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         wake_up(waitq);
  75 }
  76
  77 static void
  78 rpc_timeout_upcall_queue(struct work_struct *work)
  79 {
  80         LIST_HEAD(free_list);
  81         struct rpc_pipe *pipe =
  82                 container_of(work, struct rpc_pipe, queue_timeout.work);
  83         void (*destroy_msg)(struct rpc_pipe_msg *);
  84         struct dentry *dentry;
  85
  86         spin_lock(&pipe->lock);
  87         destroy_msg = pipe->ops->destroy_msg;
  88         if (pipe->nreaders == 0) {
  89                 list_splice_init(&pipe->pipe, &free_list);
  90                 pipe->pipelen = 0;
  91         }
  92         dentry = dget(pipe->dentry);
  93         spin_unlock(&pipe->lock);
  94         if (dentry) {
  95                 rpc_purge_list(&RPC_I(dentry->d_inode)->waitq,
  96                                &free_list, destroy_msg, -ETIMEDOUT);
  97                 dput(dentry);
  98         }
  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  * @inode: inode of 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 rpc_pipe *pipe;
 220         int first_open;
 221         int res = -ENXIO;
 222
 223         mutex_lock(&inode->i_mutex);
 224         pipe = RPC_I(inode)->pipe;
 225         if (pipe == NULL)
 226                 goto out;
 227         first_open = pipe->nreaders == 0 && pipe->nwriters == 0;
 228         if (first_open && pipe->ops->open_pipe) {
 229                 res = pipe->ops->open_pipe(inode);
 230                 if (res)
 231                         goto out;
 232         }
 233         if (filp->f_mode & FMODE_READ)
 234                 pipe->nreaders++;
 235         if (filp->f_mode & FMODE_WRITE)
 236                 pipe->nwriters++;
 237         res = 0;
 238 out:
 239         mutex_unlock(&inode->i_mutex);
 240         return res;
 241 }
 242
 243 static int
 244 rpc_pipe_release(struct inode *inode, struct file *filp)
 245 {
 246         struct rpc_pipe *pipe;
 247         struct rpc_pipe_msg *msg;
 248         int last_close;
 249
 250         mutex_lock(&inode->i_mutex);
 251         pipe = RPC_I(inode)->pipe;
 252         if (pipe == NULL)
 253                 goto out;
 254         msg = filp->private_data;
 255         if (msg != NULL) {
 256                 spin_lock(&pipe->lock);
 257                 msg->errno = -EAGAIN;
 258                 list_del_init(&msg->list);
 259                 spin_unlock(&pipe->lock);
 260                 pipe->ops->destroy_msg(msg);
 261         }
 262         if (filp->f_mode & FMODE_WRITE)
 263                 pipe->nwriters --;
 264         if (filp->f_mode & FMODE_READ) {
 265                 pipe->nreaders --;
 266                 if (pipe->nreaders == 0) {
 267                         LIST_HEAD(free_list);
 268                         spin_lock(&pipe->lock);
 269                         list_splice_init(&pipe->pipe, &free_list);
 270                         pipe->pipelen = 0;
 271                         spin_unlock(&pipe->lock);
 272                         rpc_purge_list(&RPC_I(inode)->waitq, &free_list,
 273                                         pipe->ops->destroy_msg, -EAGAIN);
 274                 }
 275         }
 276         last_close = pipe->nwriters == 0 && pipe->nreaders == 0;
 277         if (last_close && pipe->ops->release_pipe)
 278                 pipe->ops->release_pipe(inode);
 279 out:
 280         mutex_unlock(&inode->i_mutex);
 281         return 0;
 282 }
 283
 284 static ssize_t
 285 rpc_pipe_read(struct file *filp, char __user *buf, size_t len, loff_t *offset)
 286 {
 287         struct inode *inode = filp->f_path.dentry->d_inode;
 288         struct rpc_pipe *pipe;
 289         struct rpc_pipe_msg *msg;
 290         int res = 0;
 291
 292         mutex_lock(&inode->i_mutex);
 293         pipe = RPC_I(inode)->pipe;
 294         if (pipe == NULL) {
 295                 res = -EPIPE;
 296                 goto out_unlock;
 297         }
 298         msg = filp->private_data;
 299         if (msg == NULL) {
 300                 spin_lock(&pipe->lock);
 301                 if (!list_empty(&pipe->pipe)) {
 302                         msg = list_entry(pipe->pipe.next,
 303                                         struct rpc_pipe_msg,
 304                                         list);
 305                         list_move(&msg->list, &pipe->in_upcall);
 306                         pipe->pipelen -= msg->len;
 307                         filp->private_data = msg;
 308                         msg->copied = 0;
 309                 }
 310                 spin_unlock(&pipe->lock);
 311                 if (msg == NULL)
 312                         goto out_unlock;
 313         }
 314         /* NOTE: it is up to the callback to update msg->copied */
 315         res = pipe->ops->upcall(filp, msg, buf, len);
 316         if (res < 0 || msg->len == msg->copied) {
 317                 filp->private_data = NULL;
 318                 spin_lock(&pipe->lock);
 319                 list_del_init(&msg->list);
 320                 spin_unlock(&pipe->lock);
 321                 pipe->ops->destroy_msg(msg);
 322         }
 323 out_unlock:
 324         mutex_unlock(&inode->i_mutex);
 325         return res;
 326 }
 327
 328 static ssize_t
 329 rpc_pipe_write(struct file *filp, const char __user *buf, size_t len, loff_t *offset)
 330 {
 331         struct inode *inode = filp->f_path.dentry->d_inode;
 332         int res;
 333
 334         mutex_lock(&inode->i_mutex);
 335         res = -EPIPE;
 336         if (RPC_I(inode)->pipe != NULL)
 337                 res = RPC_I(inode)->pipe->ops->downcall(filp, buf, len);
 338         mutex_unlock(&inode->i_mutex);
 339         return res;
 340 }
 341
 342 static unsigned int
 343 rpc_pipe_poll(struct file *filp, struct poll_table_struct *wait)
 344 {
 345         struct inode *inode = filp->f_path.dentry->d_inode;
 346         struct rpc_inode *rpci = RPC_I(inode);
 347         unsigned int mask = POLLOUT | POLLWRNORM;
 348
 349         poll_wait(filp, &rpci->waitq, wait);
 350
 351         mutex_lock(&inode->i_mutex);
 352         if (rpci->pipe == NULL)
 353                 mask |= POLLERR | POLLHUP;
 354         else if (filp->private_data || !list_empty(&rpci->pipe->pipe))
 355                 mask |= POLLIN | POLLRDNORM;
 356         mutex_unlock(&inode->i_mutex);
 357         return mask;
 358 }
 359
 360 static long
 361 rpc_pipe_ioctl(struct file *filp, unsigned int cmd, unsigned long arg)
 362 {
 363         struct inode *inode = filp->f_path.dentry->d_inode;
 364         struct rpc_pipe *pipe;
 365         int len;
 366
 367         switch (cmd) {
 368         case FIONREAD:
 369                 mutex_lock(&inode->i_mutex);
 370                 pipe = RPC_I(inode)->pipe;
 371                 if (pipe == NULL) {
 372                         mutex_unlock(&inode->i_mutex);
 373                         return -EPIPE;
 374                 }
 375                 spin_lock(&pipe->lock);
 376                 len = pipe->pipelen;
 377                 if (filp->private_data) {
 378                         struct rpc_pipe_msg *msg;
 379                         msg = filp->private_data;
 380                         len += msg->len - msg->copied;
 381                 }
 382                 spin_unlock(&pipe->lock);
 383                 mutex_unlock(&inode->i_mutex);
 384                 return put_user(len, (int __user *)arg);
 385         default:
 386                 return -EINVAL;
 387         }
 388 }
 389
 390 static const struct file_operations rpc_pipe_fops = {
 391         .owner          = THIS_MODULE,
 392         .llseek         = no_llseek,
 393         .read           = rpc_pipe_read,
 394         .write          = rpc_pipe_write,
 395         .poll           = rpc_pipe_poll,
 396         .unlocked_ioctl = rpc_pipe_ioctl,
 397         .open           = rpc_pipe_open,
 398         .release        = rpc_pipe_release,
 399 };
 400
 401 static int
 402 rpc_show_info(struct seq_file *m, void *v)
 403 {
 404         struct rpc_clnt *clnt = m->private;
 405
 406         rcu_read_lock();
 407         seq_printf(m, "RPC server: %s\n",
 408                         rcu_dereference(clnt->cl_xprt)->servername);
 409         seq_printf(m, "service: %s (%d) version %d\n", clnt->cl_protname,
 410                         clnt->cl_prog, clnt->cl_vers);
 411         seq_printf(m, "address: %s\n", rpc_peeraddr2str(clnt, RPC_DISPLAY_ADDR));
 412         seq_printf(m, "protocol: %s\n", rpc_peeraddr2str(clnt, RPC_DISPLAY_PROTO));
 413         seq_printf(m, "port: %s\n", rpc_peeraddr2str(clnt, RPC_DISPLAY_PORT));
 414         rcu_read_unlock();
 415         return 0;
 416 }
 417
 418 static int
 419 rpc_info_open(struct inode *inode, struct file *file)
 420 {
 421         struct rpc_clnt *clnt = NULL;
 422         int ret = single_open(file, rpc_show_info, NULL);
 423
 424         if (!ret) {
 425                 struct seq_file *m = file->private_data;
 426
 427                 spin_lock(&file->f_path.dentry->d_lock);
 428                 if (!d_unhashed(file->f_path.dentry))
 429                         clnt = RPC_I(inode)->private;
 430                 if (clnt != NULL && atomic_inc_not_zero(&clnt->cl_count)) {
 431                         spin_unlock(&file->f_path.dentry->d_lock);
 432                         m->private = clnt;
 433                 } else {
 434                         spin_unlock(&file->f_path.dentry->d_lock);
 435                         single_release(inode, file);
 436                         ret = -EINVAL;
 437                 }
 438         }
 439         return ret;
 440 }
 441
 442 static int
 443 rpc_info_release(struct inode *inode, struct file *file)
 444 {
 445         struct seq_file *m = file->private_data;
 446         struct rpc_clnt *clnt = (struct rpc_clnt *)m->private;
 447
 448         if (clnt)
 449                 rpc_release_client(clnt);
 450         return single_release(inode, file);
 451 }
 452
 453 static const struct file_operations rpc_info_operations = {
 454         .owner          = THIS_MODULE,
 455         .open           = rpc_info_open,
 456         .read           = seq_read,
 457         .llseek         = seq_lseek,
 458         .release        = rpc_info_release,
 459 };
 460
 461
 462 /*
 463  * Description of fs contents.
 464  */
 465 struct rpc_filelist {
 466         const char *name;
 467         const struct file_operations *i_fop;
 468         umode_t mode;
 469 };
 470
 471 static int rpc_delete_dentry(const struct dentry *dentry)
 472 {
 473         return 1;
 474 }
 475
 476 static const struct dentry_operations rpc_dentry_operations = {
 477         .d_delete = rpc_delete_dentry,
 478 };
 479
 480 static struct inode *
 481 rpc_get_inode(struct super_block *sb, umode_t mode)
 482 {
 483         struct inode *inode = new_inode(sb);
 484         if (!inode)
 485                 return NULL;
 486         inode->i_ino = get_next_ino();
 487         inode->i_mode = mode;
 488         inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME;
 489         switch (mode & S_IFMT) {
 490         case S_IFDIR:
 491                 inode->i_fop = &simple_dir_operations;
 492                 inode->i_op = &simple_dir_inode_operations;
 493                 inc_nlink(inode);
 494         default:
 495                 break;
 496         }
 497         return inode;
 498 }
 499
 500 static int __rpc_create_common(struct inode *dir, struct dentry *dentry,
 501                                umode_t mode,
 502                                const struct file_operations *i_fop,
 503                                void *private)
 504 {
 505         struct inode *inode;
 506
 507         d_drop(dentry);
 508         inode = rpc_get_inode(dir->i_sb, mode);
 509         if (!inode)
 510                 goto out_err;
 511         inode->i_ino = iunique(dir->i_sb, 100);
 512         if (i_fop)
 513                 inode->i_fop = i_fop;
 514         if (private)
 515                 rpc_inode_setowner(inode, private);
 516         d_add(dentry, inode);
 517         return 0;
 518 out_err:
 519         printk(KERN_WARNING "%s: %s failed to allocate inode for dentry %s\n",
 520                         __FILE__, __func__, dentry->d_name.name);
 521         dput(dentry);
 522         return -ENOMEM;
 523 }
 524
 525 static int __rpc_create(struct inode *dir, struct dentry *dentry,
 526                         umode_t mode,
 527                         const struct file_operations *i_fop,
 528                         void *private)
 529 {
 530         int err;
 531
 532         err = __rpc_create_common(dir, dentry, S_IFREG | mode, i_fop, private);
 533         if (err)
 534                 return err;
 535         fsnotify_create(dir, dentry);
 536         return 0;
 537 }
 538
 539 static int __rpc_mkdir(struct inode *dir, struct dentry *dentry,
 540                        umode_t mode,
 541                        const struct file_operations *i_fop,
 542                        void *private)
 543 {
 544         int err;
 545
 546         err = __rpc_create_common(dir, dentry, S_IFDIR | mode, i_fop, private);
 547         if (err)
 548                 return err;
 549         inc_nlink(dir);
 550         fsnotify_mkdir(dir, dentry);
 551         return 0;
 552 }
 553
 554 static void
 555 init_pipe(struct rpc_pipe *pipe)
 556 {
 557         pipe->nreaders = 0;
 558         pipe->nwriters = 0;
 559         INIT_LIST_HEAD(&pipe->in_upcall);
 560         INIT_LIST_HEAD(&pipe->in_downcall);
 561         INIT_LIST_HEAD(&pipe->pipe);
 562         pipe->pipelen = 0;
 563         INIT_DELAYED_WORK(&pipe->queue_timeout,
 564                             rpc_timeout_upcall_queue);
 565         pipe->ops = NULL;
 566         spin_lock_init(&pipe->lock);
 567         pipe->dentry = NULL;
 568 }
 569
 570 void rpc_destroy_pipe_data(struct rpc_pipe *pipe)
 571 {
 572         kfree(pipe);
 573 }
 574 EXPORT_SYMBOL_GPL(rpc_destroy_pipe_data);
 575
 576 struct rpc_pipe *rpc_mkpipe_data(const struct rpc_pipe_ops *ops, int flags)
 577 {
 578         struct rpc_pipe *pipe;
 579
 580         pipe = kzalloc(sizeof(struct rpc_pipe), GFP_KERNEL);
 581         if (!pipe)
 582                 return ERR_PTR(-ENOMEM);
 583         init_pipe(pipe);
 584         pipe->ops = ops;
 585         pipe->flags = flags;
 586         return pipe;
 587 }
 588 EXPORT_SYMBOL_GPL(rpc_mkpipe_data);
 589
 590 static int __rpc_mkpipe_dentry(struct inode *dir, struct dentry *dentry,
 591                                umode_t mode,
 592                                const struct file_operations *i_fop,
 593                                void *private,
 594                                struct rpc_pipe *pipe)
 595 {
 596         struct rpc_inode *rpci;
 597         int err;
 598
 599         err = __rpc_create_common(dir, dentry, S_IFIFO | mode, i_fop, private);
 600         if (err)
 601                 return err;
 602         rpci = RPC_I(dentry->d_inode);
 603         rpci->private = private;
 604         rpci->pipe = pipe;
 605         fsnotify_create(dir, dentry);
 606         return 0;
 607 }
 608
 609 static int __rpc_rmdir(struct inode *dir, struct dentry *dentry)
 610 {
 611         int ret;
 612
 613         dget(dentry);
 614         ret = simple_rmdir(dir, dentry);
 615         d_delete(dentry);
 616         dput(dentry);
 617         return ret;
 618 }
 619
 620 int rpc_rmdir(struct dentry *dentry)
 621 {
 622         struct dentry *parent;
 623         struct inode *dir;
 624         int error;
 625
 626         parent = dget_parent(dentry);
 627         dir = parent->d_inode;
 628         mutex_lock_nested(&dir->i_mutex, I_MUTEX_PARENT);
 629         error = __rpc_rmdir(dir, dentry);
 630         mutex_unlock(&dir->i_mutex);
 631         dput(parent);
 632         return error;
 633 }
 634 EXPORT_SYMBOL_GPL(rpc_rmdir);
 635
 636 static int __rpc_unlink(struct inode *dir, struct dentry *dentry)
 637 {
 638         int ret;
 639
 640         dget(dentry);
 641         ret = simple_unlink(dir, dentry);
 642         d_delete(dentry);
 643         dput(dentry);
 644         return ret;
 645 }
 646
 647 static int __rpc_rmpipe(struct inode *dir, struct dentry *dentry)
 648 {
 649         struct inode *inode = dentry->d_inode;
 650
 651         rpc_close_pipes(inode);
 652         return __rpc_unlink(dir, dentry);
 653 }
 654
 655 static struct dentry *__rpc_lookup_create_exclusive(struct dentry *parent,
 656                                           struct qstr *name)
 657 {
 658         struct dentry *dentry;
 659
 660         dentry = d_lookup(parent, name);
 661         if (!dentry) {
 662                 dentry = d_alloc(parent, name);
 663                 if (!dentry)
 664                         return ERR_PTR(-ENOMEM);
 665         }
 666         if (dentry->d_inode == NULL) {
 667                 d_set_d_op(dentry, &rpc_dentry_operations);
 668                 return dentry;
 669         }
 670         dput(dentry);
 671         return ERR_PTR(-EEXIST);
 672 }
 673
 674 /*
 675  * FIXME: This probably has races.
 676  */
 677 static void __rpc_depopulate(struct dentry *parent,
 678                              const struct rpc_filelist *files,
 679                              int start, int eof)
 680 {
 681         struct inode *dir = parent->d_inode;
 682         struct dentry *dentry;
 683         struct qstr name;
 684         int i;
 685
 686         for (i = start; i < eof; i++) {
 687                 name.name = files[i].name;
 688                 name.len = strlen(files[i].name);
 689                 name.hash = full_name_hash(name.name, name.len);
 690                 dentry = d_lookup(parent, &name);
 691
 692                 if (dentry == NULL)
 693                         continue;
 694                 if (dentry->d_inode == NULL)
 695                         goto next;
 696                 switch (dentry->d_inode->i_mode & S_IFMT) {
 697                         default:
 698                                 BUG();
 699                         case S_IFREG:
 700                                 __rpc_unlink(dir, dentry);
 701                                 break;
 702                         case S_IFDIR:
 703                                 __rpc_rmdir(dir, dentry);
 704                 }
 705 next:
 706                 dput(dentry);
 707         }
 708 }
 709
 710 static void rpc_depopulate(struct dentry *parent,
 711                            const struct rpc_filelist *files,
 712                            int start, int eof)
 713 {
 714         struct inode *dir = parent->d_inode;
 715
 716         mutex_lock_nested(&dir->i_mutex, I_MUTEX_CHILD);
 717         __rpc_depopulate(parent, files, start, eof);
 718         mutex_unlock(&dir->i_mutex);
 719 }
 720
 721 static int rpc_populate(struct dentry *parent,
 722                         const struct rpc_filelist *files,
 723                         int start, int eof,
 724                         void *private)
 725 {
 726         struct inode *dir = parent->d_inode;
 727         struct dentry *dentry;
 728         int i, err;
 729
 730         mutex_lock(&dir->i_mutex);
 731         for (i = start; i < eof; i++) {
 732                 struct qstr q;
 733
 734                 q.name = files[i].name;
 735                 q.len = strlen(files[i].name);
 736                 q.hash = full_name_hash(q.name, q.len);
 737                 dentry = __rpc_lookup_create_exclusive(parent, &q);
 738                 err = PTR_ERR(dentry);
 739                 if (IS_ERR(dentry))
 740                         goto out_bad;
 741                 switch (files[i].mode & S_IFMT) {
 742                         default:
 743                                 BUG();
 744                         case S_IFREG:
 745                                 err = __rpc_create(dir, dentry,
 746                                                 files[i].mode,
 747                                                 files[i].i_fop,
 748                                                 private);
 749                                 break;
 750                         case S_IFDIR:
 751                                 err = __rpc_mkdir(dir, dentry,
 752                                                 files[i].mode,
 753                                                 NULL,
 754                                                 private);
 755                 }
 756                 if (err != 0)
 757                         goto out_bad;
 758         }
 759         mutex_unlock(&dir->i_mutex);
 760         return 0;
 761 out_bad:
 762         __rpc_depopulate(parent, files, start, eof);
 763         mutex_unlock(&dir->i_mutex);
 764         printk(KERN_WARNING "%s: %s failed to populate directory %s\n",
 765                         __FILE__, __func__, parent->d_name.name);
 766         return err;
 767 }
 768
 769 static struct dentry *rpc_mkdir_populate(struct dentry *parent,
 770                 struct qstr *name, umode_t mode, void *private,
 771                 int (*populate)(struct dentry *, void *), void *args_populate)
 772 {
 773         struct dentry *dentry;
 774         struct inode *dir = parent->d_inode;
 775         int error;
 776
 777         mutex_lock_nested(&dir->i_mutex, I_MUTEX_PARENT);
 778         dentry = __rpc_lookup_create_exclusive(parent, name);
 779         if (IS_ERR(dentry))
 780                 goto out;
 781         error = __rpc_mkdir(dir, dentry, mode, NULL, private);
 782         if (error != 0)
 783                 goto out_err;
 784         if (populate != NULL) {
 785                 error = populate(dentry, args_populate);
 786                 if (error)
 787                         goto err_rmdir;
 788         }
 789 out:
 790         mutex_unlock(&dir->i_mutex);
 791         return dentry;
 792 err_rmdir:
 793         __rpc_rmdir(dir, dentry);
 794 out_err:
 795         dentry = ERR_PTR(error);
 796         goto out;
 797 }
 798
 799 static int rpc_rmdir_depopulate(struct dentry *dentry,
 800                 void (*depopulate)(struct dentry *))
 801 {
 802         struct dentry *parent;
 803         struct inode *dir;
 804         int error;
 805
 806         parent = dget_parent(dentry);
 807         dir = parent->d_inode;
 808         mutex_lock_nested(&dir->i_mutex, I_MUTEX_PARENT);
 809         if (depopulate != NULL)
 810                 depopulate(dentry);
 811         error = __rpc_rmdir(dir, dentry);
 812         mutex_unlock(&dir->i_mutex);
 813         dput(parent);
 814         return error;
 815 }
 816
 817 /**
 818  * rpc_mkpipe - make an rpc_pipefs file for kernel<->userspace communication
 819  * @parent: dentry of directory to create new "pipe" in
 820  * @name: name of pipe
 821  * @private: private data to associate with the pipe, for the caller's use
 822  * @ops: operations defining the behavior of the pipe: upcall, downcall,
 823  *      release_pipe, open_pipe, and destroy_msg.
 824  * @flags: rpc_pipe flags
 825  *
 826  * Data is made available for userspace to read by calls to
 827  * rpc_queue_upcall().  The actual reads will result in calls to
 828  * @ops->upcall, which will be called with the file pointer,
 829  * message, and userspace buffer to copy to.
 830  *
 831  * Writes can come at any time, and do not necessarily have to be
 832  * responses to upcalls.  They will result in calls to @msg->downcall.
 833  *
 834  * The @private argument passed here will be available to all these methods
 835  * from the file pointer, via RPC_I(file->f_dentry->d_inode)->private.
 836  */
 837 struct dentry *rpc_mkpipe_dentry(struct dentry *parent, const char *name,
 838                                  void *private, struct rpc_pipe *pipe)
 839 {
 840         struct dentry *dentry;
 841         struct inode *dir = parent->d_inode;
 842         umode_t umode = S_IFIFO | S_IRUSR | S_IWUSR;
 843         struct qstr q;
 844         int err;
 845
 846         if (pipe->ops->upcall == NULL)
 847                 umode &= ~S_IRUGO;
 848         if (pipe->ops->downcall == NULL)
 849                 umode &= ~S_IWUGO;
 850
 851         q.name = name;
 852         q.len = strlen(name);
 853         q.hash = full_name_hash(q.name, q.len),
 854
 855         mutex_lock_nested(&dir->i_mutex, I_MUTEX_PARENT);
 856         dentry = __rpc_lookup_create_exclusive(parent, &q);
 857         if (IS_ERR(dentry))
 858                 goto out;
 859         err = __rpc_mkpipe_dentry(dir, dentry, umode, &rpc_pipe_fops,
 860                                   private, pipe);
 861         if (err)
 862                 goto out_err;
 863 out:
 864         mutex_unlock(&dir->i_mutex);
 865         return dentry;
 866 out_err:
 867         dentry = ERR_PTR(err);
 868         printk(KERN_WARNING "%s: %s() failed to create pipe %s/%s (errno = %d)\n",
 869                         __FILE__, __func__, parent->d_name.name, name,
 870                         err);
 871         goto out;
 872 }
 873 EXPORT_SYMBOL_GPL(rpc_mkpipe_dentry);
 874
 875 /**
 876  * rpc_unlink - remove a pipe
 877  * @dentry: dentry for the pipe, as returned from rpc_mkpipe
 878  *
 879  * After this call, lookups will no longer find the pipe, and any
 880  * attempts to read or write using preexisting opens of the pipe will
 881  * return -EPIPE.
 882  */
 883 int
 884 rpc_unlink(struct dentry *dentry)
 885 {
 886         struct dentry *parent;
 887         struct inode *dir;
 888         int error = 0;
 889
 890         parent = dget_parent(dentry);
 891         dir = parent->d_inode;
 892         mutex_lock_nested(&dir->i_mutex, I_MUTEX_PARENT);
 893         error = __rpc_rmpipe(dir, dentry);
 894         mutex_unlock(&dir->i_mutex);
 895         dput(parent);
 896         return error;
 897 }
 898 EXPORT_SYMBOL_GPL(rpc_unlink);
 899
 900 enum {
 901         RPCAUTH_info,
 902         RPCAUTH_EOF
 903 };
 904
 905 static const struct rpc_filelist authfiles[] = {
 906         [RPCAUTH_info] = {
 907                 .name = "info",
 908                 .i_fop = &rpc_info_operations,
 909                 .mode = S_IFREG | S_IRUSR,
 910         },
 911 };
 912
 913 static int rpc_clntdir_populate(struct dentry *dentry, void *private)
 914 {
 915         return rpc_populate(dentry,
 916                             authfiles, RPCAUTH_info, RPCAUTH_EOF,
 917                             private);
 918 }
 919
 920 static void rpc_clntdir_depopulate(struct dentry *dentry)
 921 {
 922         rpc_depopulate(dentry, authfiles, RPCAUTH_info, RPCAUTH_EOF);
 923 }
 924
 925 /**
 926  * rpc_create_client_dir - Create a new rpc_client directory in rpc_pipefs
 927  * @dentry: dentry from the rpc_pipefs root to the new directory
 928  * @name: &struct qstr for the name
 929  * @rpc_client: rpc client to associate with this directory
 930  *
 931  * This creates a directory at the given @path associated with
 932  * @rpc_clnt, which will contain a file named "info" with some basic
 933  * information about the client, together with any "pipes" that may
 934  * later be created using rpc_mkpipe().
 935  */
 936 struct dentry *rpc_create_client_dir(struct dentry *dentry,
 937                                    struct qstr *name,
 938                                    struct rpc_clnt *rpc_client)
 939 {
 940         return rpc_mkdir_populate(dentry, name, S_IRUGO | S_IXUGO, NULL,
 941                         rpc_clntdir_populate, rpc_client);
 942 }
 943
 944 /**
 945  * rpc_remove_client_dir - Remove a directory created with rpc_create_client_dir()
 946  * @clnt: rpc client
 947  */
 948 int rpc_remove_client_dir(struct dentry *dentry)
 949 {
 950         return rpc_rmdir_depopulate(dentry, rpc_clntdir_depopulate);
 951 }
 952
 953 static const struct rpc_filelist cache_pipefs_files[3] = {
 954         [0] = {
 955                 .name = "channel",
 956                 .i_fop = &cache_file_operations_pipefs,
 957                 .mode = S_IFREG|S_IRUSR|S_IWUSR,
 958         },
 959         [1] = {
 960                 .name = "content",
 961                 .i_fop = &content_file_operations_pipefs,
 962                 .mode = S_IFREG|S_IRUSR,
 963         },
 964         [2] = {
 965                 .name = "flush",
 966                 .i_fop = &cache_flush_operations_pipefs,
 967                 .mode = S_IFREG|S_IRUSR|S_IWUSR,
 968         },
 969 };
 970
 971 static int rpc_cachedir_populate(struct dentry *dentry, void *private)
 972 {
 973         return rpc_populate(dentry,
 974                             cache_pipefs_files, 0, 3,
 975                             private);
 976 }
 977
 978 static void rpc_cachedir_depopulate(struct dentry *dentry)
 979 {
 980         rpc_depopulate(dentry, cache_pipefs_files, 0, 3);
 981 }
 982
 983 struct dentry *rpc_create_cache_dir(struct dentry *parent, struct qstr *name,
 984                                     umode_t umode, struct cache_detail *cd)
 985 {
 986         return rpc_mkdir_populate(parent, name, umode, NULL,
 987                         rpc_cachedir_populate, cd);
 988 }
 989
 990 void rpc_remove_cache_dir(struct dentry *dentry)
 991 {
 992         rpc_rmdir_depopulate(dentry, rpc_cachedir_depopulate);
 993 }
 994
 995 /*
 996  * populate the filesystem
 997  */
 998 static const struct super_operations s_ops = {
 999         .alloc_inode    = rpc_alloc_inode,
1000         .destroy_inode  = rpc_destroy_inode,
1001         .statfs         = simple_statfs,
1002 };
1003
1004 #define RPCAUTH_GSSMAGIC 0x67596969
1005
1006 /*
1007  * We have a single directory with 1 node in it.
1008  */
1009 enum {
1010         RPCAUTH_lockd,
1011         RPCAUTH_mount,
1012         RPCAUTH_nfs,
1013         RPCAUTH_portmap,
1014         RPCAUTH_statd,
1015         RPCAUTH_nfsd4_cb,
1016         RPCAUTH_cache,
1017         RPCAUTH_nfsd,
1018         RPCAUTH_RootEOF
1019 };
1020
1021 static const struct rpc_filelist files[] = {
1022         [RPCAUTH_lockd] = {
1023                 .name = "lockd",
1024                 .mode = S_IFDIR | S_IRUGO | S_IXUGO,
1025         },
1026         [RPCAUTH_mount] = {
1027                 .name = "mount",
1028                 .mode = S_IFDIR | S_IRUGO | S_IXUGO,
1029         },
1030         [RPCAUTH_nfs] = {
1031                 .name = "nfs",
1032                 .mode = S_IFDIR | S_IRUGO | S_IXUGO,
1033         },
1034         [RPCAUTH_portmap] = {
1035                 .name = "portmap",
1036                 .mode = S_IFDIR | S_IRUGO | S_IXUGO,
1037         },
1038         [RPCAUTH_statd] = {
1039                 .name = "statd",
1040                 .mode = S_IFDIR | S_IRUGO | S_IXUGO,
1041         },
1042         [RPCAUTH_nfsd4_cb] = {
1043                 .name = "nfsd4_cb",
1044                 .mode = S_IFDIR | S_IRUGO | S_IXUGO,
1045         },
1046         [RPCAUTH_cache] = {
1047                 .name = "cache",
1048                 .mode = S_IFDIR | S_IRUGO | S_IXUGO,
1049         },
1050         [RPCAUTH_nfsd] = {
1051                 .name = "nfsd",
1052                 .mode = S_IFDIR | S_IRUGO | S_IXUGO,
1053         },
1054 };
1055
1056 /*
1057  * This call can be used only in RPC pipefs mount notification hooks.
1058  */
1059 struct dentry *rpc_d_lookup_sb(const struct super_block *sb,
1060                                const unsigned char *dir_name)
1061 {
1062         struct qstr dir = QSTR_INIT(dir_name, strlen(dir_name));
1063
1064         dir.hash = full_name_hash(dir.name, dir.len);
1065         return d_lookup(sb->s_root, &dir);
1066 }
1067 EXPORT_SYMBOL_GPL(rpc_d_lookup_sb);
1068
1069 void rpc_pipefs_init_net(struct net *net)
1070 {
1071         struct sunrpc_net *sn = net_generic(net, sunrpc_net_id);
1072
1073         mutex_init(&sn->pipefs_sb_lock);
1074 }
1075
1076 /*
1077  * This call will be used for per network namespace operations calls.
1078  * Note: Function will be returned with pipefs_sb_lock taken if superblock was
1079  * found. This lock have to be released by rpc_put_sb_net() when all operations
1080  * will be completed.
1081  */
1082 struct super_block *rpc_get_sb_net(const struct net *net)
1083 {
1084         struct sunrpc_net *sn = net_generic(net, sunrpc_net_id);
1085
1086         mutex_lock(&sn->pipefs_sb_lock);
1087         if (sn->pipefs_sb)
1088                 return sn->pipefs_sb;
1089         mutex_unlock(&sn->pipefs_sb_lock);
1090         return NULL;
1091 }
1092 EXPORT_SYMBOL_GPL(rpc_get_sb_net);
1093
1094 void rpc_put_sb_net(const struct net *net)
1095 {
1096         struct sunrpc_net *sn = net_generic(net, sunrpc_net_id);
1097
1098         BUG_ON(sn->pipefs_sb == NULL);
1099         mutex_unlock(&sn->pipefs_sb_lock);
1100 }
1101 EXPORT_SYMBOL_GPL(rpc_put_sb_net);
1102
1103 static int
1104 rpc_fill_super(struct super_block *sb, void *data, int silent)
1105 {
1106         struct inode *inode;
1107         struct dentry *root;
1108         struct net *net = data;
1109         struct sunrpc_net *sn = net_generic(net, sunrpc_net_id);
1110         int err;
1111
1112         sb->s_blocksize = PAGE_CACHE_SIZE;
1113         sb->s_blocksize_bits = PAGE_CACHE_SHIFT;
1114         sb->s_magic = RPCAUTH_GSSMAGIC;
1115         sb->s_op = &s_ops;
1116         sb->s_time_gran = 1;
1117
1118         inode = rpc_get_inode(sb, S_IFDIR | 0755);
1119         sb->s_root = root = d_make_root(inode);
1120         if (!root)
1121                 return -ENOMEM;
1122         if (rpc_populate(root, files, RPCAUTH_lockd, RPCAUTH_RootEOF, NULL))
1123                 return -ENOMEM;
1124         dprintk("RPC:   sending pipefs MOUNT notification for net %p%s\n", net,
1125                                                                 NET_NAME(net));
1126         sn->pipefs_sb = sb;
1127         err = blocking_notifier_call_chain(&rpc_pipefs_notifier_list,
1128                                            RPC_PIPEFS_MOUNT,
1129                                            sb);
1130         if (err)
1131                 goto err_depopulate;
1132         sb->s_fs_info = get_net(net);
1133         return 0;
1134
1135 err_depopulate:
1136         blocking_notifier_call_chain(&rpc_pipefs_notifier_list,
1137                                            RPC_PIPEFS_UMOUNT,
1138                                            sb);
1139         sn->pipefs_sb = NULL;
1140         __rpc_depopulate(root, files, RPCAUTH_lockd, RPCAUTH_RootEOF);
1141         return err;
1142 }
1143
1144 static struct dentry *
1145 rpc_mount(struct file_system_type *fs_type,
1146                 int flags, const char *dev_name, void *data)
1147 {
1148         return mount_ns(fs_type, flags, current->nsproxy->net_ns, rpc_fill_super);
1149 }
1150
1151 static void rpc_kill_sb(struct super_block *sb)
1152 {
1153         struct net *net = sb->s_fs_info;
1154         struct sunrpc_net *sn = net_generic(net, sunrpc_net_id);
1155
1156         mutex_lock(&sn->pipefs_sb_lock);
1157         sn->pipefs_sb = NULL;
1158         mutex_unlock(&sn->pipefs_sb_lock);
1159         put_net(net);
1160         dprintk("RPC:   sending pipefs UMOUNT notification for net %p%s\n", net,
1161                                                                 NET_NAME(net));
1162         blocking_notifier_call_chain(&rpc_pipefs_notifier_list,
1163                                            RPC_PIPEFS_UMOUNT,
1164                                            sb);
1165         kill_litter_super(sb);
1166 }
1167
1168 static struct file_system_type rpc_pipe_fs_type = {
1169         .owner          = THIS_MODULE,
1170         .name           = "rpc_pipefs",
1171         .mount          = rpc_mount,
1172         .kill_sb        = rpc_kill_sb,
1173 };
1174
1175 static void
1176 init_once(void *foo)
1177 {
1178         struct rpc_inode *rpci = (struct rpc_inode *) foo;
1179
1180         inode_init_once(&rpci->vfs_inode);
1181         rpci->private = NULL;
1182         rpci->pipe = NULL;
1183         init_waitqueue_head(&rpci->waitq);
1184 }
1185
1186 int register_rpc_pipefs(void)
1187 {
1188         int err;
1189
1190         rpc_inode_cachep = kmem_cache_create("rpc_inode_cache",
1191                                 sizeof(struct rpc_inode),
1192                                 0, (SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT|
1193                                                 SLAB_MEM_SPREAD),
1194                                 init_once);
1195         if (!rpc_inode_cachep)
1196                 return -ENOMEM;
1197         err = rpc_clients_notifier_register();
1198         if (err)
1199                 goto err_notifier;
1200         err = register_filesystem(&rpc_pipe_fs_type);
1201         if (err)
1202                 goto err_register;
1203         return 0;
1204
1205 err_register:
1206         rpc_clients_notifier_unregister();
1207 err_notifier:
1208         kmem_cache_destroy(rpc_inode_cachep);
1209         return err;
1210 }
1211
1212 void unregister_rpc_pipefs(void)
1213 {
1214         rpc_clients_notifier_unregister();
1215         kmem_cache_destroy(rpc_inode_cachep);
1216         unregister_filesystem(&rpc_pipe_fs_type);
1217 }
1218
1219 /* Make 'mount -t rpc_pipefs ...' autoload this module. */
1220 MODULE_ALIAS("rpc_pipefs");