xref: /openbmc/linux/fs/proc/inode.c (revision 890f0b0d) 
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_ops->proc_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 		typeof_member(struct proc_ops, proc_lseek) lseek;
204 
205 		lseek = pde->proc_ops->proc_lseek;
206 		if (!lseek)
207 			lseek = default_llseek;
208 		rv = lseek(file, offset, whence);
209 		unuse_pde(pde);
210 	}
211 	return rv;
212 }
213 
214 static ssize_t proc_reg_read(struct file *file, char __user *buf, size_t count, loff_t *ppos)
215 {
216 	struct proc_dir_entry *pde = PDE(file_inode(file));
217 	ssize_t rv = -EIO;
218 	if (use_pde(pde)) {
219 		typeof_member(struct proc_ops, proc_read) read;
220 
221 		read = pde->proc_ops->proc_read;
222 		if (read)
223 			rv = read(file, buf, count, ppos);
224 		unuse_pde(pde);
225 	}
226 	return rv;
227 }
228 
229 static ssize_t proc_reg_write(struct file *file, const char __user *buf, size_t count, loff_t *ppos)
230 {
231 	struct proc_dir_entry *pde = PDE(file_inode(file));
232 	ssize_t rv = -EIO;
233 	if (use_pde(pde)) {
234 		typeof_member(struct proc_ops, proc_write) write;
235 
236 		write = pde->proc_ops->proc_write;
237 		if (write)
238 			rv = write(file, buf, count, ppos);
239 		unuse_pde(pde);
240 	}
241 	return rv;
242 }
243 
244 static __poll_t proc_reg_poll(struct file *file, struct poll_table_struct *pts)
245 {
246 	struct proc_dir_entry *pde = PDE(file_inode(file));
247 	__poll_t rv = DEFAULT_POLLMASK;
248 	if (use_pde(pde)) {
249 		typeof_member(struct proc_ops, proc_poll) poll;
250 
251 		poll = pde->proc_ops->proc_poll;
252 		if (poll)
253 			rv = poll(file, pts);
254 		unuse_pde(pde);
255 	}
256 	return rv;
257 }
258 
259 static long proc_reg_unlocked_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 	if (use_pde(pde)) {
264 		typeof_member(struct proc_ops, proc_ioctl) ioctl;
265 
266 		ioctl = pde->proc_ops->proc_ioctl;
267 		if (ioctl)
268 			rv = ioctl(file, cmd, arg);
269 		unuse_pde(pde);
270 	}
271 	return rv;
272 }
273 
274 #ifdef CONFIG_COMPAT
275 static long proc_reg_compat_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
276 {
277 	struct proc_dir_entry *pde = PDE(file_inode(file));
278 	long rv = -ENOTTY;
279 	if (use_pde(pde)) {
280 		typeof_member(struct proc_ops, proc_compat_ioctl) compat_ioctl;
281 
282 		compat_ioctl = pde->proc_ops->proc_compat_ioctl;
283 		if (compat_ioctl)
284 			rv = compat_ioctl(file, cmd, arg);
285 		unuse_pde(pde);
286 	}
287 	return rv;
288 }
289 #endif
290 
291 static int proc_reg_mmap(struct file *file, struct vm_area_struct *vma)
292 {
293 	struct proc_dir_entry *pde = PDE(file_inode(file));
294 	int rv = -EIO;
295 	if (use_pde(pde)) {
296 		typeof_member(struct proc_ops, proc_mmap) mmap;
297 
298 		mmap = pde->proc_ops->proc_mmap;
299 		if (mmap)
300 			rv = mmap(file, vma);
301 		unuse_pde(pde);
302 	}
303 	return rv;
304 }
305 
306 static unsigned long
307 proc_reg_get_unmapped_area(struct file *file, unsigned long orig_addr,
308 			   unsigned long len, unsigned long pgoff,
309 			   unsigned long flags)
310 {
311 	struct proc_dir_entry *pde = PDE(file_inode(file));
312 	unsigned long rv = -EIO;
313 
314 	if (use_pde(pde)) {
315 		typeof_member(struct proc_ops, proc_get_unmapped_area) get_area;
316 
317 		get_area = pde->proc_ops->proc_get_unmapped_area;
318 #ifdef CONFIG_MMU
319 		if (!get_area)
320 			get_area = current->mm->get_unmapped_area;
321 #endif
322 
323 		if (get_area)
324 			rv = get_area(file, orig_addr, len, pgoff, flags);
325 		else
326 			rv = orig_addr;
327 		unuse_pde(pde);
328 	}
329 	return rv;
330 }
331 
332 static int proc_reg_open(struct inode *inode, struct file *file)
333 {
334 	struct proc_dir_entry *pde = PDE(inode);
335 	int rv = 0;
336 	typeof_member(struct proc_ops, proc_open) open;
337 	typeof_member(struct proc_ops, proc_release) release;
338 	struct pde_opener *pdeo;
339 
340 	/*
341 	 * Ensure that
342 	 * 1) PDE's ->release hook will be called no matter what
343 	 *    either normally by close()/->release, or forcefully by
344 	 *    rmmod/remove_proc_entry.
345 	 *
346 	 * 2) rmmod isn't blocked by opening file in /proc and sitting on
347 	 *    the descriptor (including "rmmod foo </proc/foo" scenario).
348 	 *
349 	 * Save every "struct file" with custom ->release hook.
350 	 */
351 	if (!use_pde(pde))
352 		return -ENOENT;
353 
354 	release = pde->proc_ops->proc_release;
355 	if (release) {
356 		pdeo = kmem_cache_alloc(pde_opener_cache, GFP_KERNEL);
357 		if (!pdeo) {
358 			rv = -ENOMEM;
359 			goto out_unuse;
360 		}
361 	}
362 
363 	open = pde->proc_ops->proc_open;
364 	if (open)
365 		rv = open(inode, file);
366 
367 	if (release) {
368 		if (rv == 0) {
369 			/* To know what to release. */
370 			pdeo->file = file;
371 			pdeo->closing = false;
372 			pdeo->c = NULL;
373 			spin_lock(&pde->pde_unload_lock);
374 			list_add(&pdeo->lh, &pde->pde_openers);
375 			spin_unlock(&pde->pde_unload_lock);
376 		} else
377 			kmem_cache_free(pde_opener_cache, pdeo);
378 	}
379 
380 out_unuse:
381 	unuse_pde(pde);
382 	return rv;
383 }
384 
385 static int proc_reg_release(struct inode *inode, struct file *file)
386 {
387 	struct proc_dir_entry *pde = PDE(inode);
388 	struct pde_opener *pdeo;
389 	spin_lock(&pde->pde_unload_lock);
390 	list_for_each_entry(pdeo, &pde->pde_openers, lh) {
391 		if (pdeo->file == file) {
392 			close_pdeo(pde, pdeo);
393 			return 0;
394 		}
395 	}
396 	spin_unlock(&pde->pde_unload_lock);
397 	return 0;
398 }
399 
400 static const struct file_operations proc_reg_file_ops = {
401 	.llseek		= proc_reg_llseek,
402 	.read		= proc_reg_read,
403 	.write		= proc_reg_write,
404 	.poll		= proc_reg_poll,
405 	.unlocked_ioctl	= proc_reg_unlocked_ioctl,
406 #ifdef CONFIG_COMPAT
407 	.compat_ioctl	= proc_reg_compat_ioctl,
408 #endif
409 	.mmap		= proc_reg_mmap,
410 	.get_unmapped_area = proc_reg_get_unmapped_area,
411 	.open		= proc_reg_open,
412 	.release	= proc_reg_release,
413 };
414 
415 #ifdef CONFIG_COMPAT
416 static const struct file_operations proc_reg_file_ops_no_compat = {
417 	.llseek		= proc_reg_llseek,
418 	.read		= proc_reg_read,
419 	.write		= proc_reg_write,
420 	.poll		= proc_reg_poll,
421 	.unlocked_ioctl	= proc_reg_unlocked_ioctl,
422 	.mmap		= proc_reg_mmap,
423 	.get_unmapped_area = proc_reg_get_unmapped_area,
424 	.open		= proc_reg_open,
425 	.release	= proc_reg_release,
426 };
427 #endif
428 
429 static void proc_put_link(void *p)
430 {
431 	unuse_pde(p);
432 }
433 
434 static const char *proc_get_link(struct dentry *dentry,
435 				 struct inode *inode,
436 				 struct delayed_call *done)
437 {
438 	struct proc_dir_entry *pde = PDE(inode);
439 	if (!use_pde(pde))
440 		return ERR_PTR(-EINVAL);
441 	set_delayed_call(done, proc_put_link, pde);
442 	return pde->data;
443 }
444 
445 const struct inode_operations proc_link_inode_operations = {
446 	.get_link	= proc_get_link,
447 };
448 
449 struct inode *proc_get_inode(struct super_block *sb, struct proc_dir_entry *de)
450 {
451 	struct inode *inode = new_inode_pseudo(sb);
452 
453 	if (inode) {
454 		inode->i_ino = de->low_ino;
455 		inode->i_mtime = inode->i_atime = inode->i_ctime = current_time(inode);
456 		PROC_I(inode)->pde = de;
457 
458 		if (is_empty_pde(de)) {
459 			make_empty_dir_inode(inode);
460 			return inode;
461 		}
462 		if (de->mode) {
463 			inode->i_mode = de->mode;
464 			inode->i_uid = de->uid;
465 			inode->i_gid = de->gid;
466 		}
467 		if (de->size)
468 			inode->i_size = de->size;
469 		if (de->nlink)
470 			set_nlink(inode, de->nlink);
471 
472 		if (S_ISREG(inode->i_mode)) {
473 			inode->i_op = de->proc_iops;
474 			inode->i_fop = &proc_reg_file_ops;
475 #ifdef CONFIG_COMPAT
476 			if (!de->proc_ops->proc_compat_ioctl) {
477 				inode->i_fop = &proc_reg_file_ops_no_compat;
478 			}
479 #endif
480 		} else if (S_ISDIR(inode->i_mode)) {
481 			inode->i_op = de->proc_iops;
482 			inode->i_fop = de->proc_dir_ops;
483 		} else if (S_ISLNK(inode->i_mode)) {
484 			inode->i_op = de->proc_iops;
485 			inode->i_fop = NULL;
486 		} else
487 			BUG();
488 	} else
489 	       pde_put(de);
490 	return inode;
491 }
492