); 473 } 474 475 static vm_fault_t erofs_dax_fault(struct vm_fault *vmf) 476 { 477 return erofs_dax_huge_fault(vmf, 0); 478 } 479 480 static const struct vm_operations_struct erofs_dax_vm_ops = { 481 .fault = erofs_dax_fault, 482 .huge_fault = erofs_dax_huge_fault, 483 }; 484 485 static int erofs_file_mmap_prepare(struct vm_area_desc *desc) 486 { 487 if (!IS_DAX(file_inode(desc->file))) 488 return generic_file_readonly_mmap_prepare(desc); 489 490 if (vma_desc_test_flags(desc, VMA_SHARED_BIT) && 491 vma_desc_test_flags(desc, VMA_MAYWRITE_BIT)) 492 return -EINVAL; 493 494 desc->vm_ops = &erofs_dax_vm_ops; 495 vma_desc_set_flags(desc, VMA_HUGEPAGE_BIT); 496 return 0; 497 } 498 #else 499 #define erofs_file_mmap_prepare generic_file_readonly_mmap_prepare 500 #endif 501 502 static loff_t erofs_file_llseek(struct file *file, loff_t offset, int whence) 503 { 504 struct inode *inode = file->f_mapping->host; 505 const struct iomap_ops *ops = &erofs_iomap_ops; 506 507 if (erofs_inode_is_data_compressed(EROFS_I(inode)->datalayout)) { 508 if (!IS_ENABLED(CONFIG_EROFS_FS_ZIP)) 509 return generic_file_llseek(file, offset, whence); 510 ops = &z_erofs_iomap_report_ops; 511 } 512 513 if (whence == SEEK_HOLE) 514 offset = iomap_seek_hole(inode, offset, ops); 515 else if (whence == SEEK_DATA) 516 offset = iomap_seek_data(inode, offset, ops); 517 else 518 return generic_file_llseek(file, offset, whence); 519 520 if (offset < 0) 521 return offset; 522 return vfs_setpos(file, offset, inode->i_sb->s_maxbytes); 523 } 524 525 const struct file_operations erofs_file_fops = { 526 .llseek = erofs_file_llseek, 527 .read_iter = erofs_file_read_iter, 528 .unlocked_ioctl = erofs_ioctl, 529 #ifdef CONFIG_COMPAT 530 .compat_ioctl = erofs_compat_ioctl, 531 #endif 532 .mmap_prepare = erofs_file_mmap_prepare, 533 .get_unmapped_area = thp_get_unmapped_area, 534 .splice_read = filemap_splice_read, 535 .setlease = generic_setlease, 536 }; 537