1 // SPDX-License-Identifier: GPL-2.0
3 * linux/fs/proc/inode.c
5 * Copyright (C) 1991, 1992 Linus Torvalds
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>
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>
29 #include <linux/uaccess.h>
33 static void proc_evict_inode(struct inode *inode)
35 struct proc_dir_entry *de;
36 struct ctl_table_header *head;
38 truncate_inode_pages_final(&inode->i_data);
41 /* Stop tracking associated processes */
42 put_pid(PROC_I(inode)->pid);
44 /* Let go of any associated proc directory entry */
49 head = PROC_I(inode)->sysctl;
51 RCU_INIT_POINTER(PROC_I(inode)->sysctl, NULL);
52 proc_sys_evict_inode(inode, head);
56 static struct kmem_cache *proc_inode_cachep __ro_after_init;
58 static struct inode *proc_alloc_inode(struct super_block *sb)
60 struct proc_inode *ei;
63 ei = kmem_cache_alloc(proc_inode_cachep, GFP_KERNEL);
68 ei->op.proc_get_link = NULL;
71 ei->sysctl_entry = NULL;
73 inode = &ei->vfs_inode;
77 static void proc_i_callback(struct rcu_head *head)
79 struct inode *inode = container_of(head, struct inode, i_rcu);
80 kmem_cache_free(proc_inode_cachep, PROC_I(inode));
83 static void proc_destroy_inode(struct inode *inode)
85 call_rcu(&inode->i_rcu, proc_i_callback);
88 static void init_once(void *foo)
90 struct proc_inode *ei = (struct proc_inode *) foo;
92 inode_init_once(&ei->vfs_inode);
95 void __init proc_init_inodecache(void)
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|
105 static int proc_show_options(struct seq_file *seq, struct dentry *root)
107 struct super_block *sb = root->d_sb;
108 struct pid_namespace *pid = sb->s_fs_info;
110 if (!gid_eq(pid->pid_gid, GLOBAL_ROOT_GID))
111 seq_printf(seq, ",gid=%u", from_kgid_munged(&init_user_ns, pid->pid_gid));
112 if (pid->hide_pid != HIDEPID_OFF)
113 seq_printf(seq, ",hidepid=%u", pid->hide_pid);
118 static const struct super_operations proc_sops = {
119 .alloc_inode = proc_alloc_inode,
120 .destroy_inode = proc_destroy_inode,
121 .drop_inode = generic_delete_inode,
122 .evict_inode = proc_evict_inode,
123 .statfs = simple_statfs,
124 .remount_fs = proc_remount,
125 .show_options = proc_show_options,
128 enum {BIAS = -1U<<31};
130 static inline int use_pde(struct proc_dir_entry *pde)
132 return likely(atomic_inc_unless_negative(&pde->in_use));
135 static void unuse_pde(struct proc_dir_entry *pde)
137 if (unlikely(atomic_dec_return(&pde->in_use) == BIAS))
138 complete(pde->pde_unload_completion);
142 static void close_pdeo(struct proc_dir_entry *pde, struct pde_opener *pdeo)
145 * close() (proc_reg_release()) can't delete an entry and proceed:
146 * ->release hook needs to be available at the right moment.
148 * rmmod (remove_proc_entry() et al) can't delete an entry and proceed:
149 * "struct file" needs to be available at the right moment.
151 * Therefore, first process to enter this function does ->release() and
152 * signals its completion to the other process which does nothing.
155 /* somebody else is doing that, just wait */
156 DECLARE_COMPLETION_ONSTACK(c);
158 spin_unlock(&pde->pde_unload_lock);
159 wait_for_completion(&c);
160 spin_lock(&pde->pde_unload_lock);
163 pdeo->closing = true;
164 spin_unlock(&pde->pde_unload_lock);
166 pde->proc_fops->release(file_inode(file), file);
167 spin_lock(&pde->pde_unload_lock);
168 /* After ->release. */
170 if (unlikely(pdeo->c))
176 void proc_entry_rundown(struct proc_dir_entry *de)
178 DECLARE_COMPLETION_ONSTACK(c);
179 /* Wait until all existing callers into module are done. */
180 de->pde_unload_completion = &c;
181 if (atomic_add_return(BIAS, &de->in_use) != BIAS)
182 wait_for_completion(&c);
184 /* ->pde_openers list can't grow from now on. */
186 spin_lock(&de->pde_unload_lock);
187 while (!list_empty(&de->pde_openers)) {
188 struct pde_opener *pdeo;
189 pdeo = list_first_entry(&de->pde_openers, struct pde_opener, lh);
190 close_pdeo(de, pdeo);
192 spin_unlock(&de->pde_unload_lock);
195 static loff_t proc_reg_llseek(struct file *file, loff_t offset, int whence)
197 struct proc_dir_entry *pde = PDE(file_inode(file));
200 loff_t (*llseek)(struct file *, loff_t, int);
201 llseek = pde->proc_fops->llseek;
203 llseek = default_llseek;
204 rv = llseek(file, offset, whence);
210 static ssize_t proc_reg_read(struct file *file, char __user *buf, size_t count, loff_t *ppos)
212 ssize_t (*read)(struct file *, char __user *, size_t, loff_t *);
213 struct proc_dir_entry *pde = PDE(file_inode(file));
216 read = pde->proc_fops->read;
218 rv = read(file, buf, count, ppos);
224 static ssize_t proc_reg_write(struct file *file, const char __user *buf, size_t count, loff_t *ppos)
226 ssize_t (*write)(struct file *, const char __user *, size_t, loff_t *);
227 struct proc_dir_entry *pde = PDE(file_inode(file));
230 write = pde->proc_fops->write;
232 rv = write(file, buf, count, ppos);
238 static __poll_t proc_reg_poll(struct file *file, struct poll_table_struct *pts)
240 struct proc_dir_entry *pde = PDE(file_inode(file));
241 __poll_t rv = DEFAULT_POLLMASK;
242 __poll_t (*poll)(struct file *, struct poll_table_struct *);
244 poll = pde->proc_fops->poll;
246 rv = poll(file, pts);
252 static long proc_reg_unlocked_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
254 struct proc_dir_entry *pde = PDE(file_inode(file));
256 long (*ioctl)(struct file *, unsigned int, unsigned long);
258 ioctl = pde->proc_fops->unlocked_ioctl;
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_inode(file));
271 long (*compat_ioctl)(struct file *, unsigned int, unsigned long);
273 compat_ioctl = pde->proc_fops->compat_ioctl;
275 rv = compat_ioctl(file, cmd, arg);
282 static int proc_reg_mmap(struct file *file, struct vm_area_struct *vma)
284 struct proc_dir_entry *pde = PDE(file_inode(file));
286 int (*mmap)(struct file *, struct vm_area_struct *);
288 mmap = pde->proc_fops->mmap;
290 rv = mmap(file, vma);
297 proc_reg_get_unmapped_area(struct file *file, unsigned long orig_addr,
298 unsigned long len, unsigned long pgoff,
301 struct proc_dir_entry *pde = PDE(file_inode(file));
302 unsigned long rv = -EIO;
305 typeof(proc_reg_get_unmapped_area) *get_area;
307 get_area = pde->proc_fops->get_unmapped_area;
310 get_area = current->mm->get_unmapped_area;
314 rv = get_area(file, orig_addr, len, pgoff, flags);
322 static int proc_reg_open(struct inode *inode, struct file *file)
324 struct proc_dir_entry *pde = PDE(inode);
326 int (*open)(struct inode *, struct file *);
327 int (*release)(struct inode *, struct file *);
328 struct pde_opener *pdeo;
332 * 1) PDE's ->release hook will be called no matter what
333 * either normally by close()/->release, or forcefully by
334 * rmmod/remove_proc_entry.
336 * 2) rmmod isn't blocked by opening file in /proc and sitting on
337 * the descriptor (including "rmmod foo </proc/foo" scenario).
339 * Save every "struct file" with custom ->release hook.
341 pdeo = kmalloc(sizeof(struct pde_opener), GFP_KERNEL);
349 open = pde->proc_fops->open;
350 release = pde->proc_fops->release;
353 rv = open(inode, file);
355 if (rv == 0 && release) {
356 /* To know what to release. */
358 pdeo->closing = false;
360 spin_lock(&pde->pde_unload_lock);
361 list_add(&pdeo->lh, &pde->pde_openers);
362 spin_unlock(&pde->pde_unload_lock);
370 static int proc_reg_release(struct inode *inode, struct file *file)
372 struct proc_dir_entry *pde = PDE(inode);
373 struct pde_opener *pdeo;
374 spin_lock(&pde->pde_unload_lock);
375 list_for_each_entry(pdeo, &pde->pde_openers, lh) {
376 if (pdeo->file == file) {
377 close_pdeo(pde, pdeo);
381 spin_unlock(&pde->pde_unload_lock);
385 static const struct file_operations proc_reg_file_ops = {
386 .llseek = proc_reg_llseek,
387 .read = proc_reg_read,
388 .write = proc_reg_write,
389 .poll = proc_reg_poll,
390 .unlocked_ioctl = proc_reg_unlocked_ioctl,
392 .compat_ioctl = proc_reg_compat_ioctl,
394 .mmap = proc_reg_mmap,
395 .get_unmapped_area = proc_reg_get_unmapped_area,
396 .open = proc_reg_open,
397 .release = proc_reg_release,
401 static const struct file_operations proc_reg_file_ops_no_compat = {
402 .llseek = proc_reg_llseek,
403 .read = proc_reg_read,
404 .write = proc_reg_write,
405 .poll = proc_reg_poll,
406 .unlocked_ioctl = proc_reg_unlocked_ioctl,
407 .mmap = proc_reg_mmap,
408 .get_unmapped_area = proc_reg_get_unmapped_area,
409 .open = proc_reg_open,
410 .release = proc_reg_release,
414 static void proc_put_link(void *p)
419 static const char *proc_get_link(struct dentry *dentry,
421 struct delayed_call *done)
423 struct proc_dir_entry *pde = PDE(inode);
425 return ERR_PTR(-EINVAL);
426 set_delayed_call(done, proc_put_link, pde);
430 const struct inode_operations proc_link_inode_operations = {
431 .get_link = proc_get_link,
434 struct inode *proc_get_inode(struct super_block *sb, struct proc_dir_entry *de)
436 struct inode *inode = new_inode_pseudo(sb);
439 inode->i_ino = de->low_ino;
440 inode->i_mtime = inode->i_atime = inode->i_ctime = current_time(inode);
441 PROC_I(inode)->pde = de;
443 if (is_empty_pde(de)) {
444 make_empty_dir_inode(inode);
448 inode->i_mode = de->mode;
449 inode->i_uid = de->uid;
450 inode->i_gid = de->gid;
453 inode->i_size = de->size;
455 set_nlink(inode, de->nlink);
456 WARN_ON(!de->proc_iops);
457 inode->i_op = de->proc_iops;
459 if (S_ISREG(inode->i_mode)) {
461 if (!de->proc_fops->compat_ioctl)
463 &proc_reg_file_ops_no_compat;
466 inode->i_fop = &proc_reg_file_ops;
468 inode->i_fop = de->proc_fops;
476 int proc_fill_super(struct super_block *s, void *data, int silent)
478 struct pid_namespace *ns = get_pid_ns(s->s_fs_info);
479 struct inode *root_inode;
482 if (!proc_parse_options(data, ns))
485 /* User space would break if executables or devices appear on proc */
486 s->s_iflags |= SB_I_USERNS_VISIBLE | SB_I_NOEXEC | SB_I_NODEV;
487 s->s_flags |= SB_NODIRATIME | SB_NOSUID | SB_NOEXEC;
488 s->s_blocksize = 1024;
489 s->s_blocksize_bits = 10;
490 s->s_magic = PROC_SUPER_MAGIC;
491 s->s_op = &proc_sops;
495 * procfs isn't actually a stacking filesystem; however, there is
496 * too much magic going on inside it to permit stacking things on
499 s->s_stack_depth = FILESYSTEM_MAX_STACK_DEPTH;
502 root_inode = proc_get_inode(s, &proc_root);
504 pr_err("proc_fill_super: get root inode failed\n");
508 s->s_root = d_make_root(root_inode);
510 pr_err("proc_fill_super: allocate dentry failed\n");
514 ret = proc_setup_self(s);
518 return proc_setup_thread_self(s);