xref: /openbmc/linux/fs/nfs/nfs4client.c (revision 5a170e9e)
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 static void
160 nfs4_cleanup_callback(struct nfs_client *clp)
161 {
162 	struct nfs4_copy_state *cp_state;
163 
164 	while (!list_empty(&clp->pending_cb_stateids)) {
165 		cp_state = list_entry(clp->pending_cb_stateids.next,
166 					struct nfs4_copy_state, copies);
167 		list_del(&cp_state->copies);
168 		kfree(cp_state);
169 	}
170 }
171 
172 void nfs41_shutdown_client(struct nfs_client *clp)
173 {
174 	if (nfs4_has_session(clp)) {
175 		nfs4_cleanup_callback(clp);
176 		nfs4_shutdown_ds_clients(clp);
177 		nfs4_destroy_session(clp->cl_session);
178 		nfs4_destroy_clientid(clp);
179 	}
180 
181 }
182 #endif	/* CONFIG_NFS_V4_1 */
183 
184 void nfs40_shutdown_client(struct nfs_client *clp)
185 {
186 	if (clp->cl_slot_tbl) {
187 		nfs4_shutdown_slot_table(clp->cl_slot_tbl);
188 		kfree(clp->cl_slot_tbl);
189 	}
190 }
191 
192 struct nfs_client *nfs4_alloc_client(const struct nfs_client_initdata *cl_init)
193 {
194 	int err;
195 	struct nfs_client *clp = nfs_alloc_client(cl_init);
196 	if (IS_ERR(clp))
197 		return clp;
198 
199 	err = nfs_get_cb_ident_idr(clp, cl_init->minorversion);
200 	if (err)
201 		goto error;
202 
203 	if (cl_init->minorversion > NFS4_MAX_MINOR_VERSION) {
204 		err = -EINVAL;
205 		goto error;
206 	}
207 
208 	spin_lock_init(&clp->cl_lock);
209 	INIT_DELAYED_WORK(&clp->cl_renewd, nfs4_renew_state);
210 	INIT_LIST_HEAD(&clp->cl_ds_clients);
211 	rpc_init_wait_queue(&clp->cl_rpcwaitq, "NFS client");
212 	clp->cl_state = 1 << NFS4CLNT_LEASE_EXPIRED;
213 	clp->cl_minorversion = cl_init->minorversion;
214 	clp->cl_mvops = nfs_v4_minor_ops[cl_init->minorversion];
215 	clp->cl_mig_gen = 1;
216 #if IS_ENABLED(CONFIG_NFS_V4_1)
217 	init_waitqueue_head(&clp->cl_lock_waitq);
218 #endif
219 	INIT_LIST_HEAD(&clp->pending_cb_stateids);
220 	return clp;
221 
222 error:
223 	nfs_free_client(clp);
224 	return ERR_PTR(err);
225 }
226 
227 /*
228  * Destroy the NFS4 callback service
229  */
230 static void nfs4_destroy_callback(struct nfs_client *clp)
231 {
232 	if (__test_and_clear_bit(NFS_CS_CALLBACK, &clp->cl_res_state))
233 		nfs_callback_down(clp->cl_mvops->minor_version, clp->cl_net);
234 }
235 
236 static void nfs4_shutdown_client(struct nfs_client *clp)
237 {
238 	if (__test_and_clear_bit(NFS_CS_RENEWD, &clp->cl_res_state))
239 		nfs4_kill_renewd(clp);
240 	clp->cl_mvops->shutdown_client(clp);
241 	nfs4_destroy_callback(clp);
242 	if (__test_and_clear_bit(NFS_CS_IDMAP, &clp->cl_res_state))
243 		nfs_idmap_delete(clp);
244 
245 	rpc_destroy_wait_queue(&clp->cl_rpcwaitq);
246 	kfree(clp->cl_serverowner);
247 	kfree(clp->cl_serverscope);
248 	kfree(clp->cl_implid);
249 	kfree(clp->cl_owner_id);
250 }
251 
252 void nfs4_free_client(struct nfs_client *clp)
253 {
254 	nfs4_shutdown_client(clp);
255 	nfs_free_client(clp);
256 }
257 
258 /*
259  * Initialize the NFS4 callback service
260  */
261 static int nfs4_init_callback(struct nfs_client *clp)
262 {
263 	struct rpc_xprt *xprt;
264 	int error;
265 
266 	xprt = rcu_dereference_raw(clp->cl_rpcclient->cl_xprt);
267 
268 	if (nfs4_has_session(clp)) {
269 		error = xprt_setup_backchannel(xprt, NFS41_BC_MIN_CALLBACKS);
270 		if (error < 0)
271 			return error;
272 	}
273 
274 	error = nfs_callback_up(clp->cl_mvops->minor_version, xprt);
275 	if (error < 0) {
276 		dprintk("%s: failed to start callback. Error = %d\n",
277 			__func__, error);
278 		return error;
279 	}
280 	__set_bit(NFS_CS_CALLBACK, &clp->cl_res_state);
281 
282 	return 0;
283 }
284 
285 /**
286  * nfs40_init_client - nfs_client initialization tasks for NFSv4.0
287  * @clp - nfs_client to initialize
288  *
289  * Returns zero on success, or a negative errno if some error occurred.
290  */
291 int nfs40_init_client(struct nfs_client *clp)
292 {
293 	struct nfs4_slot_table *tbl;
294 	int ret;
295 
296 	tbl = kzalloc(sizeof(*tbl), GFP_NOFS);
297 	if (tbl == NULL)
298 		return -ENOMEM;
299 
300 	ret = nfs4_setup_slot_table(tbl, NFS4_MAX_SLOT_TABLE,
301 					"NFSv4.0 transport Slot table");
302 	if (ret) {
303 		kfree(tbl);
304 		return ret;
305 	}
306 
307 	clp->cl_slot_tbl = tbl;
308 	return 0;
309 }
310 
311 #if defined(CONFIG_NFS_V4_1)
312 
313 /**
314  * nfs41_init_client - nfs_client initialization tasks for NFSv4.1+
315  * @clp - nfs_client to initialize
316  *
317  * Returns zero on success, or a negative errno if some error occurred.
318  */
319 int nfs41_init_client(struct nfs_client *clp)
320 {
321 	struct nfs4_session *session = NULL;
322 
323 	/*
324 	 * Create the session and mark it expired.
325 	 * When a SEQUENCE operation encounters the expired session
326 	 * it will do session recovery to initialize it.
327 	 */
328 	session = nfs4_alloc_session(clp);
329 	if (!session)
330 		return -ENOMEM;
331 
332 	clp->cl_session = session;
333 
334 	/*
335 	 * The create session reply races with the server back
336 	 * channel probe. Mark the client NFS_CS_SESSION_INITING
337 	 * so that the client back channel can find the
338 	 * nfs_client struct
339 	 */
340 	nfs_mark_client_ready(clp, NFS_CS_SESSION_INITING);
341 	return 0;
342 }
343 
344 #endif	/* CONFIG_NFS_V4_1 */
345 
346 /*
347  * Initialize the minor version specific parts of an NFS4 client record
348  */
349 static int nfs4_init_client_minor_version(struct nfs_client *clp)
350 {
351 	int ret;
352 
353 	ret = clp->cl_mvops->init_client(clp);
354 	if (ret)
355 		return ret;
356 	return nfs4_init_callback(clp);
357 }
358 
359 /**
360  * nfs4_init_client - Initialise an NFS4 client record
361  *
362  * @clp: nfs_client to initialise
363  * @timeparms: timeout parameters for underlying RPC transport
364  * @ip_addr: callback IP address in presentation format
365  * @authflavor: authentication flavor for underlying RPC transport
366  *
367  * Returns pointer to an NFS client, or an ERR_PTR value.
368  */
369 struct nfs_client *nfs4_init_client(struct nfs_client *clp,
370 				    const struct nfs_client_initdata *cl_init)
371 {
372 	char buf[INET6_ADDRSTRLEN + 1];
373 	const char *ip_addr = cl_init->ip_addr;
374 	struct nfs_client *old;
375 	int error;
376 
377 	if (clp->cl_cons_state == NFS_CS_READY)
378 		/* the client is initialised already */
379 		return clp;
380 
381 	/* Check NFS protocol revision and initialize RPC op vector */
382 	clp->rpc_ops = &nfs_v4_clientops;
383 
384 	if (clp->cl_minorversion != 0)
385 		__set_bit(NFS_CS_INFINITE_SLOTS, &clp->cl_flags);
386 	__set_bit(NFS_CS_DISCRTRY, &clp->cl_flags);
387 	__set_bit(NFS_CS_NO_RETRANS_TIMEOUT, &clp->cl_flags);
388 
389 	error = nfs_create_rpc_client(clp, cl_init, RPC_AUTH_GSS_KRB5I);
390 	if (error == -EINVAL)
391 		error = nfs_create_rpc_client(clp, cl_init, RPC_AUTH_UNIX);
392 	if (error < 0)
393 		goto error;
394 
395 	/* If no clientaddr= option was specified, find a usable cb address */
396 	if (ip_addr == NULL) {
397 		struct sockaddr_storage cb_addr;
398 		struct sockaddr *sap = (struct sockaddr *)&cb_addr;
399 
400 		error = rpc_localaddr(clp->cl_rpcclient, sap, sizeof(cb_addr));
401 		if (error < 0)
402 			goto error;
403 		error = rpc_ntop(sap, buf, sizeof(buf));
404 		if (error < 0)
405 			goto error;
406 		ip_addr = (const char *)buf;
407 	}
408 	strlcpy(clp->cl_ipaddr, ip_addr, sizeof(clp->cl_ipaddr));
409 
410 	error = nfs_idmap_new(clp);
411 	if (error < 0) {
412 		dprintk("%s: failed to create idmapper. Error = %d\n",
413 			__func__, error);
414 		goto error;
415 	}
416 	__set_bit(NFS_CS_IDMAP, &clp->cl_res_state);
417 
418 	error = nfs4_init_client_minor_version(clp);
419 	if (error < 0)
420 		goto error;
421 
422 	error = nfs4_discover_server_trunking(clp, &old);
423 	if (error < 0)
424 		goto error;
425 
426 	if (clp != old) {
427 		clp->cl_preserve_clid = true;
428 		/*
429 		 * Mark the client as having failed initialization so other
430 		 * processes walking the nfs_client_list in nfs_match_client()
431 		 * won't try to use it.
432 		 */
433 		nfs_mark_client_ready(clp, -EPERM);
434 	}
435 	nfs_put_client(clp);
436 	clear_bit(NFS_CS_TSM_POSSIBLE, &clp->cl_flags);
437 	return old;
438 
439 error:
440 	nfs_mark_client_ready(clp, error);
441 	nfs_put_client(clp);
442 	return ERR_PTR(error);
443 }
444 
445 /*
446  * SETCLIENTID just did a callback update with the callback ident in
447  * "drop," but server trunking discovery claims "drop" and "keep" are
448  * actually the same server.  Swap the callback IDs so that "keep"
449  * will continue to use the callback ident the server now knows about,
450  * and so that "keep"'s original callback ident is destroyed when
451  * "drop" is freed.
452  */
453 static void nfs4_swap_callback_idents(struct nfs_client *keep,
454 				      struct nfs_client *drop)
455 {
456 	struct nfs_net *nn = net_generic(keep->cl_net, nfs_net_id);
457 	unsigned int save = keep->cl_cb_ident;
458 
459 	if (keep->cl_cb_ident == drop->cl_cb_ident)
460 		return;
461 
462 	dprintk("%s: keeping callback ident %u and dropping ident %u\n",
463 		__func__, keep->cl_cb_ident, drop->cl_cb_ident);
464 
465 	spin_lock(&nn->nfs_client_lock);
466 
467 	idr_replace(&nn->cb_ident_idr, keep, drop->cl_cb_ident);
468 	keep->cl_cb_ident = drop->cl_cb_ident;
469 
470 	idr_replace(&nn->cb_ident_idr, drop, save);
471 	drop->cl_cb_ident = save;
472 
473 	spin_unlock(&nn->nfs_client_lock);
474 }
475 
476 static bool nfs4_match_client_owner_id(const struct nfs_client *clp1,
477 		const struct nfs_client *clp2)
478 {
479 	if (clp1->cl_owner_id == NULL || clp2->cl_owner_id == NULL)
480 		return true;
481 	return strcmp(clp1->cl_owner_id, clp2->cl_owner_id) == 0;
482 }
483 
484 static bool nfs4_same_verifier(nfs4_verifier *v1, nfs4_verifier *v2)
485 {
486 	return memcmp(v1->data, v2->data, sizeof(v1->data)) == 0;
487 }
488 
489 static int nfs4_match_client(struct nfs_client  *pos,  struct nfs_client *new,
490 			     struct nfs_client **prev, struct nfs_net *nn)
491 {
492 	int status;
493 
494 	if (pos->rpc_ops != new->rpc_ops)
495 		return 1;
496 
497 	if (pos->cl_minorversion != new->cl_minorversion)
498 		return 1;
499 
500 	/* If "pos" isn't marked ready, we can't trust the
501 	 * remaining fields in "pos", especially the client
502 	 * ID and serverowner fields.  Wait for CREATE_SESSION
503 	 * to finish. */
504 	if (pos->cl_cons_state > NFS_CS_READY) {
505 		refcount_inc(&pos->cl_count);
506 		spin_unlock(&nn->nfs_client_lock);
507 
508 		nfs_put_client(*prev);
509 		*prev = pos;
510 
511 		status = nfs_wait_client_init_complete(pos);
512 		spin_lock(&nn->nfs_client_lock);
513 
514 		if (status < 0)
515 			return status;
516 	}
517 
518 	if (pos->cl_cons_state != NFS_CS_READY)
519 		return 1;
520 
521 	if (pos->cl_clientid != new->cl_clientid)
522 		return 1;
523 
524 	/* NFSv4.1 always uses the uniform string, however someone
525 	 * might switch the uniquifier string on us.
526 	 */
527 	if (!nfs4_match_client_owner_id(pos, new))
528 		return 1;
529 
530 	return 0;
531 }
532 
533 /**
534  * nfs40_walk_client_list - Find server that recognizes a client ID
535  *
536  * @new: nfs_client with client ID to test
537  * @result: OUT: found nfs_client, or new
538  * @cred: credential to use for trunking test
539  *
540  * Returns zero, a negative errno, or a negative NFS4ERR status.
541  * If zero is returned, an nfs_client pointer is planted in "result."
542  *
543  * NB: nfs40_walk_client_list() relies on the new nfs_client being
544  *     the last nfs_client on the list.
545  */
546 int nfs40_walk_client_list(struct nfs_client *new,
547 			   struct nfs_client **result,
548 			   const struct cred *cred)
549 {
550 	struct nfs_net *nn = net_generic(new->cl_net, nfs_net_id);
551 	struct nfs_client *pos, *prev = NULL;
552 	struct nfs4_setclientid_res clid = {
553 		.clientid	= new->cl_clientid,
554 		.confirm	= new->cl_confirm,
555 	};
556 	int status = -NFS4ERR_STALE_CLIENTID;
557 
558 	spin_lock(&nn->nfs_client_lock);
559 	list_for_each_entry(pos, &nn->nfs_client_list, cl_share_link) {
560 
561 		if (pos == new)
562 			goto found;
563 
564 		status = nfs4_match_client(pos, new, &prev, nn);
565 		if (status < 0)
566 			goto out_unlock;
567 		if (status != 0)
568 			continue;
569 		/*
570 		 * We just sent a new SETCLIENTID, which should have
571 		 * caused the server to return a new cl_confirm.  So if
572 		 * cl_confirm is the same, then this is a different
573 		 * server that just returned the same cl_confirm by
574 		 * coincidence:
575 		 */
576 		if ((new != pos) && nfs4_same_verifier(&pos->cl_confirm,
577 						       &new->cl_confirm))
578 			continue;
579 		/*
580 		 * But if the cl_confirm's are different, then the only
581 		 * way that a SETCLIENTID_CONFIRM to pos can succeed is
582 		 * if new and pos point to the same server:
583 		 */
584 found:
585 		refcount_inc(&pos->cl_count);
586 		spin_unlock(&nn->nfs_client_lock);
587 
588 		nfs_put_client(prev);
589 		prev = pos;
590 
591 		status = nfs4_proc_setclientid_confirm(pos, &clid, cred);
592 		switch (status) {
593 		case -NFS4ERR_STALE_CLIENTID:
594 			break;
595 		case 0:
596 			nfs4_swap_callback_idents(pos, new);
597 			pos->cl_confirm = new->cl_confirm;
598 			nfs_mark_client_ready(pos, NFS_CS_READY);
599 
600 			prev = NULL;
601 			*result = pos;
602 			goto out;
603 		case -ERESTARTSYS:
604 		case -ETIMEDOUT:
605 			/* The callback path may have been inadvertently
606 			 * changed. Schedule recovery!
607 			 */
608 			nfs4_schedule_path_down_recovery(pos);
609 		default:
610 			goto out;
611 		}
612 
613 		spin_lock(&nn->nfs_client_lock);
614 	}
615 out_unlock:
616 	spin_unlock(&nn->nfs_client_lock);
617 
618 	/* No match found. The server lost our clientid */
619 out:
620 	nfs_put_client(prev);
621 	return status;
622 }
623 
624 #ifdef CONFIG_NFS_V4_1
625 /*
626  * Returns true if the server major ids match
627  */
628 static bool
629 nfs4_check_serverowner_major_id(struct nfs41_server_owner *o1,
630 				struct nfs41_server_owner *o2)
631 {
632 	if (o1->major_id_sz != o2->major_id_sz)
633 		return false;
634 	return memcmp(o1->major_id, o2->major_id, o1->major_id_sz) == 0;
635 }
636 
637 /*
638  * Returns true if the server scopes match
639  */
640 static bool
641 nfs4_check_server_scope(struct nfs41_server_scope *s1,
642 			struct nfs41_server_scope *s2)
643 {
644 	if (s1->server_scope_sz != s2->server_scope_sz)
645 		return false;
646 	return memcmp(s1->server_scope, s2->server_scope,
647 					s1->server_scope_sz) == 0;
648 }
649 
650 /**
651  * nfs4_detect_session_trunking - Checks for session trunking.
652  *
653  * Called after a successful EXCHANGE_ID on a multi-addr connection.
654  * Upon success, add the transport.
655  *
656  * @clp:    original mount nfs_client
657  * @res:    result structure from an exchange_id using the original mount
658  *          nfs_client with a new multi_addr transport
659  *
660  * Returns zero on success, otherwise -EINVAL
661  *
662  * Note: since the exchange_id for the new multi_addr transport uses the
663  * same nfs_client from the original mount, the cl_owner_id is reused,
664  * so eir_clientowner is the same.
665  */
666 int nfs4_detect_session_trunking(struct nfs_client *clp,
667 				 struct nfs41_exchange_id_res *res,
668 				 struct rpc_xprt *xprt)
669 {
670 	/* Check eir_clientid */
671 	if (clp->cl_clientid != res->clientid)
672 		goto out_err;
673 
674 	/* Check eir_server_owner so_major_id */
675 	if (!nfs4_check_serverowner_major_id(clp->cl_serverowner,
676 					     res->server_owner))
677 		goto out_err;
678 
679 	/* Check eir_server_owner so_minor_id */
680 	if (clp->cl_serverowner->minor_id != res->server_owner->minor_id)
681 		goto out_err;
682 
683 	/* Check eir_server_scope */
684 	if (!nfs4_check_server_scope(clp->cl_serverscope, res->server_scope))
685 		goto out_err;
686 
687 	pr_info("NFS:  %s: Session trunking succeeded for %s\n",
688 		clp->cl_hostname,
689 		xprt->address_strings[RPC_DISPLAY_ADDR]);
690 
691 	return 0;
692 out_err:
693 	pr_info("NFS:  %s: Session trunking failed for %s\n", clp->cl_hostname,
694 		xprt->address_strings[RPC_DISPLAY_ADDR]);
695 
696 	return -EINVAL;
697 }
698 
699 /**
700  * nfs41_walk_client_list - Find nfs_client that matches a client/server owner
701  *
702  * @new: nfs_client with client ID to test
703  * @result: OUT: found nfs_client, or new
704  * @cred: credential to use for trunking test
705  *
706  * Returns zero, a negative errno, or a negative NFS4ERR status.
707  * If zero is returned, an nfs_client pointer is planted in "result."
708  *
709  * NB: nfs41_walk_client_list() relies on the new nfs_client being
710  *     the last nfs_client on the list.
711  */
712 int nfs41_walk_client_list(struct nfs_client *new,
713 			   struct nfs_client **result,
714 			   const struct cred *cred)
715 {
716 	struct nfs_net *nn = net_generic(new->cl_net, nfs_net_id);
717 	struct nfs_client *pos, *prev = NULL;
718 	int status = -NFS4ERR_STALE_CLIENTID;
719 
720 	spin_lock(&nn->nfs_client_lock);
721 	list_for_each_entry(pos, &nn->nfs_client_list, cl_share_link) {
722 
723 		if (pos == new)
724 			goto found;
725 
726 		status = nfs4_match_client(pos, new, &prev, nn);
727 		if (status < 0)
728 			goto out;
729 		if (status != 0)
730 			continue;
731 
732 		/*
733 		 * Note that session trunking is just a special subcase of
734 		 * client id trunking. In either case, we want to fall back
735 		 * to using the existing nfs_client.
736 		 */
737 		if (!nfs4_check_serverowner_major_id(pos->cl_serverowner,
738 						     new->cl_serverowner))
739 			continue;
740 
741 found:
742 		refcount_inc(&pos->cl_count);
743 		*result = pos;
744 		status = 0;
745 		break;
746 	}
747 
748 out:
749 	spin_unlock(&nn->nfs_client_lock);
750 	nfs_put_client(prev);
751 	return status;
752 }
753 #endif	/* CONFIG_NFS_V4_1 */
754 
755 static void nfs4_destroy_server(struct nfs_server *server)
756 {
757 	nfs_server_return_all_delegations(server);
758 	unset_pnfs_layoutdriver(server);
759 	nfs4_purge_state_owners(server);
760 }
761 
762 /*
763  * NFSv4.0 callback thread helper
764  *
765  * Find a client by callback identifier
766  */
767 struct nfs_client *
768 nfs4_find_client_ident(struct net *net, int cb_ident)
769 {
770 	struct nfs_client *clp;
771 	struct nfs_net *nn = net_generic(net, nfs_net_id);
772 
773 	spin_lock(&nn->nfs_client_lock);
774 	clp = idr_find(&nn->cb_ident_idr, cb_ident);
775 	if (clp)
776 		refcount_inc(&clp->cl_count);
777 	spin_unlock(&nn->nfs_client_lock);
778 	return clp;
779 }
780 
781 #if defined(CONFIG_NFS_V4_1)
782 /* Common match routine for v4.0 and v4.1 callback services */
783 static bool nfs4_cb_match_client(const struct sockaddr *addr,
784 		struct nfs_client *clp, u32 minorversion)
785 {
786 	struct sockaddr *clap = (struct sockaddr *)&clp->cl_addr;
787 
788 	/* Don't match clients that failed to initialise */
789 	if (!(clp->cl_cons_state == NFS_CS_READY ||
790 	    clp->cl_cons_state == NFS_CS_SESSION_INITING))
791 		return false;
792 
793 	smp_rmb();
794 
795 	/* Match the version and minorversion */
796 	if (clp->rpc_ops->version != 4 ||
797 	    clp->cl_minorversion != minorversion)
798 		return false;
799 
800 	/* Match only the IP address, not the port number */
801 	return rpc_cmp_addr(addr, clap);
802 }
803 
804 /*
805  * NFSv4.1 callback thread helper
806  * For CB_COMPOUND calls, find a client by IP address, protocol version,
807  * minorversion, and sessionID
808  *
809  * Returns NULL if no such client
810  */
811 struct nfs_client *
812 nfs4_find_client_sessionid(struct net *net, const struct sockaddr *addr,
813 			   struct nfs4_sessionid *sid, u32 minorversion)
814 {
815 	struct nfs_client *clp;
816 	struct nfs_net *nn = net_generic(net, nfs_net_id);
817 
818 	spin_lock(&nn->nfs_client_lock);
819 	list_for_each_entry(clp, &nn->nfs_client_list, cl_share_link) {
820 		if (!nfs4_cb_match_client(addr, clp, minorversion))
821 			continue;
822 
823 		if (!nfs4_has_session(clp))
824 			continue;
825 
826 		/* Match sessionid*/
827 		if (memcmp(clp->cl_session->sess_id.data,
828 		    sid->data, NFS4_MAX_SESSIONID_LEN) != 0)
829 			continue;
830 
831 		refcount_inc(&clp->cl_count);
832 		spin_unlock(&nn->nfs_client_lock);
833 		return clp;
834 	}
835 	spin_unlock(&nn->nfs_client_lock);
836 	return NULL;
837 }
838 
839 #else /* CONFIG_NFS_V4_1 */
840 
841 struct nfs_client *
842 nfs4_find_client_sessionid(struct net *net, const struct sockaddr *addr,
843 			   struct nfs4_sessionid *sid, u32 minorversion)
844 {
845 	return NULL;
846 }
847 #endif /* CONFIG_NFS_V4_1 */
848 
849 /*
850  * Set up an NFS4 client
851  */
852 static int nfs4_set_client(struct nfs_server *server,
853 		const char *hostname,
854 		const struct sockaddr *addr,
855 		const size_t addrlen,
856 		const char *ip_addr,
857 		int proto, const struct rpc_timeout *timeparms,
858 		u32 minorversion, struct net *net)
859 {
860 	struct nfs_client_initdata cl_init = {
861 		.hostname = hostname,
862 		.addr = addr,
863 		.addrlen = addrlen,
864 		.ip_addr = ip_addr,
865 		.nfs_mod = &nfs_v4,
866 		.proto = proto,
867 		.minorversion = minorversion,
868 		.net = net,
869 		.timeparms = timeparms,
870 	};
871 	struct nfs_client *clp;
872 
873 	if (server->flags & NFS_MOUNT_NORESVPORT)
874 		set_bit(NFS_CS_NORESVPORT, &cl_init.init_flags);
875 	if (server->options & NFS_OPTION_MIGRATION)
876 		set_bit(NFS_CS_MIGRATION, &cl_init.init_flags);
877 	if (test_bit(NFS_MIG_TSM_POSSIBLE, &server->mig_status))
878 		set_bit(NFS_CS_TSM_POSSIBLE, &cl_init.init_flags);
879 	server->port = rpc_get_port(addr);
880 
881 	/* Allocate or find a client reference we can use */
882 	clp = nfs_get_client(&cl_init);
883 	if (IS_ERR(clp))
884 		return PTR_ERR(clp);
885 
886 	if (server->nfs_client == clp) {
887 		nfs_put_client(clp);
888 		return -ELOOP;
889 	}
890 
891 	/*
892 	 * Query for the lease time on clientid setup or renewal
893 	 *
894 	 * Note that this will be set on nfs_clients that were created
895 	 * only for the DS role and did not set this bit, but now will
896 	 * serve a dual role.
897 	 */
898 	set_bit(NFS_CS_CHECK_LEASE_TIME, &clp->cl_res_state);
899 
900 	server->nfs_client = clp;
901 	return 0;
902 }
903 
904 /*
905  * Set up a pNFS Data Server client.
906  *
907  * Return any existing nfs_client that matches server address,port,version
908  * and minorversion.
909  *
910  * For a new nfs_client, use a soft mount (default), a low retrans and a
911  * low timeout interval so that if a connection is lost, we retry through
912  * the MDS.
913  */
914 struct nfs_client *nfs4_set_ds_client(struct nfs_server *mds_srv,
915 		const struct sockaddr *ds_addr, int ds_addrlen,
916 		int ds_proto, unsigned int ds_timeo, unsigned int ds_retrans,
917 		u32 minor_version)
918 {
919 	struct rpc_timeout ds_timeout;
920 	struct nfs_client *mds_clp = mds_srv->nfs_client;
921 	struct nfs_client_initdata cl_init = {
922 		.addr = ds_addr,
923 		.addrlen = ds_addrlen,
924 		.nodename = mds_clp->cl_rpcclient->cl_nodename,
925 		.ip_addr = mds_clp->cl_ipaddr,
926 		.nfs_mod = &nfs_v4,
927 		.proto = ds_proto,
928 		.minorversion = minor_version,
929 		.net = mds_clp->cl_net,
930 		.timeparms = &ds_timeout,
931 	};
932 	char buf[INET6_ADDRSTRLEN + 1];
933 
934 	if (rpc_ntop(ds_addr, buf, sizeof(buf)) <= 0)
935 		return ERR_PTR(-EINVAL);
936 	cl_init.hostname = buf;
937 
938 	if (mds_srv->flags & NFS_MOUNT_NORESVPORT)
939 		__set_bit(NFS_CS_NORESVPORT, &cl_init.init_flags);
940 
941 	/*
942 	 * Set an authflavor equual to the MDS value. Use the MDS nfs_client
943 	 * cl_ipaddr so as to use the same EXCHANGE_ID co_ownerid as the MDS
944 	 * (section 13.1 RFC 5661).
945 	 */
946 	nfs_init_timeout_values(&ds_timeout, ds_proto, ds_timeo, ds_retrans);
947 	return nfs_get_client(&cl_init);
948 }
949 EXPORT_SYMBOL_GPL(nfs4_set_ds_client);
950 
951 /*
952  * Session has been established, and the client marked ready.
953  * Limit the mount rsize, wsize and dtsize using negotiated fore
954  * channel attributes.
955  */
956 static void nfs4_session_limit_rwsize(struct nfs_server *server)
957 {
958 #ifdef CONFIG_NFS_V4_1
959 	struct nfs4_session *sess;
960 	u32 server_resp_sz;
961 	u32 server_rqst_sz;
962 
963 	if (!nfs4_has_session(server->nfs_client))
964 		return;
965 	sess = server->nfs_client->cl_session;
966 	server_resp_sz = sess->fc_attrs.max_resp_sz - nfs41_maxread_overhead;
967 	server_rqst_sz = sess->fc_attrs.max_rqst_sz - nfs41_maxwrite_overhead;
968 
969 	if (server->dtsize > server_resp_sz)
970 		server->dtsize = server_resp_sz;
971 	if (server->rsize > server_resp_sz)
972 		server->rsize = server_resp_sz;
973 	if (server->wsize > server_rqst_sz)
974 		server->wsize = server_rqst_sz;
975 #endif /* CONFIG_NFS_V4_1 */
976 }
977 
978 static int nfs4_server_common_setup(struct nfs_server *server,
979 		struct nfs_fh *mntfh, bool auth_probe)
980 {
981 	struct nfs_fattr *fattr;
982 	int error;
983 
984 	/* data servers support only a subset of NFSv4.1 */
985 	if (is_ds_only_client(server->nfs_client))
986 		return -EPROTONOSUPPORT;
987 
988 	fattr = nfs_alloc_fattr();
989 	if (fattr == NULL)
990 		return -ENOMEM;
991 
992 	/* We must ensure the session is initialised first */
993 	error = nfs4_init_session(server->nfs_client);
994 	if (error < 0)
995 		goto out;
996 
997 	/* Set the basic capabilities */
998 	server->caps |= server->nfs_client->cl_mvops->init_caps;
999 	if (server->flags & NFS_MOUNT_NORDIRPLUS)
1000 			server->caps &= ~NFS_CAP_READDIRPLUS;
1001 	/*
1002 	 * Don't use NFS uid/gid mapping if we're using AUTH_SYS or lower
1003 	 * authentication.
1004 	 */
1005 	if (nfs4_disable_idmapping &&
1006 			server->client->cl_auth->au_flavor == RPC_AUTH_UNIX)
1007 		server->caps |= NFS_CAP_UIDGID_NOMAP;
1008 
1009 
1010 	/* Probe the root fh to retrieve its FSID and filehandle */
1011 	error = nfs4_get_rootfh(server, mntfh, auth_probe);
1012 	if (error < 0)
1013 		goto out;
1014 
1015 	dprintk("Server FSID: %llx:%llx\n",
1016 			(unsigned long long) server->fsid.major,
1017 			(unsigned long long) server->fsid.minor);
1018 	nfs_display_fhandle(mntfh, "Pseudo-fs root FH");
1019 
1020 	error = nfs_probe_fsinfo(server, mntfh, fattr);
1021 	if (error < 0)
1022 		goto out;
1023 
1024 	nfs4_session_limit_rwsize(server);
1025 
1026 	if (server->namelen == 0 || server->namelen > NFS4_MAXNAMLEN)
1027 		server->namelen = NFS4_MAXNAMLEN;
1028 
1029 	nfs_server_insert_lists(server);
1030 	server->mount_time = jiffies;
1031 	server->destroy = nfs4_destroy_server;
1032 out:
1033 	nfs_free_fattr(fattr);
1034 	return error;
1035 }
1036 
1037 /*
1038  * Create a version 4 volume record
1039  */
1040 static int nfs4_init_server(struct nfs_server *server,
1041 		struct nfs_parsed_mount_data *data)
1042 {
1043 	struct rpc_timeout timeparms;
1044 	int error;
1045 
1046 	nfs_init_timeout_values(&timeparms, data->nfs_server.protocol,
1047 			data->timeo, data->retrans);
1048 
1049 	/* Initialise the client representation from the mount data */
1050 	server->flags = data->flags;
1051 	server->options = data->options;
1052 	server->auth_info = data->auth_info;
1053 
1054 	/* Use the first specified auth flavor. If this flavor isn't
1055 	 * allowed by the server, use the SECINFO path to try the
1056 	 * other specified flavors */
1057 	if (data->auth_info.flavor_len >= 1)
1058 		data->selected_flavor = data->auth_info.flavors[0];
1059 	else
1060 		data->selected_flavor = RPC_AUTH_UNIX;
1061 
1062 	/* Get a client record */
1063 	error = nfs4_set_client(server,
1064 			data->nfs_server.hostname,
1065 			(const struct sockaddr *)&data->nfs_server.address,
1066 			data->nfs_server.addrlen,
1067 			data->client_address,
1068 			data->nfs_server.protocol,
1069 			&timeparms,
1070 			data->minorversion,
1071 			data->net);
1072 	if (error < 0)
1073 		return error;
1074 
1075 	if (data->rsize)
1076 		server->rsize = nfs_block_size(data->rsize, NULL);
1077 	if (data->wsize)
1078 		server->wsize = nfs_block_size(data->wsize, NULL);
1079 
1080 	server->acregmin = data->acregmin * HZ;
1081 	server->acregmax = data->acregmax * HZ;
1082 	server->acdirmin = data->acdirmin * HZ;
1083 	server->acdirmax = data->acdirmax * HZ;
1084 	server->port     = data->nfs_server.port;
1085 
1086 	return nfs_init_server_rpcclient(server, &timeparms,
1087 					 data->selected_flavor);
1088 }
1089 
1090 /*
1091  * Create a version 4 volume record
1092  * - keyed on server and FSID
1093  */
1094 /*struct nfs_server *nfs4_create_server(const struct nfs_parsed_mount_data *data,
1095 				      struct nfs_fh *mntfh)*/
1096 struct nfs_server *nfs4_create_server(struct nfs_mount_info *mount_info,
1097 				      struct nfs_subversion *nfs_mod)
1098 {
1099 	struct nfs_server *server;
1100 	bool auth_probe;
1101 	int error;
1102 
1103 	server = nfs_alloc_server();
1104 	if (!server)
1105 		return ERR_PTR(-ENOMEM);
1106 
1107 	auth_probe = mount_info->parsed->auth_info.flavor_len < 1;
1108 
1109 	/* set up the general RPC client */
1110 	error = nfs4_init_server(server, mount_info->parsed);
1111 	if (error < 0)
1112 		goto error;
1113 
1114 	error = nfs4_server_common_setup(server, mount_info->mntfh, auth_probe);
1115 	if (error < 0)
1116 		goto error;
1117 
1118 	return server;
1119 
1120 error:
1121 	nfs_free_server(server);
1122 	return ERR_PTR(error);
1123 }
1124 
1125 /*
1126  * Create an NFS4 referral server record
1127  */
1128 struct nfs_server *nfs4_create_referral_server(struct nfs_clone_mount *data,
1129 					       struct nfs_fh *mntfh)
1130 {
1131 	struct nfs_client *parent_client;
1132 	struct nfs_server *server, *parent_server;
1133 	bool auth_probe;
1134 	int error;
1135 
1136 	server = nfs_alloc_server();
1137 	if (!server)
1138 		return ERR_PTR(-ENOMEM);
1139 
1140 	parent_server = NFS_SB(data->sb);
1141 	parent_client = parent_server->nfs_client;
1142 
1143 	/* Initialise the client representation from the parent server */
1144 	nfs_server_copy_userdata(server, parent_server);
1145 
1146 	/* Get a client representation */
1147 #if IS_ENABLED(CONFIG_SUNRPC_XPRT_RDMA)
1148 	rpc_set_port(data->addr, NFS_RDMA_PORT);
1149 	error = nfs4_set_client(server, data->hostname,
1150 				data->addr,
1151 				data->addrlen,
1152 				parent_client->cl_ipaddr,
1153 				XPRT_TRANSPORT_RDMA,
1154 				parent_server->client->cl_timeout,
1155 				parent_client->cl_mvops->minor_version,
1156 				parent_client->cl_net);
1157 	if (!error)
1158 		goto init_server;
1159 #endif	/* IS_ENABLED(CONFIG_SUNRPC_XPRT_RDMA) */
1160 
1161 	rpc_set_port(data->addr, NFS_PORT);
1162 	error = nfs4_set_client(server, data->hostname,
1163 				data->addr,
1164 				data->addrlen,
1165 				parent_client->cl_ipaddr,
1166 				XPRT_TRANSPORT_TCP,
1167 				parent_server->client->cl_timeout,
1168 				parent_client->cl_mvops->minor_version,
1169 				parent_client->cl_net);
1170 	if (error < 0)
1171 		goto error;
1172 
1173 #if IS_ENABLED(CONFIG_SUNRPC_XPRT_RDMA)
1174 init_server:
1175 #endif
1176 	error = nfs_init_server_rpcclient(server, parent_server->client->cl_timeout, data->authflavor);
1177 	if (error < 0)
1178 		goto error;
1179 
1180 	auth_probe = parent_server->auth_info.flavor_len < 1;
1181 
1182 	error = nfs4_server_common_setup(server, mntfh, auth_probe);
1183 	if (error < 0)
1184 		goto error;
1185 
1186 	return server;
1187 
1188 error:
1189 	nfs_free_server(server);
1190 	return ERR_PTR(error);
1191 }
1192 
1193 /*
1194  * Grab the destination's particulars, including lease expiry time.
1195  *
1196  * Returns zero if probe succeeded and retrieved FSID matches the FSID
1197  * we have cached.
1198  */
1199 static int nfs_probe_destination(struct nfs_server *server)
1200 {
1201 	struct inode *inode = d_inode(server->super->s_root);
1202 	struct nfs_fattr *fattr;
1203 	int error;
1204 
1205 	fattr = nfs_alloc_fattr();
1206 	if (fattr == NULL)
1207 		return -ENOMEM;
1208 
1209 	/* Sanity: the probe won't work if the destination server
1210 	 * does not recognize the migrated FH. */
1211 	error = nfs_probe_fsinfo(server, NFS_FH(inode), fattr);
1212 
1213 	nfs_free_fattr(fattr);
1214 	return error;
1215 }
1216 
1217 /**
1218  * nfs4_update_server - Move an nfs_server to a different nfs_client
1219  *
1220  * @server: represents FSID to be moved
1221  * @hostname: new end-point's hostname
1222  * @sap: new end-point's socket address
1223  * @salen: size of "sap"
1224  * @net: net namespace
1225  *
1226  * The nfs_server must be quiescent before this function is invoked.
1227  * Either its session is drained (NFSv4.1+), or its transport is
1228  * plugged and drained (NFSv4.0).
1229  *
1230  * Returns zero on success, or a negative errno value.
1231  */
1232 int nfs4_update_server(struct nfs_server *server, const char *hostname,
1233 		       struct sockaddr *sap, size_t salen, struct net *net)
1234 {
1235 	struct nfs_client *clp = server->nfs_client;
1236 	struct rpc_clnt *clnt = server->client;
1237 	struct xprt_create xargs = {
1238 		.ident		= clp->cl_proto,
1239 		.net		= net,
1240 		.dstaddr	= sap,
1241 		.addrlen	= salen,
1242 		.servername	= hostname,
1243 	};
1244 	char buf[INET6_ADDRSTRLEN + 1];
1245 	struct sockaddr_storage address;
1246 	struct sockaddr *localaddr = (struct sockaddr *)&address;
1247 	int error;
1248 
1249 	error = rpc_switch_client_transport(clnt, &xargs, clnt->cl_timeout);
1250 	if (error != 0)
1251 		return error;
1252 
1253 	error = rpc_localaddr(clnt, localaddr, sizeof(address));
1254 	if (error != 0)
1255 		return error;
1256 
1257 	if (rpc_ntop(localaddr, buf, sizeof(buf)) == 0)
1258 		return -EAFNOSUPPORT;
1259 
1260 	nfs_server_remove_lists(server);
1261 	set_bit(NFS_MIG_TSM_POSSIBLE, &server->mig_status);
1262 	error = nfs4_set_client(server, hostname, sap, salen, buf,
1263 				clp->cl_proto, clnt->cl_timeout,
1264 				clp->cl_minorversion, net);
1265 	clear_bit(NFS_MIG_TSM_POSSIBLE, &server->mig_status);
1266 	if (error != 0) {
1267 		nfs_server_insert_lists(server);
1268 		return error;
1269 	}
1270 	nfs_put_client(clp);
1271 
1272 	if (server->nfs_client->cl_hostname == NULL)
1273 		server->nfs_client->cl_hostname = kstrdup(hostname, GFP_KERNEL);
1274 	nfs_server_insert_lists(server);
1275 
1276 	return nfs_probe_destination(server);
1277 }
1278