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_local(buf->base);
  17         buf->base = NULL;
  18         buf->kmap_type = EROFS_NO_KMAP;
  19 }
  20
  21 void erofs_put_metabuf(struct erofs_buf *buf)
  22 {
  23         if (!buf->page)
  24                 return;
  25         erofs_unmap_metabuf(buf);
  26         put_page(buf->page);
  27         buf->page = NULL;
  28 }
  29
  30 void *erofs_bread(struct erofs_buf *buf, struct inode *inode,
  31                   erofs_blk_t blkaddr, enum erofs_kmap_type type)
  32 {
  33         struct address_space *const mapping = inode->i_mapping;
  34         erofs_off_t offset = blknr_to_addr(blkaddr);
  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(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_read_metabuf(struct erofs_buf *buf, struct super_block *sb,
  67                          erofs_blk_t blkaddr, enum erofs_kmap_type type)
  68 {
  69         if (erofs_is_fscache_mode(sb))
  70                 return erofs_bread(buf, EROFS_SB(sb)->s_fscache->inode,
  71                                    blkaddr, type);
  72
  73         return erofs_bread(buf, sb->s_bdev->bd_inode, blkaddr, type);
  74 }
  75
  76 static int erofs_map_blocks_flatmode(struct inode *inode,
  77                                      struct erofs_map_blocks *map,
  78                                      int flags)
  79 {
  80         erofs_blk_t nblocks, lastblk;
  81         u64 offset = map->m_la;
  82         struct erofs_inode *vi = EROFS_I(inode);
  83         bool tailendpacking = (vi->datalayout == EROFS_INODE_FLAT_INLINE);
  84
  85         nblocks = DIV_ROUND_UP(inode->i_size, EROFS_BLKSIZ);
  86         lastblk = nblocks - tailendpacking;
  87
  88         /* there is no hole in flatmode */
  89         map->m_flags = EROFS_MAP_MAPPED;
  90         if (offset < blknr_to_addr(lastblk)) {
  91                 map->m_pa = blknr_to_addr(vi->raw_blkaddr) + map->m_la;
  92                 map->m_plen = blknr_to_addr(lastblk) - offset;
  93         } else if (tailendpacking) {
  94                 /* 2 - inode inline B: inode, [xattrs], inline last blk... */
  95                 struct erofs_sb_info *sbi = EROFS_SB(inode->i_sb);
  96
  97                 map->m_pa = iloc(sbi, vi->nid) + vi->inode_isize +
  98                         vi->xattr_isize + erofs_blkoff(map->m_la);
  99                 map->m_plen = inode->i_size - offset;
 100
 101                 /* inline data should be located in the same meta block */
 102                 if (erofs_blkoff(map->m_pa) + map->m_plen > EROFS_BLKSIZ) {
 103                         erofs_err(inode->i_sb,
 104                                   "inline data cross block boundary @ nid %llu",
 105                                   vi->nid);
 106                         DBG_BUGON(1);
 107                         return -EFSCORRUPTED;
 108                 }
 109                 map->m_flags |= EROFS_MAP_META;
 110         } else {
 111                 erofs_err(inode->i_sb,
 112                           "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,
 121                      struct erofs_map_blocks *map, int flags)
 122 {
 123         struct super_block *sb = inode->i_sb;
 124         struct erofs_inode *vi = EROFS_I(inode);
 125         struct erofs_inode_chunk_index *idx;
 126         struct erofs_buf buf = __EROFS_BUF_INITIALIZER;
 127         u64 chunknr;
 128         unsigned int unit;
 129         erofs_off_t pos;
 130         void *kaddr;
 131         int err = 0;
 132
 133         trace_erofs_map_blocks_enter(inode, map, flags);
 134         map->m_deviceid = 0;
 135         if (map->m_la >= inode->i_size) {
 136                 /* leave out-of-bound access unmapped */
 137                 map->m_flags = 0;
 138                 map->m_plen = 0;
 139                 goto out;
 140         }
 141
 142         if (vi->datalayout != EROFS_INODE_CHUNK_BASED) {
 143                 err = erofs_map_blocks_flatmode(inode, map, flags);
 144                 goto out;
 145         }
 146
 147         if (vi->chunkformat & EROFS_CHUNK_FORMAT_INDEXES)
 148                 unit = sizeof(*idx);                    /* chunk index */
 149         else
 150                 unit = EROFS_BLOCK_MAP_ENTRY_SIZE;      /* block map */
 151
 152         chunknr = map->m_la >> vi->chunkbits;
 153         pos = ALIGN(iloc(EROFS_SB(sb), vi->nid) + vi->inode_isize +
 154                     vi->xattr_isize, unit) + unit * chunknr;
 155
 156         kaddr = erofs_read_metabuf(&buf, sb, erofs_blknr(pos), EROFS_KMAP);
 157         if (IS_ERR(kaddr)) {
 158                 err = PTR_ERR(kaddr);
 159                 goto out;
 160         }
 161         map->m_la = chunknr << vi->chunkbits;
 162         map->m_plen = min_t(erofs_off_t, 1UL << vi->chunkbits,
 163                             roundup(inode->i_size - map->m_la, EROFS_BLKSIZ));
 164
 165         /* handle block map */
 166         if (!(vi->chunkformat & EROFS_CHUNK_FORMAT_INDEXES)) {
 167                 __le32 *blkaddr = kaddr + erofs_blkoff(pos);
 168
 169                 if (le32_to_cpu(*blkaddr) == EROFS_NULL_ADDR) {
 170                         map->m_flags = 0;
 171                 } else {
 172                         map->m_pa = blknr_to_addr(le32_to_cpu(*blkaddr));
 173                         map->m_flags = EROFS_MAP_MAPPED;
 174                 }
 175                 goto out_unlock;
 176         }
 177         /* parse chunk indexes */
 178         idx = kaddr + erofs_blkoff(pos);
 179         switch (le32_to_cpu(idx->blkaddr)) {
 180         case EROFS_NULL_ADDR:
 181                 map->m_flags = 0;
 182                 break;
 183         default:
 184                 map->m_deviceid = le16_to_cpu(idx->device_id) &
 185                         EROFS_SB(sb)->device_id_mask;
 186                 map->m_pa = blknr_to_addr(le32_to_cpu(idx->blkaddr));
 187                 map->m_flags = EROFS_MAP_MAPPED;
 188                 break;
 189         }
 190 out_unlock:
 191         erofs_put_metabuf(&buf);
 192 out:
 193         if (!err)
 194                 map->m_llen = map->m_plen;
 195         trace_erofs_map_blocks_exit(inode, map, flags, 0);
 196         return err;
 197 }
 198
 199 int erofs_map_dev(struct super_block *sb, struct erofs_map_dev *map)
 200 {
 201         struct erofs_dev_context *devs = EROFS_SB(sb)->devs;
 202         struct erofs_device_info *dif;
 203         int id;
 204
 205         /* primary device by default */
 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                 map->m_bdev = dif->bdev;
 219                 map->m_daxdev = dif->dax_dev;
 220                 map->m_dax_part_off = dif->dax_part_off;
 221                 map->m_fscache = dif->fscache;
 222                 up_read(&devs->rwsem);
 223         } else if (devs->extra_devices) {
 224                 down_read(&devs->rwsem);
 225                 idr_for_each_entry(&devs->tree, dif, id) {
 226                         erofs_off_t startoff, length;
 227
 228                         if (!dif->mapped_blkaddr)
 229                                 continue;
 230                         startoff = blknr_to_addr(dif->mapped_blkaddr);
 231                         length = blknr_to_addr(dif->blocks);
 232
 233                         if (map->m_pa >= startoff &&
 234                             map->m_pa < startoff + length) {
 235                                 map->m_pa -= startoff;
 236                                 map->m_bdev = dif->bdev;
 237                                 map->m_daxdev = dif->dax_dev;
 238                                 map->m_dax_part_off = dif->dax_part_off;
 239                                 map->m_fscache = dif->fscache;
 240                                 break;
 241                         }
 242                 }
 243                 up_read(&devs->rwsem);
 244         }
 245         return 0;
 246 }
 247
 248 static int erofs_iomap_begin(struct inode *inode, loff_t offset, loff_t length,
 249                 unsigned int flags, struct iomap *iomap, struct iomap *srcmap)
 250 {
 251         int ret;
 252         struct erofs_map_blocks map;
 253         struct erofs_map_dev mdev;
 254
 255         map.m_la = offset;
 256         map.m_llen = length;
 257
 258         ret = erofs_map_blocks(inode, &map, EROFS_GET_BLOCKS_RAW);
 259         if (ret < 0)
 260                 return ret;
 261
 262         mdev = (struct erofs_map_dev) {
 263                 .m_deviceid = map.m_deviceid,
 264                 .m_pa = map.m_pa,
 265         };
 266         ret = erofs_map_dev(inode->i_sb, &mdev);
 267         if (ret)
 268                 return ret;
 269
 270         iomap->offset = map.m_la;
 271         if (flags & IOMAP_DAX)
 272                 iomap->dax_dev = mdev.m_daxdev;
 273         else
 274                 iomap->bdev = mdev.m_bdev;
 275         iomap->length = map.m_llen;
 276         iomap->flags = 0;
 277         iomap->private = NULL;
 278
 279         if (!(map.m_flags & EROFS_MAP_MAPPED)) {
 280                 iomap->type = IOMAP_HOLE;
 281                 iomap->addr = IOMAP_NULL_ADDR;
 282                 if (!iomap->length)
 283                         iomap->length = length;
 284                 return 0;
 285         }
 286
 287         if (map.m_flags & EROFS_MAP_META) {
 288                 void *ptr;
 289                 struct erofs_buf buf = __EROFS_BUF_INITIALIZER;
 290
 291                 iomap->type = IOMAP_INLINE;
 292                 ptr = erofs_read_metabuf(&buf, inode->i_sb,
 293                                          erofs_blknr(mdev.m_pa), EROFS_KMAP);
 294                 if (IS_ERR(ptr))
 295                         return PTR_ERR(ptr);
 296                 iomap->inline_data = ptr + erofs_blkoff(mdev.m_pa);
 297                 iomap->private = buf.base;
 298         } else {
 299                 iomap->type = IOMAP_MAPPED;
 300                 iomap->addr = mdev.m_pa;
 301                 if (flags & IOMAP_DAX)
 302                         iomap->addr += mdev.m_dax_part_off;
 303         }
 304         return 0;
 305 }
 306
 307 static int erofs_iomap_end(struct inode *inode, loff_t pos, loff_t length,
 308                 ssize_t written, unsigned int flags, struct iomap *iomap)
 309 {
 310         void *ptr = iomap->private;
 311
 312         if (ptr) {
 313                 struct erofs_buf buf = {
 314                         .page = kmap_to_page(ptr),
 315                         .base = ptr,
 316                         .kmap_type = EROFS_KMAP,
 317                 };
 318
 319                 DBG_BUGON(iomap->type != IOMAP_INLINE);
 320                 erofs_put_metabuf(&buf);
 321         } else {
 322                 DBG_BUGON(iomap->type == IOMAP_INLINE);
 323         }
 324         return written;
 325 }
 326
 327 static const struct iomap_ops erofs_iomap_ops = {
 328         .iomap_begin = erofs_iomap_begin,
 329         .iomap_end = erofs_iomap_end,
 330 };
 331
 332 int erofs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
 333                  u64 start, u64 len)
 334 {
 335         if (erofs_inode_is_data_compressed(EROFS_I(inode)->datalayout)) {
 336 #ifdef CONFIG_EROFS_FS_ZIP
 337                 return iomap_fiemap(inode, fieinfo, start, len,
 338                                     &z_erofs_iomap_report_ops);
 339 #else
 340                 return -EOPNOTSUPP;
 341 #endif
 342         }
 343         return iomap_fiemap(inode, fieinfo, start, len, &erofs_iomap_ops);
 344 }
 345
 346 /*
 347  * since we dont have write or truncate flows, so no inode
 348  * locking needs to be held at the moment.
 349  */
 350 static int erofs_read_folio(struct file *file, struct folio *folio)
 351 {
 352         return iomap_read_folio(folio, &erofs_iomap_ops);
 353 }
 354
 355 static void erofs_readahead(struct readahead_control *rac)
 356 {
 357         return iomap_readahead(rac, &erofs_iomap_ops);
 358 }
 359
 360 static sector_t erofs_bmap(struct address_space *mapping, sector_t block)
 361 {
 362         return iomap_bmap(mapping, block, &erofs_iomap_ops);
 363 }
 364
 365 static ssize_t erofs_file_read_iter(struct kiocb *iocb, struct iov_iter *to)
 366 {
 367         struct inode *inode = file_inode(iocb->ki_filp);
 368
 369         /* no need taking (shared) inode lock since it's a ro filesystem */
 370         if (!iov_iter_count(to))
 371                 return 0;
 372
 373 #ifdef CONFIG_FS_DAX
 374         if (IS_DAX(inode))
 375                 return dax_iomap_rw(iocb, to, &erofs_iomap_ops);
 376 #endif
 377         if (iocb->ki_flags & IOCB_DIRECT) {
 378                 struct block_device *bdev = inode->i_sb->s_bdev;
 379                 unsigned int blksize_mask;
 380
 381                 if (bdev)
 382                         blksize_mask = bdev_logical_block_size(bdev) - 1;
 383                 else
 384                         blksize_mask = (1 << inode->i_blkbits) - 1;
 385
 386                 if ((iocb->ki_pos | iov_iter_count(to) |
 387                      iov_iter_alignment(to)) & blksize_mask)
 388                         return -EINVAL;
 389
 390                 return iomap_dio_rw(iocb, to, &erofs_iomap_ops,
 391                                     NULL, 0, NULL, 0);
 392         }
 393         return filemap_read(iocb, to, 0);
 394 }
 395
 396 /* for uncompressed (aligned) files and raw access for other files */
 397 const struct address_space_operations erofs_raw_access_aops = {
 398         .read_folio = erofs_read_folio,
 399         .readahead = erofs_readahead,
 400         .bmap = erofs_bmap,
 401         .direct_IO = noop_direct_IO,
 402         .release_folio = iomap_release_folio,
 403         .invalidate_folio = iomap_invalidate_folio,
 404 };
 405
 406 #ifdef CONFIG_FS_DAX
 407 static vm_fault_t erofs_dax_huge_fault(struct vm_fault *vmf,
 408                 enum page_entry_size pe_size)
 409 {
 410         return dax_iomap_fault(vmf, pe_size, NULL, NULL, &erofs_iomap_ops);
 411 }
 412
 413 static vm_fault_t erofs_dax_fault(struct vm_fault *vmf)
 414 {
 415         return erofs_dax_huge_fault(vmf, PE_SIZE_PTE);
 416 }
 417
 418 static const struct vm_operations_struct erofs_dax_vm_ops = {
 419         .fault          = erofs_dax_fault,
 420         .huge_fault     = erofs_dax_huge_fault,
 421 };
 422
 423 static int erofs_file_mmap(struct file *file, struct vm_area_struct *vma)
 424 {
 425         if (!IS_DAX(file_inode(file)))
 426                 return generic_file_readonly_mmap(file, vma);
 427
 428         if ((vma->vm_flags & VM_SHARED) && (vma->vm_flags & VM_MAYWRITE))
 429                 return -EINVAL;
 430
 431         vma->vm_ops = &erofs_dax_vm_ops;
 432         vma->vm_flags |= VM_HUGEPAGE;
 433         return 0;
 434 }
 435 #else
 436 #define erofs_file_mmap generic_file_readonly_mmap
 437 #endif
 438
 439 const struct file_operations erofs_file_fops = {
 440         .llseek         = generic_file_llseek,
 441         .read_iter      = erofs_file_read_iter,
 442         .mmap           = erofs_file_mmap,
 443         .splice_read    = generic_file_splice_read,
 444 };