xref: /linux/fs/nfsd/nfs4recover.c (revision 2ba9268dd603d23e17643437b2246acb6844953b)
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(struct dir_context *__ctx, const char *name, int namlen,
249 		loff_t offset, u64 ino, unsigned int d_type)
250 {
251 	struct nfs4_dir_ctx *ctx =
252 		container_of(__ctx, struct nfs4_dir_ctx, ctx);
253 	struct name_list *entry;
254 
255 	if (namlen != HEXDIR_LEN - 1)
256 		return 0;
257 	entry = kmalloc(sizeof(struct name_list), GFP_KERNEL);
258 	if (entry == NULL)
259 		return -ENOMEM;
260 	memcpy(entry->name, name, HEXDIR_LEN - 1);
261 	entry->name[HEXDIR_LEN - 1] = '\0';
262 	list_add(&entry->list, &ctx->names);
263 	return 0;
264 }
265 
266 static int
267 nfsd4_list_rec_dir(recdir_func *f, struct nfsd_net *nn)
268 {
269 	const struct cred *original_cred;
270 	struct dentry *dir = nn->rec_file->f_path.dentry;
271 	struct nfs4_dir_ctx ctx = {
272 		.ctx.actor = nfsd4_build_namelist,
273 		.names = LIST_HEAD_INIT(ctx.names)
274 	};
275 	int status;
276 
277 	status = nfs4_save_creds(&original_cred);
278 	if (status < 0)
279 		return status;
280 
281 	status = vfs_llseek(nn->rec_file, 0, SEEK_SET);
282 	if (status < 0) {
283 		nfs4_reset_creds(original_cred);
284 		return status;
285 	}
286 
287 	status = iterate_dir(nn->rec_file, &ctx.ctx);
288 	mutex_lock_nested(&dir->d_inode->i_mutex, I_MUTEX_PARENT);
289 	while (!list_empty(&ctx.names)) {
290 		struct name_list *entry;
291 		entry = list_entry(ctx.names.next, struct name_list, list);
292 		if (!status) {
293 			struct dentry *dentry;
294 			dentry = lookup_one_len(entry->name, dir, HEXDIR_LEN-1);
295 			if (IS_ERR(dentry)) {
296 				status = PTR_ERR(dentry);
297 				break;
298 			}
299 			status = f(dir, dentry, nn);
300 			dput(dentry);
301 		}
302 		list_del(&entry->list);
303 		kfree(entry);
304 	}
305 	mutex_unlock(&dir->d_inode->i_mutex);
306 	nfs4_reset_creds(original_cred);
307 	return status;
308 }
309 
310 static int
311 nfsd4_unlink_clid_dir(char *name, int namlen, struct nfsd_net *nn)
312 {
313 	struct dentry *dir, *dentry;
314 	int status;
315 
316 	dprintk("NFSD: nfsd4_unlink_clid_dir. name %.*s\n", namlen, name);
317 
318 	dir = nn->rec_file->f_path.dentry;
319 	mutex_lock_nested(&dir->d_inode->i_mutex, I_MUTEX_PARENT);
320 	dentry = lookup_one_len(name, dir, namlen);
321 	if (IS_ERR(dentry)) {
322 		status = PTR_ERR(dentry);
323 		goto out_unlock;
324 	}
325 	status = -ENOENT;
326 	if (!dentry->d_inode)
327 		goto out;
328 	status = vfs_rmdir(dir->d_inode, dentry);
329 out:
330 	dput(dentry);
331 out_unlock:
332 	mutex_unlock(&dir->d_inode->i_mutex);
333 	return status;
334 }
335 
336 static void
337 nfsd4_remove_clid_dir(struct nfs4_client *clp)
338 {
339 	const struct cred *original_cred;
340 	struct nfs4_client_reclaim *crp;
341 	char dname[HEXDIR_LEN];
342 	int status;
343 	struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
344 
345 	if (!nn->rec_file || !test_bit(NFSD4_CLIENT_STABLE, &clp->cl_flags))
346 		return;
347 
348 	status = nfs4_make_rec_clidname(dname, &clp->cl_name);
349 	if (status)
350 		return legacy_recdir_name_error(clp, status);
351 
352 	status = mnt_want_write_file(nn->rec_file);
353 	if (status)
354 		goto out;
355 	clear_bit(NFSD4_CLIENT_STABLE, &clp->cl_flags);
356 
357 	status = nfs4_save_creds(&original_cred);
358 	if (status < 0)
359 		goto out_drop_write;
360 
361 	status = nfsd4_unlink_clid_dir(dname, HEXDIR_LEN-1, nn);
362 	nfs4_reset_creds(original_cred);
363 	if (status == 0) {
364 		vfs_fsync(nn->rec_file, 0);
365 		if (nn->in_grace) {
366 			/* remove reclaim record */
367 			crp = nfsd4_find_reclaim_client(dname, nn);
368 			if (crp)
369 				nfs4_remove_reclaim_record(crp, nn);
370 		}
371 	}
372 out_drop_write:
373 	mnt_drop_write_file(nn->rec_file);
374 out:
375 	if (status)
376 		printk("NFSD: Failed to remove expired client state directory"
377 				" %.*s\n", HEXDIR_LEN, dname);
378 }
379 
380 static int
381 purge_old(struct dentry *parent, struct dentry *child, struct nfsd_net *nn)
382 {
383 	int status;
384 
385 	if (nfs4_has_reclaimed_state(child->d_name.name, nn))
386 		return 0;
387 
388 	status = vfs_rmdir(parent->d_inode, child);
389 	if (status)
390 		printk("failed to remove client recovery directory %pd\n",
391 				child);
392 	/* Keep trying, success or failure: */
393 	return 0;
394 }
395 
396 static void
397 nfsd4_recdir_purge_old(struct nfsd_net *nn)
398 {
399 	int status;
400 
401 	nn->in_grace = false;
402 	if (!nn->rec_file)
403 		return;
404 	status = mnt_want_write_file(nn->rec_file);
405 	if (status)
406 		goto out;
407 	status = nfsd4_list_rec_dir(purge_old, nn);
408 	if (status == 0)
409 		vfs_fsync(nn->rec_file, 0);
410 	mnt_drop_write_file(nn->rec_file);
411 out:
412 	nfs4_release_reclaim(nn);
413 	if (status)
414 		printk("nfsd4: failed to purge old clients from recovery"
415 			" directory %pD\n", nn->rec_file);
416 }
417 
418 static int
419 load_recdir(struct dentry *parent, struct dentry *child, struct nfsd_net *nn)
420 {
421 	if (child->d_name.len != HEXDIR_LEN - 1) {
422 		printk("nfsd4: illegal name %pd in recovery directory\n",
423 				child);
424 		/* Keep trying; maybe the others are OK: */
425 		return 0;
426 	}
427 	nfs4_client_to_reclaim(child->d_name.name, nn);
428 	return 0;
429 }
430 
431 static int
432 nfsd4_recdir_load(struct net *net) {
433 	int status;
434 	struct nfsd_net *nn =  net_generic(net, nfsd_net_id);
435 
436 	if (!nn->rec_file)
437 		return 0;
438 
439 	status = nfsd4_list_rec_dir(load_recdir, nn);
440 	if (status)
441 		printk("nfsd4: failed loading clients from recovery"
442 			" directory %pD\n", nn->rec_file);
443 	return status;
444 }
445 
446 /*
447  * Hold reference to the recovery directory.
448  */
449 
450 static int
451 nfsd4_init_recdir(struct net *net)
452 {
453 	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
454 	const struct cred *original_cred;
455 	int status;
456 
457 	printk("NFSD: Using %s as the NFSv4 state recovery directory\n",
458 			user_recovery_dirname);
459 
460 	BUG_ON(nn->rec_file);
461 
462 	status = nfs4_save_creds(&original_cred);
463 	if (status < 0) {
464 		printk("NFSD: Unable to change credentials to find recovery"
465 		       " directory: error %d\n",
466 		       status);
467 		return status;
468 	}
469 
470 	nn->rec_file = filp_open(user_recovery_dirname, O_RDONLY | O_DIRECTORY, 0);
471 	if (IS_ERR(nn->rec_file)) {
472 		printk("NFSD: unable to find recovery directory %s\n",
473 				user_recovery_dirname);
474 		status = PTR_ERR(nn->rec_file);
475 		nn->rec_file = NULL;
476 	}
477 
478 	nfs4_reset_creds(original_cred);
479 	if (!status)
480 		nn->in_grace = true;
481 	return status;
482 }
483 
484 static void
485 nfsd4_shutdown_recdir(struct net *net)
486 {
487 	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
488 
489 	if (!nn->rec_file)
490 		return;
491 	fput(nn->rec_file);
492 	nn->rec_file = NULL;
493 }
494 
495 static int
496 nfs4_legacy_state_init(struct net *net)
497 {
498 	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
499 	int i;
500 
501 	nn->reclaim_str_hashtbl = kmalloc(sizeof(struct list_head) *
502 					  CLIENT_HASH_SIZE, GFP_KERNEL);
503 	if (!nn->reclaim_str_hashtbl)
504 		return -ENOMEM;
505 
506 	for (i = 0; i < CLIENT_HASH_SIZE; i++)
507 		INIT_LIST_HEAD(&nn->reclaim_str_hashtbl[i]);
508 	nn->reclaim_str_hashtbl_size = 0;
509 
510 	return 0;
511 }
512 
513 static void
514 nfs4_legacy_state_shutdown(struct net *net)
515 {
516 	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
517 
518 	kfree(nn->reclaim_str_hashtbl);
519 }
520 
521 static int
522 nfsd4_load_reboot_recovery_data(struct net *net)
523 {
524 	int status;
525 
526 	status = nfsd4_init_recdir(net);
527 	if (status)
528 		return status;
529 
530 	status = nfsd4_recdir_load(net);
531 	if (status)
532 		nfsd4_shutdown_recdir(net);
533 
534 	return status;
535 }
536 
537 static int
538 nfsd4_legacy_tracking_init(struct net *net)
539 {
540 	int status;
541 
542 	/* XXX: The legacy code won't work in a container */
543 	if (net != &init_net) {
544 		WARN(1, KERN_ERR "NFSD: attempt to initialize legacy client "
545 			"tracking in a container!\n");
546 		return -EINVAL;
547 	}
548 
549 	status = nfs4_legacy_state_init(net);
550 	if (status)
551 		return status;
552 
553 	status = nfsd4_load_reboot_recovery_data(net);
554 	if (status)
555 		goto err;
556 	return 0;
557 
558 err:
559 	nfs4_legacy_state_shutdown(net);
560 	return status;
561 }
562 
563 static void
564 nfsd4_legacy_tracking_exit(struct net *net)
565 {
566 	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
567 
568 	nfs4_release_reclaim(nn);
569 	nfsd4_shutdown_recdir(net);
570 	nfs4_legacy_state_shutdown(net);
571 }
572 
573 /*
574  * Change the NFSv4 recovery directory to recdir.
575  */
576 int
577 nfs4_reset_recoverydir(char *recdir)
578 {
579 	int status;
580 	struct path path;
581 
582 	status = kern_path(recdir, LOOKUP_FOLLOW, &path);
583 	if (status)
584 		return status;
585 	status = -ENOTDIR;
586 	if (d_is_dir(path.dentry)) {
587 		strcpy(user_recovery_dirname, recdir);
588 		status = 0;
589 	}
590 	path_put(&path);
591 	return status;
592 }
593 
594 char *
595 nfs4_recoverydir(void)
596 {
597 	return user_recovery_dirname;
598 }
599 
600 static int
601 nfsd4_check_legacy_client(struct nfs4_client *clp)
602 {
603 	int status;
604 	char dname[HEXDIR_LEN];
605 	struct nfs4_client_reclaim *crp;
606 	struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
607 
608 	/* did we already find that this client is stable? */
609 	if (test_bit(NFSD4_CLIENT_STABLE, &clp->cl_flags))
610 		return 0;
611 
612 	status = nfs4_make_rec_clidname(dname, &clp->cl_name);
613 	if (status) {
614 		legacy_recdir_name_error(clp, status);
615 		return status;
616 	}
617 
618 	/* look for it in the reclaim hashtable otherwise */
619 	crp = nfsd4_find_reclaim_client(dname, nn);
620 	if (crp) {
621 		set_bit(NFSD4_CLIENT_STABLE, &clp->cl_flags);
622 		crp->cr_clp = clp;
623 		return 0;
624 	}
625 
626 	return -ENOENT;
627 }
628 
629 static struct nfsd4_client_tracking_ops nfsd4_legacy_tracking_ops = {
630 	.init		= nfsd4_legacy_tracking_init,
631 	.exit		= nfsd4_legacy_tracking_exit,
632 	.create		= nfsd4_create_clid_dir,
633 	.remove		= nfsd4_remove_clid_dir,
634 	.check		= nfsd4_check_legacy_client,
635 	.grace_done	= nfsd4_recdir_purge_old,
636 };
637 
638 /* Globals */
639 #define NFSD_PIPE_DIR		"nfsd"
640 #define NFSD_CLD_PIPE		"cld"
641 
642 /* per-net-ns structure for holding cld upcall info */
643 struct cld_net {
644 	struct rpc_pipe		*cn_pipe;
645 	spinlock_t		 cn_lock;
646 	struct list_head	 cn_list;
647 	unsigned int		 cn_xid;
648 };
649 
650 struct cld_upcall {
651 	struct list_head	 cu_list;
652 	struct cld_net		*cu_net;
653 	struct task_struct	*cu_task;
654 	struct cld_msg		 cu_msg;
655 };
656 
657 static int
658 __cld_pipe_upcall(struct rpc_pipe *pipe, struct cld_msg *cmsg)
659 {
660 	int ret;
661 	struct rpc_pipe_msg msg;
662 
663 	memset(&msg, 0, sizeof(msg));
664 	msg.data = cmsg;
665 	msg.len = sizeof(*cmsg);
666 
667 	/*
668 	 * Set task state before we queue the upcall. That prevents
669 	 * wake_up_process in the downcall from racing with schedule.
670 	 */
671 	set_current_state(TASK_UNINTERRUPTIBLE);
672 	ret = rpc_queue_upcall(pipe, &msg);
673 	if (ret < 0) {
674 		set_current_state(TASK_RUNNING);
675 		goto out;
676 	}
677 
678 	schedule();
679 
680 	if (msg.errno < 0)
681 		ret = msg.errno;
682 out:
683 	return ret;
684 }
685 
686 static int
687 cld_pipe_upcall(struct rpc_pipe *pipe, struct cld_msg *cmsg)
688 {
689 	int ret;
690 
691 	/*
692 	 * -EAGAIN occurs when pipe is closed and reopened while there are
693 	 *  upcalls queued.
694 	 */
695 	do {
696 		ret = __cld_pipe_upcall(pipe, cmsg);
697 	} while (ret == -EAGAIN);
698 
699 	return ret;
700 }
701 
702 static ssize_t
703 cld_pipe_downcall(struct file *filp, const char __user *src, size_t mlen)
704 {
705 	struct cld_upcall *tmp, *cup;
706 	struct cld_msg __user *cmsg = (struct cld_msg __user *)src;
707 	uint32_t xid;
708 	struct nfsd_net *nn = net_generic(file_inode(filp)->i_sb->s_fs_info,
709 						nfsd_net_id);
710 	struct cld_net *cn = nn->cld_net;
711 
712 	if (mlen != sizeof(*cmsg)) {
713 		dprintk("%s: got %zu bytes, expected %zu\n", __func__, mlen,
714 			sizeof(*cmsg));
715 		return -EINVAL;
716 	}
717 
718 	/* copy just the xid so we can try to find that */
719 	if (copy_from_user(&xid, &cmsg->cm_xid, sizeof(xid)) != 0) {
720 		dprintk("%s: error when copying xid from userspace", __func__);
721 		return -EFAULT;
722 	}
723 
724 	/* walk the list and find corresponding xid */
725 	cup = NULL;
726 	spin_lock(&cn->cn_lock);
727 	list_for_each_entry(tmp, &cn->cn_list, cu_list) {
728 		if (get_unaligned(&tmp->cu_msg.cm_xid) == xid) {
729 			cup = tmp;
730 			list_del_init(&cup->cu_list);
731 			break;
732 		}
733 	}
734 	spin_unlock(&cn->cn_lock);
735 
736 	/* couldn't find upcall? */
737 	if (!cup) {
738 		dprintk("%s: couldn't find upcall -- xid=%u\n", __func__, xid);
739 		return -EINVAL;
740 	}
741 
742 	if (copy_from_user(&cup->cu_msg, src, mlen) != 0)
743 		return -EFAULT;
744 
745 	wake_up_process(cup->cu_task);
746 	return mlen;
747 }
748 
749 static void
750 cld_pipe_destroy_msg(struct rpc_pipe_msg *msg)
751 {
752 	struct cld_msg *cmsg = msg->data;
753 	struct cld_upcall *cup = container_of(cmsg, struct cld_upcall,
754 						 cu_msg);
755 
756 	/* errno >= 0 means we got a downcall */
757 	if (msg->errno >= 0)
758 		return;
759 
760 	wake_up_process(cup->cu_task);
761 }
762 
763 static const struct rpc_pipe_ops cld_upcall_ops = {
764 	.upcall		= rpc_pipe_generic_upcall,
765 	.downcall	= cld_pipe_downcall,
766 	.destroy_msg	= cld_pipe_destroy_msg,
767 };
768 
769 static struct dentry *
770 nfsd4_cld_register_sb(struct super_block *sb, struct rpc_pipe *pipe)
771 {
772 	struct dentry *dir, *dentry;
773 
774 	dir = rpc_d_lookup_sb(sb, NFSD_PIPE_DIR);
775 	if (dir == NULL)
776 		return ERR_PTR(-ENOENT);
777 	dentry = rpc_mkpipe_dentry(dir, NFSD_CLD_PIPE, NULL, pipe);
778 	dput(dir);
779 	return dentry;
780 }
781 
782 static void
783 nfsd4_cld_unregister_sb(struct rpc_pipe *pipe)
784 {
785 	if (pipe->dentry)
786 		rpc_unlink(pipe->dentry);
787 }
788 
789 static struct dentry *
790 nfsd4_cld_register_net(struct net *net, struct rpc_pipe *pipe)
791 {
792 	struct super_block *sb;
793 	struct dentry *dentry;
794 
795 	sb = rpc_get_sb_net(net);
796 	if (!sb)
797 		return NULL;
798 	dentry = nfsd4_cld_register_sb(sb, pipe);
799 	rpc_put_sb_net(net);
800 	return dentry;
801 }
802 
803 static void
804 nfsd4_cld_unregister_net(struct net *net, struct rpc_pipe *pipe)
805 {
806 	struct super_block *sb;
807 
808 	sb = rpc_get_sb_net(net);
809 	if (sb) {
810 		nfsd4_cld_unregister_sb(pipe);
811 		rpc_put_sb_net(net);
812 	}
813 }
814 
815 /* Initialize rpc_pipefs pipe for communication with client tracking daemon */
816 static int
817 nfsd4_init_cld_pipe(struct net *net)
818 {
819 	int ret;
820 	struct dentry *dentry;
821 	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
822 	struct cld_net *cn;
823 
824 	if (nn->cld_net)
825 		return 0;
826 
827 	cn = kzalloc(sizeof(*cn), GFP_KERNEL);
828 	if (!cn) {
829 		ret = -ENOMEM;
830 		goto err;
831 	}
832 
833 	cn->cn_pipe = rpc_mkpipe_data(&cld_upcall_ops, RPC_PIPE_WAIT_FOR_OPEN);
834 	if (IS_ERR(cn->cn_pipe)) {
835 		ret = PTR_ERR(cn->cn_pipe);
836 		goto err;
837 	}
838 	spin_lock_init(&cn->cn_lock);
839 	INIT_LIST_HEAD(&cn->cn_list);
840 
841 	dentry = nfsd4_cld_register_net(net, cn->cn_pipe);
842 	if (IS_ERR(dentry)) {
843 		ret = PTR_ERR(dentry);
844 		goto err_destroy_data;
845 	}
846 
847 	cn->cn_pipe->dentry = dentry;
848 	nn->cld_net = cn;
849 	return 0;
850 
851 err_destroy_data:
852 	rpc_destroy_pipe_data(cn->cn_pipe);
853 err:
854 	kfree(cn);
855 	printk(KERN_ERR "NFSD: unable to create nfsdcld upcall pipe (%d)\n",
856 			ret);
857 	return ret;
858 }
859 
860 static void
861 nfsd4_remove_cld_pipe(struct net *net)
862 {
863 	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
864 	struct cld_net *cn = nn->cld_net;
865 
866 	nfsd4_cld_unregister_net(net, cn->cn_pipe);
867 	rpc_destroy_pipe_data(cn->cn_pipe);
868 	kfree(nn->cld_net);
869 	nn->cld_net = NULL;
870 }
871 
872 static struct cld_upcall *
873 alloc_cld_upcall(struct cld_net *cn)
874 {
875 	struct cld_upcall *new, *tmp;
876 
877 	new = kzalloc(sizeof(*new), GFP_KERNEL);
878 	if (!new)
879 		return new;
880 
881 	/* FIXME: hard cap on number in flight? */
882 restart_search:
883 	spin_lock(&cn->cn_lock);
884 	list_for_each_entry(tmp, &cn->cn_list, cu_list) {
885 		if (tmp->cu_msg.cm_xid == cn->cn_xid) {
886 			cn->cn_xid++;
887 			spin_unlock(&cn->cn_lock);
888 			goto restart_search;
889 		}
890 	}
891 	new->cu_task = current;
892 	new->cu_msg.cm_vers = CLD_UPCALL_VERSION;
893 	put_unaligned(cn->cn_xid++, &new->cu_msg.cm_xid);
894 	new->cu_net = cn;
895 	list_add(&new->cu_list, &cn->cn_list);
896 	spin_unlock(&cn->cn_lock);
897 
898 	dprintk("%s: allocated xid %u\n", __func__, new->cu_msg.cm_xid);
899 
900 	return new;
901 }
902 
903 static void
904 free_cld_upcall(struct cld_upcall *victim)
905 {
906 	struct cld_net *cn = victim->cu_net;
907 
908 	spin_lock(&cn->cn_lock);
909 	list_del(&victim->cu_list);
910 	spin_unlock(&cn->cn_lock);
911 	kfree(victim);
912 }
913 
914 /* Ask daemon to create a new record */
915 static void
916 nfsd4_cld_create(struct nfs4_client *clp)
917 {
918 	int ret;
919 	struct cld_upcall *cup;
920 	struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
921 	struct cld_net *cn = nn->cld_net;
922 
923 	/* Don't upcall if it's already stored */
924 	if (test_bit(NFSD4_CLIENT_STABLE, &clp->cl_flags))
925 		return;
926 
927 	cup = alloc_cld_upcall(cn);
928 	if (!cup) {
929 		ret = -ENOMEM;
930 		goto out_err;
931 	}
932 
933 	cup->cu_msg.cm_cmd = Cld_Create;
934 	cup->cu_msg.cm_u.cm_name.cn_len = clp->cl_name.len;
935 	memcpy(cup->cu_msg.cm_u.cm_name.cn_id, clp->cl_name.data,
936 			clp->cl_name.len);
937 
938 	ret = cld_pipe_upcall(cn->cn_pipe, &cup->cu_msg);
939 	if (!ret) {
940 		ret = cup->cu_msg.cm_status;
941 		set_bit(NFSD4_CLIENT_STABLE, &clp->cl_flags);
942 	}
943 
944 	free_cld_upcall(cup);
945 out_err:
946 	if (ret)
947 		printk(KERN_ERR "NFSD: Unable to create client "
948 				"record on stable storage: %d\n", ret);
949 }
950 
951 /* Ask daemon to create a new record */
952 static void
953 nfsd4_cld_remove(struct nfs4_client *clp)
954 {
955 	int ret;
956 	struct cld_upcall *cup;
957 	struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
958 	struct cld_net *cn = nn->cld_net;
959 
960 	/* Don't upcall if it's already removed */
961 	if (!test_bit(NFSD4_CLIENT_STABLE, &clp->cl_flags))
962 		return;
963 
964 	cup = alloc_cld_upcall(cn);
965 	if (!cup) {
966 		ret = -ENOMEM;
967 		goto out_err;
968 	}
969 
970 	cup->cu_msg.cm_cmd = Cld_Remove;
971 	cup->cu_msg.cm_u.cm_name.cn_len = clp->cl_name.len;
972 	memcpy(cup->cu_msg.cm_u.cm_name.cn_id, clp->cl_name.data,
973 			clp->cl_name.len);
974 
975 	ret = cld_pipe_upcall(cn->cn_pipe, &cup->cu_msg);
976 	if (!ret) {
977 		ret = cup->cu_msg.cm_status;
978 		clear_bit(NFSD4_CLIENT_STABLE, &clp->cl_flags);
979 	}
980 
981 	free_cld_upcall(cup);
982 out_err:
983 	if (ret)
984 		printk(KERN_ERR "NFSD: Unable to remove client "
985 				"record from stable storage: %d\n", ret);
986 }
987 
988 /* Check for presence of a record, and update its timestamp */
989 static int
990 nfsd4_cld_check(struct nfs4_client *clp)
991 {
992 	int ret;
993 	struct cld_upcall *cup;
994 	struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
995 	struct cld_net *cn = nn->cld_net;
996 
997 	/* Don't upcall if one was already stored during this grace pd */
998 	if (test_bit(NFSD4_CLIENT_STABLE, &clp->cl_flags))
999 		return 0;
1000 
1001 	cup = alloc_cld_upcall(cn);
1002 	if (!cup) {
1003 		printk(KERN_ERR "NFSD: Unable to check client record on "
1004 				"stable storage: %d\n", -ENOMEM);
1005 		return -ENOMEM;
1006 	}
1007 
1008 	cup->cu_msg.cm_cmd = Cld_Check;
1009 	cup->cu_msg.cm_u.cm_name.cn_len = clp->cl_name.len;
1010 	memcpy(cup->cu_msg.cm_u.cm_name.cn_id, clp->cl_name.data,
1011 			clp->cl_name.len);
1012 
1013 	ret = cld_pipe_upcall(cn->cn_pipe, &cup->cu_msg);
1014 	if (!ret) {
1015 		ret = cup->cu_msg.cm_status;
1016 		set_bit(NFSD4_CLIENT_STABLE, &clp->cl_flags);
1017 	}
1018 
1019 	free_cld_upcall(cup);
1020 	return ret;
1021 }
1022 
1023 static void
1024 nfsd4_cld_grace_done(struct nfsd_net *nn)
1025 {
1026 	int ret;
1027 	struct cld_upcall *cup;
1028 	struct cld_net *cn = nn->cld_net;
1029 
1030 	cup = alloc_cld_upcall(cn);
1031 	if (!cup) {
1032 		ret = -ENOMEM;
1033 		goto out_err;
1034 	}
1035 
1036 	cup->cu_msg.cm_cmd = Cld_GraceDone;
1037 	cup->cu_msg.cm_u.cm_gracetime = (int64_t)nn->boot_time;
1038 	ret = cld_pipe_upcall(cn->cn_pipe, &cup->cu_msg);
1039 	if (!ret)
1040 		ret = cup->cu_msg.cm_status;
1041 
1042 	free_cld_upcall(cup);
1043 out_err:
1044 	if (ret)
1045 		printk(KERN_ERR "NFSD: Unable to end grace period: %d\n", ret);
1046 }
1047 
1048 static struct nfsd4_client_tracking_ops nfsd4_cld_tracking_ops = {
1049 	.init		= nfsd4_init_cld_pipe,
1050 	.exit		= nfsd4_remove_cld_pipe,
1051 	.create		= nfsd4_cld_create,
1052 	.remove		= nfsd4_cld_remove,
1053 	.check		= nfsd4_cld_check,
1054 	.grace_done	= nfsd4_cld_grace_done,
1055 };
1056 
1057 /* upcall via usermodehelper */
1058 static char cltrack_prog[PATH_MAX] = "/sbin/nfsdcltrack";
1059 module_param_string(cltrack_prog, cltrack_prog, sizeof(cltrack_prog),
1060 			S_IRUGO|S_IWUSR);
1061 MODULE_PARM_DESC(cltrack_prog, "Path to the nfsdcltrack upcall program");
1062 
1063 static bool cltrack_legacy_disable;
1064 module_param(cltrack_legacy_disable, bool, S_IRUGO|S_IWUSR);
1065 MODULE_PARM_DESC(cltrack_legacy_disable,
1066 		"Disable legacy recoverydir conversion. Default: false");
1067 
1068 #define LEGACY_TOPDIR_ENV_PREFIX "NFSDCLTRACK_LEGACY_TOPDIR="
1069 #define LEGACY_RECDIR_ENV_PREFIX "NFSDCLTRACK_LEGACY_RECDIR="
1070 #define HAS_SESSION_ENV_PREFIX "NFSDCLTRACK_CLIENT_HAS_SESSION="
1071 #define GRACE_START_ENV_PREFIX "NFSDCLTRACK_GRACE_START="
1072 
1073 static char *
1074 nfsd4_cltrack_legacy_topdir(void)
1075 {
1076 	int copied;
1077 	size_t len;
1078 	char *result;
1079 
1080 	if (cltrack_legacy_disable)
1081 		return NULL;
1082 
1083 	len = strlen(LEGACY_TOPDIR_ENV_PREFIX) +
1084 		strlen(nfs4_recoverydir()) + 1;
1085 
1086 	result = kmalloc(len, GFP_KERNEL);
1087 	if (!result)
1088 		return result;
1089 
1090 	copied = snprintf(result, len, LEGACY_TOPDIR_ENV_PREFIX "%s",
1091 				nfs4_recoverydir());
1092 	if (copied >= len) {
1093 		/* just return nothing if output was truncated */
1094 		kfree(result);
1095 		return NULL;
1096 	}
1097 
1098 	return result;
1099 }
1100 
1101 static char *
1102 nfsd4_cltrack_legacy_recdir(const struct xdr_netobj *name)
1103 {
1104 	int copied;
1105 	size_t len;
1106 	char *result;
1107 
1108 	if (cltrack_legacy_disable)
1109 		return NULL;
1110 
1111 	/* +1 is for '/' between "topdir" and "recdir" */
1112 	len = strlen(LEGACY_RECDIR_ENV_PREFIX) +
1113 		strlen(nfs4_recoverydir()) + 1 + HEXDIR_LEN;
1114 
1115 	result = kmalloc(len, GFP_KERNEL);
1116 	if (!result)
1117 		return result;
1118 
1119 	copied = snprintf(result, len, LEGACY_RECDIR_ENV_PREFIX "%s/",
1120 				nfs4_recoverydir());
1121 	if (copied > (len - HEXDIR_LEN)) {
1122 		/* just return nothing if output will be truncated */
1123 		kfree(result);
1124 		return NULL;
1125 	}
1126 
1127 	copied = nfs4_make_rec_clidname(result + copied, name);
1128 	if (copied) {
1129 		kfree(result);
1130 		return NULL;
1131 	}
1132 
1133 	return result;
1134 }
1135 
1136 static char *
1137 nfsd4_cltrack_client_has_session(struct nfs4_client *clp)
1138 {
1139 	int copied;
1140 	size_t len;
1141 	char *result;
1142 
1143 	/* prefix + Y/N character + terminating NULL */
1144 	len = strlen(HAS_SESSION_ENV_PREFIX) + 1 + 1;
1145 
1146 	result = kmalloc(len, GFP_KERNEL);
1147 	if (!result)
1148 		return result;
1149 
1150 	copied = snprintf(result, len, HAS_SESSION_ENV_PREFIX "%c",
1151 				clp->cl_minorversion ? 'Y' : 'N');
1152 	if (copied >= len) {
1153 		/* just return nothing if output was truncated */
1154 		kfree(result);
1155 		return NULL;
1156 	}
1157 
1158 	return result;
1159 }
1160 
1161 static char *
1162 nfsd4_cltrack_grace_start(time_t grace_start)
1163 {
1164 	int copied;
1165 	size_t len;
1166 	char *result;
1167 
1168 	/* prefix + max width of int64_t string + terminating NULL */
1169 	len = strlen(GRACE_START_ENV_PREFIX) + 22 + 1;
1170 
1171 	result = kmalloc(len, GFP_KERNEL);
1172 	if (!result)
1173 		return result;
1174 
1175 	copied = snprintf(result, len, GRACE_START_ENV_PREFIX "%ld",
1176 				grace_start);
1177 	if (copied >= len) {
1178 		/* just return nothing if output was truncated */
1179 		kfree(result);
1180 		return NULL;
1181 	}
1182 
1183 	return result;
1184 }
1185 
1186 static int
1187 nfsd4_umh_cltrack_upcall(char *cmd, char *arg, char *env0, char *env1)
1188 {
1189 	char *envp[3];
1190 	char *argv[4];
1191 	int ret;
1192 
1193 	if (unlikely(!cltrack_prog[0])) {
1194 		dprintk("%s: cltrack_prog is disabled\n", __func__);
1195 		return -EACCES;
1196 	}
1197 
1198 	dprintk("%s: cmd: %s\n", __func__, cmd);
1199 	dprintk("%s: arg: %s\n", __func__, arg ? arg : "(null)");
1200 	dprintk("%s: env0: %s\n", __func__, env0 ? env0 : "(null)");
1201 	dprintk("%s: env1: %s\n", __func__, env1 ? env1 : "(null)");
1202 
1203 	envp[0] = env0;
1204 	envp[1] = env1;
1205 	envp[2] = NULL;
1206 
1207 	argv[0] = (char *)cltrack_prog;
1208 	argv[1] = cmd;
1209 	argv[2] = arg;
1210 	argv[3] = NULL;
1211 
1212 	ret = call_usermodehelper(argv[0], argv, envp, UMH_WAIT_PROC);
1213 	/*
1214 	 * Disable the upcall mechanism if we're getting an ENOENT or EACCES
1215 	 * error. The admin can re-enable it on the fly by using sysfs
1216 	 * once the problem has been fixed.
1217 	 */
1218 	if (ret == -ENOENT || ret == -EACCES) {
1219 		dprintk("NFSD: %s was not found or isn't executable (%d). "
1220 			"Setting cltrack_prog to blank string!",
1221 			cltrack_prog, ret);
1222 		cltrack_prog[0] = '\0';
1223 	}
1224 	dprintk("%s: %s return value: %d\n", __func__, cltrack_prog, ret);
1225 
1226 	return ret;
1227 }
1228 
1229 static char *
1230 bin_to_hex_dup(const unsigned char *src, int srclen)
1231 {
1232 	int i;
1233 	char *buf, *hex;
1234 
1235 	/* +1 for terminating NULL */
1236 	buf = kmalloc((srclen * 2) + 1, GFP_KERNEL);
1237 	if (!buf)
1238 		return buf;
1239 
1240 	hex = buf;
1241 	for (i = 0; i < srclen; i++) {
1242 		sprintf(hex, "%2.2x", *src++);
1243 		hex += 2;
1244 	}
1245 	return buf;
1246 }
1247 
1248 static int
1249 nfsd4_umh_cltrack_init(struct net *net)
1250 {
1251 	int ret;
1252 	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
1253 	char *grace_start = nfsd4_cltrack_grace_start(nn->boot_time);
1254 
1255 	/* XXX: The usermode helper s not working in container yet. */
1256 	if (net != &init_net) {
1257 		WARN(1, KERN_ERR "NFSD: attempt to initialize umh client "
1258 			"tracking in a container!\n");
1259 		return -EINVAL;
1260 	}
1261 
1262 	ret = nfsd4_umh_cltrack_upcall("init", NULL, grace_start, NULL);
1263 	kfree(grace_start);
1264 	return ret;
1265 }
1266 
1267 static void
1268 nfsd4_cltrack_upcall_lock(struct nfs4_client *clp)
1269 {
1270 	wait_on_bit_lock(&clp->cl_flags, NFSD4_CLIENT_UPCALL_LOCK,
1271 			 TASK_UNINTERRUPTIBLE);
1272 }
1273 
1274 static void
1275 nfsd4_cltrack_upcall_unlock(struct nfs4_client *clp)
1276 {
1277 	smp_mb__before_atomic();
1278 	clear_bit(NFSD4_CLIENT_UPCALL_LOCK, &clp->cl_flags);
1279 	smp_mb__after_atomic();
1280 	wake_up_bit(&clp->cl_flags, NFSD4_CLIENT_UPCALL_LOCK);
1281 }
1282 
1283 static void
1284 nfsd4_umh_cltrack_create(struct nfs4_client *clp)
1285 {
1286 	char *hexid, *has_session, *grace_start;
1287 	struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
1288 
1289 	/*
1290 	 * With v4.0 clients, there's little difference in outcome between a
1291 	 * create and check operation, and we can end up calling into this
1292 	 * function multiple times per client (once for each openowner). So,
1293 	 * for v4.0 clients skip upcalling once the client has been recorded
1294 	 * on stable storage.
1295 	 *
1296 	 * For v4.1+ clients, the outcome of the two operations is different,
1297 	 * so we must ensure that we upcall for the create operation. v4.1+
1298 	 * clients call this on RECLAIM_COMPLETE though, so we should only end
1299 	 * up doing a single create upcall per client.
1300 	 */
1301 	if (clp->cl_minorversion == 0 &&
1302 	    test_bit(NFSD4_CLIENT_STABLE, &clp->cl_flags))
1303 		return;
1304 
1305 	hexid = bin_to_hex_dup(clp->cl_name.data, clp->cl_name.len);
1306 	if (!hexid) {
1307 		dprintk("%s: can't allocate memory for upcall!\n", __func__);
1308 		return;
1309 	}
1310 
1311 	has_session = nfsd4_cltrack_client_has_session(clp);
1312 	grace_start = nfsd4_cltrack_grace_start(nn->boot_time);
1313 
1314 	nfsd4_cltrack_upcall_lock(clp);
1315 	if (!nfsd4_umh_cltrack_upcall("create", hexid, has_session, grace_start))
1316 		set_bit(NFSD4_CLIENT_STABLE, &clp->cl_flags);
1317 	nfsd4_cltrack_upcall_unlock(clp);
1318 
1319 	kfree(has_session);
1320 	kfree(grace_start);
1321 	kfree(hexid);
1322 }
1323 
1324 static void
1325 nfsd4_umh_cltrack_remove(struct nfs4_client *clp)
1326 {
1327 	char *hexid;
1328 
1329 	if (!test_bit(NFSD4_CLIENT_STABLE, &clp->cl_flags))
1330 		return;
1331 
1332 	hexid = bin_to_hex_dup(clp->cl_name.data, clp->cl_name.len);
1333 	if (!hexid) {
1334 		dprintk("%s: can't allocate memory for upcall!\n", __func__);
1335 		return;
1336 	}
1337 
1338 	nfsd4_cltrack_upcall_lock(clp);
1339 	if (test_bit(NFSD4_CLIENT_STABLE, &clp->cl_flags) &&
1340 	    nfsd4_umh_cltrack_upcall("remove", hexid, NULL, NULL) == 0)
1341 		clear_bit(NFSD4_CLIENT_STABLE, &clp->cl_flags);
1342 	nfsd4_cltrack_upcall_unlock(clp);
1343 
1344 	kfree(hexid);
1345 }
1346 
1347 static int
1348 nfsd4_umh_cltrack_check(struct nfs4_client *clp)
1349 {
1350 	int ret;
1351 	char *hexid, *has_session, *legacy;
1352 
1353 	if (test_bit(NFSD4_CLIENT_STABLE, &clp->cl_flags))
1354 		return 0;
1355 
1356 	hexid = bin_to_hex_dup(clp->cl_name.data, clp->cl_name.len);
1357 	if (!hexid) {
1358 		dprintk("%s: can't allocate memory for upcall!\n", __func__);
1359 		return -ENOMEM;
1360 	}
1361 
1362 	has_session = nfsd4_cltrack_client_has_session(clp);
1363 	legacy = nfsd4_cltrack_legacy_recdir(&clp->cl_name);
1364 
1365 	nfsd4_cltrack_upcall_lock(clp);
1366 	if (test_bit(NFSD4_CLIENT_STABLE, &clp->cl_flags)) {
1367 		ret = 0;
1368 	} else {
1369 		ret = nfsd4_umh_cltrack_upcall("check", hexid, has_session, legacy);
1370 		if (ret == 0)
1371 			set_bit(NFSD4_CLIENT_STABLE, &clp->cl_flags);
1372 	}
1373 	nfsd4_cltrack_upcall_unlock(clp);
1374 	kfree(has_session);
1375 	kfree(legacy);
1376 	kfree(hexid);
1377 
1378 	return ret;
1379 }
1380 
1381 static void
1382 nfsd4_umh_cltrack_grace_done(struct nfsd_net *nn)
1383 {
1384 	char *legacy;
1385 	char timestr[22]; /* FIXME: better way to determine max size? */
1386 
1387 	sprintf(timestr, "%ld", nn->boot_time);
1388 	legacy = nfsd4_cltrack_legacy_topdir();
1389 	nfsd4_umh_cltrack_upcall("gracedone", timestr, legacy, NULL);
1390 	kfree(legacy);
1391 }
1392 
1393 static struct nfsd4_client_tracking_ops nfsd4_umh_tracking_ops = {
1394 	.init		= nfsd4_umh_cltrack_init,
1395 	.exit		= NULL,
1396 	.create		= nfsd4_umh_cltrack_create,
1397 	.remove		= nfsd4_umh_cltrack_remove,
1398 	.check		= nfsd4_umh_cltrack_check,
1399 	.grace_done	= nfsd4_umh_cltrack_grace_done,
1400 };
1401 
1402 int
1403 nfsd4_client_tracking_init(struct net *net)
1404 {
1405 	int status;
1406 	struct path path;
1407 	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
1408 
1409 	/* just run the init if it the method is already decided */
1410 	if (nn->client_tracking_ops)
1411 		goto do_init;
1412 
1413 	/*
1414 	 * First, try a UMH upcall. It should succeed or fail quickly, so
1415 	 * there's little harm in trying that first.
1416 	 */
1417 	nn->client_tracking_ops = &nfsd4_umh_tracking_ops;
1418 	status = nn->client_tracking_ops->init(net);
1419 	if (!status)
1420 		return status;
1421 
1422 	/*
1423 	 * See if the recoverydir exists and is a directory. If it is,
1424 	 * then use the legacy ops.
1425 	 */
1426 	nn->client_tracking_ops = &nfsd4_legacy_tracking_ops;
1427 	status = kern_path(nfs4_recoverydir(), LOOKUP_FOLLOW, &path);
1428 	if (!status) {
1429 		status = d_is_dir(path.dentry);
1430 		path_put(&path);
1431 		if (status)
1432 			goto do_init;
1433 	}
1434 
1435 	/* Finally, try to use nfsdcld */
1436 	nn->client_tracking_ops = &nfsd4_cld_tracking_ops;
1437 	printk(KERN_WARNING "NFSD: the nfsdcld client tracking upcall will be "
1438 			"removed in 3.10. Please transition to using "
1439 			"nfsdcltrack.\n");
1440 do_init:
1441 	status = nn->client_tracking_ops->init(net);
1442 	if (status) {
1443 		printk(KERN_WARNING "NFSD: Unable to initialize client "
1444 				    "recovery tracking! (%d)\n", status);
1445 		nn->client_tracking_ops = NULL;
1446 	}
1447 	return status;
1448 }
1449 
1450 void
1451 nfsd4_client_tracking_exit(struct net *net)
1452 {
1453 	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
1454 
1455 	if (nn->client_tracking_ops) {
1456 		if (nn->client_tracking_ops->exit)
1457 			nn->client_tracking_ops->exit(net);
1458 		nn->client_tracking_ops = NULL;
1459 	}
1460 }
1461 
1462 void
1463 nfsd4_client_record_create(struct nfs4_client *clp)
1464 {
1465 	struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
1466 
1467 	if (nn->client_tracking_ops)
1468 		nn->client_tracking_ops->create(clp);
1469 }
1470 
1471 void
1472 nfsd4_client_record_remove(struct nfs4_client *clp)
1473 {
1474 	struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
1475 
1476 	if (nn->client_tracking_ops)
1477 		nn->client_tracking_ops->remove(clp);
1478 }
1479 
1480 int
1481 nfsd4_client_record_check(struct nfs4_client *clp)
1482 {
1483 	struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
1484 
1485 	if (nn->client_tracking_ops)
1486 		return nn->client_tracking_ops->check(clp);
1487 
1488 	return -EOPNOTSUPP;
1489 }
1490 
1491 void
1492 nfsd4_record_grace_done(struct nfsd_net *nn)
1493 {
1494 	if (nn->client_tracking_ops)
1495 		nn->client_tracking_ops->grace_done(nn);
1496 }
1497 
1498 static int
1499 rpc_pipefs_event(struct notifier_block *nb, unsigned long event, void *ptr)
1500 {
1501 	struct super_block *sb = ptr;
1502 	struct net *net = sb->s_fs_info;
1503 	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
1504 	struct cld_net *cn = nn->cld_net;
1505 	struct dentry *dentry;
1506 	int ret = 0;
1507 
1508 	if (!try_module_get(THIS_MODULE))
1509 		return 0;
1510 
1511 	if (!cn) {
1512 		module_put(THIS_MODULE);
1513 		return 0;
1514 	}
1515 
1516 	switch (event) {
1517 	case RPC_PIPEFS_MOUNT:
1518 		dentry = nfsd4_cld_register_sb(sb, cn->cn_pipe);
1519 		if (IS_ERR(dentry)) {
1520 			ret = PTR_ERR(dentry);
1521 			break;
1522 		}
1523 		cn->cn_pipe->dentry = dentry;
1524 		break;
1525 	case RPC_PIPEFS_UMOUNT:
1526 		if (cn->cn_pipe->dentry)
1527 			nfsd4_cld_unregister_sb(cn->cn_pipe);
1528 		break;
1529 	default:
1530 		ret = -ENOTSUPP;
1531 		break;
1532 	}
1533 	module_put(THIS_MODULE);
1534 	return ret;
1535 }
1536 
1537 static struct notifier_block nfsd4_cld_block = {
1538 	.notifier_call = rpc_pipefs_event,
1539 };
1540 
1541 int
1542 register_cld_notifier(void)
1543 {
1544 	return rpc_pipefs_notifier_register(&nfsd4_cld_block);
1545 }
1546 
1547 void
1548 unregister_cld_notifier(void)
1549 {
1550 	rpc_pipefs_notifier_unregister(&nfsd4_cld_block);
1551 }
1552