xref: /openbmc/linux/fs/nfs/nfs4client.c (revision 4cff79e9)
1 /*
2  * Copyright (C) 2006 Red Hat, Inc. All Rights Reserved.
3  * Written by David Howells (dhowells@redhat.com)
4  */
5 #include <linux/module.h>
6 #include <linux/nfs_fs.h>
7 #include <linux/nfs_mount.h>
8 #include <linux/sunrpc/addr.h>
9 #include <linux/sunrpc/auth.h>
10 #include <linux/sunrpc/xprt.h>
11 #include <linux/sunrpc/bc_xprt.h>
12 #include <linux/sunrpc/rpc_pipe_fs.h>
13 #include "internal.h"
14 #include "callback.h"
15 #include "delegation.h"
16 #include "nfs4session.h"
17 #include "nfs4idmap.h"
18 #include "pnfs.h"
19 #include "netns.h"
20 
21 #define NFSDBG_FACILITY		NFSDBG_CLIENT
22 
23 /*
24  * Get a unique NFSv4.0 callback identifier which will be used
25  * by the V4.0 callback service to lookup the nfs_client struct
26  */
27 static int nfs_get_cb_ident_idr(struct nfs_client *clp, int minorversion)
28 {
29 	int ret = 0;
30 	struct nfs_net *nn = net_generic(clp->cl_net, nfs_net_id);
31 
32 	if (clp->rpc_ops->version != 4 || minorversion != 0)
33 		return ret;
34 	idr_preload(GFP_KERNEL);
35 	spin_lock(&nn->nfs_client_lock);
36 	ret = idr_alloc(&nn->cb_ident_idr, clp, 1, 0, GFP_NOWAIT);
37 	if (ret >= 0)
38 		clp->cl_cb_ident = ret;
39 	spin_unlock(&nn->nfs_client_lock);
40 	idr_preload_end();
41 	return ret < 0 ? ret : 0;
42 }
43 
44 #ifdef CONFIG_NFS_V4_1
45 /**
46  * Per auth flavor data server rpc clients
47  */
48 struct nfs4_ds_server {
49 	struct list_head	list;   /* ds_clp->cl_ds_clients */
50 	struct rpc_clnt		*rpc_clnt;
51 };
52 
53 /**
54  * Common lookup case for DS I/O
55  */
56 static struct nfs4_ds_server *
57 nfs4_find_ds_client(struct nfs_client *ds_clp, rpc_authflavor_t flavor)
58 {
59 	struct nfs4_ds_server *dss;
60 
61 	rcu_read_lock();
62 	list_for_each_entry_rcu(dss, &ds_clp->cl_ds_clients, list) {
63 		if (dss->rpc_clnt->cl_auth->au_flavor != flavor)
64 			continue;
65 		goto out;
66 	}
67 	dss = NULL;
68 out:
69 	rcu_read_unlock();
70 	return dss;
71 }
72 
73 static struct nfs4_ds_server *
74 nfs4_add_ds_client(struct nfs_client *ds_clp, rpc_authflavor_t flavor,
75 			   struct nfs4_ds_server *new)
76 {
77 	struct nfs4_ds_server *dss;
78 
79 	spin_lock(&ds_clp->cl_lock);
80 	list_for_each_entry(dss, &ds_clp->cl_ds_clients, list) {
81 		if (dss->rpc_clnt->cl_auth->au_flavor != flavor)
82 			continue;
83 		goto out;
84 	}
85 	if (new)
86 		list_add_rcu(&new->list, &ds_clp->cl_ds_clients);
87 	dss = new;
88 out:
89 	spin_unlock(&ds_clp->cl_lock); /* need some lock to protect list */
90 	return dss;
91 }
92 
93 static struct nfs4_ds_server *
94 nfs4_alloc_ds_server(struct nfs_client *ds_clp, rpc_authflavor_t flavor)
95 {
96 	struct nfs4_ds_server *dss;
97 
98 	dss = kmalloc(sizeof(*dss), GFP_NOFS);
99 	if (dss == NULL)
100 		return ERR_PTR(-ENOMEM);
101 
102 	dss->rpc_clnt = rpc_clone_client_set_auth(ds_clp->cl_rpcclient, flavor);
103 	if (IS_ERR(dss->rpc_clnt)) {
104 		int err = PTR_ERR(dss->rpc_clnt);
105 		kfree (dss);
106 		return ERR_PTR(err);
107 	}
108 	INIT_LIST_HEAD(&dss->list);
109 
110 	return dss;
111 }
112 
113 static void
114 nfs4_free_ds_server(struct nfs4_ds_server *dss)
115 {
116 	rpc_release_client(dss->rpc_clnt);
117 	kfree(dss);
118 }
119 
120 /**
121 * Find or create a DS rpc client with th MDS server rpc client auth flavor
122 * in the nfs_client cl_ds_clients list.
123 */
124 struct rpc_clnt *
125 nfs4_find_or_create_ds_client(struct nfs_client *ds_clp, struct inode *inode)
126 {
127 	struct nfs4_ds_server *dss, *new;
128 	rpc_authflavor_t flavor = NFS_SERVER(inode)->client->cl_auth->au_flavor;
129 
130 	dss = nfs4_find_ds_client(ds_clp, flavor);
131 	if (dss != NULL)
132 		goto out;
133 	new = nfs4_alloc_ds_server(ds_clp, flavor);
134 	if (IS_ERR(new))
135 		return ERR_CAST(new);
136 	dss = nfs4_add_ds_client(ds_clp, flavor, new);
137 	if (dss != new)
138 		nfs4_free_ds_server(new);
139 out:
140 	return dss->rpc_clnt;
141 }
142 EXPORT_SYMBOL_GPL(nfs4_find_or_create_ds_client);
143 
144 static void
145 nfs4_shutdown_ds_clients(struct nfs_client *clp)
146 {
147 	struct nfs4_ds_server *dss;
148 	LIST_HEAD(shutdown_list);
149 
150 	while (!list_empty(&clp->cl_ds_clients)) {
151 		dss = list_entry(clp->cl_ds_clients.next,
152 					struct nfs4_ds_server, list);
153 		list_del(&dss->list);
154 		rpc_shutdown_client(dss->rpc_clnt);
155 		kfree (dss);
156 	}
157 }
158 
159 void nfs41_shutdown_client(struct nfs_client *clp)
160 {
161 	if (nfs4_has_session(clp)) {
162 		nfs4_shutdown_ds_clients(clp);
163 		nfs4_destroy_session(clp->cl_session);
164 		nfs4_destroy_clientid(clp);
165 	}
166 
167 }
168 #endif	/* CONFIG_NFS_V4_1 */
169 
170 void nfs40_shutdown_client(struct nfs_client *clp)
171 {
172 	if (clp->cl_slot_tbl) {
173 		nfs4_shutdown_slot_table(clp->cl_slot_tbl);
174 		kfree(clp->cl_slot_tbl);
175 	}
176 }
177 
178 struct nfs_client *nfs4_alloc_client(const struct nfs_client_initdata *cl_init)
179 {
180 	int err;
181 	struct nfs_client *clp = nfs_alloc_client(cl_init);
182 	if (IS_ERR(clp))
183 		return clp;
184 
185 	err = nfs_get_cb_ident_idr(clp, cl_init->minorversion);
186 	if (err)
187 		goto error;
188 
189 	if (cl_init->minorversion > NFS4_MAX_MINOR_VERSION) {
190 		err = -EINVAL;
191 		goto error;
192 	}
193 
194 	spin_lock_init(&clp->cl_lock);
195 	INIT_DELAYED_WORK(&clp->cl_renewd, nfs4_renew_state);
196 	INIT_LIST_HEAD(&clp->cl_ds_clients);
197 	rpc_init_wait_queue(&clp->cl_rpcwaitq, "NFS client");
198 	clp->cl_state = 1 << NFS4CLNT_LEASE_EXPIRED;
199 	clp->cl_minorversion = cl_init->minorversion;
200 	clp->cl_mvops = nfs_v4_minor_ops[cl_init->minorversion];
201 	clp->cl_mig_gen = 1;
202 #if IS_ENABLED(CONFIG_NFS_V4_1)
203 	init_waitqueue_head(&clp->cl_lock_waitq);
204 #endif
205 	return clp;
206 
207 error:
208 	nfs_free_client(clp);
209 	return ERR_PTR(err);
210 }
211 
212 /*
213  * Destroy the NFS4 callback service
214  */
215 static void nfs4_destroy_callback(struct nfs_client *clp)
216 {
217 	if (__test_and_clear_bit(NFS_CS_CALLBACK, &clp->cl_res_state))
218 		nfs_callback_down(clp->cl_mvops->minor_version, clp->cl_net);
219 }
220 
221 static void nfs4_shutdown_client(struct nfs_client *clp)
222 {
223 	if (__test_and_clear_bit(NFS_CS_RENEWD, &clp->cl_res_state))
224 		nfs4_kill_renewd(clp);
225 	clp->cl_mvops->shutdown_client(clp);
226 	nfs4_destroy_callback(clp);
227 	if (__test_and_clear_bit(NFS_CS_IDMAP, &clp->cl_res_state))
228 		nfs_idmap_delete(clp);
229 
230 	rpc_destroy_wait_queue(&clp->cl_rpcwaitq);
231 	kfree(clp->cl_serverowner);
232 	kfree(clp->cl_serverscope);
233 	kfree(clp->cl_implid);
234 	kfree(clp->cl_owner_id);
235 }
236 
237 void nfs4_free_client(struct nfs_client *clp)
238 {
239 	nfs4_shutdown_client(clp);
240 	nfs_free_client(clp);
241 }
242 
243 /*
244  * Initialize the NFS4 callback service
245  */
246 static int nfs4_init_callback(struct nfs_client *clp)
247 {
248 	struct rpc_xprt *xprt;
249 	int error;
250 
251 	xprt = rcu_dereference_raw(clp->cl_rpcclient->cl_xprt);
252 
253 	if (nfs4_has_session(clp)) {
254 		error = xprt_setup_backchannel(xprt, NFS41_BC_MIN_CALLBACKS);
255 		if (error < 0)
256 			return error;
257 	}
258 
259 	error = nfs_callback_up(clp->cl_mvops->minor_version, xprt);
260 	if (error < 0) {
261 		dprintk("%s: failed to start callback. Error = %d\n",
262 			__func__, error);
263 		return error;
264 	}
265 	__set_bit(NFS_CS_CALLBACK, &clp->cl_res_state);
266 
267 	return 0;
268 }
269 
270 /**
271  * nfs40_init_client - nfs_client initialization tasks for NFSv4.0
272  * @clp - nfs_client to initialize
273  *
274  * Returns zero on success, or a negative errno if some error occurred.
275  */
276 int nfs40_init_client(struct nfs_client *clp)
277 {
278 	struct nfs4_slot_table *tbl;
279 	int ret;
280 
281 	tbl = kzalloc(sizeof(*tbl), GFP_NOFS);
282 	if (tbl == NULL)
283 		return -ENOMEM;
284 
285 	ret = nfs4_setup_slot_table(tbl, NFS4_MAX_SLOT_TABLE,
286 					"NFSv4.0 transport Slot table");
287 	if (ret) {
288 		kfree(tbl);
289 		return ret;
290 	}
291 
292 	clp->cl_slot_tbl = tbl;
293 	return 0;
294 }
295 
296 #if defined(CONFIG_NFS_V4_1)
297 
298 /**
299  * nfs41_init_client - nfs_client initialization tasks for NFSv4.1+
300  * @clp - nfs_client to initialize
301  *
302  * Returns zero on success, or a negative errno if some error occurred.
303  */
304 int nfs41_init_client(struct nfs_client *clp)
305 {
306 	struct nfs4_session *session = NULL;
307 
308 	/*
309 	 * Create the session and mark it expired.
310 	 * When a SEQUENCE operation encounters the expired session
311 	 * it will do session recovery to initialize it.
312 	 */
313 	session = nfs4_alloc_session(clp);
314 	if (!session)
315 		return -ENOMEM;
316 
317 	clp->cl_session = session;
318 
319 	/*
320 	 * The create session reply races with the server back
321 	 * channel probe. Mark the client NFS_CS_SESSION_INITING
322 	 * so that the client back channel can find the
323 	 * nfs_client struct
324 	 */
325 	nfs_mark_client_ready(clp, NFS_CS_SESSION_INITING);
326 	return 0;
327 }
328 
329 #endif	/* CONFIG_NFS_V4_1 */
330 
331 /*
332  * Initialize the minor version specific parts of an NFS4 client record
333  */
334 static int nfs4_init_client_minor_version(struct nfs_client *clp)
335 {
336 	int ret;
337 
338 	ret = clp->cl_mvops->init_client(clp);
339 	if (ret)
340 		return ret;
341 	return nfs4_init_callback(clp);
342 }
343 
344 /**
345  * nfs4_init_client - Initialise an NFS4 client record
346  *
347  * @clp: nfs_client to initialise
348  * @timeparms: timeout parameters for underlying RPC transport
349  * @ip_addr: callback IP address in presentation format
350  * @authflavor: authentication flavor for underlying RPC transport
351  *
352  * Returns pointer to an NFS client, or an ERR_PTR value.
353  */
354 struct nfs_client *nfs4_init_client(struct nfs_client *clp,
355 				    const struct nfs_client_initdata *cl_init)
356 {
357 	char buf[INET6_ADDRSTRLEN + 1];
358 	const char *ip_addr = cl_init->ip_addr;
359 	struct nfs_client *old;
360 	int error;
361 
362 	if (clp->cl_cons_state == NFS_CS_READY)
363 		/* the client is initialised already */
364 		return clp;
365 
366 	/* Check NFS protocol revision and initialize RPC op vector */
367 	clp->rpc_ops = &nfs_v4_clientops;
368 
369 	if (clp->cl_minorversion != 0)
370 		__set_bit(NFS_CS_INFINITE_SLOTS, &clp->cl_flags);
371 	__set_bit(NFS_CS_DISCRTRY, &clp->cl_flags);
372 	__set_bit(NFS_CS_NO_RETRANS_TIMEOUT, &clp->cl_flags);
373 
374 	error = nfs_create_rpc_client(clp, cl_init, RPC_AUTH_GSS_KRB5I);
375 	if (error == -EINVAL)
376 		error = nfs_create_rpc_client(clp, cl_init, RPC_AUTH_UNIX);
377 	if (error < 0)
378 		goto error;
379 
380 	/* If no clientaddr= option was specified, find a usable cb address */
381 	if (ip_addr == NULL) {
382 		struct sockaddr_storage cb_addr;
383 		struct sockaddr *sap = (struct sockaddr *)&cb_addr;
384 
385 		error = rpc_localaddr(clp->cl_rpcclient, sap, sizeof(cb_addr));
386 		if (error < 0)
387 			goto error;
388 		error = rpc_ntop(sap, buf, sizeof(buf));
389 		if (error < 0)
390 			goto error;
391 		ip_addr = (const char *)buf;
392 	}
393 	strlcpy(clp->cl_ipaddr, ip_addr, sizeof(clp->cl_ipaddr));
394 
395 	error = nfs_idmap_new(clp);
396 	if (error < 0) {
397 		dprintk("%s: failed to create idmapper. Error = %d\n",
398 			__func__, error);
399 		goto error;
400 	}
401 	__set_bit(NFS_CS_IDMAP, &clp->cl_res_state);
402 
403 	error = nfs4_init_client_minor_version(clp);
404 	if (error < 0)
405 		goto error;
406 
407 	error = nfs4_discover_server_trunking(clp, &old);
408 	if (error < 0)
409 		goto error;
410 
411 	if (clp != old) {
412 		clp->cl_preserve_clid = true;
413 		/*
414 		 * Mark the client as having failed initialization so other
415 		 * processes walking the nfs_client_list in nfs_match_client()
416 		 * won't try to use it.
417 		 */
418 		nfs_mark_client_ready(clp, -EPERM);
419 	}
420 	nfs_put_client(clp);
421 	clear_bit(NFS_CS_TSM_POSSIBLE, &clp->cl_flags);
422 	return old;
423 
424 error:
425 	nfs_mark_client_ready(clp, error);
426 	nfs_put_client(clp);
427 	return ERR_PTR(error);
428 }
429 
430 /*
431  * SETCLIENTID just did a callback update with the callback ident in
432  * "drop," but server trunking discovery claims "drop" and "keep" are
433  * actually the same server.  Swap the callback IDs so that "keep"
434  * will continue to use the callback ident the server now knows about,
435  * and so that "keep"'s original callback ident is destroyed when
436  * "drop" is freed.
437  */
438 static void nfs4_swap_callback_idents(struct nfs_client *keep,
439 				      struct nfs_client *drop)
440 {
441 	struct nfs_net *nn = net_generic(keep->cl_net, nfs_net_id);
442 	unsigned int save = keep->cl_cb_ident;
443 
444 	if (keep->cl_cb_ident == drop->cl_cb_ident)
445 		return;
446 
447 	dprintk("%s: keeping callback ident %u and dropping ident %u\n",
448 		__func__, keep->cl_cb_ident, drop->cl_cb_ident);
449 
450 	spin_lock(&nn->nfs_client_lock);
451 
452 	idr_replace(&nn->cb_ident_idr, keep, drop->cl_cb_ident);
453 	keep->cl_cb_ident = drop->cl_cb_ident;
454 
455 	idr_replace(&nn->cb_ident_idr, drop, save);
456 	drop->cl_cb_ident = save;
457 
458 	spin_unlock(&nn->nfs_client_lock);
459 }
460 
461 static bool nfs4_match_client_owner_id(const struct nfs_client *clp1,
462 		const struct nfs_client *clp2)
463 {
464 	if (clp1->cl_owner_id == NULL || clp2->cl_owner_id == NULL)
465 		return true;
466 	return strcmp(clp1->cl_owner_id, clp2->cl_owner_id) == 0;
467 }
468 
469 static bool nfs4_same_verifier(nfs4_verifier *v1, nfs4_verifier *v2)
470 {
471 	return memcmp(v1->data, v2->data, sizeof(v1->data)) == 0;
472 }
473 
474 static int nfs4_match_client(struct nfs_client  *pos,  struct nfs_client *new,
475 			     struct nfs_client **prev, struct nfs_net *nn)
476 {
477 	int status;
478 
479 	if (pos->rpc_ops != new->rpc_ops)
480 		return 1;
481 
482 	if (pos->cl_minorversion != new->cl_minorversion)
483 		return 1;
484 
485 	/* If "pos" isn't marked ready, we can't trust the
486 	 * remaining fields in "pos", especially the client
487 	 * ID and serverowner fields.  Wait for CREATE_SESSION
488 	 * to finish. */
489 	if (pos->cl_cons_state > NFS_CS_READY) {
490 		refcount_inc(&pos->cl_count);
491 		spin_unlock(&nn->nfs_client_lock);
492 
493 		nfs_put_client(*prev);
494 		*prev = pos;
495 
496 		status = nfs_wait_client_init_complete(pos);
497 		spin_lock(&nn->nfs_client_lock);
498 
499 		if (status < 0)
500 			return status;
501 	}
502 
503 	if (pos->cl_cons_state != NFS_CS_READY)
504 		return 1;
505 
506 	if (pos->cl_clientid != new->cl_clientid)
507 		return 1;
508 
509 	/* NFSv4.1 always uses the uniform string, however someone
510 	 * might switch the uniquifier string on us.
511 	 */
512 	if (!nfs4_match_client_owner_id(pos, new))
513 		return 1;
514 
515 	return 0;
516 }
517 
518 /**
519  * nfs40_walk_client_list - Find server that recognizes a client ID
520  *
521  * @new: nfs_client with client ID to test
522  * @result: OUT: found nfs_client, or new
523  * @cred: credential to use for trunking test
524  *
525  * Returns zero, a negative errno, or a negative NFS4ERR status.
526  * If zero is returned, an nfs_client pointer is planted in "result."
527  *
528  * NB: nfs40_walk_client_list() relies on the new nfs_client being
529  *     the last nfs_client on the list.
530  */
531 int nfs40_walk_client_list(struct nfs_client *new,
532 			   struct nfs_client **result,
533 			   struct rpc_cred *cred)
534 {
535 	struct nfs_net *nn = net_generic(new->cl_net, nfs_net_id);
536 	struct nfs_client *pos, *prev = NULL;
537 	struct nfs4_setclientid_res clid = {
538 		.clientid	= new->cl_clientid,
539 		.confirm	= new->cl_confirm,
540 	};
541 	int status = -NFS4ERR_STALE_CLIENTID;
542 
543 	spin_lock(&nn->nfs_client_lock);
544 	list_for_each_entry(pos, &nn->nfs_client_list, cl_share_link) {
545 
546 		if (pos == new)
547 			goto found;
548 
549 		status = nfs4_match_client(pos, new, &prev, nn);
550 		if (status < 0)
551 			goto out_unlock;
552 		if (status != 0)
553 			continue;
554 		/*
555 		 * We just sent a new SETCLIENTID, which should have
556 		 * caused the server to return a new cl_confirm.  So if
557 		 * cl_confirm is the same, then this is a different
558 		 * server that just returned the same cl_confirm by
559 		 * coincidence:
560 		 */
561 		if ((new != pos) && nfs4_same_verifier(&pos->cl_confirm,
562 						       &new->cl_confirm))
563 			continue;
564 		/*
565 		 * But if the cl_confirm's are different, then the only
566 		 * way that a SETCLIENTID_CONFIRM to pos can succeed is
567 		 * if new and pos point to the same server:
568 		 */
569 found:
570 		refcount_inc(&pos->cl_count);
571 		spin_unlock(&nn->nfs_client_lock);
572 
573 		nfs_put_client(prev);
574 		prev = pos;
575 
576 		status = nfs4_proc_setclientid_confirm(pos, &clid, cred);
577 		switch (status) {
578 		case -NFS4ERR_STALE_CLIENTID:
579 			break;
580 		case 0:
581 			nfs4_swap_callback_idents(pos, new);
582 			pos->cl_confirm = new->cl_confirm;
583 			nfs_mark_client_ready(pos, NFS_CS_READY);
584 
585 			prev = NULL;
586 			*result = pos;
587 			goto out;
588 		case -ERESTARTSYS:
589 		case -ETIMEDOUT:
590 			/* The callback path may have been inadvertently
591 			 * changed. Schedule recovery!
592 			 */
593 			nfs4_schedule_path_down_recovery(pos);
594 		default:
595 			goto out;
596 		}
597 
598 		spin_lock(&nn->nfs_client_lock);
599 	}
600 out_unlock:
601 	spin_unlock(&nn->nfs_client_lock);
602 
603 	/* No match found. The server lost our clientid */
604 out:
605 	nfs_put_client(prev);
606 	return status;
607 }
608 
609 #ifdef CONFIG_NFS_V4_1
610 /*
611  * Returns true if the server major ids match
612  */
613 static bool
614 nfs4_check_serverowner_major_id(struct nfs41_server_owner *o1,
615 				struct nfs41_server_owner *o2)
616 {
617 	if (o1->major_id_sz != o2->major_id_sz)
618 		return false;
619 	return memcmp(o1->major_id, o2->major_id, o1->major_id_sz) == 0;
620 }
621 
622 /*
623  * Returns true if the server scopes match
624  */
625 static bool
626 nfs4_check_server_scope(struct nfs41_server_scope *s1,
627 			struct nfs41_server_scope *s2)
628 {
629 	if (s1->server_scope_sz != s2->server_scope_sz)
630 		return false;
631 	return memcmp(s1->server_scope, s2->server_scope,
632 					s1->server_scope_sz) == 0;
633 }
634 
635 /**
636  * nfs4_detect_session_trunking - Checks for session trunking.
637  *
638  * Called after a successful EXCHANGE_ID on a multi-addr connection.
639  * Upon success, add the transport.
640  *
641  * @clp:    original mount nfs_client
642  * @res:    result structure from an exchange_id using the original mount
643  *          nfs_client with a new multi_addr transport
644  *
645  * Returns zero on success, otherwise -EINVAL
646  *
647  * Note: since the exchange_id for the new multi_addr transport uses the
648  * same nfs_client from the original mount, the cl_owner_id is reused,
649  * so eir_clientowner is the same.
650  */
651 int nfs4_detect_session_trunking(struct nfs_client *clp,
652 				 struct nfs41_exchange_id_res *res,
653 				 struct rpc_xprt *xprt)
654 {
655 	/* Check eir_clientid */
656 	if (clp->cl_clientid != res->clientid)
657 		goto out_err;
658 
659 	/* Check eir_server_owner so_major_id */
660 	if (!nfs4_check_serverowner_major_id(clp->cl_serverowner,
661 					     res->server_owner))
662 		goto out_err;
663 
664 	/* Check eir_server_owner so_minor_id */
665 	if (clp->cl_serverowner->minor_id != res->server_owner->minor_id)
666 		goto out_err;
667 
668 	/* Check eir_server_scope */
669 	if (!nfs4_check_server_scope(clp->cl_serverscope, res->server_scope))
670 		goto out_err;
671 
672 	pr_info("NFS:  %s: Session trunking succeeded for %s\n",
673 		clp->cl_hostname,
674 		xprt->address_strings[RPC_DISPLAY_ADDR]);
675 
676 	return 0;
677 out_err:
678 	pr_info("NFS:  %s: Session trunking failed for %s\n", clp->cl_hostname,
679 		xprt->address_strings[RPC_DISPLAY_ADDR]);
680 
681 	return -EINVAL;
682 }
683 
684 /**
685  * nfs41_walk_client_list - Find nfs_client that matches a client/server owner
686  *
687  * @new: nfs_client with client ID to test
688  * @result: OUT: found nfs_client, or new
689  * @cred: credential to use for trunking test
690  *
691  * Returns zero, a negative errno, or a negative NFS4ERR status.
692  * If zero is returned, an nfs_client pointer is planted in "result."
693  *
694  * NB: nfs41_walk_client_list() relies on the new nfs_client being
695  *     the last nfs_client on the list.
696  */
697 int nfs41_walk_client_list(struct nfs_client *new,
698 			   struct nfs_client **result,
699 			   struct rpc_cred *cred)
700 {
701 	struct nfs_net *nn = net_generic(new->cl_net, nfs_net_id);
702 	struct nfs_client *pos, *prev = NULL;
703 	int status = -NFS4ERR_STALE_CLIENTID;
704 
705 	spin_lock(&nn->nfs_client_lock);
706 	list_for_each_entry(pos, &nn->nfs_client_list, cl_share_link) {
707 
708 		if (pos == new)
709 			goto found;
710 
711 		status = nfs4_match_client(pos, new, &prev, nn);
712 		if (status < 0)
713 			goto out;
714 		if (status != 0)
715 			continue;
716 
717 		/*
718 		 * Note that session trunking is just a special subcase of
719 		 * client id trunking. In either case, we want to fall back
720 		 * to using the existing nfs_client.
721 		 */
722 		if (!nfs4_check_serverowner_major_id(pos->cl_serverowner,
723 						     new->cl_serverowner))
724 			continue;
725 
726 found:
727 		refcount_inc(&pos->cl_count);
728 		*result = pos;
729 		status = 0;
730 		break;
731 	}
732 
733 out:
734 	spin_unlock(&nn->nfs_client_lock);
735 	nfs_put_client(prev);
736 	return status;
737 }
738 #endif	/* CONFIG_NFS_V4_1 */
739 
740 static void nfs4_destroy_server(struct nfs_server *server)
741 {
742 	nfs_server_return_all_delegations(server);
743 	unset_pnfs_layoutdriver(server);
744 	nfs4_purge_state_owners(server);
745 }
746 
747 /*
748  * NFSv4.0 callback thread helper
749  *
750  * Find a client by callback identifier
751  */
752 struct nfs_client *
753 nfs4_find_client_ident(struct net *net, int cb_ident)
754 {
755 	struct nfs_client *clp;
756 	struct nfs_net *nn = net_generic(net, nfs_net_id);
757 
758 	spin_lock(&nn->nfs_client_lock);
759 	clp = idr_find(&nn->cb_ident_idr, cb_ident);
760 	if (clp)
761 		refcount_inc(&clp->cl_count);
762 	spin_unlock(&nn->nfs_client_lock);
763 	return clp;
764 }
765 
766 #if defined(CONFIG_NFS_V4_1)
767 /* Common match routine for v4.0 and v4.1 callback services */
768 static bool nfs4_cb_match_client(const struct sockaddr *addr,
769 		struct nfs_client *clp, u32 minorversion)
770 {
771 	struct sockaddr *clap = (struct sockaddr *)&clp->cl_addr;
772 
773 	/* Don't match clients that failed to initialise */
774 	if (!(clp->cl_cons_state == NFS_CS_READY ||
775 	    clp->cl_cons_state == NFS_CS_SESSION_INITING))
776 		return false;
777 
778 	smp_rmb();
779 
780 	/* Match the version and minorversion */
781 	if (clp->rpc_ops->version != 4 ||
782 	    clp->cl_minorversion != minorversion)
783 		return false;
784 
785 	/* Match only the IP address, not the port number */
786 	return rpc_cmp_addr(addr, clap);
787 }
788 
789 /*
790  * NFSv4.1 callback thread helper
791  * For CB_COMPOUND calls, find a client by IP address, protocol version,
792  * minorversion, and sessionID
793  *
794  * Returns NULL if no such client
795  */
796 struct nfs_client *
797 nfs4_find_client_sessionid(struct net *net, const struct sockaddr *addr,
798 			   struct nfs4_sessionid *sid, u32 minorversion)
799 {
800 	struct nfs_client *clp;
801 	struct nfs_net *nn = net_generic(net, nfs_net_id);
802 
803 	spin_lock(&nn->nfs_client_lock);
804 	list_for_each_entry(clp, &nn->nfs_client_list, cl_share_link) {
805 		if (!nfs4_cb_match_client(addr, clp, minorversion))
806 			continue;
807 
808 		if (!nfs4_has_session(clp))
809 			continue;
810 
811 		/* Match sessionid*/
812 		if (memcmp(clp->cl_session->sess_id.data,
813 		    sid->data, NFS4_MAX_SESSIONID_LEN) != 0)
814 			continue;
815 
816 		refcount_inc(&clp->cl_count);
817 		spin_unlock(&nn->nfs_client_lock);
818 		return clp;
819 	}
820 	spin_unlock(&nn->nfs_client_lock);
821 	return NULL;
822 }
823 
824 #else /* CONFIG_NFS_V4_1 */
825 
826 struct nfs_client *
827 nfs4_find_client_sessionid(struct net *net, const struct sockaddr *addr,
828 			   struct nfs4_sessionid *sid, u32 minorversion)
829 {
830 	return NULL;
831 }
832 #endif /* CONFIG_NFS_V4_1 */
833 
834 /*
835  * Set up an NFS4 client
836  */
837 static int nfs4_set_client(struct nfs_server *server,
838 		const char *hostname,
839 		const struct sockaddr *addr,
840 		const size_t addrlen,
841 		const char *ip_addr,
842 		int proto, const struct rpc_timeout *timeparms,
843 		u32 minorversion, struct net *net)
844 {
845 	struct nfs_client_initdata cl_init = {
846 		.hostname = hostname,
847 		.addr = addr,
848 		.addrlen = addrlen,
849 		.ip_addr = ip_addr,
850 		.nfs_mod = &nfs_v4,
851 		.proto = proto,
852 		.minorversion = minorversion,
853 		.net = net,
854 		.timeparms = timeparms,
855 	};
856 	struct nfs_client *clp;
857 
858 	if (server->flags & NFS_MOUNT_NORESVPORT)
859 		set_bit(NFS_CS_NORESVPORT, &cl_init.init_flags);
860 	if (server->options & NFS_OPTION_MIGRATION)
861 		set_bit(NFS_CS_MIGRATION, &cl_init.init_flags);
862 	if (test_bit(NFS_MIG_TSM_POSSIBLE, &server->mig_status))
863 		set_bit(NFS_CS_TSM_POSSIBLE, &cl_init.init_flags);
864 	server->port = rpc_get_port(addr);
865 
866 	/* Allocate or find a client reference we can use */
867 	clp = nfs_get_client(&cl_init);
868 	if (IS_ERR(clp))
869 		return PTR_ERR(clp);
870 
871 	if (server->nfs_client == clp) {
872 		nfs_put_client(clp);
873 		return -ELOOP;
874 	}
875 
876 	/*
877 	 * Query for the lease time on clientid setup or renewal
878 	 *
879 	 * Note that this will be set on nfs_clients that were created
880 	 * only for the DS role and did not set this bit, but now will
881 	 * serve a dual role.
882 	 */
883 	set_bit(NFS_CS_CHECK_LEASE_TIME, &clp->cl_res_state);
884 
885 	server->nfs_client = clp;
886 	return 0;
887 }
888 
889 /*
890  * Set up a pNFS Data Server client.
891  *
892  * Return any existing nfs_client that matches server address,port,version
893  * and minorversion.
894  *
895  * For a new nfs_client, use a soft mount (default), a low retrans and a
896  * low timeout interval so that if a connection is lost, we retry through
897  * the MDS.
898  */
899 struct nfs_client *nfs4_set_ds_client(struct nfs_server *mds_srv,
900 		const struct sockaddr *ds_addr, int ds_addrlen,
901 		int ds_proto, unsigned int ds_timeo, unsigned int ds_retrans,
902 		u32 minor_version)
903 {
904 	struct rpc_timeout ds_timeout;
905 	struct nfs_client *mds_clp = mds_srv->nfs_client;
906 	struct nfs_client_initdata cl_init = {
907 		.addr = ds_addr,
908 		.addrlen = ds_addrlen,
909 		.nodename = mds_clp->cl_rpcclient->cl_nodename,
910 		.ip_addr = mds_clp->cl_ipaddr,
911 		.nfs_mod = &nfs_v4,
912 		.proto = ds_proto,
913 		.minorversion = minor_version,
914 		.net = mds_clp->cl_net,
915 		.timeparms = &ds_timeout,
916 	};
917 	char buf[INET6_ADDRSTRLEN + 1];
918 
919 	if (rpc_ntop(ds_addr, buf, sizeof(buf)) <= 0)
920 		return ERR_PTR(-EINVAL);
921 	cl_init.hostname = buf;
922 
923 	if (mds_srv->flags & NFS_MOUNT_NORESVPORT)
924 		__set_bit(NFS_CS_NORESVPORT, &cl_init.init_flags);
925 
926 	/*
927 	 * Set an authflavor equual to the MDS value. Use the MDS nfs_client
928 	 * cl_ipaddr so as to use the same EXCHANGE_ID co_ownerid as the MDS
929 	 * (section 13.1 RFC 5661).
930 	 */
931 	nfs_init_timeout_values(&ds_timeout, ds_proto, ds_timeo, ds_retrans);
932 	return nfs_get_client(&cl_init);
933 }
934 EXPORT_SYMBOL_GPL(nfs4_set_ds_client);
935 
936 /*
937  * Session has been established, and the client marked ready.
938  * Set the mount rsize and wsize with negotiated fore channel
939  * attributes which will be bound checked in nfs_server_set_fsinfo.
940  */
941 static void nfs4_session_set_rwsize(struct nfs_server *server)
942 {
943 #ifdef CONFIG_NFS_V4_1
944 	struct nfs4_session *sess;
945 	u32 server_resp_sz;
946 	u32 server_rqst_sz;
947 
948 	if (!nfs4_has_session(server->nfs_client))
949 		return;
950 	sess = server->nfs_client->cl_session;
951 	server_resp_sz = sess->fc_attrs.max_resp_sz - nfs41_maxread_overhead;
952 	server_rqst_sz = sess->fc_attrs.max_rqst_sz - nfs41_maxwrite_overhead;
953 
954 	if (!server->rsize || server->rsize > server_resp_sz)
955 		server->rsize = server_resp_sz;
956 	if (!server->wsize || server->wsize > server_rqst_sz)
957 		server->wsize = server_rqst_sz;
958 #endif /* CONFIG_NFS_V4_1 */
959 }
960 
961 static int nfs4_server_common_setup(struct nfs_server *server,
962 		struct nfs_fh *mntfh, bool auth_probe)
963 {
964 	struct nfs_fattr *fattr;
965 	int error;
966 
967 	/* data servers support only a subset of NFSv4.1 */
968 	if (is_ds_only_client(server->nfs_client))
969 		return -EPROTONOSUPPORT;
970 
971 	fattr = nfs_alloc_fattr();
972 	if (fattr == NULL)
973 		return -ENOMEM;
974 
975 	/* We must ensure the session is initialised first */
976 	error = nfs4_init_session(server->nfs_client);
977 	if (error < 0)
978 		goto out;
979 
980 	/* Set the basic capabilities */
981 	server->caps |= server->nfs_client->cl_mvops->init_caps;
982 	if (server->flags & NFS_MOUNT_NORDIRPLUS)
983 			server->caps &= ~NFS_CAP_READDIRPLUS;
984 	/*
985 	 * Don't use NFS uid/gid mapping if we're using AUTH_SYS or lower
986 	 * authentication.
987 	 */
988 	if (nfs4_disable_idmapping &&
989 			server->client->cl_auth->au_flavor == RPC_AUTH_UNIX)
990 		server->caps |= NFS_CAP_UIDGID_NOMAP;
991 
992 
993 	/* Probe the root fh to retrieve its FSID and filehandle */
994 	error = nfs4_get_rootfh(server, mntfh, auth_probe);
995 	if (error < 0)
996 		goto out;
997 
998 	dprintk("Server FSID: %llx:%llx\n",
999 			(unsigned long long) server->fsid.major,
1000 			(unsigned long long) server->fsid.minor);
1001 	nfs_display_fhandle(mntfh, "Pseudo-fs root FH");
1002 
1003 	nfs4_session_set_rwsize(server);
1004 
1005 	error = nfs_probe_fsinfo(server, mntfh, fattr);
1006 	if (error < 0)
1007 		goto out;
1008 
1009 	if (server->namelen == 0 || server->namelen > NFS4_MAXNAMLEN)
1010 		server->namelen = NFS4_MAXNAMLEN;
1011 
1012 	nfs_server_insert_lists(server);
1013 	server->mount_time = jiffies;
1014 	server->destroy = nfs4_destroy_server;
1015 out:
1016 	nfs_free_fattr(fattr);
1017 	return error;
1018 }
1019 
1020 /*
1021  * Create a version 4 volume record
1022  */
1023 static int nfs4_init_server(struct nfs_server *server,
1024 		struct nfs_parsed_mount_data *data)
1025 {
1026 	struct rpc_timeout timeparms;
1027 	int error;
1028 
1029 	nfs_init_timeout_values(&timeparms, data->nfs_server.protocol,
1030 			data->timeo, data->retrans);
1031 
1032 	/* Initialise the client representation from the mount data */
1033 	server->flags = data->flags;
1034 	server->options = data->options;
1035 	server->auth_info = data->auth_info;
1036 
1037 	/* Use the first specified auth flavor. If this flavor isn't
1038 	 * allowed by the server, use the SECINFO path to try the
1039 	 * other specified flavors */
1040 	if (data->auth_info.flavor_len >= 1)
1041 		data->selected_flavor = data->auth_info.flavors[0];
1042 	else
1043 		data->selected_flavor = RPC_AUTH_UNIX;
1044 
1045 	/* Get a client record */
1046 	error = nfs4_set_client(server,
1047 			data->nfs_server.hostname,
1048 			(const struct sockaddr *)&data->nfs_server.address,
1049 			data->nfs_server.addrlen,
1050 			data->client_address,
1051 			data->nfs_server.protocol,
1052 			&timeparms,
1053 			data->minorversion,
1054 			data->net);
1055 	if (error < 0)
1056 		return error;
1057 
1058 	if (data->rsize)
1059 		server->rsize = nfs_block_size(data->rsize, NULL);
1060 	if (data->wsize)
1061 		server->wsize = nfs_block_size(data->wsize, NULL);
1062 
1063 	server->acregmin = data->acregmin * HZ;
1064 	server->acregmax = data->acregmax * HZ;
1065 	server->acdirmin = data->acdirmin * HZ;
1066 	server->acdirmax = data->acdirmax * HZ;
1067 	server->port     = data->nfs_server.port;
1068 
1069 	return nfs_init_server_rpcclient(server, &timeparms,
1070 					 data->selected_flavor);
1071 }
1072 
1073 /*
1074  * Create a version 4 volume record
1075  * - keyed on server and FSID
1076  */
1077 /*struct nfs_server *nfs4_create_server(const struct nfs_parsed_mount_data *data,
1078 				      struct nfs_fh *mntfh)*/
1079 struct nfs_server *nfs4_create_server(struct nfs_mount_info *mount_info,
1080 				      struct nfs_subversion *nfs_mod)
1081 {
1082 	struct nfs_server *server;
1083 	bool auth_probe;
1084 	int error;
1085 
1086 	server = nfs_alloc_server();
1087 	if (!server)
1088 		return ERR_PTR(-ENOMEM);
1089 
1090 	auth_probe = mount_info->parsed->auth_info.flavor_len < 1;
1091 
1092 	/* set up the general RPC client */
1093 	error = nfs4_init_server(server, mount_info->parsed);
1094 	if (error < 0)
1095 		goto error;
1096 
1097 	error = nfs4_server_common_setup(server, mount_info->mntfh, auth_probe);
1098 	if (error < 0)
1099 		goto error;
1100 
1101 	return server;
1102 
1103 error:
1104 	nfs_free_server(server);
1105 	return ERR_PTR(error);
1106 }
1107 
1108 /*
1109  * Create an NFS4 referral server record
1110  */
1111 struct nfs_server *nfs4_create_referral_server(struct nfs_clone_mount *data,
1112 					       struct nfs_fh *mntfh)
1113 {
1114 	struct nfs_client *parent_client;
1115 	struct nfs_server *server, *parent_server;
1116 	bool auth_probe;
1117 	int error;
1118 
1119 	server = nfs_alloc_server();
1120 	if (!server)
1121 		return ERR_PTR(-ENOMEM);
1122 
1123 	parent_server = NFS_SB(data->sb);
1124 	parent_client = parent_server->nfs_client;
1125 
1126 	/* Initialise the client representation from the parent server */
1127 	nfs_server_copy_userdata(server, parent_server);
1128 
1129 	/* Get a client representation */
1130 #ifdef CONFIG_SUNRPC_XPRT_RDMA
1131 	rpc_set_port(data->addr, NFS_RDMA_PORT);
1132 	error = nfs4_set_client(server, data->hostname,
1133 				data->addr,
1134 				data->addrlen,
1135 				parent_client->cl_ipaddr,
1136 				XPRT_TRANSPORT_RDMA,
1137 				parent_server->client->cl_timeout,
1138 				parent_client->cl_mvops->minor_version,
1139 				parent_client->cl_net);
1140 	if (!error)
1141 		goto init_server;
1142 #endif	/* CONFIG_SUNRPC_XPRT_RDMA */
1143 
1144 	rpc_set_port(data->addr, NFS_PORT);
1145 	error = nfs4_set_client(server, data->hostname,
1146 				data->addr,
1147 				data->addrlen,
1148 				parent_client->cl_ipaddr,
1149 				XPRT_TRANSPORT_TCP,
1150 				parent_server->client->cl_timeout,
1151 				parent_client->cl_mvops->minor_version,
1152 				parent_client->cl_net);
1153 	if (error < 0)
1154 		goto error;
1155 
1156 #ifdef CONFIG_SUNRPC_XPRT_RDMA
1157 init_server:
1158 #endif
1159 	error = nfs_init_server_rpcclient(server, parent_server->client->cl_timeout, data->authflavor);
1160 	if (error < 0)
1161 		goto error;
1162 
1163 	auth_probe = parent_server->auth_info.flavor_len < 1;
1164 
1165 	error = nfs4_server_common_setup(server, mntfh, auth_probe);
1166 	if (error < 0)
1167 		goto error;
1168 
1169 	return server;
1170 
1171 error:
1172 	nfs_free_server(server);
1173 	return ERR_PTR(error);
1174 }
1175 
1176 /*
1177  * Grab the destination's particulars, including lease expiry time.
1178  *
1179  * Returns zero if probe succeeded and retrieved FSID matches the FSID
1180  * we have cached.
1181  */
1182 static int nfs_probe_destination(struct nfs_server *server)
1183 {
1184 	struct inode *inode = d_inode(server->super->s_root);
1185 	struct nfs_fattr *fattr;
1186 	int error;
1187 
1188 	fattr = nfs_alloc_fattr();
1189 	if (fattr == NULL)
1190 		return -ENOMEM;
1191 
1192 	/* Sanity: the probe won't work if the destination server
1193 	 * does not recognize the migrated FH. */
1194 	error = nfs_probe_fsinfo(server, NFS_FH(inode), fattr);
1195 
1196 	nfs_free_fattr(fattr);
1197 	return error;
1198 }
1199 
1200 /**
1201  * nfs4_update_server - Move an nfs_server to a different nfs_client
1202  *
1203  * @server: represents FSID to be moved
1204  * @hostname: new end-point's hostname
1205  * @sap: new end-point's socket address
1206  * @salen: size of "sap"
1207  * @net: net namespace
1208  *
1209  * The nfs_server must be quiescent before this function is invoked.
1210  * Either its session is drained (NFSv4.1+), or its transport is
1211  * plugged and drained (NFSv4.0).
1212  *
1213  * Returns zero on success, or a negative errno value.
1214  */
1215 int nfs4_update_server(struct nfs_server *server, const char *hostname,
1216 		       struct sockaddr *sap, size_t salen, struct net *net)
1217 {
1218 	struct nfs_client *clp = server->nfs_client;
1219 	struct rpc_clnt *clnt = server->client;
1220 	struct xprt_create xargs = {
1221 		.ident		= clp->cl_proto,
1222 		.net		= net,
1223 		.dstaddr	= sap,
1224 		.addrlen	= salen,
1225 		.servername	= hostname,
1226 	};
1227 	char buf[INET6_ADDRSTRLEN + 1];
1228 	struct sockaddr_storage address;
1229 	struct sockaddr *localaddr = (struct sockaddr *)&address;
1230 	int error;
1231 
1232 	error = rpc_switch_client_transport(clnt, &xargs, clnt->cl_timeout);
1233 	if (error != 0)
1234 		return error;
1235 
1236 	error = rpc_localaddr(clnt, localaddr, sizeof(address));
1237 	if (error != 0)
1238 		return error;
1239 
1240 	if (rpc_ntop(localaddr, buf, sizeof(buf)) == 0)
1241 		return -EAFNOSUPPORT;
1242 
1243 	nfs_server_remove_lists(server);
1244 	set_bit(NFS_MIG_TSM_POSSIBLE, &server->mig_status);
1245 	error = nfs4_set_client(server, hostname, sap, salen, buf,
1246 				clp->cl_proto, clnt->cl_timeout,
1247 				clp->cl_minorversion, net);
1248 	clear_bit(NFS_MIG_TSM_POSSIBLE, &server->mig_status);
1249 	if (error != 0) {
1250 		nfs_server_insert_lists(server);
1251 		return error;
1252 	}
1253 	nfs_put_client(clp);
1254 
1255 	if (server->nfs_client->cl_hostname == NULL)
1256 		server->nfs_client->cl_hostname = kstrdup(hostname, GFP_KERNEL);
1257 	nfs_server_insert_lists(server);
1258 
1259 	return nfs_probe_destination(server);
1260 }
1261