1 // SPDX-License-Identifier: GPL-2.0-only
3 * Copyright (C) 2017 Red Hat, Inc.
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 <linux/splice.h>
15 #include "overlayfs.h"
19 struct kiocb *orig_iocb;
23 static struct kmem_cache *ovl_aio_request_cachep;
25 static char ovl_whatisit(struct inode *inode, struct inode *realinode)
27 if (realinode != ovl_inode_upper(inode))
29 if (ovl_has_upperdata(inode))
35 static struct file *ovl_open_realfile(const struct file *file,
36 struct inode *realinode)
38 struct inode *inode = file_inode(file);
39 struct file *realfile;
40 const struct cred *old_cred;
41 int flags = file->f_flags | O_NOATIME | FMODE_NONOTIFY;
43 old_cred = ovl_override_creds(inode->i_sb);
44 realfile = open_with_fake_path(&file->f_path, flags, realinode,
46 revert_creds(old_cred);
48 pr_debug("open(%p[%pD2/%c], 0%o) -> (%p, 0%o)\n",
49 file, file, ovl_whatisit(inode, realinode), file->f_flags,
50 realfile, IS_ERR(realfile) ? 0 : realfile->f_flags);
55 #define OVL_SETFL_MASK (O_APPEND | O_NONBLOCK | O_NDELAY | O_DIRECT)
57 static int ovl_change_flags(struct file *file, unsigned int flags)
59 struct inode *inode = file_inode(file);
62 /* No atime modificaton on underlying */
63 flags |= O_NOATIME | FMODE_NONOTIFY;
65 /* If some flag changed that cannot be changed then something's amiss */
66 if (WARN_ON((file->f_flags ^ flags) & ~OVL_SETFL_MASK))
69 flags &= OVL_SETFL_MASK;
71 if (((flags ^ file->f_flags) & O_APPEND) && IS_APPEND(inode))
74 if (flags & O_DIRECT) {
75 if (!file->f_mapping->a_ops ||
76 !file->f_mapping->a_ops->direct_IO)
80 if (file->f_op->check_flags) {
81 err = file->f_op->check_flags(flags);
86 spin_lock(&file->f_lock);
87 file->f_flags = (file->f_flags & ~OVL_SETFL_MASK) | flags;
88 spin_unlock(&file->f_lock);
93 static int ovl_real_fdget_meta(const struct file *file, struct fd *real,
96 struct inode *inode = file_inode(file);
97 struct inode *realinode;
100 real->file = file->private_data;
103 realinode = ovl_inode_real(inode);
105 realinode = ovl_inode_realdata(inode);
107 /* Has it been copied up since we'd opened it? */
108 if (unlikely(file_inode(real->file) != realinode)) {
109 real->flags = FDPUT_FPUT;
110 real->file = ovl_open_realfile(file, realinode);
112 return PTR_ERR_OR_ZERO(real->file);
115 /* Did the flags change since open? */
116 if (unlikely((file->f_flags ^ real->file->f_flags) & ~O_NOATIME))
117 return ovl_change_flags(real->file, file->f_flags);
122 static int ovl_real_fdget(const struct file *file, struct fd *real)
124 return ovl_real_fdget_meta(file, real, false);
127 static int ovl_open(struct inode *inode, struct file *file)
129 struct file *realfile;
132 err = ovl_maybe_copy_up(file_dentry(file), file->f_flags);
136 /* No longer need these flags, so don't pass them on to underlying fs */
137 file->f_flags &= ~(O_CREAT | O_EXCL | O_NOCTTY | O_TRUNC);
139 realfile = ovl_open_realfile(file, ovl_inode_realdata(inode));
140 if (IS_ERR(realfile))
141 return PTR_ERR(realfile);
143 file->private_data = realfile;
148 static int ovl_release(struct inode *inode, struct file *file)
150 fput(file->private_data);
155 static loff_t ovl_llseek(struct file *file, loff_t offset, int whence)
157 struct inode *inode = file_inode(file);
159 const struct cred *old_cred;
163 * The two special cases below do not need to involve real fs,
164 * so we can optimizing concurrent callers.
167 if (whence == SEEK_CUR)
170 if (whence == SEEK_SET)
171 return vfs_setpos(file, 0, 0);
174 ret = ovl_real_fdget(file, &real);
179 * Overlay file f_pos is the master copy that is preserved
180 * through copy up and modified on read/write, but only real
181 * fs knows how to SEEK_HOLE/SEEK_DATA and real fs may impose
182 * limitations that are more strict than ->s_maxbytes for specific
183 * files, so we use the real file to perform seeks.
185 ovl_inode_lock(inode);
186 real.file->f_pos = file->f_pos;
188 old_cred = ovl_override_creds(inode->i_sb);
189 ret = vfs_llseek(real.file, offset, whence);
190 revert_creds(old_cred);
192 file->f_pos = real.file->f_pos;
193 ovl_inode_unlock(inode);
200 static void ovl_file_accessed(struct file *file)
202 struct inode *inode, *upperinode;
204 if (file->f_flags & O_NOATIME)
207 inode = file_inode(file);
208 upperinode = ovl_inode_upper(inode);
213 if ((!timespec64_equal(&inode->i_mtime, &upperinode->i_mtime) ||
214 !timespec64_equal(&inode->i_ctime, &upperinode->i_ctime))) {
215 inode->i_mtime = upperinode->i_mtime;
216 inode->i_ctime = upperinode->i_ctime;
219 touch_atime(&file->f_path);
222 static rwf_t ovl_iocb_to_rwf(struct kiocb *iocb)
224 int ifl = iocb->ki_flags;
227 if (ifl & IOCB_NOWAIT)
229 if (ifl & IOCB_HIPRI)
231 if (ifl & IOCB_DSYNC)
239 static void ovl_aio_cleanup_handler(struct ovl_aio_req *aio_req)
241 struct kiocb *iocb = &aio_req->iocb;
242 struct kiocb *orig_iocb = aio_req->orig_iocb;
244 if (iocb->ki_flags & IOCB_WRITE) {
245 struct inode *inode = file_inode(orig_iocb->ki_filp);
247 /* Actually acquired in ovl_write_iter() */
248 __sb_writers_acquired(file_inode(iocb->ki_filp)->i_sb,
250 file_end_write(iocb->ki_filp);
251 ovl_copyattr(ovl_inode_real(inode), inode);
254 orig_iocb->ki_pos = iocb->ki_pos;
256 kmem_cache_free(ovl_aio_request_cachep, aio_req);
259 static void ovl_aio_rw_complete(struct kiocb *iocb, long res, long res2)
261 struct ovl_aio_req *aio_req = container_of(iocb,
262 struct ovl_aio_req, iocb);
263 struct kiocb *orig_iocb = aio_req->orig_iocb;
265 ovl_aio_cleanup_handler(aio_req);
266 orig_iocb->ki_complete(orig_iocb, res, res2);
269 static ssize_t ovl_read_iter(struct kiocb *iocb, struct iov_iter *iter)
271 struct file *file = iocb->ki_filp;
273 const struct cred *old_cred;
276 if (!iov_iter_count(iter))
279 ret = ovl_real_fdget(file, &real);
283 old_cred = ovl_override_creds(file_inode(file)->i_sb);
284 if (is_sync_kiocb(iocb)) {
285 ret = vfs_iter_read(real.file, iter, &iocb->ki_pos,
286 ovl_iocb_to_rwf(iocb));
288 struct ovl_aio_req *aio_req;
291 aio_req = kmem_cache_zalloc(ovl_aio_request_cachep, GFP_KERNEL);
297 aio_req->orig_iocb = iocb;
298 kiocb_clone(&aio_req->iocb, iocb, real.file);
299 aio_req->iocb.ki_complete = ovl_aio_rw_complete;
300 ret = vfs_iocb_iter_read(real.file, &aio_req->iocb, iter);
301 if (ret != -EIOCBQUEUED)
302 ovl_aio_cleanup_handler(aio_req);
305 revert_creds(old_cred);
306 ovl_file_accessed(file);
313 static ssize_t ovl_write_iter(struct kiocb *iocb, struct iov_iter *iter)
315 struct file *file = iocb->ki_filp;
316 struct inode *inode = file_inode(file);
318 const struct cred *old_cred;
321 if (!iov_iter_count(iter))
326 ovl_copyattr(ovl_inode_real(inode), inode);
327 ret = file_remove_privs(file);
331 ret = ovl_real_fdget(file, &real);
335 old_cred = ovl_override_creds(file_inode(file)->i_sb);
336 if (is_sync_kiocb(iocb)) {
337 file_start_write(real.file);
338 ret = vfs_iter_write(real.file, iter, &iocb->ki_pos,
339 ovl_iocb_to_rwf(iocb));
340 file_end_write(real.file);
342 ovl_copyattr(ovl_inode_real(inode), inode);
344 struct ovl_aio_req *aio_req;
347 aio_req = kmem_cache_zalloc(ovl_aio_request_cachep, GFP_KERNEL);
351 file_start_write(real.file);
352 /* Pacify lockdep, same trick as done in aio_write() */
353 __sb_writers_release(file_inode(real.file)->i_sb,
357 aio_req->orig_iocb = iocb;
358 kiocb_clone(&aio_req->iocb, iocb, real.file);
359 aio_req->iocb.ki_complete = ovl_aio_rw_complete;
360 ret = vfs_iocb_iter_write(real.file, &aio_req->iocb, iter);
361 if (ret != -EIOCBQUEUED)
362 ovl_aio_cleanup_handler(aio_req);
365 revert_creds(old_cred);
374 static ssize_t ovl_splice_read(struct file *in, loff_t *ppos,
375 struct pipe_inode_info *pipe, size_t len,
380 const struct cred *old_cred;
382 ret = ovl_real_fdget(in, &real);
386 old_cred = ovl_override_creds(file_inode(in)->i_sb);
387 ret = generic_file_splice_read(real.file, ppos, pipe, len, flags);
388 revert_creds(old_cred);
390 ovl_file_accessed(in);
396 ovl_splice_write(struct pipe_inode_info *pipe, struct file *out,
397 loff_t *ppos, size_t len, unsigned int flags)
400 const struct cred *old_cred;
403 ret = ovl_real_fdget(out, &real);
407 old_cred = ovl_override_creds(file_inode(out)->i_sb);
408 ret = iter_file_splice_write(pipe, real.file, ppos, len, flags);
409 revert_creds(old_cred);
411 ovl_file_accessed(out);
416 static int ovl_fsync(struct file *file, loff_t start, loff_t end, int datasync)
419 const struct cred *old_cred;
422 ret = ovl_real_fdget_meta(file, &real, !datasync);
426 /* Don't sync lower file for fear of receiving EROFS error */
427 if (file_inode(real.file) == ovl_inode_upper(file_inode(file))) {
428 old_cred = ovl_override_creds(file_inode(file)->i_sb);
429 ret = vfs_fsync_range(real.file, start, end, datasync);
430 revert_creds(old_cred);
438 static int ovl_mmap(struct file *file, struct vm_area_struct *vma)
440 struct file *realfile = file->private_data;
441 const struct cred *old_cred;
444 if (!realfile->f_op->mmap)
447 if (WARN_ON(file != vma->vm_file))
450 vma->vm_file = get_file(realfile);
452 old_cred = ovl_override_creds(file_inode(file)->i_sb);
453 ret = call_mmap(vma->vm_file, vma);
454 revert_creds(old_cred);
457 /* Drop reference count from new vm_file value */
460 /* Drop reference count from previous vm_file value */
464 ovl_file_accessed(file);
469 static long ovl_fallocate(struct file *file, int mode, loff_t offset, loff_t len)
471 struct inode *inode = file_inode(file);
473 const struct cred *old_cred;
476 ret = ovl_real_fdget(file, &real);
480 old_cred = ovl_override_creds(file_inode(file)->i_sb);
481 ret = vfs_fallocate(real.file, mode, offset, len);
482 revert_creds(old_cred);
485 ovl_copyattr(ovl_inode_real(inode), inode);
492 static int ovl_fadvise(struct file *file, loff_t offset, loff_t len, int advice)
495 const struct cred *old_cred;
498 ret = ovl_real_fdget(file, &real);
502 old_cred = ovl_override_creds(file_inode(file)->i_sb);
503 ret = vfs_fadvise(real.file, offset, len, advice);
504 revert_creds(old_cred);
511 static long ovl_real_ioctl(struct file *file, unsigned int cmd,
515 const struct cred *old_cred;
518 ret = ovl_real_fdget(file, &real);
522 old_cred = ovl_override_creds(file_inode(file)->i_sb);
523 ret = vfs_ioctl(real.file, cmd, arg);
524 revert_creds(old_cred);
531 static long ovl_ioctl_set_flags(struct file *file, unsigned int cmd,
532 unsigned long arg, unsigned int iflags)
535 struct inode *inode = file_inode(file);
536 unsigned int old_iflags;
538 if (!inode_owner_or_capable(inode))
541 ret = mnt_want_write_file(file);
547 /* Check the capability before cred override */
549 old_iflags = READ_ONCE(inode->i_flags);
550 if (((iflags ^ old_iflags) & (S_APPEND | S_IMMUTABLE)) &&
551 !capable(CAP_LINUX_IMMUTABLE))
554 ret = ovl_maybe_copy_up(file_dentry(file), O_WRONLY);
558 ret = ovl_real_ioctl(file, cmd, arg);
560 ovl_copyflags(ovl_inode_real(inode), inode);
564 mnt_drop_write_file(file);
570 static unsigned int ovl_fsflags_to_iflags(unsigned int flags)
572 unsigned int iflags = 0;
574 if (flags & FS_SYNC_FL)
576 if (flags & FS_APPEND_FL)
578 if (flags & FS_IMMUTABLE_FL)
579 iflags |= S_IMMUTABLE;
580 if (flags & FS_NOATIME_FL)
586 static long ovl_ioctl_set_fsflags(struct file *file, unsigned int cmd,
591 if (get_user(flags, (int __user *) arg))
594 return ovl_ioctl_set_flags(file, cmd, arg,
595 ovl_fsflags_to_iflags(flags));
598 static unsigned int ovl_fsxflags_to_iflags(unsigned int xflags)
600 unsigned int iflags = 0;
602 if (xflags & FS_XFLAG_SYNC)
604 if (xflags & FS_XFLAG_APPEND)
606 if (xflags & FS_XFLAG_IMMUTABLE)
607 iflags |= S_IMMUTABLE;
608 if (xflags & FS_XFLAG_NOATIME)
614 static long ovl_ioctl_set_fsxflags(struct file *file, unsigned int cmd,
619 memset(&fa, 0, sizeof(fa));
620 if (copy_from_user(&fa, (void __user *) arg, sizeof(fa)))
623 return ovl_ioctl_set_flags(file, cmd, arg,
624 ovl_fsxflags_to_iflags(fa.fsx_xflags));
627 static long ovl_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
632 case FS_IOC_GETFLAGS:
633 case FS_IOC_FSGETXATTR:
634 ret = ovl_real_ioctl(file, cmd, arg);
637 case FS_IOC_SETFLAGS:
638 ret = ovl_ioctl_set_fsflags(file, cmd, arg);
641 case FS_IOC_FSSETXATTR:
642 ret = ovl_ioctl_set_fsxflags(file, cmd, arg);
652 static long ovl_compat_ioctl(struct file *file, unsigned int cmd,
656 case FS_IOC32_GETFLAGS:
657 cmd = FS_IOC_GETFLAGS;
660 case FS_IOC32_SETFLAGS:
661 cmd = FS_IOC_SETFLAGS;
668 return ovl_ioctl(file, cmd, arg);
677 static loff_t ovl_copyfile(struct file *file_in, loff_t pos_in,
678 struct file *file_out, loff_t pos_out,
679 loff_t len, unsigned int flags, enum ovl_copyop op)
681 struct inode *inode_out = file_inode(file_out);
682 struct fd real_in, real_out;
683 const struct cred *old_cred;
686 ret = ovl_real_fdget(file_out, &real_out);
690 ret = ovl_real_fdget(file_in, &real_in);
696 old_cred = ovl_override_creds(file_inode(file_out)->i_sb);
699 ret = vfs_copy_file_range(real_in.file, pos_in,
700 real_out.file, pos_out, len, flags);
704 ret = vfs_clone_file_range(real_in.file, pos_in,
705 real_out.file, pos_out, len, flags);
709 ret = vfs_dedupe_file_range_one(real_in.file, pos_in,
710 real_out.file, pos_out, len,
714 revert_creds(old_cred);
717 ovl_copyattr(ovl_inode_real(inode_out), inode_out);
725 static ssize_t ovl_copy_file_range(struct file *file_in, loff_t pos_in,
726 struct file *file_out, loff_t pos_out,
727 size_t len, unsigned int flags)
729 return ovl_copyfile(file_in, pos_in, file_out, pos_out, len, flags,
733 static loff_t ovl_remap_file_range(struct file *file_in, loff_t pos_in,
734 struct file *file_out, loff_t pos_out,
735 loff_t len, unsigned int remap_flags)
739 if (remap_flags & ~(REMAP_FILE_DEDUP | REMAP_FILE_ADVISORY))
742 if (remap_flags & REMAP_FILE_DEDUP)
748 * Don't copy up because of a dedupe request, this wouldn't make sense
749 * most of the time (data would be duplicated instead of deduplicated).
751 if (op == OVL_DEDUPE &&
752 (!ovl_inode_upper(file_inode(file_in)) ||
753 !ovl_inode_upper(file_inode(file_out))))
756 return ovl_copyfile(file_in, pos_in, file_out, pos_out, len,
760 const struct file_operations ovl_file_operations = {
762 .release = ovl_release,
763 .llseek = ovl_llseek,
764 .read_iter = ovl_read_iter,
765 .write_iter = ovl_write_iter,
768 .fallocate = ovl_fallocate,
769 .fadvise = ovl_fadvise,
770 .unlocked_ioctl = ovl_ioctl,
771 .compat_ioctl = ovl_compat_ioctl,
772 .splice_read = ovl_splice_read,
773 .splice_write = ovl_splice_write,
775 .copy_file_range = ovl_copy_file_range,
776 .remap_file_range = ovl_remap_file_range,
779 int __init ovl_aio_request_cache_init(void)
781 ovl_aio_request_cachep = kmem_cache_create("ovl_aio_req",
782 sizeof(struct ovl_aio_req),
783 0, SLAB_HWCACHE_ALIGN, NULL);
784 if (!ovl_aio_request_cachep)
790 void ovl_aio_request_cache_destroy(void)
792 kmem_cache_destroy(ovl_aio_request_cachep);