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         void *ns;
  35
  36         truncate_inode_pages(&inode->i_data, 0);
  37         clear_inode(inode);
  38
  39         /* Stop tracking associated processes */
  40         put_pid(PROC_I(inode)->pid);
  41
  42         /* Let go of any associated proc directory entry */
  43         de = PROC_I(inode)->pde;
  44         if (de)
  45                 pde_put(de);
  46         head = PROC_I(inode)->sysctl;
  47         if (head) {
  48                 rcu_assign_pointer(PROC_I(inode)->sysctl, NULL);
  49                 sysctl_head_put(head);
  50         }
  51         /* Release any associated namespace */
  52         ns_ops = PROC_I(inode)->ns_ops;
  53         ns = PROC_I(inode)->ns;
  54         if (ns_ops && ns)
  55                 ns_ops->put(ns);
  56 }
  57
  58 static struct kmem_cache * proc_inode_cachep;
  59
  60 static struct inode *proc_alloc_inode(struct super_block *sb)
  61 {
  62         struct proc_inode *ei;
  63         struct inode *inode;
  64
  65         ei = (struct proc_inode *)kmem_cache_alloc(proc_inode_cachep, GFP_KERNEL);
  66         if (!ei)
  67                 return NULL;
  68         ei->pid = NULL;
  69         ei->fd = 0;
  70         ei->op.proc_get_link = NULL;
  71         ei->pde = NULL;
  72         ei->sysctl = NULL;
  73         ei->sysctl_entry = NULL;
  74         ei->ns = NULL;
  75         ei->ns_ops = NULL;
  76         inode = &ei->vfs_inode;
  77         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
  78         return inode;
  79 }
  80
  81 static void proc_i_callback(struct rcu_head *head)
  82 {
  83         struct inode *inode = container_of(head, struct inode, i_rcu);
  84         kmem_cache_free(proc_inode_cachep, PROC_I(inode));
  85 }
  86
  87 static void proc_destroy_inode(struct inode *inode)
  88 {
  89         call_rcu(&inode->i_rcu, proc_i_callback);
  90 }
  91
  92 static void init_once(void *foo)
  93 {
  94         struct proc_inode *ei = (struct proc_inode *) foo;
  95
  96         inode_init_once(&ei->vfs_inode);
  97 }
  98
  99 void __init proc_init_inodecache(void)
 100 {
 101         proc_inode_cachep = kmem_cache_create("proc_inode_cache",
 102                                              sizeof(struct proc_inode),
 103                                              0, (SLAB_RECLAIM_ACCOUNT|
 104                                                 SLAB_MEM_SPREAD|SLAB_PANIC),
 105                                              init_once);
 106 }
 107
 108 static int proc_show_options(struct seq_file *seq, struct dentry *root)
 109 {
 110         struct super_block *sb = root->d_sb;
 111         struct pid_namespace *pid = sb->s_fs_info;
 112
 113         if (!gid_eq(pid->pid_gid, GLOBAL_ROOT_GID))
 114                 seq_printf(seq, ",gid=%u", from_kgid_munged(&init_user_ns, pid->pid_gid));
 115         if (pid->hide_pid != 0)
 116                 seq_printf(seq, ",hidepid=%u", pid->hide_pid);
 117
 118         return 0;
 119 }
 120
 121 static const struct super_operations proc_sops = {
 122         .alloc_inode    = proc_alloc_inode,
 123         .destroy_inode  = proc_destroy_inode,
 124         .drop_inode     = generic_delete_inode,
 125         .evict_inode    = proc_evict_inode,
 126         .statfs         = simple_statfs,
 127         .remount_fs     = proc_remount,
 128         .show_options   = proc_show_options,
 129 };
 130
 131 static void __pde_users_dec(struct proc_dir_entry *pde)
 132 {
 133         pde->pde_users--;
 134         if (pde->pde_unload_completion && pde->pde_users == 0)
 135                 complete(pde->pde_unload_completion);
 136 }
 137
 138 void pde_users_dec(struct proc_dir_entry *pde)
 139 {
 140         spin_lock(&pde->pde_unload_lock);
 141         __pde_users_dec(pde);
 142         spin_unlock(&pde->pde_unload_lock);
 143 }
 144
 145 static loff_t proc_reg_llseek(struct file *file, loff_t offset, int whence)
 146 {
 147         struct proc_dir_entry *pde = PDE(file->f_path.dentry->d_inode);
 148         loff_t rv = -EINVAL;
 149         loff_t (*llseek)(struct file *, loff_t, int);
 150
 151         spin_lock(&pde->pde_unload_lock);
 152         /*
 153          * remove_proc_entry() is going to delete PDE (as part of module
 154          * cleanup sequence). No new callers into module allowed.
 155          */
 156         if (!pde->proc_fops) {
 157                 spin_unlock(&pde->pde_unload_lock);
 158                 return rv;
 159         }
 160         /*
 161          * Bump refcount so that remove_proc_entry will wail for ->llseek to
 162          * complete.
 163          */
 164         pde->pde_users++;
 165         /*
 166          * Save function pointer under lock, to protect against ->proc_fops
 167          * NULL'ifying right after ->pde_unload_lock is dropped.
 168          */
 169         llseek = pde->proc_fops->llseek;
 170         spin_unlock(&pde->pde_unload_lock);
 171
 172         if (!llseek)
 173                 llseek = default_llseek;
 174         rv = llseek(file, offset, whence);
 175
 176         pde_users_dec(pde);
 177         return rv;
 178 }
 179
 180 static ssize_t proc_reg_read(struct file *file, char __user *buf, size_t count, loff_t *ppos)
 181 {
 182         struct proc_dir_entry *pde = PDE(file->f_path.dentry->d_inode);
 183         ssize_t rv = -EIO;
 184         ssize_t (*read)(struct file *, char __user *, size_t, loff_t *);
 185
 186         spin_lock(&pde->pde_unload_lock);
 187         if (!pde->proc_fops) {
 188                 spin_unlock(&pde->pde_unload_lock);
 189                 return rv;
 190         }
 191         pde->pde_users++;
 192         read = pde->proc_fops->read;
 193         spin_unlock(&pde->pde_unload_lock);
 194
 195         if (read)
 196                 rv = read(file, buf, count, ppos);
 197
 198         pde_users_dec(pde);
 199         return rv;
 200 }
 201
 202 static ssize_t proc_reg_write(struct file *file, const char __user *buf, size_t count, loff_t *ppos)
 203 {
 204         struct proc_dir_entry *pde = PDE(file->f_path.dentry->d_inode);
 205         ssize_t rv = -EIO;
 206         ssize_t (*write)(struct file *, const char __user *, size_t, loff_t *);
 207
 208         spin_lock(&pde->pde_unload_lock);
 209         if (!pde->proc_fops) {
 210                 spin_unlock(&pde->pde_unload_lock);
 211                 return rv;
 212         }
 213         pde->pde_users++;
 214         write = pde->proc_fops->write;
 215         spin_unlock(&pde->pde_unload_lock);
 216
 217         if (write)
 218                 rv = write(file, buf, count, ppos);
 219
 220         pde_users_dec(pde);
 221         return rv;
 222 }
 223
 224 static unsigned int proc_reg_poll(struct file *file, struct poll_table_struct *pts)
 225 {
 226         struct proc_dir_entry *pde = PDE(file->f_path.dentry->d_inode);
 227         unsigned int rv = DEFAULT_POLLMASK;
 228         unsigned int (*poll)(struct file *, struct poll_table_struct *);
 229
 230         spin_lock(&pde->pde_unload_lock);
 231         if (!pde->proc_fops) {
 232                 spin_unlock(&pde->pde_unload_lock);
 233                 return rv;
 234         }
 235         pde->pde_users++;
 236         poll = pde->proc_fops->poll;
 237         spin_unlock(&pde->pde_unload_lock);
 238
 239         if (poll)
 240                 rv = poll(file, pts);
 241
 242         pde_users_dec(pde);
 243         return rv;
 244 }
 245
 246 static long proc_reg_unlocked_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
 247 {
 248         struct proc_dir_entry *pde = PDE(file->f_path.dentry->d_inode);
 249         long rv = -ENOTTY;
 250         long (*ioctl)(struct file *, unsigned int, unsigned long);
 251
 252         spin_lock(&pde->pde_unload_lock);
 253         if (!pde->proc_fops) {
 254                 spin_unlock(&pde->pde_unload_lock);
 255                 return rv;
 256         }
 257         pde->pde_users++;
 258         ioctl = pde->proc_fops->unlocked_ioctl;
 259         spin_unlock(&pde->pde_unload_lock);
 260
 261         if (ioctl)
 262                 rv = ioctl(file, cmd, arg);
 263
 264         pde_users_dec(pde);
 265         return rv;
 266 }
 267
 268 #ifdef CONFIG_COMPAT
 269 static long proc_reg_compat_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
 270 {
 271         struct proc_dir_entry *pde = PDE(file->f_path.dentry->d_inode);
 272         long rv = -ENOTTY;
 273         long (*compat_ioctl)(struct file *, unsigned int, unsigned long);
 274
 275         spin_lock(&pde->pde_unload_lock);
 276         if (!pde->proc_fops) {
 277                 spin_unlock(&pde->pde_unload_lock);
 278                 return rv;
 279         }
 280         pde->pde_users++;
 281         compat_ioctl = pde->proc_fops->compat_ioctl;
 282         spin_unlock(&pde->pde_unload_lock);
 283
 284         if (compat_ioctl)
 285                 rv = compat_ioctl(file, cmd, arg);
 286
 287         pde_users_dec(pde);
 288         return rv;
 289 }
 290 #endif
 291
 292 static int proc_reg_mmap(struct file *file, struct vm_area_struct *vma)
 293 {
 294         struct proc_dir_entry *pde = PDE(file->f_path.dentry->d_inode);
 295         int rv = -EIO;
 296         int (*mmap)(struct file *, struct vm_area_struct *);
 297
 298         spin_lock(&pde->pde_unload_lock);
 299         if (!pde->proc_fops) {
 300                 spin_unlock(&pde->pde_unload_lock);
 301                 return rv;
 302         }
 303         pde->pde_users++;
 304         mmap = pde->proc_fops->mmap;
 305         spin_unlock(&pde->pde_unload_lock);
 306
 307         if (mmap)
 308                 rv = mmap(file, vma);
 309
 310         pde_users_dec(pde);
 311         return rv;
 312 }
 313
 314 static int proc_reg_open(struct inode *inode, struct file *file)
 315 {
 316         struct proc_dir_entry *pde = PDE(inode);
 317         int rv = 0;
 318         int (*open)(struct inode *, struct file *);
 319         int (*release)(struct inode *, struct file *);
 320         struct pde_opener *pdeo;
 321
 322         /*
 323          * What for, you ask? Well, we can have open, rmmod, remove_proc_entry
 324          * sequence. ->release won't be called because ->proc_fops will be
 325          * cleared. Depending on complexity of ->release, consequences vary.
 326          *
 327          * We can't wait for mercy when close will be done for real, it's
 328          * deadlockable: rmmod foo </proc/foo . So, we're going to do ->release
 329          * by hand in remove_proc_entry(). For this, save opener's credentials
 330          * for later.
 331          */
 332         pdeo = kmalloc(sizeof(struct pde_opener), GFP_KERNEL);
 333         if (!pdeo)
 334                 return -ENOMEM;
 335
 336         spin_lock(&pde->pde_unload_lock);
 337         if (!pde->proc_fops) {
 338                 spin_unlock(&pde->pde_unload_lock);
 339                 kfree(pdeo);
 340                 return -ENOENT;
 341         }
 342         pde->pde_users++;
 343         open = pde->proc_fops->open;
 344         release = pde->proc_fops->release;
 345         spin_unlock(&pde->pde_unload_lock);
 346
 347         if (open)
 348                 rv = open(inode, file);
 349
 350         spin_lock(&pde->pde_unload_lock);
 351         if (rv == 0 && release) {
 352                 /* To know what to release. */
 353                 pdeo->inode = inode;
 354                 pdeo->file = file;
 355                 /* Strictly for "too late" ->release in proc_reg_release(). */
 356                 pdeo->release = release;
 357                 list_add(&pdeo->lh, &pde->pde_openers);
 358         } else
 359                 kfree(pdeo);
 360         __pde_users_dec(pde);
 361         spin_unlock(&pde->pde_unload_lock);
 362         return rv;
 363 }
 364
 365 static struct pde_opener *find_pde_opener(struct proc_dir_entry *pde,
 366                                         struct inode *inode, struct file *file)
 367 {
 368         struct pde_opener *pdeo;
 369
 370         list_for_each_entry(pdeo, &pde->pde_openers, lh) {
 371                 if (pdeo->inode == inode && pdeo->file == file)
 372                         return pdeo;
 373         }
 374         return NULL;
 375 }
 376
 377 static int proc_reg_release(struct inode *inode, struct file *file)
 378 {
 379         struct proc_dir_entry *pde = PDE(inode);
 380         int rv = 0;
 381         int (*release)(struct inode *, struct file *);
 382         struct pde_opener *pdeo;
 383
 384         spin_lock(&pde->pde_unload_lock);
 385         pdeo = find_pde_opener(pde, inode, file);
 386         if (!pde->proc_fops) {
 387                 /*
 388                  * Can't simply exit, __fput() will think that everything is OK,
 389                  * and move on to freeing struct file. remove_proc_entry() will
 390                  * find slacker in opener's list and will try to do non-trivial
 391                  * things with struct file. Therefore, remove opener from list.
 392                  *
 393                  * But if opener is removed from list, who will ->release it?
 394                  */
 395                 if (pdeo) {
 396                         list_del(&pdeo->lh);
 397                         spin_unlock(&pde->pde_unload_lock);
 398                         rv = pdeo->release(inode, file);
 399                         kfree(pdeo);
 400                 } else
 401                         spin_unlock(&pde->pde_unload_lock);
 402                 return rv;
 403         }
 404         pde->pde_users++;
 405         release = pde->proc_fops->release;
 406         if (pdeo) {
 407                 list_del(&pdeo->lh);
 408                 kfree(pdeo);
 409         }
 410         spin_unlock(&pde->pde_unload_lock);
 411
 412         if (release)
 413                 rv = release(inode, file);
 414
 415         pde_users_dec(pde);
 416         return rv;
 417 }
 418
 419 static const struct file_operations proc_reg_file_ops = {
 420         .llseek         = proc_reg_llseek,
 421         .read           = proc_reg_read,
 422         .write          = proc_reg_write,
 423         .poll           = proc_reg_poll,
 424         .unlocked_ioctl = proc_reg_unlocked_ioctl,
 425 #ifdef CONFIG_COMPAT
 426         .compat_ioctl   = proc_reg_compat_ioctl,
 427 #endif
 428         .mmap           = proc_reg_mmap,
 429         .open           = proc_reg_open,
 430         .release        = proc_reg_release,
 431 };
 432
 433 #ifdef CONFIG_COMPAT
 434 static const struct file_operations proc_reg_file_ops_no_compat = {
 435         .llseek         = proc_reg_llseek,
 436         .read           = proc_reg_read,
 437         .write          = proc_reg_write,
 438         .poll           = proc_reg_poll,
 439         .unlocked_ioctl = proc_reg_unlocked_ioctl,
 440         .mmap           = proc_reg_mmap,
 441         .open           = proc_reg_open,
 442         .release        = proc_reg_release,
 443 };
 444 #endif
 445
 446 struct inode *proc_get_inode(struct super_block *sb, struct proc_dir_entry *de)
 447 {
 448         struct inode * inode;
 449
 450         inode = iget_locked(sb, de->low_ino);
 451         if (!inode)
 452                 return NULL;
 453         if (inode->i_state & I_NEW) {
 454                 inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
 455                 PROC_I(inode)->pde = de;
 456
 457                 if (de->mode) {
 458                         inode->i_mode = de->mode;
 459                         inode->i_uid = de->uid;
 460                         inode->i_gid = de->gid;
 461                 }
 462                 if (de->size)
 463                         inode->i_size = de->size;
 464                 if (de->nlink)
 465                         set_nlink(inode, de->nlink);
 466                 if (de->proc_iops)
 467                         inode->i_op = de->proc_iops;
 468                 if (de->proc_fops) {
 469                         if (S_ISREG(inode->i_mode)) {
 470 #ifdef CONFIG_COMPAT
 471                                 if (!de->proc_fops->compat_ioctl)
 472                                         inode->i_fop =
 473                                                 &proc_reg_file_ops_no_compat;
 474                                 else
 475 #endif
 476                                         inode->i_fop = &proc_reg_file_ops;
 477                         } else {
 478                                 inode->i_fop = de->proc_fops;
 479                         }
 480                 }
 481                 unlock_new_inode(inode);
 482         } else
 483                pde_put(de);
 484         return inode;
 485 }
 486
 487 int proc_fill_super(struct super_block *s)
 488 {
 489         s->s_flags |= MS_NODIRATIME | MS_NOSUID | MS_NOEXEC;
 490         s->s_blocksize = 1024;
 491         s->s_blocksize_bits = 10;
 492         s->s_magic = PROC_SUPER_MAGIC;
 493         s->s_op = &proc_sops;
 494         s->s_time_gran = 1;
 495
 496         pde_get(&proc_root);
 497         s->s_root = d_make_root(proc_get_inode(s, &proc_root));
 498         if (s->s_root)
 499                 return 0;
 500
 501         printk("proc_read_super: get root inode failed\n");
 502         pde_put(&proc_root);
 503         return -ENOMEM;
 504 }