fs/erofs/data.c

   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/sched/mm.h>
   9 #include <trace/events/erofs.h>
  10
  11 void erofs_unmap_metabuf(struct erofs_buf *buf)
  12 {
  13         if (!buf->base)
  14                 return;
  15         kunmap_local(buf->base);
  16         buf->base = NULL;
  17 }
  18
  19 void erofs_put_metabuf(struct erofs_buf *buf)
  20 {
  21         if (!buf->page)
  22                 return;
  23         erofs_unmap_metabuf(buf);
  24         folio_put(page_folio(buf->page));
  25         buf->page = NULL;
  26 }
  27
  28 void *erofs_bread(struct erofs_buf *buf, erofs_off_t offset,
  29                   enum erofs_kmap_type type)
  30 {
  31         pgoff_t index = 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 (!buf->base && type == EROFS_KMAP)
  47                 buf->base = kmap_local_page(buf->page);
  48         if (type == EROFS_NO_KMAP)
  49                 return NULL;
  50         return buf->base + (offset & ~PAGE_MASK);
  51 }
  52
  53 void erofs_init_metabuf(struct erofs_buf *buf, struct super_block *sb)
  54 {
  55         struct erofs_sb_info *sbi = EROFS_SB(sb);
  56
  57         buf->file = NULL;
  58         if (erofs_is_fileio_mode(sbi)) {
  59                 buf->file = sbi->fdev;          /* some fs like FUSE needs it */
  60                 buf->mapping = buf->file->f_mapping;
  61         } else if (erofs_is_fscache_mode(sb))
  62                 buf->mapping = sbi->s_fscache->inode->i_mapping;
  63         else
  64                 buf->mapping = sb->s_bdev->bd_mapping;
  65 }
  66
  67 void *erofs_read_metabuf(struct erofs_buf *buf, struct super_block *sb,
  68                          erofs_off_t offset, enum erofs_kmap_type type)
  69 {
  70         erofs_init_metabuf(buf, sb);
  71         return erofs_bread(buf, offset, type);
  72 }
  73
  74 static int erofs_map_blocks_flatmode(struct inode *inode,
  75                                      struct erofs_map_blocks *map)
  76 {
  77         struct erofs_inode *vi = EROFS_I(inode);
  78         struct super_block *sb = inode->i_sb;
  79         bool tailendpacking = (vi->datalayout == EROFS_INODE_FLAT_INLINE);
  80         erofs_blk_t lastblk = erofs_iblks(inode) - tailendpacking;
  81
  82         map->m_flags = EROFS_MAP_MAPPED;        /* no hole in flat inodes */
  83         if (map->m_la < erofs_pos(sb, lastblk)) {
  84                 map->m_pa = erofs_pos(sb, vi->raw_blkaddr) + map->m_la;
  85                 map->m_plen = erofs_pos(sb, lastblk) - map->m_la;
  86         } else {
  87                 DBG_BUGON(!tailendpacking);
  88                 map->m_pa = erofs_iloc(inode) + vi->inode_isize +
  89                         vi->xattr_isize + erofs_blkoff(sb, map->m_la);
  90                 map->m_plen = inode->i_size - map->m_la;
  91
  92                 /* inline data should be located in the same meta block */
  93                 if (erofs_blkoff(sb, map->m_pa) + map->m_plen > sb->s_blocksize) {
  94                         erofs_err(sb, "inline data across blocks @ nid %llu", vi->nid);
  95                         DBG_BUGON(1);
  96                         return -EFSCORRUPTED;
  97                 }
  98                 map->m_flags |= EROFS_MAP_META;
  99         }
 100         return 0;
 101 }
 102
 103 int erofs_map_blocks(struct inode *inode, struct erofs_map_blocks *map)
 104 {
 105         struct super_block *sb = inode->i_sb;
 106         struct erofs_inode *vi = EROFS_I(inode);
 107         struct erofs_inode_chunk_index *idx;
 108         struct erofs_buf buf = __EROFS_BUF_INITIALIZER;
 109         u64 chunknr;
 110         unsigned int unit;
 111         erofs_off_t pos;
 112         void *kaddr;
 113         int err = 0;
 114
 115         trace_erofs_map_blocks_enter(inode, map, 0);
 116         map->m_deviceid = 0;
 117         if (map->m_la >= inode->i_size) {
 118                 /* leave out-of-bound access unmapped */
 119                 map->m_flags = 0;
 120                 map->m_plen = map->m_llen;
 121                 goto out;
 122         }
 123
 124         if (vi->datalayout != EROFS_INODE_CHUNK_BASED) {
 125                 err = erofs_map_blocks_flatmode(inode, map);
 126                 goto out;
 127         }
 128
 129         if (vi->chunkformat & EROFS_CHUNK_FORMAT_INDEXES)
 130                 unit = sizeof(*idx);                    /* chunk index */
 131         else
 132                 unit = EROFS_BLOCK_MAP_ENTRY_SIZE;      /* block map */
 133
 134         chunknr = map->m_la >> vi->chunkbits;
 135         pos = ALIGN(erofs_iloc(inode) + vi->inode_isize +
 136                     vi->xattr_isize, unit) + unit * chunknr;
 137
 138         kaddr = erofs_read_metabuf(&buf, sb, pos, EROFS_KMAP);
 139         if (IS_ERR(kaddr)) {
 140                 err = PTR_ERR(kaddr);
 141                 goto out;
 142         }
 143         map->m_la = chunknr << vi->chunkbits;
 144         map->m_plen = min_t(erofs_off_t, 1UL << vi->chunkbits,
 145                         round_up(inode->i_size - map->m_la, sb->s_blocksize));
 146
 147         /* handle block map */
 148         if (!(vi->chunkformat & EROFS_CHUNK_FORMAT_INDEXES)) {
 149                 __le32 *blkaddr = kaddr;
 150
 151                 if (le32_to_cpu(*blkaddr) == EROFS_NULL_ADDR) {
 152                         map->m_flags = 0;
 153                 } else {
 154                         map->m_pa = erofs_pos(sb, le32_to_cpu(*blkaddr));
 155                         map->m_flags = EROFS_MAP_MAPPED;
 156                 }
 157                 goto out_unlock;
 158         }
 159         /* parse chunk indexes */
 160         idx = kaddr;
 161         switch (le32_to_cpu(idx->blkaddr)) {
 162         case EROFS_NULL_ADDR:
 163                 map->m_flags = 0;
 164                 break;
 165         default:
 166                 map->m_deviceid = le16_to_cpu(idx->device_id) &
 167                         EROFS_SB(sb)->device_id_mask;
 168                 map->m_pa = erofs_pos(sb, le32_to_cpu(idx->blkaddr));
 169                 map->m_flags = EROFS_MAP_MAPPED;
 170                 break;
 171         }
 172 out_unlock:
 173         erofs_put_metabuf(&buf);
 174 out:
 175         if (!err)
 176                 map->m_llen = map->m_plen;
 177         trace_erofs_map_blocks_exit(inode, map, 0, err);
 178         return err;
 179 }
 180
 181 static void erofs_fill_from_devinfo(struct erofs_map_dev *map,
 182                                     struct erofs_device_info *dif)
 183 {
 184         map->m_bdev = NULL;
 185         map->m_fp = NULL;
 186         if (dif->file) {
 187                 if (S_ISBLK(file_inode(dif->file)->i_mode))
 188                         map->m_bdev = file_bdev(dif->file);
 189                 else
 190                         map->m_fp = dif->file;
 191         }
 192         map->m_daxdev = dif->dax_dev;
 193         map->m_dax_part_off = dif->dax_part_off;
 194         map->m_fscache = dif->fscache;
 195 }
 196
 197 int erofs_map_dev(struct super_block *sb, struct erofs_map_dev *map)
 198 {
 199         struct erofs_dev_context *devs = EROFS_SB(sb)->devs;
 200         struct erofs_device_info *dif;
 201         erofs_off_t startoff, length;
 202         int id;
 203
 204         map->m_bdev = sb->s_bdev;
 205         map->m_daxdev = EROFS_SB(sb)->dax_dev;
 206         map->m_dax_part_off = EROFS_SB(sb)->dax_part_off;
 207         map->m_fscache = EROFS_SB(sb)->s_fscache;
 208         map->m_fp = EROFS_SB(sb)->fdev;
 209
 210         if (map->m_deviceid) {
 211                 down_read(&devs->rwsem);
 212                 dif = idr_find(&devs->tree, map->m_deviceid - 1);
 213                 if (!dif) {
 214                         up_read(&devs->rwsem);
 215                         return -ENODEV;
 216                 }
 217                 if (devs->flatdev) {
 218                         map->m_pa += erofs_pos(sb, dif->mapped_blkaddr);
 219                         up_read(&devs->rwsem);
 220                         return 0;
 221                 }
 222                 erofs_fill_from_devinfo(map, dif);
 223                 up_read(&devs->rwsem);
 224         } else if (devs->extra_devices && !devs->flatdev) {
 225                 down_read(&devs->rwsem);
 226                 idr_for_each_entry(&devs->tree, dif, id) {
 227                         if (!dif->mapped_blkaddr)
 228                                 continue;
 229
 230                         startoff = erofs_pos(sb, dif->mapped_blkaddr);
 231                         length = erofs_pos(sb, dif->blocks);
 232                         if (map->m_pa >= startoff &&
 233                             map->m_pa < startoff + length) {
 234                                 map->m_pa -= startoff;
 235                                 erofs_fill_from_devinfo(map, dif);
 236                                 break;
 237                         }
 238                 }
 239                 up_read(&devs->rwsem);
 240         }
 241         return 0;
 242 }
 243
 244 /*
 245  * bit 30: I/O error occurred on this folio
 246  * bit 0 - 29: remaining parts to complete this folio
 247  */
 248 #define EROFS_ONLINEFOLIO_EIO                   (1 << 30)
 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)
 266 {
 267         int orig, v;
 268
 269         do {
 270                 orig = atomic_read((atomic_t *)&folio->private);
 271                 v = (orig - 1) | (err ? EROFS_ONLINEFOLIO_EIO : 0);
 272         } while (atomic_cmpxchg((atomic_t *)&folio->private, orig, v) != orig);
 273
 274         if (v & ~EROFS_ONLINEFOLIO_EIO)
 275                 return;
 276         folio->private = 0;
 277         folio_end_read(folio, !(v & EROFS_ONLINEFOLIO_EIO));
 278 }
 279
 280 static int erofs_iomap_begin(struct inode *inode, loff_t offset, loff_t length,
 281                 unsigned int flags, struct iomap *iomap, struct iomap *srcmap)
 282 {
 283         int ret;
 284         struct super_block *sb = inode->i_sb;
 285         struct erofs_map_blocks map;
 286         struct erofs_map_dev mdev;
 287
 288         map.m_la = offset;
 289         map.m_llen = length;
 290
 291         ret = erofs_map_blocks(inode, &map);
 292         if (ret < 0)
 293                 return ret;
 294
 295         mdev = (struct erofs_map_dev) {
 296                 .m_deviceid = map.m_deviceid,
 297                 .m_pa = map.m_pa,
 298         };
 299         ret = erofs_map_dev(sb, &mdev);
 300         if (ret)
 301                 return ret;
 302
 303         iomap->offset = map.m_la;
 304         if (flags & IOMAP_DAX)
 305                 iomap->dax_dev = mdev.m_daxdev;
 306         else
 307                 iomap->bdev = mdev.m_bdev;
 308         iomap->length = map.m_llen;
 309         iomap->flags = 0;
 310         iomap->private = NULL;
 311
 312         if (!(map.m_flags & EROFS_MAP_MAPPED)) {
 313                 iomap->type = IOMAP_HOLE;
 314                 iomap->addr = IOMAP_NULL_ADDR;
 315                 if (!iomap->length)
 316                         iomap->length = length;
 317                 return 0;
 318         }
 319
 320         if (map.m_flags & EROFS_MAP_META) {
 321                 void *ptr;
 322                 struct erofs_buf buf = __EROFS_BUF_INITIALIZER;
 323
 324                 iomap->type = IOMAP_INLINE;
 325                 ptr = erofs_read_metabuf(&buf, sb, mdev.m_pa, EROFS_KMAP);
 326                 if (IS_ERR(ptr))
 327                         return PTR_ERR(ptr);
 328                 iomap->inline_data = ptr;
 329                 iomap->private = buf.base;
 330         } else {
 331                 iomap->type = IOMAP_MAPPED;
 332                 iomap->addr = mdev.m_pa;
 333                 if (flags & IOMAP_DAX)
 334                         iomap->addr += mdev.m_dax_part_off;
 335         }
 336         return 0;
 337 }
 338
 339 static int erofs_iomap_end(struct inode *inode, loff_t pos, loff_t length,
 340                 ssize_t written, unsigned int flags, struct iomap *iomap)
 341 {
 342         void *ptr = iomap->private;
 343
 344         if (ptr) {
 345                 struct erofs_buf buf = {
 346                         .page = kmap_to_page(ptr),
 347                         .base = ptr,
 348                 };
 349
 350                 DBG_BUGON(iomap->type != IOMAP_INLINE);
 351                 erofs_put_metabuf(&buf);
 352         } else {
 353                 DBG_BUGON(iomap->type == IOMAP_INLINE);
 354         }
 355         return written;
 356 }
 357
 358 static const struct iomap_ops erofs_iomap_ops = {
 359         .iomap_begin = erofs_iomap_begin,
 360         .iomap_end = erofs_iomap_end,
 361 };
 362
 363 int erofs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
 364                  u64 start, u64 len)
 365 {
 366         if (erofs_inode_is_data_compressed(EROFS_I(inode)->datalayout)) {
 367 #ifdef CONFIG_EROFS_FS_ZIP
 368                 return iomap_fiemap(inode, fieinfo, start, len,
 369                                     &z_erofs_iomap_report_ops);
 370 #else
 371                 return -EOPNOTSUPP;
 372 #endif
 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         return iomap_read_folio(folio, &erofs_iomap_ops);
 384 }
 385
 386 static void erofs_readahead(struct readahead_control *rac)
 387 {
 388         return iomap_readahead(rac, &erofs_iomap_ops);
 389 }
 390
 391 static sector_t erofs_bmap(struct address_space *mapping, sector_t block)
 392 {
 393         return iomap_bmap(mapping, block, &erofs_iomap_ops);
 394 }
 395
 396 static ssize_t erofs_file_read_iter(struct kiocb *iocb, struct iov_iter *to)
 397 {
 398         struct inode *inode = file_inode(iocb->ki_filp);
 399
 400         /* no need taking (shared) inode lock since it's a ro filesystem */
 401         if (!iov_iter_count(to))
 402                 return 0;
 403
 404 #ifdef CONFIG_FS_DAX
 405         if (IS_DAX(inode))
 406                 return dax_iomap_rw(iocb, to, &erofs_iomap_ops);
 407 #endif
 408         if ((iocb->ki_flags & IOCB_DIRECT) && inode->i_sb->s_bdev)
 409                 return iomap_dio_rw(iocb, to, &erofs_iomap_ops,
 410                                     NULL, 0, NULL, 0);
 411         return filemap_read(iocb, to, 0);
 412 }
 413
 414 /* for uncompressed (aligned) files and raw access for other files */
 415 const struct address_space_operations erofs_aops = {
 416         .read_folio = erofs_read_folio,
 417         .readahead = erofs_readahead,
 418         .bmap = erofs_bmap,
 419         .direct_IO = noop_direct_IO,
 420         .release_folio = iomap_release_folio,
 421         .invalidate_folio = iomap_invalidate_folio,
 422 };
 423
 424 #ifdef CONFIG_FS_DAX
 425 static vm_fault_t erofs_dax_huge_fault(struct vm_fault *vmf,
 426                 unsigned int order)
 427 {
 428         return dax_iomap_fault(vmf, order, NULL, NULL, &erofs_iomap_ops);
 429 }
 430
 431 static vm_fault_t erofs_dax_fault(struct vm_fault *vmf)
 432 {
 433         return erofs_dax_huge_fault(vmf, 0);
 434 }
 435
 436 static const struct vm_operations_struct erofs_dax_vm_ops = {
 437         .fault          = erofs_dax_fault,
 438         .huge_fault     = erofs_dax_huge_fault,
 439 };
 440
 441 static int erofs_file_mmap(struct file *file, struct vm_area_struct *vma)
 442 {
 443         if (!IS_DAX(file_inode(file)))
 444                 return generic_file_readonly_mmap(file, vma);
 445
 446         if ((vma->vm_flags & VM_SHARED) && (vma->vm_flags & VM_MAYWRITE))
 447                 return -EINVAL;
 448
 449         vma->vm_ops = &erofs_dax_vm_ops;
 450         vm_flags_set(vma, VM_HUGEPAGE);
 451         return 0;
 452 }
 453 #else
 454 #define erofs_file_mmap generic_file_readonly_mmap
 455 #endif
 456
 457 static loff_t erofs_file_llseek(struct file *file, loff_t offset, int whence)
 458 {
 459         struct inode *inode = file->f_mapping->host;
 460         const struct iomap_ops *ops = &erofs_iomap_ops;
 461
 462         if (erofs_inode_is_data_compressed(EROFS_I(inode)->datalayout))
 463 #ifdef CONFIG_EROFS_FS_ZIP
 464                 ops = &z_erofs_iomap_report_ops;
 465 #else
 466                 return generic_file_llseek(file, offset, whence);
 467 #endif
 468
 469         if (whence == SEEK_HOLE)
 470                 offset = iomap_seek_hole(inode, offset, ops);
 471         else if (whence == SEEK_DATA)
 472                 offset = iomap_seek_data(inode, offset, ops);
 473         else
 474                 return generic_file_llseek(file, offset, whence);
 475
 476         if (offset < 0)
 477                 return offset;
 478         return vfs_setpos(file, offset, inode->i_sb->s_maxbytes);
 479 }
 480
 481 const struct file_operations erofs_file_fops = {
 482         .llseek         = erofs_file_llseek,
 483         .read_iter      = erofs_file_read_iter,
 484         .mmap           = erofs_file_mmap,
 485         .get_unmapped_area = thp_get_unmapped_area,
 486         .splice_read    = filemap_splice_read,
 487 };