2 * linux/fs/proc/inode.c
4 * Copyright (C) 1991, 1992 Linus Torvalds
7 #include <linux/time.h>
8 #include <linux/proc_fs.h>
9 #include <linux/kernel.h>
10 #include <linux/pid_namespace.h>
12 #include <linux/string.h>
13 #include <linux/stat.h>
14 #include <linux/completion.h>
15 #include <linux/poll.h>
16 #include <linux/file.h>
17 #include <linux/limits.h>
18 #include <linux/init.h>
19 #include <linux/module.h>
20 #include <linux/sysctl.h>
21 #include <linux/seq_file.h>
22 #include <linux/slab.h>
23 #include <linux/mount.h>
25 #include <asm/uaccess.h>
29 static void proc_evict_inode(struct inode
*inode
)
31 struct proc_dir_entry
*de
;
32 struct ctl_table_header
*head
;
33 const struct proc_ns_operations
*ns_ops
;
35 truncate_inode_pages(&inode
->i_data
, 0);
38 /* Stop tracking associated processes */
39 put_pid(PROC_I(inode
)->pid
);
41 /* Let go of any associated proc directory entry */
42 de
= PROC_I(inode
)->pde
;
45 head
= PROC_I(inode
)->sysctl
;
47 rcu_assign_pointer(PROC_I(inode
)->sysctl
, NULL
);
48 sysctl_head_put(head
);
50 /* Release any associated namespace */
51 ns_ops
= PROC_I(inode
)->ns_ops
;
52 if (ns_ops
&& ns_ops
->put
)
53 ns_ops
->put(PROC_I(inode
)->ns
);
56 static struct kmem_cache
* proc_inode_cachep
;
58 static struct inode
*proc_alloc_inode(struct super_block
*sb
)
60 struct proc_inode
*ei
;
63 ei
= (struct proc_inode
*)kmem_cache_alloc(proc_inode_cachep
, GFP_KERNEL
);
68 ei
->op
.proc_get_link
= NULL
;
71 ei
->sysctl_entry
= NULL
;
74 inode
= &ei
->vfs_inode
;
75 inode
->i_mtime
= inode
->i_atime
= inode
->i_ctime
= CURRENT_TIME
;
79 static void proc_i_callback(struct rcu_head
*head
)
81 struct inode
*inode
= container_of(head
, struct inode
, i_rcu
);
82 kmem_cache_free(proc_inode_cachep
, PROC_I(inode
));
85 static void proc_destroy_inode(struct inode
*inode
)
87 call_rcu(&inode
->i_rcu
, proc_i_callback
);
90 static void init_once(void *foo
)
92 struct proc_inode
*ei
= (struct proc_inode
*) foo
;
94 inode_init_once(&ei
->vfs_inode
);
97 void __init
proc_init_inodecache(void)
99 proc_inode_cachep
= kmem_cache_create("proc_inode_cache",
100 sizeof(struct proc_inode
),
101 0, (SLAB_RECLAIM_ACCOUNT
|
102 SLAB_MEM_SPREAD
|SLAB_PANIC
),
106 static int proc_show_options(struct seq_file
*seq
, struct dentry
*root
)
108 struct super_block
*sb
= root
->d_sb
;
109 struct pid_namespace
*pid
= sb
->s_fs_info
;
111 if (!gid_eq(pid
->pid_gid
, GLOBAL_ROOT_GID
))
112 seq_printf(seq
, ",gid=%u", from_kgid_munged(&init_user_ns
, pid
->pid_gid
));
113 if (pid
->hide_pid
!= 0)
114 seq_printf(seq
, ",hidepid=%u", pid
->hide_pid
);
119 static const struct super_operations proc_sops
= {
120 .alloc_inode
= proc_alloc_inode
,
121 .destroy_inode
= proc_destroy_inode
,
122 .drop_inode
= generic_delete_inode
,
123 .evict_inode
= proc_evict_inode
,
124 .statfs
= simple_statfs
,
125 .remount_fs
= proc_remount
,
126 .show_options
= proc_show_options
,
129 static void __pde_users_dec(struct proc_dir_entry
*pde
)
132 if (pde
->pde_unload_completion
&& pde
->pde_users
== 0)
133 complete(pde
->pde_unload_completion
);
136 void pde_users_dec(struct proc_dir_entry
*pde
)
138 spin_lock(&pde
->pde_unload_lock
);
139 __pde_users_dec(pde
);
140 spin_unlock(&pde
->pde_unload_lock
);
143 static loff_t
proc_reg_llseek(struct file
*file
, loff_t offset
, int whence
)
145 struct proc_dir_entry
*pde
= PDE(file
->f_path
.dentry
->d_inode
);
147 loff_t (*llseek
)(struct file
*, loff_t
, int);
149 spin_lock(&pde
->pde_unload_lock
);
151 * remove_proc_entry() is going to delete PDE (as part of module
152 * cleanup sequence). No new callers into module allowed.
154 if (!pde
->proc_fops
) {
155 spin_unlock(&pde
->pde_unload_lock
);
159 * Bump refcount so that remove_proc_entry will wail for ->llseek to
164 * Save function pointer under lock, to protect against ->proc_fops
165 * NULL'ifying right after ->pde_unload_lock is dropped.
167 llseek
= pde
->proc_fops
->llseek
;
168 spin_unlock(&pde
->pde_unload_lock
);
171 llseek
= default_llseek
;
172 rv
= llseek(file
, offset
, whence
);
178 static ssize_t
proc_reg_read(struct file
*file
, char __user
*buf
, size_t count
, loff_t
*ppos
)
180 struct proc_dir_entry
*pde
= PDE(file
->f_path
.dentry
->d_inode
);
182 ssize_t (*read
)(struct file
*, char __user
*, size_t, loff_t
*);
184 spin_lock(&pde
->pde_unload_lock
);
185 if (!pde
->proc_fops
) {
186 spin_unlock(&pde
->pde_unload_lock
);
190 read
= pde
->proc_fops
->read
;
191 spin_unlock(&pde
->pde_unload_lock
);
194 rv
= read(file
, buf
, count
, ppos
);
200 static ssize_t
proc_reg_write(struct file
*file
, const char __user
*buf
, size_t count
, loff_t
*ppos
)
202 struct proc_dir_entry
*pde
= PDE(file
->f_path
.dentry
->d_inode
);
204 ssize_t (*write
)(struct file
*, const char __user
*, size_t, loff_t
*);
206 spin_lock(&pde
->pde_unload_lock
);
207 if (!pde
->proc_fops
) {
208 spin_unlock(&pde
->pde_unload_lock
);
212 write
= pde
->proc_fops
->write
;
213 spin_unlock(&pde
->pde_unload_lock
);
216 rv
= write(file
, buf
, count
, ppos
);
222 static unsigned int proc_reg_poll(struct file
*file
, struct poll_table_struct
*pts
)
224 struct proc_dir_entry
*pde
= PDE(file
->f_path
.dentry
->d_inode
);
225 unsigned int rv
= DEFAULT_POLLMASK
;
226 unsigned int (*poll
)(struct file
*, struct poll_table_struct
*);
228 spin_lock(&pde
->pde_unload_lock
);
229 if (!pde
->proc_fops
) {
230 spin_unlock(&pde
->pde_unload_lock
);
234 poll
= pde
->proc_fops
->poll
;
235 spin_unlock(&pde
->pde_unload_lock
);
238 rv
= poll(file
, pts
);
244 static long proc_reg_unlocked_ioctl(struct file
*file
, unsigned int cmd
, unsigned long arg
)
246 struct proc_dir_entry
*pde
= PDE(file
->f_path
.dentry
->d_inode
);
248 long (*ioctl
)(struct file
*, unsigned int, unsigned long);
250 spin_lock(&pde
->pde_unload_lock
);
251 if (!pde
->proc_fops
) {
252 spin_unlock(&pde
->pde_unload_lock
);
256 ioctl
= pde
->proc_fops
->unlocked_ioctl
;
257 spin_unlock(&pde
->pde_unload_lock
);
260 rv
= ioctl(file
, cmd
, arg
);
267 static long proc_reg_compat_ioctl(struct file
*file
, unsigned int cmd
, unsigned long arg
)
269 struct proc_dir_entry
*pde
= PDE(file
->f_path
.dentry
->d_inode
);
271 long (*compat_ioctl
)(struct file
*, unsigned int, unsigned long);
273 spin_lock(&pde
->pde_unload_lock
);
274 if (!pde
->proc_fops
) {
275 spin_unlock(&pde
->pde_unload_lock
);
279 compat_ioctl
= pde
->proc_fops
->compat_ioctl
;
280 spin_unlock(&pde
->pde_unload_lock
);
283 rv
= compat_ioctl(file
, cmd
, arg
);
290 static int proc_reg_mmap(struct file
*file
, struct vm_area_struct
*vma
)
292 struct proc_dir_entry
*pde
= PDE(file
->f_path
.dentry
->d_inode
);
294 int (*mmap
)(struct file
*, struct vm_area_struct
*);
296 spin_lock(&pde
->pde_unload_lock
);
297 if (!pde
->proc_fops
) {
298 spin_unlock(&pde
->pde_unload_lock
);
302 mmap
= pde
->proc_fops
->mmap
;
303 spin_unlock(&pde
->pde_unload_lock
);
306 rv
= mmap(file
, vma
);
312 static int proc_reg_open(struct inode
*inode
, struct file
*file
)
314 struct proc_dir_entry
*pde
= PDE(inode
);
316 int (*open
)(struct inode
*, struct file
*);
317 int (*release
)(struct inode
*, struct file
*);
318 struct pde_opener
*pdeo
;
321 * What for, you ask? Well, we can have open, rmmod, remove_proc_entry
322 * sequence. ->release won't be called because ->proc_fops will be
323 * cleared. Depending on complexity of ->release, consequences vary.
325 * We can't wait for mercy when close will be done for real, it's
326 * deadlockable: rmmod foo </proc/foo . So, we're going to do ->release
327 * by hand in remove_proc_entry(). For this, save opener's credentials
330 pdeo
= kmalloc(sizeof(struct pde_opener
), GFP_KERNEL
);
334 spin_lock(&pde
->pde_unload_lock
);
335 if (!pde
->proc_fops
) {
336 spin_unlock(&pde
->pde_unload_lock
);
341 open
= pde
->proc_fops
->open
;
342 release
= pde
->proc_fops
->release
;
343 spin_unlock(&pde
->pde_unload_lock
);
346 rv
= open(inode
, file
);
348 spin_lock(&pde
->pde_unload_lock
);
349 if (rv
== 0 && release
) {
350 /* To know what to release. */
353 /* Strictly for "too late" ->release in proc_reg_release(). */
354 pdeo
->release
= release
;
355 list_add(&pdeo
->lh
, &pde
->pde_openers
);
358 __pde_users_dec(pde
);
359 spin_unlock(&pde
->pde_unload_lock
);
363 static struct pde_opener
*find_pde_opener(struct proc_dir_entry
*pde
,
364 struct inode
*inode
, struct file
*file
)
366 struct pde_opener
*pdeo
;
368 list_for_each_entry(pdeo
, &pde
->pde_openers
, lh
) {
369 if (pdeo
->inode
== inode
&& pdeo
->file
== file
)
375 static int proc_reg_release(struct inode
*inode
, struct file
*file
)
377 struct proc_dir_entry
*pde
= PDE(inode
);
379 int (*release
)(struct inode
*, struct file
*);
380 struct pde_opener
*pdeo
;
382 spin_lock(&pde
->pde_unload_lock
);
383 pdeo
= find_pde_opener(pde
, inode
, file
);
384 if (!pde
->proc_fops
) {
386 * Can't simply exit, __fput() will think that everything is OK,
387 * and move on to freeing struct file. remove_proc_entry() will
388 * find slacker in opener's list and will try to do non-trivial
389 * things with struct file. Therefore, remove opener from list.
391 * But if opener is removed from list, who will ->release it?
395 spin_unlock(&pde
->pde_unload_lock
);
396 rv
= pdeo
->release(inode
, file
);
399 spin_unlock(&pde
->pde_unload_lock
);
403 release
= pde
->proc_fops
->release
;
408 spin_unlock(&pde
->pde_unload_lock
);
411 rv
= release(inode
, file
);
417 static const struct file_operations proc_reg_file_ops
= {
418 .llseek
= proc_reg_llseek
,
419 .read
= proc_reg_read
,
420 .write
= proc_reg_write
,
421 .poll
= proc_reg_poll
,
422 .unlocked_ioctl
= proc_reg_unlocked_ioctl
,
424 .compat_ioctl
= proc_reg_compat_ioctl
,
426 .mmap
= proc_reg_mmap
,
427 .open
= proc_reg_open
,
428 .release
= proc_reg_release
,
432 static const struct file_operations proc_reg_file_ops_no_compat
= {
433 .llseek
= proc_reg_llseek
,
434 .read
= proc_reg_read
,
435 .write
= proc_reg_write
,
436 .poll
= proc_reg_poll
,
437 .unlocked_ioctl
= proc_reg_unlocked_ioctl
,
438 .mmap
= proc_reg_mmap
,
439 .open
= proc_reg_open
,
440 .release
= proc_reg_release
,
444 struct inode
*proc_get_inode(struct super_block
*sb
, struct proc_dir_entry
*de
)
446 struct inode
* inode
;
448 inode
= iget_locked(sb
, de
->low_ino
);
451 if (inode
->i_state
& I_NEW
) {
452 inode
->i_mtime
= inode
->i_atime
= inode
->i_ctime
= CURRENT_TIME
;
453 PROC_I(inode
)->fd
= 0;
454 PROC_I(inode
)->pde
= de
;
457 inode
->i_mode
= de
->mode
;
458 inode
->i_uid
= de
->uid
;
459 inode
->i_gid
= de
->gid
;
462 inode
->i_size
= de
->size
;
464 set_nlink(inode
, de
->nlink
);
466 inode
->i_op
= de
->proc_iops
;
468 if (S_ISREG(inode
->i_mode
)) {
470 if (!de
->proc_fops
->compat_ioctl
)
472 &proc_reg_file_ops_no_compat
;
475 inode
->i_fop
= &proc_reg_file_ops
;
477 inode
->i_fop
= de
->proc_fops
;
480 unlock_new_inode(inode
);
486 int proc_fill_super(struct super_block
*s
)
488 s
->s_flags
|= MS_NODIRATIME
| MS_NOSUID
| MS_NOEXEC
;
489 s
->s_blocksize
= 1024;
490 s
->s_blocksize_bits
= 10;
491 s
->s_magic
= PROC_SUPER_MAGIC
;
492 s
->s_op
= &proc_sops
;
496 s
->s_root
= d_make_root(proc_get_inode(s
, &proc_root
));
500 printk("proc_read_super: get root inode failed\n");