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