xref: /linux/fs/nfsd/nfs4recover.c (revision e6f2a617ac53bc0753b885ffb94379ff48b2e2df)
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 <crypto/hash.h>
36 #include <linux/file.h>
37 #include <linux/slab.h>
38 #include <linux/namei.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 	uint8_t version;
63 	size_t msglen;
64 };
65 
66 static const struct nfsd4_client_tracking_ops nfsd4_cld_tracking_ops;
67 static const struct nfsd4_client_tracking_ops nfsd4_cld_tracking_ops_v2;
68 
69 /* Globals */
70 static char user_recovery_dirname[PATH_MAX] = "/var/lib/nfs/v4recovery";
71 
72 static int
73 nfs4_save_creds(const struct cred **original_creds)
74 {
75 	struct cred *new;
76 
77 	new = prepare_creds();
78 	if (!new)
79 		return -ENOMEM;
80 
81 	new->fsuid = GLOBAL_ROOT_UID;
82 	new->fsgid = GLOBAL_ROOT_GID;
83 	*original_creds = override_creds(new);
84 	put_cred(new);
85 	return 0;
86 }
87 
88 static void
89 nfs4_reset_creds(const struct cred *original)
90 {
91 	revert_creds(original);
92 }
93 
94 static void
95 md5_to_hex(char *out, char *md5)
96 {
97 	int i;
98 
99 	for (i=0; i<16; i++) {
100 		unsigned char c = md5[i];
101 
102 		*out++ = '0' + ((c&0xf0)>>4) + (c>=0xa0)*('a'-'9'-1);
103 		*out++ = '0' + (c&0x0f) + ((c&0x0f)>=0x0a)*('a'-'9'-1);
104 	}
105 	*out = '\0';
106 }
107 
108 static int
109 nfs4_make_rec_clidname(char *dname, const struct xdr_netobj *clname)
110 {
111 	struct xdr_netobj cksum;
112 	struct crypto_shash *tfm;
113 	int status;
114 
115 	dprintk("NFSD: nfs4_make_rec_clidname for %.*s\n",
116 			clname->len, clname->data);
117 	tfm = crypto_alloc_shash("md5", 0, 0);
118 	if (IS_ERR(tfm)) {
119 		status = PTR_ERR(tfm);
120 		goto out_no_tfm;
121 	}
122 
123 	cksum.len = crypto_shash_digestsize(tfm);
124 	cksum.data = kmalloc(cksum.len, GFP_KERNEL);
125 	if (cksum.data == NULL) {
126 		status = -ENOMEM;
127  		goto out;
128 	}
129 
130 	{
131 		SHASH_DESC_ON_STACK(desc, tfm);
132 
133 		desc->tfm = tfm;
134 
135 		status = crypto_shash_digest(desc, clname->data, clname->len,
136 					     cksum.data);
137 		shash_desc_zero(desc);
138 	}
139 
140 	if (status)
141 		goto out;
142 
143 	md5_to_hex(dname, cksum.data);
144 
145 	status = 0;
146 out:
147 	kfree(cksum.data);
148 	crypto_free_shash(tfm);
149 out_no_tfm:
150 	return status;
151 }
152 
153 /*
154  * If we had an error generating the recdir name for the legacy tracker
155  * then warn the admin. If the error doesn't appear to be transient,
156  * then disable recovery tracking.
157  */
158 static void
159 legacy_recdir_name_error(struct nfs4_client *clp, int error)
160 {
161 	printk(KERN_ERR "NFSD: unable to generate recoverydir "
162 			"name (%d).\n", error);
163 
164 	/*
165 	 * if the algorithm just doesn't exist, then disable the recovery
166 	 * tracker altogether. The crypto libs will generally return this if
167 	 * FIPS is enabled as well.
168 	 */
169 	if (error == -ENOENT) {
170 		printk(KERN_ERR "NFSD: disabling legacy clientid tracking. "
171 			"Reboot recovery will not function correctly!\n");
172 		nfsd4_client_tracking_exit(clp->net);
173 	}
174 }
175 
176 static void
177 __nfsd4_create_reclaim_record_grace(struct nfs4_client *clp,
178 		const char *dname, int len, struct nfsd_net *nn)
179 {
180 	struct xdr_netobj name;
181 	struct xdr_netobj princhash = { .len = 0, .data = NULL };
182 	struct nfs4_client_reclaim *crp;
183 
184 	name.data = kmemdup(dname, len, GFP_KERNEL);
185 	if (!name.data) {
186 		dprintk("%s: failed to allocate memory for name.data!\n",
187 			__func__);
188 		return;
189 	}
190 	name.len = len;
191 	crp = nfs4_client_to_reclaim(name, princhash, nn);
192 	if (!crp) {
193 		kfree(name.data);
194 		return;
195 	}
196 	crp->cr_clp = clp;
197 }
198 
199 static void
200 nfsd4_create_clid_dir(struct nfs4_client *clp)
201 {
202 	const struct cred *original_cred;
203 	char dname[HEXDIR_LEN];
204 	struct dentry *dir, *dentry;
205 	int status;
206 	struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
207 
208 	if (test_and_set_bit(NFSD4_CLIENT_STABLE, &clp->cl_flags))
209 		return;
210 	if (!nn->rec_file)
211 		return;
212 
213 	status = nfs4_make_rec_clidname(dname, &clp->cl_name);
214 	if (status)
215 		return legacy_recdir_name_error(clp, status);
216 
217 	status = nfs4_save_creds(&original_cred);
218 	if (status < 0)
219 		return;
220 
221 	status = mnt_want_write_file(nn->rec_file);
222 	if (status)
223 		goto out_creds;
224 
225 	dir = nn->rec_file->f_path.dentry;
226 	/* lock the parent */
227 	inode_lock(d_inode(dir));
228 
229 	dentry = lookup_one_len(dname, dir, HEXDIR_LEN-1);
230 	if (IS_ERR(dentry)) {
231 		status = PTR_ERR(dentry);
232 		goto out_unlock;
233 	}
234 	if (d_really_is_positive(dentry))
235 		/*
236 		 * In the 4.1 case, where we're called from
237 		 * reclaim_complete(), records from the previous reboot
238 		 * may still be left, so this is OK.
239 		 *
240 		 * In the 4.0 case, we should never get here; but we may
241 		 * as well be forgiving and just succeed silently.
242 		 */
243 		goto out_put;
244 	status = vfs_mkdir(d_inode(dir), dentry, S_IRWXU);
245 out_put:
246 	dput(dentry);
247 out_unlock:
248 	inode_unlock(d_inode(dir));
249 	if (status == 0) {
250 		if (nn->in_grace)
251 			__nfsd4_create_reclaim_record_grace(clp, dname,
252 					HEXDIR_LEN, nn);
253 		vfs_fsync(nn->rec_file, 0);
254 	} else {
255 		printk(KERN_ERR "NFSD: failed to write recovery record"
256 				" (err %d); please check that %s exists"
257 				" and is writeable", status,
258 				user_recovery_dirname);
259 	}
260 	mnt_drop_write_file(nn->rec_file);
261 out_creds:
262 	nfs4_reset_creds(original_cred);
263 }
264 
265 typedef int (recdir_func)(struct dentry *, struct dentry *, struct nfsd_net *);
266 
267 struct name_list {
268 	char name[HEXDIR_LEN];
269 	struct list_head list;
270 };
271 
272 struct nfs4_dir_ctx {
273 	struct dir_context ctx;
274 	struct list_head names;
275 };
276 
277 static int
278 nfsd4_build_namelist(struct dir_context *__ctx, const char *name, int namlen,
279 		loff_t offset, u64 ino, unsigned int d_type)
280 {
281 	struct nfs4_dir_ctx *ctx =
282 		container_of(__ctx, struct nfs4_dir_ctx, ctx);
283 	struct name_list *entry;
284 
285 	if (namlen != HEXDIR_LEN - 1)
286 		return 0;
287 	entry = kmalloc(sizeof(struct name_list), GFP_KERNEL);
288 	if (entry == NULL)
289 		return -ENOMEM;
290 	memcpy(entry->name, name, HEXDIR_LEN - 1);
291 	entry->name[HEXDIR_LEN - 1] = '\0';
292 	list_add(&entry->list, &ctx->names);
293 	return 0;
294 }
295 
296 static int
297 nfsd4_list_rec_dir(recdir_func *f, struct nfsd_net *nn)
298 {
299 	const struct cred *original_cred;
300 	struct dentry *dir = nn->rec_file->f_path.dentry;
301 	struct nfs4_dir_ctx ctx = {
302 		.ctx.actor = nfsd4_build_namelist,
303 		.names = LIST_HEAD_INIT(ctx.names)
304 	};
305 	struct name_list *entry, *tmp;
306 	int status;
307 
308 	status = nfs4_save_creds(&original_cred);
309 	if (status < 0)
310 		return status;
311 
312 	status = vfs_llseek(nn->rec_file, 0, SEEK_SET);
313 	if (status < 0) {
314 		nfs4_reset_creds(original_cred);
315 		return status;
316 	}
317 
318 	status = iterate_dir(nn->rec_file, &ctx.ctx);
319 	inode_lock_nested(d_inode(dir), I_MUTEX_PARENT);
320 
321 	list_for_each_entry_safe(entry, tmp, &ctx.names, list) {
322 		if (!status) {
323 			struct dentry *dentry;
324 			dentry = lookup_one_len(entry->name, dir, HEXDIR_LEN-1);
325 			if (IS_ERR(dentry)) {
326 				status = PTR_ERR(dentry);
327 				break;
328 			}
329 			status = f(dir, dentry, nn);
330 			dput(dentry);
331 		}
332 		list_del(&entry->list);
333 		kfree(entry);
334 	}
335 	inode_unlock(d_inode(dir));
336 	nfs4_reset_creds(original_cred);
337 
338 	list_for_each_entry_safe(entry, tmp, &ctx.names, list) {
339 		dprintk("NFSD: %s. Left entry %s\n", __func__, entry->name);
340 		list_del(&entry->list);
341 		kfree(entry);
342 	}
343 	return status;
344 }
345 
346 static int
347 nfsd4_unlink_clid_dir(char *name, int namlen, struct nfsd_net *nn)
348 {
349 	struct dentry *dir, *dentry;
350 	int status;
351 
352 	dprintk("NFSD: nfsd4_unlink_clid_dir. name %.*s\n", namlen, name);
353 
354 	dir = nn->rec_file->f_path.dentry;
355 	inode_lock_nested(d_inode(dir), I_MUTEX_PARENT);
356 	dentry = lookup_one_len(name, dir, namlen);
357 	if (IS_ERR(dentry)) {
358 		status = PTR_ERR(dentry);
359 		goto out_unlock;
360 	}
361 	status = -ENOENT;
362 	if (d_really_is_negative(dentry))
363 		goto out;
364 	status = vfs_rmdir(d_inode(dir), dentry);
365 out:
366 	dput(dentry);
367 out_unlock:
368 	inode_unlock(d_inode(dir));
369 	return status;
370 }
371 
372 static void
373 __nfsd4_remove_reclaim_record_grace(const char *dname, int len,
374 		struct nfsd_net *nn)
375 {
376 	struct xdr_netobj name;
377 	struct nfs4_client_reclaim *crp;
378 
379 	name.data = kmemdup(dname, len, GFP_KERNEL);
380 	if (!name.data) {
381 		dprintk("%s: failed to allocate memory for name.data!\n",
382 			__func__);
383 		return;
384 	}
385 	name.len = len;
386 	crp = nfsd4_find_reclaim_client(name, nn);
387 	kfree(name.data);
388 	if (crp)
389 		nfs4_remove_reclaim_record(crp, nn);
390 }
391 
392 static void
393 nfsd4_remove_clid_dir(struct nfs4_client *clp)
394 {
395 	const struct cred *original_cred;
396 	char dname[HEXDIR_LEN];
397 	int status;
398 	struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
399 
400 	if (!nn->rec_file || !test_bit(NFSD4_CLIENT_STABLE, &clp->cl_flags))
401 		return;
402 
403 	status = nfs4_make_rec_clidname(dname, &clp->cl_name);
404 	if (status)
405 		return legacy_recdir_name_error(clp, status);
406 
407 	status = mnt_want_write_file(nn->rec_file);
408 	if (status)
409 		goto out;
410 	clear_bit(NFSD4_CLIENT_STABLE, &clp->cl_flags);
411 
412 	status = nfs4_save_creds(&original_cred);
413 	if (status < 0)
414 		goto out_drop_write;
415 
416 	status = nfsd4_unlink_clid_dir(dname, HEXDIR_LEN-1, nn);
417 	nfs4_reset_creds(original_cred);
418 	if (status == 0) {
419 		vfs_fsync(nn->rec_file, 0);
420 		if (nn->in_grace)
421 			__nfsd4_remove_reclaim_record_grace(dname,
422 					HEXDIR_LEN, nn);
423 	}
424 out_drop_write:
425 	mnt_drop_write_file(nn->rec_file);
426 out:
427 	if (status)
428 		printk("NFSD: Failed to remove expired client state directory"
429 				" %.*s\n", HEXDIR_LEN, dname);
430 }
431 
432 static int
433 purge_old(struct dentry *parent, struct dentry *child, struct nfsd_net *nn)
434 {
435 	int status;
436 	struct xdr_netobj name;
437 
438 	if (child->d_name.len != HEXDIR_LEN - 1) {
439 		printk("%s: illegal name %pd in recovery directory\n",
440 				__func__, child);
441 		/* Keep trying; maybe the others are OK: */
442 		return 0;
443 	}
444 	name.data = kmemdup_nul(child->d_name.name, child->d_name.len, GFP_KERNEL);
445 	if (!name.data) {
446 		dprintk("%s: failed to allocate memory for name.data!\n",
447 			__func__);
448 		goto out;
449 	}
450 	name.len = HEXDIR_LEN;
451 	if (nfs4_has_reclaimed_state(name, nn))
452 		goto out_free;
453 
454 	status = vfs_rmdir(d_inode(parent), child);
455 	if (status)
456 		printk("failed to remove client recovery directory %pd\n",
457 				child);
458 out_free:
459 	kfree(name.data);
460 out:
461 	/* Keep trying, success or failure: */
462 	return 0;
463 }
464 
465 static void
466 nfsd4_recdir_purge_old(struct nfsd_net *nn)
467 {
468 	int status;
469 
470 	nn->in_grace = false;
471 	if (!nn->rec_file)
472 		return;
473 	status = mnt_want_write_file(nn->rec_file);
474 	if (status)
475 		goto out;
476 	status = nfsd4_list_rec_dir(purge_old, nn);
477 	if (status == 0)
478 		vfs_fsync(nn->rec_file, 0);
479 	mnt_drop_write_file(nn->rec_file);
480 out:
481 	nfs4_release_reclaim(nn);
482 	if (status)
483 		printk("nfsd4: failed to purge old clients from recovery"
484 			" directory %pD\n", nn->rec_file);
485 }
486 
487 static int
488 load_recdir(struct dentry *parent, struct dentry *child, struct nfsd_net *nn)
489 {
490 	struct xdr_netobj name;
491 	struct xdr_netobj princhash = { .len = 0, .data = NULL };
492 
493 	if (child->d_name.len != HEXDIR_LEN - 1) {
494 		printk("%s: illegal name %pd in recovery directory\n",
495 				__func__, child);
496 		/* Keep trying; maybe the others are OK: */
497 		return 0;
498 	}
499 	name.data = kmemdup_nul(child->d_name.name, child->d_name.len, GFP_KERNEL);
500 	if (!name.data) {
501 		dprintk("%s: failed to allocate memory for name.data!\n",
502 			__func__);
503 		goto out;
504 	}
505 	name.len = HEXDIR_LEN;
506 	if (!nfs4_client_to_reclaim(name, princhash, nn))
507 		kfree(name.data);
508 out:
509 	return 0;
510 }
511 
512 static int
513 nfsd4_recdir_load(struct net *net) {
514 	int status;
515 	struct nfsd_net *nn =  net_generic(net, nfsd_net_id);
516 
517 	if (!nn->rec_file)
518 		return 0;
519 
520 	status = nfsd4_list_rec_dir(load_recdir, nn);
521 	if (status)
522 		printk("nfsd4: failed loading clients from recovery"
523 			" directory %pD\n", nn->rec_file);
524 	return status;
525 }
526 
527 /*
528  * Hold reference to the recovery directory.
529  */
530 
531 static int
532 nfsd4_init_recdir(struct net *net)
533 {
534 	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
535 	const struct cred *original_cred;
536 	int status;
537 
538 	printk("NFSD: Using %s as the NFSv4 state recovery directory\n",
539 			user_recovery_dirname);
540 
541 	BUG_ON(nn->rec_file);
542 
543 	status = nfs4_save_creds(&original_cred);
544 	if (status < 0) {
545 		printk("NFSD: Unable to change credentials to find recovery"
546 		       " directory: error %d\n",
547 		       status);
548 		return status;
549 	}
550 
551 	nn->rec_file = filp_open(user_recovery_dirname, O_RDONLY | O_DIRECTORY, 0);
552 	if (IS_ERR(nn->rec_file)) {
553 		printk("NFSD: unable to find recovery directory %s\n",
554 				user_recovery_dirname);
555 		status = PTR_ERR(nn->rec_file);
556 		nn->rec_file = NULL;
557 	}
558 
559 	nfs4_reset_creds(original_cred);
560 	if (!status)
561 		nn->in_grace = true;
562 	return status;
563 }
564 
565 static void
566 nfsd4_shutdown_recdir(struct net *net)
567 {
568 	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
569 
570 	if (!nn->rec_file)
571 		return;
572 	fput(nn->rec_file);
573 	nn->rec_file = NULL;
574 }
575 
576 static int
577 nfs4_legacy_state_init(struct net *net)
578 {
579 	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
580 	int i;
581 
582 	nn->reclaim_str_hashtbl = kmalloc_array(CLIENT_HASH_SIZE,
583 						sizeof(struct list_head),
584 						GFP_KERNEL);
585 	if (!nn->reclaim_str_hashtbl)
586 		return -ENOMEM;
587 
588 	for (i = 0; i < CLIENT_HASH_SIZE; i++)
589 		INIT_LIST_HEAD(&nn->reclaim_str_hashtbl[i]);
590 	nn->reclaim_str_hashtbl_size = 0;
591 
592 	return 0;
593 }
594 
595 static void
596 nfs4_legacy_state_shutdown(struct net *net)
597 {
598 	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
599 
600 	kfree(nn->reclaim_str_hashtbl);
601 }
602 
603 static int
604 nfsd4_load_reboot_recovery_data(struct net *net)
605 {
606 	int status;
607 
608 	status = nfsd4_init_recdir(net);
609 	if (status)
610 		return status;
611 
612 	status = nfsd4_recdir_load(net);
613 	if (status)
614 		nfsd4_shutdown_recdir(net);
615 
616 	return status;
617 }
618 
619 static int
620 nfsd4_legacy_tracking_init(struct net *net)
621 {
622 	int status;
623 
624 	/* XXX: The legacy code won't work in a container */
625 	if (net != &init_net) {
626 		pr_warn("NFSD: attempt to initialize legacy client tracking in a container ignored.\n");
627 		return -EINVAL;
628 	}
629 
630 	status = nfs4_legacy_state_init(net);
631 	if (status)
632 		return status;
633 
634 	status = nfsd4_load_reboot_recovery_data(net);
635 	if (status)
636 		goto err;
637 	printk("NFSD: Using legacy client tracking operations.\n");
638 	return 0;
639 
640 err:
641 	nfs4_legacy_state_shutdown(net);
642 	return status;
643 }
644 
645 static void
646 nfsd4_legacy_tracking_exit(struct net *net)
647 {
648 	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
649 
650 	nfs4_release_reclaim(nn);
651 	nfsd4_shutdown_recdir(net);
652 	nfs4_legacy_state_shutdown(net);
653 }
654 
655 /*
656  * Change the NFSv4 recovery directory to recdir.
657  */
658 int
659 nfs4_reset_recoverydir(char *recdir)
660 {
661 	int status;
662 	struct path path;
663 
664 	status = kern_path(recdir, LOOKUP_FOLLOW, &path);
665 	if (status)
666 		return status;
667 	status = -ENOTDIR;
668 	if (d_is_dir(path.dentry)) {
669 		strcpy(user_recovery_dirname, recdir);
670 		status = 0;
671 	}
672 	path_put(&path);
673 	return status;
674 }
675 
676 char *
677 nfs4_recoverydir(void)
678 {
679 	return user_recovery_dirname;
680 }
681 
682 static int
683 nfsd4_check_legacy_client(struct nfs4_client *clp)
684 {
685 	int status;
686 	char dname[HEXDIR_LEN];
687 	struct nfs4_client_reclaim *crp;
688 	struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
689 	struct xdr_netobj name;
690 
691 	/* did we already find that this client is stable? */
692 	if (test_bit(NFSD4_CLIENT_STABLE, &clp->cl_flags))
693 		return 0;
694 
695 	status = nfs4_make_rec_clidname(dname, &clp->cl_name);
696 	if (status) {
697 		legacy_recdir_name_error(clp, status);
698 		return status;
699 	}
700 
701 	/* look for it in the reclaim hashtable otherwise */
702 	name.data = kmemdup(dname, HEXDIR_LEN, GFP_KERNEL);
703 	if (!name.data) {
704 		dprintk("%s: failed to allocate memory for name.data!\n",
705 			__func__);
706 		goto out_enoent;
707 	}
708 	name.len = HEXDIR_LEN;
709 	crp = nfsd4_find_reclaim_client(name, nn);
710 	kfree(name.data);
711 	if (crp) {
712 		set_bit(NFSD4_CLIENT_STABLE, &clp->cl_flags);
713 		crp->cr_clp = clp;
714 		return 0;
715 	}
716 
717 out_enoent:
718 	return -ENOENT;
719 }
720 
721 static const struct nfsd4_client_tracking_ops nfsd4_legacy_tracking_ops = {
722 	.init		= nfsd4_legacy_tracking_init,
723 	.exit		= nfsd4_legacy_tracking_exit,
724 	.create		= nfsd4_create_clid_dir,
725 	.remove		= nfsd4_remove_clid_dir,
726 	.check		= nfsd4_check_legacy_client,
727 	.grace_done	= nfsd4_recdir_purge_old,
728 	.version	= 1,
729 	.msglen		= 0,
730 };
731 
732 /* Globals */
733 #define NFSD_PIPE_DIR		"nfsd"
734 #define NFSD_CLD_PIPE		"cld"
735 
736 /* per-net-ns structure for holding cld upcall info */
737 struct cld_net {
738 	struct rpc_pipe		*cn_pipe;
739 	spinlock_t		 cn_lock;
740 	struct list_head	 cn_list;
741 	unsigned int		 cn_xid;
742 	bool			 cn_has_legacy;
743 	struct crypto_shash	*cn_tfm;
744 };
745 
746 struct cld_upcall {
747 	struct list_head	 cu_list;
748 	struct cld_net		*cu_net;
749 	struct completion	 cu_done;
750 	union {
751 		struct cld_msg_hdr	 cu_hdr;
752 		struct cld_msg		 cu_msg;
753 		struct cld_msg_v2	 cu_msg_v2;
754 	} cu_u;
755 };
756 
757 static int
758 __cld_pipe_upcall(struct rpc_pipe *pipe, void *cmsg)
759 {
760 	int ret;
761 	struct rpc_pipe_msg msg;
762 	struct cld_upcall *cup = container_of(cmsg, struct cld_upcall, cu_u);
763 	struct nfsd_net *nn = net_generic(pipe->dentry->d_sb->s_fs_info,
764 					  nfsd_net_id);
765 
766 	memset(&msg, 0, sizeof(msg));
767 	msg.data = cmsg;
768 	msg.len = nn->client_tracking_ops->msglen;
769 
770 	ret = rpc_queue_upcall(pipe, &msg);
771 	if (ret < 0) {
772 		goto out;
773 	}
774 
775 	wait_for_completion(&cup->cu_done);
776 
777 	if (msg.errno < 0)
778 		ret = msg.errno;
779 out:
780 	return ret;
781 }
782 
783 static int
784 cld_pipe_upcall(struct rpc_pipe *pipe, void *cmsg)
785 {
786 	int ret;
787 
788 	/*
789 	 * -EAGAIN occurs when pipe is closed and reopened while there are
790 	 *  upcalls queued.
791 	 */
792 	do {
793 		ret = __cld_pipe_upcall(pipe, cmsg);
794 	} while (ret == -EAGAIN);
795 
796 	return ret;
797 }
798 
799 static ssize_t
800 __cld_pipe_inprogress_downcall(const struct cld_msg_v2 __user *cmsg,
801 		struct nfsd_net *nn)
802 {
803 	uint8_t cmd, princhashlen;
804 	struct xdr_netobj name, princhash = { .len = 0, .data = NULL };
805 	uint16_t namelen;
806 	struct cld_net *cn = nn->cld_net;
807 
808 	if (get_user(cmd, &cmsg->cm_cmd)) {
809 		dprintk("%s: error when copying cmd from userspace", __func__);
810 		return -EFAULT;
811 	}
812 	if (cmd == Cld_GraceStart) {
813 		if (nn->client_tracking_ops->version >= 2) {
814 			const struct cld_clntinfo __user *ci;
815 
816 			ci = &cmsg->cm_u.cm_clntinfo;
817 			if (get_user(namelen, &ci->cc_name.cn_len))
818 				return -EFAULT;
819 			name.data = memdup_user(&ci->cc_name.cn_id, namelen);
820 			if (IS_ERR_OR_NULL(name.data))
821 				return -EFAULT;
822 			name.len = namelen;
823 			get_user(princhashlen, &ci->cc_princhash.cp_len);
824 			if (princhashlen > 0) {
825 				princhash.data = memdup_user(
826 						&ci->cc_princhash.cp_data,
827 						princhashlen);
828 				if (IS_ERR_OR_NULL(princhash.data))
829 					return -EFAULT;
830 				princhash.len = princhashlen;
831 			} else
832 				princhash.len = 0;
833 		} else {
834 			const struct cld_name __user *cnm;
835 
836 			cnm = &cmsg->cm_u.cm_name;
837 			if (get_user(namelen, &cnm->cn_len))
838 				return -EFAULT;
839 			name.data = memdup_user(&cnm->cn_id, namelen);
840 			if (IS_ERR_OR_NULL(name.data))
841 				return -EFAULT;
842 			name.len = namelen;
843 		}
844 		if (name.len > 5 && memcmp(name.data, "hash:", 5) == 0) {
845 			name.len = name.len - 5;
846 			memmove(name.data, name.data + 5, name.len);
847 			cn->cn_has_legacy = true;
848 		}
849 		if (!nfs4_client_to_reclaim(name, princhash, nn)) {
850 			kfree(name.data);
851 			kfree(princhash.data);
852 			return -EFAULT;
853 		}
854 		return nn->client_tracking_ops->msglen;
855 	}
856 	return -EFAULT;
857 }
858 
859 static ssize_t
860 cld_pipe_downcall(struct file *filp, const char __user *src, size_t mlen)
861 {
862 	struct cld_upcall *tmp, *cup;
863 	struct cld_msg_hdr __user *hdr = (struct cld_msg_hdr __user *)src;
864 	struct cld_msg_v2 __user *cmsg = (struct cld_msg_v2 __user *)src;
865 	uint32_t xid;
866 	struct nfsd_net *nn = net_generic(file_inode(filp)->i_sb->s_fs_info,
867 						nfsd_net_id);
868 	struct cld_net *cn = nn->cld_net;
869 	int16_t status;
870 
871 	if (mlen != nn->client_tracking_ops->msglen) {
872 		dprintk("%s: got %zu bytes, expected %zu\n", __func__, mlen,
873 			nn->client_tracking_ops->msglen);
874 		return -EINVAL;
875 	}
876 
877 	/* copy just the xid so we can try to find that */
878 	if (copy_from_user(&xid, &hdr->cm_xid, sizeof(xid)) != 0) {
879 		dprintk("%s: error when copying xid from userspace", __func__);
880 		return -EFAULT;
881 	}
882 
883 	/*
884 	 * copy the status so we know whether to remove the upcall from the
885 	 * list (for -EINPROGRESS, we just want to make sure the xid is
886 	 * valid, not remove the upcall from the list)
887 	 */
888 	if (get_user(status, &hdr->cm_status)) {
889 		dprintk("%s: error when copying status from userspace", __func__);
890 		return -EFAULT;
891 	}
892 
893 	/* walk the list and find corresponding xid */
894 	cup = NULL;
895 	spin_lock(&cn->cn_lock);
896 	list_for_each_entry(tmp, &cn->cn_list, cu_list) {
897 		if (get_unaligned(&tmp->cu_u.cu_hdr.cm_xid) == xid) {
898 			cup = tmp;
899 			if (status != -EINPROGRESS)
900 				list_del_init(&cup->cu_list);
901 			break;
902 		}
903 	}
904 	spin_unlock(&cn->cn_lock);
905 
906 	/* couldn't find upcall? */
907 	if (!cup) {
908 		dprintk("%s: couldn't find upcall -- xid=%u\n", __func__, xid);
909 		return -EINVAL;
910 	}
911 
912 	if (status == -EINPROGRESS)
913 		return __cld_pipe_inprogress_downcall(cmsg, nn);
914 
915 	if (copy_from_user(&cup->cu_u.cu_msg_v2, src, mlen) != 0)
916 		return -EFAULT;
917 
918 	complete(&cup->cu_done);
919 	return mlen;
920 }
921 
922 static void
923 cld_pipe_destroy_msg(struct rpc_pipe_msg *msg)
924 {
925 	struct cld_msg *cmsg = msg->data;
926 	struct cld_upcall *cup = container_of(cmsg, struct cld_upcall,
927 						 cu_u.cu_msg);
928 
929 	/* errno >= 0 means we got a downcall */
930 	if (msg->errno >= 0)
931 		return;
932 
933 	complete(&cup->cu_done);
934 }
935 
936 static const struct rpc_pipe_ops cld_upcall_ops = {
937 	.upcall		= rpc_pipe_generic_upcall,
938 	.downcall	= cld_pipe_downcall,
939 	.destroy_msg	= cld_pipe_destroy_msg,
940 };
941 
942 static struct dentry *
943 nfsd4_cld_register_sb(struct super_block *sb, struct rpc_pipe *pipe)
944 {
945 	struct dentry *dir, *dentry;
946 
947 	dir = rpc_d_lookup_sb(sb, NFSD_PIPE_DIR);
948 	if (dir == NULL)
949 		return ERR_PTR(-ENOENT);
950 	dentry = rpc_mkpipe_dentry(dir, NFSD_CLD_PIPE, NULL, pipe);
951 	dput(dir);
952 	return dentry;
953 }
954 
955 static void
956 nfsd4_cld_unregister_sb(struct rpc_pipe *pipe)
957 {
958 	if (pipe->dentry)
959 		rpc_unlink(pipe->dentry);
960 }
961 
962 static struct dentry *
963 nfsd4_cld_register_net(struct net *net, struct rpc_pipe *pipe)
964 {
965 	struct super_block *sb;
966 	struct dentry *dentry;
967 
968 	sb = rpc_get_sb_net(net);
969 	if (!sb)
970 		return NULL;
971 	dentry = nfsd4_cld_register_sb(sb, pipe);
972 	rpc_put_sb_net(net);
973 	return dentry;
974 }
975 
976 static void
977 nfsd4_cld_unregister_net(struct net *net, struct rpc_pipe *pipe)
978 {
979 	struct super_block *sb;
980 
981 	sb = rpc_get_sb_net(net);
982 	if (sb) {
983 		nfsd4_cld_unregister_sb(pipe);
984 		rpc_put_sb_net(net);
985 	}
986 }
987 
988 /* Initialize rpc_pipefs pipe for communication with client tracking daemon */
989 static int
990 __nfsd4_init_cld_pipe(struct net *net)
991 {
992 	int ret;
993 	struct dentry *dentry;
994 	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
995 	struct cld_net *cn;
996 
997 	if (nn->cld_net)
998 		return 0;
999 
1000 	cn = kzalloc(sizeof(*cn), GFP_KERNEL);
1001 	if (!cn) {
1002 		ret = -ENOMEM;
1003 		goto err;
1004 	}
1005 
1006 	cn->cn_pipe = rpc_mkpipe_data(&cld_upcall_ops, RPC_PIPE_WAIT_FOR_OPEN);
1007 	if (IS_ERR(cn->cn_pipe)) {
1008 		ret = PTR_ERR(cn->cn_pipe);
1009 		goto err;
1010 	}
1011 	spin_lock_init(&cn->cn_lock);
1012 	INIT_LIST_HEAD(&cn->cn_list);
1013 
1014 	dentry = nfsd4_cld_register_net(net, cn->cn_pipe);
1015 	if (IS_ERR(dentry)) {
1016 		ret = PTR_ERR(dentry);
1017 		goto err_destroy_data;
1018 	}
1019 
1020 	cn->cn_pipe->dentry = dentry;
1021 	cn->cn_has_legacy = false;
1022 	nn->cld_net = cn;
1023 	return 0;
1024 
1025 err_destroy_data:
1026 	rpc_destroy_pipe_data(cn->cn_pipe);
1027 err:
1028 	kfree(cn);
1029 	printk(KERN_ERR "NFSD: unable to create nfsdcld upcall pipe (%d)\n",
1030 			ret);
1031 	return ret;
1032 }
1033 
1034 static int
1035 nfsd4_init_cld_pipe(struct net *net)
1036 {
1037 	int status;
1038 
1039 	status = __nfsd4_init_cld_pipe(net);
1040 	if (!status)
1041 		printk("NFSD: Using old nfsdcld client tracking operations.\n");
1042 	return status;
1043 }
1044 
1045 static void
1046 nfsd4_remove_cld_pipe(struct net *net)
1047 {
1048 	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
1049 	struct cld_net *cn = nn->cld_net;
1050 
1051 	nfsd4_cld_unregister_net(net, cn->cn_pipe);
1052 	rpc_destroy_pipe_data(cn->cn_pipe);
1053 	if (cn->cn_tfm)
1054 		crypto_free_shash(cn->cn_tfm);
1055 	kfree(nn->cld_net);
1056 	nn->cld_net = NULL;
1057 }
1058 
1059 static struct cld_upcall *
1060 alloc_cld_upcall(struct nfsd_net *nn)
1061 {
1062 	struct cld_upcall *new, *tmp;
1063 	struct cld_net *cn = nn->cld_net;
1064 
1065 	new = kzalloc(sizeof(*new), GFP_KERNEL);
1066 	if (!new)
1067 		return new;
1068 
1069 	/* FIXME: hard cap on number in flight? */
1070 restart_search:
1071 	spin_lock(&cn->cn_lock);
1072 	list_for_each_entry(tmp, &cn->cn_list, cu_list) {
1073 		if (tmp->cu_u.cu_msg.cm_xid == cn->cn_xid) {
1074 			cn->cn_xid++;
1075 			spin_unlock(&cn->cn_lock);
1076 			goto restart_search;
1077 		}
1078 	}
1079 	init_completion(&new->cu_done);
1080 	new->cu_u.cu_msg.cm_vers = nn->client_tracking_ops->version;
1081 	put_unaligned(cn->cn_xid++, &new->cu_u.cu_msg.cm_xid);
1082 	new->cu_net = cn;
1083 	list_add(&new->cu_list, &cn->cn_list);
1084 	spin_unlock(&cn->cn_lock);
1085 
1086 	dprintk("%s: allocated xid %u\n", __func__, new->cu_u.cu_msg.cm_xid);
1087 
1088 	return new;
1089 }
1090 
1091 static void
1092 free_cld_upcall(struct cld_upcall *victim)
1093 {
1094 	struct cld_net *cn = victim->cu_net;
1095 
1096 	spin_lock(&cn->cn_lock);
1097 	list_del(&victim->cu_list);
1098 	spin_unlock(&cn->cn_lock);
1099 	kfree(victim);
1100 }
1101 
1102 /* Ask daemon to create a new record */
1103 static void
1104 nfsd4_cld_create(struct nfs4_client *clp)
1105 {
1106 	int ret;
1107 	struct cld_upcall *cup;
1108 	struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
1109 	struct cld_net *cn = nn->cld_net;
1110 
1111 	/* Don't upcall if it's already stored */
1112 	if (test_bit(NFSD4_CLIENT_STABLE, &clp->cl_flags))
1113 		return;
1114 
1115 	cup = alloc_cld_upcall(nn);
1116 	if (!cup) {
1117 		ret = -ENOMEM;
1118 		goto out_err;
1119 	}
1120 
1121 	cup->cu_u.cu_msg.cm_cmd = Cld_Create;
1122 	cup->cu_u.cu_msg.cm_u.cm_name.cn_len = clp->cl_name.len;
1123 	memcpy(cup->cu_u.cu_msg.cm_u.cm_name.cn_id, clp->cl_name.data,
1124 			clp->cl_name.len);
1125 
1126 	ret = cld_pipe_upcall(cn->cn_pipe, &cup->cu_u.cu_msg);
1127 	if (!ret) {
1128 		ret = cup->cu_u.cu_msg.cm_status;
1129 		set_bit(NFSD4_CLIENT_STABLE, &clp->cl_flags);
1130 	}
1131 
1132 	free_cld_upcall(cup);
1133 out_err:
1134 	if (ret)
1135 		printk(KERN_ERR "NFSD: Unable to create client "
1136 				"record on stable storage: %d\n", ret);
1137 }
1138 
1139 /* Ask daemon to create a new record */
1140 static void
1141 nfsd4_cld_create_v2(struct nfs4_client *clp)
1142 {
1143 	int ret;
1144 	struct cld_upcall *cup;
1145 	struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
1146 	struct cld_net *cn = nn->cld_net;
1147 	struct cld_msg_v2 *cmsg;
1148 	struct crypto_shash *tfm = cn->cn_tfm;
1149 	struct xdr_netobj cksum;
1150 	char *principal = NULL;
1151 	SHASH_DESC_ON_STACK(desc, tfm);
1152 
1153 	/* Don't upcall if it's already stored */
1154 	if (test_bit(NFSD4_CLIENT_STABLE, &clp->cl_flags))
1155 		return;
1156 
1157 	cup = alloc_cld_upcall(nn);
1158 	if (!cup) {
1159 		ret = -ENOMEM;
1160 		goto out_err;
1161 	}
1162 
1163 	cmsg = &cup->cu_u.cu_msg_v2;
1164 	cmsg->cm_cmd = Cld_Create;
1165 	cmsg->cm_u.cm_clntinfo.cc_name.cn_len = clp->cl_name.len;
1166 	memcpy(cmsg->cm_u.cm_clntinfo.cc_name.cn_id, clp->cl_name.data,
1167 			clp->cl_name.len);
1168 	if (clp->cl_cred.cr_raw_principal)
1169 		principal = clp->cl_cred.cr_raw_principal;
1170 	else if (clp->cl_cred.cr_principal)
1171 		principal = clp->cl_cred.cr_principal;
1172 	if (principal) {
1173 		desc->tfm = tfm;
1174 		cksum.len = crypto_shash_digestsize(tfm);
1175 		cksum.data = kmalloc(cksum.len, GFP_KERNEL);
1176 		if (cksum.data == NULL) {
1177 			ret = -ENOMEM;
1178 			goto out;
1179 		}
1180 		ret = crypto_shash_digest(desc, principal, strlen(principal),
1181 					  cksum.data);
1182 		shash_desc_zero(desc);
1183 		if (ret) {
1184 			kfree(cksum.data);
1185 			goto out;
1186 		}
1187 		cmsg->cm_u.cm_clntinfo.cc_princhash.cp_len = cksum.len;
1188 		memcpy(cmsg->cm_u.cm_clntinfo.cc_princhash.cp_data,
1189 		       cksum.data, cksum.len);
1190 		kfree(cksum.data);
1191 	} else
1192 		cmsg->cm_u.cm_clntinfo.cc_princhash.cp_len = 0;
1193 
1194 	ret = cld_pipe_upcall(cn->cn_pipe, cmsg);
1195 	if (!ret) {
1196 		ret = cmsg->cm_status;
1197 		set_bit(NFSD4_CLIENT_STABLE, &clp->cl_flags);
1198 	}
1199 
1200 out:
1201 	free_cld_upcall(cup);
1202 out_err:
1203 	if (ret)
1204 		pr_err("NFSD: Unable to create client record on stable storage: %d\n",
1205 				ret);
1206 }
1207 
1208 /* Ask daemon to create a new record */
1209 static void
1210 nfsd4_cld_remove(struct nfs4_client *clp)
1211 {
1212 	int ret;
1213 	struct cld_upcall *cup;
1214 	struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
1215 	struct cld_net *cn = nn->cld_net;
1216 
1217 	/* Don't upcall if it's already removed */
1218 	if (!test_bit(NFSD4_CLIENT_STABLE, &clp->cl_flags))
1219 		return;
1220 
1221 	cup = alloc_cld_upcall(nn);
1222 	if (!cup) {
1223 		ret = -ENOMEM;
1224 		goto out_err;
1225 	}
1226 
1227 	cup->cu_u.cu_msg.cm_cmd = Cld_Remove;
1228 	cup->cu_u.cu_msg.cm_u.cm_name.cn_len = clp->cl_name.len;
1229 	memcpy(cup->cu_u.cu_msg.cm_u.cm_name.cn_id, clp->cl_name.data,
1230 			clp->cl_name.len);
1231 
1232 	ret = cld_pipe_upcall(cn->cn_pipe, &cup->cu_u.cu_msg);
1233 	if (!ret) {
1234 		ret = cup->cu_u.cu_msg.cm_status;
1235 		clear_bit(NFSD4_CLIENT_STABLE, &clp->cl_flags);
1236 	}
1237 
1238 	free_cld_upcall(cup);
1239 out_err:
1240 	if (ret)
1241 		printk(KERN_ERR "NFSD: Unable to remove client "
1242 				"record from stable storage: %d\n", ret);
1243 }
1244 
1245 /*
1246  * For older nfsdcld's that do not allow us to "slurp" the clients
1247  * from the tracking database during startup.
1248  *
1249  * Check for presence of a record, and update its timestamp
1250  */
1251 static int
1252 nfsd4_cld_check_v0(struct nfs4_client *clp)
1253 {
1254 	int ret;
1255 	struct cld_upcall *cup;
1256 	struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
1257 	struct cld_net *cn = nn->cld_net;
1258 
1259 	/* Don't upcall if one was already stored during this grace pd */
1260 	if (test_bit(NFSD4_CLIENT_STABLE, &clp->cl_flags))
1261 		return 0;
1262 
1263 	cup = alloc_cld_upcall(nn);
1264 	if (!cup) {
1265 		printk(KERN_ERR "NFSD: Unable to check client record on "
1266 				"stable storage: %d\n", -ENOMEM);
1267 		return -ENOMEM;
1268 	}
1269 
1270 	cup->cu_u.cu_msg.cm_cmd = Cld_Check;
1271 	cup->cu_u.cu_msg.cm_u.cm_name.cn_len = clp->cl_name.len;
1272 	memcpy(cup->cu_u.cu_msg.cm_u.cm_name.cn_id, clp->cl_name.data,
1273 			clp->cl_name.len);
1274 
1275 	ret = cld_pipe_upcall(cn->cn_pipe, &cup->cu_u.cu_msg);
1276 	if (!ret) {
1277 		ret = cup->cu_u.cu_msg.cm_status;
1278 		set_bit(NFSD4_CLIENT_STABLE, &clp->cl_flags);
1279 	}
1280 
1281 	free_cld_upcall(cup);
1282 	return ret;
1283 }
1284 
1285 /*
1286  * For newer nfsdcld's that allow us to "slurp" the clients
1287  * from the tracking database during startup.
1288  *
1289  * Check for presence of a record in the reclaim_str_hashtbl
1290  */
1291 static int
1292 nfsd4_cld_check(struct nfs4_client *clp)
1293 {
1294 	struct nfs4_client_reclaim *crp;
1295 	struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
1296 	struct cld_net *cn = nn->cld_net;
1297 	int status;
1298 	char dname[HEXDIR_LEN];
1299 	struct xdr_netobj name;
1300 
1301 	/* did we already find that this client is stable? */
1302 	if (test_bit(NFSD4_CLIENT_STABLE, &clp->cl_flags))
1303 		return 0;
1304 
1305 	/* look for it in the reclaim hashtable otherwise */
1306 	crp = nfsd4_find_reclaim_client(clp->cl_name, nn);
1307 	if (crp)
1308 		goto found;
1309 
1310 	if (cn->cn_has_legacy) {
1311 		status = nfs4_make_rec_clidname(dname, &clp->cl_name);
1312 		if (status)
1313 			return -ENOENT;
1314 
1315 		name.data = kmemdup(dname, HEXDIR_LEN, GFP_KERNEL);
1316 		if (!name.data) {
1317 			dprintk("%s: failed to allocate memory for name.data!\n",
1318 				__func__);
1319 			return -ENOENT;
1320 		}
1321 		name.len = HEXDIR_LEN;
1322 		crp = nfsd4_find_reclaim_client(name, nn);
1323 		kfree(name.data);
1324 		if (crp)
1325 			goto found;
1326 
1327 	}
1328 	return -ENOENT;
1329 found:
1330 	crp->cr_clp = clp;
1331 	return 0;
1332 }
1333 
1334 static int
1335 nfsd4_cld_check_v2(struct nfs4_client *clp)
1336 {
1337 	struct nfs4_client_reclaim *crp;
1338 	struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
1339 	struct cld_net *cn = nn->cld_net;
1340 	int status;
1341 	char dname[HEXDIR_LEN];
1342 	struct xdr_netobj name;
1343 	struct crypto_shash *tfm = cn->cn_tfm;
1344 	struct xdr_netobj cksum;
1345 	char *principal = NULL;
1346 	SHASH_DESC_ON_STACK(desc, tfm);
1347 
1348 	/* did we already find that this client is stable? */
1349 	if (test_bit(NFSD4_CLIENT_STABLE, &clp->cl_flags))
1350 		return 0;
1351 
1352 	/* look for it in the reclaim hashtable otherwise */
1353 	crp = nfsd4_find_reclaim_client(clp->cl_name, nn);
1354 	if (crp)
1355 		goto found;
1356 
1357 	if (cn->cn_has_legacy) {
1358 		status = nfs4_make_rec_clidname(dname, &clp->cl_name);
1359 		if (status)
1360 			return -ENOENT;
1361 
1362 		name.data = kmemdup(dname, HEXDIR_LEN, GFP_KERNEL);
1363 		if (!name.data) {
1364 			dprintk("%s: failed to allocate memory for name.data\n",
1365 					__func__);
1366 			return -ENOENT;
1367 		}
1368 		name.len = HEXDIR_LEN;
1369 		crp = nfsd4_find_reclaim_client(name, nn);
1370 		kfree(name.data);
1371 		if (crp)
1372 			goto found;
1373 
1374 	}
1375 	return -ENOENT;
1376 found:
1377 	if (crp->cr_princhash.len) {
1378 		if (clp->cl_cred.cr_raw_principal)
1379 			principal = clp->cl_cred.cr_raw_principal;
1380 		else if (clp->cl_cred.cr_principal)
1381 			principal = clp->cl_cred.cr_principal;
1382 		if (principal == NULL)
1383 			return -ENOENT;
1384 		desc->tfm = tfm;
1385 		cksum.len = crypto_shash_digestsize(tfm);
1386 		cksum.data = kmalloc(cksum.len, GFP_KERNEL);
1387 		if (cksum.data == NULL)
1388 			return -ENOENT;
1389 		status = crypto_shash_digest(desc, principal, strlen(principal),
1390 					     cksum.data);
1391 		shash_desc_zero(desc);
1392 		if (status) {
1393 			kfree(cksum.data);
1394 			return -ENOENT;
1395 		}
1396 		if (memcmp(crp->cr_princhash.data, cksum.data,
1397 				crp->cr_princhash.len)) {
1398 			kfree(cksum.data);
1399 			return -ENOENT;
1400 		}
1401 		kfree(cksum.data);
1402 	}
1403 	crp->cr_clp = clp;
1404 	return 0;
1405 }
1406 
1407 static int
1408 nfsd4_cld_grace_start(struct nfsd_net *nn)
1409 {
1410 	int ret;
1411 	struct cld_upcall *cup;
1412 	struct cld_net *cn = nn->cld_net;
1413 
1414 	cup = alloc_cld_upcall(nn);
1415 	if (!cup) {
1416 		ret = -ENOMEM;
1417 		goto out_err;
1418 	}
1419 
1420 	cup->cu_u.cu_msg.cm_cmd = Cld_GraceStart;
1421 	ret = cld_pipe_upcall(cn->cn_pipe, &cup->cu_u.cu_msg);
1422 	if (!ret)
1423 		ret = cup->cu_u.cu_msg.cm_status;
1424 
1425 	free_cld_upcall(cup);
1426 out_err:
1427 	if (ret)
1428 		dprintk("%s: Unable to get clients from userspace: %d\n",
1429 			__func__, ret);
1430 	return ret;
1431 }
1432 
1433 /* For older nfsdcld's that need cm_gracetime */
1434 static void
1435 nfsd4_cld_grace_done_v0(struct nfsd_net *nn)
1436 {
1437 	int ret;
1438 	struct cld_upcall *cup;
1439 	struct cld_net *cn = nn->cld_net;
1440 
1441 	cup = alloc_cld_upcall(nn);
1442 	if (!cup) {
1443 		ret = -ENOMEM;
1444 		goto out_err;
1445 	}
1446 
1447 	cup->cu_u.cu_msg.cm_cmd = Cld_GraceDone;
1448 	cup->cu_u.cu_msg.cm_u.cm_gracetime = (int64_t)nn->boot_time;
1449 	ret = cld_pipe_upcall(cn->cn_pipe, &cup->cu_u.cu_msg);
1450 	if (!ret)
1451 		ret = cup->cu_u.cu_msg.cm_status;
1452 
1453 	free_cld_upcall(cup);
1454 out_err:
1455 	if (ret)
1456 		printk(KERN_ERR "NFSD: Unable to end grace period: %d\n", ret);
1457 }
1458 
1459 /*
1460  * For newer nfsdcld's that do not need cm_gracetime.  We also need to call
1461  * nfs4_release_reclaim() to clear out the reclaim_str_hashtbl.
1462  */
1463 static void
1464 nfsd4_cld_grace_done(struct nfsd_net *nn)
1465 {
1466 	int ret;
1467 	struct cld_upcall *cup;
1468 	struct cld_net *cn = nn->cld_net;
1469 
1470 	cup = alloc_cld_upcall(nn);
1471 	if (!cup) {
1472 		ret = -ENOMEM;
1473 		goto out_err;
1474 	}
1475 
1476 	cup->cu_u.cu_msg.cm_cmd = Cld_GraceDone;
1477 	ret = cld_pipe_upcall(cn->cn_pipe, &cup->cu_u.cu_msg);
1478 	if (!ret)
1479 		ret = cup->cu_u.cu_msg.cm_status;
1480 
1481 	free_cld_upcall(cup);
1482 out_err:
1483 	nfs4_release_reclaim(nn);
1484 	if (ret)
1485 		printk(KERN_ERR "NFSD: Unable to end grace period: %d\n", ret);
1486 }
1487 
1488 static int
1489 nfs4_cld_state_init(struct net *net)
1490 {
1491 	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
1492 	int i;
1493 
1494 	nn->reclaim_str_hashtbl = kmalloc_array(CLIENT_HASH_SIZE,
1495 						sizeof(struct list_head),
1496 						GFP_KERNEL);
1497 	if (!nn->reclaim_str_hashtbl)
1498 		return -ENOMEM;
1499 
1500 	for (i = 0; i < CLIENT_HASH_SIZE; i++)
1501 		INIT_LIST_HEAD(&nn->reclaim_str_hashtbl[i]);
1502 	nn->reclaim_str_hashtbl_size = 0;
1503 	nn->track_reclaim_completes = true;
1504 	atomic_set(&nn->nr_reclaim_complete, 0);
1505 
1506 	return 0;
1507 }
1508 
1509 static void
1510 nfs4_cld_state_shutdown(struct net *net)
1511 {
1512 	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
1513 
1514 	nn->track_reclaim_completes = false;
1515 	kfree(nn->reclaim_str_hashtbl);
1516 }
1517 
1518 static bool
1519 cld_running(struct nfsd_net *nn)
1520 {
1521 	struct cld_net *cn = nn->cld_net;
1522 	struct rpc_pipe *pipe = cn->cn_pipe;
1523 
1524 	return pipe->nreaders || pipe->nwriters;
1525 }
1526 
1527 static int
1528 nfsd4_cld_get_version(struct nfsd_net *nn)
1529 {
1530 	int ret = 0;
1531 	struct cld_upcall *cup;
1532 	struct cld_net *cn = nn->cld_net;
1533 	uint8_t version;
1534 
1535 	cup = alloc_cld_upcall(nn);
1536 	if (!cup) {
1537 		ret = -ENOMEM;
1538 		goto out_err;
1539 	}
1540 	cup->cu_u.cu_msg.cm_cmd = Cld_GetVersion;
1541 	ret = cld_pipe_upcall(cn->cn_pipe, &cup->cu_u.cu_msg);
1542 	if (!ret) {
1543 		ret = cup->cu_u.cu_msg.cm_status;
1544 		if (ret)
1545 			goto out_free;
1546 		version = cup->cu_u.cu_msg.cm_u.cm_version;
1547 		dprintk("%s: userspace returned version %u\n",
1548 				__func__, version);
1549 		if (version < 1)
1550 			version = 1;
1551 		else if (version > CLD_UPCALL_VERSION)
1552 			version = CLD_UPCALL_VERSION;
1553 
1554 		switch (version) {
1555 		case 1:
1556 			nn->client_tracking_ops = &nfsd4_cld_tracking_ops;
1557 			break;
1558 		case 2:
1559 			nn->client_tracking_ops = &nfsd4_cld_tracking_ops_v2;
1560 			break;
1561 		default:
1562 			break;
1563 		}
1564 	}
1565 out_free:
1566 	free_cld_upcall(cup);
1567 out_err:
1568 	if (ret)
1569 		dprintk("%s: Unable to get version from userspace: %d\n",
1570 			__func__, ret);
1571 	return ret;
1572 }
1573 
1574 static int
1575 nfsd4_cld_tracking_init(struct net *net)
1576 {
1577 	int status;
1578 	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
1579 	bool running;
1580 	int retries = 10;
1581 	struct crypto_shash *tfm;
1582 
1583 	status = nfs4_cld_state_init(net);
1584 	if (status)
1585 		return status;
1586 
1587 	status = __nfsd4_init_cld_pipe(net);
1588 	if (status)
1589 		goto err_shutdown;
1590 
1591 	/*
1592 	 * rpc pipe upcalls take 30 seconds to time out, so we don't want to
1593 	 * queue an upcall unless we know that nfsdcld is running (because we
1594 	 * want this to fail fast so that nfsd4_client_tracking_init() can try
1595 	 * the next client tracking method).  nfsdcld should already be running
1596 	 * before nfsd is started, so the wait here is for nfsdcld to open the
1597 	 * pipefs file we just created.
1598 	 */
1599 	while (!(running = cld_running(nn)) && retries--)
1600 		msleep(100);
1601 
1602 	if (!running) {
1603 		status = -ETIMEDOUT;
1604 		goto err_remove;
1605 	}
1606 	tfm = crypto_alloc_shash("sha256", 0, 0);
1607 	if (IS_ERR(tfm)) {
1608 		status = PTR_ERR(tfm);
1609 		goto err_remove;
1610 	}
1611 	nn->cld_net->cn_tfm = tfm;
1612 
1613 	status = nfsd4_cld_get_version(nn);
1614 	if (status == -EOPNOTSUPP)
1615 		pr_warn("NFSD: nfsdcld GetVersion upcall failed. Please upgrade nfsdcld.\n");
1616 
1617 	status = nfsd4_cld_grace_start(nn);
1618 	if (status) {
1619 		if (status == -EOPNOTSUPP)
1620 			pr_warn("NFSD: nfsdcld GraceStart upcall failed. Please upgrade nfsdcld.\n");
1621 		nfs4_release_reclaim(nn);
1622 		goto err_remove;
1623 	} else
1624 		printk("NFSD: Using nfsdcld client tracking operations.\n");
1625 	return 0;
1626 
1627 err_remove:
1628 	nfsd4_remove_cld_pipe(net);
1629 err_shutdown:
1630 	nfs4_cld_state_shutdown(net);
1631 	return status;
1632 }
1633 
1634 static void
1635 nfsd4_cld_tracking_exit(struct net *net)
1636 {
1637 	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
1638 
1639 	nfs4_release_reclaim(nn);
1640 	nfsd4_remove_cld_pipe(net);
1641 	nfs4_cld_state_shutdown(net);
1642 }
1643 
1644 /* For older nfsdcld's */
1645 static const struct nfsd4_client_tracking_ops nfsd4_cld_tracking_ops_v0 = {
1646 	.init		= nfsd4_init_cld_pipe,
1647 	.exit		= nfsd4_remove_cld_pipe,
1648 	.create		= nfsd4_cld_create,
1649 	.remove		= nfsd4_cld_remove,
1650 	.check		= nfsd4_cld_check_v0,
1651 	.grace_done	= nfsd4_cld_grace_done_v0,
1652 	.version	= 1,
1653 	.msglen		= sizeof(struct cld_msg),
1654 };
1655 
1656 /* For newer nfsdcld's */
1657 static const struct nfsd4_client_tracking_ops nfsd4_cld_tracking_ops = {
1658 	.init		= nfsd4_cld_tracking_init,
1659 	.exit		= nfsd4_cld_tracking_exit,
1660 	.create		= nfsd4_cld_create,
1661 	.remove		= nfsd4_cld_remove,
1662 	.check		= nfsd4_cld_check,
1663 	.grace_done	= nfsd4_cld_grace_done,
1664 	.version	= 1,
1665 	.msglen		= sizeof(struct cld_msg),
1666 };
1667 
1668 /* v2 create/check ops include the principal, if available */
1669 static const struct nfsd4_client_tracking_ops nfsd4_cld_tracking_ops_v2 = {
1670 	.init		= nfsd4_cld_tracking_init,
1671 	.exit		= nfsd4_cld_tracking_exit,
1672 	.create		= nfsd4_cld_create_v2,
1673 	.remove		= nfsd4_cld_remove,
1674 	.check		= nfsd4_cld_check_v2,
1675 	.grace_done	= nfsd4_cld_grace_done,
1676 	.version	= 2,
1677 	.msglen		= sizeof(struct cld_msg_v2),
1678 };
1679 
1680 /* upcall via usermodehelper */
1681 static char cltrack_prog[PATH_MAX] = "/sbin/nfsdcltrack";
1682 module_param_string(cltrack_prog, cltrack_prog, sizeof(cltrack_prog),
1683 			S_IRUGO|S_IWUSR);
1684 MODULE_PARM_DESC(cltrack_prog, "Path to the nfsdcltrack upcall program");
1685 
1686 static bool cltrack_legacy_disable;
1687 module_param(cltrack_legacy_disable, bool, S_IRUGO|S_IWUSR);
1688 MODULE_PARM_DESC(cltrack_legacy_disable,
1689 		"Disable legacy recoverydir conversion. Default: false");
1690 
1691 #define LEGACY_TOPDIR_ENV_PREFIX "NFSDCLTRACK_LEGACY_TOPDIR="
1692 #define LEGACY_RECDIR_ENV_PREFIX "NFSDCLTRACK_LEGACY_RECDIR="
1693 #define HAS_SESSION_ENV_PREFIX "NFSDCLTRACK_CLIENT_HAS_SESSION="
1694 #define GRACE_START_ENV_PREFIX "NFSDCLTRACK_GRACE_START="
1695 
1696 static char *
1697 nfsd4_cltrack_legacy_topdir(void)
1698 {
1699 	int copied;
1700 	size_t len;
1701 	char *result;
1702 
1703 	if (cltrack_legacy_disable)
1704 		return NULL;
1705 
1706 	len = strlen(LEGACY_TOPDIR_ENV_PREFIX) +
1707 		strlen(nfs4_recoverydir()) + 1;
1708 
1709 	result = kmalloc(len, GFP_KERNEL);
1710 	if (!result)
1711 		return result;
1712 
1713 	copied = snprintf(result, len, LEGACY_TOPDIR_ENV_PREFIX "%s",
1714 				nfs4_recoverydir());
1715 	if (copied >= len) {
1716 		/* just return nothing if output was truncated */
1717 		kfree(result);
1718 		return NULL;
1719 	}
1720 
1721 	return result;
1722 }
1723 
1724 static char *
1725 nfsd4_cltrack_legacy_recdir(const struct xdr_netobj *name)
1726 {
1727 	int copied;
1728 	size_t len;
1729 	char *result;
1730 
1731 	if (cltrack_legacy_disable)
1732 		return NULL;
1733 
1734 	/* +1 is for '/' between "topdir" and "recdir" */
1735 	len = strlen(LEGACY_RECDIR_ENV_PREFIX) +
1736 		strlen(nfs4_recoverydir()) + 1 + HEXDIR_LEN;
1737 
1738 	result = kmalloc(len, GFP_KERNEL);
1739 	if (!result)
1740 		return result;
1741 
1742 	copied = snprintf(result, len, LEGACY_RECDIR_ENV_PREFIX "%s/",
1743 				nfs4_recoverydir());
1744 	if (copied > (len - HEXDIR_LEN)) {
1745 		/* just return nothing if output will be truncated */
1746 		kfree(result);
1747 		return NULL;
1748 	}
1749 
1750 	copied = nfs4_make_rec_clidname(result + copied, name);
1751 	if (copied) {
1752 		kfree(result);
1753 		return NULL;
1754 	}
1755 
1756 	return result;
1757 }
1758 
1759 static char *
1760 nfsd4_cltrack_client_has_session(struct nfs4_client *clp)
1761 {
1762 	int copied;
1763 	size_t len;
1764 	char *result;
1765 
1766 	/* prefix + Y/N character + terminating NULL */
1767 	len = strlen(HAS_SESSION_ENV_PREFIX) + 1 + 1;
1768 
1769 	result = kmalloc(len, GFP_KERNEL);
1770 	if (!result)
1771 		return result;
1772 
1773 	copied = snprintf(result, len, HAS_SESSION_ENV_PREFIX "%c",
1774 				clp->cl_minorversion ? 'Y' : 'N');
1775 	if (copied >= len) {
1776 		/* just return nothing if output was truncated */
1777 		kfree(result);
1778 		return NULL;
1779 	}
1780 
1781 	return result;
1782 }
1783 
1784 static char *
1785 nfsd4_cltrack_grace_start(time_t grace_start)
1786 {
1787 	int copied;
1788 	size_t len;
1789 	char *result;
1790 
1791 	/* prefix + max width of int64_t string + terminating NULL */
1792 	len = strlen(GRACE_START_ENV_PREFIX) + 22 + 1;
1793 
1794 	result = kmalloc(len, GFP_KERNEL);
1795 	if (!result)
1796 		return result;
1797 
1798 	copied = snprintf(result, len, GRACE_START_ENV_PREFIX "%ld",
1799 				grace_start);
1800 	if (copied >= len) {
1801 		/* just return nothing if output was truncated */
1802 		kfree(result);
1803 		return NULL;
1804 	}
1805 
1806 	return result;
1807 }
1808 
1809 static int
1810 nfsd4_umh_cltrack_upcall(char *cmd, char *arg, char *env0, char *env1)
1811 {
1812 	char *envp[3];
1813 	char *argv[4];
1814 	int ret;
1815 
1816 	if (unlikely(!cltrack_prog[0])) {
1817 		dprintk("%s: cltrack_prog is disabled\n", __func__);
1818 		return -EACCES;
1819 	}
1820 
1821 	dprintk("%s: cmd: %s\n", __func__, cmd);
1822 	dprintk("%s: arg: %s\n", __func__, arg ? arg : "(null)");
1823 	dprintk("%s: env0: %s\n", __func__, env0 ? env0 : "(null)");
1824 	dprintk("%s: env1: %s\n", __func__, env1 ? env1 : "(null)");
1825 
1826 	envp[0] = env0;
1827 	envp[1] = env1;
1828 	envp[2] = NULL;
1829 
1830 	argv[0] = (char *)cltrack_prog;
1831 	argv[1] = cmd;
1832 	argv[2] = arg;
1833 	argv[3] = NULL;
1834 
1835 	ret = call_usermodehelper(argv[0], argv, envp, UMH_WAIT_PROC);
1836 	/*
1837 	 * Disable the upcall mechanism if we're getting an ENOENT or EACCES
1838 	 * error. The admin can re-enable it on the fly by using sysfs
1839 	 * once the problem has been fixed.
1840 	 */
1841 	if (ret == -ENOENT || ret == -EACCES) {
1842 		dprintk("NFSD: %s was not found or isn't executable (%d). "
1843 			"Setting cltrack_prog to blank string!",
1844 			cltrack_prog, ret);
1845 		cltrack_prog[0] = '\0';
1846 	}
1847 	dprintk("%s: %s return value: %d\n", __func__, cltrack_prog, ret);
1848 
1849 	return ret;
1850 }
1851 
1852 static char *
1853 bin_to_hex_dup(const unsigned char *src, int srclen)
1854 {
1855 	char *buf;
1856 
1857 	/* +1 for terminating NULL */
1858 	buf = kzalloc((srclen * 2) + 1, GFP_KERNEL);
1859 	if (!buf)
1860 		return buf;
1861 
1862 	bin2hex(buf, src, srclen);
1863 	return buf;
1864 }
1865 
1866 static int
1867 nfsd4_umh_cltrack_init(struct net *net)
1868 {
1869 	int ret;
1870 	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
1871 	char *grace_start = nfsd4_cltrack_grace_start(nn->boot_time);
1872 
1873 	/* XXX: The usermode helper s not working in container yet. */
1874 	if (net != &init_net) {
1875 		pr_warn("NFSD: attempt to initialize umh client tracking in a container ignored.\n");
1876 		kfree(grace_start);
1877 		return -EINVAL;
1878 	}
1879 
1880 	ret = nfsd4_umh_cltrack_upcall("init", NULL, grace_start, NULL);
1881 	kfree(grace_start);
1882 	if (!ret)
1883 		printk("NFSD: Using UMH upcall client tracking operations.\n");
1884 	return ret;
1885 }
1886 
1887 static void
1888 nfsd4_cltrack_upcall_lock(struct nfs4_client *clp)
1889 {
1890 	wait_on_bit_lock(&clp->cl_flags, NFSD4_CLIENT_UPCALL_LOCK,
1891 			 TASK_UNINTERRUPTIBLE);
1892 }
1893 
1894 static void
1895 nfsd4_cltrack_upcall_unlock(struct nfs4_client *clp)
1896 {
1897 	smp_mb__before_atomic();
1898 	clear_bit(NFSD4_CLIENT_UPCALL_LOCK, &clp->cl_flags);
1899 	smp_mb__after_atomic();
1900 	wake_up_bit(&clp->cl_flags, NFSD4_CLIENT_UPCALL_LOCK);
1901 }
1902 
1903 static void
1904 nfsd4_umh_cltrack_create(struct nfs4_client *clp)
1905 {
1906 	char *hexid, *has_session, *grace_start;
1907 	struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
1908 
1909 	/*
1910 	 * With v4.0 clients, there's little difference in outcome between a
1911 	 * create and check operation, and we can end up calling into this
1912 	 * function multiple times per client (once for each openowner). So,
1913 	 * for v4.0 clients skip upcalling once the client has been recorded
1914 	 * on stable storage.
1915 	 *
1916 	 * For v4.1+ clients, the outcome of the two operations is different,
1917 	 * so we must ensure that we upcall for the create operation. v4.1+
1918 	 * clients call this on RECLAIM_COMPLETE though, so we should only end
1919 	 * up doing a single create upcall per client.
1920 	 */
1921 	if (clp->cl_minorversion == 0 &&
1922 	    test_bit(NFSD4_CLIENT_STABLE, &clp->cl_flags))
1923 		return;
1924 
1925 	hexid = bin_to_hex_dup(clp->cl_name.data, clp->cl_name.len);
1926 	if (!hexid) {
1927 		dprintk("%s: can't allocate memory for upcall!\n", __func__);
1928 		return;
1929 	}
1930 
1931 	has_session = nfsd4_cltrack_client_has_session(clp);
1932 	grace_start = nfsd4_cltrack_grace_start(nn->boot_time);
1933 
1934 	nfsd4_cltrack_upcall_lock(clp);
1935 	if (!nfsd4_umh_cltrack_upcall("create", hexid, has_session, grace_start))
1936 		set_bit(NFSD4_CLIENT_STABLE, &clp->cl_flags);
1937 	nfsd4_cltrack_upcall_unlock(clp);
1938 
1939 	kfree(has_session);
1940 	kfree(grace_start);
1941 	kfree(hexid);
1942 }
1943 
1944 static void
1945 nfsd4_umh_cltrack_remove(struct nfs4_client *clp)
1946 {
1947 	char *hexid;
1948 
1949 	if (!test_bit(NFSD4_CLIENT_STABLE, &clp->cl_flags))
1950 		return;
1951 
1952 	hexid = bin_to_hex_dup(clp->cl_name.data, clp->cl_name.len);
1953 	if (!hexid) {
1954 		dprintk("%s: can't allocate memory for upcall!\n", __func__);
1955 		return;
1956 	}
1957 
1958 	nfsd4_cltrack_upcall_lock(clp);
1959 	if (test_bit(NFSD4_CLIENT_STABLE, &clp->cl_flags) &&
1960 	    nfsd4_umh_cltrack_upcall("remove", hexid, NULL, NULL) == 0)
1961 		clear_bit(NFSD4_CLIENT_STABLE, &clp->cl_flags);
1962 	nfsd4_cltrack_upcall_unlock(clp);
1963 
1964 	kfree(hexid);
1965 }
1966 
1967 static int
1968 nfsd4_umh_cltrack_check(struct nfs4_client *clp)
1969 {
1970 	int ret;
1971 	char *hexid, *has_session, *legacy;
1972 
1973 	if (test_bit(NFSD4_CLIENT_STABLE, &clp->cl_flags))
1974 		return 0;
1975 
1976 	hexid = bin_to_hex_dup(clp->cl_name.data, clp->cl_name.len);
1977 	if (!hexid) {
1978 		dprintk("%s: can't allocate memory for upcall!\n", __func__);
1979 		return -ENOMEM;
1980 	}
1981 
1982 	has_session = nfsd4_cltrack_client_has_session(clp);
1983 	legacy = nfsd4_cltrack_legacy_recdir(&clp->cl_name);
1984 
1985 	nfsd4_cltrack_upcall_lock(clp);
1986 	if (test_bit(NFSD4_CLIENT_STABLE, &clp->cl_flags)) {
1987 		ret = 0;
1988 	} else {
1989 		ret = nfsd4_umh_cltrack_upcall("check", hexid, has_session, legacy);
1990 		if (ret == 0)
1991 			set_bit(NFSD4_CLIENT_STABLE, &clp->cl_flags);
1992 	}
1993 	nfsd4_cltrack_upcall_unlock(clp);
1994 	kfree(has_session);
1995 	kfree(legacy);
1996 	kfree(hexid);
1997 
1998 	return ret;
1999 }
2000 
2001 static void
2002 nfsd4_umh_cltrack_grace_done(struct nfsd_net *nn)
2003 {
2004 	char *legacy;
2005 	char timestr[22]; /* FIXME: better way to determine max size? */
2006 
2007 	sprintf(timestr, "%ld", nn->boot_time);
2008 	legacy = nfsd4_cltrack_legacy_topdir();
2009 	nfsd4_umh_cltrack_upcall("gracedone", timestr, legacy, NULL);
2010 	kfree(legacy);
2011 }
2012 
2013 static const struct nfsd4_client_tracking_ops nfsd4_umh_tracking_ops = {
2014 	.init		= nfsd4_umh_cltrack_init,
2015 	.exit		= NULL,
2016 	.create		= nfsd4_umh_cltrack_create,
2017 	.remove		= nfsd4_umh_cltrack_remove,
2018 	.check		= nfsd4_umh_cltrack_check,
2019 	.grace_done	= nfsd4_umh_cltrack_grace_done,
2020 	.version	= 1,
2021 	.msglen		= 0,
2022 };
2023 
2024 int
2025 nfsd4_client_tracking_init(struct net *net)
2026 {
2027 	int status;
2028 	struct path path;
2029 	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
2030 
2031 	/* just run the init if it the method is already decided */
2032 	if (nn->client_tracking_ops)
2033 		goto do_init;
2034 
2035 	/* First, try to use nfsdcld */
2036 	nn->client_tracking_ops = &nfsd4_cld_tracking_ops;
2037 	status = nn->client_tracking_ops->init(net);
2038 	if (!status)
2039 		return status;
2040 	if (status != -ETIMEDOUT) {
2041 		nn->client_tracking_ops = &nfsd4_cld_tracking_ops_v0;
2042 		status = nn->client_tracking_ops->init(net);
2043 		if (!status)
2044 			return status;
2045 	}
2046 
2047 	/*
2048 	 * Next, try the UMH upcall.
2049 	 */
2050 	nn->client_tracking_ops = &nfsd4_umh_tracking_ops;
2051 	status = nn->client_tracking_ops->init(net);
2052 	if (!status)
2053 		return status;
2054 
2055 	/*
2056 	 * Finally, See if the recoverydir exists and is a directory.
2057 	 * If it is, then use the legacy ops.
2058 	 */
2059 	nn->client_tracking_ops = &nfsd4_legacy_tracking_ops;
2060 	status = kern_path(nfs4_recoverydir(), LOOKUP_FOLLOW, &path);
2061 	if (!status) {
2062 		status = d_is_dir(path.dentry);
2063 		path_put(&path);
2064 		if (!status) {
2065 			status = -EINVAL;
2066 			goto out;
2067 		}
2068 	}
2069 
2070 do_init:
2071 	status = nn->client_tracking_ops->init(net);
2072 out:
2073 	if (status) {
2074 		printk(KERN_WARNING "NFSD: Unable to initialize client "
2075 				    "recovery tracking! (%d)\n", status);
2076 		nn->client_tracking_ops = NULL;
2077 	}
2078 	return status;
2079 }
2080 
2081 void
2082 nfsd4_client_tracking_exit(struct net *net)
2083 {
2084 	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
2085 
2086 	if (nn->client_tracking_ops) {
2087 		if (nn->client_tracking_ops->exit)
2088 			nn->client_tracking_ops->exit(net);
2089 		nn->client_tracking_ops = NULL;
2090 	}
2091 }
2092 
2093 void
2094 nfsd4_client_record_create(struct nfs4_client *clp)
2095 {
2096 	struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
2097 
2098 	if (nn->client_tracking_ops)
2099 		nn->client_tracking_ops->create(clp);
2100 }
2101 
2102 void
2103 nfsd4_client_record_remove(struct nfs4_client *clp)
2104 {
2105 	struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
2106 
2107 	if (nn->client_tracking_ops)
2108 		nn->client_tracking_ops->remove(clp);
2109 }
2110 
2111 int
2112 nfsd4_client_record_check(struct nfs4_client *clp)
2113 {
2114 	struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
2115 
2116 	if (nn->client_tracking_ops)
2117 		return nn->client_tracking_ops->check(clp);
2118 
2119 	return -EOPNOTSUPP;
2120 }
2121 
2122 void
2123 nfsd4_record_grace_done(struct nfsd_net *nn)
2124 {
2125 	if (nn->client_tracking_ops)
2126 		nn->client_tracking_ops->grace_done(nn);
2127 }
2128 
2129 static int
2130 rpc_pipefs_event(struct notifier_block *nb, unsigned long event, void *ptr)
2131 {
2132 	struct super_block *sb = ptr;
2133 	struct net *net = sb->s_fs_info;
2134 	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
2135 	struct cld_net *cn = nn->cld_net;
2136 	struct dentry *dentry;
2137 	int ret = 0;
2138 
2139 	if (!try_module_get(THIS_MODULE))
2140 		return 0;
2141 
2142 	if (!cn) {
2143 		module_put(THIS_MODULE);
2144 		return 0;
2145 	}
2146 
2147 	switch (event) {
2148 	case RPC_PIPEFS_MOUNT:
2149 		dentry = nfsd4_cld_register_sb(sb, cn->cn_pipe);
2150 		if (IS_ERR(dentry)) {
2151 			ret = PTR_ERR(dentry);
2152 			break;
2153 		}
2154 		cn->cn_pipe->dentry = dentry;
2155 		break;
2156 	case RPC_PIPEFS_UMOUNT:
2157 		if (cn->cn_pipe->dentry)
2158 			nfsd4_cld_unregister_sb(cn->cn_pipe);
2159 		break;
2160 	default:
2161 		ret = -ENOTSUPP;
2162 		break;
2163 	}
2164 	module_put(THIS_MODULE);
2165 	return ret;
2166 }
2167 
2168 static struct notifier_block nfsd4_cld_block = {
2169 	.notifier_call = rpc_pipefs_event,
2170 };
2171 
2172 int
2173 register_cld_notifier(void)
2174 {
2175 	return rpc_pipefs_notifier_register(&nfsd4_cld_block);
2176 }
2177 
2178 void
2179 unregister_cld_notifier(void)
2180 {
2181 	rpc_pipefs_notifier_unregister(&nfsd4_cld_block);
2182 }
2183