]> Pileus Git - ~andy/linux/blob - net/sunrpc/rpc_pipe.c
SUNRPC: Use RCU to dereference the rpc_clnt.cl_xprt field
[~andy/linux] / net / sunrpc / rpc_pipe.c
1 /*
2  * net/sunrpc/rpc_pipe.c
3  *
4  * Userland/kernel interface for rpcauth_gss.
5  * Code shamelessly plagiarized from fs/nfsd/nfsctl.c
6  * and fs/sysfs/inode.c
7  *
8  * Copyright (c) 2002, Trond Myklebust <trond.myklebust@fys.uio.no>
9  *
10  */
11 #include <linux/module.h>
12 #include <linux/slab.h>
13 #include <linux/string.h>
14 #include <linux/pagemap.h>
15 #include <linux/mount.h>
16 #include <linux/namei.h>
17 #include <linux/fsnotify.h>
18 #include <linux/kernel.h>
19 #include <linux/rcupdate.h>
20
21 #include <asm/ioctls.h>
22 #include <linux/poll.h>
23 #include <linux/wait.h>
24 #include <linux/seq_file.h>
25
26 #include <linux/sunrpc/clnt.h>
27 #include <linux/workqueue.h>
28 #include <linux/sunrpc/rpc_pipe_fs.h>
29 #include <linux/sunrpc/cache.h>
30 #include <linux/nsproxy.h>
31 #include <linux/notifier.h>
32
33 #include "netns.h"
34 #include "sunrpc.h"
35
36 #define RPCDBG_FACILITY RPCDBG_DEBUG
37
38 #define NET_NAME(net)   ((net == &init_net) ? " (init_net)" : "")
39
40 static struct file_system_type rpc_pipe_fs_type;
41
42
43 static struct kmem_cache *rpc_inode_cachep __read_mostly;
44
45 #define RPC_UPCALL_TIMEOUT (30*HZ)
46
47 static BLOCKING_NOTIFIER_HEAD(rpc_pipefs_notifier_list);
48
49 int rpc_pipefs_notifier_register(struct notifier_block *nb)
50 {
51         return blocking_notifier_chain_cond_register(&rpc_pipefs_notifier_list, nb);
52 }
53 EXPORT_SYMBOL_GPL(rpc_pipefs_notifier_register);
54
55 void rpc_pipefs_notifier_unregister(struct notifier_block *nb)
56 {
57         blocking_notifier_chain_unregister(&rpc_pipefs_notifier_list, nb);
58 }
59 EXPORT_SYMBOL_GPL(rpc_pipefs_notifier_unregister);
60
61 static void rpc_purge_list(wait_queue_head_t *waitq, struct list_head *head,
62                 void (*destroy_msg)(struct rpc_pipe_msg *), int err)
63 {
64         struct rpc_pipe_msg *msg;
65
66         if (list_empty(head))
67                 return;
68         do {
69                 msg = list_entry(head->next, struct rpc_pipe_msg, list);
70                 list_del_init(&msg->list);
71                 msg->errno = err;
72                 destroy_msg(msg);
73         } while (!list_empty(head));
74         wake_up(waitq);
75 }
76
77 static void
78 rpc_timeout_upcall_queue(struct work_struct *work)
79 {
80         LIST_HEAD(free_list);
81         struct rpc_pipe *pipe =
82                 container_of(work, struct rpc_pipe, queue_timeout.work);
83         void (*destroy_msg)(struct rpc_pipe_msg *);
84         struct dentry *dentry;
85
86         spin_lock(&pipe->lock);
87         destroy_msg = pipe->ops->destroy_msg;
88         if (pipe->nreaders == 0) {
89                 list_splice_init(&pipe->pipe, &free_list);
90                 pipe->pipelen = 0;
91         }
92         dentry = dget(pipe->dentry);
93         spin_unlock(&pipe->lock);
94         if (dentry) {
95                 rpc_purge_list(&RPC_I(dentry->d_inode)->waitq,
96                                &free_list, destroy_msg, -ETIMEDOUT);
97                 dput(dentry);
98         }
99 }
100
101 ssize_t rpc_pipe_generic_upcall(struct file *filp, struct rpc_pipe_msg *msg,
102                                 char __user *dst, size_t buflen)
103 {
104         char *data = (char *)msg->data + msg->copied;
105         size_t mlen = min(msg->len - msg->copied, buflen);
106         unsigned long left;
107
108         left = copy_to_user(dst, data, mlen);
109         if (left == mlen) {
110                 msg->errno = -EFAULT;
111                 return -EFAULT;
112         }
113
114         mlen -= left;
115         msg->copied += mlen;
116         msg->errno = 0;
117         return mlen;
118 }
119 EXPORT_SYMBOL_GPL(rpc_pipe_generic_upcall);
120
121 /**
122  * rpc_queue_upcall - queue an upcall message to userspace
123  * @inode: inode of upcall pipe on which to queue given message
124  * @msg: message to queue
125  *
126  * Call with an @inode created by rpc_mkpipe() to queue an upcall.
127  * A userspace process may then later read the upcall by performing a
128  * read on an open file for this inode.  It is up to the caller to
129  * initialize the fields of @msg (other than @msg->list) appropriately.
130  */
131 int
132 rpc_queue_upcall(struct rpc_pipe *pipe, struct rpc_pipe_msg *msg)
133 {
134         int res = -EPIPE;
135         struct dentry *dentry;
136
137         spin_lock(&pipe->lock);
138         if (pipe->nreaders) {
139                 list_add_tail(&msg->list, &pipe->pipe);
140                 pipe->pipelen += msg->len;
141                 res = 0;
142         } else if (pipe->flags & RPC_PIPE_WAIT_FOR_OPEN) {
143                 if (list_empty(&pipe->pipe))
144                         queue_delayed_work(rpciod_workqueue,
145                                         &pipe->queue_timeout,
146                                         RPC_UPCALL_TIMEOUT);
147                 list_add_tail(&msg->list, &pipe->pipe);
148                 pipe->pipelen += msg->len;
149                 res = 0;
150         }
151         dentry = dget(pipe->dentry);
152         spin_unlock(&pipe->lock);
153         if (dentry) {
154                 wake_up(&RPC_I(dentry->d_inode)->waitq);
155                 dput(dentry);
156         }
157         return res;
158 }
159 EXPORT_SYMBOL_GPL(rpc_queue_upcall);
160
161 static inline void
162 rpc_inode_setowner(struct inode *inode, void *private)
163 {
164         RPC_I(inode)->private = private;
165 }
166
167 static void
168 rpc_close_pipes(struct inode *inode)
169 {
170         struct rpc_pipe *pipe = RPC_I(inode)->pipe;
171         int need_release;
172         LIST_HEAD(free_list);
173
174         mutex_lock(&inode->i_mutex);
175         spin_lock(&pipe->lock);
176         need_release = pipe->nreaders != 0 || pipe->nwriters != 0;
177         pipe->nreaders = 0;
178         list_splice_init(&pipe->in_upcall, &free_list);
179         list_splice_init(&pipe->pipe, &free_list);
180         pipe->pipelen = 0;
181         pipe->dentry = NULL;
182         spin_unlock(&pipe->lock);
183         rpc_purge_list(&RPC_I(inode)->waitq, &free_list, pipe->ops->destroy_msg, -EPIPE);
184         pipe->nwriters = 0;
185         if (need_release && pipe->ops->release_pipe)
186                 pipe->ops->release_pipe(inode);
187         cancel_delayed_work_sync(&pipe->queue_timeout);
188         rpc_inode_setowner(inode, NULL);
189         RPC_I(inode)->pipe = NULL;
190         mutex_unlock(&inode->i_mutex);
191 }
192
193 static struct inode *
194 rpc_alloc_inode(struct super_block *sb)
195 {
196         struct rpc_inode *rpci;
197         rpci = (struct rpc_inode *)kmem_cache_alloc(rpc_inode_cachep, GFP_KERNEL);
198         if (!rpci)
199                 return NULL;
200         return &rpci->vfs_inode;
201 }
202
203 static void
204 rpc_i_callback(struct rcu_head *head)
205 {
206         struct inode *inode = container_of(head, struct inode, i_rcu);
207         kmem_cache_free(rpc_inode_cachep, RPC_I(inode));
208 }
209
210 static void
211 rpc_destroy_inode(struct inode *inode)
212 {
213         call_rcu(&inode->i_rcu, rpc_i_callback);
214 }
215
216 static int
217 rpc_pipe_open(struct inode *inode, struct file *filp)
218 {
219         struct rpc_pipe *pipe;
220         int first_open;
221         int res = -ENXIO;
222
223         mutex_lock(&inode->i_mutex);
224         pipe = RPC_I(inode)->pipe;
225         if (pipe == NULL)
226                 goto out;
227         first_open = pipe->nreaders == 0 && pipe->nwriters == 0;
228         if (first_open && pipe->ops->open_pipe) {
229                 res = pipe->ops->open_pipe(inode);
230                 if (res)
231                         goto out;
232         }
233         if (filp->f_mode & FMODE_READ)
234                 pipe->nreaders++;
235         if (filp->f_mode & FMODE_WRITE)
236                 pipe->nwriters++;
237         res = 0;
238 out:
239         mutex_unlock(&inode->i_mutex);
240         return res;
241 }
242
243 static int
244 rpc_pipe_release(struct inode *inode, struct file *filp)
245 {
246         struct rpc_pipe *pipe;
247         struct rpc_pipe_msg *msg;
248         int last_close;
249
250         mutex_lock(&inode->i_mutex);
251         pipe = RPC_I(inode)->pipe;
252         if (pipe == NULL)
253                 goto out;
254         msg = filp->private_data;
255         if (msg != NULL) {
256                 spin_lock(&pipe->lock);
257                 msg->errno = -EAGAIN;
258                 list_del_init(&msg->list);
259                 spin_unlock(&pipe->lock);
260                 pipe->ops->destroy_msg(msg);
261         }
262         if (filp->f_mode & FMODE_WRITE)
263                 pipe->nwriters --;
264         if (filp->f_mode & FMODE_READ) {
265                 pipe->nreaders --;
266                 if (pipe->nreaders == 0) {
267                         LIST_HEAD(free_list);
268                         spin_lock(&pipe->lock);
269                         list_splice_init(&pipe->pipe, &free_list);
270                         pipe->pipelen = 0;
271                         spin_unlock(&pipe->lock);
272                         rpc_purge_list(&RPC_I(inode)->waitq, &free_list,
273                                         pipe->ops->destroy_msg, -EAGAIN);
274                 }
275         }
276         last_close = pipe->nwriters == 0 && pipe->nreaders == 0;
277         if (last_close && pipe->ops->release_pipe)
278                 pipe->ops->release_pipe(inode);
279 out:
280         mutex_unlock(&inode->i_mutex);
281         return 0;
282 }
283
284 static ssize_t
285 rpc_pipe_read(struct file *filp, char __user *buf, size_t len, loff_t *offset)
286 {
287         struct inode *inode = filp->f_path.dentry->d_inode;
288         struct rpc_pipe *pipe;
289         struct rpc_pipe_msg *msg;
290         int res = 0;
291
292         mutex_lock(&inode->i_mutex);
293         pipe = RPC_I(inode)->pipe;
294         if (pipe == NULL) {
295                 res = -EPIPE;
296                 goto out_unlock;
297         }
298         msg = filp->private_data;
299         if (msg == NULL) {
300                 spin_lock(&pipe->lock);
301                 if (!list_empty(&pipe->pipe)) {
302                         msg = list_entry(pipe->pipe.next,
303                                         struct rpc_pipe_msg,
304                                         list);
305                         list_move(&msg->list, &pipe->in_upcall);
306                         pipe->pipelen -= msg->len;
307                         filp->private_data = msg;
308                         msg->copied = 0;
309                 }
310                 spin_unlock(&pipe->lock);
311                 if (msg == NULL)
312                         goto out_unlock;
313         }
314         /* NOTE: it is up to the callback to update msg->copied */
315         res = pipe->ops->upcall(filp, msg, buf, len);
316         if (res < 0 || msg->len == msg->copied) {
317                 filp->private_data = NULL;
318                 spin_lock(&pipe->lock);
319                 list_del_init(&msg->list);
320                 spin_unlock(&pipe->lock);
321                 pipe->ops->destroy_msg(msg);
322         }
323 out_unlock:
324         mutex_unlock(&inode->i_mutex);
325         return res;
326 }
327
328 static ssize_t
329 rpc_pipe_write(struct file *filp, const char __user *buf, size_t len, loff_t *offset)
330 {
331         struct inode *inode = filp->f_path.dentry->d_inode;
332         int res;
333
334         mutex_lock(&inode->i_mutex);
335         res = -EPIPE;
336         if (RPC_I(inode)->pipe != NULL)
337                 res = RPC_I(inode)->pipe->ops->downcall(filp, buf, len);
338         mutex_unlock(&inode->i_mutex);
339         return res;
340 }
341
342 static unsigned int
343 rpc_pipe_poll(struct file *filp, struct poll_table_struct *wait)
344 {
345         struct inode *inode = filp->f_path.dentry->d_inode;
346         struct rpc_inode *rpci = RPC_I(inode);
347         unsigned int mask = POLLOUT | POLLWRNORM;
348
349         poll_wait(filp, &rpci->waitq, wait);
350
351         mutex_lock(&inode->i_mutex);
352         if (rpci->pipe == NULL)
353                 mask |= POLLERR | POLLHUP;
354         else if (filp->private_data || !list_empty(&rpci->pipe->pipe))
355                 mask |= POLLIN | POLLRDNORM;
356         mutex_unlock(&inode->i_mutex);
357         return mask;
358 }
359
360 static long
361 rpc_pipe_ioctl(struct file *filp, unsigned int cmd, unsigned long arg)
362 {
363         struct inode *inode = filp->f_path.dentry->d_inode;
364         struct rpc_pipe *pipe;
365         int len;
366
367         switch (cmd) {
368         case FIONREAD:
369                 mutex_lock(&inode->i_mutex);
370                 pipe = RPC_I(inode)->pipe;
371                 if (pipe == NULL) {
372                         mutex_unlock(&inode->i_mutex);
373                         return -EPIPE;
374                 }
375                 spin_lock(&pipe->lock);
376                 len = pipe->pipelen;
377                 if (filp->private_data) {
378                         struct rpc_pipe_msg *msg;
379                         msg = filp->private_data;
380                         len += msg->len - msg->copied;
381                 }
382                 spin_unlock(&pipe->lock);
383                 mutex_unlock(&inode->i_mutex);
384                 return put_user(len, (int __user *)arg);
385         default:
386                 return -EINVAL;
387         }
388 }
389
390 static const struct file_operations rpc_pipe_fops = {
391         .owner          = THIS_MODULE,
392         .llseek         = no_llseek,
393         .read           = rpc_pipe_read,
394         .write          = rpc_pipe_write,
395         .poll           = rpc_pipe_poll,
396         .unlocked_ioctl = rpc_pipe_ioctl,
397         .open           = rpc_pipe_open,
398         .release        = rpc_pipe_release,
399 };
400
401 static int
402 rpc_show_info(struct seq_file *m, void *v)
403 {
404         struct rpc_clnt *clnt = m->private;
405
406         rcu_read_lock();
407         seq_printf(m, "RPC server: %s\n", clnt->cl_server);
408         seq_printf(m, "service: %s (%d) version %d\n", clnt->cl_protname,
409                         clnt->cl_prog, clnt->cl_vers);
410         seq_printf(m, "address: %s\n", rpc_peeraddr2str(clnt, RPC_DISPLAY_ADDR));
411         seq_printf(m, "protocol: %s\n", rpc_peeraddr2str(clnt, RPC_DISPLAY_PROTO));
412         seq_printf(m, "port: %s\n", rpc_peeraddr2str(clnt, RPC_DISPLAY_PORT));
413         rcu_read_unlock();
414         return 0;
415 }
416
417 static int
418 rpc_info_open(struct inode *inode, struct file *file)
419 {
420         struct rpc_clnt *clnt = NULL;
421         int ret = single_open(file, rpc_show_info, NULL);
422
423         if (!ret) {
424                 struct seq_file *m = file->private_data;
425
426                 spin_lock(&file->f_path.dentry->d_lock);
427                 if (!d_unhashed(file->f_path.dentry))
428                         clnt = RPC_I(inode)->private;
429                 if (clnt != NULL && atomic_inc_not_zero(&clnt->cl_count)) {
430                         spin_unlock(&file->f_path.dentry->d_lock);
431                         m->private = clnt;
432                 } else {
433                         spin_unlock(&file->f_path.dentry->d_lock);
434                         single_release(inode, file);
435                         ret = -EINVAL;
436                 }
437         }
438         return ret;
439 }
440
441 static int
442 rpc_info_release(struct inode *inode, struct file *file)
443 {
444         struct seq_file *m = file->private_data;
445         struct rpc_clnt *clnt = (struct rpc_clnt *)m->private;
446
447         if (clnt)
448                 rpc_release_client(clnt);
449         return single_release(inode, file);
450 }
451
452 static const struct file_operations rpc_info_operations = {
453         .owner          = THIS_MODULE,
454         .open           = rpc_info_open,
455         .read           = seq_read,
456         .llseek         = seq_lseek,
457         .release        = rpc_info_release,
458 };
459
460
461 /*
462  * Description of fs contents.
463  */
464 struct rpc_filelist {
465         const char *name;
466         const struct file_operations *i_fop;
467         umode_t mode;
468 };
469
470 static int rpc_delete_dentry(const struct dentry *dentry)
471 {
472         return 1;
473 }
474
475 static const struct dentry_operations rpc_dentry_operations = {
476         .d_delete = rpc_delete_dentry,
477 };
478
479 static struct inode *
480 rpc_get_inode(struct super_block *sb, umode_t mode)
481 {
482         struct inode *inode = new_inode(sb);
483         if (!inode)
484                 return NULL;
485         inode->i_ino = get_next_ino();
486         inode->i_mode = mode;
487         inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME;
488         switch (mode & S_IFMT) {
489         case S_IFDIR:
490                 inode->i_fop = &simple_dir_operations;
491                 inode->i_op = &simple_dir_inode_operations;
492                 inc_nlink(inode);
493         default:
494                 break;
495         }
496         return inode;
497 }
498
499 static int __rpc_create_common(struct inode *dir, struct dentry *dentry,
500                                umode_t mode,
501                                const struct file_operations *i_fop,
502                                void *private)
503 {
504         struct inode *inode;
505
506         d_drop(dentry);
507         inode = rpc_get_inode(dir->i_sb, mode);
508         if (!inode)
509                 goto out_err;
510         inode->i_ino = iunique(dir->i_sb, 100);
511         if (i_fop)
512                 inode->i_fop = i_fop;
513         if (private)
514                 rpc_inode_setowner(inode, private);
515         d_add(dentry, inode);
516         return 0;
517 out_err:
518         printk(KERN_WARNING "%s: %s failed to allocate inode for dentry %s\n",
519                         __FILE__, __func__, dentry->d_name.name);
520         dput(dentry);
521         return -ENOMEM;
522 }
523
524 static int __rpc_create(struct inode *dir, struct dentry *dentry,
525                         umode_t mode,
526                         const struct file_operations *i_fop,
527                         void *private)
528 {
529         int err;
530
531         err = __rpc_create_common(dir, dentry, S_IFREG | mode, i_fop, private);
532         if (err)
533                 return err;
534         fsnotify_create(dir, dentry);
535         return 0;
536 }
537
538 static int __rpc_mkdir(struct inode *dir, struct dentry *dentry,
539                        umode_t mode,
540                        const struct file_operations *i_fop,
541                        void *private)
542 {
543         int err;
544
545         err = __rpc_create_common(dir, dentry, S_IFDIR | mode, i_fop, private);
546         if (err)
547                 return err;
548         inc_nlink(dir);
549         fsnotify_mkdir(dir, dentry);
550         return 0;
551 }
552
553 static void
554 init_pipe(struct rpc_pipe *pipe)
555 {
556         pipe->nreaders = 0;
557         pipe->nwriters = 0;
558         INIT_LIST_HEAD(&pipe->in_upcall);
559         INIT_LIST_HEAD(&pipe->in_downcall);
560         INIT_LIST_HEAD(&pipe->pipe);
561         pipe->pipelen = 0;
562         INIT_DELAYED_WORK(&pipe->queue_timeout,
563                             rpc_timeout_upcall_queue);
564         pipe->ops = NULL;
565         spin_lock_init(&pipe->lock);
566         pipe->dentry = NULL;
567 }
568
569 void rpc_destroy_pipe_data(struct rpc_pipe *pipe)
570 {
571         kfree(pipe);
572 }
573 EXPORT_SYMBOL_GPL(rpc_destroy_pipe_data);
574
575 struct rpc_pipe *rpc_mkpipe_data(const struct rpc_pipe_ops *ops, int flags)
576 {
577         struct rpc_pipe *pipe;
578
579         pipe = kzalloc(sizeof(struct rpc_pipe), GFP_KERNEL);
580         if (!pipe)
581                 return ERR_PTR(-ENOMEM);
582         init_pipe(pipe);
583         pipe->ops = ops;
584         pipe->flags = flags;
585         return pipe;
586 }
587 EXPORT_SYMBOL_GPL(rpc_mkpipe_data);
588
589 static int __rpc_mkpipe_dentry(struct inode *dir, struct dentry *dentry,
590                                umode_t mode,
591                                const struct file_operations *i_fop,
592                                void *private,
593                                struct rpc_pipe *pipe)
594 {
595         struct rpc_inode *rpci;
596         int err;
597
598         err = __rpc_create_common(dir, dentry, S_IFIFO | mode, i_fop, private);
599         if (err)
600                 return err;
601         rpci = RPC_I(dentry->d_inode);
602         rpci->private = private;
603         rpci->pipe = pipe;
604         fsnotify_create(dir, dentry);
605         return 0;
606 }
607
608 static int __rpc_rmdir(struct inode *dir, struct dentry *dentry)
609 {
610         int ret;
611
612         dget(dentry);
613         ret = simple_rmdir(dir, dentry);
614         d_delete(dentry);
615         dput(dentry);
616         return ret;
617 }
618
619 int rpc_rmdir(struct dentry *dentry)
620 {
621         struct dentry *parent;
622         struct inode *dir;
623         int error;
624
625         parent = dget_parent(dentry);
626         dir = parent->d_inode;
627         mutex_lock_nested(&dir->i_mutex, I_MUTEX_PARENT);
628         error = __rpc_rmdir(dir, dentry);
629         mutex_unlock(&dir->i_mutex);
630         dput(parent);
631         return error;
632 }
633 EXPORT_SYMBOL_GPL(rpc_rmdir);
634
635 static int __rpc_unlink(struct inode *dir, struct dentry *dentry)
636 {
637         int ret;
638
639         dget(dentry);
640         ret = simple_unlink(dir, dentry);
641         d_delete(dentry);
642         dput(dentry);
643         return ret;
644 }
645
646 static int __rpc_rmpipe(struct inode *dir, struct dentry *dentry)
647 {
648         struct inode *inode = dentry->d_inode;
649
650         rpc_close_pipes(inode);
651         return __rpc_unlink(dir, dentry);
652 }
653
654 static struct dentry *__rpc_lookup_create_exclusive(struct dentry *parent,
655                                           struct qstr *name)
656 {
657         struct dentry *dentry;
658
659         dentry = d_lookup(parent, name);
660         if (!dentry) {
661                 dentry = d_alloc(parent, name);
662                 if (!dentry)
663                         return ERR_PTR(-ENOMEM);
664         }
665         if (dentry->d_inode == NULL) {
666                 d_set_d_op(dentry, &rpc_dentry_operations);
667                 return dentry;
668         }
669         dput(dentry);
670         return ERR_PTR(-EEXIST);
671 }
672
673 /*
674  * FIXME: This probably has races.
675  */
676 static void __rpc_depopulate(struct dentry *parent,
677                              const struct rpc_filelist *files,
678                              int start, int eof)
679 {
680         struct inode *dir = parent->d_inode;
681         struct dentry *dentry;
682         struct qstr name;
683         int i;
684
685         for (i = start; i < eof; i++) {
686                 name.name = files[i].name;
687                 name.len = strlen(files[i].name);
688                 name.hash = full_name_hash(name.name, name.len);
689                 dentry = d_lookup(parent, &name);
690
691                 if (dentry == NULL)
692                         continue;
693                 if (dentry->d_inode == NULL)
694                         goto next;
695                 switch (dentry->d_inode->i_mode & S_IFMT) {
696                         default:
697                                 BUG();
698                         case S_IFREG:
699                                 __rpc_unlink(dir, dentry);
700                                 break;
701                         case S_IFDIR:
702                                 __rpc_rmdir(dir, dentry);
703                 }
704 next:
705                 dput(dentry);
706         }
707 }
708
709 static void rpc_depopulate(struct dentry *parent,
710                            const struct rpc_filelist *files,
711                            int start, int eof)
712 {
713         struct inode *dir = parent->d_inode;
714
715         mutex_lock_nested(&dir->i_mutex, I_MUTEX_CHILD);
716         __rpc_depopulate(parent, files, start, eof);
717         mutex_unlock(&dir->i_mutex);
718 }
719
720 static int rpc_populate(struct dentry *parent,
721                         const struct rpc_filelist *files,
722                         int start, int eof,
723                         void *private)
724 {
725         struct inode *dir = parent->d_inode;
726         struct dentry *dentry;
727         int i, err;
728
729         mutex_lock(&dir->i_mutex);
730         for (i = start; i < eof; i++) {
731                 struct qstr q;
732
733                 q.name = files[i].name;
734                 q.len = strlen(files[i].name);
735                 q.hash = full_name_hash(q.name, q.len);
736                 dentry = __rpc_lookup_create_exclusive(parent, &q);
737                 err = PTR_ERR(dentry);
738                 if (IS_ERR(dentry))
739                         goto out_bad;
740                 switch (files[i].mode & S_IFMT) {
741                         default:
742                                 BUG();
743                         case S_IFREG:
744                                 err = __rpc_create(dir, dentry,
745                                                 files[i].mode,
746                                                 files[i].i_fop,
747                                                 private);
748                                 break;
749                         case S_IFDIR:
750                                 err = __rpc_mkdir(dir, dentry,
751                                                 files[i].mode,
752                                                 NULL,
753                                                 private);
754                 }
755                 if (err != 0)
756                         goto out_bad;
757         }
758         mutex_unlock(&dir->i_mutex);
759         return 0;
760 out_bad:
761         __rpc_depopulate(parent, files, start, eof);
762         mutex_unlock(&dir->i_mutex);
763         printk(KERN_WARNING "%s: %s failed to populate directory %s\n",
764                         __FILE__, __func__, parent->d_name.name);
765         return err;
766 }
767
768 static struct dentry *rpc_mkdir_populate(struct dentry *parent,
769                 struct qstr *name, umode_t mode, void *private,
770                 int (*populate)(struct dentry *, void *), void *args_populate)
771 {
772         struct dentry *dentry;
773         struct inode *dir = parent->d_inode;
774         int error;
775
776         mutex_lock_nested(&dir->i_mutex, I_MUTEX_PARENT);
777         dentry = __rpc_lookup_create_exclusive(parent, name);
778         if (IS_ERR(dentry))
779                 goto out;
780         error = __rpc_mkdir(dir, dentry, mode, NULL, private);
781         if (error != 0)
782                 goto out_err;
783         if (populate != NULL) {
784                 error = populate(dentry, args_populate);
785                 if (error)
786                         goto err_rmdir;
787         }
788 out:
789         mutex_unlock(&dir->i_mutex);
790         return dentry;
791 err_rmdir:
792         __rpc_rmdir(dir, dentry);
793 out_err:
794         dentry = ERR_PTR(error);
795         goto out;
796 }
797
798 static int rpc_rmdir_depopulate(struct dentry *dentry,
799                 void (*depopulate)(struct dentry *))
800 {
801         struct dentry *parent;
802         struct inode *dir;
803         int error;
804
805         parent = dget_parent(dentry);
806         dir = parent->d_inode;
807         mutex_lock_nested(&dir->i_mutex, I_MUTEX_PARENT);
808         if (depopulate != NULL)
809                 depopulate(dentry);
810         error = __rpc_rmdir(dir, dentry);
811         mutex_unlock(&dir->i_mutex);
812         dput(parent);
813         return error;
814 }
815
816 /**
817  * rpc_mkpipe - make an rpc_pipefs file for kernel<->userspace communication
818  * @parent: dentry of directory to create new "pipe" in
819  * @name: name of pipe
820  * @private: private data to associate with the pipe, for the caller's use
821  * @ops: operations defining the behavior of the pipe: upcall, downcall,
822  *      release_pipe, open_pipe, and destroy_msg.
823  * @flags: rpc_pipe flags
824  *
825  * Data is made available for userspace to read by calls to
826  * rpc_queue_upcall().  The actual reads will result in calls to
827  * @ops->upcall, which will be called with the file pointer,
828  * message, and userspace buffer to copy to.
829  *
830  * Writes can come at any time, and do not necessarily have to be
831  * responses to upcalls.  They will result in calls to @msg->downcall.
832  *
833  * The @private argument passed here will be available to all these methods
834  * from the file pointer, via RPC_I(file->f_dentry->d_inode)->private.
835  */
836 struct dentry *rpc_mkpipe_dentry(struct dentry *parent, const char *name,
837                                  void *private, struct rpc_pipe *pipe)
838 {
839         struct dentry *dentry;
840         struct inode *dir = parent->d_inode;
841         umode_t umode = S_IFIFO | S_IRUSR | S_IWUSR;
842         struct qstr q;
843         int err;
844
845         if (pipe->ops->upcall == NULL)
846                 umode &= ~S_IRUGO;
847         if (pipe->ops->downcall == NULL)
848                 umode &= ~S_IWUGO;
849
850         q.name = name;
851         q.len = strlen(name);
852         q.hash = full_name_hash(q.name, q.len),
853
854         mutex_lock_nested(&dir->i_mutex, I_MUTEX_PARENT);
855         dentry = __rpc_lookup_create_exclusive(parent, &q);
856         if (IS_ERR(dentry))
857                 goto out;
858         err = __rpc_mkpipe_dentry(dir, dentry, umode, &rpc_pipe_fops,
859                                   private, pipe);
860         if (err)
861                 goto out_err;
862 out:
863         mutex_unlock(&dir->i_mutex);
864         return dentry;
865 out_err:
866         dentry = ERR_PTR(err);
867         printk(KERN_WARNING "%s: %s() failed to create pipe %s/%s (errno = %d)\n",
868                         __FILE__, __func__, parent->d_name.name, name,
869                         err);
870         goto out;
871 }
872 EXPORT_SYMBOL_GPL(rpc_mkpipe_dentry);
873
874 /**
875  * rpc_unlink - remove a pipe
876  * @dentry: dentry for the pipe, as returned from rpc_mkpipe
877  *
878  * After this call, lookups will no longer find the pipe, and any
879  * attempts to read or write using preexisting opens of the pipe will
880  * return -EPIPE.
881  */
882 int
883 rpc_unlink(struct dentry *dentry)
884 {
885         struct dentry *parent;
886         struct inode *dir;
887         int error = 0;
888
889         parent = dget_parent(dentry);
890         dir = parent->d_inode;
891         mutex_lock_nested(&dir->i_mutex, I_MUTEX_PARENT);
892         error = __rpc_rmpipe(dir, dentry);
893         mutex_unlock(&dir->i_mutex);
894         dput(parent);
895         return error;
896 }
897 EXPORT_SYMBOL_GPL(rpc_unlink);
898
899 enum {
900         RPCAUTH_info,
901         RPCAUTH_EOF
902 };
903
904 static const struct rpc_filelist authfiles[] = {
905         [RPCAUTH_info] = {
906                 .name = "info",
907                 .i_fop = &rpc_info_operations,
908                 .mode = S_IFREG | S_IRUSR,
909         },
910 };
911
912 static int rpc_clntdir_populate(struct dentry *dentry, void *private)
913 {
914         return rpc_populate(dentry,
915                             authfiles, RPCAUTH_info, RPCAUTH_EOF,
916                             private);
917 }
918
919 static void rpc_clntdir_depopulate(struct dentry *dentry)
920 {
921         rpc_depopulate(dentry, authfiles, RPCAUTH_info, RPCAUTH_EOF);
922 }
923
924 /**
925  * rpc_create_client_dir - Create a new rpc_client directory in rpc_pipefs
926  * @dentry: dentry from the rpc_pipefs root to the new directory
927  * @name: &struct qstr for the name
928  * @rpc_client: rpc client to associate with this directory
929  *
930  * This creates a directory at the given @path associated with
931  * @rpc_clnt, which will contain a file named "info" with some basic
932  * information about the client, together with any "pipes" that may
933  * later be created using rpc_mkpipe().
934  */
935 struct dentry *rpc_create_client_dir(struct dentry *dentry,
936                                    struct qstr *name,
937                                    struct rpc_clnt *rpc_client)
938 {
939         return rpc_mkdir_populate(dentry, name, S_IRUGO | S_IXUGO, NULL,
940                         rpc_clntdir_populate, rpc_client);
941 }
942
943 /**
944  * rpc_remove_client_dir - Remove a directory created with rpc_create_client_dir()
945  * @clnt: rpc client
946  */
947 int rpc_remove_client_dir(struct dentry *dentry)
948 {
949         return rpc_rmdir_depopulate(dentry, rpc_clntdir_depopulate);
950 }
951
952 static const struct rpc_filelist cache_pipefs_files[3] = {
953         [0] = {
954                 .name = "channel",
955                 .i_fop = &cache_file_operations_pipefs,
956                 .mode = S_IFREG|S_IRUSR|S_IWUSR,
957         },
958         [1] = {
959                 .name = "content",
960                 .i_fop = &content_file_operations_pipefs,
961                 .mode = S_IFREG|S_IRUSR,
962         },
963         [2] = {
964                 .name = "flush",
965                 .i_fop = &cache_flush_operations_pipefs,
966                 .mode = S_IFREG|S_IRUSR|S_IWUSR,
967         },
968 };
969
970 static int rpc_cachedir_populate(struct dentry *dentry, void *private)
971 {
972         return rpc_populate(dentry,
973                             cache_pipefs_files, 0, 3,
974                             private);
975 }
976
977 static void rpc_cachedir_depopulate(struct dentry *dentry)
978 {
979         rpc_depopulate(dentry, cache_pipefs_files, 0, 3);
980 }
981
982 struct dentry *rpc_create_cache_dir(struct dentry *parent, struct qstr *name,
983                                     umode_t umode, struct cache_detail *cd)
984 {
985         return rpc_mkdir_populate(parent, name, umode, NULL,
986                         rpc_cachedir_populate, cd);
987 }
988
989 void rpc_remove_cache_dir(struct dentry *dentry)
990 {
991         rpc_rmdir_depopulate(dentry, rpc_cachedir_depopulate);
992 }
993
994 /*
995  * populate the filesystem
996  */
997 static const struct super_operations s_ops = {
998         .alloc_inode    = rpc_alloc_inode,
999         .destroy_inode  = rpc_destroy_inode,
1000         .statfs         = simple_statfs,
1001 };
1002
1003 #define RPCAUTH_GSSMAGIC 0x67596969
1004
1005 /*
1006  * We have a single directory with 1 node in it.
1007  */
1008 enum {
1009         RPCAUTH_lockd,
1010         RPCAUTH_mount,
1011         RPCAUTH_nfs,
1012         RPCAUTH_portmap,
1013         RPCAUTH_statd,
1014         RPCAUTH_nfsd4_cb,
1015         RPCAUTH_cache,
1016         RPCAUTH_RootEOF
1017 };
1018
1019 static const struct rpc_filelist files[] = {
1020         [RPCAUTH_lockd] = {
1021                 .name = "lockd",
1022                 .mode = S_IFDIR | S_IRUGO | S_IXUGO,
1023         },
1024         [RPCAUTH_mount] = {
1025                 .name = "mount",
1026                 .mode = S_IFDIR | S_IRUGO | S_IXUGO,
1027         },
1028         [RPCAUTH_nfs] = {
1029                 .name = "nfs",
1030                 .mode = S_IFDIR | S_IRUGO | S_IXUGO,
1031         },
1032         [RPCAUTH_portmap] = {
1033                 .name = "portmap",
1034                 .mode = S_IFDIR | S_IRUGO | S_IXUGO,
1035         },
1036         [RPCAUTH_statd] = {
1037                 .name = "statd",
1038                 .mode = S_IFDIR | S_IRUGO | S_IXUGO,
1039         },
1040         [RPCAUTH_nfsd4_cb] = {
1041                 .name = "nfsd4_cb",
1042                 .mode = S_IFDIR | S_IRUGO | S_IXUGO,
1043         },
1044         [RPCAUTH_cache] = {
1045                 .name = "cache",
1046                 .mode = S_IFDIR | S_IRUGO | S_IXUGO,
1047         },
1048 };
1049
1050 /*
1051  * This call can be used only in RPC pipefs mount notification hooks.
1052  */
1053 struct dentry *rpc_d_lookup_sb(const struct super_block *sb,
1054                                const unsigned char *dir_name)
1055 {
1056         struct qstr dir = {
1057                 .name = dir_name,
1058                 .len = strlen(dir_name),
1059                 .hash = full_name_hash(dir_name, strlen(dir_name)),
1060         };
1061
1062         return d_lookup(sb->s_root, &dir);
1063 }
1064 EXPORT_SYMBOL_GPL(rpc_d_lookup_sb);
1065
1066 void rpc_pipefs_init_net(struct net *net)
1067 {
1068         struct sunrpc_net *sn = net_generic(net, sunrpc_net_id);
1069
1070         mutex_init(&sn->pipefs_sb_lock);
1071 }
1072
1073 /*
1074  * This call will be used for per network namespace operations calls.
1075  * Note: Function will be returned with pipefs_sb_lock taken if superblock was
1076  * found. This lock have to be released by rpc_put_sb_net() when all operations
1077  * will be completed.
1078  */
1079 struct super_block *rpc_get_sb_net(const struct net *net)
1080 {
1081         struct sunrpc_net *sn = net_generic(net, sunrpc_net_id);
1082
1083         mutex_lock(&sn->pipefs_sb_lock);
1084         if (sn->pipefs_sb)
1085                 return sn->pipefs_sb;
1086         mutex_unlock(&sn->pipefs_sb_lock);
1087         return NULL;
1088 }
1089 EXPORT_SYMBOL_GPL(rpc_get_sb_net);
1090
1091 void rpc_put_sb_net(const struct net *net)
1092 {
1093         struct sunrpc_net *sn = net_generic(net, sunrpc_net_id);
1094
1095         BUG_ON(sn->pipefs_sb == NULL);
1096         mutex_unlock(&sn->pipefs_sb_lock);
1097 }
1098 EXPORT_SYMBOL_GPL(rpc_put_sb_net);
1099
1100 static int
1101 rpc_fill_super(struct super_block *sb, void *data, int silent)
1102 {
1103         struct inode *inode;
1104         struct dentry *root;
1105         struct net *net = data;
1106         struct sunrpc_net *sn = net_generic(net, sunrpc_net_id);
1107         int err;
1108
1109         sb->s_blocksize = PAGE_CACHE_SIZE;
1110         sb->s_blocksize_bits = PAGE_CACHE_SHIFT;
1111         sb->s_magic = RPCAUTH_GSSMAGIC;
1112         sb->s_op = &s_ops;
1113         sb->s_time_gran = 1;
1114
1115         inode = rpc_get_inode(sb, S_IFDIR | 0755);
1116         if (!inode)
1117                 return -ENOMEM;
1118         sb->s_root = root = d_alloc_root(inode);
1119         if (!root) {
1120                 iput(inode);
1121                 return -ENOMEM;
1122         }
1123         if (rpc_populate(root, files, RPCAUTH_lockd, RPCAUTH_RootEOF, NULL))
1124                 return -ENOMEM;
1125         dprintk("RPC:   sending pipefs MOUNT notification for net %p%s\n", net,
1126                                                                 NET_NAME(net));
1127         err = blocking_notifier_call_chain(&rpc_pipefs_notifier_list,
1128                                            RPC_PIPEFS_MOUNT,
1129                                            sb);
1130         if (err)
1131                 goto err_depopulate;
1132         sb->s_fs_info = get_net(net);
1133         sn->pipefs_sb = sb;
1134         return 0;
1135
1136 err_depopulate:
1137         blocking_notifier_call_chain(&rpc_pipefs_notifier_list,
1138                                            RPC_PIPEFS_UMOUNT,
1139                                            sb);
1140         __rpc_depopulate(root, files, RPCAUTH_lockd, RPCAUTH_RootEOF);
1141         return err;
1142 }
1143
1144 static struct dentry *
1145 rpc_mount(struct file_system_type *fs_type,
1146                 int flags, const char *dev_name, void *data)
1147 {
1148         return mount_ns(fs_type, flags, current->nsproxy->net_ns, rpc_fill_super);
1149 }
1150
1151 void rpc_kill_sb(struct super_block *sb)
1152 {
1153         struct net *net = sb->s_fs_info;
1154         struct sunrpc_net *sn = net_generic(net, sunrpc_net_id);
1155
1156         mutex_lock(&sn->pipefs_sb_lock);
1157         sn->pipefs_sb = NULL;
1158         mutex_unlock(&sn->pipefs_sb_lock);
1159         put_net(net);
1160         dprintk("RPC:   sending pipefs UMOUNT notification for net %p%s\n", net,
1161                                                                 NET_NAME(net));
1162         blocking_notifier_call_chain(&rpc_pipefs_notifier_list,
1163                                            RPC_PIPEFS_UMOUNT,
1164                                            sb);
1165         kill_litter_super(sb);
1166 }
1167
1168 static struct file_system_type rpc_pipe_fs_type = {
1169         .owner          = THIS_MODULE,
1170         .name           = "rpc_pipefs",
1171         .mount          = rpc_mount,
1172         .kill_sb        = rpc_kill_sb,
1173 };
1174
1175 static void
1176 init_once(void *foo)
1177 {
1178         struct rpc_inode *rpci = (struct rpc_inode *) foo;
1179
1180         inode_init_once(&rpci->vfs_inode);
1181         rpci->private = NULL;
1182         rpci->pipe = NULL;
1183         init_waitqueue_head(&rpci->waitq);
1184 }
1185
1186 int register_rpc_pipefs(void)
1187 {
1188         int err;
1189
1190         rpc_inode_cachep = kmem_cache_create("rpc_inode_cache",
1191                                 sizeof(struct rpc_inode),
1192                                 0, (SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT|
1193                                                 SLAB_MEM_SPREAD),
1194                                 init_once);
1195         if (!rpc_inode_cachep)
1196                 return -ENOMEM;
1197         err = rpc_clients_notifier_register();
1198         if (err)
1199                 goto err_notifier;
1200         err = register_filesystem(&rpc_pipe_fs_type);
1201         if (err)
1202                 goto err_register;
1203         return 0;
1204
1205 err_register:
1206         rpc_clients_notifier_unregister();
1207 err_notifier:
1208         kmem_cache_destroy(rpc_inode_cachep);
1209         return err;
1210 }
1211
1212 void unregister_rpc_pipefs(void)
1213 {
1214         rpc_clients_notifier_unregister();
1215         kmem_cache_destroy(rpc_inode_cachep);
1216         unregister_filesystem(&rpc_pipe_fs_type);
1217 }
1218
1219 /* Make 'mount -t rpc_pipefs ...' autoload this module. */
1220 MODULE_ALIAS("rpc_pipefs");