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 static struct inode *
 484 rpc_get_inode(struct super_block *sb, umode_t mode)
 485 {
 486         struct inode *inode = new_inode(sb);
 487         if (!inode)
 488                 return NULL;
 489         inode->i_ino = get_next_ino();
 490         inode->i_mode = mode;
 491         inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME;
 492         switch (mode & S_IFMT) {
 493         case S_IFDIR:
 494                 inode->i_fop = &simple_dir_operations;
 495                 inode->i_op = &simple_dir_inode_operations;
 496                 inc_nlink(inode);
 497         default:
 498                 break;
 499         }
 500         return inode;
 501 }
 502
 503 static int __rpc_create_common(struct inode *dir, struct dentry *dentry,
 504                                umode_t mode,
 505                                const struct file_operations *i_fop,
 506                                void *private)
 507 {
 508         struct inode *inode;
 509
 510         d_drop(dentry);
 511         inode = rpc_get_inode(dir->i_sb, mode);
 512         if (!inode)
 513                 goto out_err;
 514         inode->i_ino = iunique(dir->i_sb, 100);
 515         if (i_fop)
 516                 inode->i_fop = i_fop;
 517         if (private)
 518                 rpc_inode_setowner(inode, private);
 519         d_add(dentry, inode);
 520         return 0;
 521 out_err:
 522         printk(KERN_WARNING "%s: %s failed to allocate inode for dentry %s\n",
 523                         __FILE__, __func__, dentry->d_name.name);
 524         dput(dentry);
 525         return -ENOMEM;
 526 }
 527
 528 static int __rpc_create(struct inode *dir, struct dentry *dentry,
 529                         umode_t mode,
 530                         const struct file_operations *i_fop,
 531                         void *private)
 532 {
 533         int err;
 534
 535         err = __rpc_create_common(dir, dentry, S_IFREG | mode, i_fop, private);
 536         if (err)
 537                 return err;
 538         fsnotify_create(dir, dentry);
 539         return 0;
 540 }
 541
 542 static int __rpc_mkdir(struct inode *dir, struct dentry *dentry,
 543                        umode_t mode,
 544                        const struct file_operations *i_fop,
 545                        void *private)
 546 {
 547         int err;
 548
 549         err = __rpc_create_common(dir, dentry, S_IFDIR | mode, i_fop, private);
 550         if (err)
 551                 return err;
 552         inc_nlink(dir);
 553         fsnotify_mkdir(dir, dentry);
 554         return 0;
 555 }
 556
 557 static void
 558 init_pipe(struct rpc_pipe *pipe)
 559 {
 560         pipe->nreaders = 0;
 561         pipe->nwriters = 0;
 562         INIT_LIST_HEAD(&pipe->in_upcall);
 563         INIT_LIST_HEAD(&pipe->in_downcall);
 564         INIT_LIST_HEAD(&pipe->pipe);
 565         pipe->pipelen = 0;
 566         INIT_DELAYED_WORK(&pipe->queue_timeout,
 567                             rpc_timeout_upcall_queue);
 568         pipe->ops = NULL;
 569         spin_lock_init(&pipe->lock);
 570         pipe->dentry = NULL;
 571 }
 572
 573 void rpc_destroy_pipe_data(struct rpc_pipe *pipe)
 574 {
 575         kfree(pipe);
 576 }
 577 EXPORT_SYMBOL_GPL(rpc_destroy_pipe_data);
 578
 579 struct rpc_pipe *rpc_mkpipe_data(const struct rpc_pipe_ops *ops, int flags)
 580 {
 581         struct rpc_pipe *pipe;
 582
 583         pipe = kzalloc(sizeof(struct rpc_pipe), GFP_KERNEL);
 584         if (!pipe)
 585                 return ERR_PTR(-ENOMEM);
 586         init_pipe(pipe);
 587         pipe->ops = ops;
 588         pipe->flags = flags;
 589         return pipe;
 590 }
 591 EXPORT_SYMBOL_GPL(rpc_mkpipe_data);
 592
 593 static int __rpc_mkpipe_dentry(struct inode *dir, struct dentry *dentry,
 594                                umode_t mode,
 595                                const struct file_operations *i_fop,
 596                                void *private,
 597                                struct rpc_pipe *pipe)
 598 {
 599         struct rpc_inode *rpci;
 600         int err;
 601
 602         err = __rpc_create_common(dir, dentry, S_IFIFO | mode, i_fop, private);
 603         if (err)
 604                 return err;
 605         rpci = RPC_I(dentry->d_inode);
 606         rpci->private = private;
 607         rpci->pipe = pipe;
 608         fsnotify_create(dir, dentry);
 609         return 0;
 610 }
 611
 612 static int __rpc_rmdir(struct inode *dir, struct dentry *dentry)
 613 {
 614         int ret;
 615
 616         dget(dentry);
 617         ret = simple_rmdir(dir, dentry);
 618         d_delete(dentry);
 619         dput(dentry);
 620         return ret;
 621 }
 622
 623 int rpc_rmdir(struct dentry *dentry)
 624 {
 625         struct dentry *parent;
 626         struct inode *dir;
 627         int error;
 628
 629         parent = dget_parent(dentry);
 630         dir = parent->d_inode;
 631         mutex_lock_nested(&dir->i_mutex, I_MUTEX_PARENT);
 632         error = __rpc_rmdir(dir, dentry);
 633         mutex_unlock(&dir->i_mutex);
 634         dput(parent);
 635         return error;
 636 }
 637 EXPORT_SYMBOL_GPL(rpc_rmdir);
 638
 639 static int __rpc_unlink(struct inode *dir, struct dentry *dentry)
 640 {
 641         int ret;
 642
 643         dget(dentry);
 644         ret = simple_unlink(dir, dentry);
 645         d_delete(dentry);
 646         dput(dentry);
 647         return ret;
 648 }
 649
 650 static int __rpc_rmpipe(struct inode *dir, struct dentry *dentry)
 651 {
 652         struct inode *inode = dentry->d_inode;
 653
 654         rpc_close_pipes(inode);
 655         return __rpc_unlink(dir, dentry);
 656 }
 657
 658 static struct dentry *__rpc_lookup_create_exclusive(struct dentry *parent,
 659                                           struct qstr *name)
 660 {
 661         struct dentry *dentry;
 662
 663         dentry = d_lookup(parent, name);
 664         if (!dentry) {
 665                 dentry = d_alloc(parent, name);
 666                 if (!dentry)
 667                         return ERR_PTR(-ENOMEM);
 668         }
 669         if (dentry->d_inode == NULL) {
 670                 if (!dentry->d_op)
 671                         d_set_d_op(dentry, &rpc_dentry_operations);
 672                 return dentry;
 673         }
 674         dput(dentry);
 675         return ERR_PTR(-EEXIST);
 676 }
 677
 678 /*
 679  * FIXME: This probably has races.
 680  */
 681 static void __rpc_depopulate(struct dentry *parent,
 682                              const struct rpc_filelist *files,
 683                              int start, int eof)
 684 {
 685         struct inode *dir = parent->d_inode;
 686         struct dentry *dentry;
 687         struct qstr name;
 688         int i;
 689
 690         for (i = start; i < eof; i++) {
 691                 name.name = files[i].name;
 692                 name.len = strlen(files[i].name);
 693                 name.hash = full_name_hash(name.name, name.len);
 694                 dentry = d_lookup(parent, &name);
 695
 696                 if (dentry == NULL)
 697                         continue;
 698                 if (dentry->d_inode == NULL)
 699                         goto next;
 700                 switch (dentry->d_inode->i_mode & S_IFMT) {
 701                         default:
 702                                 BUG();
 703                         case S_IFREG:
 704                                 __rpc_unlink(dir, dentry);
 705                                 break;
 706                         case S_IFDIR:
 707                                 __rpc_rmdir(dir, dentry);
 708                 }
 709 next:
 710                 dput(dentry);
 711         }
 712 }
 713
 714 static void rpc_depopulate(struct dentry *parent,
 715                            const struct rpc_filelist *files,
 716                            int start, int eof)
 717 {
 718         struct inode *dir = parent->d_inode;
 719
 720         mutex_lock_nested(&dir->i_mutex, I_MUTEX_CHILD);
 721         __rpc_depopulate(parent, files, start, eof);
 722         mutex_unlock(&dir->i_mutex);
 723 }
 724
 725 static int rpc_populate(struct dentry *parent,
 726                         const struct rpc_filelist *files,
 727                         int start, int eof,
 728                         void *private)
 729 {
 730         struct inode *dir = parent->d_inode;
 731         struct dentry *dentry;
 732         int i, err;
 733
 734         mutex_lock(&dir->i_mutex);
 735         for (i = start; i < eof; i++) {
 736                 struct qstr q;
 737
 738                 q.name = files[i].name;
 739                 q.len = strlen(files[i].name);
 740                 q.hash = full_name_hash(q.name, q.len);
 741                 dentry = __rpc_lookup_create_exclusive(parent, &q);
 742                 err = PTR_ERR(dentry);
 743                 if (IS_ERR(dentry))
 744                         goto out_bad;
 745                 switch (files[i].mode & S_IFMT) {
 746                         default:
 747                                 BUG();
 748                         case S_IFREG:
 749                                 err = __rpc_create(dir, dentry,
 750                                                 files[i].mode,
 751                                                 files[i].i_fop,
 752                                                 private);
 753                                 break;
 754                         case S_IFDIR:
 755                                 err = __rpc_mkdir(dir, dentry,
 756                                                 files[i].mode,
 757                                                 NULL,
 758                                                 private);
 759                 }
 760                 if (err != 0)
 761                         goto out_bad;
 762         }
 763         mutex_unlock(&dir->i_mutex);
 764         return 0;
 765 out_bad:
 766         __rpc_depopulate(parent, files, start, eof);
 767         mutex_unlock(&dir->i_mutex);
 768         printk(KERN_WARNING "%s: %s failed to populate directory %s\n",
 769                         __FILE__, __func__, parent->d_name.name);
 770         return err;
 771 }
 772
 773 static struct dentry *rpc_mkdir_populate(struct dentry *parent,
 774                 struct qstr *name, umode_t mode, void *private,
 775                 int (*populate)(struct dentry *, void *), void *args_populate)
 776 {
 777         struct dentry *dentry;
 778         struct inode *dir = parent->d_inode;
 779         int error;
 780
 781         mutex_lock_nested(&dir->i_mutex, I_MUTEX_PARENT);
 782         dentry = __rpc_lookup_create_exclusive(parent, name);
 783         if (IS_ERR(dentry))
 784                 goto out;
 785         error = __rpc_mkdir(dir, dentry, mode, NULL, private);
 786         if (error != 0)
 787                 goto out_err;
 788         if (populate != NULL) {
 789                 error = populate(dentry, args_populate);
 790                 if (error)
 791                         goto err_rmdir;
 792         }
 793 out:
 794         mutex_unlock(&dir->i_mutex);
 795         return dentry;
 796 err_rmdir:
 797         __rpc_rmdir(dir, dentry);
 798 out_err:
 799         dentry = ERR_PTR(error);
 800         goto out;
 801 }
 802
 803 static int rpc_rmdir_depopulate(struct dentry *dentry,
 804                 void (*depopulate)(struct dentry *))
 805 {
 806         struct dentry *parent;
 807         struct inode *dir;
 808         int error;
 809
 810         parent = dget_parent(dentry);
 811         dir = parent->d_inode;
 812         mutex_lock_nested(&dir->i_mutex, I_MUTEX_PARENT);
 813         if (depopulate != NULL)
 814                 depopulate(dentry);
 815         error = __rpc_rmdir(dir, dentry);
 816         mutex_unlock(&dir->i_mutex);
 817         dput(parent);
 818         return error;
 819 }
 820
 821 /**
 822  * rpc_mkpipe - make an rpc_pipefs file for kernel<->userspace communication
 823  * @parent: dentry of directory to create new "pipe" in
 824  * @name: name of pipe
 825  * @private: private data to associate with the pipe, for the caller's use
 826  * @pipe: &rpc_pipe containing input parameters
 827  *
 828  * Data is made available for userspace to read by calls to
 829  * rpc_queue_upcall().  The actual reads will result in calls to
 830  * @ops->upcall, which will be called with the file pointer,
 831  * message, and userspace buffer to copy to.
 832  *
 833  * Writes can come at any time, and do not necessarily have to be
 834  * responses to upcalls.  They will result in calls to @msg->downcall.
 835  *
 836  * The @private argument passed here will be available to all these methods
 837  * from the file pointer, via RPC_I(file_inode(file))->private.
 838  */
 839 struct dentry *rpc_mkpipe_dentry(struct dentry *parent, const char *name,
 840                                  void *private, struct rpc_pipe *pipe)
 841 {
 842         struct dentry *dentry;
 843         struct inode *dir = parent->d_inode;
 844         umode_t umode = S_IFIFO | S_IRUSR | S_IWUSR;
 845         struct qstr q;
 846         int err;
 847
 848         if (pipe->ops->upcall == NULL)
 849                 umode &= ~S_IRUGO;
 850         if (pipe->ops->downcall == NULL)
 851                 umode &= ~S_IWUGO;
 852
 853         q.name = name;
 854         q.len = strlen(name);
 855         q.hash = full_name_hash(q.name, q.len),
 856
 857         mutex_lock_nested(&dir->i_mutex, I_MUTEX_PARENT);
 858         dentry = __rpc_lookup_create_exclusive(parent, &q);
 859         if (IS_ERR(dentry))
 860                 goto out;
 861         err = __rpc_mkpipe_dentry(dir, dentry, umode, &rpc_pipe_fops,
 862                                   private, pipe);
 863         if (err)
 864                 goto out_err;
 865 out:
 866         mutex_unlock(&dir->i_mutex);
 867         return dentry;
 868 out_err:
 869         dentry = ERR_PTR(err);
 870         printk(KERN_WARNING "%s: %s() failed to create pipe %s/%s (errno = %d)\n",
 871                         __FILE__, __func__, parent->d_name.name, name,
 872                         err);
 873         goto out;
 874 }
 875 EXPORT_SYMBOL_GPL(rpc_mkpipe_dentry);
 876
 877 /**
 878  * rpc_unlink - remove a pipe
 879  * @dentry: dentry for the pipe, as returned from rpc_mkpipe
 880  *
 881  * After this call, lookups will no longer find the pipe, and any
 882  * attempts to read or write using preexisting opens of the pipe will
 883  * return -EPIPE.
 884  */
 885 int
 886 rpc_unlink(struct dentry *dentry)
 887 {
 888         struct dentry *parent;
 889         struct inode *dir;
 890         int error = 0;
 891
 892         parent = dget_parent(dentry);
 893         dir = parent->d_inode;
 894         mutex_lock_nested(&dir->i_mutex, I_MUTEX_PARENT);
 895         error = __rpc_rmpipe(dir, dentry);
 896         mutex_unlock(&dir->i_mutex);
 897         dput(parent);
 898         return error;
 899 }
 900 EXPORT_SYMBOL_GPL(rpc_unlink);
 901
 902 enum {
 903         RPCAUTH_info,
 904         RPCAUTH_EOF
 905 };
 906
 907 static const struct rpc_filelist authfiles[] = {
 908         [RPCAUTH_info] = {
 909                 .name = "info",
 910                 .i_fop = &rpc_info_operations,
 911                 .mode = S_IFREG | S_IRUSR,
 912         },
 913 };
 914
 915 static int rpc_clntdir_populate(struct dentry *dentry, void *private)
 916 {
 917         return rpc_populate(dentry,
 918                             authfiles, RPCAUTH_info, RPCAUTH_EOF,
 919                             private);
 920 }
 921
 922 static void rpc_clntdir_depopulate(struct dentry *dentry)
 923 {
 924         rpc_depopulate(dentry, authfiles, RPCAUTH_info, RPCAUTH_EOF);
 925 }
 926
 927 /**
 928  * rpc_create_client_dir - Create a new rpc_client directory in rpc_pipefs
 929  * @dentry: dentry from the rpc_pipefs root to the new directory
 930  * @name: &struct qstr for the name
 931  * @rpc_client: rpc client to associate with this directory
 932  *
 933  * This creates a directory at the given @path associated with
 934  * @rpc_clnt, which will contain a file named "info" with some basic
 935  * information about the client, together with any "pipes" that may
 936  * later be created using rpc_mkpipe().
 937  */
 938 struct dentry *rpc_create_client_dir(struct dentry *dentry,
 939                                    struct qstr *name,
 940                                    struct rpc_clnt *rpc_client)
 941 {
 942         return rpc_mkdir_populate(dentry, name, S_IRUGO | S_IXUGO, NULL,
 943                         rpc_clntdir_populate, rpc_client);
 944 }
 945
 946 /**
 947  * rpc_remove_client_dir - Remove a directory created with rpc_create_client_dir()
 948  * @dentry: dentry for the pipe
 949  */
 950 int rpc_remove_client_dir(struct dentry *dentry)
 951 {
 952         return rpc_rmdir_depopulate(dentry, rpc_clntdir_depopulate);
 953 }
 954
 955 static const struct rpc_filelist cache_pipefs_files[3] = {
 956         [0] = {
 957                 .name = "channel",
 958                 .i_fop = &cache_file_operations_pipefs,
 959                 .mode = S_IFREG|S_IRUSR|S_IWUSR,
 960         },
 961         [1] = {
 962                 .name = "content",
 963                 .i_fop = &content_file_operations_pipefs,
 964                 .mode = S_IFREG|S_IRUSR,
 965         },
 966         [2] = {
 967                 .name = "flush",
 968                 .i_fop = &cache_flush_operations_pipefs,
 969                 .mode = S_IFREG|S_IRUSR|S_IWUSR,
 970         },
 971 };
 972
 973 static int rpc_cachedir_populate(struct dentry *dentry, void *private)
 974 {
 975         return rpc_populate(dentry,
 976                             cache_pipefs_files, 0, 3,
 977                             private);
 978 }
 979
 980 static void rpc_cachedir_depopulate(struct dentry *dentry)
 981 {
 982         rpc_depopulate(dentry, cache_pipefs_files, 0, 3);
 983 }
 984
 985 struct dentry *rpc_create_cache_dir(struct dentry *parent, struct qstr *name,
 986                                     umode_t umode, struct cache_detail *cd)
 987 {
 988         return rpc_mkdir_populate(parent, name, umode, NULL,
 989                         rpc_cachedir_populate, cd);
 990 }
 991
 992 void rpc_remove_cache_dir(struct dentry *dentry)
 993 {
 994         rpc_rmdir_depopulate(dentry, rpc_cachedir_depopulate);
 995 }
 996
 997 /*
 998  * populate the filesystem
 999  */
1000 static const struct super_operations s_ops = {
1001         .alloc_inode    = rpc_alloc_inode,
1002         .destroy_inode  = rpc_destroy_inode,
1003         .statfs         = simple_statfs,
1004 };
1005
1006 #define RPCAUTH_GSSMAGIC 0x67596969
1007
1008 /*
1009  * We have a single directory with 1 node in it.
1010  */
1011 enum {
1012         RPCAUTH_lockd,
1013         RPCAUTH_mount,
1014         RPCAUTH_nfs,
1015         RPCAUTH_portmap,
1016         RPCAUTH_statd,
1017         RPCAUTH_nfsd4_cb,
1018         RPCAUTH_cache,
1019         RPCAUTH_nfsd,
1020         RPCAUTH_RootEOF
1021 };
1022
1023 static const struct rpc_filelist files[] = {
1024         [RPCAUTH_lockd] = {
1025                 .name = "lockd",
1026                 .mode = S_IFDIR | S_IRUGO | S_IXUGO,
1027         },
1028         [RPCAUTH_mount] = {
1029                 .name = "mount",
1030                 .mode = S_IFDIR | S_IRUGO | S_IXUGO,
1031         },
1032         [RPCAUTH_nfs] = {
1033                 .name = "nfs",
1034                 .mode = S_IFDIR | S_IRUGO | S_IXUGO,
1035         },
1036         [RPCAUTH_portmap] = {
1037                 .name = "portmap",
1038                 .mode = S_IFDIR | S_IRUGO | S_IXUGO,
1039         },
1040         [RPCAUTH_statd] = {
1041                 .name = "statd",
1042                 .mode = S_IFDIR | S_IRUGO | S_IXUGO,
1043         },
1044         [RPCAUTH_nfsd4_cb] = {
1045                 .name = "nfsd4_cb",
1046                 .mode = S_IFDIR | S_IRUGO | S_IXUGO,
1047         },
1048         [RPCAUTH_cache] = {
1049                 .name = "cache",
1050                 .mode = S_IFDIR | S_IRUGO | S_IXUGO,
1051         },
1052         [RPCAUTH_nfsd] = {
1053                 .name = "nfsd",
1054                 .mode = S_IFDIR | S_IRUGO | S_IXUGO,
1055         },
1056 };
1057
1058 /*
1059  * This call can be used only in RPC pipefs mount notification hooks.
1060  */
1061 struct dentry *rpc_d_lookup_sb(const struct super_block *sb,
1062                                const unsigned char *dir_name)
1063 {
1064         struct qstr dir = QSTR_INIT(dir_name, strlen(dir_name));
1065
1066         dir.hash = full_name_hash(dir.name, dir.len);
1067         return d_lookup(sb->s_root, &dir);
1068 }
1069 EXPORT_SYMBOL_GPL(rpc_d_lookup_sb);
1070
1071 void rpc_pipefs_init_net(struct net *net)
1072 {
1073         struct sunrpc_net *sn = net_generic(net, sunrpc_net_id);
1074
1075         mutex_init(&sn->pipefs_sb_lock);
1076         sn->gssd_running = 1;
1077         sn->pipe_version = -1;
1078 }
1079
1080 /*
1081  * This call will be used for per network namespace operations calls.
1082  * Note: Function will be returned with pipefs_sb_lock taken if superblock was
1083  * found. This lock have to be released by rpc_put_sb_net() when all operations
1084  * will be completed.
1085  */
1086 struct super_block *rpc_get_sb_net(const struct net *net)
1087 {
1088         struct sunrpc_net *sn = net_generic(net, sunrpc_net_id);
1089
1090         mutex_lock(&sn->pipefs_sb_lock);
1091         if (sn->pipefs_sb)
1092                 return sn->pipefs_sb;
1093         mutex_unlock(&sn->pipefs_sb_lock);
1094         return NULL;
1095 }
1096 EXPORT_SYMBOL_GPL(rpc_get_sb_net);
1097
1098 void rpc_put_sb_net(const struct net *net)
1099 {
1100         struct sunrpc_net *sn = net_generic(net, sunrpc_net_id);
1101
1102         WARN_ON(sn->pipefs_sb == NULL);
1103         mutex_unlock(&sn->pipefs_sb_lock);
1104 }
1105 EXPORT_SYMBOL_GPL(rpc_put_sb_net);
1106
1107 static int
1108 rpc_fill_super(struct super_block *sb, void *data, int silent)
1109 {
1110         struct inode *inode;
1111         struct dentry *root;
1112         struct net *net = data;
1113         struct sunrpc_net *sn = net_generic(net, sunrpc_net_id);
1114         int err;
1115
1116         sb->s_blocksize = PAGE_CACHE_SIZE;
1117         sb->s_blocksize_bits = PAGE_CACHE_SHIFT;
1118         sb->s_magic = RPCAUTH_GSSMAGIC;
1119         sb->s_op = &s_ops;
1120         sb->s_time_gran = 1;
1121
1122         inode = rpc_get_inode(sb, S_IFDIR | S_IRUGO | S_IXUGO);
1123         sb->s_root = root = d_make_root(inode);
1124         if (!root)
1125                 return -ENOMEM;
1126         if (rpc_populate(root, files, RPCAUTH_lockd, RPCAUTH_RootEOF, NULL))
1127                 return -ENOMEM;
1128         dprintk("RPC:       sending pipefs MOUNT notification for net %p%s\n",
1129                 net, NET_NAME(net));
1130         mutex_lock(&sn->pipefs_sb_lock);
1131         sn->pipefs_sb = sb;
1132         err = blocking_notifier_call_chain(&rpc_pipefs_notifier_list,
1133                                            RPC_PIPEFS_MOUNT,
1134                                            sb);
1135         if (err)
1136                 goto err_depopulate;
1137         sb->s_fs_info = get_net(net);
1138         mutex_unlock(&sn->pipefs_sb_lock);
1139         return 0;
1140
1141 err_depopulate:
1142         blocking_notifier_call_chain(&rpc_pipefs_notifier_list,
1143                                            RPC_PIPEFS_UMOUNT,
1144                                            sb);
1145         sn->pipefs_sb = NULL;
1146         __rpc_depopulate(root, files, RPCAUTH_lockd, RPCAUTH_RootEOF);
1147         mutex_unlock(&sn->pipefs_sb_lock);
1148         return err;
1149 }
1150
1151 static struct dentry *
1152 rpc_mount(struct file_system_type *fs_type,
1153                 int flags, const char *dev_name, void *data)
1154 {
1155         return mount_ns(fs_type, flags, current->nsproxy->net_ns, rpc_fill_super);
1156 }
1157
1158 static void rpc_kill_sb(struct super_block *sb)
1159 {
1160         struct net *net = sb->s_fs_info;
1161         struct sunrpc_net *sn = net_generic(net, sunrpc_net_id);
1162
1163         mutex_lock(&sn->pipefs_sb_lock);
1164         if (sn->pipefs_sb != sb) {
1165                 mutex_unlock(&sn->pipefs_sb_lock);
1166                 goto out;
1167         }
1168         sn->pipefs_sb = NULL;
1169         dprintk("RPC:       sending pipefs UMOUNT notification for net %p%s\n",
1170                 net, NET_NAME(net));
1171         blocking_notifier_call_chain(&rpc_pipefs_notifier_list,
1172                                            RPC_PIPEFS_UMOUNT,
1173                                            sb);
1174         mutex_unlock(&sn->pipefs_sb_lock);
1175         put_net(net);
1176 out:
1177         kill_litter_super(sb);
1178 }
1179
1180 static struct file_system_type rpc_pipe_fs_type = {
1181         .owner          = THIS_MODULE,
1182         .name           = "rpc_pipefs",
1183         .mount          = rpc_mount,
1184         .kill_sb        = rpc_kill_sb,
1185 };
1186 MODULE_ALIAS_FS("rpc_pipefs");
1187 MODULE_ALIAS("rpc_pipefs");
1188
1189 static void
1190 init_once(void *foo)
1191 {
1192         struct rpc_inode *rpci = (struct rpc_inode *) foo;
1193
1194         inode_init_once(&rpci->vfs_inode);
1195         rpci->private = NULL;
1196         rpci->pipe = NULL;
1197         init_waitqueue_head(&rpci->waitq);
1198 }
1199
1200 int register_rpc_pipefs(void)
1201 {
1202         int err;
1203
1204         rpc_inode_cachep = kmem_cache_create("rpc_inode_cache",
1205                                 sizeof(struct rpc_inode),
1206                                 0, (SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT|
1207                                                 SLAB_MEM_SPREAD),
1208                                 init_once);
1209         if (!rpc_inode_cachep)
1210                 return -ENOMEM;
1211         err = rpc_clients_notifier_register();
1212         if (err)
1213                 goto err_notifier;
1214         err = register_filesystem(&rpc_pipe_fs_type);
1215         if (err)
1216                 goto err_register;
1217         return 0;
1218
1219 err_register:
1220         rpc_clients_notifier_unregister();
1221 err_notifier:
1222         kmem_cache_destroy(rpc_inode_cachep);
1223         return err;
1224 }
1225
1226 void unregister_rpc_pipefs(void)
1227 {
1228         rpc_clients_notifier_unregister();
1229         kmem_cache_destroy(rpc_inode_cachep);
1230         unregister_filesystem(&rpc_pipe_fs_type);
1231 }