projects/linux·blob master · refs

426 lines · 10.7 KiB

1// SPDX-License-Identifier: GPL-2.0-only
2/*
3 * linux/fs/proc/net.c
4 *
5 * Copyright (C) 2007
6 *
7 * Author: Eric Biederman <[email protected]>
8 *
9 * proc net directory handling functions
10 */
11#include <linux/errno.h>
12#include <linux/time.h>
13#include <linux/proc_fs.h>
14#include <linux/stat.h>
15#include <linux/slab.h>
16#include <linux/init.h>
17#include <linux/sched.h>
18#include <linux/sched/task.h>
19#include <linux/module.h>
20#include <linux/bitops.h>
21#include <linux/mount.h>
22#include <linux/nsproxy.h>
23#include <linux/uidgid.h>
24#include <net/net_namespace.h>
25#include <linux/seq_file.h>
26#include <linux/security.h>
27
28#include "internal.h"
29
30static inline struct net *PDE_NET(struct proc_dir_entry *pde)
31{
32 return pde->parent->data;
33}
34
35static struct net *get_proc_net(const struct inode *inode)
36{
37 return maybe_get_net(PDE_NET(PDE(inode)));
38}
39
40static int seq_open_net(struct inode *inode, struct file *file)
41{
42 unsigned int state_size = PDE(inode)->state_size;
43 struct seq_net_private *p;
44 struct net *net;
45
46 WARN_ON_ONCE(state_size < sizeof(*p));
47
48 if (file->f_mode & FMODE_WRITE && !PDE(inode)->write)
49 return -EACCES;
50
51 net = get_proc_net(inode);
52 if (!net)
53 return -ENXIO;
54
55 p = __seq_open_private(file, PDE(inode)->seq_ops, state_size);
56 if (!p) {
57 put_net(net);
58 return -ENOMEM;
59 }
60#ifdef CONFIG_NET_NS
61 p->net = net;
62 netns_tracker_alloc(net, &p->ns_tracker, GFP_KERNEL);
63#endif
64 return 0;
65}
66
67static void seq_file_net_put_net(struct seq_file *seq)
68{
69#ifdef CONFIG_NET_NS
70 struct seq_net_private *priv = seq->private;
71
72 put_net_track(priv->net, &priv->ns_tracker);
73#else
74 put_net(&init_net);
75#endif
76}
77
78static int seq_release_net(struct inode *ino, struct file *f)
79{
80 struct seq_file *seq = f->private_data;
81
82 seq_file_net_put_net(seq);
83 seq_release_private(ino, f);
84 return 0;
85}
86
87static const struct proc_ops proc_net_seq_ops = {
88 .proc_open = seq_open_net,
89 .proc_read = seq_read,
90 .proc_write = proc_simple_write,
91 .proc_lseek = seq_lseek,
92 .proc_release = seq_release_net,
93};
94
95int bpf_iter_init_seq_net(void *priv_data, struct bpf_iter_aux_info *aux)
96{
97#ifdef CONFIG_NET_NS
98 struct seq_net_private *p = priv_data;
99
100 p->net = get_net_track(current->nsproxy->net_ns, &p->ns_tracker,
101 GFP_KERNEL);
102#endif
103 return 0;
104}
105
106void bpf_iter_fini_seq_net(void *priv_data)
107{
108#ifdef CONFIG_NET_NS
109 struct seq_net_private *p = priv_data;
110
111 put_net_track(p->net, &p->ns_tracker);
112#endif
113}
114
115struct proc_dir_entry *proc_create_net_data(const char *name, umode_t mode,
116 struct proc_dir_entry *parent, const struct seq_operations *ops,
117 unsigned int state_size, void *data)
118{
119 struct proc_dir_entry *p;
120
121 p = proc_create_reg(name, mode, &parent, data);
122 if (!p)
123 return NULL;
124 pde_force_lookup(p);
125 p->proc_ops = &proc_net_seq_ops;
126 p->seq_ops = ops;
127 p->state_size = state_size;
128 return proc_register(parent, p);
129}
130EXPORT_SYMBOL_GPL(proc_create_net_data);
131
132/**
133 * proc_create_net_data_write - Create a writable net_ns-specific proc file
134 * @name: The name of the file.
135 * @mode: The file's access mode.
136 * @parent: The parent directory in which to create.
137 * @ops: The seq_file ops with which to read the file.
138 * @write: The write method with which to 'modify' the file.
139 * @state_size: The size of the per-file private state to allocate.
140 * @data: Data for retrieval by pde_data().
141 *
142 * Create a network namespaced proc file in the @parent directory with the
143 * specified @name and @mode that allows reading of a file that displays a
144 * series of elements and also provides for the file accepting writes that have
145 * some arbitrary effect.
146 *
147 * The functions in the @ops table are used to iterate over items to be
148 * presented and extract the readable content using the seq_file interface.
149 *
150 * The @write function is called with the data copied into a kernel space
151 * scratch buffer and has a NUL appended for convenience. The buffer may be
152 * modified by the @write function. @write should return 0 on success.
153 *
154 * The @data value is accessible from the @show and @write functions by calling
155 * pde_data() on the file inode. The network namespace must be accessed by
156 * calling seq_file_net() on the seq_file struct.
157 */
158struct proc_dir_entry *proc_create_net_data_write(const char *name, umode_t mode,
159 struct proc_dir_entry *parent,
160 const struct seq_operations *ops,
161 proc_write_t write,
162 unsigned int state_size, void *data)
163{
164 struct proc_dir_entry *p;
165
166 p = proc_create_reg(name, mode, &parent, data);
167 if (!p)
168 return NULL;
169 pde_force_lookup(p);
170 p->proc_ops = &proc_net_seq_ops;
171 p->seq_ops = ops;
172 p->state_size = state_size;
173 p->write = write;
174 return proc_register(parent, p);
175}
176EXPORT_SYMBOL_GPL(proc_create_net_data_write);
177
178static int single_open_net(struct inode *inode, struct file *file)
179{
180 struct proc_dir_entry *de = PDE(inode);
181 struct net *net;
182 int err;
183
184 net = get_proc_net(inode);
185 if (!net)
186 return -ENXIO;
187
188 err = single_open(file, de->single_show, net);
189 if (err)
190 put_net(net);
191 return err;
192}
193
194static int single_release_net(struct inode *ino, struct file *f)
195{
196 struct seq_file *seq = f->private_data;
197 put_net(seq->private);
198 return single_release(ino, f);
199}
200
201static const struct proc_ops proc_net_single_ops = {
202 .proc_open = single_open_net,
203 .proc_read = seq_read,
204 .proc_write = proc_simple_write,
205 .proc_lseek = seq_lseek,
206 .proc_release = single_release_net,
207};
208
209struct proc_dir_entry *proc_create_net_single(const char *name, umode_t mode,
210 struct proc_dir_entry *parent,
211 int (*show)(struct seq_file *, void *), void *data)
212{
213 struct proc_dir_entry *p;
214
215 p = proc_create_reg(name, mode, &parent, data);
216 if (!p)
217 return NULL;
218 pde_force_lookup(p);
219 p->proc_ops = &proc_net_single_ops;
220 p->single_show = show;
221 return proc_register(parent, p);
222}
223EXPORT_SYMBOL_GPL(proc_create_net_single);
224
225/**
226 * proc_create_net_single_write - Create a writable net_ns-specific proc file
227 * @name: The name of the file.
228 * @mode: The file's access mode.
229 * @parent: The parent directory in which to create.
230 * @show: The seqfile show method with which to read the file.
231 * @write: The write method with which to 'modify' the file.
232 * @data: Data for retrieval by pde_data().
233 *
234 * Create a network-namespaced proc file in the @parent directory with the
235 * specified @name and @mode that allows reading of a file that displays a
236 * single element rather than a series and also provides for the file accepting
237 * writes that have some arbitrary effect.
238 *
239 * The @show function is called to extract the readable content via the
240 * seq_file interface.
241 *
242 * The @write function is called with the data copied into a kernel space
243 * scratch buffer and has a NUL appended for convenience. The buffer may be
244 * modified by the @write function. @write should return 0 on success.
245 *
246 * The @data value is accessible from the @show and @write functions by calling
247 * pde_data() on the file inode. The network namespace must be accessed by
248 * calling seq_file_single_net() on the seq_file struct.
249 */
250struct proc_dir_entry *proc_create_net_single_write(const char *name, umode_t mode,
251 struct proc_dir_entry *parent,
252 int (*show)(struct seq_file *, void *),
253 proc_write_t write,
254 void *data)
255{
256 struct proc_dir_entry *p;
257
258 p = proc_create_reg(name, mode, &parent, data);
259 if (!p)
260 return NULL;
261 pde_force_lookup(p);
262 p->proc_ops = &proc_net_single_ops;
263 p->single_show = show;
264 p->write = write;
265 return proc_register(parent, p);
266}
267EXPORT_SYMBOL_GPL(proc_create_net_single_write);
268
269static struct net *get_proc_task_net(struct inode *dir)
270{
271 struct task_struct *task;
272 struct nsproxy *ns;
273 struct net *net = NULL;
274 struct proc_fs_info *fs_info = proc_sb_info(dir->i_sb);
275
276 rcu_read_lock();
277 task = pid_task(proc_pid(dir), PIDTYPE_PID);
278 if (task != NULL) {
279 task_lock(task);
280 ns = task->nsproxy;
281 if (ns != NULL)
282 net = get_net(ns->net_ns);
283 task_unlock(task);
284 }
285 rcu_read_unlock();
286
287 if (net && (fs_info->pidonly == PROC_PIDONLY_ON) &&
288 security_capable(fs_info->mounter_cred, net->user_ns, CAP_NET_ADMIN, CAP_OPT_NONE) < 0) {
289 put_net(net);
290 net = NULL;
291 }
292
293 return net;
294}
295
296static struct dentry *proc_tgid_net_lookup(struct inode *dir,
297 struct dentry *dentry, unsigned int flags)
298{
299 struct dentry *de;
300 struct net *net;
301
302 de = ERR_PTR(-ENOENT);
303 net = get_proc_task_net(dir);
304 if (net != NULL) {
305 de = proc_lookup_de(dir, dentry, net->proc_net);
306 put_net(net);
307 }
308 return de;
309}
310
311static int proc_tgid_net_getattr(const struct mnt_idmap *idmap,
312 const struct path *path, struct kstat *stat,
313 u32 request_mask, unsigned int query_flags)
314{
315 struct inode *inode = d_inode(path->dentry);
316 struct net *net;
317
318 net = get_proc_task_net(inode);
319
320 generic_fillattr(&nop_mnt_idmap, request_mask, inode, stat);
321
322 if (net != NULL) {
323 stat->nlink = net->proc_net->nlink;
324 put_net(net);
325 }
326
327 return 0;
328}
329
330const struct inode_operations proc_net_inode_operations = {
331 .lookup = proc_tgid_net_lookup,
332 .getattr = proc_tgid_net_getattr,
333 .setattr = proc_nochmod_setattr,
334};
335
336static int proc_tgid_net_readdir(struct file *file, struct dir_context *ctx)
337{
338 int ret;
339 struct net *net;
340
341 ret = -EINVAL;
342 net = get_proc_task_net(file_inode(file));
343 if (net != NULL) {
344 ret = proc_readdir_de(file, ctx, net->proc_net);
345 put_net(net);
346 }
347 return ret;
348}
349
350const struct file_operations proc_net_operations = {
351 .llseek = generic_file_llseek,
352 .read = generic_read_dir,
353 .iterate_shared = proc_tgid_net_readdir,
354};
355
356static __net_init int proc_net_ns_init(struct net *net)
357{
358 struct proc_dir_entry *netd, *net_statd;
359 kuid_t uid;
360 kgid_t gid;
361 int err;
362
363 /*
364 * This PDE acts only as an anchor for /proc/${pid}/net hierarchy.
365 * Corresponding inode (PDE(inode) == net->proc_net) is never
366 * instantiated therefore blanket zeroing is fine.
367 * net->proc_net_stat inode is instantiated normally.
368 */
369 err = -ENOMEM;
370 netd = kmem_cache_zalloc(proc_dir_entry_cache, GFP_KERNEL);
371 if (!netd)
372 goto out;
373
374 netd->subdir = RB_ROOT;
375 netd->data = net;
376 netd->nlink = 2;
377 netd->namelen = 3;
378 netd->parent = &proc_root;
379 netd->name = netd->inline_name;
380 memcpy(netd->name, "net", 4);
381
382 uid = make_kuid(net->user_ns, 0);
383 if (!uid_valid(uid))
384 uid = netd->uid;
385
386 gid = make_kgid(net->user_ns, 0);
387 if (!gid_valid(gid))
388 gid = netd->gid;
389
390 proc_set_user(netd, uid, gid);
391
392 /* Seed dentry revalidation for /proc/${pid}/net */
393 pde_force_lookup(netd);
394
395 err = -EEXIST;
396 net_statd = proc_net_mkdir(net, "stat", netd);
397 if (!net_statd)
398 goto free_net;
399
400 net->proc_net = netd;
401 net->proc_net_stat = net_statd;
402 return 0;
403
404free_net:
405 pde_free(netd);
406out:
407 return err;
408}
409
410static __net_exit void proc_net_ns_exit(struct net *net)
411{
412 remove_proc_entry("stat", net->proc_net);
413 pde_free(net->proc_net);
414}
415
416static struct pernet_operations __net_initdata proc_net_ns_ops = {
417 .init = proc_net_ns_init,
418 .exit = proc_net_ns_exit,
419};
420
421int __init proc_net_init(void)
422{
423 proc_symlink("net", NULL, "self/net");
424
425 return register_pernet_subsys(&proc_net_ns_ops);
426}