xref: /linux/fs/proc/inode.c (revision 44f57d78) 
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 
28 #include <linux/uaccess.h>
29 
30 #include "internal.h"
31 
32 static void proc_evict_inode(struct inode *inode)
33 {
34 	struct proc_dir_entry *de;
35 	struct ctl_table_header *head;
36 
37 	truncate_inode_pages_final(&inode->i_data);
38 	clear_inode(inode);
39 
40 	/* Stop tracking associated processes */
41 	put_pid(PROC_I(inode)->pid);
42 
43 	/* Let go of any associated proc directory entry */
44 	de = PDE(inode);
45 	if (de)
46 		pde_put(de);
47 
48 	head = PROC_I(inode)->sysctl;
49 	if (head) {
50 		RCU_INIT_POINTER(PROC_I(inode)->sysctl, NULL);
51 		proc_sys_evict_inode(inode, head);
52 	}
53 }
54 
55 static struct kmem_cache *proc_inode_cachep __ro_after_init;
56 static struct kmem_cache *pde_opener_cache __ro_after_init;
57 
58 static struct inode *proc_alloc_inode(struct super_block *sb)
59 {
60 	struct proc_inode *ei;
61 
62 	ei = kmem_cache_alloc(proc_inode_cachep, GFP_KERNEL);
63 	if (!ei)
64 		return NULL;
65 	ei->pid = NULL;
66 	ei->fd = 0;
67 	ei->op.proc_get_link = NULL;
68 	ei->pde = NULL;
69 	ei->sysctl = NULL;
70 	ei->sysctl_entry = NULL;
71 	ei->ns_ops = NULL;
72 	return &ei->vfs_inode;
73 }
74 
75 static void proc_free_inode(struct inode *inode)
76 {
77 	kmem_cache_free(proc_inode_cachep, PROC_I(inode));
78 }
79 
80 static void init_once(void *foo)
81 {
82 	struct proc_inode *ei = (struct proc_inode *) foo;
83 
84 	inode_init_once(&ei->vfs_inode);
85 }
86 
87 void __init proc_init_kmemcache(void)
88 {
89 	proc_inode_cachep = kmem_cache_create("proc_inode_cache",
90 					     sizeof(struct proc_inode),
91 					     0, (SLAB_RECLAIM_ACCOUNT|
92 						SLAB_MEM_SPREAD|SLAB_ACCOUNT|
93 						SLAB_PANIC),
94 					     init_once);
95 	pde_opener_cache =
96 		kmem_cache_create("pde_opener", sizeof(struct pde_opener), 0,
97 				  SLAB_ACCOUNT|SLAB_PANIC, NULL);
98 	proc_dir_entry_cache = kmem_cache_create_usercopy(
99 		"proc_dir_entry", SIZEOF_PDE, 0, SLAB_PANIC,
100 		offsetof(struct proc_dir_entry, inline_name),
101 		SIZEOF_PDE_INLINE_NAME, NULL);
102 	BUILD_BUG_ON(sizeof(struct proc_dir_entry) >= SIZEOF_PDE);
103 }
104 
105 static int proc_show_options(struct seq_file *seq, struct dentry *root)
106 {
107 	struct super_block *sb = root->d_sb;
108 	struct pid_namespace *pid = sb->s_fs_info;
109 
110 	if (!gid_eq(pid->pid_gid, GLOBAL_ROOT_GID))
111 		seq_printf(seq, ",gid=%u", from_kgid_munged(&init_user_ns, pid->pid_gid));
112 	if (pid->hide_pid != HIDEPID_OFF)
113 		seq_printf(seq, ",hidepid=%u", pid->hide_pid);
114 
115 	return 0;
116 }
117 
118 const struct super_operations proc_sops = {
119 	.alloc_inode	= proc_alloc_inode,
120 	.free_inode	= proc_free_inode,
121 	.drop_inode	= generic_delete_inode,
122 	.evict_inode	= proc_evict_inode,
123 	.statfs		= simple_statfs,
124 	.show_options	= proc_show_options,
125 };
126 
127 enum {BIAS = -1U<<31};
128 
129 static inline int use_pde(struct proc_dir_entry *pde)
130 {
131 	return likely(atomic_inc_unless_negative(&pde->in_use));
132 }
133 
134 static void unuse_pde(struct proc_dir_entry *pde)
135 {
136 	if (unlikely(atomic_dec_return(&pde->in_use) == BIAS))
137 		complete(pde->pde_unload_completion);
138 }
139 
140 /* pde is locked on entry, unlocked on exit */
141 static void close_pdeo(struct proc_dir_entry *pde, struct pde_opener *pdeo)
142 {
143 	/*
144 	 * close() (proc_reg_release()) can't delete an entry and proceed:
145 	 * ->release hook needs to be available at the right moment.
146 	 *
147 	 * rmmod (remove_proc_entry() et al) can't delete an entry and proceed:
148 	 * "struct file" needs to be available at the right moment.
149 	 *
150 	 * Therefore, first process to enter this function does ->release() and
151 	 * signals its completion to the other process which does nothing.
152 	 */
153 	if (pdeo->closing) {
154 		/* somebody else is doing that, just wait */
155 		DECLARE_COMPLETION_ONSTACK(c);
156 		pdeo->c = &c;
157 		spin_unlock(&pde->pde_unload_lock);
158 		wait_for_completion(&c);
159 	} else {
160 		struct file *file;
161 		struct completion *c;
162 
163 		pdeo->closing = true;
164 		spin_unlock(&pde->pde_unload_lock);
165 		file = pdeo->file;
166 		pde->proc_fops->release(file_inode(file), file);
167 		spin_lock(&pde->pde_unload_lock);
168 		/* After ->release. */
169 		list_del(&pdeo->lh);
170 		c = pdeo->c;
171 		spin_unlock(&pde->pde_unload_lock);
172 		if (unlikely(c))
173 			complete(c);
174 		kmem_cache_free(pde_opener_cache, pdeo);
175 	}
176 }
177 
178 void proc_entry_rundown(struct proc_dir_entry *de)
179 {
180 	DECLARE_COMPLETION_ONSTACK(c);
181 	/* Wait until all existing callers into module are done. */
182 	de->pde_unload_completion = &c;
183 	if (atomic_add_return(BIAS, &de->in_use) != BIAS)
184 		wait_for_completion(&c);
185 
186 	/* ->pde_openers list can't grow from now on. */
187 
188 	spin_lock(&de->pde_unload_lock);
189 	while (!list_empty(&de->pde_openers)) {
190 		struct pde_opener *pdeo;
191 		pdeo = list_first_entry(&de->pde_openers, struct pde_opener, lh);
192 		close_pdeo(de, pdeo);
193 		spin_lock(&de->pde_unload_lock);
194 	}
195 	spin_unlock(&de->pde_unload_lock);
196 }
197 
198 static loff_t proc_reg_llseek(struct file *file, loff_t offset, int whence)
199 {
200 	struct proc_dir_entry *pde = PDE(file_inode(file));
201 	loff_t rv = -EINVAL;
202 	if (use_pde(pde)) {
203 		loff_t (*llseek)(struct file *, loff_t, int);
204 		llseek = pde->proc_fops->llseek;
205 		if (!llseek)
206 			llseek = default_llseek;
207 		rv = llseek(file, offset, whence);
208 		unuse_pde(pde);
209 	}
210 	return rv;
211 }
212 
213 static ssize_t proc_reg_read(struct file *file, char __user *buf, size_t count, loff_t *ppos)
214 {
215 	ssize_t (*read)(struct file *, char __user *, size_t, loff_t *);
216 	struct proc_dir_entry *pde = PDE(file_inode(file));
217 	ssize_t rv = -EIO;
218 	if (use_pde(pde)) {
219 		read = pde->proc_fops->read;
220 		if (read)
221 			rv = read(file, buf, count, ppos);
222 		unuse_pde(pde);
223 	}
224 	return rv;
225 }
226 
227 static ssize_t proc_reg_write(struct file *file, const char __user *buf, size_t count, loff_t *ppos)
228 {
229 	ssize_t (*write)(struct file *, const char __user *, size_t, loff_t *);
230 	struct proc_dir_entry *pde = PDE(file_inode(file));
231 	ssize_t rv = -EIO;
232 	if (use_pde(pde)) {
233 		write = pde->proc_fops->write;
234 		if (write)
235 			rv = write(file, buf, count, ppos);
236 		unuse_pde(pde);
237 	}
238 	return rv;
239 }
240 
241 static __poll_t proc_reg_poll(struct file *file, struct poll_table_struct *pts)
242 {
243 	struct proc_dir_entry *pde = PDE(file_inode(file));
244 	__poll_t rv = DEFAULT_POLLMASK;
245 	__poll_t (*poll)(struct file *, struct poll_table_struct *);
246 	if (use_pde(pde)) {
247 		poll = pde->proc_fops->poll;
248 		if (poll)
249 			rv = poll(file, pts);
250 		unuse_pde(pde);
251 	}
252 	return rv;
253 }
254 
255 static long proc_reg_unlocked_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
256 {
257 	struct proc_dir_entry *pde = PDE(file_inode(file));
258 	long rv = -ENOTTY;
259 	long (*ioctl)(struct file *, unsigned int, unsigned long);
260 	if (use_pde(pde)) {
261 		ioctl = pde->proc_fops->unlocked_ioctl;
262 		if (ioctl)
263 			rv = ioctl(file, cmd, arg);
264 		unuse_pde(pde);
265 	}
266 	return rv;
267 }
268 
269 #ifdef CONFIG_COMPAT
270 static long proc_reg_compat_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
271 {
272 	struct proc_dir_entry *pde = PDE(file_inode(file));
273 	long rv = -ENOTTY;
274 	long (*compat_ioctl)(struct file *, unsigned int, unsigned long);
275 	if (use_pde(pde)) {
276 		compat_ioctl = pde->proc_fops->compat_ioctl;
277 		if (compat_ioctl)
278 			rv = compat_ioctl(file, cmd, arg);
279 		unuse_pde(pde);
280 	}
281 	return rv;
282 }
283 #endif
284 
285 static int proc_reg_mmap(struct file *file, struct vm_area_struct *vma)
286 {
287 	struct proc_dir_entry *pde = PDE(file_inode(file));
288 	int rv = -EIO;
289 	int (*mmap)(struct file *, struct vm_area_struct *);
290 	if (use_pde(pde)) {
291 		mmap = pde->proc_fops->mmap;
292 		if (mmap)
293 			rv = mmap(file, vma);
294 		unuse_pde(pde);
295 	}
296 	return rv;
297 }
298 
299 static unsigned long
300 proc_reg_get_unmapped_area(struct file *file, unsigned long orig_addr,
301 			   unsigned long len, unsigned long pgoff,
302 			   unsigned long flags)
303 {
304 	struct proc_dir_entry *pde = PDE(file_inode(file));
305 	unsigned long rv = -EIO;
306 
307 	if (use_pde(pde)) {
308 		typeof(proc_reg_get_unmapped_area) *get_area;
309 
310 		get_area = pde->proc_fops->get_unmapped_area;
311 #ifdef CONFIG_MMU
312 		if (!get_area)
313 			get_area = current->mm->get_unmapped_area;
314 #endif
315 
316 		if (get_area)
317 			rv = get_area(file, orig_addr, len, pgoff, flags);
318 		else
319 			rv = orig_addr;
320 		unuse_pde(pde);
321 	}
322 	return rv;
323 }
324 
325 static int proc_reg_open(struct inode *inode, struct file *file)
326 {
327 	struct proc_dir_entry *pde = PDE(inode);
328 	int rv = 0;
329 	int (*open)(struct inode *, struct file *);
330 	int (*release)(struct inode *, struct file *);
331 	struct pde_opener *pdeo;
332 
333 	/*
334 	 * Ensure that
335 	 * 1) PDE's ->release hook will be called no matter what
336 	 *    either normally by close()/->release, or forcefully by
337 	 *    rmmod/remove_proc_entry.
338 	 *
339 	 * 2) rmmod isn't blocked by opening file in /proc and sitting on
340 	 *    the descriptor (including "rmmod foo </proc/foo" scenario).
341 	 *
342 	 * Save every "struct file" with custom ->release hook.
343 	 */
344 	if (!use_pde(pde))
345 		return -ENOENT;
346 
347 	release = pde->proc_fops->release;
348 	if (release) {
349 		pdeo = kmem_cache_alloc(pde_opener_cache, GFP_KERNEL);
350 		if (!pdeo) {
351 			rv = -ENOMEM;
352 			goto out_unuse;
353 		}
354 	}
355 
356 	open = pde->proc_fops->open;
357 	if (open)
358 		rv = open(inode, file);
359 
360 	if (release) {
361 		if (rv == 0) {
362 			/* To know what to release. */
363 			pdeo->file = file;
364 			pdeo->closing = false;
365 			pdeo->c = NULL;
366 			spin_lock(&pde->pde_unload_lock);
367 			list_add(&pdeo->lh, &pde->pde_openers);
368 			spin_unlock(&pde->pde_unload_lock);
369 		} else
370 			kmem_cache_free(pde_opener_cache, pdeo);
371 	}
372 
373 out_unuse:
374 	unuse_pde(pde);
375 	return rv;
376 }
377 
378 static int proc_reg_release(struct inode *inode, struct file *file)
379 {
380 	struct proc_dir_entry *pde = PDE(inode);
381 	struct pde_opener *pdeo;
382 	spin_lock(&pde->pde_unload_lock);
383 	list_for_each_entry(pdeo, &pde->pde_openers, lh) {
384 		if (pdeo->file == file) {
385 			close_pdeo(pde, pdeo);
386 			return 0;
387 		}
388 	}
389 	spin_unlock(&pde->pde_unload_lock);
390 	return 0;
391 }
392 
393 static const struct file_operations proc_reg_file_ops = {
394 	.llseek		= proc_reg_llseek,
395 	.read		= proc_reg_read,
396 	.write		= proc_reg_write,
397 	.poll		= proc_reg_poll,
398 	.unlocked_ioctl	= proc_reg_unlocked_ioctl,
399 #ifdef CONFIG_COMPAT
400 	.compat_ioctl	= proc_reg_compat_ioctl,
401 #endif
402 	.mmap		= proc_reg_mmap,
403 	.get_unmapped_area = proc_reg_get_unmapped_area,
404 	.open		= proc_reg_open,
405 	.release	= proc_reg_release,
406 };
407 
408 #ifdef CONFIG_COMPAT
409 static const struct file_operations proc_reg_file_ops_no_compat = {
410 	.llseek		= proc_reg_llseek,
411 	.read		= proc_reg_read,
412 	.write		= proc_reg_write,
413 	.poll		= proc_reg_poll,
414 	.unlocked_ioctl	= proc_reg_unlocked_ioctl,
415 	.mmap		= proc_reg_mmap,
416 	.get_unmapped_area = proc_reg_get_unmapped_area,
417 	.open		= proc_reg_open,
418 	.release	= proc_reg_release,
419 };
420 #endif
421 
422 static void proc_put_link(void *p)
423 {
424 	unuse_pde(p);
425 }
426 
427 static const char *proc_get_link(struct dentry *dentry,
428 				 struct inode *inode,
429 				 struct delayed_call *done)
430 {
431 	struct proc_dir_entry *pde = PDE(inode);
432 	if (!use_pde(pde))
433 		return ERR_PTR(-EINVAL);
434 	set_delayed_call(done, proc_put_link, pde);
435 	return pde->data;
436 }
437 
438 const struct inode_operations proc_link_inode_operations = {
439 	.get_link	= proc_get_link,
440 };
441 
442 struct inode *proc_get_inode(struct super_block *sb, struct proc_dir_entry *de)
443 {
444 	struct inode *inode = new_inode_pseudo(sb);
445 
446 	if (inode) {
447 		inode->i_ino = de->low_ino;
448 		inode->i_mtime = inode->i_atime = inode->i_ctime = current_time(inode);
449 		PROC_I(inode)->pde = de;
450 
451 		if (is_empty_pde(de)) {
452 			make_empty_dir_inode(inode);
453 			return inode;
454 		}
455 		if (de->mode) {
456 			inode->i_mode = de->mode;
457 			inode->i_uid = de->uid;
458 			inode->i_gid = de->gid;
459 		}
460 		if (de->size)
461 			inode->i_size = de->size;
462 		if (de->nlink)
463 			set_nlink(inode, de->nlink);
464 		WARN_ON(!de->proc_iops);
465 		inode->i_op = de->proc_iops;
466 		if (de->proc_fops) {
467 			if (S_ISREG(inode->i_mode)) {
468 #ifdef CONFIG_COMPAT
469 				if (!de->proc_fops->compat_ioctl)
470 					inode->i_fop =
471 						&proc_reg_file_ops_no_compat;
472 				else
473 #endif
474 					inode->i_fop = &proc_reg_file_ops;
475 			} else {
476 				inode->i_fop = de->proc_fops;
477 			}
478 		}
479 	} else
480 	       pde_put(de);
481 	return inode;
482 }
483