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/namei.h>
10#include <linux/file.h>
11#include <linux/xattr.h>
12#include <linux/rbtree.h>
13#include <linux/security.h>
14#include <linux/cred.h>
15#include <linux/ratelimit.h>
16#include <linux/overflow.h>
17#include "overlayfs.h"
18
19struct ovl_cache_entry {
20 unsigned int len;
21 unsigned int type;
22 u64 real_ino;
23 u64 ino;
24 struct list_head l_node;
25 struct rb_node node;
26 struct ovl_cache_entry *next_maybe_whiteout;
27 bool is_upper;
28 bool is_whiteout;
29 bool check_xwhiteout;
30 char name[];
31};
32
33struct ovl_dir_cache {
34 long refcount;
35 u64 version;
36 struct list_head entries;
37 struct rb_root root;
38};
39
40struct ovl_readdir_data {
41 struct dir_context ctx;
42 struct dentry *dentry;
43 bool is_lowest;
44 struct rb_root *root;
45 struct list_head *list;
46 struct list_head middle;
47 struct ovl_cache_entry *first_maybe_whiteout;
48 int count;
49 int err;
50 bool is_upper;
51 bool d_type_supported;
52 bool in_xwhiteouts_dir;
53};
54
55struct ovl_dir_file {
56 bool is_real;
57 bool is_upper;
58 struct ovl_dir_cache *cache;
59 struct list_head *cursor;
60 struct file *realfile;
61 struct file *upperfile;
62};
63
64static struct ovl_cache_entry *ovl_cache_entry_from_node(struct rb_node *n)
65{
66 return rb_entry(n, struct ovl_cache_entry, node);
67}
68
69static bool ovl_cache_entry_find_link(const char *name, int len,
70 struct rb_node ***link,
71 struct rb_node **parent)
72{
73 bool found = false;
74 struct rb_node **newp = *link;
75
76 while (!found && *newp) {
77 int cmp;
78 struct ovl_cache_entry *tmp;
79
80 *parent = *newp;
81 tmp = ovl_cache_entry_from_node(n: *newp);
82 cmp = strncmp(name, tmp->name, len);
83 if (cmp > 0)
84 newp = &tmp->node.rb_right;
85 else if (cmp < 0 || len < tmp->len)
86 newp = &tmp->node.rb_left;
87 else
88 found = true;
89 }
90 *link = newp;
91
92 return found;
93}
94
95static struct ovl_cache_entry *ovl_cache_entry_find(struct rb_root *root,
96 const char *name, int len)
97{
98 struct rb_node *node = root->rb_node;
99 int cmp;
100
101 while (node) {
102 struct ovl_cache_entry *p = ovl_cache_entry_from_node(n: node);
103
104 cmp = strncmp(name, p->name, len);
105 if (cmp > 0)
106 node = p->node.rb_right;
107 else if (cmp < 0 || len < p->len)
108 node = p->node.rb_left;
109 else
110 return p;
111 }
112
113 return NULL;
114}
115
116static bool ovl_calc_d_ino(struct ovl_readdir_data *rdd,
117 struct ovl_cache_entry *p)
118{
119 /* Don't care if not doing ovl_iter() */
120 if (!rdd->dentry)
121 return false;
122
123 /* Always recalc d_ino when remapping lower inode numbers */
124 if (ovl_xino_bits(ofs: OVL_FS(sb: rdd->dentry->d_sb)))
125 return true;
126
127 /* Always recalc d_ino for parent */
128 if (strcmp(p->name, "..") == 0)
129 return true;
130
131 /* If this is lower, then native d_ino will do */
132 if (!rdd->is_upper)
133 return false;
134
135 /*
136 * Recalc d_ino for '.' and for all entries if dir is impure (contains
137 * copied up entries)
138 */
139 if ((p->name[0] == '.' && p->len == 1) ||
140 ovl_test_flag(flag: OVL_IMPURE, inode: d_inode(dentry: rdd->dentry)))
141 return true;
142
143 return false;
144}
145
146static struct ovl_cache_entry *ovl_cache_entry_new(struct ovl_readdir_data *rdd,
147 const char *name, int len,
148 u64 ino, unsigned int d_type)
149{
150 struct ovl_cache_entry *p;
151
152 p = kmalloc(struct_size(p, name, len + 1), GFP_KERNEL);
153 if (!p)
154 return NULL;
155
156 memcpy(p->name, name, len);
157 p->name[len] = '\0';
158 p->len = len;
159 p->type = d_type;
160 p->real_ino = ino;
161 p->ino = ino;
162 /* Defer setting d_ino for upper entry to ovl_iterate() */
163 if (ovl_calc_d_ino(rdd, p))
164 p->ino = 0;
165 p->is_upper = rdd->is_upper;
166 p->is_whiteout = false;
167 /* Defer check for overlay.whiteout to ovl_iterate() */
168 p->check_xwhiteout = rdd->in_xwhiteouts_dir && d_type == DT_REG;
169
170 if (d_type == DT_CHR) {
171 p->next_maybe_whiteout = rdd->first_maybe_whiteout;
172 rdd->first_maybe_whiteout = p;
173 }
174 return p;
175}
176
177static bool ovl_cache_entry_add_rb(struct ovl_readdir_data *rdd,
178 const char *name, int len, u64 ino,
179 unsigned int d_type)
180{
181 struct rb_node **newp = &rdd->root->rb_node;
182 struct rb_node *parent = NULL;
183 struct ovl_cache_entry *p;
184
185 if (ovl_cache_entry_find_link(name, len, link: &newp, parent: &parent))
186 return true;
187
188 p = ovl_cache_entry_new(rdd, name, len, ino, d_type);
189 if (p == NULL) {
190 rdd->err = -ENOMEM;
191 return false;
192 }
193
194 list_add_tail(new: &p->l_node, head: rdd->list);
195 rb_link_node(node: &p->node, parent, rb_link: newp);
196 rb_insert_color(&p->node, rdd->root);
197
198 return true;
199}
200
201static bool ovl_fill_lowest(struct ovl_readdir_data *rdd,
202 const char *name, int namelen,
203 loff_t offset, u64 ino, unsigned int d_type)
204{
205 struct ovl_cache_entry *p;
206
207 p = ovl_cache_entry_find(root: rdd->root, name, len: namelen);
208 if (p) {
209 list_move_tail(list: &p->l_node, head: &rdd->middle);
210 } else {
211 p = ovl_cache_entry_new(rdd, name, len: namelen, ino, d_type);
212 if (p == NULL)
213 rdd->err = -ENOMEM;
214 else
215 list_add_tail(new: &p->l_node, head: &rdd->middle);
216 }
217
218 return rdd->err == 0;
219}
220
221void ovl_cache_free(struct list_head *list)
222{
223 struct ovl_cache_entry *p;
224 struct ovl_cache_entry *n;
225
226 list_for_each_entry_safe(p, n, list, l_node)
227 kfree(objp: p);
228
229 INIT_LIST_HEAD(list);
230}
231
232void ovl_dir_cache_free(struct inode *inode)
233{
234 struct ovl_dir_cache *cache = ovl_dir_cache(inode);
235
236 if (cache) {
237 ovl_cache_free(list: &cache->entries);
238 kfree(objp: cache);
239 }
240}
241
242static void ovl_cache_put(struct ovl_dir_file *od, struct inode *inode)
243{
244 struct ovl_dir_cache *cache = od->cache;
245
246 WARN_ON(cache->refcount <= 0);
247 cache->refcount--;
248 if (!cache->refcount) {
249 if (ovl_dir_cache(inode) == cache)
250 ovl_set_dir_cache(inode, NULL);
251
252 ovl_cache_free(list: &cache->entries);
253 kfree(objp: cache);
254 }
255}
256
257static bool ovl_fill_merge(struct dir_context *ctx, const char *name,
258 int namelen, loff_t offset, u64 ino,
259 unsigned int d_type)
260{
261 struct ovl_readdir_data *rdd =
262 container_of(ctx, struct ovl_readdir_data, ctx);
263
264 rdd->count++;
265 if (!rdd->is_lowest)
266 return ovl_cache_entry_add_rb(rdd, name, len: namelen, ino, d_type);
267 else
268 return ovl_fill_lowest(rdd, name, namelen, offset, ino, d_type);
269}
270
271static int ovl_check_whiteouts(const struct path *path, struct ovl_readdir_data *rdd)
272{
273 int err;
274 struct dentry *dentry, *dir = path->dentry;
275 const struct cred *old_cred;
276
277 old_cred = ovl_override_creds(sb: rdd->dentry->d_sb);
278
279 err = down_write_killable(sem: &dir->d_inode->i_rwsem);
280 if (!err) {
281 while (rdd->first_maybe_whiteout) {
282 struct ovl_cache_entry *p =
283 rdd->first_maybe_whiteout;
284 rdd->first_maybe_whiteout = p->next_maybe_whiteout;
285 dentry = lookup_one(mnt_idmap(mnt: path->mnt),
286 &QSTR_LEN(p->name, p->len), dir);
287 if (!IS_ERR(ptr: dentry)) {
288 p->is_whiteout = ovl_is_whiteout(dentry);
289 dput(dentry);
290 }
291 }
292 inode_unlock(inode: dir->d_inode);
293 }
294 ovl_revert_creds(old_cred);
295
296 return err;
297}
298
299static inline int ovl_dir_read(const struct path *realpath,
300 struct ovl_readdir_data *rdd)
301{
302 struct file *realfile;
303 int err;
304
305 realfile = ovl_path_open(path: realpath, O_RDONLY | O_LARGEFILE);
306 if (IS_ERR(ptr: realfile))
307 return PTR_ERR(ptr: realfile);
308
309 rdd->first_maybe_whiteout = NULL;
310 rdd->ctx.pos = 0;
311 do {
312 rdd->count = 0;
313 rdd->err = 0;
314 err = iterate_dir(realfile, &rdd->ctx);
315 if (err >= 0)
316 err = rdd->err;
317 } while (!err && rdd->count);
318
319 if (!err && rdd->first_maybe_whiteout && rdd->dentry)
320 err = ovl_check_whiteouts(path: realpath, rdd);
321
322 fput(realfile);
323
324 return err;
325}
326
327static void ovl_dir_reset(struct file *file)
328{
329 struct ovl_dir_file *od = file->private_data;
330 struct ovl_dir_cache *cache = od->cache;
331 struct inode *inode = file_inode(f: file);
332 bool is_real;
333
334 if (cache && ovl_inode_version_get(inode) != cache->version) {
335 ovl_cache_put(od, inode);
336 od->cache = NULL;
337 od->cursor = NULL;
338 }
339 is_real = ovl_dir_is_real(dir: inode);
340 if (od->is_real != is_real) {
341 /* is_real can only become false when dir is copied up */
342 if (WARN_ON(is_real))
343 return;
344 od->is_real = false;
345 }
346}
347
348static int ovl_dir_read_merged(struct dentry *dentry, struct list_head *list,
349 struct rb_root *root)
350{
351 int err;
352 struct path realpath;
353 struct ovl_readdir_data rdd = {
354 .ctx.actor = ovl_fill_merge,
355 .ctx.count = INT_MAX,
356 .dentry = dentry,
357 .list = list,
358 .root = root,
359 .is_lowest = false,
360 };
361 int idx, next;
362 const struct ovl_layer *layer;
363
364 for (idx = 0; idx != -1; idx = next) {
365 next = ovl_path_next(idx, dentry, path: &realpath, layer: &layer);
366 rdd.is_upper = ovl_dentry_upper(dentry) == realpath.dentry;
367 rdd.in_xwhiteouts_dir = layer->has_xwhiteouts &&
368 ovl_dentry_has_xwhiteouts(dentry);
369
370 if (next != -1) {
371 err = ovl_dir_read(realpath: &realpath, rdd: &rdd);
372 if (err)
373 break;
374 } else {
375 /*
376 * Insert lowest layer entries before upper ones, this
377 * allows offsets to be reasonably constant
378 */
379 list_add(new: &rdd.middle, head: rdd.list);
380 rdd.is_lowest = true;
381 err = ovl_dir_read(realpath: &realpath, rdd: &rdd);
382 list_del(entry: &rdd.middle);
383 }
384 }
385 return err;
386}
387
388static void ovl_seek_cursor(struct ovl_dir_file *od, loff_t pos)
389{
390 struct list_head *p;
391 loff_t off = 0;
392
393 list_for_each(p, &od->cache->entries) {
394 if (off >= pos)
395 break;
396 off++;
397 }
398 /* Cursor is safe since the cache is stable */
399 od->cursor = p;
400}
401
402static struct ovl_dir_cache *ovl_cache_get(struct dentry *dentry)
403{
404 int res;
405 struct ovl_dir_cache *cache;
406 struct inode *inode = d_inode(dentry);
407
408 cache = ovl_dir_cache(inode);
409 if (cache && ovl_inode_version_get(inode) == cache->version) {
410 WARN_ON(!cache->refcount);
411 cache->refcount++;
412 return cache;
413 }
414 ovl_set_dir_cache(inode: d_inode(dentry), NULL);
415
416 cache = kzalloc(sizeof(struct ovl_dir_cache), GFP_KERNEL);
417 if (!cache)
418 return ERR_PTR(error: -ENOMEM);
419
420 cache->refcount = 1;
421 INIT_LIST_HEAD(list: &cache->entries);
422 cache->root = RB_ROOT;
423
424 res = ovl_dir_read_merged(dentry, list: &cache->entries, root: &cache->root);
425 if (res) {
426 ovl_cache_free(list: &cache->entries);
427 kfree(objp: cache);
428 return ERR_PTR(error: res);
429 }
430
431 cache->version = ovl_inode_version_get(inode);
432 ovl_set_dir_cache(inode, cache);
433
434 return cache;
435}
436
437/* Map inode number to lower fs unique range */
438static u64 ovl_remap_lower_ino(u64 ino, int xinobits, int fsid,
439 const char *name, int namelen, bool warn)
440{
441 unsigned int xinoshift = 64 - xinobits;
442
443 if (unlikely(ino >> xinoshift)) {
444 if (warn) {
445 pr_warn_ratelimited("d_ino too big (%.*s, ino=%llu, xinobits=%d)\n",
446 namelen, name, ino, xinobits);
447 }
448 return ino;
449 }
450
451 /*
452 * The lowest xinobit is reserved for mapping the non-peresistent inode
453 * numbers range, but this range is only exposed via st_ino, not here.
454 */
455 return ino | ((u64)fsid) << (xinoshift + 1);
456}
457
458/*
459 * Set d_ino for upper entries if needed. Non-upper entries should always report
460 * the uppermost real inode ino and should not call this function.
461 *
462 * When not all layer are on same fs, report real ino also for upper.
463 *
464 * When all layers are on the same fs, and upper has a reference to
465 * copy up origin, call vfs_getattr() on the overlay entry to make
466 * sure that d_ino will be consistent with st_ino from stat(2).
467 *
468 * Also checks the overlay.whiteout xattr by doing a full lookup which will return
469 * negative in this case.
470 */
471static int ovl_cache_update(const struct path *path, struct ovl_cache_entry *p, bool update_ino)
472
473{
474 struct dentry *dir = path->dentry;
475 struct ovl_fs *ofs = OVL_FS(sb: dir->d_sb);
476 struct dentry *this = NULL;
477 enum ovl_path_type type;
478 u64 ino = p->real_ino;
479 int xinobits = ovl_xino_bits(ofs);
480 int err = 0;
481
482 if (!ovl_same_dev(ofs) && !p->check_xwhiteout)
483 goto out;
484
485 if (p->name[0] == '.') {
486 if (p->len == 1) {
487 this = dget(dentry: dir);
488 goto get;
489 }
490 if (p->len == 2 && p->name[1] == '.') {
491 /* we shall not be moved */
492 this = dget(dentry: dir->d_parent);
493 goto get;
494 }
495 }
496 /* This checks also for xwhiteouts */
497 this = lookup_one(mnt_idmap(mnt: path->mnt), &QSTR_LEN(p->name, p->len), dir);
498 if (IS_ERR_OR_NULL(ptr: this) || !this->d_inode) {
499 /* Mark a stale entry */
500 p->is_whiteout = true;
501 if (IS_ERR(ptr: this)) {
502 err = PTR_ERR(ptr: this);
503 this = NULL;
504 goto fail;
505 }
506 goto out;
507 }
508
509get:
510 if (!ovl_same_dev(ofs) || !update_ino)
511 goto out;
512
513 type = ovl_path_type(dentry: this);
514 if (OVL_TYPE_ORIGIN(type)) {
515 struct kstat stat;
516 struct path statpath = *path;
517
518 statpath.dentry = this;
519 err = vfs_getattr(&statpath, &stat, STATX_INO, 0);
520 if (err)
521 goto fail;
522
523 /*
524 * Directory inode is always on overlay st_dev.
525 * Non-dir with ovl_same_dev() could be on pseudo st_dev in case
526 * of xino bits overflow.
527 */
528 WARN_ON_ONCE(S_ISDIR(stat.mode) &&
529 dir->d_sb->s_dev != stat.dev);
530 ino = stat.ino;
531 } else if (xinobits && !OVL_TYPE_UPPER(type)) {
532 ino = ovl_remap_lower_ino(ino, xinobits,
533 fsid: ovl_layer_lower(dentry: this)->fsid,
534 name: p->name, namelen: p->len,
535 warn: ovl_xino_warn(ofs));
536 }
537
538out:
539 p->ino = ino;
540 dput(this);
541 return err;
542
543fail:
544 pr_warn_ratelimited("failed to look up (%s) for ino (%i)\n",
545 p->name, err);
546 goto out;
547}
548
549static bool ovl_fill_plain(struct dir_context *ctx, const char *name,
550 int namelen, loff_t offset, u64 ino,
551 unsigned int d_type)
552{
553 struct ovl_cache_entry *p;
554 struct ovl_readdir_data *rdd =
555 container_of(ctx, struct ovl_readdir_data, ctx);
556
557 rdd->count++;
558 p = ovl_cache_entry_new(rdd, name, len: namelen, ino, d_type);
559 if (p == NULL) {
560 rdd->err = -ENOMEM;
561 return false;
562 }
563 list_add_tail(new: &p->l_node, head: rdd->list);
564
565 return true;
566}
567
568static int ovl_dir_read_impure(const struct path *path, struct list_head *list,
569 struct rb_root *root)
570{
571 int err;
572 struct path realpath;
573 struct ovl_cache_entry *p, *n;
574 struct ovl_readdir_data rdd = {
575 .ctx.actor = ovl_fill_plain,
576 .ctx.count = INT_MAX,
577 .list = list,
578 .root = root,
579 };
580
581 INIT_LIST_HEAD(list);
582 *root = RB_ROOT;
583 ovl_path_upper(dentry: path->dentry, path: &realpath);
584
585 err = ovl_dir_read(realpath: &realpath, rdd: &rdd);
586 if (err)
587 return err;
588
589 list_for_each_entry_safe(p, n, list, l_node) {
590 if (strcmp(p->name, ".") != 0 &&
591 strcmp(p->name, "..") != 0) {
592 err = ovl_cache_update(path, p, update_ino: true);
593 if (err)
594 return err;
595 }
596 if (p->ino == p->real_ino) {
597 list_del(entry: &p->l_node);
598 kfree(objp: p);
599 } else {
600 struct rb_node **newp = &root->rb_node;
601 struct rb_node *parent = NULL;
602
603 if (WARN_ON(ovl_cache_entry_find_link(p->name, p->len,
604 &newp, &parent)))
605 return -EIO;
606
607 rb_link_node(node: &p->node, parent, rb_link: newp);
608 rb_insert_color(&p->node, root);
609 }
610 }
611 return 0;
612}
613
614static struct ovl_dir_cache *ovl_cache_get_impure(const struct path *path)
615{
616 int res;
617 struct dentry *dentry = path->dentry;
618 struct inode *inode = d_inode(dentry);
619 struct ovl_fs *ofs = OVL_FS(sb: dentry->d_sb);
620 struct ovl_dir_cache *cache;
621
622 cache = ovl_dir_cache(inode);
623 if (cache && ovl_inode_version_get(inode) == cache->version)
624 return cache;
625
626 /* Impure cache is not refcounted, free it here */
627 ovl_dir_cache_free(inode);
628 ovl_set_dir_cache(inode, NULL);
629
630 cache = kzalloc(sizeof(struct ovl_dir_cache), GFP_KERNEL);
631 if (!cache)
632 return ERR_PTR(error: -ENOMEM);
633
634 res = ovl_dir_read_impure(path, list: &cache->entries, root: &cache->root);
635 if (res) {
636 ovl_cache_free(list: &cache->entries);
637 kfree(objp: cache);
638 return ERR_PTR(error: res);
639 }
640 if (list_empty(head: &cache->entries)) {
641 /*
642 * A good opportunity to get rid of an unneeded "impure" flag.
643 * Removing the "impure" xattr is best effort.
644 */
645 if (!ovl_want_write(dentry)) {
646 ovl_removexattr(ofs, dentry: ovl_dentry_upper(dentry),
647 ox: OVL_XATTR_IMPURE);
648 ovl_drop_write(dentry);
649 }
650 ovl_clear_flag(flag: OVL_IMPURE, inode);
651 kfree(objp: cache);
652 return NULL;
653 }
654
655 cache->version = ovl_inode_version_get(inode);
656 ovl_set_dir_cache(inode, cache);
657
658 return cache;
659}
660
661struct ovl_readdir_translate {
662 struct dir_context *orig_ctx;
663 struct ovl_dir_cache *cache;
664 struct dir_context ctx;
665 u64 parent_ino;
666 int fsid;
667 int xinobits;
668 bool xinowarn;
669};
670
671static bool ovl_fill_real(struct dir_context *ctx, const char *name,
672 int namelen, loff_t offset, u64 ino,
673 unsigned int d_type)
674{
675 struct ovl_readdir_translate *rdt =
676 container_of(ctx, struct ovl_readdir_translate, ctx);
677 struct dir_context *orig_ctx = rdt->orig_ctx;
678 bool res;
679
680 if (rdt->parent_ino && strcmp(name, "..") == 0) {
681 ino = rdt->parent_ino;
682 } else if (rdt->cache) {
683 struct ovl_cache_entry *p;
684
685 p = ovl_cache_entry_find(root: &rdt->cache->root, name, len: namelen);
686 if (p)
687 ino = p->ino;
688 } else if (rdt->xinobits) {
689 ino = ovl_remap_lower_ino(ino, xinobits: rdt->xinobits, fsid: rdt->fsid,
690 name, namelen, warn: rdt->xinowarn);
691 }
692
693 res = orig_ctx->actor(orig_ctx, name, namelen, offset, ino, d_type);
694 ctx->count = orig_ctx->count;
695
696 return res;
697}
698
699static bool ovl_is_impure_dir(struct file *file)
700{
701 struct ovl_dir_file *od = file->private_data;
702 struct inode *dir = file_inode(f: file);
703
704 /*
705 * Only upper dir can be impure, but if we are in the middle of
706 * iterating a lower real dir, dir could be copied up and marked
707 * impure. We only want the impure cache if we started iterating
708 * a real upper dir to begin with.
709 */
710 return od->is_upper && ovl_test_flag(flag: OVL_IMPURE, inode: dir);
711
712}
713
714static int ovl_iterate_real(struct file *file, struct dir_context *ctx)
715{
716 int err;
717 struct ovl_dir_file *od = file->private_data;
718 struct dentry *dir = file->f_path.dentry;
719 struct ovl_fs *ofs = OVL_FS(sb: dir->d_sb);
720 const struct ovl_layer *lower_layer = ovl_layer_lower(dentry: dir);
721 struct ovl_readdir_translate rdt = {
722 .ctx.actor = ovl_fill_real,
723 .ctx.count = ctx->count,
724 .orig_ctx = ctx,
725 .xinobits = ovl_xino_bits(ofs),
726 .xinowarn = ovl_xino_warn(ofs),
727 };
728
729 if (rdt.xinobits && lower_layer)
730 rdt.fsid = lower_layer->fsid;
731
732 if (OVL_TYPE_MERGE(ovl_path_type(dir->d_parent))) {
733 struct kstat stat;
734 struct path statpath = file->f_path;
735
736 statpath.dentry = dir->d_parent;
737 err = vfs_getattr(&statpath, &stat, STATX_INO, 0);
738 if (err)
739 return err;
740
741 WARN_ON_ONCE(dir->d_sb->s_dev != stat.dev);
742 rdt.parent_ino = stat.ino;
743 }
744
745 if (ovl_is_impure_dir(file)) {
746 rdt.cache = ovl_cache_get_impure(path: &file->f_path);
747 if (IS_ERR(ptr: rdt.cache))
748 return PTR_ERR(ptr: rdt.cache);
749 }
750
751 err = iterate_dir(od->realfile, &rdt.ctx);
752 ctx->pos = rdt.ctx.pos;
753
754 return err;
755}
756
757
758static int ovl_iterate(struct file *file, struct dir_context *ctx)
759{
760 struct ovl_dir_file *od = file->private_data;
761 struct dentry *dentry = file->f_path.dentry;
762 struct ovl_fs *ofs = OVL_FS(sb: dentry->d_sb);
763 struct ovl_cache_entry *p;
764 const struct cred *old_cred;
765 int err;
766
767 old_cred = ovl_override_creds(sb: dentry->d_sb);
768 if (!ctx->pos)
769 ovl_dir_reset(file);
770
771 if (od->is_real) {
772 /*
773 * If parent is merge, then need to adjust d_ino for '..', if
774 * dir is impure then need to adjust d_ino for copied up
775 * entries.
776 */
777 if (ovl_xino_bits(ofs) ||
778 (ovl_same_fs(ofs) &&
779 (ovl_is_impure_dir(file) ||
780 OVL_TYPE_MERGE(ovl_path_type(dentry->d_parent))))) {
781 err = ovl_iterate_real(file, ctx);
782 } else {
783 err = iterate_dir(od->realfile, ctx);
784 }
785 goto out;
786 }
787
788 if (!od->cache) {
789 struct ovl_dir_cache *cache;
790
791 cache = ovl_cache_get(dentry);
792 err = PTR_ERR(ptr: cache);
793 if (IS_ERR(ptr: cache))
794 goto out;
795
796 od->cache = cache;
797 ovl_seek_cursor(od, pos: ctx->pos);
798 }
799
800 while (od->cursor != &od->cache->entries) {
801 p = list_entry(od->cursor, struct ovl_cache_entry, l_node);
802 if (!p->is_whiteout) {
803 if (!p->ino || p->check_xwhiteout) {
804 err = ovl_cache_update(path: &file->f_path, p, update_ino: !p->ino);
805 if (err)
806 goto out;
807 }
808 }
809 /* ovl_cache_update() sets is_whiteout on stale entry */
810 if (!p->is_whiteout) {
811 if (!dir_emit(ctx, name: p->name, namelen: p->len, ino: p->ino, type: p->type))
812 break;
813 }
814 od->cursor = p->l_node.next;
815 ctx->pos++;
816 }
817 err = 0;
818out:
819 ovl_revert_creds(old_cred);
820 return err;
821}
822
823static loff_t ovl_dir_llseek(struct file *file, loff_t offset, int origin)
824{
825 loff_t res;
826 struct ovl_dir_file *od = file->private_data;
827
828 inode_lock(inode: file_inode(f: file));
829 if (!file->f_pos)
830 ovl_dir_reset(file);
831
832 if (od->is_real) {
833 res = vfs_llseek(file: od->realfile, offset, whence: origin);
834 file->f_pos = od->realfile->f_pos;
835 } else {
836 res = -EINVAL;
837
838 switch (origin) {
839 case SEEK_CUR:
840 offset += file->f_pos;
841 break;
842 case SEEK_SET:
843 break;
844 default:
845 goto out_unlock;
846 }
847 if (offset < 0)
848 goto out_unlock;
849
850 if (offset != file->f_pos) {
851 file->f_pos = offset;
852 if (od->cache)
853 ovl_seek_cursor(od, pos: offset);
854 }
855 res = offset;
856 }
857out_unlock:
858 inode_unlock(inode: file_inode(f: file));
859
860 return res;
861}
862
863static struct file *ovl_dir_open_realfile(const struct file *file,
864 const struct path *realpath)
865{
866 struct file *res;
867 const struct cred *old_cred;
868
869 old_cred = ovl_override_creds(sb: file_inode(f: file)->i_sb);
870 res = ovl_path_open(path: realpath, O_RDONLY | (file->f_flags & O_LARGEFILE));
871 ovl_revert_creds(old_cred);
872
873 return res;
874}
875
876/*
877 * Like ovl_real_fdget(), returns upperfile if dir was copied up since open.
878 * Unlike ovl_real_fdget(), this caches upperfile in file->private_data.
879 *
880 * TODO: use same abstract type for file->private_data of dir and file so
881 * upperfile could also be cached for files as well.
882 */
883struct file *ovl_dir_real_file(const struct file *file, bool want_upper)
884{
885
886 struct ovl_dir_file *od = file->private_data;
887 struct dentry *dentry = file->f_path.dentry;
888 struct file *old, *realfile = od->realfile;
889
890 if (!OVL_TYPE_UPPER(ovl_path_type(dentry)))
891 return want_upper ? NULL : realfile;
892
893 /*
894 * Need to check if we started out being a lower dir, but got copied up
895 */
896 if (!od->is_upper) {
897 realfile = READ_ONCE(od->upperfile);
898 if (!realfile) {
899 struct path upperpath;
900
901 ovl_path_upper(dentry, path: &upperpath);
902 realfile = ovl_dir_open_realfile(file, realpath: &upperpath);
903 if (IS_ERR(ptr: realfile))
904 return realfile;
905
906 old = cmpxchg_release(&od->upperfile, NULL, realfile);
907 if (old) {
908 fput(realfile);
909 realfile = old;
910 }
911 }
912 }
913
914 return realfile;
915}
916
917static int ovl_dir_fsync(struct file *file, loff_t start, loff_t end,
918 int datasync)
919{
920 struct file *realfile;
921 int err;
922
923 err = ovl_sync_status(ofs: OVL_FS(sb: file_inode(f: file)->i_sb));
924 if (err <= 0)
925 return err;
926
927 realfile = ovl_dir_real_file(file, want_upper: true);
928 err = PTR_ERR_OR_ZERO(ptr: realfile);
929
930 /* Nothing to sync for lower */
931 if (!realfile || err)
932 return err;
933
934 return vfs_fsync_range(file: realfile, start, end, datasync);
935}
936
937static int ovl_dir_release(struct inode *inode, struct file *file)
938{
939 struct ovl_dir_file *od = file->private_data;
940
941 if (od->cache) {
942 inode_lock(inode);
943 ovl_cache_put(od, inode);
944 inode_unlock(inode);
945 }
946 fput(od->realfile);
947 if (od->upperfile)
948 fput(od->upperfile);
949 kfree(objp: od);
950
951 return 0;
952}
953
954static int ovl_dir_open(struct inode *inode, struct file *file)
955{
956 struct path realpath;
957 struct file *realfile;
958 struct ovl_dir_file *od;
959 enum ovl_path_type type;
960
961 od = kzalloc(sizeof(struct ovl_dir_file), GFP_KERNEL);
962 if (!od)
963 return -ENOMEM;
964
965 type = ovl_path_real(dentry: file->f_path.dentry, path: &realpath);
966 realfile = ovl_dir_open_realfile(file, realpath: &realpath);
967 if (IS_ERR(ptr: realfile)) {
968 kfree(objp: od);
969 return PTR_ERR(ptr: realfile);
970 }
971 od->realfile = realfile;
972 od->is_real = ovl_dir_is_real(dir: inode);
973 od->is_upper = OVL_TYPE_UPPER(type);
974 file->private_data = od;
975
976 return 0;
977}
978
979WRAP_DIR_ITER(ovl_iterate) // FIXME!
980const struct file_operations ovl_dir_operations = {
981 .read = generic_read_dir,
982 .open = ovl_dir_open,
983 .iterate_shared = shared_ovl_iterate,
984 .llseek = ovl_dir_llseek,
985 .fsync = ovl_dir_fsync,
986 .release = ovl_dir_release,
987};
988
989int ovl_check_empty_dir(struct dentry *dentry, struct list_head *list)
990{
991 int err;
992 struct ovl_cache_entry *p, *n;
993 struct rb_root root = RB_ROOT;
994 const struct cred *old_cred;
995
996 old_cred = ovl_override_creds(sb: dentry->d_sb);
997 err = ovl_dir_read_merged(dentry, list, root: &root);
998 ovl_revert_creds(old_cred);
999 if (err)
1000 return err;
1001
1002 err = 0;
1003
1004 list_for_each_entry_safe(p, n, list, l_node) {
1005 /*
1006 * Select whiteouts in upperdir, they should
1007 * be cleared when deleting this directory.
1008 */
1009 if (p->is_whiteout) {
1010 if (p->is_upper)
1011 continue;
1012 goto del_entry;
1013 }
1014
1015 if (p->name[0] == '.') {
1016 if (p->len == 1)
1017 goto del_entry;
1018 if (p->len == 2 && p->name[1] == '.')
1019 goto del_entry;
1020 }
1021 err = -ENOTEMPTY;
1022 break;
1023
1024del_entry:
1025 list_del(entry: &p->l_node);
1026 kfree(objp: p);
1027 }
1028
1029 return err;
1030}
1031
1032void ovl_cleanup_whiteouts(struct ovl_fs *ofs, struct dentry *upper,
1033 struct list_head *list)
1034{
1035 struct ovl_cache_entry *p;
1036
1037 inode_lock_nested(inode: upper->d_inode, subclass: I_MUTEX_CHILD);
1038 list_for_each_entry(p, list, l_node) {
1039 struct dentry *dentry;
1040
1041 if (WARN_ON(!p->is_whiteout || !p->is_upper))
1042 continue;
1043
1044 dentry = ovl_lookup_upper(ofs, name: p->name, base: upper, len: p->len);
1045 if (IS_ERR(ptr: dentry)) {
1046 pr_err("lookup '%s/%.*s' failed (%i)\n",
1047 upper->d_name.name, p->len, p->name,
1048 (int) PTR_ERR(dentry));
1049 continue;
1050 }
1051 if (dentry->d_inode)
1052 ovl_cleanup(ofs, dir: upper->d_inode, dentry);
1053 dput(dentry);
1054 }
1055 inode_unlock(inode: upper->d_inode);
1056}
1057
1058static bool ovl_check_d_type(struct dir_context *ctx, const char *name,
1059 int namelen, loff_t offset, u64 ino,
1060 unsigned int d_type)
1061{
1062 struct ovl_readdir_data *rdd =
1063 container_of(ctx, struct ovl_readdir_data, ctx);
1064
1065 /* Even if d_type is not supported, DT_DIR is returned for . and .. */
1066 if (!strncmp(name, ".", namelen) || !strncmp(name, "..", namelen))
1067 return true;
1068
1069 if (d_type != DT_UNKNOWN)
1070 rdd->d_type_supported = true;
1071
1072 return true;
1073}
1074
1075/*
1076 * Returns 1 if d_type is supported, 0 not supported/unknown. Negative values
1077 * if error is encountered.
1078 */
1079int ovl_check_d_type_supported(const struct path *realpath)
1080{
1081 int err;
1082 struct ovl_readdir_data rdd = {
1083 .ctx.actor = ovl_check_d_type,
1084 .ctx.count = INT_MAX,
1085 .d_type_supported = false,
1086 };
1087
1088 err = ovl_dir_read(realpath, rdd: &rdd);
1089 if (err)
1090 return err;
1091
1092 return rdd.d_type_supported;
1093}
1094
1095#define OVL_INCOMPATDIR_NAME "incompat"
1096
1097static int ovl_workdir_cleanup_recurse(struct ovl_fs *ofs, const struct path *path,
1098 int level)
1099{
1100 int err;
1101 struct inode *dir = path->dentry->d_inode;
1102 LIST_HEAD(list);
1103 struct ovl_cache_entry *p;
1104 struct ovl_readdir_data rdd = {
1105 .ctx.actor = ovl_fill_plain,
1106 .ctx.count = INT_MAX,
1107 .list = &list,
1108 };
1109 bool incompat = false;
1110
1111 /*
1112 * The "work/incompat" directory is treated specially - if it is not
1113 * empty, instead of printing a generic error and mounting read-only,
1114 * we will error about incompat features and fail the mount.
1115 *
1116 * When called from ovl_indexdir_cleanup(), path->dentry->d_name.name
1117 * starts with '#'.
1118 */
1119 if (level == 2 &&
1120 !strcmp(path->dentry->d_name.name, OVL_INCOMPATDIR_NAME))
1121 incompat = true;
1122
1123 err = ovl_dir_read(realpath: path, rdd: &rdd);
1124 if (err)
1125 goto out;
1126
1127 inode_lock_nested(inode: dir, subclass: I_MUTEX_PARENT);
1128 list_for_each_entry(p, &list, l_node) {
1129 struct dentry *dentry;
1130
1131 if (p->name[0] == '.') {
1132 if (p->len == 1)
1133 continue;
1134 if (p->len == 2 && p->name[1] == '.')
1135 continue;
1136 } else if (incompat) {
1137 pr_err("overlay with incompat feature '%s' cannot be mounted\n",
1138 p->name);
1139 err = -EINVAL;
1140 break;
1141 }
1142 dentry = ovl_lookup_upper(ofs, name: p->name, base: path->dentry, len: p->len);
1143 if (IS_ERR(ptr: dentry))
1144 continue;
1145 if (dentry->d_inode)
1146 err = ovl_workdir_cleanup(ofs, dir, mnt: path->mnt, dentry, level);
1147 dput(dentry);
1148 if (err)
1149 break;
1150 }
1151 inode_unlock(inode: dir);
1152out:
1153 ovl_cache_free(list: &list);
1154 return err;
1155}
1156
1157int ovl_workdir_cleanup(struct ovl_fs *ofs, struct inode *dir,
1158 struct vfsmount *mnt, struct dentry *dentry, int level)
1159{
1160 int err;
1161
1162 if (!d_is_dir(dentry) || level > 1) {
1163 return ovl_cleanup(ofs, dir, dentry);
1164 }
1165
1166 err = ovl_do_rmdir(ofs, dir, dentry);
1167 if (err) {
1168 struct path path = { .mnt = mnt, .dentry = dentry };
1169
1170 inode_unlock(inode: dir);
1171 err = ovl_workdir_cleanup_recurse(ofs, path: &path, level: level + 1);
1172 inode_lock_nested(inode: dir, subclass: I_MUTEX_PARENT);
1173 if (!err)
1174 err = ovl_cleanup(ofs, dir, dentry);
1175 }
1176
1177 return err;
1178}
1179
1180int ovl_indexdir_cleanup(struct ovl_fs *ofs)
1181{
1182 int err;
1183 struct dentry *indexdir = ofs->workdir;
1184 struct dentry *index = NULL;
1185 struct inode *dir = indexdir->d_inode;
1186 struct path path = { .mnt = ovl_upper_mnt(ofs), .dentry = indexdir };
1187 LIST_HEAD(list);
1188 struct ovl_cache_entry *p;
1189 struct ovl_readdir_data rdd = {
1190 .ctx.actor = ovl_fill_plain,
1191 .ctx.count = INT_MAX,
1192 .list = &list,
1193 };
1194
1195 err = ovl_dir_read(realpath: &path, rdd: &rdd);
1196 if (err)
1197 goto out;
1198
1199 inode_lock_nested(inode: dir, subclass: I_MUTEX_PARENT);
1200 list_for_each_entry(p, &list, l_node) {
1201 if (p->name[0] == '.') {
1202 if (p->len == 1)
1203 continue;
1204 if (p->len == 2 && p->name[1] == '.')
1205 continue;
1206 }
1207 index = ovl_lookup_upper(ofs, name: p->name, base: indexdir, len: p->len);
1208 if (IS_ERR(ptr: index)) {
1209 err = PTR_ERR(ptr: index);
1210 index = NULL;
1211 break;
1212 }
1213 /* Cleanup leftover from index create/cleanup attempt */
1214 if (index->d_name.name[0] == '#') {
1215 err = ovl_workdir_cleanup(ofs, dir, mnt: path.mnt, dentry: index, level: 1);
1216 if (err)
1217 break;
1218 goto next;
1219 }
1220 err = ovl_verify_index(ofs, index);
1221 if (!err) {
1222 goto next;
1223 } else if (err == -ESTALE) {
1224 /* Cleanup stale index entries */
1225 err = ovl_cleanup(ofs, dir, dentry: index);
1226 } else if (err != -ENOENT) {
1227 /*
1228 * Abort mount to avoid corrupting the index if
1229 * an incompatible index entry was found or on out
1230 * of memory.
1231 */
1232 break;
1233 } else if (ofs->config.nfs_export) {
1234 /*
1235 * Whiteout orphan index to block future open by
1236 * handle after overlay nlink dropped to zero.
1237 */
1238 err = ovl_cleanup_and_whiteout(ofs, dir, dentry: index);
1239 } else {
1240 /* Cleanup orphan index entries */
1241 err = ovl_cleanup(ofs, dir, dentry: index);
1242 }
1243
1244 if (err)
1245 break;
1246
1247next:
1248 dput(index);
1249 index = NULL;
1250 }
1251 dput(index);
1252 inode_unlock(inode: dir);
1253out:
1254 ovl_cache_free(list: &list);
1255 if (err)
1256 pr_err("failed index dir cleanup (%i)\n", err);
1257 return err;
1258}
1259

source code of linux/fs/overlayfs/readdir.c