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