fs/overlayfs/file.c

   1 // SPDX-License-Identifier: GPL-2.0-only
   2 /*
   3  * Copyright (C) 2017 Red Hat, Inc.
   4  */
   5
   6 #include <linux/cred.h>
   7 #include <linux/file.h>
   8 #include <linux/mount.h>
   9 #include <linux/xattr.h>
  10 #include <linux/uio.h>
  11 #include <linux/uaccess.h>
  12 #include "overlayfs.h"
  13
  14 static char ovl_whatisit(struct inode *inode, struct inode *realinode)
  15 {
  16         if (realinode != ovl_inode_upper(inode))
  17                 return 'l';
  18         if (ovl_has_upperdata(inode))
  19                 return 'u';
  20         else
  21                 return 'm';
  22 }
  23
  24 /* No atime modificaton nor notify on underlying */
  25 #define OVL_OPEN_FLAGS (O_NOATIME | FMODE_NONOTIFY)
  26
  27 static struct file *ovl_open_realfile(const struct file *file,
  28                                       struct inode *realinode)
  29 {
  30         struct inode *inode = file_inode(file);
  31         struct file *realfile;
  32         const struct cred *old_cred;
  33         int flags = file->f_flags | OVL_OPEN_FLAGS;
  34
  35         old_cred = ovl_override_creds(inode->i_sb);
  36         realfile = open_with_fake_path(&file->f_path, flags, realinode,
  37                                        current_cred());
  38         revert_creds(old_cred);
  39
  40         pr_debug("open(%p[%pD2/%c], 0%o) -> (%p, 0%o)\n",
  41                  file, file, ovl_whatisit(inode, realinode), file->f_flags,
  42                  realfile, IS_ERR(realfile) ? 0 : realfile->f_flags);
  43
  44         return realfile;
  45 }
  46
  47 #define OVL_SETFL_MASK (O_APPEND | O_NONBLOCK | O_NDELAY | O_DIRECT)
  48
  49 static int ovl_change_flags(struct file *file, unsigned int flags)
  50 {
  51         struct inode *inode = file_inode(file);
  52         int err;
  53
  54         flags |= OVL_OPEN_FLAGS;
  55
  56         /* If some flag changed that cannot be changed then something's amiss */
  57         if (WARN_ON((file->f_flags ^ flags) & ~OVL_SETFL_MASK))
  58                 return -EIO;
  59
  60         flags &= OVL_SETFL_MASK;
  61
  62         if (((flags ^ file->f_flags) & O_APPEND) && IS_APPEND(inode))
  63                 return -EPERM;
  64
  65         if (flags & O_DIRECT) {
  66                 if (!file->f_mapping->a_ops ||
  67                     !file->f_mapping->a_ops->direct_IO)
  68                         return -EINVAL;
  69         }
  70
  71         if (file->f_op->check_flags) {
  72                 err = file->f_op->check_flags(flags);
  73                 if (err)
  74                         return err;
  75         }
  76
  77         spin_lock(&file->f_lock);
  78         file->f_flags = (file->f_flags & ~OVL_SETFL_MASK) | flags;
  79         spin_unlock(&file->f_lock);
  80
  81         return 0;
  82 }
  83
  84 static int ovl_real_fdget_meta(const struct file *file, struct fd *real,
  85                                bool allow_meta)
  86 {
  87         struct inode *inode = file_inode(file);
  88         struct inode *realinode;
  89
  90         real->flags = 0;
  91         real->file = file->private_data;
  92
  93         if (allow_meta)
  94                 realinode = ovl_inode_real(inode);
  95         else
  96                 realinode = ovl_inode_realdata(inode);
  97
  98         /* Has it been copied up since we'd opened it? */
  99         if (unlikely(file_inode(real->file) != realinode)) {
 100                 real->flags = FDPUT_FPUT;
 101                 real->file = ovl_open_realfile(file, realinode);
 102
 103                 return PTR_ERR_OR_ZERO(real->file);
 104         }
 105
 106         /* Did the flags change since open? */
 107         if (unlikely((file->f_flags ^ real->file->f_flags) & ~OVL_OPEN_FLAGS))
 108                 return ovl_change_flags(real->file, file->f_flags);
 109
 110         return 0;
 111 }
 112
 113 static int ovl_real_fdget(const struct file *file, struct fd *real)
 114 {
 115         return ovl_real_fdget_meta(file, real, false);
 116 }
 117
 118 static int ovl_open(struct inode *inode, struct file *file)
 119 {
 120         struct file *realfile;
 121         int err;
 122
 123         err = ovl_maybe_copy_up(file_dentry(file), file->f_flags);
 124         if (err)
 125                 return err;
 126
 127         /* No longer need these flags, so don't pass them on to underlying fs */
 128         file->f_flags &= ~(O_CREAT | O_EXCL | O_NOCTTY | O_TRUNC);
 129
 130         realfile = ovl_open_realfile(file, ovl_inode_realdata(inode));
 131         if (IS_ERR(realfile))
 132                 return PTR_ERR(realfile);
 133
 134         file->private_data = realfile;
 135
 136         return 0;
 137 }
 138
 139 static int ovl_release(struct inode *inode, struct file *file)
 140 {
 141         fput(file->private_data);
 142
 143         return 0;
 144 }
 145
 146 static loff_t ovl_llseek(struct file *file, loff_t offset, int whence)
 147 {
 148         struct inode *inode = file_inode(file);
 149         struct fd real;
 150         const struct cred *old_cred;
 151         loff_t ret;
 152
 153         /*
 154          * The two special cases below do not need to involve real fs,
 155          * so we can optimizing concurrent callers.
 156          */
 157         if (offset == 0) {
 158                 if (whence == SEEK_CUR)
 159                         return file->f_pos;
 160
 161                 if (whence == SEEK_SET)
 162                         return vfs_setpos(file, 0, 0);
 163         }
 164
 165         ret = ovl_real_fdget(file, &real);
 166         if (ret)
 167                 return ret;
 168
 169         /*
 170          * Overlay file f_pos is the master copy that is preserved
 171          * through copy up and modified on read/write, but only real
 172          * fs knows how to SEEK_HOLE/SEEK_DATA and real fs may impose
 173          * limitations that are more strict than ->s_maxbytes for specific
 174          * files, so we use the real file to perform seeks.
 175          */
 176         inode_lock(inode);
 177         real.file->f_pos = file->f_pos;
 178
 179         old_cred = ovl_override_creds(inode->i_sb);
 180         ret = vfs_llseek(real.file, offset, whence);
 181         revert_creds(old_cred);
 182
 183         file->f_pos = real.file->f_pos;
 184         inode_unlock(inode);
 185
 186         fdput(real);
 187
 188         return ret;
 189 }
 190
 191 static void ovl_file_accessed(struct file *file)
 192 {
 193         struct inode *inode, *upperinode;
 194
 195         if (file->f_flags & O_NOATIME)
 196                 return;
 197
 198         inode = file_inode(file);
 199         upperinode = ovl_inode_upper(inode);
 200
 201         if (!upperinode)
 202                 return;
 203
 204         if ((!timespec64_equal(&inode->i_mtime, &upperinode->i_mtime) ||
 205              !timespec64_equal(&inode->i_ctime, &upperinode->i_ctime))) {
 206                 inode->i_mtime = upperinode->i_mtime;
 207                 inode->i_ctime = upperinode->i_ctime;
 208         }
 209
 210         touch_atime(&file->f_path);
 211 }
 212
 213 static rwf_t ovl_iocb_to_rwf(struct kiocb *iocb)
 214 {
 215         int ifl = iocb->ki_flags;
 216         rwf_t flags = 0;
 217
 218         if (ifl & IOCB_NOWAIT)
 219                 flags |= RWF_NOWAIT;
 220         if (ifl & IOCB_HIPRI)
 221                 flags |= RWF_HIPRI;
 222         if (ifl & IOCB_DSYNC)
 223                 flags |= RWF_DSYNC;
 224         if (ifl & IOCB_SYNC)
 225                 flags |= RWF_SYNC;
 226
 227         return flags;
 228 }
 229
 230 static ssize_t ovl_read_iter(struct kiocb *iocb, struct iov_iter *iter)
 231 {
 232         struct file *file = iocb->ki_filp;
 233         struct fd real;
 234         const struct cred *old_cred;
 235         ssize_t ret;
 236
 237         if (!iov_iter_count(iter))
 238                 return 0;
 239
 240         ret = ovl_real_fdget(file, &real);
 241         if (ret)
 242                 return ret;
 243
 244         old_cred = ovl_override_creds(file_inode(file)->i_sb);
 245         ret = vfs_iter_read(real.file, iter, &iocb->ki_pos,
 246                             ovl_iocb_to_rwf(iocb));
 247         revert_creds(old_cred);
 248
 249         ovl_file_accessed(file);
 250
 251         fdput(real);
 252
 253         return ret;
 254 }
 255
 256 static ssize_t ovl_write_iter(struct kiocb *iocb, struct iov_iter *iter)
 257 {
 258         struct file *file = iocb->ki_filp;
 259         struct inode *inode = file_inode(file);
 260         struct fd real;
 261         const struct cred *old_cred;
 262         ssize_t ret;
 263
 264         if (!iov_iter_count(iter))
 265                 return 0;
 266
 267         inode_lock(inode);
 268         /* Update mode */
 269         ovl_copyattr(ovl_inode_real(inode), inode);
 270         ret = file_remove_privs(file);
 271         if (ret)
 272                 goto out_unlock;
 273
 274         ret = ovl_real_fdget(file, &real);
 275         if (ret)
 276                 goto out_unlock;
 277
 278         old_cred = ovl_override_creds(file_inode(file)->i_sb);
 279         file_start_write(real.file);
 280         ret = vfs_iter_write(real.file, iter, &iocb->ki_pos,
 281                              ovl_iocb_to_rwf(iocb));
 282         file_end_write(real.file);
 283         revert_creds(old_cred);
 284
 285         /* Update size */
 286         ovl_copyattr(ovl_inode_real(inode), inode);
 287
 288         fdput(real);
 289
 290 out_unlock:
 291         inode_unlock(inode);
 292
 293         return ret;
 294 }
 295
 296 static int ovl_fsync(struct file *file, loff_t start, loff_t end, int datasync)
 297 {
 298         struct fd real;
 299         const struct cred *old_cred;
 300         int ret;
 301
 302         ret = ovl_real_fdget_meta(file, &real, !datasync);
 303         if (ret)
 304                 return ret;
 305
 306         /* Don't sync lower file for fear of receiving EROFS error */
 307         if (file_inode(real.file) == ovl_inode_upper(file_inode(file))) {
 308                 old_cred = ovl_override_creds(file_inode(file)->i_sb);
 309                 ret = vfs_fsync_range(real.file, start, end, datasync);
 310                 revert_creds(old_cred);
 311         }
 312
 313         fdput(real);
 314
 315         return ret;
 316 }
 317
 318 static int ovl_mmap(struct file *file, struct vm_area_struct *vma)
 319 {
 320         struct file *realfile = file->private_data;
 321         const struct cred *old_cred;
 322         int ret;
 323
 324         if (!realfile->f_op->mmap)
 325                 return -ENODEV;
 326
 327         if (WARN_ON(file != vma->vm_file))
 328                 return -EIO;
 329
 330         vma->vm_file = get_file(realfile);
 331
 332         old_cred = ovl_override_creds(file_inode(file)->i_sb);
 333         ret = call_mmap(vma->vm_file, vma);
 334         revert_creds(old_cred);
 335
 336         if (ret) {
 337                 /* Drop reference count from new vm_file value */
 338                 fput(realfile);
 339         } else {
 340                 /* Drop reference count from previous vm_file value */
 341                 fput(file);
 342         }
 343
 344         ovl_file_accessed(file);
 345
 346         return ret;
 347 }
 348
 349 static long ovl_fallocate(struct file *file, int mode, loff_t offset, loff_t len)
 350 {
 351         struct inode *inode = file_inode(file);
 352         struct fd real;
 353         const struct cred *old_cred;
 354         int ret;
 355
 356         ret = ovl_real_fdget(file, &real);
 357         if (ret)
 358                 return ret;
 359
 360         old_cred = ovl_override_creds(file_inode(file)->i_sb);
 361         ret = vfs_fallocate(real.file, mode, offset, len);
 362         revert_creds(old_cred);
 363
 364         /* Update size */
 365         ovl_copyattr(ovl_inode_real(inode), inode);
 366
 367         fdput(real);
 368
 369         return ret;
 370 }
 371
 372 static int ovl_fadvise(struct file *file, loff_t offset, loff_t len, int advice)
 373 {
 374         struct fd real;
 375         const struct cred *old_cred;
 376         int ret;
 377
 378         ret = ovl_real_fdget(file, &real);
 379         if (ret)
 380                 return ret;
 381
 382         old_cred = ovl_override_creds(file_inode(file)->i_sb);
 383         ret = vfs_fadvise(real.file, offset, len, advice);
 384         revert_creds(old_cred);
 385
 386         fdput(real);
 387
 388         return ret;
 389 }
 390
 391 static long ovl_real_ioctl(struct file *file, unsigned int cmd,
 392                            unsigned long arg)
 393 {
 394         struct fd real;
 395         const struct cred *old_cred;
 396         long ret;
 397
 398         ret = ovl_real_fdget(file, &real);
 399         if (ret)
 400                 return ret;
 401
 402         old_cred = ovl_override_creds(file_inode(file)->i_sb);
 403         ret = vfs_ioctl(real.file, cmd, arg);
 404         revert_creds(old_cred);
 405
 406         fdput(real);
 407
 408         return ret;
 409 }
 410
 411 static long ovl_ioctl_set_flags(struct file *file, unsigned int cmd,
 412                                 unsigned long arg, unsigned int iflags)
 413 {
 414         long ret;
 415         struct inode *inode = file_inode(file);
 416         unsigned int old_iflags;
 417
 418         if (!inode_owner_or_capable(inode))
 419                 return -EACCES;
 420
 421         ret = mnt_want_write_file(file);
 422         if (ret)
 423                 return ret;
 424
 425         inode_lock(inode);
 426
 427         /* Check the capability before cred override */
 428         ret = -EPERM;
 429         old_iflags = READ_ONCE(inode->i_flags);
 430         if (((iflags ^ old_iflags) & (S_APPEND | S_IMMUTABLE)) &&
 431             !capable(CAP_LINUX_IMMUTABLE))
 432                 goto unlock;
 433
 434         ret = ovl_maybe_copy_up(file_dentry(file), O_WRONLY);
 435         if (ret)
 436                 goto unlock;
 437
 438         ret = ovl_real_ioctl(file, cmd, arg);
 439
 440         ovl_copyflags(ovl_inode_real(inode), inode);
 441 unlock:
 442         inode_unlock(inode);
 443
 444         mnt_drop_write_file(file);
 445
 446         return ret;
 447
 448 }
 449
 450 static unsigned int ovl_fsflags_to_iflags(unsigned int flags)
 451 {
 452         unsigned int iflags = 0;
 453
 454         if (flags & FS_SYNC_FL)
 455                 iflags |= S_SYNC;
 456         if (flags & FS_APPEND_FL)
 457                 iflags |= S_APPEND;
 458         if (flags & FS_IMMUTABLE_FL)
 459                 iflags |= S_IMMUTABLE;
 460         if (flags & FS_NOATIME_FL)
 461                 iflags |= S_NOATIME;
 462
 463         return iflags;
 464 }
 465
 466 static long ovl_ioctl_set_fsflags(struct file *file, unsigned int cmd,
 467                                   unsigned long arg)
 468 {
 469         unsigned int flags;
 470
 471         if (get_user(flags, (int __user *) arg))
 472                 return -EFAULT;
 473
 474         return ovl_ioctl_set_flags(file, cmd, arg,
 475                                    ovl_fsflags_to_iflags(flags));
 476 }
 477
 478 static unsigned int ovl_fsxflags_to_iflags(unsigned int xflags)
 479 {
 480         unsigned int iflags = 0;
 481
 482         if (xflags & FS_XFLAG_SYNC)
 483                 iflags |= S_SYNC;
 484         if (xflags & FS_XFLAG_APPEND)
 485                 iflags |= S_APPEND;
 486         if (xflags & FS_XFLAG_IMMUTABLE)
 487                 iflags |= S_IMMUTABLE;
 488         if (xflags & FS_XFLAG_NOATIME)
 489                 iflags |= S_NOATIME;
 490
 491         return iflags;
 492 }
 493
 494 static long ovl_ioctl_set_fsxflags(struct file *file, unsigned int cmd,
 495                                    unsigned long arg)
 496 {
 497         struct fsxattr fa;
 498
 499         memset(&fa, 0, sizeof(fa));
 500         if (copy_from_user(&fa, (void __user *) arg, sizeof(fa)))
 501                 return -EFAULT;
 502
 503         return ovl_ioctl_set_flags(file, cmd, arg,
 504                                    ovl_fsxflags_to_iflags(fa.fsx_xflags));
 505 }
 506
 507 static long ovl_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
 508 {
 509         long ret;
 510
 511         switch (cmd) {
 512         case FS_IOC_GETFLAGS:
 513         case FS_IOC_FSGETXATTR:
 514                 ret = ovl_real_ioctl(file, cmd, arg);
 515                 break;
 516
 517         case FS_IOC_SETFLAGS:
 518                 ret = ovl_ioctl_set_fsflags(file, cmd, arg);
 519                 break;
 520
 521         case FS_IOC_FSSETXATTR:
 522                 ret = ovl_ioctl_set_fsxflags(file, cmd, arg);
 523                 break;
 524
 525         default:
 526                 ret = -ENOTTY;
 527         }
 528
 529         return ret;
 530 }
 531
 532 static long ovl_compat_ioctl(struct file *file, unsigned int cmd,
 533                              unsigned long arg)
 534 {
 535         switch (cmd) {
 536         case FS_IOC32_GETFLAGS:
 537                 cmd = FS_IOC_GETFLAGS;
 538                 break;
 539
 540         case FS_IOC32_SETFLAGS:
 541                 cmd = FS_IOC_SETFLAGS;
 542                 break;
 543
 544         default:
 545                 return -ENOIOCTLCMD;
 546         }
 547
 548         return ovl_ioctl(file, cmd, arg);
 549 }
 550
 551 enum ovl_copyop {
 552         OVL_COPY,
 553         OVL_CLONE,
 554         OVL_DEDUPE,
 555 };
 556
 557 static loff_t ovl_copyfile(struct file *file_in, loff_t pos_in,
 558                             struct file *file_out, loff_t pos_out,
 559                             loff_t len, unsigned int flags, enum ovl_copyop op)
 560 {
 561         struct inode *inode_out = file_inode(file_out);
 562         struct fd real_in, real_out;
 563         const struct cred *old_cred;
 564         loff_t ret;
 565
 566         ret = ovl_real_fdget(file_out, &real_out);
 567         if (ret)
 568                 return ret;
 569
 570         ret = ovl_real_fdget(file_in, &real_in);
 571         if (ret) {
 572                 fdput(real_out);
 573                 return ret;
 574         }
 575
 576         old_cred = ovl_override_creds(file_inode(file_out)->i_sb);
 577         switch (op) {
 578         case OVL_COPY:
 579                 ret = vfs_copy_file_range(real_in.file, pos_in,
 580                                           real_out.file, pos_out, len, flags);
 581                 break;
 582
 583         case OVL_CLONE:
 584                 ret = vfs_clone_file_range(real_in.file, pos_in,
 585                                            real_out.file, pos_out, len, flags);
 586                 break;
 587
 588         case OVL_DEDUPE:
 589                 ret = vfs_dedupe_file_range_one(real_in.file, pos_in,
 590                                                 real_out.file, pos_out, len,
 591                                                 flags);
 592                 break;
 593         }
 594         revert_creds(old_cred);
 595
 596         /* Update size */
 597         ovl_copyattr(ovl_inode_real(inode_out), inode_out);
 598
 599         fdput(real_in);
 600         fdput(real_out);
 601
 602         return ret;
 603 }
 604
 605 static ssize_t ovl_copy_file_range(struct file *file_in, loff_t pos_in,
 606                                    struct file *file_out, loff_t pos_out,
 607                                    size_t len, unsigned int flags)
 608 {
 609         return ovl_copyfile(file_in, pos_in, file_out, pos_out, len, flags,
 610                             OVL_COPY);
 611 }
 612
 613 static loff_t ovl_remap_file_range(struct file *file_in, loff_t pos_in,
 614                                    struct file *file_out, loff_t pos_out,
 615                                    loff_t len, unsigned int remap_flags)
 616 {
 617         enum ovl_copyop op;
 618
 619         if (remap_flags & ~(REMAP_FILE_DEDUP | REMAP_FILE_ADVISORY))
 620                 return -EINVAL;
 621
 622         if (remap_flags & REMAP_FILE_DEDUP)
 623                 op = OVL_DEDUPE;
 624         else
 625                 op = OVL_CLONE;
 626
 627         /*
 628          * Don't copy up because of a dedupe request, this wouldn't make sense
 629          * most of the time (data would be duplicated instead of deduplicated).
 630          */
 631         if (op == OVL_DEDUPE &&
 632             (!ovl_inode_upper(file_inode(file_in)) ||
 633              !ovl_inode_upper(file_inode(file_out))))
 634                 return -EPERM;
 635
 636         return ovl_copyfile(file_in, pos_in, file_out, pos_out, len,
 637                             remap_flags, op);
 638 }
 639
 640 const struct file_operations ovl_file_operations = {
 641         .open           = ovl_open,
 642         .release        = ovl_release,
 643         .llseek         = ovl_llseek,
 644         .read_iter      = ovl_read_iter,
 645         .write_iter     = ovl_write_iter,
 646         .fsync          = ovl_fsync,
 647         .mmap           = ovl_mmap,
 648         .fallocate      = ovl_fallocate,
 649         .fadvise        = ovl_fadvise,
 650         .unlocked_ioctl = ovl_ioctl,
 651         .compat_ioctl   = ovl_compat_ioctl,
 652
 653         .copy_file_range        = ovl_copy_file_range,
 654         .remap_file_range       = ovl_remap_file_range,
 655 };