2 * Copyright (c) 2001 The Regents of the University of Michigan.
5 * Kendrick Smith <kmsmith@umich.edu>
6 * Andy Adamson <kandros@umich.edu>
8 * Redistribution and use in source and binary forms, with or without
9 * modification, are permitted provided that the following conditions
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.
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.
35 #include <linux/file.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>
49 #include "current_stateid.h"
53 #include "filecache.h"
55 #define NFSDDBG_FACILITY NFSDDBG_PROC
57 #define all_ones {{~0,~0},~0}
58 static const stateid_t one_stateid = {
60 .si_opaque = all_ones,
62 static const stateid_t zero_stateid = {
65 static const stateid_t currentstateid = {
68 static const stateid_t close_stateid = {
69 .si_generation = 0xffffffffU,
72 static u64 current_sessionid = 1;
74 #define ZERO_STATEID(stateid) (!memcmp((stateid), &zero_stateid, sizeof(stateid_t)))
75 #define ONE_STATEID(stateid) (!memcmp((stateid), &one_stateid, sizeof(stateid_t)))
76 #define CURRENT_STATEID(stateid) (!memcmp((stateid), ¤tstateid, sizeof(stateid_t)))
77 #define CLOSE_STATEID(stateid) (!memcmp((stateid), &close_stateid, sizeof(stateid_t)))
79 /* forward declarations */
80 static bool check_for_locks(struct nfs4_file *fp, struct nfs4_lockowner *lowner);
81 static void nfs4_free_ol_stateid(struct nfs4_stid *stid);
82 void nfsd4_end_grace(struct nfsd_net *nn);
83 static void _free_cpntf_state_locked(struct nfsd_net *nn, struct nfs4_cpntf_state *cps);
88 * Currently used for the del_recall_lru and file hash table. In an
89 * effort to decrease the scope of the client_mutex, this spinlock may
90 * eventually cover more:
92 static DEFINE_SPINLOCK(state_lock);
94 enum nfsd4_st_mutex_lock_subclass {
95 OPEN_STATEID_MUTEX = 0,
96 LOCK_STATEID_MUTEX = 1,
100 * A waitqueue for all in-progress 4.0 CLOSE operations that are waiting for
101 * the refcount on the open stateid to drop.
103 static DECLARE_WAIT_QUEUE_HEAD(close_wq);
106 * A waitqueue where a writer to clients/#/ctl destroying a client can
107 * wait for cl_rpc_users to drop to 0 and then for the client to be
110 static DECLARE_WAIT_QUEUE_HEAD(expiry_wq);
112 static struct kmem_cache *client_slab;
113 static struct kmem_cache *openowner_slab;
114 static struct kmem_cache *lockowner_slab;
115 static struct kmem_cache *file_slab;
116 static struct kmem_cache *stateid_slab;
117 static struct kmem_cache *deleg_slab;
118 static struct kmem_cache *odstate_slab;
120 static void free_session(struct nfsd4_session *);
122 static const struct nfsd4_callback_ops nfsd4_cb_recall_ops;
123 static const struct nfsd4_callback_ops nfsd4_cb_notify_lock_ops;
125 static bool is_session_dead(struct nfsd4_session *ses)
127 return ses->se_flags & NFS4_SESSION_DEAD;
130 static __be32 mark_session_dead_locked(struct nfsd4_session *ses, int ref_held_by_me)
132 if (atomic_read(&ses->se_ref) > ref_held_by_me)
133 return nfserr_jukebox;
134 ses->se_flags |= NFS4_SESSION_DEAD;
138 static bool is_client_expired(struct nfs4_client *clp)
140 return clp->cl_time == 0;
143 static __be32 get_client_locked(struct nfs4_client *clp)
145 struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
147 lockdep_assert_held(&nn->client_lock);
149 if (is_client_expired(clp))
150 return nfserr_expired;
151 atomic_inc(&clp->cl_rpc_users);
155 /* must be called under the client_lock */
157 renew_client_locked(struct nfs4_client *clp)
159 struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
161 if (is_client_expired(clp)) {
163 printk("%s: client (clientid %08x/%08x) already expired\n",
165 clp->cl_clientid.cl_boot,
166 clp->cl_clientid.cl_id);
170 dprintk("renewing client (clientid %08x/%08x)\n",
171 clp->cl_clientid.cl_boot,
172 clp->cl_clientid.cl_id);
173 list_move_tail(&clp->cl_lru, &nn->client_lru);
174 clp->cl_time = ktime_get_boottime_seconds();
177 static void put_client_renew_locked(struct nfs4_client *clp)
179 struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
181 lockdep_assert_held(&nn->client_lock);
183 if (!atomic_dec_and_test(&clp->cl_rpc_users))
185 if (!is_client_expired(clp))
186 renew_client_locked(clp);
188 wake_up_all(&expiry_wq);
191 static void put_client_renew(struct nfs4_client *clp)
193 struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
195 if (!atomic_dec_and_lock(&clp->cl_rpc_users, &nn->client_lock))
197 if (!is_client_expired(clp))
198 renew_client_locked(clp);
200 wake_up_all(&expiry_wq);
201 spin_unlock(&nn->client_lock);
204 static __be32 nfsd4_get_session_locked(struct nfsd4_session *ses)
208 if (is_session_dead(ses))
209 return nfserr_badsession;
210 status = get_client_locked(ses->se_client);
213 atomic_inc(&ses->se_ref);
217 static void nfsd4_put_session_locked(struct nfsd4_session *ses)
219 struct nfs4_client *clp = ses->se_client;
220 struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
222 lockdep_assert_held(&nn->client_lock);
224 if (atomic_dec_and_test(&ses->se_ref) && is_session_dead(ses))
226 put_client_renew_locked(clp);
229 static void nfsd4_put_session(struct nfsd4_session *ses)
231 struct nfs4_client *clp = ses->se_client;
232 struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
234 spin_lock(&nn->client_lock);
235 nfsd4_put_session_locked(ses);
236 spin_unlock(&nn->client_lock);
239 static struct nfsd4_blocked_lock *
240 find_blocked_lock(struct nfs4_lockowner *lo, struct knfsd_fh *fh,
243 struct nfsd4_blocked_lock *cur, *found = NULL;
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);
254 spin_unlock(&nn->blocked_locks_lock);
256 locks_delete_block(&found->nbl_lock);
260 static struct nfsd4_blocked_lock *
261 find_or_allocate_block(struct nfs4_lockowner *lo, struct knfsd_fh *fh,
264 struct nfsd4_blocked_lock *nbl;
266 nbl = find_blocked_lock(lo, fh, nn);
268 nbl= kmalloc(sizeof(*nbl), GFP_KERNEL);
270 fh_copy_shallow(&nbl->nbl_fh, fh);
271 locks_init_lock(&nbl->nbl_lock);
272 nfsd4_init_cb(&nbl->nbl_cb, lo->lo_owner.so_client,
273 &nfsd4_cb_notify_lock_ops,
274 NFSPROC4_CLNT_CB_NOTIFY_LOCK);
281 free_blocked_lock(struct nfsd4_blocked_lock *nbl)
283 locks_delete_block(&nbl->nbl_lock);
284 locks_release_private(&nbl->nbl_lock);
289 remove_blocked_locks(struct nfs4_lockowner *lo)
291 struct nfs4_client *clp = lo->lo_owner.so_client;
292 struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
293 struct nfsd4_blocked_lock *nbl;
296 /* Dequeue all blocked locks */
297 spin_lock(&nn->blocked_locks_lock);
298 while (!list_empty(&lo->lo_blocked)) {
299 nbl = list_first_entry(&lo->lo_blocked,
300 struct nfsd4_blocked_lock,
302 list_del_init(&nbl->nbl_list);
303 list_move(&nbl->nbl_lru, &reaplist);
305 spin_unlock(&nn->blocked_locks_lock);
308 while (!list_empty(&reaplist)) {
309 nbl = list_first_entry(&reaplist, struct nfsd4_blocked_lock,
311 list_del_init(&nbl->nbl_lru);
312 free_blocked_lock(nbl);
317 nfsd4_cb_notify_lock_prepare(struct nfsd4_callback *cb)
319 struct nfsd4_blocked_lock *nbl = container_of(cb,
320 struct nfsd4_blocked_lock, nbl_cb);
321 locks_delete_block(&nbl->nbl_lock);
325 nfsd4_cb_notify_lock_done(struct nfsd4_callback *cb, struct rpc_task *task)
328 * Since this is just an optimization, we don't try very hard if it
329 * turns out not to succeed. We'll requeue it on NFS4ERR_DELAY, and
330 * just quit trying on anything else.
332 switch (task->tk_status) {
334 rpc_delay(task, 1 * HZ);
342 nfsd4_cb_notify_lock_release(struct nfsd4_callback *cb)
344 struct nfsd4_blocked_lock *nbl = container_of(cb,
345 struct nfsd4_blocked_lock, nbl_cb);
347 free_blocked_lock(nbl);
350 static const struct nfsd4_callback_ops nfsd4_cb_notify_lock_ops = {
351 .prepare = nfsd4_cb_notify_lock_prepare,
352 .done = nfsd4_cb_notify_lock_done,
353 .release = nfsd4_cb_notify_lock_release,
356 static inline struct nfs4_stateowner *
357 nfs4_get_stateowner(struct nfs4_stateowner *sop)
359 atomic_inc(&sop->so_count);
364 same_owner_str(struct nfs4_stateowner *sop, struct xdr_netobj *owner)
366 return (sop->so_owner.len == owner->len) &&
367 0 == memcmp(sop->so_owner.data, owner->data, owner->len);
370 static struct nfs4_openowner *
371 find_openstateowner_str_locked(unsigned int hashval, struct nfsd4_open *open,
372 struct nfs4_client *clp)
374 struct nfs4_stateowner *so;
376 lockdep_assert_held(&clp->cl_lock);
378 list_for_each_entry(so, &clp->cl_ownerstr_hashtbl[hashval],
380 if (!so->so_is_open_owner)
382 if (same_owner_str(so, &open->op_owner))
383 return openowner(nfs4_get_stateowner(so));
388 static struct nfs4_openowner *
389 find_openstateowner_str(unsigned int hashval, struct nfsd4_open *open,
390 struct nfs4_client *clp)
392 struct nfs4_openowner *oo;
394 spin_lock(&clp->cl_lock);
395 oo = find_openstateowner_str_locked(hashval, open, clp);
396 spin_unlock(&clp->cl_lock);
401 opaque_hashval(const void *ptr, int nbytes)
403 unsigned char *cptr = (unsigned char *) ptr;
413 static void nfsd4_free_file_rcu(struct rcu_head *rcu)
415 struct nfs4_file *fp = container_of(rcu, struct nfs4_file, fi_rcu);
417 kmem_cache_free(file_slab, fp);
421 put_nfs4_file(struct nfs4_file *fi)
423 might_lock(&state_lock);
425 if (refcount_dec_and_lock(&fi->fi_ref, &state_lock)) {
426 hlist_del_rcu(&fi->fi_hash);
427 spin_unlock(&state_lock);
428 WARN_ON_ONCE(!list_empty(&fi->fi_clnt_odstate));
429 WARN_ON_ONCE(!list_empty(&fi->fi_delegations));
430 call_rcu(&fi->fi_rcu, nfsd4_free_file_rcu);
434 static struct nfsd_file *
435 __nfs4_get_fd(struct nfs4_file *f, int oflag)
437 if (f->fi_fds[oflag])
438 return nfsd_file_get(f->fi_fds[oflag]);
442 static struct nfsd_file *
443 find_writeable_file_locked(struct nfs4_file *f)
445 struct nfsd_file *ret;
447 lockdep_assert_held(&f->fi_lock);
449 ret = __nfs4_get_fd(f, O_WRONLY);
451 ret = __nfs4_get_fd(f, O_RDWR);
455 static struct nfsd_file *
456 find_writeable_file(struct nfs4_file *f)
458 struct nfsd_file *ret;
460 spin_lock(&f->fi_lock);
461 ret = find_writeable_file_locked(f);
462 spin_unlock(&f->fi_lock);
467 static struct nfsd_file *
468 find_readable_file_locked(struct nfs4_file *f)
470 struct nfsd_file *ret;
472 lockdep_assert_held(&f->fi_lock);
474 ret = __nfs4_get_fd(f, O_RDONLY);
476 ret = __nfs4_get_fd(f, O_RDWR);
480 static struct nfsd_file *
481 find_readable_file(struct nfs4_file *f)
483 struct nfsd_file *ret;
485 spin_lock(&f->fi_lock);
486 ret = find_readable_file_locked(f);
487 spin_unlock(&f->fi_lock);
493 find_any_file(struct nfs4_file *f)
495 struct nfsd_file *ret;
497 spin_lock(&f->fi_lock);
498 ret = __nfs4_get_fd(f, O_RDWR);
500 ret = __nfs4_get_fd(f, O_WRONLY);
502 ret = __nfs4_get_fd(f, O_RDONLY);
504 spin_unlock(&f->fi_lock);
508 static atomic_long_t num_delegations;
509 unsigned long max_delegations;
512 * Open owner state (share locks)
515 /* hash tables for lock and open owners */
516 #define OWNER_HASH_BITS 8
517 #define OWNER_HASH_SIZE (1 << OWNER_HASH_BITS)
518 #define OWNER_HASH_MASK (OWNER_HASH_SIZE - 1)
520 static unsigned int ownerstr_hashval(struct xdr_netobj *ownername)
524 ret = opaque_hashval(ownername->data, ownername->len);
525 return ret & OWNER_HASH_MASK;
528 /* hash table for nfs4_file */
529 #define FILE_HASH_BITS 8
530 #define FILE_HASH_SIZE (1 << FILE_HASH_BITS)
532 static unsigned int nfsd_fh_hashval(struct knfsd_fh *fh)
534 return jhash2(fh->fh_base.fh_pad, XDR_QUADLEN(fh->fh_size), 0);
537 static unsigned int file_hashval(struct knfsd_fh *fh)
539 return nfsd_fh_hashval(fh) & (FILE_HASH_SIZE - 1);
542 static struct hlist_head file_hashtbl[FILE_HASH_SIZE];
545 __nfs4_file_get_access(struct nfs4_file *fp, u32 access)
547 lockdep_assert_held(&fp->fi_lock);
549 if (access & NFS4_SHARE_ACCESS_WRITE)
550 atomic_inc(&fp->fi_access[O_WRONLY]);
551 if (access & NFS4_SHARE_ACCESS_READ)
552 atomic_inc(&fp->fi_access[O_RDONLY]);
556 nfs4_file_get_access(struct nfs4_file *fp, u32 access)
558 lockdep_assert_held(&fp->fi_lock);
560 /* Does this access mode make sense? */
561 if (access & ~NFS4_SHARE_ACCESS_BOTH)
564 /* Does it conflict with a deny mode already set? */
565 if ((access & fp->fi_share_deny) != 0)
566 return nfserr_share_denied;
568 __nfs4_file_get_access(fp, access);
572 static __be32 nfs4_file_check_deny(struct nfs4_file *fp, u32 deny)
574 /* Common case is that there is no deny mode. */
576 /* Does this deny mode make sense? */
577 if (deny & ~NFS4_SHARE_DENY_BOTH)
580 if ((deny & NFS4_SHARE_DENY_READ) &&
581 atomic_read(&fp->fi_access[O_RDONLY]))
582 return nfserr_share_denied;
584 if ((deny & NFS4_SHARE_DENY_WRITE) &&
585 atomic_read(&fp->fi_access[O_WRONLY]))
586 return nfserr_share_denied;
591 static void __nfs4_file_put_access(struct nfs4_file *fp, int oflag)
593 might_lock(&fp->fi_lock);
595 if (atomic_dec_and_lock(&fp->fi_access[oflag], &fp->fi_lock)) {
596 struct nfsd_file *f1 = NULL;
597 struct nfsd_file *f2 = NULL;
599 swap(f1, fp->fi_fds[oflag]);
600 if (atomic_read(&fp->fi_access[1 - oflag]) == 0)
601 swap(f2, fp->fi_fds[O_RDWR]);
602 spin_unlock(&fp->fi_lock);
610 static void nfs4_file_put_access(struct nfs4_file *fp, u32 access)
612 WARN_ON_ONCE(access & ~NFS4_SHARE_ACCESS_BOTH);
614 if (access & NFS4_SHARE_ACCESS_WRITE)
615 __nfs4_file_put_access(fp, O_WRONLY);
616 if (access & NFS4_SHARE_ACCESS_READ)
617 __nfs4_file_put_access(fp, O_RDONLY);
621 * Allocate a new open/delegation state counter. This is needed for
622 * pNFS for proper return on close semantics.
624 * Note that we only allocate it for pNFS-enabled exports, otherwise
625 * all pointers to struct nfs4_clnt_odstate are always NULL.
627 static struct nfs4_clnt_odstate *
628 alloc_clnt_odstate(struct nfs4_client *clp)
630 struct nfs4_clnt_odstate *co;
632 co = kmem_cache_zalloc(odstate_slab, GFP_KERNEL);
635 refcount_set(&co->co_odcount, 1);
641 hash_clnt_odstate_locked(struct nfs4_clnt_odstate *co)
643 struct nfs4_file *fp = co->co_file;
645 lockdep_assert_held(&fp->fi_lock);
646 list_add(&co->co_perfile, &fp->fi_clnt_odstate);
650 get_clnt_odstate(struct nfs4_clnt_odstate *co)
653 refcount_inc(&co->co_odcount);
657 put_clnt_odstate(struct nfs4_clnt_odstate *co)
659 struct nfs4_file *fp;
665 if (refcount_dec_and_lock(&co->co_odcount, &fp->fi_lock)) {
666 list_del(&co->co_perfile);
667 spin_unlock(&fp->fi_lock);
669 nfsd4_return_all_file_layouts(co->co_client, fp);
670 kmem_cache_free(odstate_slab, co);
674 static struct nfs4_clnt_odstate *
675 find_or_hash_clnt_odstate(struct nfs4_file *fp, struct nfs4_clnt_odstate *new)
677 struct nfs4_clnt_odstate *co;
678 struct nfs4_client *cl;
685 spin_lock(&fp->fi_lock);
686 list_for_each_entry(co, &fp->fi_clnt_odstate, co_perfile) {
687 if (co->co_client == cl) {
688 get_clnt_odstate(co);
694 hash_clnt_odstate_locked(new);
696 spin_unlock(&fp->fi_lock);
700 struct nfs4_stid *nfs4_alloc_stid(struct nfs4_client *cl, struct kmem_cache *slab,
701 void (*sc_free)(struct nfs4_stid *))
703 struct nfs4_stid *stid;
706 stid = kmem_cache_zalloc(slab, GFP_KERNEL);
710 idr_preload(GFP_KERNEL);
711 spin_lock(&cl->cl_lock);
712 /* Reserving 0 for start of file in nfsdfs "states" file: */
713 new_id = idr_alloc_cyclic(&cl->cl_stateids, stid, 1, 0, GFP_NOWAIT);
714 spin_unlock(&cl->cl_lock);
719 stid->sc_free = sc_free;
720 stid->sc_client = cl;
721 stid->sc_stateid.si_opaque.so_id = new_id;
722 stid->sc_stateid.si_opaque.so_clid = cl->cl_clientid;
723 /* Will be incremented before return to client: */
724 refcount_set(&stid->sc_count, 1);
725 spin_lock_init(&stid->sc_lock);
726 INIT_LIST_HEAD(&stid->sc_cp_list);
729 * It shouldn't be a problem to reuse an opaque stateid value.
730 * I don't think it is for 4.1. But with 4.0 I worry that, for
731 * example, a stray write retransmission could be accepted by
732 * the server when it should have been rejected. Therefore,
733 * adopt a trick from the sctp code to attempt to maximize the
734 * amount of time until an id is reused, by ensuring they always
735 * "increase" (mod INT_MAX):
739 kmem_cache_free(slab, stid);
744 * Create a unique stateid_t to represent each COPY.
746 static int nfs4_init_cp_state(struct nfsd_net *nn, copy_stateid_t *stid,
747 unsigned char sc_type)
751 stid->stid.si_opaque.so_clid.cl_boot = (u32)nn->boot_time;
752 stid->stid.si_opaque.so_clid.cl_id = nn->s2s_cp_cl_id;
753 stid->sc_type = sc_type;
755 idr_preload(GFP_KERNEL);
756 spin_lock(&nn->s2s_cp_lock);
757 new_id = idr_alloc_cyclic(&nn->s2s_cp_stateids, stid, 0, 0, GFP_NOWAIT);
758 stid->stid.si_opaque.so_id = new_id;
759 spin_unlock(&nn->s2s_cp_lock);
766 int nfs4_init_copy_state(struct nfsd_net *nn, struct nfsd4_copy *copy)
768 return nfs4_init_cp_state(nn, ©->cp_stateid, NFS4_COPY_STID);
771 struct nfs4_cpntf_state *nfs4_alloc_init_cpntf_state(struct nfsd_net *nn,
772 struct nfs4_stid *p_stid)
774 struct nfs4_cpntf_state *cps;
776 cps = kzalloc(sizeof(struct nfs4_cpntf_state), GFP_KERNEL);
779 cps->cpntf_time = ktime_get_boottime_seconds();
780 refcount_set(&cps->cp_stateid.sc_count, 1);
781 if (!nfs4_init_cp_state(nn, &cps->cp_stateid, NFS4_COPYNOTIFY_STID))
783 spin_lock(&nn->s2s_cp_lock);
784 list_add(&cps->cp_list, &p_stid->sc_cp_list);
785 spin_unlock(&nn->s2s_cp_lock);
792 void nfs4_free_copy_state(struct nfsd4_copy *copy)
796 WARN_ON_ONCE(copy->cp_stateid.sc_type != NFS4_COPY_STID);
797 nn = net_generic(copy->cp_clp->net, nfsd_net_id);
798 spin_lock(&nn->s2s_cp_lock);
799 idr_remove(&nn->s2s_cp_stateids,
800 copy->cp_stateid.stid.si_opaque.so_id);
801 spin_unlock(&nn->s2s_cp_lock);
804 static void nfs4_free_cpntf_statelist(struct net *net, struct nfs4_stid *stid)
806 struct nfs4_cpntf_state *cps;
809 nn = net_generic(net, nfsd_net_id);
810 spin_lock(&nn->s2s_cp_lock);
811 while (!list_empty(&stid->sc_cp_list)) {
812 cps = list_first_entry(&stid->sc_cp_list,
813 struct nfs4_cpntf_state, cp_list);
814 _free_cpntf_state_locked(nn, cps);
816 spin_unlock(&nn->s2s_cp_lock);
819 static struct nfs4_ol_stateid * nfs4_alloc_open_stateid(struct nfs4_client *clp)
821 struct nfs4_stid *stid;
823 stid = nfs4_alloc_stid(clp, stateid_slab, nfs4_free_ol_stateid);
827 return openlockstateid(stid);
830 static void nfs4_free_deleg(struct nfs4_stid *stid)
832 kmem_cache_free(deleg_slab, stid);
833 atomic_long_dec(&num_delegations);
837 * When we recall a delegation, we should be careful not to hand it
838 * out again straight away.
839 * To ensure this we keep a pair of bloom filters ('new' and 'old')
840 * in which the filehandles of recalled delegations are "stored".
841 * If a filehandle appear in either filter, a delegation is blocked.
842 * When a delegation is recalled, the filehandle is stored in the "new"
844 * Every 30 seconds we swap the filters and clear the "new" one,
845 * unless both are empty of course.
847 * Each filter is 256 bits. We hash the filehandle to 32bit and use the
848 * low 3 bytes as hash-table indices.
850 * 'blocked_delegations_lock', which is always taken in block_delegations(),
851 * is used to manage concurrent access. Testing does not need the lock
852 * except when swapping the two filters.
854 static DEFINE_SPINLOCK(blocked_delegations_lock);
855 static struct bloom_pair {
856 int entries, old_entries;
858 int new; /* index into 'set' */
859 DECLARE_BITMAP(set[2], 256);
860 } blocked_delegations;
862 static int delegation_blocked(struct knfsd_fh *fh)
865 struct bloom_pair *bd = &blocked_delegations;
867 if (bd->entries == 0)
869 if (ktime_get_seconds() - bd->swap_time > 30) {
870 spin_lock(&blocked_delegations_lock);
871 if (ktime_get_seconds() - bd->swap_time > 30) {
872 bd->entries -= bd->old_entries;
873 bd->old_entries = bd->entries;
874 memset(bd->set[bd->new], 0,
877 bd->swap_time = ktime_get_seconds();
879 spin_unlock(&blocked_delegations_lock);
881 hash = jhash(&fh->fh_base, fh->fh_size, 0);
882 if (test_bit(hash&255, bd->set[0]) &&
883 test_bit((hash>>8)&255, bd->set[0]) &&
884 test_bit((hash>>16)&255, bd->set[0]))
887 if (test_bit(hash&255, bd->set[1]) &&
888 test_bit((hash>>8)&255, bd->set[1]) &&
889 test_bit((hash>>16)&255, bd->set[1]))
895 static void block_delegations(struct knfsd_fh *fh)
898 struct bloom_pair *bd = &blocked_delegations;
900 hash = jhash(&fh->fh_base, fh->fh_size, 0);
902 spin_lock(&blocked_delegations_lock);
903 __set_bit(hash&255, bd->set[bd->new]);
904 __set_bit((hash>>8)&255, bd->set[bd->new]);
905 __set_bit((hash>>16)&255, bd->set[bd->new]);
906 if (bd->entries == 0)
907 bd->swap_time = ktime_get_seconds();
909 spin_unlock(&blocked_delegations_lock);
912 static struct nfs4_delegation *
913 alloc_init_deleg(struct nfs4_client *clp, struct nfs4_file *fp,
914 struct svc_fh *current_fh,
915 struct nfs4_clnt_odstate *odstate)
917 struct nfs4_delegation *dp;
920 dprintk("NFSD alloc_init_deleg\n");
921 n = atomic_long_inc_return(&num_delegations);
922 if (n < 0 || n > max_delegations)
924 if (delegation_blocked(¤t_fh->fh_handle))
926 dp = delegstateid(nfs4_alloc_stid(clp, deleg_slab, nfs4_free_deleg));
931 * delegation seqid's are never incremented. The 4.1 special
932 * meaning of seqid 0 isn't meaningful, really, but let's avoid
933 * 0 anyway just for consistency and use 1:
935 dp->dl_stid.sc_stateid.si_generation = 1;
936 INIT_LIST_HEAD(&dp->dl_perfile);
937 INIT_LIST_HEAD(&dp->dl_perclnt);
938 INIT_LIST_HEAD(&dp->dl_recall_lru);
939 dp->dl_clnt_odstate = odstate;
940 get_clnt_odstate(odstate);
941 dp->dl_type = NFS4_OPEN_DELEGATE_READ;
943 nfsd4_init_cb(&dp->dl_recall, dp->dl_stid.sc_client,
944 &nfsd4_cb_recall_ops, NFSPROC4_CLNT_CB_RECALL);
946 dp->dl_stid.sc_file = fp;
949 atomic_long_dec(&num_delegations);
954 nfs4_put_stid(struct nfs4_stid *s)
956 struct nfs4_file *fp = s->sc_file;
957 struct nfs4_client *clp = s->sc_client;
959 might_lock(&clp->cl_lock);
961 if (!refcount_dec_and_lock(&s->sc_count, &clp->cl_lock)) {
962 wake_up_all(&close_wq);
965 idr_remove(&clp->cl_stateids, s->sc_stateid.si_opaque.so_id);
966 nfs4_free_cpntf_statelist(clp->net, s);
967 spin_unlock(&clp->cl_lock);
974 nfs4_inc_and_copy_stateid(stateid_t *dst, struct nfs4_stid *stid)
976 stateid_t *src = &stid->sc_stateid;
978 spin_lock(&stid->sc_lock);
979 if (unlikely(++src->si_generation == 0))
980 src->si_generation = 1;
981 memcpy(dst, src, sizeof(*dst));
982 spin_unlock(&stid->sc_lock);
985 static void put_deleg_file(struct nfs4_file *fp)
987 struct nfsd_file *nf = NULL;
989 spin_lock(&fp->fi_lock);
990 if (--fp->fi_delegees == 0)
991 swap(nf, fp->fi_deleg_file);
992 spin_unlock(&fp->fi_lock);
998 static void nfs4_unlock_deleg_lease(struct nfs4_delegation *dp)
1000 struct nfs4_file *fp = dp->dl_stid.sc_file;
1001 struct nfsd_file *nf = fp->fi_deleg_file;
1003 WARN_ON_ONCE(!fp->fi_delegees);
1005 vfs_setlease(nf->nf_file, F_UNLCK, NULL, (void **)&dp);
1009 static void destroy_unhashed_deleg(struct nfs4_delegation *dp)
1011 put_clnt_odstate(dp->dl_clnt_odstate);
1012 nfs4_unlock_deleg_lease(dp);
1013 nfs4_put_stid(&dp->dl_stid);
1016 void nfs4_unhash_stid(struct nfs4_stid *s)
1022 * nfs4_delegation_exists - Discover if this delegation already exists
1023 * @clp: a pointer to the nfs4_client we're granting a delegation to
1024 * @fp: a pointer to the nfs4_file we're granting a delegation on
1027 * On success: true iff an existing delegation is found
1031 nfs4_delegation_exists(struct nfs4_client *clp, struct nfs4_file *fp)
1033 struct nfs4_delegation *searchdp = NULL;
1034 struct nfs4_client *searchclp = NULL;
1036 lockdep_assert_held(&state_lock);
1037 lockdep_assert_held(&fp->fi_lock);
1039 list_for_each_entry(searchdp, &fp->fi_delegations, dl_perfile) {
1040 searchclp = searchdp->dl_stid.sc_client;
1041 if (clp == searchclp) {
1049 * hash_delegation_locked - Add a delegation to the appropriate lists
1050 * @dp: a pointer to the nfs4_delegation we are adding.
1051 * @fp: a pointer to the nfs4_file we're granting a delegation on
1054 * On success: NULL if the delegation was successfully hashed.
1056 * On error: -EAGAIN if one was previously granted to this
1057 * nfs4_client for this nfs4_file. Delegation is not hashed.
1062 hash_delegation_locked(struct nfs4_delegation *dp, struct nfs4_file *fp)
1064 struct nfs4_client *clp = dp->dl_stid.sc_client;
1066 lockdep_assert_held(&state_lock);
1067 lockdep_assert_held(&fp->fi_lock);
1069 if (nfs4_delegation_exists(clp, fp))
1071 refcount_inc(&dp->dl_stid.sc_count);
1072 dp->dl_stid.sc_type = NFS4_DELEG_STID;
1073 list_add(&dp->dl_perfile, &fp->fi_delegations);
1074 list_add(&dp->dl_perclnt, &clp->cl_delegations);
1079 unhash_delegation_locked(struct nfs4_delegation *dp)
1081 struct nfs4_file *fp = dp->dl_stid.sc_file;
1083 lockdep_assert_held(&state_lock);
1085 if (list_empty(&dp->dl_perfile))
1088 dp->dl_stid.sc_type = NFS4_CLOSED_DELEG_STID;
1089 /* Ensure that deleg break won't try to requeue it */
1091 spin_lock(&fp->fi_lock);
1092 list_del_init(&dp->dl_perclnt);
1093 list_del_init(&dp->dl_recall_lru);
1094 list_del_init(&dp->dl_perfile);
1095 spin_unlock(&fp->fi_lock);
1099 static void destroy_delegation(struct nfs4_delegation *dp)
1103 spin_lock(&state_lock);
1104 unhashed = unhash_delegation_locked(dp);
1105 spin_unlock(&state_lock);
1107 destroy_unhashed_deleg(dp);
1110 static void revoke_delegation(struct nfs4_delegation *dp)
1112 struct nfs4_client *clp = dp->dl_stid.sc_client;
1114 WARN_ON(!list_empty(&dp->dl_recall_lru));
1116 if (clp->cl_minorversion) {
1117 dp->dl_stid.sc_type = NFS4_REVOKED_DELEG_STID;
1118 refcount_inc(&dp->dl_stid.sc_count);
1119 spin_lock(&clp->cl_lock);
1120 list_add(&dp->dl_recall_lru, &clp->cl_revoked);
1121 spin_unlock(&clp->cl_lock);
1123 destroy_unhashed_deleg(dp);
1130 static unsigned int clientid_hashval(u32 id)
1132 return id & CLIENT_HASH_MASK;
1135 static unsigned int clientstr_hashval(struct xdr_netobj name)
1137 return opaque_hashval(name.data, 8) & CLIENT_HASH_MASK;
1141 * We store the NONE, READ, WRITE, and BOTH bits separately in the
1142 * st_{access,deny}_bmap field of the stateid, in order to track not
1143 * only what share bits are currently in force, but also what
1144 * combinations of share bits previous opens have used. This allows us
1145 * to enforce the recommendation of rfc 3530 14.2.19 that the server
1146 * return an error if the client attempt to downgrade to a combination
1147 * of share bits not explicable by closing some of its previous opens.
1149 * XXX: This enforcement is actually incomplete, since we don't keep
1150 * track of access/deny bit combinations; so, e.g., we allow:
1152 * OPEN allow read, deny write
1153 * OPEN allow both, deny none
1154 * DOWNGRADE allow read, deny none
1156 * which we should reject.
1159 bmap_to_share_mode(unsigned long bmap) {
1161 unsigned int access = 0;
1163 for (i = 1; i < 4; i++) {
1164 if (test_bit(i, &bmap))
1170 /* set share access for a given stateid */
1172 set_access(u32 access, struct nfs4_ol_stateid *stp)
1174 unsigned char mask = 1 << access;
1176 WARN_ON_ONCE(access > NFS4_SHARE_ACCESS_BOTH);
1177 stp->st_access_bmap |= mask;
1180 /* clear share access for a given stateid */
1182 clear_access(u32 access, struct nfs4_ol_stateid *stp)
1184 unsigned char mask = 1 << access;
1186 WARN_ON_ONCE(access > NFS4_SHARE_ACCESS_BOTH);
1187 stp->st_access_bmap &= ~mask;
1190 /* test whether a given stateid has access */
1192 test_access(u32 access, struct nfs4_ol_stateid *stp)
1194 unsigned char mask = 1 << access;
1196 return (bool)(stp->st_access_bmap & mask);
1199 /* set share deny for a given stateid */
1201 set_deny(u32 deny, struct nfs4_ol_stateid *stp)
1203 unsigned char mask = 1 << deny;
1205 WARN_ON_ONCE(deny > NFS4_SHARE_DENY_BOTH);
1206 stp->st_deny_bmap |= mask;
1209 /* clear share deny for a given stateid */
1211 clear_deny(u32 deny, struct nfs4_ol_stateid *stp)
1213 unsigned char mask = 1 << deny;
1215 WARN_ON_ONCE(deny > NFS4_SHARE_DENY_BOTH);
1216 stp->st_deny_bmap &= ~mask;
1219 /* test whether a given stateid is denying specific access */
1221 test_deny(u32 deny, struct nfs4_ol_stateid *stp)
1223 unsigned char mask = 1 << deny;
1225 return (bool)(stp->st_deny_bmap & mask);
1228 static int nfs4_access_to_omode(u32 access)
1230 switch (access & NFS4_SHARE_ACCESS_BOTH) {
1231 case NFS4_SHARE_ACCESS_READ:
1233 case NFS4_SHARE_ACCESS_WRITE:
1235 case NFS4_SHARE_ACCESS_BOTH:
1243 * A stateid that had a deny mode associated with it is being released
1244 * or downgraded. Recalculate the deny mode on the file.
1247 recalculate_deny_mode(struct nfs4_file *fp)
1249 struct nfs4_ol_stateid *stp;
1251 spin_lock(&fp->fi_lock);
1252 fp->fi_share_deny = 0;
1253 list_for_each_entry(stp, &fp->fi_stateids, st_perfile)
1254 fp->fi_share_deny |= bmap_to_share_mode(stp->st_deny_bmap);
1255 spin_unlock(&fp->fi_lock);
1259 reset_union_bmap_deny(u32 deny, struct nfs4_ol_stateid *stp)
1262 bool change = false;
1264 for (i = 1; i < 4; i++) {
1265 if ((i & deny) != i) {
1271 /* Recalculate per-file deny mode if there was a change */
1273 recalculate_deny_mode(stp->st_stid.sc_file);
1276 /* release all access and file references for a given stateid */
1278 release_all_access(struct nfs4_ol_stateid *stp)
1281 struct nfs4_file *fp = stp->st_stid.sc_file;
1283 if (fp && stp->st_deny_bmap != 0)
1284 recalculate_deny_mode(fp);
1286 for (i = 1; i < 4; i++) {
1287 if (test_access(i, stp))
1288 nfs4_file_put_access(stp->st_stid.sc_file, i);
1289 clear_access(i, stp);
1293 static inline void nfs4_free_stateowner(struct nfs4_stateowner *sop)
1295 kfree(sop->so_owner.data);
1296 sop->so_ops->so_free(sop);
1299 static void nfs4_put_stateowner(struct nfs4_stateowner *sop)
1301 struct nfs4_client *clp = sop->so_client;
1303 might_lock(&clp->cl_lock);
1305 if (!atomic_dec_and_lock(&sop->so_count, &clp->cl_lock))
1307 sop->so_ops->so_unhash(sop);
1308 spin_unlock(&clp->cl_lock);
1309 nfs4_free_stateowner(sop);
1312 static bool unhash_ol_stateid(struct nfs4_ol_stateid *stp)
1314 struct nfs4_file *fp = stp->st_stid.sc_file;
1316 lockdep_assert_held(&stp->st_stateowner->so_client->cl_lock);
1318 if (list_empty(&stp->st_perfile))
1321 spin_lock(&fp->fi_lock);
1322 list_del_init(&stp->st_perfile);
1323 spin_unlock(&fp->fi_lock);
1324 list_del(&stp->st_perstateowner);
1328 static void nfs4_free_ol_stateid(struct nfs4_stid *stid)
1330 struct nfs4_ol_stateid *stp = openlockstateid(stid);
1332 put_clnt_odstate(stp->st_clnt_odstate);
1333 release_all_access(stp);
1334 if (stp->st_stateowner)
1335 nfs4_put_stateowner(stp->st_stateowner);
1336 kmem_cache_free(stateid_slab, stid);
1339 static void nfs4_free_lock_stateid(struct nfs4_stid *stid)
1341 struct nfs4_ol_stateid *stp = openlockstateid(stid);
1342 struct nfs4_lockowner *lo = lockowner(stp->st_stateowner);
1343 struct nfsd_file *nf;
1345 nf = find_any_file(stp->st_stid.sc_file);
1347 get_file(nf->nf_file);
1348 filp_close(nf->nf_file, (fl_owner_t)lo);
1351 nfs4_free_ol_stateid(stid);
1355 * Put the persistent reference to an already unhashed generic stateid, while
1356 * holding the cl_lock. If it's the last reference, then put it onto the
1357 * reaplist for later destruction.
1359 static void put_ol_stateid_locked(struct nfs4_ol_stateid *stp,
1360 struct list_head *reaplist)
1362 struct nfs4_stid *s = &stp->st_stid;
1363 struct nfs4_client *clp = s->sc_client;
1365 lockdep_assert_held(&clp->cl_lock);
1367 WARN_ON_ONCE(!list_empty(&stp->st_locks));
1369 if (!refcount_dec_and_test(&s->sc_count)) {
1370 wake_up_all(&close_wq);
1374 idr_remove(&clp->cl_stateids, s->sc_stateid.si_opaque.so_id);
1375 list_add(&stp->st_locks, reaplist);
1378 static bool unhash_lock_stateid(struct nfs4_ol_stateid *stp)
1380 lockdep_assert_held(&stp->st_stid.sc_client->cl_lock);
1382 list_del_init(&stp->st_locks);
1383 nfs4_unhash_stid(&stp->st_stid);
1384 return unhash_ol_stateid(stp);
1387 static void release_lock_stateid(struct nfs4_ol_stateid *stp)
1389 struct nfs4_client *clp = stp->st_stid.sc_client;
1392 spin_lock(&clp->cl_lock);
1393 unhashed = unhash_lock_stateid(stp);
1394 spin_unlock(&clp->cl_lock);
1396 nfs4_put_stid(&stp->st_stid);
1399 static void unhash_lockowner_locked(struct nfs4_lockowner *lo)
1401 struct nfs4_client *clp = lo->lo_owner.so_client;
1403 lockdep_assert_held(&clp->cl_lock);
1405 list_del_init(&lo->lo_owner.so_strhash);
1409 * Free a list of generic stateids that were collected earlier after being
1413 free_ol_stateid_reaplist(struct list_head *reaplist)
1415 struct nfs4_ol_stateid *stp;
1416 struct nfs4_file *fp;
1420 while (!list_empty(reaplist)) {
1421 stp = list_first_entry(reaplist, struct nfs4_ol_stateid,
1423 list_del(&stp->st_locks);
1424 fp = stp->st_stid.sc_file;
1425 stp->st_stid.sc_free(&stp->st_stid);
1431 static void release_open_stateid_locks(struct nfs4_ol_stateid *open_stp,
1432 struct list_head *reaplist)
1434 struct nfs4_ol_stateid *stp;
1436 lockdep_assert_held(&open_stp->st_stid.sc_client->cl_lock);
1438 while (!list_empty(&open_stp->st_locks)) {
1439 stp = list_entry(open_stp->st_locks.next,
1440 struct nfs4_ol_stateid, st_locks);
1441 WARN_ON(!unhash_lock_stateid(stp));
1442 put_ol_stateid_locked(stp, reaplist);
1446 static bool unhash_open_stateid(struct nfs4_ol_stateid *stp,
1447 struct list_head *reaplist)
1451 lockdep_assert_held(&stp->st_stid.sc_client->cl_lock);
1453 unhashed = unhash_ol_stateid(stp);
1454 release_open_stateid_locks(stp, reaplist);
1458 static void release_open_stateid(struct nfs4_ol_stateid *stp)
1460 LIST_HEAD(reaplist);
1462 spin_lock(&stp->st_stid.sc_client->cl_lock);
1463 if (unhash_open_stateid(stp, &reaplist))
1464 put_ol_stateid_locked(stp, &reaplist);
1465 spin_unlock(&stp->st_stid.sc_client->cl_lock);
1466 free_ol_stateid_reaplist(&reaplist);
1469 static void unhash_openowner_locked(struct nfs4_openowner *oo)
1471 struct nfs4_client *clp = oo->oo_owner.so_client;
1473 lockdep_assert_held(&clp->cl_lock);
1475 list_del_init(&oo->oo_owner.so_strhash);
1476 list_del_init(&oo->oo_perclient);
1479 static void release_last_closed_stateid(struct nfs4_openowner *oo)
1481 struct nfsd_net *nn = net_generic(oo->oo_owner.so_client->net,
1483 struct nfs4_ol_stateid *s;
1485 spin_lock(&nn->client_lock);
1486 s = oo->oo_last_closed_stid;
1488 list_del_init(&oo->oo_close_lru);
1489 oo->oo_last_closed_stid = NULL;
1491 spin_unlock(&nn->client_lock);
1493 nfs4_put_stid(&s->st_stid);
1496 static void release_openowner(struct nfs4_openowner *oo)
1498 struct nfs4_ol_stateid *stp;
1499 struct nfs4_client *clp = oo->oo_owner.so_client;
1500 struct list_head reaplist;
1502 INIT_LIST_HEAD(&reaplist);
1504 spin_lock(&clp->cl_lock);
1505 unhash_openowner_locked(oo);
1506 while (!list_empty(&oo->oo_owner.so_stateids)) {
1507 stp = list_first_entry(&oo->oo_owner.so_stateids,
1508 struct nfs4_ol_stateid, st_perstateowner);
1509 if (unhash_open_stateid(stp, &reaplist))
1510 put_ol_stateid_locked(stp, &reaplist);
1512 spin_unlock(&clp->cl_lock);
1513 free_ol_stateid_reaplist(&reaplist);
1514 release_last_closed_stateid(oo);
1515 nfs4_put_stateowner(&oo->oo_owner);
1519 hash_sessionid(struct nfs4_sessionid *sessionid)
1521 struct nfsd4_sessionid *sid = (struct nfsd4_sessionid *)sessionid;
1523 return sid->sequence % SESSION_HASH_SIZE;
1526 #ifdef CONFIG_SUNRPC_DEBUG
1528 dump_sessionid(const char *fn, struct nfs4_sessionid *sessionid)
1530 u32 *ptr = (u32 *)(&sessionid->data[0]);
1531 dprintk("%s: %u:%u:%u:%u\n", fn, ptr[0], ptr[1], ptr[2], ptr[3]);
1535 dump_sessionid(const char *fn, struct nfs4_sessionid *sessionid)
1541 * Bump the seqid on cstate->replay_owner, and clear replay_owner if it
1542 * won't be used for replay.
1544 void nfsd4_bump_seqid(struct nfsd4_compound_state *cstate, __be32 nfserr)
1546 struct nfs4_stateowner *so = cstate->replay_owner;
1548 if (nfserr == nfserr_replay_me)
1551 if (!seqid_mutating_err(ntohl(nfserr))) {
1552 nfsd4_cstate_clear_replay(cstate);
1557 if (so->so_is_open_owner)
1558 release_last_closed_stateid(openowner(so));
1564 gen_sessionid(struct nfsd4_session *ses)
1566 struct nfs4_client *clp = ses->se_client;
1567 struct nfsd4_sessionid *sid;
1569 sid = (struct nfsd4_sessionid *)ses->se_sessionid.data;
1570 sid->clientid = clp->cl_clientid;
1571 sid->sequence = current_sessionid++;
1576 * The protocol defines ca_maxresponssize_cached to include the size of
1577 * the rpc header, but all we need to cache is the data starting after
1578 * the end of the initial SEQUENCE operation--the rest we regenerate
1579 * each time. Therefore we can advertise a ca_maxresponssize_cached
1580 * value that is the number of bytes in our cache plus a few additional
1581 * bytes. In order to stay on the safe side, and not promise more than
1582 * we can cache, those additional bytes must be the minimum possible: 24
1583 * bytes of rpc header (xid through accept state, with AUTH_NULL
1584 * verifier), 12 for the compound header (with zero-length tag), and 44
1585 * for the SEQUENCE op response:
1587 #define NFSD_MIN_HDR_SEQ_SZ (24 + 12 + 44)
1590 free_session_slots(struct nfsd4_session *ses)
1594 for (i = 0; i < ses->se_fchannel.maxreqs; i++) {
1595 free_svc_cred(&ses->se_slots[i]->sl_cred);
1596 kfree(ses->se_slots[i]);
1601 * We don't actually need to cache the rpc and session headers, so we
1602 * can allocate a little less for each slot:
1604 static inline u32 slot_bytes(struct nfsd4_channel_attrs *ca)
1608 if (ca->maxresp_cached < NFSD_MIN_HDR_SEQ_SZ)
1611 size = ca->maxresp_cached - NFSD_MIN_HDR_SEQ_SZ;
1612 return size + sizeof(struct nfsd4_slot);
1616 * XXX: If we run out of reserved DRC memory we could (up to a point)
1617 * re-negotiate active sessions and reduce their slot usage to make
1618 * room for new connections. For now we just fail the create session.
1620 static u32 nfsd4_get_drc_mem(struct nfsd4_channel_attrs *ca, struct nfsd_net *nn)
1622 u32 slotsize = slot_bytes(ca);
1623 u32 num = ca->maxreqs;
1624 unsigned long avail, total_avail;
1625 unsigned int scale_factor;
1627 spin_lock(&nfsd_drc_lock);
1628 if (nfsd_drc_max_mem > nfsd_drc_mem_used)
1629 total_avail = nfsd_drc_max_mem - nfsd_drc_mem_used;
1631 /* We have handed out more space than we chose in
1632 * set_max_drc() to allow. That isn't really a
1633 * problem as long as that doesn't make us think we
1634 * have lots more due to integer overflow.
1637 avail = min((unsigned long)NFSD_MAX_MEM_PER_SESSION, total_avail);
1639 * Never use more than a fraction of the remaining memory,
1640 * unless it's the only way to give this client a slot.
1641 * The chosen fraction is either 1/8 or 1/number of threads,
1642 * whichever is smaller. This ensures there are adequate
1643 * slots to support multiple clients per thread.
1644 * Give the client one slot even if that would require
1645 * over-allocation--it is better than failure.
1647 scale_factor = max_t(unsigned int, 8, nn->nfsd_serv->sv_nrthreads);
1649 avail = clamp_t(unsigned long, avail, slotsize,
1650 total_avail/scale_factor);
1651 num = min_t(int, num, avail / slotsize);
1652 num = max_t(int, num, 1);
1653 nfsd_drc_mem_used += num * slotsize;
1654 spin_unlock(&nfsd_drc_lock);
1659 static void nfsd4_put_drc_mem(struct nfsd4_channel_attrs *ca)
1661 int slotsize = slot_bytes(ca);
1663 spin_lock(&nfsd_drc_lock);
1664 nfsd_drc_mem_used -= slotsize * ca->maxreqs;
1665 spin_unlock(&nfsd_drc_lock);
1668 static struct nfsd4_session *alloc_session(struct nfsd4_channel_attrs *fattrs,
1669 struct nfsd4_channel_attrs *battrs)
1671 int numslots = fattrs->maxreqs;
1672 int slotsize = slot_bytes(fattrs);
1673 struct nfsd4_session *new;
1676 BUILD_BUG_ON(NFSD_MAX_SLOTS_PER_SESSION * sizeof(struct nfsd4_slot *)
1677 + sizeof(struct nfsd4_session) > PAGE_SIZE);
1678 mem = numslots * sizeof(struct nfsd4_slot *);
1680 new = kzalloc(sizeof(*new) + mem, GFP_KERNEL);
1683 /* allocate each struct nfsd4_slot and data cache in one piece */
1684 for (i = 0; i < numslots; i++) {
1685 new->se_slots[i] = kzalloc(slotsize, GFP_KERNEL);
1686 if (!new->se_slots[i])
1690 memcpy(&new->se_fchannel, fattrs, sizeof(struct nfsd4_channel_attrs));
1691 memcpy(&new->se_bchannel, battrs, sizeof(struct nfsd4_channel_attrs));
1696 kfree(new->se_slots[i]);
1701 static void free_conn(struct nfsd4_conn *c)
1703 svc_xprt_put(c->cn_xprt);
1707 static void nfsd4_conn_lost(struct svc_xpt_user *u)
1709 struct nfsd4_conn *c = container_of(u, struct nfsd4_conn, cn_xpt_user);
1710 struct nfs4_client *clp = c->cn_session->se_client;
1712 spin_lock(&clp->cl_lock);
1713 if (!list_empty(&c->cn_persession)) {
1714 list_del(&c->cn_persession);
1717 nfsd4_probe_callback(clp);
1718 spin_unlock(&clp->cl_lock);
1721 static struct nfsd4_conn *alloc_conn(struct svc_rqst *rqstp, u32 flags)
1723 struct nfsd4_conn *conn;
1725 conn = kmalloc(sizeof(struct nfsd4_conn), GFP_KERNEL);
1728 svc_xprt_get(rqstp->rq_xprt);
1729 conn->cn_xprt = rqstp->rq_xprt;
1730 conn->cn_flags = flags;
1731 INIT_LIST_HEAD(&conn->cn_xpt_user.list);
1735 static void __nfsd4_hash_conn(struct nfsd4_conn *conn, struct nfsd4_session *ses)
1737 conn->cn_session = ses;
1738 list_add(&conn->cn_persession, &ses->se_conns);
1741 static void nfsd4_hash_conn(struct nfsd4_conn *conn, struct nfsd4_session *ses)
1743 struct nfs4_client *clp = ses->se_client;
1745 spin_lock(&clp->cl_lock);
1746 __nfsd4_hash_conn(conn, ses);
1747 spin_unlock(&clp->cl_lock);
1750 static int nfsd4_register_conn(struct nfsd4_conn *conn)
1752 conn->cn_xpt_user.callback = nfsd4_conn_lost;
1753 return register_xpt_user(conn->cn_xprt, &conn->cn_xpt_user);
1756 static void nfsd4_init_conn(struct svc_rqst *rqstp, struct nfsd4_conn *conn, struct nfsd4_session *ses)
1760 nfsd4_hash_conn(conn, ses);
1761 ret = nfsd4_register_conn(conn);
1763 /* oops; xprt is already down: */
1764 nfsd4_conn_lost(&conn->cn_xpt_user);
1765 /* We may have gained or lost a callback channel: */
1766 nfsd4_probe_callback_sync(ses->se_client);
1769 static struct nfsd4_conn *alloc_conn_from_crses(struct svc_rqst *rqstp, struct nfsd4_create_session *cses)
1771 u32 dir = NFS4_CDFC4_FORE;
1773 if (cses->flags & SESSION4_BACK_CHAN)
1774 dir |= NFS4_CDFC4_BACK;
1775 return alloc_conn(rqstp, dir);
1778 /* must be called under client_lock */
1779 static void nfsd4_del_conns(struct nfsd4_session *s)
1781 struct nfs4_client *clp = s->se_client;
1782 struct nfsd4_conn *c;
1784 spin_lock(&clp->cl_lock);
1785 while (!list_empty(&s->se_conns)) {
1786 c = list_first_entry(&s->se_conns, struct nfsd4_conn, cn_persession);
1787 list_del_init(&c->cn_persession);
1788 spin_unlock(&clp->cl_lock);
1790 unregister_xpt_user(c->cn_xprt, &c->cn_xpt_user);
1793 spin_lock(&clp->cl_lock);
1795 spin_unlock(&clp->cl_lock);
1798 static void __free_session(struct nfsd4_session *ses)
1800 free_session_slots(ses);
1804 static void free_session(struct nfsd4_session *ses)
1806 nfsd4_del_conns(ses);
1807 nfsd4_put_drc_mem(&ses->se_fchannel);
1808 __free_session(ses);
1811 static void init_session(struct svc_rqst *rqstp, struct nfsd4_session *new, struct nfs4_client *clp, struct nfsd4_create_session *cses)
1814 struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
1816 new->se_client = clp;
1819 INIT_LIST_HEAD(&new->se_conns);
1821 new->se_cb_seq_nr = 1;
1822 new->se_flags = cses->flags;
1823 new->se_cb_prog = cses->callback_prog;
1824 new->se_cb_sec = cses->cb_sec;
1825 atomic_set(&new->se_ref, 0);
1826 idx = hash_sessionid(&new->se_sessionid);
1827 list_add(&new->se_hash, &nn->sessionid_hashtbl[idx]);
1828 spin_lock(&clp->cl_lock);
1829 list_add(&new->se_perclnt, &clp->cl_sessions);
1830 spin_unlock(&clp->cl_lock);
1833 struct sockaddr *sa = svc_addr(rqstp);
1835 * This is a little silly; with sessions there's no real
1836 * use for the callback address. Use the peer address
1837 * as a reasonable default for now, but consider fixing
1838 * the rpc client not to require an address in the
1841 rpc_copy_addr((struct sockaddr *)&clp->cl_cb_conn.cb_addr, sa);
1842 clp->cl_cb_conn.cb_addrlen = svc_addr_len(sa);
1846 /* caller must hold client_lock */
1847 static struct nfsd4_session *
1848 __find_in_sessionid_hashtbl(struct nfs4_sessionid *sessionid, struct net *net)
1850 struct nfsd4_session *elem;
1852 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
1854 lockdep_assert_held(&nn->client_lock);
1856 dump_sessionid(__func__, sessionid);
1857 idx = hash_sessionid(sessionid);
1858 /* Search in the appropriate list */
1859 list_for_each_entry(elem, &nn->sessionid_hashtbl[idx], se_hash) {
1860 if (!memcmp(elem->se_sessionid.data, sessionid->data,
1861 NFS4_MAX_SESSIONID_LEN)) {
1866 dprintk("%s: session not found\n", __func__);
1870 static struct nfsd4_session *
1871 find_in_sessionid_hashtbl(struct nfs4_sessionid *sessionid, struct net *net,
1874 struct nfsd4_session *session;
1875 __be32 status = nfserr_badsession;
1877 session = __find_in_sessionid_hashtbl(sessionid, net);
1880 status = nfsd4_get_session_locked(session);
1888 /* caller must hold client_lock */
1890 unhash_session(struct nfsd4_session *ses)
1892 struct nfs4_client *clp = ses->se_client;
1893 struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
1895 lockdep_assert_held(&nn->client_lock);
1897 list_del(&ses->se_hash);
1898 spin_lock(&ses->se_client->cl_lock);
1899 list_del(&ses->se_perclnt);
1900 spin_unlock(&ses->se_client->cl_lock);
1903 /* SETCLIENTID and SETCLIENTID_CONFIRM Helper functions */
1905 STALE_CLIENTID(clientid_t *clid, struct nfsd_net *nn)
1908 * We're assuming the clid was not given out from a boot
1909 * precisely 2^32 (about 136 years) before this one. That seems
1910 * a safe assumption:
1912 if (clid->cl_boot == (u32)nn->boot_time)
1914 dprintk("NFSD stale clientid (%08x/%08x) boot_time %08llx\n",
1915 clid->cl_boot, clid->cl_id, nn->boot_time);
1920 * XXX Should we use a slab cache ?
1921 * This type of memory management is somewhat inefficient, but we use it
1922 * anyway since SETCLIENTID is not a common operation.
1924 static struct nfs4_client *alloc_client(struct xdr_netobj name)
1926 struct nfs4_client *clp;
1929 clp = kmem_cache_zalloc(client_slab, GFP_KERNEL);
1932 xdr_netobj_dup(&clp->cl_name, &name, GFP_KERNEL);
1933 if (clp->cl_name.data == NULL)
1935 clp->cl_ownerstr_hashtbl = kmalloc_array(OWNER_HASH_SIZE,
1936 sizeof(struct list_head),
1938 if (!clp->cl_ownerstr_hashtbl)
1939 goto err_no_hashtbl;
1940 for (i = 0; i < OWNER_HASH_SIZE; i++)
1941 INIT_LIST_HEAD(&clp->cl_ownerstr_hashtbl[i]);
1942 INIT_LIST_HEAD(&clp->cl_sessions);
1943 idr_init(&clp->cl_stateids);
1944 atomic_set(&clp->cl_rpc_users, 0);
1945 clp->cl_cb_state = NFSD4_CB_UNKNOWN;
1946 INIT_LIST_HEAD(&clp->cl_idhash);
1947 INIT_LIST_HEAD(&clp->cl_openowners);
1948 INIT_LIST_HEAD(&clp->cl_delegations);
1949 INIT_LIST_HEAD(&clp->cl_lru);
1950 INIT_LIST_HEAD(&clp->cl_revoked);
1951 #ifdef CONFIG_NFSD_PNFS
1952 INIT_LIST_HEAD(&clp->cl_lo_states);
1954 INIT_LIST_HEAD(&clp->async_copies);
1955 spin_lock_init(&clp->async_lock);
1956 spin_lock_init(&clp->cl_lock);
1957 rpc_init_wait_queue(&clp->cl_cb_waitq, "Backchannel slot table");
1960 kfree(clp->cl_name.data);
1962 kmem_cache_free(client_slab, clp);
1966 static void __free_client(struct kref *k)
1968 struct nfsdfs_client *c = container_of(k, struct nfsdfs_client, cl_ref);
1969 struct nfs4_client *clp = container_of(c, struct nfs4_client, cl_nfsdfs);
1971 free_svc_cred(&clp->cl_cred);
1972 kfree(clp->cl_ownerstr_hashtbl);
1973 kfree(clp->cl_name.data);
1974 kfree(clp->cl_nii_domain.data);
1975 kfree(clp->cl_nii_name.data);
1976 idr_destroy(&clp->cl_stateids);
1977 kmem_cache_free(client_slab, clp);
1980 static void drop_client(struct nfs4_client *clp)
1982 kref_put(&clp->cl_nfsdfs.cl_ref, __free_client);
1986 free_client(struct nfs4_client *clp)
1988 while (!list_empty(&clp->cl_sessions)) {
1989 struct nfsd4_session *ses;
1990 ses = list_entry(clp->cl_sessions.next, struct nfsd4_session,
1992 list_del(&ses->se_perclnt);
1993 WARN_ON_ONCE(atomic_read(&ses->se_ref));
1996 rpc_destroy_wait_queue(&clp->cl_cb_waitq);
1997 if (clp->cl_nfsd_dentry) {
1998 nfsd_client_rmdir(clp->cl_nfsd_dentry);
1999 clp->cl_nfsd_dentry = NULL;
2000 wake_up_all(&expiry_wq);
2005 /* must be called under the client_lock */
2007 unhash_client_locked(struct nfs4_client *clp)
2009 struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
2010 struct nfsd4_session *ses;
2012 lockdep_assert_held(&nn->client_lock);
2014 /* Mark the client as expired! */
2016 /* Make it invisible */
2017 if (!list_empty(&clp->cl_idhash)) {
2018 list_del_init(&clp->cl_idhash);
2019 if (test_bit(NFSD4_CLIENT_CONFIRMED, &clp->cl_flags))
2020 rb_erase(&clp->cl_namenode, &nn->conf_name_tree);
2022 rb_erase(&clp->cl_namenode, &nn->unconf_name_tree);
2024 list_del_init(&clp->cl_lru);
2025 spin_lock(&clp->cl_lock);
2026 list_for_each_entry(ses, &clp->cl_sessions, se_perclnt)
2027 list_del_init(&ses->se_hash);
2028 spin_unlock(&clp->cl_lock);
2032 unhash_client(struct nfs4_client *clp)
2034 struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
2036 spin_lock(&nn->client_lock);
2037 unhash_client_locked(clp);
2038 spin_unlock(&nn->client_lock);
2041 static __be32 mark_client_expired_locked(struct nfs4_client *clp)
2043 if (atomic_read(&clp->cl_rpc_users))
2044 return nfserr_jukebox;
2045 unhash_client_locked(clp);
2050 __destroy_client(struct nfs4_client *clp)
2053 struct nfs4_openowner *oo;
2054 struct nfs4_delegation *dp;
2055 struct list_head reaplist;
2057 INIT_LIST_HEAD(&reaplist);
2058 spin_lock(&state_lock);
2059 while (!list_empty(&clp->cl_delegations)) {
2060 dp = list_entry(clp->cl_delegations.next, struct nfs4_delegation, dl_perclnt);
2061 WARN_ON(!unhash_delegation_locked(dp));
2062 list_add(&dp->dl_recall_lru, &reaplist);
2064 spin_unlock(&state_lock);
2065 while (!list_empty(&reaplist)) {
2066 dp = list_entry(reaplist.next, struct nfs4_delegation, dl_recall_lru);
2067 list_del_init(&dp->dl_recall_lru);
2068 destroy_unhashed_deleg(dp);
2070 while (!list_empty(&clp->cl_revoked)) {
2071 dp = list_entry(clp->cl_revoked.next, struct nfs4_delegation, dl_recall_lru);
2072 list_del_init(&dp->dl_recall_lru);
2073 nfs4_put_stid(&dp->dl_stid);
2075 while (!list_empty(&clp->cl_openowners)) {
2076 oo = list_entry(clp->cl_openowners.next, struct nfs4_openowner, oo_perclient);
2077 nfs4_get_stateowner(&oo->oo_owner);
2078 release_openowner(oo);
2080 for (i = 0; i < OWNER_HASH_SIZE; i++) {
2081 struct nfs4_stateowner *so, *tmp;
2083 list_for_each_entry_safe(so, tmp, &clp->cl_ownerstr_hashtbl[i],
2085 /* Should be no openowners at this point */
2086 WARN_ON_ONCE(so->so_is_open_owner);
2087 remove_blocked_locks(lockowner(so));
2090 nfsd4_return_all_client_layouts(clp);
2091 nfsd4_shutdown_copy(clp);
2092 nfsd4_shutdown_callback(clp);
2093 if (clp->cl_cb_conn.cb_xprt)
2094 svc_xprt_put(clp->cl_cb_conn.cb_xprt);
2096 wake_up_all(&expiry_wq);
2100 destroy_client(struct nfs4_client *clp)
2103 __destroy_client(clp);
2106 static void inc_reclaim_complete(struct nfs4_client *clp)
2108 struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
2110 if (!nn->track_reclaim_completes)
2112 if (!nfsd4_find_reclaim_client(clp->cl_name, nn))
2114 if (atomic_inc_return(&nn->nr_reclaim_complete) ==
2115 nn->reclaim_str_hashtbl_size) {
2116 printk(KERN_INFO "NFSD: all clients done reclaiming, ending NFSv4 grace period (net %x)\n",
2118 nfsd4_end_grace(nn);
2122 static void expire_client(struct nfs4_client *clp)
2125 nfsd4_client_record_remove(clp);
2126 __destroy_client(clp);
2129 static void copy_verf(struct nfs4_client *target, nfs4_verifier *source)
2131 memcpy(target->cl_verifier.data, source->data,
2132 sizeof(target->cl_verifier.data));
2135 static void copy_clid(struct nfs4_client *target, struct nfs4_client *source)
2137 target->cl_clientid.cl_boot = source->cl_clientid.cl_boot;
2138 target->cl_clientid.cl_id = source->cl_clientid.cl_id;
2141 static int copy_cred(struct svc_cred *target, struct svc_cred *source)
2143 target->cr_principal = kstrdup(source->cr_principal, GFP_KERNEL);
2144 target->cr_raw_principal = kstrdup(source->cr_raw_principal,
2146 target->cr_targ_princ = kstrdup(source->cr_targ_princ, GFP_KERNEL);
2147 if ((source->cr_principal && !target->cr_principal) ||
2148 (source->cr_raw_principal && !target->cr_raw_principal) ||
2149 (source->cr_targ_princ && !target->cr_targ_princ))
2152 target->cr_flavor = source->cr_flavor;
2153 target->cr_uid = source->cr_uid;
2154 target->cr_gid = source->cr_gid;
2155 target->cr_group_info = source->cr_group_info;
2156 get_group_info(target->cr_group_info);
2157 target->cr_gss_mech = source->cr_gss_mech;
2158 if (source->cr_gss_mech)
2159 gss_mech_get(source->cr_gss_mech);
2164 compare_blob(const struct xdr_netobj *o1, const struct xdr_netobj *o2)
2166 if (o1->len < o2->len)
2168 if (o1->len > o2->len)
2170 return memcmp(o1->data, o2->data, o1->len);
2174 same_verf(nfs4_verifier *v1, nfs4_verifier *v2)
2176 return 0 == memcmp(v1->data, v2->data, sizeof(v1->data));
2180 same_clid(clientid_t *cl1, clientid_t *cl2)
2182 return (cl1->cl_boot == cl2->cl_boot) && (cl1->cl_id == cl2->cl_id);
2185 static bool groups_equal(struct group_info *g1, struct group_info *g2)
2189 if (g1->ngroups != g2->ngroups)
2191 for (i=0; i<g1->ngroups; i++)
2192 if (!gid_eq(g1->gid[i], g2->gid[i]))
2198 * RFC 3530 language requires clid_inuse be returned when the
2199 * "principal" associated with a requests differs from that previously
2200 * used. We use uid, gid's, and gss principal string as our best
2201 * approximation. We also don't want to allow non-gss use of a client
2202 * established using gss: in theory cr_principal should catch that
2203 * change, but in practice cr_principal can be null even in the gss case
2204 * since gssd doesn't always pass down a principal string.
2206 static bool is_gss_cred(struct svc_cred *cr)
2208 /* Is cr_flavor one of the gss "pseudoflavors"?: */
2209 return (cr->cr_flavor > RPC_AUTH_MAXFLAVOR);
2214 same_creds(struct svc_cred *cr1, struct svc_cred *cr2)
2216 if ((is_gss_cred(cr1) != is_gss_cred(cr2))
2217 || (!uid_eq(cr1->cr_uid, cr2->cr_uid))
2218 || (!gid_eq(cr1->cr_gid, cr2->cr_gid))
2219 || !groups_equal(cr1->cr_group_info, cr2->cr_group_info))
2221 /* XXX: check that cr_targ_princ fields match ? */
2222 if (cr1->cr_principal == cr2->cr_principal)
2224 if (!cr1->cr_principal || !cr2->cr_principal)
2226 return 0 == strcmp(cr1->cr_principal, cr2->cr_principal);
2229 static bool svc_rqst_integrity_protected(struct svc_rqst *rqstp)
2231 struct svc_cred *cr = &rqstp->rq_cred;
2234 if (!cr->cr_gss_mech)
2236 service = gss_pseudoflavor_to_service(cr->cr_gss_mech, cr->cr_flavor);
2237 return service == RPC_GSS_SVC_INTEGRITY ||
2238 service == RPC_GSS_SVC_PRIVACY;
2241 bool nfsd4_mach_creds_match(struct nfs4_client *cl, struct svc_rqst *rqstp)
2243 struct svc_cred *cr = &rqstp->rq_cred;
2245 if (!cl->cl_mach_cred)
2247 if (cl->cl_cred.cr_gss_mech != cr->cr_gss_mech)
2249 if (!svc_rqst_integrity_protected(rqstp))
2251 if (cl->cl_cred.cr_raw_principal)
2252 return 0 == strcmp(cl->cl_cred.cr_raw_principal,
2253 cr->cr_raw_principal);
2254 if (!cr->cr_principal)
2256 return 0 == strcmp(cl->cl_cred.cr_principal, cr->cr_principal);
2259 static void gen_confirm(struct nfs4_client *clp, struct nfsd_net *nn)
2264 * This is opaque to client, so no need to byte-swap. Use
2265 * __force to keep sparse happy
2267 verf[0] = (__force __be32)(u32)ktime_get_real_seconds();
2268 verf[1] = (__force __be32)nn->clverifier_counter++;
2269 memcpy(clp->cl_confirm.data, verf, sizeof(clp->cl_confirm.data));
2272 static void gen_clid(struct nfs4_client *clp, struct nfsd_net *nn)
2274 clp->cl_clientid.cl_boot = (u32)nn->boot_time;
2275 clp->cl_clientid.cl_id = nn->clientid_counter++;
2276 gen_confirm(clp, nn);
2279 static struct nfs4_stid *
2280 find_stateid_locked(struct nfs4_client *cl, stateid_t *t)
2282 struct nfs4_stid *ret;
2284 ret = idr_find(&cl->cl_stateids, t->si_opaque.so_id);
2285 if (!ret || !ret->sc_type)
2290 static struct nfs4_stid *
2291 find_stateid_by_type(struct nfs4_client *cl, stateid_t *t, char typemask)
2293 struct nfs4_stid *s;
2295 spin_lock(&cl->cl_lock);
2296 s = find_stateid_locked(cl, t);
2298 if (typemask & s->sc_type)
2299 refcount_inc(&s->sc_count);
2303 spin_unlock(&cl->cl_lock);
2307 static struct nfs4_client *get_nfsdfs_clp(struct inode *inode)
2309 struct nfsdfs_client *nc;
2310 nc = get_nfsdfs_client(inode);
2313 return container_of(nc, struct nfs4_client, cl_nfsdfs);
2316 static void seq_quote_mem(struct seq_file *m, char *data, int len)
2318 seq_printf(m, "\"");
2319 seq_escape_mem_ascii(m, data, len);
2320 seq_printf(m, "\"");
2323 static int client_info_show(struct seq_file *m, void *v)
2325 struct inode *inode = m->private;
2326 struct nfs4_client *clp;
2329 clp = get_nfsdfs_clp(inode);
2332 memcpy(&clid, &clp->cl_clientid, sizeof(clid));
2333 seq_printf(m, "clientid: 0x%llx\n", clid);
2334 seq_printf(m, "address: \"%pISpc\"\n", (struct sockaddr *)&clp->cl_addr);
2335 seq_printf(m, "name: ");
2336 seq_quote_mem(m, clp->cl_name.data, clp->cl_name.len);
2337 seq_printf(m, "\nminor version: %d\n", clp->cl_minorversion);
2338 if (clp->cl_nii_domain.data) {
2339 seq_printf(m, "Implementation domain: ");
2340 seq_quote_mem(m, clp->cl_nii_domain.data,
2341 clp->cl_nii_domain.len);
2342 seq_printf(m, "\nImplementation name: ");
2343 seq_quote_mem(m, clp->cl_nii_name.data, clp->cl_nii_name.len);
2344 seq_printf(m, "\nImplementation time: [%lld, %ld]\n",
2345 clp->cl_nii_time.tv_sec, clp->cl_nii_time.tv_nsec);
2352 static int client_info_open(struct inode *inode, struct file *file)
2354 return single_open(file, client_info_show, inode);
2357 static const struct file_operations client_info_fops = {
2358 .open = client_info_open,
2360 .llseek = seq_lseek,
2361 .release = single_release,
2364 static void *states_start(struct seq_file *s, loff_t *pos)
2365 __acquires(&clp->cl_lock)
2367 struct nfs4_client *clp = s->private;
2368 unsigned long id = *pos;
2371 spin_lock(&clp->cl_lock);
2372 ret = idr_get_next_ul(&clp->cl_stateids, &id);
2377 static void *states_next(struct seq_file *s, void *v, loff_t *pos)
2379 struct nfs4_client *clp = s->private;
2380 unsigned long id = *pos;
2385 ret = idr_get_next_ul(&clp->cl_stateids, &id);
2390 static void states_stop(struct seq_file *s, void *v)
2391 __releases(&clp->cl_lock)
2393 struct nfs4_client *clp = s->private;
2395 spin_unlock(&clp->cl_lock);
2398 static void nfs4_show_superblock(struct seq_file *s, struct nfsd_file *f)
2400 struct inode *inode = f->nf_inode;
2402 seq_printf(s, "superblock: \"%02x:%02x:%ld\"",
2403 MAJOR(inode->i_sb->s_dev),
2404 MINOR(inode->i_sb->s_dev),
2408 static void nfs4_show_owner(struct seq_file *s, struct nfs4_stateowner *oo)
2410 seq_printf(s, "owner: ");
2411 seq_quote_mem(s, oo->so_owner.data, oo->so_owner.len);
2414 static int nfs4_show_open(struct seq_file *s, struct nfs4_stid *st)
2416 struct nfs4_ol_stateid *ols;
2417 struct nfs4_file *nf;
2418 struct nfsd_file *file;
2419 struct nfs4_stateowner *oo;
2420 unsigned int access, deny;
2422 if (st->sc_type != NFS4_OPEN_STID && st->sc_type != NFS4_LOCK_STID)
2423 return 0; /* XXX: or SEQ_SKIP? */
2424 ols = openlockstateid(st);
2425 oo = ols->st_stateowner;
2427 file = find_any_file(nf);
2429 seq_printf(s, "- 0x%16phN: { type: open, ", &st->sc_stateid);
2431 access = bmap_to_share_mode(ols->st_access_bmap);
2432 deny = bmap_to_share_mode(ols->st_deny_bmap);
2434 seq_printf(s, "access: %s%s, ",
2435 access & NFS4_SHARE_ACCESS_READ ? "r" : "-",
2436 access & NFS4_SHARE_ACCESS_WRITE ? "w" : "-");
2437 seq_printf(s, "deny: %s%s, ",
2438 deny & NFS4_SHARE_ACCESS_READ ? "r" : "-",
2439 deny & NFS4_SHARE_ACCESS_WRITE ? "w" : "-");
2441 nfs4_show_superblock(s, file);
2442 seq_printf(s, ", ");
2443 nfs4_show_owner(s, oo);
2444 seq_printf(s, " }\n");
2445 nfsd_file_put(file);
2450 static int nfs4_show_lock(struct seq_file *s, struct nfs4_stid *st)
2452 struct nfs4_ol_stateid *ols;
2453 struct nfs4_file *nf;
2454 struct nfsd_file *file;
2455 struct nfs4_stateowner *oo;
2457 ols = openlockstateid(st);
2458 oo = ols->st_stateowner;
2460 file = find_any_file(nf);
2462 seq_printf(s, "- 0x%16phN: { type: lock, ", &st->sc_stateid);
2465 * Note: a lock stateid isn't really the same thing as a lock,
2466 * it's the locking state held by one owner on a file, and there
2467 * may be multiple (or no) lock ranges associated with it.
2468 * (Same for the matter is true of open stateids.)
2471 nfs4_show_superblock(s, file);
2472 /* XXX: open stateid? */
2473 seq_printf(s, ", ");
2474 nfs4_show_owner(s, oo);
2475 seq_printf(s, " }\n");
2476 nfsd_file_put(file);
2481 static int nfs4_show_deleg(struct seq_file *s, struct nfs4_stid *st)
2483 struct nfs4_delegation *ds;
2484 struct nfs4_file *nf;
2485 struct nfsd_file *file;
2487 ds = delegstateid(st);
2489 file = nf->fi_deleg_file;
2491 seq_printf(s, "- 0x%16phN: { type: deleg, ", &st->sc_stateid);
2493 /* Kinda dead code as long as we only support read delegs: */
2494 seq_printf(s, "access: %s, ",
2495 ds->dl_type == NFS4_OPEN_DELEGATE_READ ? "r" : "w");
2497 /* XXX: lease time, whether it's being recalled. */
2499 nfs4_show_superblock(s, file);
2500 seq_printf(s, " }\n");
2505 static int nfs4_show_layout(struct seq_file *s, struct nfs4_stid *st)
2507 struct nfs4_layout_stateid *ls;
2508 struct nfsd_file *file;
2510 ls = container_of(st, struct nfs4_layout_stateid, ls_stid);
2513 seq_printf(s, "- 0x%16phN: { type: layout, ", &st->sc_stateid);
2515 /* XXX: What else would be useful? */
2517 nfs4_show_superblock(s, file);
2518 seq_printf(s, " }\n");
2523 static int states_show(struct seq_file *s, void *v)
2525 struct nfs4_stid *st = v;
2527 switch (st->sc_type) {
2528 case NFS4_OPEN_STID:
2529 return nfs4_show_open(s, st);
2530 case NFS4_LOCK_STID:
2531 return nfs4_show_lock(s, st);
2532 case NFS4_DELEG_STID:
2533 return nfs4_show_deleg(s, st);
2534 case NFS4_LAYOUT_STID:
2535 return nfs4_show_layout(s, st);
2537 return 0; /* XXX: or SEQ_SKIP? */
2539 /* XXX: copy stateids? */
2542 static struct seq_operations states_seq_ops = {
2543 .start = states_start,
2544 .next = states_next,
2545 .stop = states_stop,
2549 static int client_states_open(struct inode *inode, struct file *file)
2552 struct nfs4_client *clp;
2555 clp = get_nfsdfs_clp(inode);
2559 ret = seq_open(file, &states_seq_ops);
2562 s = file->private_data;
2567 static int client_opens_release(struct inode *inode, struct file *file)
2569 struct seq_file *m = file->private_data;
2570 struct nfs4_client *clp = m->private;
2572 /* XXX: alternatively, we could get/drop in seq start/stop */
2577 static const struct file_operations client_states_fops = {
2578 .open = client_states_open,
2580 .llseek = seq_lseek,
2581 .release = client_opens_release,
2585 * Normally we refuse to destroy clients that are in use, but here the
2586 * administrator is telling us to just do it. We also want to wait
2587 * so the caller has a guarantee that the client's locks are gone by
2588 * the time the write returns:
2590 static void force_expire_client(struct nfs4_client *clp)
2592 struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
2593 bool already_expired;
2595 spin_lock(&clp->cl_lock);
2597 spin_unlock(&clp->cl_lock);
2599 wait_event(expiry_wq, atomic_read(&clp->cl_rpc_users) == 0);
2600 spin_lock(&nn->client_lock);
2601 already_expired = list_empty(&clp->cl_lru);
2602 if (!already_expired)
2603 unhash_client_locked(clp);
2604 spin_unlock(&nn->client_lock);
2606 if (!already_expired)
2609 wait_event(expiry_wq, clp->cl_nfsd_dentry == NULL);
2612 static ssize_t client_ctl_write(struct file *file, const char __user *buf,
2613 size_t size, loff_t *pos)
2616 struct nfs4_client *clp;
2618 data = simple_transaction_get(file, buf, size);
2620 return PTR_ERR(data);
2621 if (size != 7 || 0 != memcmp(data, "expire\n", 7))
2623 clp = get_nfsdfs_clp(file_inode(file));
2626 force_expire_client(clp);
2631 static const struct file_operations client_ctl_fops = {
2632 .write = client_ctl_write,
2633 .release = simple_transaction_release,
2636 static const struct tree_descr client_files[] = {
2637 [0] = {"info", &client_info_fops, S_IRUSR},
2638 [1] = {"states", &client_states_fops, S_IRUSR},
2639 [2] = {"ctl", &client_ctl_fops, S_IRUSR|S_IWUSR},
2643 static struct nfs4_client *create_client(struct xdr_netobj name,
2644 struct svc_rqst *rqstp, nfs4_verifier *verf)
2646 struct nfs4_client *clp;
2647 struct sockaddr *sa = svc_addr(rqstp);
2649 struct net *net = SVC_NET(rqstp);
2650 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
2652 clp = alloc_client(name);
2656 ret = copy_cred(&clp->cl_cred, &rqstp->rq_cred);
2662 kref_init(&clp->cl_nfsdfs.cl_ref);
2663 nfsd4_init_cb(&clp->cl_cb_null, clp, NULL, NFSPROC4_CLNT_CB_NULL);
2664 clp->cl_time = ktime_get_boottime_seconds();
2665 clear_bit(0, &clp->cl_cb_slot_busy);
2666 copy_verf(clp, verf);
2667 memcpy(&clp->cl_addr, sa, sizeof(struct sockaddr_storage));
2668 clp->cl_cb_session = NULL;
2670 clp->cl_nfsd_dentry = nfsd_client_mkdir(nn, &clp->cl_nfsdfs,
2671 clp->cl_clientid.cl_id - nn->clientid_base,
2673 if (!clp->cl_nfsd_dentry) {
2681 add_clp_to_name_tree(struct nfs4_client *new_clp, struct rb_root *root)
2683 struct rb_node **new = &(root->rb_node), *parent = NULL;
2684 struct nfs4_client *clp;
2687 clp = rb_entry(*new, struct nfs4_client, cl_namenode);
2690 if (compare_blob(&clp->cl_name, &new_clp->cl_name) > 0)
2691 new = &((*new)->rb_left);
2693 new = &((*new)->rb_right);
2696 rb_link_node(&new_clp->cl_namenode, parent, new);
2697 rb_insert_color(&new_clp->cl_namenode, root);
2700 static struct nfs4_client *
2701 find_clp_in_name_tree(struct xdr_netobj *name, struct rb_root *root)
2704 struct rb_node *node = root->rb_node;
2705 struct nfs4_client *clp;
2708 clp = rb_entry(node, struct nfs4_client, cl_namenode);
2709 cmp = compare_blob(&clp->cl_name, name);
2711 node = node->rb_left;
2713 node = node->rb_right;
2721 add_to_unconfirmed(struct nfs4_client *clp)
2723 unsigned int idhashval;
2724 struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
2726 lockdep_assert_held(&nn->client_lock);
2728 clear_bit(NFSD4_CLIENT_CONFIRMED, &clp->cl_flags);
2729 add_clp_to_name_tree(clp, &nn->unconf_name_tree);
2730 idhashval = clientid_hashval(clp->cl_clientid.cl_id);
2731 list_add(&clp->cl_idhash, &nn->unconf_id_hashtbl[idhashval]);
2732 renew_client_locked(clp);
2736 move_to_confirmed(struct nfs4_client *clp)
2738 unsigned int idhashval = clientid_hashval(clp->cl_clientid.cl_id);
2739 struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
2741 lockdep_assert_held(&nn->client_lock);
2743 dprintk("NFSD: move_to_confirm nfs4_client %p\n", clp);
2744 list_move(&clp->cl_idhash, &nn->conf_id_hashtbl[idhashval]);
2745 rb_erase(&clp->cl_namenode, &nn->unconf_name_tree);
2746 add_clp_to_name_tree(clp, &nn->conf_name_tree);
2747 set_bit(NFSD4_CLIENT_CONFIRMED, &clp->cl_flags);
2748 renew_client_locked(clp);
2751 static struct nfs4_client *
2752 find_client_in_id_table(struct list_head *tbl, clientid_t *clid, bool sessions)
2754 struct nfs4_client *clp;
2755 unsigned int idhashval = clientid_hashval(clid->cl_id);
2757 list_for_each_entry(clp, &tbl[idhashval], cl_idhash) {
2758 if (same_clid(&clp->cl_clientid, clid)) {
2759 if ((bool)clp->cl_minorversion != sessions)
2761 renew_client_locked(clp);
2768 static struct nfs4_client *
2769 find_confirmed_client(clientid_t *clid, bool sessions, struct nfsd_net *nn)
2771 struct list_head *tbl = nn->conf_id_hashtbl;
2773 lockdep_assert_held(&nn->client_lock);
2774 return find_client_in_id_table(tbl, clid, sessions);
2777 static struct nfs4_client *
2778 find_unconfirmed_client(clientid_t *clid, bool sessions, struct nfsd_net *nn)
2780 struct list_head *tbl = nn->unconf_id_hashtbl;
2782 lockdep_assert_held(&nn->client_lock);
2783 return find_client_in_id_table(tbl, clid, sessions);
2786 static bool clp_used_exchangeid(struct nfs4_client *clp)
2788 return clp->cl_exchange_flags != 0;
2791 static struct nfs4_client *
2792 find_confirmed_client_by_name(struct xdr_netobj *name, struct nfsd_net *nn)
2794 lockdep_assert_held(&nn->client_lock);
2795 return find_clp_in_name_tree(name, &nn->conf_name_tree);
2798 static struct nfs4_client *
2799 find_unconfirmed_client_by_name(struct xdr_netobj *name, struct nfsd_net *nn)
2801 lockdep_assert_held(&nn->client_lock);
2802 return find_clp_in_name_tree(name, &nn->unconf_name_tree);
2806 gen_callback(struct nfs4_client *clp, struct nfsd4_setclientid *se, struct svc_rqst *rqstp)
2808 struct nfs4_cb_conn *conn = &clp->cl_cb_conn;
2809 struct sockaddr *sa = svc_addr(rqstp);
2810 u32 scopeid = rpc_get_scope_id(sa);
2811 unsigned short expected_family;
2813 /* Currently, we only support tcp and tcp6 for the callback channel */
2814 if (se->se_callback_netid_len == 3 &&
2815 !memcmp(se->se_callback_netid_val, "tcp", 3))
2816 expected_family = AF_INET;
2817 else if (se->se_callback_netid_len == 4 &&
2818 !memcmp(se->se_callback_netid_val, "tcp6", 4))
2819 expected_family = AF_INET6;
2823 conn->cb_addrlen = rpc_uaddr2sockaddr(clp->net, se->se_callback_addr_val,
2824 se->se_callback_addr_len,
2825 (struct sockaddr *)&conn->cb_addr,
2826 sizeof(conn->cb_addr));
2828 if (!conn->cb_addrlen || conn->cb_addr.ss_family != expected_family)
2831 if (conn->cb_addr.ss_family == AF_INET6)
2832 ((struct sockaddr_in6 *)&conn->cb_addr)->sin6_scope_id = scopeid;
2834 conn->cb_prog = se->se_callback_prog;
2835 conn->cb_ident = se->se_callback_ident;
2836 memcpy(&conn->cb_saddr, &rqstp->rq_daddr, rqstp->rq_daddrlen);
2839 conn->cb_addr.ss_family = AF_UNSPEC;
2840 conn->cb_addrlen = 0;
2841 dprintk("NFSD: this client (clientid %08x/%08x) "
2842 "will not receive delegations\n",
2843 clp->cl_clientid.cl_boot, clp->cl_clientid.cl_id);
2849 * Cache a reply. nfsd4_check_resp_size() has bounded the cache size.
2852 nfsd4_store_cache_entry(struct nfsd4_compoundres *resp)
2854 struct xdr_buf *buf = resp->xdr.buf;
2855 struct nfsd4_slot *slot = resp->cstate.slot;
2858 dprintk("--> %s slot %p\n", __func__, slot);
2860 slot->sl_flags |= NFSD4_SLOT_INITIALIZED;
2861 slot->sl_opcnt = resp->opcnt;
2862 slot->sl_status = resp->cstate.status;
2863 free_svc_cred(&slot->sl_cred);
2864 copy_cred(&slot->sl_cred, &resp->rqstp->rq_cred);
2866 if (!nfsd4_cache_this(resp)) {
2867 slot->sl_flags &= ~NFSD4_SLOT_CACHED;
2870 slot->sl_flags |= NFSD4_SLOT_CACHED;
2872 base = resp->cstate.data_offset;
2873 slot->sl_datalen = buf->len - base;
2874 if (read_bytes_from_xdr_buf(buf, base, slot->sl_data, slot->sl_datalen))
2875 WARN(1, "%s: sessions DRC could not cache compound\n",
2881 * Encode the replay sequence operation from the slot values.
2882 * If cachethis is FALSE encode the uncached rep error on the next
2883 * operation which sets resp->p and increments resp->opcnt for
2884 * nfs4svc_encode_compoundres.
2888 nfsd4_enc_sequence_replay(struct nfsd4_compoundargs *args,
2889 struct nfsd4_compoundres *resp)
2891 struct nfsd4_op *op;
2892 struct nfsd4_slot *slot = resp->cstate.slot;
2894 /* Encode the replayed sequence operation */
2895 op = &args->ops[resp->opcnt - 1];
2896 nfsd4_encode_operation(resp, op);
2898 if (slot->sl_flags & NFSD4_SLOT_CACHED)
2900 if (args->opcnt == 1) {
2902 * The original operation wasn't a solo sequence--we
2903 * always cache those--so this retry must not match the
2906 op->status = nfserr_seq_false_retry;
2908 op = &args->ops[resp->opcnt++];
2909 op->status = nfserr_retry_uncached_rep;
2910 nfsd4_encode_operation(resp, op);
2916 * The sequence operation is not cached because we can use the slot and
2920 nfsd4_replay_cache_entry(struct nfsd4_compoundres *resp,
2921 struct nfsd4_sequence *seq)
2923 struct nfsd4_slot *slot = resp->cstate.slot;
2924 struct xdr_stream *xdr = &resp->xdr;
2928 dprintk("--> %s slot %p\n", __func__, slot);
2930 status = nfsd4_enc_sequence_replay(resp->rqstp->rq_argp, resp);
2934 p = xdr_reserve_space(xdr, slot->sl_datalen);
2937 return nfserr_serverfault;
2939 xdr_encode_opaque_fixed(p, slot->sl_data, slot->sl_datalen);
2940 xdr_commit_encode(xdr);
2942 resp->opcnt = slot->sl_opcnt;
2943 return slot->sl_status;
2947 * Set the exchange_id flags returned by the server.
2950 nfsd4_set_ex_flags(struct nfs4_client *new, struct nfsd4_exchange_id *clid)
2952 #ifdef CONFIG_NFSD_PNFS
2953 new->cl_exchange_flags |= EXCHGID4_FLAG_USE_PNFS_MDS;
2955 new->cl_exchange_flags |= EXCHGID4_FLAG_USE_NON_PNFS;
2958 /* Referrals are supported, Migration is not. */
2959 new->cl_exchange_flags |= EXCHGID4_FLAG_SUPP_MOVED_REFER;
2961 /* set the wire flags to return to client. */
2962 clid->flags = new->cl_exchange_flags;
2965 static bool client_has_openowners(struct nfs4_client *clp)
2967 struct nfs4_openowner *oo;
2969 list_for_each_entry(oo, &clp->cl_openowners, oo_perclient) {
2970 if (!list_empty(&oo->oo_owner.so_stateids))
2976 static bool client_has_state(struct nfs4_client *clp)
2978 return client_has_openowners(clp)
2979 #ifdef CONFIG_NFSD_PNFS
2980 || !list_empty(&clp->cl_lo_states)
2982 || !list_empty(&clp->cl_delegations)
2983 || !list_empty(&clp->cl_sessions)
2984 || !list_empty(&clp->async_copies);
2987 static __be32 copy_impl_id(struct nfs4_client *clp,
2988 struct nfsd4_exchange_id *exid)
2990 if (!exid->nii_domain.data)
2992 xdr_netobj_dup(&clp->cl_nii_domain, &exid->nii_domain, GFP_KERNEL);
2993 if (!clp->cl_nii_domain.data)
2994 return nfserr_jukebox;
2995 xdr_netobj_dup(&clp->cl_nii_name, &exid->nii_name, GFP_KERNEL);
2996 if (!clp->cl_nii_name.data)
2997 return nfserr_jukebox;
2998 clp->cl_nii_time = exid->nii_time;
3003 nfsd4_exchange_id(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3004 union nfsd4_op_u *u)
3006 struct nfsd4_exchange_id *exid = &u->exchange_id;
3007 struct nfs4_client *conf, *new;
3008 struct nfs4_client *unconf = NULL;
3010 char addr_str[INET6_ADDRSTRLEN];
3011 nfs4_verifier verf = exid->verifier;
3012 struct sockaddr *sa = svc_addr(rqstp);
3013 bool update = exid->flags & EXCHGID4_FLAG_UPD_CONFIRMED_REC_A;
3014 struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
3016 rpc_ntop(sa, addr_str, sizeof(addr_str));
3017 dprintk("%s rqstp=%p exid=%p clname.len=%u clname.data=%p "
3018 "ip_addr=%s flags %x, spa_how %d\n",
3019 __func__, rqstp, exid, exid->clname.len, exid->clname.data,
3020 addr_str, exid->flags, exid->spa_how);
3022 if (exid->flags & ~EXCHGID4_FLAG_MASK_A)
3023 return nfserr_inval;
3025 new = create_client(exid->clname, rqstp, &verf);
3027 return nfserr_jukebox;
3028 status = copy_impl_id(new, exid);
3032 switch (exid->spa_how) {
3034 exid->spo_must_enforce[0] = 0;
3035 exid->spo_must_enforce[1] = (
3036 1 << (OP_BIND_CONN_TO_SESSION - 32) |
3037 1 << (OP_EXCHANGE_ID - 32) |
3038 1 << (OP_CREATE_SESSION - 32) |
3039 1 << (OP_DESTROY_SESSION - 32) |
3040 1 << (OP_DESTROY_CLIENTID - 32));
3042 exid->spo_must_allow[0] &= (1 << (OP_CLOSE) |
3043 1 << (OP_OPEN_DOWNGRADE) |
3045 1 << (OP_DELEGRETURN));
3047 exid->spo_must_allow[1] &= (
3048 1 << (OP_TEST_STATEID - 32) |
3049 1 << (OP_FREE_STATEID - 32));
3050 if (!svc_rqst_integrity_protected(rqstp)) {
3051 status = nfserr_inval;
3055 * Sometimes userspace doesn't give us a principal.
3056 * Which is a bug, really. Anyway, we can't enforce
3057 * MACH_CRED in that case, better to give up now:
3059 if (!new->cl_cred.cr_principal &&
3060 !new->cl_cred.cr_raw_principal) {
3061 status = nfserr_serverfault;
3064 new->cl_mach_cred = true;
3067 default: /* checked by xdr code */
3071 status = nfserr_encr_alg_unsupp;
3075 /* Cases below refer to rfc 5661 section 18.35.4: */
3076 spin_lock(&nn->client_lock);
3077 conf = find_confirmed_client_by_name(&exid->clname, nn);
3079 bool creds_match = same_creds(&conf->cl_cred, &rqstp->rq_cred);
3080 bool verfs_match = same_verf(&verf, &conf->cl_verifier);
3083 if (!clp_used_exchangeid(conf)) { /* buggy client */
3084 status = nfserr_inval;
3087 if (!nfsd4_mach_creds_match(conf, rqstp)) {
3088 status = nfserr_wrong_cred;
3091 if (!creds_match) { /* case 9 */
3092 status = nfserr_perm;
3095 if (!verfs_match) { /* case 8 */
3096 status = nfserr_not_same;
3100 exid->flags |= EXCHGID4_FLAG_CONFIRMED_R;
3103 if (!creds_match) { /* case 3 */
3104 if (client_has_state(conf)) {
3105 status = nfserr_clid_inuse;
3110 if (verfs_match) { /* case 2 */
3111 conf->cl_exchange_flags |= EXCHGID4_FLAG_CONFIRMED_R;
3114 /* case 5, client reboot */
3119 if (update) { /* case 7 */
3120 status = nfserr_noent;
3124 unconf = find_unconfirmed_client_by_name(&exid->clname, nn);
3125 if (unconf) /* case 4, possible retry or client restart */
3126 unhash_client_locked(unconf);
3128 /* case 1 (normal case) */
3131 status = mark_client_expired_locked(conf);
3135 new->cl_minorversion = cstate->minorversion;
3136 new->cl_spo_must_allow.u.words[0] = exid->spo_must_allow[0];
3137 new->cl_spo_must_allow.u.words[1] = exid->spo_must_allow[1];
3139 add_to_unconfirmed(new);
3142 exid->clientid.cl_boot = conf->cl_clientid.cl_boot;
3143 exid->clientid.cl_id = conf->cl_clientid.cl_id;
3145 exid->seqid = conf->cl_cs_slot.sl_seqid + 1;
3146 nfsd4_set_ex_flags(conf, exid);
3148 dprintk("nfsd4_exchange_id seqid %d flags %x\n",
3149 conf->cl_cs_slot.sl_seqid, conf->cl_exchange_flags);
3153 spin_unlock(&nn->client_lock);
3158 expire_client(unconf);
3163 check_slot_seqid(u32 seqid, u32 slot_seqid, int slot_inuse)
3165 dprintk("%s enter. seqid %d slot_seqid %d\n", __func__, seqid,
3168 /* The slot is in use, and no response has been sent. */
3170 if (seqid == slot_seqid)
3171 return nfserr_jukebox;
3173 return nfserr_seq_misordered;
3175 /* Note unsigned 32-bit arithmetic handles wraparound: */
3176 if (likely(seqid == slot_seqid + 1))
3178 if (seqid == slot_seqid)
3179 return nfserr_replay_cache;
3180 return nfserr_seq_misordered;
3184 * Cache the create session result into the create session single DRC
3185 * slot cache by saving the xdr structure. sl_seqid has been set.
3186 * Do this for solo or embedded create session operations.
3189 nfsd4_cache_create_session(struct nfsd4_create_session *cr_ses,
3190 struct nfsd4_clid_slot *slot, __be32 nfserr)
3192 slot->sl_status = nfserr;
3193 memcpy(&slot->sl_cr_ses, cr_ses, sizeof(*cr_ses));
3197 nfsd4_replay_create_session(struct nfsd4_create_session *cr_ses,
3198 struct nfsd4_clid_slot *slot)
3200 memcpy(cr_ses, &slot->sl_cr_ses, sizeof(*cr_ses));
3201 return slot->sl_status;
3204 #define NFSD_MIN_REQ_HDR_SEQ_SZ ((\
3205 2 * 2 + /* credential,verifier: AUTH_NULL, length 0 */ \
3206 1 + /* MIN tag is length with zero, only length */ \
3207 3 + /* version, opcount, opcode */ \
3208 XDR_QUADLEN(NFS4_MAX_SESSIONID_LEN) + \
3209 /* seqid, slotID, slotID, cache */ \
3210 4 ) * sizeof(__be32))
3212 #define NFSD_MIN_RESP_HDR_SEQ_SZ ((\
3213 2 + /* verifier: AUTH_NULL, length 0 */\
3215 1 + /* MIN tag is length with zero, only length */ \
3216 3 + /* opcount, opcode, opstatus*/ \
3217 XDR_QUADLEN(NFS4_MAX_SESSIONID_LEN) + \
3218 /* seqid, slotID, slotID, slotID, status */ \
3219 5 ) * sizeof(__be32))
3221 static __be32 check_forechannel_attrs(struct nfsd4_channel_attrs *ca, struct nfsd_net *nn)
3223 u32 maxrpc = nn->nfsd_serv->sv_max_mesg;
3225 if (ca->maxreq_sz < NFSD_MIN_REQ_HDR_SEQ_SZ)
3226 return nfserr_toosmall;
3227 if (ca->maxresp_sz < NFSD_MIN_RESP_HDR_SEQ_SZ)
3228 return nfserr_toosmall;
3229 ca->headerpadsz = 0;
3230 ca->maxreq_sz = min_t(u32, ca->maxreq_sz, maxrpc);
3231 ca->maxresp_sz = min_t(u32, ca->maxresp_sz, maxrpc);
3232 ca->maxops = min_t(u32, ca->maxops, NFSD_MAX_OPS_PER_COMPOUND);
3233 ca->maxresp_cached = min_t(u32, ca->maxresp_cached,
3234 NFSD_SLOT_CACHE_SIZE + NFSD_MIN_HDR_SEQ_SZ);
3235 ca->maxreqs = min_t(u32, ca->maxreqs, NFSD_MAX_SLOTS_PER_SESSION);
3237 * Note decreasing slot size below client's request may make it
3238 * difficult for client to function correctly, whereas
3239 * decreasing the number of slots will (just?) affect
3240 * performance. When short on memory we therefore prefer to
3241 * decrease number of slots instead of their size. Clients that
3242 * request larger slots than they need will get poor results:
3243 * Note that we always allow at least one slot, because our
3244 * accounting is soft and provides no guarantees either way.
3246 ca->maxreqs = nfsd4_get_drc_mem(ca, nn);
3252 * Server's NFSv4.1 backchannel support is AUTH_SYS-only for now.
3253 * These are based on similar macros in linux/sunrpc/msg_prot.h .
3255 #define RPC_MAX_HEADER_WITH_AUTH_SYS \
3256 (RPC_CALLHDRSIZE + 2 * (2 + UNX_CALLSLACK))
3258 #define RPC_MAX_REPHEADER_WITH_AUTH_SYS \
3259 (RPC_REPHDRSIZE + (2 + NUL_REPLYSLACK))
3261 #define NFSD_CB_MAX_REQ_SZ ((NFS4_enc_cb_recall_sz + \
3262 RPC_MAX_HEADER_WITH_AUTH_SYS) * sizeof(__be32))
3263 #define NFSD_CB_MAX_RESP_SZ ((NFS4_dec_cb_recall_sz + \
3264 RPC_MAX_REPHEADER_WITH_AUTH_SYS) * \
3267 static __be32 check_backchannel_attrs(struct nfsd4_channel_attrs *ca)
3269 ca->headerpadsz = 0;
3271 if (ca->maxreq_sz < NFSD_CB_MAX_REQ_SZ)
3272 return nfserr_toosmall;
3273 if (ca->maxresp_sz < NFSD_CB_MAX_RESP_SZ)
3274 return nfserr_toosmall;
3275 ca->maxresp_cached = 0;
3277 return nfserr_toosmall;
3282 static __be32 nfsd4_check_cb_sec(struct nfsd4_cb_sec *cbs)
3284 switch (cbs->flavor) {
3290 * GSS case: the spec doesn't allow us to return this
3291 * error. But it also doesn't allow us not to support
3293 * I'd rather this fail hard than return some error the
3294 * client might think it can already handle:
3296 return nfserr_encr_alg_unsupp;
3301 nfsd4_create_session(struct svc_rqst *rqstp,
3302 struct nfsd4_compound_state *cstate, union nfsd4_op_u *u)
3304 struct nfsd4_create_session *cr_ses = &u->create_session;
3305 struct sockaddr *sa = svc_addr(rqstp);
3306 struct nfs4_client *conf, *unconf;
3307 struct nfs4_client *old = NULL;
3308 struct nfsd4_session *new;
3309 struct nfsd4_conn *conn;
3310 struct nfsd4_clid_slot *cs_slot = NULL;
3312 struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
3314 if (cr_ses->flags & ~SESSION4_FLAG_MASK_A)
3315 return nfserr_inval;
3316 status = nfsd4_check_cb_sec(&cr_ses->cb_sec);
3319 status = check_forechannel_attrs(&cr_ses->fore_channel, nn);
3322 status = check_backchannel_attrs(&cr_ses->back_channel);
3324 goto out_release_drc_mem;
3325 status = nfserr_jukebox;
3326 new = alloc_session(&cr_ses->fore_channel, &cr_ses->back_channel);
3328 goto out_release_drc_mem;
3329 conn = alloc_conn_from_crses(rqstp, cr_ses);
3331 goto out_free_session;
3333 spin_lock(&nn->client_lock);
3334 unconf = find_unconfirmed_client(&cr_ses->clientid, true, nn);
3335 conf = find_confirmed_client(&cr_ses->clientid, true, nn);
3336 WARN_ON_ONCE(conf && unconf);
3339 status = nfserr_wrong_cred;
3340 if (!nfsd4_mach_creds_match(conf, rqstp))
3342 cs_slot = &conf->cl_cs_slot;
3343 status = check_slot_seqid(cr_ses->seqid, cs_slot->sl_seqid, 0);
3345 if (status == nfserr_replay_cache)
3346 status = nfsd4_replay_create_session(cr_ses, cs_slot);
3349 } else if (unconf) {
3350 if (!same_creds(&unconf->cl_cred, &rqstp->rq_cred) ||
3351 !rpc_cmp_addr(sa, (struct sockaddr *) &unconf->cl_addr)) {
3352 status = nfserr_clid_inuse;
3355 status = nfserr_wrong_cred;
3356 if (!nfsd4_mach_creds_match(unconf, rqstp))
3358 cs_slot = &unconf->cl_cs_slot;
3359 status = check_slot_seqid(cr_ses->seqid, cs_slot->sl_seqid, 0);
3361 /* an unconfirmed replay returns misordered */
3362 status = nfserr_seq_misordered;
3365 old = find_confirmed_client_by_name(&unconf->cl_name, nn);
3367 status = mark_client_expired_locked(old);
3373 move_to_confirmed(unconf);
3376 status = nfserr_stale_clientid;
3380 /* Persistent sessions are not supported */
3381 cr_ses->flags &= ~SESSION4_PERSIST;
3382 /* Upshifting from TCP to RDMA is not supported */
3383 cr_ses->flags &= ~SESSION4_RDMA;
3385 init_session(rqstp, new, conf, cr_ses);
3386 nfsd4_get_session_locked(new);
3388 memcpy(cr_ses->sessionid.data, new->se_sessionid.data,
3389 NFS4_MAX_SESSIONID_LEN);
3390 cs_slot->sl_seqid++;
3391 cr_ses->seqid = cs_slot->sl_seqid;
3393 /* cache solo and embedded create sessions under the client_lock */
3394 nfsd4_cache_create_session(cr_ses, cs_slot, status);
3395 spin_unlock(&nn->client_lock);
3396 /* init connection and backchannel */
3397 nfsd4_init_conn(rqstp, conn, new);
3398 nfsd4_put_session(new);
3403 spin_unlock(&nn->client_lock);
3408 __free_session(new);
3409 out_release_drc_mem:
3410 nfsd4_put_drc_mem(&cr_ses->fore_channel);
3414 static __be32 nfsd4_map_bcts_dir(u32 *dir)
3417 case NFS4_CDFC4_FORE:
3418 case NFS4_CDFC4_BACK:
3420 case NFS4_CDFC4_FORE_OR_BOTH:
3421 case NFS4_CDFC4_BACK_OR_BOTH:
3422 *dir = NFS4_CDFC4_BOTH;
3425 return nfserr_inval;
3428 __be32 nfsd4_backchannel_ctl(struct svc_rqst *rqstp,
3429 struct nfsd4_compound_state *cstate,
3430 union nfsd4_op_u *u)
3432 struct nfsd4_backchannel_ctl *bc = &u->backchannel_ctl;
3433 struct nfsd4_session *session = cstate->session;
3434 struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
3437 status = nfsd4_check_cb_sec(&bc->bc_cb_sec);
3440 spin_lock(&nn->client_lock);
3441 session->se_cb_prog = bc->bc_cb_program;
3442 session->se_cb_sec = bc->bc_cb_sec;
3443 spin_unlock(&nn->client_lock);
3445 nfsd4_probe_callback(session->se_client);
3450 __be32 nfsd4_bind_conn_to_session(struct svc_rqst *rqstp,
3451 struct nfsd4_compound_state *cstate,
3452 union nfsd4_op_u *u)
3454 struct nfsd4_bind_conn_to_session *bcts = &u->bind_conn_to_session;
3456 struct nfsd4_conn *conn;
3457 struct nfsd4_session *session;
3458 struct net *net = SVC_NET(rqstp);
3459 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
3461 if (!nfsd4_last_compound_op(rqstp))
3462 return nfserr_not_only_op;
3463 spin_lock(&nn->client_lock);
3464 session = find_in_sessionid_hashtbl(&bcts->sessionid, net, &status);
3465 spin_unlock(&nn->client_lock);
3467 goto out_no_session;
3468 status = nfserr_wrong_cred;
3469 if (!nfsd4_mach_creds_match(session->se_client, rqstp))
3471 status = nfsd4_map_bcts_dir(&bcts->dir);
3474 conn = alloc_conn(rqstp, bcts->dir);
3475 status = nfserr_jukebox;
3478 nfsd4_init_conn(rqstp, conn, session);
3481 nfsd4_put_session(session);
3486 static bool nfsd4_compound_in_session(struct nfsd4_compound_state *cstate, struct nfs4_sessionid *sid)
3488 if (!cstate->session)
3490 return !memcmp(sid, &cstate->session->se_sessionid, sizeof(*sid));
3494 nfsd4_destroy_session(struct svc_rqst *r, struct nfsd4_compound_state *cstate,
3495 union nfsd4_op_u *u)
3497 struct nfs4_sessionid *sessionid = &u->destroy_session.sessionid;
3498 struct nfsd4_session *ses;
3500 int ref_held_by_me = 0;
3501 struct net *net = SVC_NET(r);
3502 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
3504 status = nfserr_not_only_op;
3505 if (nfsd4_compound_in_session(cstate, sessionid)) {
3506 if (!nfsd4_last_compound_op(r))
3510 dump_sessionid(__func__, sessionid);
3511 spin_lock(&nn->client_lock);
3512 ses = find_in_sessionid_hashtbl(sessionid, net, &status);
3514 goto out_client_lock;
3515 status = nfserr_wrong_cred;
3516 if (!nfsd4_mach_creds_match(ses->se_client, r))
3517 goto out_put_session;
3518 status = mark_session_dead_locked(ses, 1 + ref_held_by_me);
3520 goto out_put_session;
3521 unhash_session(ses);
3522 spin_unlock(&nn->client_lock);
3524 nfsd4_probe_callback_sync(ses->se_client);
3526 spin_lock(&nn->client_lock);
3529 nfsd4_put_session_locked(ses);
3531 spin_unlock(&nn->client_lock);
3536 static struct nfsd4_conn *__nfsd4_find_conn(struct svc_xprt *xpt, struct nfsd4_session *s)
3538 struct nfsd4_conn *c;
3540 list_for_each_entry(c, &s->se_conns, cn_persession) {
3541 if (c->cn_xprt == xpt) {
3548 static __be32 nfsd4_sequence_check_conn(struct nfsd4_conn *new, struct nfsd4_session *ses)
3550 struct nfs4_client *clp = ses->se_client;
3551 struct nfsd4_conn *c;
3552 __be32 status = nfs_ok;
3555 spin_lock(&clp->cl_lock);
3556 c = __nfsd4_find_conn(new->cn_xprt, ses);
3559 status = nfserr_conn_not_bound_to_session;
3560 if (clp->cl_mach_cred)
3562 __nfsd4_hash_conn(new, ses);
3563 spin_unlock(&clp->cl_lock);
3564 ret = nfsd4_register_conn(new);
3566 /* oops; xprt is already down: */
3567 nfsd4_conn_lost(&new->cn_xpt_user);
3570 spin_unlock(&clp->cl_lock);
3575 static bool nfsd4_session_too_many_ops(struct svc_rqst *rqstp, struct nfsd4_session *session)
3577 struct nfsd4_compoundargs *args = rqstp->rq_argp;
3579 return args->opcnt > session->se_fchannel.maxops;
3582 static bool nfsd4_request_too_big(struct svc_rqst *rqstp,
3583 struct nfsd4_session *session)
3585 struct xdr_buf *xb = &rqstp->rq_arg;
3587 return xb->len > session->se_fchannel.maxreq_sz;
3590 static bool replay_matches_cache(struct svc_rqst *rqstp,
3591 struct nfsd4_sequence *seq, struct nfsd4_slot *slot)
3593 struct nfsd4_compoundargs *argp = rqstp->rq_argp;
3595 if ((bool)(slot->sl_flags & NFSD4_SLOT_CACHETHIS) !=
3596 (bool)seq->cachethis)
3599 * If there's an error then the reply can have fewer ops than
3602 if (slot->sl_opcnt < argp->opcnt && !slot->sl_status)
3605 * But if we cached a reply with *more* ops than the call you're
3606 * sending us now, then this new call is clearly not really a
3607 * replay of the old one:
3609 if (slot->sl_opcnt > argp->opcnt)
3611 /* This is the only check explicitly called by spec: */
3612 if (!same_creds(&rqstp->rq_cred, &slot->sl_cred))
3615 * There may be more comparisons we could actually do, but the
3616 * spec doesn't require us to catch every case where the calls
3617 * don't match (that would require caching the call as well as
3618 * the reply), so we don't bother.
3624 nfsd4_sequence(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3625 union nfsd4_op_u *u)
3627 struct nfsd4_sequence *seq = &u->sequence;
3628 struct nfsd4_compoundres *resp = rqstp->rq_resp;
3629 struct xdr_stream *xdr = &resp->xdr;
3630 struct nfsd4_session *session;
3631 struct nfs4_client *clp;
3632 struct nfsd4_slot *slot;
3633 struct nfsd4_conn *conn;
3636 struct net *net = SVC_NET(rqstp);
3637 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
3639 if (resp->opcnt != 1)
3640 return nfserr_sequence_pos;
3643 * Will be either used or freed by nfsd4_sequence_check_conn
3646 conn = alloc_conn(rqstp, NFS4_CDFC4_FORE);
3648 return nfserr_jukebox;
3650 spin_lock(&nn->client_lock);
3651 session = find_in_sessionid_hashtbl(&seq->sessionid, net, &status);
3653 goto out_no_session;
3654 clp = session->se_client;
3656 status = nfserr_too_many_ops;
3657 if (nfsd4_session_too_many_ops(rqstp, session))
3658 goto out_put_session;
3660 status = nfserr_req_too_big;
3661 if (nfsd4_request_too_big(rqstp, session))
3662 goto out_put_session;
3664 status = nfserr_badslot;
3665 if (seq->slotid >= session->se_fchannel.maxreqs)
3666 goto out_put_session;
3668 slot = session->se_slots[seq->slotid];
3669 dprintk("%s: slotid %d\n", __func__, seq->slotid);
3671 /* We do not negotiate the number of slots yet, so set the
3672 * maxslots to the session maxreqs which is used to encode
3673 * sr_highest_slotid and the sr_target_slot id to maxslots */
3674 seq->maxslots = session->se_fchannel.maxreqs;
3676 status = check_slot_seqid(seq->seqid, slot->sl_seqid,
3677 slot->sl_flags & NFSD4_SLOT_INUSE);
3678 if (status == nfserr_replay_cache) {
3679 status = nfserr_seq_misordered;
3680 if (!(slot->sl_flags & NFSD4_SLOT_INITIALIZED))
3681 goto out_put_session;
3682 status = nfserr_seq_false_retry;
3683 if (!replay_matches_cache(rqstp, seq, slot))
3684 goto out_put_session;
3685 cstate->slot = slot;
3686 cstate->session = session;
3688 /* Return the cached reply status and set cstate->status
3689 * for nfsd4_proc_compound processing */
3690 status = nfsd4_replay_cache_entry(resp, seq);
3691 cstate->status = nfserr_replay_cache;
3695 goto out_put_session;
3697 status = nfsd4_sequence_check_conn(conn, session);
3700 goto out_put_session;
3702 buflen = (seq->cachethis) ?
3703 session->se_fchannel.maxresp_cached :
3704 session->se_fchannel.maxresp_sz;
3705 status = (seq->cachethis) ? nfserr_rep_too_big_to_cache :
3707 if (xdr_restrict_buflen(xdr, buflen - rqstp->rq_auth_slack))
3708 goto out_put_session;
3709 svc_reserve(rqstp, buflen);
3712 /* Success! bump slot seqid */
3713 slot->sl_seqid = seq->seqid;
3714 slot->sl_flags |= NFSD4_SLOT_INUSE;
3716 slot->sl_flags |= NFSD4_SLOT_CACHETHIS;
3718 slot->sl_flags &= ~NFSD4_SLOT_CACHETHIS;
3720 cstate->slot = slot;
3721 cstate->session = session;
3725 switch (clp->cl_cb_state) {
3727 seq->status_flags = SEQ4_STATUS_CB_PATH_DOWN;
3729 case NFSD4_CB_FAULT:
3730 seq->status_flags = SEQ4_STATUS_BACKCHANNEL_FAULT;
3733 seq->status_flags = 0;
3735 if (!list_empty(&clp->cl_revoked))
3736 seq->status_flags |= SEQ4_STATUS_RECALLABLE_STATE_REVOKED;
3740 spin_unlock(&nn->client_lock);
3743 nfsd4_put_session_locked(session);
3744 goto out_no_session;
3748 nfsd4_sequence_done(struct nfsd4_compoundres *resp)
3750 struct nfsd4_compound_state *cs = &resp->cstate;
3752 if (nfsd4_has_session(cs)) {
3753 if (cs->status != nfserr_replay_cache) {
3754 nfsd4_store_cache_entry(resp);
3755 cs->slot->sl_flags &= ~NFSD4_SLOT_INUSE;
3757 /* Drop session reference that was taken in nfsd4_sequence() */
3758 nfsd4_put_session(cs->session);
3760 put_client_renew(cs->clp);
3764 nfsd4_destroy_clientid(struct svc_rqst *rqstp,
3765 struct nfsd4_compound_state *cstate,
3766 union nfsd4_op_u *u)
3768 struct nfsd4_destroy_clientid *dc = &u->destroy_clientid;
3769 struct nfs4_client *conf, *unconf;
3770 struct nfs4_client *clp = NULL;
3772 struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
3774 spin_lock(&nn->client_lock);
3775 unconf = find_unconfirmed_client(&dc->clientid, true, nn);
3776 conf = find_confirmed_client(&dc->clientid, true, nn);
3777 WARN_ON_ONCE(conf && unconf);
3780 if (client_has_state(conf)) {
3781 status = nfserr_clientid_busy;
3784 status = mark_client_expired_locked(conf);
3791 status = nfserr_stale_clientid;
3794 if (!nfsd4_mach_creds_match(clp, rqstp)) {
3796 status = nfserr_wrong_cred;
3799 unhash_client_locked(clp);
3801 spin_unlock(&nn->client_lock);
3808 nfsd4_reclaim_complete(struct svc_rqst *rqstp,
3809 struct nfsd4_compound_state *cstate, union nfsd4_op_u *u)
3811 struct nfsd4_reclaim_complete *rc = &u->reclaim_complete;
3814 if (rc->rca_one_fs) {
3815 if (!cstate->current_fh.fh_dentry)
3816 return nfserr_nofilehandle;
3818 * We don't take advantage of the rca_one_fs case.
3819 * That's OK, it's optional, we can safely ignore it.
3824 status = nfserr_complete_already;
3825 if (test_and_set_bit(NFSD4_CLIENT_RECLAIM_COMPLETE,
3826 &cstate->session->se_client->cl_flags))
3829 status = nfserr_stale_clientid;
3830 if (is_client_expired(cstate->session->se_client))
3832 * The following error isn't really legal.
3833 * But we only get here if the client just explicitly
3834 * destroyed the client. Surely it no longer cares what
3835 * error it gets back on an operation for the dead
3841 nfsd4_client_record_create(cstate->session->se_client);
3842 inc_reclaim_complete(cstate->session->se_client);
3848 nfsd4_setclientid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3849 union nfsd4_op_u *u)
3851 struct nfsd4_setclientid *setclid = &u->setclientid;
3852 struct xdr_netobj clname = setclid->se_name;
3853 nfs4_verifier clverifier = setclid->se_verf;
3854 struct nfs4_client *conf, *new;
3855 struct nfs4_client *unconf = NULL;
3857 struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
3859 new = create_client(clname, rqstp, &clverifier);
3861 return nfserr_jukebox;
3862 /* Cases below refer to rfc 3530 section 14.2.33: */
3863 spin_lock(&nn->client_lock);
3864 conf = find_confirmed_client_by_name(&clname, nn);
3865 if (conf && client_has_state(conf)) {
3867 status = nfserr_clid_inuse;
3868 if (clp_used_exchangeid(conf))
3870 if (!same_creds(&conf->cl_cred, &rqstp->rq_cred)) {
3871 char addr_str[INET6_ADDRSTRLEN];
3872 rpc_ntop((struct sockaddr *) &conf->cl_addr, addr_str,
3874 dprintk("NFSD: setclientid: string in use by client "
3875 "at %s\n", addr_str);
3879 unconf = find_unconfirmed_client_by_name(&clname, nn);
3881 unhash_client_locked(unconf);
3882 if (conf && same_verf(&conf->cl_verifier, &clverifier)) {
3883 /* case 1: probable callback update */
3884 copy_clid(new, conf);
3885 gen_confirm(new, nn);
3886 } else /* case 4 (new client) or cases 2, 3 (client reboot): */
3888 new->cl_minorversion = 0;
3889 gen_callback(new, setclid, rqstp);
3890 add_to_unconfirmed(new);
3891 setclid->se_clientid.cl_boot = new->cl_clientid.cl_boot;
3892 setclid->se_clientid.cl_id = new->cl_clientid.cl_id;
3893 memcpy(setclid->se_confirm.data, new->cl_confirm.data, sizeof(setclid->se_confirm.data));
3897 spin_unlock(&nn->client_lock);
3901 expire_client(unconf);
3907 nfsd4_setclientid_confirm(struct svc_rqst *rqstp,
3908 struct nfsd4_compound_state *cstate,
3909 union nfsd4_op_u *u)
3911 struct nfsd4_setclientid_confirm *setclientid_confirm =
3912 &u->setclientid_confirm;
3913 struct nfs4_client *conf, *unconf;
3914 struct nfs4_client *old = NULL;
3915 nfs4_verifier confirm = setclientid_confirm->sc_confirm;
3916 clientid_t * clid = &setclientid_confirm->sc_clientid;
3918 struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
3920 if (STALE_CLIENTID(clid, nn))
3921 return nfserr_stale_clientid;
3923 spin_lock(&nn->client_lock);
3924 conf = find_confirmed_client(clid, false, nn);
3925 unconf = find_unconfirmed_client(clid, false, nn);
3927 * We try hard to give out unique clientid's, so if we get an
3928 * attempt to confirm the same clientid with a different cred,
3929 * the client may be buggy; this should never happen.
3931 * Nevertheless, RFC 7530 recommends INUSE for this case:
3933 status = nfserr_clid_inuse;
3934 if (unconf && !same_creds(&unconf->cl_cred, &rqstp->rq_cred))
3936 if (conf && !same_creds(&conf->cl_cred, &rqstp->rq_cred))
3938 /* cases below refer to rfc 3530 section 14.2.34: */
3939 if (!unconf || !same_verf(&confirm, &unconf->cl_confirm)) {
3940 if (conf && same_verf(&confirm, &conf->cl_confirm)) {
3941 /* case 2: probable retransmit */
3943 } else /* case 4: client hasn't noticed we rebooted yet? */
3944 status = nfserr_stale_clientid;
3948 if (conf) { /* case 1: callback update */
3950 unhash_client_locked(old);
3951 nfsd4_change_callback(conf, &unconf->cl_cb_conn);
3952 } else { /* case 3: normal case; new or rebooted client */
3953 old = find_confirmed_client_by_name(&unconf->cl_name, nn);
3955 status = nfserr_clid_inuse;
3956 if (client_has_state(old)
3957 && !same_creds(&unconf->cl_cred,
3960 status = mark_client_expired_locked(old);
3966 move_to_confirmed(unconf);
3969 get_client_locked(conf);
3970 spin_unlock(&nn->client_lock);
3971 nfsd4_probe_callback(conf);
3972 spin_lock(&nn->client_lock);
3973 put_client_renew_locked(conf);
3975 spin_unlock(&nn->client_lock);
3981 static struct nfs4_file *nfsd4_alloc_file(void)
3983 return kmem_cache_alloc(file_slab, GFP_KERNEL);
3986 /* OPEN Share state helper functions */
3987 static void nfsd4_init_file(struct knfsd_fh *fh, unsigned int hashval,
3988 struct nfs4_file *fp)
3990 lockdep_assert_held(&state_lock);
3992 refcount_set(&fp->fi_ref, 1);
3993 spin_lock_init(&fp->fi_lock);
3994 INIT_LIST_HEAD(&fp->fi_stateids);
3995 INIT_LIST_HEAD(&fp->fi_delegations);
3996 INIT_LIST_HEAD(&fp->fi_clnt_odstate);
3997 fh_copy_shallow(&fp->fi_fhandle, fh);
3998 fp->fi_deleg_file = NULL;
3999 fp->fi_had_conflict = false;
4000 fp->fi_share_deny = 0;
4001 memset(fp->fi_fds, 0, sizeof(fp->fi_fds));
4002 memset(fp->fi_access, 0, sizeof(fp->fi_access));
4003 #ifdef CONFIG_NFSD_PNFS
4004 INIT_LIST_HEAD(&fp->fi_lo_states);
4005 atomic_set(&fp->fi_lo_recalls, 0);
4007 hlist_add_head_rcu(&fp->fi_hash, &file_hashtbl[hashval]);
4011 nfsd4_free_slabs(void)
4013 kmem_cache_destroy(client_slab);
4014 kmem_cache_destroy(openowner_slab);
4015 kmem_cache_destroy(lockowner_slab);
4016 kmem_cache_destroy(file_slab);
4017 kmem_cache_destroy(stateid_slab);
4018 kmem_cache_destroy(deleg_slab);
4019 kmem_cache_destroy(odstate_slab);
4023 nfsd4_init_slabs(void)
4025 client_slab = kmem_cache_create("nfsd4_clients",
4026 sizeof(struct nfs4_client), 0, 0, NULL);
4027 if (client_slab == NULL)
4029 openowner_slab = kmem_cache_create("nfsd4_openowners",
4030 sizeof(struct nfs4_openowner), 0, 0, NULL);
4031 if (openowner_slab == NULL)
4032 goto out_free_client_slab;
4033 lockowner_slab = kmem_cache_create("nfsd4_lockowners",
4034 sizeof(struct nfs4_lockowner), 0, 0, NULL);
4035 if (lockowner_slab == NULL)
4036 goto out_free_openowner_slab;
4037 file_slab = kmem_cache_create("nfsd4_files",
4038 sizeof(struct nfs4_file), 0, 0, NULL);
4039 if (file_slab == NULL)
4040 goto out_free_lockowner_slab;
4041 stateid_slab = kmem_cache_create("nfsd4_stateids",
4042 sizeof(struct nfs4_ol_stateid), 0, 0, NULL);
4043 if (stateid_slab == NULL)
4044 goto out_free_file_slab;
4045 deleg_slab = kmem_cache_create("nfsd4_delegations",
4046 sizeof(struct nfs4_delegation), 0, 0, NULL);
4047 if (deleg_slab == NULL)
4048 goto out_free_stateid_slab;
4049 odstate_slab = kmem_cache_create("nfsd4_odstate",
4050 sizeof(struct nfs4_clnt_odstate), 0, 0, NULL);
4051 if (odstate_slab == NULL)
4052 goto out_free_deleg_slab;
4055 out_free_deleg_slab:
4056 kmem_cache_destroy(deleg_slab);
4057 out_free_stateid_slab:
4058 kmem_cache_destroy(stateid_slab);
4060 kmem_cache_destroy(file_slab);
4061 out_free_lockowner_slab:
4062 kmem_cache_destroy(lockowner_slab);
4063 out_free_openowner_slab:
4064 kmem_cache_destroy(openowner_slab);
4065 out_free_client_slab:
4066 kmem_cache_destroy(client_slab);
4068 dprintk("nfsd4: out of memory while initializing nfsv4\n");
4072 static void init_nfs4_replay(struct nfs4_replay *rp)
4074 rp->rp_status = nfserr_serverfault;
4076 rp->rp_buf = rp->rp_ibuf;
4077 mutex_init(&rp->rp_mutex);
4080 static void nfsd4_cstate_assign_replay(struct nfsd4_compound_state *cstate,
4081 struct nfs4_stateowner *so)
4083 if (!nfsd4_has_session(cstate)) {
4084 mutex_lock(&so->so_replay.rp_mutex);
4085 cstate->replay_owner = nfs4_get_stateowner(so);
4089 void nfsd4_cstate_clear_replay(struct nfsd4_compound_state *cstate)
4091 struct nfs4_stateowner *so = cstate->replay_owner;
4094 cstate->replay_owner = NULL;
4095 mutex_unlock(&so->so_replay.rp_mutex);
4096 nfs4_put_stateowner(so);
4100 static inline void *alloc_stateowner(struct kmem_cache *slab, struct xdr_netobj *owner, struct nfs4_client *clp)
4102 struct nfs4_stateowner *sop;
4104 sop = kmem_cache_alloc(slab, GFP_KERNEL);
4108 xdr_netobj_dup(&sop->so_owner, owner, GFP_KERNEL);
4109 if (!sop->so_owner.data) {
4110 kmem_cache_free(slab, sop);
4114 INIT_LIST_HEAD(&sop->so_stateids);
4115 sop->so_client = clp;
4116 init_nfs4_replay(&sop->so_replay);
4117 atomic_set(&sop->so_count, 1);
4121 static void hash_openowner(struct nfs4_openowner *oo, struct nfs4_client *clp, unsigned int strhashval)
4123 lockdep_assert_held(&clp->cl_lock);
4125 list_add(&oo->oo_owner.so_strhash,
4126 &clp->cl_ownerstr_hashtbl[strhashval]);
4127 list_add(&oo->oo_perclient, &clp->cl_openowners);
4130 static void nfs4_unhash_openowner(struct nfs4_stateowner *so)
4132 unhash_openowner_locked(openowner(so));
4135 static void nfs4_free_openowner(struct nfs4_stateowner *so)
4137 struct nfs4_openowner *oo = openowner(so);
4139 kmem_cache_free(openowner_slab, oo);
4142 static const struct nfs4_stateowner_operations openowner_ops = {
4143 .so_unhash = nfs4_unhash_openowner,
4144 .so_free = nfs4_free_openowner,
4147 static struct nfs4_ol_stateid *
4148 nfsd4_find_existing_open(struct nfs4_file *fp, struct nfsd4_open *open)
4150 struct nfs4_ol_stateid *local, *ret = NULL;
4151 struct nfs4_openowner *oo = open->op_openowner;
4153 lockdep_assert_held(&fp->fi_lock);
4155 list_for_each_entry(local, &fp->fi_stateids, st_perfile) {
4156 /* ignore lock owners */
4157 if (local->st_stateowner->so_is_open_owner == 0)
4159 if (local->st_stateowner != &oo->oo_owner)
4161 if (local->st_stid.sc_type == NFS4_OPEN_STID) {
4163 refcount_inc(&ret->st_stid.sc_count);
4171 nfsd4_verify_open_stid(struct nfs4_stid *s)
4173 __be32 ret = nfs_ok;
4175 switch (s->sc_type) {
4179 case NFS4_CLOSED_STID:
4180 case NFS4_CLOSED_DELEG_STID:
4181 ret = nfserr_bad_stateid;
4183 case NFS4_REVOKED_DELEG_STID:
4184 ret = nfserr_deleg_revoked;
4189 /* Lock the stateid st_mutex, and deal with races with CLOSE */
4191 nfsd4_lock_ol_stateid(struct nfs4_ol_stateid *stp)
4195 mutex_lock_nested(&stp->st_mutex, LOCK_STATEID_MUTEX);
4196 ret = nfsd4_verify_open_stid(&stp->st_stid);
4198 mutex_unlock(&stp->st_mutex);
4202 static struct nfs4_ol_stateid *
4203 nfsd4_find_and_lock_existing_open(struct nfs4_file *fp, struct nfsd4_open *open)
4205 struct nfs4_ol_stateid *stp;
4207 spin_lock(&fp->fi_lock);
4208 stp = nfsd4_find_existing_open(fp, open);
4209 spin_unlock(&fp->fi_lock);
4210 if (!stp || nfsd4_lock_ol_stateid(stp) == nfs_ok)
4212 nfs4_put_stid(&stp->st_stid);
4217 static struct nfs4_openowner *
4218 alloc_init_open_stateowner(unsigned int strhashval, struct nfsd4_open *open,
4219 struct nfsd4_compound_state *cstate)
4221 struct nfs4_client *clp = cstate->clp;
4222 struct nfs4_openowner *oo, *ret;
4224 oo = alloc_stateowner(openowner_slab, &open->op_owner, clp);
4227 oo->oo_owner.so_ops = &openowner_ops;
4228 oo->oo_owner.so_is_open_owner = 1;
4229 oo->oo_owner.so_seqid = open->op_seqid;
4231 if (nfsd4_has_session(cstate))
4232 oo->oo_flags |= NFS4_OO_CONFIRMED;
4234 oo->oo_last_closed_stid = NULL;
4235 INIT_LIST_HEAD(&oo->oo_close_lru);
4236 spin_lock(&clp->cl_lock);
4237 ret = find_openstateowner_str_locked(strhashval, open, clp);
4239 hash_openowner(oo, clp, strhashval);
4242 nfs4_free_stateowner(&oo->oo_owner);
4244 spin_unlock(&clp->cl_lock);
4248 static struct nfs4_ol_stateid *
4249 init_open_stateid(struct nfs4_file *fp, struct nfsd4_open *open)
4252 struct nfs4_openowner *oo = open->op_openowner;
4253 struct nfs4_ol_stateid *retstp = NULL;
4254 struct nfs4_ol_stateid *stp;
4257 /* We are moving these outside of the spinlocks to avoid the warnings */
4258 mutex_init(&stp->st_mutex);
4259 mutex_lock_nested(&stp->st_mutex, OPEN_STATEID_MUTEX);
4262 spin_lock(&oo->oo_owner.so_client->cl_lock);
4263 spin_lock(&fp->fi_lock);
4265 retstp = nfsd4_find_existing_open(fp, open);
4269 open->op_stp = NULL;
4270 refcount_inc(&stp->st_stid.sc_count);
4271 stp->st_stid.sc_type = NFS4_OPEN_STID;
4272 INIT_LIST_HEAD(&stp->st_locks);
4273 stp->st_stateowner = nfs4_get_stateowner(&oo->oo_owner);
4275 stp->st_stid.sc_file = fp;
4276 stp->st_access_bmap = 0;
4277 stp->st_deny_bmap = 0;
4278 stp->st_openstp = NULL;
4279 list_add(&stp->st_perstateowner, &oo->oo_owner.so_stateids);
4280 list_add(&stp->st_perfile, &fp->fi_stateids);
4283 spin_unlock(&fp->fi_lock);
4284 spin_unlock(&oo->oo_owner.so_client->cl_lock);
4286 /* Handle races with CLOSE */
4287 if (nfsd4_lock_ol_stateid(retstp) != nfs_ok) {
4288 nfs4_put_stid(&retstp->st_stid);
4291 /* To keep mutex tracking happy */
4292 mutex_unlock(&stp->st_mutex);
4299 * In the 4.0 case we need to keep the owners around a little while to handle
4300 * CLOSE replay. We still do need to release any file access that is held by
4301 * them before returning however.
4304 move_to_close_lru(struct nfs4_ol_stateid *s, struct net *net)
4306 struct nfs4_ol_stateid *last;
4307 struct nfs4_openowner *oo = openowner(s->st_stateowner);
4308 struct nfsd_net *nn = net_generic(s->st_stid.sc_client->net,
4311 dprintk("NFSD: move_to_close_lru nfs4_openowner %p\n", oo);
4314 * We know that we hold one reference via nfsd4_close, and another
4315 * "persistent" reference for the client. If the refcount is higher
4316 * than 2, then there are still calls in progress that are using this
4317 * stateid. We can't put the sc_file reference until they are finished.
4318 * Wait for the refcount to drop to 2. Since it has been unhashed,
4319 * there should be no danger of the refcount going back up again at
4322 wait_event(close_wq, refcount_read(&s->st_stid.sc_count) == 2);
4324 release_all_access(s);
4325 if (s->st_stid.sc_file) {
4326 put_nfs4_file(s->st_stid.sc_file);
4327 s->st_stid.sc_file = NULL;
4330 spin_lock(&nn->client_lock);
4331 last = oo->oo_last_closed_stid;
4332 oo->oo_last_closed_stid = s;
4333 list_move_tail(&oo->oo_close_lru, &nn->close_lru);
4334 oo->oo_time = ktime_get_boottime_seconds();
4335 spin_unlock(&nn->client_lock);
4337 nfs4_put_stid(&last->st_stid);
4340 /* search file_hashtbl[] for file */
4341 static struct nfs4_file *
4342 find_file_locked(struct knfsd_fh *fh, unsigned int hashval)
4344 struct nfs4_file *fp;
4346 hlist_for_each_entry_rcu(fp, &file_hashtbl[hashval], fi_hash) {
4347 if (fh_match(&fp->fi_fhandle, fh)) {
4348 if (refcount_inc_not_zero(&fp->fi_ref))
4356 find_file(struct knfsd_fh *fh)
4358 struct nfs4_file *fp;
4359 unsigned int hashval = file_hashval(fh);
4362 fp = find_file_locked(fh, hashval);
4367 static struct nfs4_file *
4368 find_or_add_file(struct nfs4_file *new, struct knfsd_fh *fh)
4370 struct nfs4_file *fp;
4371 unsigned int hashval = file_hashval(fh);
4374 fp = find_file_locked(fh, hashval);
4379 spin_lock(&state_lock);
4380 fp = find_file_locked(fh, hashval);
4381 if (likely(fp == NULL)) {
4382 nfsd4_init_file(fh, hashval, new);
4385 spin_unlock(&state_lock);
4391 * Called to check deny when READ with all zero stateid or
4392 * WRITE with all zero or all one stateid
4395 nfs4_share_conflict(struct svc_fh *current_fh, unsigned int deny_type)
4397 struct nfs4_file *fp;
4398 __be32 ret = nfs_ok;
4400 fp = find_file(¤t_fh->fh_handle);
4403 /* Check for conflicting share reservations */
4404 spin_lock(&fp->fi_lock);
4405 if (fp->fi_share_deny & deny_type)
4406 ret = nfserr_locked;
4407 spin_unlock(&fp->fi_lock);
4412 static void nfsd4_cb_recall_prepare(struct nfsd4_callback *cb)
4414 struct nfs4_delegation *dp = cb_to_delegation(cb);
4415 struct nfsd_net *nn = net_generic(dp->dl_stid.sc_client->net,
4418 block_delegations(&dp->dl_stid.sc_file->fi_fhandle);
4421 * We can't do this in nfsd_break_deleg_cb because it is
4422 * already holding inode->i_lock.
4424 * If the dl_time != 0, then we know that it has already been
4425 * queued for a lease break. Don't queue it again.
4427 spin_lock(&state_lock);
4428 if (dp->dl_time == 0) {
4429 dp->dl_time = ktime_get_boottime_seconds();
4430 list_add_tail(&dp->dl_recall_lru, &nn->del_recall_lru);
4432 spin_unlock(&state_lock);
4435 static int nfsd4_cb_recall_done(struct nfsd4_callback *cb,
4436 struct rpc_task *task)
4438 struct nfs4_delegation *dp = cb_to_delegation(cb);
4440 if (dp->dl_stid.sc_type == NFS4_CLOSED_DELEG_STID)
4443 switch (task->tk_status) {
4446 case -NFS4ERR_DELAY:
4447 rpc_delay(task, 2 * HZ);
4450 case -NFS4ERR_BAD_STATEID:
4452 * Race: client probably got cb_recall before open reply
4453 * granting delegation.
4455 if (dp->dl_retries--) {
4456 rpc_delay(task, 2 * HZ);
4465 static void nfsd4_cb_recall_release(struct nfsd4_callback *cb)
4467 struct nfs4_delegation *dp = cb_to_delegation(cb);
4469 nfs4_put_stid(&dp->dl_stid);
4472 static const struct nfsd4_callback_ops nfsd4_cb_recall_ops = {
4473 .prepare = nfsd4_cb_recall_prepare,
4474 .done = nfsd4_cb_recall_done,
4475 .release = nfsd4_cb_recall_release,
4478 static void nfsd_break_one_deleg(struct nfs4_delegation *dp)
4481 * We're assuming the state code never drops its reference
4482 * without first removing the lease. Since we're in this lease
4483 * callback (and since the lease code is serialized by the
4484 * i_lock) we know the server hasn't removed the lease yet, and
4485 * we know it's safe to take a reference.
4487 refcount_inc(&dp->dl_stid.sc_count);
4488 nfsd4_run_cb(&dp->dl_recall);
4491 /* Called from break_lease() with i_lock held. */
4493 nfsd_break_deleg_cb(struct file_lock *fl)
4496 struct nfs4_delegation *dp = (struct nfs4_delegation *)fl->fl_owner;
4497 struct nfs4_file *fp = dp->dl_stid.sc_file;
4500 * We don't want the locks code to timeout the lease for us;
4501 * we'll remove it ourself if a delegation isn't returned
4504 fl->fl_break_time = 0;
4506 spin_lock(&fp->fi_lock);
4507 fp->fi_had_conflict = true;
4508 nfsd_break_one_deleg(dp);
4509 spin_unlock(&fp->fi_lock);
4514 nfsd_change_deleg_cb(struct file_lock *onlist, int arg,
4515 struct list_head *dispose)
4518 return lease_modify(onlist, arg, dispose);
4523 static const struct lock_manager_operations nfsd_lease_mng_ops = {
4524 .lm_break = nfsd_break_deleg_cb,
4525 .lm_change = nfsd_change_deleg_cb,
4528 static __be32 nfsd4_check_seqid(struct nfsd4_compound_state *cstate, struct nfs4_stateowner *so, u32 seqid)
4530 if (nfsd4_has_session(cstate))
4532 if (seqid == so->so_seqid - 1)
4533 return nfserr_replay_me;
4534 if (seqid == so->so_seqid)
4536 return nfserr_bad_seqid;
4539 static __be32 lookup_clientid(clientid_t *clid,
4540 struct nfsd4_compound_state *cstate,
4541 struct nfsd_net *nn,
4544 struct nfs4_client *found;
4547 found = cstate->clp;
4548 if (!same_clid(&found->cl_clientid, clid))
4549 return nfserr_stale_clientid;
4553 if (STALE_CLIENTID(clid, nn))
4554 return nfserr_stale_clientid;
4557 * For v4.1+ we get the client in the SEQUENCE op. If we don't have one
4558 * cached already then we know this is for is for v4.0 and "sessions"
4561 WARN_ON_ONCE(cstate->session);
4562 spin_lock(&nn->client_lock);
4563 found = find_confirmed_client(clid, sessions, nn);
4565 spin_unlock(&nn->client_lock);
4566 return nfserr_expired;
4568 atomic_inc(&found->cl_rpc_users);
4569 spin_unlock(&nn->client_lock);
4571 /* Cache the nfs4_client in cstate! */
4572 cstate->clp = found;
4577 nfsd4_process_open1(struct nfsd4_compound_state *cstate,
4578 struct nfsd4_open *open, struct nfsd_net *nn)
4580 clientid_t *clientid = &open->op_clientid;
4581 struct nfs4_client *clp = NULL;
4582 unsigned int strhashval;
4583 struct nfs4_openowner *oo = NULL;
4586 if (STALE_CLIENTID(&open->op_clientid, nn))
4587 return nfserr_stale_clientid;
4589 * In case we need it later, after we've already created the
4590 * file and don't want to risk a further failure:
4592 open->op_file = nfsd4_alloc_file();
4593 if (open->op_file == NULL)
4594 return nfserr_jukebox;
4596 status = lookup_clientid(clientid, cstate, nn, false);
4601 strhashval = ownerstr_hashval(&open->op_owner);
4602 oo = find_openstateowner_str(strhashval, open, clp);
4603 open->op_openowner = oo;
4607 if (!(oo->oo_flags & NFS4_OO_CONFIRMED)) {
4608 /* Replace unconfirmed owners without checking for replay. */
4609 release_openowner(oo);
4610 open->op_openowner = NULL;
4613 status = nfsd4_check_seqid(cstate, &oo->oo_owner, open->op_seqid);
4618 oo = alloc_init_open_stateowner(strhashval, open, cstate);
4620 return nfserr_jukebox;
4621 open->op_openowner = oo;
4623 open->op_stp = nfs4_alloc_open_stateid(clp);
4625 return nfserr_jukebox;
4627 if (nfsd4_has_session(cstate) &&
4628 (cstate->current_fh.fh_export->ex_flags & NFSEXP_PNFS)) {
4629 open->op_odstate = alloc_clnt_odstate(clp);
4630 if (!open->op_odstate)
4631 return nfserr_jukebox;
4637 static inline __be32
4638 nfs4_check_delegmode(struct nfs4_delegation *dp, int flags)
4640 if ((flags & WR_STATE) && (dp->dl_type == NFS4_OPEN_DELEGATE_READ))
4641 return nfserr_openmode;
4646 static int share_access_to_flags(u32 share_access)
4648 return share_access == NFS4_SHARE_ACCESS_READ ? RD_STATE : WR_STATE;
4651 static struct nfs4_delegation *find_deleg_stateid(struct nfs4_client *cl, stateid_t *s)
4653 struct nfs4_stid *ret;
4655 ret = find_stateid_by_type(cl, s,
4656 NFS4_DELEG_STID|NFS4_REVOKED_DELEG_STID);
4659 return delegstateid(ret);
4662 static bool nfsd4_is_deleg_cur(struct nfsd4_open *open)
4664 return open->op_claim_type == NFS4_OPEN_CLAIM_DELEGATE_CUR ||
4665 open->op_claim_type == NFS4_OPEN_CLAIM_DELEG_CUR_FH;
4669 nfs4_check_deleg(struct nfs4_client *cl, struct nfsd4_open *open,
4670 struct nfs4_delegation **dp)
4673 __be32 status = nfserr_bad_stateid;
4674 struct nfs4_delegation *deleg;
4676 deleg = find_deleg_stateid(cl, &open->op_delegate_stateid);
4679 if (deleg->dl_stid.sc_type == NFS4_REVOKED_DELEG_STID) {
4680 nfs4_put_stid(&deleg->dl_stid);
4681 if (cl->cl_minorversion)
4682 status = nfserr_deleg_revoked;
4685 flags = share_access_to_flags(open->op_share_access);
4686 status = nfs4_check_delegmode(deleg, flags);
4688 nfs4_put_stid(&deleg->dl_stid);
4693 if (!nfsd4_is_deleg_cur(open))
4697 open->op_openowner->oo_flags |= NFS4_OO_CONFIRMED;
4701 static inline int nfs4_access_to_access(u32 nfs4_access)
4705 if (nfs4_access & NFS4_SHARE_ACCESS_READ)
4706 flags |= NFSD_MAY_READ;
4707 if (nfs4_access & NFS4_SHARE_ACCESS_WRITE)
4708 flags |= NFSD_MAY_WRITE;
4712 static inline __be32
4713 nfsd4_truncate(struct svc_rqst *rqstp, struct svc_fh *fh,
4714 struct nfsd4_open *open)
4716 struct iattr iattr = {
4717 .ia_valid = ATTR_SIZE,
4720 if (!open->op_truncate)
4722 if (!(open->op_share_access & NFS4_SHARE_ACCESS_WRITE))
4723 return nfserr_inval;
4724 return nfsd_setattr(rqstp, fh, &iattr, 0, (time64_t)0);
4727 static __be32 nfs4_get_vfs_file(struct svc_rqst *rqstp, struct nfs4_file *fp,
4728 struct svc_fh *cur_fh, struct nfs4_ol_stateid *stp,
4729 struct nfsd4_open *open)
4731 struct nfsd_file *nf = NULL;
4733 int oflag = nfs4_access_to_omode(open->op_share_access);
4734 int access = nfs4_access_to_access(open->op_share_access);
4735 unsigned char old_access_bmap, old_deny_bmap;
4737 spin_lock(&fp->fi_lock);
4740 * Are we trying to set a deny mode that would conflict with
4743 status = nfs4_file_check_deny(fp, open->op_share_deny);
4744 if (status != nfs_ok) {
4745 spin_unlock(&fp->fi_lock);
4749 /* set access to the file */
4750 status = nfs4_file_get_access(fp, open->op_share_access);
4751 if (status != nfs_ok) {
4752 spin_unlock(&fp->fi_lock);
4756 /* Set access bits in stateid */
4757 old_access_bmap = stp->st_access_bmap;
4758 set_access(open->op_share_access, stp);
4760 /* Set new deny mask */
4761 old_deny_bmap = stp->st_deny_bmap;
4762 set_deny(open->op_share_deny, stp);
4763 fp->fi_share_deny |= (open->op_share_deny & NFS4_SHARE_DENY_BOTH);
4765 if (!fp->fi_fds[oflag]) {
4766 spin_unlock(&fp->fi_lock);
4767 status = nfsd_file_acquire(rqstp, cur_fh, access, &nf);
4769 goto out_put_access;
4770 spin_lock(&fp->fi_lock);
4771 if (!fp->fi_fds[oflag]) {
4772 fp->fi_fds[oflag] = nf;
4776 spin_unlock(&fp->fi_lock);
4780 status = nfsd4_truncate(rqstp, cur_fh, open);
4782 goto out_put_access;
4786 stp->st_access_bmap = old_access_bmap;
4787 nfs4_file_put_access(fp, open->op_share_access);
4788 reset_union_bmap_deny(bmap_to_share_mode(old_deny_bmap), stp);
4793 nfs4_upgrade_open(struct svc_rqst *rqstp, struct nfs4_file *fp, struct svc_fh *cur_fh, struct nfs4_ol_stateid *stp, struct nfsd4_open *open)
4796 unsigned char old_deny_bmap = stp->st_deny_bmap;
4798 if (!test_access(open->op_share_access, stp))
4799 return nfs4_get_vfs_file(rqstp, fp, cur_fh, stp, open);
4801 /* test and set deny mode */
4802 spin_lock(&fp->fi_lock);
4803 status = nfs4_file_check_deny(fp, open->op_share_deny);
4804 if (status == nfs_ok) {
4805 set_deny(open->op_share_deny, stp);
4806 fp->fi_share_deny |=
4807 (open->op_share_deny & NFS4_SHARE_DENY_BOTH);
4809 spin_unlock(&fp->fi_lock);
4811 if (status != nfs_ok)
4814 status = nfsd4_truncate(rqstp, cur_fh, open);
4815 if (status != nfs_ok)
4816 reset_union_bmap_deny(old_deny_bmap, stp);
4820 /* Should we give out recallable state?: */
4821 static bool nfsd4_cb_channel_good(struct nfs4_client *clp)
4823 if (clp->cl_cb_state == NFSD4_CB_UP)
4826 * In the sessions case, since we don't have to establish a
4827 * separate connection for callbacks, we assume it's OK
4828 * until we hear otherwise:
4830 return clp->cl_minorversion && clp->cl_cb_state == NFSD4_CB_UNKNOWN;
4833 static struct file_lock *nfs4_alloc_init_lease(struct nfs4_delegation *dp,
4836 struct file_lock *fl;
4838 fl = locks_alloc_lock();
4841 fl->fl_lmops = &nfsd_lease_mng_ops;
4842 fl->fl_flags = FL_DELEG;
4843 fl->fl_type = flag == NFS4_OPEN_DELEGATE_READ? F_RDLCK: F_WRLCK;
4844 fl->fl_end = OFFSET_MAX;
4845 fl->fl_owner = (fl_owner_t)dp;
4846 fl->fl_pid = current->tgid;
4847 fl->fl_file = dp->dl_stid.sc_file->fi_deleg_file->nf_file;
4851 static struct nfs4_delegation *
4852 nfs4_set_delegation(struct nfs4_client *clp, struct svc_fh *fh,
4853 struct nfs4_file *fp, struct nfs4_clnt_odstate *odstate)
4856 struct nfs4_delegation *dp;
4857 struct nfsd_file *nf;
4858 struct file_lock *fl;
4861 * The fi_had_conflict and nfs_get_existing_delegation checks
4862 * here are just optimizations; we'll need to recheck them at
4865 if (fp->fi_had_conflict)
4866 return ERR_PTR(-EAGAIN);
4868 nf = find_readable_file(fp);
4870 /* We should always have a readable file here */
4872 return ERR_PTR(-EBADF);
4874 spin_lock(&state_lock);
4875 spin_lock(&fp->fi_lock);
4876 if (nfs4_delegation_exists(clp, fp))
4878 else if (!fp->fi_deleg_file) {
4879 fp->fi_deleg_file = nf;
4880 /* increment early to prevent fi_deleg_file from being
4882 fp->fi_delegees = 1;
4886 spin_unlock(&fp->fi_lock);
4887 spin_unlock(&state_lock);
4891 return ERR_PTR(status);
4894 dp = alloc_init_deleg(clp, fp, fh, odstate);
4898 fl = nfs4_alloc_init_lease(dp, NFS4_OPEN_DELEGATE_READ);
4900 goto out_clnt_odstate;
4902 status = vfs_setlease(fp->fi_deleg_file->nf_file, fl->fl_type, &fl, NULL);
4904 locks_free_lock(fl);
4906 goto out_clnt_odstate;
4908 spin_lock(&state_lock);
4909 spin_lock(&fp->fi_lock);
4910 if (fp->fi_had_conflict)
4913 status = hash_delegation_locked(dp, fp);
4914 spin_unlock(&fp->fi_lock);
4915 spin_unlock(&state_lock);
4922 vfs_setlease(fp->fi_deleg_file->nf_file, F_UNLCK, NULL, (void **)&dp);
4924 put_clnt_odstate(dp->dl_clnt_odstate);
4925 nfs4_put_stid(&dp->dl_stid);
4928 return ERR_PTR(status);
4931 static void nfsd4_open_deleg_none_ext(struct nfsd4_open *open, int status)
4933 open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE_EXT;
4934 if (status == -EAGAIN)
4935 open->op_why_no_deleg = WND4_CONTENTION;
4937 open->op_why_no_deleg = WND4_RESOURCE;
4938 switch (open->op_deleg_want) {
4939 case NFS4_SHARE_WANT_READ_DELEG:
4940 case NFS4_SHARE_WANT_WRITE_DELEG:
4941 case NFS4_SHARE_WANT_ANY_DELEG:
4943 case NFS4_SHARE_WANT_CANCEL:
4944 open->op_why_no_deleg = WND4_CANCELLED;
4946 case NFS4_SHARE_WANT_NO_DELEG:
4953 * Attempt to hand out a delegation.
4955 * Note we don't support write delegations, and won't until the vfs has
4956 * proper support for them.
4959 nfs4_open_delegation(struct svc_fh *fh, struct nfsd4_open *open,
4960 struct nfs4_ol_stateid *stp)
4962 struct nfs4_delegation *dp;
4963 struct nfs4_openowner *oo = openowner(stp->st_stateowner);
4964 struct nfs4_client *clp = stp->st_stid.sc_client;
4968 cb_up = nfsd4_cb_channel_good(oo->oo_owner.so_client);
4969 open->op_recall = 0;
4970 switch (open->op_claim_type) {
4971 case NFS4_OPEN_CLAIM_PREVIOUS:
4973 open->op_recall = 1;
4974 if (open->op_delegate_type != NFS4_OPEN_DELEGATE_READ)
4977 case NFS4_OPEN_CLAIM_NULL:
4978 case NFS4_OPEN_CLAIM_FH:
4980 * Let's not give out any delegations till everyone's
4981 * had the chance to reclaim theirs, *and* until
4982 * NLM locks have all been reclaimed:
4984 if (locks_in_grace(clp->net))
4986 if (!cb_up || !(oo->oo_flags & NFS4_OO_CONFIRMED))
4989 * Also, if the file was opened for write or
4990 * create, there's a good chance the client's
4991 * about to write to it, resulting in an
4992 * immediate recall (since we don't support
4993 * write delegations):
4995 if (open->op_share_access & NFS4_SHARE_ACCESS_WRITE)
4997 if (open->op_create == NFS4_OPEN_CREATE)
5003 dp = nfs4_set_delegation(clp, fh, stp->st_stid.sc_file, stp->st_clnt_odstate);
5007 memcpy(&open->op_delegate_stateid, &dp->dl_stid.sc_stateid, sizeof(dp->dl_stid.sc_stateid));
5009 dprintk("NFSD: delegation stateid=" STATEID_FMT "\n",
5010 STATEID_VAL(&dp->dl_stid.sc_stateid));
5011 open->op_delegate_type = NFS4_OPEN_DELEGATE_READ;
5012 nfs4_put_stid(&dp->dl_stid);
5015 open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE;
5016 if (open->op_claim_type == NFS4_OPEN_CLAIM_PREVIOUS &&
5017 open->op_delegate_type != NFS4_OPEN_DELEGATE_NONE) {
5018 dprintk("NFSD: WARNING: refusing delegation reclaim\n");
5019 open->op_recall = 1;
5022 /* 4.1 client asking for a delegation? */
5023 if (open->op_deleg_want)
5024 nfsd4_open_deleg_none_ext(open, status);
5028 static void nfsd4_deleg_xgrade_none_ext(struct nfsd4_open *open,
5029 struct nfs4_delegation *dp)
5031 if (open->op_deleg_want == NFS4_SHARE_WANT_READ_DELEG &&
5032 dp->dl_type == NFS4_OPEN_DELEGATE_WRITE) {
5033 open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE_EXT;
5034 open->op_why_no_deleg = WND4_NOT_SUPP_DOWNGRADE;
5035 } else if (open->op_deleg_want == NFS4_SHARE_WANT_WRITE_DELEG &&
5036 dp->dl_type == NFS4_OPEN_DELEGATE_WRITE) {
5037 open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE_EXT;
5038 open->op_why_no_deleg = WND4_NOT_SUPP_UPGRADE;
5040 /* Otherwise the client must be confused wanting a delegation
5041 * it already has, therefore we don't return
5042 * NFS4_OPEN_DELEGATE_NONE_EXT and reason.
5047 nfsd4_process_open2(struct svc_rqst *rqstp, struct svc_fh *current_fh, struct nfsd4_open *open)
5049 struct nfsd4_compoundres *resp = rqstp->rq_resp;
5050 struct nfs4_client *cl = open->op_openowner->oo_owner.so_client;
5051 struct nfs4_file *fp = NULL;
5052 struct nfs4_ol_stateid *stp = NULL;
5053 struct nfs4_delegation *dp = NULL;
5055 bool new_stp = false;
5058 * Lookup file; if found, lookup stateid and check open request,
5059 * and check for delegations in the process of being recalled.
5060 * If not found, create the nfs4_file struct
5062 fp = find_or_add_file(open->op_file, ¤t_fh->fh_handle);
5063 if (fp != open->op_file) {
5064 status = nfs4_check_deleg(cl, open, &dp);
5067 stp = nfsd4_find_and_lock_existing_open(fp, open);
5069 open->op_file = NULL;
5070 status = nfserr_bad_stateid;
5071 if (nfsd4_is_deleg_cur(open))
5076 stp = init_open_stateid(fp, open);
5082 * OPEN the file, or upgrade an existing OPEN.
5083 * If truncate fails, the OPEN fails.
5085 * stp is already locked.
5088 /* Stateid was found, this is an OPEN upgrade */
5089 status = nfs4_upgrade_open(rqstp, fp, current_fh, stp, open);
5091 mutex_unlock(&stp->st_mutex);
5095 status = nfs4_get_vfs_file(rqstp, fp, current_fh, stp, open);
5097 stp->st_stid.sc_type = NFS4_CLOSED_STID;
5098 release_open_stateid(stp);
5099 mutex_unlock(&stp->st_mutex);
5103 stp->st_clnt_odstate = find_or_hash_clnt_odstate(fp,
5105 if (stp->st_clnt_odstate == open->op_odstate)
5106 open->op_odstate = NULL;
5109 nfs4_inc_and_copy_stateid(&open->op_stateid, &stp->st_stid);
5110 mutex_unlock(&stp->st_mutex);
5112 if (nfsd4_has_session(&resp->cstate)) {
5113 if (open->op_deleg_want & NFS4_SHARE_WANT_NO_DELEG) {
5114 open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE_EXT;
5115 open->op_why_no_deleg = WND4_NOT_WANTED;
5121 * Attempt to hand out a delegation. No error return, because the
5122 * OPEN succeeds even if we fail.
5124 nfs4_open_delegation(current_fh, open, stp);
5128 dprintk("%s: stateid=" STATEID_FMT "\n", __func__,
5129 STATEID_VAL(&stp->st_stid.sc_stateid));
5131 /* 4.1 client trying to upgrade/downgrade delegation? */
5132 if (open->op_delegate_type == NFS4_OPEN_DELEGATE_NONE && dp &&
5133 open->op_deleg_want)
5134 nfsd4_deleg_xgrade_none_ext(open, dp);
5138 if (status == 0 && open->op_claim_type == NFS4_OPEN_CLAIM_PREVIOUS)
5139 open->op_openowner->oo_flags |= NFS4_OO_CONFIRMED;
5141 * To finish the open response, we just need to set the rflags.
5143 open->op_rflags = NFS4_OPEN_RESULT_LOCKTYPE_POSIX;
5144 if (nfsd4_has_session(&resp->cstate))
5145 open->op_rflags |= NFS4_OPEN_RESULT_MAY_NOTIFY_LOCK;
5146 else if (!(open->op_openowner->oo_flags & NFS4_OO_CONFIRMED))
5147 open->op_rflags |= NFS4_OPEN_RESULT_CONFIRM;
5150 nfs4_put_stid(&dp->dl_stid);
5152 nfs4_put_stid(&stp->st_stid);
5157 void nfsd4_cleanup_open_state(struct nfsd4_compound_state *cstate,
5158 struct nfsd4_open *open)
5160 if (open->op_openowner) {
5161 struct nfs4_stateowner *so = &open->op_openowner->oo_owner;
5163 nfsd4_cstate_assign_replay(cstate, so);
5164 nfs4_put_stateowner(so);
5167 kmem_cache_free(file_slab, open->op_file);
5169 nfs4_put_stid(&open->op_stp->st_stid);
5170 if (open->op_odstate)
5171 kmem_cache_free(odstate_slab, open->op_odstate);
5175 nfsd4_renew(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
5176 union nfsd4_op_u *u)
5178 clientid_t *clid = &u->renew;
5179 struct nfs4_client *clp;
5181 struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
5183 dprintk("process_renew(%08x/%08x): starting\n",
5184 clid->cl_boot, clid->cl_id);
5185 status = lookup_clientid(clid, cstate, nn, false);
5189 status = nfserr_cb_path_down;
5190 if (!list_empty(&clp->cl_delegations)
5191 && clp->cl_cb_state != NFSD4_CB_UP)
5199 nfsd4_end_grace(struct nfsd_net *nn)
5201 /* do nothing if grace period already ended */
5202 if (nn->grace_ended)
5205 nn->grace_ended = true;
5207 * If the server goes down again right now, an NFSv4
5208 * client will still be allowed to reclaim after it comes back up,
5209 * even if it hasn't yet had a chance to reclaim state this time.
5212 nfsd4_record_grace_done(nn);
5214 * At this point, NFSv4 clients can still reclaim. But if the
5215 * server crashes, any that have not yet reclaimed will be out
5216 * of luck on the next boot.
5218 * (NFSv4.1+ clients are considered to have reclaimed once they
5219 * call RECLAIM_COMPLETE. NFSv4.0 clients are considered to
5220 * have reclaimed after their first OPEN.)
5222 locks_end_grace(&nn->nfsd4_manager);
5224 * At this point, and once lockd and/or any other containers
5225 * exit their grace period, further reclaims will fail and
5226 * regular locking can resume.
5231 * If we've waited a lease period but there are still clients trying to
5232 * reclaim, wait a little longer to give them a chance to finish.
5234 static bool clients_still_reclaiming(struct nfsd_net *nn)
5236 time64_t double_grace_period_end = nn->boot_time +
5237 2 * nn->nfsd4_lease;
5239 if (nn->track_reclaim_completes &&
5240 atomic_read(&nn->nr_reclaim_complete) ==
5241 nn->reclaim_str_hashtbl_size)
5243 if (!nn->somebody_reclaimed)
5245 nn->somebody_reclaimed = false;
5247 * If we've given them *two* lease times to reclaim, and they're
5248 * still not done, give up:
5250 if (ktime_get_boottime_seconds() > double_grace_period_end)
5256 nfs4_laundromat(struct nfsd_net *nn)
5258 struct nfs4_client *clp;
5259 struct nfs4_openowner *oo;
5260 struct nfs4_delegation *dp;
5261 struct nfs4_ol_stateid *stp;
5262 struct nfsd4_blocked_lock *nbl;
5263 struct list_head *pos, *next, reaplist;
5264 time64_t cutoff = ktime_get_boottime_seconds() - nn->nfsd4_lease;
5265 time64_t t, new_timeo = nn->nfsd4_lease;
5266 struct nfs4_cpntf_state *cps;
5267 copy_stateid_t *cps_t;
5270 dprintk("NFSD: laundromat service - starting\n");
5272 if (clients_still_reclaiming(nn)) {
5276 dprintk("NFSD: end of grace period\n");
5277 nfsd4_end_grace(nn);
5278 INIT_LIST_HEAD(&reaplist);
5280 spin_lock(&nn->s2s_cp_lock);
5281 idr_for_each_entry(&nn->s2s_cp_stateids, cps_t, i) {
5282 cps = container_of(cps_t, struct nfs4_cpntf_state, cp_stateid);
5283 if (cps->cp_stateid.sc_type == NFS4_COPYNOTIFY_STID &&
5284 cps->cpntf_time > cutoff)
5285 _free_cpntf_state_locked(nn, cps);
5287 spin_unlock(&nn->s2s_cp_lock);
5289 spin_lock(&nn->client_lock);
5290 list_for_each_safe(pos, next, &nn->client_lru) {
5291 clp = list_entry(pos, struct nfs4_client, cl_lru);
5292 if (clp->cl_time > cutoff) {
5293 t = clp->cl_time - cutoff;
5294 new_timeo = min(new_timeo, t);
5297 if (mark_client_expired_locked(clp)) {
5298 dprintk("NFSD: client in use (clientid %08x)\n",
5299 clp->cl_clientid.cl_id);
5302 list_add(&clp->cl_lru, &reaplist);
5304 spin_unlock(&nn->client_lock);
5305 list_for_each_safe(pos, next, &reaplist) {
5306 clp = list_entry(pos, struct nfs4_client, cl_lru);
5307 dprintk("NFSD: purging unused client (clientid %08x)\n",
5308 clp->cl_clientid.cl_id);
5309 list_del_init(&clp->cl_lru);
5312 spin_lock(&state_lock);
5313 list_for_each_safe(pos, next, &nn->del_recall_lru) {
5314 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
5315 if (dp->dl_time > cutoff) {
5316 t = dp->dl_time - cutoff;
5317 new_timeo = min(new_timeo, t);
5320 WARN_ON(!unhash_delegation_locked(dp));
5321 list_add(&dp->dl_recall_lru, &reaplist);
5323 spin_unlock(&state_lock);
5324 while (!list_empty(&reaplist)) {
5325 dp = list_first_entry(&reaplist, struct nfs4_delegation,
5327 list_del_init(&dp->dl_recall_lru);
5328 revoke_delegation(dp);
5331 spin_lock(&nn->client_lock);
5332 while (!list_empty(&nn->close_lru)) {
5333 oo = list_first_entry(&nn->close_lru, struct nfs4_openowner,
5335 if (oo->oo_time > cutoff) {
5336 t = oo->oo_time - cutoff;
5337 new_timeo = min(new_timeo, t);
5340 list_del_init(&oo->oo_close_lru);
5341 stp = oo->oo_last_closed_stid;
5342 oo->oo_last_closed_stid = NULL;
5343 spin_unlock(&nn->client_lock);
5344 nfs4_put_stid(&stp->st_stid);
5345 spin_lock(&nn->client_lock);
5347 spin_unlock(&nn->client_lock);
5350 * It's possible for a client to try and acquire an already held lock
5351 * that is being held for a long time, and then lose interest in it.
5352 * So, we clean out any un-revisited request after a lease period
5353 * under the assumption that the client is no longer interested.
5355 * RFC5661, sec. 9.6 states that the client must not rely on getting
5356 * notifications and must continue to poll for locks, even when the
5357 * server supports them. Thus this shouldn't lead to clients blocking
5358 * indefinitely once the lock does become free.
5360 BUG_ON(!list_empty(&reaplist));
5361 spin_lock(&nn->blocked_locks_lock);
5362 while (!list_empty(&nn->blocked_locks_lru)) {
5363 nbl = list_first_entry(&nn->blocked_locks_lru,
5364 struct nfsd4_blocked_lock, nbl_lru);
5365 if (nbl->nbl_time > cutoff) {
5366 t = nbl->nbl_time - cutoff;
5367 new_timeo = min(new_timeo, t);
5370 list_move(&nbl->nbl_lru, &reaplist);
5371 list_del_init(&nbl->nbl_list);
5373 spin_unlock(&nn->blocked_locks_lock);
5375 while (!list_empty(&reaplist)) {
5376 nbl = list_first_entry(&reaplist,
5377 struct nfsd4_blocked_lock, nbl_lru);
5378 list_del_init(&nbl->nbl_lru);
5379 free_blocked_lock(nbl);
5382 new_timeo = max_t(time64_t, new_timeo, NFSD_LAUNDROMAT_MINTIMEOUT);
5386 static struct workqueue_struct *laundry_wq;
5387 static void laundromat_main(struct work_struct *);
5390 laundromat_main(struct work_struct *laundry)
5393 struct delayed_work *dwork = to_delayed_work(laundry);
5394 struct nfsd_net *nn = container_of(dwork, struct nfsd_net,
5397 t = nfs4_laundromat(nn);
5398 dprintk("NFSD: laundromat_main - sleeping for %lld seconds\n", t);
5399 queue_delayed_work(laundry_wq, &nn->laundromat_work, t*HZ);
5402 static inline __be32 nfs4_check_fh(struct svc_fh *fhp, struct nfs4_stid *stp)
5404 if (!fh_match(&fhp->fh_handle, &stp->sc_file->fi_fhandle))
5405 return nfserr_bad_stateid;
5410 access_permit_read(struct nfs4_ol_stateid *stp)
5412 return test_access(NFS4_SHARE_ACCESS_READ, stp) ||
5413 test_access(NFS4_SHARE_ACCESS_BOTH, stp) ||
5414 test_access(NFS4_SHARE_ACCESS_WRITE, stp);
5418 access_permit_write(struct nfs4_ol_stateid *stp)
5420 return test_access(NFS4_SHARE_ACCESS_WRITE, stp) ||
5421 test_access(NFS4_SHARE_ACCESS_BOTH, stp);
5425 __be32 nfs4_check_openmode(struct nfs4_ol_stateid *stp, int flags)
5427 __be32 status = nfserr_openmode;
5429 /* For lock stateid's, we test the parent open, not the lock: */
5430 if (stp->st_openstp)
5431 stp = stp->st_openstp;
5432 if ((flags & WR_STATE) && !access_permit_write(stp))
5434 if ((flags & RD_STATE) && !access_permit_read(stp))
5441 static inline __be32
5442 check_special_stateids(struct net *net, svc_fh *current_fh, stateid_t *stateid, int flags)
5444 if (ONE_STATEID(stateid) && (flags & RD_STATE))
5446 else if (opens_in_grace(net)) {
5447 /* Answer in remaining cases depends on existence of
5448 * conflicting state; so we must wait out the grace period. */
5449 return nfserr_grace;
5450 } else if (flags & WR_STATE)
5451 return nfs4_share_conflict(current_fh,
5452 NFS4_SHARE_DENY_WRITE);
5453 else /* (flags & RD_STATE) && ZERO_STATEID(stateid) */
5454 return nfs4_share_conflict(current_fh,
5455 NFS4_SHARE_DENY_READ);
5459 * Allow READ/WRITE during grace period on recovered state only for files
5460 * that are not able to provide mandatory locking.
5463 grace_disallows_io(struct net *net, struct inode *inode)
5465 return opens_in_grace(net) && mandatory_lock(inode);
5468 static __be32 check_stateid_generation(stateid_t *in, stateid_t *ref, bool has_session)
5471 * When sessions are used the stateid generation number is ignored
5474 if (has_session && in->si_generation == 0)
5477 if (in->si_generation == ref->si_generation)
5480 /* If the client sends us a stateid from the future, it's buggy: */
5481 if (nfsd4_stateid_generation_after(in, ref))
5482 return nfserr_bad_stateid;
5484 * However, we could see a stateid from the past, even from a
5485 * non-buggy client. For example, if the client sends a lock
5486 * while some IO is outstanding, the lock may bump si_generation
5487 * while the IO is still in flight. The client could avoid that
5488 * situation by waiting for responses on all the IO requests,
5489 * but better performance may result in retrying IO that
5490 * receives an old_stateid error if requests are rarely
5491 * reordered in flight:
5493 return nfserr_old_stateid;
5496 static __be32 nfsd4_stid_check_stateid_generation(stateid_t *in, struct nfs4_stid *s, bool has_session)
5500 spin_lock(&s->sc_lock);
5501 ret = nfsd4_verify_open_stid(s);
5503 ret = check_stateid_generation(in, &s->sc_stateid, has_session);
5504 spin_unlock(&s->sc_lock);
5508 static __be32 nfsd4_check_openowner_confirmed(struct nfs4_ol_stateid *ols)
5510 if (ols->st_stateowner->so_is_open_owner &&
5511 !(openowner(ols->st_stateowner)->oo_flags & NFS4_OO_CONFIRMED))
5512 return nfserr_bad_stateid;
5516 static __be32 nfsd4_validate_stateid(struct nfs4_client *cl, stateid_t *stateid)
5518 struct nfs4_stid *s;
5519 __be32 status = nfserr_bad_stateid;
5521 if (ZERO_STATEID(stateid) || ONE_STATEID(stateid) ||
5522 CLOSE_STATEID(stateid))
5524 /* Client debugging aid. */
5525 if (!same_clid(&stateid->si_opaque.so_clid, &cl->cl_clientid)) {
5526 char addr_str[INET6_ADDRSTRLEN];
5527 rpc_ntop((struct sockaddr *)&cl->cl_addr, addr_str,
5529 pr_warn_ratelimited("NFSD: client %s testing state ID "
5530 "with incorrect client ID\n", addr_str);
5533 spin_lock(&cl->cl_lock);
5534 s = find_stateid_locked(cl, stateid);
5537 status = nfsd4_stid_check_stateid_generation(stateid, s, 1);
5540 switch (s->sc_type) {
5541 case NFS4_DELEG_STID:
5544 case NFS4_REVOKED_DELEG_STID:
5545 status = nfserr_deleg_revoked;
5547 case NFS4_OPEN_STID:
5548 case NFS4_LOCK_STID:
5549 status = nfsd4_check_openowner_confirmed(openlockstateid(s));
5552 printk("unknown stateid type %x\n", s->sc_type);
5554 case NFS4_CLOSED_STID:
5555 case NFS4_CLOSED_DELEG_STID:
5556 status = nfserr_bad_stateid;
5559 spin_unlock(&cl->cl_lock);
5564 nfsd4_lookup_stateid(struct nfsd4_compound_state *cstate,
5565 stateid_t *stateid, unsigned char typemask,
5566 struct nfs4_stid **s, struct nfsd_net *nn)
5569 bool return_revoked = false;
5572 * only return revoked delegations if explicitly asked.
5573 * otherwise we report revoked or bad_stateid status.
5575 if (typemask & NFS4_REVOKED_DELEG_STID)
5576 return_revoked = true;
5577 else if (typemask & NFS4_DELEG_STID)
5578 typemask |= NFS4_REVOKED_DELEG_STID;
5580 if (ZERO_STATEID(stateid) || ONE_STATEID(stateid) ||
5581 CLOSE_STATEID(stateid))
5582 return nfserr_bad_stateid;
5583 status = lookup_clientid(&stateid->si_opaque.so_clid, cstate, nn,
5585 if (status == nfserr_stale_clientid) {
5586 if (cstate->session)
5587 return nfserr_bad_stateid;
5588 return nfserr_stale_stateid;
5592 *s = find_stateid_by_type(cstate->clp, stateid, typemask);
5594 return nfserr_bad_stateid;
5595 if (((*s)->sc_type == NFS4_REVOKED_DELEG_STID) && !return_revoked) {
5597 if (cstate->minorversion)
5598 return nfserr_deleg_revoked;
5599 return nfserr_bad_stateid;
5604 static struct nfsd_file *
5605 nfs4_find_file(struct nfs4_stid *s, int flags)
5610 switch (s->sc_type) {
5611 case NFS4_DELEG_STID:
5612 if (WARN_ON_ONCE(!s->sc_file->fi_deleg_file))
5614 return nfsd_file_get(s->sc_file->fi_deleg_file);
5615 case NFS4_OPEN_STID:
5616 case NFS4_LOCK_STID:
5617 if (flags & RD_STATE)
5618 return find_readable_file(s->sc_file);
5620 return find_writeable_file(s->sc_file);
5628 nfs4_check_olstateid(struct nfs4_ol_stateid *ols, int flags)
5632 status = nfsd4_check_openowner_confirmed(ols);
5635 return nfs4_check_openmode(ols, flags);
5639 nfs4_check_file(struct svc_rqst *rqstp, struct svc_fh *fhp, struct nfs4_stid *s,
5640 struct nfsd_file **nfp, int flags)
5642 int acc = (flags & RD_STATE) ? NFSD_MAY_READ : NFSD_MAY_WRITE;
5643 struct nfsd_file *nf;
5646 nf = nfs4_find_file(s, flags);
5648 status = nfsd_permission(rqstp, fhp->fh_export, fhp->fh_dentry,
5649 acc | NFSD_MAY_OWNER_OVERRIDE);
5655 status = nfsd_file_acquire(rqstp, fhp, acc, &nf);
5664 _free_cpntf_state_locked(struct nfsd_net *nn, struct nfs4_cpntf_state *cps)
5666 WARN_ON_ONCE(cps->cp_stateid.sc_type != NFS4_COPYNOTIFY_STID);
5667 if (!refcount_dec_and_test(&cps->cp_stateid.sc_count))
5669 list_del(&cps->cp_list);
5670 idr_remove(&nn->s2s_cp_stateids,
5671 cps->cp_stateid.stid.si_opaque.so_id);
5675 * A READ from an inter server to server COPY will have a
5676 * copy stateid. Look up the copy notify stateid from the
5677 * idr structure and take a reference on it.
5679 __be32 manage_cpntf_state(struct nfsd_net *nn, stateid_t *st,
5680 struct nfs4_client *clp,
5681 struct nfs4_cpntf_state **cps)
5683 copy_stateid_t *cps_t;
5684 struct nfs4_cpntf_state *state = NULL;
5686 if (st->si_opaque.so_clid.cl_id != nn->s2s_cp_cl_id)
5687 return nfserr_bad_stateid;
5688 spin_lock(&nn->s2s_cp_lock);
5689 cps_t = idr_find(&nn->s2s_cp_stateids, st->si_opaque.so_id);
5691 state = container_of(cps_t, struct nfs4_cpntf_state,
5693 if (state->cp_stateid.sc_type != NFS4_COPYNOTIFY_STID) {
5698 refcount_inc(&state->cp_stateid.sc_count);
5700 _free_cpntf_state_locked(nn, state);
5703 spin_unlock(&nn->s2s_cp_lock);
5705 return nfserr_bad_stateid;
5711 static __be32 find_cpntf_state(struct nfsd_net *nn, stateid_t *st,
5712 struct nfs4_stid **stid)
5715 struct nfs4_cpntf_state *cps = NULL;
5716 struct nfsd4_compound_state cstate;
5718 status = manage_cpntf_state(nn, st, NULL, &cps);
5722 cps->cpntf_time = ktime_get_boottime_seconds();
5723 memset(&cstate, 0, sizeof(cstate));
5724 status = lookup_clientid(&cps->cp_p_clid, &cstate, nn, true);
5727 status = nfsd4_lookup_stateid(&cstate, &cps->cp_p_stateid,
5728 NFS4_DELEG_STID|NFS4_OPEN_STID|NFS4_LOCK_STID,
5730 put_client_renew(cstate.clp);
5732 nfs4_put_cpntf_state(nn, cps);
5736 void nfs4_put_cpntf_state(struct nfsd_net *nn, struct nfs4_cpntf_state *cps)
5738 spin_lock(&nn->s2s_cp_lock);
5739 _free_cpntf_state_locked(nn, cps);
5740 spin_unlock(&nn->s2s_cp_lock);
5744 * Checks for stateid operations
5747 nfs4_preprocess_stateid_op(struct svc_rqst *rqstp,
5748 struct nfsd4_compound_state *cstate, struct svc_fh *fhp,
5749 stateid_t *stateid, int flags, struct nfsd_file **nfp,
5750 struct nfs4_stid **cstid)
5752 struct inode *ino = d_inode(fhp->fh_dentry);
5753 struct net *net = SVC_NET(rqstp);
5754 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
5755 struct nfs4_stid *s = NULL;
5761 if (grace_disallows_io(net, ino))
5762 return nfserr_grace;
5764 if (ZERO_STATEID(stateid) || ONE_STATEID(stateid)) {
5765 status = check_special_stateids(net, fhp, stateid, flags);
5769 status = nfsd4_lookup_stateid(cstate, stateid,
5770 NFS4_DELEG_STID|NFS4_OPEN_STID|NFS4_LOCK_STID,
5772 if (status == nfserr_bad_stateid)
5773 status = find_cpntf_state(nn, stateid, &s);
5776 status = nfsd4_stid_check_stateid_generation(stateid, s,
5777 nfsd4_has_session(cstate));
5781 switch (s->sc_type) {
5782 case NFS4_DELEG_STID:
5783 status = nfs4_check_delegmode(delegstateid(s), flags);
5785 case NFS4_OPEN_STID:
5786 case NFS4_LOCK_STID:
5787 status = nfs4_check_olstateid(openlockstateid(s), flags);
5790 status = nfserr_bad_stateid;
5795 status = nfs4_check_fh(fhp, s);
5798 if (status == nfs_ok && nfp)
5799 status = nfs4_check_file(rqstp, fhp, s, nfp, flags);
5802 if (!status && cstid)
5811 * Test if the stateid is valid
5814 nfsd4_test_stateid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
5815 union nfsd4_op_u *u)
5817 struct nfsd4_test_stateid *test_stateid = &u->test_stateid;
5818 struct nfsd4_test_stateid_id *stateid;
5819 struct nfs4_client *cl = cstate->session->se_client;
5821 list_for_each_entry(stateid, &test_stateid->ts_stateid_list, ts_id_list)
5822 stateid->ts_id_status =
5823 nfsd4_validate_stateid(cl, &stateid->ts_id_stateid);
5829 nfsd4_free_lock_stateid(stateid_t *stateid, struct nfs4_stid *s)
5831 struct nfs4_ol_stateid *stp = openlockstateid(s);
5834 ret = nfsd4_lock_ol_stateid(stp);
5838 ret = check_stateid_generation(stateid, &s->sc_stateid, 1);
5842 ret = nfserr_locks_held;
5843 if (check_for_locks(stp->st_stid.sc_file,
5844 lockowner(stp->st_stateowner)))
5847 release_lock_stateid(stp);
5851 mutex_unlock(&stp->st_mutex);
5858 nfsd4_free_stateid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
5859 union nfsd4_op_u *u)
5861 struct nfsd4_free_stateid *free_stateid = &u->free_stateid;
5862 stateid_t *stateid = &free_stateid->fr_stateid;
5863 struct nfs4_stid *s;
5864 struct nfs4_delegation *dp;
5865 struct nfs4_client *cl = cstate->session->se_client;
5866 __be32 ret = nfserr_bad_stateid;
5868 spin_lock(&cl->cl_lock);
5869 s = find_stateid_locked(cl, stateid);
5872 spin_lock(&s->sc_lock);
5873 switch (s->sc_type) {
5874 case NFS4_DELEG_STID:
5875 ret = nfserr_locks_held;
5877 case NFS4_OPEN_STID:
5878 ret = check_stateid_generation(stateid, &s->sc_stateid, 1);
5881 ret = nfserr_locks_held;
5883 case NFS4_LOCK_STID:
5884 spin_unlock(&s->sc_lock);
5885 refcount_inc(&s->sc_count);
5886 spin_unlock(&cl->cl_lock);
5887 ret = nfsd4_free_lock_stateid(stateid, s);
5889 case NFS4_REVOKED_DELEG_STID:
5890 spin_unlock(&s->sc_lock);
5891 dp = delegstateid(s);
5892 list_del_init(&dp->dl_recall_lru);
5893 spin_unlock(&cl->cl_lock);
5897 /* Default falls through and returns nfserr_bad_stateid */
5899 spin_unlock(&s->sc_lock);
5901 spin_unlock(&cl->cl_lock);
5909 return (type == NFS4_READW_LT || type == NFS4_READ_LT) ?
5910 RD_STATE : WR_STATE;
5913 static __be32 nfs4_seqid_op_checks(struct nfsd4_compound_state *cstate, stateid_t *stateid, u32 seqid, struct nfs4_ol_stateid *stp)
5915 struct svc_fh *current_fh = &cstate->current_fh;
5916 struct nfs4_stateowner *sop = stp->st_stateowner;
5919 status = nfsd4_check_seqid(cstate, sop, seqid);
5922 status = nfsd4_lock_ol_stateid(stp);
5923 if (status != nfs_ok)
5925 status = check_stateid_generation(stateid, &stp->st_stid.sc_stateid, nfsd4_has_session(cstate));
5926 if (status == nfs_ok)
5927 status = nfs4_check_fh(current_fh, &stp->st_stid);
5928 if (status != nfs_ok)
5929 mutex_unlock(&stp->st_mutex);
5934 * Checks for sequence id mutating operations.
5937 nfs4_preprocess_seqid_op(struct nfsd4_compound_state *cstate, u32 seqid,
5938 stateid_t *stateid, char typemask,
5939 struct nfs4_ol_stateid **stpp,
5940 struct nfsd_net *nn)
5943 struct nfs4_stid *s;
5944 struct nfs4_ol_stateid *stp = NULL;
5946 dprintk("NFSD: %s: seqid=%d stateid = " STATEID_FMT "\n", __func__,
5947 seqid, STATEID_VAL(stateid));
5950 status = nfsd4_lookup_stateid(cstate, stateid, typemask, &s, nn);
5953 stp = openlockstateid(s);
5954 nfsd4_cstate_assign_replay(cstate, stp->st_stateowner);
5956 status = nfs4_seqid_op_checks(cstate, stateid, seqid, stp);
5960 nfs4_put_stid(&stp->st_stid);
5964 static __be32 nfs4_preprocess_confirmed_seqid_op(struct nfsd4_compound_state *cstate, u32 seqid,
5965 stateid_t *stateid, struct nfs4_ol_stateid **stpp, struct nfsd_net *nn)
5968 struct nfs4_openowner *oo;
5969 struct nfs4_ol_stateid *stp;
5971 status = nfs4_preprocess_seqid_op(cstate, seqid, stateid,
5972 NFS4_OPEN_STID, &stp, nn);
5975 oo = openowner(stp->st_stateowner);
5976 if (!(oo->oo_flags & NFS4_OO_CONFIRMED)) {
5977 mutex_unlock(&stp->st_mutex);
5978 nfs4_put_stid(&stp->st_stid);
5979 return nfserr_bad_stateid;
5986 nfsd4_open_confirm(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
5987 union nfsd4_op_u *u)
5989 struct nfsd4_open_confirm *oc = &u->open_confirm;
5991 struct nfs4_openowner *oo;
5992 struct nfs4_ol_stateid *stp;
5993 struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
5995 dprintk("NFSD: nfsd4_open_confirm on file %pd\n",
5996 cstate->current_fh.fh_dentry);
5998 status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0);
6002 status = nfs4_preprocess_seqid_op(cstate,
6003 oc->oc_seqid, &oc->oc_req_stateid,
6004 NFS4_OPEN_STID, &stp, nn);
6007 oo = openowner(stp->st_stateowner);
6008 status = nfserr_bad_stateid;
6009 if (oo->oo_flags & NFS4_OO_CONFIRMED) {
6010 mutex_unlock(&stp->st_mutex);
6013 oo->oo_flags |= NFS4_OO_CONFIRMED;
6014 nfs4_inc_and_copy_stateid(&oc->oc_resp_stateid, &stp->st_stid);
6015 mutex_unlock(&stp->st_mutex);
6016 dprintk("NFSD: %s: success, seqid=%d stateid=" STATEID_FMT "\n",
6017 __func__, oc->oc_seqid, STATEID_VAL(&stp->st_stid.sc_stateid));
6019 nfsd4_client_record_create(oo->oo_owner.so_client);
6022 nfs4_put_stid(&stp->st_stid);
6024 nfsd4_bump_seqid(cstate, status);
6028 static inline void nfs4_stateid_downgrade_bit(struct nfs4_ol_stateid *stp, u32 access)
6030 if (!test_access(access, stp))
6032 nfs4_file_put_access(stp->st_stid.sc_file, access);
6033 clear_access(access, stp);
6036 static inline void nfs4_stateid_downgrade(struct nfs4_ol_stateid *stp, u32 to_access)
6038 switch (to_access) {
6039 case NFS4_SHARE_ACCESS_READ:
6040 nfs4_stateid_downgrade_bit(stp, NFS4_SHARE_ACCESS_WRITE);
6041 nfs4_stateid_downgrade_bit(stp, NFS4_SHARE_ACCESS_BOTH);
6043 case NFS4_SHARE_ACCESS_WRITE:
6044 nfs4_stateid_downgrade_bit(stp, NFS4_SHARE_ACCESS_READ);
6045 nfs4_stateid_downgrade_bit(stp, NFS4_SHARE_ACCESS_BOTH);
6047 case NFS4_SHARE_ACCESS_BOTH:
6055 nfsd4_open_downgrade(struct svc_rqst *rqstp,
6056 struct nfsd4_compound_state *cstate, union nfsd4_op_u *u)
6058 struct nfsd4_open_downgrade *od = &u->open_downgrade;
6060 struct nfs4_ol_stateid *stp;
6061 struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
6063 dprintk("NFSD: nfsd4_open_downgrade on file %pd\n",
6064 cstate->current_fh.fh_dentry);
6066 /* We don't yet support WANT bits: */
6067 if (od->od_deleg_want)
6068 dprintk("NFSD: %s: od_deleg_want=0x%x ignored\n", __func__,
6071 status = nfs4_preprocess_confirmed_seqid_op(cstate, od->od_seqid,
6072 &od->od_stateid, &stp, nn);
6075 status = nfserr_inval;
6076 if (!test_access(od->od_share_access, stp)) {
6077 dprintk("NFSD: access not a subset of current bitmap: 0x%hhx, input access=%08x\n",
6078 stp->st_access_bmap, od->od_share_access);
6081 if (!test_deny(od->od_share_deny, stp)) {
6082 dprintk("NFSD: deny not a subset of current bitmap: 0x%hhx, input deny=%08x\n",
6083 stp->st_deny_bmap, od->od_share_deny);
6086 nfs4_stateid_downgrade(stp, od->od_share_access);
6087 reset_union_bmap_deny(od->od_share_deny, stp);
6088 nfs4_inc_and_copy_stateid(&od->od_stateid, &stp->st_stid);
6091 mutex_unlock(&stp->st_mutex);
6092 nfs4_put_stid(&stp->st_stid);
6094 nfsd4_bump_seqid(cstate, status);
6098 static void nfsd4_close_open_stateid(struct nfs4_ol_stateid *s)
6100 struct nfs4_client *clp = s->st_stid.sc_client;
6102 LIST_HEAD(reaplist);
6104 spin_lock(&clp->cl_lock);
6105 unhashed = unhash_open_stateid(s, &reaplist);
6107 if (clp->cl_minorversion) {
6109 put_ol_stateid_locked(s, &reaplist);
6110 spin_unlock(&clp->cl_lock);
6111 free_ol_stateid_reaplist(&reaplist);
6113 spin_unlock(&clp->cl_lock);
6114 free_ol_stateid_reaplist(&reaplist);
6116 move_to_close_lru(s, clp->net);
6121 * nfs4_unlock_state() called after encode
6124 nfsd4_close(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
6125 union nfsd4_op_u *u)
6127 struct nfsd4_close *close = &u->close;
6129 struct nfs4_ol_stateid *stp;
6130 struct net *net = SVC_NET(rqstp);
6131 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
6133 dprintk("NFSD: nfsd4_close on file %pd\n",
6134 cstate->current_fh.fh_dentry);
6136 status = nfs4_preprocess_seqid_op(cstate, close->cl_seqid,
6138 NFS4_OPEN_STID|NFS4_CLOSED_STID,
6140 nfsd4_bump_seqid(cstate, status);
6144 stp->st_stid.sc_type = NFS4_CLOSED_STID;
6147 * Technically we don't _really_ have to increment or copy it, since
6148 * it should just be gone after this operation and we clobber the
6149 * copied value below, but we continue to do so here just to ensure
6150 * that racing ops see that there was a state change.
6152 nfs4_inc_and_copy_stateid(&close->cl_stateid, &stp->st_stid);
6154 nfsd4_close_open_stateid(stp);
6155 mutex_unlock(&stp->st_mutex);
6157 /* v4.1+ suggests that we send a special stateid in here, since the
6158 * clients should just ignore this anyway. Since this is not useful
6159 * for v4.0 clients either, we set it to the special close_stateid
6162 * See RFC5661 section 18.2.4, and RFC7530 section 16.2.5
6164 memcpy(&close->cl_stateid, &close_stateid, sizeof(close->cl_stateid));
6166 /* put reference from nfs4_preprocess_seqid_op */
6167 nfs4_put_stid(&stp->st_stid);
6173 nfsd4_delegreturn(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
6174 union nfsd4_op_u *u)
6176 struct nfsd4_delegreturn *dr = &u->delegreturn;
6177 struct nfs4_delegation *dp;
6178 stateid_t *stateid = &dr->dr_stateid;
6179 struct nfs4_stid *s;
6181 struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
6183 if ((status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0)))
6186 status = nfsd4_lookup_stateid(cstate, stateid, NFS4_DELEG_STID, &s, nn);
6189 dp = delegstateid(s);
6190 status = nfsd4_stid_check_stateid_generation(stateid, &dp->dl_stid, nfsd4_has_session(cstate));
6194 destroy_delegation(dp);
6196 nfs4_put_stid(&dp->dl_stid);
6202 end_offset(u64 start, u64 len)
6207 return end >= start ? end: NFS4_MAX_UINT64;
6210 /* last octet in a range */
6212 last_byte_offset(u64 start, u64 len)
6218 return end > start ? end - 1: NFS4_MAX_UINT64;
6222 * TODO: Linux file offsets are _signed_ 64-bit quantities, which means that
6223 * we can't properly handle lock requests that go beyond the (2^63 - 1)-th
6224 * byte, because of sign extension problems. Since NFSv4 calls for 64-bit
6225 * locking, this prevents us from being completely protocol-compliant. The
6226 * real solution to this problem is to start using unsigned file offsets in
6227 * the VFS, but this is a very deep change!
6230 nfs4_transform_lock_offset(struct file_lock *lock)
6232 if (lock->fl_start < 0)
6233 lock->fl_start = OFFSET_MAX;
6234 if (lock->fl_end < 0)
6235 lock->fl_end = OFFSET_MAX;
6239 nfsd4_fl_get_owner(fl_owner_t owner)
6241 struct nfs4_lockowner *lo = (struct nfs4_lockowner *)owner;
6243 nfs4_get_stateowner(&lo->lo_owner);
6248 nfsd4_fl_put_owner(fl_owner_t owner)
6250 struct nfs4_lockowner *lo = (struct nfs4_lockowner *)owner;
6253 nfs4_put_stateowner(&lo->lo_owner);
6257 nfsd4_lm_notify(struct file_lock *fl)
6259 struct nfs4_lockowner *lo = (struct nfs4_lockowner *)fl->fl_owner;
6260 struct net *net = lo->lo_owner.so_client->net;
6261 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
6262 struct nfsd4_blocked_lock *nbl = container_of(fl,
6263 struct nfsd4_blocked_lock, nbl_lock);
6266 /* An empty list means that something else is going to be using it */
6267 spin_lock(&nn->blocked_locks_lock);
6268 if (!list_empty(&nbl->nbl_list)) {
6269 list_del_init(&nbl->nbl_list);
6270 list_del_init(&nbl->nbl_lru);
6273 spin_unlock(&nn->blocked_locks_lock);
6276 nfsd4_run_cb(&nbl->nbl_cb);
6279 static const struct lock_manager_operations nfsd_posix_mng_ops = {
6280 .lm_notify = nfsd4_lm_notify,
6281 .lm_get_owner = nfsd4_fl_get_owner,
6282 .lm_put_owner = nfsd4_fl_put_owner,
6286 nfs4_set_lock_denied(struct file_lock *fl, struct nfsd4_lock_denied *deny)
6288 struct nfs4_lockowner *lo;
6290 if (fl->fl_lmops == &nfsd_posix_mng_ops) {
6291 lo = (struct nfs4_lockowner *) fl->fl_owner;
6292 xdr_netobj_dup(&deny->ld_owner, &lo->lo_owner.so_owner,
6294 if (!deny->ld_owner.data)
6295 /* We just don't care that much */
6297 deny->ld_clientid = lo->lo_owner.so_client->cl_clientid;
6300 deny->ld_owner.len = 0;
6301 deny->ld_owner.data = NULL;
6302 deny->ld_clientid.cl_boot = 0;
6303 deny->ld_clientid.cl_id = 0;
6305 deny->ld_start = fl->fl_start;
6306 deny->ld_length = NFS4_MAX_UINT64;
6307 if (fl->fl_end != NFS4_MAX_UINT64)
6308 deny->ld_length = fl->fl_end - fl->fl_start + 1;
6309 deny->ld_type = NFS4_READ_LT;
6310 if (fl->fl_type != F_RDLCK)
6311 deny->ld_type = NFS4_WRITE_LT;
6314 static struct nfs4_lockowner *
6315 find_lockowner_str_locked(struct nfs4_client *clp, struct xdr_netobj *owner)
6317 unsigned int strhashval = ownerstr_hashval(owner);
6318 struct nfs4_stateowner *so;
6320 lockdep_assert_held(&clp->cl_lock);
6322 list_for_each_entry(so, &clp->cl_ownerstr_hashtbl[strhashval],
6324 if (so->so_is_open_owner)
6326 if (same_owner_str(so, owner))
6327 return lockowner(nfs4_get_stateowner(so));
6332 static struct nfs4_lockowner *
6333 find_lockowner_str(struct nfs4_client *clp, struct xdr_netobj *owner)
6335 struct nfs4_lockowner *lo;
6337 spin_lock(&clp->cl_lock);
6338 lo = find_lockowner_str_locked(clp, owner);
6339 spin_unlock(&clp->cl_lock);
6343 static void nfs4_unhash_lockowner(struct nfs4_stateowner *sop)
6345 unhash_lockowner_locked(lockowner(sop));
6348 static void nfs4_free_lockowner(struct nfs4_stateowner *sop)
6350 struct nfs4_lockowner *lo = lockowner(sop);
6352 kmem_cache_free(lockowner_slab, lo);
6355 static const struct nfs4_stateowner_operations lockowner_ops = {
6356 .so_unhash = nfs4_unhash_lockowner,
6357 .so_free = nfs4_free_lockowner,
6361 * Alloc a lock owner structure.
6362 * Called in nfsd4_lock - therefore, OPEN and OPEN_CONFIRM (if needed) has
6365 * strhashval = ownerstr_hashval
6367 static struct nfs4_lockowner *
6368 alloc_init_lock_stateowner(unsigned int strhashval, struct nfs4_client *clp,
6369 struct nfs4_ol_stateid *open_stp,
6370 struct nfsd4_lock *lock)
6372 struct nfs4_lockowner *lo, *ret;
6374 lo = alloc_stateowner(lockowner_slab, &lock->lk_new_owner, clp);
6377 INIT_LIST_HEAD(&lo->lo_blocked);
6378 INIT_LIST_HEAD(&lo->lo_owner.so_stateids);
6379 lo->lo_owner.so_is_open_owner = 0;
6380 lo->lo_owner.so_seqid = lock->lk_new_lock_seqid;
6381 lo->lo_owner.so_ops = &lockowner_ops;
6382 spin_lock(&clp->cl_lock);
6383 ret = find_lockowner_str_locked(clp, &lock->lk_new_owner);
6385 list_add(&lo->lo_owner.so_strhash,
6386 &clp->cl_ownerstr_hashtbl[strhashval]);
6389 nfs4_free_stateowner(&lo->lo_owner);
6391 spin_unlock(&clp->cl_lock);
6395 static struct nfs4_ol_stateid *
6396 find_lock_stateid(struct nfs4_lockowner *lo, struct nfs4_file *fp)
6398 struct nfs4_ol_stateid *lst;
6399 struct nfs4_client *clp = lo->lo_owner.so_client;
6401 lockdep_assert_held(&clp->cl_lock);
6403 list_for_each_entry(lst, &lo->lo_owner.so_stateids, st_perstateowner) {
6404 if (lst->st_stid.sc_type != NFS4_LOCK_STID)
6406 if (lst->st_stid.sc_file == fp) {
6407 refcount_inc(&lst->st_stid.sc_count);
6414 static struct nfs4_ol_stateid *
6415 init_lock_stateid(struct nfs4_ol_stateid *stp, struct nfs4_lockowner *lo,
6416 struct nfs4_file *fp, struct inode *inode,
6417 struct nfs4_ol_stateid *open_stp)
6419 struct nfs4_client *clp = lo->lo_owner.so_client;
6420 struct nfs4_ol_stateid *retstp;
6422 mutex_init(&stp->st_mutex);
6423 mutex_lock_nested(&stp->st_mutex, OPEN_STATEID_MUTEX);
6425 spin_lock(&clp->cl_lock);
6426 spin_lock(&fp->fi_lock);
6427 retstp = find_lock_stateid(lo, fp);
6431 refcount_inc(&stp->st_stid.sc_count);
6432 stp->st_stid.sc_type = NFS4_LOCK_STID;
6433 stp->st_stateowner = nfs4_get_stateowner(&lo->lo_owner);
6435 stp->st_stid.sc_file = fp;
6436 stp->st_access_bmap = 0;
6437 stp->st_deny_bmap = open_stp->st_deny_bmap;
6438 stp->st_openstp = open_stp;
6439 list_add(&stp->st_locks, &open_stp->st_locks);
6440 list_add(&stp->st_perstateowner, &lo->lo_owner.so_stateids);
6441 list_add(&stp->st_perfile, &fp->fi_stateids);
6443 spin_unlock(&fp->fi_lock);
6444 spin_unlock(&clp->cl_lock);
6446 if (nfsd4_lock_ol_stateid(retstp) != nfs_ok) {
6447 nfs4_put_stid(&retstp->st_stid);
6450 /* To keep mutex tracking happy */
6451 mutex_unlock(&stp->st_mutex);
6457 static struct nfs4_ol_stateid *
6458 find_or_create_lock_stateid(struct nfs4_lockowner *lo, struct nfs4_file *fi,
6459 struct inode *inode, struct nfs4_ol_stateid *ost,
6462 struct nfs4_stid *ns = NULL;
6463 struct nfs4_ol_stateid *lst;
6464 struct nfs4_openowner *oo = openowner(ost->st_stateowner);
6465 struct nfs4_client *clp = oo->oo_owner.so_client;
6468 spin_lock(&clp->cl_lock);
6469 lst = find_lock_stateid(lo, fi);
6470 spin_unlock(&clp->cl_lock);
6472 if (nfsd4_lock_ol_stateid(lst) == nfs_ok)
6474 nfs4_put_stid(&lst->st_stid);
6476 ns = nfs4_alloc_stid(clp, stateid_slab, nfs4_free_lock_stateid);
6480 lst = init_lock_stateid(openlockstateid(ns), lo, fi, inode, ost);
6481 if (lst == openlockstateid(ns))
6490 check_lock_length(u64 offset, u64 length)
6492 return ((length == 0) || ((length != NFS4_MAX_UINT64) &&
6493 (length > ~offset)));
6496 static void get_lock_access(struct nfs4_ol_stateid *lock_stp, u32 access)
6498 struct nfs4_file *fp = lock_stp->st_stid.sc_file;
6500 lockdep_assert_held(&fp->fi_lock);
6502 if (test_access(access, lock_stp))
6504 __nfs4_file_get_access(fp, access);
6505 set_access(access, lock_stp);
6509 lookup_or_create_lock_state(struct nfsd4_compound_state *cstate,
6510 struct nfs4_ol_stateid *ost,
6511 struct nfsd4_lock *lock,
6512 struct nfs4_ol_stateid **plst, bool *new)
6515 struct nfs4_file *fi = ost->st_stid.sc_file;
6516 struct nfs4_openowner *oo = openowner(ost->st_stateowner);
6517 struct nfs4_client *cl = oo->oo_owner.so_client;
6518 struct inode *inode = d_inode(cstate->current_fh.fh_dentry);
6519 struct nfs4_lockowner *lo;
6520 struct nfs4_ol_stateid *lst;
6521 unsigned int strhashval;
6523 lo = find_lockowner_str(cl, &lock->lk_new_owner);
6525 strhashval = ownerstr_hashval(&lock->lk_new_owner);
6526 lo = alloc_init_lock_stateowner(strhashval, cl, ost, lock);
6528 return nfserr_jukebox;
6530 /* with an existing lockowner, seqids must be the same */
6531 status = nfserr_bad_seqid;
6532 if (!cstate->minorversion &&
6533 lock->lk_new_lock_seqid != lo->lo_owner.so_seqid)
6537 lst = find_or_create_lock_stateid(lo, fi, inode, ost, new);
6539 status = nfserr_jukebox;
6546 nfs4_put_stateowner(&lo->lo_owner);
6554 nfsd4_lock(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
6555 union nfsd4_op_u *u)
6557 struct nfsd4_lock *lock = &u->lock;
6558 struct nfs4_openowner *open_sop = NULL;
6559 struct nfs4_lockowner *lock_sop = NULL;
6560 struct nfs4_ol_stateid *lock_stp = NULL;
6561 struct nfs4_ol_stateid *open_stp = NULL;
6562 struct nfs4_file *fp;
6563 struct nfsd_file *nf = NULL;
6564 struct nfsd4_blocked_lock *nbl = NULL;
6565 struct file_lock *file_lock = NULL;
6566 struct file_lock *conflock = NULL;
6571 unsigned char fl_type;
6572 unsigned int fl_flags = FL_POSIX;
6573 struct net *net = SVC_NET(rqstp);
6574 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
6576 dprintk("NFSD: nfsd4_lock: start=%Ld length=%Ld\n",
6577 (long long) lock->lk_offset,
6578 (long long) lock->lk_length);
6580 if (check_lock_length(lock->lk_offset, lock->lk_length))
6581 return nfserr_inval;
6583 if ((status = fh_verify(rqstp, &cstate->current_fh,
6584 S_IFREG, NFSD_MAY_LOCK))) {
6585 dprintk("NFSD: nfsd4_lock: permission denied!\n");
6589 if (lock->lk_is_new) {
6590 if (nfsd4_has_session(cstate))
6591 /* See rfc 5661 18.10.3: given clientid is ignored: */
6592 memcpy(&lock->lk_new_clientid,
6593 &cstate->session->se_client->cl_clientid,
6594 sizeof(clientid_t));
6596 status = nfserr_stale_clientid;
6597 if (STALE_CLIENTID(&lock->lk_new_clientid, nn))
6600 /* validate and update open stateid and open seqid */
6601 status = nfs4_preprocess_confirmed_seqid_op(cstate,
6602 lock->lk_new_open_seqid,
6603 &lock->lk_new_open_stateid,
6607 mutex_unlock(&open_stp->st_mutex);
6608 open_sop = openowner(open_stp->st_stateowner);
6609 status = nfserr_bad_stateid;
6610 if (!same_clid(&open_sop->oo_owner.so_client->cl_clientid,
6611 &lock->lk_new_clientid))
6613 status = lookup_or_create_lock_state(cstate, open_stp, lock,
6616 status = nfs4_preprocess_seqid_op(cstate,
6617 lock->lk_old_lock_seqid,
6618 &lock->lk_old_lock_stateid,
6619 NFS4_LOCK_STID, &lock_stp, nn);
6623 lock_sop = lockowner(lock_stp->st_stateowner);
6625 lkflg = setlkflg(lock->lk_type);
6626 status = nfs4_check_openmode(lock_stp, lkflg);
6630 status = nfserr_grace;
6631 if (locks_in_grace(net) && !lock->lk_reclaim)
6633 status = nfserr_no_grace;
6634 if (!locks_in_grace(net) && lock->lk_reclaim)
6637 fp = lock_stp->st_stid.sc_file;
6638 switch (lock->lk_type) {
6640 if (nfsd4_has_session(cstate))
6641 fl_flags |= FL_SLEEP;
6644 spin_lock(&fp->fi_lock);
6645 nf = find_readable_file_locked(fp);
6647 get_lock_access(lock_stp, NFS4_SHARE_ACCESS_READ);
6648 spin_unlock(&fp->fi_lock);
6651 case NFS4_WRITEW_LT:
6652 if (nfsd4_has_session(cstate))
6653 fl_flags |= FL_SLEEP;
6656 spin_lock(&fp->fi_lock);
6657 nf = find_writeable_file_locked(fp);
6659 get_lock_access(lock_stp, NFS4_SHARE_ACCESS_WRITE);
6660 spin_unlock(&fp->fi_lock);
6664 status = nfserr_inval;
6669 status = nfserr_openmode;
6673 nbl = find_or_allocate_block(lock_sop, &fp->fi_fhandle, nn);
6675 dprintk("NFSD: %s: unable to allocate block!\n", __func__);
6676 status = nfserr_jukebox;
6680 file_lock = &nbl->nbl_lock;
6681 file_lock->fl_type = fl_type;
6682 file_lock->fl_owner = (fl_owner_t)lockowner(nfs4_get_stateowner(&lock_sop->lo_owner));
6683 file_lock->fl_pid = current->tgid;
6684 file_lock->fl_file = nf->nf_file;
6685 file_lock->fl_flags = fl_flags;
6686 file_lock->fl_lmops = &nfsd_posix_mng_ops;
6687 file_lock->fl_start = lock->lk_offset;
6688 file_lock->fl_end = last_byte_offset(lock->lk_offset, lock->lk_length);
6689 nfs4_transform_lock_offset(file_lock);
6691 conflock = locks_alloc_lock();
6693 dprintk("NFSD: %s: unable to allocate lock!\n", __func__);
6694 status = nfserr_jukebox;
6698 if (fl_flags & FL_SLEEP) {
6699 nbl->nbl_time = ktime_get_boottime_seconds();
6700 spin_lock(&nn->blocked_locks_lock);
6701 list_add_tail(&nbl->nbl_list, &lock_sop->lo_blocked);
6702 list_add_tail(&nbl->nbl_lru, &nn->blocked_locks_lru);
6703 spin_unlock(&nn->blocked_locks_lock);
6706 err = vfs_lock_file(nf->nf_file, F_SETLK, file_lock, conflock);
6708 case 0: /* success! */
6709 nfs4_inc_and_copy_stateid(&lock->lk_resp_stateid, &lock_stp->st_stid);
6711 if (lock->lk_reclaim)
6712 nn->somebody_reclaimed = true;
6714 case FILE_LOCK_DEFERRED:
6717 case -EAGAIN: /* conflock holds conflicting lock */
6718 status = nfserr_denied;
6719 dprintk("NFSD: nfsd4_lock: conflicting lock found!\n");
6720 nfs4_set_lock_denied(conflock, &lock->lk_denied);
6723 status = nfserr_deadlock;
6726 dprintk("NFSD: nfsd4_lock: vfs_lock_file() failed! status %d\n",err);
6727 status = nfserrno(err);
6732 /* dequeue it if we queued it before */
6733 if (fl_flags & FL_SLEEP) {
6734 spin_lock(&nn->blocked_locks_lock);
6735 list_del_init(&nbl->nbl_list);
6736 list_del_init(&nbl->nbl_lru);
6737 spin_unlock(&nn->blocked_locks_lock);
6739 free_blocked_lock(nbl);
6744 /* Bump seqid manually if the 4.0 replay owner is openowner */
6745 if (cstate->replay_owner &&
6746 cstate->replay_owner != &lock_sop->lo_owner &&
6747 seqid_mutating_err(ntohl(status)))
6748 lock_sop->lo_owner.so_seqid++;
6751 * If this is a new, never-before-used stateid, and we are
6752 * returning an error, then just go ahead and release it.
6755 release_lock_stateid(lock_stp);
6757 mutex_unlock(&lock_stp->st_mutex);
6759 nfs4_put_stid(&lock_stp->st_stid);
6762 nfs4_put_stid(&open_stp->st_stid);
6763 nfsd4_bump_seqid(cstate, status);
6765 locks_free_lock(conflock);
6770 * The NFSv4 spec allows a client to do a LOCKT without holding an OPEN,
6771 * so we do a temporary open here just to get an open file to pass to
6772 * vfs_test_lock. (Arguably perhaps test_lock should be done with an
6775 static __be32 nfsd_test_lock(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file_lock *lock)
6777 struct nfsd_file *nf;
6778 __be32 err = nfsd_file_acquire(rqstp, fhp, NFSD_MAY_READ, &nf);
6780 err = nfserrno(vfs_test_lock(nf->nf_file, lock));
6790 nfsd4_lockt(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
6791 union nfsd4_op_u *u)
6793 struct nfsd4_lockt *lockt = &u->lockt;
6794 struct file_lock *file_lock = NULL;
6795 struct nfs4_lockowner *lo = NULL;
6797 struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
6799 if (locks_in_grace(SVC_NET(rqstp)))
6800 return nfserr_grace;
6802 if (check_lock_length(lockt->lt_offset, lockt->lt_length))
6803 return nfserr_inval;
6805 if (!nfsd4_has_session(cstate)) {
6806 status = lookup_clientid(&lockt->lt_clientid, cstate, nn,
6812 if ((status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0)))
6815 file_lock = locks_alloc_lock();
6817 dprintk("NFSD: %s: unable to allocate lock!\n", __func__);
6818 status = nfserr_jukebox;
6822 switch (lockt->lt_type) {
6825 file_lock->fl_type = F_RDLCK;
6828 case NFS4_WRITEW_LT:
6829 file_lock->fl_type = F_WRLCK;
6832 dprintk("NFSD: nfs4_lockt: bad lock type!\n");
6833 status = nfserr_inval;
6837 lo = find_lockowner_str(cstate->clp, &lockt->lt_owner);
6839 file_lock->fl_owner = (fl_owner_t)lo;
6840 file_lock->fl_pid = current->tgid;
6841 file_lock->fl_flags = FL_POSIX;
6843 file_lock->fl_start = lockt->lt_offset;
6844 file_lock->fl_end = last_byte_offset(lockt->lt_offset, lockt->lt_length);
6846 nfs4_transform_lock_offset(file_lock);
6848 status = nfsd_test_lock(rqstp, &cstate->current_fh, file_lock);
6852 if (file_lock->fl_type != F_UNLCK) {
6853 status = nfserr_denied;
6854 nfs4_set_lock_denied(file_lock, &lockt->lt_denied);
6858 nfs4_put_stateowner(&lo->lo_owner);
6860 locks_free_lock(file_lock);
6865 nfsd4_locku(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
6866 union nfsd4_op_u *u)
6868 struct nfsd4_locku *locku = &u->locku;
6869 struct nfs4_ol_stateid *stp;
6870 struct nfsd_file *nf = NULL;
6871 struct file_lock *file_lock = NULL;
6874 struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
6876 dprintk("NFSD: nfsd4_locku: start=%Ld length=%Ld\n",
6877 (long long) locku->lu_offset,
6878 (long long) locku->lu_length);
6880 if (check_lock_length(locku->lu_offset, locku->lu_length))
6881 return nfserr_inval;
6883 status = nfs4_preprocess_seqid_op(cstate, locku->lu_seqid,
6884 &locku->lu_stateid, NFS4_LOCK_STID,
6888 nf = find_any_file(stp->st_stid.sc_file);
6890 status = nfserr_lock_range;
6893 file_lock = locks_alloc_lock();
6895 dprintk("NFSD: %s: unable to allocate lock!\n", __func__);
6896 status = nfserr_jukebox;
6900 file_lock->fl_type = F_UNLCK;
6901 file_lock->fl_owner = (fl_owner_t)lockowner(nfs4_get_stateowner(stp->st_stateowner));
6902 file_lock->fl_pid = current->tgid;
6903 file_lock->fl_file = nf->nf_file;
6904 file_lock->fl_flags = FL_POSIX;
6905 file_lock->fl_lmops = &nfsd_posix_mng_ops;
6906 file_lock->fl_start = locku->lu_offset;
6908 file_lock->fl_end = last_byte_offset(locku->lu_offset,
6910 nfs4_transform_lock_offset(file_lock);
6912 err = vfs_lock_file(nf->nf_file, F_SETLK, file_lock, NULL);
6914 dprintk("NFSD: nfs4_locku: vfs_lock_file failed!\n");
6917 nfs4_inc_and_copy_stateid(&locku->lu_stateid, &stp->st_stid);
6921 mutex_unlock(&stp->st_mutex);
6922 nfs4_put_stid(&stp->st_stid);
6924 nfsd4_bump_seqid(cstate, status);
6926 locks_free_lock(file_lock);
6930 status = nfserrno(err);
6936 * true: locks held by lockowner
6937 * false: no locks held by lockowner
6940 check_for_locks(struct nfs4_file *fp, struct nfs4_lockowner *lowner)
6942 struct file_lock *fl;
6944 struct nfsd_file *nf = find_any_file(fp);
6945 struct inode *inode;
6946 struct file_lock_context *flctx;
6949 /* Any valid lock stateid should have some sort of access */
6954 inode = locks_inode(nf->nf_file);
6955 flctx = inode->i_flctx;
6957 if (flctx && !list_empty_careful(&flctx->flc_posix)) {
6958 spin_lock(&flctx->flc_lock);
6959 list_for_each_entry(fl, &flctx->flc_posix, fl_list) {
6960 if (fl->fl_owner == (fl_owner_t)lowner) {
6965 spin_unlock(&flctx->flc_lock);
6972 nfsd4_release_lockowner(struct svc_rqst *rqstp,
6973 struct nfsd4_compound_state *cstate,
6974 union nfsd4_op_u *u)
6976 struct nfsd4_release_lockowner *rlockowner = &u->release_lockowner;
6977 clientid_t *clid = &rlockowner->rl_clientid;
6978 struct nfs4_stateowner *sop;
6979 struct nfs4_lockowner *lo = NULL;
6980 struct nfs4_ol_stateid *stp;
6981 struct xdr_netobj *owner = &rlockowner->rl_owner;
6982 unsigned int hashval = ownerstr_hashval(owner);
6984 struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
6985 struct nfs4_client *clp;
6986 LIST_HEAD (reaplist);
6988 dprintk("nfsd4_release_lockowner clientid: (%08x/%08x):\n",
6989 clid->cl_boot, clid->cl_id);
6991 status = lookup_clientid(clid, cstate, nn, false);
6996 /* Find the matching lock stateowner */
6997 spin_lock(&clp->cl_lock);
6998 list_for_each_entry(sop, &clp->cl_ownerstr_hashtbl[hashval],
7001 if (sop->so_is_open_owner || !same_owner_str(sop, owner))
7004 /* see if there are still any locks associated with it */
7005 lo = lockowner(sop);
7006 list_for_each_entry(stp, &sop->so_stateids, st_perstateowner) {
7007 if (check_for_locks(stp->st_stid.sc_file, lo)) {
7008 status = nfserr_locks_held;
7009 spin_unlock(&clp->cl_lock);
7014 nfs4_get_stateowner(sop);
7018 spin_unlock(&clp->cl_lock);
7022 unhash_lockowner_locked(lo);
7023 while (!list_empty(&lo->lo_owner.so_stateids)) {
7024 stp = list_first_entry(&lo->lo_owner.so_stateids,
7025 struct nfs4_ol_stateid,
7027 WARN_ON(!unhash_lock_stateid(stp));
7028 put_ol_stateid_locked(stp, &reaplist);
7030 spin_unlock(&clp->cl_lock);
7031 free_ol_stateid_reaplist(&reaplist);
7032 remove_blocked_locks(lo);
7033 nfs4_put_stateowner(&lo->lo_owner);
7038 static inline struct nfs4_client_reclaim *
7041 return kmalloc(sizeof(struct nfs4_client_reclaim), GFP_KERNEL);
7045 nfs4_has_reclaimed_state(struct xdr_netobj name, struct nfsd_net *nn)
7047 struct nfs4_client_reclaim *crp;
7049 crp = nfsd4_find_reclaim_client(name, nn);
7050 return (crp && crp->cr_clp);
7054 * failure => all reset bets are off, nfserr_no_grace...
7056 * The caller is responsible for freeing name.data if NULL is returned (it
7057 * will be freed in nfs4_remove_reclaim_record in the normal case).
7059 struct nfs4_client_reclaim *
7060 nfs4_client_to_reclaim(struct xdr_netobj name, struct xdr_netobj princhash,
7061 struct nfsd_net *nn)
7063 unsigned int strhashval;
7064 struct nfs4_client_reclaim *crp;
7066 dprintk("NFSD nfs4_client_to_reclaim NAME: %.*s\n", name.len, name.data);
7067 crp = alloc_reclaim();
7069 strhashval = clientstr_hashval(name);
7070 INIT_LIST_HEAD(&crp->cr_strhash);
7071 list_add(&crp->cr_strhash, &nn->reclaim_str_hashtbl[strhashval]);
7072 crp->cr_name.data = name.data;
7073 crp->cr_name.len = name.len;
7074 crp->cr_princhash.data = princhash.data;
7075 crp->cr_princhash.len = princhash.len;
7077 nn->reclaim_str_hashtbl_size++;
7083 nfs4_remove_reclaim_record(struct nfs4_client_reclaim *crp, struct nfsd_net *nn)
7085 list_del(&crp->cr_strhash);
7086 kfree(crp->cr_name.data);
7087 kfree(crp->cr_princhash.data);
7089 nn->reclaim_str_hashtbl_size--;
7093 nfs4_release_reclaim(struct nfsd_net *nn)
7095 struct nfs4_client_reclaim *crp = NULL;
7098 for (i = 0; i < CLIENT_HASH_SIZE; i++) {
7099 while (!list_empty(&nn->reclaim_str_hashtbl[i])) {
7100 crp = list_entry(nn->reclaim_str_hashtbl[i].next,
7101 struct nfs4_client_reclaim, cr_strhash);
7102 nfs4_remove_reclaim_record(crp, nn);
7105 WARN_ON_ONCE(nn->reclaim_str_hashtbl_size);
7109 * called from OPEN, CLAIM_PREVIOUS with a new clientid. */
7110 struct nfs4_client_reclaim *
7111 nfsd4_find_reclaim_client(struct xdr_netobj name, struct nfsd_net *nn)
7113 unsigned int strhashval;
7114 struct nfs4_client_reclaim *crp = NULL;
7116 dprintk("NFSD: nfs4_find_reclaim_client for name %.*s\n", name.len, name.data);
7118 strhashval = clientstr_hashval(name);
7119 list_for_each_entry(crp, &nn->reclaim_str_hashtbl[strhashval], cr_strhash) {
7120 if (compare_blob(&crp->cr_name, &name) == 0) {
7128 * Called from OPEN. Look for clientid in reclaim list.
7131 nfs4_check_open_reclaim(clientid_t *clid,
7132 struct nfsd4_compound_state *cstate,
7133 struct nfsd_net *nn)
7137 /* find clientid in conf_id_hashtbl */
7138 status = lookup_clientid(clid, cstate, nn, false);
7140 return nfserr_reclaim_bad;
7142 if (test_bit(NFSD4_CLIENT_RECLAIM_COMPLETE, &cstate->clp->cl_flags))
7143 return nfserr_no_grace;
7145 if (nfsd4_client_record_check(cstate->clp))
7146 return nfserr_reclaim_bad;
7151 #ifdef CONFIG_NFSD_FAULT_INJECTION
7153 put_client(struct nfs4_client *clp)
7155 atomic_dec(&clp->cl_rpc_users);
7158 static struct nfs4_client *
7159 nfsd_find_client(struct sockaddr_storage *addr, size_t addr_size)
7161 struct nfs4_client *clp;
7162 struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
7165 if (!nfsd_netns_ready(nn))
7168 list_for_each_entry(clp, &nn->client_lru, cl_lru) {
7169 if (memcmp(&clp->cl_addr, addr, addr_size) == 0)
7176 nfsd_inject_print_clients(void)
7178 struct nfs4_client *clp;
7180 struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
7182 char buf[INET6_ADDRSTRLEN];
7184 if (!nfsd_netns_ready(nn))
7187 spin_lock(&nn->client_lock);
7188 list_for_each_entry(clp, &nn->client_lru, cl_lru) {
7189 rpc_ntop((struct sockaddr *)&clp->cl_addr, buf, sizeof(buf));
7190 pr_info("NFS Client: %s\n", buf);
7193 spin_unlock(&nn->client_lock);
7199 nfsd_inject_forget_client(struct sockaddr_storage *addr, size_t addr_size)
7202 struct nfs4_client *clp;
7203 struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
7206 if (!nfsd_netns_ready(nn))
7209 spin_lock(&nn->client_lock);
7210 clp = nfsd_find_client(addr, addr_size);
7212 if (mark_client_expired_locked(clp) == nfs_ok)
7217 spin_unlock(&nn->client_lock);
7226 nfsd_inject_forget_clients(u64 max)
7229 struct nfs4_client *clp, *next;
7230 struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
7232 LIST_HEAD(reaplist);
7234 if (!nfsd_netns_ready(nn))
7237 spin_lock(&nn->client_lock);
7238 list_for_each_entry_safe(clp, next, &nn->client_lru, cl_lru) {
7239 if (mark_client_expired_locked(clp) == nfs_ok) {
7240 list_add(&clp->cl_lru, &reaplist);
7241 if (max != 0 && ++count >= max)
7245 spin_unlock(&nn->client_lock);
7247 list_for_each_entry_safe(clp, next, &reaplist, cl_lru)
7253 static void nfsd_print_count(struct nfs4_client *clp, unsigned int count,
7256 char buf[INET6_ADDRSTRLEN];
7257 rpc_ntop((struct sockaddr *)&clp->cl_addr, buf, sizeof(buf));
7258 printk(KERN_INFO "NFS Client: %s has %u %s\n", buf, count, type);
7262 nfsd_inject_add_lock_to_list(struct nfs4_ol_stateid *lst,
7263 struct list_head *collect)
7265 struct nfs4_client *clp = lst->st_stid.sc_client;
7266 struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
7272 lockdep_assert_held(&nn->client_lock);
7273 atomic_inc(&clp->cl_rpc_users);
7274 list_add(&lst->st_locks, collect);
7277 static u64 nfsd_foreach_client_lock(struct nfs4_client *clp, u64 max,
7278 struct list_head *collect,
7279 bool (*func)(struct nfs4_ol_stateid *))
7281 struct nfs4_openowner *oop;
7282 struct nfs4_ol_stateid *stp, *st_next;
7283 struct nfs4_ol_stateid *lst, *lst_next;
7286 spin_lock(&clp->cl_lock);
7287 list_for_each_entry(oop, &clp->cl_openowners, oo_perclient) {
7288 list_for_each_entry_safe(stp, st_next,
7289 &oop->oo_owner.so_stateids, st_perstateowner) {
7290 list_for_each_entry_safe(lst, lst_next,
7291 &stp->st_locks, st_locks) {
7294 nfsd_inject_add_lock_to_list(lst,
7299 * Despite the fact that these functions deal
7300 * with 64-bit integers for "count", we must
7301 * ensure that it doesn't blow up the
7302 * clp->cl_rpc_users. Throw a warning if we
7303 * start to approach INT_MAX here.
7305 WARN_ON_ONCE(count == (INT_MAX / 2));
7312 spin_unlock(&clp->cl_lock);
7318 nfsd_collect_client_locks(struct nfs4_client *clp, struct list_head *collect,
7321 return nfsd_foreach_client_lock(clp, max, collect, unhash_lock_stateid);
7325 nfsd_print_client_locks(struct nfs4_client *clp)
7327 u64 count = nfsd_foreach_client_lock(clp, 0, NULL, NULL);
7328 nfsd_print_count(clp, count, "locked files");
7333 nfsd_inject_print_locks(void)
7335 struct nfs4_client *clp;
7337 struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
7340 if (!nfsd_netns_ready(nn))
7343 spin_lock(&nn->client_lock);
7344 list_for_each_entry(clp, &nn->client_lru, cl_lru)
7345 count += nfsd_print_client_locks(clp);
7346 spin_unlock(&nn->client_lock);
7352 nfsd_reap_locks(struct list_head *reaplist)
7354 struct nfs4_client *clp;
7355 struct nfs4_ol_stateid *stp, *next;
7357 list_for_each_entry_safe(stp, next, reaplist, st_locks) {
7358 list_del_init(&stp->st_locks);
7359 clp = stp->st_stid.sc_client;
7360 nfs4_put_stid(&stp->st_stid);
7366 nfsd_inject_forget_client_locks(struct sockaddr_storage *addr, size_t addr_size)
7368 unsigned int count = 0;
7369 struct nfs4_client *clp;
7370 struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
7372 LIST_HEAD(reaplist);
7374 if (!nfsd_netns_ready(nn))
7377 spin_lock(&nn->client_lock);
7378 clp = nfsd_find_client(addr, addr_size);
7380 count = nfsd_collect_client_locks(clp, &reaplist, 0);
7381 spin_unlock(&nn->client_lock);
7382 nfsd_reap_locks(&reaplist);
7387 nfsd_inject_forget_locks(u64 max)
7390 struct nfs4_client *clp;
7391 struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
7393 LIST_HEAD(reaplist);
7395 if (!nfsd_netns_ready(nn))
7398 spin_lock(&nn->client_lock);
7399 list_for_each_entry(clp, &nn->client_lru, cl_lru) {
7400 count += nfsd_collect_client_locks(clp, &reaplist, max - count);
7401 if (max != 0 && count >= max)
7404 spin_unlock(&nn->client_lock);
7405 nfsd_reap_locks(&reaplist);
7410 nfsd_foreach_client_openowner(struct nfs4_client *clp, u64 max,
7411 struct list_head *collect,
7412 void (*func)(struct nfs4_openowner *))
7414 struct nfs4_openowner *oop, *next;
7415 struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
7419 lockdep_assert_held(&nn->client_lock);
7421 spin_lock(&clp->cl_lock);
7422 list_for_each_entry_safe(oop, next, &clp->cl_openowners, oo_perclient) {
7426 atomic_inc(&clp->cl_rpc_users);
7427 list_add(&oop->oo_perclient, collect);
7432 * Despite the fact that these functions deal with
7433 * 64-bit integers for "count", we must ensure that
7434 * it doesn't blow up the clp->cl_rpc_users. Throw a
7435 * warning if we start to approach INT_MAX here.
7437 WARN_ON_ONCE(count == (INT_MAX / 2));
7441 spin_unlock(&clp->cl_lock);
7447 nfsd_print_client_openowners(struct nfs4_client *clp)
7449 u64 count = nfsd_foreach_client_openowner(clp, 0, NULL, NULL);
7451 nfsd_print_count(clp, count, "openowners");
7456 nfsd_collect_client_openowners(struct nfs4_client *clp,
7457 struct list_head *collect, u64 max)
7459 return nfsd_foreach_client_openowner(clp, max, collect,
7460 unhash_openowner_locked);
7464 nfsd_inject_print_openowners(void)
7466 struct nfs4_client *clp;
7468 struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
7471 if (!nfsd_netns_ready(nn))
7474 spin_lock(&nn->client_lock);
7475 list_for_each_entry(clp, &nn->client_lru, cl_lru)
7476 count += nfsd_print_client_openowners(clp);
7477 spin_unlock(&nn->client_lock);
7483 nfsd_reap_openowners(struct list_head *reaplist)
7485 struct nfs4_client *clp;
7486 struct nfs4_openowner *oop, *next;
7488 list_for_each_entry_safe(oop, next, reaplist, oo_perclient) {
7489 list_del_init(&oop->oo_perclient);
7490 clp = oop->oo_owner.so_client;
7491 release_openowner(oop);
7497 nfsd_inject_forget_client_openowners(struct sockaddr_storage *addr,
7500 unsigned int count = 0;
7501 struct nfs4_client *clp;
7502 struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
7504 LIST_HEAD(reaplist);
7506 if (!nfsd_netns_ready(nn))
7509 spin_lock(&nn->client_lock);
7510 clp = nfsd_find_client(addr, addr_size);
7512 count = nfsd_collect_client_openowners(clp, &reaplist, 0);
7513 spin_unlock(&nn->client_lock);
7514 nfsd_reap_openowners(&reaplist);
7519 nfsd_inject_forget_openowners(u64 max)
7522 struct nfs4_client *clp;
7523 struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
7525 LIST_HEAD(reaplist);
7527 if (!nfsd_netns_ready(nn))
7530 spin_lock(&nn->client_lock);
7531 list_for_each_entry(clp, &nn->client_lru, cl_lru) {
7532 count += nfsd_collect_client_openowners(clp, &reaplist,
7534 if (max != 0 && count >= max)
7537 spin_unlock(&nn->client_lock);
7538 nfsd_reap_openowners(&reaplist);
7542 static u64 nfsd_find_all_delegations(struct nfs4_client *clp, u64 max,
7543 struct list_head *victims)
7545 struct nfs4_delegation *dp, *next;
7546 struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
7550 lockdep_assert_held(&nn->client_lock);
7552 spin_lock(&state_lock);
7553 list_for_each_entry_safe(dp, next, &clp->cl_delegations, dl_perclnt) {
7556 * It's not safe to mess with delegations that have a
7557 * non-zero dl_time. They might have already been broken
7558 * and could be processed by the laundromat outside of
7559 * the state_lock. Just leave them be.
7561 if (dp->dl_time != 0)
7564 atomic_inc(&clp->cl_rpc_users);
7565 WARN_ON(!unhash_delegation_locked(dp));
7566 list_add(&dp->dl_recall_lru, victims);
7570 * Despite the fact that these functions deal with
7571 * 64-bit integers for "count", we must ensure that
7572 * it doesn't blow up the clp->cl_rpc_users. Throw a
7573 * warning if we start to approach INT_MAX here.
7575 WARN_ON_ONCE(count == (INT_MAX / 2));
7579 spin_unlock(&state_lock);
7584 nfsd_print_client_delegations(struct nfs4_client *clp)
7586 u64 count = nfsd_find_all_delegations(clp, 0, NULL);
7588 nfsd_print_count(clp, count, "delegations");
7593 nfsd_inject_print_delegations(void)
7595 struct nfs4_client *clp;
7597 struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
7600 if (!nfsd_netns_ready(nn))
7603 spin_lock(&nn->client_lock);
7604 list_for_each_entry(clp, &nn->client_lru, cl_lru)
7605 count += nfsd_print_client_delegations(clp);
7606 spin_unlock(&nn->client_lock);
7612 nfsd_forget_delegations(struct list_head *reaplist)
7614 struct nfs4_client *clp;
7615 struct nfs4_delegation *dp, *next;
7617 list_for_each_entry_safe(dp, next, reaplist, dl_recall_lru) {
7618 list_del_init(&dp->dl_recall_lru);
7619 clp = dp->dl_stid.sc_client;
7620 revoke_delegation(dp);
7626 nfsd_inject_forget_client_delegations(struct sockaddr_storage *addr,
7630 struct nfs4_client *clp;
7631 struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
7633 LIST_HEAD(reaplist);
7635 if (!nfsd_netns_ready(nn))
7638 spin_lock(&nn->client_lock);
7639 clp = nfsd_find_client(addr, addr_size);
7641 count = nfsd_find_all_delegations(clp, 0, &reaplist);
7642 spin_unlock(&nn->client_lock);
7644 nfsd_forget_delegations(&reaplist);
7649 nfsd_inject_forget_delegations(u64 max)
7652 struct nfs4_client *clp;
7653 struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
7655 LIST_HEAD(reaplist);
7657 if (!nfsd_netns_ready(nn))
7660 spin_lock(&nn->client_lock);
7661 list_for_each_entry(clp, &nn->client_lru, cl_lru) {
7662 count += nfsd_find_all_delegations(clp, max - count, &reaplist);
7663 if (max != 0 && count >= max)
7666 spin_unlock(&nn->client_lock);
7667 nfsd_forget_delegations(&reaplist);
7672 nfsd_recall_delegations(struct list_head *reaplist)
7674 struct nfs4_client *clp;
7675 struct nfs4_delegation *dp, *next;
7677 list_for_each_entry_safe(dp, next, reaplist, dl_recall_lru) {
7678 list_del_init(&dp->dl_recall_lru);
7679 clp = dp->dl_stid.sc_client;
7681 * We skipped all entries that had a zero dl_time before,
7682 * so we can now reset the dl_time back to 0. If a delegation
7683 * break comes in now, then it won't make any difference since
7684 * we're recalling it either way.
7686 spin_lock(&state_lock);
7688 spin_unlock(&state_lock);
7689 nfsd_break_one_deleg(dp);
7695 nfsd_inject_recall_client_delegations(struct sockaddr_storage *addr,
7699 struct nfs4_client *clp;
7700 struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
7702 LIST_HEAD(reaplist);
7704 if (!nfsd_netns_ready(nn))
7707 spin_lock(&nn->client_lock);
7708 clp = nfsd_find_client(addr, addr_size);
7710 count = nfsd_find_all_delegations(clp, 0, &reaplist);
7711 spin_unlock(&nn->client_lock);
7713 nfsd_recall_delegations(&reaplist);
7718 nfsd_inject_recall_delegations(u64 max)
7721 struct nfs4_client *clp, *next;
7722 struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
7724 LIST_HEAD(reaplist);
7726 if (!nfsd_netns_ready(nn))
7729 spin_lock(&nn->client_lock);
7730 list_for_each_entry_safe(clp, next, &nn->client_lru, cl_lru) {
7731 count += nfsd_find_all_delegations(clp, max - count, &reaplist);
7732 if (max != 0 && ++count >= max)
7735 spin_unlock(&nn->client_lock);
7736 nfsd_recall_delegations(&reaplist);
7739 #endif /* CONFIG_NFSD_FAULT_INJECTION */
7742 * Since the lifetime of a delegation isn't limited to that of an open, a
7743 * client may quite reasonably hang on to a delegation as long as it has
7744 * the inode cached. This becomes an obvious problem the first time a
7745 * client's inode cache approaches the size of the server's total memory.
7747 * For now we avoid this problem by imposing a hard limit on the number
7748 * of delegations, which varies according to the server's memory size.
7751 set_max_delegations(void)
7754 * Allow at most 4 delegations per megabyte of RAM. Quick
7755 * estimates suggest that in the worst case (where every delegation
7756 * is for a different inode), a delegation could take about 1.5K,
7757 * giving a worst case usage of about 6% of memory.
7759 max_delegations = nr_free_buffer_pages() >> (20 - 2 - PAGE_SHIFT);
7762 static int nfs4_state_create_net(struct net *net)
7764 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
7767 nn->conf_id_hashtbl = kmalloc_array(CLIENT_HASH_SIZE,
7768 sizeof(struct list_head),
7770 if (!nn->conf_id_hashtbl)
7772 nn->unconf_id_hashtbl = kmalloc_array(CLIENT_HASH_SIZE,
7773 sizeof(struct list_head),
7775 if (!nn->unconf_id_hashtbl)
7777 nn->sessionid_hashtbl = kmalloc_array(SESSION_HASH_SIZE,
7778 sizeof(struct list_head),
7780 if (!nn->sessionid_hashtbl)
7783 for (i = 0; i < CLIENT_HASH_SIZE; i++) {
7784 INIT_LIST_HEAD(&nn->conf_id_hashtbl[i]);
7785 INIT_LIST_HEAD(&nn->unconf_id_hashtbl[i]);
7787 for (i = 0; i < SESSION_HASH_SIZE; i++)
7788 INIT_LIST_HEAD(&nn->sessionid_hashtbl[i]);
7789 nn->conf_name_tree = RB_ROOT;
7790 nn->unconf_name_tree = RB_ROOT;
7791 nn->boot_time = ktime_get_real_seconds();
7792 nn->grace_ended = false;
7793 nn->nfsd4_manager.block_opens = true;
7794 INIT_LIST_HEAD(&nn->nfsd4_manager.list);
7795 INIT_LIST_HEAD(&nn->client_lru);
7796 INIT_LIST_HEAD(&nn->close_lru);
7797 INIT_LIST_HEAD(&nn->del_recall_lru);
7798 spin_lock_init(&nn->client_lock);
7799 spin_lock_init(&nn->s2s_cp_lock);
7800 idr_init(&nn->s2s_cp_stateids);
7802 spin_lock_init(&nn->blocked_locks_lock);
7803 INIT_LIST_HEAD(&nn->blocked_locks_lru);
7805 INIT_DELAYED_WORK(&nn->laundromat_work, laundromat_main);
7811 kfree(nn->unconf_id_hashtbl);
7813 kfree(nn->conf_id_hashtbl);
7819 nfs4_state_destroy_net(struct net *net)
7822 struct nfs4_client *clp = NULL;
7823 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
7825 for (i = 0; i < CLIENT_HASH_SIZE; i++) {
7826 while (!list_empty(&nn->conf_id_hashtbl[i])) {
7827 clp = list_entry(nn->conf_id_hashtbl[i].next, struct nfs4_client, cl_idhash);
7828 destroy_client(clp);
7832 WARN_ON(!list_empty(&nn->blocked_locks_lru));
7834 for (i = 0; i < CLIENT_HASH_SIZE; i++) {
7835 while (!list_empty(&nn->unconf_id_hashtbl[i])) {
7836 clp = list_entry(nn->unconf_id_hashtbl[i].next, struct nfs4_client, cl_idhash);
7837 destroy_client(clp);
7841 kfree(nn->sessionid_hashtbl);
7842 kfree(nn->unconf_id_hashtbl);
7843 kfree(nn->conf_id_hashtbl);
7848 nfs4_state_start_net(struct net *net)
7850 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
7853 ret = nfs4_state_create_net(net);
7856 locks_start_grace(net, &nn->nfsd4_manager);
7857 nfsd4_client_tracking_init(net);
7858 if (nn->track_reclaim_completes && nn->reclaim_str_hashtbl_size == 0)
7860 printk(KERN_INFO "NFSD: starting %lld-second grace period (net %x)\n",
7861 nn->nfsd4_grace, net->ns.inum);
7862 queue_delayed_work(laundry_wq, &nn->laundromat_work, nn->nfsd4_grace * HZ);
7866 printk(KERN_INFO "NFSD: no clients to reclaim, skipping NFSv4 grace period (net %x)\n",
7868 queue_delayed_work(laundry_wq, &nn->laundromat_work, nn->nfsd4_lease * HZ);
7869 nfsd4_end_grace(nn);
7873 /* initialization to perform when the nfsd service is started: */
7876 nfs4_state_start(void)
7880 laundry_wq = alloc_workqueue("%s", WQ_UNBOUND, 0, "nfsd4");
7881 if (laundry_wq == NULL) {
7885 ret = nfsd4_create_callback_queue();
7887 goto out_free_laundry;
7889 set_max_delegations();
7893 destroy_workqueue(laundry_wq);
7899 nfs4_state_shutdown_net(struct net *net)
7901 struct nfs4_delegation *dp = NULL;
7902 struct list_head *pos, *next, reaplist;
7903 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
7905 cancel_delayed_work_sync(&nn->laundromat_work);
7906 locks_end_grace(&nn->nfsd4_manager);
7908 INIT_LIST_HEAD(&reaplist);
7909 spin_lock(&state_lock);
7910 list_for_each_safe(pos, next, &nn->del_recall_lru) {
7911 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
7912 WARN_ON(!unhash_delegation_locked(dp));
7913 list_add(&dp->dl_recall_lru, &reaplist);
7915 spin_unlock(&state_lock);
7916 list_for_each_safe(pos, next, &reaplist) {
7917 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
7918 list_del_init(&dp->dl_recall_lru);
7919 destroy_unhashed_deleg(dp);
7922 nfsd4_client_tracking_exit(net);
7923 nfs4_state_destroy_net(net);
7927 nfs4_state_shutdown(void)
7929 destroy_workqueue(laundry_wq);
7930 nfsd4_destroy_callback_queue();
7934 get_stateid(struct nfsd4_compound_state *cstate, stateid_t *stateid)
7936 if (HAS_CSTATE_FLAG(cstate, CURRENT_STATE_ID_FLAG) &&
7937 CURRENT_STATEID(stateid))
7938 memcpy(stateid, &cstate->current_stateid, sizeof(stateid_t));
7942 put_stateid(struct nfsd4_compound_state *cstate, stateid_t *stateid)
7944 if (cstate->minorversion) {
7945 memcpy(&cstate->current_stateid, stateid, sizeof(stateid_t));
7946 SET_CSTATE_FLAG(cstate, CURRENT_STATE_ID_FLAG);
7951 clear_current_stateid(struct nfsd4_compound_state *cstate)
7953 CLEAR_CSTATE_FLAG(cstate, CURRENT_STATE_ID_FLAG);
7957 * functions to set current state id
7960 nfsd4_set_opendowngradestateid(struct nfsd4_compound_state *cstate,
7961 union nfsd4_op_u *u)
7963 put_stateid(cstate, &u->open_downgrade.od_stateid);
7967 nfsd4_set_openstateid(struct nfsd4_compound_state *cstate,
7968 union nfsd4_op_u *u)
7970 put_stateid(cstate, &u->open.op_stateid);
7974 nfsd4_set_closestateid(struct nfsd4_compound_state *cstate,
7975 union nfsd4_op_u *u)
7977 put_stateid(cstate, &u->close.cl_stateid);
7981 nfsd4_set_lockstateid(struct nfsd4_compound_state *cstate,
7982 union nfsd4_op_u *u)
7984 put_stateid(cstate, &u->lock.lk_resp_stateid);
7988 * functions to consume current state id
7992 nfsd4_get_opendowngradestateid(struct nfsd4_compound_state *cstate,
7993 union nfsd4_op_u *u)
7995 get_stateid(cstate, &u->open_downgrade.od_stateid);
7999 nfsd4_get_delegreturnstateid(struct nfsd4_compound_state *cstate,
8000 union nfsd4_op_u *u)
8002 get_stateid(cstate, &u->delegreturn.dr_stateid);
8006 nfsd4_get_freestateid(struct nfsd4_compound_state *cstate,
8007 union nfsd4_op_u *u)
8009 get_stateid(cstate, &u->free_stateid.fr_stateid);
8013 nfsd4_get_setattrstateid(struct nfsd4_compound_state *cstate,
8014 union nfsd4_op_u *u)
8016 get_stateid(cstate, &u->setattr.sa_stateid);
8020 nfsd4_get_closestateid(struct nfsd4_compound_state *cstate,
8021 union nfsd4_op_u *u)
8023 get_stateid(cstate, &u->close.cl_stateid);
8027 nfsd4_get_lockustateid(struct nfsd4_compound_state *cstate,
8028 union nfsd4_op_u *u)
8030 get_stateid(cstate, &u->locku.lu_stateid);
8034 nfsd4_get_readstateid(struct nfsd4_compound_state *cstate,
8035 union nfsd4_op_u *u)
8037 get_stateid(cstate, &u->read.rd_stateid);
8041 nfsd4_get_writestateid(struct nfsd4_compound_state *cstate,
8042 union nfsd4_op_u *u)
8044 get_stateid(cstate, &u->write.wr_stateid);