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 #include <linux/namei.h>
  27
  28 #include <asm/uaccess.h>
  29
  30 #include "internal.h"
  31
  32 static void proc_evict_inode(struct inode *inode)
  33 {
  34         struct proc_dir_entry *de;
  35         struct ctl_table_header *head;
  36         const struct proc_ns_operations *ns_ops;
  37         void *ns;
  38
  39         truncate_inode_pages(&inode->i_data, 0);
  40         clear_inode(inode);
  41
  42         /* Stop tracking associated processes */
  43         put_pid(PROC_I(inode)->pid);
  44
  45         /* Let go of any associated proc directory entry */
  46         de = PROC_I(inode)->pde;
  47         if (de)
  48                 pde_put(de);
  49         head = PROC_I(inode)->sysctl;
  50         if (head) {
  51                 rcu_assign_pointer(PROC_I(inode)->sysctl, NULL);
  52                 sysctl_head_put(head);
  53         }
  54         /* Release any associated namespace */
  55         ns_ops = PROC_I(inode)->ns.ns_ops;
  56         ns = PROC_I(inode)->ns.ns;
  57         if (ns_ops && ns)
  58                 ns_ops->put(ns);
  59 }
  60
  61 static struct kmem_cache * proc_inode_cachep;
  62
  63 static struct inode *proc_alloc_inode(struct super_block *sb)
  64 {
  65         struct proc_inode *ei;
  66         struct inode *inode;
  67
  68         ei = (struct proc_inode *)kmem_cache_alloc(proc_inode_cachep, GFP_KERNEL);
  69         if (!ei)
  70                 return NULL;
  71         ei->pid = NULL;
  72         ei->fd = 0;
  73         ei->op.proc_get_link = NULL;
  74         ei->pde = NULL;
  75         ei->sysctl = NULL;
  76         ei->sysctl_entry = NULL;
  77         ei->ns.ns = NULL;
  78         ei->ns.ns_ops = NULL;
  79         inode = &ei->vfs_inode;
  80         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
  81         return inode;
  82 }
  83
  84 static void proc_i_callback(struct rcu_head *head)
  85 {
  86         struct inode *inode = container_of(head, struct inode, i_rcu);
  87         kmem_cache_free(proc_inode_cachep, PROC_I(inode));
  88 }
  89
  90 static void proc_destroy_inode(struct inode *inode)
  91 {
  92         call_rcu(&inode->i_rcu, proc_i_callback);
  93 }
  94
  95 static void init_once(void *foo)
  96 {
  97         struct proc_inode *ei = (struct proc_inode *) foo;
  98
  99         inode_init_once(&ei->vfs_inode);
 100 }
 101
 102 void __init proc_init_inodecache(void)
 103 {
 104         proc_inode_cachep = kmem_cache_create("proc_inode_cache",
 105                                              sizeof(struct proc_inode),
 106                                              0, (SLAB_RECLAIM_ACCOUNT|
 107                                                 SLAB_MEM_SPREAD|SLAB_PANIC),
 108                                              init_once);
 109 }
 110
 111 static int proc_show_options(struct seq_file *seq, struct dentry *root)
 112 {
 113         struct super_block *sb = root->d_sb;
 114         struct pid_namespace *pid = sb->s_fs_info;
 115
 116         if (!gid_eq(pid->pid_gid, GLOBAL_ROOT_GID))
 117                 seq_printf(seq, ",gid=%u", from_kgid_munged(&init_user_ns, pid->pid_gid));
 118         if (pid->hide_pid != 0)
 119                 seq_printf(seq, ",hidepid=%u", pid->hide_pid);
 120
 121         return 0;
 122 }
 123
 124 static const struct super_operations proc_sops = {
 125         .alloc_inode    = proc_alloc_inode,
 126         .destroy_inode  = proc_destroy_inode,
 127         .drop_inode     = generic_delete_inode,
 128         .evict_inode    = proc_evict_inode,
 129         .statfs         = simple_statfs,
 130         .remount_fs     = proc_remount,
 131         .show_options   = proc_show_options,
 132 };
 133
 134 enum {BIAS = -1U<<31};
 135
 136 static inline int use_pde(struct proc_dir_entry *pde)
 137 {
 138         return atomic_inc_unless_negative(&pde->in_use);
 139 }
 140
 141 static void unuse_pde(struct proc_dir_entry *pde)
 142 {
 143         if (atomic_dec_return(&pde->in_use) == BIAS)
 144                 complete(pde->pde_unload_completion);
 145 }
 146
 147 /* pde is locked */
 148 static void close_pdeo(struct proc_dir_entry *pde, struct pde_opener *pdeo)
 149 {
 150         if (pdeo->closing) {
 151                 /* somebody else is doing that, just wait */
 152                 DECLARE_COMPLETION_ONSTACK(c);
 153                 pdeo->c = &c;
 154                 spin_unlock(&pde->pde_unload_lock);
 155                 wait_for_completion(&c);
 156                 spin_lock(&pde->pde_unload_lock);
 157         } else {
 158                 struct file *file;
 159                 pdeo->closing = 1;
 160                 spin_unlock(&pde->pde_unload_lock);
 161                 file = pdeo->file;
 162                 pde->proc_fops->release(file_inode(file), file);
 163                 spin_lock(&pde->pde_unload_lock);
 164                 list_del_init(&pdeo->lh);
 165                 if (pdeo->c)
 166                         complete(pdeo->c);
 167                 kfree(pdeo);
 168         }
 169 }
 170
 171 void proc_entry_rundown(struct proc_dir_entry *de)
 172 {
 173         DECLARE_COMPLETION_ONSTACK(c);
 174         /* Wait until all existing callers into module are done. */
 175         de->pde_unload_completion = &c;
 176         if (atomic_add_return(BIAS, &de->in_use) != BIAS)
 177                 wait_for_completion(&c);
 178
 179         spin_lock(&de->pde_unload_lock);
 180         while (!list_empty(&de->pde_openers)) {
 181                 struct pde_opener *pdeo;
 182                 pdeo = list_first_entry(&de->pde_openers, struct pde_opener, lh);
 183                 close_pdeo(de, pdeo);
 184         }
 185         spin_unlock(&de->pde_unload_lock);
 186 }
 187
 188 static loff_t proc_reg_llseek(struct file *file, loff_t offset, int whence)
 189 {
 190         struct proc_dir_entry *pde = PDE(file_inode(file));
 191         loff_t rv = -EINVAL;
 192         if (use_pde(pde)) {
 193                 loff_t (*llseek)(struct file *, loff_t, int);
 194                 llseek = pde->proc_fops->llseek;
 195                 if (!llseek)
 196                         llseek = default_llseek;
 197                 rv = llseek(file, offset, whence);
 198                 unuse_pde(pde);
 199         }
 200         return rv;
 201 }
 202
 203 static ssize_t proc_reg_read(struct file *file, char __user *buf, size_t count, loff_t *ppos)
 204 {
 205         ssize_t (*read)(struct file *, char __user *, size_t, loff_t *);
 206         struct proc_dir_entry *pde = PDE(file_inode(file));
 207         ssize_t rv = -EIO;
 208         if (use_pde(pde)) {
 209                 read = pde->proc_fops->read;
 210                 if (read)
 211                         rv = read(file, buf, count, ppos);
 212                 unuse_pde(pde);
 213         }
 214         return rv;
 215 }
 216
 217 static ssize_t proc_reg_write(struct file *file, const char __user *buf, size_t count, loff_t *ppos)
 218 {
 219         ssize_t (*write)(struct file *, const char __user *, size_t, loff_t *);
 220         struct proc_dir_entry *pde = PDE(file_inode(file));
 221         ssize_t rv = -EIO;
 222         if (use_pde(pde)) {
 223                 write = pde->proc_fops->write;
 224                 if (write)
 225                         rv = write(file, buf, count, ppos);
 226                 unuse_pde(pde);
 227         }
 228         return rv;
 229 }
 230
 231 static unsigned int proc_reg_poll(struct file *file, struct poll_table_struct *pts)
 232 {
 233         struct proc_dir_entry *pde = PDE(file_inode(file));
 234         unsigned int rv = DEFAULT_POLLMASK;
 235         unsigned int (*poll)(struct file *, struct poll_table_struct *);
 236         if (use_pde(pde)) {
 237                 poll = pde->proc_fops->poll;
 238                 if (poll)
 239                         rv = poll(file, pts);
 240                 unuse_pde(pde);
 241         }
 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_inode(file));
 248         long rv = -ENOTTY;
 249         long (*ioctl)(struct file *, unsigned int, unsigned long);
 250         if (use_pde(pde)) {
 251                 ioctl = pde->proc_fops->unlocked_ioctl;
 252                 if (ioctl)
 253                         rv = ioctl(file, cmd, arg);
 254                 unuse_pde(pde);
 255         }
 256         return rv;
 257 }
 258
 259 #ifdef CONFIG_COMPAT
 260 static long proc_reg_compat_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
 261 {
 262         struct proc_dir_entry *pde = PDE(file_inode(file));
 263         long rv = -ENOTTY;
 264         long (*compat_ioctl)(struct file *, unsigned int, unsigned long);
 265         if (use_pde(pde)) {
 266                 compat_ioctl = pde->proc_fops->compat_ioctl;
 267                 if (compat_ioctl)
 268                         rv = compat_ioctl(file, cmd, arg);
 269                 unuse_pde(pde);
 270         }
 271         return rv;
 272 }
 273 #endif
 274
 275 static int proc_reg_mmap(struct file *file, struct vm_area_struct *vma)
 276 {
 277         struct proc_dir_entry *pde = PDE(file_inode(file));
 278         int rv = -EIO;
 279         int (*mmap)(struct file *, struct vm_area_struct *);
 280         if (use_pde(pde)) {
 281                 mmap = pde->proc_fops->mmap;
 282                 if (mmap)
 283                         rv = mmap(file, vma);
 284                 unuse_pde(pde);
 285         }
 286         return rv;
 287 }
 288
 289 static unsigned long proc_reg_get_unmapped_area(struct file *file, unsigned long orig_addr, unsigned long len, unsigned long pgoff, unsigned long flags)
 290 {
 291         struct proc_dir_entry *pde = PDE(file_inode(file));
 292         unsigned long rv = -EIO;
 293         unsigned long (*get_unmapped_area)(struct file *, unsigned long, unsigned long, unsigned long, unsigned long) = NULL;
 294         if (use_pde(pde)) {
 295 #ifdef CONFIG_MMU
 296                 get_unmapped_area = current->mm->get_unmapped_area;
 297 #endif
 298                 if (pde->proc_fops->get_unmapped_area)
 299                         get_unmapped_area = pde->proc_fops->get_unmapped_area;
 300                 if (get_unmapped_area)
 301                         rv = get_unmapped_area(file, orig_addr, len, pgoff, flags);
 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_follow_link(struct dentry *dentry, struct nameidata *nd)
 398 {
 399         struct proc_dir_entry *pde = PDE(dentry->d_inode);
 400         if (unlikely(!use_pde(pde)))
 401                 return ERR_PTR(-EINVAL);
 402         nd_set_link(nd, pde->data);
 403         return pde;
 404 }
 405
 406 static void proc_put_link(struct dentry *dentry, struct nameidata *nd, void *p)
 407 {
 408         unuse_pde(p);
 409 }
 410
 411 const struct inode_operations proc_link_inode_operations = {
 412         .readlink       = generic_readlink,
 413         .follow_link    = proc_follow_link,
 414         .put_link       = proc_put_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;
 424                 PROC_I(inode)->pde = de;
 425
 426                 if (de->mode) {
 427                         inode->i_mode = de->mode;
 428                         inode->i_uid = de->uid;
 429                         inode->i_gid = de->gid;
 430                 }
 431                 if (de->size)
 432                         inode->i_size = de->size;
 433                 if (de->nlink)
 434                         set_nlink(inode, de->nlink);
 435                 WARN_ON(!de->proc_iops);
 436                 inode->i_op = de->proc_iops;
 437                 if (de->proc_fops) {
 438                         if (S_ISREG(inode->i_mode)) {
 439 #ifdef CONFIG_COMPAT
 440                                 if (!de->proc_fops->compat_ioctl)
 441                                         inode->i_fop =
 442                                                 &proc_reg_file_ops_no_compat;
 443                                 else
 444 #endif
 445                                         inode->i_fop = &proc_reg_file_ops;
 446                         } else {
 447                                 inode->i_fop = de->proc_fops;
 448                         }
 449                 }
 450         } else
 451                pde_put(de);
 452         return inode;
 453 }
 454
 455 int proc_fill_super(struct super_block *s)
 456 {
 457         struct inode *root_inode;
 458
 459         s->s_flags |= MS_NODIRATIME | MS_NOSUID | MS_NOEXEC;
 460         s->s_blocksize = 1024;
 461         s->s_blocksize_bits = 10;
 462         s->s_magic = PROC_SUPER_MAGIC;
 463         s->s_op = &proc_sops;
 464         s->s_time_gran = 1;
 465
 466         pde_get(&proc_root);
 467         root_inode = proc_get_inode(s, &proc_root);
 468         if (!root_inode) {
 469                 pr_err("proc_fill_super: get root inode failed\n");
 470                 return -ENOMEM;
 471         }
 472
 473         s->s_root = d_make_root(root_inode);
 474         if (!s->s_root) {
 475                 pr_err("proc_fill_super: allocate dentry failed\n");
 476                 return -ENOMEM;
 477         }
 478
 479         return proc_setup_self(s);
 480 }