xref: /linux/fs/nfsd/nfs4recover.c (revision 7f71507851fc7764b36a3221839607d3a45c2025)
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 #ifdef CONFIG_NFSD_LEGACY_CLIENT_TRACKING
70 /* Globals */
71 static char user_recovery_dirname[PATH_MAX] = "/var/lib/nfs/v4recovery";
72 
73 static int
74 nfs4_save_creds(const struct cred **original_creds)
75 {
76 	struct cred *new;
77 
78 	new = prepare_creds();
79 	if (!new)
80 		return -ENOMEM;
81 
82 	new->fsuid = GLOBAL_ROOT_UID;
83 	new->fsgid = GLOBAL_ROOT_GID;
84 	*original_creds = override_creds(new);
85 	put_cred(new);
86 	return 0;
87 }
88 
89 static void
90 nfs4_reset_creds(const struct cred *original)
91 {
92 	revert_creds(original);
93 }
94 
95 static void
96 md5_to_hex(char *out, char *md5)
97 {
98 	int i;
99 
100 	for (i=0; i<16; i++) {
101 		unsigned char c = md5[i];
102 
103 		*out++ = '0' + ((c&0xf0)>>4) + (c>=0xa0)*('a'-'9'-1);
104 		*out++ = '0' + (c&0x0f) + ((c&0x0f)>=0x0a)*('a'-'9'-1);
105 	}
106 	*out = '\0';
107 }
108 
109 static int
110 nfs4_make_rec_clidname(char *dname, const struct xdr_netobj *clname)
111 {
112 	struct xdr_netobj cksum;
113 	struct crypto_shash *tfm;
114 	int status;
115 
116 	dprintk("NFSD: nfs4_make_rec_clidname for %.*s\n",
117 			clname->len, clname->data);
118 	tfm = crypto_alloc_shash("md5", 0, 0);
119 	if (IS_ERR(tfm)) {
120 		status = PTR_ERR(tfm);
121 		goto out_no_tfm;
122 	}
123 
124 	cksum.len = crypto_shash_digestsize(tfm);
125 	cksum.data = kmalloc(cksum.len, GFP_KERNEL);
126 	if (cksum.data == NULL) {
127 		status = -ENOMEM;
128  		goto out;
129 	}
130 
131 	status = crypto_shash_tfm_digest(tfm, clname->data, clname->len,
132 					 cksum.data);
133 	if (status)
134 		goto out;
135 
136 	md5_to_hex(dname, cksum.data);
137 
138 	status = 0;
139 out:
140 	kfree(cksum.data);
141 	crypto_free_shash(tfm);
142 out_no_tfm:
143 	return status;
144 }
145 
146 /*
147  * If we had an error generating the recdir name for the legacy tracker
148  * then warn the admin. If the error doesn't appear to be transient,
149  * then disable recovery tracking.
150  */
151 static void
152 legacy_recdir_name_error(struct nfs4_client *clp, int error)
153 {
154 	printk(KERN_ERR "NFSD: unable to generate recoverydir "
155 			"name (%d).\n", error);
156 
157 	/*
158 	 * if the algorithm just doesn't exist, then disable the recovery
159 	 * tracker altogether. The crypto libs will generally return this if
160 	 * FIPS is enabled as well.
161 	 */
162 	if (error == -ENOENT) {
163 		printk(KERN_ERR "NFSD: disabling legacy clientid tracking. "
164 			"Reboot recovery will not function correctly!\n");
165 		nfsd4_client_tracking_exit(clp->net);
166 	}
167 }
168 
169 static void
170 __nfsd4_create_reclaim_record_grace(struct nfs4_client *clp,
171 		const char *dname, int len, struct nfsd_net *nn)
172 {
173 	struct xdr_netobj name;
174 	struct xdr_netobj princhash = { .len = 0, .data = NULL };
175 	struct nfs4_client_reclaim *crp;
176 
177 	name.data = kmemdup(dname, len, GFP_KERNEL);
178 	if (!name.data) {
179 		dprintk("%s: failed to allocate memory for name.data!\n",
180 			__func__);
181 		return;
182 	}
183 	name.len = len;
184 	crp = nfs4_client_to_reclaim(name, princhash, nn);
185 	if (!crp) {
186 		kfree(name.data);
187 		return;
188 	}
189 	crp->cr_clp = clp;
190 }
191 
192 static void
193 nfsd4_create_clid_dir(struct nfs4_client *clp)
194 {
195 	const struct cred *original_cred;
196 	char dname[HEXDIR_LEN];
197 	struct dentry *dir, *dentry;
198 	int status;
199 	struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
200 
201 	if (test_and_set_bit(NFSD4_CLIENT_STABLE, &clp->cl_flags))
202 		return;
203 	if (!nn->rec_file)
204 		return;
205 
206 	status = nfs4_make_rec_clidname(dname, &clp->cl_name);
207 	if (status)
208 		return legacy_recdir_name_error(clp, status);
209 
210 	status = nfs4_save_creds(&original_cred);
211 	if (status < 0)
212 		return;
213 
214 	status = mnt_want_write_file(nn->rec_file);
215 	if (status)
216 		goto out_creds;
217 
218 	dir = nn->rec_file->f_path.dentry;
219 	/* lock the parent */
220 	inode_lock(d_inode(dir));
221 
222 	dentry = lookup_one_len(dname, dir, HEXDIR_LEN-1);
223 	if (IS_ERR(dentry)) {
224 		status = PTR_ERR(dentry);
225 		goto out_unlock;
226 	}
227 	if (d_really_is_positive(dentry))
228 		/*
229 		 * In the 4.1 case, where we're called from
230 		 * reclaim_complete(), records from the previous reboot
231 		 * may still be left, so this is OK.
232 		 *
233 		 * In the 4.0 case, we should never get here; but we may
234 		 * as well be forgiving and just succeed silently.
235 		 */
236 		goto out_put;
237 	status = vfs_mkdir(&nop_mnt_idmap, d_inode(dir), dentry, S_IRWXU);
238 out_put:
239 	dput(dentry);
240 out_unlock:
241 	inode_unlock(d_inode(dir));
242 	if (status == 0) {
243 		if (nn->in_grace)
244 			__nfsd4_create_reclaim_record_grace(clp, dname,
245 					HEXDIR_LEN, nn);
246 		vfs_fsync(nn->rec_file, 0);
247 	} else {
248 		printk(KERN_ERR "NFSD: failed to write recovery record"
249 				" (err %d); please check that %s exists"
250 				" and is writeable", status,
251 				user_recovery_dirname);
252 	}
253 	mnt_drop_write_file(nn->rec_file);
254 out_creds:
255 	nfs4_reset_creds(original_cred);
256 }
257 
258 typedef int (recdir_func)(struct dentry *, struct dentry *, struct nfsd_net *);
259 
260 struct name_list {
261 	char name[HEXDIR_LEN];
262 	struct list_head list;
263 };
264 
265 struct nfs4_dir_ctx {
266 	struct dir_context ctx;
267 	struct list_head names;
268 };
269 
270 static bool
271 nfsd4_build_namelist(struct dir_context *__ctx, const char *name, int namlen,
272 		loff_t offset, u64 ino, unsigned int d_type)
273 {
274 	struct nfs4_dir_ctx *ctx =
275 		container_of(__ctx, struct nfs4_dir_ctx, ctx);
276 	struct name_list *entry;
277 
278 	if (namlen != HEXDIR_LEN - 1)
279 		return true;
280 	entry = kmalloc(sizeof(struct name_list), GFP_KERNEL);
281 	if (entry == NULL)
282 		return false;
283 	memcpy(entry->name, name, HEXDIR_LEN - 1);
284 	entry->name[HEXDIR_LEN - 1] = '\0';
285 	list_add(&entry->list, &ctx->names);
286 	return true;
287 }
288 
289 static int
290 nfsd4_list_rec_dir(recdir_func *f, struct nfsd_net *nn)
291 {
292 	const struct cred *original_cred;
293 	struct dentry *dir = nn->rec_file->f_path.dentry;
294 	struct nfs4_dir_ctx ctx = {
295 		.ctx.actor = nfsd4_build_namelist,
296 		.names = LIST_HEAD_INIT(ctx.names)
297 	};
298 	struct name_list *entry, *tmp;
299 	int status;
300 
301 	status = nfs4_save_creds(&original_cred);
302 	if (status < 0)
303 		return status;
304 
305 	status = vfs_llseek(nn->rec_file, 0, SEEK_SET);
306 	if (status < 0) {
307 		nfs4_reset_creds(original_cred);
308 		return status;
309 	}
310 
311 	status = iterate_dir(nn->rec_file, &ctx.ctx);
312 	inode_lock_nested(d_inode(dir), I_MUTEX_PARENT);
313 
314 	list_for_each_entry_safe(entry, tmp, &ctx.names, list) {
315 		if (!status) {
316 			struct dentry *dentry;
317 			dentry = lookup_one_len(entry->name, dir, HEXDIR_LEN-1);
318 			if (IS_ERR(dentry)) {
319 				status = PTR_ERR(dentry);
320 				break;
321 			}
322 			status = f(dir, dentry, nn);
323 			dput(dentry);
324 		}
325 		list_del(&entry->list);
326 		kfree(entry);
327 	}
328 	inode_unlock(d_inode(dir));
329 	nfs4_reset_creds(original_cred);
330 
331 	list_for_each_entry_safe(entry, tmp, &ctx.names, list) {
332 		dprintk("NFSD: %s. Left entry %s\n", __func__, entry->name);
333 		list_del(&entry->list);
334 		kfree(entry);
335 	}
336 	return status;
337 }
338 
339 static int
340 nfsd4_unlink_clid_dir(char *name, int namlen, struct nfsd_net *nn)
341 {
342 	struct dentry *dir, *dentry;
343 	int status;
344 
345 	dprintk("NFSD: nfsd4_unlink_clid_dir. name %.*s\n", namlen, name);
346 
347 	dir = nn->rec_file->f_path.dentry;
348 	inode_lock_nested(d_inode(dir), I_MUTEX_PARENT);
349 	dentry = lookup_one_len(name, dir, namlen);
350 	if (IS_ERR(dentry)) {
351 		status = PTR_ERR(dentry);
352 		goto out_unlock;
353 	}
354 	status = -ENOENT;
355 	if (d_really_is_negative(dentry))
356 		goto out;
357 	status = vfs_rmdir(&nop_mnt_idmap, d_inode(dir), dentry);
358 out:
359 	dput(dentry);
360 out_unlock:
361 	inode_unlock(d_inode(dir));
362 	return status;
363 }
364 
365 static void
366 __nfsd4_remove_reclaim_record_grace(const char *dname, int len,
367 		struct nfsd_net *nn)
368 {
369 	struct xdr_netobj name;
370 	struct nfs4_client_reclaim *crp;
371 
372 	name.data = kmemdup(dname, len, GFP_KERNEL);
373 	if (!name.data) {
374 		dprintk("%s: failed to allocate memory for name.data!\n",
375 			__func__);
376 		return;
377 	}
378 	name.len = len;
379 	crp = nfsd4_find_reclaim_client(name, nn);
380 	kfree(name.data);
381 	if (crp)
382 		nfs4_remove_reclaim_record(crp, nn);
383 }
384 
385 static void
386 nfsd4_remove_clid_dir(struct nfs4_client *clp)
387 {
388 	const struct cred *original_cred;
389 	char dname[HEXDIR_LEN];
390 	int status;
391 	struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
392 
393 	if (!nn->rec_file || !test_bit(NFSD4_CLIENT_STABLE, &clp->cl_flags))
394 		return;
395 
396 	status = nfs4_make_rec_clidname(dname, &clp->cl_name);
397 	if (status)
398 		return legacy_recdir_name_error(clp, status);
399 
400 	status = mnt_want_write_file(nn->rec_file);
401 	if (status)
402 		goto out;
403 	clear_bit(NFSD4_CLIENT_STABLE, &clp->cl_flags);
404 
405 	status = nfs4_save_creds(&original_cred);
406 	if (status < 0)
407 		goto out_drop_write;
408 
409 	status = nfsd4_unlink_clid_dir(dname, HEXDIR_LEN-1, nn);
410 	nfs4_reset_creds(original_cred);
411 	if (status == 0) {
412 		vfs_fsync(nn->rec_file, 0);
413 		if (nn->in_grace)
414 			__nfsd4_remove_reclaim_record_grace(dname,
415 					HEXDIR_LEN, nn);
416 	}
417 out_drop_write:
418 	mnt_drop_write_file(nn->rec_file);
419 out:
420 	if (status)
421 		printk("NFSD: Failed to remove expired client state directory"
422 				" %.*s\n", HEXDIR_LEN, dname);
423 }
424 
425 static int
426 purge_old(struct dentry *parent, struct dentry *child, struct nfsd_net *nn)
427 {
428 	int status;
429 	struct xdr_netobj name;
430 
431 	if (child->d_name.len != HEXDIR_LEN - 1) {
432 		printk("%s: illegal name %pd in recovery directory\n",
433 				__func__, child);
434 		/* Keep trying; maybe the others are OK: */
435 		return 0;
436 	}
437 	name.data = kmemdup_nul(child->d_name.name, child->d_name.len, GFP_KERNEL);
438 	if (!name.data) {
439 		dprintk("%s: failed to allocate memory for name.data!\n",
440 			__func__);
441 		goto out;
442 	}
443 	name.len = HEXDIR_LEN;
444 	if (nfs4_has_reclaimed_state(name, nn))
445 		goto out_free;
446 
447 	status = vfs_rmdir(&nop_mnt_idmap, d_inode(parent), child);
448 	if (status)
449 		printk("failed to remove client recovery directory %pd\n",
450 				child);
451 out_free:
452 	kfree(name.data);
453 out:
454 	/* Keep trying, success or failure: */
455 	return 0;
456 }
457 
458 static void
459 nfsd4_recdir_purge_old(struct nfsd_net *nn)
460 {
461 	int status;
462 
463 	nn->in_grace = false;
464 	if (!nn->rec_file)
465 		return;
466 	status = mnt_want_write_file(nn->rec_file);
467 	if (status)
468 		goto out;
469 	status = nfsd4_list_rec_dir(purge_old, nn);
470 	if (status == 0)
471 		vfs_fsync(nn->rec_file, 0);
472 	mnt_drop_write_file(nn->rec_file);
473 out:
474 	nfs4_release_reclaim(nn);
475 	if (status)
476 		printk("nfsd4: failed to purge old clients from recovery"
477 			" directory %pD\n", nn->rec_file);
478 }
479 
480 static int
481 load_recdir(struct dentry *parent, struct dentry *child, struct nfsd_net *nn)
482 {
483 	struct xdr_netobj name;
484 	struct xdr_netobj princhash = { .len = 0, .data = NULL };
485 
486 	if (child->d_name.len != HEXDIR_LEN - 1) {
487 		printk("%s: illegal name %pd in recovery directory\n",
488 				__func__, child);
489 		/* Keep trying; maybe the others are OK: */
490 		return 0;
491 	}
492 	name.data = kmemdup_nul(child->d_name.name, child->d_name.len, GFP_KERNEL);
493 	if (!name.data) {
494 		dprintk("%s: failed to allocate memory for name.data!\n",
495 			__func__);
496 		goto out;
497 	}
498 	name.len = HEXDIR_LEN;
499 	if (!nfs4_client_to_reclaim(name, princhash, nn))
500 		kfree(name.data);
501 out:
502 	return 0;
503 }
504 
505 static int
506 nfsd4_recdir_load(struct net *net) {
507 	int status;
508 	struct nfsd_net *nn =  net_generic(net, nfsd_net_id);
509 
510 	if (!nn->rec_file)
511 		return 0;
512 
513 	status = nfsd4_list_rec_dir(load_recdir, nn);
514 	if (status)
515 		printk("nfsd4: failed loading clients from recovery"
516 			" directory %pD\n", nn->rec_file);
517 	return status;
518 }
519 
520 /*
521  * Hold reference to the recovery directory.
522  */
523 
524 static int
525 nfsd4_init_recdir(struct net *net)
526 {
527 	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
528 	const struct cred *original_cred;
529 	int status;
530 
531 	printk("NFSD: Using %s as the NFSv4 state recovery directory\n",
532 			user_recovery_dirname);
533 
534 	BUG_ON(nn->rec_file);
535 
536 	status = nfs4_save_creds(&original_cred);
537 	if (status < 0) {
538 		printk("NFSD: Unable to change credentials to find recovery"
539 		       " directory: error %d\n",
540 		       status);
541 		return status;
542 	}
543 
544 	nn->rec_file = filp_open(user_recovery_dirname, O_RDONLY | O_DIRECTORY, 0);
545 	if (IS_ERR(nn->rec_file)) {
546 		printk("NFSD: unable to find recovery directory %s\n",
547 				user_recovery_dirname);
548 		status = PTR_ERR(nn->rec_file);
549 		nn->rec_file = NULL;
550 	}
551 
552 	nfs4_reset_creds(original_cred);
553 	if (!status)
554 		nn->in_grace = true;
555 	return status;
556 }
557 
558 static void
559 nfsd4_shutdown_recdir(struct net *net)
560 {
561 	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
562 
563 	if (!nn->rec_file)
564 		return;
565 	fput(nn->rec_file);
566 	nn->rec_file = NULL;
567 }
568 
569 static int
570 nfs4_legacy_state_init(struct net *net)
571 {
572 	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
573 	int i;
574 
575 	nn->reclaim_str_hashtbl = kmalloc_array(CLIENT_HASH_SIZE,
576 						sizeof(struct list_head),
577 						GFP_KERNEL);
578 	if (!nn->reclaim_str_hashtbl)
579 		return -ENOMEM;
580 
581 	for (i = 0; i < CLIENT_HASH_SIZE; i++)
582 		INIT_LIST_HEAD(&nn->reclaim_str_hashtbl[i]);
583 	nn->reclaim_str_hashtbl_size = 0;
584 
585 	return 0;
586 }
587 
588 static void
589 nfs4_legacy_state_shutdown(struct net *net)
590 {
591 	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
592 
593 	kfree(nn->reclaim_str_hashtbl);
594 }
595 
596 static int
597 nfsd4_load_reboot_recovery_data(struct net *net)
598 {
599 	int status;
600 
601 	status = nfsd4_init_recdir(net);
602 	if (status)
603 		return status;
604 
605 	status = nfsd4_recdir_load(net);
606 	if (status)
607 		nfsd4_shutdown_recdir(net);
608 
609 	return status;
610 }
611 
612 static int
613 nfsd4_legacy_tracking_init(struct net *net)
614 {
615 	int status;
616 
617 	/* XXX: The legacy code won't work in a container */
618 	if (net != &init_net) {
619 		pr_warn("NFSD: attempt to initialize legacy client tracking in a container ignored.\n");
620 		return -EINVAL;
621 	}
622 
623 	status = nfs4_legacy_state_init(net);
624 	if (status)
625 		return status;
626 
627 	status = nfsd4_load_reboot_recovery_data(net);
628 	if (status)
629 		goto err;
630 	pr_info("NFSD: Using legacy client tracking operations.\n");
631 	return 0;
632 
633 err:
634 	nfs4_legacy_state_shutdown(net);
635 	return status;
636 }
637 
638 static void
639 nfsd4_legacy_tracking_exit(struct net *net)
640 {
641 	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
642 
643 	nfs4_release_reclaim(nn);
644 	nfsd4_shutdown_recdir(net);
645 	nfs4_legacy_state_shutdown(net);
646 }
647 
648 /*
649  * Change the NFSv4 recovery directory to recdir.
650  */
651 int
652 nfs4_reset_recoverydir(char *recdir)
653 {
654 	int status;
655 	struct path path;
656 
657 	status = kern_path(recdir, LOOKUP_FOLLOW, &path);
658 	if (status)
659 		return status;
660 	status = -ENOTDIR;
661 	if (d_is_dir(path.dentry)) {
662 		strscpy(user_recovery_dirname, recdir,
663 			sizeof(user_recovery_dirname));
664 		status = 0;
665 	}
666 	path_put(&path);
667 	return status;
668 }
669 
670 char *
671 nfs4_recoverydir(void)
672 {
673 	return user_recovery_dirname;
674 }
675 
676 static int
677 nfsd4_check_legacy_client(struct nfs4_client *clp)
678 {
679 	int status;
680 	char dname[HEXDIR_LEN];
681 	struct nfs4_client_reclaim *crp;
682 	struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
683 	struct xdr_netobj name;
684 
685 	/* did we already find that this client is stable? */
686 	if (test_bit(NFSD4_CLIENT_STABLE, &clp->cl_flags))
687 		return 0;
688 
689 	status = nfs4_make_rec_clidname(dname, &clp->cl_name);
690 	if (status) {
691 		legacy_recdir_name_error(clp, status);
692 		return status;
693 	}
694 
695 	/* look for it in the reclaim hashtable otherwise */
696 	name.data = kmemdup(dname, HEXDIR_LEN, GFP_KERNEL);
697 	if (!name.data) {
698 		dprintk("%s: failed to allocate memory for name.data!\n",
699 			__func__);
700 		goto out_enoent;
701 	}
702 	name.len = HEXDIR_LEN;
703 	crp = nfsd4_find_reclaim_client(name, nn);
704 	kfree(name.data);
705 	if (crp) {
706 		set_bit(NFSD4_CLIENT_STABLE, &clp->cl_flags);
707 		crp->cr_clp = clp;
708 		return 0;
709 	}
710 
711 out_enoent:
712 	return -ENOENT;
713 }
714 
715 static const struct nfsd4_client_tracking_ops nfsd4_legacy_tracking_ops = {
716 	.init		= nfsd4_legacy_tracking_init,
717 	.exit		= nfsd4_legacy_tracking_exit,
718 	.create		= nfsd4_create_clid_dir,
719 	.remove		= nfsd4_remove_clid_dir,
720 	.check		= nfsd4_check_legacy_client,
721 	.grace_done	= nfsd4_recdir_purge_old,
722 	.version	= 1,
723 	.msglen		= 0,
724 };
725 #endif /* CONFIG_NFSD_LEGACY_CLIENT_TRACKING */
726 
727 /* Globals */
728 #define NFSD_PIPE_DIR		"nfsd"
729 #define NFSD_CLD_PIPE		"cld"
730 
731 /* per-net-ns structure for holding cld upcall info */
732 struct cld_net {
733 	struct rpc_pipe		*cn_pipe;
734 	spinlock_t		 cn_lock;
735 	struct list_head	 cn_list;
736 	unsigned int		 cn_xid;
737 	struct crypto_shash	*cn_tfm;
738 #ifdef CONFIG_NFSD_LEGACY_CLIENT_TRACKING
739 	bool			 cn_has_legacy;
740 #endif
741 };
742 
743 struct cld_upcall {
744 	struct list_head	 cu_list;
745 	struct cld_net		*cu_net;
746 	struct completion	 cu_done;
747 	union {
748 		struct cld_msg_hdr	 cu_hdr;
749 		struct cld_msg		 cu_msg;
750 		struct cld_msg_v2	 cu_msg_v2;
751 	} cu_u;
752 };
753 
754 static int
755 __cld_pipe_upcall(struct rpc_pipe *pipe, void *cmsg, struct nfsd_net *nn)
756 {
757 	int ret;
758 	struct rpc_pipe_msg msg;
759 	struct cld_upcall *cup = container_of(cmsg, struct cld_upcall, cu_u);
760 
761 	memset(&msg, 0, sizeof(msg));
762 	msg.data = cmsg;
763 	msg.len = nn->client_tracking_ops->msglen;
764 
765 	ret = rpc_queue_upcall(pipe, &msg);
766 	if (ret < 0) {
767 		goto out;
768 	}
769 
770 	wait_for_completion(&cup->cu_done);
771 
772 	if (msg.errno < 0)
773 		ret = msg.errno;
774 out:
775 	return ret;
776 }
777 
778 static int
779 cld_pipe_upcall(struct rpc_pipe *pipe, void *cmsg, struct nfsd_net *nn)
780 {
781 	int ret;
782 
783 	/*
784 	 * -EAGAIN occurs when pipe is closed and reopened while there are
785 	 *  upcalls queued.
786 	 */
787 	do {
788 		ret = __cld_pipe_upcall(pipe, cmsg, nn);
789 	} while (ret == -EAGAIN);
790 
791 	return ret;
792 }
793 
794 static ssize_t
795 __cld_pipe_inprogress_downcall(const struct cld_msg_v2 __user *cmsg,
796 		struct nfsd_net *nn)
797 {
798 	uint8_t cmd, princhashlen;
799 	struct xdr_netobj name, princhash = { .len = 0, .data = NULL };
800 	uint16_t namelen;
801 
802 	if (get_user(cmd, &cmsg->cm_cmd)) {
803 		dprintk("%s: error when copying cmd from userspace", __func__);
804 		return -EFAULT;
805 	}
806 	if (cmd == Cld_GraceStart) {
807 		if (nn->client_tracking_ops->version >= 2) {
808 			const struct cld_clntinfo __user *ci;
809 
810 			ci = &cmsg->cm_u.cm_clntinfo;
811 			if (get_user(namelen, &ci->cc_name.cn_len))
812 				return -EFAULT;
813 			if (namelen == 0 || namelen > NFS4_OPAQUE_LIMIT) {
814 				dprintk("%s: invalid namelen (%u)", __func__, namelen);
815 				return -EINVAL;
816 			}
817 			name.data = memdup_user(&ci->cc_name.cn_id, namelen);
818 			if (IS_ERR(name.data))
819 				return PTR_ERR(name.data);
820 			name.len = namelen;
821 			get_user(princhashlen, &ci->cc_princhash.cp_len);
822 			if (princhashlen > 0) {
823 				princhash.data = memdup_user(
824 						&ci->cc_princhash.cp_data,
825 						princhashlen);
826 				if (IS_ERR(princhash.data)) {
827 					kfree(name.data);
828 					return PTR_ERR(princhash.data);
829 				}
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 			if (namelen == 0 || namelen > NFS4_OPAQUE_LIMIT) {
840 				dprintk("%s: invalid namelen (%u)", __func__, namelen);
841 				return -EINVAL;
842 			}
843 			name.data = memdup_user(&cnm->cn_id, namelen);
844 			if (IS_ERR(name.data))
845 				return PTR_ERR(name.data);
846 			name.len = namelen;
847 		}
848 #ifdef CONFIG_NFSD_LEGACY_CLIENT_TRACKING
849 		if (name.len > 5 && memcmp(name.data, "hash:", 5) == 0) {
850 			struct cld_net *cn = nn->cld_net;
851 
852 			name.len = name.len - 5;
853 			memmove(name.data, name.data + 5, name.len);
854 			cn->cn_has_legacy = true;
855 		}
856 #endif
857 		if (!nfs4_client_to_reclaim(name, princhash, nn)) {
858 			kfree(name.data);
859 			kfree(princhash.data);
860 			return -EFAULT;
861 		}
862 		return nn->client_tracking_ops->msglen;
863 	}
864 	return -EFAULT;
865 }
866 
867 static ssize_t
868 cld_pipe_downcall(struct file *filp, const char __user *src, size_t mlen)
869 {
870 	struct cld_upcall *tmp, *cup;
871 	struct cld_msg_hdr __user *hdr = (struct cld_msg_hdr __user *)src;
872 	struct cld_msg_v2 __user *cmsg = (struct cld_msg_v2 __user *)src;
873 	uint32_t xid;
874 	struct nfsd_net *nn = net_generic(file_inode(filp)->i_sb->s_fs_info,
875 						nfsd_net_id);
876 	struct cld_net *cn = nn->cld_net;
877 	int16_t status;
878 
879 	if (mlen != nn->client_tracking_ops->msglen) {
880 		dprintk("%s: got %zu bytes, expected %zu\n", __func__, mlen,
881 			nn->client_tracking_ops->msglen);
882 		return -EINVAL;
883 	}
884 
885 	/* copy just the xid so we can try to find that */
886 	if (copy_from_user(&xid, &hdr->cm_xid, sizeof(xid)) != 0) {
887 		dprintk("%s: error when copying xid from userspace", __func__);
888 		return -EFAULT;
889 	}
890 
891 	/*
892 	 * copy the status so we know whether to remove the upcall from the
893 	 * list (for -EINPROGRESS, we just want to make sure the xid is
894 	 * valid, not remove the upcall from the list)
895 	 */
896 	if (get_user(status, &hdr->cm_status)) {
897 		dprintk("%s: error when copying status from userspace", __func__);
898 		return -EFAULT;
899 	}
900 
901 	/* walk the list and find corresponding xid */
902 	cup = NULL;
903 	spin_lock(&cn->cn_lock);
904 	list_for_each_entry(tmp, &cn->cn_list, cu_list) {
905 		if (get_unaligned(&tmp->cu_u.cu_hdr.cm_xid) == xid) {
906 			cup = tmp;
907 			if (status != -EINPROGRESS)
908 				list_del_init(&cup->cu_list);
909 			break;
910 		}
911 	}
912 	spin_unlock(&cn->cn_lock);
913 
914 	/* couldn't find upcall? */
915 	if (!cup) {
916 		dprintk("%s: couldn't find upcall -- xid=%u\n", __func__, xid);
917 		return -EINVAL;
918 	}
919 
920 	if (status == -EINPROGRESS)
921 		return __cld_pipe_inprogress_downcall(cmsg, nn);
922 
923 	if (copy_from_user(&cup->cu_u.cu_msg_v2, src, mlen) != 0)
924 		return -EFAULT;
925 
926 	complete(&cup->cu_done);
927 	return mlen;
928 }
929 
930 static void
931 cld_pipe_destroy_msg(struct rpc_pipe_msg *msg)
932 {
933 	struct cld_msg *cmsg = msg->data;
934 	struct cld_upcall *cup = container_of(cmsg, struct cld_upcall,
935 						 cu_u.cu_msg);
936 
937 	/* errno >= 0 means we got a downcall */
938 	if (msg->errno >= 0)
939 		return;
940 
941 	complete(&cup->cu_done);
942 }
943 
944 static const struct rpc_pipe_ops cld_upcall_ops = {
945 	.upcall		= rpc_pipe_generic_upcall,
946 	.downcall	= cld_pipe_downcall,
947 	.destroy_msg	= cld_pipe_destroy_msg,
948 };
949 
950 static struct dentry *
951 nfsd4_cld_register_sb(struct super_block *sb, struct rpc_pipe *pipe)
952 {
953 	struct dentry *dir, *dentry;
954 
955 	dir = rpc_d_lookup_sb(sb, NFSD_PIPE_DIR);
956 	if (dir == NULL)
957 		return ERR_PTR(-ENOENT);
958 	dentry = rpc_mkpipe_dentry(dir, NFSD_CLD_PIPE, NULL, pipe);
959 	dput(dir);
960 	return dentry;
961 }
962 
963 static void
964 nfsd4_cld_unregister_sb(struct rpc_pipe *pipe)
965 {
966 	if (pipe->dentry)
967 		rpc_unlink(pipe->dentry);
968 }
969 
970 static struct dentry *
971 nfsd4_cld_register_net(struct net *net, struct rpc_pipe *pipe)
972 {
973 	struct super_block *sb;
974 	struct dentry *dentry;
975 
976 	sb = rpc_get_sb_net(net);
977 	if (!sb)
978 		return NULL;
979 	dentry = nfsd4_cld_register_sb(sb, pipe);
980 	rpc_put_sb_net(net);
981 	return dentry;
982 }
983 
984 static void
985 nfsd4_cld_unregister_net(struct net *net, struct rpc_pipe *pipe)
986 {
987 	struct super_block *sb;
988 
989 	sb = rpc_get_sb_net(net);
990 	if (sb) {
991 		nfsd4_cld_unregister_sb(pipe);
992 		rpc_put_sb_net(net);
993 	}
994 }
995 
996 /* Initialize rpc_pipefs pipe for communication with client tracking daemon */
997 static int
998 __nfsd4_init_cld_pipe(struct net *net)
999 {
1000 	int ret;
1001 	struct dentry *dentry;
1002 	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
1003 	struct cld_net *cn;
1004 
1005 	if (nn->cld_net)
1006 		return 0;
1007 
1008 	cn = kzalloc(sizeof(*cn), GFP_KERNEL);
1009 	if (!cn) {
1010 		ret = -ENOMEM;
1011 		goto err;
1012 	}
1013 
1014 	cn->cn_pipe = rpc_mkpipe_data(&cld_upcall_ops, RPC_PIPE_WAIT_FOR_OPEN);
1015 	if (IS_ERR(cn->cn_pipe)) {
1016 		ret = PTR_ERR(cn->cn_pipe);
1017 		goto err;
1018 	}
1019 	spin_lock_init(&cn->cn_lock);
1020 	INIT_LIST_HEAD(&cn->cn_list);
1021 
1022 	dentry = nfsd4_cld_register_net(net, cn->cn_pipe);
1023 	if (IS_ERR(dentry)) {
1024 		ret = PTR_ERR(dentry);
1025 		goto err_destroy_data;
1026 	}
1027 
1028 	cn->cn_pipe->dentry = dentry;
1029 #ifdef CONFIG_NFSD_LEGACY_CLIENT_TRACKING
1030 	cn->cn_has_legacy = false;
1031 #endif
1032 	nn->cld_net = cn;
1033 	return 0;
1034 
1035 err_destroy_data:
1036 	rpc_destroy_pipe_data(cn->cn_pipe);
1037 err:
1038 	kfree(cn);
1039 	printk(KERN_ERR "NFSD: unable to create nfsdcld upcall pipe (%d)\n",
1040 			ret);
1041 	return ret;
1042 }
1043 
1044 static int
1045 nfsd4_init_cld_pipe(struct net *net)
1046 {
1047 	int status;
1048 
1049 	status = __nfsd4_init_cld_pipe(net);
1050 	if (!status)
1051 		pr_info("NFSD: Using old nfsdcld client tracking operations.\n");
1052 	return status;
1053 }
1054 
1055 static void
1056 nfsd4_remove_cld_pipe(struct net *net)
1057 {
1058 	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
1059 	struct cld_net *cn = nn->cld_net;
1060 
1061 	nfsd4_cld_unregister_net(net, cn->cn_pipe);
1062 	rpc_destroy_pipe_data(cn->cn_pipe);
1063 	if (cn->cn_tfm)
1064 		crypto_free_shash(cn->cn_tfm);
1065 	kfree(nn->cld_net);
1066 	nn->cld_net = NULL;
1067 }
1068 
1069 static struct cld_upcall *
1070 alloc_cld_upcall(struct nfsd_net *nn)
1071 {
1072 	struct cld_upcall *new, *tmp;
1073 	struct cld_net *cn = nn->cld_net;
1074 
1075 	new = kzalloc(sizeof(*new), GFP_KERNEL);
1076 	if (!new)
1077 		return new;
1078 
1079 	/* FIXME: hard cap on number in flight? */
1080 restart_search:
1081 	spin_lock(&cn->cn_lock);
1082 	list_for_each_entry(tmp, &cn->cn_list, cu_list) {
1083 		if (tmp->cu_u.cu_msg.cm_xid == cn->cn_xid) {
1084 			cn->cn_xid++;
1085 			spin_unlock(&cn->cn_lock);
1086 			goto restart_search;
1087 		}
1088 	}
1089 	init_completion(&new->cu_done);
1090 	new->cu_u.cu_msg.cm_vers = nn->client_tracking_ops->version;
1091 	put_unaligned(cn->cn_xid++, &new->cu_u.cu_msg.cm_xid);
1092 	new->cu_net = cn;
1093 	list_add(&new->cu_list, &cn->cn_list);
1094 	spin_unlock(&cn->cn_lock);
1095 
1096 	dprintk("%s: allocated xid %u\n", __func__, new->cu_u.cu_msg.cm_xid);
1097 
1098 	return new;
1099 }
1100 
1101 static void
1102 free_cld_upcall(struct cld_upcall *victim)
1103 {
1104 	struct cld_net *cn = victim->cu_net;
1105 
1106 	spin_lock(&cn->cn_lock);
1107 	list_del(&victim->cu_list);
1108 	spin_unlock(&cn->cn_lock);
1109 	kfree(victim);
1110 }
1111 
1112 /* Ask daemon to create a new record */
1113 static void
1114 nfsd4_cld_create(struct nfs4_client *clp)
1115 {
1116 	int ret;
1117 	struct cld_upcall *cup;
1118 	struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
1119 	struct cld_net *cn = nn->cld_net;
1120 
1121 	/* Don't upcall if it's already stored */
1122 	if (test_bit(NFSD4_CLIENT_STABLE, &clp->cl_flags))
1123 		return;
1124 
1125 	cup = alloc_cld_upcall(nn);
1126 	if (!cup) {
1127 		ret = -ENOMEM;
1128 		goto out_err;
1129 	}
1130 
1131 	cup->cu_u.cu_msg.cm_cmd = Cld_Create;
1132 	cup->cu_u.cu_msg.cm_u.cm_name.cn_len = clp->cl_name.len;
1133 	memcpy(cup->cu_u.cu_msg.cm_u.cm_name.cn_id, clp->cl_name.data,
1134 			clp->cl_name.len);
1135 
1136 	ret = cld_pipe_upcall(cn->cn_pipe, &cup->cu_u.cu_msg, nn);
1137 	if (!ret) {
1138 		ret = cup->cu_u.cu_msg.cm_status;
1139 		set_bit(NFSD4_CLIENT_STABLE, &clp->cl_flags);
1140 	}
1141 
1142 	free_cld_upcall(cup);
1143 out_err:
1144 	if (ret)
1145 		printk(KERN_ERR "NFSD: Unable to create client "
1146 				"record on stable storage: %d\n", ret);
1147 }
1148 
1149 /* Ask daemon to create a new record */
1150 static void
1151 nfsd4_cld_create_v2(struct nfs4_client *clp)
1152 {
1153 	int ret;
1154 	struct cld_upcall *cup;
1155 	struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
1156 	struct cld_net *cn = nn->cld_net;
1157 	struct cld_msg_v2 *cmsg;
1158 	struct crypto_shash *tfm = cn->cn_tfm;
1159 	struct xdr_netobj cksum;
1160 	char *principal = NULL;
1161 
1162 	/* Don't upcall if it's already stored */
1163 	if (test_bit(NFSD4_CLIENT_STABLE, &clp->cl_flags))
1164 		return;
1165 
1166 	cup = alloc_cld_upcall(nn);
1167 	if (!cup) {
1168 		ret = -ENOMEM;
1169 		goto out_err;
1170 	}
1171 
1172 	cmsg = &cup->cu_u.cu_msg_v2;
1173 	cmsg->cm_cmd = Cld_Create;
1174 	cmsg->cm_u.cm_clntinfo.cc_name.cn_len = clp->cl_name.len;
1175 	memcpy(cmsg->cm_u.cm_clntinfo.cc_name.cn_id, clp->cl_name.data,
1176 			clp->cl_name.len);
1177 	if (clp->cl_cred.cr_raw_principal)
1178 		principal = clp->cl_cred.cr_raw_principal;
1179 	else if (clp->cl_cred.cr_principal)
1180 		principal = clp->cl_cred.cr_principal;
1181 	if (principal) {
1182 		cksum.len = crypto_shash_digestsize(tfm);
1183 		cksum.data = kmalloc(cksum.len, GFP_KERNEL);
1184 		if (cksum.data == NULL) {
1185 			ret = -ENOMEM;
1186 			goto out;
1187 		}
1188 		ret = crypto_shash_tfm_digest(tfm, principal, strlen(principal),
1189 					      cksum.data);
1190 		if (ret) {
1191 			kfree(cksum.data);
1192 			goto out;
1193 		}
1194 		cmsg->cm_u.cm_clntinfo.cc_princhash.cp_len = cksum.len;
1195 		memcpy(cmsg->cm_u.cm_clntinfo.cc_princhash.cp_data,
1196 		       cksum.data, cksum.len);
1197 		kfree(cksum.data);
1198 	} else
1199 		cmsg->cm_u.cm_clntinfo.cc_princhash.cp_len = 0;
1200 
1201 	ret = cld_pipe_upcall(cn->cn_pipe, cmsg, nn);
1202 	if (!ret) {
1203 		ret = cmsg->cm_status;
1204 		set_bit(NFSD4_CLIENT_STABLE, &clp->cl_flags);
1205 	}
1206 
1207 out:
1208 	free_cld_upcall(cup);
1209 out_err:
1210 	if (ret)
1211 		pr_err("NFSD: Unable to create client record on stable storage: %d\n",
1212 				ret);
1213 }
1214 
1215 /* Ask daemon to create a new record */
1216 static void
1217 nfsd4_cld_remove(struct nfs4_client *clp)
1218 {
1219 	int ret;
1220 	struct cld_upcall *cup;
1221 	struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
1222 	struct cld_net *cn = nn->cld_net;
1223 
1224 	/* Don't upcall if it's already removed */
1225 	if (!test_bit(NFSD4_CLIENT_STABLE, &clp->cl_flags))
1226 		return;
1227 
1228 	cup = alloc_cld_upcall(nn);
1229 	if (!cup) {
1230 		ret = -ENOMEM;
1231 		goto out_err;
1232 	}
1233 
1234 	cup->cu_u.cu_msg.cm_cmd = Cld_Remove;
1235 	cup->cu_u.cu_msg.cm_u.cm_name.cn_len = clp->cl_name.len;
1236 	memcpy(cup->cu_u.cu_msg.cm_u.cm_name.cn_id, clp->cl_name.data,
1237 			clp->cl_name.len);
1238 
1239 	ret = cld_pipe_upcall(cn->cn_pipe, &cup->cu_u.cu_msg, nn);
1240 	if (!ret) {
1241 		ret = cup->cu_u.cu_msg.cm_status;
1242 		clear_bit(NFSD4_CLIENT_STABLE, &clp->cl_flags);
1243 	}
1244 
1245 	free_cld_upcall(cup);
1246 out_err:
1247 	if (ret)
1248 		printk(KERN_ERR "NFSD: Unable to remove client "
1249 				"record from stable storage: %d\n", ret);
1250 }
1251 
1252 /*
1253  * For older nfsdcld's that do not allow us to "slurp" the clients
1254  * from the tracking database during startup.
1255  *
1256  * Check for presence of a record, and update its timestamp
1257  */
1258 static int
1259 nfsd4_cld_check_v0(struct nfs4_client *clp)
1260 {
1261 	int ret;
1262 	struct cld_upcall *cup;
1263 	struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
1264 	struct cld_net *cn = nn->cld_net;
1265 
1266 	/* Don't upcall if one was already stored during this grace pd */
1267 	if (test_bit(NFSD4_CLIENT_STABLE, &clp->cl_flags))
1268 		return 0;
1269 
1270 	cup = alloc_cld_upcall(nn);
1271 	if (!cup) {
1272 		printk(KERN_ERR "NFSD: Unable to check client record on "
1273 				"stable storage: %d\n", -ENOMEM);
1274 		return -ENOMEM;
1275 	}
1276 
1277 	cup->cu_u.cu_msg.cm_cmd = Cld_Check;
1278 	cup->cu_u.cu_msg.cm_u.cm_name.cn_len = clp->cl_name.len;
1279 	memcpy(cup->cu_u.cu_msg.cm_u.cm_name.cn_id, clp->cl_name.data,
1280 			clp->cl_name.len);
1281 
1282 	ret = cld_pipe_upcall(cn->cn_pipe, &cup->cu_u.cu_msg, nn);
1283 	if (!ret) {
1284 		ret = cup->cu_u.cu_msg.cm_status;
1285 		set_bit(NFSD4_CLIENT_STABLE, &clp->cl_flags);
1286 	}
1287 
1288 	free_cld_upcall(cup);
1289 	return ret;
1290 }
1291 
1292 /*
1293  * For newer nfsdcld's that allow us to "slurp" the clients
1294  * from the tracking database during startup.
1295  *
1296  * Check for presence of a record in the reclaim_str_hashtbl
1297  */
1298 static int
1299 nfsd4_cld_check(struct nfs4_client *clp)
1300 {
1301 	struct nfs4_client_reclaim *crp;
1302 	struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
1303 
1304 	/* did we already find that this client is stable? */
1305 	if (test_bit(NFSD4_CLIENT_STABLE, &clp->cl_flags))
1306 		return 0;
1307 
1308 	/* look for it in the reclaim hashtable otherwise */
1309 	crp = nfsd4_find_reclaim_client(clp->cl_name, nn);
1310 	if (crp)
1311 		goto found;
1312 
1313 #ifdef CONFIG_NFSD_LEGACY_CLIENT_TRACKING
1314 	if (nn->cld_net->cn_has_legacy) {
1315 		int status;
1316 		char dname[HEXDIR_LEN];
1317 		struct xdr_netobj name;
1318 
1319 		status = nfs4_make_rec_clidname(dname, &clp->cl_name);
1320 		if (status)
1321 			return -ENOENT;
1322 
1323 		name.data = kmemdup(dname, HEXDIR_LEN, GFP_KERNEL);
1324 		if (!name.data) {
1325 			dprintk("%s: failed to allocate memory for name.data!\n",
1326 				__func__);
1327 			return -ENOENT;
1328 		}
1329 		name.len = HEXDIR_LEN;
1330 		crp = nfsd4_find_reclaim_client(name, nn);
1331 		kfree(name.data);
1332 		if (crp)
1333 			goto found;
1334 
1335 	}
1336 #endif
1337 	return -ENOENT;
1338 found:
1339 	crp->cr_clp = clp;
1340 	return 0;
1341 }
1342 
1343 static int
1344 nfsd4_cld_check_v2(struct nfs4_client *clp)
1345 {
1346 	struct nfs4_client_reclaim *crp;
1347 	struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
1348 	struct cld_net *cn = nn->cld_net;
1349 	int status;
1350 	struct crypto_shash *tfm = cn->cn_tfm;
1351 	struct xdr_netobj cksum;
1352 	char *principal = NULL;
1353 
1354 	/* did we already find that this client is stable? */
1355 	if (test_bit(NFSD4_CLIENT_STABLE, &clp->cl_flags))
1356 		return 0;
1357 
1358 	/* look for it in the reclaim hashtable otherwise */
1359 	crp = nfsd4_find_reclaim_client(clp->cl_name, nn);
1360 	if (crp)
1361 		goto found;
1362 
1363 #ifdef CONFIG_NFSD_LEGACY_CLIENT_TRACKING
1364 	if (cn->cn_has_legacy) {
1365 		struct xdr_netobj name;
1366 		char dname[HEXDIR_LEN];
1367 
1368 		status = nfs4_make_rec_clidname(dname, &clp->cl_name);
1369 		if (status)
1370 			return -ENOENT;
1371 
1372 		name.data = kmemdup(dname, HEXDIR_LEN, GFP_KERNEL);
1373 		if (!name.data) {
1374 			dprintk("%s: failed to allocate memory for name.data\n",
1375 					__func__);
1376 			return -ENOENT;
1377 		}
1378 		name.len = HEXDIR_LEN;
1379 		crp = nfsd4_find_reclaim_client(name, nn);
1380 		kfree(name.data);
1381 		if (crp)
1382 			goto found;
1383 
1384 	}
1385 #endif
1386 	return -ENOENT;
1387 found:
1388 	if (crp->cr_princhash.len) {
1389 		if (clp->cl_cred.cr_raw_principal)
1390 			principal = clp->cl_cred.cr_raw_principal;
1391 		else if (clp->cl_cred.cr_principal)
1392 			principal = clp->cl_cred.cr_principal;
1393 		if (principal == NULL)
1394 			return -ENOENT;
1395 		cksum.len = crypto_shash_digestsize(tfm);
1396 		cksum.data = kmalloc(cksum.len, GFP_KERNEL);
1397 		if (cksum.data == NULL)
1398 			return -ENOENT;
1399 		status = crypto_shash_tfm_digest(tfm, principal,
1400 						 strlen(principal), cksum.data);
1401 		if (status) {
1402 			kfree(cksum.data);
1403 			return -ENOENT;
1404 		}
1405 		if (memcmp(crp->cr_princhash.data, cksum.data,
1406 				crp->cr_princhash.len)) {
1407 			kfree(cksum.data);
1408 			return -ENOENT;
1409 		}
1410 		kfree(cksum.data);
1411 	}
1412 	crp->cr_clp = clp;
1413 	return 0;
1414 }
1415 
1416 static int
1417 nfsd4_cld_grace_start(struct nfsd_net *nn)
1418 {
1419 	int ret;
1420 	struct cld_upcall *cup;
1421 	struct cld_net *cn = nn->cld_net;
1422 
1423 	cup = alloc_cld_upcall(nn);
1424 	if (!cup) {
1425 		ret = -ENOMEM;
1426 		goto out_err;
1427 	}
1428 
1429 	cup->cu_u.cu_msg.cm_cmd = Cld_GraceStart;
1430 	ret = cld_pipe_upcall(cn->cn_pipe, &cup->cu_u.cu_msg, nn);
1431 	if (!ret)
1432 		ret = cup->cu_u.cu_msg.cm_status;
1433 
1434 	free_cld_upcall(cup);
1435 out_err:
1436 	if (ret)
1437 		dprintk("%s: Unable to get clients from userspace: %d\n",
1438 			__func__, ret);
1439 	return ret;
1440 }
1441 
1442 /* For older nfsdcld's that need cm_gracetime */
1443 static void
1444 nfsd4_cld_grace_done_v0(struct nfsd_net *nn)
1445 {
1446 	int ret;
1447 	struct cld_upcall *cup;
1448 	struct cld_net *cn = nn->cld_net;
1449 
1450 	cup = alloc_cld_upcall(nn);
1451 	if (!cup) {
1452 		ret = -ENOMEM;
1453 		goto out_err;
1454 	}
1455 
1456 	cup->cu_u.cu_msg.cm_cmd = Cld_GraceDone;
1457 	cup->cu_u.cu_msg.cm_u.cm_gracetime = nn->boot_time;
1458 	ret = cld_pipe_upcall(cn->cn_pipe, &cup->cu_u.cu_msg, nn);
1459 	if (!ret)
1460 		ret = cup->cu_u.cu_msg.cm_status;
1461 
1462 	free_cld_upcall(cup);
1463 out_err:
1464 	if (ret)
1465 		printk(KERN_ERR "NFSD: Unable to end grace period: %d\n", ret);
1466 }
1467 
1468 /*
1469  * For newer nfsdcld's that do not need cm_gracetime.  We also need to call
1470  * nfs4_release_reclaim() to clear out the reclaim_str_hashtbl.
1471  */
1472 static void
1473 nfsd4_cld_grace_done(struct nfsd_net *nn)
1474 {
1475 	int ret;
1476 	struct cld_upcall *cup;
1477 	struct cld_net *cn = nn->cld_net;
1478 
1479 	cup = alloc_cld_upcall(nn);
1480 	if (!cup) {
1481 		ret = -ENOMEM;
1482 		goto out_err;
1483 	}
1484 
1485 	cup->cu_u.cu_msg.cm_cmd = Cld_GraceDone;
1486 	ret = cld_pipe_upcall(cn->cn_pipe, &cup->cu_u.cu_msg, nn);
1487 	if (!ret)
1488 		ret = cup->cu_u.cu_msg.cm_status;
1489 
1490 	free_cld_upcall(cup);
1491 out_err:
1492 	nfs4_release_reclaim(nn);
1493 	if (ret)
1494 		printk(KERN_ERR "NFSD: Unable to end grace period: %d\n", ret);
1495 }
1496 
1497 static int
1498 nfs4_cld_state_init(struct net *net)
1499 {
1500 	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
1501 	int i;
1502 
1503 	nn->reclaim_str_hashtbl = kmalloc_array(CLIENT_HASH_SIZE,
1504 						sizeof(struct list_head),
1505 						GFP_KERNEL);
1506 	if (!nn->reclaim_str_hashtbl)
1507 		return -ENOMEM;
1508 
1509 	for (i = 0; i < CLIENT_HASH_SIZE; i++)
1510 		INIT_LIST_HEAD(&nn->reclaim_str_hashtbl[i]);
1511 	nn->reclaim_str_hashtbl_size = 0;
1512 	nn->track_reclaim_completes = true;
1513 	atomic_set(&nn->nr_reclaim_complete, 0);
1514 
1515 	return 0;
1516 }
1517 
1518 static void
1519 nfs4_cld_state_shutdown(struct net *net)
1520 {
1521 	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
1522 
1523 	nn->track_reclaim_completes = false;
1524 	kfree(nn->reclaim_str_hashtbl);
1525 }
1526 
1527 static bool
1528 cld_running(struct nfsd_net *nn)
1529 {
1530 	struct cld_net *cn = nn->cld_net;
1531 	struct rpc_pipe *pipe = cn->cn_pipe;
1532 
1533 	return pipe->nreaders || pipe->nwriters;
1534 }
1535 
1536 static int
1537 nfsd4_cld_get_version(struct nfsd_net *nn)
1538 {
1539 	int ret = 0;
1540 	struct cld_upcall *cup;
1541 	struct cld_net *cn = nn->cld_net;
1542 	uint8_t version;
1543 
1544 	cup = alloc_cld_upcall(nn);
1545 	if (!cup) {
1546 		ret = -ENOMEM;
1547 		goto out_err;
1548 	}
1549 	cup->cu_u.cu_msg.cm_cmd = Cld_GetVersion;
1550 	ret = cld_pipe_upcall(cn->cn_pipe, &cup->cu_u.cu_msg, nn);
1551 	if (!ret) {
1552 		ret = cup->cu_u.cu_msg.cm_status;
1553 		if (ret)
1554 			goto out_free;
1555 		version = cup->cu_u.cu_msg.cm_u.cm_version;
1556 		dprintk("%s: userspace returned version %u\n",
1557 				__func__, version);
1558 		if (version < 1)
1559 			version = 1;
1560 		else if (version > CLD_UPCALL_VERSION)
1561 			version = CLD_UPCALL_VERSION;
1562 
1563 		switch (version) {
1564 		case 1:
1565 			nn->client_tracking_ops = &nfsd4_cld_tracking_ops;
1566 			break;
1567 		case 2:
1568 			nn->client_tracking_ops = &nfsd4_cld_tracking_ops_v2;
1569 			break;
1570 		default:
1571 			break;
1572 		}
1573 	}
1574 out_free:
1575 	free_cld_upcall(cup);
1576 out_err:
1577 	if (ret)
1578 		dprintk("%s: Unable to get version from userspace: %d\n",
1579 			__func__, ret);
1580 	return ret;
1581 }
1582 
1583 static int
1584 nfsd4_cld_tracking_init(struct net *net)
1585 {
1586 	int status;
1587 	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
1588 	bool running;
1589 	int retries = 10;
1590 	struct crypto_shash *tfm;
1591 
1592 	status = nfs4_cld_state_init(net);
1593 	if (status)
1594 		return status;
1595 
1596 	status = __nfsd4_init_cld_pipe(net);
1597 	if (status)
1598 		goto err_shutdown;
1599 
1600 	/*
1601 	 * rpc pipe upcalls take 30 seconds to time out, so we don't want to
1602 	 * queue an upcall unless we know that nfsdcld is running (because we
1603 	 * want this to fail fast so that nfsd4_client_tracking_init() can try
1604 	 * the next client tracking method).  nfsdcld should already be running
1605 	 * before nfsd is started, so the wait here is for nfsdcld to open the
1606 	 * pipefs file we just created.
1607 	 */
1608 	while (!(running = cld_running(nn)) && retries--)
1609 		msleep(100);
1610 
1611 	if (!running) {
1612 		status = -ETIMEDOUT;
1613 		goto err_remove;
1614 	}
1615 	tfm = crypto_alloc_shash("sha256", 0, 0);
1616 	if (IS_ERR(tfm)) {
1617 		status = PTR_ERR(tfm);
1618 		goto err_remove;
1619 	}
1620 	nn->cld_net->cn_tfm = tfm;
1621 
1622 	status = nfsd4_cld_get_version(nn);
1623 	if (status == -EOPNOTSUPP)
1624 		pr_warn("NFSD: nfsdcld GetVersion upcall failed. Please upgrade nfsdcld.\n");
1625 
1626 	status = nfsd4_cld_grace_start(nn);
1627 	if (status) {
1628 		if (status == -EOPNOTSUPP)
1629 			pr_warn("NFSD: nfsdcld GraceStart upcall failed. Please upgrade nfsdcld.\n");
1630 		nfs4_release_reclaim(nn);
1631 		goto err_remove;
1632 	} else
1633 		pr_info("NFSD: Using nfsdcld client tracking operations.\n");
1634 	return 0;
1635 
1636 err_remove:
1637 	nfsd4_remove_cld_pipe(net);
1638 err_shutdown:
1639 	nfs4_cld_state_shutdown(net);
1640 	return status;
1641 }
1642 
1643 static void
1644 nfsd4_cld_tracking_exit(struct net *net)
1645 {
1646 	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
1647 
1648 	nfs4_release_reclaim(nn);
1649 	nfsd4_remove_cld_pipe(net);
1650 	nfs4_cld_state_shutdown(net);
1651 }
1652 
1653 /* For older nfsdcld's */
1654 static const struct nfsd4_client_tracking_ops nfsd4_cld_tracking_ops_v0 = {
1655 	.init		= nfsd4_init_cld_pipe,
1656 	.exit		= nfsd4_remove_cld_pipe,
1657 	.create		= nfsd4_cld_create,
1658 	.remove		= nfsd4_cld_remove,
1659 	.check		= nfsd4_cld_check_v0,
1660 	.grace_done	= nfsd4_cld_grace_done_v0,
1661 	.version	= 1,
1662 	.msglen		= sizeof(struct cld_msg),
1663 };
1664 
1665 /* For newer nfsdcld's */
1666 static const struct nfsd4_client_tracking_ops nfsd4_cld_tracking_ops = {
1667 	.init		= nfsd4_cld_tracking_init,
1668 	.exit		= nfsd4_cld_tracking_exit,
1669 	.create		= nfsd4_cld_create,
1670 	.remove		= nfsd4_cld_remove,
1671 	.check		= nfsd4_cld_check,
1672 	.grace_done	= nfsd4_cld_grace_done,
1673 	.version	= 1,
1674 	.msglen		= sizeof(struct cld_msg),
1675 };
1676 
1677 /* v2 create/check ops include the principal, if available */
1678 static const struct nfsd4_client_tracking_ops nfsd4_cld_tracking_ops_v2 = {
1679 	.init		= nfsd4_cld_tracking_init,
1680 	.exit		= nfsd4_cld_tracking_exit,
1681 	.create		= nfsd4_cld_create_v2,
1682 	.remove		= nfsd4_cld_remove,
1683 	.check		= nfsd4_cld_check_v2,
1684 	.grace_done	= nfsd4_cld_grace_done,
1685 	.version	= 2,
1686 	.msglen		= sizeof(struct cld_msg_v2),
1687 };
1688 
1689 #ifdef CONFIG_NFSD_LEGACY_CLIENT_TRACKING
1690 /* upcall via usermodehelper */
1691 static char cltrack_prog[PATH_MAX] = "/sbin/nfsdcltrack";
1692 module_param_string(cltrack_prog, cltrack_prog, sizeof(cltrack_prog),
1693 			S_IRUGO|S_IWUSR);
1694 MODULE_PARM_DESC(cltrack_prog, "Path to the nfsdcltrack upcall program");
1695 
1696 static bool cltrack_legacy_disable;
1697 module_param(cltrack_legacy_disable, bool, S_IRUGO|S_IWUSR);
1698 MODULE_PARM_DESC(cltrack_legacy_disable,
1699 		"Disable legacy recoverydir conversion. Default: false");
1700 
1701 #define LEGACY_TOPDIR_ENV_PREFIX "NFSDCLTRACK_LEGACY_TOPDIR="
1702 #define LEGACY_RECDIR_ENV_PREFIX "NFSDCLTRACK_LEGACY_RECDIR="
1703 #define HAS_SESSION_ENV_PREFIX "NFSDCLTRACK_CLIENT_HAS_SESSION="
1704 #define GRACE_START_ENV_PREFIX "NFSDCLTRACK_GRACE_START="
1705 
1706 static char *
1707 nfsd4_cltrack_legacy_topdir(void)
1708 {
1709 	int copied;
1710 	size_t len;
1711 	char *result;
1712 
1713 	if (cltrack_legacy_disable)
1714 		return NULL;
1715 
1716 	len = strlen(LEGACY_TOPDIR_ENV_PREFIX) +
1717 		strlen(nfs4_recoverydir()) + 1;
1718 
1719 	result = kmalloc(len, GFP_KERNEL);
1720 	if (!result)
1721 		return result;
1722 
1723 	copied = snprintf(result, len, LEGACY_TOPDIR_ENV_PREFIX "%s",
1724 				nfs4_recoverydir());
1725 	if (copied >= len) {
1726 		/* just return nothing if output was truncated */
1727 		kfree(result);
1728 		return NULL;
1729 	}
1730 
1731 	return result;
1732 }
1733 
1734 static char *
1735 nfsd4_cltrack_legacy_recdir(const struct xdr_netobj *name)
1736 {
1737 	int copied;
1738 	size_t len;
1739 	char *result;
1740 
1741 	if (cltrack_legacy_disable)
1742 		return NULL;
1743 
1744 	/* +1 is for '/' between "topdir" and "recdir" */
1745 	len = strlen(LEGACY_RECDIR_ENV_PREFIX) +
1746 		strlen(nfs4_recoverydir()) + 1 + HEXDIR_LEN;
1747 
1748 	result = kmalloc(len, GFP_KERNEL);
1749 	if (!result)
1750 		return result;
1751 
1752 	copied = snprintf(result, len, LEGACY_RECDIR_ENV_PREFIX "%s/",
1753 				nfs4_recoverydir());
1754 	if (copied > (len - HEXDIR_LEN)) {
1755 		/* just return nothing if output will be truncated */
1756 		kfree(result);
1757 		return NULL;
1758 	}
1759 
1760 	copied = nfs4_make_rec_clidname(result + copied, name);
1761 	if (copied) {
1762 		kfree(result);
1763 		return NULL;
1764 	}
1765 
1766 	return result;
1767 }
1768 
1769 static char *
1770 nfsd4_cltrack_client_has_session(struct nfs4_client *clp)
1771 {
1772 	int copied;
1773 	size_t len;
1774 	char *result;
1775 
1776 	/* prefix + Y/N character + terminating NULL */
1777 	len = strlen(HAS_SESSION_ENV_PREFIX) + 1 + 1;
1778 
1779 	result = kmalloc(len, GFP_KERNEL);
1780 	if (!result)
1781 		return result;
1782 
1783 	copied = snprintf(result, len, HAS_SESSION_ENV_PREFIX "%c",
1784 				clp->cl_minorversion ? 'Y' : 'N');
1785 	if (copied >= len) {
1786 		/* just return nothing if output was truncated */
1787 		kfree(result);
1788 		return NULL;
1789 	}
1790 
1791 	return result;
1792 }
1793 
1794 static char *
1795 nfsd4_cltrack_grace_start(time64_t grace_start)
1796 {
1797 	int copied;
1798 	size_t len;
1799 	char *result;
1800 
1801 	/* prefix + max width of int64_t string + terminating NULL */
1802 	len = strlen(GRACE_START_ENV_PREFIX) + 22 + 1;
1803 
1804 	result = kmalloc(len, GFP_KERNEL);
1805 	if (!result)
1806 		return result;
1807 
1808 	copied = snprintf(result, len, GRACE_START_ENV_PREFIX "%lld",
1809 				grace_start);
1810 	if (copied >= len) {
1811 		/* just return nothing if output was truncated */
1812 		kfree(result);
1813 		return NULL;
1814 	}
1815 
1816 	return result;
1817 }
1818 
1819 static int
1820 nfsd4_umh_cltrack_upcall(char *cmd, char *arg, char *env0, char *env1)
1821 {
1822 	char *envp[3];
1823 	char *argv[4];
1824 	int ret;
1825 
1826 	if (unlikely(!cltrack_prog[0])) {
1827 		dprintk("%s: cltrack_prog is disabled\n", __func__);
1828 		return -EACCES;
1829 	}
1830 
1831 	dprintk("%s: cmd: %s\n", __func__, cmd);
1832 	dprintk("%s: arg: %s\n", __func__, arg ? arg : "(null)");
1833 	dprintk("%s: env0: %s\n", __func__, env0 ? env0 : "(null)");
1834 	dprintk("%s: env1: %s\n", __func__, env1 ? env1 : "(null)");
1835 
1836 	envp[0] = env0;
1837 	envp[1] = env1;
1838 	envp[2] = NULL;
1839 
1840 	argv[0] = (char *)cltrack_prog;
1841 	argv[1] = cmd;
1842 	argv[2] = arg;
1843 	argv[3] = NULL;
1844 
1845 	ret = call_usermodehelper(argv[0], argv, envp, UMH_WAIT_PROC);
1846 	/*
1847 	 * Disable the upcall mechanism if we're getting an ENOENT or EACCES
1848 	 * error. The admin can re-enable it on the fly by using sysfs
1849 	 * once the problem has been fixed.
1850 	 */
1851 	if (ret == -ENOENT || ret == -EACCES) {
1852 		dprintk("NFSD: %s was not found or isn't executable (%d). "
1853 			"Setting cltrack_prog to blank string!",
1854 			cltrack_prog, ret);
1855 		cltrack_prog[0] = '\0';
1856 	}
1857 	dprintk("%s: %s return value: %d\n", __func__, cltrack_prog, ret);
1858 
1859 	return ret;
1860 }
1861 
1862 static char *
1863 bin_to_hex_dup(const unsigned char *src, int srclen)
1864 {
1865 	char *buf;
1866 
1867 	/* +1 for terminating NULL */
1868 	buf = kzalloc((srclen * 2) + 1, GFP_KERNEL);
1869 	if (!buf)
1870 		return buf;
1871 
1872 	bin2hex(buf, src, srclen);
1873 	return buf;
1874 }
1875 
1876 static int
1877 nfsd4_umh_cltrack_init(struct net *net)
1878 {
1879 	int ret;
1880 	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
1881 	char *grace_start = nfsd4_cltrack_grace_start(nn->boot_time);
1882 
1883 	/* XXX: The usermode helper s not working in container yet. */
1884 	if (net != &init_net) {
1885 		pr_warn("NFSD: attempt to initialize umh client tracking in a container ignored.\n");
1886 		kfree(grace_start);
1887 		return -EINVAL;
1888 	}
1889 
1890 	ret = nfsd4_umh_cltrack_upcall("init", NULL, grace_start, NULL);
1891 	kfree(grace_start);
1892 	if (!ret)
1893 		pr_info("NFSD: Using UMH upcall client tracking operations.\n");
1894 	return ret;
1895 }
1896 
1897 static void
1898 nfsd4_cltrack_upcall_lock(struct nfs4_client *clp)
1899 {
1900 	wait_on_bit_lock(&clp->cl_flags, NFSD4_CLIENT_UPCALL_LOCK,
1901 			 TASK_UNINTERRUPTIBLE);
1902 }
1903 
1904 static void
1905 nfsd4_cltrack_upcall_unlock(struct nfs4_client *clp)
1906 {
1907 	clear_and_wake_up_bit(NFSD4_CLIENT_UPCALL_LOCK, &clp->cl_flags);
1908 }
1909 
1910 static void
1911 nfsd4_umh_cltrack_create(struct nfs4_client *clp)
1912 {
1913 	char *hexid, *has_session, *grace_start;
1914 	struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
1915 
1916 	/*
1917 	 * With v4.0 clients, there's little difference in outcome between a
1918 	 * create and check operation, and we can end up calling into this
1919 	 * function multiple times per client (once for each openowner). So,
1920 	 * for v4.0 clients skip upcalling once the client has been recorded
1921 	 * on stable storage.
1922 	 *
1923 	 * For v4.1+ clients, the outcome of the two operations is different,
1924 	 * so we must ensure that we upcall for the create operation. v4.1+
1925 	 * clients call this on RECLAIM_COMPLETE though, so we should only end
1926 	 * up doing a single create upcall per client.
1927 	 */
1928 	if (clp->cl_minorversion == 0 &&
1929 	    test_bit(NFSD4_CLIENT_STABLE, &clp->cl_flags))
1930 		return;
1931 
1932 	hexid = bin_to_hex_dup(clp->cl_name.data, clp->cl_name.len);
1933 	if (!hexid) {
1934 		dprintk("%s: can't allocate memory for upcall!\n", __func__);
1935 		return;
1936 	}
1937 
1938 	has_session = nfsd4_cltrack_client_has_session(clp);
1939 	grace_start = nfsd4_cltrack_grace_start(nn->boot_time);
1940 
1941 	nfsd4_cltrack_upcall_lock(clp);
1942 	if (!nfsd4_umh_cltrack_upcall("create", hexid, has_session, grace_start))
1943 		set_bit(NFSD4_CLIENT_STABLE, &clp->cl_flags);
1944 	nfsd4_cltrack_upcall_unlock(clp);
1945 
1946 	kfree(has_session);
1947 	kfree(grace_start);
1948 	kfree(hexid);
1949 }
1950 
1951 static void
1952 nfsd4_umh_cltrack_remove(struct nfs4_client *clp)
1953 {
1954 	char *hexid;
1955 
1956 	if (!test_bit(NFSD4_CLIENT_STABLE, &clp->cl_flags))
1957 		return;
1958 
1959 	hexid = bin_to_hex_dup(clp->cl_name.data, clp->cl_name.len);
1960 	if (!hexid) {
1961 		dprintk("%s: can't allocate memory for upcall!\n", __func__);
1962 		return;
1963 	}
1964 
1965 	nfsd4_cltrack_upcall_lock(clp);
1966 	if (test_bit(NFSD4_CLIENT_STABLE, &clp->cl_flags) &&
1967 	    nfsd4_umh_cltrack_upcall("remove", hexid, NULL, NULL) == 0)
1968 		clear_bit(NFSD4_CLIENT_STABLE, &clp->cl_flags);
1969 	nfsd4_cltrack_upcall_unlock(clp);
1970 
1971 	kfree(hexid);
1972 }
1973 
1974 static int
1975 nfsd4_umh_cltrack_check(struct nfs4_client *clp)
1976 {
1977 	int ret;
1978 	char *hexid, *has_session, *legacy;
1979 
1980 	if (test_bit(NFSD4_CLIENT_STABLE, &clp->cl_flags))
1981 		return 0;
1982 
1983 	hexid = bin_to_hex_dup(clp->cl_name.data, clp->cl_name.len);
1984 	if (!hexid) {
1985 		dprintk("%s: can't allocate memory for upcall!\n", __func__);
1986 		return -ENOMEM;
1987 	}
1988 
1989 	has_session = nfsd4_cltrack_client_has_session(clp);
1990 	legacy = nfsd4_cltrack_legacy_recdir(&clp->cl_name);
1991 
1992 	nfsd4_cltrack_upcall_lock(clp);
1993 	if (test_bit(NFSD4_CLIENT_STABLE, &clp->cl_flags)) {
1994 		ret = 0;
1995 	} else {
1996 		ret = nfsd4_umh_cltrack_upcall("check", hexid, has_session, legacy);
1997 		if (ret == 0)
1998 			set_bit(NFSD4_CLIENT_STABLE, &clp->cl_flags);
1999 	}
2000 	nfsd4_cltrack_upcall_unlock(clp);
2001 	kfree(has_session);
2002 	kfree(legacy);
2003 	kfree(hexid);
2004 
2005 	return ret;
2006 }
2007 
2008 static void
2009 nfsd4_umh_cltrack_grace_done(struct nfsd_net *nn)
2010 {
2011 	char *legacy;
2012 	char timestr[22]; /* FIXME: better way to determine max size? */
2013 
2014 	sprintf(timestr, "%lld", nn->boot_time);
2015 	legacy = nfsd4_cltrack_legacy_topdir();
2016 	nfsd4_umh_cltrack_upcall("gracedone", timestr, legacy, NULL);
2017 	kfree(legacy);
2018 }
2019 
2020 static const struct nfsd4_client_tracking_ops nfsd4_umh_tracking_ops = {
2021 	.init		= nfsd4_umh_cltrack_init,
2022 	.exit		= NULL,
2023 	.create		= nfsd4_umh_cltrack_create,
2024 	.remove		= nfsd4_umh_cltrack_remove,
2025 	.check		= nfsd4_umh_cltrack_check,
2026 	.grace_done	= nfsd4_umh_cltrack_grace_done,
2027 	.version	= 1,
2028 	.msglen		= 0,
2029 };
2030 
2031 static inline int check_for_legacy_methods(int status, struct net *net)
2032 {
2033 	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
2034 	struct path path;
2035 
2036 	/*
2037 	 * Next, try the UMH upcall.
2038 	 */
2039 	nn->client_tracking_ops = &nfsd4_umh_tracking_ops;
2040 	status = nn->client_tracking_ops->init(net);
2041 	if (!status)
2042 		return status;
2043 
2044 	/*
2045 	 * Finally, See if the recoverydir exists and is a directory.
2046 	 * If it is, then use the legacy ops.
2047 	 */
2048 	nn->client_tracking_ops = &nfsd4_legacy_tracking_ops;
2049 	status = kern_path(nfs4_recoverydir(), LOOKUP_FOLLOW, &path);
2050 	if (!status) {
2051 		status = !d_is_dir(path.dentry);
2052 		path_put(&path);
2053 		if (status)
2054 			return -ENOTDIR;
2055 		status = nn->client_tracking_ops->init(net);
2056 	}
2057 	return status;
2058 }
2059 #else
2060 static inline int check_for_legacy_methods(int status, struct net *net)
2061 {
2062 	return status;
2063 }
2064 #endif /* CONFIG_LEGACY_NFSD_CLIENT_TRACKING */
2065 
2066 int
2067 nfsd4_client_tracking_init(struct net *net)
2068 {
2069 	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
2070 	int status;
2071 
2072 	/* just run the init if it the method is already decided */
2073 	if (nn->client_tracking_ops)
2074 		goto do_init;
2075 
2076 	/* First, try to use nfsdcld */
2077 	nn->client_tracking_ops = &nfsd4_cld_tracking_ops;
2078 	status = nn->client_tracking_ops->init(net);
2079 	if (!status)
2080 		return status;
2081 	if (status != -ETIMEDOUT) {
2082 		nn->client_tracking_ops = &nfsd4_cld_tracking_ops_v0;
2083 		status = nn->client_tracking_ops->init(net);
2084 		if (!status)
2085 			return status;
2086 	}
2087 
2088 	status = check_for_legacy_methods(status, net);
2089 	if (status)
2090 		goto out;
2091 do_init:
2092 	status = nn->client_tracking_ops->init(net);
2093 out:
2094 	if (status) {
2095 		pr_warn("NFSD: Unable to initialize client recovery tracking! (%d)\n", status);
2096 		pr_warn("NFSD: Is nfsdcld running? If not, enable CONFIG_NFSD_LEGACY_CLIENT_TRACKING.\n");
2097 		nn->client_tracking_ops = NULL;
2098 	}
2099 	return status;
2100 }
2101 
2102 void
2103 nfsd4_client_tracking_exit(struct net *net)
2104 {
2105 	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
2106 
2107 	if (nn->client_tracking_ops) {
2108 		if (nn->client_tracking_ops->exit)
2109 			nn->client_tracking_ops->exit(net);
2110 		nn->client_tracking_ops = NULL;
2111 	}
2112 }
2113 
2114 void
2115 nfsd4_client_record_create(struct nfs4_client *clp)
2116 {
2117 	struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
2118 
2119 	if (nn->client_tracking_ops)
2120 		nn->client_tracking_ops->create(clp);
2121 }
2122 
2123 void
2124 nfsd4_client_record_remove(struct nfs4_client *clp)
2125 {
2126 	struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
2127 
2128 	if (nn->client_tracking_ops)
2129 		nn->client_tracking_ops->remove(clp);
2130 }
2131 
2132 int
2133 nfsd4_client_record_check(struct nfs4_client *clp)
2134 {
2135 	struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
2136 
2137 	if (nn->client_tracking_ops)
2138 		return nn->client_tracking_ops->check(clp);
2139 
2140 	return -EOPNOTSUPP;
2141 }
2142 
2143 void
2144 nfsd4_record_grace_done(struct nfsd_net *nn)
2145 {
2146 	if (nn->client_tracking_ops)
2147 		nn->client_tracking_ops->grace_done(nn);
2148 }
2149 
2150 static int
2151 rpc_pipefs_event(struct notifier_block *nb, unsigned long event, void *ptr)
2152 {
2153 	struct super_block *sb = ptr;
2154 	struct net *net = sb->s_fs_info;
2155 	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
2156 	struct cld_net *cn = nn->cld_net;
2157 	struct dentry *dentry;
2158 	int ret = 0;
2159 
2160 	if (!try_module_get(THIS_MODULE))
2161 		return 0;
2162 
2163 	if (!cn) {
2164 		module_put(THIS_MODULE);
2165 		return 0;
2166 	}
2167 
2168 	switch (event) {
2169 	case RPC_PIPEFS_MOUNT:
2170 		dentry = nfsd4_cld_register_sb(sb, cn->cn_pipe);
2171 		if (IS_ERR(dentry)) {
2172 			ret = PTR_ERR(dentry);
2173 			break;
2174 		}
2175 		cn->cn_pipe->dentry = dentry;
2176 		break;
2177 	case RPC_PIPEFS_UMOUNT:
2178 		if (cn->cn_pipe->dentry)
2179 			nfsd4_cld_unregister_sb(cn->cn_pipe);
2180 		break;
2181 	default:
2182 		ret = -ENOTSUPP;
2183 		break;
2184 	}
2185 	module_put(THIS_MODULE);
2186 	return ret;
2187 }
2188 
2189 static struct notifier_block nfsd4_cld_block = {
2190 	.notifier_call = rpc_pipefs_event,
2191 };
2192 
2193 int
2194 register_cld_notifier(void)
2195 {
2196 	WARN_ON(!nfsd_net_id);
2197 	return rpc_pipefs_notifier_register(&nfsd4_cld_block);
2198 }
2199 
2200 void
2201 unregister_cld_notifier(void)
2202 {
2203 	rpc_pipefs_notifier_unregister(&nfsd4_cld_block);
2204 }
2205