nfs4state.c revision f7104cc1
1/*
2*  Copyright (c) 2001 The Regents of the University of Michigan.
3*  All rights reserved.
4*
5*  Kendrick Smith <kmsmith@umich.edu>
6*  Andy Adamson <kandros@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 <linux/file.h>
36#include <linux/fs.h>
37#include <linux/slab.h>
38#include <linux/namei.h>
39#include <linux/swap.h>
40#include <linux/pagemap.h>
41#include <linux/ratelimit.h>
42#include <linux/sunrpc/svcauth_gss.h>
43#include <linux/sunrpc/addr.h>
44#include <linux/jhash.h>
45#include <linux/string_helpers.h>
46#include <linux/fsnotify.h>
47#include <linux/nfs_ssc.h>
48#include "xdr4.h"
49#include "xdr4cb.h"
50#include "vfs.h"
51#include "current_stateid.h"
52
53#include "netns.h"
54#include "pnfs.h"
55#include "filecache.h"
56#include "trace.h"
57
58#define NFSDDBG_FACILITY                NFSDDBG_PROC
59
60#define all_ones {{~0,~0},~0}
61static const stateid_t one_stateid = {
62	.si_generation = ~0,
63	.si_opaque = all_ones,
64};
65static const stateid_t zero_stateid = {
66	/* all fields zero */
67};
68static const stateid_t currentstateid = {
69	.si_generation = 1,
70};
71static const stateid_t close_stateid = {
72	.si_generation = 0xffffffffU,
73};
74
75static u64 current_sessionid = 1;
76
77#define ZERO_STATEID(stateid) (!memcmp((stateid), &zero_stateid, sizeof(stateid_t)))
78#define ONE_STATEID(stateid)  (!memcmp((stateid), &one_stateid, sizeof(stateid_t)))
79#define CURRENT_STATEID(stateid) (!memcmp((stateid), &currentstateid, sizeof(stateid_t)))
80#define CLOSE_STATEID(stateid)  (!memcmp((stateid), &close_stateid, sizeof(stateid_t)))
81
82/* forward declarations */
83static bool check_for_locks(struct nfs4_file *fp, struct nfs4_lockowner *lowner);
84static void nfs4_free_ol_stateid(struct nfs4_stid *stid);
85void nfsd4_end_grace(struct nfsd_net *nn);
86static void _free_cpntf_state_locked(struct nfsd_net *nn, struct nfs4_cpntf_state *cps);
87
88/* Locking: */
89
90/*
91 * Currently used for the del_recall_lru and file hash table.  In an
92 * effort to decrease the scope of the client_mutex, this spinlock may
93 * eventually cover more:
94 */
95static DEFINE_SPINLOCK(state_lock);
96
97enum nfsd4_st_mutex_lock_subclass {
98	OPEN_STATEID_MUTEX = 0,
99	LOCK_STATEID_MUTEX = 1,
100};
101
102/*
103 * A waitqueue for all in-progress 4.0 CLOSE operations that are waiting for
104 * the refcount on the open stateid to drop.
105 */
106static DECLARE_WAIT_QUEUE_HEAD(close_wq);
107
108/*
109 * A waitqueue where a writer to clients/#/ctl destroying a client can
110 * wait for cl_rpc_users to drop to 0 and then for the client to be
111 * unhashed.
112 */
113static DECLARE_WAIT_QUEUE_HEAD(expiry_wq);
114
115static struct kmem_cache *client_slab;
116static struct kmem_cache *openowner_slab;
117static struct kmem_cache *lockowner_slab;
118static struct kmem_cache *file_slab;
119static struct kmem_cache *stateid_slab;
120static struct kmem_cache *deleg_slab;
121static struct kmem_cache *odstate_slab;
122
123static void free_session(struct nfsd4_session *);
124
125static const struct nfsd4_callback_ops nfsd4_cb_recall_ops;
126static const struct nfsd4_callback_ops nfsd4_cb_notify_lock_ops;
127
128static bool is_session_dead(struct nfsd4_session *ses)
129{
130	return ses->se_flags & NFS4_SESSION_DEAD;
131}
132
133static __be32 mark_session_dead_locked(struct nfsd4_session *ses, int ref_held_by_me)
134{
135	if (atomic_read(&ses->se_ref) > ref_held_by_me)
136		return nfserr_jukebox;
137	ses->se_flags |= NFS4_SESSION_DEAD;
138	return nfs_ok;
139}
140
141static bool is_client_expired(struct nfs4_client *clp)
142{
143	return clp->cl_time == 0;
144}
145
146static __be32 get_client_locked(struct nfs4_client *clp)
147{
148	struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
149
150	lockdep_assert_held(&nn->client_lock);
151
152	if (is_client_expired(clp))
153		return nfserr_expired;
154	atomic_inc(&clp->cl_rpc_users);
155	return nfs_ok;
156}
157
158/* must be called under the client_lock */
159static inline void
160renew_client_locked(struct nfs4_client *clp)
161{
162	struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
163
164	if (is_client_expired(clp)) {
165		WARN_ON(1);
166		printk("%s: client (clientid %08x/%08x) already expired\n",
167			__func__,
168			clp->cl_clientid.cl_boot,
169			clp->cl_clientid.cl_id);
170		return;
171	}
172
173	list_move_tail(&clp->cl_lru, &nn->client_lru);
174	clp->cl_time = ktime_get_boottime_seconds();
175}
176
177static void put_client_renew_locked(struct nfs4_client *clp)
178{
179	struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
180
181	lockdep_assert_held(&nn->client_lock);
182
183	if (!atomic_dec_and_test(&clp->cl_rpc_users))
184		return;
185	if (!is_client_expired(clp))
186		renew_client_locked(clp);
187	else
188		wake_up_all(&expiry_wq);
189}
190
191static void put_client_renew(struct nfs4_client *clp)
192{
193	struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
194
195	if (!atomic_dec_and_lock(&clp->cl_rpc_users, &nn->client_lock))
196		return;
197	if (!is_client_expired(clp))
198		renew_client_locked(clp);
199	else
200		wake_up_all(&expiry_wq);
201	spin_unlock(&nn->client_lock);
202}
203
204static __be32 nfsd4_get_session_locked(struct nfsd4_session *ses)
205{
206	__be32 status;
207
208	if (is_session_dead(ses))
209		return nfserr_badsession;
210	status = get_client_locked(ses->se_client);
211	if (status)
212		return status;
213	atomic_inc(&ses->se_ref);
214	return nfs_ok;
215}
216
217static void nfsd4_put_session_locked(struct nfsd4_session *ses)
218{
219	struct nfs4_client *clp = ses->se_client;
220	struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
221
222	lockdep_assert_held(&nn->client_lock);
223
224	if (atomic_dec_and_test(&ses->se_ref) && is_session_dead(ses))
225		free_session(ses);
226	put_client_renew_locked(clp);
227}
228
229static void nfsd4_put_session(struct nfsd4_session *ses)
230{
231	struct nfs4_client *clp = ses->se_client;
232	struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
233
234	spin_lock(&nn->client_lock);
235	nfsd4_put_session_locked(ses);
236	spin_unlock(&nn->client_lock);
237}
238
239static struct nfsd4_blocked_lock *
240find_blocked_lock(struct nfs4_lockowner *lo, struct knfsd_fh *fh,
241			struct nfsd_net *nn)
242{
243	struct nfsd4_blocked_lock *cur, *found = NULL;
244
245	spin_lock(&nn->blocked_locks_lock);
246	list_for_each_entry(cur, &lo->lo_blocked, nbl_list) {
247		if (fh_match(fh, &cur->nbl_fh)) {
248			list_del_init(&cur->nbl_list);
249			list_del_init(&cur->nbl_lru);
250			found = cur;
251			break;
252		}
253	}
254	spin_unlock(&nn->blocked_locks_lock);
255	if (found)
256		locks_delete_block(&found->nbl_lock);
257	return found;
258}
259
260static struct nfsd4_blocked_lock *
261find_or_allocate_block(struct nfs4_lockowner *lo, struct knfsd_fh *fh,
262			struct nfsd_net *nn)
263{
264	struct nfsd4_blocked_lock *nbl;
265
266	nbl = find_blocked_lock(lo, fh, nn);
267	if (!nbl) {
268		nbl= kmalloc(sizeof(*nbl), GFP_KERNEL);
269		if (nbl) {
270			INIT_LIST_HEAD(&nbl->nbl_list);
271			INIT_LIST_HEAD(&nbl->nbl_lru);
272			fh_copy_shallow(&nbl->nbl_fh, fh);
273			locks_init_lock(&nbl->nbl_lock);
274			nfsd4_init_cb(&nbl->nbl_cb, lo->lo_owner.so_client,
275					&nfsd4_cb_notify_lock_ops,
276					NFSPROC4_CLNT_CB_NOTIFY_LOCK);
277		}
278	}
279	return nbl;
280}
281
282static void
283free_blocked_lock(struct nfsd4_blocked_lock *nbl)
284{
285	locks_delete_block(&nbl->nbl_lock);
286	locks_release_private(&nbl->nbl_lock);
287	kfree(nbl);
288}
289
290static void
291remove_blocked_locks(struct nfs4_lockowner *lo)
292{
293	struct nfs4_client *clp = lo->lo_owner.so_client;
294	struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
295	struct nfsd4_blocked_lock *nbl;
296	LIST_HEAD(reaplist);
297
298	/* Dequeue all blocked locks */
299	spin_lock(&nn->blocked_locks_lock);
300	while (!list_empty(&lo->lo_blocked)) {
301		nbl = list_first_entry(&lo->lo_blocked,
302					struct nfsd4_blocked_lock,
303					nbl_list);
304		list_del_init(&nbl->nbl_list);
305		list_move(&nbl->nbl_lru, &reaplist);
306	}
307	spin_unlock(&nn->blocked_locks_lock);
308
309	/* Now free them */
310	while (!list_empty(&reaplist)) {
311		nbl = list_first_entry(&reaplist, struct nfsd4_blocked_lock,
312					nbl_lru);
313		list_del_init(&nbl->nbl_lru);
314		free_blocked_lock(nbl);
315	}
316}
317
318static void
319nfsd4_cb_notify_lock_prepare(struct nfsd4_callback *cb)
320{
321	struct nfsd4_blocked_lock	*nbl = container_of(cb,
322						struct nfsd4_blocked_lock, nbl_cb);
323	locks_delete_block(&nbl->nbl_lock);
324}
325
326static int
327nfsd4_cb_notify_lock_done(struct nfsd4_callback *cb, struct rpc_task *task)
328{
329	/*
330	 * Since this is just an optimization, we don't try very hard if it
331	 * turns out not to succeed. We'll requeue it on NFS4ERR_DELAY, and
332	 * just quit trying on anything else.
333	 */
334	switch (task->tk_status) {
335	case -NFS4ERR_DELAY:
336		rpc_delay(task, 1 * HZ);
337		return 0;
338	default:
339		return 1;
340	}
341}
342
343static void
344nfsd4_cb_notify_lock_release(struct nfsd4_callback *cb)
345{
346	struct nfsd4_blocked_lock	*nbl = container_of(cb,
347						struct nfsd4_blocked_lock, nbl_cb);
348
349	free_blocked_lock(nbl);
350}
351
352static const struct nfsd4_callback_ops nfsd4_cb_notify_lock_ops = {
353	.prepare	= nfsd4_cb_notify_lock_prepare,
354	.done		= nfsd4_cb_notify_lock_done,
355	.release	= nfsd4_cb_notify_lock_release,
356};
357
358/*
359 * We store the NONE, READ, WRITE, and BOTH bits separately in the
360 * st_{access,deny}_bmap field of the stateid, in order to track not
361 * only what share bits are currently in force, but also what
362 * combinations of share bits previous opens have used.  This allows us
363 * to enforce the recommendation of rfc 3530 14.2.19 that the server
364 * return an error if the client attempt to downgrade to a combination
365 * of share bits not explicable by closing some of its previous opens.
366 *
367 * XXX: This enforcement is actually incomplete, since we don't keep
368 * track of access/deny bit combinations; so, e.g., we allow:
369 *
370 *	OPEN allow read, deny write
371 *	OPEN allow both, deny none
372 *	DOWNGRADE allow read, deny none
373 *
374 * which we should reject.
375 */
376static unsigned int
377bmap_to_share_mode(unsigned long bmap)
378{
379	int i;
380	unsigned int access = 0;
381
382	for (i = 1; i < 4; i++) {
383		if (test_bit(i, &bmap))
384			access |= i;
385	}
386	return access;
387}
388
389/* set share access for a given stateid */
390static inline void
391set_access(u32 access, struct nfs4_ol_stateid *stp)
392{
393	unsigned char mask = 1 << access;
394
395	WARN_ON_ONCE(access > NFS4_SHARE_ACCESS_BOTH);
396	stp->st_access_bmap |= mask;
397}
398
399/* clear share access for a given stateid */
400static inline void
401clear_access(u32 access, struct nfs4_ol_stateid *stp)
402{
403	unsigned char mask = 1 << access;
404
405	WARN_ON_ONCE(access > NFS4_SHARE_ACCESS_BOTH);
406	stp->st_access_bmap &= ~mask;
407}
408
409/* test whether a given stateid has access */
410static inline bool
411test_access(u32 access, struct nfs4_ol_stateid *stp)
412{
413	unsigned char mask = 1 << access;
414
415	return (bool)(stp->st_access_bmap & mask);
416}
417
418/* set share deny for a given stateid */
419static inline void
420set_deny(u32 deny, struct nfs4_ol_stateid *stp)
421{
422	unsigned char mask = 1 << deny;
423
424	WARN_ON_ONCE(deny > NFS4_SHARE_DENY_BOTH);
425	stp->st_deny_bmap |= mask;
426}
427
428/* clear share deny for a given stateid */
429static inline void
430clear_deny(u32 deny, struct nfs4_ol_stateid *stp)
431{
432	unsigned char mask = 1 << deny;
433
434	WARN_ON_ONCE(deny > NFS4_SHARE_DENY_BOTH);
435	stp->st_deny_bmap &= ~mask;
436}
437
438/* test whether a given stateid is denying specific access */
439static inline bool
440test_deny(u32 deny, struct nfs4_ol_stateid *stp)
441{
442	unsigned char mask = 1 << deny;
443
444	return (bool)(stp->st_deny_bmap & mask);
445}
446
447static int nfs4_access_to_omode(u32 access)
448{
449	switch (access & NFS4_SHARE_ACCESS_BOTH) {
450	case NFS4_SHARE_ACCESS_READ:
451		return O_RDONLY;
452	case NFS4_SHARE_ACCESS_WRITE:
453		return O_WRONLY;
454	case NFS4_SHARE_ACCESS_BOTH:
455		return O_RDWR;
456	}
457	WARN_ON_ONCE(1);
458	return O_RDONLY;
459}
460
461static inline int
462access_permit_read(struct nfs4_ol_stateid *stp)
463{
464	return test_access(NFS4_SHARE_ACCESS_READ, stp) ||
465		test_access(NFS4_SHARE_ACCESS_BOTH, stp) ||
466		test_access(NFS4_SHARE_ACCESS_WRITE, stp);
467}
468
469static inline int
470access_permit_write(struct nfs4_ol_stateid *stp)
471{
472	return test_access(NFS4_SHARE_ACCESS_WRITE, stp) ||
473		test_access(NFS4_SHARE_ACCESS_BOTH, stp);
474}
475
476static inline struct nfs4_stateowner *
477nfs4_get_stateowner(struct nfs4_stateowner *sop)
478{
479	atomic_inc(&sop->so_count);
480	return sop;
481}
482
483static int
484same_owner_str(struct nfs4_stateowner *sop, struct xdr_netobj *owner)
485{
486	return (sop->so_owner.len == owner->len) &&
487		0 == memcmp(sop->so_owner.data, owner->data, owner->len);
488}
489
490static struct nfs4_openowner *
491find_openstateowner_str_locked(unsigned int hashval, struct nfsd4_open *open,
492			struct nfs4_client *clp)
493{
494	struct nfs4_stateowner *so;
495
496	lockdep_assert_held(&clp->cl_lock);
497
498	list_for_each_entry(so, &clp->cl_ownerstr_hashtbl[hashval],
499			    so_strhash) {
500		if (!so->so_is_open_owner)
501			continue;
502		if (same_owner_str(so, &open->op_owner))
503			return openowner(nfs4_get_stateowner(so));
504	}
505	return NULL;
506}
507
508static struct nfs4_openowner *
509find_openstateowner_str(unsigned int hashval, struct nfsd4_open *open,
510			struct nfs4_client *clp)
511{
512	struct nfs4_openowner *oo;
513
514	spin_lock(&clp->cl_lock);
515	oo = find_openstateowner_str_locked(hashval, open, clp);
516	spin_unlock(&clp->cl_lock);
517	return oo;
518}
519
520static inline u32
521opaque_hashval(const void *ptr, int nbytes)
522{
523	unsigned char *cptr = (unsigned char *) ptr;
524
525	u32 x = 0;
526	while (nbytes--) {
527		x *= 37;
528		x += *cptr++;
529	}
530	return x;
531}
532
533static void nfsd4_free_file_rcu(struct rcu_head *rcu)
534{
535	struct nfs4_file *fp = container_of(rcu, struct nfs4_file, fi_rcu);
536
537	kmem_cache_free(file_slab, fp);
538}
539
540void
541put_nfs4_file(struct nfs4_file *fi)
542{
543	might_lock(&state_lock);
544
545	if (refcount_dec_and_lock(&fi->fi_ref, &state_lock)) {
546		hlist_del_rcu(&fi->fi_hash);
547		spin_unlock(&state_lock);
548		WARN_ON_ONCE(!list_empty(&fi->fi_clnt_odstate));
549		WARN_ON_ONCE(!list_empty(&fi->fi_delegations));
550		call_rcu(&fi->fi_rcu, nfsd4_free_file_rcu);
551	}
552}
553
554static struct nfsd_file *
555__nfs4_get_fd(struct nfs4_file *f, int oflag)
556{
557	if (f->fi_fds[oflag])
558		return nfsd_file_get(f->fi_fds[oflag]);
559	return NULL;
560}
561
562static struct nfsd_file *
563find_writeable_file_locked(struct nfs4_file *f)
564{
565	struct nfsd_file *ret;
566
567	lockdep_assert_held(&f->fi_lock);
568
569	ret = __nfs4_get_fd(f, O_WRONLY);
570	if (!ret)
571		ret = __nfs4_get_fd(f, O_RDWR);
572	return ret;
573}
574
575static struct nfsd_file *
576find_writeable_file(struct nfs4_file *f)
577{
578	struct nfsd_file *ret;
579
580	spin_lock(&f->fi_lock);
581	ret = find_writeable_file_locked(f);
582	spin_unlock(&f->fi_lock);
583
584	return ret;
585}
586
587static struct nfsd_file *
588find_readable_file_locked(struct nfs4_file *f)
589{
590	struct nfsd_file *ret;
591
592	lockdep_assert_held(&f->fi_lock);
593
594	ret = __nfs4_get_fd(f, O_RDONLY);
595	if (!ret)
596		ret = __nfs4_get_fd(f, O_RDWR);
597	return ret;
598}
599
600static struct nfsd_file *
601find_readable_file(struct nfs4_file *f)
602{
603	struct nfsd_file *ret;
604
605	spin_lock(&f->fi_lock);
606	ret = find_readable_file_locked(f);
607	spin_unlock(&f->fi_lock);
608
609	return ret;
610}
611
612struct nfsd_file *
613find_any_file(struct nfs4_file *f)
614{
615	struct nfsd_file *ret;
616
617	if (!f)
618		return NULL;
619	spin_lock(&f->fi_lock);
620	ret = __nfs4_get_fd(f, O_RDWR);
621	if (!ret) {
622		ret = __nfs4_get_fd(f, O_WRONLY);
623		if (!ret)
624			ret = __nfs4_get_fd(f, O_RDONLY);
625	}
626	spin_unlock(&f->fi_lock);
627	return ret;
628}
629
630static struct nfsd_file *find_deleg_file(struct nfs4_file *f)
631{
632	struct nfsd_file *ret = NULL;
633
634	spin_lock(&f->fi_lock);
635	if (f->fi_deleg_file)
636		ret = nfsd_file_get(f->fi_deleg_file);
637	spin_unlock(&f->fi_lock);
638	return ret;
639}
640
641static atomic_long_t num_delegations;
642unsigned long max_delegations;
643
644/*
645 * Open owner state (share locks)
646 */
647
648/* hash tables for lock and open owners */
649#define OWNER_HASH_BITS              8
650#define OWNER_HASH_SIZE             (1 << OWNER_HASH_BITS)
651#define OWNER_HASH_MASK             (OWNER_HASH_SIZE - 1)
652
653static unsigned int ownerstr_hashval(struct xdr_netobj *ownername)
654{
655	unsigned int ret;
656
657	ret = opaque_hashval(ownername->data, ownername->len);
658	return ret & OWNER_HASH_MASK;
659}
660
661/* hash table for nfs4_file */
662#define FILE_HASH_BITS                   8
663#define FILE_HASH_SIZE                  (1 << FILE_HASH_BITS)
664
665static unsigned int file_hashval(struct svc_fh *fh)
666{
667	struct inode *inode = d_inode(fh->fh_dentry);
668
669	/* XXX: why not (here & in file cache) use inode? */
670	return (unsigned int)hash_long(inode->i_ino, FILE_HASH_BITS);
671}
672
673static struct hlist_head file_hashtbl[FILE_HASH_SIZE];
674
675static void
676__nfs4_file_get_access(struct nfs4_file *fp, u32 access)
677{
678	lockdep_assert_held(&fp->fi_lock);
679
680	if (access & NFS4_SHARE_ACCESS_WRITE)
681		atomic_inc(&fp->fi_access[O_WRONLY]);
682	if (access & NFS4_SHARE_ACCESS_READ)
683		atomic_inc(&fp->fi_access[O_RDONLY]);
684}
685
686static __be32
687nfs4_file_get_access(struct nfs4_file *fp, u32 access)
688{
689	lockdep_assert_held(&fp->fi_lock);
690
691	/* Does this access mode make sense? */
692	if (access & ~NFS4_SHARE_ACCESS_BOTH)
693		return nfserr_inval;
694
695	/* Does it conflict with a deny mode already set? */
696	if ((access & fp->fi_share_deny) != 0)
697		return nfserr_share_denied;
698
699	__nfs4_file_get_access(fp, access);
700	return nfs_ok;
701}
702
703static __be32 nfs4_file_check_deny(struct nfs4_file *fp, u32 deny)
704{
705	/* Common case is that there is no deny mode. */
706	if (deny) {
707		/* Does this deny mode make sense? */
708		if (deny & ~NFS4_SHARE_DENY_BOTH)
709			return nfserr_inval;
710
711		if ((deny & NFS4_SHARE_DENY_READ) &&
712		    atomic_read(&fp->fi_access[O_RDONLY]))
713			return nfserr_share_denied;
714
715		if ((deny & NFS4_SHARE_DENY_WRITE) &&
716		    atomic_read(&fp->fi_access[O_WRONLY]))
717			return nfserr_share_denied;
718	}
719	return nfs_ok;
720}
721
722static void __nfs4_file_put_access(struct nfs4_file *fp, int oflag)
723{
724	might_lock(&fp->fi_lock);
725
726	if (atomic_dec_and_lock(&fp->fi_access[oflag], &fp->fi_lock)) {
727		struct nfsd_file *f1 = NULL;
728		struct nfsd_file *f2 = NULL;
729
730		swap(f1, fp->fi_fds[oflag]);
731		if (atomic_read(&fp->fi_access[1 - oflag]) == 0)
732			swap(f2, fp->fi_fds[O_RDWR]);
733		spin_unlock(&fp->fi_lock);
734		if (f1)
735			nfsd_file_put(f1);
736		if (f2)
737			nfsd_file_put(f2);
738	}
739}
740
741static void nfs4_file_put_access(struct nfs4_file *fp, u32 access)
742{
743	WARN_ON_ONCE(access & ~NFS4_SHARE_ACCESS_BOTH);
744
745	if (access & NFS4_SHARE_ACCESS_WRITE)
746		__nfs4_file_put_access(fp, O_WRONLY);
747	if (access & NFS4_SHARE_ACCESS_READ)
748		__nfs4_file_put_access(fp, O_RDONLY);
749}
750
751/*
752 * Allocate a new open/delegation state counter. This is needed for
753 * pNFS for proper return on close semantics.
754 *
755 * Note that we only allocate it for pNFS-enabled exports, otherwise
756 * all pointers to struct nfs4_clnt_odstate are always NULL.
757 */
758static struct nfs4_clnt_odstate *
759alloc_clnt_odstate(struct nfs4_client *clp)
760{
761	struct nfs4_clnt_odstate *co;
762
763	co = kmem_cache_zalloc(odstate_slab, GFP_KERNEL);
764	if (co) {
765		co->co_client = clp;
766		refcount_set(&co->co_odcount, 1);
767	}
768	return co;
769}
770
771static void
772hash_clnt_odstate_locked(struct nfs4_clnt_odstate *co)
773{
774	struct nfs4_file *fp = co->co_file;
775
776	lockdep_assert_held(&fp->fi_lock);
777	list_add(&co->co_perfile, &fp->fi_clnt_odstate);
778}
779
780static inline void
781get_clnt_odstate(struct nfs4_clnt_odstate *co)
782{
783	if (co)
784		refcount_inc(&co->co_odcount);
785}
786
787static void
788put_clnt_odstate(struct nfs4_clnt_odstate *co)
789{
790	struct nfs4_file *fp;
791
792	if (!co)
793		return;
794
795	fp = co->co_file;
796	if (refcount_dec_and_lock(&co->co_odcount, &fp->fi_lock)) {
797		list_del(&co->co_perfile);
798		spin_unlock(&fp->fi_lock);
799
800		nfsd4_return_all_file_layouts(co->co_client, fp);
801		kmem_cache_free(odstate_slab, co);
802	}
803}
804
805static struct nfs4_clnt_odstate *
806find_or_hash_clnt_odstate(struct nfs4_file *fp, struct nfs4_clnt_odstate *new)
807{
808	struct nfs4_clnt_odstate *co;
809	struct nfs4_client *cl;
810
811	if (!new)
812		return NULL;
813
814	cl = new->co_client;
815
816	spin_lock(&fp->fi_lock);
817	list_for_each_entry(co, &fp->fi_clnt_odstate, co_perfile) {
818		if (co->co_client == cl) {
819			get_clnt_odstate(co);
820			goto out;
821		}
822	}
823	co = new;
824	co->co_file = fp;
825	hash_clnt_odstate_locked(new);
826out:
827	spin_unlock(&fp->fi_lock);
828	return co;
829}
830
831struct nfs4_stid *nfs4_alloc_stid(struct nfs4_client *cl, struct kmem_cache *slab,
832				  void (*sc_free)(struct nfs4_stid *))
833{
834	struct nfs4_stid *stid;
835	int new_id;
836
837	stid = kmem_cache_zalloc(slab, GFP_KERNEL);
838	if (!stid)
839		return NULL;
840
841	idr_preload(GFP_KERNEL);
842	spin_lock(&cl->cl_lock);
843	/* Reserving 0 for start of file in nfsdfs "states" file: */
844	new_id = idr_alloc_cyclic(&cl->cl_stateids, stid, 1, 0, GFP_NOWAIT);
845	spin_unlock(&cl->cl_lock);
846	idr_preload_end();
847	if (new_id < 0)
848		goto out_free;
849
850	stid->sc_free = sc_free;
851	stid->sc_client = cl;
852	stid->sc_stateid.si_opaque.so_id = new_id;
853	stid->sc_stateid.si_opaque.so_clid = cl->cl_clientid;
854	/* Will be incremented before return to client: */
855	refcount_set(&stid->sc_count, 1);
856	spin_lock_init(&stid->sc_lock);
857	INIT_LIST_HEAD(&stid->sc_cp_list);
858
859	/*
860	 * It shouldn't be a problem to reuse an opaque stateid value.
861	 * I don't think it is for 4.1.  But with 4.0 I worry that, for
862	 * example, a stray write retransmission could be accepted by
863	 * the server when it should have been rejected.  Therefore,
864	 * adopt a trick from the sctp code to attempt to maximize the
865	 * amount of time until an id is reused, by ensuring they always
866	 * "increase" (mod INT_MAX):
867	 */
868	return stid;
869out_free:
870	kmem_cache_free(slab, stid);
871	return NULL;
872}
873
874/*
875 * Create a unique stateid_t to represent each COPY.
876 */
877static int nfs4_init_cp_state(struct nfsd_net *nn, copy_stateid_t *stid,
878			      unsigned char sc_type)
879{
880	int new_id;
881
882	stid->stid.si_opaque.so_clid.cl_boot = (u32)nn->boot_time;
883	stid->stid.si_opaque.so_clid.cl_id = nn->s2s_cp_cl_id;
884	stid->sc_type = sc_type;
885
886	idr_preload(GFP_KERNEL);
887	spin_lock(&nn->s2s_cp_lock);
888	new_id = idr_alloc_cyclic(&nn->s2s_cp_stateids, stid, 0, 0, GFP_NOWAIT);
889	stid->stid.si_opaque.so_id = new_id;
890	stid->stid.si_generation = 1;
891	spin_unlock(&nn->s2s_cp_lock);
892	idr_preload_end();
893	if (new_id < 0)
894		return 0;
895	return 1;
896}
897
898int nfs4_init_copy_state(struct nfsd_net *nn, struct nfsd4_copy *copy)
899{
900	return nfs4_init_cp_state(nn, &copy->cp_stateid, NFS4_COPY_STID);
901}
902
903struct nfs4_cpntf_state *nfs4_alloc_init_cpntf_state(struct nfsd_net *nn,
904						     struct nfs4_stid *p_stid)
905{
906	struct nfs4_cpntf_state *cps;
907
908	cps = kzalloc(sizeof(struct nfs4_cpntf_state), GFP_KERNEL);
909	if (!cps)
910		return NULL;
911	cps->cpntf_time = ktime_get_boottime_seconds();
912	refcount_set(&cps->cp_stateid.sc_count, 1);
913	if (!nfs4_init_cp_state(nn, &cps->cp_stateid, NFS4_COPYNOTIFY_STID))
914		goto out_free;
915	spin_lock(&nn->s2s_cp_lock);
916	list_add(&cps->cp_list, &p_stid->sc_cp_list);
917	spin_unlock(&nn->s2s_cp_lock);
918	return cps;
919out_free:
920	kfree(cps);
921	return NULL;
922}
923
924void nfs4_free_copy_state(struct nfsd4_copy *copy)
925{
926	struct nfsd_net *nn;
927
928	WARN_ON_ONCE(copy->cp_stateid.sc_type != NFS4_COPY_STID);
929	nn = net_generic(copy->cp_clp->net, nfsd_net_id);
930	spin_lock(&nn->s2s_cp_lock);
931	idr_remove(&nn->s2s_cp_stateids,
932		   copy->cp_stateid.stid.si_opaque.so_id);
933	spin_unlock(&nn->s2s_cp_lock);
934}
935
936static void nfs4_free_cpntf_statelist(struct net *net, struct nfs4_stid *stid)
937{
938	struct nfs4_cpntf_state *cps;
939	struct nfsd_net *nn;
940
941	nn = net_generic(net, nfsd_net_id);
942	spin_lock(&nn->s2s_cp_lock);
943	while (!list_empty(&stid->sc_cp_list)) {
944		cps = list_first_entry(&stid->sc_cp_list,
945				       struct nfs4_cpntf_state, cp_list);
946		_free_cpntf_state_locked(nn, cps);
947	}
948	spin_unlock(&nn->s2s_cp_lock);
949}
950
951static struct nfs4_ol_stateid * nfs4_alloc_open_stateid(struct nfs4_client *clp)
952{
953	struct nfs4_stid *stid;
954
955	stid = nfs4_alloc_stid(clp, stateid_slab, nfs4_free_ol_stateid);
956	if (!stid)
957		return NULL;
958
959	return openlockstateid(stid);
960}
961
962static void nfs4_free_deleg(struct nfs4_stid *stid)
963{
964	kmem_cache_free(deleg_slab, stid);
965	atomic_long_dec(&num_delegations);
966}
967
968/*
969 * When we recall a delegation, we should be careful not to hand it
970 * out again straight away.
971 * To ensure this we keep a pair of bloom filters ('new' and 'old')
972 * in which the filehandles of recalled delegations are "stored".
973 * If a filehandle appear in either filter, a delegation is blocked.
974 * When a delegation is recalled, the filehandle is stored in the "new"
975 * filter.
976 * Every 30 seconds we swap the filters and clear the "new" one,
977 * unless both are empty of course.
978 *
979 * Each filter is 256 bits.  We hash the filehandle to 32bit and use the
980 * low 3 bytes as hash-table indices.
981 *
982 * 'blocked_delegations_lock', which is always taken in block_delegations(),
983 * is used to manage concurrent access.  Testing does not need the lock
984 * except when swapping the two filters.
985 */
986static DEFINE_SPINLOCK(blocked_delegations_lock);
987static struct bloom_pair {
988	int	entries, old_entries;
989	time64_t swap_time;
990	int	new; /* index into 'set' */
991	DECLARE_BITMAP(set[2], 256);
992} blocked_delegations;
993
994static int delegation_blocked(struct knfsd_fh *fh)
995{
996	u32 hash;
997	struct bloom_pair *bd = &blocked_delegations;
998
999	if (bd->entries == 0)
1000		return 0;
1001	if (ktime_get_seconds() - bd->swap_time > 30) {
1002		spin_lock(&blocked_delegations_lock);
1003		if (ktime_get_seconds() - bd->swap_time > 30) {
1004			bd->entries -= bd->old_entries;
1005			bd->old_entries = bd->entries;
1006			memset(bd->set[bd->new], 0,
1007			       sizeof(bd->set[0]));
1008			bd->new = 1-bd->new;
1009			bd->swap_time = ktime_get_seconds();
1010		}
1011		spin_unlock(&blocked_delegations_lock);
1012	}
1013	hash = jhash(&fh->fh_base, fh->fh_size, 0);
1014	if (test_bit(hash&255, bd->set[0]) &&
1015	    test_bit((hash>>8)&255, bd->set[0]) &&
1016	    test_bit((hash>>16)&255, bd->set[0]))
1017		return 1;
1018
1019	if (test_bit(hash&255, bd->set[1]) &&
1020	    test_bit((hash>>8)&255, bd->set[1]) &&
1021	    test_bit((hash>>16)&255, bd->set[1]))
1022		return 1;
1023
1024	return 0;
1025}
1026
1027static void block_delegations(struct knfsd_fh *fh)
1028{
1029	u32 hash;
1030	struct bloom_pair *bd = &blocked_delegations;
1031
1032	hash = jhash(&fh->fh_base, fh->fh_size, 0);
1033
1034	spin_lock(&blocked_delegations_lock);
1035	__set_bit(hash&255, bd->set[bd->new]);
1036	__set_bit((hash>>8)&255, bd->set[bd->new]);
1037	__set_bit((hash>>16)&255, bd->set[bd->new]);
1038	if (bd->entries == 0)
1039		bd->swap_time = ktime_get_seconds();
1040	bd->entries += 1;
1041	spin_unlock(&blocked_delegations_lock);
1042}
1043
1044static struct nfs4_delegation *
1045alloc_init_deleg(struct nfs4_client *clp, struct nfs4_file *fp,
1046		 struct svc_fh *current_fh,
1047		 struct nfs4_clnt_odstate *odstate)
1048{
1049	struct nfs4_delegation *dp;
1050	long n;
1051
1052	dprintk("NFSD alloc_init_deleg\n");
1053	n = atomic_long_inc_return(&num_delegations);
1054	if (n < 0 || n > max_delegations)
1055		goto out_dec;
1056	if (delegation_blocked(&current_fh->fh_handle))
1057		goto out_dec;
1058	dp = delegstateid(nfs4_alloc_stid(clp, deleg_slab, nfs4_free_deleg));
1059	if (dp == NULL)
1060		goto out_dec;
1061
1062	/*
1063	 * delegation seqid's are never incremented.  The 4.1 special
1064	 * meaning of seqid 0 isn't meaningful, really, but let's avoid
1065	 * 0 anyway just for consistency and use 1:
1066	 */
1067	dp->dl_stid.sc_stateid.si_generation = 1;
1068	INIT_LIST_HEAD(&dp->dl_perfile);
1069	INIT_LIST_HEAD(&dp->dl_perclnt);
1070	INIT_LIST_HEAD(&dp->dl_recall_lru);
1071	dp->dl_clnt_odstate = odstate;
1072	get_clnt_odstate(odstate);
1073	dp->dl_type = NFS4_OPEN_DELEGATE_READ;
1074	dp->dl_retries = 1;
1075	nfsd4_init_cb(&dp->dl_recall, dp->dl_stid.sc_client,
1076		      &nfsd4_cb_recall_ops, NFSPROC4_CLNT_CB_RECALL);
1077	get_nfs4_file(fp);
1078	dp->dl_stid.sc_file = fp;
1079	return dp;
1080out_dec:
1081	atomic_long_dec(&num_delegations);
1082	return NULL;
1083}
1084
1085void
1086nfs4_put_stid(struct nfs4_stid *s)
1087{
1088	struct nfs4_file *fp = s->sc_file;
1089	struct nfs4_client *clp = s->sc_client;
1090
1091	might_lock(&clp->cl_lock);
1092
1093	if (!refcount_dec_and_lock(&s->sc_count, &clp->cl_lock)) {
1094		wake_up_all(&close_wq);
1095		return;
1096	}
1097	idr_remove(&clp->cl_stateids, s->sc_stateid.si_opaque.so_id);
1098	nfs4_free_cpntf_statelist(clp->net, s);
1099	spin_unlock(&clp->cl_lock);
1100	s->sc_free(s);
1101	if (fp)
1102		put_nfs4_file(fp);
1103}
1104
1105void
1106nfs4_inc_and_copy_stateid(stateid_t *dst, struct nfs4_stid *stid)
1107{
1108	stateid_t *src = &stid->sc_stateid;
1109
1110	spin_lock(&stid->sc_lock);
1111	if (unlikely(++src->si_generation == 0))
1112		src->si_generation = 1;
1113	memcpy(dst, src, sizeof(*dst));
1114	spin_unlock(&stid->sc_lock);
1115}
1116
1117static void put_deleg_file(struct nfs4_file *fp)
1118{
1119	struct nfsd_file *nf = NULL;
1120
1121	spin_lock(&fp->fi_lock);
1122	if (--fp->fi_delegees == 0)
1123		swap(nf, fp->fi_deleg_file);
1124	spin_unlock(&fp->fi_lock);
1125
1126	if (nf)
1127		nfsd_file_put(nf);
1128}
1129
1130static void nfs4_unlock_deleg_lease(struct nfs4_delegation *dp)
1131{
1132	struct nfs4_file *fp = dp->dl_stid.sc_file;
1133	struct nfsd_file *nf = fp->fi_deleg_file;
1134
1135	WARN_ON_ONCE(!fp->fi_delegees);
1136
1137	vfs_setlease(nf->nf_file, F_UNLCK, NULL, (void **)&dp);
1138	put_deleg_file(fp);
1139}
1140
1141static void destroy_unhashed_deleg(struct nfs4_delegation *dp)
1142{
1143	put_clnt_odstate(dp->dl_clnt_odstate);
1144	nfs4_unlock_deleg_lease(dp);
1145	nfs4_put_stid(&dp->dl_stid);
1146}
1147
1148void nfs4_unhash_stid(struct nfs4_stid *s)
1149{
1150	s->sc_type = 0;
1151}
1152
1153/**
1154 * nfs4_delegation_exists - Discover if this delegation already exists
1155 * @clp:     a pointer to the nfs4_client we're granting a delegation to
1156 * @fp:      a pointer to the nfs4_file we're granting a delegation on
1157 *
1158 * Return:
1159 *      On success: true iff an existing delegation is found
1160 */
1161
1162static bool
1163nfs4_delegation_exists(struct nfs4_client *clp, struct nfs4_file *fp)
1164{
1165	struct nfs4_delegation *searchdp = NULL;
1166	struct nfs4_client *searchclp = NULL;
1167
1168	lockdep_assert_held(&state_lock);
1169	lockdep_assert_held(&fp->fi_lock);
1170
1171	list_for_each_entry(searchdp, &fp->fi_delegations, dl_perfile) {
1172		searchclp = searchdp->dl_stid.sc_client;
1173		if (clp == searchclp) {
1174			return true;
1175		}
1176	}
1177	return false;
1178}
1179
1180/**
1181 * hash_delegation_locked - Add a delegation to the appropriate lists
1182 * @dp:     a pointer to the nfs4_delegation we are adding.
1183 * @fp:     a pointer to the nfs4_file we're granting a delegation on
1184 *
1185 * Return:
1186 *      On success: NULL if the delegation was successfully hashed.
1187 *
1188 *      On error: -EAGAIN if one was previously granted to this
1189 *                 nfs4_client for this nfs4_file. Delegation is not hashed.
1190 *
1191 */
1192
1193static int
1194hash_delegation_locked(struct nfs4_delegation *dp, struct nfs4_file *fp)
1195{
1196	struct nfs4_client *clp = dp->dl_stid.sc_client;
1197
1198	lockdep_assert_held(&state_lock);
1199	lockdep_assert_held(&fp->fi_lock);
1200
1201	if (nfs4_delegation_exists(clp, fp))
1202		return -EAGAIN;
1203	refcount_inc(&dp->dl_stid.sc_count);
1204	dp->dl_stid.sc_type = NFS4_DELEG_STID;
1205	list_add(&dp->dl_perfile, &fp->fi_delegations);
1206	list_add(&dp->dl_perclnt, &clp->cl_delegations);
1207	return 0;
1208}
1209
1210static bool
1211unhash_delegation_locked(struct nfs4_delegation *dp)
1212{
1213	struct nfs4_file *fp = dp->dl_stid.sc_file;
1214
1215	lockdep_assert_held(&state_lock);
1216
1217	if (list_empty(&dp->dl_perfile))
1218		return false;
1219
1220	dp->dl_stid.sc_type = NFS4_CLOSED_DELEG_STID;
1221	/* Ensure that deleg break won't try to requeue it */
1222	++dp->dl_time;
1223	spin_lock(&fp->fi_lock);
1224	list_del_init(&dp->dl_perclnt);
1225	list_del_init(&dp->dl_recall_lru);
1226	list_del_init(&dp->dl_perfile);
1227	spin_unlock(&fp->fi_lock);
1228	return true;
1229}
1230
1231static void destroy_delegation(struct nfs4_delegation *dp)
1232{
1233	bool unhashed;
1234
1235	spin_lock(&state_lock);
1236	unhashed = unhash_delegation_locked(dp);
1237	spin_unlock(&state_lock);
1238	if (unhashed)
1239		destroy_unhashed_deleg(dp);
1240}
1241
1242static void revoke_delegation(struct nfs4_delegation *dp)
1243{
1244	struct nfs4_client *clp = dp->dl_stid.sc_client;
1245
1246	WARN_ON(!list_empty(&dp->dl_recall_lru));
1247
1248	if (clp->cl_minorversion) {
1249		dp->dl_stid.sc_type = NFS4_REVOKED_DELEG_STID;
1250		refcount_inc(&dp->dl_stid.sc_count);
1251		spin_lock(&clp->cl_lock);
1252		list_add(&dp->dl_recall_lru, &clp->cl_revoked);
1253		spin_unlock(&clp->cl_lock);
1254	}
1255	destroy_unhashed_deleg(dp);
1256}
1257
1258/*
1259 * SETCLIENTID state
1260 */
1261
1262static unsigned int clientid_hashval(u32 id)
1263{
1264	return id & CLIENT_HASH_MASK;
1265}
1266
1267static unsigned int clientstr_hashval(struct xdr_netobj name)
1268{
1269	return opaque_hashval(name.data, 8) & CLIENT_HASH_MASK;
1270}
1271
1272/*
1273 * A stateid that had a deny mode associated with it is being released
1274 * or downgraded. Recalculate the deny mode on the file.
1275 */
1276static void
1277recalculate_deny_mode(struct nfs4_file *fp)
1278{
1279	struct nfs4_ol_stateid *stp;
1280
1281	spin_lock(&fp->fi_lock);
1282	fp->fi_share_deny = 0;
1283	list_for_each_entry(stp, &fp->fi_stateids, st_perfile)
1284		fp->fi_share_deny |= bmap_to_share_mode(stp->st_deny_bmap);
1285	spin_unlock(&fp->fi_lock);
1286}
1287
1288static void
1289reset_union_bmap_deny(u32 deny, struct nfs4_ol_stateid *stp)
1290{
1291	int i;
1292	bool change = false;
1293
1294	for (i = 1; i < 4; i++) {
1295		if ((i & deny) != i) {
1296			change = true;
1297			clear_deny(i, stp);
1298		}
1299	}
1300
1301	/* Recalculate per-file deny mode if there was a change */
1302	if (change)
1303		recalculate_deny_mode(stp->st_stid.sc_file);
1304}
1305
1306/* release all access and file references for a given stateid */
1307static void
1308release_all_access(struct nfs4_ol_stateid *stp)
1309{
1310	int i;
1311	struct nfs4_file *fp = stp->st_stid.sc_file;
1312
1313	if (fp && stp->st_deny_bmap != 0)
1314		recalculate_deny_mode(fp);
1315
1316	for (i = 1; i < 4; i++) {
1317		if (test_access(i, stp))
1318			nfs4_file_put_access(stp->st_stid.sc_file, i);
1319		clear_access(i, stp);
1320	}
1321}
1322
1323static inline void nfs4_free_stateowner(struct nfs4_stateowner *sop)
1324{
1325	kfree(sop->so_owner.data);
1326	sop->so_ops->so_free(sop);
1327}
1328
1329static void nfs4_put_stateowner(struct nfs4_stateowner *sop)
1330{
1331	struct nfs4_client *clp = sop->so_client;
1332
1333	might_lock(&clp->cl_lock);
1334
1335	if (!atomic_dec_and_lock(&sop->so_count, &clp->cl_lock))
1336		return;
1337	sop->so_ops->so_unhash(sop);
1338	spin_unlock(&clp->cl_lock);
1339	nfs4_free_stateowner(sop);
1340}
1341
1342static bool
1343nfs4_ol_stateid_unhashed(const struct nfs4_ol_stateid *stp)
1344{
1345	return list_empty(&stp->st_perfile);
1346}
1347
1348static bool unhash_ol_stateid(struct nfs4_ol_stateid *stp)
1349{
1350	struct nfs4_file *fp = stp->st_stid.sc_file;
1351
1352	lockdep_assert_held(&stp->st_stateowner->so_client->cl_lock);
1353
1354	if (list_empty(&stp->st_perfile))
1355		return false;
1356
1357	spin_lock(&fp->fi_lock);
1358	list_del_init(&stp->st_perfile);
1359	spin_unlock(&fp->fi_lock);
1360	list_del(&stp->st_perstateowner);
1361	return true;
1362}
1363
1364static void nfs4_free_ol_stateid(struct nfs4_stid *stid)
1365{
1366	struct nfs4_ol_stateid *stp = openlockstateid(stid);
1367
1368	put_clnt_odstate(stp->st_clnt_odstate);
1369	release_all_access(stp);
1370	if (stp->st_stateowner)
1371		nfs4_put_stateowner(stp->st_stateowner);
1372	kmem_cache_free(stateid_slab, stid);
1373}
1374
1375static void nfs4_free_lock_stateid(struct nfs4_stid *stid)
1376{
1377	struct nfs4_ol_stateid *stp = openlockstateid(stid);
1378	struct nfs4_lockowner *lo = lockowner(stp->st_stateowner);
1379	struct nfsd_file *nf;
1380
1381	nf = find_any_file(stp->st_stid.sc_file);
1382	if (nf) {
1383		get_file(nf->nf_file);
1384		filp_close(nf->nf_file, (fl_owner_t)lo);
1385		nfsd_file_put(nf);
1386	}
1387	nfs4_free_ol_stateid(stid);
1388}
1389
1390/*
1391 * Put the persistent reference to an already unhashed generic stateid, while
1392 * holding the cl_lock. If it's the last reference, then put it onto the
1393 * reaplist for later destruction.
1394 */
1395static void put_ol_stateid_locked(struct nfs4_ol_stateid *stp,
1396				       struct list_head *reaplist)
1397{
1398	struct nfs4_stid *s = &stp->st_stid;
1399	struct nfs4_client *clp = s->sc_client;
1400
1401	lockdep_assert_held(&clp->cl_lock);
1402
1403	WARN_ON_ONCE(!list_empty(&stp->st_locks));
1404
1405	if (!refcount_dec_and_test(&s->sc_count)) {
1406		wake_up_all(&close_wq);
1407		return;
1408	}
1409
1410	idr_remove(&clp->cl_stateids, s->sc_stateid.si_opaque.so_id);
1411	list_add(&stp->st_locks, reaplist);
1412}
1413
1414static bool unhash_lock_stateid(struct nfs4_ol_stateid *stp)
1415{
1416	lockdep_assert_held(&stp->st_stid.sc_client->cl_lock);
1417
1418	if (!unhash_ol_stateid(stp))
1419		return false;
1420	list_del_init(&stp->st_locks);
1421	nfs4_unhash_stid(&stp->st_stid);
1422	return true;
1423}
1424
1425static void release_lock_stateid(struct nfs4_ol_stateid *stp)
1426{
1427	struct nfs4_client *clp = stp->st_stid.sc_client;
1428	bool unhashed;
1429
1430	spin_lock(&clp->cl_lock);
1431	unhashed = unhash_lock_stateid(stp);
1432	spin_unlock(&clp->cl_lock);
1433	if (unhashed)
1434		nfs4_put_stid(&stp->st_stid);
1435}
1436
1437static void unhash_lockowner_locked(struct nfs4_lockowner *lo)
1438{
1439	struct nfs4_client *clp = lo->lo_owner.so_client;
1440
1441	lockdep_assert_held(&clp->cl_lock);
1442
1443	list_del_init(&lo->lo_owner.so_strhash);
1444}
1445
1446/*
1447 * Free a list of generic stateids that were collected earlier after being
1448 * fully unhashed.
1449 */
1450static void
1451free_ol_stateid_reaplist(struct list_head *reaplist)
1452{
1453	struct nfs4_ol_stateid *stp;
1454	struct nfs4_file *fp;
1455
1456	might_sleep();
1457
1458	while (!list_empty(reaplist)) {
1459		stp = list_first_entry(reaplist, struct nfs4_ol_stateid,
1460				       st_locks);
1461		list_del(&stp->st_locks);
1462		fp = stp->st_stid.sc_file;
1463		stp->st_stid.sc_free(&stp->st_stid);
1464		if (fp)
1465			put_nfs4_file(fp);
1466	}
1467}
1468
1469static void release_open_stateid_locks(struct nfs4_ol_stateid *open_stp,
1470				       struct list_head *reaplist)
1471{
1472	struct nfs4_ol_stateid *stp;
1473
1474	lockdep_assert_held(&open_stp->st_stid.sc_client->cl_lock);
1475
1476	while (!list_empty(&open_stp->st_locks)) {
1477		stp = list_entry(open_stp->st_locks.next,
1478				struct nfs4_ol_stateid, st_locks);
1479		WARN_ON(!unhash_lock_stateid(stp));
1480		put_ol_stateid_locked(stp, reaplist);
1481	}
1482}
1483
1484static bool unhash_open_stateid(struct nfs4_ol_stateid *stp,
1485				struct list_head *reaplist)
1486{
1487	lockdep_assert_held(&stp->st_stid.sc_client->cl_lock);
1488
1489	if (!unhash_ol_stateid(stp))
1490		return false;
1491	release_open_stateid_locks(stp, reaplist);
1492	return true;
1493}
1494
1495static void release_open_stateid(struct nfs4_ol_stateid *stp)
1496{
1497	LIST_HEAD(reaplist);
1498
1499	spin_lock(&stp->st_stid.sc_client->cl_lock);
1500	if (unhash_open_stateid(stp, &reaplist))
1501		put_ol_stateid_locked(stp, &reaplist);
1502	spin_unlock(&stp->st_stid.sc_client->cl_lock);
1503	free_ol_stateid_reaplist(&reaplist);
1504}
1505
1506static void unhash_openowner_locked(struct nfs4_openowner *oo)
1507{
1508	struct nfs4_client *clp = oo->oo_owner.so_client;
1509
1510	lockdep_assert_held(&clp->cl_lock);
1511
1512	list_del_init(&oo->oo_owner.so_strhash);
1513	list_del_init(&oo->oo_perclient);
1514}
1515
1516static void release_last_closed_stateid(struct nfs4_openowner *oo)
1517{
1518	struct nfsd_net *nn = net_generic(oo->oo_owner.so_client->net,
1519					  nfsd_net_id);
1520	struct nfs4_ol_stateid *s;
1521
1522	spin_lock(&nn->client_lock);
1523	s = oo->oo_last_closed_stid;
1524	if (s) {
1525		list_del_init(&oo->oo_close_lru);
1526		oo->oo_last_closed_stid = NULL;
1527	}
1528	spin_unlock(&nn->client_lock);
1529	if (s)
1530		nfs4_put_stid(&s->st_stid);
1531}
1532
1533static void release_openowner(struct nfs4_openowner *oo)
1534{
1535	struct nfs4_ol_stateid *stp;
1536	struct nfs4_client *clp = oo->oo_owner.so_client;
1537	struct list_head reaplist;
1538
1539	INIT_LIST_HEAD(&reaplist);
1540
1541	spin_lock(&clp->cl_lock);
1542	unhash_openowner_locked(oo);
1543	while (!list_empty(&oo->oo_owner.so_stateids)) {
1544		stp = list_first_entry(&oo->oo_owner.so_stateids,
1545				struct nfs4_ol_stateid, st_perstateowner);
1546		if (unhash_open_stateid(stp, &reaplist))
1547			put_ol_stateid_locked(stp, &reaplist);
1548	}
1549	spin_unlock(&clp->cl_lock);
1550	free_ol_stateid_reaplist(&reaplist);
1551	release_last_closed_stateid(oo);
1552	nfs4_put_stateowner(&oo->oo_owner);
1553}
1554
1555static inline int
1556hash_sessionid(struct nfs4_sessionid *sessionid)
1557{
1558	struct nfsd4_sessionid *sid = (struct nfsd4_sessionid *)sessionid;
1559
1560	return sid->sequence % SESSION_HASH_SIZE;
1561}
1562
1563#ifdef CONFIG_SUNRPC_DEBUG
1564static inline void
1565dump_sessionid(const char *fn, struct nfs4_sessionid *sessionid)
1566{
1567	u32 *ptr = (u32 *)(&sessionid->data[0]);
1568	dprintk("%s: %u:%u:%u:%u\n", fn, ptr[0], ptr[1], ptr[2], ptr[3]);
1569}
1570#else
1571static inline void
1572dump_sessionid(const char *fn, struct nfs4_sessionid *sessionid)
1573{
1574}
1575#endif
1576
1577/*
1578 * Bump the seqid on cstate->replay_owner, and clear replay_owner if it
1579 * won't be used for replay.
1580 */
1581void nfsd4_bump_seqid(struct nfsd4_compound_state *cstate, __be32 nfserr)
1582{
1583	struct nfs4_stateowner *so = cstate->replay_owner;
1584
1585	if (nfserr == nfserr_replay_me)
1586		return;
1587
1588	if (!seqid_mutating_err(ntohl(nfserr))) {
1589		nfsd4_cstate_clear_replay(cstate);
1590		return;
1591	}
1592	if (!so)
1593		return;
1594	if (so->so_is_open_owner)
1595		release_last_closed_stateid(openowner(so));
1596	so->so_seqid++;
1597	return;
1598}
1599
1600static void
1601gen_sessionid(struct nfsd4_session *ses)
1602{
1603	struct nfs4_client *clp = ses->se_client;
1604	struct nfsd4_sessionid *sid;
1605
1606	sid = (struct nfsd4_sessionid *)ses->se_sessionid.data;
1607	sid->clientid = clp->cl_clientid;
1608	sid->sequence = current_sessionid++;
1609	sid->reserved = 0;
1610}
1611
1612/*
1613 * The protocol defines ca_maxresponssize_cached to include the size of
1614 * the rpc header, but all we need to cache is the data starting after
1615 * the end of the initial SEQUENCE operation--the rest we regenerate
1616 * each time.  Therefore we can advertise a ca_maxresponssize_cached
1617 * value that is the number of bytes in our cache plus a few additional
1618 * bytes.  In order to stay on the safe side, and not promise more than
1619 * we can cache, those additional bytes must be the minimum possible: 24
1620 * bytes of rpc header (xid through accept state, with AUTH_NULL
1621 * verifier), 12 for the compound header (with zero-length tag), and 44
1622 * for the SEQUENCE op response:
1623 */
1624#define NFSD_MIN_HDR_SEQ_SZ  (24 + 12 + 44)
1625
1626static void
1627free_session_slots(struct nfsd4_session *ses)
1628{
1629	int i;
1630
1631	for (i = 0; i < ses->se_fchannel.maxreqs; i++) {
1632		free_svc_cred(&ses->se_slots[i]->sl_cred);
1633		kfree(ses->se_slots[i]);
1634	}
1635}
1636
1637/*
1638 * We don't actually need to cache the rpc and session headers, so we
1639 * can allocate a little less for each slot:
1640 */
1641static inline u32 slot_bytes(struct nfsd4_channel_attrs *ca)
1642{
1643	u32 size;
1644
1645	if (ca->maxresp_cached < NFSD_MIN_HDR_SEQ_SZ)
1646		size = 0;
1647	else
1648		size = ca->maxresp_cached - NFSD_MIN_HDR_SEQ_SZ;
1649	return size + sizeof(struct nfsd4_slot);
1650}
1651
1652/*
1653 * XXX: If we run out of reserved DRC memory we could (up to a point)
1654 * re-negotiate active sessions and reduce their slot usage to make
1655 * room for new connections. For now we just fail the create session.
1656 */
1657static u32 nfsd4_get_drc_mem(struct nfsd4_channel_attrs *ca, struct nfsd_net *nn)
1658{
1659	u32 slotsize = slot_bytes(ca);
1660	u32 num = ca->maxreqs;
1661	unsigned long avail, total_avail;
1662	unsigned int scale_factor;
1663
1664	spin_lock(&nfsd_drc_lock);
1665	if (nfsd_drc_max_mem > nfsd_drc_mem_used)
1666		total_avail = nfsd_drc_max_mem - nfsd_drc_mem_used;
1667	else
1668		/* We have handed out more space than we chose in
1669		 * set_max_drc() to allow.  That isn't really a
1670		 * problem as long as that doesn't make us think we
1671		 * have lots more due to integer overflow.
1672		 */
1673		total_avail = 0;
1674	avail = min((unsigned long)NFSD_MAX_MEM_PER_SESSION, total_avail);
1675	/*
1676	 * Never use more than a fraction of the remaining memory,
1677	 * unless it's the only way to give this client a slot.
1678	 * The chosen fraction is either 1/8 or 1/number of threads,
1679	 * whichever is smaller.  This ensures there are adequate
1680	 * slots to support multiple clients per thread.
1681	 * Give the client one slot even if that would require
1682	 * over-allocation--it is better than failure.
1683	 */
1684	scale_factor = max_t(unsigned int, 8, nn->nfsd_serv->sv_nrthreads);
1685
1686	avail = clamp_t(unsigned long, avail, slotsize,
1687			total_avail/scale_factor);
1688	num = min_t(int, num, avail / slotsize);
1689	num = max_t(int, num, 1);
1690	nfsd_drc_mem_used += num * slotsize;
1691	spin_unlock(&nfsd_drc_lock);
1692
1693	return num;
1694}
1695
1696static void nfsd4_put_drc_mem(struct nfsd4_channel_attrs *ca)
1697{
1698	int slotsize = slot_bytes(ca);
1699
1700	spin_lock(&nfsd_drc_lock);
1701	nfsd_drc_mem_used -= slotsize * ca->maxreqs;
1702	spin_unlock(&nfsd_drc_lock);
1703}
1704
1705static struct nfsd4_session *alloc_session(struct nfsd4_channel_attrs *fattrs,
1706					   struct nfsd4_channel_attrs *battrs)
1707{
1708	int numslots = fattrs->maxreqs;
1709	int slotsize = slot_bytes(fattrs);
1710	struct nfsd4_session *new;
1711	int mem, i;
1712
1713	BUILD_BUG_ON(NFSD_MAX_SLOTS_PER_SESSION * sizeof(struct nfsd4_slot *)
1714			+ sizeof(struct nfsd4_session) > PAGE_SIZE);
1715	mem = numslots * sizeof(struct nfsd4_slot *);
1716
1717	new = kzalloc(sizeof(*new) + mem, GFP_KERNEL);
1718	if (!new)
1719		return NULL;
1720	/* allocate each struct nfsd4_slot and data cache in one piece */
1721	for (i = 0; i < numslots; i++) {
1722		new->se_slots[i] = kzalloc(slotsize, GFP_KERNEL);
1723		if (!new->se_slots[i])
1724			goto out_free;
1725	}
1726
1727	memcpy(&new->se_fchannel, fattrs, sizeof(struct nfsd4_channel_attrs));
1728	memcpy(&new->se_bchannel, battrs, sizeof(struct nfsd4_channel_attrs));
1729
1730	return new;
1731out_free:
1732	while (i--)
1733		kfree(new->se_slots[i]);
1734	kfree(new);
1735	return NULL;
1736}
1737
1738static void free_conn(struct nfsd4_conn *c)
1739{
1740	svc_xprt_put(c->cn_xprt);
1741	kfree(c);
1742}
1743
1744static void nfsd4_conn_lost(struct svc_xpt_user *u)
1745{
1746	struct nfsd4_conn *c = container_of(u, struct nfsd4_conn, cn_xpt_user);
1747	struct nfs4_client *clp = c->cn_session->se_client;
1748
1749	trace_nfsd_cb_lost(clp);
1750
1751	spin_lock(&clp->cl_lock);
1752	if (!list_empty(&c->cn_persession)) {
1753		list_del(&c->cn_persession);
1754		free_conn(c);
1755	}
1756	nfsd4_probe_callback(clp);
1757	spin_unlock(&clp->cl_lock);
1758}
1759
1760static struct nfsd4_conn *alloc_conn(struct svc_rqst *rqstp, u32 flags)
1761{
1762	struct nfsd4_conn *conn;
1763
1764	conn = kmalloc(sizeof(struct nfsd4_conn), GFP_KERNEL);
1765	if (!conn)
1766		return NULL;
1767	svc_xprt_get(rqstp->rq_xprt);
1768	conn->cn_xprt = rqstp->rq_xprt;
1769	conn->cn_flags = flags;
1770	INIT_LIST_HEAD(&conn->cn_xpt_user.list);
1771	return conn;
1772}
1773
1774static void __nfsd4_hash_conn(struct nfsd4_conn *conn, struct nfsd4_session *ses)
1775{
1776	conn->cn_session = ses;
1777	list_add(&conn->cn_persession, &ses->se_conns);
1778}
1779
1780static void nfsd4_hash_conn(struct nfsd4_conn *conn, struct nfsd4_session *ses)
1781{
1782	struct nfs4_client *clp = ses->se_client;
1783
1784	spin_lock(&clp->cl_lock);
1785	__nfsd4_hash_conn(conn, ses);
1786	spin_unlock(&clp->cl_lock);
1787}
1788
1789static int nfsd4_register_conn(struct nfsd4_conn *conn)
1790{
1791	conn->cn_xpt_user.callback = nfsd4_conn_lost;
1792	return register_xpt_user(conn->cn_xprt, &conn->cn_xpt_user);
1793}
1794
1795static void nfsd4_init_conn(struct svc_rqst *rqstp, struct nfsd4_conn *conn, struct nfsd4_session *ses)
1796{
1797	int ret;
1798
1799	nfsd4_hash_conn(conn, ses);
1800	ret = nfsd4_register_conn(conn);
1801	if (ret)
1802		/* oops; xprt is already down: */
1803		nfsd4_conn_lost(&conn->cn_xpt_user);
1804	/* We may have gained or lost a callback channel: */
1805	nfsd4_probe_callback_sync(ses->se_client);
1806}
1807
1808static struct nfsd4_conn *alloc_conn_from_crses(struct svc_rqst *rqstp, struct nfsd4_create_session *cses)
1809{
1810	u32 dir = NFS4_CDFC4_FORE;
1811
1812	if (cses->flags & SESSION4_BACK_CHAN)
1813		dir |= NFS4_CDFC4_BACK;
1814	return alloc_conn(rqstp, dir);
1815}
1816
1817/* must be called under client_lock */
1818static void nfsd4_del_conns(struct nfsd4_session *s)
1819{
1820	struct nfs4_client *clp = s->se_client;
1821	struct nfsd4_conn *c;
1822
1823	spin_lock(&clp->cl_lock);
1824	while (!list_empty(&s->se_conns)) {
1825		c = list_first_entry(&s->se_conns, struct nfsd4_conn, cn_persession);
1826		list_del_init(&c->cn_persession);
1827		spin_unlock(&clp->cl_lock);
1828
1829		unregister_xpt_user(c->cn_xprt, &c->cn_xpt_user);
1830		free_conn(c);
1831
1832		spin_lock(&clp->cl_lock);
1833	}
1834	spin_unlock(&clp->cl_lock);
1835}
1836
1837static void __free_session(struct nfsd4_session *ses)
1838{
1839	free_session_slots(ses);
1840	kfree(ses);
1841}
1842
1843static void free_session(struct nfsd4_session *ses)
1844{
1845	nfsd4_del_conns(ses);
1846	nfsd4_put_drc_mem(&ses->se_fchannel);
1847	__free_session(ses);
1848}
1849
1850static void init_session(struct svc_rqst *rqstp, struct nfsd4_session *new, struct nfs4_client *clp, struct nfsd4_create_session *cses)
1851{
1852	int idx;
1853	struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
1854
1855	new->se_client = clp;
1856	gen_sessionid(new);
1857
1858	INIT_LIST_HEAD(&new->se_conns);
1859
1860	new->se_cb_seq_nr = 1;
1861	new->se_flags = cses->flags;
1862	new->se_cb_prog = cses->callback_prog;
1863	new->se_cb_sec = cses->cb_sec;
1864	atomic_set(&new->se_ref, 0);
1865	idx = hash_sessionid(&new->se_sessionid);
1866	list_add(&new->se_hash, &nn->sessionid_hashtbl[idx]);
1867	spin_lock(&clp->cl_lock);
1868	list_add(&new->se_perclnt, &clp->cl_sessions);
1869	spin_unlock(&clp->cl_lock);
1870
1871	{
1872		struct sockaddr *sa = svc_addr(rqstp);
1873		/*
1874		 * This is a little silly; with sessions there's no real
1875		 * use for the callback address.  Use the peer address
1876		 * as a reasonable default for now, but consider fixing
1877		 * the rpc client not to require an address in the
1878		 * future:
1879		 */
1880		rpc_copy_addr((struct sockaddr *)&clp->cl_cb_conn.cb_addr, sa);
1881		clp->cl_cb_conn.cb_addrlen = svc_addr_len(sa);
1882	}
1883}
1884
1885/* caller must hold client_lock */
1886static struct nfsd4_session *
1887__find_in_sessionid_hashtbl(struct nfs4_sessionid *sessionid, struct net *net)
1888{
1889	struct nfsd4_session *elem;
1890	int idx;
1891	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
1892
1893	lockdep_assert_held(&nn->client_lock);
1894
1895	dump_sessionid(__func__, sessionid);
1896	idx = hash_sessionid(sessionid);
1897	/* Search in the appropriate list */
1898	list_for_each_entry(elem, &nn->sessionid_hashtbl[idx], se_hash) {
1899		if (!memcmp(elem->se_sessionid.data, sessionid->data,
1900			    NFS4_MAX_SESSIONID_LEN)) {
1901			return elem;
1902		}
1903	}
1904
1905	dprintk("%s: session not found\n", __func__);
1906	return NULL;
1907}
1908
1909static struct nfsd4_session *
1910find_in_sessionid_hashtbl(struct nfs4_sessionid *sessionid, struct net *net,
1911		__be32 *ret)
1912{
1913	struct nfsd4_session *session;
1914	__be32 status = nfserr_badsession;
1915
1916	session = __find_in_sessionid_hashtbl(sessionid, net);
1917	if (!session)
1918		goto out;
1919	status = nfsd4_get_session_locked(session);
1920	if (status)
1921		session = NULL;
1922out:
1923	*ret = status;
1924	return session;
1925}
1926
1927/* caller must hold client_lock */
1928static void
1929unhash_session(struct nfsd4_session *ses)
1930{
1931	struct nfs4_client *clp = ses->se_client;
1932	struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
1933
1934	lockdep_assert_held(&nn->client_lock);
1935
1936	list_del(&ses->se_hash);
1937	spin_lock(&ses->se_client->cl_lock);
1938	list_del(&ses->se_perclnt);
1939	spin_unlock(&ses->se_client->cl_lock);
1940}
1941
1942/* SETCLIENTID and SETCLIENTID_CONFIRM Helper functions */
1943static int
1944STALE_CLIENTID(clientid_t *clid, struct nfsd_net *nn)
1945{
1946	/*
1947	 * We're assuming the clid was not given out from a boot
1948	 * precisely 2^32 (about 136 years) before this one.  That seems
1949	 * a safe assumption:
1950	 */
1951	if (clid->cl_boot == (u32)nn->boot_time)
1952		return 0;
1953	trace_nfsd_clid_stale(clid);
1954	return 1;
1955}
1956
1957/*
1958 * XXX Should we use a slab cache ?
1959 * This type of memory management is somewhat inefficient, but we use it
1960 * anyway since SETCLIENTID is not a common operation.
1961 */
1962static struct nfs4_client *alloc_client(struct xdr_netobj name)
1963{
1964	struct nfs4_client *clp;
1965	int i;
1966
1967	clp = kmem_cache_zalloc(client_slab, GFP_KERNEL);
1968	if (clp == NULL)
1969		return NULL;
1970	xdr_netobj_dup(&clp->cl_name, &name, GFP_KERNEL);
1971	if (clp->cl_name.data == NULL)
1972		goto err_no_name;
1973	clp->cl_ownerstr_hashtbl = kmalloc_array(OWNER_HASH_SIZE,
1974						 sizeof(struct list_head),
1975						 GFP_KERNEL);
1976	if (!clp->cl_ownerstr_hashtbl)
1977		goto err_no_hashtbl;
1978	for (i = 0; i < OWNER_HASH_SIZE; i++)
1979		INIT_LIST_HEAD(&clp->cl_ownerstr_hashtbl[i]);
1980	INIT_LIST_HEAD(&clp->cl_sessions);
1981	idr_init(&clp->cl_stateids);
1982	atomic_set(&clp->cl_rpc_users, 0);
1983	clp->cl_cb_state = NFSD4_CB_UNKNOWN;
1984	INIT_LIST_HEAD(&clp->cl_idhash);
1985	INIT_LIST_HEAD(&clp->cl_openowners);
1986	INIT_LIST_HEAD(&clp->cl_delegations);
1987	INIT_LIST_HEAD(&clp->cl_lru);
1988	INIT_LIST_HEAD(&clp->cl_revoked);
1989#ifdef CONFIG_NFSD_PNFS
1990	INIT_LIST_HEAD(&clp->cl_lo_states);
1991#endif
1992	INIT_LIST_HEAD(&clp->async_copies);
1993	spin_lock_init(&clp->async_lock);
1994	spin_lock_init(&clp->cl_lock);
1995	rpc_init_wait_queue(&clp->cl_cb_waitq, "Backchannel slot table");
1996	return clp;
1997err_no_hashtbl:
1998	kfree(clp->cl_name.data);
1999err_no_name:
2000	kmem_cache_free(client_slab, clp);
2001	return NULL;
2002}
2003
2004static void __free_client(struct kref *k)
2005{
2006	struct nfsdfs_client *c = container_of(k, struct nfsdfs_client, cl_ref);
2007	struct nfs4_client *clp = container_of(c, struct nfs4_client, cl_nfsdfs);
2008
2009	free_svc_cred(&clp->cl_cred);
2010	kfree(clp->cl_ownerstr_hashtbl);
2011	kfree(clp->cl_name.data);
2012	kfree(clp->cl_nii_domain.data);
2013	kfree(clp->cl_nii_name.data);
2014	idr_destroy(&clp->cl_stateids);
2015	kmem_cache_free(client_slab, clp);
2016}
2017
2018static void drop_client(struct nfs4_client *clp)
2019{
2020	kref_put(&clp->cl_nfsdfs.cl_ref, __free_client);
2021}
2022
2023static void
2024free_client(struct nfs4_client *clp)
2025{
2026	while (!list_empty(&clp->cl_sessions)) {
2027		struct nfsd4_session *ses;
2028		ses = list_entry(clp->cl_sessions.next, struct nfsd4_session,
2029				se_perclnt);
2030		list_del(&ses->se_perclnt);
2031		WARN_ON_ONCE(atomic_read(&ses->se_ref));
2032		free_session(ses);
2033	}
2034	rpc_destroy_wait_queue(&clp->cl_cb_waitq);
2035	if (clp->cl_nfsd_dentry) {
2036		nfsd_client_rmdir(clp->cl_nfsd_dentry);
2037		clp->cl_nfsd_dentry = NULL;
2038		wake_up_all(&expiry_wq);
2039	}
2040	drop_client(clp);
2041}
2042
2043/* must be called under the client_lock */
2044static void
2045unhash_client_locked(struct nfs4_client *clp)
2046{
2047	struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
2048	struct nfsd4_session *ses;
2049
2050	lockdep_assert_held(&nn->client_lock);
2051
2052	/* Mark the client as expired! */
2053	clp->cl_time = 0;
2054	/* Make it invisible */
2055	if (!list_empty(&clp->cl_idhash)) {
2056		list_del_init(&clp->cl_idhash);
2057		if (test_bit(NFSD4_CLIENT_CONFIRMED, &clp->cl_flags))
2058			rb_erase(&clp->cl_namenode, &nn->conf_name_tree);
2059		else
2060			rb_erase(&clp->cl_namenode, &nn->unconf_name_tree);
2061	}
2062	list_del_init(&clp->cl_lru);
2063	spin_lock(&clp->cl_lock);
2064	list_for_each_entry(ses, &clp->cl_sessions, se_perclnt)
2065		list_del_init(&ses->se_hash);
2066	spin_unlock(&clp->cl_lock);
2067}
2068
2069static void
2070unhash_client(struct nfs4_client *clp)
2071{
2072	struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
2073
2074	spin_lock(&nn->client_lock);
2075	unhash_client_locked(clp);
2076	spin_unlock(&nn->client_lock);
2077}
2078
2079static __be32 mark_client_expired_locked(struct nfs4_client *clp)
2080{
2081	if (atomic_read(&clp->cl_rpc_users))
2082		return nfserr_jukebox;
2083	unhash_client_locked(clp);
2084	return nfs_ok;
2085}
2086
2087static void
2088__destroy_client(struct nfs4_client *clp)
2089{
2090	int i;
2091	struct nfs4_openowner *oo;
2092	struct nfs4_delegation *dp;
2093	struct list_head reaplist;
2094
2095	INIT_LIST_HEAD(&reaplist);
2096	spin_lock(&state_lock);
2097	while (!list_empty(&clp->cl_delegations)) {
2098		dp = list_entry(clp->cl_delegations.next, struct nfs4_delegation, dl_perclnt);
2099		WARN_ON(!unhash_delegation_locked(dp));
2100		list_add(&dp->dl_recall_lru, &reaplist);
2101	}
2102	spin_unlock(&state_lock);
2103	while (!list_empty(&reaplist)) {
2104		dp = list_entry(reaplist.next, struct nfs4_delegation, dl_recall_lru);
2105		list_del_init(&dp->dl_recall_lru);
2106		destroy_unhashed_deleg(dp);
2107	}
2108	while (!list_empty(&clp->cl_revoked)) {
2109		dp = list_entry(clp->cl_revoked.next, struct nfs4_delegation, dl_recall_lru);
2110		list_del_init(&dp->dl_recall_lru);
2111		nfs4_put_stid(&dp->dl_stid);
2112	}
2113	while (!list_empty(&clp->cl_openowners)) {
2114		oo = list_entry(clp->cl_openowners.next, struct nfs4_openowner, oo_perclient);
2115		nfs4_get_stateowner(&oo->oo_owner);
2116		release_openowner(oo);
2117	}
2118	for (i = 0; i < OWNER_HASH_SIZE; i++) {
2119		struct nfs4_stateowner *so, *tmp;
2120
2121		list_for_each_entry_safe(so, tmp, &clp->cl_ownerstr_hashtbl[i],
2122					 so_strhash) {
2123			/* Should be no openowners at this point */
2124			WARN_ON_ONCE(so->so_is_open_owner);
2125			remove_blocked_locks(lockowner(so));
2126		}
2127	}
2128	nfsd4_return_all_client_layouts(clp);
2129	nfsd4_shutdown_copy(clp);
2130	nfsd4_shutdown_callback(clp);
2131	if (clp->cl_cb_conn.cb_xprt)
2132		svc_xprt_put(clp->cl_cb_conn.cb_xprt);
2133	free_client(clp);
2134	wake_up_all(&expiry_wq);
2135}
2136
2137static void
2138destroy_client(struct nfs4_client *clp)
2139{
2140	unhash_client(clp);
2141	__destroy_client(clp);
2142}
2143
2144static void inc_reclaim_complete(struct nfs4_client *clp)
2145{
2146	struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
2147
2148	if (!nn->track_reclaim_completes)
2149		return;
2150	if (!nfsd4_find_reclaim_client(clp->cl_name, nn))
2151		return;
2152	if (atomic_inc_return(&nn->nr_reclaim_complete) ==
2153			nn->reclaim_str_hashtbl_size) {
2154		printk(KERN_INFO "NFSD: all clients done reclaiming, ending NFSv4 grace period (net %x)\n",
2155				clp->net->ns.inum);
2156		nfsd4_end_grace(nn);
2157	}
2158}
2159
2160static void expire_client(struct nfs4_client *clp)
2161{
2162	unhash_client(clp);
2163	nfsd4_client_record_remove(clp);
2164	__destroy_client(clp);
2165}
2166
2167static void copy_verf(struct nfs4_client *target, nfs4_verifier *source)
2168{
2169	memcpy(target->cl_verifier.data, source->data,
2170			sizeof(target->cl_verifier.data));
2171}
2172
2173static void copy_clid(struct nfs4_client *target, struct nfs4_client *source)
2174{
2175	target->cl_clientid.cl_boot = source->cl_clientid.cl_boot;
2176	target->cl_clientid.cl_id = source->cl_clientid.cl_id;
2177}
2178
2179static int copy_cred(struct svc_cred *target, struct svc_cred *source)
2180{
2181	target->cr_principal = kstrdup(source->cr_principal, GFP_KERNEL);
2182	target->cr_raw_principal = kstrdup(source->cr_raw_principal,
2183								GFP_KERNEL);
2184	target->cr_targ_princ = kstrdup(source->cr_targ_princ, GFP_KERNEL);
2185	if ((source->cr_principal && !target->cr_principal) ||
2186	    (source->cr_raw_principal && !target->cr_raw_principal) ||
2187	    (source->cr_targ_princ && !target->cr_targ_princ))
2188		return -ENOMEM;
2189
2190	target->cr_flavor = source->cr_flavor;
2191	target->cr_uid = source->cr_uid;
2192	target->cr_gid = source->cr_gid;
2193	target->cr_group_info = source->cr_group_info;
2194	get_group_info(target->cr_group_info);
2195	target->cr_gss_mech = source->cr_gss_mech;
2196	if (source->cr_gss_mech)
2197		gss_mech_get(source->cr_gss_mech);
2198	return 0;
2199}
2200
2201static int
2202compare_blob(const struct xdr_netobj *o1, const struct xdr_netobj *o2)
2203{
2204	if (o1->len < o2->len)
2205		return -1;
2206	if (o1->len > o2->len)
2207		return 1;
2208	return memcmp(o1->data, o2->data, o1->len);
2209}
2210
2211static int
2212same_verf(nfs4_verifier *v1, nfs4_verifier *v2)
2213{
2214	return 0 == memcmp(v1->data, v2->data, sizeof(v1->data));
2215}
2216
2217static int
2218same_clid(clientid_t *cl1, clientid_t *cl2)
2219{
2220	return (cl1->cl_boot == cl2->cl_boot) && (cl1->cl_id == cl2->cl_id);
2221}
2222
2223static bool groups_equal(struct group_info *g1, struct group_info *g2)
2224{
2225	int i;
2226
2227	if (g1->ngroups != g2->ngroups)
2228		return false;
2229	for (i=0; i<g1->ngroups; i++)
2230		if (!gid_eq(g1->gid[i], g2->gid[i]))
2231			return false;
2232	return true;
2233}
2234
2235/*
2236 * RFC 3530 language requires clid_inuse be returned when the
2237 * "principal" associated with a requests differs from that previously
2238 * used.  We use uid, gid's, and gss principal string as our best
2239 * approximation.  We also don't want to allow non-gss use of a client
2240 * established using gss: in theory cr_principal should catch that
2241 * change, but in practice cr_principal can be null even in the gss case
2242 * since gssd doesn't always pass down a principal string.
2243 */
2244static bool is_gss_cred(struct svc_cred *cr)
2245{
2246	/* Is cr_flavor one of the gss "pseudoflavors"?: */
2247	return (cr->cr_flavor > RPC_AUTH_MAXFLAVOR);
2248}
2249
2250
2251static bool
2252same_creds(struct svc_cred *cr1, struct svc_cred *cr2)
2253{
2254	if ((is_gss_cred(cr1) != is_gss_cred(cr2))
2255		|| (!uid_eq(cr1->cr_uid, cr2->cr_uid))
2256		|| (!gid_eq(cr1->cr_gid, cr2->cr_gid))
2257		|| !groups_equal(cr1->cr_group_info, cr2->cr_group_info))
2258		return false;
2259	/* XXX: check that cr_targ_princ fields match ? */
2260	if (cr1->cr_principal == cr2->cr_principal)
2261		return true;
2262	if (!cr1->cr_principal || !cr2->cr_principal)
2263		return false;
2264	return 0 == strcmp(cr1->cr_principal, cr2->cr_principal);
2265}
2266
2267static bool svc_rqst_integrity_protected(struct svc_rqst *rqstp)
2268{
2269	struct svc_cred *cr = &rqstp->rq_cred;
2270	u32 service;
2271
2272	if (!cr->cr_gss_mech)
2273		return false;
2274	service = gss_pseudoflavor_to_service(cr->cr_gss_mech, cr->cr_flavor);
2275	return service == RPC_GSS_SVC_INTEGRITY ||
2276	       service == RPC_GSS_SVC_PRIVACY;
2277}
2278
2279bool nfsd4_mach_creds_match(struct nfs4_client *cl, struct svc_rqst *rqstp)
2280{
2281	struct svc_cred *cr = &rqstp->rq_cred;
2282
2283	if (!cl->cl_mach_cred)
2284		return true;
2285	if (cl->cl_cred.cr_gss_mech != cr->cr_gss_mech)
2286		return false;
2287	if (!svc_rqst_integrity_protected(rqstp))
2288		return false;
2289	if (cl->cl_cred.cr_raw_principal)
2290		return 0 == strcmp(cl->cl_cred.cr_raw_principal,
2291						cr->cr_raw_principal);
2292	if (!cr->cr_principal)
2293		return false;
2294	return 0 == strcmp(cl->cl_cred.cr_principal, cr->cr_principal);
2295}
2296
2297static void gen_confirm(struct nfs4_client *clp, struct nfsd_net *nn)
2298{
2299	__be32 verf[2];
2300
2301	/*
2302	 * This is opaque to client, so no need to byte-swap. Use
2303	 * __force to keep sparse happy
2304	 */
2305	verf[0] = (__force __be32)(u32)ktime_get_real_seconds();
2306	verf[1] = (__force __be32)nn->clverifier_counter++;
2307	memcpy(clp->cl_confirm.data, verf, sizeof(clp->cl_confirm.data));
2308}
2309
2310static void gen_clid(struct nfs4_client *clp, struct nfsd_net *nn)
2311{
2312	clp->cl_clientid.cl_boot = (u32)nn->boot_time;
2313	clp->cl_clientid.cl_id = nn->clientid_counter++;
2314	gen_confirm(clp, nn);
2315}
2316
2317static struct nfs4_stid *
2318find_stateid_locked(struct nfs4_client *cl, stateid_t *t)
2319{
2320	struct nfs4_stid *ret;
2321
2322	ret = idr_find(&cl->cl_stateids, t->si_opaque.so_id);
2323	if (!ret || !ret->sc_type)
2324		return NULL;
2325	return ret;
2326}
2327
2328static struct nfs4_stid *
2329find_stateid_by_type(struct nfs4_client *cl, stateid_t *t, char typemask)
2330{
2331	struct nfs4_stid *s;
2332
2333	spin_lock(&cl->cl_lock);
2334	s = find_stateid_locked(cl, t);
2335	if (s != NULL) {
2336		if (typemask & s->sc_type)
2337			refcount_inc(&s->sc_count);
2338		else
2339			s = NULL;
2340	}
2341	spin_unlock(&cl->cl_lock);
2342	return s;
2343}
2344
2345static struct nfs4_client *get_nfsdfs_clp(struct inode *inode)
2346{
2347	struct nfsdfs_client *nc;
2348	nc = get_nfsdfs_client(inode);
2349	if (!nc)
2350		return NULL;
2351	return container_of(nc, struct nfs4_client, cl_nfsdfs);
2352}
2353
2354static void seq_quote_mem(struct seq_file *m, char *data, int len)
2355{
2356	seq_printf(m, "\"");
2357	seq_escape_mem(m, data, len, ESCAPE_HEX | ESCAPE_NAP | ESCAPE_APPEND, "\"\\");
2358	seq_printf(m, "\"");
2359}
2360
2361static const char *cb_state2str(int state)
2362{
2363	switch (state) {
2364	case NFSD4_CB_UP:
2365		return "UP";
2366	case NFSD4_CB_UNKNOWN:
2367		return "UNKNOWN";
2368	case NFSD4_CB_DOWN:
2369		return "DOWN";
2370	case NFSD4_CB_FAULT:
2371		return "FAULT";
2372	}
2373	return "UNDEFINED";
2374}
2375
2376static int client_info_show(struct seq_file *m, void *v)
2377{
2378	struct inode *inode = m->private;
2379	struct nfs4_client *clp;
2380	u64 clid;
2381
2382	clp = get_nfsdfs_clp(inode);
2383	if (!clp)
2384		return -ENXIO;
2385	memcpy(&clid, &clp->cl_clientid, sizeof(clid));
2386	seq_printf(m, "clientid: 0x%llx\n", clid);
2387	seq_printf(m, "address: \"%pISpc\"\n", (struct sockaddr *)&clp->cl_addr);
2388	if (test_bit(NFSD4_CLIENT_CONFIRMED, &clp->cl_flags))
2389		seq_puts(m, "status: confirmed\n");
2390	else
2391		seq_puts(m, "status: unconfirmed\n");
2392	seq_printf(m, "name: ");
2393	seq_quote_mem(m, clp->cl_name.data, clp->cl_name.len);
2394	seq_printf(m, "\nminor version: %d\n", clp->cl_minorversion);
2395	if (clp->cl_nii_domain.data) {
2396		seq_printf(m, "Implementation domain: ");
2397		seq_quote_mem(m, clp->cl_nii_domain.data,
2398					clp->cl_nii_domain.len);
2399		seq_printf(m, "\nImplementation name: ");
2400		seq_quote_mem(m, clp->cl_nii_name.data, clp->cl_nii_name.len);
2401		seq_printf(m, "\nImplementation time: [%lld, %ld]\n",
2402			clp->cl_nii_time.tv_sec, clp->cl_nii_time.tv_nsec);
2403	}
2404	seq_printf(m, "callback state: %s\n", cb_state2str(clp->cl_cb_state));
2405	seq_printf(m, "callback address: %pISpc\n", &clp->cl_cb_conn.cb_addr);
2406	drop_client(clp);
2407
2408	return 0;
2409}
2410
2411static int client_info_open(struct inode *inode, struct file *file)
2412{
2413	return single_open(file, client_info_show, inode);
2414}
2415
2416static const struct file_operations client_info_fops = {
2417	.open		= client_info_open,
2418	.read		= seq_read,
2419	.llseek		= seq_lseek,
2420	.release	= single_release,
2421};
2422
2423static void *states_start(struct seq_file *s, loff_t *pos)
2424	__acquires(&clp->cl_lock)
2425{
2426	struct nfs4_client *clp = s->private;
2427	unsigned long id = *pos;
2428	void *ret;
2429
2430	spin_lock(&clp->cl_lock);
2431	ret = idr_get_next_ul(&clp->cl_stateids, &id);
2432	*pos = id;
2433	return ret;
2434}
2435
2436static void *states_next(struct seq_file *s, void *v, loff_t *pos)
2437{
2438	struct nfs4_client *clp = s->private;
2439	unsigned long id = *pos;
2440	void *ret;
2441
2442	id = *pos;
2443	id++;
2444	ret = idr_get_next_ul(&clp->cl_stateids, &id);
2445	*pos = id;
2446	return ret;
2447}
2448
2449static void states_stop(struct seq_file *s, void *v)
2450	__releases(&clp->cl_lock)
2451{
2452	struct nfs4_client *clp = s->private;
2453
2454	spin_unlock(&clp->cl_lock);
2455}
2456
2457static void nfs4_show_fname(struct seq_file *s, struct nfsd_file *f)
2458{
2459         seq_printf(s, "filename: \"%pD2\"", f->nf_file);
2460}
2461
2462static void nfs4_show_superblock(struct seq_file *s, struct nfsd_file *f)
2463{
2464	struct inode *inode = f->nf_inode;
2465
2466	seq_printf(s, "superblock: \"%02x:%02x:%ld\"",
2467					MAJOR(inode->i_sb->s_dev),
2468					 MINOR(inode->i_sb->s_dev),
2469					 inode->i_ino);
2470}
2471
2472static void nfs4_show_owner(struct seq_file *s, struct nfs4_stateowner *oo)
2473{
2474	seq_printf(s, "owner: ");
2475	seq_quote_mem(s, oo->so_owner.data, oo->so_owner.len);
2476}
2477
2478static void nfs4_show_stateid(struct seq_file *s, stateid_t *stid)
2479{
2480	seq_printf(s, "0x%.8x", stid->si_generation);
2481	seq_printf(s, "%12phN", &stid->si_opaque);
2482}
2483
2484static int nfs4_show_open(struct seq_file *s, struct nfs4_stid *st)
2485{
2486	struct nfs4_ol_stateid *ols;
2487	struct nfs4_file *nf;
2488	struct nfsd_file *file;
2489	struct nfs4_stateowner *oo;
2490	unsigned int access, deny;
2491
2492	if (st->sc_type != NFS4_OPEN_STID && st->sc_type != NFS4_LOCK_STID)
2493		return 0; /* XXX: or SEQ_SKIP? */
2494	ols = openlockstateid(st);
2495	oo = ols->st_stateowner;
2496	nf = st->sc_file;
2497	file = find_any_file(nf);
2498	if (!file)
2499		return 0;
2500
2501	seq_printf(s, "- ");
2502	nfs4_show_stateid(s, &st->sc_stateid);
2503	seq_printf(s, ": { type: open, ");
2504
2505	access = bmap_to_share_mode(ols->st_access_bmap);
2506	deny   = bmap_to_share_mode(ols->st_deny_bmap);
2507
2508	seq_printf(s, "access: %s%s, ",
2509		access & NFS4_SHARE_ACCESS_READ ? "r" : "-",
2510		access & NFS4_SHARE_ACCESS_WRITE ? "w" : "-");
2511	seq_printf(s, "deny: %s%s, ",
2512		deny & NFS4_SHARE_ACCESS_READ ? "r" : "-",
2513		deny & NFS4_SHARE_ACCESS_WRITE ? "w" : "-");
2514
2515	nfs4_show_superblock(s, file);
2516	seq_printf(s, ", ");
2517	nfs4_show_fname(s, file);
2518	seq_printf(s, ", ");
2519	nfs4_show_owner(s, oo);
2520	seq_printf(s, " }\n");
2521	nfsd_file_put(file);
2522
2523	return 0;
2524}
2525
2526static int nfs4_show_lock(struct seq_file *s, struct nfs4_stid *st)
2527{
2528	struct nfs4_ol_stateid *ols;
2529	struct nfs4_file *nf;
2530	struct nfsd_file *file;
2531	struct nfs4_stateowner *oo;
2532
2533	ols = openlockstateid(st);
2534	oo = ols->st_stateowner;
2535	nf = st->sc_file;
2536	file = find_any_file(nf);
2537	if (!file)
2538		return 0;
2539
2540	seq_printf(s, "- ");
2541	nfs4_show_stateid(s, &st->sc_stateid);
2542	seq_printf(s, ": { type: lock, ");
2543
2544	/*
2545	 * Note: a lock stateid isn't really the same thing as a lock,
2546	 * it's the locking state held by one owner on a file, and there
2547	 * may be multiple (or no) lock ranges associated with it.
2548	 * (Same for the matter is true of open stateids.)
2549	 */
2550
2551	nfs4_show_superblock(s, file);
2552	/* XXX: open stateid? */
2553	seq_printf(s, ", ");
2554	nfs4_show_fname(s, file);
2555	seq_printf(s, ", ");
2556	nfs4_show_owner(s, oo);
2557	seq_printf(s, " }\n");
2558	nfsd_file_put(file);
2559
2560	return 0;
2561}
2562
2563static int nfs4_show_deleg(struct seq_file *s, struct nfs4_stid *st)
2564{
2565	struct nfs4_delegation *ds;
2566	struct nfs4_file *nf;
2567	struct nfsd_file *file;
2568
2569	ds = delegstateid(st);
2570	nf = st->sc_file;
2571	file = find_deleg_file(nf);
2572	if (!file)
2573		return 0;
2574
2575	seq_printf(s, "- ");
2576	nfs4_show_stateid(s, &st->sc_stateid);
2577	seq_printf(s, ": { type: deleg, ");
2578
2579	/* Kinda dead code as long as we only support read delegs: */
2580	seq_printf(s, "access: %s, ",
2581		ds->dl_type == NFS4_OPEN_DELEGATE_READ ? "r" : "w");
2582
2583	/* XXX: lease time, whether it's being recalled. */
2584
2585	nfs4_show_superblock(s, file);
2586	seq_printf(s, ", ");
2587	nfs4_show_fname(s, file);
2588	seq_printf(s, " }\n");
2589	nfsd_file_put(file);
2590
2591	return 0;
2592}
2593
2594static int nfs4_show_layout(struct seq_file *s, struct nfs4_stid *st)
2595{
2596	struct nfs4_layout_stateid *ls;
2597	struct nfsd_file *file;
2598
2599	ls = container_of(st, struct nfs4_layout_stateid, ls_stid);
2600	file = ls->ls_file;
2601
2602	seq_printf(s, "- ");
2603	nfs4_show_stateid(s, &st->sc_stateid);
2604	seq_printf(s, ": { type: layout, ");
2605
2606	/* XXX: What else would be useful? */
2607
2608	nfs4_show_superblock(s, file);
2609	seq_printf(s, ", ");
2610	nfs4_show_fname(s, file);
2611	seq_printf(s, " }\n");
2612
2613	return 0;
2614}
2615
2616static int states_show(struct seq_file *s, void *v)
2617{
2618	struct nfs4_stid *st = v;
2619
2620	switch (st->sc_type) {
2621	case NFS4_OPEN_STID:
2622		return nfs4_show_open(s, st);
2623	case NFS4_LOCK_STID:
2624		return nfs4_show_lock(s, st);
2625	case NFS4_DELEG_STID:
2626		return nfs4_show_deleg(s, st);
2627	case NFS4_LAYOUT_STID:
2628		return nfs4_show_layout(s, st);
2629	default:
2630		return 0; /* XXX: or SEQ_SKIP? */
2631	}
2632	/* XXX: copy stateids? */
2633}
2634
2635static struct seq_operations states_seq_ops = {
2636	.start = states_start,
2637	.next = states_next,
2638	.stop = states_stop,
2639	.show = states_show
2640};
2641
2642static int client_states_open(struct inode *inode, struct file *file)
2643{
2644	struct seq_file *s;
2645	struct nfs4_client *clp;
2646	int ret;
2647
2648	clp = get_nfsdfs_clp(inode);
2649	if (!clp)
2650		return -ENXIO;
2651
2652	ret = seq_open(file, &states_seq_ops);
2653	if (ret)
2654		return ret;
2655	s = file->private_data;
2656	s->private = clp;
2657	return 0;
2658}
2659
2660static int client_opens_release(struct inode *inode, struct file *file)
2661{
2662	struct seq_file *m = file->private_data;
2663	struct nfs4_client *clp = m->private;
2664
2665	/* XXX: alternatively, we could get/drop in seq start/stop */
2666	drop_client(clp);
2667	return 0;
2668}
2669
2670static const struct file_operations client_states_fops = {
2671	.open		= client_states_open,
2672	.read		= seq_read,
2673	.llseek		= seq_lseek,
2674	.release	= client_opens_release,
2675};
2676
2677/*
2678 * Normally we refuse to destroy clients that are in use, but here the
2679 * administrator is telling us to just do it.  We also want to wait
2680 * so the caller has a guarantee that the client's locks are gone by
2681 * the time the write returns:
2682 */
2683static void force_expire_client(struct nfs4_client *clp)
2684{
2685	struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
2686	bool already_expired;
2687
2688	trace_nfsd_clid_admin_expired(&clp->cl_clientid);
2689
2690	spin_lock(&nn->client_lock);
2691	clp->cl_time = 0;
2692	spin_unlock(&nn->client_lock);
2693
2694	wait_event(expiry_wq, atomic_read(&clp->cl_rpc_users) == 0);
2695	spin_lock(&nn->client_lock);
2696	already_expired = list_empty(&clp->cl_lru);
2697	if (!already_expired)
2698		unhash_client_locked(clp);
2699	spin_unlock(&nn->client_lock);
2700
2701	if (!already_expired)
2702		expire_client(clp);
2703	else
2704		wait_event(expiry_wq, clp->cl_nfsd_dentry == NULL);
2705}
2706
2707static ssize_t client_ctl_write(struct file *file, const char __user *buf,
2708				   size_t size, loff_t *pos)
2709{
2710	char *data;
2711	struct nfs4_client *clp;
2712
2713	data = simple_transaction_get(file, buf, size);
2714	if (IS_ERR(data))
2715		return PTR_ERR(data);
2716	if (size != 7 || 0 != memcmp(data, "expire\n", 7))
2717		return -EINVAL;
2718	clp = get_nfsdfs_clp(file_inode(file));
2719	if (!clp)
2720		return -ENXIO;
2721	force_expire_client(clp);
2722	drop_client(clp);
2723	return 7;
2724}
2725
2726static const struct file_operations client_ctl_fops = {
2727	.write		= client_ctl_write,
2728	.release	= simple_transaction_release,
2729};
2730
2731static const struct tree_descr client_files[] = {
2732	[0] = {"info", &client_info_fops, S_IRUSR},
2733	[1] = {"states", &client_states_fops, S_IRUSR},
2734	[2] = {"ctl", &client_ctl_fops, S_IWUSR},
2735	[3] = {""},
2736};
2737
2738static struct nfs4_client *create_client(struct xdr_netobj name,
2739		struct svc_rqst *rqstp, nfs4_verifier *verf)
2740{
2741	struct nfs4_client *clp;
2742	struct sockaddr *sa = svc_addr(rqstp);
2743	int ret;
2744	struct net *net = SVC_NET(rqstp);
2745	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
2746	struct dentry *dentries[ARRAY_SIZE(client_files)];
2747
2748	clp = alloc_client(name);
2749	if (clp == NULL)
2750		return NULL;
2751
2752	ret = copy_cred(&clp->cl_cred, &rqstp->rq_cred);
2753	if (ret) {
2754		free_client(clp);
2755		return NULL;
2756	}
2757	gen_clid(clp, nn);
2758	kref_init(&clp->cl_nfsdfs.cl_ref);
2759	nfsd4_init_cb(&clp->cl_cb_null, clp, NULL, NFSPROC4_CLNT_CB_NULL);
2760	clp->cl_time = ktime_get_boottime_seconds();
2761	clear_bit(0, &clp->cl_cb_slot_busy);
2762	copy_verf(clp, verf);
2763	memcpy(&clp->cl_addr, sa, sizeof(struct sockaddr_storage));
2764	clp->cl_cb_session = NULL;
2765	clp->net = net;
2766	clp->cl_nfsd_dentry = nfsd_client_mkdir(
2767		nn, &clp->cl_nfsdfs,
2768		clp->cl_clientid.cl_id - nn->clientid_base,
2769		client_files, dentries);
2770	clp->cl_nfsd_info_dentry = dentries[0];
2771	if (!clp->cl_nfsd_dentry) {
2772		free_client(clp);
2773		return NULL;
2774	}
2775	return clp;
2776}
2777
2778static void
2779add_clp_to_name_tree(struct nfs4_client *new_clp, struct rb_root *root)
2780{
2781	struct rb_node **new = &(root->rb_node), *parent = NULL;
2782	struct nfs4_client *clp;
2783
2784	while (*new) {
2785		clp = rb_entry(*new, struct nfs4_client, cl_namenode);
2786		parent = *new;
2787
2788		if (compare_blob(&clp->cl_name, &new_clp->cl_name) > 0)
2789			new = &((*new)->rb_left);
2790		else
2791			new = &((*new)->rb_right);
2792	}
2793
2794	rb_link_node(&new_clp->cl_namenode, parent, new);
2795	rb_insert_color(&new_clp->cl_namenode, root);
2796}
2797
2798static struct nfs4_client *
2799find_clp_in_name_tree(struct xdr_netobj *name, struct rb_root *root)
2800{
2801	int cmp;
2802	struct rb_node *node = root->rb_node;
2803	struct nfs4_client *clp;
2804
2805	while (node) {
2806		clp = rb_entry(node, struct nfs4_client, cl_namenode);
2807		cmp = compare_blob(&clp->cl_name, name);
2808		if (cmp > 0)
2809			node = node->rb_left;
2810		else if (cmp < 0)
2811			node = node->rb_right;
2812		else
2813			return clp;
2814	}
2815	return NULL;
2816}
2817
2818static void
2819add_to_unconfirmed(struct nfs4_client *clp)
2820{
2821	unsigned int idhashval;
2822	struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
2823
2824	lockdep_assert_held(&nn->client_lock);
2825
2826	clear_bit(NFSD4_CLIENT_CONFIRMED, &clp->cl_flags);
2827	add_clp_to_name_tree(clp, &nn->unconf_name_tree);
2828	idhashval = clientid_hashval(clp->cl_clientid.cl_id);
2829	list_add(&clp->cl_idhash, &nn->unconf_id_hashtbl[idhashval]);
2830	renew_client_locked(clp);
2831}
2832
2833static void
2834move_to_confirmed(struct nfs4_client *clp)
2835{
2836	unsigned int idhashval = clientid_hashval(clp->cl_clientid.cl_id);
2837	struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
2838
2839	lockdep_assert_held(&nn->client_lock);
2840
2841	list_move(&clp->cl_idhash, &nn->conf_id_hashtbl[idhashval]);
2842	rb_erase(&clp->cl_namenode, &nn->unconf_name_tree);
2843	add_clp_to_name_tree(clp, &nn->conf_name_tree);
2844	set_bit(NFSD4_CLIENT_CONFIRMED, &clp->cl_flags);
2845	trace_nfsd_clid_confirmed(&clp->cl_clientid);
2846	renew_client_locked(clp);
2847}
2848
2849static struct nfs4_client *
2850find_client_in_id_table(struct list_head *tbl, clientid_t *clid, bool sessions)
2851{
2852	struct nfs4_client *clp;
2853	unsigned int idhashval = clientid_hashval(clid->cl_id);
2854
2855	list_for_each_entry(clp, &tbl[idhashval], cl_idhash) {
2856		if (same_clid(&clp->cl_clientid, clid)) {
2857			if ((bool)clp->cl_minorversion != sessions)
2858				return NULL;
2859			renew_client_locked(clp);
2860			return clp;
2861		}
2862	}
2863	return NULL;
2864}
2865
2866static struct nfs4_client *
2867find_confirmed_client(clientid_t *clid, bool sessions, struct nfsd_net *nn)
2868{
2869	struct list_head *tbl = nn->conf_id_hashtbl;
2870
2871	lockdep_assert_held(&nn->client_lock);
2872	return find_client_in_id_table(tbl, clid, sessions);
2873}
2874
2875static struct nfs4_client *
2876find_unconfirmed_client(clientid_t *clid, bool sessions, struct nfsd_net *nn)
2877{
2878	struct list_head *tbl = nn->unconf_id_hashtbl;
2879
2880	lockdep_assert_held(&nn->client_lock);
2881	return find_client_in_id_table(tbl, clid, sessions);
2882}
2883
2884static bool clp_used_exchangeid(struct nfs4_client *clp)
2885{
2886	return clp->cl_exchange_flags != 0;
2887}
2888
2889static struct nfs4_client *
2890find_confirmed_client_by_name(struct xdr_netobj *name, struct nfsd_net *nn)
2891{
2892	lockdep_assert_held(&nn->client_lock);
2893	return find_clp_in_name_tree(name, &nn->conf_name_tree);
2894}
2895
2896static struct nfs4_client *
2897find_unconfirmed_client_by_name(struct xdr_netobj *name, struct nfsd_net *nn)
2898{
2899	lockdep_assert_held(&nn->client_lock);
2900	return find_clp_in_name_tree(name, &nn->unconf_name_tree);
2901}
2902
2903static void
2904gen_callback(struct nfs4_client *clp, struct nfsd4_setclientid *se, struct svc_rqst *rqstp)
2905{
2906	struct nfs4_cb_conn *conn = &clp->cl_cb_conn;
2907	struct sockaddr	*sa = svc_addr(rqstp);
2908	u32 scopeid = rpc_get_scope_id(sa);
2909	unsigned short expected_family;
2910
2911	/* Currently, we only support tcp and tcp6 for the callback channel */
2912	if (se->se_callback_netid_len == 3 &&
2913	    !memcmp(se->se_callback_netid_val, "tcp", 3))
2914		expected_family = AF_INET;
2915	else if (se->se_callback_netid_len == 4 &&
2916		 !memcmp(se->se_callback_netid_val, "tcp6", 4))
2917		expected_family = AF_INET6;
2918	else
2919		goto out_err;
2920
2921	conn->cb_addrlen = rpc_uaddr2sockaddr(clp->net, se->se_callback_addr_val,
2922					    se->se_callback_addr_len,
2923					    (struct sockaddr *)&conn->cb_addr,
2924					    sizeof(conn->cb_addr));
2925
2926	if (!conn->cb_addrlen || conn->cb_addr.ss_family != expected_family)
2927		goto out_err;
2928
2929	if (conn->cb_addr.ss_family == AF_INET6)
2930		((struct sockaddr_in6 *)&conn->cb_addr)->sin6_scope_id = scopeid;
2931
2932	conn->cb_prog = se->se_callback_prog;
2933	conn->cb_ident = se->se_callback_ident;
2934	memcpy(&conn->cb_saddr, &rqstp->rq_daddr, rqstp->rq_daddrlen);
2935	trace_nfsd_cb_args(clp, conn);
2936	return;
2937out_err:
2938	conn->cb_addr.ss_family = AF_UNSPEC;
2939	conn->cb_addrlen = 0;
2940	trace_nfsd_cb_nodelegs(clp);
2941	return;
2942}
2943
2944/*
2945 * Cache a reply. nfsd4_check_resp_size() has bounded the cache size.
2946 */
2947static void
2948nfsd4_store_cache_entry(struct nfsd4_compoundres *resp)
2949{
2950	struct xdr_buf *buf = resp->xdr->buf;
2951	struct nfsd4_slot *slot = resp->cstate.slot;
2952	unsigned int base;
2953
2954	dprintk("--> %s slot %p\n", __func__, slot);
2955
2956	slot->sl_flags |= NFSD4_SLOT_INITIALIZED;
2957	slot->sl_opcnt = resp->opcnt;
2958	slot->sl_status = resp->cstate.status;
2959	free_svc_cred(&slot->sl_cred);
2960	copy_cred(&slot->sl_cred, &resp->rqstp->rq_cred);
2961
2962	if (!nfsd4_cache_this(resp)) {
2963		slot->sl_flags &= ~NFSD4_SLOT_CACHED;
2964		return;
2965	}
2966	slot->sl_flags |= NFSD4_SLOT_CACHED;
2967
2968	base = resp->cstate.data_offset;
2969	slot->sl_datalen = buf->len - base;
2970	if (read_bytes_from_xdr_buf(buf, base, slot->sl_data, slot->sl_datalen))
2971		WARN(1, "%s: sessions DRC could not cache compound\n",
2972		     __func__);
2973	return;
2974}
2975
2976/*
2977 * Encode the replay sequence operation from the slot values.
2978 * If cachethis is FALSE encode the uncached rep error on the next
2979 * operation which sets resp->p and increments resp->opcnt for
2980 * nfs4svc_encode_compoundres.
2981 *
2982 */
2983static __be32
2984nfsd4_enc_sequence_replay(struct nfsd4_compoundargs *args,
2985			  struct nfsd4_compoundres *resp)
2986{
2987	struct nfsd4_op *op;
2988	struct nfsd4_slot *slot = resp->cstate.slot;
2989
2990	/* Encode the replayed sequence operation */
2991	op = &args->ops[resp->opcnt - 1];
2992	nfsd4_encode_operation(resp, op);
2993
2994	if (slot->sl_flags & NFSD4_SLOT_CACHED)
2995		return op->status;
2996	if (args->opcnt == 1) {
2997		/*
2998		 * The original operation wasn't a solo sequence--we
2999		 * always cache those--so this retry must not match the
3000		 * original:
3001		 */
3002		op->status = nfserr_seq_false_retry;
3003	} else {
3004		op = &args->ops[resp->opcnt++];
3005		op->status = nfserr_retry_uncached_rep;
3006		nfsd4_encode_operation(resp, op);
3007	}
3008	return op->status;
3009}
3010
3011/*
3012 * The sequence operation is not cached because we can use the slot and
3013 * session values.
3014 */
3015static __be32
3016nfsd4_replay_cache_entry(struct nfsd4_compoundres *resp,
3017			 struct nfsd4_sequence *seq)
3018{
3019	struct nfsd4_slot *slot = resp->cstate.slot;
3020	struct xdr_stream *xdr = resp->xdr;
3021	__be32 *p;
3022	__be32 status;
3023
3024	dprintk("--> %s slot %p\n", __func__, slot);
3025
3026	status = nfsd4_enc_sequence_replay(resp->rqstp->rq_argp, resp);
3027	if (status)
3028		return status;
3029
3030	p = xdr_reserve_space(xdr, slot->sl_datalen);
3031	if (!p) {
3032		WARN_ON_ONCE(1);
3033		return nfserr_serverfault;
3034	}
3035	xdr_encode_opaque_fixed(p, slot->sl_data, slot->sl_datalen);
3036	xdr_commit_encode(xdr);
3037
3038	resp->opcnt = slot->sl_opcnt;
3039	return slot->sl_status;
3040}
3041
3042/*
3043 * Set the exchange_id flags returned by the server.
3044 */
3045static void
3046nfsd4_set_ex_flags(struct nfs4_client *new, struct nfsd4_exchange_id *clid)
3047{
3048#ifdef CONFIG_NFSD_PNFS
3049	new->cl_exchange_flags |= EXCHGID4_FLAG_USE_PNFS_MDS;
3050#else
3051	new->cl_exchange_flags |= EXCHGID4_FLAG_USE_NON_PNFS;
3052#endif
3053
3054	/* Referrals are supported, Migration is not. */
3055	new->cl_exchange_flags |= EXCHGID4_FLAG_SUPP_MOVED_REFER;
3056
3057	/* set the wire flags to return to client. */
3058	clid->flags = new->cl_exchange_flags;
3059}
3060
3061static bool client_has_openowners(struct nfs4_client *clp)
3062{
3063	struct nfs4_openowner *oo;
3064
3065	list_for_each_entry(oo, &clp->cl_openowners, oo_perclient) {
3066		if (!list_empty(&oo->oo_owner.so_stateids))
3067			return true;
3068	}
3069	return false;
3070}
3071
3072static bool client_has_state(struct nfs4_client *clp)
3073{
3074	return client_has_openowners(clp)
3075#ifdef CONFIG_NFSD_PNFS
3076		|| !list_empty(&clp->cl_lo_states)
3077#endif
3078		|| !list_empty(&clp->cl_delegations)
3079		|| !list_empty(&clp->cl_sessions)
3080		|| !list_empty(&clp->async_copies);
3081}
3082
3083static __be32 copy_impl_id(struct nfs4_client *clp,
3084				struct nfsd4_exchange_id *exid)
3085{
3086	if (!exid->nii_domain.data)
3087		return 0;
3088	xdr_netobj_dup(&clp->cl_nii_domain, &exid->nii_domain, GFP_KERNEL);
3089	if (!clp->cl_nii_domain.data)
3090		return nfserr_jukebox;
3091	xdr_netobj_dup(&clp->cl_nii_name, &exid->nii_name, GFP_KERNEL);
3092	if (!clp->cl_nii_name.data)
3093		return nfserr_jukebox;
3094	clp->cl_nii_time = exid->nii_time;
3095	return 0;
3096}
3097
3098__be32
3099nfsd4_exchange_id(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3100		union nfsd4_op_u *u)
3101{
3102	struct nfsd4_exchange_id *exid = &u->exchange_id;
3103	struct nfs4_client *conf, *new;
3104	struct nfs4_client *unconf = NULL;
3105	__be32 status;
3106	char			addr_str[INET6_ADDRSTRLEN];
3107	nfs4_verifier		verf = exid->verifier;
3108	struct sockaddr		*sa = svc_addr(rqstp);
3109	bool	update = exid->flags & EXCHGID4_FLAG_UPD_CONFIRMED_REC_A;
3110	struct nfsd_net		*nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
3111
3112	rpc_ntop(sa, addr_str, sizeof(addr_str));
3113	dprintk("%s rqstp=%p exid=%p clname.len=%u clname.data=%p "
3114		"ip_addr=%s flags %x, spa_how %u\n",
3115		__func__, rqstp, exid, exid->clname.len, exid->clname.data,
3116		addr_str, exid->flags, exid->spa_how);
3117
3118	if (exid->flags & ~EXCHGID4_FLAG_MASK_A)
3119		return nfserr_inval;
3120
3121	new = create_client(exid->clname, rqstp, &verf);
3122	if (new == NULL)
3123		return nfserr_jukebox;
3124	status = copy_impl_id(new, exid);
3125	if (status)
3126		goto out_nolock;
3127
3128	switch (exid->spa_how) {
3129	case SP4_MACH_CRED:
3130		exid->spo_must_enforce[0] = 0;
3131		exid->spo_must_enforce[1] = (
3132			1 << (OP_BIND_CONN_TO_SESSION - 32) |
3133			1 << (OP_EXCHANGE_ID - 32) |
3134			1 << (OP_CREATE_SESSION - 32) |
3135			1 << (OP_DESTROY_SESSION - 32) |
3136			1 << (OP_DESTROY_CLIENTID - 32));
3137
3138		exid->spo_must_allow[0] &= (1 << (OP_CLOSE) |
3139					1 << (OP_OPEN_DOWNGRADE) |
3140					1 << (OP_LOCKU) |
3141					1 << (OP_DELEGRETURN));
3142
3143		exid->spo_must_allow[1] &= (
3144					1 << (OP_TEST_STATEID - 32) |
3145					1 << (OP_FREE_STATEID - 32));
3146		if (!svc_rqst_integrity_protected(rqstp)) {
3147			status = nfserr_inval;
3148			goto out_nolock;
3149		}
3150		/*
3151		 * Sometimes userspace doesn't give us a principal.
3152		 * Which is a bug, really.  Anyway, we can't enforce
3153		 * MACH_CRED in that case, better to give up now:
3154		 */
3155		if (!new->cl_cred.cr_principal &&
3156					!new->cl_cred.cr_raw_principal) {
3157			status = nfserr_serverfault;
3158			goto out_nolock;
3159		}
3160		new->cl_mach_cred = true;
3161		break;
3162	case SP4_NONE:
3163		break;
3164	default:				/* checked by xdr code */
3165		WARN_ON_ONCE(1);
3166		fallthrough;
3167	case SP4_SSV:
3168		status = nfserr_encr_alg_unsupp;
3169		goto out_nolock;
3170	}
3171
3172	/* Cases below refer to rfc 5661 section 18.35.4: */
3173	spin_lock(&nn->client_lock);
3174	conf = find_confirmed_client_by_name(&exid->clname, nn);
3175	if (conf) {
3176		bool creds_match = same_creds(&conf->cl_cred, &rqstp->rq_cred);
3177		bool verfs_match = same_verf(&verf, &conf->cl_verifier);
3178
3179		if (update) {
3180			if (!clp_used_exchangeid(conf)) { /* buggy client */
3181				status = nfserr_inval;
3182				goto out;
3183			}
3184			if (!nfsd4_mach_creds_match(conf, rqstp)) {
3185				status = nfserr_wrong_cred;
3186				goto out;
3187			}
3188			if (!creds_match) { /* case 9 */
3189				status = nfserr_perm;
3190				goto out;
3191			}
3192			if (!verfs_match) { /* case 8 */
3193				status = nfserr_not_same;
3194				goto out;
3195			}
3196			/* case 6 */
3197			exid->flags |= EXCHGID4_FLAG_CONFIRMED_R;
3198			trace_nfsd_clid_confirmed_r(conf);
3199			goto out_copy;
3200		}
3201		if (!creds_match) { /* case 3 */
3202			if (client_has_state(conf)) {
3203				status = nfserr_clid_inuse;
3204				trace_nfsd_clid_cred_mismatch(conf, rqstp);
3205				goto out;
3206			}
3207			goto out_new;
3208		}
3209		if (verfs_match) { /* case 2 */
3210			conf->cl_exchange_flags |= EXCHGID4_FLAG_CONFIRMED_R;
3211			trace_nfsd_clid_confirmed_r(conf);
3212			goto out_copy;
3213		}
3214		/* case 5, client reboot */
3215		trace_nfsd_clid_verf_mismatch(conf, rqstp, &verf);
3216		conf = NULL;
3217		goto out_new;
3218	}
3219
3220	if (update) { /* case 7 */
3221		status = nfserr_noent;
3222		goto out;
3223	}
3224
3225	unconf = find_unconfirmed_client_by_name(&exid->clname, nn);
3226	if (unconf) /* case 4, possible retry or client restart */
3227		unhash_client_locked(unconf);
3228
3229	/* case 1, new owner ID */
3230	trace_nfsd_clid_fresh(new);
3231
3232out_new:
3233	if (conf) {
3234		status = mark_client_expired_locked(conf);
3235		if (status)
3236			goto out;
3237		trace_nfsd_clid_replaced(&conf->cl_clientid);
3238	}
3239	new->cl_minorversion = cstate->minorversion;
3240	new->cl_spo_must_allow.u.words[0] = exid->spo_must_allow[0];
3241	new->cl_spo_must_allow.u.words[1] = exid->spo_must_allow[1];
3242
3243	add_to_unconfirmed(new);
3244	swap(new, conf);
3245out_copy:
3246	exid->clientid.cl_boot = conf->cl_clientid.cl_boot;
3247	exid->clientid.cl_id = conf->cl_clientid.cl_id;
3248
3249	exid->seqid = conf->cl_cs_slot.sl_seqid + 1;
3250	nfsd4_set_ex_flags(conf, exid);
3251
3252	dprintk("nfsd4_exchange_id seqid %d flags %x\n",
3253		conf->cl_cs_slot.sl_seqid, conf->cl_exchange_flags);
3254	status = nfs_ok;
3255
3256out:
3257	spin_unlock(&nn->client_lock);
3258out_nolock:
3259	if (new)
3260		expire_client(new);
3261	if (unconf) {
3262		trace_nfsd_clid_expire_unconf(&unconf->cl_clientid);
3263		expire_client(unconf);
3264	}
3265	return status;
3266}
3267
3268static __be32
3269check_slot_seqid(u32 seqid, u32 slot_seqid, int slot_inuse)
3270{
3271	dprintk("%s enter. seqid %d slot_seqid %d\n", __func__, seqid,
3272		slot_seqid);
3273
3274	/* The slot is in use, and no response has been sent. */
3275	if (slot_inuse) {
3276		if (seqid == slot_seqid)
3277			return nfserr_jukebox;
3278		else
3279			return nfserr_seq_misordered;
3280	}
3281	/* Note unsigned 32-bit arithmetic handles wraparound: */
3282	if (likely(seqid == slot_seqid + 1))
3283		return nfs_ok;
3284	if (seqid == slot_seqid)
3285		return nfserr_replay_cache;
3286	return nfserr_seq_misordered;
3287}
3288
3289/*
3290 * Cache the create session result into the create session single DRC
3291 * slot cache by saving the xdr structure. sl_seqid has been set.
3292 * Do this for solo or embedded create session operations.
3293 */
3294static void
3295nfsd4_cache_create_session(struct nfsd4_create_session *cr_ses,
3296			   struct nfsd4_clid_slot *slot, __be32 nfserr)
3297{
3298	slot->sl_status = nfserr;
3299	memcpy(&slot->sl_cr_ses, cr_ses, sizeof(*cr_ses));
3300}
3301
3302static __be32
3303nfsd4_replay_create_session(struct nfsd4_create_session *cr_ses,
3304			    struct nfsd4_clid_slot *slot)
3305{
3306	memcpy(cr_ses, &slot->sl_cr_ses, sizeof(*cr_ses));
3307	return slot->sl_status;
3308}
3309
3310#define NFSD_MIN_REQ_HDR_SEQ_SZ	((\
3311			2 * 2 + /* credential,verifier: AUTH_NULL, length 0 */ \
3312			1 +	/* MIN tag is length with zero, only length */ \
3313			3 +	/* version, opcount, opcode */ \
3314			XDR_QUADLEN(NFS4_MAX_SESSIONID_LEN) + \
3315				/* seqid, slotID, slotID, cache */ \
3316			4 ) * sizeof(__be32))
3317
3318#define NFSD_MIN_RESP_HDR_SEQ_SZ ((\
3319			2 +	/* verifier: AUTH_NULL, length 0 */\
3320			1 +	/* status */ \
3321			1 +	/* MIN tag is length with zero, only length */ \
3322			3 +	/* opcount, opcode, opstatus*/ \
3323			XDR_QUADLEN(NFS4_MAX_SESSIONID_LEN) + \
3324				/* seqid, slotID, slotID, slotID, status */ \
3325			5 ) * sizeof(__be32))
3326
3327static __be32 check_forechannel_attrs(struct nfsd4_channel_attrs *ca, struct nfsd_net *nn)
3328{
3329	u32 maxrpc = nn->nfsd_serv->sv_max_mesg;
3330
3331	if (ca->maxreq_sz < NFSD_MIN_REQ_HDR_SEQ_SZ)
3332		return nfserr_toosmall;
3333	if (ca->maxresp_sz < NFSD_MIN_RESP_HDR_SEQ_SZ)
3334		return nfserr_toosmall;
3335	ca->headerpadsz = 0;
3336	ca->maxreq_sz = min_t(u32, ca->maxreq_sz, maxrpc);
3337	ca->maxresp_sz = min_t(u32, ca->maxresp_sz, maxrpc);
3338	ca->maxops = min_t(u32, ca->maxops, NFSD_MAX_OPS_PER_COMPOUND);
3339	ca->maxresp_cached = min_t(u32, ca->maxresp_cached,
3340			NFSD_SLOT_CACHE_SIZE + NFSD_MIN_HDR_SEQ_SZ);
3341	ca->maxreqs = min_t(u32, ca->maxreqs, NFSD_MAX_SLOTS_PER_SESSION);
3342	/*
3343	 * Note decreasing slot size below client's request may make it
3344	 * difficult for client to function correctly, whereas
3345	 * decreasing the number of slots will (just?) affect
3346	 * performance.  When short on memory we therefore prefer to
3347	 * decrease number of slots instead of their size.  Clients that
3348	 * request larger slots than they need will get poor results:
3349	 * Note that we always allow at least one slot, because our
3350	 * accounting is soft and provides no guarantees either way.
3351	 */
3352	ca->maxreqs = nfsd4_get_drc_mem(ca, nn);
3353
3354	return nfs_ok;
3355}
3356
3357/*
3358 * Server's NFSv4.1 backchannel support is AUTH_SYS-only for now.
3359 * These are based on similar macros in linux/sunrpc/msg_prot.h .
3360 */
3361#define RPC_MAX_HEADER_WITH_AUTH_SYS \
3362	(RPC_CALLHDRSIZE + 2 * (2 + UNX_CALLSLACK))
3363
3364#define RPC_MAX_REPHEADER_WITH_AUTH_SYS \
3365	(RPC_REPHDRSIZE + (2 + NUL_REPLYSLACK))
3366
3367#define NFSD_CB_MAX_REQ_SZ	((NFS4_enc_cb_recall_sz + \
3368				 RPC_MAX_HEADER_WITH_AUTH_SYS) * sizeof(__be32))
3369#define NFSD_CB_MAX_RESP_SZ	((NFS4_dec_cb_recall_sz + \
3370				 RPC_MAX_REPHEADER_WITH_AUTH_SYS) * \
3371				 sizeof(__be32))
3372
3373static __be32 check_backchannel_attrs(struct nfsd4_channel_attrs *ca)
3374{
3375	ca->headerpadsz = 0;
3376
3377	if (ca->maxreq_sz < NFSD_CB_MAX_REQ_SZ)
3378		return nfserr_toosmall;
3379	if (ca->maxresp_sz < NFSD_CB_MAX_RESP_SZ)
3380		return nfserr_toosmall;
3381	ca->maxresp_cached = 0;
3382	if (ca->maxops < 2)
3383		return nfserr_toosmall;
3384
3385	return nfs_ok;
3386}
3387
3388static __be32 nfsd4_check_cb_sec(struct nfsd4_cb_sec *cbs)
3389{
3390	switch (cbs->flavor) {
3391	case RPC_AUTH_NULL:
3392	case RPC_AUTH_UNIX:
3393		return nfs_ok;
3394	default:
3395		/*
3396		 * GSS case: the spec doesn't allow us to return this
3397		 * error.  But it also doesn't allow us not to support
3398		 * GSS.
3399		 * I'd rather this fail hard than return some error the
3400		 * client might think it can already handle:
3401		 */
3402		return nfserr_encr_alg_unsupp;
3403	}
3404}
3405
3406__be32
3407nfsd4_create_session(struct svc_rqst *rqstp,
3408		struct nfsd4_compound_state *cstate, union nfsd4_op_u *u)
3409{
3410	struct nfsd4_create_session *cr_ses = &u->create_session;
3411	struct sockaddr *sa = svc_addr(rqstp);
3412	struct nfs4_client *conf, *unconf;
3413	struct nfs4_client *old = NULL;
3414	struct nfsd4_session *new;
3415	struct nfsd4_conn *conn;
3416	struct nfsd4_clid_slot *cs_slot = NULL;
3417	__be32 status = 0;
3418	struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
3419
3420	if (cr_ses->flags & ~SESSION4_FLAG_MASK_A)
3421		return nfserr_inval;
3422	status = nfsd4_check_cb_sec(&cr_ses->cb_sec);
3423	if (status)
3424		return status;
3425	status = check_forechannel_attrs(&cr_ses->fore_channel, nn);
3426	if (status)
3427		return status;
3428	status = check_backchannel_attrs(&cr_ses->back_channel);
3429	if (status)
3430		goto out_release_drc_mem;
3431	status = nfserr_jukebox;
3432	new = alloc_session(&cr_ses->fore_channel, &cr_ses->back_channel);
3433	if (!new)
3434		goto out_release_drc_mem;
3435	conn = alloc_conn_from_crses(rqstp, cr_ses);
3436	if (!conn)
3437		goto out_free_session;
3438
3439	spin_lock(&nn->client_lock);
3440	unconf = find_unconfirmed_client(&cr_ses->clientid, true, nn);
3441	conf = find_confirmed_client(&cr_ses->clientid, true, nn);
3442	WARN_ON_ONCE(conf && unconf);
3443
3444	if (conf) {
3445		status = nfserr_wrong_cred;
3446		if (!nfsd4_mach_creds_match(conf, rqstp))
3447			goto out_free_conn;
3448		cs_slot = &conf->cl_cs_slot;
3449		status = check_slot_seqid(cr_ses->seqid, cs_slot->sl_seqid, 0);
3450		if (status) {
3451			if (status == nfserr_replay_cache)
3452				status = nfsd4_replay_create_session(cr_ses, cs_slot);
3453			goto out_free_conn;
3454		}
3455	} else if (unconf) {
3456		status = nfserr_clid_inuse;
3457		if (!same_creds(&unconf->cl_cred, &rqstp->rq_cred) ||
3458		    !rpc_cmp_addr(sa, (struct sockaddr *) &unconf->cl_addr)) {
3459			trace_nfsd_clid_cred_mismatch(unconf, rqstp);
3460			goto out_free_conn;
3461		}
3462		status = nfserr_wrong_cred;
3463		if (!nfsd4_mach_creds_match(unconf, rqstp))
3464			goto out_free_conn;
3465		cs_slot = &unconf->cl_cs_slot;
3466		status = check_slot_seqid(cr_ses->seqid, cs_slot->sl_seqid, 0);
3467		if (status) {
3468			/* an unconfirmed replay returns misordered */
3469			status = nfserr_seq_misordered;
3470			goto out_free_conn;
3471		}
3472		old = find_confirmed_client_by_name(&unconf->cl_name, nn);
3473		if (old) {
3474			status = mark_client_expired_locked(old);
3475			if (status) {
3476				old = NULL;
3477				goto out_free_conn;
3478			}
3479			trace_nfsd_clid_replaced(&old->cl_clientid);
3480		}
3481		move_to_confirmed(unconf);
3482		conf = unconf;
3483	} else {
3484		status = nfserr_stale_clientid;
3485		goto out_free_conn;
3486	}
3487	status = nfs_ok;
3488	/* Persistent sessions are not supported */
3489	cr_ses->flags &= ~SESSION4_PERSIST;
3490	/* Upshifting from TCP to RDMA is not supported */
3491	cr_ses->flags &= ~SESSION4_RDMA;
3492
3493	init_session(rqstp, new, conf, cr_ses);
3494	nfsd4_get_session_locked(new);
3495
3496	memcpy(cr_ses->sessionid.data, new->se_sessionid.data,
3497	       NFS4_MAX_SESSIONID_LEN);
3498	cs_slot->sl_seqid++;
3499	cr_ses->seqid = cs_slot->sl_seqid;
3500
3501	/* cache solo and embedded create sessions under the client_lock */
3502	nfsd4_cache_create_session(cr_ses, cs_slot, status);
3503	spin_unlock(&nn->client_lock);
3504	if (conf == unconf)
3505		fsnotify_dentry(conf->cl_nfsd_info_dentry, FS_MODIFY);
3506	/* init connection and backchannel */
3507	nfsd4_init_conn(rqstp, conn, new);
3508	nfsd4_put_session(new);
3509	if (old)
3510		expire_client(old);
3511	return status;
3512out_free_conn:
3513	spin_unlock(&nn->client_lock);
3514	free_conn(conn);
3515	if (old)
3516		expire_client(old);
3517out_free_session:
3518	__free_session(new);
3519out_release_drc_mem:
3520	nfsd4_put_drc_mem(&cr_ses->fore_channel);
3521	return status;
3522}
3523
3524static __be32 nfsd4_map_bcts_dir(u32 *dir)
3525{
3526	switch (*dir) {
3527	case NFS4_CDFC4_FORE:
3528	case NFS4_CDFC4_BACK:
3529		return nfs_ok;
3530	case NFS4_CDFC4_FORE_OR_BOTH:
3531	case NFS4_CDFC4_BACK_OR_BOTH:
3532		*dir = NFS4_CDFC4_BOTH;
3533		return nfs_ok;
3534	}
3535	return nfserr_inval;
3536}
3537
3538__be32 nfsd4_backchannel_ctl(struct svc_rqst *rqstp,
3539		struct nfsd4_compound_state *cstate,
3540		union nfsd4_op_u *u)
3541{
3542	struct nfsd4_backchannel_ctl *bc = &u->backchannel_ctl;
3543	struct nfsd4_session *session = cstate->session;
3544	struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
3545	__be32 status;
3546
3547	status = nfsd4_check_cb_sec(&bc->bc_cb_sec);
3548	if (status)
3549		return status;
3550	spin_lock(&nn->client_lock);
3551	session->se_cb_prog = bc->bc_cb_program;
3552	session->se_cb_sec = bc->bc_cb_sec;
3553	spin_unlock(&nn->client_lock);
3554
3555	nfsd4_probe_callback(session->se_client);
3556
3557	return nfs_ok;
3558}
3559
3560static struct nfsd4_conn *__nfsd4_find_conn(struct svc_xprt *xpt, struct nfsd4_session *s)
3561{
3562	struct nfsd4_conn *c;
3563
3564	list_for_each_entry(c, &s->se_conns, cn_persession) {
3565		if (c->cn_xprt == xpt) {
3566			return c;
3567		}
3568	}
3569	return NULL;
3570}
3571
3572static __be32 nfsd4_match_existing_connection(struct svc_rqst *rqst,
3573				struct nfsd4_session *session, u32 req)
3574{
3575	struct nfs4_client *clp = session->se_client;
3576	struct svc_xprt *xpt = rqst->rq_xprt;
3577	struct nfsd4_conn *c;
3578	__be32 status;
3579
3580	/* Following the last paragraph of RFC 5661 Section 18.34.3: */
3581	spin_lock(&clp->cl_lock);
3582	c = __nfsd4_find_conn(xpt, session);
3583	if (!c)
3584		status = nfserr_noent;
3585	else if (req == c->cn_flags)
3586		status = nfs_ok;
3587	else if (req == NFS4_CDFC4_FORE_OR_BOTH &&
3588				c->cn_flags != NFS4_CDFC4_BACK)
3589		status = nfs_ok;
3590	else if (req == NFS4_CDFC4_BACK_OR_BOTH &&
3591				c->cn_flags != NFS4_CDFC4_FORE)
3592		status = nfs_ok;
3593	else
3594		status = nfserr_inval;
3595	spin_unlock(&clp->cl_lock);
3596	return status;
3597}
3598
3599__be32 nfsd4_bind_conn_to_session(struct svc_rqst *rqstp,
3600		     struct nfsd4_compound_state *cstate,
3601		     union nfsd4_op_u *u)
3602{
3603	struct nfsd4_bind_conn_to_session *bcts = &u->bind_conn_to_session;
3604	__be32 status;
3605	struct nfsd4_conn *conn;
3606	struct nfsd4_session *session;
3607	struct net *net = SVC_NET(rqstp);
3608	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
3609
3610	if (!nfsd4_last_compound_op(rqstp))
3611		return nfserr_not_only_op;
3612	spin_lock(&nn->client_lock);
3613	session = find_in_sessionid_hashtbl(&bcts->sessionid, net, &status);
3614	spin_unlock(&nn->client_lock);
3615	if (!session)
3616		goto out_no_session;
3617	status = nfserr_wrong_cred;
3618	if (!nfsd4_mach_creds_match(session->se_client, rqstp))
3619		goto out;
3620	status = nfsd4_match_existing_connection(rqstp, session, bcts->dir);
3621	if (status == nfs_ok || status == nfserr_inval)
3622		goto out;
3623	status = nfsd4_map_bcts_dir(&bcts->dir);
3624	if (status)
3625		goto out;
3626	conn = alloc_conn(rqstp, bcts->dir);
3627	status = nfserr_jukebox;
3628	if (!conn)
3629		goto out;
3630	nfsd4_init_conn(rqstp, conn, session);
3631	status = nfs_ok;
3632out:
3633	nfsd4_put_session(session);
3634out_no_session:
3635	return status;
3636}
3637
3638static bool nfsd4_compound_in_session(struct nfsd4_compound_state *cstate, struct nfs4_sessionid *sid)
3639{
3640	if (!cstate->session)
3641		return false;
3642	return !memcmp(sid, &cstate->session->se_sessionid, sizeof(*sid));
3643}
3644
3645__be32
3646nfsd4_destroy_session(struct svc_rqst *r, struct nfsd4_compound_state *cstate,
3647		union nfsd4_op_u *u)
3648{
3649	struct nfs4_sessionid *sessionid = &u->destroy_session.sessionid;
3650	struct nfsd4_session *ses;
3651	__be32 status;
3652	int ref_held_by_me = 0;
3653	struct net *net = SVC_NET(r);
3654	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
3655
3656	status = nfserr_not_only_op;
3657	if (nfsd4_compound_in_session(cstate, sessionid)) {
3658		if (!nfsd4_last_compound_op(r))
3659			goto out;
3660		ref_held_by_me++;
3661	}
3662	dump_sessionid(__func__, sessionid);
3663	spin_lock(&nn->client_lock);
3664	ses = find_in_sessionid_hashtbl(sessionid, net, &status);
3665	if (!ses)
3666		goto out_client_lock;
3667	status = nfserr_wrong_cred;
3668	if (!nfsd4_mach_creds_match(ses->se_client, r))
3669		goto out_put_session;
3670	status = mark_session_dead_locked(ses, 1 + ref_held_by_me);
3671	if (status)
3672		goto out_put_session;
3673	unhash_session(ses);
3674	spin_unlock(&nn->client_lock);
3675
3676	nfsd4_probe_callback_sync(ses->se_client);
3677
3678	spin_lock(&nn->client_lock);
3679	status = nfs_ok;
3680out_put_session:
3681	nfsd4_put_session_locked(ses);
3682out_client_lock:
3683	spin_unlock(&nn->client_lock);
3684out:
3685	return status;
3686}
3687
3688static __be32 nfsd4_sequence_check_conn(struct nfsd4_conn *new, struct nfsd4_session *ses)
3689{
3690	struct nfs4_client *clp = ses->se_client;
3691	struct nfsd4_conn *c;
3692	__be32 status = nfs_ok;
3693	int ret;
3694
3695	spin_lock(&clp->cl_lock);
3696	c = __nfsd4_find_conn(new->cn_xprt, ses);
3697	if (c)
3698		goto out_free;
3699	status = nfserr_conn_not_bound_to_session;
3700	if (clp->cl_mach_cred)
3701		goto out_free;
3702	__nfsd4_hash_conn(new, ses);
3703	spin_unlock(&clp->cl_lock);
3704	ret = nfsd4_register_conn(new);
3705	if (ret)
3706		/* oops; xprt is already down: */
3707		nfsd4_conn_lost(&new->cn_xpt_user);
3708	return nfs_ok;
3709out_free:
3710	spin_unlock(&clp->cl_lock);
3711	free_conn(new);
3712	return status;
3713}
3714
3715static bool nfsd4_session_too_many_ops(struct svc_rqst *rqstp, struct nfsd4_session *session)
3716{
3717	struct nfsd4_compoundargs *args = rqstp->rq_argp;
3718
3719	return args->opcnt > session->se_fchannel.maxops;
3720}
3721
3722static bool nfsd4_request_too_big(struct svc_rqst *rqstp,
3723				  struct nfsd4_session *session)
3724{
3725	struct xdr_buf *xb = &rqstp->rq_arg;
3726
3727	return xb->len > session->se_fchannel.maxreq_sz;
3728}
3729
3730static bool replay_matches_cache(struct svc_rqst *rqstp,
3731		 struct nfsd4_sequence *seq, struct nfsd4_slot *slot)
3732{
3733	struct nfsd4_compoundargs *argp = rqstp->rq_argp;
3734
3735	if ((bool)(slot->sl_flags & NFSD4_SLOT_CACHETHIS) !=
3736	    (bool)seq->cachethis)
3737		return false;
3738	/*
3739	 * If there's an error then the reply can have fewer ops than
3740	 * the call.
3741	 */
3742	if (slot->sl_opcnt < argp->opcnt && !slot->sl_status)
3743		return false;
3744	/*
3745	 * But if we cached a reply with *more* ops than the call you're
3746	 * sending us now, then this new call is clearly not really a
3747	 * replay of the old one:
3748	 */
3749	if (slot->sl_opcnt > argp->opcnt)
3750		return false;
3751	/* This is the only check explicitly called by spec: */
3752	if (!same_creds(&rqstp->rq_cred, &slot->sl_cred))
3753		return false;
3754	/*
3755	 * There may be more comparisons we could actually do, but the
3756	 * spec doesn't require us to catch every case where the calls
3757	 * don't match (that would require caching the call as well as
3758	 * the reply), so we don't bother.
3759	 */
3760	return true;
3761}
3762
3763__be32
3764nfsd4_sequence(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3765		union nfsd4_op_u *u)
3766{
3767	struct nfsd4_sequence *seq = &u->sequence;
3768	struct nfsd4_compoundres *resp = rqstp->rq_resp;
3769	struct xdr_stream *xdr = resp->xdr;
3770	struct nfsd4_session *session;
3771	struct nfs4_client *clp;
3772	struct nfsd4_slot *slot;
3773	struct nfsd4_conn *conn;
3774	__be32 status;
3775	int buflen;
3776	struct net *net = SVC_NET(rqstp);
3777	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
3778
3779	if (resp->opcnt != 1)
3780		return nfserr_sequence_pos;
3781
3782	/*
3783	 * Will be either used or freed by nfsd4_sequence_check_conn
3784	 * below.
3785	 */
3786	conn = alloc_conn(rqstp, NFS4_CDFC4_FORE);
3787	if (!conn)
3788		return nfserr_jukebox;
3789
3790	spin_lock(&nn->client_lock);
3791	session = find_in_sessionid_hashtbl(&seq->sessionid, net, &status);
3792	if (!session)
3793		goto out_no_session;
3794	clp = session->se_client;
3795
3796	status = nfserr_too_many_ops;
3797	if (nfsd4_session_too_many_ops(rqstp, session))
3798		goto out_put_session;
3799
3800	status = nfserr_req_too_big;
3801	if (nfsd4_request_too_big(rqstp, session))
3802		goto out_put_session;
3803
3804	status = nfserr_badslot;
3805	if (seq->slotid >= session->se_fchannel.maxreqs)
3806		goto out_put_session;
3807
3808	slot = session->se_slots[seq->slotid];
3809	dprintk("%s: slotid %d\n", __func__, seq->slotid);
3810
3811	/* We do not negotiate the number of slots yet, so set the
3812	 * maxslots to the session maxreqs which is used to encode
3813	 * sr_highest_slotid and the sr_target_slot id to maxslots */
3814	seq->maxslots = session->se_fchannel.maxreqs;
3815
3816	status = check_slot_seqid(seq->seqid, slot->sl_seqid,
3817					slot->sl_flags & NFSD4_SLOT_INUSE);
3818	if (status == nfserr_replay_cache) {
3819		status = nfserr_seq_misordered;
3820		if (!(slot->sl_flags & NFSD4_SLOT_INITIALIZED))
3821			goto out_put_session;
3822		status = nfserr_seq_false_retry;
3823		if (!replay_matches_cache(rqstp, seq, slot))
3824			goto out_put_session;
3825		cstate->slot = slot;
3826		cstate->session = session;
3827		cstate->clp = clp;
3828		/* Return the cached reply status and set cstate->status
3829		 * for nfsd4_proc_compound processing */
3830		status = nfsd4_replay_cache_entry(resp, seq);
3831		cstate->status = nfserr_replay_cache;
3832		goto out;
3833	}
3834	if (status)
3835		goto out_put_session;
3836
3837	status = nfsd4_sequence_check_conn(conn, session);
3838	conn = NULL;
3839	if (status)
3840		goto out_put_session;
3841
3842	buflen = (seq->cachethis) ?
3843			session->se_fchannel.maxresp_cached :
3844			session->se_fchannel.maxresp_sz;
3845	status = (seq->cachethis) ? nfserr_rep_too_big_to_cache :
3846				    nfserr_rep_too_big;
3847	if (xdr_restrict_buflen(xdr, buflen - rqstp->rq_auth_slack))
3848		goto out_put_session;
3849	svc_reserve(rqstp, buflen);
3850
3851	status = nfs_ok;
3852	/* Success! bump slot seqid */
3853	slot->sl_seqid = seq->seqid;
3854	slot->sl_flags |= NFSD4_SLOT_INUSE;
3855	if (seq->cachethis)
3856		slot->sl_flags |= NFSD4_SLOT_CACHETHIS;
3857	else
3858		slot->sl_flags &= ~NFSD4_SLOT_CACHETHIS;
3859
3860	cstate->slot = slot;
3861	cstate->session = session;
3862	cstate->clp = clp;
3863
3864out:
3865	switch (clp->cl_cb_state) {
3866	case NFSD4_CB_DOWN:
3867		seq->status_flags = SEQ4_STATUS_CB_PATH_DOWN;
3868		break;
3869	case NFSD4_CB_FAULT:
3870		seq->status_flags = SEQ4_STATUS_BACKCHANNEL_FAULT;
3871		break;
3872	default:
3873		seq->status_flags = 0;
3874	}
3875	if (!list_empty(&clp->cl_revoked))
3876		seq->status_flags |= SEQ4_STATUS_RECALLABLE_STATE_REVOKED;
3877out_no_session:
3878	if (conn)
3879		free_conn(conn);
3880	spin_unlock(&nn->client_lock);
3881	return status;
3882out_put_session:
3883	nfsd4_put_session_locked(session);
3884	goto out_no_session;
3885}
3886
3887void
3888nfsd4_sequence_done(struct nfsd4_compoundres *resp)
3889{
3890	struct nfsd4_compound_state *cs = &resp->cstate;
3891
3892	if (nfsd4_has_session(cs)) {
3893		if (cs->status != nfserr_replay_cache) {
3894			nfsd4_store_cache_entry(resp);
3895			cs->slot->sl_flags &= ~NFSD4_SLOT_INUSE;
3896		}
3897		/* Drop session reference that was taken in nfsd4_sequence() */
3898		nfsd4_put_session(cs->session);
3899	} else if (cs->clp)
3900		put_client_renew(cs->clp);
3901}
3902
3903__be32
3904nfsd4_destroy_clientid(struct svc_rqst *rqstp,
3905		struct nfsd4_compound_state *cstate,
3906		union nfsd4_op_u *u)
3907{
3908	struct nfsd4_destroy_clientid *dc = &u->destroy_clientid;
3909	struct nfs4_client *conf, *unconf;
3910	struct nfs4_client *clp = NULL;
3911	__be32 status = 0;
3912	struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
3913
3914	spin_lock(&nn->client_lock);
3915	unconf = find_unconfirmed_client(&dc->clientid, true, nn);
3916	conf = find_confirmed_client(&dc->clientid, true, nn);
3917	WARN_ON_ONCE(conf && unconf);
3918
3919	if (conf) {
3920		if (client_has_state(conf)) {
3921			status = nfserr_clientid_busy;
3922			goto out;
3923		}
3924		status = mark_client_expired_locked(conf);
3925		if (status)
3926			goto out;
3927		clp = conf;
3928	} else if (unconf)
3929		clp = unconf;
3930	else {
3931		status = nfserr_stale_clientid;
3932		goto out;
3933	}
3934	if (!nfsd4_mach_creds_match(clp, rqstp)) {
3935		clp = NULL;
3936		status = nfserr_wrong_cred;
3937		goto out;
3938	}
3939	trace_nfsd_clid_destroyed(&clp->cl_clientid);
3940	unhash_client_locked(clp);
3941out:
3942	spin_unlock(&nn->client_lock);
3943	if (clp)
3944		expire_client(clp);
3945	return status;
3946}
3947
3948__be32
3949nfsd4_reclaim_complete(struct svc_rqst *rqstp,
3950		struct nfsd4_compound_state *cstate, union nfsd4_op_u *u)
3951{
3952	struct nfsd4_reclaim_complete *rc = &u->reclaim_complete;
3953	struct nfs4_client *clp = cstate->clp;
3954	__be32 status = 0;
3955
3956	if (rc->rca_one_fs) {
3957		if (!cstate->current_fh.fh_dentry)
3958			return nfserr_nofilehandle;
3959		/*
3960		 * We don't take advantage of the rca_one_fs case.
3961		 * That's OK, it's optional, we can safely ignore it.
3962		 */
3963		return nfs_ok;
3964	}
3965
3966	status = nfserr_complete_already;
3967	if (test_and_set_bit(NFSD4_CLIENT_RECLAIM_COMPLETE, &clp->cl_flags))
3968		goto out;
3969
3970	status = nfserr_stale_clientid;
3971	if (is_client_expired(clp))
3972		/*
3973		 * The following error isn't really legal.
3974		 * But we only get here if the client just explicitly
3975		 * destroyed the client.  Surely it no longer cares what
3976		 * error it gets back on an operation for the dead
3977		 * client.
3978		 */
3979		goto out;
3980
3981	status = nfs_ok;
3982	trace_nfsd_clid_reclaim_complete(&clp->cl_clientid);
3983	nfsd4_client_record_create(clp);
3984	inc_reclaim_complete(clp);
3985out:
3986	return status;
3987}
3988
3989__be32
3990nfsd4_setclientid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3991		  union nfsd4_op_u *u)
3992{
3993	struct nfsd4_setclientid *setclid = &u->setclientid;
3994	struct xdr_netobj 	clname = setclid->se_name;
3995	nfs4_verifier		clverifier = setclid->se_verf;
3996	struct nfs4_client	*conf, *new;
3997	struct nfs4_client	*unconf = NULL;
3998	__be32 			status;
3999	struct nfsd_net		*nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
4000
4001	new = create_client(clname, rqstp, &clverifier);
4002	if (new == NULL)
4003		return nfserr_jukebox;
4004	spin_lock(&nn->client_lock);
4005	conf = find_confirmed_client_by_name(&clname, nn);
4006	if (conf && client_has_state(conf)) {
4007		status = nfserr_clid_inuse;
4008		if (clp_used_exchangeid(conf))
4009			goto out;
4010		if (!same_creds(&conf->cl_cred, &rqstp->rq_cred)) {
4011			trace_nfsd_clid_cred_mismatch(conf, rqstp);
4012			goto out;
4013		}
4014	}
4015	unconf = find_unconfirmed_client_by_name(&clname, nn);
4016	if (unconf)
4017		unhash_client_locked(unconf);
4018	if (conf) {
4019		if (same_verf(&conf->cl_verifier, &clverifier)) {
4020			copy_clid(new, conf);
4021			gen_confirm(new, nn);
4022		} else
4023			trace_nfsd_clid_verf_mismatch(conf, rqstp,
4024						      &clverifier);
4025	} else
4026		trace_nfsd_clid_fresh(new);
4027	new->cl_minorversion = 0;
4028	gen_callback(new, setclid, rqstp);
4029	add_to_unconfirmed(new);
4030	setclid->se_clientid.cl_boot = new->cl_clientid.cl_boot;
4031	setclid->se_clientid.cl_id = new->cl_clientid.cl_id;
4032	memcpy(setclid->se_confirm.data, new->cl_confirm.data, sizeof(setclid->se_confirm.data));
4033	new = NULL;
4034	status = nfs_ok;
4035out:
4036	spin_unlock(&nn->client_lock);
4037	if (new)
4038		free_client(new);
4039	if (unconf) {
4040		trace_nfsd_clid_expire_unconf(&unconf->cl_clientid);
4041		expire_client(unconf);
4042	}
4043	return status;
4044}
4045
4046__be32
4047nfsd4_setclientid_confirm(struct svc_rqst *rqstp,
4048			struct nfsd4_compound_state *cstate,
4049			union nfsd4_op_u *u)
4050{
4051	struct nfsd4_setclientid_confirm *setclientid_confirm =
4052			&u->setclientid_confirm;
4053	struct nfs4_client *conf, *unconf;
4054	struct nfs4_client *old = NULL;
4055	nfs4_verifier confirm = setclientid_confirm->sc_confirm;
4056	clientid_t * clid = &setclientid_confirm->sc_clientid;
4057	__be32 status;
4058	struct nfsd_net	*nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
4059
4060	if (STALE_CLIENTID(clid, nn))
4061		return nfserr_stale_clientid;
4062
4063	spin_lock(&nn->client_lock);
4064	conf = find_confirmed_client(clid, false, nn);
4065	unconf = find_unconfirmed_client(clid, false, nn);
4066	/*
4067	 * We try hard to give out unique clientid's, so if we get an
4068	 * attempt to confirm the same clientid with a different cred,
4069	 * the client may be buggy; this should never happen.
4070	 *
4071	 * Nevertheless, RFC 7530 recommends INUSE for this case:
4072	 */
4073	status = nfserr_clid_inuse;
4074	if (unconf && !same_creds(&unconf->cl_cred, &rqstp->rq_cred)) {
4075		trace_nfsd_clid_cred_mismatch(unconf, rqstp);
4076		goto out;
4077	}
4078	if (conf && !same_creds(&conf->cl_cred, &rqstp->rq_cred)) {
4079		trace_nfsd_clid_cred_mismatch(conf, rqstp);
4080		goto out;
4081	}
4082	if (!unconf || !same_verf(&confirm, &unconf->cl_confirm)) {
4083		if (conf && same_verf(&confirm, &conf->cl_confirm)) {
4084			status = nfs_ok;
4085		} else
4086			status = nfserr_stale_clientid;
4087		goto out;
4088	}
4089	status = nfs_ok;
4090	if (conf) {
4091		old = unconf;
4092		unhash_client_locked(old);
4093		nfsd4_change_callback(conf, &unconf->cl_cb_conn);
4094	} else {
4095		old = find_confirmed_client_by_name(&unconf->cl_name, nn);
4096		if (old) {
4097			status = nfserr_clid_inuse;
4098			if (client_has_state(old)
4099					&& !same_creds(&unconf->cl_cred,
4100							&old->cl_cred))
4101				goto out;
4102			status = mark_client_expired_locked(old);
4103			if (status) {
4104				old = NULL;
4105				goto out;
4106			}
4107			trace_nfsd_clid_replaced(&old->cl_clientid);
4108		}
4109		move_to_confirmed(unconf);
4110		conf = unconf;
4111	}
4112	get_client_locked(conf);
4113	spin_unlock(&nn->client_lock);
4114	if (conf == unconf)
4115		fsnotify_dentry(conf->cl_nfsd_info_dentry, FS_MODIFY);
4116	nfsd4_probe_callback(conf);
4117	spin_lock(&nn->client_lock);
4118	put_client_renew_locked(conf);
4119out:
4120	spin_unlock(&nn->client_lock);
4121	if (old)
4122		expire_client(old);
4123	return status;
4124}
4125
4126static struct nfs4_file *nfsd4_alloc_file(void)
4127{
4128	return kmem_cache_alloc(file_slab, GFP_KERNEL);
4129}
4130
4131/* OPEN Share state helper functions */
4132static void nfsd4_init_file(struct svc_fh *fh, unsigned int hashval,
4133				struct nfs4_file *fp)
4134{
4135	lockdep_assert_held(&state_lock);
4136
4137	refcount_set(&fp->fi_ref, 1);
4138	spin_lock_init(&fp->fi_lock);
4139	INIT_LIST_HEAD(&fp->fi_stateids);
4140	INIT_LIST_HEAD(&fp->fi_delegations);
4141	INIT_LIST_HEAD(&fp->fi_clnt_odstate);
4142	fh_copy_shallow(&fp->fi_fhandle, &fh->fh_handle);
4143	fp->fi_deleg_file = NULL;
4144	fp->fi_had_conflict = false;
4145	fp->fi_share_deny = 0;
4146	memset(fp->fi_fds, 0, sizeof(fp->fi_fds));
4147	memset(fp->fi_access, 0, sizeof(fp->fi_access));
4148	fp->fi_aliased = false;
4149	fp->fi_inode = d_inode(fh->fh_dentry);
4150#ifdef CONFIG_NFSD_PNFS
4151	INIT_LIST_HEAD(&fp->fi_lo_states);
4152	atomic_set(&fp->fi_lo_recalls, 0);
4153#endif
4154	hlist_add_head_rcu(&fp->fi_hash, &file_hashtbl[hashval]);
4155}
4156
4157void
4158nfsd4_free_slabs(void)
4159{
4160	kmem_cache_destroy(client_slab);
4161	kmem_cache_destroy(openowner_slab);
4162	kmem_cache_destroy(lockowner_slab);
4163	kmem_cache_destroy(file_slab);
4164	kmem_cache_destroy(stateid_slab);
4165	kmem_cache_destroy(deleg_slab);
4166	kmem_cache_destroy(odstate_slab);
4167}
4168
4169int
4170nfsd4_init_slabs(void)
4171{
4172	client_slab = kmem_cache_create("nfsd4_clients",
4173			sizeof(struct nfs4_client), 0, 0, NULL);
4174	if (client_slab == NULL)
4175		goto out;
4176	openowner_slab = kmem_cache_create("nfsd4_openowners",
4177			sizeof(struct nfs4_openowner), 0, 0, NULL);
4178	if (openowner_slab == NULL)
4179		goto out_free_client_slab;
4180	lockowner_slab = kmem_cache_create("nfsd4_lockowners",
4181			sizeof(struct nfs4_lockowner), 0, 0, NULL);
4182	if (lockowner_slab == NULL)
4183		goto out_free_openowner_slab;
4184	file_slab = kmem_cache_create("nfsd4_files",
4185			sizeof(struct nfs4_file), 0, 0, NULL);
4186	if (file_slab == NULL)
4187		goto out_free_lockowner_slab;
4188	stateid_slab = kmem_cache_create("nfsd4_stateids",
4189			sizeof(struct nfs4_ol_stateid), 0, 0, NULL);
4190	if (stateid_slab == NULL)
4191		goto out_free_file_slab;
4192	deleg_slab = kmem_cache_create("nfsd4_delegations",
4193			sizeof(struct nfs4_delegation), 0, 0, NULL);
4194	if (deleg_slab == NULL)
4195		goto out_free_stateid_slab;
4196	odstate_slab = kmem_cache_create("nfsd4_odstate",
4197			sizeof(struct nfs4_clnt_odstate), 0, 0, NULL);
4198	if (odstate_slab == NULL)
4199		goto out_free_deleg_slab;
4200	return 0;
4201
4202out_free_deleg_slab:
4203	kmem_cache_destroy(deleg_slab);
4204out_free_stateid_slab:
4205	kmem_cache_destroy(stateid_slab);
4206out_free_file_slab:
4207	kmem_cache_destroy(file_slab);
4208out_free_lockowner_slab:
4209	kmem_cache_destroy(lockowner_slab);
4210out_free_openowner_slab:
4211	kmem_cache_destroy(openowner_slab);
4212out_free_client_slab:
4213	kmem_cache_destroy(client_slab);
4214out:
4215	return -ENOMEM;
4216}
4217
4218static void init_nfs4_replay(struct nfs4_replay *rp)
4219{
4220	rp->rp_status = nfserr_serverfault;
4221	rp->rp_buflen = 0;
4222	rp->rp_buf = rp->rp_ibuf;
4223	mutex_init(&rp->rp_mutex);
4224}
4225
4226static void nfsd4_cstate_assign_replay(struct nfsd4_compound_state *cstate,
4227		struct nfs4_stateowner *so)
4228{
4229	if (!nfsd4_has_session(cstate)) {
4230		mutex_lock(&so->so_replay.rp_mutex);
4231		cstate->replay_owner = nfs4_get_stateowner(so);
4232	}
4233}
4234
4235void nfsd4_cstate_clear_replay(struct nfsd4_compound_state *cstate)
4236{
4237	struct nfs4_stateowner *so = cstate->replay_owner;
4238
4239	if (so != NULL) {
4240		cstate->replay_owner = NULL;
4241		mutex_unlock(&so->so_replay.rp_mutex);
4242		nfs4_put_stateowner(so);
4243	}
4244}
4245
4246static inline void *alloc_stateowner(struct kmem_cache *slab, struct xdr_netobj *owner, struct nfs4_client *clp)
4247{
4248	struct nfs4_stateowner *sop;
4249
4250	sop = kmem_cache_alloc(slab, GFP_KERNEL);
4251	if (!sop)
4252		return NULL;
4253
4254	xdr_netobj_dup(&sop->so_owner, owner, GFP_KERNEL);
4255	if (!sop->so_owner.data) {
4256		kmem_cache_free(slab, sop);
4257		return NULL;
4258	}
4259
4260	INIT_LIST_HEAD(&sop->so_stateids);
4261	sop->so_client = clp;
4262	init_nfs4_replay(&sop->so_replay);
4263	atomic_set(&sop->so_count, 1);
4264	return sop;
4265}
4266
4267static void hash_openowner(struct nfs4_openowner *oo, struct nfs4_client *clp, unsigned int strhashval)
4268{
4269	lockdep_assert_held(&clp->cl_lock);
4270
4271	list_add(&oo->oo_owner.so_strhash,
4272		 &clp->cl_ownerstr_hashtbl[strhashval]);
4273	list_add(&oo->oo_perclient, &clp->cl_openowners);
4274}
4275
4276static void nfs4_unhash_openowner(struct nfs4_stateowner *so)
4277{
4278	unhash_openowner_locked(openowner(so));
4279}
4280
4281static void nfs4_free_openowner(struct nfs4_stateowner *so)
4282{
4283	struct nfs4_openowner *oo = openowner(so);
4284
4285	kmem_cache_free(openowner_slab, oo);
4286}
4287
4288static const struct nfs4_stateowner_operations openowner_ops = {
4289	.so_unhash =	nfs4_unhash_openowner,
4290	.so_free =	nfs4_free_openowner,
4291};
4292
4293static struct nfs4_ol_stateid *
4294nfsd4_find_existing_open(struct nfs4_file *fp, struct nfsd4_open *open)
4295{
4296	struct nfs4_ol_stateid *local, *ret = NULL;
4297	struct nfs4_openowner *oo = open->op_openowner;
4298
4299	lockdep_assert_held(&fp->fi_lock);
4300
4301	list_for_each_entry(local, &fp->fi_stateids, st_perfile) {
4302		/* ignore lock owners */
4303		if (local->st_stateowner->so_is_open_owner == 0)
4304			continue;
4305		if (local->st_stateowner != &oo->oo_owner)
4306			continue;
4307		if (local->st_stid.sc_type == NFS4_OPEN_STID) {
4308			ret = local;
4309			refcount_inc(&ret->st_stid.sc_count);
4310			break;
4311		}
4312	}
4313	return ret;
4314}
4315
4316static __be32
4317nfsd4_verify_open_stid(struct nfs4_stid *s)
4318{
4319	__be32 ret = nfs_ok;
4320
4321	switch (s->sc_type) {
4322	default:
4323		break;
4324	case 0:
4325	case NFS4_CLOSED_STID:
4326	case NFS4_CLOSED_DELEG_STID:
4327		ret = nfserr_bad_stateid;
4328		break;
4329	case NFS4_REVOKED_DELEG_STID:
4330		ret = nfserr_deleg_revoked;
4331	}
4332	return ret;
4333}
4334
4335/* Lock the stateid st_mutex, and deal with races with CLOSE */
4336static __be32
4337nfsd4_lock_ol_stateid(struct nfs4_ol_stateid *stp)
4338{
4339	__be32 ret;
4340
4341	mutex_lock_nested(&stp->st_mutex, LOCK_STATEID_MUTEX);
4342	ret = nfsd4_verify_open_stid(&stp->st_stid);
4343	if (ret != nfs_ok)
4344		mutex_unlock(&stp->st_mutex);
4345	return ret;
4346}
4347
4348static struct nfs4_ol_stateid *
4349nfsd4_find_and_lock_existing_open(struct nfs4_file *fp, struct nfsd4_open *open)
4350{
4351	struct nfs4_ol_stateid *stp;
4352	for (;;) {
4353		spin_lock(&fp->fi_lock);
4354		stp = nfsd4_find_existing_open(fp, open);
4355		spin_unlock(&fp->fi_lock);
4356		if (!stp || nfsd4_lock_ol_stateid(stp) == nfs_ok)
4357			break;
4358		nfs4_put_stid(&stp->st_stid);
4359	}
4360	return stp;
4361}
4362
4363static struct nfs4_openowner *
4364alloc_init_open_stateowner(unsigned int strhashval, struct nfsd4_open *open,
4365			   struct nfsd4_compound_state *cstate)
4366{
4367	struct nfs4_client *clp = cstate->clp;
4368	struct nfs4_openowner *oo, *ret;
4369
4370	oo = alloc_stateowner(openowner_slab, &open->op_owner, clp);
4371	if (!oo)
4372		return NULL;
4373	oo->oo_owner.so_ops = &openowner_ops;
4374	oo->oo_owner.so_is_open_owner = 1;
4375	oo->oo_owner.so_seqid = open->op_seqid;
4376	oo->oo_flags = 0;
4377	if (nfsd4_has_session(cstate))
4378		oo->oo_flags |= NFS4_OO_CONFIRMED;
4379	oo->oo_time = 0;
4380	oo->oo_last_closed_stid = NULL;
4381	INIT_LIST_HEAD(&oo->oo_close_lru);
4382	spin_lock(&clp->cl_lock);
4383	ret = find_openstateowner_str_locked(strhashval, open, clp);
4384	if (ret == NULL) {
4385		hash_openowner(oo, clp, strhashval);
4386		ret = oo;
4387	} else
4388		nfs4_free_stateowner(&oo->oo_owner);
4389
4390	spin_unlock(&clp->cl_lock);
4391	return ret;
4392}
4393
4394static struct nfs4_ol_stateid *
4395init_open_stateid(struct nfs4_file *fp, struct nfsd4_open *open)
4396{
4397
4398	struct nfs4_openowner *oo = open->op_openowner;
4399	struct nfs4_ol_stateid *retstp = NULL;
4400	struct nfs4_ol_stateid *stp;
4401
4402	stp = open->op_stp;
4403	/* We are moving these outside of the spinlocks to avoid the warnings */
4404	mutex_init(&stp->st_mutex);
4405	mutex_lock_nested(&stp->st_mutex, OPEN_STATEID_MUTEX);
4406
4407retry:
4408	spin_lock(&oo->oo_owner.so_client->cl_lock);
4409	spin_lock(&fp->fi_lock);
4410
4411	retstp = nfsd4_find_existing_open(fp, open);
4412	if (retstp)
4413		goto out_unlock;
4414
4415	open->op_stp = NULL;
4416	refcount_inc(&stp->st_stid.sc_count);
4417	stp->st_stid.sc_type = NFS4_OPEN_STID;
4418	INIT_LIST_HEAD(&stp->st_locks);
4419	stp->st_stateowner = nfs4_get_stateowner(&oo->oo_owner);
4420	get_nfs4_file(fp);
4421	stp->st_stid.sc_file = fp;
4422	stp->st_access_bmap = 0;
4423	stp->st_deny_bmap = 0;
4424	stp->st_openstp = NULL;
4425	list_add(&stp->st_perstateowner, &oo->oo_owner.so_stateids);
4426	list_add(&stp->st_perfile, &fp->fi_stateids);
4427
4428out_unlock:
4429	spin_unlock(&fp->fi_lock);
4430	spin_unlock(&oo->oo_owner.so_client->cl_lock);
4431	if (retstp) {
4432		/* Handle races with CLOSE */
4433		if (nfsd4_lock_ol_stateid(retstp) != nfs_ok) {
4434			nfs4_put_stid(&retstp->st_stid);
4435			goto retry;
4436		}
4437		/* To keep mutex tracking happy */
4438		mutex_unlock(&stp->st_mutex);
4439		stp = retstp;
4440	}
4441	return stp;
4442}
4443
4444/*
4445 * In the 4.0 case we need to keep the owners around a little while to handle
4446 * CLOSE replay. We still do need to release any file access that is held by
4447 * them before returning however.
4448 */
4449static void
4450move_to_close_lru(struct nfs4_ol_stateid *s, struct net *net)
4451{
4452	struct nfs4_ol_stateid *last;
4453	struct nfs4_openowner *oo = openowner(s->st_stateowner);
4454	struct nfsd_net *nn = net_generic(s->st_stid.sc_client->net,
4455						nfsd_net_id);
4456
4457	dprintk("NFSD: move_to_close_lru nfs4_openowner %p\n", oo);
4458
4459	/*
4460	 * We know that we hold one reference via nfsd4_close, and another
4461	 * "persistent" reference for the client. If the refcount is higher
4462	 * than 2, then there are still calls in progress that are using this
4463	 * stateid. We can't put the sc_file reference until they are finished.
4464	 * Wait for the refcount to drop to 2. Since it has been unhashed,
4465	 * there should be no danger of the refcount going back up again at
4466	 * this point.
4467	 */
4468	wait_event(close_wq, refcount_read(&s->st_stid.sc_count) == 2);
4469
4470	release_all_access(s);
4471	if (s->st_stid.sc_file) {
4472		put_nfs4_file(s->st_stid.sc_file);
4473		s->st_stid.sc_file = NULL;
4474	}
4475
4476	spin_lock(&nn->client_lock);
4477	last = oo->oo_last_closed_stid;
4478	oo->oo_last_closed_stid = s;
4479	list_move_tail(&oo->oo_close_lru, &nn->close_lru);
4480	oo->oo_time = ktime_get_boottime_seconds();
4481	spin_unlock(&nn->client_lock);
4482	if (last)
4483		nfs4_put_stid(&last->st_stid);
4484}
4485
4486/* search file_hashtbl[] for file */
4487static struct nfs4_file *
4488find_file_locked(struct svc_fh *fh, unsigned int hashval)
4489{
4490	struct nfs4_file *fp;
4491
4492	hlist_for_each_entry_rcu(fp, &file_hashtbl[hashval], fi_hash,
4493				lockdep_is_held(&state_lock)) {
4494		if (fh_match(&fp->fi_fhandle, &fh->fh_handle)) {
4495			if (refcount_inc_not_zero(&fp->fi_ref))
4496				return fp;
4497		}
4498	}
4499	return NULL;
4500}
4501
4502static struct nfs4_file *insert_file(struct nfs4_file *new, struct svc_fh *fh,
4503				     unsigned int hashval)
4504{
4505	struct nfs4_file *fp;
4506	struct nfs4_file *ret = NULL;
4507	bool alias_found = false;
4508
4509	spin_lock(&state_lock);
4510	hlist_for_each_entry_rcu(fp, &file_hashtbl[hashval], fi_hash,
4511				 lockdep_is_held(&state_lock)) {
4512		if (fh_match(&fp->fi_fhandle, &fh->fh_handle)) {
4513			if (refcount_inc_not_zero(&fp->fi_ref))
4514				ret = fp;
4515		} else if (d_inode(fh->fh_dentry) == fp->fi_inode)
4516			fp->fi_aliased = alias_found = true;
4517	}
4518	if (likely(ret == NULL)) {
4519		nfsd4_init_file(fh, hashval, new);
4520		new->fi_aliased = alias_found;
4521		ret = new;
4522	}
4523	spin_unlock(&state_lock);
4524	return ret;
4525}
4526
4527static struct nfs4_file * find_file(struct svc_fh *fh)
4528{
4529	struct nfs4_file *fp;
4530	unsigned int hashval = file_hashval(fh);
4531
4532	rcu_read_lock();
4533	fp = find_file_locked(fh, hashval);
4534	rcu_read_unlock();
4535	return fp;
4536}
4537
4538static struct nfs4_file *
4539find_or_add_file(struct nfs4_file *new, struct svc_fh *fh)
4540{
4541	struct nfs4_file *fp;
4542	unsigned int hashval = file_hashval(fh);
4543
4544	rcu_read_lock();
4545	fp = find_file_locked(fh, hashval);
4546	rcu_read_unlock();
4547	if (fp)
4548		return fp;
4549
4550	return insert_file(new, fh, hashval);
4551}
4552
4553/*
4554 * Called to check deny when READ with all zero stateid or
4555 * WRITE with all zero or all one stateid
4556 */
4557static __be32
4558nfs4_share_conflict(struct svc_fh *current_fh, unsigned int deny_type)
4559{
4560	struct nfs4_file *fp;
4561	__be32 ret = nfs_ok;
4562
4563	fp = find_file(current_fh);
4564	if (!fp)
4565		return ret;
4566	/* Check for conflicting share reservations */
4567	spin_lock(&fp->fi_lock);
4568	if (fp->fi_share_deny & deny_type)
4569		ret = nfserr_locked;
4570	spin_unlock(&fp->fi_lock);
4571	put_nfs4_file(fp);
4572	return ret;
4573}
4574
4575static void nfsd4_cb_recall_prepare(struct nfsd4_callback *cb)
4576{
4577	struct nfs4_delegation *dp = cb_to_delegation(cb);
4578	struct nfsd_net *nn = net_generic(dp->dl_stid.sc_client->net,
4579					  nfsd_net_id);
4580
4581	block_delegations(&dp->dl_stid.sc_file->fi_fhandle);
4582
4583	/*
4584	 * We can't do this in nfsd_break_deleg_cb because it is
4585	 * already holding inode->i_lock.
4586	 *
4587	 * If the dl_time != 0, then we know that it has already been
4588	 * queued for a lease break. Don't queue it again.
4589	 */
4590	spin_lock(&state_lock);
4591	if (dp->dl_time == 0) {
4592		dp->dl_time = ktime_get_boottime_seconds();
4593		list_add_tail(&dp->dl_recall_lru, &nn->del_recall_lru);
4594	}
4595	spin_unlock(&state_lock);
4596}
4597
4598static int nfsd4_cb_recall_done(struct nfsd4_callback *cb,
4599		struct rpc_task *task)
4600{
4601	struct nfs4_delegation *dp = cb_to_delegation(cb);
4602
4603	if (dp->dl_stid.sc_type == NFS4_CLOSED_DELEG_STID ||
4604	    dp->dl_stid.sc_type == NFS4_REVOKED_DELEG_STID)
4605	        return 1;
4606
4607	switch (task->tk_status) {
4608	case 0:
4609		return 1;
4610	case -NFS4ERR_DELAY:
4611		rpc_delay(task, 2 * HZ);
4612		return 0;
4613	case -EBADHANDLE:
4614	case -NFS4ERR_BAD_STATEID:
4615		/*
4616		 * Race: client probably got cb_recall before open reply
4617		 * granting delegation.
4618		 */
4619		if (dp->dl_retries--) {
4620			rpc_delay(task, 2 * HZ);
4621			return 0;
4622		}
4623		fallthrough;
4624	default:
4625		return 1;
4626	}
4627}
4628
4629static void nfsd4_cb_recall_release(struct nfsd4_callback *cb)
4630{
4631	struct nfs4_delegation *dp = cb_to_delegation(cb);
4632
4633	nfs4_put_stid(&dp->dl_stid);
4634}
4635
4636static const struct nfsd4_callback_ops nfsd4_cb_recall_ops = {
4637	.prepare	= nfsd4_cb_recall_prepare,
4638	.done		= nfsd4_cb_recall_done,
4639	.release	= nfsd4_cb_recall_release,
4640};
4641
4642static void nfsd_break_one_deleg(struct nfs4_delegation *dp)
4643{
4644	/*
4645	 * We're assuming the state code never drops its reference
4646	 * without first removing the lease.  Since we're in this lease
4647	 * callback (and since the lease code is serialized by the
4648	 * i_lock) we know the server hasn't removed the lease yet, and
4649	 * we know it's safe to take a reference.
4650	 */
4651	refcount_inc(&dp->dl_stid.sc_count);
4652	nfsd4_run_cb(&dp->dl_recall);
4653}
4654
4655/* Called from break_lease() with i_lock held. */
4656static bool
4657nfsd_break_deleg_cb(struct file_lock *fl)
4658{
4659	bool ret = false;
4660	struct nfs4_delegation *dp = (struct nfs4_delegation *)fl->fl_owner;
4661	struct nfs4_file *fp = dp->dl_stid.sc_file;
4662
4663	trace_nfsd_cb_recall(&dp->dl_stid);
4664
4665	/*
4666	 * We don't want the locks code to timeout the lease for us;
4667	 * we'll remove it ourself if a delegation isn't returned
4668	 * in time:
4669	 */
4670	fl->fl_break_time = 0;
4671
4672	spin_lock(&fp->fi_lock);
4673	fp->fi_had_conflict = true;
4674	nfsd_break_one_deleg(dp);
4675	spin_unlock(&fp->fi_lock);
4676	return ret;
4677}
4678
4679static bool nfsd_breaker_owns_lease(struct file_lock *fl)
4680{
4681	struct nfs4_delegation *dl = fl->fl_owner;
4682	struct svc_rqst *rqst;
4683	struct nfs4_client *clp;
4684
4685	if (!i_am_nfsd())
4686		return NULL;
4687	rqst = kthread_data(current);
4688	/* Note rq_prog == NFS_ACL_PROGRAM is also possible: */
4689	if (rqst->rq_prog != NFS_PROGRAM || rqst->rq_vers < 4)
4690		return NULL;
4691	clp = *(rqst->rq_lease_breaker);
4692	return dl->dl_stid.sc_client == clp;
4693}
4694
4695static int
4696nfsd_change_deleg_cb(struct file_lock *onlist, int arg,
4697		     struct list_head *dispose)
4698{
4699	if (arg & F_UNLCK)
4700		return lease_modify(onlist, arg, dispose);
4701	else
4702		return -EAGAIN;
4703}
4704
4705static const struct lock_manager_operations nfsd_lease_mng_ops = {
4706	.lm_breaker_owns_lease = nfsd_breaker_owns_lease,
4707	.lm_break = nfsd_break_deleg_cb,
4708	.lm_change = nfsd_change_deleg_cb,
4709};
4710
4711static __be32 nfsd4_check_seqid(struct nfsd4_compound_state *cstate, struct nfs4_stateowner *so, u32 seqid)
4712{
4713	if (nfsd4_has_session(cstate))
4714		return nfs_ok;
4715	if (seqid == so->so_seqid - 1)
4716		return nfserr_replay_me;
4717	if (seqid == so->so_seqid)
4718		return nfs_ok;
4719	return nfserr_bad_seqid;
4720}
4721
4722static struct nfs4_client *lookup_clientid(clientid_t *clid, bool sessions,
4723						struct nfsd_net *nn)
4724{
4725	struct nfs4_client *found;
4726
4727	spin_lock(&nn->client_lock);
4728	found = find_confirmed_client(clid, sessions, nn);
4729	if (found)
4730		atomic_inc(&found->cl_rpc_users);
4731	spin_unlock(&nn->client_lock);
4732	return found;
4733}
4734
4735static __be32 set_client(clientid_t *clid,
4736		struct nfsd4_compound_state *cstate,
4737		struct nfsd_net *nn)
4738{
4739	if (cstate->clp) {
4740		if (!same_clid(&cstate->clp->cl_clientid, clid))
4741			return nfserr_stale_clientid;
4742		return nfs_ok;
4743	}
4744	if (STALE_CLIENTID(clid, nn))
4745		return nfserr_stale_clientid;
4746	/*
4747	 * We're in the 4.0 case (otherwise the SEQUENCE op would have
4748	 * set cstate->clp), so session = false:
4749	 */
4750	cstate->clp = lookup_clientid(clid, false, nn);
4751	if (!cstate->clp)
4752		return nfserr_expired;
4753	return nfs_ok;
4754}
4755
4756__be32
4757nfsd4_process_open1(struct nfsd4_compound_state *cstate,
4758		    struct nfsd4_open *open, struct nfsd_net *nn)
4759{
4760	clientid_t *clientid = &open->op_clientid;
4761	struct nfs4_client *clp = NULL;
4762	unsigned int strhashval;
4763	struct nfs4_openowner *oo = NULL;
4764	__be32 status;
4765
4766	/*
4767	 * In case we need it later, after we've already created the
4768	 * file and don't want to risk a further failure:
4769	 */
4770	open->op_file = nfsd4_alloc_file();
4771	if (open->op_file == NULL)
4772		return nfserr_jukebox;
4773
4774	status = set_client(clientid, cstate, nn);
4775	if (status)
4776		return status;
4777	clp = cstate->clp;
4778
4779	strhashval = ownerstr_hashval(&open->op_owner);
4780	oo = find_openstateowner_str(strhashval, open, clp);
4781	open->op_openowner = oo;
4782	if (!oo) {
4783		goto new_owner;
4784	}
4785	if (!(oo->oo_flags & NFS4_OO_CONFIRMED)) {
4786		/* Replace unconfirmed owners without checking for replay. */
4787		release_openowner(oo);
4788		open->op_openowner = NULL;
4789		goto new_owner;
4790	}
4791	status = nfsd4_check_seqid(cstate, &oo->oo_owner, open->op_seqid);
4792	if (status)
4793		return status;
4794	goto alloc_stateid;
4795new_owner:
4796	oo = alloc_init_open_stateowner(strhashval, open, cstate);
4797	if (oo == NULL)
4798		return nfserr_jukebox;
4799	open->op_openowner = oo;
4800alloc_stateid:
4801	open->op_stp = nfs4_alloc_open_stateid(clp);
4802	if (!open->op_stp)
4803		return nfserr_jukebox;
4804
4805	if (nfsd4_has_session(cstate) &&
4806	    (cstate->current_fh.fh_export->ex_flags & NFSEXP_PNFS)) {
4807		open->op_odstate = alloc_clnt_odstate(clp);
4808		if (!open->op_odstate)
4809			return nfserr_jukebox;
4810	}
4811
4812	return nfs_ok;
4813}
4814
4815static inline __be32
4816nfs4_check_delegmode(struct nfs4_delegation *dp, int flags)
4817{
4818	if ((flags & WR_STATE) && (dp->dl_type == NFS4_OPEN_DELEGATE_READ))
4819		return nfserr_openmode;
4820	else
4821		return nfs_ok;
4822}
4823
4824static int share_access_to_flags(u32 share_access)
4825{
4826	return share_access == NFS4_SHARE_ACCESS_READ ? RD_STATE : WR_STATE;
4827}
4828
4829static struct nfs4_delegation *find_deleg_stateid(struct nfs4_client *cl, stateid_t *s)
4830{
4831	struct nfs4_stid *ret;
4832
4833	ret = find_stateid_by_type(cl, s,
4834				NFS4_DELEG_STID|NFS4_REVOKED_DELEG_STID);
4835	if (!ret)
4836		return NULL;
4837	return delegstateid(ret);
4838}
4839
4840static bool nfsd4_is_deleg_cur(struct nfsd4_open *open)
4841{
4842	return open->op_claim_type == NFS4_OPEN_CLAIM_DELEGATE_CUR ||
4843	       open->op_claim_type == NFS4_OPEN_CLAIM_DELEG_CUR_FH;
4844}
4845
4846static __be32
4847nfs4_check_deleg(struct nfs4_client *cl, struct nfsd4_open *open,
4848		struct nfs4_delegation **dp)
4849{
4850	int flags;
4851	__be32 status = nfserr_bad_stateid;
4852	struct nfs4_delegation *deleg;
4853
4854	deleg = find_deleg_stateid(cl, &open->op_delegate_stateid);
4855	if (deleg == NULL)
4856		goto out;
4857	if (deleg->dl_stid.sc_type == NFS4_REVOKED_DELEG_STID) {
4858		nfs4_put_stid(&deleg->dl_stid);
4859		if (cl->cl_minorversion)
4860			status = nfserr_deleg_revoked;
4861		goto out;
4862	}
4863	flags = share_access_to_flags(open->op_share_access);
4864	status = nfs4_check_delegmode(deleg, flags);
4865	if (status) {
4866		nfs4_put_stid(&deleg->dl_stid);
4867		goto out;
4868	}
4869	*dp = deleg;
4870out:
4871	if (!nfsd4_is_deleg_cur(open))
4872		return nfs_ok;
4873	if (status)
4874		return status;
4875	open->op_openowner->oo_flags |= NFS4_OO_CONFIRMED;
4876	return nfs_ok;
4877}
4878
4879static inline int nfs4_access_to_access(u32 nfs4_access)
4880{
4881	int flags = 0;
4882
4883	if (nfs4_access & NFS4_SHARE_ACCESS_READ)
4884		flags |= NFSD_MAY_READ;
4885	if (nfs4_access & NFS4_SHARE_ACCESS_WRITE)
4886		flags |= NFSD_MAY_WRITE;
4887	return flags;
4888}
4889
4890static inline __be32
4891nfsd4_truncate(struct svc_rqst *rqstp, struct svc_fh *fh,
4892		struct nfsd4_open *open)
4893{
4894	struct iattr iattr = {
4895		.ia_valid = ATTR_SIZE,
4896		.ia_size = 0,
4897	};
4898	if (!open->op_truncate)
4899		return 0;
4900	if (!(open->op_share_access & NFS4_SHARE_ACCESS_WRITE))
4901		return nfserr_inval;
4902	return nfsd_setattr(rqstp, fh, &iattr, 0, (time64_t)0);
4903}
4904
4905static __be32 nfs4_get_vfs_file(struct svc_rqst *rqstp, struct nfs4_file *fp,
4906		struct svc_fh *cur_fh, struct nfs4_ol_stateid *stp,
4907		struct nfsd4_open *open)
4908{
4909	struct nfsd_file *nf = NULL;
4910	__be32 status;
4911	int oflag = nfs4_access_to_omode(open->op_share_access);
4912	int access = nfs4_access_to_access(open->op_share_access);
4913	unsigned char old_access_bmap, old_deny_bmap;
4914
4915	spin_lock(&fp->fi_lock);
4916
4917	/*
4918	 * Are we trying to set a deny mode that would conflict with
4919	 * current access?
4920	 */
4921	status = nfs4_file_check_deny(fp, open->op_share_deny);
4922	if (status != nfs_ok) {
4923		spin_unlock(&fp->fi_lock);
4924		goto out;
4925	}
4926
4927	/* set access to the file */
4928	status = nfs4_file_get_access(fp, open->op_share_access);
4929	if (status != nfs_ok) {
4930		spin_unlock(&fp->fi_lock);
4931		goto out;
4932	}
4933
4934	/* Set access bits in stateid */
4935	old_access_bmap = stp->st_access_bmap;
4936	set_access(open->op_share_access, stp);
4937
4938	/* Set new deny mask */
4939	old_deny_bmap = stp->st_deny_bmap;
4940	set_deny(open->op_share_deny, stp);
4941	fp->fi_share_deny |= (open->op_share_deny & NFS4_SHARE_DENY_BOTH);
4942
4943	if (!fp->fi_fds[oflag]) {
4944		spin_unlock(&fp->fi_lock);
4945		status = nfsd_file_acquire(rqstp, cur_fh, access, &nf);
4946		if (status)
4947			goto out_put_access;
4948		spin_lock(&fp->fi_lock);
4949		if (!fp->fi_fds[oflag]) {
4950			fp->fi_fds[oflag] = nf;
4951			nf = NULL;
4952		}
4953	}
4954	spin_unlock(&fp->fi_lock);
4955	if (nf)
4956		nfsd_file_put(nf);
4957
4958	status = nfserrno(nfsd_open_break_lease(cur_fh->fh_dentry->d_inode,
4959								access));
4960	if (status)
4961		goto out_put_access;
4962
4963	status = nfsd4_truncate(rqstp, cur_fh, open);
4964	if (status)
4965		goto out_put_access;
4966out:
4967	return status;
4968out_put_access:
4969	stp->st_access_bmap = old_access_bmap;
4970	nfs4_file_put_access(fp, open->op_share_access);
4971	reset_union_bmap_deny(bmap_to_share_mode(old_deny_bmap), stp);
4972	goto out;
4973}
4974
4975static __be32
4976nfs4_upgrade_open(struct svc_rqst *rqstp, struct nfs4_file *fp, struct svc_fh *cur_fh, struct nfs4_ol_stateid *stp, struct nfsd4_open *open)
4977{
4978	__be32 status;
4979	unsigned char old_deny_bmap = stp->st_deny_bmap;
4980
4981	if (!test_access(open->op_share_access, stp))
4982		return nfs4_get_vfs_file(rqstp, fp, cur_fh, stp, open);
4983
4984	/* test and set deny mode */
4985	spin_lock(&fp->fi_lock);
4986	status = nfs4_file_check_deny(fp, open->op_share_deny);
4987	if (status == nfs_ok) {
4988		set_deny(open->op_share_deny, stp);
4989		fp->fi_share_deny |=
4990				(open->op_share_deny & NFS4_SHARE_DENY_BOTH);
4991	}
4992	spin_unlock(&fp->fi_lock);
4993
4994	if (status != nfs_ok)
4995		return status;
4996
4997	status = nfsd4_truncate(rqstp, cur_fh, open);
4998	if (status != nfs_ok)
4999		reset_union_bmap_deny(old_deny_bmap, stp);
5000	return status;
5001}
5002
5003/* Should we give out recallable state?: */
5004static bool nfsd4_cb_channel_good(struct nfs4_client *clp)
5005{
5006	if (clp->cl_cb_state == NFSD4_CB_UP)
5007		return true;
5008	/*
5009	 * In the sessions case, since we don't have to establish a
5010	 * separate connection for callbacks, we assume it's OK
5011	 * until we hear otherwise:
5012	 */
5013	return clp->cl_minorversion && clp->cl_cb_state == NFSD4_CB_UNKNOWN;
5014}
5015
5016static struct file_lock *nfs4_alloc_init_lease(struct nfs4_delegation *dp,
5017						int flag)
5018{
5019	struct file_lock *fl;
5020
5021	fl = locks_alloc_lock();
5022	if (!fl)
5023		return NULL;
5024	fl->fl_lmops = &nfsd_lease_mng_ops;
5025	fl->fl_flags = FL_DELEG;
5026	fl->fl_type = flag == NFS4_OPEN_DELEGATE_READ? F_RDLCK: F_WRLCK;
5027	fl->fl_end = OFFSET_MAX;
5028	fl->fl_owner = (fl_owner_t)dp;
5029	fl->fl_pid = current->tgid;
5030	fl->fl_file = dp->dl_stid.sc_file->fi_deleg_file->nf_file;
5031	return fl;
5032}
5033
5034static int nfsd4_check_conflicting_opens(struct nfs4_client *clp,
5035					 struct nfs4_file *fp)
5036{
5037	struct nfs4_ol_stateid *st;
5038	struct file *f = fp->fi_deleg_file->nf_file;
5039	struct inode *ino = locks_inode(f);
5040	int writes;
5041
5042	writes = atomic_read(&ino->i_writecount);
5043	if (!writes)
5044		return 0;
5045	/*
5046	 * There could be multiple filehandles (hence multiple
5047	 * nfs4_files) referencing this file, but that's not too
5048	 * common; let's just give up in that case rather than
5049	 * trying to go look up all the clients using that other
5050	 * nfs4_file as well:
5051	 */
5052	if (fp->fi_aliased)
5053		return -EAGAIN;
5054	/*
5055	 * If there's a close in progress, make sure that we see it
5056	 * clear any fi_fds[] entries before we see it decrement
5057	 * i_writecount:
5058	 */
5059	smp_mb__after_atomic();
5060
5061	if (fp->fi_fds[O_WRONLY])
5062		writes--;
5063	if (fp->fi_fds[O_RDWR])
5064		writes--;
5065	if (writes > 0)
5066		return -EAGAIN; /* There may be non-NFSv4 writers */
5067	/*
5068	 * It's possible there are non-NFSv4 write opens in progress,
5069	 * but if they haven't incremented i_writecount yet then they
5070	 * also haven't called break lease yet; so, they'll break this
5071	 * lease soon enough.  So, all that's left to check for is NFSv4
5072	 * opens:
5073	 */
5074	spin_lock(&fp->fi_lock);
5075	list_for_each_entry(st, &fp->fi_stateids, st_perfile) {
5076		if (st->st_openstp == NULL /* it's an open */ &&
5077		    access_permit_write(st) &&
5078		    st->st_stid.sc_client != clp) {
5079			spin_unlock(&fp->fi_lock);
5080			return -EAGAIN;
5081		}
5082	}
5083	spin_unlock(&fp->fi_lock);
5084	/*
5085	 * There's a small chance that we could be racing with another
5086	 * NFSv4 open.  However, any open that hasn't added itself to
5087	 * the fi_stateids list also hasn't called break_lease yet; so,
5088	 * they'll break this lease soon enough.
5089	 */
5090	return 0;
5091}
5092
5093static struct nfs4_delegation *
5094nfs4_set_delegation(struct nfs4_client *clp, struct svc_fh *fh,
5095		    struct nfs4_file *fp, struct nfs4_clnt_odstate *odstate)
5096{
5097	int status = 0;
5098	struct nfs4_delegation *dp;
5099	struct nfsd_file *nf;
5100	struct file_lock *fl;
5101
5102	/*
5103	 * The fi_had_conflict and nfs_get_existing_delegation checks
5104	 * here are just optimizations; we'll need to recheck them at
5105	 * the end:
5106	 */
5107	if (fp->fi_had_conflict)
5108		return ERR_PTR(-EAGAIN);
5109
5110	nf = find_readable_file(fp);
5111	if (!nf) {
5112		/*
5113		 * We probably could attempt another open and get a read
5114		 * delegation, but for now, don't bother until the
5115		 * client actually sends us one.
5116		 */
5117		return ERR_PTR(-EAGAIN);
5118	}
5119	spin_lock(&state_lock);
5120	spin_lock(&fp->fi_lock);
5121	if (nfs4_delegation_exists(clp, fp))
5122		status = -EAGAIN;
5123	else if (!fp->fi_deleg_file) {
5124		fp->fi_deleg_file = nf;
5125		/* increment early to prevent fi_deleg_file from being
5126		 * cleared */
5127		fp->fi_delegees = 1;
5128		nf = NULL;
5129	} else
5130		fp->fi_delegees++;
5131	spin_unlock(&fp->fi_lock);
5132	spin_unlock(&state_lock);
5133	if (nf)
5134		nfsd_file_put(nf);
5135	if (status)
5136		return ERR_PTR(status);
5137
5138	status = -ENOMEM;
5139	dp = alloc_init_deleg(clp, fp, fh, odstate);
5140	if (!dp)
5141		goto out_delegees;
5142
5143	fl = nfs4_alloc_init_lease(dp, NFS4_OPEN_DELEGATE_READ);
5144	if (!fl)
5145		goto out_clnt_odstate;
5146
5147	status = vfs_setlease(fp->fi_deleg_file->nf_file, fl->fl_type, &fl, NULL);
5148	if (fl)
5149		locks_free_lock(fl);
5150	if (status)
5151		goto out_clnt_odstate;
5152	status = nfsd4_check_conflicting_opens(clp, fp);
5153	if (status)
5154		goto out_unlock;
5155
5156	spin_lock(&state_lock);
5157	spin_lock(&fp->fi_lock);
5158	if (fp->fi_had_conflict)
5159		status = -EAGAIN;
5160	else
5161		status = hash_delegation_locked(dp, fp);
5162	spin_unlock(&fp->fi_lock);
5163	spin_unlock(&state_lock);
5164
5165	if (status)
5166		goto out_unlock;
5167
5168	return dp;
5169out_unlock:
5170	vfs_setlease(fp->fi_deleg_file->nf_file, F_UNLCK, NULL, (void **)&dp);
5171out_clnt_odstate:
5172	put_clnt_odstate(dp->dl_clnt_odstate);
5173	nfs4_put_stid(&dp->dl_stid);
5174out_delegees:
5175	put_deleg_file(fp);
5176	return ERR_PTR(status);
5177}
5178
5179static void nfsd4_open_deleg_none_ext(struct nfsd4_open *open, int status)
5180{
5181	open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE_EXT;
5182	if (status == -EAGAIN)
5183		open->op_why_no_deleg = WND4_CONTENTION;
5184	else {
5185		open->op_why_no_deleg = WND4_RESOURCE;
5186		switch (open->op_deleg_want) {
5187		case NFS4_SHARE_WANT_READ_DELEG:
5188		case NFS4_SHARE_WANT_WRITE_DELEG:
5189		case NFS4_SHARE_WANT_ANY_DELEG:
5190			break;
5191		case NFS4_SHARE_WANT_CANCEL:
5192			open->op_why_no_deleg = WND4_CANCELLED;
5193			break;
5194		case NFS4_SHARE_WANT_NO_DELEG:
5195			WARN_ON_ONCE(1);
5196		}
5197	}
5198}
5199
5200/*
5201 * Attempt to hand out a delegation.
5202 *
5203 * Note we don't support write delegations, and won't until the vfs has
5204 * proper support for them.
5205 */
5206static void
5207nfs4_open_delegation(struct svc_fh *fh, struct nfsd4_open *open,
5208			struct nfs4_ol_stateid *stp)
5209{
5210	struct nfs4_delegation *dp;
5211	struct nfs4_openowner *oo = openowner(stp->st_stateowner);
5212	struct nfs4_client *clp = stp->st_stid.sc_client;
5213	int cb_up;
5214	int status = 0;
5215
5216	cb_up = nfsd4_cb_channel_good(oo->oo_owner.so_client);
5217	open->op_recall = 0;
5218	switch (open->op_claim_type) {
5219		case NFS4_OPEN_CLAIM_PREVIOUS:
5220			if (!cb_up)
5221				open->op_recall = 1;
5222			if (open->op_delegate_type != NFS4_OPEN_DELEGATE_READ)
5223				goto out_no_deleg;
5224			break;
5225		case NFS4_OPEN_CLAIM_NULL:
5226		case NFS4_OPEN_CLAIM_FH:
5227			/*
5228			 * Let's not give out any delegations till everyone's
5229			 * had the chance to reclaim theirs, *and* until
5230			 * NLM locks have all been reclaimed:
5231			 */
5232			if (locks_in_grace(clp->net))
5233				goto out_no_deleg;
5234			if (!cb_up || !(oo->oo_flags & NFS4_OO_CONFIRMED))
5235				goto out_no_deleg;
5236			break;
5237		default:
5238			goto out_no_deleg;
5239	}
5240	dp = nfs4_set_delegation(clp, fh, stp->st_stid.sc_file, stp->st_clnt_odstate);
5241	if (IS_ERR(dp))
5242		goto out_no_deleg;
5243
5244	memcpy(&open->op_delegate_stateid, &dp->dl_stid.sc_stateid, sizeof(dp->dl_stid.sc_stateid));
5245
5246	trace_nfsd_deleg_read(&dp->dl_stid.sc_stateid);
5247	open->op_delegate_type = NFS4_OPEN_DELEGATE_READ;
5248	nfs4_put_stid(&dp->dl_stid);
5249	return;
5250out_no_deleg:
5251	open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE;
5252	if (open->op_claim_type == NFS4_OPEN_CLAIM_PREVIOUS &&
5253	    open->op_delegate_type != NFS4_OPEN_DELEGATE_NONE) {
5254		dprintk("NFSD: WARNING: refusing delegation reclaim\n");
5255		open->op_recall = 1;
5256	}
5257
5258	/* 4.1 client asking for a delegation? */
5259	if (open->op_deleg_want)
5260		nfsd4_open_deleg_none_ext(open, status);
5261	return;
5262}
5263
5264static void nfsd4_deleg_xgrade_none_ext(struct nfsd4_open *open,
5265					struct nfs4_delegation *dp)
5266{
5267	if (open->op_deleg_want == NFS4_SHARE_WANT_READ_DELEG &&
5268	    dp->dl_type == NFS4_OPEN_DELEGATE_WRITE) {
5269		open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE_EXT;
5270		open->op_why_no_deleg = WND4_NOT_SUPP_DOWNGRADE;
5271	} else if (open->op_deleg_want == NFS4_SHARE_WANT_WRITE_DELEG &&
5272		   dp->dl_type == NFS4_OPEN_DELEGATE_WRITE) {
5273		open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE_EXT;
5274		open->op_why_no_deleg = WND4_NOT_SUPP_UPGRADE;
5275	}
5276	/* Otherwise the client must be confused wanting a delegation
5277	 * it already has, therefore we don't return
5278	 * NFS4_OPEN_DELEGATE_NONE_EXT and reason.
5279	 */
5280}
5281
5282__be32
5283nfsd4_process_open2(struct svc_rqst *rqstp, struct svc_fh *current_fh, struct nfsd4_open *open)
5284{
5285	struct nfsd4_compoundres *resp = rqstp->rq_resp;
5286	struct nfs4_client *cl = open->op_openowner->oo_owner.so_client;
5287	struct nfs4_file *fp = NULL;
5288	struct nfs4_ol_stateid *stp = NULL;
5289	struct nfs4_delegation *dp = NULL;
5290	__be32 status;
5291	bool new_stp = false;
5292
5293	/*
5294	 * Lookup file; if found, lookup stateid and check open request,
5295	 * and check for delegations in the process of being recalled.
5296	 * If not found, create the nfs4_file struct
5297	 */
5298	fp = find_or_add_file(open->op_file, current_fh);
5299	if (fp != open->op_file) {
5300		status = nfs4_check_deleg(cl, open, &dp);
5301		if (status)
5302			goto out;
5303		stp = nfsd4_find_and_lock_existing_open(fp, open);
5304	} else {
5305		open->op_file = NULL;
5306		status = nfserr_bad_stateid;
5307		if (nfsd4_is_deleg_cur(open))
5308			goto out;
5309	}
5310
5311	if (!stp) {
5312		stp = init_open_stateid(fp, open);
5313		if (!open->op_stp)
5314			new_stp = true;
5315	}
5316
5317	/*
5318	 * OPEN the file, or upgrade an existing OPEN.
5319	 * If truncate fails, the OPEN fails.
5320	 *
5321	 * stp is already locked.
5322	 */
5323	if (!new_stp) {
5324		/* Stateid was found, this is an OPEN upgrade */
5325		status = nfs4_upgrade_open(rqstp, fp, current_fh, stp, open);
5326		if (status) {
5327			mutex_unlock(&stp->st_mutex);
5328			goto out;
5329		}
5330	} else {
5331		status = nfs4_get_vfs_file(rqstp, fp, current_fh, stp, open);
5332		if (status) {
5333			stp->st_stid.sc_type = NFS4_CLOSED_STID;
5334			release_open_stateid(stp);
5335			mutex_unlock(&stp->st_mutex);
5336			goto out;
5337		}
5338
5339		stp->st_clnt_odstate = find_or_hash_clnt_odstate(fp,
5340							open->op_odstate);
5341		if (stp->st_clnt_odstate == open->op_odstate)
5342			open->op_odstate = NULL;
5343	}
5344
5345	nfs4_inc_and_copy_stateid(&open->op_stateid, &stp->st_stid);
5346	mutex_unlock(&stp->st_mutex);
5347
5348	if (nfsd4_has_session(&resp->cstate)) {
5349		if (open->op_deleg_want & NFS4_SHARE_WANT_NO_DELEG) {
5350			open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE_EXT;
5351			open->op_why_no_deleg = WND4_NOT_WANTED;
5352			goto nodeleg;
5353		}
5354	}
5355
5356	/*
5357	* Attempt to hand out a delegation. No error return, because the
5358	* OPEN succeeds even if we fail.
5359	*/
5360	nfs4_open_delegation(current_fh, open, stp);
5361nodeleg:
5362	status = nfs_ok;
5363	trace_nfsd_open(&stp->st_stid.sc_stateid);
5364out:
5365	/* 4.1 client trying to upgrade/downgrade delegation? */
5366	if (open->op_delegate_type == NFS4_OPEN_DELEGATE_NONE && dp &&
5367	    open->op_deleg_want)
5368		nfsd4_deleg_xgrade_none_ext(open, dp);
5369
5370	if (fp)
5371		put_nfs4_file(fp);
5372	if (status == 0 && open->op_claim_type == NFS4_OPEN_CLAIM_PREVIOUS)
5373		open->op_openowner->oo_flags |= NFS4_OO_CONFIRMED;
5374	/*
5375	* To finish the open response, we just need to set the rflags.
5376	*/
5377	open->op_rflags = NFS4_OPEN_RESULT_LOCKTYPE_POSIX;
5378	if (nfsd4_has_session(&resp->cstate))
5379		open->op_rflags |= NFS4_OPEN_RESULT_MAY_NOTIFY_LOCK;
5380	else if (!(open->op_openowner->oo_flags & NFS4_OO_CONFIRMED))
5381		open->op_rflags |= NFS4_OPEN_RESULT_CONFIRM;
5382
5383	if (dp)
5384		nfs4_put_stid(&dp->dl_stid);
5385	if (stp)
5386		nfs4_put_stid(&stp->st_stid);
5387
5388	return status;
5389}
5390
5391void nfsd4_cleanup_open_state(struct nfsd4_compound_state *cstate,
5392			      struct nfsd4_open *open)
5393{
5394	if (open->op_openowner) {
5395		struct nfs4_stateowner *so = &open->op_openowner->oo_owner;
5396
5397		nfsd4_cstate_assign_replay(cstate, so);
5398		nfs4_put_stateowner(so);
5399	}
5400	if (open->op_file)
5401		kmem_cache_free(file_slab, open->op_file);
5402	if (open->op_stp)
5403		nfs4_put_stid(&open->op_stp->st_stid);
5404	if (open->op_odstate)
5405		kmem_cache_free(odstate_slab, open->op_odstate);
5406}
5407
5408__be32
5409nfsd4_renew(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
5410	    union nfsd4_op_u *u)
5411{
5412	clientid_t *clid = &u->renew;
5413	struct nfs4_client *clp;
5414	__be32 status;
5415	struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
5416
5417	trace_nfsd_clid_renew(clid);
5418	status = set_client(clid, cstate, nn);
5419	if (status)
5420		return status;
5421	clp = cstate->clp;
5422	if (!list_empty(&clp->cl_delegations)
5423			&& clp->cl_cb_state != NFSD4_CB_UP)
5424		return nfserr_cb_path_down;
5425	return nfs_ok;
5426}
5427
5428void
5429nfsd4_end_grace(struct nfsd_net *nn)
5430{
5431	/* do nothing if grace period already ended */
5432	if (nn->grace_ended)
5433		return;
5434
5435	trace_nfsd_grace_complete(nn);
5436	nn->grace_ended = true;
5437	/*
5438	 * If the server goes down again right now, an NFSv4
5439	 * client will still be allowed to reclaim after it comes back up,
5440	 * even if it hasn't yet had a chance to reclaim state this time.
5441	 *
5442	 */
5443	nfsd4_record_grace_done(nn);
5444	/*
5445	 * At this point, NFSv4 clients can still reclaim.  But if the
5446	 * server crashes, any that have not yet reclaimed will be out
5447	 * of luck on the next boot.
5448	 *
5449	 * (NFSv4.1+ clients are considered to have reclaimed once they
5450	 * call RECLAIM_COMPLETE.  NFSv4.0 clients are considered to
5451	 * have reclaimed after their first OPEN.)
5452	 */
5453	locks_end_grace(&nn->nfsd4_manager);
5454	/*
5455	 * At this point, and once lockd and/or any other containers
5456	 * exit their grace period, further reclaims will fail and
5457	 * regular locking can resume.
5458	 */
5459}
5460
5461/*
5462 * If we've waited a lease period but there are still clients trying to
5463 * reclaim, wait a little longer to give them a chance to finish.
5464 */
5465static bool clients_still_reclaiming(struct nfsd_net *nn)
5466{
5467	time64_t double_grace_period_end = nn->boot_time +
5468					   2 * nn->nfsd4_lease;
5469
5470	if (nn->track_reclaim_completes &&
5471			atomic_read(&nn->nr_reclaim_complete) ==
5472			nn->reclaim_str_hashtbl_size)
5473		return false;
5474	if (!nn->somebody_reclaimed)
5475		return false;
5476	nn->somebody_reclaimed = false;
5477	/*
5478	 * If we've given them *two* lease times to reclaim, and they're
5479	 * still not done, give up:
5480	 */
5481	if (ktime_get_boottime_seconds() > double_grace_period_end)
5482		return false;
5483	return true;
5484}
5485
5486struct laundry_time {
5487	time64_t cutoff;
5488	time64_t new_timeo;
5489};
5490
5491static bool state_expired(struct laundry_time *lt, time64_t last_refresh)
5492{
5493	time64_t time_remaining;
5494
5495	if (last_refresh < lt->cutoff)
5496		return true;
5497	time_remaining = last_refresh - lt->cutoff;
5498	lt->new_timeo = min(lt->new_timeo, time_remaining);
5499	return false;
5500}
5501
5502#ifdef CONFIG_NFSD_V4_2_INTER_SSC
5503void nfsd4_ssc_init_umount_work(struct nfsd_net *nn)
5504{
5505	spin_lock_init(&nn->nfsd_ssc_lock);
5506	INIT_LIST_HEAD(&nn->nfsd_ssc_mount_list);
5507	init_waitqueue_head(&nn->nfsd_ssc_waitq);
5508}
5509EXPORT_SYMBOL_GPL(nfsd4_ssc_init_umount_work);
5510
5511/*
5512 * This is called when nfsd is being shutdown, after all inter_ssc
5513 * cleanup were done, to destroy the ssc delayed unmount list.
5514 */
5515static void nfsd4_ssc_shutdown_umount(struct nfsd_net *nn)
5516{
5517	struct nfsd4_ssc_umount_item *ni = NULL;
5518	struct nfsd4_ssc_umount_item *tmp;
5519
5520	spin_lock(&nn->nfsd_ssc_lock);
5521	list_for_each_entry_safe(ni, tmp, &nn->nfsd_ssc_mount_list, nsui_list) {
5522		list_del(&ni->nsui_list);
5523		spin_unlock(&nn->nfsd_ssc_lock);
5524		mntput(ni->nsui_vfsmount);
5525		kfree(ni);
5526		spin_lock(&nn->nfsd_ssc_lock);
5527	}
5528	spin_unlock(&nn->nfsd_ssc_lock);
5529}
5530
5531static void nfsd4_ssc_expire_umount(struct nfsd_net *nn)
5532{
5533	bool do_wakeup = false;
5534	struct nfsd4_ssc_umount_item *ni = 0;
5535	struct nfsd4_ssc_umount_item *tmp;
5536
5537	spin_lock(&nn->nfsd_ssc_lock);
5538	list_for_each_entry_safe(ni, tmp, &nn->nfsd_ssc_mount_list, nsui_list) {
5539		if (time_after(jiffies, ni->nsui_expire)) {
5540			if (refcount_read(&ni->nsui_refcnt) > 1)
5541				continue;
5542
5543			/* mark being unmount */
5544			ni->nsui_busy = true;
5545			spin_unlock(&nn->nfsd_ssc_lock);
5546			mntput(ni->nsui_vfsmount);
5547			spin_lock(&nn->nfsd_ssc_lock);
5548
5549			/* waiters need to start from begin of list */
5550			list_del(&ni->nsui_list);
5551			kfree(ni);
5552
5553			/* wakeup ssc_connect waiters */
5554			do_wakeup = true;
5555			continue;
5556		}
5557		break;
5558	}
5559	if (do_wakeup)
5560		wake_up_all(&nn->nfsd_ssc_waitq);
5561	spin_unlock(&nn->nfsd_ssc_lock);
5562}
5563#endif
5564
5565static time64_t
5566nfs4_laundromat(struct nfsd_net *nn)
5567{
5568	struct nfs4_client *clp;
5569	struct nfs4_openowner *oo;
5570	struct nfs4_delegation *dp;
5571	struct nfs4_ol_stateid *stp;
5572	struct nfsd4_blocked_lock *nbl;
5573	struct list_head *pos, *next, reaplist;
5574	struct laundry_time lt = {
5575		.cutoff = ktime_get_boottime_seconds() - nn->nfsd4_lease,
5576		.new_timeo = nn->nfsd4_lease
5577	};
5578	struct nfs4_cpntf_state *cps;
5579	copy_stateid_t *cps_t;
5580	int i;
5581
5582	if (clients_still_reclaiming(nn)) {
5583		lt.new_timeo = 0;
5584		goto out;
5585	}
5586	nfsd4_end_grace(nn);
5587	INIT_LIST_HEAD(&reaplist);
5588
5589	spin_lock(&nn->s2s_cp_lock);
5590	idr_for_each_entry(&nn->s2s_cp_stateids, cps_t, i) {
5591		cps = container_of(cps_t, struct nfs4_cpntf_state, cp_stateid);
5592		if (cps->cp_stateid.sc_type == NFS4_COPYNOTIFY_STID &&
5593				state_expired(&lt, cps->cpntf_time))
5594			_free_cpntf_state_locked(nn, cps);
5595	}
5596	spin_unlock(&nn->s2s_cp_lock);
5597
5598	spin_lock(&nn->client_lock);
5599	list_for_each_safe(pos, next, &nn->client_lru) {
5600		clp = list_entry(pos, struct nfs4_client, cl_lru);
5601		if (!state_expired(&lt, clp->cl_time))
5602			break;
5603		if (mark_client_expired_locked(clp))
5604			continue;
5605		list_add(&clp->cl_lru, &reaplist);
5606	}
5607	spin_unlock(&nn->client_lock);
5608	list_for_each_safe(pos, next, &reaplist) {
5609		clp = list_entry(pos, struct nfs4_client, cl_lru);
5610		trace_nfsd_clid_purged(&clp->cl_clientid);
5611		list_del_init(&clp->cl_lru);
5612		expire_client(clp);
5613	}
5614	spin_lock(&state_lock);
5615	list_for_each_safe(pos, next, &nn->del_recall_lru) {
5616		dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
5617		if (!state_expired(&lt, dp->dl_time))
5618			break;
5619		WARN_ON(!unhash_delegation_locked(dp));
5620		list_add(&dp->dl_recall_lru, &reaplist);
5621	}
5622	spin_unlock(&state_lock);
5623	while (!list_empty(&reaplist)) {
5624		dp = list_first_entry(&reaplist, struct nfs4_delegation,
5625					dl_recall_lru);
5626		list_del_init(&dp->dl_recall_lru);
5627		revoke_delegation(dp);
5628	}
5629
5630	spin_lock(&nn->client_lock);
5631	while (!list_empty(&nn->close_lru)) {
5632		oo = list_first_entry(&nn->close_lru, struct nfs4_openowner,
5633					oo_close_lru);
5634		if (!state_expired(&lt, oo->oo_time))
5635			break;
5636		list_del_init(&oo->oo_close_lru);
5637		stp = oo->oo_last_closed_stid;
5638		oo->oo_last_closed_stid = NULL;
5639		spin_unlock(&nn->client_lock);
5640		nfs4_put_stid(&stp->st_stid);
5641		spin_lock(&nn->client_lock);
5642	}
5643	spin_unlock(&nn->client_lock);
5644
5645	/*
5646	 * It's possible for a client to try and acquire an already held lock
5647	 * that is being held for a long time, and then lose interest in it.
5648	 * So, we clean out any un-revisited request after a lease period
5649	 * under the assumption that the client is no longer interested.
5650	 *
5651	 * RFC5661, sec. 9.6 states that the client must not rely on getting
5652	 * notifications and must continue to poll for locks, even when the
5653	 * server supports them. Thus this shouldn't lead to clients blocking
5654	 * indefinitely once the lock does become free.
5655	 */
5656	BUG_ON(!list_empty(&reaplist));
5657	spin_lock(&nn->blocked_locks_lock);
5658	while (!list_empty(&nn->blocked_locks_lru)) {
5659		nbl = list_first_entry(&nn->blocked_locks_lru,
5660					struct nfsd4_blocked_lock, nbl_lru);
5661		if (!state_expired(&lt, nbl->nbl_time))
5662			break;
5663		list_move(&nbl->nbl_lru, &reaplist);
5664		list_del_init(&nbl->nbl_list);
5665	}
5666	spin_unlock(&nn->blocked_locks_lock);
5667
5668	while (!list_empty(&reaplist)) {
5669		nbl = list_first_entry(&reaplist,
5670					struct nfsd4_blocked_lock, nbl_lru);
5671		list_del_init(&nbl->nbl_lru);
5672		free_blocked_lock(nbl);
5673	}
5674#ifdef CONFIG_NFSD_V4_2_INTER_SSC
5675	/* service the server-to-server copy delayed unmount list */
5676	nfsd4_ssc_expire_umount(nn);
5677#endif
5678out:
5679	return max_t(time64_t, lt.new_timeo, NFSD_LAUNDROMAT_MINTIMEOUT);
5680}
5681
5682static struct workqueue_struct *laundry_wq;
5683static void laundromat_main(struct work_struct *);
5684
5685static void
5686laundromat_main(struct work_struct *laundry)
5687{
5688	time64_t t;
5689	struct delayed_work *dwork = to_delayed_work(laundry);
5690	struct nfsd_net *nn = container_of(dwork, struct nfsd_net,
5691					   laundromat_work);
5692
5693	t = nfs4_laundromat(nn);
5694	queue_delayed_work(laundry_wq, &nn->laundromat_work, t*HZ);
5695}
5696
5697static inline __be32 nfs4_check_fh(struct svc_fh *fhp, struct nfs4_stid *stp)
5698{
5699	if (!fh_match(&fhp->fh_handle, &stp->sc_file->fi_fhandle))
5700		return nfserr_bad_stateid;
5701	return nfs_ok;
5702}
5703
5704static
5705__be32 nfs4_check_openmode(struct nfs4_ol_stateid *stp, int flags)
5706{
5707        __be32 status = nfserr_openmode;
5708
5709	/* For lock stateid's, we test the parent open, not the lock: */
5710	if (stp->st_openstp)
5711		stp = stp->st_openstp;
5712	if ((flags & WR_STATE) && !access_permit_write(stp))
5713                goto out;
5714	if ((flags & RD_STATE) && !access_permit_read(stp))
5715                goto out;
5716	status = nfs_ok;
5717out:
5718	return status;
5719}
5720
5721static inline __be32
5722check_special_stateids(struct net *net, svc_fh *current_fh, stateid_t *stateid, int flags)
5723{
5724	if (ONE_STATEID(stateid) && (flags & RD_STATE))
5725		return nfs_ok;
5726	else if (opens_in_grace(net)) {
5727		/* Answer in remaining cases depends on existence of
5728		 * conflicting state; so we must wait out the grace period. */
5729		return nfserr_grace;
5730	} else if (flags & WR_STATE)
5731		return nfs4_share_conflict(current_fh,
5732				NFS4_SHARE_DENY_WRITE);
5733	else /* (flags & RD_STATE) && ZERO_STATEID(stateid) */
5734		return nfs4_share_conflict(current_fh,
5735				NFS4_SHARE_DENY_READ);
5736}
5737
5738/*
5739 * Allow READ/WRITE during grace period on recovered state only for files
5740 * that are not able to provide mandatory locking.
5741 */
5742static inline int
5743grace_disallows_io(struct net *net, struct inode *inode)
5744{
5745	return opens_in_grace(net) && mandatory_lock(inode);
5746}
5747
5748static __be32 check_stateid_generation(stateid_t *in, stateid_t *ref, bool has_session)
5749{
5750	/*
5751	 * When sessions are used the stateid generation number is ignored
5752	 * when it is zero.
5753	 */
5754	if (has_session && in->si_generation == 0)
5755		return nfs_ok;
5756
5757	if (in->si_generation == ref->si_generation)
5758		return nfs_ok;
5759
5760	/* If the client sends us a stateid from the future, it's buggy: */
5761	if (nfsd4_stateid_generation_after(in, ref))
5762		return nfserr_bad_stateid;
5763	/*
5764	 * However, we could see a stateid from the past, even from a
5765	 * non-buggy client.  For example, if the client sends a lock
5766	 * while some IO is outstanding, the lock may bump si_generation
5767	 * while the IO is still in flight.  The client could avoid that
5768	 * situation by waiting for responses on all the IO requests,
5769	 * but better performance may result in retrying IO that
5770	 * receives an old_stateid error if requests are rarely
5771	 * reordered in flight:
5772	 */
5773	return nfserr_old_stateid;
5774}
5775
5776static __be32 nfsd4_stid_check_stateid_generation(stateid_t *in, struct nfs4_stid *s, bool has_session)
5777{
5778	__be32 ret;
5779
5780	spin_lock(&s->sc_lock);
5781	ret = nfsd4_verify_open_stid(s);
5782	if (ret == nfs_ok)
5783		ret = check_stateid_generation(in, &s->sc_stateid, has_session);
5784	spin_unlock(&s->sc_lock);
5785	return ret;
5786}
5787
5788static __be32 nfsd4_check_openowner_confirmed(struct nfs4_ol_stateid *