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/system.h>
  26 #include <asm/uaccess.h>
  27
  28 #include "internal.h"
  29
  30 static void proc_evict_inode(struct inode *inode)
  31 {
  32         struct proc_dir_entry *de;
  33         struct ctl_table_header *head;
  34         const struct proc_ns_operations *ns_ops;
  35
  36         truncate_inode_pages(&inode->i_data, 0);
  37         end_writeback(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         if (ns_ops && ns_ops->put)
  54                 ns_ops->put(PROC_I(inode)->ns);
  55 }
  56
  57 static struct kmem_cache * proc_inode_cachep;
  58
  59 static struct inode *proc_alloc_inode(struct super_block *sb)
  60 {
  61         struct proc_inode *ei;
  62         struct inode *inode;
  63
  64         ei = (struct proc_inode *)kmem_cache_alloc(proc_inode_cachep, GFP_KERNEL);
  65         if (!ei)
  66                 return NULL;
  67         ei->pid = NULL;
  68         ei->fd = 0;
  69         ei->op.proc_get_link = NULL;
  70         ei->pde = NULL;
  71         ei->sysctl = NULL;
  72         ei->sysctl_entry = NULL;
  73         ei->ns = NULL;
  74         ei->ns_ops = NULL;
  75         inode = &ei->vfs_inode;
  76         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
  77         return inode;
  78 }
  79
  80 static void proc_i_callback(struct rcu_head *head)
  81 {
  82         struct inode *inode = container_of(head, struct inode, i_rcu);
  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 int proc_show_options(struct seq_file *seq, struct dentry *root)
 108 {
 109         struct super_block *sb = root->d_sb;
 110         struct pid_namespace *pid = sb->s_fs_info;
 111
 112         if (pid->pid_gid)
 113                 seq_printf(seq, ",gid=%lu", (unsigned long)pid->pid_gid);
 114         if (pid->hide_pid != 0)
 115                 seq_printf(seq, ",hidepid=%u", pid->hide_pid);
 116
 117         return 0;
 118 }
 119
 120 static const struct super_operations proc_sops = {
 121         .alloc_inode    = proc_alloc_inode,
 122         .destroy_inode  = proc_destroy_inode,
 123         .drop_inode     = generic_delete_inode,
 124         .evict_inode    = proc_evict_inode,
 125         .statfs         = simple_statfs,
 126         .remount_fs     = proc_remount,
 127         .show_options   = proc_show_options,
 128 };
 129
 130 static void __pde_users_dec(struct proc_dir_entry *pde)
 131 {
 132         pde->pde_users--;
 133         if (pde->pde_unload_completion && pde->pde_users == 0)
 134                 complete(pde->pde_unload_completion);
 135 }
 136
 137 void pde_users_dec(struct proc_dir_entry *pde)
 138 {
 139         spin_lock(&pde->pde_unload_lock);
 140         __pde_users_dec(pde);
 141         spin_unlock(&pde->pde_unload_lock);
 142 }
 143
 144 static loff_t proc_reg_llseek(struct file *file, loff_t offset, int whence)
 145 {
 146         struct proc_dir_entry *pde = PDE(file->f_path.dentry->d_inode);
 147         loff_t rv = -EINVAL;
 148         loff_t (*llseek)(struct file *, loff_t, int);
 149
 150         spin_lock(&pde->pde_unload_lock);
 151         /*
 152          * remove_proc_entry() is going to delete PDE (as part of module
 153          * cleanup sequence). No new callers into module allowed.
 154          */
 155         if (!pde->proc_fops) {
 156                 spin_unlock(&pde->pde_unload_lock);
 157                 return rv;
 158         }
 159         /*
 160          * Bump refcount so that remove_proc_entry will wail for ->llseek to
 161          * complete.
 162          */
 163         pde->pde_users++;
 164         /*
 165          * Save function pointer under lock, to protect against ->proc_fops
 166          * NULL'ifying right after ->pde_unload_lock is dropped.
 167          */
 168         llseek = pde->proc_fops->llseek;
 169         spin_unlock(&pde->pde_unload_lock);
 170
 171         if (!llseek)
 172                 llseek = default_llseek;
 173         rv = llseek(file, offset, whence);
 174
 175         pde_users_dec(pde);
 176         return rv;
 177 }
 178
 179 static ssize_t proc_reg_read(struct file *file, char __user *buf, size_t count, loff_t *ppos)
 180 {
 181         struct proc_dir_entry *pde = PDE(file->f_path.dentry->d_inode);
 182         ssize_t rv = -EIO;
 183         ssize_t (*read)(struct file *, char __user *, size_t, loff_t *);
 184
 185         spin_lock(&pde->pde_unload_lock);
 186         if (!pde->proc_fops) {
 187                 spin_unlock(&pde->pde_unload_lock);
 188                 return rv;
 189         }
 190         pde->pde_users++;
 191         read = pde->proc_fops->read;
 192         spin_unlock(&pde->pde_unload_lock);
 193
 194         if (read)
 195                 rv = read(file, buf, count, ppos);
 196
 197         pde_users_dec(pde);
 198         return rv;
 199 }
 200
 201 static ssize_t proc_reg_write(struct file *file, const char __user *buf, size_t count, loff_t *ppos)
 202 {
 203         struct proc_dir_entry *pde = PDE(file->f_path.dentry->d_inode);
 204         ssize_t rv = -EIO;
 205         ssize_t (*write)(struct file *, const char __user *, size_t, loff_t *);
 206
 207         spin_lock(&pde->pde_unload_lock);
 208         if (!pde->proc_fops) {
 209                 spin_unlock(&pde->pde_unload_lock);
 210                 return rv;
 211         }
 212         pde->pde_users++;
 213         write = pde->proc_fops->write;
 214         spin_unlock(&pde->pde_unload_lock);
 215
 216         if (write)
 217                 rv = write(file, buf, count, ppos);
 218
 219         pde_users_dec(pde);
 220         return rv;
 221 }
 222
 223 static unsigned int proc_reg_poll(struct file *file, struct poll_table_struct *pts)
 224 {
 225         struct proc_dir_entry *pde = PDE(file->f_path.dentry->d_inode);
 226         unsigned int rv = DEFAULT_POLLMASK;
 227         unsigned int (*poll)(struct file *, struct poll_table_struct *);
 228
 229         spin_lock(&pde->pde_unload_lock);
 230         if (!pde->proc_fops) {
 231                 spin_unlock(&pde->pde_unload_lock);
 232                 return rv;
 233         }
 234         pde->pde_users++;
 235         poll = pde->proc_fops->poll;
 236         spin_unlock(&pde->pde_unload_lock);
 237
 238         if (poll)
 239                 rv = poll(file, pts);
 240
 241         pde_users_dec(pde);
 242         return rv;
 243 }
 244
 245 static long proc_reg_unlocked_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
 246 {
 247         struct proc_dir_entry *pde = PDE(file->f_path.dentry->d_inode);
 248         long rv = -ENOTTY;
 249         long (*ioctl)(struct file *, unsigned int, unsigned long);
 250
 251         spin_lock(&pde->pde_unload_lock);
 252         if (!pde->proc_fops) {
 253                 spin_unlock(&pde->pde_unload_lock);
 254                 return rv;
 255         }
 256         pde->pde_users++;
 257         ioctl = pde->proc_fops->unlocked_ioctl;
 258         spin_unlock(&pde->pde_unload_lock);
 259
 260         if (ioctl)
 261                 rv = ioctl(file, cmd, arg);
 262
 263         pde_users_dec(pde);
 264         return rv;
 265 }
 266
 267 #ifdef CONFIG_COMPAT
 268 static long proc_reg_compat_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
 269 {
 270         struct proc_dir_entry *pde = PDE(file->f_path.dentry->d_inode);
 271         long rv = -ENOTTY;
 272         long (*compat_ioctl)(struct file *, unsigned int, unsigned long);
 273
 274         spin_lock(&pde->pde_unload_lock);
 275         if (!pde->proc_fops) {
 276                 spin_unlock(&pde->pde_unload_lock);
 277                 return rv;
 278         }
 279         pde->pde_users++;
 280         compat_ioctl = pde->proc_fops->compat_ioctl;
 281         spin_unlock(&pde->pde_unload_lock);
 282
 283         if (compat_ioctl)
 284                 rv = compat_ioctl(file, cmd, arg);
 285
 286         pde_users_dec(pde);
 287         return rv;
 288 }
 289 #endif
 290
 291 static int proc_reg_mmap(struct file *file, struct vm_area_struct *vma)
 292 {
 293         struct proc_dir_entry *pde = PDE(file->f_path.dentry->d_inode);
 294         int rv = -EIO;
 295         int (*mmap)(struct file *, struct vm_area_struct *);
 296
 297         spin_lock(&pde->pde_unload_lock);
 298         if (!pde->proc_fops) {
 299                 spin_unlock(&pde->pde_unload_lock);
 300                 return rv;
 301         }
 302         pde->pde_users++;
 303         mmap = pde->proc_fops->mmap;
 304         spin_unlock(&pde->pde_unload_lock);
 305
 306         if (mmap)
 307                 rv = mmap(file, vma);
 308
 309         pde_users_dec(pde);
 310         return rv;
 311 }
 312
 313 static int proc_reg_open(struct inode *inode, struct file *file)
 314 {
 315         struct proc_dir_entry *pde = PDE(inode);
 316         int rv = 0;
 317         int (*open)(struct inode *, struct file *);
 318         int (*release)(struct inode *, struct file *);
 319         struct pde_opener *pdeo;
 320
 321         /*
 322          * What for, you ask? Well, we can have open, rmmod, remove_proc_entry
 323          * sequence. ->release won't be called because ->proc_fops will be
 324          * cleared. Depending on complexity of ->release, consequences vary.
 325          *
 326          * We can't wait for mercy when close will be done for real, it's
 327          * deadlockable: rmmod foo </proc/foo . So, we're going to do ->release
 328          * by hand in remove_proc_entry(). For this, save opener's credentials
 329          * for later.
 330          */
 331         pdeo = kmalloc(sizeof(struct pde_opener), GFP_KERNEL);
 332         if (!pdeo)
 333                 return -ENOMEM;
 334
 335         spin_lock(&pde->pde_unload_lock);
 336         if (!pde->proc_fops) {
 337                 spin_unlock(&pde->pde_unload_lock);
 338                 kfree(pdeo);
 339                 return -ENOENT;
 340         }
 341         pde->pde_users++;
 342         open = pde->proc_fops->open;
 343         release = pde->proc_fops->release;
 344         spin_unlock(&pde->pde_unload_lock);
 345
 346         if (open)
 347                 rv = open(inode, file);
 348
 349         spin_lock(&pde->pde_unload_lock);
 350         if (rv == 0 && release) {
 351                 /* To know what to release. */
 352                 pdeo->inode = inode;
 353                 pdeo->file = file;
 354                 /* Strictly for "too late" ->release in proc_reg_release(). */
 355                 pdeo->release = release;
 356                 list_add(&pdeo->lh, &pde->pde_openers);
 357         } else
 358                 kfree(pdeo);
 359         __pde_users_dec(pde);
 360         spin_unlock(&pde->pde_unload_lock);
 361         return rv;
 362 }
 363
 364 static struct pde_opener *find_pde_opener(struct proc_dir_entry *pde,
 365                                         struct inode *inode, struct file *file)
 366 {
 367         struct pde_opener *pdeo;
 368
 369         list_for_each_entry(pdeo, &pde->pde_openers, lh) {
 370                 if (pdeo->inode == inode && pdeo->file == file)
 371                         return pdeo;
 372         }
 373         return NULL;
 374 }
 375
 376 static int proc_reg_release(struct inode *inode, struct file *file)
 377 {
 378         struct proc_dir_entry *pde = PDE(inode);
 379         int rv = 0;
 380         int (*release)(struct inode *, struct file *);
 381         struct pde_opener *pdeo;
 382
 383         spin_lock(&pde->pde_unload_lock);
 384         pdeo = find_pde_opener(pde, inode, file);
 385         if (!pde->proc_fops) {
 386                 /*
 387                  * Can't simply exit, __fput() will think that everything is OK,
 388                  * and move on to freeing struct file. remove_proc_entry() will
 389                  * find slacker in opener's list and will try to do non-trivial
 390                  * things with struct file. Therefore, remove opener from list.
 391                  *
 392                  * But if opener is removed from list, who will ->release it?
 393                  */
 394                 if (pdeo) {
 395                         list_del(&pdeo->lh);
 396                         spin_unlock(&pde->pde_unload_lock);
 397                         rv = pdeo->release(inode, file);
 398                         kfree(pdeo);
 399                 } else
 400                         spin_unlock(&pde->pde_unload_lock);
 401                 return rv;
 402         }
 403         pde->pde_users++;
 404         release = pde->proc_fops->release;
 405         if (pdeo) {
 406                 list_del(&pdeo->lh);
 407                 kfree(pdeo);
 408         }
 409         spin_unlock(&pde->pde_unload_lock);
 410
 411         if (release)
 412                 rv = release(inode, file);
 413
 414         pde_users_dec(pde);
 415         return rv;
 416 }
 417
 418 static const struct file_operations proc_reg_file_ops = {
 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 #ifdef CONFIG_COMPAT
 425         .compat_ioctl   = proc_reg_compat_ioctl,
 426 #endif
 427         .mmap           = proc_reg_mmap,
 428         .open           = proc_reg_open,
 429         .release        = proc_reg_release,
 430 };
 431
 432 #ifdef CONFIG_COMPAT
 433 static const struct file_operations proc_reg_file_ops_no_compat = {
 434         .llseek         = proc_reg_llseek,
 435         .read           = proc_reg_read,
 436         .write          = proc_reg_write,
 437         .poll           = proc_reg_poll,
 438         .unlocked_ioctl = proc_reg_unlocked_ioctl,
 439         .mmap           = proc_reg_mmap,
 440         .open           = proc_reg_open,
 441         .release        = proc_reg_release,
 442 };
 443 #endif
 444
 445 struct inode *proc_get_inode(struct super_block *sb, struct proc_dir_entry *de)
 446 {
 447         struct inode * inode;
 448
 449         inode = iget_locked(sb, de->low_ino);
 450         if (!inode)
 451                 return NULL;
 452         if (inode->i_state & I_NEW) {
 453                 inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
 454                 PROC_I(inode)->fd = 0;
 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         struct inode * root_inode;
 490
 491         s->s_flags |= MS_NODIRATIME | MS_NOSUID | MS_NOEXEC;
 492         s->s_blocksize = 1024;
 493         s->s_blocksize_bits = 10;
 494         s->s_magic = PROC_SUPER_MAGIC;
 495         s->s_op = &proc_sops;
 496         s->s_time_gran = 1;
 497
 498         pde_get(&proc_root);
 499         root_inode = proc_get_inode(s, &proc_root);
 500         if (!root_inode)
 501                 goto out_no_root;
 502         root_inode->i_uid = 0;
 503         root_inode->i_gid = 0;
 504         s->s_root = d_alloc_root(root_inode);
 505         if (!s->s_root)
 506                 goto out_no_root;
 507         return 0;
 508
 509 out_no_root:
 510         printk("proc_read_super: get root inode failed\n");
 511         iput(root_inode);
 512         pde_put(&proc_root);
 513         return -ENOMEM;
 514 }