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>
46 #include <linux/fsnotify.h>
50 #include "current_stateid.h"
54 #include "filecache.h"
57 #define NFSDDBG_FACILITY NFSDDBG_PROC
59 #define all_ones {{~0,~0},~0}
60 static const stateid_t one_stateid = {
62 .si_opaque = all_ones,
64 static const stateid_t zero_stateid = {
67 static const stateid_t currentstateid = {
70 static const stateid_t close_stateid = {
71 .si_generation = 0xffffffffU,
74 static u64 current_sessionid = 1;
76 #define ZERO_STATEID(stateid) (!memcmp((stateid), &zero_stateid, sizeof(stateid_t)))
77 #define ONE_STATEID(stateid) (!memcmp((stateid), &one_stateid, sizeof(stateid_t)))
78 #define CURRENT_STATEID(stateid) (!memcmp((stateid), ¤tstateid, sizeof(stateid_t)))
79 #define CLOSE_STATEID(stateid) (!memcmp((stateid), &close_stateid, sizeof(stateid_t)))
81 /* forward declarations */
82 static bool check_for_locks(struct nfs4_file *fp, struct nfs4_lockowner *lowner);
83 static void nfs4_free_ol_stateid(struct nfs4_stid *stid);
84 void nfsd4_end_grace(struct nfsd_net *nn);
85 static void _free_cpntf_state_locked(struct nfsd_net *nn, struct nfs4_cpntf_state *cps);
90 * Currently used for the del_recall_lru and file hash table. In an
91 * effort to decrease the scope of the client_mutex, this spinlock may
92 * eventually cover more:
94 static DEFINE_SPINLOCK(state_lock);
96 enum nfsd4_st_mutex_lock_subclass {
97 OPEN_STATEID_MUTEX = 0,
98 LOCK_STATEID_MUTEX = 1,
102 * A waitqueue for all in-progress 4.0 CLOSE operations that are waiting for
103 * the refcount on the open stateid to drop.
105 static DECLARE_WAIT_QUEUE_HEAD(close_wq);
108 * A waitqueue where a writer to clients/#/ctl destroying a client can
109 * wait for cl_rpc_users to drop to 0 and then for the client to be
112 static DECLARE_WAIT_QUEUE_HEAD(expiry_wq);
114 static struct kmem_cache *client_slab;
115 static struct kmem_cache *openowner_slab;
116 static struct kmem_cache *lockowner_slab;
117 static struct kmem_cache *file_slab;
118 static struct kmem_cache *stateid_slab;
119 static struct kmem_cache *deleg_slab;
120 static struct kmem_cache *odstate_slab;
122 static void free_session(struct nfsd4_session *);
124 static const struct nfsd4_callback_ops nfsd4_cb_recall_ops;
125 static const struct nfsd4_callback_ops nfsd4_cb_notify_lock_ops;
127 static bool is_session_dead(struct nfsd4_session *ses)
129 return ses->se_flags & NFS4_SESSION_DEAD;
132 static __be32 mark_session_dead_locked(struct nfsd4_session *ses, int ref_held_by_me)
134 if (atomic_read(&ses->se_ref) > ref_held_by_me)
135 return nfserr_jukebox;
136 ses->se_flags |= NFS4_SESSION_DEAD;
140 static bool is_client_expired(struct nfs4_client *clp)
142 return clp->cl_time == 0;
145 static __be32 get_client_locked(struct nfs4_client *clp)
147 struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
149 lockdep_assert_held(&nn->client_lock);
151 if (is_client_expired(clp))
152 return nfserr_expired;
153 atomic_inc(&clp->cl_rpc_users);
157 /* must be called under the client_lock */
159 renew_client_locked(struct nfs4_client *clp)
161 struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
163 if (is_client_expired(clp)) {
165 printk("%s: client (clientid %08x/%08x) already expired\n",
167 clp->cl_clientid.cl_boot,
168 clp->cl_clientid.cl_id);
172 list_move_tail(&clp->cl_lru, &nn->client_lru);
173 clp->cl_time = ktime_get_boottime_seconds();
176 static void put_client_renew_locked(struct nfs4_client *clp)
178 struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
180 lockdep_assert_held(&nn->client_lock);
182 if (!atomic_dec_and_test(&clp->cl_rpc_users))
184 if (!is_client_expired(clp))
185 renew_client_locked(clp);
187 wake_up_all(&expiry_wq);
190 static void put_client_renew(struct nfs4_client *clp)
192 struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
194 if (!atomic_dec_and_lock(&clp->cl_rpc_users, &nn->client_lock))
196 if (!is_client_expired(clp))
197 renew_client_locked(clp);
199 wake_up_all(&expiry_wq);
200 spin_unlock(&nn->client_lock);
203 static __be32 nfsd4_get_session_locked(struct nfsd4_session *ses)
207 if (is_session_dead(ses))
208 return nfserr_badsession;
209 status = get_client_locked(ses->se_client);
212 atomic_inc(&ses->se_ref);
216 static void nfsd4_put_session_locked(struct nfsd4_session *ses)
218 struct nfs4_client *clp = ses->se_client;
219 struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
221 lockdep_assert_held(&nn->client_lock);
223 if (atomic_dec_and_test(&ses->se_ref) && is_session_dead(ses))
225 put_client_renew_locked(clp);
228 static void nfsd4_put_session(struct nfsd4_session *ses)
230 struct nfs4_client *clp = ses->se_client;
231 struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
233 spin_lock(&nn->client_lock);
234 nfsd4_put_session_locked(ses);
235 spin_unlock(&nn->client_lock);
238 static struct nfsd4_blocked_lock *
239 find_blocked_lock(struct nfs4_lockowner *lo, struct knfsd_fh *fh,
242 struct nfsd4_blocked_lock *cur, *found = NULL;
244 spin_lock(&nn->blocked_locks_lock);
245 list_for_each_entry(cur, &lo->lo_blocked, nbl_list) {
246 if (fh_match(fh, &cur->nbl_fh)) {
247 list_del_init(&cur->nbl_list);
248 list_del_init(&cur->nbl_lru);
253 spin_unlock(&nn->blocked_locks_lock);
255 locks_delete_block(&found->nbl_lock);
259 static struct nfsd4_blocked_lock *
260 find_or_allocate_block(struct nfs4_lockowner *lo, struct knfsd_fh *fh,
263 struct nfsd4_blocked_lock *nbl;
265 nbl = find_blocked_lock(lo, fh, nn);
267 nbl= kmalloc(sizeof(*nbl), GFP_KERNEL);
269 INIT_LIST_HEAD(&nbl->nbl_list);
270 INIT_LIST_HEAD(&nbl->nbl_lru);
271 fh_copy_shallow(&nbl->nbl_fh, fh);
272 locks_init_lock(&nbl->nbl_lock);
273 nfsd4_init_cb(&nbl->nbl_cb, lo->lo_owner.so_client,
274 &nfsd4_cb_notify_lock_ops,
275 NFSPROC4_CLNT_CB_NOTIFY_LOCK);
282 free_blocked_lock(struct nfsd4_blocked_lock *nbl)
284 locks_delete_block(&nbl->nbl_lock);
285 locks_release_private(&nbl->nbl_lock);
290 remove_blocked_locks(struct nfs4_lockowner *lo)
292 struct nfs4_client *clp = lo->lo_owner.so_client;
293 struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
294 struct nfsd4_blocked_lock *nbl;
297 /* Dequeue all blocked locks */
298 spin_lock(&nn->blocked_locks_lock);
299 while (!list_empty(&lo->lo_blocked)) {
300 nbl = list_first_entry(&lo->lo_blocked,
301 struct nfsd4_blocked_lock,
303 list_del_init(&nbl->nbl_list);
304 list_move(&nbl->nbl_lru, &reaplist);
306 spin_unlock(&nn->blocked_locks_lock);
309 while (!list_empty(&reaplist)) {
310 nbl = list_first_entry(&reaplist, struct nfsd4_blocked_lock,
312 list_del_init(&nbl->nbl_lru);
313 free_blocked_lock(nbl);
318 nfsd4_cb_notify_lock_prepare(struct nfsd4_callback *cb)
320 struct nfsd4_blocked_lock *nbl = container_of(cb,
321 struct nfsd4_blocked_lock, nbl_cb);
322 locks_delete_block(&nbl->nbl_lock);
326 nfsd4_cb_notify_lock_done(struct nfsd4_callback *cb, struct rpc_task *task)
329 * Since this is just an optimization, we don't try very hard if it
330 * turns out not to succeed. We'll requeue it on NFS4ERR_DELAY, and
331 * just quit trying on anything else.
333 switch (task->tk_status) {
335 rpc_delay(task, 1 * HZ);
343 nfsd4_cb_notify_lock_release(struct nfsd4_callback *cb)
345 struct nfsd4_blocked_lock *nbl = container_of(cb,
346 struct nfsd4_blocked_lock, nbl_cb);
348 free_blocked_lock(nbl);
351 static const struct nfsd4_callback_ops nfsd4_cb_notify_lock_ops = {
352 .prepare = nfsd4_cb_notify_lock_prepare,
353 .done = nfsd4_cb_notify_lock_done,
354 .release = nfsd4_cb_notify_lock_release,
357 static inline struct nfs4_stateowner *
358 nfs4_get_stateowner(struct nfs4_stateowner *sop)
360 atomic_inc(&sop->so_count);
365 same_owner_str(struct nfs4_stateowner *sop, struct xdr_netobj *owner)
367 return (sop->so_owner.len == owner->len) &&
368 0 == memcmp(sop->so_owner.data, owner->data, owner->len);
371 static struct nfs4_openowner *
372 find_openstateowner_str_locked(unsigned int hashval, struct nfsd4_open *open,
373 struct nfs4_client *clp)
375 struct nfs4_stateowner *so;
377 lockdep_assert_held(&clp->cl_lock);
379 list_for_each_entry(so, &clp->cl_ownerstr_hashtbl[hashval],
381 if (!so->so_is_open_owner)
383 if (same_owner_str(so, &open->op_owner))
384 return openowner(nfs4_get_stateowner(so));
389 static struct nfs4_openowner *
390 find_openstateowner_str(unsigned int hashval, struct nfsd4_open *open,
391 struct nfs4_client *clp)
393 struct nfs4_openowner *oo;
395 spin_lock(&clp->cl_lock);
396 oo = find_openstateowner_str_locked(hashval, open, clp);
397 spin_unlock(&clp->cl_lock);
402 opaque_hashval(const void *ptr, int nbytes)
404 unsigned char *cptr = (unsigned char *) ptr;
414 static void nfsd4_free_file_rcu(struct rcu_head *rcu)
416 struct nfs4_file *fp = container_of(rcu, struct nfs4_file, fi_rcu);
418 kmem_cache_free(file_slab, fp);
422 put_nfs4_file(struct nfs4_file *fi)
424 might_lock(&state_lock);
426 if (refcount_dec_and_lock(&fi->fi_ref, &state_lock)) {
427 hlist_del_rcu(&fi->fi_hash);
428 spin_unlock(&state_lock);
429 WARN_ON_ONCE(!list_empty(&fi->fi_clnt_odstate));
430 WARN_ON_ONCE(!list_empty(&fi->fi_delegations));
431 call_rcu(&fi->fi_rcu, nfsd4_free_file_rcu);
435 static struct nfsd_file *
436 __nfs4_get_fd(struct nfs4_file *f, int oflag)
438 if (f->fi_fds[oflag])
439 return nfsd_file_get(f->fi_fds[oflag]);
443 static struct nfsd_file *
444 find_writeable_file_locked(struct nfs4_file *f)
446 struct nfsd_file *ret;
448 lockdep_assert_held(&f->fi_lock);
450 ret = __nfs4_get_fd(f, O_WRONLY);
452 ret = __nfs4_get_fd(f, O_RDWR);
456 static struct nfsd_file *
457 find_writeable_file(struct nfs4_file *f)
459 struct nfsd_file *ret;
461 spin_lock(&f->fi_lock);
462 ret = find_writeable_file_locked(f);
463 spin_unlock(&f->fi_lock);
468 static struct nfsd_file *
469 find_readable_file_locked(struct nfs4_file *f)
471 struct nfsd_file *ret;
473 lockdep_assert_held(&f->fi_lock);
475 ret = __nfs4_get_fd(f, O_RDONLY);
477 ret = __nfs4_get_fd(f, O_RDWR);
481 static struct nfsd_file *
482 find_readable_file(struct nfs4_file *f)
484 struct nfsd_file *ret;
486 spin_lock(&f->fi_lock);
487 ret = find_readable_file_locked(f);
488 spin_unlock(&f->fi_lock);
494 find_any_file(struct nfs4_file *f)
496 struct nfsd_file *ret;
500 spin_lock(&f->fi_lock);
501 ret = __nfs4_get_fd(f, O_RDWR);
503 ret = __nfs4_get_fd(f, O_WRONLY);
505 ret = __nfs4_get_fd(f, O_RDONLY);
507 spin_unlock(&f->fi_lock);
511 static struct nfsd_file *find_deleg_file(struct nfs4_file *f)
513 struct nfsd_file *ret = NULL;
515 spin_lock(&f->fi_lock);
516 if (f->fi_deleg_file)
517 ret = nfsd_file_get(f->fi_deleg_file);
518 spin_unlock(&f->fi_lock);
522 static atomic_long_t num_delegations;
523 unsigned long max_delegations;
526 * Open owner state (share locks)
529 /* hash tables for lock and open owners */
530 #define OWNER_HASH_BITS 8
531 #define OWNER_HASH_SIZE (1 << OWNER_HASH_BITS)
532 #define OWNER_HASH_MASK (OWNER_HASH_SIZE - 1)
534 static unsigned int ownerstr_hashval(struct xdr_netobj *ownername)
538 ret = opaque_hashval(ownername->data, ownername->len);
539 return ret & OWNER_HASH_MASK;
542 /* hash table for nfs4_file */
543 #define FILE_HASH_BITS 8
544 #define FILE_HASH_SIZE (1 << FILE_HASH_BITS)
546 static unsigned int nfsd_fh_hashval(struct knfsd_fh *fh)
548 return jhash2(fh->fh_base.fh_pad, XDR_QUADLEN(fh->fh_size), 0);
551 static unsigned int file_hashval(struct knfsd_fh *fh)
553 return nfsd_fh_hashval(fh) & (FILE_HASH_SIZE - 1);
556 static struct hlist_head file_hashtbl[FILE_HASH_SIZE];
559 __nfs4_file_get_access(struct nfs4_file *fp, u32 access)
561 lockdep_assert_held(&fp->fi_lock);
563 if (access & NFS4_SHARE_ACCESS_WRITE)
564 atomic_inc(&fp->fi_access[O_WRONLY]);
565 if (access & NFS4_SHARE_ACCESS_READ)
566 atomic_inc(&fp->fi_access[O_RDONLY]);
570 nfs4_file_get_access(struct nfs4_file *fp, u32 access)
572 lockdep_assert_held(&fp->fi_lock);
574 /* Does this access mode make sense? */
575 if (access & ~NFS4_SHARE_ACCESS_BOTH)
578 /* Does it conflict with a deny mode already set? */
579 if ((access & fp->fi_share_deny) != 0)
580 return nfserr_share_denied;
582 __nfs4_file_get_access(fp, access);
586 static __be32 nfs4_file_check_deny(struct nfs4_file *fp, u32 deny)
588 /* Common case is that there is no deny mode. */
590 /* Does this deny mode make sense? */
591 if (deny & ~NFS4_SHARE_DENY_BOTH)
594 if ((deny & NFS4_SHARE_DENY_READ) &&
595 atomic_read(&fp->fi_access[O_RDONLY]))
596 return nfserr_share_denied;
598 if ((deny & NFS4_SHARE_DENY_WRITE) &&
599 atomic_read(&fp->fi_access[O_WRONLY]))
600 return nfserr_share_denied;
605 static void __nfs4_file_put_access(struct nfs4_file *fp, int oflag)
607 might_lock(&fp->fi_lock);
609 if (atomic_dec_and_lock(&fp->fi_access[oflag], &fp->fi_lock)) {
610 struct nfsd_file *f1 = NULL;
611 struct nfsd_file *f2 = NULL;
613 swap(f1, fp->fi_fds[oflag]);
614 if (atomic_read(&fp->fi_access[1 - oflag]) == 0)
615 swap(f2, fp->fi_fds[O_RDWR]);
616 spin_unlock(&fp->fi_lock);
624 static void nfs4_file_put_access(struct nfs4_file *fp, u32 access)
626 WARN_ON_ONCE(access & ~NFS4_SHARE_ACCESS_BOTH);
628 if (access & NFS4_SHARE_ACCESS_WRITE)
629 __nfs4_file_put_access(fp, O_WRONLY);
630 if (access & NFS4_SHARE_ACCESS_READ)
631 __nfs4_file_put_access(fp, O_RDONLY);
635 * Allocate a new open/delegation state counter. This is needed for
636 * pNFS for proper return on close semantics.
638 * Note that we only allocate it for pNFS-enabled exports, otherwise
639 * all pointers to struct nfs4_clnt_odstate are always NULL.
641 static struct nfs4_clnt_odstate *
642 alloc_clnt_odstate(struct nfs4_client *clp)
644 struct nfs4_clnt_odstate *co;
646 co = kmem_cache_zalloc(odstate_slab, GFP_KERNEL);
649 refcount_set(&co->co_odcount, 1);
655 hash_clnt_odstate_locked(struct nfs4_clnt_odstate *co)
657 struct nfs4_file *fp = co->co_file;
659 lockdep_assert_held(&fp->fi_lock);
660 list_add(&co->co_perfile, &fp->fi_clnt_odstate);
664 get_clnt_odstate(struct nfs4_clnt_odstate *co)
667 refcount_inc(&co->co_odcount);
671 put_clnt_odstate(struct nfs4_clnt_odstate *co)
673 struct nfs4_file *fp;
679 if (refcount_dec_and_lock(&co->co_odcount, &fp->fi_lock)) {
680 list_del(&co->co_perfile);
681 spin_unlock(&fp->fi_lock);
683 nfsd4_return_all_file_layouts(co->co_client, fp);
684 kmem_cache_free(odstate_slab, co);
688 static struct nfs4_clnt_odstate *
689 find_or_hash_clnt_odstate(struct nfs4_file *fp, struct nfs4_clnt_odstate *new)
691 struct nfs4_clnt_odstate *co;
692 struct nfs4_client *cl;
699 spin_lock(&fp->fi_lock);
700 list_for_each_entry(co, &fp->fi_clnt_odstate, co_perfile) {
701 if (co->co_client == cl) {
702 get_clnt_odstate(co);
708 hash_clnt_odstate_locked(new);
710 spin_unlock(&fp->fi_lock);
714 struct nfs4_stid *nfs4_alloc_stid(struct nfs4_client *cl, struct kmem_cache *slab,
715 void (*sc_free)(struct nfs4_stid *))
717 struct nfs4_stid *stid;
720 stid = kmem_cache_zalloc(slab, GFP_KERNEL);
724 idr_preload(GFP_KERNEL);
725 spin_lock(&cl->cl_lock);
726 /* Reserving 0 for start of file in nfsdfs "states" file: */
727 new_id = idr_alloc_cyclic(&cl->cl_stateids, stid, 1, 0, GFP_NOWAIT);
728 spin_unlock(&cl->cl_lock);
733 stid->sc_free = sc_free;
734 stid->sc_client = cl;
735 stid->sc_stateid.si_opaque.so_id = new_id;
736 stid->sc_stateid.si_opaque.so_clid = cl->cl_clientid;
737 /* Will be incremented before return to client: */
738 refcount_set(&stid->sc_count, 1);
739 spin_lock_init(&stid->sc_lock);
740 INIT_LIST_HEAD(&stid->sc_cp_list);
743 * It shouldn't be a problem to reuse an opaque stateid value.
744 * I don't think it is for 4.1. But with 4.0 I worry that, for
745 * example, a stray write retransmission could be accepted by
746 * the server when it should have been rejected. Therefore,
747 * adopt a trick from the sctp code to attempt to maximize the
748 * amount of time until an id is reused, by ensuring they always
749 * "increase" (mod INT_MAX):
753 kmem_cache_free(slab, stid);
758 * Create a unique stateid_t to represent each COPY.
760 static int nfs4_init_cp_state(struct nfsd_net *nn, copy_stateid_t *stid,
761 unsigned char sc_type)
765 stid->stid.si_opaque.so_clid.cl_boot = (u32)nn->boot_time;
766 stid->stid.si_opaque.so_clid.cl_id = nn->s2s_cp_cl_id;
767 stid->sc_type = sc_type;
769 idr_preload(GFP_KERNEL);
770 spin_lock(&nn->s2s_cp_lock);
771 new_id = idr_alloc_cyclic(&nn->s2s_cp_stateids, stid, 0, 0, GFP_NOWAIT);
772 stid->stid.si_opaque.so_id = new_id;
773 stid->stid.si_generation = 1;
774 spin_unlock(&nn->s2s_cp_lock);
781 int nfs4_init_copy_state(struct nfsd_net *nn, struct nfsd4_copy *copy)
783 return nfs4_init_cp_state(nn, ©->cp_stateid, NFS4_COPY_STID);
786 struct nfs4_cpntf_state *nfs4_alloc_init_cpntf_state(struct nfsd_net *nn,
787 struct nfs4_stid *p_stid)
789 struct nfs4_cpntf_state *cps;
791 cps = kzalloc(sizeof(struct nfs4_cpntf_state), GFP_KERNEL);
794 cps->cpntf_time = ktime_get_boottime_seconds();
795 refcount_set(&cps->cp_stateid.sc_count, 1);
796 if (!nfs4_init_cp_state(nn, &cps->cp_stateid, NFS4_COPYNOTIFY_STID))
798 spin_lock(&nn->s2s_cp_lock);
799 list_add(&cps->cp_list, &p_stid->sc_cp_list);
800 spin_unlock(&nn->s2s_cp_lock);
807 void nfs4_free_copy_state(struct nfsd4_copy *copy)
811 WARN_ON_ONCE(copy->cp_stateid.sc_type != NFS4_COPY_STID);
812 nn = net_generic(copy->cp_clp->net, nfsd_net_id);
813 spin_lock(&nn->s2s_cp_lock);
814 idr_remove(&nn->s2s_cp_stateids,
815 copy->cp_stateid.stid.si_opaque.so_id);
816 spin_unlock(&nn->s2s_cp_lock);
819 static void nfs4_free_cpntf_statelist(struct net *net, struct nfs4_stid *stid)
821 struct nfs4_cpntf_state *cps;
824 nn = net_generic(net, nfsd_net_id);
825 spin_lock(&nn->s2s_cp_lock);
826 while (!list_empty(&stid->sc_cp_list)) {
827 cps = list_first_entry(&stid->sc_cp_list,
828 struct nfs4_cpntf_state, cp_list);
829 _free_cpntf_state_locked(nn, cps);
831 spin_unlock(&nn->s2s_cp_lock);
834 static struct nfs4_ol_stateid * nfs4_alloc_open_stateid(struct nfs4_client *clp)
836 struct nfs4_stid *stid;
838 stid = nfs4_alloc_stid(clp, stateid_slab, nfs4_free_ol_stateid);
842 return openlockstateid(stid);
845 static void nfs4_free_deleg(struct nfs4_stid *stid)
847 kmem_cache_free(deleg_slab, stid);
848 atomic_long_dec(&num_delegations);
852 * When we recall a delegation, we should be careful not to hand it
853 * out again straight away.
854 * To ensure this we keep a pair of bloom filters ('new' and 'old')
855 * in which the filehandles of recalled delegations are "stored".
856 * If a filehandle appear in either filter, a delegation is blocked.
857 * When a delegation is recalled, the filehandle is stored in the "new"
859 * Every 30 seconds we swap the filters and clear the "new" one,
860 * unless both are empty of course.
862 * Each filter is 256 bits. We hash the filehandle to 32bit and use the
863 * low 3 bytes as hash-table indices.
865 * 'blocked_delegations_lock', which is always taken in block_delegations(),
866 * is used to manage concurrent access. Testing does not need the lock
867 * except when swapping the two filters.
869 static DEFINE_SPINLOCK(blocked_delegations_lock);
870 static struct bloom_pair {
871 int entries, old_entries;
873 int new; /* index into 'set' */
874 DECLARE_BITMAP(set[2], 256);
875 } blocked_delegations;
877 static int delegation_blocked(struct knfsd_fh *fh)
880 struct bloom_pair *bd = &blocked_delegations;
882 if (bd->entries == 0)
884 if (ktime_get_seconds() - bd->swap_time > 30) {
885 spin_lock(&blocked_delegations_lock);
886 if (ktime_get_seconds() - bd->swap_time > 30) {
887 bd->entries -= bd->old_entries;
888 bd->old_entries = bd->entries;
889 memset(bd->set[bd->new], 0,
892 bd->swap_time = ktime_get_seconds();
894 spin_unlock(&blocked_delegations_lock);
896 hash = jhash(&fh->fh_base, fh->fh_size, 0);
897 if (test_bit(hash&255, bd->set[0]) &&
898 test_bit((hash>>8)&255, bd->set[0]) &&
899 test_bit((hash>>16)&255, bd->set[0]))
902 if (test_bit(hash&255, bd->set[1]) &&
903 test_bit((hash>>8)&255, bd->set[1]) &&
904 test_bit((hash>>16)&255, bd->set[1]))
910 static void block_delegations(struct knfsd_fh *fh)
913 struct bloom_pair *bd = &blocked_delegations;
915 hash = jhash(&fh->fh_base, fh->fh_size, 0);
917 spin_lock(&blocked_delegations_lock);
918 __set_bit(hash&255, bd->set[bd->new]);
919 __set_bit((hash>>8)&255, bd->set[bd->new]);
920 __set_bit((hash>>16)&255, bd->set[bd->new]);
921 if (bd->entries == 0)
922 bd->swap_time = ktime_get_seconds();
924 spin_unlock(&blocked_delegations_lock);
927 static struct nfs4_delegation *
928 alloc_init_deleg(struct nfs4_client *clp, struct nfs4_file *fp,
929 struct svc_fh *current_fh,
930 struct nfs4_clnt_odstate *odstate)
932 struct nfs4_delegation *dp;
935 dprintk("NFSD alloc_init_deleg\n");
936 n = atomic_long_inc_return(&num_delegations);
937 if (n < 0 || n > max_delegations)
939 if (delegation_blocked(¤t_fh->fh_handle))
941 dp = delegstateid(nfs4_alloc_stid(clp, deleg_slab, nfs4_free_deleg));
946 * delegation seqid's are never incremented. The 4.1 special
947 * meaning of seqid 0 isn't meaningful, really, but let's avoid
948 * 0 anyway just for consistency and use 1:
950 dp->dl_stid.sc_stateid.si_generation = 1;
951 INIT_LIST_HEAD(&dp->dl_perfile);
952 INIT_LIST_HEAD(&dp->dl_perclnt);
953 INIT_LIST_HEAD(&dp->dl_recall_lru);
954 dp->dl_clnt_odstate = odstate;
955 get_clnt_odstate(odstate);
956 dp->dl_type = NFS4_OPEN_DELEGATE_READ;
958 nfsd4_init_cb(&dp->dl_recall, dp->dl_stid.sc_client,
959 &nfsd4_cb_recall_ops, NFSPROC4_CLNT_CB_RECALL);
961 dp->dl_stid.sc_file = fp;
964 atomic_long_dec(&num_delegations);
969 nfs4_put_stid(struct nfs4_stid *s)
971 struct nfs4_file *fp = s->sc_file;
972 struct nfs4_client *clp = s->sc_client;
974 might_lock(&clp->cl_lock);
976 if (!refcount_dec_and_lock(&s->sc_count, &clp->cl_lock)) {
977 wake_up_all(&close_wq);
980 idr_remove(&clp->cl_stateids, s->sc_stateid.si_opaque.so_id);
981 nfs4_free_cpntf_statelist(clp->net, s);
982 spin_unlock(&clp->cl_lock);
989 nfs4_inc_and_copy_stateid(stateid_t *dst, struct nfs4_stid *stid)
991 stateid_t *src = &stid->sc_stateid;
993 spin_lock(&stid->sc_lock);
994 if (unlikely(++src->si_generation == 0))
995 src->si_generation = 1;
996 memcpy(dst, src, sizeof(*dst));
997 spin_unlock(&stid->sc_lock);
1000 static void put_deleg_file(struct nfs4_file *fp)
1002 struct nfsd_file *nf = NULL;
1004 spin_lock(&fp->fi_lock);
1005 if (--fp->fi_delegees == 0)
1006 swap(nf, fp->fi_deleg_file);
1007 spin_unlock(&fp->fi_lock);
1013 static void nfs4_unlock_deleg_lease(struct nfs4_delegation *dp)
1015 struct nfs4_file *fp = dp->dl_stid.sc_file;
1016 struct nfsd_file *nf = fp->fi_deleg_file;
1018 WARN_ON_ONCE(!fp->fi_delegees);
1020 vfs_setlease(nf->nf_file, F_UNLCK, NULL, (void **)&dp);
1024 static void destroy_unhashed_deleg(struct nfs4_delegation *dp)
1026 put_clnt_odstate(dp->dl_clnt_odstate);
1027 nfs4_unlock_deleg_lease(dp);
1028 nfs4_put_stid(&dp->dl_stid);
1031 void nfs4_unhash_stid(struct nfs4_stid *s)
1037 * nfs4_delegation_exists - Discover if this delegation already exists
1038 * @clp: a pointer to the nfs4_client we're granting a delegation to
1039 * @fp: a pointer to the nfs4_file we're granting a delegation on
1042 * On success: true iff an existing delegation is found
1046 nfs4_delegation_exists(struct nfs4_client *clp, struct nfs4_file *fp)
1048 struct nfs4_delegation *searchdp = NULL;
1049 struct nfs4_client *searchclp = NULL;
1051 lockdep_assert_held(&state_lock);
1052 lockdep_assert_held(&fp->fi_lock);
1054 list_for_each_entry(searchdp, &fp->fi_delegations, dl_perfile) {
1055 searchclp = searchdp->dl_stid.sc_client;
1056 if (clp == searchclp) {
1064 * hash_delegation_locked - Add a delegation to the appropriate lists
1065 * @dp: a pointer to the nfs4_delegation we are adding.
1066 * @fp: a pointer to the nfs4_file we're granting a delegation on
1069 * On success: NULL if the delegation was successfully hashed.
1071 * On error: -EAGAIN if one was previously granted to this
1072 * nfs4_client for this nfs4_file. Delegation is not hashed.
1077 hash_delegation_locked(struct nfs4_delegation *dp, struct nfs4_file *fp)
1079 struct nfs4_client *clp = dp->dl_stid.sc_client;
1081 lockdep_assert_held(&state_lock);
1082 lockdep_assert_held(&fp->fi_lock);
1084 if (nfs4_delegation_exists(clp, fp))
1086 refcount_inc(&dp->dl_stid.sc_count);
1087 dp->dl_stid.sc_type = NFS4_DELEG_STID;
1088 list_add(&dp->dl_perfile, &fp->fi_delegations);
1089 list_add(&dp->dl_perclnt, &clp->cl_delegations);
1094 unhash_delegation_locked(struct nfs4_delegation *dp)
1096 struct nfs4_file *fp = dp->dl_stid.sc_file;
1098 lockdep_assert_held(&state_lock);
1100 if (list_empty(&dp->dl_perfile))
1103 dp->dl_stid.sc_type = NFS4_CLOSED_DELEG_STID;
1104 /* Ensure that deleg break won't try to requeue it */
1106 spin_lock(&fp->fi_lock);
1107 list_del_init(&dp->dl_perclnt);
1108 list_del_init(&dp->dl_recall_lru);
1109 list_del_init(&dp->dl_perfile);
1110 spin_unlock(&fp->fi_lock);
1114 static void destroy_delegation(struct nfs4_delegation *dp)
1118 spin_lock(&state_lock);
1119 unhashed = unhash_delegation_locked(dp);
1120 spin_unlock(&state_lock);
1122 destroy_unhashed_deleg(dp);
1125 static void revoke_delegation(struct nfs4_delegation *dp)
1127 struct nfs4_client *clp = dp->dl_stid.sc_client;
1129 WARN_ON(!list_empty(&dp->dl_recall_lru));
1131 if (clp->cl_minorversion) {
1132 dp->dl_stid.sc_type = NFS4_REVOKED_DELEG_STID;
1133 refcount_inc(&dp->dl_stid.sc_count);
1134 spin_lock(&clp->cl_lock);
1135 list_add(&dp->dl_recall_lru, &clp->cl_revoked);
1136 spin_unlock(&clp->cl_lock);
1138 destroy_unhashed_deleg(dp);
1145 static unsigned int clientid_hashval(u32 id)
1147 return id & CLIENT_HASH_MASK;
1150 static unsigned int clientstr_hashval(struct xdr_netobj name)
1152 return opaque_hashval(name.data, 8) & CLIENT_HASH_MASK;
1156 * We store the NONE, READ, WRITE, and BOTH bits separately in the
1157 * st_{access,deny}_bmap field of the stateid, in order to track not
1158 * only what share bits are currently in force, but also what
1159 * combinations of share bits previous opens have used. This allows us
1160 * to enforce the recommendation of rfc 3530 14.2.19 that the server
1161 * return an error if the client attempt to downgrade to a combination
1162 * of share bits not explicable by closing some of its previous opens.
1164 * XXX: This enforcement is actually incomplete, since we don't keep
1165 * track of access/deny bit combinations; so, e.g., we allow:
1167 * OPEN allow read, deny write
1168 * OPEN allow both, deny none
1169 * DOWNGRADE allow read, deny none
1171 * which we should reject.
1174 bmap_to_share_mode(unsigned long bmap) {
1176 unsigned int access = 0;
1178 for (i = 1; i < 4; i++) {
1179 if (test_bit(i, &bmap))
1185 /* set share access for a given stateid */
1187 set_access(u32 access, struct nfs4_ol_stateid *stp)
1189 unsigned char mask = 1 << access;
1191 WARN_ON_ONCE(access > NFS4_SHARE_ACCESS_BOTH);
1192 stp->st_access_bmap |= mask;
1195 /* clear share access for a given stateid */
1197 clear_access(u32 access, struct nfs4_ol_stateid *stp)
1199 unsigned char mask = 1 << access;
1201 WARN_ON_ONCE(access > NFS4_SHARE_ACCESS_BOTH);
1202 stp->st_access_bmap &= ~mask;
1205 /* test whether a given stateid has access */
1207 test_access(u32 access, struct nfs4_ol_stateid *stp)
1209 unsigned char mask = 1 << access;
1211 return (bool)(stp->st_access_bmap & mask);
1214 /* set share deny for a given stateid */
1216 set_deny(u32 deny, struct nfs4_ol_stateid *stp)
1218 unsigned char mask = 1 << deny;
1220 WARN_ON_ONCE(deny > NFS4_SHARE_DENY_BOTH);
1221 stp->st_deny_bmap |= mask;
1224 /* clear share deny for a given stateid */
1226 clear_deny(u32 deny, struct nfs4_ol_stateid *stp)
1228 unsigned char mask = 1 << deny;
1230 WARN_ON_ONCE(deny > NFS4_SHARE_DENY_BOTH);
1231 stp->st_deny_bmap &= ~mask;
1234 /* test whether a given stateid is denying specific access */
1236 test_deny(u32 deny, struct nfs4_ol_stateid *stp)
1238 unsigned char mask = 1 << deny;
1240 return (bool)(stp->st_deny_bmap & mask);
1243 static int nfs4_access_to_omode(u32 access)
1245 switch (access & NFS4_SHARE_ACCESS_BOTH) {
1246 case NFS4_SHARE_ACCESS_READ:
1248 case NFS4_SHARE_ACCESS_WRITE:
1250 case NFS4_SHARE_ACCESS_BOTH:
1258 * A stateid that had a deny mode associated with it is being released
1259 * or downgraded. Recalculate the deny mode on the file.
1262 recalculate_deny_mode(struct nfs4_file *fp)
1264 struct nfs4_ol_stateid *stp;
1266 spin_lock(&fp->fi_lock);
1267 fp->fi_share_deny = 0;
1268 list_for_each_entry(stp, &fp->fi_stateids, st_perfile)
1269 fp->fi_share_deny |= bmap_to_share_mode(stp->st_deny_bmap);
1270 spin_unlock(&fp->fi_lock);
1274 reset_union_bmap_deny(u32 deny, struct nfs4_ol_stateid *stp)
1277 bool change = false;
1279 for (i = 1; i < 4; i++) {
1280 if ((i & deny) != i) {
1286 /* Recalculate per-file deny mode if there was a change */
1288 recalculate_deny_mode(stp->st_stid.sc_file);
1291 /* release all access and file references for a given stateid */
1293 release_all_access(struct nfs4_ol_stateid *stp)
1296 struct nfs4_file *fp = stp->st_stid.sc_file;
1298 if (fp && stp->st_deny_bmap != 0)
1299 recalculate_deny_mode(fp);
1301 for (i = 1; i < 4; i++) {
1302 if (test_access(i, stp))
1303 nfs4_file_put_access(stp->st_stid.sc_file, i);
1304 clear_access(i, stp);
1308 static inline void nfs4_free_stateowner(struct nfs4_stateowner *sop)
1310 kfree(sop->so_owner.data);
1311 sop->so_ops->so_free(sop);
1314 static void nfs4_put_stateowner(struct nfs4_stateowner *sop)
1316 struct nfs4_client *clp = sop->so_client;
1318 might_lock(&clp->cl_lock);
1320 if (!atomic_dec_and_lock(&sop->so_count, &clp->cl_lock))
1322 sop->so_ops->so_unhash(sop);
1323 spin_unlock(&clp->cl_lock);
1324 nfs4_free_stateowner(sop);
1328 nfs4_ol_stateid_unhashed(const struct nfs4_ol_stateid *stp)
1330 return list_empty(&stp->st_perfile);
1333 static bool unhash_ol_stateid(struct nfs4_ol_stateid *stp)
1335 struct nfs4_file *fp = stp->st_stid.sc_file;
1337 lockdep_assert_held(&stp->st_stateowner->so_client->cl_lock);
1339 if (list_empty(&stp->st_perfile))
1342 spin_lock(&fp->fi_lock);
1343 list_del_init(&stp->st_perfile);
1344 spin_unlock(&fp->fi_lock);
1345 list_del(&stp->st_perstateowner);
1349 static void nfs4_free_ol_stateid(struct nfs4_stid *stid)
1351 struct nfs4_ol_stateid *stp = openlockstateid(stid);
1353 put_clnt_odstate(stp->st_clnt_odstate);
1354 release_all_access(stp);
1355 if (stp->st_stateowner)
1356 nfs4_put_stateowner(stp->st_stateowner);
1357 kmem_cache_free(stateid_slab, stid);
1360 static void nfs4_free_lock_stateid(struct nfs4_stid *stid)
1362 struct nfs4_ol_stateid *stp = openlockstateid(stid);
1363 struct nfs4_lockowner *lo = lockowner(stp->st_stateowner);
1364 struct nfsd_file *nf;
1366 nf = find_any_file(stp->st_stid.sc_file);
1368 get_file(nf->nf_file);
1369 filp_close(nf->nf_file, (fl_owner_t)lo);
1372 nfs4_free_ol_stateid(stid);
1376 * Put the persistent reference to an already unhashed generic stateid, while
1377 * holding the cl_lock. If it's the last reference, then put it onto the
1378 * reaplist for later destruction.
1380 static void put_ol_stateid_locked(struct nfs4_ol_stateid *stp,
1381 struct list_head *reaplist)
1383 struct nfs4_stid *s = &stp->st_stid;
1384 struct nfs4_client *clp = s->sc_client;
1386 lockdep_assert_held(&clp->cl_lock);
1388 WARN_ON_ONCE(!list_empty(&stp->st_locks));
1390 if (!refcount_dec_and_test(&s->sc_count)) {
1391 wake_up_all(&close_wq);
1395 idr_remove(&clp->cl_stateids, s->sc_stateid.si_opaque.so_id);
1396 list_add(&stp->st_locks, reaplist);
1399 static bool unhash_lock_stateid(struct nfs4_ol_stateid *stp)
1401 lockdep_assert_held(&stp->st_stid.sc_client->cl_lock);
1403 if (!unhash_ol_stateid(stp))
1405 list_del_init(&stp->st_locks);
1406 nfs4_unhash_stid(&stp->st_stid);
1410 static void release_lock_stateid(struct nfs4_ol_stateid *stp)
1412 struct nfs4_client *clp = stp->st_stid.sc_client;
1415 spin_lock(&clp->cl_lock);
1416 unhashed = unhash_lock_stateid(stp);
1417 spin_unlock(&clp->cl_lock);
1419 nfs4_put_stid(&stp->st_stid);
1422 static void unhash_lockowner_locked(struct nfs4_lockowner *lo)
1424 struct nfs4_client *clp = lo->lo_owner.so_client;
1426 lockdep_assert_held(&clp->cl_lock);
1428 list_del_init(&lo->lo_owner.so_strhash);
1432 * Free a list of generic stateids that were collected earlier after being
1436 free_ol_stateid_reaplist(struct list_head *reaplist)
1438 struct nfs4_ol_stateid *stp;
1439 struct nfs4_file *fp;
1443 while (!list_empty(reaplist)) {
1444 stp = list_first_entry(reaplist, struct nfs4_ol_stateid,
1446 list_del(&stp->st_locks);
1447 fp = stp->st_stid.sc_file;
1448 stp->st_stid.sc_free(&stp->st_stid);
1454 static void release_open_stateid_locks(struct nfs4_ol_stateid *open_stp,
1455 struct list_head *reaplist)
1457 struct nfs4_ol_stateid *stp;
1459 lockdep_assert_held(&open_stp->st_stid.sc_client->cl_lock);
1461 while (!list_empty(&open_stp->st_locks)) {
1462 stp = list_entry(open_stp->st_locks.next,
1463 struct nfs4_ol_stateid, st_locks);
1464 WARN_ON(!unhash_lock_stateid(stp));
1465 put_ol_stateid_locked(stp, reaplist);
1469 static bool unhash_open_stateid(struct nfs4_ol_stateid *stp,
1470 struct list_head *reaplist)
1472 lockdep_assert_held(&stp->st_stid.sc_client->cl_lock);
1474 if (!unhash_ol_stateid(stp))
1476 release_open_stateid_locks(stp, reaplist);
1480 static void release_open_stateid(struct nfs4_ol_stateid *stp)
1482 LIST_HEAD(reaplist);
1484 spin_lock(&stp->st_stid.sc_client->cl_lock);
1485 if (unhash_open_stateid(stp, &reaplist))
1486 put_ol_stateid_locked(stp, &reaplist);
1487 spin_unlock(&stp->st_stid.sc_client->cl_lock);
1488 free_ol_stateid_reaplist(&reaplist);
1491 static void unhash_openowner_locked(struct nfs4_openowner *oo)
1493 struct nfs4_client *clp = oo->oo_owner.so_client;
1495 lockdep_assert_held(&clp->cl_lock);
1497 list_del_init(&oo->oo_owner.so_strhash);
1498 list_del_init(&oo->oo_perclient);
1501 static void release_last_closed_stateid(struct nfs4_openowner *oo)
1503 struct nfsd_net *nn = net_generic(oo->oo_owner.so_client->net,
1505 struct nfs4_ol_stateid *s;
1507 spin_lock(&nn->client_lock);
1508 s = oo->oo_last_closed_stid;
1510 list_del_init(&oo->oo_close_lru);
1511 oo->oo_last_closed_stid = NULL;
1513 spin_unlock(&nn->client_lock);
1515 nfs4_put_stid(&s->st_stid);
1518 static void release_openowner(struct nfs4_openowner *oo)
1520 struct nfs4_ol_stateid *stp;
1521 struct nfs4_client *clp = oo->oo_owner.so_client;
1522 struct list_head reaplist;
1524 INIT_LIST_HEAD(&reaplist);
1526 spin_lock(&clp->cl_lock);
1527 unhash_openowner_locked(oo);
1528 while (!list_empty(&oo->oo_owner.so_stateids)) {
1529 stp = list_first_entry(&oo->oo_owner.so_stateids,
1530 struct nfs4_ol_stateid, st_perstateowner);
1531 if (unhash_open_stateid(stp, &reaplist))
1532 put_ol_stateid_locked(stp, &reaplist);
1534 spin_unlock(&clp->cl_lock);
1535 free_ol_stateid_reaplist(&reaplist);
1536 release_last_closed_stateid(oo);
1537 nfs4_put_stateowner(&oo->oo_owner);
1541 hash_sessionid(struct nfs4_sessionid *sessionid)
1543 struct nfsd4_sessionid *sid = (struct nfsd4_sessionid *)sessionid;
1545 return sid->sequence % SESSION_HASH_SIZE;
1548 #ifdef CONFIG_SUNRPC_DEBUG
1550 dump_sessionid(const char *fn, struct nfs4_sessionid *sessionid)
1552 u32 *ptr = (u32 *)(&sessionid->data[0]);
1553 dprintk("%s: %u:%u:%u:%u\n", fn, ptr[0], ptr[1], ptr[2], ptr[3]);
1557 dump_sessionid(const char *fn, struct nfs4_sessionid *sessionid)
1563 * Bump the seqid on cstate->replay_owner, and clear replay_owner if it
1564 * won't be used for replay.
1566 void nfsd4_bump_seqid(struct nfsd4_compound_state *cstate, __be32 nfserr)
1568 struct nfs4_stateowner *so = cstate->replay_owner;
1570 if (nfserr == nfserr_replay_me)
1573 if (!seqid_mutating_err(ntohl(nfserr))) {
1574 nfsd4_cstate_clear_replay(cstate);
1579 if (so->so_is_open_owner)
1580 release_last_closed_stateid(openowner(so));
1586 gen_sessionid(struct nfsd4_session *ses)
1588 struct nfs4_client *clp = ses->se_client;
1589 struct nfsd4_sessionid *sid;
1591 sid = (struct nfsd4_sessionid *)ses->se_sessionid.data;
1592 sid->clientid = clp->cl_clientid;
1593 sid->sequence = current_sessionid++;
1598 * The protocol defines ca_maxresponssize_cached to include the size of
1599 * the rpc header, but all we need to cache is the data starting after
1600 * the end of the initial SEQUENCE operation--the rest we regenerate
1601 * each time. Therefore we can advertise a ca_maxresponssize_cached
1602 * value that is the number of bytes in our cache plus a few additional
1603 * bytes. In order to stay on the safe side, and not promise more than
1604 * we can cache, those additional bytes must be the minimum possible: 24
1605 * bytes of rpc header (xid through accept state, with AUTH_NULL
1606 * verifier), 12 for the compound header (with zero-length tag), and 44
1607 * for the SEQUENCE op response:
1609 #define NFSD_MIN_HDR_SEQ_SZ (24 + 12 + 44)
1612 free_session_slots(struct nfsd4_session *ses)
1616 for (i = 0; i < ses->se_fchannel.maxreqs; i++) {
1617 free_svc_cred(&ses->se_slots[i]->sl_cred);
1618 kfree(ses->se_slots[i]);
1623 * We don't actually need to cache the rpc and session headers, so we
1624 * can allocate a little less for each slot:
1626 static inline u32 slot_bytes(struct nfsd4_channel_attrs *ca)
1630 if (ca->maxresp_cached < NFSD_MIN_HDR_SEQ_SZ)
1633 size = ca->maxresp_cached - NFSD_MIN_HDR_SEQ_SZ;
1634 return size + sizeof(struct nfsd4_slot);
1638 * XXX: If we run out of reserved DRC memory we could (up to a point)
1639 * re-negotiate active sessions and reduce their slot usage to make
1640 * room for new connections. For now we just fail the create session.
1642 static u32 nfsd4_get_drc_mem(struct nfsd4_channel_attrs *ca, struct nfsd_net *nn)
1644 u32 slotsize = slot_bytes(ca);
1645 u32 num = ca->maxreqs;
1646 unsigned long avail, total_avail;
1647 unsigned int scale_factor;
1649 spin_lock(&nfsd_drc_lock);
1650 if (nfsd_drc_max_mem > nfsd_drc_mem_used)
1651 total_avail = nfsd_drc_max_mem - nfsd_drc_mem_used;
1653 /* We have handed out more space than we chose in
1654 * set_max_drc() to allow. That isn't really a
1655 * problem as long as that doesn't make us think we
1656 * have lots more due to integer overflow.
1659 avail = min((unsigned long)NFSD_MAX_MEM_PER_SESSION, total_avail);
1661 * Never use more than a fraction of the remaining memory,
1662 * unless it's the only way to give this client a slot.
1663 * The chosen fraction is either 1/8 or 1/number of threads,
1664 * whichever is smaller. This ensures there are adequate
1665 * slots to support multiple clients per thread.
1666 * Give the client one slot even if that would require
1667 * over-allocation--it is better than failure.
1669 scale_factor = max_t(unsigned int, 8, nn->nfsd_serv->sv_nrthreads);
1671 avail = clamp_t(unsigned long, avail, slotsize,
1672 total_avail/scale_factor);
1673 num = min_t(int, num, avail / slotsize);
1674 num = max_t(int, num, 1);
1675 nfsd_drc_mem_used += num * slotsize;
1676 spin_unlock(&nfsd_drc_lock);
1681 static void nfsd4_put_drc_mem(struct nfsd4_channel_attrs *ca)
1683 int slotsize = slot_bytes(ca);
1685 spin_lock(&nfsd_drc_lock);
1686 nfsd_drc_mem_used -= slotsize * ca->maxreqs;
1687 spin_unlock(&nfsd_drc_lock);
1690 static struct nfsd4_session *alloc_session(struct nfsd4_channel_attrs *fattrs,
1691 struct nfsd4_channel_attrs *battrs)
1693 int numslots = fattrs->maxreqs;
1694 int slotsize = slot_bytes(fattrs);
1695 struct nfsd4_session *new;
1698 BUILD_BUG_ON(NFSD_MAX_SLOTS_PER_SESSION * sizeof(struct nfsd4_slot *)
1699 + sizeof(struct nfsd4_session) > PAGE_SIZE);
1700 mem = numslots * sizeof(struct nfsd4_slot *);
1702 new = kzalloc(sizeof(*new) + mem, GFP_KERNEL);
1705 /* allocate each struct nfsd4_slot and data cache in one piece */
1706 for (i = 0; i < numslots; i++) {
1707 new->se_slots[i] = kzalloc(slotsize, GFP_KERNEL);
1708 if (!new->se_slots[i])
1712 memcpy(&new->se_fchannel, fattrs, sizeof(struct nfsd4_channel_attrs));
1713 memcpy(&new->se_bchannel, battrs, sizeof(struct nfsd4_channel_attrs));
1718 kfree(new->se_slots[i]);
1723 static void free_conn(struct nfsd4_conn *c)
1725 svc_xprt_put(c->cn_xprt);
1729 static void nfsd4_conn_lost(struct svc_xpt_user *u)
1731 struct nfsd4_conn *c = container_of(u, struct nfsd4_conn, cn_xpt_user);
1732 struct nfs4_client *clp = c->cn_session->se_client;
1734 spin_lock(&clp->cl_lock);
1735 if (!list_empty(&c->cn_persession)) {
1736 list_del(&c->cn_persession);
1739 nfsd4_probe_callback(clp);
1740 spin_unlock(&clp->cl_lock);
1743 static struct nfsd4_conn *alloc_conn(struct svc_rqst *rqstp, u32 flags)
1745 struct nfsd4_conn *conn;
1747 conn = kmalloc(sizeof(struct nfsd4_conn), GFP_KERNEL);
1750 svc_xprt_get(rqstp->rq_xprt);
1751 conn->cn_xprt = rqstp->rq_xprt;
1752 conn->cn_flags = flags;
1753 INIT_LIST_HEAD(&conn->cn_xpt_user.list);
1757 static void __nfsd4_hash_conn(struct nfsd4_conn *conn, struct nfsd4_session *ses)
1759 conn->cn_session = ses;
1760 list_add(&conn->cn_persession, &ses->se_conns);
1763 static void nfsd4_hash_conn(struct nfsd4_conn *conn, struct nfsd4_session *ses)
1765 struct nfs4_client *clp = ses->se_client;
1767 spin_lock(&clp->cl_lock);
1768 __nfsd4_hash_conn(conn, ses);
1769 spin_unlock(&clp->cl_lock);
1772 static int nfsd4_register_conn(struct nfsd4_conn *conn)
1774 conn->cn_xpt_user.callback = nfsd4_conn_lost;
1775 return register_xpt_user(conn->cn_xprt, &conn->cn_xpt_user);
1778 static void nfsd4_init_conn(struct svc_rqst *rqstp, struct nfsd4_conn *conn, struct nfsd4_session *ses)
1782 nfsd4_hash_conn(conn, ses);
1783 ret = nfsd4_register_conn(conn);
1785 /* oops; xprt is already down: */
1786 nfsd4_conn_lost(&conn->cn_xpt_user);
1787 /* We may have gained or lost a callback channel: */
1788 nfsd4_probe_callback_sync(ses->se_client);
1791 static struct nfsd4_conn *alloc_conn_from_crses(struct svc_rqst *rqstp, struct nfsd4_create_session *cses)
1793 u32 dir = NFS4_CDFC4_FORE;
1795 if (cses->flags & SESSION4_BACK_CHAN)
1796 dir |= NFS4_CDFC4_BACK;
1797 return alloc_conn(rqstp, dir);
1800 /* must be called under client_lock */
1801 static void nfsd4_del_conns(struct nfsd4_session *s)
1803 struct nfs4_client *clp = s->se_client;
1804 struct nfsd4_conn *c;
1806 spin_lock(&clp->cl_lock);
1807 while (!list_empty(&s->se_conns)) {
1808 c = list_first_entry(&s->se_conns, struct nfsd4_conn, cn_persession);
1809 list_del_init(&c->cn_persession);
1810 spin_unlock(&clp->cl_lock);
1812 unregister_xpt_user(c->cn_xprt, &c->cn_xpt_user);
1815 spin_lock(&clp->cl_lock);
1817 spin_unlock(&clp->cl_lock);
1820 static void __free_session(struct nfsd4_session *ses)
1822 free_session_slots(ses);
1826 static void free_session(struct nfsd4_session *ses)
1828 nfsd4_del_conns(ses);
1829 nfsd4_put_drc_mem(&ses->se_fchannel);
1830 __free_session(ses);
1833 static void init_session(struct svc_rqst *rqstp, struct nfsd4_session *new, struct nfs4_client *clp, struct nfsd4_create_session *cses)
1836 struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
1838 new->se_client = clp;
1841 INIT_LIST_HEAD(&new->se_conns);
1843 new->se_cb_seq_nr = 1;
1844 new->se_flags = cses->flags;
1845 new->se_cb_prog = cses->callback_prog;
1846 new->se_cb_sec = cses->cb_sec;
1847 atomic_set(&new->se_ref, 0);
1848 idx = hash_sessionid(&new->se_sessionid);
1849 list_add(&new->se_hash, &nn->sessionid_hashtbl[idx]);
1850 spin_lock(&clp->cl_lock);
1851 list_add(&new->se_perclnt, &clp->cl_sessions);
1852 spin_unlock(&clp->cl_lock);
1855 struct sockaddr *sa = svc_addr(rqstp);
1857 * This is a little silly; with sessions there's no real
1858 * use for the callback address. Use the peer address
1859 * as a reasonable default for now, but consider fixing
1860 * the rpc client not to require an address in the
1863 rpc_copy_addr((struct sockaddr *)&clp->cl_cb_conn.cb_addr, sa);
1864 clp->cl_cb_conn.cb_addrlen = svc_addr_len(sa);
1868 /* caller must hold client_lock */
1869 static struct nfsd4_session *
1870 __find_in_sessionid_hashtbl(struct nfs4_sessionid *sessionid, struct net *net)
1872 struct nfsd4_session *elem;
1874 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
1876 lockdep_assert_held(&nn->client_lock);
1878 dump_sessionid(__func__, sessionid);
1879 idx = hash_sessionid(sessionid);
1880 /* Search in the appropriate list */
1881 list_for_each_entry(elem, &nn->sessionid_hashtbl[idx], se_hash) {
1882 if (!memcmp(elem->se_sessionid.data, sessionid->data,
1883 NFS4_MAX_SESSIONID_LEN)) {
1888 dprintk("%s: session not found\n", __func__);
1892 static struct nfsd4_session *
1893 find_in_sessionid_hashtbl(struct nfs4_sessionid *sessionid, struct net *net,
1896 struct nfsd4_session *session;
1897 __be32 status = nfserr_badsession;
1899 session = __find_in_sessionid_hashtbl(sessionid, net);
1902 status = nfsd4_get_session_locked(session);
1910 /* caller must hold client_lock */
1912 unhash_session(struct nfsd4_session *ses)
1914 struct nfs4_client *clp = ses->se_client;
1915 struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
1917 lockdep_assert_held(&nn->client_lock);
1919 list_del(&ses->se_hash);
1920 spin_lock(&ses->se_client->cl_lock);
1921 list_del(&ses->se_perclnt);
1922 spin_unlock(&ses->se_client->cl_lock);
1925 /* SETCLIENTID and SETCLIENTID_CONFIRM Helper functions */
1927 STALE_CLIENTID(clientid_t *clid, struct nfsd_net *nn)
1930 * We're assuming the clid was not given out from a boot
1931 * precisely 2^32 (about 136 years) before this one. That seems
1932 * a safe assumption:
1934 if (clid->cl_boot == (u32)nn->boot_time)
1936 trace_nfsd_clid_stale(clid);
1941 * XXX Should we use a slab cache ?
1942 * This type of memory management is somewhat inefficient, but we use it
1943 * anyway since SETCLIENTID is not a common operation.
1945 static struct nfs4_client *alloc_client(struct xdr_netobj name)
1947 struct nfs4_client *clp;
1950 clp = kmem_cache_zalloc(client_slab, GFP_KERNEL);
1953 xdr_netobj_dup(&clp->cl_name, &name, GFP_KERNEL);
1954 if (clp->cl_name.data == NULL)
1956 clp->cl_ownerstr_hashtbl = kmalloc_array(OWNER_HASH_SIZE,
1957 sizeof(struct list_head),
1959 if (!clp->cl_ownerstr_hashtbl)
1960 goto err_no_hashtbl;
1961 for (i = 0; i < OWNER_HASH_SIZE; i++)
1962 INIT_LIST_HEAD(&clp->cl_ownerstr_hashtbl[i]);
1963 INIT_LIST_HEAD(&clp->cl_sessions);
1964 idr_init(&clp->cl_stateids);
1965 atomic_set(&clp->cl_rpc_users, 0);
1966 clp->cl_cb_state = NFSD4_CB_UNKNOWN;
1967 INIT_LIST_HEAD(&clp->cl_idhash);
1968 INIT_LIST_HEAD(&clp->cl_openowners);
1969 INIT_LIST_HEAD(&clp->cl_delegations);
1970 INIT_LIST_HEAD(&clp->cl_lru);
1971 INIT_LIST_HEAD(&clp->cl_revoked);
1972 #ifdef CONFIG_NFSD_PNFS
1973 INIT_LIST_HEAD(&clp->cl_lo_states);
1975 INIT_LIST_HEAD(&clp->async_copies);
1976 spin_lock_init(&clp->async_lock);
1977 spin_lock_init(&clp->cl_lock);
1978 rpc_init_wait_queue(&clp->cl_cb_waitq, "Backchannel slot table");
1981 kfree(clp->cl_name.data);
1983 kmem_cache_free(client_slab, clp);
1987 static void __free_client(struct kref *k)
1989 struct nfsdfs_client *c = container_of(k, struct nfsdfs_client, cl_ref);
1990 struct nfs4_client *clp = container_of(c, struct nfs4_client, cl_nfsdfs);
1992 free_svc_cred(&clp->cl_cred);
1993 kfree(clp->cl_ownerstr_hashtbl);
1994 kfree(clp->cl_name.data);
1995 kfree(clp->cl_nii_domain.data);
1996 kfree(clp->cl_nii_name.data);
1997 idr_destroy(&clp->cl_stateids);
1998 kmem_cache_free(client_slab, clp);
2001 static void drop_client(struct nfs4_client *clp)
2003 kref_put(&clp->cl_nfsdfs.cl_ref, __free_client);
2007 free_client(struct nfs4_client *clp)
2009 while (!list_empty(&clp->cl_sessions)) {
2010 struct nfsd4_session *ses;
2011 ses = list_entry(clp->cl_sessions.next, struct nfsd4_session,
2013 list_del(&ses->se_perclnt);
2014 WARN_ON_ONCE(atomic_read(&ses->se_ref));
2017 rpc_destroy_wait_queue(&clp->cl_cb_waitq);
2018 if (clp->cl_nfsd_dentry) {
2019 nfsd_client_rmdir(clp->cl_nfsd_dentry);
2020 clp->cl_nfsd_dentry = NULL;
2021 wake_up_all(&expiry_wq);
2026 /* must be called under the client_lock */
2028 unhash_client_locked(struct nfs4_client *clp)
2030 struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
2031 struct nfsd4_session *ses;
2033 lockdep_assert_held(&nn->client_lock);
2035 /* Mark the client as expired! */
2037 /* Make it invisible */
2038 if (!list_empty(&clp->cl_idhash)) {
2039 list_del_init(&clp->cl_idhash);
2040 if (test_bit(NFSD4_CLIENT_CONFIRMED, &clp->cl_flags))
2041 rb_erase(&clp->cl_namenode, &nn->conf_name_tree);
2043 rb_erase(&clp->cl_namenode, &nn->unconf_name_tree);
2045 list_del_init(&clp->cl_lru);
2046 spin_lock(&clp->cl_lock);
2047 list_for_each_entry(ses, &clp->cl_sessions, se_perclnt)
2048 list_del_init(&ses->se_hash);
2049 spin_unlock(&clp->cl_lock);
2053 unhash_client(struct nfs4_client *clp)
2055 struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
2057 spin_lock(&nn->client_lock);
2058 unhash_client_locked(clp);
2059 spin_unlock(&nn->client_lock);
2062 static __be32 mark_client_expired_locked(struct nfs4_client *clp)
2064 if (atomic_read(&clp->cl_rpc_users))
2065 return nfserr_jukebox;
2066 unhash_client_locked(clp);
2071 __destroy_client(struct nfs4_client *clp)
2074 struct nfs4_openowner *oo;
2075 struct nfs4_delegation *dp;
2076 struct list_head reaplist;
2078 INIT_LIST_HEAD(&reaplist);
2079 spin_lock(&state_lock);
2080 while (!list_empty(&clp->cl_delegations)) {
2081 dp = list_entry(clp->cl_delegations.next, struct nfs4_delegation, dl_perclnt);
2082 WARN_ON(!unhash_delegation_locked(dp));
2083 list_add(&dp->dl_recall_lru, &reaplist);
2085 spin_unlock(&state_lock);
2086 while (!list_empty(&reaplist)) {
2087 dp = list_entry(reaplist.next, struct nfs4_delegation, dl_recall_lru);
2088 list_del_init(&dp->dl_recall_lru);
2089 destroy_unhashed_deleg(dp);
2091 while (!list_empty(&clp->cl_revoked)) {
2092 dp = list_entry(clp->cl_revoked.next, struct nfs4_delegation, dl_recall_lru);
2093 list_del_init(&dp->dl_recall_lru);
2094 nfs4_put_stid(&dp->dl_stid);
2096 while (!list_empty(&clp->cl_openowners)) {
2097 oo = list_entry(clp->cl_openowners.next, struct nfs4_openowner, oo_perclient);
2098 nfs4_get_stateowner(&oo->oo_owner);
2099 release_openowner(oo);
2101 for (i = 0; i < OWNER_HASH_SIZE; i++) {
2102 struct nfs4_stateowner *so, *tmp;
2104 list_for_each_entry_safe(so, tmp, &clp->cl_ownerstr_hashtbl[i],
2106 /* Should be no openowners at this point */
2107 WARN_ON_ONCE(so->so_is_open_owner);
2108 remove_blocked_locks(lockowner(so));
2111 nfsd4_return_all_client_layouts(clp);
2112 nfsd4_shutdown_copy(clp);
2113 nfsd4_shutdown_callback(clp);
2114 if (clp->cl_cb_conn.cb_xprt)
2115 svc_xprt_put(clp->cl_cb_conn.cb_xprt);
2117 wake_up_all(&expiry_wq);
2121 destroy_client(struct nfs4_client *clp)
2124 __destroy_client(clp);
2127 static void inc_reclaim_complete(struct nfs4_client *clp)
2129 struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
2131 if (!nn->track_reclaim_completes)
2133 if (!nfsd4_find_reclaim_client(clp->cl_name, nn))
2135 if (atomic_inc_return(&nn->nr_reclaim_complete) ==
2136 nn->reclaim_str_hashtbl_size) {
2137 printk(KERN_INFO "NFSD: all clients done reclaiming, ending NFSv4 grace period (net %x)\n",
2139 nfsd4_end_grace(nn);
2143 static void expire_client(struct nfs4_client *clp)
2146 nfsd4_client_record_remove(clp);
2147 __destroy_client(clp);
2150 static void copy_verf(struct nfs4_client *target, nfs4_verifier *source)
2152 memcpy(target->cl_verifier.data, source->data,
2153 sizeof(target->cl_verifier.data));
2156 static void copy_clid(struct nfs4_client *target, struct nfs4_client *source)
2158 target->cl_clientid.cl_boot = source->cl_clientid.cl_boot;
2159 target->cl_clientid.cl_id = source->cl_clientid.cl_id;
2162 static int copy_cred(struct svc_cred *target, struct svc_cred *source)
2164 target->cr_principal = kstrdup(source->cr_principal, GFP_KERNEL);
2165 target->cr_raw_principal = kstrdup(source->cr_raw_principal,
2167 target->cr_targ_princ = kstrdup(source->cr_targ_princ, GFP_KERNEL);
2168 if ((source->cr_principal && !target->cr_principal) ||
2169 (source->cr_raw_principal && !target->cr_raw_principal) ||
2170 (source->cr_targ_princ && !target->cr_targ_princ))
2173 target->cr_flavor = source->cr_flavor;
2174 target->cr_uid = source->cr_uid;
2175 target->cr_gid = source->cr_gid;
2176 target->cr_group_info = source->cr_group_info;
2177 get_group_info(target->cr_group_info);
2178 target->cr_gss_mech = source->cr_gss_mech;
2179 if (source->cr_gss_mech)
2180 gss_mech_get(source->cr_gss_mech);
2185 compare_blob(const struct xdr_netobj *o1, const struct xdr_netobj *o2)
2187 if (o1->len < o2->len)
2189 if (o1->len > o2->len)
2191 return memcmp(o1->data, o2->data, o1->len);
2195 same_verf(nfs4_verifier *v1, nfs4_verifier *v2)
2197 return 0 == memcmp(v1->data, v2->data, sizeof(v1->data));
2201 same_clid(clientid_t *cl1, clientid_t *cl2)
2203 return (cl1->cl_boot == cl2->cl_boot) && (cl1->cl_id == cl2->cl_id);
2206 static bool groups_equal(struct group_info *g1, struct group_info *g2)
2210 if (g1->ngroups != g2->ngroups)
2212 for (i=0; i<g1->ngroups; i++)
2213 if (!gid_eq(g1->gid[i], g2->gid[i]))
2219 * RFC 3530 language requires clid_inuse be returned when the
2220 * "principal" associated with a requests differs from that previously
2221 * used. We use uid, gid's, and gss principal string as our best
2222 * approximation. We also don't want to allow non-gss use of a client
2223 * established using gss: in theory cr_principal should catch that
2224 * change, but in practice cr_principal can be null even in the gss case
2225 * since gssd doesn't always pass down a principal string.
2227 static bool is_gss_cred(struct svc_cred *cr)
2229 /* Is cr_flavor one of the gss "pseudoflavors"?: */
2230 return (cr->cr_flavor > RPC_AUTH_MAXFLAVOR);
2235 same_creds(struct svc_cred *cr1, struct svc_cred *cr2)
2237 if ((is_gss_cred(cr1) != is_gss_cred(cr2))
2238 || (!uid_eq(cr1->cr_uid, cr2->cr_uid))
2239 || (!gid_eq(cr1->cr_gid, cr2->cr_gid))
2240 || !groups_equal(cr1->cr_group_info, cr2->cr_group_info))
2242 /* XXX: check that cr_targ_princ fields match ? */
2243 if (cr1->cr_principal == cr2->cr_principal)
2245 if (!cr1->cr_principal || !cr2->cr_principal)
2247 return 0 == strcmp(cr1->cr_principal, cr2->cr_principal);
2250 static bool svc_rqst_integrity_protected(struct svc_rqst *rqstp)
2252 struct svc_cred *cr = &rqstp->rq_cred;
2255 if (!cr->cr_gss_mech)
2257 service = gss_pseudoflavor_to_service(cr->cr_gss_mech, cr->cr_flavor);
2258 return service == RPC_GSS_SVC_INTEGRITY ||
2259 service == RPC_GSS_SVC_PRIVACY;
2262 bool nfsd4_mach_creds_match(struct nfs4_client *cl, struct svc_rqst *rqstp)
2264 struct svc_cred *cr = &rqstp->rq_cred;
2266 if (!cl->cl_mach_cred)
2268 if (cl->cl_cred.cr_gss_mech != cr->cr_gss_mech)
2270 if (!svc_rqst_integrity_protected(rqstp))
2272 if (cl->cl_cred.cr_raw_principal)
2273 return 0 == strcmp(cl->cl_cred.cr_raw_principal,
2274 cr->cr_raw_principal);
2275 if (!cr->cr_principal)
2277 return 0 == strcmp(cl->cl_cred.cr_principal, cr->cr_principal);
2280 static void gen_confirm(struct nfs4_client *clp, struct nfsd_net *nn)
2285 * This is opaque to client, so no need to byte-swap. Use
2286 * __force to keep sparse happy
2288 verf[0] = (__force __be32)(u32)ktime_get_real_seconds();
2289 verf[1] = (__force __be32)nn->clverifier_counter++;
2290 memcpy(clp->cl_confirm.data, verf, sizeof(clp->cl_confirm.data));
2293 static void gen_clid(struct nfs4_client *clp, struct nfsd_net *nn)
2295 clp->cl_clientid.cl_boot = (u32)nn->boot_time;
2296 clp->cl_clientid.cl_id = nn->clientid_counter++;
2297 gen_confirm(clp, nn);
2300 static struct nfs4_stid *
2301 find_stateid_locked(struct nfs4_client *cl, stateid_t *t)
2303 struct nfs4_stid *ret;
2305 ret = idr_find(&cl->cl_stateids, t->si_opaque.so_id);
2306 if (!ret || !ret->sc_type)
2311 static struct nfs4_stid *
2312 find_stateid_by_type(struct nfs4_client *cl, stateid_t *t, char typemask)
2314 struct nfs4_stid *s;
2316 spin_lock(&cl->cl_lock);
2317 s = find_stateid_locked(cl, t);
2319 if (typemask & s->sc_type)
2320 refcount_inc(&s->sc_count);
2324 spin_unlock(&cl->cl_lock);
2328 static struct nfs4_client *get_nfsdfs_clp(struct inode *inode)
2330 struct nfsdfs_client *nc;
2331 nc = get_nfsdfs_client(inode);
2334 return container_of(nc, struct nfs4_client, cl_nfsdfs);
2337 static void seq_quote_mem(struct seq_file *m, char *data, int len)
2339 seq_printf(m, "\"");
2340 seq_escape_mem_ascii(m, data, len);
2341 seq_printf(m, "\"");
2344 static int client_info_show(struct seq_file *m, void *v)
2346 struct inode *inode = m->private;
2347 struct nfs4_client *clp;
2350 clp = get_nfsdfs_clp(inode);
2353 memcpy(&clid, &clp->cl_clientid, sizeof(clid));
2354 seq_printf(m, "clientid: 0x%llx\n", clid);
2355 seq_printf(m, "address: \"%pISpc\"\n", (struct sockaddr *)&clp->cl_addr);
2356 if (test_bit(NFSD4_CLIENT_CONFIRMED, &clp->cl_flags))
2357 seq_puts(m, "status: confirmed\n");
2359 seq_puts(m, "status: unconfirmed\n");
2360 seq_printf(m, "name: ");
2361 seq_quote_mem(m, clp->cl_name.data, clp->cl_name.len);
2362 seq_printf(m, "\nminor version: %d\n", clp->cl_minorversion);
2363 if (clp->cl_nii_domain.data) {
2364 seq_printf(m, "Implementation domain: ");
2365 seq_quote_mem(m, clp->cl_nii_domain.data,
2366 clp->cl_nii_domain.len);
2367 seq_printf(m, "\nImplementation name: ");
2368 seq_quote_mem(m, clp->cl_nii_name.data, clp->cl_nii_name.len);
2369 seq_printf(m, "\nImplementation time: [%lld, %ld]\n",
2370 clp->cl_nii_time.tv_sec, clp->cl_nii_time.tv_nsec);
2377 static int client_info_open(struct inode *inode, struct file *file)
2379 return single_open(file, client_info_show, inode);
2382 static const struct file_operations client_info_fops = {
2383 .open = client_info_open,
2385 .llseek = seq_lseek,
2386 .release = single_release,
2389 static void *states_start(struct seq_file *s, loff_t *pos)
2390 __acquires(&clp->cl_lock)
2392 struct nfs4_client *clp = s->private;
2393 unsigned long id = *pos;
2396 spin_lock(&clp->cl_lock);
2397 ret = idr_get_next_ul(&clp->cl_stateids, &id);
2402 static void *states_next(struct seq_file *s, void *v, loff_t *pos)
2404 struct nfs4_client *clp = s->private;
2405 unsigned long id = *pos;
2410 ret = idr_get_next_ul(&clp->cl_stateids, &id);
2415 static void states_stop(struct seq_file *s, void *v)
2416 __releases(&clp->cl_lock)
2418 struct nfs4_client *clp = s->private;
2420 spin_unlock(&clp->cl_lock);
2423 static void nfs4_show_fname(struct seq_file *s, struct nfsd_file *f)
2425 seq_printf(s, "filename: \"%pD2\"", f->nf_file);
2428 static void nfs4_show_superblock(struct seq_file *s, struct nfsd_file *f)
2430 struct inode *inode = f->nf_inode;
2432 seq_printf(s, "superblock: \"%02x:%02x:%ld\"",
2433 MAJOR(inode->i_sb->s_dev),
2434 MINOR(inode->i_sb->s_dev),
2438 static void nfs4_show_owner(struct seq_file *s, struct nfs4_stateowner *oo)
2440 seq_printf(s, "owner: ");
2441 seq_quote_mem(s, oo->so_owner.data, oo->so_owner.len);
2444 static void nfs4_show_stateid(struct seq_file *s, stateid_t *stid)
2446 seq_printf(s, "0x%.8x", stid->si_generation);
2447 seq_printf(s, "%12phN", &stid->si_opaque);
2450 static int nfs4_show_open(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;
2456 unsigned int access, deny;
2458 if (st->sc_type != NFS4_OPEN_STID && st->sc_type != NFS4_LOCK_STID)
2459 return 0; /* XXX: or SEQ_SKIP? */
2460 ols = openlockstateid(st);
2461 oo = ols->st_stateowner;
2463 file = find_any_file(nf);
2467 seq_printf(s, "- ");
2468 nfs4_show_stateid(s, &st->sc_stateid);
2469 seq_printf(s, ": { type: open, ");
2471 access = bmap_to_share_mode(ols->st_access_bmap);
2472 deny = bmap_to_share_mode(ols->st_deny_bmap);
2474 seq_printf(s, "access: %s%s, ",
2475 access & NFS4_SHARE_ACCESS_READ ? "r" : "-",
2476 access & NFS4_SHARE_ACCESS_WRITE ? "w" : "-");
2477 seq_printf(s, "deny: %s%s, ",
2478 deny & NFS4_SHARE_ACCESS_READ ? "r" : "-",
2479 deny & NFS4_SHARE_ACCESS_WRITE ? "w" : "-");
2481 nfs4_show_superblock(s, file);
2482 seq_printf(s, ", ");
2483 nfs4_show_fname(s, file);
2484 seq_printf(s, ", ");
2485 nfs4_show_owner(s, oo);
2486 seq_printf(s, " }\n");
2487 nfsd_file_put(file);
2492 static int nfs4_show_lock(struct seq_file *s, struct nfs4_stid *st)
2494 struct nfs4_ol_stateid *ols;
2495 struct nfs4_file *nf;
2496 struct nfsd_file *file;
2497 struct nfs4_stateowner *oo;
2499 ols = openlockstateid(st);
2500 oo = ols->st_stateowner;
2502 file = find_any_file(nf);
2506 seq_printf(s, "- ");
2507 nfs4_show_stateid(s, &st->sc_stateid);
2508 seq_printf(s, ": { type: lock, ");
2511 * Note: a lock stateid isn't really the same thing as a lock,
2512 * it's the locking state held by one owner on a file, and there
2513 * may be multiple (or no) lock ranges associated with it.
2514 * (Same for the matter is true of open stateids.)
2517 nfs4_show_superblock(s, file);
2518 /* XXX: open stateid? */
2519 seq_printf(s, ", ");
2520 nfs4_show_fname(s, file);
2521 seq_printf(s, ", ");
2522 nfs4_show_owner(s, oo);
2523 seq_printf(s, " }\n");
2524 nfsd_file_put(file);
2529 static int nfs4_show_deleg(struct seq_file *s, struct nfs4_stid *st)
2531 struct nfs4_delegation *ds;
2532 struct nfs4_file *nf;
2533 struct nfsd_file *file;
2535 ds = delegstateid(st);
2537 file = find_deleg_file(nf);
2541 seq_printf(s, "- ");
2542 nfs4_show_stateid(s, &st->sc_stateid);
2543 seq_printf(s, ": { type: deleg, ");
2545 /* Kinda dead code as long as we only support read delegs: */
2546 seq_printf(s, "access: %s, ",
2547 ds->dl_type == NFS4_OPEN_DELEGATE_READ ? "r" : "w");
2549 /* XXX: lease time, whether it's being recalled. */
2551 nfs4_show_superblock(s, file);
2552 seq_printf(s, ", ");
2553 nfs4_show_fname(s, file);
2554 seq_printf(s, " }\n");
2555 nfsd_file_put(file);
2560 static int nfs4_show_layout(struct seq_file *s, struct nfs4_stid *st)
2562 struct nfs4_layout_stateid *ls;
2563 struct nfsd_file *file;
2565 ls = container_of(st, struct nfs4_layout_stateid, ls_stid);
2568 seq_printf(s, "- ");
2569 nfs4_show_stateid(s, &st->sc_stateid);
2570 seq_printf(s, ": { type: layout, ");
2572 /* XXX: What else would be useful? */
2574 nfs4_show_superblock(s, file);
2575 seq_printf(s, ", ");
2576 nfs4_show_fname(s, file);
2577 seq_printf(s, " }\n");
2582 static int states_show(struct seq_file *s, void *v)
2584 struct nfs4_stid *st = v;
2586 switch (st->sc_type) {
2587 case NFS4_OPEN_STID:
2588 return nfs4_show_open(s, st);
2589 case NFS4_LOCK_STID:
2590 return nfs4_show_lock(s, st);
2591 case NFS4_DELEG_STID:
2592 return nfs4_show_deleg(s, st);
2593 case NFS4_LAYOUT_STID:
2594 return nfs4_show_layout(s, st);
2596 return 0; /* XXX: or SEQ_SKIP? */
2598 /* XXX: copy stateids? */
2601 static struct seq_operations states_seq_ops = {
2602 .start = states_start,
2603 .next = states_next,
2604 .stop = states_stop,
2608 static int client_states_open(struct inode *inode, struct file *file)
2611 struct nfs4_client *clp;
2614 clp = get_nfsdfs_clp(inode);
2618 ret = seq_open(file, &states_seq_ops);
2621 s = file->private_data;
2626 static int client_opens_release(struct inode *inode, struct file *file)
2628 struct seq_file *m = file->private_data;
2629 struct nfs4_client *clp = m->private;
2631 /* XXX: alternatively, we could get/drop in seq start/stop */
2636 static const struct file_operations client_states_fops = {
2637 .open = client_states_open,
2639 .llseek = seq_lseek,
2640 .release = client_opens_release,
2644 * Normally we refuse to destroy clients that are in use, but here the
2645 * administrator is telling us to just do it. We also want to wait
2646 * so the caller has a guarantee that the client's locks are gone by
2647 * the time the write returns:
2649 static void force_expire_client(struct nfs4_client *clp)
2651 struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
2652 bool already_expired;
2654 spin_lock(&clp->cl_lock);
2656 spin_unlock(&clp->cl_lock);
2658 wait_event(expiry_wq, atomic_read(&clp->cl_rpc_users) == 0);
2659 spin_lock(&nn->client_lock);
2660 already_expired = list_empty(&clp->cl_lru);
2661 if (!already_expired)
2662 unhash_client_locked(clp);
2663 spin_unlock(&nn->client_lock);
2665 if (!already_expired)
2668 wait_event(expiry_wq, clp->cl_nfsd_dentry == NULL);
2671 static ssize_t client_ctl_write(struct file *file, const char __user *buf,
2672 size_t size, loff_t *pos)
2675 struct nfs4_client *clp;
2677 data = simple_transaction_get(file, buf, size);
2679 return PTR_ERR(data);
2680 if (size != 7 || 0 != memcmp(data, "expire\n", 7))
2682 clp = get_nfsdfs_clp(file_inode(file));
2685 force_expire_client(clp);
2690 static const struct file_operations client_ctl_fops = {
2691 .write = client_ctl_write,
2692 .release = simple_transaction_release,
2695 static const struct tree_descr client_files[] = {
2696 [0] = {"info", &client_info_fops, S_IRUSR},
2697 [1] = {"states", &client_states_fops, S_IRUSR},
2698 [2] = {"ctl", &client_ctl_fops, S_IWUSR},
2702 static struct nfs4_client *create_client(struct xdr_netobj name,
2703 struct svc_rqst *rqstp, nfs4_verifier *verf)
2705 struct nfs4_client *clp;
2706 struct sockaddr *sa = svc_addr(rqstp);
2708 struct net *net = SVC_NET(rqstp);
2709 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
2710 struct dentry *dentries[ARRAY_SIZE(client_files)];
2712 clp = alloc_client(name);
2716 ret = copy_cred(&clp->cl_cred, &rqstp->rq_cred);
2722 kref_init(&clp->cl_nfsdfs.cl_ref);
2723 nfsd4_init_cb(&clp->cl_cb_null, clp, NULL, NFSPROC4_CLNT_CB_NULL);
2724 clp->cl_time = ktime_get_boottime_seconds();
2725 clear_bit(0, &clp->cl_cb_slot_busy);
2726 copy_verf(clp, verf);
2727 memcpy(&clp->cl_addr, sa, sizeof(struct sockaddr_storage));
2728 clp->cl_cb_session = NULL;
2730 clp->cl_nfsd_dentry = nfsd_client_mkdir(
2731 nn, &clp->cl_nfsdfs,
2732 clp->cl_clientid.cl_id - nn->clientid_base,
2733 client_files, dentries);
2734 clp->cl_nfsd_info_dentry = dentries[0];
2735 if (!clp->cl_nfsd_dentry) {
2743 add_clp_to_name_tree(struct nfs4_client *new_clp, struct rb_root *root)
2745 struct rb_node **new = &(root->rb_node), *parent = NULL;
2746 struct nfs4_client *clp;
2749 clp = rb_entry(*new, struct nfs4_client, cl_namenode);
2752 if (compare_blob(&clp->cl_name, &new_clp->cl_name) > 0)
2753 new = &((*new)->rb_left);
2755 new = &((*new)->rb_right);
2758 rb_link_node(&new_clp->cl_namenode, parent, new);
2759 rb_insert_color(&new_clp->cl_namenode, root);
2762 static struct nfs4_client *
2763 find_clp_in_name_tree(struct xdr_netobj *name, struct rb_root *root)
2766 struct rb_node *node = root->rb_node;
2767 struct nfs4_client *clp;
2770 clp = rb_entry(node, struct nfs4_client, cl_namenode);
2771 cmp = compare_blob(&clp->cl_name, name);
2773 node = node->rb_left;
2775 node = node->rb_right;
2783 add_to_unconfirmed(struct nfs4_client *clp)
2785 unsigned int idhashval;
2786 struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
2788 lockdep_assert_held(&nn->client_lock);
2790 clear_bit(NFSD4_CLIENT_CONFIRMED, &clp->cl_flags);
2791 add_clp_to_name_tree(clp, &nn->unconf_name_tree);
2792 idhashval = clientid_hashval(clp->cl_clientid.cl_id);
2793 list_add(&clp->cl_idhash, &nn->unconf_id_hashtbl[idhashval]);
2794 renew_client_locked(clp);
2798 move_to_confirmed(struct nfs4_client *clp)
2800 unsigned int idhashval = clientid_hashval(clp->cl_clientid.cl_id);
2801 struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
2803 lockdep_assert_held(&nn->client_lock);
2805 dprintk("NFSD: move_to_confirm nfs4_client %p\n", clp);
2806 list_move(&clp->cl_idhash, &nn->conf_id_hashtbl[idhashval]);
2807 rb_erase(&clp->cl_namenode, &nn->unconf_name_tree);
2808 add_clp_to_name_tree(clp, &nn->conf_name_tree);
2809 if (!test_and_set_bit(NFSD4_CLIENT_CONFIRMED, &clp->cl_flags) &&
2810 clp->cl_nfsd_dentry &&
2811 clp->cl_nfsd_info_dentry)
2812 fsnotify_dentry(clp->cl_nfsd_info_dentry, FS_MODIFY);
2813 renew_client_locked(clp);
2816 static struct nfs4_client *
2817 find_client_in_id_table(struct list_head *tbl, clientid_t *clid, bool sessions)
2819 struct nfs4_client *clp;
2820 unsigned int idhashval = clientid_hashval(clid->cl_id);
2822 list_for_each_entry(clp, &tbl[idhashval], cl_idhash) {
2823 if (same_clid(&clp->cl_clientid, clid)) {
2824 if ((bool)clp->cl_minorversion != sessions)
2826 renew_client_locked(clp);
2833 static struct nfs4_client *
2834 find_confirmed_client(clientid_t *clid, bool sessions, struct nfsd_net *nn)
2836 struct list_head *tbl = nn->conf_id_hashtbl;
2838 lockdep_assert_held(&nn->client_lock);
2839 return find_client_in_id_table(tbl, clid, sessions);
2842 static struct nfs4_client *
2843 find_unconfirmed_client(clientid_t *clid, bool sessions, struct nfsd_net *nn)
2845 struct list_head *tbl = nn->unconf_id_hashtbl;
2847 lockdep_assert_held(&nn->client_lock);
2848 return find_client_in_id_table(tbl, clid, sessions);
2851 static bool clp_used_exchangeid(struct nfs4_client *clp)
2853 return clp->cl_exchange_flags != 0;
2856 static struct nfs4_client *
2857 find_confirmed_client_by_name(struct xdr_netobj *name, struct nfsd_net *nn)
2859 lockdep_assert_held(&nn->client_lock);
2860 return find_clp_in_name_tree(name, &nn->conf_name_tree);
2863 static struct nfs4_client *
2864 find_unconfirmed_client_by_name(struct xdr_netobj *name, struct nfsd_net *nn)
2866 lockdep_assert_held(&nn->client_lock);
2867 return find_clp_in_name_tree(name, &nn->unconf_name_tree);
2871 gen_callback(struct nfs4_client *clp, struct nfsd4_setclientid *se, struct svc_rqst *rqstp)
2873 struct nfs4_cb_conn *conn = &clp->cl_cb_conn;
2874 struct sockaddr *sa = svc_addr(rqstp);
2875 u32 scopeid = rpc_get_scope_id(sa);
2876 unsigned short expected_family;
2878 /* Currently, we only support tcp and tcp6 for the callback channel */
2879 if (se->se_callback_netid_len == 3 &&
2880 !memcmp(se->se_callback_netid_val, "tcp", 3))
2881 expected_family = AF_INET;
2882 else if (se->se_callback_netid_len == 4 &&
2883 !memcmp(se->se_callback_netid_val, "tcp6", 4))
2884 expected_family = AF_INET6;
2888 conn->cb_addrlen = rpc_uaddr2sockaddr(clp->net, se->se_callback_addr_val,
2889 se->se_callback_addr_len,
2890 (struct sockaddr *)&conn->cb_addr,
2891 sizeof(conn->cb_addr));
2893 if (!conn->cb_addrlen || conn->cb_addr.ss_family != expected_family)
2896 if (conn->cb_addr.ss_family == AF_INET6)
2897 ((struct sockaddr_in6 *)&conn->cb_addr)->sin6_scope_id = scopeid;
2899 conn->cb_prog = se->se_callback_prog;
2900 conn->cb_ident = se->se_callback_ident;
2901 memcpy(&conn->cb_saddr, &rqstp->rq_daddr, rqstp->rq_daddrlen);
2902 trace_nfsd_cb_args(clp, conn);
2905 conn->cb_addr.ss_family = AF_UNSPEC;
2906 conn->cb_addrlen = 0;
2907 trace_nfsd_cb_nodelegs(clp);
2912 * Cache a reply. nfsd4_check_resp_size() has bounded the cache size.
2915 nfsd4_store_cache_entry(struct nfsd4_compoundres *resp)
2917 struct xdr_buf *buf = resp->xdr->buf;
2918 struct nfsd4_slot *slot = resp->cstate.slot;
2921 dprintk("--> %s slot %p\n", __func__, slot);
2923 slot->sl_flags |= NFSD4_SLOT_INITIALIZED;
2924 slot->sl_opcnt = resp->opcnt;
2925 slot->sl_status = resp->cstate.status;
2926 free_svc_cred(&slot->sl_cred);
2927 copy_cred(&slot->sl_cred, &resp->rqstp->rq_cred);
2929 if (!nfsd4_cache_this(resp)) {
2930 slot->sl_flags &= ~NFSD4_SLOT_CACHED;
2933 slot->sl_flags |= NFSD4_SLOT_CACHED;
2935 base = resp->cstate.data_offset;
2936 slot->sl_datalen = buf->len - base;
2937 if (read_bytes_from_xdr_buf(buf, base, slot->sl_data, slot->sl_datalen))
2938 WARN(1, "%s: sessions DRC could not cache compound\n",
2944 * Encode the replay sequence operation from the slot values.
2945 * If cachethis is FALSE encode the uncached rep error on the next
2946 * operation which sets resp->p and increments resp->opcnt for
2947 * nfs4svc_encode_compoundres.
2951 nfsd4_enc_sequence_replay(struct nfsd4_compoundargs *args,
2952 struct nfsd4_compoundres *resp)
2954 struct nfsd4_op *op;
2955 struct nfsd4_slot *slot = resp->cstate.slot;
2957 /* Encode the replayed sequence operation */
2958 op = &args->ops[resp->opcnt - 1];
2959 nfsd4_encode_operation(resp, op);
2961 if (slot->sl_flags & NFSD4_SLOT_CACHED)
2963 if (args->opcnt == 1) {
2965 * The original operation wasn't a solo sequence--we
2966 * always cache those--so this retry must not match the
2969 op->status = nfserr_seq_false_retry;
2971 op = &args->ops[resp->opcnt++];
2972 op->status = nfserr_retry_uncached_rep;
2973 nfsd4_encode_operation(resp, op);
2979 * The sequence operation is not cached because we can use the slot and
2983 nfsd4_replay_cache_entry(struct nfsd4_compoundres *resp,
2984 struct nfsd4_sequence *seq)
2986 struct nfsd4_slot *slot = resp->cstate.slot;
2987 struct xdr_stream *xdr = resp->xdr;
2991 dprintk("--> %s slot %p\n", __func__, slot);
2993 status = nfsd4_enc_sequence_replay(resp->rqstp->rq_argp, resp);
2997 p = xdr_reserve_space(xdr, slot->sl_datalen);
3000 return nfserr_serverfault;
3002 xdr_encode_opaque_fixed(p, slot->sl_data, slot->sl_datalen);
3003 xdr_commit_encode(xdr);
3005 resp->opcnt = slot->sl_opcnt;
3006 return slot->sl_status;
3010 * Set the exchange_id flags returned by the server.
3013 nfsd4_set_ex_flags(struct nfs4_client *new, struct nfsd4_exchange_id *clid)
3015 #ifdef CONFIG_NFSD_PNFS
3016 new->cl_exchange_flags |= EXCHGID4_FLAG_USE_PNFS_MDS;
3018 new->cl_exchange_flags |= EXCHGID4_FLAG_USE_NON_PNFS;
3021 /* Referrals are supported, Migration is not. */
3022 new->cl_exchange_flags |= EXCHGID4_FLAG_SUPP_MOVED_REFER;
3024 /* set the wire flags to return to client. */
3025 clid->flags = new->cl_exchange_flags;
3028 static bool client_has_openowners(struct nfs4_client *clp)
3030 struct nfs4_openowner *oo;
3032 list_for_each_entry(oo, &clp->cl_openowners, oo_perclient) {
3033 if (!list_empty(&oo->oo_owner.so_stateids))
3039 static bool client_has_state(struct nfs4_client *clp)
3041 return client_has_openowners(clp)
3042 #ifdef CONFIG_NFSD_PNFS
3043 || !list_empty(&clp->cl_lo_states)
3045 || !list_empty(&clp->cl_delegations)
3046 || !list_empty(&clp->cl_sessions)
3047 || !list_empty(&clp->async_copies);
3050 static __be32 copy_impl_id(struct nfs4_client *clp,
3051 struct nfsd4_exchange_id *exid)
3053 if (!exid->nii_domain.data)
3055 xdr_netobj_dup(&clp->cl_nii_domain, &exid->nii_domain, GFP_KERNEL);
3056 if (!clp->cl_nii_domain.data)
3057 return nfserr_jukebox;
3058 xdr_netobj_dup(&clp->cl_nii_name, &exid->nii_name, GFP_KERNEL);
3059 if (!clp->cl_nii_name.data)
3060 return nfserr_jukebox;
3061 clp->cl_nii_time = exid->nii_time;
3066 nfsd4_exchange_id(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3067 union nfsd4_op_u *u)
3069 struct nfsd4_exchange_id *exid = &u->exchange_id;
3070 struct nfs4_client *conf, *new;
3071 struct nfs4_client *unconf = NULL;
3073 char addr_str[INET6_ADDRSTRLEN];
3074 nfs4_verifier verf = exid->verifier;
3075 struct sockaddr *sa = svc_addr(rqstp);
3076 bool update = exid->flags & EXCHGID4_FLAG_UPD_CONFIRMED_REC_A;
3077 struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
3079 rpc_ntop(sa, addr_str, sizeof(addr_str));
3080 dprintk("%s rqstp=%p exid=%p clname.len=%u clname.data=%p "
3081 "ip_addr=%s flags %x, spa_how %u\n",
3082 __func__, rqstp, exid, exid->clname.len, exid->clname.data,
3083 addr_str, exid->flags, exid->spa_how);
3085 if (exid->flags & ~EXCHGID4_FLAG_MASK_A)
3086 return nfserr_inval;
3088 new = create_client(exid->clname, rqstp, &verf);
3090 return nfserr_jukebox;
3091 status = copy_impl_id(new, exid);
3095 switch (exid->spa_how) {
3097 exid->spo_must_enforce[0] = 0;
3098 exid->spo_must_enforce[1] = (
3099 1 << (OP_BIND_CONN_TO_SESSION - 32) |
3100 1 << (OP_EXCHANGE_ID - 32) |
3101 1 << (OP_CREATE_SESSION - 32) |
3102 1 << (OP_DESTROY_SESSION - 32) |
3103 1 << (OP_DESTROY_CLIENTID - 32));
3105 exid->spo_must_allow[0] &= (1 << (OP_CLOSE) |
3106 1 << (OP_OPEN_DOWNGRADE) |
3108 1 << (OP_DELEGRETURN));
3110 exid->spo_must_allow[1] &= (
3111 1 << (OP_TEST_STATEID - 32) |
3112 1 << (OP_FREE_STATEID - 32));
3113 if (!svc_rqst_integrity_protected(rqstp)) {
3114 status = nfserr_inval;
3118 * Sometimes userspace doesn't give us a principal.
3119 * Which is a bug, really. Anyway, we can't enforce
3120 * MACH_CRED in that case, better to give up now:
3122 if (!new->cl_cred.cr_principal &&
3123 !new->cl_cred.cr_raw_principal) {
3124 status = nfserr_serverfault;
3127 new->cl_mach_cred = true;
3130 default: /* checked by xdr code */
3134 status = nfserr_encr_alg_unsupp;
3138 /* Cases below refer to rfc 5661 section 18.35.4: */
3139 spin_lock(&nn->client_lock);
3140 conf = find_confirmed_client_by_name(&exid->clname, nn);
3142 bool creds_match = same_creds(&conf->cl_cred, &rqstp->rq_cred);
3143 bool verfs_match = same_verf(&verf, &conf->cl_verifier);
3146 if (!clp_used_exchangeid(conf)) { /* buggy client */
3147 status = nfserr_inval;
3150 if (!nfsd4_mach_creds_match(conf, rqstp)) {
3151 status = nfserr_wrong_cred;
3154 if (!creds_match) { /* case 9 */
3155 status = nfserr_perm;
3158 if (!verfs_match) { /* case 8 */
3159 status = nfserr_not_same;
3163 exid->flags |= EXCHGID4_FLAG_CONFIRMED_R;
3166 if (!creds_match) { /* case 3 */
3167 if (client_has_state(conf)) {
3168 status = nfserr_clid_inuse;
3173 if (verfs_match) { /* case 2 */
3174 conf->cl_exchange_flags |= EXCHGID4_FLAG_CONFIRMED_R;
3177 /* case 5, client reboot */
3182 if (update) { /* case 7 */
3183 status = nfserr_noent;
3187 unconf = find_unconfirmed_client_by_name(&exid->clname, nn);
3188 if (unconf) /* case 4, possible retry or client restart */
3189 unhash_client_locked(unconf);
3191 /* case 1 (normal case) */
3194 status = mark_client_expired_locked(conf);
3198 new->cl_minorversion = cstate->minorversion;
3199 new->cl_spo_must_allow.u.words[0] = exid->spo_must_allow[0];
3200 new->cl_spo_must_allow.u.words[1] = exid->spo_must_allow[1];
3202 add_to_unconfirmed(new);
3205 exid->clientid.cl_boot = conf->cl_clientid.cl_boot;
3206 exid->clientid.cl_id = conf->cl_clientid.cl_id;
3208 exid->seqid = conf->cl_cs_slot.sl_seqid + 1;
3209 nfsd4_set_ex_flags(conf, exid);
3211 dprintk("nfsd4_exchange_id seqid %d flags %x\n",
3212 conf->cl_cs_slot.sl_seqid, conf->cl_exchange_flags);
3216 spin_unlock(&nn->client_lock);
3221 expire_client(unconf);
3226 check_slot_seqid(u32 seqid, u32 slot_seqid, int slot_inuse)
3228 dprintk("%s enter. seqid %d slot_seqid %d\n", __func__, seqid,
3231 /* The slot is in use, and no response has been sent. */
3233 if (seqid == slot_seqid)
3234 return nfserr_jukebox;
3236 return nfserr_seq_misordered;
3238 /* Note unsigned 32-bit arithmetic handles wraparound: */
3239 if (likely(seqid == slot_seqid + 1))
3241 if (seqid == slot_seqid)
3242 return nfserr_replay_cache;
3243 return nfserr_seq_misordered;
3247 * Cache the create session result into the create session single DRC
3248 * slot cache by saving the xdr structure. sl_seqid has been set.
3249 * Do this for solo or embedded create session operations.
3252 nfsd4_cache_create_session(struct nfsd4_create_session *cr_ses,
3253 struct nfsd4_clid_slot *slot, __be32 nfserr)
3255 slot->sl_status = nfserr;
3256 memcpy(&slot->sl_cr_ses, cr_ses, sizeof(*cr_ses));
3260 nfsd4_replay_create_session(struct nfsd4_create_session *cr_ses,
3261 struct nfsd4_clid_slot *slot)
3263 memcpy(cr_ses, &slot->sl_cr_ses, sizeof(*cr_ses));
3264 return slot->sl_status;
3267 #define NFSD_MIN_REQ_HDR_SEQ_SZ ((\
3268 2 * 2 + /* credential,verifier: AUTH_NULL, length 0 */ \
3269 1 + /* MIN tag is length with zero, only length */ \
3270 3 + /* version, opcount, opcode */ \
3271 XDR_QUADLEN(NFS4_MAX_SESSIONID_LEN) + \
3272 /* seqid, slotID, slotID, cache */ \
3273 4 ) * sizeof(__be32))
3275 #define NFSD_MIN_RESP_HDR_SEQ_SZ ((\
3276 2 + /* verifier: AUTH_NULL, length 0 */\
3278 1 + /* MIN tag is length with zero, only length */ \
3279 3 + /* opcount, opcode, opstatus*/ \
3280 XDR_QUADLEN(NFS4_MAX_SESSIONID_LEN) + \
3281 /* seqid, slotID, slotID, slotID, status */ \
3282 5 ) * sizeof(__be32))
3284 static __be32 check_forechannel_attrs(struct nfsd4_channel_attrs *ca, struct nfsd_net *nn)
3286 u32 maxrpc = nn->nfsd_serv->sv_max_mesg;
3288 if (ca->maxreq_sz < NFSD_MIN_REQ_HDR_SEQ_SZ)
3289 return nfserr_toosmall;
3290 if (ca->maxresp_sz < NFSD_MIN_RESP_HDR_SEQ_SZ)
3291 return nfserr_toosmall;
3292 ca->headerpadsz = 0;
3293 ca->maxreq_sz = min_t(u32, ca->maxreq_sz, maxrpc);
3294 ca->maxresp_sz = min_t(u32, ca->maxresp_sz, maxrpc);
3295 ca->maxops = min_t(u32, ca->maxops, NFSD_MAX_OPS_PER_COMPOUND);
3296 ca->maxresp_cached = min_t(u32, ca->maxresp_cached,
3297 NFSD_SLOT_CACHE_SIZE + NFSD_MIN_HDR_SEQ_SZ);
3298 ca->maxreqs = min_t(u32, ca->maxreqs, NFSD_MAX_SLOTS_PER_SESSION);
3300 * Note decreasing slot size below client's request may make it
3301 * difficult for client to function correctly, whereas
3302 * decreasing the number of slots will (just?) affect
3303 * performance. When short on memory we therefore prefer to
3304 * decrease number of slots instead of their size. Clients that
3305 * request larger slots than they need will get poor results:
3306 * Note that we always allow at least one slot, because our
3307 * accounting is soft and provides no guarantees either way.
3309 ca->maxreqs = nfsd4_get_drc_mem(ca, nn);
3315 * Server's NFSv4.1 backchannel support is AUTH_SYS-only for now.
3316 * These are based on similar macros in linux/sunrpc/msg_prot.h .
3318 #define RPC_MAX_HEADER_WITH_AUTH_SYS \
3319 (RPC_CALLHDRSIZE + 2 * (2 + UNX_CALLSLACK))
3321 #define RPC_MAX_REPHEADER_WITH_AUTH_SYS \
3322 (RPC_REPHDRSIZE + (2 + NUL_REPLYSLACK))
3324 #define NFSD_CB_MAX_REQ_SZ ((NFS4_enc_cb_recall_sz + \
3325 RPC_MAX_HEADER_WITH_AUTH_SYS) * sizeof(__be32))
3326 #define NFSD_CB_MAX_RESP_SZ ((NFS4_dec_cb_recall_sz + \
3327 RPC_MAX_REPHEADER_WITH_AUTH_SYS) * \
3330 static __be32 check_backchannel_attrs(struct nfsd4_channel_attrs *ca)
3332 ca->headerpadsz = 0;
3334 if (ca->maxreq_sz < NFSD_CB_MAX_REQ_SZ)
3335 return nfserr_toosmall;
3336 if (ca->maxresp_sz < NFSD_CB_MAX_RESP_SZ)
3337 return nfserr_toosmall;
3338 ca->maxresp_cached = 0;
3340 return nfserr_toosmall;
3345 static __be32 nfsd4_check_cb_sec(struct nfsd4_cb_sec *cbs)
3347 switch (cbs->flavor) {
3353 * GSS case: the spec doesn't allow us to return this
3354 * error. But it also doesn't allow us not to support
3356 * I'd rather this fail hard than return some error the
3357 * client might think it can already handle:
3359 return nfserr_encr_alg_unsupp;
3364 nfsd4_create_session(struct svc_rqst *rqstp,
3365 struct nfsd4_compound_state *cstate, union nfsd4_op_u *u)
3367 struct nfsd4_create_session *cr_ses = &u->create_session;
3368 struct sockaddr *sa = svc_addr(rqstp);
3369 struct nfs4_client *conf, *unconf;
3370 struct nfs4_client *old = NULL;
3371 struct nfsd4_session *new;
3372 struct nfsd4_conn *conn;
3373 struct nfsd4_clid_slot *cs_slot = NULL;
3375 struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
3377 if (cr_ses->flags & ~SESSION4_FLAG_MASK_A)
3378 return nfserr_inval;
3379 status = nfsd4_check_cb_sec(&cr_ses->cb_sec);
3382 status = check_forechannel_attrs(&cr_ses->fore_channel, nn);
3385 status = check_backchannel_attrs(&cr_ses->back_channel);
3387 goto out_release_drc_mem;
3388 status = nfserr_jukebox;
3389 new = alloc_session(&cr_ses->fore_channel, &cr_ses->back_channel);
3391 goto out_release_drc_mem;
3392 conn = alloc_conn_from_crses(rqstp, cr_ses);
3394 goto out_free_session;
3396 spin_lock(&nn->client_lock);
3397 unconf = find_unconfirmed_client(&cr_ses->clientid, true, nn);
3398 conf = find_confirmed_client(&cr_ses->clientid, true, nn);
3399 WARN_ON_ONCE(conf && unconf);
3402 status = nfserr_wrong_cred;
3403 if (!nfsd4_mach_creds_match(conf, rqstp))
3405 cs_slot = &conf->cl_cs_slot;
3406 status = check_slot_seqid(cr_ses->seqid, cs_slot->sl_seqid, 0);
3408 if (status == nfserr_replay_cache)
3409 status = nfsd4_replay_create_session(cr_ses, cs_slot);
3412 } else if (unconf) {
3413 if (!same_creds(&unconf->cl_cred, &rqstp->rq_cred) ||
3414 !rpc_cmp_addr(sa, (struct sockaddr *) &unconf->cl_addr)) {
3415 status = nfserr_clid_inuse;
3418 status = nfserr_wrong_cred;
3419 if (!nfsd4_mach_creds_match(unconf, rqstp))
3421 cs_slot = &unconf->cl_cs_slot;
3422 status = check_slot_seqid(cr_ses->seqid, cs_slot->sl_seqid, 0);
3424 /* an unconfirmed replay returns misordered */
3425 status = nfserr_seq_misordered;
3428 old = find_confirmed_client_by_name(&unconf->cl_name, nn);
3430 status = mark_client_expired_locked(old);
3436 move_to_confirmed(unconf);
3439 status = nfserr_stale_clientid;
3443 /* Persistent sessions are not supported */
3444 cr_ses->flags &= ~SESSION4_PERSIST;
3445 /* Upshifting from TCP to RDMA is not supported */
3446 cr_ses->flags &= ~SESSION4_RDMA;
3448 init_session(rqstp, new, conf, cr_ses);
3449 nfsd4_get_session_locked(new);
3451 memcpy(cr_ses->sessionid.data, new->se_sessionid.data,
3452 NFS4_MAX_SESSIONID_LEN);
3453 cs_slot->sl_seqid++;
3454 cr_ses->seqid = cs_slot->sl_seqid;
3456 /* cache solo and embedded create sessions under the client_lock */
3457 nfsd4_cache_create_session(cr_ses, cs_slot, status);
3458 spin_unlock(&nn->client_lock);
3459 /* init connection and backchannel */
3460 nfsd4_init_conn(rqstp, conn, new);
3461 nfsd4_put_session(new);
3466 spin_unlock(&nn->client_lock);
3471 __free_session(new);
3472 out_release_drc_mem:
3473 nfsd4_put_drc_mem(&cr_ses->fore_channel);
3477 static __be32 nfsd4_map_bcts_dir(u32 *dir)
3480 case NFS4_CDFC4_FORE:
3481 case NFS4_CDFC4_BACK:
3483 case NFS4_CDFC4_FORE_OR_BOTH:
3484 case NFS4_CDFC4_BACK_OR_BOTH:
3485 *dir = NFS4_CDFC4_BOTH;
3488 return nfserr_inval;
3491 __be32 nfsd4_backchannel_ctl(struct svc_rqst *rqstp,
3492 struct nfsd4_compound_state *cstate,
3493 union nfsd4_op_u *u)
3495 struct nfsd4_backchannel_ctl *bc = &u->backchannel_ctl;
3496 struct nfsd4_session *session = cstate->session;
3497 struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
3500 status = nfsd4_check_cb_sec(&bc->bc_cb_sec);
3503 spin_lock(&nn->client_lock);
3504 session->se_cb_prog = bc->bc_cb_program;
3505 session->se_cb_sec = bc->bc_cb_sec;
3506 spin_unlock(&nn->client_lock);
3508 nfsd4_probe_callback(session->se_client);
3513 static struct nfsd4_conn *__nfsd4_find_conn(struct svc_xprt *xpt, struct nfsd4_session *s)
3515 struct nfsd4_conn *c;
3517 list_for_each_entry(c, &s->se_conns, cn_persession) {
3518 if (c->cn_xprt == xpt) {
3525 static __be32 nfsd4_match_existing_connection(struct svc_rqst *rqst,
3526 struct nfsd4_session *session, u32 req)
3528 struct nfs4_client *clp = session->se_client;
3529 struct svc_xprt *xpt = rqst->rq_xprt;
3530 struct nfsd4_conn *c;
3533 /* Following the last paragraph of RFC 5661 Section 18.34.3: */
3534 spin_lock(&clp->cl_lock);
3535 c = __nfsd4_find_conn(xpt, session);
3537 status = nfserr_noent;
3538 else if (req == c->cn_flags)
3540 else if (req == NFS4_CDFC4_FORE_OR_BOTH &&
3541 c->cn_flags != NFS4_CDFC4_BACK)
3543 else if (req == NFS4_CDFC4_BACK_OR_BOTH &&
3544 c->cn_flags != NFS4_CDFC4_FORE)
3547 status = nfserr_inval;
3548 spin_unlock(&clp->cl_lock);
3552 __be32 nfsd4_bind_conn_to_session(struct svc_rqst *rqstp,
3553 struct nfsd4_compound_state *cstate,
3554 union nfsd4_op_u *u)
3556 struct nfsd4_bind_conn_to_session *bcts = &u->bind_conn_to_session;
3558 struct nfsd4_conn *conn;
3559 struct nfsd4_session *session;
3560 struct net *net = SVC_NET(rqstp);
3561 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
3563 if (!nfsd4_last_compound_op(rqstp))
3564 return nfserr_not_only_op;
3565 spin_lock(&nn->client_lock);
3566 session = find_in_sessionid_hashtbl(&bcts->sessionid, net, &status);
3567 spin_unlock(&nn->client_lock);
3569 goto out_no_session;
3570 status = nfserr_wrong_cred;
3571 if (!nfsd4_mach_creds_match(session->se_client, rqstp))
3573 status = nfsd4_match_existing_connection(rqstp, session, bcts->dir);
3574 if (status == nfs_ok || status == nfserr_inval)
3576 status = nfsd4_map_bcts_dir(&bcts->dir);
3579 conn = alloc_conn(rqstp, bcts->dir);
3580 status = nfserr_jukebox;
3583 nfsd4_init_conn(rqstp, conn, session);
3586 nfsd4_put_session(session);
3591 static bool nfsd4_compound_in_session(struct nfsd4_compound_state *cstate, struct nfs4_sessionid *sid)
3593 if (!cstate->session)
3595 return !memcmp(sid, &cstate->session->se_sessionid, sizeof(*sid));
3599 nfsd4_destroy_session(struct svc_rqst *r, struct nfsd4_compound_state *cstate,
3600 union nfsd4_op_u *u)
3602 struct nfs4_sessionid *sessionid = &u->destroy_session.sessionid;
3603 struct nfsd4_session *ses;
3605 int ref_held_by_me = 0;
3606 struct net *net = SVC_NET(r);
3607 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
3609 status = nfserr_not_only_op;
3610 if (nfsd4_compound_in_session(cstate, sessionid)) {
3611 if (!nfsd4_last_compound_op(r))
3615 dump_sessionid(__func__, sessionid);
3616 spin_lock(&nn->client_lock);
3617 ses = find_in_sessionid_hashtbl(sessionid, net, &status);
3619 goto out_client_lock;
3620 status = nfserr_wrong_cred;
3621 if (!nfsd4_mach_creds_match(ses->se_client, r))
3622 goto out_put_session;
3623 status = mark_session_dead_locked(ses, 1 + ref_held_by_me);
3625 goto out_put_session;
3626 unhash_session(ses);
3627 spin_unlock(&nn->client_lock);
3629 nfsd4_probe_callback_sync(ses->se_client);
3631 spin_lock(&nn->client_lock);
3634 nfsd4_put_session_locked(ses);
3636 spin_unlock(&nn->client_lock);
3641 static __be32 nfsd4_sequence_check_conn(struct nfsd4_conn *new, struct nfsd4_session *ses)
3643 struct nfs4_client *clp = ses->se_client;
3644 struct nfsd4_conn *c;
3645 __be32 status = nfs_ok;
3648 spin_lock(&clp->cl_lock);
3649 c = __nfsd4_find_conn(new->cn_xprt, ses);
3652 status = nfserr_conn_not_bound_to_session;
3653 if (clp->cl_mach_cred)
3655 __nfsd4_hash_conn(new, ses);
3656 spin_unlock(&clp->cl_lock);
3657 ret = nfsd4_register_conn(new);
3659 /* oops; xprt is already down: */
3660 nfsd4_conn_lost(&new->cn_xpt_user);
3663 spin_unlock(&clp->cl_lock);
3668 static bool nfsd4_session_too_many_ops(struct svc_rqst *rqstp, struct nfsd4_session *session)
3670 struct nfsd4_compoundargs *args = rqstp->rq_argp;
3672 return args->opcnt > session->se_fchannel.maxops;
3675 static bool nfsd4_request_too_big(struct svc_rqst *rqstp,
3676 struct nfsd4_session *session)
3678 struct xdr_buf *xb = &rqstp->rq_arg;
3680 return xb->len > session->se_fchannel.maxreq_sz;
3683 static bool replay_matches_cache(struct svc_rqst *rqstp,
3684 struct nfsd4_sequence *seq, struct nfsd4_slot *slot)
3686 struct nfsd4_compoundargs *argp = rqstp->rq_argp;
3688 if ((bool)(slot->sl_flags & NFSD4_SLOT_CACHETHIS) !=
3689 (bool)seq->cachethis)
3692 * If there's an error then the reply can have fewer ops than
3695 if (slot->sl_opcnt < argp->opcnt && !slot->sl_status)
3698 * But if we cached a reply with *more* ops than the call you're
3699 * sending us now, then this new call is clearly not really a
3700 * replay of the old one:
3702 if (slot->sl_opcnt > argp->opcnt)
3704 /* This is the only check explicitly called by spec: */
3705 if (!same_creds(&rqstp->rq_cred, &slot->sl_cred))
3708 * There may be more comparisons we could actually do, but the
3709 * spec doesn't require us to catch every case where the calls
3710 * don't match (that would require caching the call as well as
3711 * the reply), so we don't bother.
3717 nfsd4_sequence(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3718 union nfsd4_op_u *u)
3720 struct nfsd4_sequence *seq = &u->sequence;
3721 struct nfsd4_compoundres *resp = rqstp->rq_resp;
3722 struct xdr_stream *xdr = resp->xdr;
3723 struct nfsd4_session *session;
3724 struct nfs4_client *clp;
3725 struct nfsd4_slot *slot;
3726 struct nfsd4_conn *conn;
3729 struct net *net = SVC_NET(rqstp);
3730 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
3732 if (resp->opcnt != 1)
3733 return nfserr_sequence_pos;
3736 * Will be either used or freed by nfsd4_sequence_check_conn
3739 conn = alloc_conn(rqstp, NFS4_CDFC4_FORE);
3741 return nfserr_jukebox;
3743 spin_lock(&nn->client_lock);
3744 session = find_in_sessionid_hashtbl(&seq->sessionid, net, &status);
3746 goto out_no_session;
3747 clp = session->se_client;
3749 status = nfserr_too_many_ops;
3750 if (nfsd4_session_too_many_ops(rqstp, session))
3751 goto out_put_session;
3753 status = nfserr_req_too_big;
3754 if (nfsd4_request_too_big(rqstp, session))
3755 goto out_put_session;
3757 status = nfserr_badslot;
3758 if (seq->slotid >= session->se_fchannel.maxreqs)
3759 goto out_put_session;
3761 slot = session->se_slots[seq->slotid];
3762 dprintk("%s: slotid %d\n", __func__, seq->slotid);
3764 /* We do not negotiate the number of slots yet, so set the
3765 * maxslots to the session maxreqs which is used to encode
3766 * sr_highest_slotid and the sr_target_slot id to maxslots */
3767 seq->maxslots = session->se_fchannel.maxreqs;
3769 status = check_slot_seqid(seq->seqid, slot->sl_seqid,
3770 slot->sl_flags & NFSD4_SLOT_INUSE);
3771 if (status == nfserr_replay_cache) {
3772 status = nfserr_seq_misordered;
3773 if (!(slot->sl_flags & NFSD4_SLOT_INITIALIZED))
3774 goto out_put_session;
3775 status = nfserr_seq_false_retry;
3776 if (!replay_matches_cache(rqstp, seq, slot))
3777 goto out_put_session;
3778 cstate->slot = slot;
3779 cstate->session = session;
3781 /* Return the cached reply status and set cstate->status
3782 * for nfsd4_proc_compound processing */
3783 status = nfsd4_replay_cache_entry(resp, seq);
3784 cstate->status = nfserr_replay_cache;
3788 goto out_put_session;
3790 status = nfsd4_sequence_check_conn(conn, session);
3793 goto out_put_session;
3795 buflen = (seq->cachethis) ?
3796 session->se_fchannel.maxresp_cached :
3797 session->se_fchannel.maxresp_sz;
3798 status = (seq->cachethis) ? nfserr_rep_too_big_to_cache :
3800 if (xdr_restrict_buflen(xdr, buflen - rqstp->rq_auth_slack))
3801 goto out_put_session;
3802 svc_reserve(rqstp, buflen);
3805 /* Success! bump slot seqid */
3806 slot->sl_seqid = seq->seqid;
3807 slot->sl_flags |= NFSD4_SLOT_INUSE;
3809 slot->sl_flags |= NFSD4_SLOT_CACHETHIS;
3811 slot->sl_flags &= ~NFSD4_SLOT_CACHETHIS;
3813 cstate->slot = slot;
3814 cstate->session = session;
3818 switch (clp->cl_cb_state) {
3820 seq->status_flags = SEQ4_STATUS_CB_PATH_DOWN;
3822 case NFSD4_CB_FAULT:
3823 seq->status_flags = SEQ4_STATUS_BACKCHANNEL_FAULT;
3826 seq->status_flags = 0;
3828 if (!list_empty(&clp->cl_revoked))
3829 seq->status_flags |= SEQ4_STATUS_RECALLABLE_STATE_REVOKED;
3833 spin_unlock(&nn->client_lock);
3836 nfsd4_put_session_locked(session);
3837 goto out_no_session;
3841 nfsd4_sequence_done(struct nfsd4_compoundres *resp)
3843 struct nfsd4_compound_state *cs = &resp->cstate;
3845 if (nfsd4_has_session(cs)) {
3846 if (cs->status != nfserr_replay_cache) {
3847 nfsd4_store_cache_entry(resp);
3848 cs->slot->sl_flags &= ~NFSD4_SLOT_INUSE;
3850 /* Drop session reference that was taken in nfsd4_sequence() */
3851 nfsd4_put_session(cs->session);
3853 put_client_renew(cs->clp);
3857 nfsd4_destroy_clientid(struct svc_rqst *rqstp,
3858 struct nfsd4_compound_state *cstate,
3859 union nfsd4_op_u *u)
3861 struct nfsd4_destroy_clientid *dc = &u->destroy_clientid;
3862 struct nfs4_client *conf, *unconf;
3863 struct nfs4_client *clp = NULL;
3865 struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
3867 spin_lock(&nn->client_lock);
3868 unconf = find_unconfirmed_client(&dc->clientid, true, nn);
3869 conf = find_confirmed_client(&dc->clientid, true, nn);
3870 WARN_ON_ONCE(conf && unconf);
3873 if (client_has_state(conf)) {
3874 status = nfserr_clientid_busy;
3877 status = mark_client_expired_locked(conf);
3884 status = nfserr_stale_clientid;
3887 if (!nfsd4_mach_creds_match(clp, rqstp)) {
3889 status = nfserr_wrong_cred;
3892 unhash_client_locked(clp);
3894 spin_unlock(&nn->client_lock);
3901 nfsd4_reclaim_complete(struct svc_rqst *rqstp,
3902 struct nfsd4_compound_state *cstate, union nfsd4_op_u *u)
3904 struct nfsd4_reclaim_complete *rc = &u->reclaim_complete;
3905 struct nfs4_client *clp = cstate->clp;
3908 if (rc->rca_one_fs) {
3909 if (!cstate->current_fh.fh_dentry)
3910 return nfserr_nofilehandle;
3912 * We don't take advantage of the rca_one_fs case.
3913 * That's OK, it's optional, we can safely ignore it.
3918 status = nfserr_complete_already;
3919 if (test_and_set_bit(NFSD4_CLIENT_RECLAIM_COMPLETE, &clp->cl_flags))
3922 status = nfserr_stale_clientid;
3923 if (is_client_expired(clp))
3925 * The following error isn't really legal.
3926 * But we only get here if the client just explicitly
3927 * destroyed the client. Surely it no longer cares what
3928 * error it gets back on an operation for the dead
3934 nfsd4_client_record_create(clp);
3935 inc_reclaim_complete(clp);
3941 nfsd4_setclientid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3942 union nfsd4_op_u *u)
3944 struct nfsd4_setclientid *setclid = &u->setclientid;
3945 struct xdr_netobj clname = setclid->se_name;
3946 nfs4_verifier clverifier = setclid->se_verf;
3947 struct nfs4_client *conf, *new;
3948 struct nfs4_client *unconf = NULL;
3950 struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
3952 new = create_client(clname, rqstp, &clverifier);
3954 return nfserr_jukebox;
3955 /* Cases below refer to rfc 3530 section 14.2.33: */
3956 spin_lock(&nn->client_lock);
3957 conf = find_confirmed_client_by_name(&clname, nn);
3958 if (conf && client_has_state(conf)) {
3960 status = nfserr_clid_inuse;
3961 if (clp_used_exchangeid(conf))
3963 if (!same_creds(&conf->cl_cred, &rqstp->rq_cred)) {
3964 trace_nfsd_clid_inuse_err(conf);
3968 unconf = find_unconfirmed_client_by_name(&clname, nn);
3970 unhash_client_locked(unconf);
3971 /* We need to handle only case 1: probable callback update */
3972 if (conf && same_verf(&conf->cl_verifier, &clverifier)) {
3973 copy_clid(new, conf);
3974 gen_confirm(new, nn);
3976 new->cl_minorversion = 0;
3977 gen_callback(new, setclid, rqstp);
3978 add_to_unconfirmed(new);
3979 setclid->se_clientid.cl_boot = new->cl_clientid.cl_boot;
3980 setclid->se_clientid.cl_id = new->cl_clientid.cl_id;
3981 memcpy(setclid->se_confirm.data, new->cl_confirm.data, sizeof(setclid->se_confirm.data));
3985 spin_unlock(&nn->client_lock);
3989 expire_client(unconf);
3995 nfsd4_setclientid_confirm(struct svc_rqst *rqstp,
3996 struct nfsd4_compound_state *cstate,
3997 union nfsd4_op_u *u)
3999 struct nfsd4_setclientid_confirm *setclientid_confirm =
4000 &u->setclientid_confirm;
4001 struct nfs4_client *conf, *unconf;
4002 struct nfs4_client *old = NULL;
4003 nfs4_verifier confirm = setclientid_confirm->sc_confirm;
4004 clientid_t * clid = &setclientid_confirm->sc_clientid;
4006 struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
4008 if (STALE_CLIENTID(clid, nn))
4009 return nfserr_stale_clientid;
4011 spin_lock(&nn->client_lock);
4012 conf = find_confirmed_client(clid, false, nn);
4013 unconf = find_unconfirmed_client(clid, false, nn);
4015 * We try hard to give out unique clientid's, so if we get an
4016 * attempt to confirm the same clientid with a different cred,
4017 * the client may be buggy; this should never happen.
4019 * Nevertheless, RFC 7530 recommends INUSE for this case:
4021 status = nfserr_clid_inuse;
4022 if (unconf && !same_creds(&unconf->cl_cred, &rqstp->rq_cred))
4024 if (conf && !same_creds(&conf->cl_cred, &rqstp->rq_cred))
4026 /* cases below refer to rfc 3530 section 14.2.34: */
4027 if (!unconf || !same_verf(&confirm, &unconf->cl_confirm)) {
4028 if (conf && same_verf(&confirm, &conf->cl_confirm)) {
4029 /* case 2: probable retransmit */
4031 } else /* case 4: client hasn't noticed we rebooted yet? */
4032 status = nfserr_stale_clientid;
4036 if (conf) { /* case 1: callback update */
4038 unhash_client_locked(old);
4039 nfsd4_change_callback(conf, &unconf->cl_cb_conn);
4040 } else { /* case 3: normal case; new or rebooted client */
4041 old = find_confirmed_client_by_name(&unconf->cl_name, nn);
4043 status = nfserr_clid_inuse;
4044 if (client_has_state(old)
4045 && !same_creds(&unconf->cl_cred,
4048 status = mark_client_expired_locked(old);
4054 move_to_confirmed(unconf);
4057 get_client_locked(conf);
4058 spin_unlock(&nn->client_lock);
4059 nfsd4_probe_callback(conf);
4060 spin_lock(&nn->client_lock);
4061 put_client_renew_locked(conf);
4063 spin_unlock(&nn->client_lock);
4069 static struct nfs4_file *nfsd4_alloc_file(void)
4071 return kmem_cache_alloc(file_slab, GFP_KERNEL);
4074 /* OPEN Share state helper functions */
4075 static void nfsd4_init_file(struct knfsd_fh *fh, unsigned int hashval,
4076 struct nfs4_file *fp)
4078 lockdep_assert_held(&state_lock);
4080 refcount_set(&fp->fi_ref, 1);
4081 spin_lock_init(&fp->fi_lock);
4082 INIT_LIST_HEAD(&fp->fi_stateids);
4083 INIT_LIST_HEAD(&fp->fi_delegations);
4084 INIT_LIST_HEAD(&fp->fi_clnt_odstate);
4085 fh_copy_shallow(&fp->fi_fhandle, fh);
4086 fp->fi_deleg_file = NULL;
4087 fp->fi_had_conflict = false;
4088 fp->fi_share_deny = 0;
4089 memset(fp->fi_fds, 0, sizeof(fp->fi_fds));
4090 memset(fp->fi_access, 0, sizeof(fp->fi_access));
4091 #ifdef CONFIG_NFSD_PNFS
4092 INIT_LIST_HEAD(&fp->fi_lo_states);
4093 atomic_set(&fp->fi_lo_recalls, 0);
4095 hlist_add_head_rcu(&fp->fi_hash, &file_hashtbl[hashval]);
4099 nfsd4_free_slabs(void)
4101 kmem_cache_destroy(client_slab);
4102 kmem_cache_destroy(openowner_slab);
4103 kmem_cache_destroy(lockowner_slab);
4104 kmem_cache_destroy(file_slab);
4105 kmem_cache_destroy(stateid_slab);
4106 kmem_cache_destroy(deleg_slab);
4107 kmem_cache_destroy(odstate_slab);
4111 nfsd4_init_slabs(void)
4113 client_slab = kmem_cache_create("nfsd4_clients",
4114 sizeof(struct nfs4_client), 0, 0, NULL);
4115 if (client_slab == NULL)
4117 openowner_slab = kmem_cache_create("nfsd4_openowners",
4118 sizeof(struct nfs4_openowner), 0, 0, NULL);
4119 if (openowner_slab == NULL)
4120 goto out_free_client_slab;
4121 lockowner_slab = kmem_cache_create("nfsd4_lockowners",
4122 sizeof(struct nfs4_lockowner), 0, 0, NULL);
4123 if (lockowner_slab == NULL)
4124 goto out_free_openowner_slab;
4125 file_slab = kmem_cache_create("nfsd4_files",
4126 sizeof(struct nfs4_file), 0, 0, NULL);
4127 if (file_slab == NULL)
4128 goto out_free_lockowner_slab;
4129 stateid_slab = kmem_cache_create("nfsd4_stateids",
4130 sizeof(struct nfs4_ol_stateid), 0, 0, NULL);
4131 if (stateid_slab == NULL)
4132 goto out_free_file_slab;
4133 deleg_slab = kmem_cache_create("nfsd4_delegations",
4134 sizeof(struct nfs4_delegation), 0, 0, NULL);
4135 if (deleg_slab == NULL)
4136 goto out_free_stateid_slab;
4137 odstate_slab = kmem_cache_create("nfsd4_odstate",
4138 sizeof(struct nfs4_clnt_odstate), 0, 0, NULL);
4139 if (odstate_slab == NULL)
4140 goto out_free_deleg_slab;
4143 out_free_deleg_slab:
4144 kmem_cache_destroy(deleg_slab);
4145 out_free_stateid_slab:
4146 kmem_cache_destroy(stateid_slab);
4148 kmem_cache_destroy(file_slab);
4149 out_free_lockowner_slab:
4150 kmem_cache_destroy(lockowner_slab);
4151 out_free_openowner_slab:
4152 kmem_cache_destroy(openowner_slab);
4153 out_free_client_slab:
4154 kmem_cache_destroy(client_slab);
4159 static void init_nfs4_replay(struct nfs4_replay *rp)
4161 rp->rp_status = nfserr_serverfault;
4163 rp->rp_buf = rp->rp_ibuf;
4164 mutex_init(&rp->rp_mutex);
4167 static void nfsd4_cstate_assign_replay(struct nfsd4_compound_state *cstate,
4168 struct nfs4_stateowner *so)
4170 if (!nfsd4_has_session(cstate)) {
4171 mutex_lock(&so->so_replay.rp_mutex);
4172 cstate->replay_owner = nfs4_get_stateowner(so);
4176 void nfsd4_cstate_clear_replay(struct nfsd4_compound_state *cstate)
4178 struct nfs4_stateowner *so = cstate->replay_owner;
4181 cstate->replay_owner = NULL;
4182 mutex_unlock(&so->so_replay.rp_mutex);
4183 nfs4_put_stateowner(so);
4187 static inline void *alloc_stateowner(struct kmem_cache *slab, struct xdr_netobj *owner, struct nfs4_client *clp)
4189 struct nfs4_stateowner *sop;
4191 sop = kmem_cache_alloc(slab, GFP_KERNEL);
4195 xdr_netobj_dup(&sop->so_owner, owner, GFP_KERNEL);
4196 if (!sop->so_owner.data) {
4197 kmem_cache_free(slab, sop);
4201 INIT_LIST_HEAD(&sop->so_stateids);
4202 sop->so_client = clp;
4203 init_nfs4_replay(&sop->so_replay);
4204 atomic_set(&sop->so_count, 1);
4208 static void hash_openowner(struct nfs4_openowner *oo, struct nfs4_client *clp, unsigned int strhashval)
4210 lockdep_assert_held(&clp->cl_lock);
4212 list_add(&oo->oo_owner.so_strhash,
4213 &clp->cl_ownerstr_hashtbl[strhashval]);
4214 list_add(&oo->oo_perclient, &clp->cl_openowners);
4217 static void nfs4_unhash_openowner(struct nfs4_stateowner *so)
4219 unhash_openowner_locked(openowner(so));
4222 static void nfs4_free_openowner(struct nfs4_stateowner *so)
4224 struct nfs4_openowner *oo = openowner(so);
4226 kmem_cache_free(openowner_slab, oo);
4229 static const struct nfs4_stateowner_operations openowner_ops = {
4230 .so_unhash = nfs4_unhash_openowner,
4231 .so_free = nfs4_free_openowner,
4234 static struct nfs4_ol_stateid *
4235 nfsd4_find_existing_open(struct nfs4_file *fp, struct nfsd4_open *open)
4237 struct nfs4_ol_stateid *local, *ret = NULL;
4238 struct nfs4_openowner *oo = open->op_openowner;
4240 lockdep_assert_held(&fp->fi_lock);
4242 list_for_each_entry(local, &fp->fi_stateids, st_perfile) {
4243 /* ignore lock owners */
4244 if (local->st_stateowner->so_is_open_owner == 0)
4246 if (local->st_stateowner != &oo->oo_owner)
4248 if (local->st_stid.sc_type == NFS4_OPEN_STID) {
4250 refcount_inc(&ret->st_stid.sc_count);
4258 nfsd4_verify_open_stid(struct nfs4_stid *s)
4260 __be32 ret = nfs_ok;
4262 switch (s->sc_type) {
4266 case NFS4_CLOSED_STID:
4267 case NFS4_CLOSED_DELEG_STID:
4268 ret = nfserr_bad_stateid;
4270 case NFS4_REVOKED_DELEG_STID:
4271 ret = nfserr_deleg_revoked;
4276 /* Lock the stateid st_mutex, and deal with races with CLOSE */
4278 nfsd4_lock_ol_stateid(struct nfs4_ol_stateid *stp)
4282 mutex_lock_nested(&stp->st_mutex, LOCK_STATEID_MUTEX);
4283 ret = nfsd4_verify_open_stid(&stp->st_stid);
4285 mutex_unlock(&stp->st_mutex);
4289 static struct nfs4_ol_stateid *
4290 nfsd4_find_and_lock_existing_open(struct nfs4_file *fp, struct nfsd4_open *open)
4292 struct nfs4_ol_stateid *stp;
4294 spin_lock(&fp->fi_lock);
4295 stp = nfsd4_find_existing_open(fp, open);
4296 spin_unlock(&fp->fi_lock);
4297 if (!stp || nfsd4_lock_ol_stateid(stp) == nfs_ok)
4299 nfs4_put_stid(&stp->st_stid);
4304 static struct nfs4_openowner *
4305 alloc_init_open_stateowner(unsigned int strhashval, struct nfsd4_open *open,
4306 struct nfsd4_compound_state *cstate)
4308 struct nfs4_client *clp = cstate->clp;
4309 struct nfs4_openowner *oo, *ret;
4311 oo = alloc_stateowner(openowner_slab, &open->op_owner, clp);
4314 oo->oo_owner.so_ops = &openowner_ops;
4315 oo->oo_owner.so_is_open_owner = 1;
4316 oo->oo_owner.so_seqid = open->op_seqid;
4318 if (nfsd4_has_session(cstate))
4319 oo->oo_flags |= NFS4_OO_CONFIRMED;
4321 oo->oo_last_closed_stid = NULL;
4322 INIT_LIST_HEAD(&oo->oo_close_lru);
4323 spin_lock(&clp->cl_lock);
4324 ret = find_openstateowner_str_locked(strhashval, open, clp);
4326 hash_openowner(oo, clp, strhashval);
4329 nfs4_free_stateowner(&oo->oo_owner);
4331 spin_unlock(&clp->cl_lock);
4335 static struct nfs4_ol_stateid *
4336 init_open_stateid(struct nfs4_file *fp, struct nfsd4_open *open)
4339 struct nfs4_openowner *oo = open->op_openowner;
4340 struct nfs4_ol_stateid *retstp = NULL;
4341 struct nfs4_ol_stateid *stp;
4344 /* We are moving these outside of the spinlocks to avoid the warnings */
4345 mutex_init(&stp->st_mutex);
4346 mutex_lock_nested(&stp->st_mutex, OPEN_STATEID_MUTEX);
4349 spin_lock(&oo->oo_owner.so_client->cl_lock);
4350 spin_lock(&fp->fi_lock);
4352 retstp = nfsd4_find_existing_open(fp, open);
4356 open->op_stp = NULL;
4357 refcount_inc(&stp->st_stid.sc_count);
4358 stp->st_stid.sc_type = NFS4_OPEN_STID;
4359 INIT_LIST_HEAD(&stp->st_locks);
4360 stp->st_stateowner = nfs4_get_stateowner(&oo->oo_owner);
4362 stp->st_stid.sc_file = fp;
4363 stp->st_access_bmap = 0;
4364 stp->st_deny_bmap = 0;
4365 stp->st_openstp = NULL;
4366 list_add(&stp->st_perstateowner, &oo->oo_owner.so_stateids);
4367 list_add(&stp->st_perfile, &fp->fi_stateids);
4370 spin_unlock(&fp->fi_lock);
4371 spin_unlock(&oo->oo_owner.so_client->cl_lock);
4373 /* Handle races with CLOSE */
4374 if (nfsd4_lock_ol_stateid(retstp) != nfs_ok) {
4375 nfs4_put_stid(&retstp->st_stid);
4378 /* To keep mutex tracking happy */
4379 mutex_unlock(&stp->st_mutex);
4386 * In the 4.0 case we need to keep the owners around a little while to handle
4387 * CLOSE replay. We still do need to release any file access that is held by
4388 * them before returning however.
4391 move_to_close_lru(struct nfs4_ol_stateid *s, struct net *net)
4393 struct nfs4_ol_stateid *last;
4394 struct nfs4_openowner *oo = openowner(s->st_stateowner);
4395 struct nfsd_net *nn = net_generic(s->st_stid.sc_client->net,
4398 dprintk("NFSD: move_to_close_lru nfs4_openowner %p\n", oo);
4401 * We know that we hold one reference via nfsd4_close, and another
4402 * "persistent" reference for the client. If the refcount is higher
4403 * than 2, then there are still calls in progress that are using this
4404 * stateid. We can't put the sc_file reference until they are finished.
4405 * Wait for the refcount to drop to 2. Since it has been unhashed,
4406 * there should be no danger of the refcount going back up again at
4409 wait_event(close_wq, refcount_read(&s->st_stid.sc_count) == 2);
4411 release_all_access(s);
4412 if (s->st_stid.sc_file) {
4413 put_nfs4_file(s->st_stid.sc_file);
4414 s->st_stid.sc_file = NULL;
4417 spin_lock(&nn->client_lock);
4418 last = oo->oo_last_closed_stid;
4419 oo->oo_last_closed_stid = s;
4420 list_move_tail(&oo->oo_close_lru, &nn->close_lru);
4421 oo->oo_time = ktime_get_boottime_seconds();
4422 spin_unlock(&nn->client_lock);
4424 nfs4_put_stid(&last->st_stid);
4427 /* search file_hashtbl[] for file */
4428 static struct nfs4_file *
4429 find_file_locked(struct knfsd_fh *fh, unsigned int hashval)
4431 struct nfs4_file *fp;
4433 hlist_for_each_entry_rcu(fp, &file_hashtbl[hashval], fi_hash,
4434 lockdep_is_held(&state_lock)) {
4435 if (fh_match(&fp->fi_fhandle, fh)) {
4436 if (refcount_inc_not_zero(&fp->fi_ref))
4444 find_file(struct knfsd_fh *fh)
4446 struct nfs4_file *fp;
4447 unsigned int hashval = file_hashval(fh);
4450 fp = find_file_locked(fh, hashval);
4455 static struct nfs4_file *
4456 find_or_add_file(struct nfs4_file *new, struct knfsd_fh *fh)
4458 struct nfs4_file *fp;
4459 unsigned int hashval = file_hashval(fh);
4462 fp = find_file_locked(fh, hashval);
4467 spin_lock(&state_lock);
4468 fp = find_file_locked(fh, hashval);
4469 if (likely(fp == NULL)) {
4470 nfsd4_init_file(fh, hashval, new);
4473 spin_unlock(&state_lock);
4479 * Called to check deny when READ with all zero stateid or
4480 * WRITE with all zero or all one stateid
4483 nfs4_share_conflict(struct svc_fh *current_fh, unsigned int deny_type)
4485 struct nfs4_file *fp;
4486 __be32 ret = nfs_ok;
4488 fp = find_file(¤t_fh->fh_handle);
4491 /* Check for conflicting share reservations */
4492 spin_lock(&fp->fi_lock);
4493 if (fp->fi_share_deny & deny_type)
4494 ret = nfserr_locked;
4495 spin_unlock(&fp->fi_lock);
4500 static void nfsd4_cb_recall_prepare(struct nfsd4_callback *cb)
4502 struct nfs4_delegation *dp = cb_to_delegation(cb);
4503 struct nfsd_net *nn = net_generic(dp->dl_stid.sc_client->net,
4506 block_delegations(&dp->dl_stid.sc_file->fi_fhandle);
4509 * We can't do this in nfsd_break_deleg_cb because it is
4510 * already holding inode->i_lock.
4512 * If the dl_time != 0, then we know that it has already been
4513 * queued for a lease break. Don't queue it again.
4515 spin_lock(&state_lock);
4516 if (dp->dl_time == 0) {
4517 dp->dl_time = ktime_get_boottime_seconds();
4518 list_add_tail(&dp->dl_recall_lru, &nn->del_recall_lru);
4520 spin_unlock(&state_lock);
4523 static int nfsd4_cb_recall_done(struct nfsd4_callback *cb,
4524 struct rpc_task *task)
4526 struct nfs4_delegation *dp = cb_to_delegation(cb);
4528 if (dp->dl_stid.sc_type == NFS4_CLOSED_DELEG_STID ||
4529 dp->dl_stid.sc_type == NFS4_REVOKED_DELEG_STID)
4532 switch (task->tk_status) {
4535 case -NFS4ERR_DELAY:
4536 rpc_delay(task, 2 * HZ);
4539 case -NFS4ERR_BAD_STATEID:
4541 * Race: client probably got cb_recall before open reply
4542 * granting delegation.
4544 if (dp->dl_retries--) {
4545 rpc_delay(task, 2 * HZ);
4554 static void nfsd4_cb_recall_release(struct nfsd4_callback *cb)
4556 struct nfs4_delegation *dp = cb_to_delegation(cb);
4558 nfs4_put_stid(&dp->dl_stid);
4561 static const struct nfsd4_callback_ops nfsd4_cb_recall_ops = {
4562 .prepare = nfsd4_cb_recall_prepare,
4563 .done = nfsd4_cb_recall_done,
4564 .release = nfsd4_cb_recall_release,
4567 static void nfsd_break_one_deleg(struct nfs4_delegation *dp)
4570 * We're assuming the state code never drops its reference
4571 * without first removing the lease. Since we're in this lease
4572 * callback (and since the lease code is serialized by the
4573 * i_lock) we know the server hasn't removed the lease yet, and
4574 * we know it's safe to take a reference.
4576 refcount_inc(&dp->dl_stid.sc_count);
4577 nfsd4_run_cb(&dp->dl_recall);
4580 /* Called from break_lease() with i_lock held. */
4582 nfsd_break_deleg_cb(struct file_lock *fl)
4585 struct nfs4_delegation *dp = (struct nfs4_delegation *)fl->fl_owner;
4586 struct nfs4_file *fp = dp->dl_stid.sc_file;
4588 trace_nfsd_deleg_break(&dp->dl_stid.sc_stateid);
4591 * We don't want the locks code to timeout the lease for us;
4592 * we'll remove it ourself if a delegation isn't returned
4595 fl->fl_break_time = 0;
4597 spin_lock(&fp->fi_lock);
4598 fp->fi_had_conflict = true;
4599 nfsd_break_one_deleg(dp);
4600 spin_unlock(&fp->fi_lock);
4604 static bool nfsd_breaker_owns_lease(struct file_lock *fl)
4606 struct nfs4_delegation *dl = fl->fl_owner;
4607 struct svc_rqst *rqst;
4608 struct nfs4_client *clp;
4612 rqst = kthread_data(current);
4613 /* Note rq_prog == NFS_ACL_PROGRAM is also possible: */
4614 if (rqst->rq_prog != NFS_PROGRAM || rqst->rq_vers < 4)
4616 clp = *(rqst->rq_lease_breaker);
4617 return dl->dl_stid.sc_client == clp;
4621 nfsd_change_deleg_cb(struct file_lock *onlist, int arg,
4622 struct list_head *dispose)
4625 return lease_modify(onlist, arg, dispose);
4630 static const struct lock_manager_operations nfsd_lease_mng_ops = {
4631 .lm_breaker_owns_lease = nfsd_breaker_owns_lease,
4632 .lm_break = nfsd_break_deleg_cb,
4633 .lm_change = nfsd_change_deleg_cb,
4636 static __be32 nfsd4_check_seqid(struct nfsd4_compound_state *cstate, struct nfs4_stateowner *so, u32 seqid)
4638 if (nfsd4_has_session(cstate))
4640 if (seqid == so->so_seqid - 1)
4641 return nfserr_replay_me;
4642 if (seqid == so->so_seqid)
4644 return nfserr_bad_seqid;
4647 static struct nfs4_client *lookup_clientid(clientid_t *clid, bool sessions,
4648 struct nfsd_net *nn)
4650 struct nfs4_client *found;
4652 spin_lock(&nn->client_lock);
4653 found = find_confirmed_client(clid, sessions, nn);
4655 atomic_inc(&found->cl_rpc_users);
4656 spin_unlock(&nn->client_lock);
4660 static __be32 set_client(clientid_t *clid,
4661 struct nfsd4_compound_state *cstate,
4662 struct nfsd_net *nn)
4665 if (!same_clid(&cstate->clp->cl_clientid, clid))
4666 return nfserr_stale_clientid;
4669 if (STALE_CLIENTID(clid, nn))
4670 return nfserr_stale_clientid;
4672 * We're in the 4.0 case (otherwise the SEQUENCE op would have
4673 * set cstate->clp), so session = false:
4675 cstate->clp = lookup_clientid(clid, false, nn);
4677 return nfserr_expired;
4682 nfsd4_process_open1(struct nfsd4_compound_state *cstate,
4683 struct nfsd4_open *open, struct nfsd_net *nn)
4685 clientid_t *clientid = &open->op_clientid;
4686 struct nfs4_client *clp = NULL;
4687 unsigned int strhashval;
4688 struct nfs4_openowner *oo = NULL;
4692 * In case we need it later, after we've already created the
4693 * file and don't want to risk a further failure:
4695 open->op_file = nfsd4_alloc_file();
4696 if (open->op_file == NULL)
4697 return nfserr_jukebox;
4699 status = set_client(clientid, cstate, nn);
4704 strhashval = ownerstr_hashval(&open->op_owner);
4705 oo = find_openstateowner_str(strhashval, open, clp);
4706 open->op_openowner = oo;
4710 if (!(oo->oo_flags & NFS4_OO_CONFIRMED)) {
4711 /* Replace unconfirmed owners without checking for replay. */
4712 release_openowner(oo);
4713 open->op_openowner = NULL;
4716 status = nfsd4_check_seqid(cstate, &oo->oo_owner, open->op_seqid);
4721 oo = alloc_init_open_stateowner(strhashval, open, cstate);
4723 return nfserr_jukebox;
4724 open->op_openowner = oo;
4726 open->op_stp = nfs4_alloc_open_stateid(clp);
4728 return nfserr_jukebox;
4730 if (nfsd4_has_session(cstate) &&
4731 (cstate->current_fh.fh_export->ex_flags & NFSEXP_PNFS)) {
4732 open->op_odstate = alloc_clnt_odstate(clp);
4733 if (!open->op_odstate)
4734 return nfserr_jukebox;
4740 static inline __be32
4741 nfs4_check_delegmode(struct nfs4_delegation *dp, int flags)
4743 if ((flags & WR_STATE) && (dp->dl_type == NFS4_OPEN_DELEGATE_READ))
4744 return nfserr_openmode;
4749 static int share_access_to_flags(u32 share_access)
4751 return share_access == NFS4_SHARE_ACCESS_READ ? RD_STATE : WR_STATE;
4754 static struct nfs4_delegation *find_deleg_stateid(struct nfs4_client *cl, stateid_t *s)
4756 struct nfs4_stid *ret;
4758 ret = find_stateid_by_type(cl, s,
4759 NFS4_DELEG_STID|NFS4_REVOKED_DELEG_STID);
4762 return delegstateid(ret);
4765 static bool nfsd4_is_deleg_cur(struct nfsd4_open *open)
4767 return open->op_claim_type == NFS4_OPEN_CLAIM_DELEGATE_CUR ||
4768 open->op_claim_type == NFS4_OPEN_CLAIM_DELEG_CUR_FH;
4772 nfs4_check_deleg(struct nfs4_client *cl, struct nfsd4_open *open,
4773 struct nfs4_delegation **dp)
4776 __be32 status = nfserr_bad_stateid;
4777 struct nfs4_delegation *deleg;
4779 deleg = find_deleg_stateid(cl, &open->op_delegate_stateid);
4782 if (deleg->dl_stid.sc_type == NFS4_REVOKED_DELEG_STID) {
4783 nfs4_put_stid(&deleg->dl_stid);
4784 if (cl->cl_minorversion)
4785 status = nfserr_deleg_revoked;
4788 flags = share_access_to_flags(open->op_share_access);
4789 status = nfs4_check_delegmode(deleg, flags);
4791 nfs4_put_stid(&deleg->dl_stid);
4796 if (!nfsd4_is_deleg_cur(open))
4800 open->op_openowner->oo_flags |= NFS4_OO_CONFIRMED;
4804 static inline int nfs4_access_to_access(u32 nfs4_access)
4808 if (nfs4_access & NFS4_SHARE_ACCESS_READ)
4809 flags |= NFSD_MAY_READ;
4810 if (nfs4_access & NFS4_SHARE_ACCESS_WRITE)
4811 flags |= NFSD_MAY_WRITE;
4815 static inline __be32
4816 nfsd4_truncate(struct svc_rqst *rqstp, struct svc_fh *fh,
4817 struct nfsd4_open *open)
4819 struct iattr iattr = {
4820 .ia_valid = ATTR_SIZE,
4823 if (!open->op_truncate)
4825 if (!(open->op_share_access & NFS4_SHARE_ACCESS_WRITE))
4826 return nfserr_inval;
4827 return nfsd_setattr(rqstp, fh, &iattr, 0, (time64_t)0);
4830 static __be32 nfs4_get_vfs_file(struct svc_rqst *rqstp, struct nfs4_file *fp,
4831 struct svc_fh *cur_fh, struct nfs4_ol_stateid *stp,
4832 struct nfsd4_open *open)
4834 struct nfsd_file *nf = NULL;
4836 int oflag = nfs4_access_to_omode(open->op_share_access);
4837 int access = nfs4_access_to_access(open->op_share_access);
4838 unsigned char old_access_bmap, old_deny_bmap;
4840 spin_lock(&fp->fi_lock);
4843 * Are we trying to set a deny mode that would conflict with
4846 status = nfs4_file_check_deny(fp, open->op_share_deny);
4847 if (status != nfs_ok) {
4848 spin_unlock(&fp->fi_lock);
4852 /* set access to the file */
4853 status = nfs4_file_get_access(fp, open->op_share_access);
4854 if (status != nfs_ok) {
4855 spin_unlock(&fp->fi_lock);
4859 /* Set access bits in stateid */
4860 old_access_bmap = stp->st_access_bmap;
4861 set_access(open->op_share_access, stp);
4863 /* Set new deny mask */
4864 old_deny_bmap = stp->st_deny_bmap;
4865 set_deny(open->op_share_deny, stp);
4866 fp->fi_share_deny |= (open->op_share_deny & NFS4_SHARE_DENY_BOTH);
4868 if (!fp->fi_fds[oflag]) {
4869 spin_unlock(&fp->fi_lock);
4870 status = nfsd_file_acquire(rqstp, cur_fh, access, &nf);
4872 goto out_put_access;
4873 spin_lock(&fp->fi_lock);
4874 if (!fp->fi_fds[oflag]) {
4875 fp->fi_fds[oflag] = nf;
4879 spin_unlock(&fp->fi_lock);
4883 status = nfsd4_truncate(rqstp, cur_fh, open);
4885 goto out_put_access;
4889 stp->st_access_bmap = old_access_bmap;
4890 nfs4_file_put_access(fp, open->op_share_access);
4891 reset_union_bmap_deny(bmap_to_share_mode(old_deny_bmap), stp);
4896 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)
4899 unsigned char old_deny_bmap = stp->st_deny_bmap;
4901 if (!test_access(open->op_share_access, stp))
4902 return nfs4_get_vfs_file(rqstp, fp, cur_fh, stp, open);
4904 /* test and set deny mode */
4905 spin_lock(&fp->fi_lock);
4906 status = nfs4_file_check_deny(fp, open->op_share_deny);
4907 if (status == nfs_ok) {
4908 set_deny(open->op_share_deny, stp);
4909 fp->fi_share_deny |=
4910 (open->op_share_deny & NFS4_SHARE_DENY_BOTH);
4912 spin_unlock(&fp->fi_lock);
4914 if (status != nfs_ok)
4917 status = nfsd4_truncate(rqstp, cur_fh, open);
4918 if (status != nfs_ok)
4919 reset_union_bmap_deny(old_deny_bmap, stp);
4923 /* Should we give out recallable state?: */
4924 static bool nfsd4_cb_channel_good(struct nfs4_client *clp)
4926 if (clp->cl_cb_state == NFSD4_CB_UP)
4929 * In the sessions case, since we don't have to establish a
4930 * separate connection for callbacks, we assume it's OK
4931 * until we hear otherwise:
4933 return clp->cl_minorversion && clp->cl_cb_state == NFSD4_CB_UNKNOWN;
4936 static struct file_lock *nfs4_alloc_init_lease(struct nfs4_delegation *dp,
4939 struct file_lock *fl;
4941 fl = locks_alloc_lock();
4944 fl->fl_lmops = &nfsd_lease_mng_ops;
4945 fl->fl_flags = FL_DELEG;
4946 fl->fl_type = flag == NFS4_OPEN_DELEGATE_READ? F_RDLCK: F_WRLCK;
4947 fl->fl_end = OFFSET_MAX;
4948 fl->fl_owner = (fl_owner_t)dp;
4949 fl->fl_pid = current->tgid;
4950 fl->fl_file = dp->dl_stid.sc_file->fi_deleg_file->nf_file;
4954 static struct nfs4_delegation *
4955 nfs4_set_delegation(struct nfs4_client *clp, struct svc_fh *fh,
4956 struct nfs4_file *fp, struct nfs4_clnt_odstate *odstate)
4959 struct nfs4_delegation *dp;
4960 struct nfsd_file *nf;
4961 struct file_lock *fl;
4964 * The fi_had_conflict and nfs_get_existing_delegation checks
4965 * here are just optimizations; we'll need to recheck them at
4968 if (fp->fi_had_conflict)
4969 return ERR_PTR(-EAGAIN);
4971 nf = find_readable_file(fp);
4973 /* We should always have a readable file here */
4975 return ERR_PTR(-EBADF);
4977 spin_lock(&state_lock);
4978 spin_lock(&fp->fi_lock);
4979 if (nfs4_delegation_exists(clp, fp))
4981 else if (!fp->fi_deleg_file) {
4982 fp->fi_deleg_file = nf;
4983 /* increment early to prevent fi_deleg_file from being
4985 fp->fi_delegees = 1;
4989 spin_unlock(&fp->fi_lock);
4990 spin_unlock(&state_lock);
4994 return ERR_PTR(status);
4997 dp = alloc_init_deleg(clp, fp, fh, odstate);
5001 fl = nfs4_alloc_init_lease(dp, NFS4_OPEN_DELEGATE_READ);
5003 goto out_clnt_odstate;
5005 status = vfs_setlease(fp->fi_deleg_file->nf_file, fl->fl_type, &fl, NULL);
5007 locks_free_lock(fl);
5009 goto out_clnt_odstate;
5011 spin_lock(&state_lock);
5012 spin_lock(&fp->fi_lock);
5013 if (fp->fi_had_conflict)
5016 status = hash_delegation_locked(dp, fp);
5017 spin_unlock(&fp->fi_lock);
5018 spin_unlock(&state_lock);
5025 vfs_setlease(fp->fi_deleg_file->nf_file, F_UNLCK, NULL, (void **)&dp);
5027 put_clnt_odstate(dp->dl_clnt_odstate);
5028 nfs4_put_stid(&dp->dl_stid);
5031 return ERR_PTR(status);
5034 static void nfsd4_open_deleg_none_ext(struct nfsd4_open *open, int status)
5036 open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE_EXT;
5037 if (status == -EAGAIN)
5038 open->op_why_no_deleg = WND4_CONTENTION;
5040 open->op_why_no_deleg = WND4_RESOURCE;
5041 switch (open->op_deleg_want) {
5042 case NFS4_SHARE_WANT_READ_DELEG:
5043 case NFS4_SHARE_WANT_WRITE_DELEG:
5044 case NFS4_SHARE_WANT_ANY_DELEG:
5046 case NFS4_SHARE_WANT_CANCEL:
5047 open->op_why_no_deleg = WND4_CANCELLED;
5049 case NFS4_SHARE_WANT_NO_DELEG:
5056 * Attempt to hand out a delegation.
5058 * Note we don't support write delegations, and won't until the vfs has
5059 * proper support for them.
5062 nfs4_open_delegation(struct svc_fh *fh, struct nfsd4_open *open,
5063 struct nfs4_ol_stateid *stp)
5065 struct nfs4_delegation *dp;
5066 struct nfs4_openowner *oo = openowner(stp->st_stateowner);
5067 struct nfs4_client *clp = stp->st_stid.sc_client;
5071 cb_up = nfsd4_cb_channel_good(oo->oo_owner.so_client);
5072 open->op_recall = 0;
5073 switch (open->op_claim_type) {
5074 case NFS4_OPEN_CLAIM_PREVIOUS:
5076 open->op_recall = 1;
5077 if (open->op_delegate_type != NFS4_OPEN_DELEGATE_READ)
5080 case NFS4_OPEN_CLAIM_NULL:
5081 case NFS4_OPEN_CLAIM_FH:
5083 * Let's not give out any delegations till everyone's
5084 * had the chance to reclaim theirs, *and* until
5085 * NLM locks have all been reclaimed:
5087 if (locks_in_grace(clp->net))
5089 if (!cb_up || !(oo->oo_flags & NFS4_OO_CONFIRMED))
5092 * Also, if the file was opened for write or
5093 * create, there's a good chance the client's
5094 * about to write to it, resulting in an
5095 * immediate recall (since we don't support
5096 * write delegations):
5098 if (open->op_share_access & NFS4_SHARE_ACCESS_WRITE)
5100 if (open->op_create == NFS4_OPEN_CREATE)
5106 dp = nfs4_set_delegation(clp, fh, stp->st_stid.sc_file, stp->st_clnt_odstate);
5110 memcpy(&open->op_delegate_stateid, &dp->dl_stid.sc_stateid, sizeof(dp->dl_stid.sc_stateid));
5112 trace_nfsd_deleg_read(&dp->dl_stid.sc_stateid);
5113 open->op_delegate_type = NFS4_OPEN_DELEGATE_READ;
5114 nfs4_put_stid(&dp->dl_stid);
5117 open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE;
5118 if (open->op_claim_type == NFS4_OPEN_CLAIM_PREVIOUS &&
5119 open->op_delegate_type != NFS4_OPEN_DELEGATE_NONE) {
5120 dprintk("NFSD: WARNING: refusing delegation reclaim\n");
5121 open->op_recall = 1;
5124 /* 4.1 client asking for a delegation? */
5125 if (open->op_deleg_want)
5126 nfsd4_open_deleg_none_ext(open, status);
5130 static void nfsd4_deleg_xgrade_none_ext(struct nfsd4_open *open,
5131 struct nfs4_delegation *dp)
5133 if (open->op_deleg_want == NFS4_SHARE_WANT_READ_DELEG &&
5134 dp->dl_type == NFS4_OPEN_DELEGATE_WRITE) {
5135 open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE_EXT;
5136 open->op_why_no_deleg = WND4_NOT_SUPP_DOWNGRADE;
5137 } else if (open->op_deleg_want == NFS4_SHARE_WANT_WRITE_DELEG &&
5138 dp->dl_type == NFS4_OPEN_DELEGATE_WRITE) {
5139 open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE_EXT;
5140 open->op_why_no_deleg = WND4_NOT_SUPP_UPGRADE;
5142 /* Otherwise the client must be confused wanting a delegation
5143 * it already has, therefore we don't return
5144 * NFS4_OPEN_DELEGATE_NONE_EXT and reason.
5149 nfsd4_process_open2(struct svc_rqst *rqstp, struct svc_fh *current_fh, struct nfsd4_open *open)
5151 struct nfsd4_compoundres *resp = rqstp->rq_resp;
5152 struct nfs4_client *cl = open->op_openowner->oo_owner.so_client;
5153 struct nfs4_file *fp = NULL;
5154 struct nfs4_ol_stateid *stp = NULL;
5155 struct nfs4_delegation *dp = NULL;
5157 bool new_stp = false;
5160 * Lookup file; if found, lookup stateid and check open request,
5161 * and check for delegations in the process of being recalled.
5162 * If not found, create the nfs4_file struct
5164 fp = find_or_add_file(open->op_file, ¤t_fh->fh_handle);
5165 if (fp != open->op_file) {
5166 status = nfs4_check_deleg(cl, open, &dp);
5169 stp = nfsd4_find_and_lock_existing_open(fp, open);
5171 open->op_file = NULL;
5172 status = nfserr_bad_stateid;
5173 if (nfsd4_is_deleg_cur(open))
5178 stp = init_open_stateid(fp, open);
5184 * OPEN the file, or upgrade an existing OPEN.
5185 * If truncate fails, the OPEN fails.
5187 * stp is already locked.
5190 /* Stateid was found, this is an OPEN upgrade */
5191 status = nfs4_upgrade_open(rqstp, fp, current_fh, stp, open);
5193 mutex_unlock(&stp->st_mutex);
5197 status = nfs4_get_vfs_file(rqstp, fp, current_fh, stp, open);
5199 stp->st_stid.sc_type = NFS4_CLOSED_STID;
5200 release_open_stateid(stp);
5201 mutex_unlock(&stp->st_mutex);
5205 stp->st_clnt_odstate = find_or_hash_clnt_odstate(fp,
5207 if (stp->st_clnt_odstate == open->op_odstate)
5208 open->op_odstate = NULL;
5211 nfs4_inc_and_copy_stateid(&open->op_stateid, &stp->st_stid);
5212 mutex_unlock(&stp->st_mutex);
5214 if (nfsd4_has_session(&resp->cstate)) {
5215 if (open->op_deleg_want & NFS4_SHARE_WANT_NO_DELEG) {
5216 open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE_EXT;
5217 open->op_why_no_deleg = WND4_NOT_WANTED;
5223 * Attempt to hand out a delegation. No error return, because the
5224 * OPEN succeeds even if we fail.
5226 nfs4_open_delegation(current_fh, open, stp);
5229 trace_nfsd_open(&stp->st_stid.sc_stateid);
5231 /* 4.1 client trying to upgrade/downgrade delegation? */
5232 if (open->op_delegate_type == NFS4_OPEN_DELEGATE_NONE && dp &&
5233 open->op_deleg_want)
5234 nfsd4_deleg_xgrade_none_ext(open, dp);
5238 if (status == 0 && open->op_claim_type == NFS4_OPEN_CLAIM_PREVIOUS)
5239 open->op_openowner->oo_flags |= NFS4_OO_CONFIRMED;
5241 * To finish the open response, we just need to set the rflags.
5243 open->op_rflags = NFS4_OPEN_RESULT_LOCKTYPE_POSIX;
5244 if (nfsd4_has_session(&resp->cstate))
5245 open->op_rflags |= NFS4_OPEN_RESULT_MAY_NOTIFY_LOCK;
5246 else if (!(open->op_openowner->oo_flags & NFS4_OO_CONFIRMED))
5247 open->op_rflags |= NFS4_OPEN_RESULT_CONFIRM;
5250 nfs4_put_stid(&dp->dl_stid);
5252 nfs4_put_stid(&stp->st_stid);
5257 void nfsd4_cleanup_open_state(struct nfsd4_compound_state *cstate,
5258 struct nfsd4_open *open)
5260 if (open->op_openowner) {
5261 struct nfs4_stateowner *so = &open->op_openowner->oo_owner;
5263 nfsd4_cstate_assign_replay(cstate, so);
5264 nfs4_put_stateowner(so);
5267 kmem_cache_free(file_slab, open->op_file);
5269 nfs4_put_stid(&open->op_stp->st_stid);
5270 if (open->op_odstate)
5271 kmem_cache_free(odstate_slab, open->op_odstate);
5275 nfsd4_renew(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
5276 union nfsd4_op_u *u)
5278 clientid_t *clid = &u->renew;
5279 struct nfs4_client *clp;
5281 struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
5283 trace_nfsd_clid_renew(clid);
5284 status = set_client(clid, cstate, nn);
5288 if (!list_empty(&clp->cl_delegations)
5289 && clp->cl_cb_state != NFSD4_CB_UP)
5290 return nfserr_cb_path_down;
5295 nfsd4_end_grace(struct nfsd_net *nn)
5297 /* do nothing if grace period already ended */
5298 if (nn->grace_ended)
5301 trace_nfsd_grace_complete(nn);
5302 nn->grace_ended = true;
5304 * If the server goes down again right now, an NFSv4
5305 * client will still be allowed to reclaim after it comes back up,
5306 * even if it hasn't yet had a chance to reclaim state this time.
5309 nfsd4_record_grace_done(nn);
5311 * At this point, NFSv4 clients can still reclaim. But if the
5312 * server crashes, any that have not yet reclaimed will be out
5313 * of luck on the next boot.
5315 * (NFSv4.1+ clients are considered to have reclaimed once they
5316 * call RECLAIM_COMPLETE. NFSv4.0 clients are considered to
5317 * have reclaimed after their first OPEN.)
5319 locks_end_grace(&nn->nfsd4_manager);
5321 * At this point, and once lockd and/or any other containers
5322 * exit their grace period, further reclaims will fail and
5323 * regular locking can resume.
5328 * If we've waited a lease period but there are still clients trying to
5329 * reclaim, wait a little longer to give them a chance to finish.
5331 static bool clients_still_reclaiming(struct nfsd_net *nn)
5333 time64_t double_grace_period_end = nn->boot_time +
5334 2 * nn->nfsd4_lease;
5336 if (nn->track_reclaim_completes &&
5337 atomic_read(&nn->nr_reclaim_complete) ==
5338 nn->reclaim_str_hashtbl_size)
5340 if (!nn->somebody_reclaimed)
5342 nn->somebody_reclaimed = false;
5344 * If we've given them *two* lease times to reclaim, and they're
5345 * still not done, give up:
5347 if (ktime_get_boottime_seconds() > double_grace_period_end)
5352 struct laundry_time {
5357 static bool state_expired(struct laundry_time *lt, time64_t last_refresh)
5359 time64_t time_remaining;
5361 if (last_refresh < lt->cutoff)
5363 time_remaining = last_refresh - lt->cutoff;
5364 lt->new_timeo = min(lt->new_timeo, time_remaining);
5369 nfs4_laundromat(struct nfsd_net *nn)
5371 struct nfs4_client *clp;
5372 struct nfs4_openowner *oo;
5373 struct nfs4_delegation *dp;
5374 struct nfs4_ol_stateid *stp;
5375 struct nfsd4_blocked_lock *nbl;
5376 struct list_head *pos, *next, reaplist;
5377 struct laundry_time lt = {
5378 .cutoff = ktime_get_boottime_seconds() - nn->nfsd4_lease,
5379 .new_timeo = nn->nfsd4_lease
5381 struct nfs4_cpntf_state *cps;
5382 copy_stateid_t *cps_t;
5385 if (clients_still_reclaiming(nn)) {
5389 nfsd4_end_grace(nn);
5390 INIT_LIST_HEAD(&reaplist);
5392 spin_lock(&nn->s2s_cp_lock);
5393 idr_for_each_entry(&nn->s2s_cp_stateids, cps_t, i) {
5394 cps = container_of(cps_t, struct nfs4_cpntf_state, cp_stateid);
5395 if (cps->cp_stateid.sc_type == NFS4_COPYNOTIFY_STID &&
5396 state_expired(<, cps->cpntf_time))
5397 _free_cpntf_state_locked(nn, cps);
5399 spin_unlock(&nn->s2s_cp_lock);
5401 spin_lock(&nn->client_lock);
5402 list_for_each_safe(pos, next, &nn->client_lru) {
5403 clp = list_entry(pos, struct nfs4_client, cl_lru);
5404 if (!state_expired(<, clp->cl_time))
5406 if (mark_client_expired_locked(clp)) {
5407 trace_nfsd_clid_expired(&clp->cl_clientid);
5410 list_add(&clp->cl_lru, &reaplist);
5412 spin_unlock(&nn->client_lock);
5413 list_for_each_safe(pos, next, &reaplist) {
5414 clp = list_entry(pos, struct nfs4_client, cl_lru);
5415 trace_nfsd_clid_purged(&clp->cl_clientid);
5416 list_del_init(&clp->cl_lru);
5419 spin_lock(&state_lock);
5420 list_for_each_safe(pos, next, &nn->del_recall_lru) {
5421 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
5422 if (!state_expired(<, dp->dl_time))
5424 WARN_ON(!unhash_delegation_locked(dp));
5425 list_add(&dp->dl_recall_lru, &reaplist);
5427 spin_unlock(&state_lock);
5428 while (!list_empty(&reaplist)) {
5429 dp = list_first_entry(&reaplist, struct nfs4_delegation,
5431 list_del_init(&dp->dl_recall_lru);
5432 revoke_delegation(dp);
5435 spin_lock(&nn->client_lock);
5436 while (!list_empty(&nn->close_lru)) {
5437 oo = list_first_entry(&nn->close_lru, struct nfs4_openowner,
5439 if (!state_expired(<, oo->oo_time))
5441 list_del_init(&oo->oo_close_lru);
5442 stp = oo->oo_last_closed_stid;
5443 oo->oo_last_closed_stid = NULL;
5444 spin_unlock(&nn->client_lock);
5445 nfs4_put_stid(&stp->st_stid);
5446 spin_lock(&nn->client_lock);
5448 spin_unlock(&nn->client_lock);
5451 * It's possible for a client to try and acquire an already held lock
5452 * that is being held for a long time, and then lose interest in it.
5453 * So, we clean out any un-revisited request after a lease period
5454 * under the assumption that the client is no longer interested.
5456 * RFC5661, sec. 9.6 states that the client must not rely on getting
5457 * notifications and must continue to poll for locks, even when the
5458 * server supports them. Thus this shouldn't lead to clients blocking
5459 * indefinitely once the lock does become free.
5461 BUG_ON(!list_empty(&reaplist));
5462 spin_lock(&nn->blocked_locks_lock);
5463 while (!list_empty(&nn->blocked_locks_lru)) {
5464 nbl = list_first_entry(&nn->blocked_locks_lru,
5465 struct nfsd4_blocked_lock, nbl_lru);
5466 if (!state_expired(<, nbl->nbl_time))
5468 list_move(&nbl->nbl_lru, &reaplist);
5469 list_del_init(&nbl->nbl_list);
5471 spin_unlock(&nn->blocked_locks_lock);
5473 while (!list_empty(&reaplist)) {
5474 nbl = list_first_entry(&reaplist,
5475 struct nfsd4_blocked_lock, nbl_lru);
5476 list_del_init(&nbl->nbl_lru);
5477 free_blocked_lock(nbl);
5480 return max_t(time64_t, lt.new_timeo, NFSD_LAUNDROMAT_MINTIMEOUT);
5483 static struct workqueue_struct *laundry_wq;
5484 static void laundromat_main(struct work_struct *);
5487 laundromat_main(struct work_struct *laundry)
5490 struct delayed_work *dwork = to_delayed_work(laundry);
5491 struct nfsd_net *nn = container_of(dwork, struct nfsd_net,
5494 t = nfs4_laundromat(nn);
5495 queue_delayed_work(laundry_wq, &nn->laundromat_work, t*HZ);
5498 static inline __be32 nfs4_check_fh(struct svc_fh *fhp, struct nfs4_stid *stp)
5500 if (!fh_match(&fhp->fh_handle, &stp->sc_file->fi_fhandle))
5501 return nfserr_bad_stateid;
5506 access_permit_read(struct nfs4_ol_stateid *stp)
5508 return test_access(NFS4_SHARE_ACCESS_READ, stp) ||
5509 test_access(NFS4_SHARE_ACCESS_BOTH, stp) ||
5510 test_access(NFS4_SHARE_ACCESS_WRITE, stp);
5514 access_permit_write(struct nfs4_ol_stateid *stp)
5516 return test_access(NFS4_SHARE_ACCESS_WRITE, stp) ||
5517 test_access(NFS4_SHARE_ACCESS_BOTH, stp);
5521 __be32 nfs4_check_openmode(struct nfs4_ol_stateid *stp, int flags)
5523 __be32 status = nfserr_openmode;
5525 /* For lock stateid's, we test the parent open, not the lock: */
5526 if (stp->st_openstp)
5527 stp = stp->st_openstp;
5528 if ((flags & WR_STATE) && !access_permit_write(stp))
5530 if ((flags & RD_STATE) && !access_permit_read(stp))
5537 static inline __be32
5538 check_special_stateids(struct net *net, svc_fh *current_fh, stateid_t *stateid, int flags)
5540 if (ONE_STATEID(stateid) && (flags & RD_STATE))
5542 else if (opens_in_grace(net)) {
5543 /* Answer in remaining cases depends on existence of
5544 * conflicting state; so we must wait out the grace period. */
5545 return nfserr_grace;
5546 } else if (flags & WR_STATE)
5547 return nfs4_share_conflict(current_fh,
5548 NFS4_SHARE_DENY_WRITE);
5549 else /* (flags & RD_STATE) && ZERO_STATEID(stateid) */
5550 return nfs4_share_conflict(current_fh,
5551 NFS4_SHARE_DENY_READ);
5555 * Allow READ/WRITE during grace period on recovered state only for files
5556 * that are not able to provide mandatory locking.
5559 grace_disallows_io(struct net *net, struct inode *inode)
5561 return opens_in_grace(net) && mandatory_lock(inode);
5564 static __be32 check_stateid_generation(stateid_t *in, stateid_t *ref, bool has_session)
5567 * When sessions are used the stateid generation number is ignored
5570 if (has_session && in->si_generation == 0)
5573 if (in->si_generation == ref->si_generation)
5576 /* If the client sends us a stateid from the future, it's buggy: */
5577 if (nfsd4_stateid_generation_after(in, ref))
5578 return nfserr_bad_stateid;
5580 * However, we could see a stateid from the past, even from a
5581 * non-buggy client. For example, if the client sends a lock
5582 * while some IO is outstanding, the lock may bump si_generation
5583 * while the IO is still in flight. The client could avoid that
5584 * situation by waiting for responses on all the IO requests,
5585 * but better performance may result in retrying IO that
5586 * receives an old_stateid error if requests are rarely
5587 * reordered in flight:
5589 return nfserr_old_stateid;
5592 static __be32 nfsd4_stid_check_stateid_generation(stateid_t *in, struct nfs4_stid *s, bool has_session)
5596 spin_lock(&s->sc_lock);
5597 ret = nfsd4_verify_open_stid(s);
5599 ret = check_stateid_generation(in, &s->sc_stateid, has_session);
5600 spin_unlock(&s->sc_lock);
5604 static __be32 nfsd4_check_openowner_confirmed(struct nfs4_ol_stateid *ols)
5606 if (ols->st_stateowner->so_is_open_owner &&
5607 !(openowner(ols->st_stateowner)->oo_flags & NFS4_OO_CONFIRMED))
5608 return nfserr_bad_stateid;
5612 static __be32 nfsd4_validate_stateid(struct nfs4_client *cl, stateid_t *stateid)
5614 struct nfs4_stid *s;
5615 __be32 status = nfserr_bad_stateid;
5617 if (ZERO_STATEID(stateid) || ONE_STATEID(stateid) ||
5618 CLOSE_STATEID(stateid))
5620 if (!same_clid(&stateid->si_opaque.so_clid, &cl->cl_clientid))
5622 spin_lock(&cl->cl_lock);
5623 s = find_stateid_locked(cl, stateid);
5626 status = nfsd4_stid_check_stateid_generation(stateid, s, 1);
5629 switch (s->sc_type) {
5630 case NFS4_DELEG_STID:
5633 case NFS4_REVOKED_DELEG_STID:
5634 status = nfserr_deleg_revoked;
5636 case NFS4_OPEN_STID:
5637 case NFS4_LOCK_STID:
5638 status = nfsd4_check_openowner_confirmed(openlockstateid(s));
5641 printk("unknown stateid type %x\n", s->sc_type);
5643 case NFS4_CLOSED_STID:
5644 case NFS4_CLOSED_DELEG_STID:
5645 status = nfserr_bad_stateid;
5648 spin_unlock(&cl->cl_lock);
5653 nfsd4_lookup_stateid(struct nfsd4_compound_state *cstate,
5654 stateid_t *stateid, unsigned char typemask,
5655 struct nfs4_stid **s, struct nfsd_net *nn)
5658 bool return_revoked = false;
5661 * only return revoked delegations if explicitly asked.
5662 * otherwise we report revoked or bad_stateid status.
5664 if (typemask & NFS4_REVOKED_DELEG_STID)
5665 return_revoked = true;
5666 else if (typemask & NFS4_DELEG_STID)
5667 typemask |= NFS4_REVOKED_DELEG_STID;
5669 if (ZERO_STATEID(stateid) || ONE_STATEID(stateid) ||
5670 CLOSE_STATEID(stateid))
5671 return nfserr_bad_stateid;
5672 status = set_client(&stateid->si_opaque.so_clid, cstate, nn);
5673 if (status == nfserr_stale_clientid) {
5674 if (cstate->session)
5675 return nfserr_bad_stateid;
5676 return nfserr_stale_stateid;
5680 *s = find_stateid_by_type(cstate->clp, stateid, typemask);
5682 return nfserr_bad_stateid;
5683 if (((*s)->sc_type == NFS4_REVOKED_DELEG_STID) && !return_revoked) {
5685 if (cstate->minorversion)
5686 return nfserr_deleg_revoked;
5687 return nfserr_bad_stateid;
5692 static struct nfsd_file *
5693 nfs4_find_file(struct nfs4_stid *s, int flags)
5698 switch (s->sc_type) {
5699 case NFS4_DELEG_STID:
5700 if (WARN_ON_ONCE(!s->sc_file->fi_deleg_file))
5702 return nfsd_file_get(s->sc_file->fi_deleg_file);
5703 case NFS4_OPEN_STID:
5704 case NFS4_LOCK_STID:
5705 if (flags & RD_STATE)
5706 return find_readable_file(s->sc_file);
5708 return find_writeable_file(s->sc_file);
5715 nfs4_check_olstateid(struct nfs4_ol_stateid *ols, int flags)
5719 status = nfsd4_check_openowner_confirmed(ols);
5722 return nfs4_check_openmode(ols, flags);
5726 nfs4_check_file(struct svc_rqst *rqstp, struct svc_fh *fhp, struct nfs4_stid *s,
5727 struct nfsd_file **nfp, int flags)
5729 int acc = (flags & RD_STATE) ? NFSD_MAY_READ : NFSD_MAY_WRITE;
5730 struct nfsd_file *nf;
5733 nf = nfs4_find_file(s, flags);
5735 status = nfsd_permission(rqstp, fhp->fh_export, fhp->fh_dentry,
5736 acc | NFSD_MAY_OWNER_OVERRIDE);
5742 status = nfsd_file_acquire(rqstp, fhp, acc, &nf);
5751 _free_cpntf_state_locked(struct nfsd_net *nn, struct nfs4_cpntf_state *cps)
5753 WARN_ON_ONCE(cps->cp_stateid.sc_type != NFS4_COPYNOTIFY_STID);
5754 if (!refcount_dec_and_test(&cps->cp_stateid.sc_count))
5756 list_del(&cps->cp_list);
5757 idr_remove(&nn->s2s_cp_stateids,
5758 cps->cp_stateid.stid.si_opaque.so_id);
5762 * A READ from an inter server to server COPY will have a
5763 * copy stateid. Look up the copy notify stateid from the
5764 * idr structure and take a reference on it.
5766 __be32 manage_cpntf_state(struct nfsd_net *nn, stateid_t *st,
5767 struct nfs4_client *clp,
5768 struct nfs4_cpntf_state **cps)
5770 copy_stateid_t *cps_t;
5771 struct nfs4_cpntf_state *state = NULL;
5773 if (st->si_opaque.so_clid.cl_id != nn->s2s_cp_cl_id)
5774 return nfserr_bad_stateid;
5775 spin_lock(&nn->s2s_cp_lock);
5776 cps_t = idr_find(&nn->s2s_cp_stateids, st->si_opaque.so_id);
5778 state = container_of(cps_t, struct nfs4_cpntf_state,
5780 if (state->cp_stateid.sc_type != NFS4_COPYNOTIFY_STID) {
5785 refcount_inc(&state->cp_stateid.sc_count);
5787 _free_cpntf_state_locked(nn, state);
5790 spin_unlock(&nn->s2s_cp_lock);
5792 return nfserr_bad_stateid;
5798 static __be32 find_cpntf_state(struct nfsd_net *nn, stateid_t *st,
5799 struct nfs4_stid **stid)
5802 struct nfs4_cpntf_state *cps = NULL;
5803 struct nfs4_client *found;
5805 status = manage_cpntf_state(nn, st, NULL, &cps);
5809 cps->cpntf_time = ktime_get_boottime_seconds();
5811 status = nfserr_expired;
5812 found = lookup_clientid(&cps->cp_p_clid, true, nn);
5816 *stid = find_stateid_by_type(found, &cps->cp_p_stateid,
5817 NFS4_DELEG_STID|NFS4_OPEN_STID|NFS4_LOCK_STID);
5821 status = nfserr_bad_stateid;
5823 put_client_renew(found);
5825 nfs4_put_cpntf_state(nn, cps);
5829 void nfs4_put_cpntf_state(struct nfsd_net *nn, struct nfs4_cpntf_state *cps)
5831 spin_lock(&nn->s2s_cp_lock);
5832 _free_cpntf_state_locked(nn, cps);
5833 spin_unlock(&nn->s2s_cp_lock);
5837 * Checks for stateid operations
5840 nfs4_preprocess_stateid_op(struct svc_rqst *rqstp,
5841 struct nfsd4_compound_state *cstate, struct svc_fh *fhp,
5842 stateid_t *stateid, int flags, struct nfsd_file **nfp,
5843 struct nfs4_stid **cstid)
5845 struct inode *ino = d_inode(fhp->fh_dentry);
5846 struct net *net = SVC_NET(rqstp);
5847 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
5848 struct nfs4_stid *s = NULL;
5854 if (grace_disallows_io(net, ino))
5855 return nfserr_grace;
5857 if (ZERO_STATEID(stateid) || ONE_STATEID(stateid)) {
5858 status = check_special_stateids(net, fhp, stateid, flags);
5862 status = nfsd4_lookup_stateid(cstate, stateid,
5863 NFS4_DELEG_STID|NFS4_OPEN_STID|NFS4_LOCK_STID,
5865 if (status == nfserr_bad_stateid)
5866 status = find_cpntf_state(nn, stateid, &s);
5869 status = nfsd4_stid_check_stateid_generation(stateid, s,
5870 nfsd4_has_session(cstate));
5874 switch (s->sc_type) {
5875 case NFS4_DELEG_STID:
5876 status = nfs4_check_delegmode(delegstateid(s), flags);
5878 case NFS4_OPEN_STID:
5879 case NFS4_LOCK_STID:
5880 status = nfs4_check_olstateid(openlockstateid(s), flags);
5883 status = nfserr_bad_stateid;
5888 status = nfs4_check_fh(fhp, s);
5891 if (status == nfs_ok && nfp)
5892 status = nfs4_check_file(rqstp, fhp, s, nfp, flags);
5895 if (!status && cstid)
5904 * Test if the stateid is valid
5907 nfsd4_test_stateid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
5908 union nfsd4_op_u *u)
5910 struct nfsd4_test_stateid *test_stateid = &u->test_stateid;
5911 struct nfsd4_test_stateid_id *stateid;
5912 struct nfs4_client *cl = cstate->clp;
5914 list_for_each_entry(stateid, &test_stateid->ts_stateid_list, ts_id_list)
5915 stateid->ts_id_status =
5916 nfsd4_validate_stateid(cl, &stateid->ts_id_stateid);
5922 nfsd4_free_lock_stateid(stateid_t *stateid, struct nfs4_stid *s)
5924 struct nfs4_ol_stateid *stp = openlockstateid(s);
5927 ret = nfsd4_lock_ol_stateid(stp);
5931 ret = check_stateid_generation(stateid, &s->sc_stateid, 1);
5935 ret = nfserr_locks_held;
5936 if (check_for_locks(stp->st_stid.sc_file,
5937 lockowner(stp->st_stateowner)))
5940 release_lock_stateid(stp);
5944 mutex_unlock(&stp->st_mutex);
5951 nfsd4_free_stateid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
5952 union nfsd4_op_u *u)
5954 struct nfsd4_free_stateid *free_stateid = &u->free_stateid;
5955 stateid_t *stateid = &free_stateid->fr_stateid;
5956 struct nfs4_stid *s;
5957 struct nfs4_delegation *dp;
5958 struct nfs4_client *cl = cstate->clp;
5959 __be32 ret = nfserr_bad_stateid;
5961 spin_lock(&cl->cl_lock);
5962 s = find_stateid_locked(cl, stateid);
5965 spin_lock(&s->sc_lock);
5966 switch (s->sc_type) {
5967 case NFS4_DELEG_STID:
5968 ret = nfserr_locks_held;
5970 case NFS4_OPEN_STID:
5971 ret = check_stateid_generation(stateid, &s->sc_stateid, 1);
5974 ret = nfserr_locks_held;
5976 case NFS4_LOCK_STID:
5977 spin_unlock(&s->sc_lock);
5978 refcount_inc(&s->sc_count);
5979 spin_unlock(&cl->cl_lock);
5980 ret = nfsd4_free_lock_stateid(stateid, s);
5982 case NFS4_REVOKED_DELEG_STID:
5983 spin_unlock(&s->sc_lock);
5984 dp = delegstateid(s);
5985 list_del_init(&dp->dl_recall_lru);
5986 spin_unlock(&cl->cl_lock);
5990 /* Default falls through and returns nfserr_bad_stateid */
5992 spin_unlock(&s->sc_lock);
5994 spin_unlock(&cl->cl_lock);
6002 return (type == NFS4_READW_LT || type == NFS4_READ_LT) ?
6003 RD_STATE : WR_STATE;
6006 static __be32 nfs4_seqid_op_checks(struct nfsd4_compound_state *cstate, stateid_t *stateid, u32 seqid, struct nfs4_ol_stateid *stp)
6008 struct svc_fh *current_fh = &cstate->current_fh;
6009 struct nfs4_stateowner *sop = stp->st_stateowner;
6012 status = nfsd4_check_seqid(cstate, sop, seqid);
6015 status = nfsd4_lock_ol_stateid(stp);
6016 if (status != nfs_ok)
6018 status = check_stateid_generation(stateid, &stp->st_stid.sc_stateid, nfsd4_has_session(cstate));
6019 if (status == nfs_ok)
6020 status = nfs4_check_fh(current_fh, &stp->st_stid);
6021 if (status != nfs_ok)
6022 mutex_unlock(&stp->st_mutex);
6027 * Checks for sequence id mutating operations.
6030 nfs4_preprocess_seqid_op(struct nfsd4_compound_state *cstate, u32 seqid,
6031 stateid_t *stateid, char typemask,
6032 struct nfs4_ol_stateid **stpp,
6033 struct nfsd_net *nn)
6036 struct nfs4_stid *s;
6037 struct nfs4_ol_stateid *stp = NULL;
6039 trace_nfsd_preprocess(seqid, stateid);
6042 status = nfsd4_lookup_stateid(cstate, stateid, typemask, &s, nn);
6045 stp = openlockstateid(s);
6046 nfsd4_cstate_assign_replay(cstate, stp->st_stateowner);
6048 status = nfs4_seqid_op_checks(cstate, stateid, seqid, stp);
6052 nfs4_put_stid(&stp->st_stid);
6056 static __be32 nfs4_preprocess_confirmed_seqid_op(struct nfsd4_compound_state *cstate, u32 seqid,
6057 stateid_t *stateid, struct nfs4_ol_stateid **stpp, struct nfsd_net *nn)
6060 struct nfs4_openowner *oo;
6061 struct nfs4_ol_stateid *stp;
6063 status = nfs4_preprocess_seqid_op(cstate, seqid, stateid,
6064 NFS4_OPEN_STID, &stp, nn);
6067 oo = openowner(stp->st_stateowner);
6068 if (!(oo->oo_flags & NFS4_OO_CONFIRMED)) {
6069 mutex_unlock(&stp->st_mutex);
6070 nfs4_put_stid(&stp->st_stid);
6071 return nfserr_bad_stateid;
6078 nfsd4_open_confirm(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
6079 union nfsd4_op_u *u)
6081 struct nfsd4_open_confirm *oc = &u->open_confirm;
6083 struct nfs4_openowner *oo;
6084 struct nfs4_ol_stateid *stp;
6085 struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
6087 dprintk("NFSD: nfsd4_open_confirm on file %pd\n",
6088 cstate->current_fh.fh_dentry);
6090 status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0);
6094 status = nfs4_preprocess_seqid_op(cstate,
6095 oc->oc_seqid, &oc->oc_req_stateid,
6096 NFS4_OPEN_STID, &stp, nn);
6099 oo = openowner(stp->st_stateowner);
6100 status = nfserr_bad_stateid;
6101 if (oo->oo_flags & NFS4_OO_CONFIRMED) {
6102 mutex_unlock(&stp->st_mutex);
6105 oo->oo_flags |= NFS4_OO_CONFIRMED;
6106 nfs4_inc_and_copy_stateid(&oc->oc_resp_stateid, &stp->st_stid);
6107 mutex_unlock(&stp->st_mutex);
6108 trace_nfsd_open_confirm(oc->oc_seqid, &stp->st_stid.sc_stateid);
6109 nfsd4_client_record_create(oo->oo_owner.so_client);
6112 nfs4_put_stid(&stp->st_stid);
6114 nfsd4_bump_seqid(cstate, status);
6118 static inline void nfs4_stateid_downgrade_bit(struct nfs4_ol_stateid *stp, u32 access)
6120 if (!test_access(access, stp))
6122 nfs4_file_put_access(stp->st_stid.sc_file, access);
6123 clear_access(access, stp);
6126 static inline void nfs4_stateid_downgrade(struct nfs4_ol_stateid *stp, u32 to_access)
6128 switch (to_access) {
6129 case NFS4_SHARE_ACCESS_READ:
6130 nfs4_stateid_downgrade_bit(stp, NFS4_SHARE_ACCESS_WRITE);
6131 nfs4_stateid_downgrade_bit(stp, NFS4_SHARE_ACCESS_BOTH);
6133 case NFS4_SHARE_ACCESS_WRITE:
6134 nfs4_stateid_downgrade_bit(stp, NFS4_SHARE_ACCESS_READ);
6135 nfs4_stateid_downgrade_bit(stp, NFS4_SHARE_ACCESS_BOTH);
6137 case NFS4_SHARE_ACCESS_BOTH:
6145 nfsd4_open_downgrade(struct svc_rqst *rqstp,
6146 struct nfsd4_compound_state *cstate, union nfsd4_op_u *u)
6148 struct nfsd4_open_downgrade *od = &u->open_downgrade;
6150 struct nfs4_ol_stateid *stp;
6151 struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
6153 dprintk("NFSD: nfsd4_open_downgrade on file %pd\n",
6154 cstate->current_fh.fh_dentry);
6156 /* We don't yet support WANT bits: */
6157 if (od->od_deleg_want)
6158 dprintk("NFSD: %s: od_deleg_want=0x%x ignored\n", __func__,
6161 status = nfs4_preprocess_confirmed_seqid_op(cstate, od->od_seqid,
6162 &od->od_stateid, &stp, nn);
6165 status = nfserr_inval;
6166 if (!test_access(od->od_share_access, stp)) {
6167 dprintk("NFSD: access not a subset of current bitmap: 0x%hhx, input access=%08x\n",
6168 stp->st_access_bmap, od->od_share_access);
6171 if (!test_deny(od->od_share_deny, stp)) {
6172 dprintk("NFSD: deny not a subset of current bitmap: 0x%hhx, input deny=%08x\n",
6173 stp->st_deny_bmap, od->od_share_deny);
6176 nfs4_stateid_downgrade(stp, od->od_share_access);
6177 reset_union_bmap_deny(od->od_share_deny, stp);
6178 nfs4_inc_and_copy_stateid(&od->od_stateid, &stp->st_stid);
6181 mutex_unlock(&stp->st_mutex);
6182 nfs4_put_stid(&stp->st_stid);
6184 nfsd4_bump_seqid(cstate, status);
6188 static void nfsd4_close_open_stateid(struct nfs4_ol_stateid *s)
6190 struct nfs4_client *clp = s->st_stid.sc_client;
6192 LIST_HEAD(reaplist);
6194 spin_lock(&clp->cl_lock);
6195 unhashed = unhash_open_stateid(s, &reaplist);
6197 if (clp->cl_minorversion) {
6199 put_ol_stateid_locked(s, &reaplist);
6200 spin_unlock(&clp->cl_lock);
6201 free_ol_stateid_reaplist(&reaplist);
6203 spin_unlock(&clp->cl_lock);
6204 free_ol_stateid_reaplist(&reaplist);
6206 move_to_close_lru(s, clp->net);
6211 * nfs4_unlock_state() called after encode
6214 nfsd4_close(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
6215 union nfsd4_op_u *u)
6217 struct nfsd4_close *close = &u->close;
6219 struct nfs4_ol_stateid *stp;
6220 struct net *net = SVC_NET(rqstp);
6221 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
6223 dprintk("NFSD: nfsd4_close on file %pd\n",
6224 cstate->current_fh.fh_dentry);
6226 status = nfs4_preprocess_seqid_op(cstate, close->cl_seqid,
6228 NFS4_OPEN_STID|NFS4_CLOSED_STID,
6230 nfsd4_bump_seqid(cstate, status);
6234 stp->st_stid.sc_type = NFS4_CLOSED_STID;
6237 * Technically we don't _really_ have to increment or copy it, since
6238 * it should just be gone after this operation and we clobber the
6239 * copied value below, but we continue to do so here just to ensure
6240 * that racing ops see that there was a state change.
6242 nfs4_inc_and_copy_stateid(&close->cl_stateid, &stp->st_stid);
6244 nfsd4_close_open_stateid(stp);
6245 mutex_unlock(&stp->st_mutex);
6247 /* v4.1+ suggests that we send a special stateid in here, since the
6248 * clients should just ignore this anyway. Since this is not useful
6249 * for v4.0 clients either, we set it to the special close_stateid
6252 * See RFC5661 section 18.2.4, and RFC7530 section 16.2.5
6254 memcpy(&close->cl_stateid, &close_stateid, sizeof(close->cl_stateid));
6256 /* put reference from nfs4_preprocess_seqid_op */
6257 nfs4_put_stid(&stp->st_stid);
6263 nfsd4_delegreturn(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
6264 union nfsd4_op_u *u)
6266 struct nfsd4_delegreturn *dr = &u->delegreturn;
6267 struct nfs4_delegation *dp;
6268 stateid_t *stateid = &dr->dr_stateid;
6269 struct nfs4_stid *s;
6271 struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
6273 if ((status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0)))
6276 status = nfsd4_lookup_stateid(cstate, stateid, NFS4_DELEG_STID, &s, nn);
6279 dp = delegstateid(s);
6280 status = nfsd4_stid_check_stateid_generation(stateid, &dp->dl_stid, nfsd4_has_session(cstate));
6284 destroy_delegation(dp);
6286 nfs4_put_stid(&dp->dl_stid);
6292 end_offset(u64 start, u64 len)
6297 return end >= start ? end: NFS4_MAX_UINT64;
6300 /* last octet in a range */
6302 last_byte_offset(u64 start, u64 len)
6308 return end > start ? end - 1: NFS4_MAX_UINT64;
6312 * TODO: Linux file offsets are _signed_ 64-bit quantities, which means that
6313 * we can't properly handle lock requests that go beyond the (2^63 - 1)-th
6314 * byte, because of sign extension problems. Since NFSv4 calls for 64-bit
6315 * locking, this prevents us from being completely protocol-compliant. The
6316 * real solution to this problem is to start using unsigned file offsets in
6317 * the VFS, but this is a very deep change!
6320 nfs4_transform_lock_offset(struct file_lock *lock)
6322 if (lock->fl_start < 0)
6323 lock->fl_start = OFFSET_MAX;
6324 if (lock->fl_end < 0)
6325 lock->fl_end = OFFSET_MAX;
6329 nfsd4_fl_get_owner(fl_owner_t owner)
6331 struct nfs4_lockowner *lo = (struct nfs4_lockowner *)owner;
6333 nfs4_get_stateowner(&lo->lo_owner);
6338 nfsd4_fl_put_owner(fl_owner_t owner)
6340 struct nfs4_lockowner *lo = (struct nfs4_lockowner *)owner;
6343 nfs4_put_stateowner(&lo->lo_owner);
6347 nfsd4_lm_notify(struct file_lock *fl)
6349 struct nfs4_lockowner *lo = (struct nfs4_lockowner *)fl->fl_owner;
6350 struct net *net = lo->lo_owner.so_client->net;
6351 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
6352 struct nfsd4_blocked_lock *nbl = container_of(fl,
6353 struct nfsd4_blocked_lock, nbl_lock);
6356 /* An empty list means that something else is going to be using it */
6357 spin_lock(&nn->blocked_locks_lock);
6358 if (!list_empty(&nbl->nbl_list)) {
6359 list_del_init(&nbl->nbl_list);
6360 list_del_init(&nbl->nbl_lru);
6363 spin_unlock(&nn->blocked_locks_lock);
6366 nfsd4_run_cb(&nbl->nbl_cb);
6369 static const struct lock_manager_operations nfsd_posix_mng_ops = {
6370 .lm_notify = nfsd4_lm_notify,
6371 .lm_get_owner = nfsd4_fl_get_owner,
6372 .lm_put_owner = nfsd4_fl_put_owner,
6376 nfs4_set_lock_denied(struct file_lock *fl, struct nfsd4_lock_denied *deny)
6378 struct nfs4_lockowner *lo;
6380 if (fl->fl_lmops == &nfsd_posix_mng_ops) {
6381 lo = (struct nfs4_lockowner *) fl->fl_owner;
6382 xdr_netobj_dup(&deny->ld_owner, &lo->lo_owner.so_owner,
6384 if (!deny->ld_owner.data)
6385 /* We just don't care that much */
6387 deny->ld_clientid = lo->lo_owner.so_client->cl_clientid;
6390 deny->ld_owner.len = 0;
6391 deny->ld_owner.data = NULL;
6392 deny->ld_clientid.cl_boot = 0;
6393 deny->ld_clientid.cl_id = 0;
6395 deny->ld_start = fl->fl_start;
6396 deny->ld_length = NFS4_MAX_UINT64;
6397 if (fl->fl_end != NFS4_MAX_UINT64)
6398 deny->ld_length = fl->fl_end - fl->fl_start + 1;
6399 deny->ld_type = NFS4_READ_LT;
6400 if (fl->fl_type != F_RDLCK)
6401 deny->ld_type = NFS4_WRITE_LT;
6404 static struct nfs4_lockowner *
6405 find_lockowner_str_locked(struct nfs4_client *clp, struct xdr_netobj *owner)
6407 unsigned int strhashval = ownerstr_hashval(owner);
6408 struct nfs4_stateowner *so;
6410 lockdep_assert_held(&clp->cl_lock);
6412 list_for_each_entry(so, &clp->cl_ownerstr_hashtbl[strhashval],
6414 if (so->so_is_open_owner)
6416 if (same_owner_str(so, owner))
6417 return lockowner(nfs4_get_stateowner(so));
6422 static struct nfs4_lockowner *
6423 find_lockowner_str(struct nfs4_client *clp, struct xdr_netobj *owner)
6425 struct nfs4_lockowner *lo;
6427 spin_lock(&clp->cl_lock);
6428 lo = find_lockowner_str_locked(clp, owner);
6429 spin_unlock(&clp->cl_lock);
6433 static void nfs4_unhash_lockowner(struct nfs4_stateowner *sop)
6435 unhash_lockowner_locked(lockowner(sop));
6438 static void nfs4_free_lockowner(struct nfs4_stateowner *sop)
6440 struct nfs4_lockowner *lo = lockowner(sop);
6442 kmem_cache_free(lockowner_slab, lo);
6445 static const struct nfs4_stateowner_operations lockowner_ops = {
6446 .so_unhash = nfs4_unhash_lockowner,
6447 .so_free = nfs4_free_lockowner,
6451 * Alloc a lock owner structure.
6452 * Called in nfsd4_lock - therefore, OPEN and OPEN_CONFIRM (if needed) has
6455 * strhashval = ownerstr_hashval
6457 static struct nfs4_lockowner *
6458 alloc_init_lock_stateowner(unsigned int strhashval, struct nfs4_client *clp,
6459 struct nfs4_ol_stateid *open_stp,
6460 struct nfsd4_lock *lock)
6462 struct nfs4_lockowner *lo, *ret;
6464 lo = alloc_stateowner(lockowner_slab, &lock->lk_new_owner, clp);
6467 INIT_LIST_HEAD(&lo->lo_blocked);
6468 INIT_LIST_HEAD(&lo->lo_owner.so_stateids);
6469 lo->lo_owner.so_is_open_owner = 0;
6470 lo->lo_owner.so_seqid = lock->lk_new_lock_seqid;
6471 lo->lo_owner.so_ops = &lockowner_ops;
6472 spin_lock(&clp->cl_lock);
6473 ret = find_lockowner_str_locked(clp, &lock->lk_new_owner);
6475 list_add(&lo->lo_owner.so_strhash,
6476 &clp->cl_ownerstr_hashtbl[strhashval]);
6479 nfs4_free_stateowner(&lo->lo_owner);
6481 spin_unlock(&clp->cl_lock);
6485 static struct nfs4_ol_stateid *
6486 find_lock_stateid(const struct nfs4_lockowner *lo,
6487 const struct nfs4_ol_stateid *ost)
6489 struct nfs4_ol_stateid *lst;
6491 lockdep_assert_held(&ost->st_stid.sc_client->cl_lock);
6493 /* If ost is not hashed, ost->st_locks will not be valid */
6494 if (!nfs4_ol_stateid_unhashed(ost))
6495 list_for_each_entry(lst, &ost->st_locks, st_locks) {
6496 if (lst->st_stateowner == &lo->lo_owner) {
6497 refcount_inc(&lst->st_stid.sc_count);
6504 static struct nfs4_ol_stateid *
6505 init_lock_stateid(struct nfs4_ol_stateid *stp, struct nfs4_lockowner *lo,
6506 struct nfs4_file *fp, struct inode *inode,
6507 struct nfs4_ol_stateid *open_stp)
6509 struct nfs4_client *clp = lo->lo_owner.so_client;
6510 struct nfs4_ol_stateid *retstp;
6512 mutex_init(&stp->st_mutex);
6513 mutex_lock_nested(&stp->st_mutex, OPEN_STATEID_MUTEX);
6515 spin_lock(&clp->cl_lock);
6516 if (nfs4_ol_stateid_unhashed(open_stp))
6518 retstp = find_lock_stateid(lo, open_stp);
6521 refcount_inc(&stp->st_stid.sc_count);
6522 stp->st_stid.sc_type = NFS4_LOCK_STID;
6523 stp->st_stateowner = nfs4_get_stateowner(&lo->lo_owner);
6525 stp->st_stid.sc_file = fp;
6526 stp->st_access_bmap = 0;
6527 stp->st_deny_bmap = open_stp->st_deny_bmap;
6528 stp->st_openstp = open_stp;
6529 spin_lock(&fp->fi_lock);
6530 list_add(&stp->st_locks, &open_stp->st_locks);
6531 list_add(&stp->st_perstateowner, &lo->lo_owner.so_stateids);
6532 list_add(&stp->st_perfile, &fp->fi_stateids);
6533 spin_unlock(&fp->fi_lock);
6534 spin_unlock(&clp->cl_lock);
6537 spin_unlock(&clp->cl_lock);
6538 if (nfsd4_lock_ol_stateid(retstp) != nfs_ok) {
6539 nfs4_put_stid(&retstp->st_stid);
6542 /* To keep mutex tracking happy */
6543 mutex_unlock(&stp->st_mutex);
6546 spin_unlock(&clp->cl_lock);
6547 mutex_unlock(&stp->st_mutex);
6551 static struct nfs4_ol_stateid *
6552 find_or_create_lock_stateid(struct nfs4_lockowner *lo, struct nfs4_file *fi,
6553 struct inode *inode, struct nfs4_ol_stateid *ost,
6556 struct nfs4_stid *ns = NULL;
6557 struct nfs4_ol_stateid *lst;
6558 struct nfs4_openowner *oo = openowner(ost->st_stateowner);
6559 struct nfs4_client *clp = oo->oo_owner.so_client;
6562 spin_lock(&clp->cl_lock);
6563 lst = find_lock_stateid(lo, ost);
6564 spin_unlock(&clp->cl_lock);
6566 if (nfsd4_lock_ol_stateid(lst) == nfs_ok)
6568 nfs4_put_stid(&lst->st_stid);
6570 ns = nfs4_alloc_stid(clp, stateid_slab, nfs4_free_lock_stateid);
6574 lst = init_lock_stateid(openlockstateid(ns), lo, fi, inode, ost);
6575 if (lst == openlockstateid(ns))
6584 check_lock_length(u64 offset, u64 length)
6586 return ((length == 0) || ((length != NFS4_MAX_UINT64) &&
6587 (length > ~offset)));
6590 static void get_lock_access(struct nfs4_ol_stateid *lock_stp, u32 access)
6592 struct nfs4_file *fp = lock_stp->st_stid.sc_file;
6594 lockdep_assert_held(&fp->fi_lock);
6596 if (test_access(access, lock_stp))
6598 __nfs4_file_get_access(fp, access);
6599 set_access(access, lock_stp);
6603 lookup_or_create_lock_state(struct nfsd4_compound_state *cstate,
6604 struct nfs4_ol_stateid *ost,
6605 struct nfsd4_lock *lock,
6606 struct nfs4_ol_stateid **plst, bool *new)
6609 struct nfs4_file *fi = ost->st_stid.sc_file;
6610 struct nfs4_openowner *oo = openowner(ost->st_stateowner);
6611 struct nfs4_client *cl = oo->oo_owner.so_client;
6612 struct inode *inode = d_inode(cstate->current_fh.fh_dentry);
6613 struct nfs4_lockowner *lo;
6614 struct nfs4_ol_stateid *lst;
6615 unsigned int strhashval;
6617 lo = find_lockowner_str(cl, &lock->lk_new_owner);
6619 strhashval = ownerstr_hashval(&lock->lk_new_owner);
6620 lo = alloc_init_lock_stateowner(strhashval, cl, ost, lock);
6622 return nfserr_jukebox;
6624 /* with an existing lockowner, seqids must be the same */
6625 status = nfserr_bad_seqid;
6626 if (!cstate->minorversion &&
6627 lock->lk_new_lock_seqid != lo->lo_owner.so_seqid)
6631 lst = find_or_create_lock_stateid(lo, fi, inode, ost, new);
6633 status = nfserr_jukebox;
6640 nfs4_put_stateowner(&lo->lo_owner);
6648 nfsd4_lock(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
6649 union nfsd4_op_u *u)
6651 struct nfsd4_lock *lock = &u->lock;
6652 struct nfs4_openowner *open_sop = NULL;
6653 struct nfs4_lockowner *lock_sop = NULL;
6654 struct nfs4_ol_stateid *lock_stp = NULL;
6655 struct nfs4_ol_stateid *open_stp = NULL;
6656 struct nfs4_file *fp;
6657 struct nfsd_file *nf = NULL;
6658 struct nfsd4_blocked_lock *nbl = NULL;
6659 struct file_lock *file_lock = NULL;
6660 struct file_lock *conflock = NULL;
6665 unsigned char fl_type;
6666 unsigned int fl_flags = FL_POSIX;
6667 struct net *net = SVC_NET(rqstp);
6668 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
6670 dprintk("NFSD: nfsd4_lock: start=%Ld length=%Ld\n",
6671 (long long) lock->lk_offset,
6672 (long long) lock->lk_length);
6674 if (check_lock_length(lock->lk_offset, lock->lk_length))
6675 return nfserr_inval;
6677 if ((status = fh_verify(rqstp, &cstate->current_fh,
6678 S_IFREG, NFSD_MAY_LOCK))) {
6679 dprintk("NFSD: nfsd4_lock: permission denied!\n");
6683 if (lock->lk_is_new) {
6684 if (nfsd4_has_session(cstate))
6685 /* See rfc 5661 18.10.3: given clientid is ignored: */
6686 memcpy(&lock->lk_new_clientid,
6687 &cstate->clp->cl_clientid,
6688 sizeof(clientid_t));
6690 /* validate and update open stateid and open seqid */
6691 status = nfs4_preprocess_confirmed_seqid_op(cstate,
6692 lock->lk_new_open_seqid,
6693 &lock->lk_new_open_stateid,
6697 mutex_unlock(&open_stp->st_mutex);
6698 open_sop = openowner(open_stp->st_stateowner);
6699 status = nfserr_bad_stateid;
6700 if (!same_clid(&open_sop->oo_owner.so_client->cl_clientid,
6701 &lock->lk_new_clientid))
6703 status = lookup_or_create_lock_state(cstate, open_stp, lock,
6706 status = nfs4_preprocess_seqid_op(cstate,
6707 lock->lk_old_lock_seqid,
6708 &lock->lk_old_lock_stateid,
6709 NFS4_LOCK_STID, &lock_stp, nn);
6713 lock_sop = lockowner(lock_stp->st_stateowner);
6715 lkflg = setlkflg(lock->lk_type);
6716 status = nfs4_check_openmode(lock_stp, lkflg);
6720 status = nfserr_grace;
6721 if (locks_in_grace(net) && !lock->lk_reclaim)
6723 status = nfserr_no_grace;
6724 if (!locks_in_grace(net) && lock->lk_reclaim)
6727 fp = lock_stp->st_stid.sc_file;
6728 switch (lock->lk_type) {
6730 if (nfsd4_has_session(cstate))
6731 fl_flags |= FL_SLEEP;
6734 spin_lock(&fp->fi_lock);
6735 nf = find_readable_file_locked(fp);
6737 get_lock_access(lock_stp, NFS4_SHARE_ACCESS_READ);
6738 spin_unlock(&fp->fi_lock);
6741 case NFS4_WRITEW_LT:
6742 if (nfsd4_has_session(cstate))
6743 fl_flags |= FL_SLEEP;
6746 spin_lock(&fp->fi_lock);
6747 nf = find_writeable_file_locked(fp);
6749 get_lock_access(lock_stp, NFS4_SHARE_ACCESS_WRITE);
6750 spin_unlock(&fp->fi_lock);
6754 status = nfserr_inval;
6759 status = nfserr_openmode;
6763 nbl = find_or_allocate_block(lock_sop, &fp->fi_fhandle, nn);
6765 dprintk("NFSD: %s: unable to allocate block!\n", __func__);
6766 status = nfserr_jukebox;
6770 file_lock = &nbl->nbl_lock;
6771 file_lock->fl_type = fl_type;
6772 file_lock->fl_owner = (fl_owner_t)lockowner(nfs4_get_stateowner(&lock_sop->lo_owner));
6773 file_lock->fl_pid = current->tgid;
6774 file_lock->fl_file = nf->nf_file;
6775 file_lock->fl_flags = fl_flags;
6776 file_lock->fl_lmops = &nfsd_posix_mng_ops;
6777 file_lock->fl_start = lock->lk_offset;
6778 file_lock->fl_end = last_byte_offset(lock->lk_offset, lock->lk_length);
6779 nfs4_transform_lock_offset(file_lock);
6781 conflock = locks_alloc_lock();
6783 dprintk("NFSD: %s: unable to allocate lock!\n", __func__);
6784 status = nfserr_jukebox;
6788 if (fl_flags & FL_SLEEP) {
6789 nbl->nbl_time = ktime_get_boottime_seconds();
6790 spin_lock(&nn->blocked_locks_lock);
6791 list_add_tail(&nbl->nbl_list, &lock_sop->lo_blocked);
6792 list_add_tail(&nbl->nbl_lru, &nn->blocked_locks_lru);
6793 spin_unlock(&nn->blocked_locks_lock);
6796 err = vfs_lock_file(nf->nf_file, F_SETLK, file_lock, conflock);
6798 case 0: /* success! */
6799 nfs4_inc_and_copy_stateid(&lock->lk_resp_stateid, &lock_stp->st_stid);
6801 if (lock->lk_reclaim)
6802 nn->somebody_reclaimed = true;
6804 case FILE_LOCK_DEFERRED:
6807 case -EAGAIN: /* conflock holds conflicting lock */
6808 status = nfserr_denied;
6809 dprintk("NFSD: nfsd4_lock: conflicting lock found!\n");
6810 nfs4_set_lock_denied(conflock, &lock->lk_denied);
6813 status = nfserr_deadlock;
6816 dprintk("NFSD: nfsd4_lock: vfs_lock_file() failed! status %d\n",err);
6817 status = nfserrno(err);
6822 /* dequeue it if we queued it before */
6823 if (fl_flags & FL_SLEEP) {
6824 spin_lock(&nn->blocked_locks_lock);
6825 list_del_init(&nbl->nbl_list);
6826 list_del_init(&nbl->nbl_lru);
6827 spin_unlock(&nn->blocked_locks_lock);
6829 free_blocked_lock(nbl);
6834 /* Bump seqid manually if the 4.0 replay owner is openowner */
6835 if (cstate->replay_owner &&
6836 cstate->replay_owner != &lock_sop->lo_owner &&
6837 seqid_mutating_err(ntohl(status)))
6838 lock_sop->lo_owner.so_seqid++;
6841 * If this is a new, never-before-used stateid, and we are
6842 * returning an error, then just go ahead and release it.
6845 release_lock_stateid(lock_stp);
6847 mutex_unlock(&lock_stp->st_mutex);
6849 nfs4_put_stid(&lock_stp->st_stid);
6852 nfs4_put_stid(&open_stp->st_stid);
6853 nfsd4_bump_seqid(cstate, status);
6855 locks_free_lock(conflock);
6860 * The NFSv4 spec allows a client to do a LOCKT without holding an OPEN,
6861 * so we do a temporary open here just to get an open file to pass to
6862 * vfs_test_lock. (Arguably perhaps test_lock should be done with an
6865 static __be32 nfsd_test_lock(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file_lock *lock)
6867 struct nfsd_file *nf;
6868 __be32 err = nfsd_file_acquire(rqstp, fhp, NFSD_MAY_READ, &nf);
6870 err = nfserrno(vfs_test_lock(nf->nf_file, lock));
6880 nfsd4_lockt(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
6881 union nfsd4_op_u *u)
6883 struct nfsd4_lockt *lockt = &u->lockt;
6884 struct file_lock *file_lock = NULL;
6885 struct nfs4_lockowner *lo = NULL;
6887 struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
6889 if (locks_in_grace(SVC_NET(rqstp)))
6890 return nfserr_grace;
6892 if (check_lock_length(lockt->lt_offset, lockt->lt_length))
6893 return nfserr_inval;
6895 if (!nfsd4_has_session(cstate)) {
6896 status = set_client(&lockt->lt_clientid, cstate, nn);
6901 if ((status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0)))
6904 file_lock = locks_alloc_lock();
6906 dprintk("NFSD: %s: unable to allocate lock!\n", __func__);
6907 status = nfserr_jukebox;
6911 switch (lockt->lt_type) {
6914 file_lock->fl_type = F_RDLCK;
6917 case NFS4_WRITEW_LT:
6918 file_lock->fl_type = F_WRLCK;
6921 dprintk("NFSD: nfs4_lockt: bad lock type!\n");
6922 status = nfserr_inval;
6926 lo = find_lockowner_str(cstate->clp, &lockt->lt_owner);
6928 file_lock->fl_owner = (fl_owner_t)lo;
6929 file_lock->fl_pid = current->tgid;
6930 file_lock->fl_flags = FL_POSIX;
6932 file_lock->fl_start = lockt->lt_offset;
6933 file_lock->fl_end = last_byte_offset(lockt->lt_offset, lockt->lt_length);
6935 nfs4_transform_lock_offset(file_lock);
6937 status = nfsd_test_lock(rqstp, &cstate->current_fh, file_lock);
6941 if (file_lock->fl_type != F_UNLCK) {
6942 status = nfserr_denied;
6943 nfs4_set_lock_denied(file_lock, &lockt->lt_denied);
6947 nfs4_put_stateowner(&lo->lo_owner);
6949 locks_free_lock(file_lock);
6954 nfsd4_locku(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
6955 union nfsd4_op_u *u)
6957 struct nfsd4_locku *locku = &u->locku;
6958 struct nfs4_ol_stateid *stp;
6959 struct nfsd_file *nf = NULL;
6960 struct file_lock *file_lock = NULL;
6963 struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
6965 dprintk("NFSD: nfsd4_locku: start=%Ld length=%Ld\n",
6966 (long long) locku->lu_offset,
6967 (long long) locku->lu_length);
6969 if (check_lock_length(locku->lu_offset, locku->lu_length))
6970 return nfserr_inval;
6972 status = nfs4_preprocess_seqid_op(cstate, locku->lu_seqid,
6973 &locku->lu_stateid, NFS4_LOCK_STID,
6977 nf = find_any_file(stp->st_stid.sc_file);
6979 status = nfserr_lock_range;
6982 file_lock = locks_alloc_lock();
6984 dprintk("NFSD: %s: unable to allocate lock!\n", __func__);
6985 status = nfserr_jukebox;
6989 file_lock->fl_type = F_UNLCK;
6990 file_lock->fl_owner = (fl_owner_t)lockowner(nfs4_get_stateowner(stp->st_stateowner));
6991 file_lock->fl_pid = current->tgid;
6992 file_lock->fl_file = nf->nf_file;
6993 file_lock->fl_flags = FL_POSIX;
6994 file_lock->fl_lmops = &nfsd_posix_mng_ops;
6995 file_lock->fl_start = locku->lu_offset;
6997 file_lock->fl_end = last_byte_offset(locku->lu_offset,
6999 nfs4_transform_lock_offset(file_lock);
7001 err = vfs_lock_file(nf->nf_file, F_SETLK, file_lock, NULL);
7003 dprintk("NFSD: nfs4_locku: vfs_lock_file failed!\n");
7006 nfs4_inc_and_copy_stateid(&locku->lu_stateid, &stp->st_stid);
7010 mutex_unlock(&stp->st_mutex);
7011 nfs4_put_stid(&stp->st_stid);
7013 nfsd4_bump_seqid(cstate, status);
7015 locks_free_lock(file_lock);
7019 status = nfserrno(err);
7025 * true: locks held by lockowner
7026 * false: no locks held by lockowner
7029 check_for_locks(struct nfs4_file *fp, struct nfs4_lockowner *lowner)
7031 struct file_lock *fl;
7033 struct nfsd_file *nf = find_any_file(fp);
7034 struct inode *inode;
7035 struct file_lock_context *flctx;
7038 /* Any valid lock stateid should have some sort of access */
7043 inode = locks_inode(nf->nf_file);
7044 flctx = inode->i_flctx;
7046 if (flctx && !list_empty_careful(&flctx->flc_posix)) {
7047 spin_lock(&flctx->flc_lock);
7048 list_for_each_entry(fl, &flctx->flc_posix, fl_list) {
7049 if (fl->fl_owner == (fl_owner_t)lowner) {
7054 spin_unlock(&flctx->flc_lock);
7061 nfsd4_release_lockowner(struct svc_rqst *rqstp,
7062 struct nfsd4_compound_state *cstate,
7063 union nfsd4_op_u *u)
7065 struct nfsd4_release_lockowner *rlockowner = &u->release_lockowner;
7066 clientid_t *clid = &rlockowner->rl_clientid;
7067 struct nfs4_stateowner *sop;
7068 struct nfs4_lockowner *lo = NULL;
7069 struct nfs4_ol_stateid *stp;
7070 struct xdr_netobj *owner = &rlockowner->rl_owner;
7071 unsigned int hashval = ownerstr_hashval(owner);
7073 struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
7074 struct nfs4_client *clp;
7075 LIST_HEAD (reaplist);
7077 dprintk("nfsd4_release_lockowner clientid: (%08x/%08x):\n",
7078 clid->cl_boot, clid->cl_id);
7080 status = set_client(clid, cstate, nn);
7085 /* Find the matching lock stateowner */
7086 spin_lock(&clp->cl_lock);
7087 list_for_each_entry(sop, &clp->cl_ownerstr_hashtbl[hashval],
7090 if (sop->so_is_open_owner || !same_owner_str(sop, owner))
7093 /* see if there are still any locks associated with it */
7094 lo = lockowner(sop);
7095 list_for_each_entry(stp, &sop->so_stateids, st_perstateowner) {
7096 if (check_for_locks(stp->st_stid.sc_file, lo)) {
7097 status = nfserr_locks_held;
7098 spin_unlock(&clp->cl_lock);
7103 nfs4_get_stateowner(sop);
7107 spin_unlock(&clp->cl_lock);
7111 unhash_lockowner_locked(lo);
7112 while (!list_empty(&lo->lo_owner.so_stateids)) {
7113 stp = list_first_entry(&lo->lo_owner.so_stateids,
7114 struct nfs4_ol_stateid,
7116 WARN_ON(!unhash_lock_stateid(stp));
7117 put_ol_stateid_locked(stp, &reaplist);
7119 spin_unlock(&clp->cl_lock);
7120 free_ol_stateid_reaplist(&reaplist);
7121 remove_blocked_locks(lo);
7122 nfs4_put_stateowner(&lo->lo_owner);
7127 static inline struct nfs4_client_reclaim *
7130 return kmalloc(sizeof(struct nfs4_client_reclaim), GFP_KERNEL);
7134 nfs4_has_reclaimed_state(struct xdr_netobj name, struct nfsd_net *nn)
7136 struct nfs4_client_reclaim *crp;
7138 crp = nfsd4_find_reclaim_client(name, nn);
7139 return (crp && crp->cr_clp);
7143 * failure => all reset bets are off, nfserr_no_grace...
7145 * The caller is responsible for freeing name.data if NULL is returned (it
7146 * will be freed in nfs4_remove_reclaim_record in the normal case).
7148 struct nfs4_client_reclaim *
7149 nfs4_client_to_reclaim(struct xdr_netobj name, struct xdr_netobj princhash,
7150 struct nfsd_net *nn)
7152 unsigned int strhashval;
7153 struct nfs4_client_reclaim *crp;
7155 trace_nfsd_clid_reclaim(nn, name.len, name.data);
7156 crp = alloc_reclaim();
7158 strhashval = clientstr_hashval(name);
7159 INIT_LIST_HEAD(&crp->cr_strhash);
7160 list_add(&crp->cr_strhash, &nn->reclaim_str_hashtbl[strhashval]);
7161 crp->cr_name.data = name.data;
7162 crp->cr_name.len = name.len;
7163 crp->cr_princhash.data = princhash.data;
7164 crp->cr_princhash.len = princhash.len;
7166 nn->reclaim_str_hashtbl_size++;
7172 nfs4_remove_reclaim_record(struct nfs4_client_reclaim *crp, struct nfsd_net *nn)
7174 list_del(&crp->cr_strhash);
7175 kfree(crp->cr_name.data);
7176 kfree(crp->cr_princhash.data);
7178 nn->reclaim_str_hashtbl_size--;
7182 nfs4_release_reclaim(struct nfsd_net *nn)
7184 struct nfs4_client_reclaim *crp = NULL;
7187 for (i = 0; i < CLIENT_HASH_SIZE; i++) {
7188 while (!list_empty(&nn->reclaim_str_hashtbl[i])) {
7189 crp = list_entry(nn->reclaim_str_hashtbl[i].next,
7190 struct nfs4_client_reclaim, cr_strhash);
7191 nfs4_remove_reclaim_record(crp, nn);
7194 WARN_ON_ONCE(nn->reclaim_str_hashtbl_size);
7198 * called from OPEN, CLAIM_PREVIOUS with a new clientid. */
7199 struct nfs4_client_reclaim *
7200 nfsd4_find_reclaim_client(struct xdr_netobj name, struct nfsd_net *nn)
7202 unsigned int strhashval;
7203 struct nfs4_client_reclaim *crp = NULL;
7205 trace_nfsd_clid_find(nn, name.len, name.data);
7207 strhashval = clientstr_hashval(name);
7208 list_for_each_entry(crp, &nn->reclaim_str_hashtbl[strhashval], cr_strhash) {
7209 if (compare_blob(&crp->cr_name, &name) == 0) {
7217 nfs4_check_open_reclaim(struct nfs4_client *clp)
7219 if (test_bit(NFSD4_CLIENT_RECLAIM_COMPLETE, &clp->cl_flags))
7220 return nfserr_no_grace;
7222 if (nfsd4_client_record_check(clp))
7223 return nfserr_reclaim_bad;
7229 * Since the lifetime of a delegation isn't limited to that of an open, a
7230 * client may quite reasonably hang on to a delegation as long as it has
7231 * the inode cached. This becomes an obvious problem the first time a
7232 * client's inode cache approaches the size of the server's total memory.
7234 * For now we avoid this problem by imposing a hard limit on the number
7235 * of delegations, which varies according to the server's memory size.
7238 set_max_delegations(void)
7241 * Allow at most 4 delegations per megabyte of RAM. Quick
7242 * estimates suggest that in the worst case (where every delegation
7243 * is for a different inode), a delegation could take about 1.5K,
7244 * giving a worst case usage of about 6% of memory.
7246 max_delegations = nr_free_buffer_pages() >> (20 - 2 - PAGE_SHIFT);
7249 static int nfs4_state_create_net(struct net *net)
7251 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
7254 nn->conf_id_hashtbl = kmalloc_array(CLIENT_HASH_SIZE,
7255 sizeof(struct list_head),
7257 if (!nn->conf_id_hashtbl)
7259 nn->unconf_id_hashtbl = kmalloc_array(CLIENT_HASH_SIZE,
7260 sizeof(struct list_head),
7262 if (!nn->unconf_id_hashtbl)
7264 nn->sessionid_hashtbl = kmalloc_array(SESSION_HASH_SIZE,
7265 sizeof(struct list_head),
7267 if (!nn->sessionid_hashtbl)
7270 for (i = 0; i < CLIENT_HASH_SIZE; i++) {
7271 INIT_LIST_HEAD(&nn->conf_id_hashtbl[i]);
7272 INIT_LIST_HEAD(&nn->unconf_id_hashtbl[i]);
7274 for (i = 0; i < SESSION_HASH_SIZE; i++)
7275 INIT_LIST_HEAD(&nn->sessionid_hashtbl[i]);
7276 nn->conf_name_tree = RB_ROOT;
7277 nn->unconf_name_tree = RB_ROOT;
7278 nn->boot_time = ktime_get_real_seconds();
7279 nn->grace_ended = false;
7280 nn->nfsd4_manager.block_opens = true;
7281 INIT_LIST_HEAD(&nn->nfsd4_manager.list);
7282 INIT_LIST_HEAD(&nn->client_lru);
7283 INIT_LIST_HEAD(&nn->close_lru);
7284 INIT_LIST_HEAD(&nn->del_recall_lru);
7285 spin_lock_init(&nn->client_lock);
7286 spin_lock_init(&nn->s2s_cp_lock);
7287 idr_init(&nn->s2s_cp_stateids);
7289 spin_lock_init(&nn->blocked_locks_lock);
7290 INIT_LIST_HEAD(&nn->blocked_locks_lru);
7292 INIT_DELAYED_WORK(&nn->laundromat_work, laundromat_main);
7298 kfree(nn->unconf_id_hashtbl);
7300 kfree(nn->conf_id_hashtbl);
7306 nfs4_state_destroy_net(struct net *net)
7309 struct nfs4_client *clp = NULL;
7310 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
7312 for (i = 0; i < CLIENT_HASH_SIZE; i++) {
7313 while (!list_empty(&nn->conf_id_hashtbl[i])) {
7314 clp = list_entry(nn->conf_id_hashtbl[i].next, struct nfs4_client, cl_idhash);
7315 destroy_client(clp);
7319 WARN_ON(!list_empty(&nn->blocked_locks_lru));
7321 for (i = 0; i < CLIENT_HASH_SIZE; i++) {
7322 while (!list_empty(&nn->unconf_id_hashtbl[i])) {
7323 clp = list_entry(nn->unconf_id_hashtbl[i].next, struct nfs4_client, cl_idhash);
7324 destroy_client(clp);
7328 kfree(nn->sessionid_hashtbl);
7329 kfree(nn->unconf_id_hashtbl);
7330 kfree(nn->conf_id_hashtbl);
7335 nfs4_state_start_net(struct net *net)
7337 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
7340 ret = nfs4_state_create_net(net);
7343 locks_start_grace(net, &nn->nfsd4_manager);
7344 nfsd4_client_tracking_init(net);
7345 if (nn->track_reclaim_completes && nn->reclaim_str_hashtbl_size == 0)
7347 printk(KERN_INFO "NFSD: starting %lld-second grace period (net %x)\n",
7348 nn->nfsd4_grace, net->ns.inum);
7349 trace_nfsd_grace_start(nn);
7350 queue_delayed_work(laundry_wq, &nn->laundromat_work, nn->nfsd4_grace * HZ);
7354 printk(KERN_INFO "NFSD: no clients to reclaim, skipping NFSv4 grace period (net %x)\n",
7356 queue_delayed_work(laundry_wq, &nn->laundromat_work, nn->nfsd4_lease * HZ);
7357 nfsd4_end_grace(nn);
7361 /* initialization to perform when the nfsd service is started: */
7364 nfs4_state_start(void)
7368 laundry_wq = alloc_workqueue("%s", WQ_UNBOUND, 0, "nfsd4");
7369 if (laundry_wq == NULL) {
7373 ret = nfsd4_create_callback_queue();
7375 goto out_free_laundry;
7377 set_max_delegations();
7381 destroy_workqueue(laundry_wq);
7387 nfs4_state_shutdown_net(struct net *net)
7389 struct nfs4_delegation *dp = NULL;
7390 struct list_head *pos, *next, reaplist;
7391 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
7393 cancel_delayed_work_sync(&nn->laundromat_work);
7394 locks_end_grace(&nn->nfsd4_manager);
7396 INIT_LIST_HEAD(&reaplist);
7397 spin_lock(&state_lock);
7398 list_for_each_safe(pos, next, &nn->del_recall_lru) {
7399 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
7400 WARN_ON(!unhash_delegation_locked(dp));
7401 list_add(&dp->dl_recall_lru, &reaplist);
7403 spin_unlock(&state_lock);
7404 list_for_each_safe(pos, next, &reaplist) {
7405 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
7406 list_del_init(&dp->dl_recall_lru);
7407 destroy_unhashed_deleg(dp);
7410 nfsd4_client_tracking_exit(net);
7411 nfs4_state_destroy_net(net);
7415 nfs4_state_shutdown(void)
7417 destroy_workqueue(laundry_wq);
7418 nfsd4_destroy_callback_queue();
7422 get_stateid(struct nfsd4_compound_state *cstate, stateid_t *stateid)
7424 if (HAS_CSTATE_FLAG(cstate, CURRENT_STATE_ID_FLAG) &&
7425 CURRENT_STATEID(stateid))
7426 memcpy(stateid, &cstate->current_stateid, sizeof(stateid_t));
7430 put_stateid(struct nfsd4_compound_state *cstate, stateid_t *stateid)
7432 if (cstate->minorversion) {
7433 memcpy(&cstate->current_stateid, stateid, sizeof(stateid_t));
7434 SET_CSTATE_FLAG(cstate, CURRENT_STATE_ID_FLAG);
7439 clear_current_stateid(struct nfsd4_compound_state *cstate)
7441 CLEAR_CSTATE_FLAG(cstate, CURRENT_STATE_ID_FLAG);
7445 * functions to set current state id
7448 nfsd4_set_opendowngradestateid(struct nfsd4_compound_state *cstate,
7449 union nfsd4_op_u *u)
7451 put_stateid(cstate, &u->open_downgrade.od_stateid);
7455 nfsd4_set_openstateid(struct nfsd4_compound_state *cstate,
7456 union nfsd4_op_u *u)
7458 put_stateid(cstate, &u->open.op_stateid);
7462 nfsd4_set_closestateid(struct nfsd4_compound_state *cstate,
7463 union nfsd4_op_u *u)
7465 put_stateid(cstate, &u->close.cl_stateid);
7469 nfsd4_set_lockstateid(struct nfsd4_compound_state *cstate,
7470 union nfsd4_op_u *u)
7472 put_stateid(cstate, &u->lock.lk_resp_stateid);
7476 * functions to consume current state id
7480 nfsd4_get_opendowngradestateid(struct nfsd4_compound_state *cstate,
7481 union nfsd4_op_u *u)
7483 get_stateid(cstate, &u->open_downgrade.od_stateid);
7487 nfsd4_get_delegreturnstateid(struct nfsd4_compound_state *cstate,
7488 union nfsd4_op_u *u)
7490 get_stateid(cstate, &u->delegreturn.dr_stateid);
7494 nfsd4_get_freestateid(struct nfsd4_compound_state *cstate,
7495 union nfsd4_op_u *u)
7497 get_stateid(cstate, &u->free_stateid.fr_stateid);
7501 nfsd4_get_setattrstateid(struct nfsd4_compound_state *cstate,
7502 union nfsd4_op_u *u)
7504 get_stateid(cstate, &u->setattr.sa_stateid);
7508 nfsd4_get_closestateid(struct nfsd4_compound_state *cstate,
7509 union nfsd4_op_u *u)
7511 get_stateid(cstate, &u->close.cl_stateid);
7515 nfsd4_get_lockustateid(struct nfsd4_compound_state *cstate,
7516 union nfsd4_op_u *u)
7518 get_stateid(cstate, &u->locku.lu_stateid);
7522 nfsd4_get_readstateid(struct nfsd4_compound_state *cstate,
7523 union nfsd4_op_u *u)
7525 get_stateid(cstate, &u->read.rd_stateid);
7529 nfsd4_get_writestateid(struct nfsd4_compound_state *cstate,
7530 union nfsd4_op_u *u)
7532 get_stateid(cstate, &u->write.wr_stateid);