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>
48 #include "current_stateid.h"
53 #define NFSDDBG_FACILITY NFSDDBG_PROC
55 #define all_ones {{~0,~0},~0}
56 static const stateid_t one_stateid = {
58 .si_opaque = all_ones,
60 static const stateid_t zero_stateid = {
63 static const stateid_t currentstateid = {
67 static u64 current_sessionid = 1;
69 #define ZERO_STATEID(stateid) (!memcmp((stateid), &zero_stateid, sizeof(stateid_t)))
70 #define ONE_STATEID(stateid) (!memcmp((stateid), &one_stateid, sizeof(stateid_t)))
71 #define CURRENT_STATEID(stateid) (!memcmp((stateid), ¤tstateid, sizeof(stateid_t)))
73 /* forward declarations */
74 static bool check_for_locks(struct nfs4_file *fp, struct nfs4_lockowner *lowner);
75 static void nfs4_free_ol_stateid(struct nfs4_stid *stid);
80 * Currently used for the del_recall_lru and file hash table. In an
81 * effort to decrease the scope of the client_mutex, this spinlock may
82 * eventually cover more:
84 static DEFINE_SPINLOCK(state_lock);
87 * A waitqueue for all in-progress 4.0 CLOSE operations that are waiting for
88 * the refcount on the open stateid to drop.
90 static DECLARE_WAIT_QUEUE_HEAD(close_wq);
92 static struct kmem_cache *openowner_slab;
93 static struct kmem_cache *lockowner_slab;
94 static struct kmem_cache *file_slab;
95 static struct kmem_cache *stateid_slab;
96 static struct kmem_cache *deleg_slab;
97 static struct kmem_cache *odstate_slab;
99 static void free_session(struct nfsd4_session *);
101 static const struct nfsd4_callback_ops nfsd4_cb_recall_ops;
103 static bool is_session_dead(struct nfsd4_session *ses)
105 return ses->se_flags & NFS4_SESSION_DEAD;
108 static __be32 mark_session_dead_locked(struct nfsd4_session *ses, int ref_held_by_me)
110 if (atomic_read(&ses->se_ref) > ref_held_by_me)
111 return nfserr_jukebox;
112 ses->se_flags |= NFS4_SESSION_DEAD;
116 static bool is_client_expired(struct nfs4_client *clp)
118 return clp->cl_time == 0;
121 static __be32 get_client_locked(struct nfs4_client *clp)
123 struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
125 lockdep_assert_held(&nn->client_lock);
127 if (is_client_expired(clp))
128 return nfserr_expired;
129 atomic_inc(&clp->cl_refcount);
133 /* must be called under the client_lock */
135 renew_client_locked(struct nfs4_client *clp)
137 struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
139 if (is_client_expired(clp)) {
141 printk("%s: client (clientid %08x/%08x) already expired\n",
143 clp->cl_clientid.cl_boot,
144 clp->cl_clientid.cl_id);
148 dprintk("renewing client (clientid %08x/%08x)\n",
149 clp->cl_clientid.cl_boot,
150 clp->cl_clientid.cl_id);
151 list_move_tail(&clp->cl_lru, &nn->client_lru);
152 clp->cl_time = get_seconds();
155 static void put_client_renew_locked(struct nfs4_client *clp)
157 struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
159 lockdep_assert_held(&nn->client_lock);
161 if (!atomic_dec_and_test(&clp->cl_refcount))
163 if (!is_client_expired(clp))
164 renew_client_locked(clp);
167 static void put_client_renew(struct nfs4_client *clp)
169 struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
171 if (!atomic_dec_and_lock(&clp->cl_refcount, &nn->client_lock))
173 if (!is_client_expired(clp))
174 renew_client_locked(clp);
175 spin_unlock(&nn->client_lock);
178 static __be32 nfsd4_get_session_locked(struct nfsd4_session *ses)
182 if (is_session_dead(ses))
183 return nfserr_badsession;
184 status = get_client_locked(ses->se_client);
187 atomic_inc(&ses->se_ref);
191 static void nfsd4_put_session_locked(struct nfsd4_session *ses)
193 struct nfs4_client *clp = ses->se_client;
194 struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
196 lockdep_assert_held(&nn->client_lock);
198 if (atomic_dec_and_test(&ses->se_ref) && is_session_dead(ses))
200 put_client_renew_locked(clp);
203 static void nfsd4_put_session(struct nfsd4_session *ses)
205 struct nfs4_client *clp = ses->se_client;
206 struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
208 spin_lock(&nn->client_lock);
209 nfsd4_put_session_locked(ses);
210 spin_unlock(&nn->client_lock);
213 static inline struct nfs4_stateowner *
214 nfs4_get_stateowner(struct nfs4_stateowner *sop)
216 atomic_inc(&sop->so_count);
221 same_owner_str(struct nfs4_stateowner *sop, struct xdr_netobj *owner)
223 return (sop->so_owner.len == owner->len) &&
224 0 == memcmp(sop->so_owner.data, owner->data, owner->len);
227 static struct nfs4_openowner *
228 find_openstateowner_str_locked(unsigned int hashval, struct nfsd4_open *open,
229 struct nfs4_client *clp)
231 struct nfs4_stateowner *so;
233 lockdep_assert_held(&clp->cl_lock);
235 list_for_each_entry(so, &clp->cl_ownerstr_hashtbl[hashval],
237 if (!so->so_is_open_owner)
239 if (same_owner_str(so, &open->op_owner))
240 return openowner(nfs4_get_stateowner(so));
245 static struct nfs4_openowner *
246 find_openstateowner_str(unsigned int hashval, struct nfsd4_open *open,
247 struct nfs4_client *clp)
249 struct nfs4_openowner *oo;
251 spin_lock(&clp->cl_lock);
252 oo = find_openstateowner_str_locked(hashval, open, clp);
253 spin_unlock(&clp->cl_lock);
258 opaque_hashval(const void *ptr, int nbytes)
260 unsigned char *cptr = (unsigned char *) ptr;
270 static void nfsd4_free_file_rcu(struct rcu_head *rcu)
272 struct nfs4_file *fp = container_of(rcu, struct nfs4_file, fi_rcu);
274 kmem_cache_free(file_slab, fp);
278 put_nfs4_file(struct nfs4_file *fi)
280 might_lock(&state_lock);
282 if (atomic_dec_and_lock(&fi->fi_ref, &state_lock)) {
283 hlist_del_rcu(&fi->fi_hash);
284 spin_unlock(&state_lock);
285 WARN_ON_ONCE(!list_empty(&fi->fi_clnt_odstate));
286 WARN_ON_ONCE(!list_empty(&fi->fi_delegations));
287 call_rcu(&fi->fi_rcu, nfsd4_free_file_rcu);
292 __nfs4_get_fd(struct nfs4_file *f, int oflag)
294 if (f->fi_fds[oflag])
295 return get_file(f->fi_fds[oflag]);
300 find_writeable_file_locked(struct nfs4_file *f)
304 lockdep_assert_held(&f->fi_lock);
306 ret = __nfs4_get_fd(f, O_WRONLY);
308 ret = __nfs4_get_fd(f, O_RDWR);
313 find_writeable_file(struct nfs4_file *f)
317 spin_lock(&f->fi_lock);
318 ret = find_writeable_file_locked(f);
319 spin_unlock(&f->fi_lock);
324 static struct file *find_readable_file_locked(struct nfs4_file *f)
328 lockdep_assert_held(&f->fi_lock);
330 ret = __nfs4_get_fd(f, O_RDONLY);
332 ret = __nfs4_get_fd(f, O_RDWR);
337 find_readable_file(struct nfs4_file *f)
341 spin_lock(&f->fi_lock);
342 ret = find_readable_file_locked(f);
343 spin_unlock(&f->fi_lock);
349 find_any_file(struct nfs4_file *f)
353 spin_lock(&f->fi_lock);
354 ret = __nfs4_get_fd(f, O_RDWR);
356 ret = __nfs4_get_fd(f, O_WRONLY);
358 ret = __nfs4_get_fd(f, O_RDONLY);
360 spin_unlock(&f->fi_lock);
364 static atomic_long_t num_delegations;
365 unsigned long max_delegations;
368 * Open owner state (share locks)
371 /* hash tables for lock and open owners */
372 #define OWNER_HASH_BITS 8
373 #define OWNER_HASH_SIZE (1 << OWNER_HASH_BITS)
374 #define OWNER_HASH_MASK (OWNER_HASH_SIZE - 1)
376 static unsigned int ownerstr_hashval(struct xdr_netobj *ownername)
380 ret = opaque_hashval(ownername->data, ownername->len);
381 return ret & OWNER_HASH_MASK;
384 /* hash table for nfs4_file */
385 #define FILE_HASH_BITS 8
386 #define FILE_HASH_SIZE (1 << FILE_HASH_BITS)
388 static unsigned int nfsd_fh_hashval(struct knfsd_fh *fh)
390 return jhash2(fh->fh_base.fh_pad, XDR_QUADLEN(fh->fh_size), 0);
393 static unsigned int file_hashval(struct knfsd_fh *fh)
395 return nfsd_fh_hashval(fh) & (FILE_HASH_SIZE - 1);
398 static struct hlist_head file_hashtbl[FILE_HASH_SIZE];
401 __nfs4_file_get_access(struct nfs4_file *fp, u32 access)
403 lockdep_assert_held(&fp->fi_lock);
405 if (access & NFS4_SHARE_ACCESS_WRITE)
406 atomic_inc(&fp->fi_access[O_WRONLY]);
407 if (access & NFS4_SHARE_ACCESS_READ)
408 atomic_inc(&fp->fi_access[O_RDONLY]);
412 nfs4_file_get_access(struct nfs4_file *fp, u32 access)
414 lockdep_assert_held(&fp->fi_lock);
416 /* Does this access mode make sense? */
417 if (access & ~NFS4_SHARE_ACCESS_BOTH)
420 /* Does it conflict with a deny mode already set? */
421 if ((access & fp->fi_share_deny) != 0)
422 return nfserr_share_denied;
424 __nfs4_file_get_access(fp, access);
428 static __be32 nfs4_file_check_deny(struct nfs4_file *fp, u32 deny)
430 /* Common case is that there is no deny mode. */
432 /* Does this deny mode make sense? */
433 if (deny & ~NFS4_SHARE_DENY_BOTH)
436 if ((deny & NFS4_SHARE_DENY_READ) &&
437 atomic_read(&fp->fi_access[O_RDONLY]))
438 return nfserr_share_denied;
440 if ((deny & NFS4_SHARE_DENY_WRITE) &&
441 atomic_read(&fp->fi_access[O_WRONLY]))
442 return nfserr_share_denied;
447 static void __nfs4_file_put_access(struct nfs4_file *fp, int oflag)
449 might_lock(&fp->fi_lock);
451 if (atomic_dec_and_lock(&fp->fi_access[oflag], &fp->fi_lock)) {
452 struct file *f1 = NULL;
453 struct file *f2 = NULL;
455 swap(f1, fp->fi_fds[oflag]);
456 if (atomic_read(&fp->fi_access[1 - oflag]) == 0)
457 swap(f2, fp->fi_fds[O_RDWR]);
458 spin_unlock(&fp->fi_lock);
466 static void nfs4_file_put_access(struct nfs4_file *fp, u32 access)
468 WARN_ON_ONCE(access & ~NFS4_SHARE_ACCESS_BOTH);
470 if (access & NFS4_SHARE_ACCESS_WRITE)
471 __nfs4_file_put_access(fp, O_WRONLY);
472 if (access & NFS4_SHARE_ACCESS_READ)
473 __nfs4_file_put_access(fp, O_RDONLY);
477 * Allocate a new open/delegation state counter. This is needed for
478 * pNFS for proper return on close semantics.
480 * Note that we only allocate it for pNFS-enabled exports, otherwise
481 * all pointers to struct nfs4_clnt_odstate are always NULL.
483 static struct nfs4_clnt_odstate *
484 alloc_clnt_odstate(struct nfs4_client *clp)
486 struct nfs4_clnt_odstate *co;
488 co = kmem_cache_zalloc(odstate_slab, GFP_KERNEL);
491 atomic_set(&co->co_odcount, 1);
497 hash_clnt_odstate_locked(struct nfs4_clnt_odstate *co)
499 struct nfs4_file *fp = co->co_file;
501 lockdep_assert_held(&fp->fi_lock);
502 list_add(&co->co_perfile, &fp->fi_clnt_odstate);
506 get_clnt_odstate(struct nfs4_clnt_odstate *co)
509 atomic_inc(&co->co_odcount);
513 put_clnt_odstate(struct nfs4_clnt_odstate *co)
515 struct nfs4_file *fp;
521 if (atomic_dec_and_lock(&co->co_odcount, &fp->fi_lock)) {
522 list_del(&co->co_perfile);
523 spin_unlock(&fp->fi_lock);
525 nfsd4_return_all_file_layouts(co->co_client, fp);
526 kmem_cache_free(odstate_slab, co);
530 static struct nfs4_clnt_odstate *
531 find_or_hash_clnt_odstate(struct nfs4_file *fp, struct nfs4_clnt_odstate *new)
533 struct nfs4_clnt_odstate *co;
534 struct nfs4_client *cl;
541 spin_lock(&fp->fi_lock);
542 list_for_each_entry(co, &fp->fi_clnt_odstate, co_perfile) {
543 if (co->co_client == cl) {
544 get_clnt_odstate(co);
550 hash_clnt_odstate_locked(new);
552 spin_unlock(&fp->fi_lock);
556 struct nfs4_stid *nfs4_alloc_stid(struct nfs4_client *cl,
557 struct kmem_cache *slab)
559 struct nfs4_stid *stid;
562 stid = kmem_cache_zalloc(slab, GFP_KERNEL);
566 idr_preload(GFP_KERNEL);
567 spin_lock(&cl->cl_lock);
568 new_id = idr_alloc_cyclic(&cl->cl_stateids, stid, 0, 0, GFP_NOWAIT);
569 spin_unlock(&cl->cl_lock);
573 stid->sc_client = cl;
574 stid->sc_stateid.si_opaque.so_id = new_id;
575 stid->sc_stateid.si_opaque.so_clid = cl->cl_clientid;
576 /* Will be incremented before return to client: */
577 atomic_set(&stid->sc_count, 1);
578 spin_lock_init(&stid->sc_lock);
581 * It shouldn't be a problem to reuse an opaque stateid value.
582 * I don't think it is for 4.1. But with 4.0 I worry that, for
583 * example, a stray write retransmission could be accepted by
584 * the server when it should have been rejected. Therefore,
585 * adopt a trick from the sctp code to attempt to maximize the
586 * amount of time until an id is reused, by ensuring they always
587 * "increase" (mod INT_MAX):
591 kmem_cache_free(slab, stid);
595 static struct nfs4_ol_stateid * nfs4_alloc_open_stateid(struct nfs4_client *clp)
597 struct nfs4_stid *stid;
598 struct nfs4_ol_stateid *stp;
600 stid = nfs4_alloc_stid(clp, stateid_slab);
604 stp = openlockstateid(stid);
605 stp->st_stid.sc_free = nfs4_free_ol_stateid;
609 static void nfs4_free_deleg(struct nfs4_stid *stid)
611 kmem_cache_free(deleg_slab, stid);
612 atomic_long_dec(&num_delegations);
616 * When we recall a delegation, we should be careful not to hand it
617 * out again straight away.
618 * To ensure this we keep a pair of bloom filters ('new' and 'old')
619 * in which the filehandles of recalled delegations are "stored".
620 * If a filehandle appear in either filter, a delegation is blocked.
621 * When a delegation is recalled, the filehandle is stored in the "new"
623 * Every 30 seconds we swap the filters and clear the "new" one,
624 * unless both are empty of course.
626 * Each filter is 256 bits. We hash the filehandle to 32bit and use the
627 * low 3 bytes as hash-table indices.
629 * 'blocked_delegations_lock', which is always taken in block_delegations(),
630 * is used to manage concurrent access. Testing does not need the lock
631 * except when swapping the two filters.
633 static DEFINE_SPINLOCK(blocked_delegations_lock);
634 static struct bloom_pair {
635 int entries, old_entries;
637 int new; /* index into 'set' */
638 DECLARE_BITMAP(set[2], 256);
639 } blocked_delegations;
641 static int delegation_blocked(struct knfsd_fh *fh)
644 struct bloom_pair *bd = &blocked_delegations;
646 if (bd->entries == 0)
648 if (seconds_since_boot() - bd->swap_time > 30) {
649 spin_lock(&blocked_delegations_lock);
650 if (seconds_since_boot() - bd->swap_time > 30) {
651 bd->entries -= bd->old_entries;
652 bd->old_entries = bd->entries;
653 memset(bd->set[bd->new], 0,
656 bd->swap_time = seconds_since_boot();
658 spin_unlock(&blocked_delegations_lock);
660 hash = jhash(&fh->fh_base, fh->fh_size, 0);
661 if (test_bit(hash&255, bd->set[0]) &&
662 test_bit((hash>>8)&255, bd->set[0]) &&
663 test_bit((hash>>16)&255, bd->set[0]))
666 if (test_bit(hash&255, bd->set[1]) &&
667 test_bit((hash>>8)&255, bd->set[1]) &&
668 test_bit((hash>>16)&255, bd->set[1]))
674 static void block_delegations(struct knfsd_fh *fh)
677 struct bloom_pair *bd = &blocked_delegations;
679 hash = jhash(&fh->fh_base, fh->fh_size, 0);
681 spin_lock(&blocked_delegations_lock);
682 __set_bit(hash&255, bd->set[bd->new]);
683 __set_bit((hash>>8)&255, bd->set[bd->new]);
684 __set_bit((hash>>16)&255, bd->set[bd->new]);
685 if (bd->entries == 0)
686 bd->swap_time = seconds_since_boot();
688 spin_unlock(&blocked_delegations_lock);
691 static struct nfs4_delegation *
692 alloc_init_deleg(struct nfs4_client *clp, struct svc_fh *current_fh,
693 struct nfs4_clnt_odstate *odstate)
695 struct nfs4_delegation *dp;
698 dprintk("NFSD alloc_init_deleg\n");
699 n = atomic_long_inc_return(&num_delegations);
700 if (n < 0 || n > max_delegations)
702 if (delegation_blocked(¤t_fh->fh_handle))
704 dp = delegstateid(nfs4_alloc_stid(clp, deleg_slab));
708 dp->dl_stid.sc_free = nfs4_free_deleg;
710 * delegation seqid's are never incremented. The 4.1 special
711 * meaning of seqid 0 isn't meaningful, really, but let's avoid
712 * 0 anyway just for consistency and use 1:
714 dp->dl_stid.sc_stateid.si_generation = 1;
715 INIT_LIST_HEAD(&dp->dl_perfile);
716 INIT_LIST_HEAD(&dp->dl_perclnt);
717 INIT_LIST_HEAD(&dp->dl_recall_lru);
718 dp->dl_clnt_odstate = odstate;
719 get_clnt_odstate(odstate);
720 dp->dl_type = NFS4_OPEN_DELEGATE_READ;
722 nfsd4_init_cb(&dp->dl_recall, dp->dl_stid.sc_client,
723 &nfsd4_cb_recall_ops, NFSPROC4_CLNT_CB_RECALL);
726 atomic_long_dec(&num_delegations);
731 nfs4_put_stid(struct nfs4_stid *s)
733 struct nfs4_file *fp = s->sc_file;
734 struct nfs4_client *clp = s->sc_client;
736 might_lock(&clp->cl_lock);
738 if (!atomic_dec_and_lock(&s->sc_count, &clp->cl_lock)) {
739 wake_up_all(&close_wq);
742 idr_remove(&clp->cl_stateids, s->sc_stateid.si_opaque.so_id);
743 spin_unlock(&clp->cl_lock);
750 nfs4_inc_and_copy_stateid(stateid_t *dst, struct nfs4_stid *stid)
752 stateid_t *src = &stid->sc_stateid;
754 spin_lock(&stid->sc_lock);
755 if (unlikely(++src->si_generation == 0))
756 src->si_generation = 1;
757 memcpy(dst, src, sizeof(*dst));
758 spin_unlock(&stid->sc_lock);
761 static void nfs4_put_deleg_lease(struct nfs4_file *fp)
763 struct file *filp = NULL;
765 spin_lock(&fp->fi_lock);
766 if (fp->fi_deleg_file && --fp->fi_delegees == 0)
767 swap(filp, fp->fi_deleg_file);
768 spin_unlock(&fp->fi_lock);
771 vfs_setlease(filp, F_UNLCK, NULL, (void **)&fp);
776 void nfs4_unhash_stid(struct nfs4_stid *s)
782 * nfs4_get_existing_delegation - Discover if this delegation already exists
783 * @clp: a pointer to the nfs4_client we're granting a delegation to
784 * @fp: a pointer to the nfs4_file we're granting a delegation on
787 * On success: NULL if an existing delegation was not found.
789 * On error: -EAGAIN if one was previously granted to this nfs4_client
790 * for this nfs4_file.
795 nfs4_get_existing_delegation(struct nfs4_client *clp, struct nfs4_file *fp)
797 struct nfs4_delegation *searchdp = NULL;
798 struct nfs4_client *searchclp = NULL;
800 lockdep_assert_held(&state_lock);
801 lockdep_assert_held(&fp->fi_lock);
803 list_for_each_entry(searchdp, &fp->fi_delegations, dl_perfile) {
804 searchclp = searchdp->dl_stid.sc_client;
805 if (clp == searchclp) {
813 * hash_delegation_locked - Add a delegation to the appropriate lists
814 * @dp: a pointer to the nfs4_delegation we are adding.
815 * @fp: a pointer to the nfs4_file we're granting a delegation on
818 * On success: NULL if the delegation was successfully hashed.
820 * On error: -EAGAIN if one was previously granted to this
821 * nfs4_client for this nfs4_file. Delegation is not hashed.
826 hash_delegation_locked(struct nfs4_delegation *dp, struct nfs4_file *fp)
829 struct nfs4_client *clp = dp->dl_stid.sc_client;
831 lockdep_assert_held(&state_lock);
832 lockdep_assert_held(&fp->fi_lock);
834 status = nfs4_get_existing_delegation(clp, fp);
838 atomic_inc(&dp->dl_stid.sc_count);
839 dp->dl_stid.sc_type = NFS4_DELEG_STID;
840 list_add(&dp->dl_perfile, &fp->fi_delegations);
841 list_add(&dp->dl_perclnt, &clp->cl_delegations);
846 unhash_delegation_locked(struct nfs4_delegation *dp)
848 struct nfs4_file *fp = dp->dl_stid.sc_file;
850 lockdep_assert_held(&state_lock);
852 if (list_empty(&dp->dl_perfile))
855 dp->dl_stid.sc_type = NFS4_CLOSED_DELEG_STID;
856 /* Ensure that deleg break won't try to requeue it */
858 spin_lock(&fp->fi_lock);
859 list_del_init(&dp->dl_perclnt);
860 list_del_init(&dp->dl_recall_lru);
861 list_del_init(&dp->dl_perfile);
862 spin_unlock(&fp->fi_lock);
866 static void destroy_delegation(struct nfs4_delegation *dp)
870 spin_lock(&state_lock);
871 unhashed = unhash_delegation_locked(dp);
872 spin_unlock(&state_lock);
874 put_clnt_odstate(dp->dl_clnt_odstate);
875 nfs4_put_deleg_lease(dp->dl_stid.sc_file);
876 nfs4_put_stid(&dp->dl_stid);
880 static void revoke_delegation(struct nfs4_delegation *dp)
882 struct nfs4_client *clp = dp->dl_stid.sc_client;
884 WARN_ON(!list_empty(&dp->dl_recall_lru));
886 put_clnt_odstate(dp->dl_clnt_odstate);
887 nfs4_put_deleg_lease(dp->dl_stid.sc_file);
889 if (clp->cl_minorversion == 0)
890 nfs4_put_stid(&dp->dl_stid);
892 dp->dl_stid.sc_type = NFS4_REVOKED_DELEG_STID;
893 spin_lock(&clp->cl_lock);
894 list_add(&dp->dl_recall_lru, &clp->cl_revoked);
895 spin_unlock(&clp->cl_lock);
903 static unsigned int clientid_hashval(u32 id)
905 return id & CLIENT_HASH_MASK;
908 static unsigned int clientstr_hashval(const char *name)
910 return opaque_hashval(name, 8) & CLIENT_HASH_MASK;
914 * We store the NONE, READ, WRITE, and BOTH bits separately in the
915 * st_{access,deny}_bmap field of the stateid, in order to track not
916 * only what share bits are currently in force, but also what
917 * combinations of share bits previous opens have used. This allows us
918 * to enforce the recommendation of rfc 3530 14.2.19 that the server
919 * return an error if the client attempt to downgrade to a combination
920 * of share bits not explicable by closing some of its previous opens.
922 * XXX: This enforcement is actually incomplete, since we don't keep
923 * track of access/deny bit combinations; so, e.g., we allow:
925 * OPEN allow read, deny write
926 * OPEN allow both, deny none
927 * DOWNGRADE allow read, deny none
929 * which we should reject.
932 bmap_to_share_mode(unsigned long bmap) {
934 unsigned int access = 0;
936 for (i = 1; i < 4; i++) {
937 if (test_bit(i, &bmap))
943 /* set share access for a given stateid */
945 set_access(u32 access, struct nfs4_ol_stateid *stp)
947 unsigned char mask = 1 << access;
949 WARN_ON_ONCE(access > NFS4_SHARE_ACCESS_BOTH);
950 stp->st_access_bmap |= mask;
953 /* clear share access for a given stateid */
955 clear_access(u32 access, struct nfs4_ol_stateid *stp)
957 unsigned char mask = 1 << access;
959 WARN_ON_ONCE(access > NFS4_SHARE_ACCESS_BOTH);
960 stp->st_access_bmap &= ~mask;
963 /* test whether a given stateid has access */
965 test_access(u32 access, struct nfs4_ol_stateid *stp)
967 unsigned char mask = 1 << access;
969 return (bool)(stp->st_access_bmap & mask);
972 /* set share deny for a given stateid */
974 set_deny(u32 deny, struct nfs4_ol_stateid *stp)
976 unsigned char mask = 1 << deny;
978 WARN_ON_ONCE(deny > NFS4_SHARE_DENY_BOTH);
979 stp->st_deny_bmap |= mask;
982 /* clear share deny for a given stateid */
984 clear_deny(u32 deny, struct nfs4_ol_stateid *stp)
986 unsigned char mask = 1 << deny;
988 WARN_ON_ONCE(deny > NFS4_SHARE_DENY_BOTH);
989 stp->st_deny_bmap &= ~mask;
992 /* test whether a given stateid is denying specific access */
994 test_deny(u32 deny, struct nfs4_ol_stateid *stp)
996 unsigned char mask = 1 << deny;
998 return (bool)(stp->st_deny_bmap & mask);
1001 static int nfs4_access_to_omode(u32 access)
1003 switch (access & NFS4_SHARE_ACCESS_BOTH) {
1004 case NFS4_SHARE_ACCESS_READ:
1006 case NFS4_SHARE_ACCESS_WRITE:
1008 case NFS4_SHARE_ACCESS_BOTH:
1016 * A stateid that had a deny mode associated with it is being released
1017 * or downgraded. Recalculate the deny mode on the file.
1020 recalculate_deny_mode(struct nfs4_file *fp)
1022 struct nfs4_ol_stateid *stp;
1024 spin_lock(&fp->fi_lock);
1025 fp->fi_share_deny = 0;
1026 list_for_each_entry(stp, &fp->fi_stateids, st_perfile)
1027 fp->fi_share_deny |= bmap_to_share_mode(stp->st_deny_bmap);
1028 spin_unlock(&fp->fi_lock);
1032 reset_union_bmap_deny(u32 deny, struct nfs4_ol_stateid *stp)
1035 bool change = false;
1037 for (i = 1; i < 4; i++) {
1038 if ((i & deny) != i) {
1044 /* Recalculate per-file deny mode if there was a change */
1046 recalculate_deny_mode(stp->st_stid.sc_file);
1049 /* release all access and file references for a given stateid */
1051 release_all_access(struct nfs4_ol_stateid *stp)
1054 struct nfs4_file *fp = stp->st_stid.sc_file;
1056 if (fp && stp->st_deny_bmap != 0)
1057 recalculate_deny_mode(fp);
1059 for (i = 1; i < 4; i++) {
1060 if (test_access(i, stp))
1061 nfs4_file_put_access(stp->st_stid.sc_file, i);
1062 clear_access(i, stp);
1066 static inline void nfs4_free_stateowner(struct nfs4_stateowner *sop)
1068 kfree(sop->so_owner.data);
1069 sop->so_ops->so_free(sop);
1072 static void nfs4_put_stateowner(struct nfs4_stateowner *sop)
1074 struct nfs4_client *clp = sop->so_client;
1076 might_lock(&clp->cl_lock);
1078 if (!atomic_dec_and_lock(&sop->so_count, &clp->cl_lock))
1080 sop->so_ops->so_unhash(sop);
1081 spin_unlock(&clp->cl_lock);
1082 nfs4_free_stateowner(sop);
1085 static bool unhash_ol_stateid(struct nfs4_ol_stateid *stp)
1087 struct nfs4_file *fp = stp->st_stid.sc_file;
1089 lockdep_assert_held(&stp->st_stateowner->so_client->cl_lock);
1091 if (list_empty(&stp->st_perfile))
1094 spin_lock(&fp->fi_lock);
1095 list_del_init(&stp->st_perfile);
1096 spin_unlock(&fp->fi_lock);
1097 list_del(&stp->st_perstateowner);
1101 static void nfs4_free_ol_stateid(struct nfs4_stid *stid)
1103 struct nfs4_ol_stateid *stp = openlockstateid(stid);
1105 put_clnt_odstate(stp->st_clnt_odstate);
1106 release_all_access(stp);
1107 if (stp->st_stateowner)
1108 nfs4_put_stateowner(stp->st_stateowner);
1109 kmem_cache_free(stateid_slab, stid);
1112 static void nfs4_free_lock_stateid(struct nfs4_stid *stid)
1114 struct nfs4_ol_stateid *stp = openlockstateid(stid);
1115 struct nfs4_lockowner *lo = lockowner(stp->st_stateowner);
1118 file = find_any_file(stp->st_stid.sc_file);
1120 filp_close(file, (fl_owner_t)lo);
1121 nfs4_free_ol_stateid(stid);
1125 * Put the persistent reference to an already unhashed generic stateid, while
1126 * holding the cl_lock. If it's the last reference, then put it onto the
1127 * reaplist for later destruction.
1129 static void put_ol_stateid_locked(struct nfs4_ol_stateid *stp,
1130 struct list_head *reaplist)
1132 struct nfs4_stid *s = &stp->st_stid;
1133 struct nfs4_client *clp = s->sc_client;
1135 lockdep_assert_held(&clp->cl_lock);
1137 WARN_ON_ONCE(!list_empty(&stp->st_locks));
1139 if (!atomic_dec_and_test(&s->sc_count)) {
1140 wake_up_all(&close_wq);
1144 idr_remove(&clp->cl_stateids, s->sc_stateid.si_opaque.so_id);
1145 list_add(&stp->st_locks, reaplist);
1148 static bool unhash_lock_stateid(struct nfs4_ol_stateid *stp)
1150 struct nfs4_openowner *oo = openowner(stp->st_openstp->st_stateowner);
1152 lockdep_assert_held(&oo->oo_owner.so_client->cl_lock);
1154 list_del_init(&stp->st_locks);
1155 nfs4_unhash_stid(&stp->st_stid);
1156 return unhash_ol_stateid(stp);
1159 static void release_lock_stateid(struct nfs4_ol_stateid *stp)
1161 struct nfs4_openowner *oo = openowner(stp->st_openstp->st_stateowner);
1164 spin_lock(&oo->oo_owner.so_client->cl_lock);
1165 unhashed = unhash_lock_stateid(stp);
1166 spin_unlock(&oo->oo_owner.so_client->cl_lock);
1168 nfs4_put_stid(&stp->st_stid);
1171 static void unhash_lockowner_locked(struct nfs4_lockowner *lo)
1173 struct nfs4_client *clp = lo->lo_owner.so_client;
1175 lockdep_assert_held(&clp->cl_lock);
1177 list_del_init(&lo->lo_owner.so_strhash);
1181 * Free a list of generic stateids that were collected earlier after being
1185 free_ol_stateid_reaplist(struct list_head *reaplist)
1187 struct nfs4_ol_stateid *stp;
1188 struct nfs4_file *fp;
1192 while (!list_empty(reaplist)) {
1193 stp = list_first_entry(reaplist, struct nfs4_ol_stateid,
1195 list_del(&stp->st_locks);
1196 fp = stp->st_stid.sc_file;
1197 stp->st_stid.sc_free(&stp->st_stid);
1203 static void release_lockowner(struct nfs4_lockowner *lo)
1205 struct nfs4_client *clp = lo->lo_owner.so_client;
1206 struct nfs4_ol_stateid *stp;
1207 struct list_head reaplist;
1209 INIT_LIST_HEAD(&reaplist);
1211 spin_lock(&clp->cl_lock);
1212 unhash_lockowner_locked(lo);
1213 while (!list_empty(&lo->lo_owner.so_stateids)) {
1214 stp = list_first_entry(&lo->lo_owner.so_stateids,
1215 struct nfs4_ol_stateid, st_perstateowner);
1216 WARN_ON(!unhash_lock_stateid(stp));
1217 put_ol_stateid_locked(stp, &reaplist);
1219 spin_unlock(&clp->cl_lock);
1220 free_ol_stateid_reaplist(&reaplist);
1221 nfs4_put_stateowner(&lo->lo_owner);
1224 static void release_open_stateid_locks(struct nfs4_ol_stateid *open_stp,
1225 struct list_head *reaplist)
1227 struct nfs4_ol_stateid *stp;
1229 lockdep_assert_held(&open_stp->st_stid.sc_client->cl_lock);
1231 while (!list_empty(&open_stp->st_locks)) {
1232 stp = list_entry(open_stp->st_locks.next,
1233 struct nfs4_ol_stateid, st_locks);
1234 WARN_ON(!unhash_lock_stateid(stp));
1235 put_ol_stateid_locked(stp, reaplist);
1239 static bool unhash_open_stateid(struct nfs4_ol_stateid *stp,
1240 struct list_head *reaplist)
1244 lockdep_assert_held(&stp->st_stid.sc_client->cl_lock);
1246 unhashed = unhash_ol_stateid(stp);
1247 release_open_stateid_locks(stp, reaplist);
1251 static void release_open_stateid(struct nfs4_ol_stateid *stp)
1253 LIST_HEAD(reaplist);
1255 spin_lock(&stp->st_stid.sc_client->cl_lock);
1256 if (unhash_open_stateid(stp, &reaplist))
1257 put_ol_stateid_locked(stp, &reaplist);
1258 spin_unlock(&stp->st_stid.sc_client->cl_lock);
1259 free_ol_stateid_reaplist(&reaplist);
1262 static void unhash_openowner_locked(struct nfs4_openowner *oo)
1264 struct nfs4_client *clp = oo->oo_owner.so_client;
1266 lockdep_assert_held(&clp->cl_lock);
1268 list_del_init(&oo->oo_owner.so_strhash);
1269 list_del_init(&oo->oo_perclient);
1272 static void release_last_closed_stateid(struct nfs4_openowner *oo)
1274 struct nfsd_net *nn = net_generic(oo->oo_owner.so_client->net,
1276 struct nfs4_ol_stateid *s;
1278 spin_lock(&nn->client_lock);
1279 s = oo->oo_last_closed_stid;
1281 list_del_init(&oo->oo_close_lru);
1282 oo->oo_last_closed_stid = NULL;
1284 spin_unlock(&nn->client_lock);
1286 nfs4_put_stid(&s->st_stid);
1289 static void release_openowner(struct nfs4_openowner *oo)
1291 struct nfs4_ol_stateid *stp;
1292 struct nfs4_client *clp = oo->oo_owner.so_client;
1293 struct list_head reaplist;
1295 INIT_LIST_HEAD(&reaplist);
1297 spin_lock(&clp->cl_lock);
1298 unhash_openowner_locked(oo);
1299 while (!list_empty(&oo->oo_owner.so_stateids)) {
1300 stp = list_first_entry(&oo->oo_owner.so_stateids,
1301 struct nfs4_ol_stateid, st_perstateowner);
1302 if (unhash_open_stateid(stp, &reaplist))
1303 put_ol_stateid_locked(stp, &reaplist);
1305 spin_unlock(&clp->cl_lock);
1306 free_ol_stateid_reaplist(&reaplist);
1307 release_last_closed_stateid(oo);
1308 nfs4_put_stateowner(&oo->oo_owner);
1312 hash_sessionid(struct nfs4_sessionid *sessionid)
1314 struct nfsd4_sessionid *sid = (struct nfsd4_sessionid *)sessionid;
1316 return sid->sequence % SESSION_HASH_SIZE;
1319 #ifdef CONFIG_SUNRPC_DEBUG
1321 dump_sessionid(const char *fn, struct nfs4_sessionid *sessionid)
1323 u32 *ptr = (u32 *)(&sessionid->data[0]);
1324 dprintk("%s: %u:%u:%u:%u\n", fn, ptr[0], ptr[1], ptr[2], ptr[3]);
1328 dump_sessionid(const char *fn, struct nfs4_sessionid *sessionid)
1334 * Bump the seqid on cstate->replay_owner, and clear replay_owner if it
1335 * won't be used for replay.
1337 void nfsd4_bump_seqid(struct nfsd4_compound_state *cstate, __be32 nfserr)
1339 struct nfs4_stateowner *so = cstate->replay_owner;
1341 if (nfserr == nfserr_replay_me)
1344 if (!seqid_mutating_err(ntohl(nfserr))) {
1345 nfsd4_cstate_clear_replay(cstate);
1350 if (so->so_is_open_owner)
1351 release_last_closed_stateid(openowner(so));
1357 gen_sessionid(struct nfsd4_session *ses)
1359 struct nfs4_client *clp = ses->se_client;
1360 struct nfsd4_sessionid *sid;
1362 sid = (struct nfsd4_sessionid *)ses->se_sessionid.data;
1363 sid->clientid = clp->cl_clientid;
1364 sid->sequence = current_sessionid++;
1369 * The protocol defines ca_maxresponssize_cached to include the size of
1370 * the rpc header, but all we need to cache is the data starting after
1371 * the end of the initial SEQUENCE operation--the rest we regenerate
1372 * each time. Therefore we can advertise a ca_maxresponssize_cached
1373 * value that is the number of bytes in our cache plus a few additional
1374 * bytes. In order to stay on the safe side, and not promise more than
1375 * we can cache, those additional bytes must be the minimum possible: 24
1376 * bytes of rpc header (xid through accept state, with AUTH_NULL
1377 * verifier), 12 for the compound header (with zero-length tag), and 44
1378 * for the SEQUENCE op response:
1380 #define NFSD_MIN_HDR_SEQ_SZ (24 + 12 + 44)
1383 free_session_slots(struct nfsd4_session *ses)
1387 for (i = 0; i < ses->se_fchannel.maxreqs; i++)
1388 kfree(ses->se_slots[i]);
1392 * We don't actually need to cache the rpc and session headers, so we
1393 * can allocate a little less for each slot:
1395 static inline u32 slot_bytes(struct nfsd4_channel_attrs *ca)
1399 if (ca->maxresp_cached < NFSD_MIN_HDR_SEQ_SZ)
1402 size = ca->maxresp_cached - NFSD_MIN_HDR_SEQ_SZ;
1403 return size + sizeof(struct nfsd4_slot);
1407 * XXX: If we run out of reserved DRC memory we could (up to a point)
1408 * re-negotiate active sessions and reduce their slot usage to make
1409 * room for new connections. For now we just fail the create session.
1411 static u32 nfsd4_get_drc_mem(struct nfsd4_channel_attrs *ca)
1413 u32 slotsize = slot_bytes(ca);
1414 u32 num = ca->maxreqs;
1417 spin_lock(&nfsd_drc_lock);
1418 avail = min((unsigned long)NFSD_MAX_MEM_PER_SESSION,
1419 nfsd_drc_max_mem - nfsd_drc_mem_used);
1420 num = min_t(int, num, avail / slotsize);
1421 nfsd_drc_mem_used += num * slotsize;
1422 spin_unlock(&nfsd_drc_lock);
1427 static void nfsd4_put_drc_mem(struct nfsd4_channel_attrs *ca)
1429 int slotsize = slot_bytes(ca);
1431 spin_lock(&nfsd_drc_lock);
1432 nfsd_drc_mem_used -= slotsize * ca->maxreqs;
1433 spin_unlock(&nfsd_drc_lock);
1436 static struct nfsd4_session *alloc_session(struct nfsd4_channel_attrs *fattrs,
1437 struct nfsd4_channel_attrs *battrs)
1439 int numslots = fattrs->maxreqs;
1440 int slotsize = slot_bytes(fattrs);
1441 struct nfsd4_session *new;
1444 BUILD_BUG_ON(NFSD_MAX_SLOTS_PER_SESSION * sizeof(struct nfsd4_slot *)
1445 + sizeof(struct nfsd4_session) > PAGE_SIZE);
1446 mem = numslots * sizeof(struct nfsd4_slot *);
1448 new = kzalloc(sizeof(*new) + mem, GFP_KERNEL);
1451 /* allocate each struct nfsd4_slot and data cache in one piece */
1452 for (i = 0; i < numslots; i++) {
1453 new->se_slots[i] = kzalloc(slotsize, GFP_KERNEL);
1454 if (!new->se_slots[i])
1458 memcpy(&new->se_fchannel, fattrs, sizeof(struct nfsd4_channel_attrs));
1459 memcpy(&new->se_bchannel, battrs, sizeof(struct nfsd4_channel_attrs));
1464 kfree(new->se_slots[i]);
1469 static void free_conn(struct nfsd4_conn *c)
1471 svc_xprt_put(c->cn_xprt);
1475 static void nfsd4_conn_lost(struct svc_xpt_user *u)
1477 struct nfsd4_conn *c = container_of(u, struct nfsd4_conn, cn_xpt_user);
1478 struct nfs4_client *clp = c->cn_session->se_client;
1480 spin_lock(&clp->cl_lock);
1481 if (!list_empty(&c->cn_persession)) {
1482 list_del(&c->cn_persession);
1485 nfsd4_probe_callback(clp);
1486 spin_unlock(&clp->cl_lock);
1489 static struct nfsd4_conn *alloc_conn(struct svc_rqst *rqstp, u32 flags)
1491 struct nfsd4_conn *conn;
1493 conn = kmalloc(sizeof(struct nfsd4_conn), GFP_KERNEL);
1496 svc_xprt_get(rqstp->rq_xprt);
1497 conn->cn_xprt = rqstp->rq_xprt;
1498 conn->cn_flags = flags;
1499 INIT_LIST_HEAD(&conn->cn_xpt_user.list);
1503 static void __nfsd4_hash_conn(struct nfsd4_conn *conn, struct nfsd4_session *ses)
1505 conn->cn_session = ses;
1506 list_add(&conn->cn_persession, &ses->se_conns);
1509 static void nfsd4_hash_conn(struct nfsd4_conn *conn, struct nfsd4_session *ses)
1511 struct nfs4_client *clp = ses->se_client;
1513 spin_lock(&clp->cl_lock);
1514 __nfsd4_hash_conn(conn, ses);
1515 spin_unlock(&clp->cl_lock);
1518 static int nfsd4_register_conn(struct nfsd4_conn *conn)
1520 conn->cn_xpt_user.callback = nfsd4_conn_lost;
1521 return register_xpt_user(conn->cn_xprt, &conn->cn_xpt_user);
1524 static void nfsd4_init_conn(struct svc_rqst *rqstp, struct nfsd4_conn *conn, struct nfsd4_session *ses)
1528 nfsd4_hash_conn(conn, ses);
1529 ret = nfsd4_register_conn(conn);
1531 /* oops; xprt is already down: */
1532 nfsd4_conn_lost(&conn->cn_xpt_user);
1533 /* We may have gained or lost a callback channel: */
1534 nfsd4_probe_callback_sync(ses->se_client);
1537 static struct nfsd4_conn *alloc_conn_from_crses(struct svc_rqst *rqstp, struct nfsd4_create_session *cses)
1539 u32 dir = NFS4_CDFC4_FORE;
1541 if (cses->flags & SESSION4_BACK_CHAN)
1542 dir |= NFS4_CDFC4_BACK;
1543 return alloc_conn(rqstp, dir);
1546 /* must be called under client_lock */
1547 static void nfsd4_del_conns(struct nfsd4_session *s)
1549 struct nfs4_client *clp = s->se_client;
1550 struct nfsd4_conn *c;
1552 spin_lock(&clp->cl_lock);
1553 while (!list_empty(&s->se_conns)) {
1554 c = list_first_entry(&s->se_conns, struct nfsd4_conn, cn_persession);
1555 list_del_init(&c->cn_persession);
1556 spin_unlock(&clp->cl_lock);
1558 unregister_xpt_user(c->cn_xprt, &c->cn_xpt_user);
1561 spin_lock(&clp->cl_lock);
1563 spin_unlock(&clp->cl_lock);
1566 static void __free_session(struct nfsd4_session *ses)
1568 free_session_slots(ses);
1572 static void free_session(struct nfsd4_session *ses)
1574 nfsd4_del_conns(ses);
1575 nfsd4_put_drc_mem(&ses->se_fchannel);
1576 __free_session(ses);
1579 static void init_session(struct svc_rqst *rqstp, struct nfsd4_session *new, struct nfs4_client *clp, struct nfsd4_create_session *cses)
1582 struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
1584 new->se_client = clp;
1587 INIT_LIST_HEAD(&new->se_conns);
1589 new->se_cb_seq_nr = 1;
1590 new->se_flags = cses->flags;
1591 new->se_cb_prog = cses->callback_prog;
1592 new->se_cb_sec = cses->cb_sec;
1593 atomic_set(&new->se_ref, 0);
1594 idx = hash_sessionid(&new->se_sessionid);
1595 list_add(&new->se_hash, &nn->sessionid_hashtbl[idx]);
1596 spin_lock(&clp->cl_lock);
1597 list_add(&new->se_perclnt, &clp->cl_sessions);
1598 spin_unlock(&clp->cl_lock);
1601 struct sockaddr *sa = svc_addr(rqstp);
1603 * This is a little silly; with sessions there's no real
1604 * use for the callback address. Use the peer address
1605 * as a reasonable default for now, but consider fixing
1606 * the rpc client not to require an address in the
1609 rpc_copy_addr((struct sockaddr *)&clp->cl_cb_conn.cb_addr, sa);
1610 clp->cl_cb_conn.cb_addrlen = svc_addr_len(sa);
1614 /* caller must hold client_lock */
1615 static struct nfsd4_session *
1616 __find_in_sessionid_hashtbl(struct nfs4_sessionid *sessionid, struct net *net)
1618 struct nfsd4_session *elem;
1620 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
1622 lockdep_assert_held(&nn->client_lock);
1624 dump_sessionid(__func__, sessionid);
1625 idx = hash_sessionid(sessionid);
1626 /* Search in the appropriate list */
1627 list_for_each_entry(elem, &nn->sessionid_hashtbl[idx], se_hash) {
1628 if (!memcmp(elem->se_sessionid.data, sessionid->data,
1629 NFS4_MAX_SESSIONID_LEN)) {
1634 dprintk("%s: session not found\n", __func__);
1638 static struct nfsd4_session *
1639 find_in_sessionid_hashtbl(struct nfs4_sessionid *sessionid, struct net *net,
1642 struct nfsd4_session *session;
1643 __be32 status = nfserr_badsession;
1645 session = __find_in_sessionid_hashtbl(sessionid, net);
1648 status = nfsd4_get_session_locked(session);
1656 /* caller must hold client_lock */
1658 unhash_session(struct nfsd4_session *ses)
1660 struct nfs4_client *clp = ses->se_client;
1661 struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
1663 lockdep_assert_held(&nn->client_lock);
1665 list_del(&ses->se_hash);
1666 spin_lock(&ses->se_client->cl_lock);
1667 list_del(&ses->se_perclnt);
1668 spin_unlock(&ses->se_client->cl_lock);
1671 /* SETCLIENTID and SETCLIENTID_CONFIRM Helper functions */
1673 STALE_CLIENTID(clientid_t *clid, struct nfsd_net *nn)
1676 * We're assuming the clid was not given out from a boot
1677 * precisely 2^32 (about 136 years) before this one. That seems
1678 * a safe assumption:
1680 if (clid->cl_boot == (u32)nn->boot_time)
1682 dprintk("NFSD stale clientid (%08x/%08x) boot_time %08lx\n",
1683 clid->cl_boot, clid->cl_id, nn->boot_time);
1688 * XXX Should we use a slab cache ?
1689 * This type of memory management is somewhat inefficient, but we use it
1690 * anyway since SETCLIENTID is not a common operation.
1692 static struct nfs4_client *alloc_client(struct xdr_netobj name)
1694 struct nfs4_client *clp;
1697 clp = kzalloc(sizeof(struct nfs4_client), GFP_KERNEL);
1700 clp->cl_name.data = kmemdup(name.data, name.len, GFP_KERNEL);
1701 if (clp->cl_name.data == NULL)
1703 clp->cl_ownerstr_hashtbl = kmalloc(sizeof(struct list_head) *
1704 OWNER_HASH_SIZE, GFP_KERNEL);
1705 if (!clp->cl_ownerstr_hashtbl)
1706 goto err_no_hashtbl;
1707 for (i = 0; i < OWNER_HASH_SIZE; i++)
1708 INIT_LIST_HEAD(&clp->cl_ownerstr_hashtbl[i]);
1709 clp->cl_name.len = name.len;
1710 INIT_LIST_HEAD(&clp->cl_sessions);
1711 idr_init(&clp->cl_stateids);
1712 atomic_set(&clp->cl_refcount, 0);
1713 clp->cl_cb_state = NFSD4_CB_UNKNOWN;
1714 INIT_LIST_HEAD(&clp->cl_idhash);
1715 INIT_LIST_HEAD(&clp->cl_openowners);
1716 INIT_LIST_HEAD(&clp->cl_delegations);
1717 INIT_LIST_HEAD(&clp->cl_lru);
1718 INIT_LIST_HEAD(&clp->cl_revoked);
1719 #ifdef CONFIG_NFSD_PNFS
1720 INIT_LIST_HEAD(&clp->cl_lo_states);
1722 spin_lock_init(&clp->cl_lock);
1723 rpc_init_wait_queue(&clp->cl_cb_waitq, "Backchannel slot table");
1726 kfree(clp->cl_name.data);
1733 free_client(struct nfs4_client *clp)
1735 while (!list_empty(&clp->cl_sessions)) {
1736 struct nfsd4_session *ses;
1737 ses = list_entry(clp->cl_sessions.next, struct nfsd4_session,
1739 list_del(&ses->se_perclnt);
1740 WARN_ON_ONCE(atomic_read(&ses->se_ref));
1743 rpc_destroy_wait_queue(&clp->cl_cb_waitq);
1744 free_svc_cred(&clp->cl_cred);
1745 kfree(clp->cl_ownerstr_hashtbl);
1746 kfree(clp->cl_name.data);
1747 idr_destroy(&clp->cl_stateids);
1751 /* must be called under the client_lock */
1753 unhash_client_locked(struct nfs4_client *clp)
1755 struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
1756 struct nfsd4_session *ses;
1758 lockdep_assert_held(&nn->client_lock);
1760 /* Mark the client as expired! */
1762 /* Make it invisible */
1763 if (!list_empty(&clp->cl_idhash)) {
1764 list_del_init(&clp->cl_idhash);
1765 if (test_bit(NFSD4_CLIENT_CONFIRMED, &clp->cl_flags))
1766 rb_erase(&clp->cl_namenode, &nn->conf_name_tree);
1768 rb_erase(&clp->cl_namenode, &nn->unconf_name_tree);
1770 list_del_init(&clp->cl_lru);
1771 spin_lock(&clp->cl_lock);
1772 list_for_each_entry(ses, &clp->cl_sessions, se_perclnt)
1773 list_del_init(&ses->se_hash);
1774 spin_unlock(&clp->cl_lock);
1778 unhash_client(struct nfs4_client *clp)
1780 struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
1782 spin_lock(&nn->client_lock);
1783 unhash_client_locked(clp);
1784 spin_unlock(&nn->client_lock);
1787 static __be32 mark_client_expired_locked(struct nfs4_client *clp)
1789 if (atomic_read(&clp->cl_refcount))
1790 return nfserr_jukebox;
1791 unhash_client_locked(clp);
1796 __destroy_client(struct nfs4_client *clp)
1798 struct nfs4_openowner *oo;
1799 struct nfs4_delegation *dp;
1800 struct list_head reaplist;
1802 INIT_LIST_HEAD(&reaplist);
1803 spin_lock(&state_lock);
1804 while (!list_empty(&clp->cl_delegations)) {
1805 dp = list_entry(clp->cl_delegations.next, struct nfs4_delegation, dl_perclnt);
1806 WARN_ON(!unhash_delegation_locked(dp));
1807 list_add(&dp->dl_recall_lru, &reaplist);
1809 spin_unlock(&state_lock);
1810 while (!list_empty(&reaplist)) {
1811 dp = list_entry(reaplist.next, struct nfs4_delegation, dl_recall_lru);
1812 list_del_init(&dp->dl_recall_lru);
1813 put_clnt_odstate(dp->dl_clnt_odstate);
1814 nfs4_put_deleg_lease(dp->dl_stid.sc_file);
1815 nfs4_put_stid(&dp->dl_stid);
1817 while (!list_empty(&clp->cl_revoked)) {
1818 dp = list_entry(clp->cl_revoked.next, struct nfs4_delegation, dl_recall_lru);
1819 list_del_init(&dp->dl_recall_lru);
1820 nfs4_put_stid(&dp->dl_stid);
1822 while (!list_empty(&clp->cl_openowners)) {
1823 oo = list_entry(clp->cl_openowners.next, struct nfs4_openowner, oo_perclient);
1824 nfs4_get_stateowner(&oo->oo_owner);
1825 release_openowner(oo);
1827 nfsd4_return_all_client_layouts(clp);
1828 nfsd4_shutdown_callback(clp);
1829 if (clp->cl_cb_conn.cb_xprt)
1830 svc_xprt_put(clp->cl_cb_conn.cb_xprt);
1835 destroy_client(struct nfs4_client *clp)
1838 __destroy_client(clp);
1841 static void expire_client(struct nfs4_client *clp)
1844 nfsd4_client_record_remove(clp);
1845 __destroy_client(clp);
1848 static void copy_verf(struct nfs4_client *target, nfs4_verifier *source)
1850 memcpy(target->cl_verifier.data, source->data,
1851 sizeof(target->cl_verifier.data));
1854 static void copy_clid(struct nfs4_client *target, struct nfs4_client *source)
1856 target->cl_clientid.cl_boot = source->cl_clientid.cl_boot;
1857 target->cl_clientid.cl_id = source->cl_clientid.cl_id;
1860 int strdup_if_nonnull(char **target, char *source)
1863 *target = kstrdup(source, GFP_KERNEL);
1871 static int copy_cred(struct svc_cred *target, struct svc_cred *source)
1875 ret = strdup_if_nonnull(&target->cr_principal, source->cr_principal);
1878 ret = strdup_if_nonnull(&target->cr_raw_principal,
1879 source->cr_raw_principal);
1882 target->cr_flavor = source->cr_flavor;
1883 target->cr_uid = source->cr_uid;
1884 target->cr_gid = source->cr_gid;
1885 target->cr_group_info = source->cr_group_info;
1886 get_group_info(target->cr_group_info);
1887 target->cr_gss_mech = source->cr_gss_mech;
1888 if (source->cr_gss_mech)
1889 gss_mech_get(source->cr_gss_mech);
1894 compare_blob(const struct xdr_netobj *o1, const struct xdr_netobj *o2)
1896 if (o1->len < o2->len)
1898 if (o1->len > o2->len)
1900 return memcmp(o1->data, o2->data, o1->len);
1903 static int same_name(const char *n1, const char *n2)
1905 return 0 == memcmp(n1, n2, HEXDIR_LEN);
1909 same_verf(nfs4_verifier *v1, nfs4_verifier *v2)
1911 return 0 == memcmp(v1->data, v2->data, sizeof(v1->data));
1915 same_clid(clientid_t *cl1, clientid_t *cl2)
1917 return (cl1->cl_boot == cl2->cl_boot) && (cl1->cl_id == cl2->cl_id);
1920 static bool groups_equal(struct group_info *g1, struct group_info *g2)
1924 if (g1->ngroups != g2->ngroups)
1926 for (i=0; i<g1->ngroups; i++)
1927 if (!gid_eq(GROUP_AT(g1, i), GROUP_AT(g2, i)))
1933 * RFC 3530 language requires clid_inuse be returned when the
1934 * "principal" associated with a requests differs from that previously
1935 * used. We use uid, gid's, and gss principal string as our best
1936 * approximation. We also don't want to allow non-gss use of a client
1937 * established using gss: in theory cr_principal should catch that
1938 * change, but in practice cr_principal can be null even in the gss case
1939 * since gssd doesn't always pass down a principal string.
1941 static bool is_gss_cred(struct svc_cred *cr)
1943 /* Is cr_flavor one of the gss "pseudoflavors"?: */
1944 return (cr->cr_flavor > RPC_AUTH_MAXFLAVOR);
1949 same_creds(struct svc_cred *cr1, struct svc_cred *cr2)
1951 if ((is_gss_cred(cr1) != is_gss_cred(cr2))
1952 || (!uid_eq(cr1->cr_uid, cr2->cr_uid))
1953 || (!gid_eq(cr1->cr_gid, cr2->cr_gid))
1954 || !groups_equal(cr1->cr_group_info, cr2->cr_group_info))
1956 if (cr1->cr_principal == cr2->cr_principal)
1958 if (!cr1->cr_principal || !cr2->cr_principal)
1960 return 0 == strcmp(cr1->cr_principal, cr2->cr_principal);
1963 static bool svc_rqst_integrity_protected(struct svc_rqst *rqstp)
1965 struct svc_cred *cr = &rqstp->rq_cred;
1968 if (!cr->cr_gss_mech)
1970 service = gss_pseudoflavor_to_service(cr->cr_gss_mech, cr->cr_flavor);
1971 return service == RPC_GSS_SVC_INTEGRITY ||
1972 service == RPC_GSS_SVC_PRIVACY;
1975 static bool mach_creds_match(struct nfs4_client *cl, struct svc_rqst *rqstp)
1977 struct svc_cred *cr = &rqstp->rq_cred;
1979 if (!cl->cl_mach_cred)
1981 if (cl->cl_cred.cr_gss_mech != cr->cr_gss_mech)
1983 if (!svc_rqst_integrity_protected(rqstp))
1985 if (cl->cl_cred.cr_raw_principal)
1986 return 0 == strcmp(cl->cl_cred.cr_raw_principal,
1987 cr->cr_raw_principal);
1988 if (!cr->cr_principal)
1990 return 0 == strcmp(cl->cl_cred.cr_principal, cr->cr_principal);
1993 static void gen_confirm(struct nfs4_client *clp, struct nfsd_net *nn)
1998 * This is opaque to client, so no need to byte-swap. Use
1999 * __force to keep sparse happy
2001 verf[0] = (__force __be32)get_seconds();
2002 verf[1] = (__force __be32)nn->clverifier_counter++;
2003 memcpy(clp->cl_confirm.data, verf, sizeof(clp->cl_confirm.data));
2006 static void gen_clid(struct nfs4_client *clp, struct nfsd_net *nn)
2008 clp->cl_clientid.cl_boot = nn->boot_time;
2009 clp->cl_clientid.cl_id = nn->clientid_counter++;
2010 gen_confirm(clp, nn);
2013 static struct nfs4_stid *
2014 find_stateid_locked(struct nfs4_client *cl, stateid_t *t)
2016 struct nfs4_stid *ret;
2018 ret = idr_find(&cl->cl_stateids, t->si_opaque.so_id);
2019 if (!ret || !ret->sc_type)
2024 static struct nfs4_stid *
2025 find_stateid_by_type(struct nfs4_client *cl, stateid_t *t, char typemask)
2027 struct nfs4_stid *s;
2029 spin_lock(&cl->cl_lock);
2030 s = find_stateid_locked(cl, t);
2032 if (typemask & s->sc_type)
2033 atomic_inc(&s->sc_count);
2037 spin_unlock(&cl->cl_lock);
2041 static struct nfs4_client *create_client(struct xdr_netobj name,
2042 struct svc_rqst *rqstp, nfs4_verifier *verf)
2044 struct nfs4_client *clp;
2045 struct sockaddr *sa = svc_addr(rqstp);
2047 struct net *net = SVC_NET(rqstp);
2049 clp = alloc_client(name);
2053 ret = copy_cred(&clp->cl_cred, &rqstp->rq_cred);
2058 nfsd4_init_cb(&clp->cl_cb_null, clp, NULL, NFSPROC4_CLNT_CB_NULL);
2059 clp->cl_time = get_seconds();
2060 clear_bit(0, &clp->cl_cb_slot_busy);
2061 copy_verf(clp, verf);
2062 rpc_copy_addr((struct sockaddr *) &clp->cl_addr, sa);
2063 clp->cl_cb_session = NULL;
2069 add_clp_to_name_tree(struct nfs4_client *new_clp, struct rb_root *root)
2071 struct rb_node **new = &(root->rb_node), *parent = NULL;
2072 struct nfs4_client *clp;
2075 clp = rb_entry(*new, struct nfs4_client, cl_namenode);
2078 if (compare_blob(&clp->cl_name, &new_clp->cl_name) > 0)
2079 new = &((*new)->rb_left);
2081 new = &((*new)->rb_right);
2084 rb_link_node(&new_clp->cl_namenode, parent, new);
2085 rb_insert_color(&new_clp->cl_namenode, root);
2088 static struct nfs4_client *
2089 find_clp_in_name_tree(struct xdr_netobj *name, struct rb_root *root)
2092 struct rb_node *node = root->rb_node;
2093 struct nfs4_client *clp;
2096 clp = rb_entry(node, struct nfs4_client, cl_namenode);
2097 cmp = compare_blob(&clp->cl_name, name);
2099 node = node->rb_left;
2101 node = node->rb_right;
2109 add_to_unconfirmed(struct nfs4_client *clp)
2111 unsigned int idhashval;
2112 struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
2114 lockdep_assert_held(&nn->client_lock);
2116 clear_bit(NFSD4_CLIENT_CONFIRMED, &clp->cl_flags);
2117 add_clp_to_name_tree(clp, &nn->unconf_name_tree);
2118 idhashval = clientid_hashval(clp->cl_clientid.cl_id);
2119 list_add(&clp->cl_idhash, &nn->unconf_id_hashtbl[idhashval]);
2120 renew_client_locked(clp);
2124 move_to_confirmed(struct nfs4_client *clp)
2126 unsigned int idhashval = clientid_hashval(clp->cl_clientid.cl_id);
2127 struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
2129 lockdep_assert_held(&nn->client_lock);
2131 dprintk("NFSD: move_to_confirm nfs4_client %p\n", clp);
2132 list_move(&clp->cl_idhash, &nn->conf_id_hashtbl[idhashval]);
2133 rb_erase(&clp->cl_namenode, &nn->unconf_name_tree);
2134 add_clp_to_name_tree(clp, &nn->conf_name_tree);
2135 set_bit(NFSD4_CLIENT_CONFIRMED, &clp->cl_flags);
2136 renew_client_locked(clp);
2139 static struct nfs4_client *
2140 find_client_in_id_table(struct list_head *tbl, clientid_t *clid, bool sessions)
2142 struct nfs4_client *clp;
2143 unsigned int idhashval = clientid_hashval(clid->cl_id);
2145 list_for_each_entry(clp, &tbl[idhashval], cl_idhash) {
2146 if (same_clid(&clp->cl_clientid, clid)) {
2147 if ((bool)clp->cl_minorversion != sessions)
2149 renew_client_locked(clp);
2156 static struct nfs4_client *
2157 find_confirmed_client(clientid_t *clid, bool sessions, struct nfsd_net *nn)
2159 struct list_head *tbl = nn->conf_id_hashtbl;
2161 lockdep_assert_held(&nn->client_lock);
2162 return find_client_in_id_table(tbl, clid, sessions);
2165 static struct nfs4_client *
2166 find_unconfirmed_client(clientid_t *clid, bool sessions, struct nfsd_net *nn)
2168 struct list_head *tbl = nn->unconf_id_hashtbl;
2170 lockdep_assert_held(&nn->client_lock);
2171 return find_client_in_id_table(tbl, clid, sessions);
2174 static bool clp_used_exchangeid(struct nfs4_client *clp)
2176 return clp->cl_exchange_flags != 0;
2179 static struct nfs4_client *
2180 find_confirmed_client_by_name(struct xdr_netobj *name, struct nfsd_net *nn)
2182 lockdep_assert_held(&nn->client_lock);
2183 return find_clp_in_name_tree(name, &nn->conf_name_tree);
2186 static struct nfs4_client *
2187 find_unconfirmed_client_by_name(struct xdr_netobj *name, struct nfsd_net *nn)
2189 lockdep_assert_held(&nn->client_lock);
2190 return find_clp_in_name_tree(name, &nn->unconf_name_tree);
2194 gen_callback(struct nfs4_client *clp, struct nfsd4_setclientid *se, struct svc_rqst *rqstp)
2196 struct nfs4_cb_conn *conn = &clp->cl_cb_conn;
2197 struct sockaddr *sa = svc_addr(rqstp);
2198 u32 scopeid = rpc_get_scope_id(sa);
2199 unsigned short expected_family;
2201 /* Currently, we only support tcp and tcp6 for the callback channel */
2202 if (se->se_callback_netid_len == 3 &&
2203 !memcmp(se->se_callback_netid_val, "tcp", 3))
2204 expected_family = AF_INET;
2205 else if (se->se_callback_netid_len == 4 &&
2206 !memcmp(se->se_callback_netid_val, "tcp6", 4))
2207 expected_family = AF_INET6;
2211 conn->cb_addrlen = rpc_uaddr2sockaddr(clp->net, se->se_callback_addr_val,
2212 se->se_callback_addr_len,
2213 (struct sockaddr *)&conn->cb_addr,
2214 sizeof(conn->cb_addr));
2216 if (!conn->cb_addrlen || conn->cb_addr.ss_family != expected_family)
2219 if (conn->cb_addr.ss_family == AF_INET6)
2220 ((struct sockaddr_in6 *)&conn->cb_addr)->sin6_scope_id = scopeid;
2222 conn->cb_prog = se->se_callback_prog;
2223 conn->cb_ident = se->se_callback_ident;
2224 memcpy(&conn->cb_saddr, &rqstp->rq_daddr, rqstp->rq_daddrlen);
2227 conn->cb_addr.ss_family = AF_UNSPEC;
2228 conn->cb_addrlen = 0;
2229 dprintk(KERN_INFO "NFSD: this client (clientid %08x/%08x) "
2230 "will not receive delegations\n",
2231 clp->cl_clientid.cl_boot, clp->cl_clientid.cl_id);
2237 * Cache a reply. nfsd4_check_resp_size() has bounded the cache size.
2240 nfsd4_store_cache_entry(struct nfsd4_compoundres *resp)
2242 struct xdr_buf *buf = resp->xdr.buf;
2243 struct nfsd4_slot *slot = resp->cstate.slot;
2246 dprintk("--> %s slot %p\n", __func__, slot);
2248 slot->sl_opcnt = resp->opcnt;
2249 slot->sl_status = resp->cstate.status;
2251 slot->sl_flags |= NFSD4_SLOT_INITIALIZED;
2252 if (nfsd4_not_cached(resp)) {
2253 slot->sl_datalen = 0;
2256 base = resp->cstate.data_offset;
2257 slot->sl_datalen = buf->len - base;
2258 if (read_bytes_from_xdr_buf(buf, base, slot->sl_data, slot->sl_datalen))
2259 WARN(1, "%s: sessions DRC could not cache compound\n",
2265 * Encode the replay sequence operation from the slot values.
2266 * If cachethis is FALSE encode the uncached rep error on the next
2267 * operation which sets resp->p and increments resp->opcnt for
2268 * nfs4svc_encode_compoundres.
2272 nfsd4_enc_sequence_replay(struct nfsd4_compoundargs *args,
2273 struct nfsd4_compoundres *resp)
2275 struct nfsd4_op *op;
2276 struct nfsd4_slot *slot = resp->cstate.slot;
2278 /* Encode the replayed sequence operation */
2279 op = &args->ops[resp->opcnt - 1];
2280 nfsd4_encode_operation(resp, op);
2282 /* Return nfserr_retry_uncached_rep in next operation. */
2283 if (args->opcnt > 1 && !(slot->sl_flags & NFSD4_SLOT_CACHETHIS)) {
2284 op = &args->ops[resp->opcnt++];
2285 op->status = nfserr_retry_uncached_rep;
2286 nfsd4_encode_operation(resp, op);
2292 * The sequence operation is not cached because we can use the slot and
2296 nfsd4_replay_cache_entry(struct nfsd4_compoundres *resp,
2297 struct nfsd4_sequence *seq)
2299 struct nfsd4_slot *slot = resp->cstate.slot;
2300 struct xdr_stream *xdr = &resp->xdr;
2304 dprintk("--> %s slot %p\n", __func__, slot);
2306 status = nfsd4_enc_sequence_replay(resp->rqstp->rq_argp, resp);
2310 p = xdr_reserve_space(xdr, slot->sl_datalen);
2313 return nfserr_serverfault;
2315 xdr_encode_opaque_fixed(p, slot->sl_data, slot->sl_datalen);
2316 xdr_commit_encode(xdr);
2318 resp->opcnt = slot->sl_opcnt;
2319 return slot->sl_status;
2323 * Set the exchange_id flags returned by the server.
2326 nfsd4_set_ex_flags(struct nfs4_client *new, struct nfsd4_exchange_id *clid)
2328 #ifdef CONFIG_NFSD_PNFS
2329 new->cl_exchange_flags |= EXCHGID4_FLAG_USE_PNFS_MDS;
2331 new->cl_exchange_flags |= EXCHGID4_FLAG_USE_NON_PNFS;
2334 /* Referrals are supported, Migration is not. */
2335 new->cl_exchange_flags |= EXCHGID4_FLAG_SUPP_MOVED_REFER;
2337 /* set the wire flags to return to client. */
2338 clid->flags = new->cl_exchange_flags;
2341 static bool client_has_openowners(struct nfs4_client *clp)
2343 struct nfs4_openowner *oo;
2345 list_for_each_entry(oo, &clp->cl_openowners, oo_perclient) {
2346 if (!list_empty(&oo->oo_owner.so_stateids))
2352 static bool client_has_state(struct nfs4_client *clp)
2354 return client_has_openowners(clp)
2355 #ifdef CONFIG_NFSD_PNFS
2356 || !list_empty(&clp->cl_lo_states)
2358 || !list_empty(&clp->cl_delegations)
2359 || !list_empty(&clp->cl_sessions);
2363 nfsd4_exchange_id(struct svc_rqst *rqstp,
2364 struct nfsd4_compound_state *cstate,
2365 struct nfsd4_exchange_id *exid)
2367 struct nfs4_client *conf, *new;
2368 struct nfs4_client *unconf = NULL;
2370 char addr_str[INET6_ADDRSTRLEN];
2371 nfs4_verifier verf = exid->verifier;
2372 struct sockaddr *sa = svc_addr(rqstp);
2373 bool update = exid->flags & EXCHGID4_FLAG_UPD_CONFIRMED_REC_A;
2374 struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
2376 rpc_ntop(sa, addr_str, sizeof(addr_str));
2377 dprintk("%s rqstp=%p exid=%p clname.len=%u clname.data=%p "
2378 "ip_addr=%s flags %x, spa_how %d\n",
2379 __func__, rqstp, exid, exid->clname.len, exid->clname.data,
2380 addr_str, exid->flags, exid->spa_how);
2382 if (exid->flags & ~EXCHGID4_FLAG_MASK_A)
2383 return nfserr_inval;
2385 new = create_client(exid->clname, rqstp, &verf);
2387 return nfserr_jukebox;
2389 switch (exid->spa_how) {
2391 if (!svc_rqst_integrity_protected(rqstp)) {
2392 status = nfserr_inval;
2396 * Sometimes userspace doesn't give us a principal.
2397 * Which is a bug, really. Anyway, we can't enforce
2398 * MACH_CRED in that case, better to give up now:
2400 if (!new->cl_cred.cr_principal &&
2401 !new->cl_cred.cr_raw_principal) {
2402 status = nfserr_serverfault;
2405 new->cl_mach_cred = true;
2408 default: /* checked by xdr code */
2411 return nfserr_encr_alg_unsupp;
2414 /* Cases below refer to rfc 5661 section 18.35.4: */
2415 spin_lock(&nn->client_lock);
2416 conf = find_confirmed_client_by_name(&exid->clname, nn);
2418 bool creds_match = same_creds(&conf->cl_cred, &rqstp->rq_cred);
2419 bool verfs_match = same_verf(&verf, &conf->cl_verifier);
2422 if (!clp_used_exchangeid(conf)) { /* buggy client */
2423 status = nfserr_inval;
2426 if (!mach_creds_match(conf, rqstp)) {
2427 status = nfserr_wrong_cred;
2430 if (!creds_match) { /* case 9 */
2431 status = nfserr_perm;
2434 if (!verfs_match) { /* case 8 */
2435 status = nfserr_not_same;
2439 exid->flags |= EXCHGID4_FLAG_CONFIRMED_R;
2442 if (!creds_match) { /* case 3 */
2443 if (client_has_state(conf)) {
2444 status = nfserr_clid_inuse;
2449 if (verfs_match) { /* case 2 */
2450 conf->cl_exchange_flags |= EXCHGID4_FLAG_CONFIRMED_R;
2453 /* case 5, client reboot */
2458 if (update) { /* case 7 */
2459 status = nfserr_noent;
2463 unconf = find_unconfirmed_client_by_name(&exid->clname, nn);
2464 if (unconf) /* case 4, possible retry or client restart */
2465 unhash_client_locked(unconf);
2467 /* case 1 (normal case) */
2470 status = mark_client_expired_locked(conf);
2474 new->cl_minorversion = cstate->minorversion;
2477 add_to_unconfirmed(new);
2480 exid->clientid.cl_boot = conf->cl_clientid.cl_boot;
2481 exid->clientid.cl_id = conf->cl_clientid.cl_id;
2483 exid->seqid = conf->cl_cs_slot.sl_seqid + 1;
2484 nfsd4_set_ex_flags(conf, exid);
2486 dprintk("nfsd4_exchange_id seqid %d flags %x\n",
2487 conf->cl_cs_slot.sl_seqid, conf->cl_exchange_flags);
2491 spin_unlock(&nn->client_lock);
2496 expire_client(unconf);
2501 check_slot_seqid(u32 seqid, u32 slot_seqid, int slot_inuse)
2503 dprintk("%s enter. seqid %d slot_seqid %d\n", __func__, seqid,
2506 /* The slot is in use, and no response has been sent. */
2508 if (seqid == slot_seqid)
2509 return nfserr_jukebox;
2511 return nfserr_seq_misordered;
2513 /* Note unsigned 32-bit arithmetic handles wraparound: */
2514 if (likely(seqid == slot_seqid + 1))
2516 if (seqid == slot_seqid)
2517 return nfserr_replay_cache;
2518 return nfserr_seq_misordered;
2522 * Cache the create session result into the create session single DRC
2523 * slot cache by saving the xdr structure. sl_seqid has been set.
2524 * Do this for solo or embedded create session operations.
2527 nfsd4_cache_create_session(struct nfsd4_create_session *cr_ses,
2528 struct nfsd4_clid_slot *slot, __be32 nfserr)
2530 slot->sl_status = nfserr;
2531 memcpy(&slot->sl_cr_ses, cr_ses, sizeof(*cr_ses));
2535 nfsd4_replay_create_session(struct nfsd4_create_session *cr_ses,
2536 struct nfsd4_clid_slot *slot)
2538 memcpy(cr_ses, &slot->sl_cr_ses, sizeof(*cr_ses));
2539 return slot->sl_status;
2542 #define NFSD_MIN_REQ_HDR_SEQ_SZ ((\
2543 2 * 2 + /* credential,verifier: AUTH_NULL, length 0 */ \
2544 1 + /* MIN tag is length with zero, only length */ \
2545 3 + /* version, opcount, opcode */ \
2546 XDR_QUADLEN(NFS4_MAX_SESSIONID_LEN) + \
2547 /* seqid, slotID, slotID, cache */ \
2548 4 ) * sizeof(__be32))
2550 #define NFSD_MIN_RESP_HDR_SEQ_SZ ((\
2551 2 + /* verifier: AUTH_NULL, length 0 */\
2553 1 + /* MIN tag is length with zero, only length */ \
2554 3 + /* opcount, opcode, opstatus*/ \
2555 XDR_QUADLEN(NFS4_MAX_SESSIONID_LEN) + \
2556 /* seqid, slotID, slotID, slotID, status */ \
2557 5 ) * sizeof(__be32))
2559 static __be32 check_forechannel_attrs(struct nfsd4_channel_attrs *ca, struct nfsd_net *nn)
2561 u32 maxrpc = nn->nfsd_serv->sv_max_mesg;
2563 if (ca->maxreq_sz < NFSD_MIN_REQ_HDR_SEQ_SZ)
2564 return nfserr_toosmall;
2565 if (ca->maxresp_sz < NFSD_MIN_RESP_HDR_SEQ_SZ)
2566 return nfserr_toosmall;
2567 ca->headerpadsz = 0;
2568 ca->maxreq_sz = min_t(u32, ca->maxreq_sz, maxrpc);
2569 ca->maxresp_sz = min_t(u32, ca->maxresp_sz, maxrpc);
2570 ca->maxops = min_t(u32, ca->maxops, NFSD_MAX_OPS_PER_COMPOUND);
2571 ca->maxresp_cached = min_t(u32, ca->maxresp_cached,
2572 NFSD_SLOT_CACHE_SIZE + NFSD_MIN_HDR_SEQ_SZ);
2573 ca->maxreqs = min_t(u32, ca->maxreqs, NFSD_MAX_SLOTS_PER_SESSION);
2575 * Note decreasing slot size below client's request may make it
2576 * difficult for client to function correctly, whereas
2577 * decreasing the number of slots will (just?) affect
2578 * performance. When short on memory we therefore prefer to
2579 * decrease number of slots instead of their size. Clients that
2580 * request larger slots than they need will get poor results:
2582 ca->maxreqs = nfsd4_get_drc_mem(ca);
2584 return nfserr_jukebox;
2589 #define NFSD_CB_MAX_REQ_SZ ((NFS4_enc_cb_recall_sz + \
2590 RPC_MAX_HEADER_WITH_AUTH) * sizeof(__be32))
2591 #define NFSD_CB_MAX_RESP_SZ ((NFS4_dec_cb_recall_sz + \
2592 RPC_MAX_REPHEADER_WITH_AUTH) * sizeof(__be32))
2594 static __be32 check_backchannel_attrs(struct nfsd4_channel_attrs *ca)
2596 ca->headerpadsz = 0;
2599 * These RPC_MAX_HEADER macros are overkill, especially since we
2600 * don't even do gss on the backchannel yet. But this is still
2601 * less than 1k. Tighten up this estimate in the unlikely event
2602 * it turns out to be a problem for some client:
2604 if (ca->maxreq_sz < NFSD_CB_MAX_REQ_SZ)
2605 return nfserr_toosmall;
2606 if (ca->maxresp_sz < NFSD_CB_MAX_RESP_SZ)
2607 return nfserr_toosmall;
2608 ca->maxresp_cached = 0;
2610 return nfserr_toosmall;
2615 static __be32 nfsd4_check_cb_sec(struct nfsd4_cb_sec *cbs)
2617 switch (cbs->flavor) {
2623 * GSS case: the spec doesn't allow us to return this
2624 * error. But it also doesn't allow us not to support
2626 * I'd rather this fail hard than return some error the
2627 * client might think it can already handle:
2629 return nfserr_encr_alg_unsupp;
2634 nfsd4_create_session(struct svc_rqst *rqstp,
2635 struct nfsd4_compound_state *cstate,
2636 struct nfsd4_create_session *cr_ses)
2638 struct sockaddr *sa = svc_addr(rqstp);
2639 struct nfs4_client *conf, *unconf;
2640 struct nfs4_client *old = NULL;
2641 struct nfsd4_session *new;
2642 struct nfsd4_conn *conn;
2643 struct nfsd4_clid_slot *cs_slot = NULL;
2645 struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
2647 if (cr_ses->flags & ~SESSION4_FLAG_MASK_A)
2648 return nfserr_inval;
2649 status = nfsd4_check_cb_sec(&cr_ses->cb_sec);
2652 status = check_forechannel_attrs(&cr_ses->fore_channel, nn);
2655 status = check_backchannel_attrs(&cr_ses->back_channel);
2657 goto out_release_drc_mem;
2658 status = nfserr_jukebox;
2659 new = alloc_session(&cr_ses->fore_channel, &cr_ses->back_channel);
2661 goto out_release_drc_mem;
2662 conn = alloc_conn_from_crses(rqstp, cr_ses);
2664 goto out_free_session;
2666 spin_lock(&nn->client_lock);
2667 unconf = find_unconfirmed_client(&cr_ses->clientid, true, nn);
2668 conf = find_confirmed_client(&cr_ses->clientid, true, nn);
2669 WARN_ON_ONCE(conf && unconf);
2672 status = nfserr_wrong_cred;
2673 if (!mach_creds_match(conf, rqstp))
2675 cs_slot = &conf->cl_cs_slot;
2676 status = check_slot_seqid(cr_ses->seqid, cs_slot->sl_seqid, 0);
2678 if (status == nfserr_replay_cache)
2679 status = nfsd4_replay_create_session(cr_ses, cs_slot);
2682 } else if (unconf) {
2683 if (!same_creds(&unconf->cl_cred, &rqstp->rq_cred) ||
2684 !rpc_cmp_addr(sa, (struct sockaddr *) &unconf->cl_addr)) {
2685 status = nfserr_clid_inuse;
2688 status = nfserr_wrong_cred;
2689 if (!mach_creds_match(unconf, rqstp))
2691 cs_slot = &unconf->cl_cs_slot;
2692 status = check_slot_seqid(cr_ses->seqid, cs_slot->sl_seqid, 0);
2694 /* an unconfirmed replay returns misordered */
2695 status = nfserr_seq_misordered;
2698 old = find_confirmed_client_by_name(&unconf->cl_name, nn);
2700 status = mark_client_expired_locked(old);
2706 move_to_confirmed(unconf);
2709 status = nfserr_stale_clientid;
2714 * We do not support RDMA or persistent sessions
2716 cr_ses->flags &= ~SESSION4_PERSIST;
2717 cr_ses->flags &= ~SESSION4_RDMA;
2719 init_session(rqstp, new, conf, cr_ses);
2720 nfsd4_get_session_locked(new);
2722 memcpy(cr_ses->sessionid.data, new->se_sessionid.data,
2723 NFS4_MAX_SESSIONID_LEN);
2724 cs_slot->sl_seqid++;
2725 cr_ses->seqid = cs_slot->sl_seqid;
2727 /* cache solo and embedded create sessions under the client_lock */
2728 nfsd4_cache_create_session(cr_ses, cs_slot, status);
2729 spin_unlock(&nn->client_lock);
2730 /* init connection and backchannel */
2731 nfsd4_init_conn(rqstp, conn, new);
2732 nfsd4_put_session(new);
2737 spin_unlock(&nn->client_lock);
2742 __free_session(new);
2743 out_release_drc_mem:
2744 nfsd4_put_drc_mem(&cr_ses->fore_channel);
2748 static __be32 nfsd4_map_bcts_dir(u32 *dir)
2751 case NFS4_CDFC4_FORE:
2752 case NFS4_CDFC4_BACK:
2754 case NFS4_CDFC4_FORE_OR_BOTH:
2755 case NFS4_CDFC4_BACK_OR_BOTH:
2756 *dir = NFS4_CDFC4_BOTH;
2759 return nfserr_inval;
2762 __be32 nfsd4_backchannel_ctl(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate, struct nfsd4_backchannel_ctl *bc)
2764 struct nfsd4_session *session = cstate->session;
2765 struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
2768 status = nfsd4_check_cb_sec(&bc->bc_cb_sec);
2771 spin_lock(&nn->client_lock);
2772 session->se_cb_prog = bc->bc_cb_program;
2773 session->se_cb_sec = bc->bc_cb_sec;
2774 spin_unlock(&nn->client_lock);
2776 nfsd4_probe_callback(session->se_client);
2781 __be32 nfsd4_bind_conn_to_session(struct svc_rqst *rqstp,
2782 struct nfsd4_compound_state *cstate,
2783 struct nfsd4_bind_conn_to_session *bcts)
2786 struct nfsd4_conn *conn;
2787 struct nfsd4_session *session;
2788 struct net *net = SVC_NET(rqstp);
2789 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
2791 if (!nfsd4_last_compound_op(rqstp))
2792 return nfserr_not_only_op;
2793 spin_lock(&nn->client_lock);
2794 session = find_in_sessionid_hashtbl(&bcts->sessionid, net, &status);
2795 spin_unlock(&nn->client_lock);
2797 goto out_no_session;
2798 status = nfserr_wrong_cred;
2799 if (!mach_creds_match(session->se_client, rqstp))
2801 status = nfsd4_map_bcts_dir(&bcts->dir);
2804 conn = alloc_conn(rqstp, bcts->dir);
2805 status = nfserr_jukebox;
2808 nfsd4_init_conn(rqstp, conn, session);
2811 nfsd4_put_session(session);
2816 static bool nfsd4_compound_in_session(struct nfsd4_session *session, struct nfs4_sessionid *sid)
2820 return !memcmp(sid, &session->se_sessionid, sizeof(*sid));
2824 nfsd4_destroy_session(struct svc_rqst *r,
2825 struct nfsd4_compound_state *cstate,
2826 struct nfsd4_destroy_session *sessionid)
2828 struct nfsd4_session *ses;
2830 int ref_held_by_me = 0;
2831 struct net *net = SVC_NET(r);
2832 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
2834 status = nfserr_not_only_op;
2835 if (nfsd4_compound_in_session(cstate->session, &sessionid->sessionid)) {
2836 if (!nfsd4_last_compound_op(r))
2840 dump_sessionid(__func__, &sessionid->sessionid);
2841 spin_lock(&nn->client_lock);
2842 ses = find_in_sessionid_hashtbl(&sessionid->sessionid, net, &status);
2844 goto out_client_lock;
2845 status = nfserr_wrong_cred;
2846 if (!mach_creds_match(ses->se_client, r))
2847 goto out_put_session;
2848 status = mark_session_dead_locked(ses, 1 + ref_held_by_me);
2850 goto out_put_session;
2851 unhash_session(ses);
2852 spin_unlock(&nn->client_lock);
2854 nfsd4_probe_callback_sync(ses->se_client);
2856 spin_lock(&nn->client_lock);
2859 nfsd4_put_session_locked(ses);
2861 spin_unlock(&nn->client_lock);
2866 static struct nfsd4_conn *__nfsd4_find_conn(struct svc_xprt *xpt, struct nfsd4_session *s)
2868 struct nfsd4_conn *c;
2870 list_for_each_entry(c, &s->se_conns, cn_persession) {
2871 if (c->cn_xprt == xpt) {
2878 static __be32 nfsd4_sequence_check_conn(struct nfsd4_conn *new, struct nfsd4_session *ses)
2880 struct nfs4_client *clp = ses->se_client;
2881 struct nfsd4_conn *c;
2882 __be32 status = nfs_ok;
2885 spin_lock(&clp->cl_lock);
2886 c = __nfsd4_find_conn(new->cn_xprt, ses);
2889 status = nfserr_conn_not_bound_to_session;
2890 if (clp->cl_mach_cred)
2892 __nfsd4_hash_conn(new, ses);
2893 spin_unlock(&clp->cl_lock);
2894 ret = nfsd4_register_conn(new);
2896 /* oops; xprt is already down: */
2897 nfsd4_conn_lost(&new->cn_xpt_user);
2900 spin_unlock(&clp->cl_lock);
2905 static bool nfsd4_session_too_many_ops(struct svc_rqst *rqstp, struct nfsd4_session *session)
2907 struct nfsd4_compoundargs *args = rqstp->rq_argp;
2909 return args->opcnt > session->se_fchannel.maxops;
2912 static bool nfsd4_request_too_big(struct svc_rqst *rqstp,
2913 struct nfsd4_session *session)
2915 struct xdr_buf *xb = &rqstp->rq_arg;
2917 return xb->len > session->se_fchannel.maxreq_sz;
2921 nfsd4_sequence(struct svc_rqst *rqstp,
2922 struct nfsd4_compound_state *cstate,
2923 struct nfsd4_sequence *seq)
2925 struct nfsd4_compoundres *resp = rqstp->rq_resp;
2926 struct xdr_stream *xdr = &resp->xdr;
2927 struct nfsd4_session *session;
2928 struct nfs4_client *clp;
2929 struct nfsd4_slot *slot;
2930 struct nfsd4_conn *conn;
2933 struct net *net = SVC_NET(rqstp);
2934 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
2936 if (resp->opcnt != 1)
2937 return nfserr_sequence_pos;
2940 * Will be either used or freed by nfsd4_sequence_check_conn
2943 conn = alloc_conn(rqstp, NFS4_CDFC4_FORE);
2945 return nfserr_jukebox;
2947 spin_lock(&nn->client_lock);
2948 session = find_in_sessionid_hashtbl(&seq->sessionid, net, &status);
2950 goto out_no_session;
2951 clp = session->se_client;
2953 status = nfserr_too_many_ops;
2954 if (nfsd4_session_too_many_ops(rqstp, session))
2955 goto out_put_session;
2957 status = nfserr_req_too_big;
2958 if (nfsd4_request_too_big(rqstp, session))
2959 goto out_put_session;
2961 status = nfserr_badslot;
2962 if (seq->slotid >= session->se_fchannel.maxreqs)
2963 goto out_put_session;
2965 slot = session->se_slots[seq->slotid];
2966 dprintk("%s: slotid %d\n", __func__, seq->slotid);
2968 /* We do not negotiate the number of slots yet, so set the
2969 * maxslots to the session maxreqs which is used to encode
2970 * sr_highest_slotid and the sr_target_slot id to maxslots */
2971 seq->maxslots = session->se_fchannel.maxreqs;
2973 status = check_slot_seqid(seq->seqid, slot->sl_seqid,
2974 slot->sl_flags & NFSD4_SLOT_INUSE);
2975 if (status == nfserr_replay_cache) {
2976 status = nfserr_seq_misordered;
2977 if (!(slot->sl_flags & NFSD4_SLOT_INITIALIZED))
2978 goto out_put_session;
2979 cstate->slot = slot;
2980 cstate->session = session;
2982 /* Return the cached reply status and set cstate->status
2983 * for nfsd4_proc_compound processing */
2984 status = nfsd4_replay_cache_entry(resp, seq);
2985 cstate->status = nfserr_replay_cache;
2989 goto out_put_session;
2991 status = nfsd4_sequence_check_conn(conn, session);
2994 goto out_put_session;
2996 buflen = (seq->cachethis) ?
2997 session->se_fchannel.maxresp_cached :
2998 session->se_fchannel.maxresp_sz;
2999 status = (seq->cachethis) ? nfserr_rep_too_big_to_cache :
3001 if (xdr_restrict_buflen(xdr, buflen - rqstp->rq_auth_slack))
3002 goto out_put_session;
3003 svc_reserve(rqstp, buflen);
3006 /* Success! bump slot seqid */
3007 slot->sl_seqid = seq->seqid;
3008 slot->sl_flags |= NFSD4_SLOT_INUSE;
3010 slot->sl_flags |= NFSD4_SLOT_CACHETHIS;
3012 slot->sl_flags &= ~NFSD4_SLOT_CACHETHIS;
3014 cstate->slot = slot;
3015 cstate->session = session;
3019 switch (clp->cl_cb_state) {
3021 seq->status_flags = SEQ4_STATUS_CB_PATH_DOWN;
3023 case NFSD4_CB_FAULT:
3024 seq->status_flags = SEQ4_STATUS_BACKCHANNEL_FAULT;
3027 seq->status_flags = 0;
3029 if (!list_empty(&clp->cl_revoked))
3030 seq->status_flags |= SEQ4_STATUS_RECALLABLE_STATE_REVOKED;
3034 spin_unlock(&nn->client_lock);
3037 nfsd4_put_session_locked(session);
3038 goto out_no_session;
3042 nfsd4_sequence_done(struct nfsd4_compoundres *resp)
3044 struct nfsd4_compound_state *cs = &resp->cstate;
3046 if (nfsd4_has_session(cs)) {
3047 if (cs->status != nfserr_replay_cache) {
3048 nfsd4_store_cache_entry(resp);
3049 cs->slot->sl_flags &= ~NFSD4_SLOT_INUSE;
3051 /* Drop session reference that was taken in nfsd4_sequence() */
3052 nfsd4_put_session(cs->session);
3054 put_client_renew(cs->clp);
3058 nfsd4_destroy_clientid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate, struct nfsd4_destroy_clientid *dc)
3060 struct nfs4_client *conf, *unconf;
3061 struct nfs4_client *clp = NULL;
3063 struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
3065 spin_lock(&nn->client_lock);
3066 unconf = find_unconfirmed_client(&dc->clientid, true, nn);
3067 conf = find_confirmed_client(&dc->clientid, true, nn);
3068 WARN_ON_ONCE(conf && unconf);
3071 if (client_has_state(conf)) {
3072 status = nfserr_clientid_busy;
3075 status = mark_client_expired_locked(conf);
3082 status = nfserr_stale_clientid;
3085 if (!mach_creds_match(clp, rqstp)) {
3087 status = nfserr_wrong_cred;
3090 unhash_client_locked(clp);
3092 spin_unlock(&nn->client_lock);
3099 nfsd4_reclaim_complete(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate, struct nfsd4_reclaim_complete *rc)
3103 if (rc->rca_one_fs) {
3104 if (!cstate->current_fh.fh_dentry)
3105 return nfserr_nofilehandle;
3107 * We don't take advantage of the rca_one_fs case.
3108 * That's OK, it's optional, we can safely ignore it.
3113 status = nfserr_complete_already;
3114 if (test_and_set_bit(NFSD4_CLIENT_RECLAIM_COMPLETE,
3115 &cstate->session->se_client->cl_flags))
3118 status = nfserr_stale_clientid;
3119 if (is_client_expired(cstate->session->se_client))
3121 * The following error isn't really legal.
3122 * But we only get here if the client just explicitly
3123 * destroyed the client. Surely it no longer cares what
3124 * error it gets back on an operation for the dead
3130 nfsd4_client_record_create(cstate->session->se_client);
3136 nfsd4_setclientid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3137 struct nfsd4_setclientid *setclid)
3139 struct xdr_netobj clname = setclid->se_name;
3140 nfs4_verifier clverifier = setclid->se_verf;
3141 struct nfs4_client *conf, *new;
3142 struct nfs4_client *unconf = NULL;
3144 struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
3146 new = create_client(clname, rqstp, &clverifier);
3148 return nfserr_jukebox;
3149 /* Cases below refer to rfc 3530 section 14.2.33: */
3150 spin_lock(&nn->client_lock);
3151 conf = find_confirmed_client_by_name(&clname, nn);
3152 if (conf && client_has_state(conf)) {
3154 status = nfserr_clid_inuse;
3155 if (clp_used_exchangeid(conf))
3157 if (!same_creds(&conf->cl_cred, &rqstp->rq_cred)) {
3158 char addr_str[INET6_ADDRSTRLEN];
3159 rpc_ntop((struct sockaddr *) &conf->cl_addr, addr_str,
3161 dprintk("NFSD: setclientid: string in use by client "
3162 "at %s\n", addr_str);
3166 unconf = find_unconfirmed_client_by_name(&clname, nn);
3168 unhash_client_locked(unconf);
3169 if (conf && same_verf(&conf->cl_verifier, &clverifier)) {
3170 /* case 1: probable callback update */
3171 copy_clid(new, conf);
3172 gen_confirm(new, nn);
3173 } else /* case 4 (new client) or cases 2, 3 (client reboot): */
3175 new->cl_minorversion = 0;
3176 gen_callback(new, setclid, rqstp);
3177 add_to_unconfirmed(new);
3178 setclid->se_clientid.cl_boot = new->cl_clientid.cl_boot;
3179 setclid->se_clientid.cl_id = new->cl_clientid.cl_id;
3180 memcpy(setclid->se_confirm.data, new->cl_confirm.data, sizeof(setclid->se_confirm.data));
3184 spin_unlock(&nn->client_lock);
3188 expire_client(unconf);
3194 nfsd4_setclientid_confirm(struct svc_rqst *rqstp,
3195 struct nfsd4_compound_state *cstate,
3196 struct nfsd4_setclientid_confirm *setclientid_confirm)
3198 struct nfs4_client *conf, *unconf;
3199 struct nfs4_client *old = NULL;
3200 nfs4_verifier confirm = setclientid_confirm->sc_confirm;
3201 clientid_t * clid = &setclientid_confirm->sc_clientid;
3203 struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
3205 if (STALE_CLIENTID(clid, nn))
3206 return nfserr_stale_clientid;
3208 spin_lock(&nn->client_lock);
3209 conf = find_confirmed_client(clid, false, nn);
3210 unconf = find_unconfirmed_client(clid, false, nn);
3212 * We try hard to give out unique clientid's, so if we get an
3213 * attempt to confirm the same clientid with a different cred,
3214 * the client may be buggy; this should never happen.
3216 * Nevertheless, RFC 7530 recommends INUSE for this case:
3218 status = nfserr_clid_inuse;
3219 if (unconf && !same_creds(&unconf->cl_cred, &rqstp->rq_cred))
3221 if (conf && !same_creds(&conf->cl_cred, &rqstp->rq_cred))
3223 /* cases below refer to rfc 3530 section 14.2.34: */
3224 if (!unconf || !same_verf(&confirm, &unconf->cl_confirm)) {
3225 if (conf && !unconf) /* case 2: probable retransmit */
3227 else /* case 4: client hasn't noticed we rebooted yet? */
3228 status = nfserr_stale_clientid;
3232 if (conf) { /* case 1: callback update */
3234 unhash_client_locked(old);
3235 nfsd4_change_callback(conf, &unconf->cl_cb_conn);
3236 } else { /* case 3: normal case; new or rebooted client */
3237 old = find_confirmed_client_by_name(&unconf->cl_name, nn);
3239 status = nfserr_clid_inuse;
3240 if (client_has_state(old)
3241 && !same_creds(&unconf->cl_cred,
3244 status = mark_client_expired_locked(old);
3250 move_to_confirmed(unconf);
3253 get_client_locked(conf);
3254 spin_unlock(&nn->client_lock);
3255 nfsd4_probe_callback(conf);
3256 spin_lock(&nn->client_lock);
3257 put_client_renew_locked(conf);
3259 spin_unlock(&nn->client_lock);
3265 static struct nfs4_file *nfsd4_alloc_file(void)
3267 return kmem_cache_alloc(file_slab, GFP_KERNEL);
3270 /* OPEN Share state helper functions */
3271 static void nfsd4_init_file(struct knfsd_fh *fh, unsigned int hashval,
3272 struct nfs4_file *fp)
3274 lockdep_assert_held(&state_lock);
3276 atomic_set(&fp->fi_ref, 1);
3277 spin_lock_init(&fp->fi_lock);
3278 INIT_LIST_HEAD(&fp->fi_stateids);
3279 INIT_LIST_HEAD(&fp->fi_delegations);
3280 INIT_LIST_HEAD(&fp->fi_clnt_odstate);
3281 fh_copy_shallow(&fp->fi_fhandle, fh);
3282 fp->fi_deleg_file = NULL;
3283 fp->fi_had_conflict = false;
3284 fp->fi_share_deny = 0;
3285 memset(fp->fi_fds, 0, sizeof(fp->fi_fds));
3286 memset(fp->fi_access, 0, sizeof(fp->fi_access));
3287 #ifdef CONFIG_NFSD_PNFS
3288 INIT_LIST_HEAD(&fp->fi_lo_states);
3289 atomic_set(&fp->fi_lo_recalls, 0);
3291 hlist_add_head_rcu(&fp->fi_hash, &file_hashtbl[hashval]);
3295 nfsd4_free_slabs(void)
3297 kmem_cache_destroy(odstate_slab);
3298 kmem_cache_destroy(openowner_slab);
3299 kmem_cache_destroy(lockowner_slab);
3300 kmem_cache_destroy(file_slab);
3301 kmem_cache_destroy(stateid_slab);
3302 kmem_cache_destroy(deleg_slab);
3306 nfsd4_init_slabs(void)
3308 openowner_slab = kmem_cache_create("nfsd4_openowners",
3309 sizeof(struct nfs4_openowner), 0, 0, NULL);
3310 if (openowner_slab == NULL)
3312 lockowner_slab = kmem_cache_create("nfsd4_lockowners",
3313 sizeof(struct nfs4_lockowner), 0, 0, NULL);
3314 if (lockowner_slab == NULL)
3315 goto out_free_openowner_slab;
3316 file_slab = kmem_cache_create("nfsd4_files",
3317 sizeof(struct nfs4_file), 0, 0, NULL);
3318 if (file_slab == NULL)
3319 goto out_free_lockowner_slab;
3320 stateid_slab = kmem_cache_create("nfsd4_stateids",
3321 sizeof(struct nfs4_ol_stateid), 0, 0, NULL);
3322 if (stateid_slab == NULL)
3323 goto out_free_file_slab;
3324 deleg_slab = kmem_cache_create("nfsd4_delegations",
3325 sizeof(struct nfs4_delegation), 0, 0, NULL);
3326 if (deleg_slab == NULL)
3327 goto out_free_stateid_slab;
3328 odstate_slab = kmem_cache_create("nfsd4_odstate",
3329 sizeof(struct nfs4_clnt_odstate), 0, 0, NULL);
3330 if (odstate_slab == NULL)
3331 goto out_free_deleg_slab;
3334 out_free_deleg_slab:
3335 kmem_cache_destroy(deleg_slab);
3336 out_free_stateid_slab:
3337 kmem_cache_destroy(stateid_slab);
3339 kmem_cache_destroy(file_slab);
3340 out_free_lockowner_slab:
3341 kmem_cache_destroy(lockowner_slab);
3342 out_free_openowner_slab:
3343 kmem_cache_destroy(openowner_slab);
3345 dprintk("nfsd4: out of memory while initializing nfsv4\n");
3349 static void init_nfs4_replay(struct nfs4_replay *rp)
3351 rp->rp_status = nfserr_serverfault;
3353 rp->rp_buf = rp->rp_ibuf;
3354 mutex_init(&rp->rp_mutex);
3357 static void nfsd4_cstate_assign_replay(struct nfsd4_compound_state *cstate,
3358 struct nfs4_stateowner *so)
3360 if (!nfsd4_has_session(cstate)) {
3361 mutex_lock(&so->so_replay.rp_mutex);
3362 cstate->replay_owner = nfs4_get_stateowner(so);
3366 void nfsd4_cstate_clear_replay(struct nfsd4_compound_state *cstate)
3368 struct nfs4_stateowner *so = cstate->replay_owner;
3371 cstate->replay_owner = NULL;
3372 mutex_unlock(&so->so_replay.rp_mutex);
3373 nfs4_put_stateowner(so);
3377 static inline void *alloc_stateowner(struct kmem_cache *slab, struct xdr_netobj *owner, struct nfs4_client *clp)
3379 struct nfs4_stateowner *sop;
3381 sop = kmem_cache_alloc(slab, GFP_KERNEL);
3385 sop->so_owner.data = kmemdup(owner->data, owner->len, GFP_KERNEL);
3386 if (!sop->so_owner.data) {
3387 kmem_cache_free(slab, sop);
3390 sop->so_owner.len = owner->len;
3392 INIT_LIST_HEAD(&sop->so_stateids);
3393 sop->so_client = clp;
3394 init_nfs4_replay(&sop->so_replay);
3395 atomic_set(&sop->so_count, 1);
3399 static void hash_openowner(struct nfs4_openowner *oo, struct nfs4_client *clp, unsigned int strhashval)
3401 lockdep_assert_held(&clp->cl_lock);
3403 list_add(&oo->oo_owner.so_strhash,
3404 &clp->cl_ownerstr_hashtbl[strhashval]);
3405 list_add(&oo->oo_perclient, &clp->cl_openowners);
3408 static void nfs4_unhash_openowner(struct nfs4_stateowner *so)
3410 unhash_openowner_locked(openowner(so));
3413 static void nfs4_free_openowner(struct nfs4_stateowner *so)
3415 struct nfs4_openowner *oo = openowner(so);
3417 kmem_cache_free(openowner_slab, oo);
3420 static const struct nfs4_stateowner_operations openowner_ops = {
3421 .so_unhash = nfs4_unhash_openowner,
3422 .so_free = nfs4_free_openowner,
3425 static struct nfs4_ol_stateid *
3426 nfsd4_find_existing_open(struct nfs4_file *fp, struct nfsd4_open *open)
3428 struct nfs4_ol_stateid *local, *ret = NULL;
3429 struct nfs4_openowner *oo = open->op_openowner;
3431 lockdep_assert_held(&fp->fi_lock);
3433 list_for_each_entry(local, &fp->fi_stateids, st_perfile) {
3434 /* ignore lock owners */
3435 if (local->st_stateowner->so_is_open_owner == 0)
3437 if (local->st_stateowner == &oo->oo_owner) {
3439 atomic_inc(&ret->st_stid.sc_count);
3446 static struct nfs4_openowner *
3447 alloc_init_open_stateowner(unsigned int strhashval, struct nfsd4_open *open,
3448 struct nfsd4_compound_state *cstate)
3450 struct nfs4_client *clp = cstate->clp;
3451 struct nfs4_openowner *oo, *ret;
3453 oo = alloc_stateowner(openowner_slab, &open->op_owner, clp);
3456 oo->oo_owner.so_ops = &openowner_ops;
3457 oo->oo_owner.so_is_open_owner = 1;
3458 oo->oo_owner.so_seqid = open->op_seqid;
3460 if (nfsd4_has_session(cstate))
3461 oo->oo_flags |= NFS4_OO_CONFIRMED;
3463 oo->oo_last_closed_stid = NULL;
3464 INIT_LIST_HEAD(&oo->oo_close_lru);
3465 spin_lock(&clp->cl_lock);
3466 ret = find_openstateowner_str_locked(strhashval, open, clp);
3468 hash_openowner(oo, clp, strhashval);
3471 nfs4_free_stateowner(&oo->oo_owner);
3473 spin_unlock(&clp->cl_lock);
3477 static struct nfs4_ol_stateid *
3478 init_open_stateid(struct nfs4_ol_stateid *stp, struct nfs4_file *fp,
3479 struct nfsd4_open *open)
3482 struct nfs4_openowner *oo = open->op_openowner;
3483 struct nfs4_ol_stateid *retstp = NULL;
3485 spin_lock(&oo->oo_owner.so_client->cl_lock);
3486 spin_lock(&fp->fi_lock);
3488 retstp = nfsd4_find_existing_open(fp, open);
3491 atomic_inc(&stp->st_stid.sc_count);
3492 stp->st_stid.sc_type = NFS4_OPEN_STID;
3493 INIT_LIST_HEAD(&stp->st_locks);
3494 stp->st_stateowner = nfs4_get_stateowner(&oo->oo_owner);
3496 stp->st_stid.sc_file = fp;
3497 stp->st_access_bmap = 0;
3498 stp->st_deny_bmap = 0;
3499 stp->st_openstp = NULL;
3500 init_rwsem(&stp->st_rwsem);
3501 list_add(&stp->st_perstateowner, &oo->oo_owner.so_stateids);
3502 list_add(&stp->st_perfile, &fp->fi_stateids);
3505 spin_unlock(&fp->fi_lock);
3506 spin_unlock(&oo->oo_owner.so_client->cl_lock);
3511 * In the 4.0 case we need to keep the owners around a little while to handle
3512 * CLOSE replay. We still do need to release any file access that is held by
3513 * them before returning however.
3516 move_to_close_lru(struct nfs4_ol_stateid *s, struct net *net)
3518 struct nfs4_ol_stateid *last;
3519 struct nfs4_openowner *oo = openowner(s->st_stateowner);
3520 struct nfsd_net *nn = net_generic(s->st_stid.sc_client->net,
3523 dprintk("NFSD: move_to_close_lru nfs4_openowner %p\n", oo);
3526 * We know that we hold one reference via nfsd4_close, and another
3527 * "persistent" reference for the client. If the refcount is higher
3528 * than 2, then there are still calls in progress that are using this
3529 * stateid. We can't put the sc_file reference until they are finished.
3530 * Wait for the refcount to drop to 2. Since it has been unhashed,
3531 * there should be no danger of the refcount going back up again at
3534 wait_event(close_wq, atomic_read(&s->st_stid.sc_count) == 2);
3536 release_all_access(s);
3537 if (s->st_stid.sc_file) {
3538 put_nfs4_file(s->st_stid.sc_file);
3539 s->st_stid.sc_file = NULL;
3542 spin_lock(&nn->client_lock);
3543 last = oo->oo_last_closed_stid;
3544 oo->oo_last_closed_stid = s;
3545 list_move_tail(&oo->oo_close_lru, &nn->close_lru);
3546 oo->oo_time = get_seconds();
3547 spin_unlock(&nn->client_lock);
3549 nfs4_put_stid(&last->st_stid);
3552 /* search file_hashtbl[] for file */
3553 static struct nfs4_file *
3554 find_file_locked(struct knfsd_fh *fh, unsigned int hashval)
3556 struct nfs4_file *fp;
3558 hlist_for_each_entry_rcu(fp, &file_hashtbl[hashval], fi_hash) {
3559 if (fh_match(&fp->fi_fhandle, fh)) {
3560 if (atomic_inc_not_zero(&fp->fi_ref))
3568 find_file(struct knfsd_fh *fh)
3570 struct nfs4_file *fp;
3571 unsigned int hashval = file_hashval(fh);
3574 fp = find_file_locked(fh, hashval);
3579 static struct nfs4_file *
3580 find_or_add_file(struct nfs4_file *new, struct knfsd_fh *fh)
3582 struct nfs4_file *fp;
3583 unsigned int hashval = file_hashval(fh);
3586 fp = find_file_locked(fh, hashval);
3591 spin_lock(&state_lock);
3592 fp = find_file_locked(fh, hashval);
3593 if (likely(fp == NULL)) {
3594 nfsd4_init_file(fh, hashval, new);
3597 spin_unlock(&state_lock);
3603 * Called to check deny when READ with all zero stateid or
3604 * WRITE with all zero or all one stateid
3607 nfs4_share_conflict(struct svc_fh *current_fh, unsigned int deny_type)
3609 struct nfs4_file *fp;
3610 __be32 ret = nfs_ok;
3612 fp = find_file(¤t_fh->fh_handle);
3615 /* Check for conflicting share reservations */
3616 spin_lock(&fp->fi_lock);
3617 if (fp->fi_share_deny & deny_type)
3618 ret = nfserr_locked;
3619 spin_unlock(&fp->fi_lock);
3624 static void nfsd4_cb_recall_prepare(struct nfsd4_callback *cb)
3626 struct nfs4_delegation *dp = cb_to_delegation(cb);
3627 struct nfsd_net *nn = net_generic(dp->dl_stid.sc_client->net,
3630 block_delegations(&dp->dl_stid.sc_file->fi_fhandle);
3633 * We can't do this in nfsd_break_deleg_cb because it is
3634 * already holding inode->i_lock.
3636 * If the dl_time != 0, then we know that it has already been
3637 * queued for a lease break. Don't queue it again.
3639 spin_lock(&state_lock);
3640 if (dp->dl_time == 0) {
3641 dp->dl_time = get_seconds();
3642 list_add_tail(&dp->dl_recall_lru, &nn->del_recall_lru);
3644 spin_unlock(&state_lock);
3647 static int nfsd4_cb_recall_done(struct nfsd4_callback *cb,
3648 struct rpc_task *task)
3650 struct nfs4_delegation *dp = cb_to_delegation(cb);
3652 if (dp->dl_stid.sc_type == NFS4_CLOSED_DELEG_STID)
3655 switch (task->tk_status) {
3659 case -NFS4ERR_BAD_STATEID:
3661 * Race: client probably got cb_recall before open reply
3662 * granting delegation.
3664 if (dp->dl_retries--) {
3665 rpc_delay(task, 2 * HZ);
3674 static void nfsd4_cb_recall_release(struct nfsd4_callback *cb)
3676 struct nfs4_delegation *dp = cb_to_delegation(cb);
3678 nfs4_put_stid(&dp->dl_stid);
3681 static const struct nfsd4_callback_ops nfsd4_cb_recall_ops = {
3682 .prepare = nfsd4_cb_recall_prepare,
3683 .done = nfsd4_cb_recall_done,
3684 .release = nfsd4_cb_recall_release,
3687 static void nfsd_break_one_deleg(struct nfs4_delegation *dp)
3690 * We're assuming the state code never drops its reference
3691 * without first removing the lease. Since we're in this lease
3692 * callback (and since the lease code is serialized by the kernel
3693 * lock) we know the server hasn't removed the lease yet, we know
3694 * it's safe to take a reference.
3696 atomic_inc(&dp->dl_stid.sc_count);
3697 nfsd4_run_cb(&dp->dl_recall);
3700 /* Called from break_lease() with i_lock held. */
3702 nfsd_break_deleg_cb(struct file_lock *fl)
3705 struct nfs4_file *fp = (struct nfs4_file *)fl->fl_owner;
3706 struct nfs4_delegation *dp;
3709 WARN(1, "(%p)->fl_owner NULL\n", fl);
3712 if (fp->fi_had_conflict) {
3713 WARN(1, "duplicate break on %p\n", fp);
3717 * We don't want the locks code to timeout the lease for us;
3718 * we'll remove it ourself if a delegation isn't returned
3721 fl->fl_break_time = 0;
3723 spin_lock(&fp->fi_lock);
3724 fp->fi_had_conflict = true;
3726 * If there are no delegations on the list, then return true
3727 * so that the lease code will go ahead and delete it.
3729 if (list_empty(&fp->fi_delegations))
3732 list_for_each_entry(dp, &fp->fi_delegations, dl_perfile)
3733 nfsd_break_one_deleg(dp);
3734 spin_unlock(&fp->fi_lock);
3739 nfsd_change_deleg_cb(struct file_lock *onlist, int arg,
3740 struct list_head *dispose)
3743 return lease_modify(onlist, arg, dispose);
3748 static const struct lock_manager_operations nfsd_lease_mng_ops = {
3749 .lm_break = nfsd_break_deleg_cb,
3750 .lm_change = nfsd_change_deleg_cb,
3753 static __be32 nfsd4_check_seqid(struct nfsd4_compound_state *cstate, struct nfs4_stateowner *so, u32 seqid)
3755 if (nfsd4_has_session(cstate))
3757 if (seqid == so->so_seqid - 1)
3758 return nfserr_replay_me;
3759 if (seqid == so->so_seqid)
3761 return nfserr_bad_seqid;
3764 static __be32 lookup_clientid(clientid_t *clid,
3765 struct nfsd4_compound_state *cstate,
3766 struct nfsd_net *nn)
3768 struct nfs4_client *found;
3771 found = cstate->clp;
3772 if (!same_clid(&found->cl_clientid, clid))
3773 return nfserr_stale_clientid;
3777 if (STALE_CLIENTID(clid, nn))
3778 return nfserr_stale_clientid;
3781 * For v4.1+ we get the client in the SEQUENCE op. If we don't have one
3782 * cached already then we know this is for is for v4.0 and "sessions"
3785 WARN_ON_ONCE(cstate->session);
3786 spin_lock(&nn->client_lock);
3787 found = find_confirmed_client(clid, false, nn);
3789 spin_unlock(&nn->client_lock);
3790 return nfserr_expired;
3792 atomic_inc(&found->cl_refcount);
3793 spin_unlock(&nn->client_lock);
3795 /* Cache the nfs4_client in cstate! */
3796 cstate->clp = found;
3801 nfsd4_process_open1(struct nfsd4_compound_state *cstate,
3802 struct nfsd4_open *open, struct nfsd_net *nn)
3804 clientid_t *clientid = &open->op_clientid;
3805 struct nfs4_client *clp = NULL;
3806 unsigned int strhashval;
3807 struct nfs4_openowner *oo = NULL;
3810 if (STALE_CLIENTID(&open->op_clientid, nn))
3811 return nfserr_stale_clientid;
3813 * In case we need it later, after we've already created the
3814 * file and don't want to risk a further failure:
3816 open->op_file = nfsd4_alloc_file();
3817 if (open->op_file == NULL)
3818 return nfserr_jukebox;
3820 status = lookup_clientid(clientid, cstate, nn);
3825 strhashval = ownerstr_hashval(&open->op_owner);
3826 oo = find_openstateowner_str(strhashval, open, clp);
3827 open->op_openowner = oo;
3831 if (!(oo->oo_flags & NFS4_OO_CONFIRMED)) {
3832 /* Replace unconfirmed owners without checking for replay. */
3833 release_openowner(oo);
3834 open->op_openowner = NULL;
3837 status = nfsd4_check_seqid(cstate, &oo->oo_owner, open->op_seqid);
3842 oo = alloc_init_open_stateowner(strhashval, open, cstate);
3844 return nfserr_jukebox;
3845 open->op_openowner = oo;
3847 open->op_stp = nfs4_alloc_open_stateid(clp);
3849 return nfserr_jukebox;
3851 if (nfsd4_has_session(cstate) &&
3852 (cstate->current_fh.fh_export->ex_flags & NFSEXP_PNFS)) {
3853 open->op_odstate = alloc_clnt_odstate(clp);
3854 if (!open->op_odstate)
3855 return nfserr_jukebox;
3861 static inline __be32
3862 nfs4_check_delegmode(struct nfs4_delegation *dp, int flags)
3864 if ((flags & WR_STATE) && (dp->dl_type == NFS4_OPEN_DELEGATE_READ))
3865 return nfserr_openmode;
3870 static int share_access_to_flags(u32 share_access)
3872 return share_access == NFS4_SHARE_ACCESS_READ ? RD_STATE : WR_STATE;
3875 static struct nfs4_delegation *find_deleg_stateid(struct nfs4_client *cl, stateid_t *s)
3877 struct nfs4_stid *ret;
3879 ret = find_stateid_by_type(cl, s, NFS4_DELEG_STID);
3882 return delegstateid(ret);
3885 static bool nfsd4_is_deleg_cur(struct nfsd4_open *open)
3887 return open->op_claim_type == NFS4_OPEN_CLAIM_DELEGATE_CUR ||
3888 open->op_claim_type == NFS4_OPEN_CLAIM_DELEG_CUR_FH;
3892 nfs4_check_deleg(struct nfs4_client *cl, struct nfsd4_open *open,
3893 struct nfs4_delegation **dp)
3896 __be32 status = nfserr_bad_stateid;
3897 struct nfs4_delegation *deleg;
3899 deleg = find_deleg_stateid(cl, &open->op_delegate_stateid);
3902 flags = share_access_to_flags(open->op_share_access);
3903 status = nfs4_check_delegmode(deleg, flags);
3905 nfs4_put_stid(&deleg->dl_stid);
3910 if (!nfsd4_is_deleg_cur(open))
3914 open->op_openowner->oo_flags |= NFS4_OO_CONFIRMED;
3918 static inline int nfs4_access_to_access(u32 nfs4_access)
3922 if (nfs4_access & NFS4_SHARE_ACCESS_READ)
3923 flags |= NFSD_MAY_READ;
3924 if (nfs4_access & NFS4_SHARE_ACCESS_WRITE)
3925 flags |= NFSD_MAY_WRITE;
3929 static inline __be32
3930 nfsd4_truncate(struct svc_rqst *rqstp, struct svc_fh *fh,
3931 struct nfsd4_open *open)
3933 struct iattr iattr = {
3934 .ia_valid = ATTR_SIZE,
3937 if (!open->op_truncate)
3939 if (!(open->op_share_access & NFS4_SHARE_ACCESS_WRITE))
3940 return nfserr_inval;
3941 return nfsd_setattr(rqstp, fh, &iattr, 0, (time_t)0);
3944 static __be32 nfs4_get_vfs_file(struct svc_rqst *rqstp, struct nfs4_file *fp,
3945 struct svc_fh *cur_fh, struct nfs4_ol_stateid *stp,
3946 struct nfsd4_open *open)
3948 struct file *filp = NULL;
3950 int oflag = nfs4_access_to_omode(open->op_share_access);
3951 int access = nfs4_access_to_access(open->op_share_access);
3952 unsigned char old_access_bmap, old_deny_bmap;
3954 spin_lock(&fp->fi_lock);
3957 * Are we trying to set a deny mode that would conflict with
3960 status = nfs4_file_check_deny(fp, open->op_share_deny);
3961 if (status != nfs_ok) {
3962 spin_unlock(&fp->fi_lock);
3966 /* set access to the file */
3967 status = nfs4_file_get_access(fp, open->op_share_access);
3968 if (status != nfs_ok) {
3969 spin_unlock(&fp->fi_lock);
3973 /* Set access bits in stateid */
3974 old_access_bmap = stp->st_access_bmap;
3975 set_access(open->op_share_access, stp);
3977 /* Set new deny mask */
3978 old_deny_bmap = stp->st_deny_bmap;
3979 set_deny(open->op_share_deny, stp);
3980 fp->fi_share_deny |= (open->op_share_deny & NFS4_SHARE_DENY_BOTH);
3982 if (!fp->fi_fds[oflag]) {
3983 spin_unlock(&fp->fi_lock);
3984 status = nfsd_open(rqstp, cur_fh, S_IFREG, access, &filp);
3986 goto out_put_access;
3987 spin_lock(&fp->fi_lock);
3988 if (!fp->fi_fds[oflag]) {
3989 fp->fi_fds[oflag] = filp;
3993 spin_unlock(&fp->fi_lock);
3997 status = nfsd4_truncate(rqstp, cur_fh, open);
3999 goto out_put_access;
4003 stp->st_access_bmap = old_access_bmap;
4004 nfs4_file_put_access(fp, open->op_share_access);
4005 reset_union_bmap_deny(bmap_to_share_mode(old_deny_bmap), stp);
4010 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)
4013 unsigned char old_deny_bmap = stp->st_deny_bmap;
4015 if (!test_access(open->op_share_access, stp))
4016 return nfs4_get_vfs_file(rqstp, fp, cur_fh, stp, open);
4018 /* test and set deny mode */
4019 spin_lock(&fp->fi_lock);
4020 status = nfs4_file_check_deny(fp, open->op_share_deny);
4021 if (status == nfs_ok) {
4022 set_deny(open->op_share_deny, stp);
4023 fp->fi_share_deny |=
4024 (open->op_share_deny & NFS4_SHARE_DENY_BOTH);
4026 spin_unlock(&fp->fi_lock);
4028 if (status != nfs_ok)
4031 status = nfsd4_truncate(rqstp, cur_fh, open);
4032 if (status != nfs_ok)
4033 reset_union_bmap_deny(old_deny_bmap, stp);
4037 /* Should we give out recallable state?: */
4038 static bool nfsd4_cb_channel_good(struct nfs4_client *clp)
4040 if (clp->cl_cb_state == NFSD4_CB_UP)
4043 * In the sessions case, since we don't have to establish a
4044 * separate connection for callbacks, we assume it's OK
4045 * until we hear otherwise:
4047 return clp->cl_minorversion && clp->cl_cb_state == NFSD4_CB_UNKNOWN;
4050 static struct file_lock *nfs4_alloc_init_lease(struct nfs4_file *fp, int flag)
4052 struct file_lock *fl;
4054 fl = locks_alloc_lock();
4057 fl->fl_lmops = &nfsd_lease_mng_ops;
4058 fl->fl_flags = FL_DELEG;
4059 fl->fl_type = flag == NFS4_OPEN_DELEGATE_READ? F_RDLCK: F_WRLCK;
4060 fl->fl_end = OFFSET_MAX;
4061 fl->fl_owner = (fl_owner_t)fp;
4062 fl->fl_pid = current->tgid;
4067 * nfs4_setlease - Obtain a delegation by requesting lease from vfs layer
4068 * @dp: a pointer to the nfs4_delegation we're adding.
4071 * On success: Return code will be 0 on success.
4073 * On error: -EAGAIN if there was an existing delegation.
4074 * nonzero if there is an error in other cases.
4078 static int nfs4_setlease(struct nfs4_delegation *dp)
4080 struct nfs4_file *fp = dp->dl_stid.sc_file;
4081 struct file_lock *fl;
4085 fl = nfs4_alloc_init_lease(fp, NFS4_OPEN_DELEGATE_READ);
4088 filp = find_readable_file(fp);
4090 /* We should always have a readable file here */
4092 locks_free_lock(fl);
4096 status = vfs_setlease(filp, fl->fl_type, &fl, NULL);
4098 locks_free_lock(fl);
4101 spin_lock(&state_lock);
4102 spin_lock(&fp->fi_lock);
4103 /* Did the lease get broken before we took the lock? */
4105 if (fp->fi_had_conflict)
4108 if (fp->fi_deleg_file) {
4109 status = hash_delegation_locked(dp, fp);
4112 fp->fi_deleg_file = filp;
4113 fp->fi_delegees = 0;
4114 status = hash_delegation_locked(dp, fp);
4115 spin_unlock(&fp->fi_lock);
4116 spin_unlock(&state_lock);
4118 /* Should never happen, this is a new fi_deleg_file */
4124 spin_unlock(&fp->fi_lock);
4125 spin_unlock(&state_lock);
4131 static struct nfs4_delegation *
4132 nfs4_set_delegation(struct nfs4_client *clp, struct svc_fh *fh,
4133 struct nfs4_file *fp, struct nfs4_clnt_odstate *odstate)
4136 struct nfs4_delegation *dp;
4138 if (fp->fi_had_conflict)
4139 return ERR_PTR(-EAGAIN);
4141 spin_lock(&state_lock);
4142 spin_lock(&fp->fi_lock);
4143 status = nfs4_get_existing_delegation(clp, fp);
4144 spin_unlock(&fp->fi_lock);
4145 spin_unlock(&state_lock);
4148 return ERR_PTR(status);
4150 dp = alloc_init_deleg(clp, fh, odstate);
4152 return ERR_PTR(-ENOMEM);
4155 spin_lock(&state_lock);
4156 spin_lock(&fp->fi_lock);
4157 dp->dl_stid.sc_file = fp;
4158 if (!fp->fi_deleg_file) {
4159 spin_unlock(&fp->fi_lock);
4160 spin_unlock(&state_lock);
4161 status = nfs4_setlease(dp);
4164 if (fp->fi_had_conflict) {
4168 status = hash_delegation_locked(dp, fp);
4170 spin_unlock(&fp->fi_lock);
4171 spin_unlock(&state_lock);
4174 put_clnt_odstate(dp->dl_clnt_odstate);
4175 nfs4_put_stid(&dp->dl_stid);
4176 return ERR_PTR(status);
4181 static void nfsd4_open_deleg_none_ext(struct nfsd4_open *open, int status)
4183 open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE_EXT;
4184 if (status == -EAGAIN)
4185 open->op_why_no_deleg = WND4_CONTENTION;
4187 open->op_why_no_deleg = WND4_RESOURCE;
4188 switch (open->op_deleg_want) {
4189 case NFS4_SHARE_WANT_READ_DELEG:
4190 case NFS4_SHARE_WANT_WRITE_DELEG:
4191 case NFS4_SHARE_WANT_ANY_DELEG:
4193 case NFS4_SHARE_WANT_CANCEL:
4194 open->op_why_no_deleg = WND4_CANCELLED;
4196 case NFS4_SHARE_WANT_NO_DELEG:
4203 * Attempt to hand out a delegation.
4205 * Note we don't support write delegations, and won't until the vfs has
4206 * proper support for them.
4209 nfs4_open_delegation(struct svc_fh *fh, struct nfsd4_open *open,
4210 struct nfs4_ol_stateid *stp)
4212 struct nfs4_delegation *dp;
4213 struct nfs4_openowner *oo = openowner(stp->st_stateowner);
4214 struct nfs4_client *clp = stp->st_stid.sc_client;
4218 cb_up = nfsd4_cb_channel_good(oo->oo_owner.so_client);
4219 open->op_recall = 0;
4220 switch (open->op_claim_type) {
4221 case NFS4_OPEN_CLAIM_PREVIOUS:
4223 open->op_recall = 1;
4224 if (open->op_delegate_type != NFS4_OPEN_DELEGATE_READ)
4227 case NFS4_OPEN_CLAIM_NULL:
4228 case NFS4_OPEN_CLAIM_FH:
4230 * Let's not give out any delegations till everyone's
4231 * had the chance to reclaim theirs, *and* until
4232 * NLM locks have all been reclaimed:
4234 if (locks_in_grace(clp->net))
4236 if (!cb_up || !(oo->oo_flags & NFS4_OO_CONFIRMED))
4239 * Also, if the file was opened for write or
4240 * create, there's a good chance the client's
4241 * about to write to it, resulting in an
4242 * immediate recall (since we don't support
4243 * write delegations):
4245 if (open->op_share_access & NFS4_SHARE_ACCESS_WRITE)
4247 if (open->op_create == NFS4_OPEN_CREATE)
4253 dp = nfs4_set_delegation(clp, fh, stp->st_stid.sc_file, stp->st_clnt_odstate);
4257 memcpy(&open->op_delegate_stateid, &dp->dl_stid.sc_stateid, sizeof(dp->dl_stid.sc_stateid));
4259 dprintk("NFSD: delegation stateid=" STATEID_FMT "\n",
4260 STATEID_VAL(&dp->dl_stid.sc_stateid));
4261 open->op_delegate_type = NFS4_OPEN_DELEGATE_READ;
4262 nfs4_put_stid(&dp->dl_stid);
4265 open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE;
4266 if (open->op_claim_type == NFS4_OPEN_CLAIM_PREVIOUS &&
4267 open->op_delegate_type != NFS4_OPEN_DELEGATE_NONE) {
4268 dprintk("NFSD: WARNING: refusing delegation reclaim\n");
4269 open->op_recall = 1;
4272 /* 4.1 client asking for a delegation? */
4273 if (open->op_deleg_want)
4274 nfsd4_open_deleg_none_ext(open, status);
4278 static void nfsd4_deleg_xgrade_none_ext(struct nfsd4_open *open,
4279 struct nfs4_delegation *dp)
4281 if (open->op_deleg_want == NFS4_SHARE_WANT_READ_DELEG &&
4282 dp->dl_type == NFS4_OPEN_DELEGATE_WRITE) {
4283 open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE_EXT;
4284 open->op_why_no_deleg = WND4_NOT_SUPP_DOWNGRADE;
4285 } else if (open->op_deleg_want == NFS4_SHARE_WANT_WRITE_DELEG &&
4286 dp->dl_type == NFS4_OPEN_DELEGATE_WRITE) {
4287 open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE_EXT;
4288 open->op_why_no_deleg = WND4_NOT_SUPP_UPGRADE;
4290 /* Otherwise the client must be confused wanting a delegation
4291 * it already has, therefore we don't return
4292 * NFS4_OPEN_DELEGATE_NONE_EXT and reason.
4297 nfsd4_process_open2(struct svc_rqst *rqstp, struct svc_fh *current_fh, struct nfsd4_open *open)
4299 struct nfsd4_compoundres *resp = rqstp->rq_resp;
4300 struct nfs4_client *cl = open->op_openowner->oo_owner.so_client;
4301 struct nfs4_file *fp = NULL;
4302 struct nfs4_ol_stateid *stp = NULL;
4303 struct nfs4_ol_stateid *swapstp = NULL;
4304 struct nfs4_delegation *dp = NULL;
4308 * Lookup file; if found, lookup stateid and check open request,
4309 * and check for delegations in the process of being recalled.
4310 * If not found, create the nfs4_file struct
4312 fp = find_or_add_file(open->op_file, ¤t_fh->fh_handle);
4313 if (fp != open->op_file) {
4314 status = nfs4_check_deleg(cl, open, &dp);
4317 spin_lock(&fp->fi_lock);
4318 stp = nfsd4_find_existing_open(fp, open);
4319 spin_unlock(&fp->fi_lock);
4321 open->op_file = NULL;
4322 status = nfserr_bad_stateid;
4323 if (nfsd4_is_deleg_cur(open))
4328 * OPEN the file, or upgrade an existing OPEN.
4329 * If truncate fails, the OPEN fails.
4332 /* Stateid was found, this is an OPEN upgrade */
4333 down_read(&stp->st_rwsem);
4334 status = nfs4_upgrade_open(rqstp, fp, current_fh, stp, open);
4336 up_read(&stp->st_rwsem);
4341 open->op_stp = NULL;
4342 swapstp = init_open_stateid(stp, fp, open);
4344 nfs4_put_stid(&stp->st_stid);
4346 down_read(&stp->st_rwsem);
4347 status = nfs4_upgrade_open(rqstp, fp, current_fh,
4350 up_read(&stp->st_rwsem);
4355 down_read(&stp->st_rwsem);
4356 status = nfs4_get_vfs_file(rqstp, fp, current_fh, stp, open);
4358 up_read(&stp->st_rwsem);
4359 release_open_stateid(stp);
4363 stp->st_clnt_odstate = find_or_hash_clnt_odstate(fp,
4365 if (stp->st_clnt_odstate == open->op_odstate)
4366 open->op_odstate = NULL;
4369 nfs4_inc_and_copy_stateid(&open->op_stateid, &stp->st_stid);
4370 up_read(&stp->st_rwsem);
4372 if (nfsd4_has_session(&resp->cstate)) {
4373 if (open->op_deleg_want & NFS4_SHARE_WANT_NO_DELEG) {
4374 open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE_EXT;
4375 open->op_why_no_deleg = WND4_NOT_WANTED;
4381 * Attempt to hand out a delegation. No error return, because the
4382 * OPEN succeeds even if we fail.
4384 nfs4_open_delegation(current_fh, open, stp);
4388 dprintk("%s: stateid=" STATEID_FMT "\n", __func__,
4389 STATEID_VAL(&stp->st_stid.sc_stateid));
4391 /* 4.1 client trying to upgrade/downgrade delegation? */
4392 if (open->op_delegate_type == NFS4_OPEN_DELEGATE_NONE && dp &&
4393 open->op_deleg_want)
4394 nfsd4_deleg_xgrade_none_ext(open, dp);
4398 if (status == 0 && open->op_claim_type == NFS4_OPEN_CLAIM_PREVIOUS)
4399 open->op_openowner->oo_flags |= NFS4_OO_CONFIRMED;
4401 * To finish the open response, we just need to set the rflags.
4403 open->op_rflags = NFS4_OPEN_RESULT_LOCKTYPE_POSIX;
4404 if (!(open->op_openowner->oo_flags & NFS4_OO_CONFIRMED) &&
4405 !nfsd4_has_session(&resp->cstate))
4406 open->op_rflags |= NFS4_OPEN_RESULT_CONFIRM;
4408 nfs4_put_stid(&dp->dl_stid);
4410 nfs4_put_stid(&stp->st_stid);
4415 void nfsd4_cleanup_open_state(struct nfsd4_compound_state *cstate,
4416 struct nfsd4_open *open)
4418 if (open->op_openowner) {
4419 struct nfs4_stateowner *so = &open->op_openowner->oo_owner;
4421 nfsd4_cstate_assign_replay(cstate, so);
4422 nfs4_put_stateowner(so);
4425 kmem_cache_free(file_slab, open->op_file);
4427 nfs4_put_stid(&open->op_stp->st_stid);
4428 if (open->op_odstate)
4429 kmem_cache_free(odstate_slab, open->op_odstate);
4433 nfsd4_renew(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
4436 struct nfs4_client *clp;
4438 struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
4440 dprintk("process_renew(%08x/%08x): starting\n",
4441 clid->cl_boot, clid->cl_id);
4442 status = lookup_clientid(clid, cstate, nn);
4446 status = nfserr_cb_path_down;
4447 if (!list_empty(&clp->cl_delegations)
4448 && clp->cl_cb_state != NFSD4_CB_UP)
4456 nfsd4_end_grace(struct nfsd_net *nn)
4458 /* do nothing if grace period already ended */
4459 if (nn->grace_ended)
4462 dprintk("NFSD: end of grace period\n");
4463 nn->grace_ended = true;
4465 * If the server goes down again right now, an NFSv4
4466 * client will still be allowed to reclaim after it comes back up,
4467 * even if it hasn't yet had a chance to reclaim state this time.
4470 nfsd4_record_grace_done(nn);
4472 * At this point, NFSv4 clients can still reclaim. But if the
4473 * server crashes, any that have not yet reclaimed will be out
4474 * of luck on the next boot.
4476 * (NFSv4.1+ clients are considered to have reclaimed once they
4477 * call RECLAIM_COMPLETE. NFSv4.0 clients are considered to
4478 * have reclaimed after their first OPEN.)
4480 locks_end_grace(&nn->nfsd4_manager);
4482 * At this point, and once lockd and/or any other containers
4483 * exit their grace period, further reclaims will fail and
4484 * regular locking can resume.
4489 nfs4_laundromat(struct nfsd_net *nn)
4491 struct nfs4_client *clp;
4492 struct nfs4_openowner *oo;
4493 struct nfs4_delegation *dp;
4494 struct nfs4_ol_stateid *stp;
4495 struct list_head *pos, *next, reaplist;
4496 time_t cutoff = get_seconds() - nn->nfsd4_lease;
4497 time_t t, new_timeo = nn->nfsd4_lease;
4499 dprintk("NFSD: laundromat service - starting\n");
4500 nfsd4_end_grace(nn);
4501 INIT_LIST_HEAD(&reaplist);
4502 spin_lock(&nn->client_lock);
4503 list_for_each_safe(pos, next, &nn->client_lru) {
4504 clp = list_entry(pos, struct nfs4_client, cl_lru);
4505 if (time_after((unsigned long)clp->cl_time, (unsigned long)cutoff)) {
4506 t = clp->cl_time - cutoff;
4507 new_timeo = min(new_timeo, t);
4510 if (mark_client_expired_locked(clp)) {
4511 dprintk("NFSD: client in use (clientid %08x)\n",
4512 clp->cl_clientid.cl_id);
4515 list_add(&clp->cl_lru, &reaplist);
4517 spin_unlock(&nn->client_lock);
4518 list_for_each_safe(pos, next, &reaplist) {
4519 clp = list_entry(pos, struct nfs4_client, cl_lru);
4520 dprintk("NFSD: purging unused client (clientid %08x)\n",
4521 clp->cl_clientid.cl_id);
4522 list_del_init(&clp->cl_lru);
4525 spin_lock(&state_lock);
4526 list_for_each_safe(pos, next, &nn->del_recall_lru) {
4527 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
4528 if (time_after((unsigned long)dp->dl_time, (unsigned long)cutoff)) {
4529 t = dp->dl_time - cutoff;
4530 new_timeo = min(new_timeo, t);
4533 WARN_ON(!unhash_delegation_locked(dp));
4534 list_add(&dp->dl_recall_lru, &reaplist);
4536 spin_unlock(&state_lock);
4537 while (!list_empty(&reaplist)) {
4538 dp = list_first_entry(&reaplist, struct nfs4_delegation,
4540 list_del_init(&dp->dl_recall_lru);
4541 revoke_delegation(dp);
4544 spin_lock(&nn->client_lock);
4545 while (!list_empty(&nn->close_lru)) {
4546 oo = list_first_entry(&nn->close_lru, struct nfs4_openowner,
4548 if (time_after((unsigned long)oo->oo_time,
4549 (unsigned long)cutoff)) {
4550 t = oo->oo_time - cutoff;
4551 new_timeo = min(new_timeo, t);
4554 list_del_init(&oo->oo_close_lru);
4555 stp = oo->oo_last_closed_stid;
4556 oo->oo_last_closed_stid = NULL;
4557 spin_unlock(&nn->client_lock);
4558 nfs4_put_stid(&stp->st_stid);
4559 spin_lock(&nn->client_lock);
4561 spin_unlock(&nn->client_lock);
4563 new_timeo = max_t(time_t, new_timeo, NFSD_LAUNDROMAT_MINTIMEOUT);
4567 static struct workqueue_struct *laundry_wq;
4568 static void laundromat_main(struct work_struct *);
4571 laundromat_main(struct work_struct *laundry)
4574 struct delayed_work *dwork = to_delayed_work(laundry);
4575 struct nfsd_net *nn = container_of(dwork, struct nfsd_net,
4578 t = nfs4_laundromat(nn);
4579 dprintk("NFSD: laundromat_main - sleeping for %ld seconds\n", t);
4580 queue_delayed_work(laundry_wq, &nn->laundromat_work, t*HZ);
4583 static inline __be32 nfs4_check_fh(struct svc_fh *fhp, struct nfs4_stid *stp)
4585 if (!fh_match(&fhp->fh_handle, &stp->sc_file->fi_fhandle))
4586 return nfserr_bad_stateid;
4591 access_permit_read(struct nfs4_ol_stateid *stp)
4593 return test_access(NFS4_SHARE_ACCESS_READ, stp) ||
4594 test_access(NFS4_SHARE_ACCESS_BOTH, stp) ||
4595 test_access(NFS4_SHARE_ACCESS_WRITE, stp);
4599 access_permit_write(struct nfs4_ol_stateid *stp)
4601 return test_access(NFS4_SHARE_ACCESS_WRITE, stp) ||
4602 test_access(NFS4_SHARE_ACCESS_BOTH, stp);
4606 __be32 nfs4_check_openmode(struct nfs4_ol_stateid *stp, int flags)
4608 __be32 status = nfserr_openmode;
4610 /* For lock stateid's, we test the parent open, not the lock: */
4611 if (stp->st_openstp)
4612 stp = stp->st_openstp;
4613 if ((flags & WR_STATE) && !access_permit_write(stp))
4615 if ((flags & RD_STATE) && !access_permit_read(stp))
4622 static inline __be32
4623 check_special_stateids(struct net *net, svc_fh *current_fh, stateid_t *stateid, int flags)
4625 if (ONE_STATEID(stateid) && (flags & RD_STATE))
4627 else if (opens_in_grace(net)) {
4628 /* Answer in remaining cases depends on existence of
4629 * conflicting state; so we must wait out the grace period. */
4630 return nfserr_grace;
4631 } else if (flags & WR_STATE)
4632 return nfs4_share_conflict(current_fh,
4633 NFS4_SHARE_DENY_WRITE);
4634 else /* (flags & RD_STATE) && ZERO_STATEID(stateid) */
4635 return nfs4_share_conflict(current_fh,
4636 NFS4_SHARE_DENY_READ);
4640 * Allow READ/WRITE during grace period on recovered state only for files
4641 * that are not able to provide mandatory locking.
4644 grace_disallows_io(struct net *net, struct inode *inode)
4646 return opens_in_grace(net) && mandatory_lock(inode);
4649 /* Returns true iff a is later than b: */
4650 static bool stateid_generation_after(stateid_t *a, stateid_t *b)
4652 return (s32)(a->si_generation - b->si_generation) > 0;
4655 static __be32 check_stateid_generation(stateid_t *in, stateid_t *ref, bool has_session)
4658 * When sessions are used the stateid generation number is ignored
4661 if (has_session && in->si_generation == 0)
4664 if (in->si_generation == ref->si_generation)
4667 /* If the client sends us a stateid from the future, it's buggy: */
4668 if (stateid_generation_after(in, ref))
4669 return nfserr_bad_stateid;
4671 * However, we could see a stateid from the past, even from a
4672 * non-buggy client. For example, if the client sends a lock
4673 * while some IO is outstanding, the lock may bump si_generation
4674 * while the IO is still in flight. The client could avoid that
4675 * situation by waiting for responses on all the IO requests,
4676 * but better performance may result in retrying IO that
4677 * receives an old_stateid error if requests are rarely
4678 * reordered in flight:
4680 return nfserr_old_stateid;
4683 static __be32 nfsd4_check_openowner_confirmed(struct nfs4_ol_stateid *ols)
4685 if (ols->st_stateowner->so_is_open_owner &&
4686 !(openowner(ols->st_stateowner)->oo_flags & NFS4_OO_CONFIRMED))
4687 return nfserr_bad_stateid;
4691 static __be32 nfsd4_validate_stateid(struct nfs4_client *cl, stateid_t *stateid)
4693 struct nfs4_stid *s;
4694 __be32 status = nfserr_bad_stateid;
4696 if (ZERO_STATEID(stateid) || ONE_STATEID(stateid))
4698 /* Client debugging aid. */
4699 if (!same_clid(&stateid->si_opaque.so_clid, &cl->cl_clientid)) {
4700 char addr_str[INET6_ADDRSTRLEN];
4701 rpc_ntop((struct sockaddr *)&cl->cl_addr, addr_str,
4703 pr_warn_ratelimited("NFSD: client %s testing state ID "
4704 "with incorrect client ID\n", addr_str);
4707 spin_lock(&cl->cl_lock);
4708 s = find_stateid_locked(cl, stateid);
4711 status = check_stateid_generation(stateid, &s->sc_stateid, 1);
4714 switch (s->sc_type) {
4715 case NFS4_DELEG_STID:
4718 case NFS4_REVOKED_DELEG_STID:
4719 status = nfserr_deleg_revoked;
4721 case NFS4_OPEN_STID:
4722 case NFS4_LOCK_STID:
4723 status = nfsd4_check_openowner_confirmed(openlockstateid(s));
4726 printk("unknown stateid type %x\n", s->sc_type);
4728 case NFS4_CLOSED_STID:
4729 case NFS4_CLOSED_DELEG_STID:
4730 status = nfserr_bad_stateid;
4733 spin_unlock(&cl->cl_lock);
4738 nfsd4_lookup_stateid(struct nfsd4_compound_state *cstate,
4739 stateid_t *stateid, unsigned char typemask,
4740 struct nfs4_stid **s, struct nfsd_net *nn)
4744 if (ZERO_STATEID(stateid) || ONE_STATEID(stateid))
4745 return nfserr_bad_stateid;
4746 status = lookup_clientid(&stateid->si_opaque.so_clid, cstate, nn);
4747 if (status == nfserr_stale_clientid) {
4748 if (cstate->session)
4749 return nfserr_bad_stateid;
4750 return nfserr_stale_stateid;
4754 *s = find_stateid_by_type(cstate->clp, stateid, typemask);
4756 return nfserr_bad_stateid;
4760 static struct file *
4761 nfs4_find_file(struct nfs4_stid *s, int flags)
4766 switch (s->sc_type) {
4767 case NFS4_DELEG_STID:
4768 if (WARN_ON_ONCE(!s->sc_file->fi_deleg_file))
4770 return get_file(s->sc_file->fi_deleg_file);
4771 case NFS4_OPEN_STID:
4772 case NFS4_LOCK_STID:
4773 if (flags & RD_STATE)
4774 return find_readable_file(s->sc_file);
4776 return find_writeable_file(s->sc_file);
4784 nfs4_check_olstateid(struct svc_fh *fhp, struct nfs4_ol_stateid *ols, int flags)
4788 status = nfsd4_check_openowner_confirmed(ols);
4791 return nfs4_check_openmode(ols, flags);
4795 nfs4_check_file(struct svc_rqst *rqstp, struct svc_fh *fhp, struct nfs4_stid *s,
4796 struct file **filpp, bool *tmp_file, int flags)
4798 int acc = (flags & RD_STATE) ? NFSD_MAY_READ : NFSD_MAY_WRITE;
4802 file = nfs4_find_file(s, flags);
4804 status = nfsd_permission(rqstp, fhp->fh_export, fhp->fh_dentry,
4805 acc | NFSD_MAY_OWNER_OVERRIDE);
4813 status = nfsd_open(rqstp, fhp, S_IFREG, acc, filpp);
4825 * Checks for stateid operations
4828 nfs4_preprocess_stateid_op(struct svc_rqst *rqstp,
4829 struct nfsd4_compound_state *cstate, struct svc_fh *fhp,
4830 stateid_t *stateid, int flags, struct file **filpp, bool *tmp_file)
4832 struct inode *ino = d_inode(fhp->fh_dentry);
4833 struct net *net = SVC_NET(rqstp);
4834 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
4835 struct nfs4_stid *s = NULL;
4843 if (grace_disallows_io(net, ino))
4844 return nfserr_grace;
4846 if (ZERO_STATEID(stateid) || ONE_STATEID(stateid)) {
4847 status = check_special_stateids(net, fhp, stateid, flags);
4851 status = nfsd4_lookup_stateid(cstate, stateid,
4852 NFS4_DELEG_STID|NFS4_OPEN_STID|NFS4_LOCK_STID,
4856 status = check_stateid_generation(stateid, &s->sc_stateid,
4857 nfsd4_has_session(cstate));
4861 switch (s->sc_type) {
4862 case NFS4_DELEG_STID:
4863 status = nfs4_check_delegmode(delegstateid(s), flags);
4865 case NFS4_OPEN_STID:
4866 case NFS4_LOCK_STID:
4867 status = nfs4_check_olstateid(fhp, openlockstateid(s), flags);
4870 status = nfserr_bad_stateid;
4875 status = nfs4_check_fh(fhp, s);
4878 if (!status && filpp)
4879 status = nfs4_check_file(rqstp, fhp, s, filpp, tmp_file, flags);
4887 * Test if the stateid is valid
4890 nfsd4_test_stateid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
4891 struct nfsd4_test_stateid *test_stateid)
4893 struct nfsd4_test_stateid_id *stateid;
4894 struct nfs4_client *cl = cstate->session->se_client;
4896 list_for_each_entry(stateid, &test_stateid->ts_stateid_list, ts_id_list)
4897 stateid->ts_id_status =
4898 nfsd4_validate_stateid(cl, &stateid->ts_id_stateid);
4904 nfsd4_free_stateid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
4905 struct nfsd4_free_stateid *free_stateid)
4907 stateid_t *stateid = &free_stateid->fr_stateid;
4908 struct nfs4_stid *s;
4909 struct nfs4_delegation *dp;
4910 struct nfs4_ol_stateid *stp;
4911 struct nfs4_client *cl = cstate->session->se_client;
4912 __be32 ret = nfserr_bad_stateid;
4914 spin_lock(&cl->cl_lock);
4915 s = find_stateid_locked(cl, stateid);
4918 switch (s->sc_type) {
4919 case NFS4_DELEG_STID:
4920 ret = nfserr_locks_held;
4922 case NFS4_OPEN_STID:
4923 ret = check_stateid_generation(stateid, &s->sc_stateid, 1);
4926 ret = nfserr_locks_held;
4928 case NFS4_LOCK_STID:
4929 ret = check_stateid_generation(stateid, &s->sc_stateid, 1);
4932 stp = openlockstateid(s);
4933 ret = nfserr_locks_held;
4934 if (check_for_locks(stp->st_stid.sc_file,
4935 lockowner(stp->st_stateowner)))
4937 WARN_ON(!unhash_lock_stateid(stp));
4938 spin_unlock(&cl->cl_lock);
4942 case NFS4_REVOKED_DELEG_STID:
4943 dp = delegstateid(s);
4944 list_del_init(&dp->dl_recall_lru);
4945 spin_unlock(&cl->cl_lock);
4949 /* Default falls through and returns nfserr_bad_stateid */
4952 spin_unlock(&cl->cl_lock);
4960 return (type == NFS4_READW_LT || type == NFS4_READ_LT) ?
4961 RD_STATE : WR_STATE;
4964 static __be32 nfs4_seqid_op_checks(struct nfsd4_compound_state *cstate, stateid_t *stateid, u32 seqid, struct nfs4_ol_stateid *stp)
4966 struct svc_fh *current_fh = &cstate->current_fh;
4967 struct nfs4_stateowner *sop = stp->st_stateowner;
4970 status = nfsd4_check_seqid(cstate, sop, seqid);
4973 if (stp->st_stid.sc_type == NFS4_CLOSED_STID
4974 || stp->st_stid.sc_type == NFS4_REVOKED_DELEG_STID)
4976 * "Closed" stateid's exist *only* to return
4977 * nfserr_replay_me from the previous step, and
4978 * revoked delegations are kept only for free_stateid.
4980 return nfserr_bad_stateid;
4981 down_write(&stp->st_rwsem);
4982 status = check_stateid_generation(stateid, &stp->st_stid.sc_stateid, nfsd4_has_session(cstate));
4983 if (status == nfs_ok)
4984 status = nfs4_check_fh(current_fh, &stp->st_stid);
4985 if (status != nfs_ok)
4986 up_write(&stp->st_rwsem);
4991 * Checks for sequence id mutating operations.
4994 nfs4_preprocess_seqid_op(struct nfsd4_compound_state *cstate, u32 seqid,
4995 stateid_t *stateid, char typemask,
4996 struct nfs4_ol_stateid **stpp,
4997 struct nfsd_net *nn)
5000 struct nfs4_stid *s;
5001 struct nfs4_ol_stateid *stp = NULL;
5003 dprintk("NFSD: %s: seqid=%d stateid = " STATEID_FMT "\n", __func__,
5004 seqid, STATEID_VAL(stateid));
5007 status = nfsd4_lookup_stateid(cstate, stateid, typemask, &s, nn);
5010 stp = openlockstateid(s);
5011 nfsd4_cstate_assign_replay(cstate, stp->st_stateowner);
5013 status = nfs4_seqid_op_checks(cstate, stateid, seqid, stp);
5017 nfs4_put_stid(&stp->st_stid);
5021 static __be32 nfs4_preprocess_confirmed_seqid_op(struct nfsd4_compound_state *cstate, u32 seqid,
5022 stateid_t *stateid, struct nfs4_ol_stateid **stpp, struct nfsd_net *nn)
5025 struct nfs4_openowner *oo;
5026 struct nfs4_ol_stateid *stp;
5028 status = nfs4_preprocess_seqid_op(cstate, seqid, stateid,
5029 NFS4_OPEN_STID, &stp, nn);
5032 oo = openowner(stp->st_stateowner);
5033 if (!(oo->oo_flags & NFS4_OO_CONFIRMED)) {
5034 up_write(&stp->st_rwsem);
5035 nfs4_put_stid(&stp->st_stid);
5036 return nfserr_bad_stateid;
5043 nfsd4_open_confirm(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
5044 struct nfsd4_open_confirm *oc)
5047 struct nfs4_openowner *oo;
5048 struct nfs4_ol_stateid *stp;
5049 struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
5051 dprintk("NFSD: nfsd4_open_confirm on file %pd\n",
5052 cstate->current_fh.fh_dentry);
5054 status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0);
5058 status = nfs4_preprocess_seqid_op(cstate,
5059 oc->oc_seqid, &oc->oc_req_stateid,
5060 NFS4_OPEN_STID, &stp, nn);
5063 oo = openowner(stp->st_stateowner);
5064 status = nfserr_bad_stateid;
5065 if (oo->oo_flags & NFS4_OO_CONFIRMED) {
5066 up_write(&stp->st_rwsem);
5069 oo->oo_flags |= NFS4_OO_CONFIRMED;
5070 nfs4_inc_and_copy_stateid(&oc->oc_resp_stateid, &stp->st_stid);
5071 up_write(&stp->st_rwsem);
5072 dprintk("NFSD: %s: success, seqid=%d stateid=" STATEID_FMT "\n",
5073 __func__, oc->oc_seqid, STATEID_VAL(&stp->st_stid.sc_stateid));
5075 nfsd4_client_record_create(oo->oo_owner.so_client);
5078 nfs4_put_stid(&stp->st_stid);
5080 nfsd4_bump_seqid(cstate, status);
5084 static inline void nfs4_stateid_downgrade_bit(struct nfs4_ol_stateid *stp, u32 access)
5086 if (!test_access(access, stp))
5088 nfs4_file_put_access(stp->st_stid.sc_file, access);
5089 clear_access(access, stp);
5092 static inline void nfs4_stateid_downgrade(struct nfs4_ol_stateid *stp, u32 to_access)
5094 switch (to_access) {
5095 case NFS4_SHARE_ACCESS_READ:
5096 nfs4_stateid_downgrade_bit(stp, NFS4_SHARE_ACCESS_WRITE);
5097 nfs4_stateid_downgrade_bit(stp, NFS4_SHARE_ACCESS_BOTH);
5099 case NFS4_SHARE_ACCESS_WRITE:
5100 nfs4_stateid_downgrade_bit(stp, NFS4_SHARE_ACCESS_READ);
5101 nfs4_stateid_downgrade_bit(stp, NFS4_SHARE_ACCESS_BOTH);
5103 case NFS4_SHARE_ACCESS_BOTH:
5111 nfsd4_open_downgrade(struct svc_rqst *rqstp,
5112 struct nfsd4_compound_state *cstate,
5113 struct nfsd4_open_downgrade *od)
5116 struct nfs4_ol_stateid *stp;
5117 struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
5119 dprintk("NFSD: nfsd4_open_downgrade on file %pd\n",
5120 cstate->current_fh.fh_dentry);
5122 /* We don't yet support WANT bits: */
5123 if (od->od_deleg_want)
5124 dprintk("NFSD: %s: od_deleg_want=0x%x ignored\n", __func__,
5127 status = nfs4_preprocess_confirmed_seqid_op(cstate, od->od_seqid,
5128 &od->od_stateid, &stp, nn);
5131 status = nfserr_inval;
5132 if (!test_access(od->od_share_access, stp)) {
5133 dprintk("NFSD: access not a subset of current bitmap: 0x%hhx, input access=%08x\n",
5134 stp->st_access_bmap, od->od_share_access);
5137 if (!test_deny(od->od_share_deny, stp)) {
5138 dprintk("NFSD: deny not a subset of current bitmap: 0x%hhx, input deny=%08x\n",
5139 stp->st_deny_bmap, od->od_share_deny);
5142 nfs4_stateid_downgrade(stp, od->od_share_access);
5143 reset_union_bmap_deny(od->od_share_deny, stp);
5144 nfs4_inc_and_copy_stateid(&od->od_stateid, &stp->st_stid);
5147 up_write(&stp->st_rwsem);
5148 nfs4_put_stid(&stp->st_stid);
5150 nfsd4_bump_seqid(cstate, status);
5154 static void nfsd4_close_open_stateid(struct nfs4_ol_stateid *s)
5156 struct nfs4_client *clp = s->st_stid.sc_client;
5158 LIST_HEAD(reaplist);
5160 s->st_stid.sc_type = NFS4_CLOSED_STID;
5161 spin_lock(&clp->cl_lock);
5162 unhashed = unhash_open_stateid(s, &reaplist);
5164 if (clp->cl_minorversion) {
5166 put_ol_stateid_locked(s, &reaplist);
5167 spin_unlock(&clp->cl_lock);
5168 free_ol_stateid_reaplist(&reaplist);
5170 spin_unlock(&clp->cl_lock);
5171 free_ol_stateid_reaplist(&reaplist);
5173 move_to_close_lru(s, clp->net);
5178 * nfs4_unlock_state() called after encode
5181 nfsd4_close(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
5182 struct nfsd4_close *close)
5185 struct nfs4_ol_stateid *stp;
5186 struct net *net = SVC_NET(rqstp);
5187 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
5189 dprintk("NFSD: nfsd4_close on file %pd\n",
5190 cstate->current_fh.fh_dentry);
5192 status = nfs4_preprocess_seqid_op(cstate, close->cl_seqid,
5194 NFS4_OPEN_STID|NFS4_CLOSED_STID,
5196 nfsd4_bump_seqid(cstate, status);
5199 nfs4_inc_and_copy_stateid(&close->cl_stateid, &stp->st_stid);
5200 up_write(&stp->st_rwsem);
5202 nfsd4_close_open_stateid(stp);
5204 /* put reference from nfs4_preprocess_seqid_op */
5205 nfs4_put_stid(&stp->st_stid);
5211 nfsd4_delegreturn(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
5212 struct nfsd4_delegreturn *dr)
5214 struct nfs4_delegation *dp;
5215 stateid_t *stateid = &dr->dr_stateid;
5216 struct nfs4_stid *s;
5218 struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
5220 if ((status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0)))
5223 status = nfsd4_lookup_stateid(cstate, stateid, NFS4_DELEG_STID, &s, nn);
5226 dp = delegstateid(s);
5227 status = check_stateid_generation(stateid, &dp->dl_stid.sc_stateid, nfsd4_has_session(cstate));
5231 destroy_delegation(dp);
5233 nfs4_put_stid(&dp->dl_stid);
5239 end_offset(u64 start, u64 len)
5244 return end >= start ? end: NFS4_MAX_UINT64;
5247 /* last octet in a range */
5249 last_byte_offset(u64 start, u64 len)
5255 return end > start ? end - 1: NFS4_MAX_UINT64;
5259 * TODO: Linux file offsets are _signed_ 64-bit quantities, which means that
5260 * we can't properly handle lock requests that go beyond the (2^63 - 1)-th
5261 * byte, because of sign extension problems. Since NFSv4 calls for 64-bit
5262 * locking, this prevents us from being completely protocol-compliant. The
5263 * real solution to this problem is to start using unsigned file offsets in
5264 * the VFS, but this is a very deep change!
5267 nfs4_transform_lock_offset(struct file_lock *lock)
5269 if (lock->fl_start < 0)
5270 lock->fl_start = OFFSET_MAX;
5271 if (lock->fl_end < 0)
5272 lock->fl_end = OFFSET_MAX;
5276 nfsd4_fl_get_owner(fl_owner_t owner)
5278 struct nfs4_lockowner *lo = (struct nfs4_lockowner *)owner;
5280 nfs4_get_stateowner(&lo->lo_owner);
5285 nfsd4_fl_put_owner(fl_owner_t owner)
5287 struct nfs4_lockowner *lo = (struct nfs4_lockowner *)owner;
5290 nfs4_put_stateowner(&lo->lo_owner);
5293 static const struct lock_manager_operations nfsd_posix_mng_ops = {
5294 .lm_get_owner = nfsd4_fl_get_owner,
5295 .lm_put_owner = nfsd4_fl_put_owner,
5299 nfs4_set_lock_denied(struct file_lock *fl, struct nfsd4_lock_denied *deny)
5301 struct nfs4_lockowner *lo;
5303 if (fl->fl_lmops == &nfsd_posix_mng_ops) {
5304 lo = (struct nfs4_lockowner *) fl->fl_owner;
5305 deny->ld_owner.data = kmemdup(lo->lo_owner.so_owner.data,
5306 lo->lo_owner.so_owner.len, GFP_KERNEL);
5307 if (!deny->ld_owner.data)
5308 /* We just don't care that much */
5310 deny->ld_owner.len = lo->lo_owner.so_owner.len;
5311 deny->ld_clientid = lo->lo_owner.so_client->cl_clientid;
5314 deny->ld_owner.len = 0;
5315 deny->ld_owner.data = NULL;
5316 deny->ld_clientid.cl_boot = 0;
5317 deny->ld_clientid.cl_id = 0;
5319 deny->ld_start = fl->fl_start;
5320 deny->ld_length = NFS4_MAX_UINT64;
5321 if (fl->fl_end != NFS4_MAX_UINT64)
5322 deny->ld_length = fl->fl_end - fl->fl_start + 1;
5323 deny->ld_type = NFS4_READ_LT;
5324 if (fl->fl_type != F_RDLCK)
5325 deny->ld_type = NFS4_WRITE_LT;
5328 static struct nfs4_lockowner *
5329 find_lockowner_str_locked(struct nfs4_client *clp, struct xdr_netobj *owner)
5331 unsigned int strhashval = ownerstr_hashval(owner);
5332 struct nfs4_stateowner *so;
5334 lockdep_assert_held(&clp->cl_lock);
5336 list_for_each_entry(so, &clp->cl_ownerstr_hashtbl[strhashval],
5338 if (so->so_is_open_owner)
5340 if (same_owner_str(so, owner))
5341 return lockowner(nfs4_get_stateowner(so));
5346 static struct nfs4_lockowner *
5347 find_lockowner_str(struct nfs4_client *clp, struct xdr_netobj *owner)
5349 struct nfs4_lockowner *lo;
5351 spin_lock(&clp->cl_lock);
5352 lo = find_lockowner_str_locked(clp, owner);
5353 spin_unlock(&clp->cl_lock);
5357 static void nfs4_unhash_lockowner(struct nfs4_stateowner *sop)
5359 unhash_lockowner_locked(lockowner(sop));
5362 static void nfs4_free_lockowner(struct nfs4_stateowner *sop)
5364 struct nfs4_lockowner *lo = lockowner(sop);
5366 kmem_cache_free(lockowner_slab, lo);
5369 static const struct nfs4_stateowner_operations lockowner_ops = {
5370 .so_unhash = nfs4_unhash_lockowner,
5371 .so_free = nfs4_free_lockowner,
5375 * Alloc a lock owner structure.
5376 * Called in nfsd4_lock - therefore, OPEN and OPEN_CONFIRM (if needed) has
5379 * strhashval = ownerstr_hashval
5381 static struct nfs4_lockowner *
5382 alloc_init_lock_stateowner(unsigned int strhashval, struct nfs4_client *clp,
5383 struct nfs4_ol_stateid *open_stp,
5384 struct nfsd4_lock *lock)
5386 struct nfs4_lockowner *lo, *ret;
5388 lo = alloc_stateowner(lockowner_slab, &lock->lk_new_owner, clp);
5391 INIT_LIST_HEAD(&lo->lo_owner.so_stateids);
5392 lo->lo_owner.so_is_open_owner = 0;
5393 lo->lo_owner.so_seqid = lock->lk_new_lock_seqid;
5394 lo->lo_owner.so_ops = &lockowner_ops;
5395 spin_lock(&clp->cl_lock);
5396 ret = find_lockowner_str_locked(clp, &lock->lk_new_owner);
5398 list_add(&lo->lo_owner.so_strhash,
5399 &clp->cl_ownerstr_hashtbl[strhashval]);
5402 nfs4_free_stateowner(&lo->lo_owner);
5404 spin_unlock(&clp->cl_lock);
5409 init_lock_stateid(struct nfs4_ol_stateid *stp, struct nfs4_lockowner *lo,
5410 struct nfs4_file *fp, struct inode *inode,
5411 struct nfs4_ol_stateid *open_stp)
5413 struct nfs4_client *clp = lo->lo_owner.so_client;
5415 lockdep_assert_held(&clp->cl_lock);
5417 atomic_inc(&stp->st_stid.sc_count);
5418 stp->st_stid.sc_type = NFS4_LOCK_STID;
5419 stp->st_stateowner = nfs4_get_stateowner(&lo->lo_owner);
5421 stp->st_stid.sc_file = fp;
5422 stp->st_stid.sc_free = nfs4_free_lock_stateid;
5423 stp->st_access_bmap = 0;
5424 stp->st_deny_bmap = open_stp->st_deny_bmap;
5425 stp->st_openstp = open_stp;
5426 init_rwsem(&stp->st_rwsem);
5427 list_add(&stp->st_locks, &open_stp->st_locks);
5428 list_add(&stp->st_perstateowner, &lo->lo_owner.so_stateids);
5429 spin_lock(&fp->fi_lock);
5430 list_add(&stp->st_perfile, &fp->fi_stateids);
5431 spin_unlock(&fp->fi_lock);
5434 static struct nfs4_ol_stateid *
5435 find_lock_stateid(struct nfs4_lockowner *lo, struct nfs4_file *fp)
5437 struct nfs4_ol_stateid *lst;
5438 struct nfs4_client *clp = lo->lo_owner.so_client;
5440 lockdep_assert_held(&clp->cl_lock);
5442 list_for_each_entry(lst, &lo->lo_owner.so_stateids, st_perstateowner) {
5443 if (lst->st_stid.sc_file == fp) {
5444 atomic_inc(&lst->st_stid.sc_count);
5451 static struct nfs4_ol_stateid *
5452 find_or_create_lock_stateid(struct nfs4_lockowner *lo, struct nfs4_file *fi,
5453 struct inode *inode, struct nfs4_ol_stateid *ost,
5456 struct nfs4_stid *ns = NULL;
5457 struct nfs4_ol_stateid *lst;
5458 struct nfs4_openowner *oo = openowner(ost->st_stateowner);
5459 struct nfs4_client *clp = oo->oo_owner.so_client;
5461 spin_lock(&clp->cl_lock);
5462 lst = find_lock_stateid(lo, fi);
5464 spin_unlock(&clp->cl_lock);
5465 ns = nfs4_alloc_stid(clp, stateid_slab);
5469 spin_lock(&clp->cl_lock);
5470 lst = find_lock_stateid(lo, fi);
5472 lst = openlockstateid(ns);
5473 init_lock_stateid(lst, lo, fi, inode, ost);
5478 spin_unlock(&clp->cl_lock);
5485 check_lock_length(u64 offset, u64 length)
5487 return ((length == 0) || ((length != NFS4_MAX_UINT64) &&
5488 (length > ~offset)));
5491 static void get_lock_access(struct nfs4_ol_stateid *lock_stp, u32 access)
5493 struct nfs4_file *fp = lock_stp->st_stid.sc_file;
5495 lockdep_assert_held(&fp->fi_lock);
5497 if (test_access(access, lock_stp))
5499 __nfs4_file_get_access(fp, access);
5500 set_access(access, lock_stp);
5504 lookup_or_create_lock_state(struct nfsd4_compound_state *cstate,
5505 struct nfs4_ol_stateid *ost,
5506 struct nfsd4_lock *lock,
5507 struct nfs4_ol_stateid **lst, bool *new)
5510 struct nfs4_file *fi = ost->st_stid.sc_file;
5511 struct nfs4_openowner *oo = openowner(ost->st_stateowner);
5512 struct nfs4_client *cl = oo->oo_owner.so_client;
5513 struct inode *inode = d_inode(cstate->current_fh.fh_dentry);
5514 struct nfs4_lockowner *lo;
5515 unsigned int strhashval;
5517 lo = find_lockowner_str(cl, &lock->lk_new_owner);
5519 strhashval = ownerstr_hashval(&lock->lk_new_owner);
5520 lo = alloc_init_lock_stateowner(strhashval, cl, ost, lock);
5522 return nfserr_jukebox;
5524 /* with an existing lockowner, seqids must be the same */
5525 status = nfserr_bad_seqid;
5526 if (!cstate->minorversion &&
5527 lock->lk_new_lock_seqid != lo->lo_owner.so_seqid)
5531 *lst = find_or_create_lock_stateid(lo, fi, inode, ost, new);
5533 status = nfserr_jukebox;
5538 nfs4_put_stateowner(&lo->lo_owner);
5546 nfsd4_lock(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
5547 struct nfsd4_lock *lock)
5549 struct nfs4_openowner *open_sop = NULL;
5550 struct nfs4_lockowner *lock_sop = NULL;
5551 struct nfs4_ol_stateid *lock_stp = NULL;
5552 struct nfs4_ol_stateid *open_stp = NULL;
5553 struct nfs4_file *fp;
5554 struct file *filp = NULL;
5555 struct file_lock *file_lock = NULL;
5556 struct file_lock *conflock = NULL;
5561 struct net *net = SVC_NET(rqstp);
5562 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
5564 dprintk("NFSD: nfsd4_lock: start=%Ld length=%Ld\n",
5565 (long long) lock->lk_offset,
5566 (long long) lock->lk_length);
5568 if (check_lock_length(lock->lk_offset, lock->lk_length))
5569 return nfserr_inval;
5571 if ((status = fh_verify(rqstp, &cstate->current_fh,
5572 S_IFREG, NFSD_MAY_LOCK))) {
5573 dprintk("NFSD: nfsd4_lock: permission denied!\n");
5577 if (lock->lk_is_new) {
5578 if (nfsd4_has_session(cstate))
5579 /* See rfc 5661 18.10.3: given clientid is ignored: */
5580 memcpy(&lock->lk_new_clientid,
5581 &cstate->session->se_client->cl_clientid,
5582 sizeof(clientid_t));
5584 status = nfserr_stale_clientid;
5585 if (STALE_CLIENTID(&lock->lk_new_clientid, nn))
5588 /* validate and update open stateid and open seqid */
5589 status = nfs4_preprocess_confirmed_seqid_op(cstate,
5590 lock->lk_new_open_seqid,
5591 &lock->lk_new_open_stateid,
5595 up_write(&open_stp->st_rwsem);
5596 open_sop = openowner(open_stp->st_stateowner);
5597 status = nfserr_bad_stateid;
5598 if (!same_clid(&open_sop->oo_owner.so_client->cl_clientid,
5599 &lock->lk_new_clientid))
5601 status = lookup_or_create_lock_state(cstate, open_stp, lock,
5603 if (status == nfs_ok)
5604 down_write(&lock_stp->st_rwsem);
5606 status = nfs4_preprocess_seqid_op(cstate,
5607 lock->lk_old_lock_seqid,
5608 &lock->lk_old_lock_stateid,
5609 NFS4_LOCK_STID, &lock_stp, nn);
5613 lock_sop = lockowner(lock_stp->st_stateowner);
5615 lkflg = setlkflg(lock->lk_type);
5616 status = nfs4_check_openmode(lock_stp, lkflg);
5620 status = nfserr_grace;
5621 if (locks_in_grace(net) && !lock->lk_reclaim)
5623 status = nfserr_no_grace;
5624 if (!locks_in_grace(net) && lock->lk_reclaim)
5627 file_lock = locks_alloc_lock();
5629 dprintk("NFSD: %s: unable to allocate lock!\n", __func__);
5630 status = nfserr_jukebox;
5634 fp = lock_stp->st_stid.sc_file;
5635 switch (lock->lk_type) {
5638 spin_lock(&fp->fi_lock);
5639 filp = find_readable_file_locked(fp);
5641 get_lock_access(lock_stp, NFS4_SHARE_ACCESS_READ);
5642 spin_unlock(&fp->fi_lock);
5643 file_lock->fl_type = F_RDLCK;
5646 case NFS4_WRITEW_LT:
5647 spin_lock(&fp->fi_lock);
5648 filp = find_writeable_file_locked(fp);
5650 get_lock_access(lock_stp, NFS4_SHARE_ACCESS_WRITE);
5651 spin_unlock(&fp->fi_lock);
5652 file_lock->fl_type = F_WRLCK;
5655 status = nfserr_inval;
5659 status = nfserr_openmode;
5663 file_lock->fl_owner = (fl_owner_t)lockowner(nfs4_get_stateowner(&lock_sop->lo_owner));
5664 file_lock->fl_pid = current->tgid;
5665 file_lock->fl_file = filp;
5666 file_lock->fl_flags = FL_POSIX;
5667 file_lock->fl_lmops = &nfsd_posix_mng_ops;
5668 file_lock->fl_start = lock->lk_offset;
5669 file_lock->fl_end = last_byte_offset(lock->lk_offset, lock->lk_length);
5670 nfs4_transform_lock_offset(file_lock);
5672 conflock = locks_alloc_lock();
5674 dprintk("NFSD: %s: unable to allocate lock!\n", __func__);
5675 status = nfserr_jukebox;
5679 err = vfs_lock_file(filp, F_SETLK, file_lock, conflock);
5681 case 0: /* success! */
5682 nfs4_inc_and_copy_stateid(&lock->lk_resp_stateid, &lock_stp->st_stid);
5685 case (EAGAIN): /* conflock holds conflicting lock */
5686 status = nfserr_denied;
5687 dprintk("NFSD: nfsd4_lock: conflicting lock found!\n");
5688 nfs4_set_lock_denied(conflock, &lock->lk_denied);
5691 status = nfserr_deadlock;
5694 dprintk("NFSD: nfsd4_lock: vfs_lock_file() failed! status %d\n",err);
5695 status = nfserrno(err);
5702 /* Bump seqid manually if the 4.0 replay owner is openowner */
5703 if (cstate->replay_owner &&
5704 cstate->replay_owner != &lock_sop->lo_owner &&
5705 seqid_mutating_err(ntohl(status)))
5706 lock_sop->lo_owner.so_seqid++;
5708 up_write(&lock_stp->st_rwsem);
5711 * If this is a new, never-before-used stateid, and we are
5712 * returning an error, then just go ahead and release it.
5715 release_lock_stateid(lock_stp);
5717 nfs4_put_stid(&lock_stp->st_stid);
5720 nfs4_put_stid(&open_stp->st_stid);
5721 nfsd4_bump_seqid(cstate, status);
5723 locks_free_lock(file_lock);
5725 locks_free_lock(conflock);
5730 * The NFSv4 spec allows a client to do a LOCKT without holding an OPEN,
5731 * so we do a temporary open here just to get an open file to pass to
5732 * vfs_test_lock. (Arguably perhaps test_lock should be done with an
5735 static __be32 nfsd_test_lock(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file_lock *lock)
5738 __be32 err = nfsd_open(rqstp, fhp, S_IFREG, NFSD_MAY_READ, &file);
5740 err = nfserrno(vfs_test_lock(file, lock));
5750 nfsd4_lockt(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
5751 struct nfsd4_lockt *lockt)
5753 struct file_lock *file_lock = NULL;
5754 struct nfs4_lockowner *lo = NULL;
5756 struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
5758 if (locks_in_grace(SVC_NET(rqstp)))
5759 return nfserr_grace;
5761 if (check_lock_length(lockt->lt_offset, lockt->lt_length))
5762 return nfserr_inval;
5764 if (!nfsd4_has_session(cstate)) {
5765 status = lookup_clientid(&lockt->lt_clientid, cstate, nn);
5770 if ((status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0)))
5773 file_lock = locks_alloc_lock();
5775 dprintk("NFSD: %s: unable to allocate lock!\n", __func__);
5776 status = nfserr_jukebox;
5780 switch (lockt->lt_type) {
5783 file_lock->fl_type = F_RDLCK;
5786 case NFS4_WRITEW_LT:
5787 file_lock->fl_type = F_WRLCK;
5790 dprintk("NFSD: nfs4_lockt: bad lock type!\n");
5791 status = nfserr_inval;
5795 lo = find_lockowner_str(cstate->clp, &lockt->lt_owner);
5797 file_lock->fl_owner = (fl_owner_t)lo;
5798 file_lock->fl_pid = current->tgid;
5799 file_lock->fl_flags = FL_POSIX;
5801 file_lock->fl_start = lockt->lt_offset;
5802 file_lock->fl_end = last_byte_offset(lockt->lt_offset, lockt->lt_length);
5804 nfs4_transform_lock_offset(file_lock);
5806 status = nfsd_test_lock(rqstp, &cstate->current_fh, file_lock);
5810 if (file_lock->fl_type != F_UNLCK) {
5811 status = nfserr_denied;
5812 nfs4_set_lock_denied(file_lock, &lockt->lt_denied);
5816 nfs4_put_stateowner(&lo->lo_owner);
5818 locks_free_lock(file_lock);
5823 nfsd4_locku(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
5824 struct nfsd4_locku *locku)
5826 struct nfs4_ol_stateid *stp;
5827 struct file *filp = NULL;
5828 struct file_lock *file_lock = NULL;
5831 struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
5833 dprintk("NFSD: nfsd4_locku: start=%Ld length=%Ld\n",
5834 (long long) locku->lu_offset,
5835 (long long) locku->lu_length);
5837 if (check_lock_length(locku->lu_offset, locku->lu_length))
5838 return nfserr_inval;
5840 status = nfs4_preprocess_seqid_op(cstate, locku->lu_seqid,
5841 &locku->lu_stateid, NFS4_LOCK_STID,
5845 filp = find_any_file(stp->st_stid.sc_file);
5847 status = nfserr_lock_range;
5850 file_lock = locks_alloc_lock();
5852 dprintk("NFSD: %s: unable to allocate lock!\n", __func__);
5853 status = nfserr_jukebox;
5857 file_lock->fl_type = F_UNLCK;
5858 file_lock->fl_owner = (fl_owner_t)lockowner(nfs4_get_stateowner(stp->st_stateowner));
5859 file_lock->fl_pid = current->tgid;
5860 file_lock->fl_file = filp;
5861 file_lock->fl_flags = FL_POSIX;
5862 file_lock->fl_lmops = &nfsd_posix_mng_ops;
5863 file_lock->fl_start = locku->lu_offset;
5865 file_lock->fl_end = last_byte_offset(locku->lu_offset,
5867 nfs4_transform_lock_offset(file_lock);
5869 err = vfs_lock_file(filp, F_SETLK, file_lock, NULL);
5871 dprintk("NFSD: nfs4_locku: vfs_lock_file failed!\n");
5874 nfs4_inc_and_copy_stateid(&locku->lu_stateid, &stp->st_stid);
5878 up_write(&stp->st_rwsem);
5879 nfs4_put_stid(&stp->st_stid);
5881 nfsd4_bump_seqid(cstate, status);
5883 locks_free_lock(file_lock);
5887 status = nfserrno(err);
5893 * true: locks held by lockowner
5894 * false: no locks held by lockowner
5897 check_for_locks(struct nfs4_file *fp, struct nfs4_lockowner *lowner)
5899 struct file_lock *fl;
5901 struct file *filp = find_any_file(fp);
5902 struct inode *inode;
5903 struct file_lock_context *flctx;
5906 /* Any valid lock stateid should have some sort of access */
5911 inode = file_inode(filp);
5912 flctx = inode->i_flctx;
5914 if (flctx && !list_empty_careful(&flctx->flc_posix)) {
5915 spin_lock(&flctx->flc_lock);
5916 list_for_each_entry(fl, &flctx->flc_posix, fl_list) {
5917 if (fl->fl_owner == (fl_owner_t)lowner) {
5922 spin_unlock(&flctx->flc_lock);
5929 nfsd4_release_lockowner(struct svc_rqst *rqstp,
5930 struct nfsd4_compound_state *cstate,
5931 struct nfsd4_release_lockowner *rlockowner)
5933 clientid_t *clid = &rlockowner->rl_clientid;
5934 struct nfs4_stateowner *sop;
5935 struct nfs4_lockowner *lo = NULL;
5936 struct nfs4_ol_stateid *stp;
5937 struct xdr_netobj *owner = &rlockowner->rl_owner;
5938 unsigned int hashval = ownerstr_hashval(owner);
5940 struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
5941 struct nfs4_client *clp;
5943 dprintk("nfsd4_release_lockowner clientid: (%08x/%08x):\n",
5944 clid->cl_boot, clid->cl_id);
5946 status = lookup_clientid(clid, cstate, nn);
5951 /* Find the matching lock stateowner */
5952 spin_lock(&clp->cl_lock);
5953 list_for_each_entry(sop, &clp->cl_ownerstr_hashtbl[hashval],
5956 if (sop->so_is_open_owner || !same_owner_str(sop, owner))
5959 /* see if there are still any locks associated with it */
5960 lo = lockowner(sop);
5961 list_for_each_entry(stp, &sop->so_stateids, st_perstateowner) {
5962 if (check_for_locks(stp->st_stid.sc_file, lo)) {
5963 status = nfserr_locks_held;
5964 spin_unlock(&clp->cl_lock);
5969 nfs4_get_stateowner(sop);
5972 spin_unlock(&clp->cl_lock);
5974 release_lockowner(lo);
5978 static inline struct nfs4_client_reclaim *
5981 return kmalloc(sizeof(struct nfs4_client_reclaim), GFP_KERNEL);
5985 nfs4_has_reclaimed_state(const char *name, struct nfsd_net *nn)
5987 struct nfs4_client_reclaim *crp;
5989 crp = nfsd4_find_reclaim_client(name, nn);
5990 return (crp && crp->cr_clp);
5994 * failure => all reset bets are off, nfserr_no_grace...
5996 struct nfs4_client_reclaim *
5997 nfs4_client_to_reclaim(const char *name, struct nfsd_net *nn)
5999 unsigned int strhashval;
6000 struct nfs4_client_reclaim *crp;
6002 dprintk("NFSD nfs4_client_to_reclaim NAME: %.*s\n", HEXDIR_LEN, name);
6003 crp = alloc_reclaim();
6005 strhashval = clientstr_hashval(name);
6006 INIT_LIST_HEAD(&crp->cr_strhash);
6007 list_add(&crp->cr_strhash, &nn->reclaim_str_hashtbl[strhashval]);
6008 memcpy(crp->cr_recdir, name, HEXDIR_LEN);
6010 nn->reclaim_str_hashtbl_size++;
6016 nfs4_remove_reclaim_record(struct nfs4_client_reclaim *crp, struct nfsd_net *nn)
6018 list_del(&crp->cr_strhash);
6020 nn->reclaim_str_hashtbl_size--;
6024 nfs4_release_reclaim(struct nfsd_net *nn)
6026 struct nfs4_client_reclaim *crp = NULL;
6029 for (i = 0; i < CLIENT_HASH_SIZE; i++) {
6030 while (!list_empty(&nn->reclaim_str_hashtbl[i])) {
6031 crp = list_entry(nn->reclaim_str_hashtbl[i].next,
6032 struct nfs4_client_reclaim, cr_strhash);
6033 nfs4_remove_reclaim_record(crp, nn);
6036 WARN_ON_ONCE(nn->reclaim_str_hashtbl_size);
6040 * called from OPEN, CLAIM_PREVIOUS with a new clientid. */
6041 struct nfs4_client_reclaim *
6042 nfsd4_find_reclaim_client(const char *recdir, struct nfsd_net *nn)
6044 unsigned int strhashval;
6045 struct nfs4_client_reclaim *crp = NULL;
6047 dprintk("NFSD: nfs4_find_reclaim_client for recdir %s\n", recdir);
6049 strhashval = clientstr_hashval(recdir);
6050 list_for_each_entry(crp, &nn->reclaim_str_hashtbl[strhashval], cr_strhash) {
6051 if (same_name(crp->cr_recdir, recdir)) {
6059 * Called from OPEN. Look for clientid in reclaim list.
6062 nfs4_check_open_reclaim(clientid_t *clid,
6063 struct nfsd4_compound_state *cstate,
6064 struct nfsd_net *nn)
6068 /* find clientid in conf_id_hashtbl */
6069 status = lookup_clientid(clid, cstate, nn);
6071 return nfserr_reclaim_bad;
6073 if (test_bit(NFSD4_CLIENT_RECLAIM_COMPLETE, &cstate->clp->cl_flags))
6074 return nfserr_no_grace;
6076 if (nfsd4_client_record_check(cstate->clp))
6077 return nfserr_reclaim_bad;
6082 #ifdef CONFIG_NFSD_FAULT_INJECTION
6084 put_client(struct nfs4_client *clp)
6086 atomic_dec(&clp->cl_refcount);
6089 static struct nfs4_client *
6090 nfsd_find_client(struct sockaddr_storage *addr, size_t addr_size)
6092 struct nfs4_client *clp;
6093 struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6096 if (!nfsd_netns_ready(nn))
6099 list_for_each_entry(clp, &nn->client_lru, cl_lru) {
6100 if (memcmp(&clp->cl_addr, addr, addr_size) == 0)
6107 nfsd_inject_print_clients(void)
6109 struct nfs4_client *clp;
6111 struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6113 char buf[INET6_ADDRSTRLEN];
6115 if (!nfsd_netns_ready(nn))
6118 spin_lock(&nn->client_lock);
6119 list_for_each_entry(clp, &nn->client_lru, cl_lru) {
6120 rpc_ntop((struct sockaddr *)&clp->cl_addr, buf, sizeof(buf));
6121 pr_info("NFS Client: %s\n", buf);
6124 spin_unlock(&nn->client_lock);
6130 nfsd_inject_forget_client(struct sockaddr_storage *addr, size_t addr_size)
6133 struct nfs4_client *clp;
6134 struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6137 if (!nfsd_netns_ready(nn))
6140 spin_lock(&nn->client_lock);
6141 clp = nfsd_find_client(addr, addr_size);
6143 if (mark_client_expired_locked(clp) == nfs_ok)
6148 spin_unlock(&nn->client_lock);
6157 nfsd_inject_forget_clients(u64 max)
6160 struct nfs4_client *clp, *next;
6161 struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6163 LIST_HEAD(reaplist);
6165 if (!nfsd_netns_ready(nn))
6168 spin_lock(&nn->client_lock);
6169 list_for_each_entry_safe(clp, next, &nn->client_lru, cl_lru) {
6170 if (mark_client_expired_locked(clp) == nfs_ok) {
6171 list_add(&clp->cl_lru, &reaplist);
6172 if (max != 0 && ++count >= max)
6176 spin_unlock(&nn->client_lock);
6178 list_for_each_entry_safe(clp, next, &reaplist, cl_lru)
6184 static void nfsd_print_count(struct nfs4_client *clp, unsigned int count,
6187 char buf[INET6_ADDRSTRLEN];
6188 rpc_ntop((struct sockaddr *)&clp->cl_addr, buf, sizeof(buf));
6189 printk(KERN_INFO "NFS Client: %s has %u %s\n", buf, count, type);
6193 nfsd_inject_add_lock_to_list(struct nfs4_ol_stateid *lst,
6194 struct list_head *collect)
6196 struct nfs4_client *clp = lst->st_stid.sc_client;
6197 struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6203 lockdep_assert_held(&nn->client_lock);
6204 atomic_inc(&clp->cl_refcount);
6205 list_add(&lst->st_locks, collect);
6208 static u64 nfsd_foreach_client_lock(struct nfs4_client *clp, u64 max,
6209 struct list_head *collect,
6210 bool (*func)(struct nfs4_ol_stateid *))
6212 struct nfs4_openowner *oop;
6213 struct nfs4_ol_stateid *stp, *st_next;
6214 struct nfs4_ol_stateid *lst, *lst_next;
6217 spin_lock(&clp->cl_lock);
6218 list_for_each_entry(oop, &clp->cl_openowners, oo_perclient) {
6219 list_for_each_entry_safe(stp, st_next,
6220 &oop->oo_owner.so_stateids, st_perstateowner) {
6221 list_for_each_entry_safe(lst, lst_next,
6222 &stp->st_locks, st_locks) {
6225 nfsd_inject_add_lock_to_list(lst,
6230 * Despite the fact that these functions deal
6231 * with 64-bit integers for "count", we must
6232 * ensure that it doesn't blow up the
6233 * clp->cl_refcount. Throw a warning if we
6234 * start to approach INT_MAX here.
6236 WARN_ON_ONCE(count == (INT_MAX / 2));
6243 spin_unlock(&clp->cl_lock);
6249 nfsd_collect_client_locks(struct nfs4_client *clp, struct list_head *collect,
6252 return nfsd_foreach_client_lock(clp, max, collect, unhash_lock_stateid);
6256 nfsd_print_client_locks(struct nfs4_client *clp)
6258 u64 count = nfsd_foreach_client_lock(clp, 0, NULL, NULL);
6259 nfsd_print_count(clp, count, "locked files");
6264 nfsd_inject_print_locks(void)
6266 struct nfs4_client *clp;
6268 struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6271 if (!nfsd_netns_ready(nn))
6274 spin_lock(&nn->client_lock);
6275 list_for_each_entry(clp, &nn->client_lru, cl_lru)
6276 count += nfsd_print_client_locks(clp);
6277 spin_unlock(&nn->client_lock);
6283 nfsd_reap_locks(struct list_head *reaplist)
6285 struct nfs4_client *clp;
6286 struct nfs4_ol_stateid *stp, *next;
6288 list_for_each_entry_safe(stp, next, reaplist, st_locks) {
6289 list_del_init(&stp->st_locks);
6290 clp = stp->st_stid.sc_client;
6291 nfs4_put_stid(&stp->st_stid);
6297 nfsd_inject_forget_client_locks(struct sockaddr_storage *addr, size_t addr_size)
6299 unsigned int count = 0;
6300 struct nfs4_client *clp;
6301 struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6303 LIST_HEAD(reaplist);
6305 if (!nfsd_netns_ready(nn))
6308 spin_lock(&nn->client_lock);
6309 clp = nfsd_find_client(addr, addr_size);
6311 count = nfsd_collect_client_locks(clp, &reaplist, 0);
6312 spin_unlock(&nn->client_lock);
6313 nfsd_reap_locks(&reaplist);
6318 nfsd_inject_forget_locks(u64 max)
6321 struct nfs4_client *clp;
6322 struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6324 LIST_HEAD(reaplist);
6326 if (!nfsd_netns_ready(nn))
6329 spin_lock(&nn->client_lock);
6330 list_for_each_entry(clp, &nn->client_lru, cl_lru) {
6331 count += nfsd_collect_client_locks(clp, &reaplist, max - count);
6332 if (max != 0 && count >= max)
6335 spin_unlock(&nn->client_lock);
6336 nfsd_reap_locks(&reaplist);
6341 nfsd_foreach_client_openowner(struct nfs4_client *clp, u64 max,
6342 struct list_head *collect,
6343 void (*func)(struct nfs4_openowner *))
6345 struct nfs4_openowner *oop, *next;
6346 struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6350 lockdep_assert_held(&nn->client_lock);
6352 spin_lock(&clp->cl_lock);
6353 list_for_each_entry_safe(oop, next, &clp->cl_openowners, oo_perclient) {
6357 atomic_inc(&clp->cl_refcount);
6358 list_add(&oop->oo_perclient, collect);
6363 * Despite the fact that these functions deal with
6364 * 64-bit integers for "count", we must ensure that
6365 * it doesn't blow up the clp->cl_refcount. Throw a
6366 * warning if we start to approach INT_MAX here.
6368 WARN_ON_ONCE(count == (INT_MAX / 2));
6372 spin_unlock(&clp->cl_lock);
6378 nfsd_print_client_openowners(struct nfs4_client *clp)
6380 u64 count = nfsd_foreach_client_openowner(clp, 0, NULL, NULL);
6382 nfsd_print_count(clp, count, "openowners");
6387 nfsd_collect_client_openowners(struct nfs4_client *clp,
6388 struct list_head *collect, u64 max)
6390 return nfsd_foreach_client_openowner(clp, max, collect,
6391 unhash_openowner_locked);
6395 nfsd_inject_print_openowners(void)
6397 struct nfs4_client *clp;
6399 struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6402 if (!nfsd_netns_ready(nn))
6405 spin_lock(&nn->client_lock);
6406 list_for_each_entry(clp, &nn->client_lru, cl_lru)
6407 count += nfsd_print_client_openowners(clp);
6408 spin_unlock(&nn->client_lock);
6414 nfsd_reap_openowners(struct list_head *reaplist)
6416 struct nfs4_client *clp;
6417 struct nfs4_openowner *oop, *next;
6419 list_for_each_entry_safe(oop, next, reaplist, oo_perclient) {
6420 list_del_init(&oop->oo_perclient);
6421 clp = oop->oo_owner.so_client;
6422 release_openowner(oop);
6428 nfsd_inject_forget_client_openowners(struct sockaddr_storage *addr,
6431 unsigned int count = 0;
6432 struct nfs4_client *clp;
6433 struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6435 LIST_HEAD(reaplist);
6437 if (!nfsd_netns_ready(nn))
6440 spin_lock(&nn->client_lock);
6441 clp = nfsd_find_client(addr, addr_size);
6443 count = nfsd_collect_client_openowners(clp, &reaplist, 0);
6444 spin_unlock(&nn->client_lock);
6445 nfsd_reap_openowners(&reaplist);
6450 nfsd_inject_forget_openowners(u64 max)
6453 struct nfs4_client *clp;
6454 struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6456 LIST_HEAD(reaplist);
6458 if (!nfsd_netns_ready(nn))
6461 spin_lock(&nn->client_lock);
6462 list_for_each_entry(clp, &nn->client_lru, cl_lru) {
6463 count += nfsd_collect_client_openowners(clp, &reaplist,
6465 if (max != 0 && count >= max)
6468 spin_unlock(&nn->client_lock);
6469 nfsd_reap_openowners(&reaplist);
6473 static u64 nfsd_find_all_delegations(struct nfs4_client *clp, u64 max,
6474 struct list_head *victims)
6476 struct nfs4_delegation *dp, *next;
6477 struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6481 lockdep_assert_held(&nn->client_lock);
6483 spin_lock(&state_lock);
6484 list_for_each_entry_safe(dp, next, &clp->cl_delegations, dl_perclnt) {
6487 * It's not safe to mess with delegations that have a
6488 * non-zero dl_time. They might have already been broken
6489 * and could be processed by the laundromat outside of
6490 * the state_lock. Just leave them be.
6492 if (dp->dl_time != 0)
6495 atomic_inc(&clp->cl_refcount);
6496 WARN_ON(!unhash_delegation_locked(dp));
6497 list_add(&dp->dl_recall_lru, victims);
6501 * Despite the fact that these functions deal with
6502 * 64-bit integers for "count", we must ensure that
6503 * it doesn't blow up the clp->cl_refcount. Throw a
6504 * warning if we start to approach INT_MAX here.
6506 WARN_ON_ONCE(count == (INT_MAX / 2));
6510 spin_unlock(&state_lock);
6515 nfsd_print_client_delegations(struct nfs4_client *clp)
6517 u64 count = nfsd_find_all_delegations(clp, 0, NULL);
6519 nfsd_print_count(clp, count, "delegations");
6524 nfsd_inject_print_delegations(void)
6526 struct nfs4_client *clp;
6528 struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6531 if (!nfsd_netns_ready(nn))
6534 spin_lock(&nn->client_lock);
6535 list_for_each_entry(clp, &nn->client_lru, cl_lru)
6536 count += nfsd_print_client_delegations(clp);
6537 spin_unlock(&nn->client_lock);
6543 nfsd_forget_delegations(struct list_head *reaplist)
6545 struct nfs4_client *clp;
6546 struct nfs4_delegation *dp, *next;
6548 list_for_each_entry_safe(dp, next, reaplist, dl_recall_lru) {
6549 list_del_init(&dp->dl_recall_lru);
6550 clp = dp->dl_stid.sc_client;
6551 revoke_delegation(dp);
6557 nfsd_inject_forget_client_delegations(struct sockaddr_storage *addr,
6561 struct nfs4_client *clp;
6562 struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6564 LIST_HEAD(reaplist);
6566 if (!nfsd_netns_ready(nn))
6569 spin_lock(&nn->client_lock);
6570 clp = nfsd_find_client(addr, addr_size);
6572 count = nfsd_find_all_delegations(clp, 0, &reaplist);
6573 spin_unlock(&nn->client_lock);
6575 nfsd_forget_delegations(&reaplist);
6580 nfsd_inject_forget_delegations(u64 max)
6583 struct nfs4_client *clp;
6584 struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6586 LIST_HEAD(reaplist);
6588 if (!nfsd_netns_ready(nn))
6591 spin_lock(&nn->client_lock);
6592 list_for_each_entry(clp, &nn->client_lru, cl_lru) {
6593 count += nfsd_find_all_delegations(clp, max - count, &reaplist);
6594 if (max != 0 && count >= max)
6597 spin_unlock(&nn->client_lock);
6598 nfsd_forget_delegations(&reaplist);
6603 nfsd_recall_delegations(struct list_head *reaplist)
6605 struct nfs4_client *clp;
6606 struct nfs4_delegation *dp, *next;
6608 list_for_each_entry_safe(dp, next, reaplist, dl_recall_lru) {
6609 list_del_init(&dp->dl_recall_lru);
6610 clp = dp->dl_stid.sc_client;
6612 * We skipped all entries that had a zero dl_time before,
6613 * so we can now reset the dl_time back to 0. If a delegation
6614 * break comes in now, then it won't make any difference since
6615 * we're recalling it either way.
6617 spin_lock(&state_lock);
6619 spin_unlock(&state_lock);
6620 nfsd_break_one_deleg(dp);
6626 nfsd_inject_recall_client_delegations(struct sockaddr_storage *addr,
6630 struct nfs4_client *clp;
6631 struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6633 LIST_HEAD(reaplist);
6635 if (!nfsd_netns_ready(nn))
6638 spin_lock(&nn->client_lock);
6639 clp = nfsd_find_client(addr, addr_size);
6641 count = nfsd_find_all_delegations(clp, 0, &reaplist);
6642 spin_unlock(&nn->client_lock);
6644 nfsd_recall_delegations(&reaplist);
6649 nfsd_inject_recall_delegations(u64 max)
6652 struct nfs4_client *clp, *next;
6653 struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6655 LIST_HEAD(reaplist);
6657 if (!nfsd_netns_ready(nn))
6660 spin_lock(&nn->client_lock);
6661 list_for_each_entry_safe(clp, next, &nn->client_lru, cl_lru) {
6662 count += nfsd_find_all_delegations(clp, max - count, &reaplist);
6663 if (max != 0 && ++count >= max)
6666 spin_unlock(&nn->client_lock);
6667 nfsd_recall_delegations(&reaplist);
6670 #endif /* CONFIG_NFSD_FAULT_INJECTION */
6673 * Since the lifetime of a delegation isn't limited to that of an open, a
6674 * client may quite reasonably hang on to a delegation as long as it has
6675 * the inode cached. This becomes an obvious problem the first time a
6676 * client's inode cache approaches the size of the server's total memory.
6678 * For now we avoid this problem by imposing a hard limit on the number
6679 * of delegations, which varies according to the server's memory size.
6682 set_max_delegations(void)
6685 * Allow at most 4 delegations per megabyte of RAM. Quick
6686 * estimates suggest that in the worst case (where every delegation
6687 * is for a different inode), a delegation could take about 1.5K,
6688 * giving a worst case usage of about 6% of memory.
6690 max_delegations = nr_free_buffer_pages() >> (20 - 2 - PAGE_SHIFT);
6693 static int nfs4_state_create_net(struct net *net)
6695 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
6698 nn->conf_id_hashtbl = kmalloc(sizeof(struct list_head) *
6699 CLIENT_HASH_SIZE, GFP_KERNEL);
6700 if (!nn->conf_id_hashtbl)
6702 nn->unconf_id_hashtbl = kmalloc(sizeof(struct list_head) *
6703 CLIENT_HASH_SIZE, GFP_KERNEL);
6704 if (!nn->unconf_id_hashtbl)
6706 nn->sessionid_hashtbl = kmalloc(sizeof(struct list_head) *
6707 SESSION_HASH_SIZE, GFP_KERNEL);
6708 if (!nn->sessionid_hashtbl)
6711 for (i = 0; i < CLIENT_HASH_SIZE; i++) {
6712 INIT_LIST_HEAD(&nn->conf_id_hashtbl[i]);
6713 INIT_LIST_HEAD(&nn->unconf_id_hashtbl[i]);
6715 for (i = 0; i < SESSION_HASH_SIZE; i++)
6716 INIT_LIST_HEAD(&nn->sessionid_hashtbl[i]);
6717 nn->conf_name_tree = RB_ROOT;
6718 nn->unconf_name_tree = RB_ROOT;
6719 INIT_LIST_HEAD(&nn->client_lru);
6720 INIT_LIST_HEAD(&nn->close_lru);
6721 INIT_LIST_HEAD(&nn->del_recall_lru);
6722 spin_lock_init(&nn->client_lock);
6724 INIT_DELAYED_WORK(&nn->laundromat_work, laundromat_main);
6730 kfree(nn->unconf_id_hashtbl);
6732 kfree(nn->conf_id_hashtbl);
6738 nfs4_state_destroy_net(struct net *net)
6741 struct nfs4_client *clp = NULL;
6742 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
6744 for (i = 0; i < CLIENT_HASH_SIZE; i++) {
6745 while (!list_empty(&nn->conf_id_hashtbl[i])) {
6746 clp = list_entry(nn->conf_id_hashtbl[i].next, struct nfs4_client, cl_idhash);
6747 destroy_client(clp);
6751 for (i = 0; i < CLIENT_HASH_SIZE; i++) {
6752 while (!list_empty(&nn->unconf_id_hashtbl[i])) {
6753 clp = list_entry(nn->unconf_id_hashtbl[i].next, struct nfs4_client, cl_idhash);
6754 destroy_client(clp);
6758 kfree(nn->sessionid_hashtbl);
6759 kfree(nn->unconf_id_hashtbl);
6760 kfree(nn->conf_id_hashtbl);
6765 nfs4_state_start_net(struct net *net)
6767 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
6770 ret = nfs4_state_create_net(net);
6773 nn->boot_time = get_seconds();
6774 nn->grace_ended = false;
6775 nn->nfsd4_manager.block_opens = true;
6776 locks_start_grace(net, &nn->nfsd4_manager);
6777 nfsd4_client_tracking_init(net);
6778 printk(KERN_INFO "NFSD: starting %ld-second grace period (net %p)\n",
6779 nn->nfsd4_grace, net);
6780 queue_delayed_work(laundry_wq, &nn->laundromat_work, nn->nfsd4_grace * HZ);
6784 /* initialization to perform when the nfsd service is started: */
6787 nfs4_state_start(void)
6791 ret = set_callback_cred();
6794 laundry_wq = alloc_workqueue("%s", WQ_UNBOUND, 0, "nfsd4");
6795 if (laundry_wq == NULL) {
6799 ret = nfsd4_create_callback_queue();
6801 goto out_free_laundry;
6803 set_max_delegations();
6808 destroy_workqueue(laundry_wq);
6814 nfs4_state_shutdown_net(struct net *net)
6816 struct nfs4_delegation *dp = NULL;
6817 struct list_head *pos, *next, reaplist;
6818 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
6820 cancel_delayed_work_sync(&nn->laundromat_work);
6821 locks_end_grace(&nn->nfsd4_manager);
6823 INIT_LIST_HEAD(&reaplist);
6824 spin_lock(&state_lock);
6825 list_for_each_safe(pos, next, &nn->del_recall_lru) {
6826 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
6827 WARN_ON(!unhash_delegation_locked(dp));
6828 list_add(&dp->dl_recall_lru, &reaplist);
6830 spin_unlock(&state_lock);
6831 list_for_each_safe(pos, next, &reaplist) {
6832 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
6833 list_del_init(&dp->dl_recall_lru);
6834 put_clnt_odstate(dp->dl_clnt_odstate);
6835 nfs4_put_deleg_lease(dp->dl_stid.sc_file);
6836 nfs4_put_stid(&dp->dl_stid);
6839 nfsd4_client_tracking_exit(net);
6840 nfs4_state_destroy_net(net);
6844 nfs4_state_shutdown(void)
6846 destroy_workqueue(laundry_wq);
6847 nfsd4_destroy_callback_queue();
6851 get_stateid(struct nfsd4_compound_state *cstate, stateid_t *stateid)
6853 if (HAS_STATE_ID(cstate, CURRENT_STATE_ID_FLAG) && CURRENT_STATEID(stateid))
6854 memcpy(stateid, &cstate->current_stateid, sizeof(stateid_t));
6858 put_stateid(struct nfsd4_compound_state *cstate, stateid_t *stateid)
6860 if (cstate->minorversion) {
6861 memcpy(&cstate->current_stateid, stateid, sizeof(stateid_t));
6862 SET_STATE_ID(cstate, CURRENT_STATE_ID_FLAG);
6867 clear_current_stateid(struct nfsd4_compound_state *cstate)
6869 CLEAR_STATE_ID(cstate, CURRENT_STATE_ID_FLAG);
6873 * functions to set current state id
6876 nfsd4_set_opendowngradestateid(struct nfsd4_compound_state *cstate, struct nfsd4_open_downgrade *odp)
6878 put_stateid(cstate, &odp->od_stateid);
6882 nfsd4_set_openstateid(struct nfsd4_compound_state *cstate, struct nfsd4_open *open)
6884 put_stateid(cstate, &open->op_stateid);
6888 nfsd4_set_closestateid(struct nfsd4_compound_state *cstate, struct nfsd4_close *close)
6890 put_stateid(cstate, &close->cl_stateid);
6894 nfsd4_set_lockstateid(struct nfsd4_compound_state *cstate, struct nfsd4_lock *lock)
6896 put_stateid(cstate, &lock->lk_resp_stateid);
6900 * functions to consume current state id
6904 nfsd4_get_opendowngradestateid(struct nfsd4_compound_state *cstate, struct nfsd4_open_downgrade *odp)
6906 get_stateid(cstate, &odp->od_stateid);
6910 nfsd4_get_delegreturnstateid(struct nfsd4_compound_state *cstate, struct nfsd4_delegreturn *drp)
6912 get_stateid(cstate, &drp->dr_stateid);
6916 nfsd4_get_freestateid(struct nfsd4_compound_state *cstate, struct nfsd4_free_stateid *fsp)
6918 get_stateid(cstate, &fsp->fr_stateid);
6922 nfsd4_get_setattrstateid(struct nfsd4_compound_state *cstate, struct nfsd4_setattr *setattr)
6924 get_stateid(cstate, &setattr->sa_stateid);
6928 nfsd4_get_closestateid(struct nfsd4_compound_state *cstate, struct nfsd4_close *close)
6930 get_stateid(cstate, &close->cl_stateid);
6934 nfsd4_get_lockustateid(struct nfsd4_compound_state *cstate, struct nfsd4_locku *locku)
6936 get_stateid(cstate, &locku->lu_stateid);
6940 nfsd4_get_readstateid(struct nfsd4_compound_state *cstate, struct nfsd4_read *read)
6942 get_stateid(cstate, &read->rd_stateid);
6946 nfsd4_get_writestateid(struct nfsd4_compound_state *cstate, struct nfsd4_write *write)
6948 get_stateid(cstate, &write->wr_stateid);