xref: /linux/fs/nfsd/nfs4recover.c (revision 04eeb606a8383b306f4bc6991da8231b5f3924b0)
1 /*
2 *  Copyright (c) 2004 The Regents of the University of Michigan.
3 *  Copyright (c) 2012 Jeff Layton <jlayton@redhat.com>
4 *  All rights reserved.
5 *
6 *  Andy Adamson <andros@citi.umich.edu>
7 *
8 *  Redistribution and use in source and binary forms, with or without
9 *  modification, are permitted provided that the following conditions
10 *  are met:
11 *
12 *  1. Redistributions of source code must retain the above copyright
13 *     notice, this list of conditions and the following disclaimer.
14 *  2. Redistributions in binary form must reproduce the above copyright
15 *     notice, this list of conditions and the following disclaimer in the
16 *     documentation and/or other materials provided with the distribution.
17 *  3. Neither the name of the University nor the names of its
18 *     contributors may be used to endorse or promote products derived
19 *     from this software without specific prior written permission.
20 *
21 *  THIS SOFTWARE IS PROVIDED ``AS IS'' AND ANY EXPRESS OR IMPLIED
22 *  WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF
23 *  MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
24 *  DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
25 *  FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
26 *  CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
27 *  SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR
28 *  BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF
29 *  LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING
30 *  NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
31 *  SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
32 *
33 */
34 
35 #include <linux/file.h>
36 #include <linux/slab.h>
37 #include <linux/namei.h>
38 #include <linux/crypto.h>
39 #include <linux/sched.h>
40 #include <linux/fs.h>
41 #include <linux/module.h>
42 #include <net/net_namespace.h>
43 #include <linux/sunrpc/rpc_pipe_fs.h>
44 #include <linux/sunrpc/clnt.h>
45 #include <linux/nfsd/cld.h>
46 
47 #include "nfsd.h"
48 #include "state.h"
49 #include "vfs.h"
50 #include "netns.h"
51 
52 #define NFSDDBG_FACILITY                NFSDDBG_PROC
53 
54 /* Declarations */
55 struct nfsd4_client_tracking_ops {
56 	int (*init)(struct net *);
57 	void (*exit)(struct net *);
58 	void (*create)(struct nfs4_client *);
59 	void (*remove)(struct nfs4_client *);
60 	int (*check)(struct nfs4_client *);
61 	void (*grace_done)(struct nfsd_net *);
62 };
63 
64 /* Globals */
65 static char user_recovery_dirname[PATH_MAX] = "/var/lib/nfs/v4recovery";
66 
67 static int
68 nfs4_save_creds(const struct cred **original_creds)
69 {
70 	struct cred *new;
71 
72 	new = prepare_creds();
73 	if (!new)
74 		return -ENOMEM;
75 
76 	new->fsuid = GLOBAL_ROOT_UID;
77 	new->fsgid = GLOBAL_ROOT_GID;
78 	*original_creds = override_creds(new);
79 	put_cred(new);
80 	return 0;
81 }
82 
83 static void
84 nfs4_reset_creds(const struct cred *original)
85 {
86 	revert_creds(original);
87 }
88 
89 static void
90 md5_to_hex(char *out, char *md5)
91 {
92 	int i;
93 
94 	for (i=0; i<16; i++) {
95 		unsigned char c = md5[i];
96 
97 		*out++ = '0' + ((c&0xf0)>>4) + (c>=0xa0)*('a'-'9'-1);
98 		*out++ = '0' + (c&0x0f) + ((c&0x0f)>=0x0a)*('a'-'9'-1);
99 	}
100 	*out = '\0';
101 }
102 
103 static int
104 nfs4_make_rec_clidname(char *dname, const struct xdr_netobj *clname)
105 {
106 	struct xdr_netobj cksum;
107 	struct hash_desc desc;
108 	struct scatterlist sg;
109 	int status;
110 
111 	dprintk("NFSD: nfs4_make_rec_clidname for %.*s\n",
112 			clname->len, clname->data);
113 	desc.flags = CRYPTO_TFM_REQ_MAY_SLEEP;
114 	desc.tfm = crypto_alloc_hash("md5", 0, CRYPTO_ALG_ASYNC);
115 	if (IS_ERR(desc.tfm)) {
116 		status = PTR_ERR(desc.tfm);
117 		goto out_no_tfm;
118 	}
119 
120 	cksum.len = crypto_hash_digestsize(desc.tfm);
121 	cksum.data = kmalloc(cksum.len, GFP_KERNEL);
122 	if (cksum.data == NULL) {
123 		status = -ENOMEM;
124  		goto out;
125 	}
126 
127 	sg_init_one(&sg, clname->data, clname->len);
128 
129 	status = crypto_hash_digest(&desc, &sg, sg.length, cksum.data);
130 	if (status)
131 		goto out;
132 
133 	md5_to_hex(dname, cksum.data);
134 
135 	status = 0;
136 out:
137 	kfree(cksum.data);
138 	crypto_free_hash(desc.tfm);
139 out_no_tfm:
140 	return status;
141 }
142 
143 /*
144  * If we had an error generating the recdir name for the legacy tracker
145  * then warn the admin. If the error doesn't appear to be transient,
146  * then disable recovery tracking.
147  */
148 static void
149 legacy_recdir_name_error(struct nfs4_client *clp, int error)
150 {
151 	printk(KERN_ERR "NFSD: unable to generate recoverydir "
152 			"name (%d).\n", error);
153 
154 	/*
155 	 * if the algorithm just doesn't exist, then disable the recovery
156 	 * tracker altogether. The crypto libs will generally return this if
157 	 * FIPS is enabled as well.
158 	 */
159 	if (error == -ENOENT) {
160 		printk(KERN_ERR "NFSD: disabling legacy clientid tracking. "
161 			"Reboot recovery will not function correctly!\n");
162 		nfsd4_client_tracking_exit(clp->net);
163 	}
164 }
165 
166 static void
167 nfsd4_create_clid_dir(struct nfs4_client *clp)
168 {
169 	const struct cred *original_cred;
170 	char dname[HEXDIR_LEN];
171 	struct dentry *dir, *dentry;
172 	struct nfs4_client_reclaim *crp;
173 	int status;
174 	struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
175 
176 	if (test_and_set_bit(NFSD4_CLIENT_STABLE, &clp->cl_flags))
177 		return;
178 	if (!nn->rec_file)
179 		return;
180 
181 	status = nfs4_make_rec_clidname(dname, &clp->cl_name);
182 	if (status)
183 		return legacy_recdir_name_error(clp, status);
184 
185 	status = nfs4_save_creds(&original_cred);
186 	if (status < 0)
187 		return;
188 
189 	status = mnt_want_write_file(nn->rec_file);
190 	if (status)
191 		goto out_creds;
192 
193 	dir = nn->rec_file->f_path.dentry;
194 	/* lock the parent */
195 	mutex_lock(&dir->d_inode->i_mutex);
196 
197 	dentry = lookup_one_len(dname, dir, HEXDIR_LEN-1);
198 	if (IS_ERR(dentry)) {
199 		status = PTR_ERR(dentry);
200 		goto out_unlock;
201 	}
202 	if (dentry->d_inode)
203 		/*
204 		 * In the 4.1 case, where we're called from
205 		 * reclaim_complete(), records from the previous reboot
206 		 * may still be left, so this is OK.
207 		 *
208 		 * In the 4.0 case, we should never get here; but we may
209 		 * as well be forgiving and just succeed silently.
210 		 */
211 		goto out_put;
212 	status = vfs_mkdir(dir->d_inode, dentry, S_IRWXU);
213 out_put:
214 	dput(dentry);
215 out_unlock:
216 	mutex_unlock(&dir->d_inode->i_mutex);
217 	if (status == 0) {
218 		if (nn->in_grace) {
219 			crp = nfs4_client_to_reclaim(dname, nn);
220 			if (crp)
221 				crp->cr_clp = clp;
222 		}
223 		vfs_fsync(nn->rec_file, 0);
224 	} else {
225 		printk(KERN_ERR "NFSD: failed to write recovery record"
226 				" (err %d); please check that %s exists"
227 				" and is writeable", status,
228 				user_recovery_dirname);
229 	}
230 	mnt_drop_write_file(nn->rec_file);
231 out_creds:
232 	nfs4_reset_creds(original_cred);
233 }
234 
235 typedef int (recdir_func)(struct dentry *, struct dentry *, struct nfsd_net *);
236 
237 struct name_list {
238 	char name[HEXDIR_LEN];
239 	struct list_head list;
240 };
241 
242 struct nfs4_dir_ctx {
243 	struct dir_context ctx;
244 	struct list_head names;
245 };
246 
247 static int
248 nfsd4_build_namelist(void *arg, const char *name, int namlen,
249 		loff_t offset, u64 ino, unsigned int d_type)
250 {
251 	struct nfs4_dir_ctx *ctx = arg;
252 	struct name_list *entry;
253 
254 	if (namlen != HEXDIR_LEN - 1)
255 		return 0;
256 	entry = kmalloc(sizeof(struct name_list), GFP_KERNEL);
257 	if (entry == NULL)
258 		return -ENOMEM;
259 	memcpy(entry->name, name, HEXDIR_LEN - 1);
260 	entry->name[HEXDIR_LEN - 1] = '\0';
261 	list_add(&entry->list, &ctx->names);
262 	return 0;
263 }
264 
265 static int
266 nfsd4_list_rec_dir(recdir_func *f, struct nfsd_net *nn)
267 {
268 	const struct cred *original_cred;
269 	struct dentry *dir = nn->rec_file->f_path.dentry;
270 	struct nfs4_dir_ctx ctx = {
271 		.ctx.actor = nfsd4_build_namelist,
272 		.names = LIST_HEAD_INIT(ctx.names)
273 	};
274 	int status;
275 
276 	status = nfs4_save_creds(&original_cred);
277 	if (status < 0)
278 		return status;
279 
280 	status = vfs_llseek(nn->rec_file, 0, SEEK_SET);
281 	if (status < 0) {
282 		nfs4_reset_creds(original_cred);
283 		return status;
284 	}
285 
286 	status = iterate_dir(nn->rec_file, &ctx.ctx);
287 	mutex_lock_nested(&dir->d_inode->i_mutex, I_MUTEX_PARENT);
288 	while (!list_empty(&ctx.names)) {
289 		struct name_list *entry;
290 		entry = list_entry(ctx.names.next, struct name_list, list);
291 		if (!status) {
292 			struct dentry *dentry;
293 			dentry = lookup_one_len(entry->name, dir, HEXDIR_LEN-1);
294 			if (IS_ERR(dentry)) {
295 				status = PTR_ERR(dentry);
296 				break;
297 			}
298 			status = f(dir, dentry, nn);
299 			dput(dentry);
300 		}
301 		list_del(&entry->list);
302 		kfree(entry);
303 	}
304 	mutex_unlock(&dir->d_inode->i_mutex);
305 	nfs4_reset_creds(original_cred);
306 	return status;
307 }
308 
309 static int
310 nfsd4_unlink_clid_dir(char *name, int namlen, struct nfsd_net *nn)
311 {
312 	struct dentry *dir, *dentry;
313 	int status;
314 
315 	dprintk("NFSD: nfsd4_unlink_clid_dir. name %.*s\n", namlen, name);
316 
317 	dir = nn->rec_file->f_path.dentry;
318 	mutex_lock_nested(&dir->d_inode->i_mutex, I_MUTEX_PARENT);
319 	dentry = lookup_one_len(name, dir, namlen);
320 	if (IS_ERR(dentry)) {
321 		status = PTR_ERR(dentry);
322 		goto out_unlock;
323 	}
324 	status = -ENOENT;
325 	if (!dentry->d_inode)
326 		goto out;
327 	status = vfs_rmdir(dir->d_inode, dentry);
328 out:
329 	dput(dentry);
330 out_unlock:
331 	mutex_unlock(&dir->d_inode->i_mutex);
332 	return status;
333 }
334 
335 static void
336 nfsd4_remove_clid_dir(struct nfs4_client *clp)
337 {
338 	const struct cred *original_cred;
339 	struct nfs4_client_reclaim *crp;
340 	char dname[HEXDIR_LEN];
341 	int status;
342 	struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
343 
344 	if (!nn->rec_file || !test_bit(NFSD4_CLIENT_STABLE, &clp->cl_flags))
345 		return;
346 
347 	status = nfs4_make_rec_clidname(dname, &clp->cl_name);
348 	if (status)
349 		return legacy_recdir_name_error(clp, status);
350 
351 	status = mnt_want_write_file(nn->rec_file);
352 	if (status)
353 		goto out;
354 	clear_bit(NFSD4_CLIENT_STABLE, &clp->cl_flags);
355 
356 	status = nfs4_save_creds(&original_cred);
357 	if (status < 0)
358 		goto out_drop_write;
359 
360 	status = nfsd4_unlink_clid_dir(dname, HEXDIR_LEN-1, nn);
361 	nfs4_reset_creds(original_cred);
362 	if (status == 0) {
363 		vfs_fsync(nn->rec_file, 0);
364 		if (nn->in_grace) {
365 			/* remove reclaim record */
366 			crp = nfsd4_find_reclaim_client(dname, nn);
367 			if (crp)
368 				nfs4_remove_reclaim_record(crp, nn);
369 		}
370 	}
371 out_drop_write:
372 	mnt_drop_write_file(nn->rec_file);
373 out:
374 	if (status)
375 		printk("NFSD: Failed to remove expired client state directory"
376 				" %.*s\n", HEXDIR_LEN, dname);
377 }
378 
379 static int
380 purge_old(struct dentry *parent, struct dentry *child, struct nfsd_net *nn)
381 {
382 	int status;
383 
384 	if (nfs4_has_reclaimed_state(child->d_name.name, nn))
385 		return 0;
386 
387 	status = vfs_rmdir(parent->d_inode, child);
388 	if (status)
389 		printk("failed to remove client recovery directory %pd\n",
390 				child);
391 	/* Keep trying, success or failure: */
392 	return 0;
393 }
394 
395 static void
396 nfsd4_recdir_purge_old(struct nfsd_net *nn)
397 {
398 	int status;
399 
400 	nn->in_grace = false;
401 	if (!nn->rec_file)
402 		return;
403 	status = mnt_want_write_file(nn->rec_file);
404 	if (status)
405 		goto out;
406 	status = nfsd4_list_rec_dir(purge_old, nn);
407 	if (status == 0)
408 		vfs_fsync(nn->rec_file, 0);
409 	mnt_drop_write_file(nn->rec_file);
410 out:
411 	nfs4_release_reclaim(nn);
412 	if (status)
413 		printk("nfsd4: failed to purge old clients from recovery"
414 			" directory %pD\n", nn->rec_file);
415 }
416 
417 static int
418 load_recdir(struct dentry *parent, struct dentry *child, struct nfsd_net *nn)
419 {
420 	if (child->d_name.len != HEXDIR_LEN - 1) {
421 		printk("nfsd4: illegal name %pd in recovery directory\n",
422 				child);
423 		/* Keep trying; maybe the others are OK: */
424 		return 0;
425 	}
426 	nfs4_client_to_reclaim(child->d_name.name, nn);
427 	return 0;
428 }
429 
430 static int
431 nfsd4_recdir_load(struct net *net) {
432 	int status;
433 	struct nfsd_net *nn =  net_generic(net, nfsd_net_id);
434 
435 	if (!nn->rec_file)
436 		return 0;
437 
438 	status = nfsd4_list_rec_dir(load_recdir, nn);
439 	if (status)
440 		printk("nfsd4: failed loading clients from recovery"
441 			" directory %pD\n", nn->rec_file);
442 	return status;
443 }
444 
445 /*
446  * Hold reference to the recovery directory.
447  */
448 
449 static int
450 nfsd4_init_recdir(struct net *net)
451 {
452 	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
453 	const struct cred *original_cred;
454 	int status;
455 
456 	printk("NFSD: Using %s as the NFSv4 state recovery directory\n",
457 			user_recovery_dirname);
458 
459 	BUG_ON(nn->rec_file);
460 
461 	status = nfs4_save_creds(&original_cred);
462 	if (status < 0) {
463 		printk("NFSD: Unable to change credentials to find recovery"
464 		       " directory: error %d\n",
465 		       status);
466 		return status;
467 	}
468 
469 	nn->rec_file = filp_open(user_recovery_dirname, O_RDONLY | O_DIRECTORY, 0);
470 	if (IS_ERR(nn->rec_file)) {
471 		printk("NFSD: unable to find recovery directory %s\n",
472 				user_recovery_dirname);
473 		status = PTR_ERR(nn->rec_file);
474 		nn->rec_file = NULL;
475 	}
476 
477 	nfs4_reset_creds(original_cred);
478 	if (!status)
479 		nn->in_grace = true;
480 	return status;
481 }
482 
483 static void
484 nfsd4_shutdown_recdir(struct net *net)
485 {
486 	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
487 
488 	if (!nn->rec_file)
489 		return;
490 	fput(nn->rec_file);
491 	nn->rec_file = NULL;
492 }
493 
494 static int
495 nfs4_legacy_state_init(struct net *net)
496 {
497 	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
498 	int i;
499 
500 	nn->reclaim_str_hashtbl = kmalloc(sizeof(struct list_head) *
501 					  CLIENT_HASH_SIZE, GFP_KERNEL);
502 	if (!nn->reclaim_str_hashtbl)
503 		return -ENOMEM;
504 
505 	for (i = 0; i < CLIENT_HASH_SIZE; i++)
506 		INIT_LIST_HEAD(&nn->reclaim_str_hashtbl[i]);
507 	nn->reclaim_str_hashtbl_size = 0;
508 
509 	return 0;
510 }
511 
512 static void
513 nfs4_legacy_state_shutdown(struct net *net)
514 {
515 	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
516 
517 	kfree(nn->reclaim_str_hashtbl);
518 }
519 
520 static int
521 nfsd4_load_reboot_recovery_data(struct net *net)
522 {
523 	int status;
524 
525 	status = nfsd4_init_recdir(net);
526 	if (status)
527 		return status;
528 
529 	status = nfsd4_recdir_load(net);
530 	if (status)
531 		nfsd4_shutdown_recdir(net);
532 
533 	return status;
534 }
535 
536 static int
537 nfsd4_legacy_tracking_init(struct net *net)
538 {
539 	int status;
540 
541 	/* XXX: The legacy code won't work in a container */
542 	if (net != &init_net) {
543 		WARN(1, KERN_ERR "NFSD: attempt to initialize legacy client "
544 			"tracking in a container!\n");
545 		return -EINVAL;
546 	}
547 
548 	status = nfs4_legacy_state_init(net);
549 	if (status)
550 		return status;
551 
552 	status = nfsd4_load_reboot_recovery_data(net);
553 	if (status)
554 		goto err;
555 	return 0;
556 
557 err:
558 	nfs4_legacy_state_shutdown(net);
559 	return status;
560 }
561 
562 static void
563 nfsd4_legacy_tracking_exit(struct net *net)
564 {
565 	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
566 
567 	nfs4_release_reclaim(nn);
568 	nfsd4_shutdown_recdir(net);
569 	nfs4_legacy_state_shutdown(net);
570 }
571 
572 /*
573  * Change the NFSv4 recovery directory to recdir.
574  */
575 int
576 nfs4_reset_recoverydir(char *recdir)
577 {
578 	int status;
579 	struct path path;
580 
581 	status = kern_path(recdir, LOOKUP_FOLLOW, &path);
582 	if (status)
583 		return status;
584 	status = -ENOTDIR;
585 	if (S_ISDIR(path.dentry->d_inode->i_mode)) {
586 		strcpy(user_recovery_dirname, recdir);
587 		status = 0;
588 	}
589 	path_put(&path);
590 	return status;
591 }
592 
593 char *
594 nfs4_recoverydir(void)
595 {
596 	return user_recovery_dirname;
597 }
598 
599 static int
600 nfsd4_check_legacy_client(struct nfs4_client *clp)
601 {
602 	int status;
603 	char dname[HEXDIR_LEN];
604 	struct nfs4_client_reclaim *crp;
605 	struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
606 
607 	/* did we already find that this client is stable? */
608 	if (test_bit(NFSD4_CLIENT_STABLE, &clp->cl_flags))
609 		return 0;
610 
611 	status = nfs4_make_rec_clidname(dname, &clp->cl_name);
612 	if (status) {
613 		legacy_recdir_name_error(clp, status);
614 		return status;
615 	}
616 
617 	/* look for it in the reclaim hashtable otherwise */
618 	crp = nfsd4_find_reclaim_client(dname, nn);
619 	if (crp) {
620 		set_bit(NFSD4_CLIENT_STABLE, &clp->cl_flags);
621 		crp->cr_clp = clp;
622 		return 0;
623 	}
624 
625 	return -ENOENT;
626 }
627 
628 static struct nfsd4_client_tracking_ops nfsd4_legacy_tracking_ops = {
629 	.init		= nfsd4_legacy_tracking_init,
630 	.exit		= nfsd4_legacy_tracking_exit,
631 	.create		= nfsd4_create_clid_dir,
632 	.remove		= nfsd4_remove_clid_dir,
633 	.check		= nfsd4_check_legacy_client,
634 	.grace_done	= nfsd4_recdir_purge_old,
635 };
636 
637 /* Globals */
638 #define NFSD_PIPE_DIR		"nfsd"
639 #define NFSD_CLD_PIPE		"cld"
640 
641 /* per-net-ns structure for holding cld upcall info */
642 struct cld_net {
643 	struct rpc_pipe		*cn_pipe;
644 	spinlock_t		 cn_lock;
645 	struct list_head	 cn_list;
646 	unsigned int		 cn_xid;
647 };
648 
649 struct cld_upcall {
650 	struct list_head	 cu_list;
651 	struct cld_net		*cu_net;
652 	struct task_struct	*cu_task;
653 	struct cld_msg		 cu_msg;
654 };
655 
656 static int
657 __cld_pipe_upcall(struct rpc_pipe *pipe, struct cld_msg *cmsg)
658 {
659 	int ret;
660 	struct rpc_pipe_msg msg;
661 
662 	memset(&msg, 0, sizeof(msg));
663 	msg.data = cmsg;
664 	msg.len = sizeof(*cmsg);
665 
666 	/*
667 	 * Set task state before we queue the upcall. That prevents
668 	 * wake_up_process in the downcall from racing with schedule.
669 	 */
670 	set_current_state(TASK_UNINTERRUPTIBLE);
671 	ret = rpc_queue_upcall(pipe, &msg);
672 	if (ret < 0) {
673 		set_current_state(TASK_RUNNING);
674 		goto out;
675 	}
676 
677 	schedule();
678 
679 	if (msg.errno < 0)
680 		ret = msg.errno;
681 out:
682 	return ret;
683 }
684 
685 static int
686 cld_pipe_upcall(struct rpc_pipe *pipe, struct cld_msg *cmsg)
687 {
688 	int ret;
689 
690 	/*
691 	 * -EAGAIN occurs when pipe is closed and reopened while there are
692 	 *  upcalls queued.
693 	 */
694 	do {
695 		ret = __cld_pipe_upcall(pipe, cmsg);
696 	} while (ret == -EAGAIN);
697 
698 	return ret;
699 }
700 
701 static ssize_t
702 cld_pipe_downcall(struct file *filp, const char __user *src, size_t mlen)
703 {
704 	struct cld_upcall *tmp, *cup;
705 	struct cld_msg __user *cmsg = (struct cld_msg __user *)src;
706 	uint32_t xid;
707 	struct nfsd_net *nn = net_generic(filp->f_dentry->d_sb->s_fs_info,
708 						nfsd_net_id);
709 	struct cld_net *cn = nn->cld_net;
710 
711 	if (mlen != sizeof(*cmsg)) {
712 		dprintk("%s: got %zu bytes, expected %zu\n", __func__, mlen,
713 			sizeof(*cmsg));
714 		return -EINVAL;
715 	}
716 
717 	/* copy just the xid so we can try to find that */
718 	if (copy_from_user(&xid, &cmsg->cm_xid, sizeof(xid)) != 0) {
719 		dprintk("%s: error when copying xid from userspace", __func__);
720 		return -EFAULT;
721 	}
722 
723 	/* walk the list and find corresponding xid */
724 	cup = NULL;
725 	spin_lock(&cn->cn_lock);
726 	list_for_each_entry(tmp, &cn->cn_list, cu_list) {
727 		if (get_unaligned(&tmp->cu_msg.cm_xid) == xid) {
728 			cup = tmp;
729 			list_del_init(&cup->cu_list);
730 			break;
731 		}
732 	}
733 	spin_unlock(&cn->cn_lock);
734 
735 	/* couldn't find upcall? */
736 	if (!cup) {
737 		dprintk("%s: couldn't find upcall -- xid=%u\n", __func__, xid);
738 		return -EINVAL;
739 	}
740 
741 	if (copy_from_user(&cup->cu_msg, src, mlen) != 0)
742 		return -EFAULT;
743 
744 	wake_up_process(cup->cu_task);
745 	return mlen;
746 }
747 
748 static void
749 cld_pipe_destroy_msg(struct rpc_pipe_msg *msg)
750 {
751 	struct cld_msg *cmsg = msg->data;
752 	struct cld_upcall *cup = container_of(cmsg, struct cld_upcall,
753 						 cu_msg);
754 
755 	/* errno >= 0 means we got a downcall */
756 	if (msg->errno >= 0)
757 		return;
758 
759 	wake_up_process(cup->cu_task);
760 }
761 
762 static const struct rpc_pipe_ops cld_upcall_ops = {
763 	.upcall		= rpc_pipe_generic_upcall,
764 	.downcall	= cld_pipe_downcall,
765 	.destroy_msg	= cld_pipe_destroy_msg,
766 };
767 
768 static struct dentry *
769 nfsd4_cld_register_sb(struct super_block *sb, struct rpc_pipe *pipe)
770 {
771 	struct dentry *dir, *dentry;
772 
773 	dir = rpc_d_lookup_sb(sb, NFSD_PIPE_DIR);
774 	if (dir == NULL)
775 		return ERR_PTR(-ENOENT);
776 	dentry = rpc_mkpipe_dentry(dir, NFSD_CLD_PIPE, NULL, pipe);
777 	dput(dir);
778 	return dentry;
779 }
780 
781 static void
782 nfsd4_cld_unregister_sb(struct rpc_pipe *pipe)
783 {
784 	if (pipe->dentry)
785 		rpc_unlink(pipe->dentry);
786 }
787 
788 static struct dentry *
789 nfsd4_cld_register_net(struct net *net, struct rpc_pipe *pipe)
790 {
791 	struct super_block *sb;
792 	struct dentry *dentry;
793 
794 	sb = rpc_get_sb_net(net);
795 	if (!sb)
796 		return NULL;
797 	dentry = nfsd4_cld_register_sb(sb, pipe);
798 	rpc_put_sb_net(net);
799 	return dentry;
800 }
801 
802 static void
803 nfsd4_cld_unregister_net(struct net *net, struct rpc_pipe *pipe)
804 {
805 	struct super_block *sb;
806 
807 	sb = rpc_get_sb_net(net);
808 	if (sb) {
809 		nfsd4_cld_unregister_sb(pipe);
810 		rpc_put_sb_net(net);
811 	}
812 }
813 
814 /* Initialize rpc_pipefs pipe for communication with client tracking daemon */
815 static int
816 nfsd4_init_cld_pipe(struct net *net)
817 {
818 	int ret;
819 	struct dentry *dentry;
820 	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
821 	struct cld_net *cn;
822 
823 	if (nn->cld_net)
824 		return 0;
825 
826 	cn = kzalloc(sizeof(*cn), GFP_KERNEL);
827 	if (!cn) {
828 		ret = -ENOMEM;
829 		goto err;
830 	}
831 
832 	cn->cn_pipe = rpc_mkpipe_data(&cld_upcall_ops, RPC_PIPE_WAIT_FOR_OPEN);
833 	if (IS_ERR(cn->cn_pipe)) {
834 		ret = PTR_ERR(cn->cn_pipe);
835 		goto err;
836 	}
837 	spin_lock_init(&cn->cn_lock);
838 	INIT_LIST_HEAD(&cn->cn_list);
839 
840 	dentry = nfsd4_cld_register_net(net, cn->cn_pipe);
841 	if (IS_ERR(dentry)) {
842 		ret = PTR_ERR(dentry);
843 		goto err_destroy_data;
844 	}
845 
846 	cn->cn_pipe->dentry = dentry;
847 	nn->cld_net = cn;
848 	return 0;
849 
850 err_destroy_data:
851 	rpc_destroy_pipe_data(cn->cn_pipe);
852 err:
853 	kfree(cn);
854 	printk(KERN_ERR "NFSD: unable to create nfsdcld upcall pipe (%d)\n",
855 			ret);
856 	return ret;
857 }
858 
859 static void
860 nfsd4_remove_cld_pipe(struct net *net)
861 {
862 	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
863 	struct cld_net *cn = nn->cld_net;
864 
865 	nfsd4_cld_unregister_net(net, cn->cn_pipe);
866 	rpc_destroy_pipe_data(cn->cn_pipe);
867 	kfree(nn->cld_net);
868 	nn->cld_net = NULL;
869 }
870 
871 static struct cld_upcall *
872 alloc_cld_upcall(struct cld_net *cn)
873 {
874 	struct cld_upcall *new, *tmp;
875 
876 	new = kzalloc(sizeof(*new), GFP_KERNEL);
877 	if (!new)
878 		return new;
879 
880 	/* FIXME: hard cap on number in flight? */
881 restart_search:
882 	spin_lock(&cn->cn_lock);
883 	list_for_each_entry(tmp, &cn->cn_list, cu_list) {
884 		if (tmp->cu_msg.cm_xid == cn->cn_xid) {
885 			cn->cn_xid++;
886 			spin_unlock(&cn->cn_lock);
887 			goto restart_search;
888 		}
889 	}
890 	new->cu_task = current;
891 	new->cu_msg.cm_vers = CLD_UPCALL_VERSION;
892 	put_unaligned(cn->cn_xid++, &new->cu_msg.cm_xid);
893 	new->cu_net = cn;
894 	list_add(&new->cu_list, &cn->cn_list);
895 	spin_unlock(&cn->cn_lock);
896 
897 	dprintk("%s: allocated xid %u\n", __func__, new->cu_msg.cm_xid);
898 
899 	return new;
900 }
901 
902 static void
903 free_cld_upcall(struct cld_upcall *victim)
904 {
905 	struct cld_net *cn = victim->cu_net;
906 
907 	spin_lock(&cn->cn_lock);
908 	list_del(&victim->cu_list);
909 	spin_unlock(&cn->cn_lock);
910 	kfree(victim);
911 }
912 
913 /* Ask daemon to create a new record */
914 static void
915 nfsd4_cld_create(struct nfs4_client *clp)
916 {
917 	int ret;
918 	struct cld_upcall *cup;
919 	struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
920 	struct cld_net *cn = nn->cld_net;
921 
922 	/* Don't upcall if it's already stored */
923 	if (test_bit(NFSD4_CLIENT_STABLE, &clp->cl_flags))
924 		return;
925 
926 	cup = alloc_cld_upcall(cn);
927 	if (!cup) {
928 		ret = -ENOMEM;
929 		goto out_err;
930 	}
931 
932 	cup->cu_msg.cm_cmd = Cld_Create;
933 	cup->cu_msg.cm_u.cm_name.cn_len = clp->cl_name.len;
934 	memcpy(cup->cu_msg.cm_u.cm_name.cn_id, clp->cl_name.data,
935 			clp->cl_name.len);
936 
937 	ret = cld_pipe_upcall(cn->cn_pipe, &cup->cu_msg);
938 	if (!ret) {
939 		ret = cup->cu_msg.cm_status;
940 		set_bit(NFSD4_CLIENT_STABLE, &clp->cl_flags);
941 	}
942 
943 	free_cld_upcall(cup);
944 out_err:
945 	if (ret)
946 		printk(KERN_ERR "NFSD: Unable to create client "
947 				"record on stable storage: %d\n", ret);
948 }
949 
950 /* Ask daemon to create a new record */
951 static void
952 nfsd4_cld_remove(struct nfs4_client *clp)
953 {
954 	int ret;
955 	struct cld_upcall *cup;
956 	struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
957 	struct cld_net *cn = nn->cld_net;
958 
959 	/* Don't upcall if it's already removed */
960 	if (!test_bit(NFSD4_CLIENT_STABLE, &clp->cl_flags))
961 		return;
962 
963 	cup = alloc_cld_upcall(cn);
964 	if (!cup) {
965 		ret = -ENOMEM;
966 		goto out_err;
967 	}
968 
969 	cup->cu_msg.cm_cmd = Cld_Remove;
970 	cup->cu_msg.cm_u.cm_name.cn_len = clp->cl_name.len;
971 	memcpy(cup->cu_msg.cm_u.cm_name.cn_id, clp->cl_name.data,
972 			clp->cl_name.len);
973 
974 	ret = cld_pipe_upcall(cn->cn_pipe, &cup->cu_msg);
975 	if (!ret) {
976 		ret = cup->cu_msg.cm_status;
977 		clear_bit(NFSD4_CLIENT_STABLE, &clp->cl_flags);
978 	}
979 
980 	free_cld_upcall(cup);
981 out_err:
982 	if (ret)
983 		printk(KERN_ERR "NFSD: Unable to remove client "
984 				"record from stable storage: %d\n", ret);
985 }
986 
987 /* Check for presence of a record, and update its timestamp */
988 static int
989 nfsd4_cld_check(struct nfs4_client *clp)
990 {
991 	int ret;
992 	struct cld_upcall *cup;
993 	struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
994 	struct cld_net *cn = nn->cld_net;
995 
996 	/* Don't upcall if one was already stored during this grace pd */
997 	if (test_bit(NFSD4_CLIENT_STABLE, &clp->cl_flags))
998 		return 0;
999 
1000 	cup = alloc_cld_upcall(cn);
1001 	if (!cup) {
1002 		printk(KERN_ERR "NFSD: Unable to check client record on "
1003 				"stable storage: %d\n", -ENOMEM);
1004 		return -ENOMEM;
1005 	}
1006 
1007 	cup->cu_msg.cm_cmd = Cld_Check;
1008 	cup->cu_msg.cm_u.cm_name.cn_len = clp->cl_name.len;
1009 	memcpy(cup->cu_msg.cm_u.cm_name.cn_id, clp->cl_name.data,
1010 			clp->cl_name.len);
1011 
1012 	ret = cld_pipe_upcall(cn->cn_pipe, &cup->cu_msg);
1013 	if (!ret) {
1014 		ret = cup->cu_msg.cm_status;
1015 		set_bit(NFSD4_CLIENT_STABLE, &clp->cl_flags);
1016 	}
1017 
1018 	free_cld_upcall(cup);
1019 	return ret;
1020 }
1021 
1022 static void
1023 nfsd4_cld_grace_done(struct nfsd_net *nn)
1024 {
1025 	int ret;
1026 	struct cld_upcall *cup;
1027 	struct cld_net *cn = nn->cld_net;
1028 
1029 	cup = alloc_cld_upcall(cn);
1030 	if (!cup) {
1031 		ret = -ENOMEM;
1032 		goto out_err;
1033 	}
1034 
1035 	cup->cu_msg.cm_cmd = Cld_GraceDone;
1036 	cup->cu_msg.cm_u.cm_gracetime = (int64_t)nn->boot_time;
1037 	ret = cld_pipe_upcall(cn->cn_pipe, &cup->cu_msg);
1038 	if (!ret)
1039 		ret = cup->cu_msg.cm_status;
1040 
1041 	free_cld_upcall(cup);
1042 out_err:
1043 	if (ret)
1044 		printk(KERN_ERR "NFSD: Unable to end grace period: %d\n", ret);
1045 }
1046 
1047 static struct nfsd4_client_tracking_ops nfsd4_cld_tracking_ops = {
1048 	.init		= nfsd4_init_cld_pipe,
1049 	.exit		= nfsd4_remove_cld_pipe,
1050 	.create		= nfsd4_cld_create,
1051 	.remove		= nfsd4_cld_remove,
1052 	.check		= nfsd4_cld_check,
1053 	.grace_done	= nfsd4_cld_grace_done,
1054 };
1055 
1056 /* upcall via usermodehelper */
1057 static char cltrack_prog[PATH_MAX] = "/sbin/nfsdcltrack";
1058 module_param_string(cltrack_prog, cltrack_prog, sizeof(cltrack_prog),
1059 			S_IRUGO|S_IWUSR);
1060 MODULE_PARM_DESC(cltrack_prog, "Path to the nfsdcltrack upcall program");
1061 
1062 static bool cltrack_legacy_disable;
1063 module_param(cltrack_legacy_disable, bool, S_IRUGO|S_IWUSR);
1064 MODULE_PARM_DESC(cltrack_legacy_disable,
1065 		"Disable legacy recoverydir conversion. Default: false");
1066 
1067 #define LEGACY_TOPDIR_ENV_PREFIX "NFSDCLTRACK_LEGACY_TOPDIR="
1068 #define LEGACY_RECDIR_ENV_PREFIX "NFSDCLTRACK_LEGACY_RECDIR="
1069 #define HAS_SESSION_ENV_PREFIX "NFSDCLTRACK_CLIENT_HAS_SESSION="
1070 #define GRACE_START_ENV_PREFIX "NFSDCLTRACK_GRACE_START="
1071 
1072 static char *
1073 nfsd4_cltrack_legacy_topdir(void)
1074 {
1075 	int copied;
1076 	size_t len;
1077 	char *result;
1078 
1079 	if (cltrack_legacy_disable)
1080 		return NULL;
1081 
1082 	len = strlen(LEGACY_TOPDIR_ENV_PREFIX) +
1083 		strlen(nfs4_recoverydir()) + 1;
1084 
1085 	result = kmalloc(len, GFP_KERNEL);
1086 	if (!result)
1087 		return result;
1088 
1089 	copied = snprintf(result, len, LEGACY_TOPDIR_ENV_PREFIX "%s",
1090 				nfs4_recoverydir());
1091 	if (copied >= len) {
1092 		/* just return nothing if output was truncated */
1093 		kfree(result);
1094 		return NULL;
1095 	}
1096 
1097 	return result;
1098 }
1099 
1100 static char *
1101 nfsd4_cltrack_legacy_recdir(const struct xdr_netobj *name)
1102 {
1103 	int copied;
1104 	size_t len;
1105 	char *result;
1106 
1107 	if (cltrack_legacy_disable)
1108 		return NULL;
1109 
1110 	/* +1 is for '/' between "topdir" and "recdir" */
1111 	len = strlen(LEGACY_RECDIR_ENV_PREFIX) +
1112 		strlen(nfs4_recoverydir()) + 1 + HEXDIR_LEN;
1113 
1114 	result = kmalloc(len, GFP_KERNEL);
1115 	if (!result)
1116 		return result;
1117 
1118 	copied = snprintf(result, len, LEGACY_RECDIR_ENV_PREFIX "%s/",
1119 				nfs4_recoverydir());
1120 	if (copied > (len - HEXDIR_LEN)) {
1121 		/* just return nothing if output will be truncated */
1122 		kfree(result);
1123 		return NULL;
1124 	}
1125 
1126 	copied = nfs4_make_rec_clidname(result + copied, name);
1127 	if (copied) {
1128 		kfree(result);
1129 		return NULL;
1130 	}
1131 
1132 	return result;
1133 }
1134 
1135 static char *
1136 nfsd4_cltrack_client_has_session(struct nfs4_client *clp)
1137 {
1138 	int copied;
1139 	size_t len;
1140 	char *result;
1141 
1142 	/* prefix + Y/N character + terminating NULL */
1143 	len = strlen(HAS_SESSION_ENV_PREFIX) + 1 + 1;
1144 
1145 	result = kmalloc(len, GFP_KERNEL);
1146 	if (!result)
1147 		return result;
1148 
1149 	copied = snprintf(result, len, HAS_SESSION_ENV_PREFIX "%c",
1150 				clp->cl_minorversion ? 'Y' : 'N');
1151 	if (copied >= len) {
1152 		/* just return nothing if output was truncated */
1153 		kfree(result);
1154 		return NULL;
1155 	}
1156 
1157 	return result;
1158 }
1159 
1160 static char *
1161 nfsd4_cltrack_grace_start(time_t grace_start)
1162 {
1163 	int copied;
1164 	size_t len;
1165 	char *result;
1166 
1167 	/* prefix + max width of int64_t string + terminating NULL */
1168 	len = strlen(GRACE_START_ENV_PREFIX) + 22 + 1;
1169 
1170 	result = kmalloc(len, GFP_KERNEL);
1171 	if (!result)
1172 		return result;
1173 
1174 	copied = snprintf(result, len, GRACE_START_ENV_PREFIX "%ld",
1175 				grace_start);
1176 	if (copied >= len) {
1177 		/* just return nothing if output was truncated */
1178 		kfree(result);
1179 		return NULL;
1180 	}
1181 
1182 	return result;
1183 }
1184 
1185 static int
1186 nfsd4_umh_cltrack_upcall(char *cmd, char *arg, char *env0, char *env1)
1187 {
1188 	char *envp[3];
1189 	char *argv[4];
1190 	int ret;
1191 
1192 	if (unlikely(!cltrack_prog[0])) {
1193 		dprintk("%s: cltrack_prog is disabled\n", __func__);
1194 		return -EACCES;
1195 	}
1196 
1197 	dprintk("%s: cmd: %s\n", __func__, cmd);
1198 	dprintk("%s: arg: %s\n", __func__, arg ? arg : "(null)");
1199 	dprintk("%s: env0: %s\n", __func__, env0 ? env0 : "(null)");
1200 	dprintk("%s: env1: %s\n", __func__, env1 ? env1 : "(null)");
1201 
1202 	envp[0] = env0;
1203 	envp[1] = env1;
1204 	envp[2] = NULL;
1205 
1206 	argv[0] = (char *)cltrack_prog;
1207 	argv[1] = cmd;
1208 	argv[2] = arg;
1209 	argv[3] = NULL;
1210 
1211 	ret = call_usermodehelper(argv[0], argv, envp, UMH_WAIT_PROC);
1212 	/*
1213 	 * Disable the upcall mechanism if we're getting an ENOENT or EACCES
1214 	 * error. The admin can re-enable it on the fly by using sysfs
1215 	 * once the problem has been fixed.
1216 	 */
1217 	if (ret == -ENOENT || ret == -EACCES) {
1218 		dprintk("NFSD: %s was not found or isn't executable (%d). "
1219 			"Setting cltrack_prog to blank string!",
1220 			cltrack_prog, ret);
1221 		cltrack_prog[0] = '\0';
1222 	}
1223 	dprintk("%s: %s return value: %d\n", __func__, cltrack_prog, ret);
1224 
1225 	return ret;
1226 }
1227 
1228 static char *
1229 bin_to_hex_dup(const unsigned char *src, int srclen)
1230 {
1231 	int i;
1232 	char *buf, *hex;
1233 
1234 	/* +1 for terminating NULL */
1235 	buf = kmalloc((srclen * 2) + 1, GFP_KERNEL);
1236 	if (!buf)
1237 		return buf;
1238 
1239 	hex = buf;
1240 	for (i = 0; i < srclen; i++) {
1241 		sprintf(hex, "%2.2x", *src++);
1242 		hex += 2;
1243 	}
1244 	return buf;
1245 }
1246 
1247 static int
1248 nfsd4_umh_cltrack_init(struct net *net)
1249 {
1250 	int ret;
1251 	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
1252 	char *grace_start = nfsd4_cltrack_grace_start(nn->boot_time);
1253 
1254 	/* XXX: The usermode helper s not working in container yet. */
1255 	if (net != &init_net) {
1256 		WARN(1, KERN_ERR "NFSD: attempt to initialize umh client "
1257 			"tracking in a container!\n");
1258 		return -EINVAL;
1259 	}
1260 
1261 	ret = nfsd4_umh_cltrack_upcall("init", NULL, grace_start, NULL);
1262 	kfree(grace_start);
1263 	return ret;
1264 }
1265 
1266 static void
1267 nfsd4_cltrack_upcall_lock(struct nfs4_client *clp)
1268 {
1269 	wait_on_bit_lock(&clp->cl_flags, NFSD4_CLIENT_UPCALL_LOCK,
1270 			 TASK_UNINTERRUPTIBLE);
1271 }
1272 
1273 static void
1274 nfsd4_cltrack_upcall_unlock(struct nfs4_client *clp)
1275 {
1276 	smp_mb__before_atomic();
1277 	clear_bit(NFSD4_CLIENT_UPCALL_LOCK, &clp->cl_flags);
1278 	smp_mb__after_atomic();
1279 	wake_up_bit(&clp->cl_flags, NFSD4_CLIENT_UPCALL_LOCK);
1280 }
1281 
1282 static void
1283 nfsd4_umh_cltrack_create(struct nfs4_client *clp)
1284 {
1285 	char *hexid, *has_session, *grace_start;
1286 	struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
1287 
1288 	/*
1289 	 * With v4.0 clients, there's little difference in outcome between a
1290 	 * create and check operation, and we can end up calling into this
1291 	 * function multiple times per client (once for each openowner). So,
1292 	 * for v4.0 clients skip upcalling once the client has been recorded
1293 	 * on stable storage.
1294 	 *
1295 	 * For v4.1+ clients, the outcome of the two operations is different,
1296 	 * so we must ensure that we upcall for the create operation. v4.1+
1297 	 * clients call this on RECLAIM_COMPLETE though, so we should only end
1298 	 * up doing a single create upcall per client.
1299 	 */
1300 	if (clp->cl_minorversion == 0 &&
1301 	    test_bit(NFSD4_CLIENT_STABLE, &clp->cl_flags))
1302 		return;
1303 
1304 	hexid = bin_to_hex_dup(clp->cl_name.data, clp->cl_name.len);
1305 	if (!hexid) {
1306 		dprintk("%s: can't allocate memory for upcall!\n", __func__);
1307 		return;
1308 	}
1309 
1310 	has_session = nfsd4_cltrack_client_has_session(clp);
1311 	grace_start = nfsd4_cltrack_grace_start(nn->boot_time);
1312 
1313 	nfsd4_cltrack_upcall_lock(clp);
1314 	if (!nfsd4_umh_cltrack_upcall("create", hexid, has_session, grace_start))
1315 		set_bit(NFSD4_CLIENT_STABLE, &clp->cl_flags);
1316 	nfsd4_cltrack_upcall_unlock(clp);
1317 
1318 	kfree(has_session);
1319 	kfree(grace_start);
1320 	kfree(hexid);
1321 }
1322 
1323 static void
1324 nfsd4_umh_cltrack_remove(struct nfs4_client *clp)
1325 {
1326 	char *hexid;
1327 
1328 	if (!test_bit(NFSD4_CLIENT_STABLE, &clp->cl_flags))
1329 		return;
1330 
1331 	hexid = bin_to_hex_dup(clp->cl_name.data, clp->cl_name.len);
1332 	if (!hexid) {
1333 		dprintk("%s: can't allocate memory for upcall!\n", __func__);
1334 		return;
1335 	}
1336 
1337 	nfsd4_cltrack_upcall_lock(clp);
1338 	if (test_bit(NFSD4_CLIENT_STABLE, &clp->cl_flags) &&
1339 	    nfsd4_umh_cltrack_upcall("remove", hexid, NULL, NULL) == 0)
1340 		clear_bit(NFSD4_CLIENT_STABLE, &clp->cl_flags);
1341 	nfsd4_cltrack_upcall_unlock(clp);
1342 
1343 	kfree(hexid);
1344 }
1345 
1346 static int
1347 nfsd4_umh_cltrack_check(struct nfs4_client *clp)
1348 {
1349 	int ret;
1350 	char *hexid, *has_session, *legacy;
1351 
1352 	if (test_bit(NFSD4_CLIENT_STABLE, &clp->cl_flags))
1353 		return 0;
1354 
1355 	hexid = bin_to_hex_dup(clp->cl_name.data, clp->cl_name.len);
1356 	if (!hexid) {
1357 		dprintk("%s: can't allocate memory for upcall!\n", __func__);
1358 		return -ENOMEM;
1359 	}
1360 
1361 	has_session = nfsd4_cltrack_client_has_session(clp);
1362 	legacy = nfsd4_cltrack_legacy_recdir(&clp->cl_name);
1363 
1364 	nfsd4_cltrack_upcall_lock(clp);
1365 	if (test_bit(NFSD4_CLIENT_STABLE, &clp->cl_flags)) {
1366 		ret = 0;
1367 	} else {
1368 		ret = nfsd4_umh_cltrack_upcall("check", hexid, has_session, legacy);
1369 		if (ret == 0)
1370 			set_bit(NFSD4_CLIENT_STABLE, &clp->cl_flags);
1371 	}
1372 	nfsd4_cltrack_upcall_unlock(clp);
1373 	kfree(has_session);
1374 	kfree(legacy);
1375 	kfree(hexid);
1376 
1377 	return ret;
1378 }
1379 
1380 static void
1381 nfsd4_umh_cltrack_grace_done(struct nfsd_net *nn)
1382 {
1383 	char *legacy;
1384 	char timestr[22]; /* FIXME: better way to determine max size? */
1385 
1386 	sprintf(timestr, "%ld", nn->boot_time);
1387 	legacy = nfsd4_cltrack_legacy_topdir();
1388 	nfsd4_umh_cltrack_upcall("gracedone", timestr, legacy, NULL);
1389 	kfree(legacy);
1390 }
1391 
1392 static struct nfsd4_client_tracking_ops nfsd4_umh_tracking_ops = {
1393 	.init		= nfsd4_umh_cltrack_init,
1394 	.exit		= NULL,
1395 	.create		= nfsd4_umh_cltrack_create,
1396 	.remove		= nfsd4_umh_cltrack_remove,
1397 	.check		= nfsd4_umh_cltrack_check,
1398 	.grace_done	= nfsd4_umh_cltrack_grace_done,
1399 };
1400 
1401 int
1402 nfsd4_client_tracking_init(struct net *net)
1403 {
1404 	int status;
1405 	struct path path;
1406 	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
1407 
1408 	/* just run the init if it the method is already decided */
1409 	if (nn->client_tracking_ops)
1410 		goto do_init;
1411 
1412 	/*
1413 	 * First, try a UMH upcall. It should succeed or fail quickly, so
1414 	 * there's little harm in trying that first.
1415 	 */
1416 	nn->client_tracking_ops = &nfsd4_umh_tracking_ops;
1417 	status = nn->client_tracking_ops->init(net);
1418 	if (!status)
1419 		return status;
1420 
1421 	/*
1422 	 * See if the recoverydir exists and is a directory. If it is,
1423 	 * then use the legacy ops.
1424 	 */
1425 	nn->client_tracking_ops = &nfsd4_legacy_tracking_ops;
1426 	status = kern_path(nfs4_recoverydir(), LOOKUP_FOLLOW, &path);
1427 	if (!status) {
1428 		status = S_ISDIR(path.dentry->d_inode->i_mode);
1429 		path_put(&path);
1430 		if (status)
1431 			goto do_init;
1432 	}
1433 
1434 	/* Finally, try to use nfsdcld */
1435 	nn->client_tracking_ops = &nfsd4_cld_tracking_ops;
1436 	printk(KERN_WARNING "NFSD: the nfsdcld client tracking upcall will be "
1437 			"removed in 3.10. Please transition to using "
1438 			"nfsdcltrack.\n");
1439 do_init:
1440 	status = nn->client_tracking_ops->init(net);
1441 	if (status) {
1442 		printk(KERN_WARNING "NFSD: Unable to initialize client "
1443 				    "recovery tracking! (%d)\n", status);
1444 		nn->client_tracking_ops = NULL;
1445 	}
1446 	return status;
1447 }
1448 
1449 void
1450 nfsd4_client_tracking_exit(struct net *net)
1451 {
1452 	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
1453 
1454 	if (nn->client_tracking_ops) {
1455 		if (nn->client_tracking_ops->exit)
1456 			nn->client_tracking_ops->exit(net);
1457 		nn->client_tracking_ops = NULL;
1458 	}
1459 }
1460 
1461 void
1462 nfsd4_client_record_create(struct nfs4_client *clp)
1463 {
1464 	struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
1465 
1466 	if (nn->client_tracking_ops)
1467 		nn->client_tracking_ops->create(clp);
1468 }
1469 
1470 void
1471 nfsd4_client_record_remove(struct nfs4_client *clp)
1472 {
1473 	struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
1474 
1475 	if (nn->client_tracking_ops)
1476 		nn->client_tracking_ops->remove(clp);
1477 }
1478 
1479 int
1480 nfsd4_client_record_check(struct nfs4_client *clp)
1481 {
1482 	struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
1483 
1484 	if (nn->client_tracking_ops)
1485 		return nn->client_tracking_ops->check(clp);
1486 
1487 	return -EOPNOTSUPP;
1488 }
1489 
1490 void
1491 nfsd4_record_grace_done(struct nfsd_net *nn)
1492 {
1493 	if (nn->client_tracking_ops)
1494 		nn->client_tracking_ops->grace_done(nn);
1495 }
1496 
1497 static int
1498 rpc_pipefs_event(struct notifier_block *nb, unsigned long event, void *ptr)
1499 {
1500 	struct super_block *sb = ptr;
1501 	struct net *net = sb->s_fs_info;
1502 	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
1503 	struct cld_net *cn = nn->cld_net;
1504 	struct dentry *dentry;
1505 	int ret = 0;
1506 
1507 	if (!try_module_get(THIS_MODULE))
1508 		return 0;
1509 
1510 	if (!cn) {
1511 		module_put(THIS_MODULE);
1512 		return 0;
1513 	}
1514 
1515 	switch (event) {
1516 	case RPC_PIPEFS_MOUNT:
1517 		dentry = nfsd4_cld_register_sb(sb, cn->cn_pipe);
1518 		if (IS_ERR(dentry)) {
1519 			ret = PTR_ERR(dentry);
1520 			break;
1521 		}
1522 		cn->cn_pipe->dentry = dentry;
1523 		break;
1524 	case RPC_PIPEFS_UMOUNT:
1525 		if (cn->cn_pipe->dentry)
1526 			nfsd4_cld_unregister_sb(cn->cn_pipe);
1527 		break;
1528 	default:
1529 		ret = -ENOTSUPP;
1530 		break;
1531 	}
1532 	module_put(THIS_MODULE);
1533 	return ret;
1534 }
1535 
1536 static struct notifier_block nfsd4_cld_block = {
1537 	.notifier_call = rpc_pipefs_event,
1538 };
1539 
1540 int
1541 register_cld_notifier(void)
1542 {
1543 	return rpc_pipefs_notifier_register(&nfsd4_cld_block);
1544 }
1545 
1546 void
1547 unregister_cld_notifier(void)
1548 {
1549 	rpc_pipefs_notifier_unregister(&nfsd4_cld_block);
1550 }
1551