SUNRPC: Fix race corrupting rpc upcall
[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 #include <linux/sunrpc/cache.h>
30 #include <linux/smp_lock.h>
31
32 static struct vfsmount *rpc_mount __read_mostly;
33 static int rpc_mount_count;
34
35 static struct file_system_type rpc_pipe_fs_type;
36
37
38 static struct kmem_cache *rpc_inode_cachep __read_mostly;
39
40 #define RPC_UPCALL_TIMEOUT (30*HZ)
41
42 static void rpc_purge_list(struct rpc_inode *rpci, struct list_head *head,
43                 void (*destroy_msg)(struct rpc_pipe_msg *), int err)
44 {
45         struct rpc_pipe_msg *msg;
46
47         if (list_empty(head))
48                 return;
49         do {
50                 msg = list_entry(head->next, struct rpc_pipe_msg, list);
51                 list_del_init(&msg->list);
52                 msg->errno = err;
53                 destroy_msg(msg);
54         } while (!list_empty(head));
55         wake_up(&rpci->waitq);
56 }
57
58 static void
59 rpc_timeout_upcall_queue(struct work_struct *work)
60 {
61         LIST_HEAD(free_list);
62         struct rpc_inode *rpci =
63                 container_of(work, struct rpc_inode, queue_timeout.work);
64         struct inode *inode = &rpci->vfs_inode;
65         void (*destroy_msg)(struct rpc_pipe_msg *);
66
67         spin_lock(&inode->i_lock);
68         if (rpci->ops == NULL) {
69                 spin_unlock(&inode->i_lock);
70                 return;
71         }
72         destroy_msg = rpci->ops->destroy_msg;
73         if (rpci->nreaders == 0) {
74                 list_splice_init(&rpci->pipe, &free_list);
75                 rpci->pipelen = 0;
76         }
77         spin_unlock(&inode->i_lock);
78         rpc_purge_list(rpci, &free_list, destroy_msg, -ETIMEDOUT);
79 }
80
81 /**
82  * rpc_queue_upcall - queue an upcall message to userspace
83  * @inode: inode of upcall pipe on which to queue given message
84  * @msg: message to queue
85  *
86  * Call with an @inode created by rpc_mkpipe() to queue an upcall.
87  * A userspace process may then later read the upcall by performing a
88  * read on an open file for this inode.  It is up to the caller to
89  * initialize the fields of @msg (other than @msg->list) appropriately.
90  */
91 int
92 rpc_queue_upcall(struct inode *inode, struct rpc_pipe_msg *msg)
93 {
94         struct rpc_inode *rpci = RPC_I(inode);
95         int res = -EPIPE;
96
97         spin_lock(&inode->i_lock);
98         if (rpci->ops == NULL)
99                 goto out;
100         if (rpci->nreaders) {
101                 list_add_tail(&msg->list, &rpci->pipe);
102                 rpci->pipelen += msg->len;
103                 res = 0;
104         } else if (rpci->flags & RPC_PIPE_WAIT_FOR_OPEN) {
105                 if (list_empty(&rpci->pipe))
106                         queue_delayed_work(rpciod_workqueue,
107                                         &rpci->queue_timeout,
108                                         RPC_UPCALL_TIMEOUT);
109                 list_add_tail(&msg->list, &rpci->pipe);
110                 rpci->pipelen += msg->len;
111                 res = 0;
112         }
113 out:
114         spin_unlock(&inode->i_lock);
115         wake_up(&rpci->waitq);
116         return res;
117 }
118 EXPORT_SYMBOL_GPL(rpc_queue_upcall);
119
120 static inline void
121 rpc_inode_setowner(struct inode *inode, void *private)
122 {
123         RPC_I(inode)->private = private;
124 }
125
126 static void
127 rpc_close_pipes(struct inode *inode)
128 {
129         struct rpc_inode *rpci = RPC_I(inode);
130         const struct rpc_pipe_ops *ops;
131         int need_release;
132
133         mutex_lock(&inode->i_mutex);
134         ops = rpci->ops;
135         if (ops != NULL) {
136                 LIST_HEAD(free_list);
137                 spin_lock(&inode->i_lock);
138                 need_release = rpci->nreaders != 0 || rpci->nwriters != 0;
139                 rpci->nreaders = 0;
140                 list_splice_init(&rpci->in_upcall, &free_list);
141                 list_splice_init(&rpci->pipe, &free_list);
142                 rpci->pipelen = 0;
143                 rpci->ops = NULL;
144                 spin_unlock(&inode->i_lock);
145                 rpc_purge_list(rpci, &free_list, ops->destroy_msg, -EPIPE);
146                 rpci->nwriters = 0;
147                 if (need_release && ops->release_pipe)
148                         ops->release_pipe(inode);
149                 cancel_delayed_work_sync(&rpci->queue_timeout);
150         }
151         rpc_inode_setowner(inode, NULL);
152         mutex_unlock(&inode->i_mutex);
153 }
154
155 static struct inode *
156 rpc_alloc_inode(struct super_block *sb)
157 {
158         struct rpc_inode *rpci;
159         rpci = (struct rpc_inode *)kmem_cache_alloc(rpc_inode_cachep, GFP_KERNEL);
160         if (!rpci)
161                 return NULL;
162         return &rpci->vfs_inode;
163 }
164
165 static void
166 rpc_destroy_inode(struct inode *inode)
167 {
168         kmem_cache_free(rpc_inode_cachep, RPC_I(inode));
169 }
170
171 static int
172 rpc_pipe_open(struct inode *inode, struct file *filp)
173 {
174         struct rpc_inode *rpci = RPC_I(inode);
175         int first_open;
176         int res = -ENXIO;
177
178         mutex_lock(&inode->i_mutex);
179         if (rpci->ops == NULL)
180                 goto out;
181         first_open = rpci->nreaders == 0 && rpci->nwriters == 0;
182         if (first_open && rpci->ops->open_pipe) {
183                 res = rpci->ops->open_pipe(inode);
184                 if (res)
185                         goto out;
186         }
187         if (filp->f_mode & FMODE_READ)
188                 rpci->nreaders++;
189         if (filp->f_mode & FMODE_WRITE)
190                 rpci->nwriters++;
191         res = 0;
192 out:
193         mutex_unlock(&inode->i_mutex);
194         return res;
195 }
196
197 static int
198 rpc_pipe_release(struct inode *inode, struct file *filp)
199 {
200         struct rpc_inode *rpci = RPC_I(inode);
201         struct rpc_pipe_msg *msg;
202         int last_close;
203
204         mutex_lock(&inode->i_mutex);
205         if (rpci->ops == NULL)
206                 goto out;
207         msg = (struct rpc_pipe_msg *)filp->private_data;
208         if (msg != NULL) {
209                 spin_lock(&inode->i_lock);
210                 msg->errno = -EAGAIN;
211                 list_del_init(&msg->list);
212                 spin_unlock(&inode->i_lock);
213                 rpci->ops->destroy_msg(msg);
214         }
215         if (filp->f_mode & FMODE_WRITE)
216                 rpci->nwriters --;
217         if (filp->f_mode & FMODE_READ) {
218                 rpci->nreaders --;
219                 if (rpci->nreaders == 0) {
220                         LIST_HEAD(free_list);
221                         spin_lock(&inode->i_lock);
222                         list_splice_init(&rpci->pipe, &free_list);
223                         rpci->pipelen = 0;
224                         spin_unlock(&inode->i_lock);
225                         rpc_purge_list(rpci, &free_list,
226                                         rpci->ops->destroy_msg, -EAGAIN);
227                 }
228         }
229         last_close = rpci->nwriters == 0 && rpci->nreaders == 0;
230         if (last_close && rpci->ops->release_pipe)
231                 rpci->ops->release_pipe(inode);
232 out:
233         mutex_unlock(&inode->i_mutex);
234         return 0;
235 }
236
237 static ssize_t
238 rpc_pipe_read(struct file *filp, char __user *buf, size_t len, loff_t *offset)
239 {
240         struct inode *inode = filp->f_path.dentry->d_inode;
241         struct rpc_inode *rpci = RPC_I(inode);
242         struct rpc_pipe_msg *msg;
243         int res = 0;
244
245         mutex_lock(&inode->i_mutex);
246         if (rpci->ops == NULL) {
247                 res = -EPIPE;
248                 goto out_unlock;
249         }
250         msg = filp->private_data;
251         if (msg == NULL) {
252                 spin_lock(&inode->i_lock);
253                 if (!list_empty(&rpci->pipe)) {
254                         msg = list_entry(rpci->pipe.next,
255                                         struct rpc_pipe_msg,
256                                         list);
257                         list_move(&msg->list, &rpci->in_upcall);
258                         rpci->pipelen -= msg->len;
259                         filp->private_data = msg;
260                         msg->copied = 0;
261                 }
262                 spin_unlock(&inode->i_lock);
263                 if (msg == NULL)
264                         goto out_unlock;
265         }
266         /* NOTE: it is up to the callback to update msg->copied */
267         res = rpci->ops->upcall(filp, msg, buf, len);
268         if (res < 0 || msg->len == msg->copied) {
269                 filp->private_data = NULL;
270                 spin_lock(&inode->i_lock);
271                 list_del_init(&msg->list);
272                 spin_unlock(&inode->i_lock);
273                 rpci->ops->destroy_msg(msg);
274         }
275 out_unlock:
276         mutex_unlock(&inode->i_mutex);
277         return res;
278 }
279
280 static ssize_t
281 rpc_pipe_write(struct file *filp, const char __user *buf, size_t len, loff_t *offset)
282 {
283         struct inode *inode = filp->f_path.dentry->d_inode;
284         struct rpc_inode *rpci = RPC_I(inode);
285         int res;
286
287         mutex_lock(&inode->i_mutex);
288         res = -EPIPE;
289         if (rpci->ops != NULL)
290                 res = rpci->ops->downcall(filp, buf, len);
291         mutex_unlock(&inode->i_mutex);
292         return res;
293 }
294
295 static unsigned int
296 rpc_pipe_poll(struct file *filp, struct poll_table_struct *wait)
297 {
298         struct rpc_inode *rpci;
299         unsigned int mask = 0;
300
301         rpci = RPC_I(filp->f_path.dentry->d_inode);
302         poll_wait(filp, &rpci->waitq, wait);
303
304         mask = POLLOUT | POLLWRNORM;
305         if (rpci->ops == NULL)
306                 mask |= POLLERR | POLLHUP;
307         if (filp->private_data || !list_empty(&rpci->pipe))
308                 mask |= POLLIN | POLLRDNORM;
309         return mask;
310 }
311
312 static int
313 rpc_pipe_ioctl_unlocked(struct file *filp, unsigned int cmd, unsigned long arg)
314 {
315         struct rpc_inode *rpci = RPC_I(filp->f_path.dentry->d_inode);
316         int len;
317
318         switch (cmd) {
319         case FIONREAD:
320                 if (rpci->ops == NULL)
321                         return -EPIPE;
322                 len = rpci->pipelen;
323                 if (filp->private_data) {
324                         struct rpc_pipe_msg *msg;
325                         msg = (struct rpc_pipe_msg *)filp->private_data;
326                         len += msg->len - msg->copied;
327                 }
328                 return put_user(len, (int __user *)arg);
329         default:
330                 return -EINVAL;
331         }
332 }
333
334 static long
335 rpc_pipe_ioctl(struct file *filp, unsigned int cmd, unsigned long arg)
336 {
337         long ret;
338
339         lock_kernel();
340         ret = rpc_pipe_ioctl_unlocked(filp, cmd, arg);
341         unlock_kernel();
342
343         return ret;
344 }
345
346 static const struct file_operations rpc_pipe_fops = {
347         .owner          = THIS_MODULE,
348         .llseek         = no_llseek,
349         .read           = rpc_pipe_read,
350         .write          = rpc_pipe_write,
351         .poll           = rpc_pipe_poll,
352         .unlocked_ioctl = rpc_pipe_ioctl,
353         .open           = rpc_pipe_open,
354         .release        = rpc_pipe_release,
355 };
356
357 static int
358 rpc_show_info(struct seq_file *m, void *v)
359 {
360         struct rpc_clnt *clnt = m->private;
361
362         seq_printf(m, "RPC server: %s\n", clnt->cl_server);
363         seq_printf(m, "service: %s (%d) version %d\n", clnt->cl_protname,
364                         clnt->cl_prog, clnt->cl_vers);
365         seq_printf(m, "address: %s\n", rpc_peeraddr2str(clnt, RPC_DISPLAY_ADDR));
366         seq_printf(m, "protocol: %s\n", rpc_peeraddr2str(clnt, RPC_DISPLAY_PROTO));
367         seq_printf(m, "port: %s\n", rpc_peeraddr2str(clnt, RPC_DISPLAY_PORT));
368         return 0;
369 }
370
371 static int
372 rpc_info_open(struct inode *inode, struct file *file)
373 {
374         struct rpc_clnt *clnt;
375         int ret = single_open(file, rpc_show_info, NULL);
376
377         if (!ret) {
378                 struct seq_file *m = file->private_data;
379                 mutex_lock(&inode->i_mutex);
380                 clnt = RPC_I(inode)->private;
381                 if (clnt) {
382                         kref_get(&clnt->cl_kref);
383                         m->private = clnt;
384                 } else {
385                         single_release(inode, file);
386                         ret = -EINVAL;
387                 }
388                 mutex_unlock(&inode->i_mutex);
389         }
390         return ret;
391 }
392
393 static int
394 rpc_info_release(struct inode *inode, struct file *file)
395 {
396         struct seq_file *m = file->private_data;
397         struct rpc_clnt *clnt = (struct rpc_clnt *)m->private;
398
399         if (clnt)
400                 rpc_release_client(clnt);
401         return single_release(inode, file);
402 }
403
404 static const struct file_operations rpc_info_operations = {
405         .owner          = THIS_MODULE,
406         .open           = rpc_info_open,
407         .read           = seq_read,
408         .llseek         = seq_lseek,
409         .release        = rpc_info_release,
410 };
411
412
413 /*
414  * Description of fs contents.
415  */
416 struct rpc_filelist {
417         const char *name;
418         const struct file_operations *i_fop;
419         umode_t mode;
420 };
421
422 struct vfsmount *rpc_get_mount(void)
423 {
424         int err;
425
426         err = simple_pin_fs(&rpc_pipe_fs_type, &rpc_mount, &rpc_mount_count);
427         if (err != 0)
428                 return ERR_PTR(err);
429         return rpc_mount;
430 }
431 EXPORT_SYMBOL_GPL(rpc_get_mount);
432
433 void rpc_put_mount(void)
434 {
435         simple_release_fs(&rpc_mount, &rpc_mount_count);
436 }
437 EXPORT_SYMBOL_GPL(rpc_put_mount);
438
439 static int rpc_delete_dentry(struct dentry *dentry)
440 {
441         return 1;
442 }
443
444 static const struct dentry_operations rpc_dentry_operations = {
445         .d_delete = rpc_delete_dentry,
446 };
447
448 static struct inode *
449 rpc_get_inode(struct super_block *sb, umode_t mode)
450 {
451         struct inode *inode = new_inode(sb);
452         if (!inode)
453                 return NULL;
454         inode->i_mode = mode;
455         inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME;
456         switch(mode & S_IFMT) {
457                 case S_IFDIR:
458                         inode->i_fop = &simple_dir_operations;
459                         inode->i_op = &simple_dir_inode_operations;
460                         inc_nlink(inode);
461                 default:
462                         break;
463         }
464         return inode;
465 }
466
467 static int __rpc_create_common(struct inode *dir, struct dentry *dentry,
468                                umode_t mode,
469                                const struct file_operations *i_fop,
470                                void *private)
471 {
472         struct inode *inode;
473
474         BUG_ON(!d_unhashed(dentry));
475         inode = rpc_get_inode(dir->i_sb, mode);
476         if (!inode)
477                 goto out_err;
478         inode->i_ino = iunique(dir->i_sb, 100);
479         if (i_fop)
480                 inode->i_fop = i_fop;
481         if (private)
482                 rpc_inode_setowner(inode, private);
483         d_add(dentry, inode);
484         return 0;
485 out_err:
486         printk(KERN_WARNING "%s: %s failed to allocate inode for dentry %s\n",
487                         __FILE__, __func__, dentry->d_name.name);
488         dput(dentry);
489         return -ENOMEM;
490 }
491
492 static int __rpc_create(struct inode *dir, struct dentry *dentry,
493                         umode_t mode,
494                         const struct file_operations *i_fop,
495                         void *private)
496 {
497         int err;
498
499         err = __rpc_create_common(dir, dentry, S_IFREG | mode, i_fop, private);
500         if (err)
501                 return err;
502         fsnotify_create(dir, dentry);
503         return 0;
504 }
505
506 static int __rpc_mkdir(struct inode *dir, struct dentry *dentry,
507                        umode_t mode,
508                        const struct file_operations *i_fop,
509                        void *private)
510 {
511         int err;
512
513         err = __rpc_create_common(dir, dentry, S_IFDIR | mode, i_fop, private);
514         if (err)
515                 return err;
516         inc_nlink(dir);
517         fsnotify_mkdir(dir, dentry);
518         return 0;
519 }
520
521 static int __rpc_mkpipe(struct inode *dir, struct dentry *dentry,
522                         umode_t mode,
523                         const struct file_operations *i_fop,
524                         void *private,
525                         const struct rpc_pipe_ops *ops,
526                         int flags)
527 {
528         struct rpc_inode *rpci;
529         int err;
530
531         err = __rpc_create_common(dir, dentry, S_IFIFO | mode, i_fop, private);
532         if (err)
533                 return err;
534         rpci = RPC_I(dentry->d_inode);
535         rpci->nkern_readwriters = 1;
536         rpci->private = private;
537         rpci->flags = flags;
538         rpci->ops = ops;
539         fsnotify_create(dir, dentry);
540         return 0;
541 }
542
543 static int __rpc_rmdir(struct inode *dir, struct dentry *dentry)
544 {
545         int ret;
546
547         dget(dentry);
548         ret = simple_rmdir(dir, dentry);
549         d_delete(dentry);
550         dput(dentry);
551         return ret;
552 }
553
554 static int __rpc_unlink(struct inode *dir, struct dentry *dentry)
555 {
556         int ret;
557
558         dget(dentry);
559         ret = simple_unlink(dir, dentry);
560         d_delete(dentry);
561         dput(dentry);
562         return ret;
563 }
564
565 static int __rpc_rmpipe(struct inode *dir, struct dentry *dentry)
566 {
567         struct inode *inode = dentry->d_inode;
568         struct rpc_inode *rpci = RPC_I(inode);
569
570         rpci->nkern_readwriters--;
571         if (rpci->nkern_readwriters != 0)
572                 return 0;
573         rpc_close_pipes(inode);
574         return __rpc_unlink(dir, dentry);
575 }
576
577 static struct dentry *__rpc_lookup_create(struct dentry *parent,
578                                           struct qstr *name)
579 {
580         struct dentry *dentry;
581
582         dentry = d_lookup(parent, name);
583         if (!dentry) {
584                 dentry = d_alloc(parent, name);
585                 if (!dentry) {
586                         dentry = ERR_PTR(-ENOMEM);
587                         goto out_err;
588                 }
589         }
590         if (!dentry->d_inode)
591                 dentry->d_op = &rpc_dentry_operations;
592 out_err:
593         return dentry;
594 }
595
596 static struct dentry *__rpc_lookup_create_exclusive(struct dentry *parent,
597                                           struct qstr *name)
598 {
599         struct dentry *dentry;
600
601         dentry = __rpc_lookup_create(parent, name);
602         if (IS_ERR(dentry))
603                 return dentry;
604         if (dentry->d_inode == NULL)
605                 return dentry;
606         dput(dentry);
607         return ERR_PTR(-EEXIST);
608 }
609
610 /*
611  * FIXME: This probably has races.
612  */
613 static void __rpc_depopulate(struct dentry *parent,
614                              const struct rpc_filelist *files,
615                              int start, int eof)
616 {
617         struct inode *dir = parent->d_inode;
618         struct dentry *dentry;
619         struct qstr name;
620         int i;
621
622         for (i = start; i < eof; i++) {
623                 name.name = files[i].name;
624                 name.len = strlen(files[i].name);
625                 name.hash = full_name_hash(name.name, name.len);
626                 dentry = d_lookup(parent, &name);
627
628                 if (dentry == NULL)
629                         continue;
630                 if (dentry->d_inode == NULL)
631                         goto next;
632                 switch (dentry->d_inode->i_mode & S_IFMT) {
633                         default:
634                                 BUG();
635                         case S_IFREG:
636                                 __rpc_unlink(dir, dentry);
637                                 break;
638                         case S_IFDIR:
639                                 __rpc_rmdir(dir, dentry);
640                 }
641 next:
642                 dput(dentry);
643         }
644 }
645
646 static void rpc_depopulate(struct dentry *parent,
647                            const struct rpc_filelist *files,
648                            int start, int eof)
649 {
650         struct inode *dir = parent->d_inode;
651
652         mutex_lock_nested(&dir->i_mutex, I_MUTEX_CHILD);
653         __rpc_depopulate(parent, files, start, eof);
654         mutex_unlock(&dir->i_mutex);
655 }
656
657 static int rpc_populate(struct dentry *parent,
658                         const struct rpc_filelist *files,
659                         int start, int eof,
660                         void *private)
661 {
662         struct inode *dir = parent->d_inode;
663         struct dentry *dentry;
664         int i, err;
665
666         mutex_lock(&dir->i_mutex);
667         for (i = start; i < eof; i++) {
668                 struct qstr q;
669
670                 q.name = files[i].name;
671                 q.len = strlen(files[i].name);
672                 q.hash = full_name_hash(q.name, q.len);
673                 dentry = __rpc_lookup_create_exclusive(parent, &q);
674                 err = PTR_ERR(dentry);
675                 if (IS_ERR(dentry))
676                         goto out_bad;
677                 switch (files[i].mode & S_IFMT) {
678                         default:
679                                 BUG();
680                         case S_IFREG:
681                                 err = __rpc_create(dir, dentry,
682                                                 files[i].mode,
683                                                 files[i].i_fop,
684                                                 private);
685                                 break;
686                         case S_IFDIR:
687                                 err = __rpc_mkdir(dir, dentry,
688                                                 files[i].mode,
689                                                 NULL,
690                                                 private);
691                 }
692                 if (err != 0)
693                         goto out_bad;
694         }
695         mutex_unlock(&dir->i_mutex);
696         return 0;
697 out_bad:
698         __rpc_depopulate(parent, files, start, eof);
699         mutex_unlock(&dir->i_mutex);
700         printk(KERN_WARNING "%s: %s failed to populate directory %s\n",
701                         __FILE__, __func__, parent->d_name.name);
702         return err;
703 }
704
705 static struct dentry *rpc_mkdir_populate(struct dentry *parent,
706                 struct qstr *name, umode_t mode, void *private,
707                 int (*populate)(struct dentry *, void *), void *args_populate)
708 {
709         struct dentry *dentry;
710         struct inode *dir = parent->d_inode;
711         int error;
712
713         mutex_lock_nested(&dir->i_mutex, I_MUTEX_PARENT);
714         dentry = __rpc_lookup_create_exclusive(parent, name);
715         if (IS_ERR(dentry))
716                 goto out;
717         error = __rpc_mkdir(dir, dentry, mode, NULL, private);
718         if (error != 0)
719                 goto out_err;
720         if (populate != NULL) {
721                 error = populate(dentry, args_populate);
722                 if (error)
723                         goto err_rmdir;
724         }
725 out:
726         mutex_unlock(&dir->i_mutex);
727         return dentry;
728 err_rmdir:
729         __rpc_rmdir(dir, dentry);
730 out_err:
731         dentry = ERR_PTR(error);
732         goto out;
733 }
734
735 static int rpc_rmdir_depopulate(struct dentry *dentry,
736                 void (*depopulate)(struct dentry *))
737 {
738         struct dentry *parent;
739         struct inode *dir;
740         int error;
741
742         parent = dget_parent(dentry);
743         dir = parent->d_inode;
744         mutex_lock_nested(&dir->i_mutex, I_MUTEX_PARENT);
745         if (depopulate != NULL)
746                 depopulate(dentry);
747         error = __rpc_rmdir(dir, dentry);
748         mutex_unlock(&dir->i_mutex);
749         dput(parent);
750         return error;
751 }
752
753 /**
754  * rpc_mkpipe - make an rpc_pipefs file for kernel<->userspace communication
755  * @parent: dentry of directory to create new "pipe" in
756  * @name: name of pipe
757  * @private: private data to associate with the pipe, for the caller's use
758  * @ops: operations defining the behavior of the pipe: upcall, downcall,
759  *      release_pipe, open_pipe, and destroy_msg.
760  * @flags: rpc_inode flags
761  *
762  * Data is made available for userspace to read by calls to
763  * rpc_queue_upcall().  The actual reads will result in calls to
764  * @ops->upcall, which will be called with the file pointer,
765  * message, and userspace buffer to copy to.
766  *
767  * Writes can come at any time, and do not necessarily have to be
768  * responses to upcalls.  They will result in calls to @msg->downcall.
769  *
770  * The @private argument passed here will be available to all these methods
771  * from the file pointer, via RPC_I(file->f_dentry->d_inode)->private.
772  */
773 struct dentry *rpc_mkpipe(struct dentry *parent, const char *name,
774                           void *private, const struct rpc_pipe_ops *ops,
775                           int flags)
776 {
777         struct dentry *dentry;
778         struct inode *dir = parent->d_inode;
779         umode_t umode = S_IFIFO | S_IRUSR | S_IWUSR;
780         struct qstr q;
781         int err;
782
783         if (ops->upcall == NULL)
784                 umode &= ~S_IRUGO;
785         if (ops->downcall == NULL)
786                 umode &= ~S_IWUGO;
787
788         q.name = name;
789         q.len = strlen(name);
790         q.hash = full_name_hash(q.name, q.len),
791
792         mutex_lock_nested(&dir->i_mutex, I_MUTEX_PARENT);
793         dentry = __rpc_lookup_create(parent, &q);
794         if (IS_ERR(dentry))
795                 goto out;
796         if (dentry->d_inode) {
797                 struct rpc_inode *rpci = RPC_I(dentry->d_inode);
798                 if (rpci->private != private ||
799                                 rpci->ops != ops ||
800                                 rpci->flags != flags) {
801                         dput (dentry);
802                         err = -EBUSY;
803                         goto out_err;
804                 }
805                 rpci->nkern_readwriters++;
806                 goto out;
807         }
808
809         err = __rpc_mkpipe(dir, dentry, umode, &rpc_pipe_fops,
810                            private, ops, flags);
811         if (err)
812                 goto out_err;
813 out:
814         mutex_unlock(&dir->i_mutex);
815         return dentry;
816 out_err:
817         dentry = ERR_PTR(err);
818         printk(KERN_WARNING "%s: %s() failed to create pipe %s/%s (errno = %d)\n",
819                         __FILE__, __func__, parent->d_name.name, name,
820                         err);
821         goto out;
822 }
823 EXPORT_SYMBOL_GPL(rpc_mkpipe);
824
825 /**
826  * rpc_unlink - remove a pipe
827  * @dentry: dentry for the pipe, as returned from rpc_mkpipe
828  *
829  * After this call, lookups will no longer find the pipe, and any
830  * attempts to read or write using preexisting opens of the pipe will
831  * return -EPIPE.
832  */
833 int
834 rpc_unlink(struct dentry *dentry)
835 {
836         struct dentry *parent;
837         struct inode *dir;
838         int error = 0;
839
840         parent = dget_parent(dentry);
841         dir = parent->d_inode;
842         mutex_lock_nested(&dir->i_mutex, I_MUTEX_PARENT);
843         error = __rpc_rmpipe(dir, dentry);
844         mutex_unlock(&dir->i_mutex);
845         dput(parent);
846         return error;
847 }
848 EXPORT_SYMBOL_GPL(rpc_unlink);
849
850 enum {
851         RPCAUTH_info,
852         RPCAUTH_EOF
853 };
854
855 static const struct rpc_filelist authfiles[] = {
856         [RPCAUTH_info] = {
857                 .name = "info",
858                 .i_fop = &rpc_info_operations,
859                 .mode = S_IFREG | S_IRUSR,
860         },
861 };
862
863 static int rpc_clntdir_populate(struct dentry *dentry, void *private)
864 {
865         return rpc_populate(dentry,
866                             authfiles, RPCAUTH_info, RPCAUTH_EOF,
867                             private);
868 }
869
870 static void rpc_clntdir_depopulate(struct dentry *dentry)
871 {
872         rpc_depopulate(dentry, authfiles, RPCAUTH_info, RPCAUTH_EOF);
873 }
874
875 /**
876  * rpc_create_client_dir - Create a new rpc_client directory in rpc_pipefs
877  * @dentry: dentry from the rpc_pipefs root to the new directory
878  * @name: &struct qstr for the name
879  * @rpc_client: rpc client to associate with this directory
880  *
881  * This creates a directory at the given @path associated with
882  * @rpc_clnt, which will contain a file named "info" with some basic
883  * information about the client, together with any "pipes" that may
884  * later be created using rpc_mkpipe().
885  */
886 struct dentry *rpc_create_client_dir(struct dentry *dentry,
887                                    struct qstr *name,
888                                    struct rpc_clnt *rpc_client)
889 {
890         return rpc_mkdir_populate(dentry, name, S_IRUGO | S_IXUGO, NULL,
891                         rpc_clntdir_populate, rpc_client);
892 }
893
894 /**
895  * rpc_remove_client_dir - Remove a directory created with rpc_create_client_dir()
896  * @dentry: directory to remove
897  */
898 int rpc_remove_client_dir(struct dentry *dentry)
899 {
900         return rpc_rmdir_depopulate(dentry, rpc_clntdir_depopulate);
901 }
902
903 static const struct rpc_filelist cache_pipefs_files[3] = {
904         [0] = {
905                 .name = "channel",
906                 .i_fop = &cache_file_operations_pipefs,
907                 .mode = S_IFREG|S_IRUSR|S_IWUSR,
908         },
909         [1] = {
910                 .name = "content",
911                 .i_fop = &content_file_operations_pipefs,
912                 .mode = S_IFREG|S_IRUSR,
913         },
914         [2] = {
915                 .name = "flush",
916                 .i_fop = &cache_flush_operations_pipefs,
917                 .mode = S_IFREG|S_IRUSR|S_IWUSR,
918         },
919 };
920
921 static int rpc_cachedir_populate(struct dentry *dentry, void *private)
922 {
923         return rpc_populate(dentry,
924                             cache_pipefs_files, 0, 3,
925                             private);
926 }
927
928 static void rpc_cachedir_depopulate(struct dentry *dentry)
929 {
930         rpc_depopulate(dentry, cache_pipefs_files, 0, 3);
931 }
932
933 struct dentry *rpc_create_cache_dir(struct dentry *parent, struct qstr *name,
934                                     mode_t umode, struct cache_detail *cd)
935 {
936         return rpc_mkdir_populate(parent, name, umode, NULL,
937                         rpc_cachedir_populate, cd);
938 }
939
940 void rpc_remove_cache_dir(struct dentry *dentry)
941 {
942         rpc_rmdir_depopulate(dentry, rpc_cachedir_depopulate);
943 }
944
945 /*
946  * populate the filesystem
947  */
948 static const struct super_operations s_ops = {
949         .alloc_inode    = rpc_alloc_inode,
950         .destroy_inode  = rpc_destroy_inode,
951         .statfs         = simple_statfs,
952 };
953
954 #define RPCAUTH_GSSMAGIC 0x67596969
955
956 /*
957  * We have a single directory with 1 node in it.
958  */
959 enum {
960         RPCAUTH_lockd,
961         RPCAUTH_mount,
962         RPCAUTH_nfs,
963         RPCAUTH_portmap,
964         RPCAUTH_statd,
965         RPCAUTH_nfsd4_cb,
966         RPCAUTH_cache,
967         RPCAUTH_RootEOF
968 };
969
970 static const struct rpc_filelist files[] = {
971         [RPCAUTH_lockd] = {
972                 .name = "lockd",
973                 .mode = S_IFDIR | S_IRUGO | S_IXUGO,
974         },
975         [RPCAUTH_mount] = {
976                 .name = "mount",
977                 .mode = S_IFDIR | S_IRUGO | S_IXUGO,
978         },
979         [RPCAUTH_nfs] = {
980                 .name = "nfs",
981                 .mode = S_IFDIR | S_IRUGO | S_IXUGO,
982         },
983         [RPCAUTH_portmap] = {
984                 .name = "portmap",
985                 .mode = S_IFDIR | S_IRUGO | S_IXUGO,
986         },
987         [RPCAUTH_statd] = {
988                 .name = "statd",
989                 .mode = S_IFDIR | S_IRUGO | S_IXUGO,
990         },
991         [RPCAUTH_nfsd4_cb] = {
992                 .name = "nfsd4_cb",
993                 .mode = S_IFDIR | S_IRUGO | S_IXUGO,
994         },
995         [RPCAUTH_cache] = {
996                 .name = "cache",
997                 .mode = S_IFDIR | S_IRUGO | S_IXUGO,
998         },
999 };
1000
1001 static int
1002 rpc_fill_super(struct super_block *sb, void *data, int silent)
1003 {
1004         struct inode *inode;
1005         struct dentry *root;
1006
1007         sb->s_blocksize = PAGE_CACHE_SIZE;
1008         sb->s_blocksize_bits = PAGE_CACHE_SHIFT;
1009         sb->s_magic = RPCAUTH_GSSMAGIC;
1010         sb->s_op = &s_ops;
1011         sb->s_time_gran = 1;
1012
1013         inode = rpc_get_inode(sb, S_IFDIR | 0755);
1014         if (!inode)
1015                 return -ENOMEM;
1016         sb->s_root = root = d_alloc_root(inode);
1017         if (!root) {
1018                 iput(inode);
1019                 return -ENOMEM;
1020         }
1021         if (rpc_populate(root, files, RPCAUTH_lockd, RPCAUTH_RootEOF, NULL))
1022                 return -ENOMEM;
1023         return 0;
1024 }
1025
1026 static int
1027 rpc_get_sb(struct file_system_type *fs_type,
1028                 int flags, const char *dev_name, void *data, struct vfsmount *mnt)
1029 {
1030         return get_sb_single(fs_type, flags, data, rpc_fill_super, mnt);
1031 }
1032
1033 static struct file_system_type rpc_pipe_fs_type = {
1034         .owner          = THIS_MODULE,
1035         .name           = "rpc_pipefs",
1036         .get_sb         = rpc_get_sb,
1037         .kill_sb        = kill_litter_super,
1038 };
1039
1040 static void
1041 init_once(void *foo)
1042 {
1043         struct rpc_inode *rpci = (struct rpc_inode *) foo;
1044
1045         inode_init_once(&rpci->vfs_inode);
1046         rpci->private = NULL;
1047         rpci->nreaders = 0;
1048         rpci->nwriters = 0;
1049         INIT_LIST_HEAD(&rpci->in_upcall);
1050         INIT_LIST_HEAD(&rpci->in_downcall);
1051         INIT_LIST_HEAD(&rpci->pipe);
1052         rpci->pipelen = 0;
1053         init_waitqueue_head(&rpci->waitq);
1054         INIT_DELAYED_WORK(&rpci->queue_timeout,
1055                             rpc_timeout_upcall_queue);
1056         rpci->ops = NULL;
1057 }
1058
1059 int register_rpc_pipefs(void)
1060 {
1061         int err;
1062
1063         rpc_inode_cachep = kmem_cache_create("rpc_inode_cache",
1064                                 sizeof(struct rpc_inode),
1065                                 0, (SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT|
1066                                                 SLAB_MEM_SPREAD),
1067                                 init_once);
1068         if (!rpc_inode_cachep)
1069                 return -ENOMEM;
1070         err = register_filesystem(&rpc_pipe_fs_type);
1071         if (err) {
1072                 kmem_cache_destroy(rpc_inode_cachep);
1073                 return err;
1074         }
1075
1076         return 0;
1077 }
1078
1079 void unregister_rpc_pipefs(void)
1080 {
1081         kmem_cache_destroy(rpc_inode_cachep);
1082         unregister_filesystem(&rpc_pipe_fs_type);
1083 }