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/dax.h>
  10 #include <trace/events/erofs.h>
  11
  12 struct page *erofs_get_meta_page(struct super_block *sb, erofs_blk_t blkaddr)
  13 {
  14         struct address_space *const mapping = sb->s_bdev->bd_inode->i_mapping;
  15         struct page *page;
  16
  17         page = read_cache_page_gfp(mapping, blkaddr,
  18                                    mapping_gfp_constraint(mapping, ~__GFP_FS));
  19         /* should already be PageUptodate */
  20         if (!IS_ERR(page))
  21                 lock_page(page);
  22         return page;
  23 }
  24
  25 static int erofs_map_blocks_flatmode(struct inode *inode,
  26                                      struct erofs_map_blocks *map,
  27                                      int flags)
  28 {
  29         int err = 0;
  30         erofs_blk_t nblocks, lastblk;
  31         u64 offset = map->m_la;
  32         struct erofs_inode *vi = EROFS_I(inode);
  33         bool tailendpacking = (vi->datalayout == EROFS_INODE_FLAT_INLINE);
  34
  35         trace_erofs_map_blocks_flatmode_enter(inode, map, flags);
  36
  37         nblocks = DIV_ROUND_UP(inode->i_size, PAGE_SIZE);
  38         lastblk = nblocks - tailendpacking;
  39
  40         /* there is no hole in flatmode */
  41         map->m_flags = EROFS_MAP_MAPPED;
  42
  43         if (offset < blknr_to_addr(lastblk)) {
  44                 map->m_pa = blknr_to_addr(vi->raw_blkaddr) + map->m_la;
  45                 map->m_plen = blknr_to_addr(lastblk) - offset;
  46         } else if (tailendpacking) {
  47                 /* 2 - inode inline B: inode, [xattrs], inline last blk... */
  48                 struct erofs_sb_info *sbi = EROFS_SB(inode->i_sb);
  49
  50                 map->m_pa = iloc(sbi, vi->nid) + vi->inode_isize +
  51                         vi->xattr_isize + erofs_blkoff(map->m_la);
  52                 map->m_plen = inode->i_size - offset;
  53
  54                 /* inline data should be located in one meta block */
  55                 if (erofs_blkoff(map->m_pa) + map->m_plen > PAGE_SIZE) {
  56                         erofs_err(inode->i_sb,
  57                                   "inline data cross block boundary @ nid %llu",
  58                                   vi->nid);
  59                         DBG_BUGON(1);
  60                         err = -EFSCORRUPTED;
  61                         goto err_out;
  62                 }
  63
  64                 map->m_flags |= EROFS_MAP_META;
  65         } else {
  66                 erofs_err(inode->i_sb,
  67                           "internal error @ nid: %llu (size %llu), m_la 0x%llx",
  68                           vi->nid, inode->i_size, map->m_la);
  69                 DBG_BUGON(1);
  70                 err = -EIO;
  71                 goto err_out;
  72         }
  73
  74         map->m_llen = map->m_plen;
  75 err_out:
  76         trace_erofs_map_blocks_flatmode_exit(inode, map, flags, 0);
  77         return err;
  78 }
  79
  80 static int erofs_map_blocks(struct inode *inode,
  81                             struct erofs_map_blocks *map, int flags)
  82 {
  83         struct super_block *sb = inode->i_sb;
  84         struct erofs_inode *vi = EROFS_I(inode);
  85         struct erofs_inode_chunk_index *idx;
  86         struct page *page;
  87         u64 chunknr;
  88         unsigned int unit;
  89         erofs_off_t pos;
  90         int err = 0;
  91
  92         if (map->m_la >= inode->i_size) {
  93                 /* leave out-of-bound access unmapped */
  94                 map->m_flags = 0;
  95                 map->m_plen = 0;
  96                 goto out;
  97         }
  98
  99         if (vi->datalayout != EROFS_INODE_CHUNK_BASED)
 100                 return erofs_map_blocks_flatmode(inode, map, flags);
 101
 102         if (vi->chunkformat & EROFS_CHUNK_FORMAT_INDEXES)
 103                 unit = sizeof(*idx);                    /* chunk index */
 104         else
 105                 unit = EROFS_BLOCK_MAP_ENTRY_SIZE;      /* block map */
 106
 107         chunknr = map->m_la >> vi->chunkbits;
 108         pos = ALIGN(iloc(EROFS_SB(sb), vi->nid) + vi->inode_isize +
 109                     vi->xattr_isize, unit) + unit * chunknr;
 110
 111         page = erofs_get_meta_page(inode->i_sb, erofs_blknr(pos));
 112         if (IS_ERR(page))
 113                 return PTR_ERR(page);
 114
 115         map->m_la = chunknr << vi->chunkbits;
 116         map->m_plen = min_t(erofs_off_t, 1UL << vi->chunkbits,
 117                             roundup(inode->i_size - map->m_la, EROFS_BLKSIZ));
 118
 119         /* handle block map */
 120         if (!(vi->chunkformat & EROFS_CHUNK_FORMAT_INDEXES)) {
 121                 __le32 *blkaddr = page_address(page) + erofs_blkoff(pos);
 122
 123                 if (le32_to_cpu(*blkaddr) == EROFS_NULL_ADDR) {
 124                         map->m_flags = 0;
 125                 } else {
 126                         map->m_pa = blknr_to_addr(le32_to_cpu(*blkaddr));
 127                         map->m_flags = EROFS_MAP_MAPPED;
 128                 }
 129                 goto out_unlock;
 130         }
 131         /* parse chunk indexes */
 132         idx = page_address(page) + erofs_blkoff(pos);
 133         switch (le32_to_cpu(idx->blkaddr)) {
 134         case EROFS_NULL_ADDR:
 135                 map->m_flags = 0;
 136                 break;
 137         default:
 138                 /* only one device is supported for now */
 139                 if (idx->device_id) {
 140                         erofs_err(sb, "invalid device id %u @ %llu for nid %llu",
 141                                   le16_to_cpu(idx->device_id),
 142                                   chunknr, vi->nid);
 143                         err = -EFSCORRUPTED;
 144                         goto out_unlock;
 145                 }
 146                 map->m_pa = blknr_to_addr(le32_to_cpu(idx->blkaddr));
 147                 map->m_flags = EROFS_MAP_MAPPED;
 148                 break;
 149         }
 150 out_unlock:
 151         unlock_page(page);
 152         put_page(page);
 153 out:
 154         map->m_llen = map->m_plen;
 155         return err;
 156 }
 157
 158 static int erofs_iomap_begin(struct inode *inode, loff_t offset, loff_t length,
 159                 unsigned int flags, struct iomap *iomap, struct iomap *srcmap)
 160 {
 161         int ret;
 162         struct erofs_map_blocks map;
 163
 164         map.m_la = offset;
 165         map.m_llen = length;
 166
 167         ret = erofs_map_blocks(inode, &map, EROFS_GET_BLOCKS_RAW);
 168         if (ret < 0)
 169                 return ret;
 170
 171         iomap->bdev = inode->i_sb->s_bdev;
 172         iomap->dax_dev = EROFS_I_SB(inode)->dax_dev;
 173         iomap->offset = map.m_la;
 174         iomap->length = map.m_llen;
 175         iomap->flags = 0;
 176         iomap->private = NULL;
 177
 178         if (!(map.m_flags & EROFS_MAP_MAPPED)) {
 179                 iomap->type = IOMAP_HOLE;
 180                 iomap->addr = IOMAP_NULL_ADDR;
 181                 if (!iomap->length)
 182                         iomap->length = length;
 183                 return 0;
 184         }
 185
 186         if (map.m_flags & EROFS_MAP_META) {
 187                 struct page *ipage;
 188
 189                 iomap->type = IOMAP_INLINE;
 190                 ipage = erofs_get_meta_page(inode->i_sb,
 191                                             erofs_blknr(map.m_pa));
 192                 if (IS_ERR(ipage))
 193                         return PTR_ERR(ipage);
 194                 iomap->inline_data = page_address(ipage) +
 195                                         erofs_blkoff(map.m_pa);
 196                 iomap->private = ipage;
 197         } else {
 198                 iomap->type = IOMAP_MAPPED;
 199                 iomap->addr = map.m_pa;
 200         }
 201         return 0;
 202 }
 203
 204 static int erofs_iomap_end(struct inode *inode, loff_t pos, loff_t length,
 205                 ssize_t written, unsigned int flags, struct iomap *iomap)
 206 {
 207         struct page *ipage = iomap->private;
 208
 209         if (ipage) {
 210                 DBG_BUGON(iomap->type != IOMAP_INLINE);
 211                 unlock_page(ipage);
 212                 put_page(ipage);
 213         } else {
 214                 DBG_BUGON(iomap->type == IOMAP_INLINE);
 215         }
 216         return written;
 217 }
 218
 219 static const struct iomap_ops erofs_iomap_ops = {
 220         .iomap_begin = erofs_iomap_begin,
 221         .iomap_end = erofs_iomap_end,
 222 };
 223
 224 int erofs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
 225                  u64 start, u64 len)
 226 {
 227         if (erofs_inode_is_data_compressed(EROFS_I(inode)->datalayout)) {
 228 #ifdef CONFIG_EROFS_FS_ZIP
 229                 return iomap_fiemap(inode, fieinfo, start, len,
 230                                     &z_erofs_iomap_report_ops);
 231 #else
 232                 return -EOPNOTSUPP;
 233 #endif
 234         }
 235         return iomap_fiemap(inode, fieinfo, start, len, &erofs_iomap_ops);
 236 }
 237
 238 /*
 239  * since we dont have write or truncate flows, so no inode
 240  * locking needs to be held at the moment.
 241  */
 242 static int erofs_readpage(struct file *file, struct page *page)
 243 {
 244         return iomap_readpage(page, &erofs_iomap_ops);
 245 }
 246
 247 static void erofs_readahead(struct readahead_control *rac)
 248 {
 249         return iomap_readahead(rac, &erofs_iomap_ops);
 250 }
 251
 252 static sector_t erofs_bmap(struct address_space *mapping, sector_t block)
 253 {
 254         return iomap_bmap(mapping, block, &erofs_iomap_ops);
 255 }
 256
 257 static int erofs_prepare_dio(struct kiocb *iocb, struct iov_iter *to)
 258 {
 259         struct inode *inode = file_inode(iocb->ki_filp);
 260         loff_t align = iocb->ki_pos | iov_iter_count(to) |
 261                 iov_iter_alignment(to);
 262         struct block_device *bdev = inode->i_sb->s_bdev;
 263         unsigned int blksize_mask;
 264
 265         if (bdev)
 266                 blksize_mask = (1 << ilog2(bdev_logical_block_size(bdev))) - 1;
 267         else
 268                 blksize_mask = (1 << inode->i_blkbits) - 1;
 269
 270         if (align & blksize_mask)
 271                 return -EINVAL;
 272         return 0;
 273 }
 274
 275 static ssize_t erofs_file_read_iter(struct kiocb *iocb, struct iov_iter *to)
 276 {
 277         /* no need taking (shared) inode lock since it's a ro filesystem */
 278         if (!iov_iter_count(to))
 279                 return 0;
 280
 281 #ifdef CONFIG_FS_DAX
 282         if (IS_DAX(iocb->ki_filp->f_mapping->host))
 283                 return dax_iomap_rw(iocb, to, &erofs_iomap_ops);
 284 #endif
 285         if (iocb->ki_flags & IOCB_DIRECT) {
 286                 int err = erofs_prepare_dio(iocb, to);
 287
 288                 if (!err)
 289                         return iomap_dio_rw(iocb, to, &erofs_iomap_ops,
 290                                             NULL, 0);
 291                 if (err < 0)
 292                         return err;
 293         }
 294         return filemap_read(iocb, to, 0);
 295 }
 296
 297 /* for uncompressed (aligned) files and raw access for other files */
 298 const struct address_space_operations erofs_raw_access_aops = {
 299         .readpage = erofs_readpage,
 300         .readahead = erofs_readahead,
 301         .bmap = erofs_bmap,
 302         .direct_IO = noop_direct_IO,
 303 };
 304
 305 #ifdef CONFIG_FS_DAX
 306 static vm_fault_t erofs_dax_huge_fault(struct vm_fault *vmf,
 307                 enum page_entry_size pe_size)
 308 {
 309         return dax_iomap_fault(vmf, pe_size, NULL, NULL, &erofs_iomap_ops);
 310 }
 311
 312 static vm_fault_t erofs_dax_fault(struct vm_fault *vmf)
 313 {
 314         return erofs_dax_huge_fault(vmf, PE_SIZE_PTE);
 315 }
 316
 317 static const struct vm_operations_struct erofs_dax_vm_ops = {
 318         .fault          = erofs_dax_fault,
 319         .huge_fault     = erofs_dax_huge_fault,
 320 };
 321
 322 static int erofs_file_mmap(struct file *file, struct vm_area_struct *vma)
 323 {
 324         if (!IS_DAX(file_inode(file)))
 325                 return generic_file_readonly_mmap(file, vma);
 326
 327         if ((vma->vm_flags & VM_SHARED) && (vma->vm_flags & VM_MAYWRITE))
 328                 return -EINVAL;
 329
 330         vma->vm_ops = &erofs_dax_vm_ops;
 331         vma->vm_flags |= VM_HUGEPAGE;
 332         return 0;
 333 }
 334 #else
 335 #define erofs_file_mmap generic_file_readonly_mmap
 336 #endif
 337
 338 const struct file_operations erofs_file_fops = {
 339         .llseek         = generic_file_llseek,
 340         .read_iter      = erofs_file_read_iter,
 341         .mmap           = erofs_file_mmap,
 342         .splice_read    = generic_file_splice_read,
 343 };