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/printk.h>
  17 #include <linux/file.h>
  18 #include <linux/limits.h>
  19 #include <linux/init.h>
  20 #include <linux/module.h>
  21 #include <linux/sysctl.h>
  22 #include <linux/seq_file.h>
  23 #include <linux/slab.h>
  24 #include <linux/mount.h>
  25 #include <linux/magic.h>
  26
  27 #include <asm/uaccess.h>
  28
  29 #include "internal.h"
  30
  31 static void proc_evict_inode(struct inode *inode)
  32 {
  33         struct proc_dir_entry *de;
  34         struct ctl_table_header *head;
  35
  36         truncate_inode_pages_final(&inode->i_data);
  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 = PDE(inode);
  44         if (de)
  45                 pde_put(de);
  46         head = PROC_I(inode)->sysctl;
  47         if (head) {
  48                 RCU_INIT_POINTER(PROC_I(inode)->sysctl, NULL);
  49                 sysctl_head_put(head);
  50         }
  51 }
  52
  53 static struct kmem_cache * proc_inode_cachep;
  54
  55 static struct inode *proc_alloc_inode(struct super_block *sb)
  56 {
  57         struct proc_inode *ei;
  58         struct inode *inode;
  59
  60         ei = (struct proc_inode *)kmem_cache_alloc(proc_inode_cachep, GFP_KERNEL);
  61         if (!ei)
  62                 return NULL;
  63         ei->pid = NULL;
  64         ei->fd = 0;
  65         ei->op.proc_get_link = NULL;
  66         ei->pde = NULL;
  67         ei->sysctl = NULL;
  68         ei->sysctl_entry = NULL;
  69         ei->ns_ops = NULL;
  70         inode = &ei->vfs_inode;
  71         return inode;
  72 }
  73
  74 static void proc_i_callback(struct rcu_head *head)
  75 {
  76         struct inode *inode = container_of(head, struct inode, i_rcu);
  77         kmem_cache_free(proc_inode_cachep, PROC_I(inode));
  78 }
  79
  80 static void proc_destroy_inode(struct inode *inode)
  81 {
  82         call_rcu(&inode->i_rcu, proc_i_callback);
  83 }
  84
  85 static void init_once(void *foo)
  86 {
  87         struct proc_inode *ei = (struct proc_inode *) foo;
  88
  89         inode_init_once(&ei->vfs_inode);
  90 }
  91
  92 void __init proc_init_inodecache(void)
  93 {
  94         proc_inode_cachep = kmem_cache_create("proc_inode_cache",
  95                                              sizeof(struct proc_inode),
  96                                              0, (SLAB_RECLAIM_ACCOUNT|
  97                                                 SLAB_MEM_SPREAD|SLAB_ACCOUNT|
  98                                                 SLAB_PANIC),
  99                                              init_once);
 100 }
 101
 102 static int proc_show_options(struct seq_file *seq, struct dentry *root)
 103 {
 104         struct super_block *sb = root->d_sb;
 105         struct pid_namespace *pid = sb->s_fs_info;
 106
 107         if (!gid_eq(pid->pid_gid, GLOBAL_ROOT_GID))
 108                 seq_printf(seq, ",gid=%u", from_kgid_munged(&init_user_ns, pid->pid_gid));
 109         if (pid->hide_pid != 0)
 110                 seq_printf(seq, ",hidepid=%u", pid->hide_pid);
 111
 112         return 0;
 113 }
 114
 115 static const struct super_operations proc_sops = {
 116         .alloc_inode    = proc_alloc_inode,
 117         .destroy_inode  = proc_destroy_inode,
 118         .drop_inode     = generic_delete_inode,
 119         .evict_inode    = proc_evict_inode,
 120         .statfs         = simple_statfs,
 121         .remount_fs     = proc_remount,
 122         .show_options   = proc_show_options,
 123 };
 124
 125 enum {BIAS = -1U<<31};
 126
 127 static inline int use_pde(struct proc_dir_entry *pde)
 128 {
 129         return atomic_inc_unless_negative(&pde->in_use);
 130 }
 131
 132 static void unuse_pde(struct proc_dir_entry *pde)
 133 {
 134         if (atomic_dec_return(&pde->in_use) == BIAS)
 135                 complete(pde->pde_unload_completion);
 136 }
 137
 138 /* pde is locked */
 139 static void close_pdeo(struct proc_dir_entry *pde, struct pde_opener *pdeo)
 140 {
 141         if (pdeo->closing) {
 142                 /* somebody else is doing that, just wait */
 143                 DECLARE_COMPLETION_ONSTACK(c);
 144                 pdeo->c = &c;
 145                 spin_unlock(&pde->pde_unload_lock);
 146                 wait_for_completion(&c);
 147                 spin_lock(&pde->pde_unload_lock);
 148         } else {
 149                 struct file *file;
 150                 pdeo->closing = 1;
 151                 spin_unlock(&pde->pde_unload_lock);
 152                 file = pdeo->file;
 153                 pde->proc_fops->release(file_inode(file), file);
 154                 spin_lock(&pde->pde_unload_lock);
 155                 list_del_init(&pdeo->lh);
 156                 if (pdeo->c)
 157                         complete(pdeo->c);
 158                 kfree(pdeo);
 159         }
 160 }
 161
 162 void proc_entry_rundown(struct proc_dir_entry *de)
 163 {
 164         DECLARE_COMPLETION_ONSTACK(c);
 165         /* Wait until all existing callers into module are done. */
 166         de->pde_unload_completion = &c;
 167         if (atomic_add_return(BIAS, &de->in_use) != BIAS)
 168                 wait_for_completion(&c);
 169
 170         spin_lock(&de->pde_unload_lock);
 171         while (!list_empty(&de->pde_openers)) {
 172                 struct pde_opener *pdeo;
 173                 pdeo = list_first_entry(&de->pde_openers, struct pde_opener, lh);
 174                 close_pdeo(de, pdeo);
 175         }
 176         spin_unlock(&de->pde_unload_lock);
 177 }
 178
 179 static loff_t proc_reg_llseek(struct file *file, loff_t offset, int whence)
 180 {
 181         struct proc_dir_entry *pde = PDE(file_inode(file));
 182         loff_t rv = -EINVAL;
 183         if (use_pde(pde)) {
 184                 loff_t (*llseek)(struct file *, loff_t, int);
 185                 llseek = pde->proc_fops->llseek;
 186                 if (!llseek)
 187                         llseek = default_llseek;
 188                 rv = llseek(file, offset, whence);
 189                 unuse_pde(pde);
 190         }
 191         return rv;
 192 }
 193
 194 static ssize_t proc_reg_read(struct file *file, char __user *buf, size_t count, loff_t *ppos)
 195 {
 196         ssize_t (*read)(struct file *, char __user *, size_t, loff_t *);
 197         struct proc_dir_entry *pde = PDE(file_inode(file));
 198         ssize_t rv = -EIO;
 199         if (use_pde(pde)) {
 200                 read = pde->proc_fops->read;
 201                 if (read)
 202                         rv = read(file, buf, count, ppos);
 203                 unuse_pde(pde);
 204         }
 205         return rv;
 206 }
 207
 208 static ssize_t proc_reg_write(struct file *file, const char __user *buf, size_t count, loff_t *ppos)
 209 {
 210         ssize_t (*write)(struct file *, const char __user *, size_t, loff_t *);
 211         struct proc_dir_entry *pde = PDE(file_inode(file));
 212         ssize_t rv = -EIO;
 213         if (use_pde(pde)) {
 214                 write = pde->proc_fops->write;
 215                 if (write)
 216                         rv = write(file, buf, count, ppos);
 217                 unuse_pde(pde);
 218         }
 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_inode(file));
 225         unsigned int rv = DEFAULT_POLLMASK;
 226         unsigned int (*poll)(struct file *, struct poll_table_struct *);
 227         if (use_pde(pde)) {
 228                 poll = pde->proc_fops->poll;
 229                 if (poll)
 230                         rv = poll(file, pts);
 231                 unuse_pde(pde);
 232         }
 233         return rv;
 234 }
 235
 236 static long proc_reg_unlocked_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
 237 {
 238         struct proc_dir_entry *pde = PDE(file_inode(file));
 239         long rv = -ENOTTY;
 240         long (*ioctl)(struct file *, unsigned int, unsigned long);
 241         if (use_pde(pde)) {
 242                 ioctl = pde->proc_fops->unlocked_ioctl;
 243                 if (ioctl)
 244                         rv = ioctl(file, cmd, arg);
 245                 unuse_pde(pde);
 246         }
 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_inode(file));
 254         long rv = -ENOTTY;
 255         long (*compat_ioctl)(struct file *, unsigned int, unsigned long);
 256         if (use_pde(pde)) {
 257                 compat_ioctl = pde->proc_fops->compat_ioctl;
 258                 if (compat_ioctl)
 259                         rv = compat_ioctl(file, cmd, arg);
 260                 unuse_pde(pde);
 261         }
 262         return rv;
 263 }
 264 #endif
 265
 266 static int proc_reg_mmap(struct file *file, struct vm_area_struct *vma)
 267 {
 268         struct proc_dir_entry *pde = PDE(file_inode(file));
 269         int rv = -EIO;
 270         int (*mmap)(struct file *, struct vm_area_struct *);
 271         if (use_pde(pde)) {
 272                 mmap = pde->proc_fops->mmap;
 273                 if (mmap)
 274                         rv = mmap(file, vma);
 275                 unuse_pde(pde);
 276         }
 277         return rv;
 278 }
 279
 280 static unsigned long
 281 proc_reg_get_unmapped_area(struct file *file, unsigned long orig_addr,
 282                            unsigned long len, unsigned long pgoff,
 283                            unsigned long flags)
 284 {
 285         struct proc_dir_entry *pde = PDE(file_inode(file));
 286         unsigned long rv = -EIO;
 287
 288         if (use_pde(pde)) {
 289                 typeof(proc_reg_get_unmapped_area) *get_area;
 290
 291                 get_area = pde->proc_fops->get_unmapped_area;
 292 #ifdef CONFIG_MMU
 293                 if (!get_area)
 294                         get_area = current->mm->get_unmapped_area;
 295 #endif
 296
 297                 if (get_area)
 298                         rv = get_area(file, orig_addr, len, pgoff, flags);
 299                 else
 300                         rv = orig_addr;
 301                 unuse_pde(pde);
 302         }
 303         return rv;
 304 }
 305
 306 static int proc_reg_open(struct inode *inode, struct file *file)
 307 {
 308         struct proc_dir_entry *pde = PDE(inode);
 309         int rv = 0;
 310         int (*open)(struct inode *, struct file *);
 311         int (*release)(struct inode *, struct file *);
 312         struct pde_opener *pdeo;
 313
 314         /*
 315          * What for, you ask? Well, we can have open, rmmod, remove_proc_entry
 316          * sequence. ->release won't be called because ->proc_fops will be
 317          * cleared. Depending on complexity of ->release, consequences vary.
 318          *
 319          * We can't wait for mercy when close will be done for real, it's
 320          * deadlockable: rmmod foo </proc/foo . So, we're going to do ->release
 321          * by hand in remove_proc_entry(). For this, save opener's credentials
 322          * for later.
 323          */
 324         pdeo = kzalloc(sizeof(struct pde_opener), GFP_KERNEL);
 325         if (!pdeo)
 326                 return -ENOMEM;
 327
 328         if (!use_pde(pde)) {
 329                 kfree(pdeo);
 330                 return -ENOENT;
 331         }
 332         open = pde->proc_fops->open;
 333         release = pde->proc_fops->release;
 334
 335         if (open)
 336                 rv = open(inode, file);
 337
 338         if (rv == 0 && release) {
 339                 /* To know what to release. */
 340                 pdeo->file = file;
 341                 /* Strictly for "too late" ->release in proc_reg_release(). */
 342                 spin_lock(&pde->pde_unload_lock);
 343                 list_add(&pdeo->lh, &pde->pde_openers);
 344                 spin_unlock(&pde->pde_unload_lock);
 345         } else
 346                 kfree(pdeo);
 347
 348         unuse_pde(pde);
 349         return rv;
 350 }
 351
 352 static int proc_reg_release(struct inode *inode, struct file *file)
 353 {
 354         struct proc_dir_entry *pde = PDE(inode);
 355         struct pde_opener *pdeo;
 356         spin_lock(&pde->pde_unload_lock);
 357         list_for_each_entry(pdeo, &pde->pde_openers, lh) {
 358                 if (pdeo->file == file) {
 359                         close_pdeo(pde, pdeo);
 360                         break;
 361                 }
 362         }
 363         spin_unlock(&pde->pde_unload_lock);
 364         return 0;
 365 }
 366
 367 static const struct file_operations proc_reg_file_ops = {
 368         .llseek         = proc_reg_llseek,
 369         .read           = proc_reg_read,
 370         .write          = proc_reg_write,
 371         .poll           = proc_reg_poll,
 372         .unlocked_ioctl = proc_reg_unlocked_ioctl,
 373 #ifdef CONFIG_COMPAT
 374         .compat_ioctl   = proc_reg_compat_ioctl,
 375 #endif
 376         .mmap           = proc_reg_mmap,
 377         .get_unmapped_area = proc_reg_get_unmapped_area,
 378         .open           = proc_reg_open,
 379         .release        = proc_reg_release,
 380 };
 381
 382 #ifdef CONFIG_COMPAT
 383 static const struct file_operations proc_reg_file_ops_no_compat = {
 384         .llseek         = proc_reg_llseek,
 385         .read           = proc_reg_read,
 386         .write          = proc_reg_write,
 387         .poll           = proc_reg_poll,
 388         .unlocked_ioctl = proc_reg_unlocked_ioctl,
 389         .mmap           = proc_reg_mmap,
 390         .get_unmapped_area = proc_reg_get_unmapped_area,
 391         .open           = proc_reg_open,
 392         .release        = proc_reg_release,
 393 };
 394 #endif
 395
 396 static void proc_put_link(void *p)
 397 {
 398         unuse_pde(p);
 399 }
 400
 401 static const char *proc_get_link(struct dentry *dentry,
 402                                  struct inode *inode,
 403                                  struct delayed_call *done)
 404 {
 405         struct proc_dir_entry *pde = PDE(inode);
 406         if (unlikely(!use_pde(pde)))
 407                 return ERR_PTR(-EINVAL);
 408         set_delayed_call(done, proc_put_link, pde);
 409         return pde->data;
 410 }
 411
 412 const struct inode_operations proc_link_inode_operations = {
 413         .readlink       = generic_readlink,
 414         .get_link       = proc_get_link,
 415 };
 416
 417 struct inode *proc_get_inode(struct super_block *sb, struct proc_dir_entry *de)
 418 {
 419         struct inode *inode = new_inode_pseudo(sb);
 420
 421         if (inode) {
 422                 inode->i_ino = de->low_ino;
 423                 inode->i_mtime = inode->i_atime = inode->i_ctime = current_time(inode);
 424                 PROC_I(inode)->pde = de;
 425
 426                 if (is_empty_pde(de)) {
 427                         make_empty_dir_inode(inode);
 428                         return inode;
 429                 }
 430                 if (de->mode) {
 431                         inode->i_mode = de->mode;
 432                         inode->i_uid = de->uid;
 433                         inode->i_gid = de->gid;
 434                 }
 435                 if (de->size)
 436                         inode->i_size = de->size;
 437                 if (de->nlink)
 438                         set_nlink(inode, de->nlink);
 439                 WARN_ON(!de->proc_iops);
 440                 inode->i_op = de->proc_iops;
 441                 if (de->proc_fops) {
 442                         if (S_ISREG(inode->i_mode)) {
 443 #ifdef CONFIG_COMPAT
 444                                 if (!de->proc_fops->compat_ioctl)
 445                                         inode->i_fop =
 446                                                 &proc_reg_file_ops_no_compat;
 447                                 else
 448 #endif
 449                                         inode->i_fop = &proc_reg_file_ops;
 450                         } else {
 451                                 inode->i_fop = de->proc_fops;
 452                         }
 453                 }
 454         } else
 455                pde_put(de);
 456         return inode;
 457 }
 458
 459 int proc_fill_super(struct super_block *s, void *data, int silent)
 460 {
 461         struct pid_namespace *ns = get_pid_ns(s->s_fs_info);
 462         struct inode *root_inode;
 463         int ret;
 464
 465         if (!proc_parse_options(data, ns))
 466                 return -EINVAL;
 467
 468         /* User space would break if executables or devices appear on proc */
 469         s->s_iflags |= SB_I_USERNS_VISIBLE | SB_I_NOEXEC | SB_I_NODEV;
 470         s->s_flags |= MS_NODIRATIME | MS_NOSUID | MS_NOEXEC;
 471         s->s_blocksize = 1024;
 472         s->s_blocksize_bits = 10;
 473         s->s_magic = PROC_SUPER_MAGIC;
 474         s->s_op = &proc_sops;
 475         s->s_time_gran = 1;
 476
 477         /*
 478          * procfs isn't actually a stacking filesystem; however, there is
 479          * too much magic going on inside it to permit stacking things on
 480          * top of it
 481          */
 482         s->s_stack_depth = FILESYSTEM_MAX_STACK_DEPTH;
 483
 484         pde_get(&proc_root);
 485         root_inode = proc_get_inode(s, &proc_root);
 486         if (!root_inode) {
 487                 pr_err("proc_fill_super: get root inode failed\n");
 488                 return -ENOMEM;
 489         }
 490
 491         s->s_root = d_make_root(root_inode);
 492         if (!s->s_root) {
 493                 pr_err("proc_fill_super: allocate dentry failed\n");
 494                 return -ENOMEM;
 495         }
 496
 497         ret = proc_setup_self(s);
 498         if (ret) {
 499                 return ret;
 500         }
 501         return proc_setup_thread_self(s);
 502 }