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 find_writeable_file_locked(struct nfs4_file *f)
605 struct nfsd_file *ret;
607 lockdep_assert_held(&f->fi_lock);
609 ret = nfsd_file_get(f->fi_fds[O_WRONLY]);
611 ret = nfsd_file_get(f->fi_fds[O_RDWR]);
615 static struct nfsd_file *
616 find_writeable_file(struct nfs4_file *f)
618 struct nfsd_file *ret;
620 spin_lock(&f->fi_lock);
621 ret = find_writeable_file_locked(f);
622 spin_unlock(&f->fi_lock);
627 static struct nfsd_file *
628 find_readable_file_locked(struct nfs4_file *f)
630 struct nfsd_file *ret;
632 lockdep_assert_held(&f->fi_lock);
634 ret = nfsd_file_get(f->fi_fds[O_RDONLY]);
636 ret = nfsd_file_get(f->fi_fds[O_RDWR]);
640 static struct nfsd_file *
641 find_readable_file(struct nfs4_file *f)
643 struct nfsd_file *ret;
645 spin_lock(&f->fi_lock);
646 ret = find_readable_file_locked(f);
647 spin_unlock(&f->fi_lock);
653 find_any_file(struct nfs4_file *f)
655 struct nfsd_file *ret;
659 spin_lock(&f->fi_lock);
660 ret = nfsd_file_get(f->fi_fds[O_RDWR]);
662 ret = nfsd_file_get(f->fi_fds[O_WRONLY]);
664 ret = nfsd_file_get(f->fi_fds[O_RDONLY]);
666 spin_unlock(&f->fi_lock);
670 static struct nfsd_file *find_any_file_locked(struct nfs4_file *f)
672 lockdep_assert_held(&f->fi_lock);
674 if (f->fi_fds[O_RDWR])
675 return f->fi_fds[O_RDWR];
676 if (f->fi_fds[O_WRONLY])
677 return f->fi_fds[O_WRONLY];
678 if (f->fi_fds[O_RDONLY])
679 return f->fi_fds[O_RDONLY];
683 static atomic_long_t num_delegations;
684 unsigned long max_delegations;
687 * Open owner state (share locks)
690 /* hash tables for lock and open owners */
691 #define OWNER_HASH_BITS 8
692 #define OWNER_HASH_SIZE (1 << OWNER_HASH_BITS)
693 #define OWNER_HASH_MASK (OWNER_HASH_SIZE - 1)
695 static unsigned int ownerstr_hashval(struct xdr_netobj *ownername)
699 ret = opaque_hashval(ownername->data, ownername->len);
700 return ret & OWNER_HASH_MASK;
703 static struct rhltable nfs4_file_rhltable ____cacheline_aligned_in_smp;
705 static const struct rhashtable_params nfs4_file_rhash_params = {
706 .key_len = sizeof_field(struct nfs4_file, fi_inode),
707 .key_offset = offsetof(struct nfs4_file, fi_inode),
708 .head_offset = offsetof(struct nfs4_file, fi_rlist),
711 * Start with a single page hash table to reduce resizing churn
712 * on light workloads.
715 .automatic_shrinking = true,
719 * Check if courtesy clients have conflicting access and resolve it if possible
721 * access: is op_share_access if share_access is true.
722 * Check if access mode, op_share_access, would conflict with
723 * the current deny mode of the file 'fp'.
724 * access: is op_share_deny if share_access is false.
725 * Check if the deny mode, op_share_deny, would conflict with
726 * current access of the file 'fp'.
727 * stp: skip checking this entry.
728 * new_stp: normal open, not open upgrade.
731 * false - access/deny mode conflict with normal client.
732 * true - no conflict or conflict with courtesy client(s) is resolved.
735 nfs4_resolve_deny_conflicts_locked(struct nfs4_file *fp, bool new_stp,
736 struct nfs4_ol_stateid *stp, u32 access, bool share_access)
738 struct nfs4_ol_stateid *st;
739 bool resolvable = true;
742 struct nfs4_client *clp;
744 lockdep_assert_held(&fp->fi_lock);
745 list_for_each_entry(st, &fp->fi_stateids, st_perfile) {
746 /* ignore lock stateid */
749 if (st == stp && new_stp)
751 /* check file access against deny mode or vice versa */
752 bmap = share_access ? st->st_deny_bmap : st->st_access_bmap;
753 if (!(access & bmap_to_share_mode(bmap)))
755 clp = st->st_stid.sc_client;
756 if (try_to_expire_client(clp))
762 clp = stp->st_stid.sc_client;
763 nn = net_generic(clp->net, nfsd_net_id);
764 mod_delayed_work(laundry_wq, &nn->laundromat_work, 0);
770 __nfs4_file_get_access(struct nfs4_file *fp, u32 access)
772 lockdep_assert_held(&fp->fi_lock);
774 if (access & NFS4_SHARE_ACCESS_WRITE)
775 atomic_inc(&fp->fi_access[O_WRONLY]);
776 if (access & NFS4_SHARE_ACCESS_READ)
777 atomic_inc(&fp->fi_access[O_RDONLY]);
781 nfs4_file_get_access(struct nfs4_file *fp, u32 access)
783 lockdep_assert_held(&fp->fi_lock);
785 /* Does this access mode make sense? */
786 if (access & ~NFS4_SHARE_ACCESS_BOTH)
789 /* Does it conflict with a deny mode already set? */
790 if ((access & fp->fi_share_deny) != 0)
791 return nfserr_share_denied;
793 __nfs4_file_get_access(fp, access);
797 static __be32 nfs4_file_check_deny(struct nfs4_file *fp, u32 deny)
799 /* Common case is that there is no deny mode. */
801 /* Does this deny mode make sense? */
802 if (deny & ~NFS4_SHARE_DENY_BOTH)
805 if ((deny & NFS4_SHARE_DENY_READ) &&
806 atomic_read(&fp->fi_access[O_RDONLY]))
807 return nfserr_share_denied;
809 if ((deny & NFS4_SHARE_DENY_WRITE) &&
810 atomic_read(&fp->fi_access[O_WRONLY]))
811 return nfserr_share_denied;
816 static void __nfs4_file_put_access(struct nfs4_file *fp, int oflag)
818 might_lock(&fp->fi_lock);
820 if (atomic_dec_and_lock(&fp->fi_access[oflag], &fp->fi_lock)) {
821 struct nfsd_file *f1 = NULL;
822 struct nfsd_file *f2 = NULL;
824 swap(f1, fp->fi_fds[oflag]);
825 if (atomic_read(&fp->fi_access[1 - oflag]) == 0)
826 swap(f2, fp->fi_fds[O_RDWR]);
827 spin_unlock(&fp->fi_lock);
835 static void nfs4_file_put_access(struct nfs4_file *fp, u32 access)
837 WARN_ON_ONCE(access & ~NFS4_SHARE_ACCESS_BOTH);
839 if (access & NFS4_SHARE_ACCESS_WRITE)
840 __nfs4_file_put_access(fp, O_WRONLY);
841 if (access & NFS4_SHARE_ACCESS_READ)
842 __nfs4_file_put_access(fp, O_RDONLY);
846 * Allocate a new open/delegation state counter. This is needed for
847 * pNFS for proper return on close semantics.
849 * Note that we only allocate it for pNFS-enabled exports, otherwise
850 * all pointers to struct nfs4_clnt_odstate are always NULL.
852 static struct nfs4_clnt_odstate *
853 alloc_clnt_odstate(struct nfs4_client *clp)
855 struct nfs4_clnt_odstate *co;
857 co = kmem_cache_zalloc(odstate_slab, GFP_KERNEL);
860 refcount_set(&co->co_odcount, 1);
866 hash_clnt_odstate_locked(struct nfs4_clnt_odstate *co)
868 struct nfs4_file *fp = co->co_file;
870 lockdep_assert_held(&fp->fi_lock);
871 list_add(&co->co_perfile, &fp->fi_clnt_odstate);
875 get_clnt_odstate(struct nfs4_clnt_odstate *co)
878 refcount_inc(&co->co_odcount);
882 put_clnt_odstate(struct nfs4_clnt_odstate *co)
884 struct nfs4_file *fp;
890 if (refcount_dec_and_lock(&co->co_odcount, &fp->fi_lock)) {
891 list_del(&co->co_perfile);
892 spin_unlock(&fp->fi_lock);
894 nfsd4_return_all_file_layouts(co->co_client, fp);
895 kmem_cache_free(odstate_slab, co);
899 static struct nfs4_clnt_odstate *
900 find_or_hash_clnt_odstate(struct nfs4_file *fp, struct nfs4_clnt_odstate *new)
902 struct nfs4_clnt_odstate *co;
903 struct nfs4_client *cl;
910 spin_lock(&fp->fi_lock);
911 list_for_each_entry(co, &fp->fi_clnt_odstate, co_perfile) {
912 if (co->co_client == cl) {
913 get_clnt_odstate(co);
919 hash_clnt_odstate_locked(new);
921 spin_unlock(&fp->fi_lock);
925 struct nfs4_stid *nfs4_alloc_stid(struct nfs4_client *cl, struct kmem_cache *slab,
926 void (*sc_free)(struct nfs4_stid *))
928 struct nfs4_stid *stid;
931 stid = kmem_cache_zalloc(slab, GFP_KERNEL);
935 idr_preload(GFP_KERNEL);
936 spin_lock(&cl->cl_lock);
937 /* Reserving 0 for start of file in nfsdfs "states" file: */
938 new_id = idr_alloc_cyclic(&cl->cl_stateids, stid, 1, 0, GFP_NOWAIT);
939 spin_unlock(&cl->cl_lock);
944 stid->sc_free = sc_free;
945 stid->sc_client = cl;
946 stid->sc_stateid.si_opaque.so_id = new_id;
947 stid->sc_stateid.si_opaque.so_clid = cl->cl_clientid;
948 /* Will be incremented before return to client: */
949 refcount_set(&stid->sc_count, 1);
950 spin_lock_init(&stid->sc_lock);
951 INIT_LIST_HEAD(&stid->sc_cp_list);
954 * It shouldn't be a problem to reuse an opaque stateid value.
955 * I don't think it is for 4.1. But with 4.0 I worry that, for
956 * example, a stray write retransmission could be accepted by
957 * the server when it should have been rejected. Therefore,
958 * adopt a trick from the sctp code to attempt to maximize the
959 * amount of time until an id is reused, by ensuring they always
960 * "increase" (mod INT_MAX):
964 kmem_cache_free(slab, stid);
969 * Create a unique stateid_t to represent each COPY.
971 static int nfs4_init_cp_state(struct nfsd_net *nn, copy_stateid_t *stid,
972 unsigned char cs_type)
976 stid->cs_stid.si_opaque.so_clid.cl_boot = (u32)nn->boot_time;
977 stid->cs_stid.si_opaque.so_clid.cl_id = nn->s2s_cp_cl_id;
979 idr_preload(GFP_KERNEL);
980 spin_lock(&nn->s2s_cp_lock);
981 new_id = idr_alloc_cyclic(&nn->s2s_cp_stateids, stid, 0, 0, GFP_NOWAIT);
982 stid->cs_stid.si_opaque.so_id = new_id;
983 stid->cs_stid.si_generation = 1;
984 spin_unlock(&nn->s2s_cp_lock);
988 stid->cs_type = cs_type;
992 int nfs4_init_copy_state(struct nfsd_net *nn, struct nfsd4_copy *copy)
994 return nfs4_init_cp_state(nn, ©->cp_stateid, NFS4_COPY_STID);
997 struct nfs4_cpntf_state *nfs4_alloc_init_cpntf_state(struct nfsd_net *nn,
998 struct nfs4_stid *p_stid)
1000 struct nfs4_cpntf_state *cps;
1002 cps = kzalloc(sizeof(struct nfs4_cpntf_state), GFP_KERNEL);
1005 cps->cpntf_time = ktime_get_boottime_seconds();
1006 refcount_set(&cps->cp_stateid.cs_count, 1);
1007 if (!nfs4_init_cp_state(nn, &cps->cp_stateid, NFS4_COPYNOTIFY_STID))
1009 spin_lock(&nn->s2s_cp_lock);
1010 list_add(&cps->cp_list, &p_stid->sc_cp_list);
1011 spin_unlock(&nn->s2s_cp_lock);
1018 void nfs4_free_copy_state(struct nfsd4_copy *copy)
1020 struct nfsd_net *nn;
1022 if (copy->cp_stateid.cs_type != NFS4_COPY_STID)
1024 nn = net_generic(copy->cp_clp->net, nfsd_net_id);
1025 spin_lock(&nn->s2s_cp_lock);
1026 idr_remove(&nn->s2s_cp_stateids,
1027 copy->cp_stateid.cs_stid.si_opaque.so_id);
1028 spin_unlock(&nn->s2s_cp_lock);
1031 static void nfs4_free_cpntf_statelist(struct net *net, struct nfs4_stid *stid)
1033 struct nfs4_cpntf_state *cps;
1034 struct nfsd_net *nn;
1036 nn = net_generic(net, nfsd_net_id);
1037 spin_lock(&nn->s2s_cp_lock);
1038 while (!list_empty(&stid->sc_cp_list)) {
1039 cps = list_first_entry(&stid->sc_cp_list,
1040 struct nfs4_cpntf_state, cp_list);
1041 _free_cpntf_state_locked(nn, cps);
1043 spin_unlock(&nn->s2s_cp_lock);
1046 static struct nfs4_ol_stateid * nfs4_alloc_open_stateid(struct nfs4_client *clp)
1048 struct nfs4_stid *stid;
1050 stid = nfs4_alloc_stid(clp, stateid_slab, nfs4_free_ol_stateid);
1054 return openlockstateid(stid);
1057 static void nfs4_free_deleg(struct nfs4_stid *stid)
1059 struct nfs4_delegation *dp = delegstateid(stid);
1061 WARN_ON_ONCE(!list_empty(&stid->sc_cp_list));
1062 WARN_ON_ONCE(!list_empty(&dp->dl_perfile));
1063 WARN_ON_ONCE(!list_empty(&dp->dl_perclnt));
1064 WARN_ON_ONCE(!list_empty(&dp->dl_recall_lru));
1065 kmem_cache_free(deleg_slab, stid);
1066 atomic_long_dec(&num_delegations);
1070 * When we recall a delegation, we should be careful not to hand it
1071 * out again straight away.
1072 * To ensure this we keep a pair of bloom filters ('new' and 'old')
1073 * in which the filehandles of recalled delegations are "stored".
1074 * If a filehandle appear in either filter, a delegation is blocked.
1075 * When a delegation is recalled, the filehandle is stored in the "new"
1077 * Every 30 seconds we swap the filters and clear the "new" one,
1078 * unless both are empty of course.
1080 * Each filter is 256 bits. We hash the filehandle to 32bit and use the
1081 * low 3 bytes as hash-table indices.
1083 * 'blocked_delegations_lock', which is always taken in block_delegations(),
1084 * is used to manage concurrent access. Testing does not need the lock
1085 * except when swapping the two filters.
1087 static DEFINE_SPINLOCK(blocked_delegations_lock);
1088 static struct bloom_pair {
1089 int entries, old_entries;
1091 int new; /* index into 'set' */
1092 DECLARE_BITMAP(set[2], 256);
1093 } blocked_delegations;
1095 static int delegation_blocked(struct knfsd_fh *fh)
1098 struct bloom_pair *bd = &blocked_delegations;
1100 if (bd->entries == 0)
1102 if (ktime_get_seconds() - bd->swap_time > 30) {
1103 spin_lock(&blocked_delegations_lock);
1104 if (ktime_get_seconds() - bd->swap_time > 30) {
1105 bd->entries -= bd->old_entries;
1106 bd->old_entries = bd->entries;
1107 memset(bd->set[bd->new], 0,
1108 sizeof(bd->set[0]));
1109 bd->new = 1-bd->new;
1110 bd->swap_time = ktime_get_seconds();
1112 spin_unlock(&blocked_delegations_lock);
1114 hash = jhash(&fh->fh_raw, fh->fh_size, 0);
1115 if (test_bit(hash&255, bd->set[0]) &&
1116 test_bit((hash>>8)&255, bd->set[0]) &&
1117 test_bit((hash>>16)&255, bd->set[0]))
1120 if (test_bit(hash&255, bd->set[1]) &&
1121 test_bit((hash>>8)&255, bd->set[1]) &&
1122 test_bit((hash>>16)&255, bd->set[1]))
1128 static void block_delegations(struct knfsd_fh *fh)
1131 struct bloom_pair *bd = &blocked_delegations;
1133 hash = jhash(&fh->fh_raw, fh->fh_size, 0);
1135 spin_lock(&blocked_delegations_lock);
1136 __set_bit(hash&255, bd->set[bd->new]);
1137 __set_bit((hash>>8)&255, bd->set[bd->new]);
1138 __set_bit((hash>>16)&255, bd->set[bd->new]);
1139 if (bd->entries == 0)
1140 bd->swap_time = ktime_get_seconds();
1142 spin_unlock(&blocked_delegations_lock);
1145 static struct nfs4_delegation *
1146 alloc_init_deleg(struct nfs4_client *clp, struct nfs4_file *fp,
1147 struct nfs4_clnt_odstate *odstate)
1149 struct nfs4_delegation *dp;
1152 dprintk("NFSD alloc_init_deleg\n");
1153 n = atomic_long_inc_return(&num_delegations);
1154 if (n < 0 || n > max_delegations)
1156 if (delegation_blocked(&fp->fi_fhandle))
1158 dp = delegstateid(nfs4_alloc_stid(clp, deleg_slab, nfs4_free_deleg));
1163 * delegation seqid's are never incremented. The 4.1 special
1164 * meaning of seqid 0 isn't meaningful, really, but let's avoid
1165 * 0 anyway just for consistency and use 1:
1167 dp->dl_stid.sc_stateid.si_generation = 1;
1168 INIT_LIST_HEAD(&dp->dl_perfile);
1169 INIT_LIST_HEAD(&dp->dl_perclnt);
1170 INIT_LIST_HEAD(&dp->dl_recall_lru);
1171 dp->dl_clnt_odstate = odstate;
1172 get_clnt_odstate(odstate);
1173 dp->dl_type = NFS4_OPEN_DELEGATE_READ;
1175 dp->dl_recalled = false;
1176 nfsd4_init_cb(&dp->dl_recall, dp->dl_stid.sc_client,
1177 &nfsd4_cb_recall_ops, NFSPROC4_CLNT_CB_RECALL);
1179 dp->dl_stid.sc_file = fp;
1182 atomic_long_dec(&num_delegations);
1187 nfs4_put_stid(struct nfs4_stid *s)
1189 struct nfs4_file *fp = s->sc_file;
1190 struct nfs4_client *clp = s->sc_client;
1192 might_lock(&clp->cl_lock);
1194 if (!refcount_dec_and_lock(&s->sc_count, &clp->cl_lock)) {
1195 wake_up_all(&close_wq);
1198 idr_remove(&clp->cl_stateids, s->sc_stateid.si_opaque.so_id);
1199 nfs4_free_cpntf_statelist(clp->net, s);
1200 spin_unlock(&clp->cl_lock);
1207 nfs4_inc_and_copy_stateid(stateid_t *dst, struct nfs4_stid *stid)
1209 stateid_t *src = &stid->sc_stateid;
1211 spin_lock(&stid->sc_lock);
1212 if (unlikely(++src->si_generation == 0))
1213 src->si_generation = 1;
1214 memcpy(dst, src, sizeof(*dst));
1215 spin_unlock(&stid->sc_lock);
1218 static void put_deleg_file(struct nfs4_file *fp)
1220 struct nfsd_file *nf = NULL;
1222 spin_lock(&fp->fi_lock);
1223 if (--fp->fi_delegees == 0)
1224 swap(nf, fp->fi_deleg_file);
1225 spin_unlock(&fp->fi_lock);
1231 static void nfs4_unlock_deleg_lease(struct nfs4_delegation *dp)
1233 struct nfs4_file *fp = dp->dl_stid.sc_file;
1234 struct nfsd_file *nf = fp->fi_deleg_file;
1236 WARN_ON_ONCE(!fp->fi_delegees);
1238 vfs_setlease(nf->nf_file, F_UNLCK, NULL, (void **)&dp);
1242 static void destroy_unhashed_deleg(struct nfs4_delegation *dp)
1244 put_clnt_odstate(dp->dl_clnt_odstate);
1245 nfs4_unlock_deleg_lease(dp);
1246 nfs4_put_stid(&dp->dl_stid);
1249 void nfs4_unhash_stid(struct nfs4_stid *s)
1255 * nfs4_delegation_exists - Discover if this delegation already exists
1256 * @clp: a pointer to the nfs4_client we're granting a delegation to
1257 * @fp: a pointer to the nfs4_file we're granting a delegation on
1260 * On success: true iff an existing delegation is found
1264 nfs4_delegation_exists(struct nfs4_client *clp, struct nfs4_file *fp)
1266 struct nfs4_delegation *searchdp = NULL;
1267 struct nfs4_client *searchclp = NULL;
1269 lockdep_assert_held(&state_lock);
1270 lockdep_assert_held(&fp->fi_lock);
1272 list_for_each_entry(searchdp, &fp->fi_delegations, dl_perfile) {
1273 searchclp = searchdp->dl_stid.sc_client;
1274 if (clp == searchclp) {
1282 * hash_delegation_locked - Add a delegation to the appropriate lists
1283 * @dp: a pointer to the nfs4_delegation we are adding.
1284 * @fp: a pointer to the nfs4_file we're granting a delegation on
1287 * On success: NULL if the delegation was successfully hashed.
1289 * On error: -EAGAIN if one was previously granted to this
1290 * nfs4_client for this nfs4_file. Delegation is not hashed.
1295 hash_delegation_locked(struct nfs4_delegation *dp, struct nfs4_file *fp)
1297 struct nfs4_client *clp = dp->dl_stid.sc_client;
1299 lockdep_assert_held(&state_lock);
1300 lockdep_assert_held(&fp->fi_lock);
1302 if (nfs4_delegation_exists(clp, fp))
1304 refcount_inc(&dp->dl_stid.sc_count);
1305 dp->dl_stid.sc_type = NFS4_DELEG_STID;
1306 list_add(&dp->dl_perfile, &fp->fi_delegations);
1307 list_add(&dp->dl_perclnt, &clp->cl_delegations);
1311 static bool delegation_hashed(struct nfs4_delegation *dp)
1313 return !(list_empty(&dp->dl_perfile));
1317 unhash_delegation_locked(struct nfs4_delegation *dp)
1319 struct nfs4_file *fp = dp->dl_stid.sc_file;
1321 lockdep_assert_held(&state_lock);
1323 if (!delegation_hashed(dp))
1326 dp->dl_stid.sc_type = NFS4_CLOSED_DELEG_STID;
1327 /* Ensure that deleg break won't try to requeue it */
1329 spin_lock(&fp->fi_lock);
1330 list_del_init(&dp->dl_perclnt);
1331 list_del_init(&dp->dl_recall_lru);
1332 list_del_init(&dp->dl_perfile);
1333 spin_unlock(&fp->fi_lock);
1337 static void destroy_delegation(struct nfs4_delegation *dp)
1341 spin_lock(&state_lock);
1342 unhashed = unhash_delegation_locked(dp);
1343 spin_unlock(&state_lock);
1345 destroy_unhashed_deleg(dp);
1348 static void revoke_delegation(struct nfs4_delegation *dp)
1350 struct nfs4_client *clp = dp->dl_stid.sc_client;
1352 WARN_ON(!list_empty(&dp->dl_recall_lru));
1354 trace_nfsd_stid_revoke(&dp->dl_stid);
1356 if (clp->cl_minorversion) {
1357 dp->dl_stid.sc_type = NFS4_REVOKED_DELEG_STID;
1358 refcount_inc(&dp->dl_stid.sc_count);
1359 spin_lock(&clp->cl_lock);
1360 list_add(&dp->dl_recall_lru, &clp->cl_revoked);
1361 spin_unlock(&clp->cl_lock);
1363 destroy_unhashed_deleg(dp);
1370 static unsigned int clientid_hashval(u32 id)
1372 return id & CLIENT_HASH_MASK;
1375 static unsigned int clientstr_hashval(struct xdr_netobj name)
1377 return opaque_hashval(name.data, 8) & CLIENT_HASH_MASK;
1381 * A stateid that had a deny mode associated with it is being released
1382 * or downgraded. Recalculate the deny mode on the file.
1385 recalculate_deny_mode(struct nfs4_file *fp)
1387 struct nfs4_ol_stateid *stp;
1389 spin_lock(&fp->fi_lock);
1390 fp->fi_share_deny = 0;
1391 list_for_each_entry(stp, &fp->fi_stateids, st_perfile)
1392 fp->fi_share_deny |= bmap_to_share_mode(stp->st_deny_bmap);
1393 spin_unlock(&fp->fi_lock);
1397 reset_union_bmap_deny(u32 deny, struct nfs4_ol_stateid *stp)
1400 bool change = false;
1402 for (i = 1; i < 4; i++) {
1403 if ((i & deny) != i) {
1409 /* Recalculate per-file deny mode if there was a change */
1411 recalculate_deny_mode(stp->st_stid.sc_file);
1414 /* release all access and file references for a given stateid */
1416 release_all_access(struct nfs4_ol_stateid *stp)
1419 struct nfs4_file *fp = stp->st_stid.sc_file;
1421 if (fp && stp->st_deny_bmap != 0)
1422 recalculate_deny_mode(fp);
1424 for (i = 1; i < 4; i++) {
1425 if (test_access(i, stp))
1426 nfs4_file_put_access(stp->st_stid.sc_file, i);
1427 clear_access(i, stp);
1431 static inline void nfs4_free_stateowner(struct nfs4_stateowner *sop)
1433 kfree(sop->so_owner.data);
1434 sop->so_ops->so_free(sop);
1437 static void nfs4_put_stateowner(struct nfs4_stateowner *sop)
1439 struct nfs4_client *clp = sop->so_client;
1441 might_lock(&clp->cl_lock);
1443 if (!atomic_dec_and_lock(&sop->so_count, &clp->cl_lock))
1445 sop->so_ops->so_unhash(sop);
1446 spin_unlock(&clp->cl_lock);
1447 nfs4_free_stateowner(sop);
1451 nfs4_ol_stateid_unhashed(const struct nfs4_ol_stateid *stp)
1453 return list_empty(&stp->st_perfile);
1456 static bool unhash_ol_stateid(struct nfs4_ol_stateid *stp)
1458 struct nfs4_file *fp = stp->st_stid.sc_file;
1460 lockdep_assert_held(&stp->st_stateowner->so_client->cl_lock);
1462 if (list_empty(&stp->st_perfile))
1465 spin_lock(&fp->fi_lock);
1466 list_del_init(&stp->st_perfile);
1467 spin_unlock(&fp->fi_lock);
1468 list_del(&stp->st_perstateowner);
1472 static void nfs4_free_ol_stateid(struct nfs4_stid *stid)
1474 struct nfs4_ol_stateid *stp = openlockstateid(stid);
1476 put_clnt_odstate(stp->st_clnt_odstate);
1477 release_all_access(stp);
1478 if (stp->st_stateowner)
1479 nfs4_put_stateowner(stp->st_stateowner);
1480 WARN_ON(!list_empty(&stid->sc_cp_list));
1481 kmem_cache_free(stateid_slab, stid);
1484 static void nfs4_free_lock_stateid(struct nfs4_stid *stid)
1486 struct nfs4_ol_stateid *stp = openlockstateid(stid);
1487 struct nfs4_lockowner *lo = lockowner(stp->st_stateowner);
1488 struct nfsd_file *nf;
1490 nf = find_any_file(stp->st_stid.sc_file);
1492 get_file(nf->nf_file);
1493 filp_close(nf->nf_file, (fl_owner_t)lo);
1496 nfs4_free_ol_stateid(stid);
1500 * Put the persistent reference to an already unhashed generic stateid, while
1501 * holding the cl_lock. If it's the last reference, then put it onto the
1502 * reaplist for later destruction.
1504 static void put_ol_stateid_locked(struct nfs4_ol_stateid *stp,
1505 struct list_head *reaplist)
1507 struct nfs4_stid *s = &stp->st_stid;
1508 struct nfs4_client *clp = s->sc_client;
1510 lockdep_assert_held(&clp->cl_lock);
1512 WARN_ON_ONCE(!list_empty(&stp->st_locks));
1514 if (!refcount_dec_and_test(&s->sc_count)) {
1515 wake_up_all(&close_wq);
1519 idr_remove(&clp->cl_stateids, s->sc_stateid.si_opaque.so_id);
1520 list_add(&stp->st_locks, reaplist);
1523 static bool unhash_lock_stateid(struct nfs4_ol_stateid *stp)
1525 lockdep_assert_held(&stp->st_stid.sc_client->cl_lock);
1527 if (!unhash_ol_stateid(stp))
1529 list_del_init(&stp->st_locks);
1530 nfs4_unhash_stid(&stp->st_stid);
1534 static void release_lock_stateid(struct nfs4_ol_stateid *stp)
1536 struct nfs4_client *clp = stp->st_stid.sc_client;
1539 spin_lock(&clp->cl_lock);
1540 unhashed = unhash_lock_stateid(stp);
1541 spin_unlock(&clp->cl_lock);
1543 nfs4_put_stid(&stp->st_stid);
1546 static void unhash_lockowner_locked(struct nfs4_lockowner *lo)
1548 struct nfs4_client *clp = lo->lo_owner.so_client;
1550 lockdep_assert_held(&clp->cl_lock);
1552 list_del_init(&lo->lo_owner.so_strhash);
1556 * Free a list of generic stateids that were collected earlier after being
1560 free_ol_stateid_reaplist(struct list_head *reaplist)
1562 struct nfs4_ol_stateid *stp;
1563 struct nfs4_file *fp;
1567 while (!list_empty(reaplist)) {
1568 stp = list_first_entry(reaplist, struct nfs4_ol_stateid,
1570 list_del(&stp->st_locks);
1571 fp = stp->st_stid.sc_file;
1572 stp->st_stid.sc_free(&stp->st_stid);
1578 static void release_open_stateid_locks(struct nfs4_ol_stateid *open_stp,
1579 struct list_head *reaplist)
1581 struct nfs4_ol_stateid *stp;
1583 lockdep_assert_held(&open_stp->st_stid.sc_client->cl_lock);
1585 while (!list_empty(&open_stp->st_locks)) {
1586 stp = list_entry(open_stp->st_locks.next,
1587 struct nfs4_ol_stateid, st_locks);
1588 WARN_ON(!unhash_lock_stateid(stp));
1589 put_ol_stateid_locked(stp, reaplist);
1593 static bool unhash_open_stateid(struct nfs4_ol_stateid *stp,
1594 struct list_head *reaplist)
1596 lockdep_assert_held(&stp->st_stid.sc_client->cl_lock);
1598 if (!unhash_ol_stateid(stp))
1600 release_open_stateid_locks(stp, reaplist);
1604 static void release_open_stateid(struct nfs4_ol_stateid *stp)
1606 LIST_HEAD(reaplist);
1608 spin_lock(&stp->st_stid.sc_client->cl_lock);
1609 if (unhash_open_stateid(stp, &reaplist))
1610 put_ol_stateid_locked(stp, &reaplist);
1611 spin_unlock(&stp->st_stid.sc_client->cl_lock);
1612 free_ol_stateid_reaplist(&reaplist);
1615 static void unhash_openowner_locked(struct nfs4_openowner *oo)
1617 struct nfs4_client *clp = oo->oo_owner.so_client;
1619 lockdep_assert_held(&clp->cl_lock);
1621 list_del_init(&oo->oo_owner.so_strhash);
1622 list_del_init(&oo->oo_perclient);
1625 static void release_last_closed_stateid(struct nfs4_openowner *oo)
1627 struct nfsd_net *nn = net_generic(oo->oo_owner.so_client->net,
1629 struct nfs4_ol_stateid *s;
1631 spin_lock(&nn->client_lock);
1632 s = oo->oo_last_closed_stid;
1634 list_del_init(&oo->oo_close_lru);
1635 oo->oo_last_closed_stid = NULL;
1637 spin_unlock(&nn->client_lock);
1639 nfs4_put_stid(&s->st_stid);
1642 static void release_openowner(struct nfs4_openowner *oo)
1644 struct nfs4_ol_stateid *stp;
1645 struct nfs4_client *clp = oo->oo_owner.so_client;
1646 struct list_head reaplist;
1648 INIT_LIST_HEAD(&reaplist);
1650 spin_lock(&clp->cl_lock);
1651 unhash_openowner_locked(oo);
1652 while (!list_empty(&oo->oo_owner.so_stateids)) {
1653 stp = list_first_entry(&oo->oo_owner.so_stateids,
1654 struct nfs4_ol_stateid, st_perstateowner);
1655 if (unhash_open_stateid(stp, &reaplist))
1656 put_ol_stateid_locked(stp, &reaplist);
1658 spin_unlock(&clp->cl_lock);
1659 free_ol_stateid_reaplist(&reaplist);
1660 release_last_closed_stateid(oo);
1661 nfs4_put_stateowner(&oo->oo_owner);
1665 hash_sessionid(struct nfs4_sessionid *sessionid)
1667 struct nfsd4_sessionid *sid = (struct nfsd4_sessionid *)sessionid;
1669 return sid->sequence % SESSION_HASH_SIZE;
1672 #ifdef CONFIG_SUNRPC_DEBUG
1674 dump_sessionid(const char *fn, struct nfs4_sessionid *sessionid)
1676 u32 *ptr = (u32 *)(&sessionid->data[0]);
1677 dprintk("%s: %u:%u:%u:%u\n", fn, ptr[0], ptr[1], ptr[2], ptr[3]);
1681 dump_sessionid(const char *fn, struct nfs4_sessionid *sessionid)
1687 * Bump the seqid on cstate->replay_owner, and clear replay_owner if it
1688 * won't be used for replay.
1690 void nfsd4_bump_seqid(struct nfsd4_compound_state *cstate, __be32 nfserr)
1692 struct nfs4_stateowner *so = cstate->replay_owner;
1694 if (nfserr == nfserr_replay_me)
1697 if (!seqid_mutating_err(ntohl(nfserr))) {
1698 nfsd4_cstate_clear_replay(cstate);
1703 if (so->so_is_open_owner)
1704 release_last_closed_stateid(openowner(so));
1710 gen_sessionid(struct nfsd4_session *ses)
1712 struct nfs4_client *clp = ses->se_client;
1713 struct nfsd4_sessionid *sid;
1715 sid = (struct nfsd4_sessionid *)ses->se_sessionid.data;
1716 sid->clientid = clp->cl_clientid;
1717 sid->sequence = current_sessionid++;
1722 * The protocol defines ca_maxresponssize_cached to include the size of
1723 * the rpc header, but all we need to cache is the data starting after
1724 * the end of the initial SEQUENCE operation--the rest we regenerate
1725 * each time. Therefore we can advertise a ca_maxresponssize_cached
1726 * value that is the number of bytes in our cache plus a few additional
1727 * bytes. In order to stay on the safe side, and not promise more than
1728 * we can cache, those additional bytes must be the minimum possible: 24
1729 * bytes of rpc header (xid through accept state, with AUTH_NULL
1730 * verifier), 12 for the compound header (with zero-length tag), and 44
1731 * for the SEQUENCE op response:
1733 #define NFSD_MIN_HDR_SEQ_SZ (24 + 12 + 44)
1736 free_session_slots(struct nfsd4_session *ses)
1740 for (i = 0; i < ses->se_fchannel.maxreqs; i++) {
1741 free_svc_cred(&ses->se_slots[i]->sl_cred);
1742 kfree(ses->se_slots[i]);
1747 * We don't actually need to cache the rpc and session headers, so we
1748 * can allocate a little less for each slot:
1750 static inline u32 slot_bytes(struct nfsd4_channel_attrs *ca)
1754 if (ca->maxresp_cached < NFSD_MIN_HDR_SEQ_SZ)
1757 size = ca->maxresp_cached - NFSD_MIN_HDR_SEQ_SZ;
1758 return size + sizeof(struct nfsd4_slot);
1762 * XXX: If we run out of reserved DRC memory we could (up to a point)
1763 * re-negotiate active sessions and reduce their slot usage to make
1764 * room for new connections. For now we just fail the create session.
1766 static u32 nfsd4_get_drc_mem(struct nfsd4_channel_attrs *ca, struct nfsd_net *nn)
1768 u32 slotsize = slot_bytes(ca);
1769 u32 num = ca->maxreqs;
1770 unsigned long avail, total_avail;
1771 unsigned int scale_factor;
1773 spin_lock(&nfsd_drc_lock);
1774 if (nfsd_drc_max_mem > nfsd_drc_mem_used)
1775 total_avail = nfsd_drc_max_mem - nfsd_drc_mem_used;
1777 /* We have handed out more space than we chose in
1778 * set_max_drc() to allow. That isn't really a
1779 * problem as long as that doesn't make us think we
1780 * have lots more due to integer overflow.
1783 avail = min((unsigned long)NFSD_MAX_MEM_PER_SESSION, total_avail);
1785 * Never use more than a fraction of the remaining memory,
1786 * unless it's the only way to give this client a slot.
1787 * The chosen fraction is either 1/8 or 1/number of threads,
1788 * whichever is smaller. This ensures there are adequate
1789 * slots to support multiple clients per thread.
1790 * Give the client one slot even if that would require
1791 * over-allocation--it is better than failure.
1793 scale_factor = max_t(unsigned int, 8, nn->nfsd_serv->sv_nrthreads);
1795 avail = clamp_t(unsigned long, avail, slotsize,
1796 total_avail/scale_factor);
1797 num = min_t(int, num, avail / slotsize);
1798 num = max_t(int, num, 1);
1799 nfsd_drc_mem_used += num * slotsize;
1800 spin_unlock(&nfsd_drc_lock);
1805 static void nfsd4_put_drc_mem(struct nfsd4_channel_attrs *ca)
1807 int slotsize = slot_bytes(ca);
1809 spin_lock(&nfsd_drc_lock);
1810 nfsd_drc_mem_used -= slotsize * ca->maxreqs;
1811 spin_unlock(&nfsd_drc_lock);
1814 static struct nfsd4_session *alloc_session(struct nfsd4_channel_attrs *fattrs,
1815 struct nfsd4_channel_attrs *battrs)
1817 int numslots = fattrs->maxreqs;
1818 int slotsize = slot_bytes(fattrs);
1819 struct nfsd4_session *new;
1822 BUILD_BUG_ON(struct_size(new, se_slots, NFSD_MAX_SLOTS_PER_SESSION)
1825 new = kzalloc(struct_size(new, se_slots, numslots), GFP_KERNEL);
1828 /* allocate each struct nfsd4_slot and data cache in one piece */
1829 for (i = 0; i < numslots; i++) {
1830 new->se_slots[i] = kzalloc(slotsize, GFP_KERNEL);
1831 if (!new->se_slots[i])
1835 memcpy(&new->se_fchannel, fattrs, sizeof(struct nfsd4_channel_attrs));
1836 memcpy(&new->se_bchannel, battrs, sizeof(struct nfsd4_channel_attrs));
1841 kfree(new->se_slots[i]);
1846 static void free_conn(struct nfsd4_conn *c)
1848 svc_xprt_put(c->cn_xprt);
1852 static void nfsd4_conn_lost(struct svc_xpt_user *u)
1854 struct nfsd4_conn *c = container_of(u, struct nfsd4_conn, cn_xpt_user);
1855 struct nfs4_client *clp = c->cn_session->se_client;
1857 trace_nfsd_cb_lost(clp);
1859 spin_lock(&clp->cl_lock);
1860 if (!list_empty(&c->cn_persession)) {
1861 list_del(&c->cn_persession);
1864 nfsd4_probe_callback(clp);
1865 spin_unlock(&clp->cl_lock);
1868 static struct nfsd4_conn *alloc_conn(struct svc_rqst *rqstp, u32 flags)
1870 struct nfsd4_conn *conn;
1872 conn = kmalloc(sizeof(struct nfsd4_conn), GFP_KERNEL);
1875 svc_xprt_get(rqstp->rq_xprt);
1876 conn->cn_xprt = rqstp->rq_xprt;
1877 conn->cn_flags = flags;
1878 INIT_LIST_HEAD(&conn->cn_xpt_user.list);
1882 static void __nfsd4_hash_conn(struct nfsd4_conn *conn, struct nfsd4_session *ses)
1884 conn->cn_session = ses;
1885 list_add(&conn->cn_persession, &ses->se_conns);
1888 static void nfsd4_hash_conn(struct nfsd4_conn *conn, struct nfsd4_session *ses)
1890 struct nfs4_client *clp = ses->se_client;
1892 spin_lock(&clp->cl_lock);
1893 __nfsd4_hash_conn(conn, ses);
1894 spin_unlock(&clp->cl_lock);
1897 static int nfsd4_register_conn(struct nfsd4_conn *conn)
1899 conn->cn_xpt_user.callback = nfsd4_conn_lost;
1900 return register_xpt_user(conn->cn_xprt, &conn->cn_xpt_user);
1903 static void nfsd4_init_conn(struct svc_rqst *rqstp, struct nfsd4_conn *conn, struct nfsd4_session *ses)
1907 nfsd4_hash_conn(conn, ses);
1908 ret = nfsd4_register_conn(conn);
1910 /* oops; xprt is already down: */
1911 nfsd4_conn_lost(&conn->cn_xpt_user);
1912 /* We may have gained or lost a callback channel: */
1913 nfsd4_probe_callback_sync(ses->se_client);
1916 static struct nfsd4_conn *alloc_conn_from_crses(struct svc_rqst *rqstp, struct nfsd4_create_session *cses)
1918 u32 dir = NFS4_CDFC4_FORE;
1920 if (cses->flags & SESSION4_BACK_CHAN)
1921 dir |= NFS4_CDFC4_BACK;
1922 return alloc_conn(rqstp, dir);
1925 /* must be called under client_lock */
1926 static void nfsd4_del_conns(struct nfsd4_session *s)
1928 struct nfs4_client *clp = s->se_client;
1929 struct nfsd4_conn *c;
1931 spin_lock(&clp->cl_lock);
1932 while (!list_empty(&s->se_conns)) {
1933 c = list_first_entry(&s->se_conns, struct nfsd4_conn, cn_persession);
1934 list_del_init(&c->cn_persession);
1935 spin_unlock(&clp->cl_lock);
1937 unregister_xpt_user(c->cn_xprt, &c->cn_xpt_user);
1940 spin_lock(&clp->cl_lock);
1942 spin_unlock(&clp->cl_lock);
1945 static void __free_session(struct nfsd4_session *ses)
1947 free_session_slots(ses);
1951 static void free_session(struct nfsd4_session *ses)
1953 nfsd4_del_conns(ses);
1954 nfsd4_put_drc_mem(&ses->se_fchannel);
1955 __free_session(ses);
1958 static void init_session(struct svc_rqst *rqstp, struct nfsd4_session *new, struct nfs4_client *clp, struct nfsd4_create_session *cses)
1961 struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
1963 new->se_client = clp;
1966 INIT_LIST_HEAD(&new->se_conns);
1968 new->se_cb_seq_nr = 1;
1969 new->se_flags = cses->flags;
1970 new->se_cb_prog = cses->callback_prog;
1971 new->se_cb_sec = cses->cb_sec;
1972 atomic_set(&new->se_ref, 0);
1973 idx = hash_sessionid(&new->se_sessionid);
1974 list_add(&new->se_hash, &nn->sessionid_hashtbl[idx]);
1975 spin_lock(&clp->cl_lock);
1976 list_add(&new->se_perclnt, &clp->cl_sessions);
1977 spin_unlock(&clp->cl_lock);
1980 struct sockaddr *sa = svc_addr(rqstp);
1982 * This is a little silly; with sessions there's no real
1983 * use for the callback address. Use the peer address
1984 * as a reasonable default for now, but consider fixing
1985 * the rpc client not to require an address in the
1988 rpc_copy_addr((struct sockaddr *)&clp->cl_cb_conn.cb_addr, sa);
1989 clp->cl_cb_conn.cb_addrlen = svc_addr_len(sa);
1993 /* caller must hold client_lock */
1994 static struct nfsd4_session *
1995 __find_in_sessionid_hashtbl(struct nfs4_sessionid *sessionid, struct net *net)
1997 struct nfsd4_session *elem;
1999 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
2001 lockdep_assert_held(&nn->client_lock);
2003 dump_sessionid(__func__, sessionid);
2004 idx = hash_sessionid(sessionid);
2005 /* Search in the appropriate list */
2006 list_for_each_entry(elem, &nn->sessionid_hashtbl[idx], se_hash) {
2007 if (!memcmp(elem->se_sessionid.data, sessionid->data,
2008 NFS4_MAX_SESSIONID_LEN)) {
2013 dprintk("%s: session not found\n", __func__);
2017 static struct nfsd4_session *
2018 find_in_sessionid_hashtbl(struct nfs4_sessionid *sessionid, struct net *net,
2021 struct nfsd4_session *session;
2022 __be32 status = nfserr_badsession;
2024 session = __find_in_sessionid_hashtbl(sessionid, net);
2027 status = nfsd4_get_session_locked(session);
2035 /* caller must hold client_lock */
2037 unhash_session(struct nfsd4_session *ses)
2039 struct nfs4_client *clp = ses->se_client;
2040 struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
2042 lockdep_assert_held(&nn->client_lock);
2044 list_del(&ses->se_hash);
2045 spin_lock(&ses->se_client->cl_lock);
2046 list_del(&ses->se_perclnt);
2047 spin_unlock(&ses->se_client->cl_lock);
2050 /* SETCLIENTID and SETCLIENTID_CONFIRM Helper functions */
2052 STALE_CLIENTID(clientid_t *clid, struct nfsd_net *nn)
2055 * We're assuming the clid was not given out from a boot
2056 * precisely 2^32 (about 136 years) before this one. That seems
2057 * a safe assumption:
2059 if (clid->cl_boot == (u32)nn->boot_time)
2061 trace_nfsd_clid_stale(clid);
2066 * XXX Should we use a slab cache ?
2067 * This type of memory management is somewhat inefficient, but we use it
2068 * anyway since SETCLIENTID is not a common operation.
2070 static struct nfs4_client *alloc_client(struct xdr_netobj name,
2071 struct nfsd_net *nn)
2073 struct nfs4_client *clp;
2076 if (atomic_read(&nn->nfs4_client_count) >= nn->nfs4_max_clients) {
2077 mod_delayed_work(laundry_wq, &nn->laundromat_work, 0);
2080 clp = kmem_cache_zalloc(client_slab, GFP_KERNEL);
2083 xdr_netobj_dup(&clp->cl_name, &name, GFP_KERNEL);
2084 if (clp->cl_name.data == NULL)
2086 clp->cl_ownerstr_hashtbl = kmalloc_array(OWNER_HASH_SIZE,
2087 sizeof(struct list_head),
2089 if (!clp->cl_ownerstr_hashtbl)
2090 goto err_no_hashtbl;
2091 for (i = 0; i < OWNER_HASH_SIZE; i++)
2092 INIT_LIST_HEAD(&clp->cl_ownerstr_hashtbl[i]);
2093 INIT_LIST_HEAD(&clp->cl_sessions);
2094 idr_init(&clp->cl_stateids);
2095 atomic_set(&clp->cl_rpc_users, 0);
2096 clp->cl_cb_state = NFSD4_CB_UNKNOWN;
2097 clp->cl_state = NFSD4_ACTIVE;
2098 atomic_inc(&nn->nfs4_client_count);
2099 atomic_set(&clp->cl_delegs_in_recall, 0);
2100 INIT_LIST_HEAD(&clp->cl_idhash);
2101 INIT_LIST_HEAD(&clp->cl_openowners);
2102 INIT_LIST_HEAD(&clp->cl_delegations);
2103 INIT_LIST_HEAD(&clp->cl_lru);
2104 INIT_LIST_HEAD(&clp->cl_revoked);
2105 #ifdef CONFIG_NFSD_PNFS
2106 INIT_LIST_HEAD(&clp->cl_lo_states);
2108 INIT_LIST_HEAD(&clp->async_copies);
2109 spin_lock_init(&clp->async_lock);
2110 spin_lock_init(&clp->cl_lock);
2111 rpc_init_wait_queue(&clp->cl_cb_waitq, "Backchannel slot table");
2114 kfree(clp->cl_name.data);
2116 kmem_cache_free(client_slab, clp);
2120 static void __free_client(struct kref *k)
2122 struct nfsdfs_client *c = container_of(k, struct nfsdfs_client, cl_ref);
2123 struct nfs4_client *clp = container_of(c, struct nfs4_client, cl_nfsdfs);
2125 free_svc_cred(&clp->cl_cred);
2126 kfree(clp->cl_ownerstr_hashtbl);
2127 kfree(clp->cl_name.data);
2128 kfree(clp->cl_nii_domain.data);
2129 kfree(clp->cl_nii_name.data);
2130 idr_destroy(&clp->cl_stateids);
2132 kmem_cache_free(client_slab, clp);
2135 static void drop_client(struct nfs4_client *clp)
2137 kref_put(&clp->cl_nfsdfs.cl_ref, __free_client);
2141 free_client(struct nfs4_client *clp)
2143 while (!list_empty(&clp->cl_sessions)) {
2144 struct nfsd4_session *ses;
2145 ses = list_entry(clp->cl_sessions.next, struct nfsd4_session,
2147 list_del(&ses->se_perclnt);
2148 WARN_ON_ONCE(atomic_read(&ses->se_ref));
2151 rpc_destroy_wait_queue(&clp->cl_cb_waitq);
2152 if (clp->cl_nfsd_dentry) {
2153 nfsd_client_rmdir(clp->cl_nfsd_dentry);
2154 clp->cl_nfsd_dentry = NULL;
2155 wake_up_all(&expiry_wq);
2160 /* must be called under the client_lock */
2162 unhash_client_locked(struct nfs4_client *clp)
2164 struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
2165 struct nfsd4_session *ses;
2167 lockdep_assert_held(&nn->client_lock);
2169 /* Mark the client as expired! */
2171 /* Make it invisible */
2172 if (!list_empty(&clp->cl_idhash)) {
2173 list_del_init(&clp->cl_idhash);
2174 if (test_bit(NFSD4_CLIENT_CONFIRMED, &clp->cl_flags))
2175 rb_erase(&clp->cl_namenode, &nn->conf_name_tree);
2177 rb_erase(&clp->cl_namenode, &nn->unconf_name_tree);
2179 list_del_init(&clp->cl_lru);
2180 spin_lock(&clp->cl_lock);
2181 list_for_each_entry(ses, &clp->cl_sessions, se_perclnt)
2182 list_del_init(&ses->se_hash);
2183 spin_unlock(&clp->cl_lock);
2187 unhash_client(struct nfs4_client *clp)
2189 struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
2191 spin_lock(&nn->client_lock);
2192 unhash_client_locked(clp);
2193 spin_unlock(&nn->client_lock);
2196 static __be32 mark_client_expired_locked(struct nfs4_client *clp)
2198 if (atomic_read(&clp->cl_rpc_users))
2199 return nfserr_jukebox;
2200 unhash_client_locked(clp);
2205 __destroy_client(struct nfs4_client *clp)
2207 struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
2209 struct nfs4_openowner *oo;
2210 struct nfs4_delegation *dp;
2211 struct list_head reaplist;
2213 INIT_LIST_HEAD(&reaplist);
2214 spin_lock(&state_lock);
2215 while (!list_empty(&clp->cl_delegations)) {
2216 dp = list_entry(clp->cl_delegations.next, struct nfs4_delegation, dl_perclnt);
2217 WARN_ON(!unhash_delegation_locked(dp));
2218 list_add(&dp->dl_recall_lru, &reaplist);
2220 spin_unlock(&state_lock);
2221 while (!list_empty(&reaplist)) {
2222 dp = list_entry(reaplist.next, struct nfs4_delegation, dl_recall_lru);
2223 list_del_init(&dp->dl_recall_lru);
2224 destroy_unhashed_deleg(dp);
2226 while (!list_empty(&clp->cl_revoked)) {
2227 dp = list_entry(clp->cl_revoked.next, struct nfs4_delegation, dl_recall_lru);
2228 list_del_init(&dp->dl_recall_lru);
2229 nfs4_put_stid(&dp->dl_stid);
2231 while (!list_empty(&clp->cl_openowners)) {
2232 oo = list_entry(clp->cl_openowners.next, struct nfs4_openowner, oo_perclient);
2233 nfs4_get_stateowner(&oo->oo_owner);
2234 release_openowner(oo);
2236 for (i = 0; i < OWNER_HASH_SIZE; i++) {
2237 struct nfs4_stateowner *so, *tmp;
2239 list_for_each_entry_safe(so, tmp, &clp->cl_ownerstr_hashtbl[i],
2241 /* Should be no openowners at this point */
2242 WARN_ON_ONCE(so->so_is_open_owner);
2243 remove_blocked_locks(lockowner(so));
2246 nfsd4_return_all_client_layouts(clp);
2247 nfsd4_shutdown_copy(clp);
2248 nfsd4_shutdown_callback(clp);
2249 if (clp->cl_cb_conn.cb_xprt)
2250 svc_xprt_put(clp->cl_cb_conn.cb_xprt);
2251 atomic_add_unless(&nn->nfs4_client_count, -1, 0);
2252 nfsd4_dec_courtesy_client_count(nn, clp);
2254 wake_up_all(&expiry_wq);
2258 destroy_client(struct nfs4_client *clp)
2261 __destroy_client(clp);
2264 static void inc_reclaim_complete(struct nfs4_client *clp)
2266 struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
2268 if (!nn->track_reclaim_completes)
2270 if (!nfsd4_find_reclaim_client(clp->cl_name, nn))
2272 if (atomic_inc_return(&nn->nr_reclaim_complete) ==
2273 nn->reclaim_str_hashtbl_size) {
2274 printk(KERN_INFO "NFSD: all clients done reclaiming, ending NFSv4 grace period (net %x)\n",
2276 nfsd4_end_grace(nn);
2280 static void expire_client(struct nfs4_client *clp)
2283 nfsd4_client_record_remove(clp);
2284 __destroy_client(clp);
2287 static void copy_verf(struct nfs4_client *target, nfs4_verifier *source)
2289 memcpy(target->cl_verifier.data, source->data,
2290 sizeof(target->cl_verifier.data));
2293 static void copy_clid(struct nfs4_client *target, struct nfs4_client *source)
2295 target->cl_clientid.cl_boot = source->cl_clientid.cl_boot;
2296 target->cl_clientid.cl_id = source->cl_clientid.cl_id;
2299 static int copy_cred(struct svc_cred *target, struct svc_cred *source)
2301 target->cr_principal = kstrdup(source->cr_principal, GFP_KERNEL);
2302 target->cr_raw_principal = kstrdup(source->cr_raw_principal,
2304 target->cr_targ_princ = kstrdup(source->cr_targ_princ, GFP_KERNEL);
2305 if ((source->cr_principal && !target->cr_principal) ||
2306 (source->cr_raw_principal && !target->cr_raw_principal) ||
2307 (source->cr_targ_princ && !target->cr_targ_princ))
2310 target->cr_flavor = source->cr_flavor;
2311 target->cr_uid = source->cr_uid;
2312 target->cr_gid = source->cr_gid;
2313 target->cr_group_info = source->cr_group_info;
2314 get_group_info(target->cr_group_info);
2315 target->cr_gss_mech = source->cr_gss_mech;
2316 if (source->cr_gss_mech)
2317 gss_mech_get(source->cr_gss_mech);
2322 compare_blob(const struct xdr_netobj *o1, const struct xdr_netobj *o2)
2324 if (o1->len < o2->len)
2326 if (o1->len > o2->len)
2328 return memcmp(o1->data, o2->data, o1->len);
2332 same_verf(nfs4_verifier *v1, nfs4_verifier *v2)
2334 return 0 == memcmp(v1->data, v2->data, sizeof(v1->data));
2338 same_clid(clientid_t *cl1, clientid_t *cl2)
2340 return (cl1->cl_boot == cl2->cl_boot) && (cl1->cl_id == cl2->cl_id);
2343 static bool groups_equal(struct group_info *g1, struct group_info *g2)
2347 if (g1->ngroups != g2->ngroups)
2349 for (i=0; i<g1->ngroups; i++)
2350 if (!gid_eq(g1->gid[i], g2->gid[i]))
2356 * RFC 3530 language requires clid_inuse be returned when the
2357 * "principal" associated with a requests differs from that previously
2358 * used. We use uid, gid's, and gss principal string as our best
2359 * approximation. We also don't want to allow non-gss use of a client
2360 * established using gss: in theory cr_principal should catch that
2361 * change, but in practice cr_principal can be null even in the gss case
2362 * since gssd doesn't always pass down a principal string.
2364 static bool is_gss_cred(struct svc_cred *cr)
2366 /* Is cr_flavor one of the gss "pseudoflavors"?: */
2367 return (cr->cr_flavor > RPC_AUTH_MAXFLAVOR);
2372 same_creds(struct svc_cred *cr1, struct svc_cred *cr2)
2374 if ((is_gss_cred(cr1) != is_gss_cred(cr2))
2375 || (!uid_eq(cr1->cr_uid, cr2->cr_uid))
2376 || (!gid_eq(cr1->cr_gid, cr2->cr_gid))
2377 || !groups_equal(cr1->cr_group_info, cr2->cr_group_info))
2379 /* XXX: check that cr_targ_princ fields match ? */
2380 if (cr1->cr_principal == cr2->cr_principal)
2382 if (!cr1->cr_principal || !cr2->cr_principal)
2384 return 0 == strcmp(cr1->cr_principal, cr2->cr_principal);
2387 static bool svc_rqst_integrity_protected(struct svc_rqst *rqstp)
2389 struct svc_cred *cr = &rqstp->rq_cred;
2392 if (!cr->cr_gss_mech)
2394 service = gss_pseudoflavor_to_service(cr->cr_gss_mech, cr->cr_flavor);
2395 return service == RPC_GSS_SVC_INTEGRITY ||
2396 service == RPC_GSS_SVC_PRIVACY;
2399 bool nfsd4_mach_creds_match(struct nfs4_client *cl, struct svc_rqst *rqstp)
2401 struct svc_cred *cr = &rqstp->rq_cred;
2403 if (!cl->cl_mach_cred)
2405 if (cl->cl_cred.cr_gss_mech != cr->cr_gss_mech)
2407 if (!svc_rqst_integrity_protected(rqstp))
2409 if (cl->cl_cred.cr_raw_principal)
2410 return 0 == strcmp(cl->cl_cred.cr_raw_principal,
2411 cr->cr_raw_principal);
2412 if (!cr->cr_principal)
2414 return 0 == strcmp(cl->cl_cred.cr_principal, cr->cr_principal);
2417 static void gen_confirm(struct nfs4_client *clp, struct nfsd_net *nn)
2422 * This is opaque to client, so no need to byte-swap. Use
2423 * __force to keep sparse happy
2425 verf[0] = (__force __be32)(u32)ktime_get_real_seconds();
2426 verf[1] = (__force __be32)nn->clverifier_counter++;
2427 memcpy(clp->cl_confirm.data, verf, sizeof(clp->cl_confirm.data));
2430 static void gen_clid(struct nfs4_client *clp, struct nfsd_net *nn)
2432 clp->cl_clientid.cl_boot = (u32)nn->boot_time;
2433 clp->cl_clientid.cl_id = nn->clientid_counter++;
2434 gen_confirm(clp, nn);
2437 static struct nfs4_stid *
2438 find_stateid_locked(struct nfs4_client *cl, stateid_t *t)
2440 struct nfs4_stid *ret;
2442 ret = idr_find(&cl->cl_stateids, t->si_opaque.so_id);
2443 if (!ret || !ret->sc_type)
2448 static struct nfs4_stid *
2449 find_stateid_by_type(struct nfs4_client *cl, stateid_t *t, char typemask)
2451 struct nfs4_stid *s;
2453 spin_lock(&cl->cl_lock);
2454 s = find_stateid_locked(cl, t);
2456 if (typemask & s->sc_type)
2457 refcount_inc(&s->sc_count);
2461 spin_unlock(&cl->cl_lock);
2465 static struct nfs4_client *get_nfsdfs_clp(struct inode *inode)
2467 struct nfsdfs_client *nc;
2468 nc = get_nfsdfs_client(inode);
2471 return container_of(nc, struct nfs4_client, cl_nfsdfs);
2474 static void seq_quote_mem(struct seq_file *m, char *data, int len)
2476 seq_printf(m, "\"");
2477 seq_escape_mem(m, data, len, ESCAPE_HEX | ESCAPE_NAP | ESCAPE_APPEND, "\"\\");
2478 seq_printf(m, "\"");
2481 static const char *cb_state2str(int state)
2486 case NFSD4_CB_UNKNOWN:
2490 case NFSD4_CB_FAULT:
2496 static int client_info_show(struct seq_file *m, void *v)
2498 struct inode *inode = file_inode(m->file);
2499 struct nfs4_client *clp;
2502 clp = get_nfsdfs_clp(inode);
2505 memcpy(&clid, &clp->cl_clientid, sizeof(clid));
2506 seq_printf(m, "clientid: 0x%llx\n", clid);
2507 seq_printf(m, "address: \"%pISpc\"\n", (struct sockaddr *)&clp->cl_addr);
2509 if (clp->cl_state == NFSD4_COURTESY)
2510 seq_puts(m, "status: courtesy\n");
2511 else if (clp->cl_state == NFSD4_EXPIRABLE)
2512 seq_puts(m, "status: expirable\n");
2513 else if (test_bit(NFSD4_CLIENT_CONFIRMED, &clp->cl_flags))
2514 seq_puts(m, "status: confirmed\n");
2516 seq_puts(m, "status: unconfirmed\n");
2517 seq_printf(m, "seconds from last renew: %lld\n",
2518 ktime_get_boottime_seconds() - clp->cl_time);
2519 seq_printf(m, "name: ");
2520 seq_quote_mem(m, clp->cl_name.data, clp->cl_name.len);
2521 seq_printf(m, "\nminor version: %d\n", clp->cl_minorversion);
2522 if (clp->cl_nii_domain.data) {
2523 seq_printf(m, "Implementation domain: ");
2524 seq_quote_mem(m, clp->cl_nii_domain.data,
2525 clp->cl_nii_domain.len);
2526 seq_printf(m, "\nImplementation name: ");
2527 seq_quote_mem(m, clp->cl_nii_name.data, clp->cl_nii_name.len);
2528 seq_printf(m, "\nImplementation time: [%lld, %ld]\n",
2529 clp->cl_nii_time.tv_sec, clp->cl_nii_time.tv_nsec);
2531 seq_printf(m, "callback state: %s\n", cb_state2str(clp->cl_cb_state));
2532 seq_printf(m, "callback address: %pISpc\n", &clp->cl_cb_conn.cb_addr);
2538 DEFINE_SHOW_ATTRIBUTE(client_info);
2540 static void *states_start(struct seq_file *s, loff_t *pos)
2541 __acquires(&clp->cl_lock)
2543 struct nfs4_client *clp = s->private;
2544 unsigned long id = *pos;
2547 spin_lock(&clp->cl_lock);
2548 ret = idr_get_next_ul(&clp->cl_stateids, &id);
2553 static void *states_next(struct seq_file *s, void *v, loff_t *pos)
2555 struct nfs4_client *clp = s->private;
2556 unsigned long id = *pos;
2561 ret = idr_get_next_ul(&clp->cl_stateids, &id);
2566 static void states_stop(struct seq_file *s, void *v)
2567 __releases(&clp->cl_lock)
2569 struct nfs4_client *clp = s->private;
2571 spin_unlock(&clp->cl_lock);
2574 static void nfs4_show_fname(struct seq_file *s, struct nfsd_file *f)
2576 seq_printf(s, "filename: \"%pD2\"", f->nf_file);
2579 static void nfs4_show_superblock(struct seq_file *s, struct nfsd_file *f)
2581 struct inode *inode = file_inode(f->nf_file);
2583 seq_printf(s, "superblock: \"%02x:%02x:%ld\"",
2584 MAJOR(inode->i_sb->s_dev),
2585 MINOR(inode->i_sb->s_dev),
2589 static void nfs4_show_owner(struct seq_file *s, struct nfs4_stateowner *oo)
2591 seq_printf(s, "owner: ");
2592 seq_quote_mem(s, oo->so_owner.data, oo->so_owner.len);
2595 static void nfs4_show_stateid(struct seq_file *s, stateid_t *stid)
2597 seq_printf(s, "0x%.8x", stid->si_generation);
2598 seq_printf(s, "%12phN", &stid->si_opaque);
2601 static int nfs4_show_open(struct seq_file *s, struct nfs4_stid *st)
2603 struct nfs4_ol_stateid *ols;
2604 struct nfs4_file *nf;
2605 struct nfsd_file *file;
2606 struct nfs4_stateowner *oo;
2607 unsigned int access, deny;
2609 if (st->sc_type != NFS4_OPEN_STID && st->sc_type != NFS4_LOCK_STID)
2610 return 0; /* XXX: or SEQ_SKIP? */
2611 ols = openlockstateid(st);
2612 oo = ols->st_stateowner;
2615 spin_lock(&nf->fi_lock);
2616 file = find_any_file_locked(nf);
2620 seq_printf(s, "- ");
2621 nfs4_show_stateid(s, &st->sc_stateid);
2622 seq_printf(s, ": { type: open, ");
2624 access = bmap_to_share_mode(ols->st_access_bmap);
2625 deny = bmap_to_share_mode(ols->st_deny_bmap);
2627 seq_printf(s, "access: %s%s, ",
2628 access & NFS4_SHARE_ACCESS_READ ? "r" : "-",
2629 access & NFS4_SHARE_ACCESS_WRITE ? "w" : "-");
2630 seq_printf(s, "deny: %s%s, ",
2631 deny & NFS4_SHARE_ACCESS_READ ? "r" : "-",
2632 deny & NFS4_SHARE_ACCESS_WRITE ? "w" : "-");
2634 nfs4_show_superblock(s, file);
2635 seq_printf(s, ", ");
2636 nfs4_show_fname(s, file);
2637 seq_printf(s, ", ");
2638 nfs4_show_owner(s, oo);
2639 seq_printf(s, " }\n");
2641 spin_unlock(&nf->fi_lock);
2645 static int nfs4_show_lock(struct seq_file *s, struct nfs4_stid *st)
2647 struct nfs4_ol_stateid *ols;
2648 struct nfs4_file *nf;
2649 struct nfsd_file *file;
2650 struct nfs4_stateowner *oo;
2652 ols = openlockstateid(st);
2653 oo = ols->st_stateowner;
2655 spin_lock(&nf->fi_lock);
2656 file = find_any_file_locked(nf);
2660 seq_printf(s, "- ");
2661 nfs4_show_stateid(s, &st->sc_stateid);
2662 seq_printf(s, ": { type: lock, ");
2665 * Note: a lock stateid isn't really the same thing as a lock,
2666 * it's the locking state held by one owner on a file, and there
2667 * may be multiple (or no) lock ranges associated with it.
2668 * (Same for the matter is true of open stateids.)
2671 nfs4_show_superblock(s, file);
2672 /* XXX: open stateid? */
2673 seq_printf(s, ", ");
2674 nfs4_show_fname(s, file);
2675 seq_printf(s, ", ");
2676 nfs4_show_owner(s, oo);
2677 seq_printf(s, " }\n");
2679 spin_unlock(&nf->fi_lock);
2683 static int nfs4_show_deleg(struct seq_file *s, struct nfs4_stid *st)
2685 struct nfs4_delegation *ds;
2686 struct nfs4_file *nf;
2687 struct nfsd_file *file;
2689 ds = delegstateid(st);
2691 spin_lock(&nf->fi_lock);
2692 file = nf->fi_deleg_file;
2696 seq_printf(s, "- ");
2697 nfs4_show_stateid(s, &st->sc_stateid);
2698 seq_printf(s, ": { type: deleg, ");
2700 /* Kinda dead code as long as we only support read delegs: */
2701 seq_printf(s, "access: %s, ",
2702 ds->dl_type == NFS4_OPEN_DELEGATE_READ ? "r" : "w");
2704 /* XXX: lease time, whether it's being recalled. */
2706 nfs4_show_superblock(s, file);
2707 seq_printf(s, ", ");
2708 nfs4_show_fname(s, file);
2709 seq_printf(s, " }\n");
2711 spin_unlock(&nf->fi_lock);
2715 static int nfs4_show_layout(struct seq_file *s, struct nfs4_stid *st)
2717 struct nfs4_layout_stateid *ls;
2718 struct nfsd_file *file;
2720 ls = container_of(st, struct nfs4_layout_stateid, ls_stid);
2723 seq_printf(s, "- ");
2724 nfs4_show_stateid(s, &st->sc_stateid);
2725 seq_printf(s, ": { type: layout, ");
2727 /* XXX: What else would be useful? */
2729 nfs4_show_superblock(s, file);
2730 seq_printf(s, ", ");
2731 nfs4_show_fname(s, file);
2732 seq_printf(s, " }\n");
2737 static int states_show(struct seq_file *s, void *v)
2739 struct nfs4_stid *st = v;
2741 switch (st->sc_type) {
2742 case NFS4_OPEN_STID:
2743 return nfs4_show_open(s, st);
2744 case NFS4_LOCK_STID:
2745 return nfs4_show_lock(s, st);
2746 case NFS4_DELEG_STID:
2747 return nfs4_show_deleg(s, st);
2748 case NFS4_LAYOUT_STID:
2749 return nfs4_show_layout(s, st);
2751 return 0; /* XXX: or SEQ_SKIP? */
2753 /* XXX: copy stateids? */
2756 static struct seq_operations states_seq_ops = {
2757 .start = states_start,
2758 .next = states_next,
2759 .stop = states_stop,
2763 static int client_states_open(struct inode *inode, struct file *file)
2766 struct nfs4_client *clp;
2769 clp = get_nfsdfs_clp(inode);
2773 ret = seq_open(file, &states_seq_ops);
2776 s = file->private_data;
2781 static int client_opens_release(struct inode *inode, struct file *file)
2783 struct seq_file *m = file->private_data;
2784 struct nfs4_client *clp = m->private;
2786 /* XXX: alternatively, we could get/drop in seq start/stop */
2791 static const struct file_operations client_states_fops = {
2792 .open = client_states_open,
2794 .llseek = seq_lseek,
2795 .release = client_opens_release,
2799 * Normally we refuse to destroy clients that are in use, but here the
2800 * administrator is telling us to just do it. We also want to wait
2801 * so the caller has a guarantee that the client's locks are gone by
2802 * the time the write returns:
2804 static void force_expire_client(struct nfs4_client *clp)
2806 struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
2807 bool already_expired;
2809 trace_nfsd_clid_admin_expired(&clp->cl_clientid);
2811 spin_lock(&nn->client_lock);
2813 spin_unlock(&nn->client_lock);
2815 wait_event(expiry_wq, atomic_read(&clp->cl_rpc_users) == 0);
2816 spin_lock(&nn->client_lock);
2817 already_expired = list_empty(&clp->cl_lru);
2818 if (!already_expired)
2819 unhash_client_locked(clp);
2820 spin_unlock(&nn->client_lock);
2822 if (!already_expired)
2825 wait_event(expiry_wq, clp->cl_nfsd_dentry == NULL);
2828 static ssize_t client_ctl_write(struct file *file, const char __user *buf,
2829 size_t size, loff_t *pos)
2832 struct nfs4_client *clp;
2834 data = simple_transaction_get(file, buf, size);
2836 return PTR_ERR(data);
2837 if (size != 7 || 0 != memcmp(data, "expire\n", 7))
2839 clp = get_nfsdfs_clp(file_inode(file));
2842 force_expire_client(clp);
2847 static const struct file_operations client_ctl_fops = {
2848 .write = client_ctl_write,
2849 .release = simple_transaction_release,
2852 static const struct tree_descr client_files[] = {
2853 [0] = {"info", &client_info_fops, S_IRUSR},
2854 [1] = {"states", &client_states_fops, S_IRUSR},
2855 [2] = {"ctl", &client_ctl_fops, S_IWUSR},
2860 nfsd4_cb_recall_any_done(struct nfsd4_callback *cb,
2861 struct rpc_task *task)
2863 trace_nfsd_cb_recall_any_done(cb, task);
2864 switch (task->tk_status) {
2865 case -NFS4ERR_DELAY:
2866 rpc_delay(task, 2 * HZ);
2874 nfsd4_cb_recall_any_release(struct nfsd4_callback *cb)
2876 struct nfs4_client *clp = cb->cb_clp;
2877 struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
2879 spin_lock(&nn->client_lock);
2880 clear_bit(NFSD4_CLIENT_CB_RECALL_ANY, &clp->cl_flags);
2881 put_client_renew_locked(clp);
2882 spin_unlock(&nn->client_lock);
2885 static const struct nfsd4_callback_ops nfsd4_cb_recall_any_ops = {
2886 .done = nfsd4_cb_recall_any_done,
2887 .release = nfsd4_cb_recall_any_release,
2890 static struct nfs4_client *create_client(struct xdr_netobj name,
2891 struct svc_rqst *rqstp, nfs4_verifier *verf)
2893 struct nfs4_client *clp;
2894 struct sockaddr *sa = svc_addr(rqstp);
2896 struct net *net = SVC_NET(rqstp);
2897 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
2898 struct dentry *dentries[ARRAY_SIZE(client_files)];
2900 clp = alloc_client(name, nn);
2904 ret = copy_cred(&clp->cl_cred, &rqstp->rq_cred);
2910 kref_init(&clp->cl_nfsdfs.cl_ref);
2911 nfsd4_init_cb(&clp->cl_cb_null, clp, NULL, NFSPROC4_CLNT_CB_NULL);
2912 clp->cl_time = ktime_get_boottime_seconds();
2913 clear_bit(0, &clp->cl_cb_slot_busy);
2914 copy_verf(clp, verf);
2915 memcpy(&clp->cl_addr, sa, sizeof(struct sockaddr_storage));
2916 clp->cl_cb_session = NULL;
2918 clp->cl_nfsd_dentry = nfsd_client_mkdir(
2919 nn, &clp->cl_nfsdfs,
2920 clp->cl_clientid.cl_id - nn->clientid_base,
2921 client_files, dentries);
2922 clp->cl_nfsd_info_dentry = dentries[0];
2923 if (!clp->cl_nfsd_dentry) {
2927 clp->cl_ra = kzalloc(sizeof(*clp->cl_ra), GFP_KERNEL);
2932 clp->cl_ra_time = 0;
2933 nfsd4_init_cb(&clp->cl_ra->ra_cb, clp, &nfsd4_cb_recall_any_ops,
2934 NFSPROC4_CLNT_CB_RECALL_ANY);
2939 add_clp_to_name_tree(struct nfs4_client *new_clp, struct rb_root *root)
2941 struct rb_node **new = &(root->rb_node), *parent = NULL;
2942 struct nfs4_client *clp;
2945 clp = rb_entry(*new, struct nfs4_client, cl_namenode);
2948 if (compare_blob(&clp->cl_name, &new_clp->cl_name) > 0)
2949 new = &((*new)->rb_left);
2951 new = &((*new)->rb_right);
2954 rb_link_node(&new_clp->cl_namenode, parent, new);
2955 rb_insert_color(&new_clp->cl_namenode, root);
2958 static struct nfs4_client *
2959 find_clp_in_name_tree(struct xdr_netobj *name, struct rb_root *root)
2962 struct rb_node *node = root->rb_node;
2963 struct nfs4_client *clp;
2966 clp = rb_entry(node, struct nfs4_client, cl_namenode);
2967 cmp = compare_blob(&clp->cl_name, name);
2969 node = node->rb_left;
2971 node = node->rb_right;
2979 add_to_unconfirmed(struct nfs4_client *clp)
2981 unsigned int idhashval;
2982 struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
2984 lockdep_assert_held(&nn->client_lock);
2986 clear_bit(NFSD4_CLIENT_CONFIRMED, &clp->cl_flags);
2987 add_clp_to_name_tree(clp, &nn->unconf_name_tree);
2988 idhashval = clientid_hashval(clp->cl_clientid.cl_id);
2989 list_add(&clp->cl_idhash, &nn->unconf_id_hashtbl[idhashval]);
2990 renew_client_locked(clp);
2994 move_to_confirmed(struct nfs4_client *clp)
2996 unsigned int idhashval = clientid_hashval(clp->cl_clientid.cl_id);
2997 struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
2999 lockdep_assert_held(&nn->client_lock);
3001 list_move(&clp->cl_idhash, &nn->conf_id_hashtbl[idhashval]);
3002 rb_erase(&clp->cl_namenode, &nn->unconf_name_tree);
3003 add_clp_to_name_tree(clp, &nn->conf_name_tree);
3004 set_bit(NFSD4_CLIENT_CONFIRMED, &clp->cl_flags);
3005 trace_nfsd_clid_confirmed(&clp->cl_clientid);
3006 renew_client_locked(clp);
3009 static struct nfs4_client *
3010 find_client_in_id_table(struct list_head *tbl, clientid_t *clid, bool sessions)
3012 struct nfs4_client *clp;
3013 unsigned int idhashval = clientid_hashval(clid->cl_id);
3015 list_for_each_entry(clp, &tbl[idhashval], cl_idhash) {
3016 if (same_clid(&clp->cl_clientid, clid)) {
3017 if ((bool)clp->cl_minorversion != sessions)
3019 renew_client_locked(clp);
3026 static struct nfs4_client *
3027 find_confirmed_client(clientid_t *clid, bool sessions, struct nfsd_net *nn)
3029 struct list_head *tbl = nn->conf_id_hashtbl;
3031 lockdep_assert_held(&nn->client_lock);
3032 return find_client_in_id_table(tbl, clid, sessions);
3035 static struct nfs4_client *
3036 find_unconfirmed_client(clientid_t *clid, bool sessions, struct nfsd_net *nn)
3038 struct list_head *tbl = nn->unconf_id_hashtbl;
3040 lockdep_assert_held(&nn->client_lock);
3041 return find_client_in_id_table(tbl, clid, sessions);
3044 static bool clp_used_exchangeid(struct nfs4_client *clp)
3046 return clp->cl_exchange_flags != 0;
3049 static struct nfs4_client *
3050 find_confirmed_client_by_name(struct xdr_netobj *name, struct nfsd_net *nn)
3052 lockdep_assert_held(&nn->client_lock);
3053 return find_clp_in_name_tree(name, &nn->conf_name_tree);
3056 static struct nfs4_client *
3057 find_unconfirmed_client_by_name(struct xdr_netobj *name, struct nfsd_net *nn)
3059 lockdep_assert_held(&nn->client_lock);
3060 return find_clp_in_name_tree(name, &nn->unconf_name_tree);
3064 gen_callback(struct nfs4_client *clp, struct nfsd4_setclientid *se, struct svc_rqst *rqstp)
3066 struct nfs4_cb_conn *conn = &clp->cl_cb_conn;
3067 struct sockaddr *sa = svc_addr(rqstp);
3068 u32 scopeid = rpc_get_scope_id(sa);
3069 unsigned short expected_family;
3071 /* Currently, we only support tcp and tcp6 for the callback channel */
3072 if (se->se_callback_netid_len == 3 &&
3073 !memcmp(se->se_callback_netid_val, "tcp", 3))
3074 expected_family = AF_INET;
3075 else if (se->se_callback_netid_len == 4 &&
3076 !memcmp(se->se_callback_netid_val, "tcp6", 4))
3077 expected_family = AF_INET6;
3081 conn->cb_addrlen = rpc_uaddr2sockaddr(clp->net, se->se_callback_addr_val,
3082 se->se_callback_addr_len,
3083 (struct sockaddr *)&conn->cb_addr,
3084 sizeof(conn->cb_addr));
3086 if (!conn->cb_addrlen || conn->cb_addr.ss_family != expected_family)
3089 if (conn->cb_addr.ss_family == AF_INET6)
3090 ((struct sockaddr_in6 *)&conn->cb_addr)->sin6_scope_id = scopeid;
3092 conn->cb_prog = se->se_callback_prog;
3093 conn->cb_ident = se->se_callback_ident;
3094 memcpy(&conn->cb_saddr, &rqstp->rq_daddr, rqstp->rq_daddrlen);
3095 trace_nfsd_cb_args(clp, conn);
3098 conn->cb_addr.ss_family = AF_UNSPEC;
3099 conn->cb_addrlen = 0;
3100 trace_nfsd_cb_nodelegs(clp);
3105 * Cache a reply. nfsd4_check_resp_size() has bounded the cache size.
3108 nfsd4_store_cache_entry(struct nfsd4_compoundres *resp)
3110 struct xdr_buf *buf = resp->xdr->buf;
3111 struct nfsd4_slot *slot = resp->cstate.slot;
3114 dprintk("--> %s slot %p\n", __func__, slot);
3116 slot->sl_flags |= NFSD4_SLOT_INITIALIZED;
3117 slot->sl_opcnt = resp->opcnt;
3118 slot->sl_status = resp->cstate.status;
3119 free_svc_cred(&slot->sl_cred);
3120 copy_cred(&slot->sl_cred, &resp->rqstp->rq_cred);
3122 if (!nfsd4_cache_this(resp)) {
3123 slot->sl_flags &= ~NFSD4_SLOT_CACHED;
3126 slot->sl_flags |= NFSD4_SLOT_CACHED;
3128 base = resp->cstate.data_offset;
3129 slot->sl_datalen = buf->len - base;
3130 if (read_bytes_from_xdr_buf(buf, base, slot->sl_data, slot->sl_datalen))
3131 WARN(1, "%s: sessions DRC could not cache compound\n",
3137 * Encode the replay sequence operation from the slot values.
3138 * If cachethis is FALSE encode the uncached rep error on the next
3139 * operation which sets resp->p and increments resp->opcnt for
3140 * nfs4svc_encode_compoundres.
3144 nfsd4_enc_sequence_replay(struct nfsd4_compoundargs *args,
3145 struct nfsd4_compoundres *resp)
3147 struct nfsd4_op *op;
3148 struct nfsd4_slot *slot = resp->cstate.slot;
3150 /* Encode the replayed sequence operation */
3151 op = &args->ops[resp->opcnt - 1];
3152 nfsd4_encode_operation(resp, op);
3154 if (slot->sl_flags & NFSD4_SLOT_CACHED)
3156 if (args->opcnt == 1) {
3158 * The original operation wasn't a solo sequence--we
3159 * always cache those--so this retry must not match the
3162 op->status = nfserr_seq_false_retry;
3164 op = &args->ops[resp->opcnt++];
3165 op->status = nfserr_retry_uncached_rep;
3166 nfsd4_encode_operation(resp, op);
3172 * The sequence operation is not cached because we can use the slot and
3176 nfsd4_replay_cache_entry(struct nfsd4_compoundres *resp,
3177 struct nfsd4_sequence *seq)
3179 struct nfsd4_slot *slot = resp->cstate.slot;
3180 struct xdr_stream *xdr = resp->xdr;
3184 dprintk("--> %s slot %p\n", __func__, slot);
3186 status = nfsd4_enc_sequence_replay(resp->rqstp->rq_argp, resp);
3190 p = xdr_reserve_space(xdr, slot->sl_datalen);
3193 return nfserr_serverfault;
3195 xdr_encode_opaque_fixed(p, slot->sl_data, slot->sl_datalen);
3196 xdr_commit_encode(xdr);
3198 resp->opcnt = slot->sl_opcnt;
3199 return slot->sl_status;
3203 * Set the exchange_id flags returned by the server.
3206 nfsd4_set_ex_flags(struct nfs4_client *new, struct nfsd4_exchange_id *clid)
3208 #ifdef CONFIG_NFSD_PNFS
3209 new->cl_exchange_flags |= EXCHGID4_FLAG_USE_PNFS_MDS;
3211 new->cl_exchange_flags |= EXCHGID4_FLAG_USE_NON_PNFS;
3214 /* Referrals are supported, Migration is not. */
3215 new->cl_exchange_flags |= EXCHGID4_FLAG_SUPP_MOVED_REFER;
3217 /* set the wire flags to return to client. */
3218 clid->flags = new->cl_exchange_flags;
3221 static bool client_has_openowners(struct nfs4_client *clp)
3223 struct nfs4_openowner *oo;
3225 list_for_each_entry(oo, &clp->cl_openowners, oo_perclient) {
3226 if (!list_empty(&oo->oo_owner.so_stateids))
3232 static bool client_has_state(struct nfs4_client *clp)
3234 return client_has_openowners(clp)
3235 #ifdef CONFIG_NFSD_PNFS
3236 || !list_empty(&clp->cl_lo_states)
3238 || !list_empty(&clp->cl_delegations)
3239 || !list_empty(&clp->cl_sessions)
3240 || !list_empty(&clp->async_copies);
3243 static __be32 copy_impl_id(struct nfs4_client *clp,
3244 struct nfsd4_exchange_id *exid)
3246 if (!exid->nii_domain.data)
3248 xdr_netobj_dup(&clp->cl_nii_domain, &exid->nii_domain, GFP_KERNEL);
3249 if (!clp->cl_nii_domain.data)
3250 return nfserr_jukebox;
3251 xdr_netobj_dup(&clp->cl_nii_name, &exid->nii_name, GFP_KERNEL);
3252 if (!clp->cl_nii_name.data)
3253 return nfserr_jukebox;
3254 clp->cl_nii_time = exid->nii_time;
3259 nfsd4_exchange_id(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3260 union nfsd4_op_u *u)
3262 struct nfsd4_exchange_id *exid = &u->exchange_id;
3263 struct nfs4_client *conf, *new;
3264 struct nfs4_client *unconf = NULL;
3266 char addr_str[INET6_ADDRSTRLEN];
3267 nfs4_verifier verf = exid->verifier;
3268 struct sockaddr *sa = svc_addr(rqstp);
3269 bool update = exid->flags & EXCHGID4_FLAG_UPD_CONFIRMED_REC_A;
3270 struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
3272 rpc_ntop(sa, addr_str, sizeof(addr_str));
3273 dprintk("%s rqstp=%p exid=%p clname.len=%u clname.data=%p "
3274 "ip_addr=%s flags %x, spa_how %u\n",
3275 __func__, rqstp, exid, exid->clname.len, exid->clname.data,
3276 addr_str, exid->flags, exid->spa_how);
3278 if (exid->flags & ~EXCHGID4_FLAG_MASK_A)
3279 return nfserr_inval;
3281 new = create_client(exid->clname, rqstp, &verf);
3283 return nfserr_jukebox;
3284 status = copy_impl_id(new, exid);
3288 switch (exid->spa_how) {
3290 exid->spo_must_enforce[0] = 0;
3291 exid->spo_must_enforce[1] = (
3292 1 << (OP_BIND_CONN_TO_SESSION - 32) |
3293 1 << (OP_EXCHANGE_ID - 32) |
3294 1 << (OP_CREATE_SESSION - 32) |
3295 1 << (OP_DESTROY_SESSION - 32) |
3296 1 << (OP_DESTROY_CLIENTID - 32));
3298 exid->spo_must_allow[0] &= (1 << (OP_CLOSE) |
3299 1 << (OP_OPEN_DOWNGRADE) |
3301 1 << (OP_DELEGRETURN));
3303 exid->spo_must_allow[1] &= (
3304 1 << (OP_TEST_STATEID - 32) |
3305 1 << (OP_FREE_STATEID - 32));
3306 if (!svc_rqst_integrity_protected(rqstp)) {
3307 status = nfserr_inval;
3311 * Sometimes userspace doesn't give us a principal.
3312 * Which is a bug, really. Anyway, we can't enforce
3313 * MACH_CRED in that case, better to give up now:
3315 if (!new->cl_cred.cr_principal &&
3316 !new->cl_cred.cr_raw_principal) {
3317 status = nfserr_serverfault;
3320 new->cl_mach_cred = true;
3324 default: /* checked by xdr code */
3328 status = nfserr_encr_alg_unsupp;
3332 /* Cases below refer to rfc 5661 section 18.35.4: */
3333 spin_lock(&nn->client_lock);
3334 conf = find_confirmed_client_by_name(&exid->clname, nn);
3336 bool creds_match = same_creds(&conf->cl_cred, &rqstp->rq_cred);
3337 bool verfs_match = same_verf(&verf, &conf->cl_verifier);
3340 if (!clp_used_exchangeid(conf)) { /* buggy client */
3341 status = nfserr_inval;
3344 if (!nfsd4_mach_creds_match(conf, rqstp)) {
3345 status = nfserr_wrong_cred;
3348 if (!creds_match) { /* case 9 */
3349 status = nfserr_perm;
3352 if (!verfs_match) { /* case 8 */
3353 status = nfserr_not_same;
3357 exid->flags |= EXCHGID4_FLAG_CONFIRMED_R;
3358 trace_nfsd_clid_confirmed_r(conf);
3361 if (!creds_match) { /* case 3 */
3362 if (client_has_state(conf)) {
3363 status = nfserr_clid_inuse;
3364 trace_nfsd_clid_cred_mismatch(conf, rqstp);
3369 if (verfs_match) { /* case 2 */
3370 conf->cl_exchange_flags |= EXCHGID4_FLAG_CONFIRMED_R;
3371 trace_nfsd_clid_confirmed_r(conf);
3374 /* case 5, client reboot */
3375 trace_nfsd_clid_verf_mismatch(conf, rqstp, &verf);
3380 if (update) { /* case 7 */
3381 status = nfserr_noent;
3385 unconf = find_unconfirmed_client_by_name(&exid->clname, nn);
3386 if (unconf) /* case 4, possible retry or client restart */
3387 unhash_client_locked(unconf);
3389 /* case 1, new owner ID */
3390 trace_nfsd_clid_fresh(new);
3394 status = mark_client_expired_locked(conf);
3397 trace_nfsd_clid_replaced(&conf->cl_clientid);
3399 new->cl_minorversion = cstate->minorversion;
3400 new->cl_spo_must_allow.u.words[0] = exid->spo_must_allow[0];
3401 new->cl_spo_must_allow.u.words[1] = exid->spo_must_allow[1];
3403 add_to_unconfirmed(new);
3406 exid->clientid.cl_boot = conf->cl_clientid.cl_boot;
3407 exid->clientid.cl_id = conf->cl_clientid.cl_id;
3409 exid->seqid = conf->cl_cs_slot.sl_seqid + 1;
3410 nfsd4_set_ex_flags(conf, exid);
3412 dprintk("nfsd4_exchange_id seqid %d flags %x\n",
3413 conf->cl_cs_slot.sl_seqid, conf->cl_exchange_flags);
3417 spin_unlock(&nn->client_lock);
3422 trace_nfsd_clid_expire_unconf(&unconf->cl_clientid);
3423 expire_client(unconf);
3429 check_slot_seqid(u32 seqid, u32 slot_seqid, int slot_inuse)
3431 dprintk("%s enter. seqid %d slot_seqid %d\n", __func__, seqid,
3434 /* The slot is in use, and no response has been sent. */
3436 if (seqid == slot_seqid)
3437 return nfserr_jukebox;
3439 return nfserr_seq_misordered;
3441 /* Note unsigned 32-bit arithmetic handles wraparound: */
3442 if (likely(seqid == slot_seqid + 1))
3444 if (seqid == slot_seqid)
3445 return nfserr_replay_cache;
3446 return nfserr_seq_misordered;
3450 * Cache the create session result into the create session single DRC
3451 * slot cache by saving the xdr structure. sl_seqid has been set.
3452 * Do this for solo or embedded create session operations.
3455 nfsd4_cache_create_session(struct nfsd4_create_session *cr_ses,
3456 struct nfsd4_clid_slot *slot, __be32 nfserr)
3458 slot->sl_status = nfserr;
3459 memcpy(&slot->sl_cr_ses, cr_ses, sizeof(*cr_ses));
3463 nfsd4_replay_create_session(struct nfsd4_create_session *cr_ses,
3464 struct nfsd4_clid_slot *slot)
3466 memcpy(cr_ses, &slot->sl_cr_ses, sizeof(*cr_ses));
3467 return slot->sl_status;
3470 #define NFSD_MIN_REQ_HDR_SEQ_SZ ((\
3471 2 * 2 + /* credential,verifier: AUTH_NULL, length 0 */ \
3472 1 + /* MIN tag is length with zero, only length */ \
3473 3 + /* version, opcount, opcode */ \
3474 XDR_QUADLEN(NFS4_MAX_SESSIONID_LEN) + \
3475 /* seqid, slotID, slotID, cache */ \
3476 4 ) * sizeof(__be32))
3478 #define NFSD_MIN_RESP_HDR_SEQ_SZ ((\
3479 2 + /* verifier: AUTH_NULL, length 0 */\
3481 1 + /* MIN tag is length with zero, only length */ \
3482 3 + /* opcount, opcode, opstatus*/ \
3483 XDR_QUADLEN(NFS4_MAX_SESSIONID_LEN) + \
3484 /* seqid, slotID, slotID, slotID, status */ \
3485 5 ) * sizeof(__be32))
3487 static __be32 check_forechannel_attrs(struct nfsd4_channel_attrs *ca, struct nfsd_net *nn)
3489 u32 maxrpc = nn->nfsd_serv->sv_max_mesg;
3491 if (ca->maxreq_sz < NFSD_MIN_REQ_HDR_SEQ_SZ)
3492 return nfserr_toosmall;
3493 if (ca->maxresp_sz < NFSD_MIN_RESP_HDR_SEQ_SZ)
3494 return nfserr_toosmall;
3495 ca->headerpadsz = 0;
3496 ca->maxreq_sz = min_t(u32, ca->maxreq_sz, maxrpc);
3497 ca->maxresp_sz = min_t(u32, ca->maxresp_sz, maxrpc);
3498 ca->maxops = min_t(u32, ca->maxops, NFSD_MAX_OPS_PER_COMPOUND);
3499 ca->maxresp_cached = min_t(u32, ca->maxresp_cached,
3500 NFSD_SLOT_CACHE_SIZE + NFSD_MIN_HDR_SEQ_SZ);
3501 ca->maxreqs = min_t(u32, ca->maxreqs, NFSD_MAX_SLOTS_PER_SESSION);
3503 * Note decreasing slot size below client's request may make it
3504 * difficult for client to function correctly, whereas
3505 * decreasing the number of slots will (just?) affect
3506 * performance. When short on memory we therefore prefer to
3507 * decrease number of slots instead of their size. Clients that
3508 * request larger slots than they need will get poor results:
3509 * Note that we always allow at least one slot, because our
3510 * accounting is soft and provides no guarantees either way.
3512 ca->maxreqs = nfsd4_get_drc_mem(ca, nn);
3518 * Server's NFSv4.1 backchannel support is AUTH_SYS-only for now.
3519 * These are based on similar macros in linux/sunrpc/msg_prot.h .
3521 #define RPC_MAX_HEADER_WITH_AUTH_SYS \
3522 (RPC_CALLHDRSIZE + 2 * (2 + UNX_CALLSLACK))
3524 #define RPC_MAX_REPHEADER_WITH_AUTH_SYS \
3525 (RPC_REPHDRSIZE + (2 + NUL_REPLYSLACK))
3527 #define NFSD_CB_MAX_REQ_SZ ((NFS4_enc_cb_recall_sz + \
3528 RPC_MAX_HEADER_WITH_AUTH_SYS) * sizeof(__be32))
3529 #define NFSD_CB_MAX_RESP_SZ ((NFS4_dec_cb_recall_sz + \
3530 RPC_MAX_REPHEADER_WITH_AUTH_SYS) * \
3533 static __be32 check_backchannel_attrs(struct nfsd4_channel_attrs *ca)
3535 ca->headerpadsz = 0;
3537 if (ca->maxreq_sz < NFSD_CB_MAX_REQ_SZ)
3538 return nfserr_toosmall;
3539 if (ca->maxresp_sz < NFSD_CB_MAX_RESP_SZ)
3540 return nfserr_toosmall;
3541 ca->maxresp_cached = 0;
3543 return nfserr_toosmall;
3548 static __be32 nfsd4_check_cb_sec(struct nfsd4_cb_sec *cbs)
3550 switch (cbs->flavor) {
3556 * GSS case: the spec doesn't allow us to return this
3557 * error. But it also doesn't allow us not to support
3559 * I'd rather this fail hard than return some error the
3560 * client might think it can already handle:
3562 return nfserr_encr_alg_unsupp;
3567 nfsd4_create_session(struct svc_rqst *rqstp,
3568 struct nfsd4_compound_state *cstate, union nfsd4_op_u *u)
3570 struct nfsd4_create_session *cr_ses = &u->create_session;
3571 struct sockaddr *sa = svc_addr(rqstp);
3572 struct nfs4_client *conf, *unconf;
3573 struct nfs4_client *old = NULL;
3574 struct nfsd4_session *new;
3575 struct nfsd4_conn *conn;
3576 struct nfsd4_clid_slot *cs_slot = NULL;
3578 struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
3580 if (cr_ses->flags & ~SESSION4_FLAG_MASK_A)
3581 return nfserr_inval;
3582 status = nfsd4_check_cb_sec(&cr_ses->cb_sec);
3585 status = check_forechannel_attrs(&cr_ses->fore_channel, nn);
3588 status = check_backchannel_attrs(&cr_ses->back_channel);
3590 goto out_release_drc_mem;
3591 status = nfserr_jukebox;
3592 new = alloc_session(&cr_ses->fore_channel, &cr_ses->back_channel);
3594 goto out_release_drc_mem;
3595 conn = alloc_conn_from_crses(rqstp, cr_ses);
3597 goto out_free_session;
3599 spin_lock(&nn->client_lock);
3600 unconf = find_unconfirmed_client(&cr_ses->clientid, true, nn);
3601 conf = find_confirmed_client(&cr_ses->clientid, true, nn);
3602 WARN_ON_ONCE(conf && unconf);
3605 status = nfserr_wrong_cred;
3606 if (!nfsd4_mach_creds_match(conf, rqstp))
3608 cs_slot = &conf->cl_cs_slot;
3609 status = check_slot_seqid(cr_ses->seqid, cs_slot->sl_seqid, 0);
3611 if (status == nfserr_replay_cache)
3612 status = nfsd4_replay_create_session(cr_ses, cs_slot);
3615 } else if (unconf) {
3616 status = nfserr_clid_inuse;
3617 if (!same_creds(&unconf->cl_cred, &rqstp->rq_cred) ||
3618 !rpc_cmp_addr(sa, (struct sockaddr *) &unconf->cl_addr)) {
3619 trace_nfsd_clid_cred_mismatch(unconf, rqstp);
3622 status = nfserr_wrong_cred;
3623 if (!nfsd4_mach_creds_match(unconf, rqstp))
3625 cs_slot = &unconf->cl_cs_slot;
3626 status = check_slot_seqid(cr_ses->seqid, cs_slot->sl_seqid, 0);
3628 /* an unconfirmed replay returns misordered */
3629 status = nfserr_seq_misordered;
3632 old = find_confirmed_client_by_name(&unconf->cl_name, nn);
3634 status = mark_client_expired_locked(old);
3639 trace_nfsd_clid_replaced(&old->cl_clientid);
3641 move_to_confirmed(unconf);
3644 status = nfserr_stale_clientid;
3648 /* Persistent sessions are not supported */
3649 cr_ses->flags &= ~SESSION4_PERSIST;
3650 /* Upshifting from TCP to RDMA is not supported */
3651 cr_ses->flags &= ~SESSION4_RDMA;
3653 init_session(rqstp, new, conf, cr_ses);
3654 nfsd4_get_session_locked(new);
3656 memcpy(cr_ses->sessionid.data, new->se_sessionid.data,
3657 NFS4_MAX_SESSIONID_LEN);
3658 cs_slot->sl_seqid++;
3659 cr_ses->seqid = cs_slot->sl_seqid;
3661 /* cache solo and embedded create sessions under the client_lock */
3662 nfsd4_cache_create_session(cr_ses, cs_slot, status);
3663 spin_unlock(&nn->client_lock);
3665 fsnotify_dentry(conf->cl_nfsd_info_dentry, FS_MODIFY);
3666 /* init connection and backchannel */
3667 nfsd4_init_conn(rqstp, conn, new);
3668 nfsd4_put_session(new);
3673 spin_unlock(&nn->client_lock);
3678 __free_session(new);
3679 out_release_drc_mem:
3680 nfsd4_put_drc_mem(&cr_ses->fore_channel);
3684 static __be32 nfsd4_map_bcts_dir(u32 *dir)
3687 case NFS4_CDFC4_FORE:
3688 case NFS4_CDFC4_BACK:
3690 case NFS4_CDFC4_FORE_OR_BOTH:
3691 case NFS4_CDFC4_BACK_OR_BOTH:
3692 *dir = NFS4_CDFC4_BOTH;
3695 return nfserr_inval;
3698 __be32 nfsd4_backchannel_ctl(struct svc_rqst *rqstp,
3699 struct nfsd4_compound_state *cstate,
3700 union nfsd4_op_u *u)
3702 struct nfsd4_backchannel_ctl *bc = &u->backchannel_ctl;
3703 struct nfsd4_session *session = cstate->session;
3704 struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
3707 status = nfsd4_check_cb_sec(&bc->bc_cb_sec);
3710 spin_lock(&nn->client_lock);
3711 session->se_cb_prog = bc->bc_cb_program;
3712 session->se_cb_sec = bc->bc_cb_sec;
3713 spin_unlock(&nn->client_lock);
3715 nfsd4_probe_callback(session->se_client);
3720 static struct nfsd4_conn *__nfsd4_find_conn(struct svc_xprt *xpt, struct nfsd4_session *s)
3722 struct nfsd4_conn *c;
3724 list_for_each_entry(c, &s->se_conns, cn_persession) {
3725 if (c->cn_xprt == xpt) {
3732 static __be32 nfsd4_match_existing_connection(struct svc_rqst *rqst,
3733 struct nfsd4_session *session, u32 req, struct nfsd4_conn **conn)
3735 struct nfs4_client *clp = session->se_client;
3736 struct svc_xprt *xpt = rqst->rq_xprt;
3737 struct nfsd4_conn *c;
3740 /* Following the last paragraph of RFC 5661 Section 18.34.3: */
3741 spin_lock(&clp->cl_lock);
3742 c = __nfsd4_find_conn(xpt, session);
3744 status = nfserr_noent;
3745 else if (req == c->cn_flags)
3747 else if (req == NFS4_CDFC4_FORE_OR_BOTH &&
3748 c->cn_flags != NFS4_CDFC4_BACK)
3750 else if (req == NFS4_CDFC4_BACK_OR_BOTH &&
3751 c->cn_flags != NFS4_CDFC4_FORE)
3754 status = nfserr_inval;
3755 spin_unlock(&clp->cl_lock);
3756 if (status == nfs_ok && conn)
3761 __be32 nfsd4_bind_conn_to_session(struct svc_rqst *rqstp,
3762 struct nfsd4_compound_state *cstate,
3763 union nfsd4_op_u *u)
3765 struct nfsd4_bind_conn_to_session *bcts = &u->bind_conn_to_session;
3767 struct nfsd4_conn *conn;
3768 struct nfsd4_session *session;
3769 struct net *net = SVC_NET(rqstp);
3770 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
3772 if (!nfsd4_last_compound_op(rqstp))
3773 return nfserr_not_only_op;
3774 spin_lock(&nn->client_lock);
3775 session = find_in_sessionid_hashtbl(&bcts->sessionid, net, &status);
3776 spin_unlock(&nn->client_lock);
3778 goto out_no_session;
3779 status = nfserr_wrong_cred;
3780 if (!nfsd4_mach_creds_match(session->se_client, rqstp))
3782 status = nfsd4_match_existing_connection(rqstp, session,
3784 if (status == nfs_ok) {
3785 if (bcts->dir == NFS4_CDFC4_FORE_OR_BOTH ||
3786 bcts->dir == NFS4_CDFC4_BACK)
3787 conn->cn_flags |= NFS4_CDFC4_BACK;
3788 nfsd4_probe_callback(session->se_client);
3791 if (status == nfserr_inval)
3793 status = nfsd4_map_bcts_dir(&bcts->dir);
3796 conn = alloc_conn(rqstp, bcts->dir);
3797 status = nfserr_jukebox;
3800 nfsd4_init_conn(rqstp, conn, session);
3803 nfsd4_put_session(session);
3808 static bool nfsd4_compound_in_session(struct nfsd4_compound_state *cstate, struct nfs4_sessionid *sid)
3810 if (!cstate->session)
3812 return !memcmp(sid, &cstate->session->se_sessionid, sizeof(*sid));
3816 nfsd4_destroy_session(struct svc_rqst *r, struct nfsd4_compound_state *cstate,
3817 union nfsd4_op_u *u)
3819 struct nfs4_sessionid *sessionid = &u->destroy_session.sessionid;
3820 struct nfsd4_session *ses;
3822 int ref_held_by_me = 0;
3823 struct net *net = SVC_NET(r);
3824 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
3826 status = nfserr_not_only_op;
3827 if (nfsd4_compound_in_session(cstate, sessionid)) {
3828 if (!nfsd4_last_compound_op(r))
3832 dump_sessionid(__func__, sessionid);
3833 spin_lock(&nn->client_lock);
3834 ses = find_in_sessionid_hashtbl(sessionid, net, &status);
3836 goto out_client_lock;
3837 status = nfserr_wrong_cred;
3838 if (!nfsd4_mach_creds_match(ses->se_client, r))
3839 goto out_put_session;
3840 status = mark_session_dead_locked(ses, 1 + ref_held_by_me);
3842 goto out_put_session;
3843 unhash_session(ses);
3844 spin_unlock(&nn->client_lock);
3846 nfsd4_probe_callback_sync(ses->se_client);
3848 spin_lock(&nn->client_lock);
3851 nfsd4_put_session_locked(ses);
3853 spin_unlock(&nn->client_lock);
3858 static __be32 nfsd4_sequence_check_conn(struct nfsd4_conn *new, struct nfsd4_session *ses)
3860 struct nfs4_client *clp = ses->se_client;
3861 struct nfsd4_conn *c;
3862 __be32 status = nfs_ok;
3865 spin_lock(&clp->cl_lock);
3866 c = __nfsd4_find_conn(new->cn_xprt, ses);
3869 status = nfserr_conn_not_bound_to_session;
3870 if (clp->cl_mach_cred)
3872 __nfsd4_hash_conn(new, ses);
3873 spin_unlock(&clp->cl_lock);
3874 ret = nfsd4_register_conn(new);
3876 /* oops; xprt is already down: */
3877 nfsd4_conn_lost(&new->cn_xpt_user);
3880 spin_unlock(&clp->cl_lock);
3885 static bool nfsd4_session_too_many_ops(struct svc_rqst *rqstp, struct nfsd4_session *session)
3887 struct nfsd4_compoundargs *args = rqstp->rq_argp;
3889 return args->opcnt > session->se_fchannel.maxops;
3892 static bool nfsd4_request_too_big(struct svc_rqst *rqstp,
3893 struct nfsd4_session *session)
3895 struct xdr_buf *xb = &rqstp->rq_arg;
3897 return xb->len > session->se_fchannel.maxreq_sz;
3900 static bool replay_matches_cache(struct svc_rqst *rqstp,
3901 struct nfsd4_sequence *seq, struct nfsd4_slot *slot)
3903 struct nfsd4_compoundargs *argp = rqstp->rq_argp;
3905 if ((bool)(slot->sl_flags & NFSD4_SLOT_CACHETHIS) !=
3906 (bool)seq->cachethis)
3909 * If there's an error then the reply can have fewer ops than
3912 if (slot->sl_opcnt < argp->opcnt && !slot->sl_status)
3915 * But if we cached a reply with *more* ops than the call you're
3916 * sending us now, then this new call is clearly not really a
3917 * replay of the old one:
3919 if (slot->sl_opcnt > argp->opcnt)
3921 /* This is the only check explicitly called by spec: */
3922 if (!same_creds(&rqstp->rq_cred, &slot->sl_cred))
3925 * There may be more comparisons we could actually do, but the
3926 * spec doesn't require us to catch every case where the calls
3927 * don't match (that would require caching the call as well as
3928 * the reply), so we don't bother.
3934 nfsd4_sequence(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3935 union nfsd4_op_u *u)
3937 struct nfsd4_sequence *seq = &u->sequence;
3938 struct nfsd4_compoundres *resp = rqstp->rq_resp;
3939 struct xdr_stream *xdr = resp->xdr;
3940 struct nfsd4_session *session;
3941 struct nfs4_client *clp;
3942 struct nfsd4_slot *slot;
3943 struct nfsd4_conn *conn;
3946 struct net *net = SVC_NET(rqstp);
3947 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
3949 if (resp->opcnt != 1)
3950 return nfserr_sequence_pos;
3953 * Will be either used or freed by nfsd4_sequence_check_conn
3956 conn = alloc_conn(rqstp, NFS4_CDFC4_FORE);
3958 return nfserr_jukebox;
3960 spin_lock(&nn->client_lock);
3961 session = find_in_sessionid_hashtbl(&seq->sessionid, net, &status);
3963 goto out_no_session;
3964 clp = session->se_client;
3966 status = nfserr_too_many_ops;
3967 if (nfsd4_session_too_many_ops(rqstp, session))
3968 goto out_put_session;
3970 status = nfserr_req_too_big;
3971 if (nfsd4_request_too_big(rqstp, session))
3972 goto out_put_session;
3974 status = nfserr_badslot;
3975 if (seq->slotid >= session->se_fchannel.maxreqs)
3976 goto out_put_session;
3978 slot = session->se_slots[seq->slotid];
3979 dprintk("%s: slotid %d\n", __func__, seq->slotid);
3981 /* We do not negotiate the number of slots yet, so set the
3982 * maxslots to the session maxreqs which is used to encode
3983 * sr_highest_slotid and the sr_target_slot id to maxslots */
3984 seq->maxslots = session->se_fchannel.maxreqs;
3986 status = check_slot_seqid(seq->seqid, slot->sl_seqid,
3987 slot->sl_flags & NFSD4_SLOT_INUSE);
3988 if (status == nfserr_replay_cache) {
3989 status = nfserr_seq_misordered;
3990 if (!(slot->sl_flags & NFSD4_SLOT_INITIALIZED))
3991 goto out_put_session;
3992 status = nfserr_seq_false_retry;
3993 if (!replay_matches_cache(rqstp, seq, slot))
3994 goto out_put_session;
3995 cstate->slot = slot;
3996 cstate->session = session;
3998 /* Return the cached reply status and set cstate->status
3999 * for nfsd4_proc_compound processing */
4000 status = nfsd4_replay_cache_entry(resp, seq);
4001 cstate->status = nfserr_replay_cache;
4005 goto out_put_session;
4007 status = nfsd4_sequence_check_conn(conn, session);
4010 goto out_put_session;
4012 buflen = (seq->cachethis) ?
4013 session->se_fchannel.maxresp_cached :
4014 session->se_fchannel.maxresp_sz;
4015 status = (seq->cachethis) ? nfserr_rep_too_big_to_cache :
4017 if (xdr_restrict_buflen(xdr, buflen - rqstp->rq_auth_slack))
4018 goto out_put_session;
4019 svc_reserve(rqstp, buflen);
4022 /* Success! bump slot seqid */
4023 slot->sl_seqid = seq->seqid;
4024 slot->sl_flags |= NFSD4_SLOT_INUSE;
4026 slot->sl_flags |= NFSD4_SLOT_CACHETHIS;
4028 slot->sl_flags &= ~NFSD4_SLOT_CACHETHIS;
4030 cstate->slot = slot;
4031 cstate->session = session;
4035 switch (clp->cl_cb_state) {
4037 seq->status_flags = SEQ4_STATUS_CB_PATH_DOWN;
4039 case NFSD4_CB_FAULT:
4040 seq->status_flags = SEQ4_STATUS_BACKCHANNEL_FAULT;
4043 seq->status_flags = 0;
4045 if (!list_empty(&clp->cl_revoked))
4046 seq->status_flags |= SEQ4_STATUS_RECALLABLE_STATE_REVOKED;
4050 spin_unlock(&nn->client_lock);
4053 nfsd4_put_session_locked(session);
4054 goto out_no_session;
4058 nfsd4_sequence_done(struct nfsd4_compoundres *resp)
4060 struct nfsd4_compound_state *cs = &resp->cstate;
4062 if (nfsd4_has_session(cs)) {
4063 if (cs->status != nfserr_replay_cache) {
4064 nfsd4_store_cache_entry(resp);
4065 cs->slot->sl_flags &= ~NFSD4_SLOT_INUSE;
4067 /* Drop session reference that was taken in nfsd4_sequence() */
4068 nfsd4_put_session(cs->session);
4070 put_client_renew(cs->clp);
4074 nfsd4_destroy_clientid(struct svc_rqst *rqstp,
4075 struct nfsd4_compound_state *cstate,
4076 union nfsd4_op_u *u)
4078 struct nfsd4_destroy_clientid *dc = &u->destroy_clientid;
4079 struct nfs4_client *conf, *unconf;
4080 struct nfs4_client *clp = NULL;
4082 struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
4084 spin_lock(&nn->client_lock);
4085 unconf = find_unconfirmed_client(&dc->clientid, true, nn);
4086 conf = find_confirmed_client(&dc->clientid, true, nn);
4087 WARN_ON_ONCE(conf && unconf);
4090 if (client_has_state(conf)) {
4091 status = nfserr_clientid_busy;
4094 status = mark_client_expired_locked(conf);
4101 status = nfserr_stale_clientid;
4104 if (!nfsd4_mach_creds_match(clp, rqstp)) {
4106 status = nfserr_wrong_cred;
4109 trace_nfsd_clid_destroyed(&clp->cl_clientid);
4110 unhash_client_locked(clp);
4112 spin_unlock(&nn->client_lock);
4119 nfsd4_reclaim_complete(struct svc_rqst *rqstp,
4120 struct nfsd4_compound_state *cstate, union nfsd4_op_u *u)
4122 struct nfsd4_reclaim_complete *rc = &u->reclaim_complete;
4123 struct nfs4_client *clp = cstate->clp;
4126 if (rc->rca_one_fs) {
4127 if (!cstate->current_fh.fh_dentry)
4128 return nfserr_nofilehandle;
4130 * We don't take advantage of the rca_one_fs case.
4131 * That's OK, it's optional, we can safely ignore it.
4136 status = nfserr_complete_already;
4137 if (test_and_set_bit(NFSD4_CLIENT_RECLAIM_COMPLETE, &clp->cl_flags))
4140 status = nfserr_stale_clientid;
4141 if (is_client_expired(clp))
4143 * The following error isn't really legal.
4144 * But we only get here if the client just explicitly
4145 * destroyed the client. Surely it no longer cares what
4146 * error it gets back on an operation for the dead
4152 trace_nfsd_clid_reclaim_complete(&clp->cl_clientid);
4153 nfsd4_client_record_create(clp);
4154 inc_reclaim_complete(clp);
4160 nfsd4_setclientid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
4161 union nfsd4_op_u *u)
4163 struct nfsd4_setclientid *setclid = &u->setclientid;
4164 struct xdr_netobj clname = setclid->se_name;
4165 nfs4_verifier clverifier = setclid->se_verf;
4166 struct nfs4_client *conf, *new;
4167 struct nfs4_client *unconf = NULL;
4169 struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
4171 new = create_client(clname, rqstp, &clverifier);
4173 return nfserr_jukebox;
4174 spin_lock(&nn->client_lock);
4175 conf = find_confirmed_client_by_name(&clname, nn);
4176 if (conf && client_has_state(conf)) {
4177 status = nfserr_clid_inuse;
4178 if (clp_used_exchangeid(conf))
4180 if (!same_creds(&conf->cl_cred, &rqstp->rq_cred)) {
4181 trace_nfsd_clid_cred_mismatch(conf, rqstp);
4185 unconf = find_unconfirmed_client_by_name(&clname, nn);
4187 unhash_client_locked(unconf);
4189 if (same_verf(&conf->cl_verifier, &clverifier)) {
4190 copy_clid(new, conf);
4191 gen_confirm(new, nn);
4193 trace_nfsd_clid_verf_mismatch(conf, rqstp,
4196 trace_nfsd_clid_fresh(new);
4197 new->cl_minorversion = 0;
4198 gen_callback(new, setclid, rqstp);
4199 add_to_unconfirmed(new);
4200 setclid->se_clientid.cl_boot = new->cl_clientid.cl_boot;
4201 setclid->se_clientid.cl_id = new->cl_clientid.cl_id;
4202 memcpy(setclid->se_confirm.data, new->cl_confirm.data, sizeof(setclid->se_confirm.data));
4206 spin_unlock(&nn->client_lock);
4210 trace_nfsd_clid_expire_unconf(&unconf->cl_clientid);
4211 expire_client(unconf);
4217 nfsd4_setclientid_confirm(struct svc_rqst *rqstp,
4218 struct nfsd4_compound_state *cstate,
4219 union nfsd4_op_u *u)
4221 struct nfsd4_setclientid_confirm *setclientid_confirm =
4222 &u->setclientid_confirm;
4223 struct nfs4_client *conf, *unconf;
4224 struct nfs4_client *old = NULL;
4225 nfs4_verifier confirm = setclientid_confirm->sc_confirm;
4226 clientid_t * clid = &setclientid_confirm->sc_clientid;
4228 struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
4230 if (STALE_CLIENTID(clid, nn))
4231 return nfserr_stale_clientid;
4233 spin_lock(&nn->client_lock);
4234 conf = find_confirmed_client(clid, false, nn);
4235 unconf = find_unconfirmed_client(clid, false, nn);
4237 * We try hard to give out unique clientid's, so if we get an
4238 * attempt to confirm the same clientid with a different cred,
4239 * the client may be buggy; this should never happen.
4241 * Nevertheless, RFC 7530 recommends INUSE for this case:
4243 status = nfserr_clid_inuse;
4244 if (unconf && !same_creds(&unconf->cl_cred, &rqstp->rq_cred)) {
4245 trace_nfsd_clid_cred_mismatch(unconf, rqstp);
4248 if (conf && !same_creds(&conf->cl_cred, &rqstp->rq_cred)) {
4249 trace_nfsd_clid_cred_mismatch(conf, rqstp);
4252 if (!unconf || !same_verf(&confirm, &unconf->cl_confirm)) {
4253 if (conf && same_verf(&confirm, &conf->cl_confirm)) {
4256 status = nfserr_stale_clientid;
4262 unhash_client_locked(old);
4263 nfsd4_change_callback(conf, &unconf->cl_cb_conn);
4265 old = find_confirmed_client_by_name(&unconf->cl_name, nn);
4267 status = nfserr_clid_inuse;
4268 if (client_has_state(old)
4269 && !same_creds(&unconf->cl_cred,
4274 status = mark_client_expired_locked(old);
4279 trace_nfsd_clid_replaced(&old->cl_clientid);
4281 move_to_confirmed(unconf);
4284 get_client_locked(conf);
4285 spin_unlock(&nn->client_lock);
4287 fsnotify_dentry(conf->cl_nfsd_info_dentry, FS_MODIFY);
4288 nfsd4_probe_callback(conf);
4289 spin_lock(&nn->client_lock);
4290 put_client_renew_locked(conf);
4292 spin_unlock(&nn->client_lock);
4298 static struct nfs4_file *nfsd4_alloc_file(void)
4300 return kmem_cache_alloc(file_slab, GFP_KERNEL);
4303 /* OPEN Share state helper functions */
4305 static void nfsd4_file_init(const struct svc_fh *fh, struct nfs4_file *fp)
4307 refcount_set(&fp->fi_ref, 1);
4308 spin_lock_init(&fp->fi_lock);
4309 INIT_LIST_HEAD(&fp->fi_stateids);
4310 INIT_LIST_HEAD(&fp->fi_delegations);
4311 INIT_LIST_HEAD(&fp->fi_clnt_odstate);
4312 fh_copy_shallow(&fp->fi_fhandle, &fh->fh_handle);
4313 fp->fi_deleg_file = NULL;
4314 fp->fi_had_conflict = false;
4315 fp->fi_share_deny = 0;
4316 memset(fp->fi_fds, 0, sizeof(fp->fi_fds));
4317 memset(fp->fi_access, 0, sizeof(fp->fi_access));
4318 fp->fi_aliased = false;
4319 fp->fi_inode = d_inode(fh->fh_dentry);
4320 #ifdef CONFIG_NFSD_PNFS
4321 INIT_LIST_HEAD(&fp->fi_lo_states);
4322 atomic_set(&fp->fi_lo_recalls, 0);
4327 nfsd4_free_slabs(void)
4329 kmem_cache_destroy(client_slab);
4330 kmem_cache_destroy(openowner_slab);
4331 kmem_cache_destroy(lockowner_slab);
4332 kmem_cache_destroy(file_slab);
4333 kmem_cache_destroy(stateid_slab);
4334 kmem_cache_destroy(deleg_slab);
4335 kmem_cache_destroy(odstate_slab);
4339 nfsd4_init_slabs(void)
4341 client_slab = kmem_cache_create("nfsd4_clients",
4342 sizeof(struct nfs4_client), 0, 0, NULL);
4343 if (client_slab == NULL)
4345 openowner_slab = kmem_cache_create("nfsd4_openowners",
4346 sizeof(struct nfs4_openowner), 0, 0, NULL);
4347 if (openowner_slab == NULL)
4348 goto out_free_client_slab;
4349 lockowner_slab = kmem_cache_create("nfsd4_lockowners",
4350 sizeof(struct nfs4_lockowner), 0, 0, NULL);
4351 if (lockowner_slab == NULL)
4352 goto out_free_openowner_slab;
4353 file_slab = kmem_cache_create("nfsd4_files",
4354 sizeof(struct nfs4_file), 0, 0, NULL);
4355 if (file_slab == NULL)
4356 goto out_free_lockowner_slab;
4357 stateid_slab = kmem_cache_create("nfsd4_stateids",
4358 sizeof(struct nfs4_ol_stateid), 0, 0, NULL);
4359 if (stateid_slab == NULL)
4360 goto out_free_file_slab;
4361 deleg_slab = kmem_cache_create("nfsd4_delegations",
4362 sizeof(struct nfs4_delegation), 0, 0, NULL);
4363 if (deleg_slab == NULL)
4364 goto out_free_stateid_slab;
4365 odstate_slab = kmem_cache_create("nfsd4_odstate",
4366 sizeof(struct nfs4_clnt_odstate), 0, 0, NULL);
4367 if (odstate_slab == NULL)
4368 goto out_free_deleg_slab;
4371 out_free_deleg_slab:
4372 kmem_cache_destroy(deleg_slab);
4373 out_free_stateid_slab:
4374 kmem_cache_destroy(stateid_slab);
4376 kmem_cache_destroy(file_slab);
4377 out_free_lockowner_slab:
4378 kmem_cache_destroy(lockowner_slab);
4379 out_free_openowner_slab:
4380 kmem_cache_destroy(openowner_slab);
4381 out_free_client_slab:
4382 kmem_cache_destroy(client_slab);
4387 static unsigned long
4388 nfsd4_state_shrinker_count(struct shrinker *shrink, struct shrink_control *sc)
4391 struct nfsd_net *nn = container_of(shrink,
4392 struct nfsd_net, nfsd_client_shrinker);
4394 count = atomic_read(&nn->nfsd_courtesy_clients);
4396 count = atomic_long_read(&num_delegations);
4398 queue_work(laundry_wq, &nn->nfsd_shrinker_work);
4399 return (unsigned long)count;
4402 static unsigned long
4403 nfsd4_state_shrinker_scan(struct shrinker *shrink, struct shrink_control *sc)
4409 nfsd4_init_leases_net(struct nfsd_net *nn)
4414 nn->nfsd4_lease = 90; /* default lease time */
4415 nn->nfsd4_grace = 90;
4416 nn->somebody_reclaimed = false;
4417 nn->track_reclaim_completes = false;
4418 nn->clverifier_counter = get_random_u32();
4419 nn->clientid_base = get_random_u32();
4420 nn->clientid_counter = nn->clientid_base + 1;
4421 nn->s2s_cp_cl_id = nn->clientid_counter++;
4423 atomic_set(&nn->nfs4_client_count, 0);
4425 max_clients = (u64)si.totalram * si.mem_unit / (1024 * 1024 * 1024);
4426 max_clients *= NFS4_CLIENTS_PER_GB;
4427 nn->nfs4_max_clients = max_t(int, max_clients, NFS4_CLIENTS_PER_GB);
4429 atomic_set(&nn->nfsd_courtesy_clients, 0);
4432 static void init_nfs4_replay(struct nfs4_replay *rp)
4434 rp->rp_status = nfserr_serverfault;
4436 rp->rp_buf = rp->rp_ibuf;
4437 mutex_init(&rp->rp_mutex);
4440 static void nfsd4_cstate_assign_replay(struct nfsd4_compound_state *cstate,
4441 struct nfs4_stateowner *so)
4443 if (!nfsd4_has_session(cstate)) {
4444 mutex_lock(&so->so_replay.rp_mutex);
4445 cstate->replay_owner = nfs4_get_stateowner(so);
4449 void nfsd4_cstate_clear_replay(struct nfsd4_compound_state *cstate)
4451 struct nfs4_stateowner *so = cstate->replay_owner;
4454 cstate->replay_owner = NULL;
4455 mutex_unlock(&so->so_replay.rp_mutex);
4456 nfs4_put_stateowner(so);
4460 static inline void *alloc_stateowner(struct kmem_cache *slab, struct xdr_netobj *owner, struct nfs4_client *clp)
4462 struct nfs4_stateowner *sop;
4464 sop = kmem_cache_alloc(slab, GFP_KERNEL);
4468 xdr_netobj_dup(&sop->so_owner, owner, GFP_KERNEL);
4469 if (!sop->so_owner.data) {
4470 kmem_cache_free(slab, sop);
4474 INIT_LIST_HEAD(&sop->so_stateids);
4475 sop->so_client = clp;
4476 init_nfs4_replay(&sop->so_replay);
4477 atomic_set(&sop->so_count, 1);
4481 static void hash_openowner(struct nfs4_openowner *oo, struct nfs4_client *clp, unsigned int strhashval)
4483 lockdep_assert_held(&clp->cl_lock);
4485 list_add(&oo->oo_owner.so_strhash,
4486 &clp->cl_ownerstr_hashtbl[strhashval]);
4487 list_add(&oo->oo_perclient, &clp->cl_openowners);
4490 static void nfs4_unhash_openowner(struct nfs4_stateowner *so)
4492 unhash_openowner_locked(openowner(so));
4495 static void nfs4_free_openowner(struct nfs4_stateowner *so)
4497 struct nfs4_openowner *oo = openowner(so);
4499 kmem_cache_free(openowner_slab, oo);
4502 static const struct nfs4_stateowner_operations openowner_ops = {
4503 .so_unhash = nfs4_unhash_openowner,
4504 .so_free = nfs4_free_openowner,
4507 static struct nfs4_ol_stateid *
4508 nfsd4_find_existing_open(struct nfs4_file *fp, struct nfsd4_open *open)
4510 struct nfs4_ol_stateid *local, *ret = NULL;
4511 struct nfs4_openowner *oo = open->op_openowner;
4513 lockdep_assert_held(&fp->fi_lock);
4515 list_for_each_entry(local, &fp->fi_stateids, st_perfile) {
4516 /* ignore lock owners */
4517 if (local->st_stateowner->so_is_open_owner == 0)
4519 if (local->st_stateowner != &oo->oo_owner)
4521 if (local->st_stid.sc_type == NFS4_OPEN_STID) {
4523 refcount_inc(&ret->st_stid.sc_count);
4531 nfsd4_verify_open_stid(struct nfs4_stid *s)
4533 __be32 ret = nfs_ok;
4535 switch (s->sc_type) {
4539 case NFS4_CLOSED_STID:
4540 case NFS4_CLOSED_DELEG_STID:
4541 ret = nfserr_bad_stateid;
4543 case NFS4_REVOKED_DELEG_STID:
4544 ret = nfserr_deleg_revoked;
4549 /* Lock the stateid st_mutex, and deal with races with CLOSE */
4551 nfsd4_lock_ol_stateid(struct nfs4_ol_stateid *stp)
4555 mutex_lock_nested(&stp->st_mutex, LOCK_STATEID_MUTEX);
4556 ret = nfsd4_verify_open_stid(&stp->st_stid);
4558 mutex_unlock(&stp->st_mutex);
4562 static struct nfs4_ol_stateid *
4563 nfsd4_find_and_lock_existing_open(struct nfs4_file *fp, struct nfsd4_open *open)
4565 struct nfs4_ol_stateid *stp;
4567 spin_lock(&fp->fi_lock);
4568 stp = nfsd4_find_existing_open(fp, open);
4569 spin_unlock(&fp->fi_lock);
4570 if (!stp || nfsd4_lock_ol_stateid(stp) == nfs_ok)
4572 nfs4_put_stid(&stp->st_stid);
4577 static struct nfs4_openowner *
4578 alloc_init_open_stateowner(unsigned int strhashval, struct nfsd4_open *open,
4579 struct nfsd4_compound_state *cstate)
4581 struct nfs4_client *clp = cstate->clp;
4582 struct nfs4_openowner *oo, *ret;
4584 oo = alloc_stateowner(openowner_slab, &open->op_owner, clp);
4587 oo->oo_owner.so_ops = &openowner_ops;
4588 oo->oo_owner.so_is_open_owner = 1;
4589 oo->oo_owner.so_seqid = open->op_seqid;
4591 if (nfsd4_has_session(cstate))
4592 oo->oo_flags |= NFS4_OO_CONFIRMED;
4594 oo->oo_last_closed_stid = NULL;
4595 INIT_LIST_HEAD(&oo->oo_close_lru);
4596 spin_lock(&clp->cl_lock);
4597 ret = find_openstateowner_str_locked(strhashval, open, clp);
4599 hash_openowner(oo, clp, strhashval);
4602 nfs4_free_stateowner(&oo->oo_owner);
4604 spin_unlock(&clp->cl_lock);
4608 static struct nfs4_ol_stateid *
4609 init_open_stateid(struct nfs4_file *fp, struct nfsd4_open *open)
4612 struct nfs4_openowner *oo = open->op_openowner;
4613 struct nfs4_ol_stateid *retstp = NULL;
4614 struct nfs4_ol_stateid *stp;
4617 /* We are moving these outside of the spinlocks to avoid the warnings */
4618 mutex_init(&stp->st_mutex);
4619 mutex_lock_nested(&stp->st_mutex, OPEN_STATEID_MUTEX);
4622 spin_lock(&oo->oo_owner.so_client->cl_lock);
4623 spin_lock(&fp->fi_lock);
4625 retstp = nfsd4_find_existing_open(fp, open);
4629 open->op_stp = NULL;
4630 refcount_inc(&stp->st_stid.sc_count);
4631 stp->st_stid.sc_type = NFS4_OPEN_STID;
4632 INIT_LIST_HEAD(&stp->st_locks);
4633 stp->st_stateowner = nfs4_get_stateowner(&oo->oo_owner);
4635 stp->st_stid.sc_file = fp;
4636 stp->st_access_bmap = 0;
4637 stp->st_deny_bmap = 0;
4638 stp->st_openstp = NULL;
4639 list_add(&stp->st_perstateowner, &oo->oo_owner.so_stateids);
4640 list_add(&stp->st_perfile, &fp->fi_stateids);
4643 spin_unlock(&fp->fi_lock);
4644 spin_unlock(&oo->oo_owner.so_client->cl_lock);
4646 /* Handle races with CLOSE */
4647 if (nfsd4_lock_ol_stateid(retstp) != nfs_ok) {
4648 nfs4_put_stid(&retstp->st_stid);
4651 /* To keep mutex tracking happy */
4652 mutex_unlock(&stp->st_mutex);
4659 * In the 4.0 case we need to keep the owners around a little while to handle
4660 * CLOSE replay. We still do need to release any file access that is held by
4661 * them before returning however.
4664 move_to_close_lru(struct nfs4_ol_stateid *s, struct net *net)
4666 struct nfs4_ol_stateid *last;
4667 struct nfs4_openowner *oo = openowner(s->st_stateowner);
4668 struct nfsd_net *nn = net_generic(s->st_stid.sc_client->net,
4671 dprintk("NFSD: move_to_close_lru nfs4_openowner %p\n", oo);
4674 * We know that we hold one reference via nfsd4_close, and another
4675 * "persistent" reference for the client. If the refcount is higher
4676 * than 2, then there are still calls in progress that are using this
4677 * stateid. We can't put the sc_file reference until they are finished.
4678 * Wait for the refcount to drop to 2. Since it has been unhashed,
4679 * there should be no danger of the refcount going back up again at
4682 wait_event(close_wq, refcount_read(&s->st_stid.sc_count) == 2);
4684 release_all_access(s);
4685 if (s->st_stid.sc_file) {
4686 put_nfs4_file(s->st_stid.sc_file);
4687 s->st_stid.sc_file = NULL;
4690 spin_lock(&nn->client_lock);
4691 last = oo->oo_last_closed_stid;
4692 oo->oo_last_closed_stid = s;
4693 list_move_tail(&oo->oo_close_lru, &nn->close_lru);
4694 oo->oo_time = ktime_get_boottime_seconds();
4695 spin_unlock(&nn->client_lock);
4697 nfs4_put_stid(&last->st_stid);
4700 static noinline_for_stack struct nfs4_file *
4701 nfsd4_file_hash_lookup(const struct svc_fh *fhp)
4703 struct inode *inode = d_inode(fhp->fh_dentry);
4704 struct rhlist_head *tmp, *list;
4705 struct nfs4_file *fi;
4708 list = rhltable_lookup(&nfs4_file_rhltable, &inode,
4709 nfs4_file_rhash_params);
4710 rhl_for_each_entry_rcu(fi, tmp, list, fi_rlist) {
4711 if (fh_match(&fi->fi_fhandle, &fhp->fh_handle)) {
4712 if (refcount_inc_not_zero(&fi->fi_ref)) {
4723 * On hash insertion, identify entries with the same inode but
4724 * distinct filehandles. They will all be on the list returned
4725 * by rhltable_lookup().
4727 * inode->i_lock prevents racing insertions from adding an entry
4728 * for the same inode/fhp pair twice.
4730 static noinline_for_stack struct nfs4_file *
4731 nfsd4_file_hash_insert(struct nfs4_file *new, const struct svc_fh *fhp)
4733 struct inode *inode = d_inode(fhp->fh_dentry);
4734 struct rhlist_head *tmp, *list;
4735 struct nfs4_file *ret = NULL;
4736 bool alias_found = false;
4737 struct nfs4_file *fi;
4741 spin_lock(&inode->i_lock);
4743 list = rhltable_lookup(&nfs4_file_rhltable, &inode,
4744 nfs4_file_rhash_params);
4745 rhl_for_each_entry_rcu(fi, tmp, list, fi_rlist) {
4746 if (fh_match(&fi->fi_fhandle, &fhp->fh_handle)) {
4747 if (refcount_inc_not_zero(&fi->fi_ref))
4750 fi->fi_aliased = alias_found = true;
4755 nfsd4_file_init(fhp, new);
4756 err = rhltable_insert(&nfs4_file_rhltable, &new->fi_rlist,
4757 nfs4_file_rhash_params);
4761 new->fi_aliased = alias_found;
4765 spin_unlock(&inode->i_lock);
4770 static noinline_for_stack void nfsd4_file_hash_remove(struct nfs4_file *fi)
4772 rhltable_remove(&nfs4_file_rhltable, &fi->fi_rlist,
4773 nfs4_file_rhash_params);
4777 * Called to check deny when READ with all zero stateid or
4778 * WRITE with all zero or all one stateid
4781 nfs4_share_conflict(struct svc_fh *current_fh, unsigned int deny_type)
4783 struct nfs4_file *fp;
4784 __be32 ret = nfs_ok;
4786 fp = nfsd4_file_hash_lookup(current_fh);
4790 /* Check for conflicting share reservations */
4791 spin_lock(&fp->fi_lock);
4792 if (fp->fi_share_deny & deny_type)
4793 ret = nfserr_locked;
4794 spin_unlock(&fp->fi_lock);
4799 static bool nfsd4_deleg_present(const struct inode *inode)
4801 struct file_lock_context *ctx = locks_inode_context(inode);
4803 return ctx && !list_empty_careful(&ctx->flc_lease);
4807 * nfsd_wait_for_delegreturn - wait for delegations to be returned
4808 * @rqstp: the RPC transaction being executed
4809 * @inode: in-core inode of the file being waited for
4811 * The timeout prevents deadlock if all nfsd threads happen to be
4812 * tied up waiting for returning delegations.
4815 * %true: delegation was returned
4816 * %false: timed out waiting for delegreturn
4818 bool nfsd_wait_for_delegreturn(struct svc_rqst *rqstp, struct inode *inode)
4820 long __maybe_unused timeo;
4822 timeo = wait_var_event_timeout(inode, !nfsd4_deleg_present(inode),
4823 NFSD_DELEGRETURN_TIMEOUT);
4824 trace_nfsd_delegret_wakeup(rqstp, inode, timeo);
4828 static void nfsd4_cb_recall_prepare(struct nfsd4_callback *cb)
4830 struct nfs4_delegation *dp = cb_to_delegation(cb);
4831 struct nfsd_net *nn = net_generic(dp->dl_stid.sc_client->net,
4834 block_delegations(&dp->dl_stid.sc_file->fi_fhandle);
4837 * We can't do this in nfsd_break_deleg_cb because it is
4838 * already holding inode->i_lock.
4840 * If the dl_time != 0, then we know that it has already been
4841 * queued for a lease break. Don't queue it again.
4843 spin_lock(&state_lock);
4844 if (delegation_hashed(dp) && dp->dl_time == 0) {
4845 dp->dl_time = ktime_get_boottime_seconds();
4846 list_add_tail(&dp->dl_recall_lru, &nn->del_recall_lru);
4848 spin_unlock(&state_lock);
4851 static int nfsd4_cb_recall_done(struct nfsd4_callback *cb,
4852 struct rpc_task *task)
4854 struct nfs4_delegation *dp = cb_to_delegation(cb);
4856 trace_nfsd_cb_recall_done(&dp->dl_stid.sc_stateid, task);
4858 if (dp->dl_stid.sc_type == NFS4_CLOSED_DELEG_STID ||
4859 dp->dl_stid.sc_type == NFS4_REVOKED_DELEG_STID)
4862 switch (task->tk_status) {
4865 case -NFS4ERR_DELAY:
4866 rpc_delay(task, 2 * HZ);
4869 case -NFS4ERR_BAD_STATEID:
4871 * Race: client probably got cb_recall before open reply
4872 * granting delegation.
4874 if (dp->dl_retries--) {
4875 rpc_delay(task, 2 * HZ);
4884 static void nfsd4_cb_recall_release(struct nfsd4_callback *cb)
4886 struct nfs4_delegation *dp = cb_to_delegation(cb);
4888 nfs4_put_stid(&dp->dl_stid);
4891 static const struct nfsd4_callback_ops nfsd4_cb_recall_ops = {
4892 .prepare = nfsd4_cb_recall_prepare,
4893 .done = nfsd4_cb_recall_done,
4894 .release = nfsd4_cb_recall_release,
4897 static void nfsd_break_one_deleg(struct nfs4_delegation *dp)
4900 * We're assuming the state code never drops its reference
4901 * without first removing the lease. Since we're in this lease
4902 * callback (and since the lease code is serialized by the
4903 * flc_lock) we know the server hasn't removed the lease yet, and
4904 * we know it's safe to take a reference.
4906 refcount_inc(&dp->dl_stid.sc_count);
4907 WARN_ON_ONCE(!nfsd4_run_cb(&dp->dl_recall));
4910 /* Called from break_lease() with flc_lock held. */
4912 nfsd_break_deleg_cb(struct file_lock *fl)
4914 struct nfs4_delegation *dp = (struct nfs4_delegation *)fl->fl_owner;
4915 struct nfs4_file *fp = dp->dl_stid.sc_file;
4916 struct nfs4_client *clp = dp->dl_stid.sc_client;
4917 struct nfsd_net *nn;
4919 trace_nfsd_cb_recall(&dp->dl_stid);
4921 dp->dl_recalled = true;
4922 atomic_inc(&clp->cl_delegs_in_recall);
4923 if (try_to_expire_client(clp)) {
4924 nn = net_generic(clp->net, nfsd_net_id);
4925 mod_delayed_work(laundry_wq, &nn->laundromat_work, 0);
4929 * We don't want the locks code to timeout the lease for us;
4930 * we'll remove it ourself if a delegation isn't returned
4933 fl->fl_break_time = 0;
4935 spin_lock(&fp->fi_lock);
4936 fp->fi_had_conflict = true;
4937 nfsd_break_one_deleg(dp);
4938 spin_unlock(&fp->fi_lock);
4943 * nfsd_breaker_owns_lease - Check if lease conflict was resolved
4944 * @fl: Lock state to check
4947 * %true: Lease conflict was resolved
4948 * %false: Lease conflict was not resolved.
4950 static bool nfsd_breaker_owns_lease(struct file_lock *fl)
4952 struct nfs4_delegation *dl = fl->fl_owner;
4953 struct svc_rqst *rqst;
4954 struct nfs4_client *clp;
4958 rqst = kthread_data(current);
4959 /* Note rq_prog == NFS_ACL_PROGRAM is also possible: */
4960 if (rqst->rq_prog != NFS_PROGRAM || rqst->rq_vers < 4)
4962 clp = *(rqst->rq_lease_breaker);
4963 return dl->dl_stid.sc_client == clp;
4967 nfsd_change_deleg_cb(struct file_lock *onlist, int arg,
4968 struct list_head *dispose)
4970 struct nfs4_delegation *dp = (struct nfs4_delegation *)onlist->fl_owner;
4971 struct nfs4_client *clp = dp->dl_stid.sc_client;
4973 if (arg & F_UNLCK) {
4974 if (dp->dl_recalled)
4975 atomic_dec(&clp->cl_delegs_in_recall);
4976 return lease_modify(onlist, arg, dispose);
4981 static const struct lock_manager_operations nfsd_lease_mng_ops = {
4982 .lm_breaker_owns_lease = nfsd_breaker_owns_lease,
4983 .lm_break = nfsd_break_deleg_cb,
4984 .lm_change = nfsd_change_deleg_cb,
4987 static __be32 nfsd4_check_seqid(struct nfsd4_compound_state *cstate, struct nfs4_stateowner *so, u32 seqid)
4989 if (nfsd4_has_session(cstate))
4991 if (seqid == so->so_seqid - 1)
4992 return nfserr_replay_me;
4993 if (seqid == so->so_seqid)
4995 return nfserr_bad_seqid;
4998 static struct nfs4_client *lookup_clientid(clientid_t *clid, bool sessions,
4999 struct nfsd_net *nn)
5001 struct nfs4_client *found;
5003 spin_lock(&nn->client_lock);
5004 found = find_confirmed_client(clid, sessions, nn);
5006 atomic_inc(&found->cl_rpc_users);
5007 spin_unlock(&nn->client_lock);
5011 static __be32 set_client(clientid_t *clid,
5012 struct nfsd4_compound_state *cstate,
5013 struct nfsd_net *nn)
5016 if (!same_clid(&cstate->clp->cl_clientid, clid))
5017 return nfserr_stale_clientid;
5020 if (STALE_CLIENTID(clid, nn))
5021 return nfserr_stale_clientid;
5023 * We're in the 4.0 case (otherwise the SEQUENCE op would have
5024 * set cstate->clp), so session = false:
5026 cstate->clp = lookup_clientid(clid, false, nn);
5028 return nfserr_expired;
5033 nfsd4_process_open1(struct nfsd4_compound_state *cstate,
5034 struct nfsd4_open *open, struct nfsd_net *nn)
5036 clientid_t *clientid = &open->op_clientid;
5037 struct nfs4_client *clp = NULL;
5038 unsigned int strhashval;
5039 struct nfs4_openowner *oo = NULL;
5043 * In case we need it later, after we've already created the
5044 * file and don't want to risk a further failure:
5046 open->op_file = nfsd4_alloc_file();
5047 if (open->op_file == NULL)
5048 return nfserr_jukebox;
5050 status = set_client(clientid, cstate, nn);
5055 strhashval = ownerstr_hashval(&open->op_owner);
5056 oo = find_openstateowner_str(strhashval, open, clp);
5057 open->op_openowner = oo;
5061 if (!(oo->oo_flags & NFS4_OO_CONFIRMED)) {
5062 /* Replace unconfirmed owners without checking for replay. */
5063 release_openowner(oo);
5064 open->op_openowner = NULL;
5067 status = nfsd4_check_seqid(cstate, &oo->oo_owner, open->op_seqid);
5072 oo = alloc_init_open_stateowner(strhashval, open, cstate);
5074 return nfserr_jukebox;
5075 open->op_openowner = oo;
5077 open->op_stp = nfs4_alloc_open_stateid(clp);
5079 return nfserr_jukebox;
5081 if (nfsd4_has_session(cstate) &&
5082 (cstate->current_fh.fh_export->ex_flags & NFSEXP_PNFS)) {
5083 open->op_odstate = alloc_clnt_odstate(clp);
5084 if (!open->op_odstate)
5085 return nfserr_jukebox;
5091 static inline __be32
5092 nfs4_check_delegmode(struct nfs4_delegation *dp, int flags)
5094 if ((flags & WR_STATE) && (dp->dl_type == NFS4_OPEN_DELEGATE_READ))
5095 return nfserr_openmode;
5100 static int share_access_to_flags(u32 share_access)
5102 return share_access == NFS4_SHARE_ACCESS_READ ? RD_STATE : WR_STATE;
5105 static struct nfs4_delegation *find_deleg_stateid(struct nfs4_client *cl, stateid_t *s)
5107 struct nfs4_stid *ret;
5109 ret = find_stateid_by_type(cl, s,
5110 NFS4_DELEG_STID|NFS4_REVOKED_DELEG_STID);
5113 return delegstateid(ret);
5116 static bool nfsd4_is_deleg_cur(struct nfsd4_open *open)
5118 return open->op_claim_type == NFS4_OPEN_CLAIM_DELEGATE_CUR ||
5119 open->op_claim_type == NFS4_OPEN_CLAIM_DELEG_CUR_FH;
5123 nfs4_check_deleg(struct nfs4_client *cl, struct nfsd4_open *open,
5124 struct nfs4_delegation **dp)
5127 __be32 status = nfserr_bad_stateid;
5128 struct nfs4_delegation *deleg;
5130 deleg = find_deleg_stateid(cl, &open->op_delegate_stateid);
5133 if (deleg->dl_stid.sc_type == NFS4_REVOKED_DELEG_STID) {
5134 nfs4_put_stid(&deleg->dl_stid);
5135 if (cl->cl_minorversion)
5136 status = nfserr_deleg_revoked;
5139 flags = share_access_to_flags(open->op_share_access);
5140 status = nfs4_check_delegmode(deleg, flags);
5142 nfs4_put_stid(&deleg->dl_stid);
5147 if (!nfsd4_is_deleg_cur(open))
5151 open->op_openowner->oo_flags |= NFS4_OO_CONFIRMED;
5155 static inline int nfs4_access_to_access(u32 nfs4_access)
5159 if (nfs4_access & NFS4_SHARE_ACCESS_READ)
5160 flags |= NFSD_MAY_READ;
5161 if (nfs4_access & NFS4_SHARE_ACCESS_WRITE)
5162 flags |= NFSD_MAY_WRITE;
5166 static inline __be32
5167 nfsd4_truncate(struct svc_rqst *rqstp, struct svc_fh *fh,
5168 struct nfsd4_open *open)
5170 struct iattr iattr = {
5171 .ia_valid = ATTR_SIZE,
5174 struct nfsd_attrs attrs = {
5177 if (!open->op_truncate)
5179 if (!(open->op_share_access & NFS4_SHARE_ACCESS_WRITE))
5180 return nfserr_inval;
5181 return nfsd_setattr(rqstp, fh, &attrs, 0, (time64_t)0);
5184 static __be32 nfs4_get_vfs_file(struct svc_rqst *rqstp, struct nfs4_file *fp,
5185 struct svc_fh *cur_fh, struct nfs4_ol_stateid *stp,
5186 struct nfsd4_open *open, bool new_stp)
5188 struct nfsd_file *nf = NULL;
5190 int oflag = nfs4_access_to_omode(open->op_share_access);
5191 int access = nfs4_access_to_access(open->op_share_access);
5192 unsigned char old_access_bmap, old_deny_bmap;
5194 spin_lock(&fp->fi_lock);
5197 * Are we trying to set a deny mode that would conflict with
5200 status = nfs4_file_check_deny(fp, open->op_share_deny);
5201 if (status != nfs_ok) {
5202 if (status != nfserr_share_denied) {
5203 spin_unlock(&fp->fi_lock);
5206 if (nfs4_resolve_deny_conflicts_locked(fp, new_stp,
5207 stp, open->op_share_deny, false))
5208 status = nfserr_jukebox;
5209 spin_unlock(&fp->fi_lock);
5213 /* set access to the file */
5214 status = nfs4_file_get_access(fp, open->op_share_access);
5215 if (status != nfs_ok) {
5216 if (status != nfserr_share_denied) {
5217 spin_unlock(&fp->fi_lock);
5220 if (nfs4_resolve_deny_conflicts_locked(fp, new_stp,
5221 stp, open->op_share_access, true))
5222 status = nfserr_jukebox;
5223 spin_unlock(&fp->fi_lock);
5227 /* Set access bits in stateid */
5228 old_access_bmap = stp->st_access_bmap;
5229 set_access(open->op_share_access, stp);
5231 /* Set new deny mask */
5232 old_deny_bmap = stp->st_deny_bmap;
5233 set_deny(open->op_share_deny, stp);
5234 fp->fi_share_deny |= (open->op_share_deny & NFS4_SHARE_DENY_BOTH);
5236 if (!fp->fi_fds[oflag]) {
5237 spin_unlock(&fp->fi_lock);
5239 status = nfsd_file_acquire_opened(rqstp, cur_fh, access,
5240 open->op_filp, &nf);
5241 if (status != nfs_ok)
5242 goto out_put_access;
5244 spin_lock(&fp->fi_lock);
5245 if (!fp->fi_fds[oflag]) {
5246 fp->fi_fds[oflag] = nf;
5250 spin_unlock(&fp->fi_lock);
5254 status = nfserrno(nfsd_open_break_lease(cur_fh->fh_dentry->d_inode,
5257 goto out_put_access;
5259 status = nfsd4_truncate(rqstp, cur_fh, open);
5261 goto out_put_access;
5265 stp->st_access_bmap = old_access_bmap;
5266 nfs4_file_put_access(fp, open->op_share_access);
5267 reset_union_bmap_deny(bmap_to_share_mode(old_deny_bmap), stp);
5272 nfs4_upgrade_open(struct svc_rqst *rqstp, struct nfs4_file *fp,
5273 struct svc_fh *cur_fh, struct nfs4_ol_stateid *stp,
5274 struct nfsd4_open *open)
5277 unsigned char old_deny_bmap = stp->st_deny_bmap;
5279 if (!test_access(open->op_share_access, stp))
5280 return nfs4_get_vfs_file(rqstp, fp, cur_fh, stp, open, false);
5282 /* test and set deny mode */
5283 spin_lock(&fp->fi_lock);
5284 status = nfs4_file_check_deny(fp, open->op_share_deny);
5287 set_deny(open->op_share_deny, stp);
5288 fp->fi_share_deny |=
5289 (open->op_share_deny & NFS4_SHARE_DENY_BOTH);
5291 case nfserr_share_denied:
5292 if (nfs4_resolve_deny_conflicts_locked(fp, false,
5293 stp, open->op_share_deny, false))
5294 status = nfserr_jukebox;
5297 spin_unlock(&fp->fi_lock);
5299 if (status != nfs_ok)
5302 status = nfsd4_truncate(rqstp, cur_fh, open);
5303 if (status != nfs_ok)
5304 reset_union_bmap_deny(old_deny_bmap, stp);
5308 /* Should we give out recallable state?: */
5309 static bool nfsd4_cb_channel_good(struct nfs4_client *clp)
5311 if (clp->cl_cb_state == NFSD4_CB_UP)
5314 * In the sessions case, since we don't have to establish a
5315 * separate connection for callbacks, we assume it's OK
5316 * until we hear otherwise:
5318 return clp->cl_minorversion && clp->cl_cb_state == NFSD4_CB_UNKNOWN;
5321 static struct file_lock *nfs4_alloc_init_lease(struct nfs4_delegation *dp,
5324 struct file_lock *fl;
5326 fl = locks_alloc_lock();
5329 fl->fl_lmops = &nfsd_lease_mng_ops;
5330 fl->fl_flags = FL_DELEG;
5331 fl->fl_type = flag == NFS4_OPEN_DELEGATE_READ? F_RDLCK: F_WRLCK;
5332 fl->fl_end = OFFSET_MAX;
5333 fl->fl_owner = (fl_owner_t)dp;
5334 fl->fl_pid = current->tgid;
5335 fl->fl_file = dp->dl_stid.sc_file->fi_deleg_file->nf_file;
5339 static int nfsd4_check_conflicting_opens(struct nfs4_client *clp,
5340 struct nfs4_file *fp)
5342 struct nfs4_ol_stateid *st;
5343 struct file *f = fp->fi_deleg_file->nf_file;
5344 struct inode *ino = file_inode(f);
5347 writes = atomic_read(&ino->i_writecount);
5351 * There could be multiple filehandles (hence multiple
5352 * nfs4_files) referencing this file, but that's not too
5353 * common; let's just give up in that case rather than
5354 * trying to go look up all the clients using that other
5355 * nfs4_file as well:
5360 * If there's a close in progress, make sure that we see it
5361 * clear any fi_fds[] entries before we see it decrement
5364 smp_mb__after_atomic();
5366 if (fp->fi_fds[O_WRONLY])
5368 if (fp->fi_fds[O_RDWR])
5371 return -EAGAIN; /* There may be non-NFSv4 writers */
5373 * It's possible there are non-NFSv4 write opens in progress,
5374 * but if they haven't incremented i_writecount yet then they
5375 * also haven't called break lease yet; so, they'll break this
5376 * lease soon enough. So, all that's left to check for is NFSv4
5379 spin_lock(&fp->fi_lock);
5380 list_for_each_entry(st, &fp->fi_stateids, st_perfile) {
5381 if (st->st_openstp == NULL /* it's an open */ &&
5382 access_permit_write(st) &&
5383 st->st_stid.sc_client != clp) {
5384 spin_unlock(&fp->fi_lock);
5388 spin_unlock(&fp->fi_lock);
5390 * There's a small chance that we could be racing with another
5391 * NFSv4 open. However, any open that hasn't added itself to
5392 * the fi_stateids list also hasn't called break_lease yet; so,
5393 * they'll break this lease soon enough.
5399 * It's possible that between opening the dentry and setting the delegation,
5400 * that it has been renamed or unlinked. Redo the lookup to verify that this
5404 nfsd4_verify_deleg_dentry(struct nfsd4_open *open, struct nfs4_file *fp,
5405 struct svc_fh *parent)
5407 struct svc_export *exp;
5408 struct dentry *child;
5411 err = nfsd_lookup_dentry(open->op_rqstp, parent,
5412 open->op_fname, open->op_fnamelen,
5420 if (child != file_dentry(fp->fi_deleg_file->nf_file))
5427 * We avoid breaking delegations held by a client due to its own activity, but
5428 * clearing setuid/setgid bits on a write is an implicit activity and the client
5429 * may not notice and continue using the old mode. Avoid giving out a delegation
5430 * on setuid/setgid files when the client is requesting an open for write.
5433 nfsd4_verify_setuid_write(struct nfsd4_open *open, struct nfsd_file *nf)
5435 struct inode *inode = file_inode(nf->nf_file);
5437 if ((open->op_share_access & NFS4_SHARE_ACCESS_WRITE) &&
5438 (inode->i_mode & (S_ISUID|S_ISGID)))
5443 static struct nfs4_delegation *
5444 nfs4_set_delegation(struct nfsd4_open *open, struct nfs4_ol_stateid *stp,
5445 struct svc_fh *parent)
5448 struct nfs4_client *clp = stp->st_stid.sc_client;
5449 struct nfs4_file *fp = stp->st_stid.sc_file;
5450 struct nfs4_clnt_odstate *odstate = stp->st_clnt_odstate;
5451 struct nfs4_delegation *dp;
5452 struct nfsd_file *nf;
5453 struct file_lock *fl;
5456 * The fi_had_conflict and nfs_get_existing_delegation checks
5457 * here are just optimizations; we'll need to recheck them at
5460 if (fp->fi_had_conflict)
5461 return ERR_PTR(-EAGAIN);
5463 nf = find_readable_file(fp);
5466 * We probably could attempt another open and get a read
5467 * delegation, but for now, don't bother until the
5468 * client actually sends us one.
5470 return ERR_PTR(-EAGAIN);
5472 spin_lock(&state_lock);
5473 spin_lock(&fp->fi_lock);
5474 if (nfs4_delegation_exists(clp, fp))
5476 else if (nfsd4_verify_setuid_write(open, nf))
5478 else if (!fp->fi_deleg_file) {
5479 fp->fi_deleg_file = nf;
5480 /* increment early to prevent fi_deleg_file from being
5482 fp->fi_delegees = 1;
5486 spin_unlock(&fp->fi_lock);
5487 spin_unlock(&state_lock);
5491 return ERR_PTR(status);
5494 dp = alloc_init_deleg(clp, fp, odstate);
5498 fl = nfs4_alloc_init_lease(dp, NFS4_OPEN_DELEGATE_READ);
5500 goto out_clnt_odstate;
5502 status = vfs_setlease(fp->fi_deleg_file->nf_file, fl->fl_type, &fl, NULL);
5504 locks_free_lock(fl);
5506 goto out_clnt_odstate;
5509 status = nfsd4_verify_deleg_dentry(open, fp, parent);
5514 status = nfsd4_check_conflicting_opens(clp, fp);
5519 * Now that the deleg is set, check again to ensure that nothing
5520 * raced in and changed the mode while we weren't lookng.
5522 status = nfsd4_verify_setuid_write(open, fp->fi_deleg_file);
5526 spin_lock(&state_lock);
5527 spin_lock(&fp->fi_lock);
5528 if (fp->fi_had_conflict)
5531 status = hash_delegation_locked(dp, fp);
5532 spin_unlock(&fp->fi_lock);
5533 spin_unlock(&state_lock);
5540 vfs_setlease(fp->fi_deleg_file->nf_file, F_UNLCK, NULL, (void **)&dp);
5542 put_clnt_odstate(dp->dl_clnt_odstate);
5543 nfs4_put_stid(&dp->dl_stid);
5546 return ERR_PTR(status);
5549 static void nfsd4_open_deleg_none_ext(struct nfsd4_open *open, int status)
5551 open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE_EXT;
5552 if (status == -EAGAIN)
5553 open->op_why_no_deleg = WND4_CONTENTION;
5555 open->op_why_no_deleg = WND4_RESOURCE;
5556 switch (open->op_deleg_want) {
5557 case NFS4_SHARE_WANT_READ_DELEG:
5558 case NFS4_SHARE_WANT_WRITE_DELEG:
5559 case NFS4_SHARE_WANT_ANY_DELEG:
5561 case NFS4_SHARE_WANT_CANCEL:
5562 open->op_why_no_deleg = WND4_CANCELLED;
5564 case NFS4_SHARE_WANT_NO_DELEG:
5571 * Attempt to hand out a delegation.
5573 * Note we don't support write delegations, and won't until the vfs has
5574 * proper support for them.
5577 nfs4_open_delegation(struct nfsd4_open *open, struct nfs4_ol_stateid *stp,
5578 struct svc_fh *currentfh)
5580 struct nfs4_delegation *dp;
5581 struct nfs4_openowner *oo = openowner(stp->st_stateowner);
5582 struct nfs4_client *clp = stp->st_stid.sc_client;
5583 struct svc_fh *parent = NULL;
5587 cb_up = nfsd4_cb_channel_good(oo->oo_owner.so_client);
5588 open->op_recall = 0;
5589 switch (open->op_claim_type) {
5590 case NFS4_OPEN_CLAIM_PREVIOUS:
5592 open->op_recall = 1;
5593 if (open->op_delegate_type != NFS4_OPEN_DELEGATE_READ)
5596 case NFS4_OPEN_CLAIM_NULL:
5599 case NFS4_OPEN_CLAIM_FH:
5601 * Let's not give out any delegations till everyone's
5602 * had the chance to reclaim theirs, *and* until
5603 * NLM locks have all been reclaimed:
5605 if (locks_in_grace(clp->net))
5607 if (!cb_up || !(oo->oo_flags & NFS4_OO_CONFIRMED))
5613 dp = nfs4_set_delegation(open, stp, parent);
5617 memcpy(&open->op_delegate_stateid, &dp->dl_stid.sc_stateid, sizeof(dp->dl_stid.sc_stateid));
5619 trace_nfsd_deleg_read(&dp->dl_stid.sc_stateid);
5620 open->op_delegate_type = NFS4_OPEN_DELEGATE_READ;
5621 nfs4_put_stid(&dp->dl_stid);
5624 open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE;
5625 if (open->op_claim_type == NFS4_OPEN_CLAIM_PREVIOUS &&
5626 open->op_delegate_type != NFS4_OPEN_DELEGATE_NONE) {
5627 dprintk("NFSD: WARNING: refusing delegation reclaim\n");
5628 open->op_recall = 1;
5631 /* 4.1 client asking for a delegation? */
5632 if (open->op_deleg_want)
5633 nfsd4_open_deleg_none_ext(open, status);
5637 static void nfsd4_deleg_xgrade_none_ext(struct nfsd4_open *open,
5638 struct nfs4_delegation *dp)
5640 if (open->op_deleg_want == NFS4_SHARE_WANT_READ_DELEG &&
5641 dp->dl_type == NFS4_OPEN_DELEGATE_WRITE) {
5642 open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE_EXT;
5643 open->op_why_no_deleg = WND4_NOT_SUPP_DOWNGRADE;
5644 } else if (open->op_deleg_want == NFS4_SHARE_WANT_WRITE_DELEG &&
5645 dp->dl_type == NFS4_OPEN_DELEGATE_WRITE) {
5646 open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE_EXT;
5647 open->op_why_no_deleg = WND4_NOT_SUPP_UPGRADE;
5649 /* Otherwise the client must be confused wanting a delegation
5650 * it already has, therefore we don't return
5651 * NFS4_OPEN_DELEGATE_NONE_EXT and reason.
5656 * nfsd4_process_open2 - finish open processing
5657 * @rqstp: the RPC transaction being executed
5658 * @current_fh: NFSv4 COMPOUND's current filehandle
5659 * @open: OPEN arguments
5661 * If successful, (1) truncate the file if open->op_truncate was
5662 * set, (2) set open->op_stateid, (3) set open->op_delegation.
5664 * Returns %nfs_ok on success; otherwise an nfs4stat value in
5665 * network byte order is returned.
5668 nfsd4_process_open2(struct svc_rqst *rqstp, struct svc_fh *current_fh, struct nfsd4_open *open)
5670 struct nfsd4_compoundres *resp = rqstp->rq_resp;
5671 struct nfs4_client *cl = open->op_openowner->oo_owner.so_client;
5672 struct nfs4_file *fp = NULL;
5673 struct nfs4_ol_stateid *stp = NULL;
5674 struct nfs4_delegation *dp = NULL;
5676 bool new_stp = false;
5679 * Lookup file; if found, lookup stateid and check open request,
5680 * and check for delegations in the process of being recalled.
5681 * If not found, create the nfs4_file struct
5683 fp = nfsd4_file_hash_insert(open->op_file, current_fh);
5685 return nfserr_jukebox;
5686 if (fp != open->op_file) {
5687 status = nfs4_check_deleg(cl, open, &dp);
5690 stp = nfsd4_find_and_lock_existing_open(fp, open);
5692 open->op_file = NULL;
5693 status = nfserr_bad_stateid;
5694 if (nfsd4_is_deleg_cur(open))
5699 stp = init_open_stateid(fp, open);
5705 * OPEN the file, or upgrade an existing OPEN.
5706 * If truncate fails, the OPEN fails.
5708 * stp is already locked.
5711 /* Stateid was found, this is an OPEN upgrade */
5712 status = nfs4_upgrade_open(rqstp, fp, current_fh, stp, open);
5714 mutex_unlock(&stp->st_mutex);
5718 status = nfs4_get_vfs_file(rqstp, fp, current_fh, stp, open, true);
5720 stp->st_stid.sc_type = NFS4_CLOSED_STID;
5721 release_open_stateid(stp);
5722 mutex_unlock(&stp->st_mutex);
5726 stp->st_clnt_odstate = find_or_hash_clnt_odstate(fp,
5728 if (stp->st_clnt_odstate == open->op_odstate)
5729 open->op_odstate = NULL;
5732 nfs4_inc_and_copy_stateid(&open->op_stateid, &stp->st_stid);
5733 mutex_unlock(&stp->st_mutex);
5735 if (nfsd4_has_session(&resp->cstate)) {
5736 if (open->op_deleg_want & NFS4_SHARE_WANT_NO_DELEG) {
5737 open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE_EXT;
5738 open->op_why_no_deleg = WND4_NOT_WANTED;
5744 * Attempt to hand out a delegation. No error return, because the
5745 * OPEN succeeds even if we fail.
5747 nfs4_open_delegation(open, stp, &resp->cstate.current_fh);
5750 trace_nfsd_open(&stp->st_stid.sc_stateid);
5752 /* 4.1 client trying to upgrade/downgrade delegation? */
5753 if (open->op_delegate_type == NFS4_OPEN_DELEGATE_NONE && dp &&
5754 open->op_deleg_want)
5755 nfsd4_deleg_xgrade_none_ext(open, dp);
5759 if (status == 0 && open->op_claim_type == NFS4_OPEN_CLAIM_PREVIOUS)
5760 open->op_openowner->oo_flags |= NFS4_OO_CONFIRMED;
5762 * To finish the open response, we just need to set the rflags.
5764 open->op_rflags = NFS4_OPEN_RESULT_LOCKTYPE_POSIX;
5765 if (nfsd4_has_session(&resp->cstate))
5766 open->op_rflags |= NFS4_OPEN_RESULT_MAY_NOTIFY_LOCK;
5767 else if (!(open->op_openowner->oo_flags & NFS4_OO_CONFIRMED))
5768 open->op_rflags |= NFS4_OPEN_RESULT_CONFIRM;
5771 nfs4_put_stid(&dp->dl_stid);
5773 nfs4_put_stid(&stp->st_stid);
5778 void nfsd4_cleanup_open_state(struct nfsd4_compound_state *cstate,
5779 struct nfsd4_open *open)
5781 if (open->op_openowner) {
5782 struct nfs4_stateowner *so = &open->op_openowner->oo_owner;
5784 nfsd4_cstate_assign_replay(cstate, so);
5785 nfs4_put_stateowner(so);
5788 kmem_cache_free(file_slab, open->op_file);
5790 nfs4_put_stid(&open->op_stp->st_stid);
5791 if (open->op_odstate)
5792 kmem_cache_free(odstate_slab, open->op_odstate);
5796 nfsd4_renew(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
5797 union nfsd4_op_u *u)
5799 clientid_t *clid = &u->renew;
5800 struct nfs4_client *clp;
5802 struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
5804 trace_nfsd_clid_renew(clid);
5805 status = set_client(clid, cstate, nn);
5809 if (!list_empty(&clp->cl_delegations)
5810 && clp->cl_cb_state != NFSD4_CB_UP)
5811 return nfserr_cb_path_down;
5816 nfsd4_end_grace(struct nfsd_net *nn)
5818 /* do nothing if grace period already ended */
5819 if (nn->grace_ended)
5822 trace_nfsd_grace_complete(nn);
5823 nn->grace_ended = true;
5825 * If the server goes down again right now, an NFSv4
5826 * client will still be allowed to reclaim after it comes back up,
5827 * even if it hasn't yet had a chance to reclaim state this time.
5830 nfsd4_record_grace_done(nn);
5832 * At this point, NFSv4 clients can still reclaim. But if the
5833 * server crashes, any that have not yet reclaimed will be out
5834 * of luck on the next boot.
5836 * (NFSv4.1+ clients are considered to have reclaimed once they
5837 * call RECLAIM_COMPLETE. NFSv4.0 clients are considered to
5838 * have reclaimed after their first OPEN.)
5840 locks_end_grace(&nn->nfsd4_manager);
5842 * At this point, and once lockd and/or any other containers
5843 * exit their grace period, further reclaims will fail and
5844 * regular locking can resume.
5849 * If we've waited a lease period but there are still clients trying to
5850 * reclaim, wait a little longer to give them a chance to finish.
5852 static bool clients_still_reclaiming(struct nfsd_net *nn)
5854 time64_t double_grace_period_end = nn->boot_time +
5855 2 * nn->nfsd4_lease;
5857 if (nn->track_reclaim_completes &&
5858 atomic_read(&nn->nr_reclaim_complete) ==
5859 nn->reclaim_str_hashtbl_size)
5861 if (!nn->somebody_reclaimed)
5863 nn->somebody_reclaimed = false;
5865 * If we've given them *two* lease times to reclaim, and they're
5866 * still not done, give up:
5868 if (ktime_get_boottime_seconds() > double_grace_period_end)
5873 struct laundry_time {
5878 static bool state_expired(struct laundry_time *lt, time64_t last_refresh)
5880 time64_t time_remaining;
5882 if (last_refresh < lt->cutoff)
5884 time_remaining = last_refresh - lt->cutoff;
5885 lt->new_timeo = min(lt->new_timeo, time_remaining);
5889 #ifdef CONFIG_NFSD_V4_2_INTER_SSC
5890 void nfsd4_ssc_init_umount_work(struct nfsd_net *nn)
5892 spin_lock_init(&nn->nfsd_ssc_lock);
5893 INIT_LIST_HEAD(&nn->nfsd_ssc_mount_list);
5894 init_waitqueue_head(&nn->nfsd_ssc_waitq);
5896 EXPORT_SYMBOL_GPL(nfsd4_ssc_init_umount_work);
5899 * This is called when nfsd is being shutdown, after all inter_ssc
5900 * cleanup were done, to destroy the ssc delayed unmount list.
5902 static void nfsd4_ssc_shutdown_umount(struct nfsd_net *nn)
5904 struct nfsd4_ssc_umount_item *ni = NULL;
5905 struct nfsd4_ssc_umount_item *tmp;
5907 spin_lock(&nn->nfsd_ssc_lock);
5908 list_for_each_entry_safe(ni, tmp, &nn->nfsd_ssc_mount_list, nsui_list) {
5909 list_del(&ni->nsui_list);
5910 spin_unlock(&nn->nfsd_ssc_lock);
5911 mntput(ni->nsui_vfsmount);
5913 spin_lock(&nn->nfsd_ssc_lock);
5915 spin_unlock(&nn->nfsd_ssc_lock);
5918 static void nfsd4_ssc_expire_umount(struct nfsd_net *nn)
5920 bool do_wakeup = false;
5921 struct nfsd4_ssc_umount_item *ni = NULL;
5922 struct nfsd4_ssc_umount_item *tmp;
5924 spin_lock(&nn->nfsd_ssc_lock);
5925 list_for_each_entry_safe(ni, tmp, &nn->nfsd_ssc_mount_list, nsui_list) {
5926 if (time_after(jiffies, ni->nsui_expire)) {
5927 if (refcount_read(&ni->nsui_refcnt) > 1)
5930 /* mark being unmount */
5931 ni->nsui_busy = true;
5932 spin_unlock(&nn->nfsd_ssc_lock);
5933 mntput(ni->nsui_vfsmount);
5934 spin_lock(&nn->nfsd_ssc_lock);
5936 /* waiters need to start from begin of list */
5937 list_del(&ni->nsui_list);
5940 /* wakeup ssc_connect waiters */
5947 wake_up_all(&nn->nfsd_ssc_waitq);
5948 spin_unlock(&nn->nfsd_ssc_lock);
5952 /* Check if any lock belonging to this lockowner has any blockers */
5954 nfs4_lockowner_has_blockers(struct nfs4_lockowner *lo)
5956 struct file_lock_context *ctx;
5957 struct nfs4_ol_stateid *stp;
5958 struct nfs4_file *nf;
5960 list_for_each_entry(stp, &lo->lo_owner.so_stateids, st_perstateowner) {
5961 nf = stp->st_stid.sc_file;
5962 ctx = locks_inode_context(nf->fi_inode);
5965 if (locks_owner_has_blockers(ctx, lo))
5972 nfs4_anylock_blockers(struct nfs4_client *clp)
5975 struct nfs4_stateowner *so;
5976 struct nfs4_lockowner *lo;
5978 if (atomic_read(&clp->cl_delegs_in_recall))
5980 spin_lock(&clp->cl_lock);
5981 for (i = 0; i < OWNER_HASH_SIZE; i++) {
5982 list_for_each_entry(so, &clp->cl_ownerstr_hashtbl[i],
5984 if (so->so_is_open_owner)
5987 if (nfs4_lockowner_has_blockers(lo)) {
5988 spin_unlock(&clp->cl_lock);
5993 spin_unlock(&clp->cl_lock);
5998 nfs4_get_client_reaplist(struct nfsd_net *nn, struct list_head *reaplist,
5999 struct laundry_time *lt)
6001 unsigned int maxreap, reapcnt = 0;
6002 struct list_head *pos, *next;
6003 struct nfs4_client *clp;
6005 maxreap = (atomic_read(&nn->nfs4_client_count) >= nn->nfs4_max_clients) ?
6006 NFSD_CLIENT_MAX_TRIM_PER_RUN : 0;
6007 INIT_LIST_HEAD(reaplist);
6008 spin_lock(&nn->client_lock);
6009 list_for_each_safe(pos, next, &nn->client_lru) {
6010 clp = list_entry(pos, struct nfs4_client, cl_lru);
6011 if (clp->cl_state == NFSD4_EXPIRABLE)
6013 if (!state_expired(lt, clp->cl_time))
6015 if (!atomic_read(&clp->cl_rpc_users)) {
6016 if (clp->cl_state == NFSD4_ACTIVE)
6017 atomic_inc(&nn->nfsd_courtesy_clients);
6018 clp->cl_state = NFSD4_COURTESY;
6020 if (!client_has_state(clp))
6022 if (!nfs4_anylock_blockers(clp))
6023 if (reapcnt >= maxreap)
6026 if (!mark_client_expired_locked(clp)) {
6027 list_add(&clp->cl_lru, reaplist);
6031 spin_unlock(&nn->client_lock);
6035 nfs4_get_courtesy_client_reaplist(struct nfsd_net *nn,
6036 struct list_head *reaplist)
6038 unsigned int maxreap = 0, reapcnt = 0;
6039 struct list_head *pos, *next;
6040 struct nfs4_client *clp;
6042 maxreap = NFSD_CLIENT_MAX_TRIM_PER_RUN;
6043 INIT_LIST_HEAD(reaplist);
6045 spin_lock(&nn->client_lock);
6046 list_for_each_safe(pos, next, &nn->client_lru) {
6047 clp = list_entry(pos, struct nfs4_client, cl_lru);
6048 if (clp->cl_state == NFSD4_ACTIVE)
6050 if (reapcnt >= maxreap)
6052 if (!mark_client_expired_locked(clp)) {
6053 list_add(&clp->cl_lru, reaplist);
6057 spin_unlock(&nn->client_lock);
6061 nfs4_process_client_reaplist(struct list_head *reaplist)
6063 struct list_head *pos, *next;
6064 struct nfs4_client *clp;
6066 list_for_each_safe(pos, next, reaplist) {
6067 clp = list_entry(pos, struct nfs4_client, cl_lru);
6068 trace_nfsd_clid_purged(&clp->cl_clientid);
6069 list_del_init(&clp->cl_lru);
6075 nfs4_laundromat(struct nfsd_net *nn)
6077 struct nfs4_openowner *oo;
6078 struct nfs4_delegation *dp;
6079 struct nfs4_ol_stateid *stp;
6080 struct nfsd4_blocked_lock *nbl;
6081 struct list_head *pos, *next, reaplist;
6082 struct laundry_time lt = {
6083 .cutoff = ktime_get_boottime_seconds() - nn->nfsd4_lease,
6084 .new_timeo = nn->nfsd4_lease
6086 struct nfs4_cpntf_state *cps;
6087 copy_stateid_t *cps_t;
6090 if (clients_still_reclaiming(nn)) {
6094 nfsd4_end_grace(nn);
6096 spin_lock(&nn->s2s_cp_lock);
6097 idr_for_each_entry(&nn->s2s_cp_stateids, cps_t, i) {
6098 cps = container_of(cps_t, struct nfs4_cpntf_state, cp_stateid);
6099 if (cps->cp_stateid.cs_type == NFS4_COPYNOTIFY_STID &&
6100 state_expired(<, cps->cpntf_time))
6101 _free_cpntf_state_locked(nn, cps);
6103 spin_unlock(&nn->s2s_cp_lock);
6104 nfs4_get_client_reaplist(nn, &reaplist, <);
6105 nfs4_process_client_reaplist(&reaplist);
6107 spin_lock(&state_lock);
6108 list_for_each_safe(pos, next, &nn->del_recall_lru) {
6109 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
6110 if (!state_expired(<, dp->dl_time))
6112 WARN_ON(!unhash_delegation_locked(dp));
6113 list_add(&dp->dl_recall_lru, &reaplist);
6115 spin_unlock(&state_lock);
6116 while (!list_empty(&reaplist)) {
6117 dp = list_first_entry(&reaplist, struct nfs4_delegation,
6119 list_del_init(&dp->dl_recall_lru);
6120 revoke_delegation(dp);
6123 spin_lock(&nn->client_lock);
6124 while (!list_empty(&nn->close_lru)) {
6125 oo = list_first_entry(&nn->close_lru, struct nfs4_openowner,
6127 if (!state_expired(<, oo->oo_time))
6129 list_del_init(&oo->oo_close_lru);
6130 stp = oo->oo_last_closed_stid;
6131 oo->oo_last_closed_stid = NULL;
6132 spin_unlock(&nn->client_lock);
6133 nfs4_put_stid(&stp->st_stid);
6134 spin_lock(&nn->client_lock);
6136 spin_unlock(&nn->client_lock);
6139 * It's possible for a client to try and acquire an already held lock
6140 * that is being held for a long time, and then lose interest in it.
6141 * So, we clean out any un-revisited request after a lease period
6142 * under the assumption that the client is no longer interested.
6144 * RFC5661, sec. 9.6 states that the client must not rely on getting
6145 * notifications and must continue to poll for locks, even when the
6146 * server supports them. Thus this shouldn't lead to clients blocking
6147 * indefinitely once the lock does become free.
6149 BUG_ON(!list_empty(&reaplist));
6150 spin_lock(&nn->blocked_locks_lock);
6151 while (!list_empty(&nn->blocked_locks_lru)) {
6152 nbl = list_first_entry(&nn->blocked_locks_lru,
6153 struct nfsd4_blocked_lock, nbl_lru);
6154 if (!state_expired(<, nbl->nbl_time))
6156 list_move(&nbl->nbl_lru, &reaplist);
6157 list_del_init(&nbl->nbl_list);
6159 spin_unlock(&nn->blocked_locks_lock);
6161 while (!list_empty(&reaplist)) {
6162 nbl = list_first_entry(&reaplist,
6163 struct nfsd4_blocked_lock, nbl_lru);
6164 list_del_init(&nbl->nbl_lru);
6165 free_blocked_lock(nbl);
6167 #ifdef CONFIG_NFSD_V4_2_INTER_SSC
6168 /* service the server-to-server copy delayed unmount list */
6169 nfsd4_ssc_expire_umount(nn);
6172 return max_t(time64_t, lt.new_timeo, NFSD_LAUNDROMAT_MINTIMEOUT);
6175 static void laundromat_main(struct work_struct *);
6178 laundromat_main(struct work_struct *laundry)
6181 struct delayed_work *dwork = to_delayed_work(laundry);
6182 struct nfsd_net *nn = container_of(dwork, struct nfsd_net,
6185 t = nfs4_laundromat(nn);
6186 queue_delayed_work(laundry_wq, &nn->laundromat_work, t*HZ);
6190 courtesy_client_reaper(struct nfsd_net *nn)
6192 struct list_head reaplist;
6194 nfs4_get_courtesy_client_reaplist(nn, &reaplist);
6195 nfs4_process_client_reaplist(&reaplist);
6199 deleg_reaper(struct nfsd_net *nn)
6201 struct list_head *pos, *next;
6202 struct nfs4_client *clp;
6203 struct list_head cblist;
6205 INIT_LIST_HEAD(&cblist);
6206 spin_lock(&nn->client_lock);
6207 list_for_each_safe(pos, next, &nn->client_lru) {
6208 clp = list_entry(pos, struct nfs4_client, cl_lru);
6209 if (clp->cl_state != NFSD4_ACTIVE ||
6210 list_empty(&clp->cl_delegations) ||
6211 atomic_read(&clp->cl_delegs_in_recall) ||
6212 test_bit(NFSD4_CLIENT_CB_RECALL_ANY, &clp->cl_flags) ||
6213 (ktime_get_boottime_seconds() -
6214 clp->cl_ra_time < 5)) {
6217 list_add(&clp->cl_ra_cblist, &cblist);
6219 /* release in nfsd4_cb_recall_any_release */
6220 atomic_inc(&clp->cl_rpc_users);
6221 set_bit(NFSD4_CLIENT_CB_RECALL_ANY, &clp->cl_flags);
6222 clp->cl_ra_time = ktime_get_boottime_seconds();
6224 spin_unlock(&nn->client_lock);
6226 while (!list_empty(&cblist)) {
6227 clp = list_first_entry(&cblist, struct nfs4_client,
6229 list_del_init(&clp->cl_ra_cblist);
6230 clp->cl_ra->ra_keep = 0;
6231 clp->cl_ra->ra_bmval[0] = BIT(RCA4_TYPE_MASK_RDATA_DLG);
6232 trace_nfsd_cb_recall_any(clp->cl_ra);
6233 nfsd4_run_cb(&clp->cl_ra->ra_cb);
6238 nfsd4_state_shrinker_worker(struct work_struct *work)
6240 struct nfsd_net *nn = container_of(work, struct nfsd_net,
6241 nfsd_shrinker_work);
6243 courtesy_client_reaper(nn);
6247 static inline __be32 nfs4_check_fh(struct svc_fh *fhp, struct nfs4_stid *stp)
6249 if (!fh_match(&fhp->fh_handle, &stp->sc_file->fi_fhandle))
6250 return nfserr_bad_stateid;
6255 __be32 nfs4_check_openmode(struct nfs4_ol_stateid *stp, int flags)
6257 __be32 status = nfserr_openmode;
6259 /* For lock stateid's, we test the parent open, not the lock: */
6260 if (stp->st_openstp)
6261 stp = stp->st_openstp;
6262 if ((flags & WR_STATE) && !access_permit_write(stp))
6264 if ((flags & RD_STATE) && !access_permit_read(stp))
6271 static inline __be32
6272 check_special_stateids(struct net *net, svc_fh *current_fh, stateid_t *stateid, int flags)
6274 if (ONE_STATEID(stateid) && (flags & RD_STATE))
6276 else if (opens_in_grace(net)) {
6277 /* Answer in remaining cases depends on existence of
6278 * conflicting state; so we must wait out the grace period. */
6279 return nfserr_grace;
6280 } else if (flags & WR_STATE)
6281 return nfs4_share_conflict(current_fh,
6282 NFS4_SHARE_DENY_WRITE);
6283 else /* (flags & RD_STATE) && ZERO_STATEID(stateid) */
6284 return nfs4_share_conflict(current_fh,
6285 NFS4_SHARE_DENY_READ);
6288 static __be32 check_stateid_generation(stateid_t *in, stateid_t *ref, bool has_session)
6291 * When sessions are used the stateid generation number is ignored
6294 if (has_session && in->si_generation == 0)
6297 if (in->si_generation == ref->si_generation)
6300 /* If the client sends us a stateid from the future, it's buggy: */
6301 if (nfsd4_stateid_generation_after(in, ref))
6302 return nfserr_bad_stateid;
6304 * However, we could see a stateid from the past, even from a
6305 * non-buggy client. For example, if the client sends a lock
6306 * while some IO is outstanding, the lock may bump si_generation
6307 * while the IO is still in flight. The client could avoid that
6308 * situation by waiting for responses on all the IO requests,
6309 * but better performance may result in retrying IO that
6310 * receives an old_stateid error if requests are rarely
6311 * reordered in flight:
6313 return nfserr_old_stateid;
6316 static __be32 nfsd4_stid_check_stateid_generation(stateid_t *in, struct nfs4_stid *s, bool has_session)
6320 spin_lock(&s->sc_lock);
6321 ret = nfsd4_verify_open_stid(s);
6323 ret = check_stateid_generation(in, &s->sc_stateid, has_session);
6324 spin_unlock(&s->sc_lock);
6328 static __be32 nfsd4_check_openowner_confirmed(struct nfs4_ol_stateid *ols)
6330 if (ols->st_stateowner->so_is_open_owner &&
6331 !(openowner(ols->st_stateowner)->oo_flags & NFS4_OO_CONFIRMED))
6332 return nfserr_bad_stateid;
6336 static __be32 nfsd4_validate_stateid(struct nfs4_client *cl, stateid_t *stateid)
6338 struct nfs4_stid *s;
6339 __be32 status = nfserr_bad_stateid;
6341 if (ZERO_STATEID(stateid) || ONE_STATEID(stateid) ||
6342 CLOSE_STATEID(stateid))
6344 if (!same_clid(&stateid->si_opaque.so_clid, &cl->cl_clientid))
6346 spin_lock(&cl->cl_lock);
6347 s = find_stateid_locked(cl, stateid);
6350 status = nfsd4_stid_check_stateid_generation(stateid, s, 1);
6353 switch (s->sc_type) {
6354 case NFS4_DELEG_STID:
6357 case NFS4_REVOKED_DELEG_STID:
6358 status = nfserr_deleg_revoked;
6360 case NFS4_OPEN_STID:
6361 case NFS4_LOCK_STID:
6362 status = nfsd4_check_openowner_confirmed(openlockstateid(s));
6365 printk("unknown stateid type %x\n", s->sc_type);
6367 case NFS4_CLOSED_STID:
6368 case NFS4_CLOSED_DELEG_STID:
6369 status = nfserr_bad_stateid;
6372 spin_unlock(&cl->cl_lock);
6377 nfsd4_lookup_stateid(struct nfsd4_compound_state *cstate,
6378 stateid_t *stateid, unsigned char typemask,
6379 struct nfs4_stid **s, struct nfsd_net *nn)
6382 struct nfs4_stid *stid;
6383 bool return_revoked = false;
6386 * only return revoked delegations if explicitly asked.
6387 * otherwise we report revoked or bad_stateid status.
6389 if (typemask & NFS4_REVOKED_DELEG_STID)
6390 return_revoked = true;
6391 else if (typemask & NFS4_DELEG_STID)
6392 typemask |= NFS4_REVOKED_DELEG_STID;
6394 if (ZERO_STATEID(stateid) || ONE_STATEID(stateid) ||
6395 CLOSE_STATEID(stateid))
6396 return nfserr_bad_stateid;
6397 status = set_client(&stateid->si_opaque.so_clid, cstate, nn);
6398 if (status == nfserr_stale_clientid) {
6399 if (cstate->session)
6400 return nfserr_bad_stateid;
6401 return nfserr_stale_stateid;
6405 stid = find_stateid_by_type(cstate->clp, stateid, typemask);
6407 return nfserr_bad_stateid;
6408 if ((stid->sc_type == NFS4_REVOKED_DELEG_STID) && !return_revoked) {
6409 nfs4_put_stid(stid);
6410 if (cstate->minorversion)
6411 return nfserr_deleg_revoked;
6412 return nfserr_bad_stateid;
6418 static struct nfsd_file *
6419 nfs4_find_file(struct nfs4_stid *s, int flags)
6421 struct nfsd_file *ret = NULL;
6426 switch (s->sc_type) {
6427 case NFS4_DELEG_STID:
6428 spin_lock(&s->sc_file->fi_lock);
6429 ret = nfsd_file_get(s->sc_file->fi_deleg_file);
6430 spin_unlock(&s->sc_file->fi_lock);
6432 case NFS4_OPEN_STID:
6433 case NFS4_LOCK_STID:
6434 if (flags & RD_STATE)
6435 ret = find_readable_file(s->sc_file);
6437 ret = find_writeable_file(s->sc_file);
6444 nfs4_check_olstateid(struct nfs4_ol_stateid *ols, int flags)
6448 status = nfsd4_check_openowner_confirmed(ols);
6451 return nfs4_check_openmode(ols, flags);
6455 nfs4_check_file(struct svc_rqst *rqstp, struct svc_fh *fhp, struct nfs4_stid *s,
6456 struct nfsd_file **nfp, int flags)
6458 int acc = (flags & RD_STATE) ? NFSD_MAY_READ : NFSD_MAY_WRITE;
6459 struct nfsd_file *nf;
6462 nf = nfs4_find_file(s, flags);
6464 status = nfsd_permission(rqstp, fhp->fh_export, fhp->fh_dentry,
6465 acc | NFSD_MAY_OWNER_OVERRIDE);
6471 status = nfsd_file_acquire(rqstp, fhp, acc, &nf);
6480 _free_cpntf_state_locked(struct nfsd_net *nn, struct nfs4_cpntf_state *cps)
6482 WARN_ON_ONCE(cps->cp_stateid.cs_type != NFS4_COPYNOTIFY_STID);
6483 if (!refcount_dec_and_test(&cps->cp_stateid.cs_count))
6485 list_del(&cps->cp_list);
6486 idr_remove(&nn->s2s_cp_stateids,
6487 cps->cp_stateid.cs_stid.si_opaque.so_id);
6491 * A READ from an inter server to server COPY will have a
6492 * copy stateid. Look up the copy notify stateid from the
6493 * idr structure and take a reference on it.
6495 __be32 manage_cpntf_state(struct nfsd_net *nn, stateid_t *st,
6496 struct nfs4_client *clp,
6497 struct nfs4_cpntf_state **cps)
6499 copy_stateid_t *cps_t;
6500 struct nfs4_cpntf_state *state = NULL;
6502 if (st->si_opaque.so_clid.cl_id != nn->s2s_cp_cl_id)
6503 return nfserr_bad_stateid;
6504 spin_lock(&nn->s2s_cp_lock);
6505 cps_t = idr_find(&nn->s2s_cp_stateids, st->si_opaque.so_id);
6507 state = container_of(cps_t, struct nfs4_cpntf_state,
6509 if (state->cp_stateid.cs_type != NFS4_COPYNOTIFY_STID) {
6514 refcount_inc(&state->cp_stateid.cs_count);
6516 _free_cpntf_state_locked(nn, state);
6519 spin_unlock(&nn->s2s_cp_lock);
6521 return nfserr_bad_stateid;
6527 static __be32 find_cpntf_state(struct nfsd_net *nn, stateid_t *st,
6528 struct nfs4_stid **stid)
6531 struct nfs4_cpntf_state *cps = NULL;
6532 struct nfs4_client *found;
6534 status = manage_cpntf_state(nn, st, NULL, &cps);
6538 cps->cpntf_time = ktime_get_boottime_seconds();
6540 status = nfserr_expired;
6541 found = lookup_clientid(&cps->cp_p_clid, true, nn);
6545 *stid = find_stateid_by_type(found, &cps->cp_p_stateid,
6546 NFS4_DELEG_STID|NFS4_OPEN_STID|NFS4_LOCK_STID);
6550 status = nfserr_bad_stateid;
6552 put_client_renew(found);
6554 nfs4_put_cpntf_state(nn, cps);
6558 void nfs4_put_cpntf_state(struct nfsd_net *nn, struct nfs4_cpntf_state *cps)
6560 spin_lock(&nn->s2s_cp_lock);
6561 _free_cpntf_state_locked(nn, cps);
6562 spin_unlock(&nn->s2s_cp_lock);
6566 * nfs4_preprocess_stateid_op - find and prep stateid for an operation
6567 * @rqstp: incoming request from client
6568 * @cstate: current compound state
6569 * @fhp: filehandle associated with requested stateid
6570 * @stateid: stateid (provided by client)
6571 * @flags: flags describing type of operation to be done
6572 * @nfp: optional nfsd_file return pointer (may be NULL)
6573 * @cstid: optional returned nfs4_stid pointer (may be NULL)
6575 * Given info from the client, look up a nfs4_stid for the operation. On
6576 * success, it returns a reference to the nfs4_stid and/or the nfsd_file
6577 * associated with it.
6580 nfs4_preprocess_stateid_op(struct svc_rqst *rqstp,
6581 struct nfsd4_compound_state *cstate, struct svc_fh *fhp,
6582 stateid_t *stateid, int flags, struct nfsd_file **nfp,
6583 struct nfs4_stid **cstid)
6585 struct net *net = SVC_NET(rqstp);
6586 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
6587 struct nfs4_stid *s = NULL;
6593 if (ZERO_STATEID(stateid) || ONE_STATEID(stateid)) {
6595 status = nfserr_bad_stateid;
6597 status = check_special_stateids(net, fhp, stateid,
6602 status = nfsd4_lookup_stateid(cstate, stateid,
6603 NFS4_DELEG_STID|NFS4_OPEN_STID|NFS4_LOCK_STID,
6605 if (status == nfserr_bad_stateid)
6606 status = find_cpntf_state(nn, stateid, &s);
6609 status = nfsd4_stid_check_stateid_generation(stateid, s,
6610 nfsd4_has_session(cstate));
6614 switch (s->sc_type) {
6615 case NFS4_DELEG_STID:
6616 status = nfs4_check_delegmode(delegstateid(s), flags);
6618 case NFS4_OPEN_STID:
6619 case NFS4_LOCK_STID:
6620 status = nfs4_check_olstateid(openlockstateid(s), flags);
6623 status = nfserr_bad_stateid;
6628 status = nfs4_check_fh(fhp, s);
6631 if (status == nfs_ok && nfp)
6632 status = nfs4_check_file(rqstp, fhp, s, nfp, flags);
6635 if (!status && cstid)
6644 * Test if the stateid is valid
6647 nfsd4_test_stateid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
6648 union nfsd4_op_u *u)
6650 struct nfsd4_test_stateid *test_stateid = &u->test_stateid;
6651 struct nfsd4_test_stateid_id *stateid;
6652 struct nfs4_client *cl = cstate->clp;
6654 list_for_each_entry(stateid, &test_stateid->ts_stateid_list, ts_id_list)
6655 stateid->ts_id_status =
6656 nfsd4_validate_stateid(cl, &stateid->ts_id_stateid);
6662 nfsd4_free_lock_stateid(stateid_t *stateid, struct nfs4_stid *s)
6664 struct nfs4_ol_stateid *stp = openlockstateid(s);
6667 ret = nfsd4_lock_ol_stateid(stp);
6671 ret = check_stateid_generation(stateid, &s->sc_stateid, 1);
6675 ret = nfserr_locks_held;
6676 if (check_for_locks(stp->st_stid.sc_file,
6677 lockowner(stp->st_stateowner)))
6680 release_lock_stateid(stp);
6684 mutex_unlock(&stp->st_mutex);
6691 nfsd4_free_stateid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
6692 union nfsd4_op_u *u)
6694 struct nfsd4_free_stateid *free_stateid = &u->free_stateid;
6695 stateid_t *stateid = &free_stateid->fr_stateid;
6696 struct nfs4_stid *s;
6697 struct nfs4_delegation *dp;
6698 struct nfs4_client *cl = cstate->clp;
6699 __be32 ret = nfserr_bad_stateid;
6701 spin_lock(&cl->cl_lock);
6702 s = find_stateid_locked(cl, stateid);
6705 spin_lock(&s->sc_lock);
6706 switch (s->sc_type) {
6707 case NFS4_DELEG_STID:
6708 ret = nfserr_locks_held;
6710 case NFS4_OPEN_STID:
6711 ret = check_stateid_generation(stateid, &s->sc_stateid, 1);
6714 ret = nfserr_locks_held;
6716 case NFS4_LOCK_STID:
6717 spin_unlock(&s->sc_lock);
6718 refcount_inc(&s->sc_count);
6719 spin_unlock(&cl->cl_lock);
6720 ret = nfsd4_free_lock_stateid(stateid, s);
6722 case NFS4_REVOKED_DELEG_STID:
6723 spin_unlock(&s->sc_lock);
6724 dp = delegstateid(s);
6725 list_del_init(&dp->dl_recall_lru);
6726 spin_unlock(&cl->cl_lock);
6730 /* Default falls through and returns nfserr_bad_stateid */
6732 spin_unlock(&s->sc_lock);
6734 spin_unlock(&cl->cl_lock);
6742 return (type == NFS4_READW_LT || type == NFS4_READ_LT) ?
6743 RD_STATE : WR_STATE;
6746 static __be32 nfs4_seqid_op_checks(struct nfsd4_compound_state *cstate, stateid_t *stateid, u32 seqid, struct nfs4_ol_stateid *stp)
6748 struct svc_fh *current_fh = &cstate->current_fh;
6749 struct nfs4_stateowner *sop = stp->st_stateowner;
6752 status = nfsd4_check_seqid(cstate, sop, seqid);
6755 status = nfsd4_lock_ol_stateid(stp);
6756 if (status != nfs_ok)
6758 status = check_stateid_generation(stateid, &stp->st_stid.sc_stateid, nfsd4_has_session(cstate));
6759 if (status == nfs_ok)
6760 status = nfs4_check_fh(current_fh, &stp->st_stid);
6761 if (status != nfs_ok)
6762 mutex_unlock(&stp->st_mutex);
6767 * nfs4_preprocess_seqid_op - find and prep an ol_stateid for a seqid-morphing op
6768 * @cstate: compund state
6769 * @seqid: seqid (provided by client)
6770 * @stateid: stateid (provided by client)
6771 * @typemask: mask of allowable types for this operation
6772 * @stpp: return pointer for the stateid found
6773 * @nn: net namespace for request
6775 * Given a stateid+seqid from a client, look up an nfs4_ol_stateid and
6776 * return it in @stpp. On a nfs_ok return, the returned stateid will
6777 * have its st_mutex locked.
6780 nfs4_preprocess_seqid_op(struct nfsd4_compound_state *cstate, u32 seqid,
6781 stateid_t *stateid, char typemask,
6782 struct nfs4_ol_stateid **stpp,
6783 struct nfsd_net *nn)
6786 struct nfs4_stid *s;
6787 struct nfs4_ol_stateid *stp = NULL;
6789 trace_nfsd_preprocess(seqid, stateid);
6792 status = nfsd4_lookup_stateid(cstate, stateid, typemask, &s, nn);
6795 stp = openlockstateid(s);
6796 nfsd4_cstate_assign_replay(cstate, stp->st_stateowner);
6798 status = nfs4_seqid_op_checks(cstate, stateid, seqid, stp);
6802 nfs4_put_stid(&stp->st_stid);
6806 static __be32 nfs4_preprocess_confirmed_seqid_op(struct nfsd4_compound_state *cstate, u32 seqid,
6807 stateid_t *stateid, struct nfs4_ol_stateid **stpp, struct nfsd_net *nn)
6810 struct nfs4_openowner *oo;
6811 struct nfs4_ol_stateid *stp;
6813 status = nfs4_preprocess_seqid_op(cstate, seqid, stateid,
6814 NFS4_OPEN_STID, &stp, nn);
6817 oo = openowner(stp->st_stateowner);
6818 if (!(oo->oo_flags & NFS4_OO_CONFIRMED)) {
6819 mutex_unlock(&stp->st_mutex);
6820 nfs4_put_stid(&stp->st_stid);
6821 return nfserr_bad_stateid;
6828 nfsd4_open_confirm(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
6829 union nfsd4_op_u *u)
6831 struct nfsd4_open_confirm *oc = &u->open_confirm;
6833 struct nfs4_openowner *oo;
6834 struct nfs4_ol_stateid *stp;
6835 struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
6837 dprintk("NFSD: nfsd4_open_confirm on file %pd\n",
6838 cstate->current_fh.fh_dentry);
6840 status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0);
6844 status = nfs4_preprocess_seqid_op(cstate,
6845 oc->oc_seqid, &oc->oc_req_stateid,
6846 NFS4_OPEN_STID, &stp, nn);
6849 oo = openowner(stp->st_stateowner);
6850 status = nfserr_bad_stateid;
6851 if (oo->oo_flags & NFS4_OO_CONFIRMED) {
6852 mutex_unlock(&stp->st_mutex);
6855 oo->oo_flags |= NFS4_OO_CONFIRMED;
6856 nfs4_inc_and_copy_stateid(&oc->oc_resp_stateid, &stp->st_stid);
6857 mutex_unlock(&stp->st_mutex);
6858 trace_nfsd_open_confirm(oc->oc_seqid, &stp->st_stid.sc_stateid);
6859 nfsd4_client_record_create(oo->oo_owner.so_client);
6862 nfs4_put_stid(&stp->st_stid);
6864 nfsd4_bump_seqid(cstate, status);
6868 static inline void nfs4_stateid_downgrade_bit(struct nfs4_ol_stateid *stp, u32 access)
6870 if (!test_access(access, stp))
6872 nfs4_file_put_access(stp->st_stid.sc_file, access);
6873 clear_access(access, stp);
6876 static inline void nfs4_stateid_downgrade(struct nfs4_ol_stateid *stp, u32 to_access)
6878 switch (to_access) {
6879 case NFS4_SHARE_ACCESS_READ:
6880 nfs4_stateid_downgrade_bit(stp, NFS4_SHARE_ACCESS_WRITE);
6881 nfs4_stateid_downgrade_bit(stp, NFS4_SHARE_ACCESS_BOTH);
6883 case NFS4_SHARE_ACCESS_WRITE:
6884 nfs4_stateid_downgrade_bit(stp, NFS4_SHARE_ACCESS_READ);
6885 nfs4_stateid_downgrade_bit(stp, NFS4_SHARE_ACCESS_BOTH);
6887 case NFS4_SHARE_ACCESS_BOTH:
6895 nfsd4_open_downgrade(struct svc_rqst *rqstp,
6896 struct nfsd4_compound_state *cstate, union nfsd4_op_u *u)
6898 struct nfsd4_open_downgrade *od = &u->open_downgrade;
6900 struct nfs4_ol_stateid *stp;
6901 struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
6903 dprintk("NFSD: nfsd4_open_downgrade on file %pd\n",
6904 cstate->current_fh.fh_dentry);
6906 /* We don't yet support WANT bits: */
6907 if (od->od_deleg_want)
6908 dprintk("NFSD: %s: od_deleg_want=0x%x ignored\n", __func__,
6911 status = nfs4_preprocess_confirmed_seqid_op(cstate, od->od_seqid,
6912 &od->od_stateid, &stp, nn);
6915 status = nfserr_inval;
6916 if (!test_access(od->od_share_access, stp)) {
6917 dprintk("NFSD: access not a subset of current bitmap: 0x%hhx, input access=%08x\n",
6918 stp->st_access_bmap, od->od_share_access);
6921 if (!test_deny(od->od_share_deny, stp)) {
6922 dprintk("NFSD: deny not a subset of current bitmap: 0x%hhx, input deny=%08x\n",
6923 stp->st_deny_bmap, od->od_share_deny);
6926 nfs4_stateid_downgrade(stp, od->od_share_access);
6927 reset_union_bmap_deny(od->od_share_deny, stp);
6928 nfs4_inc_and_copy_stateid(&od->od_stateid, &stp->st_stid);
6931 mutex_unlock(&stp->st_mutex);
6932 nfs4_put_stid(&stp->st_stid);
6934 nfsd4_bump_seqid(cstate, status);
6938 static void nfsd4_close_open_stateid(struct nfs4_ol_stateid *s)
6940 struct nfs4_client *clp = s->st_stid.sc_client;
6942 LIST_HEAD(reaplist);
6943 struct nfs4_ol_stateid *stp;
6945 spin_lock(&clp->cl_lock);
6946 unhashed = unhash_open_stateid(s, &reaplist);
6948 if (clp->cl_minorversion) {
6950 put_ol_stateid_locked(s, &reaplist);
6951 spin_unlock(&clp->cl_lock);
6952 list_for_each_entry(stp, &reaplist, st_locks)
6953 nfs4_free_cpntf_statelist(clp->net, &stp->st_stid);
6954 free_ol_stateid_reaplist(&reaplist);
6956 spin_unlock(&clp->cl_lock);
6957 free_ol_stateid_reaplist(&reaplist);
6959 move_to_close_lru(s, clp->net);
6964 * nfs4_unlock_state() called after encode
6967 nfsd4_close(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
6968 union nfsd4_op_u *u)
6970 struct nfsd4_close *close = &u->close;
6972 struct nfs4_ol_stateid *stp;
6973 struct net *net = SVC_NET(rqstp);
6974 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
6976 dprintk("NFSD: nfsd4_close on file %pd\n",
6977 cstate->current_fh.fh_dentry);
6979 status = nfs4_preprocess_seqid_op(cstate, close->cl_seqid,
6981 NFS4_OPEN_STID|NFS4_CLOSED_STID,
6983 nfsd4_bump_seqid(cstate, status);
6987 stp->st_stid.sc_type = NFS4_CLOSED_STID;
6990 * Technically we don't _really_ have to increment or copy it, since
6991 * it should just be gone after this operation and we clobber the
6992 * copied value below, but we continue to do so here just to ensure
6993 * that racing ops see that there was a state change.
6995 nfs4_inc_and_copy_stateid(&close->cl_stateid, &stp->st_stid);
6997 nfsd4_close_open_stateid(stp);
6998 mutex_unlock(&stp->st_mutex);
7000 /* v4.1+ suggests that we send a special stateid in here, since the
7001 * clients should just ignore this anyway. Since this is not useful
7002 * for v4.0 clients either, we set it to the special close_stateid
7005 * See RFC5661 section 18.2.4, and RFC7530 section 16.2.5
7007 memcpy(&close->cl_stateid, &close_stateid, sizeof(close->cl_stateid));
7009 /* put reference from nfs4_preprocess_seqid_op */
7010 nfs4_put_stid(&stp->st_stid);
7016 nfsd4_delegreturn(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
7017 union nfsd4_op_u *u)
7019 struct nfsd4_delegreturn *dr = &u->delegreturn;
7020 struct nfs4_delegation *dp;
7021 stateid_t *stateid = &dr->dr_stateid;
7022 struct nfs4_stid *s;
7024 struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
7026 if ((status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0)))
7029 status = nfsd4_lookup_stateid(cstate, stateid, NFS4_DELEG_STID, &s, nn);
7032 dp = delegstateid(s);
7033 status = nfsd4_stid_check_stateid_generation(stateid, &dp->dl_stid, nfsd4_has_session(cstate));
7037 trace_nfsd_deleg_return(stateid);
7038 wake_up_var(d_inode(cstate->current_fh.fh_dentry));
7039 destroy_delegation(dp);
7041 nfs4_put_stid(&dp->dl_stid);
7046 /* last octet in a range */
7048 last_byte_offset(u64 start, u64 len)
7054 return end > start ? end - 1: NFS4_MAX_UINT64;
7058 * TODO: Linux file offsets are _signed_ 64-bit quantities, which means that
7059 * we can't properly handle lock requests that go beyond the (2^63 - 1)-th
7060 * byte, because of sign extension problems. Since NFSv4 calls for 64-bit
7061 * locking, this prevents us from being completely protocol-compliant. The
7062 * real solution to this problem is to start using unsigned file offsets in
7063 * the VFS, but this is a very deep change!
7066 nfs4_transform_lock_offset(struct file_lock *lock)
7068 if (lock->fl_start < 0)
7069 lock->fl_start = OFFSET_MAX;
7070 if (lock->fl_end < 0)
7071 lock->fl_end = OFFSET_MAX;
7075 nfsd4_lm_get_owner(fl_owner_t owner)
7077 struct nfs4_lockowner *lo = (struct nfs4_lockowner *)owner;
7079 nfs4_get_stateowner(&lo->lo_owner);
7084 nfsd4_lm_put_owner(fl_owner_t owner)
7086 struct nfs4_lockowner *lo = (struct nfs4_lockowner *)owner;
7089 nfs4_put_stateowner(&lo->lo_owner);
7092 /* return pointer to struct nfs4_client if client is expirable */
7094 nfsd4_lm_lock_expirable(struct file_lock *cfl)
7096 struct nfs4_lockowner *lo = (struct nfs4_lockowner *)cfl->fl_owner;
7097 struct nfs4_client *clp = lo->lo_owner.so_client;
7098 struct nfsd_net *nn;
7100 if (try_to_expire_client(clp)) {
7101 nn = net_generic(clp->net, nfsd_net_id);
7102 mod_delayed_work(laundry_wq, &nn->laundromat_work, 0);
7108 /* schedule laundromat to run immediately and wait for it to complete */
7110 nfsd4_lm_expire_lock(void)
7112 flush_workqueue(laundry_wq);
7116 nfsd4_lm_notify(struct file_lock *fl)
7118 struct nfs4_lockowner *lo = (struct nfs4_lockowner *)fl->fl_owner;
7119 struct net *net = lo->lo_owner.so_client->net;
7120 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
7121 struct nfsd4_blocked_lock *nbl = container_of(fl,
7122 struct nfsd4_blocked_lock, nbl_lock);
7125 /* An empty list means that something else is going to be using it */
7126 spin_lock(&nn->blocked_locks_lock);
7127 if (!list_empty(&nbl->nbl_list)) {
7128 list_del_init(&nbl->nbl_list);
7129 list_del_init(&nbl->nbl_lru);
7132 spin_unlock(&nn->blocked_locks_lock);
7135 trace_nfsd_cb_notify_lock(lo, nbl);
7136 nfsd4_run_cb(&nbl->nbl_cb);
7140 static const struct lock_manager_operations nfsd_posix_mng_ops = {
7141 .lm_mod_owner = THIS_MODULE,
7142 .lm_notify = nfsd4_lm_notify,
7143 .lm_get_owner = nfsd4_lm_get_owner,
7144 .lm_put_owner = nfsd4_lm_put_owner,
7145 .lm_lock_expirable = nfsd4_lm_lock_expirable,
7146 .lm_expire_lock = nfsd4_lm_expire_lock,
7150 nfs4_set_lock_denied(struct file_lock *fl, struct nfsd4_lock_denied *deny)
7152 struct nfs4_lockowner *lo;
7154 if (fl->fl_lmops == &nfsd_posix_mng_ops) {
7155 lo = (struct nfs4_lockowner *) fl->fl_owner;
7156 xdr_netobj_dup(&deny->ld_owner, &lo->lo_owner.so_owner,
7158 if (!deny->ld_owner.data)
7159 /* We just don't care that much */
7161 deny->ld_clientid = lo->lo_owner.so_client->cl_clientid;
7164 deny->ld_owner.len = 0;
7165 deny->ld_owner.data = NULL;
7166 deny->ld_clientid.cl_boot = 0;
7167 deny->ld_clientid.cl_id = 0;
7169 deny->ld_start = fl->fl_start;
7170 deny->ld_length = NFS4_MAX_UINT64;
7171 if (fl->fl_end != NFS4_MAX_UINT64)
7172 deny->ld_length = fl->fl_end - fl->fl_start + 1;
7173 deny->ld_type = NFS4_READ_LT;
7174 if (fl->fl_type != F_RDLCK)
7175 deny->ld_type = NFS4_WRITE_LT;
7178 static struct nfs4_lockowner *
7179 find_lockowner_str_locked(struct nfs4_client *clp, struct xdr_netobj *owner)
7181 unsigned int strhashval = ownerstr_hashval(owner);
7182 struct nfs4_stateowner *so;
7184 lockdep_assert_held(&clp->cl_lock);
7186 list_for_each_entry(so, &clp->cl_ownerstr_hashtbl[strhashval],
7188 if (so->so_is_open_owner)
7190 if (same_owner_str(so, owner))
7191 return lockowner(nfs4_get_stateowner(so));
7196 static struct nfs4_lockowner *
7197 find_lockowner_str(struct nfs4_client *clp, struct xdr_netobj *owner)
7199 struct nfs4_lockowner *lo;
7201 spin_lock(&clp->cl_lock);
7202 lo = find_lockowner_str_locked(clp, owner);
7203 spin_unlock(&clp->cl_lock);
7207 static void nfs4_unhash_lockowner(struct nfs4_stateowner *sop)
7209 unhash_lockowner_locked(lockowner(sop));
7212 static void nfs4_free_lockowner(struct nfs4_stateowner *sop)
7214 struct nfs4_lockowner *lo = lockowner(sop);
7216 kmem_cache_free(lockowner_slab, lo);
7219 static const struct nfs4_stateowner_operations lockowner_ops = {
7220 .so_unhash = nfs4_unhash_lockowner,
7221 .so_free = nfs4_free_lockowner,
7225 * Alloc a lock owner structure.
7226 * Called in nfsd4_lock - therefore, OPEN and OPEN_CONFIRM (if needed) has
7229 * strhashval = ownerstr_hashval
7231 static struct nfs4_lockowner *
7232 alloc_init_lock_stateowner(unsigned int strhashval, struct nfs4_client *clp,
7233 struct nfs4_ol_stateid *open_stp,
7234 struct nfsd4_lock *lock)
7236 struct nfs4_lockowner *lo, *ret;
7238 lo = alloc_stateowner(lockowner_slab, &lock->lk_new_owner, clp);
7241 INIT_LIST_HEAD(&lo->lo_blocked);
7242 INIT_LIST_HEAD(&lo->lo_owner.so_stateids);
7243 lo->lo_owner.so_is_open_owner = 0;
7244 lo->lo_owner.so_seqid = lock->lk_new_lock_seqid;
7245 lo->lo_owner.so_ops = &lockowner_ops;
7246 spin_lock(&clp->cl_lock);
7247 ret = find_lockowner_str_locked(clp, &lock->lk_new_owner);
7249 list_add(&lo->lo_owner.so_strhash,
7250 &clp->cl_ownerstr_hashtbl[strhashval]);
7253 nfs4_free_stateowner(&lo->lo_owner);
7255 spin_unlock(&clp->cl_lock);
7259 static struct nfs4_ol_stateid *
7260 find_lock_stateid(const struct nfs4_lockowner *lo,
7261 const struct nfs4_ol_stateid *ost)
7263 struct nfs4_ol_stateid *lst;
7265 lockdep_assert_held(&ost->st_stid.sc_client->cl_lock);
7267 /* If ost is not hashed, ost->st_locks will not be valid */
7268 if (!nfs4_ol_stateid_unhashed(ost))
7269 list_for_each_entry(lst, &ost->st_locks, st_locks) {
7270 if (lst->st_stateowner == &lo->lo_owner) {
7271 refcount_inc(&lst->st_stid.sc_count);
7278 static struct nfs4_ol_stateid *
7279 init_lock_stateid(struct nfs4_ol_stateid *stp, struct nfs4_lockowner *lo,
7280 struct nfs4_file *fp, struct inode *inode,
7281 struct nfs4_ol_stateid *open_stp)
7283 struct nfs4_client *clp = lo->lo_owner.so_client;
7284 struct nfs4_ol_stateid *retstp;
7286 mutex_init(&stp->st_mutex);
7287 mutex_lock_nested(&stp->st_mutex, OPEN_STATEID_MUTEX);
7289 spin_lock(&clp->cl_lock);
7290 if (nfs4_ol_stateid_unhashed(open_stp))
7292 retstp = find_lock_stateid(lo, open_stp);
7295 refcount_inc(&stp->st_stid.sc_count);
7296 stp->st_stid.sc_type = NFS4_LOCK_STID;
7297 stp->st_stateowner = nfs4_get_stateowner(&lo->lo_owner);
7299 stp->st_stid.sc_file = fp;
7300 stp->st_access_bmap = 0;
7301 stp->st_deny_bmap = open_stp->st_deny_bmap;
7302 stp->st_openstp = open_stp;
7303 spin_lock(&fp->fi_lock);
7304 list_add(&stp->st_locks, &open_stp->st_locks);
7305 list_add(&stp->st_perstateowner, &lo->lo_owner.so_stateids);
7306 list_add(&stp->st_perfile, &fp->fi_stateids);
7307 spin_unlock(&fp->fi_lock);
7308 spin_unlock(&clp->cl_lock);
7311 spin_unlock(&clp->cl_lock);
7312 if (nfsd4_lock_ol_stateid(retstp) != nfs_ok) {
7313 nfs4_put_stid(&retstp->st_stid);
7316 /* To keep mutex tracking happy */
7317 mutex_unlock(&stp->st_mutex);
7320 spin_unlock(&clp->cl_lock);
7321 mutex_unlock(&stp->st_mutex);
7325 static struct nfs4_ol_stateid *
7326 find_or_create_lock_stateid(struct nfs4_lockowner *lo, struct nfs4_file *fi,
7327 struct inode *inode, struct nfs4_ol_stateid *ost,
7330 struct nfs4_stid *ns = NULL;
7331 struct nfs4_ol_stateid *lst;
7332 struct nfs4_openowner *oo = openowner(ost->st_stateowner);
7333 struct nfs4_client *clp = oo->oo_owner.so_client;
7336 spin_lock(&clp->cl_lock);
7337 lst = find_lock_stateid(lo, ost);
7338 spin_unlock(&clp->cl_lock);
7340 if (nfsd4_lock_ol_stateid(lst) == nfs_ok)
7342 nfs4_put_stid(&lst->st_stid);
7344 ns = nfs4_alloc_stid(clp, stateid_slab, nfs4_free_lock_stateid);
7348 lst = init_lock_stateid(openlockstateid(ns), lo, fi, inode, ost);
7349 if (lst == openlockstateid(ns))
7358 check_lock_length(u64 offset, u64 length)
7360 return ((length == 0) || ((length != NFS4_MAX_UINT64) &&
7361 (length > ~offset)));
7364 static void get_lock_access(struct nfs4_ol_stateid *lock_stp, u32 access)
7366 struct nfs4_file *fp = lock_stp->st_stid.sc_file;
7368 lockdep_assert_held(&fp->fi_lock);
7370 if (test_access(access, lock_stp))
7372 __nfs4_file_get_access(fp, access);
7373 set_access(access, lock_stp);
7377 lookup_or_create_lock_state(struct nfsd4_compound_state *cstate,
7378 struct nfs4_ol_stateid *ost,
7379 struct nfsd4_lock *lock,
7380 struct nfs4_ol_stateid **plst, bool *new)
7383 struct nfs4_file *fi = ost->st_stid.sc_file;
7384 struct nfs4_openowner *oo = openowner(ost->st_stateowner);
7385 struct nfs4_client *cl = oo->oo_owner.so_client;
7386 struct inode *inode = d_inode(cstate->current_fh.fh_dentry);
7387 struct nfs4_lockowner *lo;
7388 struct nfs4_ol_stateid *lst;
7389 unsigned int strhashval;
7391 lo = find_lockowner_str(cl, &lock->lk_new_owner);
7393 strhashval = ownerstr_hashval(&lock->lk_new_owner);
7394 lo = alloc_init_lock_stateowner(strhashval, cl, ost, lock);
7396 return nfserr_jukebox;
7398 /* with an existing lockowner, seqids must be the same */
7399 status = nfserr_bad_seqid;
7400 if (!cstate->minorversion &&
7401 lock->lk_new_lock_seqid != lo->lo_owner.so_seqid)
7405 lst = find_or_create_lock_stateid(lo, fi, inode, ost, new);
7407 status = nfserr_jukebox;
7414 nfs4_put_stateowner(&lo->lo_owner);
7422 nfsd4_lock(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
7423 union nfsd4_op_u *u)
7425 struct nfsd4_lock *lock = &u->lock;
7426 struct nfs4_openowner *open_sop = NULL;
7427 struct nfs4_lockowner *lock_sop = NULL;
7428 struct nfs4_ol_stateid *lock_stp = NULL;
7429 struct nfs4_ol_stateid *open_stp = NULL;
7430 struct nfs4_file *fp;
7431 struct nfsd_file *nf = NULL;
7432 struct nfsd4_blocked_lock *nbl = NULL;
7433 struct file_lock *file_lock = NULL;
7434 struct file_lock *conflock = NULL;
7439 unsigned char fl_type;
7440 unsigned int fl_flags = FL_POSIX;
7441 struct net *net = SVC_NET(rqstp);
7442 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
7444 dprintk("NFSD: nfsd4_lock: start=%Ld length=%Ld\n",
7445 (long long) lock->lk_offset,
7446 (long long) lock->lk_length);
7448 if (check_lock_length(lock->lk_offset, lock->lk_length))
7449 return nfserr_inval;
7451 if ((status = fh_verify(rqstp, &cstate->current_fh,
7452 S_IFREG, NFSD_MAY_LOCK))) {
7453 dprintk("NFSD: nfsd4_lock: permission denied!\n");
7457 if (lock->lk_is_new) {
7458 if (nfsd4_has_session(cstate))
7459 /* See rfc 5661 18.10.3: given clientid is ignored: */
7460 memcpy(&lock->lk_new_clientid,
7461 &cstate->clp->cl_clientid,
7462 sizeof(clientid_t));
7464 /* validate and update open stateid and open seqid */
7465 status = nfs4_preprocess_confirmed_seqid_op(cstate,
7466 lock->lk_new_open_seqid,
7467 &lock->lk_new_open_stateid,
7471 mutex_unlock(&open_stp->st_mutex);
7472 open_sop = openowner(open_stp->st_stateowner);
7473 status = nfserr_bad_stateid;
7474 if (!same_clid(&open_sop->oo_owner.so_client->cl_clientid,
7475 &lock->lk_new_clientid))
7477 status = lookup_or_create_lock_state(cstate, open_stp, lock,
7480 status = nfs4_preprocess_seqid_op(cstate,
7481 lock->lk_old_lock_seqid,
7482 &lock->lk_old_lock_stateid,
7483 NFS4_LOCK_STID, &lock_stp, nn);
7487 lock_sop = lockowner(lock_stp->st_stateowner);
7489 lkflg = setlkflg(lock->lk_type);
7490 status = nfs4_check_openmode(lock_stp, lkflg);
7494 status = nfserr_grace;
7495 if (locks_in_grace(net) && !lock->lk_reclaim)
7497 status = nfserr_no_grace;
7498 if (!locks_in_grace(net) && lock->lk_reclaim)
7501 if (lock->lk_reclaim)
7502 fl_flags |= FL_RECLAIM;
7504 fp = lock_stp->st_stid.sc_file;
7505 switch (lock->lk_type) {
7507 if (nfsd4_has_session(cstate))
7508 fl_flags |= FL_SLEEP;
7511 spin_lock(&fp->fi_lock);
7512 nf = find_readable_file_locked(fp);
7514 get_lock_access(lock_stp, NFS4_SHARE_ACCESS_READ);
7515 spin_unlock(&fp->fi_lock);
7518 case NFS4_WRITEW_LT:
7519 if (nfsd4_has_session(cstate))
7520 fl_flags |= FL_SLEEP;
7523 spin_lock(&fp->fi_lock);
7524 nf = find_writeable_file_locked(fp);
7526 get_lock_access(lock_stp, NFS4_SHARE_ACCESS_WRITE);
7527 spin_unlock(&fp->fi_lock);
7531 status = nfserr_inval;
7536 status = nfserr_openmode;
7541 * Most filesystems with their own ->lock operations will block
7542 * the nfsd thread waiting to acquire the lock. That leads to
7543 * deadlocks (we don't want every nfsd thread tied up waiting
7544 * for file locks), so don't attempt blocking lock notifications
7545 * on those filesystems:
7547 if (nf->nf_file->f_op->lock)
7548 fl_flags &= ~FL_SLEEP;
7550 nbl = find_or_allocate_block(lock_sop, &fp->fi_fhandle, nn);
7552 dprintk("NFSD: %s: unable to allocate block!\n", __func__);
7553 status = nfserr_jukebox;
7557 file_lock = &nbl->nbl_lock;
7558 file_lock->fl_type = fl_type;
7559 file_lock->fl_owner = (fl_owner_t)lockowner(nfs4_get_stateowner(&lock_sop->lo_owner));
7560 file_lock->fl_pid = current->tgid;
7561 file_lock->fl_file = nf->nf_file;
7562 file_lock->fl_flags = fl_flags;
7563 file_lock->fl_lmops = &nfsd_posix_mng_ops;
7564 file_lock->fl_start = lock->lk_offset;
7565 file_lock->fl_end = last_byte_offset(lock->lk_offset, lock->lk_length);
7566 nfs4_transform_lock_offset(file_lock);
7568 conflock = locks_alloc_lock();
7570 dprintk("NFSD: %s: unable to allocate lock!\n", __func__);
7571 status = nfserr_jukebox;
7575 if (fl_flags & FL_SLEEP) {
7576 nbl->nbl_time = ktime_get_boottime_seconds();
7577 spin_lock(&nn->blocked_locks_lock);
7578 list_add_tail(&nbl->nbl_list, &lock_sop->lo_blocked);
7579 list_add_tail(&nbl->nbl_lru, &nn->blocked_locks_lru);
7580 kref_get(&nbl->nbl_kref);
7581 spin_unlock(&nn->blocked_locks_lock);
7584 err = vfs_lock_file(nf->nf_file, F_SETLK, file_lock, conflock);
7586 case 0: /* success! */
7587 nfs4_inc_and_copy_stateid(&lock->lk_resp_stateid, &lock_stp->st_stid);
7589 if (lock->lk_reclaim)
7590 nn->somebody_reclaimed = true;
7592 case FILE_LOCK_DEFERRED:
7593 kref_put(&nbl->nbl_kref, free_nbl);
7596 case -EAGAIN: /* conflock holds conflicting lock */
7597 status = nfserr_denied;
7598 dprintk("NFSD: nfsd4_lock: conflicting lock found!\n");
7599 nfs4_set_lock_denied(conflock, &lock->lk_denied);
7602 status = nfserr_deadlock;
7605 dprintk("NFSD: nfsd4_lock: vfs_lock_file() failed! status %d\n",err);
7606 status = nfserrno(err);
7611 /* dequeue it if we queued it before */
7612 if (fl_flags & FL_SLEEP) {
7613 spin_lock(&nn->blocked_locks_lock);
7614 if (!list_empty(&nbl->nbl_list) &&
7615 !list_empty(&nbl->nbl_lru)) {
7616 list_del_init(&nbl->nbl_list);
7617 list_del_init(&nbl->nbl_lru);
7618 kref_put(&nbl->nbl_kref, free_nbl);
7620 /* nbl can use one of lists to be linked to reaplist */
7621 spin_unlock(&nn->blocked_locks_lock);
7623 free_blocked_lock(nbl);
7628 /* Bump seqid manually if the 4.0 replay owner is openowner */
7629 if (cstate->replay_owner &&
7630 cstate->replay_owner != &lock_sop->lo_owner &&
7631 seqid_mutating_err(ntohl(status)))
7632 lock_sop->lo_owner.so_seqid++;
7635 * If this is a new, never-before-used stateid, and we are
7636 * returning an error, then just go ahead and release it.
7639 release_lock_stateid(lock_stp);
7641 mutex_unlock(&lock_stp->st_mutex);
7643 nfs4_put_stid(&lock_stp->st_stid);
7646 nfs4_put_stid(&open_stp->st_stid);
7647 nfsd4_bump_seqid(cstate, status);
7649 locks_free_lock(conflock);
7654 * The NFSv4 spec allows a client to do a LOCKT without holding an OPEN,
7655 * so we do a temporary open here just to get an open file to pass to
7658 static __be32 nfsd_test_lock(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file_lock *lock)
7660 struct nfsd_file *nf;
7661 struct inode *inode;
7664 err = nfsd_file_acquire(rqstp, fhp, NFSD_MAY_READ, &nf);
7667 inode = fhp->fh_dentry->d_inode;
7668 inode_lock(inode); /* to block new leases till after test_lock: */
7669 err = nfserrno(nfsd_open_break_lease(inode, NFSD_MAY_READ));
7672 lock->fl_file = nf->nf_file;
7673 err = nfserrno(vfs_test_lock(nf->nf_file, lock));
7674 lock->fl_file = NULL;
7676 inode_unlock(inode);
7685 nfsd4_lockt(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
7686 union nfsd4_op_u *u)
7688 struct nfsd4_lockt *lockt = &u->lockt;
7689 struct file_lock *file_lock = NULL;
7690 struct nfs4_lockowner *lo = NULL;
7692 struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
7694 if (locks_in_grace(SVC_NET(rqstp)))
7695 return nfserr_grace;
7697 if (check_lock_length(lockt->lt_offset, lockt->lt_length))
7698 return nfserr_inval;
7700 if (!nfsd4_has_session(cstate)) {
7701 status = set_client(&lockt->lt_clientid, cstate, nn);
7706 if ((status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0)))
7709 file_lock = locks_alloc_lock();
7711 dprintk("NFSD: %s: unable to allocate lock!\n", __func__);
7712 status = nfserr_jukebox;
7716 switch (lockt->lt_type) {
7719 file_lock->fl_type = F_RDLCK;
7722 case NFS4_WRITEW_LT:
7723 file_lock->fl_type = F_WRLCK;
7726 dprintk("NFSD: nfs4_lockt: bad lock type!\n");
7727 status = nfserr_inval;
7731 lo = find_lockowner_str(cstate->clp, &lockt->lt_owner);
7733 file_lock->fl_owner = (fl_owner_t)lo;
7734 file_lock->fl_pid = current->tgid;
7735 file_lock->fl_flags = FL_POSIX;
7737 file_lock->fl_start = lockt->lt_offset;
7738 file_lock->fl_end = last_byte_offset(lockt->lt_offset, lockt->lt_length);
7740 nfs4_transform_lock_offset(file_lock);
7742 status = nfsd_test_lock(rqstp, &cstate->current_fh, file_lock);
7746 if (file_lock->fl_type != F_UNLCK) {
7747 status = nfserr_denied;
7748 nfs4_set_lock_denied(file_lock, &lockt->lt_denied);
7752 nfs4_put_stateowner(&lo->lo_owner);
7754 locks_free_lock(file_lock);
7759 nfsd4_locku(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
7760 union nfsd4_op_u *u)
7762 struct nfsd4_locku *locku = &u->locku;
7763 struct nfs4_ol_stateid *stp;
7764 struct nfsd_file *nf = NULL;
7765 struct file_lock *file_lock = NULL;
7768 struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
7770 dprintk("NFSD: nfsd4_locku: start=%Ld length=%Ld\n",
7771 (long long) locku->lu_offset,
7772 (long long) locku->lu_length);
7774 if (check_lock_length(locku->lu_offset, locku->lu_length))
7775 return nfserr_inval;
7777 status = nfs4_preprocess_seqid_op(cstate, locku->lu_seqid,
7778 &locku->lu_stateid, NFS4_LOCK_STID,
7782 nf = find_any_file(stp->st_stid.sc_file);
7784 status = nfserr_lock_range;
7787 file_lock = locks_alloc_lock();
7789 dprintk("NFSD: %s: unable to allocate lock!\n", __func__);
7790 status = nfserr_jukebox;
7794 file_lock->fl_type = F_UNLCK;
7795 file_lock->fl_owner = (fl_owner_t)lockowner(nfs4_get_stateowner(stp->st_stateowner));
7796 file_lock->fl_pid = current->tgid;
7797 file_lock->fl_file = nf->nf_file;
7798 file_lock->fl_flags = FL_POSIX;
7799 file_lock->fl_lmops = &nfsd_posix_mng_ops;
7800 file_lock->fl_start = locku->lu_offset;
7802 file_lock->fl_end = last_byte_offset(locku->lu_offset,
7804 nfs4_transform_lock_offset(file_lock);
7806 err = vfs_lock_file(nf->nf_file, F_SETLK, file_lock, NULL);
7808 dprintk("NFSD: nfs4_locku: vfs_lock_file failed!\n");
7811 nfs4_inc_and_copy_stateid(&locku->lu_stateid, &stp->st_stid);
7815 mutex_unlock(&stp->st_mutex);
7816 nfs4_put_stid(&stp->st_stid);
7818 nfsd4_bump_seqid(cstate, status);
7820 locks_free_lock(file_lock);
7824 status = nfserrno(err);
7830 * true: locks held by lockowner
7831 * false: no locks held by lockowner
7834 check_for_locks(struct nfs4_file *fp, struct nfs4_lockowner *lowner)
7836 struct file_lock *fl;
7838 struct nfsd_file *nf = find_any_file(fp);
7839 struct inode *inode;
7840 struct file_lock_context *flctx;
7843 /* Any valid lock stateid should have some sort of access */
7848 inode = file_inode(nf->nf_file);
7849 flctx = locks_inode_context(inode);
7851 if (flctx && !list_empty_careful(&flctx->flc_posix)) {
7852 spin_lock(&flctx->flc_lock);
7853 list_for_each_entry(fl, &flctx->flc_posix, fl_list) {
7854 if (fl->fl_owner == (fl_owner_t)lowner) {
7859 spin_unlock(&flctx->flc_lock);
7866 * nfsd4_release_lockowner - process NFSv4.0 RELEASE_LOCKOWNER operations
7867 * @rqstp: RPC transaction
7868 * @cstate: NFSv4 COMPOUND state
7869 * @u: RELEASE_LOCKOWNER arguments
7871 * The lockowner's so_count is bumped when a lock record is added
7872 * or when copying a conflicting lock. The latter case is brief,
7873 * but can lead to fleeting false positives when looking for
7877 * %nfs_ok: lockowner released or not found
7878 * %nfserr_locks_held: lockowner still in use
7879 * %nfserr_stale_clientid: clientid no longer active
7880 * %nfserr_expired: clientid not recognized
7883 nfsd4_release_lockowner(struct svc_rqst *rqstp,
7884 struct nfsd4_compound_state *cstate,
7885 union nfsd4_op_u *u)
7887 struct nfsd4_release_lockowner *rlockowner = &u->release_lockowner;
7888 struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
7889 clientid_t *clid = &rlockowner->rl_clientid;
7890 struct nfs4_ol_stateid *stp;
7891 struct nfs4_lockowner *lo;
7892 struct nfs4_client *clp;
7893 LIST_HEAD(reaplist);
7896 dprintk("nfsd4_release_lockowner clientid: (%08x/%08x):\n",
7897 clid->cl_boot, clid->cl_id);
7899 status = set_client(clid, cstate, nn);
7904 spin_lock(&clp->cl_lock);
7905 lo = find_lockowner_str_locked(clp, &rlockowner->rl_owner);
7907 spin_unlock(&clp->cl_lock);
7910 if (atomic_read(&lo->lo_owner.so_count) != 2) {
7911 spin_unlock(&clp->cl_lock);
7912 nfs4_put_stateowner(&lo->lo_owner);
7913 return nfserr_locks_held;
7915 unhash_lockowner_locked(lo);
7916 while (!list_empty(&lo->lo_owner.so_stateids)) {
7917 stp = list_first_entry(&lo->lo_owner.so_stateids,
7918 struct nfs4_ol_stateid,
7920 WARN_ON(!unhash_lock_stateid(stp));
7921 put_ol_stateid_locked(stp, &reaplist);
7923 spin_unlock(&clp->cl_lock);
7925 free_ol_stateid_reaplist(&reaplist);
7926 remove_blocked_locks(lo);
7927 nfs4_put_stateowner(&lo->lo_owner);
7931 static inline struct nfs4_client_reclaim *
7934 return kmalloc(sizeof(struct nfs4_client_reclaim), GFP_KERNEL);
7938 nfs4_has_reclaimed_state(struct xdr_netobj name, struct nfsd_net *nn)
7940 struct nfs4_client_reclaim *crp;
7942 crp = nfsd4_find_reclaim_client(name, nn);
7943 return (crp && crp->cr_clp);
7947 * failure => all reset bets are off, nfserr_no_grace...
7949 * The caller is responsible for freeing name.data if NULL is returned (it
7950 * will be freed in nfs4_remove_reclaim_record in the normal case).
7952 struct nfs4_client_reclaim *
7953 nfs4_client_to_reclaim(struct xdr_netobj name, struct xdr_netobj princhash,
7954 struct nfsd_net *nn)
7956 unsigned int strhashval;
7957 struct nfs4_client_reclaim *crp;
7959 crp = alloc_reclaim();
7961 strhashval = clientstr_hashval(name);
7962 INIT_LIST_HEAD(&crp->cr_strhash);
7963 list_add(&crp->cr_strhash, &nn->reclaim_str_hashtbl[strhashval]);
7964 crp->cr_name.data = name.data;
7965 crp->cr_name.len = name.len;
7966 crp->cr_princhash.data = princhash.data;
7967 crp->cr_princhash.len = princhash.len;
7969 nn->reclaim_str_hashtbl_size++;
7975 nfs4_remove_reclaim_record(struct nfs4_client_reclaim *crp, struct nfsd_net *nn)
7977 list_del(&crp->cr_strhash);
7978 kfree(crp->cr_name.data);
7979 kfree(crp->cr_princhash.data);
7981 nn->reclaim_str_hashtbl_size--;
7985 nfs4_release_reclaim(struct nfsd_net *nn)
7987 struct nfs4_client_reclaim *crp = NULL;
7990 for (i = 0; i < CLIENT_HASH_SIZE; i++) {
7991 while (!list_empty(&nn->reclaim_str_hashtbl[i])) {
7992 crp = list_entry(nn->reclaim_str_hashtbl[i].next,
7993 struct nfs4_client_reclaim, cr_strhash);
7994 nfs4_remove_reclaim_record(crp, nn);
7997 WARN_ON_ONCE(nn->reclaim_str_hashtbl_size);
8001 * called from OPEN, CLAIM_PREVIOUS with a new clientid. */
8002 struct nfs4_client_reclaim *
8003 nfsd4_find_reclaim_client(struct xdr_netobj name, struct nfsd_net *nn)
8005 unsigned int strhashval;
8006 struct nfs4_client_reclaim *crp = NULL;
8008 strhashval = clientstr_hashval(name);
8009 list_for_each_entry(crp, &nn->reclaim_str_hashtbl[strhashval], cr_strhash) {
8010 if (compare_blob(&crp->cr_name, &name) == 0) {
8018 nfs4_check_open_reclaim(struct nfs4_client *clp)
8020 if (test_bit(NFSD4_CLIENT_RECLAIM_COMPLETE, &clp->cl_flags))
8021 return nfserr_no_grace;
8023 if (nfsd4_client_record_check(clp))
8024 return nfserr_reclaim_bad;
8030 * Since the lifetime of a delegation isn't limited to that of an open, a
8031 * client may quite reasonably hang on to a delegation as long as it has
8032 * the inode cached. This becomes an obvious problem the first time a
8033 * client's inode cache approaches the size of the server's total memory.
8035 * For now we avoid this problem by imposing a hard limit on the number
8036 * of delegations, which varies according to the server's memory size.
8039 set_max_delegations(void)
8042 * Allow at most 4 delegations per megabyte of RAM. Quick
8043 * estimates suggest that in the worst case (where every delegation
8044 * is for a different inode), a delegation could take about 1.5K,
8045 * giving a worst case usage of about 6% of memory.
8047 max_delegations = nr_free_buffer_pages() >> (20 - 2 - PAGE_SHIFT);
8050 static int nfs4_state_create_net(struct net *net)
8052 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
8055 nn->conf_id_hashtbl = kmalloc_array(CLIENT_HASH_SIZE,
8056 sizeof(struct list_head),
8058 if (!nn->conf_id_hashtbl)
8060 nn->unconf_id_hashtbl = kmalloc_array(CLIENT_HASH_SIZE,
8061 sizeof(struct list_head),
8063 if (!nn->unconf_id_hashtbl)
8065 nn->sessionid_hashtbl = kmalloc_array(SESSION_HASH_SIZE,
8066 sizeof(struct list_head),
8068 if (!nn->sessionid_hashtbl)
8071 for (i = 0; i < CLIENT_HASH_SIZE; i++) {
8072 INIT_LIST_HEAD(&nn->conf_id_hashtbl[i]);
8073 INIT_LIST_HEAD(&nn->unconf_id_hashtbl[i]);
8075 for (i = 0; i < SESSION_HASH_SIZE; i++)
8076 INIT_LIST_HEAD(&nn->sessionid_hashtbl[i]);
8077 nn->conf_name_tree = RB_ROOT;
8078 nn->unconf_name_tree = RB_ROOT;
8079 nn->boot_time = ktime_get_real_seconds();
8080 nn->grace_ended = false;
8081 nn->nfsd4_manager.block_opens = true;
8082 INIT_LIST_HEAD(&nn->nfsd4_manager.list);
8083 INIT_LIST_HEAD(&nn->client_lru);
8084 INIT_LIST_HEAD(&nn->close_lru);
8085 INIT_LIST_HEAD(&nn->del_recall_lru);
8086 spin_lock_init(&nn->client_lock);
8087 spin_lock_init(&nn->s2s_cp_lock);
8088 idr_init(&nn->s2s_cp_stateids);
8090 spin_lock_init(&nn->blocked_locks_lock);
8091 INIT_LIST_HEAD(&nn->blocked_locks_lru);
8093 INIT_DELAYED_WORK(&nn->laundromat_work, laundromat_main);
8094 INIT_WORK(&nn->nfsd_shrinker_work, nfsd4_state_shrinker_worker);
8097 nn->nfsd_client_shrinker.scan_objects = nfsd4_state_shrinker_scan;
8098 nn->nfsd_client_shrinker.count_objects = nfsd4_state_shrinker_count;
8099 nn->nfsd_client_shrinker.seeks = DEFAULT_SEEKS;
8101 if (register_shrinker(&nn->nfsd_client_shrinker, "nfsd-client"))
8107 kfree(nn->sessionid_hashtbl);
8109 kfree(nn->unconf_id_hashtbl);
8111 kfree(nn->conf_id_hashtbl);
8117 nfs4_state_destroy_net(struct net *net)
8120 struct nfs4_client *clp = NULL;
8121 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
8123 for (i = 0; i < CLIENT_HASH_SIZE; i++) {
8124 while (!list_empty(&nn->conf_id_hashtbl[i])) {
8125 clp = list_entry(nn->conf_id_hashtbl[i].next, struct nfs4_client, cl_idhash);
8126 destroy_client(clp);
8130 WARN_ON(!list_empty(&nn->blocked_locks_lru));
8132 for (i = 0; i < CLIENT_HASH_SIZE; i++) {
8133 while (!list_empty(&nn->unconf_id_hashtbl[i])) {
8134 clp = list_entry(nn->unconf_id_hashtbl[i].next, struct nfs4_client, cl_idhash);
8135 destroy_client(clp);
8139 kfree(nn->sessionid_hashtbl);
8140 kfree(nn->unconf_id_hashtbl);
8141 kfree(nn->conf_id_hashtbl);
8146 nfs4_state_start_net(struct net *net)
8148 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
8151 ret = nfs4_state_create_net(net);
8154 locks_start_grace(net, &nn->nfsd4_manager);
8155 nfsd4_client_tracking_init(net);
8156 if (nn->track_reclaim_completes && nn->reclaim_str_hashtbl_size == 0)
8158 printk(KERN_INFO "NFSD: starting %lld-second grace period (net %x)\n",
8159 nn->nfsd4_grace, net->ns.inum);
8160 trace_nfsd_grace_start(nn);
8161 queue_delayed_work(laundry_wq, &nn->laundromat_work, nn->nfsd4_grace * HZ);
8165 printk(KERN_INFO "NFSD: no clients to reclaim, skipping NFSv4 grace period (net %x)\n",
8167 queue_delayed_work(laundry_wq, &nn->laundromat_work, nn->nfsd4_lease * HZ);
8168 nfsd4_end_grace(nn);
8172 /* initialization to perform when the nfsd service is started: */
8175 nfs4_state_start(void)
8179 ret = rhltable_init(&nfs4_file_rhltable, &nfs4_file_rhash_params);
8183 ret = nfsd4_create_callback_queue();
8185 rhltable_destroy(&nfs4_file_rhltable);
8189 set_max_delegations();
8194 nfs4_state_shutdown_net(struct net *net)
8196 struct nfs4_delegation *dp = NULL;
8197 struct list_head *pos, *next, reaplist;
8198 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
8200 unregister_shrinker(&nn->nfsd_client_shrinker);
8201 cancel_work(&nn->nfsd_shrinker_work);
8202 cancel_delayed_work_sync(&nn->laundromat_work);
8203 locks_end_grace(&nn->nfsd4_manager);
8205 INIT_LIST_HEAD(&reaplist);
8206 spin_lock(&state_lock);
8207 list_for_each_safe(pos, next, &nn->del_recall_lru) {
8208 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
8209 WARN_ON(!unhash_delegation_locked(dp));
8210 list_add(&dp->dl_recall_lru, &reaplist);
8212 spin_unlock(&state_lock);
8213 list_for_each_safe(pos, next, &reaplist) {
8214 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
8215 list_del_init(&dp->dl_recall_lru);
8216 destroy_unhashed_deleg(dp);
8219 nfsd4_client_tracking_exit(net);
8220 nfs4_state_destroy_net(net);
8221 #ifdef CONFIG_NFSD_V4_2_INTER_SSC
8222 nfsd4_ssc_shutdown_umount(nn);
8227 nfs4_state_shutdown(void)
8229 nfsd4_destroy_callback_queue();
8230 rhltable_destroy(&nfs4_file_rhltable);
8234 get_stateid(struct nfsd4_compound_state *cstate, stateid_t *stateid)
8236 if (HAS_CSTATE_FLAG(cstate, CURRENT_STATE_ID_FLAG) &&
8237 CURRENT_STATEID(stateid))
8238 memcpy(stateid, &cstate->current_stateid, sizeof(stateid_t));
8242 put_stateid(struct nfsd4_compound_state *cstate, stateid_t *stateid)
8244 if (cstate->minorversion) {
8245 memcpy(&cstate->current_stateid, stateid, sizeof(stateid_t));
8246 SET_CSTATE_FLAG(cstate, CURRENT_STATE_ID_FLAG);
8251 clear_current_stateid(struct nfsd4_compound_state *cstate)
8253 CLEAR_CSTATE_FLAG(cstate, CURRENT_STATE_ID_FLAG);
8257 * functions to set current state id
8260 nfsd4_set_opendowngradestateid(struct nfsd4_compound_state *cstate,
8261 union nfsd4_op_u *u)
8263 put_stateid(cstate, &u->open_downgrade.od_stateid);
8267 nfsd4_set_openstateid(struct nfsd4_compound_state *cstate,
8268 union nfsd4_op_u *u)
8270 put_stateid(cstate, &u->open.op_stateid);
8274 nfsd4_set_closestateid(struct nfsd4_compound_state *cstate,
8275 union nfsd4_op_u *u)
8277 put_stateid(cstate, &u->close.cl_stateid);
8281 nfsd4_set_lockstateid(struct nfsd4_compound_state *cstate,
8282 union nfsd4_op_u *u)
8284 put_stateid(cstate, &u->lock.lk_resp_stateid);
8288 * functions to consume current state id
8292 nfsd4_get_opendowngradestateid(struct nfsd4_compound_state *cstate,
8293 union nfsd4_op_u *u)
8295 get_stateid(cstate, &u->open_downgrade.od_stateid);
8299 nfsd4_get_delegreturnstateid(struct nfsd4_compound_state *cstate,
8300 union nfsd4_op_u *u)
8302 get_stateid(cstate, &u->delegreturn.dr_stateid);
8306 nfsd4_get_freestateid(struct nfsd4_compound_state *cstate,
8307 union nfsd4_op_u *u)
8309 get_stateid(cstate, &u->free_stateid.fr_stateid);
8313 nfsd4_get_setattrstateid(struct nfsd4_compound_state *cstate,
8314 union nfsd4_op_u *u)
8316 get_stateid(cstate, &u->setattr.sa_stateid);
8320 nfsd4_get_closestateid(struct nfsd4_compound_state *cstate,
8321 union nfsd4_op_u *u)
8323 get_stateid(cstate, &u->close.cl_stateid);
8327 nfsd4_get_lockustateid(struct nfsd4_compound_state *cstate,
8328 union nfsd4_op_u *u)
8330 get_stateid(cstate, &u->locku.lu_stateid);
8334 nfsd4_get_readstateid(struct nfsd4_compound_state *cstate,
8335 union nfsd4_op_u *u)
8337 get_stateid(cstate, &u->read.rd_stateid);
8341 nfsd4_get_writestateid(struct nfsd4_compound_state *cstate,
8342 union nfsd4_op_u *u)
8344 get_stateid(cstate, &u->write.wr_stateid);