Line data Source code
1 : // SPDX-License-Identifier: GPL-2.0-only
2 : /*
3 : *
4 : * Copyright (C) 2011 Novell Inc.
5 : */
6 :
7 : #include <linux/fs.h>
8 : #include <linux/slab.h>
9 : #include <linux/cred.h>
10 : #include <linux/xattr.h>
11 : #include <linux/posix_acl.h>
12 : #include <linux/ratelimit.h>
13 : #include <linux/fiemap.h>
14 : #include "overlayfs.h"
15 :
16 :
17 0 : int ovl_setattr(struct user_namespace *mnt_userns, struct dentry *dentry,
18 : struct iattr *attr)
19 : {
20 0 : int err;
21 0 : bool full_copy_up = false;
22 0 : struct dentry *upperdentry;
23 0 : const struct cred *old_cred;
24 :
25 0 : err = setattr_prepare(&init_user_ns, dentry, attr);
26 0 : if (err)
27 : return err;
28 :
29 0 : err = ovl_want_write(dentry);
30 0 : if (err)
31 0 : goto out;
32 :
33 0 : if (attr->ia_valid & ATTR_SIZE) {
34 0 : struct inode *realinode = d_inode(ovl_dentry_real(dentry));
35 :
36 0 : err = -ETXTBSY;
37 0 : if (atomic_read(&realinode->i_writecount) < 0)
38 0 : goto out_drop_write;
39 :
40 : /* Truncate should trigger data copy up as well */
41 : full_copy_up = true;
42 : }
43 :
44 0 : if (!full_copy_up)
45 0 : err = ovl_copy_up(dentry);
46 : else
47 0 : err = ovl_copy_up_with_data(dentry);
48 0 : if (!err) {
49 0 : struct inode *winode = NULL;
50 :
51 0 : upperdentry = ovl_dentry_upper(dentry);
52 :
53 0 : if (attr->ia_valid & ATTR_SIZE) {
54 0 : winode = d_inode(upperdentry);
55 0 : err = get_write_access(winode);
56 0 : if (err)
57 0 : goto out_drop_write;
58 : }
59 :
60 0 : if (attr->ia_valid & (ATTR_KILL_SUID|ATTR_KILL_SGID))
61 0 : attr->ia_valid &= ~ATTR_MODE;
62 :
63 : /*
64 : * We might have to translate ovl file into real file object
65 : * once use cases emerge. For now, simply don't let underlying
66 : * filesystem rely on attr->ia_file
67 : */
68 0 : attr->ia_valid &= ~ATTR_FILE;
69 :
70 : /*
71 : * If open(O_TRUNC) is done, VFS calls ->setattr with ATTR_OPEN
72 : * set. Overlayfs does not pass O_TRUNC flag to underlying
73 : * filesystem during open -> do not pass ATTR_OPEN. This
74 : * disables optimization in fuse which assumes open(O_TRUNC)
75 : * already set file size to 0. But we never passed O_TRUNC to
76 : * fuse. So by clearing ATTR_OPEN, fuse will be forced to send
77 : * setattr request to server.
78 : */
79 0 : attr->ia_valid &= ~ATTR_OPEN;
80 :
81 0 : inode_lock(upperdentry->d_inode);
82 0 : old_cred = ovl_override_creds(dentry->d_sb);
83 0 : err = notify_change(&init_user_ns, upperdentry, attr, NULL);
84 0 : revert_creds(old_cred);
85 0 : if (!err)
86 0 : ovl_copyattr(upperdentry->d_inode, dentry->d_inode);
87 0 : inode_unlock(upperdentry->d_inode);
88 :
89 0 : if (winode)
90 0 : put_write_access(winode);
91 : }
92 0 : out_drop_write:
93 0 : ovl_drop_write(dentry);
94 : out:
95 : return err;
96 : }
97 :
98 0 : static int ovl_map_dev_ino(struct dentry *dentry, struct kstat *stat, int fsid)
99 : {
100 0 : bool samefs = ovl_same_fs(dentry->d_sb);
101 0 : unsigned int xinobits = ovl_xino_bits(dentry->d_sb);
102 0 : unsigned int xinoshift = 64 - xinobits;
103 :
104 0 : if (samefs) {
105 : /*
106 : * When all layers are on the same fs, all real inode
107 : * number are unique, so we use the overlay st_dev,
108 : * which is friendly to du -x.
109 : */
110 0 : stat->dev = dentry->d_sb->s_dev;
111 0 : return 0;
112 0 : } else if (xinobits) {
113 : /*
114 : * All inode numbers of underlying fs should not be using the
115 : * high xinobits, so we use high xinobits to partition the
116 : * overlay st_ino address space. The high bits holds the fsid
117 : * (upper fsid is 0). The lowest xinobit is reserved for mapping
118 : * the non-peresistent inode numbers range in case of overflow.
119 : * This way all overlay inode numbers are unique and use the
120 : * overlay st_dev.
121 : */
122 0 : if (likely(!(stat->ino >> xinoshift))) {
123 0 : stat->ino |= ((u64)fsid) << (xinoshift + 1);
124 0 : stat->dev = dentry->d_sb->s_dev;
125 0 : return 0;
126 0 : } else if (ovl_xino_warn(dentry->d_sb)) {
127 0 : pr_warn_ratelimited("inode number too big (%pd2, ino=%llu, xinobits=%d)\n",
128 : dentry, stat->ino, xinobits);
129 : }
130 : }
131 :
132 : /* The inode could not be mapped to a unified st_ino address space */
133 0 : if (S_ISDIR(dentry->d_inode->i_mode)) {
134 : /*
135 : * Always use the overlay st_dev for directories, so 'find
136 : * -xdev' will scan the entire overlay mount and won't cross the
137 : * overlay mount boundaries.
138 : *
139 : * If not all layers are on the same fs the pair {real st_ino;
140 : * overlay st_dev} is not unique, so use the non persistent
141 : * overlay st_ino for directories.
142 : */
143 0 : stat->dev = dentry->d_sb->s_dev;
144 0 : stat->ino = dentry->d_inode->i_ino;
145 : } else {
146 : /*
147 : * For non-samefs setup, if we cannot map all layers st_ino
148 : * to a unified address space, we need to make sure that st_dev
149 : * is unique per underlying fs, so we use the unique anonymous
150 : * bdev assigned to the underlying fs.
151 : */
152 0 : stat->dev = OVL_FS(dentry->d_sb)->fs[fsid].pseudo_dev;
153 : }
154 :
155 : return 0;
156 : }
157 :
158 0 : int ovl_getattr(struct user_namespace *mnt_userns, const struct path *path,
159 : struct kstat *stat, u32 request_mask, unsigned int flags)
160 : {
161 0 : struct dentry *dentry = path->dentry;
162 0 : enum ovl_path_type type;
163 0 : struct path realpath;
164 0 : const struct cred *old_cred;
165 0 : bool is_dir = S_ISDIR(dentry->d_inode->i_mode);
166 0 : int fsid = 0;
167 0 : int err;
168 0 : bool metacopy_blocks = false;
169 :
170 0 : metacopy_blocks = ovl_is_metacopy_dentry(dentry);
171 :
172 0 : type = ovl_path_real(dentry, &realpath);
173 0 : old_cred = ovl_override_creds(dentry->d_sb);
174 0 : err = vfs_getattr(&realpath, stat, request_mask, flags);
175 0 : if (err)
176 0 : goto out;
177 :
178 : /*
179 : * For non-dir or same fs, we use st_ino of the copy up origin.
180 : * This guaranties constant st_dev/st_ino across copy up.
181 : * With xino feature and non-samefs, we use st_ino of the copy up
182 : * origin masked with high bits that represent the layer id.
183 : *
184 : * If lower filesystem supports NFS file handles, this also guaranties
185 : * persistent st_ino across mount cycle.
186 : */
187 0 : if (!is_dir || ovl_same_dev(dentry->d_sb)) {
188 0 : if (!OVL_TYPE_UPPER(type)) {
189 0 : fsid = ovl_layer_lower(dentry)->fsid;
190 0 : } else if (OVL_TYPE_ORIGIN(type)) {
191 0 : struct kstat lowerstat;
192 0 : u32 lowermask = STATX_INO | STATX_BLOCKS |
193 : (!is_dir ? STATX_NLINK : 0);
194 :
195 0 : ovl_path_lower(dentry, &realpath);
196 0 : err = vfs_getattr(&realpath, &lowerstat,
197 : lowermask, flags);
198 0 : if (err)
199 0 : goto out;
200 :
201 : /*
202 : * Lower hardlinks may be broken on copy up to different
203 : * upper files, so we cannot use the lower origin st_ino
204 : * for those different files, even for the same fs case.
205 : *
206 : * Similarly, several redirected dirs can point to the
207 : * same dir on a lower layer. With the "verify_lower"
208 : * feature, we do not use the lower origin st_ino, if
209 : * we haven't verified that this redirect is unique.
210 : *
211 : * With inodes index enabled, it is safe to use st_ino
212 : * of an indexed origin. The index validates that the
213 : * upper hardlink is not broken and that a redirected
214 : * dir is the only redirect to that origin.
215 : */
216 0 : if (ovl_test_flag(OVL_INDEX, d_inode(dentry)) ||
217 0 : (!ovl_verify_lower(dentry->d_sb) &&
218 0 : (is_dir || lowerstat.nlink == 1))) {
219 0 : fsid = ovl_layer_lower(dentry)->fsid;
220 0 : stat->ino = lowerstat.ino;
221 : }
222 :
223 : /*
224 : * If we are querying a metacopy dentry and lower
225 : * dentry is data dentry, then use the blocks we
226 : * queried just now. We don't have to do additional
227 : * vfs_getattr(). If lower itself is metacopy, then
228 : * additional vfs_getattr() is unavoidable.
229 : */
230 0 : if (metacopy_blocks &&
231 0 : realpath.dentry == ovl_dentry_lowerdata(dentry)) {
232 0 : stat->blocks = lowerstat.blocks;
233 0 : metacopy_blocks = false;
234 : }
235 : }
236 :
237 0 : if (metacopy_blocks) {
238 : /*
239 : * If lower is not same as lowerdata or if there was
240 : * no origin on upper, we can end up here.
241 : */
242 0 : struct kstat lowerdatastat;
243 0 : u32 lowermask = STATX_BLOCKS;
244 :
245 0 : ovl_path_lowerdata(dentry, &realpath);
246 0 : err = vfs_getattr(&realpath, &lowerdatastat,
247 : lowermask, flags);
248 0 : if (err)
249 0 : goto out;
250 0 : stat->blocks = lowerdatastat.blocks;
251 : }
252 : }
253 :
254 0 : err = ovl_map_dev_ino(dentry, stat, fsid);
255 0 : if (err)
256 0 : goto out;
257 :
258 : /*
259 : * It's probably not worth it to count subdirs to get the
260 : * correct link count. nlink=1 seems to pacify 'find' and
261 : * other utilities.
262 : */
263 0 : if (is_dir && OVL_TYPE_MERGE(type))
264 0 : stat->nlink = 1;
265 :
266 : /*
267 : * Return the overlay inode nlinks for indexed upper inodes.
268 : * Overlay inode nlink counts the union of the upper hardlinks
269 : * and non-covered lower hardlinks. It does not include the upper
270 : * index hardlink.
271 : */
272 0 : if (!is_dir && ovl_test_flag(OVL_INDEX, d_inode(dentry)))
273 0 : stat->nlink = dentry->d_inode->i_nlink;
274 :
275 0 : out:
276 0 : revert_creds(old_cred);
277 :
278 0 : return err;
279 : }
280 :
281 175 : int ovl_permission(struct user_namespace *mnt_userns,
282 : struct inode *inode, int mask)
283 : {
284 175 : struct inode *upperinode = ovl_inode_upper(inode);
285 175 : struct inode *realinode = upperinode ?: ovl_inode_lower(inode);
286 175 : const struct cred *old_cred;
287 175 : int err;
288 :
289 : /* Careful in RCU walk mode */
290 175 : if (!realinode) {
291 0 : WARN_ON(!(mask & MAY_NOT_BLOCK));
292 : return -ECHILD;
293 : }
294 :
295 : /*
296 : * Check overlay inode with the creds of task and underlying inode
297 : * with creds of mounter
298 : */
299 175 : err = generic_permission(&init_user_ns, inode, mask);
300 175 : if (err)
301 : return err;
302 :
303 175 : old_cred = ovl_override_creds(inode->i_sb);
304 175 : if (!upperinode &&
305 10 : !special_file(realinode->i_mode) && mask & MAY_WRITE) {
306 3 : mask &= ~(MAY_WRITE | MAY_APPEND);
307 : /* Make sure mounter can read file for copy up later */
308 3 : mask |= MAY_READ;
309 : }
310 175 : err = inode_permission(&init_user_ns, realinode, mask);
311 175 : revert_creds(old_cred);
312 :
313 175 : return err;
314 : }
315 :
316 0 : static const char *ovl_get_link(struct dentry *dentry,
317 : struct inode *inode,
318 : struct delayed_call *done)
319 : {
320 0 : const struct cred *old_cred;
321 0 : const char *p;
322 :
323 0 : if (!dentry)
324 0 : return ERR_PTR(-ECHILD);
325 :
326 0 : old_cred = ovl_override_creds(dentry->d_sb);
327 0 : p = vfs_get_link(ovl_dentry_real(dentry), done);
328 0 : revert_creds(old_cred);
329 0 : return p;
330 : }
331 :
332 0 : bool ovl_is_private_xattr(struct super_block *sb, const char *name)
333 : {
334 0 : struct ovl_fs *ofs = sb->s_fs_info;
335 :
336 0 : if (ofs->config.userxattr)
337 0 : return strncmp(name, OVL_XATTR_USER_PREFIX,
338 0 : sizeof(OVL_XATTR_USER_PREFIX) - 1) == 0;
339 : else
340 0 : return strncmp(name, OVL_XATTR_TRUSTED_PREFIX,
341 0 : sizeof(OVL_XATTR_TRUSTED_PREFIX) - 1) == 0;
342 : }
343 :
344 0 : int ovl_xattr_set(struct dentry *dentry, struct inode *inode, const char *name,
345 : const void *value, size_t size, int flags)
346 : {
347 0 : int err;
348 0 : struct dentry *upperdentry = ovl_i_dentry_upper(inode);
349 0 : struct dentry *realdentry = upperdentry ?: ovl_dentry_lower(dentry);
350 0 : const struct cred *old_cred;
351 :
352 0 : err = ovl_want_write(dentry);
353 0 : if (err)
354 0 : goto out;
355 :
356 0 : if (!value && !upperdentry) {
357 0 : old_cred = ovl_override_creds(dentry->d_sb);
358 0 : err = vfs_getxattr(&init_user_ns, realdentry, name, NULL, 0);
359 0 : revert_creds(old_cred);
360 0 : if (err < 0)
361 0 : goto out_drop_write;
362 : }
363 :
364 0 : if (!upperdentry) {
365 0 : err = ovl_copy_up(dentry);
366 0 : if (err)
367 0 : goto out_drop_write;
368 :
369 0 : realdentry = ovl_dentry_upper(dentry);
370 : }
371 :
372 0 : old_cred = ovl_override_creds(dentry->d_sb);
373 0 : if (value)
374 0 : err = vfs_setxattr(&init_user_ns, realdentry, name, value, size,
375 : flags);
376 : else {
377 0 : WARN_ON(flags != XATTR_REPLACE);
378 0 : err = vfs_removexattr(&init_user_ns, realdentry, name);
379 : }
380 0 : revert_creds(old_cred);
381 :
382 : /* copy c/mtime */
383 0 : ovl_copyattr(d_inode(realdentry), inode);
384 :
385 0 : out_drop_write:
386 0 : ovl_drop_write(dentry);
387 0 : out:
388 0 : return err;
389 : }
390 :
391 0 : int ovl_xattr_get(struct dentry *dentry, struct inode *inode, const char *name,
392 : void *value, size_t size)
393 : {
394 0 : ssize_t res;
395 0 : const struct cred *old_cred;
396 0 : struct dentry *realdentry =
397 0 : ovl_i_dentry_upper(inode) ?: ovl_dentry_lower(dentry);
398 :
399 0 : old_cred = ovl_override_creds(dentry->d_sb);
400 0 : res = vfs_getxattr(&init_user_ns, realdentry, name, value, size);
401 0 : revert_creds(old_cred);
402 0 : return res;
403 : }
404 :
405 0 : static bool ovl_can_list(struct super_block *sb, const char *s)
406 : {
407 : /* Never list private (.overlay) */
408 0 : if (ovl_is_private_xattr(sb, s))
409 : return false;
410 :
411 : /* List all non-trusted xatts */
412 0 : if (strncmp(s, XATTR_TRUSTED_PREFIX, XATTR_TRUSTED_PREFIX_LEN) != 0)
413 : return true;
414 :
415 : /* list other trusted for superuser only */
416 0 : return ns_capable_noaudit(&init_user_ns, CAP_SYS_ADMIN);
417 : }
418 :
419 0 : ssize_t ovl_listxattr(struct dentry *dentry, char *list, size_t size)
420 : {
421 0 : struct dentry *realdentry = ovl_dentry_real(dentry);
422 0 : ssize_t res;
423 0 : size_t len;
424 0 : char *s;
425 0 : const struct cred *old_cred;
426 :
427 0 : old_cred = ovl_override_creds(dentry->d_sb);
428 0 : res = vfs_listxattr(realdentry, list, size);
429 0 : revert_creds(old_cred);
430 0 : if (res <= 0 || size == 0)
431 : return res;
432 :
433 : /* filter out private xattrs */
434 0 : for (s = list, len = res; len;) {
435 0 : size_t slen = strnlen(s, len) + 1;
436 :
437 : /* underlying fs providing us with an broken xattr list? */
438 0 : if (WARN_ON(slen > len))
439 : return -EIO;
440 :
441 0 : len -= slen;
442 0 : if (!ovl_can_list(dentry->d_sb, s)) {
443 0 : res -= slen;
444 0 : memmove(s, s + slen, len);
445 : } else {
446 0 : s += slen;
447 : }
448 : }
449 :
450 : return res;
451 : }
452 :
453 0 : struct posix_acl *ovl_get_acl(struct inode *inode, int type)
454 : {
455 0 : struct inode *realinode = ovl_inode_real(inode);
456 0 : const struct cred *old_cred;
457 0 : struct posix_acl *acl;
458 :
459 0 : if (!IS_ENABLED(CONFIG_FS_POSIX_ACL) || !IS_POSIXACL(realinode))
460 0 : return NULL;
461 :
462 : old_cred = ovl_override_creds(inode->i_sb);
463 : acl = get_acl(realinode, type);
464 : revert_creds(old_cred);
465 :
466 : return acl;
467 : }
468 :
469 0 : int ovl_update_time(struct inode *inode, struct timespec64 *ts, int flags)
470 : {
471 0 : if (flags & S_ATIME) {
472 0 : struct ovl_fs *ofs = inode->i_sb->s_fs_info;
473 0 : struct path upperpath = {
474 0 : .mnt = ovl_upper_mnt(ofs),
475 0 : .dentry = ovl_upperdentry_dereference(OVL_I(inode)),
476 : };
477 :
478 0 : if (upperpath.dentry) {
479 0 : touch_atime(&upperpath);
480 0 : inode->i_atime = d_inode(upperpath.dentry)->i_atime;
481 : }
482 : }
483 0 : return 0;
484 : }
485 :
486 0 : static int ovl_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
487 : u64 start, u64 len)
488 : {
489 0 : int err;
490 0 : struct inode *realinode = ovl_inode_realdata(inode);
491 0 : const struct cred *old_cred;
492 :
493 0 : if (!realinode->i_op->fiemap)
494 : return -EOPNOTSUPP;
495 :
496 0 : old_cred = ovl_override_creds(inode->i_sb);
497 0 : err = realinode->i_op->fiemap(realinode, fieinfo, start, len);
498 0 : revert_creds(old_cred);
499 :
500 0 : return err;
501 : }
502 :
503 : static const struct inode_operations ovl_file_inode_operations = {
504 : .setattr = ovl_setattr,
505 : .permission = ovl_permission,
506 : .getattr = ovl_getattr,
507 : .listxattr = ovl_listxattr,
508 : .get_acl = ovl_get_acl,
509 : .update_time = ovl_update_time,
510 : .fiemap = ovl_fiemap,
511 : };
512 :
513 : static const struct inode_operations ovl_symlink_inode_operations = {
514 : .setattr = ovl_setattr,
515 : .get_link = ovl_get_link,
516 : .getattr = ovl_getattr,
517 : .listxattr = ovl_listxattr,
518 : .update_time = ovl_update_time,
519 : };
520 :
521 : static const struct inode_operations ovl_special_inode_operations = {
522 : .setattr = ovl_setattr,
523 : .permission = ovl_permission,
524 : .getattr = ovl_getattr,
525 : .listxattr = ovl_listxattr,
526 : .get_acl = ovl_get_acl,
527 : .update_time = ovl_update_time,
528 : };
529 :
530 : static const struct address_space_operations ovl_aops = {
531 : /* For O_DIRECT dentry_open() checks f_mapping->a_ops->direct_IO */
532 : .direct_IO = noop_direct_IO,
533 : };
534 :
535 : /*
536 : * It is possible to stack overlayfs instance on top of another
537 : * overlayfs instance as lower layer. We need to annotate the
538 : * stackable i_mutex locks according to stack level of the super
539 : * block instance. An overlayfs instance can never be in stack
540 : * depth 0 (there is always a real fs below it). An overlayfs
541 : * inode lock will use the lockdep annotaion ovl_i_mutex_key[depth].
542 : *
543 : * For example, here is a snip from /proc/lockdep_chains after
544 : * dir_iterate of nested overlayfs:
545 : *
546 : * [...] &ovl_i_mutex_dir_key[depth] (stack_depth=2)
547 : * [...] &ovl_i_mutex_dir_key[depth]#2 (stack_depth=1)
548 : * [...] &type->i_mutex_dir_key (stack_depth=0)
549 : *
550 : * Locking order w.r.t ovl_want_write() is important for nested overlayfs.
551 : *
552 : * This chain is valid:
553 : * - inode->i_rwsem (inode_lock[2])
554 : * - upper_mnt->mnt_sb->s_writers (ovl_want_write[0])
555 : * - OVL_I(inode)->lock (ovl_inode_lock[2])
556 : * - OVL_I(lowerinode)->lock (ovl_inode_lock[1])
557 : *
558 : * And this chain is valid:
559 : * - inode->i_rwsem (inode_lock[2])
560 : * - OVL_I(inode)->lock (ovl_inode_lock[2])
561 : * - lowerinode->i_rwsem (inode_lock[1])
562 : * - OVL_I(lowerinode)->lock (ovl_inode_lock[1])
563 : *
564 : * But lowerinode->i_rwsem SHOULD NOT be acquired while ovl_want_write() is
565 : * held, because it is in reverse order of the non-nested case using the same
566 : * upper fs:
567 : * - inode->i_rwsem (inode_lock[1])
568 : * - upper_mnt->mnt_sb->s_writers (ovl_want_write[0])
569 : * - OVL_I(inode)->lock (ovl_inode_lock[1])
570 : */
571 : #define OVL_MAX_NESTING FILESYSTEM_MAX_STACK_DEPTH
572 :
573 24 : static inline void ovl_lockdep_annotate_inode_mutex_key(struct inode *inode)
574 : {
575 : #ifdef CONFIG_LOCKDEP
576 24 : static struct lock_class_key ovl_i_mutex_key[OVL_MAX_NESTING];
577 24 : static struct lock_class_key ovl_i_mutex_dir_key[OVL_MAX_NESTING];
578 24 : static struct lock_class_key ovl_i_lock_key[OVL_MAX_NESTING];
579 :
580 24 : int depth = inode->i_sb->s_stack_depth - 1;
581 :
582 24 : if (WARN_ON_ONCE(depth < 0 || depth >= OVL_MAX_NESTING))
583 0 : depth = 0;
584 :
585 24 : if (S_ISDIR(inode->i_mode))
586 8 : lockdep_set_class(&inode->i_rwsem, &ovl_i_mutex_dir_key[depth]);
587 : else
588 16 : lockdep_set_class(&inode->i_rwsem, &ovl_i_mutex_key[depth]);
589 :
590 24 : lockdep_set_class(&OVL_I(inode)->lock, &ovl_i_lock_key[depth]);
591 : #endif
592 24 : }
593 :
594 8 : static void ovl_next_ino(struct inode *inode)
595 : {
596 8 : struct ovl_fs *ofs = inode->i_sb->s_fs_info;
597 :
598 8 : inode->i_ino = atomic_long_inc_return(&ofs->last_ino);
599 8 : if (unlikely(!inode->i_ino))
600 0 : inode->i_ino = atomic_long_inc_return(&ofs->last_ino);
601 8 : }
602 :
603 24 : static void ovl_map_ino(struct inode *inode, unsigned long ino, int fsid)
604 : {
605 24 : int xinobits = ovl_xino_bits(inode->i_sb);
606 24 : unsigned int xinoshift = 64 - xinobits;
607 :
608 : /*
609 : * When d_ino is consistent with st_ino (samefs or i_ino has enough
610 : * bits to encode layer), set the same value used for st_ino to i_ino,
611 : * so inode number exposed via /proc/locks and a like will be
612 : * consistent with d_ino and st_ino values. An i_ino value inconsistent
613 : * with d_ino also causes nfsd readdirplus to fail.
614 : */
615 24 : inode->i_ino = ino;
616 24 : if (ovl_same_fs(inode->i_sb)) {
617 : return;
618 24 : } else if (xinobits && likely(!(ino >> xinoshift))) {
619 0 : inode->i_ino |= (unsigned long)fsid << (xinoshift + 1);
620 0 : return;
621 : }
622 :
623 : /*
624 : * For directory inodes on non-samefs with xino disabled or xino
625 : * overflow, we allocate a non-persistent inode number, to be used for
626 : * resolving st_ino collisions in ovl_map_dev_ino().
627 : *
628 : * To avoid ino collision with legitimate xino values from upper
629 : * layer (fsid 0), use the lowest xinobit to map the non
630 : * persistent inode numbers to the unified st_ino address space.
631 : */
632 24 : if (S_ISDIR(inode->i_mode)) {
633 8 : ovl_next_ino(inode);
634 8 : if (xinobits) {
635 0 : inode->i_ino &= ~0UL >> xinobits;
636 0 : inode->i_ino |= 1UL << xinoshift;
637 : }
638 : }
639 : }
640 :
641 24 : void ovl_inode_init(struct inode *inode, struct ovl_inode_params *oip,
642 : unsigned long ino, int fsid)
643 : {
644 24 : struct inode *realinode;
645 :
646 24 : if (oip->upperdentry)
647 16 : OVL_I(inode)->__upperdentry = oip->upperdentry;
648 24 : if (oip->lowerpath && oip->lowerpath->dentry)
649 12 : OVL_I(inode)->lower = igrab(d_inode(oip->lowerpath->dentry));
650 24 : if (oip->lowerdata)
651 0 : OVL_I(inode)->lowerdata = igrab(d_inode(oip->lowerdata));
652 :
653 24 : realinode = ovl_inode_real(inode);
654 24 : ovl_copyattr(realinode, inode);
655 24 : ovl_copyflags(realinode, inode);
656 24 : ovl_map_ino(inode, ino, fsid);
657 24 : }
658 :
659 24 : static void ovl_fill_inode(struct inode *inode, umode_t mode, dev_t rdev)
660 : {
661 24 : inode->i_mode = mode;
662 24 : inode->i_flags |= S_NOCMTIME;
663 : #ifdef CONFIG_FS_POSIX_ACL
664 : inode->i_acl = inode->i_default_acl = ACL_DONT_CACHE;
665 : #endif
666 :
667 24 : ovl_lockdep_annotate_inode_mutex_key(inode);
668 :
669 24 : switch (mode & S_IFMT) {
670 16 : case S_IFREG:
671 16 : inode->i_op = &ovl_file_inode_operations;
672 16 : inode->i_fop = &ovl_file_operations;
673 16 : inode->i_mapping->a_ops = &ovl_aops;
674 16 : break;
675 :
676 8 : case S_IFDIR:
677 8 : inode->i_op = &ovl_dir_inode_operations;
678 8 : inode->i_fop = &ovl_dir_operations;
679 8 : break;
680 :
681 0 : case S_IFLNK:
682 0 : inode->i_op = &ovl_symlink_inode_operations;
683 0 : break;
684 :
685 0 : default:
686 0 : inode->i_op = &ovl_special_inode_operations;
687 0 : init_special_inode(inode, mode, rdev);
688 0 : break;
689 : }
690 24 : }
691 :
692 : /*
693 : * With inodes index enabled, an overlay inode nlink counts the union of upper
694 : * hardlinks and non-covered lower hardlinks. During the lifetime of a non-pure
695 : * upper inode, the following nlink modifying operations can happen:
696 : *
697 : * 1. Lower hardlink copy up
698 : * 2. Upper hardlink created, unlinked or renamed over
699 : * 3. Lower hardlink whiteout or renamed over
700 : *
701 : * For the first, copy up case, the union nlink does not change, whether the
702 : * operation succeeds or fails, but the upper inode nlink may change.
703 : * Therefore, before copy up, we store the union nlink value relative to the
704 : * lower inode nlink in the index inode xattr .overlay.nlink.
705 : *
706 : * For the second, upper hardlink case, the union nlink should be incremented
707 : * or decremented IFF the operation succeeds, aligned with nlink change of the
708 : * upper inode. Therefore, before link/unlink/rename, we store the union nlink
709 : * value relative to the upper inode nlink in the index inode.
710 : *
711 : * For the last, lower cover up case, we simplify things by preceding the
712 : * whiteout or cover up with copy up. This makes sure that there is an index
713 : * upper inode where the nlink xattr can be stored before the copied up upper
714 : * entry is unlink.
715 : */
716 : #define OVL_NLINK_ADD_UPPER (1 << 0)
717 :
718 : /*
719 : * On-disk format for indexed nlink:
720 : *
721 : * nlink relative to the upper inode - "U[+-]NUM"
722 : * nlink relative to the lower inode - "L[+-]NUM"
723 : */
724 :
725 0 : static int ovl_set_nlink_common(struct dentry *dentry,
726 : struct dentry *realdentry, const char *format)
727 : {
728 0 : struct inode *inode = d_inode(dentry);
729 0 : struct inode *realinode = d_inode(realdentry);
730 0 : char buf[13];
731 0 : int len;
732 :
733 0 : len = snprintf(buf, sizeof(buf), format,
734 0 : (int) (inode->i_nlink - realinode->i_nlink));
735 :
736 0 : if (WARN_ON(len >= sizeof(buf)))
737 : return -EIO;
738 :
739 0 : return ovl_do_setxattr(OVL_FS(inode->i_sb), ovl_dentry_upper(dentry),
740 : OVL_XATTR_NLINK, buf, len);
741 : }
742 :
743 0 : int ovl_set_nlink_upper(struct dentry *dentry)
744 : {
745 0 : return ovl_set_nlink_common(dentry, ovl_dentry_upper(dentry), "U%+i");
746 : }
747 :
748 0 : int ovl_set_nlink_lower(struct dentry *dentry)
749 : {
750 0 : return ovl_set_nlink_common(dentry, ovl_dentry_lower(dentry), "L%+i");
751 : }
752 :
753 16 : unsigned int ovl_get_nlink(struct ovl_fs *ofs, struct dentry *lowerdentry,
754 : struct dentry *upperdentry,
755 : unsigned int fallback)
756 : {
757 16 : int nlink_diff;
758 16 : int nlink;
759 16 : char buf[13];
760 16 : int err;
761 :
762 16 : if (!lowerdentry || !upperdentry || d_inode(lowerdentry)->i_nlink == 1)
763 : return fallback;
764 :
765 0 : err = ovl_do_getxattr(ofs, upperdentry, OVL_XATTR_NLINK,
766 : &buf, sizeof(buf) - 1);
767 0 : if (err < 0)
768 0 : goto fail;
769 :
770 0 : buf[err] = '\0';
771 0 : if ((buf[0] != 'L' && buf[0] != 'U') ||
772 0 : (buf[1] != '+' && buf[1] != '-'))
773 0 : goto fail;
774 :
775 0 : err = kstrtoint(buf + 1, 10, &nlink_diff);
776 0 : if (err < 0)
777 0 : goto fail;
778 :
779 0 : nlink = d_inode(buf[0] == 'L' ? lowerdentry : upperdentry)->i_nlink;
780 0 : nlink += nlink_diff;
781 :
782 0 : if (nlink <= 0)
783 0 : goto fail;
784 :
785 0 : return nlink;
786 :
787 0 : fail:
788 0 : pr_warn_ratelimited("failed to get index nlink (%pd2, err=%i)\n",
789 : upperdentry, err);
790 : return fallback;
791 : }
792 :
793 2 : struct inode *ovl_new_inode(struct super_block *sb, umode_t mode, dev_t rdev)
794 : {
795 2 : struct inode *inode;
796 :
797 2 : inode = new_inode(sb);
798 2 : if (inode)
799 2 : ovl_fill_inode(inode, mode, rdev);
800 :
801 2 : return inode;
802 : }
803 :
804 0 : static int ovl_inode_test(struct inode *inode, void *data)
805 : {
806 0 : return inode->i_private == data;
807 : }
808 :
809 30 : static int ovl_inode_set(struct inode *inode, void *data)
810 : {
811 30 : inode->i_private = data;
812 30 : return 0;
813 : }
814 :
815 0 : static bool ovl_verify_inode(struct inode *inode, struct dentry *lowerdentry,
816 : struct dentry *upperdentry, bool strict)
817 : {
818 : /*
819 : * For directories, @strict verify from lookup path performs consistency
820 : * checks, so NULL lower/upper in dentry must match NULL lower/upper in
821 : * inode. Non @strict verify from NFS handle decode path passes NULL for
822 : * 'unknown' lower/upper.
823 : */
824 0 : if (S_ISDIR(inode->i_mode) && strict) {
825 : /* Real lower dir moved to upper layer under us? */
826 0 : if (!lowerdentry && ovl_inode_lower(inode))
827 : return false;
828 :
829 : /* Lookup of an uncovered redirect origin? */
830 0 : if (!upperdentry && ovl_inode_upper(inode))
831 : return false;
832 : }
833 :
834 : /*
835 : * Allow non-NULL lower inode in ovl_inode even if lowerdentry is NULL.
836 : * This happens when finding a copied up overlay inode for a renamed
837 : * or hardlinked overlay dentry and lower dentry cannot be followed
838 : * by origin because lower fs does not support file handles.
839 : */
840 0 : if (lowerdentry && ovl_inode_lower(inode) != d_inode(lowerdentry))
841 : return false;
842 :
843 : /*
844 : * Allow non-NULL __upperdentry in inode even if upperdentry is NULL.
845 : * This happens when finding a lower alias for a copied up hard link.
846 : */
847 0 : if (upperdentry && ovl_inode_upper(inode) != d_inode(upperdentry))
848 0 : return false;
849 :
850 : return true;
851 : }
852 :
853 0 : struct inode *ovl_lookup_inode(struct super_block *sb, struct dentry *real,
854 : bool is_upper)
855 : {
856 0 : struct inode *inode, *key = d_inode(real);
857 :
858 0 : inode = ilookup5(sb, (unsigned long) key, ovl_inode_test, key);
859 0 : if (!inode)
860 : return NULL;
861 :
862 0 : if (!ovl_verify_inode(inode, is_upper ? NULL : real,
863 : is_upper ? real : NULL, false)) {
864 0 : iput(inode);
865 0 : return ERR_PTR(-ESTALE);
866 : }
867 :
868 : return inode;
869 : }
870 :
871 14 : bool ovl_lookup_trap_inode(struct super_block *sb, struct dentry *dir)
872 : {
873 14 : struct inode *key = d_inode(dir);
874 14 : struct inode *trap;
875 14 : bool res;
876 :
877 14 : trap = ilookup5(sb, (unsigned long) key, ovl_inode_test, key);
878 14 : if (!trap)
879 : return false;
880 :
881 0 : res = IS_DEADDIR(trap) && !ovl_inode_upper(trap) &&
882 0 : !ovl_inode_lower(trap);
883 :
884 0 : iput(trap);
885 0 : return res;
886 : }
887 :
888 : /*
889 : * Create an inode cache entry for layer root dir, that will intentionally
890 : * fail ovl_verify_inode(), so any lookup that will find some layer root
891 : * will fail.
892 : */
893 8 : struct inode *ovl_get_trap_inode(struct super_block *sb, struct dentry *dir)
894 : {
895 8 : struct inode *key = d_inode(dir);
896 8 : struct inode *trap;
897 :
898 8 : if (!d_is_dir(dir))
899 8 : return ERR_PTR(-ENOTDIR);
900 :
901 8 : trap = iget5_locked(sb, (unsigned long) key, ovl_inode_test,
902 : ovl_inode_set, key);
903 8 : if (!trap)
904 8 : return ERR_PTR(-ENOMEM);
905 :
906 8 : if (!(trap->i_state & I_NEW)) {
907 : /* Conflicting layer roots? */
908 0 : iput(trap);
909 0 : return ERR_PTR(-ELOOP);
910 : }
911 :
912 8 : trap->i_mode = S_IFDIR;
913 8 : trap->i_flags = S_DEAD;
914 8 : unlock_new_inode(trap);
915 :
916 8 : return trap;
917 : }
918 :
919 : /*
920 : * Does overlay inode need to be hashed by lower inode?
921 : */
922 22 : static bool ovl_hash_bylower(struct super_block *sb, struct dentry *upper,
923 : struct dentry *lower, bool index)
924 : {
925 22 : struct ovl_fs *ofs = sb->s_fs_info;
926 :
927 : /* No, if pure upper */
928 22 : if (!lower)
929 : return false;
930 :
931 : /* Yes, if already indexed */
932 10 : if (index)
933 : return true;
934 :
935 : /* Yes, if won't be copied up */
936 10 : if (!ovl_upper_mnt(ofs))
937 : return true;
938 :
939 : /* No, if lower hardlink is or will be broken on copy up */
940 10 : if ((upper || !ovl_indexdir(sb)) &&
941 10 : !d_is_dir(lower) && d_inode(lower)->i_nlink > 1)
942 : return false;
943 :
944 : /* No, if non-indexed upper with NFS export */
945 10 : if (sb->s_export_op && upper)
946 0 : return false;
947 :
948 : /* Otherwise, hash by lower inode for fsnotify */
949 : return true;
950 : }
951 :
952 22 : static struct inode *ovl_iget5(struct super_block *sb, struct inode *newinode,
953 : struct inode *key)
954 : {
955 0 : return newinode ? inode_insert5(newinode, (unsigned long) key,
956 22 : ovl_inode_test, ovl_inode_set, key) :
957 22 : iget5_locked(sb, (unsigned long) key,
958 : ovl_inode_test, ovl_inode_set, key);
959 : }
960 :
961 22 : struct inode *ovl_get_inode(struct super_block *sb,
962 : struct ovl_inode_params *oip)
963 : {
964 22 : struct ovl_fs *ofs = OVL_FS(sb);
965 22 : struct dentry *upperdentry = oip->upperdentry;
966 22 : struct ovl_path *lowerpath = oip->lowerpath;
967 22 : struct inode *realinode = upperdentry ? d_inode(upperdentry) : NULL;
968 22 : struct inode *inode;
969 22 : struct dentry *lowerdentry = lowerpath ? lowerpath->dentry : NULL;
970 44 : bool bylower = ovl_hash_bylower(sb, upperdentry, lowerdentry,
971 22 : oip->index);
972 22 : int fsid = bylower ? lowerpath->layer->fsid : 0;
973 22 : bool is_dir;
974 22 : unsigned long ino = 0;
975 22 : int err = oip->newinode ? -EEXIST : -ENOMEM;
976 :
977 22 : if (!realinode)
978 8 : realinode = d_inode(lowerdentry);
979 :
980 : /*
981 : * Copy up origin (lower) may exist for non-indexed upper, but we must
982 : * not use lower as hash key if this is a broken hardlink.
983 : */
984 22 : is_dir = S_ISDIR(realinode->i_mode);
985 22 : if (upperdentry || bylower) {
986 22 : struct inode *key = d_inode(bylower ? lowerdentry :
987 : upperdentry);
988 22 : unsigned int nlink = is_dir ? 1 : realinode->i_nlink;
989 :
990 22 : inode = ovl_iget5(sb, oip->newinode, key);
991 22 : if (!inode)
992 0 : goto out_err;
993 22 : if (!(inode->i_state & I_NEW)) {
994 : /*
995 : * Verify that the underlying files stored in the inode
996 : * match those in the dentry.
997 : */
998 0 : if (!ovl_verify_inode(inode, lowerdentry, upperdentry,
999 : true)) {
1000 0 : iput(inode);
1001 0 : err = -ESTALE;
1002 0 : goto out_err;
1003 : }
1004 :
1005 0 : dput(upperdentry);
1006 0 : kfree(oip->redirect);
1007 0 : goto out;
1008 : }
1009 :
1010 : /* Recalculate nlink for non-dir due to indexing */
1011 22 : if (!is_dir)
1012 16 : nlink = ovl_get_nlink(ofs, lowerdentry, upperdentry,
1013 : nlink);
1014 22 : set_nlink(inode, nlink);
1015 22 : ino = key->i_ino;
1016 : } else {
1017 : /* Lower hardlink that will be broken on copy up */
1018 0 : inode = new_inode(sb);
1019 0 : if (!inode) {
1020 0 : err = -ENOMEM;
1021 0 : goto out_err;
1022 : }
1023 0 : ino = realinode->i_ino;
1024 0 : fsid = lowerpath->layer->fsid;
1025 : }
1026 22 : ovl_fill_inode(inode, realinode->i_mode, realinode->i_rdev);
1027 22 : ovl_inode_init(inode, oip, ino, fsid);
1028 :
1029 36 : if (upperdentry && ovl_is_impuredir(sb, upperdentry))
1030 0 : ovl_set_flag(OVL_IMPURE, inode);
1031 :
1032 22 : if (oip->index)
1033 0 : ovl_set_flag(OVL_INDEX, inode);
1034 :
1035 22 : OVL_I(inode)->redirect = oip->redirect;
1036 :
1037 22 : if (bylower)
1038 10 : ovl_set_flag(OVL_CONST_INO, inode);
1039 :
1040 : /* Check for non-merge dir that may have whiteouts */
1041 22 : if (is_dir) {
1042 10 : if (((upperdentry && lowerdentry) || oip->numlower > 1) ||
1043 6 : ovl_check_origin_xattr(ofs, upperdentry ?: lowerdentry)) {
1044 2 : ovl_set_flag(OVL_WHITEOUTS, inode);
1045 : }
1046 : }
1047 :
1048 22 : if (inode->i_state & I_NEW)
1049 22 : unlock_new_inode(inode);
1050 0 : out:
1051 22 : return inode;
1052 :
1053 0 : out_err:
1054 0 : pr_warn_ratelimited("failed to get inode (%i)\n", err);
1055 0 : inode = ERR_PTR(err);
1056 0 : goto out;
1057 : }
|
