Loading...
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#include <linux/bug.h>
28
29#include "internal.h"
30
31static void proc_evict_inode(struct inode *inode)
32{
33 struct proc_dir_entry *de;
34 struct ctl_table_header *head;
35 struct proc_inode *ei = PROC_I(inode);
36
37 truncate_inode_pages_final(&inode->i_data);
38 clear_inode(inode);
39
40 /* Stop tracking associated processes */
41 if (ei->pid) {
42 proc_pid_evict_inode(ei);
43 ei->pid = NULL;
44 }
45
46 /* Let go of any associated proc directory entry */
47 de = ei->pde;
48 if (de) {
49 pde_put(de);
50 ei->pde = NULL;
51 }
52
53 head = ei->sysctl;
54 if (head) {
55 RCU_INIT_POINTER(ei->sysctl, NULL);
56 proc_sys_evict_inode(inode, head);
57 }
58}
59
60static struct kmem_cache *proc_inode_cachep __ro_after_init;
61static struct kmem_cache *pde_opener_cache __ro_after_init;
62
63static struct inode *proc_alloc_inode(struct super_block *sb)
64{
65 struct proc_inode *ei;
66
67 ei = alloc_inode_sb(sb, 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 INIT_HLIST_NODE(&ei->sibling_inodes);
77 ei->ns_ops = NULL;
78 return &ei->vfs_inode;
79}
80
81static void proc_free_inode(struct inode *inode)
82{
83 kmem_cache_free(proc_inode_cachep, PROC_I(inode));
84}
85
86static 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
93void __init proc_init_kmemcache(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 pde_opener_cache =
102 kmem_cache_create("pde_opener", sizeof(struct pde_opener), 0,
103 SLAB_ACCOUNT|SLAB_PANIC, NULL);
104 proc_dir_entry_cache = kmem_cache_create_usercopy(
105 "proc_dir_entry", SIZEOF_PDE, 0, SLAB_PANIC,
106 offsetof(struct proc_dir_entry, inline_name),
107 SIZEOF_PDE_INLINE_NAME, NULL);
108 BUILD_BUG_ON(sizeof(struct proc_dir_entry) >= SIZEOF_PDE);
109}
110
111void proc_invalidate_siblings_dcache(struct hlist_head *inodes, spinlock_t *lock)
112{
113 struct inode *inode;
114 struct proc_inode *ei;
115 struct hlist_node *node;
116 struct super_block *old_sb = NULL;
117
118 rcu_read_lock();
119 for (;;) {
120 struct super_block *sb;
121 node = hlist_first_rcu(inodes);
122 if (!node)
123 break;
124 ei = hlist_entry(node, struct proc_inode, sibling_inodes);
125 spin_lock(lock);
126 hlist_del_init_rcu(&ei->sibling_inodes);
127 spin_unlock(lock);
128
129 inode = &ei->vfs_inode;
130 sb = inode->i_sb;
131 if ((sb != old_sb) && !atomic_inc_not_zero(&sb->s_active))
132 continue;
133 inode = igrab(inode);
134 rcu_read_unlock();
135 if (sb != old_sb) {
136 if (old_sb)
137 deactivate_super(old_sb);
138 old_sb = sb;
139 }
140 if (unlikely(!inode)) {
141 rcu_read_lock();
142 continue;
143 }
144
145 if (S_ISDIR(inode->i_mode)) {
146 struct dentry *dir = d_find_any_alias(inode);
147 if (dir) {
148 d_invalidate(dir);
149 dput(dir);
150 }
151 } else {
152 struct dentry *dentry;
153 while ((dentry = d_find_alias(inode))) {
154 d_invalidate(dentry);
155 dput(dentry);
156 }
157 }
158 iput(inode);
159
160 rcu_read_lock();
161 }
162 rcu_read_unlock();
163 if (old_sb)
164 deactivate_super(old_sb);
165}
166
167static inline const char *hidepid2str(enum proc_hidepid v)
168{
169 switch (v) {
170 case HIDEPID_OFF: return "off";
171 case HIDEPID_NO_ACCESS: return "noaccess";
172 case HIDEPID_INVISIBLE: return "invisible";
173 case HIDEPID_NOT_PTRACEABLE: return "ptraceable";
174 }
175 WARN_ONCE(1, "bad hide_pid value: %d\n", v);
176 return "unknown";
177}
178
179static int proc_show_options(struct seq_file *seq, struct dentry *root)
180{
181 struct proc_fs_info *fs_info = proc_sb_info(root->d_sb);
182
183 if (!gid_eq(fs_info->pid_gid, GLOBAL_ROOT_GID))
184 seq_printf(seq, ",gid=%u", from_kgid_munged(&init_user_ns, fs_info->pid_gid));
185 if (fs_info->hide_pid != HIDEPID_OFF)
186 seq_printf(seq, ",hidepid=%s", hidepid2str(fs_info->hide_pid));
187 if (fs_info->pidonly != PROC_PIDONLY_OFF)
188 seq_printf(seq, ",subset=pid");
189
190 return 0;
191}
192
193const struct super_operations proc_sops = {
194 .alloc_inode = proc_alloc_inode,
195 .free_inode = proc_free_inode,
196 .drop_inode = generic_delete_inode,
197 .evict_inode = proc_evict_inode,
198 .statfs = simple_statfs,
199 .show_options = proc_show_options,
200};
201
202enum {BIAS = -1U<<31};
203
204static inline int use_pde(struct proc_dir_entry *pde)
205{
206 return likely(atomic_inc_unless_negative(&pde->in_use));
207}
208
209static void unuse_pde(struct proc_dir_entry *pde)
210{
211 if (unlikely(atomic_dec_return(&pde->in_use) == BIAS))
212 complete(pde->pde_unload_completion);
213}
214
215/*
216 * At most 2 contexts can enter this function: the one doing the last
217 * close on the descriptor and whoever is deleting PDE itself.
218 *
219 * First to enter calls ->proc_release hook and signals its completion
220 * to the second one which waits and then does nothing.
221 *
222 * PDE is locked on entry, unlocked on exit.
223 */
224static void close_pdeo(struct proc_dir_entry *pde, struct pde_opener *pdeo)
225 __releases(&pde->pde_unload_lock)
226{
227 /*
228 * close() (proc_reg_release()) can't delete an entry and proceed:
229 * ->release hook needs to be available at the right moment.
230 *
231 * rmmod (remove_proc_entry() et al) can't delete an entry and proceed:
232 * "struct file" needs to be available at the right moment.
233 */
234 if (pdeo->closing) {
235 /* somebody else is doing that, just wait */
236 DECLARE_COMPLETION_ONSTACK(c);
237 pdeo->c = &c;
238 spin_unlock(&pde->pde_unload_lock);
239 wait_for_completion(&c);
240 } else {
241 struct file *file;
242 struct completion *c;
243
244 pdeo->closing = true;
245 spin_unlock(&pde->pde_unload_lock);
246
247 file = pdeo->file;
248 pde->proc_ops->proc_release(file_inode(file), file);
249
250 spin_lock(&pde->pde_unload_lock);
251 /* Strictly after ->proc_release, see above. */
252 list_del(&pdeo->lh);
253 c = pdeo->c;
254 spin_unlock(&pde->pde_unload_lock);
255 if (unlikely(c))
256 complete(c);
257 kmem_cache_free(pde_opener_cache, pdeo);
258 }
259}
260
261void proc_entry_rundown(struct proc_dir_entry *de)
262{
263 DECLARE_COMPLETION_ONSTACK(c);
264 /* Wait until all existing callers into module are done. */
265 de->pde_unload_completion = &c;
266 if (atomic_add_return(BIAS, &de->in_use) != BIAS)
267 wait_for_completion(&c);
268
269 /* ->pde_openers list can't grow from now on. */
270
271 spin_lock(&de->pde_unload_lock);
272 while (!list_empty(&de->pde_openers)) {
273 struct pde_opener *pdeo;
274 pdeo = list_first_entry(&de->pde_openers, struct pde_opener, lh);
275 close_pdeo(de, pdeo);
276 spin_lock(&de->pde_unload_lock);
277 }
278 spin_unlock(&de->pde_unload_lock);
279}
280
281static loff_t proc_reg_llseek(struct file *file, loff_t offset, int whence)
282{
283 struct proc_dir_entry *pde = PDE(file_inode(file));
284 loff_t rv = -EINVAL;
285
286 if (pde_is_permanent(pde)) {
287 return pde->proc_ops->proc_lseek(file, offset, whence);
288 } else if (use_pde(pde)) {
289 rv = pde->proc_ops->proc_lseek(file, offset, whence);
290 unuse_pde(pde);
291 }
292 return rv;
293}
294
295static ssize_t proc_reg_read_iter(struct kiocb *iocb, struct iov_iter *iter)
296{
297 struct proc_dir_entry *pde = PDE(file_inode(iocb->ki_filp));
298 ssize_t ret;
299
300 if (pde_is_permanent(pde))
301 return pde->proc_ops->proc_read_iter(iocb, iter);
302
303 if (!use_pde(pde))
304 return -EIO;
305 ret = pde->proc_ops->proc_read_iter(iocb, iter);
306 unuse_pde(pde);
307 return ret;
308}
309
310static ssize_t pde_read(struct proc_dir_entry *pde, struct file *file, char __user *buf, size_t count, loff_t *ppos)
311{
312 typeof_member(struct proc_ops, proc_read) read;
313
314 read = pde->proc_ops->proc_read;
315 if (read)
316 return read(file, buf, count, ppos);
317 return -EIO;
318}
319
320static ssize_t proc_reg_read(struct file *file, char __user *buf, size_t count, loff_t *ppos)
321{
322 struct proc_dir_entry *pde = PDE(file_inode(file));
323 ssize_t rv = -EIO;
324
325 if (pde_is_permanent(pde)) {
326 return pde_read(pde, file, buf, count, ppos);
327 } else if (use_pde(pde)) {
328 rv = pde_read(pde, file, buf, count, ppos);
329 unuse_pde(pde);
330 }
331 return rv;
332}
333
334static ssize_t pde_write(struct proc_dir_entry *pde, struct file *file, const char __user *buf, size_t count, loff_t *ppos)
335{
336 typeof_member(struct proc_ops, proc_write) write;
337
338 write = pde->proc_ops->proc_write;
339 if (write)
340 return write(file, buf, count, ppos);
341 return -EIO;
342}
343
344static ssize_t proc_reg_write(struct file *file, const char __user *buf, size_t count, loff_t *ppos)
345{
346 struct proc_dir_entry *pde = PDE(file_inode(file));
347 ssize_t rv = -EIO;
348
349 if (pde_is_permanent(pde)) {
350 return pde_write(pde, file, buf, count, ppos);
351 } else if (use_pde(pde)) {
352 rv = pde_write(pde, file, buf, count, ppos);
353 unuse_pde(pde);
354 }
355 return rv;
356}
357
358static __poll_t pde_poll(struct proc_dir_entry *pde, struct file *file, struct poll_table_struct *pts)
359{
360 typeof_member(struct proc_ops, proc_poll) poll;
361
362 poll = pde->proc_ops->proc_poll;
363 if (poll)
364 return poll(file, pts);
365 return DEFAULT_POLLMASK;
366}
367
368static __poll_t proc_reg_poll(struct file *file, struct poll_table_struct *pts)
369{
370 struct proc_dir_entry *pde = PDE(file_inode(file));
371 __poll_t rv = DEFAULT_POLLMASK;
372
373 if (pde_is_permanent(pde)) {
374 return pde_poll(pde, file, pts);
375 } else if (use_pde(pde)) {
376 rv = pde_poll(pde, file, pts);
377 unuse_pde(pde);
378 }
379 return rv;
380}
381
382static long pde_ioctl(struct proc_dir_entry *pde, struct file *file, unsigned int cmd, unsigned long arg)
383{
384 typeof_member(struct proc_ops, proc_ioctl) ioctl;
385
386 ioctl = pde->proc_ops->proc_ioctl;
387 if (ioctl)
388 return ioctl(file, cmd, arg);
389 return -ENOTTY;
390}
391
392static long proc_reg_unlocked_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
393{
394 struct proc_dir_entry *pde = PDE(file_inode(file));
395 long rv = -ENOTTY;
396
397 if (pde_is_permanent(pde)) {
398 return pde_ioctl(pde, file, cmd, arg);
399 } else if (use_pde(pde)) {
400 rv = pde_ioctl(pde, file, cmd, arg);
401 unuse_pde(pde);
402 }
403 return rv;
404}
405
406#ifdef CONFIG_COMPAT
407static long pde_compat_ioctl(struct proc_dir_entry *pde, struct file *file, unsigned int cmd, unsigned long arg)
408{
409 typeof_member(struct proc_ops, proc_compat_ioctl) compat_ioctl;
410
411 compat_ioctl = pde->proc_ops->proc_compat_ioctl;
412 if (compat_ioctl)
413 return compat_ioctl(file, cmd, arg);
414 return -ENOTTY;
415}
416
417static long proc_reg_compat_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
418{
419 struct proc_dir_entry *pde = PDE(file_inode(file));
420 long rv = -ENOTTY;
421 if (pde_is_permanent(pde)) {
422 return pde_compat_ioctl(pde, file, cmd, arg);
423 } else if (use_pde(pde)) {
424 rv = pde_compat_ioctl(pde, file, cmd, arg);
425 unuse_pde(pde);
426 }
427 return rv;
428}
429#endif
430
431static int pde_mmap(struct proc_dir_entry *pde, struct file *file, struct vm_area_struct *vma)
432{
433 typeof_member(struct proc_ops, proc_mmap) mmap;
434
435 mmap = pde->proc_ops->proc_mmap;
436 if (mmap)
437 return mmap(file, vma);
438 return -EIO;
439}
440
441static int proc_reg_mmap(struct file *file, struct vm_area_struct *vma)
442{
443 struct proc_dir_entry *pde = PDE(file_inode(file));
444 int rv = -EIO;
445
446 if (pde_is_permanent(pde)) {
447 return pde_mmap(pde, file, vma);
448 } else if (use_pde(pde)) {
449 rv = pde_mmap(pde, file, vma);
450 unuse_pde(pde);
451 }
452 return rv;
453}
454
455static unsigned long
456pde_get_unmapped_area(struct proc_dir_entry *pde, struct file *file, unsigned long orig_addr,
457 unsigned long len, unsigned long pgoff,
458 unsigned long flags)
459{
460 typeof_member(struct proc_ops, proc_get_unmapped_area) get_area;
461
462 get_area = pde->proc_ops->proc_get_unmapped_area;
463#ifdef CONFIG_MMU
464 if (!get_area)
465 get_area = current->mm->get_unmapped_area;
466#endif
467 if (get_area)
468 return get_area(file, orig_addr, len, pgoff, flags);
469 return orig_addr;
470}
471
472static unsigned long
473proc_reg_get_unmapped_area(struct file *file, unsigned long orig_addr,
474 unsigned long len, unsigned long pgoff,
475 unsigned long flags)
476{
477 struct proc_dir_entry *pde = PDE(file_inode(file));
478 unsigned long rv = -EIO;
479
480 if (pde_is_permanent(pde)) {
481 return pde_get_unmapped_area(pde, file, orig_addr, len, pgoff, flags);
482 } else if (use_pde(pde)) {
483 rv = pde_get_unmapped_area(pde, file, orig_addr, len, pgoff, flags);
484 unuse_pde(pde);
485 }
486 return rv;
487}
488
489static int proc_reg_open(struct inode *inode, struct file *file)
490{
491 struct proc_dir_entry *pde = PDE(inode);
492 int rv = 0;
493 typeof_member(struct proc_ops, proc_open) open;
494 typeof_member(struct proc_ops, proc_release) release;
495 struct pde_opener *pdeo;
496
497 if (!pde->proc_ops->proc_lseek)
498 file->f_mode &= ~FMODE_LSEEK;
499
500 if (pde_is_permanent(pde)) {
501 open = pde->proc_ops->proc_open;
502 if (open)
503 rv = open(inode, file);
504 return rv;
505 }
506
507 /*
508 * Ensure that
509 * 1) PDE's ->release hook will be called no matter what
510 * either normally by close()/->release, or forcefully by
511 * rmmod/remove_proc_entry.
512 *
513 * 2) rmmod isn't blocked by opening file in /proc and sitting on
514 * the descriptor (including "rmmod foo </proc/foo" scenario).
515 *
516 * Save every "struct file" with custom ->release hook.
517 */
518 if (!use_pde(pde))
519 return -ENOENT;
520
521 release = pde->proc_ops->proc_release;
522 if (release) {
523 pdeo = kmem_cache_alloc(pde_opener_cache, GFP_KERNEL);
524 if (!pdeo) {
525 rv = -ENOMEM;
526 goto out_unuse;
527 }
528 }
529
530 open = pde->proc_ops->proc_open;
531 if (open)
532 rv = open(inode, file);
533
534 if (release) {
535 if (rv == 0) {
536 /* To know what to release. */
537 pdeo->file = file;
538 pdeo->closing = false;
539 pdeo->c = NULL;
540 spin_lock(&pde->pde_unload_lock);
541 list_add(&pdeo->lh, &pde->pde_openers);
542 spin_unlock(&pde->pde_unload_lock);
543 } else
544 kmem_cache_free(pde_opener_cache, pdeo);
545 }
546
547out_unuse:
548 unuse_pde(pde);
549 return rv;
550}
551
552static int proc_reg_release(struct inode *inode, struct file *file)
553{
554 struct proc_dir_entry *pde = PDE(inode);
555 struct pde_opener *pdeo;
556
557 if (pde_is_permanent(pde)) {
558 typeof_member(struct proc_ops, proc_release) release;
559
560 release = pde->proc_ops->proc_release;
561 if (release) {
562 return release(inode, file);
563 }
564 return 0;
565 }
566
567 spin_lock(&pde->pde_unload_lock);
568 list_for_each_entry(pdeo, &pde->pde_openers, lh) {
569 if (pdeo->file == file) {
570 close_pdeo(pde, pdeo);
571 return 0;
572 }
573 }
574 spin_unlock(&pde->pde_unload_lock);
575 return 0;
576}
577
578static const struct file_operations proc_reg_file_ops = {
579 .llseek = proc_reg_llseek,
580 .read = proc_reg_read,
581 .write = proc_reg_write,
582 .poll = proc_reg_poll,
583 .unlocked_ioctl = proc_reg_unlocked_ioctl,
584 .mmap = proc_reg_mmap,
585 .get_unmapped_area = proc_reg_get_unmapped_area,
586 .open = proc_reg_open,
587 .release = proc_reg_release,
588};
589
590static const struct file_operations proc_iter_file_ops = {
591 .llseek = proc_reg_llseek,
592 .read_iter = proc_reg_read_iter,
593 .write = proc_reg_write,
594 .splice_read = generic_file_splice_read,
595 .poll = proc_reg_poll,
596 .unlocked_ioctl = proc_reg_unlocked_ioctl,
597 .mmap = proc_reg_mmap,
598 .get_unmapped_area = proc_reg_get_unmapped_area,
599 .open = proc_reg_open,
600 .release = proc_reg_release,
601};
602
603#ifdef CONFIG_COMPAT
604static const struct file_operations proc_reg_file_ops_compat = {
605 .llseek = proc_reg_llseek,
606 .read = proc_reg_read,
607 .write = proc_reg_write,
608 .poll = proc_reg_poll,
609 .unlocked_ioctl = proc_reg_unlocked_ioctl,
610 .compat_ioctl = proc_reg_compat_ioctl,
611 .mmap = proc_reg_mmap,
612 .get_unmapped_area = proc_reg_get_unmapped_area,
613 .open = proc_reg_open,
614 .release = proc_reg_release,
615};
616
617static const struct file_operations proc_iter_file_ops_compat = {
618 .llseek = proc_reg_llseek,
619 .read_iter = proc_reg_read_iter,
620 .splice_read = generic_file_splice_read,
621 .write = proc_reg_write,
622 .poll = proc_reg_poll,
623 .unlocked_ioctl = proc_reg_unlocked_ioctl,
624 .compat_ioctl = proc_reg_compat_ioctl,
625 .mmap = proc_reg_mmap,
626 .get_unmapped_area = proc_reg_get_unmapped_area,
627 .open = proc_reg_open,
628 .release = proc_reg_release,
629};
630#endif
631
632static void proc_put_link(void *p)
633{
634 unuse_pde(p);
635}
636
637static const char *proc_get_link(struct dentry *dentry,
638 struct inode *inode,
639 struct delayed_call *done)
640{
641 struct proc_dir_entry *pde = PDE(inode);
642 if (!use_pde(pde))
643 return ERR_PTR(-EINVAL);
644 set_delayed_call(done, proc_put_link, pde);
645 return pde->data;
646}
647
648const struct inode_operations proc_link_inode_operations = {
649 .get_link = proc_get_link,
650};
651
652struct inode *proc_get_inode(struct super_block *sb, struct proc_dir_entry *de)
653{
654 struct inode *inode = new_inode(sb);
655
656 if (!inode) {
657 pde_put(de);
658 return NULL;
659 }
660
661 inode->i_private = de->data;
662 inode->i_ino = de->low_ino;
663 inode->i_mtime = inode->i_atime = inode->i_ctime = current_time(inode);
664 PROC_I(inode)->pde = de;
665 if (is_empty_pde(de)) {
666 make_empty_dir_inode(inode);
667 return inode;
668 }
669
670 if (de->mode) {
671 inode->i_mode = de->mode;
672 inode->i_uid = de->uid;
673 inode->i_gid = de->gid;
674 }
675 if (de->size)
676 inode->i_size = de->size;
677 if (de->nlink)
678 set_nlink(inode, de->nlink);
679
680 if (S_ISREG(inode->i_mode)) {
681 inode->i_op = de->proc_iops;
682 if (de->proc_ops->proc_read_iter)
683 inode->i_fop = &proc_iter_file_ops;
684 else
685 inode->i_fop = &proc_reg_file_ops;
686#ifdef CONFIG_COMPAT
687 if (de->proc_ops->proc_compat_ioctl) {
688 if (de->proc_ops->proc_read_iter)
689 inode->i_fop = &proc_iter_file_ops_compat;
690 else
691 inode->i_fop = &proc_reg_file_ops_compat;
692 }
693#endif
694 } else if (S_ISDIR(inode->i_mode)) {
695 inode->i_op = de->proc_iops;
696 inode->i_fop = de->proc_dir_ops;
697 } else if (S_ISLNK(inode->i_mode)) {
698 inode->i_op = de->proc_iops;
699 inode->i_fop = NULL;
700 } else {
701 BUG();
702 }
703 return inode;
704}
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#include <linux/magic.h>
28
29#include <linux/uaccess.h>
30
31#include "internal.h"
32
33static void proc_evict_inode(struct inode *inode)
34{
35 struct proc_dir_entry *de;
36 struct ctl_table_header *head;
37
38 truncate_inode_pages_final(&inode->i_data);
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 = PDE(inode);
46 if (de)
47 pde_put(de);
48
49 head = PROC_I(inode)->sysctl;
50 if (head) {
51 RCU_INIT_POINTER(PROC_I(inode)->sysctl, NULL);
52 proc_sys_evict_inode(inode, head);
53 }
54}
55
56static struct kmem_cache *proc_inode_cachep __ro_after_init;
57static struct kmem_cache *pde_opener_cache __ro_after_init;
58
59static struct inode *proc_alloc_inode(struct super_block *sb)
60{
61 struct proc_inode *ei;
62 struct inode *inode;
63
64 ei = kmem_cache_alloc(proc_inode_cachep, GFP_KERNEL);
65 if (!ei)
66 return NULL;
67 ei->pid = NULL;
68 ei->fd = 0;
69 ei->op.proc_get_link = NULL;
70 ei->pde = NULL;
71 ei->sysctl = NULL;
72 ei->sysctl_entry = NULL;
73 ei->ns_ops = NULL;
74 inode = &ei->vfs_inode;
75 return inode;
76}
77
78static void proc_i_callback(struct rcu_head *head)
79{
80 struct inode *inode = container_of(head, struct inode, i_rcu);
81 kmem_cache_free(proc_inode_cachep, PROC_I(inode));
82}
83
84static void proc_destroy_inode(struct inode *inode)
85{
86 call_rcu(&inode->i_rcu, proc_i_callback);
87}
88
89static void init_once(void *foo)
90{
91 struct proc_inode *ei = (struct proc_inode *) foo;
92
93 inode_init_once(&ei->vfs_inode);
94}
95
96void __init proc_init_kmemcache(void)
97{
98 proc_inode_cachep = kmem_cache_create("proc_inode_cache",
99 sizeof(struct proc_inode),
100 0, (SLAB_RECLAIM_ACCOUNT|
101 SLAB_MEM_SPREAD|SLAB_ACCOUNT|
102 SLAB_PANIC),
103 init_once);
104 pde_opener_cache =
105 kmem_cache_create("pde_opener", sizeof(struct pde_opener), 0,
106 SLAB_ACCOUNT|SLAB_PANIC, NULL);
107 proc_dir_entry_cache = kmem_cache_create_usercopy(
108 "proc_dir_entry", sizeof(struct proc_dir_entry), 0, SLAB_PANIC,
109 offsetof(struct proc_dir_entry, inline_name),
110 sizeof_field(struct proc_dir_entry, inline_name), NULL);
111}
112
113static int proc_show_options(struct seq_file *seq, struct dentry *root)
114{
115 struct super_block *sb = root->d_sb;
116 struct pid_namespace *pid = sb->s_fs_info;
117
118 if (!gid_eq(pid->pid_gid, GLOBAL_ROOT_GID))
119 seq_printf(seq, ",gid=%u", from_kgid_munged(&init_user_ns, pid->pid_gid));
120 if (pid->hide_pid != HIDEPID_OFF)
121 seq_printf(seq, ",hidepid=%u", pid->hide_pid);
122
123 return 0;
124}
125
126static const struct super_operations proc_sops = {
127 .alloc_inode = proc_alloc_inode,
128 .destroy_inode = proc_destroy_inode,
129 .drop_inode = generic_delete_inode,
130 .evict_inode = proc_evict_inode,
131 .statfs = simple_statfs,
132 .remount_fs = proc_remount,
133 .show_options = proc_show_options,
134};
135
136enum {BIAS = -1U<<31};
137
138static inline int use_pde(struct proc_dir_entry *pde)
139{
140 return likely(atomic_inc_unless_negative(&pde->in_use));
141}
142
143static void unuse_pde(struct proc_dir_entry *pde)
144{
145 if (unlikely(atomic_dec_return(&pde->in_use) == BIAS))
146 complete(pde->pde_unload_completion);
147}
148
149/* pde is locked on entry, unlocked on exit */
150static void close_pdeo(struct proc_dir_entry *pde, struct pde_opener *pdeo)
151{
152 /*
153 * close() (proc_reg_release()) can't delete an entry and proceed:
154 * ->release hook needs to be available at the right moment.
155 *
156 * rmmod (remove_proc_entry() et al) can't delete an entry and proceed:
157 * "struct file" needs to be available at the right moment.
158 *
159 * Therefore, first process to enter this function does ->release() and
160 * signals its completion to the other process which does nothing.
161 */
162 if (pdeo->closing) {
163 /* somebody else is doing that, just wait */
164 DECLARE_COMPLETION_ONSTACK(c);
165 pdeo->c = &c;
166 spin_unlock(&pde->pde_unload_lock);
167 wait_for_completion(&c);
168 } else {
169 struct file *file;
170 struct completion *c;
171
172 pdeo->closing = true;
173 spin_unlock(&pde->pde_unload_lock);
174 file = pdeo->file;
175 pde->proc_fops->release(file_inode(file), file);
176 spin_lock(&pde->pde_unload_lock);
177 /* After ->release. */
178 list_del(&pdeo->lh);
179 c = pdeo->c;
180 spin_unlock(&pde->pde_unload_lock);
181 if (unlikely(c))
182 complete(c);
183 kmem_cache_free(pde_opener_cache, pdeo);
184 }
185}
186
187void proc_entry_rundown(struct proc_dir_entry *de)
188{
189 DECLARE_COMPLETION_ONSTACK(c);
190 /* Wait until all existing callers into module are done. */
191 de->pde_unload_completion = &c;
192 if (atomic_add_return(BIAS, &de->in_use) != BIAS)
193 wait_for_completion(&c);
194
195 /* ->pde_openers list can't grow from now on. */
196
197 spin_lock(&de->pde_unload_lock);
198 while (!list_empty(&de->pde_openers)) {
199 struct pde_opener *pdeo;
200 pdeo = list_first_entry(&de->pde_openers, struct pde_opener, lh);
201 close_pdeo(de, pdeo);
202 spin_lock(&de->pde_unload_lock);
203 }
204 spin_unlock(&de->pde_unload_lock);
205}
206
207static loff_t proc_reg_llseek(struct file *file, loff_t offset, int whence)
208{
209 struct proc_dir_entry *pde = PDE(file_inode(file));
210 loff_t rv = -EINVAL;
211 if (use_pde(pde)) {
212 loff_t (*llseek)(struct file *, loff_t, int);
213 llseek = pde->proc_fops->llseek;
214 if (!llseek)
215 llseek = default_llseek;
216 rv = llseek(file, offset, whence);
217 unuse_pde(pde);
218 }
219 return rv;
220}
221
222static ssize_t proc_reg_read(struct file *file, char __user *buf, size_t count, loff_t *ppos)
223{
224 ssize_t (*read)(struct file *, 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 read = pde->proc_fops->read;
229 if (read)
230 rv = read(file, buf, count, ppos);
231 unuse_pde(pde);
232 }
233 return rv;
234}
235
236static ssize_t proc_reg_write(struct file *file, const char __user *buf, size_t count, loff_t *ppos)
237{
238 ssize_t (*write)(struct file *, const char __user *, size_t, loff_t *);
239 struct proc_dir_entry *pde = PDE(file_inode(file));
240 ssize_t rv = -EIO;
241 if (use_pde(pde)) {
242 write = pde->proc_fops->write;
243 if (write)
244 rv = write(file, buf, count, ppos);
245 unuse_pde(pde);
246 }
247 return rv;
248}
249
250static __poll_t proc_reg_poll(struct file *file, struct poll_table_struct *pts)
251{
252 struct proc_dir_entry *pde = PDE(file_inode(file));
253 __poll_t rv = DEFAULT_POLLMASK;
254 __poll_t (*poll)(struct file *, struct poll_table_struct *);
255 if (use_pde(pde)) {
256 poll = pde->proc_fops->poll;
257 if (poll)
258 rv = poll(file, pts);
259 unuse_pde(pde);
260 }
261 return rv;
262}
263
264static long proc_reg_unlocked_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 (*ioctl)(struct file *, unsigned int, unsigned long);
269 if (use_pde(pde)) {
270 ioctl = pde->proc_fops->unlocked_ioctl;
271 if (ioctl)
272 rv = ioctl(file, cmd, arg);
273 unuse_pde(pde);
274 }
275 return rv;
276}
277
278#ifdef CONFIG_COMPAT
279static long proc_reg_compat_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
280{
281 struct proc_dir_entry *pde = PDE(file_inode(file));
282 long rv = -ENOTTY;
283 long (*compat_ioctl)(struct file *, unsigned int, unsigned long);
284 if (use_pde(pde)) {
285 compat_ioctl = pde->proc_fops->compat_ioctl;
286 if (compat_ioctl)
287 rv = compat_ioctl(file, cmd, arg);
288 unuse_pde(pde);
289 }
290 return rv;
291}
292#endif
293
294static int proc_reg_mmap(struct file *file, struct vm_area_struct *vma)
295{
296 struct proc_dir_entry *pde = PDE(file_inode(file));
297 int rv = -EIO;
298 int (*mmap)(struct file *, struct vm_area_struct *);
299 if (use_pde(pde)) {
300 mmap = pde->proc_fops->mmap;
301 if (mmap)
302 rv = mmap(file, vma);
303 unuse_pde(pde);
304 }
305 return rv;
306}
307
308static unsigned long
309proc_reg_get_unmapped_area(struct file *file, unsigned long orig_addr,
310 unsigned long len, unsigned long pgoff,
311 unsigned long flags)
312{
313 struct proc_dir_entry *pde = PDE(file_inode(file));
314 unsigned long rv = -EIO;
315
316 if (use_pde(pde)) {
317 typeof(proc_reg_get_unmapped_area) *get_area;
318
319 get_area = pde->proc_fops->get_unmapped_area;
320#ifdef CONFIG_MMU
321 if (!get_area)
322 get_area = current->mm->get_unmapped_area;
323#endif
324
325 if (get_area)
326 rv = get_area(file, orig_addr, len, pgoff, flags);
327 else
328 rv = orig_addr;
329 unuse_pde(pde);
330 }
331 return rv;
332}
333
334static int proc_reg_open(struct inode *inode, struct file *file)
335{
336 struct proc_dir_entry *pde = PDE(inode);
337 int rv = 0;
338 int (*open)(struct inode *, struct file *);
339 int (*release)(struct inode *, struct file *);
340 struct pde_opener *pdeo;
341
342 /*
343 * Ensure that
344 * 1) PDE's ->release hook will be called no matter what
345 * either normally by close()/->release, or forcefully by
346 * rmmod/remove_proc_entry.
347 *
348 * 2) rmmod isn't blocked by opening file in /proc and sitting on
349 * the descriptor (including "rmmod foo </proc/foo" scenario).
350 *
351 * Save every "struct file" with custom ->release hook.
352 */
353 if (!use_pde(pde))
354 return -ENOENT;
355
356 release = pde->proc_fops->release;
357 if (release) {
358 pdeo = kmem_cache_alloc(pde_opener_cache, GFP_KERNEL);
359 if (!pdeo) {
360 rv = -ENOMEM;
361 goto out_unuse;
362 }
363 }
364
365 open = pde->proc_fops->open;
366 if (open)
367 rv = open(inode, file);
368
369 if (release) {
370 if (rv == 0) {
371 /* To know what to release. */
372 pdeo->file = file;
373 pdeo->closing = false;
374 pdeo->c = NULL;
375 spin_lock(&pde->pde_unload_lock);
376 list_add(&pdeo->lh, &pde->pde_openers);
377 spin_unlock(&pde->pde_unload_lock);
378 } else
379 kmem_cache_free(pde_opener_cache, pdeo);
380 }
381
382out_unuse:
383 unuse_pde(pde);
384 return rv;
385}
386
387static int proc_reg_release(struct inode *inode, struct file *file)
388{
389 struct proc_dir_entry *pde = PDE(inode);
390 struct pde_opener *pdeo;
391 spin_lock(&pde->pde_unload_lock);
392 list_for_each_entry(pdeo, &pde->pde_openers, lh) {
393 if (pdeo->file == file) {
394 close_pdeo(pde, pdeo);
395 return 0;
396 }
397 }
398 spin_unlock(&pde->pde_unload_lock);
399 return 0;
400}
401
402static const struct file_operations proc_reg_file_ops = {
403 .llseek = proc_reg_llseek,
404 .read = proc_reg_read,
405 .write = proc_reg_write,
406 .poll = proc_reg_poll,
407 .unlocked_ioctl = proc_reg_unlocked_ioctl,
408#ifdef CONFIG_COMPAT
409 .compat_ioctl = proc_reg_compat_ioctl,
410#endif
411 .mmap = proc_reg_mmap,
412 .get_unmapped_area = proc_reg_get_unmapped_area,
413 .open = proc_reg_open,
414 .release = proc_reg_release,
415};
416
417#ifdef CONFIG_COMPAT
418static const struct file_operations proc_reg_file_ops_no_compat = {
419 .llseek = proc_reg_llseek,
420 .read = proc_reg_read,
421 .write = proc_reg_write,
422 .poll = proc_reg_poll,
423 .unlocked_ioctl = proc_reg_unlocked_ioctl,
424 .mmap = proc_reg_mmap,
425 .get_unmapped_area = proc_reg_get_unmapped_area,
426 .open = proc_reg_open,
427 .release = proc_reg_release,
428};
429#endif
430
431static void proc_put_link(void *p)
432{
433 unuse_pde(p);
434}
435
436static const char *proc_get_link(struct dentry *dentry,
437 struct inode *inode,
438 struct delayed_call *done)
439{
440 struct proc_dir_entry *pde = PDE(inode);
441 if (!use_pde(pde))
442 return ERR_PTR(-EINVAL);
443 set_delayed_call(done, proc_put_link, pde);
444 return pde->data;
445}
446
447const struct inode_operations proc_link_inode_operations = {
448 .get_link = proc_get_link,
449};
450
451struct inode *proc_get_inode(struct super_block *sb, struct proc_dir_entry *de)
452{
453 struct inode *inode = new_inode_pseudo(sb);
454
455 if (inode) {
456 inode->i_ino = de->low_ino;
457 inode->i_mtime = inode->i_atime = inode->i_ctime = current_time(inode);
458 PROC_I(inode)->pde = de;
459
460 if (is_empty_pde(de)) {
461 make_empty_dir_inode(inode);
462 return inode;
463 }
464 if (de->mode) {
465 inode->i_mode = de->mode;
466 inode->i_uid = de->uid;
467 inode->i_gid = de->gid;
468 }
469 if (de->size)
470 inode->i_size = de->size;
471 if (de->nlink)
472 set_nlink(inode, de->nlink);
473 WARN_ON(!de->proc_iops);
474 inode->i_op = de->proc_iops;
475 if (de->proc_fops) {
476 if (S_ISREG(inode->i_mode)) {
477#ifdef CONFIG_COMPAT
478 if (!de->proc_fops->compat_ioctl)
479 inode->i_fop =
480 &proc_reg_file_ops_no_compat;
481 else
482#endif
483 inode->i_fop = &proc_reg_file_ops;
484 } else {
485 inode->i_fop = de->proc_fops;
486 }
487 }
488 } else
489 pde_put(de);
490 return inode;
491}
492
493int proc_fill_super(struct super_block *s, void *data, int silent)
494{
495 struct pid_namespace *ns = get_pid_ns(s->s_fs_info);
496 struct inode *root_inode;
497 int ret;
498
499 if (!proc_parse_options(data, ns))
500 return -EINVAL;
501
502 /* User space would break if executables or devices appear on proc */
503 s->s_iflags |= SB_I_USERNS_VISIBLE | SB_I_NOEXEC | SB_I_NODEV;
504 s->s_flags |= SB_NODIRATIME | SB_NOSUID | SB_NOEXEC;
505 s->s_blocksize = 1024;
506 s->s_blocksize_bits = 10;
507 s->s_magic = PROC_SUPER_MAGIC;
508 s->s_op = &proc_sops;
509 s->s_time_gran = 1;
510
511 /*
512 * procfs isn't actually a stacking filesystem; however, there is
513 * too much magic going on inside it to permit stacking things on
514 * top of it
515 */
516 s->s_stack_depth = FILESYSTEM_MAX_STACK_DEPTH;
517
518 pde_get(&proc_root);
519 root_inode = proc_get_inode(s, &proc_root);
520 if (!root_inode) {
521 pr_err("proc_fill_super: get root inode failed\n");
522 return -ENOMEM;
523 }
524
525 s->s_root = d_make_root(root_inode);
526 if (!s->s_root) {
527 pr_err("proc_fill_super: allocate dentry failed\n");
528 return -ENOMEM;
529 }
530
531 ret = proc_setup_self(s);
532 if (ret) {
533 return ret;
534 }
535 return proc_setup_thread_self(s);
536}