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>
13 #include <linux/security.h>
16 #include "overlayfs.h"
21 struct kiocb *orig_iocb;
24 static struct kmem_cache *ovl_aio_request_cachep;
26 static char ovl_whatisit(struct inode *inode, struct inode *realinode)
28 if (realinode != ovl_inode_upper(inode))
30 if (ovl_has_upperdata(inode))
36 /* No atime modification on underlying */
37 #define OVL_OPEN_FLAGS (O_NOATIME)
39 static struct file *ovl_open_realfile(const struct file *file,
40 const struct path *realpath)
42 struct inode *realinode = d_inode(realpath->dentry);
43 struct inode *inode = file_inode(file);
44 struct mnt_idmap *real_idmap;
45 struct file *realfile;
46 const struct cred *old_cred;
47 int flags = file->f_flags | OVL_OPEN_FLAGS;
48 int acc_mode = ACC_MODE(flags);
52 acc_mode |= MAY_APPEND;
54 old_cred = ovl_override_creds(inode->i_sb);
55 real_idmap = mnt_idmap(realpath->mnt);
56 err = inode_permission(real_idmap, realinode, MAY_OPEN | acc_mode);
58 realfile = ERR_PTR(err);
60 if (!inode_owner_or_capable(real_idmap, realinode))
63 realfile = backing_file_open(&file->f_path, flags, realpath,
66 revert_creds(old_cred);
68 pr_debug("open(%p[%pD2/%c], 0%o) -> (%p, 0%o)\n",
69 file, file, ovl_whatisit(inode, realinode), file->f_flags,
70 realfile, IS_ERR(realfile) ? 0 : realfile->f_flags);
75 #define OVL_SETFL_MASK (O_APPEND | O_NONBLOCK | O_NDELAY | O_DIRECT)
77 static int ovl_change_flags(struct file *file, unsigned int flags)
79 struct inode *inode = file_inode(file);
82 flags &= OVL_SETFL_MASK;
84 if (((flags ^ file->f_flags) & O_APPEND) && IS_APPEND(inode))
87 if ((flags & O_DIRECT) && !(file->f_mode & FMODE_CAN_ODIRECT))
90 if (file->f_op->check_flags) {
91 err = file->f_op->check_flags(flags);
96 spin_lock(&file->f_lock);
97 file->f_flags = (file->f_flags & ~OVL_SETFL_MASK) | flags;
98 file->f_iocb_flags = iocb_flags(file);
99 spin_unlock(&file->f_lock);
104 static int ovl_real_fdget_meta(const struct file *file, struct fd *real,
107 struct dentry *dentry = file_dentry(file);
108 struct path realpath;
112 real->file = file->private_data;
115 ovl_path_real(dentry, &realpath);
117 /* lazy lookup and verify of lowerdata */
118 err = ovl_verify_lowerdata(dentry);
122 ovl_path_realdata(dentry, &realpath);
124 if (!realpath.dentry)
127 /* Has it been copied up since we'd opened it? */
128 if (unlikely(file_inode(real->file) != d_inode(realpath.dentry))) {
129 real->flags = FDPUT_FPUT;
130 real->file = ovl_open_realfile(file, &realpath);
132 return PTR_ERR_OR_ZERO(real->file);
135 /* Did the flags change since open? */
136 if (unlikely((file->f_flags ^ real->file->f_flags) & ~OVL_OPEN_FLAGS))
137 return ovl_change_flags(real->file, file->f_flags);
142 static int ovl_real_fdget(const struct file *file, struct fd *real)
144 if (d_is_dir(file_dentry(file))) {
146 real->file = ovl_dir_real_file(file, false);
148 return PTR_ERR_OR_ZERO(real->file);
151 return ovl_real_fdget_meta(file, real, false);
154 static int ovl_open(struct inode *inode, struct file *file)
156 struct dentry *dentry = file_dentry(file);
157 struct file *realfile;
158 struct path realpath;
161 /* lazy lookup and verify lowerdata */
162 err = ovl_verify_lowerdata(dentry);
166 err = ovl_maybe_copy_up(dentry, file->f_flags);
170 /* No longer need these flags, so don't pass them on to underlying fs */
171 file->f_flags &= ~(O_CREAT | O_EXCL | O_NOCTTY | O_TRUNC);
173 ovl_path_realdata(dentry, &realpath);
174 if (!realpath.dentry)
177 realfile = ovl_open_realfile(file, &realpath);
178 if (IS_ERR(realfile))
179 return PTR_ERR(realfile);
181 file->private_data = realfile;
186 static int ovl_release(struct inode *inode, struct file *file)
188 fput(file->private_data);
193 static loff_t ovl_llseek(struct file *file, loff_t offset, int whence)
195 struct inode *inode = file_inode(file);
197 const struct cred *old_cred;
201 * The two special cases below do not need to involve real fs,
202 * so we can optimizing concurrent callers.
205 if (whence == SEEK_CUR)
208 if (whence == SEEK_SET)
209 return vfs_setpos(file, 0, 0);
212 ret = ovl_real_fdget(file, &real);
217 * Overlay file f_pos is the master copy that is preserved
218 * through copy up and modified on read/write, but only real
219 * fs knows how to SEEK_HOLE/SEEK_DATA and real fs may impose
220 * limitations that are more strict than ->s_maxbytes for specific
221 * files, so we use the real file to perform seeks.
223 ovl_inode_lock(inode);
224 real.file->f_pos = file->f_pos;
226 old_cred = ovl_override_creds(inode->i_sb);
227 ret = vfs_llseek(real.file, offset, whence);
228 revert_creds(old_cred);
230 file->f_pos = real.file->f_pos;
231 ovl_inode_unlock(inode);
238 static void ovl_file_accessed(struct file *file)
240 struct inode *inode, *upperinode;
241 struct timespec64 ctime, uctime;
242 struct timespec64 mtime, umtime;
244 if (file->f_flags & O_NOATIME)
247 inode = file_inode(file);
248 upperinode = ovl_inode_upper(inode);
253 ctime = inode_get_ctime(inode);
254 uctime = inode_get_ctime(upperinode);
255 mtime = inode_get_mtime(inode);
256 umtime = inode_get_mtime(upperinode);
257 if ((!timespec64_equal(&mtime, &umtime)) ||
258 !timespec64_equal(&ctime, &uctime)) {
259 inode_set_mtime_to_ts(inode, inode_get_mtime(upperinode));
260 inode_set_ctime_to_ts(inode, uctime);
263 touch_atime(&file->f_path);
266 static rwf_t ovl_iocb_to_rwf(int ifl)
270 if (ifl & IOCB_NOWAIT)
272 if (ifl & IOCB_HIPRI)
274 if (ifl & IOCB_DSYNC)
282 static inline void ovl_aio_put(struct ovl_aio_req *aio_req)
284 if (refcount_dec_and_test(&aio_req->ref)) {
285 fput(aio_req->iocb.ki_filp);
286 kmem_cache_free(ovl_aio_request_cachep, aio_req);
290 static void ovl_aio_cleanup_handler(struct ovl_aio_req *aio_req)
292 struct kiocb *iocb = &aio_req->iocb;
293 struct kiocb *orig_iocb = aio_req->orig_iocb;
295 if (iocb->ki_flags & IOCB_WRITE) {
296 struct inode *inode = file_inode(orig_iocb->ki_filp);
298 kiocb_end_write(iocb);
302 orig_iocb->ki_pos = iocb->ki_pos;
303 ovl_aio_put(aio_req);
306 static void ovl_aio_rw_complete(struct kiocb *iocb, long res)
308 struct ovl_aio_req *aio_req = container_of(iocb,
309 struct ovl_aio_req, iocb);
310 struct kiocb *orig_iocb = aio_req->orig_iocb;
312 ovl_aio_cleanup_handler(aio_req);
313 orig_iocb->ki_complete(orig_iocb, res);
316 static ssize_t ovl_read_iter(struct kiocb *iocb, struct iov_iter *iter)
318 struct file *file = iocb->ki_filp;
320 const struct cred *old_cred;
323 if (!iov_iter_count(iter))
326 ret = ovl_real_fdget(file, &real);
331 if (iocb->ki_flags & IOCB_DIRECT &&
332 !(real.file->f_mode & FMODE_CAN_ODIRECT))
335 old_cred = ovl_override_creds(file_inode(file)->i_sb);
336 if (is_sync_kiocb(iocb)) {
337 ret = vfs_iter_read(real.file, iter, &iocb->ki_pos,
338 ovl_iocb_to_rwf(iocb->ki_flags));
340 struct ovl_aio_req *aio_req;
343 aio_req = kmem_cache_zalloc(ovl_aio_request_cachep, GFP_KERNEL);
347 aio_req->orig_iocb = iocb;
348 kiocb_clone(&aio_req->iocb, iocb, get_file(real.file));
349 aio_req->iocb.ki_complete = ovl_aio_rw_complete;
350 refcount_set(&aio_req->ref, 2);
351 ret = vfs_iocb_iter_read(real.file, &aio_req->iocb, iter);
352 ovl_aio_put(aio_req);
353 if (ret != -EIOCBQUEUED)
354 ovl_aio_cleanup_handler(aio_req);
357 revert_creds(old_cred);
358 ovl_file_accessed(file);
365 static ssize_t ovl_write_iter(struct kiocb *iocb, struct iov_iter *iter)
367 struct file *file = iocb->ki_filp;
368 struct inode *inode = file_inode(file);
370 const struct cred *old_cred;
372 int ifl = iocb->ki_flags;
374 if (!iov_iter_count(iter))
380 ret = file_remove_privs(file);
384 ret = ovl_real_fdget(file, &real);
389 if (iocb->ki_flags & IOCB_DIRECT &&
390 !(real.file->f_mode & FMODE_CAN_ODIRECT))
393 if (!ovl_should_sync(OVL_FS(inode->i_sb)))
394 ifl &= ~(IOCB_DSYNC | IOCB_SYNC);
397 * Overlayfs doesn't support deferred completions, don't copy
398 * this property in case it is set by the issuer.
400 ifl &= ~IOCB_DIO_CALLER_COMP;
402 old_cred = ovl_override_creds(file_inode(file)->i_sb);
403 if (is_sync_kiocb(iocb)) {
404 file_start_write(real.file);
405 ret = vfs_iter_write(real.file, iter, &iocb->ki_pos,
406 ovl_iocb_to_rwf(ifl));
407 file_end_write(real.file);
411 struct ovl_aio_req *aio_req;
414 aio_req = kmem_cache_zalloc(ovl_aio_request_cachep, GFP_KERNEL);
418 aio_req->orig_iocb = iocb;
419 kiocb_clone(&aio_req->iocb, iocb, get_file(real.file));
420 aio_req->iocb.ki_flags = ifl;
421 aio_req->iocb.ki_complete = ovl_aio_rw_complete;
422 refcount_set(&aio_req->ref, 2);
423 kiocb_start_write(&aio_req->iocb);
424 ret = vfs_iocb_iter_write(real.file, &aio_req->iocb, iter);
425 ovl_aio_put(aio_req);
426 if (ret != -EIOCBQUEUED)
427 ovl_aio_cleanup_handler(aio_req);
430 revert_creds(old_cred);
440 static ssize_t ovl_splice_read(struct file *in, loff_t *ppos,
441 struct pipe_inode_info *pipe, size_t len,
444 const struct cred *old_cred;
448 ret = ovl_real_fdget(in, &real);
452 old_cred = ovl_override_creds(file_inode(in)->i_sb);
453 ret = vfs_splice_read(real.file, ppos, pipe, len, flags);
454 revert_creds(old_cred);
455 ovl_file_accessed(in);
462 * Calling iter_file_splice_write() directly from overlay's f_op may deadlock
463 * due to lock order inversion between pipe->mutex in iter_file_splice_write()
464 * and file_start_write(real.file) in ovl_write_iter().
466 * So do everything ovl_write_iter() does and call iter_file_splice_write() on
469 static ssize_t ovl_splice_write(struct pipe_inode_info *pipe, struct file *out,
470 loff_t *ppos, size_t len, unsigned int flags)
473 const struct cred *old_cred;
474 struct inode *inode = file_inode(out);
480 ret = file_remove_privs(out);
484 ret = ovl_real_fdget(out, &real);
488 old_cred = ovl_override_creds(inode->i_sb);
489 file_start_write(real.file);
491 ret = iter_file_splice_write(pipe, real.file, ppos, len, flags);
493 file_end_write(real.file);
496 revert_creds(old_cred);
505 static int ovl_fsync(struct file *file, loff_t start, loff_t end, int datasync)
508 const struct cred *old_cred;
511 ret = ovl_sync_status(OVL_FS(file_inode(file)->i_sb));
515 ret = ovl_real_fdget_meta(file, &real, !datasync);
519 /* Don't sync lower file for fear of receiving EROFS error */
520 if (file_inode(real.file) == ovl_inode_upper(file_inode(file))) {
521 old_cred = ovl_override_creds(file_inode(file)->i_sb);
522 ret = vfs_fsync_range(real.file, start, end, datasync);
523 revert_creds(old_cred);
531 static int ovl_mmap(struct file *file, struct vm_area_struct *vma)
533 struct file *realfile = file->private_data;
534 const struct cred *old_cred;
537 if (!realfile->f_op->mmap)
540 if (WARN_ON(file != vma->vm_file))
543 vma_set_file(vma, realfile);
545 old_cred = ovl_override_creds(file_inode(file)->i_sb);
546 ret = call_mmap(vma->vm_file, vma);
547 revert_creds(old_cred);
548 ovl_file_accessed(file);
553 static long ovl_fallocate(struct file *file, int mode, loff_t offset, loff_t len)
555 struct inode *inode = file_inode(file);
557 const struct cred *old_cred;
563 ret = file_remove_privs(file);
567 ret = ovl_real_fdget(file, &real);
571 old_cred = ovl_override_creds(file_inode(file)->i_sb);
572 ret = vfs_fallocate(real.file, mode, offset, len);
573 revert_creds(old_cred);
586 static int ovl_fadvise(struct file *file, loff_t offset, loff_t len, int advice)
589 const struct cred *old_cred;
592 ret = ovl_real_fdget(file, &real);
596 old_cred = ovl_override_creds(file_inode(file)->i_sb);
597 ret = vfs_fadvise(real.file, offset, len, advice);
598 revert_creds(old_cred);
611 static loff_t ovl_copyfile(struct file *file_in, loff_t pos_in,
612 struct file *file_out, loff_t pos_out,
613 loff_t len, unsigned int flags, enum ovl_copyop op)
615 struct inode *inode_out = file_inode(file_out);
616 struct fd real_in, real_out;
617 const struct cred *old_cred;
620 inode_lock(inode_out);
621 if (op != OVL_DEDUPE) {
623 ovl_copyattr(inode_out);
624 ret = file_remove_privs(file_out);
629 ret = ovl_real_fdget(file_out, &real_out);
633 ret = ovl_real_fdget(file_in, &real_in);
639 old_cred = ovl_override_creds(file_inode(file_out)->i_sb);
642 ret = vfs_copy_file_range(real_in.file, pos_in,
643 real_out.file, pos_out, len, flags);
647 ret = vfs_clone_file_range(real_in.file, pos_in,
648 real_out.file, pos_out, len, flags);
652 ret = vfs_dedupe_file_range_one(real_in.file, pos_in,
653 real_out.file, pos_out, len,
657 revert_creds(old_cred);
660 ovl_copyattr(inode_out);
666 inode_unlock(inode_out);
671 static ssize_t ovl_copy_file_range(struct file *file_in, loff_t pos_in,
672 struct file *file_out, loff_t pos_out,
673 size_t len, unsigned int flags)
675 return ovl_copyfile(file_in, pos_in, file_out, pos_out, len, flags,
679 static loff_t ovl_remap_file_range(struct file *file_in, loff_t pos_in,
680 struct file *file_out, loff_t pos_out,
681 loff_t len, unsigned int remap_flags)
685 if (remap_flags & ~(REMAP_FILE_DEDUP | REMAP_FILE_ADVISORY))
688 if (remap_flags & REMAP_FILE_DEDUP)
694 * Don't copy up because of a dedupe request, this wouldn't make sense
695 * most of the time (data would be duplicated instead of deduplicated).
697 if (op == OVL_DEDUPE &&
698 (!ovl_inode_upper(file_inode(file_in)) ||
699 !ovl_inode_upper(file_inode(file_out))))
702 return ovl_copyfile(file_in, pos_in, file_out, pos_out, len,
706 static int ovl_flush(struct file *file, fl_owner_t id)
709 const struct cred *old_cred;
712 err = ovl_real_fdget(file, &real);
716 if (real.file->f_op->flush) {
717 old_cred = ovl_override_creds(file_inode(file)->i_sb);
718 err = real.file->f_op->flush(real.file, id);
719 revert_creds(old_cred);
726 const struct file_operations ovl_file_operations = {
728 .release = ovl_release,
729 .llseek = ovl_llseek,
730 .read_iter = ovl_read_iter,
731 .write_iter = ovl_write_iter,
734 .fallocate = ovl_fallocate,
735 .fadvise = ovl_fadvise,
737 .splice_read = ovl_splice_read,
738 .splice_write = ovl_splice_write,
740 .copy_file_range = ovl_copy_file_range,
741 .remap_file_range = ovl_remap_file_range,
744 int __init ovl_aio_request_cache_init(void)
746 ovl_aio_request_cachep = kmem_cache_create("ovl_aio_req",
747 sizeof(struct ovl_aio_req),
748 0, SLAB_HWCACHE_ALIGN, NULL);
749 if (!ovl_aio_request_cachep)
755 void ovl_aio_request_cache_destroy(void)
757 kmem_cache_destroy(ovl_aio_request_cachep);