fs/proc/inode.c

   1 // SPDX-License-Identifier: GPL-2.0
   2 /*
   3  *  linux/fs/proc/inode.c
   4  *
   5  *  Copyright (C) 1991, 1992  Linus Torvalds
   6  */
   7
   8 #include <linux/time.h>
   9 #include <linux/proc_fs.h>
  10 #include <linux/kernel.h>
  11 #include <linux/pid_namespace.h>
  12 #include <linux/mm.h>
  13 #include <linux/string.h>
  14 #include <linux/stat.h>
  15 #include <linux/completion.h>
  16 #include <linux/poll.h>
  17 #include <linux/printk.h>
  18 #include <linux/file.h>
  19 #include <linux/limits.h>
  20 #include <linux/init.h>
  21 #include <linux/module.h>
  22 #include <linux/sysctl.h>
  23 #include <linux/seq_file.h>
  24 #include <linux/slab.h>
  25 #include <linux/mount.h>
  26 #include <linux/magic.h>
  27
  28 #include <linux/uaccess.h>
  29
  30 #include "internal.h"
  31
  32 static void proc_evict_inode(struct inode *inode)
  33 {
  34         struct proc_dir_entry *de;
  35         struct ctl_table_header *head;
  36
  37         truncate_inode_pages_final(&inode->i_data);
  38         clear_inode(inode);
  39
  40         /* Stop tracking associated processes */
  41         put_pid(PROC_I(inode)->pid);
  42
  43         /* Let go of any associated proc directory entry */
  44         de = PDE(inode);
  45         if (de)
  46                 pde_put(de);
  47
  48         head = PROC_I(inode)->sysctl;
  49         if (head) {
  50                 RCU_INIT_POINTER(PROC_I(inode)->sysctl, NULL);
  51                 proc_sys_evict_inode(inode, head);
  52         }
  53 }
  54
  55 static struct kmem_cache * proc_inode_cachep;
  56
  57 static struct inode *proc_alloc_inode(struct super_block *sb)
  58 {
  59         struct proc_inode *ei;
  60         struct inode *inode;
  61
  62         ei = kmem_cache_alloc(proc_inode_cachep, GFP_KERNEL);
  63         if (!ei)
  64                 return NULL;
  65         ei->pid = NULL;
  66         ei->fd = 0;
  67         ei->op.proc_get_link = NULL;
  68         ei->pde = NULL;
  69         ei->sysctl = NULL;
  70         ei->sysctl_entry = NULL;
  71         ei->ns_ops = NULL;
  72         inode = &ei->vfs_inode;
  73         return inode;
  74 }
  75
  76 static void proc_i_callback(struct rcu_head *head)
  77 {
  78         struct inode *inode = container_of(head, struct inode, i_rcu);
  79         kmem_cache_free(proc_inode_cachep, PROC_I(inode));
  80 }
  81
  82 static void proc_destroy_inode(struct inode *inode)
  83 {
  84         call_rcu(&inode->i_rcu, proc_i_callback);
  85 }
  86
  87 static void init_once(void *foo)
  88 {
  89         struct proc_inode *ei = (struct proc_inode *) foo;
  90
  91         inode_init_once(&ei->vfs_inode);
  92 }
  93
  94 void __init proc_init_inodecache(void)
  95 {
  96         proc_inode_cachep = kmem_cache_create("proc_inode_cache",
  97                                              sizeof(struct proc_inode),
  98                                              0, (SLAB_RECLAIM_ACCOUNT|
  99                                                 SLAB_MEM_SPREAD|SLAB_ACCOUNT|
 100                                                 SLAB_PANIC),
 101                                              init_once);
 102 }
 103
 104 static int proc_show_options(struct seq_file *seq, struct dentry *root)
 105 {
 106         struct super_block *sb = root->d_sb;
 107         struct pid_namespace *pid = sb->s_fs_info;
 108
 109         if (!gid_eq(pid->pid_gid, GLOBAL_ROOT_GID))
 110                 seq_printf(seq, ",gid=%u", from_kgid_munged(&init_user_ns, pid->pid_gid));
 111         if (pid->hide_pid != HIDEPID_OFF)
 112                 seq_printf(seq, ",hidepid=%u", pid->hide_pid);
 113
 114         return 0;
 115 }
 116
 117 static const struct super_operations proc_sops = {
 118         .alloc_inode    = proc_alloc_inode,
 119         .destroy_inode  = proc_destroy_inode,
 120         .drop_inode     = generic_delete_inode,
 121         .evict_inode    = proc_evict_inode,
 122         .statfs         = simple_statfs,
 123         .remount_fs     = proc_remount,
 124         .show_options   = proc_show_options,
 125 };
 126
 127 enum {BIAS = -1U<<31};
 128
 129 static inline int use_pde(struct proc_dir_entry *pde)
 130 {
 131         return atomic_inc_unless_negative(&pde->in_use);
 132 }
 133
 134 static void unuse_pde(struct proc_dir_entry *pde)
 135 {
 136         if (atomic_dec_return(&pde->in_use) == BIAS)
 137                 complete(pde->pde_unload_completion);
 138 }
 139
 140 /* pde is locked */
 141 static void close_pdeo(struct proc_dir_entry *pde, struct pde_opener *pdeo)
 142 {
 143         /*
 144          * close() (proc_reg_release()) can't delete an entry and proceed:
 145          * ->release hook needs to be available at the right moment.
 146          *
 147          * rmmod (remove_proc_entry() et al) can't delete an entry and proceed:
 148          * "struct file" needs to be available at the right moment.
 149          *
 150          * Therefore, first process to enter this function does ->release() and
 151          * signals its completion to the other process which does nothing.
 152          */
 153         if (pdeo->closing) {
 154                 /* somebody else is doing that, just wait */
 155                 DECLARE_COMPLETION_ONSTACK(c);
 156                 pdeo->c = &c;
 157                 spin_unlock(&pde->pde_unload_lock);
 158                 wait_for_completion(&c);
 159                 spin_lock(&pde->pde_unload_lock);
 160         } else {
 161                 struct file *file;
 162                 pdeo->closing = true;
 163                 spin_unlock(&pde->pde_unload_lock);
 164                 file = pdeo->file;
 165                 pde->proc_fops->release(file_inode(file), file);
 166                 spin_lock(&pde->pde_unload_lock);
 167                 /* After ->release. */
 168                 list_del(&pdeo->lh);
 169                 if (pdeo->c)
 170                         complete(pdeo->c);
 171                 kfree(pdeo);
 172         }
 173 }
 174
 175 void proc_entry_rundown(struct proc_dir_entry *de)
 176 {
 177         DECLARE_COMPLETION_ONSTACK(c);
 178         /* Wait until all existing callers into module are done. */
 179         de->pde_unload_completion = &c;
 180         if (atomic_add_return(BIAS, &de->in_use) != BIAS)
 181                 wait_for_completion(&c);
 182
 183         /* ->pde_openers list can't grow from now on. */
 184
 185         spin_lock(&de->pde_unload_lock);
 186         while (!list_empty(&de->pde_openers)) {
 187                 struct pde_opener *pdeo;
 188                 pdeo = list_first_entry(&de->pde_openers, struct pde_opener, lh);
 189                 close_pdeo(de, pdeo);
 190         }
 191         spin_unlock(&de->pde_unload_lock);
 192 }
 193
 194 static loff_t proc_reg_llseek(struct file *file, loff_t offset, int whence)
 195 {
 196         struct proc_dir_entry *pde = PDE(file_inode(file));
 197         loff_t rv = -EINVAL;
 198         if (use_pde(pde)) {
 199                 loff_t (*llseek)(struct file *, loff_t, int);
 200                 llseek = pde->proc_fops->llseek;
 201                 if (!llseek)
 202                         llseek = default_llseek;
 203                 rv = llseek(file, offset, whence);
 204                 unuse_pde(pde);
 205         }
 206         return rv;
 207 }
 208
 209 static ssize_t proc_reg_read(struct file *file, char __user *buf, size_t count, loff_t *ppos)
 210 {
 211         ssize_t (*read)(struct file *, char __user *, size_t, loff_t *);
 212         struct proc_dir_entry *pde = PDE(file_inode(file));
 213         ssize_t rv = -EIO;
 214         if (use_pde(pde)) {
 215                 read = pde->proc_fops->read;
 216                 if (read)
 217                         rv = read(file, buf, count, ppos);
 218                 unuse_pde(pde);
 219         }
 220         return rv;
 221 }
 222
 223 static ssize_t proc_reg_write(struct file *file, const char __user *buf, size_t count, loff_t *ppos)
 224 {
 225         ssize_t (*write)(struct file *, const char __user *, size_t, loff_t *);
 226         struct proc_dir_entry *pde = PDE(file_inode(file));
 227         ssize_t rv = -EIO;
 228         if (use_pde(pde)) {
 229                 write = pde->proc_fops->write;
 230                 if (write)
 231                         rv = write(file, buf, count, ppos);
 232                 unuse_pde(pde);
 233         }
 234         return rv;
 235 }
 236
 237 static unsigned int proc_reg_poll(struct file *file, struct poll_table_struct *pts)
 238 {
 239         struct proc_dir_entry *pde = PDE(file_inode(file));
 240         unsigned int rv = DEFAULT_POLLMASK;
 241         unsigned int (*poll)(struct file *, struct poll_table_struct *);
 242         if (use_pde(pde)) {
 243                 poll = pde->proc_fops->poll;
 244                 if (poll)
 245                         rv = poll(file, pts);
 246                 unuse_pde(pde);
 247         }
 248         return rv;
 249 }
 250
 251 static long proc_reg_unlocked_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
 252 {
 253         struct proc_dir_entry *pde = PDE(file_inode(file));
 254         long rv = -ENOTTY;
 255         long (*ioctl)(struct file *, unsigned int, unsigned long);
 256         if (use_pde(pde)) {
 257                 ioctl = pde->proc_fops->unlocked_ioctl;
 258                 if (ioctl)
 259                         rv = ioctl(file, cmd, arg);
 260                 unuse_pde(pde);
 261         }
 262         return rv;
 263 }
 264
 265 #ifdef CONFIG_COMPAT
 266 static long proc_reg_compat_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
 267 {
 268         struct proc_dir_entry *pde = PDE(file_inode(file));
 269         long rv = -ENOTTY;
 270         long (*compat_ioctl)(struct file *, unsigned int, unsigned long);
 271         if (use_pde(pde)) {
 272                 compat_ioctl = pde->proc_fops->compat_ioctl;
 273                 if (compat_ioctl)
 274                         rv = compat_ioctl(file, cmd, arg);
 275                 unuse_pde(pde);
 276         }
 277         return rv;
 278 }
 279 #endif
 280
 281 static int proc_reg_mmap(struct file *file, struct vm_area_struct *vma)
 282 {
 283         struct proc_dir_entry *pde = PDE(file_inode(file));
 284         int rv = -EIO;
 285         int (*mmap)(struct file *, struct vm_area_struct *);
 286         if (use_pde(pde)) {
 287                 mmap = pde->proc_fops->mmap;
 288                 if (mmap)
 289                         rv = mmap(file, vma);
 290                 unuse_pde(pde);
 291         }
 292         return rv;
 293 }
 294
 295 static unsigned long
 296 proc_reg_get_unmapped_area(struct file *file, unsigned long orig_addr,
 297                            unsigned long len, unsigned long pgoff,
 298                            unsigned long flags)
 299 {
 300         struct proc_dir_entry *pde = PDE(file_inode(file));
 301         unsigned long rv = -EIO;
 302
 303         if (use_pde(pde)) {
 304                 typeof(proc_reg_get_unmapped_area) *get_area;
 305
 306                 get_area = pde->proc_fops->get_unmapped_area;
 307 #ifdef CONFIG_MMU
 308                 if (!get_area)
 309                         get_area = current->mm->get_unmapped_area;
 310 #endif
 311
 312                 if (get_area)
 313                         rv = get_area(file, orig_addr, len, pgoff, flags);
 314                 else
 315                         rv = orig_addr;
 316                 unuse_pde(pde);
 317         }
 318         return rv;
 319 }
 320
 321 static int proc_reg_open(struct inode *inode, struct file *file)
 322 {
 323         struct proc_dir_entry *pde = PDE(inode);
 324         int rv = 0;
 325         int (*open)(struct inode *, struct file *);
 326         int (*release)(struct inode *, struct file *);
 327         struct pde_opener *pdeo;
 328
 329         /*
 330          * Ensure that
 331          * 1) PDE's ->release hook will be called no matter what
 332          *    either normally by close()/->release, or forcefully by
 333          *    rmmod/remove_proc_entry.
 334          *
 335          * 2) rmmod isn't blocked by opening file in /proc and sitting on
 336          *    the descriptor (including "rmmod foo </proc/foo" scenario).
 337          *
 338          * Save every "struct file" with custom ->release hook.
 339          */
 340         pdeo = kmalloc(sizeof(struct pde_opener), GFP_KERNEL);
 341         if (!pdeo)
 342                 return -ENOMEM;
 343
 344         if (!use_pde(pde)) {
 345                 kfree(pdeo);
 346                 return -ENOENT;
 347         }
 348         open = pde->proc_fops->open;
 349         release = pde->proc_fops->release;
 350
 351         if (open)
 352                 rv = open(inode, file);
 353
 354         if (rv == 0 && release) {
 355                 /* To know what to release. */
 356                 pdeo->file = file;
 357                 pdeo->closing = false;
 358                 pdeo->c = NULL;
 359                 spin_lock(&pde->pde_unload_lock);
 360                 list_add(&pdeo->lh, &pde->pde_openers);
 361                 spin_unlock(&pde->pde_unload_lock);
 362         } else
 363                 kfree(pdeo);
 364
 365         unuse_pde(pde);
 366         return rv;
 367 }
 368
 369 static int proc_reg_release(struct inode *inode, struct file *file)
 370 {
 371         struct proc_dir_entry *pde = PDE(inode);
 372         struct pde_opener *pdeo;
 373         spin_lock(&pde->pde_unload_lock);
 374         list_for_each_entry(pdeo, &pde->pde_openers, lh) {
 375                 if (pdeo->file == file) {
 376                         close_pdeo(pde, pdeo);
 377                         break;
 378                 }
 379         }
 380         spin_unlock(&pde->pde_unload_lock);
 381         return 0;
 382 }
 383
 384 static const struct file_operations proc_reg_file_ops = {
 385         .llseek         = proc_reg_llseek,
 386         .read           = proc_reg_read,
 387         .write          = proc_reg_write,
 388         .poll           = proc_reg_poll,
 389         .unlocked_ioctl = proc_reg_unlocked_ioctl,
 390 #ifdef CONFIG_COMPAT
 391         .compat_ioctl   = proc_reg_compat_ioctl,
 392 #endif
 393         .mmap           = proc_reg_mmap,
 394         .get_unmapped_area = proc_reg_get_unmapped_area,
 395         .open           = proc_reg_open,
 396         .release        = proc_reg_release,
 397 };
 398
 399 #ifdef CONFIG_COMPAT
 400 static const struct file_operations proc_reg_file_ops_no_compat = {
 401         .llseek         = proc_reg_llseek,
 402         .read           = proc_reg_read,
 403         .write          = proc_reg_write,
 404         .poll           = proc_reg_poll,
 405         .unlocked_ioctl = proc_reg_unlocked_ioctl,
 406         .mmap           = proc_reg_mmap,
 407         .get_unmapped_area = proc_reg_get_unmapped_area,
 408         .open           = proc_reg_open,
 409         .release        = proc_reg_release,
 410 };
 411 #endif
 412
 413 static void proc_put_link(void *p)
 414 {
 415         unuse_pde(p);
 416 }
 417
 418 static const char *proc_get_link(struct dentry *dentry,
 419                                  struct inode *inode,
 420                                  struct delayed_call *done)
 421 {
 422         struct proc_dir_entry *pde = PDE(inode);
 423         if (unlikely(!use_pde(pde)))
 424                 return ERR_PTR(-EINVAL);
 425         set_delayed_call(done, proc_put_link, pde);
 426         return pde->data;
 427 }
 428
 429 const struct inode_operations proc_link_inode_operations = {
 430         .get_link       = proc_get_link,
 431 };
 432
 433 struct inode *proc_get_inode(struct super_block *sb, struct proc_dir_entry *de)
 434 {
 435         struct inode *inode = new_inode_pseudo(sb);
 436
 437         if (inode) {
 438                 inode->i_ino = de->low_ino;
 439                 inode->i_mtime = inode->i_atime = inode->i_ctime = current_time(inode);
 440                 PROC_I(inode)->pde = de;
 441
 442                 if (is_empty_pde(de)) {
 443                         make_empty_dir_inode(inode);
 444                         return inode;
 445                 }
 446                 if (de->mode) {
 447                         inode->i_mode = de->mode;
 448                         inode->i_uid = de->uid;
 449                         inode->i_gid = de->gid;
 450                 }
 451                 if (de->size)
 452                         inode->i_size = de->size;
 453                 if (de->nlink)
 454                         set_nlink(inode, de->nlink);
 455                 WARN_ON(!de->proc_iops);
 456                 inode->i_op = de->proc_iops;
 457                 if (de->proc_fops) {
 458                         if (S_ISREG(inode->i_mode)) {
 459 #ifdef CONFIG_COMPAT
 460                                 if (!de->proc_fops->compat_ioctl)
 461                                         inode->i_fop =
 462                                                 &proc_reg_file_ops_no_compat;
 463                                 else
 464 #endif
 465                                         inode->i_fop = &proc_reg_file_ops;
 466                         } else {
 467                                 inode->i_fop = de->proc_fops;
 468                         }
 469                 }
 470         } else
 471                pde_put(de);
 472         return inode;
 473 }
 474
 475 int proc_fill_super(struct super_block *s, void *data, int silent)
 476 {
 477         struct pid_namespace *ns = get_pid_ns(s->s_fs_info);
 478         struct inode *root_inode;
 479         int ret;
 480
 481         if (!proc_parse_options(data, ns))
 482                 return -EINVAL;
 483
 484         /* User space would break if executables or devices appear on proc */
 485         s->s_iflags |= SB_I_USERNS_VISIBLE | SB_I_NOEXEC | SB_I_NODEV;
 486         s->s_flags |= MS_NODIRATIME | MS_NOSUID | MS_NOEXEC;
 487         s->s_blocksize = 1024;
 488         s->s_blocksize_bits = 10;
 489         s->s_magic = PROC_SUPER_MAGIC;
 490         s->s_op = &proc_sops;
 491         s->s_time_gran = 1;
 492
 493         /*
 494          * procfs isn't actually a stacking filesystem; however, there is
 495          * too much magic going on inside it to permit stacking things on
 496          * top of it
 497          */
 498         s->s_stack_depth = FILESYSTEM_MAX_STACK_DEPTH;
 499
 500         pde_get(&proc_root);
 501         root_inode = proc_get_inode(s, &proc_root);
 502         if (!root_inode) {
 503                 pr_err("proc_fill_super: get root inode failed\n");
 504                 return -ENOMEM;
 505         }
 506
 507         s->s_root = d_make_root(root_inode);
 508         if (!s->s_root) {
 509                 pr_err("proc_fill_super: allocate dentry failed\n");
 510                 return -ENOMEM;
 511         }
 512
 513         ret = proc_setup_self(s);
 514         if (ret) {
 515                 return ret;
 516         }
 517         return proc_setup_thread_self(s);
 518 }