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