xref: /linux/fs/nfsd/nfs4recover.c (revision 109d59b900e78834c66657dd4748fcedb9a1fe8d)
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 };
63 
64 /* Globals */
65 static char user_recovery_dirname[PATH_MAX] = "/var/lib/nfs/v4recovery";
66 
67 static int
68 nfs4_save_creds(const struct cred **original_creds)
69 {
70 	struct cred *new;
71 
72 	new = prepare_creds();
73 	if (!new)
74 		return -ENOMEM;
75 
76 	new->fsuid = GLOBAL_ROOT_UID;
77 	new->fsgid = GLOBAL_ROOT_GID;
78 	*original_creds = override_creds(new);
79 	put_cred(new);
80 	return 0;
81 }
82 
83 static void
84 nfs4_reset_creds(const struct cred *original)
85 {
86 	revert_creds(original);
87 }
88 
89 static void
90 md5_to_hex(char *out, char *md5)
91 {
92 	int i;
93 
94 	for (i=0; i<16; i++) {
95 		unsigned char c = md5[i];
96 
97 		*out++ = '0' + ((c&0xf0)>>4) + (c>=0xa0)*('a'-'9'-1);
98 		*out++ = '0' + (c&0x0f) + ((c&0x0f)>=0x0a)*('a'-'9'-1);
99 	}
100 	*out = '\0';
101 }
102 
103 static int
104 nfs4_make_rec_clidname(char *dname, const struct xdr_netobj *clname)
105 {
106 	struct xdr_netobj cksum;
107 	struct crypto_shash *tfm;
108 	int status;
109 
110 	dprintk("NFSD: nfs4_make_rec_clidname for %.*s\n",
111 			clname->len, clname->data);
112 	tfm = crypto_alloc_shash("md5", 0, 0);
113 	if (IS_ERR(tfm)) {
114 		status = PTR_ERR(tfm);
115 		goto out_no_tfm;
116 	}
117 
118 	cksum.len = crypto_shash_digestsize(tfm);
119 	cksum.data = kmalloc(cksum.len, GFP_KERNEL);
120 	if (cksum.data == NULL) {
121 		status = -ENOMEM;
122  		goto out;
123 	}
124 
125 	{
126 		SHASH_DESC_ON_STACK(desc, tfm);
127 
128 		desc->tfm = tfm;
129 		desc->flags = CRYPTO_TFM_REQ_MAY_SLEEP;
130 
131 		status = crypto_shash_digest(desc, clname->data, clname->len,
132 					     cksum.data);
133 		shash_desc_zero(desc);
134 	}
135 
136 	if (status)
137 		goto out;
138 
139 	md5_to_hex(dname, cksum.data);
140 
141 	status = 0;
142 out:
143 	kfree(cksum.data);
144 	crypto_free_shash(tfm);
145 out_no_tfm:
146 	return status;
147 }
148 
149 /*
150  * If we had an error generating the recdir name for the legacy tracker
151  * then warn the admin. If the error doesn't appear to be transient,
152  * then disable recovery tracking.
153  */
154 static void
155 legacy_recdir_name_error(struct nfs4_client *clp, int error)
156 {
157 	printk(KERN_ERR "NFSD: unable to generate recoverydir "
158 			"name (%d).\n", error);
159 
160 	/*
161 	 * if the algorithm just doesn't exist, then disable the recovery
162 	 * tracker altogether. The crypto libs will generally return this if
163 	 * FIPS is enabled as well.
164 	 */
165 	if (error == -ENOENT) {
166 		printk(KERN_ERR "NFSD: disabling legacy clientid tracking. "
167 			"Reboot recovery will not function correctly!\n");
168 		nfsd4_client_tracking_exit(clp->net);
169 	}
170 }
171 
172 static void
173 nfsd4_create_clid_dir(struct nfs4_client *clp)
174 {
175 	const struct cred *original_cred;
176 	char dname[HEXDIR_LEN];
177 	struct dentry *dir, *dentry;
178 	struct nfs4_client_reclaim *crp;
179 	int status;
180 	struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
181 
182 	if (test_and_set_bit(NFSD4_CLIENT_STABLE, &clp->cl_flags))
183 		return;
184 	if (!nn->rec_file)
185 		return;
186 
187 	status = nfs4_make_rec_clidname(dname, &clp->cl_name);
188 	if (status)
189 		return legacy_recdir_name_error(clp, status);
190 
191 	status = nfs4_save_creds(&original_cred);
192 	if (status < 0)
193 		return;
194 
195 	status = mnt_want_write_file(nn->rec_file);
196 	if (status)
197 		goto out_creds;
198 
199 	dir = nn->rec_file->f_path.dentry;
200 	/* lock the parent */
201 	inode_lock(d_inode(dir));
202 
203 	dentry = lookup_one_len(dname, dir, HEXDIR_LEN-1);
204 	if (IS_ERR(dentry)) {
205 		status = PTR_ERR(dentry);
206 		goto out_unlock;
207 	}
208 	if (d_really_is_positive(dentry))
209 		/*
210 		 * In the 4.1 case, where we're called from
211 		 * reclaim_complete(), records from the previous reboot
212 		 * may still be left, so this is OK.
213 		 *
214 		 * In the 4.0 case, we should never get here; but we may
215 		 * as well be forgiving and just succeed silently.
216 		 */
217 		goto out_put;
218 	status = vfs_mkdir(d_inode(dir), dentry, S_IRWXU);
219 out_put:
220 	dput(dentry);
221 out_unlock:
222 	inode_unlock(d_inode(dir));
223 	if (status == 0) {
224 		if (nn->in_grace) {
225 			crp = nfs4_client_to_reclaim(dname, nn);
226 			if (crp)
227 				crp->cr_clp = clp;
228 		}
229 		vfs_fsync(nn->rec_file, 0);
230 	} else {
231 		printk(KERN_ERR "NFSD: failed to write recovery record"
232 				" (err %d); please check that %s exists"
233 				" and is writeable", status,
234 				user_recovery_dirname);
235 	}
236 	mnt_drop_write_file(nn->rec_file);
237 out_creds:
238 	nfs4_reset_creds(original_cred);
239 }
240 
241 typedef int (recdir_func)(struct dentry *, struct dentry *, struct nfsd_net *);
242 
243 struct name_list {
244 	char name[HEXDIR_LEN];
245 	struct list_head list;
246 };
247 
248 struct nfs4_dir_ctx {
249 	struct dir_context ctx;
250 	struct list_head names;
251 };
252 
253 static int
254 nfsd4_build_namelist(struct dir_context *__ctx, const char *name, int namlen,
255 		loff_t offset, u64 ino, unsigned int d_type)
256 {
257 	struct nfs4_dir_ctx *ctx =
258 		container_of(__ctx, struct nfs4_dir_ctx, ctx);
259 	struct name_list *entry;
260 
261 	if (namlen != HEXDIR_LEN - 1)
262 		return 0;
263 	entry = kmalloc(sizeof(struct name_list), GFP_KERNEL);
264 	if (entry == NULL)
265 		return -ENOMEM;
266 	memcpy(entry->name, name, HEXDIR_LEN - 1);
267 	entry->name[HEXDIR_LEN - 1] = '\0';
268 	list_add(&entry->list, &ctx->names);
269 	return 0;
270 }
271 
272 static int
273 nfsd4_list_rec_dir(recdir_func *f, struct nfsd_net *nn)
274 {
275 	const struct cred *original_cred;
276 	struct dentry *dir = nn->rec_file->f_path.dentry;
277 	struct nfs4_dir_ctx ctx = {
278 		.ctx.actor = nfsd4_build_namelist,
279 		.names = LIST_HEAD_INIT(ctx.names)
280 	};
281 	struct name_list *entry, *tmp;
282 	int status;
283 
284 	status = nfs4_save_creds(&original_cred);
285 	if (status < 0)
286 		return status;
287 
288 	status = vfs_llseek(nn->rec_file, 0, SEEK_SET);
289 	if (status < 0) {
290 		nfs4_reset_creds(original_cred);
291 		return status;
292 	}
293 
294 	status = iterate_dir(nn->rec_file, &ctx.ctx);
295 	inode_lock_nested(d_inode(dir), I_MUTEX_PARENT);
296 
297 	list_for_each_entry_safe(entry, tmp, &ctx.names, list) {
298 		if (!status) {
299 			struct dentry *dentry;
300 			dentry = lookup_one_len(entry->name, dir, HEXDIR_LEN-1);
301 			if (IS_ERR(dentry)) {
302 				status = PTR_ERR(dentry);
303 				break;
304 			}
305 			status = f(dir, dentry, nn);
306 			dput(dentry);
307 		}
308 		list_del(&entry->list);
309 		kfree(entry);
310 	}
311 	inode_unlock(d_inode(dir));
312 	nfs4_reset_creds(original_cred);
313 
314 	list_for_each_entry_safe(entry, tmp, &ctx.names, list) {
315 		dprintk("NFSD: %s. Left entry %s\n", __func__, entry->name);
316 		list_del(&entry->list);
317 		kfree(entry);
318 	}
319 	return status;
320 }
321 
322 static int
323 nfsd4_unlink_clid_dir(char *name, int namlen, struct nfsd_net *nn)
324 {
325 	struct dentry *dir, *dentry;
326 	int status;
327 
328 	dprintk("NFSD: nfsd4_unlink_clid_dir. name %.*s\n", namlen, name);
329 
330 	dir = nn->rec_file->f_path.dentry;
331 	inode_lock_nested(d_inode(dir), I_MUTEX_PARENT);
332 	dentry = lookup_one_len(name, dir, namlen);
333 	if (IS_ERR(dentry)) {
334 		status = PTR_ERR(dentry);
335 		goto out_unlock;
336 	}
337 	status = -ENOENT;
338 	if (d_really_is_negative(dentry))
339 		goto out;
340 	status = vfs_rmdir(d_inode(dir), dentry);
341 out:
342 	dput(dentry);
343 out_unlock:
344 	inode_unlock(d_inode(dir));
345 	return status;
346 }
347 
348 static void
349 nfsd4_remove_clid_dir(struct nfs4_client *clp)
350 {
351 	const struct cred *original_cred;
352 	struct nfs4_client_reclaim *crp;
353 	char dname[HEXDIR_LEN];
354 	int status;
355 	struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
356 
357 	if (!nn->rec_file || !test_bit(NFSD4_CLIENT_STABLE, &clp->cl_flags))
358 		return;
359 
360 	status = nfs4_make_rec_clidname(dname, &clp->cl_name);
361 	if (status)
362 		return legacy_recdir_name_error(clp, status);
363 
364 	status = mnt_want_write_file(nn->rec_file);
365 	if (status)
366 		goto out;
367 	clear_bit(NFSD4_CLIENT_STABLE, &clp->cl_flags);
368 
369 	status = nfs4_save_creds(&original_cred);
370 	if (status < 0)
371 		goto out_drop_write;
372 
373 	status = nfsd4_unlink_clid_dir(dname, HEXDIR_LEN-1, nn);
374 	nfs4_reset_creds(original_cred);
375 	if (status == 0) {
376 		vfs_fsync(nn->rec_file, 0);
377 		if (nn->in_grace) {
378 			/* remove reclaim record */
379 			crp = nfsd4_find_reclaim_client(dname, nn);
380 			if (crp)
381 				nfs4_remove_reclaim_record(crp, nn);
382 		}
383 	}
384 out_drop_write:
385 	mnt_drop_write_file(nn->rec_file);
386 out:
387 	if (status)
388 		printk("NFSD: Failed to remove expired client state directory"
389 				" %.*s\n", HEXDIR_LEN, dname);
390 }
391 
392 static int
393 purge_old(struct dentry *parent, struct dentry *child, struct nfsd_net *nn)
394 {
395 	int status;
396 
397 	if (nfs4_has_reclaimed_state(child->d_name.name, nn))
398 		return 0;
399 
400 	status = vfs_rmdir(d_inode(parent), child);
401 	if (status)
402 		printk("failed to remove client recovery directory %pd\n",
403 				child);
404 	/* Keep trying, success or failure: */
405 	return 0;
406 }
407 
408 static void
409 nfsd4_recdir_purge_old(struct nfsd_net *nn)
410 {
411 	int status;
412 
413 	nn->in_grace = false;
414 	if (!nn->rec_file)
415 		return;
416 	status = mnt_want_write_file(nn->rec_file);
417 	if (status)
418 		goto out;
419 	status = nfsd4_list_rec_dir(purge_old, nn);
420 	if (status == 0)
421 		vfs_fsync(nn->rec_file, 0);
422 	mnt_drop_write_file(nn->rec_file);
423 out:
424 	nfs4_release_reclaim(nn);
425 	if (status)
426 		printk("nfsd4: failed to purge old clients from recovery"
427 			" directory %pD\n", nn->rec_file);
428 }
429 
430 static int
431 load_recdir(struct dentry *parent, struct dentry *child, struct nfsd_net *nn)
432 {
433 	if (child->d_name.len != HEXDIR_LEN - 1) {
434 		printk("nfsd4: illegal name %pd in recovery directory\n",
435 				child);
436 		/* Keep trying; maybe the others are OK: */
437 		return 0;
438 	}
439 	nfs4_client_to_reclaim(child->d_name.name, nn);
440 	return 0;
441 }
442 
443 static int
444 nfsd4_recdir_load(struct net *net) {
445 	int status;
446 	struct nfsd_net *nn =  net_generic(net, nfsd_net_id);
447 
448 	if (!nn->rec_file)
449 		return 0;
450 
451 	status = nfsd4_list_rec_dir(load_recdir, nn);
452 	if (status)
453 		printk("nfsd4: failed loading clients from recovery"
454 			" directory %pD\n", nn->rec_file);
455 	return status;
456 }
457 
458 /*
459  * Hold reference to the recovery directory.
460  */
461 
462 static int
463 nfsd4_init_recdir(struct net *net)
464 {
465 	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
466 	const struct cred *original_cred;
467 	int status;
468 
469 	printk("NFSD: Using %s as the NFSv4 state recovery directory\n",
470 			user_recovery_dirname);
471 
472 	BUG_ON(nn->rec_file);
473 
474 	status = nfs4_save_creds(&original_cred);
475 	if (status < 0) {
476 		printk("NFSD: Unable to change credentials to find recovery"
477 		       " directory: error %d\n",
478 		       status);
479 		return status;
480 	}
481 
482 	nn->rec_file = filp_open(user_recovery_dirname, O_RDONLY | O_DIRECTORY, 0);
483 	if (IS_ERR(nn->rec_file)) {
484 		printk("NFSD: unable to find recovery directory %s\n",
485 				user_recovery_dirname);
486 		status = PTR_ERR(nn->rec_file);
487 		nn->rec_file = NULL;
488 	}
489 
490 	nfs4_reset_creds(original_cred);
491 	if (!status)
492 		nn->in_grace = true;
493 	return status;
494 }
495 
496 static void
497 nfsd4_shutdown_recdir(struct net *net)
498 {
499 	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
500 
501 	if (!nn->rec_file)
502 		return;
503 	fput(nn->rec_file);
504 	nn->rec_file = NULL;
505 }
506 
507 static int
508 nfs4_legacy_state_init(struct net *net)
509 {
510 	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
511 	int i;
512 
513 	nn->reclaim_str_hashtbl = kmalloc(sizeof(struct list_head) *
514 					  CLIENT_HASH_SIZE, GFP_KERNEL);
515 	if (!nn->reclaim_str_hashtbl)
516 		return -ENOMEM;
517 
518 	for (i = 0; i < CLIENT_HASH_SIZE; i++)
519 		INIT_LIST_HEAD(&nn->reclaim_str_hashtbl[i]);
520 	nn->reclaim_str_hashtbl_size = 0;
521 
522 	return 0;
523 }
524 
525 static void
526 nfs4_legacy_state_shutdown(struct net *net)
527 {
528 	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
529 
530 	kfree(nn->reclaim_str_hashtbl);
531 }
532 
533 static int
534 nfsd4_load_reboot_recovery_data(struct net *net)
535 {
536 	int status;
537 
538 	status = nfsd4_init_recdir(net);
539 	if (status)
540 		return status;
541 
542 	status = nfsd4_recdir_load(net);
543 	if (status)
544 		nfsd4_shutdown_recdir(net);
545 
546 	return status;
547 }
548 
549 static int
550 nfsd4_legacy_tracking_init(struct net *net)
551 {
552 	int status;
553 
554 	/* XXX: The legacy code won't work in a container */
555 	if (net != &init_net) {
556 		pr_warn("NFSD: attempt to initialize legacy client tracking in a container ignored.\n");
557 		return -EINVAL;
558 	}
559 
560 	status = nfs4_legacy_state_init(net);
561 	if (status)
562 		return status;
563 
564 	status = nfsd4_load_reboot_recovery_data(net);
565 	if (status)
566 		goto err;
567 	return 0;
568 
569 err:
570 	nfs4_legacy_state_shutdown(net);
571 	return status;
572 }
573 
574 static void
575 nfsd4_legacy_tracking_exit(struct net *net)
576 {
577 	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
578 
579 	nfs4_release_reclaim(nn);
580 	nfsd4_shutdown_recdir(net);
581 	nfs4_legacy_state_shutdown(net);
582 }
583 
584 /*
585  * Change the NFSv4 recovery directory to recdir.
586  */
587 int
588 nfs4_reset_recoverydir(char *recdir)
589 {
590 	int status;
591 	struct path path;
592 
593 	status = kern_path(recdir, LOOKUP_FOLLOW, &path);
594 	if (status)
595 		return status;
596 	status = -ENOTDIR;
597 	if (d_is_dir(path.dentry)) {
598 		strcpy(user_recovery_dirname, recdir);
599 		status = 0;
600 	}
601 	path_put(&path);
602 	return status;
603 }
604 
605 char *
606 nfs4_recoverydir(void)
607 {
608 	return user_recovery_dirname;
609 }
610 
611 static int
612 nfsd4_check_legacy_client(struct nfs4_client *clp)
613 {
614 	int status;
615 	char dname[HEXDIR_LEN];
616 	struct nfs4_client_reclaim *crp;
617 	struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
618 
619 	/* did we already find that this client is stable? */
620 	if (test_bit(NFSD4_CLIENT_STABLE, &clp->cl_flags))
621 		return 0;
622 
623 	status = nfs4_make_rec_clidname(dname, &clp->cl_name);
624 	if (status) {
625 		legacy_recdir_name_error(clp, status);
626 		return status;
627 	}
628 
629 	/* look for it in the reclaim hashtable otherwise */
630 	crp = nfsd4_find_reclaim_client(dname, nn);
631 	if (crp) {
632 		set_bit(NFSD4_CLIENT_STABLE, &clp->cl_flags);
633 		crp->cr_clp = clp;
634 		return 0;
635 	}
636 
637 	return -ENOENT;
638 }
639 
640 static const struct nfsd4_client_tracking_ops nfsd4_legacy_tracking_ops = {
641 	.init		= nfsd4_legacy_tracking_init,
642 	.exit		= nfsd4_legacy_tracking_exit,
643 	.create		= nfsd4_create_clid_dir,
644 	.remove		= nfsd4_remove_clid_dir,
645 	.check		= nfsd4_check_legacy_client,
646 	.grace_done	= nfsd4_recdir_purge_old,
647 };
648 
649 /* Globals */
650 #define NFSD_PIPE_DIR		"nfsd"
651 #define NFSD_CLD_PIPE		"cld"
652 
653 /* per-net-ns structure for holding cld upcall info */
654 struct cld_net {
655 	struct rpc_pipe		*cn_pipe;
656 	spinlock_t		 cn_lock;
657 	struct list_head	 cn_list;
658 	unsigned int		 cn_xid;
659 };
660 
661 struct cld_upcall {
662 	struct list_head	 cu_list;
663 	struct cld_net		*cu_net;
664 	struct task_struct	*cu_task;
665 	struct cld_msg		 cu_msg;
666 };
667 
668 static int
669 __cld_pipe_upcall(struct rpc_pipe *pipe, struct cld_msg *cmsg)
670 {
671 	int ret;
672 	struct rpc_pipe_msg msg;
673 
674 	memset(&msg, 0, sizeof(msg));
675 	msg.data = cmsg;
676 	msg.len = sizeof(*cmsg);
677 
678 	/*
679 	 * Set task state before we queue the upcall. That prevents
680 	 * wake_up_process in the downcall from racing with schedule.
681 	 */
682 	set_current_state(TASK_UNINTERRUPTIBLE);
683 	ret = rpc_queue_upcall(pipe, &msg);
684 	if (ret < 0) {
685 		set_current_state(TASK_RUNNING);
686 		goto out;
687 	}
688 
689 	schedule();
690 
691 	if (msg.errno < 0)
692 		ret = msg.errno;
693 out:
694 	return ret;
695 }
696 
697 static int
698 cld_pipe_upcall(struct rpc_pipe *pipe, struct cld_msg *cmsg)
699 {
700 	int ret;
701 
702 	/*
703 	 * -EAGAIN occurs when pipe is closed and reopened while there are
704 	 *  upcalls queued.
705 	 */
706 	do {
707 		ret = __cld_pipe_upcall(pipe, cmsg);
708 	} while (ret == -EAGAIN);
709 
710 	return ret;
711 }
712 
713 static ssize_t
714 cld_pipe_downcall(struct file *filp, const char __user *src, size_t mlen)
715 {
716 	struct cld_upcall *tmp, *cup;
717 	struct cld_msg __user *cmsg = (struct cld_msg __user *)src;
718 	uint32_t xid;
719 	struct nfsd_net *nn = net_generic(file_inode(filp)->i_sb->s_fs_info,
720 						nfsd_net_id);
721 	struct cld_net *cn = nn->cld_net;
722 
723 	if (mlen != sizeof(*cmsg)) {
724 		dprintk("%s: got %zu bytes, expected %zu\n", __func__, mlen,
725 			sizeof(*cmsg));
726 		return -EINVAL;
727 	}
728 
729 	/* copy just the xid so we can try to find that */
730 	if (copy_from_user(&xid, &cmsg->cm_xid, sizeof(xid)) != 0) {
731 		dprintk("%s: error when copying xid from userspace", __func__);
732 		return -EFAULT;
733 	}
734 
735 	/* walk the list and find corresponding xid */
736 	cup = NULL;
737 	spin_lock(&cn->cn_lock);
738 	list_for_each_entry(tmp, &cn->cn_list, cu_list) {
739 		if (get_unaligned(&tmp->cu_msg.cm_xid) == xid) {
740 			cup = tmp;
741 			list_del_init(&cup->cu_list);
742 			break;
743 		}
744 	}
745 	spin_unlock(&cn->cn_lock);
746 
747 	/* couldn't find upcall? */
748 	if (!cup) {
749 		dprintk("%s: couldn't find upcall -- xid=%u\n", __func__, xid);
750 		return -EINVAL;
751 	}
752 
753 	if (copy_from_user(&cup->cu_msg, src, mlen) != 0)
754 		return -EFAULT;
755 
756 	wake_up_process(cup->cu_task);
757 	return mlen;
758 }
759 
760 static void
761 cld_pipe_destroy_msg(struct rpc_pipe_msg *msg)
762 {
763 	struct cld_msg *cmsg = msg->data;
764 	struct cld_upcall *cup = container_of(cmsg, struct cld_upcall,
765 						 cu_msg);
766 
767 	/* errno >= 0 means we got a downcall */
768 	if (msg->errno >= 0)
769 		return;
770 
771 	wake_up_process(cup->cu_task);
772 }
773 
774 static const struct rpc_pipe_ops cld_upcall_ops = {
775 	.upcall		= rpc_pipe_generic_upcall,
776 	.downcall	= cld_pipe_downcall,
777 	.destroy_msg	= cld_pipe_destroy_msg,
778 };
779 
780 static struct dentry *
781 nfsd4_cld_register_sb(struct super_block *sb, struct rpc_pipe *pipe)
782 {
783 	struct dentry *dir, *dentry;
784 
785 	dir = rpc_d_lookup_sb(sb, NFSD_PIPE_DIR);
786 	if (dir == NULL)
787 		return ERR_PTR(-ENOENT);
788 	dentry = rpc_mkpipe_dentry(dir, NFSD_CLD_PIPE, NULL, pipe);
789 	dput(dir);
790 	return dentry;
791 }
792 
793 static void
794 nfsd4_cld_unregister_sb(struct rpc_pipe *pipe)
795 {
796 	if (pipe->dentry)
797 		rpc_unlink(pipe->dentry);
798 }
799 
800 static struct dentry *
801 nfsd4_cld_register_net(struct net *net, struct rpc_pipe *pipe)
802 {
803 	struct super_block *sb;
804 	struct dentry *dentry;
805 
806 	sb = rpc_get_sb_net(net);
807 	if (!sb)
808 		return NULL;
809 	dentry = nfsd4_cld_register_sb(sb, pipe);
810 	rpc_put_sb_net(net);
811 	return dentry;
812 }
813 
814 static void
815 nfsd4_cld_unregister_net(struct net *net, struct rpc_pipe *pipe)
816 {
817 	struct super_block *sb;
818 
819 	sb = rpc_get_sb_net(net);
820 	if (sb) {
821 		nfsd4_cld_unregister_sb(pipe);
822 		rpc_put_sb_net(net);
823 	}
824 }
825 
826 /* Initialize rpc_pipefs pipe for communication with client tracking daemon */
827 static int
828 nfsd4_init_cld_pipe(struct net *net)
829 {
830 	int ret;
831 	struct dentry *dentry;
832 	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
833 	struct cld_net *cn;
834 
835 	if (nn->cld_net)
836 		return 0;
837 
838 	cn = kzalloc(sizeof(*cn), GFP_KERNEL);
839 	if (!cn) {
840 		ret = -ENOMEM;
841 		goto err;
842 	}
843 
844 	cn->cn_pipe = rpc_mkpipe_data(&cld_upcall_ops, RPC_PIPE_WAIT_FOR_OPEN);
845 	if (IS_ERR(cn->cn_pipe)) {
846 		ret = PTR_ERR(cn->cn_pipe);
847 		goto err;
848 	}
849 	spin_lock_init(&cn->cn_lock);
850 	INIT_LIST_HEAD(&cn->cn_list);
851 
852 	dentry = nfsd4_cld_register_net(net, cn->cn_pipe);
853 	if (IS_ERR(dentry)) {
854 		ret = PTR_ERR(dentry);
855 		goto err_destroy_data;
856 	}
857 
858 	cn->cn_pipe->dentry = dentry;
859 	nn->cld_net = cn;
860 	return 0;
861 
862 err_destroy_data:
863 	rpc_destroy_pipe_data(cn->cn_pipe);
864 err:
865 	kfree(cn);
866 	printk(KERN_ERR "NFSD: unable to create nfsdcld upcall pipe (%d)\n",
867 			ret);
868 	return ret;
869 }
870 
871 static void
872 nfsd4_remove_cld_pipe(struct net *net)
873 {
874 	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
875 	struct cld_net *cn = nn->cld_net;
876 
877 	nfsd4_cld_unregister_net(net, cn->cn_pipe);
878 	rpc_destroy_pipe_data(cn->cn_pipe);
879 	kfree(nn->cld_net);
880 	nn->cld_net = NULL;
881 }
882 
883 static struct cld_upcall *
884 alloc_cld_upcall(struct cld_net *cn)
885 {
886 	struct cld_upcall *new, *tmp;
887 
888 	new = kzalloc(sizeof(*new), GFP_KERNEL);
889 	if (!new)
890 		return new;
891 
892 	/* FIXME: hard cap on number in flight? */
893 restart_search:
894 	spin_lock(&cn->cn_lock);
895 	list_for_each_entry(tmp, &cn->cn_list, cu_list) {
896 		if (tmp->cu_msg.cm_xid == cn->cn_xid) {
897 			cn->cn_xid++;
898 			spin_unlock(&cn->cn_lock);
899 			goto restart_search;
900 		}
901 	}
902 	new->cu_task = current;
903 	new->cu_msg.cm_vers = CLD_UPCALL_VERSION;
904 	put_unaligned(cn->cn_xid++, &new->cu_msg.cm_xid);
905 	new->cu_net = cn;
906 	list_add(&new->cu_list, &cn->cn_list);
907 	spin_unlock(&cn->cn_lock);
908 
909 	dprintk("%s: allocated xid %u\n", __func__, new->cu_msg.cm_xid);
910 
911 	return new;
912 }
913 
914 static void
915 free_cld_upcall(struct cld_upcall *victim)
916 {
917 	struct cld_net *cn = victim->cu_net;
918 
919 	spin_lock(&cn->cn_lock);
920 	list_del(&victim->cu_list);
921 	spin_unlock(&cn->cn_lock);
922 	kfree(victim);
923 }
924 
925 /* Ask daemon to create a new record */
926 static void
927 nfsd4_cld_create(struct nfs4_client *clp)
928 {
929 	int ret;
930 	struct cld_upcall *cup;
931 	struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
932 	struct cld_net *cn = nn->cld_net;
933 
934 	/* Don't upcall if it's already stored */
935 	if (test_bit(NFSD4_CLIENT_STABLE, &clp->cl_flags))
936 		return;
937 
938 	cup = alloc_cld_upcall(cn);
939 	if (!cup) {
940 		ret = -ENOMEM;
941 		goto out_err;
942 	}
943 
944 	cup->cu_msg.cm_cmd = Cld_Create;
945 	cup->cu_msg.cm_u.cm_name.cn_len = clp->cl_name.len;
946 	memcpy(cup->cu_msg.cm_u.cm_name.cn_id, clp->cl_name.data,
947 			clp->cl_name.len);
948 
949 	ret = cld_pipe_upcall(cn->cn_pipe, &cup->cu_msg);
950 	if (!ret) {
951 		ret = cup->cu_msg.cm_status;
952 		set_bit(NFSD4_CLIENT_STABLE, &clp->cl_flags);
953 	}
954 
955 	free_cld_upcall(cup);
956 out_err:
957 	if (ret)
958 		printk(KERN_ERR "NFSD: Unable to create client "
959 				"record on stable storage: %d\n", ret);
960 }
961 
962 /* Ask daemon to create a new record */
963 static void
964 nfsd4_cld_remove(struct nfs4_client *clp)
965 {
966 	int ret;
967 	struct cld_upcall *cup;
968 	struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
969 	struct cld_net *cn = nn->cld_net;
970 
971 	/* Don't upcall if it's already removed */
972 	if (!test_bit(NFSD4_CLIENT_STABLE, &clp->cl_flags))
973 		return;
974 
975 	cup = alloc_cld_upcall(cn);
976 	if (!cup) {
977 		ret = -ENOMEM;
978 		goto out_err;
979 	}
980 
981 	cup->cu_msg.cm_cmd = Cld_Remove;
982 	cup->cu_msg.cm_u.cm_name.cn_len = clp->cl_name.len;
983 	memcpy(cup->cu_msg.cm_u.cm_name.cn_id, clp->cl_name.data,
984 			clp->cl_name.len);
985 
986 	ret = cld_pipe_upcall(cn->cn_pipe, &cup->cu_msg);
987 	if (!ret) {
988 		ret = cup->cu_msg.cm_status;
989 		clear_bit(NFSD4_CLIENT_STABLE, &clp->cl_flags);
990 	}
991 
992 	free_cld_upcall(cup);
993 out_err:
994 	if (ret)
995 		printk(KERN_ERR "NFSD: Unable to remove client "
996 				"record from stable storage: %d\n", ret);
997 }
998 
999 /* Check for presence of a record, and update its timestamp */
1000 static int
1001 nfsd4_cld_check(struct nfs4_client *clp)
1002 {
1003 	int ret;
1004 	struct cld_upcall *cup;
1005 	struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
1006 	struct cld_net *cn = nn->cld_net;
1007 
1008 	/* Don't upcall if one was already stored during this grace pd */
1009 	if (test_bit(NFSD4_CLIENT_STABLE, &clp->cl_flags))
1010 		return 0;
1011 
1012 	cup = alloc_cld_upcall(cn);
1013 	if (!cup) {
1014 		printk(KERN_ERR "NFSD: Unable to check client record on "
1015 				"stable storage: %d\n", -ENOMEM);
1016 		return -ENOMEM;
1017 	}
1018 
1019 	cup->cu_msg.cm_cmd = Cld_Check;
1020 	cup->cu_msg.cm_u.cm_name.cn_len = clp->cl_name.len;
1021 	memcpy(cup->cu_msg.cm_u.cm_name.cn_id, clp->cl_name.data,
1022 			clp->cl_name.len);
1023 
1024 	ret = cld_pipe_upcall(cn->cn_pipe, &cup->cu_msg);
1025 	if (!ret) {
1026 		ret = cup->cu_msg.cm_status;
1027 		set_bit(NFSD4_CLIENT_STABLE, &clp->cl_flags);
1028 	}
1029 
1030 	free_cld_upcall(cup);
1031 	return ret;
1032 }
1033 
1034 static void
1035 nfsd4_cld_grace_done(struct nfsd_net *nn)
1036 {
1037 	int ret;
1038 	struct cld_upcall *cup;
1039 	struct cld_net *cn = nn->cld_net;
1040 
1041 	cup = alloc_cld_upcall(cn);
1042 	if (!cup) {
1043 		ret = -ENOMEM;
1044 		goto out_err;
1045 	}
1046 
1047 	cup->cu_msg.cm_cmd = Cld_GraceDone;
1048 	cup->cu_msg.cm_u.cm_gracetime = (int64_t)nn->boot_time;
1049 	ret = cld_pipe_upcall(cn->cn_pipe, &cup->cu_msg);
1050 	if (!ret)
1051 		ret = cup->cu_msg.cm_status;
1052 
1053 	free_cld_upcall(cup);
1054 out_err:
1055 	if (ret)
1056 		printk(KERN_ERR "NFSD: Unable to end grace period: %d\n", ret);
1057 }
1058 
1059 static const struct nfsd4_client_tracking_ops nfsd4_cld_tracking_ops = {
1060 	.init		= nfsd4_init_cld_pipe,
1061 	.exit		= nfsd4_remove_cld_pipe,
1062 	.create		= nfsd4_cld_create,
1063 	.remove		= nfsd4_cld_remove,
1064 	.check		= nfsd4_cld_check,
1065 	.grace_done	= nfsd4_cld_grace_done,
1066 };
1067 
1068 /* upcall via usermodehelper */
1069 static char cltrack_prog[PATH_MAX] = "/sbin/nfsdcltrack";
1070 module_param_string(cltrack_prog, cltrack_prog, sizeof(cltrack_prog),
1071 			S_IRUGO|S_IWUSR);
1072 MODULE_PARM_DESC(cltrack_prog, "Path to the nfsdcltrack upcall program");
1073 
1074 static bool cltrack_legacy_disable;
1075 module_param(cltrack_legacy_disable, bool, S_IRUGO|S_IWUSR);
1076 MODULE_PARM_DESC(cltrack_legacy_disable,
1077 		"Disable legacy recoverydir conversion. Default: false");
1078 
1079 #define LEGACY_TOPDIR_ENV_PREFIX "NFSDCLTRACK_LEGACY_TOPDIR="
1080 #define LEGACY_RECDIR_ENV_PREFIX "NFSDCLTRACK_LEGACY_RECDIR="
1081 #define HAS_SESSION_ENV_PREFIX "NFSDCLTRACK_CLIENT_HAS_SESSION="
1082 #define GRACE_START_ENV_PREFIX "NFSDCLTRACK_GRACE_START="
1083 
1084 static char *
1085 nfsd4_cltrack_legacy_topdir(void)
1086 {
1087 	int copied;
1088 	size_t len;
1089 	char *result;
1090 
1091 	if (cltrack_legacy_disable)
1092 		return NULL;
1093 
1094 	len = strlen(LEGACY_TOPDIR_ENV_PREFIX) +
1095 		strlen(nfs4_recoverydir()) + 1;
1096 
1097 	result = kmalloc(len, GFP_KERNEL);
1098 	if (!result)
1099 		return result;
1100 
1101 	copied = snprintf(result, len, LEGACY_TOPDIR_ENV_PREFIX "%s",
1102 				nfs4_recoverydir());
1103 	if (copied >= len) {
1104 		/* just return nothing if output was truncated */
1105 		kfree(result);
1106 		return NULL;
1107 	}
1108 
1109 	return result;
1110 }
1111 
1112 static char *
1113 nfsd4_cltrack_legacy_recdir(const struct xdr_netobj *name)
1114 {
1115 	int copied;
1116 	size_t len;
1117 	char *result;
1118 
1119 	if (cltrack_legacy_disable)
1120 		return NULL;
1121 
1122 	/* +1 is for '/' between "topdir" and "recdir" */
1123 	len = strlen(LEGACY_RECDIR_ENV_PREFIX) +
1124 		strlen(nfs4_recoverydir()) + 1 + HEXDIR_LEN;
1125 
1126 	result = kmalloc(len, GFP_KERNEL);
1127 	if (!result)
1128 		return result;
1129 
1130 	copied = snprintf(result, len, LEGACY_RECDIR_ENV_PREFIX "%s/",
1131 				nfs4_recoverydir());
1132 	if (copied > (len - HEXDIR_LEN)) {
1133 		/* just return nothing if output will be truncated */
1134 		kfree(result);
1135 		return NULL;
1136 	}
1137 
1138 	copied = nfs4_make_rec_clidname(result + copied, name);
1139 	if (copied) {
1140 		kfree(result);
1141 		return NULL;
1142 	}
1143 
1144 	return result;
1145 }
1146 
1147 static char *
1148 nfsd4_cltrack_client_has_session(struct nfs4_client *clp)
1149 {
1150 	int copied;
1151 	size_t len;
1152 	char *result;
1153 
1154 	/* prefix + Y/N character + terminating NULL */
1155 	len = strlen(HAS_SESSION_ENV_PREFIX) + 1 + 1;
1156 
1157 	result = kmalloc(len, GFP_KERNEL);
1158 	if (!result)
1159 		return result;
1160 
1161 	copied = snprintf(result, len, HAS_SESSION_ENV_PREFIX "%c",
1162 				clp->cl_minorversion ? 'Y' : 'N');
1163 	if (copied >= len) {
1164 		/* just return nothing if output was truncated */
1165 		kfree(result);
1166 		return NULL;
1167 	}
1168 
1169 	return result;
1170 }
1171 
1172 static char *
1173 nfsd4_cltrack_grace_start(time_t grace_start)
1174 {
1175 	int copied;
1176 	size_t len;
1177 	char *result;
1178 
1179 	/* prefix + max width of int64_t string + terminating NULL */
1180 	len = strlen(GRACE_START_ENV_PREFIX) + 22 + 1;
1181 
1182 	result = kmalloc(len, GFP_KERNEL);
1183 	if (!result)
1184 		return result;
1185 
1186 	copied = snprintf(result, len, GRACE_START_ENV_PREFIX "%ld",
1187 				grace_start);
1188 	if (copied >= len) {
1189 		/* just return nothing if output was truncated */
1190 		kfree(result);
1191 		return NULL;
1192 	}
1193 
1194 	return result;
1195 }
1196 
1197 static int
1198 nfsd4_umh_cltrack_upcall(char *cmd, char *arg, char *env0, char *env1)
1199 {
1200 	char *envp[3];
1201 	char *argv[4];
1202 	int ret;
1203 
1204 	if (unlikely(!cltrack_prog[0])) {
1205 		dprintk("%s: cltrack_prog is disabled\n", __func__);
1206 		return -EACCES;
1207 	}
1208 
1209 	dprintk("%s: cmd: %s\n", __func__, cmd);
1210 	dprintk("%s: arg: %s\n", __func__, arg ? arg : "(null)");
1211 	dprintk("%s: env0: %s\n", __func__, env0 ? env0 : "(null)");
1212 	dprintk("%s: env1: %s\n", __func__, env1 ? env1 : "(null)");
1213 
1214 	envp[0] = env0;
1215 	envp[1] = env1;
1216 	envp[2] = NULL;
1217 
1218 	argv[0] = (char *)cltrack_prog;
1219 	argv[1] = cmd;
1220 	argv[2] = arg;
1221 	argv[3] = NULL;
1222 
1223 	ret = call_usermodehelper(argv[0], argv, envp, UMH_WAIT_PROC);
1224 	/*
1225 	 * Disable the upcall mechanism if we're getting an ENOENT or EACCES
1226 	 * error. The admin can re-enable it on the fly by using sysfs
1227 	 * once the problem has been fixed.
1228 	 */
1229 	if (ret == -ENOENT || ret == -EACCES) {
1230 		dprintk("NFSD: %s was not found or isn't executable (%d). "
1231 			"Setting cltrack_prog to blank string!",
1232 			cltrack_prog, ret);
1233 		cltrack_prog[0] = '\0';
1234 	}
1235 	dprintk("%s: %s return value: %d\n", __func__, cltrack_prog, ret);
1236 
1237 	return ret;
1238 }
1239 
1240 static char *
1241 bin_to_hex_dup(const unsigned char *src, int srclen)
1242 {
1243 	int i;
1244 	char *buf, *hex;
1245 
1246 	/* +1 for terminating NULL */
1247 	buf = kmalloc((srclen * 2) + 1, GFP_KERNEL);
1248 	if (!buf)
1249 		return buf;
1250 
1251 	hex = buf;
1252 	for (i = 0; i < srclen; i++) {
1253 		sprintf(hex, "%2.2x", *src++);
1254 		hex += 2;
1255 	}
1256 	return buf;
1257 }
1258 
1259 static int
1260 nfsd4_umh_cltrack_init(struct net *net)
1261 {
1262 	int ret;
1263 	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
1264 	char *grace_start = nfsd4_cltrack_grace_start(nn->boot_time);
1265 
1266 	/* XXX: The usermode helper s not working in container yet. */
1267 	if (net != &init_net) {
1268 		pr_warn("NFSD: attempt to initialize umh client tracking in a container ignored.\n");
1269 		kfree(grace_start);
1270 		return -EINVAL;
1271 	}
1272 
1273 	ret = nfsd4_umh_cltrack_upcall("init", NULL, grace_start, NULL);
1274 	kfree(grace_start);
1275 	return ret;
1276 }
1277 
1278 static void
1279 nfsd4_cltrack_upcall_lock(struct nfs4_client *clp)
1280 {
1281 	wait_on_bit_lock(&clp->cl_flags, NFSD4_CLIENT_UPCALL_LOCK,
1282 			 TASK_UNINTERRUPTIBLE);
1283 }
1284 
1285 static void
1286 nfsd4_cltrack_upcall_unlock(struct nfs4_client *clp)
1287 {
1288 	smp_mb__before_atomic();
1289 	clear_bit(NFSD4_CLIENT_UPCALL_LOCK, &clp->cl_flags);
1290 	smp_mb__after_atomic();
1291 	wake_up_bit(&clp->cl_flags, NFSD4_CLIENT_UPCALL_LOCK);
1292 }
1293 
1294 static void
1295 nfsd4_umh_cltrack_create(struct nfs4_client *clp)
1296 {
1297 	char *hexid, *has_session, *grace_start;
1298 	struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
1299 
1300 	/*
1301 	 * With v4.0 clients, there's little difference in outcome between a
1302 	 * create and check operation, and we can end up calling into this
1303 	 * function multiple times per client (once for each openowner). So,
1304 	 * for v4.0 clients skip upcalling once the client has been recorded
1305 	 * on stable storage.
1306 	 *
1307 	 * For v4.1+ clients, the outcome of the two operations is different,
1308 	 * so we must ensure that we upcall for the create operation. v4.1+
1309 	 * clients call this on RECLAIM_COMPLETE though, so we should only end
1310 	 * up doing a single create upcall per client.
1311 	 */
1312 	if (clp->cl_minorversion == 0 &&
1313 	    test_bit(NFSD4_CLIENT_STABLE, &clp->cl_flags))
1314 		return;
1315 
1316 	hexid = bin_to_hex_dup(clp->cl_name.data, clp->cl_name.len);
1317 	if (!hexid) {
1318 		dprintk("%s: can't allocate memory for upcall!\n", __func__);
1319 		return;
1320 	}
1321 
1322 	has_session = nfsd4_cltrack_client_has_session(clp);
1323 	grace_start = nfsd4_cltrack_grace_start(nn->boot_time);
1324 
1325 	nfsd4_cltrack_upcall_lock(clp);
1326 	if (!nfsd4_umh_cltrack_upcall("create", hexid, has_session, grace_start))
1327 		set_bit(NFSD4_CLIENT_STABLE, &clp->cl_flags);
1328 	nfsd4_cltrack_upcall_unlock(clp);
1329 
1330 	kfree(has_session);
1331 	kfree(grace_start);
1332 	kfree(hexid);
1333 }
1334 
1335 static void
1336 nfsd4_umh_cltrack_remove(struct nfs4_client *clp)
1337 {
1338 	char *hexid;
1339 
1340 	if (!test_bit(NFSD4_CLIENT_STABLE, &clp->cl_flags))
1341 		return;
1342 
1343 	hexid = bin_to_hex_dup(clp->cl_name.data, clp->cl_name.len);
1344 	if (!hexid) {
1345 		dprintk("%s: can't allocate memory for upcall!\n", __func__);
1346 		return;
1347 	}
1348 
1349 	nfsd4_cltrack_upcall_lock(clp);
1350 	if (test_bit(NFSD4_CLIENT_STABLE, &clp->cl_flags) &&
1351 	    nfsd4_umh_cltrack_upcall("remove", hexid, NULL, NULL) == 0)
1352 		clear_bit(NFSD4_CLIENT_STABLE, &clp->cl_flags);
1353 	nfsd4_cltrack_upcall_unlock(clp);
1354 
1355 	kfree(hexid);
1356 }
1357 
1358 static int
1359 nfsd4_umh_cltrack_check(struct nfs4_client *clp)
1360 {
1361 	int ret;
1362 	char *hexid, *has_session, *legacy;
1363 
1364 	if (test_bit(NFSD4_CLIENT_STABLE, &clp->cl_flags))
1365 		return 0;
1366 
1367 	hexid = bin_to_hex_dup(clp->cl_name.data, clp->cl_name.len);
1368 	if (!hexid) {
1369 		dprintk("%s: can't allocate memory for upcall!\n", __func__);
1370 		return -ENOMEM;
1371 	}
1372 
1373 	has_session = nfsd4_cltrack_client_has_session(clp);
1374 	legacy = nfsd4_cltrack_legacy_recdir(&clp->cl_name);
1375 
1376 	nfsd4_cltrack_upcall_lock(clp);
1377 	if (test_bit(NFSD4_CLIENT_STABLE, &clp->cl_flags)) {
1378 		ret = 0;
1379 	} else {
1380 		ret = nfsd4_umh_cltrack_upcall("check", hexid, has_session, legacy);
1381 		if (ret == 0)
1382 			set_bit(NFSD4_CLIENT_STABLE, &clp->cl_flags);
1383 	}
1384 	nfsd4_cltrack_upcall_unlock(clp);
1385 	kfree(has_session);
1386 	kfree(legacy);
1387 	kfree(hexid);
1388 
1389 	return ret;
1390 }
1391 
1392 static void
1393 nfsd4_umh_cltrack_grace_done(struct nfsd_net *nn)
1394 {
1395 	char *legacy;
1396 	char timestr[22]; /* FIXME: better way to determine max size? */
1397 
1398 	sprintf(timestr, "%ld", nn->boot_time);
1399 	legacy = nfsd4_cltrack_legacy_topdir();
1400 	nfsd4_umh_cltrack_upcall("gracedone", timestr, legacy, NULL);
1401 	kfree(legacy);
1402 }
1403 
1404 static const struct nfsd4_client_tracking_ops nfsd4_umh_tracking_ops = {
1405 	.init		= nfsd4_umh_cltrack_init,
1406 	.exit		= NULL,
1407 	.create		= nfsd4_umh_cltrack_create,
1408 	.remove		= nfsd4_umh_cltrack_remove,
1409 	.check		= nfsd4_umh_cltrack_check,
1410 	.grace_done	= nfsd4_umh_cltrack_grace_done,
1411 };
1412 
1413 int
1414 nfsd4_client_tracking_init(struct net *net)
1415 {
1416 	int status;
1417 	struct path path;
1418 	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
1419 
1420 	/* just run the init if it the method is already decided */
1421 	if (nn->client_tracking_ops)
1422 		goto do_init;
1423 
1424 	/*
1425 	 * First, try a UMH upcall. It should succeed or fail quickly, so
1426 	 * there's little harm in trying that first.
1427 	 */
1428 	nn->client_tracking_ops = &nfsd4_umh_tracking_ops;
1429 	status = nn->client_tracking_ops->init(net);
1430 	if (!status)
1431 		return status;
1432 
1433 	/*
1434 	 * See if the recoverydir exists and is a directory. If it is,
1435 	 * then use the legacy ops.
1436 	 */
1437 	nn->client_tracking_ops = &nfsd4_legacy_tracking_ops;
1438 	status = kern_path(nfs4_recoverydir(), LOOKUP_FOLLOW, &path);
1439 	if (!status) {
1440 		status = d_is_dir(path.dentry);
1441 		path_put(&path);
1442 		if (status)
1443 			goto do_init;
1444 	}
1445 
1446 	/* Finally, try to use nfsdcld */
1447 	nn->client_tracking_ops = &nfsd4_cld_tracking_ops;
1448 	printk(KERN_WARNING "NFSD: the nfsdcld client tracking upcall will be "
1449 			"removed in 3.10. Please transition to using "
1450 			"nfsdcltrack.\n");
1451 do_init:
1452 	status = nn->client_tracking_ops->init(net);
1453 	if (status) {
1454 		printk(KERN_WARNING "NFSD: Unable to initialize client "
1455 				    "recovery tracking! (%d)\n", status);
1456 		nn->client_tracking_ops = NULL;
1457 	}
1458 	return status;
1459 }
1460 
1461 void
1462 nfsd4_client_tracking_exit(struct net *net)
1463 {
1464 	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
1465 
1466 	if (nn->client_tracking_ops) {
1467 		if (nn->client_tracking_ops->exit)
1468 			nn->client_tracking_ops->exit(net);
1469 		nn->client_tracking_ops = NULL;
1470 	}
1471 }
1472 
1473 void
1474 nfsd4_client_record_create(struct nfs4_client *clp)
1475 {
1476 	struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
1477 
1478 	if (nn->client_tracking_ops)
1479 		nn->client_tracking_ops->create(clp);
1480 }
1481 
1482 void
1483 nfsd4_client_record_remove(struct nfs4_client *clp)
1484 {
1485 	struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
1486 
1487 	if (nn->client_tracking_ops)
1488 		nn->client_tracking_ops->remove(clp);
1489 }
1490 
1491 int
1492 nfsd4_client_record_check(struct nfs4_client *clp)
1493 {
1494 	struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
1495 
1496 	if (nn->client_tracking_ops)
1497 		return nn->client_tracking_ops->check(clp);
1498 
1499 	return -EOPNOTSUPP;
1500 }
1501 
1502 void
1503 nfsd4_record_grace_done(struct nfsd_net *nn)
1504 {
1505 	if (nn->client_tracking_ops)
1506 		nn->client_tracking_ops->grace_done(nn);
1507 }
1508 
1509 static int
1510 rpc_pipefs_event(struct notifier_block *nb, unsigned long event, void *ptr)
1511 {
1512 	struct super_block *sb = ptr;
1513 	struct net *net = sb->s_fs_info;
1514 	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
1515 	struct cld_net *cn = nn->cld_net;
1516 	struct dentry *dentry;
1517 	int ret = 0;
1518 
1519 	if (!try_module_get(THIS_MODULE))
1520 		return 0;
1521 
1522 	if (!cn) {
1523 		module_put(THIS_MODULE);
1524 		return 0;
1525 	}
1526 
1527 	switch (event) {
1528 	case RPC_PIPEFS_MOUNT:
1529 		dentry = nfsd4_cld_register_sb(sb, cn->cn_pipe);
1530 		if (IS_ERR(dentry)) {
1531 			ret = PTR_ERR(dentry);
1532 			break;
1533 		}
1534 		cn->cn_pipe->dentry = dentry;
1535 		break;
1536 	case RPC_PIPEFS_UMOUNT:
1537 		if (cn->cn_pipe->dentry)
1538 			nfsd4_cld_unregister_sb(cn->cn_pipe);
1539 		break;
1540 	default:
1541 		ret = -ENOTSUPP;
1542 		break;
1543 	}
1544 	module_put(THIS_MODULE);
1545 	return ret;
1546 }
1547 
1548 static struct notifier_block nfsd4_cld_block = {
1549 	.notifier_call = rpc_pipefs_event,
1550 };
1551 
1552 int
1553 register_cld_notifier(void)
1554 {
1555 	return rpc_pipefs_notifier_register(&nfsd4_cld_block);
1556 }
1557 
1558 void
1559 unregister_cld_notifier(void)
1560 {
1561 	rpc_pipefs_notifier_unregister(&nfsd4_cld_block);
1562 }
1563