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/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
  27 static void proc_evict_inode(struct inode *inode)
  28 {
  29         struct proc_dir_entry *de;
  30
  31         truncate_inode_pages(&inode->i_data, 0);
  32         end_writeback(inode);
  33
  34         /* Stop tracking associated processes */
  35         put_pid(PROC_I(inode)->pid);
  36
  37         /* Let go of any associated proc directory entry */
  38         de = PROC_I(inode)->pde;
  39         if (de)
  40                 pde_put(de);
  41         if (PROC_I(inode)->sysctl)
  42                 sysctl_head_put(PROC_I(inode)->sysctl);
  43 }
  44
  45 struct vfsmount *proc_mnt;
  46
  47 static struct kmem_cache * proc_inode_cachep;
  48
  49 static struct inode *proc_alloc_inode(struct super_block *sb)
  50 {
  51         struct proc_inode *ei;
  52         struct inode *inode;
  53
  54         ei = (struct proc_inode *)kmem_cache_alloc(proc_inode_cachep, GFP_KERNEL);
  55         if (!ei)
  56                 return NULL;
  57         ei->pid = NULL;
  58         ei->fd = 0;
  59         ei->op.proc_get_link = NULL;
  60         ei->pde = NULL;
  61         ei->sysctl = NULL;
  62         ei->sysctl_entry = NULL;
  63         inode = &ei->vfs_inode;
  64         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
  65         return inode;
  66 }
  67
  68 static void proc_i_callback(struct rcu_head *head)
  69 {
  70         struct inode *inode = container_of(head, struct inode, i_rcu);
  71         INIT_LIST_HEAD(&inode->i_dentry);
  72         kmem_cache_free(proc_inode_cachep, PROC_I(inode));
  73 }
  74
  75 static void proc_destroy_inode(struct inode *inode)
  76 {
  77         call_rcu(&inode->i_rcu, proc_i_callback);
  78 }
  79
  80 static void init_once(void *foo)
  81 {
  82         struct proc_inode *ei = (struct proc_inode *) foo;
  83
  84         inode_init_once(&ei->vfs_inode);
  85 }
  86
  87 void __init proc_init_inodecache(void)
  88 {
  89         proc_inode_cachep = kmem_cache_create("proc_inode_cache",
  90                                              sizeof(struct proc_inode),
  91                                              0, (SLAB_RECLAIM_ACCOUNT|
  92                                                 SLAB_MEM_SPREAD|SLAB_PANIC),
  93                                              init_once);
  94 }
  95
  96 static const struct super_operations proc_sops = {
  97         .alloc_inode    = proc_alloc_inode,
  98         .destroy_inode  = proc_destroy_inode,
  99         .drop_inode     = generic_delete_inode,
 100         .evict_inode    = proc_evict_inode,
 101         .statfs         = simple_statfs,
 102 };
 103
 104 static void __pde_users_dec(struct proc_dir_entry *pde)
 105 {
 106         pde->pde_users--;
 107         if (pde->pde_unload_completion && pde->pde_users == 0)
 108                 complete(pde->pde_unload_completion);
 109 }
 110
 111 void pde_users_dec(struct proc_dir_entry *pde)
 112 {
 113         spin_lock(&pde->pde_unload_lock);
 114         __pde_users_dec(pde);
 115         spin_unlock(&pde->pde_unload_lock);
 116 }
 117
 118 static loff_t proc_reg_llseek(struct file *file, loff_t offset, int whence)
 119 {
 120         struct proc_dir_entry *pde = PDE(file->f_path.dentry->d_inode);
 121         loff_t rv = -EINVAL;
 122         loff_t (*llseek)(struct file *, loff_t, int);
 123
 124         spin_lock(&pde->pde_unload_lock);
 125         /*
 126          * remove_proc_entry() is going to delete PDE (as part of module
 127          * cleanup sequence). No new callers into module allowed.
 128          */
 129         if (!pde->proc_fops) {
 130                 spin_unlock(&pde->pde_unload_lock);
 131                 return rv;
 132         }
 133         /*
 134          * Bump refcount so that remove_proc_entry will wail for ->llseek to
 135          * complete.
 136          */
 137         pde->pde_users++;
 138         /*
 139          * Save function pointer under lock, to protect against ->proc_fops
 140          * NULL'ifying right after ->pde_unload_lock is dropped.
 141          */
 142         llseek = pde->proc_fops->llseek;
 143         spin_unlock(&pde->pde_unload_lock);
 144
 145         if (!llseek)
 146                 llseek = default_llseek;
 147         rv = llseek(file, offset, whence);
 148
 149         pde_users_dec(pde);
 150         return rv;
 151 }
 152
 153 static ssize_t proc_reg_read(struct file *file, char __user *buf, size_t count, loff_t *ppos)
 154 {
 155         struct proc_dir_entry *pde = PDE(file->f_path.dentry->d_inode);
 156         ssize_t rv = -EIO;
 157         ssize_t (*read)(struct file *, char __user *, size_t, loff_t *);
 158
 159         spin_lock(&pde->pde_unload_lock);
 160         if (!pde->proc_fops) {
 161                 spin_unlock(&pde->pde_unload_lock);
 162                 return rv;
 163         }
 164         pde->pde_users++;
 165         read = pde->proc_fops->read;
 166         spin_unlock(&pde->pde_unload_lock);
 167
 168         if (read)
 169                 rv = read(file, buf, count, ppos);
 170
 171         pde_users_dec(pde);
 172         return rv;
 173 }
 174
 175 static ssize_t proc_reg_write(struct file *file, const char __user *buf, size_t count, loff_t *ppos)
 176 {
 177         struct proc_dir_entry *pde = PDE(file->f_path.dentry->d_inode);
 178         ssize_t rv = -EIO;
 179         ssize_t (*write)(struct file *, const char __user *, size_t, loff_t *);
 180
 181         spin_lock(&pde->pde_unload_lock);
 182         if (!pde->proc_fops) {
 183                 spin_unlock(&pde->pde_unload_lock);
 184                 return rv;
 185         }
 186         pde->pde_users++;
 187         write = pde->proc_fops->write;
 188         spin_unlock(&pde->pde_unload_lock);
 189
 190         if (write)
 191                 rv = write(file, buf, count, ppos);
 192
 193         pde_users_dec(pde);
 194         return rv;
 195 }
 196
 197 static unsigned int proc_reg_poll(struct file *file, struct poll_table_struct *pts)
 198 {
 199         struct proc_dir_entry *pde = PDE(file->f_path.dentry->d_inode);
 200         unsigned int rv = DEFAULT_POLLMASK;
 201         unsigned int (*poll)(struct file *, struct poll_table_struct *);
 202
 203         spin_lock(&pde->pde_unload_lock);
 204         if (!pde->proc_fops) {
 205                 spin_unlock(&pde->pde_unload_lock);
 206                 return rv;
 207         }
 208         pde->pde_users++;
 209         poll = pde->proc_fops->poll;
 210         spin_unlock(&pde->pde_unload_lock);
 211
 212         if (poll)
 213                 rv = poll(file, pts);
 214
 215         pde_users_dec(pde);
 216         return rv;
 217 }
 218
 219 static long proc_reg_unlocked_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
 220 {
 221         struct proc_dir_entry *pde = PDE(file->f_path.dentry->d_inode);
 222         long rv = -ENOTTY;
 223         long (*ioctl)(struct file *, unsigned int, unsigned long);
 224
 225         spin_lock(&pde->pde_unload_lock);
 226         if (!pde->proc_fops) {
 227                 spin_unlock(&pde->pde_unload_lock);
 228                 return rv;
 229         }
 230         pde->pde_users++;
 231         ioctl = pde->proc_fops->unlocked_ioctl;
 232         spin_unlock(&pde->pde_unload_lock);
 233
 234         if (ioctl)
 235                 rv = ioctl(file, cmd, arg);
 236
 237         pde_users_dec(pde);
 238         return rv;
 239 }
 240
 241 #ifdef CONFIG_COMPAT
 242 static long proc_reg_compat_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
 243 {
 244         struct proc_dir_entry *pde = PDE(file->f_path.dentry->d_inode);
 245         long rv = -ENOTTY;
 246         long (*compat_ioctl)(struct file *, unsigned int, unsigned long);
 247
 248         spin_lock(&pde->pde_unload_lock);
 249         if (!pde->proc_fops) {
 250                 spin_unlock(&pde->pde_unload_lock);
 251                 return rv;
 252         }
 253         pde->pde_users++;
 254         compat_ioctl = pde->proc_fops->compat_ioctl;
 255         spin_unlock(&pde->pde_unload_lock);
 256
 257         if (compat_ioctl)
 258                 rv = compat_ioctl(file, cmd, arg);
 259
 260         pde_users_dec(pde);
 261         return rv;
 262 }
 263 #endif
 264
 265 static int proc_reg_mmap(struct file *file, struct vm_area_struct *vma)
 266 {
 267         struct proc_dir_entry *pde = PDE(file->f_path.dentry->d_inode);
 268         int rv = -EIO;
 269         int (*mmap)(struct file *, struct vm_area_struct *);
 270
 271         spin_lock(&pde->pde_unload_lock);
 272         if (!pde->proc_fops) {
 273                 spin_unlock(&pde->pde_unload_lock);
 274                 return rv;
 275         }
 276         pde->pde_users++;
 277         mmap = pde->proc_fops->mmap;
 278         spin_unlock(&pde->pde_unload_lock);
 279
 280         if (mmap)
 281                 rv = mmap(file, vma);
 282
 283         pde_users_dec(pde);
 284         return rv;
 285 }
 286
 287 static int proc_reg_open(struct inode *inode, struct file *file)
 288 {
 289         struct proc_dir_entry *pde = PDE(inode);
 290         int rv = 0;
 291         int (*open)(struct inode *, struct file *);
 292         int (*release)(struct inode *, struct file *);
 293         struct pde_opener *pdeo;
 294
 295         /*
 296          * What for, you ask? Well, we can have open, rmmod, remove_proc_entry
 297          * sequence. ->release won't be called because ->proc_fops will be
 298          * cleared. Depending on complexity of ->release, consequences vary.
 299          *
 300          * We can't wait for mercy when close will be done for real, it's
 301          * deadlockable: rmmod foo </proc/foo . So, we're going to do ->release
 302          * by hand in remove_proc_entry(). For this, save opener's credentials
 303          * for later.
 304          */
 305         pdeo = kmalloc(sizeof(struct pde_opener), GFP_KERNEL);
 306         if (!pdeo)
 307                 return -ENOMEM;
 308
 309         spin_lock(&pde->pde_unload_lock);
 310         if (!pde->proc_fops) {
 311                 spin_unlock(&pde->pde_unload_lock);
 312                 kfree(pdeo);
 313                 return -EINVAL;
 314         }
 315         pde->pde_users++;
 316         open = pde->proc_fops->open;
 317         release = pde->proc_fops->release;
 318         spin_unlock(&pde->pde_unload_lock);
 319
 320         if (open)
 321                 rv = open(inode, file);
 322
 323         spin_lock(&pde->pde_unload_lock);
 324         if (rv == 0 && release) {
 325                 /* To know what to release. */
 326                 pdeo->inode = inode;
 327                 pdeo->file = file;
 328                 /* Strictly for "too late" ->release in proc_reg_release(). */
 329                 pdeo->release = release;
 330                 list_add(&pdeo->lh, &pde->pde_openers);
 331         } else
 332                 kfree(pdeo);
 333         __pde_users_dec(pde);
 334         spin_unlock(&pde->pde_unload_lock);
 335         return rv;
 336 }
 337
 338 static struct pde_opener *find_pde_opener(struct proc_dir_entry *pde,
 339                                         struct inode *inode, struct file *file)
 340 {
 341         struct pde_opener *pdeo;
 342
 343         list_for_each_entry(pdeo, &pde->pde_openers, lh) {
 344                 if (pdeo->inode == inode && pdeo->file == file)
 345                         return pdeo;
 346         }
 347         return NULL;
 348 }
 349
 350 static int proc_reg_release(struct inode *inode, struct file *file)
 351 {
 352         struct proc_dir_entry *pde = PDE(inode);
 353         int rv = 0;
 354         int (*release)(struct inode *, struct file *);
 355         struct pde_opener *pdeo;
 356
 357         spin_lock(&pde->pde_unload_lock);
 358         pdeo = find_pde_opener(pde, inode, file);
 359         if (!pde->proc_fops) {
 360                 /*
 361                  * Can't simply exit, __fput() will think that everything is OK,
 362                  * and move on to freeing struct file. remove_proc_entry() will
 363                  * find slacker in opener's list and will try to do non-trivial
 364                  * things with struct file. Therefore, remove opener from list.
 365                  *
 366                  * But if opener is removed from list, who will ->release it?
 367                  */
 368                 if (pdeo) {
 369                         list_del(&pdeo->lh);
 370                         spin_unlock(&pde->pde_unload_lock);
 371                         rv = pdeo->release(inode, file);
 372                         kfree(pdeo);
 373                 } else
 374                         spin_unlock(&pde->pde_unload_lock);
 375                 return rv;
 376         }
 377         pde->pde_users++;
 378         release = pde->proc_fops->release;
 379         if (pdeo) {
 380                 list_del(&pdeo->lh);
 381                 kfree(pdeo);
 382         }
 383         spin_unlock(&pde->pde_unload_lock);
 384
 385         if (release)
 386                 rv = release(inode, file);
 387
 388         pde_users_dec(pde);
 389         return rv;
 390 }
 391
 392 static const struct file_operations proc_reg_file_ops = {
 393         .llseek         = proc_reg_llseek,
 394         .read           = proc_reg_read,
 395         .write          = proc_reg_write,
 396         .poll           = proc_reg_poll,
 397         .unlocked_ioctl = proc_reg_unlocked_ioctl,
 398 #ifdef CONFIG_COMPAT
 399         .compat_ioctl   = proc_reg_compat_ioctl,
 400 #endif
 401         .mmap           = proc_reg_mmap,
 402         .open           = proc_reg_open,
 403         .release        = proc_reg_release,
 404 };
 405
 406 #ifdef CONFIG_COMPAT
 407 static const struct file_operations proc_reg_file_ops_no_compat = {
 408         .llseek         = proc_reg_llseek,
 409         .read           = proc_reg_read,
 410         .write          = proc_reg_write,
 411         .poll           = proc_reg_poll,
 412         .unlocked_ioctl = proc_reg_unlocked_ioctl,
 413         .mmap           = proc_reg_mmap,
 414         .open           = proc_reg_open,
 415         .release        = proc_reg_release,
 416 };
 417 #endif
 418
 419 struct inode *proc_get_inode(struct super_block *sb, struct proc_dir_entry *de)
 420 {
 421         struct inode * inode;
 422
 423         inode = iget_locked(sb, de->low_ino);
 424         if (!inode)
 425                 return NULL;
 426         if (inode->i_state & I_NEW) {
 427                 inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
 428                 PROC_I(inode)->fd = 0;
 429                 PROC_I(inode)->pde = de;
 430
 431                 if (de->mode) {
 432                         inode->i_mode = de->mode;
 433                         inode->i_uid = de->uid;
 434                         inode->i_gid = de->gid;
 435                 }
 436                 if (de->size)
 437                         inode->i_size = de->size;
 438                 if (de->nlink)
 439                         inode->i_nlink = de->nlink;
 440                 if (de->proc_iops)
 441                         inode->i_op = de->proc_iops;
 442                 if (de->proc_fops) {
 443                         if (S_ISREG(inode->i_mode)) {
 444 #ifdef CONFIG_COMPAT
 445                                 if (!de->proc_fops->compat_ioctl)
 446                                         inode->i_fop =
 447                                                 &proc_reg_file_ops_no_compat;
 448                                 else
 449 #endif
 450                                         inode->i_fop = &proc_reg_file_ops;
 451                         } else {
 452                                 inode->i_fop = de->proc_fops;
 453                         }
 454                 }
 455                 unlock_new_inode(inode);
 456         } else
 457                pde_put(de);
 458         return inode;
 459 }
 460
 461 int proc_fill_super(struct super_block *s)
 462 {
 463         struct inode * root_inode;
 464
 465         s->s_flags |= MS_NODIRATIME | MS_NOSUID | MS_NOEXEC;
 466         s->s_blocksize = 1024;
 467         s->s_blocksize_bits = 10;
 468         s->s_magic = PROC_SUPER_MAGIC;
 469         s->s_op = &proc_sops;
 470         s->s_time_gran = 1;
 471
 472         pde_get(&proc_root);
 473         root_inode = proc_get_inode(s, &proc_root);
 474         if (!root_inode)
 475                 goto out_no_root;
 476         root_inode->i_uid = 0;
 477         root_inode->i_gid = 0;
 478         s->s_root = d_alloc_root(root_inode);
 479         if (!s->s_root)
 480                 goto out_no_root;
 481         return 0;
 482
 483 out_no_root:
 484         printk("proc_read_super: get root inode failed\n");
 485         iput(root_inode);
 486         pde_put(&proc_root);
 487         return -ENOMEM;
 488 }