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 = {
66 static const stateid_t close_stateid = {
67 .si_generation = 0xffffffffU,
70 static u64 current_sessionid = 1;
72 #define ZERO_STATEID(stateid) (!memcmp((stateid), &zero_stateid, sizeof(stateid_t)))
73 #define ONE_STATEID(stateid) (!memcmp((stateid), &one_stateid, sizeof(stateid_t)))
74 #define CURRENT_STATEID(stateid) (!memcmp((stateid), ¤tstateid, sizeof(stateid_t)))
75 #define CLOSE_STATEID(stateid) (!memcmp((stateid), &close_stateid, sizeof(stateid_t)))
77 /* forward declarations */
78 static bool check_for_locks(struct nfs4_file *fp, struct nfs4_lockowner *lowner);
79 static void nfs4_free_ol_stateid(struct nfs4_stid *stid);
84 * Currently used for the del_recall_lru and file hash table. In an
85 * effort to decrease the scope of the client_mutex, this spinlock may
86 * eventually cover more:
88 static DEFINE_SPINLOCK(state_lock);
90 enum nfsd4_st_mutex_lock_subclass {
91 OPEN_STATEID_MUTEX = 0,
92 LOCK_STATEID_MUTEX = 1,
96 * A waitqueue for all in-progress 4.0 CLOSE operations that are waiting for
97 * the refcount on the open stateid to drop.
99 static DECLARE_WAIT_QUEUE_HEAD(close_wq);
101 static struct kmem_cache *client_slab;
102 static struct kmem_cache *openowner_slab;
103 static struct kmem_cache *lockowner_slab;
104 static struct kmem_cache *file_slab;
105 static struct kmem_cache *stateid_slab;
106 static struct kmem_cache *deleg_slab;
107 static struct kmem_cache *odstate_slab;
109 static void free_session(struct nfsd4_session *);
111 static const struct nfsd4_callback_ops nfsd4_cb_recall_ops;
112 static const struct nfsd4_callback_ops nfsd4_cb_notify_lock_ops;
114 static bool is_session_dead(struct nfsd4_session *ses)
116 return ses->se_flags & NFS4_SESSION_DEAD;
119 static __be32 mark_session_dead_locked(struct nfsd4_session *ses, int ref_held_by_me)
121 if (atomic_read(&ses->se_ref) > ref_held_by_me)
122 return nfserr_jukebox;
123 ses->se_flags |= NFS4_SESSION_DEAD;
127 static bool is_client_expired(struct nfs4_client *clp)
129 return clp->cl_time == 0;
132 static __be32 get_client_locked(struct nfs4_client *clp)
134 struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
136 lockdep_assert_held(&nn->client_lock);
138 if (is_client_expired(clp))
139 return nfserr_expired;
140 atomic_inc(&clp->cl_refcount);
144 /* must be called under the client_lock */
146 renew_client_locked(struct nfs4_client *clp)
148 struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
150 if (is_client_expired(clp)) {
152 printk("%s: client (clientid %08x/%08x) already expired\n",
154 clp->cl_clientid.cl_boot,
155 clp->cl_clientid.cl_id);
159 dprintk("renewing client (clientid %08x/%08x)\n",
160 clp->cl_clientid.cl_boot,
161 clp->cl_clientid.cl_id);
162 list_move_tail(&clp->cl_lru, &nn->client_lru);
163 clp->cl_time = get_seconds();
166 static void put_client_renew_locked(struct nfs4_client *clp)
168 struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
170 lockdep_assert_held(&nn->client_lock);
172 if (!atomic_dec_and_test(&clp->cl_refcount))
174 if (!is_client_expired(clp))
175 renew_client_locked(clp);
178 static void put_client_renew(struct nfs4_client *clp)
180 struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
182 if (!atomic_dec_and_lock(&clp->cl_refcount, &nn->client_lock))
184 if (!is_client_expired(clp))
185 renew_client_locked(clp);
186 spin_unlock(&nn->client_lock);
189 static __be32 nfsd4_get_session_locked(struct nfsd4_session *ses)
193 if (is_session_dead(ses))
194 return nfserr_badsession;
195 status = get_client_locked(ses->se_client);
198 atomic_inc(&ses->se_ref);
202 static void nfsd4_put_session_locked(struct nfsd4_session *ses)
204 struct nfs4_client *clp = ses->se_client;
205 struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
207 lockdep_assert_held(&nn->client_lock);
209 if (atomic_dec_and_test(&ses->se_ref) && is_session_dead(ses))
211 put_client_renew_locked(clp);
214 static void nfsd4_put_session(struct nfsd4_session *ses)
216 struct nfs4_client *clp = ses->se_client;
217 struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
219 spin_lock(&nn->client_lock);
220 nfsd4_put_session_locked(ses);
221 spin_unlock(&nn->client_lock);
224 static struct nfsd4_blocked_lock *
225 find_blocked_lock(struct nfs4_lockowner *lo, struct knfsd_fh *fh,
228 struct nfsd4_blocked_lock *cur, *found = NULL;
230 spin_lock(&nn->blocked_locks_lock);
231 list_for_each_entry(cur, &lo->lo_blocked, nbl_list) {
232 if (fh_match(fh, &cur->nbl_fh)) {
233 list_del_init(&cur->nbl_list);
234 list_del_init(&cur->nbl_lru);
239 spin_unlock(&nn->blocked_locks_lock);
241 posix_unblock_lock(&found->nbl_lock);
245 static struct nfsd4_blocked_lock *
246 find_or_allocate_block(struct nfs4_lockowner *lo, struct knfsd_fh *fh,
249 struct nfsd4_blocked_lock *nbl;
251 nbl = find_blocked_lock(lo, fh, nn);
253 nbl= kmalloc(sizeof(*nbl), GFP_KERNEL);
255 fh_copy_shallow(&nbl->nbl_fh, fh);
256 locks_init_lock(&nbl->nbl_lock);
257 nfsd4_init_cb(&nbl->nbl_cb, lo->lo_owner.so_client,
258 &nfsd4_cb_notify_lock_ops,
259 NFSPROC4_CLNT_CB_NOTIFY_LOCK);
266 free_blocked_lock(struct nfsd4_blocked_lock *nbl)
268 locks_release_private(&nbl->nbl_lock);
273 remove_blocked_locks(struct nfs4_lockowner *lo)
275 struct nfs4_client *clp = lo->lo_owner.so_client;
276 struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
277 struct nfsd4_blocked_lock *nbl;
280 /* Dequeue all blocked locks */
281 spin_lock(&nn->blocked_locks_lock);
282 while (!list_empty(&lo->lo_blocked)) {
283 nbl = list_first_entry(&lo->lo_blocked,
284 struct nfsd4_blocked_lock,
286 list_del_init(&nbl->nbl_list);
287 list_move(&nbl->nbl_lru, &reaplist);
289 spin_unlock(&nn->blocked_locks_lock);
292 while (!list_empty(&reaplist)) {
293 nbl = list_first_entry(&reaplist, struct nfsd4_blocked_lock,
295 list_del_init(&nbl->nbl_lru);
296 posix_unblock_lock(&nbl->nbl_lock);
297 free_blocked_lock(nbl);
302 nfsd4_cb_notify_lock_done(struct nfsd4_callback *cb, struct rpc_task *task)
305 * Since this is just an optimization, we don't try very hard if it
306 * turns out not to succeed. We'll requeue it on NFS4ERR_DELAY, and
307 * just quit trying on anything else.
309 switch (task->tk_status) {
311 rpc_delay(task, 1 * HZ);
319 nfsd4_cb_notify_lock_release(struct nfsd4_callback *cb)
321 struct nfsd4_blocked_lock *nbl = container_of(cb,
322 struct nfsd4_blocked_lock, nbl_cb);
324 free_blocked_lock(nbl);
327 static const struct nfsd4_callback_ops nfsd4_cb_notify_lock_ops = {
328 .done = nfsd4_cb_notify_lock_done,
329 .release = nfsd4_cb_notify_lock_release,
332 static inline struct nfs4_stateowner *
333 nfs4_get_stateowner(struct nfs4_stateowner *sop)
335 atomic_inc(&sop->so_count);
340 same_owner_str(struct nfs4_stateowner *sop, struct xdr_netobj *owner)
342 return (sop->so_owner.len == owner->len) &&
343 0 == memcmp(sop->so_owner.data, owner->data, owner->len);
346 static struct nfs4_openowner *
347 find_openstateowner_str_locked(unsigned int hashval, struct nfsd4_open *open,
348 struct nfs4_client *clp)
350 struct nfs4_stateowner *so;
352 lockdep_assert_held(&clp->cl_lock);
354 list_for_each_entry(so, &clp->cl_ownerstr_hashtbl[hashval],
356 if (!so->so_is_open_owner)
358 if (same_owner_str(so, &open->op_owner))
359 return openowner(nfs4_get_stateowner(so));
364 static struct nfs4_openowner *
365 find_openstateowner_str(unsigned int hashval, struct nfsd4_open *open,
366 struct nfs4_client *clp)
368 struct nfs4_openowner *oo;
370 spin_lock(&clp->cl_lock);
371 oo = find_openstateowner_str_locked(hashval, open, clp);
372 spin_unlock(&clp->cl_lock);
377 opaque_hashval(const void *ptr, int nbytes)
379 unsigned char *cptr = (unsigned char *) ptr;
389 static void nfsd4_free_file_rcu(struct rcu_head *rcu)
391 struct nfs4_file *fp = container_of(rcu, struct nfs4_file, fi_rcu);
393 kmem_cache_free(file_slab, fp);
397 put_nfs4_file(struct nfs4_file *fi)
399 might_lock(&state_lock);
401 if (refcount_dec_and_lock(&fi->fi_ref, &state_lock)) {
402 hlist_del_rcu(&fi->fi_hash);
403 spin_unlock(&state_lock);
404 WARN_ON_ONCE(!list_empty(&fi->fi_clnt_odstate));
405 WARN_ON_ONCE(!list_empty(&fi->fi_delegations));
406 call_rcu(&fi->fi_rcu, nfsd4_free_file_rcu);
411 __nfs4_get_fd(struct nfs4_file *f, int oflag)
413 if (f->fi_fds[oflag])
414 return get_file(f->fi_fds[oflag]);
419 find_writeable_file_locked(struct nfs4_file *f)
423 lockdep_assert_held(&f->fi_lock);
425 ret = __nfs4_get_fd(f, O_WRONLY);
427 ret = __nfs4_get_fd(f, O_RDWR);
432 find_writeable_file(struct nfs4_file *f)
436 spin_lock(&f->fi_lock);
437 ret = find_writeable_file_locked(f);
438 spin_unlock(&f->fi_lock);
443 static struct file *find_readable_file_locked(struct nfs4_file *f)
447 lockdep_assert_held(&f->fi_lock);
449 ret = __nfs4_get_fd(f, O_RDONLY);
451 ret = __nfs4_get_fd(f, O_RDWR);
456 find_readable_file(struct nfs4_file *f)
460 spin_lock(&f->fi_lock);
461 ret = find_readable_file_locked(f);
462 spin_unlock(&f->fi_lock);
468 find_any_file(struct nfs4_file *f)
472 spin_lock(&f->fi_lock);
473 ret = __nfs4_get_fd(f, O_RDWR);
475 ret = __nfs4_get_fd(f, O_WRONLY);
477 ret = __nfs4_get_fd(f, O_RDONLY);
479 spin_unlock(&f->fi_lock);
483 static atomic_long_t num_delegations;
484 unsigned long max_delegations;
487 * Open owner state (share locks)
490 /* hash tables for lock and open owners */
491 #define OWNER_HASH_BITS 8
492 #define OWNER_HASH_SIZE (1 << OWNER_HASH_BITS)
493 #define OWNER_HASH_MASK (OWNER_HASH_SIZE - 1)
495 static unsigned int ownerstr_hashval(struct xdr_netobj *ownername)
499 ret = opaque_hashval(ownername->data, ownername->len);
500 return ret & OWNER_HASH_MASK;
503 /* hash table for nfs4_file */
504 #define FILE_HASH_BITS 8
505 #define FILE_HASH_SIZE (1 << FILE_HASH_BITS)
507 static unsigned int nfsd_fh_hashval(struct knfsd_fh *fh)
509 return jhash2(fh->fh_base.fh_pad, XDR_QUADLEN(fh->fh_size), 0);
512 static unsigned int file_hashval(struct knfsd_fh *fh)
514 return nfsd_fh_hashval(fh) & (FILE_HASH_SIZE - 1);
517 static struct hlist_head file_hashtbl[FILE_HASH_SIZE];
520 __nfs4_file_get_access(struct nfs4_file *fp, u32 access)
522 lockdep_assert_held(&fp->fi_lock);
524 if (access & NFS4_SHARE_ACCESS_WRITE)
525 atomic_inc(&fp->fi_access[O_WRONLY]);
526 if (access & NFS4_SHARE_ACCESS_READ)
527 atomic_inc(&fp->fi_access[O_RDONLY]);
531 nfs4_file_get_access(struct nfs4_file *fp, u32 access)
533 lockdep_assert_held(&fp->fi_lock);
535 /* Does this access mode make sense? */
536 if (access & ~NFS4_SHARE_ACCESS_BOTH)
539 /* Does it conflict with a deny mode already set? */
540 if ((access & fp->fi_share_deny) != 0)
541 return nfserr_share_denied;
543 __nfs4_file_get_access(fp, access);
547 static __be32 nfs4_file_check_deny(struct nfs4_file *fp, u32 deny)
549 /* Common case is that there is no deny mode. */
551 /* Does this deny mode make sense? */
552 if (deny & ~NFS4_SHARE_DENY_BOTH)
555 if ((deny & NFS4_SHARE_DENY_READ) &&
556 atomic_read(&fp->fi_access[O_RDONLY]))
557 return nfserr_share_denied;
559 if ((deny & NFS4_SHARE_DENY_WRITE) &&
560 atomic_read(&fp->fi_access[O_WRONLY]))
561 return nfserr_share_denied;
566 static void __nfs4_file_put_access(struct nfs4_file *fp, int oflag)
568 might_lock(&fp->fi_lock);
570 if (atomic_dec_and_lock(&fp->fi_access[oflag], &fp->fi_lock)) {
571 struct file *f1 = NULL;
572 struct file *f2 = NULL;
574 swap(f1, fp->fi_fds[oflag]);
575 if (atomic_read(&fp->fi_access[1 - oflag]) == 0)
576 swap(f2, fp->fi_fds[O_RDWR]);
577 spin_unlock(&fp->fi_lock);
585 static void nfs4_file_put_access(struct nfs4_file *fp, u32 access)
587 WARN_ON_ONCE(access & ~NFS4_SHARE_ACCESS_BOTH);
589 if (access & NFS4_SHARE_ACCESS_WRITE)
590 __nfs4_file_put_access(fp, O_WRONLY);
591 if (access & NFS4_SHARE_ACCESS_READ)
592 __nfs4_file_put_access(fp, O_RDONLY);
596 * Allocate a new open/delegation state counter. This is needed for
597 * pNFS for proper return on close semantics.
599 * Note that we only allocate it for pNFS-enabled exports, otherwise
600 * all pointers to struct nfs4_clnt_odstate are always NULL.
602 static struct nfs4_clnt_odstate *
603 alloc_clnt_odstate(struct nfs4_client *clp)
605 struct nfs4_clnt_odstate *co;
607 co = kmem_cache_zalloc(odstate_slab, GFP_KERNEL);
610 refcount_set(&co->co_odcount, 1);
616 hash_clnt_odstate_locked(struct nfs4_clnt_odstate *co)
618 struct nfs4_file *fp = co->co_file;
620 lockdep_assert_held(&fp->fi_lock);
621 list_add(&co->co_perfile, &fp->fi_clnt_odstate);
625 get_clnt_odstate(struct nfs4_clnt_odstate *co)
628 refcount_inc(&co->co_odcount);
632 put_clnt_odstate(struct nfs4_clnt_odstate *co)
634 struct nfs4_file *fp;
640 if (refcount_dec_and_lock(&co->co_odcount, &fp->fi_lock)) {
641 list_del(&co->co_perfile);
642 spin_unlock(&fp->fi_lock);
644 nfsd4_return_all_file_layouts(co->co_client, fp);
645 kmem_cache_free(odstate_slab, co);
649 static struct nfs4_clnt_odstate *
650 find_or_hash_clnt_odstate(struct nfs4_file *fp, struct nfs4_clnt_odstate *new)
652 struct nfs4_clnt_odstate *co;
653 struct nfs4_client *cl;
660 spin_lock(&fp->fi_lock);
661 list_for_each_entry(co, &fp->fi_clnt_odstate, co_perfile) {
662 if (co->co_client == cl) {
663 get_clnt_odstate(co);
669 hash_clnt_odstate_locked(new);
671 spin_unlock(&fp->fi_lock);
675 struct nfs4_stid *nfs4_alloc_stid(struct nfs4_client *cl, struct kmem_cache *slab,
676 void (*sc_free)(struct nfs4_stid *))
678 struct nfs4_stid *stid;
681 stid = kmem_cache_zalloc(slab, GFP_KERNEL);
685 idr_preload(GFP_KERNEL);
686 spin_lock(&cl->cl_lock);
687 new_id = idr_alloc_cyclic(&cl->cl_stateids, stid, 0, 0, GFP_NOWAIT);
688 spin_unlock(&cl->cl_lock);
693 stid->sc_free = sc_free;
694 stid->sc_client = cl;
695 stid->sc_stateid.si_opaque.so_id = new_id;
696 stid->sc_stateid.si_opaque.so_clid = cl->cl_clientid;
697 /* Will be incremented before return to client: */
698 refcount_set(&stid->sc_count, 1);
699 spin_lock_init(&stid->sc_lock);
702 * It shouldn't be a problem to reuse an opaque stateid value.
703 * I don't think it is for 4.1. But with 4.0 I worry that, for
704 * example, a stray write retransmission could be accepted by
705 * the server when it should have been rejected. Therefore,
706 * adopt a trick from the sctp code to attempt to maximize the
707 * amount of time until an id is reused, by ensuring they always
708 * "increase" (mod INT_MAX):
712 kmem_cache_free(slab, stid);
716 static struct nfs4_ol_stateid * nfs4_alloc_open_stateid(struct nfs4_client *clp)
718 struct nfs4_stid *stid;
720 stid = nfs4_alloc_stid(clp, stateid_slab, nfs4_free_ol_stateid);
724 return openlockstateid(stid);
727 static void nfs4_free_deleg(struct nfs4_stid *stid)
729 kmem_cache_free(deleg_slab, stid);
730 atomic_long_dec(&num_delegations);
734 * When we recall a delegation, we should be careful not to hand it
735 * out again straight away.
736 * To ensure this we keep a pair of bloom filters ('new' and 'old')
737 * in which the filehandles of recalled delegations are "stored".
738 * If a filehandle appear in either filter, a delegation is blocked.
739 * When a delegation is recalled, the filehandle is stored in the "new"
741 * Every 30 seconds we swap the filters and clear the "new" one,
742 * unless both are empty of course.
744 * Each filter is 256 bits. We hash the filehandle to 32bit and use the
745 * low 3 bytes as hash-table indices.
747 * 'blocked_delegations_lock', which is always taken in block_delegations(),
748 * is used to manage concurrent access. Testing does not need the lock
749 * except when swapping the two filters.
751 static DEFINE_SPINLOCK(blocked_delegations_lock);
752 static struct bloom_pair {
753 int entries, old_entries;
755 int new; /* index into 'set' */
756 DECLARE_BITMAP(set[2], 256);
757 } blocked_delegations;
759 static int delegation_blocked(struct knfsd_fh *fh)
762 struct bloom_pair *bd = &blocked_delegations;
764 if (bd->entries == 0)
766 if (seconds_since_boot() - bd->swap_time > 30) {
767 spin_lock(&blocked_delegations_lock);
768 if (seconds_since_boot() - bd->swap_time > 30) {
769 bd->entries -= bd->old_entries;
770 bd->old_entries = bd->entries;
771 memset(bd->set[bd->new], 0,
774 bd->swap_time = seconds_since_boot();
776 spin_unlock(&blocked_delegations_lock);
778 hash = jhash(&fh->fh_base, fh->fh_size, 0);
779 if (test_bit(hash&255, bd->set[0]) &&
780 test_bit((hash>>8)&255, bd->set[0]) &&
781 test_bit((hash>>16)&255, bd->set[0]))
784 if (test_bit(hash&255, bd->set[1]) &&
785 test_bit((hash>>8)&255, bd->set[1]) &&
786 test_bit((hash>>16)&255, bd->set[1]))
792 static void block_delegations(struct knfsd_fh *fh)
795 struct bloom_pair *bd = &blocked_delegations;
797 hash = jhash(&fh->fh_base, fh->fh_size, 0);
799 spin_lock(&blocked_delegations_lock);
800 __set_bit(hash&255, bd->set[bd->new]);
801 __set_bit((hash>>8)&255, bd->set[bd->new]);
802 __set_bit((hash>>16)&255, bd->set[bd->new]);
803 if (bd->entries == 0)
804 bd->swap_time = seconds_since_boot();
806 spin_unlock(&blocked_delegations_lock);
809 static struct nfs4_delegation *
810 alloc_init_deleg(struct nfs4_client *clp, struct nfs4_file *fp,
811 struct svc_fh *current_fh,
812 struct nfs4_clnt_odstate *odstate)
814 struct nfs4_delegation *dp;
817 dprintk("NFSD alloc_init_deleg\n");
818 n = atomic_long_inc_return(&num_delegations);
819 if (n < 0 || n > max_delegations)
821 if (delegation_blocked(¤t_fh->fh_handle))
823 dp = delegstateid(nfs4_alloc_stid(clp, deleg_slab, nfs4_free_deleg));
828 * delegation seqid's are never incremented. The 4.1 special
829 * meaning of seqid 0 isn't meaningful, really, but let's avoid
830 * 0 anyway just for consistency and use 1:
832 dp->dl_stid.sc_stateid.si_generation = 1;
833 INIT_LIST_HEAD(&dp->dl_perfile);
834 INIT_LIST_HEAD(&dp->dl_perclnt);
835 INIT_LIST_HEAD(&dp->dl_recall_lru);
836 dp->dl_clnt_odstate = odstate;
837 get_clnt_odstate(odstate);
838 dp->dl_type = NFS4_OPEN_DELEGATE_READ;
840 nfsd4_init_cb(&dp->dl_recall, dp->dl_stid.sc_client,
841 &nfsd4_cb_recall_ops, NFSPROC4_CLNT_CB_RECALL);
843 dp->dl_stid.sc_file = fp;
846 atomic_long_dec(&num_delegations);
851 nfs4_put_stid(struct nfs4_stid *s)
853 struct nfs4_file *fp = s->sc_file;
854 struct nfs4_client *clp = s->sc_client;
856 might_lock(&clp->cl_lock);
858 if (!refcount_dec_and_lock(&s->sc_count, &clp->cl_lock)) {
859 wake_up_all(&close_wq);
862 idr_remove(&clp->cl_stateids, s->sc_stateid.si_opaque.so_id);
863 spin_unlock(&clp->cl_lock);
870 nfs4_inc_and_copy_stateid(stateid_t *dst, struct nfs4_stid *stid)
872 stateid_t *src = &stid->sc_stateid;
874 spin_lock(&stid->sc_lock);
875 if (unlikely(++src->si_generation == 0))
876 src->si_generation = 1;
877 memcpy(dst, src, sizeof(*dst));
878 spin_unlock(&stid->sc_lock);
881 static void put_deleg_file(struct nfs4_file *fp)
883 struct file *filp = NULL;
885 spin_lock(&fp->fi_lock);
886 if (--fp->fi_delegees == 0)
887 swap(filp, fp->fi_deleg_file);
888 spin_unlock(&fp->fi_lock);
894 static void nfs4_unlock_deleg_lease(struct nfs4_delegation *dp)
896 struct nfs4_file *fp = dp->dl_stid.sc_file;
897 struct file *filp = fp->fi_deleg_file;
899 WARN_ON_ONCE(!fp->fi_delegees);
901 vfs_setlease(filp, F_UNLCK, NULL, (void **)&dp);
905 static void destroy_unhashed_deleg(struct nfs4_delegation *dp)
907 put_clnt_odstate(dp->dl_clnt_odstate);
908 nfs4_unlock_deleg_lease(dp);
909 nfs4_put_stid(&dp->dl_stid);
912 void nfs4_unhash_stid(struct nfs4_stid *s)
918 * nfs4_delegation_exists - Discover if this delegation already exists
919 * @clp: a pointer to the nfs4_client we're granting a delegation to
920 * @fp: a pointer to the nfs4_file we're granting a delegation on
923 * On success: true iff an existing delegation is found
927 nfs4_delegation_exists(struct nfs4_client *clp, struct nfs4_file *fp)
929 struct nfs4_delegation *searchdp = NULL;
930 struct nfs4_client *searchclp = NULL;
932 lockdep_assert_held(&state_lock);
933 lockdep_assert_held(&fp->fi_lock);
935 list_for_each_entry(searchdp, &fp->fi_delegations, dl_perfile) {
936 searchclp = searchdp->dl_stid.sc_client;
937 if (clp == searchclp) {
945 * hash_delegation_locked - Add a delegation to the appropriate lists
946 * @dp: a pointer to the nfs4_delegation we are adding.
947 * @fp: a pointer to the nfs4_file we're granting a delegation on
950 * On success: NULL if the delegation was successfully hashed.
952 * On error: -EAGAIN if one was previously granted to this
953 * nfs4_client for this nfs4_file. Delegation is not hashed.
958 hash_delegation_locked(struct nfs4_delegation *dp, struct nfs4_file *fp)
960 struct nfs4_client *clp = dp->dl_stid.sc_client;
962 lockdep_assert_held(&state_lock);
963 lockdep_assert_held(&fp->fi_lock);
965 if (nfs4_delegation_exists(clp, fp))
967 refcount_inc(&dp->dl_stid.sc_count);
968 dp->dl_stid.sc_type = NFS4_DELEG_STID;
969 list_add(&dp->dl_perfile, &fp->fi_delegations);
970 list_add(&dp->dl_perclnt, &clp->cl_delegations);
975 unhash_delegation_locked(struct nfs4_delegation *dp)
977 struct nfs4_file *fp = dp->dl_stid.sc_file;
979 lockdep_assert_held(&state_lock);
981 if (list_empty(&dp->dl_perfile))
984 dp->dl_stid.sc_type = NFS4_CLOSED_DELEG_STID;
985 /* Ensure that deleg break won't try to requeue it */
987 spin_lock(&fp->fi_lock);
988 list_del_init(&dp->dl_perclnt);
989 list_del_init(&dp->dl_recall_lru);
990 list_del_init(&dp->dl_perfile);
991 spin_unlock(&fp->fi_lock);
995 static void destroy_delegation(struct nfs4_delegation *dp)
999 spin_lock(&state_lock);
1000 unhashed = unhash_delegation_locked(dp);
1001 spin_unlock(&state_lock);
1003 destroy_unhashed_deleg(dp);
1006 static void revoke_delegation(struct nfs4_delegation *dp)
1008 struct nfs4_client *clp = dp->dl_stid.sc_client;
1010 WARN_ON(!list_empty(&dp->dl_recall_lru));
1012 if (clp->cl_minorversion) {
1013 dp->dl_stid.sc_type = NFS4_REVOKED_DELEG_STID;
1014 refcount_inc(&dp->dl_stid.sc_count);
1015 spin_lock(&clp->cl_lock);
1016 list_add(&dp->dl_recall_lru, &clp->cl_revoked);
1017 spin_unlock(&clp->cl_lock);
1019 destroy_unhashed_deleg(dp);
1026 static unsigned int clientid_hashval(u32 id)
1028 return id & CLIENT_HASH_MASK;
1031 static unsigned int clientstr_hashval(const char *name)
1033 return opaque_hashval(name, 8) & CLIENT_HASH_MASK;
1037 * We store the NONE, READ, WRITE, and BOTH bits separately in the
1038 * st_{access,deny}_bmap field of the stateid, in order to track not
1039 * only what share bits are currently in force, but also what
1040 * combinations of share bits previous opens have used. This allows us
1041 * to enforce the recommendation of rfc 3530 14.2.19 that the server
1042 * return an error if the client attempt to downgrade to a combination
1043 * of share bits not explicable by closing some of its previous opens.
1045 * XXX: This enforcement is actually incomplete, since we don't keep
1046 * track of access/deny bit combinations; so, e.g., we allow:
1048 * OPEN allow read, deny write
1049 * OPEN allow both, deny none
1050 * DOWNGRADE allow read, deny none
1052 * which we should reject.
1055 bmap_to_share_mode(unsigned long bmap) {
1057 unsigned int access = 0;
1059 for (i = 1; i < 4; i++) {
1060 if (test_bit(i, &bmap))
1066 /* set share access for a given stateid */
1068 set_access(u32 access, struct nfs4_ol_stateid *stp)
1070 unsigned char mask = 1 << access;
1072 WARN_ON_ONCE(access > NFS4_SHARE_ACCESS_BOTH);
1073 stp->st_access_bmap |= mask;
1076 /* clear share access for a given stateid */
1078 clear_access(u32 access, struct nfs4_ol_stateid *stp)
1080 unsigned char mask = 1 << access;
1082 WARN_ON_ONCE(access > NFS4_SHARE_ACCESS_BOTH);
1083 stp->st_access_bmap &= ~mask;
1086 /* test whether a given stateid has access */
1088 test_access(u32 access, struct nfs4_ol_stateid *stp)
1090 unsigned char mask = 1 << access;
1092 return (bool)(stp->st_access_bmap & mask);
1095 /* set share deny for a given stateid */
1097 set_deny(u32 deny, struct nfs4_ol_stateid *stp)
1099 unsigned char mask = 1 << deny;
1101 WARN_ON_ONCE(deny > NFS4_SHARE_DENY_BOTH);
1102 stp->st_deny_bmap |= mask;
1105 /* clear share deny for a given stateid */
1107 clear_deny(u32 deny, struct nfs4_ol_stateid *stp)
1109 unsigned char mask = 1 << deny;
1111 WARN_ON_ONCE(deny > NFS4_SHARE_DENY_BOTH);
1112 stp->st_deny_bmap &= ~mask;
1115 /* test whether a given stateid is denying specific access */
1117 test_deny(u32 deny, struct nfs4_ol_stateid *stp)
1119 unsigned char mask = 1 << deny;
1121 return (bool)(stp->st_deny_bmap & mask);
1124 static int nfs4_access_to_omode(u32 access)
1126 switch (access & NFS4_SHARE_ACCESS_BOTH) {
1127 case NFS4_SHARE_ACCESS_READ:
1129 case NFS4_SHARE_ACCESS_WRITE:
1131 case NFS4_SHARE_ACCESS_BOTH:
1139 * A stateid that had a deny mode associated with it is being released
1140 * or downgraded. Recalculate the deny mode on the file.
1143 recalculate_deny_mode(struct nfs4_file *fp)
1145 struct nfs4_ol_stateid *stp;
1147 spin_lock(&fp->fi_lock);
1148 fp->fi_share_deny = 0;
1149 list_for_each_entry(stp, &fp->fi_stateids, st_perfile)
1150 fp->fi_share_deny |= bmap_to_share_mode(stp->st_deny_bmap);
1151 spin_unlock(&fp->fi_lock);
1155 reset_union_bmap_deny(u32 deny, struct nfs4_ol_stateid *stp)
1158 bool change = false;
1160 for (i = 1; i < 4; i++) {
1161 if ((i & deny) != i) {
1167 /* Recalculate per-file deny mode if there was a change */
1169 recalculate_deny_mode(stp->st_stid.sc_file);
1172 /* release all access and file references for a given stateid */
1174 release_all_access(struct nfs4_ol_stateid *stp)
1177 struct nfs4_file *fp = stp->st_stid.sc_file;
1179 if (fp && stp->st_deny_bmap != 0)
1180 recalculate_deny_mode(fp);
1182 for (i = 1; i < 4; i++) {
1183 if (test_access(i, stp))
1184 nfs4_file_put_access(stp->st_stid.sc_file, i);
1185 clear_access(i, stp);
1189 static inline void nfs4_free_stateowner(struct nfs4_stateowner *sop)
1191 kfree(sop->so_owner.data);
1192 sop->so_ops->so_free(sop);
1195 static void nfs4_put_stateowner(struct nfs4_stateowner *sop)
1197 struct nfs4_client *clp = sop->so_client;
1199 might_lock(&clp->cl_lock);
1201 if (!atomic_dec_and_lock(&sop->so_count, &clp->cl_lock))
1203 sop->so_ops->so_unhash(sop);
1204 spin_unlock(&clp->cl_lock);
1205 nfs4_free_stateowner(sop);
1208 static bool unhash_ol_stateid(struct nfs4_ol_stateid *stp)
1210 struct nfs4_file *fp = stp->st_stid.sc_file;
1212 lockdep_assert_held(&stp->st_stateowner->so_client->cl_lock);
1214 if (list_empty(&stp->st_perfile))
1217 spin_lock(&fp->fi_lock);
1218 list_del_init(&stp->st_perfile);
1219 spin_unlock(&fp->fi_lock);
1220 list_del(&stp->st_perstateowner);
1224 static void nfs4_free_ol_stateid(struct nfs4_stid *stid)
1226 struct nfs4_ol_stateid *stp = openlockstateid(stid);
1228 put_clnt_odstate(stp->st_clnt_odstate);
1229 release_all_access(stp);
1230 if (stp->st_stateowner)
1231 nfs4_put_stateowner(stp->st_stateowner);
1232 kmem_cache_free(stateid_slab, stid);
1235 static void nfs4_free_lock_stateid(struct nfs4_stid *stid)
1237 struct nfs4_ol_stateid *stp = openlockstateid(stid);
1238 struct nfs4_lockowner *lo = lockowner(stp->st_stateowner);
1241 file = find_any_file(stp->st_stid.sc_file);
1243 filp_close(file, (fl_owner_t)lo);
1244 nfs4_free_ol_stateid(stid);
1248 * Put the persistent reference to an already unhashed generic stateid, while
1249 * holding the cl_lock. If it's the last reference, then put it onto the
1250 * reaplist for later destruction.
1252 static void put_ol_stateid_locked(struct nfs4_ol_stateid *stp,
1253 struct list_head *reaplist)
1255 struct nfs4_stid *s = &stp->st_stid;
1256 struct nfs4_client *clp = s->sc_client;
1258 lockdep_assert_held(&clp->cl_lock);
1260 WARN_ON_ONCE(!list_empty(&stp->st_locks));
1262 if (!refcount_dec_and_test(&s->sc_count)) {
1263 wake_up_all(&close_wq);
1267 idr_remove(&clp->cl_stateids, s->sc_stateid.si_opaque.so_id);
1268 list_add(&stp->st_locks, reaplist);
1271 static bool unhash_lock_stateid(struct nfs4_ol_stateid *stp)
1273 lockdep_assert_held(&stp->st_stid.sc_client->cl_lock);
1275 list_del_init(&stp->st_locks);
1276 nfs4_unhash_stid(&stp->st_stid);
1277 return unhash_ol_stateid(stp);
1280 static void release_lock_stateid(struct nfs4_ol_stateid *stp)
1282 struct nfs4_client *clp = stp->st_stid.sc_client;
1285 spin_lock(&clp->cl_lock);
1286 unhashed = unhash_lock_stateid(stp);
1287 spin_unlock(&clp->cl_lock);
1289 nfs4_put_stid(&stp->st_stid);
1292 static void unhash_lockowner_locked(struct nfs4_lockowner *lo)
1294 struct nfs4_client *clp = lo->lo_owner.so_client;
1296 lockdep_assert_held(&clp->cl_lock);
1298 list_del_init(&lo->lo_owner.so_strhash);
1302 * Free a list of generic stateids that were collected earlier after being
1306 free_ol_stateid_reaplist(struct list_head *reaplist)
1308 struct nfs4_ol_stateid *stp;
1309 struct nfs4_file *fp;
1313 while (!list_empty(reaplist)) {
1314 stp = list_first_entry(reaplist, struct nfs4_ol_stateid,
1316 list_del(&stp->st_locks);
1317 fp = stp->st_stid.sc_file;
1318 stp->st_stid.sc_free(&stp->st_stid);
1324 static void release_open_stateid_locks(struct nfs4_ol_stateid *open_stp,
1325 struct list_head *reaplist)
1327 struct nfs4_ol_stateid *stp;
1329 lockdep_assert_held(&open_stp->st_stid.sc_client->cl_lock);
1331 while (!list_empty(&open_stp->st_locks)) {
1332 stp = list_entry(open_stp->st_locks.next,
1333 struct nfs4_ol_stateid, st_locks);
1334 WARN_ON(!unhash_lock_stateid(stp));
1335 put_ol_stateid_locked(stp, reaplist);
1339 static bool unhash_open_stateid(struct nfs4_ol_stateid *stp,
1340 struct list_head *reaplist)
1344 lockdep_assert_held(&stp->st_stid.sc_client->cl_lock);
1346 unhashed = unhash_ol_stateid(stp);
1347 release_open_stateid_locks(stp, reaplist);
1351 static void release_open_stateid(struct nfs4_ol_stateid *stp)
1353 LIST_HEAD(reaplist);
1355 spin_lock(&stp->st_stid.sc_client->cl_lock);
1356 if (unhash_open_stateid(stp, &reaplist))
1357 put_ol_stateid_locked(stp, &reaplist);
1358 spin_unlock(&stp->st_stid.sc_client->cl_lock);
1359 free_ol_stateid_reaplist(&reaplist);
1362 static void unhash_openowner_locked(struct nfs4_openowner *oo)
1364 struct nfs4_client *clp = oo->oo_owner.so_client;
1366 lockdep_assert_held(&clp->cl_lock);
1368 list_del_init(&oo->oo_owner.so_strhash);
1369 list_del_init(&oo->oo_perclient);
1372 static void release_last_closed_stateid(struct nfs4_openowner *oo)
1374 struct nfsd_net *nn = net_generic(oo->oo_owner.so_client->net,
1376 struct nfs4_ol_stateid *s;
1378 spin_lock(&nn->client_lock);
1379 s = oo->oo_last_closed_stid;
1381 list_del_init(&oo->oo_close_lru);
1382 oo->oo_last_closed_stid = NULL;
1384 spin_unlock(&nn->client_lock);
1386 nfs4_put_stid(&s->st_stid);
1389 static void release_openowner(struct nfs4_openowner *oo)
1391 struct nfs4_ol_stateid *stp;
1392 struct nfs4_client *clp = oo->oo_owner.so_client;
1393 struct list_head reaplist;
1395 INIT_LIST_HEAD(&reaplist);
1397 spin_lock(&clp->cl_lock);
1398 unhash_openowner_locked(oo);
1399 while (!list_empty(&oo->oo_owner.so_stateids)) {
1400 stp = list_first_entry(&oo->oo_owner.so_stateids,
1401 struct nfs4_ol_stateid, st_perstateowner);
1402 if (unhash_open_stateid(stp, &reaplist))
1403 put_ol_stateid_locked(stp, &reaplist);
1405 spin_unlock(&clp->cl_lock);
1406 free_ol_stateid_reaplist(&reaplist);
1407 release_last_closed_stateid(oo);
1408 nfs4_put_stateowner(&oo->oo_owner);
1412 hash_sessionid(struct nfs4_sessionid *sessionid)
1414 struct nfsd4_sessionid *sid = (struct nfsd4_sessionid *)sessionid;
1416 return sid->sequence % SESSION_HASH_SIZE;
1419 #ifdef CONFIG_SUNRPC_DEBUG
1421 dump_sessionid(const char *fn, struct nfs4_sessionid *sessionid)
1423 u32 *ptr = (u32 *)(&sessionid->data[0]);
1424 dprintk("%s: %u:%u:%u:%u\n", fn, ptr[0], ptr[1], ptr[2], ptr[3]);
1428 dump_sessionid(const char *fn, struct nfs4_sessionid *sessionid)
1434 * Bump the seqid on cstate->replay_owner, and clear replay_owner if it
1435 * won't be used for replay.
1437 void nfsd4_bump_seqid(struct nfsd4_compound_state *cstate, __be32 nfserr)
1439 struct nfs4_stateowner *so = cstate->replay_owner;
1441 if (nfserr == nfserr_replay_me)
1444 if (!seqid_mutating_err(ntohl(nfserr))) {
1445 nfsd4_cstate_clear_replay(cstate);
1450 if (so->so_is_open_owner)
1451 release_last_closed_stateid(openowner(so));
1457 gen_sessionid(struct nfsd4_session *ses)
1459 struct nfs4_client *clp = ses->se_client;
1460 struct nfsd4_sessionid *sid;
1462 sid = (struct nfsd4_sessionid *)ses->se_sessionid.data;
1463 sid->clientid = clp->cl_clientid;
1464 sid->sequence = current_sessionid++;
1469 * The protocol defines ca_maxresponssize_cached to include the size of
1470 * the rpc header, but all we need to cache is the data starting after
1471 * the end of the initial SEQUENCE operation--the rest we regenerate
1472 * each time. Therefore we can advertise a ca_maxresponssize_cached
1473 * value that is the number of bytes in our cache plus a few additional
1474 * bytes. In order to stay on the safe side, and not promise more than
1475 * we can cache, those additional bytes must be the minimum possible: 24
1476 * bytes of rpc header (xid through accept state, with AUTH_NULL
1477 * verifier), 12 for the compound header (with zero-length tag), and 44
1478 * for the SEQUENCE op response:
1480 #define NFSD_MIN_HDR_SEQ_SZ (24 + 12 + 44)
1483 free_session_slots(struct nfsd4_session *ses)
1487 for (i = 0; i < ses->se_fchannel.maxreqs; i++) {
1488 free_svc_cred(&ses->se_slots[i]->sl_cred);
1489 kfree(ses->se_slots[i]);
1494 * We don't actually need to cache the rpc and session headers, so we
1495 * can allocate a little less for each slot:
1497 static inline u32 slot_bytes(struct nfsd4_channel_attrs *ca)
1501 if (ca->maxresp_cached < NFSD_MIN_HDR_SEQ_SZ)
1504 size = ca->maxresp_cached - NFSD_MIN_HDR_SEQ_SZ;
1505 return size + sizeof(struct nfsd4_slot);
1509 * XXX: If we run out of reserved DRC memory we could (up to a point)
1510 * re-negotiate active sessions and reduce their slot usage to make
1511 * room for new connections. For now we just fail the create session.
1513 static u32 nfsd4_get_drc_mem(struct nfsd4_channel_attrs *ca)
1515 u32 slotsize = slot_bytes(ca);
1516 u32 num = ca->maxreqs;
1519 spin_lock(&nfsd_drc_lock);
1520 avail = min((unsigned long)NFSD_MAX_MEM_PER_SESSION,
1521 nfsd_drc_max_mem - nfsd_drc_mem_used);
1523 * Never use more than a third of the remaining memory,
1524 * unless it's the only way to give this client a slot:
1526 avail = clamp_t(int, avail, slotsize, avail/3);
1527 num = min_t(int, num, avail / slotsize);
1528 nfsd_drc_mem_used += num * slotsize;
1529 spin_unlock(&nfsd_drc_lock);
1534 static void nfsd4_put_drc_mem(struct nfsd4_channel_attrs *ca)
1536 int slotsize = slot_bytes(ca);
1538 spin_lock(&nfsd_drc_lock);
1539 nfsd_drc_mem_used -= slotsize * ca->maxreqs;
1540 spin_unlock(&nfsd_drc_lock);
1543 static struct nfsd4_session *alloc_session(struct nfsd4_channel_attrs *fattrs,
1544 struct nfsd4_channel_attrs *battrs)
1546 int numslots = fattrs->maxreqs;
1547 int slotsize = slot_bytes(fattrs);
1548 struct nfsd4_session *new;
1551 BUILD_BUG_ON(NFSD_MAX_SLOTS_PER_SESSION * sizeof(struct nfsd4_slot *)
1552 + sizeof(struct nfsd4_session) > PAGE_SIZE);
1553 mem = numslots * sizeof(struct nfsd4_slot *);
1555 new = kzalloc(sizeof(*new) + mem, GFP_KERNEL);
1558 /* allocate each struct nfsd4_slot and data cache in one piece */
1559 for (i = 0; i < numslots; i++) {
1560 new->se_slots[i] = kzalloc(slotsize, GFP_KERNEL);
1561 if (!new->se_slots[i])
1565 memcpy(&new->se_fchannel, fattrs, sizeof(struct nfsd4_channel_attrs));
1566 memcpy(&new->se_bchannel, battrs, sizeof(struct nfsd4_channel_attrs));
1571 kfree(new->se_slots[i]);
1576 static void free_conn(struct nfsd4_conn *c)
1578 svc_xprt_put(c->cn_xprt);
1582 static void nfsd4_conn_lost(struct svc_xpt_user *u)
1584 struct nfsd4_conn *c = container_of(u, struct nfsd4_conn, cn_xpt_user);
1585 struct nfs4_client *clp = c->cn_session->se_client;
1587 spin_lock(&clp->cl_lock);
1588 if (!list_empty(&c->cn_persession)) {
1589 list_del(&c->cn_persession);
1592 nfsd4_probe_callback(clp);
1593 spin_unlock(&clp->cl_lock);
1596 static struct nfsd4_conn *alloc_conn(struct svc_rqst *rqstp, u32 flags)
1598 struct nfsd4_conn *conn;
1600 conn = kmalloc(sizeof(struct nfsd4_conn), GFP_KERNEL);
1603 svc_xprt_get(rqstp->rq_xprt);
1604 conn->cn_xprt = rqstp->rq_xprt;
1605 conn->cn_flags = flags;
1606 INIT_LIST_HEAD(&conn->cn_xpt_user.list);
1610 static void __nfsd4_hash_conn(struct nfsd4_conn *conn, struct nfsd4_session *ses)
1612 conn->cn_session = ses;
1613 list_add(&conn->cn_persession, &ses->se_conns);
1616 static void nfsd4_hash_conn(struct nfsd4_conn *conn, struct nfsd4_session *ses)
1618 struct nfs4_client *clp = ses->se_client;
1620 spin_lock(&clp->cl_lock);
1621 __nfsd4_hash_conn(conn, ses);
1622 spin_unlock(&clp->cl_lock);
1625 static int nfsd4_register_conn(struct nfsd4_conn *conn)
1627 conn->cn_xpt_user.callback = nfsd4_conn_lost;
1628 return register_xpt_user(conn->cn_xprt, &conn->cn_xpt_user);
1631 static void nfsd4_init_conn(struct svc_rqst *rqstp, struct nfsd4_conn *conn, struct nfsd4_session *ses)
1635 nfsd4_hash_conn(conn, ses);
1636 ret = nfsd4_register_conn(conn);
1638 /* oops; xprt is already down: */
1639 nfsd4_conn_lost(&conn->cn_xpt_user);
1640 /* We may have gained or lost a callback channel: */
1641 nfsd4_probe_callback_sync(ses->se_client);
1644 static struct nfsd4_conn *alloc_conn_from_crses(struct svc_rqst *rqstp, struct nfsd4_create_session *cses)
1646 u32 dir = NFS4_CDFC4_FORE;
1648 if (cses->flags & SESSION4_BACK_CHAN)
1649 dir |= NFS4_CDFC4_BACK;
1650 return alloc_conn(rqstp, dir);
1653 /* must be called under client_lock */
1654 static void nfsd4_del_conns(struct nfsd4_session *s)
1656 struct nfs4_client *clp = s->se_client;
1657 struct nfsd4_conn *c;
1659 spin_lock(&clp->cl_lock);
1660 while (!list_empty(&s->se_conns)) {
1661 c = list_first_entry(&s->se_conns, struct nfsd4_conn, cn_persession);
1662 list_del_init(&c->cn_persession);
1663 spin_unlock(&clp->cl_lock);
1665 unregister_xpt_user(c->cn_xprt, &c->cn_xpt_user);
1668 spin_lock(&clp->cl_lock);
1670 spin_unlock(&clp->cl_lock);
1673 static void __free_session(struct nfsd4_session *ses)
1675 free_session_slots(ses);
1679 static void free_session(struct nfsd4_session *ses)
1681 nfsd4_del_conns(ses);
1682 nfsd4_put_drc_mem(&ses->se_fchannel);
1683 __free_session(ses);
1686 static void init_session(struct svc_rqst *rqstp, struct nfsd4_session *new, struct nfs4_client *clp, struct nfsd4_create_session *cses)
1689 struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
1691 new->se_client = clp;
1694 INIT_LIST_HEAD(&new->se_conns);
1696 new->se_cb_seq_nr = 1;
1697 new->se_flags = cses->flags;
1698 new->se_cb_prog = cses->callback_prog;
1699 new->se_cb_sec = cses->cb_sec;
1700 atomic_set(&new->se_ref, 0);
1701 idx = hash_sessionid(&new->se_sessionid);
1702 list_add(&new->se_hash, &nn->sessionid_hashtbl[idx]);
1703 spin_lock(&clp->cl_lock);
1704 list_add(&new->se_perclnt, &clp->cl_sessions);
1705 spin_unlock(&clp->cl_lock);
1708 struct sockaddr *sa = svc_addr(rqstp);
1710 * This is a little silly; with sessions there's no real
1711 * use for the callback address. Use the peer address
1712 * as a reasonable default for now, but consider fixing
1713 * the rpc client not to require an address in the
1716 rpc_copy_addr((struct sockaddr *)&clp->cl_cb_conn.cb_addr, sa);
1717 clp->cl_cb_conn.cb_addrlen = svc_addr_len(sa);
1721 /* caller must hold client_lock */
1722 static struct nfsd4_session *
1723 __find_in_sessionid_hashtbl(struct nfs4_sessionid *sessionid, struct net *net)
1725 struct nfsd4_session *elem;
1727 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
1729 lockdep_assert_held(&nn->client_lock);
1731 dump_sessionid(__func__, sessionid);
1732 idx = hash_sessionid(sessionid);
1733 /* Search in the appropriate list */
1734 list_for_each_entry(elem, &nn->sessionid_hashtbl[idx], se_hash) {
1735 if (!memcmp(elem->se_sessionid.data, sessionid->data,
1736 NFS4_MAX_SESSIONID_LEN)) {
1741 dprintk("%s: session not found\n", __func__);
1745 static struct nfsd4_session *
1746 find_in_sessionid_hashtbl(struct nfs4_sessionid *sessionid, struct net *net,
1749 struct nfsd4_session *session;
1750 __be32 status = nfserr_badsession;
1752 session = __find_in_sessionid_hashtbl(sessionid, net);
1755 status = nfsd4_get_session_locked(session);
1763 /* caller must hold client_lock */
1765 unhash_session(struct nfsd4_session *ses)
1767 struct nfs4_client *clp = ses->se_client;
1768 struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
1770 lockdep_assert_held(&nn->client_lock);
1772 list_del(&ses->se_hash);
1773 spin_lock(&ses->se_client->cl_lock);
1774 list_del(&ses->se_perclnt);
1775 spin_unlock(&ses->se_client->cl_lock);
1778 /* SETCLIENTID and SETCLIENTID_CONFIRM Helper functions */
1780 STALE_CLIENTID(clientid_t *clid, struct nfsd_net *nn)
1783 * We're assuming the clid was not given out from a boot
1784 * precisely 2^32 (about 136 years) before this one. That seems
1785 * a safe assumption:
1787 if (clid->cl_boot == (u32)nn->boot_time)
1789 dprintk("NFSD stale clientid (%08x/%08x) boot_time %08lx\n",
1790 clid->cl_boot, clid->cl_id, nn->boot_time);
1795 * XXX Should we use a slab cache ?
1796 * This type of memory management is somewhat inefficient, but we use it
1797 * anyway since SETCLIENTID is not a common operation.
1799 static struct nfs4_client *alloc_client(struct xdr_netobj name)
1801 struct nfs4_client *clp;
1804 clp = kmem_cache_zalloc(client_slab, GFP_KERNEL);
1807 clp->cl_name.data = kmemdup(name.data, name.len, GFP_KERNEL);
1808 if (clp->cl_name.data == NULL)
1810 clp->cl_ownerstr_hashtbl = kmalloc_array(OWNER_HASH_SIZE,
1811 sizeof(struct list_head),
1813 if (!clp->cl_ownerstr_hashtbl)
1814 goto err_no_hashtbl;
1815 for (i = 0; i < OWNER_HASH_SIZE; i++)
1816 INIT_LIST_HEAD(&clp->cl_ownerstr_hashtbl[i]);
1817 clp->cl_name.len = name.len;
1818 INIT_LIST_HEAD(&clp->cl_sessions);
1819 idr_init(&clp->cl_stateids);
1820 atomic_set(&clp->cl_refcount, 0);
1821 clp->cl_cb_state = NFSD4_CB_UNKNOWN;
1822 INIT_LIST_HEAD(&clp->cl_idhash);
1823 INIT_LIST_HEAD(&clp->cl_openowners);
1824 INIT_LIST_HEAD(&clp->cl_delegations);
1825 INIT_LIST_HEAD(&clp->cl_lru);
1826 INIT_LIST_HEAD(&clp->cl_revoked);
1827 #ifdef CONFIG_NFSD_PNFS
1828 INIT_LIST_HEAD(&clp->cl_lo_states);
1830 spin_lock_init(&clp->cl_lock);
1831 rpc_init_wait_queue(&clp->cl_cb_waitq, "Backchannel slot table");
1834 kfree(clp->cl_name.data);
1836 kmem_cache_free(client_slab, clp);
1841 free_client(struct nfs4_client *clp)
1843 while (!list_empty(&clp->cl_sessions)) {
1844 struct nfsd4_session *ses;
1845 ses = list_entry(clp->cl_sessions.next, struct nfsd4_session,
1847 list_del(&ses->se_perclnt);
1848 WARN_ON_ONCE(atomic_read(&ses->se_ref));
1851 rpc_destroy_wait_queue(&clp->cl_cb_waitq);
1852 free_svc_cred(&clp->cl_cred);
1853 kfree(clp->cl_ownerstr_hashtbl);
1854 kfree(clp->cl_name.data);
1855 idr_destroy(&clp->cl_stateids);
1856 kmem_cache_free(client_slab, clp);
1859 /* must be called under the client_lock */
1861 unhash_client_locked(struct nfs4_client *clp)
1863 struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
1864 struct nfsd4_session *ses;
1866 lockdep_assert_held(&nn->client_lock);
1868 /* Mark the client as expired! */
1870 /* Make it invisible */
1871 if (!list_empty(&clp->cl_idhash)) {
1872 list_del_init(&clp->cl_idhash);
1873 if (test_bit(NFSD4_CLIENT_CONFIRMED, &clp->cl_flags))
1874 rb_erase(&clp->cl_namenode, &nn->conf_name_tree);
1876 rb_erase(&clp->cl_namenode, &nn->unconf_name_tree);
1878 list_del_init(&clp->cl_lru);
1879 spin_lock(&clp->cl_lock);
1880 list_for_each_entry(ses, &clp->cl_sessions, se_perclnt)
1881 list_del_init(&ses->se_hash);
1882 spin_unlock(&clp->cl_lock);
1886 unhash_client(struct nfs4_client *clp)
1888 struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
1890 spin_lock(&nn->client_lock);
1891 unhash_client_locked(clp);
1892 spin_unlock(&nn->client_lock);
1895 static __be32 mark_client_expired_locked(struct nfs4_client *clp)
1897 if (atomic_read(&clp->cl_refcount))
1898 return nfserr_jukebox;
1899 unhash_client_locked(clp);
1904 __destroy_client(struct nfs4_client *clp)
1907 struct nfs4_openowner *oo;
1908 struct nfs4_delegation *dp;
1909 struct list_head reaplist;
1911 INIT_LIST_HEAD(&reaplist);
1912 spin_lock(&state_lock);
1913 while (!list_empty(&clp->cl_delegations)) {
1914 dp = list_entry(clp->cl_delegations.next, struct nfs4_delegation, dl_perclnt);
1915 WARN_ON(!unhash_delegation_locked(dp));
1916 list_add(&dp->dl_recall_lru, &reaplist);
1918 spin_unlock(&state_lock);
1919 while (!list_empty(&reaplist)) {
1920 dp = list_entry(reaplist.next, struct nfs4_delegation, dl_recall_lru);
1921 list_del_init(&dp->dl_recall_lru);
1922 destroy_unhashed_deleg(dp);
1924 while (!list_empty(&clp->cl_revoked)) {
1925 dp = list_entry(clp->cl_revoked.next, struct nfs4_delegation, dl_recall_lru);
1926 list_del_init(&dp->dl_recall_lru);
1927 nfs4_put_stid(&dp->dl_stid);
1929 while (!list_empty(&clp->cl_openowners)) {
1930 oo = list_entry(clp->cl_openowners.next, struct nfs4_openowner, oo_perclient);
1931 nfs4_get_stateowner(&oo->oo_owner);
1932 release_openowner(oo);
1934 for (i = 0; i < OWNER_HASH_SIZE; i++) {
1935 struct nfs4_stateowner *so, *tmp;
1937 list_for_each_entry_safe(so, tmp, &clp->cl_ownerstr_hashtbl[i],
1939 /* Should be no openowners at this point */
1940 WARN_ON_ONCE(so->so_is_open_owner);
1941 remove_blocked_locks(lockowner(so));
1944 nfsd4_return_all_client_layouts(clp);
1945 nfsd4_shutdown_callback(clp);
1946 if (clp->cl_cb_conn.cb_xprt)
1947 svc_xprt_put(clp->cl_cb_conn.cb_xprt);
1952 destroy_client(struct nfs4_client *clp)
1955 __destroy_client(clp);
1958 static void expire_client(struct nfs4_client *clp)
1961 nfsd4_client_record_remove(clp);
1962 __destroy_client(clp);
1965 static void copy_verf(struct nfs4_client *target, nfs4_verifier *source)
1967 memcpy(target->cl_verifier.data, source->data,
1968 sizeof(target->cl_verifier.data));
1971 static void copy_clid(struct nfs4_client *target, struct nfs4_client *source)
1973 target->cl_clientid.cl_boot = source->cl_clientid.cl_boot;
1974 target->cl_clientid.cl_id = source->cl_clientid.cl_id;
1977 static int copy_cred(struct svc_cred *target, struct svc_cred *source)
1979 target->cr_principal = kstrdup(source->cr_principal, GFP_KERNEL);
1980 target->cr_raw_principal = kstrdup(source->cr_raw_principal,
1982 if ((source->cr_principal && ! target->cr_principal) ||
1983 (source->cr_raw_principal && ! target->cr_raw_principal))
1986 target->cr_flavor = source->cr_flavor;
1987 target->cr_uid = source->cr_uid;
1988 target->cr_gid = source->cr_gid;
1989 target->cr_group_info = source->cr_group_info;
1990 get_group_info(target->cr_group_info);
1991 target->cr_gss_mech = source->cr_gss_mech;
1992 if (source->cr_gss_mech)
1993 gss_mech_get(source->cr_gss_mech);
1998 compare_blob(const struct xdr_netobj *o1, const struct xdr_netobj *o2)
2000 if (o1->len < o2->len)
2002 if (o1->len > o2->len)
2004 return memcmp(o1->data, o2->data, o1->len);
2007 static int same_name(const char *n1, const char *n2)
2009 return 0 == memcmp(n1, n2, HEXDIR_LEN);
2013 same_verf(nfs4_verifier *v1, nfs4_verifier *v2)
2015 return 0 == memcmp(v1->data, v2->data, sizeof(v1->data));
2019 same_clid(clientid_t *cl1, clientid_t *cl2)
2021 return (cl1->cl_boot == cl2->cl_boot) && (cl1->cl_id == cl2->cl_id);
2024 static bool groups_equal(struct group_info *g1, struct group_info *g2)
2028 if (g1->ngroups != g2->ngroups)
2030 for (i=0; i<g1->ngroups; i++)
2031 if (!gid_eq(g1->gid[i], g2->gid[i]))
2037 * RFC 3530 language requires clid_inuse be returned when the
2038 * "principal" associated with a requests differs from that previously
2039 * used. We use uid, gid's, and gss principal string as our best
2040 * approximation. We also don't want to allow non-gss use of a client
2041 * established using gss: in theory cr_principal should catch that
2042 * change, but in practice cr_principal can be null even in the gss case
2043 * since gssd doesn't always pass down a principal string.
2045 static bool is_gss_cred(struct svc_cred *cr)
2047 /* Is cr_flavor one of the gss "pseudoflavors"?: */
2048 return (cr->cr_flavor > RPC_AUTH_MAXFLAVOR);
2053 same_creds(struct svc_cred *cr1, struct svc_cred *cr2)
2055 if ((is_gss_cred(cr1) != is_gss_cred(cr2))
2056 || (!uid_eq(cr1->cr_uid, cr2->cr_uid))
2057 || (!gid_eq(cr1->cr_gid, cr2->cr_gid))
2058 || !groups_equal(cr1->cr_group_info, cr2->cr_group_info))
2060 if (cr1->cr_principal == cr2->cr_principal)
2062 if (!cr1->cr_principal || !cr2->cr_principal)
2064 return 0 == strcmp(cr1->cr_principal, cr2->cr_principal);
2067 static bool svc_rqst_integrity_protected(struct svc_rqst *rqstp)
2069 struct svc_cred *cr = &rqstp->rq_cred;
2072 if (!cr->cr_gss_mech)
2074 service = gss_pseudoflavor_to_service(cr->cr_gss_mech, cr->cr_flavor);
2075 return service == RPC_GSS_SVC_INTEGRITY ||
2076 service == RPC_GSS_SVC_PRIVACY;
2079 bool nfsd4_mach_creds_match(struct nfs4_client *cl, struct svc_rqst *rqstp)
2081 struct svc_cred *cr = &rqstp->rq_cred;
2083 if (!cl->cl_mach_cred)
2085 if (cl->cl_cred.cr_gss_mech != cr->cr_gss_mech)
2087 if (!svc_rqst_integrity_protected(rqstp))
2089 if (cl->cl_cred.cr_raw_principal)
2090 return 0 == strcmp(cl->cl_cred.cr_raw_principal,
2091 cr->cr_raw_principal);
2092 if (!cr->cr_principal)
2094 return 0 == strcmp(cl->cl_cred.cr_principal, cr->cr_principal);
2097 static void gen_confirm(struct nfs4_client *clp, struct nfsd_net *nn)
2102 * This is opaque to client, so no need to byte-swap. Use
2103 * __force to keep sparse happy
2105 verf[0] = (__force __be32)get_seconds();
2106 verf[1] = (__force __be32)nn->clverifier_counter++;
2107 memcpy(clp->cl_confirm.data, verf, sizeof(clp->cl_confirm.data));
2110 static void gen_clid(struct nfs4_client *clp, struct nfsd_net *nn)
2112 clp->cl_clientid.cl_boot = nn->boot_time;
2113 clp->cl_clientid.cl_id = nn->clientid_counter++;
2114 gen_confirm(clp, nn);
2117 static struct nfs4_stid *
2118 find_stateid_locked(struct nfs4_client *cl, stateid_t *t)
2120 struct nfs4_stid *ret;
2122 ret = idr_find(&cl->cl_stateids, t->si_opaque.so_id);
2123 if (!ret || !ret->sc_type)
2128 static struct nfs4_stid *
2129 find_stateid_by_type(struct nfs4_client *cl, stateid_t *t, char typemask)
2131 struct nfs4_stid *s;
2133 spin_lock(&cl->cl_lock);
2134 s = find_stateid_locked(cl, t);
2136 if (typemask & s->sc_type)
2137 refcount_inc(&s->sc_count);
2141 spin_unlock(&cl->cl_lock);
2145 static struct nfs4_client *create_client(struct xdr_netobj name,
2146 struct svc_rqst *rqstp, nfs4_verifier *verf)
2148 struct nfs4_client *clp;
2149 struct sockaddr *sa = svc_addr(rqstp);
2151 struct net *net = SVC_NET(rqstp);
2153 clp = alloc_client(name);
2157 ret = copy_cred(&clp->cl_cred, &rqstp->rq_cred);
2162 nfsd4_init_cb(&clp->cl_cb_null, clp, NULL, NFSPROC4_CLNT_CB_NULL);
2163 clp->cl_time = get_seconds();
2164 clear_bit(0, &clp->cl_cb_slot_busy);
2165 copy_verf(clp, verf);
2166 rpc_copy_addr((struct sockaddr *) &clp->cl_addr, sa);
2167 clp->cl_cb_session = NULL;
2173 add_clp_to_name_tree(struct nfs4_client *new_clp, struct rb_root *root)
2175 struct rb_node **new = &(root->rb_node), *parent = NULL;
2176 struct nfs4_client *clp;
2179 clp = rb_entry(*new, struct nfs4_client, cl_namenode);
2182 if (compare_blob(&clp->cl_name, &new_clp->cl_name) > 0)
2183 new = &((*new)->rb_left);
2185 new = &((*new)->rb_right);
2188 rb_link_node(&new_clp->cl_namenode, parent, new);
2189 rb_insert_color(&new_clp->cl_namenode, root);
2192 static struct nfs4_client *
2193 find_clp_in_name_tree(struct xdr_netobj *name, struct rb_root *root)
2196 struct rb_node *node = root->rb_node;
2197 struct nfs4_client *clp;
2200 clp = rb_entry(node, struct nfs4_client, cl_namenode);
2201 cmp = compare_blob(&clp->cl_name, name);
2203 node = node->rb_left;
2205 node = node->rb_right;
2213 add_to_unconfirmed(struct nfs4_client *clp)
2215 unsigned int idhashval;
2216 struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
2218 lockdep_assert_held(&nn->client_lock);
2220 clear_bit(NFSD4_CLIENT_CONFIRMED, &clp->cl_flags);
2221 add_clp_to_name_tree(clp, &nn->unconf_name_tree);
2222 idhashval = clientid_hashval(clp->cl_clientid.cl_id);
2223 list_add(&clp->cl_idhash, &nn->unconf_id_hashtbl[idhashval]);
2224 renew_client_locked(clp);
2228 move_to_confirmed(struct nfs4_client *clp)
2230 unsigned int idhashval = clientid_hashval(clp->cl_clientid.cl_id);
2231 struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
2233 lockdep_assert_held(&nn->client_lock);
2235 dprintk("NFSD: move_to_confirm nfs4_client %p\n", clp);
2236 list_move(&clp->cl_idhash, &nn->conf_id_hashtbl[idhashval]);
2237 rb_erase(&clp->cl_namenode, &nn->unconf_name_tree);
2238 add_clp_to_name_tree(clp, &nn->conf_name_tree);
2239 set_bit(NFSD4_CLIENT_CONFIRMED, &clp->cl_flags);
2240 renew_client_locked(clp);
2243 static struct nfs4_client *
2244 find_client_in_id_table(struct list_head *tbl, clientid_t *clid, bool sessions)
2246 struct nfs4_client *clp;
2247 unsigned int idhashval = clientid_hashval(clid->cl_id);
2249 list_for_each_entry(clp, &tbl[idhashval], cl_idhash) {
2250 if (same_clid(&clp->cl_clientid, clid)) {
2251 if ((bool)clp->cl_minorversion != sessions)
2253 renew_client_locked(clp);
2260 static struct nfs4_client *
2261 find_confirmed_client(clientid_t *clid, bool sessions, struct nfsd_net *nn)
2263 struct list_head *tbl = nn->conf_id_hashtbl;
2265 lockdep_assert_held(&nn->client_lock);
2266 return find_client_in_id_table(tbl, clid, sessions);
2269 static struct nfs4_client *
2270 find_unconfirmed_client(clientid_t *clid, bool sessions, struct nfsd_net *nn)
2272 struct list_head *tbl = nn->unconf_id_hashtbl;
2274 lockdep_assert_held(&nn->client_lock);
2275 return find_client_in_id_table(tbl, clid, sessions);
2278 static bool clp_used_exchangeid(struct nfs4_client *clp)
2280 return clp->cl_exchange_flags != 0;
2283 static struct nfs4_client *
2284 find_confirmed_client_by_name(struct xdr_netobj *name, struct nfsd_net *nn)
2286 lockdep_assert_held(&nn->client_lock);
2287 return find_clp_in_name_tree(name, &nn->conf_name_tree);
2290 static struct nfs4_client *
2291 find_unconfirmed_client_by_name(struct xdr_netobj *name, struct nfsd_net *nn)
2293 lockdep_assert_held(&nn->client_lock);
2294 return find_clp_in_name_tree(name, &nn->unconf_name_tree);
2298 gen_callback(struct nfs4_client *clp, struct nfsd4_setclientid *se, struct svc_rqst *rqstp)
2300 struct nfs4_cb_conn *conn = &clp->cl_cb_conn;
2301 struct sockaddr *sa = svc_addr(rqstp);
2302 u32 scopeid = rpc_get_scope_id(sa);
2303 unsigned short expected_family;
2305 /* Currently, we only support tcp and tcp6 for the callback channel */
2306 if (se->se_callback_netid_len == 3 &&
2307 !memcmp(se->se_callback_netid_val, "tcp", 3))
2308 expected_family = AF_INET;
2309 else if (se->se_callback_netid_len == 4 &&
2310 !memcmp(se->se_callback_netid_val, "tcp6", 4))
2311 expected_family = AF_INET6;
2315 conn->cb_addrlen = rpc_uaddr2sockaddr(clp->net, se->se_callback_addr_val,
2316 se->se_callback_addr_len,
2317 (struct sockaddr *)&conn->cb_addr,
2318 sizeof(conn->cb_addr));
2320 if (!conn->cb_addrlen || conn->cb_addr.ss_family != expected_family)
2323 if (conn->cb_addr.ss_family == AF_INET6)
2324 ((struct sockaddr_in6 *)&conn->cb_addr)->sin6_scope_id = scopeid;
2326 conn->cb_prog = se->se_callback_prog;
2327 conn->cb_ident = se->se_callback_ident;
2328 memcpy(&conn->cb_saddr, &rqstp->rq_daddr, rqstp->rq_daddrlen);
2331 conn->cb_addr.ss_family = AF_UNSPEC;
2332 conn->cb_addrlen = 0;
2333 dprintk("NFSD: this client (clientid %08x/%08x) "
2334 "will not receive delegations\n",
2335 clp->cl_clientid.cl_boot, clp->cl_clientid.cl_id);
2341 * Cache a reply. nfsd4_check_resp_size() has bounded the cache size.
2344 nfsd4_store_cache_entry(struct nfsd4_compoundres *resp)
2346 struct xdr_buf *buf = resp->xdr.buf;
2347 struct nfsd4_slot *slot = resp->cstate.slot;
2350 dprintk("--> %s slot %p\n", __func__, slot);
2352 slot->sl_flags |= NFSD4_SLOT_INITIALIZED;
2353 slot->sl_opcnt = resp->opcnt;
2354 slot->sl_status = resp->cstate.status;
2355 free_svc_cred(&slot->sl_cred);
2356 copy_cred(&slot->sl_cred, &resp->rqstp->rq_cred);
2358 if (!nfsd4_cache_this(resp)) {
2359 slot->sl_flags &= ~NFSD4_SLOT_CACHED;
2362 slot->sl_flags |= NFSD4_SLOT_CACHED;
2364 base = resp->cstate.data_offset;
2365 slot->sl_datalen = buf->len - base;
2366 if (read_bytes_from_xdr_buf(buf, base, slot->sl_data, slot->sl_datalen))
2367 WARN(1, "%s: sessions DRC could not cache compound\n",
2373 * Encode the replay sequence operation from the slot values.
2374 * If cachethis is FALSE encode the uncached rep error on the next
2375 * operation which sets resp->p and increments resp->opcnt for
2376 * nfs4svc_encode_compoundres.
2380 nfsd4_enc_sequence_replay(struct nfsd4_compoundargs *args,
2381 struct nfsd4_compoundres *resp)
2383 struct nfsd4_op *op;
2384 struct nfsd4_slot *slot = resp->cstate.slot;
2386 /* Encode the replayed sequence operation */
2387 op = &args->ops[resp->opcnt - 1];
2388 nfsd4_encode_operation(resp, op);
2390 if (slot->sl_flags & NFSD4_SLOT_CACHED)
2392 if (args->opcnt == 1) {
2394 * The original operation wasn't a solo sequence--we
2395 * always cache those--so this retry must not match the
2398 op->status = nfserr_seq_false_retry;
2400 op = &args->ops[resp->opcnt++];
2401 op->status = nfserr_retry_uncached_rep;
2402 nfsd4_encode_operation(resp, op);
2408 * The sequence operation is not cached because we can use the slot and
2412 nfsd4_replay_cache_entry(struct nfsd4_compoundres *resp,
2413 struct nfsd4_sequence *seq)
2415 struct nfsd4_slot *slot = resp->cstate.slot;
2416 struct xdr_stream *xdr = &resp->xdr;
2420 dprintk("--> %s slot %p\n", __func__, slot);
2422 status = nfsd4_enc_sequence_replay(resp->rqstp->rq_argp, resp);
2426 p = xdr_reserve_space(xdr, slot->sl_datalen);
2429 return nfserr_serverfault;
2431 xdr_encode_opaque_fixed(p, slot->sl_data, slot->sl_datalen);
2432 xdr_commit_encode(xdr);
2434 resp->opcnt = slot->sl_opcnt;
2435 return slot->sl_status;
2439 * Set the exchange_id flags returned by the server.
2442 nfsd4_set_ex_flags(struct nfs4_client *new, struct nfsd4_exchange_id *clid)
2444 #ifdef CONFIG_NFSD_PNFS
2445 new->cl_exchange_flags |= EXCHGID4_FLAG_USE_PNFS_MDS;
2447 new->cl_exchange_flags |= EXCHGID4_FLAG_USE_NON_PNFS;
2450 /* Referrals are supported, Migration is not. */
2451 new->cl_exchange_flags |= EXCHGID4_FLAG_SUPP_MOVED_REFER;
2453 /* set the wire flags to return to client. */
2454 clid->flags = new->cl_exchange_flags;
2457 static bool client_has_openowners(struct nfs4_client *clp)
2459 struct nfs4_openowner *oo;
2461 list_for_each_entry(oo, &clp->cl_openowners, oo_perclient) {
2462 if (!list_empty(&oo->oo_owner.so_stateids))
2468 static bool client_has_state(struct nfs4_client *clp)
2470 return client_has_openowners(clp)
2471 #ifdef CONFIG_NFSD_PNFS
2472 || !list_empty(&clp->cl_lo_states)
2474 || !list_empty(&clp->cl_delegations)
2475 || !list_empty(&clp->cl_sessions);
2479 nfsd4_exchange_id(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
2480 union nfsd4_op_u *u)
2482 struct nfsd4_exchange_id *exid = &u->exchange_id;
2483 struct nfs4_client *conf, *new;
2484 struct nfs4_client *unconf = NULL;
2486 char addr_str[INET6_ADDRSTRLEN];
2487 nfs4_verifier verf = exid->verifier;
2488 struct sockaddr *sa = svc_addr(rqstp);
2489 bool update = exid->flags & EXCHGID4_FLAG_UPD_CONFIRMED_REC_A;
2490 struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
2492 rpc_ntop(sa, addr_str, sizeof(addr_str));
2493 dprintk("%s rqstp=%p exid=%p clname.len=%u clname.data=%p "
2494 "ip_addr=%s flags %x, spa_how %d\n",
2495 __func__, rqstp, exid, exid->clname.len, exid->clname.data,
2496 addr_str, exid->flags, exid->spa_how);
2498 if (exid->flags & ~EXCHGID4_FLAG_MASK_A)
2499 return nfserr_inval;
2501 new = create_client(exid->clname, rqstp, &verf);
2503 return nfserr_jukebox;
2505 switch (exid->spa_how) {
2507 exid->spo_must_enforce[0] = 0;
2508 exid->spo_must_enforce[1] = (
2509 1 << (OP_BIND_CONN_TO_SESSION - 32) |
2510 1 << (OP_EXCHANGE_ID - 32) |
2511 1 << (OP_CREATE_SESSION - 32) |
2512 1 << (OP_DESTROY_SESSION - 32) |
2513 1 << (OP_DESTROY_CLIENTID - 32));
2515 exid->spo_must_allow[0] &= (1 << (OP_CLOSE) |
2516 1 << (OP_OPEN_DOWNGRADE) |
2518 1 << (OP_DELEGRETURN));
2520 exid->spo_must_allow[1] &= (
2521 1 << (OP_TEST_STATEID - 32) |
2522 1 << (OP_FREE_STATEID - 32));
2523 if (!svc_rqst_integrity_protected(rqstp)) {
2524 status = nfserr_inval;
2528 * Sometimes userspace doesn't give us a principal.
2529 * Which is a bug, really. Anyway, we can't enforce
2530 * MACH_CRED in that case, better to give up now:
2532 if (!new->cl_cred.cr_principal &&
2533 !new->cl_cred.cr_raw_principal) {
2534 status = nfserr_serverfault;
2537 new->cl_mach_cred = true;
2540 default: /* checked by xdr code */
2543 status = nfserr_encr_alg_unsupp;
2547 /* Cases below refer to rfc 5661 section 18.35.4: */
2548 spin_lock(&nn->client_lock);
2549 conf = find_confirmed_client_by_name(&exid->clname, nn);
2551 bool creds_match = same_creds(&conf->cl_cred, &rqstp->rq_cred);
2552 bool verfs_match = same_verf(&verf, &conf->cl_verifier);
2555 if (!clp_used_exchangeid(conf)) { /* buggy client */
2556 status = nfserr_inval;
2559 if (!nfsd4_mach_creds_match(conf, rqstp)) {
2560 status = nfserr_wrong_cred;
2563 if (!creds_match) { /* case 9 */
2564 status = nfserr_perm;
2567 if (!verfs_match) { /* case 8 */
2568 status = nfserr_not_same;
2572 exid->flags |= EXCHGID4_FLAG_CONFIRMED_R;
2575 if (!creds_match) { /* case 3 */
2576 if (client_has_state(conf)) {
2577 status = nfserr_clid_inuse;
2582 if (verfs_match) { /* case 2 */
2583 conf->cl_exchange_flags |= EXCHGID4_FLAG_CONFIRMED_R;
2586 /* case 5, client reboot */
2591 if (update) { /* case 7 */
2592 status = nfserr_noent;
2596 unconf = find_unconfirmed_client_by_name(&exid->clname, nn);
2597 if (unconf) /* case 4, possible retry or client restart */
2598 unhash_client_locked(unconf);
2600 /* case 1 (normal case) */
2603 status = mark_client_expired_locked(conf);
2607 new->cl_minorversion = cstate->minorversion;
2608 new->cl_spo_must_allow.u.words[0] = exid->spo_must_allow[0];
2609 new->cl_spo_must_allow.u.words[1] = exid->spo_must_allow[1];
2612 add_to_unconfirmed(new);
2615 exid->clientid.cl_boot = conf->cl_clientid.cl_boot;
2616 exid->clientid.cl_id = conf->cl_clientid.cl_id;
2618 exid->seqid = conf->cl_cs_slot.sl_seqid + 1;
2619 nfsd4_set_ex_flags(conf, exid);
2621 dprintk("nfsd4_exchange_id seqid %d flags %x\n",
2622 conf->cl_cs_slot.sl_seqid, conf->cl_exchange_flags);
2626 spin_unlock(&nn->client_lock);
2631 expire_client(unconf);
2636 check_slot_seqid(u32 seqid, u32 slot_seqid, int slot_inuse)
2638 dprintk("%s enter. seqid %d slot_seqid %d\n", __func__, seqid,
2641 /* The slot is in use, and no response has been sent. */
2643 if (seqid == slot_seqid)
2644 return nfserr_jukebox;
2646 return nfserr_seq_misordered;
2648 /* Note unsigned 32-bit arithmetic handles wraparound: */
2649 if (likely(seqid == slot_seqid + 1))
2651 if (seqid == slot_seqid)
2652 return nfserr_replay_cache;
2653 return nfserr_seq_misordered;
2657 * Cache the create session result into the create session single DRC
2658 * slot cache by saving the xdr structure. sl_seqid has been set.
2659 * Do this for solo or embedded create session operations.
2662 nfsd4_cache_create_session(struct nfsd4_create_session *cr_ses,
2663 struct nfsd4_clid_slot *slot, __be32 nfserr)
2665 slot->sl_status = nfserr;
2666 memcpy(&slot->sl_cr_ses, cr_ses, sizeof(*cr_ses));
2670 nfsd4_replay_create_session(struct nfsd4_create_session *cr_ses,
2671 struct nfsd4_clid_slot *slot)
2673 memcpy(cr_ses, &slot->sl_cr_ses, sizeof(*cr_ses));
2674 return slot->sl_status;
2677 #define NFSD_MIN_REQ_HDR_SEQ_SZ ((\
2678 2 * 2 + /* credential,verifier: AUTH_NULL, length 0 */ \
2679 1 + /* MIN tag is length with zero, only length */ \
2680 3 + /* version, opcount, opcode */ \
2681 XDR_QUADLEN(NFS4_MAX_SESSIONID_LEN) + \
2682 /* seqid, slotID, slotID, cache */ \
2683 4 ) * sizeof(__be32))
2685 #define NFSD_MIN_RESP_HDR_SEQ_SZ ((\
2686 2 + /* verifier: AUTH_NULL, length 0 */\
2688 1 + /* MIN tag is length with zero, only length */ \
2689 3 + /* opcount, opcode, opstatus*/ \
2690 XDR_QUADLEN(NFS4_MAX_SESSIONID_LEN) + \
2691 /* seqid, slotID, slotID, slotID, status */ \
2692 5 ) * sizeof(__be32))
2694 static __be32 check_forechannel_attrs(struct nfsd4_channel_attrs *ca, struct nfsd_net *nn)
2696 u32 maxrpc = nn->nfsd_serv->sv_max_mesg;
2698 if (ca->maxreq_sz < NFSD_MIN_REQ_HDR_SEQ_SZ)
2699 return nfserr_toosmall;
2700 if (ca->maxresp_sz < NFSD_MIN_RESP_HDR_SEQ_SZ)
2701 return nfserr_toosmall;
2702 ca->headerpadsz = 0;
2703 ca->maxreq_sz = min_t(u32, ca->maxreq_sz, maxrpc);
2704 ca->maxresp_sz = min_t(u32, ca->maxresp_sz, maxrpc);
2705 ca->maxops = min_t(u32, ca->maxops, NFSD_MAX_OPS_PER_COMPOUND);
2706 ca->maxresp_cached = min_t(u32, ca->maxresp_cached,
2707 NFSD_SLOT_CACHE_SIZE + NFSD_MIN_HDR_SEQ_SZ);
2708 ca->maxreqs = min_t(u32, ca->maxreqs, NFSD_MAX_SLOTS_PER_SESSION);
2710 * Note decreasing slot size below client's request may make it
2711 * difficult for client to function correctly, whereas
2712 * decreasing the number of slots will (just?) affect
2713 * performance. When short on memory we therefore prefer to
2714 * decrease number of slots instead of their size. Clients that
2715 * request larger slots than they need will get poor results:
2717 ca->maxreqs = nfsd4_get_drc_mem(ca);
2719 return nfserr_jukebox;
2725 * Server's NFSv4.1 backchannel support is AUTH_SYS-only for now.
2726 * These are based on similar macros in linux/sunrpc/msg_prot.h .
2728 #define RPC_MAX_HEADER_WITH_AUTH_SYS \
2729 (RPC_CALLHDRSIZE + 2 * (2 + UNX_CALLSLACK))
2731 #define RPC_MAX_REPHEADER_WITH_AUTH_SYS \
2732 (RPC_REPHDRSIZE + (2 + NUL_REPLYSLACK))
2734 #define NFSD_CB_MAX_REQ_SZ ((NFS4_enc_cb_recall_sz + \
2735 RPC_MAX_HEADER_WITH_AUTH_SYS) * sizeof(__be32))
2736 #define NFSD_CB_MAX_RESP_SZ ((NFS4_dec_cb_recall_sz + \
2737 RPC_MAX_REPHEADER_WITH_AUTH_SYS) * \
2740 static __be32 check_backchannel_attrs(struct nfsd4_channel_attrs *ca)
2742 ca->headerpadsz = 0;
2744 if (ca->maxreq_sz < NFSD_CB_MAX_REQ_SZ)
2745 return nfserr_toosmall;
2746 if (ca->maxresp_sz < NFSD_CB_MAX_RESP_SZ)
2747 return nfserr_toosmall;
2748 ca->maxresp_cached = 0;
2750 return nfserr_toosmall;
2755 static __be32 nfsd4_check_cb_sec(struct nfsd4_cb_sec *cbs)
2757 switch (cbs->flavor) {
2763 * GSS case: the spec doesn't allow us to return this
2764 * error. But it also doesn't allow us not to support
2766 * I'd rather this fail hard than return some error the
2767 * client might think it can already handle:
2769 return nfserr_encr_alg_unsupp;
2774 nfsd4_create_session(struct svc_rqst *rqstp,
2775 struct nfsd4_compound_state *cstate, union nfsd4_op_u *u)
2777 struct nfsd4_create_session *cr_ses = &u->create_session;
2778 struct sockaddr *sa = svc_addr(rqstp);
2779 struct nfs4_client *conf, *unconf;
2780 struct nfs4_client *old = NULL;
2781 struct nfsd4_session *new;
2782 struct nfsd4_conn *conn;
2783 struct nfsd4_clid_slot *cs_slot = NULL;
2785 struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
2787 if (cr_ses->flags & ~SESSION4_FLAG_MASK_A)
2788 return nfserr_inval;
2789 status = nfsd4_check_cb_sec(&cr_ses->cb_sec);
2792 status = check_forechannel_attrs(&cr_ses->fore_channel, nn);
2795 status = check_backchannel_attrs(&cr_ses->back_channel);
2797 goto out_release_drc_mem;
2798 status = nfserr_jukebox;
2799 new = alloc_session(&cr_ses->fore_channel, &cr_ses->back_channel);
2801 goto out_release_drc_mem;
2802 conn = alloc_conn_from_crses(rqstp, cr_ses);
2804 goto out_free_session;
2806 spin_lock(&nn->client_lock);
2807 unconf = find_unconfirmed_client(&cr_ses->clientid, true, nn);
2808 conf = find_confirmed_client(&cr_ses->clientid, true, nn);
2809 WARN_ON_ONCE(conf && unconf);
2812 status = nfserr_wrong_cred;
2813 if (!nfsd4_mach_creds_match(conf, rqstp))
2815 cs_slot = &conf->cl_cs_slot;
2816 status = check_slot_seqid(cr_ses->seqid, cs_slot->sl_seqid, 0);
2818 if (status == nfserr_replay_cache)
2819 status = nfsd4_replay_create_session(cr_ses, cs_slot);
2822 } else if (unconf) {
2823 if (!same_creds(&unconf->cl_cred, &rqstp->rq_cred) ||
2824 !rpc_cmp_addr(sa, (struct sockaddr *) &unconf->cl_addr)) {
2825 status = nfserr_clid_inuse;
2828 status = nfserr_wrong_cred;
2829 if (!nfsd4_mach_creds_match(unconf, rqstp))
2831 cs_slot = &unconf->cl_cs_slot;
2832 status = check_slot_seqid(cr_ses->seqid, cs_slot->sl_seqid, 0);
2834 /* an unconfirmed replay returns misordered */
2835 status = nfserr_seq_misordered;
2838 old = find_confirmed_client_by_name(&unconf->cl_name, nn);
2840 status = mark_client_expired_locked(old);
2846 move_to_confirmed(unconf);
2849 status = nfserr_stale_clientid;
2853 /* Persistent sessions are not supported */
2854 cr_ses->flags &= ~SESSION4_PERSIST;
2855 /* Upshifting from TCP to RDMA is not supported */
2856 cr_ses->flags &= ~SESSION4_RDMA;
2858 init_session(rqstp, new, conf, cr_ses);
2859 nfsd4_get_session_locked(new);
2861 memcpy(cr_ses->sessionid.data, new->se_sessionid.data,
2862 NFS4_MAX_SESSIONID_LEN);
2863 cs_slot->sl_seqid++;
2864 cr_ses->seqid = cs_slot->sl_seqid;
2866 /* cache solo and embedded create sessions under the client_lock */
2867 nfsd4_cache_create_session(cr_ses, cs_slot, status);
2868 spin_unlock(&nn->client_lock);
2869 /* init connection and backchannel */
2870 nfsd4_init_conn(rqstp, conn, new);
2871 nfsd4_put_session(new);
2876 spin_unlock(&nn->client_lock);
2881 __free_session(new);
2882 out_release_drc_mem:
2883 nfsd4_put_drc_mem(&cr_ses->fore_channel);
2887 static __be32 nfsd4_map_bcts_dir(u32 *dir)
2890 case NFS4_CDFC4_FORE:
2891 case NFS4_CDFC4_BACK:
2893 case NFS4_CDFC4_FORE_OR_BOTH:
2894 case NFS4_CDFC4_BACK_OR_BOTH:
2895 *dir = NFS4_CDFC4_BOTH;
2898 return nfserr_inval;
2901 __be32 nfsd4_backchannel_ctl(struct svc_rqst *rqstp,
2902 struct nfsd4_compound_state *cstate,
2903 union nfsd4_op_u *u)
2905 struct nfsd4_backchannel_ctl *bc = &u->backchannel_ctl;
2906 struct nfsd4_session *session = cstate->session;
2907 struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
2910 status = nfsd4_check_cb_sec(&bc->bc_cb_sec);
2913 spin_lock(&nn->client_lock);
2914 session->se_cb_prog = bc->bc_cb_program;
2915 session->se_cb_sec = bc->bc_cb_sec;
2916 spin_unlock(&nn->client_lock);
2918 nfsd4_probe_callback(session->se_client);
2923 __be32 nfsd4_bind_conn_to_session(struct svc_rqst *rqstp,
2924 struct nfsd4_compound_state *cstate,
2925 union nfsd4_op_u *u)
2927 struct nfsd4_bind_conn_to_session *bcts = &u->bind_conn_to_session;
2929 struct nfsd4_conn *conn;
2930 struct nfsd4_session *session;
2931 struct net *net = SVC_NET(rqstp);
2932 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
2934 if (!nfsd4_last_compound_op(rqstp))
2935 return nfserr_not_only_op;
2936 spin_lock(&nn->client_lock);
2937 session = find_in_sessionid_hashtbl(&bcts->sessionid, net, &status);
2938 spin_unlock(&nn->client_lock);
2940 goto out_no_session;
2941 status = nfserr_wrong_cred;
2942 if (!nfsd4_mach_creds_match(session->se_client, rqstp))
2944 status = nfsd4_map_bcts_dir(&bcts->dir);
2947 conn = alloc_conn(rqstp, bcts->dir);
2948 status = nfserr_jukebox;
2951 nfsd4_init_conn(rqstp, conn, session);
2954 nfsd4_put_session(session);
2959 static bool nfsd4_compound_in_session(struct nfsd4_session *session, struct nfs4_sessionid *sid)
2963 return !memcmp(sid, &session->se_sessionid, sizeof(*sid));
2967 nfsd4_destroy_session(struct svc_rqst *r, struct nfsd4_compound_state *cstate,
2968 union nfsd4_op_u *u)
2970 struct nfsd4_destroy_session *sessionid = &u->destroy_session;
2971 struct nfsd4_session *ses;
2973 int ref_held_by_me = 0;
2974 struct net *net = SVC_NET(r);
2975 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
2977 status = nfserr_not_only_op;
2978 if (nfsd4_compound_in_session(cstate->session, &sessionid->sessionid)) {
2979 if (!nfsd4_last_compound_op(r))
2983 dump_sessionid(__func__, &sessionid->sessionid);
2984 spin_lock(&nn->client_lock);
2985 ses = find_in_sessionid_hashtbl(&sessionid->sessionid, net, &status);
2987 goto out_client_lock;
2988 status = nfserr_wrong_cred;
2989 if (!nfsd4_mach_creds_match(ses->se_client, r))
2990 goto out_put_session;
2991 status = mark_session_dead_locked(ses, 1 + ref_held_by_me);
2993 goto out_put_session;
2994 unhash_session(ses);
2995 spin_unlock(&nn->client_lock);
2997 nfsd4_probe_callback_sync(ses->se_client);
2999 spin_lock(&nn->client_lock);
3002 nfsd4_put_session_locked(ses);
3004 spin_unlock(&nn->client_lock);
3009 static struct nfsd4_conn *__nfsd4_find_conn(struct svc_xprt *xpt, struct nfsd4_session *s)
3011 struct nfsd4_conn *c;
3013 list_for_each_entry(c, &s->se_conns, cn_persession) {
3014 if (c->cn_xprt == xpt) {
3021 static __be32 nfsd4_sequence_check_conn(struct nfsd4_conn *new, struct nfsd4_session *ses)
3023 struct nfs4_client *clp = ses->se_client;
3024 struct nfsd4_conn *c;
3025 __be32 status = nfs_ok;
3028 spin_lock(&clp->cl_lock);
3029 c = __nfsd4_find_conn(new->cn_xprt, ses);
3032 status = nfserr_conn_not_bound_to_session;
3033 if (clp->cl_mach_cred)
3035 __nfsd4_hash_conn(new, ses);
3036 spin_unlock(&clp->cl_lock);
3037 ret = nfsd4_register_conn(new);
3039 /* oops; xprt is already down: */
3040 nfsd4_conn_lost(&new->cn_xpt_user);
3043 spin_unlock(&clp->cl_lock);
3048 static bool nfsd4_session_too_many_ops(struct svc_rqst *rqstp, struct nfsd4_session *session)
3050 struct nfsd4_compoundargs *args = rqstp->rq_argp;
3052 return args->opcnt > session->se_fchannel.maxops;
3055 static bool nfsd4_request_too_big(struct svc_rqst *rqstp,
3056 struct nfsd4_session *session)
3058 struct xdr_buf *xb = &rqstp->rq_arg;
3060 return xb->len > session->se_fchannel.maxreq_sz;
3063 static bool replay_matches_cache(struct svc_rqst *rqstp,
3064 struct nfsd4_sequence *seq, struct nfsd4_slot *slot)
3066 struct nfsd4_compoundargs *argp = rqstp->rq_argp;
3068 if ((bool)(slot->sl_flags & NFSD4_SLOT_CACHETHIS) !=
3069 (bool)seq->cachethis)
3072 * If there's an error than the reply can have fewer ops than
3073 * the call. But if we cached a reply with *more* ops than the
3074 * call you're sending us now, then this new call is clearly not
3075 * really a replay of the old one:
3077 if (slot->sl_opcnt < argp->opcnt)
3079 /* This is the only check explicitly called by spec: */
3080 if (!same_creds(&rqstp->rq_cred, &slot->sl_cred))
3083 * There may be more comparisons we could actually do, but the
3084 * spec doesn't require us to catch every case where the calls
3085 * don't match (that would require caching the call as well as
3086 * the reply), so we don't bother.
3092 nfsd4_sequence(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3093 union nfsd4_op_u *u)
3095 struct nfsd4_sequence *seq = &u->sequence;
3096 struct nfsd4_compoundres *resp = rqstp->rq_resp;
3097 struct xdr_stream *xdr = &resp->xdr;
3098 struct nfsd4_session *session;
3099 struct nfs4_client *clp;
3100 struct nfsd4_slot *slot;
3101 struct nfsd4_conn *conn;
3104 struct net *net = SVC_NET(rqstp);
3105 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
3107 if (resp->opcnt != 1)
3108 return nfserr_sequence_pos;
3111 * Will be either used or freed by nfsd4_sequence_check_conn
3114 conn = alloc_conn(rqstp, NFS4_CDFC4_FORE);
3116 return nfserr_jukebox;
3118 spin_lock(&nn->client_lock);
3119 session = find_in_sessionid_hashtbl(&seq->sessionid, net, &status);
3121 goto out_no_session;
3122 clp = session->se_client;
3124 status = nfserr_too_many_ops;
3125 if (nfsd4_session_too_many_ops(rqstp, session))
3126 goto out_put_session;
3128 status = nfserr_req_too_big;
3129 if (nfsd4_request_too_big(rqstp, session))
3130 goto out_put_session;
3132 status = nfserr_badslot;
3133 if (seq->slotid >= session->se_fchannel.maxreqs)
3134 goto out_put_session;
3136 slot = session->se_slots[seq->slotid];
3137 dprintk("%s: slotid %d\n", __func__, seq->slotid);
3139 /* We do not negotiate the number of slots yet, so set the
3140 * maxslots to the session maxreqs which is used to encode
3141 * sr_highest_slotid and the sr_target_slot id to maxslots */
3142 seq->maxslots = session->se_fchannel.maxreqs;
3144 status = check_slot_seqid(seq->seqid, slot->sl_seqid,
3145 slot->sl_flags & NFSD4_SLOT_INUSE);
3146 if (status == nfserr_replay_cache) {
3147 status = nfserr_seq_misordered;
3148 if (!(slot->sl_flags & NFSD4_SLOT_INITIALIZED))
3149 goto out_put_session;
3150 status = nfserr_seq_false_retry;
3151 if (!replay_matches_cache(rqstp, seq, slot))
3152 goto out_put_session;
3153 cstate->slot = slot;
3154 cstate->session = session;
3156 /* Return the cached reply status and set cstate->status
3157 * for nfsd4_proc_compound processing */
3158 status = nfsd4_replay_cache_entry(resp, seq);
3159 cstate->status = nfserr_replay_cache;
3163 goto out_put_session;
3165 status = nfsd4_sequence_check_conn(conn, session);
3168 goto out_put_session;
3170 buflen = (seq->cachethis) ?
3171 session->se_fchannel.maxresp_cached :
3172 session->se_fchannel.maxresp_sz;
3173 status = (seq->cachethis) ? nfserr_rep_too_big_to_cache :
3175 if (xdr_restrict_buflen(xdr, buflen - rqstp->rq_auth_slack))
3176 goto out_put_session;
3177 svc_reserve(rqstp, buflen);
3180 /* Success! bump slot seqid */
3181 slot->sl_seqid = seq->seqid;
3182 slot->sl_flags |= NFSD4_SLOT_INUSE;
3184 slot->sl_flags |= NFSD4_SLOT_CACHETHIS;
3186 slot->sl_flags &= ~NFSD4_SLOT_CACHETHIS;
3188 cstate->slot = slot;
3189 cstate->session = session;
3193 switch (clp->cl_cb_state) {
3195 seq->status_flags = SEQ4_STATUS_CB_PATH_DOWN;
3197 case NFSD4_CB_FAULT:
3198 seq->status_flags = SEQ4_STATUS_BACKCHANNEL_FAULT;
3201 seq->status_flags = 0;
3203 if (!list_empty(&clp->cl_revoked))
3204 seq->status_flags |= SEQ4_STATUS_RECALLABLE_STATE_REVOKED;
3208 spin_unlock(&nn->client_lock);
3211 nfsd4_put_session_locked(session);
3212 goto out_no_session;
3216 nfsd4_sequence_done(struct nfsd4_compoundres *resp)
3218 struct nfsd4_compound_state *cs = &resp->cstate;
3220 if (nfsd4_has_session(cs)) {
3221 if (cs->status != nfserr_replay_cache) {
3222 nfsd4_store_cache_entry(resp);
3223 cs->slot->sl_flags &= ~NFSD4_SLOT_INUSE;
3225 /* Drop session reference that was taken in nfsd4_sequence() */
3226 nfsd4_put_session(cs->session);
3228 put_client_renew(cs->clp);
3232 nfsd4_destroy_clientid(struct svc_rqst *rqstp,
3233 struct nfsd4_compound_state *cstate,
3234 union nfsd4_op_u *u)
3236 struct nfsd4_destroy_clientid *dc = &u->destroy_clientid;
3237 struct nfs4_client *conf, *unconf;
3238 struct nfs4_client *clp = NULL;
3240 struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
3242 spin_lock(&nn->client_lock);
3243 unconf = find_unconfirmed_client(&dc->clientid, true, nn);
3244 conf = find_confirmed_client(&dc->clientid, true, nn);
3245 WARN_ON_ONCE(conf && unconf);
3248 if (client_has_state(conf)) {
3249 status = nfserr_clientid_busy;
3252 status = mark_client_expired_locked(conf);
3259 status = nfserr_stale_clientid;
3262 if (!nfsd4_mach_creds_match(clp, rqstp)) {
3264 status = nfserr_wrong_cred;
3267 unhash_client_locked(clp);
3269 spin_unlock(&nn->client_lock);
3276 nfsd4_reclaim_complete(struct svc_rqst *rqstp,
3277 struct nfsd4_compound_state *cstate, union nfsd4_op_u *u)
3279 struct nfsd4_reclaim_complete *rc = &u->reclaim_complete;
3282 if (rc->rca_one_fs) {
3283 if (!cstate->current_fh.fh_dentry)
3284 return nfserr_nofilehandle;
3286 * We don't take advantage of the rca_one_fs case.
3287 * That's OK, it's optional, we can safely ignore it.
3292 status = nfserr_complete_already;
3293 if (test_and_set_bit(NFSD4_CLIENT_RECLAIM_COMPLETE,
3294 &cstate->session->se_client->cl_flags))
3297 status = nfserr_stale_clientid;
3298 if (is_client_expired(cstate->session->se_client))
3300 * The following error isn't really legal.
3301 * But we only get here if the client just explicitly
3302 * destroyed the client. Surely it no longer cares what
3303 * error it gets back on an operation for the dead
3309 nfsd4_client_record_create(cstate->session->se_client);
3315 nfsd4_setclientid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3316 union nfsd4_op_u *u)
3318 struct nfsd4_setclientid *setclid = &u->setclientid;
3319 struct xdr_netobj clname = setclid->se_name;
3320 nfs4_verifier clverifier = setclid->se_verf;
3321 struct nfs4_client *conf, *new;
3322 struct nfs4_client *unconf = NULL;
3324 struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
3326 new = create_client(clname, rqstp, &clverifier);
3328 return nfserr_jukebox;
3329 /* Cases below refer to rfc 3530 section 14.2.33: */
3330 spin_lock(&nn->client_lock);
3331 conf = find_confirmed_client_by_name(&clname, nn);
3332 if (conf && client_has_state(conf)) {
3334 status = nfserr_clid_inuse;
3335 if (clp_used_exchangeid(conf))
3337 if (!same_creds(&conf->cl_cred, &rqstp->rq_cred)) {
3338 char addr_str[INET6_ADDRSTRLEN];
3339 rpc_ntop((struct sockaddr *) &conf->cl_addr, addr_str,
3341 dprintk("NFSD: setclientid: string in use by client "
3342 "at %s\n", addr_str);
3346 unconf = find_unconfirmed_client_by_name(&clname, nn);
3348 unhash_client_locked(unconf);
3349 if (conf && same_verf(&conf->cl_verifier, &clverifier)) {
3350 /* case 1: probable callback update */
3351 copy_clid(new, conf);
3352 gen_confirm(new, nn);
3353 } else /* case 4 (new client) or cases 2, 3 (client reboot): */
3355 new->cl_minorversion = 0;
3356 gen_callback(new, setclid, rqstp);
3357 add_to_unconfirmed(new);
3358 setclid->se_clientid.cl_boot = new->cl_clientid.cl_boot;
3359 setclid->se_clientid.cl_id = new->cl_clientid.cl_id;
3360 memcpy(setclid->se_confirm.data, new->cl_confirm.data, sizeof(setclid->se_confirm.data));
3364 spin_unlock(&nn->client_lock);
3368 expire_client(unconf);
3374 nfsd4_setclientid_confirm(struct svc_rqst *rqstp,
3375 struct nfsd4_compound_state *cstate,
3376 union nfsd4_op_u *u)
3378 struct nfsd4_setclientid_confirm *setclientid_confirm =
3379 &u->setclientid_confirm;
3380 struct nfs4_client *conf, *unconf;
3381 struct nfs4_client *old = NULL;
3382 nfs4_verifier confirm = setclientid_confirm->sc_confirm;
3383 clientid_t * clid = &setclientid_confirm->sc_clientid;
3385 struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
3387 if (STALE_CLIENTID(clid, nn))
3388 return nfserr_stale_clientid;
3390 spin_lock(&nn->client_lock);
3391 conf = find_confirmed_client(clid, false, nn);
3392 unconf = find_unconfirmed_client(clid, false, nn);
3394 * We try hard to give out unique clientid's, so if we get an
3395 * attempt to confirm the same clientid with a different cred,
3396 * the client may be buggy; this should never happen.
3398 * Nevertheless, RFC 7530 recommends INUSE for this case:
3400 status = nfserr_clid_inuse;
3401 if (unconf && !same_creds(&unconf->cl_cred, &rqstp->rq_cred))
3403 if (conf && !same_creds(&conf->cl_cred, &rqstp->rq_cred))
3405 /* cases below refer to rfc 3530 section 14.2.34: */
3406 if (!unconf || !same_verf(&confirm, &unconf->cl_confirm)) {
3407 if (conf && same_verf(&confirm, &conf->cl_confirm)) {
3408 /* case 2: probable retransmit */
3410 } else /* case 4: client hasn't noticed we rebooted yet? */
3411 status = nfserr_stale_clientid;
3415 if (conf) { /* case 1: callback update */
3417 unhash_client_locked(old);
3418 nfsd4_change_callback(conf, &unconf->cl_cb_conn);
3419 } else { /* case 3: normal case; new or rebooted client */
3420 old = find_confirmed_client_by_name(&unconf->cl_name, nn);
3422 status = nfserr_clid_inuse;
3423 if (client_has_state(old)
3424 && !same_creds(&unconf->cl_cred,
3427 status = mark_client_expired_locked(old);
3433 move_to_confirmed(unconf);
3436 get_client_locked(conf);
3437 spin_unlock(&nn->client_lock);
3438 nfsd4_probe_callback(conf);
3439 spin_lock(&nn->client_lock);
3440 put_client_renew_locked(conf);
3442 spin_unlock(&nn->client_lock);
3448 static struct nfs4_file *nfsd4_alloc_file(void)
3450 return kmem_cache_alloc(file_slab, GFP_KERNEL);
3453 /* OPEN Share state helper functions */
3454 static void nfsd4_init_file(struct knfsd_fh *fh, unsigned int hashval,
3455 struct nfs4_file *fp)
3457 lockdep_assert_held(&state_lock);
3459 refcount_set(&fp->fi_ref, 1);
3460 spin_lock_init(&fp->fi_lock);
3461 INIT_LIST_HEAD(&fp->fi_stateids);
3462 INIT_LIST_HEAD(&fp->fi_delegations);
3463 INIT_LIST_HEAD(&fp->fi_clnt_odstate);
3464 fh_copy_shallow(&fp->fi_fhandle, fh);
3465 fp->fi_deleg_file = NULL;
3466 fp->fi_had_conflict = false;
3467 fp->fi_share_deny = 0;
3468 memset(fp->fi_fds, 0, sizeof(fp->fi_fds));
3469 memset(fp->fi_access, 0, sizeof(fp->fi_access));
3470 #ifdef CONFIG_NFSD_PNFS
3471 INIT_LIST_HEAD(&fp->fi_lo_states);
3472 atomic_set(&fp->fi_lo_recalls, 0);
3474 hlist_add_head_rcu(&fp->fi_hash, &file_hashtbl[hashval]);
3478 nfsd4_free_slabs(void)
3480 kmem_cache_destroy(client_slab);
3481 kmem_cache_destroy(openowner_slab);
3482 kmem_cache_destroy(lockowner_slab);
3483 kmem_cache_destroy(file_slab);
3484 kmem_cache_destroy(stateid_slab);
3485 kmem_cache_destroy(deleg_slab);
3486 kmem_cache_destroy(odstate_slab);
3490 nfsd4_init_slabs(void)
3492 client_slab = kmem_cache_create("nfsd4_clients",
3493 sizeof(struct nfs4_client), 0, 0, NULL);
3494 if (client_slab == NULL)
3496 openowner_slab = kmem_cache_create("nfsd4_openowners",
3497 sizeof(struct nfs4_openowner), 0, 0, NULL);
3498 if (openowner_slab == NULL)
3499 goto out_free_client_slab;
3500 lockowner_slab = kmem_cache_create("nfsd4_lockowners",
3501 sizeof(struct nfs4_lockowner), 0, 0, NULL);
3502 if (lockowner_slab == NULL)
3503 goto out_free_openowner_slab;
3504 file_slab = kmem_cache_create("nfsd4_files",
3505 sizeof(struct nfs4_file), 0, 0, NULL);
3506 if (file_slab == NULL)
3507 goto out_free_lockowner_slab;
3508 stateid_slab = kmem_cache_create("nfsd4_stateids",
3509 sizeof(struct nfs4_ol_stateid), 0, 0, NULL);
3510 if (stateid_slab == NULL)
3511 goto out_free_file_slab;
3512 deleg_slab = kmem_cache_create("nfsd4_delegations",
3513 sizeof(struct nfs4_delegation), 0, 0, NULL);
3514 if (deleg_slab == NULL)
3515 goto out_free_stateid_slab;
3516 odstate_slab = kmem_cache_create("nfsd4_odstate",
3517 sizeof(struct nfs4_clnt_odstate), 0, 0, NULL);
3518 if (odstate_slab == NULL)
3519 goto out_free_deleg_slab;
3522 out_free_deleg_slab:
3523 kmem_cache_destroy(deleg_slab);
3524 out_free_stateid_slab:
3525 kmem_cache_destroy(stateid_slab);
3527 kmem_cache_destroy(file_slab);
3528 out_free_lockowner_slab:
3529 kmem_cache_destroy(lockowner_slab);
3530 out_free_openowner_slab:
3531 kmem_cache_destroy(openowner_slab);
3532 out_free_client_slab:
3533 kmem_cache_destroy(client_slab);
3535 dprintk("nfsd4: out of memory while initializing nfsv4\n");
3539 static void init_nfs4_replay(struct nfs4_replay *rp)
3541 rp->rp_status = nfserr_serverfault;
3543 rp->rp_buf = rp->rp_ibuf;
3544 mutex_init(&rp->rp_mutex);
3547 static void nfsd4_cstate_assign_replay(struct nfsd4_compound_state *cstate,
3548 struct nfs4_stateowner *so)
3550 if (!nfsd4_has_session(cstate)) {
3551 mutex_lock(&so->so_replay.rp_mutex);
3552 cstate->replay_owner = nfs4_get_stateowner(so);
3556 void nfsd4_cstate_clear_replay(struct nfsd4_compound_state *cstate)
3558 struct nfs4_stateowner *so = cstate->replay_owner;
3561 cstate->replay_owner = NULL;
3562 mutex_unlock(&so->so_replay.rp_mutex);
3563 nfs4_put_stateowner(so);
3567 static inline void *alloc_stateowner(struct kmem_cache *slab, struct xdr_netobj *owner, struct nfs4_client *clp)
3569 struct nfs4_stateowner *sop;
3571 sop = kmem_cache_alloc(slab, GFP_KERNEL);
3575 sop->so_owner.data = kmemdup(owner->data, owner->len, GFP_KERNEL);
3576 if (!sop->so_owner.data) {
3577 kmem_cache_free(slab, sop);
3580 sop->so_owner.len = owner->len;
3582 INIT_LIST_HEAD(&sop->so_stateids);
3583 sop->so_client = clp;
3584 init_nfs4_replay(&sop->so_replay);
3585 atomic_set(&sop->so_count, 1);
3589 static void hash_openowner(struct nfs4_openowner *oo, struct nfs4_client *clp, unsigned int strhashval)
3591 lockdep_assert_held(&clp->cl_lock);
3593 list_add(&oo->oo_owner.so_strhash,
3594 &clp->cl_ownerstr_hashtbl[strhashval]);
3595 list_add(&oo->oo_perclient, &clp->cl_openowners);
3598 static void nfs4_unhash_openowner(struct nfs4_stateowner *so)
3600 unhash_openowner_locked(openowner(so));
3603 static void nfs4_free_openowner(struct nfs4_stateowner *so)
3605 struct nfs4_openowner *oo = openowner(so);
3607 kmem_cache_free(openowner_slab, oo);
3610 static const struct nfs4_stateowner_operations openowner_ops = {
3611 .so_unhash = nfs4_unhash_openowner,
3612 .so_free = nfs4_free_openowner,
3615 static struct nfs4_ol_stateid *
3616 nfsd4_find_existing_open(struct nfs4_file *fp, struct nfsd4_open *open)
3618 struct nfs4_ol_stateid *local, *ret = NULL;
3619 struct nfs4_openowner *oo = open->op_openowner;
3621 lockdep_assert_held(&fp->fi_lock);
3623 list_for_each_entry(local, &fp->fi_stateids, st_perfile) {
3624 /* ignore lock owners */
3625 if (local->st_stateowner->so_is_open_owner == 0)
3627 if (local->st_stateowner != &oo->oo_owner)
3629 if (local->st_stid.sc_type == NFS4_OPEN_STID) {
3631 refcount_inc(&ret->st_stid.sc_count);
3639 nfsd4_verify_open_stid(struct nfs4_stid *s)
3641 __be32 ret = nfs_ok;
3643 switch (s->sc_type) {
3647 case NFS4_CLOSED_STID:
3648 case NFS4_CLOSED_DELEG_STID:
3649 ret = nfserr_bad_stateid;
3651 case NFS4_REVOKED_DELEG_STID:
3652 ret = nfserr_deleg_revoked;
3657 /* Lock the stateid st_mutex, and deal with races with CLOSE */
3659 nfsd4_lock_ol_stateid(struct nfs4_ol_stateid *stp)
3663 mutex_lock_nested(&stp->st_mutex, LOCK_STATEID_MUTEX);
3664 ret = nfsd4_verify_open_stid(&stp->st_stid);
3666 mutex_unlock(&stp->st_mutex);
3670 static struct nfs4_ol_stateid *
3671 nfsd4_find_and_lock_existing_open(struct nfs4_file *fp, struct nfsd4_open *open)
3673 struct nfs4_ol_stateid *stp;
3675 spin_lock(&fp->fi_lock);
3676 stp = nfsd4_find_existing_open(fp, open);
3677 spin_unlock(&fp->fi_lock);
3678 if (!stp || nfsd4_lock_ol_stateid(stp) == nfs_ok)
3680 nfs4_put_stid(&stp->st_stid);
3685 static struct nfs4_openowner *
3686 alloc_init_open_stateowner(unsigned int strhashval, struct nfsd4_open *open,
3687 struct nfsd4_compound_state *cstate)
3689 struct nfs4_client *clp = cstate->clp;
3690 struct nfs4_openowner *oo, *ret;
3692 oo = alloc_stateowner(openowner_slab, &open->op_owner, clp);
3695 oo->oo_owner.so_ops = &openowner_ops;
3696 oo->oo_owner.so_is_open_owner = 1;
3697 oo->oo_owner.so_seqid = open->op_seqid;
3699 if (nfsd4_has_session(cstate))
3700 oo->oo_flags |= NFS4_OO_CONFIRMED;
3702 oo->oo_last_closed_stid = NULL;
3703 INIT_LIST_HEAD(&oo->oo_close_lru);
3704 spin_lock(&clp->cl_lock);
3705 ret = find_openstateowner_str_locked(strhashval, open, clp);
3707 hash_openowner(oo, clp, strhashval);
3710 nfs4_free_stateowner(&oo->oo_owner);
3712 spin_unlock(&clp->cl_lock);
3716 static struct nfs4_ol_stateid *
3717 init_open_stateid(struct nfs4_file *fp, struct nfsd4_open *open)
3720 struct nfs4_openowner *oo = open->op_openowner;
3721 struct nfs4_ol_stateid *retstp = NULL;
3722 struct nfs4_ol_stateid *stp;
3725 /* We are moving these outside of the spinlocks to avoid the warnings */
3726 mutex_init(&stp->st_mutex);
3727 mutex_lock_nested(&stp->st_mutex, OPEN_STATEID_MUTEX);
3730 spin_lock(&oo->oo_owner.so_client->cl_lock);
3731 spin_lock(&fp->fi_lock);
3733 retstp = nfsd4_find_existing_open(fp, open);
3737 open->op_stp = NULL;
3738 refcount_inc(&stp->st_stid.sc_count);
3739 stp->st_stid.sc_type = NFS4_OPEN_STID;
3740 INIT_LIST_HEAD(&stp->st_locks);
3741 stp->st_stateowner = nfs4_get_stateowner(&oo->oo_owner);
3743 stp->st_stid.sc_file = fp;
3744 stp->st_access_bmap = 0;
3745 stp->st_deny_bmap = 0;
3746 stp->st_openstp = NULL;
3747 list_add(&stp->st_perstateowner, &oo->oo_owner.so_stateids);
3748 list_add(&stp->st_perfile, &fp->fi_stateids);
3751 spin_unlock(&fp->fi_lock);
3752 spin_unlock(&oo->oo_owner.so_client->cl_lock);
3754 /* Handle races with CLOSE */
3755 if (nfsd4_lock_ol_stateid(retstp) != nfs_ok) {
3756 nfs4_put_stid(&retstp->st_stid);
3759 /* To keep mutex tracking happy */
3760 mutex_unlock(&stp->st_mutex);
3767 * In the 4.0 case we need to keep the owners around a little while to handle
3768 * CLOSE replay. We still do need to release any file access that is held by
3769 * them before returning however.
3772 move_to_close_lru(struct nfs4_ol_stateid *s, struct net *net)
3774 struct nfs4_ol_stateid *last;
3775 struct nfs4_openowner *oo = openowner(s->st_stateowner);
3776 struct nfsd_net *nn = net_generic(s->st_stid.sc_client->net,
3779 dprintk("NFSD: move_to_close_lru nfs4_openowner %p\n", oo);
3782 * We know that we hold one reference via nfsd4_close, and another
3783 * "persistent" reference for the client. If the refcount is higher
3784 * than 2, then there are still calls in progress that are using this
3785 * stateid. We can't put the sc_file reference until they are finished.
3786 * Wait for the refcount to drop to 2. Since it has been unhashed,
3787 * there should be no danger of the refcount going back up again at
3790 wait_event(close_wq, refcount_read(&s->st_stid.sc_count) == 2);
3792 release_all_access(s);
3793 if (s->st_stid.sc_file) {
3794 put_nfs4_file(s->st_stid.sc_file);
3795 s->st_stid.sc_file = NULL;
3798 spin_lock(&nn->client_lock);
3799 last = oo->oo_last_closed_stid;
3800 oo->oo_last_closed_stid = s;
3801 list_move_tail(&oo->oo_close_lru, &nn->close_lru);
3802 oo->oo_time = get_seconds();
3803 spin_unlock(&nn->client_lock);
3805 nfs4_put_stid(&last->st_stid);
3808 /* search file_hashtbl[] for file */
3809 static struct nfs4_file *
3810 find_file_locked(struct knfsd_fh *fh, unsigned int hashval)
3812 struct nfs4_file *fp;
3814 hlist_for_each_entry_rcu(fp, &file_hashtbl[hashval], fi_hash) {
3815 if (fh_match(&fp->fi_fhandle, fh)) {
3816 if (refcount_inc_not_zero(&fp->fi_ref))
3824 find_file(struct knfsd_fh *fh)
3826 struct nfs4_file *fp;
3827 unsigned int hashval = file_hashval(fh);
3830 fp = find_file_locked(fh, hashval);
3835 static struct nfs4_file *
3836 find_or_add_file(struct nfs4_file *new, struct knfsd_fh *fh)
3838 struct nfs4_file *fp;
3839 unsigned int hashval = file_hashval(fh);
3842 fp = find_file_locked(fh, hashval);
3847 spin_lock(&state_lock);
3848 fp = find_file_locked(fh, hashval);
3849 if (likely(fp == NULL)) {
3850 nfsd4_init_file(fh, hashval, new);
3853 spin_unlock(&state_lock);
3859 * Called to check deny when READ with all zero stateid or
3860 * WRITE with all zero or all one stateid
3863 nfs4_share_conflict(struct svc_fh *current_fh, unsigned int deny_type)
3865 struct nfs4_file *fp;
3866 __be32 ret = nfs_ok;
3868 fp = find_file(¤t_fh->fh_handle);
3871 /* Check for conflicting share reservations */
3872 spin_lock(&fp->fi_lock);
3873 if (fp->fi_share_deny & deny_type)
3874 ret = nfserr_locked;
3875 spin_unlock(&fp->fi_lock);
3880 static void nfsd4_cb_recall_prepare(struct nfsd4_callback *cb)
3882 struct nfs4_delegation *dp = cb_to_delegation(cb);
3883 struct nfsd_net *nn = net_generic(dp->dl_stid.sc_client->net,
3886 block_delegations(&dp->dl_stid.sc_file->fi_fhandle);
3889 * We can't do this in nfsd_break_deleg_cb because it is
3890 * already holding inode->i_lock.
3892 * If the dl_time != 0, then we know that it has already been
3893 * queued for a lease break. Don't queue it again.
3895 spin_lock(&state_lock);
3896 if (dp->dl_time == 0) {
3897 dp->dl_time = get_seconds();
3898 list_add_tail(&dp->dl_recall_lru, &nn->del_recall_lru);
3900 spin_unlock(&state_lock);
3903 static int nfsd4_cb_recall_done(struct nfsd4_callback *cb,
3904 struct rpc_task *task)
3906 struct nfs4_delegation *dp = cb_to_delegation(cb);
3908 if (dp->dl_stid.sc_type == NFS4_CLOSED_DELEG_STID)
3911 switch (task->tk_status) {
3915 case -NFS4ERR_BAD_STATEID:
3917 * Race: client probably got cb_recall before open reply
3918 * granting delegation.
3920 if (dp->dl_retries--) {
3921 rpc_delay(task, 2 * HZ);
3930 static void nfsd4_cb_recall_release(struct nfsd4_callback *cb)
3932 struct nfs4_delegation *dp = cb_to_delegation(cb);
3934 nfs4_put_stid(&dp->dl_stid);
3937 static const struct nfsd4_callback_ops nfsd4_cb_recall_ops = {
3938 .prepare = nfsd4_cb_recall_prepare,
3939 .done = nfsd4_cb_recall_done,
3940 .release = nfsd4_cb_recall_release,
3943 static void nfsd_break_one_deleg(struct nfs4_delegation *dp)
3946 * We're assuming the state code never drops its reference
3947 * without first removing the lease. Since we're in this lease
3948 * callback (and since the lease code is serialized by the kernel
3949 * lock) we know the server hasn't removed the lease yet, we know
3950 * it's safe to take a reference.
3952 refcount_inc(&dp->dl_stid.sc_count);
3953 nfsd4_run_cb(&dp->dl_recall);
3956 /* Called from break_lease() with i_lock held. */
3958 nfsd_break_deleg_cb(struct file_lock *fl)
3961 struct nfs4_delegation *dp = (struct nfs4_delegation *)fl->fl_owner;
3962 struct nfs4_file *fp = dp->dl_stid.sc_file;
3965 * We don't want the locks code to timeout the lease for us;
3966 * we'll remove it ourself if a delegation isn't returned
3969 fl->fl_break_time = 0;
3971 spin_lock(&fp->fi_lock);
3972 fp->fi_had_conflict = true;
3973 nfsd_break_one_deleg(dp);
3974 spin_unlock(&fp->fi_lock);
3979 nfsd_change_deleg_cb(struct file_lock *onlist, int arg,
3980 struct list_head *dispose)
3983 return lease_modify(onlist, arg, dispose);
3988 static const struct lock_manager_operations nfsd_lease_mng_ops = {
3989 .lm_break = nfsd_break_deleg_cb,
3990 .lm_change = nfsd_change_deleg_cb,
3993 static __be32 nfsd4_check_seqid(struct nfsd4_compound_state *cstate, struct nfs4_stateowner *so, u32 seqid)
3995 if (nfsd4_has_session(cstate))
3997 if (seqid == so->so_seqid - 1)
3998 return nfserr_replay_me;
3999 if (seqid == so->so_seqid)
4001 return nfserr_bad_seqid;
4004 static __be32 lookup_clientid(clientid_t *clid,
4005 struct nfsd4_compound_state *cstate,
4006 struct nfsd_net *nn)
4008 struct nfs4_client *found;
4011 found = cstate->clp;
4012 if (!same_clid(&found->cl_clientid, clid))
4013 return nfserr_stale_clientid;
4017 if (STALE_CLIENTID(clid, nn))
4018 return nfserr_stale_clientid;
4021 * For v4.1+ we get the client in the SEQUENCE op. If we don't have one
4022 * cached already then we know this is for is for v4.0 and "sessions"
4025 WARN_ON_ONCE(cstate->session);
4026 spin_lock(&nn->client_lock);
4027 found = find_confirmed_client(clid, false, nn);
4029 spin_unlock(&nn->client_lock);
4030 return nfserr_expired;
4032 atomic_inc(&found->cl_refcount);
4033 spin_unlock(&nn->client_lock);
4035 /* Cache the nfs4_client in cstate! */
4036 cstate->clp = found;
4041 nfsd4_process_open1(struct nfsd4_compound_state *cstate,
4042 struct nfsd4_open *open, struct nfsd_net *nn)
4044 clientid_t *clientid = &open->op_clientid;
4045 struct nfs4_client *clp = NULL;
4046 unsigned int strhashval;
4047 struct nfs4_openowner *oo = NULL;
4050 if (STALE_CLIENTID(&open->op_clientid, nn))
4051 return nfserr_stale_clientid;
4053 * In case we need it later, after we've already created the
4054 * file and don't want to risk a further failure:
4056 open->op_file = nfsd4_alloc_file();
4057 if (open->op_file == NULL)
4058 return nfserr_jukebox;
4060 status = lookup_clientid(clientid, cstate, nn);
4065 strhashval = ownerstr_hashval(&open->op_owner);
4066 oo = find_openstateowner_str(strhashval, open, clp);
4067 open->op_openowner = oo;
4071 if (!(oo->oo_flags & NFS4_OO_CONFIRMED)) {
4072 /* Replace unconfirmed owners without checking for replay. */
4073 release_openowner(oo);
4074 open->op_openowner = NULL;
4077 status = nfsd4_check_seqid(cstate, &oo->oo_owner, open->op_seqid);
4082 oo = alloc_init_open_stateowner(strhashval, open, cstate);
4084 return nfserr_jukebox;
4085 open->op_openowner = oo;
4087 open->op_stp = nfs4_alloc_open_stateid(clp);
4089 return nfserr_jukebox;
4091 if (nfsd4_has_session(cstate) &&
4092 (cstate->current_fh.fh_export->ex_flags & NFSEXP_PNFS)) {
4093 open->op_odstate = alloc_clnt_odstate(clp);
4094 if (!open->op_odstate)
4095 return nfserr_jukebox;
4101 static inline __be32
4102 nfs4_check_delegmode(struct nfs4_delegation *dp, int flags)
4104 if ((flags & WR_STATE) && (dp->dl_type == NFS4_OPEN_DELEGATE_READ))
4105 return nfserr_openmode;
4110 static int share_access_to_flags(u32 share_access)
4112 return share_access == NFS4_SHARE_ACCESS_READ ? RD_STATE : WR_STATE;
4115 static struct nfs4_delegation *find_deleg_stateid(struct nfs4_client *cl, stateid_t *s)
4117 struct nfs4_stid *ret;
4119 ret = find_stateid_by_type(cl, s,
4120 NFS4_DELEG_STID|NFS4_REVOKED_DELEG_STID);
4123 return delegstateid(ret);
4126 static bool nfsd4_is_deleg_cur(struct nfsd4_open *open)
4128 return open->op_claim_type == NFS4_OPEN_CLAIM_DELEGATE_CUR ||
4129 open->op_claim_type == NFS4_OPEN_CLAIM_DELEG_CUR_FH;
4133 nfs4_check_deleg(struct nfs4_client *cl, struct nfsd4_open *open,
4134 struct nfs4_delegation **dp)
4137 __be32 status = nfserr_bad_stateid;
4138 struct nfs4_delegation *deleg;
4140 deleg = find_deleg_stateid(cl, &open->op_delegate_stateid);
4143 if (deleg->dl_stid.sc_type == NFS4_REVOKED_DELEG_STID) {
4144 nfs4_put_stid(&deleg->dl_stid);
4145 if (cl->cl_minorversion)
4146 status = nfserr_deleg_revoked;
4149 flags = share_access_to_flags(open->op_share_access);
4150 status = nfs4_check_delegmode(deleg, flags);
4152 nfs4_put_stid(&deleg->dl_stid);
4157 if (!nfsd4_is_deleg_cur(open))
4161 open->op_openowner->oo_flags |= NFS4_OO_CONFIRMED;
4165 static inline int nfs4_access_to_access(u32 nfs4_access)
4169 if (nfs4_access & NFS4_SHARE_ACCESS_READ)
4170 flags |= NFSD_MAY_READ;
4171 if (nfs4_access & NFS4_SHARE_ACCESS_WRITE)
4172 flags |= NFSD_MAY_WRITE;
4176 static inline __be32
4177 nfsd4_truncate(struct svc_rqst *rqstp, struct svc_fh *fh,
4178 struct nfsd4_open *open)
4180 struct iattr iattr = {
4181 .ia_valid = ATTR_SIZE,
4184 if (!open->op_truncate)
4186 if (!(open->op_share_access & NFS4_SHARE_ACCESS_WRITE))
4187 return nfserr_inval;
4188 return nfsd_setattr(rqstp, fh, &iattr, 0, (time_t)0);
4191 static __be32 nfs4_get_vfs_file(struct svc_rqst *rqstp, struct nfs4_file *fp,
4192 struct svc_fh *cur_fh, struct nfs4_ol_stateid *stp,
4193 struct nfsd4_open *open)
4195 struct file *filp = NULL;
4197 int oflag = nfs4_access_to_omode(open->op_share_access);
4198 int access = nfs4_access_to_access(open->op_share_access);
4199 unsigned char old_access_bmap, old_deny_bmap;
4201 spin_lock(&fp->fi_lock);
4204 * Are we trying to set a deny mode that would conflict with
4207 status = nfs4_file_check_deny(fp, open->op_share_deny);
4208 if (status != nfs_ok) {
4209 spin_unlock(&fp->fi_lock);
4213 /* set access to the file */
4214 status = nfs4_file_get_access(fp, open->op_share_access);
4215 if (status != nfs_ok) {
4216 spin_unlock(&fp->fi_lock);
4220 /* Set access bits in stateid */
4221 old_access_bmap = stp->st_access_bmap;
4222 set_access(open->op_share_access, stp);
4224 /* Set new deny mask */
4225 old_deny_bmap = stp->st_deny_bmap;
4226 set_deny(open->op_share_deny, stp);
4227 fp->fi_share_deny |= (open->op_share_deny & NFS4_SHARE_DENY_BOTH);
4229 if (!fp->fi_fds[oflag]) {
4230 spin_unlock(&fp->fi_lock);
4231 status = nfsd_open(rqstp, cur_fh, S_IFREG, access, &filp);
4233 goto out_put_access;
4234 spin_lock(&fp->fi_lock);
4235 if (!fp->fi_fds[oflag]) {
4236 fp->fi_fds[oflag] = filp;
4240 spin_unlock(&fp->fi_lock);
4244 status = nfsd4_truncate(rqstp, cur_fh, open);
4246 goto out_put_access;
4250 stp->st_access_bmap = old_access_bmap;
4251 nfs4_file_put_access(fp, open->op_share_access);
4252 reset_union_bmap_deny(bmap_to_share_mode(old_deny_bmap), stp);
4257 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)
4260 unsigned char old_deny_bmap = stp->st_deny_bmap;
4262 if (!test_access(open->op_share_access, stp))
4263 return nfs4_get_vfs_file(rqstp, fp, cur_fh, stp, open);
4265 /* test and set deny mode */
4266 spin_lock(&fp->fi_lock);
4267 status = nfs4_file_check_deny(fp, open->op_share_deny);
4268 if (status == nfs_ok) {
4269 set_deny(open->op_share_deny, stp);
4270 fp->fi_share_deny |=
4271 (open->op_share_deny & NFS4_SHARE_DENY_BOTH);
4273 spin_unlock(&fp->fi_lock);
4275 if (status != nfs_ok)
4278 status = nfsd4_truncate(rqstp, cur_fh, open);
4279 if (status != nfs_ok)
4280 reset_union_bmap_deny(old_deny_bmap, stp);
4284 /* Should we give out recallable state?: */
4285 static bool nfsd4_cb_channel_good(struct nfs4_client *clp)
4287 if (clp->cl_cb_state == NFSD4_CB_UP)
4290 * In the sessions case, since we don't have to establish a
4291 * separate connection for callbacks, we assume it's OK
4292 * until we hear otherwise:
4294 return clp->cl_minorversion && clp->cl_cb_state == NFSD4_CB_UNKNOWN;
4297 static struct file_lock *nfs4_alloc_init_lease(struct nfs4_delegation *dp,
4300 struct file_lock *fl;
4302 fl = locks_alloc_lock();
4305 fl->fl_lmops = &nfsd_lease_mng_ops;
4306 fl->fl_flags = FL_DELEG;
4307 fl->fl_type = flag == NFS4_OPEN_DELEGATE_READ? F_RDLCK: F_WRLCK;
4308 fl->fl_end = OFFSET_MAX;
4309 fl->fl_owner = (fl_owner_t)dp;
4310 fl->fl_pid = current->tgid;
4311 fl->fl_file = dp->dl_stid.sc_file->fi_deleg_file;
4315 static struct nfs4_delegation *
4316 nfs4_set_delegation(struct nfs4_client *clp, struct svc_fh *fh,
4317 struct nfs4_file *fp, struct nfs4_clnt_odstate *odstate)
4320 struct nfs4_delegation *dp;
4322 struct file_lock *fl;
4325 * The fi_had_conflict and nfs_get_existing_delegation checks
4326 * here are just optimizations; we'll need to recheck them at
4329 if (fp->fi_had_conflict)
4330 return ERR_PTR(-EAGAIN);
4332 filp = find_readable_file(fp);
4334 /* We should always have a readable file here */
4336 return ERR_PTR(-EBADF);
4338 spin_lock(&state_lock);
4339 spin_lock(&fp->fi_lock);
4340 if (nfs4_delegation_exists(clp, fp))
4342 else if (!fp->fi_deleg_file) {
4343 fp->fi_deleg_file = filp;
4344 /* increment early to prevent fi_deleg_file from being
4346 fp->fi_delegees = 1;
4350 spin_unlock(&fp->fi_lock);
4351 spin_unlock(&state_lock);
4355 return ERR_PTR(status);
4358 dp = alloc_init_deleg(clp, fp, fh, odstate);
4362 fl = nfs4_alloc_init_lease(dp, NFS4_OPEN_DELEGATE_READ);
4366 status = vfs_setlease(fp->fi_deleg_file, fl->fl_type, &fl, NULL);
4368 locks_free_lock(fl);
4370 goto out_clnt_odstate;
4372 spin_lock(&state_lock);
4373 spin_lock(&fp->fi_lock);
4374 if (fp->fi_had_conflict)
4377 status = hash_delegation_locked(dp, fp);
4378 spin_unlock(&fp->fi_lock);
4379 spin_unlock(&state_lock);
4382 destroy_unhashed_deleg(dp);
4385 put_clnt_odstate(dp->dl_clnt_odstate);
4387 nfs4_put_stid(&dp->dl_stid);
4390 return ERR_PTR(status);
4393 static void nfsd4_open_deleg_none_ext(struct nfsd4_open *open, int status)
4395 open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE_EXT;
4396 if (status == -EAGAIN)
4397 open->op_why_no_deleg = WND4_CONTENTION;
4399 open->op_why_no_deleg = WND4_RESOURCE;
4400 switch (open->op_deleg_want) {
4401 case NFS4_SHARE_WANT_READ_DELEG:
4402 case NFS4_SHARE_WANT_WRITE_DELEG:
4403 case NFS4_SHARE_WANT_ANY_DELEG:
4405 case NFS4_SHARE_WANT_CANCEL:
4406 open->op_why_no_deleg = WND4_CANCELLED;
4408 case NFS4_SHARE_WANT_NO_DELEG:
4415 * Attempt to hand out a delegation.
4417 * Note we don't support write delegations, and won't until the vfs has
4418 * proper support for them.
4421 nfs4_open_delegation(struct svc_fh *fh, struct nfsd4_open *open,
4422 struct nfs4_ol_stateid *stp)
4424 struct nfs4_delegation *dp;
4425 struct nfs4_openowner *oo = openowner(stp->st_stateowner);
4426 struct nfs4_client *clp = stp->st_stid.sc_client;
4430 cb_up = nfsd4_cb_channel_good(oo->oo_owner.so_client);
4431 open->op_recall = 0;
4432 switch (open->op_claim_type) {
4433 case NFS4_OPEN_CLAIM_PREVIOUS:
4435 open->op_recall = 1;
4436 if (open->op_delegate_type != NFS4_OPEN_DELEGATE_READ)
4439 case NFS4_OPEN_CLAIM_NULL:
4440 case NFS4_OPEN_CLAIM_FH:
4442 * Let's not give out any delegations till everyone's
4443 * had the chance to reclaim theirs, *and* until
4444 * NLM locks have all been reclaimed:
4446 if (locks_in_grace(clp->net))
4448 if (!cb_up || !(oo->oo_flags & NFS4_OO_CONFIRMED))
4451 * Also, if the file was opened for write or
4452 * create, there's a good chance the client's
4453 * about to write to it, resulting in an
4454 * immediate recall (since we don't support
4455 * write delegations):
4457 if (open->op_share_access & NFS4_SHARE_ACCESS_WRITE)
4459 if (open->op_create == NFS4_OPEN_CREATE)
4465 dp = nfs4_set_delegation(clp, fh, stp->st_stid.sc_file, stp->st_clnt_odstate);
4469 memcpy(&open->op_delegate_stateid, &dp->dl_stid.sc_stateid, sizeof(dp->dl_stid.sc_stateid));
4471 dprintk("NFSD: delegation stateid=" STATEID_FMT "\n",
4472 STATEID_VAL(&dp->dl_stid.sc_stateid));
4473 open->op_delegate_type = NFS4_OPEN_DELEGATE_READ;
4474 nfs4_put_stid(&dp->dl_stid);
4477 open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE;
4478 if (open->op_claim_type == NFS4_OPEN_CLAIM_PREVIOUS &&
4479 open->op_delegate_type != NFS4_OPEN_DELEGATE_NONE) {
4480 dprintk("NFSD: WARNING: refusing delegation reclaim\n");
4481 open->op_recall = 1;
4484 /* 4.1 client asking for a delegation? */
4485 if (open->op_deleg_want)
4486 nfsd4_open_deleg_none_ext(open, status);
4490 static void nfsd4_deleg_xgrade_none_ext(struct nfsd4_open *open,
4491 struct nfs4_delegation *dp)
4493 if (open->op_deleg_want == NFS4_SHARE_WANT_READ_DELEG &&
4494 dp->dl_type == NFS4_OPEN_DELEGATE_WRITE) {
4495 open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE_EXT;
4496 open->op_why_no_deleg = WND4_NOT_SUPP_DOWNGRADE;
4497 } else if (open->op_deleg_want == NFS4_SHARE_WANT_WRITE_DELEG &&
4498 dp->dl_type == NFS4_OPEN_DELEGATE_WRITE) {
4499 open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE_EXT;
4500 open->op_why_no_deleg = WND4_NOT_SUPP_UPGRADE;
4502 /* Otherwise the client must be confused wanting a delegation
4503 * it already has, therefore we don't return
4504 * NFS4_OPEN_DELEGATE_NONE_EXT and reason.
4509 nfsd4_process_open2(struct svc_rqst *rqstp, struct svc_fh *current_fh, struct nfsd4_open *open)
4511 struct nfsd4_compoundres *resp = rqstp->rq_resp;
4512 struct nfs4_client *cl = open->op_openowner->oo_owner.so_client;
4513 struct nfs4_file *fp = NULL;
4514 struct nfs4_ol_stateid *stp = NULL;
4515 struct nfs4_delegation *dp = NULL;
4517 bool new_stp = false;
4520 * Lookup file; if found, lookup stateid and check open request,
4521 * and check for delegations in the process of being recalled.
4522 * If not found, create the nfs4_file struct
4524 fp = find_or_add_file(open->op_file, ¤t_fh->fh_handle);
4525 if (fp != open->op_file) {
4526 status = nfs4_check_deleg(cl, open, &dp);
4529 stp = nfsd4_find_and_lock_existing_open(fp, open);
4531 open->op_file = NULL;
4532 status = nfserr_bad_stateid;
4533 if (nfsd4_is_deleg_cur(open))
4538 stp = init_open_stateid(fp, open);
4544 * OPEN the file, or upgrade an existing OPEN.
4545 * If truncate fails, the OPEN fails.
4547 * stp is already locked.
4550 /* Stateid was found, this is an OPEN upgrade */
4551 status = nfs4_upgrade_open(rqstp, fp, current_fh, stp, open);
4553 mutex_unlock(&stp->st_mutex);
4557 status = nfs4_get_vfs_file(rqstp, fp, current_fh, stp, open);
4559 stp->st_stid.sc_type = NFS4_CLOSED_STID;
4560 release_open_stateid(stp);
4561 mutex_unlock(&stp->st_mutex);
4565 stp->st_clnt_odstate = find_or_hash_clnt_odstate(fp,
4567 if (stp->st_clnt_odstate == open->op_odstate)
4568 open->op_odstate = NULL;
4571 nfs4_inc_and_copy_stateid(&open->op_stateid, &stp->st_stid);
4572 mutex_unlock(&stp->st_mutex);
4574 if (nfsd4_has_session(&resp->cstate)) {
4575 if (open->op_deleg_want & NFS4_SHARE_WANT_NO_DELEG) {
4576 open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE_EXT;
4577 open->op_why_no_deleg = WND4_NOT_WANTED;
4583 * Attempt to hand out a delegation. No error return, because the
4584 * OPEN succeeds even if we fail.
4586 nfs4_open_delegation(current_fh, open, stp);
4590 dprintk("%s: stateid=" STATEID_FMT "\n", __func__,
4591 STATEID_VAL(&stp->st_stid.sc_stateid));
4593 /* 4.1 client trying to upgrade/downgrade delegation? */
4594 if (open->op_delegate_type == NFS4_OPEN_DELEGATE_NONE && dp &&
4595 open->op_deleg_want)
4596 nfsd4_deleg_xgrade_none_ext(open, dp);
4600 if (status == 0 && open->op_claim_type == NFS4_OPEN_CLAIM_PREVIOUS)
4601 open->op_openowner->oo_flags |= NFS4_OO_CONFIRMED;
4603 * To finish the open response, we just need to set the rflags.
4605 open->op_rflags = NFS4_OPEN_RESULT_LOCKTYPE_POSIX;
4606 if (nfsd4_has_session(&resp->cstate))
4607 open->op_rflags |= NFS4_OPEN_RESULT_MAY_NOTIFY_LOCK;
4608 else if (!(open->op_openowner->oo_flags & NFS4_OO_CONFIRMED))
4609 open->op_rflags |= NFS4_OPEN_RESULT_CONFIRM;
4612 nfs4_put_stid(&dp->dl_stid);
4614 nfs4_put_stid(&stp->st_stid);
4619 void nfsd4_cleanup_open_state(struct nfsd4_compound_state *cstate,
4620 struct nfsd4_open *open)
4622 if (open->op_openowner) {
4623 struct nfs4_stateowner *so = &open->op_openowner->oo_owner;
4625 nfsd4_cstate_assign_replay(cstate, so);
4626 nfs4_put_stateowner(so);
4629 kmem_cache_free(file_slab, open->op_file);
4631 nfs4_put_stid(&open->op_stp->st_stid);
4632 if (open->op_odstate)
4633 kmem_cache_free(odstate_slab, open->op_odstate);
4637 nfsd4_renew(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
4638 union nfsd4_op_u *u)
4640 clientid_t *clid = &u->renew;
4641 struct nfs4_client *clp;
4643 struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
4645 dprintk("process_renew(%08x/%08x): starting\n",
4646 clid->cl_boot, clid->cl_id);
4647 status = lookup_clientid(clid, cstate, nn);
4651 status = nfserr_cb_path_down;
4652 if (!list_empty(&clp->cl_delegations)
4653 && clp->cl_cb_state != NFSD4_CB_UP)
4661 nfsd4_end_grace(struct nfsd_net *nn)
4663 /* do nothing if grace period already ended */
4664 if (nn->grace_ended)
4667 dprintk("NFSD: end of grace period\n");
4668 nn->grace_ended = true;
4670 * If the server goes down again right now, an NFSv4
4671 * client will still be allowed to reclaim after it comes back up,
4672 * even if it hasn't yet had a chance to reclaim state this time.
4675 nfsd4_record_grace_done(nn);
4677 * At this point, NFSv4 clients can still reclaim. But if the
4678 * server crashes, any that have not yet reclaimed will be out
4679 * of luck on the next boot.
4681 * (NFSv4.1+ clients are considered to have reclaimed once they
4682 * call RECLAIM_COMPLETE. NFSv4.0 clients are considered to
4683 * have reclaimed after their first OPEN.)
4685 locks_end_grace(&nn->nfsd4_manager);
4687 * At this point, and once lockd and/or any other containers
4688 * exit their grace period, further reclaims will fail and
4689 * regular locking can resume.
4694 nfs4_laundromat(struct nfsd_net *nn)
4696 struct nfs4_client *clp;
4697 struct nfs4_openowner *oo;
4698 struct nfs4_delegation *dp;
4699 struct nfs4_ol_stateid *stp;
4700 struct nfsd4_blocked_lock *nbl;
4701 struct list_head *pos, *next, reaplist;
4702 time_t cutoff = get_seconds() - nn->nfsd4_lease;
4703 time_t t, new_timeo = nn->nfsd4_lease;
4705 dprintk("NFSD: laundromat service - starting\n");
4706 nfsd4_end_grace(nn);
4707 INIT_LIST_HEAD(&reaplist);
4708 spin_lock(&nn->client_lock);
4709 list_for_each_safe(pos, next, &nn->client_lru) {
4710 clp = list_entry(pos, struct nfs4_client, cl_lru);
4711 if (time_after((unsigned long)clp->cl_time, (unsigned long)cutoff)) {
4712 t = clp->cl_time - cutoff;
4713 new_timeo = min(new_timeo, t);
4716 if (mark_client_expired_locked(clp)) {
4717 dprintk("NFSD: client in use (clientid %08x)\n",
4718 clp->cl_clientid.cl_id);
4721 list_add(&clp->cl_lru, &reaplist);
4723 spin_unlock(&nn->client_lock);
4724 list_for_each_safe(pos, next, &reaplist) {
4725 clp = list_entry(pos, struct nfs4_client, cl_lru);
4726 dprintk("NFSD: purging unused client (clientid %08x)\n",
4727 clp->cl_clientid.cl_id);
4728 list_del_init(&clp->cl_lru);
4731 spin_lock(&state_lock);
4732 list_for_each_safe(pos, next, &nn->del_recall_lru) {
4733 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
4734 if (time_after((unsigned long)dp->dl_time, (unsigned long)cutoff)) {
4735 t = dp->dl_time - cutoff;
4736 new_timeo = min(new_timeo, t);
4739 WARN_ON(!unhash_delegation_locked(dp));
4740 list_add(&dp->dl_recall_lru, &reaplist);
4742 spin_unlock(&state_lock);
4743 while (!list_empty(&reaplist)) {
4744 dp = list_first_entry(&reaplist, struct nfs4_delegation,
4746 list_del_init(&dp->dl_recall_lru);
4747 revoke_delegation(dp);
4750 spin_lock(&nn->client_lock);
4751 while (!list_empty(&nn->close_lru)) {
4752 oo = list_first_entry(&nn->close_lru, struct nfs4_openowner,
4754 if (time_after((unsigned long)oo->oo_time,
4755 (unsigned long)cutoff)) {
4756 t = oo->oo_time - cutoff;
4757 new_timeo = min(new_timeo, t);
4760 list_del_init(&oo->oo_close_lru);
4761 stp = oo->oo_last_closed_stid;
4762 oo->oo_last_closed_stid = NULL;
4763 spin_unlock(&nn->client_lock);
4764 nfs4_put_stid(&stp->st_stid);
4765 spin_lock(&nn->client_lock);
4767 spin_unlock(&nn->client_lock);
4770 * It's possible for a client to try and acquire an already held lock
4771 * that is being held for a long time, and then lose interest in it.
4772 * So, we clean out any un-revisited request after a lease period
4773 * under the assumption that the client is no longer interested.
4775 * RFC5661, sec. 9.6 states that the client must not rely on getting
4776 * notifications and must continue to poll for locks, even when the
4777 * server supports them. Thus this shouldn't lead to clients blocking
4778 * indefinitely once the lock does become free.
4780 BUG_ON(!list_empty(&reaplist));
4781 spin_lock(&nn->blocked_locks_lock);
4782 while (!list_empty(&nn->blocked_locks_lru)) {
4783 nbl = list_first_entry(&nn->blocked_locks_lru,
4784 struct nfsd4_blocked_lock, nbl_lru);
4785 if (time_after((unsigned long)nbl->nbl_time,
4786 (unsigned long)cutoff)) {
4787 t = nbl->nbl_time - cutoff;
4788 new_timeo = min(new_timeo, t);
4791 list_move(&nbl->nbl_lru, &reaplist);
4792 list_del_init(&nbl->nbl_list);
4794 spin_unlock(&nn->blocked_locks_lock);
4796 while (!list_empty(&reaplist)) {
4797 nbl = list_first_entry(&reaplist,
4798 struct nfsd4_blocked_lock, nbl_lru);
4799 list_del_init(&nbl->nbl_lru);
4800 posix_unblock_lock(&nbl->nbl_lock);
4801 free_blocked_lock(nbl);
4804 new_timeo = max_t(time_t, new_timeo, NFSD_LAUNDROMAT_MINTIMEOUT);
4808 static struct workqueue_struct *laundry_wq;
4809 static void laundromat_main(struct work_struct *);
4812 laundromat_main(struct work_struct *laundry)
4815 struct delayed_work *dwork = to_delayed_work(laundry);
4816 struct nfsd_net *nn = container_of(dwork, struct nfsd_net,
4819 t = nfs4_laundromat(nn);
4820 dprintk("NFSD: laundromat_main - sleeping for %ld seconds\n", t);
4821 queue_delayed_work(laundry_wq, &nn->laundromat_work, t*HZ);
4824 static inline __be32 nfs4_check_fh(struct svc_fh *fhp, struct nfs4_stid *stp)
4826 if (!fh_match(&fhp->fh_handle, &stp->sc_file->fi_fhandle))
4827 return nfserr_bad_stateid;
4832 access_permit_read(struct nfs4_ol_stateid *stp)
4834 return test_access(NFS4_SHARE_ACCESS_READ, stp) ||
4835 test_access(NFS4_SHARE_ACCESS_BOTH, stp) ||
4836 test_access(NFS4_SHARE_ACCESS_WRITE, stp);
4840 access_permit_write(struct nfs4_ol_stateid *stp)
4842 return test_access(NFS4_SHARE_ACCESS_WRITE, stp) ||
4843 test_access(NFS4_SHARE_ACCESS_BOTH, stp);
4847 __be32 nfs4_check_openmode(struct nfs4_ol_stateid *stp, int flags)
4849 __be32 status = nfserr_openmode;
4851 /* For lock stateid's, we test the parent open, not the lock: */
4852 if (stp->st_openstp)
4853 stp = stp->st_openstp;
4854 if ((flags & WR_STATE) && !access_permit_write(stp))
4856 if ((flags & RD_STATE) && !access_permit_read(stp))
4863 static inline __be32
4864 check_special_stateids(struct net *net, svc_fh *current_fh, stateid_t *stateid, int flags)
4866 if (ONE_STATEID(stateid) && (flags & RD_STATE))
4868 else if (opens_in_grace(net)) {
4869 /* Answer in remaining cases depends on existence of
4870 * conflicting state; so we must wait out the grace period. */
4871 return nfserr_grace;
4872 } else if (flags & WR_STATE)
4873 return nfs4_share_conflict(current_fh,
4874 NFS4_SHARE_DENY_WRITE);
4875 else /* (flags & RD_STATE) && ZERO_STATEID(stateid) */
4876 return nfs4_share_conflict(current_fh,
4877 NFS4_SHARE_DENY_READ);
4881 * Allow READ/WRITE during grace period on recovered state only for files
4882 * that are not able to provide mandatory locking.
4885 grace_disallows_io(struct net *net, struct inode *inode)
4887 return opens_in_grace(net) && mandatory_lock(inode);
4890 static __be32 check_stateid_generation(stateid_t *in, stateid_t *ref, bool has_session)
4893 * When sessions are used the stateid generation number is ignored
4896 if (has_session && in->si_generation == 0)
4899 if (in->si_generation == ref->si_generation)
4902 /* If the client sends us a stateid from the future, it's buggy: */
4903 if (nfsd4_stateid_generation_after(in, ref))
4904 return nfserr_bad_stateid;
4906 * However, we could see a stateid from the past, even from a
4907 * non-buggy client. For example, if the client sends a lock
4908 * while some IO is outstanding, the lock may bump si_generation
4909 * while the IO is still in flight. The client could avoid that
4910 * situation by waiting for responses on all the IO requests,
4911 * but better performance may result in retrying IO that
4912 * receives an old_stateid error if requests are rarely
4913 * reordered in flight:
4915 return nfserr_old_stateid;
4918 static __be32 nfsd4_stid_check_stateid_generation(stateid_t *in, struct nfs4_stid *s, bool has_session)
4922 spin_lock(&s->sc_lock);
4923 ret = nfsd4_verify_open_stid(s);
4925 ret = check_stateid_generation(in, &s->sc_stateid, has_session);
4926 spin_unlock(&s->sc_lock);
4930 static __be32 nfsd4_check_openowner_confirmed(struct nfs4_ol_stateid *ols)
4932 if (ols->st_stateowner->so_is_open_owner &&
4933 !(openowner(ols->st_stateowner)->oo_flags & NFS4_OO_CONFIRMED))
4934 return nfserr_bad_stateid;
4938 static __be32 nfsd4_validate_stateid(struct nfs4_client *cl, stateid_t *stateid)
4940 struct nfs4_stid *s;
4941 __be32 status = nfserr_bad_stateid;
4943 if (ZERO_STATEID(stateid) || ONE_STATEID(stateid) ||
4944 CLOSE_STATEID(stateid))
4946 /* Client debugging aid. */
4947 if (!same_clid(&stateid->si_opaque.so_clid, &cl->cl_clientid)) {
4948 char addr_str[INET6_ADDRSTRLEN];
4949 rpc_ntop((struct sockaddr *)&cl->cl_addr, addr_str,
4951 pr_warn_ratelimited("NFSD: client %s testing state ID "
4952 "with incorrect client ID\n", addr_str);
4955 spin_lock(&cl->cl_lock);
4956 s = find_stateid_locked(cl, stateid);
4959 status = nfsd4_stid_check_stateid_generation(stateid, s, 1);
4962 switch (s->sc_type) {
4963 case NFS4_DELEG_STID:
4966 case NFS4_REVOKED_DELEG_STID:
4967 status = nfserr_deleg_revoked;
4969 case NFS4_OPEN_STID:
4970 case NFS4_LOCK_STID:
4971 status = nfsd4_check_openowner_confirmed(openlockstateid(s));
4974 printk("unknown stateid type %x\n", s->sc_type);
4976 case NFS4_CLOSED_STID:
4977 case NFS4_CLOSED_DELEG_STID:
4978 status = nfserr_bad_stateid;
4981 spin_unlock(&cl->cl_lock);
4986 nfsd4_lookup_stateid(struct nfsd4_compound_state *cstate,
4987 stateid_t *stateid, unsigned char typemask,
4988 struct nfs4_stid **s, struct nfsd_net *nn)
4991 bool return_revoked = false;
4994 * only return revoked delegations if explicitly asked.
4995 * otherwise we report revoked or bad_stateid status.
4997 if (typemask & NFS4_REVOKED_DELEG_STID)
4998 return_revoked = true;
4999 else if (typemask & NFS4_DELEG_STID)
5000 typemask |= NFS4_REVOKED_DELEG_STID;
5002 if (ZERO_STATEID(stateid) || ONE_STATEID(stateid) ||
5003 CLOSE_STATEID(stateid))
5004 return nfserr_bad_stateid;
5005 status = lookup_clientid(&stateid->si_opaque.so_clid, cstate, nn);
5006 if (status == nfserr_stale_clientid) {
5007 if (cstate->session)
5008 return nfserr_bad_stateid;
5009 return nfserr_stale_stateid;
5013 *s = find_stateid_by_type(cstate->clp, stateid, typemask);
5015 return nfserr_bad_stateid;
5016 if (((*s)->sc_type == NFS4_REVOKED_DELEG_STID) && !return_revoked) {
5018 if (cstate->minorversion)
5019 return nfserr_deleg_revoked;
5020 return nfserr_bad_stateid;
5025 static struct file *
5026 nfs4_find_file(struct nfs4_stid *s, int flags)
5031 switch (s->sc_type) {
5032 case NFS4_DELEG_STID:
5033 if (WARN_ON_ONCE(!s->sc_file->fi_deleg_file))
5035 return get_file(s->sc_file->fi_deleg_file);
5036 case NFS4_OPEN_STID:
5037 case NFS4_LOCK_STID:
5038 if (flags & RD_STATE)
5039 return find_readable_file(s->sc_file);
5041 return find_writeable_file(s->sc_file);
5049 nfs4_check_olstateid(struct svc_fh *fhp, struct nfs4_ol_stateid *ols, int flags)
5053 status = nfsd4_check_openowner_confirmed(ols);
5056 return nfs4_check_openmode(ols, flags);
5060 nfs4_check_file(struct svc_rqst *rqstp, struct svc_fh *fhp, struct nfs4_stid *s,
5061 struct file **filpp, bool *tmp_file, int flags)
5063 int acc = (flags & RD_STATE) ? NFSD_MAY_READ : NFSD_MAY_WRITE;
5067 file = nfs4_find_file(s, flags);
5069 status = nfsd_permission(rqstp, fhp->fh_export, fhp->fh_dentry,
5070 acc | NFSD_MAY_OWNER_OVERRIDE);
5078 status = nfsd_open(rqstp, fhp, S_IFREG, acc, filpp);
5090 * Checks for stateid operations
5093 nfs4_preprocess_stateid_op(struct svc_rqst *rqstp,
5094 struct nfsd4_compound_state *cstate, struct svc_fh *fhp,
5095 stateid_t *stateid, int flags, struct file **filpp, bool *tmp_file)
5097 struct inode *ino = d_inode(fhp->fh_dentry);
5098 struct net *net = SVC_NET(rqstp);
5099 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
5100 struct nfs4_stid *s = NULL;
5108 if (grace_disallows_io(net, ino))
5109 return nfserr_grace;
5111 if (ZERO_STATEID(stateid) || ONE_STATEID(stateid)) {
5112 status = check_special_stateids(net, fhp, stateid, flags);
5116 status = nfsd4_lookup_stateid(cstate, stateid,
5117 NFS4_DELEG_STID|NFS4_OPEN_STID|NFS4_LOCK_STID,
5121 status = nfsd4_stid_check_stateid_generation(stateid, s,
5122 nfsd4_has_session(cstate));
5126 switch (s->sc_type) {
5127 case NFS4_DELEG_STID:
5128 status = nfs4_check_delegmode(delegstateid(s), flags);
5130 case NFS4_OPEN_STID:
5131 case NFS4_LOCK_STID:
5132 status = nfs4_check_olstateid(fhp, openlockstateid(s), flags);
5135 status = nfserr_bad_stateid;
5140 status = nfs4_check_fh(fhp, s);
5143 if (!status && filpp)
5144 status = nfs4_check_file(rqstp, fhp, s, filpp, tmp_file, flags);
5152 * Test if the stateid is valid
5155 nfsd4_test_stateid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
5156 union nfsd4_op_u *u)
5158 struct nfsd4_test_stateid *test_stateid = &u->test_stateid;
5159 struct nfsd4_test_stateid_id *stateid;
5160 struct nfs4_client *cl = cstate->session->se_client;
5162 list_for_each_entry(stateid, &test_stateid->ts_stateid_list, ts_id_list)
5163 stateid->ts_id_status =
5164 nfsd4_validate_stateid(cl, &stateid->ts_id_stateid);
5170 nfsd4_free_lock_stateid(stateid_t *stateid, struct nfs4_stid *s)
5172 struct nfs4_ol_stateid *stp = openlockstateid(s);
5175 ret = nfsd4_lock_ol_stateid(stp);
5179 ret = check_stateid_generation(stateid, &s->sc_stateid, 1);
5183 ret = nfserr_locks_held;
5184 if (check_for_locks(stp->st_stid.sc_file,
5185 lockowner(stp->st_stateowner)))
5188 release_lock_stateid(stp);
5192 mutex_unlock(&stp->st_mutex);
5199 nfsd4_free_stateid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
5200 union nfsd4_op_u *u)
5202 struct nfsd4_free_stateid *free_stateid = &u->free_stateid;
5203 stateid_t *stateid = &free_stateid->fr_stateid;
5204 struct nfs4_stid *s;
5205 struct nfs4_delegation *dp;
5206 struct nfs4_client *cl = cstate->session->se_client;
5207 __be32 ret = nfserr_bad_stateid;
5209 spin_lock(&cl->cl_lock);
5210 s = find_stateid_locked(cl, stateid);
5213 spin_lock(&s->sc_lock);
5214 switch (s->sc_type) {
5215 case NFS4_DELEG_STID:
5216 ret = nfserr_locks_held;
5218 case NFS4_OPEN_STID:
5219 ret = check_stateid_generation(stateid, &s->sc_stateid, 1);
5222 ret = nfserr_locks_held;
5224 case NFS4_LOCK_STID:
5225 spin_unlock(&s->sc_lock);
5226 refcount_inc(&s->sc_count);
5227 spin_unlock(&cl->cl_lock);
5228 ret = nfsd4_free_lock_stateid(stateid, s);
5230 case NFS4_REVOKED_DELEG_STID:
5231 spin_unlock(&s->sc_lock);
5232 dp = delegstateid(s);
5233 list_del_init(&dp->dl_recall_lru);
5234 spin_unlock(&cl->cl_lock);
5238 /* Default falls through and returns nfserr_bad_stateid */
5240 spin_unlock(&s->sc_lock);
5242 spin_unlock(&cl->cl_lock);
5250 return (type == NFS4_READW_LT || type == NFS4_READ_LT) ?
5251 RD_STATE : WR_STATE;
5254 static __be32 nfs4_seqid_op_checks(struct nfsd4_compound_state *cstate, stateid_t *stateid, u32 seqid, struct nfs4_ol_stateid *stp)
5256 struct svc_fh *current_fh = &cstate->current_fh;
5257 struct nfs4_stateowner *sop = stp->st_stateowner;
5260 status = nfsd4_check_seqid(cstate, sop, seqid);
5263 status = nfsd4_lock_ol_stateid(stp);
5264 if (status != nfs_ok)
5266 status = check_stateid_generation(stateid, &stp->st_stid.sc_stateid, nfsd4_has_session(cstate));
5267 if (status == nfs_ok)
5268 status = nfs4_check_fh(current_fh, &stp->st_stid);
5269 if (status != nfs_ok)
5270 mutex_unlock(&stp->st_mutex);
5275 * Checks for sequence id mutating operations.
5278 nfs4_preprocess_seqid_op(struct nfsd4_compound_state *cstate, u32 seqid,
5279 stateid_t *stateid, char typemask,
5280 struct nfs4_ol_stateid **stpp,
5281 struct nfsd_net *nn)
5284 struct nfs4_stid *s;
5285 struct nfs4_ol_stateid *stp = NULL;
5287 dprintk("NFSD: %s: seqid=%d stateid = " STATEID_FMT "\n", __func__,
5288 seqid, STATEID_VAL(stateid));
5291 status = nfsd4_lookup_stateid(cstate, stateid, typemask, &s, nn);
5294 stp = openlockstateid(s);
5295 nfsd4_cstate_assign_replay(cstate, stp->st_stateowner);
5297 status = nfs4_seqid_op_checks(cstate, stateid, seqid, stp);
5301 nfs4_put_stid(&stp->st_stid);
5305 static __be32 nfs4_preprocess_confirmed_seqid_op(struct nfsd4_compound_state *cstate, u32 seqid,
5306 stateid_t *stateid, struct nfs4_ol_stateid **stpp, struct nfsd_net *nn)
5309 struct nfs4_openowner *oo;
5310 struct nfs4_ol_stateid *stp;
5312 status = nfs4_preprocess_seqid_op(cstate, seqid, stateid,
5313 NFS4_OPEN_STID, &stp, nn);
5316 oo = openowner(stp->st_stateowner);
5317 if (!(oo->oo_flags & NFS4_OO_CONFIRMED)) {
5318 mutex_unlock(&stp->st_mutex);
5319 nfs4_put_stid(&stp->st_stid);
5320 return nfserr_bad_stateid;
5327 nfsd4_open_confirm(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
5328 union nfsd4_op_u *u)
5330 struct nfsd4_open_confirm *oc = &u->open_confirm;
5332 struct nfs4_openowner *oo;
5333 struct nfs4_ol_stateid *stp;
5334 struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
5336 dprintk("NFSD: nfsd4_open_confirm on file %pd\n",
5337 cstate->current_fh.fh_dentry);
5339 status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0);
5343 status = nfs4_preprocess_seqid_op(cstate,
5344 oc->oc_seqid, &oc->oc_req_stateid,
5345 NFS4_OPEN_STID, &stp, nn);
5348 oo = openowner(stp->st_stateowner);
5349 status = nfserr_bad_stateid;
5350 if (oo->oo_flags & NFS4_OO_CONFIRMED) {
5351 mutex_unlock(&stp->st_mutex);
5354 oo->oo_flags |= NFS4_OO_CONFIRMED;
5355 nfs4_inc_and_copy_stateid(&oc->oc_resp_stateid, &stp->st_stid);
5356 mutex_unlock(&stp->st_mutex);
5357 dprintk("NFSD: %s: success, seqid=%d stateid=" STATEID_FMT "\n",
5358 __func__, oc->oc_seqid, STATEID_VAL(&stp->st_stid.sc_stateid));
5360 nfsd4_client_record_create(oo->oo_owner.so_client);
5363 nfs4_put_stid(&stp->st_stid);
5365 nfsd4_bump_seqid(cstate, status);
5369 static inline void nfs4_stateid_downgrade_bit(struct nfs4_ol_stateid *stp, u32 access)
5371 if (!test_access(access, stp))
5373 nfs4_file_put_access(stp->st_stid.sc_file, access);
5374 clear_access(access, stp);
5377 static inline void nfs4_stateid_downgrade(struct nfs4_ol_stateid *stp, u32 to_access)
5379 switch (to_access) {
5380 case NFS4_SHARE_ACCESS_READ:
5381 nfs4_stateid_downgrade_bit(stp, NFS4_SHARE_ACCESS_WRITE);
5382 nfs4_stateid_downgrade_bit(stp, NFS4_SHARE_ACCESS_BOTH);
5384 case NFS4_SHARE_ACCESS_WRITE:
5385 nfs4_stateid_downgrade_bit(stp, NFS4_SHARE_ACCESS_READ);
5386 nfs4_stateid_downgrade_bit(stp, NFS4_SHARE_ACCESS_BOTH);
5388 case NFS4_SHARE_ACCESS_BOTH:
5396 nfsd4_open_downgrade(struct svc_rqst *rqstp,
5397 struct nfsd4_compound_state *cstate, union nfsd4_op_u *u)
5399 struct nfsd4_open_downgrade *od = &u->open_downgrade;
5401 struct nfs4_ol_stateid *stp;
5402 struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
5404 dprintk("NFSD: nfsd4_open_downgrade on file %pd\n",
5405 cstate->current_fh.fh_dentry);
5407 /* We don't yet support WANT bits: */
5408 if (od->od_deleg_want)
5409 dprintk("NFSD: %s: od_deleg_want=0x%x ignored\n", __func__,
5412 status = nfs4_preprocess_confirmed_seqid_op(cstate, od->od_seqid,
5413 &od->od_stateid, &stp, nn);
5416 status = nfserr_inval;
5417 if (!test_access(od->od_share_access, stp)) {
5418 dprintk("NFSD: access not a subset of current bitmap: 0x%hhx, input access=%08x\n",
5419 stp->st_access_bmap, od->od_share_access);
5422 if (!test_deny(od->od_share_deny, stp)) {
5423 dprintk("NFSD: deny not a subset of current bitmap: 0x%hhx, input deny=%08x\n",
5424 stp->st_deny_bmap, od->od_share_deny);
5427 nfs4_stateid_downgrade(stp, od->od_share_access);
5428 reset_union_bmap_deny(od->od_share_deny, stp);
5429 nfs4_inc_and_copy_stateid(&od->od_stateid, &stp->st_stid);
5432 mutex_unlock(&stp->st_mutex);
5433 nfs4_put_stid(&stp->st_stid);
5435 nfsd4_bump_seqid(cstate, status);
5439 static void nfsd4_close_open_stateid(struct nfs4_ol_stateid *s)
5441 struct nfs4_client *clp = s->st_stid.sc_client;
5443 LIST_HEAD(reaplist);
5445 spin_lock(&clp->cl_lock);
5446 unhashed = unhash_open_stateid(s, &reaplist);
5448 if (clp->cl_minorversion) {
5450 put_ol_stateid_locked(s, &reaplist);
5451 spin_unlock(&clp->cl_lock);
5452 free_ol_stateid_reaplist(&reaplist);
5454 spin_unlock(&clp->cl_lock);
5455 free_ol_stateid_reaplist(&reaplist);
5457 move_to_close_lru(s, clp->net);
5462 * nfs4_unlock_state() called after encode
5465 nfsd4_close(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
5466 union nfsd4_op_u *u)
5468 struct nfsd4_close *close = &u->close;
5470 struct nfs4_ol_stateid *stp;
5471 struct net *net = SVC_NET(rqstp);
5472 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
5474 dprintk("NFSD: nfsd4_close on file %pd\n",
5475 cstate->current_fh.fh_dentry);
5477 status = nfs4_preprocess_seqid_op(cstate, close->cl_seqid,
5479 NFS4_OPEN_STID|NFS4_CLOSED_STID,
5481 nfsd4_bump_seqid(cstate, status);
5485 stp->st_stid.sc_type = NFS4_CLOSED_STID;
5488 * Technically we don't _really_ have to increment or copy it, since
5489 * it should just be gone after this operation and we clobber the
5490 * copied value below, but we continue to do so here just to ensure
5491 * that racing ops see that there was a state change.
5493 nfs4_inc_and_copy_stateid(&close->cl_stateid, &stp->st_stid);
5495 nfsd4_close_open_stateid(stp);
5496 mutex_unlock(&stp->st_mutex);
5498 /* v4.1+ suggests that we send a special stateid in here, since the
5499 * clients should just ignore this anyway. Since this is not useful
5500 * for v4.0 clients either, we set it to the special close_stateid
5503 * See RFC5661 section 18.2.4, and RFC7530 section 16.2.5
5505 memcpy(&close->cl_stateid, &close_stateid, sizeof(close->cl_stateid));
5507 /* put reference from nfs4_preprocess_seqid_op */
5508 nfs4_put_stid(&stp->st_stid);
5514 nfsd4_delegreturn(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
5515 union nfsd4_op_u *u)
5517 struct nfsd4_delegreturn *dr = &u->delegreturn;
5518 struct nfs4_delegation *dp;
5519 stateid_t *stateid = &dr->dr_stateid;
5520 struct nfs4_stid *s;
5522 struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
5524 if ((status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0)))
5527 status = nfsd4_lookup_stateid(cstate, stateid, NFS4_DELEG_STID, &s, nn);
5530 dp = delegstateid(s);
5531 status = nfsd4_stid_check_stateid_generation(stateid, &dp->dl_stid, nfsd4_has_session(cstate));
5535 destroy_delegation(dp);
5537 nfs4_put_stid(&dp->dl_stid);
5543 end_offset(u64 start, u64 len)
5548 return end >= start ? end: NFS4_MAX_UINT64;
5551 /* last octet in a range */
5553 last_byte_offset(u64 start, u64 len)
5559 return end > start ? end - 1: NFS4_MAX_UINT64;
5563 * TODO: Linux file offsets are _signed_ 64-bit quantities, which means that
5564 * we can't properly handle lock requests that go beyond the (2^63 - 1)-th
5565 * byte, because of sign extension problems. Since NFSv4 calls for 64-bit
5566 * locking, this prevents us from being completely protocol-compliant. The
5567 * real solution to this problem is to start using unsigned file offsets in
5568 * the VFS, but this is a very deep change!
5571 nfs4_transform_lock_offset(struct file_lock *lock)
5573 if (lock->fl_start < 0)
5574 lock->fl_start = OFFSET_MAX;
5575 if (lock->fl_end < 0)
5576 lock->fl_end = OFFSET_MAX;
5580 nfsd4_fl_get_owner(fl_owner_t owner)
5582 struct nfs4_lockowner *lo = (struct nfs4_lockowner *)owner;
5584 nfs4_get_stateowner(&lo->lo_owner);
5589 nfsd4_fl_put_owner(fl_owner_t owner)
5591 struct nfs4_lockowner *lo = (struct nfs4_lockowner *)owner;
5594 nfs4_put_stateowner(&lo->lo_owner);
5598 nfsd4_lm_notify(struct file_lock *fl)
5600 struct nfs4_lockowner *lo = (struct nfs4_lockowner *)fl->fl_owner;
5601 struct net *net = lo->lo_owner.so_client->net;
5602 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
5603 struct nfsd4_blocked_lock *nbl = container_of(fl,
5604 struct nfsd4_blocked_lock, nbl_lock);
5607 /* An empty list means that something else is going to be using it */
5608 spin_lock(&nn->blocked_locks_lock);
5609 if (!list_empty(&nbl->nbl_list)) {
5610 list_del_init(&nbl->nbl_list);
5611 list_del_init(&nbl->nbl_lru);
5614 spin_unlock(&nn->blocked_locks_lock);
5617 nfsd4_run_cb(&nbl->nbl_cb);
5620 static const struct lock_manager_operations nfsd_posix_mng_ops = {
5621 .lm_notify = nfsd4_lm_notify,
5622 .lm_get_owner = nfsd4_fl_get_owner,
5623 .lm_put_owner = nfsd4_fl_put_owner,
5627 nfs4_set_lock_denied(struct file_lock *fl, struct nfsd4_lock_denied *deny)
5629 struct nfs4_lockowner *lo;
5631 if (fl->fl_lmops == &nfsd_posix_mng_ops) {
5632 lo = (struct nfs4_lockowner *) fl->fl_owner;
5633 deny->ld_owner.data = kmemdup(lo->lo_owner.so_owner.data,
5634 lo->lo_owner.so_owner.len, GFP_KERNEL);
5635 if (!deny->ld_owner.data)
5636 /* We just don't care that much */
5638 deny->ld_owner.len = lo->lo_owner.so_owner.len;
5639 deny->ld_clientid = lo->lo_owner.so_client->cl_clientid;
5642 deny->ld_owner.len = 0;
5643 deny->ld_owner.data = NULL;
5644 deny->ld_clientid.cl_boot = 0;
5645 deny->ld_clientid.cl_id = 0;
5647 deny->ld_start = fl->fl_start;
5648 deny->ld_length = NFS4_MAX_UINT64;
5649 if (fl->fl_end != NFS4_MAX_UINT64)
5650 deny->ld_length = fl->fl_end - fl->fl_start + 1;
5651 deny->ld_type = NFS4_READ_LT;
5652 if (fl->fl_type != F_RDLCK)
5653 deny->ld_type = NFS4_WRITE_LT;
5656 static struct nfs4_lockowner *
5657 find_lockowner_str_locked(struct nfs4_client *clp, struct xdr_netobj *owner)
5659 unsigned int strhashval = ownerstr_hashval(owner);
5660 struct nfs4_stateowner *so;
5662 lockdep_assert_held(&clp->cl_lock);
5664 list_for_each_entry(so, &clp->cl_ownerstr_hashtbl[strhashval],
5666 if (so->so_is_open_owner)
5668 if (same_owner_str(so, owner))
5669 return lockowner(nfs4_get_stateowner(so));
5674 static struct nfs4_lockowner *
5675 find_lockowner_str(struct nfs4_client *clp, struct xdr_netobj *owner)
5677 struct nfs4_lockowner *lo;
5679 spin_lock(&clp->cl_lock);
5680 lo = find_lockowner_str_locked(clp, owner);
5681 spin_unlock(&clp->cl_lock);
5685 static void nfs4_unhash_lockowner(struct nfs4_stateowner *sop)
5687 unhash_lockowner_locked(lockowner(sop));
5690 static void nfs4_free_lockowner(struct nfs4_stateowner *sop)
5692 struct nfs4_lockowner *lo = lockowner(sop);
5694 kmem_cache_free(lockowner_slab, lo);
5697 static const struct nfs4_stateowner_operations lockowner_ops = {
5698 .so_unhash = nfs4_unhash_lockowner,
5699 .so_free = nfs4_free_lockowner,
5703 * Alloc a lock owner structure.
5704 * Called in nfsd4_lock - therefore, OPEN and OPEN_CONFIRM (if needed) has
5707 * strhashval = ownerstr_hashval
5709 static struct nfs4_lockowner *
5710 alloc_init_lock_stateowner(unsigned int strhashval, struct nfs4_client *clp,
5711 struct nfs4_ol_stateid *open_stp,
5712 struct nfsd4_lock *lock)
5714 struct nfs4_lockowner *lo, *ret;
5716 lo = alloc_stateowner(lockowner_slab, &lock->lk_new_owner, clp);
5719 INIT_LIST_HEAD(&lo->lo_blocked);
5720 INIT_LIST_HEAD(&lo->lo_owner.so_stateids);
5721 lo->lo_owner.so_is_open_owner = 0;
5722 lo->lo_owner.so_seqid = lock->lk_new_lock_seqid;
5723 lo->lo_owner.so_ops = &lockowner_ops;
5724 spin_lock(&clp->cl_lock);
5725 ret = find_lockowner_str_locked(clp, &lock->lk_new_owner);
5727 list_add(&lo->lo_owner.so_strhash,
5728 &clp->cl_ownerstr_hashtbl[strhashval]);
5731 nfs4_free_stateowner(&lo->lo_owner);
5733 spin_unlock(&clp->cl_lock);
5737 static struct nfs4_ol_stateid *
5738 find_lock_stateid(struct nfs4_lockowner *lo, struct nfs4_file *fp)
5740 struct nfs4_ol_stateid *lst;
5741 struct nfs4_client *clp = lo->lo_owner.so_client;
5743 lockdep_assert_held(&clp->cl_lock);
5745 list_for_each_entry(lst, &lo->lo_owner.so_stateids, st_perstateowner) {
5746 if (lst->st_stid.sc_type != NFS4_LOCK_STID)
5748 if (lst->st_stid.sc_file == fp) {
5749 refcount_inc(&lst->st_stid.sc_count);
5756 static struct nfs4_ol_stateid *
5757 init_lock_stateid(struct nfs4_ol_stateid *stp, struct nfs4_lockowner *lo,
5758 struct nfs4_file *fp, struct inode *inode,
5759 struct nfs4_ol_stateid *open_stp)
5761 struct nfs4_client *clp = lo->lo_owner.so_client;
5762 struct nfs4_ol_stateid *retstp;
5764 mutex_init(&stp->st_mutex);
5765 mutex_lock_nested(&stp->st_mutex, OPEN_STATEID_MUTEX);
5767 spin_lock(&clp->cl_lock);
5768 spin_lock(&fp->fi_lock);
5769 retstp = find_lock_stateid(lo, fp);
5773 refcount_inc(&stp->st_stid.sc_count);
5774 stp->st_stid.sc_type = NFS4_LOCK_STID;
5775 stp->st_stateowner = nfs4_get_stateowner(&lo->lo_owner);
5777 stp->st_stid.sc_file = fp;
5778 stp->st_access_bmap = 0;
5779 stp->st_deny_bmap = open_stp->st_deny_bmap;
5780 stp->st_openstp = open_stp;
5781 list_add(&stp->st_locks, &open_stp->st_locks);
5782 list_add(&stp->st_perstateowner, &lo->lo_owner.so_stateids);
5783 list_add(&stp->st_perfile, &fp->fi_stateids);
5785 spin_unlock(&fp->fi_lock);
5786 spin_unlock(&clp->cl_lock);
5788 if (nfsd4_lock_ol_stateid(retstp) != nfs_ok) {
5789 nfs4_put_stid(&retstp->st_stid);
5792 /* To keep mutex tracking happy */
5793 mutex_unlock(&stp->st_mutex);
5799 static struct nfs4_ol_stateid *
5800 find_or_create_lock_stateid(struct nfs4_lockowner *lo, struct nfs4_file *fi,
5801 struct inode *inode, struct nfs4_ol_stateid *ost,
5804 struct nfs4_stid *ns = NULL;
5805 struct nfs4_ol_stateid *lst;
5806 struct nfs4_openowner *oo = openowner(ost->st_stateowner);
5807 struct nfs4_client *clp = oo->oo_owner.so_client;
5810 spin_lock(&clp->cl_lock);
5811 lst = find_lock_stateid(lo, fi);
5812 spin_unlock(&clp->cl_lock);
5814 if (nfsd4_lock_ol_stateid(lst) == nfs_ok)
5816 nfs4_put_stid(&lst->st_stid);
5818 ns = nfs4_alloc_stid(clp, stateid_slab, nfs4_free_lock_stateid);
5822 lst = init_lock_stateid(openlockstateid(ns), lo, fi, inode, ost);
5823 if (lst == openlockstateid(ns))
5832 check_lock_length(u64 offset, u64 length)
5834 return ((length == 0) || ((length != NFS4_MAX_UINT64) &&
5835 (length > ~offset)));
5838 static void get_lock_access(struct nfs4_ol_stateid *lock_stp, u32 access)
5840 struct nfs4_file *fp = lock_stp->st_stid.sc_file;
5842 lockdep_assert_held(&fp->fi_lock);
5844 if (test_access(access, lock_stp))
5846 __nfs4_file_get_access(fp, access);
5847 set_access(access, lock_stp);
5851 lookup_or_create_lock_state(struct nfsd4_compound_state *cstate,
5852 struct nfs4_ol_stateid *ost,
5853 struct nfsd4_lock *lock,
5854 struct nfs4_ol_stateid **plst, bool *new)
5857 struct nfs4_file *fi = ost->st_stid.sc_file;
5858 struct nfs4_openowner *oo = openowner(ost->st_stateowner);
5859 struct nfs4_client *cl = oo->oo_owner.so_client;
5860 struct inode *inode = d_inode(cstate->current_fh.fh_dentry);
5861 struct nfs4_lockowner *lo;
5862 struct nfs4_ol_stateid *lst;
5863 unsigned int strhashval;
5865 lo = find_lockowner_str(cl, &lock->lk_new_owner);
5867 strhashval = ownerstr_hashval(&lock->lk_new_owner);
5868 lo = alloc_init_lock_stateowner(strhashval, cl, ost, lock);
5870 return nfserr_jukebox;
5872 /* with an existing lockowner, seqids must be the same */
5873 status = nfserr_bad_seqid;
5874 if (!cstate->minorversion &&
5875 lock->lk_new_lock_seqid != lo->lo_owner.so_seqid)
5879 lst = find_or_create_lock_stateid(lo, fi, inode, ost, new);
5881 status = nfserr_jukebox;
5888 nfs4_put_stateowner(&lo->lo_owner);
5896 nfsd4_lock(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
5897 union nfsd4_op_u *u)
5899 struct nfsd4_lock *lock = &u->lock;
5900 struct nfs4_openowner *open_sop = NULL;
5901 struct nfs4_lockowner *lock_sop = NULL;
5902 struct nfs4_ol_stateid *lock_stp = NULL;
5903 struct nfs4_ol_stateid *open_stp = NULL;
5904 struct nfs4_file *fp;
5905 struct file *filp = NULL;
5906 struct nfsd4_blocked_lock *nbl = NULL;
5907 struct file_lock *file_lock = NULL;
5908 struct file_lock *conflock = NULL;
5913 unsigned char fl_type;
5914 unsigned int fl_flags = FL_POSIX;
5915 struct net *net = SVC_NET(rqstp);
5916 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
5918 dprintk("NFSD: nfsd4_lock: start=%Ld length=%Ld\n",
5919 (long long) lock->lk_offset,
5920 (long long) lock->lk_length);
5922 if (check_lock_length(lock->lk_offset, lock->lk_length))
5923 return nfserr_inval;
5925 if ((status = fh_verify(rqstp, &cstate->current_fh,
5926 S_IFREG, NFSD_MAY_LOCK))) {
5927 dprintk("NFSD: nfsd4_lock: permission denied!\n");
5931 if (lock->lk_is_new) {
5932 if (nfsd4_has_session(cstate))
5933 /* See rfc 5661 18.10.3: given clientid is ignored: */
5934 memcpy(&lock->lk_new_clientid,
5935 &cstate->session->se_client->cl_clientid,
5936 sizeof(clientid_t));
5938 status = nfserr_stale_clientid;
5939 if (STALE_CLIENTID(&lock->lk_new_clientid, nn))
5942 /* validate and update open stateid and open seqid */
5943 status = nfs4_preprocess_confirmed_seqid_op(cstate,
5944 lock->lk_new_open_seqid,
5945 &lock->lk_new_open_stateid,
5949 mutex_unlock(&open_stp->st_mutex);
5950 open_sop = openowner(open_stp->st_stateowner);
5951 status = nfserr_bad_stateid;
5952 if (!same_clid(&open_sop->oo_owner.so_client->cl_clientid,
5953 &lock->lk_new_clientid))
5955 status = lookup_or_create_lock_state(cstate, open_stp, lock,
5958 status = nfs4_preprocess_seqid_op(cstate,
5959 lock->lk_old_lock_seqid,
5960 &lock->lk_old_lock_stateid,
5961 NFS4_LOCK_STID, &lock_stp, nn);
5965 lock_sop = lockowner(lock_stp->st_stateowner);
5967 lkflg = setlkflg(lock->lk_type);
5968 status = nfs4_check_openmode(lock_stp, lkflg);
5972 status = nfserr_grace;
5973 if (locks_in_grace(net) && !lock->lk_reclaim)
5975 status = nfserr_no_grace;
5976 if (!locks_in_grace(net) && lock->lk_reclaim)
5979 fp = lock_stp->st_stid.sc_file;
5980 switch (lock->lk_type) {
5982 if (nfsd4_has_session(cstate))
5983 fl_flags |= FL_SLEEP;
5986 spin_lock(&fp->fi_lock);
5987 filp = find_readable_file_locked(fp);
5989 get_lock_access(lock_stp, NFS4_SHARE_ACCESS_READ);
5990 spin_unlock(&fp->fi_lock);
5993 case NFS4_WRITEW_LT:
5994 if (nfsd4_has_session(cstate))
5995 fl_flags |= FL_SLEEP;
5998 spin_lock(&fp->fi_lock);
5999 filp = find_writeable_file_locked(fp);
6001 get_lock_access(lock_stp, NFS4_SHARE_ACCESS_WRITE);
6002 spin_unlock(&fp->fi_lock);
6006 status = nfserr_inval;
6011 status = nfserr_openmode;
6015 nbl = find_or_allocate_block(lock_sop, &fp->fi_fhandle, nn);
6017 dprintk("NFSD: %s: unable to allocate block!\n", __func__);
6018 status = nfserr_jukebox;
6022 file_lock = &nbl->nbl_lock;
6023 file_lock->fl_type = fl_type;
6024 file_lock->fl_owner = (fl_owner_t)lockowner(nfs4_get_stateowner(&lock_sop->lo_owner));
6025 file_lock->fl_pid = current->tgid;
6026 file_lock->fl_file = filp;
6027 file_lock->fl_flags = fl_flags;
6028 file_lock->fl_lmops = &nfsd_posix_mng_ops;
6029 file_lock->fl_start = lock->lk_offset;
6030 file_lock->fl_end = last_byte_offset(lock->lk_offset, lock->lk_length);
6031 nfs4_transform_lock_offset(file_lock);
6033 conflock = locks_alloc_lock();
6035 dprintk("NFSD: %s: unable to allocate lock!\n", __func__);
6036 status = nfserr_jukebox;
6040 if (fl_flags & FL_SLEEP) {
6041 nbl->nbl_time = jiffies;
6042 spin_lock(&nn->blocked_locks_lock);
6043 list_add_tail(&nbl->nbl_list, &lock_sop->lo_blocked);
6044 list_add_tail(&nbl->nbl_lru, &nn->blocked_locks_lru);
6045 spin_unlock(&nn->blocked_locks_lock);
6048 err = vfs_lock_file(filp, F_SETLK, file_lock, conflock);
6050 case 0: /* success! */
6051 nfs4_inc_and_copy_stateid(&lock->lk_resp_stateid, &lock_stp->st_stid);
6054 case FILE_LOCK_DEFERRED:
6057 case -EAGAIN: /* conflock holds conflicting lock */
6058 status = nfserr_denied;
6059 dprintk("NFSD: nfsd4_lock: conflicting lock found!\n");
6060 nfs4_set_lock_denied(conflock, &lock->lk_denied);
6063 status = nfserr_deadlock;
6066 dprintk("NFSD: nfsd4_lock: vfs_lock_file() failed! status %d\n",err);
6067 status = nfserrno(err);
6072 /* dequeue it if we queued it before */
6073 if (fl_flags & FL_SLEEP) {
6074 spin_lock(&nn->blocked_locks_lock);
6075 list_del_init(&nbl->nbl_list);
6076 list_del_init(&nbl->nbl_lru);
6077 spin_unlock(&nn->blocked_locks_lock);
6079 free_blocked_lock(nbl);
6084 /* Bump seqid manually if the 4.0 replay owner is openowner */
6085 if (cstate->replay_owner &&
6086 cstate->replay_owner != &lock_sop->lo_owner &&
6087 seqid_mutating_err(ntohl(status)))
6088 lock_sop->lo_owner.so_seqid++;
6091 * If this is a new, never-before-used stateid, and we are
6092 * returning an error, then just go ahead and release it.
6095 release_lock_stateid(lock_stp);
6097 mutex_unlock(&lock_stp->st_mutex);
6099 nfs4_put_stid(&lock_stp->st_stid);
6102 nfs4_put_stid(&open_stp->st_stid);
6103 nfsd4_bump_seqid(cstate, status);
6105 locks_free_lock(conflock);
6110 * The NFSv4 spec allows a client to do a LOCKT without holding an OPEN,
6111 * so we do a temporary open here just to get an open file to pass to
6112 * vfs_test_lock. (Arguably perhaps test_lock should be done with an
6115 static __be32 nfsd_test_lock(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file_lock *lock)
6118 __be32 err = nfsd_open(rqstp, fhp, S_IFREG, NFSD_MAY_READ, &file);
6120 err = nfserrno(vfs_test_lock(file, lock));
6130 nfsd4_lockt(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
6131 union nfsd4_op_u *u)
6133 struct nfsd4_lockt *lockt = &u->lockt;
6134 struct file_lock *file_lock = NULL;
6135 struct nfs4_lockowner *lo = NULL;
6137 struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
6139 if (locks_in_grace(SVC_NET(rqstp)))
6140 return nfserr_grace;
6142 if (check_lock_length(lockt->lt_offset, lockt->lt_length))
6143 return nfserr_inval;
6145 if (!nfsd4_has_session(cstate)) {
6146 status = lookup_clientid(&lockt->lt_clientid, cstate, nn);
6151 if ((status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0)))
6154 file_lock = locks_alloc_lock();
6156 dprintk("NFSD: %s: unable to allocate lock!\n", __func__);
6157 status = nfserr_jukebox;
6161 switch (lockt->lt_type) {
6164 file_lock->fl_type = F_RDLCK;
6167 case NFS4_WRITEW_LT:
6168 file_lock->fl_type = F_WRLCK;
6171 dprintk("NFSD: nfs4_lockt: bad lock type!\n");
6172 status = nfserr_inval;
6176 lo = find_lockowner_str(cstate->clp, &lockt->lt_owner);
6178 file_lock->fl_owner = (fl_owner_t)lo;
6179 file_lock->fl_pid = current->tgid;
6180 file_lock->fl_flags = FL_POSIX;
6182 file_lock->fl_start = lockt->lt_offset;
6183 file_lock->fl_end = last_byte_offset(lockt->lt_offset, lockt->lt_length);
6185 nfs4_transform_lock_offset(file_lock);
6187 status = nfsd_test_lock(rqstp, &cstate->current_fh, file_lock);
6191 if (file_lock->fl_type != F_UNLCK) {
6192 status = nfserr_denied;
6193 nfs4_set_lock_denied(file_lock, &lockt->lt_denied);
6197 nfs4_put_stateowner(&lo->lo_owner);
6199 locks_free_lock(file_lock);
6204 nfsd4_locku(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
6205 union nfsd4_op_u *u)
6207 struct nfsd4_locku *locku = &u->locku;
6208 struct nfs4_ol_stateid *stp;
6209 struct file *filp = NULL;
6210 struct file_lock *file_lock = NULL;
6213 struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
6215 dprintk("NFSD: nfsd4_locku: start=%Ld length=%Ld\n",
6216 (long long) locku->lu_offset,
6217 (long long) locku->lu_length);
6219 if (check_lock_length(locku->lu_offset, locku->lu_length))
6220 return nfserr_inval;
6222 status = nfs4_preprocess_seqid_op(cstate, locku->lu_seqid,
6223 &locku->lu_stateid, NFS4_LOCK_STID,
6227 filp = find_any_file(stp->st_stid.sc_file);
6229 status = nfserr_lock_range;
6232 file_lock = locks_alloc_lock();
6234 dprintk("NFSD: %s: unable to allocate lock!\n", __func__);
6235 status = nfserr_jukebox;
6239 file_lock->fl_type = F_UNLCK;
6240 file_lock->fl_owner = (fl_owner_t)lockowner(nfs4_get_stateowner(stp->st_stateowner));
6241 file_lock->fl_pid = current->tgid;
6242 file_lock->fl_file = filp;
6243 file_lock->fl_flags = FL_POSIX;
6244 file_lock->fl_lmops = &nfsd_posix_mng_ops;
6245 file_lock->fl_start = locku->lu_offset;
6247 file_lock->fl_end = last_byte_offset(locku->lu_offset,
6249 nfs4_transform_lock_offset(file_lock);
6251 err = vfs_lock_file(filp, F_SETLK, file_lock, NULL);
6253 dprintk("NFSD: nfs4_locku: vfs_lock_file failed!\n");
6256 nfs4_inc_and_copy_stateid(&locku->lu_stateid, &stp->st_stid);
6260 mutex_unlock(&stp->st_mutex);
6261 nfs4_put_stid(&stp->st_stid);
6263 nfsd4_bump_seqid(cstate, status);
6265 locks_free_lock(file_lock);
6269 status = nfserrno(err);
6275 * true: locks held by lockowner
6276 * false: no locks held by lockowner
6279 check_for_locks(struct nfs4_file *fp, struct nfs4_lockowner *lowner)
6281 struct file_lock *fl;
6283 struct file *filp = find_any_file(fp);
6284 struct inode *inode;
6285 struct file_lock_context *flctx;
6288 /* Any valid lock stateid should have some sort of access */
6293 inode = file_inode(filp);
6294 flctx = inode->i_flctx;
6296 if (flctx && !list_empty_careful(&flctx->flc_posix)) {
6297 spin_lock(&flctx->flc_lock);
6298 list_for_each_entry(fl, &flctx->flc_posix, fl_list) {
6299 if (fl->fl_owner == (fl_owner_t)lowner) {
6304 spin_unlock(&flctx->flc_lock);
6311 nfsd4_release_lockowner(struct svc_rqst *rqstp,
6312 struct nfsd4_compound_state *cstate,
6313 union nfsd4_op_u *u)
6315 struct nfsd4_release_lockowner *rlockowner = &u->release_lockowner;
6316 clientid_t *clid = &rlockowner->rl_clientid;
6317 struct nfs4_stateowner *sop;
6318 struct nfs4_lockowner *lo = NULL;
6319 struct nfs4_ol_stateid *stp;
6320 struct xdr_netobj *owner = &rlockowner->rl_owner;
6321 unsigned int hashval = ownerstr_hashval(owner);
6323 struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
6324 struct nfs4_client *clp;
6325 LIST_HEAD (reaplist);
6327 dprintk("nfsd4_release_lockowner clientid: (%08x/%08x):\n",
6328 clid->cl_boot, clid->cl_id);
6330 status = lookup_clientid(clid, cstate, nn);
6335 /* Find the matching lock stateowner */
6336 spin_lock(&clp->cl_lock);
6337 list_for_each_entry(sop, &clp->cl_ownerstr_hashtbl[hashval],
6340 if (sop->so_is_open_owner || !same_owner_str(sop, owner))
6343 /* see if there are still any locks associated with it */
6344 lo = lockowner(sop);
6345 list_for_each_entry(stp, &sop->so_stateids, st_perstateowner) {
6346 if (check_for_locks(stp->st_stid.sc_file, lo)) {
6347 status = nfserr_locks_held;
6348 spin_unlock(&clp->cl_lock);
6353 nfs4_get_stateowner(sop);
6357 spin_unlock(&clp->cl_lock);
6361 unhash_lockowner_locked(lo);
6362 while (!list_empty(&lo->lo_owner.so_stateids)) {
6363 stp = list_first_entry(&lo->lo_owner.so_stateids,
6364 struct nfs4_ol_stateid,
6366 WARN_ON(!unhash_lock_stateid(stp));
6367 put_ol_stateid_locked(stp, &reaplist);
6369 spin_unlock(&clp->cl_lock);
6370 free_ol_stateid_reaplist(&reaplist);
6371 remove_blocked_locks(lo);
6372 nfs4_put_stateowner(&lo->lo_owner);
6377 static inline struct nfs4_client_reclaim *
6380 return kmalloc(sizeof(struct nfs4_client_reclaim), GFP_KERNEL);
6384 nfs4_has_reclaimed_state(const char *name, struct nfsd_net *nn)
6386 struct nfs4_client_reclaim *crp;
6388 crp = nfsd4_find_reclaim_client(name, nn);
6389 return (crp && crp->cr_clp);
6393 * failure => all reset bets are off, nfserr_no_grace...
6395 struct nfs4_client_reclaim *
6396 nfs4_client_to_reclaim(const char *name, struct nfsd_net *nn)
6398 unsigned int strhashval;
6399 struct nfs4_client_reclaim *crp;
6401 dprintk("NFSD nfs4_client_to_reclaim NAME: %.*s\n", HEXDIR_LEN, name);
6402 crp = alloc_reclaim();
6404 strhashval = clientstr_hashval(name);
6405 INIT_LIST_HEAD(&crp->cr_strhash);
6406 list_add(&crp->cr_strhash, &nn->reclaim_str_hashtbl[strhashval]);
6407 memcpy(crp->cr_recdir, name, HEXDIR_LEN);
6409 nn->reclaim_str_hashtbl_size++;
6415 nfs4_remove_reclaim_record(struct nfs4_client_reclaim *crp, struct nfsd_net *nn)
6417 list_del(&crp->cr_strhash);
6419 nn->reclaim_str_hashtbl_size--;
6423 nfs4_release_reclaim(struct nfsd_net *nn)
6425 struct nfs4_client_reclaim *crp = NULL;
6428 for (i = 0; i < CLIENT_HASH_SIZE; i++) {
6429 while (!list_empty(&nn->reclaim_str_hashtbl[i])) {
6430 crp = list_entry(nn->reclaim_str_hashtbl[i].next,
6431 struct nfs4_client_reclaim, cr_strhash);
6432 nfs4_remove_reclaim_record(crp, nn);
6435 WARN_ON_ONCE(nn->reclaim_str_hashtbl_size);
6439 * called from OPEN, CLAIM_PREVIOUS with a new clientid. */
6440 struct nfs4_client_reclaim *
6441 nfsd4_find_reclaim_client(const char *recdir, struct nfsd_net *nn)
6443 unsigned int strhashval;
6444 struct nfs4_client_reclaim *crp = NULL;
6446 dprintk("NFSD: nfs4_find_reclaim_client for recdir %s\n", recdir);
6448 strhashval = clientstr_hashval(recdir);
6449 list_for_each_entry(crp, &nn->reclaim_str_hashtbl[strhashval], cr_strhash) {
6450 if (same_name(crp->cr_recdir, recdir)) {
6458 * Called from OPEN. Look for clientid in reclaim list.
6461 nfs4_check_open_reclaim(clientid_t *clid,
6462 struct nfsd4_compound_state *cstate,
6463 struct nfsd_net *nn)
6467 /* find clientid in conf_id_hashtbl */
6468 status = lookup_clientid(clid, cstate, nn);
6470 return nfserr_reclaim_bad;
6472 if (test_bit(NFSD4_CLIENT_RECLAIM_COMPLETE, &cstate->clp->cl_flags))
6473 return nfserr_no_grace;
6475 if (nfsd4_client_record_check(cstate->clp))
6476 return nfserr_reclaim_bad;
6481 #ifdef CONFIG_NFSD_FAULT_INJECTION
6483 put_client(struct nfs4_client *clp)
6485 atomic_dec(&clp->cl_refcount);
6488 static struct nfs4_client *
6489 nfsd_find_client(struct sockaddr_storage *addr, size_t addr_size)
6491 struct nfs4_client *clp;
6492 struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6495 if (!nfsd_netns_ready(nn))
6498 list_for_each_entry(clp, &nn->client_lru, cl_lru) {
6499 if (memcmp(&clp->cl_addr, addr, addr_size) == 0)
6506 nfsd_inject_print_clients(void)
6508 struct nfs4_client *clp;
6510 struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6512 char buf[INET6_ADDRSTRLEN];
6514 if (!nfsd_netns_ready(nn))
6517 spin_lock(&nn->client_lock);
6518 list_for_each_entry(clp, &nn->client_lru, cl_lru) {
6519 rpc_ntop((struct sockaddr *)&clp->cl_addr, buf, sizeof(buf));
6520 pr_info("NFS Client: %s\n", buf);
6523 spin_unlock(&nn->client_lock);
6529 nfsd_inject_forget_client(struct sockaddr_storage *addr, size_t addr_size)
6532 struct nfs4_client *clp;
6533 struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6536 if (!nfsd_netns_ready(nn))
6539 spin_lock(&nn->client_lock);
6540 clp = nfsd_find_client(addr, addr_size);
6542 if (mark_client_expired_locked(clp) == nfs_ok)
6547 spin_unlock(&nn->client_lock);
6556 nfsd_inject_forget_clients(u64 max)
6559 struct nfs4_client *clp, *next;
6560 struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6562 LIST_HEAD(reaplist);
6564 if (!nfsd_netns_ready(nn))
6567 spin_lock(&nn->client_lock);
6568 list_for_each_entry_safe(clp, next, &nn->client_lru, cl_lru) {
6569 if (mark_client_expired_locked(clp) == nfs_ok) {
6570 list_add(&clp->cl_lru, &reaplist);
6571 if (max != 0 && ++count >= max)
6575 spin_unlock(&nn->client_lock);
6577 list_for_each_entry_safe(clp, next, &reaplist, cl_lru)
6583 static void nfsd_print_count(struct nfs4_client *clp, unsigned int count,
6586 char buf[INET6_ADDRSTRLEN];
6587 rpc_ntop((struct sockaddr *)&clp->cl_addr, buf, sizeof(buf));
6588 printk(KERN_INFO "NFS Client: %s has %u %s\n", buf, count, type);
6592 nfsd_inject_add_lock_to_list(struct nfs4_ol_stateid *lst,
6593 struct list_head *collect)
6595 struct nfs4_client *clp = lst->st_stid.sc_client;
6596 struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6602 lockdep_assert_held(&nn->client_lock);
6603 atomic_inc(&clp->cl_refcount);
6604 list_add(&lst->st_locks, collect);
6607 static u64 nfsd_foreach_client_lock(struct nfs4_client *clp, u64 max,
6608 struct list_head *collect,
6609 bool (*func)(struct nfs4_ol_stateid *))
6611 struct nfs4_openowner *oop;
6612 struct nfs4_ol_stateid *stp, *st_next;
6613 struct nfs4_ol_stateid *lst, *lst_next;
6616 spin_lock(&clp->cl_lock);
6617 list_for_each_entry(oop, &clp->cl_openowners, oo_perclient) {
6618 list_for_each_entry_safe(stp, st_next,
6619 &oop->oo_owner.so_stateids, st_perstateowner) {
6620 list_for_each_entry_safe(lst, lst_next,
6621 &stp->st_locks, st_locks) {
6624 nfsd_inject_add_lock_to_list(lst,
6629 * Despite the fact that these functions deal
6630 * with 64-bit integers for "count", we must
6631 * ensure that it doesn't blow up the
6632 * clp->cl_refcount. Throw a warning if we
6633 * start to approach INT_MAX here.
6635 WARN_ON_ONCE(count == (INT_MAX / 2));
6642 spin_unlock(&clp->cl_lock);
6648 nfsd_collect_client_locks(struct nfs4_client *clp, struct list_head *collect,
6651 return nfsd_foreach_client_lock(clp, max, collect, unhash_lock_stateid);
6655 nfsd_print_client_locks(struct nfs4_client *clp)
6657 u64 count = nfsd_foreach_client_lock(clp, 0, NULL, NULL);
6658 nfsd_print_count(clp, count, "locked files");
6663 nfsd_inject_print_locks(void)
6665 struct nfs4_client *clp;
6667 struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6670 if (!nfsd_netns_ready(nn))
6673 spin_lock(&nn->client_lock);
6674 list_for_each_entry(clp, &nn->client_lru, cl_lru)
6675 count += nfsd_print_client_locks(clp);
6676 spin_unlock(&nn->client_lock);
6682 nfsd_reap_locks(struct list_head *reaplist)
6684 struct nfs4_client *clp;
6685 struct nfs4_ol_stateid *stp, *next;
6687 list_for_each_entry_safe(stp, next, reaplist, st_locks) {
6688 list_del_init(&stp->st_locks);
6689 clp = stp->st_stid.sc_client;
6690 nfs4_put_stid(&stp->st_stid);
6696 nfsd_inject_forget_client_locks(struct sockaddr_storage *addr, size_t addr_size)
6698 unsigned int count = 0;
6699 struct nfs4_client *clp;
6700 struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6702 LIST_HEAD(reaplist);
6704 if (!nfsd_netns_ready(nn))
6707 spin_lock(&nn->client_lock);
6708 clp = nfsd_find_client(addr, addr_size);
6710 count = nfsd_collect_client_locks(clp, &reaplist, 0);
6711 spin_unlock(&nn->client_lock);
6712 nfsd_reap_locks(&reaplist);
6717 nfsd_inject_forget_locks(u64 max)
6720 struct nfs4_client *clp;
6721 struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6723 LIST_HEAD(reaplist);
6725 if (!nfsd_netns_ready(nn))
6728 spin_lock(&nn->client_lock);
6729 list_for_each_entry(clp, &nn->client_lru, cl_lru) {
6730 count += nfsd_collect_client_locks(clp, &reaplist, max - count);
6731 if (max != 0 && count >= max)
6734 spin_unlock(&nn->client_lock);
6735 nfsd_reap_locks(&reaplist);
6740 nfsd_foreach_client_openowner(struct nfs4_client *clp, u64 max,
6741 struct list_head *collect,
6742 void (*func)(struct nfs4_openowner *))
6744 struct nfs4_openowner *oop, *next;
6745 struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6749 lockdep_assert_held(&nn->client_lock);
6751 spin_lock(&clp->cl_lock);
6752 list_for_each_entry_safe(oop, next, &clp->cl_openowners, oo_perclient) {
6756 atomic_inc(&clp->cl_refcount);
6757 list_add(&oop->oo_perclient, collect);
6762 * Despite the fact that these functions deal with
6763 * 64-bit integers for "count", we must ensure that
6764 * it doesn't blow up the clp->cl_refcount. Throw a
6765 * warning if we start to approach INT_MAX here.
6767 WARN_ON_ONCE(count == (INT_MAX / 2));
6771 spin_unlock(&clp->cl_lock);
6777 nfsd_print_client_openowners(struct nfs4_client *clp)
6779 u64 count = nfsd_foreach_client_openowner(clp, 0, NULL, NULL);
6781 nfsd_print_count(clp, count, "openowners");
6786 nfsd_collect_client_openowners(struct nfs4_client *clp,
6787 struct list_head *collect, u64 max)
6789 return nfsd_foreach_client_openowner(clp, max, collect,
6790 unhash_openowner_locked);
6794 nfsd_inject_print_openowners(void)
6796 struct nfs4_client *clp;
6798 struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6801 if (!nfsd_netns_ready(nn))
6804 spin_lock(&nn->client_lock);
6805 list_for_each_entry(clp, &nn->client_lru, cl_lru)
6806 count += nfsd_print_client_openowners(clp);
6807 spin_unlock(&nn->client_lock);
6813 nfsd_reap_openowners(struct list_head *reaplist)
6815 struct nfs4_client *clp;
6816 struct nfs4_openowner *oop, *next;
6818 list_for_each_entry_safe(oop, next, reaplist, oo_perclient) {
6819 list_del_init(&oop->oo_perclient);
6820 clp = oop->oo_owner.so_client;
6821 release_openowner(oop);
6827 nfsd_inject_forget_client_openowners(struct sockaddr_storage *addr,
6830 unsigned int count = 0;
6831 struct nfs4_client *clp;
6832 struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6834 LIST_HEAD(reaplist);
6836 if (!nfsd_netns_ready(nn))
6839 spin_lock(&nn->client_lock);
6840 clp = nfsd_find_client(addr, addr_size);
6842 count = nfsd_collect_client_openowners(clp, &reaplist, 0);
6843 spin_unlock(&nn->client_lock);
6844 nfsd_reap_openowners(&reaplist);
6849 nfsd_inject_forget_openowners(u64 max)
6852 struct nfs4_client *clp;
6853 struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6855 LIST_HEAD(reaplist);
6857 if (!nfsd_netns_ready(nn))
6860 spin_lock(&nn->client_lock);
6861 list_for_each_entry(clp, &nn->client_lru, cl_lru) {
6862 count += nfsd_collect_client_openowners(clp, &reaplist,
6864 if (max != 0 && count >= max)
6867 spin_unlock(&nn->client_lock);
6868 nfsd_reap_openowners(&reaplist);
6872 static u64 nfsd_find_all_delegations(struct nfs4_client *clp, u64 max,
6873 struct list_head *victims)
6875 struct nfs4_delegation *dp, *next;
6876 struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6880 lockdep_assert_held(&nn->client_lock);
6882 spin_lock(&state_lock);
6883 list_for_each_entry_safe(dp, next, &clp->cl_delegations, dl_perclnt) {
6886 * It's not safe to mess with delegations that have a
6887 * non-zero dl_time. They might have already been broken
6888 * and could be processed by the laundromat outside of
6889 * the state_lock. Just leave them be.
6891 if (dp->dl_time != 0)
6894 atomic_inc(&clp->cl_refcount);
6895 WARN_ON(!unhash_delegation_locked(dp));
6896 list_add(&dp->dl_recall_lru, victims);
6900 * Despite the fact that these functions deal with
6901 * 64-bit integers for "count", we must ensure that
6902 * it doesn't blow up the clp->cl_refcount. Throw a
6903 * warning if we start to approach INT_MAX here.
6905 WARN_ON_ONCE(count == (INT_MAX / 2));
6909 spin_unlock(&state_lock);
6914 nfsd_print_client_delegations(struct nfs4_client *clp)
6916 u64 count = nfsd_find_all_delegations(clp, 0, NULL);
6918 nfsd_print_count(clp, count, "delegations");
6923 nfsd_inject_print_delegations(void)
6925 struct nfs4_client *clp;
6927 struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6930 if (!nfsd_netns_ready(nn))
6933 spin_lock(&nn->client_lock);
6934 list_for_each_entry(clp, &nn->client_lru, cl_lru)
6935 count += nfsd_print_client_delegations(clp);
6936 spin_unlock(&nn->client_lock);
6942 nfsd_forget_delegations(struct list_head *reaplist)
6944 struct nfs4_client *clp;
6945 struct nfs4_delegation *dp, *next;
6947 list_for_each_entry_safe(dp, next, reaplist, dl_recall_lru) {
6948 list_del_init(&dp->dl_recall_lru);
6949 clp = dp->dl_stid.sc_client;
6950 revoke_delegation(dp);
6956 nfsd_inject_forget_client_delegations(struct sockaddr_storage *addr,
6960 struct nfs4_client *clp;
6961 struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6963 LIST_HEAD(reaplist);
6965 if (!nfsd_netns_ready(nn))
6968 spin_lock(&nn->client_lock);
6969 clp = nfsd_find_client(addr, addr_size);
6971 count = nfsd_find_all_delegations(clp, 0, &reaplist);
6972 spin_unlock(&nn->client_lock);
6974 nfsd_forget_delegations(&reaplist);
6979 nfsd_inject_forget_delegations(u64 max)
6982 struct nfs4_client *clp;
6983 struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6985 LIST_HEAD(reaplist);
6987 if (!nfsd_netns_ready(nn))
6990 spin_lock(&nn->client_lock);
6991 list_for_each_entry(clp, &nn->client_lru, cl_lru) {
6992 count += nfsd_find_all_delegations(clp, max - count, &reaplist);
6993 if (max != 0 && count >= max)
6996 spin_unlock(&nn->client_lock);
6997 nfsd_forget_delegations(&reaplist);
7002 nfsd_recall_delegations(struct list_head *reaplist)
7004 struct nfs4_client *clp;
7005 struct nfs4_delegation *dp, *next;
7007 list_for_each_entry_safe(dp, next, reaplist, dl_recall_lru) {
7008 list_del_init(&dp->dl_recall_lru);
7009 clp = dp->dl_stid.sc_client;
7011 * We skipped all entries that had a zero dl_time before,
7012 * so we can now reset the dl_time back to 0. If a delegation
7013 * break comes in now, then it won't make any difference since
7014 * we're recalling it either way.
7016 spin_lock(&state_lock);
7018 spin_unlock(&state_lock);
7019 nfsd_break_one_deleg(dp);
7025 nfsd_inject_recall_client_delegations(struct sockaddr_storage *addr,
7029 struct nfs4_client *clp;
7030 struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
7032 LIST_HEAD(reaplist);
7034 if (!nfsd_netns_ready(nn))
7037 spin_lock(&nn->client_lock);
7038 clp = nfsd_find_client(addr, addr_size);
7040 count = nfsd_find_all_delegations(clp, 0, &reaplist);
7041 spin_unlock(&nn->client_lock);
7043 nfsd_recall_delegations(&reaplist);
7048 nfsd_inject_recall_delegations(u64 max)
7051 struct nfs4_client *clp, *next;
7052 struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
7054 LIST_HEAD(reaplist);
7056 if (!nfsd_netns_ready(nn))
7059 spin_lock(&nn->client_lock);
7060 list_for_each_entry_safe(clp, next, &nn->client_lru, cl_lru) {
7061 count += nfsd_find_all_delegations(clp, max - count, &reaplist);
7062 if (max != 0 && ++count >= max)
7065 spin_unlock(&nn->client_lock);
7066 nfsd_recall_delegations(&reaplist);
7069 #endif /* CONFIG_NFSD_FAULT_INJECTION */
7072 * Since the lifetime of a delegation isn't limited to that of an open, a
7073 * client may quite reasonably hang on to a delegation as long as it has
7074 * the inode cached. This becomes an obvious problem the first time a
7075 * client's inode cache approaches the size of the server's total memory.
7077 * For now we avoid this problem by imposing a hard limit on the number
7078 * of delegations, which varies according to the server's memory size.
7081 set_max_delegations(void)
7084 * Allow at most 4 delegations per megabyte of RAM. Quick
7085 * estimates suggest that in the worst case (where every delegation
7086 * is for a different inode), a delegation could take about 1.5K,
7087 * giving a worst case usage of about 6% of memory.
7089 max_delegations = nr_free_buffer_pages() >> (20 - 2 - PAGE_SHIFT);
7092 static int nfs4_state_create_net(struct net *net)
7094 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
7097 nn->conf_id_hashtbl = kmalloc_array(CLIENT_HASH_SIZE,
7098 sizeof(struct list_head),
7100 if (!nn->conf_id_hashtbl)
7102 nn->unconf_id_hashtbl = kmalloc_array(CLIENT_HASH_SIZE,
7103 sizeof(struct list_head),
7105 if (!nn->unconf_id_hashtbl)
7107 nn->sessionid_hashtbl = kmalloc_array(SESSION_HASH_SIZE,
7108 sizeof(struct list_head),
7110 if (!nn->sessionid_hashtbl)
7113 for (i = 0; i < CLIENT_HASH_SIZE; i++) {
7114 INIT_LIST_HEAD(&nn->conf_id_hashtbl[i]);
7115 INIT_LIST_HEAD(&nn->unconf_id_hashtbl[i]);
7117 for (i = 0; i < SESSION_HASH_SIZE; i++)
7118 INIT_LIST_HEAD(&nn->sessionid_hashtbl[i]);
7119 nn->conf_name_tree = RB_ROOT;
7120 nn->unconf_name_tree = RB_ROOT;
7121 nn->boot_time = get_seconds();
7122 nn->grace_ended = false;
7123 nn->nfsd4_manager.block_opens = true;
7124 INIT_LIST_HEAD(&nn->nfsd4_manager.list);
7125 INIT_LIST_HEAD(&nn->client_lru);
7126 INIT_LIST_HEAD(&nn->close_lru);
7127 INIT_LIST_HEAD(&nn->del_recall_lru);
7128 spin_lock_init(&nn->client_lock);
7130 spin_lock_init(&nn->blocked_locks_lock);
7131 INIT_LIST_HEAD(&nn->blocked_locks_lru);
7133 INIT_DELAYED_WORK(&nn->laundromat_work, laundromat_main);
7139 kfree(nn->unconf_id_hashtbl);
7141 kfree(nn->conf_id_hashtbl);
7147 nfs4_state_destroy_net(struct net *net)
7150 struct nfs4_client *clp = NULL;
7151 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
7153 for (i = 0; i < CLIENT_HASH_SIZE; i++) {
7154 while (!list_empty(&nn->conf_id_hashtbl[i])) {
7155 clp = list_entry(nn->conf_id_hashtbl[i].next, struct nfs4_client, cl_idhash);
7156 destroy_client(clp);
7160 WARN_ON(!list_empty(&nn->blocked_locks_lru));
7162 for (i = 0; i < CLIENT_HASH_SIZE; i++) {
7163 while (!list_empty(&nn->unconf_id_hashtbl[i])) {
7164 clp = list_entry(nn->unconf_id_hashtbl[i].next, struct nfs4_client, cl_idhash);
7165 destroy_client(clp);
7169 kfree(nn->sessionid_hashtbl);
7170 kfree(nn->unconf_id_hashtbl);
7171 kfree(nn->conf_id_hashtbl);
7176 nfs4_state_start_net(struct net *net)
7178 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
7181 ret = nfs4_state_create_net(net);
7184 locks_start_grace(net, &nn->nfsd4_manager);
7185 nfsd4_client_tracking_init(net);
7186 printk(KERN_INFO "NFSD: starting %ld-second grace period (net %x)\n",
7187 nn->nfsd4_grace, net->ns.inum);
7188 queue_delayed_work(laundry_wq, &nn->laundromat_work, nn->nfsd4_grace * HZ);
7192 /* initialization to perform when the nfsd service is started: */
7195 nfs4_state_start(void)
7199 ret = set_callback_cred();
7203 laundry_wq = alloc_workqueue("%s", WQ_UNBOUND, 0, "nfsd4");
7204 if (laundry_wq == NULL) {
7206 goto out_cleanup_cred;
7208 ret = nfsd4_create_callback_queue();
7210 goto out_free_laundry;
7212 set_max_delegations();
7216 destroy_workqueue(laundry_wq);
7218 cleanup_callback_cred();
7223 nfs4_state_shutdown_net(struct net *net)
7225 struct nfs4_delegation *dp = NULL;
7226 struct list_head *pos, *next, reaplist;
7227 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
7229 cancel_delayed_work_sync(&nn->laundromat_work);
7230 locks_end_grace(&nn->nfsd4_manager);
7232 INIT_LIST_HEAD(&reaplist);
7233 spin_lock(&state_lock);
7234 list_for_each_safe(pos, next, &nn->del_recall_lru) {
7235 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
7236 WARN_ON(!unhash_delegation_locked(dp));
7237 list_add(&dp->dl_recall_lru, &reaplist);
7239 spin_unlock(&state_lock);
7240 list_for_each_safe(pos, next, &reaplist) {
7241 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
7242 list_del_init(&dp->dl_recall_lru);
7243 destroy_unhashed_deleg(dp);
7246 nfsd4_client_tracking_exit(net);
7247 nfs4_state_destroy_net(net);
7251 nfs4_state_shutdown(void)
7253 destroy_workqueue(laundry_wq);
7254 nfsd4_destroy_callback_queue();
7255 cleanup_callback_cred();
7259 get_stateid(struct nfsd4_compound_state *cstate, stateid_t *stateid)
7261 if (HAS_STATE_ID(cstate, CURRENT_STATE_ID_FLAG) && CURRENT_STATEID(stateid))
7262 memcpy(stateid, &cstate->current_stateid, sizeof(stateid_t));
7266 put_stateid(struct nfsd4_compound_state *cstate, stateid_t *stateid)
7268 if (cstate->minorversion) {
7269 memcpy(&cstate->current_stateid, stateid, sizeof(stateid_t));
7270 SET_STATE_ID(cstate, CURRENT_STATE_ID_FLAG);
7275 clear_current_stateid(struct nfsd4_compound_state *cstate)
7277 CLEAR_STATE_ID(cstate, CURRENT_STATE_ID_FLAG);
7281 * functions to set current state id
7284 nfsd4_set_opendowngradestateid(struct nfsd4_compound_state *cstate,
7285 union nfsd4_op_u *u)
7287 put_stateid(cstate, &u->open_downgrade.od_stateid);
7291 nfsd4_set_openstateid(struct nfsd4_compound_state *cstate,
7292 union nfsd4_op_u *u)
7294 put_stateid(cstate, &u->open.op_stateid);
7298 nfsd4_set_closestateid(struct nfsd4_compound_state *cstate,
7299 union nfsd4_op_u *u)
7301 put_stateid(cstate, &u->close.cl_stateid);
7305 nfsd4_set_lockstateid(struct nfsd4_compound_state *cstate,
7306 union nfsd4_op_u *u)
7308 put_stateid(cstate, &u->lock.lk_resp_stateid);
7312 * functions to consume current state id
7316 nfsd4_get_opendowngradestateid(struct nfsd4_compound_state *cstate,
7317 union nfsd4_op_u *u)
7319 get_stateid(cstate, &u->open_downgrade.od_stateid);
7323 nfsd4_get_delegreturnstateid(struct nfsd4_compound_state *cstate,
7324 union nfsd4_op_u *u)
7326 get_stateid(cstate, &u->delegreturn.dr_stateid);
7330 nfsd4_get_freestateid(struct nfsd4_compound_state *cstate,
7331 union nfsd4_op_u *u)
7333 get_stateid(cstate, &u->free_stateid.fr_stateid);
7337 nfsd4_get_setattrstateid(struct nfsd4_compound_state *cstate,
7338 union nfsd4_op_u *u)
7340 get_stateid(cstate, &u->setattr.sa_stateid);
7344 nfsd4_get_closestateid(struct nfsd4_compound_state *cstate,
7345 union nfsd4_op_u *u)
7347 get_stateid(cstate, &u->close.cl_stateid);
7351 nfsd4_get_lockustateid(struct nfsd4_compound_state *cstate,
7352 union nfsd4_op_u *u)
7354 get_stateid(cstate, &u->locku.lu_stateid);
7358 nfsd4_get_readstateid(struct nfsd4_compound_state *cstate,
7359 union nfsd4_op_u *u)
7361 get_stateid(cstate, &u->read.rd_stateid);
7365 nfsd4_get_writestateid(struct nfsd4_compound_state *cstate,
7366 union nfsd4_op_u *u)
7368 get_stateid(cstate, &u->write.wr_stateid);