fs/proc/inode.c

   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
  26 #include <asm/uaccess.h>
  27
  28 #include "internal.h"
  29
  30 static void proc_evict_inode(struct inode *inode)
  31 {
  32         struct proc_dir_entry *de;
  33         struct ctl_table_header *head;
  34         const struct proc_ns_operations *ns_ops;
  35         void *ns;
  36
  37         truncate_inode_pages(&inode->i_data, 0);
  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 = PROC_I(inode)->pde;
  45         if (de)
  46                 pde_put(de);
  47         head = PROC_I(inode)->sysctl;
  48         if (head) {
  49                 rcu_assign_pointer(PROC_I(inode)->sysctl, NULL);
  50                 sysctl_head_put(head);
  51         }
  52         /* Release any associated namespace */
  53         ns_ops = PROC_I(inode)->ns_ops;
  54         ns = PROC_I(inode)->ns;
  55         if (ns_ops && ns)
  56                 ns_ops->put(ns);
  57 }
  58
  59 static struct kmem_cache * proc_inode_cachep;
  60
  61 static struct inode *proc_alloc_inode(struct super_block *sb)
  62 {
  63         struct proc_inode *ei;
  64         struct inode *inode;
  65
  66         ei = (struct proc_inode *)kmem_cache_alloc(proc_inode_cachep, GFP_KERNEL);
  67         if (!ei)
  68                 return NULL;
  69         ei->pid = NULL;
  70         ei->fd = 0;
  71         ei->op.proc_get_link = NULL;
  72         ei->pde = NULL;
  73         ei->sysctl = NULL;
  74         ei->sysctl_entry = NULL;
  75         ei->ns = NULL;
  76         ei->ns_ops = NULL;
  77         inode = &ei->vfs_inode;
  78         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
  79         return inode;
  80 }
  81
  82 static void proc_i_callback(struct rcu_head *head)
  83 {
  84         struct inode *inode = container_of(head, struct inode, i_rcu);
  85         kmem_cache_free(proc_inode_cachep, PROC_I(inode));
  86 }
  87
  88 static void proc_destroy_inode(struct inode *inode)
  89 {
  90         call_rcu(&inode->i_rcu, proc_i_callback);
  91 }
  92
  93 static void init_once(void *foo)
  94 {
  95         struct proc_inode *ei = (struct proc_inode *) foo;
  96
  97         inode_init_once(&ei->vfs_inode);
  98 }
  99
 100 void __init proc_init_inodecache(void)
 101 {
 102         proc_inode_cachep = kmem_cache_create("proc_inode_cache",
 103                                              sizeof(struct proc_inode),
 104                                              0, (SLAB_RECLAIM_ACCOUNT|
 105                                                 SLAB_MEM_SPREAD|SLAB_PANIC),
 106                                              init_once);
 107 }
 108
 109 static int proc_show_options(struct seq_file *seq, struct dentry *root)
 110 {
 111         struct super_block *sb = root->d_sb;
 112         struct pid_namespace *pid = sb->s_fs_info;
 113
 114         if (!gid_eq(pid->pid_gid, GLOBAL_ROOT_GID))
 115                 seq_printf(seq, ",gid=%u", from_kgid_munged(&init_user_ns, pid->pid_gid));
 116         if (pid->hide_pid != 0)
 117                 seq_printf(seq, ",hidepid=%u", pid->hide_pid);
 118
 119         return 0;
 120 }
 121
 122 static const struct super_operations proc_sops = {
 123         .alloc_inode    = proc_alloc_inode,
 124         .destroy_inode  = proc_destroy_inode,
 125         .drop_inode     = generic_delete_inode,
 126         .evict_inode    = proc_evict_inode,
 127         .statfs         = simple_statfs,
 128         .remount_fs     = proc_remount,
 129         .show_options   = proc_show_options,
 130 };
 131
 132 static void __pde_users_dec(struct proc_dir_entry *pde)
 133 {
 134         pde->pde_users--;
 135         if (pde->pde_unload_completion && pde->pde_users == 0)
 136                 complete(pde->pde_unload_completion);
 137 }
 138
 139 void pde_users_dec(struct proc_dir_entry *pde)
 140 {
 141         spin_lock(&pde->pde_unload_lock);
 142         __pde_users_dec(pde);
 143         spin_unlock(&pde->pde_unload_lock);
 144 }
 145
 146 static loff_t proc_reg_llseek(struct file *file, loff_t offset, int whence)
 147 {
 148         struct proc_dir_entry *pde = PDE(file_inode(file));
 149         loff_t rv = -EINVAL;
 150         loff_t (*llseek)(struct file *, loff_t, int);
 151
 152         spin_lock(&pde->pde_unload_lock);
 153         /*
 154          * remove_proc_entry() is going to delete PDE (as part of module
 155          * cleanup sequence). No new callers into module allowed.
 156          */
 157         if (!pde->proc_fops) {
 158                 spin_unlock(&pde->pde_unload_lock);
 159                 return rv;
 160         }
 161         /*
 162          * Bump refcount so that remove_proc_entry will wail for ->llseek to
 163          * complete.
 164          */
 165         pde->pde_users++;
 166         /*
 167          * Save function pointer under lock, to protect against ->proc_fops
 168          * NULL'ifying right after ->pde_unload_lock is dropped.
 169          */
 170         llseek = pde->proc_fops->llseek;
 171         spin_unlock(&pde->pde_unload_lock);
 172
 173         if (!llseek)
 174                 llseek = default_llseek;
 175         rv = llseek(file, offset, whence);
 176
 177         pde_users_dec(pde);
 178         return rv;
 179 }
 180
 181 static ssize_t proc_reg_read(struct file *file, char __user *buf, size_t count, loff_t *ppos)
 182 {
 183         struct proc_dir_entry *pde = PDE(file_inode(file));
 184         ssize_t rv = -EIO;
 185         ssize_t (*read)(struct file *, char __user *, size_t, loff_t *);
 186
 187         spin_lock(&pde->pde_unload_lock);
 188         if (!pde->proc_fops) {
 189                 spin_unlock(&pde->pde_unload_lock);
 190                 return rv;
 191         }
 192         pde->pde_users++;
 193         read = pde->proc_fops->read;
 194         spin_unlock(&pde->pde_unload_lock);
 195
 196         if (read)
 197                 rv = read(file, buf, count, ppos);
 198
 199         pde_users_dec(pde);
 200         return rv;
 201 }
 202
 203 static ssize_t proc_reg_write(struct file *file, const char __user *buf, size_t count, loff_t *ppos)
 204 {
 205         struct proc_dir_entry *pde = PDE(file_inode(file));
 206         ssize_t rv = -EIO;
 207         ssize_t (*write)(struct file *, const char __user *, size_t, loff_t *);
 208
 209         spin_lock(&pde->pde_unload_lock);
 210         if (!pde->proc_fops) {
 211                 spin_unlock(&pde->pde_unload_lock);
 212                 return rv;
 213         }
 214         pde->pde_users++;
 215         write = pde->proc_fops->write;
 216         spin_unlock(&pde->pde_unload_lock);
 217
 218         if (write)
 219                 rv = write(file, buf, count, ppos);
 220
 221         pde_users_dec(pde);
 222         return rv;
 223 }
 224
 225 static unsigned int proc_reg_poll(struct file *file, struct poll_table_struct *pts)
 226 {
 227         struct proc_dir_entry *pde = PDE(file_inode(file));
 228         unsigned int rv = DEFAULT_POLLMASK;
 229         unsigned int (*poll)(struct file *, struct poll_table_struct *);
 230
 231         spin_lock(&pde->pde_unload_lock);
 232         if (!pde->proc_fops) {
 233                 spin_unlock(&pde->pde_unload_lock);
 234                 return rv;
 235         }
 236         pde->pde_users++;
 237         poll = pde->proc_fops->poll;
 238         spin_unlock(&pde->pde_unload_lock);
 239
 240         if (poll)
 241                 rv = poll(file, pts);
 242
 243         pde_users_dec(pde);
 244         return rv;
 245 }
 246
 247 static long proc_reg_unlocked_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
 248 {
 249         struct proc_dir_entry *pde = PDE(file_inode(file));
 250         long rv = -ENOTTY;
 251         long (*ioctl)(struct file *, unsigned int, unsigned long);
 252
 253         spin_lock(&pde->pde_unload_lock);
 254         if (!pde->proc_fops) {
 255                 spin_unlock(&pde->pde_unload_lock);
 256                 return rv;
 257         }
 258         pde->pde_users++;
 259         ioctl = pde->proc_fops->unlocked_ioctl;
 260         spin_unlock(&pde->pde_unload_lock);
 261
 262         if (ioctl)
 263                 rv = ioctl(file, cmd, arg);
 264
 265         pde_users_dec(pde);
 266         return rv;
 267 }
 268
 269 #ifdef CONFIG_COMPAT
 270 static long proc_reg_compat_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
 271 {
 272         struct proc_dir_entry *pde = PDE(file_inode(file));
 273         long rv = -ENOTTY;
 274         long (*compat_ioctl)(struct file *, unsigned int, unsigned long);
 275
 276         spin_lock(&pde->pde_unload_lock);
 277         if (!pde->proc_fops) {
 278                 spin_unlock(&pde->pde_unload_lock);
 279                 return rv;
 280         }
 281         pde->pde_users++;
 282         compat_ioctl = pde->proc_fops->compat_ioctl;
 283         spin_unlock(&pde->pde_unload_lock);
 284
 285         if (compat_ioctl)
 286                 rv = compat_ioctl(file, cmd, arg);
 287
 288         pde_users_dec(pde);
 289         return rv;
 290 }
 291 #endif
 292
 293 static int proc_reg_mmap(struct file *file, struct vm_area_struct *vma)
 294 {
 295         struct proc_dir_entry *pde = PDE(file_inode(file));
 296         int rv = -EIO;
 297         int (*mmap)(struct file *, struct vm_area_struct *);
 298
 299         spin_lock(&pde->pde_unload_lock);
 300         if (!pde->proc_fops) {
 301                 spin_unlock(&pde->pde_unload_lock);
 302                 return rv;
 303         }
 304         pde->pde_users++;
 305         mmap = pde->proc_fops->mmap;
 306         spin_unlock(&pde->pde_unload_lock);
 307
 308         if (mmap)
 309                 rv = mmap(file, vma);
 310
 311         pde_users_dec(pde);
 312         return rv;
 313 }
 314
 315 static int proc_reg_open(struct inode *inode, struct file *file)
 316 {
 317         struct proc_dir_entry *pde = PDE(inode);
 318         int rv = 0;
 319         int (*open)(struct inode *, struct file *);
 320         int (*release)(struct inode *, struct file *);
 321         struct pde_opener *pdeo;
 322
 323         /*
 324          * What for, you ask? Well, we can have open, rmmod, remove_proc_entry
 325          * sequence. ->release won't be called because ->proc_fops will be
 326          * cleared. Depending on complexity of ->release, consequences vary.
 327          *
 328          * We can't wait for mercy when close will be done for real, it's
 329          * deadlockable: rmmod foo </proc/foo . So, we're going to do ->release
 330          * by hand in remove_proc_entry(). For this, save opener's credentials
 331          * for later.
 332          */
 333         pdeo = kmalloc(sizeof(struct pde_opener), GFP_KERNEL);
 334         if (!pdeo)
 335                 return -ENOMEM;
 336
 337         spin_lock(&pde->pde_unload_lock);
 338         if (!pde->proc_fops) {
 339                 spin_unlock(&pde->pde_unload_lock);
 340                 kfree(pdeo);
 341                 return -ENOENT;
 342         }
 343         pde->pde_users++;
 344         open = pde->proc_fops->open;
 345         release = pde->proc_fops->release;
 346         spin_unlock(&pde->pde_unload_lock);
 347
 348         if (open)
 349                 rv = open(inode, file);
 350
 351         spin_lock(&pde->pde_unload_lock);
 352         if (rv == 0 && release) {
 353                 /* To know what to release. */
 354                 pdeo->inode = inode;
 355                 pdeo->file = file;
 356                 /* Strictly for "too late" ->release in proc_reg_release(). */
 357                 pdeo->release = release;
 358                 list_add(&pdeo->lh, &pde->pde_openers);
 359         } else
 360                 kfree(pdeo);
 361         __pde_users_dec(pde);
 362         spin_unlock(&pde->pde_unload_lock);
 363         return rv;
 364 }
 365
 366 static struct pde_opener *find_pde_opener(struct proc_dir_entry *pde,
 367                                         struct inode *inode, struct file *file)
 368 {
 369         struct pde_opener *pdeo;
 370
 371         list_for_each_entry(pdeo, &pde->pde_openers, lh) {
 372                 if (pdeo->inode == inode && pdeo->file == file)
 373                         return pdeo;
 374         }
 375         return NULL;
 376 }
 377
 378 static int proc_reg_release(struct inode *inode, struct file *file)
 379 {
 380         struct proc_dir_entry *pde = PDE(inode);
 381         int rv = 0;
 382         int (*release)(struct inode *, struct file *);
 383         struct pde_opener *pdeo;
 384
 385         spin_lock(&pde->pde_unload_lock);
 386         pdeo = find_pde_opener(pde, inode, file);
 387         if (!pde->proc_fops) {
 388                 /*
 389                  * Can't simply exit, __fput() will think that everything is OK,
 390                  * and move on to freeing struct file. remove_proc_entry() will
 391                  * find slacker in opener's list and will try to do non-trivial
 392                  * things with struct file. Therefore, remove opener from list.
 393                  *
 394                  * But if opener is removed from list, who will ->release it?
 395                  */
 396                 if (pdeo) {
 397                         list_del(&pdeo->lh);
 398                         spin_unlock(&pde->pde_unload_lock);
 399                         rv = pdeo->release(inode, file);
 400                         kfree(pdeo);
 401                 } else
 402                         spin_unlock(&pde->pde_unload_lock);
 403                 return rv;
 404         }
 405         pde->pde_users++;
 406         release = pde->proc_fops->release;
 407         if (pdeo) {
 408                 list_del(&pdeo->lh);
 409                 kfree(pdeo);
 410         }
 411         spin_unlock(&pde->pde_unload_lock);
 412
 413         if (release)
 414                 rv = release(inode, file);
 415
 416         pde_users_dec(pde);
 417         return rv;
 418 }
 419
 420 static const struct file_operations proc_reg_file_ops = {
 421         .llseek         = proc_reg_llseek,
 422         .read           = proc_reg_read,
 423         .write          = proc_reg_write,
 424         .poll           = proc_reg_poll,
 425         .unlocked_ioctl = proc_reg_unlocked_ioctl,
 426 #ifdef CONFIG_COMPAT
 427         .compat_ioctl   = proc_reg_compat_ioctl,
 428 #endif
 429         .mmap           = proc_reg_mmap,
 430         .open           = proc_reg_open,
 431         .release        = proc_reg_release,
 432 };
 433
 434 #ifdef CONFIG_COMPAT
 435 static const struct file_operations proc_reg_file_ops_no_compat = {
 436         .llseek         = proc_reg_llseek,
 437         .read           = proc_reg_read,
 438         .write          = proc_reg_write,
 439         .poll           = proc_reg_poll,
 440         .unlocked_ioctl = proc_reg_unlocked_ioctl,
 441         .mmap           = proc_reg_mmap,
 442         .open           = proc_reg_open,
 443         .release        = proc_reg_release,
 444 };
 445 #endif
 446
 447 struct inode *proc_get_inode(struct super_block *sb, struct proc_dir_entry *de)
 448 {
 449         struct inode *inode = iget_locked(sb, de->low_ino);
 450
 451         if (inode && (inode->i_state & I_NEW)) {
 452                 inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
 453                 PROC_I(inode)->pde = de;
 454
 455                 if (de->mode) {
 456                         inode->i_mode = de->mode;
 457                         inode->i_uid = de->uid;
 458                         inode->i_gid = de->gid;
 459                 }
 460                 if (de->size)
 461                         inode->i_size = de->size;
 462                 if (de->nlink)
 463                         set_nlink(inode, de->nlink);
 464                 if (de->proc_iops)
 465                         inode->i_op = de->proc_iops;
 466                 if (de->proc_fops) {
 467                         if (S_ISREG(inode->i_mode)) {
 468 #ifdef CONFIG_COMPAT
 469                                 if (!de->proc_fops->compat_ioctl)
 470                                         inode->i_fop =
 471                                                 &proc_reg_file_ops_no_compat;
 472                                 else
 473 #endif
 474                                         inode->i_fop = &proc_reg_file_ops;
 475                         } else {
 476                                 inode->i_fop = de->proc_fops;
 477                         }
 478                 }
 479                 unlock_new_inode(inode);
 480         } else
 481                pde_put(de);
 482         return inode;
 483 }
 484
 485 int proc_fill_super(struct super_block *s)
 486 {
 487         struct inode *root_inode;
 488
 489         s->s_flags |= MS_NODIRATIME | MS_NOSUID | MS_NOEXEC;
 490         s->s_blocksize = 1024;
 491         s->s_blocksize_bits = 10;
 492         s->s_magic = PROC_SUPER_MAGIC;
 493         s->s_op = &proc_sops;
 494         s->s_time_gran = 1;
 495
 496         pde_get(&proc_root);
 497         root_inode = proc_get_inode(s, &proc_root);
 498         if (!root_inode) {
 499                 pr_err("proc_fill_super: get root inode failed\n");
 500                 return -ENOMEM;
 501         }
 502
 503         s->s_root = d_make_root(root_inode);
 504         if (!s->s_root) {
 505                 pr_err("proc_fill_super: allocate dentry failed\n");
 506                 return -ENOMEM;
 507         }
 508
 509         return 0;
 510 }