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/smp_lock.h>
  20 #include <linux/sysctl.h>
  21
  22 #include <asm/system.h>
  23 #include <asm/uaccess.h>
  24
  25 #include "internal.h"
  26
  27 static void proc_delete_inode(struct inode *inode)
  28 {
  29         struct proc_dir_entry *de;
  30
  31         truncate_inode_pages(&inode->i_data, 0);
  32
  33         /* Stop tracking associated processes */
  34         put_pid(PROC_I(inode)->pid);
  35
  36         /* Let go of any associated proc directory entry */
  37         de = PROC_I(inode)->pde;
  38         if (de)
  39                 pde_put(de);
  40         if (PROC_I(inode)->sysctl)
  41                 sysctl_head_put(PROC_I(inode)->sysctl);
  42         clear_inode(inode);
  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_destroy_inode(struct inode *inode)
  69 {
  70         kmem_cache_free(proc_inode_cachep, PROC_I(inode));
  71 }
  72
  73 static void init_once(void *foo)
  74 {
  75         struct proc_inode *ei = (struct proc_inode *) foo;
  76
  77         inode_init_once(&ei->vfs_inode);
  78 }
  79
  80 void __init proc_init_inodecache(void)
  81 {
  82         proc_inode_cachep = kmem_cache_create("proc_inode_cache",
  83                                              sizeof(struct proc_inode),
  84                                              0, (SLAB_RECLAIM_ACCOUNT|
  85                                                 SLAB_MEM_SPREAD|SLAB_PANIC),
  86                                              init_once);
  87 }
  88
  89 static const struct super_operations proc_sops = {
  90         .alloc_inode    = proc_alloc_inode,
  91         .destroy_inode  = proc_destroy_inode,
  92         .drop_inode     = generic_delete_inode,
  93         .delete_inode   = proc_delete_inode,
  94         .statfs         = simple_statfs,
  95 };
  96
  97 static void __pde_users_dec(struct proc_dir_entry *pde)
  98 {
  99         pde->pde_users--;
 100         if (pde->pde_unload_completion && pde->pde_users == 0)
 101                 complete(pde->pde_unload_completion);
 102 }
 103
 104 void pde_users_dec(struct proc_dir_entry *pde)
 105 {
 106         spin_lock(&pde->pde_unload_lock);
 107         __pde_users_dec(pde);
 108         spin_unlock(&pde->pde_unload_lock);
 109 }
 110
 111 static loff_t proc_reg_llseek(struct file *file, loff_t offset, int whence)
 112 {
 113         struct proc_dir_entry *pde = PDE(file->f_path.dentry->d_inode);
 114         loff_t rv = -EINVAL;
 115         loff_t (*llseek)(struct file *, loff_t, int);
 116
 117         spin_lock(&pde->pde_unload_lock);
 118         /*
 119          * remove_proc_entry() is going to delete PDE (as part of module
 120          * cleanup sequence). No new callers into module allowed.
 121          */
 122         if (!pde->proc_fops) {
 123                 spin_unlock(&pde->pde_unload_lock);
 124                 return rv;
 125         }
 126         /*
 127          * Bump refcount so that remove_proc_entry will wail for ->llseek to
 128          * complete.
 129          */
 130         pde->pde_users++;
 131         /*
 132          * Save function pointer under lock, to protect against ->proc_fops
 133          * NULL'ifying right after ->pde_unload_lock is dropped.
 134          */
 135         llseek = pde->proc_fops->llseek;
 136         spin_unlock(&pde->pde_unload_lock);
 137
 138         if (!llseek)
 139                 llseek = default_llseek;
 140         rv = llseek(file, offset, whence);
 141
 142         pde_users_dec(pde);
 143         return rv;
 144 }
 145
 146 static ssize_t proc_reg_read(struct file *file, char __user *buf, size_t count, loff_t *ppos)
 147 {
 148         struct proc_dir_entry *pde = PDE(file->f_path.dentry->d_inode);
 149         ssize_t rv = -EIO;
 150         ssize_t (*read)(struct file *, char __user *, size_t, loff_t *);
 151
 152         spin_lock(&pde->pde_unload_lock);
 153         if (!pde->proc_fops) {
 154                 spin_unlock(&pde->pde_unload_lock);
 155                 return rv;
 156         }
 157         pde->pde_users++;
 158         read = pde->proc_fops->read;
 159         spin_unlock(&pde->pde_unload_lock);
 160
 161         if (read)
 162                 rv = read(file, buf, count, ppos);
 163
 164         pde_users_dec(pde);
 165         return rv;
 166 }
 167
 168 static ssize_t proc_reg_write(struct file *file, const char __user *buf, size_t count, loff_t *ppos)
 169 {
 170         struct proc_dir_entry *pde = PDE(file->f_path.dentry->d_inode);
 171         ssize_t rv = -EIO;
 172         ssize_t (*write)(struct file *, const char __user *, size_t, loff_t *);
 173
 174         spin_lock(&pde->pde_unload_lock);
 175         if (!pde->proc_fops) {
 176                 spin_unlock(&pde->pde_unload_lock);
 177                 return rv;
 178         }
 179         pde->pde_users++;
 180         write = pde->proc_fops->write;
 181         spin_unlock(&pde->pde_unload_lock);
 182
 183         if (write)
 184                 rv = write(file, buf, count, ppos);
 185
 186         pde_users_dec(pde);
 187         return rv;
 188 }
 189
 190 static unsigned int proc_reg_poll(struct file *file, struct poll_table_struct *pts)
 191 {
 192         struct proc_dir_entry *pde = PDE(file->f_path.dentry->d_inode);
 193         unsigned int rv = DEFAULT_POLLMASK;
 194         unsigned int (*poll)(struct file *, struct poll_table_struct *);
 195
 196         spin_lock(&pde->pde_unload_lock);
 197         if (!pde->proc_fops) {
 198                 spin_unlock(&pde->pde_unload_lock);
 199                 return rv;
 200         }
 201         pde->pde_users++;
 202         poll = pde->proc_fops->poll;
 203         spin_unlock(&pde->pde_unload_lock);
 204
 205         if (poll)
 206                 rv = poll(file, pts);
 207
 208         pde_users_dec(pde);
 209         return rv;
 210 }
 211
 212 static long proc_reg_unlocked_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
 213 {
 214         struct proc_dir_entry *pde = PDE(file->f_path.dentry->d_inode);
 215         long rv = -ENOTTY;
 216         long (*unlocked_ioctl)(struct file *, unsigned int, unsigned long);
 217         int (*ioctl)(struct inode *, struct file *, unsigned int, unsigned long);
 218
 219         spin_lock(&pde->pde_unload_lock);
 220         if (!pde->proc_fops) {
 221                 spin_unlock(&pde->pde_unload_lock);
 222                 return rv;
 223         }
 224         pde->pde_users++;
 225         unlocked_ioctl = pde->proc_fops->unlocked_ioctl;
 226         ioctl = pde->proc_fops->ioctl;
 227         spin_unlock(&pde->pde_unload_lock);
 228
 229         if (unlocked_ioctl) {
 230                 rv = unlocked_ioctl(file, cmd, arg);
 231                 if (rv == -ENOIOCTLCMD)
 232                         rv = -EINVAL;
 233         } else if (ioctl) {
 234                 lock_kernel();
 235                 rv = ioctl(file->f_path.dentry->d_inode, file, cmd, arg);
 236                 unlock_kernel();
 237         }
 238
 239         pde_users_dec(pde);
 240         return rv;
 241 }
 242
 243 #ifdef CONFIG_COMPAT
 244 static long proc_reg_compat_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
 245 {
 246         struct proc_dir_entry *pde = PDE(file->f_path.dentry->d_inode);
 247         long rv = -ENOTTY;
 248         long (*compat_ioctl)(struct file *, unsigned int, unsigned long);
 249
 250         spin_lock(&pde->pde_unload_lock);
 251         if (!pde->proc_fops) {
 252                 spin_unlock(&pde->pde_unload_lock);
 253                 return rv;
 254         }
 255         pde->pde_users++;
 256         compat_ioctl = pde->proc_fops->compat_ioctl;
 257         spin_unlock(&pde->pde_unload_lock);
 258
 259         if (compat_ioctl)
 260                 rv = compat_ioctl(file, cmd, arg);
 261
 262         pde_users_dec(pde);
 263         return rv;
 264 }
 265 #endif
 266
 267 static int proc_reg_mmap(struct file *file, struct vm_area_struct *vma)
 268 {
 269         struct proc_dir_entry *pde = PDE(file->f_path.dentry->d_inode);
 270         int rv = -EIO;
 271         int (*mmap)(struct file *, struct vm_area_struct *);
 272
 273         spin_lock(&pde->pde_unload_lock);
 274         if (!pde->proc_fops) {
 275                 spin_unlock(&pde->pde_unload_lock);
 276                 return rv;
 277         }
 278         pde->pde_users++;
 279         mmap = pde->proc_fops->mmap;
 280         spin_unlock(&pde->pde_unload_lock);
 281
 282         if (mmap)
 283                 rv = mmap(file, vma);
 284
 285         pde_users_dec(pde);
 286         return rv;
 287 }
 288
 289 static int proc_reg_open(struct inode *inode, struct file *file)
 290 {
 291         struct proc_dir_entry *pde = PDE(inode);
 292         int rv = 0;
 293         int (*open)(struct inode *, struct file *);
 294         int (*release)(struct inode *, struct file *);
 295         struct pde_opener *pdeo;
 296
 297         /*
 298          * What for, you ask? Well, we can have open, rmmod, remove_proc_entry
 299          * sequence. ->release won't be called because ->proc_fops will be
 300          * cleared. Depending on complexity of ->release, consequences vary.
 301          *
 302          * We can't wait for mercy when close will be done for real, it's
 303          * deadlockable: rmmod foo </proc/foo . So, we're going to do ->release
 304          * by hand in remove_proc_entry(). For this, save opener's credentials
 305          * for later.
 306          */
 307         pdeo = kmalloc(sizeof(struct pde_opener), GFP_KERNEL);
 308         if (!pdeo)
 309                 return -ENOMEM;
 310
 311         spin_lock(&pde->pde_unload_lock);
 312         if (!pde->proc_fops) {
 313                 spin_unlock(&pde->pde_unload_lock);
 314                 kfree(pdeo);
 315                 return -EINVAL;
 316         }
 317         pde->pde_users++;
 318         open = pde->proc_fops->open;
 319         release = pde->proc_fops->release;
 320         spin_unlock(&pde->pde_unload_lock);
 321
 322         if (open)
 323                 rv = open(inode, file);
 324
 325         spin_lock(&pde->pde_unload_lock);
 326         if (rv == 0 && release) {
 327                 /* To know what to release. */
 328                 pdeo->inode = inode;
 329                 pdeo->file = file;
 330                 /* Strictly for "too late" ->release in proc_reg_release(). */
 331                 pdeo->release = release;
 332                 list_add(&pdeo->lh, &pde->pde_openers);
 333         } else
 334                 kfree(pdeo);
 335         __pde_users_dec(pde);
 336         spin_unlock(&pde->pde_unload_lock);
 337         return rv;
 338 }
 339
 340 static struct pde_opener *find_pde_opener(struct proc_dir_entry *pde,
 341                                         struct inode *inode, struct file *file)
 342 {
 343         struct pde_opener *pdeo;
 344
 345         list_for_each_entry(pdeo, &pde->pde_openers, lh) {
 346                 if (pdeo->inode == inode && pdeo->file == file)
 347                         return pdeo;
 348         }
 349         return NULL;
 350 }
 351
 352 static int proc_reg_release(struct inode *inode, struct file *file)
 353 {
 354         struct proc_dir_entry *pde = PDE(inode);
 355         int rv = 0;
 356         int (*release)(struct inode *, struct file *);
 357         struct pde_opener *pdeo;
 358
 359         spin_lock(&pde->pde_unload_lock);
 360         pdeo = find_pde_opener(pde, inode, file);
 361         if (!pde->proc_fops) {
 362                 /*
 363                  * Can't simply exit, __fput() will think that everything is OK,
 364                  * and move on to freeing struct file. remove_proc_entry() will
 365                  * find slacker in opener's list and will try to do non-trivial
 366                  * things with struct file. Therefore, remove opener from list.
 367                  *
 368                  * But if opener is removed from list, who will ->release it?
 369                  */
 370                 if (pdeo) {
 371                         list_del(&pdeo->lh);
 372                         spin_unlock(&pde->pde_unload_lock);
 373                         rv = pdeo->release(inode, file);
 374                         kfree(pdeo);
 375                 } else
 376                         spin_unlock(&pde->pde_unload_lock);
 377                 return rv;
 378         }
 379         pde->pde_users++;
 380         release = pde->proc_fops->release;
 381         if (pdeo) {
 382                 list_del(&pdeo->lh);
 383                 kfree(pdeo);
 384         }
 385         spin_unlock(&pde->pde_unload_lock);
 386
 387         if (release)
 388                 rv = release(inode, file);
 389
 390         pde_users_dec(pde);
 391         return rv;
 392 }
 393
 394 static const struct file_operations proc_reg_file_ops = {
 395         .llseek         = proc_reg_llseek,
 396         .read           = proc_reg_read,
 397         .write          = proc_reg_write,
 398         .poll           = proc_reg_poll,
 399         .unlocked_ioctl = proc_reg_unlocked_ioctl,
 400 #ifdef CONFIG_COMPAT
 401         .compat_ioctl   = proc_reg_compat_ioctl,
 402 #endif
 403         .mmap           = proc_reg_mmap,
 404         .open           = proc_reg_open,
 405         .release        = proc_reg_release,
 406 };
 407
 408 #ifdef CONFIG_COMPAT
 409 static const struct file_operations proc_reg_file_ops_no_compat = {
 410         .llseek         = proc_reg_llseek,
 411         .read           = proc_reg_read,
 412         .write          = proc_reg_write,
 413         .poll           = proc_reg_poll,
 414         .unlocked_ioctl = proc_reg_unlocked_ioctl,
 415         .mmap           = proc_reg_mmap,
 416         .open           = proc_reg_open,
 417         .release        = proc_reg_release,
 418 };
 419 #endif
 420
 421 struct inode *proc_get_inode(struct super_block *sb, unsigned int ino,
 422                                 struct proc_dir_entry *de)
 423 {
 424         struct inode * inode;
 425
 426         inode = iget_locked(sb, ino);
 427         if (!inode)
 428                 return NULL;
 429         if (inode->i_state & I_NEW) {
 430                 inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
 431                 PROC_I(inode)->fd = 0;
 432                 PROC_I(inode)->pde = de;
 433
 434                 if (de->mode) {
 435                         inode->i_mode = de->mode;
 436                         inode->i_uid = de->uid;
 437                         inode->i_gid = de->gid;
 438                 }
 439                 if (de->size)
 440                         inode->i_size = de->size;
 441                 if (de->nlink)
 442                         inode->i_nlink = de->nlink;
 443                 if (de->proc_iops)
 444                         inode->i_op = de->proc_iops;
 445                 if (de->proc_fops) {
 446                         if (S_ISREG(inode->i_mode)) {
 447 #ifdef CONFIG_COMPAT
 448                                 if (!de->proc_fops->compat_ioctl)
 449                                         inode->i_fop =
 450                                                 &proc_reg_file_ops_no_compat;
 451                                 else
 452 #endif
 453                                         inode->i_fop = &proc_reg_file_ops;
 454                         } else {
 455                                 inode->i_fop = de->proc_fops;
 456                         }
 457                 }
 458                 unlock_new_inode(inode);
 459         } else
 460                pde_put(de);
 461         return inode;
 462 }
 463
 464 int proc_fill_super(struct super_block *s)
 465 {
 466         struct inode * root_inode;
 467
 468         s->s_flags |= MS_NODIRATIME | MS_NOSUID | MS_NOEXEC;
 469         s->s_blocksize = 1024;
 470         s->s_blocksize_bits = 10;
 471         s->s_magic = PROC_SUPER_MAGIC;
 472         s->s_op = &proc_sops;
 473         s->s_time_gran = 1;
 474
 475         pde_get(&proc_root);
 476         root_inode = proc_get_inode(s, PROC_ROOT_INO, &proc_root);
 477         if (!root_inode)
 478                 goto out_no_root;
 479         root_inode->i_uid = 0;
 480         root_inode->i_gid = 0;
 481         s->s_root = d_alloc_root(root_inode);
 482         if (!s->s_root)
 483                 goto out_no_root;
 484         return 0;
 485
 486 out_no_root:
 487         printk("proc_read_super: get root inode failed\n");
 488         iput(root_inode);
 489         pde_put(&proc_root);
 490         return -ENOMEM;
 491 }