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         struct ctl_table_header *head;
  31
  32         truncate_inode_pages(&inode->i_data, 0);
  33         end_writeback(inode);
  34
  35         /* Stop tracking associated processes */
  36         put_pid(PROC_I(inode)->pid);
  37
  38         /* Let go of any associated proc directory entry */
  39         de = PROC_I(inode)->pde;
  40         if (de)
  41                 pde_put(de);
  42         head = PROC_I(inode)->sysctl;
  43         if (head) {
  44                 rcu_assign_pointer(PROC_I(inode)->sysctl, NULL);
  45                 sysctl_head_put(head);
  46         }
  47 }
  48
  49 static struct kmem_cache * proc_inode_cachep;
  50
  51 static struct inode *proc_alloc_inode(struct super_block *sb)
  52 {
  53         struct proc_inode *ei;
  54         struct inode *inode;
  55
  56         ei = (struct proc_inode *)kmem_cache_alloc(proc_inode_cachep, GFP_KERNEL);
  57         if (!ei)
  58                 return NULL;
  59         ei->pid = NULL;
  60         ei->fd = 0;
  61         ei->op.proc_get_link = NULL;
  62         ei->pde = NULL;
  63         ei->sysctl = NULL;
  64         ei->sysctl_entry = NULL;
  65         inode = &ei->vfs_inode;
  66         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
  67         return inode;
  68 }
  69
  70 static void proc_i_callback(struct rcu_head *head)
  71 {
  72         struct inode *inode = container_of(head, struct inode, i_rcu);
  73         INIT_LIST_HEAD(&inode->i_dentry);
  74         kmem_cache_free(proc_inode_cachep, PROC_I(inode));
  75 }
  76
  77 static void proc_destroy_inode(struct inode *inode)
  78 {
  79         call_rcu(&inode->i_rcu, proc_i_callback);
  80 }
  81
  82 static void init_once(void *foo)
  83 {
  84         struct proc_inode *ei = (struct proc_inode *) foo;
  85
  86         inode_init_once(&ei->vfs_inode);
  87 }
  88
  89 void __init proc_init_inodecache(void)
  90 {
  91         proc_inode_cachep = kmem_cache_create("proc_inode_cache",
  92                                              sizeof(struct proc_inode),
  93                                              0, (SLAB_RECLAIM_ACCOUNT|
  94                                                 SLAB_MEM_SPREAD|SLAB_PANIC),
  95                                              init_once);
  96 }
  97
  98 static const struct super_operations proc_sops = {
  99         .alloc_inode    = proc_alloc_inode,
 100         .destroy_inode  = proc_destroy_inode,
 101         .drop_inode     = generic_delete_inode,
 102         .evict_inode    = proc_evict_inode,
 103         .statfs         = simple_statfs,
 104 };
 105
 106 static void __pde_users_dec(struct proc_dir_entry *pde)
 107 {
 108         pde->pde_users--;
 109         if (pde->pde_unload_completion && pde->pde_users == 0)
 110                 complete(pde->pde_unload_completion);
 111 }
 112
 113 void pde_users_dec(struct proc_dir_entry *pde)
 114 {
 115         spin_lock(&pde->pde_unload_lock);
 116         __pde_users_dec(pde);
 117         spin_unlock(&pde->pde_unload_lock);
 118 }
 119
 120 static loff_t proc_reg_llseek(struct file *file, loff_t offset, int whence)
 121 {
 122         struct proc_dir_entry *pde = PDE(file->f_path.dentry->d_inode);
 123         loff_t rv = -EINVAL;
 124         loff_t (*llseek)(struct file *, loff_t, int);
 125
 126         spin_lock(&pde->pde_unload_lock);
 127         /*
 128          * remove_proc_entry() is going to delete PDE (as part of module
 129          * cleanup sequence). No new callers into module allowed.
 130          */
 131         if (!pde->proc_fops) {
 132                 spin_unlock(&pde->pde_unload_lock);
 133                 return rv;
 134         }
 135         /*
 136          * Bump refcount so that remove_proc_entry will wail for ->llseek to
 137          * complete.
 138          */
 139         pde->pde_users++;
 140         /*
 141          * Save function pointer under lock, to protect against ->proc_fops
 142          * NULL'ifying right after ->pde_unload_lock is dropped.
 143          */
 144         llseek = pde->proc_fops->llseek;
 145         spin_unlock(&pde->pde_unload_lock);
 146
 147         if (!llseek)
 148                 llseek = default_llseek;
 149         rv = llseek(file, offset, whence);
 150
 151         pde_users_dec(pde);
 152         return rv;
 153 }
 154
 155 static ssize_t proc_reg_read(struct file *file, char __user *buf, size_t count, loff_t *ppos)
 156 {
 157         struct proc_dir_entry *pde = PDE(file->f_path.dentry->d_inode);
 158         ssize_t rv = -EIO;
 159         ssize_t (*read)(struct file *, char __user *, size_t, loff_t *);
 160
 161         spin_lock(&pde->pde_unload_lock);
 162         if (!pde->proc_fops) {
 163                 spin_unlock(&pde->pde_unload_lock);
 164                 return rv;
 165         }
 166         pde->pde_users++;
 167         read = pde->proc_fops->read;
 168         spin_unlock(&pde->pde_unload_lock);
 169
 170         if (read)
 171                 rv = read(file, buf, count, ppos);
 172
 173         pde_users_dec(pde);
 174         return rv;
 175 }
 176
 177 static ssize_t proc_reg_write(struct file *file, const char __user *buf, size_t count, loff_t *ppos)
 178 {
 179         struct proc_dir_entry *pde = PDE(file->f_path.dentry->d_inode);
 180         ssize_t rv = -EIO;
 181         ssize_t (*write)(struct file *, const char __user *, size_t, loff_t *);
 182
 183         spin_lock(&pde->pde_unload_lock);
 184         if (!pde->proc_fops) {
 185                 spin_unlock(&pde->pde_unload_lock);
 186                 return rv;
 187         }
 188         pde->pde_users++;
 189         write = pde->proc_fops->write;
 190         spin_unlock(&pde->pde_unload_lock);
 191
 192         if (write)
 193                 rv = write(file, buf, count, ppos);
 194
 195         pde_users_dec(pde);
 196         return rv;
 197 }
 198
 199 static unsigned int proc_reg_poll(struct file *file, struct poll_table_struct *pts)
 200 {
 201         struct proc_dir_entry *pde = PDE(file->f_path.dentry->d_inode);
 202         unsigned int rv = DEFAULT_POLLMASK;
 203         unsigned int (*poll)(struct file *, struct poll_table_struct *);
 204
 205         spin_lock(&pde->pde_unload_lock);
 206         if (!pde->proc_fops) {
 207                 spin_unlock(&pde->pde_unload_lock);
 208                 return rv;
 209         }
 210         pde->pde_users++;
 211         poll = pde->proc_fops->poll;
 212         spin_unlock(&pde->pde_unload_lock);
 213
 214         if (poll)
 215                 rv = poll(file, pts);
 216
 217         pde_users_dec(pde);
 218         return rv;
 219 }
 220
 221 static long proc_reg_unlocked_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
 222 {
 223         struct proc_dir_entry *pde = PDE(file->f_path.dentry->d_inode);
 224         long rv = -ENOTTY;
 225         long (*ioctl)(struct file *, unsigned int, unsigned long);
 226
 227         spin_lock(&pde->pde_unload_lock);
 228         if (!pde->proc_fops) {
 229                 spin_unlock(&pde->pde_unload_lock);
 230                 return rv;
 231         }
 232         pde->pde_users++;
 233         ioctl = pde->proc_fops->unlocked_ioctl;
 234         spin_unlock(&pde->pde_unload_lock);
 235
 236         if (ioctl)
 237                 rv = ioctl(file, cmd, arg);
 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, struct proc_dir_entry *de)
 422 {
 423         struct inode * inode;
 424
 425         inode = iget_locked(sb, de->low_ino);
 426         if (!inode)
 427                 return NULL;
 428         if (inode->i_state & I_NEW) {
 429                 inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
 430                 PROC_I(inode)->fd = 0;
 431                 PROC_I(inode)->pde = de;
 432
 433                 if (de->mode) {
 434                         inode->i_mode = de->mode;
 435                         inode->i_uid = de->uid;
 436                         inode->i_gid = de->gid;
 437                 }
 438                 if (de->size)
 439                         inode->i_size = de->size;
 440                 if (de->nlink)
 441                         inode->i_nlink = de->nlink;
 442                 if (de->proc_iops)
 443                         inode->i_op = de->proc_iops;
 444                 if (de->proc_fops) {
 445                         if (S_ISREG(inode->i_mode)) {
 446 #ifdef CONFIG_COMPAT
 447                                 if (!de->proc_fops->compat_ioctl)
 448                                         inode->i_fop =
 449                                                 &proc_reg_file_ops_no_compat;
 450                                 else
 451 #endif
 452                                         inode->i_fop = &proc_reg_file_ops;
 453                         } else {
 454                                 inode->i_fop = de->proc_fops;
 455                         }
 456                 }
 457                 unlock_new_inode(inode);
 458         } else
 459                pde_put(de);
 460         return inode;
 461 }
 462
 463 int proc_fill_super(struct super_block *s)
 464 {
 465         struct inode * root_inode;
 466
 467         s->s_flags |= MS_NODIRATIME | MS_NOSUID | MS_NOEXEC;
 468         s->s_blocksize = 1024;
 469         s->s_blocksize_bits = 10;
 470         s->s_magic = PROC_SUPER_MAGIC;
 471         s->s_op = &proc_sops;
 472         s->s_time_gran = 1;
 473
 474         pde_get(&proc_root);
 475         root_inode = proc_get_inode(s, &proc_root);
 476         if (!root_inode)
 477                 goto out_no_root;
 478         root_inode->i_uid = 0;
 479         root_inode->i_gid = 0;
 480         s->s_root = d_alloc_root(root_inode);
 481         if (!s->s_root)
 482                 goto out_no_root;
 483         return 0;
 484
 485 out_no_root:
 486         printk("proc_read_super: get root inode failed\n");
 487         iput(root_inode);
 488         pde_put(&proc_root);
 489         return -ENOMEM;
 490 }