2 * linux/fs/nfsd/nfs4state.c
4 * Copyright (c) 2001 The Regents of the University of Michigan.
7 * Kendrick Smith <kmsmith@umich.edu>
8 * Andy Adamson <kandros@umich.edu>
10 * Redistribution and use in source and binary forms, with or without
11 * modification, are permitted provided that the following conditions
14 * 1. Redistributions of source code must retain the above copyright
15 * notice, this list of conditions and the following disclaimer.
16 * 2. Redistributions in binary form must reproduce the above copyright
17 * notice, this list of conditions and the following disclaimer in the
18 * documentation and/or other materials provided with the distribution.
19 * 3. Neither the name of the University nor the names of its
20 * contributors may be used to endorse or promote products derived
21 * from this software without specific prior written permission.
23 * THIS SOFTWARE IS PROVIDED ``AS IS'' AND ANY EXPRESS OR IMPLIED
24 * WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF
25 * MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
26 * DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
27 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
28 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
29 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR
30 * BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF
31 * LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING
32 * NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
33 * SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
37 #include <linux/param.h>
38 #include <linux/major.h>
39 #include <linux/slab.h>
41 #include <linux/sunrpc/svc.h>
42 #include <linux/nfsd/nfsd.h>
43 #include <linux/nfsd/cache.h>
44 #include <linux/file.h>
45 #include <linux/mount.h>
46 #include <linux/workqueue.h>
47 #include <linux/smp_lock.h>
48 #include <linux/kthread.h>
49 #include <linux/nfs4.h>
50 #include <linux/nfsd/state.h>
51 #include <linux/nfsd/xdr4.h>
52 #include <linux/namei.h>
53 #include <linux/swap.h>
54 #include <linux/mutex.h>
55 #include <linux/lockd/bind.h>
56 #include <linux/module.h>
57 #include <linux/sunrpc/svcauth_gss.h>
59 #define NFSDDBG_FACILITY NFSDDBG_PROC
62 static time_t lease_time = 90; /* default lease time */
63 static time_t user_lease_time = 90;
64 static time_t boot_time;
65 static u32 current_ownerid = 1;
66 static u32 current_fileid = 1;
67 static u32 current_delegid = 1;
69 static stateid_t zerostateid; /* bits all 0 */
70 static stateid_t onestateid; /* bits all 1 */
72 #define ZERO_STATEID(stateid) (!memcmp((stateid), &zerostateid, sizeof(stateid_t)))
73 #define ONE_STATEID(stateid) (!memcmp((stateid), &onestateid, sizeof(stateid_t)))
75 /* forward declarations */
76 static struct nfs4_stateid * find_stateid(stateid_t *stid, int flags);
77 static struct nfs4_delegation * find_delegation_stateid(struct inode *ino, stateid_t *stid);
78 static void release_stateid_lockowners(struct nfs4_stateid *open_stp);
79 static char user_recovery_dirname[PATH_MAX] = "/var/lib/nfs/v4recovery";
80 static void nfs4_set_recdir(char *recdir);
85 * protects clientid_hashtbl[], clientstr_hashtbl[],
86 * unconfstr_hashtbl[], uncofid_hashtbl[].
88 static DEFINE_MUTEX(client_mutex);
90 static struct kmem_cache *stateowner_slab = NULL;
91 static struct kmem_cache *file_slab = NULL;
92 static struct kmem_cache *stateid_slab = NULL;
93 static struct kmem_cache *deleg_slab = NULL;
98 mutex_lock(&client_mutex);
102 nfs4_unlock_state(void)
104 mutex_unlock(&client_mutex);
108 opaque_hashval(const void *ptr, int nbytes)
110 unsigned char *cptr = (unsigned char *) ptr;
120 /* forward declarations */
121 static void release_stateowner(struct nfs4_stateowner *sop);
122 static void release_stateid(struct nfs4_stateid *stp, int flags);
128 /* recall_lock protects the del_recall_lru */
129 static DEFINE_SPINLOCK(recall_lock);
130 static struct list_head del_recall_lru;
133 free_nfs4_file(struct kref *kref)
135 struct nfs4_file *fp = container_of(kref, struct nfs4_file, fi_ref);
136 list_del(&fp->fi_hash);
138 kmem_cache_free(file_slab, fp);
142 put_nfs4_file(struct nfs4_file *fi)
144 kref_put(&fi->fi_ref, free_nfs4_file);
148 get_nfs4_file(struct nfs4_file *fi)
150 kref_get(&fi->fi_ref);
153 static int num_delegations;
154 unsigned int max_delegations;
157 * Open owner state (share locks)
160 /* hash tables for nfs4_stateowner */
161 #define OWNER_HASH_BITS 8
162 #define OWNER_HASH_SIZE (1 << OWNER_HASH_BITS)
163 #define OWNER_HASH_MASK (OWNER_HASH_SIZE - 1)
165 #define ownerid_hashval(id) \
166 ((id) & OWNER_HASH_MASK)
167 #define ownerstr_hashval(clientid, ownername) \
168 (((clientid) + opaque_hashval((ownername.data), (ownername.len))) & OWNER_HASH_MASK)
170 static struct list_head ownerid_hashtbl[OWNER_HASH_SIZE];
171 static struct list_head ownerstr_hashtbl[OWNER_HASH_SIZE];
173 /* hash table for nfs4_file */
174 #define FILE_HASH_BITS 8
175 #define FILE_HASH_SIZE (1 << FILE_HASH_BITS)
176 #define FILE_HASH_MASK (FILE_HASH_SIZE - 1)
177 /* hash table for (open)nfs4_stateid */
178 #define STATEID_HASH_BITS 10
179 #define STATEID_HASH_SIZE (1 << STATEID_HASH_BITS)
180 #define STATEID_HASH_MASK (STATEID_HASH_SIZE - 1)
182 #define file_hashval(x) \
183 hash_ptr(x, FILE_HASH_BITS)
184 #define stateid_hashval(owner_id, file_id) \
185 (((owner_id) + (file_id)) & STATEID_HASH_MASK)
187 static struct list_head file_hashtbl[FILE_HASH_SIZE];
188 static struct list_head stateid_hashtbl[STATEID_HASH_SIZE];
190 static struct nfs4_delegation *
191 alloc_init_deleg(struct nfs4_client *clp, struct nfs4_stateid *stp, struct svc_fh *current_fh, u32 type)
193 struct nfs4_delegation *dp;
194 struct nfs4_file *fp = stp->st_file;
195 struct nfs4_callback *cb = &stp->st_stateowner->so_client->cl_callback;
197 dprintk("NFSD alloc_init_deleg\n");
198 if (fp->fi_had_conflict)
200 if (num_delegations > max_delegations)
202 dp = kmem_cache_alloc(deleg_slab, GFP_KERNEL);
206 INIT_LIST_HEAD(&dp->dl_perfile);
207 INIT_LIST_HEAD(&dp->dl_perclnt);
208 INIT_LIST_HEAD(&dp->dl_recall_lru);
213 get_file(stp->st_vfs_file);
214 dp->dl_vfs_file = stp->st_vfs_file;
216 dp->dl_recall.cbr_dp = NULL;
217 dp->dl_recall.cbr_ident = cb->cb_ident;
218 dp->dl_recall.cbr_trunc = 0;
219 dp->dl_stateid.si_boot = boot_time;
220 dp->dl_stateid.si_stateownerid = current_delegid++;
221 dp->dl_stateid.si_fileid = 0;
222 dp->dl_stateid.si_generation = 0;
223 dp->dl_fhlen = current_fh->fh_handle.fh_size;
224 memcpy(dp->dl_fhval, ¤t_fh->fh_handle.fh_base,
225 current_fh->fh_handle.fh_size);
227 atomic_set(&dp->dl_count, 1);
228 list_add(&dp->dl_perfile, &fp->fi_delegations);
229 list_add(&dp->dl_perclnt, &clp->cl_delegations);
234 nfs4_put_delegation(struct nfs4_delegation *dp)
236 if (atomic_dec_and_test(&dp->dl_count)) {
237 dprintk("NFSD: freeing dp %p\n",dp);
238 put_nfs4_file(dp->dl_file);
239 kmem_cache_free(deleg_slab, dp);
244 /* Remove the associated file_lock first, then remove the delegation.
245 * lease_modify() is called to remove the FS_LEASE file_lock from
246 * the i_flock list, eventually calling nfsd's lock_manager
247 * fl_release_callback.
250 nfs4_close_delegation(struct nfs4_delegation *dp)
252 struct file *filp = dp->dl_vfs_file;
254 dprintk("NFSD: close_delegation dp %p\n",dp);
255 dp->dl_vfs_file = NULL;
256 /* The following nfsd_close may not actually close the file,
257 * but we want to remove the lease in any case. */
259 vfs_setlease(filp, F_UNLCK, &dp->dl_flock);
263 /* Called under the state lock. */
265 unhash_delegation(struct nfs4_delegation *dp)
267 list_del_init(&dp->dl_perfile);
268 list_del_init(&dp->dl_perclnt);
269 spin_lock(&recall_lock);
270 list_del_init(&dp->dl_recall_lru);
271 spin_unlock(&recall_lock);
272 nfs4_close_delegation(dp);
273 nfs4_put_delegation(dp);
280 /* Hash tables for nfs4_clientid state */
281 #define CLIENT_HASH_BITS 4
282 #define CLIENT_HASH_SIZE (1 << CLIENT_HASH_BITS)
283 #define CLIENT_HASH_MASK (CLIENT_HASH_SIZE - 1)
285 #define clientid_hashval(id) \
286 ((id) & CLIENT_HASH_MASK)
287 #define clientstr_hashval(name) \
288 (opaque_hashval((name), 8) & CLIENT_HASH_MASK)
290 * reclaim_str_hashtbl[] holds known client info from previous reset/reboot
291 * used in reboot/reset lease grace period processing
293 * conf_id_hashtbl[], and conf_str_hashtbl[] hold confirmed
294 * setclientid_confirmed info.
296 * unconf_str_hastbl[] and unconf_id_hashtbl[] hold unconfirmed
299 * client_lru holds client queue ordered by nfs4_client.cl_time
302 * close_lru holds (open) stateowner queue ordered by nfs4_stateowner.so_time
303 * for last close replay.
305 static struct list_head reclaim_str_hashtbl[CLIENT_HASH_SIZE];
306 static int reclaim_str_hashtbl_size = 0;
307 static struct list_head conf_id_hashtbl[CLIENT_HASH_SIZE];
308 static struct list_head conf_str_hashtbl[CLIENT_HASH_SIZE];
309 static struct list_head unconf_str_hashtbl[CLIENT_HASH_SIZE];
310 static struct list_head unconf_id_hashtbl[CLIENT_HASH_SIZE];
311 static struct list_head client_lru;
312 static struct list_head close_lru;
315 renew_client(struct nfs4_client *clp)
318 * Move client to the end to the LRU list.
320 dprintk("renewing client (clientid %08x/%08x)\n",
321 clp->cl_clientid.cl_boot,
322 clp->cl_clientid.cl_id);
323 list_move_tail(&clp->cl_lru, &client_lru);
324 clp->cl_time = get_seconds();
327 /* SETCLIENTID and SETCLIENTID_CONFIRM Helper functions */
329 STALE_CLIENTID(clientid_t *clid)
331 if (clid->cl_boot == boot_time)
333 dprintk("NFSD stale clientid (%08x/%08x)\n",
334 clid->cl_boot, clid->cl_id);
339 * XXX Should we use a slab cache ?
340 * This type of memory management is somewhat inefficient, but we use it
341 * anyway since SETCLIENTID is not a common operation.
343 static struct nfs4_client *alloc_client(struct xdr_netobj name)
345 struct nfs4_client *clp;
347 clp = kzalloc(sizeof(struct nfs4_client), GFP_KERNEL);
350 clp->cl_name.data = kmalloc(name.len, GFP_KERNEL);
351 if (clp->cl_name.data == NULL) {
355 memcpy(clp->cl_name.data, name.data, name.len);
356 clp->cl_name.len = name.len;
361 shutdown_callback_client(struct nfs4_client *clp)
363 struct rpc_clnt *clnt = clp->cl_callback.cb_client;
367 * Callback threads take a reference on the client, so there
368 * should be no outstanding callbacks at this point.
370 clp->cl_callback.cb_client = NULL;
371 rpc_shutdown_client(clnt);
376 free_client(struct nfs4_client *clp)
378 shutdown_callback_client(clp);
379 if (clp->cl_cred.cr_group_info)
380 put_group_info(clp->cl_cred.cr_group_info);
381 kfree(clp->cl_principal);
382 kfree(clp->cl_name.data);
387 put_nfs4_client(struct nfs4_client *clp)
389 if (atomic_dec_and_test(&clp->cl_count))
394 expire_client(struct nfs4_client *clp)
396 struct nfs4_stateowner *sop;
397 struct nfs4_delegation *dp;
398 struct list_head reaplist;
400 dprintk("NFSD: expire_client cl_count %d\n",
401 atomic_read(&clp->cl_count));
403 INIT_LIST_HEAD(&reaplist);
404 spin_lock(&recall_lock);
405 while (!list_empty(&clp->cl_delegations)) {
406 dp = list_entry(clp->cl_delegations.next, struct nfs4_delegation, dl_perclnt);
407 dprintk("NFSD: expire client. dp %p, fp %p\n", dp,
409 list_del_init(&dp->dl_perclnt);
410 list_move(&dp->dl_recall_lru, &reaplist);
412 spin_unlock(&recall_lock);
413 while (!list_empty(&reaplist)) {
414 dp = list_entry(reaplist.next, struct nfs4_delegation, dl_recall_lru);
415 list_del_init(&dp->dl_recall_lru);
416 unhash_delegation(dp);
418 list_del(&clp->cl_idhash);
419 list_del(&clp->cl_strhash);
420 list_del(&clp->cl_lru);
421 while (!list_empty(&clp->cl_openowners)) {
422 sop = list_entry(clp->cl_openowners.next, struct nfs4_stateowner, so_perclient);
423 release_stateowner(sop);
425 put_nfs4_client(clp);
428 static struct nfs4_client *create_client(struct xdr_netobj name, char *recdir)
430 struct nfs4_client *clp;
432 clp = alloc_client(name);
435 memcpy(clp->cl_recdir, recdir, HEXDIR_LEN);
436 atomic_set(&clp->cl_count, 1);
437 atomic_set(&clp->cl_callback.cb_set, 0);
438 INIT_LIST_HEAD(&clp->cl_idhash);
439 INIT_LIST_HEAD(&clp->cl_strhash);
440 INIT_LIST_HEAD(&clp->cl_openowners);
441 INIT_LIST_HEAD(&clp->cl_delegations);
442 INIT_LIST_HEAD(&clp->cl_lru);
446 static void copy_verf(struct nfs4_client *target, nfs4_verifier *source)
448 memcpy(target->cl_verifier.data, source->data,
449 sizeof(target->cl_verifier.data));
452 static void copy_clid(struct nfs4_client *target, struct nfs4_client *source)
454 target->cl_clientid.cl_boot = source->cl_clientid.cl_boot;
455 target->cl_clientid.cl_id = source->cl_clientid.cl_id;
458 static void copy_cred(struct svc_cred *target, struct svc_cred *source)
460 target->cr_uid = source->cr_uid;
461 target->cr_gid = source->cr_gid;
462 target->cr_group_info = source->cr_group_info;
463 get_group_info(target->cr_group_info);
466 static int same_name(const char *n1, const char *n2)
468 return 0 == memcmp(n1, n2, HEXDIR_LEN);
472 same_verf(nfs4_verifier *v1, nfs4_verifier *v2)
474 return 0 == memcmp(v1->data, v2->data, sizeof(v1->data));
478 same_clid(clientid_t *cl1, clientid_t *cl2)
480 return (cl1->cl_boot == cl2->cl_boot) && (cl1->cl_id == cl2->cl_id);
483 /* XXX what about NGROUP */
485 same_creds(struct svc_cred *cr1, struct svc_cred *cr2)
487 return cr1->cr_uid == cr2->cr_uid;
490 static void gen_clid(struct nfs4_client *clp)
492 static u32 current_clientid = 1;
494 clp->cl_clientid.cl_boot = boot_time;
495 clp->cl_clientid.cl_id = current_clientid++;
498 static void gen_confirm(struct nfs4_client *clp)
503 p = (u32 *)clp->cl_confirm.data;
504 *p++ = get_seconds();
508 static int check_name(struct xdr_netobj name)
512 if (name.len > NFS4_OPAQUE_LIMIT) {
513 dprintk("NFSD: check_name: name too long(%d)!\n", name.len);
520 add_to_unconfirmed(struct nfs4_client *clp, unsigned int strhashval)
522 unsigned int idhashval;
524 list_add(&clp->cl_strhash, &unconf_str_hashtbl[strhashval]);
525 idhashval = clientid_hashval(clp->cl_clientid.cl_id);
526 list_add(&clp->cl_idhash, &unconf_id_hashtbl[idhashval]);
527 list_add_tail(&clp->cl_lru, &client_lru);
528 clp->cl_time = get_seconds();
532 move_to_confirmed(struct nfs4_client *clp)
534 unsigned int idhashval = clientid_hashval(clp->cl_clientid.cl_id);
535 unsigned int strhashval;
537 dprintk("NFSD: move_to_confirm nfs4_client %p\n", clp);
538 list_del_init(&clp->cl_strhash);
539 list_move(&clp->cl_idhash, &conf_id_hashtbl[idhashval]);
540 strhashval = clientstr_hashval(clp->cl_recdir);
541 list_add(&clp->cl_strhash, &conf_str_hashtbl[strhashval]);
545 static struct nfs4_client *
546 find_confirmed_client(clientid_t *clid)
548 struct nfs4_client *clp;
549 unsigned int idhashval = clientid_hashval(clid->cl_id);
551 list_for_each_entry(clp, &conf_id_hashtbl[idhashval], cl_idhash) {
552 if (same_clid(&clp->cl_clientid, clid))
558 static struct nfs4_client *
559 find_unconfirmed_client(clientid_t *clid)
561 struct nfs4_client *clp;
562 unsigned int idhashval = clientid_hashval(clid->cl_id);
564 list_for_each_entry(clp, &unconf_id_hashtbl[idhashval], cl_idhash) {
565 if (same_clid(&clp->cl_clientid, clid))
571 static struct nfs4_client *
572 find_confirmed_client_by_str(const char *dname, unsigned int hashval)
574 struct nfs4_client *clp;
576 list_for_each_entry(clp, &conf_str_hashtbl[hashval], cl_strhash) {
577 if (same_name(clp->cl_recdir, dname))
583 static struct nfs4_client *
584 find_unconfirmed_client_by_str(const char *dname, unsigned int hashval)
586 struct nfs4_client *clp;
588 list_for_each_entry(clp, &unconf_str_hashtbl[hashval], cl_strhash) {
589 if (same_name(clp->cl_recdir, dname))
595 /* a helper function for parse_callback */
597 parse_octet(unsigned int *lenp, char **addrp)
599 unsigned int len = *lenp;
611 if ((c < '0') || (c > '9')) {
617 n = (n * 10) + (c - '0');
628 /* parse and set the setclientid ipv4 callback address */
630 parse_ipv4(unsigned int addr_len, char *addr_val, unsigned int *cbaddrp, unsigned short *cbportp)
635 u32 addrlen = addr_len;
636 char *addr = addr_val;
641 for(i = 4; i > 0 ; i--) {
642 if ((temp = parse_octet(&addrlen, &addr)) < 0) {
645 cbaddr |= (temp << shift);
653 for(i = 2; i > 0 ; i--) {
654 if ((temp = parse_octet(&addrlen, &addr)) < 0) {
657 cbport |= (temp << shift);
666 gen_callback(struct nfs4_client *clp, struct nfsd4_setclientid *se)
668 struct nfs4_callback *cb = &clp->cl_callback;
670 /* Currently, we only support tcp for the callback channel */
671 if ((se->se_callback_netid_len != 3) || memcmp((char *)se->se_callback_netid_val, "tcp", 3))
674 if ( !(parse_ipv4(se->se_callback_addr_len, se->se_callback_addr_val,
675 &cb->cb_addr, &cb->cb_port)))
677 cb->cb_prog = se->se_callback_prog;
678 cb->cb_ident = se->se_callback_ident;
681 dprintk(KERN_INFO "NFSD: this client (clientid %08x/%08x) "
682 "will not receive delegations\n",
683 clp->cl_clientid.cl_boot, clp->cl_clientid.cl_id);
689 nfsd4_setclientid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
690 struct nfsd4_setclientid *setclid)
692 struct sockaddr_in *sin = svc_addr_in(rqstp);
693 struct xdr_netobj clname = {
694 .len = setclid->se_namelen,
695 .data = setclid->se_name,
697 nfs4_verifier clverifier = setclid->se_verf;
698 unsigned int strhashval;
699 struct nfs4_client *conf, *unconf, *new;
702 char dname[HEXDIR_LEN];
704 if (!check_name(clname))
707 status = nfs4_make_rec_clidname(dname, &clname);
712 * XXX The Duplicate Request Cache (DRC) has been checked (??)
713 * We get here on a DRC miss.
716 strhashval = clientstr_hashval(dname);
719 conf = find_confirmed_client_by_str(dname, strhashval);
721 /* RFC 3530 14.2.33 CASE 0: */
722 status = nfserr_clid_inuse;
723 if (!same_creds(&conf->cl_cred, &rqstp->rq_cred)
724 || conf->cl_addr != sin->sin_addr.s_addr) {
725 dprintk("NFSD: setclientid: string in use by client"
726 "at %u.%u.%u.%u\n", NIPQUAD(conf->cl_addr));
731 * section 14.2.33 of RFC 3530 (under the heading "IMPLEMENTATION")
732 * has a description of SETCLIENTID request processing consisting
733 * of 5 bullet points, labeled as CASE0 - CASE4 below.
735 unconf = find_unconfirmed_client_by_str(dname, strhashval);
736 status = nfserr_resource;
739 * RFC 3530 14.2.33 CASE 4:
740 * placed first, because it is the normal case
743 expire_client(unconf);
744 new = create_client(clname, dname);
748 } else if (same_verf(&conf->cl_verifier, &clverifier)) {
750 * RFC 3530 14.2.33 CASE 1:
751 * probable callback update
754 /* Note this is removing unconfirmed {*x***},
755 * which is stronger than RFC recommended {vxc**}.
756 * This has the advantage that there is at most
757 * one {*x***} in either list at any time.
759 expire_client(unconf);
761 new = create_client(clname, dname);
764 copy_clid(new, conf);
765 } else if (!unconf) {
767 * RFC 3530 14.2.33 CASE 2:
768 * probable client reboot; state will be removed if
771 new = create_client(clname, dname);
777 * RFC 3530 14.2.33 CASE 3:
778 * probable client reboot; state will be removed if
781 expire_client(unconf);
782 new = create_client(clname, dname);
787 copy_verf(new, &clverifier);
788 new->cl_addr = sin->sin_addr.s_addr;
789 princ = svc_gss_principal(rqstp);
791 new->cl_principal = kstrdup(princ, GFP_KERNEL);
792 if (new->cl_principal == NULL) {
797 copy_cred(&new->cl_cred, &rqstp->rq_cred);
799 gen_callback(new, setclid);
800 add_to_unconfirmed(new, strhashval);
801 setclid->se_clientid.cl_boot = new->cl_clientid.cl_boot;
802 setclid->se_clientid.cl_id = new->cl_clientid.cl_id;
803 memcpy(setclid->se_confirm.data, new->cl_confirm.data, sizeof(setclid->se_confirm.data));
812 * Section 14.2.34 of RFC 3530 (under the heading "IMPLEMENTATION") has
813 * a description of SETCLIENTID_CONFIRM request processing consisting of 4
814 * bullets, labeled as CASE1 - CASE4 below.
817 nfsd4_setclientid_confirm(struct svc_rqst *rqstp,
818 struct nfsd4_compound_state *cstate,
819 struct nfsd4_setclientid_confirm *setclientid_confirm)
821 struct sockaddr_in *sin = svc_addr_in(rqstp);
822 struct nfs4_client *conf, *unconf;
823 nfs4_verifier confirm = setclientid_confirm->sc_confirm;
824 clientid_t * clid = &setclientid_confirm->sc_clientid;
827 if (STALE_CLIENTID(clid))
828 return nfserr_stale_clientid;
830 * XXX The Duplicate Request Cache (DRC) has been checked (??)
831 * We get here on a DRC miss.
836 conf = find_confirmed_client(clid);
837 unconf = find_unconfirmed_client(clid);
839 status = nfserr_clid_inuse;
840 if (conf && conf->cl_addr != sin->sin_addr.s_addr)
842 if (unconf && unconf->cl_addr != sin->sin_addr.s_addr)
846 * section 14.2.34 of RFC 3530 has a description of
847 * SETCLIENTID_CONFIRM request processing consisting
848 * of 4 bullet points, labeled as CASE1 - CASE4 below.
850 if (conf && unconf && same_verf(&confirm, &unconf->cl_confirm)) {
852 * RFC 3530 14.2.34 CASE 1:
855 if (!same_creds(&conf->cl_cred, &unconf->cl_cred))
856 status = nfserr_clid_inuse;
858 /* XXX: We just turn off callbacks until we can handle
859 * change request correctly. */
860 atomic_set(&conf->cl_callback.cb_set, 0);
862 nfsd4_remove_clid_dir(unconf);
863 expire_client(unconf);
867 } else if (conf && !unconf) {
869 * RFC 3530 14.2.34 CASE 2:
870 * probable retransmitted request; play it safe and
873 if (!same_creds(&conf->cl_cred, &rqstp->rq_cred))
874 status = nfserr_clid_inuse;
877 } else if (!conf && unconf
878 && same_verf(&unconf->cl_confirm, &confirm)) {
880 * RFC 3530 14.2.34 CASE 3:
881 * Normal case; new or rebooted client:
883 if (!same_creds(&unconf->cl_cred, &rqstp->rq_cred)) {
884 status = nfserr_clid_inuse;
887 clientstr_hashval(unconf->cl_recdir);
888 conf = find_confirmed_client_by_str(unconf->cl_recdir,
891 nfsd4_remove_clid_dir(conf);
894 move_to_confirmed(unconf);
896 nfsd4_probe_callback(conf);
899 } else if ((!conf || (conf && !same_verf(&conf->cl_confirm, &confirm)))
900 && (!unconf || (unconf && !same_verf(&unconf->cl_confirm,
903 * RFC 3530 14.2.34 CASE 4:
904 * Client probably hasn't noticed that we rebooted yet.
906 status = nfserr_stale_clientid;
908 /* check that we have hit one of the cases...*/
909 status = nfserr_clid_inuse;
916 /* OPEN Share state helper functions */
917 static inline struct nfs4_file *
918 alloc_init_file(struct inode *ino)
920 struct nfs4_file *fp;
921 unsigned int hashval = file_hashval(ino);
923 fp = kmem_cache_alloc(file_slab, GFP_KERNEL);
925 kref_init(&fp->fi_ref);
926 INIT_LIST_HEAD(&fp->fi_hash);
927 INIT_LIST_HEAD(&fp->fi_stateids);
928 INIT_LIST_HEAD(&fp->fi_delegations);
929 list_add(&fp->fi_hash, &file_hashtbl[hashval]);
930 fp->fi_inode = igrab(ino);
931 fp->fi_id = current_fileid++;
932 fp->fi_had_conflict = false;
939 nfsd4_free_slab(struct kmem_cache **slab)
943 kmem_cache_destroy(*slab);
948 nfsd4_free_slabs(void)
950 nfsd4_free_slab(&stateowner_slab);
951 nfsd4_free_slab(&file_slab);
952 nfsd4_free_slab(&stateid_slab);
953 nfsd4_free_slab(&deleg_slab);
957 nfsd4_init_slabs(void)
959 stateowner_slab = kmem_cache_create("nfsd4_stateowners",
960 sizeof(struct nfs4_stateowner), 0, 0, NULL);
961 if (stateowner_slab == NULL)
963 file_slab = kmem_cache_create("nfsd4_files",
964 sizeof(struct nfs4_file), 0, 0, NULL);
965 if (file_slab == NULL)
967 stateid_slab = kmem_cache_create("nfsd4_stateids",
968 sizeof(struct nfs4_stateid), 0, 0, NULL);
969 if (stateid_slab == NULL)
971 deleg_slab = kmem_cache_create("nfsd4_delegations",
972 sizeof(struct nfs4_delegation), 0, 0, NULL);
973 if (deleg_slab == NULL)
978 dprintk("nfsd4: out of memory while initializing nfsv4\n");
983 nfs4_free_stateowner(struct kref *kref)
985 struct nfs4_stateowner *sop =
986 container_of(kref, struct nfs4_stateowner, so_ref);
987 kfree(sop->so_owner.data);
988 kmem_cache_free(stateowner_slab, sop);
991 static inline struct nfs4_stateowner *
992 alloc_stateowner(struct xdr_netobj *owner)
994 struct nfs4_stateowner *sop;
996 if ((sop = kmem_cache_alloc(stateowner_slab, GFP_KERNEL))) {
997 if ((sop->so_owner.data = kmalloc(owner->len, GFP_KERNEL))) {
998 memcpy(sop->so_owner.data, owner->data, owner->len);
999 sop->so_owner.len = owner->len;
1000 kref_init(&sop->so_ref);
1003 kmem_cache_free(stateowner_slab, sop);
1008 static struct nfs4_stateowner *
1009 alloc_init_open_stateowner(unsigned int strhashval, struct nfs4_client *clp, struct nfsd4_open *open) {
1010 struct nfs4_stateowner *sop;
1011 struct nfs4_replay *rp;
1012 unsigned int idhashval;
1014 if (!(sop = alloc_stateowner(&open->op_owner)))
1016 idhashval = ownerid_hashval(current_ownerid);
1017 INIT_LIST_HEAD(&sop->so_idhash);
1018 INIT_LIST_HEAD(&sop->so_strhash);
1019 INIT_LIST_HEAD(&sop->so_perclient);
1020 INIT_LIST_HEAD(&sop->so_stateids);
1021 INIT_LIST_HEAD(&sop->so_perstateid); /* not used */
1022 INIT_LIST_HEAD(&sop->so_close_lru);
1024 list_add(&sop->so_idhash, &ownerid_hashtbl[idhashval]);
1025 list_add(&sop->so_strhash, &ownerstr_hashtbl[strhashval]);
1026 list_add(&sop->so_perclient, &clp->cl_openowners);
1027 sop->so_is_open_owner = 1;
1028 sop->so_id = current_ownerid++;
1029 sop->so_client = clp;
1030 sop->so_seqid = open->op_seqid;
1031 sop->so_confirmed = 0;
1032 rp = &sop->so_replay;
1033 rp->rp_status = nfserr_serverfault;
1035 rp->rp_buf = rp->rp_ibuf;
1040 release_stateid_lockowners(struct nfs4_stateid *open_stp)
1042 struct nfs4_stateowner *lock_sop;
1044 while (!list_empty(&open_stp->st_lockowners)) {
1045 lock_sop = list_entry(open_stp->st_lockowners.next,
1046 struct nfs4_stateowner, so_perstateid);
1047 /* list_del(&open_stp->st_lockowners); */
1048 BUG_ON(lock_sop->so_is_open_owner);
1049 release_stateowner(lock_sop);
1054 unhash_stateowner(struct nfs4_stateowner *sop)
1056 struct nfs4_stateid *stp;
1058 list_del(&sop->so_idhash);
1059 list_del(&sop->so_strhash);
1060 if (sop->so_is_open_owner)
1061 list_del(&sop->so_perclient);
1062 list_del(&sop->so_perstateid);
1063 while (!list_empty(&sop->so_stateids)) {
1064 stp = list_entry(sop->so_stateids.next,
1065 struct nfs4_stateid, st_perstateowner);
1066 if (sop->so_is_open_owner)
1067 release_stateid(stp, OPEN_STATE);
1069 release_stateid(stp, LOCK_STATE);
1074 release_stateowner(struct nfs4_stateowner *sop)
1076 unhash_stateowner(sop);
1077 list_del(&sop->so_close_lru);
1078 nfs4_put_stateowner(sop);
1082 init_stateid(struct nfs4_stateid *stp, struct nfs4_file *fp, struct nfsd4_open *open) {
1083 struct nfs4_stateowner *sop = open->op_stateowner;
1084 unsigned int hashval = stateid_hashval(sop->so_id, fp->fi_id);
1086 INIT_LIST_HEAD(&stp->st_hash);
1087 INIT_LIST_HEAD(&stp->st_perstateowner);
1088 INIT_LIST_HEAD(&stp->st_lockowners);
1089 INIT_LIST_HEAD(&stp->st_perfile);
1090 list_add(&stp->st_hash, &stateid_hashtbl[hashval]);
1091 list_add(&stp->st_perstateowner, &sop->so_stateids);
1092 list_add(&stp->st_perfile, &fp->fi_stateids);
1093 stp->st_stateowner = sop;
1096 stp->st_stateid.si_boot = boot_time;
1097 stp->st_stateid.si_stateownerid = sop->so_id;
1098 stp->st_stateid.si_fileid = fp->fi_id;
1099 stp->st_stateid.si_generation = 0;
1100 stp->st_access_bmap = 0;
1101 stp->st_deny_bmap = 0;
1102 __set_bit(open->op_share_access, &stp->st_access_bmap);
1103 __set_bit(open->op_share_deny, &stp->st_deny_bmap);
1104 stp->st_openstp = NULL;
1108 release_stateid(struct nfs4_stateid *stp, int flags)
1110 struct file *filp = stp->st_vfs_file;
1112 list_del(&stp->st_hash);
1113 list_del(&stp->st_perfile);
1114 list_del(&stp->st_perstateowner);
1115 if (flags & OPEN_STATE) {
1116 release_stateid_lockowners(stp);
1117 stp->st_vfs_file = NULL;
1119 } else if (flags & LOCK_STATE)
1120 locks_remove_posix(filp, (fl_owner_t) stp->st_stateowner);
1121 put_nfs4_file(stp->st_file);
1122 kmem_cache_free(stateid_slab, stp);
1126 move_to_close_lru(struct nfs4_stateowner *sop)
1128 dprintk("NFSD: move_to_close_lru nfs4_stateowner %p\n", sop);
1130 list_move_tail(&sop->so_close_lru, &close_lru);
1131 sop->so_time = get_seconds();
1135 same_owner_str(struct nfs4_stateowner *sop, struct xdr_netobj *owner,
1138 return (sop->so_owner.len == owner->len) &&
1139 0 == memcmp(sop->so_owner.data, owner->data, owner->len) &&
1140 (sop->so_client->cl_clientid.cl_id == clid->cl_id);
1143 static struct nfs4_stateowner *
1144 find_openstateowner_str(unsigned int hashval, struct nfsd4_open *open)
1146 struct nfs4_stateowner *so = NULL;
1148 list_for_each_entry(so, &ownerstr_hashtbl[hashval], so_strhash) {
1149 if (same_owner_str(so, &open->op_owner, &open->op_clientid))
1155 /* search file_hashtbl[] for file */
1156 static struct nfs4_file *
1157 find_file(struct inode *ino)
1159 unsigned int hashval = file_hashval(ino);
1160 struct nfs4_file *fp;
1162 list_for_each_entry(fp, &file_hashtbl[hashval], fi_hash) {
1163 if (fp->fi_inode == ino) {
1171 static inline int access_valid(u32 x)
1173 if (x < NFS4_SHARE_ACCESS_READ)
1175 if (x > NFS4_SHARE_ACCESS_BOTH)
1180 static inline int deny_valid(u32 x)
1182 /* Note: unlike access bits, deny bits may be zero. */
1183 return x <= NFS4_SHARE_DENY_BOTH;
1187 * We store the NONE, READ, WRITE, and BOTH bits separately in the
1188 * st_{access,deny}_bmap field of the stateid, in order to track not
1189 * only what share bits are currently in force, but also what
1190 * combinations of share bits previous opens have used. This allows us
1191 * to enforce the recommendation of rfc 3530 14.2.19 that the server
1192 * return an error if the client attempt to downgrade to a combination
1193 * of share bits not explicable by closing some of its previous opens.
1195 * XXX: This enforcement is actually incomplete, since we don't keep
1196 * track of access/deny bit combinations; so, e.g., we allow:
1198 * OPEN allow read, deny write
1199 * OPEN allow both, deny none
1200 * DOWNGRADE allow read, deny none
1202 * which we should reject.
1205 set_access(unsigned int *access, unsigned long bmap) {
1209 for (i = 1; i < 4; i++) {
1210 if (test_bit(i, &bmap))
1216 set_deny(unsigned int *deny, unsigned long bmap) {
1220 for (i = 0; i < 4; i++) {
1221 if (test_bit(i, &bmap))
1227 test_share(struct nfs4_stateid *stp, struct nfsd4_open *open) {
1228 unsigned int access, deny;
1230 set_access(&access, stp->st_access_bmap);
1231 set_deny(&deny, stp->st_deny_bmap);
1232 if ((access & open->op_share_deny) || (deny & open->op_share_access))
1238 * Called to check deny when READ with all zero stateid or
1239 * WRITE with all zero or all one stateid
1242 nfs4_share_conflict(struct svc_fh *current_fh, unsigned int deny_type)
1244 struct inode *ino = current_fh->fh_dentry->d_inode;
1245 struct nfs4_file *fp;
1246 struct nfs4_stateid *stp;
1249 dprintk("NFSD: nfs4_share_conflict\n");
1251 fp = find_file(ino);
1254 ret = nfserr_locked;
1255 /* Search for conflicting share reservations */
1256 list_for_each_entry(stp, &fp->fi_stateids, st_perfile) {
1257 if (test_bit(deny_type, &stp->st_deny_bmap) ||
1258 test_bit(NFS4_SHARE_DENY_BOTH, &stp->st_deny_bmap))
1268 nfs4_file_downgrade(struct file *filp, unsigned int share_access)
1270 if (share_access & NFS4_SHARE_ACCESS_WRITE) {
1271 drop_file_write_access(filp);
1272 filp->f_mode = (filp->f_mode | FMODE_READ) & ~FMODE_WRITE;
1277 * Recall a delegation
1280 do_recall(void *__dp)
1282 struct nfs4_delegation *dp = __dp;
1284 dp->dl_file->fi_had_conflict = true;
1285 nfsd4_cb_recall(dp);
1290 * Spawn a thread to perform a recall on the delegation represented
1291 * by the lease (file_lock)
1293 * Called from break_lease() with lock_kernel() held.
1294 * Note: we assume break_lease will only call this *once* for any given
1298 void nfsd_break_deleg_cb(struct file_lock *fl)
1300 struct nfs4_delegation *dp= (struct nfs4_delegation *)fl->fl_owner;
1301 struct task_struct *t;
1303 dprintk("NFSD nfsd_break_deleg_cb: dp %p fl %p\n",dp,fl);
1307 /* We're assuming the state code never drops its reference
1308 * without first removing the lease. Since we're in this lease
1309 * callback (and since the lease code is serialized by the kernel
1310 * lock) we know the server hasn't removed the lease yet, we know
1311 * it's safe to take a reference: */
1312 atomic_inc(&dp->dl_count);
1313 atomic_inc(&dp->dl_client->cl_count);
1315 spin_lock(&recall_lock);
1316 list_add_tail(&dp->dl_recall_lru, &del_recall_lru);
1317 spin_unlock(&recall_lock);
1319 /* only place dl_time is set. protected by lock_kernel*/
1320 dp->dl_time = get_seconds();
1323 * We don't want the locks code to timeout the lease for us;
1324 * we'll remove it ourself if the delegation isn't returned
1327 fl->fl_break_time = 0;
1329 t = kthread_run(do_recall, dp, "%s", "nfs4_cb_recall");
1331 struct nfs4_client *clp = dp->dl_client;
1333 printk(KERN_INFO "NFSD: Callback thread failed for "
1334 "for client (clientid %08x/%08x)\n",
1335 clp->cl_clientid.cl_boot, clp->cl_clientid.cl_id);
1336 put_nfs4_client(dp->dl_client);
1337 nfs4_put_delegation(dp);
1342 * The file_lock is being reapd.
1344 * Called by locks_free_lock() with lock_kernel() held.
1347 void nfsd_release_deleg_cb(struct file_lock *fl)
1349 struct nfs4_delegation *dp = (struct nfs4_delegation *)fl->fl_owner;
1351 dprintk("NFSD nfsd_release_deleg_cb: fl %p dp %p dl_count %d\n", fl,dp, atomic_read(&dp->dl_count));
1353 if (!(fl->fl_flags & FL_LEASE) || !dp)
1355 dp->dl_flock = NULL;
1359 * Set the delegation file_lock back pointer.
1361 * Called from setlease() with lock_kernel() held.
1364 void nfsd_copy_lock_deleg_cb(struct file_lock *new, struct file_lock *fl)
1366 struct nfs4_delegation *dp = (struct nfs4_delegation *)new->fl_owner;
1368 dprintk("NFSD: nfsd_copy_lock_deleg_cb: new fl %p dp %p\n", new, dp);
1375 * Called from setlease() with lock_kernel() held
1378 int nfsd_same_client_deleg_cb(struct file_lock *onlist, struct file_lock *try)
1380 struct nfs4_delegation *onlistd =
1381 (struct nfs4_delegation *)onlist->fl_owner;
1382 struct nfs4_delegation *tryd =
1383 (struct nfs4_delegation *)try->fl_owner;
1385 if (onlist->fl_lmops != try->fl_lmops)
1388 return onlistd->dl_client == tryd->dl_client;
1393 int nfsd_change_deleg_cb(struct file_lock **onlist, int arg)
1396 return lease_modify(onlist, arg);
1401 static struct lock_manager_operations nfsd_lease_mng_ops = {
1402 .fl_break = nfsd_break_deleg_cb,
1403 .fl_release_private = nfsd_release_deleg_cb,
1404 .fl_copy_lock = nfsd_copy_lock_deleg_cb,
1405 .fl_mylease = nfsd_same_client_deleg_cb,
1406 .fl_change = nfsd_change_deleg_cb,
1411 nfsd4_process_open1(struct nfsd4_open *open)
1413 clientid_t *clientid = &open->op_clientid;
1414 struct nfs4_client *clp = NULL;
1415 unsigned int strhashval;
1416 struct nfs4_stateowner *sop = NULL;
1418 if (!check_name(open->op_owner))
1419 return nfserr_inval;
1421 if (STALE_CLIENTID(&open->op_clientid))
1422 return nfserr_stale_clientid;
1424 strhashval = ownerstr_hashval(clientid->cl_id, open->op_owner);
1425 sop = find_openstateowner_str(strhashval, open);
1426 open->op_stateowner = sop;
1428 /* Make sure the client's lease hasn't expired. */
1429 clp = find_confirmed_client(clientid);
1431 return nfserr_expired;
1434 if (!sop->so_confirmed) {
1435 /* Replace unconfirmed owners without checking for replay. */
1436 clp = sop->so_client;
1437 release_stateowner(sop);
1438 open->op_stateowner = NULL;
1441 if (open->op_seqid == sop->so_seqid - 1) {
1442 if (sop->so_replay.rp_buflen)
1443 return nfserr_replay_me;
1444 /* The original OPEN failed so spectacularly
1445 * that we don't even have replay data saved!
1446 * Therefore, we have no choice but to continue
1447 * processing this OPEN; presumably, we'll
1448 * fail again for the same reason.
1450 dprintk("nfsd4_process_open1: replay with no replay cache\n");
1453 if (open->op_seqid != sop->so_seqid)
1454 return nfserr_bad_seqid;
1456 if (open->op_stateowner == NULL) {
1457 sop = alloc_init_open_stateowner(strhashval, clp, open);
1459 return nfserr_resource;
1460 open->op_stateowner = sop;
1462 list_del_init(&sop->so_close_lru);
1463 renew_client(sop->so_client);
1467 static inline __be32
1468 nfs4_check_delegmode(struct nfs4_delegation *dp, int flags)
1470 if ((flags & WR_STATE) && (dp->dl_type == NFS4_OPEN_DELEGATE_READ))
1471 return nfserr_openmode;
1476 static struct nfs4_delegation *
1477 find_delegation_file(struct nfs4_file *fp, stateid_t *stid)
1479 struct nfs4_delegation *dp;
1481 list_for_each_entry(dp, &fp->fi_delegations, dl_perfile) {
1482 if (dp->dl_stateid.si_stateownerid == stid->si_stateownerid)
1489 nfs4_check_deleg(struct nfs4_file *fp, struct nfsd4_open *open,
1490 struct nfs4_delegation **dp)
1493 __be32 status = nfserr_bad_stateid;
1495 *dp = find_delegation_file(fp, &open->op_delegate_stateid);
1498 flags = open->op_share_access == NFS4_SHARE_ACCESS_READ ?
1499 RD_STATE : WR_STATE;
1500 status = nfs4_check_delegmode(*dp, flags);
1504 if (open->op_claim_type != NFS4_OPEN_CLAIM_DELEGATE_CUR)
1508 open->op_stateowner->so_confirmed = 1;
1513 nfs4_check_open(struct nfs4_file *fp, struct nfsd4_open *open, struct nfs4_stateid **stpp)
1515 struct nfs4_stateid *local;
1516 __be32 status = nfserr_share_denied;
1517 struct nfs4_stateowner *sop = open->op_stateowner;
1519 list_for_each_entry(local, &fp->fi_stateids, st_perfile) {
1520 /* ignore lock owners */
1521 if (local->st_stateowner->so_is_open_owner == 0)
1523 /* remember if we have seen this open owner */
1524 if (local->st_stateowner == sop)
1526 /* check for conflicting share reservations */
1527 if (!test_share(local, open))
1535 static inline struct nfs4_stateid *
1536 nfs4_alloc_stateid(void)
1538 return kmem_cache_alloc(stateid_slab, GFP_KERNEL);
1542 nfs4_new_open(struct svc_rqst *rqstp, struct nfs4_stateid **stpp,
1543 struct nfs4_delegation *dp,
1544 struct svc_fh *cur_fh, int flags)
1546 struct nfs4_stateid *stp;
1548 stp = nfs4_alloc_stateid();
1550 return nfserr_resource;
1553 get_file(dp->dl_vfs_file);
1554 stp->st_vfs_file = dp->dl_vfs_file;
1557 status = nfsd_open(rqstp, cur_fh, S_IFREG, flags,
1560 if (status == nfserr_dropit)
1561 status = nfserr_jukebox;
1562 kmem_cache_free(stateid_slab, stp);
1570 static inline __be32
1571 nfsd4_truncate(struct svc_rqst *rqstp, struct svc_fh *fh,
1572 struct nfsd4_open *open)
1574 struct iattr iattr = {
1575 .ia_valid = ATTR_SIZE,
1578 if (!open->op_truncate)
1580 if (!(open->op_share_access & NFS4_SHARE_ACCESS_WRITE))
1581 return nfserr_inval;
1582 return nfsd_setattr(rqstp, fh, &iattr, 0, (time_t)0);
1586 nfs4_upgrade_open(struct svc_rqst *rqstp, struct svc_fh *cur_fh, struct nfs4_stateid *stp, struct nfsd4_open *open)
1588 struct file *filp = stp->st_vfs_file;
1589 struct inode *inode = filp->f_path.dentry->d_inode;
1590 unsigned int share_access, new_writer;
1593 set_access(&share_access, stp->st_access_bmap);
1594 new_writer = (~share_access) & open->op_share_access
1595 & NFS4_SHARE_ACCESS_WRITE;
1598 int err = get_write_access(inode);
1600 return nfserrno(err);
1601 err = mnt_want_write(cur_fh->fh_export->ex_path.mnt);
1603 return nfserrno(err);
1604 file_take_write(filp);
1606 status = nfsd4_truncate(rqstp, cur_fh, open);
1609 put_write_access(inode);
1612 /* remember the open */
1613 filp->f_mode |= open->op_share_access;
1614 __set_bit(open->op_share_access, &stp->st_access_bmap);
1615 __set_bit(open->op_share_deny, &stp->st_deny_bmap);
1622 nfs4_set_claim_prev(struct nfsd4_open *open)
1624 open->op_stateowner->so_confirmed = 1;
1625 open->op_stateowner->so_client->cl_firststate = 1;
1629 * Attempt to hand out a delegation.
1632 nfs4_open_delegation(struct svc_fh *fh, struct nfsd4_open *open, struct nfs4_stateid *stp)
1634 struct nfs4_delegation *dp;
1635 struct nfs4_stateowner *sop = stp->st_stateowner;
1636 struct nfs4_callback *cb = &sop->so_client->cl_callback;
1637 struct file_lock fl, *flp = &fl;
1638 int status, flag = 0;
1640 flag = NFS4_OPEN_DELEGATE_NONE;
1641 open->op_recall = 0;
1642 switch (open->op_claim_type) {
1643 case NFS4_OPEN_CLAIM_PREVIOUS:
1644 if (!atomic_read(&cb->cb_set))
1645 open->op_recall = 1;
1646 flag = open->op_delegate_type;
1647 if (flag == NFS4_OPEN_DELEGATE_NONE)
1650 case NFS4_OPEN_CLAIM_NULL:
1651 /* Let's not give out any delegations till everyone's
1652 * had the chance to reclaim theirs.... */
1653 if (locks_in_grace())
1655 if (!atomic_read(&cb->cb_set) || !sop->so_confirmed)
1657 if (open->op_share_access & NFS4_SHARE_ACCESS_WRITE)
1658 flag = NFS4_OPEN_DELEGATE_WRITE;
1660 flag = NFS4_OPEN_DELEGATE_READ;
1666 dp = alloc_init_deleg(sop->so_client, stp, fh, flag);
1668 flag = NFS4_OPEN_DELEGATE_NONE;
1671 locks_init_lock(&fl);
1672 fl.fl_lmops = &nfsd_lease_mng_ops;
1673 fl.fl_flags = FL_LEASE;
1674 fl.fl_type = flag == NFS4_OPEN_DELEGATE_READ? F_RDLCK: F_WRLCK;
1675 fl.fl_end = OFFSET_MAX;
1676 fl.fl_owner = (fl_owner_t)dp;
1677 fl.fl_file = stp->st_vfs_file;
1678 fl.fl_pid = current->tgid;
1680 /* vfs_setlease checks to see if delegation should be handed out.
1681 * the lock_manager callbacks fl_mylease and fl_change are used
1683 if ((status = vfs_setlease(stp->st_vfs_file, fl.fl_type, &flp))) {
1684 dprintk("NFSD: setlease failed [%d], no delegation\n", status);
1685 unhash_delegation(dp);
1686 flag = NFS4_OPEN_DELEGATE_NONE;
1690 memcpy(&open->op_delegate_stateid, &dp->dl_stateid, sizeof(dp->dl_stateid));
1692 dprintk("NFSD: delegation stateid=(%08x/%08x/%08x/%08x)\n\n",
1693 dp->dl_stateid.si_boot,
1694 dp->dl_stateid.si_stateownerid,
1695 dp->dl_stateid.si_fileid,
1696 dp->dl_stateid.si_generation);
1698 if (open->op_claim_type == NFS4_OPEN_CLAIM_PREVIOUS
1699 && flag == NFS4_OPEN_DELEGATE_NONE
1700 && open->op_delegate_type != NFS4_OPEN_DELEGATE_NONE)
1701 dprintk("NFSD: WARNING: refusing delegation reclaim\n");
1702 open->op_delegate_type = flag;
1706 * called with nfs4_lock_state() held.
1709 nfsd4_process_open2(struct svc_rqst *rqstp, struct svc_fh *current_fh, struct nfsd4_open *open)
1711 struct nfs4_file *fp = NULL;
1712 struct inode *ino = current_fh->fh_dentry->d_inode;
1713 struct nfs4_stateid *stp = NULL;
1714 struct nfs4_delegation *dp = NULL;
1717 status = nfserr_inval;
1718 if (!access_valid(open->op_share_access)
1719 || !deny_valid(open->op_share_deny))
1722 * Lookup file; if found, lookup stateid and check open request,
1723 * and check for delegations in the process of being recalled.
1724 * If not found, create the nfs4_file struct
1726 fp = find_file(ino);
1728 if ((status = nfs4_check_open(fp, open, &stp)))
1730 status = nfs4_check_deleg(fp, open, &dp);
1734 status = nfserr_bad_stateid;
1735 if (open->op_claim_type == NFS4_OPEN_CLAIM_DELEGATE_CUR)
1737 status = nfserr_resource;
1738 fp = alloc_init_file(ino);
1744 * OPEN the file, or upgrade an existing OPEN.
1745 * If truncate fails, the OPEN fails.
1748 /* Stateid was found, this is an OPEN upgrade */
1749 status = nfs4_upgrade_open(rqstp, current_fh, stp, open);
1752 update_stateid(&stp->st_stateid);
1754 /* Stateid was not found, this is a new OPEN */
1756 if (open->op_share_access & NFS4_SHARE_ACCESS_READ)
1757 flags |= NFSD_MAY_READ;
1758 if (open->op_share_access & NFS4_SHARE_ACCESS_WRITE)
1759 flags |= NFSD_MAY_WRITE;
1760 status = nfs4_new_open(rqstp, &stp, dp, current_fh, flags);
1763 init_stateid(stp, fp, open);
1764 status = nfsd4_truncate(rqstp, current_fh, open);
1766 release_stateid(stp, OPEN_STATE);
1770 memcpy(&open->op_stateid, &stp->st_stateid, sizeof(stateid_t));
1773 * Attempt to hand out a delegation. No error return, because the
1774 * OPEN succeeds even if we fail.
1776 nfs4_open_delegation(current_fh, open, stp);
1780 dprintk("nfs4_process_open2: stateid=(%08x/%08x/%08x/%08x)\n",
1781 stp->st_stateid.si_boot, stp->st_stateid.si_stateownerid,
1782 stp->st_stateid.si_fileid, stp->st_stateid.si_generation);
1786 if (status == 0 && open->op_claim_type == NFS4_OPEN_CLAIM_PREVIOUS)
1787 nfs4_set_claim_prev(open);
1789 * To finish the open response, we just need to set the rflags.
1791 open->op_rflags = NFS4_OPEN_RESULT_LOCKTYPE_POSIX;
1792 if (!open->op_stateowner->so_confirmed)
1793 open->op_rflags |= NFS4_OPEN_RESULT_CONFIRM;
1799 nfsd4_renew(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
1802 struct nfs4_client *clp;
1806 dprintk("process_renew(%08x/%08x): starting\n",
1807 clid->cl_boot, clid->cl_id);
1808 status = nfserr_stale_clientid;
1809 if (STALE_CLIENTID(clid))
1811 clp = find_confirmed_client(clid);
1812 status = nfserr_expired;
1814 /* We assume the client took too long to RENEW. */
1815 dprintk("nfsd4_renew: clientid not found!\n");
1819 status = nfserr_cb_path_down;
1820 if (!list_empty(&clp->cl_delegations)
1821 && !atomic_read(&clp->cl_callback.cb_set))
1825 nfs4_unlock_state();
1829 struct lock_manager nfsd4_manager = {
1833 nfsd4_end_grace(void)
1835 dprintk("NFSD: end of grace period\n");
1836 nfsd4_recdir_purge_old();
1837 locks_end_grace(&nfsd4_manager);
1841 nfs4_laundromat(void)
1843 struct nfs4_client *clp;
1844 struct nfs4_stateowner *sop;
1845 struct nfs4_delegation *dp;
1846 struct list_head *pos, *next, reaplist;
1847 time_t cutoff = get_seconds() - NFSD_LEASE_TIME;
1848 time_t t, clientid_val = NFSD_LEASE_TIME;
1849 time_t u, test_val = NFSD_LEASE_TIME;
1853 dprintk("NFSD: laundromat service - starting\n");
1854 if (locks_in_grace())
1856 list_for_each_safe(pos, next, &client_lru) {
1857 clp = list_entry(pos, struct nfs4_client, cl_lru);
1858 if (time_after((unsigned long)clp->cl_time, (unsigned long)cutoff)) {
1859 t = clp->cl_time - cutoff;
1860 if (clientid_val > t)
1864 dprintk("NFSD: purging unused client (clientid %08x)\n",
1865 clp->cl_clientid.cl_id);
1866 nfsd4_remove_clid_dir(clp);
1869 INIT_LIST_HEAD(&reaplist);
1870 spin_lock(&recall_lock);
1871 list_for_each_safe(pos, next, &del_recall_lru) {
1872 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
1873 if (time_after((unsigned long)dp->dl_time, (unsigned long)cutoff)) {
1874 u = dp->dl_time - cutoff;
1879 dprintk("NFSD: purging unused delegation dp %p, fp %p\n",
1881 list_move(&dp->dl_recall_lru, &reaplist);
1883 spin_unlock(&recall_lock);
1884 list_for_each_safe(pos, next, &reaplist) {
1885 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
1886 list_del_init(&dp->dl_recall_lru);
1887 unhash_delegation(dp);
1889 test_val = NFSD_LEASE_TIME;
1890 list_for_each_safe(pos, next, &close_lru) {
1891 sop = list_entry(pos, struct nfs4_stateowner, so_close_lru);
1892 if (time_after((unsigned long)sop->so_time, (unsigned long)cutoff)) {
1893 u = sop->so_time - cutoff;
1898 dprintk("NFSD: purging unused open stateowner (so_id %d)\n",
1900 release_stateowner(sop);
1902 if (clientid_val < NFSD_LAUNDROMAT_MINTIMEOUT)
1903 clientid_val = NFSD_LAUNDROMAT_MINTIMEOUT;
1904 nfs4_unlock_state();
1905 return clientid_val;
1908 static struct workqueue_struct *laundry_wq;
1909 static void laundromat_main(struct work_struct *);
1910 static DECLARE_DELAYED_WORK(laundromat_work, laundromat_main);
1913 laundromat_main(struct work_struct *not_used)
1917 t = nfs4_laundromat();
1918 dprintk("NFSD: laundromat_main - sleeping for %ld seconds\n", t);
1919 queue_delayed_work(laundry_wq, &laundromat_work, t*HZ);
1922 static struct nfs4_stateowner *
1923 search_close_lru(u32 st_id, int flags)
1925 struct nfs4_stateowner *local = NULL;
1927 if (flags & CLOSE_STATE) {
1928 list_for_each_entry(local, &close_lru, so_close_lru) {
1929 if (local->so_id == st_id)
1937 nfs4_check_fh(struct svc_fh *fhp, struct nfs4_stateid *stp)
1939 return fhp->fh_dentry->d_inode != stp->st_vfs_file->f_path.dentry->d_inode;
1943 STALE_STATEID(stateid_t *stateid)
1945 if (stateid->si_boot == boot_time)
1947 dprintk("NFSD: stale stateid (%08x/%08x/%08x/%08x)!\n",
1948 stateid->si_boot, stateid->si_stateownerid, stateid->si_fileid,
1949 stateid->si_generation);
1954 access_permit_read(unsigned long access_bmap)
1956 return test_bit(NFS4_SHARE_ACCESS_READ, &access_bmap) ||
1957 test_bit(NFS4_SHARE_ACCESS_BOTH, &access_bmap) ||
1958 test_bit(NFS4_SHARE_ACCESS_WRITE, &access_bmap);
1962 access_permit_write(unsigned long access_bmap)
1964 return test_bit(NFS4_SHARE_ACCESS_WRITE, &access_bmap) ||
1965 test_bit(NFS4_SHARE_ACCESS_BOTH, &access_bmap);
1969 __be32 nfs4_check_openmode(struct nfs4_stateid *stp, int flags)
1971 __be32 status = nfserr_openmode;
1973 if ((flags & WR_STATE) && (!access_permit_write(stp->st_access_bmap)))
1975 if ((flags & RD_STATE) && (!access_permit_read(stp->st_access_bmap)))
1982 static inline __be32
1983 check_special_stateids(svc_fh *current_fh, stateid_t *stateid, int flags)
1985 /* Trying to call delegreturn with a special stateid? Yuch: */
1986 if (!(flags & (RD_STATE | WR_STATE)))
1987 return nfserr_bad_stateid;
1988 else if (ONE_STATEID(stateid) && (flags & RD_STATE))
1990 else if (locks_in_grace()) {
1991 /* Answer in remaining cases depends on existance of
1992 * conflicting state; so we must wait out the grace period. */
1993 return nfserr_grace;
1994 } else if (flags & WR_STATE)
1995 return nfs4_share_conflict(current_fh,
1996 NFS4_SHARE_DENY_WRITE);
1997 else /* (flags & RD_STATE) && ZERO_STATEID(stateid) */
1998 return nfs4_share_conflict(current_fh,
1999 NFS4_SHARE_DENY_READ);
2003 * Allow READ/WRITE during grace period on recovered state only for files
2004 * that are not able to provide mandatory locking.
2007 io_during_grace_disallowed(struct inode *inode, int flags)
2009 return locks_in_grace() && (flags & (RD_STATE | WR_STATE))
2010 && mandatory_lock(inode);
2013 static int check_stateid_generation(stateid_t *in, stateid_t *ref)
2015 /* If the client sends us a stateid from the future, it's buggy: */
2016 if (in->si_generation > ref->si_generation)
2017 return nfserr_bad_stateid;
2019 * The following, however, can happen. For example, if the
2020 * client sends an open and some IO at the same time, the open
2021 * may bump si_generation while the IO is still in flight.
2022 * Thanks to hard links and renames, the client never knows what
2023 * file an open will affect. So it could avoid that situation
2024 * only by serializing all opens and IO from the same open
2025 * owner. To recover from the old_stateid error, the client
2026 * will just have to retry the IO:
2028 if (in->si_generation < ref->si_generation)
2029 return nfserr_old_stateid;
2034 * Checks for stateid operations
2037 nfs4_preprocess_stateid_op(struct svc_fh *current_fh, stateid_t *stateid, int flags, struct file **filpp)
2039 struct nfs4_stateid *stp = NULL;
2040 struct nfs4_delegation *dp = NULL;
2042 struct inode *ino = current_fh->fh_dentry->d_inode;
2045 dprintk("NFSD: preprocess_stateid_op: stateid = (%08x/%08x/%08x/%08x)\n",
2046 stateid->si_boot, stateid->si_stateownerid,
2047 stateid->si_fileid, stateid->si_generation);
2051 if (io_during_grace_disallowed(ino, flags))
2052 return nfserr_grace;
2054 if (ZERO_STATEID(stateid) || ONE_STATEID(stateid))
2055 return check_special_stateids(current_fh, stateid, flags);
2058 status = nfserr_stale_stateid;
2059 if (STALE_STATEID(stateid))
2063 status = nfserr_bad_stateid;
2064 if (!stateid->si_fileid) { /* delegation stateid */
2065 if(!(dp = find_delegation_stateid(ino, stateid))) {
2066 dprintk("NFSD: delegation stateid not found\n");
2069 stidp = &dp->dl_stateid;
2070 } else { /* open or lock stateid */
2071 if (!(stp = find_stateid(stateid, flags))) {
2072 dprintk("NFSD: open or lock stateid not found\n");
2075 if ((flags & CHECK_FH) && nfs4_check_fh(current_fh, stp))
2077 if (!stp->st_stateowner->so_confirmed)
2079 stidp = &stp->st_stateid;
2081 status = check_stateid_generation(stateid, stidp);
2085 if ((status = nfs4_check_openmode(stp,flags)))
2087 renew_client(stp->st_stateowner->so_client);
2089 *filpp = stp->st_vfs_file;
2091 if ((status = nfs4_check_delegmode(dp, flags)))
2093 renew_client(dp->dl_client);
2094 if (flags & DELEG_RET)
2095 unhash_delegation(dp);
2097 *filpp = dp->dl_vfs_file;
2107 return (type == NFS4_READW_LT || type == NFS4_READ_LT) ?
2108 RD_STATE : WR_STATE;
2112 * Checks for sequence id mutating operations.
2115 nfs4_preprocess_seqid_op(struct svc_fh *current_fh, u32 seqid, stateid_t *stateid, int flags, struct nfs4_stateowner **sopp, struct nfs4_stateid **stpp, struct nfsd4_lock *lock)
2117 struct nfs4_stateid *stp;
2118 struct nfs4_stateowner *sop;
2121 dprintk("NFSD: preprocess_seqid_op: seqid=%d "
2122 "stateid = (%08x/%08x/%08x/%08x)\n", seqid,
2123 stateid->si_boot, stateid->si_stateownerid, stateid->si_fileid,
2124 stateid->si_generation);
2129 if (ZERO_STATEID(stateid) || ONE_STATEID(stateid)) {
2130 dprintk("NFSD: preprocess_seqid_op: magic stateid!\n");
2131 return nfserr_bad_stateid;
2134 if (STALE_STATEID(stateid))
2135 return nfserr_stale_stateid;
2137 * We return BAD_STATEID if filehandle doesn't match stateid,
2138 * the confirmed flag is incorrecly set, or the generation
2139 * number is incorrect.
2141 stp = find_stateid(stateid, flags);
2144 * Also, we should make sure this isn't just the result of
2147 sop = search_close_lru(stateid->si_stateownerid, flags);
2149 return nfserr_bad_stateid;
2155 *sopp = sop = stp->st_stateowner;
2158 clientid_t *lockclid = &lock->v.new.clientid;
2159 struct nfs4_client *clp = sop->so_client;
2163 lkflg = setlkflg(lock->lk_type);
2165 if (lock->lk_is_new) {
2166 if (!sop->so_is_open_owner)
2167 return nfserr_bad_stateid;
2168 if (!same_clid(&clp->cl_clientid, lockclid))
2169 return nfserr_bad_stateid;
2170 /* stp is the open stateid */
2171 status = nfs4_check_openmode(stp, lkflg);
2175 /* stp is the lock stateid */
2176 status = nfs4_check_openmode(stp->st_openstp, lkflg);
2182 if (nfs4_check_fh(current_fh, stp)) {
2183 dprintk("NFSD: preprocess_seqid_op: fh-stateid mismatch!\n");
2184 return nfserr_bad_stateid;
2188 * We now validate the seqid and stateid generation numbers.
2189 * For the moment, we ignore the possibility of
2190 * generation number wraparound.
2192 if (seqid != sop->so_seqid)
2195 if (sop->so_confirmed && flags & CONFIRM) {
2196 dprintk("NFSD: preprocess_seqid_op: expected"
2197 " unconfirmed stateowner!\n");
2198 return nfserr_bad_stateid;
2200 if (!sop->so_confirmed && !(flags & CONFIRM)) {
2201 dprintk("NFSD: preprocess_seqid_op: stateowner not"
2202 " confirmed yet!\n");
2203 return nfserr_bad_stateid;
2205 status = check_stateid_generation(stateid, &stp->st_stateid);
2208 renew_client(sop->so_client);
2212 if (seqid == sop->so_seqid - 1) {
2213 dprintk("NFSD: preprocess_seqid_op: retransmission?\n");
2214 /* indicate replay to calling function */
2215 return nfserr_replay_me;
2217 dprintk("NFSD: preprocess_seqid_op: bad seqid (expected %d, got %d)\n",
2218 sop->so_seqid, seqid);
2220 return nfserr_bad_seqid;
2224 nfsd4_open_confirm(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
2225 struct nfsd4_open_confirm *oc)
2228 struct nfs4_stateowner *sop;
2229 struct nfs4_stateid *stp;
2231 dprintk("NFSD: nfsd4_open_confirm on file %.*s\n",
2232 (int)cstate->current_fh.fh_dentry->d_name.len,
2233 cstate->current_fh.fh_dentry->d_name.name);
2235 status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0);
2241 if ((status = nfs4_preprocess_seqid_op(&cstate->current_fh,
2242 oc->oc_seqid, &oc->oc_req_stateid,
2243 CONFIRM | OPEN_STATE,
2244 &oc->oc_stateowner, &stp, NULL)))
2247 sop = oc->oc_stateowner;
2248 sop->so_confirmed = 1;
2249 update_stateid(&stp->st_stateid);
2250 memcpy(&oc->oc_resp_stateid, &stp->st_stateid, sizeof(stateid_t));
2251 dprintk("NFSD: nfsd4_open_confirm: success, seqid=%d "
2252 "stateid=(%08x/%08x/%08x/%08x)\n", oc->oc_seqid,
2253 stp->st_stateid.si_boot,
2254 stp->st_stateid.si_stateownerid,
2255 stp->st_stateid.si_fileid,
2256 stp->st_stateid.si_generation);
2258 nfsd4_create_clid_dir(sop->so_client);
2260 if (oc->oc_stateowner) {
2261 nfs4_get_stateowner(oc->oc_stateowner);
2262 cstate->replay_owner = oc->oc_stateowner;
2264 nfs4_unlock_state();
2270 * unset all bits in union bitmap (bmap) that
2271 * do not exist in share (from successful OPEN_DOWNGRADE)
2274 reset_union_bmap_access(unsigned long access, unsigned long *bmap)
2277 for (i = 1; i < 4; i++) {
2278 if ((i & access) != i)
2279 __clear_bit(i, bmap);
2284 reset_union_bmap_deny(unsigned long deny, unsigned long *bmap)
2287 for (i = 0; i < 4; i++) {
2288 if ((i & deny) != i)
2289 __clear_bit(i, bmap);
2294 nfsd4_open_downgrade(struct svc_rqst *rqstp,
2295 struct nfsd4_compound_state *cstate,
2296 struct nfsd4_open_downgrade *od)
2299 struct nfs4_stateid *stp;
2300 unsigned int share_access;
2302 dprintk("NFSD: nfsd4_open_downgrade on file %.*s\n",
2303 (int)cstate->current_fh.fh_dentry->d_name.len,
2304 cstate->current_fh.fh_dentry->d_name.name);
2306 if (!access_valid(od->od_share_access)
2307 || !deny_valid(od->od_share_deny))
2308 return nfserr_inval;
2311 if ((status = nfs4_preprocess_seqid_op(&cstate->current_fh,
2315 &od->od_stateowner, &stp, NULL)))
2318 status = nfserr_inval;
2319 if (!test_bit(od->od_share_access, &stp->st_access_bmap)) {
2320 dprintk("NFSD:access not a subset current bitmap: 0x%lx, input access=%08x\n",
2321 stp->st_access_bmap, od->od_share_access);
2324 if (!test_bit(od->od_share_deny, &stp->st_deny_bmap)) {
2325 dprintk("NFSD:deny not a subset current bitmap: 0x%lx, input deny=%08x\n",
2326 stp->st_deny_bmap, od->od_share_deny);
2329 set_access(&share_access, stp->st_access_bmap);
2330 nfs4_file_downgrade(stp->st_vfs_file,
2331 share_access & ~od->od_share_access);
2333 reset_union_bmap_access(od->od_share_access, &stp->st_access_bmap);
2334 reset_union_bmap_deny(od->od_share_deny, &stp->st_deny_bmap);
2336 update_stateid(&stp->st_stateid);
2337 memcpy(&od->od_stateid, &stp->st_stateid, sizeof(stateid_t));
2340 if (od->od_stateowner) {
2341 nfs4_get_stateowner(od->od_stateowner);
2342 cstate->replay_owner = od->od_stateowner;
2344 nfs4_unlock_state();
2349 * nfs4_unlock_state() called after encode
2352 nfsd4_close(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
2353 struct nfsd4_close *close)
2356 struct nfs4_stateid *stp;
2358 dprintk("NFSD: nfsd4_close on file %.*s\n",
2359 (int)cstate->current_fh.fh_dentry->d_name.len,
2360 cstate->current_fh.fh_dentry->d_name.name);
2363 /* check close_lru for replay */
2364 if ((status = nfs4_preprocess_seqid_op(&cstate->current_fh,
2367 OPEN_STATE | CLOSE_STATE,
2368 &close->cl_stateowner, &stp, NULL)))
2371 update_stateid(&stp->st_stateid);
2372 memcpy(&close->cl_stateid, &stp->st_stateid, sizeof(stateid_t));
2374 /* release_stateid() calls nfsd_close() if needed */
2375 release_stateid(stp, OPEN_STATE);
2377 /* place unused nfs4_stateowners on so_close_lru list to be
2378 * released by the laundromat service after the lease period
2379 * to enable us to handle CLOSE replay
2381 if (list_empty(&close->cl_stateowner->so_stateids))
2382 move_to_close_lru(close->cl_stateowner);
2384 if (close->cl_stateowner) {
2385 nfs4_get_stateowner(close->cl_stateowner);
2386 cstate->replay_owner = close->cl_stateowner;
2388 nfs4_unlock_state();
2393 nfsd4_delegreturn(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
2394 struct nfsd4_delegreturn *dr)
2398 if ((status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0)))
2402 status = nfs4_preprocess_stateid_op(&cstate->current_fh,
2403 &dr->dr_stateid, DELEG_RET, NULL);
2404 nfs4_unlock_state();
2411 * Lock owner state (byte-range locks)
2413 #define LOFF_OVERFLOW(start, len) ((u64)(len) > ~(u64)(start))
2414 #define LOCK_HASH_BITS 8
2415 #define LOCK_HASH_SIZE (1 << LOCK_HASH_BITS)
2416 #define LOCK_HASH_MASK (LOCK_HASH_SIZE - 1)
2418 #define lockownerid_hashval(id) \
2419 ((id) & LOCK_HASH_MASK)
2421 static inline unsigned int
2422 lock_ownerstr_hashval(struct inode *inode, u32 cl_id,
2423 struct xdr_netobj *ownername)
2425 return (file_hashval(inode) + cl_id
2426 + opaque_hashval(ownername->data, ownername->len))
2430 static struct list_head lock_ownerid_hashtbl[LOCK_HASH_SIZE];
2431 static struct list_head lock_ownerstr_hashtbl[LOCK_HASH_SIZE];
2432 static struct list_head lockstateid_hashtbl[STATEID_HASH_SIZE];
2434 static struct nfs4_stateid *
2435 find_stateid(stateid_t *stid, int flags)
2437 struct nfs4_stateid *local = NULL;
2438 u32 st_id = stid->si_stateownerid;
2439 u32 f_id = stid->si_fileid;
2440 unsigned int hashval;
2442 dprintk("NFSD: find_stateid flags 0x%x\n",flags);
2443 if ((flags & LOCK_STATE) || (flags & RD_STATE) || (flags & WR_STATE)) {
2444 hashval = stateid_hashval(st_id, f_id);
2445 list_for_each_entry(local, &lockstateid_hashtbl[hashval], st_hash) {
2446 if ((local->st_stateid.si_stateownerid == st_id) &&
2447 (local->st_stateid.si_fileid == f_id))
2451 if ((flags & OPEN_STATE) || (flags & RD_STATE) || (flags & WR_STATE)) {
2452 hashval = stateid_hashval(st_id, f_id);
2453 list_for_each_entry(local, &stateid_hashtbl[hashval], st_hash) {
2454 if ((local->st_stateid.si_stateownerid == st_id) &&
2455 (local->st_stateid.si_fileid == f_id))
2462 static struct nfs4_delegation *
2463 find_delegation_stateid(struct inode *ino, stateid_t *stid)
2465 struct nfs4_file *fp;
2466 struct nfs4_delegation *dl;
2468 dprintk("NFSD:find_delegation_stateid stateid=(%08x/%08x/%08x/%08x)\n",
2469 stid->si_boot, stid->si_stateownerid,
2470 stid->si_fileid, stid->si_generation);
2472 fp = find_file(ino);
2475 dl = find_delegation_file(fp, stid);
2481 * TODO: Linux file offsets are _signed_ 64-bit quantities, which means that
2482 * we can't properly handle lock requests that go beyond the (2^63 - 1)-th
2483 * byte, because of sign extension problems. Since NFSv4 calls for 64-bit
2484 * locking, this prevents us from being completely protocol-compliant. The
2485 * real solution to this problem is to start using unsigned file offsets in
2486 * the VFS, but this is a very deep change!
2489 nfs4_transform_lock_offset(struct file_lock *lock)
2491 if (lock->fl_start < 0)
2492 lock->fl_start = OFFSET_MAX;
2493 if (lock->fl_end < 0)
2494 lock->fl_end = OFFSET_MAX;
2497 /* Hack!: For now, we're defining this just so we can use a pointer to it
2498 * as a unique cookie to identify our (NFSv4's) posix locks. */
2499 static struct lock_manager_operations nfsd_posix_mng_ops = {
2503 nfs4_set_lock_denied(struct file_lock *fl, struct nfsd4_lock_denied *deny)
2505 struct nfs4_stateowner *sop;
2508 if (fl->fl_lmops == &nfsd_posix_mng_ops) {
2509 sop = (struct nfs4_stateowner *) fl->fl_owner;
2510 hval = lockownerid_hashval(sop->so_id);
2511 kref_get(&sop->so_ref);
2513 deny->ld_clientid = sop->so_client->cl_clientid;
2515 deny->ld_sop = NULL;
2516 deny->ld_clientid.cl_boot = 0;
2517 deny->ld_clientid.cl_id = 0;
2519 deny->ld_start = fl->fl_start;
2520 deny->ld_length = ~(u64)0;
2521 if (fl->fl_end != ~(u64)0)
2522 deny->ld_length = fl->fl_end - fl->fl_start + 1;
2523 deny->ld_type = NFS4_READ_LT;
2524 if (fl->fl_type != F_RDLCK)
2525 deny->ld_type = NFS4_WRITE_LT;
2528 static struct nfs4_stateowner *
2529 find_lockstateowner_str(struct inode *inode, clientid_t *clid,
2530 struct xdr_netobj *owner)
2532 unsigned int hashval = lock_ownerstr_hashval(inode, clid->cl_id, owner);
2533 struct nfs4_stateowner *op;
2535 list_for_each_entry(op, &lock_ownerstr_hashtbl[hashval], so_strhash) {
2536 if (same_owner_str(op, owner, clid))
2543 * Alloc a lock owner structure.
2544 * Called in nfsd4_lock - therefore, OPEN and OPEN_CONFIRM (if needed) has
2547 * strhashval = lock_ownerstr_hashval
2550 static struct nfs4_stateowner *
2551 alloc_init_lock_stateowner(unsigned int strhashval, struct nfs4_client *clp, struct nfs4_stateid *open_stp, struct nfsd4_lock *lock) {
2552 struct nfs4_stateowner *sop;
2553 struct nfs4_replay *rp;
2554 unsigned int idhashval;
2556 if (!(sop = alloc_stateowner(&lock->lk_new_owner)))
2558 idhashval = lockownerid_hashval(current_ownerid);
2559 INIT_LIST_HEAD(&sop->so_idhash);
2560 INIT_LIST_HEAD(&sop->so_strhash);
2561 INIT_LIST_HEAD(&sop->so_perclient);
2562 INIT_LIST_HEAD(&sop->so_stateids);
2563 INIT_LIST_HEAD(&sop->so_perstateid);
2564 INIT_LIST_HEAD(&sop->so_close_lru); /* not used */
2566 list_add(&sop->so_idhash, &lock_ownerid_hashtbl[idhashval]);
2567 list_add(&sop->so_strhash, &lock_ownerstr_hashtbl[strhashval]);
2568 list_add(&sop->so_perstateid, &open_stp->st_lockowners);
2569 sop->so_is_open_owner = 0;
2570 sop->so_id = current_ownerid++;
2571 sop->so_client = clp;
2572 /* It is the openowner seqid that will be incremented in encode in the
2573 * case of new lockowners; so increment the lock seqid manually: */
2574 sop->so_seqid = lock->lk_new_lock_seqid + 1;
2575 sop->so_confirmed = 1;
2576 rp = &sop->so_replay;
2577 rp->rp_status = nfserr_serverfault;
2579 rp->rp_buf = rp->rp_ibuf;
2583 static struct nfs4_stateid *
2584 alloc_init_lock_stateid(struct nfs4_stateowner *sop, struct nfs4_file *fp, struct nfs4_stateid *open_stp)
2586 struct nfs4_stateid *stp;
2587 unsigned int hashval = stateid_hashval(sop->so_id, fp->fi_id);
2589 stp = nfs4_alloc_stateid();
2592 INIT_LIST_HEAD(&stp->st_hash);
2593 INIT_LIST_HEAD(&stp->st_perfile);
2594 INIT_LIST_HEAD(&stp->st_perstateowner);
2595 INIT_LIST_HEAD(&stp->st_lockowners); /* not used */
2596 list_add(&stp->st_hash, &lockstateid_hashtbl[hashval]);
2597 list_add(&stp->st_perfile, &fp->fi_stateids);
2598 list_add(&stp->st_perstateowner, &sop->so_stateids);
2599 stp->st_stateowner = sop;
2602 stp->st_stateid.si_boot = boot_time;
2603 stp->st_stateid.si_stateownerid = sop->so_id;
2604 stp->st_stateid.si_fileid = fp->fi_id;
2605 stp->st_stateid.si_generation = 0;
2606 stp->st_vfs_file = open_stp->st_vfs_file; /* FIXME refcount?? */
2607 stp->st_access_bmap = open_stp->st_access_bmap;
2608 stp->st_deny_bmap = open_stp->st_deny_bmap;
2609 stp->st_openstp = open_stp;
2616 check_lock_length(u64 offset, u64 length)
2618 return ((length == 0) || ((length != ~(u64)0) &&
2619 LOFF_OVERFLOW(offset, length)));
2626 nfsd4_lock(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
2627 struct nfsd4_lock *lock)
2629 struct nfs4_stateowner *open_sop = NULL;
2630 struct nfs4_stateowner *lock_sop = NULL;
2631 struct nfs4_stateid *lock_stp;
2633 struct file_lock file_lock;
2634 struct file_lock conflock;
2636 unsigned int strhashval;
2640 dprintk("NFSD: nfsd4_lock: start=%Ld length=%Ld\n",
2641 (long long) lock->lk_offset,
2642 (long long) lock->lk_length);
2644 if (check_lock_length(lock->lk_offset, lock->lk_length))
2645 return nfserr_inval;
2647 if ((status = fh_verify(rqstp, &cstate->current_fh,
2648 S_IFREG, NFSD_MAY_LOCK))) {
2649 dprintk("NFSD: nfsd4_lock: permission denied!\n");
2655 if (lock->lk_is_new) {
2657 * Client indicates that this is a new lockowner.
2658 * Use open owner and open stateid to create lock owner and
2661 struct nfs4_stateid *open_stp = NULL;
2662 struct nfs4_file *fp;
2664 status = nfserr_stale_clientid;
2665 if (STALE_CLIENTID(&lock->lk_new_clientid))
2668 /* validate and update open stateid and open seqid */
2669 status = nfs4_preprocess_seqid_op(&cstate->current_fh,
2670 lock->lk_new_open_seqid,
2671 &lock->lk_new_open_stateid,
2673 &lock->lk_replay_owner, &open_stp,
2677 open_sop = lock->lk_replay_owner;
2678 /* create lockowner and lock stateid */
2679 fp = open_stp->st_file;
2680 strhashval = lock_ownerstr_hashval(fp->fi_inode,
2681 open_sop->so_client->cl_clientid.cl_id,
2682 &lock->v.new.owner);
2683 /* XXX: Do we need to check for duplicate stateowners on
2684 * the same file, or should they just be allowed (and
2685 * create new stateids)? */
2686 status = nfserr_resource;
2687 lock_sop = alloc_init_lock_stateowner(strhashval,
2688 open_sop->so_client, open_stp, lock);
2689 if (lock_sop == NULL)
2691 lock_stp = alloc_init_lock_stateid(lock_sop, fp, open_stp);
2692 if (lock_stp == NULL)
2695 /* lock (lock owner + lock stateid) already exists */
2696 status = nfs4_preprocess_seqid_op(&cstate->current_fh,
2697 lock->lk_old_lock_seqid,
2698 &lock->lk_old_lock_stateid,
2700 &lock->lk_replay_owner, &lock_stp, lock);
2703 lock_sop = lock->lk_replay_owner;
2705 /* lock->lk_replay_owner and lock_stp have been created or found */
2706 filp = lock_stp->st_vfs_file;
2708 status = nfserr_grace;
2709 if (locks_in_grace() && !lock->lk_reclaim)
2711 status = nfserr_no_grace;
2712 if (!locks_in_grace() && lock->lk_reclaim)
2715 locks_init_lock(&file_lock);
2716 switch (lock->lk_type) {
2719 file_lock.fl_type = F_RDLCK;
2723 case NFS4_WRITEW_LT:
2724 file_lock.fl_type = F_WRLCK;
2728 status = nfserr_inval;
2731 file_lock.fl_owner = (fl_owner_t)lock_sop;
2732 file_lock.fl_pid = current->tgid;
2733 file_lock.fl_file = filp;
2734 file_lock.fl_flags = FL_POSIX;
2735 file_lock.fl_lmops = &nfsd_posix_mng_ops;
2737 file_lock.fl_start = lock->lk_offset;
2738 if ((lock->lk_length == ~(u64)0) ||
2739 LOFF_OVERFLOW(lock->lk_offset, lock->lk_length))
2740 file_lock.fl_end = ~(u64)0;
2742 file_lock.fl_end = lock->lk_offset + lock->lk_length - 1;
2743 nfs4_transform_lock_offset(&file_lock);
2746 * Try to lock the file in the VFS.
2747 * Note: locks.c uses the BKL to protect the inode's lock list.
2750 err = vfs_lock_file(filp, cmd, &file_lock, &conflock);
2752 case 0: /* success! */
2753 update_stateid(&lock_stp->st_stateid);
2754 memcpy(&lock->lk_resp_stateid, &lock_stp->st_stateid,
2758 case (EAGAIN): /* conflock holds conflicting lock */
2759 status = nfserr_denied;
2760 dprintk("NFSD: nfsd4_lock: conflicting lock found!\n");
2761 nfs4_set_lock_denied(&conflock, &lock->lk_denied);
2764 status = nfserr_deadlock;
2767 dprintk("NFSD: nfsd4_lock: vfs_lock_file() failed! status %d\n",err);
2768 status = nfserr_resource;
2772 if (status && lock->lk_is_new && lock_sop)
2773 release_stateowner(lock_sop);
2774 if (lock->lk_replay_owner) {
2775 nfs4_get_stateowner(lock->lk_replay_owner);
2776 cstate->replay_owner = lock->lk_replay_owner;
2778 nfs4_unlock_state();
2786 nfsd4_lockt(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
2787 struct nfsd4_lockt *lockt)
2789 struct inode *inode;
2791 struct file_lock file_lock;
2795 if (locks_in_grace())
2796 return nfserr_grace;
2798 if (check_lock_length(lockt->lt_offset, lockt->lt_length))
2799 return nfserr_inval;
2801 lockt->lt_stateowner = NULL;
2804 status = nfserr_stale_clientid;
2805 if (STALE_CLIENTID(&lockt->lt_clientid))
2808 if ((status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0))) {
2809 dprintk("NFSD: nfsd4_lockt: fh_verify() failed!\n");
2810 if (status == nfserr_symlink)
2811 status = nfserr_inval;
2815 inode = cstate->current_fh.fh_dentry->d_inode;
2816 locks_init_lock(&file_lock);
2817 switch (lockt->lt_type) {
2820 file_lock.fl_type = F_RDLCK;
2823 case NFS4_WRITEW_LT:
2824 file_lock.fl_type = F_WRLCK;
2827 dprintk("NFSD: nfs4_lockt: bad lock type!\n");
2828 status = nfserr_inval;
2832 lockt->lt_stateowner = find_lockstateowner_str(inode,
2833 &lockt->lt_clientid, &lockt->lt_owner);
2834 if (lockt->lt_stateowner)
2835 file_lock.fl_owner = (fl_owner_t)lockt->lt_stateowner;
2836 file_lock.fl_pid = current->tgid;
2837 file_lock.fl_flags = FL_POSIX;
2838 file_lock.fl_lmops = &nfsd_posix_mng_ops;
2840 file_lock.fl_start = lockt->lt_offset;
2841 if ((lockt->lt_length == ~(u64)0) || LOFF_OVERFLOW(lockt->lt_offset, lockt->lt_length))
2842 file_lock.fl_end = ~(u64)0;
2844 file_lock.fl_end = lockt->lt_offset + lockt->lt_length - 1;
2846 nfs4_transform_lock_offset(&file_lock);
2848 /* vfs_test_lock uses the struct file _only_ to resolve the inode.
2849 * since LOCKT doesn't require an OPEN, and therefore a struct
2850 * file may not exist, pass vfs_test_lock a struct file with
2851 * only the dentry:inode set.
2853 memset(&file, 0, sizeof (struct file));
2854 file.f_path.dentry = cstate->current_fh.fh_dentry;
2857 error = vfs_test_lock(&file, &file_lock);
2859 status = nfserrno(error);
2862 if (file_lock.fl_type != F_UNLCK) {
2863 status = nfserr_denied;
2864 nfs4_set_lock_denied(&file_lock, &lockt->lt_denied);
2867 nfs4_unlock_state();
2872 nfsd4_locku(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
2873 struct nfsd4_locku *locku)
2875 struct nfs4_stateid *stp;
2876 struct file *filp = NULL;
2877 struct file_lock file_lock;
2881 dprintk("NFSD: nfsd4_locku: start=%Ld length=%Ld\n",
2882 (long long) locku->lu_offset,
2883 (long long) locku->lu_length);
2885 if (check_lock_length(locku->lu_offset, locku->lu_length))
2886 return nfserr_inval;
2890 if ((status = nfs4_preprocess_seqid_op(&cstate->current_fh,
2894 &locku->lu_stateowner, &stp, NULL)))
2897 filp = stp->st_vfs_file;
2899 locks_init_lock(&file_lock);
2900 file_lock.fl_type = F_UNLCK;
2901 file_lock.fl_owner = (fl_owner_t) locku->lu_stateowner;
2902 file_lock.fl_pid = current->tgid;
2903 file_lock.fl_file = filp;
2904 file_lock.fl_flags = FL_POSIX;
2905 file_lock.fl_lmops = &nfsd_posix_mng_ops;
2906 file_lock.fl_start = locku->lu_offset;
2908 if ((locku->lu_length == ~(u64)0) || LOFF_OVERFLOW(locku->lu_offset, locku->lu_length))
2909 file_lock.fl_end = ~(u64)0;
2911 file_lock.fl_end = locku->lu_offset + locku->lu_length - 1;
2912 nfs4_transform_lock_offset(&file_lock);
2915 * Try to unlock the file in the VFS.
2917 err = vfs_lock_file(filp, F_SETLK, &file_lock, NULL);
2919 dprintk("NFSD: nfs4_locku: vfs_lock_file failed!\n");
2923 * OK, unlock succeeded; the only thing left to do is update the stateid.
2925 update_stateid(&stp->st_stateid);
2926 memcpy(&locku->lu_stateid, &stp->st_stateid, sizeof(stateid_t));
2929 if (locku->lu_stateowner) {
2930 nfs4_get_stateowner(locku->lu_stateowner);
2931 cstate->replay_owner = locku->lu_stateowner;
2933 nfs4_unlock_state();
2937 status = nfserrno(err);
2943 * 1: locks held by lockowner
2944 * 0: no locks held by lockowner
2947 check_for_locks(struct file *filp, struct nfs4_stateowner *lowner)
2949 struct file_lock **flpp;
2950 struct inode *inode = filp->f_path.dentry->d_inode;
2954 for (flpp = &inode->i_flock; *flpp != NULL; flpp = &(*flpp)->fl_next) {
2955 if ((*flpp)->fl_owner == (fl_owner_t)lowner) {
2966 nfsd4_release_lockowner(struct svc_rqst *rqstp,
2967 struct nfsd4_compound_state *cstate,
2968 struct nfsd4_release_lockowner *rlockowner)
2970 clientid_t *clid = &rlockowner->rl_clientid;
2971 struct nfs4_stateowner *sop;
2972 struct nfs4_stateid *stp;
2973 struct xdr_netobj *owner = &rlockowner->rl_owner;
2974 struct list_head matches;
2978 dprintk("nfsd4_release_lockowner clientid: (%08x/%08x):\n",
2979 clid->cl_boot, clid->cl_id);
2981 /* XXX check for lease expiration */
2983 status = nfserr_stale_clientid;
2984 if (STALE_CLIENTID(clid))
2989 status = nfserr_locks_held;
2990 /* XXX: we're doing a linear search through all the lockowners.
2991 * Yipes! For now we'll just hope clients aren't really using
2992 * release_lockowner much, but eventually we have to fix these
2993 * data structures. */
2994 INIT_LIST_HEAD(&matches);
2995 for (i = 0; i < LOCK_HASH_SIZE; i++) {
2996 list_for_each_entry(sop, &lock_ownerid_hashtbl[i], so_idhash) {
2997 if (!same_owner_str(sop, owner, clid))
2999 list_for_each_entry(stp, &sop->so_stateids,
3001 if (check_for_locks(stp->st_vfs_file, sop))
3003 /* Note: so_perclient unused for lockowners,
3004 * so it's OK to fool with here. */
3005 list_add(&sop->so_perclient, &matches);
3009 /* Clients probably won't expect us to return with some (but not all)
3010 * of the lockowner state released; so don't release any until all
3011 * have been checked. */
3013 while (!list_empty(&matches)) {
3014 sop = list_entry(matches.next, struct nfs4_stateowner,
3016 /* unhash_stateowner deletes so_perclient only
3017 * for openowners. */
3018 list_del(&sop->so_perclient);
3019 release_stateowner(sop);
3022 nfs4_unlock_state();
3026 static inline struct nfs4_client_reclaim *
3029 return kmalloc(sizeof(struct nfs4_client_reclaim), GFP_KERNEL);
3033 nfs4_has_reclaimed_state(const char *name)
3035 unsigned int strhashval = clientstr_hashval(name);
3036 struct nfs4_client *clp;
3038 clp = find_confirmed_client_by_str(name, strhashval);
3043 * failure => all reset bets are off, nfserr_no_grace...
3046 nfs4_client_to_reclaim(const char *name)
3048 unsigned int strhashval;
3049 struct nfs4_client_reclaim *crp = NULL;
3051 dprintk("NFSD nfs4_client_to_reclaim NAME: %.*s\n", HEXDIR_LEN, name);
3052 crp = alloc_reclaim();
3055 strhashval = clientstr_hashval(name);
3056 INIT_LIST_HEAD(&crp->cr_strhash);
3057 list_add(&crp->cr_strhash, &reclaim_str_hashtbl[strhashval]);
3058 memcpy(crp->cr_recdir, name, HEXDIR_LEN);
3059 reclaim_str_hashtbl_size++;
3064 nfs4_release_reclaim(void)
3066 struct nfs4_client_reclaim *crp = NULL;
3069 for (i = 0; i < CLIENT_HASH_SIZE; i++) {
3070 while (!list_empty(&reclaim_str_hashtbl[i])) {
3071 crp = list_entry(reclaim_str_hashtbl[i].next,
3072 struct nfs4_client_reclaim, cr_strhash);
3073 list_del(&crp->cr_strhash);
3075 reclaim_str_hashtbl_size--;
3078 BUG_ON(reclaim_str_hashtbl_size);
3082 * called from OPEN, CLAIM_PREVIOUS with a new clientid. */
3083 static struct nfs4_client_reclaim *
3084 nfs4_find_reclaim_client(clientid_t *clid)
3086 unsigned int strhashval;
3087 struct nfs4_client *clp;
3088 struct nfs4_client_reclaim *crp = NULL;
3091 /* find clientid in conf_id_hashtbl */
3092 clp = find_confirmed_client(clid);
3096 dprintk("NFSD: nfs4_find_reclaim_client for %.*s with recdir %s\n",
3097 clp->cl_name.len, clp->cl_name.data,
3100 /* find clp->cl_name in reclaim_str_hashtbl */
3101 strhashval = clientstr_hashval(clp->cl_recdir);
3102 list_for_each_entry(crp, &reclaim_str_hashtbl[strhashval], cr_strhash) {
3103 if (same_name(crp->cr_recdir, clp->cl_recdir)) {
3111 * Called from OPEN. Look for clientid in reclaim list.
3114 nfs4_check_open_reclaim(clientid_t *clid)
3116 return nfs4_find_reclaim_client(clid) ? nfs_ok : nfserr_reclaim_bad;
3119 /* initialization to perform at module load time: */
3122 nfs4_state_init(void)
3126 status = nfsd4_init_slabs();
3129 for (i = 0; i < CLIENT_HASH_SIZE; i++) {
3130 INIT_LIST_HEAD(&conf_id_hashtbl[i]);
3131 INIT_LIST_HEAD(&conf_str_hashtbl[i]);
3132 INIT_LIST_HEAD(&unconf_str_hashtbl[i]);
3133 INIT_LIST_HEAD(&unconf_id_hashtbl[i]);
3135 for (i = 0; i < FILE_HASH_SIZE; i++) {
3136 INIT_LIST_HEAD(&file_hashtbl[i]);
3138 for (i = 0; i < OWNER_HASH_SIZE; i++) {
3139 INIT_LIST_HEAD(&ownerstr_hashtbl[i]);
3140 INIT_LIST_HEAD(&ownerid_hashtbl[i]);
3142 for (i = 0; i < STATEID_HASH_SIZE; i++) {
3143 INIT_LIST_HEAD(&stateid_hashtbl[i]);
3144 INIT_LIST_HEAD(&lockstateid_hashtbl[i]);
3146 for (i = 0; i < LOCK_HASH_SIZE; i++) {
3147 INIT_LIST_HEAD(&lock_ownerid_hashtbl[i]);
3148 INIT_LIST_HEAD(&lock_ownerstr_hashtbl[i]);
3150 memset(&onestateid, ~0, sizeof(stateid_t));
3151 INIT_LIST_HEAD(&close_lru);
3152 INIT_LIST_HEAD(&client_lru);
3153 INIT_LIST_HEAD(&del_recall_lru);
3154 for (i = 0; i < CLIENT_HASH_SIZE; i++)
3155 INIT_LIST_HEAD(&reclaim_str_hashtbl[i]);
3156 reclaim_str_hashtbl_size = 0;
3161 nfsd4_load_reboot_recovery_data(void)
3166 nfsd4_init_recdir(user_recovery_dirname);
3167 status = nfsd4_recdir_load();
3168 nfs4_unlock_state();
3170 printk("NFSD: Failure reading reboot recovery data\n");
3174 get_nfs4_grace_period(void)
3176 return max(user_lease_time, lease_time) * HZ;
3180 * Since the lifetime of a delegation isn't limited to that of an open, a
3181 * client may quite reasonably hang on to a delegation as long as it has
3182 * the inode cached. This becomes an obvious problem the first time a
3183 * client's inode cache approaches the size of the server's total memory.
3185 * For now we avoid this problem by imposing a hard limit on the number
3186 * of delegations, which varies according to the server's memory size.
3189 set_max_delegations(void)
3192 * Allow at most 4 delegations per megabyte of RAM. Quick
3193 * estimates suggest that in the worst case (where every delegation
3194 * is for a different inode), a delegation could take about 1.5K,
3195 * giving a worst case usage of about 6% of memory.
3197 max_delegations = nr_free_buffer_pages() >> (20 - 2 - PAGE_SHIFT);
3200 /* initialization to perform when the nfsd service is started: */
3203 __nfs4_state_start(void)
3205 unsigned long grace_time;
3207 boot_time = get_seconds();
3208 grace_time = get_nfs4_grace_period();
3209 lease_time = user_lease_time;
3210 locks_start_grace(&nfsd4_manager);
3211 printk(KERN_INFO "NFSD: starting %ld-second grace period\n",
3213 laundry_wq = create_singlethread_workqueue("nfsd4");
3214 queue_delayed_work(laundry_wq, &laundromat_work, grace_time);
3215 set_max_delegations();
3219 nfs4_state_start(void)
3223 nfsd4_load_reboot_recovery_data();
3224 __nfs4_state_start();
3230 nfs4_lease_time(void)
3236 __nfs4_state_shutdown(void)
3239 struct nfs4_client *clp = NULL;
3240 struct nfs4_delegation *dp = NULL;
3241 struct list_head *pos, *next, reaplist;
3243 for (i = 0; i < CLIENT_HASH_SIZE; i++) {
3244 while (!list_empty(&conf_id_hashtbl[i])) {
3245 clp = list_entry(conf_id_hashtbl[i].next, struct nfs4_client, cl_idhash);
3248 while (!list_empty(&unconf_str_hashtbl[i])) {
3249 clp = list_entry(unconf_str_hashtbl[i].next, struct nfs4_client, cl_strhash);
3253 INIT_LIST_HEAD(&reaplist);
3254 spin_lock(&recall_lock);
3255 list_for_each_safe(pos, next, &del_recall_lru) {
3256 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
3257 list_move(&dp->dl_recall_lru, &reaplist);
3259 spin_unlock(&recall_lock);
3260 list_for_each_safe(pos, next, &reaplist) {
3261 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
3262 list_del_init(&dp->dl_recall_lru);
3263 unhash_delegation(dp);
3266 nfsd4_shutdown_recdir();
3271 nfs4_state_shutdown(void)
3273 cancel_rearming_delayed_workqueue(laundry_wq, &laundromat_work);
3274 destroy_workqueue(laundry_wq);
3275 locks_end_grace(&nfsd4_manager);
3277 nfs4_release_reclaim();
3278 __nfs4_state_shutdown();
3279 nfs4_unlock_state();
3283 * user_recovery_dirname is protected by the nfsd_mutex since it's only
3284 * accessed when nfsd is starting.
3287 nfs4_set_recdir(char *recdir)
3289 strcpy(user_recovery_dirname, recdir);
3293 * Change the NFSv4 recovery directory to recdir.
3296 nfs4_reset_recoverydir(char *recdir)
3301 status = kern_path(recdir, LOOKUP_FOLLOW, &path);
3305 if (S_ISDIR(path.dentry->d_inode->i_mode)) {
3306 nfs4_set_recdir(recdir);
3314 nfs4_recoverydir(void)
3316 return user_recovery_dirname;
3320 * Called when leasetime is changed.
3322 * The only way the protocol gives us to handle on-the-fly lease changes is to
3323 * simulate a reboot. Instead of doing that, we just wait till the next time
3324 * we start to register any changes in lease time. If the administrator
3325 * really wants to change the lease time *now*, they can go ahead and bring
3326 * nfsd down and then back up again after changing the lease time.
3328 * user_lease_time is protected by nfsd_mutex since it's only really accessed
3329 * when nfsd is starting
3332 nfs4_reset_lease(time_t leasetime)
3334 user_lease_time = leasetime;