xref: /linux/fs/proc/inode.c (revision bd628c1bed7902ec1f24ba0fe70758949146abbe) 
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  *  linux/fs/proc/inode.c
4  *
5  *  Copyright (C) 1991, 1992  Linus Torvalds
6  */
7 
8 #include <linux/cache.h>
9 #include <linux/time.h>
10 #include <linux/proc_fs.h>
11 #include <linux/kernel.h>
12 #include <linux/pid_namespace.h>
13 #include <linux/mm.h>
14 #include <linux/string.h>
15 #include <linux/stat.h>
16 #include <linux/completion.h>
17 #include <linux/poll.h>
18 #include <linux/printk.h>
19 #include <linux/file.h>
20 #include <linux/limits.h>
21 #include <linux/init.h>
22 #include <linux/module.h>
23 #include <linux/sysctl.h>
24 #include <linux/seq_file.h>
25 #include <linux/slab.h>
26 #include <linux/mount.h>
27 #include <linux/magic.h>
28 
29 #include <linux/uaccess.h>
30 
31 #include "internal.h"
32 
33 static void proc_evict_inode(struct inode *inode)
34 {
35 	struct proc_dir_entry *de;
36 	struct ctl_table_header *head;
37 
38 	truncate_inode_pages_final(&inode->i_data);
39 	clear_inode(inode);
40 
41 	/* Stop tracking associated processes */
42 	put_pid(PROC_I(inode)->pid);
43 
44 	/* Let go of any associated proc directory entry */
45 	de = PDE(inode);
46 	if (de)
47 		pde_put(de);
48 
49 	head = PROC_I(inode)->sysctl;
50 	if (head) {
51 		RCU_INIT_POINTER(PROC_I(inode)->sysctl, NULL);
52 		proc_sys_evict_inode(inode, head);
53 	}
54 }
55 
56 static struct kmem_cache *proc_inode_cachep __ro_after_init;
57 static struct kmem_cache *pde_opener_cache __ro_after_init;
58 
59 static struct inode *proc_alloc_inode(struct super_block *sb)
60 {
61 	struct proc_inode *ei;
62 
63 	ei = kmem_cache_alloc(proc_inode_cachep, GFP_KERNEL);
64 	if (!ei)
65 		return NULL;
66 	ei->pid = NULL;
67 	ei->fd = 0;
68 	ei->op.proc_get_link = NULL;
69 	ei->pde = NULL;
70 	ei->sysctl = NULL;
71 	ei->sysctl_entry = NULL;
72 	ei->ns_ops = NULL;
73 	return &ei->vfs_inode;
74 }
75 
76 static void proc_i_callback(struct rcu_head *head)
77 {
78 	struct inode *inode = container_of(head, struct inode, i_rcu);
79 	kmem_cache_free(proc_inode_cachep, PROC_I(inode));
80 }
81 
82 static void proc_destroy_inode(struct inode *inode)
83 {
84 	call_rcu(&inode->i_rcu, proc_i_callback);
85 }
86 
87 static void init_once(void *foo)
88 {
89 	struct proc_inode *ei = (struct proc_inode *) foo;
90 
91 	inode_init_once(&ei->vfs_inode);
92 }
93 
94 void __init proc_init_kmemcache(void)
95 {
96 	proc_inode_cachep = kmem_cache_create("proc_inode_cache",
97 					     sizeof(struct proc_inode),
98 					     0, (SLAB_RECLAIM_ACCOUNT|
99 						SLAB_MEM_SPREAD|SLAB_ACCOUNT|
100 						SLAB_PANIC),
101 					     init_once);
102 	pde_opener_cache =
103 		kmem_cache_create("pde_opener", sizeof(struct pde_opener), 0,
104 				  SLAB_ACCOUNT|SLAB_PANIC, NULL);
105 	proc_dir_entry_cache = kmem_cache_create_usercopy(
106 		"proc_dir_entry", SIZEOF_PDE, 0, SLAB_PANIC,
107 		offsetof(struct proc_dir_entry, inline_name),
108 		SIZEOF_PDE_INLINE_NAME, NULL);
109 	BUILD_BUG_ON(sizeof(struct proc_dir_entry) >= SIZEOF_PDE);
110 }
111 
112 static int proc_show_options(struct seq_file *seq, struct dentry *root)
113 {
114 	struct super_block *sb = root->d_sb;
115 	struct pid_namespace *pid = sb->s_fs_info;
116 
117 	if (!gid_eq(pid->pid_gid, GLOBAL_ROOT_GID))
118 		seq_printf(seq, ",gid=%u", from_kgid_munged(&init_user_ns, pid->pid_gid));
119 	if (pid->hide_pid != HIDEPID_OFF)
120 		seq_printf(seq, ",hidepid=%u", pid->hide_pid);
121 
122 	return 0;
123 }
124 
125 static const struct super_operations proc_sops = {
126 	.alloc_inode	= proc_alloc_inode,
127 	.destroy_inode	= proc_destroy_inode,
128 	.drop_inode	= generic_delete_inode,
129 	.evict_inode	= proc_evict_inode,
130 	.statfs		= simple_statfs,
131 	.remount_fs	= proc_remount,
132 	.show_options	= proc_show_options,
133 };
134 
135 enum {BIAS = -1U<<31};
136 
137 static inline int use_pde(struct proc_dir_entry *pde)
138 {
139 	return likely(atomic_inc_unless_negative(&pde->in_use));
140 }
141 
142 static void unuse_pde(struct proc_dir_entry *pde)
143 {
144 	if (unlikely(atomic_dec_return(&pde->in_use) == BIAS))
145 		complete(pde->pde_unload_completion);
146 }
147 
148 /* pde is locked on entry, unlocked on exit */
149 static void close_pdeo(struct proc_dir_entry *pde, struct pde_opener *pdeo)
150 {
151 	/*
152 	 * close() (proc_reg_release()) can't delete an entry and proceed:
153 	 * ->release hook needs to be available at the right moment.
154 	 *
155 	 * rmmod (remove_proc_entry() et al) can't delete an entry and proceed:
156 	 * "struct file" needs to be available at the right moment.
157 	 *
158 	 * Therefore, first process to enter this function does ->release() and
159 	 * signals its completion to the other process which does nothing.
160 	 */
161 	if (pdeo->closing) {
162 		/* somebody else is doing that, just wait */
163 		DECLARE_COMPLETION_ONSTACK(c);
164 		pdeo->c = &c;
165 		spin_unlock(&pde->pde_unload_lock);
166 		wait_for_completion(&c);
167 	} else {
168 		struct file *file;
169 		struct completion *c;
170 
171 		pdeo->closing = true;
172 		spin_unlock(&pde->pde_unload_lock);
173 		file = pdeo->file;
174 		pde->proc_fops->release(file_inode(file), file);
175 		spin_lock(&pde->pde_unload_lock);
176 		/* After ->release. */
177 		list_del(&pdeo->lh);
178 		c = pdeo->c;
179 		spin_unlock(&pde->pde_unload_lock);
180 		if (unlikely(c))
181 			complete(c);
182 		kmem_cache_free(pde_opener_cache, pdeo);
183 	}
184 }
185 
186 void proc_entry_rundown(struct proc_dir_entry *de)
187 {
188 	DECLARE_COMPLETION_ONSTACK(c);
189 	/* Wait until all existing callers into module are done. */
190 	de->pde_unload_completion = &c;
191 	if (atomic_add_return(BIAS, &de->in_use) != BIAS)
192 		wait_for_completion(&c);
193 
194 	/* ->pde_openers list can't grow from now on. */
195 
196 	spin_lock(&de->pde_unload_lock);
197 	while (!list_empty(&de->pde_openers)) {
198 		struct pde_opener *pdeo;
199 		pdeo = list_first_entry(&de->pde_openers, struct pde_opener, lh);
200 		close_pdeo(de, pdeo);
201 		spin_lock(&de->pde_unload_lock);
202 	}
203 	spin_unlock(&de->pde_unload_lock);
204 }
205 
206 static loff_t proc_reg_llseek(struct file *file, loff_t offset, int whence)
207 {
208 	struct proc_dir_entry *pde = PDE(file_inode(file));
209 	loff_t rv = -EINVAL;
210 	if (use_pde(pde)) {
211 		loff_t (*llseek)(struct file *, loff_t, int);
212 		llseek = pde->proc_fops->llseek;
213 		if (!llseek)
214 			llseek = default_llseek;
215 		rv = llseek(file, offset, whence);
216 		unuse_pde(pde);
217 	}
218 	return rv;
219 }
220 
221 static ssize_t proc_reg_read(struct file *file, char __user *buf, size_t count, loff_t *ppos)
222 {
223 	ssize_t (*read)(struct file *, char __user *, size_t, loff_t *);
224 	struct proc_dir_entry *pde = PDE(file_inode(file));
225 	ssize_t rv = -EIO;
226 	if (use_pde(pde)) {
227 		read = pde->proc_fops->read;
228 		if (read)
229 			rv = read(file, buf, count, ppos);
230 		unuse_pde(pde);
231 	}
232 	return rv;
233 }
234 
235 static ssize_t proc_reg_write(struct file *file, const char __user *buf, size_t count, loff_t *ppos)
236 {
237 	ssize_t (*write)(struct file *, const char __user *, size_t, loff_t *);
238 	struct proc_dir_entry *pde = PDE(file_inode(file));
239 	ssize_t rv = -EIO;
240 	if (use_pde(pde)) {
241 		write = pde->proc_fops->write;
242 		if (write)
243 			rv = write(file, buf, count, ppos);
244 		unuse_pde(pde);
245 	}
246 	return rv;
247 }
248 
249 static __poll_t proc_reg_poll(struct file *file, struct poll_table_struct *pts)
250 {
251 	struct proc_dir_entry *pde = PDE(file_inode(file));
252 	__poll_t rv = DEFAULT_POLLMASK;
253 	__poll_t (*poll)(struct file *, struct poll_table_struct *);
254 	if (use_pde(pde)) {
255 		poll = pde->proc_fops->poll;
256 		if (poll)
257 			rv = poll(file, pts);
258 		unuse_pde(pde);
259 	}
260 	return rv;
261 }
262 
263 static long proc_reg_unlocked_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
264 {
265 	struct proc_dir_entry *pde = PDE(file_inode(file));
266 	long rv = -ENOTTY;
267 	long (*ioctl)(struct file *, unsigned int, unsigned long);
268 	if (use_pde(pde)) {
269 		ioctl = pde->proc_fops->unlocked_ioctl;
270 		if (ioctl)
271 			rv = ioctl(file, cmd, arg);
272 		unuse_pde(pde);
273 	}
274 	return rv;
275 }
276 
277 #ifdef CONFIG_COMPAT
278 static long proc_reg_compat_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
279 {
280 	struct proc_dir_entry *pde = PDE(file_inode(file));
281 	long rv = -ENOTTY;
282 	long (*compat_ioctl)(struct file *, unsigned int, unsigned long);
283 	if (use_pde(pde)) {
284 		compat_ioctl = pde->proc_fops->compat_ioctl;
285 		if (compat_ioctl)
286 			rv = compat_ioctl(file, cmd, arg);
287 		unuse_pde(pde);
288 	}
289 	return rv;
290 }
291 #endif
292 
293 static int proc_reg_mmap(struct file *file, struct vm_area_struct *vma)
294 {
295 	struct proc_dir_entry *pde = PDE(file_inode(file));
296 	int rv = -EIO;
297 	int (*mmap)(struct file *, struct vm_area_struct *);
298 	if (use_pde(pde)) {
299 		mmap = pde->proc_fops->mmap;
300 		if (mmap)
301 			rv = mmap(file, vma);
302 		unuse_pde(pde);
303 	}
304 	return rv;
305 }
306 
307 static unsigned long
308 proc_reg_get_unmapped_area(struct file *file, unsigned long orig_addr,
309 			   unsigned long len, unsigned long pgoff,
310 			   unsigned long flags)
311 {
312 	struct proc_dir_entry *pde = PDE(file_inode(file));
313 	unsigned long rv = -EIO;
314 
315 	if (use_pde(pde)) {
316 		typeof(proc_reg_get_unmapped_area) *get_area;
317 
318 		get_area = pde->proc_fops->get_unmapped_area;
319 #ifdef CONFIG_MMU
320 		if (!get_area)
321 			get_area = current->mm->get_unmapped_area;
322 #endif
323 
324 		if (get_area)
325 			rv = get_area(file, orig_addr, len, pgoff, flags);
326 		else
327 			rv = orig_addr;
328 		unuse_pde(pde);
329 	}
330 	return rv;
331 }
332 
333 static int proc_reg_open(struct inode *inode, struct file *file)
334 {
335 	struct proc_dir_entry *pde = PDE(inode);
336 	int rv = 0;
337 	int (*open)(struct inode *, struct file *);
338 	int (*release)(struct inode *, struct file *);
339 	struct pde_opener *pdeo;
340 
341 	/*
342 	 * Ensure that
343 	 * 1) PDE's ->release hook will be called no matter what
344 	 *    either normally by close()/->release, or forcefully by
345 	 *    rmmod/remove_proc_entry.
346 	 *
347 	 * 2) rmmod isn't blocked by opening file in /proc and sitting on
348 	 *    the descriptor (including "rmmod foo </proc/foo" scenario).
349 	 *
350 	 * Save every "struct file" with custom ->release hook.
351 	 */
352 	if (!use_pde(pde))
353 		return -ENOENT;
354 
355 	release = pde->proc_fops->release;
356 	if (release) {
357 		pdeo = kmem_cache_alloc(pde_opener_cache, GFP_KERNEL);
358 		if (!pdeo) {
359 			rv = -ENOMEM;
360 			goto out_unuse;
361 		}
362 	}
363 
364 	open = pde->proc_fops->open;
365 	if (open)
366 		rv = open(inode, file);
367 
368 	if (release) {
369 		if (rv == 0) {
370 			/* To know what to release. */
371 			pdeo->file = file;
372 			pdeo->closing = false;
373 			pdeo->c = NULL;
374 			spin_lock(&pde->pde_unload_lock);
375 			list_add(&pdeo->lh, &pde->pde_openers);
376 			spin_unlock(&pde->pde_unload_lock);
377 		} else
378 			kmem_cache_free(pde_opener_cache, pdeo);
379 	}
380 
381 out_unuse:
382 	unuse_pde(pde);
383 	return rv;
384 }
385 
386 static int proc_reg_release(struct inode *inode, struct file *file)
387 {
388 	struct proc_dir_entry *pde = PDE(inode);
389 	struct pde_opener *pdeo;
390 	spin_lock(&pde->pde_unload_lock);
391 	list_for_each_entry(pdeo, &pde->pde_openers, lh) {
392 		if (pdeo->file == file) {
393 			close_pdeo(pde, pdeo);
394 			return 0;
395 		}
396 	}
397 	spin_unlock(&pde->pde_unload_lock);
398 	return 0;
399 }
400 
401 static const struct file_operations proc_reg_file_ops = {
402 	.llseek		= proc_reg_llseek,
403 	.read		= proc_reg_read,
404 	.write		= proc_reg_write,
405 	.poll		= proc_reg_poll,
406 	.unlocked_ioctl	= proc_reg_unlocked_ioctl,
407 #ifdef CONFIG_COMPAT
408 	.compat_ioctl	= proc_reg_compat_ioctl,
409 #endif
410 	.mmap		= proc_reg_mmap,
411 	.get_unmapped_area = proc_reg_get_unmapped_area,
412 	.open		= proc_reg_open,
413 	.release	= proc_reg_release,
414 };
415 
416 #ifdef CONFIG_COMPAT
417 static const struct file_operations proc_reg_file_ops_no_compat = {
418 	.llseek		= proc_reg_llseek,
419 	.read		= proc_reg_read,
420 	.write		= proc_reg_write,
421 	.poll		= proc_reg_poll,
422 	.unlocked_ioctl	= proc_reg_unlocked_ioctl,
423 	.mmap		= proc_reg_mmap,
424 	.get_unmapped_area = proc_reg_get_unmapped_area,
425 	.open		= proc_reg_open,
426 	.release	= proc_reg_release,
427 };
428 #endif
429 
430 static void proc_put_link(void *p)
431 {
432 	unuse_pde(p);
433 }
434 
435 static const char *proc_get_link(struct dentry *dentry,
436 				 struct inode *inode,
437 				 struct delayed_call *done)
438 {
439 	struct proc_dir_entry *pde = PDE(inode);
440 	if (!use_pde(pde))
441 		return ERR_PTR(-EINVAL);
442 	set_delayed_call(done, proc_put_link, pde);
443 	return pde->data;
444 }
445 
446 const struct inode_operations proc_link_inode_operations = {
447 	.get_link	= proc_get_link,
448 };
449 
450 struct inode *proc_get_inode(struct super_block *sb, struct proc_dir_entry *de)
451 {
452 	struct inode *inode = new_inode_pseudo(sb);
453 
454 	if (inode) {
455 		inode->i_ino = de->low_ino;
456 		inode->i_mtime = inode->i_atime = inode->i_ctime = current_time(inode);
457 		PROC_I(inode)->pde = de;
458 
459 		if (is_empty_pde(de)) {
460 			make_empty_dir_inode(inode);
461 			return inode;
462 		}
463 		if (de->mode) {
464 			inode->i_mode = de->mode;
465 			inode->i_uid = de->uid;
466 			inode->i_gid = de->gid;
467 		}
468 		if (de->size)
469 			inode->i_size = de->size;
470 		if (de->nlink)
471 			set_nlink(inode, de->nlink);
472 		WARN_ON(!de->proc_iops);
473 		inode->i_op = de->proc_iops;
474 		if (de->proc_fops) {
475 			if (S_ISREG(inode->i_mode)) {
476 #ifdef CONFIG_COMPAT
477 				if (!de->proc_fops->compat_ioctl)
478 					inode->i_fop =
479 						&proc_reg_file_ops_no_compat;
480 				else
481 #endif
482 					inode->i_fop = &proc_reg_file_ops;
483 			} else {
484 				inode->i_fop = de->proc_fops;
485 			}
486 		}
487 	} else
488 	       pde_put(de);
489 	return inode;
490 }
491 
492 int proc_fill_super(struct super_block *s, void *data, int silent)
493 {
494 	struct pid_namespace *ns = get_pid_ns(s->s_fs_info);
495 	struct inode *root_inode;
496 	int ret;
497 
498 	if (!proc_parse_options(data, ns))
499 		return -EINVAL;
500 
501 	/* User space would break if executables or devices appear on proc */
502 	s->s_iflags |= SB_I_USERNS_VISIBLE | SB_I_NOEXEC | SB_I_NODEV;
503 	s->s_flags |= SB_NODIRATIME | SB_NOSUID | SB_NOEXEC;
504 	s->s_blocksize = 1024;
505 	s->s_blocksize_bits = 10;
506 	s->s_magic = PROC_SUPER_MAGIC;
507 	s->s_op = &proc_sops;
508 	s->s_time_gran = 1;
509 
510 	/*
511 	 * procfs isn't actually a stacking filesystem; however, there is
512 	 * too much magic going on inside it to permit stacking things on
513 	 * top of it
514 	 */
515 	s->s_stack_depth = FILESYSTEM_MAX_STACK_DEPTH;
516 
517 	/* procfs dentries and inodes don't require IO to create */
518 	s->s_shrink.seeks = 0;
519 
520 	pde_get(&proc_root);
521 	root_inode = proc_get_inode(s, &proc_root);
522 	if (!root_inode) {
523 		pr_err("proc_fill_super: get root inode failed\n");
524 		return -ENOMEM;
525 	}
526 
527 	s->s_root = d_make_root(root_inode);
528 	if (!s->s_root) {
529 		pr_err("proc_fill_super: allocate dentry failed\n");
530 		return -ENOMEM;
531 	}
532 
533 	ret = proc_setup_self(s);
534 	if (ret) {
535 		return ret;
536 	}
537 	return proc_setup_thread_self(s);
538 }
539