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 /*
  12  * if inode is successfully read, return its inode page (or sometimes
  13  * the inode payload page if it's an extended inode) in order to fill
  14  * inline data if possible.
  15  */
  16 static struct page *erofs_read_inode(struct inode *inode,
  17                                      unsigned int *ofs)
  18 {
  19         struct super_block *sb = inode->i_sb;
  20         struct erofs_sb_info *sbi = EROFS_SB(sb);
  21         struct erofs_inode *vi = EROFS_I(inode);
  22         const erofs_off_t inode_loc = iloc(sbi, vi->nid);
  23
  24         erofs_blk_t blkaddr, nblks = 0;
  25         struct page *page;
  26         struct erofs_inode_compact *dic;
  27         struct erofs_inode_extended *die, *copied = NULL;
  28         unsigned int ifmt;
  29         int err;
  30
  31         blkaddr = erofs_blknr(inode_loc);
  32         *ofs = erofs_blkoff(inode_loc);
  33
  34         erofs_dbg("%s, reading inode nid %llu at %u of blkaddr %u",
  35                   __func__, vi->nid, *ofs, blkaddr);
  36
  37         page = erofs_get_meta_page(sb, blkaddr);
  38         if (IS_ERR(page)) {
  39                 erofs_err(sb, "failed to get inode (nid: %llu) page, err %ld",
  40                           vi->nid, PTR_ERR(page));
  41                 return page;
  42         }
  43
  44         dic = page_address(page) + *ofs;
  45         ifmt = le16_to_cpu(dic->i_format);
  46
  47         if (ifmt & ~EROFS_I_ALL) {
  48                 erofs_err(inode->i_sb, "unsupported i_format %u of nid %llu",
  49                           ifmt, vi->nid);
  50                 err = -EOPNOTSUPP;
  51                 goto err_out;
  52         }
  53
  54         vi->datalayout = erofs_inode_datalayout(ifmt);
  55         if (vi->datalayout >= EROFS_INODE_DATALAYOUT_MAX) {
  56                 erofs_err(inode->i_sb, "unsupported datalayout %u of nid %llu",
  57                           vi->datalayout, vi->nid);
  58                 err = -EOPNOTSUPP;
  59                 goto err_out;
  60         }
  61
  62         switch (erofs_inode_version(ifmt)) {
  63         case EROFS_INODE_LAYOUT_EXTENDED:
  64                 vi->inode_isize = sizeof(struct erofs_inode_extended);
  65                 /* check if the inode acrosses page boundary */
  66                 if (*ofs + vi->inode_isize <= PAGE_SIZE) {
  67                         *ofs += vi->inode_isize;
  68                         die = (struct erofs_inode_extended *)dic;
  69                 } else {
  70                         const unsigned int gotten = PAGE_SIZE - *ofs;
  71
  72                         copied = kmalloc(vi->inode_isize, GFP_NOFS);
  73                         if (!copied) {
  74                                 err = -ENOMEM;
  75                                 goto err_out;
  76                         }
  77                         memcpy(copied, dic, gotten);
  78                         unlock_page(page);
  79                         put_page(page);
  80
  81                         page = erofs_get_meta_page(sb, blkaddr + 1);
  82                         if (IS_ERR(page)) {
  83                                 erofs_err(sb, "failed to get inode payload page (nid: %llu), err %ld",
  84                                           vi->nid, PTR_ERR(page));
  85                                 kfree(copied);
  86                                 return page;
  87                         }
  88                         *ofs = vi->inode_isize - gotten;
  89                         memcpy((u8 *)copied + gotten, page_address(page), *ofs);
  90                         die = copied;
  91                 }
  92                 vi->xattr_isize = erofs_xattr_ibody_size(die->i_xattr_icount);
  93
  94                 inode->i_mode = le16_to_cpu(die->i_mode);
  95                 switch (inode->i_mode & S_IFMT) {
  96                 case S_IFREG:
  97                 case S_IFDIR:
  98                 case S_IFLNK:
  99                         vi->raw_blkaddr = le32_to_cpu(die->i_u.raw_blkaddr);
 100                         break;
 101                 case S_IFCHR:
 102                 case S_IFBLK:
 103                         inode->i_rdev =
 104                                 new_decode_dev(le32_to_cpu(die->i_u.rdev));
 105                         break;
 106                 case S_IFIFO:
 107                 case S_IFSOCK:
 108                         inode->i_rdev = 0;
 109                         break;
 110                 default:
 111                         goto bogusimode;
 112                 }
 113                 i_uid_write(inode, le32_to_cpu(die->i_uid));
 114                 i_gid_write(inode, le32_to_cpu(die->i_gid));
 115                 set_nlink(inode, le32_to_cpu(die->i_nlink));
 116
 117                 /* extended inode has its own timestamp */
 118                 inode->i_ctime.tv_sec = le64_to_cpu(die->i_ctime);
 119                 inode->i_ctime.tv_nsec = le32_to_cpu(die->i_ctime_nsec);
 120
 121                 inode->i_size = le64_to_cpu(die->i_size);
 122
 123                 /* total blocks for compressed files */
 124                 if (erofs_inode_is_data_compressed(vi->datalayout))
 125                         nblks = le32_to_cpu(die->i_u.compressed_blocks);
 126                 else if (vi->datalayout == EROFS_INODE_CHUNK_BASED)
 127                         /* fill chunked inode summary info */
 128                         vi->chunkformat = le16_to_cpu(die->i_u.c.format);
 129                 kfree(copied);
 130                 copied = NULL;
 131                 break;
 132         case EROFS_INODE_LAYOUT_COMPACT:
 133                 vi->inode_isize = sizeof(struct erofs_inode_compact);
 134                 *ofs += vi->inode_isize;
 135                 vi->xattr_isize = erofs_xattr_ibody_size(dic->i_xattr_icount);
 136
 137                 inode->i_mode = le16_to_cpu(dic->i_mode);
 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(dic->i_u.raw_blkaddr);
 143                         break;
 144                 case S_IFCHR:
 145                 case S_IFBLK:
 146                         inode->i_rdev =
 147                                 new_decode_dev(le32_to_cpu(dic->i_u.rdev));
 148                         break;
 149                 case S_IFIFO:
 150                 case S_IFSOCK:
 151                         inode->i_rdev = 0;
 152                         break;
 153                 default:
 154                         goto bogusimode;
 155                 }
 156                 i_uid_write(inode, le16_to_cpu(dic->i_uid));
 157                 i_gid_write(inode, le16_to_cpu(dic->i_gid));
 158                 set_nlink(inode, le16_to_cpu(dic->i_nlink));
 159
 160                 /* use build time for compact inodes */
 161                 inode->i_ctime.tv_sec = sbi->build_time;
 162                 inode->i_ctime.tv_nsec = sbi->build_time_nsec;
 163
 164                 inode->i_size = le32_to_cpu(dic->i_size);
 165                 if (erofs_inode_is_data_compressed(vi->datalayout))
 166                         nblks = le32_to_cpu(dic->i_u.compressed_blocks);
 167                 else if (vi->datalayout == EROFS_INODE_CHUNK_BASED)
 168                         vi->chunkformat = le16_to_cpu(dic->i_u.c.format);
 169                 break;
 170         default:
 171                 erofs_err(inode->i_sb,
 172                           "unsupported on-disk inode version %u of nid %llu",
 173                           erofs_inode_version(ifmt), vi->nid);
 174                 err = -EOPNOTSUPP;
 175                 goto err_out;
 176         }
 177
 178         if (vi->datalayout == EROFS_INODE_CHUNK_BASED) {
 179                 if (vi->chunkformat & ~EROFS_CHUNK_FORMAT_ALL) {
 180                         erofs_err(inode->i_sb,
 181                                   "unsupported chunk format %x of nid %llu",
 182                                   vi->chunkformat, vi->nid);
 183                         err = -EOPNOTSUPP;
 184                         goto err_out;
 185                 }
 186                 vi->chunkbits = LOG_BLOCK_SIZE +
 187                         (vi->chunkformat & EROFS_CHUNK_FORMAT_BLKBITS_MASK);
 188         }
 189         inode->i_mtime.tv_sec = inode->i_ctime.tv_sec;
 190         inode->i_atime.tv_sec = inode->i_ctime.tv_sec;
 191         inode->i_mtime.tv_nsec = inode->i_ctime.tv_nsec;
 192         inode->i_atime.tv_nsec = inode->i_ctime.tv_nsec;
 193
 194         inode->i_flags &= ~S_DAX;
 195         if (test_opt(&sbi->opt, DAX_ALWAYS) && S_ISREG(inode->i_mode) &&
 196             (vi->datalayout == EROFS_INODE_FLAT_PLAIN ||
 197              vi->datalayout == EROFS_INODE_CHUNK_BASED))
 198                 inode->i_flags |= S_DAX;
 199         if (!nblks)
 200                 /* measure inode.i_blocks as generic filesystems */
 201                 inode->i_blocks = roundup(inode->i_size, EROFS_BLKSIZ) >> 9;
 202         else
 203                 inode->i_blocks = nblks << LOG_SECTORS_PER_BLOCK;
 204         return page;
 205
 206 bogusimode:
 207         erofs_err(inode->i_sb, "bogus i_mode (%o) @ nid %llu",
 208                   inode->i_mode, vi->nid);
 209         err = -EFSCORRUPTED;
 210 err_out:
 211         DBG_BUGON(1);
 212         kfree(copied);
 213         unlock_page(page);
 214         put_page(page);
 215         return ERR_PTR(err);
 216 }
 217
 218 static int erofs_fill_symlink(struct inode *inode, void *data,
 219                               unsigned int m_pofs)
 220 {
 221         struct erofs_inode *vi = EROFS_I(inode);
 222         char *lnk;
 223
 224         /* if it cannot be handled with fast symlink scheme */
 225         if (vi->datalayout != EROFS_INODE_FLAT_INLINE ||
 226             inode->i_size >= PAGE_SIZE || inode->i_size < 0) {
 227                 inode->i_op = &erofs_symlink_iops;
 228                 return 0;
 229         }
 230
 231         lnk = kmalloc(inode->i_size + 1, GFP_KERNEL);
 232         if (!lnk)
 233                 return -ENOMEM;
 234
 235         m_pofs += vi->xattr_isize;
 236         /* inline symlink data shouldn't cross page boundary as well */
 237         if (m_pofs + inode->i_size > PAGE_SIZE) {
 238                 kfree(lnk);
 239                 erofs_err(inode->i_sb,
 240                           "inline data cross block boundary @ nid %llu",
 241                           vi->nid);
 242                 DBG_BUGON(1);
 243                 return -EFSCORRUPTED;
 244         }
 245
 246         memcpy(lnk, data + m_pofs, inode->i_size);
 247         lnk[inode->i_size] = '\0';
 248
 249         inode->i_link = lnk;
 250         inode->i_op = &erofs_fast_symlink_iops;
 251         return 0;
 252 }
 253
 254 static int erofs_fill_inode(struct inode *inode, int isdir)
 255 {
 256         struct erofs_inode *vi = EROFS_I(inode);
 257         struct page *page;
 258         unsigned int ofs;
 259         int err = 0;
 260
 261         trace_erofs_fill_inode(inode, isdir);
 262
 263         /* read inode base data from disk */
 264         page = erofs_read_inode(inode, &ofs);
 265         if (IS_ERR(page))
 266                 return PTR_ERR(page);
 267
 268         /* setup the new inode */
 269         switch (inode->i_mode & S_IFMT) {
 270         case S_IFREG:
 271                 inode->i_op = &erofs_generic_iops;
 272                 if (erofs_inode_is_data_compressed(vi->datalayout))
 273                         inode->i_fop = &generic_ro_fops;
 274                 else
 275                         inode->i_fop = &erofs_file_fops;
 276                 break;
 277         case S_IFDIR:
 278                 inode->i_op = &erofs_dir_iops;
 279                 inode->i_fop = &erofs_dir_fops;
 280                 break;
 281         case S_IFLNK:
 282                 err = erofs_fill_symlink(inode, page_address(page), ofs);
 283                 if (err)
 284                         goto out_unlock;
 285                 inode_nohighmem(inode);
 286                 break;
 287         case S_IFCHR:
 288         case S_IFBLK:
 289         case S_IFIFO:
 290         case S_IFSOCK:
 291                 inode->i_op = &erofs_generic_iops;
 292                 init_special_inode(inode, inode->i_mode, inode->i_rdev);
 293                 goto out_unlock;
 294         default:
 295                 err = -EFSCORRUPTED;
 296                 goto out_unlock;
 297         }
 298
 299         if (erofs_inode_is_data_compressed(vi->datalayout)) {
 300                 err = z_erofs_fill_inode(inode);
 301                 goto out_unlock;
 302         }
 303         inode->i_mapping->a_ops = &erofs_raw_access_aops;
 304
 305 out_unlock:
 306         unlock_page(page);
 307         put_page(page);
 308         return err;
 309 }
 310
 311 /*
 312  * erofs nid is 64bits, but i_ino is 'unsigned long', therefore
 313  * we should do more for 32-bit platform to find the right inode.
 314  */
 315 static int erofs_ilookup_test_actor(struct inode *inode, void *opaque)
 316 {
 317         const erofs_nid_t nid = *(erofs_nid_t *)opaque;
 318
 319         return EROFS_I(inode)->nid == nid;
 320 }
 321
 322 static int erofs_iget_set_actor(struct inode *inode, void *opaque)
 323 {
 324         const erofs_nid_t nid = *(erofs_nid_t *)opaque;
 325
 326         inode->i_ino = erofs_inode_hash(nid);
 327         return 0;
 328 }
 329
 330 static inline struct inode *erofs_iget_locked(struct super_block *sb,
 331                                               erofs_nid_t nid)
 332 {
 333         const unsigned long hashval = erofs_inode_hash(nid);
 334
 335         return iget5_locked(sb, hashval, erofs_ilookup_test_actor,
 336                 erofs_iget_set_actor, &nid);
 337 }
 338
 339 struct inode *erofs_iget(struct super_block *sb,
 340                          erofs_nid_t nid,
 341                          bool isdir)
 342 {
 343         struct inode *inode = erofs_iget_locked(sb, nid);
 344
 345         if (!inode)
 346                 return ERR_PTR(-ENOMEM);
 347
 348         if (inode->i_state & I_NEW) {
 349                 int err;
 350                 struct erofs_inode *vi = EROFS_I(inode);
 351
 352                 vi->nid = nid;
 353
 354                 err = erofs_fill_inode(inode, isdir);
 355                 if (!err)
 356                         unlock_new_inode(inode);
 357                 else {
 358                         iget_failed(inode);
 359                         inode = ERR_PTR(err);
 360                 }
 361         }
 362         return inode;
 363 }
 364
 365 int erofs_getattr(struct user_namespace *mnt_userns, const struct path *path,
 366                   struct kstat *stat, u32 request_mask,
 367                   unsigned int query_flags)
 368 {
 369         struct inode *const inode = d_inode(path->dentry);
 370
 371         if (erofs_inode_is_data_compressed(EROFS_I(inode)->datalayout))
 372                 stat->attributes |= STATX_ATTR_COMPRESSED;
 373
 374         stat->attributes |= STATX_ATTR_IMMUTABLE;
 375         stat->attributes_mask |= (STATX_ATTR_COMPRESSED |
 376                                   STATX_ATTR_IMMUTABLE);
 377
 378         generic_fillattr(&init_user_ns, inode, stat);
 379         return 0;
 380 }
 381
 382 const struct inode_operations erofs_generic_iops = {
 383         .getattr = erofs_getattr,
 384         .listxattr = erofs_listxattr,
 385         .get_acl = erofs_get_acl,
 386         .fiemap = erofs_fiemap,
 387 };
 388
 389 const struct inode_operations erofs_symlink_iops = {
 390         .get_link = page_get_link,
 391         .getattr = erofs_getattr,
 392         .listxattr = erofs_listxattr,
 393         .get_acl = erofs_get_acl,
 394 };
 395
 396 const struct inode_operations erofs_fast_symlink_iops = {
 397         .get_link = simple_get_link,
 398         .getattr = erofs_getattr,
 399         .listxattr = erofs_listxattr,
 400         .get_acl = erofs_get_acl,
 401 };