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_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         .evict_inode    = proc_evict_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 (*ioctl)(struct file *, unsigned int, unsigned long);
 217
 218         spin_lock(&pde->pde_unload_lock);
 219         if (!pde->proc_fops) {
 220                 spin_unlock(&pde->pde_unload_lock);
 221                 return rv;
 222         }
 223         pde->pde_users++;
 224         ioctl = pde->proc_fops->unlocked_ioctl;
 225         spin_unlock(&pde->pde_unload_lock);
 226
 227         if (ioctl)
 228                 rv = ioctl(file, cmd, arg);
 229
 230         pde_users_dec(pde);
 231         return rv;
 232 }
 233
 234 #ifdef CONFIG_COMPAT
 235 static long proc_reg_compat_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
 236 {
 237         struct proc_dir_entry *pde = PDE(file->f_path.dentry->d_inode);
 238         long rv = -ENOTTY;
 239         long (*compat_ioctl)(struct file *, unsigned int, unsigned long);
 240
 241         spin_lock(&pde->pde_unload_lock);
 242         if (!pde->proc_fops) {
 243                 spin_unlock(&pde->pde_unload_lock);
 244                 return rv;
 245         }
 246         pde->pde_users++;
 247         compat_ioctl = pde->proc_fops->compat_ioctl;
 248         spin_unlock(&pde->pde_unload_lock);
 249
 250         if (compat_ioctl)
 251                 rv = compat_ioctl(file, cmd, arg);
 252
 253         pde_users_dec(pde);
 254         return rv;
 255 }
 256 #endif
 257
 258 static int proc_reg_mmap(struct file *file, struct vm_area_struct *vma)
 259 {
 260         struct proc_dir_entry *pde = PDE(file->f_path.dentry->d_inode);
 261         int rv = -EIO;
 262         int (*mmap)(struct file *, struct vm_area_struct *);
 263
 264         spin_lock(&pde->pde_unload_lock);
 265         if (!pde->proc_fops) {
 266                 spin_unlock(&pde->pde_unload_lock);
 267                 return rv;
 268         }
 269         pde->pde_users++;
 270         mmap = pde->proc_fops->mmap;
 271         spin_unlock(&pde->pde_unload_lock);
 272
 273         if (mmap)
 274                 rv = mmap(file, vma);
 275
 276         pde_users_dec(pde);
 277         return rv;
 278 }
 279
 280 static int proc_reg_open(struct inode *inode, struct file *file)
 281 {
 282         struct proc_dir_entry *pde = PDE(inode);
 283         int rv = 0;
 284         int (*open)(struct inode *, struct file *);
 285         int (*release)(struct inode *, struct file *);
 286         struct pde_opener *pdeo;
 287
 288         /*
 289          * What for, you ask? Well, we can have open, rmmod, remove_proc_entry
 290          * sequence. ->release won't be called because ->proc_fops will be
 291          * cleared. Depending on complexity of ->release, consequences vary.
 292          *
 293          * We can't wait for mercy when close will be done for real, it's
 294          * deadlockable: rmmod foo </proc/foo . So, we're going to do ->release
 295          * by hand in remove_proc_entry(). For this, save opener's credentials
 296          * for later.
 297          */
 298         pdeo = kmalloc(sizeof(struct pde_opener), GFP_KERNEL);
 299         if (!pdeo)
 300                 return -ENOMEM;
 301
 302         spin_lock(&pde->pde_unload_lock);
 303         if (!pde->proc_fops) {
 304                 spin_unlock(&pde->pde_unload_lock);
 305                 kfree(pdeo);
 306                 return -EINVAL;
 307         }
 308         pde->pde_users++;
 309         open = pde->proc_fops->open;
 310         release = pde->proc_fops->release;
 311         spin_unlock(&pde->pde_unload_lock);
 312
 313         if (open)
 314                 rv = open(inode, file);
 315
 316         spin_lock(&pde->pde_unload_lock);
 317         if (rv == 0 && release) {
 318                 /* To know what to release. */
 319                 pdeo->inode = inode;
 320                 pdeo->file = file;
 321                 /* Strictly for "too late" ->release in proc_reg_release(). */
 322                 pdeo->release = release;
 323                 list_add(&pdeo->lh, &pde->pde_openers);
 324         } else
 325                 kfree(pdeo);
 326         __pde_users_dec(pde);
 327         spin_unlock(&pde->pde_unload_lock);
 328         return rv;
 329 }
 330
 331 static struct pde_opener *find_pde_opener(struct proc_dir_entry *pde,
 332                                         struct inode *inode, struct file *file)
 333 {
 334         struct pde_opener *pdeo;
 335
 336         list_for_each_entry(pdeo, &pde->pde_openers, lh) {
 337                 if (pdeo->inode == inode && pdeo->file == file)
 338                         return pdeo;
 339         }
 340         return NULL;
 341 }
 342
 343 static int proc_reg_release(struct inode *inode, struct file *file)
 344 {
 345         struct proc_dir_entry *pde = PDE(inode);
 346         int rv = 0;
 347         int (*release)(struct inode *, struct file *);
 348         struct pde_opener *pdeo;
 349
 350         spin_lock(&pde->pde_unload_lock);
 351         pdeo = find_pde_opener(pde, inode, file);
 352         if (!pde->proc_fops) {
 353                 /*
 354                  * Can't simply exit, __fput() will think that everything is OK,
 355                  * and move on to freeing struct file. remove_proc_entry() will
 356                  * find slacker in opener's list and will try to do non-trivial
 357                  * things with struct file. Therefore, remove opener from list.
 358                  *
 359                  * But if opener is removed from list, who will ->release it?
 360                  */
 361                 if (pdeo) {
 362                         list_del(&pdeo->lh);
 363                         spin_unlock(&pde->pde_unload_lock);
 364                         rv = pdeo->release(inode, file);
 365                         kfree(pdeo);
 366                 } else
 367                         spin_unlock(&pde->pde_unload_lock);
 368                 return rv;
 369         }
 370         pde->pde_users++;
 371         release = pde->proc_fops->release;
 372         if (pdeo) {
 373                 list_del(&pdeo->lh);
 374                 kfree(pdeo);
 375         }
 376         spin_unlock(&pde->pde_unload_lock);
 377
 378         if (release)
 379                 rv = release(inode, file);
 380
 381         pde_users_dec(pde);
 382         return rv;
 383 }
 384
 385 static const struct file_operations proc_reg_file_ops = {
 386         .llseek         = proc_reg_llseek,
 387         .read           = proc_reg_read,
 388         .write          = proc_reg_write,
 389         .poll           = proc_reg_poll,
 390         .unlocked_ioctl = proc_reg_unlocked_ioctl,
 391 #ifdef CONFIG_COMPAT
 392         .compat_ioctl   = proc_reg_compat_ioctl,
 393 #endif
 394         .mmap           = proc_reg_mmap,
 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         .open           = proc_reg_open,
 408         .release        = proc_reg_release,
 409 };
 410 #endif
 411
 412 struct inode *proc_get_inode(struct super_block *sb, unsigned int ino,
 413                                 struct proc_dir_entry *de)
 414 {
 415         struct inode * inode;
 416
 417         inode = iget_locked(sb, ino);
 418         if (!inode)
 419                 return NULL;
 420         if (inode->i_state & I_NEW) {
 421                 inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
 422                 PROC_I(inode)->fd = 0;
 423                 PROC_I(inode)->pde = de;
 424
 425                 if (de->mode) {
 426                         inode->i_mode = de->mode;
 427                         inode->i_uid = de->uid;
 428                         inode->i_gid = de->gid;
 429                 }
 430                 if (de->size)
 431                         inode->i_size = de->size;
 432                 if (de->nlink)
 433                         inode->i_nlink = de->nlink;
 434                 if (de->proc_iops)
 435                         inode->i_op = de->proc_iops;
 436                 if (de->proc_fops) {
 437                         if (S_ISREG(inode->i_mode)) {
 438 #ifdef CONFIG_COMPAT
 439                                 if (!de->proc_fops->compat_ioctl)
 440                                         inode->i_fop =
 441                                                 &proc_reg_file_ops_no_compat;
 442                                 else
 443 #endif
 444                                         inode->i_fop = &proc_reg_file_ops;
 445                         } else {
 446                                 inode->i_fop = de->proc_fops;
 447                         }
 448                 }
 449                 unlock_new_inode(inode);
 450         } else
 451                pde_put(de);
 452         return inode;
 453 }
 454
 455 int proc_fill_super(struct super_block *s)
 456 {
 457         struct inode * root_inode;
 458
 459         s->s_flags |= MS_NODIRATIME | MS_NOSUID | MS_NOEXEC;
 460         s->s_blocksize = 1024;
 461         s->s_blocksize_bits = 10;
 462         s->s_magic = PROC_SUPER_MAGIC;
 463         s->s_op = &proc_sops;
 464         s->s_time_gran = 1;
 465
 466         pde_get(&proc_root);
 467         root_inode = proc_get_inode(s, PROC_ROOT_INO, &proc_root);
 468         if (!root_inode)
 469                 goto out_no_root;
 470         root_inode->i_uid = 0;
 471         root_inode->i_gid = 0;
 472         s->s_root = d_alloc_root(root_inode);
 473         if (!s->s_root)
 474                 goto out_no_root;
 475         return 0;
 476
 477 out_no_root:
 478         printk("proc_read_super: get root inode failed\n");
 479         iput(root_inode);
 480         pde_put(&proc_root);
 481         return -ENOMEM;
 482 }