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         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
  72         return inode;
  73 }
  74
  75 static void proc_i_callback(struct rcu_head *head)
  76 {
  77         struct inode *inode = container_of(head, struct inode, i_rcu);
  78         kmem_cache_free(proc_inode_cachep, PROC_I(inode));
  79 }
  80
  81 static void proc_destroy_inode(struct inode *inode)
  82 {
  83         call_rcu(&inode->i_rcu, proc_i_callback);
  84 }
  85
  86 static void init_once(void *foo)
  87 {
  88         struct proc_inode *ei = (struct proc_inode *) foo;
  89
  90         inode_init_once(&ei->vfs_inode);
  91 }
  92
  93 void __init proc_init_inodecache(void)
  94 {
  95         proc_inode_cachep = kmem_cache_create("proc_inode_cache",
  96                                              sizeof(struct proc_inode),
  97                                              0, (SLAB_RECLAIM_ACCOUNT|
  98                                                 SLAB_MEM_SPREAD|SLAB_ACCOUNT|
  99                                                 SLAB_PANIC),
 100                                              init_once);
 101 }
 102
 103 static int proc_show_options(struct seq_file *seq, struct dentry *root)
 104 {
 105         struct super_block *sb = root->d_sb;
 106         struct pid_namespace *pid = sb->s_fs_info;
 107
 108         if (!gid_eq(pid->pid_gid, GLOBAL_ROOT_GID))
 109                 seq_printf(seq, ",gid=%u", from_kgid_munged(&init_user_ns, pid->pid_gid));
 110         if (pid->hide_pid != 0)
 111                 seq_printf(seq, ",hidepid=%u", pid->hide_pid);
 112
 113         return 0;
 114 }
 115
 116 static const struct super_operations proc_sops = {
 117         .alloc_inode    = proc_alloc_inode,
 118         .destroy_inode  = proc_destroy_inode,
 119         .drop_inode     = generic_delete_inode,
 120         .evict_inode    = proc_evict_inode,
 121         .statfs         = simple_statfs,
 122         .remount_fs     = proc_remount,
 123         .show_options   = proc_show_options,
 124 };
 125
 126 enum {BIAS = -1U<<31};
 127
 128 static inline int use_pde(struct proc_dir_entry *pde)
 129 {
 130         return atomic_inc_unless_negative(&pde->in_use);
 131 }
 132
 133 static void unuse_pde(struct proc_dir_entry *pde)
 134 {
 135         if (atomic_dec_return(&pde->in_use) == BIAS)
 136                 complete(pde->pde_unload_completion);
 137 }
 138
 139 /* pde is locked */
 140 static void close_pdeo(struct proc_dir_entry *pde, struct pde_opener *pdeo)
 141 {
 142         if (pdeo->closing) {
 143                 /* somebody else is doing that, just wait */
 144                 DECLARE_COMPLETION_ONSTACK(c);
 145                 pdeo->c = &c;
 146                 spin_unlock(&pde->pde_unload_lock);
 147                 wait_for_completion(&c);
 148                 spin_lock(&pde->pde_unload_lock);
 149         } else {
 150                 struct file *file;
 151                 pdeo->closing = 1;
 152                 spin_unlock(&pde->pde_unload_lock);
 153                 file = pdeo->file;
 154                 pde->proc_fops->release(file_inode(file), file);
 155                 spin_lock(&pde->pde_unload_lock);
 156                 list_del_init(&pdeo->lh);
 157                 if (pdeo->c)
 158                         complete(pdeo->c);
 159                 kfree(pdeo);
 160         }
 161 }
 162
 163 void proc_entry_rundown(struct proc_dir_entry *de)
 164 {
 165         DECLARE_COMPLETION_ONSTACK(c);
 166         /* Wait until all existing callers into module are done. */
 167         de->pde_unload_completion = &c;
 168         if (atomic_add_return(BIAS, &de->in_use) != BIAS)
 169                 wait_for_completion(&c);
 170
 171         spin_lock(&de->pde_unload_lock);
 172         while (!list_empty(&de->pde_openers)) {
 173                 struct pde_opener *pdeo;
 174                 pdeo = list_first_entry(&de->pde_openers, struct pde_opener, lh);
 175                 close_pdeo(de, pdeo);
 176         }
 177         spin_unlock(&de->pde_unload_lock);
 178 }
 179
 180 static loff_t proc_reg_llseek(struct file *file, loff_t offset, int whence)
 181 {
 182         struct proc_dir_entry *pde = PDE(file_inode(file));
 183         loff_t rv = -EINVAL;
 184         if (use_pde(pde)) {
 185                 loff_t (*llseek)(struct file *, loff_t, int);
 186                 llseek = pde->proc_fops->llseek;
 187                 if (!llseek)
 188                         llseek = default_llseek;
 189                 rv = llseek(file, offset, whence);
 190                 unuse_pde(pde);
 191         }
 192         return rv;
 193 }
 194
 195 static ssize_t proc_reg_read(struct file *file, char __user *buf, size_t count, loff_t *ppos)
 196 {
 197         ssize_t (*read)(struct file *, char __user *, size_t, loff_t *);
 198         struct proc_dir_entry *pde = PDE(file_inode(file));
 199         ssize_t rv = -EIO;
 200         if (use_pde(pde)) {
 201                 read = pde->proc_fops->read;
 202                 if (read)
 203                         rv = read(file, buf, count, ppos);
 204                 unuse_pde(pde);
 205         }
 206         return rv;
 207 }
 208
 209 static ssize_t proc_reg_write(struct file *file, const char __user *buf, size_t count, loff_t *ppos)
 210 {
 211         ssize_t (*write)(struct file *, const char __user *, size_t, loff_t *);
 212         struct proc_dir_entry *pde = PDE(file_inode(file));
 213         ssize_t rv = -EIO;
 214         if (use_pde(pde)) {
 215                 write = pde->proc_fops->write;
 216                 if (write)
 217                         rv = write(file, buf, count, ppos);
 218                 unuse_pde(pde);
 219         }
 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_inode(file));
 226         unsigned int rv = DEFAULT_POLLMASK;
 227         unsigned int (*poll)(struct file *, struct poll_table_struct *);
 228         if (use_pde(pde)) {
 229                 poll = pde->proc_fops->poll;
 230                 if (poll)
 231                         rv = poll(file, pts);
 232                 unuse_pde(pde);
 233         }
 234         return rv;
 235 }
 236
 237 static long proc_reg_unlocked_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
 238 {
 239         struct proc_dir_entry *pde = PDE(file_inode(file));
 240         long rv = -ENOTTY;
 241         long (*ioctl)(struct file *, unsigned int, unsigned long);
 242         if (use_pde(pde)) {
 243                 ioctl = pde->proc_fops->unlocked_ioctl;
 244                 if (ioctl)
 245                         rv = ioctl(file, cmd, arg);
 246                 unuse_pde(pde);
 247         }
 248         return rv;
 249 }
 250
 251 #ifdef CONFIG_COMPAT
 252 static long proc_reg_compat_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
 253 {
 254         struct proc_dir_entry *pde = PDE(file_inode(file));
 255         long rv = -ENOTTY;
 256         long (*compat_ioctl)(struct file *, unsigned int, unsigned long);
 257         if (use_pde(pde)) {
 258                 compat_ioctl = pde->proc_fops->compat_ioctl;
 259                 if (compat_ioctl)
 260                         rv = compat_ioctl(file, cmd, arg);
 261                 unuse_pde(pde);
 262         }
 263         return rv;
 264 }
 265 #endif
 266
 267 static int proc_reg_mmap(struct file *file, struct vm_area_struct *vma)
 268 {
 269         struct proc_dir_entry *pde = PDE(file_inode(file));
 270         int rv = -EIO;
 271         int (*mmap)(struct file *, struct vm_area_struct *);
 272         if (use_pde(pde)) {
 273                 mmap = pde->proc_fops->mmap;
 274                 if (mmap)
 275                         rv = mmap(file, vma);
 276                 unuse_pde(pde);
 277         }
 278         return rv;
 279 }
 280
 281 static unsigned long
 282 proc_reg_get_unmapped_area(struct file *file, unsigned long orig_addr,
 283                            unsigned long len, unsigned long pgoff,
 284                            unsigned long flags)
 285 {
 286         struct proc_dir_entry *pde = PDE(file_inode(file));
 287         unsigned long rv = -EIO;
 288
 289         if (use_pde(pde)) {
 290                 typeof(proc_reg_get_unmapped_area) *get_area;
 291
 292                 get_area = pde->proc_fops->get_unmapped_area;
 293 #ifdef CONFIG_MMU
 294                 if (!get_area)
 295                         get_area = current->mm->get_unmapped_area;
 296 #endif
 297
 298                 if (get_area)
 299                         rv = get_area(file, orig_addr, len, pgoff, flags);
 300                 else
 301                         rv = orig_addr;
 302                 unuse_pde(pde);
 303         }
 304         return rv;
 305 }
 306
 307 static int proc_reg_open(struct inode *inode, struct file *file)
 308 {
 309         struct proc_dir_entry *pde = PDE(inode);
 310         int rv = 0;
 311         int (*open)(struct inode *, struct file *);
 312         int (*release)(struct inode *, struct file *);
 313         struct pde_opener *pdeo;
 314
 315         /*
 316          * What for, you ask? Well, we can have open, rmmod, remove_proc_entry
 317          * sequence. ->release won't be called because ->proc_fops will be
 318          * cleared. Depending on complexity of ->release, consequences vary.
 319          *
 320          * We can't wait for mercy when close will be done for real, it's
 321          * deadlockable: rmmod foo </proc/foo . So, we're going to do ->release
 322          * by hand in remove_proc_entry(). For this, save opener's credentials
 323          * for later.
 324          */
 325         pdeo = kzalloc(sizeof(struct pde_opener), GFP_KERNEL);
 326         if (!pdeo)
 327                 return -ENOMEM;
 328
 329         if (!use_pde(pde)) {
 330                 kfree(pdeo);
 331                 return -ENOENT;
 332         }
 333         open = pde->proc_fops->open;
 334         release = pde->proc_fops->release;
 335
 336         if (open)
 337                 rv = open(inode, file);
 338
 339         if (rv == 0 && release) {
 340                 /* To know what to release. */
 341                 pdeo->file = file;
 342                 /* Strictly for "too late" ->release in proc_reg_release(). */
 343                 spin_lock(&pde->pde_unload_lock);
 344                 list_add(&pdeo->lh, &pde->pde_openers);
 345                 spin_unlock(&pde->pde_unload_lock);
 346         } else
 347                 kfree(pdeo);
 348
 349         unuse_pde(pde);
 350         return rv;
 351 }
 352
 353 static int proc_reg_release(struct inode *inode, struct file *file)
 354 {
 355         struct proc_dir_entry *pde = PDE(inode);
 356         struct pde_opener *pdeo;
 357         spin_lock(&pde->pde_unload_lock);
 358         list_for_each_entry(pdeo, &pde->pde_openers, lh) {
 359                 if (pdeo->file == file) {
 360                         close_pdeo(pde, pdeo);
 361                         break;
 362                 }
 363         }
 364         spin_unlock(&pde->pde_unload_lock);
 365         return 0;
 366 }
 367
 368 static const struct file_operations proc_reg_file_ops = {
 369         .llseek         = proc_reg_llseek,
 370         .read           = proc_reg_read,
 371         .write          = proc_reg_write,
 372         .poll           = proc_reg_poll,
 373         .unlocked_ioctl = proc_reg_unlocked_ioctl,
 374 #ifdef CONFIG_COMPAT
 375         .compat_ioctl   = proc_reg_compat_ioctl,
 376 #endif
 377         .mmap           = proc_reg_mmap,
 378         .get_unmapped_area = proc_reg_get_unmapped_area,
 379         .open           = proc_reg_open,
 380         .release        = proc_reg_release,
 381 };
 382
 383 #ifdef CONFIG_COMPAT
 384 static const struct file_operations proc_reg_file_ops_no_compat = {
 385         .llseek         = proc_reg_llseek,
 386         .read           = proc_reg_read,
 387         .write          = proc_reg_write,
 388         .poll           = proc_reg_poll,
 389         .unlocked_ioctl = proc_reg_unlocked_ioctl,
 390         .mmap           = proc_reg_mmap,
 391         .get_unmapped_area = proc_reg_get_unmapped_area,
 392         .open           = proc_reg_open,
 393         .release        = proc_reg_release,
 394 };
 395 #endif
 396
 397 static void proc_put_link(void *p)
 398 {
 399         unuse_pde(p);
 400 }
 401
 402 static const char *proc_get_link(struct dentry *dentry,
 403                                  struct inode *inode,
 404                                  struct delayed_call *done)
 405 {
 406         struct proc_dir_entry *pde = PDE(inode);
 407         if (unlikely(!use_pde(pde)))
 408                 return ERR_PTR(-EINVAL);
 409         set_delayed_call(done, proc_put_link, pde);
 410         return pde->data;
 411 }
 412
 413 const struct inode_operations proc_link_inode_operations = {
 414         .readlink       = generic_readlink,
 415         .get_link       = proc_get_link,
 416 };
 417
 418 struct inode *proc_get_inode(struct super_block *sb, struct proc_dir_entry *de)
 419 {
 420         struct inode *inode = new_inode_pseudo(sb);
 421
 422         if (inode) {
 423                 inode->i_ino = de->low_ino;
 424                 inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
 425                 PROC_I(inode)->pde = de;
 426
 427                 if (is_empty_pde(de)) {
 428                         make_empty_dir_inode(inode);
 429                         return inode;
 430                 }
 431                 if (de->mode) {
 432                         inode->i_mode = de->mode;
 433                         inode->i_uid = de->uid;
 434                         inode->i_gid = de->gid;
 435                 }
 436                 if (de->size)
 437                         inode->i_size = de->size;
 438                 if (de->nlink)
 439                         set_nlink(inode, de->nlink);
 440                 WARN_ON(!de->proc_iops);
 441                 inode->i_op = de->proc_iops;
 442                 if (de->proc_fops) {
 443                         if (S_ISREG(inode->i_mode)) {
 444 #ifdef CONFIG_COMPAT
 445                                 if (!de->proc_fops->compat_ioctl)
 446                                         inode->i_fop =
 447                                                 &proc_reg_file_ops_no_compat;
 448                                 else
 449 #endif
 450                                         inode->i_fop = &proc_reg_file_ops;
 451                         } else {
 452                                 inode->i_fop = de->proc_fops;
 453                         }
 454                 }
 455         } else
 456                pde_put(de);
 457         return inode;
 458 }
 459
 460 int proc_fill_super(struct super_block *s, void *data, int silent)
 461 {
 462         struct pid_namespace *ns = get_pid_ns(s->s_fs_info);
 463         struct inode *root_inode;
 464         int ret;
 465
 466         if (!proc_parse_options(data, ns))
 467                 return -EINVAL;
 468
 469         /* User space would break if executables or devices appear on proc */
 470         s->s_iflags |= SB_I_USERNS_VISIBLE | SB_I_NOEXEC | SB_I_NODEV;
 471         s->s_flags |= MS_NODIRATIME | MS_NOSUID | MS_NOEXEC;
 472         s->s_blocksize = 1024;
 473         s->s_blocksize_bits = 10;
 474         s->s_magic = PROC_SUPER_MAGIC;
 475         s->s_op = &proc_sops;
 476         s->s_time_gran = 1;
 477
 478         /*
 479          * procfs isn't actually a stacking filesystem; however, there is
 480          * too much magic going on inside it to permit stacking things on
 481          * top of it
 482          */
 483         s->s_stack_depth = FILESYSTEM_MAX_STACK_DEPTH;
 484
 485         pde_get(&proc_root);
 486         root_inode = proc_get_inode(s, &proc_root);
 487         if (!root_inode) {
 488                 pr_err("proc_fill_super: get root inode failed\n");
 489                 return -ENOMEM;
 490         }
 491
 492         s->s_root = d_make_root(root_inode);
 493         if (!s->s_root) {
 494                 pr_err("proc_fill_super: allocate dentry failed\n");
 495                 return -ENOMEM;
 496         }
 497
 498         ret = proc_setup_self(s);
 499         if (ret) {
 500                 return ret;
 501         }
 502         return proc_setup_thread_self(s);
 503 }