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