rpc: call release_pipe only on last close
[linux-3.10.git] / 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
20 #include <asm/ioctls.h>
21 #include <linux/fs.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
30 static struct vfsmount *rpc_mount __read_mostly;
31 static int rpc_mount_count;
32
33 static struct file_system_type rpc_pipe_fs_type;
34
35
36 static struct kmem_cache *rpc_inode_cachep __read_mostly;
37
38 #define RPC_UPCALL_TIMEOUT (30*HZ)
39
40 static void rpc_purge_list(struct rpc_inode *rpci, struct list_head *head,
41                 void (*destroy_msg)(struct rpc_pipe_msg *), int err)
42 {
43         struct rpc_pipe_msg *msg;
44
45         if (list_empty(head))
46                 return;
47         do {
48                 msg = list_entry(head->next, struct rpc_pipe_msg, list);
49                 list_del(&msg->list);
50                 msg->errno = err;
51                 destroy_msg(msg);
52         } while (!list_empty(head));
53         wake_up(&rpci->waitq);
54 }
55
56 static void
57 rpc_timeout_upcall_queue(struct work_struct *work)
58 {
59         LIST_HEAD(free_list);
60         struct rpc_inode *rpci =
61                 container_of(work, struct rpc_inode, queue_timeout.work);
62         struct inode *inode = &rpci->vfs_inode;
63         void (*destroy_msg)(struct rpc_pipe_msg *);
64
65         spin_lock(&inode->i_lock);
66         if (rpci->ops == NULL) {
67                 spin_unlock(&inode->i_lock);
68                 return;
69         }
70         destroy_msg = rpci->ops->destroy_msg;
71         if (rpci->nreaders == 0) {
72                 list_splice_init(&rpci->pipe, &free_list);
73                 rpci->pipelen = 0;
74         }
75         spin_unlock(&inode->i_lock);
76         rpc_purge_list(rpci, &free_list, destroy_msg, -ETIMEDOUT);
77 }
78
79 /**
80  * rpc_queue_upcall
81  * @inode: inode of upcall pipe on which to queue given message
82  * @msg: message to queue
83  *
84  * Call with an @inode created by rpc_mkpipe() to queue an upcall.
85  * A userspace process may then later read the upcall by performing a
86  * read on an open file for this inode.  It is up to the caller to
87  * initialize the fields of @msg (other than @msg->list) appropriately.
88  */
89 int
90 rpc_queue_upcall(struct inode *inode, struct rpc_pipe_msg *msg)
91 {
92         struct rpc_inode *rpci = RPC_I(inode);
93         int res = -EPIPE;
94
95         spin_lock(&inode->i_lock);
96         if (rpci->ops == NULL)
97                 goto out;
98         if (rpci->nreaders) {
99                 list_add_tail(&msg->list, &rpci->pipe);
100                 rpci->pipelen += msg->len;
101                 res = 0;
102         } else if (rpci->flags & RPC_PIPE_WAIT_FOR_OPEN) {
103                 if (list_empty(&rpci->pipe))
104                         queue_delayed_work(rpciod_workqueue,
105                                         &rpci->queue_timeout,
106                                         RPC_UPCALL_TIMEOUT);
107                 list_add_tail(&msg->list, &rpci->pipe);
108                 rpci->pipelen += msg->len;
109                 res = 0;
110         }
111 out:
112         spin_unlock(&inode->i_lock);
113         wake_up(&rpci->waitq);
114         return res;
115 }
116 EXPORT_SYMBOL_GPL(rpc_queue_upcall);
117
118 static inline void
119 rpc_inode_setowner(struct inode *inode, void *private)
120 {
121         RPC_I(inode)->private = private;
122 }
123
124 static void
125 rpc_close_pipes(struct inode *inode)
126 {
127         struct rpc_inode *rpci = RPC_I(inode);
128         struct rpc_pipe_ops *ops;
129         int need_release;
130
131         mutex_lock(&inode->i_mutex);
132         ops = rpci->ops;
133         if (ops != NULL) {
134                 LIST_HEAD(free_list);
135                 spin_lock(&inode->i_lock);
136                 need_release = rpci->nreaders != 0 || rpci->nwriters != 0;
137                 rpci->nreaders = 0;
138                 list_splice_init(&rpci->in_upcall, &free_list);
139                 list_splice_init(&rpci->pipe, &free_list);
140                 rpci->pipelen = 0;
141                 rpci->ops = NULL;
142                 spin_unlock(&inode->i_lock);
143                 rpc_purge_list(rpci, &free_list, ops->destroy_msg, -EPIPE);
144                 rpci->nwriters = 0;
145                 if (need_release && ops->release_pipe)
146                         ops->release_pipe(inode);
147                 cancel_delayed_work_sync(&rpci->queue_timeout);
148         }
149         rpc_inode_setowner(inode, NULL);
150         mutex_unlock(&inode->i_mutex);
151 }
152
153 static struct inode *
154 rpc_alloc_inode(struct super_block *sb)
155 {
156         struct rpc_inode *rpci;
157         rpci = (struct rpc_inode *)kmem_cache_alloc(rpc_inode_cachep, GFP_KERNEL);
158         if (!rpci)
159                 return NULL;
160         return &rpci->vfs_inode;
161 }
162
163 static void
164 rpc_destroy_inode(struct inode *inode)
165 {
166         kmem_cache_free(rpc_inode_cachep, RPC_I(inode));
167 }
168
169 static int
170 rpc_pipe_open(struct inode *inode, struct file *filp)
171 {
172         struct rpc_inode *rpci = RPC_I(inode);
173         int first_open;
174         int res = -ENXIO;
175
176         mutex_lock(&inode->i_mutex);
177         if (rpci->ops == NULL)
178                 goto out;
179         first_open = rpci->nreaders == 0 && rpci->nwriters == 0;
180         if (first_open && rpci->ops->open_pipe) {
181                 res = rpci->ops->open_pipe(inode);
182                 if (res)
183                         goto out;
184         }
185         if (filp->f_mode & FMODE_READ)
186                 rpci->nreaders++;
187         if (filp->f_mode & FMODE_WRITE)
188                 rpci->nwriters++;
189         res = 0;
190 out:
191         mutex_unlock(&inode->i_mutex);
192         return res;
193 }
194
195 static int
196 rpc_pipe_release(struct inode *inode, struct file *filp)
197 {
198         struct rpc_inode *rpci = RPC_I(inode);
199         struct rpc_pipe_msg *msg;
200         int last_close;
201
202         mutex_lock(&inode->i_mutex);
203         if (rpci->ops == NULL)
204                 goto out;
205         msg = (struct rpc_pipe_msg *)filp->private_data;
206         if (msg != NULL) {
207                 spin_lock(&inode->i_lock);
208                 msg->errno = -EAGAIN;
209                 list_del(&msg->list);
210                 spin_unlock(&inode->i_lock);
211                 rpci->ops->destroy_msg(msg);
212         }
213         if (filp->f_mode & FMODE_WRITE)
214                 rpci->nwriters --;
215         if (filp->f_mode & FMODE_READ) {
216                 rpci->nreaders --;
217                 if (rpci->nreaders == 0) {
218                         LIST_HEAD(free_list);
219                         spin_lock(&inode->i_lock);
220                         list_splice_init(&rpci->pipe, &free_list);
221                         rpci->pipelen = 0;
222                         spin_unlock(&inode->i_lock);
223                         rpc_purge_list(rpci, &free_list,
224                                         rpci->ops->destroy_msg, -EAGAIN);
225                 }
226         }
227         last_close = rpci->nwriters == 0 && rpci->nreaders == 0;
228         if (last_close && rpci->ops->release_pipe)
229                 rpci->ops->release_pipe(inode);
230 out:
231         mutex_unlock(&inode->i_mutex);
232         return 0;
233 }
234
235 static ssize_t
236 rpc_pipe_read(struct file *filp, char __user *buf, size_t len, loff_t *offset)
237 {
238         struct inode *inode = filp->f_path.dentry->d_inode;
239         struct rpc_inode *rpci = RPC_I(inode);
240         struct rpc_pipe_msg *msg;
241         int res = 0;
242
243         mutex_lock(&inode->i_mutex);
244         if (rpci->ops == NULL) {
245                 res = -EPIPE;
246                 goto out_unlock;
247         }
248         msg = filp->private_data;
249         if (msg == NULL) {
250                 spin_lock(&inode->i_lock);
251                 if (!list_empty(&rpci->pipe)) {
252                         msg = list_entry(rpci->pipe.next,
253                                         struct rpc_pipe_msg,
254                                         list);
255                         list_move(&msg->list, &rpci->in_upcall);
256                         rpci->pipelen -= msg->len;
257                         filp->private_data = msg;
258                         msg->copied = 0;
259                 }
260                 spin_unlock(&inode->i_lock);
261                 if (msg == NULL)
262                         goto out_unlock;
263         }
264         /* NOTE: it is up to the callback to update msg->copied */
265         res = rpci->ops->upcall(filp, msg, buf, len);
266         if (res < 0 || msg->len == msg->copied) {
267                 filp->private_data = NULL;
268                 spin_lock(&inode->i_lock);
269                 list_del(&msg->list);
270                 spin_unlock(&inode->i_lock);
271                 rpci->ops->destroy_msg(msg);
272         }
273 out_unlock:
274         mutex_unlock(&inode->i_mutex);
275         return res;
276 }
277
278 static ssize_t
279 rpc_pipe_write(struct file *filp, const char __user *buf, size_t len, loff_t *offset)
280 {
281         struct inode *inode = filp->f_path.dentry->d_inode;
282         struct rpc_inode *rpci = RPC_I(inode);
283         int res;
284
285         mutex_lock(&inode->i_mutex);
286         res = -EPIPE;
287         if (rpci->ops != NULL)
288                 res = rpci->ops->downcall(filp, buf, len);
289         mutex_unlock(&inode->i_mutex);
290         return res;
291 }
292
293 static unsigned int
294 rpc_pipe_poll(struct file *filp, struct poll_table_struct *wait)
295 {
296         struct rpc_inode *rpci;
297         unsigned int mask = 0;
298
299         rpci = RPC_I(filp->f_path.dentry->d_inode);
300         poll_wait(filp, &rpci->waitq, wait);
301
302         mask = POLLOUT | POLLWRNORM;
303         if (rpci->ops == NULL)
304                 mask |= POLLERR | POLLHUP;
305         if (filp->private_data || !list_empty(&rpci->pipe))
306                 mask |= POLLIN | POLLRDNORM;
307         return mask;
308 }
309
310 static int
311 rpc_pipe_ioctl(struct inode *ino, struct file *filp,
312                 unsigned int cmd, unsigned long arg)
313 {
314         struct rpc_inode *rpci = RPC_I(filp->f_path.dentry->d_inode);
315         int len;
316
317         switch (cmd) {
318         case FIONREAD:
319                 if (rpci->ops == NULL)
320                         return -EPIPE;
321                 len = rpci->pipelen;
322                 if (filp->private_data) {
323                         struct rpc_pipe_msg *msg;
324                         msg = (struct rpc_pipe_msg *)filp->private_data;
325                         len += msg->len - msg->copied;
326                 }
327                 return put_user(len, (int __user *)arg);
328         default:
329                 return -EINVAL;
330         }
331 }
332
333 static const struct file_operations rpc_pipe_fops = {
334         .owner          = THIS_MODULE,
335         .llseek         = no_llseek,
336         .read           = rpc_pipe_read,
337         .write          = rpc_pipe_write,
338         .poll           = rpc_pipe_poll,
339         .ioctl          = rpc_pipe_ioctl,
340         .open           = rpc_pipe_open,
341         .release        = rpc_pipe_release,
342 };
343
344 static int
345 rpc_show_info(struct seq_file *m, void *v)
346 {
347         struct rpc_clnt *clnt = m->private;
348
349         seq_printf(m, "RPC server: %s\n", clnt->cl_server);
350         seq_printf(m, "service: %s (%d) version %d\n", clnt->cl_protname,
351                         clnt->cl_prog, clnt->cl_vers);
352         seq_printf(m, "address: %s\n", rpc_peeraddr2str(clnt, RPC_DISPLAY_ADDR));
353         seq_printf(m, "protocol: %s\n", rpc_peeraddr2str(clnt, RPC_DISPLAY_PROTO));
354         seq_printf(m, "port: %s\n", rpc_peeraddr2str(clnt, RPC_DISPLAY_PORT));
355         return 0;
356 }
357
358 static int
359 rpc_info_open(struct inode *inode, struct file *file)
360 {
361         struct rpc_clnt *clnt;
362         int ret = single_open(file, rpc_show_info, NULL);
363
364         if (!ret) {
365                 struct seq_file *m = file->private_data;
366                 mutex_lock(&inode->i_mutex);
367                 clnt = RPC_I(inode)->private;
368                 if (clnt) {
369                         kref_get(&clnt->cl_kref);
370                         m->private = clnt;
371                 } else {
372                         single_release(inode, file);
373                         ret = -EINVAL;
374                 }
375                 mutex_unlock(&inode->i_mutex);
376         }
377         return ret;
378 }
379
380 static int
381 rpc_info_release(struct inode *inode, struct file *file)
382 {
383         struct seq_file *m = file->private_data;
384         struct rpc_clnt *clnt = (struct rpc_clnt *)m->private;
385
386         if (clnt)
387                 rpc_release_client(clnt);
388         return single_release(inode, file);
389 }
390
391 static const struct file_operations rpc_info_operations = {
392         .owner          = THIS_MODULE,
393         .open           = rpc_info_open,
394         .read           = seq_read,
395         .llseek         = seq_lseek,
396         .release        = rpc_info_release,
397 };
398
399
400 /*
401  * We have a single directory with 1 node in it.
402  */
403 enum {
404         RPCAUTH_Root = 1,
405         RPCAUTH_lockd,
406         RPCAUTH_mount,
407         RPCAUTH_nfs,
408         RPCAUTH_portmap,
409         RPCAUTH_statd,
410         RPCAUTH_RootEOF
411 };
412
413 /*
414  * Description of fs contents.
415  */
416 struct rpc_filelist {
417         char *name;
418         const struct file_operations *i_fop;
419         int mode;
420 };
421
422 static struct rpc_filelist files[] = {
423         [RPCAUTH_lockd] = {
424                 .name = "lockd",
425                 .mode = S_IFDIR | S_IRUGO | S_IXUGO,
426         },
427         [RPCAUTH_mount] = {
428                 .name = "mount",
429                 .mode = S_IFDIR | S_IRUGO | S_IXUGO,
430         },
431         [RPCAUTH_nfs] = {
432                 .name = "nfs",
433                 .mode = S_IFDIR | S_IRUGO | S_IXUGO,
434         },
435         [RPCAUTH_portmap] = {
436                 .name = "portmap",
437                 .mode = S_IFDIR | S_IRUGO | S_IXUGO,
438         },
439         [RPCAUTH_statd] = {
440                 .name = "statd",
441                 .mode = S_IFDIR | S_IRUGO | S_IXUGO,
442         },
443 };
444
445 enum {
446         RPCAUTH_info = 2,
447         RPCAUTH_EOF
448 };
449
450 static struct rpc_filelist authfiles[] = {
451         [RPCAUTH_info] = {
452                 .name = "info",
453                 .i_fop = &rpc_info_operations,
454                 .mode = S_IFREG | S_IRUSR,
455         },
456 };
457
458 struct vfsmount *rpc_get_mount(void)
459 {
460         int err;
461
462         err = simple_pin_fs(&rpc_pipe_fs_type, &rpc_mount, &rpc_mount_count);
463         if (err != 0)
464                 return ERR_PTR(err);
465         return rpc_mount;
466 }
467
468 void rpc_put_mount(void)
469 {
470         simple_release_fs(&rpc_mount, &rpc_mount_count);
471 }
472
473 static int rpc_delete_dentry(struct dentry *dentry)
474 {
475         return 1;
476 }
477
478 static struct dentry_operations rpc_dentry_operations = {
479         .d_delete = rpc_delete_dentry,
480 };
481
482 static int
483 rpc_lookup_parent(char *path, struct nameidata *nd)
484 {
485         struct vfsmount *mnt;
486
487         if (path[0] == '\0')
488                 return -ENOENT;
489
490         mnt = rpc_get_mount();
491         if (IS_ERR(mnt)) {
492                 printk(KERN_WARNING "%s: %s failed to mount "
493                                "pseudofilesystem \n", __FILE__, __func__);
494                 return PTR_ERR(mnt);
495         }
496
497         if (vfs_path_lookup(mnt->mnt_root, mnt, path, LOOKUP_PARENT, nd)) {
498                 printk(KERN_WARNING "%s: %s failed to find path %s\n",
499                                 __FILE__, __func__, path);
500                 rpc_put_mount();
501                 return -ENOENT;
502         }
503         return 0;
504 }
505
506 static void
507 rpc_release_path(struct nameidata *nd)
508 {
509         path_put(&nd->path);
510         rpc_put_mount();
511 }
512
513 static struct inode *
514 rpc_get_inode(struct super_block *sb, int mode)
515 {
516         struct inode *inode = new_inode(sb);
517         if (!inode)
518                 return NULL;
519         inode->i_mode = mode;
520         inode->i_uid = inode->i_gid = 0;
521         inode->i_blocks = 0;
522         inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME;
523         switch(mode & S_IFMT) {
524                 case S_IFDIR:
525                         inode->i_fop = &simple_dir_operations;
526                         inode->i_op = &simple_dir_inode_operations;
527                         inc_nlink(inode);
528                 default:
529                         break;
530         }
531         return inode;
532 }
533
534 /*
535  * FIXME: This probably has races.
536  */
537 static void rpc_depopulate(struct dentry *parent,
538                            unsigned long start, unsigned long eof)
539 {
540         struct inode *dir = parent->d_inode;
541         struct list_head *pos, *next;
542         struct dentry *dentry, *dvec[10];
543         int n = 0;
544
545         mutex_lock_nested(&dir->i_mutex, I_MUTEX_CHILD);
546 repeat:
547         spin_lock(&dcache_lock);
548         list_for_each_safe(pos, next, &parent->d_subdirs) {
549                 dentry = list_entry(pos, struct dentry, d_u.d_child);
550                 if (!dentry->d_inode ||
551                                 dentry->d_inode->i_ino < start ||
552                                 dentry->d_inode->i_ino >= eof)
553                         continue;
554                 spin_lock(&dentry->d_lock);
555                 if (!d_unhashed(dentry)) {
556                         dget_locked(dentry);
557                         __d_drop(dentry);
558                         spin_unlock(&dentry->d_lock);
559                         dvec[n++] = dentry;
560                         if (n == ARRAY_SIZE(dvec))
561                                 break;
562                 } else
563                         spin_unlock(&dentry->d_lock);
564         }
565         spin_unlock(&dcache_lock);
566         if (n) {
567                 do {
568                         dentry = dvec[--n];
569                         if (S_ISREG(dentry->d_inode->i_mode))
570                                 simple_unlink(dir, dentry);
571                         else if (S_ISDIR(dentry->d_inode->i_mode))
572                                 simple_rmdir(dir, dentry);
573                         d_delete(dentry);
574                         dput(dentry);
575                 } while (n);
576                 goto repeat;
577         }
578         mutex_unlock(&dir->i_mutex);
579 }
580
581 static int
582 rpc_populate(struct dentry *parent,
583                 struct rpc_filelist *files,
584                 int start, int eof)
585 {
586         struct inode *inode, *dir = parent->d_inode;
587         void *private = RPC_I(dir)->private;
588         struct dentry *dentry;
589         int mode, i;
590
591         mutex_lock(&dir->i_mutex);
592         for (i = start; i < eof; i++) {
593                 dentry = d_alloc_name(parent, files[i].name);
594                 if (!dentry)
595                         goto out_bad;
596                 dentry->d_op = &rpc_dentry_operations;
597                 mode = files[i].mode;
598                 inode = rpc_get_inode(dir->i_sb, mode);
599                 if (!inode) {
600                         dput(dentry);
601                         goto out_bad;
602                 }
603                 inode->i_ino = i;
604                 if (files[i].i_fop)
605                         inode->i_fop = files[i].i_fop;
606                 if (private)
607                         rpc_inode_setowner(inode, private);
608                 if (S_ISDIR(mode))
609                         inc_nlink(dir);
610                 d_add(dentry, inode);
611                 fsnotify_create(dir, dentry);
612         }
613         mutex_unlock(&dir->i_mutex);
614         return 0;
615 out_bad:
616         mutex_unlock(&dir->i_mutex);
617         printk(KERN_WARNING "%s: %s failed to populate directory %s\n",
618                         __FILE__, __func__, parent->d_name.name);
619         return -ENOMEM;
620 }
621
622 static int
623 __rpc_mkdir(struct inode *dir, struct dentry *dentry)
624 {
625         struct inode *inode;
626
627         inode = rpc_get_inode(dir->i_sb, S_IFDIR | S_IRUGO | S_IXUGO);
628         if (!inode)
629                 goto out_err;
630         inode->i_ino = iunique(dir->i_sb, 100);
631         d_instantiate(dentry, inode);
632         inc_nlink(dir);
633         fsnotify_mkdir(dir, dentry);
634         return 0;
635 out_err:
636         printk(KERN_WARNING "%s: %s failed to allocate inode for dentry %s\n",
637                         __FILE__, __func__, dentry->d_name.name);
638         return -ENOMEM;
639 }
640
641 static int
642 __rpc_rmdir(struct inode *dir, struct dentry *dentry)
643 {
644         int error;
645         error = simple_rmdir(dir, dentry);
646         if (!error)
647                 d_delete(dentry);
648         return error;
649 }
650
651 static struct dentry *
652 rpc_lookup_create(struct dentry *parent, const char *name, int len, int exclusive)
653 {
654         struct inode *dir = parent->d_inode;
655         struct dentry *dentry;
656
657         mutex_lock_nested(&dir->i_mutex, I_MUTEX_PARENT);
658         dentry = lookup_one_len(name, parent, len);
659         if (IS_ERR(dentry))
660                 goto out_err;
661         if (!dentry->d_inode)
662                 dentry->d_op = &rpc_dentry_operations;
663         else if (exclusive) {
664                 dput(dentry);
665                 dentry = ERR_PTR(-EEXIST);
666                 goto out_err;
667         }
668         return dentry;
669 out_err:
670         mutex_unlock(&dir->i_mutex);
671         return dentry;
672 }
673
674 static struct dentry *
675 rpc_lookup_negative(char *path, struct nameidata *nd)
676 {
677         struct dentry *dentry;
678         int error;
679
680         if ((error = rpc_lookup_parent(path, nd)) != 0)
681                 return ERR_PTR(error);
682         dentry = rpc_lookup_create(nd->path.dentry, nd->last.name, nd->last.len,
683                                    1);
684         if (IS_ERR(dentry))
685                 rpc_release_path(nd);
686         return dentry;
687 }
688
689 /**
690  * rpc_mkdir - Create a new directory in rpc_pipefs
691  * @path: path from the rpc_pipefs root to the new directory
692  * @rpc_client: rpc client to associate with this directory
693  *
694  * This creates a directory at the given @path associated with
695  * @rpc_clnt, which will contain a file named "info" with some basic
696  * information about the client, together with any "pipes" that may
697  * later be created using rpc_mkpipe().
698  */
699 struct dentry *
700 rpc_mkdir(char *path, struct rpc_clnt *rpc_client)
701 {
702         struct nameidata nd;
703         struct dentry *dentry;
704         struct inode *dir;
705         int error;
706
707         dentry = rpc_lookup_negative(path, &nd);
708         if (IS_ERR(dentry))
709                 return dentry;
710         dir = nd.path.dentry->d_inode;
711         if ((error = __rpc_mkdir(dir, dentry)) != 0)
712                 goto err_dput;
713         RPC_I(dentry->d_inode)->private = rpc_client;
714         error = rpc_populate(dentry, authfiles,
715                         RPCAUTH_info, RPCAUTH_EOF);
716         if (error)
717                 goto err_depopulate;
718         dget(dentry);
719 out:
720         mutex_unlock(&dir->i_mutex);
721         rpc_release_path(&nd);
722         return dentry;
723 err_depopulate:
724         rpc_depopulate(dentry, RPCAUTH_info, RPCAUTH_EOF);
725         __rpc_rmdir(dir, dentry);
726 err_dput:
727         dput(dentry);
728         printk(KERN_WARNING "%s: %s() failed to create directory %s (errno = %d)\n",
729                         __FILE__, __func__, path, error);
730         dentry = ERR_PTR(error);
731         goto out;
732 }
733
734 /**
735  * rpc_rmdir - Remove a directory created with rpc_mkdir()
736  * @dentry: directory to remove
737  */
738 int
739 rpc_rmdir(struct dentry *dentry)
740 {
741         struct dentry *parent;
742         struct inode *dir;
743         int error;
744
745         parent = dget_parent(dentry);
746         dir = parent->d_inode;
747         mutex_lock_nested(&dir->i_mutex, I_MUTEX_PARENT);
748         rpc_depopulate(dentry, RPCAUTH_info, RPCAUTH_EOF);
749         error = __rpc_rmdir(dir, dentry);
750         dput(dentry);
751         mutex_unlock(&dir->i_mutex);
752         dput(parent);
753         return error;
754 }
755
756 /**
757  * rpc_mkpipe - make an rpc_pipefs file for kernel<->userspace communication
758  * @parent: dentry of directory to create new "pipe" in
759  * @name: name of pipe
760  * @private: private data to associate with the pipe, for the caller's use
761  * @ops: operations defining the behavior of the pipe: upcall, downcall,
762  *      release_pipe, open_pipe, and destroy_msg.
763  * @flags: rpc_inode flags
764  *
765  * Data is made available for userspace to read by calls to
766  * rpc_queue_upcall().  The actual reads will result in calls to
767  * @ops->upcall, which will be called with the file pointer,
768  * message, and userspace buffer to copy to.
769  *
770  * Writes can come at any time, and do not necessarily have to be
771  * responses to upcalls.  They will result in calls to @msg->downcall.
772  *
773  * The @private argument passed here will be available to all these methods
774  * from the file pointer, via RPC_I(file->f_dentry->d_inode)->private.
775  */
776 struct dentry *
777 rpc_mkpipe(struct dentry *parent, const char *name, void *private, struct rpc_pipe_ops *ops, int flags)
778 {
779         struct dentry *dentry;
780         struct inode *dir, *inode;
781         struct rpc_inode *rpci;
782
783         dentry = rpc_lookup_create(parent, name, strlen(name), 0);
784         if (IS_ERR(dentry))
785                 return dentry;
786         dir = parent->d_inode;
787         if (dentry->d_inode) {
788                 rpci = RPC_I(dentry->d_inode);
789                 if (rpci->private != private ||
790                                 rpci->ops != ops ||
791                                 rpci->flags != flags) {
792                         dput (dentry);
793                         dentry = ERR_PTR(-EBUSY);
794                 }
795                 rpci->nkern_readwriters++;
796                 goto out;
797         }
798         inode = rpc_get_inode(dir->i_sb, S_IFIFO | S_IRUSR | S_IWUSR);
799         if (!inode)
800                 goto err_dput;
801         inode->i_ino = iunique(dir->i_sb, 100);
802         inode->i_fop = &rpc_pipe_fops;
803         d_instantiate(dentry, inode);
804         rpci = RPC_I(inode);
805         rpci->private = private;
806         rpci->flags = flags;
807         rpci->ops = ops;
808         rpci->nkern_readwriters = 1;
809         fsnotify_create(dir, dentry);
810         dget(dentry);
811 out:
812         mutex_unlock(&dir->i_mutex);
813         return dentry;
814 err_dput:
815         dput(dentry);
816         dentry = ERR_PTR(-ENOMEM);
817         printk(KERN_WARNING "%s: %s() failed to create pipe %s/%s (errno = %d)\n",
818                         __FILE__, __func__, parent->d_name.name, name,
819                         -ENOMEM);
820         goto out;
821 }
822 EXPORT_SYMBOL_GPL(rpc_mkpipe);
823
824 /**
825  * rpc_unlink - remove a pipe
826  * @dentry: dentry for the pipe, as returned from rpc_mkpipe
827  *
828  * After this call, lookups will no longer find the pipe, and any
829  * attempts to read or write using preexisting opens of the pipe will
830  * return -EPIPE.
831  */
832 int
833 rpc_unlink(struct dentry *dentry)
834 {
835         struct dentry *parent;
836         struct inode *dir;
837         int error = 0;
838
839         parent = dget_parent(dentry);
840         dir = parent->d_inode;
841         mutex_lock_nested(&dir->i_mutex, I_MUTEX_PARENT);
842         if (--RPC_I(dentry->d_inode)->nkern_readwriters == 0) {
843                 rpc_close_pipes(dentry->d_inode);
844                 error = simple_unlink(dir, dentry);
845                 if (!error)
846                         d_delete(dentry);
847         }
848         dput(dentry);
849         mutex_unlock(&dir->i_mutex);
850         dput(parent);
851         return error;
852 }
853 EXPORT_SYMBOL_GPL(rpc_unlink);
854
855 /*
856  * populate the filesystem
857  */
858 static struct super_operations s_ops = {
859         .alloc_inode    = rpc_alloc_inode,
860         .destroy_inode  = rpc_destroy_inode,
861         .statfs         = simple_statfs,
862 };
863
864 #define RPCAUTH_GSSMAGIC 0x67596969
865
866 static int
867 rpc_fill_super(struct super_block *sb, void *data, int silent)
868 {
869         struct inode *inode;
870         struct dentry *root;
871
872         sb->s_blocksize = PAGE_CACHE_SIZE;
873         sb->s_blocksize_bits = PAGE_CACHE_SHIFT;
874         sb->s_magic = RPCAUTH_GSSMAGIC;
875         sb->s_op = &s_ops;
876         sb->s_time_gran = 1;
877
878         inode = rpc_get_inode(sb, S_IFDIR | 0755);
879         if (!inode)
880                 return -ENOMEM;
881         root = d_alloc_root(inode);
882         if (!root) {
883                 iput(inode);
884                 return -ENOMEM;
885         }
886         if (rpc_populate(root, files, RPCAUTH_Root + 1, RPCAUTH_RootEOF))
887                 goto out;
888         sb->s_root = root;
889         return 0;
890 out:
891         d_genocide(root);
892         dput(root);
893         return -ENOMEM;
894 }
895
896 static int
897 rpc_get_sb(struct file_system_type *fs_type,
898                 int flags, const char *dev_name, void *data, struct vfsmount *mnt)
899 {
900         return get_sb_single(fs_type, flags, data, rpc_fill_super, mnt);
901 }
902
903 static struct file_system_type rpc_pipe_fs_type = {
904         .owner          = THIS_MODULE,
905         .name           = "rpc_pipefs",
906         .get_sb         = rpc_get_sb,
907         .kill_sb        = kill_litter_super,
908 };
909
910 static void
911 init_once(void *foo)
912 {
913         struct rpc_inode *rpci = (struct rpc_inode *) foo;
914
915         inode_init_once(&rpci->vfs_inode);
916         rpci->private = NULL;
917         rpci->nreaders = 0;
918         rpci->nwriters = 0;
919         INIT_LIST_HEAD(&rpci->in_upcall);
920         INIT_LIST_HEAD(&rpci->in_downcall);
921         INIT_LIST_HEAD(&rpci->pipe);
922         rpci->pipelen = 0;
923         init_waitqueue_head(&rpci->waitq);
924         INIT_DELAYED_WORK(&rpci->queue_timeout,
925                             rpc_timeout_upcall_queue);
926         rpci->ops = NULL;
927 }
928
929 int register_rpc_pipefs(void)
930 {
931         int err;
932
933         rpc_inode_cachep = kmem_cache_create("rpc_inode_cache",
934                                 sizeof(struct rpc_inode),
935                                 0, (SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT|
936                                                 SLAB_MEM_SPREAD),
937                                 init_once);
938         if (!rpc_inode_cachep)
939                 return -ENOMEM;
940         err = register_filesystem(&rpc_pipe_fs_type);
941         if (err) {
942                 kmem_cache_destroy(rpc_inode_cachep);
943                 return err;
944         }
945
946         return 0;
947 }
948
949 void unregister_rpc_pipefs(void)
950 {
951         kmem_cache_destroy(rpc_inode_cachep);
952         unregister_filesystem(&rpc_pipe_fs_type);
953 }