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         struct ns_common *ns;
  36
  37         truncate_inode_pages_final(&inode->i_data);
  38         clear_inode(inode);
  39
  40         /* Stop tracking associated processes */
  41         put_pid(PROC_I(inode)->pid);
  42
  43         /* Let go of any associated proc directory entry */
  44         de = PROC_I(inode)->pde;
  45         if (de)
  46                 pde_put(de);
  47         head = PROC_I(inode)->sysctl;
  48         if (head) {
  49                 RCU_INIT_POINTER(PROC_I(inode)->sysctl, NULL);
  50                 sysctl_head_put(head);
  51         }
  52         /* Release any associated namespace */
  53         ns = PROC_I(inode)->ns.ns;
  54         if (ns && ns->ops)
  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.ns = NULL;
  75         ei->ns.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 enum {BIAS = -1U<<31};
 132
 133 static inline int use_pde(struct proc_dir_entry *pde)
 134 {
 135         return atomic_inc_unless_negative(&pde->in_use);
 136 }
 137
 138 static void unuse_pde(struct proc_dir_entry *pde)
 139 {
 140         if (atomic_dec_return(&pde->in_use) == BIAS)
 141                 complete(pde->pde_unload_completion);
 142 }
 143
 144 /* pde is locked */
 145 static void close_pdeo(struct proc_dir_entry *pde, struct pde_opener *pdeo)
 146 {
 147         if (pdeo->closing) {
 148                 /* somebody else is doing that, just wait */
 149                 DECLARE_COMPLETION_ONSTACK(c);
 150                 pdeo->c = &c;
 151                 spin_unlock(&pde->pde_unload_lock);
 152                 wait_for_completion(&c);
 153                 spin_lock(&pde->pde_unload_lock);
 154         } else {
 155                 struct file *file;
 156                 pdeo->closing = 1;
 157                 spin_unlock(&pde->pde_unload_lock);
 158                 file = pdeo->file;
 159                 pde->proc_fops->release(file_inode(file), file);
 160                 spin_lock(&pde->pde_unload_lock);
 161                 list_del_init(&pdeo->lh);
 162                 if (pdeo->c)
 163                         complete(pdeo->c);
 164                 kfree(pdeo);
 165         }
 166 }
 167
 168 void proc_entry_rundown(struct proc_dir_entry *de)
 169 {
 170         DECLARE_COMPLETION_ONSTACK(c);
 171         /* Wait until all existing callers into module are done. */
 172         de->pde_unload_completion = &c;
 173         if (atomic_add_return(BIAS, &de->in_use) != BIAS)
 174                 wait_for_completion(&c);
 175
 176         spin_lock(&de->pde_unload_lock);
 177         while (!list_empty(&de->pde_openers)) {
 178                 struct pde_opener *pdeo;
 179                 pdeo = list_first_entry(&de->pde_openers, struct pde_opener, lh);
 180                 close_pdeo(de, pdeo);
 181         }
 182         spin_unlock(&de->pde_unload_lock);
 183 }
 184
 185 static loff_t proc_reg_llseek(struct file *file, loff_t offset, int whence)
 186 {
 187         struct proc_dir_entry *pde = PDE(file_inode(file));
 188         loff_t rv = -EINVAL;
 189         if (use_pde(pde)) {
 190                 loff_t (*llseek)(struct file *, loff_t, int);
 191                 llseek = pde->proc_fops->llseek;
 192                 if (!llseek)
 193                         llseek = default_llseek;
 194                 rv = llseek(file, offset, whence);
 195                 unuse_pde(pde);
 196         }
 197         return rv;
 198 }
 199
 200 static ssize_t proc_reg_read(struct file *file, char __user *buf, size_t count, loff_t *ppos)
 201 {
 202         ssize_t (*read)(struct file *, char __user *, size_t, loff_t *);
 203         struct proc_dir_entry *pde = PDE(file_inode(file));
 204         ssize_t rv = -EIO;
 205         if (use_pde(pde)) {
 206                 read = pde->proc_fops->read;
 207                 if (read)
 208                         rv = read(file, buf, count, ppos);
 209                 unuse_pde(pde);
 210         }
 211         return rv;
 212 }
 213
 214 static ssize_t proc_reg_write(struct file *file, const char __user *buf, size_t count, loff_t *ppos)
 215 {
 216         ssize_t (*write)(struct file *, const char __user *, size_t, loff_t *);
 217         struct proc_dir_entry *pde = PDE(file_inode(file));
 218         ssize_t rv = -EIO;
 219         if (use_pde(pde)) {
 220                 write = pde->proc_fops->write;
 221                 if (write)
 222                         rv = write(file, buf, count, ppos);
 223                 unuse_pde(pde);
 224         }
 225         return rv;
 226 }
 227
 228 static unsigned int proc_reg_poll(struct file *file, struct poll_table_struct *pts)
 229 {
 230         struct proc_dir_entry *pde = PDE(file_inode(file));
 231         unsigned int rv = DEFAULT_POLLMASK;
 232         unsigned int (*poll)(struct file *, struct poll_table_struct *);
 233         if (use_pde(pde)) {
 234                 poll = pde->proc_fops->poll;
 235                 if (poll)
 236                         rv = poll(file, pts);
 237                 unuse_pde(pde);
 238         }
 239         return rv;
 240 }
 241
 242 static long proc_reg_unlocked_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
 243 {
 244         struct proc_dir_entry *pde = PDE(file_inode(file));
 245         long rv = -ENOTTY;
 246         long (*ioctl)(struct file *, unsigned int, unsigned long);
 247         if (use_pde(pde)) {
 248                 ioctl = pde->proc_fops->unlocked_ioctl;
 249                 if (ioctl)
 250                         rv = ioctl(file, cmd, arg);
 251                 unuse_pde(pde);
 252         }
 253         return rv;
 254 }
 255
 256 #ifdef CONFIG_COMPAT
 257 static long proc_reg_compat_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
 258 {
 259         struct proc_dir_entry *pde = PDE(file_inode(file));
 260         long rv = -ENOTTY;
 261         long (*compat_ioctl)(struct file *, unsigned int, unsigned long);
 262         if (use_pde(pde)) {
 263                 compat_ioctl = pde->proc_fops->compat_ioctl;
 264                 if (compat_ioctl)
 265                         rv = compat_ioctl(file, cmd, arg);
 266                 unuse_pde(pde);
 267         }
 268         return rv;
 269 }
 270 #endif
 271
 272 static int proc_reg_mmap(struct file *file, struct vm_area_struct *vma)
 273 {
 274         struct proc_dir_entry *pde = PDE(file_inode(file));
 275         int rv = -EIO;
 276         int (*mmap)(struct file *, struct vm_area_struct *);
 277         if (use_pde(pde)) {
 278                 mmap = pde->proc_fops->mmap;
 279                 if (mmap)
 280                         rv = mmap(file, vma);
 281                 unuse_pde(pde);
 282         }
 283         return rv;
 284 }
 285
 286 static unsigned long
 287 proc_reg_get_unmapped_area(struct file *file, unsigned long orig_addr,
 288                            unsigned long len, unsigned long pgoff,
 289                            unsigned long flags)
 290 {
 291         struct proc_dir_entry *pde = PDE(file_inode(file));
 292         unsigned long rv = -EIO;
 293
 294         if (use_pde(pde)) {
 295                 typeof(proc_reg_get_unmapped_area) *get_area;
 296
 297                 get_area = pde->proc_fops->get_unmapped_area;
 298 #ifdef CONFIG_MMU
 299                 if (!get_area)
 300                         get_area = current->mm->get_unmapped_area;
 301 #endif
 302
 303                 if (get_area)
 304                         rv = get_area(file, orig_addr, len, pgoff, flags);
 305                 else
 306                         rv = orig_addr;
 307                 unuse_pde(pde);
 308         }
 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 = kzalloc(sizeof(struct pde_opener), GFP_KERNEL);
 331         if (!pdeo)
 332                 return -ENOMEM;
 333
 334         if (!use_pde(pde)) {
 335                 kfree(pdeo);
 336                 return -ENOENT;
 337         }
 338         open = pde->proc_fops->open;
 339         release = pde->proc_fops->release;
 340
 341         if (open)
 342                 rv = open(inode, file);
 343
 344         if (rv == 0 && release) {
 345                 /* To know what to release. */
 346                 pdeo->file = file;
 347                 /* Strictly for "too late" ->release in proc_reg_release(). */
 348                 spin_lock(&pde->pde_unload_lock);
 349                 list_add(&pdeo->lh, &pde->pde_openers);
 350                 spin_unlock(&pde->pde_unload_lock);
 351         } else
 352                 kfree(pdeo);
 353
 354         unuse_pde(pde);
 355         return rv;
 356 }
 357
 358 static int proc_reg_release(struct inode *inode, struct file *file)
 359 {
 360         struct proc_dir_entry *pde = PDE(inode);
 361         struct pde_opener *pdeo;
 362         spin_lock(&pde->pde_unload_lock);
 363         list_for_each_entry(pdeo, &pde->pde_openers, lh) {
 364                 if (pdeo->file == file) {
 365                         close_pdeo(pde, pdeo);
 366                         break;
 367                 }
 368         }
 369         spin_unlock(&pde->pde_unload_lock);
 370         return 0;
 371 }
 372
 373 static const struct file_operations proc_reg_file_ops = {
 374         .llseek         = proc_reg_llseek,
 375         .read           = proc_reg_read,
 376         .write          = proc_reg_write,
 377         .poll           = proc_reg_poll,
 378         .unlocked_ioctl = proc_reg_unlocked_ioctl,
 379 #ifdef CONFIG_COMPAT
 380         .compat_ioctl   = proc_reg_compat_ioctl,
 381 #endif
 382         .mmap           = proc_reg_mmap,
 383         .get_unmapped_area = proc_reg_get_unmapped_area,
 384         .open           = proc_reg_open,
 385         .release        = proc_reg_release,
 386 };
 387
 388 #ifdef CONFIG_COMPAT
 389 static const struct file_operations proc_reg_file_ops_no_compat = {
 390         .llseek         = proc_reg_llseek,
 391         .read           = proc_reg_read,
 392         .write          = proc_reg_write,
 393         .poll           = proc_reg_poll,
 394         .unlocked_ioctl = proc_reg_unlocked_ioctl,
 395         .mmap           = proc_reg_mmap,
 396         .get_unmapped_area = proc_reg_get_unmapped_area,
 397         .open           = proc_reg_open,
 398         .release        = proc_reg_release,
 399 };
 400 #endif
 401
 402 struct inode *proc_get_inode(struct super_block *sb, struct proc_dir_entry *de)
 403 {
 404         struct inode *inode = new_inode_pseudo(sb);
 405
 406         if (inode) {
 407                 inode->i_ino = de->low_ino;
 408                 inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
 409                 PROC_I(inode)->pde = de;
 410
 411                 if (de->mode) {
 412                         inode->i_mode = de->mode;
 413                         inode->i_uid = de->uid;
 414                         inode->i_gid = de->gid;
 415                 }
 416                 if (de->size)
 417                         inode->i_size = de->size;
 418                 if (de->nlink)
 419                         set_nlink(inode, de->nlink);
 420                 WARN_ON(!de->proc_iops);
 421                 inode->i_op = de->proc_iops;
 422                 if (de->proc_fops) {
 423                         if (S_ISREG(inode->i_mode)) {
 424 #ifdef CONFIG_COMPAT
 425                                 if (!de->proc_fops->compat_ioctl)
 426                                         inode->i_fop =
 427                                                 &proc_reg_file_ops_no_compat;
 428                                 else
 429 #endif
 430                                         inode->i_fop = &proc_reg_file_ops;
 431                         } else {
 432                                 inode->i_fop = de->proc_fops;
 433                         }
 434                 }
 435         } else
 436                pde_put(de);
 437         return inode;
 438 }
 439
 440 int proc_fill_super(struct super_block *s)
 441 {
 442         struct inode *root_inode;
 443         int ret;
 444
 445         s->s_flags |= MS_NODIRATIME | MS_NOSUID | MS_NOEXEC;
 446         s->s_blocksize = 1024;
 447         s->s_blocksize_bits = 10;
 448         s->s_magic = PROC_SUPER_MAGIC;
 449         s->s_op = &proc_sops;
 450         s->s_time_gran = 1;
 451
 452         pde_get(&proc_root);
 453         root_inode = proc_get_inode(s, &proc_root);
 454         if (!root_inode) {
 455                 pr_err("proc_fill_super: get root inode failed\n");
 456                 return -ENOMEM;
 457         }
 458
 459         s->s_root = d_make_root(root_inode);
 460         if (!s->s_root) {
 461                 pr_err("proc_fill_super: allocate dentry failed\n");
 462                 return -ENOMEM;
 463         }
 464
 465         ret = proc_setup_self(s);
 466         if (ret) {
 467                 return ret;
 468         }
 469         return proc_setup_thread_self(s);
 470 }