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 */
47 head = PROC_I(inode)->sysctl;
49 RCU_INIT_POINTER(PROC_I(inode)->sysctl, NULL);
50 proc_sys_evict_inode(inode, head);
54 static struct kmem_cache * proc_inode_cachep;
56 static struct inode *proc_alloc_inode(struct super_block *sb)
58 struct proc_inode *ei;
61 ei = kmem_cache_alloc(proc_inode_cachep, GFP_KERNEL);
66 ei->op.proc_get_link = NULL;
69 ei->sysctl_entry = NULL;
71 inode = &ei->vfs_inode;
75 static void proc_i_callback(struct rcu_head *head)
77 struct inode *inode = container_of(head, struct inode, i_rcu);
78 kmem_cache_free(proc_inode_cachep, PROC_I(inode));
81 static void proc_destroy_inode(struct inode *inode)
83 call_rcu(&inode->i_rcu, proc_i_callback);
86 static void init_once(void *foo)
88 struct proc_inode *ei = (struct proc_inode *) foo;
90 inode_init_once(&ei->vfs_inode);
93 void __init proc_init_inodecache(void)
95 proc_inode_cachep = kmem_cache_create("proc_inode_cache",
96 sizeof(struct proc_inode),
97 0, (SLAB_RECLAIM_ACCOUNT|
98 SLAB_MEM_SPREAD|SLAB_ACCOUNT|
103 static int proc_show_options(struct seq_file *seq, struct dentry *root)
105 struct super_block *sb = root->d_sb;
106 struct pid_namespace *pid = sb->s_fs_info;
108 if (!gid_eq(pid->pid_gid, GLOBAL_ROOT_GID))
109 seq_printf(seq, ",gid=%u", from_kgid_munged(&init_user_ns, pid->pid_gid));
110 if (pid->hide_pid != HIDEPID_OFF)
111 seq_printf(seq, ",hidepid=%u", pid->hide_pid);
116 static const struct super_operations proc_sops = {
117 .alloc_inode = proc_alloc_inode,
118 .destroy_inode = proc_destroy_inode,
119 .drop_inode = generic_delete_inode,
120 .evict_inode = proc_evict_inode,
121 .statfs = simple_statfs,
122 .remount_fs = proc_remount,
123 .show_options = proc_show_options,
126 enum {BIAS = -1U<<31};
128 static inline int use_pde(struct proc_dir_entry *pde)
130 return atomic_inc_unless_negative(&pde->in_use);
133 static void unuse_pde(struct proc_dir_entry *pde)
135 if (atomic_dec_return(&pde->in_use) == BIAS)
136 complete(pde->pde_unload_completion);
140 static void close_pdeo(struct proc_dir_entry *pde, struct pde_opener *pdeo)
143 * close() (proc_reg_release()) can't delete an entry and proceed:
144 * ->release hook needs to be available at the right moment.
146 * rmmod (remove_proc_entry() et al) can't delete an entry and proceed:
147 * "struct file" needs to be available at the right moment.
149 * Therefore, first process to enter this function does ->release() and
150 * signals its completion to the other process which does nothing.
153 /* somebody else is doing that, just wait */
154 DECLARE_COMPLETION_ONSTACK(c);
156 spin_unlock(&pde->pde_unload_lock);
157 wait_for_completion(&c);
158 spin_lock(&pde->pde_unload_lock);
161 pdeo->closing = true;
162 spin_unlock(&pde->pde_unload_lock);
164 pde->proc_fops->release(file_inode(file), file);
165 spin_lock(&pde->pde_unload_lock);
166 /* After ->release. */
174 void proc_entry_rundown(struct proc_dir_entry *de)
176 DECLARE_COMPLETION_ONSTACK(c);
177 /* Wait until all existing callers into module are done. */
178 de->pde_unload_completion = &c;
179 if (atomic_add_return(BIAS, &de->in_use) != BIAS)
180 wait_for_completion(&c);
182 /* ->pde_openers list can't grow from now on. */
184 spin_lock(&de->pde_unload_lock);
185 while (!list_empty(&de->pde_openers)) {
186 struct pde_opener *pdeo;
187 pdeo = list_first_entry(&de->pde_openers, struct pde_opener, lh);
188 close_pdeo(de, pdeo);
190 spin_unlock(&de->pde_unload_lock);
193 static loff_t proc_reg_llseek(struct file *file, loff_t offset, int whence)
195 struct proc_dir_entry *pde = PDE(file_inode(file));
198 loff_t (*llseek)(struct file *, loff_t, int);
199 llseek = pde->proc_fops->llseek;
201 llseek = default_llseek;
202 rv = llseek(file, offset, whence);
208 static ssize_t proc_reg_read(struct file *file, char __user *buf, size_t count, loff_t *ppos)
210 ssize_t (*read)(struct file *, char __user *, size_t, loff_t *);
211 struct proc_dir_entry *pde = PDE(file_inode(file));
214 read = pde->proc_fops->read;
216 rv = read(file, buf, count, ppos);
222 static ssize_t proc_reg_write(struct file *file, const char __user *buf, size_t count, loff_t *ppos)
224 ssize_t (*write)(struct file *, const char __user *, size_t, loff_t *);
225 struct proc_dir_entry *pde = PDE(file_inode(file));
228 write = pde->proc_fops->write;
230 rv = write(file, buf, count, ppos);
236 static unsigned int proc_reg_poll(struct file *file, struct poll_table_struct *pts)
238 struct proc_dir_entry *pde = PDE(file_inode(file));
239 unsigned int rv = DEFAULT_POLLMASK;
240 unsigned int (*poll)(struct file *, struct poll_table_struct *);
242 poll = pde->proc_fops->poll;
244 rv = poll(file, pts);
250 static long proc_reg_unlocked_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
252 struct proc_dir_entry *pde = PDE(file_inode(file));
254 long (*ioctl)(struct file *, unsigned int, unsigned long);
256 ioctl = pde->proc_fops->unlocked_ioctl;
258 rv = ioctl(file, cmd, arg);
265 static long proc_reg_compat_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
267 struct proc_dir_entry *pde = PDE(file_inode(file));
269 long (*compat_ioctl)(struct file *, unsigned int, unsigned long);
271 compat_ioctl = pde->proc_fops->compat_ioctl;
273 rv = compat_ioctl(file, cmd, arg);
280 static int proc_reg_mmap(struct file *file, struct vm_area_struct *vma)
282 struct proc_dir_entry *pde = PDE(file_inode(file));
284 int (*mmap)(struct file *, struct vm_area_struct *);
286 mmap = pde->proc_fops->mmap;
288 rv = mmap(file, vma);
295 proc_reg_get_unmapped_area(struct file *file, unsigned long orig_addr,
296 unsigned long len, unsigned long pgoff,
299 struct proc_dir_entry *pde = PDE(file_inode(file));
300 unsigned long rv = -EIO;
303 typeof(proc_reg_get_unmapped_area) *get_area;
305 get_area = pde->proc_fops->get_unmapped_area;
308 get_area = current->mm->get_unmapped_area;
312 rv = get_area(file, orig_addr, len, pgoff, flags);
320 static int proc_reg_open(struct inode *inode, struct file *file)
322 struct proc_dir_entry *pde = PDE(inode);
324 int (*open)(struct inode *, struct file *);
325 int (*release)(struct inode *, struct file *);
326 struct pde_opener *pdeo;
330 * 1) PDE's ->release hook will be called no matter what
331 * either normally by close()/->release, or forcefully by
332 * rmmod/remove_proc_entry.
334 * 2) rmmod isn't blocked by opening file in /proc and sitting on
335 * the descriptor (including "rmmod foo </proc/foo" scenario).
337 * Save every "struct file" with custom ->release hook.
339 pdeo = kmalloc(sizeof(struct pde_opener), GFP_KERNEL);
347 open = pde->proc_fops->open;
348 release = pde->proc_fops->release;
351 rv = open(inode, file);
353 if (rv == 0 && release) {
354 /* To know what to release. */
356 pdeo->closing = false;
358 spin_lock(&pde->pde_unload_lock);
359 list_add(&pdeo->lh, &pde->pde_openers);
360 spin_unlock(&pde->pde_unload_lock);
368 static int proc_reg_release(struct inode *inode, struct file *file)
370 struct proc_dir_entry *pde = PDE(inode);
371 struct pde_opener *pdeo;
372 spin_lock(&pde->pde_unload_lock);
373 list_for_each_entry(pdeo, &pde->pde_openers, lh) {
374 if (pdeo->file == file) {
375 close_pdeo(pde, pdeo);
379 spin_unlock(&pde->pde_unload_lock);
383 static const struct file_operations proc_reg_file_ops = {
384 .llseek = proc_reg_llseek,
385 .read = proc_reg_read,
386 .write = proc_reg_write,
387 .poll = proc_reg_poll,
388 .unlocked_ioctl = proc_reg_unlocked_ioctl,
390 .compat_ioctl = proc_reg_compat_ioctl,
392 .mmap = proc_reg_mmap,
393 .get_unmapped_area = proc_reg_get_unmapped_area,
394 .open = proc_reg_open,
395 .release = proc_reg_release,
399 static const struct file_operations proc_reg_file_ops_no_compat = {
400 .llseek = proc_reg_llseek,
401 .read = proc_reg_read,
402 .write = proc_reg_write,
403 .poll = proc_reg_poll,
404 .unlocked_ioctl = proc_reg_unlocked_ioctl,
405 .mmap = proc_reg_mmap,
406 .get_unmapped_area = proc_reg_get_unmapped_area,
407 .open = proc_reg_open,
408 .release = proc_reg_release,
412 static void proc_put_link(void *p)
417 static const char *proc_get_link(struct dentry *dentry,
419 struct delayed_call *done)
421 struct proc_dir_entry *pde = PDE(inode);
422 if (unlikely(!use_pde(pde)))
423 return ERR_PTR(-EINVAL);
424 set_delayed_call(done, proc_put_link, pde);
428 const struct inode_operations proc_link_inode_operations = {
429 .get_link = proc_get_link,
432 struct inode *proc_get_inode(struct super_block *sb, struct proc_dir_entry *de)
434 struct inode *inode = new_inode_pseudo(sb);
437 inode->i_ino = de->low_ino;
438 inode->i_mtime = inode->i_atime = inode->i_ctime = current_time(inode);
439 PROC_I(inode)->pde = de;
441 if (is_empty_pde(de)) {
442 make_empty_dir_inode(inode);
446 inode->i_mode = de->mode;
447 inode->i_uid = de->uid;
448 inode->i_gid = de->gid;
451 inode->i_size = de->size;
453 set_nlink(inode, de->nlink);
454 WARN_ON(!de->proc_iops);
455 inode->i_op = de->proc_iops;
457 if (S_ISREG(inode->i_mode)) {
459 if (!de->proc_fops->compat_ioctl)
461 &proc_reg_file_ops_no_compat;
464 inode->i_fop = &proc_reg_file_ops;
466 inode->i_fop = de->proc_fops;
474 int proc_fill_super(struct super_block *s, void *data, int silent)
476 struct pid_namespace *ns = get_pid_ns(s->s_fs_info);
477 struct inode *root_inode;
480 if (!proc_parse_options(data, ns))
483 /* User space would break if executables or devices appear on proc */
484 s->s_iflags |= SB_I_USERNS_VISIBLE | SB_I_NOEXEC | SB_I_NODEV;
485 s->s_flags |= MS_NODIRATIME | MS_NOSUID | MS_NOEXEC;
486 s->s_blocksize = 1024;
487 s->s_blocksize_bits = 10;
488 s->s_magic = PROC_SUPER_MAGIC;
489 s->s_op = &proc_sops;
493 * procfs isn't actually a stacking filesystem; however, there is
494 * too much magic going on inside it to permit stacking things on
497 s->s_stack_depth = FILESYSTEM_MAX_STACK_DEPTH;
500 root_inode = proc_get_inode(s, &proc_root);
502 pr_err("proc_fill_super: get root inode failed\n");
506 s->s_root = d_make_root(root_inode);
508 pr_err("proc_fill_super: allocate dentry failed\n");
512 ret = proc_setup_self(s);
516 return proc_setup_thread_self(s);