Commit | Line | Data |
---|---|---|
d1d04ef8 MS |
1 | /* |
2 | * Copyright (C) 2017 Red Hat, Inc. | |
3 | * | |
4 | * This program is free software; you can redistribute it and/or modify it | |
5 | * under the terms of the GNU General Public License version 2 as published by | |
6 | * the Free Software Foundation. | |
7 | */ | |
8 | ||
9 | #include <linux/cred.h> | |
10 | #include <linux/file.h> | |
dab5ca8f | 11 | #include <linux/mount.h> |
d1d04ef8 | 12 | #include <linux/xattr.h> |
16914e6f | 13 | #include <linux/uio.h> |
d1d04ef8 MS |
14 | #include "overlayfs.h" |
15 | ||
16 | static struct file *ovl_open_realfile(const struct file *file) | |
17 | { | |
18 | struct inode *inode = file_inode(file); | |
19 | struct inode *upperinode = ovl_inode_upper(inode); | |
20 | struct inode *realinode = upperinode ?: ovl_inode_lower(inode); | |
21 | struct file *realfile; | |
22 | const struct cred *old_cred; | |
23 | ||
24 | old_cred = ovl_override_creds(inode->i_sb); | |
25 | realfile = open_with_fake_path(&file->f_path, file->f_flags | O_NOATIME, | |
26 | realinode, current_cred()); | |
27 | revert_creds(old_cred); | |
28 | ||
29 | pr_debug("open(%p[%pD2/%c], 0%o) -> (%p, 0%o)\n", | |
30 | file, file, upperinode ? 'u' : 'l', file->f_flags, | |
31 | realfile, IS_ERR(realfile) ? 0 : realfile->f_flags); | |
32 | ||
33 | return realfile; | |
34 | } | |
35 | ||
2ef66b8a MS |
36 | #define OVL_SETFL_MASK (O_APPEND | O_NONBLOCK | O_NDELAY | O_DIRECT) |
37 | ||
38 | static int ovl_change_flags(struct file *file, unsigned int flags) | |
39 | { | |
40 | struct inode *inode = file_inode(file); | |
41 | int err; | |
42 | ||
43 | /* No atime modificaton on underlying */ | |
44 | flags |= O_NOATIME; | |
45 | ||
46 | /* If some flag changed that cannot be changed then something's amiss */ | |
47 | if (WARN_ON((file->f_flags ^ flags) & ~OVL_SETFL_MASK)) | |
48 | return -EIO; | |
49 | ||
50 | flags &= OVL_SETFL_MASK; | |
51 | ||
52 | if (((flags ^ file->f_flags) & O_APPEND) && IS_APPEND(inode)) | |
53 | return -EPERM; | |
54 | ||
55 | if (flags & O_DIRECT) { | |
56 | if (!file->f_mapping->a_ops || | |
57 | !file->f_mapping->a_ops->direct_IO) | |
58 | return -EINVAL; | |
59 | } | |
60 | ||
61 | if (file->f_op->check_flags) { | |
62 | err = file->f_op->check_flags(flags); | |
63 | if (err) | |
64 | return err; | |
65 | } | |
66 | ||
67 | spin_lock(&file->f_lock); | |
68 | file->f_flags = (file->f_flags & ~OVL_SETFL_MASK) | flags; | |
69 | spin_unlock(&file->f_lock); | |
70 | ||
71 | return 0; | |
72 | } | |
73 | ||
74 | static int ovl_real_fdget(const struct file *file, struct fd *real) | |
75 | { | |
76 | struct inode *inode = file_inode(file); | |
77 | ||
78 | real->flags = 0; | |
79 | real->file = file->private_data; | |
80 | ||
81 | /* Has it been copied up since we'd opened it? */ | |
82 | if (unlikely(file_inode(real->file) != ovl_inode_real(inode))) { | |
83 | real->flags = FDPUT_FPUT; | |
84 | real->file = ovl_open_realfile(file); | |
85 | ||
86 | return PTR_ERR_OR_ZERO(real->file); | |
87 | } | |
88 | ||
89 | /* Did the flags change since open? */ | |
90 | if (unlikely((file->f_flags ^ real->file->f_flags) & ~O_NOATIME)) | |
91 | return ovl_change_flags(real->file, file->f_flags); | |
92 | ||
93 | return 0; | |
94 | } | |
95 | ||
d1d04ef8 MS |
96 | static int ovl_open(struct inode *inode, struct file *file) |
97 | { | |
98 | struct dentry *dentry = file_dentry(file); | |
99 | struct file *realfile; | |
100 | int err; | |
101 | ||
102 | err = ovl_open_maybe_copy_up(dentry, file->f_flags); | |
103 | if (err) | |
104 | return err; | |
105 | ||
106 | /* No longer need these flags, so don't pass them on to underlying fs */ | |
107 | file->f_flags &= ~(O_CREAT | O_EXCL | O_NOCTTY | O_TRUNC); | |
108 | ||
109 | realfile = ovl_open_realfile(file); | |
110 | if (IS_ERR(realfile)) | |
111 | return PTR_ERR(realfile); | |
112 | ||
f7c72396 MS |
113 | /* For O_DIRECT dentry_open() checks f_mapping->a_ops->direct_IO */ |
114 | file->f_mapping = realfile->f_mapping; | |
115 | ||
d1d04ef8 MS |
116 | file->private_data = realfile; |
117 | ||
118 | return 0; | |
119 | } | |
120 | ||
121 | static int ovl_release(struct inode *inode, struct file *file) | |
122 | { | |
123 | fput(file->private_data); | |
124 | ||
125 | return 0; | |
126 | } | |
127 | ||
128 | static loff_t ovl_llseek(struct file *file, loff_t offset, int whence) | |
129 | { | |
130 | struct inode *realinode = ovl_inode_real(file_inode(file)); | |
131 | ||
132 | return generic_file_llseek_size(file, offset, whence, | |
133 | realinode->i_sb->s_maxbytes, | |
134 | i_size_read(realinode)); | |
135 | } | |
136 | ||
16914e6f MS |
137 | static void ovl_file_accessed(struct file *file) |
138 | { | |
139 | struct inode *inode, *upperinode; | |
140 | ||
141 | if (file->f_flags & O_NOATIME) | |
142 | return; | |
143 | ||
144 | inode = file_inode(file); | |
145 | upperinode = ovl_inode_upper(inode); | |
146 | ||
147 | if (!upperinode) | |
148 | return; | |
149 | ||
150 | if ((!timespec64_equal(&inode->i_mtime, &upperinode->i_mtime) || | |
151 | !timespec64_equal(&inode->i_ctime, &upperinode->i_ctime))) { | |
152 | inode->i_mtime = upperinode->i_mtime; | |
153 | inode->i_ctime = upperinode->i_ctime; | |
154 | } | |
155 | ||
156 | touch_atime(&file->f_path); | |
157 | } | |
158 | ||
159 | static rwf_t ovl_iocb_to_rwf(struct kiocb *iocb) | |
160 | { | |
161 | int ifl = iocb->ki_flags; | |
162 | rwf_t flags = 0; | |
163 | ||
164 | if (ifl & IOCB_NOWAIT) | |
165 | flags |= RWF_NOWAIT; | |
166 | if (ifl & IOCB_HIPRI) | |
167 | flags |= RWF_HIPRI; | |
168 | if (ifl & IOCB_DSYNC) | |
169 | flags |= RWF_DSYNC; | |
170 | if (ifl & IOCB_SYNC) | |
171 | flags |= RWF_SYNC; | |
172 | ||
173 | return flags; | |
174 | } | |
175 | ||
176 | static ssize_t ovl_read_iter(struct kiocb *iocb, struct iov_iter *iter) | |
177 | { | |
178 | struct file *file = iocb->ki_filp; | |
179 | struct fd real; | |
180 | const struct cred *old_cred; | |
181 | ssize_t ret; | |
182 | ||
183 | if (!iov_iter_count(iter)) | |
184 | return 0; | |
185 | ||
186 | ret = ovl_real_fdget(file, &real); | |
187 | if (ret) | |
188 | return ret; | |
189 | ||
190 | old_cred = ovl_override_creds(file_inode(file)->i_sb); | |
191 | ret = vfs_iter_read(real.file, iter, &iocb->ki_pos, | |
192 | ovl_iocb_to_rwf(iocb)); | |
193 | revert_creds(old_cred); | |
194 | ||
195 | ovl_file_accessed(file); | |
196 | ||
197 | fdput(real); | |
198 | ||
199 | return ret; | |
200 | } | |
201 | ||
2a92e07e MS |
202 | static ssize_t ovl_write_iter(struct kiocb *iocb, struct iov_iter *iter) |
203 | { | |
204 | struct file *file = iocb->ki_filp; | |
205 | struct inode *inode = file_inode(file); | |
206 | struct fd real; | |
207 | const struct cred *old_cred; | |
208 | ssize_t ret; | |
209 | ||
210 | if (!iov_iter_count(iter)) | |
211 | return 0; | |
212 | ||
213 | inode_lock(inode); | |
214 | /* Update mode */ | |
215 | ovl_copyattr(ovl_inode_real(inode), inode); | |
216 | ret = file_remove_privs(file); | |
217 | if (ret) | |
218 | goto out_unlock; | |
219 | ||
220 | ret = ovl_real_fdget(file, &real); | |
221 | if (ret) | |
222 | goto out_unlock; | |
223 | ||
224 | old_cred = ovl_override_creds(file_inode(file)->i_sb); | |
225 | ret = vfs_iter_write(real.file, iter, &iocb->ki_pos, | |
226 | ovl_iocb_to_rwf(iocb)); | |
227 | revert_creds(old_cred); | |
228 | ||
229 | /* Update size */ | |
230 | ovl_copyattr(ovl_inode_real(inode), inode); | |
231 | ||
232 | fdput(real); | |
233 | ||
234 | out_unlock: | |
235 | inode_unlock(inode); | |
236 | ||
237 | return ret; | |
238 | } | |
239 | ||
de30dfd6 MS |
240 | static int ovl_fsync(struct file *file, loff_t start, loff_t end, int datasync) |
241 | { | |
242 | struct fd real; | |
243 | const struct cred *old_cred; | |
244 | int ret; | |
245 | ||
246 | ret = ovl_real_fdget(file, &real); | |
247 | if (ret) | |
248 | return ret; | |
249 | ||
250 | /* Don't sync lower file for fear of receiving EROFS error */ | |
251 | if (file_inode(real.file) == ovl_inode_upper(file_inode(file))) { | |
252 | old_cred = ovl_override_creds(file_inode(file)->i_sb); | |
253 | ret = vfs_fsync_range(real.file, start, end, datasync); | |
254 | revert_creds(old_cred); | |
255 | } | |
256 | ||
257 | fdput(real); | |
258 | ||
259 | return ret; | |
260 | } | |
261 | ||
2f502839 MS |
262 | static int ovl_mmap(struct file *file, struct vm_area_struct *vma) |
263 | { | |
264 | struct file *realfile = file->private_data; | |
265 | const struct cred *old_cred; | |
266 | int ret; | |
267 | ||
268 | if (!realfile->f_op->mmap) | |
269 | return -ENODEV; | |
270 | ||
271 | if (WARN_ON(file != vma->vm_file)) | |
272 | return -EIO; | |
273 | ||
274 | vma->vm_file = get_file(realfile); | |
275 | ||
276 | old_cred = ovl_override_creds(file_inode(file)->i_sb); | |
277 | ret = call_mmap(vma->vm_file, vma); | |
278 | revert_creds(old_cred); | |
279 | ||
280 | if (ret) { | |
281 | /* Drop reference count from new vm_file value */ | |
282 | fput(realfile); | |
283 | } else { | |
284 | /* Drop reference count from previous vm_file value */ | |
285 | fput(file); | |
286 | } | |
287 | ||
288 | ovl_file_accessed(file); | |
289 | ||
290 | return ret; | |
291 | } | |
292 | ||
aab8848c MS |
293 | static long ovl_fallocate(struct file *file, int mode, loff_t offset, loff_t len) |
294 | { | |
295 | struct inode *inode = file_inode(file); | |
296 | struct fd real; | |
297 | const struct cred *old_cred; | |
298 | int ret; | |
299 | ||
300 | ret = ovl_real_fdget(file, &real); | |
301 | if (ret) | |
302 | return ret; | |
303 | ||
304 | old_cred = ovl_override_creds(file_inode(file)->i_sb); | |
305 | ret = vfs_fallocate(real.file, mode, offset, len); | |
306 | revert_creds(old_cred); | |
307 | ||
308 | /* Update size */ | |
309 | ovl_copyattr(ovl_inode_real(inode), inode); | |
310 | ||
311 | fdput(real); | |
312 | ||
313 | return ret; | |
314 | } | |
315 | ||
dab5ca8f MS |
316 | static long ovl_real_ioctl(struct file *file, unsigned int cmd, |
317 | unsigned long arg) | |
318 | { | |
319 | struct fd real; | |
320 | const struct cred *old_cred; | |
321 | long ret; | |
322 | ||
323 | ret = ovl_real_fdget(file, &real); | |
324 | if (ret) | |
325 | return ret; | |
326 | ||
327 | old_cred = ovl_override_creds(file_inode(file)->i_sb); | |
328 | ret = vfs_ioctl(real.file, cmd, arg); | |
329 | revert_creds(old_cred); | |
330 | ||
331 | fdput(real); | |
332 | ||
333 | return ret; | |
334 | } | |
335 | ||
336 | static long ovl_ioctl(struct file *file, unsigned int cmd, unsigned long arg) | |
337 | { | |
338 | long ret; | |
339 | struct inode *inode = file_inode(file); | |
340 | ||
341 | switch (cmd) { | |
342 | case FS_IOC_GETFLAGS: | |
343 | ret = ovl_real_ioctl(file, cmd, arg); | |
344 | break; | |
345 | ||
346 | case FS_IOC_SETFLAGS: | |
347 | if (!inode_owner_or_capable(inode)) | |
348 | return -EACCES; | |
349 | ||
350 | ret = mnt_want_write_file(file); | |
351 | if (ret) | |
352 | return ret; | |
353 | ||
354 | ret = ovl_copy_up(file_dentry(file)); | |
355 | if (!ret) { | |
356 | ret = ovl_real_ioctl(file, cmd, arg); | |
357 | ||
358 | inode_lock(inode); | |
359 | ovl_copyflags(ovl_inode_real(inode), inode); | |
360 | inode_unlock(inode); | |
361 | } | |
362 | ||
363 | mnt_drop_write_file(file); | |
364 | break; | |
365 | ||
366 | default: | |
367 | ret = -ENOTTY; | |
368 | } | |
369 | ||
370 | return ret; | |
371 | } | |
372 | ||
373 | static long ovl_compat_ioctl(struct file *file, unsigned int cmd, | |
374 | unsigned long arg) | |
375 | { | |
376 | switch (cmd) { | |
377 | case FS_IOC32_GETFLAGS: | |
378 | cmd = FS_IOC_GETFLAGS; | |
379 | break; | |
380 | ||
381 | case FS_IOC32_SETFLAGS: | |
382 | cmd = FS_IOC_SETFLAGS; | |
383 | break; | |
384 | ||
385 | default: | |
386 | return -ENOIOCTLCMD; | |
387 | } | |
388 | ||
389 | return ovl_ioctl(file, cmd, arg); | |
390 | } | |
391 | ||
8ede2055 MS |
392 | enum ovl_copyop { |
393 | OVL_COPY, | |
394 | OVL_CLONE, | |
395 | OVL_DEDUPE, | |
396 | }; | |
397 | ||
398 | static ssize_t ovl_copyfile(struct file *file_in, loff_t pos_in, | |
399 | struct file *file_out, loff_t pos_out, | |
400 | u64 len, unsigned int flags, enum ovl_copyop op) | |
401 | { | |
402 | struct inode *inode_out = file_inode(file_out); | |
403 | struct fd real_in, real_out; | |
404 | const struct cred *old_cred; | |
405 | ssize_t ret; | |
406 | ||
407 | ret = ovl_real_fdget(file_out, &real_out); | |
408 | if (ret) | |
409 | return ret; | |
410 | ||
411 | ret = ovl_real_fdget(file_in, &real_in); | |
412 | if (ret) { | |
413 | fdput(real_out); | |
414 | return ret; | |
415 | } | |
416 | ||
417 | old_cred = ovl_override_creds(file_inode(file_out)->i_sb); | |
418 | switch (op) { | |
419 | case OVL_COPY: | |
420 | ret = vfs_copy_file_range(real_in.file, pos_in, | |
421 | real_out.file, pos_out, len, flags); | |
422 | break; | |
423 | ||
424 | case OVL_CLONE: | |
425 | ret = vfs_clone_file_range(real_in.file, pos_in, | |
426 | real_out.file, pos_out, len); | |
427 | break; | |
428 | ||
429 | case OVL_DEDUPE: | |
430 | ret = vfs_dedupe_file_range_one(real_in.file, pos_in, | |
431 | real_out.file, pos_out, len); | |
432 | break; | |
433 | } | |
434 | revert_creds(old_cred); | |
435 | ||
436 | /* Update size */ | |
437 | ovl_copyattr(ovl_inode_real(inode_out), inode_out); | |
438 | ||
439 | fdput(real_in); | |
440 | fdput(real_out); | |
441 | ||
442 | return ret; | |
443 | } | |
444 | ||
445 | static ssize_t ovl_copy_file_range(struct file *file_in, loff_t pos_in, | |
446 | struct file *file_out, loff_t pos_out, | |
447 | size_t len, unsigned int flags) | |
448 | { | |
449 | return ovl_copyfile(file_in, pos_in, file_out, pos_out, len, flags, | |
450 | OVL_COPY); | |
451 | } | |
452 | ||
453 | static int ovl_clone_file_range(struct file *file_in, loff_t pos_in, | |
454 | struct file *file_out, loff_t pos_out, u64 len) | |
455 | { | |
456 | return ovl_copyfile(file_in, pos_in, file_out, pos_out, len, 0, | |
457 | OVL_CLONE); | |
458 | } | |
459 | ||
460 | static int ovl_dedupe_file_range(struct file *file_in, loff_t pos_in, | |
461 | struct file *file_out, loff_t pos_out, u64 len) | |
462 | { | |
463 | /* | |
464 | * Don't copy up because of a dedupe request, this wouldn't make sense | |
465 | * most of the time (data would be duplicated instead of deduplicated). | |
466 | */ | |
467 | if (!ovl_inode_upper(file_inode(file_in)) || | |
468 | !ovl_inode_upper(file_inode(file_out))) | |
469 | return -EPERM; | |
470 | ||
471 | return ovl_copyfile(file_in, pos_in, file_out, pos_out, len, 0, | |
472 | OVL_DEDUPE); | |
473 | } | |
474 | ||
d1d04ef8 MS |
475 | const struct file_operations ovl_file_operations = { |
476 | .open = ovl_open, | |
477 | .release = ovl_release, | |
478 | .llseek = ovl_llseek, | |
16914e6f | 479 | .read_iter = ovl_read_iter, |
2a92e07e | 480 | .write_iter = ovl_write_iter, |
de30dfd6 | 481 | .fsync = ovl_fsync, |
2f502839 | 482 | .mmap = ovl_mmap, |
aab8848c | 483 | .fallocate = ovl_fallocate, |
dab5ca8f MS |
484 | .unlocked_ioctl = ovl_ioctl, |
485 | .compat_ioctl = ovl_compat_ioctl, | |
8ede2055 MS |
486 | |
487 | .copy_file_range = ovl_copy_file_range, | |
488 | .clone_file_range = ovl_clone_file_range, | |
489 | .dedupe_file_range = ovl_dedupe_file_range, | |
d1d04ef8 | 490 | }; |