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