blob: b584dca845baa21b15e212c25bacb2ac176c4371 [file] [log] [blame]
Thomas Gleixnerd2912cb2019-06-04 10:11:33 +02001// SPDX-License-Identifier: GPL-2.0-only
Miklos Szeredie9be9d52014-10-24 00:14:38 +02002/*
3 *
4 * Copyright (C) 2011 Novell Inc.
Miklos Szeredie9be9d52014-10-24 00:14:38 +02005 */
6
7#include <linux/fs.h>
8#include <linux/slab.h>
Ingo Molnar5b825c32017-02-02 17:54:15 +01009#include <linux/cred.h>
Miklos Szeredie9be9d52014-10-24 00:14:38 +020010#include <linux/xattr.h>
Miklos Szeredi5201dc42016-09-01 11:11:59 +020011#include <linux/posix_acl.h>
Amir Goldstein5f8415d2017-06-20 15:35:14 +030012#include <linux/ratelimit.h>
Christoph Hellwig10c5db22020-05-23 09:30:11 +020013#include <linux/fiemap.h>
Miklos Szeredie9be9d52014-10-24 00:14:38 +020014#include "overlayfs.h"
15
Chandan Rajendraba1e5632017-07-24 01:57:54 -050016
Miklos Szeredie9be9d52014-10-24 00:14:38 +020017int ovl_setattr(struct dentry *dentry, struct iattr *attr)
18{
19 int err;
Vivek Goyal997336f2018-05-11 11:49:33 -040020 bool full_copy_up = false;
Miklos Szeredie9be9d52014-10-24 00:14:38 +020021 struct dentry *upperdentry;
Vivek Goyal1175b6b2016-07-01 16:34:28 -040022 const struct cred *old_cred;
Miklos Szeredie9be9d52014-10-24 00:14:38 +020023
Jan Kara31051c82016-05-26 16:55:18 +020024 err = setattr_prepare(dentry, attr);
Miklos Szeredicf9a6782015-12-11 16:30:49 +010025 if (err)
26 return err;
27
Miklos Szeredie9be9d52014-10-24 00:14:38 +020028 err = ovl_want_write(dentry);
29 if (err)
30 goto out;
31
Miklos Szeredi58121602018-07-18 15:44:41 +020032 if (attr->ia_valid & ATTR_SIZE) {
33 struct inode *realinode = d_inode(ovl_dentry_real(dentry));
34
35 err = -ETXTBSY;
36 if (atomic_read(&realinode->i_writecount) < 0)
37 goto out_drop_write;
Vivek Goyal997336f2018-05-11 11:49:33 -040038
39 /* Truncate should trigger data copy up as well */
40 full_copy_up = true;
Miklos Szeredi58121602018-07-18 15:44:41 +020041 }
42
Vivek Goyal997336f2018-05-11 11:49:33 -040043 if (!full_copy_up)
44 err = ovl_copy_up(dentry);
45 else
46 err = ovl_copy_up_with_data(dentry);
Miklos Szerediacff81e2015-12-04 19:18:48 +010047 if (!err) {
Miklos Szeredi58121602018-07-18 15:44:41 +020048 struct inode *winode = NULL;
49
Miklos Szerediacff81e2015-12-04 19:18:48 +010050 upperdentry = ovl_dentry_upper(dentry);
51
Miklos Szeredi58121602018-07-18 15:44:41 +020052 if (attr->ia_valid & ATTR_SIZE) {
53 winode = d_inode(upperdentry);
54 err = get_write_access(winode);
55 if (err)
56 goto out_drop_write;
57 }
58
Miklos Szeredib99c2d92016-07-04 16:49:48 +020059 if (attr->ia_valid & (ATTR_KILL_SUID|ATTR_KILL_SGID))
60 attr->ia_valid &= ~ATTR_MODE;
61
Vivek Goyale67f0212020-04-22 09:08:49 -040062 /*
Vivek Goyal15fd2ea2020-04-22 09:08:50 -040063 * We might have to translate ovl file into real file object
64 * once use cases emerge. For now, simply don't let underlying
65 * filesystem rely on attr->ia_file
Vivek Goyale67f0212020-04-22 09:08:49 -040066 */
67 attr->ia_valid &= ~ATTR_FILE;
68
Vivek Goyal15fd2ea2020-04-22 09:08:50 -040069 /*
70 * If open(O_TRUNC) is done, VFS calls ->setattr with ATTR_OPEN
71 * set. Overlayfs does not pass O_TRUNC flag to underlying
72 * filesystem during open -> do not pass ATTR_OPEN. This
73 * disables optimization in fuse which assumes open(O_TRUNC)
74 * already set file size to 0. But we never passed O_TRUNC to
75 * fuse. So by clearing ATTR_OPEN, fuse will be forced to send
76 * setattr request to server.
77 */
78 attr->ia_valid &= ~ATTR_OPEN;
79
Al Viro59551022016-01-22 15:40:57 -050080 inode_lock(upperdentry->d_inode);
Vivek Goyal1175b6b2016-07-01 16:34:28 -040081 old_cred = ovl_override_creds(dentry->d_sb);
Miklos Szeredie9be9d52014-10-24 00:14:38 +020082 err = notify_change(upperdentry, attr, NULL);
Vivek Goyal1175b6b2016-07-01 16:34:28 -040083 revert_creds(old_cred);
Konstantin Khlebnikovb81de062016-01-31 16:21:29 +030084 if (!err)
85 ovl_copyattr(upperdentry->d_inode, dentry->d_inode);
Al Viro59551022016-01-22 15:40:57 -050086 inode_unlock(upperdentry->d_inode);
Miklos Szeredi58121602018-07-18 15:44:41 +020087
88 if (winode)
89 put_write_access(winode);
Miklos Szeredie9be9d52014-10-24 00:14:38 +020090 }
Miklos Szeredi58121602018-07-18 15:44:41 +020091out_drop_write:
Miklos Szeredie9be9d52014-10-24 00:14:38 +020092 ovl_drop_write(dentry);
93out:
94 return err;
95}
96
Amir Goldstein07f1e592020-01-14 21:59:22 +020097static int ovl_map_dev_ino(struct dentry *dentry, struct kstat *stat, int fsid)
Amir Goldsteinda309e82017-11-08 19:39:51 +020098{
Amir Goldstein0f831ec2019-11-16 18:14:41 +020099 bool samefs = ovl_same_fs(dentry->d_sb);
Amir Goldsteine487d882017-11-07 13:55:04 +0200100 unsigned int xinobits = ovl_xino_bits(dentry->d_sb);
Amir Goldsteindfe51d42020-02-21 16:34:44 +0200101 unsigned int xinoshift = 64 - xinobits;
Amir Goldsteinda309e82017-11-08 19:39:51 +0200102
103 if (samefs) {
104 /*
105 * When all layers are on the same fs, all real inode
106 * number are unique, so we use the overlay st_dev,
107 * which is friendly to du -x.
108 */
109 stat->dev = dentry->d_sb->s_dev;
Amir Goldsteine487d882017-11-07 13:55:04 +0200110 return 0;
111 } else if (xinobits) {
Amir Goldsteine487d882017-11-07 13:55:04 +0200112 /*
113 * All inode numbers of underlying fs should not be using the
114 * high xinobits, so we use high xinobits to partition the
115 * overlay st_ino address space. The high bits holds the fsid
Amir Goldsteindfe51d42020-02-21 16:34:44 +0200116 * (upper fsid is 0). The lowest xinobit is reserved for mapping
117 * the non-peresistent inode numbers range in case of overflow.
118 * This way all overlay inode numbers are unique and use the
119 * overlay st_dev.
Amir Goldsteine487d882017-11-07 13:55:04 +0200120 */
Amir Goldstein926e94d2020-02-21 16:34:45 +0200121 if (likely(!(stat->ino >> xinoshift))) {
Amir Goldsteindfe51d42020-02-21 16:34:44 +0200122 stat->ino |= ((u64)fsid) << (xinoshift + 1);
Amir Goldsteine487d882017-11-07 13:55:04 +0200123 stat->dev = dentry->d_sb->s_dev;
124 return 0;
Amir Goldstein926e94d2020-02-21 16:34:45 +0200125 } else if (ovl_xino_warn(dentry->d_sb)) {
126 pr_warn_ratelimited("inode number too big (%pd2, ino=%llu, xinobits=%d)\n",
127 dentry, stat->ino, xinobits);
Amir Goldsteine487d882017-11-07 13:55:04 +0200128 }
129 }
130
131 /* The inode could not be mapped to a unified st_ino address space */
132 if (S_ISDIR(dentry->d_inode->i_mode)) {
Amir Goldsteinda309e82017-11-08 19:39:51 +0200133 /*
134 * Always use the overlay st_dev for directories, so 'find
135 * -xdev' will scan the entire overlay mount and won't cross the
136 * overlay mount boundaries.
137 *
138 * If not all layers are on the same fs the pair {real st_ino;
139 * overlay st_dev} is not unique, so use the non persistent
140 * overlay st_ino for directories.
141 */
142 stat->dev = dentry->d_sb->s_dev;
143 stat->ino = dentry->d_inode->i_ino;
Amir Goldsteinb7bf9902020-01-14 22:17:25 +0200144 } else {
Amir Goldsteinda309e82017-11-08 19:39:51 +0200145 /*
146 * For non-samefs setup, if we cannot map all layers st_ino
147 * to a unified address space, we need to make sure that st_dev
Amir Goldsteinb7bf9902020-01-14 22:17:25 +0200148 * is unique per underlying fs, so we use the unique anonymous
149 * bdev assigned to the underlying fs.
Amir Goldsteinda309e82017-11-08 19:39:51 +0200150 */
Amir Goldstein07f1e592020-01-14 21:59:22 +0200151 stat->dev = OVL_FS(dentry->d_sb)->fs[fsid].pseudo_dev;
Amir Goldsteinda309e82017-11-08 19:39:51 +0200152 }
153
154 return 0;
155}
156
Miklos Szeredi5b712092017-05-05 11:38:58 +0200157int ovl_getattr(const struct path *path, struct kstat *stat,
158 u32 request_mask, unsigned int flags)
Miklos Szeredie9be9d52014-10-24 00:14:38 +0200159{
David Howellsa528d352017-01-31 16:46:22 +0000160 struct dentry *dentry = path->dentry;
Amir Goldstein72b608f2017-04-24 00:25:49 +0300161 enum ovl_path_type type;
Miklos Szeredie9be9d52014-10-24 00:14:38 +0200162 struct path realpath;
Vivek Goyal1175b6b2016-07-01 16:34:28 -0400163 const struct cred *old_cred;
Miklos Szeredi5b712092017-05-05 11:38:58 +0200164 bool is_dir = S_ISDIR(dentry->d_inode->i_mode);
Amir Goldstein07f1e592020-01-14 21:59:22 +0200165 int fsid = 0;
Vivek Goyal1175b6b2016-07-01 16:34:28 -0400166 int err;
Vivek Goyal67d756c22018-05-11 11:49:30 -0400167 bool metacopy_blocks = false;
168
169 metacopy_blocks = ovl_is_metacopy_dentry(dentry);
Miklos Szeredie9be9d52014-10-24 00:14:38 +0200170
Amir Goldstein72b608f2017-04-24 00:25:49 +0300171 type = ovl_path_real(dentry, &realpath);
Vivek Goyal1175b6b2016-07-01 16:34:28 -0400172 old_cred = ovl_override_creds(dentry->d_sb);
David Howellsa528d352017-01-31 16:46:22 +0000173 err = vfs_getattr(&realpath, stat, request_mask, flags);
Amir Goldstein72b608f2017-04-24 00:25:49 +0300174 if (err)
175 goto out;
176
177 /*
Amir Goldsteinda309e82017-11-08 19:39:51 +0200178 * For non-dir or same fs, we use st_ino of the copy up origin.
179 * This guaranties constant st_dev/st_ino across copy up.
Amir Goldsteine487d882017-11-07 13:55:04 +0200180 * With xino feature and non-samefs, we use st_ino of the copy up
181 * origin masked with high bits that represent the layer id.
Amir Goldstein72b608f2017-04-24 00:25:49 +0300182 *
Amir Goldsteinda309e82017-11-08 19:39:51 +0200183 * If lower filesystem supports NFS file handles, this also guaranties
Amir Goldstein72b608f2017-04-24 00:25:49 +0300184 * persistent st_ino across mount cycle.
185 */
Amir Goldstein0f831ec2019-11-16 18:14:41 +0200186 if (!is_dir || ovl_same_dev(dentry->d_sb)) {
Amir Goldsteinda309e82017-11-08 19:39:51 +0200187 if (!OVL_TYPE_UPPER(type)) {
Amir Goldstein07f1e592020-01-14 21:59:22 +0200188 fsid = ovl_layer_lower(dentry)->fsid;
Amir Goldsteinda309e82017-11-08 19:39:51 +0200189 } else if (OVL_TYPE_ORIGIN(type)) {
Amir Goldstein72b608f2017-04-24 00:25:49 +0300190 struct kstat lowerstat;
Vivek Goyal67d756c22018-05-11 11:49:30 -0400191 u32 lowermask = STATX_INO | STATX_BLOCKS |
192 (!is_dir ? STATX_NLINK : 0);
Amir Goldstein72b608f2017-04-24 00:25:49 +0300193
194 ovl_path_lower(dentry, &realpath);
195 err = vfs_getattr(&realpath, &lowerstat,
Miklos Szeredi5b712092017-05-05 11:38:58 +0200196 lowermask, flags);
Amir Goldstein72b608f2017-04-24 00:25:49 +0300197 if (err)
198 goto out;
199
Amir Goldstein72b608f2017-04-24 00:25:49 +0300200 /*
Amir Goldstein359f3922017-06-21 15:28:41 +0300201 * Lower hardlinks may be broken on copy up to different
Amir Goldstein72b608f2017-04-24 00:25:49 +0300202 * upper files, so we cannot use the lower origin st_ino
203 * for those different files, even for the same fs case.
Amir Goldstein86eaa132017-11-21 13:55:51 +0200204 *
205 * Similarly, several redirected dirs can point to the
206 * same dir on a lower layer. With the "verify_lower"
207 * feature, we do not use the lower origin st_ino, if
208 * we haven't verified that this redirect is unique.
209 *
Amir Goldstein359f3922017-06-21 15:28:41 +0300210 * With inodes index enabled, it is safe to use st_ino
Amir Goldstein86eaa132017-11-21 13:55:51 +0200211 * of an indexed origin. The index validates that the
212 * upper hardlink is not broken and that a redirected
213 * dir is the only redirect to that origin.
Amir Goldstein72b608f2017-04-24 00:25:49 +0300214 */
Amir Goldstein86eaa132017-11-21 13:55:51 +0200215 if (ovl_test_flag(OVL_INDEX, d_inode(dentry)) ||
216 (!ovl_verify_lower(dentry->d_sb) &&
Amir Goldstein9f99e502018-04-11 20:09:29 +0300217 (is_dir || lowerstat.nlink == 1))) {
Amir Goldstein07f1e592020-01-14 21:59:22 +0200218 fsid = ovl_layer_lower(dentry)->fsid;
Amir Goldsteinb7bf9902020-01-14 22:17:25 +0200219 stat->ino = lowerstat.ino;
Amir Goldstein9f99e502018-04-11 20:09:29 +0300220 }
Vivek Goyal67d756c22018-05-11 11:49:30 -0400221
222 /*
223 * If we are querying a metacopy dentry and lower
224 * dentry is data dentry, then use the blocks we
225 * queried just now. We don't have to do additional
226 * vfs_getattr(). If lower itself is metacopy, then
227 * additional vfs_getattr() is unavoidable.
228 */
229 if (metacopy_blocks &&
230 realpath.dentry == ovl_dentry_lowerdata(dentry)) {
231 stat->blocks = lowerstat.blocks;
232 metacopy_blocks = false;
233 }
234 }
235
236 if (metacopy_blocks) {
237 /*
238 * If lower is not same as lowerdata or if there was
239 * no origin on upper, we can end up here.
240 */
241 struct kstat lowerdatastat;
242 u32 lowermask = STATX_BLOCKS;
243
244 ovl_path_lowerdata(dentry, &realpath);
245 err = vfs_getattr(&realpath, &lowerdatastat,
246 lowermask, flags);
247 if (err)
248 goto out;
249 stat->blocks = lowerdatastat.blocks;
Amir Goldstein72b608f2017-04-24 00:25:49 +0300250 }
Amir Goldstein72b608f2017-04-24 00:25:49 +0300251 }
Miklos Szeredi5b712092017-05-05 11:38:58 +0200252
Amir Goldstein07f1e592020-01-14 21:59:22 +0200253 err = ovl_map_dev_ino(dentry, stat, fsid);
Amir Goldsteinda309e82017-11-08 19:39:51 +0200254 if (err)
255 goto out;
256
Miklos Szeredi5b712092017-05-05 11:38:58 +0200257 /*
258 * It's probably not worth it to count subdirs to get the
259 * correct link count. nlink=1 seems to pacify 'find' and
260 * other utilities.
261 */
262 if (is_dir && OVL_TYPE_MERGE(type))
263 stat->nlink = 1;
264
Amir Goldstein5f8415d2017-06-20 15:35:14 +0300265 /*
266 * Return the overlay inode nlinks for indexed upper inodes.
267 * Overlay inode nlink counts the union of the upper hardlinks
268 * and non-covered lower hardlinks. It does not include the upper
269 * index hardlink.
270 */
271 if (!is_dir && ovl_test_flag(OVL_INDEX, d_inode(dentry)))
272 stat->nlink = dentry->d_inode->i_nlink;
273
Amir Goldstein72b608f2017-04-24 00:25:49 +0300274out:
Vivek Goyal1175b6b2016-07-01 16:34:28 -0400275 revert_creds(old_cred);
Amir Goldstein72b608f2017-04-24 00:25:49 +0300276
Vivek Goyal1175b6b2016-07-01 16:34:28 -0400277 return err;
Miklos Szeredie9be9d52014-10-24 00:14:38 +0200278}
279
280int ovl_permission(struct inode *inode, int mask)
281{
Miklos Szeredi09d8b582017-07-04 22:03:16 +0200282 struct inode *upperinode = ovl_inode_upper(inode);
283 struct inode *realinode = upperinode ?: ovl_inode_lower(inode);
Vivek Goyalc0ca3d72016-07-01 16:34:27 -0400284 const struct cred *old_cred;
Miklos Szeredie9be9d52014-10-24 00:14:38 +0200285 int err;
286
Miklos Szeredie9be9d52014-10-24 00:14:38 +0200287 /* Careful in RCU walk mode */
Miklos Szeredie9be9d52014-10-24 00:14:38 +0200288 if (!realinode) {
289 WARN_ON(!(mask & MAY_NOT_BLOCK));
Miklos Szeredia999d7e2016-07-29 12:05:23 +0200290 return -ECHILD;
Miklos Szeredie9be9d52014-10-24 00:14:38 +0200291 }
292
Vivek Goyalc0ca3d72016-07-01 16:34:27 -0400293 /*
294 * Check overlay inode with the creds of task and underlying inode
295 * with creds of mounter
296 */
297 err = generic_permission(inode, mask);
298 if (err)
299 return err;
300
Miklos Szerediec7ba112018-12-04 11:31:30 +0100301 old_cred = ovl_override_creds(inode->i_sb);
302 if (!upperinode &&
303 !special_file(realinode->i_mode) && mask & MAY_WRITE) {
Vivek Goyal754f8cb2016-07-01 16:34:29 -0400304 mask &= ~(MAY_WRITE | MAY_APPEND);
Miklos Szerediec7ba112018-12-04 11:31:30 +0100305 /* Make sure mounter can read file for copy up later */
Vivek Goyal500cac32016-07-13 11:00:14 -0400306 mask |= MAY_READ;
307 }
Miklos Szeredi9c630eb2016-07-29 12:05:23 +0200308 err = inode_permission(realinode, mask);
Vivek Goyalc0ca3d72016-07-01 16:34:27 -0400309 revert_creds(old_cred);
310
311 return err;
Miklos Szeredie9be9d52014-10-24 00:14:38 +0200312}
313
Al Viro6b255392015-11-17 10:20:54 -0500314static const char *ovl_get_link(struct dentry *dentry,
Al Virofceef392015-12-29 15:58:39 -0500315 struct inode *inode,
316 struct delayed_call *done)
Miklos Szeredie9be9d52014-10-24 00:14:38 +0200317{
Vivek Goyal1175b6b2016-07-01 16:34:28 -0400318 const struct cred *old_cred;
319 const char *p;
Miklos Szeredie9be9d52014-10-24 00:14:38 +0200320
Al Viro6b255392015-11-17 10:20:54 -0500321 if (!dentry)
322 return ERR_PTR(-ECHILD);
323
Vivek Goyal1175b6b2016-07-01 16:34:28 -0400324 old_cred = ovl_override_creds(dentry->d_sb);
Miklos Szeredi7764235b2016-10-04 14:40:45 +0200325 p = vfs_get_link(ovl_dentry_real(dentry), done);
Vivek Goyal1175b6b2016-07-01 16:34:28 -0400326 revert_creds(old_cred);
327 return p;
Miklos Szeredie9be9d52014-10-24 00:14:38 +0200328}
329
Miklos Szeredi610afc02020-09-02 10:58:49 +0200330bool ovl_is_private_xattr(struct super_block *sb, const char *name)
Miklos Szeredie9be9d52014-10-24 00:14:38 +0200331{
Andreas Gruenbacherfe2b7592016-08-22 17:59:22 +0200332 return strncmp(name, OVL_XATTR_PREFIX,
333 sizeof(OVL_XATTR_PREFIX) - 1) == 0;
Miklos Szeredie9be9d52014-10-24 00:14:38 +0200334}
335
Miklos Szeredi1d88f182017-07-20 11:08:21 +0200336int ovl_xattr_set(struct dentry *dentry, struct inode *inode, const char *name,
337 const void *value, size_t size, int flags)
Miklos Szeredie9be9d52014-10-24 00:14:38 +0200338{
339 int err;
Miklos Szeredi1d88f182017-07-20 11:08:21 +0200340 struct dentry *upperdentry = ovl_i_dentry_upper(inode);
341 struct dentry *realdentry = upperdentry ?: ovl_dentry_lower(dentry);
Vivek Goyal1175b6b2016-07-01 16:34:28 -0400342 const struct cred *old_cred;
Miklos Szeredie9be9d52014-10-24 00:14:38 +0200343
344 err = ovl_want_write(dentry);
345 if (err)
346 goto out;
347
Miklos Szeredi1d88f182017-07-20 11:08:21 +0200348 if (!value && !upperdentry) {
349 err = vfs_getxattr(realdentry, name, NULL, 0);
Andreas Gruenbacher0e585cc2016-08-22 17:22:11 +0200350 if (err < 0)
351 goto out_drop_write;
352 }
353
Miklos Szeredi1d88f182017-07-20 11:08:21 +0200354 if (!upperdentry) {
355 err = ovl_copy_up(dentry);
356 if (err)
357 goto out_drop_write;
Miklos Szeredie9be9d52014-10-24 00:14:38 +0200358
Miklos Szeredi1d88f182017-07-20 11:08:21 +0200359 realdentry = ovl_dentry_upper(dentry);
360 }
Andreas Gruenbacher0e585cc2016-08-22 17:22:11 +0200361
Vivek Goyal1175b6b2016-07-01 16:34:28 -0400362 old_cred = ovl_override_creds(dentry->d_sb);
Andreas Gruenbacher0e585cc2016-08-22 17:22:11 +0200363 if (value)
Miklos Szeredi1d88f182017-07-20 11:08:21 +0200364 err = vfs_setxattr(realdentry, name, value, size, flags);
Andreas Gruenbacher0e585cc2016-08-22 17:22:11 +0200365 else {
366 WARN_ON(flags != XATTR_REPLACE);
Miklos Szeredi1d88f182017-07-20 11:08:21 +0200367 err = vfs_removexattr(realdentry, name);
Andreas Gruenbacher0e585cc2016-08-22 17:22:11 +0200368 }
Vivek Goyal1175b6b2016-07-01 16:34:28 -0400369 revert_creds(old_cred);
Miklos Szeredie9be9d52014-10-24 00:14:38 +0200370
Miklos Szeredid9854c82018-07-18 15:44:40 +0200371 /* copy c/mtime */
372 ovl_copyattr(d_inode(realdentry), inode);
373
Miklos Szeredie9be9d52014-10-24 00:14:38 +0200374out_drop_write:
375 ovl_drop_write(dentry);
376out:
377 return err;
378}
379
Miklos Szeredi1d88f182017-07-20 11:08:21 +0200380int ovl_xattr_get(struct dentry *dentry, struct inode *inode, const char *name,
Andreas Gruenbacher0eb45fc2016-08-22 17:52:55 +0200381 void *value, size_t size)
Miklos Szeredie9be9d52014-10-24 00:14:38 +0200382{
Vivek Goyal1175b6b2016-07-01 16:34:28 -0400383 ssize_t res;
384 const struct cred *old_cred;
Miklos Szeredi1d88f182017-07-20 11:08:21 +0200385 struct dentry *realdentry =
386 ovl_i_dentry_upper(inode) ?: ovl_dentry_lower(dentry);
Miklos Szeredi52148462014-11-20 16:40:00 +0100387
Vivek Goyal1175b6b2016-07-01 16:34:28 -0400388 old_cred = ovl_override_creds(dentry->d_sb);
389 res = vfs_getxattr(realdentry, name, value, size);
390 revert_creds(old_cred);
391 return res;
Miklos Szeredie9be9d52014-10-24 00:14:38 +0200392}
393
Miklos Szeredi610afc02020-09-02 10:58:49 +0200394static bool ovl_can_list(struct super_block *sb, const char *s)
Miklos Szeredia082c6f2017-05-29 15:15:27 +0200395{
Miklos Szeredi8f6ee742020-09-02 10:58:49 +0200396 /* Never list private (.overlay) */
397 if (ovl_is_private_xattr(sb, s))
398 return false;
399
Miklos Szeredia082c6f2017-05-29 15:15:27 +0200400 /* List all non-trusted xatts */
401 if (strncmp(s, XATTR_TRUSTED_PREFIX, XATTR_TRUSTED_PREFIX_LEN) != 0)
402 return true;
403
Miklos Szeredi8f6ee742020-09-02 10:58:49 +0200404 /* list other trusted for superuser only */
405 return ns_capable_noaudit(&init_user_ns, CAP_SYS_ADMIN);
Miklos Szeredia082c6f2017-05-29 15:15:27 +0200406}
407
Miklos Szeredie9be9d52014-10-24 00:14:38 +0200408ssize_t ovl_listxattr(struct dentry *dentry, char *list, size_t size)
409{
Miklos Szeredib5817552016-06-06 16:21:37 +0200410 struct dentry *realdentry = ovl_dentry_real(dentry);
Miklos Szeredie9be9d52014-10-24 00:14:38 +0200411 ssize_t res;
Miklos Szeredi7cb35112016-09-01 11:12:00 +0200412 size_t len;
413 char *s;
Vivek Goyal1175b6b2016-07-01 16:34:28 -0400414 const struct cred *old_cred;
Miklos Szeredie9be9d52014-10-24 00:14:38 +0200415
Vivek Goyal1175b6b2016-07-01 16:34:28 -0400416 old_cred = ovl_override_creds(dentry->d_sb);
Miklos Szeredib5817552016-06-06 16:21:37 +0200417 res = vfs_listxattr(realdentry, list, size);
Vivek Goyal1175b6b2016-07-01 16:34:28 -0400418 revert_creds(old_cred);
Miklos Szeredie9be9d52014-10-24 00:14:38 +0200419 if (res <= 0 || size == 0)
420 return res;
421
Miklos Szeredie9be9d52014-10-24 00:14:38 +0200422 /* filter out private xattrs */
Miklos Szeredi7cb35112016-09-01 11:12:00 +0200423 for (s = list, len = res; len;) {
424 size_t slen = strnlen(s, len) + 1;
Miklos Szeredie9be9d52014-10-24 00:14:38 +0200425
Miklos Szeredi7cb35112016-09-01 11:12:00 +0200426 /* underlying fs providing us with an broken xattr list? */
427 if (WARN_ON(slen > len))
428 return -EIO;
Miklos Szeredie9be9d52014-10-24 00:14:38 +0200429
Miklos Szeredi7cb35112016-09-01 11:12:00 +0200430 len -= slen;
Miklos Szeredi610afc02020-09-02 10:58:49 +0200431 if (!ovl_can_list(dentry->d_sb, s)) {
Miklos Szeredie9be9d52014-10-24 00:14:38 +0200432 res -= slen;
Miklos Szeredi7cb35112016-09-01 11:12:00 +0200433 memmove(s, s + slen, len);
Miklos Szeredie9be9d52014-10-24 00:14:38 +0200434 } else {
Miklos Szeredi7cb35112016-09-01 11:12:00 +0200435 s += slen;
Miklos Szeredie9be9d52014-10-24 00:14:38 +0200436 }
437 }
438
439 return res;
440}
441
Vivek Goyal39a25b22016-07-01 16:34:26 -0400442struct posix_acl *ovl_get_acl(struct inode *inode, int type)
443{
Miklos Szeredi09d8b582017-07-04 22:03:16 +0200444 struct inode *realinode = ovl_inode_real(inode);
Vivek Goyal1175b6b2016-07-01 16:34:28 -0400445 const struct cred *old_cred;
446 struct posix_acl *acl;
Vivek Goyal39a25b22016-07-01 16:34:26 -0400447
Miklos Szeredi5201dc42016-09-01 11:11:59 +0200448 if (!IS_ENABLED(CONFIG_FS_POSIX_ACL) || !IS_POSIXACL(realinode))
Vivek Goyal39a25b22016-07-01 16:34:26 -0400449 return NULL;
450
Vivek Goyal1175b6b2016-07-01 16:34:28 -0400451 old_cred = ovl_override_creds(inode->i_sb);
Miklos Szeredi5201dc42016-09-01 11:11:59 +0200452 acl = get_acl(realinode, type);
Vivek Goyal1175b6b2016-07-01 16:34:28 -0400453 revert_creds(old_cred);
454
455 return acl;
Vivek Goyal39a25b22016-07-01 16:34:26 -0400456}
457
Deepa Dinamani95582b02018-05-08 19:36:02 -0700458int ovl_update_time(struct inode *inode, struct timespec64 *ts, int flags)
Miklos Szeredid719e8f2016-07-29 12:05:23 +0200459{
Miklos Szeredi8f35cf52018-04-12 12:04:50 +0200460 if (flags & S_ATIME) {
461 struct ovl_fs *ofs = inode->i_sb->s_fs_info;
462 struct path upperpath = {
Miklos Szeredi08f4c7c2020-06-04 10:48:19 +0200463 .mnt = ovl_upper_mnt(ofs),
Miklos Szeredi8f35cf52018-04-12 12:04:50 +0200464 .dentry = ovl_upperdentry_dereference(OVL_I(inode)),
465 };
Miklos Szeredid719e8f2016-07-29 12:05:23 +0200466
Miklos Szeredi8f35cf52018-04-12 12:04:50 +0200467 if (upperpath.dentry) {
468 touch_atime(&upperpath);
469 inode->i_atime = d_inode(upperpath.dentry)->i_atime;
470 }
Miklos Szeredid719e8f2016-07-29 12:05:23 +0200471 }
Miklos Szeredid719e8f2016-07-29 12:05:23 +0200472 return 0;
473}
474
Miklos Szeredi9e142c42018-07-18 15:44:42 +0200475static int ovl_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
476 u64 start, u64 len)
477{
478 int err;
479 struct inode *realinode = ovl_inode_real(inode);
480 const struct cred *old_cred;
481
482 if (!realinode->i_op->fiemap)
483 return -EOPNOTSUPP;
484
485 old_cred = ovl_override_creds(inode->i_sb);
486 err = realinode->i_op->fiemap(realinode, fieinfo, start, len);
487 revert_creds(old_cred);
488
489 return err;
490}
491
Miklos Szeredie9be9d52014-10-24 00:14:38 +0200492static const struct inode_operations ovl_file_inode_operations = {
493 .setattr = ovl_setattr,
494 .permission = ovl_permission,
495 .getattr = ovl_getattr,
Miklos Szeredie9be9d52014-10-24 00:14:38 +0200496 .listxattr = ovl_listxattr,
Vivek Goyal39a25b22016-07-01 16:34:26 -0400497 .get_acl = ovl_get_acl,
Miklos Szeredid719e8f2016-07-29 12:05:23 +0200498 .update_time = ovl_update_time,
Miklos Szeredi9e142c42018-07-18 15:44:42 +0200499 .fiemap = ovl_fiemap,
Miklos Szeredie9be9d52014-10-24 00:14:38 +0200500};
501
502static const struct inode_operations ovl_symlink_inode_operations = {
503 .setattr = ovl_setattr,
Al Viro6b255392015-11-17 10:20:54 -0500504 .get_link = ovl_get_link,
Miklos Szeredie9be9d52014-10-24 00:14:38 +0200505 .getattr = ovl_getattr,
Miklos Szeredie9be9d52014-10-24 00:14:38 +0200506 .listxattr = ovl_listxattr,
Miklos Szeredid719e8f2016-07-29 12:05:23 +0200507 .update_time = ovl_update_time,
Miklos Szeredie9be9d52014-10-24 00:14:38 +0200508};
509
Miklos Szeredi9e142c42018-07-18 15:44:42 +0200510static const struct inode_operations ovl_special_inode_operations = {
511 .setattr = ovl_setattr,
512 .permission = ovl_permission,
513 .getattr = ovl_getattr,
514 .listxattr = ovl_listxattr,
515 .get_acl = ovl_get_acl,
516 .update_time = ovl_update_time,
517};
518
Wei Yongjun69383c52018-09-25 14:57:42 +0000519static const struct address_space_operations ovl_aops = {
Amir Goldstein5b910bd2018-08-27 15:56:00 +0300520 /* For O_DIRECT dentry_open() checks f_mapping->a_ops->direct_IO */
521 .direct_IO = noop_direct_IO,
522};
523
Amir Goldsteinb1eaa952017-01-11 12:07:46 +0200524/*
525 * It is possible to stack overlayfs instance on top of another
Chengguang Xua5a84682020-02-10 11:11:14 +0800526 * overlayfs instance as lower layer. We need to annotate the
Amir Goldsteinb1eaa952017-01-11 12:07:46 +0200527 * stackable i_mutex locks according to stack level of the super
528 * block instance. An overlayfs instance can never be in stack
529 * depth 0 (there is always a real fs below it). An overlayfs
530 * inode lock will use the lockdep annotaion ovl_i_mutex_key[depth].
531 *
532 * For example, here is a snip from /proc/lockdep_chains after
533 * dir_iterate of nested overlayfs:
534 *
535 * [...] &ovl_i_mutex_dir_key[depth] (stack_depth=2)
536 * [...] &ovl_i_mutex_dir_key[depth]#2 (stack_depth=1)
537 * [...] &type->i_mutex_dir_key (stack_depth=0)
Amir Goldsteinb1f9d382019-12-21 11:42:29 +0200538 *
539 * Locking order w.r.t ovl_want_write() is important for nested overlayfs.
540 *
541 * This chain is valid:
542 * - inode->i_rwsem (inode_lock[2])
543 * - upper_mnt->mnt_sb->s_writers (ovl_want_write[0])
544 * - OVL_I(inode)->lock (ovl_inode_lock[2])
545 * - OVL_I(lowerinode)->lock (ovl_inode_lock[1])
546 *
547 * And this chain is valid:
548 * - inode->i_rwsem (inode_lock[2])
549 * - OVL_I(inode)->lock (ovl_inode_lock[2])
550 * - lowerinode->i_rwsem (inode_lock[1])
551 * - OVL_I(lowerinode)->lock (ovl_inode_lock[1])
552 *
553 * But lowerinode->i_rwsem SHOULD NOT be acquired while ovl_want_write() is
554 * held, because it is in reverse order of the non-nested case using the same
555 * upper fs:
556 * - inode->i_rwsem (inode_lock[1])
557 * - upper_mnt->mnt_sb->s_writers (ovl_want_write[0])
558 * - OVL_I(inode)->lock (ovl_inode_lock[1])
Amir Goldsteinb1eaa952017-01-11 12:07:46 +0200559 */
560#define OVL_MAX_NESTING FILESYSTEM_MAX_STACK_DEPTH
561
562static inline void ovl_lockdep_annotate_inode_mutex_key(struct inode *inode)
563{
564#ifdef CONFIG_LOCKDEP
565 static struct lock_class_key ovl_i_mutex_key[OVL_MAX_NESTING];
566 static struct lock_class_key ovl_i_mutex_dir_key[OVL_MAX_NESTING];
Amir Goldstein4eae06d2017-10-27 15:44:08 +0300567 static struct lock_class_key ovl_i_lock_key[OVL_MAX_NESTING];
Amir Goldsteinb1eaa952017-01-11 12:07:46 +0200568
569 int depth = inode->i_sb->s_stack_depth - 1;
570
571 if (WARN_ON_ONCE(depth < 0 || depth >= OVL_MAX_NESTING))
572 depth = 0;
573
574 if (S_ISDIR(inode->i_mode))
575 lockdep_set_class(&inode->i_rwsem, &ovl_i_mutex_dir_key[depth]);
576 else
577 lockdep_set_class(&inode->i_rwsem, &ovl_i_mutex_key[depth]);
Amir Goldstein4eae06d2017-10-27 15:44:08 +0300578
579 lockdep_set_class(&OVL_I(inode)->lock, &ovl_i_lock_key[depth]);
Amir Goldsteinb1eaa952017-01-11 12:07:46 +0200580#endif
581}
582
Amir Goldstein4d314f72020-02-21 16:34:43 +0200583static void ovl_next_ino(struct inode *inode)
584{
585 struct ovl_fs *ofs = inode->i_sb->s_fs_info;
586
587 inode->i_ino = atomic_long_inc_return(&ofs->last_ino);
588 if (unlikely(!inode->i_ino))
589 inode->i_ino = atomic_long_inc_return(&ofs->last_ino);
590}
591
Amir Goldstein62c832e2019-11-19 15:31:46 +0200592static void ovl_map_ino(struct inode *inode, unsigned long ino, int fsid)
Miklos Szeredie9be9d52014-10-24 00:14:38 +0200593{
Amir Goldstein12574a92018-03-16 10:39:37 +0200594 int xinobits = ovl_xino_bits(inode->i_sb);
Amir Goldsteindfe51d42020-02-21 16:34:44 +0200595 unsigned int xinoshift = 64 - xinobits;
Amir Goldstein12574a92018-03-16 10:39:37 +0200596
Amir Goldstein695b46e2018-03-15 23:39:01 +0200597 /*
Amir Goldstein6dde1e42019-06-09 19:03:44 +0300598 * When d_ino is consistent with st_ino (samefs or i_ino has enough
599 * bits to encode layer), set the same value used for st_ino to i_ino,
600 * so inode number exposed via /proc/locks and a like will be
601 * consistent with d_ino and st_ino values. An i_ino value inconsistent
Amir Goldstein735c9072019-11-19 17:14:55 +0200602 * with d_ino also causes nfsd readdirplus to fail.
Amir Goldstein695b46e2018-03-15 23:39:01 +0200603 */
Amir Goldstein4d314f72020-02-21 16:34:43 +0200604 inode->i_ino = ino;
Amir Goldsteindfe51d42020-02-21 16:34:44 +0200605 if (ovl_same_fs(inode->i_sb)) {
606 return;
607 } else if (xinobits && likely(!(ino >> xinoshift))) {
608 inode->i_ino |= (unsigned long)fsid << (xinoshift + 1);
609 return;
610 }
611
612 /*
613 * For directory inodes on non-samefs with xino disabled or xino
614 * overflow, we allocate a non-persistent inode number, to be used for
615 * resolving st_ino collisions in ovl_map_dev_ino().
616 *
617 * To avoid ino collision with legitimate xino values from upper
618 * layer (fsid 0), use the lowest xinobit to map the non
619 * persistent inode numbers to the unified st_ino address space.
620 */
621 if (S_ISDIR(inode->i_mode)) {
Amir Goldstein4d314f72020-02-21 16:34:43 +0200622 ovl_next_ino(inode);
Amir Goldsteindfe51d42020-02-21 16:34:44 +0200623 if (xinobits) {
624 inode->i_ino &= ~0UL >> xinobits;
625 inode->i_ino |= 1UL << xinoshift;
626 }
Amir Goldstein12574a92018-03-16 10:39:37 +0200627 }
Amir Goldstein62c832e2019-11-19 15:31:46 +0200628}
629
630void ovl_inode_init(struct inode *inode, struct ovl_inode_params *oip,
631 unsigned long ino, int fsid)
632{
633 struct inode *realinode;
634
635 if (oip->upperdentry)
636 OVL_I(inode)->__upperdentry = oip->upperdentry;
637 if (oip->lowerpath && oip->lowerpath->dentry)
638 OVL_I(inode)->lower = igrab(d_inode(oip->lowerpath->dentry));
639 if (oip->lowerdata)
640 OVL_I(inode)->lowerdata = igrab(d_inode(oip->lowerdata));
641
642 realinode = ovl_inode_real(inode);
643 ovl_copyattr(realinode, inode);
644 ovl_copyflags(realinode, inode);
645 ovl_map_ino(inode, ino, fsid);
646}
647
648static void ovl_fill_inode(struct inode *inode, umode_t mode, dev_t rdev)
649{
Miklos Szeredie9be9d52014-10-24 00:14:38 +0200650 inode->i_mode = mode;
Miklos Szeredid719e8f2016-07-29 12:05:23 +0200651 inode->i_flags |= S_NOCMTIME;
Miklos Szeredi2a3a2a32016-09-01 11:11:59 +0200652#ifdef CONFIG_FS_POSIX_ACL
653 inode->i_acl = inode->i_default_acl = ACL_DONT_CACHE;
654#endif
Miklos Szeredie9be9d52014-10-24 00:14:38 +0200655
Amir Goldsteinb1eaa952017-01-11 12:07:46 +0200656 ovl_lockdep_annotate_inode_mutex_key(inode);
657
Miklos Szeredica4c8a32016-12-16 11:02:55 +0100658 switch (mode & S_IFMT) {
659 case S_IFREG:
660 inode->i_op = &ovl_file_inode_operations;
Miklos Szeredid1d04ef2018-07-18 15:44:41 +0200661 inode->i_fop = &ovl_file_operations;
Amir Goldstein5b910bd2018-08-27 15:56:00 +0300662 inode->i_mapping->a_ops = &ovl_aops;
Miklos Szeredica4c8a32016-12-16 11:02:55 +0100663 break;
664
Miklos Szeredie9be9d52014-10-24 00:14:38 +0200665 case S_IFDIR:
Miklos Szeredie9be9d52014-10-24 00:14:38 +0200666 inode->i_op = &ovl_dir_inode_operations;
667 inode->i_fop = &ovl_dir_operations;
668 break;
669
670 case S_IFLNK:
671 inode->i_op = &ovl_symlink_inode_operations;
672 break;
673
Miklos Szeredi51f7e522016-07-29 12:05:24 +0200674 default:
Miklos Szeredi9e142c42018-07-18 15:44:42 +0200675 inode->i_op = &ovl_special_inode_operations;
Miklos Szeredica4c8a32016-12-16 11:02:55 +0100676 init_special_inode(inode, mode, rdev);
Miklos Szeredie9be9d52014-10-24 00:14:38 +0200677 break;
Miklos Szeredi51f7e522016-07-29 12:05:24 +0200678 }
679}
Miklos Szeredie9be9d52014-10-24 00:14:38 +0200680
Amir Goldstein5f8415d2017-06-20 15:35:14 +0300681/*
682 * With inodes index enabled, an overlay inode nlink counts the union of upper
683 * hardlinks and non-covered lower hardlinks. During the lifetime of a non-pure
684 * upper inode, the following nlink modifying operations can happen:
685 *
686 * 1. Lower hardlink copy up
687 * 2. Upper hardlink created, unlinked or renamed over
688 * 3. Lower hardlink whiteout or renamed over
689 *
690 * For the first, copy up case, the union nlink does not change, whether the
691 * operation succeeds or fails, but the upper inode nlink may change.
692 * Therefore, before copy up, we store the union nlink value relative to the
693 * lower inode nlink in the index inode xattr trusted.overlay.nlink.
694 *
695 * For the second, upper hardlink case, the union nlink should be incremented
696 * or decremented IFF the operation succeeds, aligned with nlink change of the
697 * upper inode. Therefore, before link/unlink/rename, we store the union nlink
698 * value relative to the upper inode nlink in the index inode.
699 *
700 * For the last, lower cover up case, we simplify things by preceding the
701 * whiteout or cover up with copy up. This makes sure that there is an index
702 * upper inode where the nlink xattr can be stored before the copied up upper
703 * entry is unlink.
704 */
705#define OVL_NLINK_ADD_UPPER (1 << 0)
706
707/*
708 * On-disk format for indexed nlink:
709 *
710 * nlink relative to the upper inode - "U[+-]NUM"
711 * nlink relative to the lower inode - "L[+-]NUM"
712 */
713
714static int ovl_set_nlink_common(struct dentry *dentry,
715 struct dentry *realdentry, const char *format)
716{
717 struct inode *inode = d_inode(dentry);
718 struct inode *realinode = d_inode(realdentry);
719 char buf[13];
720 int len;
721
722 len = snprintf(buf, sizeof(buf), format,
723 (int) (inode->i_nlink - realinode->i_nlink));
724
Miklos Szeredi67873412017-07-27 21:54:05 +0200725 if (WARN_ON(len >= sizeof(buf)))
726 return -EIO;
727
Miklos Szeredi610afc02020-09-02 10:58:49 +0200728 return ovl_do_setxattr(OVL_FS(inode->i_sb), ovl_dentry_upper(dentry),
Miklos Szeredi26150ab2020-09-02 10:58:48 +0200729 OVL_XATTR_NLINK, buf, len);
Amir Goldstein5f8415d2017-06-20 15:35:14 +0300730}
731
732int ovl_set_nlink_upper(struct dentry *dentry)
733{
734 return ovl_set_nlink_common(dentry, ovl_dentry_upper(dentry), "U%+i");
735}
736
737int ovl_set_nlink_lower(struct dentry *dentry)
738{
739 return ovl_set_nlink_common(dentry, ovl_dentry_lower(dentry), "L%+i");
740}
741
Miklos Szeredi610afc02020-09-02 10:58:49 +0200742unsigned int ovl_get_nlink(struct ovl_fs *ofs, struct dentry *lowerdentry,
Amir Goldsteincaf70cb2017-06-21 13:46:12 +0300743 struct dentry *upperdentry,
744 unsigned int fallback)
Amir Goldstein5f8415d2017-06-20 15:35:14 +0300745{
746 int nlink_diff;
747 int nlink;
748 char buf[13];
749 int err;
750
751 if (!lowerdentry || !upperdentry || d_inode(lowerdentry)->i_nlink == 1)
752 return fallback;
753
Miklos Szeredi610afc02020-09-02 10:58:49 +0200754 err = ovl_do_getxattr(ofs, upperdentry, OVL_XATTR_NLINK,
Miklos Szeredid5dc7482020-09-02 10:58:48 +0200755 &buf, sizeof(buf) - 1);
Amir Goldstein5f8415d2017-06-20 15:35:14 +0300756 if (err < 0)
757 goto fail;
758
759 buf[err] = '\0';
760 if ((buf[0] != 'L' && buf[0] != 'U') ||
761 (buf[1] != '+' && buf[1] != '-'))
762 goto fail;
763
764 err = kstrtoint(buf + 1, 10, &nlink_diff);
765 if (err < 0)
766 goto fail;
767
768 nlink = d_inode(buf[0] == 'L' ? lowerdentry : upperdentry)->i_nlink;
769 nlink += nlink_diff;
770
771 if (nlink <= 0)
772 goto fail;
773
774 return nlink;
775
776fail:
lijiazi1bd0a3a2019-12-16 19:12:32 +0800777 pr_warn_ratelimited("failed to get index nlink (%pd2, err=%i)\n",
Amir Goldstein5f8415d2017-06-20 15:35:14 +0300778 upperdentry, err);
779 return fallback;
780}
781
Miklos Szeredica4c8a32016-12-16 11:02:55 +0100782struct inode *ovl_new_inode(struct super_block *sb, umode_t mode, dev_t rdev)
Miklos Szeredi51f7e522016-07-29 12:05:24 +0200783{
784 struct inode *inode;
785
786 inode = new_inode(sb);
787 if (inode)
Amir Goldstein62c832e2019-11-19 15:31:46 +0200788 ovl_fill_inode(inode, mode, rdev);
Miklos Szeredi51f7e522016-07-29 12:05:24 +0200789
790 return inode;
791}
792
793static int ovl_inode_test(struct inode *inode, void *data)
794{
Miklos Szeredi25b77132017-07-04 22:03:16 +0200795 return inode->i_private == data;
Miklos Szeredi51f7e522016-07-29 12:05:24 +0200796}
797
798static int ovl_inode_set(struct inode *inode, void *data)
799{
Miklos Szeredi25b77132017-07-04 22:03:16 +0200800 inode->i_private = data;
Miklos Szeredi51f7e522016-07-29 12:05:24 +0200801 return 0;
802}
803
Miklos Szeredib9ac5c272017-07-04 22:03:17 +0200804static bool ovl_verify_inode(struct inode *inode, struct dentry *lowerdentry,
Amir Goldstein4b91c302018-01-18 16:39:13 +0200805 struct dentry *upperdentry, bool strict)
Miklos Szeredib9ac5c272017-07-04 22:03:17 +0200806{
Amir Goldstein4b91c302018-01-18 16:39:13 +0200807 /*
808 * For directories, @strict verify from lookup path performs consistency
809 * checks, so NULL lower/upper in dentry must match NULL lower/upper in
810 * inode. Non @strict verify from NFS handle decode path passes NULL for
811 * 'unknown' lower/upper.
812 */
813 if (S_ISDIR(inode->i_mode) && strict) {
Amir Goldstein31747ed2018-01-14 18:35:40 +0200814 /* Real lower dir moved to upper layer under us? */
815 if (!lowerdentry && ovl_inode_lower(inode))
816 return false;
817
818 /* Lookup of an uncovered redirect origin? */
819 if (!upperdentry && ovl_inode_upper(inode))
820 return false;
821 }
822
Amir Goldstein939ae4e2017-09-11 16:30:15 +0300823 /*
824 * Allow non-NULL lower inode in ovl_inode even if lowerdentry is NULL.
825 * This happens when finding a copied up overlay inode for a renamed
826 * or hardlinked overlay dentry and lower dentry cannot be followed
827 * by origin because lower fs does not support file handles.
828 */
829 if (lowerdentry && ovl_inode_lower(inode) != d_inode(lowerdentry))
Miklos Szeredib9ac5c272017-07-04 22:03:17 +0200830 return false;
831
832 /*
833 * Allow non-NULL __upperdentry in inode even if upperdentry is NULL.
834 * This happens when finding a lower alias for a copied up hard link.
835 */
836 if (upperdentry && ovl_inode_upper(inode) != d_inode(upperdentry))
837 return false;
838
839 return true;
840}
841
Amir Goldstein4b91c302018-01-18 16:39:13 +0200842struct inode *ovl_lookup_inode(struct super_block *sb, struct dentry *real,
843 bool is_upper)
Amir Goldstein9436a1a2017-12-24 18:28:04 +0200844{
Amir Goldstein4b91c302018-01-18 16:39:13 +0200845 struct inode *inode, *key = d_inode(real);
Amir Goldstein9436a1a2017-12-24 18:28:04 +0200846
847 inode = ilookup5(sb, (unsigned long) key, ovl_inode_test, key);
848 if (!inode)
849 return NULL;
850
Amir Goldstein4b91c302018-01-18 16:39:13 +0200851 if (!ovl_verify_inode(inode, is_upper ? NULL : real,
852 is_upper ? real : NULL, false)) {
Amir Goldstein9436a1a2017-12-24 18:28:04 +0200853 iput(inode);
854 return ERR_PTR(-ESTALE);
855 }
856
857 return inode;
858}
859
Amir Goldstein146d62e2019-04-18 17:42:08 +0300860bool ovl_lookup_trap_inode(struct super_block *sb, struct dentry *dir)
861{
862 struct inode *key = d_inode(dir);
863 struct inode *trap;
864 bool res;
865
866 trap = ilookup5(sb, (unsigned long) key, ovl_inode_test, key);
867 if (!trap)
868 return false;
869
870 res = IS_DEADDIR(trap) && !ovl_inode_upper(trap) &&
871 !ovl_inode_lower(trap);
872
873 iput(trap);
874 return res;
875}
876
877/*
878 * Create an inode cache entry for layer root dir, that will intentionally
879 * fail ovl_verify_inode(), so any lookup that will find some layer root
880 * will fail.
881 */
882struct inode *ovl_get_trap_inode(struct super_block *sb, struct dentry *dir)
883{
884 struct inode *key = d_inode(dir);
885 struct inode *trap;
886
887 if (!d_is_dir(dir))
888 return ERR_PTR(-ENOTDIR);
889
890 trap = iget5_locked(sb, (unsigned long) key, ovl_inode_test,
891 ovl_inode_set, key);
892 if (!trap)
893 return ERR_PTR(-ENOMEM);
894
895 if (!(trap->i_state & I_NEW)) {
896 /* Conflicting layer roots? */
897 iput(trap);
898 return ERR_PTR(-ELOOP);
899 }
900
901 trap->i_mode = S_IFDIR;
902 trap->i_flags = S_DEAD;
903 unlock_new_inode(trap);
904
905 return trap;
906}
907
Amir Goldstein764baba2018-02-04 15:35:09 +0200908/*
909 * Does overlay inode need to be hashed by lower inode?
910 */
911static bool ovl_hash_bylower(struct super_block *sb, struct dentry *upper,
Miklos Szeredi74c6e382020-06-04 10:48:19 +0200912 struct dentry *lower, bool index)
Amir Goldstein764baba2018-02-04 15:35:09 +0200913{
914 struct ovl_fs *ofs = sb->s_fs_info;
915
916 /* No, if pure upper */
917 if (!lower)
918 return false;
919
920 /* Yes, if already indexed */
921 if (index)
922 return true;
923
924 /* Yes, if won't be copied up */
Miklos Szeredi08f4c7c2020-06-04 10:48:19 +0200925 if (!ovl_upper_mnt(ofs))
Amir Goldstein764baba2018-02-04 15:35:09 +0200926 return true;
927
928 /* No, if lower hardlink is or will be broken on copy up */
929 if ((upper || !ovl_indexdir(sb)) &&
930 !d_is_dir(lower) && d_inode(lower)->i_nlink > 1)
931 return false;
932
933 /* No, if non-indexed upper with NFS export */
934 if (sb->s_export_op && upper)
935 return false;
936
937 /* Otherwise, hash by lower inode for fsnotify */
938 return true;
939}
940
Amir Goldstein01b39dc2018-05-11 11:15:15 +0300941static struct inode *ovl_iget5(struct super_block *sb, struct inode *newinode,
942 struct inode *key)
943{
944 return newinode ? inode_insert5(newinode, (unsigned long) key,
945 ovl_inode_test, ovl_inode_set, key) :
946 iget5_locked(sb, (unsigned long) key,
947 ovl_inode_test, ovl_inode_set, key);
948}
949
Vivek Goyalac6a52e2018-05-08 09:27:21 -0400950struct inode *ovl_get_inode(struct super_block *sb,
951 struct ovl_inode_params *oip)
Miklos Szeredi51f7e522016-07-29 12:05:24 +0200952{
Miklos Szeredi610afc02020-09-02 10:58:49 +0200953 struct ovl_fs *ofs = OVL_FS(sb);
Vivek Goyalac6a52e2018-05-08 09:27:21 -0400954 struct dentry *upperdentry = oip->upperdentry;
955 struct ovl_path *lowerpath = oip->lowerpath;
Miklos Szeredi09d8b582017-07-04 22:03:16 +0200956 struct inode *realinode = upperdentry ? d_inode(upperdentry) : NULL;
Miklos Szeredi51f7e522016-07-29 12:05:24 +0200957 struct inode *inode;
Amir Goldstein12574a92018-03-16 10:39:37 +0200958 struct dentry *lowerdentry = lowerpath ? lowerpath->dentry : NULL;
Vivek Goyalac6a52e2018-05-08 09:27:21 -0400959 bool bylower = ovl_hash_bylower(sb, upperdentry, lowerdentry,
960 oip->index);
Amir Goldstein300b1242019-11-19 15:36:14 +0200961 int fsid = bylower ? lowerpath->layer->fsid : 0;
Vivek Goyal28166ab2020-06-01 11:56:52 -0400962 bool is_dir;
Amir Goldstein695b46e2018-03-15 23:39:01 +0200963 unsigned long ino = 0;
Amir Goldsteinacf30622019-03-28 17:38:29 +0200964 int err = oip->newinode ? -EEXIST : -ENOMEM;
Amir Goldstein6eaf0112017-10-12 19:03:04 +0300965
Miklos Szeredi09d8b582017-07-04 22:03:16 +0200966 if (!realinode)
967 realinode = d_inode(lowerdentry);
968
Amir Goldstein6eaf0112017-10-12 19:03:04 +0300969 /*
Amir Goldstein764baba2018-02-04 15:35:09 +0200970 * Copy up origin (lower) may exist for non-indexed upper, but we must
971 * not use lower as hash key if this is a broken hardlink.
Amir Goldstein6eaf0112017-10-12 19:03:04 +0300972 */
Amir Goldstein31747ed2018-01-14 18:35:40 +0200973 is_dir = S_ISDIR(realinode->i_mode);
Amir Goldstein764baba2018-02-04 15:35:09 +0200974 if (upperdentry || bylower) {
975 struct inode *key = d_inode(bylower ? lowerdentry :
976 upperdentry);
Amir Goldstein31747ed2018-01-14 18:35:40 +0200977 unsigned int nlink = is_dir ? 1 : realinode->i_nlink;
Miklos Szeredib9ac5c272017-07-04 22:03:17 +0200978
Amir Goldstein01b39dc2018-05-11 11:15:15 +0300979 inode = ovl_iget5(sb, oip->newinode, key);
Miklos Szeredi09d8b582017-07-04 22:03:16 +0200980 if (!inode)
Vivek Goyal027065b2018-05-11 11:49:28 -0400981 goto out_err;
Miklos Szeredi09d8b582017-07-04 22:03:16 +0200982 if (!(inode->i_state & I_NEW)) {
Miklos Szeredib9ac5c272017-07-04 22:03:17 +0200983 /*
984 * Verify that the underlying files stored in the inode
985 * match those in the dentry.
986 */
Amir Goldstein4b91c302018-01-18 16:39:13 +0200987 if (!ovl_verify_inode(inode, lowerdentry, upperdentry,
988 true)) {
Miklos Szeredib9ac5c272017-07-04 22:03:17 +0200989 iput(inode);
Vivek Goyal027065b2018-05-11 11:49:28 -0400990 err = -ESTALE;
991 goto out_err;
Miklos Szeredib9ac5c272017-07-04 22:03:17 +0200992 }
993
Miklos Szeredi09d8b582017-07-04 22:03:16 +0200994 dput(upperdentry);
Vivek Goyal9cec54c2018-05-11 11:49:27 -0400995 kfree(oip->redirect);
Miklos Szeredi09d8b582017-07-04 22:03:16 +0200996 goto out;
997 }
Miklos Szeredie9be9d52014-10-24 00:14:38 +0200998
Amir Goldstein31747ed2018-01-14 18:35:40 +0200999 /* Recalculate nlink for non-dir due to indexing */
1000 if (!is_dir)
Miklos Szeredi610afc02020-09-02 10:58:49 +02001001 nlink = ovl_get_nlink(ofs, lowerdentry, upperdentry,
1002 nlink);
Amir Goldstein5f8415d2017-06-20 15:35:14 +03001003 set_nlink(inode, nlink);
Amir Goldstein695b46e2018-03-15 23:39:01 +02001004 ino = key->i_ino;
Miklos Szeredie6d2ebd2017-07-04 22:03:16 +02001005 } else {
Amir Goldstein764baba2018-02-04 15:35:09 +02001006 /* Lower hardlink that will be broken on copy up */
Amir Goldstein0aceb532017-12-12 23:43:16 +02001007 inode = new_inode(sb);
Vivek Goyal027065b2018-05-11 11:49:28 -04001008 if (!inode) {
1009 err = -ENOMEM;
1010 goto out_err;
1011 }
Amir Goldstein300b1242019-11-19 15:36:14 +02001012 ino = realinode->i_ino;
1013 fsid = lowerpath->layer->fsid;
Miklos Szeredie6d2ebd2017-07-04 22:03:16 +02001014 }
Amir Goldstein62c832e2019-11-19 15:31:46 +02001015 ovl_fill_inode(inode, realinode->i_mode, realinode->i_rdev);
1016 ovl_inode_init(inode, oip, ino, fsid);
Miklos Szeredi13c72072017-07-04 22:03:16 +02001017
Miklos Szeredi610afc02020-09-02 10:58:49 +02001018 if (upperdentry && ovl_is_impuredir(sb, upperdentry))
Miklos Szeredi13c72072017-07-04 22:03:16 +02001019 ovl_set_flag(OVL_IMPURE, inode);
1020
Vivek Goyalac6a52e2018-05-08 09:27:21 -04001021 if (oip->index)
Vivek Goyal0471a9c2018-03-20 16:35:40 -04001022 ovl_set_flag(OVL_INDEX, inode);
1023
Vivek Goyal9cec54c2018-05-11 11:49:27 -04001024 OVL_I(inode)->redirect = oip->redirect;
1025
Vivek Goyala00c2d52018-05-11 11:49:32 -04001026 if (bylower)
1027 ovl_set_flag(OVL_CONST_INO, inode);
1028
Amir Goldsteinb79e05a2017-06-25 16:37:17 +03001029 /* Check for non-merge dir that may have whiteouts */
Amir Goldstein31747ed2018-01-14 18:35:40 +02001030 if (is_dir) {
Vivek Goyalac6a52e2018-05-08 09:27:21 -04001031 if (((upperdentry && lowerdentry) || oip->numlower > 1) ||
Miklos Szeredi610afc02020-09-02 10:58:49 +02001032 ovl_check_origin_xattr(ofs, upperdentry ?: lowerdentry)) {
Amir Goldsteinb79e05a2017-06-25 16:37:17 +03001033 ovl_set_flag(OVL_WHITEOUTS, inode);
1034 }
1035 }
1036
Miklos Szeredie6d2ebd2017-07-04 22:03:16 +02001037 if (inode->i_state & I_NEW)
1038 unlock_new_inode(inode);
1039out:
Miklos Szeredie9be9d52014-10-24 00:14:38 +02001040 return inode;
Miklos Szeredib9ac5c272017-07-04 22:03:17 +02001041
Vivek Goyal027065b2018-05-11 11:49:28 -04001042out_err:
lijiazi1bd0a3a2019-12-16 19:12:32 +08001043 pr_warn_ratelimited("failed to get inode (%i)\n", err);
Vivek Goyal027065b2018-05-11 11:49:28 -04001044 inode = ERR_PTR(err);
Miklos Szeredib9ac5c272017-07-04 22:03:17 +02001045 goto out;
Miklos Szeredie9be9d52014-10-24 00:14:38 +02001046}