net/sunrpc/rpc_pipe.c

   1 /*
   2  * net/sunrpc/rpc_pipe.c
   3  *
   4  * Userland/kernel interface for rpcauth_gss.
   5  * Code shamelessly plagiarized from fs/nfsd/nfsctl.c
   6  * and fs/sysfs/inode.c
   7  *
   8  * Copyright (c) 2002, Trond Myklebust <trond.myklebust@fys.uio.no>
   9  *
  10  */
  11 #include <linux/module.h>
  12 #include <linux/slab.h>
  13 #include <linux/string.h>
  14 #include <linux/pagemap.h>
  15 #include <linux/mount.h>
  16 #include <linux/namei.h>
  17 #include <linux/fsnotify.h>
  18 #include <linux/kernel.h>
  19 #include <linux/rcupdate.h>
  20
  21 #include <asm/ioctls.h>
  22 #include <linux/poll.h>
  23 #include <linux/wait.h>
  24 #include <linux/seq_file.h>
  25
  26 #include <linux/sunrpc/clnt.h>
  27 #include <linux/workqueue.h>
  28 #include <linux/sunrpc/rpc_pipe_fs.h>
  29 #include <linux/sunrpc/cache.h>
  30 #include <linux/nsproxy.h>
  31 #include <linux/notifier.h>
  32
  33 #include "netns.h"
  34 #include "sunrpc.h"
  35
  36 #define RPCDBG_FACILITY RPCDBG_DEBUG
  37
  38 #define NET_NAME(net)   ((net == &init_net) ? " (init_net)" : "")
  39
  40 static struct file_system_type rpc_pipe_fs_type;
  41
  42
  43 static struct kmem_cache *rpc_inode_cachep __read_mostly;
  44
  45 #define RPC_UPCALL_TIMEOUT (30*HZ)
  46
  47 static BLOCKING_NOTIFIER_HEAD(rpc_pipefs_notifier_list);
  48
  49 int rpc_pipefs_notifier_register(struct notifier_block *nb)
  50 {
  51         return blocking_notifier_chain_cond_register(&rpc_pipefs_notifier_list, nb);
  52 }
  53 EXPORT_SYMBOL_GPL(rpc_pipefs_notifier_register);
  54
  55 void rpc_pipefs_notifier_unregister(struct notifier_block *nb)
  56 {
  57         blocking_notifier_chain_unregister(&rpc_pipefs_notifier_list, nb);
  58 }
  59 EXPORT_SYMBOL_GPL(rpc_pipefs_notifier_unregister);
  60
  61 static void rpc_purge_list(wait_queue_head_t *waitq, struct list_head *head,
  62                 void (*destroy_msg)(struct rpc_pipe_msg *), int err)
  63 {
  64         struct rpc_pipe_msg *msg;
  65
  66         if (list_empty(head))
  67                 return;
  68         do {
  69                 msg = list_entry(head->next, struct rpc_pipe_msg, list);
  70                 list_del_init(&msg->list);
  71                 msg->errno = err;
  72                 destroy_msg(msg);
  73         } while (!list_empty(head));
  74
  75         if (waitq)
  76                 wake_up(waitq);
  77 }
  78
  79 static void
  80 rpc_timeout_upcall_queue(struct work_struct *work)
  81 {
  82         LIST_HEAD(free_list);
  83         struct rpc_pipe *pipe =
  84                 container_of(work, struct rpc_pipe, queue_timeout.work);
  85         void (*destroy_msg)(struct rpc_pipe_msg *);
  86         struct dentry *dentry;
  87
  88         spin_lock(&pipe->lock);
  89         destroy_msg = pipe->ops->destroy_msg;
  90         if (pipe->nreaders == 0) {
  91                 list_splice_init(&pipe->pipe, &free_list);
  92                 pipe->pipelen = 0;
  93         }
  94         dentry = dget(pipe->dentry);
  95         spin_unlock(&pipe->lock);
  96         rpc_purge_list(dentry ? &RPC_I(dentry->d_inode)->waitq : NULL,
  97                         &free_list, destroy_msg, -ETIMEDOUT);
  98         dput(dentry);
  99 }
 100
 101 ssize_t rpc_pipe_generic_upcall(struct file *filp, struct rpc_pipe_msg *msg,
 102                                 char __user *dst, size_t buflen)
 103 {
 104         char *data = (char *)msg->data + msg->copied;
 105         size_t mlen = min(msg->len - msg->copied, buflen);
 106         unsigned long left;
 107
 108         left = copy_to_user(dst, data, mlen);
 109         if (left == mlen) {
 110                 msg->errno = -EFAULT;
 111                 return -EFAULT;
 112         }
 113
 114         mlen -= left;
 115         msg->copied += mlen;
 116         msg->errno = 0;
 117         return mlen;
 118 }
 119 EXPORT_SYMBOL_GPL(rpc_pipe_generic_upcall);
 120
 121 /**
 122  * rpc_queue_upcall - queue an upcall message to userspace
 123  * @pipe: upcall pipe on which to queue given message
 124  * @msg: message to queue
 125  *
 126  * Call with an @inode created by rpc_mkpipe() to queue an upcall.
 127  * A userspace process may then later read the upcall by performing a
 128  * read on an open file for this inode.  It is up to the caller to
 129  * initialize the fields of @msg (other than @msg->list) appropriately.
 130  */
 131 int
 132 rpc_queue_upcall(struct rpc_pipe *pipe, struct rpc_pipe_msg *msg)
 133 {
 134         int res = -EPIPE;
 135         struct dentry *dentry;
 136
 137         spin_lock(&pipe->lock);
 138         if (pipe->nreaders) {
 139                 list_add_tail(&msg->list, &pipe->pipe);
 140                 pipe->pipelen += msg->len;
 141                 res = 0;
 142         } else if (pipe->flags & RPC_PIPE_WAIT_FOR_OPEN) {
 143                 if (list_empty(&pipe->pipe))
 144                         queue_delayed_work(rpciod_workqueue,
 145                                         &pipe->queue_timeout,
 146                                         RPC_UPCALL_TIMEOUT);
 147                 list_add_tail(&msg->list, &pipe->pipe);
 148                 pipe->pipelen += msg->len;
 149                 res = 0;
 150         }
 151         dentry = dget(pipe->dentry);
 152         spin_unlock(&pipe->lock);
 153         if (dentry) {
 154                 wake_up(&RPC_I(dentry->d_inode)->waitq);
 155                 dput(dentry);
 156         }
 157         return res;
 158 }
 159 EXPORT_SYMBOL_GPL(rpc_queue_upcall);
 160
 161 static inline void
 162 rpc_inode_setowner(struct inode *inode, void *private)
 163 {
 164         RPC_I(inode)->private = private;
 165 }
 166
 167 static void
 168 rpc_close_pipes(struct inode *inode)
 169 {
 170         struct rpc_pipe *pipe = RPC_I(inode)->pipe;
 171         int need_release;
 172         LIST_HEAD(free_list);
 173
 174         mutex_lock(&inode->i_mutex);
 175         spin_lock(&pipe->lock);
 176         need_release = pipe->nreaders != 0 || pipe->nwriters != 0;
 177         pipe->nreaders = 0;
 178         list_splice_init(&pipe->in_upcall, &free_list);
 179         list_splice_init(&pipe->pipe, &free_list);
 180         pipe->pipelen = 0;
 181         pipe->dentry = NULL;
 182         spin_unlock(&pipe->lock);
 183         rpc_purge_list(&RPC_I(inode)->waitq, &free_list, pipe->ops->destroy_msg, -EPIPE);
 184         pipe->nwriters = 0;
 185         if (need_release && pipe->ops->release_pipe)
 186                 pipe->ops->release_pipe(inode);
 187         cancel_delayed_work_sync(&pipe->queue_timeout);
 188         rpc_inode_setowner(inode, NULL);
 189         RPC_I(inode)->pipe = NULL;
 190         mutex_unlock(&inode->i_mutex);
 191 }
 192
 193 static struct inode *
 194 rpc_alloc_inode(struct super_block *sb)
 195 {
 196         struct rpc_inode *rpci;
 197         rpci = (struct rpc_inode *)kmem_cache_alloc(rpc_inode_cachep, GFP_KERNEL);
 198         if (!rpci)
 199                 return NULL;
 200         return &rpci->vfs_inode;
 201 }
 202
 203 static void
 204 rpc_i_callback(struct rcu_head *head)
 205 {
 206         struct inode *inode = container_of(head, struct inode, i_rcu);
 207         kmem_cache_free(rpc_inode_cachep, RPC_I(inode));
 208 }
 209
 210 static void
 211 rpc_destroy_inode(struct inode *inode)
 212 {
 213         call_rcu(&inode->i_rcu, rpc_i_callback);
 214 }
 215
 216 static int
 217 rpc_pipe_open(struct inode *inode, struct file *filp)
 218 {
 219         struct net *net = inode->i_sb->s_fs_info;
 220         struct sunrpc_net *sn = net_generic(net, sunrpc_net_id);
 221         struct rpc_pipe *pipe;
 222         int first_open;
 223         int res = -ENXIO;
 224
 225         mutex_lock(&inode->i_mutex);
 226         sn->gssd_running = 1;
 227         pipe = RPC_I(inode)->pipe;
 228         if (pipe == NULL)
 229                 goto out;
 230         first_open = pipe->nreaders == 0 && pipe->nwriters == 0;
 231         if (first_open && pipe->ops->open_pipe) {
 232                 res = pipe->ops->open_pipe(inode);
 233                 if (res)
 234                         goto out;
 235         }
 236         if (filp->f_mode & FMODE_READ)
 237                 pipe->nreaders++;
 238         if (filp->f_mode & FMODE_WRITE)
 239                 pipe->nwriters++;
 240         res = 0;
 241 out:
 242         mutex_unlock(&inode->i_mutex);
 243         return res;
 244 }
 245
 246 static int
 247 rpc_pipe_release(struct inode *inode, struct file *filp)
 248 {
 249         struct rpc_pipe *pipe;
 250         struct rpc_pipe_msg *msg;
 251         int last_close;
 252
 253         mutex_lock(&inode->i_mutex);
 254         pipe = RPC_I(inode)->pipe;
 255         if (pipe == NULL)
 256                 goto out;
 257         msg = filp->private_data;
 258         if (msg != NULL) {
 259                 spin_lock(&pipe->lock);
 260                 msg->errno = -EAGAIN;
 261                 list_del_init(&msg->list);
 262                 spin_unlock(&pipe->lock);
 263                 pipe->ops->destroy_msg(msg);
 264         }
 265         if (filp->f_mode & FMODE_WRITE)
 266                 pipe->nwriters --;
 267         if (filp->f_mode & FMODE_READ) {
 268                 pipe->nreaders --;
 269                 if (pipe->nreaders == 0) {
 270                         LIST_HEAD(free_list);
 271                         spin_lock(&pipe->lock);
 272                         list_splice_init(&pipe->pipe, &free_list);
 273                         pipe->pipelen = 0;
 274                         spin_unlock(&pipe->lock);
 275                         rpc_purge_list(&RPC_I(inode)->waitq, &free_list,
 276                                         pipe->ops->destroy_msg, -EAGAIN);
 277                 }
 278         }
 279         last_close = pipe->nwriters == 0 && pipe->nreaders == 0;
 280         if (last_close && pipe->ops->release_pipe)
 281                 pipe->ops->release_pipe(inode);
 282 out:
 283         mutex_unlock(&inode->i_mutex);
 284         return 0;
 285 }
 286
 287 static ssize_t
 288 rpc_pipe_read(struct file *filp, char __user *buf, size_t len, loff_t *offset)
 289 {
 290         struct inode *inode = file_inode(filp);
 291         struct rpc_pipe *pipe;
 292         struct rpc_pipe_msg *msg;
 293         int res = 0;
 294
 295         mutex_lock(&inode->i_mutex);
 296         pipe = RPC_I(inode)->pipe;
 297         if (pipe == NULL) {
 298                 res = -EPIPE;
 299                 goto out_unlock;
 300         }
 301         msg = filp->private_data;
 302         if (msg == NULL) {
 303                 spin_lock(&pipe->lock);
 304                 if (!list_empty(&pipe->pipe)) {
 305                         msg = list_entry(pipe->pipe.next,
 306                                         struct rpc_pipe_msg,
 307                                         list);
 308                         list_move(&msg->list, &pipe->in_upcall);
 309                         pipe->pipelen -= msg->len;
 310                         filp->private_data = msg;
 311                         msg->copied = 0;
 312                 }
 313                 spin_unlock(&pipe->lock);
 314                 if (msg == NULL)
 315                         goto out_unlock;
 316         }
 317         /* NOTE: it is up to the callback to update msg->copied */
 318         res = pipe->ops->upcall(filp, msg, buf, len);
 319         if (res < 0 || msg->len == msg->copied) {
 320                 filp->private_data = NULL;
 321                 spin_lock(&pipe->lock);
 322                 list_del_init(&msg->list);
 323                 spin_unlock(&pipe->lock);
 324                 pipe->ops->destroy_msg(msg);
 325         }
 326 out_unlock:
 327         mutex_unlock(&inode->i_mutex);
 328         return res;
 329 }
 330
 331 static ssize_t
 332 rpc_pipe_write(struct file *filp, const char __user *buf, size_t len, loff_t *offset)
 333 {
 334         struct inode *inode = file_inode(filp);
 335         int res;
 336
 337         mutex_lock(&inode->i_mutex);
 338         res = -EPIPE;
 339         if (RPC_I(inode)->pipe != NULL)
 340                 res = RPC_I(inode)->pipe->ops->downcall(filp, buf, len);
 341         mutex_unlock(&inode->i_mutex);
 342         return res;
 343 }
 344
 345 static unsigned int
 346 rpc_pipe_poll(struct file *filp, struct poll_table_struct *wait)
 347 {
 348         struct inode *inode = file_inode(filp);
 349         struct rpc_inode *rpci = RPC_I(inode);
 350         unsigned int mask = POLLOUT | POLLWRNORM;
 351
 352         poll_wait(filp, &rpci->waitq, wait);
 353
 354         mutex_lock(&inode->i_mutex);
 355         if (rpci->pipe == NULL)
 356                 mask |= POLLERR | POLLHUP;
 357         else if (filp->private_data || !list_empty(&rpci->pipe->pipe))
 358                 mask |= POLLIN | POLLRDNORM;
 359         mutex_unlock(&inode->i_mutex);
 360         return mask;
 361 }
 362
 363 static long
 364 rpc_pipe_ioctl(struct file *filp, unsigned int cmd, unsigned long arg)
 365 {
 366         struct inode *inode = file_inode(filp);
 367         struct rpc_pipe *pipe;
 368         int len;
 369
 370         switch (cmd) {
 371         case FIONREAD:
 372                 mutex_lock(&inode->i_mutex);
 373                 pipe = RPC_I(inode)->pipe;
 374                 if (pipe == NULL) {
 375                         mutex_unlock(&inode->i_mutex);
 376                         return -EPIPE;
 377                 }
 378                 spin_lock(&pipe->lock);
 379                 len = pipe->pipelen;
 380                 if (filp->private_data) {
 381                         struct rpc_pipe_msg *msg;
 382                         msg = filp->private_data;
 383                         len += msg->len - msg->copied;
 384                 }
 385                 spin_unlock(&pipe->lock);
 386                 mutex_unlock(&inode->i_mutex);
 387                 return put_user(len, (int __user *)arg);
 388         default:
 389                 return -EINVAL;
 390         }
 391 }
 392
 393 static const struct file_operations rpc_pipe_fops = {
 394         .owner          = THIS_MODULE,
 395         .llseek         = no_llseek,
 396         .read           = rpc_pipe_read,
 397         .write          = rpc_pipe_write,
 398         .poll           = rpc_pipe_poll,
 399         .unlocked_ioctl = rpc_pipe_ioctl,
 400         .open           = rpc_pipe_open,
 401         .release        = rpc_pipe_release,
 402 };
 403
 404 static int
 405 rpc_show_info(struct seq_file *m, void *v)
 406 {
 407         struct rpc_clnt *clnt = m->private;
 408
 409         rcu_read_lock();
 410         seq_printf(m, "RPC server: %s\n",
 411                         rcu_dereference(clnt->cl_xprt)->servername);
 412         seq_printf(m, "service: %s (%d) version %d\n", clnt->cl_protname,
 413                         clnt->cl_prog, clnt->cl_vers);
 414         seq_printf(m, "address: %s\n", rpc_peeraddr2str(clnt, RPC_DISPLAY_ADDR));
 415         seq_printf(m, "protocol: %s\n", rpc_peeraddr2str(clnt, RPC_DISPLAY_PROTO));
 416         seq_printf(m, "port: %s\n", rpc_peeraddr2str(clnt, RPC_DISPLAY_PORT));
 417         rcu_read_unlock();
 418         return 0;
 419 }
 420
 421 static int
 422 rpc_info_open(struct inode *inode, struct file *file)
 423 {
 424         struct rpc_clnt *clnt = NULL;
 425         int ret = single_open(file, rpc_show_info, NULL);
 426
 427         if (!ret) {
 428                 struct seq_file *m = file->private_data;
 429
 430                 spin_lock(&file->f_path.dentry->d_lock);
 431                 if (!d_unhashed(file->f_path.dentry))
 432                         clnt = RPC_I(inode)->private;
 433                 if (clnt != NULL && atomic_inc_not_zero(&clnt->cl_count)) {
 434                         spin_unlock(&file->f_path.dentry->d_lock);
 435                         m->private = clnt;
 436                 } else {
 437                         spin_unlock(&file->f_path.dentry->d_lock);
 438                         single_release(inode, file);
 439                         ret = -EINVAL;
 440                 }
 441         }
 442         return ret;
 443 }
 444
 445 static int
 446 rpc_info_release(struct inode *inode, struct file *file)
 447 {
 448         struct seq_file *m = file->private_data;
 449         struct rpc_clnt *clnt = (struct rpc_clnt *)m->private;
 450
 451         if (clnt)
 452                 rpc_release_client(clnt);
 453         return single_release(inode, file);
 454 }
 455
 456 static const struct file_operations rpc_info_operations = {
 457         .owner          = THIS_MODULE,
 458         .open           = rpc_info_open,
 459         .read           = seq_read,
 460         .llseek         = seq_lseek,
 461         .release        = rpc_info_release,
 462 };
 463
 464
 465 /*
 466  * Description of fs contents.
 467  */
 468 struct rpc_filelist {
 469         const char *name;
 470         const struct file_operations *i_fop;
 471         umode_t mode;
 472 };
 473
 474 static int rpc_delete_dentry(const struct dentry *dentry)
 475 {
 476         return 1;
 477 }
 478
 479 static const struct dentry_operations rpc_dentry_operations = {
 480         .d_delete = rpc_delete_dentry,
 481 };
 482
 483 /*
 484  * Lookup the data. This is trivial - if the dentry didn't already
 485  * exist, we know it is negative.
 486  */
 487 static struct dentry *
 488 rpc_lookup(struct inode *dir, struct dentry *dentry, unsigned int flags)
 489 {
 490         if (dentry->d_name.len > NAME_MAX)
 491                 return ERR_PTR(-ENAMETOOLONG);
 492         d_add(dentry, NULL);
 493         return NULL;
 494 }
 495
 496 static const struct inode_operations rpc_dir_inode_operations = {
 497         .lookup         = rpc_lookup,
 498 };
 499
 500 static struct inode *
 501 rpc_get_inode(struct super_block *sb, umode_t mode)
 502 {
 503         struct inode *inode = new_inode(sb);
 504         if (!inode)
 505                 return NULL;
 506         inode->i_ino = get_next_ino();
 507         inode->i_mode = mode;
 508         inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME;
 509         switch (mode & S_IFMT) {
 510         case S_IFDIR:
 511                 inode->i_fop = &simple_dir_operations;
 512                 inode->i_op = &rpc_dir_inode_operations;
 513                 inc_nlink(inode);
 514         default:
 515                 break;
 516         }
 517         return inode;
 518 }
 519
 520 static int __rpc_create_common(struct inode *dir, struct dentry *dentry,
 521                                umode_t mode,
 522                                const struct file_operations *i_fop,
 523                                void *private)
 524 {
 525         struct inode *inode;
 526
 527         d_drop(dentry);
 528         inode = rpc_get_inode(dir->i_sb, mode);
 529         if (!inode)
 530                 goto out_err;
 531         inode->i_ino = iunique(dir->i_sb, 100);
 532         if (i_fop)
 533                 inode->i_fop = i_fop;
 534         if (private)
 535                 rpc_inode_setowner(inode, private);
 536         d_add(dentry, inode);
 537         return 0;
 538 out_err:
 539         printk(KERN_WARNING "%s: %s failed to allocate inode for dentry %s\n",
 540                         __FILE__, __func__, dentry->d_name.name);
 541         dput(dentry);
 542         return -ENOMEM;
 543 }
 544
 545 static int __rpc_create(struct inode *dir, struct dentry *dentry,
 546                         umode_t mode,
 547                         const struct file_operations *i_fop,
 548                         void *private)
 549 {
 550         int err;
 551
 552         err = __rpc_create_common(dir, dentry, S_IFREG | mode, i_fop, private);
 553         if (err)
 554                 return err;
 555         fsnotify_create(dir, dentry);
 556         return 0;
 557 }
 558
 559 static int __rpc_mkdir(struct inode *dir, struct dentry *dentry,
 560                        umode_t mode,
 561                        const struct file_operations *i_fop,
 562                        void *private)
 563 {
 564         int err;
 565
 566         err = __rpc_create_common(dir, dentry, S_IFDIR | mode, i_fop, private);
 567         if (err)
 568                 return err;
 569         inc_nlink(dir);
 570         fsnotify_mkdir(dir, dentry);
 571         return 0;
 572 }
 573
 574 static void
 575 init_pipe(struct rpc_pipe *pipe)
 576 {
 577         pipe->nreaders = 0;
 578         pipe->nwriters = 0;
 579         INIT_LIST_HEAD(&pipe->in_upcall);
 580         INIT_LIST_HEAD(&pipe->in_downcall);
 581         INIT_LIST_HEAD(&pipe->pipe);
 582         pipe->pipelen = 0;
 583         INIT_DELAYED_WORK(&pipe->queue_timeout,
 584                             rpc_timeout_upcall_queue);
 585         pipe->ops = NULL;
 586         spin_lock_init(&pipe->lock);
 587         pipe->dentry = NULL;
 588 }
 589
 590 void rpc_destroy_pipe_data(struct rpc_pipe *pipe)
 591 {
 592         kfree(pipe);
 593 }
 594 EXPORT_SYMBOL_GPL(rpc_destroy_pipe_data);
 595
 596 struct rpc_pipe *rpc_mkpipe_data(const struct rpc_pipe_ops *ops, int flags)
 597 {
 598         struct rpc_pipe *pipe;
 599
 600         pipe = kzalloc(sizeof(struct rpc_pipe), GFP_KERNEL);
 601         if (!pipe)
 602                 return ERR_PTR(-ENOMEM);
 603         init_pipe(pipe);
 604         pipe->ops = ops;
 605         pipe->flags = flags;
 606         return pipe;
 607 }
 608 EXPORT_SYMBOL_GPL(rpc_mkpipe_data);
 609
 610 static int __rpc_mkpipe_dentry(struct inode *dir, struct dentry *dentry,
 611                                umode_t mode,
 612                                const struct file_operations *i_fop,
 613                                void *private,
 614                                struct rpc_pipe *pipe)
 615 {
 616         struct rpc_inode *rpci;
 617         int err;
 618
 619         err = __rpc_create_common(dir, dentry, S_IFIFO | mode, i_fop, private);
 620         if (err)
 621                 return err;
 622         rpci = RPC_I(dentry->d_inode);
 623         rpci->private = private;
 624         rpci->pipe = pipe;
 625         fsnotify_create(dir, dentry);
 626         return 0;
 627 }
 628
 629 static int __rpc_rmdir(struct inode *dir, struct dentry *dentry)
 630 {
 631         int ret;
 632
 633         dget(dentry);
 634         ret = simple_rmdir(dir, dentry);
 635         d_delete(dentry);
 636         dput(dentry);
 637         return ret;
 638 }
 639
 640 int rpc_rmdir(struct dentry *dentry)
 641 {
 642         struct dentry *parent;
 643         struct inode *dir;
 644         int error;
 645
 646         parent = dget_parent(dentry);
 647         dir = parent->d_inode;
 648         mutex_lock_nested(&dir->i_mutex, I_MUTEX_PARENT);
 649         error = __rpc_rmdir(dir, dentry);
 650         mutex_unlock(&dir->i_mutex);
 651         dput(parent);
 652         return error;
 653 }
 654 EXPORT_SYMBOL_GPL(rpc_rmdir);
 655
 656 static int __rpc_unlink(struct inode *dir, struct dentry *dentry)
 657 {
 658         int ret;
 659
 660         dget(dentry);
 661         ret = simple_unlink(dir, dentry);
 662         d_delete(dentry);
 663         dput(dentry);
 664         return ret;
 665 }
 666
 667 static int __rpc_rmpipe(struct inode *dir, struct dentry *dentry)
 668 {
 669         struct inode *inode = dentry->d_inode;
 670
 671         rpc_close_pipes(inode);
 672         return __rpc_unlink(dir, dentry);
 673 }
 674
 675 static struct dentry *__rpc_lookup_create_exclusive(struct dentry *parent,
 676                                           struct qstr *name)
 677 {
 678         struct dentry *dentry;
 679
 680         dentry = d_lookup(parent, name);
 681         if (!dentry) {
 682                 dentry = d_alloc(parent, name);
 683                 if (!dentry)
 684                         return ERR_PTR(-ENOMEM);
 685         }
 686         if (dentry->d_inode == NULL)
 687                 return dentry;
 688         dput(dentry);
 689         return ERR_PTR(-EEXIST);
 690 }
 691
 692 /*
 693  * FIXME: This probably has races.
 694  */
 695 static void __rpc_depopulate(struct dentry *parent,
 696                              const struct rpc_filelist *files,
 697                              int start, int eof)
 698 {
 699         struct inode *dir = parent->d_inode;
 700         struct dentry *dentry;
 701         struct qstr name;
 702         int i;
 703
 704         for (i = start; i < eof; i++) {
 705                 name.name = files[i].name;
 706                 name.len = strlen(files[i].name);
 707                 name.hash = full_name_hash(name.name, name.len);
 708                 dentry = d_lookup(parent, &name);
 709
 710                 if (dentry == NULL)
 711                         continue;
 712                 if (dentry->d_inode == NULL)
 713                         goto next;
 714                 switch (dentry->d_inode->i_mode & S_IFMT) {
 715                         default:
 716                                 BUG();
 717                         case S_IFREG:
 718                                 __rpc_unlink(dir, dentry);
 719                                 break;
 720                         case S_IFDIR:
 721                                 __rpc_rmdir(dir, dentry);
 722                 }
 723 next:
 724                 dput(dentry);
 725         }
 726 }
 727
 728 static void rpc_depopulate(struct dentry *parent,
 729                            const struct rpc_filelist *files,
 730                            int start, int eof)
 731 {
 732         struct inode *dir = parent->d_inode;
 733
 734         mutex_lock_nested(&dir->i_mutex, I_MUTEX_CHILD);
 735         __rpc_depopulate(parent, files, start, eof);
 736         mutex_unlock(&dir->i_mutex);
 737 }
 738
 739 static int rpc_populate(struct dentry *parent,
 740                         const struct rpc_filelist *files,
 741                         int start, int eof,
 742                         void *private)
 743 {
 744         struct inode *dir = parent->d_inode;
 745         struct dentry *dentry;
 746         int i, err;
 747
 748         mutex_lock(&dir->i_mutex);
 749         for (i = start; i < eof; i++) {
 750                 struct qstr q;
 751
 752                 q.name = files[i].name;
 753                 q.len = strlen(files[i].name);
 754                 q.hash = full_name_hash(q.name, q.len);
 755                 dentry = __rpc_lookup_create_exclusive(parent, &q);
 756                 err = PTR_ERR(dentry);
 757                 if (IS_ERR(dentry))
 758                         goto out_bad;
 759                 switch (files[i].mode & S_IFMT) {
 760                         default:
 761                                 BUG();
 762                         case S_IFREG:
 763                                 err = __rpc_create(dir, dentry,
 764                                                 files[i].mode,
 765                                                 files[i].i_fop,
 766                                                 private);
 767                                 break;
 768                         case S_IFDIR:
 769                                 err = __rpc_mkdir(dir, dentry,
 770                                                 files[i].mode,
 771                                                 NULL,
 772                                                 private);
 773                 }
 774                 if (err != 0)
 775                         goto out_bad;
 776         }
 777         mutex_unlock(&dir->i_mutex);
 778         return 0;
 779 out_bad:
 780         __rpc_depopulate(parent, files, start, eof);
 781         mutex_unlock(&dir->i_mutex);
 782         printk(KERN_WARNING "%s: %s failed to populate directory %s\n",
 783                         __FILE__, __func__, parent->d_name.name);
 784         return err;
 785 }
 786
 787 static struct dentry *rpc_mkdir_populate(struct dentry *parent,
 788                 struct qstr *name, umode_t mode, void *private,
 789                 int (*populate)(struct dentry *, void *), void *args_populate)
 790 {
 791         struct dentry *dentry;
 792         struct inode *dir = parent->d_inode;
 793         int error;
 794
 795         mutex_lock_nested(&dir->i_mutex, I_MUTEX_PARENT);
 796         dentry = __rpc_lookup_create_exclusive(parent, name);
 797         if (IS_ERR(dentry))
 798                 goto out;
 799         error = __rpc_mkdir(dir, dentry, mode, NULL, private);
 800         if (error != 0)
 801                 goto out_err;
 802         if (populate != NULL) {
 803                 error = populate(dentry, args_populate);
 804                 if (error)
 805                         goto err_rmdir;
 806         }
 807 out:
 808         mutex_unlock(&dir->i_mutex);
 809         return dentry;
 810 err_rmdir:
 811         __rpc_rmdir(dir, dentry);
 812 out_err:
 813         dentry = ERR_PTR(error);
 814         goto out;
 815 }
 816
 817 static int rpc_rmdir_depopulate(struct dentry *dentry,
 818                 void (*depopulate)(struct dentry *))
 819 {
 820         struct dentry *parent;
 821         struct inode *dir;
 822         int error;
 823
 824         parent = dget_parent(dentry);
 825         dir = parent->d_inode;
 826         mutex_lock_nested(&dir->i_mutex, I_MUTEX_PARENT);
 827         if (depopulate != NULL)
 828                 depopulate(dentry);
 829         error = __rpc_rmdir(dir, dentry);
 830         mutex_unlock(&dir->i_mutex);
 831         dput(parent);
 832         return error;
 833 }
 834
 835 /**
 836  * rpc_mkpipe - make an rpc_pipefs file for kernel<->userspace communication
 837  * @parent: dentry of directory to create new "pipe" in
 838  * @name: name of pipe
 839  * @private: private data to associate with the pipe, for the caller's use
 840  * @pipe: &rpc_pipe containing input parameters
 841  *
 842  * Data is made available for userspace to read by calls to
 843  * rpc_queue_upcall().  The actual reads will result in calls to
 844  * @ops->upcall, which will be called with the file pointer,
 845  * message, and userspace buffer to copy to.
 846  *
 847  * Writes can come at any time, and do not necessarily have to be
 848  * responses to upcalls.  They will result in calls to @msg->downcall.
 849  *
 850  * The @private argument passed here will be available to all these methods
 851  * from the file pointer, via RPC_I(file_inode(file))->private.
 852  */
 853 struct dentry *rpc_mkpipe_dentry(struct dentry *parent, const char *name,
 854                                  void *private, struct rpc_pipe *pipe)
 855 {
 856         struct dentry *dentry;
 857         struct inode *dir = parent->d_inode;
 858         umode_t umode = S_IFIFO | S_IRUSR | S_IWUSR;
 859         struct qstr q;
 860         int err;
 861
 862         if (pipe->ops->upcall == NULL)
 863                 umode &= ~S_IRUGO;
 864         if (pipe->ops->downcall == NULL)
 865                 umode &= ~S_IWUGO;
 866
 867         q.name = name;
 868         q.len = strlen(name);
 869         q.hash = full_name_hash(q.name, q.len),
 870
 871         mutex_lock_nested(&dir->i_mutex, I_MUTEX_PARENT);
 872         dentry = __rpc_lookup_create_exclusive(parent, &q);
 873         if (IS_ERR(dentry))
 874                 goto out;
 875         err = __rpc_mkpipe_dentry(dir, dentry, umode, &rpc_pipe_fops,
 876                                   private, pipe);
 877         if (err)
 878                 goto out_err;
 879 out:
 880         mutex_unlock(&dir->i_mutex);
 881         return dentry;
 882 out_err:
 883         dentry = ERR_PTR(err);
 884         printk(KERN_WARNING "%s: %s() failed to create pipe %s/%s (errno = %d)\n",
 885                         __FILE__, __func__, parent->d_name.name, name,
 886                         err);
 887         goto out;
 888 }
 889 EXPORT_SYMBOL_GPL(rpc_mkpipe_dentry);
 890
 891 /**
 892  * rpc_unlink - remove a pipe
 893  * @dentry: dentry for the pipe, as returned from rpc_mkpipe
 894  *
 895  * After this call, lookups will no longer find the pipe, and any
 896  * attempts to read or write using preexisting opens of the pipe will
 897  * return -EPIPE.
 898  */
 899 int
 900 rpc_unlink(struct dentry *dentry)
 901 {
 902         struct dentry *parent;
 903         struct inode *dir;
 904         int error = 0;
 905
 906         parent = dget_parent(dentry);
 907         dir = parent->d_inode;
 908         mutex_lock_nested(&dir->i_mutex, I_MUTEX_PARENT);
 909         error = __rpc_rmpipe(dir, dentry);
 910         mutex_unlock(&dir->i_mutex);
 911         dput(parent);
 912         return error;
 913 }
 914 EXPORT_SYMBOL_GPL(rpc_unlink);
 915
 916 enum {
 917         RPCAUTH_info,
 918         RPCAUTH_EOF
 919 };
 920
 921 static const struct rpc_filelist authfiles[] = {
 922         [RPCAUTH_info] = {
 923                 .name = "info",
 924                 .i_fop = &rpc_info_operations,
 925                 .mode = S_IFREG | S_IRUSR,
 926         },
 927 };
 928
 929 static int rpc_clntdir_populate(struct dentry *dentry, void *private)
 930 {
 931         return rpc_populate(dentry,
 932                             authfiles, RPCAUTH_info, RPCAUTH_EOF,
 933                             private);
 934 }
 935
 936 static void rpc_clntdir_depopulate(struct dentry *dentry)
 937 {
 938         rpc_depopulate(dentry, authfiles, RPCAUTH_info, RPCAUTH_EOF);
 939 }
 940
 941 /**
 942  * rpc_create_client_dir - Create a new rpc_client directory in rpc_pipefs
 943  * @dentry: dentry from the rpc_pipefs root to the new directory
 944  * @name: &struct qstr for the name
 945  * @rpc_client: rpc client to associate with this directory
 946  *
 947  * This creates a directory at the given @path associated with
 948  * @rpc_clnt, which will contain a file named "info" with some basic
 949  * information about the client, together with any "pipes" that may
 950  * later be created using rpc_mkpipe().
 951  */
 952 struct dentry *rpc_create_client_dir(struct dentry *dentry,
 953                                    struct qstr *name,
 954                                    struct rpc_clnt *rpc_client)
 955 {
 956         return rpc_mkdir_populate(dentry, name, S_IRUGO | S_IXUGO, NULL,
 957                         rpc_clntdir_populate, rpc_client);
 958 }
 959
 960 /**
 961  * rpc_remove_client_dir - Remove a directory created with rpc_create_client_dir()
 962  * @dentry: dentry for the pipe
 963  */
 964 int rpc_remove_client_dir(struct dentry *dentry)
 965 {
 966         return rpc_rmdir_depopulate(dentry, rpc_clntdir_depopulate);
 967 }
 968
 969 static const struct rpc_filelist cache_pipefs_files[3] = {
 970         [0] = {
 971                 .name = "channel",
 972                 .i_fop = &cache_file_operations_pipefs,
 973                 .mode = S_IFREG|S_IRUSR|S_IWUSR,
 974         },
 975         [1] = {
 976                 .name = "content",
 977                 .i_fop = &content_file_operations_pipefs,
 978                 .mode = S_IFREG|S_IRUSR,
 979         },
 980         [2] = {
 981                 .name = "flush",
 982                 .i_fop = &cache_flush_operations_pipefs,
 983                 .mode = S_IFREG|S_IRUSR|S_IWUSR,
 984         },
 985 };
 986
 987 static int rpc_cachedir_populate(struct dentry *dentry, void *private)
 988 {
 989         return rpc_populate(dentry,
 990                             cache_pipefs_files, 0, 3,
 991                             private);
 992 }
 993
 994 static void rpc_cachedir_depopulate(struct dentry *dentry)
 995 {
 996         rpc_depopulate(dentry, cache_pipefs_files, 0, 3);
 997 }
 998
 999 struct dentry *rpc_create_cache_dir(struct dentry *parent, struct qstr *name,
1000                                     umode_t umode, struct cache_detail *cd)
1001 {
1002         return rpc_mkdir_populate(parent, name, umode, NULL,
1003                         rpc_cachedir_populate, cd);
1004 }
1005
1006 void rpc_remove_cache_dir(struct dentry *dentry)
1007 {
1008         rpc_rmdir_depopulate(dentry, rpc_cachedir_depopulate);
1009 }
1010
1011 /*
1012  * populate the filesystem
1013  */
1014 static const struct super_operations s_ops = {
1015         .alloc_inode    = rpc_alloc_inode,
1016         .destroy_inode  = rpc_destroy_inode,
1017         .statfs         = simple_statfs,
1018 };
1019
1020 #define RPCAUTH_GSSMAGIC 0x67596969
1021
1022 /*
1023  * We have a single directory with 1 node in it.
1024  */
1025 enum {
1026         RPCAUTH_lockd,
1027         RPCAUTH_mount,
1028         RPCAUTH_nfs,
1029         RPCAUTH_portmap,
1030         RPCAUTH_statd,
1031         RPCAUTH_nfsd4_cb,
1032         RPCAUTH_cache,
1033         RPCAUTH_nfsd,
1034         RPCAUTH_RootEOF
1035 };
1036
1037 static const struct rpc_filelist files[] = {
1038         [RPCAUTH_lockd] = {
1039                 .name = "lockd",
1040                 .mode = S_IFDIR | S_IRUGO | S_IXUGO,
1041         },
1042         [RPCAUTH_mount] = {
1043                 .name = "mount",
1044                 .mode = S_IFDIR | S_IRUGO | S_IXUGO,
1045         },
1046         [RPCAUTH_nfs] = {
1047                 .name = "nfs",
1048                 .mode = S_IFDIR | S_IRUGO | S_IXUGO,
1049         },
1050         [RPCAUTH_portmap] = {
1051                 .name = "portmap",
1052                 .mode = S_IFDIR | S_IRUGO | S_IXUGO,
1053         },
1054         [RPCAUTH_statd] = {
1055                 .name = "statd",
1056                 .mode = S_IFDIR | S_IRUGO | S_IXUGO,
1057         },
1058         [RPCAUTH_nfsd4_cb] = {
1059                 .name = "nfsd4_cb",
1060                 .mode = S_IFDIR | S_IRUGO | S_IXUGO,
1061         },
1062         [RPCAUTH_cache] = {
1063                 .name = "cache",
1064                 .mode = S_IFDIR | S_IRUGO | S_IXUGO,
1065         },
1066         [RPCAUTH_nfsd] = {
1067                 .name = "nfsd",
1068                 .mode = S_IFDIR | S_IRUGO | S_IXUGO,
1069         },
1070 };
1071
1072 /*
1073  * This call can be used only in RPC pipefs mount notification hooks.
1074  */
1075 struct dentry *rpc_d_lookup_sb(const struct super_block *sb,
1076                                const unsigned char *dir_name)
1077 {
1078         struct qstr dir = QSTR_INIT(dir_name, strlen(dir_name));
1079
1080         dir.hash = full_name_hash(dir.name, dir.len);
1081         return d_lookup(sb->s_root, &dir);
1082 }
1083 EXPORT_SYMBOL_GPL(rpc_d_lookup_sb);
1084
1085 void rpc_pipefs_init_net(struct net *net)
1086 {
1087         struct sunrpc_net *sn = net_generic(net, sunrpc_net_id);
1088
1089         mutex_init(&sn->pipefs_sb_lock);
1090         sn->gssd_running = 1;
1091         sn->pipe_version = -1;
1092 }
1093
1094 /*
1095  * This call will be used for per network namespace operations calls.
1096  * Note: Function will be returned with pipefs_sb_lock taken if superblock was
1097  * found. This lock have to be released by rpc_put_sb_net() when all operations
1098  * will be completed.
1099  */
1100 struct super_block *rpc_get_sb_net(const struct net *net)
1101 {
1102         struct sunrpc_net *sn = net_generic(net, sunrpc_net_id);
1103
1104         mutex_lock(&sn->pipefs_sb_lock);
1105         if (sn->pipefs_sb)
1106                 return sn->pipefs_sb;
1107         mutex_unlock(&sn->pipefs_sb_lock);
1108         return NULL;
1109 }
1110 EXPORT_SYMBOL_GPL(rpc_get_sb_net);
1111
1112 void rpc_put_sb_net(const struct net *net)
1113 {
1114         struct sunrpc_net *sn = net_generic(net, sunrpc_net_id);
1115
1116         WARN_ON(sn->pipefs_sb == NULL);
1117         mutex_unlock(&sn->pipefs_sb_lock);
1118 }
1119 EXPORT_SYMBOL_GPL(rpc_put_sb_net);
1120
1121 static int
1122 rpc_fill_super(struct super_block *sb, void *data, int silent)
1123 {
1124         struct inode *inode;
1125         struct dentry *root;
1126         struct net *net = data;
1127         struct sunrpc_net *sn = net_generic(net, sunrpc_net_id);
1128         int err;
1129
1130         sb->s_blocksize = PAGE_CACHE_SIZE;
1131         sb->s_blocksize_bits = PAGE_CACHE_SHIFT;
1132         sb->s_magic = RPCAUTH_GSSMAGIC;
1133         sb->s_op = &s_ops;
1134         sb->s_d_op = &rpc_dentry_operations;
1135         sb->s_time_gran = 1;
1136
1137         inode = rpc_get_inode(sb, S_IFDIR | S_IRUGO | S_IXUGO);
1138         sb->s_root = root = d_make_root(inode);
1139         if (!root)
1140                 return -ENOMEM;
1141         if (rpc_populate(root, files, RPCAUTH_lockd, RPCAUTH_RootEOF, NULL))
1142                 return -ENOMEM;
1143         dprintk("RPC:       sending pipefs MOUNT notification for net %p%s\n",
1144                 net, NET_NAME(net));
1145         mutex_lock(&sn->pipefs_sb_lock);
1146         sn->pipefs_sb = sb;
1147         err = blocking_notifier_call_chain(&rpc_pipefs_notifier_list,
1148                                            RPC_PIPEFS_MOUNT,
1149                                            sb);
1150         if (err)
1151                 goto err_depopulate;
1152         sb->s_fs_info = get_net(net);
1153         mutex_unlock(&sn->pipefs_sb_lock);
1154         return 0;
1155
1156 err_depopulate:
1157         blocking_notifier_call_chain(&rpc_pipefs_notifier_list,
1158                                            RPC_PIPEFS_UMOUNT,
1159                                            sb);
1160         sn->pipefs_sb = NULL;
1161         __rpc_depopulate(root, files, RPCAUTH_lockd, RPCAUTH_RootEOF);
1162         mutex_unlock(&sn->pipefs_sb_lock);
1163         return err;
1164 }
1165
1166 static struct dentry *
1167 rpc_mount(struct file_system_type *fs_type,
1168                 int flags, const char *dev_name, void *data)
1169 {
1170         return mount_ns(fs_type, flags, current->nsproxy->net_ns, rpc_fill_super);
1171 }
1172
1173 static void rpc_kill_sb(struct super_block *sb)
1174 {
1175         struct net *net = sb->s_fs_info;
1176         struct sunrpc_net *sn = net_generic(net, sunrpc_net_id);
1177
1178         mutex_lock(&sn->pipefs_sb_lock);
1179         if (sn->pipefs_sb != sb) {
1180                 mutex_unlock(&sn->pipefs_sb_lock);
1181                 goto out;
1182         }
1183         sn->pipefs_sb = NULL;
1184         dprintk("RPC:       sending pipefs UMOUNT notification for net %p%s\n",
1185                 net, NET_NAME(net));
1186         blocking_notifier_call_chain(&rpc_pipefs_notifier_list,
1187                                            RPC_PIPEFS_UMOUNT,
1188                                            sb);
1189         mutex_unlock(&sn->pipefs_sb_lock);
1190         put_net(net);
1191 out:
1192         kill_litter_super(sb);
1193 }
1194
1195 static struct file_system_type rpc_pipe_fs_type = {
1196         .owner          = THIS_MODULE,
1197         .name           = "rpc_pipefs",
1198         .mount          = rpc_mount,
1199         .kill_sb        = rpc_kill_sb,
1200 };
1201 MODULE_ALIAS_FS("rpc_pipefs");
1202 MODULE_ALIAS("rpc_pipefs");
1203
1204 static void
1205 init_once(void *foo)
1206 {
1207         struct rpc_inode *rpci = (struct rpc_inode *) foo;
1208
1209         inode_init_once(&rpci->vfs_inode);
1210         rpci->private = NULL;
1211         rpci->pipe = NULL;
1212         init_waitqueue_head(&rpci->waitq);
1213 }
1214
1215 int register_rpc_pipefs(void)
1216 {
1217         int err;
1218
1219         rpc_inode_cachep = kmem_cache_create("rpc_inode_cache",
1220                                 sizeof(struct rpc_inode),
1221                                 0, (SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT|
1222                                                 SLAB_MEM_SPREAD),
1223                                 init_once);
1224         if (!rpc_inode_cachep)
1225                 return -ENOMEM;
1226         err = rpc_clients_notifier_register();
1227         if (err)
1228                 goto err_notifier;
1229         err = register_filesystem(&rpc_pipe_fs_type);
1230         if (err)
1231                 goto err_register;
1232         return 0;
1233
1234 err_register:
1235         rpc_clients_notifier_unregister();
1236 err_notifier:
1237         kmem_cache_destroy(rpc_inode_cachep);
1238         return err;
1239 }
1240
1241 void unregister_rpc_pipefs(void)
1242 {
1243         rpc_clients_notifier_unregister();
1244         kmem_cache_destroy(rpc_inode_cachep);
1245         unregister_filesystem(&rpc_pipe_fs_type);
1246 }