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 status = nfserr_encr_alg_unsupp;
2415 /* Cases below refer to rfc 5661 section 18.35.4: */
2416 spin_lock(&nn->client_lock);
2417 conf = find_confirmed_client_by_name(&exid->clname, nn);
2419 bool creds_match = same_creds(&conf->cl_cred, &rqstp->rq_cred);
2420 bool verfs_match = same_verf(&verf, &conf->cl_verifier);
2423 if (!clp_used_exchangeid(conf)) { /* buggy client */
2424 status = nfserr_inval;
2427 if (!mach_creds_match(conf, rqstp)) {
2428 status = nfserr_wrong_cred;
2431 if (!creds_match) { /* case 9 */
2432 status = nfserr_perm;
2435 if (!verfs_match) { /* case 8 */
2436 status = nfserr_not_same;
2440 exid->flags |= EXCHGID4_FLAG_CONFIRMED_R;
2443 if (!creds_match) { /* case 3 */
2444 if (client_has_state(conf)) {
2445 status = nfserr_clid_inuse;
2450 if (verfs_match) { /* case 2 */
2451 conf->cl_exchange_flags |= EXCHGID4_FLAG_CONFIRMED_R;
2454 /* case 5, client reboot */
2459 if (update) { /* case 7 */
2460 status = nfserr_noent;
2464 unconf = find_unconfirmed_client_by_name(&exid->clname, nn);
2465 if (unconf) /* case 4, possible retry or client restart */
2466 unhash_client_locked(unconf);
2468 /* case 1 (normal case) */
2471 status = mark_client_expired_locked(conf);
2475 new->cl_minorversion = cstate->minorversion;
2478 add_to_unconfirmed(new);
2481 exid->clientid.cl_boot = conf->cl_clientid.cl_boot;
2482 exid->clientid.cl_id = conf->cl_clientid.cl_id;
2484 exid->seqid = conf->cl_cs_slot.sl_seqid + 1;
2485 nfsd4_set_ex_flags(conf, exid);
2487 dprintk("nfsd4_exchange_id seqid %d flags %x\n",
2488 conf->cl_cs_slot.sl_seqid, conf->cl_exchange_flags);
2492 spin_unlock(&nn->client_lock);
2497 expire_client(unconf);
2502 check_slot_seqid(u32 seqid, u32 slot_seqid, int slot_inuse)
2504 dprintk("%s enter. seqid %d slot_seqid %d\n", __func__, seqid,
2507 /* The slot is in use, and no response has been sent. */
2509 if (seqid == slot_seqid)
2510 return nfserr_jukebox;
2512 return nfserr_seq_misordered;
2514 /* Note unsigned 32-bit arithmetic handles wraparound: */
2515 if (likely(seqid == slot_seqid + 1))
2517 if (seqid == slot_seqid)
2518 return nfserr_replay_cache;
2519 return nfserr_seq_misordered;
2523 * Cache the create session result into the create session single DRC
2524 * slot cache by saving the xdr structure. sl_seqid has been set.
2525 * Do this for solo or embedded create session operations.
2528 nfsd4_cache_create_session(struct nfsd4_create_session *cr_ses,
2529 struct nfsd4_clid_slot *slot, __be32 nfserr)
2531 slot->sl_status = nfserr;
2532 memcpy(&slot->sl_cr_ses, cr_ses, sizeof(*cr_ses));
2536 nfsd4_replay_create_session(struct nfsd4_create_session *cr_ses,
2537 struct nfsd4_clid_slot *slot)
2539 memcpy(cr_ses, &slot->sl_cr_ses, sizeof(*cr_ses));
2540 return slot->sl_status;
2543 #define NFSD_MIN_REQ_HDR_SEQ_SZ ((\
2544 2 * 2 + /* credential,verifier: AUTH_NULL, length 0 */ \
2545 1 + /* MIN tag is length with zero, only length */ \
2546 3 + /* version, opcount, opcode */ \
2547 XDR_QUADLEN(NFS4_MAX_SESSIONID_LEN) + \
2548 /* seqid, slotID, slotID, cache */ \
2549 4 ) * sizeof(__be32))
2551 #define NFSD_MIN_RESP_HDR_SEQ_SZ ((\
2552 2 + /* verifier: AUTH_NULL, length 0 */\
2554 1 + /* MIN tag is length with zero, only length */ \
2555 3 + /* opcount, opcode, opstatus*/ \
2556 XDR_QUADLEN(NFS4_MAX_SESSIONID_LEN) + \
2557 /* seqid, slotID, slotID, slotID, status */ \
2558 5 ) * sizeof(__be32))
2560 static __be32 check_forechannel_attrs(struct nfsd4_channel_attrs *ca, struct nfsd_net *nn)
2562 u32 maxrpc = nn->nfsd_serv->sv_max_mesg;
2564 if (ca->maxreq_sz < NFSD_MIN_REQ_HDR_SEQ_SZ)
2565 return nfserr_toosmall;
2566 if (ca->maxresp_sz < NFSD_MIN_RESP_HDR_SEQ_SZ)
2567 return nfserr_toosmall;
2568 ca->headerpadsz = 0;
2569 ca->maxreq_sz = min_t(u32, ca->maxreq_sz, maxrpc);
2570 ca->maxresp_sz = min_t(u32, ca->maxresp_sz, maxrpc);
2571 ca->maxops = min_t(u32, ca->maxops, NFSD_MAX_OPS_PER_COMPOUND);
2572 ca->maxresp_cached = min_t(u32, ca->maxresp_cached,
2573 NFSD_SLOT_CACHE_SIZE + NFSD_MIN_HDR_SEQ_SZ);
2574 ca->maxreqs = min_t(u32, ca->maxreqs, NFSD_MAX_SLOTS_PER_SESSION);
2576 * Note decreasing slot size below client's request may make it
2577 * difficult for client to function correctly, whereas
2578 * decreasing the number of slots will (just?) affect
2579 * performance. When short on memory we therefore prefer to
2580 * decrease number of slots instead of their size. Clients that
2581 * request larger slots than they need will get poor results:
2583 ca->maxreqs = nfsd4_get_drc_mem(ca);
2585 return nfserr_jukebox;
2591 * Server's NFSv4.1 backchannel support is AUTH_SYS-only for now.
2592 * These are based on similar macros in linux/sunrpc/msg_prot.h .
2594 #define RPC_MAX_HEADER_WITH_AUTH_SYS \
2595 (RPC_CALLHDRSIZE + 2 * (2 + UNX_CALLSLACK))
2597 #define RPC_MAX_REPHEADER_WITH_AUTH_SYS \
2598 (RPC_REPHDRSIZE + (2 + NUL_REPLYSLACK))
2600 #define NFSD_CB_MAX_REQ_SZ ((NFS4_enc_cb_recall_sz + \
2601 RPC_MAX_HEADER_WITH_AUTH_SYS) * sizeof(__be32))
2602 #define NFSD_CB_MAX_RESP_SZ ((NFS4_dec_cb_recall_sz + \
2603 RPC_MAX_REPHEADER_WITH_AUTH_SYS) * \
2606 static __be32 check_backchannel_attrs(struct nfsd4_channel_attrs *ca)
2608 ca->headerpadsz = 0;
2610 if (ca->maxreq_sz < NFSD_CB_MAX_REQ_SZ)
2611 return nfserr_toosmall;
2612 if (ca->maxresp_sz < NFSD_CB_MAX_RESP_SZ)
2613 return nfserr_toosmall;
2614 ca->maxresp_cached = 0;
2616 return nfserr_toosmall;
2621 static __be32 nfsd4_check_cb_sec(struct nfsd4_cb_sec *cbs)
2623 switch (cbs->flavor) {
2629 * GSS case: the spec doesn't allow us to return this
2630 * error. But it also doesn't allow us not to support
2632 * I'd rather this fail hard than return some error the
2633 * client might think it can already handle:
2635 return nfserr_encr_alg_unsupp;
2640 nfsd4_create_session(struct svc_rqst *rqstp,
2641 struct nfsd4_compound_state *cstate,
2642 struct nfsd4_create_session *cr_ses)
2644 struct sockaddr *sa = svc_addr(rqstp);
2645 struct nfs4_client *conf, *unconf;
2646 struct nfs4_client *old = NULL;
2647 struct nfsd4_session *new;
2648 struct nfsd4_conn *conn;
2649 struct nfsd4_clid_slot *cs_slot = NULL;
2651 struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
2653 if (cr_ses->flags & ~SESSION4_FLAG_MASK_A)
2654 return nfserr_inval;
2655 status = nfsd4_check_cb_sec(&cr_ses->cb_sec);
2658 status = check_forechannel_attrs(&cr_ses->fore_channel, nn);
2661 status = check_backchannel_attrs(&cr_ses->back_channel);
2663 goto out_release_drc_mem;
2664 status = nfserr_jukebox;
2665 new = alloc_session(&cr_ses->fore_channel, &cr_ses->back_channel);
2667 goto out_release_drc_mem;
2668 conn = alloc_conn_from_crses(rqstp, cr_ses);
2670 goto out_free_session;
2672 spin_lock(&nn->client_lock);
2673 unconf = find_unconfirmed_client(&cr_ses->clientid, true, nn);
2674 conf = find_confirmed_client(&cr_ses->clientid, true, nn);
2675 WARN_ON_ONCE(conf && unconf);
2678 status = nfserr_wrong_cred;
2679 if (!mach_creds_match(conf, rqstp))
2681 cs_slot = &conf->cl_cs_slot;
2682 status = check_slot_seqid(cr_ses->seqid, cs_slot->sl_seqid, 0);
2684 if (status == nfserr_replay_cache)
2685 status = nfsd4_replay_create_session(cr_ses, cs_slot);
2688 } else if (unconf) {
2689 if (!same_creds(&unconf->cl_cred, &rqstp->rq_cred) ||
2690 !rpc_cmp_addr(sa, (struct sockaddr *) &unconf->cl_addr)) {
2691 status = nfserr_clid_inuse;
2694 status = nfserr_wrong_cred;
2695 if (!mach_creds_match(unconf, rqstp))
2697 cs_slot = &unconf->cl_cs_slot;
2698 status = check_slot_seqid(cr_ses->seqid, cs_slot->sl_seqid, 0);
2700 /* an unconfirmed replay returns misordered */
2701 status = nfserr_seq_misordered;
2704 old = find_confirmed_client_by_name(&unconf->cl_name, nn);
2706 status = mark_client_expired_locked(old);
2712 move_to_confirmed(unconf);
2715 status = nfserr_stale_clientid;
2719 /* Persistent sessions are not supported */
2720 cr_ses->flags &= ~SESSION4_PERSIST;
2721 /* Upshifting from TCP to RDMA is not supported */
2722 cr_ses->flags &= ~SESSION4_RDMA;
2724 init_session(rqstp, new, conf, cr_ses);
2725 nfsd4_get_session_locked(new);
2727 memcpy(cr_ses->sessionid.data, new->se_sessionid.data,
2728 NFS4_MAX_SESSIONID_LEN);
2729 cs_slot->sl_seqid++;
2730 cr_ses->seqid = cs_slot->sl_seqid;
2732 /* cache solo and embedded create sessions under the client_lock */
2733 nfsd4_cache_create_session(cr_ses, cs_slot, status);
2734 spin_unlock(&nn->client_lock);
2735 /* init connection and backchannel */
2736 nfsd4_init_conn(rqstp, conn, new);
2737 nfsd4_put_session(new);
2742 spin_unlock(&nn->client_lock);
2747 __free_session(new);
2748 out_release_drc_mem:
2749 nfsd4_put_drc_mem(&cr_ses->fore_channel);
2753 static __be32 nfsd4_map_bcts_dir(u32 *dir)
2756 case NFS4_CDFC4_FORE:
2757 case NFS4_CDFC4_BACK:
2759 case NFS4_CDFC4_FORE_OR_BOTH:
2760 case NFS4_CDFC4_BACK_OR_BOTH:
2761 *dir = NFS4_CDFC4_BOTH;
2764 return nfserr_inval;
2767 __be32 nfsd4_backchannel_ctl(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate, struct nfsd4_backchannel_ctl *bc)
2769 struct nfsd4_session *session = cstate->session;
2770 struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
2773 status = nfsd4_check_cb_sec(&bc->bc_cb_sec);
2776 spin_lock(&nn->client_lock);
2777 session->se_cb_prog = bc->bc_cb_program;
2778 session->se_cb_sec = bc->bc_cb_sec;
2779 spin_unlock(&nn->client_lock);
2781 nfsd4_probe_callback(session->se_client);
2786 __be32 nfsd4_bind_conn_to_session(struct svc_rqst *rqstp,
2787 struct nfsd4_compound_state *cstate,
2788 struct nfsd4_bind_conn_to_session *bcts)
2791 struct nfsd4_conn *conn;
2792 struct nfsd4_session *session;
2793 struct net *net = SVC_NET(rqstp);
2794 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
2796 if (!nfsd4_last_compound_op(rqstp))
2797 return nfserr_not_only_op;
2798 spin_lock(&nn->client_lock);
2799 session = find_in_sessionid_hashtbl(&bcts->sessionid, net, &status);
2800 spin_unlock(&nn->client_lock);
2802 goto out_no_session;
2803 status = nfserr_wrong_cred;
2804 if (!mach_creds_match(session->se_client, rqstp))
2806 status = nfsd4_map_bcts_dir(&bcts->dir);
2809 conn = alloc_conn(rqstp, bcts->dir);
2810 status = nfserr_jukebox;
2813 nfsd4_init_conn(rqstp, conn, session);
2816 nfsd4_put_session(session);
2821 static bool nfsd4_compound_in_session(struct nfsd4_session *session, struct nfs4_sessionid *sid)
2825 return !memcmp(sid, &session->se_sessionid, sizeof(*sid));
2829 nfsd4_destroy_session(struct svc_rqst *r,
2830 struct nfsd4_compound_state *cstate,
2831 struct nfsd4_destroy_session *sessionid)
2833 struct nfsd4_session *ses;
2835 int ref_held_by_me = 0;
2836 struct net *net = SVC_NET(r);
2837 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
2839 status = nfserr_not_only_op;
2840 if (nfsd4_compound_in_session(cstate->session, &sessionid->sessionid)) {
2841 if (!nfsd4_last_compound_op(r))
2845 dump_sessionid(__func__, &sessionid->sessionid);
2846 spin_lock(&nn->client_lock);
2847 ses = find_in_sessionid_hashtbl(&sessionid->sessionid, net, &status);
2849 goto out_client_lock;
2850 status = nfserr_wrong_cred;
2851 if (!mach_creds_match(ses->se_client, r))
2852 goto out_put_session;
2853 status = mark_session_dead_locked(ses, 1 + ref_held_by_me);
2855 goto out_put_session;
2856 unhash_session(ses);
2857 spin_unlock(&nn->client_lock);
2859 nfsd4_probe_callback_sync(ses->se_client);
2861 spin_lock(&nn->client_lock);
2864 nfsd4_put_session_locked(ses);
2866 spin_unlock(&nn->client_lock);
2871 static struct nfsd4_conn *__nfsd4_find_conn(struct svc_xprt *xpt, struct nfsd4_session *s)
2873 struct nfsd4_conn *c;
2875 list_for_each_entry(c, &s->se_conns, cn_persession) {
2876 if (c->cn_xprt == xpt) {
2883 static __be32 nfsd4_sequence_check_conn(struct nfsd4_conn *new, struct nfsd4_session *ses)
2885 struct nfs4_client *clp = ses->se_client;
2886 struct nfsd4_conn *c;
2887 __be32 status = nfs_ok;
2890 spin_lock(&clp->cl_lock);
2891 c = __nfsd4_find_conn(new->cn_xprt, ses);
2894 status = nfserr_conn_not_bound_to_session;
2895 if (clp->cl_mach_cred)
2897 __nfsd4_hash_conn(new, ses);
2898 spin_unlock(&clp->cl_lock);
2899 ret = nfsd4_register_conn(new);
2901 /* oops; xprt is already down: */
2902 nfsd4_conn_lost(&new->cn_xpt_user);
2905 spin_unlock(&clp->cl_lock);
2910 static bool nfsd4_session_too_many_ops(struct svc_rqst *rqstp, struct nfsd4_session *session)
2912 struct nfsd4_compoundargs *args = rqstp->rq_argp;
2914 return args->opcnt > session->se_fchannel.maxops;
2917 static bool nfsd4_request_too_big(struct svc_rqst *rqstp,
2918 struct nfsd4_session *session)
2920 struct xdr_buf *xb = &rqstp->rq_arg;
2922 return xb->len > session->se_fchannel.maxreq_sz;
2926 nfsd4_sequence(struct svc_rqst *rqstp,
2927 struct nfsd4_compound_state *cstate,
2928 struct nfsd4_sequence *seq)
2930 struct nfsd4_compoundres *resp = rqstp->rq_resp;
2931 struct xdr_stream *xdr = &resp->xdr;
2932 struct nfsd4_session *session;
2933 struct nfs4_client *clp;
2934 struct nfsd4_slot *slot;
2935 struct nfsd4_conn *conn;
2938 struct net *net = SVC_NET(rqstp);
2939 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
2941 if (resp->opcnt != 1)
2942 return nfserr_sequence_pos;
2945 * Will be either used or freed by nfsd4_sequence_check_conn
2948 conn = alloc_conn(rqstp, NFS4_CDFC4_FORE);
2950 return nfserr_jukebox;
2952 spin_lock(&nn->client_lock);
2953 session = find_in_sessionid_hashtbl(&seq->sessionid, net, &status);
2955 goto out_no_session;
2956 clp = session->se_client;
2958 status = nfserr_too_many_ops;
2959 if (nfsd4_session_too_many_ops(rqstp, session))
2960 goto out_put_session;
2962 status = nfserr_req_too_big;
2963 if (nfsd4_request_too_big(rqstp, session))
2964 goto out_put_session;
2966 status = nfserr_badslot;
2967 if (seq->slotid >= session->se_fchannel.maxreqs)
2968 goto out_put_session;
2970 slot = session->se_slots[seq->slotid];
2971 dprintk("%s: slotid %d\n", __func__, seq->slotid);
2973 /* We do not negotiate the number of slots yet, so set the
2974 * maxslots to the session maxreqs which is used to encode
2975 * sr_highest_slotid and the sr_target_slot id to maxslots */
2976 seq->maxslots = session->se_fchannel.maxreqs;
2978 status = check_slot_seqid(seq->seqid, slot->sl_seqid,
2979 slot->sl_flags & NFSD4_SLOT_INUSE);
2980 if (status == nfserr_replay_cache) {
2981 status = nfserr_seq_misordered;
2982 if (!(slot->sl_flags & NFSD4_SLOT_INITIALIZED))
2983 goto out_put_session;
2984 cstate->slot = slot;
2985 cstate->session = session;
2987 /* Return the cached reply status and set cstate->status
2988 * for nfsd4_proc_compound processing */
2989 status = nfsd4_replay_cache_entry(resp, seq);
2990 cstate->status = nfserr_replay_cache;
2994 goto out_put_session;
2996 status = nfsd4_sequence_check_conn(conn, session);
2999 goto out_put_session;
3001 buflen = (seq->cachethis) ?
3002 session->se_fchannel.maxresp_cached :
3003 session->se_fchannel.maxresp_sz;
3004 status = (seq->cachethis) ? nfserr_rep_too_big_to_cache :
3006 if (xdr_restrict_buflen(xdr, buflen - rqstp->rq_auth_slack))
3007 goto out_put_session;
3008 svc_reserve(rqstp, buflen);
3011 /* Success! bump slot seqid */
3012 slot->sl_seqid = seq->seqid;
3013 slot->sl_flags |= NFSD4_SLOT_INUSE;
3015 slot->sl_flags |= NFSD4_SLOT_CACHETHIS;
3017 slot->sl_flags &= ~NFSD4_SLOT_CACHETHIS;
3019 cstate->slot = slot;
3020 cstate->session = session;
3024 switch (clp->cl_cb_state) {
3026 seq->status_flags = SEQ4_STATUS_CB_PATH_DOWN;
3028 case NFSD4_CB_FAULT:
3029 seq->status_flags = SEQ4_STATUS_BACKCHANNEL_FAULT;
3032 seq->status_flags = 0;
3034 if (!list_empty(&clp->cl_revoked))
3035 seq->status_flags |= SEQ4_STATUS_RECALLABLE_STATE_REVOKED;
3039 spin_unlock(&nn->client_lock);
3042 nfsd4_put_session_locked(session);
3043 goto out_no_session;
3047 nfsd4_sequence_done(struct nfsd4_compoundres *resp)
3049 struct nfsd4_compound_state *cs = &resp->cstate;
3051 if (nfsd4_has_session(cs)) {
3052 if (cs->status != nfserr_replay_cache) {
3053 nfsd4_store_cache_entry(resp);
3054 cs->slot->sl_flags &= ~NFSD4_SLOT_INUSE;
3056 /* Drop session reference that was taken in nfsd4_sequence() */
3057 nfsd4_put_session(cs->session);
3059 put_client_renew(cs->clp);
3063 nfsd4_destroy_clientid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate, struct nfsd4_destroy_clientid *dc)
3065 struct nfs4_client *conf, *unconf;
3066 struct nfs4_client *clp = NULL;
3068 struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
3070 spin_lock(&nn->client_lock);
3071 unconf = find_unconfirmed_client(&dc->clientid, true, nn);
3072 conf = find_confirmed_client(&dc->clientid, true, nn);
3073 WARN_ON_ONCE(conf && unconf);
3076 if (client_has_state(conf)) {
3077 status = nfserr_clientid_busy;
3080 status = mark_client_expired_locked(conf);
3087 status = nfserr_stale_clientid;
3090 if (!mach_creds_match(clp, rqstp)) {
3092 status = nfserr_wrong_cred;
3095 unhash_client_locked(clp);
3097 spin_unlock(&nn->client_lock);
3104 nfsd4_reclaim_complete(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate, struct nfsd4_reclaim_complete *rc)
3108 if (rc->rca_one_fs) {
3109 if (!cstate->current_fh.fh_dentry)
3110 return nfserr_nofilehandle;
3112 * We don't take advantage of the rca_one_fs case.
3113 * That's OK, it's optional, we can safely ignore it.
3118 status = nfserr_complete_already;
3119 if (test_and_set_bit(NFSD4_CLIENT_RECLAIM_COMPLETE,
3120 &cstate->session->se_client->cl_flags))
3123 status = nfserr_stale_clientid;
3124 if (is_client_expired(cstate->session->se_client))
3126 * The following error isn't really legal.
3127 * But we only get here if the client just explicitly
3128 * destroyed the client. Surely it no longer cares what
3129 * error it gets back on an operation for the dead
3135 nfsd4_client_record_create(cstate->session->se_client);
3141 nfsd4_setclientid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3142 struct nfsd4_setclientid *setclid)
3144 struct xdr_netobj clname = setclid->se_name;
3145 nfs4_verifier clverifier = setclid->se_verf;
3146 struct nfs4_client *conf, *new;
3147 struct nfs4_client *unconf = NULL;
3149 struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
3151 new = create_client(clname, rqstp, &clverifier);
3153 return nfserr_jukebox;
3154 /* Cases below refer to rfc 3530 section 14.2.33: */
3155 spin_lock(&nn->client_lock);
3156 conf = find_confirmed_client_by_name(&clname, nn);
3157 if (conf && client_has_state(conf)) {
3159 status = nfserr_clid_inuse;
3160 if (clp_used_exchangeid(conf))
3162 if (!same_creds(&conf->cl_cred, &rqstp->rq_cred)) {
3163 char addr_str[INET6_ADDRSTRLEN];
3164 rpc_ntop((struct sockaddr *) &conf->cl_addr, addr_str,
3166 dprintk("NFSD: setclientid: string in use by client "
3167 "at %s\n", addr_str);
3171 unconf = find_unconfirmed_client_by_name(&clname, nn);
3173 unhash_client_locked(unconf);
3174 if (conf && same_verf(&conf->cl_verifier, &clverifier)) {
3175 /* case 1: probable callback update */
3176 copy_clid(new, conf);
3177 gen_confirm(new, nn);
3178 } else /* case 4 (new client) or cases 2, 3 (client reboot): */
3180 new->cl_minorversion = 0;
3181 gen_callback(new, setclid, rqstp);
3182 add_to_unconfirmed(new);
3183 setclid->se_clientid.cl_boot = new->cl_clientid.cl_boot;
3184 setclid->se_clientid.cl_id = new->cl_clientid.cl_id;
3185 memcpy(setclid->se_confirm.data, new->cl_confirm.data, sizeof(setclid->se_confirm.data));
3189 spin_unlock(&nn->client_lock);
3193 expire_client(unconf);
3199 nfsd4_setclientid_confirm(struct svc_rqst *rqstp,
3200 struct nfsd4_compound_state *cstate,
3201 struct nfsd4_setclientid_confirm *setclientid_confirm)
3203 struct nfs4_client *conf, *unconf;
3204 struct nfs4_client *old = NULL;
3205 nfs4_verifier confirm = setclientid_confirm->sc_confirm;
3206 clientid_t * clid = &setclientid_confirm->sc_clientid;
3208 struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
3210 if (STALE_CLIENTID(clid, nn))
3211 return nfserr_stale_clientid;
3213 spin_lock(&nn->client_lock);
3214 conf = find_confirmed_client(clid, false, nn);
3215 unconf = find_unconfirmed_client(clid, false, nn);
3217 * We try hard to give out unique clientid's, so if we get an
3218 * attempt to confirm the same clientid with a different cred,
3219 * the client may be buggy; this should never happen.
3221 * Nevertheless, RFC 7530 recommends INUSE for this case:
3223 status = nfserr_clid_inuse;
3224 if (unconf && !same_creds(&unconf->cl_cred, &rqstp->rq_cred))
3226 if (conf && !same_creds(&conf->cl_cred, &rqstp->rq_cred))
3228 /* cases below refer to rfc 3530 section 14.2.34: */
3229 if (!unconf || !same_verf(&confirm, &unconf->cl_confirm)) {
3230 if (conf && !unconf) /* case 2: probable retransmit */
3232 else /* case 4: client hasn't noticed we rebooted yet? */
3233 status = nfserr_stale_clientid;
3237 if (conf) { /* case 1: callback update */
3239 unhash_client_locked(old);
3240 nfsd4_change_callback(conf, &unconf->cl_cb_conn);
3241 } else { /* case 3: normal case; new or rebooted client */
3242 old = find_confirmed_client_by_name(&unconf->cl_name, nn);
3244 status = nfserr_clid_inuse;
3245 if (client_has_state(old)
3246 && !same_creds(&unconf->cl_cred,
3249 status = mark_client_expired_locked(old);
3255 move_to_confirmed(unconf);
3258 get_client_locked(conf);
3259 spin_unlock(&nn->client_lock);
3260 nfsd4_probe_callback(conf);
3261 spin_lock(&nn->client_lock);
3262 put_client_renew_locked(conf);
3264 spin_unlock(&nn->client_lock);
3270 static struct nfs4_file *nfsd4_alloc_file(void)
3272 return kmem_cache_alloc(file_slab, GFP_KERNEL);
3275 /* OPEN Share state helper functions */
3276 static void nfsd4_init_file(struct knfsd_fh *fh, unsigned int hashval,
3277 struct nfs4_file *fp)
3279 lockdep_assert_held(&state_lock);
3281 atomic_set(&fp->fi_ref, 1);
3282 spin_lock_init(&fp->fi_lock);
3283 INIT_LIST_HEAD(&fp->fi_stateids);
3284 INIT_LIST_HEAD(&fp->fi_delegations);
3285 INIT_LIST_HEAD(&fp->fi_clnt_odstate);
3286 fh_copy_shallow(&fp->fi_fhandle, fh);
3287 fp->fi_deleg_file = NULL;
3288 fp->fi_had_conflict = false;
3289 fp->fi_share_deny = 0;
3290 memset(fp->fi_fds, 0, sizeof(fp->fi_fds));
3291 memset(fp->fi_access, 0, sizeof(fp->fi_access));
3292 #ifdef CONFIG_NFSD_PNFS
3293 INIT_LIST_HEAD(&fp->fi_lo_states);
3294 atomic_set(&fp->fi_lo_recalls, 0);
3296 hlist_add_head_rcu(&fp->fi_hash, &file_hashtbl[hashval]);
3300 nfsd4_free_slabs(void)
3302 kmem_cache_destroy(odstate_slab);
3303 kmem_cache_destroy(openowner_slab);
3304 kmem_cache_destroy(lockowner_slab);
3305 kmem_cache_destroy(file_slab);
3306 kmem_cache_destroy(stateid_slab);
3307 kmem_cache_destroy(deleg_slab);
3311 nfsd4_init_slabs(void)
3313 openowner_slab = kmem_cache_create("nfsd4_openowners",
3314 sizeof(struct nfs4_openowner), 0, 0, NULL);
3315 if (openowner_slab == NULL)
3317 lockowner_slab = kmem_cache_create("nfsd4_lockowners",
3318 sizeof(struct nfs4_lockowner), 0, 0, NULL);
3319 if (lockowner_slab == NULL)
3320 goto out_free_openowner_slab;
3321 file_slab = kmem_cache_create("nfsd4_files",
3322 sizeof(struct nfs4_file), 0, 0, NULL);
3323 if (file_slab == NULL)
3324 goto out_free_lockowner_slab;
3325 stateid_slab = kmem_cache_create("nfsd4_stateids",
3326 sizeof(struct nfs4_ol_stateid), 0, 0, NULL);
3327 if (stateid_slab == NULL)
3328 goto out_free_file_slab;
3329 deleg_slab = kmem_cache_create("nfsd4_delegations",
3330 sizeof(struct nfs4_delegation), 0, 0, NULL);
3331 if (deleg_slab == NULL)
3332 goto out_free_stateid_slab;
3333 odstate_slab = kmem_cache_create("nfsd4_odstate",
3334 sizeof(struct nfs4_clnt_odstate), 0, 0, NULL);
3335 if (odstate_slab == NULL)
3336 goto out_free_deleg_slab;
3339 out_free_deleg_slab:
3340 kmem_cache_destroy(deleg_slab);
3341 out_free_stateid_slab:
3342 kmem_cache_destroy(stateid_slab);
3344 kmem_cache_destroy(file_slab);
3345 out_free_lockowner_slab:
3346 kmem_cache_destroy(lockowner_slab);
3347 out_free_openowner_slab:
3348 kmem_cache_destroy(openowner_slab);
3350 dprintk("nfsd4: out of memory while initializing nfsv4\n");
3354 static void init_nfs4_replay(struct nfs4_replay *rp)
3356 rp->rp_status = nfserr_serverfault;
3358 rp->rp_buf = rp->rp_ibuf;
3359 mutex_init(&rp->rp_mutex);
3362 static void nfsd4_cstate_assign_replay(struct nfsd4_compound_state *cstate,
3363 struct nfs4_stateowner *so)
3365 if (!nfsd4_has_session(cstate)) {
3366 mutex_lock(&so->so_replay.rp_mutex);
3367 cstate->replay_owner = nfs4_get_stateowner(so);
3371 void nfsd4_cstate_clear_replay(struct nfsd4_compound_state *cstate)
3373 struct nfs4_stateowner *so = cstate->replay_owner;
3376 cstate->replay_owner = NULL;
3377 mutex_unlock(&so->so_replay.rp_mutex);
3378 nfs4_put_stateowner(so);
3382 static inline void *alloc_stateowner(struct kmem_cache *slab, struct xdr_netobj *owner, struct nfs4_client *clp)
3384 struct nfs4_stateowner *sop;
3386 sop = kmem_cache_alloc(slab, GFP_KERNEL);
3390 sop->so_owner.data = kmemdup(owner->data, owner->len, GFP_KERNEL);
3391 if (!sop->so_owner.data) {
3392 kmem_cache_free(slab, sop);
3395 sop->so_owner.len = owner->len;
3397 INIT_LIST_HEAD(&sop->so_stateids);
3398 sop->so_client = clp;
3399 init_nfs4_replay(&sop->so_replay);
3400 atomic_set(&sop->so_count, 1);
3404 static void hash_openowner(struct nfs4_openowner *oo, struct nfs4_client *clp, unsigned int strhashval)
3406 lockdep_assert_held(&clp->cl_lock);
3408 list_add(&oo->oo_owner.so_strhash,
3409 &clp->cl_ownerstr_hashtbl[strhashval]);
3410 list_add(&oo->oo_perclient, &clp->cl_openowners);
3413 static void nfs4_unhash_openowner(struct nfs4_stateowner *so)
3415 unhash_openowner_locked(openowner(so));
3418 static void nfs4_free_openowner(struct nfs4_stateowner *so)
3420 struct nfs4_openowner *oo = openowner(so);
3422 kmem_cache_free(openowner_slab, oo);
3425 static const struct nfs4_stateowner_operations openowner_ops = {
3426 .so_unhash = nfs4_unhash_openowner,
3427 .so_free = nfs4_free_openowner,
3430 static struct nfs4_ol_stateid *
3431 nfsd4_find_existing_open(struct nfs4_file *fp, struct nfsd4_open *open)
3433 struct nfs4_ol_stateid *local, *ret = NULL;
3434 struct nfs4_openowner *oo = open->op_openowner;
3436 lockdep_assert_held(&fp->fi_lock);
3438 list_for_each_entry(local, &fp->fi_stateids, st_perfile) {
3439 /* ignore lock owners */
3440 if (local->st_stateowner->so_is_open_owner == 0)
3442 if (local->st_stateowner == &oo->oo_owner) {
3444 atomic_inc(&ret->st_stid.sc_count);
3451 static struct nfs4_openowner *
3452 alloc_init_open_stateowner(unsigned int strhashval, struct nfsd4_open *open,
3453 struct nfsd4_compound_state *cstate)
3455 struct nfs4_client *clp = cstate->clp;
3456 struct nfs4_openowner *oo, *ret;
3458 oo = alloc_stateowner(openowner_slab, &open->op_owner, clp);
3461 oo->oo_owner.so_ops = &openowner_ops;
3462 oo->oo_owner.so_is_open_owner = 1;
3463 oo->oo_owner.so_seqid = open->op_seqid;
3465 if (nfsd4_has_session(cstate))
3466 oo->oo_flags |= NFS4_OO_CONFIRMED;
3468 oo->oo_last_closed_stid = NULL;
3469 INIT_LIST_HEAD(&oo->oo_close_lru);
3470 spin_lock(&clp->cl_lock);
3471 ret = find_openstateowner_str_locked(strhashval, open, clp);
3473 hash_openowner(oo, clp, strhashval);
3476 nfs4_free_stateowner(&oo->oo_owner);
3478 spin_unlock(&clp->cl_lock);
3482 static struct nfs4_ol_stateid *
3483 init_open_stateid(struct nfs4_file *fp, struct nfsd4_open *open)
3486 struct nfs4_openowner *oo = open->op_openowner;
3487 struct nfs4_ol_stateid *retstp = NULL;
3488 struct nfs4_ol_stateid *stp;
3491 /* We are moving these outside of the spinlocks to avoid the warnings */
3492 mutex_init(&stp->st_mutex);
3493 mutex_lock(&stp->st_mutex);
3495 spin_lock(&oo->oo_owner.so_client->cl_lock);
3496 spin_lock(&fp->fi_lock);
3498 retstp = nfsd4_find_existing_open(fp, open);
3502 open->op_stp = NULL;
3503 atomic_inc(&stp->st_stid.sc_count);
3504 stp->st_stid.sc_type = NFS4_OPEN_STID;
3505 INIT_LIST_HEAD(&stp->st_locks);
3506 stp->st_stateowner = nfs4_get_stateowner(&oo->oo_owner);
3508 stp->st_stid.sc_file = fp;
3509 stp->st_access_bmap = 0;
3510 stp->st_deny_bmap = 0;
3511 stp->st_openstp = NULL;
3512 list_add(&stp->st_perstateowner, &oo->oo_owner.so_stateids);
3513 list_add(&stp->st_perfile, &fp->fi_stateids);
3516 spin_unlock(&fp->fi_lock);
3517 spin_unlock(&oo->oo_owner.so_client->cl_lock);
3519 mutex_lock(&retstp->st_mutex);
3520 /* To keep mutex tracking happy */
3521 mutex_unlock(&stp->st_mutex);
3528 * In the 4.0 case we need to keep the owners around a little while to handle
3529 * CLOSE replay. We still do need to release any file access that is held by
3530 * them before returning however.
3533 move_to_close_lru(struct nfs4_ol_stateid *s, struct net *net)
3535 struct nfs4_ol_stateid *last;
3536 struct nfs4_openowner *oo = openowner(s->st_stateowner);
3537 struct nfsd_net *nn = net_generic(s->st_stid.sc_client->net,
3540 dprintk("NFSD: move_to_close_lru nfs4_openowner %p\n", oo);
3543 * We know that we hold one reference via nfsd4_close, and another
3544 * "persistent" reference for the client. If the refcount is higher
3545 * than 2, then there are still calls in progress that are using this
3546 * stateid. We can't put the sc_file reference until they are finished.
3547 * Wait for the refcount to drop to 2. Since it has been unhashed,
3548 * there should be no danger of the refcount going back up again at
3551 wait_event(close_wq, atomic_read(&s->st_stid.sc_count) == 2);
3553 release_all_access(s);
3554 if (s->st_stid.sc_file) {
3555 put_nfs4_file(s->st_stid.sc_file);
3556 s->st_stid.sc_file = NULL;
3559 spin_lock(&nn->client_lock);
3560 last = oo->oo_last_closed_stid;
3561 oo->oo_last_closed_stid = s;
3562 list_move_tail(&oo->oo_close_lru, &nn->close_lru);
3563 oo->oo_time = get_seconds();
3564 spin_unlock(&nn->client_lock);
3566 nfs4_put_stid(&last->st_stid);
3569 /* search file_hashtbl[] for file */
3570 static struct nfs4_file *
3571 find_file_locked(struct knfsd_fh *fh, unsigned int hashval)
3573 struct nfs4_file *fp;
3575 hlist_for_each_entry_rcu(fp, &file_hashtbl[hashval], fi_hash) {
3576 if (fh_match(&fp->fi_fhandle, fh)) {
3577 if (atomic_inc_not_zero(&fp->fi_ref))
3585 find_file(struct knfsd_fh *fh)
3587 struct nfs4_file *fp;
3588 unsigned int hashval = file_hashval(fh);
3591 fp = find_file_locked(fh, hashval);
3596 static struct nfs4_file *
3597 find_or_add_file(struct nfs4_file *new, struct knfsd_fh *fh)
3599 struct nfs4_file *fp;
3600 unsigned int hashval = file_hashval(fh);
3603 fp = find_file_locked(fh, hashval);
3608 spin_lock(&state_lock);
3609 fp = find_file_locked(fh, hashval);
3610 if (likely(fp == NULL)) {
3611 nfsd4_init_file(fh, hashval, new);
3614 spin_unlock(&state_lock);
3620 * Called to check deny when READ with all zero stateid or
3621 * WRITE with all zero or all one stateid
3624 nfs4_share_conflict(struct svc_fh *current_fh, unsigned int deny_type)
3626 struct nfs4_file *fp;
3627 __be32 ret = nfs_ok;
3629 fp = find_file(¤t_fh->fh_handle);
3632 /* Check for conflicting share reservations */
3633 spin_lock(&fp->fi_lock);
3634 if (fp->fi_share_deny & deny_type)
3635 ret = nfserr_locked;
3636 spin_unlock(&fp->fi_lock);
3641 static void nfsd4_cb_recall_prepare(struct nfsd4_callback *cb)
3643 struct nfs4_delegation *dp = cb_to_delegation(cb);
3644 struct nfsd_net *nn = net_generic(dp->dl_stid.sc_client->net,
3647 block_delegations(&dp->dl_stid.sc_file->fi_fhandle);
3650 * We can't do this in nfsd_break_deleg_cb because it is
3651 * already holding inode->i_lock.
3653 * If the dl_time != 0, then we know that it has already been
3654 * queued for a lease break. Don't queue it again.
3656 spin_lock(&state_lock);
3657 if (dp->dl_time == 0) {
3658 dp->dl_time = get_seconds();
3659 list_add_tail(&dp->dl_recall_lru, &nn->del_recall_lru);
3661 spin_unlock(&state_lock);
3664 static int nfsd4_cb_recall_done(struct nfsd4_callback *cb,
3665 struct rpc_task *task)
3667 struct nfs4_delegation *dp = cb_to_delegation(cb);
3669 if (dp->dl_stid.sc_type == NFS4_CLOSED_DELEG_STID)
3672 switch (task->tk_status) {
3676 case -NFS4ERR_BAD_STATEID:
3678 * Race: client probably got cb_recall before open reply
3679 * granting delegation.
3681 if (dp->dl_retries--) {
3682 rpc_delay(task, 2 * HZ);
3691 static void nfsd4_cb_recall_release(struct nfsd4_callback *cb)
3693 struct nfs4_delegation *dp = cb_to_delegation(cb);
3695 nfs4_put_stid(&dp->dl_stid);
3698 static const struct nfsd4_callback_ops nfsd4_cb_recall_ops = {
3699 .prepare = nfsd4_cb_recall_prepare,
3700 .done = nfsd4_cb_recall_done,
3701 .release = nfsd4_cb_recall_release,
3704 static void nfsd_break_one_deleg(struct nfs4_delegation *dp)
3707 * We're assuming the state code never drops its reference
3708 * without first removing the lease. Since we're in this lease
3709 * callback (and since the lease code is serialized by the kernel
3710 * lock) we know the server hasn't removed the lease yet, we know
3711 * it's safe to take a reference.
3713 atomic_inc(&dp->dl_stid.sc_count);
3714 nfsd4_run_cb(&dp->dl_recall);
3717 /* Called from break_lease() with i_lock held. */
3719 nfsd_break_deleg_cb(struct file_lock *fl)
3722 struct nfs4_file *fp = (struct nfs4_file *)fl->fl_owner;
3723 struct nfs4_delegation *dp;
3726 WARN(1, "(%p)->fl_owner NULL\n", fl);
3729 if (fp->fi_had_conflict) {
3730 WARN(1, "duplicate break on %p\n", fp);
3734 * We don't want the locks code to timeout the lease for us;
3735 * we'll remove it ourself if a delegation isn't returned
3738 fl->fl_break_time = 0;
3740 spin_lock(&fp->fi_lock);
3741 fp->fi_had_conflict = true;
3743 * If there are no delegations on the list, then return true
3744 * so that the lease code will go ahead and delete it.
3746 if (list_empty(&fp->fi_delegations))
3749 list_for_each_entry(dp, &fp->fi_delegations, dl_perfile)
3750 nfsd_break_one_deleg(dp);
3751 spin_unlock(&fp->fi_lock);
3756 nfsd_change_deleg_cb(struct file_lock *onlist, int arg,
3757 struct list_head *dispose)
3760 return lease_modify(onlist, arg, dispose);
3765 static const struct lock_manager_operations nfsd_lease_mng_ops = {
3766 .lm_break = nfsd_break_deleg_cb,
3767 .lm_change = nfsd_change_deleg_cb,
3770 static __be32 nfsd4_check_seqid(struct nfsd4_compound_state *cstate, struct nfs4_stateowner *so, u32 seqid)
3772 if (nfsd4_has_session(cstate))
3774 if (seqid == so->so_seqid - 1)
3775 return nfserr_replay_me;
3776 if (seqid == so->so_seqid)
3778 return nfserr_bad_seqid;
3781 static __be32 lookup_clientid(clientid_t *clid,
3782 struct nfsd4_compound_state *cstate,
3783 struct nfsd_net *nn)
3785 struct nfs4_client *found;
3788 found = cstate->clp;
3789 if (!same_clid(&found->cl_clientid, clid))
3790 return nfserr_stale_clientid;
3794 if (STALE_CLIENTID(clid, nn))
3795 return nfserr_stale_clientid;
3798 * For v4.1+ we get the client in the SEQUENCE op. If we don't have one
3799 * cached already then we know this is for is for v4.0 and "sessions"
3802 WARN_ON_ONCE(cstate->session);
3803 spin_lock(&nn->client_lock);
3804 found = find_confirmed_client(clid, false, nn);
3806 spin_unlock(&nn->client_lock);
3807 return nfserr_expired;
3809 atomic_inc(&found->cl_refcount);
3810 spin_unlock(&nn->client_lock);
3812 /* Cache the nfs4_client in cstate! */
3813 cstate->clp = found;
3818 nfsd4_process_open1(struct nfsd4_compound_state *cstate,
3819 struct nfsd4_open *open, struct nfsd_net *nn)
3821 clientid_t *clientid = &open->op_clientid;
3822 struct nfs4_client *clp = NULL;
3823 unsigned int strhashval;
3824 struct nfs4_openowner *oo = NULL;
3827 if (STALE_CLIENTID(&open->op_clientid, nn))
3828 return nfserr_stale_clientid;
3830 * In case we need it later, after we've already created the
3831 * file and don't want to risk a further failure:
3833 open->op_file = nfsd4_alloc_file();
3834 if (open->op_file == NULL)
3835 return nfserr_jukebox;
3837 status = lookup_clientid(clientid, cstate, nn);
3842 strhashval = ownerstr_hashval(&open->op_owner);
3843 oo = find_openstateowner_str(strhashval, open, clp);
3844 open->op_openowner = oo;
3848 if (!(oo->oo_flags & NFS4_OO_CONFIRMED)) {
3849 /* Replace unconfirmed owners without checking for replay. */
3850 release_openowner(oo);
3851 open->op_openowner = NULL;
3854 status = nfsd4_check_seqid(cstate, &oo->oo_owner, open->op_seqid);
3859 oo = alloc_init_open_stateowner(strhashval, open, cstate);
3861 return nfserr_jukebox;
3862 open->op_openowner = oo;
3864 open->op_stp = nfs4_alloc_open_stateid(clp);
3866 return nfserr_jukebox;
3868 if (nfsd4_has_session(cstate) &&
3869 (cstate->current_fh.fh_export->ex_flags & NFSEXP_PNFS)) {
3870 open->op_odstate = alloc_clnt_odstate(clp);
3871 if (!open->op_odstate)
3872 return nfserr_jukebox;
3878 static inline __be32
3879 nfs4_check_delegmode(struct nfs4_delegation *dp, int flags)
3881 if ((flags & WR_STATE) && (dp->dl_type == NFS4_OPEN_DELEGATE_READ))
3882 return nfserr_openmode;
3887 static int share_access_to_flags(u32 share_access)
3889 return share_access == NFS4_SHARE_ACCESS_READ ? RD_STATE : WR_STATE;
3892 static struct nfs4_delegation *find_deleg_stateid(struct nfs4_client *cl, stateid_t *s)
3894 struct nfs4_stid *ret;
3896 ret = find_stateid_by_type(cl, s, NFS4_DELEG_STID);
3899 return delegstateid(ret);
3902 static bool nfsd4_is_deleg_cur(struct nfsd4_open *open)
3904 return open->op_claim_type == NFS4_OPEN_CLAIM_DELEGATE_CUR ||
3905 open->op_claim_type == NFS4_OPEN_CLAIM_DELEG_CUR_FH;
3909 nfs4_check_deleg(struct nfs4_client *cl, struct nfsd4_open *open,
3910 struct nfs4_delegation **dp)
3913 __be32 status = nfserr_bad_stateid;
3914 struct nfs4_delegation *deleg;
3916 deleg = find_deleg_stateid(cl, &open->op_delegate_stateid);
3919 flags = share_access_to_flags(open->op_share_access);
3920 status = nfs4_check_delegmode(deleg, flags);
3922 nfs4_put_stid(&deleg->dl_stid);
3927 if (!nfsd4_is_deleg_cur(open))
3931 open->op_openowner->oo_flags |= NFS4_OO_CONFIRMED;
3935 static inline int nfs4_access_to_access(u32 nfs4_access)
3939 if (nfs4_access & NFS4_SHARE_ACCESS_READ)
3940 flags |= NFSD_MAY_READ;
3941 if (nfs4_access & NFS4_SHARE_ACCESS_WRITE)
3942 flags |= NFSD_MAY_WRITE;
3946 static inline __be32
3947 nfsd4_truncate(struct svc_rqst *rqstp, struct svc_fh *fh,
3948 struct nfsd4_open *open)
3950 struct iattr iattr = {
3951 .ia_valid = ATTR_SIZE,
3954 if (!open->op_truncate)
3956 if (!(open->op_share_access & NFS4_SHARE_ACCESS_WRITE))
3957 return nfserr_inval;
3958 return nfsd_setattr(rqstp, fh, &iattr, 0, (time_t)0);
3961 static __be32 nfs4_get_vfs_file(struct svc_rqst *rqstp, struct nfs4_file *fp,
3962 struct svc_fh *cur_fh, struct nfs4_ol_stateid *stp,
3963 struct nfsd4_open *open)
3965 struct file *filp = NULL;
3967 int oflag = nfs4_access_to_omode(open->op_share_access);
3968 int access = nfs4_access_to_access(open->op_share_access);
3969 unsigned char old_access_bmap, old_deny_bmap;
3971 spin_lock(&fp->fi_lock);
3974 * Are we trying to set a deny mode that would conflict with
3977 status = nfs4_file_check_deny(fp, open->op_share_deny);
3978 if (status != nfs_ok) {
3979 spin_unlock(&fp->fi_lock);
3983 /* set access to the file */
3984 status = nfs4_file_get_access(fp, open->op_share_access);
3985 if (status != nfs_ok) {
3986 spin_unlock(&fp->fi_lock);
3990 /* Set access bits in stateid */
3991 old_access_bmap = stp->st_access_bmap;
3992 set_access(open->op_share_access, stp);
3994 /* Set new deny mask */
3995 old_deny_bmap = stp->st_deny_bmap;
3996 set_deny(open->op_share_deny, stp);
3997 fp->fi_share_deny |= (open->op_share_deny & NFS4_SHARE_DENY_BOTH);
3999 if (!fp->fi_fds[oflag]) {
4000 spin_unlock(&fp->fi_lock);
4001 status = nfsd_open(rqstp, cur_fh, S_IFREG, access, &filp);
4003 goto out_put_access;
4004 spin_lock(&fp->fi_lock);
4005 if (!fp->fi_fds[oflag]) {
4006 fp->fi_fds[oflag] = filp;
4010 spin_unlock(&fp->fi_lock);
4014 status = nfsd4_truncate(rqstp, cur_fh, open);
4016 goto out_put_access;
4020 stp->st_access_bmap = old_access_bmap;
4021 nfs4_file_put_access(fp, open->op_share_access);
4022 reset_union_bmap_deny(bmap_to_share_mode(old_deny_bmap), stp);
4027 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)
4030 unsigned char old_deny_bmap = stp->st_deny_bmap;
4032 if (!test_access(open->op_share_access, stp))
4033 return nfs4_get_vfs_file(rqstp, fp, cur_fh, stp, open);
4035 /* test and set deny mode */
4036 spin_lock(&fp->fi_lock);
4037 status = nfs4_file_check_deny(fp, open->op_share_deny);
4038 if (status == nfs_ok) {
4039 set_deny(open->op_share_deny, stp);
4040 fp->fi_share_deny |=
4041 (open->op_share_deny & NFS4_SHARE_DENY_BOTH);
4043 spin_unlock(&fp->fi_lock);
4045 if (status != nfs_ok)
4048 status = nfsd4_truncate(rqstp, cur_fh, open);
4049 if (status != nfs_ok)
4050 reset_union_bmap_deny(old_deny_bmap, stp);
4054 /* Should we give out recallable state?: */
4055 static bool nfsd4_cb_channel_good(struct nfs4_client *clp)
4057 if (clp->cl_cb_state == NFSD4_CB_UP)
4060 * In the sessions case, since we don't have to establish a
4061 * separate connection for callbacks, we assume it's OK
4062 * until we hear otherwise:
4064 return clp->cl_minorversion && clp->cl_cb_state == NFSD4_CB_UNKNOWN;
4067 static struct file_lock *nfs4_alloc_init_lease(struct nfs4_file *fp, int flag)
4069 struct file_lock *fl;
4071 fl = locks_alloc_lock();
4074 fl->fl_lmops = &nfsd_lease_mng_ops;
4075 fl->fl_flags = FL_DELEG;
4076 fl->fl_type = flag == NFS4_OPEN_DELEGATE_READ? F_RDLCK: F_WRLCK;
4077 fl->fl_end = OFFSET_MAX;
4078 fl->fl_owner = (fl_owner_t)fp;
4079 fl->fl_pid = current->tgid;
4084 * nfs4_setlease - Obtain a delegation by requesting lease from vfs layer
4085 * @dp: a pointer to the nfs4_delegation we're adding.
4088 * On success: Return code will be 0 on success.
4090 * On error: -EAGAIN if there was an existing delegation.
4091 * nonzero if there is an error in other cases.
4095 static int nfs4_setlease(struct nfs4_delegation *dp)
4097 struct nfs4_file *fp = dp->dl_stid.sc_file;
4098 struct file_lock *fl;
4102 fl = nfs4_alloc_init_lease(fp, NFS4_OPEN_DELEGATE_READ);
4105 filp = find_readable_file(fp);
4107 /* We should always have a readable file here */
4109 locks_free_lock(fl);
4113 status = vfs_setlease(filp, fl->fl_type, &fl, NULL);
4115 locks_free_lock(fl);
4118 spin_lock(&state_lock);
4119 spin_lock(&fp->fi_lock);
4120 /* Did the lease get broken before we took the lock? */
4122 if (fp->fi_had_conflict)
4125 if (fp->fi_deleg_file) {
4126 status = hash_delegation_locked(dp, fp);
4129 fp->fi_deleg_file = filp;
4130 fp->fi_delegees = 0;
4131 status = hash_delegation_locked(dp, fp);
4132 spin_unlock(&fp->fi_lock);
4133 spin_unlock(&state_lock);
4135 /* Should never happen, this is a new fi_deleg_file */
4141 spin_unlock(&fp->fi_lock);
4142 spin_unlock(&state_lock);
4148 static struct nfs4_delegation *
4149 nfs4_set_delegation(struct nfs4_client *clp, struct svc_fh *fh,
4150 struct nfs4_file *fp, struct nfs4_clnt_odstate *odstate)
4153 struct nfs4_delegation *dp;
4155 if (fp->fi_had_conflict)
4156 return ERR_PTR(-EAGAIN);
4158 spin_lock(&state_lock);
4159 spin_lock(&fp->fi_lock);
4160 status = nfs4_get_existing_delegation(clp, fp);
4161 spin_unlock(&fp->fi_lock);
4162 spin_unlock(&state_lock);
4165 return ERR_PTR(status);
4167 dp = alloc_init_deleg(clp, fh, odstate);
4169 return ERR_PTR(-ENOMEM);
4172 spin_lock(&state_lock);
4173 spin_lock(&fp->fi_lock);
4174 dp->dl_stid.sc_file = fp;
4175 if (!fp->fi_deleg_file) {
4176 spin_unlock(&fp->fi_lock);
4177 spin_unlock(&state_lock);
4178 status = nfs4_setlease(dp);
4181 if (fp->fi_had_conflict) {
4185 status = hash_delegation_locked(dp, fp);
4187 spin_unlock(&fp->fi_lock);
4188 spin_unlock(&state_lock);
4191 put_clnt_odstate(dp->dl_clnt_odstate);
4192 nfs4_put_stid(&dp->dl_stid);
4193 return ERR_PTR(status);
4198 static void nfsd4_open_deleg_none_ext(struct nfsd4_open *open, int status)
4200 open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE_EXT;
4201 if (status == -EAGAIN)
4202 open->op_why_no_deleg = WND4_CONTENTION;
4204 open->op_why_no_deleg = WND4_RESOURCE;
4205 switch (open->op_deleg_want) {
4206 case NFS4_SHARE_WANT_READ_DELEG:
4207 case NFS4_SHARE_WANT_WRITE_DELEG:
4208 case NFS4_SHARE_WANT_ANY_DELEG:
4210 case NFS4_SHARE_WANT_CANCEL:
4211 open->op_why_no_deleg = WND4_CANCELLED;
4213 case NFS4_SHARE_WANT_NO_DELEG:
4220 * Attempt to hand out a delegation.
4222 * Note we don't support write delegations, and won't until the vfs has
4223 * proper support for them.
4226 nfs4_open_delegation(struct svc_fh *fh, struct nfsd4_open *open,
4227 struct nfs4_ol_stateid *stp)
4229 struct nfs4_delegation *dp;
4230 struct nfs4_openowner *oo = openowner(stp->st_stateowner);
4231 struct nfs4_client *clp = stp->st_stid.sc_client;
4235 cb_up = nfsd4_cb_channel_good(oo->oo_owner.so_client);
4236 open->op_recall = 0;
4237 switch (open->op_claim_type) {
4238 case NFS4_OPEN_CLAIM_PREVIOUS:
4240 open->op_recall = 1;
4241 if (open->op_delegate_type != NFS4_OPEN_DELEGATE_READ)
4244 case NFS4_OPEN_CLAIM_NULL:
4245 case NFS4_OPEN_CLAIM_FH:
4247 * Let's not give out any delegations till everyone's
4248 * had the chance to reclaim theirs, *and* until
4249 * NLM locks have all been reclaimed:
4251 if (locks_in_grace(clp->net))
4253 if (!cb_up || !(oo->oo_flags & NFS4_OO_CONFIRMED))
4256 * Also, if the file was opened for write or
4257 * create, there's a good chance the client's
4258 * about to write to it, resulting in an
4259 * immediate recall (since we don't support
4260 * write delegations):
4262 if (open->op_share_access & NFS4_SHARE_ACCESS_WRITE)
4264 if (open->op_create == NFS4_OPEN_CREATE)
4270 dp = nfs4_set_delegation(clp, fh, stp->st_stid.sc_file, stp->st_clnt_odstate);
4274 memcpy(&open->op_delegate_stateid, &dp->dl_stid.sc_stateid, sizeof(dp->dl_stid.sc_stateid));
4276 dprintk("NFSD: delegation stateid=" STATEID_FMT "\n",
4277 STATEID_VAL(&dp->dl_stid.sc_stateid));
4278 open->op_delegate_type = NFS4_OPEN_DELEGATE_READ;
4279 nfs4_put_stid(&dp->dl_stid);
4282 open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE;
4283 if (open->op_claim_type == NFS4_OPEN_CLAIM_PREVIOUS &&
4284 open->op_delegate_type != NFS4_OPEN_DELEGATE_NONE) {
4285 dprintk("NFSD: WARNING: refusing delegation reclaim\n");
4286 open->op_recall = 1;
4289 /* 4.1 client asking for a delegation? */
4290 if (open->op_deleg_want)
4291 nfsd4_open_deleg_none_ext(open, status);
4295 static void nfsd4_deleg_xgrade_none_ext(struct nfsd4_open *open,
4296 struct nfs4_delegation *dp)
4298 if (open->op_deleg_want == NFS4_SHARE_WANT_READ_DELEG &&
4299 dp->dl_type == NFS4_OPEN_DELEGATE_WRITE) {
4300 open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE_EXT;
4301 open->op_why_no_deleg = WND4_NOT_SUPP_DOWNGRADE;
4302 } else if (open->op_deleg_want == NFS4_SHARE_WANT_WRITE_DELEG &&
4303 dp->dl_type == NFS4_OPEN_DELEGATE_WRITE) {
4304 open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE_EXT;
4305 open->op_why_no_deleg = WND4_NOT_SUPP_UPGRADE;
4307 /* Otherwise the client must be confused wanting a delegation
4308 * it already has, therefore we don't return
4309 * NFS4_OPEN_DELEGATE_NONE_EXT and reason.
4314 nfsd4_process_open2(struct svc_rqst *rqstp, struct svc_fh *current_fh, struct nfsd4_open *open)
4316 struct nfsd4_compoundres *resp = rqstp->rq_resp;
4317 struct nfs4_client *cl = open->op_openowner->oo_owner.so_client;
4318 struct nfs4_file *fp = NULL;
4319 struct nfs4_ol_stateid *stp = NULL;
4320 struct nfs4_delegation *dp = NULL;
4324 * Lookup file; if found, lookup stateid and check open request,
4325 * and check for delegations in the process of being recalled.
4326 * If not found, create the nfs4_file struct
4328 fp = find_or_add_file(open->op_file, ¤t_fh->fh_handle);
4329 if (fp != open->op_file) {
4330 status = nfs4_check_deleg(cl, open, &dp);
4333 spin_lock(&fp->fi_lock);
4334 stp = nfsd4_find_existing_open(fp, open);
4335 spin_unlock(&fp->fi_lock);
4337 open->op_file = NULL;
4338 status = nfserr_bad_stateid;
4339 if (nfsd4_is_deleg_cur(open))
4344 * OPEN the file, or upgrade an existing OPEN.
4345 * If truncate fails, the OPEN fails.
4348 /* Stateid was found, this is an OPEN upgrade */
4349 mutex_lock(&stp->st_mutex);
4350 status = nfs4_upgrade_open(rqstp, fp, current_fh, stp, open);
4352 mutex_unlock(&stp->st_mutex);
4356 /* stp is returned locked. */
4357 stp = init_open_stateid(fp, open);
4358 /* See if we lost the race to some other thread */
4359 if (stp->st_access_bmap != 0) {
4360 status = nfs4_upgrade_open(rqstp, fp, current_fh,
4363 mutex_unlock(&stp->st_mutex);
4368 status = nfs4_get_vfs_file(rqstp, fp, current_fh, stp, open);
4370 mutex_unlock(&stp->st_mutex);
4371 release_open_stateid(stp);
4375 stp->st_clnt_odstate = find_or_hash_clnt_odstate(fp,
4377 if (stp->st_clnt_odstate == open->op_odstate)
4378 open->op_odstate = NULL;
4381 nfs4_inc_and_copy_stateid(&open->op_stateid, &stp->st_stid);
4382 mutex_unlock(&stp->st_mutex);
4384 if (nfsd4_has_session(&resp->cstate)) {
4385 if (open->op_deleg_want & NFS4_SHARE_WANT_NO_DELEG) {
4386 open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE_EXT;
4387 open->op_why_no_deleg = WND4_NOT_WANTED;
4393 * Attempt to hand out a delegation. No error return, because the
4394 * OPEN succeeds even if we fail.
4396 nfs4_open_delegation(current_fh, open, stp);
4400 dprintk("%s: stateid=" STATEID_FMT "\n", __func__,
4401 STATEID_VAL(&stp->st_stid.sc_stateid));
4403 /* 4.1 client trying to upgrade/downgrade delegation? */
4404 if (open->op_delegate_type == NFS4_OPEN_DELEGATE_NONE && dp &&
4405 open->op_deleg_want)
4406 nfsd4_deleg_xgrade_none_ext(open, dp);
4410 if (status == 0 && open->op_claim_type == NFS4_OPEN_CLAIM_PREVIOUS)
4411 open->op_openowner->oo_flags |= NFS4_OO_CONFIRMED;
4413 * To finish the open response, we just need to set the rflags.
4415 open->op_rflags = NFS4_OPEN_RESULT_LOCKTYPE_POSIX;
4416 if (!(open->op_openowner->oo_flags & NFS4_OO_CONFIRMED) &&
4417 !nfsd4_has_session(&resp->cstate))
4418 open->op_rflags |= NFS4_OPEN_RESULT_CONFIRM;
4420 nfs4_put_stid(&dp->dl_stid);
4422 nfs4_put_stid(&stp->st_stid);
4427 void nfsd4_cleanup_open_state(struct nfsd4_compound_state *cstate,
4428 struct nfsd4_open *open)
4430 if (open->op_openowner) {
4431 struct nfs4_stateowner *so = &open->op_openowner->oo_owner;
4433 nfsd4_cstate_assign_replay(cstate, so);
4434 nfs4_put_stateowner(so);
4437 kmem_cache_free(file_slab, open->op_file);
4439 nfs4_put_stid(&open->op_stp->st_stid);
4440 if (open->op_odstate)
4441 kmem_cache_free(odstate_slab, open->op_odstate);
4445 nfsd4_renew(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
4448 struct nfs4_client *clp;
4450 struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
4452 dprintk("process_renew(%08x/%08x): starting\n",
4453 clid->cl_boot, clid->cl_id);
4454 status = lookup_clientid(clid, cstate, nn);
4458 status = nfserr_cb_path_down;
4459 if (!list_empty(&clp->cl_delegations)
4460 && clp->cl_cb_state != NFSD4_CB_UP)
4468 nfsd4_end_grace(struct nfsd_net *nn)
4470 /* do nothing if grace period already ended */
4471 if (nn->grace_ended)
4474 dprintk("NFSD: end of grace period\n");
4475 nn->grace_ended = true;
4477 * If the server goes down again right now, an NFSv4
4478 * client will still be allowed to reclaim after it comes back up,
4479 * even if it hasn't yet had a chance to reclaim state this time.
4482 nfsd4_record_grace_done(nn);
4484 * At this point, NFSv4 clients can still reclaim. But if the
4485 * server crashes, any that have not yet reclaimed will be out
4486 * of luck on the next boot.
4488 * (NFSv4.1+ clients are considered to have reclaimed once they
4489 * call RECLAIM_COMPLETE. NFSv4.0 clients are considered to
4490 * have reclaimed after their first OPEN.)
4492 locks_end_grace(&nn->nfsd4_manager);
4494 * At this point, and once lockd and/or any other containers
4495 * exit their grace period, further reclaims will fail and
4496 * regular locking can resume.
4501 nfs4_laundromat(struct nfsd_net *nn)
4503 struct nfs4_client *clp;
4504 struct nfs4_openowner *oo;
4505 struct nfs4_delegation *dp;
4506 struct nfs4_ol_stateid *stp;
4507 struct list_head *pos, *next, reaplist;
4508 time_t cutoff = get_seconds() - nn->nfsd4_lease;
4509 time_t t, new_timeo = nn->nfsd4_lease;
4511 dprintk("NFSD: laundromat service - starting\n");
4512 nfsd4_end_grace(nn);
4513 INIT_LIST_HEAD(&reaplist);
4514 spin_lock(&nn->client_lock);
4515 list_for_each_safe(pos, next, &nn->client_lru) {
4516 clp = list_entry(pos, struct nfs4_client, cl_lru);
4517 if (time_after((unsigned long)clp->cl_time, (unsigned long)cutoff)) {
4518 t = clp->cl_time - cutoff;
4519 new_timeo = min(new_timeo, t);
4522 if (mark_client_expired_locked(clp)) {
4523 dprintk("NFSD: client in use (clientid %08x)\n",
4524 clp->cl_clientid.cl_id);
4527 list_add(&clp->cl_lru, &reaplist);
4529 spin_unlock(&nn->client_lock);
4530 list_for_each_safe(pos, next, &reaplist) {
4531 clp = list_entry(pos, struct nfs4_client, cl_lru);
4532 dprintk("NFSD: purging unused client (clientid %08x)\n",
4533 clp->cl_clientid.cl_id);
4534 list_del_init(&clp->cl_lru);
4537 spin_lock(&state_lock);
4538 list_for_each_safe(pos, next, &nn->del_recall_lru) {
4539 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
4540 if (time_after((unsigned long)dp->dl_time, (unsigned long)cutoff)) {
4541 t = dp->dl_time - cutoff;
4542 new_timeo = min(new_timeo, t);
4545 WARN_ON(!unhash_delegation_locked(dp));
4546 list_add(&dp->dl_recall_lru, &reaplist);
4548 spin_unlock(&state_lock);
4549 while (!list_empty(&reaplist)) {
4550 dp = list_first_entry(&reaplist, struct nfs4_delegation,
4552 list_del_init(&dp->dl_recall_lru);
4553 revoke_delegation(dp);
4556 spin_lock(&nn->client_lock);
4557 while (!list_empty(&nn->close_lru)) {
4558 oo = list_first_entry(&nn->close_lru, struct nfs4_openowner,
4560 if (time_after((unsigned long)oo->oo_time,
4561 (unsigned long)cutoff)) {
4562 t = oo->oo_time - cutoff;
4563 new_timeo = min(new_timeo, t);
4566 list_del_init(&oo->oo_close_lru);
4567 stp = oo->oo_last_closed_stid;
4568 oo->oo_last_closed_stid = NULL;
4569 spin_unlock(&nn->client_lock);
4570 nfs4_put_stid(&stp->st_stid);
4571 spin_lock(&nn->client_lock);
4573 spin_unlock(&nn->client_lock);
4575 new_timeo = max_t(time_t, new_timeo, NFSD_LAUNDROMAT_MINTIMEOUT);
4579 static struct workqueue_struct *laundry_wq;
4580 static void laundromat_main(struct work_struct *);
4583 laundromat_main(struct work_struct *laundry)
4586 struct delayed_work *dwork = to_delayed_work(laundry);
4587 struct nfsd_net *nn = container_of(dwork, struct nfsd_net,
4590 t = nfs4_laundromat(nn);
4591 dprintk("NFSD: laundromat_main - sleeping for %ld seconds\n", t);
4592 queue_delayed_work(laundry_wq, &nn->laundromat_work, t*HZ);
4595 static inline __be32 nfs4_check_fh(struct svc_fh *fhp, struct nfs4_stid *stp)
4597 if (!fh_match(&fhp->fh_handle, &stp->sc_file->fi_fhandle))
4598 return nfserr_bad_stateid;
4603 access_permit_read(struct nfs4_ol_stateid *stp)
4605 return test_access(NFS4_SHARE_ACCESS_READ, stp) ||
4606 test_access(NFS4_SHARE_ACCESS_BOTH, stp) ||
4607 test_access(NFS4_SHARE_ACCESS_WRITE, stp);
4611 access_permit_write(struct nfs4_ol_stateid *stp)
4613 return test_access(NFS4_SHARE_ACCESS_WRITE, stp) ||
4614 test_access(NFS4_SHARE_ACCESS_BOTH, stp);
4618 __be32 nfs4_check_openmode(struct nfs4_ol_stateid *stp, int flags)
4620 __be32 status = nfserr_openmode;
4622 /* For lock stateid's, we test the parent open, not the lock: */
4623 if (stp->st_openstp)
4624 stp = stp->st_openstp;
4625 if ((flags & WR_STATE) && !access_permit_write(stp))
4627 if ((flags & RD_STATE) && !access_permit_read(stp))
4634 static inline __be32
4635 check_special_stateids(struct net *net, svc_fh *current_fh, stateid_t *stateid, int flags)
4637 if (ONE_STATEID(stateid) && (flags & RD_STATE))
4639 else if (opens_in_grace(net)) {
4640 /* Answer in remaining cases depends on existence of
4641 * conflicting state; so we must wait out the grace period. */
4642 return nfserr_grace;
4643 } else if (flags & WR_STATE)
4644 return nfs4_share_conflict(current_fh,
4645 NFS4_SHARE_DENY_WRITE);
4646 else /* (flags & RD_STATE) && ZERO_STATEID(stateid) */
4647 return nfs4_share_conflict(current_fh,
4648 NFS4_SHARE_DENY_READ);
4652 * Allow READ/WRITE during grace period on recovered state only for files
4653 * that are not able to provide mandatory locking.
4656 grace_disallows_io(struct net *net, struct inode *inode)
4658 return opens_in_grace(net) && mandatory_lock(inode);
4661 static __be32 check_stateid_generation(stateid_t *in, stateid_t *ref, bool has_session)
4664 * When sessions are used the stateid generation number is ignored
4667 if (has_session && in->si_generation == 0)
4670 if (in->si_generation == ref->si_generation)
4673 /* If the client sends us a stateid from the future, it's buggy: */
4674 if (nfsd4_stateid_generation_after(in, ref))
4675 return nfserr_bad_stateid;
4677 * However, we could see a stateid from the past, even from a
4678 * non-buggy client. For example, if the client sends a lock
4679 * while some IO is outstanding, the lock may bump si_generation
4680 * while the IO is still in flight. The client could avoid that
4681 * situation by waiting for responses on all the IO requests,
4682 * but better performance may result in retrying IO that
4683 * receives an old_stateid error if requests are rarely
4684 * reordered in flight:
4686 return nfserr_old_stateid;
4689 static __be32 nfsd4_check_openowner_confirmed(struct nfs4_ol_stateid *ols)
4691 if (ols->st_stateowner->so_is_open_owner &&
4692 !(openowner(ols->st_stateowner)->oo_flags & NFS4_OO_CONFIRMED))
4693 return nfserr_bad_stateid;
4697 static __be32 nfsd4_validate_stateid(struct nfs4_client *cl, stateid_t *stateid)
4699 struct nfs4_stid *s;
4700 __be32 status = nfserr_bad_stateid;
4702 if (ZERO_STATEID(stateid) || ONE_STATEID(stateid))
4704 /* Client debugging aid. */
4705 if (!same_clid(&stateid->si_opaque.so_clid, &cl->cl_clientid)) {
4706 char addr_str[INET6_ADDRSTRLEN];
4707 rpc_ntop((struct sockaddr *)&cl->cl_addr, addr_str,
4709 pr_warn_ratelimited("NFSD: client %s testing state ID "
4710 "with incorrect client ID\n", addr_str);
4713 spin_lock(&cl->cl_lock);
4714 s = find_stateid_locked(cl, stateid);
4717 status = check_stateid_generation(stateid, &s->sc_stateid, 1);
4720 switch (s->sc_type) {
4721 case NFS4_DELEG_STID:
4724 case NFS4_REVOKED_DELEG_STID:
4725 status = nfserr_deleg_revoked;
4727 case NFS4_OPEN_STID:
4728 case NFS4_LOCK_STID:
4729 status = nfsd4_check_openowner_confirmed(openlockstateid(s));
4732 printk("unknown stateid type %x\n", s->sc_type);
4734 case NFS4_CLOSED_STID:
4735 case NFS4_CLOSED_DELEG_STID:
4736 status = nfserr_bad_stateid;
4739 spin_unlock(&cl->cl_lock);
4744 nfsd4_lookup_stateid(struct nfsd4_compound_state *cstate,
4745 stateid_t *stateid, unsigned char typemask,
4746 struct nfs4_stid **s, struct nfsd_net *nn)
4750 if (ZERO_STATEID(stateid) || ONE_STATEID(stateid))
4751 return nfserr_bad_stateid;
4752 status = lookup_clientid(&stateid->si_opaque.so_clid, cstate, nn);
4753 if (status == nfserr_stale_clientid) {
4754 if (cstate->session)
4755 return nfserr_bad_stateid;
4756 return nfserr_stale_stateid;
4760 *s = find_stateid_by_type(cstate->clp, stateid, typemask);
4762 return nfserr_bad_stateid;
4766 static struct file *
4767 nfs4_find_file(struct nfs4_stid *s, int flags)
4772 switch (s->sc_type) {
4773 case NFS4_DELEG_STID:
4774 if (WARN_ON_ONCE(!s->sc_file->fi_deleg_file))
4776 return get_file(s->sc_file->fi_deleg_file);
4777 case NFS4_OPEN_STID:
4778 case NFS4_LOCK_STID:
4779 if (flags & RD_STATE)
4780 return find_readable_file(s->sc_file);
4782 return find_writeable_file(s->sc_file);
4790 nfs4_check_olstateid(struct svc_fh *fhp, struct nfs4_ol_stateid *ols, int flags)
4794 status = nfsd4_check_openowner_confirmed(ols);
4797 return nfs4_check_openmode(ols, flags);
4801 nfs4_check_file(struct svc_rqst *rqstp, struct svc_fh *fhp, struct nfs4_stid *s,
4802 struct file **filpp, bool *tmp_file, int flags)
4804 int acc = (flags & RD_STATE) ? NFSD_MAY_READ : NFSD_MAY_WRITE;
4808 file = nfs4_find_file(s, flags);
4810 status = nfsd_permission(rqstp, fhp->fh_export, fhp->fh_dentry,
4811 acc | NFSD_MAY_OWNER_OVERRIDE);
4819 status = nfsd_open(rqstp, fhp, S_IFREG, acc, filpp);
4831 * Checks for stateid operations
4834 nfs4_preprocess_stateid_op(struct svc_rqst *rqstp,
4835 struct nfsd4_compound_state *cstate, struct svc_fh *fhp,
4836 stateid_t *stateid, int flags, struct file **filpp, bool *tmp_file)
4838 struct inode *ino = d_inode(fhp->fh_dentry);
4839 struct net *net = SVC_NET(rqstp);
4840 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
4841 struct nfs4_stid *s = NULL;
4849 if (grace_disallows_io(net, ino))
4850 return nfserr_grace;
4852 if (ZERO_STATEID(stateid) || ONE_STATEID(stateid)) {
4853 status = check_special_stateids(net, fhp, stateid, flags);
4857 status = nfsd4_lookup_stateid(cstate, stateid,
4858 NFS4_DELEG_STID|NFS4_OPEN_STID|NFS4_LOCK_STID,
4862 status = check_stateid_generation(stateid, &s->sc_stateid,
4863 nfsd4_has_session(cstate));
4867 switch (s->sc_type) {
4868 case NFS4_DELEG_STID:
4869 status = nfs4_check_delegmode(delegstateid(s), flags);
4871 case NFS4_OPEN_STID:
4872 case NFS4_LOCK_STID:
4873 status = nfs4_check_olstateid(fhp, openlockstateid(s), flags);
4876 status = nfserr_bad_stateid;
4881 status = nfs4_check_fh(fhp, s);
4884 if (!status && filpp)
4885 status = nfs4_check_file(rqstp, fhp, s, filpp, tmp_file, flags);
4893 * Test if the stateid is valid
4896 nfsd4_test_stateid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
4897 struct nfsd4_test_stateid *test_stateid)
4899 struct nfsd4_test_stateid_id *stateid;
4900 struct nfs4_client *cl = cstate->session->se_client;
4902 list_for_each_entry(stateid, &test_stateid->ts_stateid_list, ts_id_list)
4903 stateid->ts_id_status =
4904 nfsd4_validate_stateid(cl, &stateid->ts_id_stateid);
4910 nfsd4_free_stateid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
4911 struct nfsd4_free_stateid *free_stateid)
4913 stateid_t *stateid = &free_stateid->fr_stateid;
4914 struct nfs4_stid *s;
4915 struct nfs4_delegation *dp;
4916 struct nfs4_ol_stateid *stp;
4917 struct nfs4_client *cl = cstate->session->se_client;
4918 __be32 ret = nfserr_bad_stateid;
4920 spin_lock(&cl->cl_lock);
4921 s = find_stateid_locked(cl, stateid);
4924 switch (s->sc_type) {
4925 case NFS4_DELEG_STID:
4926 ret = nfserr_locks_held;
4928 case NFS4_OPEN_STID:
4929 ret = check_stateid_generation(stateid, &s->sc_stateid, 1);
4932 ret = nfserr_locks_held;
4934 case NFS4_LOCK_STID:
4935 ret = check_stateid_generation(stateid, &s->sc_stateid, 1);
4938 stp = openlockstateid(s);
4939 ret = nfserr_locks_held;
4940 if (check_for_locks(stp->st_stid.sc_file,
4941 lockowner(stp->st_stateowner)))
4943 WARN_ON(!unhash_lock_stateid(stp));
4944 spin_unlock(&cl->cl_lock);
4948 case NFS4_REVOKED_DELEG_STID:
4949 dp = delegstateid(s);
4950 list_del_init(&dp->dl_recall_lru);
4951 spin_unlock(&cl->cl_lock);
4955 /* Default falls through and returns nfserr_bad_stateid */
4958 spin_unlock(&cl->cl_lock);
4966 return (type == NFS4_READW_LT || type == NFS4_READ_LT) ?
4967 RD_STATE : WR_STATE;
4970 static __be32 nfs4_seqid_op_checks(struct nfsd4_compound_state *cstate, stateid_t *stateid, u32 seqid, struct nfs4_ol_stateid *stp)
4972 struct svc_fh *current_fh = &cstate->current_fh;
4973 struct nfs4_stateowner *sop = stp->st_stateowner;
4976 status = nfsd4_check_seqid(cstate, sop, seqid);
4979 if (stp->st_stid.sc_type == NFS4_CLOSED_STID
4980 || stp->st_stid.sc_type == NFS4_REVOKED_DELEG_STID)
4982 * "Closed" stateid's exist *only* to return
4983 * nfserr_replay_me from the previous step, and
4984 * revoked delegations are kept only for free_stateid.
4986 return nfserr_bad_stateid;
4987 mutex_lock(&stp->st_mutex);
4988 status = check_stateid_generation(stateid, &stp->st_stid.sc_stateid, nfsd4_has_session(cstate));
4989 if (status == nfs_ok)
4990 status = nfs4_check_fh(current_fh, &stp->st_stid);
4991 if (status != nfs_ok)
4992 mutex_unlock(&stp->st_mutex);
4997 * Checks for sequence id mutating operations.
5000 nfs4_preprocess_seqid_op(struct nfsd4_compound_state *cstate, u32 seqid,
5001 stateid_t *stateid, char typemask,
5002 struct nfs4_ol_stateid **stpp,
5003 struct nfsd_net *nn)
5006 struct nfs4_stid *s;
5007 struct nfs4_ol_stateid *stp = NULL;
5009 dprintk("NFSD: %s: seqid=%d stateid = " STATEID_FMT "\n", __func__,
5010 seqid, STATEID_VAL(stateid));
5013 status = nfsd4_lookup_stateid(cstate, stateid, typemask, &s, nn);
5016 stp = openlockstateid(s);
5017 nfsd4_cstate_assign_replay(cstate, stp->st_stateowner);
5019 status = nfs4_seqid_op_checks(cstate, stateid, seqid, stp);
5023 nfs4_put_stid(&stp->st_stid);
5027 static __be32 nfs4_preprocess_confirmed_seqid_op(struct nfsd4_compound_state *cstate, u32 seqid,
5028 stateid_t *stateid, struct nfs4_ol_stateid **stpp, struct nfsd_net *nn)
5031 struct nfs4_openowner *oo;
5032 struct nfs4_ol_stateid *stp;
5034 status = nfs4_preprocess_seqid_op(cstate, seqid, stateid,
5035 NFS4_OPEN_STID, &stp, nn);
5038 oo = openowner(stp->st_stateowner);
5039 if (!(oo->oo_flags & NFS4_OO_CONFIRMED)) {
5040 mutex_unlock(&stp->st_mutex);
5041 nfs4_put_stid(&stp->st_stid);
5042 return nfserr_bad_stateid;
5049 nfsd4_open_confirm(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
5050 struct nfsd4_open_confirm *oc)
5053 struct nfs4_openowner *oo;
5054 struct nfs4_ol_stateid *stp;
5055 struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
5057 dprintk("NFSD: nfsd4_open_confirm on file %pd\n",
5058 cstate->current_fh.fh_dentry);
5060 status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0);
5064 status = nfs4_preprocess_seqid_op(cstate,
5065 oc->oc_seqid, &oc->oc_req_stateid,
5066 NFS4_OPEN_STID, &stp, nn);
5069 oo = openowner(stp->st_stateowner);
5070 status = nfserr_bad_stateid;
5071 if (oo->oo_flags & NFS4_OO_CONFIRMED) {
5072 mutex_unlock(&stp->st_mutex);
5075 oo->oo_flags |= NFS4_OO_CONFIRMED;
5076 nfs4_inc_and_copy_stateid(&oc->oc_resp_stateid, &stp->st_stid);
5077 mutex_unlock(&stp->st_mutex);
5078 dprintk("NFSD: %s: success, seqid=%d stateid=" STATEID_FMT "\n",
5079 __func__, oc->oc_seqid, STATEID_VAL(&stp->st_stid.sc_stateid));
5081 nfsd4_client_record_create(oo->oo_owner.so_client);
5084 nfs4_put_stid(&stp->st_stid);
5086 nfsd4_bump_seqid(cstate, status);
5090 static inline void nfs4_stateid_downgrade_bit(struct nfs4_ol_stateid *stp, u32 access)
5092 if (!test_access(access, stp))
5094 nfs4_file_put_access(stp->st_stid.sc_file, access);
5095 clear_access(access, stp);
5098 static inline void nfs4_stateid_downgrade(struct nfs4_ol_stateid *stp, u32 to_access)
5100 switch (to_access) {
5101 case NFS4_SHARE_ACCESS_READ:
5102 nfs4_stateid_downgrade_bit(stp, NFS4_SHARE_ACCESS_WRITE);
5103 nfs4_stateid_downgrade_bit(stp, NFS4_SHARE_ACCESS_BOTH);
5105 case NFS4_SHARE_ACCESS_WRITE:
5106 nfs4_stateid_downgrade_bit(stp, NFS4_SHARE_ACCESS_READ);
5107 nfs4_stateid_downgrade_bit(stp, NFS4_SHARE_ACCESS_BOTH);
5109 case NFS4_SHARE_ACCESS_BOTH:
5117 nfsd4_open_downgrade(struct svc_rqst *rqstp,
5118 struct nfsd4_compound_state *cstate,
5119 struct nfsd4_open_downgrade *od)
5122 struct nfs4_ol_stateid *stp;
5123 struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
5125 dprintk("NFSD: nfsd4_open_downgrade on file %pd\n",
5126 cstate->current_fh.fh_dentry);
5128 /* We don't yet support WANT bits: */
5129 if (od->od_deleg_want)
5130 dprintk("NFSD: %s: od_deleg_want=0x%x ignored\n", __func__,
5133 status = nfs4_preprocess_confirmed_seqid_op(cstate, od->od_seqid,
5134 &od->od_stateid, &stp, nn);
5137 status = nfserr_inval;
5138 if (!test_access(od->od_share_access, stp)) {
5139 dprintk("NFSD: access not a subset of current bitmap: 0x%hhx, input access=%08x\n",
5140 stp->st_access_bmap, od->od_share_access);
5143 if (!test_deny(od->od_share_deny, stp)) {
5144 dprintk("NFSD: deny not a subset of current bitmap: 0x%hhx, input deny=%08x\n",
5145 stp->st_deny_bmap, od->od_share_deny);
5148 nfs4_stateid_downgrade(stp, od->od_share_access);
5149 reset_union_bmap_deny(od->od_share_deny, stp);
5150 nfs4_inc_and_copy_stateid(&od->od_stateid, &stp->st_stid);
5153 mutex_unlock(&stp->st_mutex);
5154 nfs4_put_stid(&stp->st_stid);
5156 nfsd4_bump_seqid(cstate, status);
5160 static void nfsd4_close_open_stateid(struct nfs4_ol_stateid *s)
5162 struct nfs4_client *clp = s->st_stid.sc_client;
5164 LIST_HEAD(reaplist);
5166 s->st_stid.sc_type = NFS4_CLOSED_STID;
5167 spin_lock(&clp->cl_lock);
5168 unhashed = unhash_open_stateid(s, &reaplist);
5170 if (clp->cl_minorversion) {
5172 put_ol_stateid_locked(s, &reaplist);
5173 spin_unlock(&clp->cl_lock);
5174 free_ol_stateid_reaplist(&reaplist);
5176 spin_unlock(&clp->cl_lock);
5177 free_ol_stateid_reaplist(&reaplist);
5179 move_to_close_lru(s, clp->net);
5184 * nfs4_unlock_state() called after encode
5187 nfsd4_close(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
5188 struct nfsd4_close *close)
5191 struct nfs4_ol_stateid *stp;
5192 struct net *net = SVC_NET(rqstp);
5193 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
5195 dprintk("NFSD: nfsd4_close on file %pd\n",
5196 cstate->current_fh.fh_dentry);
5198 status = nfs4_preprocess_seqid_op(cstate, close->cl_seqid,
5200 NFS4_OPEN_STID|NFS4_CLOSED_STID,
5202 nfsd4_bump_seqid(cstate, status);
5205 nfs4_inc_and_copy_stateid(&close->cl_stateid, &stp->st_stid);
5206 mutex_unlock(&stp->st_mutex);
5208 nfsd4_close_open_stateid(stp);
5210 /* put reference from nfs4_preprocess_seqid_op */
5211 nfs4_put_stid(&stp->st_stid);
5217 nfsd4_delegreturn(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
5218 struct nfsd4_delegreturn *dr)
5220 struct nfs4_delegation *dp;
5221 stateid_t *stateid = &dr->dr_stateid;
5222 struct nfs4_stid *s;
5224 struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
5226 if ((status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0)))
5229 status = nfsd4_lookup_stateid(cstate, stateid, NFS4_DELEG_STID, &s, nn);
5232 dp = delegstateid(s);
5233 status = check_stateid_generation(stateid, &dp->dl_stid.sc_stateid, nfsd4_has_session(cstate));
5237 destroy_delegation(dp);
5239 nfs4_put_stid(&dp->dl_stid);
5245 end_offset(u64 start, u64 len)
5250 return end >= start ? end: NFS4_MAX_UINT64;
5253 /* last octet in a range */
5255 last_byte_offset(u64 start, u64 len)
5261 return end > start ? end - 1: NFS4_MAX_UINT64;
5265 * TODO: Linux file offsets are _signed_ 64-bit quantities, which means that
5266 * we can't properly handle lock requests that go beyond the (2^63 - 1)-th
5267 * byte, because of sign extension problems. Since NFSv4 calls for 64-bit
5268 * locking, this prevents us from being completely protocol-compliant. The
5269 * real solution to this problem is to start using unsigned file offsets in
5270 * the VFS, but this is a very deep change!
5273 nfs4_transform_lock_offset(struct file_lock *lock)
5275 if (lock->fl_start < 0)
5276 lock->fl_start = OFFSET_MAX;
5277 if (lock->fl_end < 0)
5278 lock->fl_end = OFFSET_MAX;
5282 nfsd4_fl_get_owner(fl_owner_t owner)
5284 struct nfs4_lockowner *lo = (struct nfs4_lockowner *)owner;
5286 nfs4_get_stateowner(&lo->lo_owner);
5291 nfsd4_fl_put_owner(fl_owner_t owner)
5293 struct nfs4_lockowner *lo = (struct nfs4_lockowner *)owner;
5296 nfs4_put_stateowner(&lo->lo_owner);
5299 static const struct lock_manager_operations nfsd_posix_mng_ops = {
5300 .lm_get_owner = nfsd4_fl_get_owner,
5301 .lm_put_owner = nfsd4_fl_put_owner,
5305 nfs4_set_lock_denied(struct file_lock *fl, struct nfsd4_lock_denied *deny)
5307 struct nfs4_lockowner *lo;
5309 if (fl->fl_lmops == &nfsd_posix_mng_ops) {
5310 lo = (struct nfs4_lockowner *) fl->fl_owner;
5311 deny->ld_owner.data = kmemdup(lo->lo_owner.so_owner.data,
5312 lo->lo_owner.so_owner.len, GFP_KERNEL);
5313 if (!deny->ld_owner.data)
5314 /* We just don't care that much */
5316 deny->ld_owner.len = lo->lo_owner.so_owner.len;
5317 deny->ld_clientid = lo->lo_owner.so_client->cl_clientid;
5320 deny->ld_owner.len = 0;
5321 deny->ld_owner.data = NULL;
5322 deny->ld_clientid.cl_boot = 0;
5323 deny->ld_clientid.cl_id = 0;
5325 deny->ld_start = fl->fl_start;
5326 deny->ld_length = NFS4_MAX_UINT64;
5327 if (fl->fl_end != NFS4_MAX_UINT64)
5328 deny->ld_length = fl->fl_end - fl->fl_start + 1;
5329 deny->ld_type = NFS4_READ_LT;
5330 if (fl->fl_type != F_RDLCK)
5331 deny->ld_type = NFS4_WRITE_LT;
5334 static struct nfs4_lockowner *
5335 find_lockowner_str_locked(struct nfs4_client *clp, struct xdr_netobj *owner)
5337 unsigned int strhashval = ownerstr_hashval(owner);
5338 struct nfs4_stateowner *so;
5340 lockdep_assert_held(&clp->cl_lock);
5342 list_for_each_entry(so, &clp->cl_ownerstr_hashtbl[strhashval],
5344 if (so->so_is_open_owner)
5346 if (same_owner_str(so, owner))
5347 return lockowner(nfs4_get_stateowner(so));
5352 static struct nfs4_lockowner *
5353 find_lockowner_str(struct nfs4_client *clp, struct xdr_netobj *owner)
5355 struct nfs4_lockowner *lo;
5357 spin_lock(&clp->cl_lock);
5358 lo = find_lockowner_str_locked(clp, owner);
5359 spin_unlock(&clp->cl_lock);
5363 static void nfs4_unhash_lockowner(struct nfs4_stateowner *sop)
5365 unhash_lockowner_locked(lockowner(sop));
5368 static void nfs4_free_lockowner(struct nfs4_stateowner *sop)
5370 struct nfs4_lockowner *lo = lockowner(sop);
5372 kmem_cache_free(lockowner_slab, lo);
5375 static const struct nfs4_stateowner_operations lockowner_ops = {
5376 .so_unhash = nfs4_unhash_lockowner,
5377 .so_free = nfs4_free_lockowner,
5381 * Alloc a lock owner structure.
5382 * Called in nfsd4_lock - therefore, OPEN and OPEN_CONFIRM (if needed) has
5385 * strhashval = ownerstr_hashval
5387 static struct nfs4_lockowner *
5388 alloc_init_lock_stateowner(unsigned int strhashval, struct nfs4_client *clp,
5389 struct nfs4_ol_stateid *open_stp,
5390 struct nfsd4_lock *lock)
5392 struct nfs4_lockowner *lo, *ret;
5394 lo = alloc_stateowner(lockowner_slab, &lock->lk_new_owner, clp);
5397 INIT_LIST_HEAD(&lo->lo_owner.so_stateids);
5398 lo->lo_owner.so_is_open_owner = 0;
5399 lo->lo_owner.so_seqid = lock->lk_new_lock_seqid;
5400 lo->lo_owner.so_ops = &lockowner_ops;
5401 spin_lock(&clp->cl_lock);
5402 ret = find_lockowner_str_locked(clp, &lock->lk_new_owner);
5404 list_add(&lo->lo_owner.so_strhash,
5405 &clp->cl_ownerstr_hashtbl[strhashval]);
5408 nfs4_free_stateowner(&lo->lo_owner);
5410 spin_unlock(&clp->cl_lock);
5415 init_lock_stateid(struct nfs4_ol_stateid *stp, struct nfs4_lockowner *lo,
5416 struct nfs4_file *fp, struct inode *inode,
5417 struct nfs4_ol_stateid *open_stp)
5419 struct nfs4_client *clp = lo->lo_owner.so_client;
5421 lockdep_assert_held(&clp->cl_lock);
5423 atomic_inc(&stp->st_stid.sc_count);
5424 stp->st_stid.sc_type = NFS4_LOCK_STID;
5425 stp->st_stateowner = nfs4_get_stateowner(&lo->lo_owner);
5427 stp->st_stid.sc_file = fp;
5428 stp->st_stid.sc_free = nfs4_free_lock_stateid;
5429 stp->st_access_bmap = 0;
5430 stp->st_deny_bmap = open_stp->st_deny_bmap;
5431 stp->st_openstp = open_stp;
5432 mutex_init(&stp->st_mutex);
5433 list_add(&stp->st_locks, &open_stp->st_locks);
5434 list_add(&stp->st_perstateowner, &lo->lo_owner.so_stateids);
5435 spin_lock(&fp->fi_lock);
5436 list_add(&stp->st_perfile, &fp->fi_stateids);
5437 spin_unlock(&fp->fi_lock);
5440 static struct nfs4_ol_stateid *
5441 find_lock_stateid(struct nfs4_lockowner *lo, struct nfs4_file *fp)
5443 struct nfs4_ol_stateid *lst;
5444 struct nfs4_client *clp = lo->lo_owner.so_client;
5446 lockdep_assert_held(&clp->cl_lock);
5448 list_for_each_entry(lst, &lo->lo_owner.so_stateids, st_perstateowner) {
5449 if (lst->st_stid.sc_file == fp) {
5450 atomic_inc(&lst->st_stid.sc_count);
5457 static struct nfs4_ol_stateid *
5458 find_or_create_lock_stateid(struct nfs4_lockowner *lo, struct nfs4_file *fi,
5459 struct inode *inode, struct nfs4_ol_stateid *ost,
5462 struct nfs4_stid *ns = NULL;
5463 struct nfs4_ol_stateid *lst;
5464 struct nfs4_openowner *oo = openowner(ost->st_stateowner);
5465 struct nfs4_client *clp = oo->oo_owner.so_client;
5467 spin_lock(&clp->cl_lock);
5468 lst = find_lock_stateid(lo, fi);
5470 spin_unlock(&clp->cl_lock);
5471 ns = nfs4_alloc_stid(clp, stateid_slab);
5475 spin_lock(&clp->cl_lock);
5476 lst = find_lock_stateid(lo, fi);
5478 lst = openlockstateid(ns);
5479 init_lock_stateid(lst, lo, fi, inode, ost);
5484 spin_unlock(&clp->cl_lock);
5491 check_lock_length(u64 offset, u64 length)
5493 return ((length == 0) || ((length != NFS4_MAX_UINT64) &&
5494 (length > ~offset)));
5497 static void get_lock_access(struct nfs4_ol_stateid *lock_stp, u32 access)
5499 struct nfs4_file *fp = lock_stp->st_stid.sc_file;
5501 lockdep_assert_held(&fp->fi_lock);
5503 if (test_access(access, lock_stp))
5505 __nfs4_file_get_access(fp, access);
5506 set_access(access, lock_stp);
5510 lookup_or_create_lock_state(struct nfsd4_compound_state *cstate,
5511 struct nfs4_ol_stateid *ost,
5512 struct nfsd4_lock *lock,
5513 struct nfs4_ol_stateid **lst, bool *new)
5516 struct nfs4_file *fi = ost->st_stid.sc_file;
5517 struct nfs4_openowner *oo = openowner(ost->st_stateowner);
5518 struct nfs4_client *cl = oo->oo_owner.so_client;
5519 struct inode *inode = d_inode(cstate->current_fh.fh_dentry);
5520 struct nfs4_lockowner *lo;
5521 unsigned int strhashval;
5523 lo = find_lockowner_str(cl, &lock->lk_new_owner);
5525 strhashval = ownerstr_hashval(&lock->lk_new_owner);
5526 lo = alloc_init_lock_stateowner(strhashval, cl, ost, lock);
5528 return nfserr_jukebox;
5530 /* with an existing lockowner, seqids must be the same */
5531 status = nfserr_bad_seqid;
5532 if (!cstate->minorversion &&
5533 lock->lk_new_lock_seqid != lo->lo_owner.so_seqid)
5537 *lst = find_or_create_lock_stateid(lo, fi, inode, ost, new);
5539 status = nfserr_jukebox;
5544 nfs4_put_stateowner(&lo->lo_owner);
5552 nfsd4_lock(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
5553 struct nfsd4_lock *lock)
5555 struct nfs4_openowner *open_sop = NULL;
5556 struct nfs4_lockowner *lock_sop = NULL;
5557 struct nfs4_ol_stateid *lock_stp = NULL;
5558 struct nfs4_ol_stateid *open_stp = NULL;
5559 struct nfs4_file *fp;
5560 struct file *filp = NULL;
5561 struct file_lock *file_lock = NULL;
5562 struct file_lock *conflock = NULL;
5567 struct net *net = SVC_NET(rqstp);
5568 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
5570 dprintk("NFSD: nfsd4_lock: start=%Ld length=%Ld\n",
5571 (long long) lock->lk_offset,
5572 (long long) lock->lk_length);
5574 if (check_lock_length(lock->lk_offset, lock->lk_length))
5575 return nfserr_inval;
5577 if ((status = fh_verify(rqstp, &cstate->current_fh,
5578 S_IFREG, NFSD_MAY_LOCK))) {
5579 dprintk("NFSD: nfsd4_lock: permission denied!\n");
5583 if (lock->lk_is_new) {
5584 if (nfsd4_has_session(cstate))
5585 /* See rfc 5661 18.10.3: given clientid is ignored: */
5586 memcpy(&lock->lk_new_clientid,
5587 &cstate->session->se_client->cl_clientid,
5588 sizeof(clientid_t));
5590 status = nfserr_stale_clientid;
5591 if (STALE_CLIENTID(&lock->lk_new_clientid, nn))
5594 /* validate and update open stateid and open seqid */
5595 status = nfs4_preprocess_confirmed_seqid_op(cstate,
5596 lock->lk_new_open_seqid,
5597 &lock->lk_new_open_stateid,
5601 mutex_unlock(&open_stp->st_mutex);
5602 open_sop = openowner(open_stp->st_stateowner);
5603 status = nfserr_bad_stateid;
5604 if (!same_clid(&open_sop->oo_owner.so_client->cl_clientid,
5605 &lock->lk_new_clientid))
5607 status = lookup_or_create_lock_state(cstate, open_stp, lock,
5609 if (status == nfs_ok)
5610 mutex_lock(&lock_stp->st_mutex);
5612 status = nfs4_preprocess_seqid_op(cstate,
5613 lock->lk_old_lock_seqid,
5614 &lock->lk_old_lock_stateid,
5615 NFS4_LOCK_STID, &lock_stp, nn);
5619 lock_sop = lockowner(lock_stp->st_stateowner);
5621 lkflg = setlkflg(lock->lk_type);
5622 status = nfs4_check_openmode(lock_stp, lkflg);
5626 status = nfserr_grace;
5627 if (locks_in_grace(net) && !lock->lk_reclaim)
5629 status = nfserr_no_grace;
5630 if (!locks_in_grace(net) && lock->lk_reclaim)
5633 file_lock = locks_alloc_lock();
5635 dprintk("NFSD: %s: unable to allocate lock!\n", __func__);
5636 status = nfserr_jukebox;
5640 fp = lock_stp->st_stid.sc_file;
5641 switch (lock->lk_type) {
5644 spin_lock(&fp->fi_lock);
5645 filp = find_readable_file_locked(fp);
5647 get_lock_access(lock_stp, NFS4_SHARE_ACCESS_READ);
5648 spin_unlock(&fp->fi_lock);
5649 file_lock->fl_type = F_RDLCK;
5652 case NFS4_WRITEW_LT:
5653 spin_lock(&fp->fi_lock);
5654 filp = find_writeable_file_locked(fp);
5656 get_lock_access(lock_stp, NFS4_SHARE_ACCESS_WRITE);
5657 spin_unlock(&fp->fi_lock);
5658 file_lock->fl_type = F_WRLCK;
5661 status = nfserr_inval;
5665 status = nfserr_openmode;
5669 file_lock->fl_owner = (fl_owner_t)lockowner(nfs4_get_stateowner(&lock_sop->lo_owner));
5670 file_lock->fl_pid = current->tgid;
5671 file_lock->fl_file = filp;
5672 file_lock->fl_flags = FL_POSIX;
5673 file_lock->fl_lmops = &nfsd_posix_mng_ops;
5674 file_lock->fl_start = lock->lk_offset;
5675 file_lock->fl_end = last_byte_offset(lock->lk_offset, lock->lk_length);
5676 nfs4_transform_lock_offset(file_lock);
5678 conflock = locks_alloc_lock();
5680 dprintk("NFSD: %s: unable to allocate lock!\n", __func__);
5681 status = nfserr_jukebox;
5685 err = vfs_lock_file(filp, F_SETLK, file_lock, conflock);
5687 case 0: /* success! */
5688 nfs4_inc_and_copy_stateid(&lock->lk_resp_stateid, &lock_stp->st_stid);
5691 case (EAGAIN): /* conflock holds conflicting lock */
5692 status = nfserr_denied;
5693 dprintk("NFSD: nfsd4_lock: conflicting lock found!\n");
5694 nfs4_set_lock_denied(conflock, &lock->lk_denied);
5697 status = nfserr_deadlock;
5700 dprintk("NFSD: nfsd4_lock: vfs_lock_file() failed! status %d\n",err);
5701 status = nfserrno(err);
5708 /* Bump seqid manually if the 4.0 replay owner is openowner */
5709 if (cstate->replay_owner &&
5710 cstate->replay_owner != &lock_sop->lo_owner &&
5711 seqid_mutating_err(ntohl(status)))
5712 lock_sop->lo_owner.so_seqid++;
5714 mutex_unlock(&lock_stp->st_mutex);
5717 * If this is a new, never-before-used stateid, and we are
5718 * returning an error, then just go ahead and release it.
5721 release_lock_stateid(lock_stp);
5723 nfs4_put_stid(&lock_stp->st_stid);
5726 nfs4_put_stid(&open_stp->st_stid);
5727 nfsd4_bump_seqid(cstate, status);
5729 locks_free_lock(file_lock);
5731 locks_free_lock(conflock);
5736 * The NFSv4 spec allows a client to do a LOCKT without holding an OPEN,
5737 * so we do a temporary open here just to get an open file to pass to
5738 * vfs_test_lock. (Arguably perhaps test_lock should be done with an
5741 static __be32 nfsd_test_lock(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file_lock *lock)
5744 __be32 err = nfsd_open(rqstp, fhp, S_IFREG, NFSD_MAY_READ, &file);
5746 err = nfserrno(vfs_test_lock(file, lock));
5756 nfsd4_lockt(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
5757 struct nfsd4_lockt *lockt)
5759 struct file_lock *file_lock = NULL;
5760 struct nfs4_lockowner *lo = NULL;
5762 struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
5764 if (locks_in_grace(SVC_NET(rqstp)))
5765 return nfserr_grace;
5767 if (check_lock_length(lockt->lt_offset, lockt->lt_length))
5768 return nfserr_inval;
5770 if (!nfsd4_has_session(cstate)) {
5771 status = lookup_clientid(&lockt->lt_clientid, cstate, nn);
5776 if ((status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0)))
5779 file_lock = locks_alloc_lock();
5781 dprintk("NFSD: %s: unable to allocate lock!\n", __func__);
5782 status = nfserr_jukebox;
5786 switch (lockt->lt_type) {
5789 file_lock->fl_type = F_RDLCK;
5792 case NFS4_WRITEW_LT:
5793 file_lock->fl_type = F_WRLCK;
5796 dprintk("NFSD: nfs4_lockt: bad lock type!\n");
5797 status = nfserr_inval;
5801 lo = find_lockowner_str(cstate->clp, &lockt->lt_owner);
5803 file_lock->fl_owner = (fl_owner_t)lo;
5804 file_lock->fl_pid = current->tgid;
5805 file_lock->fl_flags = FL_POSIX;
5807 file_lock->fl_start = lockt->lt_offset;
5808 file_lock->fl_end = last_byte_offset(lockt->lt_offset, lockt->lt_length);
5810 nfs4_transform_lock_offset(file_lock);
5812 status = nfsd_test_lock(rqstp, &cstate->current_fh, file_lock);
5816 if (file_lock->fl_type != F_UNLCK) {
5817 status = nfserr_denied;
5818 nfs4_set_lock_denied(file_lock, &lockt->lt_denied);
5822 nfs4_put_stateowner(&lo->lo_owner);
5824 locks_free_lock(file_lock);
5829 nfsd4_locku(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
5830 struct nfsd4_locku *locku)
5832 struct nfs4_ol_stateid *stp;
5833 struct file *filp = NULL;
5834 struct file_lock *file_lock = NULL;
5837 struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
5839 dprintk("NFSD: nfsd4_locku: start=%Ld length=%Ld\n",
5840 (long long) locku->lu_offset,
5841 (long long) locku->lu_length);
5843 if (check_lock_length(locku->lu_offset, locku->lu_length))
5844 return nfserr_inval;
5846 status = nfs4_preprocess_seqid_op(cstate, locku->lu_seqid,
5847 &locku->lu_stateid, NFS4_LOCK_STID,
5851 filp = find_any_file(stp->st_stid.sc_file);
5853 status = nfserr_lock_range;
5856 file_lock = locks_alloc_lock();
5858 dprintk("NFSD: %s: unable to allocate lock!\n", __func__);
5859 status = nfserr_jukebox;
5863 file_lock->fl_type = F_UNLCK;
5864 file_lock->fl_owner = (fl_owner_t)lockowner(nfs4_get_stateowner(stp->st_stateowner));
5865 file_lock->fl_pid = current->tgid;
5866 file_lock->fl_file = filp;
5867 file_lock->fl_flags = FL_POSIX;
5868 file_lock->fl_lmops = &nfsd_posix_mng_ops;
5869 file_lock->fl_start = locku->lu_offset;
5871 file_lock->fl_end = last_byte_offset(locku->lu_offset,
5873 nfs4_transform_lock_offset(file_lock);
5875 err = vfs_lock_file(filp, F_SETLK, file_lock, NULL);
5877 dprintk("NFSD: nfs4_locku: vfs_lock_file failed!\n");
5880 nfs4_inc_and_copy_stateid(&locku->lu_stateid, &stp->st_stid);
5884 mutex_unlock(&stp->st_mutex);
5885 nfs4_put_stid(&stp->st_stid);
5887 nfsd4_bump_seqid(cstate, status);
5889 locks_free_lock(file_lock);
5893 status = nfserrno(err);
5899 * true: locks held by lockowner
5900 * false: no locks held by lockowner
5903 check_for_locks(struct nfs4_file *fp, struct nfs4_lockowner *lowner)
5905 struct file_lock *fl;
5907 struct file *filp = find_any_file(fp);
5908 struct inode *inode;
5909 struct file_lock_context *flctx;
5912 /* Any valid lock stateid should have some sort of access */
5917 inode = file_inode(filp);
5918 flctx = inode->i_flctx;
5920 if (flctx && !list_empty_careful(&flctx->flc_posix)) {
5921 spin_lock(&flctx->flc_lock);
5922 list_for_each_entry(fl, &flctx->flc_posix, fl_list) {
5923 if (fl->fl_owner == (fl_owner_t)lowner) {
5928 spin_unlock(&flctx->flc_lock);
5935 nfsd4_release_lockowner(struct svc_rqst *rqstp,
5936 struct nfsd4_compound_state *cstate,
5937 struct nfsd4_release_lockowner *rlockowner)
5939 clientid_t *clid = &rlockowner->rl_clientid;
5940 struct nfs4_stateowner *sop;
5941 struct nfs4_lockowner *lo = NULL;
5942 struct nfs4_ol_stateid *stp;
5943 struct xdr_netobj *owner = &rlockowner->rl_owner;
5944 unsigned int hashval = ownerstr_hashval(owner);
5946 struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
5947 struct nfs4_client *clp;
5949 dprintk("nfsd4_release_lockowner clientid: (%08x/%08x):\n",
5950 clid->cl_boot, clid->cl_id);
5952 status = lookup_clientid(clid, cstate, nn);
5957 /* Find the matching lock stateowner */
5958 spin_lock(&clp->cl_lock);
5959 list_for_each_entry(sop, &clp->cl_ownerstr_hashtbl[hashval],
5962 if (sop->so_is_open_owner || !same_owner_str(sop, owner))
5965 /* see if there are still any locks associated with it */
5966 lo = lockowner(sop);
5967 list_for_each_entry(stp, &sop->so_stateids, st_perstateowner) {
5968 if (check_for_locks(stp->st_stid.sc_file, lo)) {
5969 status = nfserr_locks_held;
5970 spin_unlock(&clp->cl_lock);
5975 nfs4_get_stateowner(sop);
5978 spin_unlock(&clp->cl_lock);
5980 release_lockowner(lo);
5984 static inline struct nfs4_client_reclaim *
5987 return kmalloc(sizeof(struct nfs4_client_reclaim), GFP_KERNEL);
5991 nfs4_has_reclaimed_state(const char *name, struct nfsd_net *nn)
5993 struct nfs4_client_reclaim *crp;
5995 crp = nfsd4_find_reclaim_client(name, nn);
5996 return (crp && crp->cr_clp);
6000 * failure => all reset bets are off, nfserr_no_grace...
6002 struct nfs4_client_reclaim *
6003 nfs4_client_to_reclaim(const char *name, struct nfsd_net *nn)
6005 unsigned int strhashval;
6006 struct nfs4_client_reclaim *crp;
6008 dprintk("NFSD nfs4_client_to_reclaim NAME: %.*s\n", HEXDIR_LEN, name);
6009 crp = alloc_reclaim();
6011 strhashval = clientstr_hashval(name);
6012 INIT_LIST_HEAD(&crp->cr_strhash);
6013 list_add(&crp->cr_strhash, &nn->reclaim_str_hashtbl[strhashval]);
6014 memcpy(crp->cr_recdir, name, HEXDIR_LEN);
6016 nn->reclaim_str_hashtbl_size++;
6022 nfs4_remove_reclaim_record(struct nfs4_client_reclaim *crp, struct nfsd_net *nn)
6024 list_del(&crp->cr_strhash);
6026 nn->reclaim_str_hashtbl_size--;
6030 nfs4_release_reclaim(struct nfsd_net *nn)
6032 struct nfs4_client_reclaim *crp = NULL;
6035 for (i = 0; i < CLIENT_HASH_SIZE; i++) {
6036 while (!list_empty(&nn->reclaim_str_hashtbl[i])) {
6037 crp = list_entry(nn->reclaim_str_hashtbl[i].next,
6038 struct nfs4_client_reclaim, cr_strhash);
6039 nfs4_remove_reclaim_record(crp, nn);
6042 WARN_ON_ONCE(nn->reclaim_str_hashtbl_size);
6046 * called from OPEN, CLAIM_PREVIOUS with a new clientid. */
6047 struct nfs4_client_reclaim *
6048 nfsd4_find_reclaim_client(const char *recdir, struct nfsd_net *nn)
6050 unsigned int strhashval;
6051 struct nfs4_client_reclaim *crp = NULL;
6053 dprintk("NFSD: nfs4_find_reclaim_client for recdir %s\n", recdir);
6055 strhashval = clientstr_hashval(recdir);
6056 list_for_each_entry(crp, &nn->reclaim_str_hashtbl[strhashval], cr_strhash) {
6057 if (same_name(crp->cr_recdir, recdir)) {
6065 * Called from OPEN. Look for clientid in reclaim list.
6068 nfs4_check_open_reclaim(clientid_t *clid,
6069 struct nfsd4_compound_state *cstate,
6070 struct nfsd_net *nn)
6074 /* find clientid in conf_id_hashtbl */
6075 status = lookup_clientid(clid, cstate, nn);
6077 return nfserr_reclaim_bad;
6079 if (test_bit(NFSD4_CLIENT_RECLAIM_COMPLETE, &cstate->clp->cl_flags))
6080 return nfserr_no_grace;
6082 if (nfsd4_client_record_check(cstate->clp))
6083 return nfserr_reclaim_bad;
6088 #ifdef CONFIG_NFSD_FAULT_INJECTION
6090 put_client(struct nfs4_client *clp)
6092 atomic_dec(&clp->cl_refcount);
6095 static struct nfs4_client *
6096 nfsd_find_client(struct sockaddr_storage *addr, size_t addr_size)
6098 struct nfs4_client *clp;
6099 struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6102 if (!nfsd_netns_ready(nn))
6105 list_for_each_entry(clp, &nn->client_lru, cl_lru) {
6106 if (memcmp(&clp->cl_addr, addr, addr_size) == 0)
6113 nfsd_inject_print_clients(void)
6115 struct nfs4_client *clp;
6117 struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6119 char buf[INET6_ADDRSTRLEN];
6121 if (!nfsd_netns_ready(nn))
6124 spin_lock(&nn->client_lock);
6125 list_for_each_entry(clp, &nn->client_lru, cl_lru) {
6126 rpc_ntop((struct sockaddr *)&clp->cl_addr, buf, sizeof(buf));
6127 pr_info("NFS Client: %s\n", buf);
6130 spin_unlock(&nn->client_lock);
6136 nfsd_inject_forget_client(struct sockaddr_storage *addr, size_t addr_size)
6139 struct nfs4_client *clp;
6140 struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6143 if (!nfsd_netns_ready(nn))
6146 spin_lock(&nn->client_lock);
6147 clp = nfsd_find_client(addr, addr_size);
6149 if (mark_client_expired_locked(clp) == nfs_ok)
6154 spin_unlock(&nn->client_lock);
6163 nfsd_inject_forget_clients(u64 max)
6166 struct nfs4_client *clp, *next;
6167 struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6169 LIST_HEAD(reaplist);
6171 if (!nfsd_netns_ready(nn))
6174 spin_lock(&nn->client_lock);
6175 list_for_each_entry_safe(clp, next, &nn->client_lru, cl_lru) {
6176 if (mark_client_expired_locked(clp) == nfs_ok) {
6177 list_add(&clp->cl_lru, &reaplist);
6178 if (max != 0 && ++count >= max)
6182 spin_unlock(&nn->client_lock);
6184 list_for_each_entry_safe(clp, next, &reaplist, cl_lru)
6190 static void nfsd_print_count(struct nfs4_client *clp, unsigned int count,
6193 char buf[INET6_ADDRSTRLEN];
6194 rpc_ntop((struct sockaddr *)&clp->cl_addr, buf, sizeof(buf));
6195 printk(KERN_INFO "NFS Client: %s has %u %s\n", buf, count, type);
6199 nfsd_inject_add_lock_to_list(struct nfs4_ol_stateid *lst,
6200 struct list_head *collect)
6202 struct nfs4_client *clp = lst->st_stid.sc_client;
6203 struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6209 lockdep_assert_held(&nn->client_lock);
6210 atomic_inc(&clp->cl_refcount);
6211 list_add(&lst->st_locks, collect);
6214 static u64 nfsd_foreach_client_lock(struct nfs4_client *clp, u64 max,
6215 struct list_head *collect,
6216 bool (*func)(struct nfs4_ol_stateid *))
6218 struct nfs4_openowner *oop;
6219 struct nfs4_ol_stateid *stp, *st_next;
6220 struct nfs4_ol_stateid *lst, *lst_next;
6223 spin_lock(&clp->cl_lock);
6224 list_for_each_entry(oop, &clp->cl_openowners, oo_perclient) {
6225 list_for_each_entry_safe(stp, st_next,
6226 &oop->oo_owner.so_stateids, st_perstateowner) {
6227 list_for_each_entry_safe(lst, lst_next,
6228 &stp->st_locks, st_locks) {
6231 nfsd_inject_add_lock_to_list(lst,
6236 * Despite the fact that these functions deal
6237 * with 64-bit integers for "count", we must
6238 * ensure that it doesn't blow up the
6239 * clp->cl_refcount. Throw a warning if we
6240 * start to approach INT_MAX here.
6242 WARN_ON_ONCE(count == (INT_MAX / 2));
6249 spin_unlock(&clp->cl_lock);
6255 nfsd_collect_client_locks(struct nfs4_client *clp, struct list_head *collect,
6258 return nfsd_foreach_client_lock(clp, max, collect, unhash_lock_stateid);
6262 nfsd_print_client_locks(struct nfs4_client *clp)
6264 u64 count = nfsd_foreach_client_lock(clp, 0, NULL, NULL);
6265 nfsd_print_count(clp, count, "locked files");
6270 nfsd_inject_print_locks(void)
6272 struct nfs4_client *clp;
6274 struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6277 if (!nfsd_netns_ready(nn))
6280 spin_lock(&nn->client_lock);
6281 list_for_each_entry(clp, &nn->client_lru, cl_lru)
6282 count += nfsd_print_client_locks(clp);
6283 spin_unlock(&nn->client_lock);
6289 nfsd_reap_locks(struct list_head *reaplist)
6291 struct nfs4_client *clp;
6292 struct nfs4_ol_stateid *stp, *next;
6294 list_for_each_entry_safe(stp, next, reaplist, st_locks) {
6295 list_del_init(&stp->st_locks);
6296 clp = stp->st_stid.sc_client;
6297 nfs4_put_stid(&stp->st_stid);
6303 nfsd_inject_forget_client_locks(struct sockaddr_storage *addr, size_t addr_size)
6305 unsigned int count = 0;
6306 struct nfs4_client *clp;
6307 struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6309 LIST_HEAD(reaplist);
6311 if (!nfsd_netns_ready(nn))
6314 spin_lock(&nn->client_lock);
6315 clp = nfsd_find_client(addr, addr_size);
6317 count = nfsd_collect_client_locks(clp, &reaplist, 0);
6318 spin_unlock(&nn->client_lock);
6319 nfsd_reap_locks(&reaplist);
6324 nfsd_inject_forget_locks(u64 max)
6327 struct nfs4_client *clp;
6328 struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6330 LIST_HEAD(reaplist);
6332 if (!nfsd_netns_ready(nn))
6335 spin_lock(&nn->client_lock);
6336 list_for_each_entry(clp, &nn->client_lru, cl_lru) {
6337 count += nfsd_collect_client_locks(clp, &reaplist, max - count);
6338 if (max != 0 && count >= max)
6341 spin_unlock(&nn->client_lock);
6342 nfsd_reap_locks(&reaplist);
6347 nfsd_foreach_client_openowner(struct nfs4_client *clp, u64 max,
6348 struct list_head *collect,
6349 void (*func)(struct nfs4_openowner *))
6351 struct nfs4_openowner *oop, *next;
6352 struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6356 lockdep_assert_held(&nn->client_lock);
6358 spin_lock(&clp->cl_lock);
6359 list_for_each_entry_safe(oop, next, &clp->cl_openowners, oo_perclient) {
6363 atomic_inc(&clp->cl_refcount);
6364 list_add(&oop->oo_perclient, collect);
6369 * Despite the fact that these functions deal with
6370 * 64-bit integers for "count", we must ensure that
6371 * it doesn't blow up the clp->cl_refcount. Throw a
6372 * warning if we start to approach INT_MAX here.
6374 WARN_ON_ONCE(count == (INT_MAX / 2));
6378 spin_unlock(&clp->cl_lock);
6384 nfsd_print_client_openowners(struct nfs4_client *clp)
6386 u64 count = nfsd_foreach_client_openowner(clp, 0, NULL, NULL);
6388 nfsd_print_count(clp, count, "openowners");
6393 nfsd_collect_client_openowners(struct nfs4_client *clp,
6394 struct list_head *collect, u64 max)
6396 return nfsd_foreach_client_openowner(clp, max, collect,
6397 unhash_openowner_locked);
6401 nfsd_inject_print_openowners(void)
6403 struct nfs4_client *clp;
6405 struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6408 if (!nfsd_netns_ready(nn))
6411 spin_lock(&nn->client_lock);
6412 list_for_each_entry(clp, &nn->client_lru, cl_lru)
6413 count += nfsd_print_client_openowners(clp);
6414 spin_unlock(&nn->client_lock);
6420 nfsd_reap_openowners(struct list_head *reaplist)
6422 struct nfs4_client *clp;
6423 struct nfs4_openowner *oop, *next;
6425 list_for_each_entry_safe(oop, next, reaplist, oo_perclient) {
6426 list_del_init(&oop->oo_perclient);
6427 clp = oop->oo_owner.so_client;
6428 release_openowner(oop);
6434 nfsd_inject_forget_client_openowners(struct sockaddr_storage *addr,
6437 unsigned int count = 0;
6438 struct nfs4_client *clp;
6439 struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6441 LIST_HEAD(reaplist);
6443 if (!nfsd_netns_ready(nn))
6446 spin_lock(&nn->client_lock);
6447 clp = nfsd_find_client(addr, addr_size);
6449 count = nfsd_collect_client_openowners(clp, &reaplist, 0);
6450 spin_unlock(&nn->client_lock);
6451 nfsd_reap_openowners(&reaplist);
6456 nfsd_inject_forget_openowners(u64 max)
6459 struct nfs4_client *clp;
6460 struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6462 LIST_HEAD(reaplist);
6464 if (!nfsd_netns_ready(nn))
6467 spin_lock(&nn->client_lock);
6468 list_for_each_entry(clp, &nn->client_lru, cl_lru) {
6469 count += nfsd_collect_client_openowners(clp, &reaplist,
6471 if (max != 0 && count >= max)
6474 spin_unlock(&nn->client_lock);
6475 nfsd_reap_openowners(&reaplist);
6479 static u64 nfsd_find_all_delegations(struct nfs4_client *clp, u64 max,
6480 struct list_head *victims)
6482 struct nfs4_delegation *dp, *next;
6483 struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6487 lockdep_assert_held(&nn->client_lock);
6489 spin_lock(&state_lock);
6490 list_for_each_entry_safe(dp, next, &clp->cl_delegations, dl_perclnt) {
6493 * It's not safe to mess with delegations that have a
6494 * non-zero dl_time. They might have already been broken
6495 * and could be processed by the laundromat outside of
6496 * the state_lock. Just leave them be.
6498 if (dp->dl_time != 0)
6501 atomic_inc(&clp->cl_refcount);
6502 WARN_ON(!unhash_delegation_locked(dp));
6503 list_add(&dp->dl_recall_lru, victims);
6507 * Despite the fact that these functions deal with
6508 * 64-bit integers for "count", we must ensure that
6509 * it doesn't blow up the clp->cl_refcount. Throw a
6510 * warning if we start to approach INT_MAX here.
6512 WARN_ON_ONCE(count == (INT_MAX / 2));
6516 spin_unlock(&state_lock);
6521 nfsd_print_client_delegations(struct nfs4_client *clp)
6523 u64 count = nfsd_find_all_delegations(clp, 0, NULL);
6525 nfsd_print_count(clp, count, "delegations");
6530 nfsd_inject_print_delegations(void)
6532 struct nfs4_client *clp;
6534 struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6537 if (!nfsd_netns_ready(nn))
6540 spin_lock(&nn->client_lock);
6541 list_for_each_entry(clp, &nn->client_lru, cl_lru)
6542 count += nfsd_print_client_delegations(clp);
6543 spin_unlock(&nn->client_lock);
6549 nfsd_forget_delegations(struct list_head *reaplist)
6551 struct nfs4_client *clp;
6552 struct nfs4_delegation *dp, *next;
6554 list_for_each_entry_safe(dp, next, reaplist, dl_recall_lru) {
6555 list_del_init(&dp->dl_recall_lru);
6556 clp = dp->dl_stid.sc_client;
6557 revoke_delegation(dp);
6563 nfsd_inject_forget_client_delegations(struct sockaddr_storage *addr,
6567 struct nfs4_client *clp;
6568 struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6570 LIST_HEAD(reaplist);
6572 if (!nfsd_netns_ready(nn))
6575 spin_lock(&nn->client_lock);
6576 clp = nfsd_find_client(addr, addr_size);
6578 count = nfsd_find_all_delegations(clp, 0, &reaplist);
6579 spin_unlock(&nn->client_lock);
6581 nfsd_forget_delegations(&reaplist);
6586 nfsd_inject_forget_delegations(u64 max)
6589 struct nfs4_client *clp;
6590 struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6592 LIST_HEAD(reaplist);
6594 if (!nfsd_netns_ready(nn))
6597 spin_lock(&nn->client_lock);
6598 list_for_each_entry(clp, &nn->client_lru, cl_lru) {
6599 count += nfsd_find_all_delegations(clp, max - count, &reaplist);
6600 if (max != 0 && count >= max)
6603 spin_unlock(&nn->client_lock);
6604 nfsd_forget_delegations(&reaplist);
6609 nfsd_recall_delegations(struct list_head *reaplist)
6611 struct nfs4_client *clp;
6612 struct nfs4_delegation *dp, *next;
6614 list_for_each_entry_safe(dp, next, reaplist, dl_recall_lru) {
6615 list_del_init(&dp->dl_recall_lru);
6616 clp = dp->dl_stid.sc_client;
6618 * We skipped all entries that had a zero dl_time before,
6619 * so we can now reset the dl_time back to 0. If a delegation
6620 * break comes in now, then it won't make any difference since
6621 * we're recalling it either way.
6623 spin_lock(&state_lock);
6625 spin_unlock(&state_lock);
6626 nfsd_break_one_deleg(dp);
6632 nfsd_inject_recall_client_delegations(struct sockaddr_storage *addr,
6636 struct nfs4_client *clp;
6637 struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6639 LIST_HEAD(reaplist);
6641 if (!nfsd_netns_ready(nn))
6644 spin_lock(&nn->client_lock);
6645 clp = nfsd_find_client(addr, addr_size);
6647 count = nfsd_find_all_delegations(clp, 0, &reaplist);
6648 spin_unlock(&nn->client_lock);
6650 nfsd_recall_delegations(&reaplist);
6655 nfsd_inject_recall_delegations(u64 max)
6658 struct nfs4_client *clp, *next;
6659 struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6661 LIST_HEAD(reaplist);
6663 if (!nfsd_netns_ready(nn))
6666 spin_lock(&nn->client_lock);
6667 list_for_each_entry_safe(clp, next, &nn->client_lru, cl_lru) {
6668 count += nfsd_find_all_delegations(clp, max - count, &reaplist);
6669 if (max != 0 && ++count >= max)
6672 spin_unlock(&nn->client_lock);
6673 nfsd_recall_delegations(&reaplist);
6676 #endif /* CONFIG_NFSD_FAULT_INJECTION */
6679 * Since the lifetime of a delegation isn't limited to that of an open, a
6680 * client may quite reasonably hang on to a delegation as long as it has
6681 * the inode cached. This becomes an obvious problem the first time a
6682 * client's inode cache approaches the size of the server's total memory.
6684 * For now we avoid this problem by imposing a hard limit on the number
6685 * of delegations, which varies according to the server's memory size.
6688 set_max_delegations(void)
6691 * Allow at most 4 delegations per megabyte of RAM. Quick
6692 * estimates suggest that in the worst case (where every delegation
6693 * is for a different inode), a delegation could take about 1.5K,
6694 * giving a worst case usage of about 6% of memory.
6696 max_delegations = nr_free_buffer_pages() >> (20 - 2 - PAGE_SHIFT);
6699 static int nfs4_state_create_net(struct net *net)
6701 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
6704 nn->conf_id_hashtbl = kmalloc(sizeof(struct list_head) *
6705 CLIENT_HASH_SIZE, GFP_KERNEL);
6706 if (!nn->conf_id_hashtbl)
6708 nn->unconf_id_hashtbl = kmalloc(sizeof(struct list_head) *
6709 CLIENT_HASH_SIZE, GFP_KERNEL);
6710 if (!nn->unconf_id_hashtbl)
6712 nn->sessionid_hashtbl = kmalloc(sizeof(struct list_head) *
6713 SESSION_HASH_SIZE, GFP_KERNEL);
6714 if (!nn->sessionid_hashtbl)
6717 for (i = 0; i < CLIENT_HASH_SIZE; i++) {
6718 INIT_LIST_HEAD(&nn->conf_id_hashtbl[i]);
6719 INIT_LIST_HEAD(&nn->unconf_id_hashtbl[i]);
6721 for (i = 0; i < SESSION_HASH_SIZE; i++)
6722 INIT_LIST_HEAD(&nn->sessionid_hashtbl[i]);
6723 nn->conf_name_tree = RB_ROOT;
6724 nn->unconf_name_tree = RB_ROOT;
6725 INIT_LIST_HEAD(&nn->client_lru);
6726 INIT_LIST_HEAD(&nn->close_lru);
6727 INIT_LIST_HEAD(&nn->del_recall_lru);
6728 spin_lock_init(&nn->client_lock);
6730 INIT_DELAYED_WORK(&nn->laundromat_work, laundromat_main);
6736 kfree(nn->unconf_id_hashtbl);
6738 kfree(nn->conf_id_hashtbl);
6744 nfs4_state_destroy_net(struct net *net)
6747 struct nfs4_client *clp = NULL;
6748 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
6750 for (i = 0; i < CLIENT_HASH_SIZE; i++) {
6751 while (!list_empty(&nn->conf_id_hashtbl[i])) {
6752 clp = list_entry(nn->conf_id_hashtbl[i].next, struct nfs4_client, cl_idhash);
6753 destroy_client(clp);
6757 for (i = 0; i < CLIENT_HASH_SIZE; i++) {
6758 while (!list_empty(&nn->unconf_id_hashtbl[i])) {
6759 clp = list_entry(nn->unconf_id_hashtbl[i].next, struct nfs4_client, cl_idhash);
6760 destroy_client(clp);
6764 kfree(nn->sessionid_hashtbl);
6765 kfree(nn->unconf_id_hashtbl);
6766 kfree(nn->conf_id_hashtbl);
6771 nfs4_state_start_net(struct net *net)
6773 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
6776 ret = nfs4_state_create_net(net);
6779 nn->boot_time = get_seconds();
6780 nn->grace_ended = false;
6781 nn->nfsd4_manager.block_opens = true;
6782 locks_start_grace(net, &nn->nfsd4_manager);
6783 nfsd4_client_tracking_init(net);
6784 printk(KERN_INFO "NFSD: starting %ld-second grace period (net %p)\n",
6785 nn->nfsd4_grace, net);
6786 queue_delayed_work(laundry_wq, &nn->laundromat_work, nn->nfsd4_grace * HZ);
6790 /* initialization to perform when the nfsd service is started: */
6793 nfs4_state_start(void)
6797 ret = set_callback_cred();
6800 laundry_wq = alloc_workqueue("%s", WQ_UNBOUND, 0, "nfsd4");
6801 if (laundry_wq == NULL) {
6805 ret = nfsd4_create_callback_queue();
6807 goto out_free_laundry;
6809 set_max_delegations();
6814 destroy_workqueue(laundry_wq);
6820 nfs4_state_shutdown_net(struct net *net)
6822 struct nfs4_delegation *dp = NULL;
6823 struct list_head *pos, *next, reaplist;
6824 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
6826 cancel_delayed_work_sync(&nn->laundromat_work);
6827 locks_end_grace(&nn->nfsd4_manager);
6829 INIT_LIST_HEAD(&reaplist);
6830 spin_lock(&state_lock);
6831 list_for_each_safe(pos, next, &nn->del_recall_lru) {
6832 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
6833 WARN_ON(!unhash_delegation_locked(dp));
6834 list_add(&dp->dl_recall_lru, &reaplist);
6836 spin_unlock(&state_lock);
6837 list_for_each_safe(pos, next, &reaplist) {
6838 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
6839 list_del_init(&dp->dl_recall_lru);
6840 put_clnt_odstate(dp->dl_clnt_odstate);
6841 nfs4_put_deleg_lease(dp->dl_stid.sc_file);
6842 nfs4_put_stid(&dp->dl_stid);
6845 nfsd4_client_tracking_exit(net);
6846 nfs4_state_destroy_net(net);
6850 nfs4_state_shutdown(void)
6852 destroy_workqueue(laundry_wq);
6853 nfsd4_destroy_callback_queue();
6857 get_stateid(struct nfsd4_compound_state *cstate, stateid_t *stateid)
6859 if (HAS_STATE_ID(cstate, CURRENT_STATE_ID_FLAG) && CURRENT_STATEID(stateid))
6860 memcpy(stateid, &cstate->current_stateid, sizeof(stateid_t));
6864 put_stateid(struct nfsd4_compound_state *cstate, stateid_t *stateid)
6866 if (cstate->minorversion) {
6867 memcpy(&cstate->current_stateid, stateid, sizeof(stateid_t));
6868 SET_STATE_ID(cstate, CURRENT_STATE_ID_FLAG);
6873 clear_current_stateid(struct nfsd4_compound_state *cstate)
6875 CLEAR_STATE_ID(cstate, CURRENT_STATE_ID_FLAG);
6879 * functions to set current state id
6882 nfsd4_set_opendowngradestateid(struct nfsd4_compound_state *cstate, struct nfsd4_open_downgrade *odp)
6884 put_stateid(cstate, &odp->od_stateid);
6888 nfsd4_set_openstateid(struct nfsd4_compound_state *cstate, struct nfsd4_open *open)
6890 put_stateid(cstate, &open->op_stateid);
6894 nfsd4_set_closestateid(struct nfsd4_compound_state *cstate, struct nfsd4_close *close)
6896 put_stateid(cstate, &close->cl_stateid);
6900 nfsd4_set_lockstateid(struct nfsd4_compound_state *cstate, struct nfsd4_lock *lock)
6902 put_stateid(cstate, &lock->lk_resp_stateid);
6906 * functions to consume current state id
6910 nfsd4_get_opendowngradestateid(struct nfsd4_compound_state *cstate, struct nfsd4_open_downgrade *odp)
6912 get_stateid(cstate, &odp->od_stateid);
6916 nfsd4_get_delegreturnstateid(struct nfsd4_compound_state *cstate, struct nfsd4_delegreturn *drp)
6918 get_stateid(cstate, &drp->dr_stateid);
6922 nfsd4_get_freestateid(struct nfsd4_compound_state *cstate, struct nfsd4_free_stateid *fsp)
6924 get_stateid(cstate, &fsp->fr_stateid);
6928 nfsd4_get_setattrstateid(struct nfsd4_compound_state *cstate, struct nfsd4_setattr *setattr)
6930 get_stateid(cstate, &setattr->sa_stateid);
6934 nfsd4_get_closestateid(struct nfsd4_compound_state *cstate, struct nfsd4_close *close)
6936 get_stateid(cstate, &close->cl_stateid);
6940 nfsd4_get_lockustateid(struct nfsd4_compound_state *cstate, struct nfsd4_locku *locku)
6942 get_stateid(cstate, &locku->lu_stateid);
6946 nfsd4_get_readstateid(struct nfsd4_compound_state *cstate, struct nfsd4_read *read)
6948 get_stateid(cstate, &read->rd_stateid);
6952 nfsd4_get_writestateid(struct nfsd4_compound_state *cstate, struct nfsd4_write *write)
6954 get_stateid(cstate, &write->wr_stateid);