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/prefetch.h>
   9 #include <linux/sched/mm.h>
  10 #include <linux/dax.h>
  11 #include <trace/events/erofs.h>
  12
  13 void erofs_unmap_metabuf(struct erofs_buf *buf)
  14 {
  15         if (buf->kmap_type == EROFS_KMAP)
  16                 kunmap(buf->page);
  17         else if (buf->kmap_type == EROFS_KMAP_ATOMIC)
  18                 kunmap_atomic(buf->base);
  19         buf->base = NULL;
  20         buf->kmap_type = EROFS_NO_KMAP;
  21 }
  22
  23 void erofs_put_metabuf(struct erofs_buf *buf)
  24 {
  25         if (!buf->page)
  26                 return;
  27         erofs_unmap_metabuf(buf);
  28         put_page(buf->page);
  29         buf->page = NULL;
  30 }
  31
  32 void *erofs_bread(struct erofs_buf *buf, struct inode *inode,
  33                   erofs_blk_t blkaddr, enum erofs_kmap_type type)
  34 {
  35         struct address_space *const mapping = inode->i_mapping;
  36         erofs_off_t offset = blknr_to_addr(blkaddr);
  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(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(page);
  58                 else if (type == EROFS_KMAP_ATOMIC)
  59                         buf->base = kmap_atomic(page);
  60                 buf->kmap_type = type;
  61         } else if (buf->kmap_type != type) {
  62                 DBG_BUGON(1);
  63                 return ERR_PTR(-EFAULT);
  64         }
  65         if (type == EROFS_NO_KMAP)
  66                 return NULL;
  67         return buf->base + (offset & ~PAGE_MASK);
  68 }
  69
  70 void *erofs_read_metabuf(struct erofs_buf *buf, struct super_block *sb,
  71                          erofs_blk_t blkaddr, enum erofs_kmap_type type)
  72 {
  73         if (erofs_is_fscache_mode(sb))
  74                 return erofs_bread(buf, EROFS_SB(sb)->s_fscache->inode,
  75                                    blkaddr, type);
  76
  77         return erofs_bread(buf, sb->s_bdev->bd_inode, blkaddr, type);
  78 }
  79
  80 static int erofs_map_blocks_flatmode(struct inode *inode,
  81                                      struct erofs_map_blocks *map,
  82                                      int flags)
  83 {
  84         erofs_blk_t nblocks, lastblk;
  85         u64 offset = map->m_la;
  86         struct erofs_inode *vi = EROFS_I(inode);
  87         bool tailendpacking = (vi->datalayout == EROFS_INODE_FLAT_INLINE);
  88
  89         nblocks = DIV_ROUND_UP(inode->i_size, EROFS_BLKSIZ);
  90         lastblk = nblocks - tailendpacking;
  91
  92         /* there is no hole in flatmode */
  93         map->m_flags = EROFS_MAP_MAPPED;
  94         if (offset < blknr_to_addr(lastblk)) {
  95                 map->m_pa = blknr_to_addr(vi->raw_blkaddr) + map->m_la;
  96                 map->m_plen = blknr_to_addr(lastblk) - offset;
  97         } else if (tailendpacking) {
  98                 /* 2 - inode inline B: inode, [xattrs], inline last blk... */
  99                 struct erofs_sb_info *sbi = EROFS_SB(inode->i_sb);
 100
 101                 map->m_pa = iloc(sbi, vi->nid) + vi->inode_isize +
 102                         vi->xattr_isize + erofs_blkoff(map->m_la);
 103                 map->m_plen = inode->i_size - offset;
 104
 105                 /* inline data should be located in the same meta block */
 106                 if (erofs_blkoff(map->m_pa) + map->m_plen > EROFS_BLKSIZ) {
 107                         erofs_err(inode->i_sb,
 108                                   "inline data cross block boundary @ nid %llu",
 109                                   vi->nid);
 110                         DBG_BUGON(1);
 111                         return -EFSCORRUPTED;
 112                 }
 113                 map->m_flags |= EROFS_MAP_META;
 114         } else {
 115                 erofs_err(inode->i_sb,
 116                           "internal error @ nid: %llu (size %llu), m_la 0x%llx",
 117                           vi->nid, inode->i_size, map->m_la);
 118                 DBG_BUGON(1);
 119                 return -EIO;
 120         }
 121         return 0;
 122 }
 123
 124 int erofs_map_blocks(struct inode *inode,
 125                      struct erofs_map_blocks *map, int flags)
 126 {
 127         struct super_block *sb = inode->i_sb;
 128         struct erofs_inode *vi = EROFS_I(inode);
 129         struct erofs_inode_chunk_index *idx;
 130         struct erofs_buf buf = __EROFS_BUF_INITIALIZER;
 131         u64 chunknr;
 132         unsigned int unit;
 133         erofs_off_t pos;
 134         void *kaddr;
 135         int err = 0;
 136
 137         trace_erofs_map_blocks_enter(inode, map, flags);
 138         map->m_deviceid = 0;
 139         if (map->m_la >= inode->i_size) {
 140                 /* leave out-of-bound access unmapped */
 141                 map->m_flags = 0;
 142                 map->m_plen = 0;
 143                 goto out;
 144         }
 145
 146         if (vi->datalayout != EROFS_INODE_CHUNK_BASED) {
 147                 err = erofs_map_blocks_flatmode(inode, map, flags);
 148                 goto out;
 149         }
 150
 151         if (vi->chunkformat & EROFS_CHUNK_FORMAT_INDEXES)
 152                 unit = sizeof(*idx);                    /* chunk index */
 153         else
 154                 unit = EROFS_BLOCK_MAP_ENTRY_SIZE;      /* block map */
 155
 156         chunknr = map->m_la >> vi->chunkbits;
 157         pos = ALIGN(iloc(EROFS_SB(sb), vi->nid) + vi->inode_isize +
 158                     vi->xattr_isize, unit) + unit * chunknr;
 159
 160         kaddr = erofs_read_metabuf(&buf, sb, erofs_blknr(pos), EROFS_KMAP);
 161         if (IS_ERR(kaddr)) {
 162                 err = PTR_ERR(kaddr);
 163                 goto out;
 164         }
 165         map->m_la = chunknr << vi->chunkbits;
 166         map->m_plen = min_t(erofs_off_t, 1UL << vi->chunkbits,
 167                             roundup(inode->i_size - map->m_la, EROFS_BLKSIZ));
 168
 169         /* handle block map */
 170         if (!(vi->chunkformat & EROFS_CHUNK_FORMAT_INDEXES)) {
 171                 __le32 *blkaddr = kaddr + erofs_blkoff(pos);
 172
 173                 if (le32_to_cpu(*blkaddr) == EROFS_NULL_ADDR) {
 174                         map->m_flags = 0;
 175                 } else {
 176                         map->m_pa = blknr_to_addr(le32_to_cpu(*blkaddr));
 177                         map->m_flags = EROFS_MAP_MAPPED;
 178                 }
 179                 goto out_unlock;
 180         }
 181         /* parse chunk indexes */
 182         idx = kaddr + erofs_blkoff(pos);
 183         switch (le32_to_cpu(idx->blkaddr)) {
 184         case EROFS_NULL_ADDR:
 185                 map->m_flags = 0;
 186                 break;
 187         default:
 188                 map->m_deviceid = le16_to_cpu(idx->device_id) &
 189                         EROFS_SB(sb)->device_id_mask;
 190                 map->m_pa = blknr_to_addr(le32_to_cpu(idx->blkaddr));
 191                 map->m_flags = EROFS_MAP_MAPPED;
 192                 break;
 193         }
 194 out_unlock:
 195         erofs_put_metabuf(&buf);
 196 out:
 197         if (!err)
 198                 map->m_llen = map->m_plen;
 199         trace_erofs_map_blocks_exit(inode, map, flags, 0);
 200         return err;
 201 }
 202
 203 int erofs_map_dev(struct super_block *sb, struct erofs_map_dev *map)
 204 {
 205         struct erofs_dev_context *devs = EROFS_SB(sb)->devs;
 206         struct erofs_device_info *dif;
 207         int id;
 208
 209         /* primary device by default */
 210         map->m_bdev = sb->s_bdev;
 211         map->m_daxdev = EROFS_SB(sb)->dax_dev;
 212         map->m_dax_part_off = EROFS_SB(sb)->dax_part_off;
 213         map->m_fscache = EROFS_SB(sb)->s_fscache;
 214
 215         if (map->m_deviceid) {
 216                 down_read(&devs->rwsem);
 217                 dif = idr_find(&devs->tree, map->m_deviceid - 1);
 218                 if (!dif) {
 219                         up_read(&devs->rwsem);
 220                         return -ENODEV;
 221                 }
 222                 map->m_bdev = dif->bdev;
 223                 map->m_daxdev = dif->dax_dev;
 224                 map->m_dax_part_off = dif->dax_part_off;
 225                 map->m_fscache = dif->fscache;
 226                 up_read(&devs->rwsem);
 227         } else if (devs->extra_devices) {
 228                 down_read(&devs->rwsem);
 229                 idr_for_each_entry(&devs->tree, dif, id) {
 230                         erofs_off_t startoff, length;
 231
 232                         if (!dif->mapped_blkaddr)
 233                                 continue;
 234                         startoff = blknr_to_addr(dif->mapped_blkaddr);
 235                         length = blknr_to_addr(dif->blocks);
 236
 237                         if (map->m_pa >= startoff &&
 238                             map->m_pa < startoff + length) {
 239                                 map->m_pa -= startoff;
 240                                 map->m_bdev = dif->bdev;
 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 erofs_map_blocks map;
 257         struct erofs_map_dev mdev;
 258
 259         map.m_la = offset;
 260         map.m_llen = length;
 261
 262         ret = erofs_map_blocks(inode, &map, EROFS_GET_BLOCKS_RAW);
 263         if (ret < 0)
 264                 return ret;
 265
 266         mdev = (struct erofs_map_dev) {
 267                 .m_deviceid = map.m_deviceid,
 268                 .m_pa = map.m_pa,
 269         };
 270         ret = erofs_map_dev(inode->i_sb, &mdev);
 271         if (ret)
 272                 return ret;
 273
 274         iomap->offset = map.m_la;
 275         if (flags & IOMAP_DAX)
 276                 iomap->dax_dev = mdev.m_daxdev;
 277         else
 278                 iomap->bdev = mdev.m_bdev;
 279         iomap->length = map.m_llen;
 280         iomap->flags = 0;
 281         iomap->private = NULL;
 282
 283         if (!(map.m_flags & EROFS_MAP_MAPPED)) {
 284                 iomap->type = IOMAP_HOLE;
 285                 iomap->addr = IOMAP_NULL_ADDR;
 286                 if (!iomap->length)
 287                         iomap->length = length;
 288                 return 0;
 289         }
 290
 291         if (map.m_flags & EROFS_MAP_META) {
 292                 void *ptr;
 293                 struct erofs_buf buf = __EROFS_BUF_INITIALIZER;
 294
 295                 iomap->type = IOMAP_INLINE;
 296                 ptr = erofs_read_metabuf(&buf, inode->i_sb,
 297                                          erofs_blknr(mdev.m_pa), EROFS_KMAP);
 298                 if (IS_ERR(ptr))
 299                         return PTR_ERR(ptr);
 300                 iomap->inline_data = ptr + erofs_blkoff(mdev.m_pa);
 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 = (1 << inode->i_blkbits) - 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 };
 407
 408 #ifdef CONFIG_FS_DAX
 409 static vm_fault_t erofs_dax_huge_fault(struct vm_fault *vmf,
 410                 enum page_entry_size pe_size)
 411 {
 412         return dax_iomap_fault(vmf, pe_size, NULL, NULL, &erofs_iomap_ops);
 413 }
 414
 415 static vm_fault_t erofs_dax_fault(struct vm_fault *vmf)
 416 {
 417         return erofs_dax_huge_fault(vmf, PE_SIZE_PTE);
 418 }
 419
 420 static const struct vm_operations_struct erofs_dax_vm_ops = {
 421         .fault          = erofs_dax_fault,
 422         .huge_fault     = erofs_dax_huge_fault,
 423 };
 424
 425 static int erofs_file_mmap(struct file *file, struct vm_area_struct *vma)
 426 {
 427         if (!IS_DAX(file_inode(file)))
 428                 return generic_file_readonly_mmap(file, vma);
 429
 430         if ((vma->vm_flags & VM_SHARED) && (vma->vm_flags & VM_MAYWRITE))
 431                 return -EINVAL;
 432
 433         vma->vm_ops = &erofs_dax_vm_ops;
 434         vma->vm_flags |= VM_HUGEPAGE;
 435         return 0;
 436 }
 437 #else
 438 #define erofs_file_mmap generic_file_readonly_mmap
 439 #endif
 440
 441 const struct file_operations erofs_file_fops = {
 442         .llseek         = generic_file_llseek,
 443         .read_iter      = erofs_file_read_iter,
 444         .mmap           = erofs_file_mmap,
 445         .splice_read    = generic_file_splice_read,
 446 };