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