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
   9 #include <trace/events/erofs.h>
  10
  11 static void *erofs_read_inode(struct erofs_buf *buf,
  12                               struct inode *inode, unsigned int *ofs)
  13 {
  14         struct super_block *sb = inode->i_sb;
  15         struct erofs_sb_info *sbi = EROFS_SB(sb);
  16         struct erofs_inode *vi = EROFS_I(inode);
  17         const erofs_off_t inode_loc = erofs_iloc(inode);
  18
  19         erofs_blk_t blkaddr, nblks = 0;
  20         void *kaddr;
  21         struct erofs_inode_compact *dic;
  22         struct erofs_inode_extended *die, *copied = NULL;
  23         unsigned int ifmt;
  24         int err;
  25
  26         blkaddr = erofs_blknr(sb, inode_loc);
  27         *ofs = erofs_blkoff(sb, inode_loc);
  28
  29         kaddr = erofs_read_metabuf(buf, sb, blkaddr, EROFS_KMAP);
  30         if (IS_ERR(kaddr)) {
  31                 erofs_err(sb, "failed to get inode (nid: %llu) page, err %ld",
  32                           vi->nid, PTR_ERR(kaddr));
  33                 return kaddr;
  34         }
  35
  36         dic = kaddr + *ofs;
  37         ifmt = le16_to_cpu(dic->i_format);
  38
  39         if (ifmt & ~EROFS_I_ALL) {
  40                 erofs_err(inode->i_sb, "unsupported i_format %u of nid %llu",
  41                           ifmt, vi->nid);
  42                 err = -EOPNOTSUPP;
  43                 goto err_out;
  44         }
  45
  46         vi->datalayout = erofs_inode_datalayout(ifmt);
  47         if (vi->datalayout >= EROFS_INODE_DATALAYOUT_MAX) {
  48                 erofs_err(inode->i_sb, "unsupported datalayout %u of nid %llu",
  49                           vi->datalayout, vi->nid);
  50                 err = -EOPNOTSUPP;
  51                 goto err_out;
  52         }
  53
  54         switch (erofs_inode_version(ifmt)) {
  55         case EROFS_INODE_LAYOUT_EXTENDED:
  56                 vi->inode_isize = sizeof(struct erofs_inode_extended);
  57                 /* check if the extended inode acrosses block boundary */
  58                 if (*ofs + vi->inode_isize <= sb->s_blocksize) {
  59                         *ofs += vi->inode_isize;
  60                         die = (struct erofs_inode_extended *)dic;
  61                 } else {
  62                         const unsigned int gotten = sb->s_blocksize - *ofs;
  63
  64                         copied = kmalloc(vi->inode_isize, GFP_NOFS);
  65                         if (!copied) {
  66                                 err = -ENOMEM;
  67                                 goto err_out;
  68                         }
  69                         memcpy(copied, dic, gotten);
  70                         kaddr = erofs_read_metabuf(buf, sb, blkaddr + 1,
  71                                                    EROFS_KMAP);
  72                         if (IS_ERR(kaddr)) {
  73                                 erofs_err(sb, "failed to get inode payload block (nid: %llu), err %ld",
  74                                           vi->nid, PTR_ERR(kaddr));
  75                                 kfree(copied);
  76                                 return kaddr;
  77                         }
  78                         *ofs = vi->inode_isize - gotten;
  79                         memcpy((u8 *)copied + gotten, kaddr, *ofs);
  80                         die = copied;
  81                 }
  82                 vi->xattr_isize = erofs_xattr_ibody_size(die->i_xattr_icount);
  83
  84                 inode->i_mode = le16_to_cpu(die->i_mode);
  85                 switch (inode->i_mode & S_IFMT) {
  86                 case S_IFREG:
  87                 case S_IFDIR:
  88                 case S_IFLNK:
  89                         vi->raw_blkaddr = le32_to_cpu(die->i_u.raw_blkaddr);
  90                         break;
  91                 case S_IFCHR:
  92                 case S_IFBLK:
  93                         inode->i_rdev =
  94                                 new_decode_dev(le32_to_cpu(die->i_u.rdev));
  95                         break;
  96                 case S_IFIFO:
  97                 case S_IFSOCK:
  98                         inode->i_rdev = 0;
  99                         break;
 100                 default:
 101                         goto bogusimode;
 102                 }
 103                 i_uid_write(inode, le32_to_cpu(die->i_uid));
 104                 i_gid_write(inode, le32_to_cpu(die->i_gid));
 105                 set_nlink(inode, le32_to_cpu(die->i_nlink));
 106
 107                 /* extended inode has its own timestamp */
 108                 inode_set_ctime(inode, le64_to_cpu(die->i_mtime),
 109                                 le32_to_cpu(die->i_mtime_nsec));
 110
 111                 inode->i_size = le64_to_cpu(die->i_size);
 112
 113                 /* total blocks for compressed files */
 114                 if (erofs_inode_is_data_compressed(vi->datalayout))
 115                         nblks = le32_to_cpu(die->i_u.compressed_blocks);
 116                 else if (vi->datalayout == EROFS_INODE_CHUNK_BASED)
 117                         /* fill chunked inode summary info */
 118                         vi->chunkformat = le16_to_cpu(die->i_u.c.format);
 119                 kfree(copied);
 120                 copied = NULL;
 121                 break;
 122         case EROFS_INODE_LAYOUT_COMPACT:
 123                 vi->inode_isize = sizeof(struct erofs_inode_compact);
 124                 *ofs += vi->inode_isize;
 125                 vi->xattr_isize = erofs_xattr_ibody_size(dic->i_xattr_icount);
 126
 127                 inode->i_mode = le16_to_cpu(dic->i_mode);
 128                 switch (inode->i_mode & S_IFMT) {
 129                 case S_IFREG:
 130                 case S_IFDIR:
 131                 case S_IFLNK:
 132                         vi->raw_blkaddr = le32_to_cpu(dic->i_u.raw_blkaddr);
 133                         break;
 134                 case S_IFCHR:
 135                 case S_IFBLK:
 136                         inode->i_rdev =
 137                                 new_decode_dev(le32_to_cpu(dic->i_u.rdev));
 138                         break;
 139                 case S_IFIFO:
 140                 case S_IFSOCK:
 141                         inode->i_rdev = 0;
 142                         break;
 143                 default:
 144                         goto bogusimode;
 145                 }
 146                 i_uid_write(inode, le16_to_cpu(dic->i_uid));
 147                 i_gid_write(inode, le16_to_cpu(dic->i_gid));
 148                 set_nlink(inode, le16_to_cpu(dic->i_nlink));
 149
 150                 /* use build time for compact inodes */
 151                 inode_set_ctime(inode, sbi->build_time, sbi->build_time_nsec);
 152
 153                 inode->i_size = le32_to_cpu(dic->i_size);
 154                 if (erofs_inode_is_data_compressed(vi->datalayout))
 155                         nblks = le32_to_cpu(dic->i_u.compressed_blocks);
 156                 else if (vi->datalayout == EROFS_INODE_CHUNK_BASED)
 157                         vi->chunkformat = le16_to_cpu(dic->i_u.c.format);
 158                 break;
 159         default:
 160                 erofs_err(inode->i_sb,
 161                           "unsupported on-disk inode version %u of nid %llu",
 162                           erofs_inode_version(ifmt), vi->nid);
 163                 err = -EOPNOTSUPP;
 164                 goto err_out;
 165         }
 166
 167         if (vi->datalayout == EROFS_INODE_CHUNK_BASED) {
 168                 if (vi->chunkformat & ~EROFS_CHUNK_FORMAT_ALL) {
 169                         erofs_err(inode->i_sb,
 170                                   "unsupported chunk format %x of nid %llu",
 171                                   vi->chunkformat, vi->nid);
 172                         err = -EOPNOTSUPP;
 173                         goto err_out;
 174                 }
 175                 vi->chunkbits = sb->s_blocksize_bits +
 176                         (vi->chunkformat & EROFS_CHUNK_FORMAT_BLKBITS_MASK);
 177         }
 178         inode->i_mtime = inode->i_atime = inode_get_ctime(inode);
 179
 180         inode->i_flags &= ~S_DAX;
 181         if (test_opt(&sbi->opt, DAX_ALWAYS) && S_ISREG(inode->i_mode) &&
 182             (vi->datalayout == EROFS_INODE_FLAT_PLAIN ||
 183              vi->datalayout == EROFS_INODE_CHUNK_BASED))
 184                 inode->i_flags |= S_DAX;
 185
 186         if (!nblks)
 187                 /* measure inode.i_blocks as generic filesystems */
 188                 inode->i_blocks = round_up(inode->i_size, sb->s_blocksize) >> 9;
 189         else
 190                 inode->i_blocks = nblks << (sb->s_blocksize_bits - 9);
 191         return kaddr;
 192
 193 bogusimode:
 194         erofs_err(inode->i_sb, "bogus i_mode (%o) @ nid %llu",
 195                   inode->i_mode, vi->nid);
 196         err = -EFSCORRUPTED;
 197 err_out:
 198         DBG_BUGON(1);
 199         kfree(copied);
 200         erofs_put_metabuf(buf);
 201         return ERR_PTR(err);
 202 }
 203
 204 static int erofs_fill_symlink(struct inode *inode, void *kaddr,
 205                               unsigned int m_pofs)
 206 {
 207         struct erofs_inode *vi = EROFS_I(inode);
 208         unsigned int bsz = i_blocksize(inode);
 209         char *lnk;
 210
 211         /* if it cannot be handled with fast symlink scheme */
 212         if (vi->datalayout != EROFS_INODE_FLAT_INLINE ||
 213             inode->i_size >= bsz || inode->i_size < 0) {
 214                 inode->i_op = &erofs_symlink_iops;
 215                 return 0;
 216         }
 217
 218         lnk = kmalloc(inode->i_size + 1, GFP_KERNEL);
 219         if (!lnk)
 220                 return -ENOMEM;
 221
 222         m_pofs += vi->xattr_isize;
 223         /* inline symlink data shouldn't cross block boundary */
 224         if (m_pofs + inode->i_size > bsz) {
 225                 kfree(lnk);
 226                 erofs_err(inode->i_sb,
 227                           "inline data cross block boundary @ nid %llu",
 228                           vi->nid);
 229                 DBG_BUGON(1);
 230                 return -EFSCORRUPTED;
 231         }
 232         memcpy(lnk, kaddr + m_pofs, inode->i_size);
 233         lnk[inode->i_size] = '\0';
 234
 235         inode->i_link = lnk;
 236         inode->i_op = &erofs_fast_symlink_iops;
 237         return 0;
 238 }
 239
 240 static int erofs_fill_inode(struct inode *inode)
 241 {
 242         struct erofs_inode *vi = EROFS_I(inode);
 243         struct erofs_buf buf = __EROFS_BUF_INITIALIZER;
 244         void *kaddr;
 245         unsigned int ofs;
 246         int err = 0;
 247
 248         trace_erofs_fill_inode(inode);
 249
 250         /* read inode base data from disk */
 251         kaddr = erofs_read_inode(&buf, inode, &ofs);
 252         if (IS_ERR(kaddr))
 253                 return PTR_ERR(kaddr);
 254
 255         /* setup the new inode */
 256         switch (inode->i_mode & S_IFMT) {
 257         case S_IFREG:
 258                 inode->i_op = &erofs_generic_iops;
 259                 if (erofs_inode_is_data_compressed(vi->datalayout))
 260                         inode->i_fop = &generic_ro_fops;
 261                 else
 262                         inode->i_fop = &erofs_file_fops;
 263                 break;
 264         case S_IFDIR:
 265                 inode->i_op = &erofs_dir_iops;
 266                 inode->i_fop = &erofs_dir_fops;
 267                 inode_nohighmem(inode);
 268                 break;
 269         case S_IFLNK:
 270                 err = erofs_fill_symlink(inode, kaddr, ofs);
 271                 if (err)
 272                         goto out_unlock;
 273                 inode_nohighmem(inode);
 274                 break;
 275         case S_IFCHR:
 276         case S_IFBLK:
 277         case S_IFIFO:
 278         case S_IFSOCK:
 279                 inode->i_op = &erofs_generic_iops;
 280                 init_special_inode(inode, inode->i_mode, inode->i_rdev);
 281                 goto out_unlock;
 282         default:
 283                 err = -EFSCORRUPTED;
 284                 goto out_unlock;
 285         }
 286
 287         if (erofs_inode_is_data_compressed(vi->datalayout)) {
 288 #ifdef CONFIG_EROFS_FS_ZIP
 289                 if (!erofs_is_fscache_mode(inode->i_sb) &&
 290                     inode->i_sb->s_blocksize_bits == PAGE_SHIFT) {
 291                         inode->i_mapping->a_ops = &z_erofs_aops;
 292                         err = 0;
 293                         goto out_unlock;
 294                 }
 295 #endif
 296                 err = -EOPNOTSUPP;
 297                 goto out_unlock;
 298         }
 299         inode->i_mapping->a_ops = &erofs_raw_access_aops;
 300         mapping_set_large_folios(inode->i_mapping);
 301 #ifdef CONFIG_EROFS_FS_ONDEMAND
 302         if (erofs_is_fscache_mode(inode->i_sb))
 303                 inode->i_mapping->a_ops = &erofs_fscache_access_aops;
 304 #endif
 305
 306 out_unlock:
 307         erofs_put_metabuf(&buf);
 308         return err;
 309 }
 310
 311 /*
 312  * ino_t is 32-bits on 32-bit arch. We have to squash the 64-bit value down
 313  * so that it will fit.
 314  */
 315 static ino_t erofs_squash_ino(erofs_nid_t nid)
 316 {
 317         ino_t ino = (ino_t)nid;
 318
 319         if (sizeof(ino_t) < sizeof(erofs_nid_t))
 320                 ino ^= nid >> (sizeof(erofs_nid_t) - sizeof(ino_t)) * 8;
 321         return ino;
 322 }
 323
 324 static int erofs_iget5_eq(struct inode *inode, void *opaque)
 325 {
 326         return EROFS_I(inode)->nid == *(erofs_nid_t *)opaque;
 327 }
 328
 329 static int erofs_iget5_set(struct inode *inode, void *opaque)
 330 {
 331         const erofs_nid_t nid = *(erofs_nid_t *)opaque;
 332
 333         inode->i_ino = erofs_squash_ino(nid);
 334         EROFS_I(inode)->nid = nid;
 335         return 0;
 336 }
 337
 338 struct inode *erofs_iget(struct super_block *sb, erofs_nid_t nid)
 339 {
 340         struct inode *inode;
 341
 342         inode = iget5_locked(sb, erofs_squash_ino(nid), erofs_iget5_eq,
 343                              erofs_iget5_set, &nid);
 344         if (!inode)
 345                 return ERR_PTR(-ENOMEM);
 346
 347         if (inode->i_state & I_NEW) {
 348                 int err = erofs_fill_inode(inode);
 349
 350                 if (err) {
 351                         iget_failed(inode);
 352                         return ERR_PTR(err);
 353                 }
 354                 unlock_new_inode(inode);
 355         }
 356         return inode;
 357 }
 358
 359 int erofs_getattr(struct mnt_idmap *idmap, const struct path *path,
 360                   struct kstat *stat, u32 request_mask,
 361                   unsigned int query_flags)
 362 {
 363         struct inode *const inode = d_inode(path->dentry);
 364
 365         if (erofs_inode_is_data_compressed(EROFS_I(inode)->datalayout))
 366                 stat->attributes |= STATX_ATTR_COMPRESSED;
 367
 368         stat->attributes |= STATX_ATTR_IMMUTABLE;
 369         stat->attributes_mask |= (STATX_ATTR_COMPRESSED |
 370                                   STATX_ATTR_IMMUTABLE);
 371
 372         generic_fillattr(idmap, request_mask, inode, stat);
 373         return 0;
 374 }
 375
 376 const struct inode_operations erofs_generic_iops = {
 377         .getattr = erofs_getattr,
 378         .listxattr = erofs_listxattr,
 379         .get_inode_acl = erofs_get_acl,
 380         .fiemap = erofs_fiemap,
 381 };
 382
 383 const struct inode_operations erofs_symlink_iops = {
 384         .get_link = page_get_link,
 385         .getattr = erofs_getattr,
 386         .listxattr = erofs_listxattr,
 387         .get_inode_acl = erofs_get_acl,
 388 };
 389
 390 const struct inode_operations erofs_fast_symlink_iops = {
 391         .get_link = simple_get_link,
 392         .getattr = erofs_getattr,
 393         .listxattr = erofs_listxattr,
 394         .get_inode_acl = erofs_get_acl,
 395 };