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

source code of linux/fs/proc/inode.c