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 int __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         return 0;
 117 }
 118
 119 static const struct super_operations proc_sops = {
 120         .alloc_inode    = proc_alloc_inode,
 121         .destroy_inode  = proc_destroy_inode,
 122         .drop_inode     = generic_delete_inode,
 123         .delete_inode   = proc_delete_inode,
 124         .statfs         = simple_statfs,
 125 };
 126
 127 static void __pde_users_dec(struct proc_dir_entry *pde)
 128 {
 129         pde->pde_users--;
 130         if (pde->pde_unload_completion && pde->pde_users == 0)
 131                 complete(pde->pde_unload_completion);
 132 }
 133
 134 static void pde_users_dec(struct proc_dir_entry *pde)
 135 {
 136         spin_lock(&pde->pde_unload_lock);
 137         __pde_users_dec(pde);
 138         spin_unlock(&pde->pde_unload_lock);
 139 }
 140
 141 static loff_t proc_reg_llseek(struct file *file, loff_t offset, int whence)
 142 {
 143         struct proc_dir_entry *pde = PDE(file->f_path.dentry->d_inode);
 144         loff_t rv = -EINVAL;
 145         loff_t (*llseek)(struct file *, loff_t, int);
 146
 147         spin_lock(&pde->pde_unload_lock);
 148         /*
 149          * remove_proc_entry() is going to delete PDE (as part of module
 150          * cleanup sequence). No new callers into module allowed.
 151          */
 152         if (!pde->proc_fops) {
 153                 spin_unlock(&pde->pde_unload_lock);
 154                 return rv;
 155         }
 156         /*
 157          * Bump refcount so that remove_proc_entry will wail for ->llseek to
 158          * complete.
 159          */
 160         pde->pde_users++;
 161         /*
 162          * Save function pointer under lock, to protect against ->proc_fops
 163          * NULL'ifying right after ->pde_unload_lock is dropped.
 164          */
 165         llseek = pde->proc_fops->llseek;
 166         spin_unlock(&pde->pde_unload_lock);
 167
 168         if (!llseek)
 169                 llseek = default_llseek;
 170         rv = llseek(file, offset, whence);
 171
 172         pde_users_dec(pde);
 173         return rv;
 174 }
 175
 176 static ssize_t proc_reg_read(struct file *file, char __user *buf, size_t count, loff_t *ppos)
 177 {
 178         struct proc_dir_entry *pde = PDE(file->f_path.dentry->d_inode);
 179         ssize_t rv = -EIO;
 180         ssize_t (*read)(struct file *, char __user *, size_t, loff_t *);
 181
 182         spin_lock(&pde->pde_unload_lock);
 183         if (!pde->proc_fops) {
 184                 spin_unlock(&pde->pde_unload_lock);
 185                 return rv;
 186         }
 187         pde->pde_users++;
 188         read = pde->proc_fops->read;
 189         spin_unlock(&pde->pde_unload_lock);
 190
 191         if (read)
 192                 rv = read(file, buf, count, ppos);
 193
 194         pde_users_dec(pde);
 195         return rv;
 196 }
 197
 198 static ssize_t proc_reg_write(struct file *file, const char __user *buf, size_t count, loff_t *ppos)
 199 {
 200         struct proc_dir_entry *pde = PDE(file->f_path.dentry->d_inode);
 201         ssize_t rv = -EIO;
 202         ssize_t (*write)(struct file *, const char __user *, size_t, loff_t *);
 203
 204         spin_lock(&pde->pde_unload_lock);
 205         if (!pde->proc_fops) {
 206                 spin_unlock(&pde->pde_unload_lock);
 207                 return rv;
 208         }
 209         pde->pde_users++;
 210         write = pde->proc_fops->write;
 211         spin_unlock(&pde->pde_unload_lock);
 212
 213         if (write)
 214                 rv = write(file, buf, count, ppos);
 215
 216         pde_users_dec(pde);
 217         return rv;
 218 }
 219
 220 static unsigned int proc_reg_poll(struct file *file, struct poll_table_struct *pts)
 221 {
 222         struct proc_dir_entry *pde = PDE(file->f_path.dentry->d_inode);
 223         unsigned int rv = DEFAULT_POLLMASK;
 224         unsigned int (*poll)(struct file *, struct poll_table_struct *);
 225
 226         spin_lock(&pde->pde_unload_lock);
 227         if (!pde->proc_fops) {
 228                 spin_unlock(&pde->pde_unload_lock);
 229                 return rv;
 230         }
 231         pde->pde_users++;
 232         poll = pde->proc_fops->poll;
 233         spin_unlock(&pde->pde_unload_lock);
 234
 235         if (poll)
 236                 rv = poll(file, pts);
 237
 238         pde_users_dec(pde);
 239         return rv;
 240 }
 241
 242 static long proc_reg_unlocked_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 (*unlocked_ioctl)(struct file *, unsigned int, unsigned long);
 247         int (*ioctl)(struct inode *, struct file *, unsigned int, unsigned long);
 248
 249         spin_lock(&pde->pde_unload_lock);
 250         if (!pde->proc_fops) {
 251                 spin_unlock(&pde->pde_unload_lock);
 252                 return rv;
 253         }
 254         pde->pde_users++;
 255         unlocked_ioctl = pde->proc_fops->unlocked_ioctl;
 256         ioctl = pde->proc_fops->ioctl;
 257         spin_unlock(&pde->pde_unload_lock);
 258
 259         if (unlocked_ioctl) {
 260                 rv = unlocked_ioctl(file, cmd, arg);
 261                 if (rv == -ENOIOCTLCMD)
 262                         rv = -EINVAL;
 263         } else if (ioctl) {
 264                 lock_kernel();
 265                 rv = ioctl(file->f_path.dentry->d_inode, file, cmd, arg);
 266                 unlock_kernel();
 267         }
 268
 269         pde_users_dec(pde);
 270         return rv;
 271 }
 272
 273 #ifdef CONFIG_COMPAT
 274 static long proc_reg_compat_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
 275 {
 276         struct proc_dir_entry *pde = PDE(file->f_path.dentry->d_inode);
 277         long rv = -ENOTTY;
 278         long (*compat_ioctl)(struct file *, unsigned int, unsigned long);
 279
 280         spin_lock(&pde->pde_unload_lock);
 281         if (!pde->proc_fops) {
 282                 spin_unlock(&pde->pde_unload_lock);
 283                 return rv;
 284         }
 285         pde->pde_users++;
 286         compat_ioctl = pde->proc_fops->compat_ioctl;
 287         spin_unlock(&pde->pde_unload_lock);
 288
 289         if (compat_ioctl)
 290                 rv = compat_ioctl(file, cmd, arg);
 291
 292         pde_users_dec(pde);
 293         return rv;
 294 }
 295 #endif
 296
 297 static int proc_reg_mmap(struct file *file, struct vm_area_struct *vma)
 298 {
 299         struct proc_dir_entry *pde = PDE(file->f_path.dentry->d_inode);
 300         int rv = -EIO;
 301         int (*mmap)(struct file *, struct vm_area_struct *);
 302
 303         spin_lock(&pde->pde_unload_lock);
 304         if (!pde->proc_fops) {
 305                 spin_unlock(&pde->pde_unload_lock);
 306                 return rv;
 307         }
 308         pde->pde_users++;
 309         mmap = pde->proc_fops->mmap;
 310         spin_unlock(&pde->pde_unload_lock);
 311
 312         if (mmap)
 313                 rv = mmap(file, vma);
 314
 315         pde_users_dec(pde);
 316         return rv;
 317 }
 318
 319 static int proc_reg_open(struct inode *inode, struct file *file)
 320 {
 321         struct proc_dir_entry *pde = PDE(inode);
 322         int rv = 0;
 323         int (*open)(struct inode *, struct file *);
 324         int (*release)(struct inode *, struct file *);
 325         struct pde_opener *pdeo;
 326
 327         /*
 328          * What for, you ask? Well, we can have open, rmmod, remove_proc_entry
 329          * sequence. ->release won't be called because ->proc_fops will be
 330          * cleared. Depending on complexity of ->release, consequences vary.
 331          *
 332          * We can't wait for mercy when close will be done for real, it's
 333          * deadlockable: rmmod foo </proc/foo . So, we're going to do ->release
 334          * by hand in remove_proc_entry(). For this, save opener's credentials
 335          * for later.
 336          */
 337         pdeo = kmalloc(sizeof(struct pde_opener), GFP_KERNEL);
 338         if (!pdeo)
 339                 return -ENOMEM;
 340
 341         spin_lock(&pde->pde_unload_lock);
 342         if (!pde->proc_fops) {
 343                 spin_unlock(&pde->pde_unload_lock);
 344                 kfree(pdeo);
 345                 return rv;
 346         }
 347         pde->pde_users++;
 348         open = pde->proc_fops->open;
 349         release = pde->proc_fops->release;
 350         spin_unlock(&pde->pde_unload_lock);
 351
 352         if (open)
 353                 rv = open(inode, file);
 354
 355         spin_lock(&pde->pde_unload_lock);
 356         if (rv == 0 && release) {
 357                 /* To know what to release. */
 358                 pdeo->inode = inode;
 359                 pdeo->file = file;
 360                 /* Strictly for "too late" ->release in proc_reg_release(). */
 361                 pdeo->release = release;
 362                 list_add(&pdeo->lh, &pde->pde_openers);
 363         } else
 364                 kfree(pdeo);
 365         __pde_users_dec(pde);
 366         spin_unlock(&pde->pde_unload_lock);
 367         return rv;
 368 }
 369
 370 static struct pde_opener *find_pde_opener(struct proc_dir_entry *pde,
 371                                         struct inode *inode, struct file *file)
 372 {
 373         struct pde_opener *pdeo;
 374
 375         list_for_each_entry(pdeo, &pde->pde_openers, lh) {
 376                 if (pdeo->inode == inode && pdeo->file == file)
 377                         return pdeo;
 378         }
 379         return NULL;
 380 }
 381
 382 static int proc_reg_release(struct inode *inode, struct file *file)
 383 {
 384         struct proc_dir_entry *pde = PDE(inode);
 385         int rv = 0;
 386         int (*release)(struct inode *, struct file *);
 387         struct pde_opener *pdeo;
 388
 389         spin_lock(&pde->pde_unload_lock);
 390         pdeo = find_pde_opener(pde, inode, file);
 391         if (!pde->proc_fops) {
 392                 /*
 393                  * Can't simply exit, __fput() will think that everything is OK,
 394                  * and move on to freeing struct file. remove_proc_entry() will
 395                  * find slacker in opener's list and will try to do non-trivial
 396                  * things with struct file. Therefore, remove opener from list.
 397                  *
 398                  * But if opener is removed from list, who will ->release it?
 399                  */
 400                 if (pdeo) {
 401                         list_del(&pdeo->lh);
 402                         spin_unlock(&pde->pde_unload_lock);
 403                         rv = pdeo->release(inode, file);
 404                         kfree(pdeo);
 405                 } else
 406                         spin_unlock(&pde->pde_unload_lock);
 407                 return rv;
 408         }
 409         pde->pde_users++;
 410         release = pde->proc_fops->release;
 411         if (pdeo) {
 412                 list_del(&pdeo->lh);
 413                 kfree(pdeo);
 414         }
 415         spin_unlock(&pde->pde_unload_lock);
 416
 417         if (release)
 418                 rv = release(inode, file);
 419
 420         pde_users_dec(pde);
 421         return rv;
 422 }
 423
 424 static const struct file_operations proc_reg_file_ops = {
 425         .llseek         = proc_reg_llseek,
 426         .read           = proc_reg_read,
 427         .write          = proc_reg_write,
 428         .poll           = proc_reg_poll,
 429         .unlocked_ioctl = proc_reg_unlocked_ioctl,
 430 #ifdef CONFIG_COMPAT
 431         .compat_ioctl   = proc_reg_compat_ioctl,
 432 #endif
 433         .mmap           = proc_reg_mmap,
 434         .open           = proc_reg_open,
 435         .release        = proc_reg_release,
 436 };
 437
 438 #ifdef CONFIG_COMPAT
 439 static const struct file_operations proc_reg_file_ops_no_compat = {
 440         .llseek         = proc_reg_llseek,
 441         .read           = proc_reg_read,
 442         .write          = proc_reg_write,
 443         .poll           = proc_reg_poll,
 444         .unlocked_ioctl = proc_reg_unlocked_ioctl,
 445         .mmap           = proc_reg_mmap,
 446         .open           = proc_reg_open,
 447         .release        = proc_reg_release,
 448 };
 449 #endif
 450
 451 struct inode *proc_get_inode(struct super_block *sb, unsigned int ino,
 452                                 struct proc_dir_entry *de)
 453 {
 454         struct inode * inode;
 455
 456         if (!try_module_get(de->owner))
 457                 goto out_mod;
 458
 459         inode = iget_locked(sb, ino);
 460         if (!inode)
 461                 goto out_ino;
 462         if (inode->i_state & I_NEW) {
 463                 inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
 464                 PROC_I(inode)->fd = 0;
 465                 PROC_I(inode)->pde = de;
 466
 467                 if (de->mode) {
 468                         inode->i_mode = de->mode;
 469                         inode->i_uid = de->uid;
 470                         inode->i_gid = de->gid;
 471                 }
 472                 if (de->size)
 473                         inode->i_size = de->size;
 474                 if (de->nlink)
 475                         inode->i_nlink = de->nlink;
 476                 if (de->proc_iops)
 477                         inode->i_op = de->proc_iops;
 478                 if (de->proc_fops) {
 479                         if (S_ISREG(inode->i_mode)) {
 480 #ifdef CONFIG_COMPAT
 481                                 if (!de->proc_fops->compat_ioctl)
 482                                         inode->i_fop =
 483                                                 &proc_reg_file_ops_no_compat;
 484                                 else
 485 #endif
 486                                         inode->i_fop = &proc_reg_file_ops;
 487                         } else {
 488                                 inode->i_fop = de->proc_fops;
 489                         }
 490                 }
 491                 unlock_new_inode(inode);
 492         } else
 493                module_put(de->owner);
 494         return inode;
 495
 496 out_ino:
 497         module_put(de->owner);
 498 out_mod:
 499         return NULL;
 500 }
 501
 502 int proc_fill_super(struct super_block *s)
 503 {
 504         struct inode * root_inode;
 505
 506         s->s_flags |= MS_NODIRATIME | MS_NOSUID | MS_NOEXEC;
 507         s->s_blocksize = 1024;
 508         s->s_blocksize_bits = 10;
 509         s->s_magic = PROC_SUPER_MAGIC;
 510         s->s_op = &proc_sops;
 511         s->s_time_gran = 1;
 512
 513         de_get(&proc_root);
 514         root_inode = proc_get_inode(s, PROC_ROOT_INO, &proc_root);
 515         if (!root_inode)
 516                 goto out_no_root;
 517         root_inode->i_uid = 0;
 518         root_inode->i_gid = 0;
 519         s->s_root = d_alloc_root(root_inode);
 520         if (!s->s_root)
 521                 goto out_no_root;
 522         return 0;
 523
 524 out_no_root:
 525         printk("proc_read_super: get root inode failed\n");
 526         iput(root_inode);
 527         de_put(&proc_root);
 528         return -ENOMEM;
 529 }