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->kmap_type == EROFS_KMAP)
  14                 kunmap_local(buf->base);
  15         buf->base = NULL;
  16         buf->kmap_type = EROFS_NO_KMAP;
  17 }
  18
  19 void erofs_put_metabuf(struct erofs_buf *buf)
  20 {
  21         if (!buf->page)
  22                 return;
  23         erofs_unmap_metabuf(buf);
  24         put_page(buf->page);
  25         buf->page = NULL;
  26 }
  27
  28 /*
  29  * Derive the block size from inode->i_blkbits to make compatible with
  30  * anonymous inode in fscache mode.
  31  */
  32 void *erofs_bread(struct erofs_buf *buf, erofs_off_t offset,
  33                   enum erofs_kmap_type type)
  34 {
  35         pgoff_t index = offset >> PAGE_SHIFT;
  36         struct page *page = buf->page;
  37         struct folio *folio;
  38         unsigned int nofs_flag;
  39
  40         if (!page || page->index != index) {
  41                 erofs_put_metabuf(buf);
  42
  43                 nofs_flag = memalloc_nofs_save();
  44                 folio = read_cache_folio(buf->mapping, index, NULL, NULL);
  45                 memalloc_nofs_restore(nofs_flag);
  46                 if (IS_ERR(folio))
  47                         return folio;
  48
  49                 /* should already be PageUptodate, no need to lock page */
  50                 page = folio_file_page(folio, index);
  51                 buf->page = page;
  52         }
  53         if (buf->kmap_type == EROFS_NO_KMAP) {
  54                 if (type == EROFS_KMAP)
  55                         buf->base = kmap_local_page(page);
  56                 buf->kmap_type = type;
  57         } else if (buf->kmap_type != type) {
  58                 DBG_BUGON(1);
  59                 return ERR_PTR(-EFAULT);
  60         }
  61         if (type == EROFS_NO_KMAP)
  62                 return NULL;
  63         return buf->base + (offset & ~PAGE_MASK);
  64 }
  65
  66 void erofs_init_metabuf(struct erofs_buf *buf, struct super_block *sb)
  67 {
  68         if (erofs_is_fscache_mode(sb))
  69                 buf->mapping = EROFS_SB(sb)->s_fscache->inode->i_mapping;
  70         else
  71                 buf->mapping = sb->s_bdev->bd_mapping;
  72 }
  73
  74 void *erofs_read_metabuf(struct erofs_buf *buf, struct super_block *sb,
  75                          erofs_off_t offset, enum erofs_kmap_type type)
  76 {
  77         erofs_init_metabuf(buf, sb);
  78         return erofs_bread(buf, offset, type);
  79 }
  80
  81 static int erofs_map_blocks_flatmode(struct inode *inode,
  82                                      struct erofs_map_blocks *map)
  83 {
  84         erofs_blk_t nblocks, lastblk;
  85         u64 offset = map->m_la;
  86         struct erofs_inode *vi = EROFS_I(inode);
  87         struct super_block *sb = inode->i_sb;
  88         bool tailendpacking = (vi->datalayout == EROFS_INODE_FLAT_INLINE);
  89
  90         nblocks = erofs_iblks(inode);
  91         lastblk = nblocks - tailendpacking;
  92
  93         /* there is no hole in flatmode */
  94         map->m_flags = EROFS_MAP_MAPPED;
  95         if (offset < erofs_pos(sb, lastblk)) {
  96                 map->m_pa = erofs_pos(sb, vi->raw_blkaddr) + map->m_la;
  97                 map->m_plen = erofs_pos(sb, lastblk) - offset;
  98         } else if (tailendpacking) {
  99                 map->m_pa = erofs_iloc(inode) + vi->inode_isize +
 100                         vi->xattr_isize + erofs_blkoff(sb, offset);
 101                 map->m_plen = inode->i_size - offset;
 102
 103                 /* inline data should be located in the same meta block */
 104                 if (erofs_blkoff(sb, map->m_pa) + map->m_plen > sb->s_blocksize) {
 105                         erofs_err(sb, "inline data cross block boundary @ nid %llu",
 106                                   vi->nid);
 107                         DBG_BUGON(1);
 108                         return -EFSCORRUPTED;
 109                 }
 110                 map->m_flags |= EROFS_MAP_META;
 111         } else {
 112                 erofs_err(sb, "internal error @ nid: %llu (size %llu), m_la 0x%llx",
 113                           vi->nid, inode->i_size, map->m_la);
 114                 DBG_BUGON(1);
 115                 return -EIO;
 116         }
 117         return 0;
 118 }
 119
 120 int erofs_map_blocks(struct inode *inode, struct erofs_map_blocks *map)
 121 {
 122         struct super_block *sb = inode->i_sb;
 123         struct erofs_inode *vi = EROFS_I(inode);
 124         struct erofs_inode_chunk_index *idx;
 125         struct erofs_buf buf = __EROFS_BUF_INITIALIZER;
 126         u64 chunknr;
 127         unsigned int unit;
 128         erofs_off_t pos;
 129         void *kaddr;
 130         int err = 0;
 131
 132         trace_erofs_map_blocks_enter(inode, map, 0);
 133         map->m_deviceid = 0;
 134         if (map->m_la >= inode->i_size) {
 135                 /* leave out-of-bound access unmapped */
 136                 map->m_flags = 0;
 137                 map->m_plen = 0;
 138                 goto out;
 139         }
 140
 141         if (vi->datalayout != EROFS_INODE_CHUNK_BASED) {
 142                 err = erofs_map_blocks_flatmode(inode, map);
 143                 goto out;
 144         }
 145
 146         if (vi->chunkformat & EROFS_CHUNK_FORMAT_INDEXES)
 147                 unit = sizeof(*idx);                    /* chunk index */
 148         else
 149                 unit = EROFS_BLOCK_MAP_ENTRY_SIZE;      /* block map */
 150
 151         chunknr = map->m_la >> vi->chunkbits;
 152         pos = ALIGN(erofs_iloc(inode) + vi->inode_isize +
 153                     vi->xattr_isize, unit) + unit * chunknr;
 154
 155         kaddr = erofs_read_metabuf(&buf, sb, pos, EROFS_KMAP);
 156         if (IS_ERR(kaddr)) {
 157                 err = PTR_ERR(kaddr);
 158                 goto out;
 159         }
 160         map->m_la = chunknr << vi->chunkbits;
 161         map->m_plen = min_t(erofs_off_t, 1UL << vi->chunkbits,
 162                         round_up(inode->i_size - map->m_la, sb->s_blocksize));
 163
 164         /* handle block map */
 165         if (!(vi->chunkformat & EROFS_CHUNK_FORMAT_INDEXES)) {
 166                 __le32 *blkaddr = kaddr;
 167
 168                 if (le32_to_cpu(*blkaddr) == EROFS_NULL_ADDR) {
 169                         map->m_flags = 0;
 170                 } else {
 171                         map->m_pa = erofs_pos(sb, le32_to_cpu(*blkaddr));
 172                         map->m_flags = EROFS_MAP_MAPPED;
 173                 }
 174                 goto out_unlock;
 175         }
 176         /* parse chunk indexes */
 177         idx = kaddr;
 178         switch (le32_to_cpu(idx->blkaddr)) {
 179         case EROFS_NULL_ADDR:
 180                 map->m_flags = 0;
 181                 break;
 182         default:
 183                 map->m_deviceid = le16_to_cpu(idx->device_id) &
 184                         EROFS_SB(sb)->device_id_mask;
 185                 map->m_pa = erofs_pos(sb, le32_to_cpu(idx->blkaddr));
 186                 map->m_flags = EROFS_MAP_MAPPED;
 187                 break;
 188         }
 189 out_unlock:
 190         erofs_put_metabuf(&buf);
 191 out:
 192         if (!err)
 193                 map->m_llen = map->m_plen;
 194         trace_erofs_map_blocks_exit(inode, map, 0, err);
 195         return err;
 196 }
 197
 198 int erofs_map_dev(struct super_block *sb, struct erofs_map_dev *map)
 199 {
 200         struct erofs_dev_context *devs = EROFS_SB(sb)->devs;
 201         struct erofs_device_info *dif;
 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
 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->mapped_blkaddr);
 218                         up_read(&devs->rwsem);
 219                         return 0;
 220                 }
 221                 map->m_bdev = dif->bdev_file ? file_bdev(dif->bdev_file) : NULL;
 222                 map->m_daxdev = dif->dax_dev;
 223                 map->m_dax_part_off = dif->dax_part_off;
 224                 map->m_fscache = dif->fscache;
 225                 up_read(&devs->rwsem);
 226         } else if (devs->extra_devices && !devs->flatdev) {
 227                 down_read(&devs->rwsem);
 228                 idr_for_each_entry(&devs->tree, dif, id) {
 229                         erofs_off_t startoff, length;
 230
 231                         if (!dif->mapped_blkaddr)
 232                                 continue;
 233                         startoff = erofs_pos(sb, dif->mapped_blkaddr);
 234                         length = erofs_pos(sb, dif->blocks);
 235
 236                         if (map->m_pa >= startoff &&
 237                             map->m_pa < startoff + length) {
 238                                 map->m_pa -= startoff;
 239                                 map->m_bdev = dif->bdev_file ?
 240                                               file_bdev(dif->bdev_file) : NULL;
 241                                 map->m_daxdev = dif->dax_dev;
 242                                 map->m_dax_part_off = dif->dax_part_off;
 243                                 map->m_fscache = dif->fscache;
 244                                 break;
 245                         }
 246                 }
 247                 up_read(&devs->rwsem);
 248         }
 249         return 0;
 250 }
 251
 252 static int erofs_iomap_begin(struct inode *inode, loff_t offset, loff_t length,
 253                 unsigned int flags, struct iomap *iomap, struct iomap *srcmap)
 254 {
 255         int ret;
 256         struct super_block *sb = inode->i_sb;
 257         struct erofs_map_blocks map;
 258         struct erofs_map_dev mdev;
 259
 260         map.m_la = offset;
 261         map.m_llen = length;
 262
 263         ret = erofs_map_blocks(inode, &map);
 264         if (ret < 0)
 265                 return ret;
 266
 267         mdev = (struct erofs_map_dev) {
 268                 .m_deviceid = map.m_deviceid,
 269                 .m_pa = map.m_pa,
 270         };
 271         ret = erofs_map_dev(sb, &mdev);
 272         if (ret)
 273                 return ret;
 274
 275         iomap->offset = map.m_la;
 276         if (flags & IOMAP_DAX)
 277                 iomap->dax_dev = mdev.m_daxdev;
 278         else
 279                 iomap->bdev = mdev.m_bdev;
 280         iomap->length = map.m_llen;
 281         iomap->flags = 0;
 282         iomap->private = NULL;
 283
 284         if (!(map.m_flags & EROFS_MAP_MAPPED)) {
 285                 iomap->type = IOMAP_HOLE;
 286                 iomap->addr = IOMAP_NULL_ADDR;
 287                 if (!iomap->length)
 288                         iomap->length = length;
 289                 return 0;
 290         }
 291
 292         if (map.m_flags & EROFS_MAP_META) {
 293                 void *ptr;
 294                 struct erofs_buf buf = __EROFS_BUF_INITIALIZER;
 295
 296                 iomap->type = IOMAP_INLINE;
 297                 ptr = erofs_read_metabuf(&buf, sb, mdev.m_pa, EROFS_KMAP);
 298                 if (IS_ERR(ptr))
 299                         return PTR_ERR(ptr);
 300                 iomap->inline_data = ptr;
 301                 iomap->private = buf.base;
 302         } else {
 303                 iomap->type = IOMAP_MAPPED;
 304                 iomap->addr = mdev.m_pa;
 305                 if (flags & IOMAP_DAX)
 306                         iomap->addr += mdev.m_dax_part_off;
 307         }
 308         return 0;
 309 }
 310
 311 static int erofs_iomap_end(struct inode *inode, loff_t pos, loff_t length,
 312                 ssize_t written, unsigned int flags, struct iomap *iomap)
 313 {
 314         void *ptr = iomap->private;
 315
 316         if (ptr) {
 317                 struct erofs_buf buf = {
 318                         .page = kmap_to_page(ptr),
 319                         .base = ptr,
 320                         .kmap_type = EROFS_KMAP,
 321                 };
 322
 323                 DBG_BUGON(iomap->type != IOMAP_INLINE);
 324                 erofs_put_metabuf(&buf);
 325         } else {
 326                 DBG_BUGON(iomap->type == IOMAP_INLINE);
 327         }
 328         return written;
 329 }
 330
 331 static const struct iomap_ops erofs_iomap_ops = {
 332         .iomap_begin = erofs_iomap_begin,
 333         .iomap_end = erofs_iomap_end,
 334 };
 335
 336 int erofs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
 337                  u64 start, u64 len)
 338 {
 339         if (erofs_inode_is_data_compressed(EROFS_I(inode)->datalayout)) {
 340 #ifdef CONFIG_EROFS_FS_ZIP
 341                 return iomap_fiemap(inode, fieinfo, start, len,
 342                                     &z_erofs_iomap_report_ops);
 343 #else
 344                 return -EOPNOTSUPP;
 345 #endif
 346         }
 347         return iomap_fiemap(inode, fieinfo, start, len, &erofs_iomap_ops);
 348 }
 349
 350 /*
 351  * since we dont have write or truncate flows, so no inode
 352  * locking needs to be held at the moment.
 353  */
 354 static int erofs_read_folio(struct file *file, struct folio *folio)
 355 {
 356         return iomap_read_folio(folio, &erofs_iomap_ops);
 357 }
 358
 359 static void erofs_readahead(struct readahead_control *rac)
 360 {
 361         return iomap_readahead(rac, &erofs_iomap_ops);
 362 }
 363
 364 static sector_t erofs_bmap(struct address_space *mapping, sector_t block)
 365 {
 366         return iomap_bmap(mapping, block, &erofs_iomap_ops);
 367 }
 368
 369 static ssize_t erofs_file_read_iter(struct kiocb *iocb, struct iov_iter *to)
 370 {
 371         struct inode *inode = file_inode(iocb->ki_filp);
 372
 373         /* no need taking (shared) inode lock since it's a ro filesystem */
 374         if (!iov_iter_count(to))
 375                 return 0;
 376
 377 #ifdef CONFIG_FS_DAX
 378         if (IS_DAX(inode))
 379                 return dax_iomap_rw(iocb, to, &erofs_iomap_ops);
 380 #endif
 381         if (iocb->ki_flags & IOCB_DIRECT) {
 382                 struct block_device *bdev = inode->i_sb->s_bdev;
 383                 unsigned int blksize_mask;
 384
 385                 if (bdev)
 386                         blksize_mask = bdev_logical_block_size(bdev) - 1;
 387                 else
 388                         blksize_mask = i_blocksize(inode) - 1;
 389
 390                 if ((iocb->ki_pos | iov_iter_count(to) |
 391                      iov_iter_alignment(to)) & blksize_mask)
 392                         return -EINVAL;
 393
 394                 return iomap_dio_rw(iocb, to, &erofs_iomap_ops,
 395                                     NULL, 0, NULL, 0);
 396         }
 397         return filemap_read(iocb, to, 0);
 398 }
 399
 400 /* for uncompressed (aligned) files and raw access for other files */
 401 const struct address_space_operations erofs_raw_access_aops = {
 402         .read_folio = erofs_read_folio,
 403         .readahead = erofs_readahead,
 404         .bmap = erofs_bmap,
 405         .direct_IO = noop_direct_IO,
 406         .release_folio = iomap_release_folio,
 407         .invalidate_folio = iomap_invalidate_folio,
 408 };
 409
 410 #ifdef CONFIG_FS_DAX
 411 static vm_fault_t erofs_dax_huge_fault(struct vm_fault *vmf,
 412                 unsigned int order)
 413 {
 414         return dax_iomap_fault(vmf, order, NULL, NULL, &erofs_iomap_ops);
 415 }
 416
 417 static vm_fault_t erofs_dax_fault(struct vm_fault *vmf)
 418 {
 419         return erofs_dax_huge_fault(vmf, 0);
 420 }
 421
 422 static const struct vm_operations_struct erofs_dax_vm_ops = {
 423         .fault          = erofs_dax_fault,
 424         .huge_fault     = erofs_dax_huge_fault,
 425 };
 426
 427 static int erofs_file_mmap(struct file *file, struct vm_area_struct *vma)
 428 {
 429         if (!IS_DAX(file_inode(file)))
 430                 return generic_file_readonly_mmap(file, vma);
 431
 432         if ((vma->vm_flags & VM_SHARED) && (vma->vm_flags & VM_MAYWRITE))
 433                 return -EINVAL;
 434
 435         vma->vm_ops = &erofs_dax_vm_ops;
 436         vm_flags_set(vma, VM_HUGEPAGE);
 437         return 0;
 438 }
 439 #else
 440 #define erofs_file_mmap generic_file_readonly_mmap
 441 #endif
 442
 443 const struct file_operations erofs_file_fops = {
 444         .llseek         = generic_file_llseek,
 445         .read_iter      = erofs_file_read_iter,
 446         .mmap           = erofs_file_mmap,
 447         .get_unmapped_area = thp_get_unmapped_area,
 448         .splice_read    = filemap_splice_read,
 449 };