2 * Copyright (c) 2001 The Regents of the University of Michigan.
5 * Kendrick Smith <kmsmith@umich.edu>
6 * Andy Adamson <kandros@umich.edu>
8 * Redistribution and use in source and binary forms, with or without
9 * modification, are permitted provided that the following conditions
12 * 1. Redistributions of source code must retain the above copyright
13 * notice, this list of conditions and the following disclaimer.
14 * 2. Redistributions in binary form must reproduce the above copyright
15 * notice, this list of conditions and the following disclaimer in the
16 * documentation and/or other materials provided with the distribution.
17 * 3. Neither the name of the University nor the names of its
18 * contributors may be used to endorse or promote products derived
19 * from this software without specific prior written permission.
21 * THIS SOFTWARE IS PROVIDED ``AS IS'' AND ANY EXPRESS OR IMPLIED
22 * WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF
23 * MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
24 * DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
25 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
26 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
27 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR
28 * BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF
29 * LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING
30 * NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
31 * SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
35 #include <linux/file.h>
37 #include <linux/slab.h>
38 #include <linux/namei.h>
39 #include <linux/swap.h>
40 #include <linux/pagemap.h>
41 #include <linux/ratelimit.h>
42 #include <linux/sunrpc/svcauth_gss.h>
43 #include <linux/sunrpc/addr.h>
44 #include <linux/jhash.h>
45 #include <linux/string_helpers.h>
46 #include <linux/fsnotify.h>
47 #include <linux/rhashtable.h>
48 #include <linux/nfs_ssc.h>
53 #include "current_stateid.h"
57 #include "filecache.h"
60 #define NFSDDBG_FACILITY NFSDDBG_PROC
62 #define all_ones {{ ~0, ~0}, ~0}
63 static const stateid_t one_stateid = {
65 .si_opaque = all_ones,
67 static const stateid_t zero_stateid = {
70 static const stateid_t currentstateid = {
73 static const stateid_t close_stateid = {
74 .si_generation = 0xffffffffU,
77 static u64 current_sessionid = 1;
79 #define ZERO_STATEID(stateid) (!memcmp((stateid), &zero_stateid, sizeof(stateid_t)))
80 #define ONE_STATEID(stateid) (!memcmp((stateid), &one_stateid, sizeof(stateid_t)))
81 #define CURRENT_STATEID(stateid) (!memcmp((stateid), ¤tstateid, sizeof(stateid_t)))
82 #define CLOSE_STATEID(stateid) (!memcmp((stateid), &close_stateid, sizeof(stateid_t)))
84 /* forward declarations */
85 static bool check_for_locks(struct nfs4_file *fp, struct nfs4_lockowner *lowner);
86 static void nfs4_free_ol_stateid(struct nfs4_stid *stid);
87 void nfsd4_end_grace(struct nfsd_net *nn);
88 static void _free_cpntf_state_locked(struct nfsd_net *nn, struct nfs4_cpntf_state *cps);
89 static void nfsd4_file_hash_remove(struct nfs4_file *fi);
90 static void deleg_reaper(struct nfsd_net *nn);
95 * Currently used for the del_recall_lru and file hash table. In an
96 * effort to decrease the scope of the client_mutex, this spinlock may
97 * eventually cover more:
99 static DEFINE_SPINLOCK(state_lock);
101 enum nfsd4_st_mutex_lock_subclass {
102 OPEN_STATEID_MUTEX = 0,
103 LOCK_STATEID_MUTEX = 1,
107 * A waitqueue for all in-progress 4.0 CLOSE operations that are waiting for
108 * the refcount on the open stateid to drop.
110 static DECLARE_WAIT_QUEUE_HEAD(close_wq);
113 * A waitqueue where a writer to clients/#/ctl destroying a client can
114 * wait for cl_rpc_users to drop to 0 and then for the client to be
117 static DECLARE_WAIT_QUEUE_HEAD(expiry_wq);
119 static struct kmem_cache *client_slab;
120 static struct kmem_cache *openowner_slab;
121 static struct kmem_cache *lockowner_slab;
122 static struct kmem_cache *file_slab;
123 static struct kmem_cache *stateid_slab;
124 static struct kmem_cache *deleg_slab;
125 static struct kmem_cache *odstate_slab;
127 static void free_session(struct nfsd4_session *);
129 static const struct nfsd4_callback_ops nfsd4_cb_recall_ops;
130 static const struct nfsd4_callback_ops nfsd4_cb_notify_lock_ops;
131 static const struct nfsd4_callback_ops nfsd4_cb_getattr_ops;
133 static struct workqueue_struct *laundry_wq;
135 int nfsd4_create_laundry_wq(void)
139 laundry_wq = alloc_workqueue("%s", WQ_UNBOUND, 0, "nfsd4");
140 if (laundry_wq == NULL)
145 void nfsd4_destroy_laundry_wq(void)
147 destroy_workqueue(laundry_wq);
150 static bool is_session_dead(struct nfsd4_session *ses)
152 return ses->se_flags & NFS4_SESSION_DEAD;
155 static __be32 mark_session_dead_locked(struct nfsd4_session *ses, int ref_held_by_me)
157 if (atomic_read(&ses->se_ref) > ref_held_by_me)
158 return nfserr_jukebox;
159 ses->se_flags |= NFS4_SESSION_DEAD;
163 static bool is_client_expired(struct nfs4_client *clp)
165 return clp->cl_time == 0;
168 static void nfsd4_dec_courtesy_client_count(struct nfsd_net *nn,
169 struct nfs4_client *clp)
171 if (clp->cl_state != NFSD4_ACTIVE)
172 atomic_add_unless(&nn->nfsd_courtesy_clients, -1, 0);
175 static __be32 get_client_locked(struct nfs4_client *clp)
177 struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
179 lockdep_assert_held(&nn->client_lock);
181 if (is_client_expired(clp))
182 return nfserr_expired;
183 atomic_inc(&clp->cl_rpc_users);
184 nfsd4_dec_courtesy_client_count(nn, clp);
185 clp->cl_state = NFSD4_ACTIVE;
189 /* must be called under the client_lock */
191 renew_client_locked(struct nfs4_client *clp)
193 struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
195 if (is_client_expired(clp)) {
197 printk("%s: client (clientid %08x/%08x) already expired\n",
199 clp->cl_clientid.cl_boot,
200 clp->cl_clientid.cl_id);
204 list_move_tail(&clp->cl_lru, &nn->client_lru);
205 clp->cl_time = ktime_get_boottime_seconds();
206 nfsd4_dec_courtesy_client_count(nn, clp);
207 clp->cl_state = NFSD4_ACTIVE;
210 static void put_client_renew_locked(struct nfs4_client *clp)
212 struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
214 lockdep_assert_held(&nn->client_lock);
216 if (!atomic_dec_and_test(&clp->cl_rpc_users))
218 if (!is_client_expired(clp))
219 renew_client_locked(clp);
221 wake_up_all(&expiry_wq);
224 static void put_client_renew(struct nfs4_client *clp)
226 struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
228 if (!atomic_dec_and_lock(&clp->cl_rpc_users, &nn->client_lock))
230 if (!is_client_expired(clp))
231 renew_client_locked(clp);
233 wake_up_all(&expiry_wq);
234 spin_unlock(&nn->client_lock);
237 static __be32 nfsd4_get_session_locked(struct nfsd4_session *ses)
241 if (is_session_dead(ses))
242 return nfserr_badsession;
243 status = get_client_locked(ses->se_client);
246 atomic_inc(&ses->se_ref);
250 static void nfsd4_put_session_locked(struct nfsd4_session *ses)
252 struct nfs4_client *clp = ses->se_client;
253 struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
255 lockdep_assert_held(&nn->client_lock);
257 if (atomic_dec_and_test(&ses->se_ref) && is_session_dead(ses))
259 put_client_renew_locked(clp);
262 static void nfsd4_put_session(struct nfsd4_session *ses)
264 struct nfs4_client *clp = ses->se_client;
265 struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
267 spin_lock(&nn->client_lock);
268 nfsd4_put_session_locked(ses);
269 spin_unlock(&nn->client_lock);
272 static struct nfsd4_blocked_lock *
273 find_blocked_lock(struct nfs4_lockowner *lo, struct knfsd_fh *fh,
276 struct nfsd4_blocked_lock *cur, *found = NULL;
278 spin_lock(&nn->blocked_locks_lock);
279 list_for_each_entry(cur, &lo->lo_blocked, nbl_list) {
280 if (fh_match(fh, &cur->nbl_fh)) {
281 list_del_init(&cur->nbl_list);
282 WARN_ON(list_empty(&cur->nbl_lru));
283 list_del_init(&cur->nbl_lru);
288 spin_unlock(&nn->blocked_locks_lock);
290 locks_delete_block(&found->nbl_lock);
294 static struct nfsd4_blocked_lock *
295 find_or_allocate_block(struct nfs4_lockowner *lo, struct knfsd_fh *fh,
298 struct nfsd4_blocked_lock *nbl;
300 nbl = find_blocked_lock(lo, fh, nn);
302 nbl = kmalloc(sizeof(*nbl), GFP_KERNEL);
304 INIT_LIST_HEAD(&nbl->nbl_list);
305 INIT_LIST_HEAD(&nbl->nbl_lru);
306 fh_copy_shallow(&nbl->nbl_fh, fh);
307 locks_init_lock(&nbl->nbl_lock);
308 kref_init(&nbl->nbl_kref);
309 nfsd4_init_cb(&nbl->nbl_cb, lo->lo_owner.so_client,
310 &nfsd4_cb_notify_lock_ops,
311 NFSPROC4_CLNT_CB_NOTIFY_LOCK);
318 free_nbl(struct kref *kref)
320 struct nfsd4_blocked_lock *nbl;
322 nbl = container_of(kref, struct nfsd4_blocked_lock, nbl_kref);
323 locks_release_private(&nbl->nbl_lock);
328 free_blocked_lock(struct nfsd4_blocked_lock *nbl)
330 locks_delete_block(&nbl->nbl_lock);
331 kref_put(&nbl->nbl_kref, free_nbl);
335 remove_blocked_locks(struct nfs4_lockowner *lo)
337 struct nfs4_client *clp = lo->lo_owner.so_client;
338 struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
339 struct nfsd4_blocked_lock *nbl;
342 /* Dequeue all blocked locks */
343 spin_lock(&nn->blocked_locks_lock);
344 while (!list_empty(&lo->lo_blocked)) {
345 nbl = list_first_entry(&lo->lo_blocked,
346 struct nfsd4_blocked_lock,
348 list_del_init(&nbl->nbl_list);
349 WARN_ON(list_empty(&nbl->nbl_lru));
350 list_move(&nbl->nbl_lru, &reaplist);
352 spin_unlock(&nn->blocked_locks_lock);
355 while (!list_empty(&reaplist)) {
356 nbl = list_first_entry(&reaplist, struct nfsd4_blocked_lock,
358 list_del_init(&nbl->nbl_lru);
359 free_blocked_lock(nbl);
364 nfsd4_cb_notify_lock_prepare(struct nfsd4_callback *cb)
366 struct nfsd4_blocked_lock *nbl = container_of(cb,
367 struct nfsd4_blocked_lock, nbl_cb);
368 locks_delete_block(&nbl->nbl_lock);
372 nfsd4_cb_notify_lock_done(struct nfsd4_callback *cb, struct rpc_task *task)
374 trace_nfsd_cb_notify_lock_done(&zero_stateid, task);
377 * Since this is just an optimization, we don't try very hard if it
378 * turns out not to succeed. We'll requeue it on NFS4ERR_DELAY, and
379 * just quit trying on anything else.
381 switch (task->tk_status) {
383 rpc_delay(task, 1 * HZ);
391 nfsd4_cb_notify_lock_release(struct nfsd4_callback *cb)
393 struct nfsd4_blocked_lock *nbl = container_of(cb,
394 struct nfsd4_blocked_lock, nbl_cb);
396 free_blocked_lock(nbl);
399 static const struct nfsd4_callback_ops nfsd4_cb_notify_lock_ops = {
400 .prepare = nfsd4_cb_notify_lock_prepare,
401 .done = nfsd4_cb_notify_lock_done,
402 .release = nfsd4_cb_notify_lock_release,
406 * We store the NONE, READ, WRITE, and BOTH bits separately in the
407 * st_{access,deny}_bmap field of the stateid, in order to track not
408 * only what share bits are currently in force, but also what
409 * combinations of share bits previous opens have used. This allows us
410 * to enforce the recommendation in
411 * https://datatracker.ietf.org/doc/html/rfc7530#section-16.19.4 that
412 * the server return an error if the client attempt to downgrade to a
413 * combination of share bits not explicable by closing some of its
416 * This enforcement is arguably incomplete, since we don't keep
417 * track of access/deny bit combinations; so, e.g., we allow:
419 * OPEN allow read, deny write
420 * OPEN allow both, deny none
421 * DOWNGRADE allow read, deny none
423 * which we should reject.
425 * But you could also argue that our current code is already overkill,
426 * since it only exists to return NFS4ERR_INVAL on incorrect client
430 bmap_to_share_mode(unsigned long bmap)
433 unsigned int access = 0;
435 for (i = 1; i < 4; i++) {
436 if (test_bit(i, &bmap))
442 /* set share access for a given stateid */
444 set_access(u32 access, struct nfs4_ol_stateid *stp)
446 unsigned char mask = 1 << access;
448 WARN_ON_ONCE(access > NFS4_SHARE_ACCESS_BOTH);
449 stp->st_access_bmap |= mask;
452 /* clear share access for a given stateid */
454 clear_access(u32 access, struct nfs4_ol_stateid *stp)
456 unsigned char mask = 1 << access;
458 WARN_ON_ONCE(access > NFS4_SHARE_ACCESS_BOTH);
459 stp->st_access_bmap &= ~mask;
462 /* test whether a given stateid has access */
464 test_access(u32 access, struct nfs4_ol_stateid *stp)
466 unsigned char mask = 1 << access;
468 return (bool)(stp->st_access_bmap & mask);
471 /* set share deny for a given stateid */
473 set_deny(u32 deny, struct nfs4_ol_stateid *stp)
475 unsigned char mask = 1 << deny;
477 WARN_ON_ONCE(deny > NFS4_SHARE_DENY_BOTH);
478 stp->st_deny_bmap |= mask;
481 /* clear share deny for a given stateid */
483 clear_deny(u32 deny, struct nfs4_ol_stateid *stp)
485 unsigned char mask = 1 << deny;
487 WARN_ON_ONCE(deny > NFS4_SHARE_DENY_BOTH);
488 stp->st_deny_bmap &= ~mask;
491 /* test whether a given stateid is denying specific access */
493 test_deny(u32 deny, struct nfs4_ol_stateid *stp)
495 unsigned char mask = 1 << deny;
497 return (bool)(stp->st_deny_bmap & mask);
500 static int nfs4_access_to_omode(u32 access)
502 switch (access & NFS4_SHARE_ACCESS_BOTH) {
503 case NFS4_SHARE_ACCESS_READ:
505 case NFS4_SHARE_ACCESS_WRITE:
507 case NFS4_SHARE_ACCESS_BOTH:
515 access_permit_read(struct nfs4_ol_stateid *stp)
517 return test_access(NFS4_SHARE_ACCESS_READ, stp) ||
518 test_access(NFS4_SHARE_ACCESS_BOTH, stp) ||
519 test_access(NFS4_SHARE_ACCESS_WRITE, stp);
523 access_permit_write(struct nfs4_ol_stateid *stp)
525 return test_access(NFS4_SHARE_ACCESS_WRITE, stp) ||
526 test_access(NFS4_SHARE_ACCESS_BOTH, stp);
529 static inline struct nfs4_stateowner *
530 nfs4_get_stateowner(struct nfs4_stateowner *sop)
532 atomic_inc(&sop->so_count);
537 same_owner_str(struct nfs4_stateowner *sop, struct xdr_netobj *owner)
539 return (sop->so_owner.len == owner->len) &&
540 0 == memcmp(sop->so_owner.data, owner->data, owner->len);
543 static struct nfs4_openowner *
544 find_openstateowner_str_locked(unsigned int hashval, struct nfsd4_open *open,
545 struct nfs4_client *clp)
547 struct nfs4_stateowner *so;
549 lockdep_assert_held(&clp->cl_lock);
551 list_for_each_entry(so, &clp->cl_ownerstr_hashtbl[hashval],
553 if (!so->so_is_open_owner)
555 if (same_owner_str(so, &open->op_owner))
556 return openowner(nfs4_get_stateowner(so));
561 static struct nfs4_openowner *
562 find_openstateowner_str(unsigned int hashval, struct nfsd4_open *open,
563 struct nfs4_client *clp)
565 struct nfs4_openowner *oo;
567 spin_lock(&clp->cl_lock);
568 oo = find_openstateowner_str_locked(hashval, open, clp);
569 spin_unlock(&clp->cl_lock);
574 opaque_hashval(const void *ptr, int nbytes)
576 unsigned char *cptr = (unsigned char *) ptr;
586 static void nfsd4_free_file_rcu(struct rcu_head *rcu)
588 struct nfs4_file *fp = container_of(rcu, struct nfs4_file, fi_rcu);
590 kmem_cache_free(file_slab, fp);
594 put_nfs4_file(struct nfs4_file *fi)
596 if (refcount_dec_and_test(&fi->fi_ref)) {
597 nfsd4_file_hash_remove(fi);
598 WARN_ON_ONCE(!list_empty(&fi->fi_clnt_odstate));
599 WARN_ON_ONCE(!list_empty(&fi->fi_delegations));
600 call_rcu(&fi->fi_rcu, nfsd4_free_file_rcu);
604 static struct nfsd_file *
605 find_writeable_file_locked(struct nfs4_file *f)
607 struct nfsd_file *ret;
609 lockdep_assert_held(&f->fi_lock);
611 ret = nfsd_file_get(f->fi_fds[O_WRONLY]);
613 ret = nfsd_file_get(f->fi_fds[O_RDWR]);
617 static struct nfsd_file *
618 find_writeable_file(struct nfs4_file *f)
620 struct nfsd_file *ret;
622 spin_lock(&f->fi_lock);
623 ret = find_writeable_file_locked(f);
624 spin_unlock(&f->fi_lock);
629 static struct nfsd_file *
630 find_readable_file_locked(struct nfs4_file *f)
632 struct nfsd_file *ret;
634 lockdep_assert_held(&f->fi_lock);
636 ret = nfsd_file_get(f->fi_fds[O_RDONLY]);
638 ret = nfsd_file_get(f->fi_fds[O_RDWR]);
642 static struct nfsd_file *
643 find_readable_file(struct nfs4_file *f)
645 struct nfsd_file *ret;
647 spin_lock(&f->fi_lock);
648 ret = find_readable_file_locked(f);
649 spin_unlock(&f->fi_lock);
654 static struct nfsd_file *
655 find_rw_file(struct nfs4_file *f)
657 struct nfsd_file *ret;
659 spin_lock(&f->fi_lock);
660 ret = nfsd_file_get(f->fi_fds[O_RDWR]);
661 spin_unlock(&f->fi_lock);
667 find_any_file(struct nfs4_file *f)
669 struct nfsd_file *ret;
673 spin_lock(&f->fi_lock);
674 ret = nfsd_file_get(f->fi_fds[O_RDWR]);
676 ret = nfsd_file_get(f->fi_fds[O_WRONLY]);
678 ret = nfsd_file_get(f->fi_fds[O_RDONLY]);
680 spin_unlock(&f->fi_lock);
684 static struct nfsd_file *find_any_file_locked(struct nfs4_file *f)
686 lockdep_assert_held(&f->fi_lock);
688 if (f->fi_fds[O_RDWR])
689 return f->fi_fds[O_RDWR];
690 if (f->fi_fds[O_WRONLY])
691 return f->fi_fds[O_WRONLY];
692 if (f->fi_fds[O_RDONLY])
693 return f->fi_fds[O_RDONLY];
697 static atomic_long_t num_delegations;
698 unsigned long max_delegations;
701 * Open owner state (share locks)
704 /* hash tables for lock and open owners */
705 #define OWNER_HASH_BITS 8
706 #define OWNER_HASH_SIZE (1 << OWNER_HASH_BITS)
707 #define OWNER_HASH_MASK (OWNER_HASH_SIZE - 1)
709 static unsigned int ownerstr_hashval(struct xdr_netobj *ownername)
713 ret = opaque_hashval(ownername->data, ownername->len);
714 return ret & OWNER_HASH_MASK;
717 static struct rhltable nfs4_file_rhltable ____cacheline_aligned_in_smp;
719 static const struct rhashtable_params nfs4_file_rhash_params = {
720 .key_len = sizeof_field(struct nfs4_file, fi_inode),
721 .key_offset = offsetof(struct nfs4_file, fi_inode),
722 .head_offset = offsetof(struct nfs4_file, fi_rlist),
725 * Start with a single page hash table to reduce resizing churn
726 * on light workloads.
729 .automatic_shrinking = true,
733 * Check if courtesy clients have conflicting access and resolve it if possible
735 * access: is op_share_access if share_access is true.
736 * Check if access mode, op_share_access, would conflict with
737 * the current deny mode of the file 'fp'.
738 * access: is op_share_deny if share_access is false.
739 * Check if the deny mode, op_share_deny, would conflict with
740 * current access of the file 'fp'.
741 * stp: skip checking this entry.
742 * new_stp: normal open, not open upgrade.
745 * false - access/deny mode conflict with normal client.
746 * true - no conflict or conflict with courtesy client(s) is resolved.
749 nfs4_resolve_deny_conflicts_locked(struct nfs4_file *fp, bool new_stp,
750 struct nfs4_ol_stateid *stp, u32 access, bool share_access)
752 struct nfs4_ol_stateid *st;
753 bool resolvable = true;
756 struct nfs4_client *clp;
758 lockdep_assert_held(&fp->fi_lock);
759 list_for_each_entry(st, &fp->fi_stateids, st_perfile) {
760 /* ignore lock stateid */
763 if (st == stp && new_stp)
765 /* check file access against deny mode or vice versa */
766 bmap = share_access ? st->st_deny_bmap : st->st_access_bmap;
767 if (!(access & bmap_to_share_mode(bmap)))
769 clp = st->st_stid.sc_client;
770 if (try_to_expire_client(clp))
776 clp = stp->st_stid.sc_client;
777 nn = net_generic(clp->net, nfsd_net_id);
778 mod_delayed_work(laundry_wq, &nn->laundromat_work, 0);
784 __nfs4_file_get_access(struct nfs4_file *fp, u32 access)
786 lockdep_assert_held(&fp->fi_lock);
788 if (access & NFS4_SHARE_ACCESS_WRITE)
789 atomic_inc(&fp->fi_access[O_WRONLY]);
790 if (access & NFS4_SHARE_ACCESS_READ)
791 atomic_inc(&fp->fi_access[O_RDONLY]);
795 nfs4_file_get_access(struct nfs4_file *fp, u32 access)
797 lockdep_assert_held(&fp->fi_lock);
799 /* Does this access mode make sense? */
800 if (access & ~NFS4_SHARE_ACCESS_BOTH)
803 /* Does it conflict with a deny mode already set? */
804 if ((access & fp->fi_share_deny) != 0)
805 return nfserr_share_denied;
807 __nfs4_file_get_access(fp, access);
811 static __be32 nfs4_file_check_deny(struct nfs4_file *fp, u32 deny)
813 /* Common case is that there is no deny mode. */
815 /* Does this deny mode make sense? */
816 if (deny & ~NFS4_SHARE_DENY_BOTH)
819 if ((deny & NFS4_SHARE_DENY_READ) &&
820 atomic_read(&fp->fi_access[O_RDONLY]))
821 return nfserr_share_denied;
823 if ((deny & NFS4_SHARE_DENY_WRITE) &&
824 atomic_read(&fp->fi_access[O_WRONLY]))
825 return nfserr_share_denied;
830 static void __nfs4_file_put_access(struct nfs4_file *fp, int oflag)
832 might_lock(&fp->fi_lock);
834 if (atomic_dec_and_lock(&fp->fi_access[oflag], &fp->fi_lock)) {
835 struct nfsd_file *f1 = NULL;
836 struct nfsd_file *f2 = NULL;
838 swap(f1, fp->fi_fds[oflag]);
839 if (atomic_read(&fp->fi_access[1 - oflag]) == 0)
840 swap(f2, fp->fi_fds[O_RDWR]);
841 spin_unlock(&fp->fi_lock);
849 static void nfs4_file_put_access(struct nfs4_file *fp, u32 access)
851 WARN_ON_ONCE(access & ~NFS4_SHARE_ACCESS_BOTH);
853 if (access & NFS4_SHARE_ACCESS_WRITE)
854 __nfs4_file_put_access(fp, O_WRONLY);
855 if (access & NFS4_SHARE_ACCESS_READ)
856 __nfs4_file_put_access(fp, O_RDONLY);
860 * Allocate a new open/delegation state counter. This is needed for
861 * pNFS for proper return on close semantics.
863 * Note that we only allocate it for pNFS-enabled exports, otherwise
864 * all pointers to struct nfs4_clnt_odstate are always NULL.
866 static struct nfs4_clnt_odstate *
867 alloc_clnt_odstate(struct nfs4_client *clp)
869 struct nfs4_clnt_odstate *co;
871 co = kmem_cache_zalloc(odstate_slab, GFP_KERNEL);
874 refcount_set(&co->co_odcount, 1);
880 hash_clnt_odstate_locked(struct nfs4_clnt_odstate *co)
882 struct nfs4_file *fp = co->co_file;
884 lockdep_assert_held(&fp->fi_lock);
885 list_add(&co->co_perfile, &fp->fi_clnt_odstate);
889 get_clnt_odstate(struct nfs4_clnt_odstate *co)
892 refcount_inc(&co->co_odcount);
896 put_clnt_odstate(struct nfs4_clnt_odstate *co)
898 struct nfs4_file *fp;
904 if (refcount_dec_and_lock(&co->co_odcount, &fp->fi_lock)) {
905 list_del(&co->co_perfile);
906 spin_unlock(&fp->fi_lock);
908 nfsd4_return_all_file_layouts(co->co_client, fp);
909 kmem_cache_free(odstate_slab, co);
913 static struct nfs4_clnt_odstate *
914 find_or_hash_clnt_odstate(struct nfs4_file *fp, struct nfs4_clnt_odstate *new)
916 struct nfs4_clnt_odstate *co;
917 struct nfs4_client *cl;
924 spin_lock(&fp->fi_lock);
925 list_for_each_entry(co, &fp->fi_clnt_odstate, co_perfile) {
926 if (co->co_client == cl) {
927 get_clnt_odstate(co);
933 hash_clnt_odstate_locked(new);
935 spin_unlock(&fp->fi_lock);
939 struct nfs4_stid *nfs4_alloc_stid(struct nfs4_client *cl, struct kmem_cache *slab,
940 void (*sc_free)(struct nfs4_stid *))
942 struct nfs4_stid *stid;
945 stid = kmem_cache_zalloc(slab, GFP_KERNEL);
949 idr_preload(GFP_KERNEL);
950 spin_lock(&cl->cl_lock);
951 /* Reserving 0 for start of file in nfsdfs "states" file: */
952 new_id = idr_alloc_cyclic(&cl->cl_stateids, stid, 1, 0, GFP_NOWAIT);
953 spin_unlock(&cl->cl_lock);
958 stid->sc_free = sc_free;
959 stid->sc_client = cl;
960 stid->sc_stateid.si_opaque.so_id = new_id;
961 stid->sc_stateid.si_opaque.so_clid = cl->cl_clientid;
962 /* Will be incremented before return to client: */
963 refcount_set(&stid->sc_count, 1);
964 spin_lock_init(&stid->sc_lock);
965 INIT_LIST_HEAD(&stid->sc_cp_list);
968 * It shouldn't be a problem to reuse an opaque stateid value.
969 * I don't think it is for 4.1. But with 4.0 I worry that, for
970 * example, a stray write retransmission could be accepted by
971 * the server when it should have been rejected. Therefore,
972 * adopt a trick from the sctp code to attempt to maximize the
973 * amount of time until an id is reused, by ensuring they always
974 * "increase" (mod INT_MAX):
978 kmem_cache_free(slab, stid);
983 * Create a unique stateid_t to represent each COPY.
985 static int nfs4_init_cp_state(struct nfsd_net *nn, copy_stateid_t *stid,
986 unsigned char cs_type)
990 stid->cs_stid.si_opaque.so_clid.cl_boot = (u32)nn->boot_time;
991 stid->cs_stid.si_opaque.so_clid.cl_id = nn->s2s_cp_cl_id;
993 idr_preload(GFP_KERNEL);
994 spin_lock(&nn->s2s_cp_lock);
995 new_id = idr_alloc_cyclic(&nn->s2s_cp_stateids, stid, 0, 0, GFP_NOWAIT);
996 stid->cs_stid.si_opaque.so_id = new_id;
997 stid->cs_stid.si_generation = 1;
998 spin_unlock(&nn->s2s_cp_lock);
1002 stid->cs_type = cs_type;
1006 int nfs4_init_copy_state(struct nfsd_net *nn, struct nfsd4_copy *copy)
1008 return nfs4_init_cp_state(nn, ©->cp_stateid, NFS4_COPY_STID);
1011 struct nfs4_cpntf_state *nfs4_alloc_init_cpntf_state(struct nfsd_net *nn,
1012 struct nfs4_stid *p_stid)
1014 struct nfs4_cpntf_state *cps;
1016 cps = kzalloc(sizeof(struct nfs4_cpntf_state), GFP_KERNEL);
1019 cps->cpntf_time = ktime_get_boottime_seconds();
1020 refcount_set(&cps->cp_stateid.cs_count, 1);
1021 if (!nfs4_init_cp_state(nn, &cps->cp_stateid, NFS4_COPYNOTIFY_STID))
1023 spin_lock(&nn->s2s_cp_lock);
1024 list_add(&cps->cp_list, &p_stid->sc_cp_list);
1025 spin_unlock(&nn->s2s_cp_lock);
1032 void nfs4_free_copy_state(struct nfsd4_copy *copy)
1034 struct nfsd_net *nn;
1036 if (copy->cp_stateid.cs_type != NFS4_COPY_STID)
1038 nn = net_generic(copy->cp_clp->net, nfsd_net_id);
1039 spin_lock(&nn->s2s_cp_lock);
1040 idr_remove(&nn->s2s_cp_stateids,
1041 copy->cp_stateid.cs_stid.si_opaque.so_id);
1042 spin_unlock(&nn->s2s_cp_lock);
1045 static void nfs4_free_cpntf_statelist(struct net *net, struct nfs4_stid *stid)
1047 struct nfs4_cpntf_state *cps;
1048 struct nfsd_net *nn;
1050 nn = net_generic(net, nfsd_net_id);
1051 spin_lock(&nn->s2s_cp_lock);
1052 while (!list_empty(&stid->sc_cp_list)) {
1053 cps = list_first_entry(&stid->sc_cp_list,
1054 struct nfs4_cpntf_state, cp_list);
1055 _free_cpntf_state_locked(nn, cps);
1057 spin_unlock(&nn->s2s_cp_lock);
1060 static struct nfs4_ol_stateid * nfs4_alloc_open_stateid(struct nfs4_client *clp)
1062 struct nfs4_stid *stid;
1064 stid = nfs4_alloc_stid(clp, stateid_slab, nfs4_free_ol_stateid);
1068 return openlockstateid(stid);
1071 static void nfs4_free_deleg(struct nfs4_stid *stid)
1073 struct nfs4_delegation *dp = delegstateid(stid);
1075 WARN_ON_ONCE(!list_empty(&stid->sc_cp_list));
1076 WARN_ON_ONCE(!list_empty(&dp->dl_perfile));
1077 WARN_ON_ONCE(!list_empty(&dp->dl_perclnt));
1078 WARN_ON_ONCE(!list_empty(&dp->dl_recall_lru));
1079 kmem_cache_free(deleg_slab, stid);
1080 atomic_long_dec(&num_delegations);
1084 * When we recall a delegation, we should be careful not to hand it
1085 * out again straight away.
1086 * To ensure this we keep a pair of bloom filters ('new' and 'old')
1087 * in which the filehandles of recalled delegations are "stored".
1088 * If a filehandle appear in either filter, a delegation is blocked.
1089 * When a delegation is recalled, the filehandle is stored in the "new"
1091 * Every 30 seconds we swap the filters and clear the "new" one,
1092 * unless both are empty of course.
1094 * Each filter is 256 bits. We hash the filehandle to 32bit and use the
1095 * low 3 bytes as hash-table indices.
1097 * 'blocked_delegations_lock', which is always taken in block_delegations(),
1098 * is used to manage concurrent access. Testing does not need the lock
1099 * except when swapping the two filters.
1101 static DEFINE_SPINLOCK(blocked_delegations_lock);
1102 static struct bloom_pair {
1103 int entries, old_entries;
1105 int new; /* index into 'set' */
1106 DECLARE_BITMAP(set[2], 256);
1107 } blocked_delegations;
1109 static int delegation_blocked(struct knfsd_fh *fh)
1112 struct bloom_pair *bd = &blocked_delegations;
1114 if (bd->entries == 0)
1116 if (ktime_get_seconds() - bd->swap_time > 30) {
1117 spin_lock(&blocked_delegations_lock);
1118 if (ktime_get_seconds() - bd->swap_time > 30) {
1119 bd->entries -= bd->old_entries;
1120 bd->old_entries = bd->entries;
1121 memset(bd->set[bd->new], 0,
1122 sizeof(bd->set[0]));
1123 bd->new = 1-bd->new;
1124 bd->swap_time = ktime_get_seconds();
1126 spin_unlock(&blocked_delegations_lock);
1128 hash = jhash(&fh->fh_raw, fh->fh_size, 0);
1129 if (test_bit(hash&255, bd->set[0]) &&
1130 test_bit((hash>>8)&255, bd->set[0]) &&
1131 test_bit((hash>>16)&255, bd->set[0]))
1134 if (test_bit(hash&255, bd->set[1]) &&
1135 test_bit((hash>>8)&255, bd->set[1]) &&
1136 test_bit((hash>>16)&255, bd->set[1]))
1142 static void block_delegations(struct knfsd_fh *fh)
1145 struct bloom_pair *bd = &blocked_delegations;
1147 hash = jhash(&fh->fh_raw, fh->fh_size, 0);
1149 spin_lock(&blocked_delegations_lock);
1150 __set_bit(hash&255, bd->set[bd->new]);
1151 __set_bit((hash>>8)&255, bd->set[bd->new]);
1152 __set_bit((hash>>16)&255, bd->set[bd->new]);
1153 if (bd->entries == 0)
1154 bd->swap_time = ktime_get_seconds();
1156 spin_unlock(&blocked_delegations_lock);
1159 static struct nfs4_delegation *
1160 alloc_init_deleg(struct nfs4_client *clp, struct nfs4_file *fp,
1161 struct nfs4_clnt_odstate *odstate, u32 dl_type)
1163 struct nfs4_delegation *dp;
1164 struct nfs4_stid *stid;
1167 dprintk("NFSD alloc_init_deleg\n");
1168 n = atomic_long_inc_return(&num_delegations);
1169 if (n < 0 || n > max_delegations)
1171 if (delegation_blocked(&fp->fi_fhandle))
1173 stid = nfs4_alloc_stid(clp, deleg_slab, nfs4_free_deleg);
1176 dp = delegstateid(stid);
1179 * delegation seqid's are never incremented. The 4.1 special
1180 * meaning of seqid 0 isn't meaningful, really, but let's avoid
1181 * 0 anyway just for consistency and use 1:
1183 dp->dl_stid.sc_stateid.si_generation = 1;
1184 INIT_LIST_HEAD(&dp->dl_perfile);
1185 INIT_LIST_HEAD(&dp->dl_perclnt);
1186 INIT_LIST_HEAD(&dp->dl_recall_lru);
1187 dp->dl_clnt_odstate = odstate;
1188 get_clnt_odstate(odstate);
1189 dp->dl_type = dl_type;
1191 dp->dl_recalled = false;
1192 nfsd4_init_cb(&dp->dl_recall, dp->dl_stid.sc_client,
1193 &nfsd4_cb_recall_ops, NFSPROC4_CLNT_CB_RECALL);
1194 nfsd4_init_cb(&dp->dl_cb_fattr.ncf_getattr, dp->dl_stid.sc_client,
1195 &nfsd4_cb_getattr_ops, NFSPROC4_CLNT_CB_GETATTR);
1196 dp->dl_cb_fattr.ncf_file_modified = false;
1197 dp->dl_cb_fattr.ncf_cb_bmap[0] = FATTR4_WORD0_CHANGE | FATTR4_WORD0_SIZE;
1199 dp->dl_stid.sc_file = fp;
1202 atomic_long_dec(&num_delegations);
1207 nfs4_put_stid(struct nfs4_stid *s)
1209 struct nfs4_file *fp = s->sc_file;
1210 struct nfs4_client *clp = s->sc_client;
1212 might_lock(&clp->cl_lock);
1214 if (!refcount_dec_and_lock(&s->sc_count, &clp->cl_lock)) {
1215 wake_up_all(&close_wq);
1218 idr_remove(&clp->cl_stateids, s->sc_stateid.si_opaque.so_id);
1219 if (s->sc_status & SC_STATUS_ADMIN_REVOKED)
1220 atomic_dec(&s->sc_client->cl_admin_revoked);
1221 nfs4_free_cpntf_statelist(clp->net, s);
1222 spin_unlock(&clp->cl_lock);
1229 nfs4_inc_and_copy_stateid(stateid_t *dst, struct nfs4_stid *stid)
1231 stateid_t *src = &stid->sc_stateid;
1233 spin_lock(&stid->sc_lock);
1234 if (unlikely(++src->si_generation == 0))
1235 src->si_generation = 1;
1236 memcpy(dst, src, sizeof(*dst));
1237 spin_unlock(&stid->sc_lock);
1240 static void put_deleg_file(struct nfs4_file *fp)
1242 struct nfsd_file *nf = NULL;
1244 spin_lock(&fp->fi_lock);
1245 if (--fp->fi_delegees == 0)
1246 swap(nf, fp->fi_deleg_file);
1247 spin_unlock(&fp->fi_lock);
1253 static void nfs4_unlock_deleg_lease(struct nfs4_delegation *dp)
1255 struct nfs4_file *fp = dp->dl_stid.sc_file;
1256 struct nfsd_file *nf = fp->fi_deleg_file;
1258 WARN_ON_ONCE(!fp->fi_delegees);
1260 kernel_setlease(nf->nf_file, F_UNLCK, NULL, (void **)&dp);
1264 static void destroy_unhashed_deleg(struct nfs4_delegation *dp)
1266 put_clnt_odstate(dp->dl_clnt_odstate);
1267 nfs4_unlock_deleg_lease(dp);
1268 nfs4_put_stid(&dp->dl_stid);
1272 * nfs4_delegation_exists - Discover if this delegation already exists
1273 * @clp: a pointer to the nfs4_client we're granting a delegation to
1274 * @fp: a pointer to the nfs4_file we're granting a delegation on
1277 * On success: true iff an existing delegation is found
1281 nfs4_delegation_exists(struct nfs4_client *clp, struct nfs4_file *fp)
1283 struct nfs4_delegation *searchdp = NULL;
1284 struct nfs4_client *searchclp = NULL;
1286 lockdep_assert_held(&state_lock);
1287 lockdep_assert_held(&fp->fi_lock);
1289 list_for_each_entry(searchdp, &fp->fi_delegations, dl_perfile) {
1290 searchclp = searchdp->dl_stid.sc_client;
1291 if (clp == searchclp) {
1299 * hash_delegation_locked - Add a delegation to the appropriate lists
1300 * @dp: a pointer to the nfs4_delegation we are adding.
1301 * @fp: a pointer to the nfs4_file we're granting a delegation on
1304 * On success: NULL if the delegation was successfully hashed.
1306 * On error: -EAGAIN if one was previously granted to this
1307 * nfs4_client for this nfs4_file. Delegation is not hashed.
1312 hash_delegation_locked(struct nfs4_delegation *dp, struct nfs4_file *fp)
1314 struct nfs4_client *clp = dp->dl_stid.sc_client;
1316 lockdep_assert_held(&state_lock);
1317 lockdep_assert_held(&fp->fi_lock);
1318 lockdep_assert_held(&clp->cl_lock);
1320 if (nfs4_delegation_exists(clp, fp))
1322 refcount_inc(&dp->dl_stid.sc_count);
1323 dp->dl_stid.sc_type = SC_TYPE_DELEG;
1324 list_add(&dp->dl_perfile, &fp->fi_delegations);
1325 list_add(&dp->dl_perclnt, &clp->cl_delegations);
1329 static bool delegation_hashed(struct nfs4_delegation *dp)
1331 return !(list_empty(&dp->dl_perfile));
1335 unhash_delegation_locked(struct nfs4_delegation *dp, unsigned short statusmask)
1337 struct nfs4_file *fp = dp->dl_stid.sc_file;
1339 lockdep_assert_held(&state_lock);
1341 if (!delegation_hashed(dp))
1344 if (statusmask == SC_STATUS_REVOKED &&
1345 dp->dl_stid.sc_client->cl_minorversion == 0)
1346 statusmask = SC_STATUS_CLOSED;
1347 dp->dl_stid.sc_status |= statusmask;
1348 if (statusmask & SC_STATUS_ADMIN_REVOKED)
1349 atomic_inc(&dp->dl_stid.sc_client->cl_admin_revoked);
1351 /* Ensure that deleg break won't try to requeue it */
1353 spin_lock(&fp->fi_lock);
1354 list_del_init(&dp->dl_perclnt);
1355 list_del_init(&dp->dl_recall_lru);
1356 list_del_init(&dp->dl_perfile);
1357 spin_unlock(&fp->fi_lock);
1361 static void destroy_delegation(struct nfs4_delegation *dp)
1365 spin_lock(&state_lock);
1366 unhashed = unhash_delegation_locked(dp, SC_STATUS_CLOSED);
1367 spin_unlock(&state_lock);
1369 destroy_unhashed_deleg(dp);
1372 static void revoke_delegation(struct nfs4_delegation *dp)
1374 struct nfs4_client *clp = dp->dl_stid.sc_client;
1376 WARN_ON(!list_empty(&dp->dl_recall_lru));
1378 trace_nfsd_stid_revoke(&dp->dl_stid);
1380 if (dp->dl_stid.sc_status &
1381 (SC_STATUS_REVOKED | SC_STATUS_ADMIN_REVOKED)) {
1382 spin_lock(&clp->cl_lock);
1383 refcount_inc(&dp->dl_stid.sc_count);
1384 list_add(&dp->dl_recall_lru, &clp->cl_revoked);
1385 spin_unlock(&clp->cl_lock);
1387 destroy_unhashed_deleg(dp);
1394 static unsigned int clientid_hashval(u32 id)
1396 return id & CLIENT_HASH_MASK;
1399 static unsigned int clientstr_hashval(struct xdr_netobj name)
1401 return opaque_hashval(name.data, 8) & CLIENT_HASH_MASK;
1405 * A stateid that had a deny mode associated with it is being released
1406 * or downgraded. Recalculate the deny mode on the file.
1409 recalculate_deny_mode(struct nfs4_file *fp)
1411 struct nfs4_ol_stateid *stp;
1413 spin_lock(&fp->fi_lock);
1414 fp->fi_share_deny = 0;
1415 list_for_each_entry(stp, &fp->fi_stateids, st_perfile)
1416 fp->fi_share_deny |= bmap_to_share_mode(stp->st_deny_bmap);
1417 spin_unlock(&fp->fi_lock);
1421 reset_union_bmap_deny(u32 deny, struct nfs4_ol_stateid *stp)
1424 bool change = false;
1426 for (i = 1; i < 4; i++) {
1427 if ((i & deny) != i) {
1433 /* Recalculate per-file deny mode if there was a change */
1435 recalculate_deny_mode(stp->st_stid.sc_file);
1438 /* release all access and file references for a given stateid */
1440 release_all_access(struct nfs4_ol_stateid *stp)
1443 struct nfs4_file *fp = stp->st_stid.sc_file;
1445 if (fp && stp->st_deny_bmap != 0)
1446 recalculate_deny_mode(fp);
1448 for (i = 1; i < 4; i++) {
1449 if (test_access(i, stp))
1450 nfs4_file_put_access(stp->st_stid.sc_file, i);
1451 clear_access(i, stp);
1455 static inline void nfs4_free_stateowner(struct nfs4_stateowner *sop)
1457 kfree(sop->so_owner.data);
1458 sop->so_ops->so_free(sop);
1461 static void nfs4_put_stateowner(struct nfs4_stateowner *sop)
1463 struct nfs4_client *clp = sop->so_client;
1465 might_lock(&clp->cl_lock);
1467 if (!atomic_dec_and_lock(&sop->so_count, &clp->cl_lock))
1469 sop->so_ops->so_unhash(sop);
1470 spin_unlock(&clp->cl_lock);
1471 nfs4_free_stateowner(sop);
1475 nfs4_ol_stateid_unhashed(const struct nfs4_ol_stateid *stp)
1477 return list_empty(&stp->st_perfile);
1480 static bool unhash_ol_stateid(struct nfs4_ol_stateid *stp)
1482 struct nfs4_file *fp = stp->st_stid.sc_file;
1484 lockdep_assert_held(&stp->st_stateowner->so_client->cl_lock);
1486 if (list_empty(&stp->st_perfile))
1489 spin_lock(&fp->fi_lock);
1490 list_del_init(&stp->st_perfile);
1491 spin_unlock(&fp->fi_lock);
1492 list_del(&stp->st_perstateowner);
1496 static void nfs4_free_ol_stateid(struct nfs4_stid *stid)
1498 struct nfs4_ol_stateid *stp = openlockstateid(stid);
1500 put_clnt_odstate(stp->st_clnt_odstate);
1501 release_all_access(stp);
1502 if (stp->st_stateowner)
1503 nfs4_put_stateowner(stp->st_stateowner);
1504 WARN_ON(!list_empty(&stid->sc_cp_list));
1505 kmem_cache_free(stateid_slab, stid);
1508 static void nfs4_free_lock_stateid(struct nfs4_stid *stid)
1510 struct nfs4_ol_stateid *stp = openlockstateid(stid);
1511 struct nfs4_lockowner *lo = lockowner(stp->st_stateowner);
1512 struct nfsd_file *nf;
1514 nf = find_any_file(stp->st_stid.sc_file);
1516 get_file(nf->nf_file);
1517 filp_close(nf->nf_file, (fl_owner_t)lo);
1520 nfs4_free_ol_stateid(stid);
1524 * Put the persistent reference to an already unhashed generic stateid, while
1525 * holding the cl_lock. If it's the last reference, then put it onto the
1526 * reaplist for later destruction.
1528 static void put_ol_stateid_locked(struct nfs4_ol_stateid *stp,
1529 struct list_head *reaplist)
1531 struct nfs4_stid *s = &stp->st_stid;
1532 struct nfs4_client *clp = s->sc_client;
1534 lockdep_assert_held(&clp->cl_lock);
1536 WARN_ON_ONCE(!list_empty(&stp->st_locks));
1538 if (!refcount_dec_and_test(&s->sc_count)) {
1539 wake_up_all(&close_wq);
1543 idr_remove(&clp->cl_stateids, s->sc_stateid.si_opaque.so_id);
1544 if (s->sc_status & SC_STATUS_ADMIN_REVOKED)
1545 atomic_dec(&s->sc_client->cl_admin_revoked);
1546 list_add(&stp->st_locks, reaplist);
1549 static bool unhash_lock_stateid(struct nfs4_ol_stateid *stp)
1551 lockdep_assert_held(&stp->st_stid.sc_client->cl_lock);
1553 if (!unhash_ol_stateid(stp))
1555 list_del_init(&stp->st_locks);
1556 stp->st_stid.sc_status |= SC_STATUS_CLOSED;
1560 static void release_lock_stateid(struct nfs4_ol_stateid *stp)
1562 struct nfs4_client *clp = stp->st_stid.sc_client;
1565 spin_lock(&clp->cl_lock);
1566 unhashed = unhash_lock_stateid(stp);
1567 spin_unlock(&clp->cl_lock);
1569 nfs4_put_stid(&stp->st_stid);
1572 static void unhash_lockowner_locked(struct nfs4_lockowner *lo)
1574 struct nfs4_client *clp = lo->lo_owner.so_client;
1576 lockdep_assert_held(&clp->cl_lock);
1578 list_del_init(&lo->lo_owner.so_strhash);
1582 * Free a list of generic stateids that were collected earlier after being
1586 free_ol_stateid_reaplist(struct list_head *reaplist)
1588 struct nfs4_ol_stateid *stp;
1589 struct nfs4_file *fp;
1593 while (!list_empty(reaplist)) {
1594 stp = list_first_entry(reaplist, struct nfs4_ol_stateid,
1596 list_del(&stp->st_locks);
1597 fp = stp->st_stid.sc_file;
1598 stp->st_stid.sc_free(&stp->st_stid);
1604 static void release_open_stateid_locks(struct nfs4_ol_stateid *open_stp,
1605 struct list_head *reaplist)
1607 struct nfs4_ol_stateid *stp;
1609 lockdep_assert_held(&open_stp->st_stid.sc_client->cl_lock);
1611 while (!list_empty(&open_stp->st_locks)) {
1612 stp = list_entry(open_stp->st_locks.next,
1613 struct nfs4_ol_stateid, st_locks);
1614 unhash_lock_stateid(stp);
1615 put_ol_stateid_locked(stp, reaplist);
1619 static bool unhash_open_stateid(struct nfs4_ol_stateid *stp,
1620 struct list_head *reaplist)
1622 lockdep_assert_held(&stp->st_stid.sc_client->cl_lock);
1624 if (!unhash_ol_stateid(stp))
1626 release_open_stateid_locks(stp, reaplist);
1630 static void release_open_stateid(struct nfs4_ol_stateid *stp)
1632 LIST_HEAD(reaplist);
1634 spin_lock(&stp->st_stid.sc_client->cl_lock);
1635 stp->st_stid.sc_status |= SC_STATUS_CLOSED;
1636 if (unhash_open_stateid(stp, &reaplist))
1637 put_ol_stateid_locked(stp, &reaplist);
1638 spin_unlock(&stp->st_stid.sc_client->cl_lock);
1639 free_ol_stateid_reaplist(&reaplist);
1642 static void unhash_openowner_locked(struct nfs4_openowner *oo)
1644 struct nfs4_client *clp = oo->oo_owner.so_client;
1646 lockdep_assert_held(&clp->cl_lock);
1648 list_del_init(&oo->oo_owner.so_strhash);
1649 list_del_init(&oo->oo_perclient);
1652 static void release_last_closed_stateid(struct nfs4_openowner *oo)
1654 struct nfsd_net *nn = net_generic(oo->oo_owner.so_client->net,
1656 struct nfs4_ol_stateid *s;
1658 spin_lock(&nn->client_lock);
1659 s = oo->oo_last_closed_stid;
1661 list_del_init(&oo->oo_close_lru);
1662 oo->oo_last_closed_stid = NULL;
1664 spin_unlock(&nn->client_lock);
1666 nfs4_put_stid(&s->st_stid);
1669 static void release_openowner(struct nfs4_openowner *oo)
1671 struct nfs4_ol_stateid *stp;
1672 struct nfs4_client *clp = oo->oo_owner.so_client;
1673 struct list_head reaplist;
1675 INIT_LIST_HEAD(&reaplist);
1677 spin_lock(&clp->cl_lock);
1678 unhash_openowner_locked(oo);
1679 while (!list_empty(&oo->oo_owner.so_stateids)) {
1680 stp = list_first_entry(&oo->oo_owner.so_stateids,
1681 struct nfs4_ol_stateid, st_perstateowner);
1682 if (unhash_open_stateid(stp, &reaplist))
1683 put_ol_stateid_locked(stp, &reaplist);
1685 spin_unlock(&clp->cl_lock);
1686 free_ol_stateid_reaplist(&reaplist);
1687 release_last_closed_stateid(oo);
1688 nfs4_put_stateowner(&oo->oo_owner);
1691 static struct nfs4_stid *find_one_sb_stid(struct nfs4_client *clp,
1692 struct super_block *sb,
1693 unsigned int sc_types)
1695 unsigned long id, tmp;
1696 struct nfs4_stid *stid;
1698 spin_lock(&clp->cl_lock);
1699 idr_for_each_entry_ul(&clp->cl_stateids, stid, tmp, id)
1700 if ((stid->sc_type & sc_types) &&
1701 stid->sc_status == 0 &&
1702 stid->sc_file->fi_inode->i_sb == sb) {
1703 refcount_inc(&stid->sc_count);
1706 spin_unlock(&clp->cl_lock);
1711 * nfsd4_revoke_states - revoke all nfsv4 states associated with given filesystem
1712 * @net: used to identify instance of nfsd (there is one per net namespace)
1713 * @sb: super_block used to identify target filesystem
1715 * All nfs4 states (open, lock, delegation, layout) held by the server instance
1716 * and associated with a file on the given filesystem will be revoked resulting
1717 * in any files being closed and so all references from nfsd to the filesystem
1718 * being released. Thus nfsd will no longer prevent the filesystem from being
1721 * The clients which own the states will subsequently being notified that the
1722 * states have been "admin-revoked".
1724 void nfsd4_revoke_states(struct net *net, struct super_block *sb)
1726 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
1727 unsigned int idhashval;
1728 unsigned int sc_types;
1730 sc_types = SC_TYPE_OPEN | SC_TYPE_LOCK | SC_TYPE_DELEG | SC_TYPE_LAYOUT;
1732 spin_lock(&nn->client_lock);
1733 for (idhashval = 0; idhashval < CLIENT_HASH_MASK; idhashval++) {
1734 struct list_head *head = &nn->conf_id_hashtbl[idhashval];
1735 struct nfs4_client *clp;
1737 list_for_each_entry(clp, head, cl_idhash) {
1738 struct nfs4_stid *stid = find_one_sb_stid(clp, sb,
1741 struct nfs4_ol_stateid *stp;
1742 struct nfs4_delegation *dp;
1743 struct nfs4_layout_stateid *ls;
1745 spin_unlock(&nn->client_lock);
1746 switch (stid->sc_type) {
1748 stp = openlockstateid(stid);
1749 mutex_lock_nested(&stp->st_mutex,
1750 OPEN_STATEID_MUTEX);
1752 spin_lock(&clp->cl_lock);
1753 if (stid->sc_status == 0) {
1755 SC_STATUS_ADMIN_REVOKED;
1756 atomic_inc(&clp->cl_admin_revoked);
1757 spin_unlock(&clp->cl_lock);
1758 release_all_access(stp);
1760 spin_unlock(&clp->cl_lock);
1761 mutex_unlock(&stp->st_mutex);
1764 stp = openlockstateid(stid);
1765 mutex_lock_nested(&stp->st_mutex,
1766 LOCK_STATEID_MUTEX);
1767 spin_lock(&clp->cl_lock);
1768 if (stid->sc_status == 0) {
1769 struct nfs4_lockowner *lo =
1770 lockowner(stp->st_stateowner);
1771 struct nfsd_file *nf;
1774 SC_STATUS_ADMIN_REVOKED;
1775 atomic_inc(&clp->cl_admin_revoked);
1776 spin_unlock(&clp->cl_lock);
1777 nf = find_any_file(stp->st_stid.sc_file);
1779 get_file(nf->nf_file);
1780 filp_close(nf->nf_file,
1784 release_all_access(stp);
1786 spin_unlock(&clp->cl_lock);
1787 mutex_unlock(&stp->st_mutex);
1790 dp = delegstateid(stid);
1791 spin_lock(&state_lock);
1792 if (!unhash_delegation_locked(
1793 dp, SC_STATUS_ADMIN_REVOKED))
1795 spin_unlock(&state_lock);
1797 revoke_delegation(dp);
1799 case SC_TYPE_LAYOUT:
1800 ls = layoutstateid(stid);
1801 nfsd4_close_layout(ls);
1804 nfs4_put_stid(stid);
1805 spin_lock(&nn->client_lock);
1806 if (clp->cl_minorversion == 0)
1807 /* Allow cleanup after a lease period.
1808 * store_release ensures cleanup will
1809 * see any newly revoked states if it
1810 * sees the time updated.
1812 nn->nfs40_last_revoke =
1813 ktime_get_boottime_seconds();
1818 spin_unlock(&nn->client_lock);
1822 hash_sessionid(struct nfs4_sessionid *sessionid)
1824 struct nfsd4_sessionid *sid = (struct nfsd4_sessionid *)sessionid;
1826 return sid->sequence % SESSION_HASH_SIZE;
1829 #ifdef CONFIG_SUNRPC_DEBUG
1831 dump_sessionid(const char *fn, struct nfs4_sessionid *sessionid)
1833 u32 *ptr = (u32 *)(&sessionid->data[0]);
1834 dprintk("%s: %u:%u:%u:%u\n", fn, ptr[0], ptr[1], ptr[2], ptr[3]);
1838 dump_sessionid(const char *fn, struct nfs4_sessionid *sessionid)
1844 * Bump the seqid on cstate->replay_owner, and clear replay_owner if it
1845 * won't be used for replay.
1847 void nfsd4_bump_seqid(struct nfsd4_compound_state *cstate, __be32 nfserr)
1849 struct nfs4_stateowner *so = cstate->replay_owner;
1851 if (nfserr == nfserr_replay_me)
1854 if (!seqid_mutating_err(ntohl(nfserr))) {
1855 nfsd4_cstate_clear_replay(cstate);
1860 if (so->so_is_open_owner)
1861 release_last_closed_stateid(openowner(so));
1867 gen_sessionid(struct nfsd4_session *ses)
1869 struct nfs4_client *clp = ses->se_client;
1870 struct nfsd4_sessionid *sid;
1872 sid = (struct nfsd4_sessionid *)ses->se_sessionid.data;
1873 sid->clientid = clp->cl_clientid;
1874 sid->sequence = current_sessionid++;
1879 * The protocol defines ca_maxresponssize_cached to include the size of
1880 * the rpc header, but all we need to cache is the data starting after
1881 * the end of the initial SEQUENCE operation--the rest we regenerate
1882 * each time. Therefore we can advertise a ca_maxresponssize_cached
1883 * value that is the number of bytes in our cache plus a few additional
1884 * bytes. In order to stay on the safe side, and not promise more than
1885 * we can cache, those additional bytes must be the minimum possible: 24
1886 * bytes of rpc header (xid through accept state, with AUTH_NULL
1887 * verifier), 12 for the compound header (with zero-length tag), and 44
1888 * for the SEQUENCE op response:
1890 #define NFSD_MIN_HDR_SEQ_SZ (24 + 12 + 44)
1893 free_session_slots(struct nfsd4_session *ses)
1897 for (i = 0; i < ses->se_fchannel.maxreqs; i++) {
1898 free_svc_cred(&ses->se_slots[i]->sl_cred);
1899 kfree(ses->se_slots[i]);
1904 * We don't actually need to cache the rpc and session headers, so we
1905 * can allocate a little less for each slot:
1907 static inline u32 slot_bytes(struct nfsd4_channel_attrs *ca)
1911 if (ca->maxresp_cached < NFSD_MIN_HDR_SEQ_SZ)
1914 size = ca->maxresp_cached - NFSD_MIN_HDR_SEQ_SZ;
1915 return size + sizeof(struct nfsd4_slot);
1919 * XXX: If we run out of reserved DRC memory we could (up to a point)
1920 * re-negotiate active sessions and reduce their slot usage to make
1921 * room for new connections. For now we just fail the create session.
1923 static u32 nfsd4_get_drc_mem(struct nfsd4_channel_attrs *ca, struct nfsd_net *nn)
1925 u32 slotsize = slot_bytes(ca);
1926 u32 num = ca->maxreqs;
1927 unsigned long avail, total_avail;
1928 unsigned int scale_factor;
1930 spin_lock(&nfsd_drc_lock);
1931 if (nfsd_drc_max_mem > nfsd_drc_mem_used)
1932 total_avail = nfsd_drc_max_mem - nfsd_drc_mem_used;
1934 /* We have handed out more space than we chose in
1935 * set_max_drc() to allow. That isn't really a
1936 * problem as long as that doesn't make us think we
1937 * have lots more due to integer overflow.
1940 avail = min((unsigned long)NFSD_MAX_MEM_PER_SESSION, total_avail);
1942 * Never use more than a fraction of the remaining memory,
1943 * unless it's the only way to give this client a slot.
1944 * The chosen fraction is either 1/8 or 1/number of threads,
1945 * whichever is smaller. This ensures there are adequate
1946 * slots to support multiple clients per thread.
1947 * Give the client one slot even if that would require
1948 * over-allocation--it is better than failure.
1950 scale_factor = max_t(unsigned int, 8, nn->nfsd_serv->sv_nrthreads);
1952 avail = clamp_t(unsigned long, avail, slotsize,
1953 total_avail/scale_factor);
1954 num = min_t(int, num, avail / slotsize);
1955 num = max_t(int, num, 1);
1956 nfsd_drc_mem_used += num * slotsize;
1957 spin_unlock(&nfsd_drc_lock);
1962 static void nfsd4_put_drc_mem(struct nfsd4_channel_attrs *ca)
1964 int slotsize = slot_bytes(ca);
1966 spin_lock(&nfsd_drc_lock);
1967 nfsd_drc_mem_used -= slotsize * ca->maxreqs;
1968 spin_unlock(&nfsd_drc_lock);
1971 static struct nfsd4_session *alloc_session(struct nfsd4_channel_attrs *fattrs,
1972 struct nfsd4_channel_attrs *battrs)
1974 int numslots = fattrs->maxreqs;
1975 int slotsize = slot_bytes(fattrs);
1976 struct nfsd4_session *new;
1979 BUILD_BUG_ON(struct_size(new, se_slots, NFSD_MAX_SLOTS_PER_SESSION)
1982 new = kzalloc(struct_size(new, se_slots, numslots), GFP_KERNEL);
1985 /* allocate each struct nfsd4_slot and data cache in one piece */
1986 for (i = 0; i < numslots; i++) {
1987 new->se_slots[i] = kzalloc(slotsize, GFP_KERNEL);
1988 if (!new->se_slots[i])
1992 memcpy(&new->se_fchannel, fattrs, sizeof(struct nfsd4_channel_attrs));
1993 memcpy(&new->se_bchannel, battrs, sizeof(struct nfsd4_channel_attrs));
1998 kfree(new->se_slots[i]);
2003 static void free_conn(struct nfsd4_conn *c)
2005 svc_xprt_put(c->cn_xprt);
2009 static void nfsd4_conn_lost(struct svc_xpt_user *u)
2011 struct nfsd4_conn *c = container_of(u, struct nfsd4_conn, cn_xpt_user);
2012 struct nfs4_client *clp = c->cn_session->se_client;
2014 trace_nfsd_cb_lost(clp);
2016 spin_lock(&clp->cl_lock);
2017 if (!list_empty(&c->cn_persession)) {
2018 list_del(&c->cn_persession);
2021 nfsd4_probe_callback(clp);
2022 spin_unlock(&clp->cl_lock);
2025 static struct nfsd4_conn *alloc_conn(struct svc_rqst *rqstp, u32 flags)
2027 struct nfsd4_conn *conn;
2029 conn = kmalloc(sizeof(struct nfsd4_conn), GFP_KERNEL);
2032 svc_xprt_get(rqstp->rq_xprt);
2033 conn->cn_xprt = rqstp->rq_xprt;
2034 conn->cn_flags = flags;
2035 INIT_LIST_HEAD(&conn->cn_xpt_user.list);
2039 static void __nfsd4_hash_conn(struct nfsd4_conn *conn, struct nfsd4_session *ses)
2041 conn->cn_session = ses;
2042 list_add(&conn->cn_persession, &ses->se_conns);
2045 static void nfsd4_hash_conn(struct nfsd4_conn *conn, struct nfsd4_session *ses)
2047 struct nfs4_client *clp = ses->se_client;
2049 spin_lock(&clp->cl_lock);
2050 __nfsd4_hash_conn(conn, ses);
2051 spin_unlock(&clp->cl_lock);
2054 static int nfsd4_register_conn(struct nfsd4_conn *conn)
2056 conn->cn_xpt_user.callback = nfsd4_conn_lost;
2057 return register_xpt_user(conn->cn_xprt, &conn->cn_xpt_user);
2060 static void nfsd4_init_conn(struct svc_rqst *rqstp, struct nfsd4_conn *conn, struct nfsd4_session *ses)
2064 nfsd4_hash_conn(conn, ses);
2065 ret = nfsd4_register_conn(conn);
2067 /* oops; xprt is already down: */
2068 nfsd4_conn_lost(&conn->cn_xpt_user);
2069 /* We may have gained or lost a callback channel: */
2070 nfsd4_probe_callback_sync(ses->se_client);
2073 static struct nfsd4_conn *alloc_conn_from_crses(struct svc_rqst *rqstp, struct nfsd4_create_session *cses)
2075 u32 dir = NFS4_CDFC4_FORE;
2077 if (cses->flags & SESSION4_BACK_CHAN)
2078 dir |= NFS4_CDFC4_BACK;
2079 return alloc_conn(rqstp, dir);
2082 /* must be called under client_lock */
2083 static void nfsd4_del_conns(struct nfsd4_session *s)
2085 struct nfs4_client *clp = s->se_client;
2086 struct nfsd4_conn *c;
2088 spin_lock(&clp->cl_lock);
2089 while (!list_empty(&s->se_conns)) {
2090 c = list_first_entry(&s->se_conns, struct nfsd4_conn, cn_persession);
2091 list_del_init(&c->cn_persession);
2092 spin_unlock(&clp->cl_lock);
2094 unregister_xpt_user(c->cn_xprt, &c->cn_xpt_user);
2097 spin_lock(&clp->cl_lock);
2099 spin_unlock(&clp->cl_lock);
2102 static void __free_session(struct nfsd4_session *ses)
2104 free_session_slots(ses);
2108 static void free_session(struct nfsd4_session *ses)
2110 nfsd4_del_conns(ses);
2111 nfsd4_put_drc_mem(&ses->se_fchannel);
2112 __free_session(ses);
2115 static void init_session(struct svc_rqst *rqstp, struct nfsd4_session *new, struct nfs4_client *clp, struct nfsd4_create_session *cses)
2118 struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
2120 new->se_client = clp;
2123 INIT_LIST_HEAD(&new->se_conns);
2125 new->se_cb_seq_nr = 1;
2126 new->se_flags = cses->flags;
2127 new->se_cb_prog = cses->callback_prog;
2128 new->se_cb_sec = cses->cb_sec;
2129 atomic_set(&new->se_ref, 0);
2130 idx = hash_sessionid(&new->se_sessionid);
2131 list_add(&new->se_hash, &nn->sessionid_hashtbl[idx]);
2132 spin_lock(&clp->cl_lock);
2133 list_add(&new->se_perclnt, &clp->cl_sessions);
2134 spin_unlock(&clp->cl_lock);
2137 struct sockaddr *sa = svc_addr(rqstp);
2139 * This is a little silly; with sessions there's no real
2140 * use for the callback address. Use the peer address
2141 * as a reasonable default for now, but consider fixing
2142 * the rpc client not to require an address in the
2145 rpc_copy_addr((struct sockaddr *)&clp->cl_cb_conn.cb_addr, sa);
2146 clp->cl_cb_conn.cb_addrlen = svc_addr_len(sa);
2150 /* caller must hold client_lock */
2151 static struct nfsd4_session *
2152 __find_in_sessionid_hashtbl(struct nfs4_sessionid *sessionid, struct net *net)
2154 struct nfsd4_session *elem;
2156 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
2158 lockdep_assert_held(&nn->client_lock);
2160 dump_sessionid(__func__, sessionid);
2161 idx = hash_sessionid(sessionid);
2162 /* Search in the appropriate list */
2163 list_for_each_entry(elem, &nn->sessionid_hashtbl[idx], se_hash) {
2164 if (!memcmp(elem->se_sessionid.data, sessionid->data,
2165 NFS4_MAX_SESSIONID_LEN)) {
2170 dprintk("%s: session not found\n", __func__);
2174 static struct nfsd4_session *
2175 find_in_sessionid_hashtbl(struct nfs4_sessionid *sessionid, struct net *net,
2178 struct nfsd4_session *session;
2179 __be32 status = nfserr_badsession;
2181 session = __find_in_sessionid_hashtbl(sessionid, net);
2184 status = nfsd4_get_session_locked(session);
2192 /* caller must hold client_lock */
2194 unhash_session(struct nfsd4_session *ses)
2196 struct nfs4_client *clp = ses->se_client;
2197 struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
2199 lockdep_assert_held(&nn->client_lock);
2201 list_del(&ses->se_hash);
2202 spin_lock(&ses->se_client->cl_lock);
2203 list_del(&ses->se_perclnt);
2204 spin_unlock(&ses->se_client->cl_lock);
2207 /* SETCLIENTID and SETCLIENTID_CONFIRM Helper functions */
2209 STALE_CLIENTID(clientid_t *clid, struct nfsd_net *nn)
2212 * We're assuming the clid was not given out from a boot
2213 * precisely 2^32 (about 136 years) before this one. That seems
2214 * a safe assumption:
2216 if (clid->cl_boot == (u32)nn->boot_time)
2218 trace_nfsd_clid_stale(clid);
2223 * XXX Should we use a slab cache ?
2224 * This type of memory management is somewhat inefficient, but we use it
2225 * anyway since SETCLIENTID is not a common operation.
2227 static struct nfs4_client *alloc_client(struct xdr_netobj name,
2228 struct nfsd_net *nn)
2230 struct nfs4_client *clp;
2233 if (atomic_read(&nn->nfs4_client_count) >= nn->nfs4_max_clients) {
2234 mod_delayed_work(laundry_wq, &nn->laundromat_work, 0);
2237 clp = kmem_cache_zalloc(client_slab, GFP_KERNEL);
2240 xdr_netobj_dup(&clp->cl_name, &name, GFP_KERNEL);
2241 if (clp->cl_name.data == NULL)
2243 clp->cl_ownerstr_hashtbl = kmalloc_array(OWNER_HASH_SIZE,
2244 sizeof(struct list_head),
2246 if (!clp->cl_ownerstr_hashtbl)
2247 goto err_no_hashtbl;
2248 for (i = 0; i < OWNER_HASH_SIZE; i++)
2249 INIT_LIST_HEAD(&clp->cl_ownerstr_hashtbl[i]);
2250 INIT_LIST_HEAD(&clp->cl_sessions);
2251 idr_init(&clp->cl_stateids);
2252 atomic_set(&clp->cl_rpc_users, 0);
2253 clp->cl_cb_state = NFSD4_CB_UNKNOWN;
2254 clp->cl_state = NFSD4_ACTIVE;
2255 atomic_inc(&nn->nfs4_client_count);
2256 atomic_set(&clp->cl_delegs_in_recall, 0);
2257 INIT_LIST_HEAD(&clp->cl_idhash);
2258 INIT_LIST_HEAD(&clp->cl_openowners);
2259 INIT_LIST_HEAD(&clp->cl_delegations);
2260 INIT_LIST_HEAD(&clp->cl_lru);
2261 INIT_LIST_HEAD(&clp->cl_revoked);
2262 #ifdef CONFIG_NFSD_PNFS
2263 INIT_LIST_HEAD(&clp->cl_lo_states);
2265 INIT_LIST_HEAD(&clp->async_copies);
2266 spin_lock_init(&clp->async_lock);
2267 spin_lock_init(&clp->cl_lock);
2268 rpc_init_wait_queue(&clp->cl_cb_waitq, "Backchannel slot table");
2271 kfree(clp->cl_name.data);
2273 kmem_cache_free(client_slab, clp);
2277 static void __free_client(struct kref *k)
2279 struct nfsdfs_client *c = container_of(k, struct nfsdfs_client, cl_ref);
2280 struct nfs4_client *clp = container_of(c, struct nfs4_client, cl_nfsdfs);
2282 free_svc_cred(&clp->cl_cred);
2283 kfree(clp->cl_ownerstr_hashtbl);
2284 kfree(clp->cl_name.data);
2285 kfree(clp->cl_nii_domain.data);
2286 kfree(clp->cl_nii_name.data);
2287 idr_destroy(&clp->cl_stateids);
2289 kmem_cache_free(client_slab, clp);
2292 static void drop_client(struct nfs4_client *clp)
2294 kref_put(&clp->cl_nfsdfs.cl_ref, __free_client);
2298 free_client(struct nfs4_client *clp)
2300 while (!list_empty(&clp->cl_sessions)) {
2301 struct nfsd4_session *ses;
2302 ses = list_entry(clp->cl_sessions.next, struct nfsd4_session,
2304 list_del(&ses->se_perclnt);
2305 WARN_ON_ONCE(atomic_read(&ses->se_ref));
2308 rpc_destroy_wait_queue(&clp->cl_cb_waitq);
2309 if (clp->cl_nfsd_dentry) {
2310 nfsd_client_rmdir(clp->cl_nfsd_dentry);
2311 clp->cl_nfsd_dentry = NULL;
2312 wake_up_all(&expiry_wq);
2317 /* must be called under the client_lock */
2319 unhash_client_locked(struct nfs4_client *clp)
2321 struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
2322 struct nfsd4_session *ses;
2324 lockdep_assert_held(&nn->client_lock);
2326 /* Mark the client as expired! */
2328 /* Make it invisible */
2329 if (!list_empty(&clp->cl_idhash)) {
2330 list_del_init(&clp->cl_idhash);
2331 if (test_bit(NFSD4_CLIENT_CONFIRMED, &clp->cl_flags))
2332 rb_erase(&clp->cl_namenode, &nn->conf_name_tree);
2334 rb_erase(&clp->cl_namenode, &nn->unconf_name_tree);
2336 list_del_init(&clp->cl_lru);
2337 spin_lock(&clp->cl_lock);
2338 list_for_each_entry(ses, &clp->cl_sessions, se_perclnt)
2339 list_del_init(&ses->se_hash);
2340 spin_unlock(&clp->cl_lock);
2344 unhash_client(struct nfs4_client *clp)
2346 struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
2348 spin_lock(&nn->client_lock);
2349 unhash_client_locked(clp);
2350 spin_unlock(&nn->client_lock);
2353 static __be32 mark_client_expired_locked(struct nfs4_client *clp)
2355 if (atomic_read(&clp->cl_rpc_users))
2356 return nfserr_jukebox;
2357 unhash_client_locked(clp);
2362 __destroy_client(struct nfs4_client *clp)
2364 struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
2366 struct nfs4_openowner *oo;
2367 struct nfs4_delegation *dp;
2368 struct list_head reaplist;
2370 INIT_LIST_HEAD(&reaplist);
2371 spin_lock(&state_lock);
2372 while (!list_empty(&clp->cl_delegations)) {
2373 dp = list_entry(clp->cl_delegations.next, struct nfs4_delegation, dl_perclnt);
2374 unhash_delegation_locked(dp, SC_STATUS_CLOSED);
2375 list_add(&dp->dl_recall_lru, &reaplist);
2377 spin_unlock(&state_lock);
2378 while (!list_empty(&reaplist)) {
2379 dp = list_entry(reaplist.next, struct nfs4_delegation, dl_recall_lru);
2380 list_del_init(&dp->dl_recall_lru);
2381 destroy_unhashed_deleg(dp);
2383 while (!list_empty(&clp->cl_revoked)) {
2384 dp = list_entry(clp->cl_revoked.next, struct nfs4_delegation, dl_recall_lru);
2385 list_del_init(&dp->dl_recall_lru);
2386 nfs4_put_stid(&dp->dl_stid);
2388 while (!list_empty(&clp->cl_openowners)) {
2389 oo = list_entry(clp->cl_openowners.next, struct nfs4_openowner, oo_perclient);
2390 nfs4_get_stateowner(&oo->oo_owner);
2391 release_openowner(oo);
2393 for (i = 0; i < OWNER_HASH_SIZE; i++) {
2394 struct nfs4_stateowner *so, *tmp;
2396 list_for_each_entry_safe(so, tmp, &clp->cl_ownerstr_hashtbl[i],
2398 /* Should be no openowners at this point */
2399 WARN_ON_ONCE(so->so_is_open_owner);
2400 remove_blocked_locks(lockowner(so));
2403 nfsd4_return_all_client_layouts(clp);
2404 nfsd4_shutdown_copy(clp);
2405 nfsd4_shutdown_callback(clp);
2406 if (clp->cl_cb_conn.cb_xprt)
2407 svc_xprt_put(clp->cl_cb_conn.cb_xprt);
2408 atomic_add_unless(&nn->nfs4_client_count, -1, 0);
2409 nfsd4_dec_courtesy_client_count(nn, clp);
2411 wake_up_all(&expiry_wq);
2415 destroy_client(struct nfs4_client *clp)
2418 __destroy_client(clp);
2421 static void inc_reclaim_complete(struct nfs4_client *clp)
2423 struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
2425 if (!nn->track_reclaim_completes)
2427 if (!nfsd4_find_reclaim_client(clp->cl_name, nn))
2429 if (atomic_inc_return(&nn->nr_reclaim_complete) ==
2430 nn->reclaim_str_hashtbl_size) {
2431 printk(KERN_INFO "NFSD: all clients done reclaiming, ending NFSv4 grace period (net %x)\n",
2433 nfsd4_end_grace(nn);
2437 static void expire_client(struct nfs4_client *clp)
2440 nfsd4_client_record_remove(clp);
2441 __destroy_client(clp);
2444 static void copy_verf(struct nfs4_client *target, nfs4_verifier *source)
2446 memcpy(target->cl_verifier.data, source->data,
2447 sizeof(target->cl_verifier.data));
2450 static void copy_clid(struct nfs4_client *target, struct nfs4_client *source)
2452 target->cl_clientid.cl_boot = source->cl_clientid.cl_boot;
2453 target->cl_clientid.cl_id = source->cl_clientid.cl_id;
2456 static int copy_cred(struct svc_cred *target, struct svc_cred *source)
2458 target->cr_principal = kstrdup(source->cr_principal, GFP_KERNEL);
2459 target->cr_raw_principal = kstrdup(source->cr_raw_principal,
2461 target->cr_targ_princ = kstrdup(source->cr_targ_princ, GFP_KERNEL);
2462 if ((source->cr_principal && !target->cr_principal) ||
2463 (source->cr_raw_principal && !target->cr_raw_principal) ||
2464 (source->cr_targ_princ && !target->cr_targ_princ))
2467 target->cr_flavor = source->cr_flavor;
2468 target->cr_uid = source->cr_uid;
2469 target->cr_gid = source->cr_gid;
2470 target->cr_group_info = source->cr_group_info;
2471 get_group_info(target->cr_group_info);
2472 target->cr_gss_mech = source->cr_gss_mech;
2473 if (source->cr_gss_mech)
2474 gss_mech_get(source->cr_gss_mech);
2479 compare_blob(const struct xdr_netobj *o1, const struct xdr_netobj *o2)
2481 if (o1->len < o2->len)
2483 if (o1->len > o2->len)
2485 return memcmp(o1->data, o2->data, o1->len);
2489 same_verf(nfs4_verifier *v1, nfs4_verifier *v2)
2491 return 0 == memcmp(v1->data, v2->data, sizeof(v1->data));
2495 same_clid(clientid_t *cl1, clientid_t *cl2)
2497 return (cl1->cl_boot == cl2->cl_boot) && (cl1->cl_id == cl2->cl_id);
2500 static bool groups_equal(struct group_info *g1, struct group_info *g2)
2504 if (g1->ngroups != g2->ngroups)
2506 for (i=0; i<g1->ngroups; i++)
2507 if (!gid_eq(g1->gid[i], g2->gid[i]))
2513 * RFC 3530 language requires clid_inuse be returned when the
2514 * "principal" associated with a requests differs from that previously
2515 * used. We use uid, gid's, and gss principal string as our best
2516 * approximation. We also don't want to allow non-gss use of a client
2517 * established using gss: in theory cr_principal should catch that
2518 * change, but in practice cr_principal can be null even in the gss case
2519 * since gssd doesn't always pass down a principal string.
2521 static bool is_gss_cred(struct svc_cred *cr)
2523 /* Is cr_flavor one of the gss "pseudoflavors"?: */
2524 return (cr->cr_flavor > RPC_AUTH_MAXFLAVOR);
2529 same_creds(struct svc_cred *cr1, struct svc_cred *cr2)
2531 if ((is_gss_cred(cr1) != is_gss_cred(cr2))
2532 || (!uid_eq(cr1->cr_uid, cr2->cr_uid))
2533 || (!gid_eq(cr1->cr_gid, cr2->cr_gid))
2534 || !groups_equal(cr1->cr_group_info, cr2->cr_group_info))
2536 /* XXX: check that cr_targ_princ fields match ? */
2537 if (cr1->cr_principal == cr2->cr_principal)
2539 if (!cr1->cr_principal || !cr2->cr_principal)
2541 return 0 == strcmp(cr1->cr_principal, cr2->cr_principal);
2544 static bool svc_rqst_integrity_protected(struct svc_rqst *rqstp)
2546 struct svc_cred *cr = &rqstp->rq_cred;
2549 if (!cr->cr_gss_mech)
2551 service = gss_pseudoflavor_to_service(cr->cr_gss_mech, cr->cr_flavor);
2552 return service == RPC_GSS_SVC_INTEGRITY ||
2553 service == RPC_GSS_SVC_PRIVACY;
2556 bool nfsd4_mach_creds_match(struct nfs4_client *cl, struct svc_rqst *rqstp)
2558 struct svc_cred *cr = &rqstp->rq_cred;
2560 if (!cl->cl_mach_cred)
2562 if (cl->cl_cred.cr_gss_mech != cr->cr_gss_mech)
2564 if (!svc_rqst_integrity_protected(rqstp))
2566 if (cl->cl_cred.cr_raw_principal)
2567 return 0 == strcmp(cl->cl_cred.cr_raw_principal,
2568 cr->cr_raw_principal);
2569 if (!cr->cr_principal)
2571 return 0 == strcmp(cl->cl_cred.cr_principal, cr->cr_principal);
2574 static void gen_confirm(struct nfs4_client *clp, struct nfsd_net *nn)
2579 * This is opaque to client, so no need to byte-swap. Use
2580 * __force to keep sparse happy
2582 verf[0] = (__force __be32)(u32)ktime_get_real_seconds();
2583 verf[1] = (__force __be32)nn->clverifier_counter++;
2584 memcpy(clp->cl_confirm.data, verf, sizeof(clp->cl_confirm.data));
2587 static void gen_clid(struct nfs4_client *clp, struct nfsd_net *nn)
2589 clp->cl_clientid.cl_boot = (u32)nn->boot_time;
2590 clp->cl_clientid.cl_id = nn->clientid_counter++;
2591 gen_confirm(clp, nn);
2594 static struct nfs4_stid *
2595 find_stateid_locked(struct nfs4_client *cl, stateid_t *t)
2597 struct nfs4_stid *ret;
2599 ret = idr_find(&cl->cl_stateids, t->si_opaque.so_id);
2600 if (!ret || !ret->sc_type)
2605 static struct nfs4_stid *
2606 find_stateid_by_type(struct nfs4_client *cl, stateid_t *t,
2607 unsigned short typemask, unsigned short ok_states)
2609 struct nfs4_stid *s;
2611 spin_lock(&cl->cl_lock);
2612 s = find_stateid_locked(cl, t);
2614 if ((s->sc_status & ~ok_states) == 0 &&
2615 (typemask & s->sc_type))
2616 refcount_inc(&s->sc_count);
2620 spin_unlock(&cl->cl_lock);
2624 static struct nfs4_client *get_nfsdfs_clp(struct inode *inode)
2626 struct nfsdfs_client *nc;
2627 nc = get_nfsdfs_client(inode);
2630 return container_of(nc, struct nfs4_client, cl_nfsdfs);
2633 static void seq_quote_mem(struct seq_file *m, char *data, int len)
2636 seq_escape_mem(m, data, len, ESCAPE_HEX | ESCAPE_NAP | ESCAPE_APPEND, "\"\\");
2640 static const char *cb_state2str(int state)
2645 case NFSD4_CB_UNKNOWN:
2649 case NFSD4_CB_FAULT:
2655 static int client_info_show(struct seq_file *m, void *v)
2657 struct inode *inode = file_inode(m->file);
2658 struct nfs4_client *clp;
2661 clp = get_nfsdfs_clp(inode);
2664 memcpy(&clid, &clp->cl_clientid, sizeof(clid));
2665 seq_printf(m, "clientid: 0x%llx\n", clid);
2666 seq_printf(m, "address: \"%pISpc\"\n", (struct sockaddr *)&clp->cl_addr);
2668 if (clp->cl_state == NFSD4_COURTESY)
2669 seq_puts(m, "status: courtesy\n");
2670 else if (clp->cl_state == NFSD4_EXPIRABLE)
2671 seq_puts(m, "status: expirable\n");
2672 else if (test_bit(NFSD4_CLIENT_CONFIRMED, &clp->cl_flags))
2673 seq_puts(m, "status: confirmed\n");
2675 seq_puts(m, "status: unconfirmed\n");
2676 seq_printf(m, "seconds from last renew: %lld\n",
2677 ktime_get_boottime_seconds() - clp->cl_time);
2678 seq_puts(m, "name: ");
2679 seq_quote_mem(m, clp->cl_name.data, clp->cl_name.len);
2680 seq_printf(m, "\nminor version: %d\n", clp->cl_minorversion);
2681 if (clp->cl_nii_domain.data) {
2682 seq_puts(m, "Implementation domain: ");
2683 seq_quote_mem(m, clp->cl_nii_domain.data,
2684 clp->cl_nii_domain.len);
2685 seq_puts(m, "\nImplementation name: ");
2686 seq_quote_mem(m, clp->cl_nii_name.data, clp->cl_nii_name.len);
2687 seq_printf(m, "\nImplementation time: [%lld, %ld]\n",
2688 clp->cl_nii_time.tv_sec, clp->cl_nii_time.tv_nsec);
2690 seq_printf(m, "callback state: %s\n", cb_state2str(clp->cl_cb_state));
2691 seq_printf(m, "callback address: %pISpc\n", &clp->cl_cb_conn.cb_addr);
2692 seq_printf(m, "admin-revoked states: %d\n",
2693 atomic_read(&clp->cl_admin_revoked));
2699 DEFINE_SHOW_ATTRIBUTE(client_info);
2701 static void *states_start(struct seq_file *s, loff_t *pos)
2702 __acquires(&clp->cl_lock)
2704 struct nfs4_client *clp = s->private;
2705 unsigned long id = *pos;
2708 spin_lock(&clp->cl_lock);
2709 ret = idr_get_next_ul(&clp->cl_stateids, &id);
2714 static void *states_next(struct seq_file *s, void *v, loff_t *pos)
2716 struct nfs4_client *clp = s->private;
2717 unsigned long id = *pos;
2722 ret = idr_get_next_ul(&clp->cl_stateids, &id);
2727 static void states_stop(struct seq_file *s, void *v)
2728 __releases(&clp->cl_lock)
2730 struct nfs4_client *clp = s->private;
2732 spin_unlock(&clp->cl_lock);
2735 static void nfs4_show_fname(struct seq_file *s, struct nfsd_file *f)
2737 seq_printf(s, "filename: \"%pD2\"", f->nf_file);
2740 static void nfs4_show_superblock(struct seq_file *s, struct nfsd_file *f)
2742 struct inode *inode = file_inode(f->nf_file);
2744 seq_printf(s, "superblock: \"%02x:%02x:%ld\"",
2745 MAJOR(inode->i_sb->s_dev),
2746 MINOR(inode->i_sb->s_dev),
2750 static void nfs4_show_owner(struct seq_file *s, struct nfs4_stateowner *oo)
2752 seq_puts(s, "owner: ");
2753 seq_quote_mem(s, oo->so_owner.data, oo->so_owner.len);
2756 static void nfs4_show_stateid(struct seq_file *s, stateid_t *stid)
2758 seq_printf(s, "0x%.8x", stid->si_generation);
2759 seq_printf(s, "%12phN", &stid->si_opaque);
2762 static int nfs4_show_open(struct seq_file *s, struct nfs4_stid *st)
2764 struct nfs4_ol_stateid *ols;
2765 struct nfs4_file *nf;
2766 struct nfsd_file *file;
2767 struct nfs4_stateowner *oo;
2768 unsigned int access, deny;
2770 ols = openlockstateid(st);
2771 oo = ols->st_stateowner;
2775 nfs4_show_stateid(s, &st->sc_stateid);
2776 seq_puts(s, ": { type: open, ");
2778 access = bmap_to_share_mode(ols->st_access_bmap);
2779 deny = bmap_to_share_mode(ols->st_deny_bmap);
2781 seq_printf(s, "access: %s%s, ",
2782 access & NFS4_SHARE_ACCESS_READ ? "r" : "-",
2783 access & NFS4_SHARE_ACCESS_WRITE ? "w" : "-");
2784 seq_printf(s, "deny: %s%s, ",
2785 deny & NFS4_SHARE_ACCESS_READ ? "r" : "-",
2786 deny & NFS4_SHARE_ACCESS_WRITE ? "w" : "-");
2788 spin_lock(&nf->fi_lock);
2789 file = find_any_file_locked(nf);
2791 nfs4_show_superblock(s, file);
2793 nfs4_show_fname(s, file);
2796 spin_unlock(&nf->fi_lock);
2797 nfs4_show_owner(s, oo);
2798 if (st->sc_status & SC_STATUS_ADMIN_REVOKED)
2799 seq_puts(s, ", admin-revoked");
2800 seq_puts(s, " }\n");
2804 static int nfs4_show_lock(struct seq_file *s, struct nfs4_stid *st)
2806 struct nfs4_ol_stateid *ols;
2807 struct nfs4_file *nf;
2808 struct nfsd_file *file;
2809 struct nfs4_stateowner *oo;
2811 ols = openlockstateid(st);
2812 oo = ols->st_stateowner;
2816 nfs4_show_stateid(s, &st->sc_stateid);
2817 seq_puts(s, ": { type: lock, ");
2819 spin_lock(&nf->fi_lock);
2820 file = find_any_file_locked(nf);
2823 * Note: a lock stateid isn't really the same thing as a lock,
2824 * it's the locking state held by one owner on a file, and there
2825 * may be multiple (or no) lock ranges associated with it.
2826 * (Same for the matter is true of open stateids.)
2829 nfs4_show_superblock(s, file);
2830 /* XXX: open stateid? */
2832 nfs4_show_fname(s, file);
2835 nfs4_show_owner(s, oo);
2836 if (st->sc_status & SC_STATUS_ADMIN_REVOKED)
2837 seq_puts(s, ", admin-revoked");
2838 seq_puts(s, " }\n");
2839 spin_unlock(&nf->fi_lock);
2843 static int nfs4_show_deleg(struct seq_file *s, struct nfs4_stid *st)
2845 struct nfs4_delegation *ds;
2846 struct nfs4_file *nf;
2847 struct nfsd_file *file;
2849 ds = delegstateid(st);
2853 nfs4_show_stateid(s, &st->sc_stateid);
2854 seq_puts(s, ": { type: deleg, ");
2856 seq_printf(s, "access: %s",
2857 ds->dl_type == NFS4_OPEN_DELEGATE_READ ? "r" : "w");
2859 /* XXX: lease time, whether it's being recalled. */
2861 spin_lock(&nf->fi_lock);
2862 file = nf->fi_deleg_file;
2865 nfs4_show_superblock(s, file);
2867 nfs4_show_fname(s, file);
2869 spin_unlock(&nf->fi_lock);
2870 if (st->sc_status & SC_STATUS_ADMIN_REVOKED)
2871 seq_puts(s, ", admin-revoked");
2872 seq_puts(s, " }\n");
2876 static int nfs4_show_layout(struct seq_file *s, struct nfs4_stid *st)
2878 struct nfs4_layout_stateid *ls;
2879 struct nfsd_file *file;
2881 ls = container_of(st, struct nfs4_layout_stateid, ls_stid);
2884 nfs4_show_stateid(s, &st->sc_stateid);
2885 seq_puts(s, ": { type: layout");
2887 /* XXX: What else would be useful? */
2889 spin_lock(&ls->ls_stid.sc_file->fi_lock);
2893 nfs4_show_superblock(s, file);
2895 nfs4_show_fname(s, file);
2897 spin_unlock(&ls->ls_stid.sc_file->fi_lock);
2898 if (st->sc_status & SC_STATUS_ADMIN_REVOKED)
2899 seq_puts(s, ", admin-revoked");
2900 seq_puts(s, " }\n");
2905 static int states_show(struct seq_file *s, void *v)
2907 struct nfs4_stid *st = v;
2909 switch (st->sc_type) {
2911 return nfs4_show_open(s, st);
2913 return nfs4_show_lock(s, st);
2915 return nfs4_show_deleg(s, st);
2916 case SC_TYPE_LAYOUT:
2917 return nfs4_show_layout(s, st);
2919 return 0; /* XXX: or SEQ_SKIP? */
2921 /* XXX: copy stateids? */
2924 static struct seq_operations states_seq_ops = {
2925 .start = states_start,
2926 .next = states_next,
2927 .stop = states_stop,
2931 static int client_states_open(struct inode *inode, struct file *file)
2934 struct nfs4_client *clp;
2937 clp = get_nfsdfs_clp(inode);
2941 ret = seq_open(file, &states_seq_ops);
2944 s = file->private_data;
2949 static int client_opens_release(struct inode *inode, struct file *file)
2951 struct seq_file *m = file->private_data;
2952 struct nfs4_client *clp = m->private;
2954 /* XXX: alternatively, we could get/drop in seq start/stop */
2956 return seq_release(inode, file);
2959 static const struct file_operations client_states_fops = {
2960 .open = client_states_open,
2962 .llseek = seq_lseek,
2963 .release = client_opens_release,
2967 * Normally we refuse to destroy clients that are in use, but here the
2968 * administrator is telling us to just do it. We also want to wait
2969 * so the caller has a guarantee that the client's locks are gone by
2970 * the time the write returns:
2972 static void force_expire_client(struct nfs4_client *clp)
2974 struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
2975 bool already_expired;
2977 trace_nfsd_clid_admin_expired(&clp->cl_clientid);
2979 spin_lock(&nn->client_lock);
2981 spin_unlock(&nn->client_lock);
2983 wait_event(expiry_wq, atomic_read(&clp->cl_rpc_users) == 0);
2984 spin_lock(&nn->client_lock);
2985 already_expired = list_empty(&clp->cl_lru);
2986 if (!already_expired)
2987 unhash_client_locked(clp);
2988 spin_unlock(&nn->client_lock);
2990 if (!already_expired)
2993 wait_event(expiry_wq, clp->cl_nfsd_dentry == NULL);
2996 static ssize_t client_ctl_write(struct file *file, const char __user *buf,
2997 size_t size, loff_t *pos)
3000 struct nfs4_client *clp;
3002 data = simple_transaction_get(file, buf, size);
3004 return PTR_ERR(data);
3005 if (size != 7 || 0 != memcmp(data, "expire\n", 7))
3007 clp = get_nfsdfs_clp(file_inode(file));
3010 force_expire_client(clp);
3015 static const struct file_operations client_ctl_fops = {
3016 .write = client_ctl_write,
3017 .release = simple_transaction_release,
3020 static const struct tree_descr client_files[] = {
3021 [0] = {"info", &client_info_fops, S_IRUSR},
3022 [1] = {"states", &client_states_fops, S_IRUSR},
3023 [2] = {"ctl", &client_ctl_fops, S_IWUSR},
3028 nfsd4_cb_recall_any_done(struct nfsd4_callback *cb,
3029 struct rpc_task *task)
3031 trace_nfsd_cb_recall_any_done(cb, task);
3032 switch (task->tk_status) {
3033 case -NFS4ERR_DELAY:
3034 rpc_delay(task, 2 * HZ);
3042 nfsd4_cb_recall_any_release(struct nfsd4_callback *cb)
3044 struct nfs4_client *clp = cb->cb_clp;
3046 clear_bit(NFSD4_CLIENT_CB_RECALL_ANY, &clp->cl_flags);
3051 nfsd4_cb_getattr_done(struct nfsd4_callback *cb, struct rpc_task *task)
3053 struct nfs4_cb_fattr *ncf =
3054 container_of(cb, struct nfs4_cb_fattr, ncf_getattr);
3056 ncf->ncf_cb_status = task->tk_status;
3057 switch (task->tk_status) {
3058 case -NFS4ERR_DELAY:
3059 rpc_delay(task, 2 * HZ);
3067 nfsd4_cb_getattr_release(struct nfsd4_callback *cb)
3069 struct nfs4_cb_fattr *ncf =
3070 container_of(cb, struct nfs4_cb_fattr, ncf_getattr);
3071 struct nfs4_delegation *dp =
3072 container_of(ncf, struct nfs4_delegation, dl_cb_fattr);
3074 nfs4_put_stid(&dp->dl_stid);
3075 clear_bit(CB_GETATTR_BUSY, &ncf->ncf_cb_flags);
3076 wake_up_bit(&ncf->ncf_cb_flags, CB_GETATTR_BUSY);
3079 static const struct nfsd4_callback_ops nfsd4_cb_recall_any_ops = {
3080 .done = nfsd4_cb_recall_any_done,
3081 .release = nfsd4_cb_recall_any_release,
3084 static const struct nfsd4_callback_ops nfsd4_cb_getattr_ops = {
3085 .done = nfsd4_cb_getattr_done,
3086 .release = nfsd4_cb_getattr_release,
3089 static void nfs4_cb_getattr(struct nfs4_cb_fattr *ncf)
3091 struct nfs4_delegation *dp =
3092 container_of(ncf, struct nfs4_delegation, dl_cb_fattr);
3094 if (test_and_set_bit(CB_GETATTR_BUSY, &ncf->ncf_cb_flags))
3096 /* set to proper status when nfsd4_cb_getattr_done runs */
3097 ncf->ncf_cb_status = NFS4ERR_IO;
3099 refcount_inc(&dp->dl_stid.sc_count);
3100 nfsd4_run_cb(&ncf->ncf_getattr);
3103 static struct nfs4_client *create_client(struct xdr_netobj name,
3104 struct svc_rqst *rqstp, nfs4_verifier *verf)
3106 struct nfs4_client *clp;
3107 struct sockaddr *sa = svc_addr(rqstp);
3109 struct net *net = SVC_NET(rqstp);
3110 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
3111 struct dentry *dentries[ARRAY_SIZE(client_files)];
3113 clp = alloc_client(name, nn);
3117 ret = copy_cred(&clp->cl_cred, &rqstp->rq_cred);
3123 kref_init(&clp->cl_nfsdfs.cl_ref);
3124 nfsd4_init_cb(&clp->cl_cb_null, clp, NULL, NFSPROC4_CLNT_CB_NULL);
3125 clp->cl_time = ktime_get_boottime_seconds();
3126 clear_bit(0, &clp->cl_cb_slot_busy);
3127 copy_verf(clp, verf);
3128 memcpy(&clp->cl_addr, sa, sizeof(struct sockaddr_storage));
3129 clp->cl_cb_session = NULL;
3131 clp->cl_nfsd_dentry = nfsd_client_mkdir(
3132 nn, &clp->cl_nfsdfs,
3133 clp->cl_clientid.cl_id - nn->clientid_base,
3134 client_files, dentries);
3135 clp->cl_nfsd_info_dentry = dentries[0];
3136 if (!clp->cl_nfsd_dentry) {
3140 clp->cl_ra = kzalloc(sizeof(*clp->cl_ra), GFP_KERNEL);
3145 clp->cl_ra_time = 0;
3146 nfsd4_init_cb(&clp->cl_ra->ra_cb, clp, &nfsd4_cb_recall_any_ops,
3147 NFSPROC4_CLNT_CB_RECALL_ANY);
3152 add_clp_to_name_tree(struct nfs4_client *new_clp, struct rb_root *root)
3154 struct rb_node **new = &(root->rb_node), *parent = NULL;
3155 struct nfs4_client *clp;
3158 clp = rb_entry(*new, struct nfs4_client, cl_namenode);
3161 if (compare_blob(&clp->cl_name, &new_clp->cl_name) > 0)
3162 new = &((*new)->rb_left);
3164 new = &((*new)->rb_right);
3167 rb_link_node(&new_clp->cl_namenode, parent, new);
3168 rb_insert_color(&new_clp->cl_namenode, root);
3171 static struct nfs4_client *
3172 find_clp_in_name_tree(struct xdr_netobj *name, struct rb_root *root)
3175 struct rb_node *node = root->rb_node;
3176 struct nfs4_client *clp;
3179 clp = rb_entry(node, struct nfs4_client, cl_namenode);
3180 cmp = compare_blob(&clp->cl_name, name);
3182 node = node->rb_left;
3184 node = node->rb_right;
3192 add_to_unconfirmed(struct nfs4_client *clp)
3194 unsigned int idhashval;
3195 struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
3197 lockdep_assert_held(&nn->client_lock);
3199 clear_bit(NFSD4_CLIENT_CONFIRMED, &clp->cl_flags);
3200 add_clp_to_name_tree(clp, &nn->unconf_name_tree);
3201 idhashval = clientid_hashval(clp->cl_clientid.cl_id);
3202 list_add(&clp->cl_idhash, &nn->unconf_id_hashtbl[idhashval]);
3203 renew_client_locked(clp);
3207 move_to_confirmed(struct nfs4_client *clp)
3209 unsigned int idhashval = clientid_hashval(clp->cl_clientid.cl_id);
3210 struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
3212 lockdep_assert_held(&nn->client_lock);
3214 list_move(&clp->cl_idhash, &nn->conf_id_hashtbl[idhashval]);
3215 rb_erase(&clp->cl_namenode, &nn->unconf_name_tree);
3216 add_clp_to_name_tree(clp, &nn->conf_name_tree);
3217 set_bit(NFSD4_CLIENT_CONFIRMED, &clp->cl_flags);
3218 trace_nfsd_clid_confirmed(&clp->cl_clientid);
3219 renew_client_locked(clp);
3222 static struct nfs4_client *
3223 find_client_in_id_table(struct list_head *tbl, clientid_t *clid, bool sessions)
3225 struct nfs4_client *clp;
3226 unsigned int idhashval = clientid_hashval(clid->cl_id);
3228 list_for_each_entry(clp, &tbl[idhashval], cl_idhash) {
3229 if (same_clid(&clp->cl_clientid, clid)) {
3230 if ((bool)clp->cl_minorversion != sessions)
3232 renew_client_locked(clp);
3239 static struct nfs4_client *
3240 find_confirmed_client(clientid_t *clid, bool sessions, struct nfsd_net *nn)
3242 struct list_head *tbl = nn->conf_id_hashtbl;
3244 lockdep_assert_held(&nn->client_lock);
3245 return find_client_in_id_table(tbl, clid, sessions);
3248 static struct nfs4_client *
3249 find_unconfirmed_client(clientid_t *clid, bool sessions, struct nfsd_net *nn)
3251 struct list_head *tbl = nn->unconf_id_hashtbl;
3253 lockdep_assert_held(&nn->client_lock);
3254 return find_client_in_id_table(tbl, clid, sessions);
3257 static bool clp_used_exchangeid(struct nfs4_client *clp)
3259 return clp->cl_exchange_flags != 0;
3262 static struct nfs4_client *
3263 find_confirmed_client_by_name(struct xdr_netobj *name, struct nfsd_net *nn)
3265 lockdep_assert_held(&nn->client_lock);
3266 return find_clp_in_name_tree(name, &nn->conf_name_tree);
3269 static struct nfs4_client *
3270 find_unconfirmed_client_by_name(struct xdr_netobj *name, struct nfsd_net *nn)
3272 lockdep_assert_held(&nn->client_lock);
3273 return find_clp_in_name_tree(name, &nn->unconf_name_tree);
3277 gen_callback(struct nfs4_client *clp, struct nfsd4_setclientid *se, struct svc_rqst *rqstp)
3279 struct nfs4_cb_conn *conn = &clp->cl_cb_conn;
3280 struct sockaddr *sa = svc_addr(rqstp);
3281 u32 scopeid = rpc_get_scope_id(sa);
3282 unsigned short expected_family;
3284 /* Currently, we only support tcp and tcp6 for the callback channel */
3285 if (se->se_callback_netid_len == 3 &&
3286 !memcmp(se->se_callback_netid_val, "tcp", 3))
3287 expected_family = AF_INET;
3288 else if (se->se_callback_netid_len == 4 &&
3289 !memcmp(se->se_callback_netid_val, "tcp6", 4))
3290 expected_family = AF_INET6;
3294 conn->cb_addrlen = rpc_uaddr2sockaddr(clp->net, se->se_callback_addr_val,
3295 se->se_callback_addr_len,
3296 (struct sockaddr *)&conn->cb_addr,
3297 sizeof(conn->cb_addr));
3299 if (!conn->cb_addrlen || conn->cb_addr.ss_family != expected_family)
3302 if (conn->cb_addr.ss_family == AF_INET6)
3303 ((struct sockaddr_in6 *)&conn->cb_addr)->sin6_scope_id = scopeid;
3305 conn->cb_prog = se->se_callback_prog;
3306 conn->cb_ident = se->se_callback_ident;
3307 memcpy(&conn->cb_saddr, &rqstp->rq_daddr, rqstp->rq_daddrlen);
3308 trace_nfsd_cb_args(clp, conn);
3311 conn->cb_addr.ss_family = AF_UNSPEC;
3312 conn->cb_addrlen = 0;
3313 trace_nfsd_cb_nodelegs(clp);
3318 * Cache a reply. nfsd4_check_resp_size() has bounded the cache size.
3321 nfsd4_store_cache_entry(struct nfsd4_compoundres *resp)
3323 struct xdr_buf *buf = resp->xdr->buf;
3324 struct nfsd4_slot *slot = resp->cstate.slot;
3327 dprintk("--> %s slot %p\n", __func__, slot);
3329 slot->sl_flags |= NFSD4_SLOT_INITIALIZED;
3330 slot->sl_opcnt = resp->opcnt;
3331 slot->sl_status = resp->cstate.status;
3332 free_svc_cred(&slot->sl_cred);
3333 copy_cred(&slot->sl_cred, &resp->rqstp->rq_cred);
3335 if (!nfsd4_cache_this(resp)) {
3336 slot->sl_flags &= ~NFSD4_SLOT_CACHED;
3339 slot->sl_flags |= NFSD4_SLOT_CACHED;
3341 base = resp->cstate.data_offset;
3342 slot->sl_datalen = buf->len - base;
3343 if (read_bytes_from_xdr_buf(buf, base, slot->sl_data, slot->sl_datalen))
3344 WARN(1, "%s: sessions DRC could not cache compound\n",
3350 * Encode the replay sequence operation from the slot values.
3351 * If cachethis is FALSE encode the uncached rep error on the next
3352 * operation which sets resp->p and increments resp->opcnt for
3353 * nfs4svc_encode_compoundres.
3357 nfsd4_enc_sequence_replay(struct nfsd4_compoundargs *args,
3358 struct nfsd4_compoundres *resp)
3360 struct nfsd4_op *op;
3361 struct nfsd4_slot *slot = resp->cstate.slot;
3363 /* Encode the replayed sequence operation */
3364 op = &args->ops[resp->opcnt - 1];
3365 nfsd4_encode_operation(resp, op);
3367 if (slot->sl_flags & NFSD4_SLOT_CACHED)
3369 if (args->opcnt == 1) {
3371 * The original operation wasn't a solo sequence--we
3372 * always cache those--so this retry must not match the
3375 op->status = nfserr_seq_false_retry;
3377 op = &args->ops[resp->opcnt++];
3378 op->status = nfserr_retry_uncached_rep;
3379 nfsd4_encode_operation(resp, op);
3385 * The sequence operation is not cached because we can use the slot and
3389 nfsd4_replay_cache_entry(struct nfsd4_compoundres *resp,
3390 struct nfsd4_sequence *seq)
3392 struct nfsd4_slot *slot = resp->cstate.slot;
3393 struct xdr_stream *xdr = resp->xdr;
3397 dprintk("--> %s slot %p\n", __func__, slot);
3399 status = nfsd4_enc_sequence_replay(resp->rqstp->rq_argp, resp);
3403 p = xdr_reserve_space(xdr, slot->sl_datalen);
3406 return nfserr_serverfault;
3408 xdr_encode_opaque_fixed(p, slot->sl_data, slot->sl_datalen);
3409 xdr_commit_encode(xdr);
3411 resp->opcnt = slot->sl_opcnt;
3412 return slot->sl_status;
3416 * Set the exchange_id flags returned by the server.
3419 nfsd4_set_ex_flags(struct nfs4_client *new, struct nfsd4_exchange_id *clid)
3421 #ifdef CONFIG_NFSD_PNFS
3422 new->cl_exchange_flags |= EXCHGID4_FLAG_USE_PNFS_MDS;
3424 new->cl_exchange_flags |= EXCHGID4_FLAG_USE_NON_PNFS;
3427 /* Referrals are supported, Migration is not. */
3428 new->cl_exchange_flags |= EXCHGID4_FLAG_SUPP_MOVED_REFER;
3430 /* set the wire flags to return to client. */
3431 clid->flags = new->cl_exchange_flags;
3434 static bool client_has_openowners(struct nfs4_client *clp)
3436 struct nfs4_openowner *oo;
3438 list_for_each_entry(oo, &clp->cl_openowners, oo_perclient) {
3439 if (!list_empty(&oo->oo_owner.so_stateids))
3445 static bool client_has_state(struct nfs4_client *clp)
3447 return client_has_openowners(clp)
3448 #ifdef CONFIG_NFSD_PNFS
3449 || !list_empty(&clp->cl_lo_states)
3451 || !list_empty(&clp->cl_delegations)
3452 || !list_empty(&clp->cl_sessions)
3453 || !list_empty(&clp->async_copies);
3456 static __be32 copy_impl_id(struct nfs4_client *clp,
3457 struct nfsd4_exchange_id *exid)
3459 if (!exid->nii_domain.data)
3461 xdr_netobj_dup(&clp->cl_nii_domain, &exid->nii_domain, GFP_KERNEL);
3462 if (!clp->cl_nii_domain.data)
3463 return nfserr_jukebox;
3464 xdr_netobj_dup(&clp->cl_nii_name, &exid->nii_name, GFP_KERNEL);
3465 if (!clp->cl_nii_name.data)
3466 return nfserr_jukebox;
3467 clp->cl_nii_time = exid->nii_time;
3472 nfsd4_exchange_id(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3473 union nfsd4_op_u *u)
3475 struct nfsd4_exchange_id *exid = &u->exchange_id;
3476 struct nfs4_client *conf, *new;
3477 struct nfs4_client *unconf = NULL;
3479 char addr_str[INET6_ADDRSTRLEN];
3480 nfs4_verifier verf = exid->verifier;
3481 struct sockaddr *sa = svc_addr(rqstp);
3482 bool update = exid->flags & EXCHGID4_FLAG_UPD_CONFIRMED_REC_A;
3483 struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
3485 rpc_ntop(sa, addr_str, sizeof(addr_str));
3486 dprintk("%s rqstp=%p exid=%p clname.len=%u clname.data=%p "
3487 "ip_addr=%s flags %x, spa_how %u\n",
3488 __func__, rqstp, exid, exid->clname.len, exid->clname.data,
3489 addr_str, exid->flags, exid->spa_how);
3491 if (exid->flags & ~EXCHGID4_FLAG_MASK_A)
3492 return nfserr_inval;
3494 new = create_client(exid->clname, rqstp, &verf);
3496 return nfserr_jukebox;
3497 status = copy_impl_id(new, exid);
3501 switch (exid->spa_how) {
3503 exid->spo_must_enforce[0] = 0;
3504 exid->spo_must_enforce[1] = (
3505 1 << (OP_BIND_CONN_TO_SESSION - 32) |
3506 1 << (OP_EXCHANGE_ID - 32) |
3507 1 << (OP_CREATE_SESSION - 32) |
3508 1 << (OP_DESTROY_SESSION - 32) |
3509 1 << (OP_DESTROY_CLIENTID - 32));
3511 exid->spo_must_allow[0] &= (1 << (OP_CLOSE) |
3512 1 << (OP_OPEN_DOWNGRADE) |
3514 1 << (OP_DELEGRETURN));
3516 exid->spo_must_allow[1] &= (
3517 1 << (OP_TEST_STATEID - 32) |
3518 1 << (OP_FREE_STATEID - 32));
3519 if (!svc_rqst_integrity_protected(rqstp)) {
3520 status = nfserr_inval;
3524 * Sometimes userspace doesn't give us a principal.
3525 * Which is a bug, really. Anyway, we can't enforce
3526 * MACH_CRED in that case, better to give up now:
3528 if (!new->cl_cred.cr_principal &&
3529 !new->cl_cred.cr_raw_principal) {
3530 status = nfserr_serverfault;
3533 new->cl_mach_cred = true;
3537 default: /* checked by xdr code */
3541 status = nfserr_encr_alg_unsupp;
3545 /* Cases below refer to rfc 5661 section 18.35.4: */
3546 spin_lock(&nn->client_lock);
3547 conf = find_confirmed_client_by_name(&exid->clname, nn);
3549 bool creds_match = same_creds(&conf->cl_cred, &rqstp->rq_cred);
3550 bool verfs_match = same_verf(&verf, &conf->cl_verifier);
3553 if (!clp_used_exchangeid(conf)) { /* buggy client */
3554 status = nfserr_inval;
3557 if (!nfsd4_mach_creds_match(conf, rqstp)) {
3558 status = nfserr_wrong_cred;
3561 if (!creds_match) { /* case 9 */
3562 status = nfserr_perm;
3565 if (!verfs_match) { /* case 8 */
3566 status = nfserr_not_same;
3570 exid->flags |= EXCHGID4_FLAG_CONFIRMED_R;
3571 trace_nfsd_clid_confirmed_r(conf);
3574 if (!creds_match) { /* case 3 */
3575 if (client_has_state(conf)) {
3576 status = nfserr_clid_inuse;
3577 trace_nfsd_clid_cred_mismatch(conf, rqstp);
3582 if (verfs_match) { /* case 2 */
3583 conf->cl_exchange_flags |= EXCHGID4_FLAG_CONFIRMED_R;
3584 trace_nfsd_clid_confirmed_r(conf);
3587 /* case 5, client reboot */
3588 trace_nfsd_clid_verf_mismatch(conf, rqstp, &verf);
3593 if (update) { /* case 7 */
3594 status = nfserr_noent;
3598 unconf = find_unconfirmed_client_by_name(&exid->clname, nn);
3599 if (unconf) /* case 4, possible retry or client restart */
3600 unhash_client_locked(unconf);
3602 /* case 1, new owner ID */
3603 trace_nfsd_clid_fresh(new);
3607 status = mark_client_expired_locked(conf);
3610 trace_nfsd_clid_replaced(&conf->cl_clientid);
3612 new->cl_minorversion = cstate->minorversion;
3613 new->cl_spo_must_allow.u.words[0] = exid->spo_must_allow[0];
3614 new->cl_spo_must_allow.u.words[1] = exid->spo_must_allow[1];
3616 /* Contrived initial CREATE_SESSION response */
3617 new->cl_cs_slot.sl_status = nfserr_seq_misordered;
3619 add_to_unconfirmed(new);
3622 exid->clientid.cl_boot = conf->cl_clientid.cl_boot;
3623 exid->clientid.cl_id = conf->cl_clientid.cl_id;
3625 exid->seqid = conf->cl_cs_slot.sl_seqid + 1;
3626 nfsd4_set_ex_flags(conf, exid);
3628 dprintk("nfsd4_exchange_id seqid %d flags %x\n",
3629 conf->cl_cs_slot.sl_seqid, conf->cl_exchange_flags);
3633 spin_unlock(&nn->client_lock);
3638 trace_nfsd_clid_expire_unconf(&unconf->cl_clientid);
3639 expire_client(unconf);
3645 check_slot_seqid(u32 seqid, u32 slot_seqid, int slot_inuse)
3647 dprintk("%s enter. seqid %d slot_seqid %d\n", __func__, seqid,
3650 /* The slot is in use, and no response has been sent. */
3652 if (seqid == slot_seqid)
3653 return nfserr_jukebox;
3655 return nfserr_seq_misordered;
3657 /* Note unsigned 32-bit arithmetic handles wraparound: */
3658 if (likely(seqid == slot_seqid + 1))
3660 if (seqid == slot_seqid)
3661 return nfserr_replay_cache;
3662 return nfserr_seq_misordered;
3666 * Cache the create session result into the create session single DRC
3667 * slot cache by saving the xdr structure. sl_seqid has been set.
3668 * Do this for solo or embedded create session operations.
3671 nfsd4_cache_create_session(struct nfsd4_create_session *cr_ses,
3672 struct nfsd4_clid_slot *slot, __be32 nfserr)
3674 slot->sl_status = nfserr;
3675 memcpy(&slot->sl_cr_ses, cr_ses, sizeof(*cr_ses));
3679 nfsd4_replay_create_session(struct nfsd4_create_session *cr_ses,
3680 struct nfsd4_clid_slot *slot)
3682 memcpy(cr_ses, &slot->sl_cr_ses, sizeof(*cr_ses));
3683 return slot->sl_status;
3686 #define NFSD_MIN_REQ_HDR_SEQ_SZ ((\
3687 2 * 2 + /* credential,verifier: AUTH_NULL, length 0 */ \
3688 1 + /* MIN tag is length with zero, only length */ \
3689 3 + /* version, opcount, opcode */ \
3690 XDR_QUADLEN(NFS4_MAX_SESSIONID_LEN) + \
3691 /* seqid, slotID, slotID, cache */ \
3692 4 ) * sizeof(__be32))
3694 #define NFSD_MIN_RESP_HDR_SEQ_SZ ((\
3695 2 + /* verifier: AUTH_NULL, length 0 */\
3697 1 + /* MIN tag is length with zero, only length */ \
3698 3 + /* opcount, opcode, opstatus*/ \
3699 XDR_QUADLEN(NFS4_MAX_SESSIONID_LEN) + \
3700 /* seqid, slotID, slotID, slotID, status */ \
3701 5 ) * sizeof(__be32))
3703 static __be32 check_forechannel_attrs(struct nfsd4_channel_attrs *ca, struct nfsd_net *nn)
3705 u32 maxrpc = nn->nfsd_serv->sv_max_mesg;
3707 if (ca->maxreq_sz < NFSD_MIN_REQ_HDR_SEQ_SZ)
3708 return nfserr_toosmall;
3709 if (ca->maxresp_sz < NFSD_MIN_RESP_HDR_SEQ_SZ)
3710 return nfserr_toosmall;
3711 ca->headerpadsz = 0;
3712 ca->maxreq_sz = min_t(u32, ca->maxreq_sz, maxrpc);
3713 ca->maxresp_sz = min_t(u32, ca->maxresp_sz, maxrpc);
3714 ca->maxops = min_t(u32, ca->maxops, NFSD_MAX_OPS_PER_COMPOUND);
3715 ca->maxresp_cached = min_t(u32, ca->maxresp_cached,
3716 NFSD_SLOT_CACHE_SIZE + NFSD_MIN_HDR_SEQ_SZ);
3717 ca->maxreqs = min_t(u32, ca->maxreqs, NFSD_MAX_SLOTS_PER_SESSION);
3719 * Note decreasing slot size below client's request may make it
3720 * difficult for client to function correctly, whereas
3721 * decreasing the number of slots will (just?) affect
3722 * performance. When short on memory we therefore prefer to
3723 * decrease number of slots instead of their size. Clients that
3724 * request larger slots than they need will get poor results:
3725 * Note that we always allow at least one slot, because our
3726 * accounting is soft and provides no guarantees either way.
3728 ca->maxreqs = nfsd4_get_drc_mem(ca, nn);
3734 * Server's NFSv4.1 backchannel support is AUTH_SYS-only for now.
3735 * These are based on similar macros in linux/sunrpc/msg_prot.h .
3737 #define RPC_MAX_HEADER_WITH_AUTH_SYS \
3738 (RPC_CALLHDRSIZE + 2 * (2 + UNX_CALLSLACK))
3740 #define RPC_MAX_REPHEADER_WITH_AUTH_SYS \
3741 (RPC_REPHDRSIZE + (2 + NUL_REPLYSLACK))
3743 #define NFSD_CB_MAX_REQ_SZ ((NFS4_enc_cb_recall_sz + \
3744 RPC_MAX_HEADER_WITH_AUTH_SYS) * sizeof(__be32))
3745 #define NFSD_CB_MAX_RESP_SZ ((NFS4_dec_cb_recall_sz + \
3746 RPC_MAX_REPHEADER_WITH_AUTH_SYS) * \
3749 static __be32 check_backchannel_attrs(struct nfsd4_channel_attrs *ca)
3751 ca->headerpadsz = 0;
3753 if (ca->maxreq_sz < NFSD_CB_MAX_REQ_SZ)
3754 return nfserr_toosmall;
3755 if (ca->maxresp_sz < NFSD_CB_MAX_RESP_SZ)
3756 return nfserr_toosmall;
3757 ca->maxresp_cached = 0;
3759 return nfserr_toosmall;
3764 static __be32 nfsd4_check_cb_sec(struct nfsd4_cb_sec *cbs)
3766 switch (cbs->flavor) {
3772 * GSS case: the spec doesn't allow us to return this
3773 * error. But it also doesn't allow us not to support
3775 * I'd rather this fail hard than return some error the
3776 * client might think it can already handle:
3778 return nfserr_encr_alg_unsupp;
3783 nfsd4_create_session(struct svc_rqst *rqstp,
3784 struct nfsd4_compound_state *cstate, union nfsd4_op_u *u)
3786 struct nfsd4_create_session *cr_ses = &u->create_session;
3787 struct sockaddr *sa = svc_addr(rqstp);
3788 struct nfs4_client *conf, *unconf;
3789 struct nfsd4_clid_slot *cs_slot;
3790 struct nfs4_client *old = NULL;
3791 struct nfsd4_session *new;
3792 struct nfsd4_conn *conn;
3794 struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
3796 if (cr_ses->flags & ~SESSION4_FLAG_MASK_A)
3797 return nfserr_inval;
3798 status = nfsd4_check_cb_sec(&cr_ses->cb_sec);
3801 status = check_forechannel_attrs(&cr_ses->fore_channel, nn);
3804 status = check_backchannel_attrs(&cr_ses->back_channel);
3806 goto out_release_drc_mem;
3807 status = nfserr_jukebox;
3808 new = alloc_session(&cr_ses->fore_channel, &cr_ses->back_channel);
3810 goto out_release_drc_mem;
3811 conn = alloc_conn_from_crses(rqstp, cr_ses);
3813 goto out_free_session;
3815 spin_lock(&nn->client_lock);
3817 /* RFC 8881 Section 18.36.4 Phase 1: Client record look-up. */
3818 unconf = find_unconfirmed_client(&cr_ses->clientid, true, nn);
3819 conf = find_confirmed_client(&cr_ses->clientid, true, nn);
3820 if (!conf && !unconf) {
3821 status = nfserr_stale_clientid;
3825 /* RFC 8881 Section 18.36.4 Phase 2: Sequence ID processing. */
3827 cs_slot = &conf->cl_cs_slot;
3829 cs_slot = &unconf->cl_cs_slot;
3830 status = check_slot_seqid(cr_ses->seqid, cs_slot->sl_seqid, 0);
3833 cs_slot->sl_seqid++;
3834 cr_ses->seqid = cs_slot->sl_seqid;
3836 case nfserr_replay_cache:
3837 status = nfsd4_replay_create_session(cr_ses, cs_slot);
3839 case nfserr_jukebox:
3840 /* The server MUST NOT cache NFS4ERR_DELAY */
3843 goto out_cache_error;
3846 /* RFC 8881 Section 18.36.4 Phase 3: Client ID confirmation. */
3848 status = nfserr_wrong_cred;
3849 if (!nfsd4_mach_creds_match(conf, rqstp))
3850 goto out_cache_error;
3852 status = nfserr_clid_inuse;
3853 if (!same_creds(&unconf->cl_cred, &rqstp->rq_cred) ||
3854 !rpc_cmp_addr(sa, (struct sockaddr *) &unconf->cl_addr)) {
3855 trace_nfsd_clid_cred_mismatch(unconf, rqstp);
3856 goto out_cache_error;
3858 status = nfserr_wrong_cred;
3859 if (!nfsd4_mach_creds_match(unconf, rqstp))
3860 goto out_cache_error;
3861 old = find_confirmed_client_by_name(&unconf->cl_name, nn);
3863 status = mark_client_expired_locked(old);
3865 goto out_expired_error;
3866 trace_nfsd_clid_replaced(&old->cl_clientid);
3868 move_to_confirmed(unconf);
3872 /* RFC 8881 Section 18.36.4 Phase 4: Session creation. */
3874 /* Persistent sessions are not supported */
3875 cr_ses->flags &= ~SESSION4_PERSIST;
3876 /* Upshifting from TCP to RDMA is not supported */
3877 cr_ses->flags &= ~SESSION4_RDMA;
3879 init_session(rqstp, new, conf, cr_ses);
3880 nfsd4_get_session_locked(new);
3882 memcpy(cr_ses->sessionid.data, new->se_sessionid.data,
3883 NFS4_MAX_SESSIONID_LEN);
3885 /* cache solo and embedded create sessions under the client_lock */
3886 nfsd4_cache_create_session(cr_ses, cs_slot, status);
3887 spin_unlock(&nn->client_lock);
3889 fsnotify_dentry(conf->cl_nfsd_info_dentry, FS_MODIFY);
3890 /* init connection and backchannel */
3891 nfsd4_init_conn(rqstp, conn, new);
3892 nfsd4_put_session(new);
3900 * Revert the slot seq_nr change so the server will process
3901 * the client's resend instead of returning a cached response.
3903 if (status == nfserr_jukebox) {
3904 cs_slot->sl_seqid--;
3905 cr_ses->seqid = cs_slot->sl_seqid;
3909 nfsd4_cache_create_session(cr_ses, cs_slot, status);
3911 spin_unlock(&nn->client_lock);
3916 __free_session(new);
3917 out_release_drc_mem:
3918 nfsd4_put_drc_mem(&cr_ses->fore_channel);
3922 static __be32 nfsd4_map_bcts_dir(u32 *dir)
3925 case NFS4_CDFC4_FORE:
3926 case NFS4_CDFC4_BACK:
3928 case NFS4_CDFC4_FORE_OR_BOTH:
3929 case NFS4_CDFC4_BACK_OR_BOTH:
3930 *dir = NFS4_CDFC4_BOTH;
3933 return nfserr_inval;
3936 __be32 nfsd4_backchannel_ctl(struct svc_rqst *rqstp,
3937 struct nfsd4_compound_state *cstate,
3938 union nfsd4_op_u *u)
3940 struct nfsd4_backchannel_ctl *bc = &u->backchannel_ctl;
3941 struct nfsd4_session *session = cstate->session;
3942 struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
3945 status = nfsd4_check_cb_sec(&bc->bc_cb_sec);
3948 spin_lock(&nn->client_lock);
3949 session->se_cb_prog = bc->bc_cb_program;
3950 session->se_cb_sec = bc->bc_cb_sec;
3951 spin_unlock(&nn->client_lock);
3953 nfsd4_probe_callback(session->se_client);
3958 static struct nfsd4_conn *__nfsd4_find_conn(struct svc_xprt *xpt, struct nfsd4_session *s)
3960 struct nfsd4_conn *c;
3962 list_for_each_entry(c, &s->se_conns, cn_persession) {
3963 if (c->cn_xprt == xpt) {
3970 static __be32 nfsd4_match_existing_connection(struct svc_rqst *rqst,
3971 struct nfsd4_session *session, u32 req, struct nfsd4_conn **conn)
3973 struct nfs4_client *clp = session->se_client;
3974 struct svc_xprt *xpt = rqst->rq_xprt;
3975 struct nfsd4_conn *c;
3978 /* Following the last paragraph of RFC 5661 Section 18.34.3: */
3979 spin_lock(&clp->cl_lock);
3980 c = __nfsd4_find_conn(xpt, session);
3982 status = nfserr_noent;
3983 else if (req == c->cn_flags)
3985 else if (req == NFS4_CDFC4_FORE_OR_BOTH &&
3986 c->cn_flags != NFS4_CDFC4_BACK)
3988 else if (req == NFS4_CDFC4_BACK_OR_BOTH &&
3989 c->cn_flags != NFS4_CDFC4_FORE)
3992 status = nfserr_inval;
3993 spin_unlock(&clp->cl_lock);
3994 if (status == nfs_ok && conn)
3999 __be32 nfsd4_bind_conn_to_session(struct svc_rqst *rqstp,
4000 struct nfsd4_compound_state *cstate,
4001 union nfsd4_op_u *u)
4003 struct nfsd4_bind_conn_to_session *bcts = &u->bind_conn_to_session;
4005 struct nfsd4_conn *conn;
4006 struct nfsd4_session *session;
4007 struct net *net = SVC_NET(rqstp);
4008 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
4010 if (!nfsd4_last_compound_op(rqstp))
4011 return nfserr_not_only_op;
4012 spin_lock(&nn->client_lock);
4013 session = find_in_sessionid_hashtbl(&bcts->sessionid, net, &status);
4014 spin_unlock(&nn->client_lock);
4016 goto out_no_session;
4017 status = nfserr_wrong_cred;
4018 if (!nfsd4_mach_creds_match(session->se_client, rqstp))
4020 status = nfsd4_match_existing_connection(rqstp, session,
4022 if (status == nfs_ok) {
4023 if (bcts->dir == NFS4_CDFC4_FORE_OR_BOTH ||
4024 bcts->dir == NFS4_CDFC4_BACK)
4025 conn->cn_flags |= NFS4_CDFC4_BACK;
4026 nfsd4_probe_callback(session->se_client);
4029 if (status == nfserr_inval)
4031 status = nfsd4_map_bcts_dir(&bcts->dir);
4034 conn = alloc_conn(rqstp, bcts->dir);
4035 status = nfserr_jukebox;
4038 nfsd4_init_conn(rqstp, conn, session);
4041 nfsd4_put_session(session);
4046 static bool nfsd4_compound_in_session(struct nfsd4_compound_state *cstate, struct nfs4_sessionid *sid)
4048 if (!cstate->session)
4050 return !memcmp(sid, &cstate->session->se_sessionid, sizeof(*sid));
4054 nfsd4_destroy_session(struct svc_rqst *r, struct nfsd4_compound_state *cstate,
4055 union nfsd4_op_u *u)
4057 struct nfs4_sessionid *sessionid = &u->destroy_session.sessionid;
4058 struct nfsd4_session *ses;
4060 int ref_held_by_me = 0;
4061 struct net *net = SVC_NET(r);
4062 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
4064 status = nfserr_not_only_op;
4065 if (nfsd4_compound_in_session(cstate, sessionid)) {
4066 if (!nfsd4_last_compound_op(r))
4070 dump_sessionid(__func__, sessionid);
4071 spin_lock(&nn->client_lock);
4072 ses = find_in_sessionid_hashtbl(sessionid, net, &status);
4074 goto out_client_lock;
4075 status = nfserr_wrong_cred;
4076 if (!nfsd4_mach_creds_match(ses->se_client, r))
4077 goto out_put_session;
4078 status = mark_session_dead_locked(ses, 1 + ref_held_by_me);
4080 goto out_put_session;
4081 unhash_session(ses);
4082 spin_unlock(&nn->client_lock);
4084 nfsd4_probe_callback_sync(ses->se_client);
4086 spin_lock(&nn->client_lock);
4089 nfsd4_put_session_locked(ses);
4091 spin_unlock(&nn->client_lock);
4096 static __be32 nfsd4_sequence_check_conn(struct nfsd4_conn *new, struct nfsd4_session *ses)
4098 struct nfs4_client *clp = ses->se_client;
4099 struct nfsd4_conn *c;
4100 __be32 status = nfs_ok;
4103 spin_lock(&clp->cl_lock);
4104 c = __nfsd4_find_conn(new->cn_xprt, ses);
4107 status = nfserr_conn_not_bound_to_session;
4108 if (clp->cl_mach_cred)
4110 __nfsd4_hash_conn(new, ses);
4111 spin_unlock(&clp->cl_lock);
4112 ret = nfsd4_register_conn(new);
4114 /* oops; xprt is already down: */
4115 nfsd4_conn_lost(&new->cn_xpt_user);
4118 spin_unlock(&clp->cl_lock);
4123 static bool nfsd4_session_too_many_ops(struct svc_rqst *rqstp, struct nfsd4_session *session)
4125 struct nfsd4_compoundargs *args = rqstp->rq_argp;
4127 return args->opcnt > session->se_fchannel.maxops;
4130 static bool nfsd4_request_too_big(struct svc_rqst *rqstp,
4131 struct nfsd4_session *session)
4133 struct xdr_buf *xb = &rqstp->rq_arg;
4135 return xb->len > session->se_fchannel.maxreq_sz;
4138 static bool replay_matches_cache(struct svc_rqst *rqstp,
4139 struct nfsd4_sequence *seq, struct nfsd4_slot *slot)
4141 struct nfsd4_compoundargs *argp = rqstp->rq_argp;
4143 if ((bool)(slot->sl_flags & NFSD4_SLOT_CACHETHIS) !=
4144 (bool)seq->cachethis)
4147 * If there's an error then the reply can have fewer ops than
4150 if (slot->sl_opcnt < argp->opcnt && !slot->sl_status)
4153 * But if we cached a reply with *more* ops than the call you're
4154 * sending us now, then this new call is clearly not really a
4155 * replay of the old one:
4157 if (slot->sl_opcnt > argp->opcnt)
4159 /* This is the only check explicitly called by spec: */
4160 if (!same_creds(&rqstp->rq_cred, &slot->sl_cred))
4163 * There may be more comparisons we could actually do, but the
4164 * spec doesn't require us to catch every case where the calls
4165 * don't match (that would require caching the call as well as
4166 * the reply), so we don't bother.
4172 nfsd4_sequence(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
4173 union nfsd4_op_u *u)
4175 struct nfsd4_sequence *seq = &u->sequence;
4176 struct nfsd4_compoundres *resp = rqstp->rq_resp;
4177 struct xdr_stream *xdr = resp->xdr;
4178 struct nfsd4_session *session;
4179 struct nfs4_client *clp;
4180 struct nfsd4_slot *slot;
4181 struct nfsd4_conn *conn;
4184 struct net *net = SVC_NET(rqstp);
4185 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
4187 if (resp->opcnt != 1)
4188 return nfserr_sequence_pos;
4191 * Will be either used or freed by nfsd4_sequence_check_conn
4194 conn = alloc_conn(rqstp, NFS4_CDFC4_FORE);
4196 return nfserr_jukebox;
4198 spin_lock(&nn->client_lock);
4199 session = find_in_sessionid_hashtbl(&seq->sessionid, net, &status);
4201 goto out_no_session;
4202 clp = session->se_client;
4204 status = nfserr_too_many_ops;
4205 if (nfsd4_session_too_many_ops(rqstp, session))
4206 goto out_put_session;
4208 status = nfserr_req_too_big;
4209 if (nfsd4_request_too_big(rqstp, session))
4210 goto out_put_session;
4212 status = nfserr_badslot;
4213 if (seq->slotid >= session->se_fchannel.maxreqs)
4214 goto out_put_session;
4216 slot = session->se_slots[seq->slotid];
4217 dprintk("%s: slotid %d\n", __func__, seq->slotid);
4219 /* We do not negotiate the number of slots yet, so set the
4220 * maxslots to the session maxreqs which is used to encode
4221 * sr_highest_slotid and the sr_target_slot id to maxslots */
4222 seq->maxslots = session->se_fchannel.maxreqs;
4224 status = check_slot_seqid(seq->seqid, slot->sl_seqid,
4225 slot->sl_flags & NFSD4_SLOT_INUSE);
4226 if (status == nfserr_replay_cache) {
4227 status = nfserr_seq_misordered;
4228 if (!(slot->sl_flags & NFSD4_SLOT_INITIALIZED))
4229 goto out_put_session;
4230 status = nfserr_seq_false_retry;
4231 if (!replay_matches_cache(rqstp, seq, slot))
4232 goto out_put_session;
4233 cstate->slot = slot;
4234 cstate->session = session;
4236 /* Return the cached reply status and set cstate->status
4237 * for nfsd4_proc_compound processing */
4238 status = nfsd4_replay_cache_entry(resp, seq);
4239 cstate->status = nfserr_replay_cache;
4243 goto out_put_session;
4245 status = nfsd4_sequence_check_conn(conn, session);
4248 goto out_put_session;
4250 buflen = (seq->cachethis) ?
4251 session->se_fchannel.maxresp_cached :
4252 session->se_fchannel.maxresp_sz;
4253 status = (seq->cachethis) ? nfserr_rep_too_big_to_cache :
4255 if (xdr_restrict_buflen(xdr, buflen - rqstp->rq_auth_slack))
4256 goto out_put_session;
4257 svc_reserve(rqstp, buflen);
4260 /* Success! bump slot seqid */
4261 slot->sl_seqid = seq->seqid;
4262 slot->sl_flags |= NFSD4_SLOT_INUSE;
4264 slot->sl_flags |= NFSD4_SLOT_CACHETHIS;
4266 slot->sl_flags &= ~NFSD4_SLOT_CACHETHIS;
4268 cstate->slot = slot;
4269 cstate->session = session;
4273 switch (clp->cl_cb_state) {
4275 seq->status_flags = SEQ4_STATUS_CB_PATH_DOWN;
4277 case NFSD4_CB_FAULT:
4278 seq->status_flags = SEQ4_STATUS_BACKCHANNEL_FAULT;
4281 seq->status_flags = 0;
4283 if (!list_empty(&clp->cl_revoked))
4284 seq->status_flags |= SEQ4_STATUS_RECALLABLE_STATE_REVOKED;
4285 if (atomic_read(&clp->cl_admin_revoked))
4286 seq->status_flags |= SEQ4_STATUS_ADMIN_STATE_REVOKED;
4287 trace_nfsd_seq4_status(rqstp, seq);
4291 spin_unlock(&nn->client_lock);
4294 nfsd4_put_session_locked(session);
4295 goto out_no_session;
4299 nfsd4_sequence_done(struct nfsd4_compoundres *resp)
4301 struct nfsd4_compound_state *cs = &resp->cstate;
4303 if (nfsd4_has_session(cs)) {
4304 if (cs->status != nfserr_replay_cache) {
4305 nfsd4_store_cache_entry(resp);
4306 cs->slot->sl_flags &= ~NFSD4_SLOT_INUSE;
4308 /* Drop session reference that was taken in nfsd4_sequence() */
4309 nfsd4_put_session(cs->session);
4311 put_client_renew(cs->clp);
4315 nfsd4_destroy_clientid(struct svc_rqst *rqstp,
4316 struct nfsd4_compound_state *cstate,
4317 union nfsd4_op_u *u)
4319 struct nfsd4_destroy_clientid *dc = &u->destroy_clientid;
4320 struct nfs4_client *conf, *unconf;
4321 struct nfs4_client *clp = NULL;
4323 struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
4325 spin_lock(&nn->client_lock);
4326 unconf = find_unconfirmed_client(&dc->clientid, true, nn);
4327 conf = find_confirmed_client(&dc->clientid, true, nn);
4328 WARN_ON_ONCE(conf && unconf);
4331 if (client_has_state(conf)) {
4332 status = nfserr_clientid_busy;
4335 status = mark_client_expired_locked(conf);
4342 status = nfserr_stale_clientid;
4345 if (!nfsd4_mach_creds_match(clp, rqstp)) {
4347 status = nfserr_wrong_cred;
4350 trace_nfsd_clid_destroyed(&clp->cl_clientid);
4351 unhash_client_locked(clp);
4353 spin_unlock(&nn->client_lock);
4360 nfsd4_reclaim_complete(struct svc_rqst *rqstp,
4361 struct nfsd4_compound_state *cstate, union nfsd4_op_u *u)
4363 struct nfsd4_reclaim_complete *rc = &u->reclaim_complete;
4364 struct nfs4_client *clp = cstate->clp;
4367 if (rc->rca_one_fs) {
4368 if (!cstate->current_fh.fh_dentry)
4369 return nfserr_nofilehandle;
4371 * We don't take advantage of the rca_one_fs case.
4372 * That's OK, it's optional, we can safely ignore it.
4377 status = nfserr_complete_already;
4378 if (test_and_set_bit(NFSD4_CLIENT_RECLAIM_COMPLETE, &clp->cl_flags))
4381 status = nfserr_stale_clientid;
4382 if (is_client_expired(clp))
4384 * The following error isn't really legal.
4385 * But we only get here if the client just explicitly
4386 * destroyed the client. Surely it no longer cares what
4387 * error it gets back on an operation for the dead
4393 trace_nfsd_clid_reclaim_complete(&clp->cl_clientid);
4394 nfsd4_client_record_create(clp);
4395 inc_reclaim_complete(clp);
4401 nfsd4_setclientid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
4402 union nfsd4_op_u *u)
4404 struct nfsd4_setclientid *setclid = &u->setclientid;
4405 struct xdr_netobj clname = setclid->se_name;
4406 nfs4_verifier clverifier = setclid->se_verf;
4407 struct nfs4_client *conf, *new;
4408 struct nfs4_client *unconf = NULL;
4410 struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
4412 new = create_client(clname, rqstp, &clverifier);
4414 return nfserr_jukebox;
4415 spin_lock(&nn->client_lock);
4416 conf = find_confirmed_client_by_name(&clname, nn);
4417 if (conf && client_has_state(conf)) {
4418 status = nfserr_clid_inuse;
4419 if (clp_used_exchangeid(conf))
4421 if (!same_creds(&conf->cl_cred, &rqstp->rq_cred)) {
4422 trace_nfsd_clid_cred_mismatch(conf, rqstp);
4426 unconf = find_unconfirmed_client_by_name(&clname, nn);
4428 unhash_client_locked(unconf);
4430 if (same_verf(&conf->cl_verifier, &clverifier)) {
4431 copy_clid(new, conf);
4432 gen_confirm(new, nn);
4434 trace_nfsd_clid_verf_mismatch(conf, rqstp,
4437 trace_nfsd_clid_fresh(new);
4438 new->cl_minorversion = 0;
4439 gen_callback(new, setclid, rqstp);
4440 add_to_unconfirmed(new);
4441 setclid->se_clientid.cl_boot = new->cl_clientid.cl_boot;
4442 setclid->se_clientid.cl_id = new->cl_clientid.cl_id;
4443 memcpy(setclid->se_confirm.data, new->cl_confirm.data, sizeof(setclid->se_confirm.data));
4447 spin_unlock(&nn->client_lock);
4451 trace_nfsd_clid_expire_unconf(&unconf->cl_clientid);
4452 expire_client(unconf);
4458 nfsd4_setclientid_confirm(struct svc_rqst *rqstp,
4459 struct nfsd4_compound_state *cstate,
4460 union nfsd4_op_u *u)
4462 struct nfsd4_setclientid_confirm *setclientid_confirm =
4463 &u->setclientid_confirm;
4464 struct nfs4_client *conf, *unconf;
4465 struct nfs4_client *old = NULL;
4466 nfs4_verifier confirm = setclientid_confirm->sc_confirm;
4467 clientid_t * clid = &setclientid_confirm->sc_clientid;
4469 struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
4471 if (STALE_CLIENTID(clid, nn))
4472 return nfserr_stale_clientid;
4474 spin_lock(&nn->client_lock);
4475 conf = find_confirmed_client(clid, false, nn);
4476 unconf = find_unconfirmed_client(clid, false, nn);
4478 * We try hard to give out unique clientid's, so if we get an
4479 * attempt to confirm the same clientid with a different cred,
4480 * the client may be buggy; this should never happen.
4482 * Nevertheless, RFC 7530 recommends INUSE for this case:
4484 status = nfserr_clid_inuse;
4485 if (unconf && !same_creds(&unconf->cl_cred, &rqstp->rq_cred)) {
4486 trace_nfsd_clid_cred_mismatch(unconf, rqstp);
4489 if (conf && !same_creds(&conf->cl_cred, &rqstp->rq_cred)) {
4490 trace_nfsd_clid_cred_mismatch(conf, rqstp);
4493 if (!unconf || !same_verf(&confirm, &unconf->cl_confirm)) {
4494 if (conf && same_verf(&confirm, &conf->cl_confirm)) {
4497 status = nfserr_stale_clientid;
4503 unhash_client_locked(old);
4504 nfsd4_change_callback(conf, &unconf->cl_cb_conn);
4506 old = find_confirmed_client_by_name(&unconf->cl_name, nn);
4508 status = nfserr_clid_inuse;
4509 if (client_has_state(old)
4510 && !same_creds(&unconf->cl_cred,
4515 status = mark_client_expired_locked(old);
4520 trace_nfsd_clid_replaced(&old->cl_clientid);
4522 move_to_confirmed(unconf);
4525 get_client_locked(conf);
4526 spin_unlock(&nn->client_lock);
4528 fsnotify_dentry(conf->cl_nfsd_info_dentry, FS_MODIFY);
4529 nfsd4_probe_callback(conf);
4530 spin_lock(&nn->client_lock);
4531 put_client_renew_locked(conf);
4533 spin_unlock(&nn->client_lock);
4539 static struct nfs4_file *nfsd4_alloc_file(void)
4541 return kmem_cache_alloc(file_slab, GFP_KERNEL);
4544 /* OPEN Share state helper functions */
4546 static void nfsd4_file_init(const struct svc_fh *fh, struct nfs4_file *fp)
4548 refcount_set(&fp->fi_ref, 1);
4549 spin_lock_init(&fp->fi_lock);
4550 INIT_LIST_HEAD(&fp->fi_stateids);
4551 INIT_LIST_HEAD(&fp->fi_delegations);
4552 INIT_LIST_HEAD(&fp->fi_clnt_odstate);
4553 fh_copy_shallow(&fp->fi_fhandle, &fh->fh_handle);
4554 fp->fi_deleg_file = NULL;
4555 fp->fi_had_conflict = false;
4556 fp->fi_share_deny = 0;
4557 memset(fp->fi_fds, 0, sizeof(fp->fi_fds));
4558 memset(fp->fi_access, 0, sizeof(fp->fi_access));
4559 fp->fi_aliased = false;
4560 fp->fi_inode = d_inode(fh->fh_dentry);
4561 #ifdef CONFIG_NFSD_PNFS
4562 INIT_LIST_HEAD(&fp->fi_lo_states);
4563 atomic_set(&fp->fi_lo_recalls, 0);
4568 nfsd4_free_slabs(void)
4570 kmem_cache_destroy(client_slab);
4571 kmem_cache_destroy(openowner_slab);
4572 kmem_cache_destroy(lockowner_slab);
4573 kmem_cache_destroy(file_slab);
4574 kmem_cache_destroy(stateid_slab);
4575 kmem_cache_destroy(deleg_slab);
4576 kmem_cache_destroy(odstate_slab);
4580 nfsd4_init_slabs(void)
4582 client_slab = KMEM_CACHE(nfs4_client, 0);
4583 if (client_slab == NULL)
4585 openowner_slab = KMEM_CACHE(nfs4_openowner, 0);
4586 if (openowner_slab == NULL)
4587 goto out_free_client_slab;
4588 lockowner_slab = KMEM_CACHE(nfs4_lockowner, 0);
4589 if (lockowner_slab == NULL)
4590 goto out_free_openowner_slab;
4591 file_slab = KMEM_CACHE(nfs4_file, 0);
4592 if (file_slab == NULL)
4593 goto out_free_lockowner_slab;
4594 stateid_slab = KMEM_CACHE(nfs4_ol_stateid, 0);
4595 if (stateid_slab == NULL)
4596 goto out_free_file_slab;
4597 deleg_slab = KMEM_CACHE(nfs4_delegation, 0);
4598 if (deleg_slab == NULL)
4599 goto out_free_stateid_slab;
4600 odstate_slab = KMEM_CACHE(nfs4_clnt_odstate, 0);
4601 if (odstate_slab == NULL)
4602 goto out_free_deleg_slab;
4605 out_free_deleg_slab:
4606 kmem_cache_destroy(deleg_slab);
4607 out_free_stateid_slab:
4608 kmem_cache_destroy(stateid_slab);
4610 kmem_cache_destroy(file_slab);
4611 out_free_lockowner_slab:
4612 kmem_cache_destroy(lockowner_slab);
4613 out_free_openowner_slab:
4614 kmem_cache_destroy(openowner_slab);
4615 out_free_client_slab:
4616 kmem_cache_destroy(client_slab);
4621 static unsigned long
4622 nfsd4_state_shrinker_count(struct shrinker *shrink, struct shrink_control *sc)
4625 struct nfsd_net *nn = shrink->private_data;
4627 count = atomic_read(&nn->nfsd_courtesy_clients);
4629 count = atomic_long_read(&num_delegations);
4631 queue_work(laundry_wq, &nn->nfsd_shrinker_work);
4632 return (unsigned long)count;
4635 static unsigned long
4636 nfsd4_state_shrinker_scan(struct shrinker *shrink, struct shrink_control *sc)
4642 nfsd4_init_leases_net(struct nfsd_net *nn)
4647 nn->nfsd4_lease = 90; /* default lease time */
4648 nn->nfsd4_grace = 90;
4649 nn->somebody_reclaimed = false;
4650 nn->track_reclaim_completes = false;
4651 nn->clverifier_counter = get_random_u32();
4652 nn->clientid_base = get_random_u32();
4653 nn->clientid_counter = nn->clientid_base + 1;
4654 nn->s2s_cp_cl_id = nn->clientid_counter++;
4656 atomic_set(&nn->nfs4_client_count, 0);
4658 max_clients = (u64)si.totalram * si.mem_unit / (1024 * 1024 * 1024);
4659 max_clients *= NFS4_CLIENTS_PER_GB;
4660 nn->nfs4_max_clients = max_t(int, max_clients, NFS4_CLIENTS_PER_GB);
4662 atomic_set(&nn->nfsd_courtesy_clients, 0);
4665 static void init_nfs4_replay(struct nfs4_replay *rp)
4667 rp->rp_status = nfserr_serverfault;
4669 rp->rp_buf = rp->rp_ibuf;
4670 mutex_init(&rp->rp_mutex);
4673 static void nfsd4_cstate_assign_replay(struct nfsd4_compound_state *cstate,
4674 struct nfs4_stateowner *so)
4676 if (!nfsd4_has_session(cstate)) {
4677 mutex_lock(&so->so_replay.rp_mutex);
4678 cstate->replay_owner = nfs4_get_stateowner(so);
4682 void nfsd4_cstate_clear_replay(struct nfsd4_compound_state *cstate)
4684 struct nfs4_stateowner *so = cstate->replay_owner;
4687 cstate->replay_owner = NULL;
4688 mutex_unlock(&so->so_replay.rp_mutex);
4689 nfs4_put_stateowner(so);
4693 static inline void *alloc_stateowner(struct kmem_cache *slab, struct xdr_netobj *owner, struct nfs4_client *clp)
4695 struct nfs4_stateowner *sop;
4697 sop = kmem_cache_alloc(slab, GFP_KERNEL);
4701 xdr_netobj_dup(&sop->so_owner, owner, GFP_KERNEL);
4702 if (!sop->so_owner.data) {
4703 kmem_cache_free(slab, sop);
4707 INIT_LIST_HEAD(&sop->so_stateids);
4708 sop->so_client = clp;
4709 init_nfs4_replay(&sop->so_replay);
4710 atomic_set(&sop->so_count, 1);
4714 static void hash_openowner(struct nfs4_openowner *oo, struct nfs4_client *clp, unsigned int strhashval)
4716 lockdep_assert_held(&clp->cl_lock);
4718 list_add(&oo->oo_owner.so_strhash,
4719 &clp->cl_ownerstr_hashtbl[strhashval]);
4720 list_add(&oo->oo_perclient, &clp->cl_openowners);
4723 static void nfs4_unhash_openowner(struct nfs4_stateowner *so)
4725 unhash_openowner_locked(openowner(so));
4728 static void nfs4_free_openowner(struct nfs4_stateowner *so)
4730 struct nfs4_openowner *oo = openowner(so);
4732 kmem_cache_free(openowner_slab, oo);
4735 static const struct nfs4_stateowner_operations openowner_ops = {
4736 .so_unhash = nfs4_unhash_openowner,
4737 .so_free = nfs4_free_openowner,
4740 static struct nfs4_ol_stateid *
4741 nfsd4_find_existing_open(struct nfs4_file *fp, struct nfsd4_open *open)
4743 struct nfs4_ol_stateid *local, *ret = NULL;
4744 struct nfs4_openowner *oo = open->op_openowner;
4746 lockdep_assert_held(&fp->fi_lock);
4748 list_for_each_entry(local, &fp->fi_stateids, st_perfile) {
4749 /* ignore lock owners */
4750 if (local->st_stateowner->so_is_open_owner == 0)
4752 if (local->st_stateowner != &oo->oo_owner)
4754 if (local->st_stid.sc_type == SC_TYPE_OPEN &&
4755 !local->st_stid.sc_status) {
4757 refcount_inc(&ret->st_stid.sc_count);
4764 static void nfsd4_drop_revoked_stid(struct nfs4_stid *s)
4765 __releases(&s->sc_client->cl_lock)
4767 struct nfs4_client *cl = s->sc_client;
4768 LIST_HEAD(reaplist);
4769 struct nfs4_ol_stateid *stp;
4770 struct nfs4_delegation *dp;
4773 switch (s->sc_type) {
4775 stp = openlockstateid(s);
4776 if (unhash_open_stateid(stp, &reaplist))
4777 put_ol_stateid_locked(stp, &reaplist);
4778 spin_unlock(&cl->cl_lock);
4779 free_ol_stateid_reaplist(&reaplist);
4782 stp = openlockstateid(s);
4783 unhashed = unhash_lock_stateid(stp);
4784 spin_unlock(&cl->cl_lock);
4789 dp = delegstateid(s);
4790 list_del_init(&dp->dl_recall_lru);
4791 spin_unlock(&cl->cl_lock);
4795 spin_unlock(&cl->cl_lock);
4799 static void nfsd40_drop_revoked_stid(struct nfs4_client *cl,
4802 /* NFSv4.0 has no way for the client to tell the server
4803 * that it can forget an admin-revoked stateid.
4804 * So we keep it around until the first time that the
4805 * client uses it, and drop it the first time
4806 * nfserr_admin_revoked is returned.
4807 * For v4.1 and later we wait until explicitly told
4808 * to free the stateid.
4810 if (cl->cl_minorversion == 0) {
4811 struct nfs4_stid *st;
4813 spin_lock(&cl->cl_lock);
4814 st = find_stateid_locked(cl, stid);
4816 nfsd4_drop_revoked_stid(st);
4818 spin_unlock(&cl->cl_lock);
4823 nfsd4_verify_open_stid(struct nfs4_stid *s)
4825 __be32 ret = nfs_ok;
4827 if (s->sc_status & SC_STATUS_ADMIN_REVOKED)
4828 ret = nfserr_admin_revoked;
4829 else if (s->sc_status & SC_STATUS_REVOKED)
4830 ret = nfserr_deleg_revoked;
4831 else if (s->sc_status & SC_STATUS_CLOSED)
4832 ret = nfserr_bad_stateid;
4836 /* Lock the stateid st_mutex, and deal with races with CLOSE */
4838 nfsd4_lock_ol_stateid(struct nfs4_ol_stateid *stp)
4842 mutex_lock_nested(&stp->st_mutex, LOCK_STATEID_MUTEX);
4843 ret = nfsd4_verify_open_stid(&stp->st_stid);
4844 if (ret == nfserr_admin_revoked)
4845 nfsd40_drop_revoked_stid(stp->st_stid.sc_client,
4846 &stp->st_stid.sc_stateid);
4849 mutex_unlock(&stp->st_mutex);
4853 static struct nfs4_ol_stateid *
4854 nfsd4_find_and_lock_existing_open(struct nfs4_file *fp, struct nfsd4_open *open)
4856 struct nfs4_ol_stateid *stp;
4858 spin_lock(&fp->fi_lock);
4859 stp = nfsd4_find_existing_open(fp, open);
4860 spin_unlock(&fp->fi_lock);
4861 if (!stp || nfsd4_lock_ol_stateid(stp) == nfs_ok)
4863 nfs4_put_stid(&stp->st_stid);
4868 static struct nfs4_openowner *
4869 alloc_init_open_stateowner(unsigned int strhashval, struct nfsd4_open *open,
4870 struct nfsd4_compound_state *cstate)
4872 struct nfs4_client *clp = cstate->clp;
4873 struct nfs4_openowner *oo, *ret;
4875 oo = alloc_stateowner(openowner_slab, &open->op_owner, clp);
4878 oo->oo_owner.so_ops = &openowner_ops;
4879 oo->oo_owner.so_is_open_owner = 1;
4880 oo->oo_owner.so_seqid = open->op_seqid;
4882 if (nfsd4_has_session(cstate))
4883 oo->oo_flags |= NFS4_OO_CONFIRMED;
4885 oo->oo_last_closed_stid = NULL;
4886 INIT_LIST_HEAD(&oo->oo_close_lru);
4887 spin_lock(&clp->cl_lock);
4888 ret = find_openstateowner_str_locked(strhashval, open, clp);
4890 hash_openowner(oo, clp, strhashval);
4893 nfs4_free_stateowner(&oo->oo_owner);
4895 spin_unlock(&clp->cl_lock);
4899 static struct nfs4_ol_stateid *
4900 init_open_stateid(struct nfs4_file *fp, struct nfsd4_open *open)
4903 struct nfs4_openowner *oo = open->op_openowner;
4904 struct nfs4_ol_stateid *retstp = NULL;
4905 struct nfs4_ol_stateid *stp;
4908 /* We are moving these outside of the spinlocks to avoid the warnings */
4909 mutex_init(&stp->st_mutex);
4910 mutex_lock_nested(&stp->st_mutex, OPEN_STATEID_MUTEX);
4913 spin_lock(&oo->oo_owner.so_client->cl_lock);
4914 spin_lock(&fp->fi_lock);
4916 retstp = nfsd4_find_existing_open(fp, open);
4920 open->op_stp = NULL;
4921 refcount_inc(&stp->st_stid.sc_count);
4922 stp->st_stid.sc_type = SC_TYPE_OPEN;
4923 INIT_LIST_HEAD(&stp->st_locks);
4924 stp->st_stateowner = nfs4_get_stateowner(&oo->oo_owner);
4926 stp->st_stid.sc_file = fp;
4927 stp->st_access_bmap = 0;
4928 stp->st_deny_bmap = 0;
4929 stp->st_openstp = NULL;
4930 list_add(&stp->st_perstateowner, &oo->oo_owner.so_stateids);
4931 list_add(&stp->st_perfile, &fp->fi_stateids);
4934 spin_unlock(&fp->fi_lock);
4935 spin_unlock(&oo->oo_owner.so_client->cl_lock);
4937 /* Handle races with CLOSE */
4938 if (nfsd4_lock_ol_stateid(retstp) != nfs_ok) {
4939 nfs4_put_stid(&retstp->st_stid);
4942 /* To keep mutex tracking happy */
4943 mutex_unlock(&stp->st_mutex);
4950 * In the 4.0 case we need to keep the owners around a little while to handle
4951 * CLOSE replay. We still do need to release any file access that is held by
4952 * them before returning however.
4955 move_to_close_lru(struct nfs4_ol_stateid *s, struct net *net)
4957 struct nfs4_ol_stateid *last;
4958 struct nfs4_openowner *oo = openowner(s->st_stateowner);
4959 struct nfsd_net *nn = net_generic(s->st_stid.sc_client->net,
4962 dprintk("NFSD: move_to_close_lru nfs4_openowner %p\n", oo);
4965 * We know that we hold one reference via nfsd4_close, and another
4966 * "persistent" reference for the client. If the refcount is higher
4967 * than 2, then there are still calls in progress that are using this
4968 * stateid. We can't put the sc_file reference until they are finished.
4969 * Wait for the refcount to drop to 2. Since it has been unhashed,
4970 * there should be no danger of the refcount going back up again at
4973 wait_event(close_wq, refcount_read(&s->st_stid.sc_count) == 2);
4975 release_all_access(s);
4976 if (s->st_stid.sc_file) {
4977 put_nfs4_file(s->st_stid.sc_file);
4978 s->st_stid.sc_file = NULL;
4981 spin_lock(&nn->client_lock);
4982 last = oo->oo_last_closed_stid;
4983 oo->oo_last_closed_stid = s;
4984 list_move_tail(&oo->oo_close_lru, &nn->close_lru);
4985 oo->oo_time = ktime_get_boottime_seconds();
4986 spin_unlock(&nn->client_lock);
4988 nfs4_put_stid(&last->st_stid);
4991 static noinline_for_stack struct nfs4_file *
4992 nfsd4_file_hash_lookup(const struct svc_fh *fhp)
4994 struct inode *inode = d_inode(fhp->fh_dentry);
4995 struct rhlist_head *tmp, *list;
4996 struct nfs4_file *fi;
4999 list = rhltable_lookup(&nfs4_file_rhltable, &inode,
5000 nfs4_file_rhash_params);
5001 rhl_for_each_entry_rcu(fi, tmp, list, fi_rlist) {
5002 if (fh_match(&fi->fi_fhandle, &fhp->fh_handle)) {
5003 if (refcount_inc_not_zero(&fi->fi_ref)) {
5014 * On hash insertion, identify entries with the same inode but
5015 * distinct filehandles. They will all be on the list returned
5016 * by rhltable_lookup().
5018 * inode->i_lock prevents racing insertions from adding an entry
5019 * for the same inode/fhp pair twice.
5021 static noinline_for_stack struct nfs4_file *
5022 nfsd4_file_hash_insert(struct nfs4_file *new, const struct svc_fh *fhp)
5024 struct inode *inode = d_inode(fhp->fh_dentry);
5025 struct rhlist_head *tmp, *list;
5026 struct nfs4_file *ret = NULL;
5027 bool alias_found = false;
5028 struct nfs4_file *fi;
5032 spin_lock(&inode->i_lock);
5034 list = rhltable_lookup(&nfs4_file_rhltable, &inode,
5035 nfs4_file_rhash_params);
5036 rhl_for_each_entry_rcu(fi, tmp, list, fi_rlist) {
5037 if (fh_match(&fi->fi_fhandle, &fhp->fh_handle)) {
5038 if (refcount_inc_not_zero(&fi->fi_ref))
5041 fi->fi_aliased = alias_found = true;
5046 nfsd4_file_init(fhp, new);
5047 err = rhltable_insert(&nfs4_file_rhltable, &new->fi_rlist,
5048 nfs4_file_rhash_params);
5052 new->fi_aliased = alias_found;
5056 spin_unlock(&inode->i_lock);
5061 static noinline_for_stack void nfsd4_file_hash_remove(struct nfs4_file *fi)
5063 rhltable_remove(&nfs4_file_rhltable, &fi->fi_rlist,
5064 nfs4_file_rhash_params);
5068 * Called to check deny when READ with all zero stateid or
5069 * WRITE with all zero or all one stateid
5072 nfs4_share_conflict(struct svc_fh *current_fh, unsigned int deny_type)
5074 struct nfs4_file *fp;
5075 __be32 ret = nfs_ok;
5077 fp = nfsd4_file_hash_lookup(current_fh);
5081 /* Check for conflicting share reservations */
5082 spin_lock(&fp->fi_lock);
5083 if (fp->fi_share_deny & deny_type)
5084 ret = nfserr_locked;
5085 spin_unlock(&fp->fi_lock);
5090 static bool nfsd4_deleg_present(const struct inode *inode)
5092 struct file_lock_context *ctx = locks_inode_context(inode);
5094 return ctx && !list_empty_careful(&ctx->flc_lease);
5098 * nfsd_wait_for_delegreturn - wait for delegations to be returned
5099 * @rqstp: the RPC transaction being executed
5100 * @inode: in-core inode of the file being waited for
5102 * The timeout prevents deadlock if all nfsd threads happen to be
5103 * tied up waiting for returning delegations.
5106 * %true: delegation was returned
5107 * %false: timed out waiting for delegreturn
5109 bool nfsd_wait_for_delegreturn(struct svc_rqst *rqstp, struct inode *inode)
5111 long __maybe_unused timeo;
5113 timeo = wait_var_event_timeout(inode, !nfsd4_deleg_present(inode),
5114 NFSD_DELEGRETURN_TIMEOUT);
5115 trace_nfsd_delegret_wakeup(rqstp, inode, timeo);
5119 static void nfsd4_cb_recall_prepare(struct nfsd4_callback *cb)
5121 struct nfs4_delegation *dp = cb_to_delegation(cb);
5122 struct nfsd_net *nn = net_generic(dp->dl_stid.sc_client->net,
5125 block_delegations(&dp->dl_stid.sc_file->fi_fhandle);
5128 * We can't do this in nfsd_break_deleg_cb because it is
5129 * already holding inode->i_lock.
5131 * If the dl_time != 0, then we know that it has already been
5132 * queued for a lease break. Don't queue it again.
5134 spin_lock(&state_lock);
5135 if (delegation_hashed(dp) && dp->dl_time == 0) {
5136 dp->dl_time = ktime_get_boottime_seconds();
5137 list_add_tail(&dp->dl_recall_lru, &nn->del_recall_lru);
5139 spin_unlock(&state_lock);
5142 static int nfsd4_cb_recall_done(struct nfsd4_callback *cb,
5143 struct rpc_task *task)
5145 struct nfs4_delegation *dp = cb_to_delegation(cb);
5147 trace_nfsd_cb_recall_done(&dp->dl_stid.sc_stateid, task);
5149 if (dp->dl_stid.sc_status)
5150 /* CLOSED or REVOKED */
5153 switch (task->tk_status) {
5156 case -NFS4ERR_DELAY:
5157 rpc_delay(task, 2 * HZ);
5160 case -NFS4ERR_BAD_STATEID:
5162 * Race: client probably got cb_recall before open reply
5163 * granting delegation.
5165 if (dp->dl_retries--) {
5166 rpc_delay(task, 2 * HZ);
5175 static void nfsd4_cb_recall_release(struct nfsd4_callback *cb)
5177 struct nfs4_delegation *dp = cb_to_delegation(cb);
5179 nfs4_put_stid(&dp->dl_stid);
5182 static const struct nfsd4_callback_ops nfsd4_cb_recall_ops = {
5183 .prepare = nfsd4_cb_recall_prepare,
5184 .done = nfsd4_cb_recall_done,
5185 .release = nfsd4_cb_recall_release,
5188 static void nfsd_break_one_deleg(struct nfs4_delegation *dp)
5191 * We're assuming the state code never drops its reference
5192 * without first removing the lease. Since we're in this lease
5193 * callback (and since the lease code is serialized by the
5194 * flc_lock) we know the server hasn't removed the lease yet, and
5195 * we know it's safe to take a reference.
5197 refcount_inc(&dp->dl_stid.sc_count);
5198 WARN_ON_ONCE(!nfsd4_run_cb(&dp->dl_recall));
5201 /* Called from break_lease() with flc_lock held. */
5203 nfsd_break_deleg_cb(struct file_lease *fl)
5205 struct nfs4_delegation *dp = (struct nfs4_delegation *) fl->c.flc_owner;
5206 struct nfs4_file *fp = dp->dl_stid.sc_file;
5207 struct nfs4_client *clp = dp->dl_stid.sc_client;
5208 struct nfsd_net *nn;
5210 trace_nfsd_cb_recall(&dp->dl_stid);
5212 dp->dl_recalled = true;
5213 atomic_inc(&clp->cl_delegs_in_recall);
5214 if (try_to_expire_client(clp)) {
5215 nn = net_generic(clp->net, nfsd_net_id);
5216 mod_delayed_work(laundry_wq, &nn->laundromat_work, 0);
5220 * We don't want the locks code to timeout the lease for us;
5221 * we'll remove it ourself if a delegation isn't returned
5224 fl->fl_break_time = 0;
5226 fp->fi_had_conflict = true;
5227 nfsd_break_one_deleg(dp);
5232 * nfsd_breaker_owns_lease - Check if lease conflict was resolved
5233 * @fl: Lock state to check
5236 * %true: Lease conflict was resolved
5237 * %false: Lease conflict was not resolved.
5239 static bool nfsd_breaker_owns_lease(struct file_lease *fl)
5241 struct nfs4_delegation *dl = fl->c.flc_owner;
5242 struct svc_rqst *rqst;
5243 struct nfs4_client *clp;
5247 rqst = kthread_data(current);
5248 /* Note rq_prog == NFS_ACL_PROGRAM is also possible: */
5249 if (rqst->rq_prog != NFS_PROGRAM || rqst->rq_vers < 4)
5251 clp = *(rqst->rq_lease_breaker);
5252 return dl->dl_stid.sc_client == clp;
5256 nfsd_change_deleg_cb(struct file_lease *onlist, int arg,
5257 struct list_head *dispose)
5259 struct nfs4_delegation *dp = (struct nfs4_delegation *) onlist->c.flc_owner;
5260 struct nfs4_client *clp = dp->dl_stid.sc_client;
5262 if (arg & F_UNLCK) {
5263 if (dp->dl_recalled)
5264 atomic_dec(&clp->cl_delegs_in_recall);
5265 return lease_modify(onlist, arg, dispose);
5270 static const struct lease_manager_operations nfsd_lease_mng_ops = {
5271 .lm_breaker_owns_lease = nfsd_breaker_owns_lease,
5272 .lm_break = nfsd_break_deleg_cb,
5273 .lm_change = nfsd_change_deleg_cb,
5276 static __be32 nfsd4_check_seqid(struct nfsd4_compound_state *cstate, struct nfs4_stateowner *so, u32 seqid)
5278 if (nfsd4_has_session(cstate))
5280 if (seqid == so->so_seqid - 1)
5281 return nfserr_replay_me;
5282 if (seqid == so->so_seqid)
5284 return nfserr_bad_seqid;
5287 static struct nfs4_client *lookup_clientid(clientid_t *clid, bool sessions,
5288 struct nfsd_net *nn)
5290 struct nfs4_client *found;
5292 spin_lock(&nn->client_lock);
5293 found = find_confirmed_client(clid, sessions, nn);
5295 atomic_inc(&found->cl_rpc_users);
5296 spin_unlock(&nn->client_lock);
5300 static __be32 set_client(clientid_t *clid,
5301 struct nfsd4_compound_state *cstate,
5302 struct nfsd_net *nn)
5305 if (!same_clid(&cstate->clp->cl_clientid, clid))
5306 return nfserr_stale_clientid;
5309 if (STALE_CLIENTID(clid, nn))
5310 return nfserr_stale_clientid;
5312 * We're in the 4.0 case (otherwise the SEQUENCE op would have
5313 * set cstate->clp), so session = false:
5315 cstate->clp = lookup_clientid(clid, false, nn);
5317 return nfserr_expired;
5322 nfsd4_process_open1(struct nfsd4_compound_state *cstate,
5323 struct nfsd4_open *open, struct nfsd_net *nn)
5325 clientid_t *clientid = &open->op_clientid;
5326 struct nfs4_client *clp = NULL;
5327 unsigned int strhashval;
5328 struct nfs4_openowner *oo = NULL;
5332 * In case we need it later, after we've already created the
5333 * file and don't want to risk a further failure:
5335 open->op_file = nfsd4_alloc_file();
5336 if (open->op_file == NULL)
5337 return nfserr_jukebox;
5339 status = set_client(clientid, cstate, nn);
5344 strhashval = ownerstr_hashval(&open->op_owner);
5345 oo = find_openstateowner_str(strhashval, open, clp);
5346 open->op_openowner = oo;
5350 if (!(oo->oo_flags & NFS4_OO_CONFIRMED)) {
5351 /* Replace unconfirmed owners without checking for replay. */
5352 release_openowner(oo);
5353 open->op_openowner = NULL;
5356 status = nfsd4_check_seqid(cstate, &oo->oo_owner, open->op_seqid);
5361 oo = alloc_init_open_stateowner(strhashval, open, cstate);
5363 return nfserr_jukebox;
5364 open->op_openowner = oo;
5366 open->op_stp = nfs4_alloc_open_stateid(clp);
5368 return nfserr_jukebox;
5370 if (nfsd4_has_session(cstate) &&
5371 (cstate->current_fh.fh_export->ex_flags & NFSEXP_PNFS)) {
5372 open->op_odstate = alloc_clnt_odstate(clp);
5373 if (!open->op_odstate)
5374 return nfserr_jukebox;
5380 static inline __be32
5381 nfs4_check_delegmode(struct nfs4_delegation *dp, int flags)
5383 if ((flags & WR_STATE) && (dp->dl_type == NFS4_OPEN_DELEGATE_READ))
5384 return nfserr_openmode;
5389 static int share_access_to_flags(u32 share_access)
5391 return share_access == NFS4_SHARE_ACCESS_READ ? RD_STATE : WR_STATE;
5394 static struct nfs4_delegation *find_deleg_stateid(struct nfs4_client *cl,
5397 struct nfs4_stid *ret;
5399 ret = find_stateid_by_type(cl, s, SC_TYPE_DELEG, SC_STATUS_REVOKED);
5402 return delegstateid(ret);
5405 static bool nfsd4_is_deleg_cur(struct nfsd4_open *open)
5407 return open->op_claim_type == NFS4_OPEN_CLAIM_DELEGATE_CUR ||
5408 open->op_claim_type == NFS4_OPEN_CLAIM_DELEG_CUR_FH;
5412 nfs4_check_deleg(struct nfs4_client *cl, struct nfsd4_open *open,
5413 struct nfs4_delegation **dp)
5416 __be32 status = nfserr_bad_stateid;
5417 struct nfs4_delegation *deleg;
5419 deleg = find_deleg_stateid(cl, &open->op_delegate_stateid);
5422 if (deleg->dl_stid.sc_status & SC_STATUS_ADMIN_REVOKED) {
5423 nfs4_put_stid(&deleg->dl_stid);
5424 status = nfserr_admin_revoked;
5427 if (deleg->dl_stid.sc_status & SC_STATUS_REVOKED) {
5428 nfs4_put_stid(&deleg->dl_stid);
5429 nfsd40_drop_revoked_stid(cl, &open->op_delegate_stateid);
5430 status = nfserr_deleg_revoked;
5433 flags = share_access_to_flags(open->op_share_access);
5434 status = nfs4_check_delegmode(deleg, flags);
5436 nfs4_put_stid(&deleg->dl_stid);
5441 if (!nfsd4_is_deleg_cur(open))
5445 open->op_openowner->oo_flags |= NFS4_OO_CONFIRMED;
5449 static inline int nfs4_access_to_access(u32 nfs4_access)
5453 if (nfs4_access & NFS4_SHARE_ACCESS_READ)
5454 flags |= NFSD_MAY_READ;
5455 if (nfs4_access & NFS4_SHARE_ACCESS_WRITE)
5456 flags |= NFSD_MAY_WRITE;
5460 static inline __be32
5461 nfsd4_truncate(struct svc_rqst *rqstp, struct svc_fh *fh,
5462 struct nfsd4_open *open)
5464 struct iattr iattr = {
5465 .ia_valid = ATTR_SIZE,
5468 struct nfsd_attrs attrs = {
5471 if (!open->op_truncate)
5473 if (!(open->op_share_access & NFS4_SHARE_ACCESS_WRITE))
5474 return nfserr_inval;
5475 return nfsd_setattr(rqstp, fh, &attrs, NULL);
5478 static __be32 nfs4_get_vfs_file(struct svc_rqst *rqstp, struct nfs4_file *fp,
5479 struct svc_fh *cur_fh, struct nfs4_ol_stateid *stp,
5480 struct nfsd4_open *open, bool new_stp)
5482 struct nfsd_file *nf = NULL;
5484 int oflag = nfs4_access_to_omode(open->op_share_access);
5485 int access = nfs4_access_to_access(open->op_share_access);
5486 unsigned char old_access_bmap, old_deny_bmap;
5488 spin_lock(&fp->fi_lock);
5491 * Are we trying to set a deny mode that would conflict with
5494 status = nfs4_file_check_deny(fp, open->op_share_deny);
5495 if (status != nfs_ok) {
5496 if (status != nfserr_share_denied) {
5497 spin_unlock(&fp->fi_lock);
5500 if (nfs4_resolve_deny_conflicts_locked(fp, new_stp,
5501 stp, open->op_share_deny, false))
5502 status = nfserr_jukebox;
5503 spin_unlock(&fp->fi_lock);
5507 /* set access to the file */
5508 status = nfs4_file_get_access(fp, open->op_share_access);
5509 if (status != nfs_ok) {
5510 if (status != nfserr_share_denied) {
5511 spin_unlock(&fp->fi_lock);
5514 if (nfs4_resolve_deny_conflicts_locked(fp, new_stp,
5515 stp, open->op_share_access, true))
5516 status = nfserr_jukebox;
5517 spin_unlock(&fp->fi_lock);
5521 /* Set access bits in stateid */
5522 old_access_bmap = stp->st_access_bmap;
5523 set_access(open->op_share_access, stp);
5525 /* Set new deny mask */
5526 old_deny_bmap = stp->st_deny_bmap;
5527 set_deny(open->op_share_deny, stp);
5528 fp->fi_share_deny |= (open->op_share_deny & NFS4_SHARE_DENY_BOTH);
5530 if (!fp->fi_fds[oflag]) {
5531 spin_unlock(&fp->fi_lock);
5533 status = nfsd_file_acquire_opened(rqstp, cur_fh, access,
5534 open->op_filp, &nf);
5535 if (status != nfs_ok)
5536 goto out_put_access;
5538 spin_lock(&fp->fi_lock);
5539 if (!fp->fi_fds[oflag]) {
5540 fp->fi_fds[oflag] = nf;
5544 spin_unlock(&fp->fi_lock);
5548 status = nfserrno(nfsd_open_break_lease(cur_fh->fh_dentry->d_inode,
5551 goto out_put_access;
5553 status = nfsd4_truncate(rqstp, cur_fh, open);
5555 goto out_put_access;
5559 stp->st_access_bmap = old_access_bmap;
5560 nfs4_file_put_access(fp, open->op_share_access);
5561 reset_union_bmap_deny(bmap_to_share_mode(old_deny_bmap), stp);
5566 nfs4_upgrade_open(struct svc_rqst *rqstp, struct nfs4_file *fp,
5567 struct svc_fh *cur_fh, struct nfs4_ol_stateid *stp,
5568 struct nfsd4_open *open)
5571 unsigned char old_deny_bmap = stp->st_deny_bmap;
5573 if (!test_access(open->op_share_access, stp))
5574 return nfs4_get_vfs_file(rqstp, fp, cur_fh, stp, open, false);
5576 /* test and set deny mode */
5577 spin_lock(&fp->fi_lock);
5578 status = nfs4_file_check_deny(fp, open->op_share_deny);
5581 set_deny(open->op_share_deny, stp);
5582 fp->fi_share_deny |=
5583 (open->op_share_deny & NFS4_SHARE_DENY_BOTH);
5585 case nfserr_share_denied:
5586 if (nfs4_resolve_deny_conflicts_locked(fp, false,
5587 stp, open->op_share_deny, false))
5588 status = nfserr_jukebox;
5591 spin_unlock(&fp->fi_lock);
5593 if (status != nfs_ok)
5596 status = nfsd4_truncate(rqstp, cur_fh, open);
5597 if (status != nfs_ok)
5598 reset_union_bmap_deny(old_deny_bmap, stp);
5602 /* Should we give out recallable state?: */
5603 static bool nfsd4_cb_channel_good(struct nfs4_client *clp)
5605 if (clp->cl_cb_state == NFSD4_CB_UP)
5608 * In the sessions case, since we don't have to establish a
5609 * separate connection for callbacks, we assume it's OK
5610 * until we hear otherwise:
5612 return clp->cl_minorversion && clp->cl_cb_state == NFSD4_CB_UNKNOWN;
5615 static struct file_lease *nfs4_alloc_init_lease(struct nfs4_delegation *dp,
5618 struct file_lease *fl;
5620 fl = locks_alloc_lease();
5623 fl->fl_lmops = &nfsd_lease_mng_ops;
5624 fl->c.flc_flags = FL_DELEG;
5625 fl->c.flc_type = flag == NFS4_OPEN_DELEGATE_READ? F_RDLCK: F_WRLCK;
5626 fl->c.flc_owner = (fl_owner_t)dp;
5627 fl->c.flc_pid = current->tgid;
5628 fl->c.flc_file = dp->dl_stid.sc_file->fi_deleg_file->nf_file;
5632 static int nfsd4_check_conflicting_opens(struct nfs4_client *clp,
5633 struct nfs4_file *fp)
5635 struct nfs4_ol_stateid *st;
5636 struct file *f = fp->fi_deleg_file->nf_file;
5637 struct inode *ino = file_inode(f);
5640 writes = atomic_read(&ino->i_writecount);
5644 * There could be multiple filehandles (hence multiple
5645 * nfs4_files) referencing this file, but that's not too
5646 * common; let's just give up in that case rather than
5647 * trying to go look up all the clients using that other
5648 * nfs4_file as well:
5653 * If there's a close in progress, make sure that we see it
5654 * clear any fi_fds[] entries before we see it decrement
5657 smp_mb__after_atomic();
5659 if (fp->fi_fds[O_WRONLY])
5661 if (fp->fi_fds[O_RDWR])
5664 return -EAGAIN; /* There may be non-NFSv4 writers */
5666 * It's possible there are non-NFSv4 write opens in progress,
5667 * but if they haven't incremented i_writecount yet then they
5668 * also haven't called break lease yet; so, they'll break this
5669 * lease soon enough. So, all that's left to check for is NFSv4
5672 spin_lock(&fp->fi_lock);
5673 list_for_each_entry(st, &fp->fi_stateids, st_perfile) {
5674 if (st->st_openstp == NULL /* it's an open */ &&
5675 access_permit_write(st) &&
5676 st->st_stid.sc_client != clp) {
5677 spin_unlock(&fp->fi_lock);
5681 spin_unlock(&fp->fi_lock);
5683 * There's a small chance that we could be racing with another
5684 * NFSv4 open. However, any open that hasn't added itself to
5685 * the fi_stateids list also hasn't called break_lease yet; so,
5686 * they'll break this lease soon enough.
5692 * It's possible that between opening the dentry and setting the delegation,
5693 * that it has been renamed or unlinked. Redo the lookup to verify that this
5697 nfsd4_verify_deleg_dentry(struct nfsd4_open *open, struct nfs4_file *fp,
5698 struct svc_fh *parent)
5700 struct svc_export *exp;
5701 struct dentry *child;
5704 err = nfsd_lookup_dentry(open->op_rqstp, parent,
5705 open->op_fname, open->op_fnamelen,
5713 if (child != file_dentry(fp->fi_deleg_file->nf_file))
5720 * We avoid breaking delegations held by a client due to its own activity, but
5721 * clearing setuid/setgid bits on a write is an implicit activity and the client
5722 * may not notice and continue using the old mode. Avoid giving out a delegation
5723 * on setuid/setgid files when the client is requesting an open for write.
5726 nfsd4_verify_setuid_write(struct nfsd4_open *open, struct nfsd_file *nf)
5728 struct inode *inode = file_inode(nf->nf_file);
5730 if ((open->op_share_access & NFS4_SHARE_ACCESS_WRITE) &&
5731 (inode->i_mode & (S_ISUID|S_ISGID)))
5736 static struct nfs4_delegation *
5737 nfs4_set_delegation(struct nfsd4_open *open, struct nfs4_ol_stateid *stp,
5738 struct svc_fh *parent)
5741 struct nfs4_client *clp = stp->st_stid.sc_client;
5742 struct nfs4_file *fp = stp->st_stid.sc_file;
5743 struct nfs4_clnt_odstate *odstate = stp->st_clnt_odstate;
5744 struct nfs4_delegation *dp;
5745 struct nfsd_file *nf = NULL;
5746 struct file_lease *fl;
5750 * The fi_had_conflict and nfs_get_existing_delegation checks
5751 * here are just optimizations; we'll need to recheck them at
5754 if (fp->fi_had_conflict)
5755 return ERR_PTR(-EAGAIN);
5758 * Try for a write delegation first. RFC8881 section 10.4 says:
5760 * "An OPEN_DELEGATE_WRITE delegation allows the client to handle,
5761 * on its own, all opens."
5763 * Furthermore the client can use a write delegation for most READ
5764 * operations as well, so we require a O_RDWR file here.
5766 * Offer a write delegation in the case of a BOTH open, and ensure
5767 * we get the O_RDWR descriptor.
5769 if ((open->op_share_access & NFS4_SHARE_ACCESS_BOTH) == NFS4_SHARE_ACCESS_BOTH) {
5770 nf = find_rw_file(fp);
5771 dl_type = NFS4_OPEN_DELEGATE_WRITE;
5775 * If the file is being opened O_RDONLY or we couldn't get a O_RDWR
5776 * file for some reason, then try for a read delegation instead.
5778 if (!nf && (open->op_share_access & NFS4_SHARE_ACCESS_READ)) {
5779 nf = find_readable_file(fp);
5780 dl_type = NFS4_OPEN_DELEGATE_READ;
5784 return ERR_PTR(-EAGAIN);
5786 spin_lock(&state_lock);
5787 spin_lock(&fp->fi_lock);
5788 if (nfs4_delegation_exists(clp, fp))
5790 else if (nfsd4_verify_setuid_write(open, nf))
5792 else if (!fp->fi_deleg_file) {
5793 fp->fi_deleg_file = nf;
5794 /* increment early to prevent fi_deleg_file from being
5796 fp->fi_delegees = 1;
5800 spin_unlock(&fp->fi_lock);
5801 spin_unlock(&state_lock);
5805 return ERR_PTR(status);
5808 dp = alloc_init_deleg(clp, fp, odstate, dl_type);
5812 fl = nfs4_alloc_init_lease(dp, dl_type);
5814 goto out_clnt_odstate;
5816 status = kernel_setlease(fp->fi_deleg_file->nf_file,
5817 fl->c.flc_type, &fl, NULL);
5819 locks_free_lease(fl);
5821 goto out_clnt_odstate;
5824 status = nfsd4_verify_deleg_dentry(open, fp, parent);
5829 status = nfsd4_check_conflicting_opens(clp, fp);
5834 * Now that the deleg is set, check again to ensure that nothing
5835 * raced in and changed the mode while we weren't lookng.
5837 status = nfsd4_verify_setuid_write(open, fp->fi_deleg_file);
5842 if (fp->fi_had_conflict)
5845 spin_lock(&state_lock);
5846 spin_lock(&clp->cl_lock);
5847 spin_lock(&fp->fi_lock);
5848 status = hash_delegation_locked(dp, fp);
5849 spin_unlock(&fp->fi_lock);
5850 spin_unlock(&clp->cl_lock);
5851 spin_unlock(&state_lock);
5858 kernel_setlease(fp->fi_deleg_file->nf_file, F_UNLCK, NULL, (void **)&dp);
5860 put_clnt_odstate(dp->dl_clnt_odstate);
5861 nfs4_put_stid(&dp->dl_stid);
5864 return ERR_PTR(status);
5867 static void nfsd4_open_deleg_none_ext(struct nfsd4_open *open, int status)
5869 open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE_EXT;
5870 if (status == -EAGAIN)
5871 open->op_why_no_deleg = WND4_CONTENTION;
5873 open->op_why_no_deleg = WND4_RESOURCE;
5874 switch (open->op_deleg_want) {
5875 case NFS4_SHARE_WANT_READ_DELEG:
5876 case NFS4_SHARE_WANT_WRITE_DELEG:
5877 case NFS4_SHARE_WANT_ANY_DELEG:
5879 case NFS4_SHARE_WANT_CANCEL:
5880 open->op_why_no_deleg = WND4_CANCELLED;
5882 case NFS4_SHARE_WANT_NO_DELEG:
5889 * The Linux NFS server does not offer write delegations to NFSv4.0
5890 * clients in order to avoid conflicts between write delegations and
5891 * GETATTRs requesting CHANGE or SIZE attributes.
5893 * With NFSv4.1 and later minorversions, the SEQUENCE operation that
5894 * begins each COMPOUND contains a client ID. Delegation recall can
5895 * be avoided when the server recognizes the client sending a
5896 * GETATTR also holds write delegation it conflicts with.
5898 * However, the NFSv4.0 protocol does not enable a server to
5899 * determine that a GETATTR originated from the client holding the
5900 * conflicting delegation versus coming from some other client. Per
5901 * RFC 7530 Section 16.7.5, the server must recall or send a
5902 * CB_GETATTR even when the GETATTR originates from the client that
5903 * holds the conflicting delegation.
5905 * An NFSv4.0 client can trigger a pathological situation if it
5906 * always sends a DELEGRETURN preceded by a conflicting GETATTR in
5907 * the same COMPOUND. COMPOUND execution will always stop at the
5908 * GETATTR and the DELEGRETURN will never get executed. The server
5909 * eventually revokes the delegation, which can result in loss of
5910 * open or lock state.
5913 nfs4_open_delegation(struct nfsd4_open *open, struct nfs4_ol_stateid *stp,
5914 struct svc_fh *currentfh)
5916 struct nfs4_delegation *dp;
5917 struct nfs4_openowner *oo = openowner(stp->st_stateowner);
5918 struct nfs4_client *clp = stp->st_stid.sc_client;
5919 struct svc_fh *parent = NULL;
5925 cb_up = nfsd4_cb_channel_good(oo->oo_owner.so_client);
5926 open->op_recall = false;
5927 switch (open->op_claim_type) {
5928 case NFS4_OPEN_CLAIM_PREVIOUS:
5930 open->op_recall = true;
5932 case NFS4_OPEN_CLAIM_NULL:
5935 case NFS4_OPEN_CLAIM_FH:
5937 * Let's not give out any delegations till everyone's
5938 * had the chance to reclaim theirs, *and* until
5939 * NLM locks have all been reclaimed:
5941 if (locks_in_grace(clp->net))
5943 if (!cb_up || !(oo->oo_flags & NFS4_OO_CONFIRMED))
5945 if (open->op_share_access & NFS4_SHARE_ACCESS_WRITE &&
5946 !clp->cl_minorversion)
5952 dp = nfs4_set_delegation(open, stp, parent);
5956 memcpy(&open->op_delegate_stateid, &dp->dl_stid.sc_stateid, sizeof(dp->dl_stid.sc_stateid));
5958 if (open->op_share_access & NFS4_SHARE_ACCESS_WRITE) {
5959 open->op_delegate_type = NFS4_OPEN_DELEGATE_WRITE;
5960 trace_nfsd_deleg_write(&dp->dl_stid.sc_stateid);
5961 path.mnt = currentfh->fh_export->ex_path.mnt;
5962 path.dentry = currentfh->fh_dentry;
5963 if (vfs_getattr(&path, &stat,
5964 (STATX_SIZE | STATX_CTIME | STATX_CHANGE_COOKIE),
5965 AT_STATX_SYNC_AS_STAT)) {
5966 nfs4_put_stid(&dp->dl_stid);
5967 destroy_delegation(dp);
5970 dp->dl_cb_fattr.ncf_cur_fsize = stat.size;
5971 dp->dl_cb_fattr.ncf_initial_cinfo =
5972 nfsd4_change_attribute(&stat, d_inode(currentfh->fh_dentry));
5974 open->op_delegate_type = NFS4_OPEN_DELEGATE_READ;
5975 trace_nfsd_deleg_read(&dp->dl_stid.sc_stateid);
5977 nfs4_put_stid(&dp->dl_stid);
5980 open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE;
5981 if (open->op_claim_type == NFS4_OPEN_CLAIM_PREVIOUS &&
5982 open->op_delegate_type != NFS4_OPEN_DELEGATE_NONE) {
5983 dprintk("NFSD: WARNING: refusing delegation reclaim\n");
5984 open->op_recall = true;
5987 /* 4.1 client asking for a delegation? */
5988 if (open->op_deleg_want)
5989 nfsd4_open_deleg_none_ext(open, status);
5993 static void nfsd4_deleg_xgrade_none_ext(struct nfsd4_open *open,
5994 struct nfs4_delegation *dp)
5996 if (open->op_deleg_want == NFS4_SHARE_WANT_READ_DELEG &&
5997 dp->dl_type == NFS4_OPEN_DELEGATE_WRITE) {
5998 open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE_EXT;
5999 open->op_why_no_deleg = WND4_NOT_SUPP_DOWNGRADE;
6000 } else if (open->op_deleg_want == NFS4_SHARE_WANT_WRITE_DELEG &&
6001 dp->dl_type == NFS4_OPEN_DELEGATE_WRITE) {
6002 open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE_EXT;
6003 open->op_why_no_deleg = WND4_NOT_SUPP_UPGRADE;
6005 /* Otherwise the client must be confused wanting a delegation
6006 * it already has, therefore we don't return
6007 * NFS4_OPEN_DELEGATE_NONE_EXT and reason.
6012 * nfsd4_process_open2 - finish open processing
6013 * @rqstp: the RPC transaction being executed
6014 * @current_fh: NFSv4 COMPOUND's current filehandle
6015 * @open: OPEN arguments
6017 * If successful, (1) truncate the file if open->op_truncate was
6018 * set, (2) set open->op_stateid, (3) set open->op_delegation.
6020 * Returns %nfs_ok on success; otherwise an nfs4stat value in
6021 * network byte order is returned.
6024 nfsd4_process_open2(struct svc_rqst *rqstp, struct svc_fh *current_fh, struct nfsd4_open *open)
6026 struct nfsd4_compoundres *resp = rqstp->rq_resp;
6027 struct nfs4_client *cl = open->op_openowner->oo_owner.so_client;
6028 struct nfs4_file *fp = NULL;
6029 struct nfs4_ol_stateid *stp = NULL;
6030 struct nfs4_delegation *dp = NULL;
6032 bool new_stp = false;
6035 * Lookup file; if found, lookup stateid and check open request,
6036 * and check for delegations in the process of being recalled.
6037 * If not found, create the nfs4_file struct
6039 fp = nfsd4_file_hash_insert(open->op_file, current_fh);
6041 return nfserr_jukebox;
6042 if (fp != open->op_file) {
6043 status = nfs4_check_deleg(cl, open, &dp);
6046 stp = nfsd4_find_and_lock_existing_open(fp, open);
6048 open->op_file = NULL;
6049 status = nfserr_bad_stateid;
6050 if (nfsd4_is_deleg_cur(open))
6055 stp = init_open_stateid(fp, open);
6061 * OPEN the file, or upgrade an existing OPEN.
6062 * If truncate fails, the OPEN fails.
6064 * stp is already locked.
6067 /* Stateid was found, this is an OPEN upgrade */
6068 status = nfs4_upgrade_open(rqstp, fp, current_fh, stp, open);
6070 mutex_unlock(&stp->st_mutex);
6074 status = nfs4_get_vfs_file(rqstp, fp, current_fh, stp, open, true);
6076 release_open_stateid(stp);
6077 mutex_unlock(&stp->st_mutex);
6081 stp->st_clnt_odstate = find_or_hash_clnt_odstate(fp,
6083 if (stp->st_clnt_odstate == open->op_odstate)
6084 open->op_odstate = NULL;
6087 nfs4_inc_and_copy_stateid(&open->op_stateid, &stp->st_stid);
6088 mutex_unlock(&stp->st_mutex);
6090 if (nfsd4_has_session(&resp->cstate)) {
6091 if (open->op_deleg_want & NFS4_SHARE_WANT_NO_DELEG) {
6092 open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE_EXT;
6093 open->op_why_no_deleg = WND4_NOT_WANTED;
6099 * Attempt to hand out a delegation. No error return, because the
6100 * OPEN succeeds even if we fail.
6102 nfs4_open_delegation(open, stp, &resp->cstate.current_fh);
6105 trace_nfsd_open(&stp->st_stid.sc_stateid);
6107 /* 4.1 client trying to upgrade/downgrade delegation? */
6108 if (open->op_delegate_type == NFS4_OPEN_DELEGATE_NONE && dp &&
6109 open->op_deleg_want)
6110 nfsd4_deleg_xgrade_none_ext(open, dp);
6114 if (status == 0 && open->op_claim_type == NFS4_OPEN_CLAIM_PREVIOUS)
6115 open->op_openowner->oo_flags |= NFS4_OO_CONFIRMED;
6117 * To finish the open response, we just need to set the rflags.
6119 open->op_rflags = NFS4_OPEN_RESULT_LOCKTYPE_POSIX;
6120 if (nfsd4_has_session(&resp->cstate))
6121 open->op_rflags |= NFS4_OPEN_RESULT_MAY_NOTIFY_LOCK;
6122 else if (!(open->op_openowner->oo_flags & NFS4_OO_CONFIRMED))
6123 open->op_rflags |= NFS4_OPEN_RESULT_CONFIRM;
6126 nfs4_put_stid(&dp->dl_stid);
6128 nfs4_put_stid(&stp->st_stid);
6133 void nfsd4_cleanup_open_state(struct nfsd4_compound_state *cstate,
6134 struct nfsd4_open *open)
6136 if (open->op_openowner) {
6137 struct nfs4_stateowner *so = &open->op_openowner->oo_owner;
6139 nfsd4_cstate_assign_replay(cstate, so);
6140 nfs4_put_stateowner(so);
6143 kmem_cache_free(file_slab, open->op_file);
6145 nfs4_put_stid(&open->op_stp->st_stid);
6146 if (open->op_odstate)
6147 kmem_cache_free(odstate_slab, open->op_odstate);
6151 nfsd4_renew(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
6152 union nfsd4_op_u *u)
6154 clientid_t *clid = &u->renew;
6155 struct nfs4_client *clp;
6157 struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
6159 trace_nfsd_clid_renew(clid);
6160 status = set_client(clid, cstate, nn);
6164 if (!list_empty(&clp->cl_delegations)
6165 && clp->cl_cb_state != NFSD4_CB_UP)
6166 return nfserr_cb_path_down;
6171 nfsd4_end_grace(struct nfsd_net *nn)
6173 /* do nothing if grace period already ended */
6174 if (nn->grace_ended)
6177 trace_nfsd_grace_complete(nn);
6178 nn->grace_ended = true;
6180 * If the server goes down again right now, an NFSv4
6181 * client will still be allowed to reclaim after it comes back up,
6182 * even if it hasn't yet had a chance to reclaim state this time.
6185 nfsd4_record_grace_done(nn);
6187 * At this point, NFSv4 clients can still reclaim. But if the
6188 * server crashes, any that have not yet reclaimed will be out
6189 * of luck on the next boot.
6191 * (NFSv4.1+ clients are considered to have reclaimed once they
6192 * call RECLAIM_COMPLETE. NFSv4.0 clients are considered to
6193 * have reclaimed after their first OPEN.)
6195 locks_end_grace(&nn->nfsd4_manager);
6197 * At this point, and once lockd and/or any other containers
6198 * exit their grace period, further reclaims will fail and
6199 * regular locking can resume.
6204 * If we've waited a lease period but there are still clients trying to
6205 * reclaim, wait a little longer to give them a chance to finish.
6207 static bool clients_still_reclaiming(struct nfsd_net *nn)
6209 time64_t double_grace_period_end = nn->boot_time +
6210 2 * nn->nfsd4_lease;
6212 if (nn->track_reclaim_completes &&
6213 atomic_read(&nn->nr_reclaim_complete) ==
6214 nn->reclaim_str_hashtbl_size)
6216 if (!nn->somebody_reclaimed)
6218 nn->somebody_reclaimed = false;
6220 * If we've given them *two* lease times to reclaim, and they're
6221 * still not done, give up:
6223 if (ktime_get_boottime_seconds() > double_grace_period_end)
6228 struct laundry_time {
6233 static bool state_expired(struct laundry_time *lt, time64_t last_refresh)
6235 time64_t time_remaining;
6237 if (last_refresh < lt->cutoff)
6239 time_remaining = last_refresh - lt->cutoff;
6240 lt->new_timeo = min(lt->new_timeo, time_remaining);
6244 #ifdef CONFIG_NFSD_V4_2_INTER_SSC
6245 void nfsd4_ssc_init_umount_work(struct nfsd_net *nn)
6247 spin_lock_init(&nn->nfsd_ssc_lock);
6248 INIT_LIST_HEAD(&nn->nfsd_ssc_mount_list);
6249 init_waitqueue_head(&nn->nfsd_ssc_waitq);
6251 EXPORT_SYMBOL_GPL(nfsd4_ssc_init_umount_work);
6254 * This is called when nfsd is being shutdown, after all inter_ssc
6255 * cleanup were done, to destroy the ssc delayed unmount list.
6257 static void nfsd4_ssc_shutdown_umount(struct nfsd_net *nn)
6259 struct nfsd4_ssc_umount_item *ni = NULL;
6260 struct nfsd4_ssc_umount_item *tmp;
6262 spin_lock(&nn->nfsd_ssc_lock);
6263 list_for_each_entry_safe(ni, tmp, &nn->nfsd_ssc_mount_list, nsui_list) {
6264 list_del(&ni->nsui_list);
6265 spin_unlock(&nn->nfsd_ssc_lock);
6266 mntput(ni->nsui_vfsmount);
6268 spin_lock(&nn->nfsd_ssc_lock);
6270 spin_unlock(&nn->nfsd_ssc_lock);
6273 static void nfsd4_ssc_expire_umount(struct nfsd_net *nn)
6275 bool do_wakeup = false;
6276 struct nfsd4_ssc_umount_item *ni = NULL;
6277 struct nfsd4_ssc_umount_item *tmp;
6279 spin_lock(&nn->nfsd_ssc_lock);
6280 list_for_each_entry_safe(ni, tmp, &nn->nfsd_ssc_mount_list, nsui_list) {
6281 if (time_after(jiffies, ni->nsui_expire)) {
6282 if (refcount_read(&ni->nsui_refcnt) > 1)
6285 /* mark being unmount */
6286 ni->nsui_busy = true;
6287 spin_unlock(&nn->nfsd_ssc_lock);
6288 mntput(ni->nsui_vfsmount);
6289 spin_lock(&nn->nfsd_ssc_lock);
6291 /* waiters need to start from begin of list */
6292 list_del(&ni->nsui_list);
6295 /* wakeup ssc_connect waiters */
6302 wake_up_all(&nn->nfsd_ssc_waitq);
6303 spin_unlock(&nn->nfsd_ssc_lock);
6307 /* Check if any lock belonging to this lockowner has any blockers */
6309 nfs4_lockowner_has_blockers(struct nfs4_lockowner *lo)
6311 struct file_lock_context *ctx;
6312 struct nfs4_ol_stateid *stp;
6313 struct nfs4_file *nf;
6315 list_for_each_entry(stp, &lo->lo_owner.so_stateids, st_perstateowner) {
6316 nf = stp->st_stid.sc_file;
6317 ctx = locks_inode_context(nf->fi_inode);
6320 if (locks_owner_has_blockers(ctx, lo))
6327 nfs4_anylock_blockers(struct nfs4_client *clp)
6330 struct nfs4_stateowner *so;
6331 struct nfs4_lockowner *lo;
6333 if (atomic_read(&clp->cl_delegs_in_recall))
6335 spin_lock(&clp->cl_lock);
6336 for (i = 0; i < OWNER_HASH_SIZE; i++) {
6337 list_for_each_entry(so, &clp->cl_ownerstr_hashtbl[i],
6339 if (so->so_is_open_owner)
6342 if (nfs4_lockowner_has_blockers(lo)) {
6343 spin_unlock(&clp->cl_lock);
6348 spin_unlock(&clp->cl_lock);
6353 nfs4_get_client_reaplist(struct nfsd_net *nn, struct list_head *reaplist,
6354 struct laundry_time *lt)
6356 unsigned int maxreap, reapcnt = 0;
6357 struct list_head *pos, *next;
6358 struct nfs4_client *clp;
6360 maxreap = (atomic_read(&nn->nfs4_client_count) >= nn->nfs4_max_clients) ?
6361 NFSD_CLIENT_MAX_TRIM_PER_RUN : 0;
6362 INIT_LIST_HEAD(reaplist);
6363 spin_lock(&nn->client_lock);
6364 list_for_each_safe(pos, next, &nn->client_lru) {
6365 clp = list_entry(pos, struct nfs4_client, cl_lru);
6366 if (clp->cl_state == NFSD4_EXPIRABLE)
6368 if (!state_expired(lt, clp->cl_time))
6370 if (!atomic_read(&clp->cl_rpc_users)) {
6371 if (clp->cl_state == NFSD4_ACTIVE)
6372 atomic_inc(&nn->nfsd_courtesy_clients);
6373 clp->cl_state = NFSD4_COURTESY;
6375 if (!client_has_state(clp))
6377 if (!nfs4_anylock_blockers(clp))
6378 if (reapcnt >= maxreap)
6381 if (!mark_client_expired_locked(clp)) {
6382 list_add(&clp->cl_lru, reaplist);
6386 spin_unlock(&nn->client_lock);
6390 nfs4_get_courtesy_client_reaplist(struct nfsd_net *nn,
6391 struct list_head *reaplist)
6393 unsigned int maxreap = 0, reapcnt = 0;
6394 struct list_head *pos, *next;
6395 struct nfs4_client *clp;
6397 maxreap = NFSD_CLIENT_MAX_TRIM_PER_RUN;
6398 INIT_LIST_HEAD(reaplist);
6400 spin_lock(&nn->client_lock);
6401 list_for_each_safe(pos, next, &nn->client_lru) {
6402 clp = list_entry(pos, struct nfs4_client, cl_lru);
6403 if (clp->cl_state == NFSD4_ACTIVE)
6405 if (reapcnt >= maxreap)
6407 if (!mark_client_expired_locked(clp)) {
6408 list_add(&clp->cl_lru, reaplist);
6412 spin_unlock(&nn->client_lock);
6416 nfs4_process_client_reaplist(struct list_head *reaplist)
6418 struct list_head *pos, *next;
6419 struct nfs4_client *clp;
6421 list_for_each_safe(pos, next, reaplist) {
6422 clp = list_entry(pos, struct nfs4_client, cl_lru);
6423 trace_nfsd_clid_purged(&clp->cl_clientid);
6424 list_del_init(&clp->cl_lru);
6429 static void nfs40_clean_admin_revoked(struct nfsd_net *nn,
6430 struct laundry_time *lt)
6432 struct nfs4_client *clp;
6434 spin_lock(&nn->client_lock);
6435 if (nn->nfs40_last_revoke == 0 ||
6436 nn->nfs40_last_revoke > lt->cutoff) {
6437 spin_unlock(&nn->client_lock);
6440 nn->nfs40_last_revoke = 0;
6443 list_for_each_entry(clp, &nn->client_lru, cl_lru) {
6444 unsigned long id, tmp;
6445 struct nfs4_stid *stid;
6447 if (atomic_read(&clp->cl_admin_revoked) == 0)
6450 spin_lock(&clp->cl_lock);
6451 idr_for_each_entry_ul(&clp->cl_stateids, stid, tmp, id)
6452 if (stid->sc_status & SC_STATUS_ADMIN_REVOKED) {
6453 refcount_inc(&stid->sc_count);
6454 spin_unlock(&nn->client_lock);
6455 /* this function drops ->cl_lock */
6456 nfsd4_drop_revoked_stid(stid);
6457 nfs4_put_stid(stid);
6458 spin_lock(&nn->client_lock);
6461 spin_unlock(&clp->cl_lock);
6463 spin_unlock(&nn->client_lock);
6467 nfs4_laundromat(struct nfsd_net *nn)
6469 struct nfs4_openowner *oo;
6470 struct nfs4_delegation *dp;
6471 struct nfs4_ol_stateid *stp;
6472 struct nfsd4_blocked_lock *nbl;
6473 struct list_head *pos, *next, reaplist;
6474 struct laundry_time lt = {
6475 .cutoff = ktime_get_boottime_seconds() - nn->nfsd4_lease,
6476 .new_timeo = nn->nfsd4_lease
6478 struct nfs4_cpntf_state *cps;
6479 copy_stateid_t *cps_t;
6482 if (clients_still_reclaiming(nn)) {
6486 nfsd4_end_grace(nn);
6488 spin_lock(&nn->s2s_cp_lock);
6489 idr_for_each_entry(&nn->s2s_cp_stateids, cps_t, i) {
6490 cps = container_of(cps_t, struct nfs4_cpntf_state, cp_stateid);
6491 if (cps->cp_stateid.cs_type == NFS4_COPYNOTIFY_STID &&
6492 state_expired(<, cps->cpntf_time))
6493 _free_cpntf_state_locked(nn, cps);
6495 spin_unlock(&nn->s2s_cp_lock);
6496 nfs4_get_client_reaplist(nn, &reaplist, <);
6497 nfs4_process_client_reaplist(&reaplist);
6499 nfs40_clean_admin_revoked(nn, <);
6501 spin_lock(&state_lock);
6502 list_for_each_safe(pos, next, &nn->del_recall_lru) {
6503 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
6504 if (!state_expired(<, dp->dl_time))
6506 unhash_delegation_locked(dp, SC_STATUS_REVOKED);
6507 list_add(&dp->dl_recall_lru, &reaplist);
6509 spin_unlock(&state_lock);
6510 while (!list_empty(&reaplist)) {
6511 dp = list_first_entry(&reaplist, struct nfs4_delegation,
6513 list_del_init(&dp->dl_recall_lru);
6514 revoke_delegation(dp);
6517 spin_lock(&nn->client_lock);
6518 while (!list_empty(&nn->close_lru)) {
6519 oo = list_first_entry(&nn->close_lru, struct nfs4_openowner,
6521 if (!state_expired(<, oo->oo_time))
6523 list_del_init(&oo->oo_close_lru);
6524 stp = oo->oo_last_closed_stid;
6525 oo->oo_last_closed_stid = NULL;
6526 spin_unlock(&nn->client_lock);
6527 nfs4_put_stid(&stp->st_stid);
6528 spin_lock(&nn->client_lock);
6530 spin_unlock(&nn->client_lock);
6533 * It's possible for a client to try and acquire an already held lock
6534 * that is being held for a long time, and then lose interest in it.
6535 * So, we clean out any un-revisited request after a lease period
6536 * under the assumption that the client is no longer interested.
6538 * RFC5661, sec. 9.6 states that the client must not rely on getting
6539 * notifications and must continue to poll for locks, even when the
6540 * server supports them. Thus this shouldn't lead to clients blocking
6541 * indefinitely once the lock does become free.
6543 BUG_ON(!list_empty(&reaplist));
6544 spin_lock(&nn->blocked_locks_lock);
6545 while (!list_empty(&nn->blocked_locks_lru)) {
6546 nbl = list_first_entry(&nn->blocked_locks_lru,
6547 struct nfsd4_blocked_lock, nbl_lru);
6548 if (!state_expired(<, nbl->nbl_time))
6550 list_move(&nbl->nbl_lru, &reaplist);
6551 list_del_init(&nbl->nbl_list);
6553 spin_unlock(&nn->blocked_locks_lock);
6555 while (!list_empty(&reaplist)) {
6556 nbl = list_first_entry(&reaplist,
6557 struct nfsd4_blocked_lock, nbl_lru);
6558 list_del_init(&nbl->nbl_lru);
6559 free_blocked_lock(nbl);
6561 #ifdef CONFIG_NFSD_V4_2_INTER_SSC
6562 /* service the server-to-server copy delayed unmount list */
6563 nfsd4_ssc_expire_umount(nn);
6565 if (atomic_long_read(&num_delegations) >= max_delegations)
6568 return max_t(time64_t, lt.new_timeo, NFSD_LAUNDROMAT_MINTIMEOUT);
6571 static void laundromat_main(struct work_struct *);
6574 laundromat_main(struct work_struct *laundry)
6577 struct delayed_work *dwork = to_delayed_work(laundry);
6578 struct nfsd_net *nn = container_of(dwork, struct nfsd_net,
6581 t = nfs4_laundromat(nn);
6582 queue_delayed_work(laundry_wq, &nn->laundromat_work, t*HZ);
6586 courtesy_client_reaper(struct nfsd_net *nn)
6588 struct list_head reaplist;
6590 nfs4_get_courtesy_client_reaplist(nn, &reaplist);
6591 nfs4_process_client_reaplist(&reaplist);
6595 deleg_reaper(struct nfsd_net *nn)
6597 struct list_head *pos, *next;
6598 struct nfs4_client *clp;
6599 struct list_head cblist;
6601 INIT_LIST_HEAD(&cblist);
6602 spin_lock(&nn->client_lock);
6603 list_for_each_safe(pos, next, &nn->client_lru) {
6604 clp = list_entry(pos, struct nfs4_client, cl_lru);
6605 if (clp->cl_state != NFSD4_ACTIVE ||
6606 list_empty(&clp->cl_delegations) ||
6607 atomic_read(&clp->cl_delegs_in_recall) ||
6608 test_bit(NFSD4_CLIENT_CB_RECALL_ANY, &clp->cl_flags) ||
6609 (ktime_get_boottime_seconds() -
6610 clp->cl_ra_time < 5)) {
6613 list_add(&clp->cl_ra_cblist, &cblist);
6615 /* release in nfsd4_cb_recall_any_release */
6616 kref_get(&clp->cl_nfsdfs.cl_ref);
6617 set_bit(NFSD4_CLIENT_CB_RECALL_ANY, &clp->cl_flags);
6618 clp->cl_ra_time = ktime_get_boottime_seconds();
6620 spin_unlock(&nn->client_lock);
6622 while (!list_empty(&cblist)) {
6623 clp = list_first_entry(&cblist, struct nfs4_client,
6625 list_del_init(&clp->cl_ra_cblist);
6626 clp->cl_ra->ra_keep = 0;
6627 clp->cl_ra->ra_bmval[0] = BIT(RCA4_TYPE_MASK_RDATA_DLG);
6628 clp->cl_ra->ra_bmval[0] = BIT(RCA4_TYPE_MASK_RDATA_DLG) |
6629 BIT(RCA4_TYPE_MASK_WDATA_DLG);
6630 trace_nfsd_cb_recall_any(clp->cl_ra);
6631 nfsd4_run_cb(&clp->cl_ra->ra_cb);
6636 nfsd4_state_shrinker_worker(struct work_struct *work)
6638 struct nfsd_net *nn = container_of(work, struct nfsd_net,
6639 nfsd_shrinker_work);
6641 courtesy_client_reaper(nn);
6645 static inline __be32 nfs4_check_fh(struct svc_fh *fhp, struct nfs4_stid *stp)
6647 if (!fh_match(&fhp->fh_handle, &stp->sc_file->fi_fhandle))
6648 return nfserr_bad_stateid;
6653 __be32 nfs4_check_openmode(struct nfs4_ol_stateid *stp, int flags)
6655 __be32 status = nfserr_openmode;
6657 /* For lock stateid's, we test the parent open, not the lock: */
6658 if (stp->st_openstp)
6659 stp = stp->st_openstp;
6660 if ((flags & WR_STATE) && !access_permit_write(stp))
6662 if ((flags & RD_STATE) && !access_permit_read(stp))
6669 static inline __be32
6670 check_special_stateids(struct net *net, svc_fh *current_fh, stateid_t *stateid, int flags)
6672 if (ONE_STATEID(stateid) && (flags & RD_STATE))
6674 else if (opens_in_grace(net)) {
6675 /* Answer in remaining cases depends on existence of
6676 * conflicting state; so we must wait out the grace period. */
6677 return nfserr_grace;
6678 } else if (flags & WR_STATE)
6679 return nfs4_share_conflict(current_fh,
6680 NFS4_SHARE_DENY_WRITE);
6681 else /* (flags & RD_STATE) && ZERO_STATEID(stateid) */
6682 return nfs4_share_conflict(current_fh,
6683 NFS4_SHARE_DENY_READ);
6686 static __be32 check_stateid_generation(stateid_t *in, stateid_t *ref, bool has_session)
6689 * When sessions are used the stateid generation number is ignored
6692 if (has_session && in->si_generation == 0)
6695 if (in->si_generation == ref->si_generation)
6698 /* If the client sends us a stateid from the future, it's buggy: */
6699 if (nfsd4_stateid_generation_after(in, ref))
6700 return nfserr_bad_stateid;
6702 * However, we could see a stateid from the past, even from a
6703 * non-buggy client. For example, if the client sends a lock
6704 * while some IO is outstanding, the lock may bump si_generation
6705 * while the IO is still in flight. The client could avoid that
6706 * situation by waiting for responses on all the IO requests,
6707 * but better performance may result in retrying IO that
6708 * receives an old_stateid error if requests are rarely
6709 * reordered in flight:
6711 return nfserr_old_stateid;
6714 static __be32 nfsd4_stid_check_stateid_generation(stateid_t *in, struct nfs4_stid *s, bool has_session)
6718 spin_lock(&s->sc_lock);
6719 ret = nfsd4_verify_open_stid(s);
6721 ret = check_stateid_generation(in, &s->sc_stateid, has_session);
6722 spin_unlock(&s->sc_lock);
6723 if (ret == nfserr_admin_revoked)
6724 nfsd40_drop_revoked_stid(s->sc_client,
6729 static __be32 nfsd4_check_openowner_confirmed(struct nfs4_ol_stateid *ols)
6731 if (ols->st_stateowner->so_is_open_owner &&
6732 !(openowner(ols->st_stateowner)->oo_flags & NFS4_OO_CONFIRMED))
6733 return nfserr_bad_stateid;
6737 static __be32 nfsd4_validate_stateid(struct nfs4_client *cl, stateid_t *stateid)
6739 struct nfs4_stid *s;
6740 __be32 status = nfserr_bad_stateid;
6742 if (ZERO_STATEID(stateid) || ONE_STATEID(stateid) ||
6743 CLOSE_STATEID(stateid))
6745 spin_lock(&cl->cl_lock);
6746 s = find_stateid_locked(cl, stateid);
6749 status = nfsd4_stid_check_stateid_generation(stateid, s, 1);
6752 status = nfsd4_verify_open_stid(s);
6756 switch (s->sc_type) {
6762 status = nfsd4_check_openowner_confirmed(openlockstateid(s));
6765 printk("unknown stateid type %x\n", s->sc_type);
6766 status = nfserr_bad_stateid;
6769 spin_unlock(&cl->cl_lock);
6770 if (status == nfserr_admin_revoked)
6771 nfsd40_drop_revoked_stid(cl, stateid);
6776 nfsd4_lookup_stateid(struct nfsd4_compound_state *cstate,
6778 unsigned short typemask, unsigned short statusmask,
6779 struct nfs4_stid **s, struct nfsd_net *nn)
6782 struct nfs4_stid *stid;
6783 bool return_revoked = false;
6786 * only return revoked delegations if explicitly asked.
6787 * otherwise we report revoked or bad_stateid status.
6789 if (statusmask & SC_STATUS_REVOKED)
6790 return_revoked = true;
6791 if (typemask & SC_TYPE_DELEG)
6792 /* Always allow REVOKED for DELEG so we can
6793 * retturn the appropriate error.
6795 statusmask |= SC_STATUS_REVOKED;
6797 statusmask |= SC_STATUS_ADMIN_REVOKED;
6799 if (ZERO_STATEID(stateid) || ONE_STATEID(stateid) ||
6800 CLOSE_STATEID(stateid))
6801 return nfserr_bad_stateid;
6802 status = set_client(&stateid->si_opaque.so_clid, cstate, nn);
6803 if (status == nfserr_stale_clientid) {
6804 if (cstate->session)
6805 return nfserr_bad_stateid;
6806 return nfserr_stale_stateid;
6810 stid = find_stateid_by_type(cstate->clp, stateid, typemask, statusmask);
6812 return nfserr_bad_stateid;
6813 if ((stid->sc_status & SC_STATUS_REVOKED) && !return_revoked) {
6814 nfs4_put_stid(stid);
6815 return nfserr_deleg_revoked;
6817 if (stid->sc_status & SC_STATUS_ADMIN_REVOKED) {
6818 nfsd40_drop_revoked_stid(cstate->clp, stateid);
6819 nfs4_put_stid(stid);
6820 return nfserr_admin_revoked;
6826 static struct nfsd_file *
6827 nfs4_find_file(struct nfs4_stid *s, int flags)
6829 struct nfsd_file *ret = NULL;
6831 if (!s || s->sc_status)
6834 switch (s->sc_type) {
6836 spin_lock(&s->sc_file->fi_lock);
6837 ret = nfsd_file_get(s->sc_file->fi_deleg_file);
6838 spin_unlock(&s->sc_file->fi_lock);
6842 if (flags & RD_STATE)
6843 ret = find_readable_file(s->sc_file);
6845 ret = find_writeable_file(s->sc_file);
6852 nfs4_check_olstateid(struct nfs4_ol_stateid *ols, int flags)
6856 status = nfsd4_check_openowner_confirmed(ols);
6859 return nfs4_check_openmode(ols, flags);
6863 nfs4_check_file(struct svc_rqst *rqstp, struct svc_fh *fhp, struct nfs4_stid *s,
6864 struct nfsd_file **nfp, int flags)
6866 int acc = (flags & RD_STATE) ? NFSD_MAY_READ : NFSD_MAY_WRITE;
6867 struct nfsd_file *nf;
6870 nf = nfs4_find_file(s, flags);
6872 status = nfsd_permission(rqstp, fhp->fh_export, fhp->fh_dentry,
6873 acc | NFSD_MAY_OWNER_OVERRIDE);
6879 status = nfsd_file_acquire(rqstp, fhp, acc, &nf);
6888 _free_cpntf_state_locked(struct nfsd_net *nn, struct nfs4_cpntf_state *cps)
6890 WARN_ON_ONCE(cps->cp_stateid.cs_type != NFS4_COPYNOTIFY_STID);
6891 if (!refcount_dec_and_test(&cps->cp_stateid.cs_count))
6893 list_del(&cps->cp_list);
6894 idr_remove(&nn->s2s_cp_stateids,
6895 cps->cp_stateid.cs_stid.si_opaque.so_id);
6899 * A READ from an inter server to server COPY will have a
6900 * copy stateid. Look up the copy notify stateid from the
6901 * idr structure and take a reference on it.
6903 __be32 manage_cpntf_state(struct nfsd_net *nn, stateid_t *st,
6904 struct nfs4_client *clp,
6905 struct nfs4_cpntf_state **cps)
6907 copy_stateid_t *cps_t;
6908 struct nfs4_cpntf_state *state = NULL;
6910 if (st->si_opaque.so_clid.cl_id != nn->s2s_cp_cl_id)
6911 return nfserr_bad_stateid;
6912 spin_lock(&nn->s2s_cp_lock);
6913 cps_t = idr_find(&nn->s2s_cp_stateids, st->si_opaque.so_id);
6915 state = container_of(cps_t, struct nfs4_cpntf_state,
6917 if (state->cp_stateid.cs_type != NFS4_COPYNOTIFY_STID) {
6922 refcount_inc(&state->cp_stateid.cs_count);
6924 _free_cpntf_state_locked(nn, state);
6927 spin_unlock(&nn->s2s_cp_lock);
6929 return nfserr_bad_stateid;
6935 static __be32 find_cpntf_state(struct nfsd_net *nn, stateid_t *st,
6936 struct nfs4_stid **stid)
6939 struct nfs4_cpntf_state *cps = NULL;
6940 struct nfs4_client *found;
6942 status = manage_cpntf_state(nn, st, NULL, &cps);
6946 cps->cpntf_time = ktime_get_boottime_seconds();
6948 status = nfserr_expired;
6949 found = lookup_clientid(&cps->cp_p_clid, true, nn);
6953 *stid = find_stateid_by_type(found, &cps->cp_p_stateid,
6954 SC_TYPE_DELEG|SC_TYPE_OPEN|SC_TYPE_LOCK,
6959 status = nfserr_bad_stateid;
6961 put_client_renew(found);
6963 nfs4_put_cpntf_state(nn, cps);
6967 void nfs4_put_cpntf_state(struct nfsd_net *nn, struct nfs4_cpntf_state *cps)
6969 spin_lock(&nn->s2s_cp_lock);
6970 _free_cpntf_state_locked(nn, cps);
6971 spin_unlock(&nn->s2s_cp_lock);
6975 * nfs4_preprocess_stateid_op - find and prep stateid for an operation
6976 * @rqstp: incoming request from client
6977 * @cstate: current compound state
6978 * @fhp: filehandle associated with requested stateid
6979 * @stateid: stateid (provided by client)
6980 * @flags: flags describing type of operation to be done
6981 * @nfp: optional nfsd_file return pointer (may be NULL)
6982 * @cstid: optional returned nfs4_stid pointer (may be NULL)
6984 * Given info from the client, look up a nfs4_stid for the operation. On
6985 * success, it returns a reference to the nfs4_stid and/or the nfsd_file
6986 * associated with it.
6989 nfs4_preprocess_stateid_op(struct svc_rqst *rqstp,
6990 struct nfsd4_compound_state *cstate, struct svc_fh *fhp,
6991 stateid_t *stateid, int flags, struct nfsd_file **nfp,
6992 struct nfs4_stid **cstid)
6994 struct net *net = SVC_NET(rqstp);
6995 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
6996 struct nfs4_stid *s = NULL;
7002 if (ZERO_STATEID(stateid) || ONE_STATEID(stateid)) {
7004 status = nfserr_bad_stateid;
7006 status = check_special_stateids(net, fhp, stateid,
7011 status = nfsd4_lookup_stateid(cstate, stateid,
7012 SC_TYPE_DELEG|SC_TYPE_OPEN|SC_TYPE_LOCK,
7014 if (status == nfserr_bad_stateid)
7015 status = find_cpntf_state(nn, stateid, &s);
7018 status = nfsd4_stid_check_stateid_generation(stateid, s,
7019 nfsd4_has_session(cstate));
7023 switch (s->sc_type) {
7025 status = nfs4_check_delegmode(delegstateid(s), flags);
7029 status = nfs4_check_olstateid(openlockstateid(s), flags);
7034 status = nfs4_check_fh(fhp, s);
7037 if (status == nfs_ok && nfp)
7038 status = nfs4_check_file(rqstp, fhp, s, nfp, flags);
7041 if (!status && cstid)
7050 * Test if the stateid is valid
7053 nfsd4_test_stateid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
7054 union nfsd4_op_u *u)
7056 struct nfsd4_test_stateid *test_stateid = &u->test_stateid;
7057 struct nfsd4_test_stateid_id *stateid;
7058 struct nfs4_client *cl = cstate->clp;
7060 list_for_each_entry(stateid, &test_stateid->ts_stateid_list, ts_id_list)
7061 stateid->ts_id_status =
7062 nfsd4_validate_stateid(cl, &stateid->ts_id_stateid);
7068 nfsd4_free_lock_stateid(stateid_t *stateid, struct nfs4_stid *s)
7070 struct nfs4_ol_stateid *stp = openlockstateid(s);
7073 ret = nfsd4_lock_ol_stateid(stp);
7077 ret = check_stateid_generation(stateid, &s->sc_stateid, 1);
7081 ret = nfserr_locks_held;
7082 if (check_for_locks(stp->st_stid.sc_file,
7083 lockowner(stp->st_stateowner)))
7086 release_lock_stateid(stp);
7090 mutex_unlock(&stp->st_mutex);
7097 nfsd4_free_stateid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
7098 union nfsd4_op_u *u)
7100 struct nfsd4_free_stateid *free_stateid = &u->free_stateid;
7101 stateid_t *stateid = &free_stateid->fr_stateid;
7102 struct nfs4_stid *s;
7103 struct nfs4_delegation *dp;
7104 struct nfs4_client *cl = cstate->clp;
7105 __be32 ret = nfserr_bad_stateid;
7107 spin_lock(&cl->cl_lock);
7108 s = find_stateid_locked(cl, stateid);
7109 if (!s || s->sc_status & SC_STATUS_CLOSED)
7111 if (s->sc_status & SC_STATUS_ADMIN_REVOKED) {
7112 nfsd4_drop_revoked_stid(s);
7116 spin_lock(&s->sc_lock);
7117 switch (s->sc_type) {
7119 if (s->sc_status & SC_STATUS_REVOKED) {
7120 spin_unlock(&s->sc_lock);
7121 dp = delegstateid(s);
7122 list_del_init(&dp->dl_recall_lru);
7123 spin_unlock(&cl->cl_lock);
7128 ret = nfserr_locks_held;
7131 ret = check_stateid_generation(stateid, &s->sc_stateid, 1);
7134 ret = nfserr_locks_held;
7137 spin_unlock(&s->sc_lock);
7138 refcount_inc(&s->sc_count);
7139 spin_unlock(&cl->cl_lock);
7140 ret = nfsd4_free_lock_stateid(stateid, s);
7143 spin_unlock(&s->sc_lock);
7145 spin_unlock(&cl->cl_lock);
7153 return (type == NFS4_READW_LT || type == NFS4_READ_LT) ?
7154 RD_STATE : WR_STATE;
7157 static __be32 nfs4_seqid_op_checks(struct nfsd4_compound_state *cstate, stateid_t *stateid, u32 seqid, struct nfs4_ol_stateid *stp)
7159 struct svc_fh *current_fh = &cstate->current_fh;
7160 struct nfs4_stateowner *sop = stp->st_stateowner;
7163 status = nfsd4_check_seqid(cstate, sop, seqid);
7166 status = nfsd4_lock_ol_stateid(stp);
7167 if (status != nfs_ok)
7169 status = check_stateid_generation(stateid, &stp->st_stid.sc_stateid, nfsd4_has_session(cstate));
7170 if (status == nfs_ok)
7171 status = nfs4_check_fh(current_fh, &stp->st_stid);
7172 if (status != nfs_ok)
7173 mutex_unlock(&stp->st_mutex);
7178 * nfs4_preprocess_seqid_op - find and prep an ol_stateid for a seqid-morphing op
7179 * @cstate: compund state
7180 * @seqid: seqid (provided by client)
7181 * @stateid: stateid (provided by client)
7182 * @typemask: mask of allowable types for this operation
7183 * @statusmask: mask of allowed states: 0 or STID_CLOSED
7184 * @stpp: return pointer for the stateid found
7185 * @nn: net namespace for request
7187 * Given a stateid+seqid from a client, look up an nfs4_ol_stateid and
7188 * return it in @stpp. On a nfs_ok return, the returned stateid will
7189 * have its st_mutex locked.
7192 nfs4_preprocess_seqid_op(struct nfsd4_compound_state *cstate, u32 seqid,
7194 unsigned short typemask, unsigned short statusmask,
7195 struct nfs4_ol_stateid **stpp,
7196 struct nfsd_net *nn)
7199 struct nfs4_stid *s;
7200 struct nfs4_ol_stateid *stp = NULL;
7202 trace_nfsd_preprocess(seqid, stateid);
7205 status = nfsd4_lookup_stateid(cstate, stateid,
7206 typemask, statusmask, &s, nn);
7209 stp = openlockstateid(s);
7210 nfsd4_cstate_assign_replay(cstate, stp->st_stateowner);
7212 status = nfs4_seqid_op_checks(cstate, stateid, seqid, stp);
7216 nfs4_put_stid(&stp->st_stid);
7220 static __be32 nfs4_preprocess_confirmed_seqid_op(struct nfsd4_compound_state *cstate, u32 seqid,
7221 stateid_t *stateid, struct nfs4_ol_stateid **stpp, struct nfsd_net *nn)
7224 struct nfs4_openowner *oo;
7225 struct nfs4_ol_stateid *stp;
7227 status = nfs4_preprocess_seqid_op(cstate, seqid, stateid,
7228 SC_TYPE_OPEN, 0, &stp, nn);
7231 oo = openowner(stp->st_stateowner);
7232 if (!(oo->oo_flags & NFS4_OO_CONFIRMED)) {
7233 mutex_unlock(&stp->st_mutex);
7234 nfs4_put_stid(&stp->st_stid);
7235 return nfserr_bad_stateid;
7242 nfsd4_open_confirm(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
7243 union nfsd4_op_u *u)
7245 struct nfsd4_open_confirm *oc = &u->open_confirm;
7247 struct nfs4_openowner *oo;
7248 struct nfs4_ol_stateid *stp;
7249 struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
7251 dprintk("NFSD: nfsd4_open_confirm on file %pd\n",
7252 cstate->current_fh.fh_dentry);
7254 status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0);
7258 status = nfs4_preprocess_seqid_op(cstate,
7259 oc->oc_seqid, &oc->oc_req_stateid,
7260 SC_TYPE_OPEN, 0, &stp, nn);
7263 oo = openowner(stp->st_stateowner);
7264 status = nfserr_bad_stateid;
7265 if (oo->oo_flags & NFS4_OO_CONFIRMED) {
7266 mutex_unlock(&stp->st_mutex);
7269 oo->oo_flags |= NFS4_OO_CONFIRMED;
7270 nfs4_inc_and_copy_stateid(&oc->oc_resp_stateid, &stp->st_stid);
7271 mutex_unlock(&stp->st_mutex);
7272 trace_nfsd_open_confirm(oc->oc_seqid, &stp->st_stid.sc_stateid);
7273 nfsd4_client_record_create(oo->oo_owner.so_client);
7276 nfs4_put_stid(&stp->st_stid);
7278 nfsd4_bump_seqid(cstate, status);
7282 static inline void nfs4_stateid_downgrade_bit(struct nfs4_ol_stateid *stp, u32 access)
7284 if (!test_access(access, stp))
7286 nfs4_file_put_access(stp->st_stid.sc_file, access);
7287 clear_access(access, stp);
7290 static inline void nfs4_stateid_downgrade(struct nfs4_ol_stateid *stp, u32 to_access)
7292 switch (to_access) {
7293 case NFS4_SHARE_ACCESS_READ:
7294 nfs4_stateid_downgrade_bit(stp, NFS4_SHARE_ACCESS_WRITE);
7295 nfs4_stateid_downgrade_bit(stp, NFS4_SHARE_ACCESS_BOTH);
7297 case NFS4_SHARE_ACCESS_WRITE:
7298 nfs4_stateid_downgrade_bit(stp, NFS4_SHARE_ACCESS_READ);
7299 nfs4_stateid_downgrade_bit(stp, NFS4_SHARE_ACCESS_BOTH);
7301 case NFS4_SHARE_ACCESS_BOTH:
7309 nfsd4_open_downgrade(struct svc_rqst *rqstp,
7310 struct nfsd4_compound_state *cstate, union nfsd4_op_u *u)
7312 struct nfsd4_open_downgrade *od = &u->open_downgrade;
7314 struct nfs4_ol_stateid *stp;
7315 struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
7317 dprintk("NFSD: nfsd4_open_downgrade on file %pd\n",
7318 cstate->current_fh.fh_dentry);
7320 /* We don't yet support WANT bits: */
7321 if (od->od_deleg_want)
7322 dprintk("NFSD: %s: od_deleg_want=0x%x ignored\n", __func__,
7325 status = nfs4_preprocess_confirmed_seqid_op(cstate, od->od_seqid,
7326 &od->od_stateid, &stp, nn);
7329 status = nfserr_inval;
7330 if (!test_access(od->od_share_access, stp)) {
7331 dprintk("NFSD: access not a subset of current bitmap: 0x%hhx, input access=%08x\n",
7332 stp->st_access_bmap, od->od_share_access);
7335 if (!test_deny(od->od_share_deny, stp)) {
7336 dprintk("NFSD: deny not a subset of current bitmap: 0x%hhx, input deny=%08x\n",
7337 stp->st_deny_bmap, od->od_share_deny);
7340 nfs4_stateid_downgrade(stp, od->od_share_access);
7341 reset_union_bmap_deny(od->od_share_deny, stp);
7342 nfs4_inc_and_copy_stateid(&od->od_stateid, &stp->st_stid);
7345 mutex_unlock(&stp->st_mutex);
7346 nfs4_put_stid(&stp->st_stid);
7348 nfsd4_bump_seqid(cstate, status);
7352 static void nfsd4_close_open_stateid(struct nfs4_ol_stateid *s)
7354 struct nfs4_client *clp = s->st_stid.sc_client;
7356 LIST_HEAD(reaplist);
7357 struct nfs4_ol_stateid *stp;
7359 spin_lock(&clp->cl_lock);
7360 unhashed = unhash_open_stateid(s, &reaplist);
7362 if (clp->cl_minorversion) {
7364 put_ol_stateid_locked(s, &reaplist);
7365 spin_unlock(&clp->cl_lock);
7366 list_for_each_entry(stp, &reaplist, st_locks)
7367 nfs4_free_cpntf_statelist(clp->net, &stp->st_stid);
7368 free_ol_stateid_reaplist(&reaplist);
7370 spin_unlock(&clp->cl_lock);
7371 free_ol_stateid_reaplist(&reaplist);
7373 move_to_close_lru(s, clp->net);
7378 * nfs4_unlock_state() called after encode
7381 nfsd4_close(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
7382 union nfsd4_op_u *u)
7384 struct nfsd4_close *close = &u->close;
7386 struct nfs4_ol_stateid *stp;
7387 struct net *net = SVC_NET(rqstp);
7388 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
7390 dprintk("NFSD: nfsd4_close on file %pd\n",
7391 cstate->current_fh.fh_dentry);
7393 status = nfs4_preprocess_seqid_op(cstate, close->cl_seqid,
7395 SC_TYPE_OPEN, SC_STATUS_CLOSED,
7397 nfsd4_bump_seqid(cstate, status);
7401 spin_lock(&stp->st_stid.sc_client->cl_lock);
7402 stp->st_stid.sc_status |= SC_STATUS_CLOSED;
7403 spin_unlock(&stp->st_stid.sc_client->cl_lock);
7406 * Technically we don't _really_ have to increment or copy it, since
7407 * it should just be gone after this operation and we clobber the
7408 * copied value below, but we continue to do so here just to ensure
7409 * that racing ops see that there was a state change.
7411 nfs4_inc_and_copy_stateid(&close->cl_stateid, &stp->st_stid);
7413 nfsd4_close_open_stateid(stp);
7414 mutex_unlock(&stp->st_mutex);
7416 /* v4.1+ suggests that we send a special stateid in here, since the
7417 * clients should just ignore this anyway. Since this is not useful
7418 * for v4.0 clients either, we set it to the special close_stateid
7421 * See RFC5661 section 18.2.4, and RFC7530 section 16.2.5
7423 memcpy(&close->cl_stateid, &close_stateid, sizeof(close->cl_stateid));
7425 /* put reference from nfs4_preprocess_seqid_op */
7426 nfs4_put_stid(&stp->st_stid);
7432 nfsd4_delegreturn(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
7433 union nfsd4_op_u *u)
7435 struct nfsd4_delegreturn *dr = &u->delegreturn;
7436 struct nfs4_delegation *dp;
7437 stateid_t *stateid = &dr->dr_stateid;
7438 struct nfs4_stid *s;
7440 struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
7442 if ((status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0)))
7445 status = nfsd4_lookup_stateid(cstate, stateid, SC_TYPE_DELEG, 0, &s, nn);
7448 dp = delegstateid(s);
7449 status = nfsd4_stid_check_stateid_generation(stateid, &dp->dl_stid, nfsd4_has_session(cstate));
7453 trace_nfsd_deleg_return(stateid);
7454 wake_up_var(d_inode(cstate->current_fh.fh_dentry));
7455 destroy_delegation(dp);
7457 nfs4_put_stid(&dp->dl_stid);
7462 /* last octet in a range */
7464 last_byte_offset(u64 start, u64 len)
7470 return end > start ? end - 1: NFS4_MAX_UINT64;
7474 * TODO: Linux file offsets are _signed_ 64-bit quantities, which means that
7475 * we can't properly handle lock requests that go beyond the (2^63 - 1)-th
7476 * byte, because of sign extension problems. Since NFSv4 calls for 64-bit
7477 * locking, this prevents us from being completely protocol-compliant. The
7478 * real solution to this problem is to start using unsigned file offsets in
7479 * the VFS, but this is a very deep change!
7482 nfs4_transform_lock_offset(struct file_lock *lock)
7484 if (lock->fl_start < 0)
7485 lock->fl_start = OFFSET_MAX;
7486 if (lock->fl_end < 0)
7487 lock->fl_end = OFFSET_MAX;
7491 nfsd4_lm_get_owner(fl_owner_t owner)
7493 struct nfs4_lockowner *lo = (struct nfs4_lockowner *)owner;
7495 nfs4_get_stateowner(&lo->lo_owner);
7500 nfsd4_lm_put_owner(fl_owner_t owner)
7502 struct nfs4_lockowner *lo = (struct nfs4_lockowner *)owner;
7505 nfs4_put_stateowner(&lo->lo_owner);
7508 /* return pointer to struct nfs4_client if client is expirable */
7510 nfsd4_lm_lock_expirable(struct file_lock *cfl)
7512 struct nfs4_lockowner *lo = (struct nfs4_lockowner *) cfl->c.flc_owner;
7513 struct nfs4_client *clp = lo->lo_owner.so_client;
7514 struct nfsd_net *nn;
7516 if (try_to_expire_client(clp)) {
7517 nn = net_generic(clp->net, nfsd_net_id);
7518 mod_delayed_work(laundry_wq, &nn->laundromat_work, 0);
7524 /* schedule laundromat to run immediately and wait for it to complete */
7526 nfsd4_lm_expire_lock(void)
7528 flush_workqueue(laundry_wq);
7532 nfsd4_lm_notify(struct file_lock *fl)
7534 struct nfs4_lockowner *lo = (struct nfs4_lockowner *) fl->c.flc_owner;
7535 struct net *net = lo->lo_owner.so_client->net;
7536 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
7537 struct nfsd4_blocked_lock *nbl = container_of(fl,
7538 struct nfsd4_blocked_lock, nbl_lock);
7541 /* An empty list means that something else is going to be using it */
7542 spin_lock(&nn->blocked_locks_lock);
7543 if (!list_empty(&nbl->nbl_list)) {
7544 list_del_init(&nbl->nbl_list);
7545 list_del_init(&nbl->nbl_lru);
7548 spin_unlock(&nn->blocked_locks_lock);
7551 trace_nfsd_cb_notify_lock(lo, nbl);
7552 nfsd4_run_cb(&nbl->nbl_cb);
7556 static const struct lock_manager_operations nfsd_posix_mng_ops = {
7557 .lm_mod_owner = THIS_MODULE,
7558 .lm_notify = nfsd4_lm_notify,
7559 .lm_get_owner = nfsd4_lm_get_owner,
7560 .lm_put_owner = nfsd4_lm_put_owner,
7561 .lm_lock_expirable = nfsd4_lm_lock_expirable,
7562 .lm_expire_lock = nfsd4_lm_expire_lock,
7566 nfs4_set_lock_denied(struct file_lock *fl, struct nfsd4_lock_denied *deny)
7568 struct nfs4_lockowner *lo;
7570 if (fl->fl_lmops == &nfsd_posix_mng_ops) {
7571 lo = (struct nfs4_lockowner *) fl->c.flc_owner;
7572 xdr_netobj_dup(&deny->ld_owner, &lo->lo_owner.so_owner,
7574 if (!deny->ld_owner.data)
7575 /* We just don't care that much */
7577 deny->ld_clientid = lo->lo_owner.so_client->cl_clientid;
7580 deny->ld_owner.len = 0;
7581 deny->ld_owner.data = NULL;
7582 deny->ld_clientid.cl_boot = 0;
7583 deny->ld_clientid.cl_id = 0;
7585 deny->ld_start = fl->fl_start;
7586 deny->ld_length = NFS4_MAX_UINT64;
7587 if (fl->fl_end != NFS4_MAX_UINT64)
7588 deny->ld_length = fl->fl_end - fl->fl_start + 1;
7589 deny->ld_type = NFS4_READ_LT;
7590 if (fl->c.flc_type != F_RDLCK)
7591 deny->ld_type = NFS4_WRITE_LT;
7594 static struct nfs4_lockowner *
7595 find_lockowner_str_locked(struct nfs4_client *clp, struct xdr_netobj *owner)
7597 unsigned int strhashval = ownerstr_hashval(owner);
7598 struct nfs4_stateowner *so;
7600 lockdep_assert_held(&clp->cl_lock);
7602 list_for_each_entry(so, &clp->cl_ownerstr_hashtbl[strhashval],
7604 if (so->so_is_open_owner)
7606 if (same_owner_str(so, owner))
7607 return lockowner(nfs4_get_stateowner(so));
7612 static struct nfs4_lockowner *
7613 find_lockowner_str(struct nfs4_client *clp, struct xdr_netobj *owner)
7615 struct nfs4_lockowner *lo;
7617 spin_lock(&clp->cl_lock);
7618 lo = find_lockowner_str_locked(clp, owner);
7619 spin_unlock(&clp->cl_lock);
7623 static void nfs4_unhash_lockowner(struct nfs4_stateowner *sop)
7625 unhash_lockowner_locked(lockowner(sop));
7628 static void nfs4_free_lockowner(struct nfs4_stateowner *sop)
7630 struct nfs4_lockowner *lo = lockowner(sop);
7632 kmem_cache_free(lockowner_slab, lo);
7635 static const struct nfs4_stateowner_operations lockowner_ops = {
7636 .so_unhash = nfs4_unhash_lockowner,
7637 .so_free = nfs4_free_lockowner,
7641 * Alloc a lock owner structure.
7642 * Called in nfsd4_lock - therefore, OPEN and OPEN_CONFIRM (if needed) has
7645 * strhashval = ownerstr_hashval
7647 static struct nfs4_lockowner *
7648 alloc_init_lock_stateowner(unsigned int strhashval, struct nfs4_client *clp,
7649 struct nfs4_ol_stateid *open_stp,
7650 struct nfsd4_lock *lock)
7652 struct nfs4_lockowner *lo, *ret;
7654 lo = alloc_stateowner(lockowner_slab, &lock->lk_new_owner, clp);
7657 INIT_LIST_HEAD(&lo->lo_blocked);
7658 INIT_LIST_HEAD(&lo->lo_owner.so_stateids);
7659 lo->lo_owner.so_is_open_owner = 0;
7660 lo->lo_owner.so_seqid = lock->lk_new_lock_seqid;
7661 lo->lo_owner.so_ops = &lockowner_ops;
7662 spin_lock(&clp->cl_lock);
7663 ret = find_lockowner_str_locked(clp, &lock->lk_new_owner);
7665 list_add(&lo->lo_owner.so_strhash,
7666 &clp->cl_ownerstr_hashtbl[strhashval]);
7669 nfs4_free_stateowner(&lo->lo_owner);
7671 spin_unlock(&clp->cl_lock);
7675 static struct nfs4_ol_stateid *
7676 find_lock_stateid(const struct nfs4_lockowner *lo,
7677 const struct nfs4_ol_stateid *ost)
7679 struct nfs4_ol_stateid *lst;
7681 lockdep_assert_held(&ost->st_stid.sc_client->cl_lock);
7683 /* If ost is not hashed, ost->st_locks will not be valid */
7684 if (!nfs4_ol_stateid_unhashed(ost))
7685 list_for_each_entry(lst, &ost->st_locks, st_locks) {
7686 if (lst->st_stateowner == &lo->lo_owner) {
7687 refcount_inc(&lst->st_stid.sc_count);
7694 static struct nfs4_ol_stateid *
7695 init_lock_stateid(struct nfs4_ol_stateid *stp, struct nfs4_lockowner *lo,
7696 struct nfs4_file *fp, struct inode *inode,
7697 struct nfs4_ol_stateid *open_stp)
7699 struct nfs4_client *clp = lo->lo_owner.so_client;
7700 struct nfs4_ol_stateid *retstp;
7702 mutex_init(&stp->st_mutex);
7703 mutex_lock_nested(&stp->st_mutex, OPEN_STATEID_MUTEX);
7705 spin_lock(&clp->cl_lock);
7706 if (nfs4_ol_stateid_unhashed(open_stp))
7708 retstp = find_lock_stateid(lo, open_stp);
7711 refcount_inc(&stp->st_stid.sc_count);
7712 stp->st_stid.sc_type = SC_TYPE_LOCK;
7713 stp->st_stateowner = nfs4_get_stateowner(&lo->lo_owner);
7715 stp->st_stid.sc_file = fp;
7716 stp->st_access_bmap = 0;
7717 stp->st_deny_bmap = open_stp->st_deny_bmap;
7718 stp->st_openstp = open_stp;
7719 spin_lock(&fp->fi_lock);
7720 list_add(&stp->st_locks, &open_stp->st_locks);
7721 list_add(&stp->st_perstateowner, &lo->lo_owner.so_stateids);
7722 list_add(&stp->st_perfile, &fp->fi_stateids);
7723 spin_unlock(&fp->fi_lock);
7724 spin_unlock(&clp->cl_lock);
7727 spin_unlock(&clp->cl_lock);
7728 if (nfsd4_lock_ol_stateid(retstp) != nfs_ok) {
7729 nfs4_put_stid(&retstp->st_stid);
7732 /* To keep mutex tracking happy */
7733 mutex_unlock(&stp->st_mutex);
7736 spin_unlock(&clp->cl_lock);
7737 mutex_unlock(&stp->st_mutex);
7741 static struct nfs4_ol_stateid *
7742 find_or_create_lock_stateid(struct nfs4_lockowner *lo, struct nfs4_file *fi,
7743 struct inode *inode, struct nfs4_ol_stateid *ost,
7746 struct nfs4_stid *ns = NULL;
7747 struct nfs4_ol_stateid *lst;
7748 struct nfs4_openowner *oo = openowner(ost->st_stateowner);
7749 struct nfs4_client *clp = oo->oo_owner.so_client;
7752 spin_lock(&clp->cl_lock);
7753 lst = find_lock_stateid(lo, ost);
7754 spin_unlock(&clp->cl_lock);
7756 if (nfsd4_lock_ol_stateid(lst) == nfs_ok)
7758 nfs4_put_stid(&lst->st_stid);
7760 ns = nfs4_alloc_stid(clp, stateid_slab, nfs4_free_lock_stateid);
7764 lst = init_lock_stateid(openlockstateid(ns), lo, fi, inode, ost);
7765 if (lst == openlockstateid(ns))
7774 check_lock_length(u64 offset, u64 length)
7776 return ((length == 0) || ((length != NFS4_MAX_UINT64) &&
7777 (length > ~offset)));
7780 static void get_lock_access(struct nfs4_ol_stateid *lock_stp, u32 access)
7782 struct nfs4_file *fp = lock_stp->st_stid.sc_file;
7784 lockdep_assert_held(&fp->fi_lock);
7786 if (test_access(access, lock_stp))
7788 __nfs4_file_get_access(fp, access);
7789 set_access(access, lock_stp);
7793 lookup_or_create_lock_state(struct nfsd4_compound_state *cstate,
7794 struct nfs4_ol_stateid *ost,
7795 struct nfsd4_lock *lock,
7796 struct nfs4_ol_stateid **plst, bool *new)
7799 struct nfs4_file *fi = ost->st_stid.sc_file;
7800 struct nfs4_openowner *oo = openowner(ost->st_stateowner);
7801 struct nfs4_client *cl = oo->oo_owner.so_client;
7802 struct inode *inode = d_inode(cstate->current_fh.fh_dentry);
7803 struct nfs4_lockowner *lo;
7804 struct nfs4_ol_stateid *lst;
7805 unsigned int strhashval;
7807 lo = find_lockowner_str(cl, &lock->lk_new_owner);
7809 strhashval = ownerstr_hashval(&lock->lk_new_owner);
7810 lo = alloc_init_lock_stateowner(strhashval, cl, ost, lock);
7812 return nfserr_jukebox;
7814 /* with an existing lockowner, seqids must be the same */
7815 status = nfserr_bad_seqid;
7816 if (!cstate->minorversion &&
7817 lock->lk_new_lock_seqid != lo->lo_owner.so_seqid)
7821 lst = find_or_create_lock_stateid(lo, fi, inode, ost, new);
7823 status = nfserr_jukebox;
7830 nfs4_put_stateowner(&lo->lo_owner);
7838 nfsd4_lock(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
7839 union nfsd4_op_u *u)
7841 struct nfsd4_lock *lock = &u->lock;
7842 struct nfs4_openowner *open_sop = NULL;
7843 struct nfs4_lockowner *lock_sop = NULL;
7844 struct nfs4_ol_stateid *lock_stp = NULL;
7845 struct nfs4_ol_stateid *open_stp = NULL;
7846 struct nfs4_file *fp;
7847 struct nfsd_file *nf = NULL;
7848 struct nfsd4_blocked_lock *nbl = NULL;
7849 struct file_lock *file_lock = NULL;
7850 struct file_lock *conflock = NULL;
7851 struct super_block *sb;
7857 unsigned int flags = FL_POSIX;
7858 struct net *net = SVC_NET(rqstp);
7859 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
7861 dprintk("NFSD: nfsd4_lock: start=%Ld length=%Ld\n",
7862 (long long) lock->lk_offset,
7863 (long long) lock->lk_length);
7865 if (check_lock_length(lock->lk_offset, lock->lk_length))
7866 return nfserr_inval;
7868 if ((status = fh_verify(rqstp, &cstate->current_fh,
7869 S_IFREG, NFSD_MAY_LOCK))) {
7870 dprintk("NFSD: nfsd4_lock: permission denied!\n");
7873 sb = cstate->current_fh.fh_dentry->d_sb;
7875 if (lock->lk_is_new) {
7876 if (nfsd4_has_session(cstate))
7877 /* See rfc 5661 18.10.3: given clientid is ignored: */
7878 memcpy(&lock->lk_new_clientid,
7879 &cstate->clp->cl_clientid,
7880 sizeof(clientid_t));
7882 /* validate and update open stateid and open seqid */
7883 status = nfs4_preprocess_confirmed_seqid_op(cstate,
7884 lock->lk_new_open_seqid,
7885 &lock->lk_new_open_stateid,
7889 mutex_unlock(&open_stp->st_mutex);
7890 open_sop = openowner(open_stp->st_stateowner);
7891 status = nfserr_bad_stateid;
7892 if (!same_clid(&open_sop->oo_owner.so_client->cl_clientid,
7893 &lock->lk_new_clientid))
7895 status = lookup_or_create_lock_state(cstate, open_stp, lock,
7898 status = nfs4_preprocess_seqid_op(cstate,
7899 lock->lk_old_lock_seqid,
7900 &lock->lk_old_lock_stateid,
7901 SC_TYPE_LOCK, 0, &lock_stp,
7906 lock_sop = lockowner(lock_stp->st_stateowner);
7908 lkflg = setlkflg(lock->lk_type);
7909 status = nfs4_check_openmode(lock_stp, lkflg);
7913 status = nfserr_grace;
7914 if (locks_in_grace(net) && !lock->lk_reclaim)
7916 status = nfserr_no_grace;
7917 if (!locks_in_grace(net) && lock->lk_reclaim)
7920 if (lock->lk_reclaim)
7921 flags |= FL_RECLAIM;
7923 fp = lock_stp->st_stid.sc_file;
7924 switch (lock->lk_type) {
7926 if (nfsd4_has_session(cstate) ||
7927 exportfs_lock_op_is_async(sb->s_export_op))
7931 spin_lock(&fp->fi_lock);
7932 nf = find_readable_file_locked(fp);
7934 get_lock_access(lock_stp, NFS4_SHARE_ACCESS_READ);
7935 spin_unlock(&fp->fi_lock);
7938 case NFS4_WRITEW_LT:
7939 if (nfsd4_has_session(cstate) ||
7940 exportfs_lock_op_is_async(sb->s_export_op))
7944 spin_lock(&fp->fi_lock);
7945 nf = find_writeable_file_locked(fp);
7947 get_lock_access(lock_stp, NFS4_SHARE_ACCESS_WRITE);
7948 spin_unlock(&fp->fi_lock);
7952 status = nfserr_inval;
7957 status = nfserr_openmode;
7962 * Most filesystems with their own ->lock operations will block
7963 * the nfsd thread waiting to acquire the lock. That leads to
7964 * deadlocks (we don't want every nfsd thread tied up waiting
7965 * for file locks), so don't attempt blocking lock notifications
7966 * on those filesystems:
7968 if (!exportfs_lock_op_is_async(sb->s_export_op))
7971 nbl = find_or_allocate_block(lock_sop, &fp->fi_fhandle, nn);
7973 dprintk("NFSD: %s: unable to allocate block!\n", __func__);
7974 status = nfserr_jukebox;
7978 file_lock = &nbl->nbl_lock;
7979 file_lock->c.flc_type = type;
7980 file_lock->c.flc_owner = (fl_owner_t)lockowner(nfs4_get_stateowner(&lock_sop->lo_owner));
7981 file_lock->c.flc_pid = current->tgid;
7982 file_lock->c.flc_file = nf->nf_file;
7983 file_lock->c.flc_flags = flags;
7984 file_lock->fl_lmops = &nfsd_posix_mng_ops;
7985 file_lock->fl_start = lock->lk_offset;
7986 file_lock->fl_end = last_byte_offset(lock->lk_offset, lock->lk_length);
7987 nfs4_transform_lock_offset(file_lock);
7989 conflock = locks_alloc_lock();
7991 dprintk("NFSD: %s: unable to allocate lock!\n", __func__);
7992 status = nfserr_jukebox;
7996 if (flags & FL_SLEEP) {
7997 nbl->nbl_time = ktime_get_boottime_seconds();
7998 spin_lock(&nn->blocked_locks_lock);
7999 list_add_tail(&nbl->nbl_list, &lock_sop->lo_blocked);
8000 list_add_tail(&nbl->nbl_lru, &nn->blocked_locks_lru);
8001 kref_get(&nbl->nbl_kref);
8002 spin_unlock(&nn->blocked_locks_lock);
8005 err = vfs_lock_file(nf->nf_file, F_SETLK, file_lock, conflock);
8007 case 0: /* success! */
8008 nfs4_inc_and_copy_stateid(&lock->lk_resp_stateid, &lock_stp->st_stid);
8010 if (lock->lk_reclaim)
8011 nn->somebody_reclaimed = true;
8013 case FILE_LOCK_DEFERRED:
8014 kref_put(&nbl->nbl_kref, free_nbl);
8017 case -EAGAIN: /* conflock holds conflicting lock */
8018 status = nfserr_denied;
8019 dprintk("NFSD: nfsd4_lock: conflicting lock found!\n");
8020 nfs4_set_lock_denied(conflock, &lock->lk_denied);
8023 status = nfserr_deadlock;
8026 dprintk("NFSD: nfsd4_lock: vfs_lock_file() failed! status %d\n",err);
8027 status = nfserrno(err);
8032 /* dequeue it if we queued it before */
8033 if (flags & FL_SLEEP) {
8034 spin_lock(&nn->blocked_locks_lock);
8035 if (!list_empty(&nbl->nbl_list) &&
8036 !list_empty(&nbl->nbl_lru)) {
8037 list_del_init(&nbl->nbl_list);
8038 list_del_init(&nbl->nbl_lru);
8039 kref_put(&nbl->nbl_kref, free_nbl);
8041 /* nbl can use one of lists to be linked to reaplist */
8042 spin_unlock(&nn->blocked_locks_lock);
8044 free_blocked_lock(nbl);
8049 /* Bump seqid manually if the 4.0 replay owner is openowner */
8050 if (cstate->replay_owner &&
8051 cstate->replay_owner != &lock_sop->lo_owner &&
8052 seqid_mutating_err(ntohl(status)))
8053 lock_sop->lo_owner.so_seqid++;
8056 * If this is a new, never-before-used stateid, and we are
8057 * returning an error, then just go ahead and release it.
8060 release_lock_stateid(lock_stp);
8062 mutex_unlock(&lock_stp->st_mutex);
8064 nfs4_put_stid(&lock_stp->st_stid);
8067 nfs4_put_stid(&open_stp->st_stid);
8068 nfsd4_bump_seqid(cstate, status);
8070 locks_free_lock(conflock);
8074 void nfsd4_lock_release(union nfsd4_op_u *u)
8076 struct nfsd4_lock *lock = &u->lock;
8077 struct nfsd4_lock_denied *deny = &lock->lk_denied;
8079 kfree(deny->ld_owner.data);
8083 * The NFSv4 spec allows a client to do a LOCKT without holding an OPEN,
8084 * so we do a temporary open here just to get an open file to pass to
8087 static __be32 nfsd_test_lock(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file_lock *lock)
8089 struct nfsd_file *nf;
8090 struct inode *inode;
8093 err = nfsd_file_acquire(rqstp, fhp, NFSD_MAY_READ, &nf);
8096 inode = fhp->fh_dentry->d_inode;
8097 inode_lock(inode); /* to block new leases till after test_lock: */
8098 err = nfserrno(nfsd_open_break_lease(inode, NFSD_MAY_READ));
8101 lock->c.flc_file = nf->nf_file;
8102 err = nfserrno(vfs_test_lock(nf->nf_file, lock));
8103 lock->c.flc_file = NULL;
8105 inode_unlock(inode);
8114 nfsd4_lockt(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
8115 union nfsd4_op_u *u)
8117 struct nfsd4_lockt *lockt = &u->lockt;
8118 struct file_lock *file_lock = NULL;
8119 struct nfs4_lockowner *lo = NULL;
8121 struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
8123 if (locks_in_grace(SVC_NET(rqstp)))
8124 return nfserr_grace;
8126 if (check_lock_length(lockt->lt_offset, lockt->lt_length))
8127 return nfserr_inval;
8129 if (!nfsd4_has_session(cstate)) {
8130 status = set_client(&lockt->lt_clientid, cstate, nn);
8135 if ((status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0)))
8138 file_lock = locks_alloc_lock();
8140 dprintk("NFSD: %s: unable to allocate lock!\n", __func__);
8141 status = nfserr_jukebox;
8145 switch (lockt->lt_type) {
8148 file_lock->c.flc_type = F_RDLCK;
8151 case NFS4_WRITEW_LT:
8152 file_lock->c.flc_type = F_WRLCK;
8155 dprintk("NFSD: nfs4_lockt: bad lock type!\n");
8156 status = nfserr_inval;
8160 lo = find_lockowner_str(cstate->clp, &lockt->lt_owner);
8162 file_lock->c.flc_owner = (fl_owner_t)lo;
8163 file_lock->c.flc_pid = current->tgid;
8164 file_lock->c.flc_flags = FL_POSIX;
8166 file_lock->fl_start = lockt->lt_offset;
8167 file_lock->fl_end = last_byte_offset(lockt->lt_offset, lockt->lt_length);
8169 nfs4_transform_lock_offset(file_lock);
8171 status = nfsd_test_lock(rqstp, &cstate->current_fh, file_lock);
8175 if (file_lock->c.flc_type != F_UNLCK) {
8176 status = nfserr_denied;
8177 nfs4_set_lock_denied(file_lock, &lockt->lt_denied);
8181 nfs4_put_stateowner(&lo->lo_owner);
8183 locks_free_lock(file_lock);
8187 void nfsd4_lockt_release(union nfsd4_op_u *u)
8189 struct nfsd4_lockt *lockt = &u->lockt;
8190 struct nfsd4_lock_denied *deny = &lockt->lt_denied;
8192 kfree(deny->ld_owner.data);
8196 nfsd4_locku(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
8197 union nfsd4_op_u *u)
8199 struct nfsd4_locku *locku = &u->locku;
8200 struct nfs4_ol_stateid *stp;
8201 struct nfsd_file *nf = NULL;
8202 struct file_lock *file_lock = NULL;
8205 struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
8207 dprintk("NFSD: nfsd4_locku: start=%Ld length=%Ld\n",
8208 (long long) locku->lu_offset,
8209 (long long) locku->lu_length);
8211 if (check_lock_length(locku->lu_offset, locku->lu_length))
8212 return nfserr_inval;
8214 status = nfs4_preprocess_seqid_op(cstate, locku->lu_seqid,
8215 &locku->lu_stateid, SC_TYPE_LOCK, 0,
8219 nf = find_any_file(stp->st_stid.sc_file);
8221 status = nfserr_lock_range;
8224 file_lock = locks_alloc_lock();
8226 dprintk("NFSD: %s: unable to allocate lock!\n", __func__);
8227 status = nfserr_jukebox;
8231 file_lock->c.flc_type = F_UNLCK;
8232 file_lock->c.flc_owner = (fl_owner_t)lockowner(nfs4_get_stateowner(stp->st_stateowner));
8233 file_lock->c.flc_pid = current->tgid;
8234 file_lock->c.flc_file = nf->nf_file;
8235 file_lock->c.flc_flags = FL_POSIX;
8236 file_lock->fl_lmops = &nfsd_posix_mng_ops;
8237 file_lock->fl_start = locku->lu_offset;
8239 file_lock->fl_end = last_byte_offset(locku->lu_offset,
8241 nfs4_transform_lock_offset(file_lock);
8243 err = vfs_lock_file(nf->nf_file, F_SETLK, file_lock, NULL);
8245 dprintk("NFSD: nfs4_locku: vfs_lock_file failed!\n");
8248 nfs4_inc_and_copy_stateid(&locku->lu_stateid, &stp->st_stid);
8252 mutex_unlock(&stp->st_mutex);
8253 nfs4_put_stid(&stp->st_stid);
8255 nfsd4_bump_seqid(cstate, status);
8257 locks_free_lock(file_lock);
8261 status = nfserrno(err);
8267 * true: locks held by lockowner
8268 * false: no locks held by lockowner
8271 check_for_locks(struct nfs4_file *fp, struct nfs4_lockowner *lowner)
8273 struct file_lock *fl;
8275 struct nfsd_file *nf;
8276 struct inode *inode;
8277 struct file_lock_context *flctx;
8279 spin_lock(&fp->fi_lock);
8280 nf = find_any_file_locked(fp);
8282 /* Any valid lock stateid should have some sort of access */
8287 inode = file_inode(nf->nf_file);
8288 flctx = locks_inode_context(inode);
8290 if (flctx && !list_empty_careful(&flctx->flc_posix)) {
8291 spin_lock(&flctx->flc_lock);
8292 for_each_file_lock(fl, &flctx->flc_posix) {
8293 if (fl->c.flc_owner == (fl_owner_t)lowner) {
8298 spin_unlock(&flctx->flc_lock);
8301 spin_unlock(&fp->fi_lock);
8306 * nfsd4_release_lockowner - process NFSv4.0 RELEASE_LOCKOWNER operations
8307 * @rqstp: RPC transaction
8308 * @cstate: NFSv4 COMPOUND state
8309 * @u: RELEASE_LOCKOWNER arguments
8311 * Check if theree are any locks still held and if not - free the lockowner
8312 * and any lock state that is owned.
8315 * %nfs_ok: lockowner released or not found
8316 * %nfserr_locks_held: lockowner still in use
8317 * %nfserr_stale_clientid: clientid no longer active
8318 * %nfserr_expired: clientid not recognized
8321 nfsd4_release_lockowner(struct svc_rqst *rqstp,
8322 struct nfsd4_compound_state *cstate,
8323 union nfsd4_op_u *u)
8325 struct nfsd4_release_lockowner *rlockowner = &u->release_lockowner;
8326 struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
8327 clientid_t *clid = &rlockowner->rl_clientid;
8328 struct nfs4_ol_stateid *stp;
8329 struct nfs4_lockowner *lo;
8330 struct nfs4_client *clp;
8331 LIST_HEAD(reaplist);
8334 dprintk("nfsd4_release_lockowner clientid: (%08x/%08x):\n",
8335 clid->cl_boot, clid->cl_id);
8337 status = set_client(clid, cstate, nn);
8342 spin_lock(&clp->cl_lock);
8343 lo = find_lockowner_str_locked(clp, &rlockowner->rl_owner);
8345 spin_unlock(&clp->cl_lock);
8349 list_for_each_entry(stp, &lo->lo_owner.so_stateids, st_perstateowner) {
8350 if (check_for_locks(stp->st_stid.sc_file, lo)) {
8351 spin_unlock(&clp->cl_lock);
8352 nfs4_put_stateowner(&lo->lo_owner);
8353 return nfserr_locks_held;
8356 unhash_lockowner_locked(lo);
8357 while (!list_empty(&lo->lo_owner.so_stateids)) {
8358 stp = list_first_entry(&lo->lo_owner.so_stateids,
8359 struct nfs4_ol_stateid,
8361 unhash_lock_stateid(stp);
8362 put_ol_stateid_locked(stp, &reaplist);
8364 spin_unlock(&clp->cl_lock);
8366 free_ol_stateid_reaplist(&reaplist);
8367 remove_blocked_locks(lo);
8368 nfs4_put_stateowner(&lo->lo_owner);
8372 static inline struct nfs4_client_reclaim *
8375 return kmalloc(sizeof(struct nfs4_client_reclaim), GFP_KERNEL);
8379 nfs4_has_reclaimed_state(struct xdr_netobj name, struct nfsd_net *nn)
8381 struct nfs4_client_reclaim *crp;
8383 crp = nfsd4_find_reclaim_client(name, nn);
8384 return (crp && crp->cr_clp);
8388 * failure => all reset bets are off, nfserr_no_grace...
8390 * The caller is responsible for freeing name.data if NULL is returned (it
8391 * will be freed in nfs4_remove_reclaim_record in the normal case).
8393 struct nfs4_client_reclaim *
8394 nfs4_client_to_reclaim(struct xdr_netobj name, struct xdr_netobj princhash,
8395 struct nfsd_net *nn)
8397 unsigned int strhashval;
8398 struct nfs4_client_reclaim *crp;
8400 crp = alloc_reclaim();
8402 strhashval = clientstr_hashval(name);
8403 INIT_LIST_HEAD(&crp->cr_strhash);
8404 list_add(&crp->cr_strhash, &nn->reclaim_str_hashtbl[strhashval]);
8405 crp->cr_name.data = name.data;
8406 crp->cr_name.len = name.len;
8407 crp->cr_princhash.data = princhash.data;
8408 crp->cr_princhash.len = princhash.len;
8410 nn->reclaim_str_hashtbl_size++;
8416 nfs4_remove_reclaim_record(struct nfs4_client_reclaim *crp, struct nfsd_net *nn)
8418 list_del(&crp->cr_strhash);
8419 kfree(crp->cr_name.data);
8420 kfree(crp->cr_princhash.data);
8422 nn->reclaim_str_hashtbl_size--;
8426 nfs4_release_reclaim(struct nfsd_net *nn)
8428 struct nfs4_client_reclaim *crp = NULL;
8431 for (i = 0; i < CLIENT_HASH_SIZE; i++) {
8432 while (!list_empty(&nn->reclaim_str_hashtbl[i])) {
8433 crp = list_entry(nn->reclaim_str_hashtbl[i].next,
8434 struct nfs4_client_reclaim, cr_strhash);
8435 nfs4_remove_reclaim_record(crp, nn);
8438 WARN_ON_ONCE(nn->reclaim_str_hashtbl_size);
8442 * called from OPEN, CLAIM_PREVIOUS with a new clientid. */
8443 struct nfs4_client_reclaim *
8444 nfsd4_find_reclaim_client(struct xdr_netobj name, struct nfsd_net *nn)
8446 unsigned int strhashval;
8447 struct nfs4_client_reclaim *crp = NULL;
8449 strhashval = clientstr_hashval(name);
8450 list_for_each_entry(crp, &nn->reclaim_str_hashtbl[strhashval], cr_strhash) {
8451 if (compare_blob(&crp->cr_name, &name) == 0) {
8459 nfs4_check_open_reclaim(struct nfs4_client *clp)
8461 if (test_bit(NFSD4_CLIENT_RECLAIM_COMPLETE, &clp->cl_flags))
8462 return nfserr_no_grace;
8464 if (nfsd4_client_record_check(clp))
8465 return nfserr_reclaim_bad;
8471 * Since the lifetime of a delegation isn't limited to that of an open, a
8472 * client may quite reasonably hang on to a delegation as long as it has
8473 * the inode cached. This becomes an obvious problem the first time a
8474 * client's inode cache approaches the size of the server's total memory.
8476 * For now we avoid this problem by imposing a hard limit on the number
8477 * of delegations, which varies according to the server's memory size.
8480 set_max_delegations(void)
8483 * Allow at most 4 delegations per megabyte of RAM. Quick
8484 * estimates suggest that in the worst case (where every delegation
8485 * is for a different inode), a delegation could take about 1.5K,
8486 * giving a worst case usage of about 6% of memory.
8488 max_delegations = nr_free_buffer_pages() >> (20 - 2 - PAGE_SHIFT);
8491 static int nfs4_state_create_net(struct net *net)
8493 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
8496 nn->conf_id_hashtbl = kmalloc_array(CLIENT_HASH_SIZE,
8497 sizeof(struct list_head),
8499 if (!nn->conf_id_hashtbl)
8501 nn->unconf_id_hashtbl = kmalloc_array(CLIENT_HASH_SIZE,
8502 sizeof(struct list_head),
8504 if (!nn->unconf_id_hashtbl)
8506 nn->sessionid_hashtbl = kmalloc_array(SESSION_HASH_SIZE,
8507 sizeof(struct list_head),
8509 if (!nn->sessionid_hashtbl)
8512 for (i = 0; i < CLIENT_HASH_SIZE; i++) {
8513 INIT_LIST_HEAD(&nn->conf_id_hashtbl[i]);
8514 INIT_LIST_HEAD(&nn->unconf_id_hashtbl[i]);
8516 for (i = 0; i < SESSION_HASH_SIZE; i++)
8517 INIT_LIST_HEAD(&nn->sessionid_hashtbl[i]);
8518 nn->conf_name_tree = RB_ROOT;
8519 nn->unconf_name_tree = RB_ROOT;
8520 nn->boot_time = ktime_get_real_seconds();
8521 nn->grace_ended = false;
8522 nn->nfsd4_manager.block_opens = true;
8523 INIT_LIST_HEAD(&nn->nfsd4_manager.list);
8524 INIT_LIST_HEAD(&nn->client_lru);
8525 INIT_LIST_HEAD(&nn->close_lru);
8526 INIT_LIST_HEAD(&nn->del_recall_lru);
8527 spin_lock_init(&nn->client_lock);
8528 spin_lock_init(&nn->s2s_cp_lock);
8529 idr_init(&nn->s2s_cp_stateids);
8531 spin_lock_init(&nn->blocked_locks_lock);
8532 INIT_LIST_HEAD(&nn->blocked_locks_lru);
8534 INIT_DELAYED_WORK(&nn->laundromat_work, laundromat_main);
8535 INIT_WORK(&nn->nfsd_shrinker_work, nfsd4_state_shrinker_worker);
8538 nn->nfsd_client_shrinker = shrinker_alloc(0, "nfsd-client");
8539 if (!nn->nfsd_client_shrinker)
8542 nn->nfsd_client_shrinker->scan_objects = nfsd4_state_shrinker_scan;
8543 nn->nfsd_client_shrinker->count_objects = nfsd4_state_shrinker_count;
8544 nn->nfsd_client_shrinker->private_data = nn;
8546 shrinker_register(nn->nfsd_client_shrinker);
8552 kfree(nn->sessionid_hashtbl);
8554 kfree(nn->unconf_id_hashtbl);
8556 kfree(nn->conf_id_hashtbl);
8562 nfs4_state_destroy_net(struct net *net)
8565 struct nfs4_client *clp = NULL;
8566 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
8568 for (i = 0; i < CLIENT_HASH_SIZE; i++) {
8569 while (!list_empty(&nn->conf_id_hashtbl[i])) {
8570 clp = list_entry(nn->conf_id_hashtbl[i].next, struct nfs4_client, cl_idhash);
8571 destroy_client(clp);
8575 WARN_ON(!list_empty(&nn->blocked_locks_lru));
8577 for (i = 0; i < CLIENT_HASH_SIZE; i++) {
8578 while (!list_empty(&nn->unconf_id_hashtbl[i])) {
8579 clp = list_entry(nn->unconf_id_hashtbl[i].next, struct nfs4_client, cl_idhash);
8580 destroy_client(clp);
8584 kfree(nn->sessionid_hashtbl);
8585 kfree(nn->unconf_id_hashtbl);
8586 kfree(nn->conf_id_hashtbl);
8591 nfs4_state_start_net(struct net *net)
8593 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
8596 ret = nfs4_state_create_net(net);
8599 locks_start_grace(net, &nn->nfsd4_manager);
8600 nfsd4_client_tracking_init(net);
8601 if (nn->track_reclaim_completes && nn->reclaim_str_hashtbl_size == 0)
8603 printk(KERN_INFO "NFSD: starting %lld-second grace period (net %x)\n",
8604 nn->nfsd4_grace, net->ns.inum);
8605 trace_nfsd_grace_start(nn);
8606 queue_delayed_work(laundry_wq, &nn->laundromat_work, nn->nfsd4_grace * HZ);
8610 printk(KERN_INFO "NFSD: no clients to reclaim, skipping NFSv4 grace period (net %x)\n",
8612 queue_delayed_work(laundry_wq, &nn->laundromat_work, nn->nfsd4_lease * HZ);
8613 nfsd4_end_grace(nn);
8617 /* initialization to perform when the nfsd service is started: */
8620 nfs4_state_start(void)
8624 ret = rhltable_init(&nfs4_file_rhltable, &nfs4_file_rhash_params);
8628 ret = nfsd4_create_callback_queue();
8630 rhltable_destroy(&nfs4_file_rhltable);
8634 set_max_delegations();
8639 nfs4_state_shutdown_net(struct net *net)
8641 struct nfs4_delegation *dp = NULL;
8642 struct list_head *pos, *next, reaplist;
8643 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
8645 shrinker_free(nn->nfsd_client_shrinker);
8646 cancel_work(&nn->nfsd_shrinker_work);
8647 cancel_delayed_work_sync(&nn->laundromat_work);
8648 locks_end_grace(&nn->nfsd4_manager);
8650 INIT_LIST_HEAD(&reaplist);
8651 spin_lock(&state_lock);
8652 list_for_each_safe(pos, next, &nn->del_recall_lru) {
8653 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
8654 unhash_delegation_locked(dp, SC_STATUS_CLOSED);
8655 list_add(&dp->dl_recall_lru, &reaplist);
8657 spin_unlock(&state_lock);
8658 list_for_each_safe(pos, next, &reaplist) {
8659 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
8660 list_del_init(&dp->dl_recall_lru);
8661 destroy_unhashed_deleg(dp);
8664 nfsd4_client_tracking_exit(net);
8665 nfs4_state_destroy_net(net);
8666 #ifdef CONFIG_NFSD_V4_2_INTER_SSC
8667 nfsd4_ssc_shutdown_umount(nn);
8672 nfs4_state_shutdown(void)
8674 nfsd4_destroy_callback_queue();
8675 rhltable_destroy(&nfs4_file_rhltable);
8679 get_stateid(struct nfsd4_compound_state *cstate, stateid_t *stateid)
8681 if (HAS_CSTATE_FLAG(cstate, CURRENT_STATE_ID_FLAG) &&
8682 CURRENT_STATEID(stateid))
8683 memcpy(stateid, &cstate->current_stateid, sizeof(stateid_t));
8687 put_stateid(struct nfsd4_compound_state *cstate, stateid_t *stateid)
8689 if (cstate->minorversion) {
8690 memcpy(&cstate->current_stateid, stateid, sizeof(stateid_t));
8691 SET_CSTATE_FLAG(cstate, CURRENT_STATE_ID_FLAG);
8696 clear_current_stateid(struct nfsd4_compound_state *cstate)
8698 CLEAR_CSTATE_FLAG(cstate, CURRENT_STATE_ID_FLAG);
8702 * functions to set current state id
8705 nfsd4_set_opendowngradestateid(struct nfsd4_compound_state *cstate,
8706 union nfsd4_op_u *u)
8708 put_stateid(cstate, &u->open_downgrade.od_stateid);
8712 nfsd4_set_openstateid(struct nfsd4_compound_state *cstate,
8713 union nfsd4_op_u *u)
8715 put_stateid(cstate, &u->open.op_stateid);
8719 nfsd4_set_closestateid(struct nfsd4_compound_state *cstate,
8720 union nfsd4_op_u *u)
8722 put_stateid(cstate, &u->close.cl_stateid);
8726 nfsd4_set_lockstateid(struct nfsd4_compound_state *cstate,
8727 union nfsd4_op_u *u)
8729 put_stateid(cstate, &u->lock.lk_resp_stateid);
8733 * functions to consume current state id
8737 nfsd4_get_opendowngradestateid(struct nfsd4_compound_state *cstate,
8738 union nfsd4_op_u *u)
8740 get_stateid(cstate, &u->open_downgrade.od_stateid);
8744 nfsd4_get_delegreturnstateid(struct nfsd4_compound_state *cstate,
8745 union nfsd4_op_u *u)
8747 get_stateid(cstate, &u->delegreturn.dr_stateid);
8751 nfsd4_get_freestateid(struct nfsd4_compound_state *cstate,
8752 union nfsd4_op_u *u)
8754 get_stateid(cstate, &u->free_stateid.fr_stateid);
8758 nfsd4_get_setattrstateid(struct nfsd4_compound_state *cstate,
8759 union nfsd4_op_u *u)
8761 get_stateid(cstate, &u->setattr.sa_stateid);
8765 nfsd4_get_closestateid(struct nfsd4_compound_state *cstate,
8766 union nfsd4_op_u *u)
8768 get_stateid(cstate, &u->close.cl_stateid);
8772 nfsd4_get_lockustateid(struct nfsd4_compound_state *cstate,
8773 union nfsd4_op_u *u)
8775 get_stateid(cstate, &u->locku.lu_stateid);
8779 nfsd4_get_readstateid(struct nfsd4_compound_state *cstate,
8780 union nfsd4_op_u *u)
8782 get_stateid(cstate, &u->read.rd_stateid);
8786 nfsd4_get_writestateid(struct nfsd4_compound_state *cstate,
8787 union nfsd4_op_u *u)
8789 get_stateid(cstate, &u->write.wr_stateid);
8793 * nfsd4_deleg_getattr_conflict - Recall if GETATTR causes conflict
8794 * @rqstp: RPC transaction context
8795 * @inode: file to be checked for a conflict
8796 * @modified: return true if file was modified
8797 * @size: new size of file if modified is true
8799 * This function is called when there is a conflict between a write
8800 * delegation and a change/size GETATTR from another client. The server
8801 * must either use the CB_GETATTR to get the current values of the
8802 * attributes from the client that holds the delegation or recall the
8803 * delegation before replying to the GETATTR. See RFC 8881 section
8806 * Returns 0 if there is no conflict; otherwise an nfs_stat
8810 nfsd4_deleg_getattr_conflict(struct svc_rqst *rqstp, struct inode *inode,
8811 bool *modified, u64 *size)
8814 struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
8815 struct file_lock_context *ctx;
8816 struct file_lease *fl;
8817 struct nfs4_delegation *dp;
8819 struct nfs4_cb_fattr *ncf;
8822 ctx = locks_inode_context(inode);
8825 spin_lock(&ctx->flc_lock);
8826 for_each_file_lock(fl, &ctx->flc_lease) {
8827 unsigned char type = fl->c.flc_type;
8829 if (fl->c.flc_flags == FL_LAYOUT)
8831 if (fl->fl_lmops != &nfsd_lease_mng_ops) {
8833 * non-nfs lease, if it's a lease with F_RDLCK then
8834 * we are done; there isn't any write delegation
8837 if (type == F_RDLCK)
8841 if (type == F_WRLCK) {
8842 dp = fl->c.flc_owner;
8843 if (dp->dl_recall.cb_clp == *(rqstp->rq_lease_breaker)) {
8844 spin_unlock(&ctx->flc_lock);
8848 nfsd_stats_wdeleg_getattr_inc(nn);
8849 dp = fl->c.flc_owner;
8850 ncf = &dp->dl_cb_fattr;
8851 nfs4_cb_getattr(&dp->dl_cb_fattr);
8852 spin_unlock(&ctx->flc_lock);
8853 wait_on_bit_timeout(&ncf->ncf_cb_flags, CB_GETATTR_BUSY,
8854 TASK_INTERRUPTIBLE, NFSD_CB_GETATTR_TIMEOUT);
8855 if (ncf->ncf_cb_status) {
8856 /* Recall delegation only if client didn't respond */
8857 status = nfserrno(nfsd_open_break_lease(inode, NFSD_MAY_READ));
8858 if (status != nfserr_jukebox ||
8859 !nfsd_wait_for_delegreturn(rqstp, inode))
8862 if (!ncf->ncf_file_modified &&
8863 (ncf->ncf_initial_cinfo != ncf->ncf_cb_change ||
8864 ncf->ncf_cur_fsize != ncf->ncf_cb_fsize))
8865 ncf->ncf_file_modified = true;
8866 if (ncf->ncf_file_modified) {
8868 * Per section 10.4.3 of RFC 8881, the server would
8869 * not update the file's metadata with the client's
8872 attrs.ia_mtime = attrs.ia_ctime = current_time(inode);
8873 attrs.ia_valid = ATTR_MTIME | ATTR_CTIME;
8874 setattr_copy(&nop_mnt_idmap, inode, &attrs);
8875 mark_inode_dirty(inode);
8876 ncf->ncf_cur_fsize = ncf->ncf_cb_fsize;
8877 *size = ncf->ncf_cur_fsize;
8884 spin_unlock(&ctx->flc_lock);