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

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