fs/erofs/inode.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 "xattr.h"
   8 #include <trace/events/erofs.h>
   9
  10 static int erofs_fill_symlink(struct inode *inode, void *kaddr,
  11                               unsigned int m_pofs)
  12 {
  13         struct erofs_inode *vi = EROFS_I(inode);
  14         loff_t off;
  15
  16         m_pofs += vi->xattr_isize;
  17         /* check if it cannot be handled with fast symlink scheme */
  18         if (vi->datalayout != EROFS_INODE_FLAT_INLINE ||
  19             check_add_overflow(m_pofs, inode->i_size, &off) ||
  20             off > i_blocksize(inode))
  21                 return 0;
  22
  23         inode->i_link = kmemdup_nul(kaddr + m_pofs, inode->i_size, GFP_KERNEL);
  24         return inode->i_link ? 0 : -ENOMEM;
  25 }
  26
  27 static int erofs_read_inode(struct inode *inode)
  28 {
  29         struct super_block *sb = inode->i_sb;
  30         struct erofs_sb_info *sbi = EROFS_SB(sb);
  31         struct erofs_inode *vi = EROFS_I(inode);
  32         const erofs_off_t inode_loc = erofs_iloc(inode);
  33         erofs_blk_t blkaddr, nblks = 0;
  34         void *kaddr;
  35         struct erofs_inode_compact *dic;
  36         struct erofs_inode_extended *die, *copied = NULL;
  37         union erofs_inode_i_u iu;
  38         struct erofs_buf buf = __EROFS_BUF_INITIALIZER;
  39         unsigned int ifmt, ofs;
  40         int err = 0;
  41
  42         blkaddr = erofs_blknr(sb, inode_loc);
  43         ofs = erofs_blkoff(sb, inode_loc);
  44
  45         kaddr = erofs_read_metabuf(&buf, sb, erofs_pos(sb, blkaddr), EROFS_KMAP);
  46         if (IS_ERR(kaddr)) {
  47                 erofs_err(sb, "failed to get inode (nid: %llu) page, err %ld",
  48                           vi->nid, PTR_ERR(kaddr));
  49                 return PTR_ERR(kaddr);
  50         }
  51
  52         dic = kaddr + ofs;
  53         ifmt = le16_to_cpu(dic->i_format);
  54         if (ifmt & ~EROFS_I_ALL) {
  55                 erofs_err(sb, "unsupported i_format %u of nid %llu",
  56                           ifmt, vi->nid);
  57                 err = -EOPNOTSUPP;
  58                 goto err_out;
  59         }
  60
  61         vi->datalayout = erofs_inode_datalayout(ifmt);
  62         if (vi->datalayout >= EROFS_INODE_DATALAYOUT_MAX) {
  63                 erofs_err(sb, "unsupported datalayout %u of nid %llu",
  64                           vi->datalayout, vi->nid);
  65                 err = -EOPNOTSUPP;
  66                 goto err_out;
  67         }
  68
  69         switch (erofs_inode_version(ifmt)) {
  70         case EROFS_INODE_LAYOUT_EXTENDED:
  71                 vi->inode_isize = sizeof(struct erofs_inode_extended);
  72                 /* check if the extended inode acrosses block boundary */
  73                 if (ofs + vi->inode_isize <= sb->s_blocksize) {
  74                         ofs += vi->inode_isize;
  75                         die = (struct erofs_inode_extended *)dic;
  76                 } else {
  77                         const unsigned int gotten = sb->s_blocksize - ofs;
  78
  79                         copied = kmalloc(vi->inode_isize, GFP_KERNEL);
  80                         if (!copied) {
  81                                 err = -ENOMEM;
  82                                 goto err_out;
  83                         }
  84                         memcpy(copied, dic, gotten);
  85                         kaddr = erofs_read_metabuf(&buf, sb, erofs_pos(sb, blkaddr + 1),
  86                                                    EROFS_KMAP);
  87                         if (IS_ERR(kaddr)) {
  88                                 erofs_err(sb, "failed to get inode payload block (nid: %llu), err %ld",
  89                                           vi->nid, PTR_ERR(kaddr));
  90                                 kfree(copied);
  91                                 return PTR_ERR(kaddr);
  92                         }
  93                         ofs = vi->inode_isize - gotten;
  94                         memcpy((u8 *)copied + gotten, kaddr, ofs);
  95                         die = copied;
  96                 }
  97                 vi->xattr_isize = erofs_xattr_ibody_size(die->i_xattr_icount);
  98
  99                 inode->i_mode = le16_to_cpu(die->i_mode);
 100                 iu = die->i_u;
 101                 i_uid_write(inode, le32_to_cpu(die->i_uid));
 102                 i_gid_write(inode, le32_to_cpu(die->i_gid));
 103                 set_nlink(inode, le32_to_cpu(die->i_nlink));
 104                 /* each extended inode has its own timestamp */
 105                 inode_set_ctime(inode, le64_to_cpu(die->i_mtime),
 106                                 le32_to_cpu(die->i_mtime_nsec));
 107
 108                 inode->i_size = le64_to_cpu(die->i_size);
 109                 kfree(copied);
 110                 break;
 111         case EROFS_INODE_LAYOUT_COMPACT:
 112                 vi->inode_isize = sizeof(struct erofs_inode_compact);
 113                 ofs += vi->inode_isize;
 114                 vi->xattr_isize = erofs_xattr_ibody_size(dic->i_xattr_icount);
 115
 116                 inode->i_mode = le16_to_cpu(dic->i_mode);
 117                 iu = dic->i_u;
 118                 i_uid_write(inode, le16_to_cpu(dic->i_uid));
 119                 i_gid_write(inode, le16_to_cpu(dic->i_gid));
 120                 set_nlink(inode, le16_to_cpu(dic->i_nlink));
 121                 /* use build time for compact inodes */
 122                 inode_set_ctime(inode, sbi->build_time, sbi->build_time_nsec);
 123
 124                 inode->i_size = le32_to_cpu(dic->i_size);
 125                 break;
 126         default:
 127                 erofs_err(sb, "unsupported on-disk inode version %u of nid %llu",
 128                           erofs_inode_version(ifmt), vi->nid);
 129                 err = -EOPNOTSUPP;
 130                 goto err_out;
 131         }
 132
 133         if (unlikely(inode->i_size < 0)) {
 134                 erofs_err(sb, "negative i_size @ nid %llu", vi->nid);
 135                 err = -EFSCORRUPTED;
 136                 goto err_out;
 137         }
 138         switch (inode->i_mode & S_IFMT) {
 139         case S_IFREG:
 140         case S_IFDIR:
 141         case S_IFLNK:
 142                 vi->raw_blkaddr = le32_to_cpu(iu.raw_blkaddr);
 143                 if(S_ISLNK(inode->i_mode)) {
 144                         err = erofs_fill_symlink(inode, kaddr, ofs);
 145                         if (err)
 146                                 goto err_out;
 147                 }
 148                 break;
 149         case S_IFCHR:
 150         case S_IFBLK:
 151                 inode->i_rdev = new_decode_dev(le32_to_cpu(iu.rdev));
 152                 break;
 153         case S_IFIFO:
 154         case S_IFSOCK:
 155                 inode->i_rdev = 0;
 156                 break;
 157         default:
 158                 erofs_err(sb, "bogus i_mode (%o) @ nid %llu", inode->i_mode,
 159                           vi->nid);
 160                 err = -EFSCORRUPTED;
 161                 goto err_out;
 162         }
 163
 164         /* total blocks for compressed files */
 165         if (erofs_inode_is_data_compressed(vi->datalayout)) {
 166                 nblks = le32_to_cpu(iu.compressed_blocks);
 167         } else if (vi->datalayout == EROFS_INODE_CHUNK_BASED) {
 168                 /* fill chunked inode summary info */
 169                 vi->chunkformat = le16_to_cpu(iu.c.format);
 170                 if (vi->chunkformat & ~EROFS_CHUNK_FORMAT_ALL) {
 171                         erofs_err(sb, "unsupported chunk format %x of nid %llu",
 172                                   vi->chunkformat, vi->nid);
 173                         err = -EOPNOTSUPP;
 174                         goto err_out;
 175                 }
 176                 vi->chunkbits = sb->s_blocksize_bits +
 177                         (vi->chunkformat & EROFS_CHUNK_FORMAT_BLKBITS_MASK);
 178         }
 179         inode_set_mtime_to_ts(inode,
 180                               inode_set_atime_to_ts(inode, inode_get_ctime(inode)));
 181
 182         inode->i_flags &= ~S_DAX;
 183         if (test_opt(&sbi->opt, DAX_ALWAYS) && S_ISREG(inode->i_mode) &&
 184             (vi->datalayout == EROFS_INODE_FLAT_PLAIN ||
 185              vi->datalayout == EROFS_INODE_CHUNK_BASED))
 186                 inode->i_flags |= S_DAX;
 187
 188         if (!nblks)
 189                 /* measure inode.i_blocks as generic filesystems */
 190                 inode->i_blocks = round_up(inode->i_size, sb->s_blocksize) >> 9;
 191         else
 192                 inode->i_blocks = nblks << (sb->s_blocksize_bits - 9);
 193 err_out:
 194         DBG_BUGON(err);
 195         erofs_put_metabuf(&buf);
 196         return err;
 197 }
 198
 199 static int erofs_fill_inode(struct inode *inode)
 200 {
 201         struct erofs_inode *vi = EROFS_I(inode);
 202         int err;
 203
 204         trace_erofs_fill_inode(inode);
 205
 206         /* read inode base data from disk */
 207         err = erofs_read_inode(inode);
 208         if (err)
 209                 return err;
 210
 211         /* setup the new inode */
 212         switch (inode->i_mode & S_IFMT) {
 213         case S_IFREG:
 214                 inode->i_op = &erofs_generic_iops;
 215                 if (erofs_inode_is_data_compressed(vi->datalayout))
 216                         inode->i_fop = &generic_ro_fops;
 217                 else
 218                         inode->i_fop = &erofs_file_fops;
 219                 break;
 220         case S_IFDIR:
 221                 inode->i_op = &erofs_dir_iops;
 222                 inode->i_fop = &erofs_dir_fops;
 223                 inode_nohighmem(inode);
 224                 break;
 225         case S_IFLNK:
 226                 if (inode->i_link)
 227                         inode->i_op = &erofs_fast_symlink_iops;
 228                 else
 229                         inode->i_op = &erofs_symlink_iops;
 230                 inode_nohighmem(inode);
 231                 break;
 232         case S_IFCHR:
 233         case S_IFBLK:
 234         case S_IFIFO:
 235         case S_IFSOCK:
 236                 inode->i_op = &erofs_generic_iops;
 237                 init_special_inode(inode, inode->i_mode, inode->i_rdev);
 238                 return 0;
 239         default:
 240                 return -EFSCORRUPTED;
 241         }
 242
 243         mapping_set_large_folios(inode->i_mapping);
 244         if (erofs_inode_is_data_compressed(vi->datalayout)) {
 245 #ifdef CONFIG_EROFS_FS_ZIP
 246                 DO_ONCE_LITE_IF(inode->i_blkbits != PAGE_SHIFT,
 247                           erofs_info, inode->i_sb,
 248                           "EXPERIMENTAL EROFS subpage compressed block support in use. Use at your own risk!");
 249                 inode->i_mapping->a_ops = &z_erofs_aops;
 250 #else
 251                 err = -EOPNOTSUPP;
 252 #endif
 253         } else {
 254                 inode->i_mapping->a_ops = &erofs_aops;
 255 #ifdef CONFIG_EROFS_FS_ONDEMAND
 256                 if (erofs_is_fscache_mode(inode->i_sb))
 257                         inode->i_mapping->a_ops = &erofs_fscache_access_aops;
 258 #endif
 259 #ifdef CONFIG_EROFS_FS_BACKED_BY_FILE
 260                 if (erofs_is_fileio_mode(EROFS_SB(inode->i_sb)))
 261                         inode->i_mapping->a_ops = &erofs_fileio_aops;
 262 #endif
 263         }
 264
 265         return err;
 266 }
 267
 268 /*
 269  * ino_t is 32-bits on 32-bit arch. We have to squash the 64-bit value down
 270  * so that it will fit.
 271  */
 272 static ino_t erofs_squash_ino(erofs_nid_t nid)
 273 {
 274         ino_t ino = (ino_t)nid;
 275
 276         if (sizeof(ino_t) < sizeof(erofs_nid_t))
 277                 ino ^= nid >> (sizeof(erofs_nid_t) - sizeof(ino_t)) * 8;
 278         return ino;
 279 }
 280
 281 static int erofs_iget5_eq(struct inode *inode, void *opaque)
 282 {
 283         return EROFS_I(inode)->nid == *(erofs_nid_t *)opaque;
 284 }
 285
 286 static int erofs_iget5_set(struct inode *inode, void *opaque)
 287 {
 288         const erofs_nid_t nid = *(erofs_nid_t *)opaque;
 289
 290         inode->i_ino = erofs_squash_ino(nid);
 291         EROFS_I(inode)->nid = nid;
 292         return 0;
 293 }
 294
 295 struct inode *erofs_iget(struct super_block *sb, erofs_nid_t nid)
 296 {
 297         struct inode *inode;
 298
 299         inode = iget5_locked(sb, erofs_squash_ino(nid), erofs_iget5_eq,
 300                              erofs_iget5_set, &nid);
 301         if (!inode)
 302                 return ERR_PTR(-ENOMEM);
 303
 304         if (inode->i_state & I_NEW) {
 305                 int err = erofs_fill_inode(inode);
 306
 307                 if (err) {
 308                         iget_failed(inode);
 309                         return ERR_PTR(err);
 310                 }
 311                 unlock_new_inode(inode);
 312         }
 313         return inode;
 314 }
 315
 316 int erofs_getattr(struct mnt_idmap *idmap, const struct path *path,
 317                   struct kstat *stat, u32 request_mask,
 318                   unsigned int query_flags)
 319 {
 320         struct inode *const inode = d_inode(path->dentry);
 321         struct block_device *bdev = inode->i_sb->s_bdev;
 322         bool compressed =
 323                 erofs_inode_is_data_compressed(EROFS_I(inode)->datalayout);
 324
 325         if (compressed)
 326                 stat->attributes |= STATX_ATTR_COMPRESSED;
 327         stat->attributes |= STATX_ATTR_IMMUTABLE;
 328         stat->attributes_mask |= (STATX_ATTR_COMPRESSED |
 329                                   STATX_ATTR_IMMUTABLE);
 330
 331         /*
 332          * Return the DIO alignment restrictions if requested.
 333          *
 334          * In EROFS, STATX_DIOALIGN is only supported in bdev-based mode
 335          * and uncompressed inodes, otherwise we report no DIO support.
 336          */
 337         if ((request_mask & STATX_DIOALIGN) && S_ISREG(inode->i_mode)) {
 338                 stat->result_mask |= STATX_DIOALIGN;
 339                 if (bdev && !compressed) {
 340                         stat->dio_mem_align = bdev_dma_alignment(bdev) + 1;
 341                         stat->dio_offset_align = bdev_logical_block_size(bdev);
 342                 }
 343         }
 344         generic_fillattr(idmap, request_mask, inode, stat);
 345         return 0;
 346 }
 347
 348 const struct inode_operations erofs_generic_iops = {
 349         .getattr = erofs_getattr,
 350         .listxattr = erofs_listxattr,
 351         .get_inode_acl = erofs_get_acl,
 352         .fiemap = erofs_fiemap,
 353 };
 354
 355 const struct inode_operations erofs_symlink_iops = {
 356         .get_link = page_get_link,
 357         .getattr = erofs_getattr,
 358         .listxattr = erofs_listxattr,
 359         .get_inode_acl = erofs_get_acl,
 360 };
 361
 362 const struct inode_operations erofs_fast_symlink_iops = {
 363         .get_link = simple_get_link,
 364         .getattr = erofs_getattr,
 365         .listxattr = erofs_listxattr,
 366         .get_inode_acl = erofs_get_acl,
 367 };