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 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, 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(&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(ei->pid);
77 /* Let go of any associated proc directory entry */
78 if (ei->pde)
79 pde_put(ei->pde);
80 kmem_cache_free(proc_inode_cachep, 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("proc_inode_cache",
93 sizeof(struct proc_inode),
94 0, (SLAB_RECLAIM_ACCOUNT|
95 SLAB_ACCOUNT|
96 SLAB_PANIC),
97 init_once);
98 pde_opener_cache =
99 kmem_cache_create("pde_opener", sizeof(struct pde_opener), 0,
100 SLAB_ACCOUNT|SLAB_PANIC, NULL);
101 proc_dir_entry_cache = kmem_cache_create_usercopy(
102 "proc_dir_entry", SIZEOF_PDE, 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(&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(&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(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(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(root->d_sb);
176
177 if (!gid_eq(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(&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(&pde->pde_unload_lock);
240
241 file = pdeo->file;
242 pde->proc_ops->proc_release(file_inode(file), file);
243
244 spin_lock(&pde->pde_unload_lock);
245 /* Strictly after ->proc_release, see above. */
246 list_del(&pdeo->lh);
247 c = pdeo->c;
248 spin_unlock(&pde->pde_unload_lock);
249 if (unlikely(c))
250 complete(c);
251 kmem_cache_free(pde_opener_cache, 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(BIAS, &de->in_use) != BIAS)
261 wait_for_completion(&c);
262
263 /* ->pde_openers list can't grow from now on. */
264
265 spin_lock(&de->pde_unload_lock);
266 while (!list_empty(&de->pde_openers)) {
267 struct pde_opener *pdeo;
268 pdeo = list_first_entry(&de->pde_openers, struct pde_opener, lh);
269 close_pdeo(de, pdeo);
270 spin_lock(&de->pde_unload_lock);
271 }
272 spin_unlock(&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(file_inode(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(file_inode(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 __auto_type read = pde->proc_ops->proc_read;
307 if (read)
308 return read(file, buf, count, ppos);
309 return -EIO;
310}
311
312static ssize_t proc_reg_read(struct file *file, char __user *buf, size_t count, loff_t *ppos)
313{
314 struct proc_dir_entry *pde = PDE(file_inode(file));
315 ssize_t rv = -EIO;
316
317 if (pde_is_permanent(pde)) {
318 return pde_read(pde, file, buf, count, ppos);
319 } else if (use_pde(pde)) {
320 rv = pde_read(pde, file, buf, count, ppos);
321 unuse_pde(pde);
322 }
323 return rv;
324}
325
326static ssize_t pde_write(struct proc_dir_entry *pde, struct file *file, const char __user *buf, size_t count, loff_t *ppos)
327{
328 __auto_type write = pde->proc_ops->proc_write;
329 if (write)
330 return write(file, buf, count, ppos);
331 return -EIO;
332}
333
334static ssize_t proc_reg_write(struct file *file, const char __user *buf, size_t count, loff_t *ppos)
335{
336 struct proc_dir_entry *pde = PDE(file_inode(file));
337 ssize_t rv = -EIO;
338
339 if (pde_is_permanent(pde)) {
340 return pde_write(pde, file, buf, count, ppos);
341 } else if (use_pde(pde)) {
342 rv = pde_write(pde, file, buf, count, ppos);
343 unuse_pde(pde);
344 }
345 return rv;
346}
347
348static __poll_t pde_poll(struct proc_dir_entry *pde, struct file *file, struct poll_table_struct *pts)
349{
350 __auto_type poll = pde->proc_ops->proc_poll;
351 if (poll)
352 return poll(file, pts);
353 return DEFAULT_POLLMASK;
354}
355
356static __poll_t proc_reg_poll(struct file *file, struct poll_table_struct *pts)
357{
358 struct proc_dir_entry *pde = PDE(file_inode(file));
359 __poll_t rv = DEFAULT_POLLMASK;
360
361 if (pde_is_permanent(pde)) {
362 return pde_poll(pde, file, pts);
363 } else if (use_pde(pde)) {
364 rv = pde_poll(pde, file, pts);
365 unuse_pde(pde);
366 }
367 return rv;
368}
369
370static long pde_ioctl(struct proc_dir_entry *pde, struct file *file, unsigned int cmd, unsigned long arg)
371{
372 __auto_type ioctl = pde->proc_ops->proc_ioctl;
373 if (ioctl)
374 return ioctl(file, cmd, arg);
375 return -ENOTTY;
376}
377
378static long proc_reg_unlocked_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
379{
380 struct proc_dir_entry *pde = PDE(file_inode(file));
381 long rv = -ENOTTY;
382
383 if (pde_is_permanent(pde)) {
384 return pde_ioctl(pde, file, cmd, arg);
385 } else if (use_pde(pde)) {
386 rv = pde_ioctl(pde, file, cmd, arg);
387 unuse_pde(pde);
388 }
389 return rv;
390}
391
392#ifdef CONFIG_COMPAT
393static long pde_compat_ioctl(struct proc_dir_entry *pde, struct file *file, unsigned int cmd, unsigned long arg)
394{
395 __auto_type compat_ioctl = pde->proc_ops->proc_compat_ioctl;
396 if (compat_ioctl)
397 return compat_ioctl(file, cmd, arg);
398 return -ENOTTY;
399}
400
401static long proc_reg_compat_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
402{
403 struct proc_dir_entry *pde = PDE(file_inode(file));
404 long rv = -ENOTTY;
405 if (pde_is_permanent(pde)) {
406 return pde_compat_ioctl(pde, file, cmd, arg);
407 } else if (use_pde(pde)) {
408 rv = pde_compat_ioctl(pde, file, cmd, arg);
409 unuse_pde(pde);
410 }
411 return rv;
412}
413#endif
414
415static int pde_mmap(struct proc_dir_entry *pde, struct file *file, struct vm_area_struct *vma)
416{
417 __auto_type mmap = pde->proc_ops->proc_mmap;
418 if (mmap)
419 return mmap(file, vma);
420 return -EIO;
421}
422
423static int proc_reg_mmap(struct file *file, struct vm_area_struct *vma)
424{
425 struct proc_dir_entry *pde = PDE(file_inode(file));
426 int rv = -EIO;
427
428 if (pde_is_permanent(pde)) {
429 return pde_mmap(pde, file, vma);
430 } else if (use_pde(pde)) {
431 rv = pde_mmap(pde, file, vma);
432 unuse_pde(pde);
433 }
434 return rv;
435}
436
437static unsigned long
438pde_get_unmapped_area(struct proc_dir_entry *pde, struct file *file, unsigned long orig_addr,
439 unsigned long len, unsigned long pgoff,
440 unsigned long flags)
441{
442 if (pde->proc_ops->proc_get_unmapped_area)
443 return pde->proc_ops->proc_get_unmapped_area(file, orig_addr, len, pgoff, flags);
444
445#ifdef CONFIG_MMU
446 return mm_get_unmapped_area(current->mm, file, orig_addr, len, pgoff, flags);
447#endif
448
449 return orig_addr;
450}
451
452static unsigned long
453proc_reg_get_unmapped_area(struct file *file, unsigned long orig_addr,
454 unsigned long len, unsigned long pgoff,
455 unsigned long flags)
456{
457 struct proc_dir_entry *pde = PDE(file_inode(file));
458 unsigned long rv = -EIO;
459
460 if (pde_is_permanent(pde)) {
461 return pde_get_unmapped_area(pde, file, orig_addr, len, pgoff, flags);
462 } else if (use_pde(pde)) {
463 rv = pde_get_unmapped_area(pde, file, orig_addr, len, pgoff, flags);
464 unuse_pde(pde);
465 }
466 return rv;
467}
468
469static int proc_reg_open(struct inode *inode, struct file *file)
470{
471 struct proc_dir_entry *pde = PDE(inode);
472 int rv = 0;
473 typeof_member(struct proc_ops, proc_open) open;
474 struct pde_opener *pdeo;
475
476 if (!pde->proc_ops->proc_lseek)
477 file->f_mode &= ~FMODE_LSEEK;
478
479 if (pde_is_permanent(pde)) {
480 open = pde->proc_ops->proc_open;
481 if (open)
482 rv = open(inode, file);
483 return rv;
484 }
485
486 /*
487 * Ensure that
488 * 1) PDE's ->release hook will be called no matter what
489 * either normally by close()/->release, or forcefully by
490 * rmmod/remove_proc_entry.
491 *
492 * 2) rmmod isn't blocked by opening file in /proc and sitting on
493 * the descriptor (including "rmmod foo </proc/foo" scenario).
494 *
495 * Save every "struct file" with custom ->release hook.
496 */
497 if (!use_pde(pde))
498 return -ENOENT;
499
500 __auto_type release = pde->proc_ops->proc_release;
501 if (release) {
502 pdeo = kmem_cache_alloc(pde_opener_cache, GFP_KERNEL);
503 if (!pdeo) {
504 rv = -ENOMEM;
505 goto out_unuse;
506 }
507 }
508
509 open = pde->proc_ops->proc_open;
510 if (open)
511 rv = open(inode, file);
512
513 if (release) {
514 if (rv == 0) {
515 /* To know what to release. */
516 pdeo->file = file;
517 pdeo->closing = false;
518 pdeo->c = NULL;
519 spin_lock(&pde->pde_unload_lock);
520 list_add(&pdeo->lh, &pde->pde_openers);
521 spin_unlock(&pde->pde_unload_lock);
522 } else
523 kmem_cache_free(pde_opener_cache, pdeo);
524 }
525
526out_unuse:
527 unuse_pde(pde);
528 return rv;
529}
530
531static int proc_reg_release(struct inode *inode, struct file *file)
532{
533 struct proc_dir_entry *pde = PDE(inode);
534 struct pde_opener *pdeo;
535
536 if (pde_is_permanent(pde)) {
537 __auto_type release = pde->proc_ops->proc_release;
538 if (release) {
539 return release(inode, file);
540 }
541 return 0;
542 }
543
544 spin_lock(&pde->pde_unload_lock);
545 list_for_each_entry(pdeo, &pde->pde_openers, lh) {
546 if (pdeo->file == file) {
547 close_pdeo(pde, pdeo);
548 return 0;
549 }
550 }
551 spin_unlock(&pde->pde_unload_lock);
552 return 0;
553}
554
555static const struct file_operations proc_reg_file_ops = {
556 .llseek = proc_reg_llseek,
557 .read = proc_reg_read,
558 .write = proc_reg_write,
559 .poll = proc_reg_poll,
560 .unlocked_ioctl = proc_reg_unlocked_ioctl,
561 .mmap = proc_reg_mmap,
562 .get_unmapped_area = proc_reg_get_unmapped_area,
563 .open = proc_reg_open,
564 .release = proc_reg_release,
565};
566
567static const struct file_operations proc_iter_file_ops = {
568 .llseek = proc_reg_llseek,
569 .read_iter = proc_reg_read_iter,
570 .write = proc_reg_write,
571 .splice_read = copy_splice_read,
572 .poll = proc_reg_poll,
573 .unlocked_ioctl = proc_reg_unlocked_ioctl,
574 .mmap = proc_reg_mmap,
575 .get_unmapped_area = proc_reg_get_unmapped_area,
576 .open = proc_reg_open,
577 .release = proc_reg_release,
578};
579
580#ifdef CONFIG_COMPAT
581static const struct file_operations proc_reg_file_ops_compat = {
582 .llseek = proc_reg_llseek,
583 .read = proc_reg_read,
584 .write = proc_reg_write,
585 .poll = proc_reg_poll,
586 .unlocked_ioctl = proc_reg_unlocked_ioctl,
587 .compat_ioctl = proc_reg_compat_ioctl,
588 .mmap = proc_reg_mmap,
589 .get_unmapped_area = proc_reg_get_unmapped_area,
590 .open = proc_reg_open,
591 .release = proc_reg_release,
592};
593
594static const struct file_operations proc_iter_file_ops_compat = {
595 .llseek = proc_reg_llseek,
596 .read_iter = proc_reg_read_iter,
597 .splice_read = copy_splice_read,
598 .write = proc_reg_write,
599 .poll = proc_reg_poll,
600 .unlocked_ioctl = proc_reg_unlocked_ioctl,
601 .compat_ioctl = proc_reg_compat_ioctl,
602 .mmap = proc_reg_mmap,
603 .get_unmapped_area = proc_reg_get_unmapped_area,
604 .open = proc_reg_open,
605 .release = proc_reg_release,
606};
607#endif
608
609static void proc_put_link(void *p)
610{
611 unuse_pde(p);
612}
613
614static const char *proc_get_link(struct dentry *dentry,
615 struct inode *inode,
616 struct delayed_call *done)
617{
618 struct proc_dir_entry *pde = PDE(inode);
619 if (!use_pde(pde))
620 return ERR_PTR(-EINVAL);
621 set_delayed_call(done, proc_put_link, pde);
622 return pde->data;
623}
624
625const struct inode_operations proc_link_inode_operations = {
626 .get_link = proc_get_link,
627};
628
629struct inode *proc_get_inode(struct super_block *sb, struct proc_dir_entry *de)
630{
631 struct inode *inode = new_inode(sb);
632
633 if (!inode) {
634 pde_put(de);
635 return NULL;
636 }
637
638 inode->i_private = de->data;
639 inode->i_ino = de->low_ino;
640 simple_inode_init_ts(inode);
641 PROC_I(inode)->pde = de;
642 if (is_empty_pde(de)) {
643 make_empty_dir_inode(inode);
644 return inode;
645 }
646
647 if (de->mode) {
648 inode->i_mode = de->mode;
649 inode->i_uid = de->uid;
650 inode->i_gid = de->gid;
651 }
652 if (de->size)
653 inode->i_size = de->size;
654 if (de->nlink)
655 set_nlink(inode, de->nlink);
656
657 if (S_ISREG(inode->i_mode)) {
658 inode->i_op = de->proc_iops;
659 if (de->proc_ops->proc_read_iter)
660 inode->i_fop = &proc_iter_file_ops;
661 else
662 inode->i_fop = &proc_reg_file_ops;
663#ifdef CONFIG_COMPAT
664 if (de->proc_ops->proc_compat_ioctl) {
665 if (de->proc_ops->proc_read_iter)
666 inode->i_fop = &proc_iter_file_ops_compat;
667 else
668 inode->i_fop = &proc_reg_file_ops_compat;
669 }
670#endif
671 } else if (S_ISDIR(inode->i_mode)) {
672 inode->i_op = de->proc_iops;
673 inode->i_fop = de->proc_dir_ops;
674 } else if (S_ISLNK(inode->i_mode)) {
675 inode->i_op = de->proc_iops;
676 inode->i_fop = NULL;
677 } else {
678 BUG();
679 }
680 return inode;
681}
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}