xref: /linux/fs/proc/inode.c (revision a5f6ea29f9a918403629f8369ae55fac6b09cb53) 
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 <asm/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 	const struct proc_ns_operations *ns_ops;
36 	void *ns;
37 
38 	truncate_inode_pages(&inode->i_data, 0);
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 = PROC_I(inode)->pde;
46 	if (de)
47 		pde_put(de);
48 	head = PROC_I(inode)->sysctl;
49 	if (head) {
50 		rcu_assign_pointer(PROC_I(inode)->sysctl, NULL);
51 		sysctl_head_put(head);
52 	}
53 	/* Release any associated namespace */
54 	ns_ops = PROC_I(inode)->ns.ns_ops;
55 	ns = PROC_I(inode)->ns.ns;
56 	if (ns_ops && ns)
57 		ns_ops->put(ns);
58 }
59 
60 static struct kmem_cache * proc_inode_cachep;
61 
62 static struct inode *proc_alloc_inode(struct super_block *sb)
63 {
64 	struct proc_inode *ei;
65 	struct inode *inode;
66 
67 	ei = (struct proc_inode *)kmem_cache_alloc(proc_inode_cachep, GFP_KERNEL);
68 	if (!ei)
69 		return NULL;
70 	ei->pid = NULL;
71 	ei->fd = 0;
72 	ei->op.proc_get_link = NULL;
73 	ei->pde = NULL;
74 	ei->sysctl = NULL;
75 	ei->sysctl_entry = NULL;
76 	ei->ns.ns = NULL;
77 	ei->ns.ns_ops = NULL;
78 	inode = &ei->vfs_inode;
79 	inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
80 	return inode;
81 }
82 
83 static void proc_i_callback(struct rcu_head *head)
84 {
85 	struct inode *inode = container_of(head, struct inode, i_rcu);
86 	kmem_cache_free(proc_inode_cachep, PROC_I(inode));
87 }
88 
89 static void proc_destroy_inode(struct inode *inode)
90 {
91 	call_rcu(&inode->i_rcu, proc_i_callback);
92 }
93 
94 static void init_once(void *foo)
95 {
96 	struct proc_inode *ei = (struct proc_inode *) foo;
97 
98 	inode_init_once(&ei->vfs_inode);
99 }
100 
101 void __init proc_init_inodecache(void)
102 {
103 	proc_inode_cachep = kmem_cache_create("proc_inode_cache",
104 					     sizeof(struct proc_inode),
105 					     0, (SLAB_RECLAIM_ACCOUNT|
106 						SLAB_MEM_SPREAD|SLAB_PANIC),
107 					     init_once);
108 }
109 
110 static int proc_show_options(struct seq_file *seq, struct dentry *root)
111 {
112 	struct super_block *sb = root->d_sb;
113 	struct pid_namespace *pid = sb->s_fs_info;
114 
115 	if (!gid_eq(pid->pid_gid, GLOBAL_ROOT_GID))
116 		seq_printf(seq, ",gid=%u", from_kgid_munged(&init_user_ns, pid->pid_gid));
117 	if (pid->hide_pid != 0)
118 		seq_printf(seq, ",hidepid=%u", pid->hide_pid);
119 
120 	return 0;
121 }
122 
123 static const struct super_operations proc_sops = {
124 	.alloc_inode	= proc_alloc_inode,
125 	.destroy_inode	= proc_destroy_inode,
126 	.drop_inode	= generic_delete_inode,
127 	.evict_inode	= proc_evict_inode,
128 	.statfs		= simple_statfs,
129 	.remount_fs	= proc_remount,
130 	.show_options	= proc_show_options,
131 };
132 
133 enum {BIAS = -1U<<31};
134 
135 static inline int use_pde(struct proc_dir_entry *pde)
136 {
137 	return atomic_inc_unless_negative(&pde->in_use);
138 }
139 
140 static void unuse_pde(struct proc_dir_entry *pde)
141 {
142 	if (atomic_dec_return(&pde->in_use) == BIAS)
143 		complete(pde->pde_unload_completion);
144 }
145 
146 /* pde is locked */
147 static void close_pdeo(struct proc_dir_entry *pde, struct pde_opener *pdeo)
148 {
149 	if (pdeo->closing) {
150 		/* somebody else is doing that, just wait */
151 		DECLARE_COMPLETION_ONSTACK(c);
152 		pdeo->c = &c;
153 		spin_unlock(&pde->pde_unload_lock);
154 		wait_for_completion(&c);
155 		spin_lock(&pde->pde_unload_lock);
156 	} else {
157 		struct file *file;
158 		pdeo->closing = 1;
159 		spin_unlock(&pde->pde_unload_lock);
160 		file = pdeo->file;
161 		pde->proc_fops->release(file_inode(file), file);
162 		spin_lock(&pde->pde_unload_lock);
163 		list_del_init(&pdeo->lh);
164 		if (pdeo->c)
165 			complete(pdeo->c);
166 		kfree(pdeo);
167 	}
168 }
169 
170 void proc_entry_rundown(struct proc_dir_entry *de)
171 {
172 	DECLARE_COMPLETION_ONSTACK(c);
173 	/* Wait until all existing callers into module are done. */
174 	de->pde_unload_completion = &c;
175 	if (atomic_add_return(BIAS, &de->in_use) != BIAS)
176 		wait_for_completion(&c);
177 
178 	spin_lock(&de->pde_unload_lock);
179 	while (!list_empty(&de->pde_openers)) {
180 		struct pde_opener *pdeo;
181 		pdeo = list_first_entry(&de->pde_openers, struct pde_opener, lh);
182 		close_pdeo(de, pdeo);
183 	}
184 	spin_unlock(&de->pde_unload_lock);
185 }
186 
187 static loff_t proc_reg_llseek(struct file *file, loff_t offset, int whence)
188 {
189 	struct proc_dir_entry *pde = PDE(file_inode(file));
190 	loff_t rv = -EINVAL;
191 	if (use_pde(pde)) {
192 		loff_t (*llseek)(struct file *, loff_t, int);
193 		llseek = pde->proc_fops->llseek;
194 		if (!llseek)
195 			llseek = default_llseek;
196 		rv = llseek(file, offset, whence);
197 		unuse_pde(pde);
198 	}
199 	return rv;
200 }
201 
202 static ssize_t proc_reg_read(struct file *file, char __user *buf, size_t count, loff_t *ppos)
203 {
204 	ssize_t (*read)(struct file *, char __user *, size_t, loff_t *);
205 	struct proc_dir_entry *pde = PDE(file_inode(file));
206 	ssize_t rv = -EIO;
207 	if (use_pde(pde)) {
208 		read = pde->proc_fops->read;
209 		if (read)
210 			rv = read(file, buf, count, ppos);
211 		unuse_pde(pde);
212 	}
213 	return rv;
214 }
215 
216 static ssize_t proc_reg_write(struct file *file, const char __user *buf, size_t count, loff_t *ppos)
217 {
218 	ssize_t (*write)(struct file *, const char __user *, size_t, loff_t *);
219 	struct proc_dir_entry *pde = PDE(file_inode(file));
220 	ssize_t rv = -EIO;
221 	if (use_pde(pde)) {
222 		write = pde->proc_fops->write;
223 		if (write)
224 			rv = write(file, buf, count, ppos);
225 		unuse_pde(pde);
226 	}
227 	return rv;
228 }
229 
230 static unsigned int proc_reg_poll(struct file *file, struct poll_table_struct *pts)
231 {
232 	struct proc_dir_entry *pde = PDE(file_inode(file));
233 	unsigned int rv = DEFAULT_POLLMASK;
234 	unsigned int (*poll)(struct file *, struct poll_table_struct *);
235 	if (use_pde(pde)) {
236 		poll = pde->proc_fops->poll;
237 		if (poll)
238 			rv = poll(file, pts);
239 		unuse_pde(pde);
240 	}
241 	return rv;
242 }
243 
244 static long proc_reg_unlocked_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
245 {
246 	struct proc_dir_entry *pde = PDE(file_inode(file));
247 	long rv = -ENOTTY;
248 	long (*ioctl)(struct file *, unsigned int, unsigned long);
249 	if (use_pde(pde)) {
250 		ioctl = pde->proc_fops->unlocked_ioctl;
251 		if (ioctl)
252 			rv = ioctl(file, cmd, arg);
253 		unuse_pde(pde);
254 	}
255 	return rv;
256 }
257 
258 #ifdef CONFIG_COMPAT
259 static long proc_reg_compat_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
260 {
261 	struct proc_dir_entry *pde = PDE(file_inode(file));
262 	long rv = -ENOTTY;
263 	long (*compat_ioctl)(struct file *, unsigned int, unsigned long);
264 	if (use_pde(pde)) {
265 		compat_ioctl = pde->proc_fops->compat_ioctl;
266 		if (compat_ioctl)
267 			rv = compat_ioctl(file, cmd, arg);
268 		unuse_pde(pde);
269 	}
270 	return rv;
271 }
272 #endif
273 
274 static int proc_reg_mmap(struct file *file, struct vm_area_struct *vma)
275 {
276 	struct proc_dir_entry *pde = PDE(file_inode(file));
277 	int rv = -EIO;
278 	int (*mmap)(struct file *, struct vm_area_struct *);
279 	if (use_pde(pde)) {
280 		mmap = pde->proc_fops->mmap;
281 		if (mmap)
282 			rv = mmap(file, vma);
283 		unuse_pde(pde);
284 	}
285 	return rv;
286 }
287 
288 static unsigned long
289 proc_reg_get_unmapped_area(struct file *file, unsigned long orig_addr,
290 			   unsigned long len, unsigned long pgoff,
291 			   unsigned long flags)
292 {
293 	struct proc_dir_entry *pde = PDE(file_inode(file));
294 	unsigned long rv = -EIO;
295 
296 	if (use_pde(pde)) {
297 		typeof(proc_reg_get_unmapped_area) *get_area;
298 
299 		get_area = pde->proc_fops->get_unmapped_area;
300 #ifdef CONFIG_MMU
301 		if (!get_area)
302 			get_area = current->mm->get_unmapped_area;
303 #endif
304 
305 		if (get_area)
306 			rv = get_area(file, orig_addr, len, pgoff, flags);
307 		else
308 			rv = orig_addr;
309 		unuse_pde(pde);
310 	}
311 	return rv;
312 }
313 
314 static int proc_reg_open(struct inode *inode, struct file *file)
315 {
316 	struct proc_dir_entry *pde = PDE(inode);
317 	int rv = 0;
318 	int (*open)(struct inode *, struct file *);
319 	int (*release)(struct inode *, struct file *);
320 	struct pde_opener *pdeo;
321 
322 	/*
323 	 * What for, you ask? Well, we can have open, rmmod, remove_proc_entry
324 	 * sequence. ->release won't be called because ->proc_fops will be
325 	 * cleared. Depending on complexity of ->release, consequences vary.
326 	 *
327 	 * We can't wait for mercy when close will be done for real, it's
328 	 * deadlockable: rmmod foo </proc/foo . So, we're going to do ->release
329 	 * by hand in remove_proc_entry(). For this, save opener's credentials
330 	 * for later.
331 	 */
332 	pdeo = kzalloc(sizeof(struct pde_opener), GFP_KERNEL);
333 	if (!pdeo)
334 		return -ENOMEM;
335 
336 	if (!use_pde(pde)) {
337 		kfree(pdeo);
338 		return -ENOENT;
339 	}
340 	open = pde->proc_fops->open;
341 	release = pde->proc_fops->release;
342 
343 	if (open)
344 		rv = open(inode, file);
345 
346 	if (rv == 0 && release) {
347 		/* To know what to release. */
348 		pdeo->file = file;
349 		/* Strictly for "too late" ->release in proc_reg_release(). */
350 		spin_lock(&pde->pde_unload_lock);
351 		list_add(&pdeo->lh, &pde->pde_openers);
352 		spin_unlock(&pde->pde_unload_lock);
353 	} else
354 		kfree(pdeo);
355 
356 	unuse_pde(pde);
357 	return rv;
358 }
359 
360 static int proc_reg_release(struct inode *inode, struct file *file)
361 {
362 	struct proc_dir_entry *pde = PDE(inode);
363 	struct pde_opener *pdeo;
364 	spin_lock(&pde->pde_unload_lock);
365 	list_for_each_entry(pdeo, &pde->pde_openers, lh) {
366 		if (pdeo->file == file) {
367 			close_pdeo(pde, pdeo);
368 			break;
369 		}
370 	}
371 	spin_unlock(&pde->pde_unload_lock);
372 	return 0;
373 }
374 
375 static const struct file_operations proc_reg_file_ops = {
376 	.llseek		= proc_reg_llseek,
377 	.read		= proc_reg_read,
378 	.write		= proc_reg_write,
379 	.poll		= proc_reg_poll,
380 	.unlocked_ioctl	= proc_reg_unlocked_ioctl,
381 #ifdef CONFIG_COMPAT
382 	.compat_ioctl	= proc_reg_compat_ioctl,
383 #endif
384 	.mmap		= proc_reg_mmap,
385 	.get_unmapped_area = proc_reg_get_unmapped_area,
386 	.open		= proc_reg_open,
387 	.release	= proc_reg_release,
388 };
389 
390 #ifdef CONFIG_COMPAT
391 static const struct file_operations proc_reg_file_ops_no_compat = {
392 	.llseek		= proc_reg_llseek,
393 	.read		= proc_reg_read,
394 	.write		= proc_reg_write,
395 	.poll		= proc_reg_poll,
396 	.unlocked_ioctl	= proc_reg_unlocked_ioctl,
397 	.mmap		= proc_reg_mmap,
398 	.get_unmapped_area = proc_reg_get_unmapped_area,
399 	.open		= proc_reg_open,
400 	.release	= proc_reg_release,
401 };
402 #endif
403 
404 struct inode *proc_get_inode(struct super_block *sb, struct proc_dir_entry *de)
405 {
406 	struct inode *inode = new_inode_pseudo(sb);
407 
408 	if (inode) {
409 		inode->i_ino = de->low_ino;
410 		inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
411 		PROC_I(inode)->pde = de;
412 
413 		if (de->mode) {
414 			inode->i_mode = de->mode;
415 			inode->i_uid = de->uid;
416 			inode->i_gid = de->gid;
417 		}
418 		if (de->size)
419 			inode->i_size = de->size;
420 		if (de->nlink)
421 			set_nlink(inode, de->nlink);
422 		WARN_ON(!de->proc_iops);
423 		inode->i_op = de->proc_iops;
424 		if (de->proc_fops) {
425 			if (S_ISREG(inode->i_mode)) {
426 #ifdef CONFIG_COMPAT
427 				if (!de->proc_fops->compat_ioctl)
428 					inode->i_fop =
429 						&proc_reg_file_ops_no_compat;
430 				else
431 #endif
432 					inode->i_fop = &proc_reg_file_ops;
433 			} else {
434 				inode->i_fop = de->proc_fops;
435 			}
436 		}
437 	} else
438 	       pde_put(de);
439 	return inode;
440 }
441 
442 int proc_fill_super(struct super_block *s)
443 {
444 	struct inode *root_inode;
445 
446 	s->s_flags |= MS_NODIRATIME | MS_NOSUID | MS_NOEXEC;
447 	s->s_blocksize = 1024;
448 	s->s_blocksize_bits = 10;
449 	s->s_magic = PROC_SUPER_MAGIC;
450 	s->s_op = &proc_sops;
451 	s->s_time_gran = 1;
452 
453 	pde_get(&proc_root);
454 	root_inode = proc_get_inode(s, &proc_root);
455 	if (!root_inode) {
456 		pr_err("proc_fill_super: get root inode failed\n");
457 		return -ENOMEM;
458 	}
459 
460 	s->s_root = d_make_root(root_inode);
461 	if (!s->s_root) {
462 		pr_err("proc_fill_super: allocate dentry failed\n");
463 		return -ENOMEM;
464 	}
465 
466 	return proc_setup_self(s);
467 }
468