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_blk_t blkaddr,
  33                   enum erofs_kmap_type type)
  34 {
  35         struct inode *inode = buf->inode;
  36         erofs_off_t offset = (erofs_off_t)blkaddr << inode->i_blkbits;
  37         pgoff_t index = offset >> PAGE_SHIFT;
  38         struct page *page = buf->page;
  39         struct folio *folio;
  40         unsigned int nofs_flag;
  41
  42         if (!page || page->index != index) {
  43                 erofs_put_metabuf(buf);
  44
  45                 nofs_flag = memalloc_nofs_save();
  46                 folio = read_cache_folio(inode->i_mapping, index, NULL, NULL);
  47                 memalloc_nofs_restore(nofs_flag);
  48                 if (IS_ERR(folio))
  49                         return folio;
  50
  51                 /* should already be PageUptodate, no need to lock page */
  52                 page = folio_file_page(folio, index);
  53                 buf->page = page;
  54         }
  55         if (buf->kmap_type == EROFS_NO_KMAP) {
  56                 if (type == EROFS_KMAP)
  57                         buf->base = kmap_local_page(page);
  58                 buf->kmap_type = type;
  59         } else if (buf->kmap_type != type) {
  60                 DBG_BUGON(1);
  61                 return ERR_PTR(-EFAULT);
  62         }
  63         if (type == EROFS_NO_KMAP)
  64                 return NULL;
  65         return buf->base + (offset & ~PAGE_MASK);
  66 }
  67
  68 void erofs_init_metabuf(struct erofs_buf *buf, struct super_block *sb)
  69 {
  70         if (erofs_is_fscache_mode(sb))
  71                 buf->inode = EROFS_SB(sb)->s_fscache->inode;
  72         else
  73                 buf->inode = sb->s_bdev->bd_inode;
  74 }
  75
  76 void *erofs_read_metabuf(struct erofs_buf *buf, struct super_block *sb,
  77                          erofs_blk_t blkaddr, enum erofs_kmap_type type)
  78 {
  79         erofs_init_metabuf(buf, sb);
  80         return erofs_bread(buf, blkaddr, type);
  81 }
  82
  83 static int erofs_map_blocks_flatmode(struct inode *inode,
  84                                      struct erofs_map_blocks *map)
  85 {
  86         erofs_blk_t nblocks, lastblk;
  87         u64 offset = map->m_la;
  88         struct erofs_inode *vi = EROFS_I(inode);
  89         struct super_block *sb = inode->i_sb;
  90         bool tailendpacking = (vi->datalayout == EROFS_INODE_FLAT_INLINE);
  91
  92         nblocks = erofs_iblks(inode);
  93         lastblk = nblocks - tailendpacking;
  94
  95         /* there is no hole in flatmode */
  96         map->m_flags = EROFS_MAP_MAPPED;
  97         if (offset < erofs_pos(sb, lastblk)) {
  98                 map->m_pa = erofs_pos(sb, vi->raw_blkaddr) + map->m_la;
  99                 map->m_plen = erofs_pos(sb, lastblk) - offset;
 100         } else if (tailendpacking) {
 101                 map->m_pa = erofs_iloc(inode) + vi->inode_isize +
 102                         vi->xattr_isize + erofs_blkoff(sb, offset);
 103                 map->m_plen = inode->i_size - offset;
 104
 105                 /* inline data should be located in the same meta block */
 106                 if (erofs_blkoff(sb, map->m_pa) + map->m_plen > sb->s_blocksize) {
 107                         erofs_err(sb, "inline data cross block boundary @ nid %llu",
 108                                   vi->nid);
 109                         DBG_BUGON(1);
 110                         return -EFSCORRUPTED;
 111                 }
 112                 map->m_flags |= EROFS_MAP_META;
 113         } else {
 114                 erofs_err(sb, "internal error @ nid: %llu (size %llu), m_la 0x%llx",
 115                           vi->nid, inode->i_size, map->m_la);
 116                 DBG_BUGON(1);
 117                 return -EIO;
 118         }
 119         return 0;
 120 }
 121
 122 int erofs_map_blocks(struct inode *inode, struct erofs_map_blocks *map)
 123 {
 124         struct super_block *sb = inode->i_sb;
 125         struct erofs_inode *vi = EROFS_I(inode);
 126         struct erofs_inode_chunk_index *idx;
 127         struct erofs_buf buf = __EROFS_BUF_INITIALIZER;
 128         u64 chunknr;
 129         unsigned int unit;
 130         erofs_off_t pos;
 131         void *kaddr;
 132         int err = 0;
 133
 134         trace_erofs_map_blocks_enter(inode, map, 0);
 135         map->m_deviceid = 0;
 136         if (map->m_la >= inode->i_size) {
 137                 /* leave out-of-bound access unmapped */
 138                 map->m_flags = 0;
 139                 map->m_plen = 0;
 140                 goto out;
 141         }
 142
 143         if (vi->datalayout != EROFS_INODE_CHUNK_BASED) {
 144                 err = erofs_map_blocks_flatmode(inode, map);
 145                 goto out;
 146         }
 147
 148         if (vi->chunkformat & EROFS_CHUNK_FORMAT_INDEXES)
 149                 unit = sizeof(*idx);                    /* chunk index */
 150         else
 151                 unit = EROFS_BLOCK_MAP_ENTRY_SIZE;      /* block map */
 152
 153         chunknr = map->m_la >> vi->chunkbits;
 154         pos = ALIGN(erofs_iloc(inode) + vi->inode_isize +
 155                     vi->xattr_isize, unit) + unit * chunknr;
 156
 157         kaddr = erofs_read_metabuf(&buf, sb, erofs_blknr(sb, pos), EROFS_KMAP);
 158         if (IS_ERR(kaddr)) {
 159                 err = PTR_ERR(kaddr);
 160                 goto out;
 161         }
 162         map->m_la = chunknr << vi->chunkbits;
 163         map->m_plen = min_t(erofs_off_t, 1UL << vi->chunkbits,
 164                         round_up(inode->i_size - map->m_la, sb->s_blocksize));
 165
 166         /* handle block map */
 167         if (!(vi->chunkformat & EROFS_CHUNK_FORMAT_INDEXES)) {
 168                 __le32 *blkaddr = kaddr + erofs_blkoff(sb, pos);
 169
 170                 if (le32_to_cpu(*blkaddr) == EROFS_NULL_ADDR) {
 171                         map->m_flags = 0;
 172                 } else {
 173                         map->m_pa = erofs_pos(sb, le32_to_cpu(*blkaddr));
 174                         map->m_flags = EROFS_MAP_MAPPED;
 175                 }
 176                 goto out_unlock;
 177         }
 178         /* parse chunk indexes */
 179         idx = kaddr + erofs_blkoff(sb, pos);
 180         switch (le32_to_cpu(idx->blkaddr)) {
 181         case EROFS_NULL_ADDR:
 182                 map->m_flags = 0;
 183                 break;
 184         default:
 185                 map->m_deviceid = le16_to_cpu(idx->device_id) &
 186                         EROFS_SB(sb)->device_id_mask;
 187                 map->m_pa = erofs_pos(sb, le32_to_cpu(idx->blkaddr));
 188                 map->m_flags = EROFS_MAP_MAPPED;
 189                 break;
 190         }
 191 out_unlock:
 192         erofs_put_metabuf(&buf);
 193 out:
 194         if (!err)
 195                 map->m_llen = map->m_plen;
 196         trace_erofs_map_blocks_exit(inode, map, 0, err);
 197         return err;
 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         int id;
 205
 206         map->m_bdev = sb->s_bdev;
 207         map->m_daxdev = EROFS_SB(sb)->dax_dev;
 208         map->m_dax_part_off = EROFS_SB(sb)->dax_part_off;
 209         map->m_fscache = EROFS_SB(sb)->s_fscache;
 210
 211         if (map->m_deviceid) {
 212                 down_read(&devs->rwsem);
 213                 dif = idr_find(&devs->tree, map->m_deviceid - 1);
 214                 if (!dif) {
 215                         up_read(&devs->rwsem);
 216                         return -ENODEV;
 217                 }
 218                 if (devs->flatdev) {
 219                         map->m_pa += erofs_pos(sb, dif->mapped_blkaddr);
 220                         up_read(&devs->rwsem);
 221                         return 0;
 222                 }
 223                 map->m_bdev = dif->bdev_handle ? dif->bdev_handle->bdev : NULL;
 224                 map->m_daxdev = dif->dax_dev;
 225                 map->m_dax_part_off = dif->dax_part_off;
 226                 map->m_fscache = dif->fscache;
 227                 up_read(&devs->rwsem);
 228         } else if (devs->extra_devices && !devs->flatdev) {
 229                 down_read(&devs->rwsem);
 230                 idr_for_each_entry(&devs->tree, dif, id) {
 231                         erofs_off_t startoff, length;
 232
 233                         if (!dif->mapped_blkaddr)
 234                                 continue;
 235                         startoff = erofs_pos(sb, dif->mapped_blkaddr);
 236                         length = erofs_pos(sb, dif->blocks);
 237
 238                         if (map->m_pa >= startoff &&
 239                             map->m_pa < startoff + length) {
 240                                 map->m_pa -= startoff;
 241                                 map->m_bdev = dif->bdev_handle ?
 242                                               dif->bdev_handle->bdev : NULL;
 243                                 map->m_daxdev = dif->dax_dev;
 244                                 map->m_dax_part_off = dif->dax_part_off;
 245                                 map->m_fscache = dif->fscache;
 246                                 break;
 247                         }
 248                 }
 249                 up_read(&devs->rwsem);
 250         }
 251         return 0;
 252 }
 253
 254 static int erofs_iomap_begin(struct inode *inode, loff_t offset, loff_t length,
 255                 unsigned int flags, struct iomap *iomap, struct iomap *srcmap)
 256 {
 257         int ret;
 258         struct super_block *sb = inode->i_sb;
 259         struct erofs_map_blocks map;
 260         struct erofs_map_dev mdev;
 261
 262         map.m_la = offset;
 263         map.m_llen = length;
 264
 265         ret = erofs_map_blocks(inode, &map);
 266         if (ret < 0)
 267                 return ret;
 268
 269         mdev = (struct erofs_map_dev) {
 270                 .m_deviceid = map.m_deviceid,
 271                 .m_pa = map.m_pa,
 272         };
 273         ret = erofs_map_dev(sb, &mdev);
 274         if (ret)
 275                 return ret;
 276
 277         iomap->offset = map.m_la;
 278         if (flags & IOMAP_DAX)
 279                 iomap->dax_dev = mdev.m_daxdev;
 280         else
 281                 iomap->bdev = mdev.m_bdev;
 282         iomap->length = map.m_llen;
 283         iomap->flags = 0;
 284         iomap->private = NULL;
 285
 286         if (!(map.m_flags & EROFS_MAP_MAPPED)) {
 287                 iomap->type = IOMAP_HOLE;
 288                 iomap->addr = IOMAP_NULL_ADDR;
 289                 if (!iomap->length)
 290                         iomap->length = length;
 291                 return 0;
 292         }
 293
 294         if (map.m_flags & EROFS_MAP_META) {
 295                 void *ptr;
 296                 struct erofs_buf buf = __EROFS_BUF_INITIALIZER;
 297
 298                 iomap->type = IOMAP_INLINE;
 299                 ptr = erofs_read_metabuf(&buf, sb,
 300                                 erofs_blknr(sb, mdev.m_pa), EROFS_KMAP);
 301                 if (IS_ERR(ptr))
 302                         return PTR_ERR(ptr);
 303                 iomap->inline_data = ptr + erofs_blkoff(sb, mdev.m_pa);
 304                 iomap->private = buf.base;
 305         } else {
 306                 iomap->type = IOMAP_MAPPED;
 307                 iomap->addr = mdev.m_pa;
 308                 if (flags & IOMAP_DAX)
 309                         iomap->addr += mdev.m_dax_part_off;
 310         }
 311         return 0;
 312 }
 313
 314 static int erofs_iomap_end(struct inode *inode, loff_t pos, loff_t length,
 315                 ssize_t written, unsigned int flags, struct iomap *iomap)
 316 {
 317         void *ptr = iomap->private;
 318
 319         if (ptr) {
 320                 struct erofs_buf buf = {
 321                         .page = kmap_to_page(ptr),
 322                         .base = ptr,
 323                         .kmap_type = EROFS_KMAP,
 324                 };
 325
 326                 DBG_BUGON(iomap->type != IOMAP_INLINE);
 327                 erofs_put_metabuf(&buf);
 328         } else {
 329                 DBG_BUGON(iomap->type == IOMAP_INLINE);
 330         }
 331         return written;
 332 }
 333
 334 static const struct iomap_ops erofs_iomap_ops = {
 335         .iomap_begin = erofs_iomap_begin,
 336         .iomap_end = erofs_iomap_end,
 337 };
 338
 339 int erofs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
 340                  u64 start, u64 len)
 341 {
 342         if (erofs_inode_is_data_compressed(EROFS_I(inode)->datalayout)) {
 343 #ifdef CONFIG_EROFS_FS_ZIP
 344                 return iomap_fiemap(inode, fieinfo, start, len,
 345                                     &z_erofs_iomap_report_ops);
 346 #else
 347                 return -EOPNOTSUPP;
 348 #endif
 349         }
 350         return iomap_fiemap(inode, fieinfo, start, len, &erofs_iomap_ops);
 351 }
 352
 353 /*
 354  * since we dont have write or truncate flows, so no inode
 355  * locking needs to be held at the moment.
 356  */
 357 static int erofs_read_folio(struct file *file, struct folio *folio)
 358 {
 359         return iomap_read_folio(folio, &erofs_iomap_ops);
 360 }
 361
 362 static void erofs_readahead(struct readahead_control *rac)
 363 {
 364         return iomap_readahead(rac, &erofs_iomap_ops);
 365 }
 366
 367 static sector_t erofs_bmap(struct address_space *mapping, sector_t block)
 368 {
 369         return iomap_bmap(mapping, block, &erofs_iomap_ops);
 370 }
 371
 372 static ssize_t erofs_file_read_iter(struct kiocb *iocb, struct iov_iter *to)
 373 {
 374         struct inode *inode = file_inode(iocb->ki_filp);
 375
 376         /* no need taking (shared) inode lock since it's a ro filesystem */
 377         if (!iov_iter_count(to))
 378                 return 0;
 379
 380 #ifdef CONFIG_FS_DAX
 381         if (IS_DAX(inode))
 382                 return dax_iomap_rw(iocb, to, &erofs_iomap_ops);
 383 #endif
 384         if (iocb->ki_flags & IOCB_DIRECT) {
 385                 struct block_device *bdev = inode->i_sb->s_bdev;
 386                 unsigned int blksize_mask;
 387
 388                 if (bdev)
 389                         blksize_mask = bdev_logical_block_size(bdev) - 1;
 390                 else
 391                         blksize_mask = i_blocksize(inode) - 1;
 392
 393                 if ((iocb->ki_pos | iov_iter_count(to) |
 394                      iov_iter_alignment(to)) & blksize_mask)
 395                         return -EINVAL;
 396
 397                 return iomap_dio_rw(iocb, to, &erofs_iomap_ops,
 398                                     NULL, 0, NULL, 0);
 399         }
 400         return filemap_read(iocb, to, 0);
 401 }
 402
 403 /* for uncompressed (aligned) files and raw access for other files */
 404 const struct address_space_operations erofs_raw_access_aops = {
 405         .read_folio = erofs_read_folio,
 406         .readahead = erofs_readahead,
 407         .bmap = erofs_bmap,
 408         .direct_IO = noop_direct_IO,
 409         .release_folio = iomap_release_folio,
 410         .invalidate_folio = iomap_invalidate_folio,
 411 };
 412
 413 #ifdef CONFIG_FS_DAX
 414 static vm_fault_t erofs_dax_huge_fault(struct vm_fault *vmf,
 415                 unsigned int order)
 416 {
 417         return dax_iomap_fault(vmf, order, NULL, NULL, &erofs_iomap_ops);
 418 }
 419
 420 static vm_fault_t erofs_dax_fault(struct vm_fault *vmf)
 421 {
 422         return erofs_dax_huge_fault(vmf, 0);
 423 }
 424
 425 static const struct vm_operations_struct erofs_dax_vm_ops = {
 426         .fault          = erofs_dax_fault,
 427         .huge_fault     = erofs_dax_huge_fault,
 428 };
 429
 430 static int erofs_file_mmap(struct file *file, struct vm_area_struct *vma)
 431 {
 432         if (!IS_DAX(file_inode(file)))
 433                 return generic_file_readonly_mmap(file, vma);
 434
 435         if ((vma->vm_flags & VM_SHARED) && (vma->vm_flags & VM_MAYWRITE))
 436                 return -EINVAL;
 437
 438         vma->vm_ops = &erofs_dax_vm_ops;
 439         vm_flags_set(vma, VM_HUGEPAGE);
 440         return 0;
 441 }
 442 #else
 443 #define erofs_file_mmap generic_file_readonly_mmap
 444 #endif
 445
 446 const struct file_operations erofs_file_fops = {
 447         .llseek         = generic_file_llseek,
 448         .read_iter      = erofs_file_read_iter,
 449         .mmap           = erofs_file_mmap,
 450         .splice_read    = filemap_splice_read,
 451 };