blk-mq: don't count completed flush data request as inflight in case of quiesce
[linux-block.git] / fs / nsfs.c
CommitLineData
b2441318 1// SPDX-License-Identifier: GPL-2.0
e149ed2b 2#include <linux/mount.h>
059b20d9 3#include <linux/pseudo_fs.h>
e149ed2b
AV
4#include <linux/file.h>
5#include <linux/fs.h>
7bebd69e 6#include <linux/proc_fs.h>
e149ed2b
AV
7#include <linux/proc_ns.h>
8#include <linux/magic.h>
9#include <linux/ktime.h>
75509fd8 10#include <linux/seq_file.h>
6786741d
AV
11#include <linux/user_namespace.h>
12#include <linux/nsfs.h>
d95fa3c7 13#include <linux/uaccess.h>
e149ed2b 14
7bebd69e
EB
15#include "internal.h"
16
e149ed2b
AV
17static struct vfsmount *nsfs_mnt;
18
6786741d
AV
19static long ns_ioctl(struct file *filp, unsigned int ioctl,
20 unsigned long arg);
e149ed2b
AV
21static const struct file_operations ns_file_operations = {
22 .llseek = no_llseek,
6786741d 23 .unlocked_ioctl = ns_ioctl,
1cb925c0 24 .compat_ioctl = compat_ptr_ioctl,
e149ed2b
AV
25};
26
27static char *ns_dname(struct dentry *dentry, char *buffer, int buflen)
28{
75c3cfa8 29 struct inode *inode = d_inode(dentry);
e149ed2b
AV
30 const struct proc_ns_operations *ns_ops = dentry->d_fsdata;
31
0f60d288 32 return dynamic_dname(buffer, buflen, "%s:[%lu]",
e149ed2b
AV
33 ns_ops->name, inode->i_ino);
34}
35
36static void ns_prune_dentry(struct dentry *dentry)
37{
75c3cfa8 38 struct inode *inode = d_inode(dentry);
e149ed2b
AV
39 if (inode) {
40 struct ns_common *ns = inode->i_private;
41 atomic_long_set(&ns->stashed, 0);
42 }
43}
44
45const struct dentry_operations ns_dentry_operations =
46{
47 .d_prune = ns_prune_dentry,
48 .d_delete = always_delete_dentry,
49 .d_dname = ns_dname,
50};
51
52static void nsfs_evict(struct inode *inode)
53{
54 struct ns_common *ns = inode->i_private;
55 clear_inode(inode);
56 ns->ops->put(ns);
57}
58
ce623f89 59static int __ns_get_path(struct path *path, struct ns_common *ns)
e149ed2b 60{
213b067c 61 struct vfsmount *mnt = nsfs_mnt;
e149ed2b
AV
62 struct dentry *dentry;
63 struct inode *inode;
e149ed2b
AV
64 unsigned long d;
65
e149ed2b
AV
66 rcu_read_lock();
67 d = atomic_long_read(&ns->stashed);
68 if (!d)
69 goto slow;
70 dentry = (struct dentry *)d;
71 if (!lockref_get_not_dead(&dentry->d_lockref))
72 goto slow;
73 rcu_read_unlock();
6786741d 74 ns->ops->put(ns);
e149ed2b 75got_it:
213b067c 76 path->mnt = mntget(mnt);
e149ed2b 77 path->dentry = dentry;
ce623f89 78 return 0;
e149ed2b
AV
79slow:
80 rcu_read_unlock();
81 inode = new_inode_pseudo(mnt->mnt_sb);
82 if (!inode) {
6786741d 83 ns->ops->put(ns);
ce623f89 84 return -ENOMEM;
e149ed2b
AV
85 }
86 inode->i_ino = ns->inum;
16a94965 87 simple_inode_init_ts(inode);
e149ed2b
AV
88 inode->i_flags |= S_IMMUTABLE;
89 inode->i_mode = S_IFREG | S_IRUGO;
90 inode->i_fop = &ns_file_operations;
91 inode->i_private = ns;
92
5467a68c 93 dentry = d_alloc_anon(mnt->mnt_sb);
e149ed2b
AV
94 if (!dentry) {
95 iput(inode);
ce623f89 96 return -ENOMEM;
e149ed2b
AV
97 }
98 d_instantiate(dentry, inode);
6786741d 99 dentry->d_fsdata = (void *)ns->ops;
e149ed2b
AV
100 d = atomic_long_cmpxchg(&ns->stashed, 0, (unsigned long)dentry);
101 if (d) {
102 d_delete(dentry); /* make sure ->d_prune() does nothing */
103 dput(dentry);
104 cpu_relax();
ce623f89 105 return -EAGAIN;
e149ed2b
AV
106 }
107 goto got_it;
108}
109
ce623f89 110int ns_get_path_cb(struct path *path, ns_get_path_helper_t *ns_get_cb,
cdab6ba8 111 void *private_data)
6786741d 112{
ce623f89 113 int ret;
6786741d 114
357ab5b5
AV
115 do {
116 struct ns_common *ns = ns_get_cb(private_data);
117 if (!ns)
ce623f89 118 return -ENOENT;
357ab5b5 119 ret = __ns_get_path(path, ns);
ce623f89 120 } while (ret == -EAGAIN);
6786741d 121
6786741d
AV
122 return ret;
123}
124
cdab6ba8
JK
125struct ns_get_path_task_args {
126 const struct proc_ns_operations *ns_ops;
127 struct task_struct *task;
128};
129
130static struct ns_common *ns_get_path_task(void *private_data)
131{
132 struct ns_get_path_task_args *args = private_data;
133
134 return args->ns_ops->get(args->task);
135}
136
ce623f89 137int ns_get_path(struct path *path, struct task_struct *task,
cdab6ba8
JK
138 const struct proc_ns_operations *ns_ops)
139{
140 struct ns_get_path_task_args args = {
141 .ns_ops = ns_ops,
142 .task = task,
143 };
144
145 return ns_get_path_cb(path, ns_get_path_task, &args);
146}
147
c62cce2c 148int open_related_ns(struct ns_common *ns,
6786741d
AV
149 struct ns_common *(*get_ns)(struct ns_common *ns))
150{
151 struct path path = {};
152 struct file *f;
ce623f89 153 int err;
6786741d
AV
154 int fd;
155
156 fd = get_unused_fd_flags(O_CLOEXEC);
157 if (fd < 0)
158 return fd;
159
357ab5b5 160 do {
6786741d
AV
161 struct ns_common *relative;
162
163 relative = get_ns(ns);
164 if (IS_ERR(relative)) {
165 put_unused_fd(fd);
166 return PTR_ERR(relative);
167 }
168
169 err = __ns_get_path(&path, relative);
ce623f89 170 } while (err == -EAGAIN);
357ab5b5 171
ce623f89 172 if (err) {
6786741d 173 put_unused_fd(fd);
ce623f89 174 return err;
6786741d
AV
175 }
176
177 f = dentry_open(&path, O_RDONLY, current_cred());
178 path_put(&path);
179 if (IS_ERR(f)) {
180 put_unused_fd(fd);
181 fd = PTR_ERR(f);
182 } else
183 fd_install(fd, f);
184
185 return fd;
186}
24dce080 187EXPORT_SYMBOL_GPL(open_related_ns);
6786741d
AV
188
189static long ns_ioctl(struct file *filp, unsigned int ioctl,
190 unsigned long arg)
191{
d95fa3c7 192 struct user_namespace *user_ns;
6786741d 193 struct ns_common *ns = get_proc_ns(file_inode(filp));
d95fa3c7
MK
194 uid_t __user *argp;
195 uid_t uid;
6786741d
AV
196
197 switch (ioctl) {
198 case NS_GET_USERNS:
199 return open_related_ns(ns, ns_get_owner);
a7306ed8
AV
200 case NS_GET_PARENT:
201 if (!ns->ops->get_parent)
202 return -EINVAL;
203 return open_related_ns(ns, ns->ops->get_parent);
e5ff5ce6
MK
204 case NS_GET_NSTYPE:
205 return ns->ops->type;
d95fa3c7
MK
206 case NS_GET_OWNER_UID:
207 if (ns->ops->type != CLONE_NEWUSER)
208 return -EINVAL;
209 user_ns = container_of(ns, struct user_namespace, ns);
210 argp = (uid_t __user *) arg;
211 uid = from_kuid_munged(current_user_ns(), user_ns->owner);
212 return put_user(uid, argp);
6786741d
AV
213 default:
214 return -ENOTTY;
215 }
216}
217
e149ed2b
AV
218int ns_get_name(char *buf, size_t size, struct task_struct *task,
219 const struct proc_ns_operations *ns_ops)
220{
221 struct ns_common *ns;
222 int res = -ENOENT;
25b14e92 223 const char *name;
e149ed2b
AV
224 ns = ns_ops->get(task);
225 if (ns) {
25b14e92
KT
226 name = ns_ops->real_ns_name ? : ns_ops->name;
227 res = snprintf(buf, size, "%s:[%u]", name, ns->inum);
e149ed2b
AV
228 ns_ops->put(ns);
229 }
230 return res;
231}
232
303cc571
CB
233bool proc_ns_file(const struct file *file)
234{
235 return file->f_op == &ns_file_operations;
236}
237
1e2328e7
CN
238/**
239 * ns_match() - Returns true if current namespace matches dev/ino provided.
39ecb653 240 * @ns: current namespace
1e2328e7
CN
241 * @dev: dev_t from nsfs that will be matched against current nsfs
242 * @ino: ino_t from nsfs that will be matched against current nsfs
243 *
244 * Return: true if dev and ino matches the current nsfs.
245 */
246bool ns_match(const struct ns_common *ns, dev_t dev, ino_t ino)
247{
248 return (ns->inum == ino) && (nsfs_mnt->mnt_sb->s_dev == dev);
249}
250
251
75509fd8
EB
252static int nsfs_show_path(struct seq_file *seq, struct dentry *dentry)
253{
254 struct inode *inode = d_inode(dentry);
255 const struct proc_ns_operations *ns_ops = dentry->d_fsdata;
256
6798a8ca
JP
257 seq_printf(seq, "%s:[%lu]", ns_ops->name, inode->i_ino);
258 return 0;
75509fd8
EB
259}
260
e149ed2b
AV
261static const struct super_operations nsfs_ops = {
262 .statfs = simple_statfs,
263 .evict_inode = nsfs_evict,
75509fd8 264 .show_path = nsfs_show_path,
e149ed2b 265};
059b20d9
DH
266
267static int nsfs_init_fs_context(struct fs_context *fc)
e149ed2b 268{
059b20d9
DH
269 struct pseudo_fs_context *ctx = init_pseudo(fc, NSFS_MAGIC);
270 if (!ctx)
271 return -ENOMEM;
272 ctx->ops = &nsfs_ops;
273 ctx->dops = &ns_dentry_operations;
274 return 0;
e149ed2b 275}
059b20d9 276
e149ed2b
AV
277static struct file_system_type nsfs = {
278 .name = "nsfs",
059b20d9 279 .init_fs_context = nsfs_init_fs_context,
e149ed2b
AV
280 .kill_sb = kill_anon_super,
281};
282
283void __init nsfs_init(void)
284{
285 nsfs_mnt = kern_mount(&nsfs);
286 if (IS_ERR(nsfs_mnt))
287 panic("can't set nsfs up\n");
1751e8a6 288 nsfs_mnt->mnt_sb->s_flags &= ~SB_NOUSER;
e149ed2b 289}