xref: /linux/fs/proc/inode.c (revision e58e871becec2d3b04ed91c0c16fe8deac9c9dfa) 
1 /*
2  *  linux/fs/proc/inode.c
3  *
4  *  Copyright (C) 1991, 1992  Linus Torvalds
5  */
6 
7 #include <linux/time.h>
8 #include <linux/proc_fs.h>
9 #include <linux/kernel.h>
10 #include <linux/pid_namespace.h>
11 #include <linux/mm.h>
12 #include <linux/string.h>
13 #include <linux/stat.h>
14 #include <linux/completion.h>
15 #include <linux/poll.h>
16 #include <linux/printk.h>
17 #include <linux/file.h>
18 #include <linux/limits.h>
19 #include <linux/init.h>
20 #include <linux/module.h>
21 #include <linux/sysctl.h>
22 #include <linux/seq_file.h>
23 #include <linux/slab.h>
24 #include <linux/mount.h>
25 #include <linux/magic.h>
26 
27 #include <linux/uaccess.h>
28 
29 #include "internal.h"
30 
31 static void proc_evict_inode(struct inode *inode)
32 {
33 	struct proc_dir_entry *de;
34 	struct ctl_table_header *head;
35 
36 	truncate_inode_pages_final(&inode->i_data);
37 	clear_inode(inode);
38 
39 	/* Stop tracking associated processes */
40 	put_pid(PROC_I(inode)->pid);
41 
42 	/* Let go of any associated proc directory entry */
43 	de = PDE(inode);
44 	if (de)
45 		pde_put(de);
46 
47 	head = PROC_I(inode)->sysctl;
48 	if (head) {
49 		RCU_INIT_POINTER(PROC_I(inode)->sysctl, NULL);
50 		proc_sys_evict_inode(inode, head);
51 	}
52 }
53 
54 static struct kmem_cache * proc_inode_cachep;
55 
56 static struct inode *proc_alloc_inode(struct super_block *sb)
57 {
58 	struct proc_inode *ei;
59 	struct inode *inode;
60 
61 	ei = kmem_cache_alloc(proc_inode_cachep, GFP_KERNEL);
62 	if (!ei)
63 		return NULL;
64 	ei->pid = NULL;
65 	ei->fd = 0;
66 	ei->op.proc_get_link = NULL;
67 	ei->pde = NULL;
68 	ei->sysctl = NULL;
69 	ei->sysctl_entry = NULL;
70 	ei->ns_ops = NULL;
71 	inode = &ei->vfs_inode;
72 	return inode;
73 }
74 
75 static void proc_i_callback(struct rcu_head *head)
76 {
77 	struct inode *inode = container_of(head, struct inode, i_rcu);
78 	kmem_cache_free(proc_inode_cachep, PROC_I(inode));
79 }
80 
81 static void proc_destroy_inode(struct inode *inode)
82 {
83 	call_rcu(&inode->i_rcu, proc_i_callback);
84 }
85 
86 static void init_once(void *foo)
87 {
88 	struct proc_inode *ei = (struct proc_inode *) foo;
89 
90 	inode_init_once(&ei->vfs_inode);
91 }
92 
93 void __init proc_init_inodecache(void)
94 {
95 	proc_inode_cachep = kmem_cache_create("proc_inode_cache",
96 					     sizeof(struct proc_inode),
97 					     0, (SLAB_RECLAIM_ACCOUNT|
98 						SLAB_MEM_SPREAD|SLAB_ACCOUNT|
99 						SLAB_PANIC),
100 					     init_once);
101 }
102 
103 static int proc_show_options(struct seq_file *seq, struct dentry *root)
104 {
105 	struct super_block *sb = root->d_sb;
106 	struct pid_namespace *pid = sb->s_fs_info;
107 
108 	if (!gid_eq(pid->pid_gid, GLOBAL_ROOT_GID))
109 		seq_printf(seq, ",gid=%u", from_kgid_munged(&init_user_ns, pid->pid_gid));
110 	if (pid->hide_pid != HIDEPID_OFF)
111 		seq_printf(seq, ",hidepid=%u", pid->hide_pid);
112 
113 	return 0;
114 }
115 
116 static const struct super_operations proc_sops = {
117 	.alloc_inode	= proc_alloc_inode,
118 	.destroy_inode	= proc_destroy_inode,
119 	.drop_inode	= generic_delete_inode,
120 	.evict_inode	= proc_evict_inode,
121 	.statfs		= simple_statfs,
122 	.remount_fs	= proc_remount,
123 	.show_options	= proc_show_options,
124 };
125 
126 enum {BIAS = -1U<<31};
127 
128 static inline int use_pde(struct proc_dir_entry *pde)
129 {
130 	return atomic_inc_unless_negative(&pde->in_use);
131 }
132 
133 static void unuse_pde(struct proc_dir_entry *pde)
134 {
135 	if (atomic_dec_return(&pde->in_use) == BIAS)
136 		complete(pde->pde_unload_completion);
137 }
138 
139 /* pde is locked */
140 static void close_pdeo(struct proc_dir_entry *pde, struct pde_opener *pdeo)
141 {
142 	/*
143 	 * close() (proc_reg_release()) can't delete an entry and proceed:
144 	 * ->release hook needs to be available at the right moment.
145 	 *
146 	 * rmmod (remove_proc_entry() et al) can't delete an entry and proceed:
147 	 * "struct file" needs to be available at the right moment.
148 	 *
149 	 * Therefore, first process to enter this function does ->release() and
150 	 * signals its completion to the other process which does nothing.
151 	 */
152 	if (pdeo->closing) {
153 		/* somebody else is doing that, just wait */
154 		DECLARE_COMPLETION_ONSTACK(c);
155 		pdeo->c = &c;
156 		spin_unlock(&pde->pde_unload_lock);
157 		wait_for_completion(&c);
158 		spin_lock(&pde->pde_unload_lock);
159 	} else {
160 		struct file *file;
161 		pdeo->closing = true;
162 		spin_unlock(&pde->pde_unload_lock);
163 		file = pdeo->file;
164 		pde->proc_fops->release(file_inode(file), file);
165 		spin_lock(&pde->pde_unload_lock);
166 		/* After ->release. */
167 		list_del(&pdeo->lh);
168 		if (pdeo->c)
169 			complete(pdeo->c);
170 		kfree(pdeo);
171 	}
172 }
173 
174 void proc_entry_rundown(struct proc_dir_entry *de)
175 {
176 	DECLARE_COMPLETION_ONSTACK(c);
177 	/* Wait until all existing callers into module are done. */
178 	de->pde_unload_completion = &c;
179 	if (atomic_add_return(BIAS, &de->in_use) != BIAS)
180 		wait_for_completion(&c);
181 
182 	/* ->pde_openers list can't grow from now on. */
183 
184 	spin_lock(&de->pde_unload_lock);
185 	while (!list_empty(&de->pde_openers)) {
186 		struct pde_opener *pdeo;
187 		pdeo = list_first_entry(&de->pde_openers, struct pde_opener, lh);
188 		close_pdeo(de, pdeo);
189 	}
190 	spin_unlock(&de->pde_unload_lock);
191 }
192 
193 static loff_t proc_reg_llseek(struct file *file, loff_t offset, int whence)
194 {
195 	struct proc_dir_entry *pde = PDE(file_inode(file));
196 	loff_t rv = -EINVAL;
197 	if (use_pde(pde)) {
198 		loff_t (*llseek)(struct file *, loff_t, int);
199 		llseek = pde->proc_fops->llseek;
200 		if (!llseek)
201 			llseek = default_llseek;
202 		rv = llseek(file, offset, whence);
203 		unuse_pde(pde);
204 	}
205 	return rv;
206 }
207 
208 static ssize_t proc_reg_read(struct file *file, char __user *buf, size_t count, loff_t *ppos)
209 {
210 	ssize_t (*read)(struct file *, char __user *, size_t, loff_t *);
211 	struct proc_dir_entry *pde = PDE(file_inode(file));
212 	ssize_t rv = -EIO;
213 	if (use_pde(pde)) {
214 		read = pde->proc_fops->read;
215 		if (read)
216 			rv = read(file, buf, count, ppos);
217 		unuse_pde(pde);
218 	}
219 	return rv;
220 }
221 
222 static ssize_t proc_reg_write(struct file *file, const char __user *buf, size_t count, loff_t *ppos)
223 {
224 	ssize_t (*write)(struct file *, const char __user *, size_t, loff_t *);
225 	struct proc_dir_entry *pde = PDE(file_inode(file));
226 	ssize_t rv = -EIO;
227 	if (use_pde(pde)) {
228 		write = pde->proc_fops->write;
229 		if (write)
230 			rv = write(file, buf, count, ppos);
231 		unuse_pde(pde);
232 	}
233 	return rv;
234 }
235 
236 static unsigned int proc_reg_poll(struct file *file, struct poll_table_struct *pts)
237 {
238 	struct proc_dir_entry *pde = PDE(file_inode(file));
239 	unsigned int rv = DEFAULT_POLLMASK;
240 	unsigned int (*poll)(struct file *, struct poll_table_struct *);
241 	if (use_pde(pde)) {
242 		poll = pde->proc_fops->poll;
243 		if (poll)
244 			rv = poll(file, pts);
245 		unuse_pde(pde);
246 	}
247 	return rv;
248 }
249 
250 static long proc_reg_unlocked_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
251 {
252 	struct proc_dir_entry *pde = PDE(file_inode(file));
253 	long rv = -ENOTTY;
254 	long (*ioctl)(struct file *, unsigned int, unsigned long);
255 	if (use_pde(pde)) {
256 		ioctl = pde->proc_fops->unlocked_ioctl;
257 		if (ioctl)
258 			rv = ioctl(file, cmd, arg);
259 		unuse_pde(pde);
260 	}
261 	return rv;
262 }
263 
264 #ifdef CONFIG_COMPAT
265 static long proc_reg_compat_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
266 {
267 	struct proc_dir_entry *pde = PDE(file_inode(file));
268 	long rv = -ENOTTY;
269 	long (*compat_ioctl)(struct file *, unsigned int, unsigned long);
270 	if (use_pde(pde)) {
271 		compat_ioctl = pde->proc_fops->compat_ioctl;
272 		if (compat_ioctl)
273 			rv = compat_ioctl(file, cmd, arg);
274 		unuse_pde(pde);
275 	}
276 	return rv;
277 }
278 #endif
279 
280 static int proc_reg_mmap(struct file *file, struct vm_area_struct *vma)
281 {
282 	struct proc_dir_entry *pde = PDE(file_inode(file));
283 	int rv = -EIO;
284 	int (*mmap)(struct file *, struct vm_area_struct *);
285 	if (use_pde(pde)) {
286 		mmap = pde->proc_fops->mmap;
287 		if (mmap)
288 			rv = mmap(file, vma);
289 		unuse_pde(pde);
290 	}
291 	return rv;
292 }
293 
294 static unsigned long
295 proc_reg_get_unmapped_area(struct file *file, unsigned long orig_addr,
296 			   unsigned long len, unsigned long pgoff,
297 			   unsigned long flags)
298 {
299 	struct proc_dir_entry *pde = PDE(file_inode(file));
300 	unsigned long rv = -EIO;
301 
302 	if (use_pde(pde)) {
303 		typeof(proc_reg_get_unmapped_area) *get_area;
304 
305 		get_area = pde->proc_fops->get_unmapped_area;
306 #ifdef CONFIG_MMU
307 		if (!get_area)
308 			get_area = current->mm->get_unmapped_area;
309 #endif
310 
311 		if (get_area)
312 			rv = get_area(file, orig_addr, len, pgoff, flags);
313 		else
314 			rv = orig_addr;
315 		unuse_pde(pde);
316 	}
317 	return rv;
318 }
319 
320 static int proc_reg_open(struct inode *inode, struct file *file)
321 {
322 	struct proc_dir_entry *pde = PDE(inode);
323 	int rv = 0;
324 	int (*open)(struct inode *, struct file *);
325 	int (*release)(struct inode *, struct file *);
326 	struct pde_opener *pdeo;
327 
328 	/*
329 	 * Ensure that
330 	 * 1) PDE's ->release hook will be called no matter what
331 	 *    either normally by close()/->release, or forcefully by
332 	 *    rmmod/remove_proc_entry.
333 	 *
334 	 * 2) rmmod isn't blocked by opening file in /proc and sitting on
335 	 *    the descriptor (including "rmmod foo </proc/foo" scenario).
336 	 *
337 	 * Save every "struct file" with custom ->release hook.
338 	 */
339 	pdeo = kmalloc(sizeof(struct pde_opener), GFP_KERNEL);
340 	if (!pdeo)
341 		return -ENOMEM;
342 
343 	if (!use_pde(pde)) {
344 		kfree(pdeo);
345 		return -ENOENT;
346 	}
347 	open = pde->proc_fops->open;
348 	release = pde->proc_fops->release;
349 
350 	if (open)
351 		rv = open(inode, file);
352 
353 	if (rv == 0 && release) {
354 		/* To know what to release. */
355 		pdeo->file = file;
356 		pdeo->closing = false;
357 		pdeo->c = NULL;
358 		spin_lock(&pde->pde_unload_lock);
359 		list_add(&pdeo->lh, &pde->pde_openers);
360 		spin_unlock(&pde->pde_unload_lock);
361 	} else
362 		kfree(pdeo);
363 
364 	unuse_pde(pde);
365 	return rv;
366 }
367 
368 static int proc_reg_release(struct inode *inode, struct file *file)
369 {
370 	struct proc_dir_entry *pde = PDE(inode);
371 	struct pde_opener *pdeo;
372 	spin_lock(&pde->pde_unload_lock);
373 	list_for_each_entry(pdeo, &pde->pde_openers, lh) {
374 		if (pdeo->file == file) {
375 			close_pdeo(pde, pdeo);
376 			break;
377 		}
378 	}
379 	spin_unlock(&pde->pde_unload_lock);
380 	return 0;
381 }
382 
383 static const struct file_operations proc_reg_file_ops = {
384 	.llseek		= proc_reg_llseek,
385 	.read		= proc_reg_read,
386 	.write		= proc_reg_write,
387 	.poll		= proc_reg_poll,
388 	.unlocked_ioctl	= proc_reg_unlocked_ioctl,
389 #ifdef CONFIG_COMPAT
390 	.compat_ioctl	= proc_reg_compat_ioctl,
391 #endif
392 	.mmap		= proc_reg_mmap,
393 	.get_unmapped_area = proc_reg_get_unmapped_area,
394 	.open		= proc_reg_open,
395 	.release	= proc_reg_release,
396 };
397 
398 #ifdef CONFIG_COMPAT
399 static const struct file_operations proc_reg_file_ops_no_compat = {
400 	.llseek		= proc_reg_llseek,
401 	.read		= proc_reg_read,
402 	.write		= proc_reg_write,
403 	.poll		= proc_reg_poll,
404 	.unlocked_ioctl	= proc_reg_unlocked_ioctl,
405 	.mmap		= proc_reg_mmap,
406 	.get_unmapped_area = proc_reg_get_unmapped_area,
407 	.open		= proc_reg_open,
408 	.release	= proc_reg_release,
409 };
410 #endif
411 
412 static void proc_put_link(void *p)
413 {
414 	unuse_pde(p);
415 }
416 
417 static const char *proc_get_link(struct dentry *dentry,
418 				 struct inode *inode,
419 				 struct delayed_call *done)
420 {
421 	struct proc_dir_entry *pde = PDE(inode);
422 	if (unlikely(!use_pde(pde)))
423 		return ERR_PTR(-EINVAL);
424 	set_delayed_call(done, proc_put_link, pde);
425 	return pde->data;
426 }
427 
428 const struct inode_operations proc_link_inode_operations = {
429 	.get_link	= proc_get_link,
430 };
431 
432 struct inode *proc_get_inode(struct super_block *sb, struct proc_dir_entry *de)
433 {
434 	struct inode *inode = new_inode_pseudo(sb);
435 
436 	if (inode) {
437 		inode->i_ino = de->low_ino;
438 		inode->i_mtime = inode->i_atime = inode->i_ctime = current_time(inode);
439 		PROC_I(inode)->pde = de;
440 
441 		if (is_empty_pde(de)) {
442 			make_empty_dir_inode(inode);
443 			return inode;
444 		}
445 		if (de->mode) {
446 			inode->i_mode = de->mode;
447 			inode->i_uid = de->uid;
448 			inode->i_gid = de->gid;
449 		}
450 		if (de->size)
451 			inode->i_size = de->size;
452 		if (de->nlink)
453 			set_nlink(inode, de->nlink);
454 		WARN_ON(!de->proc_iops);
455 		inode->i_op = de->proc_iops;
456 		if (de->proc_fops) {
457 			if (S_ISREG(inode->i_mode)) {
458 #ifdef CONFIG_COMPAT
459 				if (!de->proc_fops->compat_ioctl)
460 					inode->i_fop =
461 						&proc_reg_file_ops_no_compat;
462 				else
463 #endif
464 					inode->i_fop = &proc_reg_file_ops;
465 			} else {
466 				inode->i_fop = de->proc_fops;
467 			}
468 		}
469 	} else
470 	       pde_put(de);
471 	return inode;
472 }
473 
474 int proc_fill_super(struct super_block *s, void *data, int silent)
475 {
476 	struct pid_namespace *ns = get_pid_ns(s->s_fs_info);
477 	struct inode *root_inode;
478 	int ret;
479 
480 	if (!proc_parse_options(data, ns))
481 		return -EINVAL;
482 
483 	/* User space would break if executables or devices appear on proc */
484 	s->s_iflags |= SB_I_USERNS_VISIBLE | SB_I_NOEXEC | SB_I_NODEV;
485 	s->s_flags |= MS_NODIRATIME | MS_NOSUID | MS_NOEXEC;
486 	s->s_blocksize = 1024;
487 	s->s_blocksize_bits = 10;
488 	s->s_magic = PROC_SUPER_MAGIC;
489 	s->s_op = &proc_sops;
490 	s->s_time_gran = 1;
491 
492 	/*
493 	 * procfs isn't actually a stacking filesystem; however, there is
494 	 * too much magic going on inside it to permit stacking things on
495 	 * top of it
496 	 */
497 	s->s_stack_depth = FILESYSTEM_MAX_STACK_DEPTH;
498 
499 	pde_get(&proc_root);
500 	root_inode = proc_get_inode(s, &proc_root);
501 	if (!root_inode) {
502 		pr_err("proc_fill_super: get root inode failed\n");
503 		return -ENOMEM;
504 	}
505 
506 	s->s_root = d_make_root(root_inode);
507 	if (!s->s_root) {
508 		pr_err("proc_fill_super: allocate dentry failed\n");
509 		return -ENOMEM;
510 	}
511 
512 	ret = proc_setup_self(s);
513 	if (ret) {
514 		return ret;
515 	}
516 	return proc_setup_thread_self(s);
517 }
518