Loading...
1// SPDX-License-Identifier: GPL-2.0-only
2/*
3 * proc/fs/generic.c --- generic routines for the proc-fs
4 *
5 * This file contains generic proc-fs routines for handling
6 * directories and files.
7 *
8 * Copyright (C) 1991, 1992 Linus Torvalds.
9 * Copyright (C) 1997 Theodore Ts'o
10 */
11
12#include <linux/cache.h>
13#include <linux/errno.h>
14#include <linux/time.h>
15#include <linux/proc_fs.h>
16#include <linux/stat.h>
17#include <linux/mm.h>
18#include <linux/module.h>
19#include <linux/namei.h>
20#include <linux/slab.h>
21#include <linux/printk.h>
22#include <linux/mount.h>
23#include <linux/init.h>
24#include <linux/idr.h>
25#include <linux/bitops.h>
26#include <linux/spinlock.h>
27#include <linux/completion.h>
28#include <linux/uaccess.h>
29#include <linux/seq_file.h>
30
31#include "internal.h"
32
33static DEFINE_RWLOCK(proc_subdir_lock);
34
35struct kmem_cache *proc_dir_entry_cache __ro_after_init;
36
37void pde_free(struct proc_dir_entry *pde)
38{
39 if (S_ISLNK(pde->mode))
40 kfree(pde->data);
41 if (pde->name != pde->inline_name)
42 kfree(pde->name);
43 kmem_cache_free(proc_dir_entry_cache, pde);
44}
45
46static int proc_match(const char *name, struct proc_dir_entry *de, unsigned int len)
47{
48 if (len < de->namelen)
49 return -1;
50 if (len > de->namelen)
51 return 1;
52
53 return memcmp(name, de->name, len);
54}
55
56static struct proc_dir_entry *pde_subdir_first(struct proc_dir_entry *dir)
57{
58 return rb_entry_safe(rb_first(&dir->subdir), struct proc_dir_entry,
59 subdir_node);
60}
61
62static struct proc_dir_entry *pde_subdir_next(struct proc_dir_entry *dir)
63{
64 return rb_entry_safe(rb_next(&dir->subdir_node), struct proc_dir_entry,
65 subdir_node);
66}
67
68static struct proc_dir_entry *pde_subdir_find(struct proc_dir_entry *dir,
69 const char *name,
70 unsigned int len)
71{
72 struct rb_node *node = dir->subdir.rb_node;
73
74 while (node) {
75 struct proc_dir_entry *de = rb_entry(node,
76 struct proc_dir_entry,
77 subdir_node);
78 int result = proc_match(name, de, len);
79
80 if (result < 0)
81 node = node->rb_left;
82 else if (result > 0)
83 node = node->rb_right;
84 else
85 return de;
86 }
87 return NULL;
88}
89
90static bool pde_subdir_insert(struct proc_dir_entry *dir,
91 struct proc_dir_entry *de)
92{
93 struct rb_root *root = &dir->subdir;
94 struct rb_node **new = &root->rb_node, *parent = NULL;
95
96 /* Figure out where to put new node */
97 while (*new) {
98 struct proc_dir_entry *this = rb_entry(*new,
99 struct proc_dir_entry,
100 subdir_node);
101 int result = proc_match(de->name, this, de->namelen);
102
103 parent = *new;
104 if (result < 0)
105 new = &(*new)->rb_left;
106 else if (result > 0)
107 new = &(*new)->rb_right;
108 else
109 return false;
110 }
111
112 /* Add new node and rebalance tree. */
113 rb_link_node(&de->subdir_node, parent, new);
114 rb_insert_color(&de->subdir_node, root);
115 return true;
116}
117
118static int proc_notify_change(struct dentry *dentry, struct iattr *iattr)
119{
120 struct inode *inode = d_inode(dentry);
121 struct proc_dir_entry *de = PDE(inode);
122 int error;
123
124 error = setattr_prepare(dentry, iattr);
125 if (error)
126 return error;
127
128 setattr_copy(inode, iattr);
129 mark_inode_dirty(inode);
130
131 proc_set_user(de, inode->i_uid, inode->i_gid);
132 de->mode = inode->i_mode;
133 return 0;
134}
135
136static int proc_getattr(const struct path *path, struct kstat *stat,
137 u32 request_mask, unsigned int query_flags)
138{
139 struct inode *inode = d_inode(path->dentry);
140 struct proc_dir_entry *de = PDE(inode);
141 if (de) {
142 nlink_t nlink = READ_ONCE(de->nlink);
143 if (nlink > 0) {
144 set_nlink(inode, nlink);
145 }
146 }
147
148 generic_fillattr(inode, stat);
149 return 0;
150}
151
152static const struct inode_operations proc_file_inode_operations = {
153 .setattr = proc_notify_change,
154};
155
156/*
157 * This function parses a name such as "tty/driver/serial", and
158 * returns the struct proc_dir_entry for "/proc/tty/driver", and
159 * returns "serial" in residual.
160 */
161static int __xlate_proc_name(const char *name, struct proc_dir_entry **ret,
162 const char **residual)
163{
164 const char *cp = name, *next;
165 struct proc_dir_entry *de;
166
167 de = *ret;
168 if (!de)
169 de = &proc_root;
170
171 while (1) {
172 next = strchr(cp, '/');
173 if (!next)
174 break;
175
176 de = pde_subdir_find(de, cp, next - cp);
177 if (!de) {
178 WARN(1, "name '%s'\n", name);
179 return -ENOENT;
180 }
181 cp = next + 1;
182 }
183 *residual = cp;
184 *ret = de;
185 return 0;
186}
187
188static int xlate_proc_name(const char *name, struct proc_dir_entry **ret,
189 const char **residual)
190{
191 int rv;
192
193 read_lock(&proc_subdir_lock);
194 rv = __xlate_proc_name(name, ret, residual);
195 read_unlock(&proc_subdir_lock);
196 return rv;
197}
198
199static DEFINE_IDA(proc_inum_ida);
200
201#define PROC_DYNAMIC_FIRST 0xF0000000U
202
203/*
204 * Return an inode number between PROC_DYNAMIC_FIRST and
205 * 0xffffffff, or zero on failure.
206 */
207int proc_alloc_inum(unsigned int *inum)
208{
209 int i;
210
211 i = ida_simple_get(&proc_inum_ida, 0, UINT_MAX - PROC_DYNAMIC_FIRST + 1,
212 GFP_KERNEL);
213 if (i < 0)
214 return i;
215
216 *inum = PROC_DYNAMIC_FIRST + (unsigned int)i;
217 return 0;
218}
219
220void proc_free_inum(unsigned int inum)
221{
222 ida_simple_remove(&proc_inum_ida, inum - PROC_DYNAMIC_FIRST);
223}
224
225static int proc_misc_d_revalidate(struct dentry *dentry, unsigned int flags)
226{
227 if (flags & LOOKUP_RCU)
228 return -ECHILD;
229
230 if (atomic_read(&PDE(d_inode(dentry))->in_use) < 0)
231 return 0; /* revalidate */
232 return 1;
233}
234
235static int proc_misc_d_delete(const struct dentry *dentry)
236{
237 return atomic_read(&PDE(d_inode(dentry))->in_use) < 0;
238}
239
240static const struct dentry_operations proc_misc_dentry_ops = {
241 .d_revalidate = proc_misc_d_revalidate,
242 .d_delete = proc_misc_d_delete,
243};
244
245/*
246 * Don't create negative dentries here, return -ENOENT by hand
247 * instead.
248 */
249struct dentry *proc_lookup_de(struct inode *dir, struct dentry *dentry,
250 struct proc_dir_entry *de)
251{
252 struct inode *inode;
253
254 read_lock(&proc_subdir_lock);
255 de = pde_subdir_find(de, dentry->d_name.name, dentry->d_name.len);
256 if (de) {
257 pde_get(de);
258 read_unlock(&proc_subdir_lock);
259 inode = proc_get_inode(dir->i_sb, de);
260 if (!inode)
261 return ERR_PTR(-ENOMEM);
262 d_set_d_op(dentry, de->proc_dops);
263 return d_splice_alias(inode, dentry);
264 }
265 read_unlock(&proc_subdir_lock);
266 return ERR_PTR(-ENOENT);
267}
268
269struct dentry *proc_lookup(struct inode *dir, struct dentry *dentry,
270 unsigned int flags)
271{
272 struct proc_fs_info *fs_info = proc_sb_info(dir->i_sb);
273
274 if (fs_info->pidonly == PROC_PIDONLY_ON)
275 return ERR_PTR(-ENOENT);
276
277 return proc_lookup_de(dir, dentry, PDE(dir));
278}
279
280/*
281 * This returns non-zero if at EOF, so that the /proc
282 * root directory can use this and check if it should
283 * continue with the <pid> entries..
284 *
285 * Note that the VFS-layer doesn't care about the return
286 * value of the readdir() call, as long as it's non-negative
287 * for success..
288 */
289int proc_readdir_de(struct file *file, struct dir_context *ctx,
290 struct proc_dir_entry *de)
291{
292 int i;
293
294 if (!dir_emit_dots(file, ctx))
295 return 0;
296
297 i = ctx->pos - 2;
298 read_lock(&proc_subdir_lock);
299 de = pde_subdir_first(de);
300 for (;;) {
301 if (!de) {
302 read_unlock(&proc_subdir_lock);
303 return 0;
304 }
305 if (!i)
306 break;
307 de = pde_subdir_next(de);
308 i--;
309 }
310
311 do {
312 struct proc_dir_entry *next;
313 pde_get(de);
314 read_unlock(&proc_subdir_lock);
315 if (!dir_emit(ctx, de->name, de->namelen,
316 de->low_ino, de->mode >> 12)) {
317 pde_put(de);
318 return 0;
319 }
320 ctx->pos++;
321 read_lock(&proc_subdir_lock);
322 next = pde_subdir_next(de);
323 pde_put(de);
324 de = next;
325 } while (de);
326 read_unlock(&proc_subdir_lock);
327 return 1;
328}
329
330int proc_readdir(struct file *file, struct dir_context *ctx)
331{
332 struct inode *inode = file_inode(file);
333 struct proc_fs_info *fs_info = proc_sb_info(inode->i_sb);
334
335 if (fs_info->pidonly == PROC_PIDONLY_ON)
336 return 1;
337
338 return proc_readdir_de(file, ctx, PDE(inode));
339}
340
341/*
342 * These are the generic /proc directory operations. They
343 * use the in-memory "struct proc_dir_entry" tree to parse
344 * the /proc directory.
345 */
346static const struct file_operations proc_dir_operations = {
347 .llseek = generic_file_llseek,
348 .read = generic_read_dir,
349 .iterate_shared = proc_readdir,
350};
351
352/*
353 * proc directories can do almost nothing..
354 */
355static const struct inode_operations proc_dir_inode_operations = {
356 .lookup = proc_lookup,
357 .getattr = proc_getattr,
358 .setattr = proc_notify_change,
359};
360
361/* returns the registered entry, or frees dp and returns NULL on failure */
362struct proc_dir_entry *proc_register(struct proc_dir_entry *dir,
363 struct proc_dir_entry *dp)
364{
365 if (proc_alloc_inum(&dp->low_ino))
366 goto out_free_entry;
367
368 write_lock(&proc_subdir_lock);
369 dp->parent = dir;
370 if (pde_subdir_insert(dir, dp) == false) {
371 WARN(1, "proc_dir_entry '%s/%s' already registered\n",
372 dir->name, dp->name);
373 write_unlock(&proc_subdir_lock);
374 goto out_free_inum;
375 }
376 dir->nlink++;
377 write_unlock(&proc_subdir_lock);
378
379 return dp;
380out_free_inum:
381 proc_free_inum(dp->low_ino);
382out_free_entry:
383 pde_free(dp);
384 return NULL;
385}
386
387static struct proc_dir_entry *__proc_create(struct proc_dir_entry **parent,
388 const char *name,
389 umode_t mode,
390 nlink_t nlink)
391{
392 struct proc_dir_entry *ent = NULL;
393 const char *fn;
394 struct qstr qstr;
395
396 if (xlate_proc_name(name, parent, &fn) != 0)
397 goto out;
398 qstr.name = fn;
399 qstr.len = strlen(fn);
400 if (qstr.len == 0 || qstr.len >= 256) {
401 WARN(1, "name len %u\n", qstr.len);
402 return NULL;
403 }
404 if (qstr.len == 1 && fn[0] == '.') {
405 WARN(1, "name '.'\n");
406 return NULL;
407 }
408 if (qstr.len == 2 && fn[0] == '.' && fn[1] == '.') {
409 WARN(1, "name '..'\n");
410 return NULL;
411 }
412 if (*parent == &proc_root && name_to_int(&qstr) != ~0U) {
413 WARN(1, "create '/proc/%s' by hand\n", qstr.name);
414 return NULL;
415 }
416 if (is_empty_pde(*parent)) {
417 WARN(1, "attempt to add to permanently empty directory");
418 return NULL;
419 }
420
421 ent = kmem_cache_zalloc(proc_dir_entry_cache, GFP_KERNEL);
422 if (!ent)
423 goto out;
424
425 if (qstr.len + 1 <= SIZEOF_PDE_INLINE_NAME) {
426 ent->name = ent->inline_name;
427 } else {
428 ent->name = kmalloc(qstr.len + 1, GFP_KERNEL);
429 if (!ent->name) {
430 pde_free(ent);
431 return NULL;
432 }
433 }
434
435 memcpy(ent->name, fn, qstr.len + 1);
436 ent->namelen = qstr.len;
437 ent->mode = mode;
438 ent->nlink = nlink;
439 ent->subdir = RB_ROOT;
440 refcount_set(&ent->refcnt, 1);
441 spin_lock_init(&ent->pde_unload_lock);
442 INIT_LIST_HEAD(&ent->pde_openers);
443 proc_set_user(ent, (*parent)->uid, (*parent)->gid);
444
445 ent->proc_dops = &proc_misc_dentry_ops;
446
447out:
448 return ent;
449}
450
451struct proc_dir_entry *proc_symlink(const char *name,
452 struct proc_dir_entry *parent, const char *dest)
453{
454 struct proc_dir_entry *ent;
455
456 ent = __proc_create(&parent, name,
457 (S_IFLNK | S_IRUGO | S_IWUGO | S_IXUGO),1);
458
459 if (ent) {
460 ent->data = kmalloc((ent->size=strlen(dest))+1, GFP_KERNEL);
461 if (ent->data) {
462 strcpy((char*)ent->data,dest);
463 ent->proc_iops = &proc_link_inode_operations;
464 ent = proc_register(parent, ent);
465 } else {
466 pde_free(ent);
467 ent = NULL;
468 }
469 }
470 return ent;
471}
472EXPORT_SYMBOL(proc_symlink);
473
474struct proc_dir_entry *proc_mkdir_data(const char *name, umode_t mode,
475 struct proc_dir_entry *parent, void *data)
476{
477 struct proc_dir_entry *ent;
478
479 if (mode == 0)
480 mode = S_IRUGO | S_IXUGO;
481
482 ent = __proc_create(&parent, name, S_IFDIR | mode, 2);
483 if (ent) {
484 ent->data = data;
485 ent->proc_dir_ops = &proc_dir_operations;
486 ent->proc_iops = &proc_dir_inode_operations;
487 ent = proc_register(parent, ent);
488 }
489 return ent;
490}
491EXPORT_SYMBOL_GPL(proc_mkdir_data);
492
493struct proc_dir_entry *proc_mkdir_mode(const char *name, umode_t mode,
494 struct proc_dir_entry *parent)
495{
496 return proc_mkdir_data(name, mode, parent, NULL);
497}
498EXPORT_SYMBOL(proc_mkdir_mode);
499
500struct proc_dir_entry *proc_mkdir(const char *name,
501 struct proc_dir_entry *parent)
502{
503 return proc_mkdir_data(name, 0, parent, NULL);
504}
505EXPORT_SYMBOL(proc_mkdir);
506
507struct proc_dir_entry *proc_create_mount_point(const char *name)
508{
509 umode_t mode = S_IFDIR | S_IRUGO | S_IXUGO;
510 struct proc_dir_entry *ent, *parent = NULL;
511
512 ent = __proc_create(&parent, name, mode, 2);
513 if (ent) {
514 ent->data = NULL;
515 ent->proc_dir_ops = NULL;
516 ent->proc_iops = NULL;
517 ent = proc_register(parent, ent);
518 }
519 return ent;
520}
521EXPORT_SYMBOL(proc_create_mount_point);
522
523struct proc_dir_entry *proc_create_reg(const char *name, umode_t mode,
524 struct proc_dir_entry **parent, void *data)
525{
526 struct proc_dir_entry *p;
527
528 if ((mode & S_IFMT) == 0)
529 mode |= S_IFREG;
530 if ((mode & S_IALLUGO) == 0)
531 mode |= S_IRUGO;
532 if (WARN_ON_ONCE(!S_ISREG(mode)))
533 return NULL;
534
535 p = __proc_create(parent, name, mode, 1);
536 if (p) {
537 p->proc_iops = &proc_file_inode_operations;
538 p->data = data;
539 }
540 return p;
541}
542
543static inline void pde_set_flags(struct proc_dir_entry *pde)
544{
545 if (pde->proc_ops->proc_flags & PROC_ENTRY_PERMANENT)
546 pde->flags |= PROC_ENTRY_PERMANENT;
547}
548
549struct proc_dir_entry *proc_create_data(const char *name, umode_t mode,
550 struct proc_dir_entry *parent,
551 const struct proc_ops *proc_ops, void *data)
552{
553 struct proc_dir_entry *p;
554
555 p = proc_create_reg(name, mode, &parent, data);
556 if (!p)
557 return NULL;
558 p->proc_ops = proc_ops;
559 pde_set_flags(p);
560 return proc_register(parent, p);
561}
562EXPORT_SYMBOL(proc_create_data);
563
564struct proc_dir_entry *proc_create(const char *name, umode_t mode,
565 struct proc_dir_entry *parent,
566 const struct proc_ops *proc_ops)
567{
568 return proc_create_data(name, mode, parent, proc_ops, NULL);
569}
570EXPORT_SYMBOL(proc_create);
571
572static int proc_seq_open(struct inode *inode, struct file *file)
573{
574 struct proc_dir_entry *de = PDE(inode);
575
576 if (de->state_size)
577 return seq_open_private(file, de->seq_ops, de->state_size);
578 return seq_open(file, de->seq_ops);
579}
580
581static int proc_seq_release(struct inode *inode, struct file *file)
582{
583 struct proc_dir_entry *de = PDE(inode);
584
585 if (de->state_size)
586 return seq_release_private(inode, file);
587 return seq_release(inode, file);
588}
589
590static const struct proc_ops proc_seq_ops = {
591 /* not permanent -- can call into arbitrary seq_operations */
592 .proc_open = proc_seq_open,
593 .proc_read = seq_read,
594 .proc_lseek = seq_lseek,
595 .proc_release = proc_seq_release,
596};
597
598struct proc_dir_entry *proc_create_seq_private(const char *name, umode_t mode,
599 struct proc_dir_entry *parent, const struct seq_operations *ops,
600 unsigned int state_size, void *data)
601{
602 struct proc_dir_entry *p;
603
604 p = proc_create_reg(name, mode, &parent, data);
605 if (!p)
606 return NULL;
607 p->proc_ops = &proc_seq_ops;
608 p->seq_ops = ops;
609 p->state_size = state_size;
610 return proc_register(parent, p);
611}
612EXPORT_SYMBOL(proc_create_seq_private);
613
614static int proc_single_open(struct inode *inode, struct file *file)
615{
616 struct proc_dir_entry *de = PDE(inode);
617
618 return single_open(file, de->single_show, de->data);
619}
620
621static const struct proc_ops proc_single_ops = {
622 /* not permanent -- can call into arbitrary ->single_show */
623 .proc_open = proc_single_open,
624 .proc_read = seq_read,
625 .proc_lseek = seq_lseek,
626 .proc_release = single_release,
627};
628
629struct proc_dir_entry *proc_create_single_data(const char *name, umode_t mode,
630 struct proc_dir_entry *parent,
631 int (*show)(struct seq_file *, void *), void *data)
632{
633 struct proc_dir_entry *p;
634
635 p = proc_create_reg(name, mode, &parent, data);
636 if (!p)
637 return NULL;
638 p->proc_ops = &proc_single_ops;
639 p->single_show = show;
640 return proc_register(parent, p);
641}
642EXPORT_SYMBOL(proc_create_single_data);
643
644void proc_set_size(struct proc_dir_entry *de, loff_t size)
645{
646 de->size = size;
647}
648EXPORT_SYMBOL(proc_set_size);
649
650void proc_set_user(struct proc_dir_entry *de, kuid_t uid, kgid_t gid)
651{
652 de->uid = uid;
653 de->gid = gid;
654}
655EXPORT_SYMBOL(proc_set_user);
656
657void pde_put(struct proc_dir_entry *pde)
658{
659 if (refcount_dec_and_test(&pde->refcnt)) {
660 proc_free_inum(pde->low_ino);
661 pde_free(pde);
662 }
663}
664
665/*
666 * Remove a /proc entry and free it if it's not currently in use.
667 */
668void remove_proc_entry(const char *name, struct proc_dir_entry *parent)
669{
670 struct proc_dir_entry *de = NULL;
671 const char *fn = name;
672 unsigned int len;
673
674 write_lock(&proc_subdir_lock);
675 if (__xlate_proc_name(name, &parent, &fn) != 0) {
676 write_unlock(&proc_subdir_lock);
677 return;
678 }
679 len = strlen(fn);
680
681 de = pde_subdir_find(parent, fn, len);
682 if (de) {
683 if (unlikely(pde_is_permanent(de))) {
684 WARN(1, "removing permanent /proc entry '%s'", de->name);
685 de = NULL;
686 } else {
687 rb_erase(&de->subdir_node, &parent->subdir);
688 if (S_ISDIR(de->mode))
689 parent->nlink--;
690 }
691 }
692 write_unlock(&proc_subdir_lock);
693 if (!de) {
694 WARN(1, "name '%s'\n", name);
695 return;
696 }
697
698 proc_entry_rundown(de);
699
700 WARN(pde_subdir_first(de),
701 "%s: removing non-empty directory '%s/%s', leaking at least '%s'\n",
702 __func__, de->parent->name, de->name, pde_subdir_first(de)->name);
703 pde_put(de);
704}
705EXPORT_SYMBOL(remove_proc_entry);
706
707int remove_proc_subtree(const char *name, struct proc_dir_entry *parent)
708{
709 struct proc_dir_entry *root = NULL, *de, *next;
710 const char *fn = name;
711 unsigned int len;
712
713 write_lock(&proc_subdir_lock);
714 if (__xlate_proc_name(name, &parent, &fn) != 0) {
715 write_unlock(&proc_subdir_lock);
716 return -ENOENT;
717 }
718 len = strlen(fn);
719
720 root = pde_subdir_find(parent, fn, len);
721 if (!root) {
722 write_unlock(&proc_subdir_lock);
723 return -ENOENT;
724 }
725 if (unlikely(pde_is_permanent(root))) {
726 write_unlock(&proc_subdir_lock);
727 WARN(1, "removing permanent /proc entry '%s/%s'",
728 root->parent->name, root->name);
729 return -EINVAL;
730 }
731 rb_erase(&root->subdir_node, &parent->subdir);
732
733 de = root;
734 while (1) {
735 next = pde_subdir_first(de);
736 if (next) {
737 if (unlikely(pde_is_permanent(root))) {
738 write_unlock(&proc_subdir_lock);
739 WARN(1, "removing permanent /proc entry '%s/%s'",
740 next->parent->name, next->name);
741 return -EINVAL;
742 }
743 rb_erase(&next->subdir_node, &de->subdir);
744 de = next;
745 continue;
746 }
747 next = de->parent;
748 if (S_ISDIR(de->mode))
749 next->nlink--;
750 write_unlock(&proc_subdir_lock);
751
752 proc_entry_rundown(de);
753 if (de == root)
754 break;
755 pde_put(de);
756
757 write_lock(&proc_subdir_lock);
758 de = next;
759 }
760 pde_put(root);
761 return 0;
762}
763EXPORT_SYMBOL(remove_proc_subtree);
764
765void *proc_get_parent_data(const struct inode *inode)
766{
767 struct proc_dir_entry *de = PDE(inode);
768 return de->parent->data;
769}
770EXPORT_SYMBOL_GPL(proc_get_parent_data);
771
772void proc_remove(struct proc_dir_entry *de)
773{
774 if (de)
775 remove_proc_subtree(de->name, de->parent);
776}
777EXPORT_SYMBOL(proc_remove);
778
779void *PDE_DATA(const struct inode *inode)
780{
781 return __PDE_DATA(inode);
782}
783EXPORT_SYMBOL(PDE_DATA);
784
785/*
786 * Pull a user buffer into memory and pass it to the file's write handler if
787 * one is supplied. The ->write() method is permitted to modify the
788 * kernel-side buffer.
789 */
790ssize_t proc_simple_write(struct file *f, const char __user *ubuf, size_t size,
791 loff_t *_pos)
792{
793 struct proc_dir_entry *pde = PDE(file_inode(f));
794 char *buf;
795 int ret;
796
797 if (!pde->write)
798 return -EACCES;
799 if (size == 0 || size > PAGE_SIZE - 1)
800 return -EINVAL;
801 buf = memdup_user_nul(ubuf, size);
802 if (IS_ERR(buf))
803 return PTR_ERR(buf);
804 ret = pde->write(f, buf, size);
805 kfree(buf);
806 return ret == 0 ? size : ret;
807}
1// SPDX-License-Identifier: GPL-2.0-only
2/*
3 * proc/fs/generic.c --- generic routines for the proc-fs
4 *
5 * This file contains generic proc-fs routines for handling
6 * directories and files.
7 *
8 * Copyright (C) 1991, 1992 Linus Torvalds.
9 * Copyright (C) 1997 Theodore Ts'o
10 */
11
12#include <linux/cache.h>
13#include <linux/errno.h>
14#include <linux/time.h>
15#include <linux/proc_fs.h>
16#include <linux/stat.h>
17#include <linux/mm.h>
18#include <linux/module.h>
19#include <linux/namei.h>
20#include <linux/slab.h>
21#include <linux/printk.h>
22#include <linux/mount.h>
23#include <linux/init.h>
24#include <linux/idr.h>
25#include <linux/bitops.h>
26#include <linux/spinlock.h>
27#include <linux/completion.h>
28#include <linux/uaccess.h>
29#include <linux/seq_file.h>
30
31#include "internal.h"
32
33static DEFINE_RWLOCK(proc_subdir_lock);
34
35struct kmem_cache *proc_dir_entry_cache __ro_after_init;
36
37void pde_free(struct proc_dir_entry *pde)
38{
39 if (S_ISLNK(pde->mode))
40 kfree(pde->data);
41 if (pde->name != pde->inline_name)
42 kfree(pde->name);
43 kmem_cache_free(proc_dir_entry_cache, pde);
44}
45
46static int proc_match(const char *name, struct proc_dir_entry *de, unsigned int len)
47{
48 if (len < de->namelen)
49 return -1;
50 if (len > de->namelen)
51 return 1;
52
53 return memcmp(name, de->name, len);
54}
55
56static struct proc_dir_entry *pde_subdir_first(struct proc_dir_entry *dir)
57{
58 return rb_entry_safe(rb_first(&dir->subdir), struct proc_dir_entry,
59 subdir_node);
60}
61
62static struct proc_dir_entry *pde_subdir_next(struct proc_dir_entry *dir)
63{
64 return rb_entry_safe(rb_next(&dir->subdir_node), struct proc_dir_entry,
65 subdir_node);
66}
67
68static struct proc_dir_entry *pde_subdir_find(struct proc_dir_entry *dir,
69 const char *name,
70 unsigned int len)
71{
72 struct rb_node *node = dir->subdir.rb_node;
73
74 while (node) {
75 struct proc_dir_entry *de = rb_entry(node,
76 struct proc_dir_entry,
77 subdir_node);
78 int result = proc_match(name, de, len);
79
80 if (result < 0)
81 node = node->rb_left;
82 else if (result > 0)
83 node = node->rb_right;
84 else
85 return de;
86 }
87 return NULL;
88}
89
90static bool pde_subdir_insert(struct proc_dir_entry *dir,
91 struct proc_dir_entry *de)
92{
93 struct rb_root *root = &dir->subdir;
94 struct rb_node **new = &root->rb_node, *parent = NULL;
95
96 /* Figure out where to put new node */
97 while (*new) {
98 struct proc_dir_entry *this = rb_entry(*new,
99 struct proc_dir_entry,
100 subdir_node);
101 int result = proc_match(de->name, this, de->namelen);
102
103 parent = *new;
104 if (result < 0)
105 new = &(*new)->rb_left;
106 else if (result > 0)
107 new = &(*new)->rb_right;
108 else
109 return false;
110 }
111
112 /* Add new node and rebalance tree. */
113 rb_link_node(&de->subdir_node, parent, new);
114 rb_insert_color(&de->subdir_node, root);
115 return true;
116}
117
118static int proc_notify_change(struct dentry *dentry, struct iattr *iattr)
119{
120 struct inode *inode = d_inode(dentry);
121 struct proc_dir_entry *de = PDE(inode);
122 int error;
123
124 error = setattr_prepare(dentry, iattr);
125 if (error)
126 return error;
127
128 setattr_copy(inode, iattr);
129 mark_inode_dirty(inode);
130
131 proc_set_user(de, inode->i_uid, inode->i_gid);
132 de->mode = inode->i_mode;
133 return 0;
134}
135
136static int proc_getattr(const struct path *path, struct kstat *stat,
137 u32 request_mask, unsigned int query_flags)
138{
139 struct inode *inode = d_inode(path->dentry);
140 struct proc_dir_entry *de = PDE(inode);
141 if (de && de->nlink)
142 set_nlink(inode, de->nlink);
143
144 generic_fillattr(inode, stat);
145 return 0;
146}
147
148static const struct inode_operations proc_file_inode_operations = {
149 .setattr = proc_notify_change,
150};
151
152/*
153 * This function parses a name such as "tty/driver/serial", and
154 * returns the struct proc_dir_entry for "/proc/tty/driver", and
155 * returns "serial" in residual.
156 */
157static int __xlate_proc_name(const char *name, struct proc_dir_entry **ret,
158 const char **residual)
159{
160 const char *cp = name, *next;
161 struct proc_dir_entry *de;
162 unsigned int len;
163
164 de = *ret;
165 if (!de)
166 de = &proc_root;
167
168 while (1) {
169 next = strchr(cp, '/');
170 if (!next)
171 break;
172
173 len = next - cp;
174 de = pde_subdir_find(de, cp, len);
175 if (!de) {
176 WARN(1, "name '%s'\n", name);
177 return -ENOENT;
178 }
179 cp += len + 1;
180 }
181 *residual = cp;
182 *ret = de;
183 return 0;
184}
185
186static int xlate_proc_name(const char *name, struct proc_dir_entry **ret,
187 const char **residual)
188{
189 int rv;
190
191 read_lock(&proc_subdir_lock);
192 rv = __xlate_proc_name(name, ret, residual);
193 read_unlock(&proc_subdir_lock);
194 return rv;
195}
196
197static DEFINE_IDA(proc_inum_ida);
198
199#define PROC_DYNAMIC_FIRST 0xF0000000U
200
201/*
202 * Return an inode number between PROC_DYNAMIC_FIRST and
203 * 0xffffffff, or zero on failure.
204 */
205int proc_alloc_inum(unsigned int *inum)
206{
207 int i;
208
209 i = ida_simple_get(&proc_inum_ida, 0, UINT_MAX - PROC_DYNAMIC_FIRST + 1,
210 GFP_KERNEL);
211 if (i < 0)
212 return i;
213
214 *inum = PROC_DYNAMIC_FIRST + (unsigned int)i;
215 return 0;
216}
217
218void proc_free_inum(unsigned int inum)
219{
220 ida_simple_remove(&proc_inum_ida, inum - PROC_DYNAMIC_FIRST);
221}
222
223static int proc_misc_d_revalidate(struct dentry *dentry, unsigned int flags)
224{
225 if (flags & LOOKUP_RCU)
226 return -ECHILD;
227
228 if (atomic_read(&PDE(d_inode(dentry))->in_use) < 0)
229 return 0; /* revalidate */
230 return 1;
231}
232
233static int proc_misc_d_delete(const struct dentry *dentry)
234{
235 return atomic_read(&PDE(d_inode(dentry))->in_use) < 0;
236}
237
238static const struct dentry_operations proc_misc_dentry_ops = {
239 .d_revalidate = proc_misc_d_revalidate,
240 .d_delete = proc_misc_d_delete,
241};
242
243/*
244 * Don't create negative dentries here, return -ENOENT by hand
245 * instead.
246 */
247struct dentry *proc_lookup_de(struct inode *dir, struct dentry *dentry,
248 struct proc_dir_entry *de)
249{
250 struct inode *inode;
251
252 read_lock(&proc_subdir_lock);
253 de = pde_subdir_find(de, dentry->d_name.name, dentry->d_name.len);
254 if (de) {
255 pde_get(de);
256 read_unlock(&proc_subdir_lock);
257 inode = proc_get_inode(dir->i_sb, de);
258 if (!inode)
259 return ERR_PTR(-ENOMEM);
260 d_set_d_op(dentry, de->proc_dops);
261 return d_splice_alias(inode, dentry);
262 }
263 read_unlock(&proc_subdir_lock);
264 return ERR_PTR(-ENOENT);
265}
266
267struct dentry *proc_lookup(struct inode *dir, struct dentry *dentry,
268 unsigned int flags)
269{
270 return proc_lookup_de(dir, dentry, PDE(dir));
271}
272
273/*
274 * This returns non-zero if at EOF, so that the /proc
275 * root directory can use this and check if it should
276 * continue with the <pid> entries..
277 *
278 * Note that the VFS-layer doesn't care about the return
279 * value of the readdir() call, as long as it's non-negative
280 * for success..
281 */
282int proc_readdir_de(struct file *file, struct dir_context *ctx,
283 struct proc_dir_entry *de)
284{
285 int i;
286
287 if (!dir_emit_dots(file, ctx))
288 return 0;
289
290 i = ctx->pos - 2;
291 read_lock(&proc_subdir_lock);
292 de = pde_subdir_first(de);
293 for (;;) {
294 if (!de) {
295 read_unlock(&proc_subdir_lock);
296 return 0;
297 }
298 if (!i)
299 break;
300 de = pde_subdir_next(de);
301 i--;
302 }
303
304 do {
305 struct proc_dir_entry *next;
306 pde_get(de);
307 read_unlock(&proc_subdir_lock);
308 if (!dir_emit(ctx, de->name, de->namelen,
309 de->low_ino, de->mode >> 12)) {
310 pde_put(de);
311 return 0;
312 }
313 ctx->pos++;
314 read_lock(&proc_subdir_lock);
315 next = pde_subdir_next(de);
316 pde_put(de);
317 de = next;
318 } while (de);
319 read_unlock(&proc_subdir_lock);
320 return 1;
321}
322
323int proc_readdir(struct file *file, struct dir_context *ctx)
324{
325 struct inode *inode = file_inode(file);
326
327 return proc_readdir_de(file, ctx, PDE(inode));
328}
329
330/*
331 * These are the generic /proc directory operations. They
332 * use the in-memory "struct proc_dir_entry" tree to parse
333 * the /proc directory.
334 */
335static const struct file_operations proc_dir_operations = {
336 .llseek = generic_file_llseek,
337 .read = generic_read_dir,
338 .iterate_shared = proc_readdir,
339};
340
341/*
342 * proc directories can do almost nothing..
343 */
344static const struct inode_operations proc_dir_inode_operations = {
345 .lookup = proc_lookup,
346 .getattr = proc_getattr,
347 .setattr = proc_notify_change,
348};
349
350/* returns the registered entry, or frees dp and returns NULL on failure */
351struct proc_dir_entry *proc_register(struct proc_dir_entry *dir,
352 struct proc_dir_entry *dp)
353{
354 if (proc_alloc_inum(&dp->low_ino))
355 goto out_free_entry;
356
357 write_lock(&proc_subdir_lock);
358 dp->parent = dir;
359 if (pde_subdir_insert(dir, dp) == false) {
360 WARN(1, "proc_dir_entry '%s/%s' already registered\n",
361 dir->name, dp->name);
362 write_unlock(&proc_subdir_lock);
363 goto out_free_inum;
364 }
365 write_unlock(&proc_subdir_lock);
366
367 return dp;
368out_free_inum:
369 proc_free_inum(dp->low_ino);
370out_free_entry:
371 pde_free(dp);
372 return NULL;
373}
374
375static struct proc_dir_entry *__proc_create(struct proc_dir_entry **parent,
376 const char *name,
377 umode_t mode,
378 nlink_t nlink)
379{
380 struct proc_dir_entry *ent = NULL;
381 const char *fn;
382 struct qstr qstr;
383
384 if (xlate_proc_name(name, parent, &fn) != 0)
385 goto out;
386 qstr.name = fn;
387 qstr.len = strlen(fn);
388 if (qstr.len == 0 || qstr.len >= 256) {
389 WARN(1, "name len %u\n", qstr.len);
390 return NULL;
391 }
392 if (qstr.len == 1 && fn[0] == '.') {
393 WARN(1, "name '.'\n");
394 return NULL;
395 }
396 if (qstr.len == 2 && fn[0] == '.' && fn[1] == '.') {
397 WARN(1, "name '..'\n");
398 return NULL;
399 }
400 if (*parent == &proc_root && name_to_int(&qstr) != ~0U) {
401 WARN(1, "create '/proc/%s' by hand\n", qstr.name);
402 return NULL;
403 }
404 if (is_empty_pde(*parent)) {
405 WARN(1, "attempt to add to permanently empty directory");
406 return NULL;
407 }
408
409 ent = kmem_cache_zalloc(proc_dir_entry_cache, GFP_KERNEL);
410 if (!ent)
411 goto out;
412
413 if (qstr.len + 1 <= SIZEOF_PDE_INLINE_NAME) {
414 ent->name = ent->inline_name;
415 } else {
416 ent->name = kmalloc(qstr.len + 1, GFP_KERNEL);
417 if (!ent->name) {
418 pde_free(ent);
419 return NULL;
420 }
421 }
422
423 memcpy(ent->name, fn, qstr.len + 1);
424 ent->namelen = qstr.len;
425 ent->mode = mode;
426 ent->nlink = nlink;
427 ent->subdir = RB_ROOT;
428 refcount_set(&ent->refcnt, 1);
429 spin_lock_init(&ent->pde_unload_lock);
430 INIT_LIST_HEAD(&ent->pde_openers);
431 proc_set_user(ent, (*parent)->uid, (*parent)->gid);
432
433 ent->proc_dops = &proc_misc_dentry_ops;
434
435out:
436 return ent;
437}
438
439struct proc_dir_entry *proc_symlink(const char *name,
440 struct proc_dir_entry *parent, const char *dest)
441{
442 struct proc_dir_entry *ent;
443
444 ent = __proc_create(&parent, name,
445 (S_IFLNK | S_IRUGO | S_IWUGO | S_IXUGO),1);
446
447 if (ent) {
448 ent->data = kmalloc((ent->size=strlen(dest))+1, GFP_KERNEL);
449 if (ent->data) {
450 strcpy((char*)ent->data,dest);
451 ent->proc_iops = &proc_link_inode_operations;
452 ent = proc_register(parent, ent);
453 } else {
454 pde_free(ent);
455 ent = NULL;
456 }
457 }
458 return ent;
459}
460EXPORT_SYMBOL(proc_symlink);
461
462struct proc_dir_entry *proc_mkdir_data(const char *name, umode_t mode,
463 struct proc_dir_entry *parent, void *data)
464{
465 struct proc_dir_entry *ent;
466
467 if (mode == 0)
468 mode = S_IRUGO | S_IXUGO;
469
470 ent = __proc_create(&parent, name, S_IFDIR | mode, 2);
471 if (ent) {
472 ent->data = data;
473 ent->proc_fops = &proc_dir_operations;
474 ent->proc_iops = &proc_dir_inode_operations;
475 parent->nlink++;
476 ent = proc_register(parent, ent);
477 if (!ent)
478 parent->nlink--;
479 }
480 return ent;
481}
482EXPORT_SYMBOL_GPL(proc_mkdir_data);
483
484struct proc_dir_entry *proc_mkdir_mode(const char *name, umode_t mode,
485 struct proc_dir_entry *parent)
486{
487 return proc_mkdir_data(name, mode, parent, NULL);
488}
489EXPORT_SYMBOL(proc_mkdir_mode);
490
491struct proc_dir_entry *proc_mkdir(const char *name,
492 struct proc_dir_entry *parent)
493{
494 return proc_mkdir_data(name, 0, parent, NULL);
495}
496EXPORT_SYMBOL(proc_mkdir);
497
498struct proc_dir_entry *proc_create_mount_point(const char *name)
499{
500 umode_t mode = S_IFDIR | S_IRUGO | S_IXUGO;
501 struct proc_dir_entry *ent, *parent = NULL;
502
503 ent = __proc_create(&parent, name, mode, 2);
504 if (ent) {
505 ent->data = NULL;
506 ent->proc_fops = NULL;
507 ent->proc_iops = NULL;
508 parent->nlink++;
509 ent = proc_register(parent, ent);
510 if (!ent)
511 parent->nlink--;
512 }
513 return ent;
514}
515EXPORT_SYMBOL(proc_create_mount_point);
516
517struct proc_dir_entry *proc_create_reg(const char *name, umode_t mode,
518 struct proc_dir_entry **parent, void *data)
519{
520 struct proc_dir_entry *p;
521
522 if ((mode & S_IFMT) == 0)
523 mode |= S_IFREG;
524 if ((mode & S_IALLUGO) == 0)
525 mode |= S_IRUGO;
526 if (WARN_ON_ONCE(!S_ISREG(mode)))
527 return NULL;
528
529 p = __proc_create(parent, name, mode, 1);
530 if (p) {
531 p->proc_iops = &proc_file_inode_operations;
532 p->data = data;
533 }
534 return p;
535}
536
537struct proc_dir_entry *proc_create_data(const char *name, umode_t mode,
538 struct proc_dir_entry *parent,
539 const struct file_operations *proc_fops, void *data)
540{
541 struct proc_dir_entry *p;
542
543 BUG_ON(proc_fops == NULL);
544
545 p = proc_create_reg(name, mode, &parent, data);
546 if (!p)
547 return NULL;
548 p->proc_fops = proc_fops;
549 return proc_register(parent, p);
550}
551EXPORT_SYMBOL(proc_create_data);
552
553struct proc_dir_entry *proc_create(const char *name, umode_t mode,
554 struct proc_dir_entry *parent,
555 const struct file_operations *proc_fops)
556{
557 return proc_create_data(name, mode, parent, proc_fops, NULL);
558}
559EXPORT_SYMBOL(proc_create);
560
561static int proc_seq_open(struct inode *inode, struct file *file)
562{
563 struct proc_dir_entry *de = PDE(inode);
564
565 if (de->state_size)
566 return seq_open_private(file, de->seq_ops, de->state_size);
567 return seq_open(file, de->seq_ops);
568}
569
570static int proc_seq_release(struct inode *inode, struct file *file)
571{
572 struct proc_dir_entry *de = PDE(inode);
573
574 if (de->state_size)
575 return seq_release_private(inode, file);
576 return seq_release(inode, file);
577}
578
579static const struct file_operations proc_seq_fops = {
580 .open = proc_seq_open,
581 .read = seq_read,
582 .llseek = seq_lseek,
583 .release = proc_seq_release,
584};
585
586struct proc_dir_entry *proc_create_seq_private(const char *name, umode_t mode,
587 struct proc_dir_entry *parent, const struct seq_operations *ops,
588 unsigned int state_size, void *data)
589{
590 struct proc_dir_entry *p;
591
592 p = proc_create_reg(name, mode, &parent, data);
593 if (!p)
594 return NULL;
595 p->proc_fops = &proc_seq_fops;
596 p->seq_ops = ops;
597 p->state_size = state_size;
598 return proc_register(parent, p);
599}
600EXPORT_SYMBOL(proc_create_seq_private);
601
602static int proc_single_open(struct inode *inode, struct file *file)
603{
604 struct proc_dir_entry *de = PDE(inode);
605
606 return single_open(file, de->single_show, de->data);
607}
608
609static const struct file_operations proc_single_fops = {
610 .open = proc_single_open,
611 .read = seq_read,
612 .llseek = seq_lseek,
613 .release = single_release,
614};
615
616struct proc_dir_entry *proc_create_single_data(const char *name, umode_t mode,
617 struct proc_dir_entry *parent,
618 int (*show)(struct seq_file *, void *), void *data)
619{
620 struct proc_dir_entry *p;
621
622 p = proc_create_reg(name, mode, &parent, data);
623 if (!p)
624 return NULL;
625 p->proc_fops = &proc_single_fops;
626 p->single_show = show;
627 return proc_register(parent, p);
628}
629EXPORT_SYMBOL(proc_create_single_data);
630
631void proc_set_size(struct proc_dir_entry *de, loff_t size)
632{
633 de->size = size;
634}
635EXPORT_SYMBOL(proc_set_size);
636
637void proc_set_user(struct proc_dir_entry *de, kuid_t uid, kgid_t gid)
638{
639 de->uid = uid;
640 de->gid = gid;
641}
642EXPORT_SYMBOL(proc_set_user);
643
644void pde_put(struct proc_dir_entry *pde)
645{
646 if (refcount_dec_and_test(&pde->refcnt)) {
647 proc_free_inum(pde->low_ino);
648 pde_free(pde);
649 }
650}
651
652/*
653 * Remove a /proc entry and free it if it's not currently in use.
654 */
655void remove_proc_entry(const char *name, struct proc_dir_entry *parent)
656{
657 struct proc_dir_entry *de = NULL;
658 const char *fn = name;
659 unsigned int len;
660
661 write_lock(&proc_subdir_lock);
662 if (__xlate_proc_name(name, &parent, &fn) != 0) {
663 write_unlock(&proc_subdir_lock);
664 return;
665 }
666 len = strlen(fn);
667
668 de = pde_subdir_find(parent, fn, len);
669 if (de)
670 rb_erase(&de->subdir_node, &parent->subdir);
671 write_unlock(&proc_subdir_lock);
672 if (!de) {
673 WARN(1, "name '%s'\n", name);
674 return;
675 }
676
677 proc_entry_rundown(de);
678
679 if (S_ISDIR(de->mode))
680 parent->nlink--;
681 de->nlink = 0;
682 WARN(pde_subdir_first(de),
683 "%s: removing non-empty directory '%s/%s', leaking at least '%s'\n",
684 __func__, de->parent->name, de->name, pde_subdir_first(de)->name);
685 pde_put(de);
686}
687EXPORT_SYMBOL(remove_proc_entry);
688
689int remove_proc_subtree(const char *name, struct proc_dir_entry *parent)
690{
691 struct proc_dir_entry *root = NULL, *de, *next;
692 const char *fn = name;
693 unsigned int len;
694
695 write_lock(&proc_subdir_lock);
696 if (__xlate_proc_name(name, &parent, &fn) != 0) {
697 write_unlock(&proc_subdir_lock);
698 return -ENOENT;
699 }
700 len = strlen(fn);
701
702 root = pde_subdir_find(parent, fn, len);
703 if (!root) {
704 write_unlock(&proc_subdir_lock);
705 return -ENOENT;
706 }
707 rb_erase(&root->subdir_node, &parent->subdir);
708
709 de = root;
710 while (1) {
711 next = pde_subdir_first(de);
712 if (next) {
713 rb_erase(&next->subdir_node, &de->subdir);
714 de = next;
715 continue;
716 }
717 write_unlock(&proc_subdir_lock);
718
719 proc_entry_rundown(de);
720 next = de->parent;
721 if (S_ISDIR(de->mode))
722 next->nlink--;
723 de->nlink = 0;
724 if (de == root)
725 break;
726 pde_put(de);
727
728 write_lock(&proc_subdir_lock);
729 de = next;
730 }
731 pde_put(root);
732 return 0;
733}
734EXPORT_SYMBOL(remove_proc_subtree);
735
736void *proc_get_parent_data(const struct inode *inode)
737{
738 struct proc_dir_entry *de = PDE(inode);
739 return de->parent->data;
740}
741EXPORT_SYMBOL_GPL(proc_get_parent_data);
742
743void proc_remove(struct proc_dir_entry *de)
744{
745 if (de)
746 remove_proc_subtree(de->name, de->parent);
747}
748EXPORT_SYMBOL(proc_remove);
749
750void *PDE_DATA(const struct inode *inode)
751{
752 return __PDE_DATA(inode);
753}
754EXPORT_SYMBOL(PDE_DATA);
755
756/*
757 * Pull a user buffer into memory and pass it to the file's write handler if
758 * one is supplied. The ->write() method is permitted to modify the
759 * kernel-side buffer.
760 */
761ssize_t proc_simple_write(struct file *f, const char __user *ubuf, size_t size,
762 loff_t *_pos)
763{
764 struct proc_dir_entry *pde = PDE(file_inode(f));
765 char *buf;
766 int ret;
767
768 if (!pde->write)
769 return -EACCES;
770 if (size == 0 || size > PAGE_SIZE - 1)
771 return -EINVAL;
772 buf = memdup_user_nul(ubuf, size);
773 if (IS_ERR(buf))
774 return PTR_ERR(buf);
775 ret = pde->write(f, buf, size);
776 kfree(buf);
777 return ret == 0 ? size : ret;
778}