]> git.ipfire.org Git - people/ms/linux.git/blob - fs/proc/inode.c
Merge tag 'xfs-5.20-merge-8' of git://git.kernel.org/pub/scm/fs/xfs/xfs-linux
[people/ms/linux.git] / fs / proc / inode.c
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
31 static 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
60 static struct kmem_cache *proc_inode_cachep __ro_after_init;
61 static struct kmem_cache *pde_opener_cache __ro_after_init;
62
63 static 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
81 static void proc_free_inode(struct inode *inode)
82 {
83 kmem_cache_free(proc_inode_cachep, PROC_I(inode));
84 }
85
86 static 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
93 void __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
111 void 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
167 static 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
179 static 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
193 const 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
202 enum {BIAS = -1U<<31};
203
204 static inline int use_pde(struct proc_dir_entry *pde)
205 {
206 return likely(atomic_inc_unless_negative(&pde->in_use));
207 }
208
209 static 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 */
224 static 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
261 void 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
281 static 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
295 static 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
310 static 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
320 static 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
334 static 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
344 static 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
358 static __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
368 static __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
382 static 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
392 static 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
407 static 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
417 static 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
431 static 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
441 static 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
455 static unsigned long
456 pde_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
472 static unsigned long
473 proc_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
489 static 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_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(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(&pde->pde_unload_lock);
538 list_add(&pdeo->lh, &pde->pde_openers);
539 spin_unlock(&pde->pde_unload_lock);
540 } else
541 kmem_cache_free(pde_opener_cache, pdeo);
542 }
543
544 out_unuse:
545 unuse_pde(pde);
546 return rv;
547 }
548
549 static 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(&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(&pde->pde_unload_lock);
572 return 0;
573 }
574
575 static 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
587 static 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 = generic_file_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
601 static 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
614 static const struct file_operations proc_iter_file_ops_compat = {
615 .llseek = proc_reg_llseek,
616 .read_iter = proc_reg_read_iter,
617 .splice_read = generic_file_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
629 static void proc_put_link(void *p)
630 {
631 unuse_pde(p);
632 }
633
634 static 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(-EINVAL);
641 set_delayed_call(done, proc_put_link, pde);
642 return pde->data;
643 }
644
645 const struct inode_operations proc_link_inode_operations = {
646 .get_link = proc_get_link,
647 };
648
649 struct 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 inode->i_mtime = inode->i_atime = inode->i_ctime = current_time(inode);
661 PROC_I(inode)->pde = de;
662 if (is_empty_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, 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 }