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/printk.h>
17 #include <linux/file.h>
18 #include <linux/limits.h>
19 #include <linux/init.h>
20 #include <linux/module.h>
21 #include <linux/sysctl.h>
22 #include <linux/seq_file.h>
23 #include <linux/slab.h>
24 #include <linux/mount.h>
25 #include <linux/magic.h>
27 #include <linux/uaccess.h>
31 static void proc_evict_inode(struct inode *inode)
33 struct proc_dir_entry *de;
34 struct ctl_table_header *head;
36 truncate_inode_pages_final(&inode->i_data);
39 /* Stop tracking associated processes */
40 put_pid(PROC_I(inode)->pid);
42 /* Let go of any associated proc directory entry */
46 head = PROC_I(inode)->sysctl;
48 RCU_INIT_POINTER(PROC_I(inode)->sysctl, NULL);
49 sysctl_head_put(head);
53 static struct kmem_cache * proc_inode_cachep;
55 static struct inode *proc_alloc_inode(struct super_block *sb)
57 struct proc_inode *ei;
60 ei = (struct proc_inode *)kmem_cache_alloc(proc_inode_cachep, GFP_KERNEL);
65 ei->op.proc_get_link = NULL;
68 ei->sysctl_entry = NULL;
70 inode = &ei->vfs_inode;
74 static void proc_i_callback(struct rcu_head *head)
76 struct inode *inode = container_of(head, struct inode, i_rcu);
77 kmem_cache_free(proc_inode_cachep, PROC_I(inode));
80 static void proc_destroy_inode(struct inode *inode)
82 call_rcu(&inode->i_rcu, proc_i_callback);
85 static void init_once(void *foo)
87 struct proc_inode *ei = (struct proc_inode *) foo;
89 inode_init_once(&ei->vfs_inode);
92 void __init proc_init_inodecache(void)
94 proc_inode_cachep = kmem_cache_create("proc_inode_cache",
95 sizeof(struct proc_inode),
96 0, (SLAB_RECLAIM_ACCOUNT|
97 SLAB_MEM_SPREAD|SLAB_ACCOUNT|
102 static int proc_show_options(struct seq_file *seq, struct dentry *root)
104 struct super_block *sb = root->d_sb;
105 struct pid_namespace *pid = sb->s_fs_info;
107 if (!gid_eq(pid->pid_gid, GLOBAL_ROOT_GID))
108 seq_printf(seq, ",gid=%u", from_kgid_munged(&init_user_ns, pid->pid_gid));
109 if (pid->hide_pid != 0)
110 seq_printf(seq, ",hidepid=%u", pid->hide_pid);
115 static const struct super_operations proc_sops = {
116 .alloc_inode = proc_alloc_inode,
117 .destroy_inode = proc_destroy_inode,
118 .drop_inode = generic_delete_inode,
119 .evict_inode = proc_evict_inode,
120 .statfs = simple_statfs,
121 .remount_fs = proc_remount,
122 .show_options = proc_show_options,
125 enum {BIAS = -1U<<31};
127 static inline int use_pde(struct proc_dir_entry *pde)
129 return atomic_inc_unless_negative(&pde->in_use);
132 static void unuse_pde(struct proc_dir_entry *pde)
134 if (atomic_dec_return(&pde->in_use) == BIAS)
135 complete(pde->pde_unload_completion);
139 static void close_pdeo(struct proc_dir_entry *pde, struct pde_opener *pdeo)
142 * close() (proc_reg_release()) can't delete an entry and proceed:
143 * ->release hook needs to be available at the right moment.
145 * rmmod (remove_proc_entry() et al) can't delete an entry and proceed:
146 * "struct file" needs to be available at the right moment.
148 * Therefore, first process to enter this function does ->release() and
149 * signals its completion to the other process which does nothing.
152 /* somebody else is doing that, just wait */
153 DECLARE_COMPLETION_ONSTACK(c);
155 spin_unlock(&pde->pde_unload_lock);
156 wait_for_completion(&c);
157 spin_lock(&pde->pde_unload_lock);
160 pdeo->closing = true;
161 spin_unlock(&pde->pde_unload_lock);
163 pde->proc_fops->release(file_inode(file), file);
164 spin_lock(&pde->pde_unload_lock);
165 /* After ->release. */
173 void proc_entry_rundown(struct proc_dir_entry *de)
175 DECLARE_COMPLETION_ONSTACK(c);
176 /* Wait until all existing callers into module are done. */
177 de->pde_unload_completion = &c;
178 if (atomic_add_return(BIAS, &de->in_use) != BIAS)
179 wait_for_completion(&c);
181 /* ->pde_openers list can't grow from now on. */
183 spin_lock(&de->pde_unload_lock);
184 while (!list_empty(&de->pde_openers)) {
185 struct pde_opener *pdeo;
186 pdeo = list_first_entry(&de->pde_openers, struct pde_opener, lh);
187 close_pdeo(de, pdeo);
189 spin_unlock(&de->pde_unload_lock);
192 static loff_t proc_reg_llseek(struct file *file, loff_t offset, int whence)
194 struct proc_dir_entry *pde = PDE(file_inode(file));
197 loff_t (*llseek)(struct file *, loff_t, int);
198 llseek = pde->proc_fops->llseek;
200 llseek = default_llseek;
201 rv = llseek(file, offset, whence);
207 static ssize_t proc_reg_read(struct file *file, char __user *buf, size_t count, loff_t *ppos)
209 ssize_t (*read)(struct file *, char __user *, size_t, loff_t *);
210 struct proc_dir_entry *pde = PDE(file_inode(file));
213 read = pde->proc_fops->read;
215 rv = read(file, buf, count, ppos);
221 static ssize_t proc_reg_write(struct file *file, const char __user *buf, size_t count, loff_t *ppos)
223 ssize_t (*write)(struct file *, const char __user *, size_t, loff_t *);
224 struct proc_dir_entry *pde = PDE(file_inode(file));
227 write = pde->proc_fops->write;
229 rv = write(file, buf, count, ppos);
235 static unsigned int proc_reg_poll(struct file *file, struct poll_table_struct *pts)
237 struct proc_dir_entry *pde = PDE(file_inode(file));
238 unsigned int rv = DEFAULT_POLLMASK;
239 unsigned int (*poll)(struct file *, struct poll_table_struct *);
241 poll = pde->proc_fops->poll;
243 rv = poll(file, pts);
249 static long proc_reg_unlocked_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
251 struct proc_dir_entry *pde = PDE(file_inode(file));
253 long (*ioctl)(struct file *, unsigned int, unsigned long);
255 ioctl = pde->proc_fops->unlocked_ioctl;
257 rv = ioctl(file, cmd, arg);
264 static long proc_reg_compat_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
266 struct proc_dir_entry *pde = PDE(file_inode(file));
268 long (*compat_ioctl)(struct file *, unsigned int, unsigned long);
270 compat_ioctl = pde->proc_fops->compat_ioctl;
272 rv = compat_ioctl(file, cmd, arg);
279 static int proc_reg_mmap(struct file *file, struct vm_area_struct *vma)
281 struct proc_dir_entry *pde = PDE(file_inode(file));
283 int (*mmap)(struct file *, struct vm_area_struct *);
285 mmap = pde->proc_fops->mmap;
287 rv = mmap(file, vma);
294 proc_reg_get_unmapped_area(struct file *file, unsigned long orig_addr,
295 unsigned long len, unsigned long pgoff,
298 struct proc_dir_entry *pde = PDE(file_inode(file));
299 unsigned long rv = -EIO;
302 typeof(proc_reg_get_unmapped_area) *get_area;
304 get_area = pde->proc_fops->get_unmapped_area;
307 get_area = current->mm->get_unmapped_area;
311 rv = get_area(file, orig_addr, len, pgoff, flags);
319 static int proc_reg_open(struct inode *inode, struct file *file)
321 struct proc_dir_entry *pde = PDE(inode);
323 int (*open)(struct inode *, struct file *);
324 int (*release)(struct inode *, struct file *);
325 struct pde_opener *pdeo;
329 * 1) PDE's ->release hook will be called no matter what
330 * either normally by close()/->release, or forcefully by
331 * rmmod/remove_proc_entry.
333 * 2) rmmod isn't blocked by opening file in /proc and sitting on
334 * the descriptor (including "rmmod foo </proc/foo" scenario).
336 * Save every "struct file" with custom ->release hook.
338 pdeo = kmalloc(sizeof(struct pde_opener), GFP_KERNEL);
346 open = pde->proc_fops->open;
347 release = pde->proc_fops->release;
350 rv = open(inode, file);
352 if (rv == 0 && release) {
353 /* To know what to release. */
355 pdeo->closing = false;
357 spin_lock(&pde->pde_unload_lock);
358 list_add(&pdeo->lh, &pde->pde_openers);
359 spin_unlock(&pde->pde_unload_lock);
367 static int proc_reg_release(struct inode *inode, struct file *file)
369 struct proc_dir_entry *pde = PDE(inode);
370 struct pde_opener *pdeo;
371 spin_lock(&pde->pde_unload_lock);
372 list_for_each_entry(pdeo, &pde->pde_openers, lh) {
373 if (pdeo->file == file) {
374 close_pdeo(pde, pdeo);
378 spin_unlock(&pde->pde_unload_lock);
382 static const struct file_operations proc_reg_file_ops = {
383 .llseek = proc_reg_llseek,
384 .read = proc_reg_read,
385 .write = proc_reg_write,
386 .poll = proc_reg_poll,
387 .unlocked_ioctl = proc_reg_unlocked_ioctl,
389 .compat_ioctl = proc_reg_compat_ioctl,
391 .mmap = proc_reg_mmap,
392 .get_unmapped_area = proc_reg_get_unmapped_area,
393 .open = proc_reg_open,
394 .release = proc_reg_release,
398 static const struct file_operations proc_reg_file_ops_no_compat = {
399 .llseek = proc_reg_llseek,
400 .read = proc_reg_read,
401 .write = proc_reg_write,
402 .poll = proc_reg_poll,
403 .unlocked_ioctl = proc_reg_unlocked_ioctl,
404 .mmap = proc_reg_mmap,
405 .get_unmapped_area = proc_reg_get_unmapped_area,
406 .open = proc_reg_open,
407 .release = proc_reg_release,
411 static void proc_put_link(void *p)
416 static const char *proc_get_link(struct dentry *dentry,
418 struct delayed_call *done)
420 struct proc_dir_entry *pde = PDE(inode);
421 if (unlikely(!use_pde(pde)))
422 return ERR_PTR(-EINVAL);
423 set_delayed_call(done, proc_put_link, pde);
427 const struct inode_operations proc_link_inode_operations = {
428 .get_link = proc_get_link,
431 struct inode *proc_get_inode(struct super_block *sb, struct proc_dir_entry *de)
433 struct inode *inode = new_inode_pseudo(sb);
436 inode->i_ino = de->low_ino;
437 inode->i_mtime = inode->i_atime = inode->i_ctime = current_time(inode);
438 PROC_I(inode)->pde = de;
440 if (is_empty_pde(de)) {
441 make_empty_dir_inode(inode);
445 inode->i_mode = de->mode;
446 inode->i_uid = de->uid;
447 inode->i_gid = de->gid;
450 inode->i_size = de->size;
452 set_nlink(inode, de->nlink);
453 WARN_ON(!de->proc_iops);
454 inode->i_op = de->proc_iops;
456 if (S_ISREG(inode->i_mode)) {
458 if (!de->proc_fops->compat_ioctl)
460 &proc_reg_file_ops_no_compat;
463 inode->i_fop = &proc_reg_file_ops;
465 inode->i_fop = de->proc_fops;
473 int proc_fill_super(struct super_block *s, void *data, int silent)
475 struct pid_namespace *ns = get_pid_ns(s->s_fs_info);
476 struct inode *root_inode;
479 if (!proc_parse_options(data, ns))
482 /* User space would break if executables or devices appear on proc */
483 s->s_iflags |= SB_I_USERNS_VISIBLE | SB_I_NOEXEC | SB_I_NODEV;
484 s->s_flags |= MS_NODIRATIME | MS_NOSUID | MS_NOEXEC;
485 s->s_blocksize = 1024;
486 s->s_blocksize_bits = 10;
487 s->s_magic = PROC_SUPER_MAGIC;
488 s->s_op = &proc_sops;
492 * procfs isn't actually a stacking filesystem; however, there is
493 * too much magic going on inside it to permit stacking things on
496 s->s_stack_depth = FILESYSTEM_MAX_STACK_DEPTH;
499 root_inode = proc_get_inode(s, &proc_root);
501 pr_err("proc_fill_super: get root inode failed\n");
505 s->s_root = d_make_root(root_inode);
507 pr_err("proc_fill_super: allocate dentry failed\n");
511 ret = proc_setup_self(s);
515 return proc_setup_thread_self(s);