Loading...
1/*
2 * linux/fs/proc/inode.c
3 *
4 * Copyright (C) 1991, 1992 Linus Torvalds
5 */
6
7#include <linux/time.h>
8#include <linux/proc_fs.h>
9#include <linux/kernel.h>
10#include <linux/mm.h>
11#include <linux/string.h>
12#include <linux/stat.h>
13#include <linux/completion.h>
14#include <linux/poll.h>
15#include <linux/file.h>
16#include <linux/limits.h>
17#include <linux/init.h>
18#include <linux/module.h>
19#include <linux/sysctl.h>
20#include <linux/slab.h>
21
22#include <asm/system.h>
23#include <asm/uaccess.h>
24
25#include "internal.h"
26
27static void proc_evict_inode(struct inode *inode)
28{
29 struct proc_dir_entry *de;
30 struct ctl_table_header *head;
31 const struct proc_ns_operations *ns_ops;
32
33 truncate_inode_pages(&inode->i_data, 0);
34 end_writeback(inode);
35
36 /* Stop tracking associated processes */
37 put_pid(PROC_I(inode)->pid);
38
39 /* Let go of any associated proc directory entry */
40 de = PROC_I(inode)->pde;
41 if (de)
42 pde_put(de);
43 head = PROC_I(inode)->sysctl;
44 if (head) {
45 rcu_assign_pointer(PROC_I(inode)->sysctl, NULL);
46 sysctl_head_put(head);
47 }
48 /* Release any associated namespace */
49 ns_ops = PROC_I(inode)->ns_ops;
50 if (ns_ops && ns_ops->put)
51 ns_ops->put(PROC_I(inode)->ns);
52}
53
54static struct kmem_cache * proc_inode_cachep;
55
56static struct inode *proc_alloc_inode(struct super_block *sb)
57{
58 struct proc_inode *ei;
59 struct inode *inode;
60
61 ei = (struct proc_inode *)kmem_cache_alloc(proc_inode_cachep, GFP_KERNEL);
62 if (!ei)
63 return NULL;
64 ei->pid = NULL;
65 ei->fd = 0;
66 ei->op.proc_get_link = NULL;
67 ei->pde = NULL;
68 ei->sysctl = NULL;
69 ei->sysctl_entry = NULL;
70 ei->ns = NULL;
71 ei->ns_ops = NULL;
72 inode = &ei->vfs_inode;
73 inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
74 return inode;
75}
76
77static void proc_i_callback(struct rcu_head *head)
78{
79 struct inode *inode = container_of(head, struct inode, i_rcu);
80 INIT_LIST_HEAD(&inode->i_dentry);
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_inodecache(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_PANIC),
102 init_once);
103}
104
105static const struct super_operations proc_sops = {
106 .alloc_inode = proc_alloc_inode,
107 .destroy_inode = proc_destroy_inode,
108 .drop_inode = generic_delete_inode,
109 .evict_inode = proc_evict_inode,
110 .statfs = simple_statfs,
111};
112
113static void __pde_users_dec(struct proc_dir_entry *pde)
114{
115 pde->pde_users--;
116 if (pde->pde_unload_completion && pde->pde_users == 0)
117 complete(pde->pde_unload_completion);
118}
119
120void pde_users_dec(struct proc_dir_entry *pde)
121{
122 spin_lock(&pde->pde_unload_lock);
123 __pde_users_dec(pde);
124 spin_unlock(&pde->pde_unload_lock);
125}
126
127static loff_t proc_reg_llseek(struct file *file, loff_t offset, int whence)
128{
129 struct proc_dir_entry *pde = PDE(file->f_path.dentry->d_inode);
130 loff_t rv = -EINVAL;
131 loff_t (*llseek)(struct file *, loff_t, int);
132
133 spin_lock(&pde->pde_unload_lock);
134 /*
135 * remove_proc_entry() is going to delete PDE (as part of module
136 * cleanup sequence). No new callers into module allowed.
137 */
138 if (!pde->proc_fops) {
139 spin_unlock(&pde->pde_unload_lock);
140 return rv;
141 }
142 /*
143 * Bump refcount so that remove_proc_entry will wail for ->llseek to
144 * complete.
145 */
146 pde->pde_users++;
147 /*
148 * Save function pointer under lock, to protect against ->proc_fops
149 * NULL'ifying right after ->pde_unload_lock is dropped.
150 */
151 llseek = pde->proc_fops->llseek;
152 spin_unlock(&pde->pde_unload_lock);
153
154 if (!llseek)
155 llseek = default_llseek;
156 rv = llseek(file, offset, whence);
157
158 pde_users_dec(pde);
159 return rv;
160}
161
162static ssize_t proc_reg_read(struct file *file, char __user *buf, size_t count, loff_t *ppos)
163{
164 struct proc_dir_entry *pde = PDE(file->f_path.dentry->d_inode);
165 ssize_t rv = -EIO;
166 ssize_t (*read)(struct file *, char __user *, size_t, loff_t *);
167
168 spin_lock(&pde->pde_unload_lock);
169 if (!pde->proc_fops) {
170 spin_unlock(&pde->pde_unload_lock);
171 return rv;
172 }
173 pde->pde_users++;
174 read = pde->proc_fops->read;
175 spin_unlock(&pde->pde_unload_lock);
176
177 if (read)
178 rv = read(file, buf, count, ppos);
179
180 pde_users_dec(pde);
181 return rv;
182}
183
184static ssize_t proc_reg_write(struct file *file, const char __user *buf, size_t count, loff_t *ppos)
185{
186 struct proc_dir_entry *pde = PDE(file->f_path.dentry->d_inode);
187 ssize_t rv = -EIO;
188 ssize_t (*write)(struct file *, const char __user *, size_t, loff_t *);
189
190 spin_lock(&pde->pde_unload_lock);
191 if (!pde->proc_fops) {
192 spin_unlock(&pde->pde_unload_lock);
193 return rv;
194 }
195 pde->pde_users++;
196 write = pde->proc_fops->write;
197 spin_unlock(&pde->pde_unload_lock);
198
199 if (write)
200 rv = write(file, buf, count, ppos);
201
202 pde_users_dec(pde);
203 return rv;
204}
205
206static unsigned int proc_reg_poll(struct file *file, struct poll_table_struct *pts)
207{
208 struct proc_dir_entry *pde = PDE(file->f_path.dentry->d_inode);
209 unsigned int rv = DEFAULT_POLLMASK;
210 unsigned int (*poll)(struct file *, struct poll_table_struct *);
211
212 spin_lock(&pde->pde_unload_lock);
213 if (!pde->proc_fops) {
214 spin_unlock(&pde->pde_unload_lock);
215 return rv;
216 }
217 pde->pde_users++;
218 poll = pde->proc_fops->poll;
219 spin_unlock(&pde->pde_unload_lock);
220
221 if (poll)
222 rv = poll(file, pts);
223
224 pde_users_dec(pde);
225 return rv;
226}
227
228static long proc_reg_unlocked_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
229{
230 struct proc_dir_entry *pde = PDE(file->f_path.dentry->d_inode);
231 long rv = -ENOTTY;
232 long (*ioctl)(struct file *, unsigned int, unsigned long);
233
234 spin_lock(&pde->pde_unload_lock);
235 if (!pde->proc_fops) {
236 spin_unlock(&pde->pde_unload_lock);
237 return rv;
238 }
239 pde->pde_users++;
240 ioctl = pde->proc_fops->unlocked_ioctl;
241 spin_unlock(&pde->pde_unload_lock);
242
243 if (ioctl)
244 rv = ioctl(file, cmd, arg);
245
246 pde_users_dec(pde);
247 return rv;
248}
249
250#ifdef CONFIG_COMPAT
251static long proc_reg_compat_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
252{
253 struct proc_dir_entry *pde = PDE(file->f_path.dentry->d_inode);
254 long rv = -ENOTTY;
255 long (*compat_ioctl)(struct file *, unsigned int, unsigned long);
256
257 spin_lock(&pde->pde_unload_lock);
258 if (!pde->proc_fops) {
259 spin_unlock(&pde->pde_unload_lock);
260 return rv;
261 }
262 pde->pde_users++;
263 compat_ioctl = pde->proc_fops->compat_ioctl;
264 spin_unlock(&pde->pde_unload_lock);
265
266 if (compat_ioctl)
267 rv = compat_ioctl(file, cmd, arg);
268
269 pde_users_dec(pde);
270 return rv;
271}
272#endif
273
274static int proc_reg_mmap(struct file *file, struct vm_area_struct *vma)
275{
276 struct proc_dir_entry *pde = PDE(file->f_path.dentry->d_inode);
277 int rv = -EIO;
278 int (*mmap)(struct file *, struct vm_area_struct *);
279
280 spin_lock(&pde->pde_unload_lock);
281 if (!pde->proc_fops) {
282 spin_unlock(&pde->pde_unload_lock);
283 return rv;
284 }
285 pde->pde_users++;
286 mmap = pde->proc_fops->mmap;
287 spin_unlock(&pde->pde_unload_lock);
288
289 if (mmap)
290 rv = mmap(file, vma);
291
292 pde_users_dec(pde);
293 return rv;
294}
295
296static int proc_reg_open(struct inode *inode, struct file *file)
297{
298 struct proc_dir_entry *pde = PDE(inode);
299 int rv = 0;
300 int (*open)(struct inode *, struct file *);
301 int (*release)(struct inode *, struct file *);
302 struct pde_opener *pdeo;
303
304 /*
305 * What for, you ask? Well, we can have open, rmmod, remove_proc_entry
306 * sequence. ->release won't be called because ->proc_fops will be
307 * cleared. Depending on complexity of ->release, consequences vary.
308 *
309 * We can't wait for mercy when close will be done for real, it's
310 * deadlockable: rmmod foo </proc/foo . So, we're going to do ->release
311 * by hand in remove_proc_entry(). For this, save opener's credentials
312 * for later.
313 */
314 pdeo = kmalloc(sizeof(struct pde_opener), GFP_KERNEL);
315 if (!pdeo)
316 return -ENOMEM;
317
318 spin_lock(&pde->pde_unload_lock);
319 if (!pde->proc_fops) {
320 spin_unlock(&pde->pde_unload_lock);
321 kfree(pdeo);
322 return -ENOENT;
323 }
324 pde->pde_users++;
325 open = pde->proc_fops->open;
326 release = pde->proc_fops->release;
327 spin_unlock(&pde->pde_unload_lock);
328
329 if (open)
330 rv = open(inode, file);
331
332 spin_lock(&pde->pde_unload_lock);
333 if (rv == 0 && release) {
334 /* To know what to release. */
335 pdeo->inode = inode;
336 pdeo->file = file;
337 /* Strictly for "too late" ->release in proc_reg_release(). */
338 pdeo->release = release;
339 list_add(&pdeo->lh, &pde->pde_openers);
340 } else
341 kfree(pdeo);
342 __pde_users_dec(pde);
343 spin_unlock(&pde->pde_unload_lock);
344 return rv;
345}
346
347static struct pde_opener *find_pde_opener(struct proc_dir_entry *pde,
348 struct inode *inode, struct file *file)
349{
350 struct pde_opener *pdeo;
351
352 list_for_each_entry(pdeo, &pde->pde_openers, lh) {
353 if (pdeo->inode == inode && pdeo->file == file)
354 return pdeo;
355 }
356 return NULL;
357}
358
359static int proc_reg_release(struct inode *inode, struct file *file)
360{
361 struct proc_dir_entry *pde = PDE(inode);
362 int rv = 0;
363 int (*release)(struct inode *, struct file *);
364 struct pde_opener *pdeo;
365
366 spin_lock(&pde->pde_unload_lock);
367 pdeo = find_pde_opener(pde, inode, file);
368 if (!pde->proc_fops) {
369 /*
370 * Can't simply exit, __fput() will think that everything is OK,
371 * and move on to freeing struct file. remove_proc_entry() will
372 * find slacker in opener's list and will try to do non-trivial
373 * things with struct file. Therefore, remove opener from list.
374 *
375 * But if opener is removed from list, who will ->release it?
376 */
377 if (pdeo) {
378 list_del(&pdeo->lh);
379 spin_unlock(&pde->pde_unload_lock);
380 rv = pdeo->release(inode, file);
381 kfree(pdeo);
382 } else
383 spin_unlock(&pde->pde_unload_lock);
384 return rv;
385 }
386 pde->pde_users++;
387 release = pde->proc_fops->release;
388 if (pdeo) {
389 list_del(&pdeo->lh);
390 kfree(pdeo);
391 }
392 spin_unlock(&pde->pde_unload_lock);
393
394 if (release)
395 rv = release(inode, file);
396
397 pde_users_dec(pde);
398 return rv;
399}
400
401static const struct file_operations proc_reg_file_ops = {
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#ifdef CONFIG_COMPAT
408 .compat_ioctl = proc_reg_compat_ioctl,
409#endif
410 .mmap = proc_reg_mmap,
411 .open = proc_reg_open,
412 .release = proc_reg_release,
413};
414
415#ifdef CONFIG_COMPAT
416static const struct file_operations proc_reg_file_ops_no_compat = {
417 .llseek = proc_reg_llseek,
418 .read = proc_reg_read,
419 .write = proc_reg_write,
420 .poll = proc_reg_poll,
421 .unlocked_ioctl = proc_reg_unlocked_ioctl,
422 .mmap = proc_reg_mmap,
423 .open = proc_reg_open,
424 .release = proc_reg_release,
425};
426#endif
427
428struct inode *proc_get_inode(struct super_block *sb, struct proc_dir_entry *de)
429{
430 struct inode * inode;
431
432 inode = iget_locked(sb, de->low_ino);
433 if (!inode)
434 return NULL;
435 if (inode->i_state & I_NEW) {
436 inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
437 PROC_I(inode)->fd = 0;
438 PROC_I(inode)->pde = de;
439
440 if (de->mode) {
441 inode->i_mode = de->mode;
442 inode->i_uid = de->uid;
443 inode->i_gid = de->gid;
444 }
445 if (de->size)
446 inode->i_size = de->size;
447 if (de->nlink)
448 inode->i_nlink = de->nlink;
449 if (de->proc_iops)
450 inode->i_op = de->proc_iops;
451 if (de->proc_fops) {
452 if (S_ISREG(inode->i_mode)) {
453#ifdef CONFIG_COMPAT
454 if (!de->proc_fops->compat_ioctl)
455 inode->i_fop =
456 &proc_reg_file_ops_no_compat;
457 else
458#endif
459 inode->i_fop = &proc_reg_file_ops;
460 } else {
461 inode->i_fop = de->proc_fops;
462 }
463 }
464 unlock_new_inode(inode);
465 } else
466 pde_put(de);
467 return inode;
468}
469
470int proc_fill_super(struct super_block *s)
471{
472 struct inode * root_inode;
473
474 s->s_flags |= MS_NODIRATIME | MS_NOSUID | MS_NOEXEC;
475 s->s_blocksize = 1024;
476 s->s_blocksize_bits = 10;
477 s->s_magic = PROC_SUPER_MAGIC;
478 s->s_op = &proc_sops;
479 s->s_time_gran = 1;
480
481 pde_get(&proc_root);
482 root_inode = proc_get_inode(s, &proc_root);
483 if (!root_inode)
484 goto out_no_root;
485 root_inode->i_uid = 0;
486 root_inode->i_gid = 0;
487 s->s_root = d_alloc_root(root_inode);
488 if (!s->s_root)
489 goto out_no_root;
490 return 0;
491
492out_no_root:
493 printk("proc_read_super: get root inode failed\n");
494 iput(root_inode);
495 pde_put(&proc_root);
496 return -ENOMEM;
497}
1/*
2 * linux/fs/proc/inode.c
3 *
4 * Copyright (C) 1991, 1992 Linus Torvalds
5 */
6
7#include <linux/time.h>
8#include <linux/proc_fs.h>
9#include <linux/kernel.h>
10#include <linux/pid_namespace.h>
11#include <linux/mm.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>
26
27#include <linux/uaccess.h>
28
29#include "internal.h"
30
31static void proc_evict_inode(struct inode *inode)
32{
33 struct proc_dir_entry *de;
34 struct ctl_table_header *head;
35
36 truncate_inode_pages_final(&inode->i_data);
37 clear_inode(inode);
38
39 /* Stop tracking associated processes */
40 put_pid(PROC_I(inode)->pid);
41
42 /* Let go of any associated proc directory entry */
43 de = PDE(inode);
44 if (de)
45 pde_put(de);
46 head = PROC_I(inode)->sysctl;
47 if (head) {
48 RCU_INIT_POINTER(PROC_I(inode)->sysctl, NULL);
49 sysctl_head_put(head);
50 }
51}
52
53static struct kmem_cache * proc_inode_cachep;
54
55static struct inode *proc_alloc_inode(struct super_block *sb)
56{
57 struct proc_inode *ei;
58 struct inode *inode;
59
60 ei = (struct proc_inode *)kmem_cache_alloc(proc_inode_cachep, GFP_KERNEL);
61 if (!ei)
62 return NULL;
63 ei->pid = NULL;
64 ei->fd = 0;
65 ei->op.proc_get_link = NULL;
66 ei->pde = NULL;
67 ei->sysctl = NULL;
68 ei->sysctl_entry = NULL;
69 ei->ns_ops = NULL;
70 inode = &ei->vfs_inode;
71 return inode;
72}
73
74static void proc_i_callback(struct rcu_head *head)
75{
76 struct inode *inode = container_of(head, struct inode, i_rcu);
77 kmem_cache_free(proc_inode_cachep, PROC_I(inode));
78}
79
80static void proc_destroy_inode(struct inode *inode)
81{
82 call_rcu(&inode->i_rcu, proc_i_callback);
83}
84
85static void init_once(void *foo)
86{
87 struct proc_inode *ei = (struct proc_inode *) foo;
88
89 inode_init_once(&ei->vfs_inode);
90}
91
92void __init proc_init_inodecache(void)
93{
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|
98 SLAB_PANIC),
99 init_once);
100}
101
102static int proc_show_options(struct seq_file *seq, struct dentry *root)
103{
104 struct super_block *sb = root->d_sb;
105 struct pid_namespace *pid = sb->s_fs_info;
106
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);
111
112 return 0;
113}
114
115static 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,
123};
124
125enum {BIAS = -1U<<31};
126
127static inline int use_pde(struct proc_dir_entry *pde)
128{
129 return atomic_inc_unless_negative(&pde->in_use);
130}
131
132static void unuse_pde(struct proc_dir_entry *pde)
133{
134 if (atomic_dec_return(&pde->in_use) == BIAS)
135 complete(pde->pde_unload_completion);
136}
137
138/* pde is locked */
139static void close_pdeo(struct proc_dir_entry *pde, struct pde_opener *pdeo)
140{
141 /*
142 * close() (proc_reg_release()) can't delete an entry and proceed:
143 * ->release hook needs to be available at the right moment.
144 *
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.
147 *
148 * Therefore, first process to enter this function does ->release() and
149 * signals its completion to the other process which does nothing.
150 */
151 if (pdeo->closing) {
152 /* somebody else is doing that, just wait */
153 DECLARE_COMPLETION_ONSTACK(c);
154 pdeo->c = &c;
155 spin_unlock(&pde->pde_unload_lock);
156 wait_for_completion(&c);
157 spin_lock(&pde->pde_unload_lock);
158 } else {
159 struct file *file;
160 pdeo->closing = true;
161 spin_unlock(&pde->pde_unload_lock);
162 file = pdeo->file;
163 pde->proc_fops->release(file_inode(file), file);
164 spin_lock(&pde->pde_unload_lock);
165 /* After ->release. */
166 list_del(&pdeo->lh);
167 if (pdeo->c)
168 complete(pdeo->c);
169 kfree(pdeo);
170 }
171}
172
173void proc_entry_rundown(struct proc_dir_entry *de)
174{
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);
180
181 /* ->pde_openers list can't grow from now on. */
182
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);
188 }
189 spin_unlock(&de->pde_unload_lock);
190}
191
192static loff_t proc_reg_llseek(struct file *file, loff_t offset, int whence)
193{
194 struct proc_dir_entry *pde = PDE(file_inode(file));
195 loff_t rv = -EINVAL;
196 if (use_pde(pde)) {
197 loff_t (*llseek)(struct file *, loff_t, int);
198 llseek = pde->proc_fops->llseek;
199 if (!llseek)
200 llseek = default_llseek;
201 rv = llseek(file, offset, whence);
202 unuse_pde(pde);
203 }
204 return rv;
205}
206
207static ssize_t proc_reg_read(struct file *file, char __user *buf, size_t count, loff_t *ppos)
208{
209 ssize_t (*read)(struct file *, char __user *, size_t, loff_t *);
210 struct proc_dir_entry *pde = PDE(file_inode(file));
211 ssize_t rv = -EIO;
212 if (use_pde(pde)) {
213 read = pde->proc_fops->read;
214 if (read)
215 rv = read(file, buf, count, ppos);
216 unuse_pde(pde);
217 }
218 return rv;
219}
220
221static ssize_t proc_reg_write(struct file *file, const char __user *buf, size_t count, loff_t *ppos)
222{
223 ssize_t (*write)(struct file *, const char __user *, size_t, loff_t *);
224 struct proc_dir_entry *pde = PDE(file_inode(file));
225 ssize_t rv = -EIO;
226 if (use_pde(pde)) {
227 write = pde->proc_fops->write;
228 if (write)
229 rv = write(file, buf, count, ppos);
230 unuse_pde(pde);
231 }
232 return rv;
233}
234
235static unsigned int proc_reg_poll(struct file *file, struct poll_table_struct *pts)
236{
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 *);
240 if (use_pde(pde)) {
241 poll = pde->proc_fops->poll;
242 if (poll)
243 rv = poll(file, pts);
244 unuse_pde(pde);
245 }
246 return rv;
247}
248
249static long proc_reg_unlocked_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
250{
251 struct proc_dir_entry *pde = PDE(file_inode(file));
252 long rv = -ENOTTY;
253 long (*ioctl)(struct file *, unsigned int, unsigned long);
254 if (use_pde(pde)) {
255 ioctl = pde->proc_fops->unlocked_ioctl;
256 if (ioctl)
257 rv = ioctl(file, cmd, arg);
258 unuse_pde(pde);
259 }
260 return rv;
261}
262
263#ifdef CONFIG_COMPAT
264static long proc_reg_compat_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 (*compat_ioctl)(struct file *, unsigned int, unsigned long);
269 if (use_pde(pde)) {
270 compat_ioctl = pde->proc_fops->compat_ioctl;
271 if (compat_ioctl)
272 rv = compat_ioctl(file, cmd, arg);
273 unuse_pde(pde);
274 }
275 return rv;
276}
277#endif
278
279static int proc_reg_mmap(struct file *file, struct vm_area_struct *vma)
280{
281 struct proc_dir_entry *pde = PDE(file_inode(file));
282 int rv = -EIO;
283 int (*mmap)(struct file *, struct vm_area_struct *);
284 if (use_pde(pde)) {
285 mmap = pde->proc_fops->mmap;
286 if (mmap)
287 rv = mmap(file, vma);
288 unuse_pde(pde);
289 }
290 return rv;
291}
292
293static unsigned long
294proc_reg_get_unmapped_area(struct file *file, unsigned long orig_addr,
295 unsigned long len, unsigned long pgoff,
296 unsigned long flags)
297{
298 struct proc_dir_entry *pde = PDE(file_inode(file));
299 unsigned long rv = -EIO;
300
301 if (use_pde(pde)) {
302 typeof(proc_reg_get_unmapped_area) *get_area;
303
304 get_area = pde->proc_fops->get_unmapped_area;
305#ifdef CONFIG_MMU
306 if (!get_area)
307 get_area = current->mm->get_unmapped_area;
308#endif
309
310 if (get_area)
311 rv = get_area(file, orig_addr, len, pgoff, flags);
312 else
313 rv = orig_addr;
314 unuse_pde(pde);
315 }
316 return rv;
317}
318
319static int proc_reg_open(struct inode *inode, struct file *file)
320{
321 struct proc_dir_entry *pde = PDE(inode);
322 int rv = 0;
323 int (*open)(struct inode *, struct file *);
324 int (*release)(struct inode *, struct file *);
325 struct pde_opener *pdeo;
326
327 /*
328 * Ensure that
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.
332 *
333 * 2) rmmod isn't blocked by opening file in /proc and sitting on
334 * the descriptor (including "rmmod foo </proc/foo" scenario).
335 *
336 * Save every "struct file" with custom ->release hook.
337 */
338 pdeo = kmalloc(sizeof(struct pde_opener), GFP_KERNEL);
339 if (!pdeo)
340 return -ENOMEM;
341
342 if (!use_pde(pde)) {
343 kfree(pdeo);
344 return -ENOENT;
345 }
346 open = pde->proc_fops->open;
347 release = pde->proc_fops->release;
348
349 if (open)
350 rv = open(inode, file);
351
352 if (rv == 0 && release) {
353 /* To know what to release. */
354 pdeo->file = file;
355 pdeo->closing = false;
356 pdeo->c = NULL;
357 spin_lock(&pde->pde_unload_lock);
358 list_add(&pdeo->lh, &pde->pde_openers);
359 spin_unlock(&pde->pde_unload_lock);
360 } else
361 kfree(pdeo);
362
363 unuse_pde(pde);
364 return rv;
365}
366
367static int proc_reg_release(struct inode *inode, struct file *file)
368{
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);
375 break;
376 }
377 }
378 spin_unlock(&pde->pde_unload_lock);
379 return 0;
380}
381
382static 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,
388#ifdef CONFIG_COMPAT
389 .compat_ioctl = proc_reg_compat_ioctl,
390#endif
391 .mmap = proc_reg_mmap,
392 .get_unmapped_area = proc_reg_get_unmapped_area,
393 .open = proc_reg_open,
394 .release = proc_reg_release,
395};
396
397#ifdef CONFIG_COMPAT
398static 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,
408};
409#endif
410
411static void proc_put_link(void *p)
412{
413 unuse_pde(p);
414}
415
416static const char *proc_get_link(struct dentry *dentry,
417 struct inode *inode,
418 struct delayed_call *done)
419{
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);
424 return pde->data;
425}
426
427const struct inode_operations proc_link_inode_operations = {
428 .get_link = proc_get_link,
429};
430
431struct inode *proc_get_inode(struct super_block *sb, struct proc_dir_entry *de)
432{
433 struct inode *inode = new_inode_pseudo(sb);
434
435 if (inode) {
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;
439
440 if (is_empty_pde(de)) {
441 make_empty_dir_inode(inode);
442 return inode;
443 }
444 if (de->mode) {
445 inode->i_mode = de->mode;
446 inode->i_uid = de->uid;
447 inode->i_gid = de->gid;
448 }
449 if (de->size)
450 inode->i_size = de->size;
451 if (de->nlink)
452 set_nlink(inode, de->nlink);
453 WARN_ON(!de->proc_iops);
454 inode->i_op = de->proc_iops;
455 if (de->proc_fops) {
456 if (S_ISREG(inode->i_mode)) {
457#ifdef CONFIG_COMPAT
458 if (!de->proc_fops->compat_ioctl)
459 inode->i_fop =
460 &proc_reg_file_ops_no_compat;
461 else
462#endif
463 inode->i_fop = &proc_reg_file_ops;
464 } else {
465 inode->i_fop = de->proc_fops;
466 }
467 }
468 } else
469 pde_put(de);
470 return inode;
471}
472
473int proc_fill_super(struct super_block *s, void *data, int silent)
474{
475 struct pid_namespace *ns = get_pid_ns(s->s_fs_info);
476 struct inode *root_inode;
477 int ret;
478
479 if (!proc_parse_options(data, ns))
480 return -EINVAL;
481
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;
489 s->s_time_gran = 1;
490
491 /*
492 * procfs isn't actually a stacking filesystem; however, there is
493 * too much magic going on inside it to permit stacking things on
494 * top of it
495 */
496 s->s_stack_depth = FILESYSTEM_MAX_STACK_DEPTH;
497
498 pde_get(&proc_root);
499 root_inode = proc_get_inode(s, &proc_root);
500 if (!root_inode) {
501 pr_err("proc_fill_super: get root inode failed\n");
502 return -ENOMEM;
503 }
504
505 s->s_root = d_make_root(root_inode);
506 if (!s->s_root) {
507 pr_err("proc_fill_super: allocate dentry failed\n");
508 return -ENOMEM;
509 }
510
511 ret = proc_setup_self(s);
512 if (ret) {
513 return ret;
514 }
515 return proc_setup_thread_self(s);
516}