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
  21 #include <asm/system.h>
  22 #include <asm/uaccess.h>
  23
  24 #include "internal.h"
  25
  26 struct proc_dir_entry *de_get(struct proc_dir_entry *de)
  27 {
  28         atomic_inc(&de->count);
  29         return de;
  30 }
  31
  32 /*
  33  * Decrements the use count and checks for deferred deletion.
  34  */
  35 void de_put(struct proc_dir_entry *de)
  36 {
  37         lock_kernel();
  38         if (!atomic_read(&de->count)) {
  39                 printk("de_put: entry %s already free!\n", de->name);
  40                 unlock_kernel();
  41                 return;
  42         }
  43
  44         if (atomic_dec_and_test(&de->count))
  45                 free_proc_entry(de);
  46         unlock_kernel();
  47 }
  48
  49 /*
  50  * Decrement the use count of the proc_dir_entry.
  51  */
  52 static void proc_delete_inode(struct inode *inode)
  53 {
  54         struct proc_dir_entry *de;
  55
  56         truncate_inode_pages(&inode->i_data, 0);
  57
  58         /* Stop tracking associated processes */
  59         put_pid(PROC_I(inode)->pid);
  60
  61         /* Let go of any associated proc directory entry */
  62         de = PROC_I(inode)->pde;
  63         if (de) {
  64                 if (de->owner)
  65                         module_put(de->owner);
  66                 de_put(de);
  67         }
  68         clear_inode(inode);
  69 }
  70
  71 struct vfsmount *proc_mnt;
  72
  73 static struct kmem_cache * proc_inode_cachep;
  74
  75 static struct inode *proc_alloc_inode(struct super_block *sb)
  76 {
  77         struct proc_inode *ei;
  78         struct inode *inode;
  79
  80         ei = (struct proc_inode *)kmem_cache_alloc(proc_inode_cachep, GFP_KERNEL);
  81         if (!ei)
  82                 return NULL;
  83         ei->pid = NULL;
  84         ei->fd = 0;
  85         ei->op.proc_get_link = NULL;
  86         ei->pde = NULL;
  87         inode = &ei->vfs_inode;
  88         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
  89         return inode;
  90 }
  91
  92 static void proc_destroy_inode(struct inode *inode)
  93 {
  94         kmem_cache_free(proc_inode_cachep, PROC_I(inode));
  95 }
  96
  97 static void init_once(void *foo)
  98 {
  99         struct proc_inode *ei = (struct proc_inode *) foo;
 100
 101         inode_init_once(&ei->vfs_inode);
 102 }
 103
 104 int __init proc_init_inodecache(void)
 105 {
 106         proc_inode_cachep = kmem_cache_create("proc_inode_cache",
 107                                              sizeof(struct proc_inode),
 108                                              0, (SLAB_RECLAIM_ACCOUNT|
 109                                                 SLAB_MEM_SPREAD|SLAB_PANIC),
 110                                              init_once);
 111         return 0;
 112 }
 113
 114 static const struct super_operations proc_sops = {
 115         .alloc_inode    = proc_alloc_inode,
 116         .destroy_inode  = proc_destroy_inode,
 117         .drop_inode     = generic_delete_inode,
 118         .delete_inode   = proc_delete_inode,
 119         .statfs         = simple_statfs,
 120 };
 121
 122 static void __pde_users_dec(struct proc_dir_entry *pde)
 123 {
 124         pde->pde_users--;
 125         if (pde->pde_unload_completion && pde->pde_users == 0)
 126                 complete(pde->pde_unload_completion);
 127 }
 128
 129 static void pde_users_dec(struct proc_dir_entry *pde)
 130 {
 131         spin_lock(&pde->pde_unload_lock);
 132         __pde_users_dec(pde);
 133         spin_unlock(&pde->pde_unload_lock);
 134 }
 135
 136 static loff_t proc_reg_llseek(struct file *file, loff_t offset, int whence)
 137 {
 138         struct proc_dir_entry *pde = PDE(file->f_path.dentry->d_inode);
 139         loff_t rv = -EINVAL;
 140         loff_t (*llseek)(struct file *, loff_t, int);
 141
 142         spin_lock(&pde->pde_unload_lock);
 143         /*
 144          * remove_proc_entry() is going to delete PDE (as part of module
 145          * cleanup sequence). No new callers into module allowed.
 146          */
 147         if (!pde->proc_fops) {
 148                 spin_unlock(&pde->pde_unload_lock);
 149                 return rv;
 150         }
 151         /*
 152          * Bump refcount so that remove_proc_entry will wail for ->llseek to
 153          * complete.
 154          */
 155         pde->pde_users++;
 156         /*
 157          * Save function pointer under lock, to protect against ->proc_fops
 158          * NULL'ifying right after ->pde_unload_lock is dropped.
 159          */
 160         llseek = pde->proc_fops->llseek;
 161         spin_unlock(&pde->pde_unload_lock);
 162
 163         if (!llseek)
 164                 llseek = default_llseek;
 165         rv = llseek(file, offset, whence);
 166
 167         pde_users_dec(pde);
 168         return rv;
 169 }
 170
 171 static ssize_t proc_reg_read(struct file *file, char __user *buf, size_t count, loff_t *ppos)
 172 {
 173         struct proc_dir_entry *pde = PDE(file->f_path.dentry->d_inode);
 174         ssize_t rv = -EIO;
 175         ssize_t (*read)(struct file *, char __user *, size_t, loff_t *);
 176
 177         spin_lock(&pde->pde_unload_lock);
 178         if (!pde->proc_fops) {
 179                 spin_unlock(&pde->pde_unload_lock);
 180                 return rv;
 181         }
 182         pde->pde_users++;
 183         read = pde->proc_fops->read;
 184         spin_unlock(&pde->pde_unload_lock);
 185
 186         if (read)
 187                 rv = read(file, buf, count, ppos);
 188
 189         pde_users_dec(pde);
 190         return rv;
 191 }
 192
 193 static ssize_t proc_reg_write(struct file *file, const char __user *buf, size_t count, loff_t *ppos)
 194 {
 195         struct proc_dir_entry *pde = PDE(file->f_path.dentry->d_inode);
 196         ssize_t rv = -EIO;
 197         ssize_t (*write)(struct file *, const char __user *, size_t, loff_t *);
 198
 199         spin_lock(&pde->pde_unload_lock);
 200         if (!pde->proc_fops) {
 201                 spin_unlock(&pde->pde_unload_lock);
 202                 return rv;
 203         }
 204         pde->pde_users++;
 205         write = pde->proc_fops->write;
 206         spin_unlock(&pde->pde_unload_lock);
 207
 208         if (write)
 209                 rv = write(file, buf, count, ppos);
 210
 211         pde_users_dec(pde);
 212         return rv;
 213 }
 214
 215 static unsigned int proc_reg_poll(struct file *file, struct poll_table_struct *pts)
 216 {
 217         struct proc_dir_entry *pde = PDE(file->f_path.dentry->d_inode);
 218         unsigned int rv = DEFAULT_POLLMASK;
 219         unsigned int (*poll)(struct file *, struct poll_table_struct *);
 220
 221         spin_lock(&pde->pde_unload_lock);
 222         if (!pde->proc_fops) {
 223                 spin_unlock(&pde->pde_unload_lock);
 224                 return rv;
 225         }
 226         pde->pde_users++;
 227         poll = pde->proc_fops->poll;
 228         spin_unlock(&pde->pde_unload_lock);
 229
 230         if (poll)
 231                 rv = poll(file, pts);
 232
 233         pde_users_dec(pde);
 234         return rv;
 235 }
 236
 237 static long proc_reg_unlocked_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
 238 {
 239         struct proc_dir_entry *pde = PDE(file->f_path.dentry->d_inode);
 240         long rv = -ENOTTY;
 241         long (*unlocked_ioctl)(struct file *, unsigned int, unsigned long);
 242         int (*ioctl)(struct inode *, struct file *, unsigned int, unsigned long);
 243
 244         spin_lock(&pde->pde_unload_lock);
 245         if (!pde->proc_fops) {
 246                 spin_unlock(&pde->pde_unload_lock);
 247                 return rv;
 248         }
 249         pde->pde_users++;
 250         unlocked_ioctl = pde->proc_fops->unlocked_ioctl;
 251         ioctl = pde->proc_fops->ioctl;
 252         spin_unlock(&pde->pde_unload_lock);
 253
 254         if (unlocked_ioctl) {
 255                 rv = unlocked_ioctl(file, cmd, arg);
 256                 if (rv == -ENOIOCTLCMD)
 257                         rv = -EINVAL;
 258         } else if (ioctl) {
 259                 lock_kernel();
 260                 rv = ioctl(file->f_path.dentry->d_inode, file, cmd, arg);
 261                 unlock_kernel();
 262         }
 263
 264         pde_users_dec(pde);
 265         return rv;
 266 }
 267
 268 #ifdef CONFIG_COMPAT
 269 static long proc_reg_compat_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
 270 {
 271         struct proc_dir_entry *pde = PDE(file->f_path.dentry->d_inode);
 272         long rv = -ENOTTY;
 273         long (*compat_ioctl)(struct file *, unsigned int, unsigned long);
 274
 275         spin_lock(&pde->pde_unload_lock);
 276         if (!pde->proc_fops) {
 277                 spin_unlock(&pde->pde_unload_lock);
 278                 return rv;
 279         }
 280         pde->pde_users++;
 281         compat_ioctl = pde->proc_fops->compat_ioctl;
 282         spin_unlock(&pde->pde_unload_lock);
 283
 284         if (compat_ioctl)
 285                 rv = compat_ioctl(file, cmd, arg);
 286
 287         pde_users_dec(pde);
 288         return rv;
 289 }
 290 #endif
 291
 292 static int proc_reg_mmap(struct file *file, struct vm_area_struct *vma)
 293 {
 294         struct proc_dir_entry *pde = PDE(file->f_path.dentry->d_inode);
 295         int rv = -EIO;
 296         int (*mmap)(struct file *, struct vm_area_struct *);
 297
 298         spin_lock(&pde->pde_unload_lock);
 299         if (!pde->proc_fops) {
 300                 spin_unlock(&pde->pde_unload_lock);
 301                 return rv;
 302         }
 303         pde->pde_users++;
 304         mmap = pde->proc_fops->mmap;
 305         spin_unlock(&pde->pde_unload_lock);
 306
 307         if (mmap)
 308                 rv = mmap(file, vma);
 309
 310         pde_users_dec(pde);
 311         return rv;
 312 }
 313
 314 static int proc_reg_open(struct inode *inode, struct file *file)
 315 {
 316         struct proc_dir_entry *pde = PDE(inode);
 317         int rv = 0;
 318         int (*open)(struct inode *, struct file *);
 319         int (*release)(struct inode *, struct file *);
 320         struct pde_opener *pdeo;
 321
 322         /*
 323          * What for, you ask? Well, we can have open, rmmod, remove_proc_entry
 324          * sequence. ->release won't be called because ->proc_fops will be
 325          * cleared. Depending on complexity of ->release, consequences vary.
 326          *
 327          * We can't wait for mercy when close will be done for real, it's
 328          * deadlockable: rmmod foo </proc/foo . So, we're going to do ->release
 329          * by hand in remove_proc_entry(). For this, save opener's credentials
 330          * for later.
 331          */
 332         pdeo = kmalloc(sizeof(struct pde_opener), GFP_KERNEL);
 333         if (!pdeo)
 334                 return -ENOMEM;
 335
 336         spin_lock(&pde->pde_unload_lock);
 337         if (!pde->proc_fops) {
 338                 spin_unlock(&pde->pde_unload_lock);
 339                 kfree(pdeo);
 340                 return rv;
 341         }
 342         pde->pde_users++;
 343         open = pde->proc_fops->open;
 344         release = pde->proc_fops->release;
 345         spin_unlock(&pde->pde_unload_lock);
 346
 347         if (open)
 348                 rv = open(inode, file);
 349
 350         spin_lock(&pde->pde_unload_lock);
 351         if (rv == 0 && release) {
 352                 /* To know what to release. */
 353                 pdeo->inode = inode;
 354                 pdeo->file = file;
 355                 /* Strictly for "too late" ->release in proc_reg_release(). */
 356                 pdeo->release = release;
 357                 list_add(&pdeo->lh, &pde->pde_openers);
 358         } else
 359                 kfree(pdeo);
 360         __pde_users_dec(pde);
 361         spin_unlock(&pde->pde_unload_lock);
 362         return rv;
 363 }
 364
 365 static struct pde_opener *find_pde_opener(struct proc_dir_entry *pde,
 366                                         struct inode *inode, struct file *file)
 367 {
 368         struct pde_opener *pdeo;
 369
 370         list_for_each_entry(pdeo, &pde->pde_openers, lh) {
 371                 if (pdeo->inode == inode && pdeo->file == file)
 372                         return pdeo;
 373         }
 374         return NULL;
 375 }
 376
 377 static int proc_reg_release(struct inode *inode, struct file *file)
 378 {
 379         struct proc_dir_entry *pde = PDE(inode);
 380         int rv = 0;
 381         int (*release)(struct inode *, struct file *);
 382         struct pde_opener *pdeo;
 383
 384         spin_lock(&pde->pde_unload_lock);
 385         pdeo = find_pde_opener(pde, inode, file);
 386         if (!pde->proc_fops) {
 387                 /*
 388                  * Can't simply exit, __fput() will think that everything is OK,
 389                  * and move on to freeing struct file. remove_proc_entry() will
 390                  * find slacker in opener's list and will try to do non-trivial
 391                  * things with struct file. Therefore, remove opener from list.
 392                  *
 393                  * But if opener is removed from list, who will ->release it?
 394                  */
 395                 if (pdeo) {
 396                         list_del(&pdeo->lh);
 397                         spin_unlock(&pde->pde_unload_lock);
 398                         rv = pdeo->release(inode, file);
 399                         kfree(pdeo);
 400                 } else
 401                         spin_unlock(&pde->pde_unload_lock);
 402                 return rv;
 403         }
 404         pde->pde_users++;
 405         release = pde->proc_fops->release;
 406         if (pdeo) {
 407                 list_del(&pdeo->lh);
 408                 kfree(pdeo);
 409         }
 410         spin_unlock(&pde->pde_unload_lock);
 411
 412         if (release)
 413                 rv = release(inode, file);
 414
 415         pde_users_dec(pde);
 416         return rv;
 417 }
 418
 419 static const struct file_operations proc_reg_file_ops = {
 420         .llseek         = proc_reg_llseek,
 421         .read           = proc_reg_read,
 422         .write          = proc_reg_write,
 423         .poll           = proc_reg_poll,
 424         .unlocked_ioctl = proc_reg_unlocked_ioctl,
 425 #ifdef CONFIG_COMPAT
 426         .compat_ioctl   = proc_reg_compat_ioctl,
 427 #endif
 428         .mmap           = proc_reg_mmap,
 429         .open           = proc_reg_open,
 430         .release        = proc_reg_release,
 431 };
 432
 433 #ifdef CONFIG_COMPAT
 434 static const struct file_operations proc_reg_file_ops_no_compat = {
 435         .llseek         = proc_reg_llseek,
 436         .read           = proc_reg_read,
 437         .write          = proc_reg_write,
 438         .poll           = proc_reg_poll,
 439         .unlocked_ioctl = proc_reg_unlocked_ioctl,
 440         .mmap           = proc_reg_mmap,
 441         .open           = proc_reg_open,
 442         .release        = proc_reg_release,
 443 };
 444 #endif
 445
 446 struct inode *proc_get_inode(struct super_block *sb, unsigned int ino,
 447                                 struct proc_dir_entry *de)
 448 {
 449         struct inode * inode;
 450
 451         if (!try_module_get(de->owner))
 452                 goto out_mod;
 453
 454         inode = iget_locked(sb, ino);
 455         if (!inode)
 456                 goto out_ino;
 457         if (inode->i_state & I_NEW) {
 458                 inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
 459                 PROC_I(inode)->fd = 0;
 460                 PROC_I(inode)->pde = de;
 461
 462                 if (de->mode) {
 463                         inode->i_mode = de->mode;
 464                         inode->i_uid = de->uid;
 465                         inode->i_gid = de->gid;
 466                 }
 467                 if (de->size)
 468                         inode->i_size = de->size;
 469                 if (de->nlink)
 470                         inode->i_nlink = de->nlink;
 471                 if (de->proc_iops)
 472                         inode->i_op = de->proc_iops;
 473                 if (de->proc_fops) {
 474                         if (S_ISREG(inode->i_mode)) {
 475 #ifdef CONFIG_COMPAT
 476                                 if (!de->proc_fops->compat_ioctl)
 477                                         inode->i_fop =
 478                                                 &proc_reg_file_ops_no_compat;
 479                                 else
 480 #endif
 481                                         inode->i_fop = &proc_reg_file_ops;
 482                         } else {
 483                                 inode->i_fop = de->proc_fops;
 484                         }
 485                 }
 486                 unlock_new_inode(inode);
 487         } else
 488                module_put(de->owner);
 489         return inode;
 490
 491 out_ino:
 492         module_put(de->owner);
 493 out_mod:
 494         return NULL;
 495 }
 496
 497 int proc_fill_super(struct super_block *s)
 498 {
 499         struct inode * root_inode;
 500
 501         s->s_flags |= MS_NODIRATIME | MS_NOSUID | MS_NOEXEC;
 502         s->s_blocksize = 1024;
 503         s->s_blocksize_bits = 10;
 504         s->s_magic = PROC_SUPER_MAGIC;
 505         s->s_op = &proc_sops;
 506         s->s_time_gran = 1;
 507
 508         de_get(&proc_root);
 509         root_inode = proc_get_inode(s, PROC_ROOT_INO, &proc_root);
 510         if (!root_inode)
 511                 goto out_no_root;
 512         root_inode->i_uid = 0;
 513         root_inode->i_gid = 0;
 514         s->s_root = d_alloc_root(root_inode);
 515         if (!s->s_root)
 516                 goto out_no_root;
 517         return 0;
 518
 519 out_no_root:
 520         printk("proc_read_super: get root inode failed\n");
 521         iput(root_inode);
 522         de_put(&proc_root);
 523         return -ENOMEM;
 524 }