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
28#include <linux/uaccess.h>
29
30#include "internal.h"
31
32static 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
55static struct kmem_cache *proc_inode_cachep __ro_after_init;
56static struct kmem_cache *pde_opener_cache __ro_after_init;
57
58static 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
75static void proc_free_inode(struct inode *inode)
76{
77 kmem_cache_free(proc_inode_cachep, PROC_I(inode));
78}
79
80static 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
87void __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
105static 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
118const 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
127enum {BIAS = -1U<<31};
128
129static inline int use_pde(struct proc_dir_entry *pde)
130{
131 return likely(atomic_inc_unless_negative(&pde->in_use));
132}
133
134static 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 */
141static 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_fops->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
178void 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
198static 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 file_operations, llseek) llseek;
204
205 llseek = pde->proc_fops->llseek;
206 if (!llseek)
207 llseek = default_llseek;
208 rv = llseek(file, offset, whence);
209 unuse_pde(pde);
210 }
211 return rv;
212}
213
214static 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 file_operations, read) read;
220
221 read = pde->proc_fops->read;
222 if (read)
223 rv = read(file, buf, count, ppos);
224 unuse_pde(pde);
225 }
226 return rv;
227}
228
229static 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 file_operations, write) write;
235
236 write = pde->proc_fops->write;
237 if (write)
238 rv = write(file, buf, count, ppos);
239 unuse_pde(pde);
240 }
241 return rv;
242}
243
244static __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 file_operations, poll) poll;
250
251 poll = pde->proc_fops->poll;
252 if (poll)
253 rv = poll(file, pts);
254 unuse_pde(pde);
255 }
256 return rv;
257}
258
259static 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 file_operations, unlocked_ioctl) ioctl;
265
266 ioctl = pde->proc_fops->unlocked_ioctl;
267 if (ioctl)
268 rv = ioctl(file, cmd, arg);
269 unuse_pde(pde);
270 }
271 return rv;
272}
273
274#ifdef CONFIG_COMPAT
275static 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 file_operations, compat_ioctl) compat_ioctl;
281
282 compat_ioctl = pde->proc_fops->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
291static 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 file_operations, mmap) mmap;
297
298 mmap = pde->proc_fops->mmap;
299 if (mmap)
300 rv = mmap(file, vma);
301 unuse_pde(pde);
302 }
303 return rv;
304}
305
306static unsigned long
307proc_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 file_operations, get_unmapped_area) get_area;
316
317 get_area = pde->proc_fops->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
332static 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 file_operations, open) open;
337 typeof_member(struct file_operations, 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_fops->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_fops->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
380out_unuse:
381 unuse_pde(pde);
382 return rv;
383}
384
385static 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
400static 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
416static 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
429static void proc_put_link(void *p)
430{
431 unuse_pde(p);
432}
433
434static 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
445const struct inode_operations proc_link_inode_operations = {
446 .get_link = proc_get_link,
447};
448
449struct 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 WARN_ON(!de->proc_iops);
472 inode->i_op = de->proc_iops;
473 if (de->proc_fops) {
474 if (S_ISREG(inode->i_mode)) {
475#ifdef CONFIG_COMPAT
476 if (!de->proc_fops->compat_ioctl)
477 inode->i_fop =
478 &proc_reg_file_ops_no_compat;
479 else
480#endif
481 inode->i_fop = &proc_reg_file_ops;
482 } else {
483 inode->i_fop = de->proc_fops;
484 }
485 }
486 } else
487 pde_put(de);
488 return inode;
489}