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);
94 * Currently used for the del_recall_lru and file hash table. In an
95 * effort to decrease the scope of the client_mutex, this spinlock may
96 * eventually cover more:
98 static DEFINE_SPINLOCK(state_lock);
100 enum nfsd4_st_mutex_lock_subclass {
101 OPEN_STATEID_MUTEX = 0,
102 LOCK_STATEID_MUTEX = 1,
106 * A waitqueue for all in-progress 4.0 CLOSE operations that are waiting for
107 * the refcount on the open stateid to drop.
109 static DECLARE_WAIT_QUEUE_HEAD(close_wq);
112 * A waitqueue where a writer to clients/#/ctl destroying a client can
113 * wait for cl_rpc_users to drop to 0 and then for the client to be
116 static DECLARE_WAIT_QUEUE_HEAD(expiry_wq);
118 static struct kmem_cache *client_slab;
119 static struct kmem_cache *openowner_slab;
120 static struct kmem_cache *lockowner_slab;
121 static struct kmem_cache *file_slab;
122 static struct kmem_cache *stateid_slab;
123 static struct kmem_cache *deleg_slab;
124 static struct kmem_cache *odstate_slab;
126 static void free_session(struct nfsd4_session *);
128 static const struct nfsd4_callback_ops nfsd4_cb_recall_ops;
129 static const struct nfsd4_callback_ops nfsd4_cb_notify_lock_ops;
131 static struct workqueue_struct *laundry_wq;
133 int nfsd4_create_laundry_wq(void)
137 laundry_wq = alloc_workqueue("%s", WQ_UNBOUND, 0, "nfsd4");
138 if (laundry_wq == NULL)
143 void nfsd4_destroy_laundry_wq(void)
145 destroy_workqueue(laundry_wq);
148 static bool is_session_dead(struct nfsd4_session *ses)
150 return ses->se_flags & NFS4_SESSION_DEAD;
153 static __be32 mark_session_dead_locked(struct nfsd4_session *ses, int ref_held_by_me)
155 if (atomic_read(&ses->se_ref) > ref_held_by_me)
156 return nfserr_jukebox;
157 ses->se_flags |= NFS4_SESSION_DEAD;
161 static bool is_client_expired(struct nfs4_client *clp)
163 return clp->cl_time == 0;
166 static void nfsd4_dec_courtesy_client_count(struct nfsd_net *nn,
167 struct nfs4_client *clp)
169 if (clp->cl_state != NFSD4_ACTIVE)
170 atomic_add_unless(&nn->nfsd_courtesy_clients, -1, 0);
173 static __be32 get_client_locked(struct nfs4_client *clp)
175 struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
177 lockdep_assert_held(&nn->client_lock);
179 if (is_client_expired(clp))
180 return nfserr_expired;
181 atomic_inc(&clp->cl_rpc_users);
182 nfsd4_dec_courtesy_client_count(nn, clp);
183 clp->cl_state = NFSD4_ACTIVE;
187 /* must be called under the client_lock */
189 renew_client_locked(struct nfs4_client *clp)
191 struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
193 if (is_client_expired(clp)) {
195 printk("%s: client (clientid %08x/%08x) already expired\n",
197 clp->cl_clientid.cl_boot,
198 clp->cl_clientid.cl_id);
202 list_move_tail(&clp->cl_lru, &nn->client_lru);
203 clp->cl_time = ktime_get_boottime_seconds();
204 nfsd4_dec_courtesy_client_count(nn, clp);
205 clp->cl_state = NFSD4_ACTIVE;
208 static void put_client_renew_locked(struct nfs4_client *clp)
210 struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
212 lockdep_assert_held(&nn->client_lock);
214 if (!atomic_dec_and_test(&clp->cl_rpc_users))
216 if (!is_client_expired(clp))
217 renew_client_locked(clp);
219 wake_up_all(&expiry_wq);
222 static void put_client_renew(struct nfs4_client *clp)
224 struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
226 if (!atomic_dec_and_lock(&clp->cl_rpc_users, &nn->client_lock))
228 if (!is_client_expired(clp))
229 renew_client_locked(clp);
231 wake_up_all(&expiry_wq);
232 spin_unlock(&nn->client_lock);
235 static __be32 nfsd4_get_session_locked(struct nfsd4_session *ses)
239 if (is_session_dead(ses))
240 return nfserr_badsession;
241 status = get_client_locked(ses->se_client);
244 atomic_inc(&ses->se_ref);
248 static void nfsd4_put_session_locked(struct nfsd4_session *ses)
250 struct nfs4_client *clp = ses->se_client;
251 struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
253 lockdep_assert_held(&nn->client_lock);
255 if (atomic_dec_and_test(&ses->se_ref) && is_session_dead(ses))
257 put_client_renew_locked(clp);
260 static void nfsd4_put_session(struct nfsd4_session *ses)
262 struct nfs4_client *clp = ses->se_client;
263 struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
265 spin_lock(&nn->client_lock);
266 nfsd4_put_session_locked(ses);
267 spin_unlock(&nn->client_lock);
270 static struct nfsd4_blocked_lock *
271 find_blocked_lock(struct nfs4_lockowner *lo, struct knfsd_fh *fh,
274 struct nfsd4_blocked_lock *cur, *found = NULL;
276 spin_lock(&nn->blocked_locks_lock);
277 list_for_each_entry(cur, &lo->lo_blocked, nbl_list) {
278 if (fh_match(fh, &cur->nbl_fh)) {
279 list_del_init(&cur->nbl_list);
280 WARN_ON(list_empty(&cur->nbl_lru));
281 list_del_init(&cur->nbl_lru);
286 spin_unlock(&nn->blocked_locks_lock);
288 locks_delete_block(&found->nbl_lock);
292 static struct nfsd4_blocked_lock *
293 find_or_allocate_block(struct nfs4_lockowner *lo, struct knfsd_fh *fh,
296 struct nfsd4_blocked_lock *nbl;
298 nbl = find_blocked_lock(lo, fh, nn);
300 nbl= kmalloc(sizeof(*nbl), GFP_KERNEL);
302 INIT_LIST_HEAD(&nbl->nbl_list);
303 INIT_LIST_HEAD(&nbl->nbl_lru);
304 fh_copy_shallow(&nbl->nbl_fh, fh);
305 locks_init_lock(&nbl->nbl_lock);
306 kref_init(&nbl->nbl_kref);
307 nfsd4_init_cb(&nbl->nbl_cb, lo->lo_owner.so_client,
308 &nfsd4_cb_notify_lock_ops,
309 NFSPROC4_CLNT_CB_NOTIFY_LOCK);
316 free_nbl(struct kref *kref)
318 struct nfsd4_blocked_lock *nbl;
320 nbl = container_of(kref, struct nfsd4_blocked_lock, nbl_kref);
325 free_blocked_lock(struct nfsd4_blocked_lock *nbl)
327 locks_delete_block(&nbl->nbl_lock);
328 locks_release_private(&nbl->nbl_lock);
329 kref_put(&nbl->nbl_kref, free_nbl);
333 remove_blocked_locks(struct nfs4_lockowner *lo)
335 struct nfs4_client *clp = lo->lo_owner.so_client;
336 struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
337 struct nfsd4_blocked_lock *nbl;
340 /* Dequeue all blocked locks */
341 spin_lock(&nn->blocked_locks_lock);
342 while (!list_empty(&lo->lo_blocked)) {
343 nbl = list_first_entry(&lo->lo_blocked,
344 struct nfsd4_blocked_lock,
346 list_del_init(&nbl->nbl_list);
347 WARN_ON(list_empty(&nbl->nbl_lru));
348 list_move(&nbl->nbl_lru, &reaplist);
350 spin_unlock(&nn->blocked_locks_lock);
353 while (!list_empty(&reaplist)) {
354 nbl = list_first_entry(&reaplist, struct nfsd4_blocked_lock,
356 list_del_init(&nbl->nbl_lru);
357 free_blocked_lock(nbl);
362 nfsd4_cb_notify_lock_prepare(struct nfsd4_callback *cb)
364 struct nfsd4_blocked_lock *nbl = container_of(cb,
365 struct nfsd4_blocked_lock, nbl_cb);
366 locks_delete_block(&nbl->nbl_lock);
370 nfsd4_cb_notify_lock_done(struct nfsd4_callback *cb, struct rpc_task *task)
372 trace_nfsd_cb_notify_lock_done(&zero_stateid, task);
375 * Since this is just an optimization, we don't try very hard if it
376 * turns out not to succeed. We'll requeue it on NFS4ERR_DELAY, and
377 * just quit trying on anything else.
379 switch (task->tk_status) {
381 rpc_delay(task, 1 * HZ);
389 nfsd4_cb_notify_lock_release(struct nfsd4_callback *cb)
391 struct nfsd4_blocked_lock *nbl = container_of(cb,
392 struct nfsd4_blocked_lock, nbl_cb);
394 free_blocked_lock(nbl);
397 static const struct nfsd4_callback_ops nfsd4_cb_notify_lock_ops = {
398 .prepare = nfsd4_cb_notify_lock_prepare,
399 .done = nfsd4_cb_notify_lock_done,
400 .release = nfsd4_cb_notify_lock_release,
404 * We store the NONE, READ, WRITE, and BOTH bits separately in the
405 * st_{access,deny}_bmap field of the stateid, in order to track not
406 * only what share bits are currently in force, but also what
407 * combinations of share bits previous opens have used. This allows us
408 * to enforce the recommendation in
409 * https://datatracker.ietf.org/doc/html/rfc7530#section-16.19.4 that
410 * the server return an error if the client attempt to downgrade to a
411 * combination of share bits not explicable by closing some of its
414 * This enforcement is arguably incomplete, since we don't keep
415 * track of access/deny bit combinations; so, e.g., we allow:
417 * OPEN allow read, deny write
418 * OPEN allow both, deny none
419 * DOWNGRADE allow read, deny none
421 * which we should reject.
423 * But you could also argue that our current code is already overkill,
424 * since it only exists to return NFS4ERR_INVAL on incorrect client
428 bmap_to_share_mode(unsigned long bmap)
431 unsigned int access = 0;
433 for (i = 1; i < 4; i++) {
434 if (test_bit(i, &bmap))
440 /* set share access for a given stateid */
442 set_access(u32 access, struct nfs4_ol_stateid *stp)
444 unsigned char mask = 1 << access;
446 WARN_ON_ONCE(access > NFS4_SHARE_ACCESS_BOTH);
447 stp->st_access_bmap |= mask;
450 /* clear share access for a given stateid */
452 clear_access(u32 access, struct nfs4_ol_stateid *stp)
454 unsigned char mask = 1 << access;
456 WARN_ON_ONCE(access > NFS4_SHARE_ACCESS_BOTH);
457 stp->st_access_bmap &= ~mask;
460 /* test whether a given stateid has access */
462 test_access(u32 access, struct nfs4_ol_stateid *stp)
464 unsigned char mask = 1 << access;
466 return (bool)(stp->st_access_bmap & mask);
469 /* set share deny for a given stateid */
471 set_deny(u32 deny, struct nfs4_ol_stateid *stp)
473 unsigned char mask = 1 << deny;
475 WARN_ON_ONCE(deny > NFS4_SHARE_DENY_BOTH);
476 stp->st_deny_bmap |= mask;
479 /* clear share deny for a given stateid */
481 clear_deny(u32 deny, struct nfs4_ol_stateid *stp)
483 unsigned char mask = 1 << deny;
485 WARN_ON_ONCE(deny > NFS4_SHARE_DENY_BOTH);
486 stp->st_deny_bmap &= ~mask;
489 /* test whether a given stateid is denying specific access */
491 test_deny(u32 deny, struct nfs4_ol_stateid *stp)
493 unsigned char mask = 1 << deny;
495 return (bool)(stp->st_deny_bmap & mask);
498 static int nfs4_access_to_omode(u32 access)
500 switch (access & NFS4_SHARE_ACCESS_BOTH) {
501 case NFS4_SHARE_ACCESS_READ:
503 case NFS4_SHARE_ACCESS_WRITE:
505 case NFS4_SHARE_ACCESS_BOTH:
513 access_permit_read(struct nfs4_ol_stateid *stp)
515 return test_access(NFS4_SHARE_ACCESS_READ, stp) ||
516 test_access(NFS4_SHARE_ACCESS_BOTH, stp) ||
517 test_access(NFS4_SHARE_ACCESS_WRITE, stp);
521 access_permit_write(struct nfs4_ol_stateid *stp)
523 return test_access(NFS4_SHARE_ACCESS_WRITE, stp) ||
524 test_access(NFS4_SHARE_ACCESS_BOTH, stp);
527 static inline struct nfs4_stateowner *
528 nfs4_get_stateowner(struct nfs4_stateowner *sop)
530 atomic_inc(&sop->so_count);
535 same_owner_str(struct nfs4_stateowner *sop, struct xdr_netobj *owner)
537 return (sop->so_owner.len == owner->len) &&
538 0 == memcmp(sop->so_owner.data, owner->data, owner->len);
541 static struct nfs4_openowner *
542 find_openstateowner_str_locked(unsigned int hashval, struct nfsd4_open *open,
543 struct nfs4_client *clp)
545 struct nfs4_stateowner *so;
547 lockdep_assert_held(&clp->cl_lock);
549 list_for_each_entry(so, &clp->cl_ownerstr_hashtbl[hashval],
551 if (!so->so_is_open_owner)
553 if (same_owner_str(so, &open->op_owner))
554 return openowner(nfs4_get_stateowner(so));
559 static struct nfs4_openowner *
560 find_openstateowner_str(unsigned int hashval, struct nfsd4_open *open,
561 struct nfs4_client *clp)
563 struct nfs4_openowner *oo;
565 spin_lock(&clp->cl_lock);
566 oo = find_openstateowner_str_locked(hashval, open, clp);
567 spin_unlock(&clp->cl_lock);
572 opaque_hashval(const void *ptr, int nbytes)
574 unsigned char *cptr = (unsigned char *) ptr;
584 static void nfsd4_free_file_rcu(struct rcu_head *rcu)
586 struct nfs4_file *fp = container_of(rcu, struct nfs4_file, fi_rcu);
588 kmem_cache_free(file_slab, fp);
592 put_nfs4_file(struct nfs4_file *fi)
594 if (refcount_dec_and_test(&fi->fi_ref)) {
595 nfsd4_file_hash_remove(fi);
596 WARN_ON_ONCE(!list_empty(&fi->fi_clnt_odstate));
597 WARN_ON_ONCE(!list_empty(&fi->fi_delegations));
598 call_rcu(&fi->fi_rcu, nfsd4_free_file_rcu);
602 static struct nfsd_file *
603 __nfs4_get_fd(struct nfs4_file *f, int oflag)
605 if (f->fi_fds[oflag])
606 return nfsd_file_get(f->fi_fds[oflag]);
610 static struct nfsd_file *
611 find_writeable_file_locked(struct nfs4_file *f)
613 struct nfsd_file *ret;
615 lockdep_assert_held(&f->fi_lock);
617 ret = __nfs4_get_fd(f, O_WRONLY);
619 ret = __nfs4_get_fd(f, O_RDWR);
623 static struct nfsd_file *
624 find_writeable_file(struct nfs4_file *f)
626 struct nfsd_file *ret;
628 spin_lock(&f->fi_lock);
629 ret = find_writeable_file_locked(f);
630 spin_unlock(&f->fi_lock);
635 static struct nfsd_file *
636 find_readable_file_locked(struct nfs4_file *f)
638 struct nfsd_file *ret;
640 lockdep_assert_held(&f->fi_lock);
642 ret = __nfs4_get_fd(f, O_RDONLY);
644 ret = __nfs4_get_fd(f, O_RDWR);
648 static struct nfsd_file *
649 find_readable_file(struct nfs4_file *f)
651 struct nfsd_file *ret;
653 spin_lock(&f->fi_lock);
654 ret = find_readable_file_locked(f);
655 spin_unlock(&f->fi_lock);
661 find_any_file(struct nfs4_file *f)
663 struct nfsd_file *ret;
667 spin_lock(&f->fi_lock);
668 ret = __nfs4_get_fd(f, O_RDWR);
670 ret = __nfs4_get_fd(f, O_WRONLY);
672 ret = __nfs4_get_fd(f, O_RDONLY);
674 spin_unlock(&f->fi_lock);
678 static struct nfsd_file *find_any_file_locked(struct nfs4_file *f)
680 lockdep_assert_held(&f->fi_lock);
682 if (f->fi_fds[O_RDWR])
683 return f->fi_fds[O_RDWR];
684 if (f->fi_fds[O_WRONLY])
685 return f->fi_fds[O_WRONLY];
686 if (f->fi_fds[O_RDONLY])
687 return f->fi_fds[O_RDONLY];
691 static struct nfsd_file *find_deleg_file_locked(struct nfs4_file *f)
693 lockdep_assert_held(&f->fi_lock);
695 if (f->fi_deleg_file)
696 return f->fi_deleg_file;
700 static atomic_long_t num_delegations;
701 unsigned long max_delegations;
704 * Open owner state (share locks)
707 /* hash tables for lock and open owners */
708 #define OWNER_HASH_BITS 8
709 #define OWNER_HASH_SIZE (1 << OWNER_HASH_BITS)
710 #define OWNER_HASH_MASK (OWNER_HASH_SIZE - 1)
712 static unsigned int ownerstr_hashval(struct xdr_netobj *ownername)
716 ret = opaque_hashval(ownername->data, ownername->len);
717 return ret & OWNER_HASH_MASK;
720 static struct rhltable nfs4_file_rhltable ____cacheline_aligned_in_smp;
722 static const struct rhashtable_params nfs4_file_rhash_params = {
723 .key_len = sizeof_field(struct nfs4_file, fi_inode),
724 .key_offset = offsetof(struct nfs4_file, fi_inode),
725 .head_offset = offsetof(struct nfs4_file, fi_rlist),
728 * Start with a single page hash table to reduce resizing churn
729 * on light workloads.
732 .automatic_shrinking = true,
736 * Check if courtesy clients have conflicting access and resolve it if possible
738 * access: is op_share_access if share_access is true.
739 * Check if access mode, op_share_access, would conflict with
740 * the current deny mode of the file 'fp'.
741 * access: is op_share_deny if share_access is false.
742 * Check if the deny mode, op_share_deny, would conflict with
743 * current access of the file 'fp'.
744 * stp: skip checking this entry.
745 * new_stp: normal open, not open upgrade.
748 * false - access/deny mode conflict with normal client.
749 * true - no conflict or conflict with courtesy client(s) is resolved.
752 nfs4_resolve_deny_conflicts_locked(struct nfs4_file *fp, bool new_stp,
753 struct nfs4_ol_stateid *stp, u32 access, bool share_access)
755 struct nfs4_ol_stateid *st;
756 bool resolvable = true;
759 struct nfs4_client *clp;
761 lockdep_assert_held(&fp->fi_lock);
762 list_for_each_entry(st, &fp->fi_stateids, st_perfile) {
763 /* ignore lock stateid */
766 if (st == stp && new_stp)
768 /* check file access against deny mode or vice versa */
769 bmap = share_access ? st->st_deny_bmap : st->st_access_bmap;
770 if (!(access & bmap_to_share_mode(bmap)))
772 clp = st->st_stid.sc_client;
773 if (try_to_expire_client(clp))
779 clp = stp->st_stid.sc_client;
780 nn = net_generic(clp->net, nfsd_net_id);
781 mod_delayed_work(laundry_wq, &nn->laundromat_work, 0);
787 __nfs4_file_get_access(struct nfs4_file *fp, u32 access)
789 lockdep_assert_held(&fp->fi_lock);
791 if (access & NFS4_SHARE_ACCESS_WRITE)
792 atomic_inc(&fp->fi_access[O_WRONLY]);
793 if (access & NFS4_SHARE_ACCESS_READ)
794 atomic_inc(&fp->fi_access[O_RDONLY]);
798 nfs4_file_get_access(struct nfs4_file *fp, u32 access)
800 lockdep_assert_held(&fp->fi_lock);
802 /* Does this access mode make sense? */
803 if (access & ~NFS4_SHARE_ACCESS_BOTH)
806 /* Does it conflict with a deny mode already set? */
807 if ((access & fp->fi_share_deny) != 0)
808 return nfserr_share_denied;
810 __nfs4_file_get_access(fp, access);
814 static __be32 nfs4_file_check_deny(struct nfs4_file *fp, u32 deny)
816 /* Common case is that there is no deny mode. */
818 /* Does this deny mode make sense? */
819 if (deny & ~NFS4_SHARE_DENY_BOTH)
822 if ((deny & NFS4_SHARE_DENY_READ) &&
823 atomic_read(&fp->fi_access[O_RDONLY]))
824 return nfserr_share_denied;
826 if ((deny & NFS4_SHARE_DENY_WRITE) &&
827 atomic_read(&fp->fi_access[O_WRONLY]))
828 return nfserr_share_denied;
833 static void __nfs4_file_put_access(struct nfs4_file *fp, int oflag)
835 might_lock(&fp->fi_lock);
837 if (atomic_dec_and_lock(&fp->fi_access[oflag], &fp->fi_lock)) {
838 struct nfsd_file *f1 = NULL;
839 struct nfsd_file *f2 = NULL;
841 swap(f1, fp->fi_fds[oflag]);
842 if (atomic_read(&fp->fi_access[1 - oflag]) == 0)
843 swap(f2, fp->fi_fds[O_RDWR]);
844 spin_unlock(&fp->fi_lock);
852 static void nfs4_file_put_access(struct nfs4_file *fp, u32 access)
854 WARN_ON_ONCE(access & ~NFS4_SHARE_ACCESS_BOTH);
856 if (access & NFS4_SHARE_ACCESS_WRITE)
857 __nfs4_file_put_access(fp, O_WRONLY);
858 if (access & NFS4_SHARE_ACCESS_READ)
859 __nfs4_file_put_access(fp, O_RDONLY);
863 * Allocate a new open/delegation state counter. This is needed for
864 * pNFS for proper return on close semantics.
866 * Note that we only allocate it for pNFS-enabled exports, otherwise
867 * all pointers to struct nfs4_clnt_odstate are always NULL.
869 static struct nfs4_clnt_odstate *
870 alloc_clnt_odstate(struct nfs4_client *clp)
872 struct nfs4_clnt_odstate *co;
874 co = kmem_cache_zalloc(odstate_slab, GFP_KERNEL);
877 refcount_set(&co->co_odcount, 1);
883 hash_clnt_odstate_locked(struct nfs4_clnt_odstate *co)
885 struct nfs4_file *fp = co->co_file;
887 lockdep_assert_held(&fp->fi_lock);
888 list_add(&co->co_perfile, &fp->fi_clnt_odstate);
892 get_clnt_odstate(struct nfs4_clnt_odstate *co)
895 refcount_inc(&co->co_odcount);
899 put_clnt_odstate(struct nfs4_clnt_odstate *co)
901 struct nfs4_file *fp;
907 if (refcount_dec_and_lock(&co->co_odcount, &fp->fi_lock)) {
908 list_del(&co->co_perfile);
909 spin_unlock(&fp->fi_lock);
911 nfsd4_return_all_file_layouts(co->co_client, fp);
912 kmem_cache_free(odstate_slab, co);
916 static struct nfs4_clnt_odstate *
917 find_or_hash_clnt_odstate(struct nfs4_file *fp, struct nfs4_clnt_odstate *new)
919 struct nfs4_clnt_odstate *co;
920 struct nfs4_client *cl;
927 spin_lock(&fp->fi_lock);
928 list_for_each_entry(co, &fp->fi_clnt_odstate, co_perfile) {
929 if (co->co_client == cl) {
930 get_clnt_odstate(co);
936 hash_clnt_odstate_locked(new);
938 spin_unlock(&fp->fi_lock);
942 struct nfs4_stid *nfs4_alloc_stid(struct nfs4_client *cl, struct kmem_cache *slab,
943 void (*sc_free)(struct nfs4_stid *))
945 struct nfs4_stid *stid;
948 stid = kmem_cache_zalloc(slab, GFP_KERNEL);
952 idr_preload(GFP_KERNEL);
953 spin_lock(&cl->cl_lock);
954 /* Reserving 0 for start of file in nfsdfs "states" file: */
955 new_id = idr_alloc_cyclic(&cl->cl_stateids, stid, 1, 0, GFP_NOWAIT);
956 spin_unlock(&cl->cl_lock);
961 stid->sc_free = sc_free;
962 stid->sc_client = cl;
963 stid->sc_stateid.si_opaque.so_id = new_id;
964 stid->sc_stateid.si_opaque.so_clid = cl->cl_clientid;
965 /* Will be incremented before return to client: */
966 refcount_set(&stid->sc_count, 1);
967 spin_lock_init(&stid->sc_lock);
968 INIT_LIST_HEAD(&stid->sc_cp_list);
971 * It shouldn't be a problem to reuse an opaque stateid value.
972 * I don't think it is for 4.1. But with 4.0 I worry that, for
973 * example, a stray write retransmission could be accepted by
974 * the server when it should have been rejected. Therefore,
975 * adopt a trick from the sctp code to attempt to maximize the
976 * amount of time until an id is reused, by ensuring they always
977 * "increase" (mod INT_MAX):
981 kmem_cache_free(slab, stid);
986 * Create a unique stateid_t to represent each COPY.
988 static int nfs4_init_cp_state(struct nfsd_net *nn, copy_stateid_t *stid,
989 unsigned char cs_type)
993 stid->cs_stid.si_opaque.so_clid.cl_boot = (u32)nn->boot_time;
994 stid->cs_stid.si_opaque.so_clid.cl_id = nn->s2s_cp_cl_id;
995 stid->cs_type = cs_type;
997 idr_preload(GFP_KERNEL);
998 spin_lock(&nn->s2s_cp_lock);
999 new_id = idr_alloc_cyclic(&nn->s2s_cp_stateids, stid, 0, 0, GFP_NOWAIT);
1000 stid->cs_stid.si_opaque.so_id = new_id;
1001 stid->cs_stid.si_generation = 1;
1002 spin_unlock(&nn->s2s_cp_lock);
1009 int nfs4_init_copy_state(struct nfsd_net *nn, struct nfsd4_copy *copy)
1011 return nfs4_init_cp_state(nn, ©->cp_stateid, NFS4_COPY_STID);
1014 struct nfs4_cpntf_state *nfs4_alloc_init_cpntf_state(struct nfsd_net *nn,
1015 struct nfs4_stid *p_stid)
1017 struct nfs4_cpntf_state *cps;
1019 cps = kzalloc(sizeof(struct nfs4_cpntf_state), GFP_KERNEL);
1022 cps->cpntf_time = ktime_get_boottime_seconds();
1023 refcount_set(&cps->cp_stateid.cs_count, 1);
1024 if (!nfs4_init_cp_state(nn, &cps->cp_stateid, NFS4_COPYNOTIFY_STID))
1026 spin_lock(&nn->s2s_cp_lock);
1027 list_add(&cps->cp_list, &p_stid->sc_cp_list);
1028 spin_unlock(&nn->s2s_cp_lock);
1035 void nfs4_free_copy_state(struct nfsd4_copy *copy)
1037 struct nfsd_net *nn;
1039 WARN_ON_ONCE(copy->cp_stateid.cs_type != NFS4_COPY_STID);
1040 nn = net_generic(copy->cp_clp->net, nfsd_net_id);
1041 spin_lock(&nn->s2s_cp_lock);
1042 idr_remove(&nn->s2s_cp_stateids,
1043 copy->cp_stateid.cs_stid.si_opaque.so_id);
1044 spin_unlock(&nn->s2s_cp_lock);
1047 static void nfs4_free_cpntf_statelist(struct net *net, struct nfs4_stid *stid)
1049 struct nfs4_cpntf_state *cps;
1050 struct nfsd_net *nn;
1052 nn = net_generic(net, nfsd_net_id);
1053 spin_lock(&nn->s2s_cp_lock);
1054 while (!list_empty(&stid->sc_cp_list)) {
1055 cps = list_first_entry(&stid->sc_cp_list,
1056 struct nfs4_cpntf_state, cp_list);
1057 _free_cpntf_state_locked(nn, cps);
1059 spin_unlock(&nn->s2s_cp_lock);
1062 static struct nfs4_ol_stateid * nfs4_alloc_open_stateid(struct nfs4_client *clp)
1064 struct nfs4_stid *stid;
1066 stid = nfs4_alloc_stid(clp, stateid_slab, nfs4_free_ol_stateid);
1070 return openlockstateid(stid);
1073 static void nfs4_free_deleg(struct nfs4_stid *stid)
1075 struct nfs4_delegation *dp = delegstateid(stid);
1077 WARN_ON_ONCE(!list_empty(&stid->sc_cp_list));
1078 WARN_ON_ONCE(!list_empty(&dp->dl_perfile));
1079 WARN_ON_ONCE(!list_empty(&dp->dl_perclnt));
1080 WARN_ON_ONCE(!list_empty(&dp->dl_recall_lru));
1081 kmem_cache_free(deleg_slab, stid);
1082 atomic_long_dec(&num_delegations);
1086 * When we recall a delegation, we should be careful not to hand it
1087 * out again straight away.
1088 * To ensure this we keep a pair of bloom filters ('new' and 'old')
1089 * in which the filehandles of recalled delegations are "stored".
1090 * If a filehandle appear in either filter, a delegation is blocked.
1091 * When a delegation is recalled, the filehandle is stored in the "new"
1093 * Every 30 seconds we swap the filters and clear the "new" one,
1094 * unless both are empty of course.
1096 * Each filter is 256 bits. We hash the filehandle to 32bit and use the
1097 * low 3 bytes as hash-table indices.
1099 * 'blocked_delegations_lock', which is always taken in block_delegations(),
1100 * is used to manage concurrent access. Testing does not need the lock
1101 * except when swapping the two filters.
1103 static DEFINE_SPINLOCK(blocked_delegations_lock);
1104 static struct bloom_pair {
1105 int entries, old_entries;
1107 int new; /* index into 'set' */
1108 DECLARE_BITMAP(set[2], 256);
1109 } blocked_delegations;
1111 static int delegation_blocked(struct knfsd_fh *fh)
1114 struct bloom_pair *bd = &blocked_delegations;
1116 if (bd->entries == 0)
1118 if (ktime_get_seconds() - bd->swap_time > 30) {
1119 spin_lock(&blocked_delegations_lock);
1120 if (ktime_get_seconds() - bd->swap_time > 30) {
1121 bd->entries -= bd->old_entries;
1122 bd->old_entries = bd->entries;
1123 memset(bd->set[bd->new], 0,
1124 sizeof(bd->set[0]));
1125 bd->new = 1-bd->new;
1126 bd->swap_time = ktime_get_seconds();
1128 spin_unlock(&blocked_delegations_lock);
1130 hash = jhash(&fh->fh_raw, fh->fh_size, 0);
1131 if (test_bit(hash&255, bd->set[0]) &&
1132 test_bit((hash>>8)&255, bd->set[0]) &&
1133 test_bit((hash>>16)&255, bd->set[0]))
1136 if (test_bit(hash&255, bd->set[1]) &&
1137 test_bit((hash>>8)&255, bd->set[1]) &&
1138 test_bit((hash>>16)&255, bd->set[1]))
1144 static void block_delegations(struct knfsd_fh *fh)
1147 struct bloom_pair *bd = &blocked_delegations;
1149 hash = jhash(&fh->fh_raw, fh->fh_size, 0);
1151 spin_lock(&blocked_delegations_lock);
1152 __set_bit(hash&255, bd->set[bd->new]);
1153 __set_bit((hash>>8)&255, bd->set[bd->new]);
1154 __set_bit((hash>>16)&255, bd->set[bd->new]);
1155 if (bd->entries == 0)
1156 bd->swap_time = ktime_get_seconds();
1158 spin_unlock(&blocked_delegations_lock);
1161 static struct nfs4_delegation *
1162 alloc_init_deleg(struct nfs4_client *clp, struct nfs4_file *fp,
1163 struct nfs4_clnt_odstate *odstate)
1165 struct nfs4_delegation *dp;
1168 dprintk("NFSD alloc_init_deleg\n");
1169 n = atomic_long_inc_return(&num_delegations);
1170 if (n < 0 || n > max_delegations)
1172 if (delegation_blocked(&fp->fi_fhandle))
1174 dp = delegstateid(nfs4_alloc_stid(clp, deleg_slab, nfs4_free_deleg));
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 = NFS4_OPEN_DELEGATE_READ;
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);
1195 dp->dl_stid.sc_file = fp;
1198 atomic_long_dec(&num_delegations);
1203 nfs4_put_stid(struct nfs4_stid *s)
1205 struct nfs4_file *fp = s->sc_file;
1206 struct nfs4_client *clp = s->sc_client;
1208 might_lock(&clp->cl_lock);
1210 if (!refcount_dec_and_lock(&s->sc_count, &clp->cl_lock)) {
1211 wake_up_all(&close_wq);
1214 idr_remove(&clp->cl_stateids, s->sc_stateid.si_opaque.so_id);
1215 nfs4_free_cpntf_statelist(clp->net, s);
1216 spin_unlock(&clp->cl_lock);
1223 nfs4_inc_and_copy_stateid(stateid_t *dst, struct nfs4_stid *stid)
1225 stateid_t *src = &stid->sc_stateid;
1227 spin_lock(&stid->sc_lock);
1228 if (unlikely(++src->si_generation == 0))
1229 src->si_generation = 1;
1230 memcpy(dst, src, sizeof(*dst));
1231 spin_unlock(&stid->sc_lock);
1234 static void put_deleg_file(struct nfs4_file *fp)
1236 struct nfsd_file *nf = NULL;
1238 spin_lock(&fp->fi_lock);
1239 if (--fp->fi_delegees == 0)
1240 swap(nf, fp->fi_deleg_file);
1241 spin_unlock(&fp->fi_lock);
1247 static void nfs4_unlock_deleg_lease(struct nfs4_delegation *dp)
1249 struct nfs4_file *fp = dp->dl_stid.sc_file;
1250 struct nfsd_file *nf = fp->fi_deleg_file;
1252 WARN_ON_ONCE(!fp->fi_delegees);
1254 vfs_setlease(nf->nf_file, F_UNLCK, NULL, (void **)&dp);
1258 static void destroy_unhashed_deleg(struct nfs4_delegation *dp)
1260 put_clnt_odstate(dp->dl_clnt_odstate);
1261 nfs4_unlock_deleg_lease(dp);
1262 nfs4_put_stid(&dp->dl_stid);
1265 void nfs4_unhash_stid(struct nfs4_stid *s)
1271 * nfs4_delegation_exists - Discover if this delegation already exists
1272 * @clp: a pointer to the nfs4_client we're granting a delegation to
1273 * @fp: a pointer to the nfs4_file we're granting a delegation on
1276 * On success: true iff an existing delegation is found
1280 nfs4_delegation_exists(struct nfs4_client *clp, struct nfs4_file *fp)
1282 struct nfs4_delegation *searchdp = NULL;
1283 struct nfs4_client *searchclp = NULL;
1285 lockdep_assert_held(&state_lock);
1286 lockdep_assert_held(&fp->fi_lock);
1288 list_for_each_entry(searchdp, &fp->fi_delegations, dl_perfile) {
1289 searchclp = searchdp->dl_stid.sc_client;
1290 if (clp == searchclp) {
1298 * hash_delegation_locked - Add a delegation to the appropriate lists
1299 * @dp: a pointer to the nfs4_delegation we are adding.
1300 * @fp: a pointer to the nfs4_file we're granting a delegation on
1303 * On success: NULL if the delegation was successfully hashed.
1305 * On error: -EAGAIN if one was previously granted to this
1306 * nfs4_client for this nfs4_file. Delegation is not hashed.
1311 hash_delegation_locked(struct nfs4_delegation *dp, struct nfs4_file *fp)
1313 struct nfs4_client *clp = dp->dl_stid.sc_client;
1315 lockdep_assert_held(&state_lock);
1316 lockdep_assert_held(&fp->fi_lock);
1318 if (nfs4_delegation_exists(clp, fp))
1320 refcount_inc(&dp->dl_stid.sc_count);
1321 dp->dl_stid.sc_type = NFS4_DELEG_STID;
1322 list_add(&dp->dl_perfile, &fp->fi_delegations);
1323 list_add(&dp->dl_perclnt, &clp->cl_delegations);
1327 static bool delegation_hashed(struct nfs4_delegation *dp)
1329 return !(list_empty(&dp->dl_perfile));
1333 unhash_delegation_locked(struct nfs4_delegation *dp)
1335 struct nfs4_file *fp = dp->dl_stid.sc_file;
1337 lockdep_assert_held(&state_lock);
1339 if (!delegation_hashed(dp))
1342 dp->dl_stid.sc_type = NFS4_CLOSED_DELEG_STID;
1343 /* Ensure that deleg break won't try to requeue it */
1345 spin_lock(&fp->fi_lock);
1346 list_del_init(&dp->dl_perclnt);
1347 list_del_init(&dp->dl_recall_lru);
1348 list_del_init(&dp->dl_perfile);
1349 spin_unlock(&fp->fi_lock);
1353 static void destroy_delegation(struct nfs4_delegation *dp)
1357 spin_lock(&state_lock);
1358 unhashed = unhash_delegation_locked(dp);
1359 spin_unlock(&state_lock);
1361 destroy_unhashed_deleg(dp);
1364 static void revoke_delegation(struct nfs4_delegation *dp)
1366 struct nfs4_client *clp = dp->dl_stid.sc_client;
1368 WARN_ON(!list_empty(&dp->dl_recall_lru));
1370 trace_nfsd_stid_revoke(&dp->dl_stid);
1372 if (clp->cl_minorversion) {
1373 dp->dl_stid.sc_type = NFS4_REVOKED_DELEG_STID;
1374 refcount_inc(&dp->dl_stid.sc_count);
1375 spin_lock(&clp->cl_lock);
1376 list_add(&dp->dl_recall_lru, &clp->cl_revoked);
1377 spin_unlock(&clp->cl_lock);
1379 destroy_unhashed_deleg(dp);
1386 static unsigned int clientid_hashval(u32 id)
1388 return id & CLIENT_HASH_MASK;
1391 static unsigned int clientstr_hashval(struct xdr_netobj name)
1393 return opaque_hashval(name.data, 8) & CLIENT_HASH_MASK;
1397 * A stateid that had a deny mode associated with it is being released
1398 * or downgraded. Recalculate the deny mode on the file.
1401 recalculate_deny_mode(struct nfs4_file *fp)
1403 struct nfs4_ol_stateid *stp;
1405 spin_lock(&fp->fi_lock);
1406 fp->fi_share_deny = 0;
1407 list_for_each_entry(stp, &fp->fi_stateids, st_perfile)
1408 fp->fi_share_deny |= bmap_to_share_mode(stp->st_deny_bmap);
1409 spin_unlock(&fp->fi_lock);
1413 reset_union_bmap_deny(u32 deny, struct nfs4_ol_stateid *stp)
1416 bool change = false;
1418 for (i = 1; i < 4; i++) {
1419 if ((i & deny) != i) {
1425 /* Recalculate per-file deny mode if there was a change */
1427 recalculate_deny_mode(stp->st_stid.sc_file);
1430 /* release all access and file references for a given stateid */
1432 release_all_access(struct nfs4_ol_stateid *stp)
1435 struct nfs4_file *fp = stp->st_stid.sc_file;
1437 if (fp && stp->st_deny_bmap != 0)
1438 recalculate_deny_mode(fp);
1440 for (i = 1; i < 4; i++) {
1441 if (test_access(i, stp))
1442 nfs4_file_put_access(stp->st_stid.sc_file, i);
1443 clear_access(i, stp);
1447 static inline void nfs4_free_stateowner(struct nfs4_stateowner *sop)
1449 kfree(sop->so_owner.data);
1450 sop->so_ops->so_free(sop);
1453 static void nfs4_put_stateowner(struct nfs4_stateowner *sop)
1455 struct nfs4_client *clp = sop->so_client;
1457 might_lock(&clp->cl_lock);
1459 if (!atomic_dec_and_lock(&sop->so_count, &clp->cl_lock))
1461 sop->so_ops->so_unhash(sop);
1462 spin_unlock(&clp->cl_lock);
1463 nfs4_free_stateowner(sop);
1467 nfs4_ol_stateid_unhashed(const struct nfs4_ol_stateid *stp)
1469 return list_empty(&stp->st_perfile);
1472 static bool unhash_ol_stateid(struct nfs4_ol_stateid *stp)
1474 struct nfs4_file *fp = stp->st_stid.sc_file;
1476 lockdep_assert_held(&stp->st_stateowner->so_client->cl_lock);
1478 if (list_empty(&stp->st_perfile))
1481 spin_lock(&fp->fi_lock);
1482 list_del_init(&stp->st_perfile);
1483 spin_unlock(&fp->fi_lock);
1484 list_del(&stp->st_perstateowner);
1488 static void nfs4_free_ol_stateid(struct nfs4_stid *stid)
1490 struct nfs4_ol_stateid *stp = openlockstateid(stid);
1492 put_clnt_odstate(stp->st_clnt_odstate);
1493 release_all_access(stp);
1494 if (stp->st_stateowner)
1495 nfs4_put_stateowner(stp->st_stateowner);
1496 WARN_ON(!list_empty(&stid->sc_cp_list));
1497 kmem_cache_free(stateid_slab, stid);
1500 static void nfs4_free_lock_stateid(struct nfs4_stid *stid)
1502 struct nfs4_ol_stateid *stp = openlockstateid(stid);
1503 struct nfs4_lockowner *lo = lockowner(stp->st_stateowner);
1504 struct nfsd_file *nf;
1506 nf = find_any_file(stp->st_stid.sc_file);
1508 get_file(nf->nf_file);
1509 filp_close(nf->nf_file, (fl_owner_t)lo);
1512 nfs4_free_ol_stateid(stid);
1516 * Put the persistent reference to an already unhashed generic stateid, while
1517 * holding the cl_lock. If it's the last reference, then put it onto the
1518 * reaplist for later destruction.
1520 static void put_ol_stateid_locked(struct nfs4_ol_stateid *stp,
1521 struct list_head *reaplist)
1523 struct nfs4_stid *s = &stp->st_stid;
1524 struct nfs4_client *clp = s->sc_client;
1526 lockdep_assert_held(&clp->cl_lock);
1528 WARN_ON_ONCE(!list_empty(&stp->st_locks));
1530 if (!refcount_dec_and_test(&s->sc_count)) {
1531 wake_up_all(&close_wq);
1535 idr_remove(&clp->cl_stateids, s->sc_stateid.si_opaque.so_id);
1536 list_add(&stp->st_locks, reaplist);
1539 static bool unhash_lock_stateid(struct nfs4_ol_stateid *stp)
1541 lockdep_assert_held(&stp->st_stid.sc_client->cl_lock);
1543 if (!unhash_ol_stateid(stp))
1545 list_del_init(&stp->st_locks);
1546 nfs4_unhash_stid(&stp->st_stid);
1550 static void release_lock_stateid(struct nfs4_ol_stateid *stp)
1552 struct nfs4_client *clp = stp->st_stid.sc_client;
1555 spin_lock(&clp->cl_lock);
1556 unhashed = unhash_lock_stateid(stp);
1557 spin_unlock(&clp->cl_lock);
1559 nfs4_put_stid(&stp->st_stid);
1562 static void unhash_lockowner_locked(struct nfs4_lockowner *lo)
1564 struct nfs4_client *clp = lo->lo_owner.so_client;
1566 lockdep_assert_held(&clp->cl_lock);
1568 list_del_init(&lo->lo_owner.so_strhash);
1572 * Free a list of generic stateids that were collected earlier after being
1576 free_ol_stateid_reaplist(struct list_head *reaplist)
1578 struct nfs4_ol_stateid *stp;
1579 struct nfs4_file *fp;
1583 while (!list_empty(reaplist)) {
1584 stp = list_first_entry(reaplist, struct nfs4_ol_stateid,
1586 list_del(&stp->st_locks);
1587 fp = stp->st_stid.sc_file;
1588 stp->st_stid.sc_free(&stp->st_stid);
1594 static void release_open_stateid_locks(struct nfs4_ol_stateid *open_stp,
1595 struct list_head *reaplist)
1597 struct nfs4_ol_stateid *stp;
1599 lockdep_assert_held(&open_stp->st_stid.sc_client->cl_lock);
1601 while (!list_empty(&open_stp->st_locks)) {
1602 stp = list_entry(open_stp->st_locks.next,
1603 struct nfs4_ol_stateid, st_locks);
1604 WARN_ON(!unhash_lock_stateid(stp));
1605 put_ol_stateid_locked(stp, reaplist);
1609 static bool unhash_open_stateid(struct nfs4_ol_stateid *stp,
1610 struct list_head *reaplist)
1612 lockdep_assert_held(&stp->st_stid.sc_client->cl_lock);
1614 if (!unhash_ol_stateid(stp))
1616 release_open_stateid_locks(stp, reaplist);
1620 static void release_open_stateid(struct nfs4_ol_stateid *stp)
1622 LIST_HEAD(reaplist);
1624 spin_lock(&stp->st_stid.sc_client->cl_lock);
1625 if (unhash_open_stateid(stp, &reaplist))
1626 put_ol_stateid_locked(stp, &reaplist);
1627 spin_unlock(&stp->st_stid.sc_client->cl_lock);
1628 free_ol_stateid_reaplist(&reaplist);
1631 static void unhash_openowner_locked(struct nfs4_openowner *oo)
1633 struct nfs4_client *clp = oo->oo_owner.so_client;
1635 lockdep_assert_held(&clp->cl_lock);
1637 list_del_init(&oo->oo_owner.so_strhash);
1638 list_del_init(&oo->oo_perclient);
1641 static void release_last_closed_stateid(struct nfs4_openowner *oo)
1643 struct nfsd_net *nn = net_generic(oo->oo_owner.so_client->net,
1645 struct nfs4_ol_stateid *s;
1647 spin_lock(&nn->client_lock);
1648 s = oo->oo_last_closed_stid;
1650 list_del_init(&oo->oo_close_lru);
1651 oo->oo_last_closed_stid = NULL;
1653 spin_unlock(&nn->client_lock);
1655 nfs4_put_stid(&s->st_stid);
1658 static void release_openowner(struct nfs4_openowner *oo)
1660 struct nfs4_ol_stateid *stp;
1661 struct nfs4_client *clp = oo->oo_owner.so_client;
1662 struct list_head reaplist;
1664 INIT_LIST_HEAD(&reaplist);
1666 spin_lock(&clp->cl_lock);
1667 unhash_openowner_locked(oo);
1668 while (!list_empty(&oo->oo_owner.so_stateids)) {
1669 stp = list_first_entry(&oo->oo_owner.so_stateids,
1670 struct nfs4_ol_stateid, st_perstateowner);
1671 if (unhash_open_stateid(stp, &reaplist))
1672 put_ol_stateid_locked(stp, &reaplist);
1674 spin_unlock(&clp->cl_lock);
1675 free_ol_stateid_reaplist(&reaplist);
1676 release_last_closed_stateid(oo);
1677 nfs4_put_stateowner(&oo->oo_owner);
1681 hash_sessionid(struct nfs4_sessionid *sessionid)
1683 struct nfsd4_sessionid *sid = (struct nfsd4_sessionid *)sessionid;
1685 return sid->sequence % SESSION_HASH_SIZE;
1688 #ifdef CONFIG_SUNRPC_DEBUG
1690 dump_sessionid(const char *fn, struct nfs4_sessionid *sessionid)
1692 u32 *ptr = (u32 *)(&sessionid->data[0]);
1693 dprintk("%s: %u:%u:%u:%u\n", fn, ptr[0], ptr[1], ptr[2], ptr[3]);
1697 dump_sessionid(const char *fn, struct nfs4_sessionid *sessionid)
1703 * Bump the seqid on cstate->replay_owner, and clear replay_owner if it
1704 * won't be used for replay.
1706 void nfsd4_bump_seqid(struct nfsd4_compound_state *cstate, __be32 nfserr)
1708 struct nfs4_stateowner *so = cstate->replay_owner;
1710 if (nfserr == nfserr_replay_me)
1713 if (!seqid_mutating_err(ntohl(nfserr))) {
1714 nfsd4_cstate_clear_replay(cstate);
1719 if (so->so_is_open_owner)
1720 release_last_closed_stateid(openowner(so));
1726 gen_sessionid(struct nfsd4_session *ses)
1728 struct nfs4_client *clp = ses->se_client;
1729 struct nfsd4_sessionid *sid;
1731 sid = (struct nfsd4_sessionid *)ses->se_sessionid.data;
1732 sid->clientid = clp->cl_clientid;
1733 sid->sequence = current_sessionid++;
1738 * The protocol defines ca_maxresponssize_cached to include the size of
1739 * the rpc header, but all we need to cache is the data starting after
1740 * the end of the initial SEQUENCE operation--the rest we regenerate
1741 * each time. Therefore we can advertise a ca_maxresponssize_cached
1742 * value that is the number of bytes in our cache plus a few additional
1743 * bytes. In order to stay on the safe side, and not promise more than
1744 * we can cache, those additional bytes must be the minimum possible: 24
1745 * bytes of rpc header (xid through accept state, with AUTH_NULL
1746 * verifier), 12 for the compound header (with zero-length tag), and 44
1747 * for the SEQUENCE op response:
1749 #define NFSD_MIN_HDR_SEQ_SZ (24 + 12 + 44)
1752 free_session_slots(struct nfsd4_session *ses)
1756 for (i = 0; i < ses->se_fchannel.maxreqs; i++) {
1757 free_svc_cred(&ses->se_slots[i]->sl_cred);
1758 kfree(ses->se_slots[i]);
1763 * We don't actually need to cache the rpc and session headers, so we
1764 * can allocate a little less for each slot:
1766 static inline u32 slot_bytes(struct nfsd4_channel_attrs *ca)
1770 if (ca->maxresp_cached < NFSD_MIN_HDR_SEQ_SZ)
1773 size = ca->maxresp_cached - NFSD_MIN_HDR_SEQ_SZ;
1774 return size + sizeof(struct nfsd4_slot);
1778 * XXX: If we run out of reserved DRC memory we could (up to a point)
1779 * re-negotiate active sessions and reduce their slot usage to make
1780 * room for new connections. For now we just fail the create session.
1782 static u32 nfsd4_get_drc_mem(struct nfsd4_channel_attrs *ca, struct nfsd_net *nn)
1784 u32 slotsize = slot_bytes(ca);
1785 u32 num = ca->maxreqs;
1786 unsigned long avail, total_avail;
1787 unsigned int scale_factor;
1789 spin_lock(&nfsd_drc_lock);
1790 if (nfsd_drc_max_mem > nfsd_drc_mem_used)
1791 total_avail = nfsd_drc_max_mem - nfsd_drc_mem_used;
1793 /* We have handed out more space than we chose in
1794 * set_max_drc() to allow. That isn't really a
1795 * problem as long as that doesn't make us think we
1796 * have lots more due to integer overflow.
1799 avail = min((unsigned long)NFSD_MAX_MEM_PER_SESSION, total_avail);
1801 * Never use more than a fraction of the remaining memory,
1802 * unless it's the only way to give this client a slot.
1803 * The chosen fraction is either 1/8 or 1/number of threads,
1804 * whichever is smaller. This ensures there are adequate
1805 * slots to support multiple clients per thread.
1806 * Give the client one slot even if that would require
1807 * over-allocation--it is better than failure.
1809 scale_factor = max_t(unsigned int, 8, nn->nfsd_serv->sv_nrthreads);
1811 avail = clamp_t(unsigned long, avail, slotsize,
1812 total_avail/scale_factor);
1813 num = min_t(int, num, avail / slotsize);
1814 num = max_t(int, num, 1);
1815 nfsd_drc_mem_used += num * slotsize;
1816 spin_unlock(&nfsd_drc_lock);
1821 static void nfsd4_put_drc_mem(struct nfsd4_channel_attrs *ca)
1823 int slotsize = slot_bytes(ca);
1825 spin_lock(&nfsd_drc_lock);
1826 nfsd_drc_mem_used -= slotsize * ca->maxreqs;
1827 spin_unlock(&nfsd_drc_lock);
1830 static struct nfsd4_session *alloc_session(struct nfsd4_channel_attrs *fattrs,
1831 struct nfsd4_channel_attrs *battrs)
1833 int numslots = fattrs->maxreqs;
1834 int slotsize = slot_bytes(fattrs);
1835 struct nfsd4_session *new;
1838 BUILD_BUG_ON(struct_size(new, se_slots, NFSD_MAX_SLOTS_PER_SESSION)
1841 new = kzalloc(struct_size(new, se_slots, numslots), GFP_KERNEL);
1844 /* allocate each struct nfsd4_slot and data cache in one piece */
1845 for (i = 0; i < numslots; i++) {
1846 new->se_slots[i] = kzalloc(slotsize, GFP_KERNEL);
1847 if (!new->se_slots[i])
1851 memcpy(&new->se_fchannel, fattrs, sizeof(struct nfsd4_channel_attrs));
1852 memcpy(&new->se_bchannel, battrs, sizeof(struct nfsd4_channel_attrs));
1857 kfree(new->se_slots[i]);
1862 static void free_conn(struct nfsd4_conn *c)
1864 svc_xprt_put(c->cn_xprt);
1868 static void nfsd4_conn_lost(struct svc_xpt_user *u)
1870 struct nfsd4_conn *c = container_of(u, struct nfsd4_conn, cn_xpt_user);
1871 struct nfs4_client *clp = c->cn_session->se_client;
1873 trace_nfsd_cb_lost(clp);
1875 spin_lock(&clp->cl_lock);
1876 if (!list_empty(&c->cn_persession)) {
1877 list_del(&c->cn_persession);
1880 nfsd4_probe_callback(clp);
1881 spin_unlock(&clp->cl_lock);
1884 static struct nfsd4_conn *alloc_conn(struct svc_rqst *rqstp, u32 flags)
1886 struct nfsd4_conn *conn;
1888 conn = kmalloc(sizeof(struct nfsd4_conn), GFP_KERNEL);
1891 svc_xprt_get(rqstp->rq_xprt);
1892 conn->cn_xprt = rqstp->rq_xprt;
1893 conn->cn_flags = flags;
1894 INIT_LIST_HEAD(&conn->cn_xpt_user.list);
1898 static void __nfsd4_hash_conn(struct nfsd4_conn *conn, struct nfsd4_session *ses)
1900 conn->cn_session = ses;
1901 list_add(&conn->cn_persession, &ses->se_conns);
1904 static void nfsd4_hash_conn(struct nfsd4_conn *conn, struct nfsd4_session *ses)
1906 struct nfs4_client *clp = ses->se_client;
1908 spin_lock(&clp->cl_lock);
1909 __nfsd4_hash_conn(conn, ses);
1910 spin_unlock(&clp->cl_lock);
1913 static int nfsd4_register_conn(struct nfsd4_conn *conn)
1915 conn->cn_xpt_user.callback = nfsd4_conn_lost;
1916 return register_xpt_user(conn->cn_xprt, &conn->cn_xpt_user);
1919 static void nfsd4_init_conn(struct svc_rqst *rqstp, struct nfsd4_conn *conn, struct nfsd4_session *ses)
1923 nfsd4_hash_conn(conn, ses);
1924 ret = nfsd4_register_conn(conn);
1926 /* oops; xprt is already down: */
1927 nfsd4_conn_lost(&conn->cn_xpt_user);
1928 /* We may have gained or lost a callback channel: */
1929 nfsd4_probe_callback_sync(ses->se_client);
1932 static struct nfsd4_conn *alloc_conn_from_crses(struct svc_rqst *rqstp, struct nfsd4_create_session *cses)
1934 u32 dir = NFS4_CDFC4_FORE;
1936 if (cses->flags & SESSION4_BACK_CHAN)
1937 dir |= NFS4_CDFC4_BACK;
1938 return alloc_conn(rqstp, dir);
1941 /* must be called under client_lock */
1942 static void nfsd4_del_conns(struct nfsd4_session *s)
1944 struct nfs4_client *clp = s->se_client;
1945 struct nfsd4_conn *c;
1947 spin_lock(&clp->cl_lock);
1948 while (!list_empty(&s->se_conns)) {
1949 c = list_first_entry(&s->se_conns, struct nfsd4_conn, cn_persession);
1950 list_del_init(&c->cn_persession);
1951 spin_unlock(&clp->cl_lock);
1953 unregister_xpt_user(c->cn_xprt, &c->cn_xpt_user);
1956 spin_lock(&clp->cl_lock);
1958 spin_unlock(&clp->cl_lock);
1961 static void __free_session(struct nfsd4_session *ses)
1963 free_session_slots(ses);
1967 static void free_session(struct nfsd4_session *ses)
1969 nfsd4_del_conns(ses);
1970 nfsd4_put_drc_mem(&ses->se_fchannel);
1971 __free_session(ses);
1974 static void init_session(struct svc_rqst *rqstp, struct nfsd4_session *new, struct nfs4_client *clp, struct nfsd4_create_session *cses)
1977 struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
1979 new->se_client = clp;
1982 INIT_LIST_HEAD(&new->se_conns);
1984 new->se_cb_seq_nr = 1;
1985 new->se_flags = cses->flags;
1986 new->se_cb_prog = cses->callback_prog;
1987 new->se_cb_sec = cses->cb_sec;
1988 atomic_set(&new->se_ref, 0);
1989 idx = hash_sessionid(&new->se_sessionid);
1990 list_add(&new->se_hash, &nn->sessionid_hashtbl[idx]);
1991 spin_lock(&clp->cl_lock);
1992 list_add(&new->se_perclnt, &clp->cl_sessions);
1993 spin_unlock(&clp->cl_lock);
1996 struct sockaddr *sa = svc_addr(rqstp);
1998 * This is a little silly; with sessions there's no real
1999 * use for the callback address. Use the peer address
2000 * as a reasonable default for now, but consider fixing
2001 * the rpc client not to require an address in the
2004 rpc_copy_addr((struct sockaddr *)&clp->cl_cb_conn.cb_addr, sa);
2005 clp->cl_cb_conn.cb_addrlen = svc_addr_len(sa);
2009 /* caller must hold client_lock */
2010 static struct nfsd4_session *
2011 __find_in_sessionid_hashtbl(struct nfs4_sessionid *sessionid, struct net *net)
2013 struct nfsd4_session *elem;
2015 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
2017 lockdep_assert_held(&nn->client_lock);
2019 dump_sessionid(__func__, sessionid);
2020 idx = hash_sessionid(sessionid);
2021 /* Search in the appropriate list */
2022 list_for_each_entry(elem, &nn->sessionid_hashtbl[idx], se_hash) {
2023 if (!memcmp(elem->se_sessionid.data, sessionid->data,
2024 NFS4_MAX_SESSIONID_LEN)) {
2029 dprintk("%s: session not found\n", __func__);
2033 static struct nfsd4_session *
2034 find_in_sessionid_hashtbl(struct nfs4_sessionid *sessionid, struct net *net,
2037 struct nfsd4_session *session;
2038 __be32 status = nfserr_badsession;
2040 session = __find_in_sessionid_hashtbl(sessionid, net);
2043 status = nfsd4_get_session_locked(session);
2051 /* caller must hold client_lock */
2053 unhash_session(struct nfsd4_session *ses)
2055 struct nfs4_client *clp = ses->se_client;
2056 struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
2058 lockdep_assert_held(&nn->client_lock);
2060 list_del(&ses->se_hash);
2061 spin_lock(&ses->se_client->cl_lock);
2062 list_del(&ses->se_perclnt);
2063 spin_unlock(&ses->se_client->cl_lock);
2066 /* SETCLIENTID and SETCLIENTID_CONFIRM Helper functions */
2068 STALE_CLIENTID(clientid_t *clid, struct nfsd_net *nn)
2071 * We're assuming the clid was not given out from a boot
2072 * precisely 2^32 (about 136 years) before this one. That seems
2073 * a safe assumption:
2075 if (clid->cl_boot == (u32)nn->boot_time)
2077 trace_nfsd_clid_stale(clid);
2082 * XXX Should we use a slab cache ?
2083 * This type of memory management is somewhat inefficient, but we use it
2084 * anyway since SETCLIENTID is not a common operation.
2086 static struct nfs4_client *alloc_client(struct xdr_netobj name,
2087 struct nfsd_net *nn)
2089 struct nfs4_client *clp;
2092 if (atomic_read(&nn->nfs4_client_count) >= nn->nfs4_max_clients) {
2093 mod_delayed_work(laundry_wq, &nn->laundromat_work, 0);
2096 clp = kmem_cache_zalloc(client_slab, GFP_KERNEL);
2099 xdr_netobj_dup(&clp->cl_name, &name, GFP_KERNEL);
2100 if (clp->cl_name.data == NULL)
2102 clp->cl_ownerstr_hashtbl = kmalloc_array(OWNER_HASH_SIZE,
2103 sizeof(struct list_head),
2105 if (!clp->cl_ownerstr_hashtbl)
2106 goto err_no_hashtbl;
2107 for (i = 0; i < OWNER_HASH_SIZE; i++)
2108 INIT_LIST_HEAD(&clp->cl_ownerstr_hashtbl[i]);
2109 INIT_LIST_HEAD(&clp->cl_sessions);
2110 idr_init(&clp->cl_stateids);
2111 atomic_set(&clp->cl_rpc_users, 0);
2112 clp->cl_cb_state = NFSD4_CB_UNKNOWN;
2113 clp->cl_state = NFSD4_ACTIVE;
2114 atomic_inc(&nn->nfs4_client_count);
2115 atomic_set(&clp->cl_delegs_in_recall, 0);
2116 INIT_LIST_HEAD(&clp->cl_idhash);
2117 INIT_LIST_HEAD(&clp->cl_openowners);
2118 INIT_LIST_HEAD(&clp->cl_delegations);
2119 INIT_LIST_HEAD(&clp->cl_lru);
2120 INIT_LIST_HEAD(&clp->cl_revoked);
2121 #ifdef CONFIG_NFSD_PNFS
2122 INIT_LIST_HEAD(&clp->cl_lo_states);
2124 INIT_LIST_HEAD(&clp->async_copies);
2125 spin_lock_init(&clp->async_lock);
2126 spin_lock_init(&clp->cl_lock);
2127 rpc_init_wait_queue(&clp->cl_cb_waitq, "Backchannel slot table");
2130 kfree(clp->cl_name.data);
2132 kmem_cache_free(client_slab, clp);
2136 static void __free_client(struct kref *k)
2138 struct nfsdfs_client *c = container_of(k, struct nfsdfs_client, cl_ref);
2139 struct nfs4_client *clp = container_of(c, struct nfs4_client, cl_nfsdfs);
2141 free_svc_cred(&clp->cl_cred);
2142 kfree(clp->cl_ownerstr_hashtbl);
2143 kfree(clp->cl_name.data);
2144 kfree(clp->cl_nii_domain.data);
2145 kfree(clp->cl_nii_name.data);
2146 idr_destroy(&clp->cl_stateids);
2148 kmem_cache_free(client_slab, clp);
2151 static void drop_client(struct nfs4_client *clp)
2153 kref_put(&clp->cl_nfsdfs.cl_ref, __free_client);
2157 free_client(struct nfs4_client *clp)
2159 while (!list_empty(&clp->cl_sessions)) {
2160 struct nfsd4_session *ses;
2161 ses = list_entry(clp->cl_sessions.next, struct nfsd4_session,
2163 list_del(&ses->se_perclnt);
2164 WARN_ON_ONCE(atomic_read(&ses->se_ref));
2167 rpc_destroy_wait_queue(&clp->cl_cb_waitq);
2168 if (clp->cl_nfsd_dentry) {
2169 nfsd_client_rmdir(clp->cl_nfsd_dentry);
2170 clp->cl_nfsd_dentry = NULL;
2171 wake_up_all(&expiry_wq);
2176 /* must be called under the client_lock */
2178 unhash_client_locked(struct nfs4_client *clp)
2180 struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
2181 struct nfsd4_session *ses;
2183 lockdep_assert_held(&nn->client_lock);
2185 /* Mark the client as expired! */
2187 /* Make it invisible */
2188 if (!list_empty(&clp->cl_idhash)) {
2189 list_del_init(&clp->cl_idhash);
2190 if (test_bit(NFSD4_CLIENT_CONFIRMED, &clp->cl_flags))
2191 rb_erase(&clp->cl_namenode, &nn->conf_name_tree);
2193 rb_erase(&clp->cl_namenode, &nn->unconf_name_tree);
2195 list_del_init(&clp->cl_lru);
2196 spin_lock(&clp->cl_lock);
2197 list_for_each_entry(ses, &clp->cl_sessions, se_perclnt)
2198 list_del_init(&ses->se_hash);
2199 spin_unlock(&clp->cl_lock);
2203 unhash_client(struct nfs4_client *clp)
2205 struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
2207 spin_lock(&nn->client_lock);
2208 unhash_client_locked(clp);
2209 spin_unlock(&nn->client_lock);
2212 static __be32 mark_client_expired_locked(struct nfs4_client *clp)
2214 if (atomic_read(&clp->cl_rpc_users))
2215 return nfserr_jukebox;
2216 unhash_client_locked(clp);
2221 __destroy_client(struct nfs4_client *clp)
2223 struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
2225 struct nfs4_openowner *oo;
2226 struct nfs4_delegation *dp;
2227 struct list_head reaplist;
2229 INIT_LIST_HEAD(&reaplist);
2230 spin_lock(&state_lock);
2231 while (!list_empty(&clp->cl_delegations)) {
2232 dp = list_entry(clp->cl_delegations.next, struct nfs4_delegation, dl_perclnt);
2233 WARN_ON(!unhash_delegation_locked(dp));
2234 list_add(&dp->dl_recall_lru, &reaplist);
2236 spin_unlock(&state_lock);
2237 while (!list_empty(&reaplist)) {
2238 dp = list_entry(reaplist.next, struct nfs4_delegation, dl_recall_lru);
2239 list_del_init(&dp->dl_recall_lru);
2240 destroy_unhashed_deleg(dp);
2242 while (!list_empty(&clp->cl_revoked)) {
2243 dp = list_entry(clp->cl_revoked.next, struct nfs4_delegation, dl_recall_lru);
2244 list_del_init(&dp->dl_recall_lru);
2245 nfs4_put_stid(&dp->dl_stid);
2247 while (!list_empty(&clp->cl_openowners)) {
2248 oo = list_entry(clp->cl_openowners.next, struct nfs4_openowner, oo_perclient);
2249 nfs4_get_stateowner(&oo->oo_owner);
2250 release_openowner(oo);
2252 for (i = 0; i < OWNER_HASH_SIZE; i++) {
2253 struct nfs4_stateowner *so, *tmp;
2255 list_for_each_entry_safe(so, tmp, &clp->cl_ownerstr_hashtbl[i],
2257 /* Should be no openowners at this point */
2258 WARN_ON_ONCE(so->so_is_open_owner);
2259 remove_blocked_locks(lockowner(so));
2262 nfsd4_return_all_client_layouts(clp);
2263 nfsd4_shutdown_copy(clp);
2264 nfsd4_shutdown_callback(clp);
2265 if (clp->cl_cb_conn.cb_xprt)
2266 svc_xprt_put(clp->cl_cb_conn.cb_xprt);
2267 atomic_add_unless(&nn->nfs4_client_count, -1, 0);
2268 nfsd4_dec_courtesy_client_count(nn, clp);
2270 wake_up_all(&expiry_wq);
2274 destroy_client(struct nfs4_client *clp)
2277 __destroy_client(clp);
2280 static void inc_reclaim_complete(struct nfs4_client *clp)
2282 struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
2284 if (!nn->track_reclaim_completes)
2286 if (!nfsd4_find_reclaim_client(clp->cl_name, nn))
2288 if (atomic_inc_return(&nn->nr_reclaim_complete) ==
2289 nn->reclaim_str_hashtbl_size) {
2290 printk(KERN_INFO "NFSD: all clients done reclaiming, ending NFSv4 grace period (net %x)\n",
2292 nfsd4_end_grace(nn);
2296 static void expire_client(struct nfs4_client *clp)
2299 nfsd4_client_record_remove(clp);
2300 __destroy_client(clp);
2303 static void copy_verf(struct nfs4_client *target, nfs4_verifier *source)
2305 memcpy(target->cl_verifier.data, source->data,
2306 sizeof(target->cl_verifier.data));
2309 static void copy_clid(struct nfs4_client *target, struct nfs4_client *source)
2311 target->cl_clientid.cl_boot = source->cl_clientid.cl_boot;
2312 target->cl_clientid.cl_id = source->cl_clientid.cl_id;
2315 static int copy_cred(struct svc_cred *target, struct svc_cred *source)
2317 target->cr_principal = kstrdup(source->cr_principal, GFP_KERNEL);
2318 target->cr_raw_principal = kstrdup(source->cr_raw_principal,
2320 target->cr_targ_princ = kstrdup(source->cr_targ_princ, GFP_KERNEL);
2321 if ((source->cr_principal && !target->cr_principal) ||
2322 (source->cr_raw_principal && !target->cr_raw_principal) ||
2323 (source->cr_targ_princ && !target->cr_targ_princ))
2326 target->cr_flavor = source->cr_flavor;
2327 target->cr_uid = source->cr_uid;
2328 target->cr_gid = source->cr_gid;
2329 target->cr_group_info = source->cr_group_info;
2330 get_group_info(target->cr_group_info);
2331 target->cr_gss_mech = source->cr_gss_mech;
2332 if (source->cr_gss_mech)
2333 gss_mech_get(source->cr_gss_mech);
2338 compare_blob(const struct xdr_netobj *o1, const struct xdr_netobj *o2)
2340 if (o1->len < o2->len)
2342 if (o1->len > o2->len)
2344 return memcmp(o1->data, o2->data, o1->len);
2348 same_verf(nfs4_verifier *v1, nfs4_verifier *v2)
2350 return 0 == memcmp(v1->data, v2->data, sizeof(v1->data));
2354 same_clid(clientid_t *cl1, clientid_t *cl2)
2356 return (cl1->cl_boot == cl2->cl_boot) && (cl1->cl_id == cl2->cl_id);
2359 static bool groups_equal(struct group_info *g1, struct group_info *g2)
2363 if (g1->ngroups != g2->ngroups)
2365 for (i=0; i<g1->ngroups; i++)
2366 if (!gid_eq(g1->gid[i], g2->gid[i]))
2372 * RFC 3530 language requires clid_inuse be returned when the
2373 * "principal" associated with a requests differs from that previously
2374 * used. We use uid, gid's, and gss principal string as our best
2375 * approximation. We also don't want to allow non-gss use of a client
2376 * established using gss: in theory cr_principal should catch that
2377 * change, but in practice cr_principal can be null even in the gss case
2378 * since gssd doesn't always pass down a principal string.
2380 static bool is_gss_cred(struct svc_cred *cr)
2382 /* Is cr_flavor one of the gss "pseudoflavors"?: */
2383 return (cr->cr_flavor > RPC_AUTH_MAXFLAVOR);
2388 same_creds(struct svc_cred *cr1, struct svc_cred *cr2)
2390 if ((is_gss_cred(cr1) != is_gss_cred(cr2))
2391 || (!uid_eq(cr1->cr_uid, cr2->cr_uid))
2392 || (!gid_eq(cr1->cr_gid, cr2->cr_gid))
2393 || !groups_equal(cr1->cr_group_info, cr2->cr_group_info))
2395 /* XXX: check that cr_targ_princ fields match ? */
2396 if (cr1->cr_principal == cr2->cr_principal)
2398 if (!cr1->cr_principal || !cr2->cr_principal)
2400 return 0 == strcmp(cr1->cr_principal, cr2->cr_principal);
2403 static bool svc_rqst_integrity_protected(struct svc_rqst *rqstp)
2405 struct svc_cred *cr = &rqstp->rq_cred;
2408 if (!cr->cr_gss_mech)
2410 service = gss_pseudoflavor_to_service(cr->cr_gss_mech, cr->cr_flavor);
2411 return service == RPC_GSS_SVC_INTEGRITY ||
2412 service == RPC_GSS_SVC_PRIVACY;
2415 bool nfsd4_mach_creds_match(struct nfs4_client *cl, struct svc_rqst *rqstp)
2417 struct svc_cred *cr = &rqstp->rq_cred;
2419 if (!cl->cl_mach_cred)
2421 if (cl->cl_cred.cr_gss_mech != cr->cr_gss_mech)
2423 if (!svc_rqst_integrity_protected(rqstp))
2425 if (cl->cl_cred.cr_raw_principal)
2426 return 0 == strcmp(cl->cl_cred.cr_raw_principal,
2427 cr->cr_raw_principal);
2428 if (!cr->cr_principal)
2430 return 0 == strcmp(cl->cl_cred.cr_principal, cr->cr_principal);
2433 static void gen_confirm(struct nfs4_client *clp, struct nfsd_net *nn)
2438 * This is opaque to client, so no need to byte-swap. Use
2439 * __force to keep sparse happy
2441 verf[0] = (__force __be32)(u32)ktime_get_real_seconds();
2442 verf[1] = (__force __be32)nn->clverifier_counter++;
2443 memcpy(clp->cl_confirm.data, verf, sizeof(clp->cl_confirm.data));
2446 static void gen_clid(struct nfs4_client *clp, struct nfsd_net *nn)
2448 clp->cl_clientid.cl_boot = (u32)nn->boot_time;
2449 clp->cl_clientid.cl_id = nn->clientid_counter++;
2450 gen_confirm(clp, nn);
2453 static struct nfs4_stid *
2454 find_stateid_locked(struct nfs4_client *cl, stateid_t *t)
2456 struct nfs4_stid *ret;
2458 ret = idr_find(&cl->cl_stateids, t->si_opaque.so_id);
2459 if (!ret || !ret->sc_type)
2464 static struct nfs4_stid *
2465 find_stateid_by_type(struct nfs4_client *cl, stateid_t *t, char typemask)
2467 struct nfs4_stid *s;
2469 spin_lock(&cl->cl_lock);
2470 s = find_stateid_locked(cl, t);
2472 if (typemask & s->sc_type)
2473 refcount_inc(&s->sc_count);
2477 spin_unlock(&cl->cl_lock);
2481 static struct nfs4_client *get_nfsdfs_clp(struct inode *inode)
2483 struct nfsdfs_client *nc;
2484 nc = get_nfsdfs_client(inode);
2487 return container_of(nc, struct nfs4_client, cl_nfsdfs);
2490 static void seq_quote_mem(struct seq_file *m, char *data, int len)
2492 seq_printf(m, "\"");
2493 seq_escape_mem(m, data, len, ESCAPE_HEX | ESCAPE_NAP | ESCAPE_APPEND, "\"\\");
2494 seq_printf(m, "\"");
2497 static const char *cb_state2str(int state)
2502 case NFSD4_CB_UNKNOWN:
2506 case NFSD4_CB_FAULT:
2512 static int client_info_show(struct seq_file *m, void *v)
2514 struct inode *inode = file_inode(m->file);
2515 struct nfs4_client *clp;
2518 clp = get_nfsdfs_clp(inode);
2521 memcpy(&clid, &clp->cl_clientid, sizeof(clid));
2522 seq_printf(m, "clientid: 0x%llx\n", clid);
2523 seq_printf(m, "address: \"%pISpc\"\n", (struct sockaddr *)&clp->cl_addr);
2525 if (clp->cl_state == NFSD4_COURTESY)
2526 seq_puts(m, "status: courtesy\n");
2527 else if (clp->cl_state == NFSD4_EXPIRABLE)
2528 seq_puts(m, "status: expirable\n");
2529 else if (test_bit(NFSD4_CLIENT_CONFIRMED, &clp->cl_flags))
2530 seq_puts(m, "status: confirmed\n");
2532 seq_puts(m, "status: unconfirmed\n");
2533 seq_printf(m, "seconds from last renew: %lld\n",
2534 ktime_get_boottime_seconds() - clp->cl_time);
2535 seq_printf(m, "name: ");
2536 seq_quote_mem(m, clp->cl_name.data, clp->cl_name.len);
2537 seq_printf(m, "\nminor version: %d\n", clp->cl_minorversion);
2538 if (clp->cl_nii_domain.data) {
2539 seq_printf(m, "Implementation domain: ");
2540 seq_quote_mem(m, clp->cl_nii_domain.data,
2541 clp->cl_nii_domain.len);
2542 seq_printf(m, "\nImplementation name: ");
2543 seq_quote_mem(m, clp->cl_nii_name.data, clp->cl_nii_name.len);
2544 seq_printf(m, "\nImplementation time: [%lld, %ld]\n",
2545 clp->cl_nii_time.tv_sec, clp->cl_nii_time.tv_nsec);
2547 seq_printf(m, "callback state: %s\n", cb_state2str(clp->cl_cb_state));
2548 seq_printf(m, "callback address: %pISpc\n", &clp->cl_cb_conn.cb_addr);
2554 DEFINE_SHOW_ATTRIBUTE(client_info);
2556 static void *states_start(struct seq_file *s, loff_t *pos)
2557 __acquires(&clp->cl_lock)
2559 struct nfs4_client *clp = s->private;
2560 unsigned long id = *pos;
2563 spin_lock(&clp->cl_lock);
2564 ret = idr_get_next_ul(&clp->cl_stateids, &id);
2569 static void *states_next(struct seq_file *s, void *v, loff_t *pos)
2571 struct nfs4_client *clp = s->private;
2572 unsigned long id = *pos;
2577 ret = idr_get_next_ul(&clp->cl_stateids, &id);
2582 static void states_stop(struct seq_file *s, void *v)
2583 __releases(&clp->cl_lock)
2585 struct nfs4_client *clp = s->private;
2587 spin_unlock(&clp->cl_lock);
2590 static void nfs4_show_fname(struct seq_file *s, struct nfsd_file *f)
2592 seq_printf(s, "filename: \"%pD2\"", f->nf_file);
2595 static void nfs4_show_superblock(struct seq_file *s, struct nfsd_file *f)
2597 struct inode *inode = file_inode(f->nf_file);
2599 seq_printf(s, "superblock: \"%02x:%02x:%ld\"",
2600 MAJOR(inode->i_sb->s_dev),
2601 MINOR(inode->i_sb->s_dev),
2605 static void nfs4_show_owner(struct seq_file *s, struct nfs4_stateowner *oo)
2607 seq_printf(s, "owner: ");
2608 seq_quote_mem(s, oo->so_owner.data, oo->so_owner.len);
2611 static void nfs4_show_stateid(struct seq_file *s, stateid_t *stid)
2613 seq_printf(s, "0x%.8x", stid->si_generation);
2614 seq_printf(s, "%12phN", &stid->si_opaque);
2617 static int nfs4_show_open(struct seq_file *s, struct nfs4_stid *st)
2619 struct nfs4_ol_stateid *ols;
2620 struct nfs4_file *nf;
2621 struct nfsd_file *file;
2622 struct nfs4_stateowner *oo;
2623 unsigned int access, deny;
2625 if (st->sc_type != NFS4_OPEN_STID && st->sc_type != NFS4_LOCK_STID)
2626 return 0; /* XXX: or SEQ_SKIP? */
2627 ols = openlockstateid(st);
2628 oo = ols->st_stateowner;
2631 spin_lock(&nf->fi_lock);
2632 file = find_any_file_locked(nf);
2636 seq_printf(s, "- ");
2637 nfs4_show_stateid(s, &st->sc_stateid);
2638 seq_printf(s, ": { type: open, ");
2640 access = bmap_to_share_mode(ols->st_access_bmap);
2641 deny = bmap_to_share_mode(ols->st_deny_bmap);
2643 seq_printf(s, "access: %s%s, ",
2644 access & NFS4_SHARE_ACCESS_READ ? "r" : "-",
2645 access & NFS4_SHARE_ACCESS_WRITE ? "w" : "-");
2646 seq_printf(s, "deny: %s%s, ",
2647 deny & NFS4_SHARE_ACCESS_READ ? "r" : "-",
2648 deny & NFS4_SHARE_ACCESS_WRITE ? "w" : "-");
2650 nfs4_show_superblock(s, file);
2651 seq_printf(s, ", ");
2652 nfs4_show_fname(s, file);
2653 seq_printf(s, ", ");
2654 nfs4_show_owner(s, oo);
2655 seq_printf(s, " }\n");
2657 spin_unlock(&nf->fi_lock);
2661 static int nfs4_show_lock(struct seq_file *s, struct nfs4_stid *st)
2663 struct nfs4_ol_stateid *ols;
2664 struct nfs4_file *nf;
2665 struct nfsd_file *file;
2666 struct nfs4_stateowner *oo;
2668 ols = openlockstateid(st);
2669 oo = ols->st_stateowner;
2671 spin_lock(&nf->fi_lock);
2672 file = find_any_file_locked(nf);
2676 seq_printf(s, "- ");
2677 nfs4_show_stateid(s, &st->sc_stateid);
2678 seq_printf(s, ": { type: lock, ");
2681 * Note: a lock stateid isn't really the same thing as a lock,
2682 * it's the locking state held by one owner on a file, and there
2683 * may be multiple (or no) lock ranges associated with it.
2684 * (Same for the matter is true of open stateids.)
2687 nfs4_show_superblock(s, file);
2688 /* XXX: open stateid? */
2689 seq_printf(s, ", ");
2690 nfs4_show_fname(s, file);
2691 seq_printf(s, ", ");
2692 nfs4_show_owner(s, oo);
2693 seq_printf(s, " }\n");
2695 spin_unlock(&nf->fi_lock);
2699 static int nfs4_show_deleg(struct seq_file *s, struct nfs4_stid *st)
2701 struct nfs4_delegation *ds;
2702 struct nfs4_file *nf;
2703 struct nfsd_file *file;
2705 ds = delegstateid(st);
2707 spin_lock(&nf->fi_lock);
2708 file = find_deleg_file_locked(nf);
2712 seq_printf(s, "- ");
2713 nfs4_show_stateid(s, &st->sc_stateid);
2714 seq_printf(s, ": { type: deleg, ");
2716 /* Kinda dead code as long as we only support read delegs: */
2717 seq_printf(s, "access: %s, ",
2718 ds->dl_type == NFS4_OPEN_DELEGATE_READ ? "r" : "w");
2720 /* XXX: lease time, whether it's being recalled. */
2722 nfs4_show_superblock(s, file);
2723 seq_printf(s, ", ");
2724 nfs4_show_fname(s, file);
2725 seq_printf(s, " }\n");
2727 spin_unlock(&nf->fi_lock);
2731 static int nfs4_show_layout(struct seq_file *s, struct nfs4_stid *st)
2733 struct nfs4_layout_stateid *ls;
2734 struct nfsd_file *file;
2736 ls = container_of(st, struct nfs4_layout_stateid, ls_stid);
2739 seq_printf(s, "- ");
2740 nfs4_show_stateid(s, &st->sc_stateid);
2741 seq_printf(s, ": { type: layout, ");
2743 /* XXX: What else would be useful? */
2745 nfs4_show_superblock(s, file);
2746 seq_printf(s, ", ");
2747 nfs4_show_fname(s, file);
2748 seq_printf(s, " }\n");
2753 static int states_show(struct seq_file *s, void *v)
2755 struct nfs4_stid *st = v;
2757 switch (st->sc_type) {
2758 case NFS4_OPEN_STID:
2759 return nfs4_show_open(s, st);
2760 case NFS4_LOCK_STID:
2761 return nfs4_show_lock(s, st);
2762 case NFS4_DELEG_STID:
2763 return nfs4_show_deleg(s, st);
2764 case NFS4_LAYOUT_STID:
2765 return nfs4_show_layout(s, st);
2767 return 0; /* XXX: or SEQ_SKIP? */
2769 /* XXX: copy stateids? */
2772 static struct seq_operations states_seq_ops = {
2773 .start = states_start,
2774 .next = states_next,
2775 .stop = states_stop,
2779 static int client_states_open(struct inode *inode, struct file *file)
2782 struct nfs4_client *clp;
2785 clp = get_nfsdfs_clp(inode);
2789 ret = seq_open(file, &states_seq_ops);
2792 s = file->private_data;
2797 static int client_opens_release(struct inode *inode, struct file *file)
2799 struct seq_file *m = file->private_data;
2800 struct nfs4_client *clp = m->private;
2802 /* XXX: alternatively, we could get/drop in seq start/stop */
2807 static const struct file_operations client_states_fops = {
2808 .open = client_states_open,
2810 .llseek = seq_lseek,
2811 .release = client_opens_release,
2815 * Normally we refuse to destroy clients that are in use, but here the
2816 * administrator is telling us to just do it. We also want to wait
2817 * so the caller has a guarantee that the client's locks are gone by
2818 * the time the write returns:
2820 static void force_expire_client(struct nfs4_client *clp)
2822 struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
2823 bool already_expired;
2825 trace_nfsd_clid_admin_expired(&clp->cl_clientid);
2827 spin_lock(&nn->client_lock);
2829 spin_unlock(&nn->client_lock);
2831 wait_event(expiry_wq, atomic_read(&clp->cl_rpc_users) == 0);
2832 spin_lock(&nn->client_lock);
2833 already_expired = list_empty(&clp->cl_lru);
2834 if (!already_expired)
2835 unhash_client_locked(clp);
2836 spin_unlock(&nn->client_lock);
2838 if (!already_expired)
2841 wait_event(expiry_wq, clp->cl_nfsd_dentry == NULL);
2844 static ssize_t client_ctl_write(struct file *file, const char __user *buf,
2845 size_t size, loff_t *pos)
2848 struct nfs4_client *clp;
2850 data = simple_transaction_get(file, buf, size);
2852 return PTR_ERR(data);
2853 if (size != 7 || 0 != memcmp(data, "expire\n", 7))
2855 clp = get_nfsdfs_clp(file_inode(file));
2858 force_expire_client(clp);
2863 static const struct file_operations client_ctl_fops = {
2864 .write = client_ctl_write,
2865 .release = simple_transaction_release,
2868 static const struct tree_descr client_files[] = {
2869 [0] = {"info", &client_info_fops, S_IRUSR},
2870 [1] = {"states", &client_states_fops, S_IRUSR},
2871 [2] = {"ctl", &client_ctl_fops, S_IWUSR},
2876 nfsd4_cb_recall_any_done(struct nfsd4_callback *cb,
2877 struct rpc_task *task)
2879 trace_nfsd_cb_recall_any_done(cb, task);
2880 switch (task->tk_status) {
2881 case -NFS4ERR_DELAY:
2882 rpc_delay(task, 2 * HZ);
2890 nfsd4_cb_recall_any_release(struct nfsd4_callback *cb)
2892 struct nfs4_client *clp = cb->cb_clp;
2893 struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
2895 spin_lock(&nn->client_lock);
2896 clear_bit(NFSD4_CLIENT_CB_RECALL_ANY, &clp->cl_flags);
2897 put_client_renew_locked(clp);
2898 spin_unlock(&nn->client_lock);
2901 static const struct nfsd4_callback_ops nfsd4_cb_recall_any_ops = {
2902 .done = nfsd4_cb_recall_any_done,
2903 .release = nfsd4_cb_recall_any_release,
2906 static struct nfs4_client *create_client(struct xdr_netobj name,
2907 struct svc_rqst *rqstp, nfs4_verifier *verf)
2909 struct nfs4_client *clp;
2910 struct sockaddr *sa = svc_addr(rqstp);
2912 struct net *net = SVC_NET(rqstp);
2913 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
2914 struct dentry *dentries[ARRAY_SIZE(client_files)];
2916 clp = alloc_client(name, nn);
2920 ret = copy_cred(&clp->cl_cred, &rqstp->rq_cred);
2926 kref_init(&clp->cl_nfsdfs.cl_ref);
2927 nfsd4_init_cb(&clp->cl_cb_null, clp, NULL, NFSPROC4_CLNT_CB_NULL);
2928 clp->cl_time = ktime_get_boottime_seconds();
2929 clear_bit(0, &clp->cl_cb_slot_busy);
2930 copy_verf(clp, verf);
2931 memcpy(&clp->cl_addr, sa, sizeof(struct sockaddr_storage));
2932 clp->cl_cb_session = NULL;
2934 clp->cl_nfsd_dentry = nfsd_client_mkdir(
2935 nn, &clp->cl_nfsdfs,
2936 clp->cl_clientid.cl_id - nn->clientid_base,
2937 client_files, dentries);
2938 clp->cl_nfsd_info_dentry = dentries[0];
2939 if (!clp->cl_nfsd_dentry) {
2943 clp->cl_ra = kzalloc(sizeof(*clp->cl_ra), GFP_KERNEL);
2948 clp->cl_ra_time = 0;
2949 nfsd4_init_cb(&clp->cl_ra->ra_cb, clp, &nfsd4_cb_recall_any_ops,
2950 NFSPROC4_CLNT_CB_RECALL_ANY);
2955 add_clp_to_name_tree(struct nfs4_client *new_clp, struct rb_root *root)
2957 struct rb_node **new = &(root->rb_node), *parent = NULL;
2958 struct nfs4_client *clp;
2961 clp = rb_entry(*new, struct nfs4_client, cl_namenode);
2964 if (compare_blob(&clp->cl_name, &new_clp->cl_name) > 0)
2965 new = &((*new)->rb_left);
2967 new = &((*new)->rb_right);
2970 rb_link_node(&new_clp->cl_namenode, parent, new);
2971 rb_insert_color(&new_clp->cl_namenode, root);
2974 static struct nfs4_client *
2975 find_clp_in_name_tree(struct xdr_netobj *name, struct rb_root *root)
2978 struct rb_node *node = root->rb_node;
2979 struct nfs4_client *clp;
2982 clp = rb_entry(node, struct nfs4_client, cl_namenode);
2983 cmp = compare_blob(&clp->cl_name, name);
2985 node = node->rb_left;
2987 node = node->rb_right;
2995 add_to_unconfirmed(struct nfs4_client *clp)
2997 unsigned int idhashval;
2998 struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
3000 lockdep_assert_held(&nn->client_lock);
3002 clear_bit(NFSD4_CLIENT_CONFIRMED, &clp->cl_flags);
3003 add_clp_to_name_tree(clp, &nn->unconf_name_tree);
3004 idhashval = clientid_hashval(clp->cl_clientid.cl_id);
3005 list_add(&clp->cl_idhash, &nn->unconf_id_hashtbl[idhashval]);
3006 renew_client_locked(clp);
3010 move_to_confirmed(struct nfs4_client *clp)
3012 unsigned int idhashval = clientid_hashval(clp->cl_clientid.cl_id);
3013 struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
3015 lockdep_assert_held(&nn->client_lock);
3017 list_move(&clp->cl_idhash, &nn->conf_id_hashtbl[idhashval]);
3018 rb_erase(&clp->cl_namenode, &nn->unconf_name_tree);
3019 add_clp_to_name_tree(clp, &nn->conf_name_tree);
3020 set_bit(NFSD4_CLIENT_CONFIRMED, &clp->cl_flags);
3021 trace_nfsd_clid_confirmed(&clp->cl_clientid);
3022 renew_client_locked(clp);
3025 static struct nfs4_client *
3026 find_client_in_id_table(struct list_head *tbl, clientid_t *clid, bool sessions)
3028 struct nfs4_client *clp;
3029 unsigned int idhashval = clientid_hashval(clid->cl_id);
3031 list_for_each_entry(clp, &tbl[idhashval], cl_idhash) {
3032 if (same_clid(&clp->cl_clientid, clid)) {
3033 if ((bool)clp->cl_minorversion != sessions)
3035 renew_client_locked(clp);
3042 static struct nfs4_client *
3043 find_confirmed_client(clientid_t *clid, bool sessions, struct nfsd_net *nn)
3045 struct list_head *tbl = nn->conf_id_hashtbl;
3047 lockdep_assert_held(&nn->client_lock);
3048 return find_client_in_id_table(tbl, clid, sessions);
3051 static struct nfs4_client *
3052 find_unconfirmed_client(clientid_t *clid, bool sessions, struct nfsd_net *nn)
3054 struct list_head *tbl = nn->unconf_id_hashtbl;
3056 lockdep_assert_held(&nn->client_lock);
3057 return find_client_in_id_table(tbl, clid, sessions);
3060 static bool clp_used_exchangeid(struct nfs4_client *clp)
3062 return clp->cl_exchange_flags != 0;
3065 static struct nfs4_client *
3066 find_confirmed_client_by_name(struct xdr_netobj *name, struct nfsd_net *nn)
3068 lockdep_assert_held(&nn->client_lock);
3069 return find_clp_in_name_tree(name, &nn->conf_name_tree);
3072 static struct nfs4_client *
3073 find_unconfirmed_client_by_name(struct xdr_netobj *name, struct nfsd_net *nn)
3075 lockdep_assert_held(&nn->client_lock);
3076 return find_clp_in_name_tree(name, &nn->unconf_name_tree);
3080 gen_callback(struct nfs4_client *clp, struct nfsd4_setclientid *se, struct svc_rqst *rqstp)
3082 struct nfs4_cb_conn *conn = &clp->cl_cb_conn;
3083 struct sockaddr *sa = svc_addr(rqstp);
3084 u32 scopeid = rpc_get_scope_id(sa);
3085 unsigned short expected_family;
3087 /* Currently, we only support tcp and tcp6 for the callback channel */
3088 if (se->se_callback_netid_len == 3 &&
3089 !memcmp(se->se_callback_netid_val, "tcp", 3))
3090 expected_family = AF_INET;
3091 else if (se->se_callback_netid_len == 4 &&
3092 !memcmp(se->se_callback_netid_val, "tcp6", 4))
3093 expected_family = AF_INET6;
3097 conn->cb_addrlen = rpc_uaddr2sockaddr(clp->net, se->se_callback_addr_val,
3098 se->se_callback_addr_len,
3099 (struct sockaddr *)&conn->cb_addr,
3100 sizeof(conn->cb_addr));
3102 if (!conn->cb_addrlen || conn->cb_addr.ss_family != expected_family)
3105 if (conn->cb_addr.ss_family == AF_INET6)
3106 ((struct sockaddr_in6 *)&conn->cb_addr)->sin6_scope_id = scopeid;
3108 conn->cb_prog = se->se_callback_prog;
3109 conn->cb_ident = se->se_callback_ident;
3110 memcpy(&conn->cb_saddr, &rqstp->rq_daddr, rqstp->rq_daddrlen);
3111 trace_nfsd_cb_args(clp, conn);
3114 conn->cb_addr.ss_family = AF_UNSPEC;
3115 conn->cb_addrlen = 0;
3116 trace_nfsd_cb_nodelegs(clp);
3121 * Cache a reply. nfsd4_check_resp_size() has bounded the cache size.
3124 nfsd4_store_cache_entry(struct nfsd4_compoundres *resp)
3126 struct xdr_buf *buf = resp->xdr->buf;
3127 struct nfsd4_slot *slot = resp->cstate.slot;
3130 dprintk("--> %s slot %p\n", __func__, slot);
3132 slot->sl_flags |= NFSD4_SLOT_INITIALIZED;
3133 slot->sl_opcnt = resp->opcnt;
3134 slot->sl_status = resp->cstate.status;
3135 free_svc_cred(&slot->sl_cred);
3136 copy_cred(&slot->sl_cred, &resp->rqstp->rq_cred);
3138 if (!nfsd4_cache_this(resp)) {
3139 slot->sl_flags &= ~NFSD4_SLOT_CACHED;
3142 slot->sl_flags |= NFSD4_SLOT_CACHED;
3144 base = resp->cstate.data_offset;
3145 slot->sl_datalen = buf->len - base;
3146 if (read_bytes_from_xdr_buf(buf, base, slot->sl_data, slot->sl_datalen))
3147 WARN(1, "%s: sessions DRC could not cache compound\n",
3153 * Encode the replay sequence operation from the slot values.
3154 * If cachethis is FALSE encode the uncached rep error on the next
3155 * operation which sets resp->p and increments resp->opcnt for
3156 * nfs4svc_encode_compoundres.
3160 nfsd4_enc_sequence_replay(struct nfsd4_compoundargs *args,
3161 struct nfsd4_compoundres *resp)
3163 struct nfsd4_op *op;
3164 struct nfsd4_slot *slot = resp->cstate.slot;
3166 /* Encode the replayed sequence operation */
3167 op = &args->ops[resp->opcnt - 1];
3168 nfsd4_encode_operation(resp, op);
3170 if (slot->sl_flags & NFSD4_SLOT_CACHED)
3172 if (args->opcnt == 1) {
3174 * The original operation wasn't a solo sequence--we
3175 * always cache those--so this retry must not match the
3178 op->status = nfserr_seq_false_retry;
3180 op = &args->ops[resp->opcnt++];
3181 op->status = nfserr_retry_uncached_rep;
3182 nfsd4_encode_operation(resp, op);
3188 * The sequence operation is not cached because we can use the slot and
3192 nfsd4_replay_cache_entry(struct nfsd4_compoundres *resp,
3193 struct nfsd4_sequence *seq)
3195 struct nfsd4_slot *slot = resp->cstate.slot;
3196 struct xdr_stream *xdr = resp->xdr;
3200 dprintk("--> %s slot %p\n", __func__, slot);
3202 status = nfsd4_enc_sequence_replay(resp->rqstp->rq_argp, resp);
3206 p = xdr_reserve_space(xdr, slot->sl_datalen);
3209 return nfserr_serverfault;
3211 xdr_encode_opaque_fixed(p, slot->sl_data, slot->sl_datalen);
3212 xdr_commit_encode(xdr);
3214 resp->opcnt = slot->sl_opcnt;
3215 return slot->sl_status;
3219 * Set the exchange_id flags returned by the server.
3222 nfsd4_set_ex_flags(struct nfs4_client *new, struct nfsd4_exchange_id *clid)
3224 #ifdef CONFIG_NFSD_PNFS
3225 new->cl_exchange_flags |= EXCHGID4_FLAG_USE_PNFS_MDS;
3227 new->cl_exchange_flags |= EXCHGID4_FLAG_USE_NON_PNFS;
3230 /* Referrals are supported, Migration is not. */
3231 new->cl_exchange_flags |= EXCHGID4_FLAG_SUPP_MOVED_REFER;
3233 /* set the wire flags to return to client. */
3234 clid->flags = new->cl_exchange_flags;
3237 static bool client_has_openowners(struct nfs4_client *clp)
3239 struct nfs4_openowner *oo;
3241 list_for_each_entry(oo, &clp->cl_openowners, oo_perclient) {
3242 if (!list_empty(&oo->oo_owner.so_stateids))
3248 static bool client_has_state(struct nfs4_client *clp)
3250 return client_has_openowners(clp)
3251 #ifdef CONFIG_NFSD_PNFS
3252 || !list_empty(&clp->cl_lo_states)
3254 || !list_empty(&clp->cl_delegations)
3255 || !list_empty(&clp->cl_sessions)
3256 || !list_empty(&clp->async_copies);
3259 static __be32 copy_impl_id(struct nfs4_client *clp,
3260 struct nfsd4_exchange_id *exid)
3262 if (!exid->nii_domain.data)
3264 xdr_netobj_dup(&clp->cl_nii_domain, &exid->nii_domain, GFP_KERNEL);
3265 if (!clp->cl_nii_domain.data)
3266 return nfserr_jukebox;
3267 xdr_netobj_dup(&clp->cl_nii_name, &exid->nii_name, GFP_KERNEL);
3268 if (!clp->cl_nii_name.data)
3269 return nfserr_jukebox;
3270 clp->cl_nii_time = exid->nii_time;
3275 nfsd4_exchange_id(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3276 union nfsd4_op_u *u)
3278 struct nfsd4_exchange_id *exid = &u->exchange_id;
3279 struct nfs4_client *conf, *new;
3280 struct nfs4_client *unconf = NULL;
3282 char addr_str[INET6_ADDRSTRLEN];
3283 nfs4_verifier verf = exid->verifier;
3284 struct sockaddr *sa = svc_addr(rqstp);
3285 bool update = exid->flags & EXCHGID4_FLAG_UPD_CONFIRMED_REC_A;
3286 struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
3288 rpc_ntop(sa, addr_str, sizeof(addr_str));
3289 dprintk("%s rqstp=%p exid=%p clname.len=%u clname.data=%p "
3290 "ip_addr=%s flags %x, spa_how %u\n",
3291 __func__, rqstp, exid, exid->clname.len, exid->clname.data,
3292 addr_str, exid->flags, exid->spa_how);
3294 if (exid->flags & ~EXCHGID4_FLAG_MASK_A)
3295 return nfserr_inval;
3297 new = create_client(exid->clname, rqstp, &verf);
3299 return nfserr_jukebox;
3300 status = copy_impl_id(new, exid);
3304 switch (exid->spa_how) {
3306 exid->spo_must_enforce[0] = 0;
3307 exid->spo_must_enforce[1] = (
3308 1 << (OP_BIND_CONN_TO_SESSION - 32) |
3309 1 << (OP_EXCHANGE_ID - 32) |
3310 1 << (OP_CREATE_SESSION - 32) |
3311 1 << (OP_DESTROY_SESSION - 32) |
3312 1 << (OP_DESTROY_CLIENTID - 32));
3314 exid->spo_must_allow[0] &= (1 << (OP_CLOSE) |
3315 1 << (OP_OPEN_DOWNGRADE) |
3317 1 << (OP_DELEGRETURN));
3319 exid->spo_must_allow[1] &= (
3320 1 << (OP_TEST_STATEID - 32) |
3321 1 << (OP_FREE_STATEID - 32));
3322 if (!svc_rqst_integrity_protected(rqstp)) {
3323 status = nfserr_inval;
3327 * Sometimes userspace doesn't give us a principal.
3328 * Which is a bug, really. Anyway, we can't enforce
3329 * MACH_CRED in that case, better to give up now:
3331 if (!new->cl_cred.cr_principal &&
3332 !new->cl_cred.cr_raw_principal) {
3333 status = nfserr_serverfault;
3336 new->cl_mach_cred = true;
3340 default: /* checked by xdr code */
3344 status = nfserr_encr_alg_unsupp;
3348 /* Cases below refer to rfc 5661 section 18.35.4: */
3349 spin_lock(&nn->client_lock);
3350 conf = find_confirmed_client_by_name(&exid->clname, nn);
3352 bool creds_match = same_creds(&conf->cl_cred, &rqstp->rq_cred);
3353 bool verfs_match = same_verf(&verf, &conf->cl_verifier);
3356 if (!clp_used_exchangeid(conf)) { /* buggy client */
3357 status = nfserr_inval;
3360 if (!nfsd4_mach_creds_match(conf, rqstp)) {
3361 status = nfserr_wrong_cred;
3364 if (!creds_match) { /* case 9 */
3365 status = nfserr_perm;
3368 if (!verfs_match) { /* case 8 */
3369 status = nfserr_not_same;
3373 exid->flags |= EXCHGID4_FLAG_CONFIRMED_R;
3374 trace_nfsd_clid_confirmed_r(conf);
3377 if (!creds_match) { /* case 3 */
3378 if (client_has_state(conf)) {
3379 status = nfserr_clid_inuse;
3380 trace_nfsd_clid_cred_mismatch(conf, rqstp);
3385 if (verfs_match) { /* case 2 */
3386 conf->cl_exchange_flags |= EXCHGID4_FLAG_CONFIRMED_R;
3387 trace_nfsd_clid_confirmed_r(conf);
3390 /* case 5, client reboot */
3391 trace_nfsd_clid_verf_mismatch(conf, rqstp, &verf);
3396 if (update) { /* case 7 */
3397 status = nfserr_noent;
3401 unconf = find_unconfirmed_client_by_name(&exid->clname, nn);
3402 if (unconf) /* case 4, possible retry or client restart */
3403 unhash_client_locked(unconf);
3405 /* case 1, new owner ID */
3406 trace_nfsd_clid_fresh(new);
3410 status = mark_client_expired_locked(conf);
3413 trace_nfsd_clid_replaced(&conf->cl_clientid);
3415 new->cl_minorversion = cstate->minorversion;
3416 new->cl_spo_must_allow.u.words[0] = exid->spo_must_allow[0];
3417 new->cl_spo_must_allow.u.words[1] = exid->spo_must_allow[1];
3419 add_to_unconfirmed(new);
3422 exid->clientid.cl_boot = conf->cl_clientid.cl_boot;
3423 exid->clientid.cl_id = conf->cl_clientid.cl_id;
3425 exid->seqid = conf->cl_cs_slot.sl_seqid + 1;
3426 nfsd4_set_ex_flags(conf, exid);
3428 dprintk("nfsd4_exchange_id seqid %d flags %x\n",
3429 conf->cl_cs_slot.sl_seqid, conf->cl_exchange_flags);
3433 spin_unlock(&nn->client_lock);
3438 trace_nfsd_clid_expire_unconf(&unconf->cl_clientid);
3439 expire_client(unconf);
3445 check_slot_seqid(u32 seqid, u32 slot_seqid, int slot_inuse)
3447 dprintk("%s enter. seqid %d slot_seqid %d\n", __func__, seqid,
3450 /* The slot is in use, and no response has been sent. */
3452 if (seqid == slot_seqid)
3453 return nfserr_jukebox;
3455 return nfserr_seq_misordered;
3457 /* Note unsigned 32-bit arithmetic handles wraparound: */
3458 if (likely(seqid == slot_seqid + 1))
3460 if (seqid == slot_seqid)
3461 return nfserr_replay_cache;
3462 return nfserr_seq_misordered;
3466 * Cache the create session result into the create session single DRC
3467 * slot cache by saving the xdr structure. sl_seqid has been set.
3468 * Do this for solo or embedded create session operations.
3471 nfsd4_cache_create_session(struct nfsd4_create_session *cr_ses,
3472 struct nfsd4_clid_slot *slot, __be32 nfserr)
3474 slot->sl_status = nfserr;
3475 memcpy(&slot->sl_cr_ses, cr_ses, sizeof(*cr_ses));
3479 nfsd4_replay_create_session(struct nfsd4_create_session *cr_ses,
3480 struct nfsd4_clid_slot *slot)
3482 memcpy(cr_ses, &slot->sl_cr_ses, sizeof(*cr_ses));
3483 return slot->sl_status;
3486 #define NFSD_MIN_REQ_HDR_SEQ_SZ ((\
3487 2 * 2 + /* credential,verifier: AUTH_NULL, length 0 */ \
3488 1 + /* MIN tag is length with zero, only length */ \
3489 3 + /* version, opcount, opcode */ \
3490 XDR_QUADLEN(NFS4_MAX_SESSIONID_LEN) + \
3491 /* seqid, slotID, slotID, cache */ \
3492 4 ) * sizeof(__be32))
3494 #define NFSD_MIN_RESP_HDR_SEQ_SZ ((\
3495 2 + /* verifier: AUTH_NULL, length 0 */\
3497 1 + /* MIN tag is length with zero, only length */ \
3498 3 + /* opcount, opcode, opstatus*/ \
3499 XDR_QUADLEN(NFS4_MAX_SESSIONID_LEN) + \
3500 /* seqid, slotID, slotID, slotID, status */ \
3501 5 ) * sizeof(__be32))
3503 static __be32 check_forechannel_attrs(struct nfsd4_channel_attrs *ca, struct nfsd_net *nn)
3505 u32 maxrpc = nn->nfsd_serv->sv_max_mesg;
3507 if (ca->maxreq_sz < NFSD_MIN_REQ_HDR_SEQ_SZ)
3508 return nfserr_toosmall;
3509 if (ca->maxresp_sz < NFSD_MIN_RESP_HDR_SEQ_SZ)
3510 return nfserr_toosmall;
3511 ca->headerpadsz = 0;
3512 ca->maxreq_sz = min_t(u32, ca->maxreq_sz, maxrpc);
3513 ca->maxresp_sz = min_t(u32, ca->maxresp_sz, maxrpc);
3514 ca->maxops = min_t(u32, ca->maxops, NFSD_MAX_OPS_PER_COMPOUND);
3515 ca->maxresp_cached = min_t(u32, ca->maxresp_cached,
3516 NFSD_SLOT_CACHE_SIZE + NFSD_MIN_HDR_SEQ_SZ);
3517 ca->maxreqs = min_t(u32, ca->maxreqs, NFSD_MAX_SLOTS_PER_SESSION);
3519 * Note decreasing slot size below client's request may make it
3520 * difficult for client to function correctly, whereas
3521 * decreasing the number of slots will (just?) affect
3522 * performance. When short on memory we therefore prefer to
3523 * decrease number of slots instead of their size. Clients that
3524 * request larger slots than they need will get poor results:
3525 * Note that we always allow at least one slot, because our
3526 * accounting is soft and provides no guarantees either way.
3528 ca->maxreqs = nfsd4_get_drc_mem(ca, nn);
3534 * Server's NFSv4.1 backchannel support is AUTH_SYS-only for now.
3535 * These are based on similar macros in linux/sunrpc/msg_prot.h .
3537 #define RPC_MAX_HEADER_WITH_AUTH_SYS \
3538 (RPC_CALLHDRSIZE + 2 * (2 + UNX_CALLSLACK))
3540 #define RPC_MAX_REPHEADER_WITH_AUTH_SYS \
3541 (RPC_REPHDRSIZE + (2 + NUL_REPLYSLACK))
3543 #define NFSD_CB_MAX_REQ_SZ ((NFS4_enc_cb_recall_sz + \
3544 RPC_MAX_HEADER_WITH_AUTH_SYS) * sizeof(__be32))
3545 #define NFSD_CB_MAX_RESP_SZ ((NFS4_dec_cb_recall_sz + \
3546 RPC_MAX_REPHEADER_WITH_AUTH_SYS) * \
3549 static __be32 check_backchannel_attrs(struct nfsd4_channel_attrs *ca)
3551 ca->headerpadsz = 0;
3553 if (ca->maxreq_sz < NFSD_CB_MAX_REQ_SZ)
3554 return nfserr_toosmall;
3555 if (ca->maxresp_sz < NFSD_CB_MAX_RESP_SZ)
3556 return nfserr_toosmall;
3557 ca->maxresp_cached = 0;
3559 return nfserr_toosmall;
3564 static __be32 nfsd4_check_cb_sec(struct nfsd4_cb_sec *cbs)
3566 switch (cbs->flavor) {
3572 * GSS case: the spec doesn't allow us to return this
3573 * error. But it also doesn't allow us not to support
3575 * I'd rather this fail hard than return some error the
3576 * client might think it can already handle:
3578 return nfserr_encr_alg_unsupp;
3583 nfsd4_create_session(struct svc_rqst *rqstp,
3584 struct nfsd4_compound_state *cstate, union nfsd4_op_u *u)
3586 struct nfsd4_create_session *cr_ses = &u->create_session;
3587 struct sockaddr *sa = svc_addr(rqstp);
3588 struct nfs4_client *conf, *unconf;
3589 struct nfs4_client *old = NULL;
3590 struct nfsd4_session *new;
3591 struct nfsd4_conn *conn;
3592 struct nfsd4_clid_slot *cs_slot = NULL;
3594 struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
3596 if (cr_ses->flags & ~SESSION4_FLAG_MASK_A)
3597 return nfserr_inval;
3598 status = nfsd4_check_cb_sec(&cr_ses->cb_sec);
3601 status = check_forechannel_attrs(&cr_ses->fore_channel, nn);
3604 status = check_backchannel_attrs(&cr_ses->back_channel);
3606 goto out_release_drc_mem;
3607 status = nfserr_jukebox;
3608 new = alloc_session(&cr_ses->fore_channel, &cr_ses->back_channel);
3610 goto out_release_drc_mem;
3611 conn = alloc_conn_from_crses(rqstp, cr_ses);
3613 goto out_free_session;
3615 spin_lock(&nn->client_lock);
3616 unconf = find_unconfirmed_client(&cr_ses->clientid, true, nn);
3617 conf = find_confirmed_client(&cr_ses->clientid, true, nn);
3618 WARN_ON_ONCE(conf && unconf);
3621 status = nfserr_wrong_cred;
3622 if (!nfsd4_mach_creds_match(conf, rqstp))
3624 cs_slot = &conf->cl_cs_slot;
3625 status = check_slot_seqid(cr_ses->seqid, cs_slot->sl_seqid, 0);
3627 if (status == nfserr_replay_cache)
3628 status = nfsd4_replay_create_session(cr_ses, cs_slot);
3631 } else if (unconf) {
3632 status = nfserr_clid_inuse;
3633 if (!same_creds(&unconf->cl_cred, &rqstp->rq_cred) ||
3634 !rpc_cmp_addr(sa, (struct sockaddr *) &unconf->cl_addr)) {
3635 trace_nfsd_clid_cred_mismatch(unconf, rqstp);
3638 status = nfserr_wrong_cred;
3639 if (!nfsd4_mach_creds_match(unconf, rqstp))
3641 cs_slot = &unconf->cl_cs_slot;
3642 status = check_slot_seqid(cr_ses->seqid, cs_slot->sl_seqid, 0);
3644 /* an unconfirmed replay returns misordered */
3645 status = nfserr_seq_misordered;
3648 old = find_confirmed_client_by_name(&unconf->cl_name, nn);
3650 status = mark_client_expired_locked(old);
3655 trace_nfsd_clid_replaced(&old->cl_clientid);
3657 move_to_confirmed(unconf);
3660 status = nfserr_stale_clientid;
3664 /* Persistent sessions are not supported */
3665 cr_ses->flags &= ~SESSION4_PERSIST;
3666 /* Upshifting from TCP to RDMA is not supported */
3667 cr_ses->flags &= ~SESSION4_RDMA;
3669 init_session(rqstp, new, conf, cr_ses);
3670 nfsd4_get_session_locked(new);
3672 memcpy(cr_ses->sessionid.data, new->se_sessionid.data,
3673 NFS4_MAX_SESSIONID_LEN);
3674 cs_slot->sl_seqid++;
3675 cr_ses->seqid = cs_slot->sl_seqid;
3677 /* cache solo and embedded create sessions under the client_lock */
3678 nfsd4_cache_create_session(cr_ses, cs_slot, status);
3679 spin_unlock(&nn->client_lock);
3681 fsnotify_dentry(conf->cl_nfsd_info_dentry, FS_MODIFY);
3682 /* init connection and backchannel */
3683 nfsd4_init_conn(rqstp, conn, new);
3684 nfsd4_put_session(new);
3689 spin_unlock(&nn->client_lock);
3694 __free_session(new);
3695 out_release_drc_mem:
3696 nfsd4_put_drc_mem(&cr_ses->fore_channel);
3700 static __be32 nfsd4_map_bcts_dir(u32 *dir)
3703 case NFS4_CDFC4_FORE:
3704 case NFS4_CDFC4_BACK:
3706 case NFS4_CDFC4_FORE_OR_BOTH:
3707 case NFS4_CDFC4_BACK_OR_BOTH:
3708 *dir = NFS4_CDFC4_BOTH;
3711 return nfserr_inval;
3714 __be32 nfsd4_backchannel_ctl(struct svc_rqst *rqstp,
3715 struct nfsd4_compound_state *cstate,
3716 union nfsd4_op_u *u)
3718 struct nfsd4_backchannel_ctl *bc = &u->backchannel_ctl;
3719 struct nfsd4_session *session = cstate->session;
3720 struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
3723 status = nfsd4_check_cb_sec(&bc->bc_cb_sec);
3726 spin_lock(&nn->client_lock);
3727 session->se_cb_prog = bc->bc_cb_program;
3728 session->se_cb_sec = bc->bc_cb_sec;
3729 spin_unlock(&nn->client_lock);
3731 nfsd4_probe_callback(session->se_client);
3736 static struct nfsd4_conn *__nfsd4_find_conn(struct svc_xprt *xpt, struct nfsd4_session *s)
3738 struct nfsd4_conn *c;
3740 list_for_each_entry(c, &s->se_conns, cn_persession) {
3741 if (c->cn_xprt == xpt) {
3748 static __be32 nfsd4_match_existing_connection(struct svc_rqst *rqst,
3749 struct nfsd4_session *session, u32 req, struct nfsd4_conn **conn)
3751 struct nfs4_client *clp = session->se_client;
3752 struct svc_xprt *xpt = rqst->rq_xprt;
3753 struct nfsd4_conn *c;
3756 /* Following the last paragraph of RFC 5661 Section 18.34.3: */
3757 spin_lock(&clp->cl_lock);
3758 c = __nfsd4_find_conn(xpt, session);
3760 status = nfserr_noent;
3761 else if (req == c->cn_flags)
3763 else if (req == NFS4_CDFC4_FORE_OR_BOTH &&
3764 c->cn_flags != NFS4_CDFC4_BACK)
3766 else if (req == NFS4_CDFC4_BACK_OR_BOTH &&
3767 c->cn_flags != NFS4_CDFC4_FORE)
3770 status = nfserr_inval;
3771 spin_unlock(&clp->cl_lock);
3772 if (status == nfs_ok && conn)
3777 __be32 nfsd4_bind_conn_to_session(struct svc_rqst *rqstp,
3778 struct nfsd4_compound_state *cstate,
3779 union nfsd4_op_u *u)
3781 struct nfsd4_bind_conn_to_session *bcts = &u->bind_conn_to_session;
3783 struct nfsd4_conn *conn;
3784 struct nfsd4_session *session;
3785 struct net *net = SVC_NET(rqstp);
3786 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
3788 if (!nfsd4_last_compound_op(rqstp))
3789 return nfserr_not_only_op;
3790 spin_lock(&nn->client_lock);
3791 session = find_in_sessionid_hashtbl(&bcts->sessionid, net, &status);
3792 spin_unlock(&nn->client_lock);
3794 goto out_no_session;
3795 status = nfserr_wrong_cred;
3796 if (!nfsd4_mach_creds_match(session->se_client, rqstp))
3798 status = nfsd4_match_existing_connection(rqstp, session,
3800 if (status == nfs_ok) {
3801 if (bcts->dir == NFS4_CDFC4_FORE_OR_BOTH ||
3802 bcts->dir == NFS4_CDFC4_BACK)
3803 conn->cn_flags |= NFS4_CDFC4_BACK;
3804 nfsd4_probe_callback(session->se_client);
3807 if (status == nfserr_inval)
3809 status = nfsd4_map_bcts_dir(&bcts->dir);
3812 conn = alloc_conn(rqstp, bcts->dir);
3813 status = nfserr_jukebox;
3816 nfsd4_init_conn(rqstp, conn, session);
3819 nfsd4_put_session(session);
3824 static bool nfsd4_compound_in_session(struct nfsd4_compound_state *cstate, struct nfs4_sessionid *sid)
3826 if (!cstate->session)
3828 return !memcmp(sid, &cstate->session->se_sessionid, sizeof(*sid));
3832 nfsd4_destroy_session(struct svc_rqst *r, struct nfsd4_compound_state *cstate,
3833 union nfsd4_op_u *u)
3835 struct nfs4_sessionid *sessionid = &u->destroy_session.sessionid;
3836 struct nfsd4_session *ses;
3838 int ref_held_by_me = 0;
3839 struct net *net = SVC_NET(r);
3840 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
3842 status = nfserr_not_only_op;
3843 if (nfsd4_compound_in_session(cstate, sessionid)) {
3844 if (!nfsd4_last_compound_op(r))
3848 dump_sessionid(__func__, sessionid);
3849 spin_lock(&nn->client_lock);
3850 ses = find_in_sessionid_hashtbl(sessionid, net, &status);
3852 goto out_client_lock;
3853 status = nfserr_wrong_cred;
3854 if (!nfsd4_mach_creds_match(ses->se_client, r))
3855 goto out_put_session;
3856 status = mark_session_dead_locked(ses, 1 + ref_held_by_me);
3858 goto out_put_session;
3859 unhash_session(ses);
3860 spin_unlock(&nn->client_lock);
3862 nfsd4_probe_callback_sync(ses->se_client);
3864 spin_lock(&nn->client_lock);
3867 nfsd4_put_session_locked(ses);
3869 spin_unlock(&nn->client_lock);
3874 static __be32 nfsd4_sequence_check_conn(struct nfsd4_conn *new, struct nfsd4_session *ses)
3876 struct nfs4_client *clp = ses->se_client;
3877 struct nfsd4_conn *c;
3878 __be32 status = nfs_ok;
3881 spin_lock(&clp->cl_lock);
3882 c = __nfsd4_find_conn(new->cn_xprt, ses);
3885 status = nfserr_conn_not_bound_to_session;
3886 if (clp->cl_mach_cred)
3888 __nfsd4_hash_conn(new, ses);
3889 spin_unlock(&clp->cl_lock);
3890 ret = nfsd4_register_conn(new);
3892 /* oops; xprt is already down: */
3893 nfsd4_conn_lost(&new->cn_xpt_user);
3896 spin_unlock(&clp->cl_lock);
3901 static bool nfsd4_session_too_many_ops(struct svc_rqst *rqstp, struct nfsd4_session *session)
3903 struct nfsd4_compoundargs *args = rqstp->rq_argp;
3905 return args->opcnt > session->se_fchannel.maxops;
3908 static bool nfsd4_request_too_big(struct svc_rqst *rqstp,
3909 struct nfsd4_session *session)
3911 struct xdr_buf *xb = &rqstp->rq_arg;
3913 return xb->len > session->se_fchannel.maxreq_sz;
3916 static bool replay_matches_cache(struct svc_rqst *rqstp,
3917 struct nfsd4_sequence *seq, struct nfsd4_slot *slot)
3919 struct nfsd4_compoundargs *argp = rqstp->rq_argp;
3921 if ((bool)(slot->sl_flags & NFSD4_SLOT_CACHETHIS) !=
3922 (bool)seq->cachethis)
3925 * If there's an error then the reply can have fewer ops than
3928 if (slot->sl_opcnt < argp->opcnt && !slot->sl_status)
3931 * But if we cached a reply with *more* ops than the call you're
3932 * sending us now, then this new call is clearly not really a
3933 * replay of the old one:
3935 if (slot->sl_opcnt > argp->opcnt)
3937 /* This is the only check explicitly called by spec: */
3938 if (!same_creds(&rqstp->rq_cred, &slot->sl_cred))
3941 * There may be more comparisons we could actually do, but the
3942 * spec doesn't require us to catch every case where the calls
3943 * don't match (that would require caching the call as well as
3944 * the reply), so we don't bother.
3950 nfsd4_sequence(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3951 union nfsd4_op_u *u)
3953 struct nfsd4_sequence *seq = &u->sequence;
3954 struct nfsd4_compoundres *resp = rqstp->rq_resp;
3955 struct xdr_stream *xdr = resp->xdr;
3956 struct nfsd4_session *session;
3957 struct nfs4_client *clp;
3958 struct nfsd4_slot *slot;
3959 struct nfsd4_conn *conn;
3962 struct net *net = SVC_NET(rqstp);
3963 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
3965 if (resp->opcnt != 1)
3966 return nfserr_sequence_pos;
3969 * Will be either used or freed by nfsd4_sequence_check_conn
3972 conn = alloc_conn(rqstp, NFS4_CDFC4_FORE);
3974 return nfserr_jukebox;
3976 spin_lock(&nn->client_lock);
3977 session = find_in_sessionid_hashtbl(&seq->sessionid, net, &status);
3979 goto out_no_session;
3980 clp = session->se_client;
3982 status = nfserr_too_many_ops;
3983 if (nfsd4_session_too_many_ops(rqstp, session))
3984 goto out_put_session;
3986 status = nfserr_req_too_big;
3987 if (nfsd4_request_too_big(rqstp, session))
3988 goto out_put_session;
3990 status = nfserr_badslot;
3991 if (seq->slotid >= session->se_fchannel.maxreqs)
3992 goto out_put_session;
3994 slot = session->se_slots[seq->slotid];
3995 dprintk("%s: slotid %d\n", __func__, seq->slotid);
3997 /* We do not negotiate the number of slots yet, so set the
3998 * maxslots to the session maxreqs which is used to encode
3999 * sr_highest_slotid and the sr_target_slot id to maxslots */
4000 seq->maxslots = session->se_fchannel.maxreqs;
4002 status = check_slot_seqid(seq->seqid, slot->sl_seqid,
4003 slot->sl_flags & NFSD4_SLOT_INUSE);
4004 if (status == nfserr_replay_cache) {
4005 status = nfserr_seq_misordered;
4006 if (!(slot->sl_flags & NFSD4_SLOT_INITIALIZED))
4007 goto out_put_session;
4008 status = nfserr_seq_false_retry;
4009 if (!replay_matches_cache(rqstp, seq, slot))
4010 goto out_put_session;
4011 cstate->slot = slot;
4012 cstate->session = session;
4014 /* Return the cached reply status and set cstate->status
4015 * for nfsd4_proc_compound processing */
4016 status = nfsd4_replay_cache_entry(resp, seq);
4017 cstate->status = nfserr_replay_cache;
4021 goto out_put_session;
4023 status = nfsd4_sequence_check_conn(conn, session);
4026 goto out_put_session;
4028 buflen = (seq->cachethis) ?
4029 session->se_fchannel.maxresp_cached :
4030 session->se_fchannel.maxresp_sz;
4031 status = (seq->cachethis) ? nfserr_rep_too_big_to_cache :
4033 if (xdr_restrict_buflen(xdr, buflen - rqstp->rq_auth_slack))
4034 goto out_put_session;
4035 svc_reserve(rqstp, buflen);
4038 /* Success! bump slot seqid */
4039 slot->sl_seqid = seq->seqid;
4040 slot->sl_flags |= NFSD4_SLOT_INUSE;
4042 slot->sl_flags |= NFSD4_SLOT_CACHETHIS;
4044 slot->sl_flags &= ~NFSD4_SLOT_CACHETHIS;
4046 cstate->slot = slot;
4047 cstate->session = session;
4051 switch (clp->cl_cb_state) {
4053 seq->status_flags = SEQ4_STATUS_CB_PATH_DOWN;
4055 case NFSD4_CB_FAULT:
4056 seq->status_flags = SEQ4_STATUS_BACKCHANNEL_FAULT;
4059 seq->status_flags = 0;
4061 if (!list_empty(&clp->cl_revoked))
4062 seq->status_flags |= SEQ4_STATUS_RECALLABLE_STATE_REVOKED;
4066 spin_unlock(&nn->client_lock);
4069 nfsd4_put_session_locked(session);
4070 goto out_no_session;
4074 nfsd4_sequence_done(struct nfsd4_compoundres *resp)
4076 struct nfsd4_compound_state *cs = &resp->cstate;
4078 if (nfsd4_has_session(cs)) {
4079 if (cs->status != nfserr_replay_cache) {
4080 nfsd4_store_cache_entry(resp);
4081 cs->slot->sl_flags &= ~NFSD4_SLOT_INUSE;
4083 /* Drop session reference that was taken in nfsd4_sequence() */
4084 nfsd4_put_session(cs->session);
4086 put_client_renew(cs->clp);
4090 nfsd4_destroy_clientid(struct svc_rqst *rqstp,
4091 struct nfsd4_compound_state *cstate,
4092 union nfsd4_op_u *u)
4094 struct nfsd4_destroy_clientid *dc = &u->destroy_clientid;
4095 struct nfs4_client *conf, *unconf;
4096 struct nfs4_client *clp = NULL;
4098 struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
4100 spin_lock(&nn->client_lock);
4101 unconf = find_unconfirmed_client(&dc->clientid, true, nn);
4102 conf = find_confirmed_client(&dc->clientid, true, nn);
4103 WARN_ON_ONCE(conf && unconf);
4106 if (client_has_state(conf)) {
4107 status = nfserr_clientid_busy;
4110 status = mark_client_expired_locked(conf);
4117 status = nfserr_stale_clientid;
4120 if (!nfsd4_mach_creds_match(clp, rqstp)) {
4122 status = nfserr_wrong_cred;
4125 trace_nfsd_clid_destroyed(&clp->cl_clientid);
4126 unhash_client_locked(clp);
4128 spin_unlock(&nn->client_lock);
4135 nfsd4_reclaim_complete(struct svc_rqst *rqstp,
4136 struct nfsd4_compound_state *cstate, union nfsd4_op_u *u)
4138 struct nfsd4_reclaim_complete *rc = &u->reclaim_complete;
4139 struct nfs4_client *clp = cstate->clp;
4142 if (rc->rca_one_fs) {
4143 if (!cstate->current_fh.fh_dentry)
4144 return nfserr_nofilehandle;
4146 * We don't take advantage of the rca_one_fs case.
4147 * That's OK, it's optional, we can safely ignore it.
4152 status = nfserr_complete_already;
4153 if (test_and_set_bit(NFSD4_CLIENT_RECLAIM_COMPLETE, &clp->cl_flags))
4156 status = nfserr_stale_clientid;
4157 if (is_client_expired(clp))
4159 * The following error isn't really legal.
4160 * But we only get here if the client just explicitly
4161 * destroyed the client. Surely it no longer cares what
4162 * error it gets back on an operation for the dead
4168 trace_nfsd_clid_reclaim_complete(&clp->cl_clientid);
4169 nfsd4_client_record_create(clp);
4170 inc_reclaim_complete(clp);
4176 nfsd4_setclientid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
4177 union nfsd4_op_u *u)
4179 struct nfsd4_setclientid *setclid = &u->setclientid;
4180 struct xdr_netobj clname = setclid->se_name;
4181 nfs4_verifier clverifier = setclid->se_verf;
4182 struct nfs4_client *conf, *new;
4183 struct nfs4_client *unconf = NULL;
4185 struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
4187 new = create_client(clname, rqstp, &clverifier);
4189 return nfserr_jukebox;
4190 spin_lock(&nn->client_lock);
4191 conf = find_confirmed_client_by_name(&clname, nn);
4192 if (conf && client_has_state(conf)) {
4193 status = nfserr_clid_inuse;
4194 if (clp_used_exchangeid(conf))
4196 if (!same_creds(&conf->cl_cred, &rqstp->rq_cred)) {
4197 trace_nfsd_clid_cred_mismatch(conf, rqstp);
4201 unconf = find_unconfirmed_client_by_name(&clname, nn);
4203 unhash_client_locked(unconf);
4205 if (same_verf(&conf->cl_verifier, &clverifier)) {
4206 copy_clid(new, conf);
4207 gen_confirm(new, nn);
4209 trace_nfsd_clid_verf_mismatch(conf, rqstp,
4212 trace_nfsd_clid_fresh(new);
4213 new->cl_minorversion = 0;
4214 gen_callback(new, setclid, rqstp);
4215 add_to_unconfirmed(new);
4216 setclid->se_clientid.cl_boot = new->cl_clientid.cl_boot;
4217 setclid->se_clientid.cl_id = new->cl_clientid.cl_id;
4218 memcpy(setclid->se_confirm.data, new->cl_confirm.data, sizeof(setclid->se_confirm.data));
4222 spin_unlock(&nn->client_lock);
4226 trace_nfsd_clid_expire_unconf(&unconf->cl_clientid);
4227 expire_client(unconf);
4233 nfsd4_setclientid_confirm(struct svc_rqst *rqstp,
4234 struct nfsd4_compound_state *cstate,
4235 union nfsd4_op_u *u)
4237 struct nfsd4_setclientid_confirm *setclientid_confirm =
4238 &u->setclientid_confirm;
4239 struct nfs4_client *conf, *unconf;
4240 struct nfs4_client *old = NULL;
4241 nfs4_verifier confirm = setclientid_confirm->sc_confirm;
4242 clientid_t * clid = &setclientid_confirm->sc_clientid;
4244 struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
4246 if (STALE_CLIENTID(clid, nn))
4247 return nfserr_stale_clientid;
4249 spin_lock(&nn->client_lock);
4250 conf = find_confirmed_client(clid, false, nn);
4251 unconf = find_unconfirmed_client(clid, false, nn);
4253 * We try hard to give out unique clientid's, so if we get an
4254 * attempt to confirm the same clientid with a different cred,
4255 * the client may be buggy; this should never happen.
4257 * Nevertheless, RFC 7530 recommends INUSE for this case:
4259 status = nfserr_clid_inuse;
4260 if (unconf && !same_creds(&unconf->cl_cred, &rqstp->rq_cred)) {
4261 trace_nfsd_clid_cred_mismatch(unconf, rqstp);
4264 if (conf && !same_creds(&conf->cl_cred, &rqstp->rq_cred)) {
4265 trace_nfsd_clid_cred_mismatch(conf, rqstp);
4268 if (!unconf || !same_verf(&confirm, &unconf->cl_confirm)) {
4269 if (conf && same_verf(&confirm, &conf->cl_confirm)) {
4272 status = nfserr_stale_clientid;
4278 unhash_client_locked(old);
4279 nfsd4_change_callback(conf, &unconf->cl_cb_conn);
4281 old = find_confirmed_client_by_name(&unconf->cl_name, nn);
4283 status = nfserr_clid_inuse;
4284 if (client_has_state(old)
4285 && !same_creds(&unconf->cl_cred,
4290 status = mark_client_expired_locked(old);
4295 trace_nfsd_clid_replaced(&old->cl_clientid);
4297 move_to_confirmed(unconf);
4300 get_client_locked(conf);
4301 spin_unlock(&nn->client_lock);
4303 fsnotify_dentry(conf->cl_nfsd_info_dentry, FS_MODIFY);
4304 nfsd4_probe_callback(conf);
4305 spin_lock(&nn->client_lock);
4306 put_client_renew_locked(conf);
4308 spin_unlock(&nn->client_lock);
4314 static struct nfs4_file *nfsd4_alloc_file(void)
4316 return kmem_cache_alloc(file_slab, GFP_KERNEL);
4319 /* OPEN Share state helper functions */
4321 static void nfsd4_file_init(const struct svc_fh *fh, struct nfs4_file *fp)
4323 refcount_set(&fp->fi_ref, 1);
4324 spin_lock_init(&fp->fi_lock);
4325 INIT_LIST_HEAD(&fp->fi_stateids);
4326 INIT_LIST_HEAD(&fp->fi_delegations);
4327 INIT_LIST_HEAD(&fp->fi_clnt_odstate);
4328 fh_copy_shallow(&fp->fi_fhandle, &fh->fh_handle);
4329 fp->fi_deleg_file = NULL;
4330 fp->fi_had_conflict = false;
4331 fp->fi_share_deny = 0;
4332 memset(fp->fi_fds, 0, sizeof(fp->fi_fds));
4333 memset(fp->fi_access, 0, sizeof(fp->fi_access));
4334 fp->fi_aliased = false;
4335 fp->fi_inode = d_inode(fh->fh_dentry);
4336 #ifdef CONFIG_NFSD_PNFS
4337 INIT_LIST_HEAD(&fp->fi_lo_states);
4338 atomic_set(&fp->fi_lo_recalls, 0);
4343 nfsd4_free_slabs(void)
4345 kmem_cache_destroy(client_slab);
4346 kmem_cache_destroy(openowner_slab);
4347 kmem_cache_destroy(lockowner_slab);
4348 kmem_cache_destroy(file_slab);
4349 kmem_cache_destroy(stateid_slab);
4350 kmem_cache_destroy(deleg_slab);
4351 kmem_cache_destroy(odstate_slab);
4355 nfsd4_init_slabs(void)
4357 client_slab = kmem_cache_create("nfsd4_clients",
4358 sizeof(struct nfs4_client), 0, 0, NULL);
4359 if (client_slab == NULL)
4361 openowner_slab = kmem_cache_create("nfsd4_openowners",
4362 sizeof(struct nfs4_openowner), 0, 0, NULL);
4363 if (openowner_slab == NULL)
4364 goto out_free_client_slab;
4365 lockowner_slab = kmem_cache_create("nfsd4_lockowners",
4366 sizeof(struct nfs4_lockowner), 0, 0, NULL);
4367 if (lockowner_slab == NULL)
4368 goto out_free_openowner_slab;
4369 file_slab = kmem_cache_create("nfsd4_files",
4370 sizeof(struct nfs4_file), 0, 0, NULL);
4371 if (file_slab == NULL)
4372 goto out_free_lockowner_slab;
4373 stateid_slab = kmem_cache_create("nfsd4_stateids",
4374 sizeof(struct nfs4_ol_stateid), 0, 0, NULL);
4375 if (stateid_slab == NULL)
4376 goto out_free_file_slab;
4377 deleg_slab = kmem_cache_create("nfsd4_delegations",
4378 sizeof(struct nfs4_delegation), 0, 0, NULL);
4379 if (deleg_slab == NULL)
4380 goto out_free_stateid_slab;
4381 odstate_slab = kmem_cache_create("nfsd4_odstate",
4382 sizeof(struct nfs4_clnt_odstate), 0, 0, NULL);
4383 if (odstate_slab == NULL)
4384 goto out_free_deleg_slab;
4387 out_free_deleg_slab:
4388 kmem_cache_destroy(deleg_slab);
4389 out_free_stateid_slab:
4390 kmem_cache_destroy(stateid_slab);
4392 kmem_cache_destroy(file_slab);
4393 out_free_lockowner_slab:
4394 kmem_cache_destroy(lockowner_slab);
4395 out_free_openowner_slab:
4396 kmem_cache_destroy(openowner_slab);
4397 out_free_client_slab:
4398 kmem_cache_destroy(client_slab);
4403 static unsigned long
4404 nfsd4_state_shrinker_count(struct shrinker *shrink, struct shrink_control *sc)
4407 struct nfsd_net *nn = container_of(shrink,
4408 struct nfsd_net, nfsd_client_shrinker);
4410 count = atomic_read(&nn->nfsd_courtesy_clients);
4412 count = atomic_long_read(&num_delegations);
4414 queue_work(laundry_wq, &nn->nfsd_shrinker_work);
4415 return (unsigned long)count;
4418 static unsigned long
4419 nfsd4_state_shrinker_scan(struct shrinker *shrink, struct shrink_control *sc)
4425 nfsd4_init_leases_net(struct nfsd_net *nn)
4430 nn->nfsd4_lease = 90; /* default lease time */
4431 nn->nfsd4_grace = 90;
4432 nn->somebody_reclaimed = false;
4433 nn->track_reclaim_completes = false;
4434 nn->clverifier_counter = get_random_u32();
4435 nn->clientid_base = get_random_u32();
4436 nn->clientid_counter = nn->clientid_base + 1;
4437 nn->s2s_cp_cl_id = nn->clientid_counter++;
4439 atomic_set(&nn->nfs4_client_count, 0);
4441 max_clients = (u64)si.totalram * si.mem_unit / (1024 * 1024 * 1024);
4442 max_clients *= NFS4_CLIENTS_PER_GB;
4443 nn->nfs4_max_clients = max_t(int, max_clients, NFS4_CLIENTS_PER_GB);
4445 atomic_set(&nn->nfsd_courtesy_clients, 0);
4448 static void init_nfs4_replay(struct nfs4_replay *rp)
4450 rp->rp_status = nfserr_serverfault;
4452 rp->rp_buf = rp->rp_ibuf;
4453 mutex_init(&rp->rp_mutex);
4456 static void nfsd4_cstate_assign_replay(struct nfsd4_compound_state *cstate,
4457 struct nfs4_stateowner *so)
4459 if (!nfsd4_has_session(cstate)) {
4460 mutex_lock(&so->so_replay.rp_mutex);
4461 cstate->replay_owner = nfs4_get_stateowner(so);
4465 void nfsd4_cstate_clear_replay(struct nfsd4_compound_state *cstate)
4467 struct nfs4_stateowner *so = cstate->replay_owner;
4470 cstate->replay_owner = NULL;
4471 mutex_unlock(&so->so_replay.rp_mutex);
4472 nfs4_put_stateowner(so);
4476 static inline void *alloc_stateowner(struct kmem_cache *slab, struct xdr_netobj *owner, struct nfs4_client *clp)
4478 struct nfs4_stateowner *sop;
4480 sop = kmem_cache_alloc(slab, GFP_KERNEL);
4484 xdr_netobj_dup(&sop->so_owner, owner, GFP_KERNEL);
4485 if (!sop->so_owner.data) {
4486 kmem_cache_free(slab, sop);
4490 INIT_LIST_HEAD(&sop->so_stateids);
4491 sop->so_client = clp;
4492 init_nfs4_replay(&sop->so_replay);
4493 atomic_set(&sop->so_count, 1);
4497 static void hash_openowner(struct nfs4_openowner *oo, struct nfs4_client *clp, unsigned int strhashval)
4499 lockdep_assert_held(&clp->cl_lock);
4501 list_add(&oo->oo_owner.so_strhash,
4502 &clp->cl_ownerstr_hashtbl[strhashval]);
4503 list_add(&oo->oo_perclient, &clp->cl_openowners);
4506 static void nfs4_unhash_openowner(struct nfs4_stateowner *so)
4508 unhash_openowner_locked(openowner(so));
4511 static void nfs4_free_openowner(struct nfs4_stateowner *so)
4513 struct nfs4_openowner *oo = openowner(so);
4515 kmem_cache_free(openowner_slab, oo);
4518 static const struct nfs4_stateowner_operations openowner_ops = {
4519 .so_unhash = nfs4_unhash_openowner,
4520 .so_free = nfs4_free_openowner,
4523 static struct nfs4_ol_stateid *
4524 nfsd4_find_existing_open(struct nfs4_file *fp, struct nfsd4_open *open)
4526 struct nfs4_ol_stateid *local, *ret = NULL;
4527 struct nfs4_openowner *oo = open->op_openowner;
4529 lockdep_assert_held(&fp->fi_lock);
4531 list_for_each_entry(local, &fp->fi_stateids, st_perfile) {
4532 /* ignore lock owners */
4533 if (local->st_stateowner->so_is_open_owner == 0)
4535 if (local->st_stateowner != &oo->oo_owner)
4537 if (local->st_stid.sc_type == NFS4_OPEN_STID) {
4539 refcount_inc(&ret->st_stid.sc_count);
4547 nfsd4_verify_open_stid(struct nfs4_stid *s)
4549 __be32 ret = nfs_ok;
4551 switch (s->sc_type) {
4555 case NFS4_CLOSED_STID:
4556 case NFS4_CLOSED_DELEG_STID:
4557 ret = nfserr_bad_stateid;
4559 case NFS4_REVOKED_DELEG_STID:
4560 ret = nfserr_deleg_revoked;
4565 /* Lock the stateid st_mutex, and deal with races with CLOSE */
4567 nfsd4_lock_ol_stateid(struct nfs4_ol_stateid *stp)
4571 mutex_lock_nested(&stp->st_mutex, LOCK_STATEID_MUTEX);
4572 ret = nfsd4_verify_open_stid(&stp->st_stid);
4574 mutex_unlock(&stp->st_mutex);
4578 static struct nfs4_ol_stateid *
4579 nfsd4_find_and_lock_existing_open(struct nfs4_file *fp, struct nfsd4_open *open)
4581 struct nfs4_ol_stateid *stp;
4583 spin_lock(&fp->fi_lock);
4584 stp = nfsd4_find_existing_open(fp, open);
4585 spin_unlock(&fp->fi_lock);
4586 if (!stp || nfsd4_lock_ol_stateid(stp) == nfs_ok)
4588 nfs4_put_stid(&stp->st_stid);
4593 static struct nfs4_openowner *
4594 alloc_init_open_stateowner(unsigned int strhashval, struct nfsd4_open *open,
4595 struct nfsd4_compound_state *cstate)
4597 struct nfs4_client *clp = cstate->clp;
4598 struct nfs4_openowner *oo, *ret;
4600 oo = alloc_stateowner(openowner_slab, &open->op_owner, clp);
4603 oo->oo_owner.so_ops = &openowner_ops;
4604 oo->oo_owner.so_is_open_owner = 1;
4605 oo->oo_owner.so_seqid = open->op_seqid;
4607 if (nfsd4_has_session(cstate))
4608 oo->oo_flags |= NFS4_OO_CONFIRMED;
4610 oo->oo_last_closed_stid = NULL;
4611 INIT_LIST_HEAD(&oo->oo_close_lru);
4612 spin_lock(&clp->cl_lock);
4613 ret = find_openstateowner_str_locked(strhashval, open, clp);
4615 hash_openowner(oo, clp, strhashval);
4618 nfs4_free_stateowner(&oo->oo_owner);
4620 spin_unlock(&clp->cl_lock);
4624 static struct nfs4_ol_stateid *
4625 init_open_stateid(struct nfs4_file *fp, struct nfsd4_open *open)
4628 struct nfs4_openowner *oo = open->op_openowner;
4629 struct nfs4_ol_stateid *retstp = NULL;
4630 struct nfs4_ol_stateid *stp;
4633 /* We are moving these outside of the spinlocks to avoid the warnings */
4634 mutex_init(&stp->st_mutex);
4635 mutex_lock_nested(&stp->st_mutex, OPEN_STATEID_MUTEX);
4638 spin_lock(&oo->oo_owner.so_client->cl_lock);
4639 spin_lock(&fp->fi_lock);
4641 retstp = nfsd4_find_existing_open(fp, open);
4645 open->op_stp = NULL;
4646 refcount_inc(&stp->st_stid.sc_count);
4647 stp->st_stid.sc_type = NFS4_OPEN_STID;
4648 INIT_LIST_HEAD(&stp->st_locks);
4649 stp->st_stateowner = nfs4_get_stateowner(&oo->oo_owner);
4651 stp->st_stid.sc_file = fp;
4652 stp->st_access_bmap = 0;
4653 stp->st_deny_bmap = 0;
4654 stp->st_openstp = NULL;
4655 list_add(&stp->st_perstateowner, &oo->oo_owner.so_stateids);
4656 list_add(&stp->st_perfile, &fp->fi_stateids);
4659 spin_unlock(&fp->fi_lock);
4660 spin_unlock(&oo->oo_owner.so_client->cl_lock);
4662 /* Handle races with CLOSE */
4663 if (nfsd4_lock_ol_stateid(retstp) != nfs_ok) {
4664 nfs4_put_stid(&retstp->st_stid);
4667 /* To keep mutex tracking happy */
4668 mutex_unlock(&stp->st_mutex);
4675 * In the 4.0 case we need to keep the owners around a little while to handle
4676 * CLOSE replay. We still do need to release any file access that is held by
4677 * them before returning however.
4680 move_to_close_lru(struct nfs4_ol_stateid *s, struct net *net)
4682 struct nfs4_ol_stateid *last;
4683 struct nfs4_openowner *oo = openowner(s->st_stateowner);
4684 struct nfsd_net *nn = net_generic(s->st_stid.sc_client->net,
4687 dprintk("NFSD: move_to_close_lru nfs4_openowner %p\n", oo);
4690 * We know that we hold one reference via nfsd4_close, and another
4691 * "persistent" reference for the client. If the refcount is higher
4692 * than 2, then there are still calls in progress that are using this
4693 * stateid. We can't put the sc_file reference until they are finished.
4694 * Wait for the refcount to drop to 2. Since it has been unhashed,
4695 * there should be no danger of the refcount going back up again at
4698 wait_event(close_wq, refcount_read(&s->st_stid.sc_count) == 2);
4700 release_all_access(s);
4701 if (s->st_stid.sc_file) {
4702 put_nfs4_file(s->st_stid.sc_file);
4703 s->st_stid.sc_file = NULL;
4706 spin_lock(&nn->client_lock);
4707 last = oo->oo_last_closed_stid;
4708 oo->oo_last_closed_stid = s;
4709 list_move_tail(&oo->oo_close_lru, &nn->close_lru);
4710 oo->oo_time = ktime_get_boottime_seconds();
4711 spin_unlock(&nn->client_lock);
4713 nfs4_put_stid(&last->st_stid);
4716 static noinline_for_stack struct nfs4_file *
4717 nfsd4_file_hash_lookup(const struct svc_fh *fhp)
4719 struct inode *inode = d_inode(fhp->fh_dentry);
4720 struct rhlist_head *tmp, *list;
4721 struct nfs4_file *fi;
4724 list = rhltable_lookup(&nfs4_file_rhltable, &inode,
4725 nfs4_file_rhash_params);
4726 rhl_for_each_entry_rcu(fi, tmp, list, fi_rlist) {
4727 if (fh_match(&fi->fi_fhandle, &fhp->fh_handle)) {
4728 if (refcount_inc_not_zero(&fi->fi_ref)) {
4739 * On hash insertion, identify entries with the same inode but
4740 * distinct filehandles. They will all be on the list returned
4741 * by rhltable_lookup().
4743 * inode->i_lock prevents racing insertions from adding an entry
4744 * for the same inode/fhp pair twice.
4746 static noinline_for_stack struct nfs4_file *
4747 nfsd4_file_hash_insert(struct nfs4_file *new, const struct svc_fh *fhp)
4749 struct inode *inode = d_inode(fhp->fh_dentry);
4750 struct rhlist_head *tmp, *list;
4751 struct nfs4_file *ret = NULL;
4752 bool alias_found = false;
4753 struct nfs4_file *fi;
4757 spin_lock(&inode->i_lock);
4759 list = rhltable_lookup(&nfs4_file_rhltable, &inode,
4760 nfs4_file_rhash_params);
4761 rhl_for_each_entry_rcu(fi, tmp, list, fi_rlist) {
4762 if (fh_match(&fi->fi_fhandle, &fhp->fh_handle)) {
4763 if (refcount_inc_not_zero(&fi->fi_ref))
4766 fi->fi_aliased = alias_found = true;
4771 nfsd4_file_init(fhp, new);
4772 err = rhltable_insert(&nfs4_file_rhltable, &new->fi_rlist,
4773 nfs4_file_rhash_params);
4777 new->fi_aliased = alias_found;
4781 spin_unlock(&inode->i_lock);
4786 static noinline_for_stack void nfsd4_file_hash_remove(struct nfs4_file *fi)
4788 rhltable_remove(&nfs4_file_rhltable, &fi->fi_rlist,
4789 nfs4_file_rhash_params);
4793 * Called to check deny when READ with all zero stateid or
4794 * WRITE with all zero or all one stateid
4797 nfs4_share_conflict(struct svc_fh *current_fh, unsigned int deny_type)
4799 struct nfs4_file *fp;
4800 __be32 ret = nfs_ok;
4802 fp = nfsd4_file_hash_lookup(current_fh);
4806 /* Check for conflicting share reservations */
4807 spin_lock(&fp->fi_lock);
4808 if (fp->fi_share_deny & deny_type)
4809 ret = nfserr_locked;
4810 spin_unlock(&fp->fi_lock);
4815 static bool nfsd4_deleg_present(const struct inode *inode)
4817 struct file_lock_context *ctx = locks_inode_context(inode);
4819 return ctx && !list_empty_careful(&ctx->flc_lease);
4823 * nfsd_wait_for_delegreturn - wait for delegations to be returned
4824 * @rqstp: the RPC transaction being executed
4825 * @inode: in-core inode of the file being waited for
4827 * The timeout prevents deadlock if all nfsd threads happen to be
4828 * tied up waiting for returning delegations.
4831 * %true: delegation was returned
4832 * %false: timed out waiting for delegreturn
4834 bool nfsd_wait_for_delegreturn(struct svc_rqst *rqstp, struct inode *inode)
4836 long __maybe_unused timeo;
4838 timeo = wait_var_event_timeout(inode, !nfsd4_deleg_present(inode),
4839 NFSD_DELEGRETURN_TIMEOUT);
4840 trace_nfsd_delegret_wakeup(rqstp, inode, timeo);
4844 static void nfsd4_cb_recall_prepare(struct nfsd4_callback *cb)
4846 struct nfs4_delegation *dp = cb_to_delegation(cb);
4847 struct nfsd_net *nn = net_generic(dp->dl_stid.sc_client->net,
4850 block_delegations(&dp->dl_stid.sc_file->fi_fhandle);
4853 * We can't do this in nfsd_break_deleg_cb because it is
4854 * already holding inode->i_lock.
4856 * If the dl_time != 0, then we know that it has already been
4857 * queued for a lease break. Don't queue it again.
4859 spin_lock(&state_lock);
4860 if (delegation_hashed(dp) && dp->dl_time == 0) {
4861 dp->dl_time = ktime_get_boottime_seconds();
4862 list_add_tail(&dp->dl_recall_lru, &nn->del_recall_lru);
4864 spin_unlock(&state_lock);
4867 static int nfsd4_cb_recall_done(struct nfsd4_callback *cb,
4868 struct rpc_task *task)
4870 struct nfs4_delegation *dp = cb_to_delegation(cb);
4872 trace_nfsd_cb_recall_done(&dp->dl_stid.sc_stateid, task);
4874 if (dp->dl_stid.sc_type == NFS4_CLOSED_DELEG_STID ||
4875 dp->dl_stid.sc_type == NFS4_REVOKED_DELEG_STID)
4878 switch (task->tk_status) {
4881 case -NFS4ERR_DELAY:
4882 rpc_delay(task, 2 * HZ);
4885 case -NFS4ERR_BAD_STATEID:
4887 * Race: client probably got cb_recall before open reply
4888 * granting delegation.
4890 if (dp->dl_retries--) {
4891 rpc_delay(task, 2 * HZ);
4900 static void nfsd4_cb_recall_release(struct nfsd4_callback *cb)
4902 struct nfs4_delegation *dp = cb_to_delegation(cb);
4904 nfs4_put_stid(&dp->dl_stid);
4907 static const struct nfsd4_callback_ops nfsd4_cb_recall_ops = {
4908 .prepare = nfsd4_cb_recall_prepare,
4909 .done = nfsd4_cb_recall_done,
4910 .release = nfsd4_cb_recall_release,
4913 static void nfsd_break_one_deleg(struct nfs4_delegation *dp)
4916 * We're assuming the state code never drops its reference
4917 * without first removing the lease. Since we're in this lease
4918 * callback (and since the lease code is serialized by the
4919 * flc_lock) we know the server hasn't removed the lease yet, and
4920 * we know it's safe to take a reference.
4922 refcount_inc(&dp->dl_stid.sc_count);
4923 WARN_ON_ONCE(!nfsd4_run_cb(&dp->dl_recall));
4926 /* Called from break_lease() with flc_lock held. */
4928 nfsd_break_deleg_cb(struct file_lock *fl)
4930 struct nfs4_delegation *dp = (struct nfs4_delegation *)fl->fl_owner;
4931 struct nfs4_file *fp = dp->dl_stid.sc_file;
4932 struct nfs4_client *clp = dp->dl_stid.sc_client;
4933 struct nfsd_net *nn;
4935 trace_nfsd_cb_recall(&dp->dl_stid);
4937 dp->dl_recalled = true;
4938 atomic_inc(&clp->cl_delegs_in_recall);
4939 if (try_to_expire_client(clp)) {
4940 nn = net_generic(clp->net, nfsd_net_id);
4941 mod_delayed_work(laundry_wq, &nn->laundromat_work, 0);
4945 * We don't want the locks code to timeout the lease for us;
4946 * we'll remove it ourself if a delegation isn't returned
4949 fl->fl_break_time = 0;
4951 spin_lock(&fp->fi_lock);
4952 fp->fi_had_conflict = true;
4953 nfsd_break_one_deleg(dp);
4954 spin_unlock(&fp->fi_lock);
4959 * nfsd_breaker_owns_lease - Check if lease conflict was resolved
4960 * @fl: Lock state to check
4963 * %true: Lease conflict was resolved
4964 * %false: Lease conflict was not resolved.
4966 static bool nfsd_breaker_owns_lease(struct file_lock *fl)
4968 struct nfs4_delegation *dl = fl->fl_owner;
4969 struct svc_rqst *rqst;
4970 struct nfs4_client *clp;
4974 rqst = kthread_data(current);
4975 /* Note rq_prog == NFS_ACL_PROGRAM is also possible: */
4976 if (rqst->rq_prog != NFS_PROGRAM || rqst->rq_vers < 4)
4978 clp = *(rqst->rq_lease_breaker);
4979 return dl->dl_stid.sc_client == clp;
4983 nfsd_change_deleg_cb(struct file_lock *onlist, int arg,
4984 struct list_head *dispose)
4986 struct nfs4_delegation *dp = (struct nfs4_delegation *)onlist->fl_owner;
4987 struct nfs4_client *clp = dp->dl_stid.sc_client;
4989 if (arg & F_UNLCK) {
4990 if (dp->dl_recalled)
4991 atomic_dec(&clp->cl_delegs_in_recall);
4992 return lease_modify(onlist, arg, dispose);
4997 static const struct lock_manager_operations nfsd_lease_mng_ops = {
4998 .lm_breaker_owns_lease = nfsd_breaker_owns_lease,
4999 .lm_break = nfsd_break_deleg_cb,
5000 .lm_change = nfsd_change_deleg_cb,
5003 static __be32 nfsd4_check_seqid(struct nfsd4_compound_state *cstate, struct nfs4_stateowner *so, u32 seqid)
5005 if (nfsd4_has_session(cstate))
5007 if (seqid == so->so_seqid - 1)
5008 return nfserr_replay_me;
5009 if (seqid == so->so_seqid)
5011 return nfserr_bad_seqid;
5014 static struct nfs4_client *lookup_clientid(clientid_t *clid, bool sessions,
5015 struct nfsd_net *nn)
5017 struct nfs4_client *found;
5019 spin_lock(&nn->client_lock);
5020 found = find_confirmed_client(clid, sessions, nn);
5022 atomic_inc(&found->cl_rpc_users);
5023 spin_unlock(&nn->client_lock);
5027 static __be32 set_client(clientid_t *clid,
5028 struct nfsd4_compound_state *cstate,
5029 struct nfsd_net *nn)
5032 if (!same_clid(&cstate->clp->cl_clientid, clid))
5033 return nfserr_stale_clientid;
5036 if (STALE_CLIENTID(clid, nn))
5037 return nfserr_stale_clientid;
5039 * We're in the 4.0 case (otherwise the SEQUENCE op would have
5040 * set cstate->clp), so session = false:
5042 cstate->clp = lookup_clientid(clid, false, nn);
5044 return nfserr_expired;
5049 nfsd4_process_open1(struct nfsd4_compound_state *cstate,
5050 struct nfsd4_open *open, struct nfsd_net *nn)
5052 clientid_t *clientid = &open->op_clientid;
5053 struct nfs4_client *clp = NULL;
5054 unsigned int strhashval;
5055 struct nfs4_openowner *oo = NULL;
5059 * In case we need it later, after we've already created the
5060 * file and don't want to risk a further failure:
5062 open->op_file = nfsd4_alloc_file();
5063 if (open->op_file == NULL)
5064 return nfserr_jukebox;
5066 status = set_client(clientid, cstate, nn);
5071 strhashval = ownerstr_hashval(&open->op_owner);
5072 oo = find_openstateowner_str(strhashval, open, clp);
5073 open->op_openowner = oo;
5077 if (!(oo->oo_flags & NFS4_OO_CONFIRMED)) {
5078 /* Replace unconfirmed owners without checking for replay. */
5079 release_openowner(oo);
5080 open->op_openowner = NULL;
5083 status = nfsd4_check_seqid(cstate, &oo->oo_owner, open->op_seqid);
5088 oo = alloc_init_open_stateowner(strhashval, open, cstate);
5090 return nfserr_jukebox;
5091 open->op_openowner = oo;
5093 open->op_stp = nfs4_alloc_open_stateid(clp);
5095 return nfserr_jukebox;
5097 if (nfsd4_has_session(cstate) &&
5098 (cstate->current_fh.fh_export->ex_flags & NFSEXP_PNFS)) {
5099 open->op_odstate = alloc_clnt_odstate(clp);
5100 if (!open->op_odstate)
5101 return nfserr_jukebox;
5107 static inline __be32
5108 nfs4_check_delegmode(struct nfs4_delegation *dp, int flags)
5110 if ((flags & WR_STATE) && (dp->dl_type == NFS4_OPEN_DELEGATE_READ))
5111 return nfserr_openmode;
5116 static int share_access_to_flags(u32 share_access)
5118 return share_access == NFS4_SHARE_ACCESS_READ ? RD_STATE : WR_STATE;
5121 static struct nfs4_delegation *find_deleg_stateid(struct nfs4_client *cl, stateid_t *s)
5123 struct nfs4_stid *ret;
5125 ret = find_stateid_by_type(cl, s,
5126 NFS4_DELEG_STID|NFS4_REVOKED_DELEG_STID);
5129 return delegstateid(ret);
5132 static bool nfsd4_is_deleg_cur(struct nfsd4_open *open)
5134 return open->op_claim_type == NFS4_OPEN_CLAIM_DELEGATE_CUR ||
5135 open->op_claim_type == NFS4_OPEN_CLAIM_DELEG_CUR_FH;
5139 nfs4_check_deleg(struct nfs4_client *cl, struct nfsd4_open *open,
5140 struct nfs4_delegation **dp)
5143 __be32 status = nfserr_bad_stateid;
5144 struct nfs4_delegation *deleg;
5146 deleg = find_deleg_stateid(cl, &open->op_delegate_stateid);
5149 if (deleg->dl_stid.sc_type == NFS4_REVOKED_DELEG_STID) {
5150 nfs4_put_stid(&deleg->dl_stid);
5151 if (cl->cl_minorversion)
5152 status = nfserr_deleg_revoked;
5155 flags = share_access_to_flags(open->op_share_access);
5156 status = nfs4_check_delegmode(deleg, flags);
5158 nfs4_put_stid(&deleg->dl_stid);
5163 if (!nfsd4_is_deleg_cur(open))
5167 open->op_openowner->oo_flags |= NFS4_OO_CONFIRMED;
5171 static inline int nfs4_access_to_access(u32 nfs4_access)
5175 if (nfs4_access & NFS4_SHARE_ACCESS_READ)
5176 flags |= NFSD_MAY_READ;
5177 if (nfs4_access & NFS4_SHARE_ACCESS_WRITE)
5178 flags |= NFSD_MAY_WRITE;
5182 static inline __be32
5183 nfsd4_truncate(struct svc_rqst *rqstp, struct svc_fh *fh,
5184 struct nfsd4_open *open)
5186 struct iattr iattr = {
5187 .ia_valid = ATTR_SIZE,
5190 struct nfsd_attrs attrs = {
5193 if (!open->op_truncate)
5195 if (!(open->op_share_access & NFS4_SHARE_ACCESS_WRITE))
5196 return nfserr_inval;
5197 return nfsd_setattr(rqstp, fh, &attrs, 0, (time64_t)0);
5200 static __be32 nfs4_get_vfs_file(struct svc_rqst *rqstp, struct nfs4_file *fp,
5201 struct svc_fh *cur_fh, struct nfs4_ol_stateid *stp,
5202 struct nfsd4_open *open, bool new_stp)
5204 struct nfsd_file *nf = NULL;
5206 int oflag = nfs4_access_to_omode(open->op_share_access);
5207 int access = nfs4_access_to_access(open->op_share_access);
5208 unsigned char old_access_bmap, old_deny_bmap;
5210 spin_lock(&fp->fi_lock);
5213 * Are we trying to set a deny mode that would conflict with
5216 status = nfs4_file_check_deny(fp, open->op_share_deny);
5217 if (status != nfs_ok) {
5218 if (status != nfserr_share_denied) {
5219 spin_unlock(&fp->fi_lock);
5222 if (nfs4_resolve_deny_conflicts_locked(fp, new_stp,
5223 stp, open->op_share_deny, false))
5224 status = nfserr_jukebox;
5225 spin_unlock(&fp->fi_lock);
5229 /* set access to the file */
5230 status = nfs4_file_get_access(fp, open->op_share_access);
5231 if (status != nfs_ok) {
5232 if (status != nfserr_share_denied) {
5233 spin_unlock(&fp->fi_lock);
5236 if (nfs4_resolve_deny_conflicts_locked(fp, new_stp,
5237 stp, open->op_share_access, true))
5238 status = nfserr_jukebox;
5239 spin_unlock(&fp->fi_lock);
5243 /* Set access bits in stateid */
5244 old_access_bmap = stp->st_access_bmap;
5245 set_access(open->op_share_access, stp);
5247 /* Set new deny mask */
5248 old_deny_bmap = stp->st_deny_bmap;
5249 set_deny(open->op_share_deny, stp);
5250 fp->fi_share_deny |= (open->op_share_deny & NFS4_SHARE_DENY_BOTH);
5252 if (!fp->fi_fds[oflag]) {
5253 spin_unlock(&fp->fi_lock);
5255 status = nfsd_file_acquire_opened(rqstp, cur_fh, access,
5256 open->op_filp, &nf);
5257 if (status != nfs_ok)
5258 goto out_put_access;
5260 spin_lock(&fp->fi_lock);
5261 if (!fp->fi_fds[oflag]) {
5262 fp->fi_fds[oflag] = nf;
5266 spin_unlock(&fp->fi_lock);
5270 status = nfserrno(nfsd_open_break_lease(cur_fh->fh_dentry->d_inode,
5273 goto out_put_access;
5275 status = nfsd4_truncate(rqstp, cur_fh, open);
5277 goto out_put_access;
5281 stp->st_access_bmap = old_access_bmap;
5282 nfs4_file_put_access(fp, open->op_share_access);
5283 reset_union_bmap_deny(bmap_to_share_mode(old_deny_bmap), stp);
5288 nfs4_upgrade_open(struct svc_rqst *rqstp, struct nfs4_file *fp,
5289 struct svc_fh *cur_fh, struct nfs4_ol_stateid *stp,
5290 struct nfsd4_open *open)
5293 unsigned char old_deny_bmap = stp->st_deny_bmap;
5295 if (!test_access(open->op_share_access, stp))
5296 return nfs4_get_vfs_file(rqstp, fp, cur_fh, stp, open, false);
5298 /* test and set deny mode */
5299 spin_lock(&fp->fi_lock);
5300 status = nfs4_file_check_deny(fp, open->op_share_deny);
5301 if (status == nfs_ok) {
5302 if (status != nfserr_share_denied) {
5303 set_deny(open->op_share_deny, stp);
5304 fp->fi_share_deny |=
5305 (open->op_share_deny & NFS4_SHARE_DENY_BOTH);
5307 if (nfs4_resolve_deny_conflicts_locked(fp, false,
5308 stp, open->op_share_deny, false))
5309 status = nfserr_jukebox;
5312 spin_unlock(&fp->fi_lock);
5314 if (status != nfs_ok)
5317 status = nfsd4_truncate(rqstp, cur_fh, open);
5318 if (status != nfs_ok)
5319 reset_union_bmap_deny(old_deny_bmap, stp);
5323 /* Should we give out recallable state?: */
5324 static bool nfsd4_cb_channel_good(struct nfs4_client *clp)
5326 if (clp->cl_cb_state == NFSD4_CB_UP)
5329 * In the sessions case, since we don't have to establish a
5330 * separate connection for callbacks, we assume it's OK
5331 * until we hear otherwise:
5333 return clp->cl_minorversion && clp->cl_cb_state == NFSD4_CB_UNKNOWN;
5336 static struct file_lock *nfs4_alloc_init_lease(struct nfs4_delegation *dp,
5339 struct file_lock *fl;
5341 fl = locks_alloc_lock();
5344 fl->fl_lmops = &nfsd_lease_mng_ops;
5345 fl->fl_flags = FL_DELEG;
5346 fl->fl_type = flag == NFS4_OPEN_DELEGATE_READ? F_RDLCK: F_WRLCK;
5347 fl->fl_end = OFFSET_MAX;
5348 fl->fl_owner = (fl_owner_t)dp;
5349 fl->fl_pid = current->tgid;
5350 fl->fl_file = dp->dl_stid.sc_file->fi_deleg_file->nf_file;
5354 static int nfsd4_check_conflicting_opens(struct nfs4_client *clp,
5355 struct nfs4_file *fp)
5357 struct nfs4_ol_stateid *st;
5358 struct file *f = fp->fi_deleg_file->nf_file;
5359 struct inode *ino = locks_inode(f);
5362 writes = atomic_read(&ino->i_writecount);
5366 * There could be multiple filehandles (hence multiple
5367 * nfs4_files) referencing this file, but that's not too
5368 * common; let's just give up in that case rather than
5369 * trying to go look up all the clients using that other
5370 * nfs4_file as well:
5375 * If there's a close in progress, make sure that we see it
5376 * clear any fi_fds[] entries before we see it decrement
5379 smp_mb__after_atomic();
5381 if (fp->fi_fds[O_WRONLY])
5383 if (fp->fi_fds[O_RDWR])
5386 return -EAGAIN; /* There may be non-NFSv4 writers */
5388 * It's possible there are non-NFSv4 write opens in progress,
5389 * but if they haven't incremented i_writecount yet then they
5390 * also haven't called break lease yet; so, they'll break this
5391 * lease soon enough. So, all that's left to check for is NFSv4
5394 spin_lock(&fp->fi_lock);
5395 list_for_each_entry(st, &fp->fi_stateids, st_perfile) {
5396 if (st->st_openstp == NULL /* it's an open */ &&
5397 access_permit_write(st) &&
5398 st->st_stid.sc_client != clp) {
5399 spin_unlock(&fp->fi_lock);
5403 spin_unlock(&fp->fi_lock);
5405 * There's a small chance that we could be racing with another
5406 * NFSv4 open. However, any open that hasn't added itself to
5407 * the fi_stateids list also hasn't called break_lease yet; so,
5408 * they'll break this lease soon enough.
5414 * It's possible that between opening the dentry and setting the delegation,
5415 * that it has been renamed or unlinked. Redo the lookup to verify that this
5419 nfsd4_verify_deleg_dentry(struct nfsd4_open *open, struct nfs4_file *fp,
5420 struct svc_fh *parent)
5422 struct svc_export *exp;
5423 struct dentry *child;
5426 err = nfsd_lookup_dentry(open->op_rqstp, parent,
5427 open->op_fname, open->op_fnamelen,
5435 if (child != file_dentry(fp->fi_deleg_file->nf_file))
5441 static struct nfs4_delegation *
5442 nfs4_set_delegation(struct nfsd4_open *open, struct nfs4_ol_stateid *stp,
5443 struct svc_fh *parent)
5446 struct nfs4_client *clp = stp->st_stid.sc_client;
5447 struct nfs4_file *fp = stp->st_stid.sc_file;
5448 struct nfs4_clnt_odstate *odstate = stp->st_clnt_odstate;
5449 struct nfs4_delegation *dp;
5450 struct nfsd_file *nf;
5451 struct file_lock *fl;
5454 * The fi_had_conflict and nfs_get_existing_delegation checks
5455 * here are just optimizations; we'll need to recheck them at
5458 if (fp->fi_had_conflict)
5459 return ERR_PTR(-EAGAIN);
5461 nf = find_readable_file(fp);
5464 * We probably could attempt another open and get a read
5465 * delegation, but for now, don't bother until the
5466 * client actually sends us one.
5468 return ERR_PTR(-EAGAIN);
5470 spin_lock(&state_lock);
5471 spin_lock(&fp->fi_lock);
5472 if (nfs4_delegation_exists(clp, fp))
5474 else if (!fp->fi_deleg_file) {
5475 fp->fi_deleg_file = nf;
5476 /* increment early to prevent fi_deleg_file from being
5478 fp->fi_delegees = 1;
5482 spin_unlock(&fp->fi_lock);
5483 spin_unlock(&state_lock);
5487 return ERR_PTR(status);
5490 dp = alloc_init_deleg(clp, fp, odstate);
5494 fl = nfs4_alloc_init_lease(dp, NFS4_OPEN_DELEGATE_READ);
5496 goto out_clnt_odstate;
5498 status = vfs_setlease(fp->fi_deleg_file->nf_file, fl->fl_type, &fl, NULL);
5500 locks_free_lock(fl);
5502 goto out_clnt_odstate;
5505 status = nfsd4_verify_deleg_dentry(open, fp, parent);
5510 status = nfsd4_check_conflicting_opens(clp, fp);
5514 spin_lock(&state_lock);
5515 spin_lock(&fp->fi_lock);
5516 if (fp->fi_had_conflict)
5519 status = hash_delegation_locked(dp, fp);
5520 spin_unlock(&fp->fi_lock);
5521 spin_unlock(&state_lock);
5528 vfs_setlease(fp->fi_deleg_file->nf_file, F_UNLCK, NULL, (void **)&dp);
5530 put_clnt_odstate(dp->dl_clnt_odstate);
5531 nfs4_put_stid(&dp->dl_stid);
5534 return ERR_PTR(status);
5537 static void nfsd4_open_deleg_none_ext(struct nfsd4_open *open, int status)
5539 open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE_EXT;
5540 if (status == -EAGAIN)
5541 open->op_why_no_deleg = WND4_CONTENTION;
5543 open->op_why_no_deleg = WND4_RESOURCE;
5544 switch (open->op_deleg_want) {
5545 case NFS4_SHARE_WANT_READ_DELEG:
5546 case NFS4_SHARE_WANT_WRITE_DELEG:
5547 case NFS4_SHARE_WANT_ANY_DELEG:
5549 case NFS4_SHARE_WANT_CANCEL:
5550 open->op_why_no_deleg = WND4_CANCELLED;
5552 case NFS4_SHARE_WANT_NO_DELEG:
5559 * Attempt to hand out a delegation.
5561 * Note we don't support write delegations, and won't until the vfs has
5562 * proper support for them.
5565 nfs4_open_delegation(struct nfsd4_open *open, struct nfs4_ol_stateid *stp,
5566 struct svc_fh *currentfh)
5568 struct nfs4_delegation *dp;
5569 struct nfs4_openowner *oo = openowner(stp->st_stateowner);
5570 struct nfs4_client *clp = stp->st_stid.sc_client;
5571 struct svc_fh *parent = NULL;
5575 cb_up = nfsd4_cb_channel_good(oo->oo_owner.so_client);
5576 open->op_recall = 0;
5577 switch (open->op_claim_type) {
5578 case NFS4_OPEN_CLAIM_PREVIOUS:
5580 open->op_recall = 1;
5581 if (open->op_delegate_type != NFS4_OPEN_DELEGATE_READ)
5584 case NFS4_OPEN_CLAIM_NULL:
5587 case NFS4_OPEN_CLAIM_FH:
5589 * Let's not give out any delegations till everyone's
5590 * had the chance to reclaim theirs, *and* until
5591 * NLM locks have all been reclaimed:
5593 if (locks_in_grace(clp->net))
5595 if (!cb_up || !(oo->oo_flags & NFS4_OO_CONFIRMED))
5601 dp = nfs4_set_delegation(open, stp, parent);
5605 memcpy(&open->op_delegate_stateid, &dp->dl_stid.sc_stateid, sizeof(dp->dl_stid.sc_stateid));
5607 trace_nfsd_deleg_read(&dp->dl_stid.sc_stateid);
5608 open->op_delegate_type = NFS4_OPEN_DELEGATE_READ;
5609 nfs4_put_stid(&dp->dl_stid);
5612 open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE;
5613 if (open->op_claim_type == NFS4_OPEN_CLAIM_PREVIOUS &&
5614 open->op_delegate_type != NFS4_OPEN_DELEGATE_NONE) {
5615 dprintk("NFSD: WARNING: refusing delegation reclaim\n");
5616 open->op_recall = 1;
5619 /* 4.1 client asking for a delegation? */
5620 if (open->op_deleg_want)
5621 nfsd4_open_deleg_none_ext(open, status);
5625 static void nfsd4_deleg_xgrade_none_ext(struct nfsd4_open *open,
5626 struct nfs4_delegation *dp)
5628 if (open->op_deleg_want == NFS4_SHARE_WANT_READ_DELEG &&
5629 dp->dl_type == NFS4_OPEN_DELEGATE_WRITE) {
5630 open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE_EXT;
5631 open->op_why_no_deleg = WND4_NOT_SUPP_DOWNGRADE;
5632 } else if (open->op_deleg_want == NFS4_SHARE_WANT_WRITE_DELEG &&
5633 dp->dl_type == NFS4_OPEN_DELEGATE_WRITE) {
5634 open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE_EXT;
5635 open->op_why_no_deleg = WND4_NOT_SUPP_UPGRADE;
5637 /* Otherwise the client must be confused wanting a delegation
5638 * it already has, therefore we don't return
5639 * NFS4_OPEN_DELEGATE_NONE_EXT and reason.
5644 * nfsd4_process_open2 - finish open processing
5645 * @rqstp: the RPC transaction being executed
5646 * @current_fh: NFSv4 COMPOUND's current filehandle
5647 * @open: OPEN arguments
5649 * If successful, (1) truncate the file if open->op_truncate was
5650 * set, (2) set open->op_stateid, (3) set open->op_delegation.
5652 * Returns %nfs_ok on success; otherwise an nfs4stat value in
5653 * network byte order is returned.
5656 nfsd4_process_open2(struct svc_rqst *rqstp, struct svc_fh *current_fh, struct nfsd4_open *open)
5658 struct nfsd4_compoundres *resp = rqstp->rq_resp;
5659 struct nfs4_client *cl = open->op_openowner->oo_owner.so_client;
5660 struct nfs4_file *fp = NULL;
5661 struct nfs4_ol_stateid *stp = NULL;
5662 struct nfs4_delegation *dp = NULL;
5664 bool new_stp = false;
5667 * Lookup file; if found, lookup stateid and check open request,
5668 * and check for delegations in the process of being recalled.
5669 * If not found, create the nfs4_file struct
5671 fp = nfsd4_file_hash_insert(open->op_file, current_fh);
5673 return nfserr_jukebox;
5674 if (fp != open->op_file) {
5675 status = nfs4_check_deleg(cl, open, &dp);
5678 stp = nfsd4_find_and_lock_existing_open(fp, open);
5680 open->op_file = NULL;
5681 status = nfserr_bad_stateid;
5682 if (nfsd4_is_deleg_cur(open))
5687 stp = init_open_stateid(fp, open);
5693 * OPEN the file, or upgrade an existing OPEN.
5694 * If truncate fails, the OPEN fails.
5696 * stp is already locked.
5699 /* Stateid was found, this is an OPEN upgrade */
5700 status = nfs4_upgrade_open(rqstp, fp, current_fh, stp, open);
5702 mutex_unlock(&stp->st_mutex);
5706 status = nfs4_get_vfs_file(rqstp, fp, current_fh, stp, open, true);
5708 stp->st_stid.sc_type = NFS4_CLOSED_STID;
5709 release_open_stateid(stp);
5710 mutex_unlock(&stp->st_mutex);
5714 stp->st_clnt_odstate = find_or_hash_clnt_odstate(fp,
5716 if (stp->st_clnt_odstate == open->op_odstate)
5717 open->op_odstate = NULL;
5720 nfs4_inc_and_copy_stateid(&open->op_stateid, &stp->st_stid);
5721 mutex_unlock(&stp->st_mutex);
5723 if (nfsd4_has_session(&resp->cstate)) {
5724 if (open->op_deleg_want & NFS4_SHARE_WANT_NO_DELEG) {
5725 open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE_EXT;
5726 open->op_why_no_deleg = WND4_NOT_WANTED;
5732 * Attempt to hand out a delegation. No error return, because the
5733 * OPEN succeeds even if we fail.
5735 nfs4_open_delegation(open, stp, &resp->cstate.current_fh);
5738 trace_nfsd_open(&stp->st_stid.sc_stateid);
5740 /* 4.1 client trying to upgrade/downgrade delegation? */
5741 if (open->op_delegate_type == NFS4_OPEN_DELEGATE_NONE && dp &&
5742 open->op_deleg_want)
5743 nfsd4_deleg_xgrade_none_ext(open, dp);
5747 if (status == 0 && open->op_claim_type == NFS4_OPEN_CLAIM_PREVIOUS)
5748 open->op_openowner->oo_flags |= NFS4_OO_CONFIRMED;
5750 * To finish the open response, we just need to set the rflags.
5752 open->op_rflags = NFS4_OPEN_RESULT_LOCKTYPE_POSIX;
5753 if (nfsd4_has_session(&resp->cstate))
5754 open->op_rflags |= NFS4_OPEN_RESULT_MAY_NOTIFY_LOCK;
5755 else if (!(open->op_openowner->oo_flags & NFS4_OO_CONFIRMED))
5756 open->op_rflags |= NFS4_OPEN_RESULT_CONFIRM;
5759 nfs4_put_stid(&dp->dl_stid);
5761 nfs4_put_stid(&stp->st_stid);
5766 void nfsd4_cleanup_open_state(struct nfsd4_compound_state *cstate,
5767 struct nfsd4_open *open)
5769 if (open->op_openowner) {
5770 struct nfs4_stateowner *so = &open->op_openowner->oo_owner;
5772 nfsd4_cstate_assign_replay(cstate, so);
5773 nfs4_put_stateowner(so);
5776 kmem_cache_free(file_slab, open->op_file);
5778 nfs4_put_stid(&open->op_stp->st_stid);
5779 if (open->op_odstate)
5780 kmem_cache_free(odstate_slab, open->op_odstate);
5784 nfsd4_renew(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
5785 union nfsd4_op_u *u)
5787 clientid_t *clid = &u->renew;
5788 struct nfs4_client *clp;
5790 struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
5792 trace_nfsd_clid_renew(clid);
5793 status = set_client(clid, cstate, nn);
5797 if (!list_empty(&clp->cl_delegations)
5798 && clp->cl_cb_state != NFSD4_CB_UP)
5799 return nfserr_cb_path_down;
5804 nfsd4_end_grace(struct nfsd_net *nn)
5806 /* do nothing if grace period already ended */
5807 if (nn->grace_ended)
5810 trace_nfsd_grace_complete(nn);
5811 nn->grace_ended = true;
5813 * If the server goes down again right now, an NFSv4
5814 * client will still be allowed to reclaim after it comes back up,
5815 * even if it hasn't yet had a chance to reclaim state this time.
5818 nfsd4_record_grace_done(nn);
5820 * At this point, NFSv4 clients can still reclaim. But if the
5821 * server crashes, any that have not yet reclaimed will be out
5822 * of luck on the next boot.
5824 * (NFSv4.1+ clients are considered to have reclaimed once they
5825 * call RECLAIM_COMPLETE. NFSv4.0 clients are considered to
5826 * have reclaimed after their first OPEN.)
5828 locks_end_grace(&nn->nfsd4_manager);
5830 * At this point, and once lockd and/or any other containers
5831 * exit their grace period, further reclaims will fail and
5832 * regular locking can resume.
5837 * If we've waited a lease period but there are still clients trying to
5838 * reclaim, wait a little longer to give them a chance to finish.
5840 static bool clients_still_reclaiming(struct nfsd_net *nn)
5842 time64_t double_grace_period_end = nn->boot_time +
5843 2 * nn->nfsd4_lease;
5845 if (nn->track_reclaim_completes &&
5846 atomic_read(&nn->nr_reclaim_complete) ==
5847 nn->reclaim_str_hashtbl_size)
5849 if (!nn->somebody_reclaimed)
5851 nn->somebody_reclaimed = false;
5853 * If we've given them *two* lease times to reclaim, and they're
5854 * still not done, give up:
5856 if (ktime_get_boottime_seconds() > double_grace_period_end)
5861 struct laundry_time {
5866 static bool state_expired(struct laundry_time *lt, time64_t last_refresh)
5868 time64_t time_remaining;
5870 if (last_refresh < lt->cutoff)
5872 time_remaining = last_refresh - lt->cutoff;
5873 lt->new_timeo = min(lt->new_timeo, time_remaining);
5877 #ifdef CONFIG_NFSD_V4_2_INTER_SSC
5878 void nfsd4_ssc_init_umount_work(struct nfsd_net *nn)
5880 spin_lock_init(&nn->nfsd_ssc_lock);
5881 INIT_LIST_HEAD(&nn->nfsd_ssc_mount_list);
5882 init_waitqueue_head(&nn->nfsd_ssc_waitq);
5884 EXPORT_SYMBOL_GPL(nfsd4_ssc_init_umount_work);
5887 * This is called when nfsd is being shutdown, after all inter_ssc
5888 * cleanup were done, to destroy the ssc delayed unmount list.
5890 static void nfsd4_ssc_shutdown_umount(struct nfsd_net *nn)
5892 struct nfsd4_ssc_umount_item *ni = NULL;
5893 struct nfsd4_ssc_umount_item *tmp;
5895 spin_lock(&nn->nfsd_ssc_lock);
5896 list_for_each_entry_safe(ni, tmp, &nn->nfsd_ssc_mount_list, nsui_list) {
5897 list_del(&ni->nsui_list);
5898 spin_unlock(&nn->nfsd_ssc_lock);
5899 mntput(ni->nsui_vfsmount);
5901 spin_lock(&nn->nfsd_ssc_lock);
5903 spin_unlock(&nn->nfsd_ssc_lock);
5906 static void nfsd4_ssc_expire_umount(struct nfsd_net *nn)
5908 bool do_wakeup = false;
5909 struct nfsd4_ssc_umount_item *ni = NULL;
5910 struct nfsd4_ssc_umount_item *tmp;
5912 spin_lock(&nn->nfsd_ssc_lock);
5913 list_for_each_entry_safe(ni, tmp, &nn->nfsd_ssc_mount_list, nsui_list) {
5914 if (time_after(jiffies, ni->nsui_expire)) {
5915 if (refcount_read(&ni->nsui_refcnt) > 1)
5918 /* mark being unmount */
5919 ni->nsui_busy = true;
5920 spin_unlock(&nn->nfsd_ssc_lock);
5921 mntput(ni->nsui_vfsmount);
5922 spin_lock(&nn->nfsd_ssc_lock);
5924 /* waiters need to start from begin of list */
5925 list_del(&ni->nsui_list);
5928 /* wakeup ssc_connect waiters */
5935 wake_up_all(&nn->nfsd_ssc_waitq);
5936 spin_unlock(&nn->nfsd_ssc_lock);
5940 /* Check if any lock belonging to this lockowner has any blockers */
5942 nfs4_lockowner_has_blockers(struct nfs4_lockowner *lo)
5944 struct file_lock_context *ctx;
5945 struct nfs4_ol_stateid *stp;
5946 struct nfs4_file *nf;
5948 list_for_each_entry(stp, &lo->lo_owner.so_stateids, st_perstateowner) {
5949 nf = stp->st_stid.sc_file;
5950 ctx = locks_inode_context(nf->fi_inode);
5953 if (locks_owner_has_blockers(ctx, lo))
5960 nfs4_anylock_blockers(struct nfs4_client *clp)
5963 struct nfs4_stateowner *so;
5964 struct nfs4_lockowner *lo;
5966 if (atomic_read(&clp->cl_delegs_in_recall))
5968 spin_lock(&clp->cl_lock);
5969 for (i = 0; i < OWNER_HASH_SIZE; i++) {
5970 list_for_each_entry(so, &clp->cl_ownerstr_hashtbl[i],
5972 if (so->so_is_open_owner)
5975 if (nfs4_lockowner_has_blockers(lo)) {
5976 spin_unlock(&clp->cl_lock);
5981 spin_unlock(&clp->cl_lock);
5986 nfs4_get_client_reaplist(struct nfsd_net *nn, struct list_head *reaplist,
5987 struct laundry_time *lt)
5989 unsigned int maxreap, reapcnt = 0;
5990 struct list_head *pos, *next;
5991 struct nfs4_client *clp;
5993 maxreap = (atomic_read(&nn->nfs4_client_count) >= nn->nfs4_max_clients) ?
5994 NFSD_CLIENT_MAX_TRIM_PER_RUN : 0;
5995 INIT_LIST_HEAD(reaplist);
5996 spin_lock(&nn->client_lock);
5997 list_for_each_safe(pos, next, &nn->client_lru) {
5998 clp = list_entry(pos, struct nfs4_client, cl_lru);
5999 if (clp->cl_state == NFSD4_EXPIRABLE)
6001 if (!state_expired(lt, clp->cl_time))
6003 if (!atomic_read(&clp->cl_rpc_users)) {
6004 if (clp->cl_state == NFSD4_ACTIVE)
6005 atomic_inc(&nn->nfsd_courtesy_clients);
6006 clp->cl_state = NFSD4_COURTESY;
6008 if (!client_has_state(clp))
6010 if (!nfs4_anylock_blockers(clp))
6011 if (reapcnt >= maxreap)
6014 if (!mark_client_expired_locked(clp)) {
6015 list_add(&clp->cl_lru, reaplist);
6019 spin_unlock(&nn->client_lock);
6023 nfs4_get_courtesy_client_reaplist(struct nfsd_net *nn,
6024 struct list_head *reaplist)
6026 unsigned int maxreap = 0, reapcnt = 0;
6027 struct list_head *pos, *next;
6028 struct nfs4_client *clp;
6030 maxreap = NFSD_CLIENT_MAX_TRIM_PER_RUN;
6031 INIT_LIST_HEAD(reaplist);
6033 spin_lock(&nn->client_lock);
6034 list_for_each_safe(pos, next, &nn->client_lru) {
6035 clp = list_entry(pos, struct nfs4_client, cl_lru);
6036 if (clp->cl_state == NFSD4_ACTIVE)
6038 if (reapcnt >= maxreap)
6040 if (!mark_client_expired_locked(clp)) {
6041 list_add(&clp->cl_lru, reaplist);
6045 spin_unlock(&nn->client_lock);
6049 nfs4_process_client_reaplist(struct list_head *reaplist)
6051 struct list_head *pos, *next;
6052 struct nfs4_client *clp;
6054 list_for_each_safe(pos, next, reaplist) {
6055 clp = list_entry(pos, struct nfs4_client, cl_lru);
6056 trace_nfsd_clid_purged(&clp->cl_clientid);
6057 list_del_init(&clp->cl_lru);
6063 nfs4_laundromat(struct nfsd_net *nn)
6065 struct nfs4_openowner *oo;
6066 struct nfs4_delegation *dp;
6067 struct nfs4_ol_stateid *stp;
6068 struct nfsd4_blocked_lock *nbl;
6069 struct list_head *pos, *next, reaplist;
6070 struct laundry_time lt = {
6071 .cutoff = ktime_get_boottime_seconds() - nn->nfsd4_lease,
6072 .new_timeo = nn->nfsd4_lease
6074 struct nfs4_cpntf_state *cps;
6075 copy_stateid_t *cps_t;
6078 if (clients_still_reclaiming(nn)) {
6082 nfsd4_end_grace(nn);
6084 spin_lock(&nn->s2s_cp_lock);
6085 idr_for_each_entry(&nn->s2s_cp_stateids, cps_t, i) {
6086 cps = container_of(cps_t, struct nfs4_cpntf_state, cp_stateid);
6087 if (cps->cp_stateid.cs_type == NFS4_COPYNOTIFY_STID &&
6088 state_expired(<, cps->cpntf_time))
6089 _free_cpntf_state_locked(nn, cps);
6091 spin_unlock(&nn->s2s_cp_lock);
6092 nfs4_get_client_reaplist(nn, &reaplist, <);
6093 nfs4_process_client_reaplist(&reaplist);
6095 spin_lock(&state_lock);
6096 list_for_each_safe(pos, next, &nn->del_recall_lru) {
6097 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
6098 if (!state_expired(<, dp->dl_time))
6100 WARN_ON(!unhash_delegation_locked(dp));
6101 list_add(&dp->dl_recall_lru, &reaplist);
6103 spin_unlock(&state_lock);
6104 while (!list_empty(&reaplist)) {
6105 dp = list_first_entry(&reaplist, struct nfs4_delegation,
6107 list_del_init(&dp->dl_recall_lru);
6108 revoke_delegation(dp);
6111 spin_lock(&nn->client_lock);
6112 while (!list_empty(&nn->close_lru)) {
6113 oo = list_first_entry(&nn->close_lru, struct nfs4_openowner,
6115 if (!state_expired(<, oo->oo_time))
6117 list_del_init(&oo->oo_close_lru);
6118 stp = oo->oo_last_closed_stid;
6119 oo->oo_last_closed_stid = NULL;
6120 spin_unlock(&nn->client_lock);
6121 nfs4_put_stid(&stp->st_stid);
6122 spin_lock(&nn->client_lock);
6124 spin_unlock(&nn->client_lock);
6127 * It's possible for a client to try and acquire an already held lock
6128 * that is being held for a long time, and then lose interest in it.
6129 * So, we clean out any un-revisited request after a lease period
6130 * under the assumption that the client is no longer interested.
6132 * RFC5661, sec. 9.6 states that the client must not rely on getting
6133 * notifications and must continue to poll for locks, even when the
6134 * server supports them. Thus this shouldn't lead to clients blocking
6135 * indefinitely once the lock does become free.
6137 BUG_ON(!list_empty(&reaplist));
6138 spin_lock(&nn->blocked_locks_lock);
6139 while (!list_empty(&nn->blocked_locks_lru)) {
6140 nbl = list_first_entry(&nn->blocked_locks_lru,
6141 struct nfsd4_blocked_lock, nbl_lru);
6142 if (!state_expired(<, nbl->nbl_time))
6144 list_move(&nbl->nbl_lru, &reaplist);
6145 list_del_init(&nbl->nbl_list);
6147 spin_unlock(&nn->blocked_locks_lock);
6149 while (!list_empty(&reaplist)) {
6150 nbl = list_first_entry(&reaplist,
6151 struct nfsd4_blocked_lock, nbl_lru);
6152 list_del_init(&nbl->nbl_lru);
6153 free_blocked_lock(nbl);
6155 #ifdef CONFIG_NFSD_V4_2_INTER_SSC
6156 /* service the server-to-server copy delayed unmount list */
6157 nfsd4_ssc_expire_umount(nn);
6160 return max_t(time64_t, lt.new_timeo, NFSD_LAUNDROMAT_MINTIMEOUT);
6163 static void laundromat_main(struct work_struct *);
6166 laundromat_main(struct work_struct *laundry)
6169 struct delayed_work *dwork = to_delayed_work(laundry);
6170 struct nfsd_net *nn = container_of(dwork, struct nfsd_net,
6173 t = nfs4_laundromat(nn);
6174 queue_delayed_work(laundry_wq, &nn->laundromat_work, t*HZ);
6178 courtesy_client_reaper(struct nfsd_net *nn)
6180 struct list_head reaplist;
6182 nfs4_get_courtesy_client_reaplist(nn, &reaplist);
6183 nfs4_process_client_reaplist(&reaplist);
6187 deleg_reaper(struct nfsd_net *nn)
6189 struct list_head *pos, *next;
6190 struct nfs4_client *clp;
6191 struct list_head cblist;
6193 INIT_LIST_HEAD(&cblist);
6194 spin_lock(&nn->client_lock);
6195 list_for_each_safe(pos, next, &nn->client_lru) {
6196 clp = list_entry(pos, struct nfs4_client, cl_lru);
6197 if (clp->cl_state != NFSD4_ACTIVE ||
6198 list_empty(&clp->cl_delegations) ||
6199 atomic_read(&clp->cl_delegs_in_recall) ||
6200 test_bit(NFSD4_CLIENT_CB_RECALL_ANY, &clp->cl_flags) ||
6201 (ktime_get_boottime_seconds() -
6202 clp->cl_ra_time < 5)) {
6205 list_add(&clp->cl_ra_cblist, &cblist);
6207 /* release in nfsd4_cb_recall_any_release */
6208 atomic_inc(&clp->cl_rpc_users);
6209 set_bit(NFSD4_CLIENT_CB_RECALL_ANY, &clp->cl_flags);
6210 clp->cl_ra_time = ktime_get_boottime_seconds();
6212 spin_unlock(&nn->client_lock);
6214 while (!list_empty(&cblist)) {
6215 clp = list_first_entry(&cblist, struct nfs4_client,
6217 list_del_init(&clp->cl_ra_cblist);
6218 clp->cl_ra->ra_keep = 0;
6219 clp->cl_ra->ra_bmval[0] = BIT(RCA4_TYPE_MASK_RDATA_DLG);
6220 trace_nfsd_cb_recall_any(clp->cl_ra);
6221 nfsd4_run_cb(&clp->cl_ra->ra_cb);
6226 nfsd4_state_shrinker_worker(struct work_struct *work)
6228 struct nfsd_net *nn = container_of(work, struct nfsd_net,
6229 nfsd_shrinker_work);
6231 courtesy_client_reaper(nn);
6235 static inline __be32 nfs4_check_fh(struct svc_fh *fhp, struct nfs4_stid *stp)
6237 if (!fh_match(&fhp->fh_handle, &stp->sc_file->fi_fhandle))
6238 return nfserr_bad_stateid;
6243 __be32 nfs4_check_openmode(struct nfs4_ol_stateid *stp, int flags)
6245 __be32 status = nfserr_openmode;
6247 /* For lock stateid's, we test the parent open, not the lock: */
6248 if (stp->st_openstp)
6249 stp = stp->st_openstp;
6250 if ((flags & WR_STATE) && !access_permit_write(stp))
6252 if ((flags & RD_STATE) && !access_permit_read(stp))
6259 static inline __be32
6260 check_special_stateids(struct net *net, svc_fh *current_fh, stateid_t *stateid, int flags)
6262 if (ONE_STATEID(stateid) && (flags & RD_STATE))
6264 else if (opens_in_grace(net)) {
6265 /* Answer in remaining cases depends on existence of
6266 * conflicting state; so we must wait out the grace period. */
6267 return nfserr_grace;
6268 } else if (flags & WR_STATE)
6269 return nfs4_share_conflict(current_fh,
6270 NFS4_SHARE_DENY_WRITE);
6271 else /* (flags & RD_STATE) && ZERO_STATEID(stateid) */
6272 return nfs4_share_conflict(current_fh,
6273 NFS4_SHARE_DENY_READ);
6276 static __be32 check_stateid_generation(stateid_t *in, stateid_t *ref, bool has_session)
6279 * When sessions are used the stateid generation number is ignored
6282 if (has_session && in->si_generation == 0)
6285 if (in->si_generation == ref->si_generation)
6288 /* If the client sends us a stateid from the future, it's buggy: */
6289 if (nfsd4_stateid_generation_after(in, ref))
6290 return nfserr_bad_stateid;
6292 * However, we could see a stateid from the past, even from a
6293 * non-buggy client. For example, if the client sends a lock
6294 * while some IO is outstanding, the lock may bump si_generation
6295 * while the IO is still in flight. The client could avoid that
6296 * situation by waiting for responses on all the IO requests,
6297 * but better performance may result in retrying IO that
6298 * receives an old_stateid error if requests are rarely
6299 * reordered in flight:
6301 return nfserr_old_stateid;
6304 static __be32 nfsd4_stid_check_stateid_generation(stateid_t *in, struct nfs4_stid *s, bool has_session)
6308 spin_lock(&s->sc_lock);
6309 ret = nfsd4_verify_open_stid(s);
6311 ret = check_stateid_generation(in, &s->sc_stateid, has_session);
6312 spin_unlock(&s->sc_lock);
6316 static __be32 nfsd4_check_openowner_confirmed(struct nfs4_ol_stateid *ols)
6318 if (ols->st_stateowner->so_is_open_owner &&
6319 !(openowner(ols->st_stateowner)->oo_flags & NFS4_OO_CONFIRMED))
6320 return nfserr_bad_stateid;
6324 static __be32 nfsd4_validate_stateid(struct nfs4_client *cl, stateid_t *stateid)
6326 struct nfs4_stid *s;
6327 __be32 status = nfserr_bad_stateid;
6329 if (ZERO_STATEID(stateid) || ONE_STATEID(stateid) ||
6330 CLOSE_STATEID(stateid))
6332 if (!same_clid(&stateid->si_opaque.so_clid, &cl->cl_clientid))
6334 spin_lock(&cl->cl_lock);
6335 s = find_stateid_locked(cl, stateid);
6338 status = nfsd4_stid_check_stateid_generation(stateid, s, 1);
6341 switch (s->sc_type) {
6342 case NFS4_DELEG_STID:
6345 case NFS4_REVOKED_DELEG_STID:
6346 status = nfserr_deleg_revoked;
6348 case NFS4_OPEN_STID:
6349 case NFS4_LOCK_STID:
6350 status = nfsd4_check_openowner_confirmed(openlockstateid(s));
6353 printk("unknown stateid type %x\n", s->sc_type);
6355 case NFS4_CLOSED_STID:
6356 case NFS4_CLOSED_DELEG_STID:
6357 status = nfserr_bad_stateid;
6360 spin_unlock(&cl->cl_lock);
6365 nfsd4_lookup_stateid(struct nfsd4_compound_state *cstate,
6366 stateid_t *stateid, unsigned char typemask,
6367 struct nfs4_stid **s, struct nfsd_net *nn)
6370 struct nfs4_stid *stid;
6371 bool return_revoked = false;
6374 * only return revoked delegations if explicitly asked.
6375 * otherwise we report revoked or bad_stateid status.
6377 if (typemask & NFS4_REVOKED_DELEG_STID)
6378 return_revoked = true;
6379 else if (typemask & NFS4_DELEG_STID)
6380 typemask |= NFS4_REVOKED_DELEG_STID;
6382 if (ZERO_STATEID(stateid) || ONE_STATEID(stateid) ||
6383 CLOSE_STATEID(stateid))
6384 return nfserr_bad_stateid;
6385 status = set_client(&stateid->si_opaque.so_clid, cstate, nn);
6386 if (status == nfserr_stale_clientid) {
6387 if (cstate->session)
6388 return nfserr_bad_stateid;
6389 return nfserr_stale_stateid;
6393 stid = find_stateid_by_type(cstate->clp, stateid, typemask);
6395 return nfserr_bad_stateid;
6396 if ((stid->sc_type == NFS4_REVOKED_DELEG_STID) && !return_revoked) {
6397 nfs4_put_stid(stid);
6398 if (cstate->minorversion)
6399 return nfserr_deleg_revoked;
6400 return nfserr_bad_stateid;
6406 static struct nfsd_file *
6407 nfs4_find_file(struct nfs4_stid *s, int flags)
6412 switch (s->sc_type) {
6413 case NFS4_DELEG_STID:
6414 if (WARN_ON_ONCE(!s->sc_file->fi_deleg_file))
6416 return nfsd_file_get(s->sc_file->fi_deleg_file);
6417 case NFS4_OPEN_STID:
6418 case NFS4_LOCK_STID:
6419 if (flags & RD_STATE)
6420 return find_readable_file(s->sc_file);
6422 return find_writeable_file(s->sc_file);
6429 nfs4_check_olstateid(struct nfs4_ol_stateid *ols, int flags)
6433 status = nfsd4_check_openowner_confirmed(ols);
6436 return nfs4_check_openmode(ols, flags);
6440 nfs4_check_file(struct svc_rqst *rqstp, struct svc_fh *fhp, struct nfs4_stid *s,
6441 struct nfsd_file **nfp, int flags)
6443 int acc = (flags & RD_STATE) ? NFSD_MAY_READ : NFSD_MAY_WRITE;
6444 struct nfsd_file *nf;
6447 nf = nfs4_find_file(s, flags);
6449 status = nfsd_permission(rqstp, fhp->fh_export, fhp->fh_dentry,
6450 acc | NFSD_MAY_OWNER_OVERRIDE);
6456 status = nfsd_file_acquire(rqstp, fhp, acc, &nf);
6465 _free_cpntf_state_locked(struct nfsd_net *nn, struct nfs4_cpntf_state *cps)
6467 WARN_ON_ONCE(cps->cp_stateid.cs_type != NFS4_COPYNOTIFY_STID);
6468 if (!refcount_dec_and_test(&cps->cp_stateid.cs_count))
6470 list_del(&cps->cp_list);
6471 idr_remove(&nn->s2s_cp_stateids,
6472 cps->cp_stateid.cs_stid.si_opaque.so_id);
6476 * A READ from an inter server to server COPY will have a
6477 * copy stateid. Look up the copy notify stateid from the
6478 * idr structure and take a reference on it.
6480 __be32 manage_cpntf_state(struct nfsd_net *nn, stateid_t *st,
6481 struct nfs4_client *clp,
6482 struct nfs4_cpntf_state **cps)
6484 copy_stateid_t *cps_t;
6485 struct nfs4_cpntf_state *state = NULL;
6487 if (st->si_opaque.so_clid.cl_id != nn->s2s_cp_cl_id)
6488 return nfserr_bad_stateid;
6489 spin_lock(&nn->s2s_cp_lock);
6490 cps_t = idr_find(&nn->s2s_cp_stateids, st->si_opaque.so_id);
6492 state = container_of(cps_t, struct nfs4_cpntf_state,
6494 if (state->cp_stateid.cs_type != NFS4_COPYNOTIFY_STID) {
6499 refcount_inc(&state->cp_stateid.cs_count);
6501 _free_cpntf_state_locked(nn, state);
6504 spin_unlock(&nn->s2s_cp_lock);
6506 return nfserr_bad_stateid;
6512 static __be32 find_cpntf_state(struct nfsd_net *nn, stateid_t *st,
6513 struct nfs4_stid **stid)
6516 struct nfs4_cpntf_state *cps = NULL;
6517 struct nfs4_client *found;
6519 status = manage_cpntf_state(nn, st, NULL, &cps);
6523 cps->cpntf_time = ktime_get_boottime_seconds();
6525 status = nfserr_expired;
6526 found = lookup_clientid(&cps->cp_p_clid, true, nn);
6530 *stid = find_stateid_by_type(found, &cps->cp_p_stateid,
6531 NFS4_DELEG_STID|NFS4_OPEN_STID|NFS4_LOCK_STID);
6535 status = nfserr_bad_stateid;
6537 put_client_renew(found);
6539 nfs4_put_cpntf_state(nn, cps);
6543 void nfs4_put_cpntf_state(struct nfsd_net *nn, struct nfs4_cpntf_state *cps)
6545 spin_lock(&nn->s2s_cp_lock);
6546 _free_cpntf_state_locked(nn, cps);
6547 spin_unlock(&nn->s2s_cp_lock);
6551 * Checks for stateid operations
6554 nfs4_preprocess_stateid_op(struct svc_rqst *rqstp,
6555 struct nfsd4_compound_state *cstate, struct svc_fh *fhp,
6556 stateid_t *stateid, int flags, struct nfsd_file **nfp,
6557 struct nfs4_stid **cstid)
6559 struct net *net = SVC_NET(rqstp);
6560 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
6561 struct nfs4_stid *s = NULL;
6567 if (ZERO_STATEID(stateid) || ONE_STATEID(stateid)) {
6569 status = nfserr_bad_stateid;
6571 status = check_special_stateids(net, fhp, stateid,
6576 status = nfsd4_lookup_stateid(cstate, stateid,
6577 NFS4_DELEG_STID|NFS4_OPEN_STID|NFS4_LOCK_STID,
6579 if (status == nfserr_bad_stateid)
6580 status = find_cpntf_state(nn, stateid, &s);
6583 status = nfsd4_stid_check_stateid_generation(stateid, s,
6584 nfsd4_has_session(cstate));
6588 switch (s->sc_type) {
6589 case NFS4_DELEG_STID:
6590 status = nfs4_check_delegmode(delegstateid(s), flags);
6592 case NFS4_OPEN_STID:
6593 case NFS4_LOCK_STID:
6594 status = nfs4_check_olstateid(openlockstateid(s), flags);
6597 status = nfserr_bad_stateid;
6602 status = nfs4_check_fh(fhp, s);
6605 if (status == nfs_ok && nfp)
6606 status = nfs4_check_file(rqstp, fhp, s, nfp, flags);
6609 if (!status && cstid)
6618 * Test if the stateid is valid
6621 nfsd4_test_stateid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
6622 union nfsd4_op_u *u)
6624 struct nfsd4_test_stateid *test_stateid = &u->test_stateid;
6625 struct nfsd4_test_stateid_id *stateid;
6626 struct nfs4_client *cl = cstate->clp;
6628 list_for_each_entry(stateid, &test_stateid->ts_stateid_list, ts_id_list)
6629 stateid->ts_id_status =
6630 nfsd4_validate_stateid(cl, &stateid->ts_id_stateid);
6636 nfsd4_free_lock_stateid(stateid_t *stateid, struct nfs4_stid *s)
6638 struct nfs4_ol_stateid *stp = openlockstateid(s);
6641 ret = nfsd4_lock_ol_stateid(stp);
6645 ret = check_stateid_generation(stateid, &s->sc_stateid, 1);
6649 ret = nfserr_locks_held;
6650 if (check_for_locks(stp->st_stid.sc_file,
6651 lockowner(stp->st_stateowner)))
6654 release_lock_stateid(stp);
6658 mutex_unlock(&stp->st_mutex);
6665 nfsd4_free_stateid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
6666 union nfsd4_op_u *u)
6668 struct nfsd4_free_stateid *free_stateid = &u->free_stateid;
6669 stateid_t *stateid = &free_stateid->fr_stateid;
6670 struct nfs4_stid *s;
6671 struct nfs4_delegation *dp;
6672 struct nfs4_client *cl = cstate->clp;
6673 __be32 ret = nfserr_bad_stateid;
6675 spin_lock(&cl->cl_lock);
6676 s = find_stateid_locked(cl, stateid);
6679 spin_lock(&s->sc_lock);
6680 switch (s->sc_type) {
6681 case NFS4_DELEG_STID:
6682 ret = nfserr_locks_held;
6684 case NFS4_OPEN_STID:
6685 ret = check_stateid_generation(stateid, &s->sc_stateid, 1);
6688 ret = nfserr_locks_held;
6690 case NFS4_LOCK_STID:
6691 spin_unlock(&s->sc_lock);
6692 refcount_inc(&s->sc_count);
6693 spin_unlock(&cl->cl_lock);
6694 ret = nfsd4_free_lock_stateid(stateid, s);
6696 case NFS4_REVOKED_DELEG_STID:
6697 spin_unlock(&s->sc_lock);
6698 dp = delegstateid(s);
6699 list_del_init(&dp->dl_recall_lru);
6700 spin_unlock(&cl->cl_lock);
6704 /* Default falls through and returns nfserr_bad_stateid */
6706 spin_unlock(&s->sc_lock);
6708 spin_unlock(&cl->cl_lock);
6716 return (type == NFS4_READW_LT || type == NFS4_READ_LT) ?
6717 RD_STATE : WR_STATE;
6720 static __be32 nfs4_seqid_op_checks(struct nfsd4_compound_state *cstate, stateid_t *stateid, u32 seqid, struct nfs4_ol_stateid *stp)
6722 struct svc_fh *current_fh = &cstate->current_fh;
6723 struct nfs4_stateowner *sop = stp->st_stateowner;
6726 status = nfsd4_check_seqid(cstate, sop, seqid);
6729 status = nfsd4_lock_ol_stateid(stp);
6730 if (status != nfs_ok)
6732 status = check_stateid_generation(stateid, &stp->st_stid.sc_stateid, nfsd4_has_session(cstate));
6733 if (status == nfs_ok)
6734 status = nfs4_check_fh(current_fh, &stp->st_stid);
6735 if (status != nfs_ok)
6736 mutex_unlock(&stp->st_mutex);
6741 * Checks for sequence id mutating operations.
6744 nfs4_preprocess_seqid_op(struct nfsd4_compound_state *cstate, u32 seqid,
6745 stateid_t *stateid, char typemask,
6746 struct nfs4_ol_stateid **stpp,
6747 struct nfsd_net *nn)
6750 struct nfs4_stid *s;
6751 struct nfs4_ol_stateid *stp = NULL;
6753 trace_nfsd_preprocess(seqid, stateid);
6756 status = nfsd4_lookup_stateid(cstate, stateid, typemask, &s, nn);
6759 stp = openlockstateid(s);
6760 nfsd4_cstate_assign_replay(cstate, stp->st_stateowner);
6762 status = nfs4_seqid_op_checks(cstate, stateid, seqid, stp);
6766 nfs4_put_stid(&stp->st_stid);
6770 static __be32 nfs4_preprocess_confirmed_seqid_op(struct nfsd4_compound_state *cstate, u32 seqid,
6771 stateid_t *stateid, struct nfs4_ol_stateid **stpp, struct nfsd_net *nn)
6774 struct nfs4_openowner *oo;
6775 struct nfs4_ol_stateid *stp;
6777 status = nfs4_preprocess_seqid_op(cstate, seqid, stateid,
6778 NFS4_OPEN_STID, &stp, nn);
6781 oo = openowner(stp->st_stateowner);
6782 if (!(oo->oo_flags & NFS4_OO_CONFIRMED)) {
6783 mutex_unlock(&stp->st_mutex);
6784 nfs4_put_stid(&stp->st_stid);
6785 return nfserr_bad_stateid;
6792 nfsd4_open_confirm(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
6793 union nfsd4_op_u *u)
6795 struct nfsd4_open_confirm *oc = &u->open_confirm;
6797 struct nfs4_openowner *oo;
6798 struct nfs4_ol_stateid *stp;
6799 struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
6801 dprintk("NFSD: nfsd4_open_confirm on file %pd\n",
6802 cstate->current_fh.fh_dentry);
6804 status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0);
6808 status = nfs4_preprocess_seqid_op(cstate,
6809 oc->oc_seqid, &oc->oc_req_stateid,
6810 NFS4_OPEN_STID, &stp, nn);
6813 oo = openowner(stp->st_stateowner);
6814 status = nfserr_bad_stateid;
6815 if (oo->oo_flags & NFS4_OO_CONFIRMED) {
6816 mutex_unlock(&stp->st_mutex);
6819 oo->oo_flags |= NFS4_OO_CONFIRMED;
6820 nfs4_inc_and_copy_stateid(&oc->oc_resp_stateid, &stp->st_stid);
6821 mutex_unlock(&stp->st_mutex);
6822 trace_nfsd_open_confirm(oc->oc_seqid, &stp->st_stid.sc_stateid);
6823 nfsd4_client_record_create(oo->oo_owner.so_client);
6826 nfs4_put_stid(&stp->st_stid);
6828 nfsd4_bump_seqid(cstate, status);
6832 static inline void nfs4_stateid_downgrade_bit(struct nfs4_ol_stateid *stp, u32 access)
6834 if (!test_access(access, stp))
6836 nfs4_file_put_access(stp->st_stid.sc_file, access);
6837 clear_access(access, stp);
6840 static inline void nfs4_stateid_downgrade(struct nfs4_ol_stateid *stp, u32 to_access)
6842 switch (to_access) {
6843 case NFS4_SHARE_ACCESS_READ:
6844 nfs4_stateid_downgrade_bit(stp, NFS4_SHARE_ACCESS_WRITE);
6845 nfs4_stateid_downgrade_bit(stp, NFS4_SHARE_ACCESS_BOTH);
6847 case NFS4_SHARE_ACCESS_WRITE:
6848 nfs4_stateid_downgrade_bit(stp, NFS4_SHARE_ACCESS_READ);
6849 nfs4_stateid_downgrade_bit(stp, NFS4_SHARE_ACCESS_BOTH);
6851 case NFS4_SHARE_ACCESS_BOTH:
6859 nfsd4_open_downgrade(struct svc_rqst *rqstp,
6860 struct nfsd4_compound_state *cstate, union nfsd4_op_u *u)
6862 struct nfsd4_open_downgrade *od = &u->open_downgrade;
6864 struct nfs4_ol_stateid *stp;
6865 struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
6867 dprintk("NFSD: nfsd4_open_downgrade on file %pd\n",
6868 cstate->current_fh.fh_dentry);
6870 /* We don't yet support WANT bits: */
6871 if (od->od_deleg_want)
6872 dprintk("NFSD: %s: od_deleg_want=0x%x ignored\n", __func__,
6875 status = nfs4_preprocess_confirmed_seqid_op(cstate, od->od_seqid,
6876 &od->od_stateid, &stp, nn);
6879 status = nfserr_inval;
6880 if (!test_access(od->od_share_access, stp)) {
6881 dprintk("NFSD: access not a subset of current bitmap: 0x%hhx, input access=%08x\n",
6882 stp->st_access_bmap, od->od_share_access);
6885 if (!test_deny(od->od_share_deny, stp)) {
6886 dprintk("NFSD: deny not a subset of current bitmap: 0x%hhx, input deny=%08x\n",
6887 stp->st_deny_bmap, od->od_share_deny);
6890 nfs4_stateid_downgrade(stp, od->od_share_access);
6891 reset_union_bmap_deny(od->od_share_deny, stp);
6892 nfs4_inc_and_copy_stateid(&od->od_stateid, &stp->st_stid);
6895 mutex_unlock(&stp->st_mutex);
6896 nfs4_put_stid(&stp->st_stid);
6898 nfsd4_bump_seqid(cstate, status);
6902 static void nfsd4_close_open_stateid(struct nfs4_ol_stateid *s)
6904 struct nfs4_client *clp = s->st_stid.sc_client;
6906 LIST_HEAD(reaplist);
6907 struct nfs4_ol_stateid *stp;
6909 spin_lock(&clp->cl_lock);
6910 unhashed = unhash_open_stateid(s, &reaplist);
6912 if (clp->cl_minorversion) {
6914 put_ol_stateid_locked(s, &reaplist);
6915 spin_unlock(&clp->cl_lock);
6916 list_for_each_entry(stp, &reaplist, st_locks)
6917 nfs4_free_cpntf_statelist(clp->net, &stp->st_stid);
6918 free_ol_stateid_reaplist(&reaplist);
6920 spin_unlock(&clp->cl_lock);
6921 free_ol_stateid_reaplist(&reaplist);
6923 move_to_close_lru(s, clp->net);
6928 * nfs4_unlock_state() called after encode
6931 nfsd4_close(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
6932 union nfsd4_op_u *u)
6934 struct nfsd4_close *close = &u->close;
6936 struct nfs4_ol_stateid *stp;
6937 struct net *net = SVC_NET(rqstp);
6938 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
6940 dprintk("NFSD: nfsd4_close on file %pd\n",
6941 cstate->current_fh.fh_dentry);
6943 status = nfs4_preprocess_seqid_op(cstate, close->cl_seqid,
6945 NFS4_OPEN_STID|NFS4_CLOSED_STID,
6947 nfsd4_bump_seqid(cstate, status);
6951 stp->st_stid.sc_type = NFS4_CLOSED_STID;
6954 * Technically we don't _really_ have to increment or copy it, since
6955 * it should just be gone after this operation and we clobber the
6956 * copied value below, but we continue to do so here just to ensure
6957 * that racing ops see that there was a state change.
6959 nfs4_inc_and_copy_stateid(&close->cl_stateid, &stp->st_stid);
6961 nfsd4_close_open_stateid(stp);
6962 mutex_unlock(&stp->st_mutex);
6964 /* v4.1+ suggests that we send a special stateid in here, since the
6965 * clients should just ignore this anyway. Since this is not useful
6966 * for v4.0 clients either, we set it to the special close_stateid
6969 * See RFC5661 section 18.2.4, and RFC7530 section 16.2.5
6971 memcpy(&close->cl_stateid, &close_stateid, sizeof(close->cl_stateid));
6973 /* put reference from nfs4_preprocess_seqid_op */
6974 nfs4_put_stid(&stp->st_stid);
6980 nfsd4_delegreturn(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
6981 union nfsd4_op_u *u)
6983 struct nfsd4_delegreturn *dr = &u->delegreturn;
6984 struct nfs4_delegation *dp;
6985 stateid_t *stateid = &dr->dr_stateid;
6986 struct nfs4_stid *s;
6988 struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
6990 if ((status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0)))
6993 status = nfsd4_lookup_stateid(cstate, stateid, NFS4_DELEG_STID, &s, nn);
6996 dp = delegstateid(s);
6997 status = nfsd4_stid_check_stateid_generation(stateid, &dp->dl_stid, nfsd4_has_session(cstate));
7001 trace_nfsd_deleg_return(stateid);
7002 wake_up_var(d_inode(cstate->current_fh.fh_dentry));
7003 destroy_delegation(dp);
7005 nfs4_put_stid(&dp->dl_stid);
7010 /* last octet in a range */
7012 last_byte_offset(u64 start, u64 len)
7018 return end > start ? end - 1: NFS4_MAX_UINT64;
7022 * TODO: Linux file offsets are _signed_ 64-bit quantities, which means that
7023 * we can't properly handle lock requests that go beyond the (2^63 - 1)-th
7024 * byte, because of sign extension problems. Since NFSv4 calls for 64-bit
7025 * locking, this prevents us from being completely protocol-compliant. The
7026 * real solution to this problem is to start using unsigned file offsets in
7027 * the VFS, but this is a very deep change!
7030 nfs4_transform_lock_offset(struct file_lock *lock)
7032 if (lock->fl_start < 0)
7033 lock->fl_start = OFFSET_MAX;
7034 if (lock->fl_end < 0)
7035 lock->fl_end = OFFSET_MAX;
7039 nfsd4_lm_get_owner(fl_owner_t owner)
7041 struct nfs4_lockowner *lo = (struct nfs4_lockowner *)owner;
7043 nfs4_get_stateowner(&lo->lo_owner);
7048 nfsd4_lm_put_owner(fl_owner_t owner)
7050 struct nfs4_lockowner *lo = (struct nfs4_lockowner *)owner;
7053 nfs4_put_stateowner(&lo->lo_owner);
7056 /* return pointer to struct nfs4_client if client is expirable */
7058 nfsd4_lm_lock_expirable(struct file_lock *cfl)
7060 struct nfs4_lockowner *lo = (struct nfs4_lockowner *)cfl->fl_owner;
7061 struct nfs4_client *clp = lo->lo_owner.so_client;
7062 struct nfsd_net *nn;
7064 if (try_to_expire_client(clp)) {
7065 nn = net_generic(clp->net, nfsd_net_id);
7066 mod_delayed_work(laundry_wq, &nn->laundromat_work, 0);
7072 /* schedule laundromat to run immediately and wait for it to complete */
7074 nfsd4_lm_expire_lock(void)
7076 flush_workqueue(laundry_wq);
7080 nfsd4_lm_notify(struct file_lock *fl)
7082 struct nfs4_lockowner *lo = (struct nfs4_lockowner *)fl->fl_owner;
7083 struct net *net = lo->lo_owner.so_client->net;
7084 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
7085 struct nfsd4_blocked_lock *nbl = container_of(fl,
7086 struct nfsd4_blocked_lock, nbl_lock);
7089 /* An empty list means that something else is going to be using it */
7090 spin_lock(&nn->blocked_locks_lock);
7091 if (!list_empty(&nbl->nbl_list)) {
7092 list_del_init(&nbl->nbl_list);
7093 list_del_init(&nbl->nbl_lru);
7096 spin_unlock(&nn->blocked_locks_lock);
7099 trace_nfsd_cb_notify_lock(lo, nbl);
7100 nfsd4_run_cb(&nbl->nbl_cb);
7104 static const struct lock_manager_operations nfsd_posix_mng_ops = {
7105 .lm_mod_owner = THIS_MODULE,
7106 .lm_notify = nfsd4_lm_notify,
7107 .lm_get_owner = nfsd4_lm_get_owner,
7108 .lm_put_owner = nfsd4_lm_put_owner,
7109 .lm_lock_expirable = nfsd4_lm_lock_expirable,
7110 .lm_expire_lock = nfsd4_lm_expire_lock,
7114 nfs4_set_lock_denied(struct file_lock *fl, struct nfsd4_lock_denied *deny)
7116 struct nfs4_lockowner *lo;
7118 if (fl->fl_lmops == &nfsd_posix_mng_ops) {
7119 lo = (struct nfs4_lockowner *) fl->fl_owner;
7120 xdr_netobj_dup(&deny->ld_owner, &lo->lo_owner.so_owner,
7122 if (!deny->ld_owner.data)
7123 /* We just don't care that much */
7125 deny->ld_clientid = lo->lo_owner.so_client->cl_clientid;
7128 deny->ld_owner.len = 0;
7129 deny->ld_owner.data = NULL;
7130 deny->ld_clientid.cl_boot = 0;
7131 deny->ld_clientid.cl_id = 0;
7133 deny->ld_start = fl->fl_start;
7134 deny->ld_length = NFS4_MAX_UINT64;
7135 if (fl->fl_end != NFS4_MAX_UINT64)
7136 deny->ld_length = fl->fl_end - fl->fl_start + 1;
7137 deny->ld_type = NFS4_READ_LT;
7138 if (fl->fl_type != F_RDLCK)
7139 deny->ld_type = NFS4_WRITE_LT;
7142 static struct nfs4_lockowner *
7143 find_lockowner_str_locked(struct nfs4_client *clp, struct xdr_netobj *owner)
7145 unsigned int strhashval = ownerstr_hashval(owner);
7146 struct nfs4_stateowner *so;
7148 lockdep_assert_held(&clp->cl_lock);
7150 list_for_each_entry(so, &clp->cl_ownerstr_hashtbl[strhashval],
7152 if (so->so_is_open_owner)
7154 if (same_owner_str(so, owner))
7155 return lockowner(nfs4_get_stateowner(so));
7160 static struct nfs4_lockowner *
7161 find_lockowner_str(struct nfs4_client *clp, struct xdr_netobj *owner)
7163 struct nfs4_lockowner *lo;
7165 spin_lock(&clp->cl_lock);
7166 lo = find_lockowner_str_locked(clp, owner);
7167 spin_unlock(&clp->cl_lock);
7171 static void nfs4_unhash_lockowner(struct nfs4_stateowner *sop)
7173 unhash_lockowner_locked(lockowner(sop));
7176 static void nfs4_free_lockowner(struct nfs4_stateowner *sop)
7178 struct nfs4_lockowner *lo = lockowner(sop);
7180 kmem_cache_free(lockowner_slab, lo);
7183 static const struct nfs4_stateowner_operations lockowner_ops = {
7184 .so_unhash = nfs4_unhash_lockowner,
7185 .so_free = nfs4_free_lockowner,
7189 * Alloc a lock owner structure.
7190 * Called in nfsd4_lock - therefore, OPEN and OPEN_CONFIRM (if needed) has
7193 * strhashval = ownerstr_hashval
7195 static struct nfs4_lockowner *
7196 alloc_init_lock_stateowner(unsigned int strhashval, struct nfs4_client *clp,
7197 struct nfs4_ol_stateid *open_stp,
7198 struct nfsd4_lock *lock)
7200 struct nfs4_lockowner *lo, *ret;
7202 lo = alloc_stateowner(lockowner_slab, &lock->lk_new_owner, clp);
7205 INIT_LIST_HEAD(&lo->lo_blocked);
7206 INIT_LIST_HEAD(&lo->lo_owner.so_stateids);
7207 lo->lo_owner.so_is_open_owner = 0;
7208 lo->lo_owner.so_seqid = lock->lk_new_lock_seqid;
7209 lo->lo_owner.so_ops = &lockowner_ops;
7210 spin_lock(&clp->cl_lock);
7211 ret = find_lockowner_str_locked(clp, &lock->lk_new_owner);
7213 list_add(&lo->lo_owner.so_strhash,
7214 &clp->cl_ownerstr_hashtbl[strhashval]);
7217 nfs4_free_stateowner(&lo->lo_owner);
7219 spin_unlock(&clp->cl_lock);
7223 static struct nfs4_ol_stateid *
7224 find_lock_stateid(const struct nfs4_lockowner *lo,
7225 const struct nfs4_ol_stateid *ost)
7227 struct nfs4_ol_stateid *lst;
7229 lockdep_assert_held(&ost->st_stid.sc_client->cl_lock);
7231 /* If ost is not hashed, ost->st_locks will not be valid */
7232 if (!nfs4_ol_stateid_unhashed(ost))
7233 list_for_each_entry(lst, &ost->st_locks, st_locks) {
7234 if (lst->st_stateowner == &lo->lo_owner) {
7235 refcount_inc(&lst->st_stid.sc_count);
7242 static struct nfs4_ol_stateid *
7243 init_lock_stateid(struct nfs4_ol_stateid *stp, struct nfs4_lockowner *lo,
7244 struct nfs4_file *fp, struct inode *inode,
7245 struct nfs4_ol_stateid *open_stp)
7247 struct nfs4_client *clp = lo->lo_owner.so_client;
7248 struct nfs4_ol_stateid *retstp;
7250 mutex_init(&stp->st_mutex);
7251 mutex_lock_nested(&stp->st_mutex, OPEN_STATEID_MUTEX);
7253 spin_lock(&clp->cl_lock);
7254 if (nfs4_ol_stateid_unhashed(open_stp))
7256 retstp = find_lock_stateid(lo, open_stp);
7259 refcount_inc(&stp->st_stid.sc_count);
7260 stp->st_stid.sc_type = NFS4_LOCK_STID;
7261 stp->st_stateowner = nfs4_get_stateowner(&lo->lo_owner);
7263 stp->st_stid.sc_file = fp;
7264 stp->st_access_bmap = 0;
7265 stp->st_deny_bmap = open_stp->st_deny_bmap;
7266 stp->st_openstp = open_stp;
7267 spin_lock(&fp->fi_lock);
7268 list_add(&stp->st_locks, &open_stp->st_locks);
7269 list_add(&stp->st_perstateowner, &lo->lo_owner.so_stateids);
7270 list_add(&stp->st_perfile, &fp->fi_stateids);
7271 spin_unlock(&fp->fi_lock);
7272 spin_unlock(&clp->cl_lock);
7275 spin_unlock(&clp->cl_lock);
7276 if (nfsd4_lock_ol_stateid(retstp) != nfs_ok) {
7277 nfs4_put_stid(&retstp->st_stid);
7280 /* To keep mutex tracking happy */
7281 mutex_unlock(&stp->st_mutex);
7284 spin_unlock(&clp->cl_lock);
7285 mutex_unlock(&stp->st_mutex);
7289 static struct nfs4_ol_stateid *
7290 find_or_create_lock_stateid(struct nfs4_lockowner *lo, struct nfs4_file *fi,
7291 struct inode *inode, struct nfs4_ol_stateid *ost,
7294 struct nfs4_stid *ns = NULL;
7295 struct nfs4_ol_stateid *lst;
7296 struct nfs4_openowner *oo = openowner(ost->st_stateowner);
7297 struct nfs4_client *clp = oo->oo_owner.so_client;
7300 spin_lock(&clp->cl_lock);
7301 lst = find_lock_stateid(lo, ost);
7302 spin_unlock(&clp->cl_lock);
7304 if (nfsd4_lock_ol_stateid(lst) == nfs_ok)
7306 nfs4_put_stid(&lst->st_stid);
7308 ns = nfs4_alloc_stid(clp, stateid_slab, nfs4_free_lock_stateid);
7312 lst = init_lock_stateid(openlockstateid(ns), lo, fi, inode, ost);
7313 if (lst == openlockstateid(ns))
7322 check_lock_length(u64 offset, u64 length)
7324 return ((length == 0) || ((length != NFS4_MAX_UINT64) &&
7325 (length > ~offset)));
7328 static void get_lock_access(struct nfs4_ol_stateid *lock_stp, u32 access)
7330 struct nfs4_file *fp = lock_stp->st_stid.sc_file;
7332 lockdep_assert_held(&fp->fi_lock);
7334 if (test_access(access, lock_stp))
7336 __nfs4_file_get_access(fp, access);
7337 set_access(access, lock_stp);
7341 lookup_or_create_lock_state(struct nfsd4_compound_state *cstate,
7342 struct nfs4_ol_stateid *ost,
7343 struct nfsd4_lock *lock,
7344 struct nfs4_ol_stateid **plst, bool *new)
7347 struct nfs4_file *fi = ost->st_stid.sc_file;
7348 struct nfs4_openowner *oo = openowner(ost->st_stateowner);
7349 struct nfs4_client *cl = oo->oo_owner.so_client;
7350 struct inode *inode = d_inode(cstate->current_fh.fh_dentry);
7351 struct nfs4_lockowner *lo;
7352 struct nfs4_ol_stateid *lst;
7353 unsigned int strhashval;
7355 lo = find_lockowner_str(cl, &lock->lk_new_owner);
7357 strhashval = ownerstr_hashval(&lock->lk_new_owner);
7358 lo = alloc_init_lock_stateowner(strhashval, cl, ost, lock);
7360 return nfserr_jukebox;
7362 /* with an existing lockowner, seqids must be the same */
7363 status = nfserr_bad_seqid;
7364 if (!cstate->minorversion &&
7365 lock->lk_new_lock_seqid != lo->lo_owner.so_seqid)
7369 lst = find_or_create_lock_stateid(lo, fi, inode, ost, new);
7371 status = nfserr_jukebox;
7378 nfs4_put_stateowner(&lo->lo_owner);
7386 nfsd4_lock(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
7387 union nfsd4_op_u *u)
7389 struct nfsd4_lock *lock = &u->lock;
7390 struct nfs4_openowner *open_sop = NULL;
7391 struct nfs4_lockowner *lock_sop = NULL;
7392 struct nfs4_ol_stateid *lock_stp = NULL;
7393 struct nfs4_ol_stateid *open_stp = NULL;
7394 struct nfs4_file *fp;
7395 struct nfsd_file *nf = NULL;
7396 struct nfsd4_blocked_lock *nbl = NULL;
7397 struct file_lock *file_lock = NULL;
7398 struct file_lock *conflock = NULL;
7403 unsigned char fl_type;
7404 unsigned int fl_flags = FL_POSIX;
7405 struct net *net = SVC_NET(rqstp);
7406 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
7408 dprintk("NFSD: nfsd4_lock: start=%Ld length=%Ld\n",
7409 (long long) lock->lk_offset,
7410 (long long) lock->lk_length);
7412 if (check_lock_length(lock->lk_offset, lock->lk_length))
7413 return nfserr_inval;
7415 if ((status = fh_verify(rqstp, &cstate->current_fh,
7416 S_IFREG, NFSD_MAY_LOCK))) {
7417 dprintk("NFSD: nfsd4_lock: permission denied!\n");
7421 if (lock->lk_is_new) {
7422 if (nfsd4_has_session(cstate))
7423 /* See rfc 5661 18.10.3: given clientid is ignored: */
7424 memcpy(&lock->lk_new_clientid,
7425 &cstate->clp->cl_clientid,
7426 sizeof(clientid_t));
7428 /* validate and update open stateid and open seqid */
7429 status = nfs4_preprocess_confirmed_seqid_op(cstate,
7430 lock->lk_new_open_seqid,
7431 &lock->lk_new_open_stateid,
7435 mutex_unlock(&open_stp->st_mutex);
7436 open_sop = openowner(open_stp->st_stateowner);
7437 status = nfserr_bad_stateid;
7438 if (!same_clid(&open_sop->oo_owner.so_client->cl_clientid,
7439 &lock->lk_new_clientid))
7441 status = lookup_or_create_lock_state(cstate, open_stp, lock,
7444 status = nfs4_preprocess_seqid_op(cstate,
7445 lock->lk_old_lock_seqid,
7446 &lock->lk_old_lock_stateid,
7447 NFS4_LOCK_STID, &lock_stp, nn);
7451 lock_sop = lockowner(lock_stp->st_stateowner);
7453 lkflg = setlkflg(lock->lk_type);
7454 status = nfs4_check_openmode(lock_stp, lkflg);
7458 status = nfserr_grace;
7459 if (locks_in_grace(net) && !lock->lk_reclaim)
7461 status = nfserr_no_grace;
7462 if (!locks_in_grace(net) && lock->lk_reclaim)
7465 if (lock->lk_reclaim)
7466 fl_flags |= FL_RECLAIM;
7468 fp = lock_stp->st_stid.sc_file;
7469 switch (lock->lk_type) {
7471 if (nfsd4_has_session(cstate))
7472 fl_flags |= FL_SLEEP;
7475 spin_lock(&fp->fi_lock);
7476 nf = find_readable_file_locked(fp);
7478 get_lock_access(lock_stp, NFS4_SHARE_ACCESS_READ);
7479 spin_unlock(&fp->fi_lock);
7482 case NFS4_WRITEW_LT:
7483 if (nfsd4_has_session(cstate))
7484 fl_flags |= FL_SLEEP;
7487 spin_lock(&fp->fi_lock);
7488 nf = find_writeable_file_locked(fp);
7490 get_lock_access(lock_stp, NFS4_SHARE_ACCESS_WRITE);
7491 spin_unlock(&fp->fi_lock);
7495 status = nfserr_inval;
7500 status = nfserr_openmode;
7505 * Most filesystems with their own ->lock operations will block
7506 * the nfsd thread waiting to acquire the lock. That leads to
7507 * deadlocks (we don't want every nfsd thread tied up waiting
7508 * for file locks), so don't attempt blocking lock notifications
7509 * on those filesystems:
7511 if (nf->nf_file->f_op->lock)
7512 fl_flags &= ~FL_SLEEP;
7514 nbl = find_or_allocate_block(lock_sop, &fp->fi_fhandle, nn);
7516 dprintk("NFSD: %s: unable to allocate block!\n", __func__);
7517 status = nfserr_jukebox;
7521 file_lock = &nbl->nbl_lock;
7522 file_lock->fl_type = fl_type;
7523 file_lock->fl_owner = (fl_owner_t)lockowner(nfs4_get_stateowner(&lock_sop->lo_owner));
7524 file_lock->fl_pid = current->tgid;
7525 file_lock->fl_file = nf->nf_file;
7526 file_lock->fl_flags = fl_flags;
7527 file_lock->fl_lmops = &nfsd_posix_mng_ops;
7528 file_lock->fl_start = lock->lk_offset;
7529 file_lock->fl_end = last_byte_offset(lock->lk_offset, lock->lk_length);
7530 nfs4_transform_lock_offset(file_lock);
7532 conflock = locks_alloc_lock();
7534 dprintk("NFSD: %s: unable to allocate lock!\n", __func__);
7535 status = nfserr_jukebox;
7539 if (fl_flags & FL_SLEEP) {
7540 nbl->nbl_time = ktime_get_boottime_seconds();
7541 spin_lock(&nn->blocked_locks_lock);
7542 list_add_tail(&nbl->nbl_list, &lock_sop->lo_blocked);
7543 list_add_tail(&nbl->nbl_lru, &nn->blocked_locks_lru);
7544 kref_get(&nbl->nbl_kref);
7545 spin_unlock(&nn->blocked_locks_lock);
7548 err = vfs_lock_file(nf->nf_file, F_SETLK, file_lock, conflock);
7550 case 0: /* success! */
7551 nfs4_inc_and_copy_stateid(&lock->lk_resp_stateid, &lock_stp->st_stid);
7553 if (lock->lk_reclaim)
7554 nn->somebody_reclaimed = true;
7556 case FILE_LOCK_DEFERRED:
7557 kref_put(&nbl->nbl_kref, free_nbl);
7560 case -EAGAIN: /* conflock holds conflicting lock */
7561 status = nfserr_denied;
7562 dprintk("NFSD: nfsd4_lock: conflicting lock found!\n");
7563 nfs4_set_lock_denied(conflock, &lock->lk_denied);
7566 status = nfserr_deadlock;
7569 dprintk("NFSD: nfsd4_lock: vfs_lock_file() failed! status %d\n",err);
7570 status = nfserrno(err);
7575 /* dequeue it if we queued it before */
7576 if (fl_flags & FL_SLEEP) {
7577 spin_lock(&nn->blocked_locks_lock);
7578 if (!list_empty(&nbl->nbl_list) &&
7579 !list_empty(&nbl->nbl_lru)) {
7580 list_del_init(&nbl->nbl_list);
7581 list_del_init(&nbl->nbl_lru);
7582 kref_put(&nbl->nbl_kref, free_nbl);
7584 /* nbl can use one of lists to be linked to reaplist */
7585 spin_unlock(&nn->blocked_locks_lock);
7587 free_blocked_lock(nbl);
7592 /* Bump seqid manually if the 4.0 replay owner is openowner */
7593 if (cstate->replay_owner &&
7594 cstate->replay_owner != &lock_sop->lo_owner &&
7595 seqid_mutating_err(ntohl(status)))
7596 lock_sop->lo_owner.so_seqid++;
7599 * If this is a new, never-before-used stateid, and we are
7600 * returning an error, then just go ahead and release it.
7603 release_lock_stateid(lock_stp);
7605 mutex_unlock(&lock_stp->st_mutex);
7607 nfs4_put_stid(&lock_stp->st_stid);
7610 nfs4_put_stid(&open_stp->st_stid);
7611 nfsd4_bump_seqid(cstate, status);
7613 locks_free_lock(conflock);
7618 * The NFSv4 spec allows a client to do a LOCKT without holding an OPEN,
7619 * so we do a temporary open here just to get an open file to pass to
7622 static __be32 nfsd_test_lock(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file_lock *lock)
7624 struct nfsd_file *nf;
7625 struct inode *inode;
7628 err = nfsd_file_acquire(rqstp, fhp, NFSD_MAY_READ, &nf);
7631 inode = fhp->fh_dentry->d_inode;
7632 inode_lock(inode); /* to block new leases till after test_lock: */
7633 err = nfserrno(nfsd_open_break_lease(inode, NFSD_MAY_READ));
7636 lock->fl_file = nf->nf_file;
7637 err = nfserrno(vfs_test_lock(nf->nf_file, lock));
7638 lock->fl_file = NULL;
7640 inode_unlock(inode);
7649 nfsd4_lockt(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
7650 union nfsd4_op_u *u)
7652 struct nfsd4_lockt *lockt = &u->lockt;
7653 struct file_lock *file_lock = NULL;
7654 struct nfs4_lockowner *lo = NULL;
7656 struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
7658 if (locks_in_grace(SVC_NET(rqstp)))
7659 return nfserr_grace;
7661 if (check_lock_length(lockt->lt_offset, lockt->lt_length))
7662 return nfserr_inval;
7664 if (!nfsd4_has_session(cstate)) {
7665 status = set_client(&lockt->lt_clientid, cstate, nn);
7670 if ((status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0)))
7673 file_lock = locks_alloc_lock();
7675 dprintk("NFSD: %s: unable to allocate lock!\n", __func__);
7676 status = nfserr_jukebox;
7680 switch (lockt->lt_type) {
7683 file_lock->fl_type = F_RDLCK;
7686 case NFS4_WRITEW_LT:
7687 file_lock->fl_type = F_WRLCK;
7690 dprintk("NFSD: nfs4_lockt: bad lock type!\n");
7691 status = nfserr_inval;
7695 lo = find_lockowner_str(cstate->clp, &lockt->lt_owner);
7697 file_lock->fl_owner = (fl_owner_t)lo;
7698 file_lock->fl_pid = current->tgid;
7699 file_lock->fl_flags = FL_POSIX;
7701 file_lock->fl_start = lockt->lt_offset;
7702 file_lock->fl_end = last_byte_offset(lockt->lt_offset, lockt->lt_length);
7704 nfs4_transform_lock_offset(file_lock);
7706 status = nfsd_test_lock(rqstp, &cstate->current_fh, file_lock);
7710 if (file_lock->fl_type != F_UNLCK) {
7711 status = nfserr_denied;
7712 nfs4_set_lock_denied(file_lock, &lockt->lt_denied);
7716 nfs4_put_stateowner(&lo->lo_owner);
7718 locks_free_lock(file_lock);
7723 nfsd4_locku(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
7724 union nfsd4_op_u *u)
7726 struct nfsd4_locku *locku = &u->locku;
7727 struct nfs4_ol_stateid *stp;
7728 struct nfsd_file *nf = NULL;
7729 struct file_lock *file_lock = NULL;
7732 struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
7734 dprintk("NFSD: nfsd4_locku: start=%Ld length=%Ld\n",
7735 (long long) locku->lu_offset,
7736 (long long) locku->lu_length);
7738 if (check_lock_length(locku->lu_offset, locku->lu_length))
7739 return nfserr_inval;
7741 status = nfs4_preprocess_seqid_op(cstate, locku->lu_seqid,
7742 &locku->lu_stateid, NFS4_LOCK_STID,
7746 nf = find_any_file(stp->st_stid.sc_file);
7748 status = nfserr_lock_range;
7751 file_lock = locks_alloc_lock();
7753 dprintk("NFSD: %s: unable to allocate lock!\n", __func__);
7754 status = nfserr_jukebox;
7758 file_lock->fl_type = F_UNLCK;
7759 file_lock->fl_owner = (fl_owner_t)lockowner(nfs4_get_stateowner(stp->st_stateowner));
7760 file_lock->fl_pid = current->tgid;
7761 file_lock->fl_file = nf->nf_file;
7762 file_lock->fl_flags = FL_POSIX;
7763 file_lock->fl_lmops = &nfsd_posix_mng_ops;
7764 file_lock->fl_start = locku->lu_offset;
7766 file_lock->fl_end = last_byte_offset(locku->lu_offset,
7768 nfs4_transform_lock_offset(file_lock);
7770 err = vfs_lock_file(nf->nf_file, F_SETLK, file_lock, NULL);
7772 dprintk("NFSD: nfs4_locku: vfs_lock_file failed!\n");
7775 nfs4_inc_and_copy_stateid(&locku->lu_stateid, &stp->st_stid);
7779 mutex_unlock(&stp->st_mutex);
7780 nfs4_put_stid(&stp->st_stid);
7782 nfsd4_bump_seqid(cstate, status);
7784 locks_free_lock(file_lock);
7788 status = nfserrno(err);
7794 * true: locks held by lockowner
7795 * false: no locks held by lockowner
7798 check_for_locks(struct nfs4_file *fp, struct nfs4_lockowner *lowner)
7800 struct file_lock *fl;
7802 struct nfsd_file *nf = find_any_file(fp);
7803 struct inode *inode;
7804 struct file_lock_context *flctx;
7807 /* Any valid lock stateid should have some sort of access */
7812 inode = locks_inode(nf->nf_file);
7813 flctx = locks_inode_context(inode);
7815 if (flctx && !list_empty_careful(&flctx->flc_posix)) {
7816 spin_lock(&flctx->flc_lock);
7817 list_for_each_entry(fl, &flctx->flc_posix, fl_list) {
7818 if (fl->fl_owner == (fl_owner_t)lowner) {
7823 spin_unlock(&flctx->flc_lock);
7830 * nfsd4_release_lockowner - process NFSv4.0 RELEASE_LOCKOWNER operations
7831 * @rqstp: RPC transaction
7832 * @cstate: NFSv4 COMPOUND state
7833 * @u: RELEASE_LOCKOWNER arguments
7835 * The lockowner's so_count is bumped when a lock record is added
7836 * or when copying a conflicting lock. The latter case is brief,
7837 * but can lead to fleeting false positives when looking for
7841 * %nfs_ok: lockowner released or not found
7842 * %nfserr_locks_held: lockowner still in use
7843 * %nfserr_stale_clientid: clientid no longer active
7844 * %nfserr_expired: clientid not recognized
7847 nfsd4_release_lockowner(struct svc_rqst *rqstp,
7848 struct nfsd4_compound_state *cstate,
7849 union nfsd4_op_u *u)
7851 struct nfsd4_release_lockowner *rlockowner = &u->release_lockowner;
7852 struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
7853 clientid_t *clid = &rlockowner->rl_clientid;
7854 struct nfs4_ol_stateid *stp;
7855 struct nfs4_lockowner *lo;
7856 struct nfs4_client *clp;
7857 LIST_HEAD(reaplist);
7860 dprintk("nfsd4_release_lockowner clientid: (%08x/%08x):\n",
7861 clid->cl_boot, clid->cl_id);
7863 status = set_client(clid, cstate, nn);
7868 spin_lock(&clp->cl_lock);
7869 lo = find_lockowner_str_locked(clp, &rlockowner->rl_owner);
7871 spin_unlock(&clp->cl_lock);
7874 if (atomic_read(&lo->lo_owner.so_count) != 2) {
7875 spin_unlock(&clp->cl_lock);
7876 nfs4_put_stateowner(&lo->lo_owner);
7877 return nfserr_locks_held;
7879 unhash_lockowner_locked(lo);
7880 while (!list_empty(&lo->lo_owner.so_stateids)) {
7881 stp = list_first_entry(&lo->lo_owner.so_stateids,
7882 struct nfs4_ol_stateid,
7884 WARN_ON(!unhash_lock_stateid(stp));
7885 put_ol_stateid_locked(stp, &reaplist);
7887 spin_unlock(&clp->cl_lock);
7889 free_ol_stateid_reaplist(&reaplist);
7890 remove_blocked_locks(lo);
7891 nfs4_put_stateowner(&lo->lo_owner);
7895 static inline struct nfs4_client_reclaim *
7898 return kmalloc(sizeof(struct nfs4_client_reclaim), GFP_KERNEL);
7902 nfs4_has_reclaimed_state(struct xdr_netobj name, struct nfsd_net *nn)
7904 struct nfs4_client_reclaim *crp;
7906 crp = nfsd4_find_reclaim_client(name, nn);
7907 return (crp && crp->cr_clp);
7911 * failure => all reset bets are off, nfserr_no_grace...
7913 * The caller is responsible for freeing name.data if NULL is returned (it
7914 * will be freed in nfs4_remove_reclaim_record in the normal case).
7916 struct nfs4_client_reclaim *
7917 nfs4_client_to_reclaim(struct xdr_netobj name, struct xdr_netobj princhash,
7918 struct nfsd_net *nn)
7920 unsigned int strhashval;
7921 struct nfs4_client_reclaim *crp;
7923 crp = alloc_reclaim();
7925 strhashval = clientstr_hashval(name);
7926 INIT_LIST_HEAD(&crp->cr_strhash);
7927 list_add(&crp->cr_strhash, &nn->reclaim_str_hashtbl[strhashval]);
7928 crp->cr_name.data = name.data;
7929 crp->cr_name.len = name.len;
7930 crp->cr_princhash.data = princhash.data;
7931 crp->cr_princhash.len = princhash.len;
7933 nn->reclaim_str_hashtbl_size++;
7939 nfs4_remove_reclaim_record(struct nfs4_client_reclaim *crp, struct nfsd_net *nn)
7941 list_del(&crp->cr_strhash);
7942 kfree(crp->cr_name.data);
7943 kfree(crp->cr_princhash.data);
7945 nn->reclaim_str_hashtbl_size--;
7949 nfs4_release_reclaim(struct nfsd_net *nn)
7951 struct nfs4_client_reclaim *crp = NULL;
7954 for (i = 0; i < CLIENT_HASH_SIZE; i++) {
7955 while (!list_empty(&nn->reclaim_str_hashtbl[i])) {
7956 crp = list_entry(nn->reclaim_str_hashtbl[i].next,
7957 struct nfs4_client_reclaim, cr_strhash);
7958 nfs4_remove_reclaim_record(crp, nn);
7961 WARN_ON_ONCE(nn->reclaim_str_hashtbl_size);
7965 * called from OPEN, CLAIM_PREVIOUS with a new clientid. */
7966 struct nfs4_client_reclaim *
7967 nfsd4_find_reclaim_client(struct xdr_netobj name, struct nfsd_net *nn)
7969 unsigned int strhashval;
7970 struct nfs4_client_reclaim *crp = NULL;
7972 strhashval = clientstr_hashval(name);
7973 list_for_each_entry(crp, &nn->reclaim_str_hashtbl[strhashval], cr_strhash) {
7974 if (compare_blob(&crp->cr_name, &name) == 0) {
7982 nfs4_check_open_reclaim(struct nfs4_client *clp)
7984 if (test_bit(NFSD4_CLIENT_RECLAIM_COMPLETE, &clp->cl_flags))
7985 return nfserr_no_grace;
7987 if (nfsd4_client_record_check(clp))
7988 return nfserr_reclaim_bad;
7994 * Since the lifetime of a delegation isn't limited to that of an open, a
7995 * client may quite reasonably hang on to a delegation as long as it has
7996 * the inode cached. This becomes an obvious problem the first time a
7997 * client's inode cache approaches the size of the server's total memory.
7999 * For now we avoid this problem by imposing a hard limit on the number
8000 * of delegations, which varies according to the server's memory size.
8003 set_max_delegations(void)
8006 * Allow at most 4 delegations per megabyte of RAM. Quick
8007 * estimates suggest that in the worst case (where every delegation
8008 * is for a different inode), a delegation could take about 1.5K,
8009 * giving a worst case usage of about 6% of memory.
8011 max_delegations = nr_free_buffer_pages() >> (20 - 2 - PAGE_SHIFT);
8014 static int nfs4_state_create_net(struct net *net)
8016 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
8019 nn->conf_id_hashtbl = kmalloc_array(CLIENT_HASH_SIZE,
8020 sizeof(struct list_head),
8022 if (!nn->conf_id_hashtbl)
8024 nn->unconf_id_hashtbl = kmalloc_array(CLIENT_HASH_SIZE,
8025 sizeof(struct list_head),
8027 if (!nn->unconf_id_hashtbl)
8029 nn->sessionid_hashtbl = kmalloc_array(SESSION_HASH_SIZE,
8030 sizeof(struct list_head),
8032 if (!nn->sessionid_hashtbl)
8035 for (i = 0; i < CLIENT_HASH_SIZE; i++) {
8036 INIT_LIST_HEAD(&nn->conf_id_hashtbl[i]);
8037 INIT_LIST_HEAD(&nn->unconf_id_hashtbl[i]);
8039 for (i = 0; i < SESSION_HASH_SIZE; i++)
8040 INIT_LIST_HEAD(&nn->sessionid_hashtbl[i]);
8041 nn->conf_name_tree = RB_ROOT;
8042 nn->unconf_name_tree = RB_ROOT;
8043 nn->boot_time = ktime_get_real_seconds();
8044 nn->grace_ended = false;
8045 nn->nfsd4_manager.block_opens = true;
8046 INIT_LIST_HEAD(&nn->nfsd4_manager.list);
8047 INIT_LIST_HEAD(&nn->client_lru);
8048 INIT_LIST_HEAD(&nn->close_lru);
8049 INIT_LIST_HEAD(&nn->del_recall_lru);
8050 spin_lock_init(&nn->client_lock);
8051 spin_lock_init(&nn->s2s_cp_lock);
8052 idr_init(&nn->s2s_cp_stateids);
8054 spin_lock_init(&nn->blocked_locks_lock);
8055 INIT_LIST_HEAD(&nn->blocked_locks_lru);
8057 INIT_DELAYED_WORK(&nn->laundromat_work, laundromat_main);
8058 INIT_WORK(&nn->nfsd_shrinker_work, nfsd4_state_shrinker_worker);
8061 nn->nfsd_client_shrinker.scan_objects = nfsd4_state_shrinker_scan;
8062 nn->nfsd_client_shrinker.count_objects = nfsd4_state_shrinker_count;
8063 nn->nfsd_client_shrinker.seeks = DEFAULT_SEEKS;
8065 if (register_shrinker(&nn->nfsd_client_shrinker, "nfsd-client"))
8071 kfree(nn->sessionid_hashtbl);
8073 kfree(nn->unconf_id_hashtbl);
8075 kfree(nn->conf_id_hashtbl);
8081 nfs4_state_destroy_net(struct net *net)
8084 struct nfs4_client *clp = NULL;
8085 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
8087 for (i = 0; i < CLIENT_HASH_SIZE; i++) {
8088 while (!list_empty(&nn->conf_id_hashtbl[i])) {
8089 clp = list_entry(nn->conf_id_hashtbl[i].next, struct nfs4_client, cl_idhash);
8090 destroy_client(clp);
8094 WARN_ON(!list_empty(&nn->blocked_locks_lru));
8096 for (i = 0; i < CLIENT_HASH_SIZE; i++) {
8097 while (!list_empty(&nn->unconf_id_hashtbl[i])) {
8098 clp = list_entry(nn->unconf_id_hashtbl[i].next, struct nfs4_client, cl_idhash);
8099 destroy_client(clp);
8103 kfree(nn->sessionid_hashtbl);
8104 kfree(nn->unconf_id_hashtbl);
8105 kfree(nn->conf_id_hashtbl);
8110 nfs4_state_start_net(struct net *net)
8112 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
8115 ret = nfs4_state_create_net(net);
8118 locks_start_grace(net, &nn->nfsd4_manager);
8119 nfsd4_client_tracking_init(net);
8120 if (nn->track_reclaim_completes && nn->reclaim_str_hashtbl_size == 0)
8122 printk(KERN_INFO "NFSD: starting %lld-second grace period (net %x)\n",
8123 nn->nfsd4_grace, net->ns.inum);
8124 trace_nfsd_grace_start(nn);
8125 queue_delayed_work(laundry_wq, &nn->laundromat_work, nn->nfsd4_grace * HZ);
8129 printk(KERN_INFO "NFSD: no clients to reclaim, skipping NFSv4 grace period (net %x)\n",
8131 queue_delayed_work(laundry_wq, &nn->laundromat_work, nn->nfsd4_lease * HZ);
8132 nfsd4_end_grace(nn);
8136 /* initialization to perform when the nfsd service is started: */
8139 nfs4_state_start(void)
8143 ret = rhltable_init(&nfs4_file_rhltable, &nfs4_file_rhash_params);
8147 ret = nfsd4_create_callback_queue();
8149 rhltable_destroy(&nfs4_file_rhltable);
8153 set_max_delegations();
8158 nfs4_state_shutdown_net(struct net *net)
8160 struct nfs4_delegation *dp = NULL;
8161 struct list_head *pos, *next, reaplist;
8162 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
8164 unregister_shrinker(&nn->nfsd_client_shrinker);
8165 cancel_work(&nn->nfsd_shrinker_work);
8166 cancel_delayed_work_sync(&nn->laundromat_work);
8167 locks_end_grace(&nn->nfsd4_manager);
8169 INIT_LIST_HEAD(&reaplist);
8170 spin_lock(&state_lock);
8171 list_for_each_safe(pos, next, &nn->del_recall_lru) {
8172 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
8173 WARN_ON(!unhash_delegation_locked(dp));
8174 list_add(&dp->dl_recall_lru, &reaplist);
8176 spin_unlock(&state_lock);
8177 list_for_each_safe(pos, next, &reaplist) {
8178 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
8179 list_del_init(&dp->dl_recall_lru);
8180 destroy_unhashed_deleg(dp);
8183 nfsd4_client_tracking_exit(net);
8184 nfs4_state_destroy_net(net);
8185 #ifdef CONFIG_NFSD_V4_2_INTER_SSC
8186 nfsd4_ssc_shutdown_umount(nn);
8191 nfs4_state_shutdown(void)
8193 nfsd4_destroy_callback_queue();
8194 rhltable_destroy(&nfs4_file_rhltable);
8198 get_stateid(struct nfsd4_compound_state *cstate, stateid_t *stateid)
8200 if (HAS_CSTATE_FLAG(cstate, CURRENT_STATE_ID_FLAG) &&
8201 CURRENT_STATEID(stateid))
8202 memcpy(stateid, &cstate->current_stateid, sizeof(stateid_t));
8206 put_stateid(struct nfsd4_compound_state *cstate, stateid_t *stateid)
8208 if (cstate->minorversion) {
8209 memcpy(&cstate->current_stateid, stateid, sizeof(stateid_t));
8210 SET_CSTATE_FLAG(cstate, CURRENT_STATE_ID_FLAG);
8215 clear_current_stateid(struct nfsd4_compound_state *cstate)
8217 CLEAR_CSTATE_FLAG(cstate, CURRENT_STATE_ID_FLAG);
8221 * functions to set current state id
8224 nfsd4_set_opendowngradestateid(struct nfsd4_compound_state *cstate,
8225 union nfsd4_op_u *u)
8227 put_stateid(cstate, &u->open_downgrade.od_stateid);
8231 nfsd4_set_openstateid(struct nfsd4_compound_state *cstate,
8232 union nfsd4_op_u *u)
8234 put_stateid(cstate, &u->open.op_stateid);
8238 nfsd4_set_closestateid(struct nfsd4_compound_state *cstate,
8239 union nfsd4_op_u *u)
8241 put_stateid(cstate, &u->close.cl_stateid);
8245 nfsd4_set_lockstateid(struct nfsd4_compound_state *cstate,
8246 union nfsd4_op_u *u)
8248 put_stateid(cstate, &u->lock.lk_resp_stateid);
8252 * functions to consume current state id
8256 nfsd4_get_opendowngradestateid(struct nfsd4_compound_state *cstate,
8257 union nfsd4_op_u *u)
8259 get_stateid(cstate, &u->open_downgrade.od_stateid);
8263 nfsd4_get_delegreturnstateid(struct nfsd4_compound_state *cstate,
8264 union nfsd4_op_u *u)
8266 get_stateid(cstate, &u->delegreturn.dr_stateid);
8270 nfsd4_get_freestateid(struct nfsd4_compound_state *cstate,
8271 union nfsd4_op_u *u)
8273 get_stateid(cstate, &u->free_stateid.fr_stateid);
8277 nfsd4_get_setattrstateid(struct nfsd4_compound_state *cstate,
8278 union nfsd4_op_u *u)
8280 get_stateid(cstate, &u->setattr.sa_stateid);
8284 nfsd4_get_closestateid(struct nfsd4_compound_state *cstate,
8285 union nfsd4_op_u *u)
8287 get_stateid(cstate, &u->close.cl_stateid);
8291 nfsd4_get_lockustateid(struct nfsd4_compound_state *cstate,
8292 union nfsd4_op_u *u)
8294 get_stateid(cstate, &u->locku.lu_stateid);
8298 nfsd4_get_readstateid(struct nfsd4_compound_state *cstate,
8299 union nfsd4_op_u *u)
8301 get_stateid(cstate, &u->read.rd_stateid);
8305 nfsd4_get_writestateid(struct nfsd4_compound_state *cstate,
8306 union nfsd4_op_u *u)
8308 get_stateid(cstate, &u->write.wr_stateid);