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 struct proc_dir_entry *de_get(struct proc_dir_entry *de)
  28 {
  29         atomic_inc(&de->count);
  30         return de;
  31 }
  32
  33 /*
  34  * Decrements the use count and checks for deferred deletion.
  35  */
  36 void de_put(struct proc_dir_entry *de)
  37 {
  38         lock_kernel();
  39         if (!atomic_read(&de->count)) {
  40                 printk("de_put: entry %s already free!\n", de->name);
  41                 unlock_kernel();
  42                 return;
  43         }
  44
  45         if (atomic_dec_and_test(&de->count))
  46                 free_proc_entry(de);
  47         unlock_kernel();
  48 }
  49
  50 /*
  51  * Decrement the use count of the proc_dir_entry.
  52  */
  53 static void proc_delete_inode(struct inode *inode)
  54 {
  55         struct proc_dir_entry *de;
  56
  57         truncate_inode_pages(&inode->i_data, 0);
  58
  59         /* Stop tracking associated processes */
  60         put_pid(PROC_I(inode)->pid);
  61
  62         /* Let go of any associated proc directory entry */
  63         de = PROC_I(inode)->pde;
  64         if (de) {
  65                 if (de->owner)
  66                         module_put(de->owner);
  67                 de_put(de);
  68         }
  69         if (PROC_I(inode)->sysctl)
  70                 sysctl_head_put(PROC_I(inode)->sysctl);
  71         clear_inode(inode);
  72 }
  73
  74 struct vfsmount *proc_mnt;
  75
  76 static struct kmem_cache * proc_inode_cachep;
  77
  78 static struct inode *proc_alloc_inode(struct super_block *sb)
  79 {
  80         struct proc_inode *ei;
  81         struct inode *inode;
  82
  83         ei = (struct proc_inode *)kmem_cache_alloc(proc_inode_cachep, GFP_KERNEL);
  84         if (!ei)
  85                 return NULL;
  86         ei->pid = NULL;
  87         ei->fd = 0;
  88         ei->op.proc_get_link = NULL;
  89         ei->pde = NULL;
  90         ei->sysctl = NULL;
  91         ei->sysctl_entry = NULL;
  92         inode = &ei->vfs_inode;
  93         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
  94         return inode;
  95 }
  96
  97 static void proc_destroy_inode(struct inode *inode)
  98 {
  99         kmem_cache_free(proc_inode_cachep, PROC_I(inode));
 100 }
 101
 102 static void init_once(void *foo)
 103 {
 104         struct proc_inode *ei = (struct proc_inode *) foo;
 105
 106         inode_init_once(&ei->vfs_inode);
 107 }
 108
 109 void __init proc_init_inodecache(void)
 110 {
 111         proc_inode_cachep = kmem_cache_create("proc_inode_cache",
 112                                              sizeof(struct proc_inode),
 113                                              0, (SLAB_RECLAIM_ACCOUNT|
 114                                                 SLAB_MEM_SPREAD|SLAB_PANIC),
 115                                              init_once);
 116 }
 117
 118 static const struct super_operations proc_sops = {
 119         .alloc_inode    = proc_alloc_inode,
 120         .destroy_inode  = proc_destroy_inode,
 121         .drop_inode     = generic_delete_inode,
 122         .delete_inode   = proc_delete_inode,
 123         .statfs         = simple_statfs,
 124 };
 125
 126 static void __pde_users_dec(struct proc_dir_entry *pde)
 127 {
 128         pde->pde_users--;
 129         if (pde->pde_unload_completion && pde->pde_users == 0)
 130                 complete(pde->pde_unload_completion);
 131 }
 132
 133 static void pde_users_dec(struct proc_dir_entry *pde)
 134 {
 135         spin_lock(&pde->pde_unload_lock);
 136         __pde_users_dec(pde);
 137         spin_unlock(&pde->pde_unload_lock);
 138 }
 139
 140 static loff_t proc_reg_llseek(struct file *file, loff_t offset, int whence)
 141 {
 142         struct proc_dir_entry *pde = PDE(file->f_path.dentry->d_inode);
 143         loff_t rv = -EINVAL;
 144         loff_t (*llseek)(struct file *, loff_t, int);
 145
 146         spin_lock(&pde->pde_unload_lock);
 147         /*
 148          * remove_proc_entry() is going to delete PDE (as part of module
 149          * cleanup sequence). No new callers into module allowed.
 150          */
 151         if (!pde->proc_fops) {
 152                 spin_unlock(&pde->pde_unload_lock);
 153                 return rv;
 154         }
 155         /*
 156          * Bump refcount so that remove_proc_entry will wail for ->llseek to
 157          * complete.
 158          */
 159         pde->pde_users++;
 160         /*
 161          * Save function pointer under lock, to protect against ->proc_fops
 162          * NULL'ifying right after ->pde_unload_lock is dropped.
 163          */
 164         llseek = pde->proc_fops->llseek;
 165         spin_unlock(&pde->pde_unload_lock);
 166
 167         if (!llseek)
 168                 llseek = default_llseek;
 169         rv = llseek(file, offset, whence);
 170
 171         pde_users_dec(pde);
 172         return rv;
 173 }
 174
 175 static ssize_t proc_reg_read(struct file *file, 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 (*read)(struct file *, 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         read = pde->proc_fops->read;
 188         spin_unlock(&pde->pde_unload_lock);
 189
 190         if (read)
 191                 rv = read(file, buf, count, ppos);
 192
 193         pde_users_dec(pde);
 194         return rv;
 195 }
 196
 197 static ssize_t proc_reg_write(struct file *file, const char __user *buf, size_t count, loff_t *ppos)
 198 {
 199         struct proc_dir_entry *pde = PDE(file->f_path.dentry->d_inode);
 200         ssize_t rv = -EIO;
 201         ssize_t (*write)(struct file *, const char __user *, size_t, loff_t *);
 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         write = pde->proc_fops->write;
 210         spin_unlock(&pde->pde_unload_lock);
 211
 212         if (write)
 213                 rv = write(file, buf, count, ppos);
 214
 215         pde_users_dec(pde);
 216         return rv;
 217 }
 218
 219 static unsigned int proc_reg_poll(struct file *file, struct poll_table_struct *pts)
 220 {
 221         struct proc_dir_entry *pde = PDE(file->f_path.dentry->d_inode);
 222         unsigned int rv = DEFAULT_POLLMASK;
 223         unsigned int (*poll)(struct file *, struct poll_table_struct *);
 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         poll = pde->proc_fops->poll;
 232         spin_unlock(&pde->pde_unload_lock);
 233
 234         if (poll)
 235                 rv = poll(file, pts);
 236
 237         pde_users_dec(pde);
 238         return rv;
 239 }
 240
 241 static long proc_reg_unlocked_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
 242 {
 243         struct proc_dir_entry *pde = PDE(file->f_path.dentry->d_inode);
 244         long rv = -ENOTTY;
 245         long (*unlocked_ioctl)(struct file *, unsigned int, unsigned long);
 246         int (*ioctl)(struct inode *, 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         unlocked_ioctl = pde->proc_fops->unlocked_ioctl;
 255         ioctl = pde->proc_fops->ioctl;
 256         spin_unlock(&pde->pde_unload_lock);
 257
 258         if (unlocked_ioctl) {
 259                 rv = unlocked_ioctl(file, cmd, arg);
 260                 if (rv == -ENOIOCTLCMD)
 261                         rv = -EINVAL;
 262         } else if (ioctl) {
 263                 lock_kernel();
 264                 rv = ioctl(file->f_path.dentry->d_inode, file, cmd, arg);
 265                 unlock_kernel();
 266         }
 267
 268         pde_users_dec(pde);
 269         return rv;
 270 }
 271
 272 #ifdef CONFIG_COMPAT
 273 static long proc_reg_compat_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
 274 {
 275         struct proc_dir_entry *pde = PDE(file->f_path.dentry->d_inode);
 276         long rv = -ENOTTY;
 277         long (*compat_ioctl)(struct file *, unsigned int, unsigned long);
 278
 279         spin_lock(&pde->pde_unload_lock);
 280         if (!pde->proc_fops) {
 281                 spin_unlock(&pde->pde_unload_lock);
 282                 return rv;
 283         }
 284         pde->pde_users++;
 285         compat_ioctl = pde->proc_fops->compat_ioctl;
 286         spin_unlock(&pde->pde_unload_lock);
 287
 288         if (compat_ioctl)
 289                 rv = compat_ioctl(file, cmd, arg);
 290
 291         pde_users_dec(pde);
 292         return rv;
 293 }
 294 #endif
 295
 296 static int proc_reg_mmap(struct file *file, struct vm_area_struct *vma)
 297 {
 298         struct proc_dir_entry *pde = PDE(file->f_path.dentry->d_inode);
 299         int rv = -EIO;
 300         int (*mmap)(struct file *, struct vm_area_struct *);
 301
 302         spin_lock(&pde->pde_unload_lock);
 303         if (!pde->proc_fops) {
 304                 spin_unlock(&pde->pde_unload_lock);
 305                 return rv;
 306         }
 307         pde->pde_users++;
 308         mmap = pde->proc_fops->mmap;
 309         spin_unlock(&pde->pde_unload_lock);
 310
 311         if (mmap)
 312                 rv = mmap(file, vma);
 313
 314         pde_users_dec(pde);
 315         return rv;
 316 }
 317
 318 static int proc_reg_open(struct inode *inode, struct file *file)
 319 {
 320         struct proc_dir_entry *pde = PDE(inode);
 321         int rv = 0;
 322         int (*open)(struct inode *, struct file *);
 323         int (*release)(struct inode *, struct file *);
 324         struct pde_opener *pdeo;
 325
 326         /*
 327          * What for, you ask? Well, we can have open, rmmod, remove_proc_entry
 328          * sequence. ->release won't be called because ->proc_fops will be
 329          * cleared. Depending on complexity of ->release, consequences vary.
 330          *
 331          * We can't wait for mercy when close will be done for real, it's
 332          * deadlockable: rmmod foo </proc/foo . So, we're going to do ->release
 333          * by hand in remove_proc_entry(). For this, save opener's credentials
 334          * for later.
 335          */
 336         pdeo = kmalloc(sizeof(struct pde_opener), GFP_KERNEL);
 337         if (!pdeo)
 338                 return -ENOMEM;
 339
 340         spin_lock(&pde->pde_unload_lock);
 341         if (!pde->proc_fops) {
 342                 spin_unlock(&pde->pde_unload_lock);
 343                 kfree(pdeo);
 344                 return -EINVAL;
 345         }
 346         pde->pde_users++;
 347         open = pde->proc_fops->open;
 348         release = pde->proc_fops->release;
 349         spin_unlock(&pde->pde_unload_lock);
 350
 351         if (open)
 352                 rv = open(inode, file);
 353
 354         spin_lock(&pde->pde_unload_lock);
 355         if (rv == 0 && release) {
 356                 /* To know what to release. */
 357                 pdeo->inode = inode;
 358                 pdeo->file = file;
 359                 /* Strictly for "too late" ->release in proc_reg_release(). */
 360                 pdeo->release = release;
 361                 list_add(&pdeo->lh, &pde->pde_openers);
 362         } else
 363                 kfree(pdeo);
 364         __pde_users_dec(pde);
 365         spin_unlock(&pde->pde_unload_lock);
 366         return rv;
 367 }
 368
 369 static struct pde_opener *find_pde_opener(struct proc_dir_entry *pde,
 370                                         struct inode *inode, struct file *file)
 371 {
 372         struct pde_opener *pdeo;
 373
 374         list_for_each_entry(pdeo, &pde->pde_openers, lh) {
 375                 if (pdeo->inode == inode && pdeo->file == file)
 376                         return pdeo;
 377         }
 378         return NULL;
 379 }
 380
 381 static int proc_reg_release(struct inode *inode, struct file *file)
 382 {
 383         struct proc_dir_entry *pde = PDE(inode);
 384         int rv = 0;
 385         int (*release)(struct inode *, struct file *);
 386         struct pde_opener *pdeo;
 387
 388         spin_lock(&pde->pde_unload_lock);
 389         pdeo = find_pde_opener(pde, inode, file);
 390         if (!pde->proc_fops) {
 391                 /*
 392                  * Can't simply exit, __fput() will think that everything is OK,
 393                  * and move on to freeing struct file. remove_proc_entry() will
 394                  * find slacker in opener's list and will try to do non-trivial
 395                  * things with struct file. Therefore, remove opener from list.
 396                  *
 397                  * But if opener is removed from list, who will ->release it?
 398                  */
 399                 if (pdeo) {
 400                         list_del(&pdeo->lh);
 401                         spin_unlock(&pde->pde_unload_lock);
 402                         rv = pdeo->release(inode, file);
 403                         kfree(pdeo);
 404                 } else
 405                         spin_unlock(&pde->pde_unload_lock);
 406                 return rv;
 407         }
 408         pde->pde_users++;
 409         release = pde->proc_fops->release;
 410         if (pdeo) {
 411                 list_del(&pdeo->lh);
 412                 kfree(pdeo);
 413         }
 414         spin_unlock(&pde->pde_unload_lock);
 415
 416         if (release)
 417                 rv = release(inode, file);
 418
 419         pde_users_dec(pde);
 420         return rv;
 421 }
 422
 423 static const struct file_operations proc_reg_file_ops = {
 424         .llseek         = proc_reg_llseek,
 425         .read           = proc_reg_read,
 426         .write          = proc_reg_write,
 427         .poll           = proc_reg_poll,
 428         .unlocked_ioctl = proc_reg_unlocked_ioctl,
 429 #ifdef CONFIG_COMPAT
 430         .compat_ioctl   = proc_reg_compat_ioctl,
 431 #endif
 432         .mmap           = proc_reg_mmap,
 433         .open           = proc_reg_open,
 434         .release        = proc_reg_release,
 435 };
 436
 437 #ifdef CONFIG_COMPAT
 438 static const struct file_operations proc_reg_file_ops_no_compat = {
 439         .llseek         = proc_reg_llseek,
 440         .read           = proc_reg_read,
 441         .write          = proc_reg_write,
 442         .poll           = proc_reg_poll,
 443         .unlocked_ioctl = proc_reg_unlocked_ioctl,
 444         .mmap           = proc_reg_mmap,
 445         .open           = proc_reg_open,
 446         .release        = proc_reg_release,
 447 };
 448 #endif
 449
 450 struct inode *proc_get_inode(struct super_block *sb, unsigned int ino,
 451                                 struct proc_dir_entry *de)
 452 {
 453         struct inode * inode;
 454
 455         if (!try_module_get(de->owner))
 456                 goto out_mod;
 457
 458         inode = iget_locked(sb, ino);
 459         if (!inode)
 460                 goto out_ino;
 461         if (inode->i_state & I_NEW) {
 462                 inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
 463                 PROC_I(inode)->fd = 0;
 464                 PROC_I(inode)->pde = de;
 465
 466                 if (de->mode) {
 467                         inode->i_mode = de->mode;
 468                         inode->i_uid = de->uid;
 469                         inode->i_gid = de->gid;
 470                 }
 471                 if (de->size)
 472                         inode->i_size = de->size;
 473                 if (de->nlink)
 474                         inode->i_nlink = de->nlink;
 475                 if (de->proc_iops)
 476                         inode->i_op = de->proc_iops;
 477                 if (de->proc_fops) {
 478                         if (S_ISREG(inode->i_mode)) {
 479 #ifdef CONFIG_COMPAT
 480                                 if (!de->proc_fops->compat_ioctl)
 481                                         inode->i_fop =
 482                                                 &proc_reg_file_ops_no_compat;
 483                                 else
 484 #endif
 485                                         inode->i_fop = &proc_reg_file_ops;
 486                         } else {
 487                                 inode->i_fop = de->proc_fops;
 488                         }
 489                 }
 490                 unlock_new_inode(inode);
 491         } else
 492                module_put(de->owner);
 493         return inode;
 494
 495 out_ino:
 496         module_put(de->owner);
 497 out_mod:
 498         return NULL;
 499 }
 500
 501 int proc_fill_super(struct super_block *s)
 502 {
 503         struct inode * root_inode;
 504
 505         s->s_flags |= MS_NODIRATIME | MS_NOSUID | MS_NOEXEC;
 506         s->s_blocksize = 1024;
 507         s->s_blocksize_bits = 10;
 508         s->s_magic = PROC_SUPER_MAGIC;
 509         s->s_op = &proc_sops;
 510         s->s_time_gran = 1;
 511
 512         de_get(&proc_root);
 513         root_inode = proc_get_inode(s, PROC_ROOT_INO, &proc_root);
 514         if (!root_inode)
 515                 goto out_no_root;
 516         root_inode->i_uid = 0;
 517         root_inode->i_gid = 0;
 518         s->s_root = d_alloc_root(root_inode);
 519         if (!s->s_root)
 520                 goto out_no_root;
 521         return 0;
 522
 523 out_no_root:
 524         printk("proc_read_super: get root inode failed\n");
 525         iput(root_inode);
 526         de_put(&proc_root);
 527         return -ENOMEM;
 528 }