xref: /linux/fs/proc/inode.c (revision 80d443e8876602be2c130f79c4de81e12e2a700d) 
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 	head = PROC_I(inode)->sysctl;
47 	if (head) {
48 		RCU_INIT_POINTER(PROC_I(inode)->sysctl, NULL);
49 		sysctl_head_put(head);
50 	}
51 }
52 
53 static struct kmem_cache * proc_inode_cachep;
54 
55 static struct inode *proc_alloc_inode(struct super_block *sb)
56 {
57 	struct proc_inode *ei;
58 	struct inode *inode;
59 
60 	ei = (struct proc_inode *)kmem_cache_alloc(proc_inode_cachep, GFP_KERNEL);
61 	if (!ei)
62 		return NULL;
63 	ei->pid = NULL;
64 	ei->fd = 0;
65 	ei->op.proc_get_link = NULL;
66 	ei->pde = NULL;
67 	ei->sysctl = NULL;
68 	ei->sysctl_entry = NULL;
69 	ei->ns_ops = NULL;
70 	inode = &ei->vfs_inode;
71 	return inode;
72 }
73 
74 static void proc_i_callback(struct rcu_head *head)
75 {
76 	struct inode *inode = container_of(head, struct inode, i_rcu);
77 	kmem_cache_free(proc_inode_cachep, PROC_I(inode));
78 }
79 
80 static void proc_destroy_inode(struct inode *inode)
81 {
82 	call_rcu(&inode->i_rcu, proc_i_callback);
83 }
84 
85 static void init_once(void *foo)
86 {
87 	struct proc_inode *ei = (struct proc_inode *) foo;
88 
89 	inode_init_once(&ei->vfs_inode);
90 }
91 
92 void __init proc_init_inodecache(void)
93 {
94 	proc_inode_cachep = kmem_cache_create("proc_inode_cache",
95 					     sizeof(struct proc_inode),
96 					     0, (SLAB_RECLAIM_ACCOUNT|
97 						SLAB_MEM_SPREAD|SLAB_ACCOUNT|
98 						SLAB_PANIC),
99 					     init_once);
100 }
101 
102 static int proc_show_options(struct seq_file *seq, struct dentry *root)
103 {
104 	struct super_block *sb = root->d_sb;
105 	struct pid_namespace *pid = sb->s_fs_info;
106 
107 	if (!gid_eq(pid->pid_gid, GLOBAL_ROOT_GID))
108 		seq_printf(seq, ",gid=%u", from_kgid_munged(&init_user_ns, pid->pid_gid));
109 	if (pid->hide_pid != 0)
110 		seq_printf(seq, ",hidepid=%u", pid->hide_pid);
111 
112 	return 0;
113 }
114 
115 static const struct super_operations proc_sops = {
116 	.alloc_inode	= proc_alloc_inode,
117 	.destroy_inode	= proc_destroy_inode,
118 	.drop_inode	= generic_delete_inode,
119 	.evict_inode	= proc_evict_inode,
120 	.statfs		= simple_statfs,
121 	.remount_fs	= proc_remount,
122 	.show_options	= proc_show_options,
123 };
124 
125 enum {BIAS = -1U<<31};
126 
127 static inline int use_pde(struct proc_dir_entry *pde)
128 {
129 	return atomic_inc_unless_negative(&pde->in_use);
130 }
131 
132 static void unuse_pde(struct proc_dir_entry *pde)
133 {
134 	if (atomic_dec_return(&pde->in_use) == BIAS)
135 		complete(pde->pde_unload_completion);
136 }
137 
138 /* pde is locked */
139 static void close_pdeo(struct proc_dir_entry *pde, struct pde_opener *pdeo)
140 {
141 	/*
142 	 * close() (proc_reg_release()) can't delete an entry and proceed:
143 	 * ->release hook needs to be available at the right moment.
144 	 *
145 	 * rmmod (remove_proc_entry() et al) can't delete an entry and proceed:
146 	 * "struct file" needs to be available at the right moment.
147 	 *
148 	 * Therefore, first process to enter this function does ->release() and
149 	 * signals its completion to the other process which does nothing.
150 	 */
151 	if (pdeo->closing) {
152 		/* somebody else is doing that, just wait */
153 		DECLARE_COMPLETION_ONSTACK(c);
154 		pdeo->c = &c;
155 		spin_unlock(&pde->pde_unload_lock);
156 		wait_for_completion(&c);
157 		spin_lock(&pde->pde_unload_lock);
158 	} else {
159 		struct file *file;
160 		pdeo->closing = true;
161 		spin_unlock(&pde->pde_unload_lock);
162 		file = pdeo->file;
163 		pde->proc_fops->release(file_inode(file), file);
164 		spin_lock(&pde->pde_unload_lock);
165 		/* After ->release. */
166 		list_del(&pdeo->lh);
167 		if (pdeo->c)
168 			complete(pdeo->c);
169 		kfree(pdeo);
170 	}
171 }
172 
173 void proc_entry_rundown(struct proc_dir_entry *de)
174 {
175 	DECLARE_COMPLETION_ONSTACK(c);
176 	/* Wait until all existing callers into module are done. */
177 	de->pde_unload_completion = &c;
178 	if (atomic_add_return(BIAS, &de->in_use) != BIAS)
179 		wait_for_completion(&c);
180 
181 	/* ->pde_openers list can't grow from now on. */
182 
183 	spin_lock(&de->pde_unload_lock);
184 	while (!list_empty(&de->pde_openers)) {
185 		struct pde_opener *pdeo;
186 		pdeo = list_first_entry(&de->pde_openers, struct pde_opener, lh);
187 		close_pdeo(de, pdeo);
188 	}
189 	spin_unlock(&de->pde_unload_lock);
190 }
191 
192 static loff_t proc_reg_llseek(struct file *file, loff_t offset, int whence)
193 {
194 	struct proc_dir_entry *pde = PDE(file_inode(file));
195 	loff_t rv = -EINVAL;
196 	if (use_pde(pde)) {
197 		loff_t (*llseek)(struct file *, loff_t, int);
198 		llseek = pde->proc_fops->llseek;
199 		if (!llseek)
200 			llseek = default_llseek;
201 		rv = llseek(file, offset, whence);
202 		unuse_pde(pde);
203 	}
204 	return rv;
205 }
206 
207 static ssize_t proc_reg_read(struct file *file, char __user *buf, size_t count, loff_t *ppos)
208 {
209 	ssize_t (*read)(struct file *, char __user *, size_t, loff_t *);
210 	struct proc_dir_entry *pde = PDE(file_inode(file));
211 	ssize_t rv = -EIO;
212 	if (use_pde(pde)) {
213 		read = pde->proc_fops->read;
214 		if (read)
215 			rv = read(file, buf, count, ppos);
216 		unuse_pde(pde);
217 	}
218 	return rv;
219 }
220 
221 static ssize_t proc_reg_write(struct file *file, const char __user *buf, size_t count, loff_t *ppos)
222 {
223 	ssize_t (*write)(struct file *, const 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 		write = pde->proc_fops->write;
228 		if (write)
229 			rv = write(file, buf, count, ppos);
230 		unuse_pde(pde);
231 	}
232 	return rv;
233 }
234 
235 static unsigned int proc_reg_poll(struct file *file, struct poll_table_struct *pts)
236 {
237 	struct proc_dir_entry *pde = PDE(file_inode(file));
238 	unsigned int rv = DEFAULT_POLLMASK;
239 	unsigned int (*poll)(struct file *, struct poll_table_struct *);
240 	if (use_pde(pde)) {
241 		poll = pde->proc_fops->poll;
242 		if (poll)
243 			rv = poll(file, pts);
244 		unuse_pde(pde);
245 	}
246 	return rv;
247 }
248 
249 static long proc_reg_unlocked_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
250 {
251 	struct proc_dir_entry *pde = PDE(file_inode(file));
252 	long rv = -ENOTTY;
253 	long (*ioctl)(struct file *, unsigned int, unsigned long);
254 	if (use_pde(pde)) {
255 		ioctl = pde->proc_fops->unlocked_ioctl;
256 		if (ioctl)
257 			rv = ioctl(file, cmd, arg);
258 		unuse_pde(pde);
259 	}
260 	return rv;
261 }
262 
263 #ifdef CONFIG_COMPAT
264 static long proc_reg_compat_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
265 {
266 	struct proc_dir_entry *pde = PDE(file_inode(file));
267 	long rv = -ENOTTY;
268 	long (*compat_ioctl)(struct file *, unsigned int, unsigned long);
269 	if (use_pde(pde)) {
270 		compat_ioctl = pde->proc_fops->compat_ioctl;
271 		if (compat_ioctl)
272 			rv = compat_ioctl(file, cmd, arg);
273 		unuse_pde(pde);
274 	}
275 	return rv;
276 }
277 #endif
278 
279 static int proc_reg_mmap(struct file *file, struct vm_area_struct *vma)
280 {
281 	struct proc_dir_entry *pde = PDE(file_inode(file));
282 	int rv = -EIO;
283 	int (*mmap)(struct file *, struct vm_area_struct *);
284 	if (use_pde(pde)) {
285 		mmap = pde->proc_fops->mmap;
286 		if (mmap)
287 			rv = mmap(file, vma);
288 		unuse_pde(pde);
289 	}
290 	return rv;
291 }
292 
293 static unsigned long
294 proc_reg_get_unmapped_area(struct file *file, unsigned long orig_addr,
295 			   unsigned long len, unsigned long pgoff,
296 			   unsigned long flags)
297 {
298 	struct proc_dir_entry *pde = PDE(file_inode(file));
299 	unsigned long rv = -EIO;
300 
301 	if (use_pde(pde)) {
302 		typeof(proc_reg_get_unmapped_area) *get_area;
303 
304 		get_area = pde->proc_fops->get_unmapped_area;
305 #ifdef CONFIG_MMU
306 		if (!get_area)
307 			get_area = current->mm->get_unmapped_area;
308 #endif
309 
310 		if (get_area)
311 			rv = get_area(file, orig_addr, len, pgoff, flags);
312 		else
313 			rv = orig_addr;
314 		unuse_pde(pde);
315 	}
316 	return rv;
317 }
318 
319 static int proc_reg_open(struct inode *inode, struct file *file)
320 {
321 	struct proc_dir_entry *pde = PDE(inode);
322 	int rv = 0;
323 	int (*open)(struct inode *, struct file *);
324 	int (*release)(struct inode *, struct file *);
325 	struct pde_opener *pdeo;
326 
327 	/*
328 	 * Ensure that
329 	 * 1) PDE's ->release hook will be called no matter what
330 	 *    either normally by close()/->release, or forcefully by
331 	 *    rmmod/remove_proc_entry.
332 	 *
333 	 * 2) rmmod isn't blocked by opening file in /proc and sitting on
334 	 *    the descriptor (including "rmmod foo </proc/foo" scenario).
335 	 *
336 	 * Save every "struct file" with custom ->release hook.
337 	 */
338 	pdeo = kmalloc(sizeof(struct pde_opener), GFP_KERNEL);
339 	if (!pdeo)
340 		return -ENOMEM;
341 
342 	if (!use_pde(pde)) {
343 		kfree(pdeo);
344 		return -ENOENT;
345 	}
346 	open = pde->proc_fops->open;
347 	release = pde->proc_fops->release;
348 
349 	if (open)
350 		rv = open(inode, file);
351 
352 	if (rv == 0 && release) {
353 		/* To know what to release. */
354 		pdeo->file = file;
355 		pdeo->closing = false;
356 		pdeo->c = NULL;
357 		spin_lock(&pde->pde_unload_lock);
358 		list_add(&pdeo->lh, &pde->pde_openers);
359 		spin_unlock(&pde->pde_unload_lock);
360 	} else
361 		kfree(pdeo);
362 
363 	unuse_pde(pde);
364 	return rv;
365 }
366 
367 static int proc_reg_release(struct inode *inode, struct file *file)
368 {
369 	struct proc_dir_entry *pde = PDE(inode);
370 	struct pde_opener *pdeo;
371 	spin_lock(&pde->pde_unload_lock);
372 	list_for_each_entry(pdeo, &pde->pde_openers, lh) {
373 		if (pdeo->file == file) {
374 			close_pdeo(pde, pdeo);
375 			break;
376 		}
377 	}
378 	spin_unlock(&pde->pde_unload_lock);
379 	return 0;
380 }
381 
382 static const struct file_operations proc_reg_file_ops = {
383 	.llseek		= proc_reg_llseek,
384 	.read		= proc_reg_read,
385 	.write		= proc_reg_write,
386 	.poll		= proc_reg_poll,
387 	.unlocked_ioctl	= proc_reg_unlocked_ioctl,
388 #ifdef CONFIG_COMPAT
389 	.compat_ioctl	= proc_reg_compat_ioctl,
390 #endif
391 	.mmap		= proc_reg_mmap,
392 	.get_unmapped_area = proc_reg_get_unmapped_area,
393 	.open		= proc_reg_open,
394 	.release	= proc_reg_release,
395 };
396 
397 #ifdef CONFIG_COMPAT
398 static const struct file_operations proc_reg_file_ops_no_compat = {
399 	.llseek		= proc_reg_llseek,
400 	.read		= proc_reg_read,
401 	.write		= proc_reg_write,
402 	.poll		= proc_reg_poll,
403 	.unlocked_ioctl	= proc_reg_unlocked_ioctl,
404 	.mmap		= proc_reg_mmap,
405 	.get_unmapped_area = proc_reg_get_unmapped_area,
406 	.open		= proc_reg_open,
407 	.release	= proc_reg_release,
408 };
409 #endif
410 
411 static void proc_put_link(void *p)
412 {
413 	unuse_pde(p);
414 }
415 
416 static const char *proc_get_link(struct dentry *dentry,
417 				 struct inode *inode,
418 				 struct delayed_call *done)
419 {
420 	struct proc_dir_entry *pde = PDE(inode);
421 	if (unlikely(!use_pde(pde)))
422 		return ERR_PTR(-EINVAL);
423 	set_delayed_call(done, proc_put_link, pde);
424 	return pde->data;
425 }
426 
427 const struct inode_operations proc_link_inode_operations = {
428 	.get_link	= proc_get_link,
429 };
430 
431 struct inode *proc_get_inode(struct super_block *sb, struct proc_dir_entry *de)
432 {
433 	struct inode *inode = new_inode_pseudo(sb);
434 
435 	if (inode) {
436 		inode->i_ino = de->low_ino;
437 		inode->i_mtime = inode->i_atime = inode->i_ctime = current_time(inode);
438 		PROC_I(inode)->pde = de;
439 
440 		if (is_empty_pde(de)) {
441 			make_empty_dir_inode(inode);
442 			return inode;
443 		}
444 		if (de->mode) {
445 			inode->i_mode = de->mode;
446 			inode->i_uid = de->uid;
447 			inode->i_gid = de->gid;
448 		}
449 		if (de->size)
450 			inode->i_size = de->size;
451 		if (de->nlink)
452 			set_nlink(inode, de->nlink);
453 		WARN_ON(!de->proc_iops);
454 		inode->i_op = de->proc_iops;
455 		if (de->proc_fops) {
456 			if (S_ISREG(inode->i_mode)) {
457 #ifdef CONFIG_COMPAT
458 				if (!de->proc_fops->compat_ioctl)
459 					inode->i_fop =
460 						&proc_reg_file_ops_no_compat;
461 				else
462 #endif
463 					inode->i_fop = &proc_reg_file_ops;
464 			} else {
465 				inode->i_fop = de->proc_fops;
466 			}
467 		}
468 	} else
469 	       pde_put(de);
470 	return inode;
471 }
472 
473 int proc_fill_super(struct super_block *s, void *data, int silent)
474 {
475 	struct pid_namespace *ns = get_pid_ns(s->s_fs_info);
476 	struct inode *root_inode;
477 	int ret;
478 
479 	if (!proc_parse_options(data, ns))
480 		return -EINVAL;
481 
482 	/* User space would break if executables or devices appear on proc */
483 	s->s_iflags |= SB_I_USERNS_VISIBLE | SB_I_NOEXEC | SB_I_NODEV;
484 	s->s_flags |= MS_NODIRATIME | MS_NOSUID | MS_NOEXEC;
485 	s->s_blocksize = 1024;
486 	s->s_blocksize_bits = 10;
487 	s->s_magic = PROC_SUPER_MAGIC;
488 	s->s_op = &proc_sops;
489 	s->s_time_gran = 1;
490 
491 	/*
492 	 * procfs isn't actually a stacking filesystem; however, there is
493 	 * too much magic going on inside it to permit stacking things on
494 	 * top of it
495 	 */
496 	s->s_stack_depth = FILESYSTEM_MAX_STACK_DEPTH;
497 
498 	pde_get(&proc_root);
499 	root_inode = proc_get_inode(s, &proc_root);
500 	if (!root_inode) {
501 		pr_err("proc_fill_super: get root inode failed\n");
502 		return -ENOMEM;
503 	}
504 
505 	s->s_root = d_make_root(root_inode);
506 	if (!s->s_root) {
507 		pr_err("proc_fill_super: allocate dentry failed\n");
508 		return -ENOMEM;
509 	}
510 
511 	ret = proc_setup_self(s);
512 	if (ret) {
513 		return ret;
514 	}
515 	return proc_setup_thread_self(s);
516 }
517