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
  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         void *ns;
  36
  37         truncate_inode_pages(&inode->i_data, 0);
  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_assign_pointer(PROC_I(inode)->sysctl, NULL);
  50                 sysctl_head_put(head);
  51         }
  52         /* Release any associated namespace */
  53         ns_ops = PROC_I(inode)->ns_ops;
  54         ns = PROC_I(inode)->ns;
  55         if (ns_ops && ns)
  56                 ns_ops->put(ns);
  57 }
  58
  59 static struct kmem_cache * proc_inode_cachep;
  60
  61 static struct inode *proc_alloc_inode(struct super_block *sb)
  62 {
  63         struct proc_inode *ei;
  64         struct inode *inode;
  65
  66         ei = (struct proc_inode *)kmem_cache_alloc(proc_inode_cachep, GFP_KERNEL);
  67         if (!ei)
  68                 return NULL;
  69         ei->pid = NULL;
  70         ei->fd = 0;
  71         ei->op.proc_get_link = NULL;
  72         ei->pde = NULL;
  73         ei->sysctl = NULL;
  74         ei->sysctl_entry = NULL;
  75         ei->ns = NULL;
  76         ei->ns_ops = NULL;
  77         inode = &ei->vfs_inode;
  78         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
  79         return inode;
  80 }
  81
  82 static void proc_i_callback(struct rcu_head *head)
  83 {
  84         struct inode *inode = container_of(head, struct inode, i_rcu);
  85         kmem_cache_free(proc_inode_cachep, PROC_I(inode));
  86 }
  87
  88 static void proc_destroy_inode(struct inode *inode)
  89 {
  90         call_rcu(&inode->i_rcu, proc_i_callback);
  91 }
  92
  93 static void init_once(void *foo)
  94 {
  95         struct proc_inode *ei = (struct proc_inode *) foo;
  96
  97         inode_init_once(&ei->vfs_inode);
  98 }
  99
 100 void __init proc_init_inodecache(void)
 101 {
 102         proc_inode_cachep = kmem_cache_create("proc_inode_cache",
 103                                              sizeof(struct proc_inode),
 104                                              0, (SLAB_RECLAIM_ACCOUNT|
 105                                                 SLAB_MEM_SPREAD|SLAB_PANIC),
 106                                              init_once);
 107 }
 108
 109 static int proc_show_options(struct seq_file *seq, struct dentry *root)
 110 {
 111         struct super_block *sb = root->d_sb;
 112         struct pid_namespace *pid = sb->s_fs_info;
 113
 114         if (!gid_eq(pid->pid_gid, GLOBAL_ROOT_GID))
 115                 seq_printf(seq, ",gid=%u", from_kgid_munged(&init_user_ns, pid->pid_gid));
 116         if (pid->hide_pid != 0)
 117                 seq_printf(seq, ",hidepid=%u", pid->hide_pid);
 118
 119         return 0;
 120 }
 121
 122 static const struct super_operations proc_sops = {
 123         .alloc_inode    = proc_alloc_inode,
 124         .destroy_inode  = proc_destroy_inode,
 125         .drop_inode     = generic_delete_inode,
 126         .evict_inode    = proc_evict_inode,
 127         .statfs         = simple_statfs,
 128         .remount_fs     = proc_remount,
 129         .show_options   = proc_show_options,
 130 };
 131
 132 enum {BIAS = -1U<<31};
 133
 134 static inline int use_pde(struct proc_dir_entry *pde)
 135 {
 136         return atomic_inc_unless_negative(&pde->in_use);
 137 }
 138
 139 static void unuse_pde(struct proc_dir_entry *pde)
 140 {
 141         if (atomic_dec_return(&pde->in_use) == BIAS)
 142                 complete(pde->pde_unload_completion);
 143 }
 144
 145 /* pde is locked */
 146 static void close_pdeo(struct proc_dir_entry *pde, struct pde_opener *pdeo)
 147 {
 148         if (pdeo->closing) {
 149                 /* somebody else is doing that, just wait */
 150                 DECLARE_COMPLETION_ONSTACK(c);
 151                 pdeo->c = &c;
 152                 spin_unlock(&pde->pde_unload_lock);
 153                 wait_for_completion(&c);
 154                 spin_lock(&pde->pde_unload_lock);
 155         } else {
 156                 struct file *file;
 157                 pdeo->closing = 1;
 158                 spin_unlock(&pde->pde_unload_lock);
 159                 file = pdeo->file;
 160                 pde->proc_fops->release(file_inode(file), file);
 161                 spin_lock(&pde->pde_unload_lock);
 162                 list_del_init(&pdeo->lh);
 163                 if (pdeo->c)
 164                         complete(pdeo->c);
 165                 kfree(pdeo);
 166         }
 167 }
 168
 169 void proc_entry_rundown(struct proc_dir_entry *de)
 170 {
 171         DECLARE_COMPLETION_ONSTACK(c);
 172         /* Wait until all existing callers into module are done. */
 173         de->pde_unload_completion = &c;
 174         if (atomic_add_return(BIAS, &de->in_use) != BIAS)
 175                 wait_for_completion(&c);
 176
 177         spin_lock(&de->pde_unload_lock);
 178         while (!list_empty(&de->pde_openers)) {
 179                 struct pde_opener *pdeo;
 180                 pdeo = list_first_entry(&de->pde_openers, struct pde_opener, lh);
 181                 close_pdeo(de, pdeo);
 182         }
 183         spin_unlock(&de->pde_unload_lock);
 184 }
 185
 186 static loff_t proc_reg_llseek(struct file *file, loff_t offset, int whence)
 187 {
 188         struct proc_dir_entry *pde = PDE(file_inode(file));
 189         loff_t rv = -EINVAL;
 190         if (use_pde(pde)) {
 191                 loff_t (*llseek)(struct file *, loff_t, int);
 192                 llseek = pde->proc_fops->llseek;
 193                 if (!llseek)
 194                         llseek = default_llseek;
 195                 rv = llseek(file, offset, whence);
 196                 unuse_pde(pde);
 197         }
 198         return rv;
 199 }
 200
 201 static ssize_t proc_reg_read(struct file *file, char __user *buf, size_t count, loff_t *ppos)
 202 {
 203         ssize_t (*read)(struct file *, char __user *, size_t, loff_t *);
 204         struct proc_dir_entry *pde = PDE(file_inode(file));
 205         ssize_t rv = -EIO;
 206         if (use_pde(pde)) {
 207                 read = pde->proc_fops->read;
 208                 if (read)
 209                         rv = read(file, buf, count, ppos);
 210                 unuse_pde(pde);
 211         }
 212         return rv;
 213 }
 214
 215 static ssize_t proc_reg_write(struct file *file, const char __user *buf, size_t count, loff_t *ppos)
 216 {
 217         ssize_t (*write)(struct file *, const char __user *, size_t, loff_t *);
 218         struct proc_dir_entry *pde = PDE(file_inode(file));
 219         ssize_t rv = -EIO;
 220         if (use_pde(pde)) {
 221                 write = pde->proc_fops->write;
 222                 if (write)
 223                         rv = write(file, buf, count, ppos);
 224                 unuse_pde(pde);
 225         }
 226         return rv;
 227 }
 228
 229 static unsigned int proc_reg_poll(struct file *file, struct poll_table_struct *pts)
 230 {
 231         struct proc_dir_entry *pde = PDE(file_inode(file));
 232         unsigned int rv = DEFAULT_POLLMASK;
 233         unsigned int (*poll)(struct file *, struct poll_table_struct *);
 234         if (use_pde(pde)) {
 235                 poll = pde->proc_fops->poll;
 236                 if (poll)
 237                         rv = poll(file, pts);
 238                 unuse_pde(pde);
 239         }
 240         return rv;
 241 }
 242
 243 static long proc_reg_unlocked_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
 244 {
 245         struct proc_dir_entry *pde = PDE(file_inode(file));
 246         long rv = -ENOTTY;
 247         long (*ioctl)(struct file *, unsigned int, unsigned long);
 248         if (use_pde(pde)) {
 249                 ioctl = pde->proc_fops->unlocked_ioctl;
 250                 if (ioctl)
 251                         rv = ioctl(file, cmd, arg);
 252                 unuse_pde(pde);
 253         }
 254         return rv;
 255 }
 256
 257 #ifdef CONFIG_COMPAT
 258 static long proc_reg_compat_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
 259 {
 260         struct proc_dir_entry *pde = PDE(file_inode(file));
 261         long rv = -ENOTTY;
 262         long (*compat_ioctl)(struct file *, unsigned int, unsigned long);
 263         if (use_pde(pde)) {
 264                 compat_ioctl = pde->proc_fops->compat_ioctl;
 265                 if (compat_ioctl)
 266                         rv = compat_ioctl(file, cmd, arg);
 267                 unuse_pde(pde);
 268         }
 269         return rv;
 270 }
 271 #endif
 272
 273 static int proc_reg_mmap(struct file *file, struct vm_area_struct *vma)
 274 {
 275         struct proc_dir_entry *pde = PDE(file_inode(file));
 276         int rv = -EIO;
 277         int (*mmap)(struct file *, struct vm_area_struct *);
 278         if (use_pde(pde)) {
 279                 mmap = pde->proc_fops->mmap;
 280                 if (mmap)
 281                         rv = mmap(file, vma);
 282                 unuse_pde(pde);
 283         }
 284         return rv;
 285 }
 286
 287 static int proc_reg_open(struct inode *inode, struct file *file)
 288 {
 289         struct proc_dir_entry *pde = PDE(inode);
 290         int rv = 0;
 291         int (*open)(struct inode *, struct file *);
 292         int (*release)(struct inode *, struct file *);
 293         struct pde_opener *pdeo;
 294
 295         /*
 296          * What for, you ask? Well, we can have open, rmmod, remove_proc_entry
 297          * sequence. ->release won't be called because ->proc_fops will be
 298          * cleared. Depending on complexity of ->release, consequences vary.
 299          *
 300          * We can't wait for mercy when close will be done for real, it's
 301          * deadlockable: rmmod foo </proc/foo . So, we're going to do ->release
 302          * by hand in remove_proc_entry(). For this, save opener's credentials
 303          * for later.
 304          */
 305         pdeo = kzalloc(sizeof(struct pde_opener), GFP_KERNEL);
 306         if (!pdeo)
 307                 return -ENOMEM;
 308
 309         if (!use_pde(pde)) {
 310                 kfree(pdeo);
 311                 return -ENOENT;
 312         }
 313         open = pde->proc_fops->open;
 314         release = pde->proc_fops->release;
 315
 316         if (open)
 317                 rv = open(inode, file);
 318
 319         if (rv == 0 && release) {
 320                 /* To know what to release. */
 321                 pdeo->file = file;
 322                 /* Strictly for "too late" ->release in proc_reg_release(). */
 323                 spin_lock(&pde->pde_unload_lock);
 324                 list_add(&pdeo->lh, &pde->pde_openers);
 325                 spin_unlock(&pde->pde_unload_lock);
 326         } else
 327                 kfree(pdeo);
 328
 329         unuse_pde(pde);
 330         return rv;
 331 }
 332
 333 static int proc_reg_release(struct inode *inode, struct file *file)
 334 {
 335         struct proc_dir_entry *pde = PDE(inode);
 336         struct pde_opener *pdeo;
 337         spin_lock(&pde->pde_unload_lock);
 338         list_for_each_entry(pdeo, &pde->pde_openers, lh) {
 339                 if (pdeo->file == file) {
 340                         close_pdeo(pde, pdeo);
 341                         break;
 342                 }
 343         }
 344         spin_unlock(&pde->pde_unload_lock);
 345         return 0;
 346 }
 347
 348 static const struct file_operations proc_reg_file_ops = {
 349         .llseek         = proc_reg_llseek,
 350         .read           = proc_reg_read,
 351         .write          = proc_reg_write,
 352         .poll           = proc_reg_poll,
 353         .unlocked_ioctl = proc_reg_unlocked_ioctl,
 354 #ifdef CONFIG_COMPAT
 355         .compat_ioctl   = proc_reg_compat_ioctl,
 356 #endif
 357         .mmap           = proc_reg_mmap,
 358         .open           = proc_reg_open,
 359         .release        = proc_reg_release,
 360 };
 361
 362 #ifdef CONFIG_COMPAT
 363 static const struct file_operations proc_reg_file_ops_no_compat = {
 364         .llseek         = proc_reg_llseek,
 365         .read           = proc_reg_read,
 366         .write          = proc_reg_write,
 367         .poll           = proc_reg_poll,
 368         .unlocked_ioctl = proc_reg_unlocked_ioctl,
 369         .mmap           = proc_reg_mmap,
 370         .open           = proc_reg_open,
 371         .release        = proc_reg_release,
 372 };
 373 #endif
 374
 375 struct inode *proc_get_inode(struct super_block *sb, struct proc_dir_entry *de)
 376 {
 377         struct inode *inode = new_inode_pseudo(sb);
 378
 379         if (inode) {
 380                 inode->i_ino = de->low_ino;
 381                 inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
 382                 PROC_I(inode)->pde = de;
 383
 384                 if (de->mode) {
 385                         inode->i_mode = de->mode;
 386                         inode->i_uid = de->uid;
 387                         inode->i_gid = de->gid;
 388                 }
 389                 if (de->size)
 390                         inode->i_size = de->size;
 391                 if (de->nlink)
 392                         set_nlink(inode, de->nlink);
 393                 WARN_ON(!de->proc_iops);
 394                 inode->i_op = de->proc_iops;
 395                 if (de->proc_fops) {
 396                         if (S_ISREG(inode->i_mode)) {
 397 #ifdef CONFIG_COMPAT
 398                                 if (!de->proc_fops->compat_ioctl)
 399                                         inode->i_fop =
 400                                                 &proc_reg_file_ops_no_compat;
 401                                 else
 402 #endif
 403                                         inode->i_fop = &proc_reg_file_ops;
 404                         } else {
 405                                 inode->i_fop = de->proc_fops;
 406                         }
 407                 }
 408         } else
 409                pde_put(de);
 410         return inode;
 411 }
 412
 413 int proc_fill_super(struct super_block *s)
 414 {
 415         struct inode *root_inode;
 416
 417         s->s_flags |= MS_NODIRATIME | MS_NOSUID | MS_NOEXEC;
 418         s->s_blocksize = 1024;
 419         s->s_blocksize_bits = 10;
 420         s->s_magic = PROC_SUPER_MAGIC;
 421         s->s_op = &proc_sops;
 422         s->s_time_gran = 1;
 423
 424         pde_get(&proc_root);
 425         root_inode = proc_get_inode(s, &proc_root);
 426         if (!root_inode) {
 427                 pr_err("proc_fill_super: get root inode failed\n");
 428                 return -ENOMEM;
 429         }
 430
 431         s->s_root = d_make_root(root_inode);
 432         if (!s->s_root) {
 433                 pr_err("proc_fill_super: allocate dentry failed\n");
 434                 return -ENOMEM;
 435         }
 436
 437         return proc_setup_self(s);
 438 }