Root/fs/nfsd/nfs4state.c

1/*
2* Copyright (c) 2001 The Regents of the University of Michigan.
3* All rights reserved.
4*
5* Kendrick Smith <kmsmith@umich.edu>
6* Andy Adamson <kandros@umich.edu>
7*
8* Redistribution and use in source and binary forms, with or without
9* modification, are permitted provided that the following conditions
10* are met:
11*
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.
20*
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.
32*
33*/
34
35#include <linux/file.h>
36#include <linux/fs.h>
37#include <linux/slab.h>
38#include <linux/namei.h>
39#include <linux/swap.h>
40#include <linux/sunrpc/svcauth_gss.h>
41#include <linux/sunrpc/clnt.h>
42#include "xdr4.h"
43#include "vfs.h"
44
45#define NFSDDBG_FACILITY NFSDDBG_PROC
46
47/* Globals */
48time_t nfsd4_lease = 90; /* default lease time */
49time_t nfsd4_grace = 90;
50static time_t boot_time;
51static u32 current_ownerid = 1;
52static u32 current_fileid = 1;
53static u32 current_delegid = 1;
54static stateid_t zerostateid; /* bits all 0 */
55static stateid_t onestateid; /* bits all 1 */
56static u64 current_sessionid = 1;
57
58#define ZERO_STATEID(stateid) (!memcmp((stateid), &zerostateid, sizeof(stateid_t)))
59#define ONE_STATEID(stateid) (!memcmp((stateid), &onestateid, sizeof(stateid_t)))
60
61/* forward declarations */
62static struct nfs4_stateid * find_stateid(stateid_t *stid, int flags);
63static struct nfs4_delegation * find_delegation_stateid(struct inode *ino, stateid_t *stid);
64static char user_recovery_dirname[PATH_MAX] = "/var/lib/nfs/v4recovery";
65static void nfs4_set_recdir(char *recdir);
66
67/* Locking: */
68
69/* Currently used for almost all code touching nfsv4 state: */
70static DEFINE_MUTEX(client_mutex);
71
72/*
73 * Currently used for the del_recall_lru and file hash table. In an
74 * effort to decrease the scope of the client_mutex, this spinlock may
75 * eventually cover more:
76 */
77static DEFINE_SPINLOCK(recall_lock);
78
79static struct kmem_cache *stateowner_slab = NULL;
80static struct kmem_cache *file_slab = NULL;
81static struct kmem_cache *stateid_slab = NULL;
82static struct kmem_cache *deleg_slab = NULL;
83
84void
85nfs4_lock_state(void)
86{
87    mutex_lock(&client_mutex);
88}
89
90void
91nfs4_unlock_state(void)
92{
93    mutex_unlock(&client_mutex);
94}
95
96static inline u32
97opaque_hashval(const void *ptr, int nbytes)
98{
99    unsigned char *cptr = (unsigned char *) ptr;
100
101    u32 x = 0;
102    while (nbytes--) {
103        x *= 37;
104        x += *cptr++;
105    }
106    return x;
107}
108
109static struct list_head del_recall_lru;
110
111static inline void
112put_nfs4_file(struct nfs4_file *fi)
113{
114    if (atomic_dec_and_lock(&fi->fi_ref, &recall_lock)) {
115        list_del(&fi->fi_hash);
116        spin_unlock(&recall_lock);
117        iput(fi->fi_inode);
118        kmem_cache_free(file_slab, fi);
119    }
120}
121
122static inline void
123get_nfs4_file(struct nfs4_file *fi)
124{
125    atomic_inc(&fi->fi_ref);
126}
127
128static int num_delegations;
129unsigned int max_delegations;
130
131/*
132 * Open owner state (share locks)
133 */
134
135/* hash tables for nfs4_stateowner */
136#define OWNER_HASH_BITS 8
137#define OWNER_HASH_SIZE (1 << OWNER_HASH_BITS)
138#define OWNER_HASH_MASK (OWNER_HASH_SIZE - 1)
139
140#define ownerid_hashval(id) \
141        ((id) & OWNER_HASH_MASK)
142#define ownerstr_hashval(clientid, ownername) \
143        (((clientid) + opaque_hashval((ownername.data), (ownername.len))) & OWNER_HASH_MASK)
144
145static struct list_head ownerid_hashtbl[OWNER_HASH_SIZE];
146static struct list_head ownerstr_hashtbl[OWNER_HASH_SIZE];
147
148/* hash table for nfs4_file */
149#define FILE_HASH_BITS 8
150#define FILE_HASH_SIZE (1 << FILE_HASH_BITS)
151
152/* hash table for (open)nfs4_stateid */
153#define STATEID_HASH_BITS 10
154#define STATEID_HASH_SIZE (1 << STATEID_HASH_BITS)
155#define STATEID_HASH_MASK (STATEID_HASH_SIZE - 1)
156
157#define file_hashval(x) \
158        hash_ptr(x, FILE_HASH_BITS)
159#define stateid_hashval(owner_id, file_id) \
160        (((owner_id) + (file_id)) & STATEID_HASH_MASK)
161
162static struct list_head file_hashtbl[FILE_HASH_SIZE];
163static struct list_head stateid_hashtbl[STATEID_HASH_SIZE];
164
165static void __nfs4_file_get_access(struct nfs4_file *fp, int oflag)
166{
167    BUG_ON(!(fp->fi_fds[oflag] || fp->fi_fds[O_RDWR]));
168    atomic_inc(&fp->fi_access[oflag]);
169}
170
171static void nfs4_file_get_access(struct nfs4_file *fp, int oflag)
172{
173    if (oflag == O_RDWR) {
174        __nfs4_file_get_access(fp, O_RDONLY);
175        __nfs4_file_get_access(fp, O_WRONLY);
176    } else
177        __nfs4_file_get_access(fp, oflag);
178}
179
180static void nfs4_file_put_fd(struct nfs4_file *fp, int oflag)
181{
182    if (fp->fi_fds[oflag]) {
183        fput(fp->fi_fds[oflag]);
184        fp->fi_fds[oflag] = NULL;
185    }
186}
187
188static void __nfs4_file_put_access(struct nfs4_file *fp, int oflag)
189{
190    if (atomic_dec_and_test(&fp->fi_access[oflag])) {
191        nfs4_file_put_fd(fp, O_RDWR);
192        nfs4_file_put_fd(fp, oflag);
193    }
194}
195
196static void nfs4_file_put_access(struct nfs4_file *fp, int oflag)
197{
198    if (oflag == O_RDWR) {
199        __nfs4_file_put_access(fp, O_RDONLY);
200        __nfs4_file_put_access(fp, O_WRONLY);
201    } else
202        __nfs4_file_put_access(fp, oflag);
203}
204
205static struct nfs4_delegation *
206alloc_init_deleg(struct nfs4_client *clp, struct nfs4_stateid *stp, struct svc_fh *current_fh, u32 type)
207{
208    struct nfs4_delegation *dp;
209    struct nfs4_file *fp = stp->st_file;
210
211    dprintk("NFSD alloc_init_deleg\n");
212    /*
213     * Major work on the lease subsystem (for example, to support
214     * calbacks on stat) will be required before we can support
215     * write delegations properly.
216     */
217    if (type != NFS4_OPEN_DELEGATE_READ)
218        return NULL;
219    if (fp->fi_had_conflict)
220        return NULL;
221    if (num_delegations > max_delegations)
222        return NULL;
223    dp = kmem_cache_alloc(deleg_slab, GFP_KERNEL);
224    if (dp == NULL)
225        return dp;
226    num_delegations++;
227    INIT_LIST_HEAD(&dp->dl_perfile);
228    INIT_LIST_HEAD(&dp->dl_perclnt);
229    INIT_LIST_HEAD(&dp->dl_recall_lru);
230    dp->dl_client = clp;
231    get_nfs4_file(fp);
232    dp->dl_file = fp;
233    dp->dl_type = type;
234    dp->dl_stateid.si_boot = boot_time;
235    dp->dl_stateid.si_stateownerid = current_delegid++;
236    dp->dl_stateid.si_fileid = 0;
237    dp->dl_stateid.si_generation = 0;
238    fh_copy_shallow(&dp->dl_fh, &current_fh->fh_handle);
239    dp->dl_time = 0;
240    atomic_set(&dp->dl_count, 1);
241    INIT_WORK(&dp->dl_recall.cb_work, nfsd4_do_callback_rpc);
242    return dp;
243}
244
245void
246nfs4_put_delegation(struct nfs4_delegation *dp)
247{
248    if (atomic_dec_and_test(&dp->dl_count)) {
249        dprintk("NFSD: freeing dp %p\n",dp);
250        put_nfs4_file(dp->dl_file);
251        kmem_cache_free(deleg_slab, dp);
252        num_delegations--;
253    }
254}
255
256static void nfs4_put_deleg_lease(struct nfs4_file *fp)
257{
258    if (atomic_dec_and_test(&fp->fi_delegees)) {
259        vfs_setlease(fp->fi_deleg_file, F_UNLCK, &fp->fi_lease);
260        fp->fi_lease = NULL;
261        fput(fp->fi_deleg_file);
262        fp->fi_deleg_file = NULL;
263    }
264}
265
266/* Called under the state lock. */
267static void
268unhash_delegation(struct nfs4_delegation *dp)
269{
270    list_del_init(&dp->dl_perclnt);
271    spin_lock(&recall_lock);
272    list_del_init(&dp->dl_perfile);
273    list_del_init(&dp->dl_recall_lru);
274    spin_unlock(&recall_lock);
275    nfs4_put_deleg_lease(dp->dl_file);
276    nfs4_put_delegation(dp);
277}
278
279/*
280 * SETCLIENTID state
281 */
282
283/* client_lock protects the client lru list and session hash table */
284static DEFINE_SPINLOCK(client_lock);
285
286/* Hash tables for nfs4_clientid state */
287#define CLIENT_HASH_BITS 4
288#define CLIENT_HASH_SIZE (1 << CLIENT_HASH_BITS)
289#define CLIENT_HASH_MASK (CLIENT_HASH_SIZE - 1)
290
291#define clientid_hashval(id) \
292    ((id) & CLIENT_HASH_MASK)
293#define clientstr_hashval(name) \
294    (opaque_hashval((name), 8) & CLIENT_HASH_MASK)
295/*
296 * reclaim_str_hashtbl[] holds known client info from previous reset/reboot
297 * used in reboot/reset lease grace period processing
298 *
299 * conf_id_hashtbl[], and conf_str_hashtbl[] hold confirmed
300 * setclientid_confirmed info.
301 *
302 * unconf_str_hastbl[] and unconf_id_hashtbl[] hold unconfirmed
303 * setclientid info.
304 *
305 * client_lru holds client queue ordered by nfs4_client.cl_time
306 * for lease renewal.
307 *
308 * close_lru holds (open) stateowner queue ordered by nfs4_stateowner.so_time
309 * for last close replay.
310 */
311static struct list_head reclaim_str_hashtbl[CLIENT_HASH_SIZE];
312static int reclaim_str_hashtbl_size = 0;
313static struct list_head conf_id_hashtbl[CLIENT_HASH_SIZE];
314static struct list_head conf_str_hashtbl[CLIENT_HASH_SIZE];
315static struct list_head unconf_str_hashtbl[CLIENT_HASH_SIZE];
316static struct list_head unconf_id_hashtbl[CLIENT_HASH_SIZE];
317static struct list_head client_lru;
318static struct list_head close_lru;
319
320/*
321 * We store the NONE, READ, WRITE, and BOTH bits separately in the
322 * st_{access,deny}_bmap field of the stateid, in order to track not
323 * only what share bits are currently in force, but also what
324 * combinations of share bits previous opens have used. This allows us
325 * to enforce the recommendation of rfc 3530 14.2.19 that the server
326 * return an error if the client attempt to downgrade to a combination
327 * of share bits not explicable by closing some of its previous opens.
328 *
329 * XXX: This enforcement is actually incomplete, since we don't keep
330 * track of access/deny bit combinations; so, e.g., we allow:
331 *
332 * OPEN allow read, deny write
333 * OPEN allow both, deny none
334 * DOWNGRADE allow read, deny none
335 *
336 * which we should reject.
337 */
338static void
339set_access(unsigned int *access, unsigned long bmap) {
340    int i;
341
342    *access = 0;
343    for (i = 1; i < 4; i++) {
344        if (test_bit(i, &bmap))
345            *access |= i;
346    }
347}
348
349static void
350set_deny(unsigned int *deny, unsigned long bmap) {
351    int i;
352
353    *deny = 0;
354    for (i = 0; i < 4; i++) {
355        if (test_bit(i, &bmap))
356            *deny |= i ;
357    }
358}
359
360static int
361test_share(struct nfs4_stateid *stp, struct nfsd4_open *open) {
362    unsigned int access, deny;
363
364    set_access(&access, stp->st_access_bmap);
365    set_deny(&deny, stp->st_deny_bmap);
366    if ((access & open->op_share_deny) || (deny & open->op_share_access))
367        return 0;
368    return 1;
369}
370
371static int nfs4_access_to_omode(u32 access)
372{
373    switch (access & NFS4_SHARE_ACCESS_BOTH) {
374    case NFS4_SHARE_ACCESS_READ:
375        return O_RDONLY;
376    case NFS4_SHARE_ACCESS_WRITE:
377        return O_WRONLY;
378    case NFS4_SHARE_ACCESS_BOTH:
379        return O_RDWR;
380    }
381    BUG();
382}
383
384static int nfs4_access_bmap_to_omode(struct nfs4_stateid *stp)
385{
386    unsigned int access;
387
388    set_access(&access, stp->st_access_bmap);
389    return nfs4_access_to_omode(access);
390}
391
392static void unhash_generic_stateid(struct nfs4_stateid *stp)
393{
394    list_del(&stp->st_hash);
395    list_del(&stp->st_perfile);
396    list_del(&stp->st_perstateowner);
397}
398
399static void free_generic_stateid(struct nfs4_stateid *stp)
400{
401    int oflag;
402
403    if (stp->st_access_bmap) {
404        oflag = nfs4_access_bmap_to_omode(stp);
405        nfs4_file_put_access(stp->st_file, oflag);
406    }
407    put_nfs4_file(stp->st_file);
408    kmem_cache_free(stateid_slab, stp);
409}
410
411static void release_lock_stateid(struct nfs4_stateid *stp)
412{
413    struct file *file;
414
415    unhash_generic_stateid(stp);
416    file = find_any_file(stp->st_file);
417    if (file)
418        locks_remove_posix(file, (fl_owner_t)stp->st_stateowner);
419    free_generic_stateid(stp);
420}
421
422static void unhash_lockowner(struct nfs4_stateowner *sop)
423{
424    struct nfs4_stateid *stp;
425
426    list_del(&sop->so_idhash);
427    list_del(&sop->so_strhash);
428    list_del(&sop->so_perstateid);
429    while (!list_empty(&sop->so_stateids)) {
430        stp = list_first_entry(&sop->so_stateids,
431                struct nfs4_stateid, st_perstateowner);
432        release_lock_stateid(stp);
433    }
434}
435
436static void release_lockowner(struct nfs4_stateowner *sop)
437{
438    unhash_lockowner(sop);
439    nfs4_put_stateowner(sop);
440}
441
442static void
443release_stateid_lockowners(struct nfs4_stateid *open_stp)
444{
445    struct nfs4_stateowner *lock_sop;
446
447    while (!list_empty(&open_stp->st_lockowners)) {
448        lock_sop = list_entry(open_stp->st_lockowners.next,
449                struct nfs4_stateowner, so_perstateid);
450        /* list_del(&open_stp->st_lockowners); */
451        BUG_ON(lock_sop->so_is_open_owner);
452        release_lockowner(lock_sop);
453    }
454}
455
456static void release_open_stateid(struct nfs4_stateid *stp)
457{
458    unhash_generic_stateid(stp);
459    release_stateid_lockowners(stp);
460    free_generic_stateid(stp);
461}
462
463static void unhash_openowner(struct nfs4_stateowner *sop)
464{
465    struct nfs4_stateid *stp;
466
467    list_del(&sop->so_idhash);
468    list_del(&sop->so_strhash);
469    list_del(&sop->so_perclient);
470    list_del(&sop->so_perstateid); /* XXX: necessary? */
471    while (!list_empty(&sop->so_stateids)) {
472        stp = list_first_entry(&sop->so_stateids,
473                struct nfs4_stateid, st_perstateowner);
474        release_open_stateid(stp);
475    }
476}
477
478static void release_openowner(struct nfs4_stateowner *sop)
479{
480    unhash_openowner(sop);
481    list_del(&sop->so_close_lru);
482    nfs4_put_stateowner(sop);
483}
484
485#define SESSION_HASH_SIZE 512
486static struct list_head sessionid_hashtbl[SESSION_HASH_SIZE];
487
488static inline int
489hash_sessionid(struct nfs4_sessionid *sessionid)
490{
491    struct nfsd4_sessionid *sid = (struct nfsd4_sessionid *)sessionid;
492
493    return sid->sequence % SESSION_HASH_SIZE;
494}
495
496static inline void
497dump_sessionid(const char *fn, struct nfs4_sessionid *sessionid)
498{
499    u32 *ptr = (u32 *)(&sessionid->data[0]);
500    dprintk("%s: %u:%u:%u:%u\n", fn, ptr[0], ptr[1], ptr[2], ptr[3]);
501}
502
503static void
504gen_sessionid(struct nfsd4_session *ses)
505{
506    struct nfs4_client *clp = ses->se_client;
507    struct nfsd4_sessionid *sid;
508
509    sid = (struct nfsd4_sessionid *)ses->se_sessionid.data;
510    sid->clientid = clp->cl_clientid;
511    sid->sequence = current_sessionid++;
512    sid->reserved = 0;
513}
514
515/*
516 * The protocol defines ca_maxresponssize_cached to include the size of
517 * the rpc header, but all we need to cache is the data starting after
518 * the end of the initial SEQUENCE operation--the rest we regenerate
519 * each time. Therefore we can advertise a ca_maxresponssize_cached
520 * value that is the number of bytes in our cache plus a few additional
521 * bytes. In order to stay on the safe side, and not promise more than
522 * we can cache, those additional bytes must be the minimum possible: 24
523 * bytes of rpc header (xid through accept state, with AUTH_NULL
524 * verifier), 12 for the compound header (with zero-length tag), and 44
525 * for the SEQUENCE op response:
526 */
527#define NFSD_MIN_HDR_SEQ_SZ (24 + 12 + 44)
528
529static void
530free_session_slots(struct nfsd4_session *ses)
531{
532    int i;
533
534    for (i = 0; i < ses->se_fchannel.maxreqs; i++)
535        kfree(ses->se_slots[i]);
536}
537
538/*
539 * We don't actually need to cache the rpc and session headers, so we
540 * can allocate a little less for each slot:
541 */
542static inline int slot_bytes(struct nfsd4_channel_attrs *ca)
543{
544    return ca->maxresp_cached - NFSD_MIN_HDR_SEQ_SZ;
545}
546
547static int nfsd4_sanitize_slot_size(u32 size)
548{
549    size -= NFSD_MIN_HDR_SEQ_SZ; /* We don't cache the rpc header */
550    size = min_t(u32, size, NFSD_SLOT_CACHE_SIZE);
551
552    return size;
553}
554
555/*
556 * XXX: If we run out of reserved DRC memory we could (up to a point)
557 * re-negotiate active sessions and reduce their slot usage to make
558 * rooom for new connections. For now we just fail the create session.
559 */
560static int nfsd4_get_drc_mem(int slotsize, u32 num)
561{
562    int avail;
563
564    num = min_t(u32, num, NFSD_MAX_SLOTS_PER_SESSION);
565
566    spin_lock(&nfsd_drc_lock);
567    avail = min_t(int, NFSD_MAX_MEM_PER_SESSION,
568            nfsd_drc_max_mem - nfsd_drc_mem_used);
569    num = min_t(int, num, avail / slotsize);
570    nfsd_drc_mem_used += num * slotsize;
571    spin_unlock(&nfsd_drc_lock);
572
573    return num;
574}
575
576static void nfsd4_put_drc_mem(int slotsize, int num)
577{
578    spin_lock(&nfsd_drc_lock);
579    nfsd_drc_mem_used -= slotsize * num;
580    spin_unlock(&nfsd_drc_lock);
581}
582
583static struct nfsd4_session *alloc_session(int slotsize, int numslots)
584{
585    struct nfsd4_session *new;
586    int mem, i;
587
588    BUILD_BUG_ON(NFSD_MAX_SLOTS_PER_SESSION * sizeof(struct nfsd4_slot *)
589            + sizeof(struct nfsd4_session) > PAGE_SIZE);
590    mem = numslots * sizeof(struct nfsd4_slot *);
591
592    new = kzalloc(sizeof(*new) + mem, GFP_KERNEL);
593    if (!new)
594        return NULL;
595    /* allocate each struct nfsd4_slot and data cache in one piece */
596    for (i = 0; i < numslots; i++) {
597        mem = sizeof(struct nfsd4_slot) + slotsize;
598        new->se_slots[i] = kzalloc(mem, GFP_KERNEL);
599        if (!new->se_slots[i])
600            goto out_free;
601    }
602    return new;
603out_free:
604    while (i--)
605        kfree(new->se_slots[i]);
606    kfree(new);
607    return NULL;
608}
609
610static void init_forechannel_attrs(struct nfsd4_channel_attrs *new, struct nfsd4_channel_attrs *req, int numslots, int slotsize)
611{
612    u32 maxrpc = nfsd_serv->sv_max_mesg;
613
614    new->maxreqs = numslots;
615    new->maxresp_cached = min_t(u32, req->maxresp_cached,
616                    slotsize + NFSD_MIN_HDR_SEQ_SZ);
617    new->maxreq_sz = min_t(u32, req->maxreq_sz, maxrpc);
618    new->maxresp_sz = min_t(u32, req->maxresp_sz, maxrpc);
619    new->maxops = min_t(u32, req->maxops, NFSD_MAX_OPS_PER_COMPOUND);
620}
621
622static void free_conn(struct nfsd4_conn *c)
623{
624    svc_xprt_put(c->cn_xprt);
625    kfree(c);
626}
627
628static void nfsd4_conn_lost(struct svc_xpt_user *u)
629{
630    struct nfsd4_conn *c = container_of(u, struct nfsd4_conn, cn_xpt_user);
631    struct nfs4_client *clp = c->cn_session->se_client;
632
633    spin_lock(&clp->cl_lock);
634    if (!list_empty(&c->cn_persession)) {
635        list_del(&c->cn_persession);
636        free_conn(c);
637    }
638    spin_unlock(&clp->cl_lock);
639    nfsd4_probe_callback(clp);
640}
641
642static struct nfsd4_conn *alloc_conn(struct svc_rqst *rqstp, u32 flags)
643{
644    struct nfsd4_conn *conn;
645
646    conn = kmalloc(sizeof(struct nfsd4_conn), GFP_KERNEL);
647    if (!conn)
648        return NULL;
649    svc_xprt_get(rqstp->rq_xprt);
650    conn->cn_xprt = rqstp->rq_xprt;
651    conn->cn_flags = flags;
652    INIT_LIST_HEAD(&conn->cn_xpt_user.list);
653    return conn;
654}
655
656static void __nfsd4_hash_conn(struct nfsd4_conn *conn, struct nfsd4_session *ses)
657{
658    conn->cn_session = ses;
659    list_add(&conn->cn_persession, &ses->se_conns);
660}
661
662static void nfsd4_hash_conn(struct nfsd4_conn *conn, struct nfsd4_session *ses)
663{
664    struct nfs4_client *clp = ses->se_client;
665
666    spin_lock(&clp->cl_lock);
667    __nfsd4_hash_conn(conn, ses);
668    spin_unlock(&clp->cl_lock);
669}
670
671static int nfsd4_register_conn(struct nfsd4_conn *conn)
672{
673    conn->cn_xpt_user.callback = nfsd4_conn_lost;
674    return register_xpt_user(conn->cn_xprt, &conn->cn_xpt_user);
675}
676
677static __be32 nfsd4_new_conn(struct svc_rqst *rqstp, struct nfsd4_session *ses, u32 dir)
678{
679    struct nfsd4_conn *conn;
680    int ret;
681
682    conn = alloc_conn(rqstp, dir);
683    if (!conn)
684        return nfserr_jukebox;
685    nfsd4_hash_conn(conn, ses);
686    ret = nfsd4_register_conn(conn);
687    if (ret)
688        /* oops; xprt is already down: */
689        nfsd4_conn_lost(&conn->cn_xpt_user);
690    return nfs_ok;
691}
692
693static __be32 nfsd4_new_conn_from_crses(struct svc_rqst *rqstp, struct nfsd4_session *ses)
694{
695    u32 dir = NFS4_CDFC4_FORE;
696
697    if (ses->se_flags & SESSION4_BACK_CHAN)
698        dir |= NFS4_CDFC4_BACK;
699
700    return nfsd4_new_conn(rqstp, ses, dir);
701}
702
703/* must be called under client_lock */
704static void nfsd4_del_conns(struct nfsd4_session *s)
705{
706    struct nfs4_client *clp = s->se_client;
707    struct nfsd4_conn *c;
708
709    spin_lock(&clp->cl_lock);
710    while (!list_empty(&s->se_conns)) {
711        c = list_first_entry(&s->se_conns, struct nfsd4_conn, cn_persession);
712        list_del_init(&c->cn_persession);
713        spin_unlock(&clp->cl_lock);
714
715        unregister_xpt_user(c->cn_xprt, &c->cn_xpt_user);
716        free_conn(c);
717
718        spin_lock(&clp->cl_lock);
719    }
720    spin_unlock(&clp->cl_lock);
721}
722
723void free_session(struct kref *kref)
724{
725    struct nfsd4_session *ses;
726    int mem;
727
728    ses = container_of(kref, struct nfsd4_session, se_ref);
729    nfsd4_del_conns(ses);
730    spin_lock(&nfsd_drc_lock);
731    mem = ses->se_fchannel.maxreqs * slot_bytes(&ses->se_fchannel);
732    nfsd_drc_mem_used -= mem;
733    spin_unlock(&nfsd_drc_lock);
734    free_session_slots(ses);
735    kfree(ses);
736}
737
738static struct nfsd4_session *alloc_init_session(struct svc_rqst *rqstp, struct nfs4_client *clp, struct nfsd4_create_session *cses)
739{
740    struct nfsd4_session *new;
741    struct nfsd4_channel_attrs *fchan = &cses->fore_channel;
742    int numslots, slotsize;
743    int status;
744    int idx;
745
746    /*
747     * Note decreasing slot size below client's request may
748     * make it difficult for client to function correctly, whereas
749     * decreasing the number of slots will (just?) affect
750     * performance. When short on memory we therefore prefer to
751     * decrease number of slots instead of their size.
752     */
753    slotsize = nfsd4_sanitize_slot_size(fchan->maxresp_cached);
754    numslots = nfsd4_get_drc_mem(slotsize, fchan->maxreqs);
755    if (numslots < 1)
756        return NULL;
757
758    new = alloc_session(slotsize, numslots);
759    if (!new) {
760        nfsd4_put_drc_mem(slotsize, fchan->maxreqs);
761        return NULL;
762    }
763    init_forechannel_attrs(&new->se_fchannel, fchan, numslots, slotsize);
764
765    new->se_client = clp;
766    gen_sessionid(new);
767
768    INIT_LIST_HEAD(&new->se_conns);
769
770    new->se_cb_seq_nr = 1;
771    new->se_flags = cses->flags;
772    new->se_cb_prog = cses->callback_prog;
773    kref_init(&new->se_ref);
774    idx = hash_sessionid(&new->se_sessionid);
775    spin_lock(&client_lock);
776    list_add(&new->se_hash, &sessionid_hashtbl[idx]);
777    spin_lock(&clp->cl_lock);
778    list_add(&new->se_perclnt, &clp->cl_sessions);
779    spin_unlock(&clp->cl_lock);
780    spin_unlock(&client_lock);
781
782    status = nfsd4_new_conn_from_crses(rqstp, new);
783    /* whoops: benny points out, status is ignored! (err, or bogus) */
784    if (status) {
785        free_session(&new->se_ref);
786        return NULL;
787    }
788    if (cses->flags & SESSION4_BACK_CHAN) {
789        struct sockaddr *sa = svc_addr(rqstp);
790        /*
791         * This is a little silly; with sessions there's no real
792         * use for the callback address. Use the peer address
793         * as a reasonable default for now, but consider fixing
794         * the rpc client not to require an address in the
795         * future:
796         */
797        rpc_copy_addr((struct sockaddr *)&clp->cl_cb_conn.cb_addr, sa);
798        clp->cl_cb_conn.cb_addrlen = svc_addr_len(sa);
799    }
800    nfsd4_probe_callback(clp);
801    return new;
802}
803
804/* caller must hold client_lock */
805static struct nfsd4_session *
806find_in_sessionid_hashtbl(struct nfs4_sessionid *sessionid)
807{
808    struct nfsd4_session *elem;
809    int idx;
810
811    dump_sessionid(__func__, sessionid);
812    idx = hash_sessionid(sessionid);
813    /* Search in the appropriate list */
814    list_for_each_entry(elem, &sessionid_hashtbl[idx], se_hash) {
815        if (!memcmp(elem->se_sessionid.data, sessionid->data,
816                NFS4_MAX_SESSIONID_LEN)) {
817            return elem;
818        }
819    }
820
821    dprintk("%s: session not found\n", __func__);
822    return NULL;
823}
824
825/* caller must hold client_lock */
826static void
827unhash_session(struct nfsd4_session *ses)
828{
829    list_del(&ses->se_hash);
830    spin_lock(&ses->se_client->cl_lock);
831    list_del(&ses->se_perclnt);
832    spin_unlock(&ses->se_client->cl_lock);
833}
834
835/* must be called under the client_lock */
836static inline void
837renew_client_locked(struct nfs4_client *clp)
838{
839    if (is_client_expired(clp)) {
840        dprintk("%s: client (clientid %08x/%08x) already expired\n",
841            __func__,
842            clp->cl_clientid.cl_boot,
843            clp->cl_clientid.cl_id);
844        return;
845    }
846
847    /*
848    * Move client to the end to the LRU list.
849    */
850    dprintk("renewing client (clientid %08x/%08x)\n",
851            clp->cl_clientid.cl_boot,
852            clp->cl_clientid.cl_id);
853    list_move_tail(&clp->cl_lru, &client_lru);
854    clp->cl_time = get_seconds();
855}
856
857static inline void
858renew_client(struct nfs4_client *clp)
859{
860    spin_lock(&client_lock);
861    renew_client_locked(clp);
862    spin_unlock(&client_lock);
863}
864
865/* SETCLIENTID and SETCLIENTID_CONFIRM Helper functions */
866static int
867STALE_CLIENTID(clientid_t *clid)
868{
869    if (clid->cl_boot == boot_time)
870        return 0;
871    dprintk("NFSD stale clientid (%08x/%08x) boot_time %08lx\n",
872        clid->cl_boot, clid->cl_id, boot_time);
873    return 1;
874}
875
876/*
877 * XXX Should we use a slab cache ?
878 * This type of memory management is somewhat inefficient, but we use it
879 * anyway since SETCLIENTID is not a common operation.
880 */
881static struct nfs4_client *alloc_client(struct xdr_netobj name)
882{
883    struct nfs4_client *clp;
884
885    clp = kzalloc(sizeof(struct nfs4_client), GFP_KERNEL);
886    if (clp == NULL)
887        return NULL;
888    clp->cl_name.data = kmalloc(name.len, GFP_KERNEL);
889    if (clp->cl_name.data == NULL) {
890        kfree(clp);
891        return NULL;
892    }
893    memcpy(clp->cl_name.data, name.data, name.len);
894    clp->cl_name.len = name.len;
895    return clp;
896}
897
898static inline void
899free_client(struct nfs4_client *clp)
900{
901    while (!list_empty(&clp->cl_sessions)) {
902        struct nfsd4_session *ses;
903        ses = list_entry(clp->cl_sessions.next, struct nfsd4_session,
904                se_perclnt);
905        list_del(&ses->se_perclnt);
906        nfsd4_put_session(ses);
907    }
908    if (clp->cl_cred.cr_group_info)
909        put_group_info(clp->cl_cred.cr_group_info);
910    kfree(clp->cl_principal);
911    kfree(clp->cl_name.data);
912    kfree(clp);
913}
914
915void
916release_session_client(struct nfsd4_session *session)
917{
918    struct nfs4_client *clp = session->se_client;
919
920    if (!atomic_dec_and_lock(&clp->cl_refcount, &client_lock))
921        return;
922    if (is_client_expired(clp)) {
923        free_client(clp);
924        session->se_client = NULL;
925    } else
926        renew_client_locked(clp);
927    spin_unlock(&client_lock);
928}
929
930/* must be called under the client_lock */
931static inline void
932unhash_client_locked(struct nfs4_client *clp)
933{
934    struct nfsd4_session *ses;
935
936    mark_client_expired(clp);
937    list_del(&clp->cl_lru);
938    spin_lock(&clp->cl_lock);
939    list_for_each_entry(ses, &clp->cl_sessions, se_perclnt)
940        list_del_init(&ses->se_hash);
941    spin_unlock(&clp->cl_lock);
942}
943
944static void
945expire_client(struct nfs4_client *clp)
946{
947    struct nfs4_stateowner *sop;
948    struct nfs4_delegation *dp;
949    struct list_head reaplist;
950
951    INIT_LIST_HEAD(&reaplist);
952    spin_lock(&recall_lock);
953    while (!list_empty(&clp->cl_delegations)) {
954        dp = list_entry(clp->cl_delegations.next, struct nfs4_delegation, dl_perclnt);
955        list_del_init(&dp->dl_perclnt);
956        list_move(&dp->dl_recall_lru, &reaplist);
957    }
958    spin_unlock(&recall_lock);
959    while (!list_empty(&reaplist)) {
960        dp = list_entry(reaplist.next, struct nfs4_delegation, dl_recall_lru);
961        list_del_init(&dp->dl_recall_lru);
962        unhash_delegation(dp);
963    }
964    while (!list_empty(&clp->cl_openowners)) {
965        sop = list_entry(clp->cl_openowners.next, struct nfs4_stateowner, so_perclient);
966        release_openowner(sop);
967    }
968    nfsd4_shutdown_callback(clp);
969    if (clp->cl_cb_conn.cb_xprt)
970        svc_xprt_put(clp->cl_cb_conn.cb_xprt);
971    list_del(&clp->cl_idhash);
972    list_del(&clp->cl_strhash);
973    spin_lock(&client_lock);
974    unhash_client_locked(clp);
975    if (atomic_read(&clp->cl_refcount) == 0)
976        free_client(clp);
977    spin_unlock(&client_lock);
978}
979
980static void copy_verf(struct nfs4_client *target, nfs4_verifier *source)
981{
982    memcpy(target->cl_verifier.data, source->data,
983            sizeof(target->cl_verifier.data));
984}
985
986static void copy_clid(struct nfs4_client *target, struct nfs4_client *source)
987{
988    target->cl_clientid.cl_boot = source->cl_clientid.cl_boot;
989    target->cl_clientid.cl_id = source->cl_clientid.cl_id;
990}
991
992static void copy_cred(struct svc_cred *target, struct svc_cred *source)
993{
994    target->cr_uid = source->cr_uid;
995    target->cr_gid = source->cr_gid;
996    target->cr_group_info = source->cr_group_info;
997    get_group_info(target->cr_group_info);
998}
999
1000static int same_name(const char *n1, const char *n2)
1001{
1002    return 0 == memcmp(n1, n2, HEXDIR_LEN);
1003}
1004
1005static int
1006same_verf(nfs4_verifier *v1, nfs4_verifier *v2)
1007{
1008    return 0 == memcmp(v1->data, v2->data, sizeof(v1->data));
1009}
1010
1011static int
1012same_clid(clientid_t *cl1, clientid_t *cl2)
1013{
1014    return (cl1->cl_boot == cl2->cl_boot) && (cl1->cl_id == cl2->cl_id);
1015}
1016
1017/* XXX what about NGROUP */
1018static int
1019same_creds(struct svc_cred *cr1, struct svc_cred *cr2)
1020{
1021    return cr1->cr_uid == cr2->cr_uid;
1022}
1023
1024static void gen_clid(struct nfs4_client *clp)
1025{
1026    static u32 current_clientid = 1;
1027
1028    clp->cl_clientid.cl_boot = boot_time;
1029    clp->cl_clientid.cl_id = current_clientid++;
1030}
1031
1032static void gen_confirm(struct nfs4_client *clp)
1033{
1034    static u32 i;
1035    u32 *p;
1036
1037    p = (u32 *)clp->cl_confirm.data;
1038    *p++ = get_seconds();
1039    *p++ = i++;
1040}
1041
1042static struct nfs4_client *create_client(struct xdr_netobj name, char *recdir,
1043        struct svc_rqst *rqstp, nfs4_verifier *verf)
1044{
1045    struct nfs4_client *clp;
1046    struct sockaddr *sa = svc_addr(rqstp);
1047    char *princ;
1048
1049    clp = alloc_client(name);
1050    if (clp == NULL)
1051        return NULL;
1052
1053    INIT_LIST_HEAD(&clp->cl_sessions);
1054
1055    princ = svc_gss_principal(rqstp);
1056    if (princ) {
1057        clp->cl_principal = kstrdup(princ, GFP_KERNEL);
1058        if (clp->cl_principal == NULL) {
1059            free_client(clp);
1060            return NULL;
1061        }
1062    }
1063
1064    memcpy(clp->cl_recdir, recdir, HEXDIR_LEN);
1065    atomic_set(&clp->cl_refcount, 0);
1066    clp->cl_cb_state = NFSD4_CB_UNKNOWN;
1067    INIT_LIST_HEAD(&clp->cl_idhash);
1068    INIT_LIST_HEAD(&clp->cl_strhash);
1069    INIT_LIST_HEAD(&clp->cl_openowners);
1070    INIT_LIST_HEAD(&clp->cl_delegations);
1071    INIT_LIST_HEAD(&clp->cl_lru);
1072    INIT_LIST_HEAD(&clp->cl_callbacks);
1073    spin_lock_init(&clp->cl_lock);
1074    INIT_WORK(&clp->cl_cb_null.cb_work, nfsd4_do_callback_rpc);
1075    clp->cl_time = get_seconds();
1076    clear_bit(0, &clp->cl_cb_slot_busy);
1077    rpc_init_wait_queue(&clp->cl_cb_waitq, "Backchannel slot table");
1078    copy_verf(clp, verf);
1079    rpc_copy_addr((struct sockaddr *) &clp->cl_addr, sa);
1080    clp->cl_flavor = rqstp->rq_flavor;
1081    copy_cred(&clp->cl_cred, &rqstp->rq_cred);
1082    gen_confirm(clp);
1083    clp->cl_cb_session = NULL;
1084    return clp;
1085}
1086
1087static int check_name(struct xdr_netobj name)
1088{
1089    if (name.len == 0)
1090        return 0;
1091    if (name.len > NFS4_OPAQUE_LIMIT) {
1092        dprintk("NFSD: check_name: name too long(%d)!\n", name.len);
1093        return 0;
1094    }
1095    return 1;
1096}
1097
1098static void
1099add_to_unconfirmed(struct nfs4_client *clp, unsigned int strhashval)
1100{
1101    unsigned int idhashval;
1102
1103    list_add(&clp->cl_strhash, &unconf_str_hashtbl[strhashval]);
1104    idhashval = clientid_hashval(clp->cl_clientid.cl_id);
1105    list_add(&clp->cl_idhash, &unconf_id_hashtbl[idhashval]);
1106    renew_client(clp);
1107}
1108
1109static void
1110move_to_confirmed(struct nfs4_client *clp)
1111{
1112    unsigned int idhashval = clientid_hashval(clp->cl_clientid.cl_id);
1113    unsigned int strhashval;
1114
1115    dprintk("NFSD: move_to_confirm nfs4_client %p\n", clp);
1116    list_move(&clp->cl_idhash, &conf_id_hashtbl[idhashval]);
1117    strhashval = clientstr_hashval(clp->cl_recdir);
1118    list_move(&clp->cl_strhash, &conf_str_hashtbl[strhashval]);
1119    renew_client(clp);
1120}
1121
1122static struct nfs4_client *
1123find_confirmed_client(clientid_t *clid)
1124{
1125    struct nfs4_client *clp;
1126    unsigned int idhashval = clientid_hashval(clid->cl_id);
1127
1128    list_for_each_entry(clp, &conf_id_hashtbl[idhashval], cl_idhash) {
1129        if (same_clid(&clp->cl_clientid, clid))
1130            return clp;
1131    }
1132    return NULL;
1133}
1134
1135static struct nfs4_client *
1136find_unconfirmed_client(clientid_t *clid)
1137{
1138    struct nfs4_client *clp;
1139    unsigned int idhashval = clientid_hashval(clid->cl_id);
1140
1141    list_for_each_entry(clp, &unconf_id_hashtbl[idhashval], cl_idhash) {
1142        if (same_clid(&clp->cl_clientid, clid))
1143            return clp;
1144    }
1145    return NULL;
1146}
1147
1148static bool clp_used_exchangeid(struct nfs4_client *clp)
1149{
1150    return clp->cl_exchange_flags != 0;
1151}
1152
1153static struct nfs4_client *
1154find_confirmed_client_by_str(const char *dname, unsigned int hashval)
1155{
1156    struct nfs4_client *clp;
1157
1158    list_for_each_entry(clp, &conf_str_hashtbl[hashval], cl_strhash) {
1159        if (same_name(clp->cl_recdir, dname))
1160            return clp;
1161    }
1162    return NULL;
1163}
1164
1165static struct nfs4_client *
1166find_unconfirmed_client_by_str(const char *dname, unsigned int hashval)
1167{
1168    struct nfs4_client *clp;
1169
1170    list_for_each_entry(clp, &unconf_str_hashtbl[hashval], cl_strhash) {
1171        if (same_name(clp->cl_recdir, dname))
1172            return clp;
1173    }
1174    return NULL;
1175}
1176
1177static void rpc_svcaddr2sockaddr(struct sockaddr *sa, unsigned short family, union svc_addr_u *svcaddr)
1178{
1179    switch (family) {
1180    case AF_INET:
1181        ((struct sockaddr_in *)sa)->sin_family = AF_INET;
1182        ((struct sockaddr_in *)sa)->sin_addr = svcaddr->addr;
1183        return;
1184    case AF_INET6:
1185        ((struct sockaddr_in6 *)sa)->sin6_family = AF_INET6;
1186        ((struct sockaddr_in6 *)sa)->sin6_addr = svcaddr->addr6;
1187        return;
1188    }
1189}
1190
1191static void
1192gen_callback(struct nfs4_client *clp, struct nfsd4_setclientid *se, struct svc_rqst *rqstp)
1193{
1194    struct nfs4_cb_conn *conn = &clp->cl_cb_conn;
1195    struct sockaddr *sa = svc_addr(rqstp);
1196    u32 scopeid = rpc_get_scope_id(sa);
1197    unsigned short expected_family;
1198
1199    /* Currently, we only support tcp and tcp6 for the callback channel */
1200    if (se->se_callback_netid_len == 3 &&
1201        !memcmp(se->se_callback_netid_val, "tcp", 3))
1202        expected_family = AF_INET;
1203    else if (se->se_callback_netid_len == 4 &&
1204         !memcmp(se->se_callback_netid_val, "tcp6", 4))
1205        expected_family = AF_INET6;
1206    else
1207        goto out_err;
1208
1209    conn->cb_addrlen = rpc_uaddr2sockaddr(se->se_callback_addr_val,
1210                        se->se_callback_addr_len,
1211                        (struct sockaddr *)&conn->cb_addr,
1212                        sizeof(conn->cb_addr));
1213
1214    if (!conn->cb_addrlen || conn->cb_addr.ss_family != expected_family)
1215        goto out_err;
1216
1217    if (conn->cb_addr.ss_family == AF_INET6)
1218        ((struct sockaddr_in6 *)&conn->cb_addr)->sin6_scope_id = scopeid;
1219
1220    conn->cb_prog = se->se_callback_prog;
1221    conn->cb_ident = se->se_callback_ident;
1222    rpc_svcaddr2sockaddr((struct sockaddr *)&conn->cb_saddr, expected_family, &rqstp->rq_daddr);
1223    return;
1224out_err:
1225    conn->cb_addr.ss_family = AF_UNSPEC;
1226    conn->cb_addrlen = 0;
1227    dprintk(KERN_INFO "NFSD: this client (clientid %08x/%08x) "
1228        "will not receive delegations\n",
1229        clp->cl_clientid.cl_boot, clp->cl_clientid.cl_id);
1230
1231    return;
1232}
1233
1234/*
1235 * Cache a reply. nfsd4_check_drc_limit() has bounded the cache size.
1236 */
1237void
1238nfsd4_store_cache_entry(struct nfsd4_compoundres *resp)
1239{
1240    struct nfsd4_slot *slot = resp->cstate.slot;
1241    unsigned int base;
1242
1243    dprintk("--> %s slot %p\n", __func__, slot);
1244
1245    slot->sl_opcnt = resp->opcnt;
1246    slot->sl_status = resp->cstate.status;
1247
1248    if (nfsd4_not_cached(resp)) {
1249        slot->sl_datalen = 0;
1250        return;
1251    }
1252    slot->sl_datalen = (char *)resp->p - (char *)resp->cstate.datap;
1253    base = (char *)resp->cstate.datap -
1254                    (char *)resp->xbuf->head[0].iov_base;
1255    if (read_bytes_from_xdr_buf(resp->xbuf, base, slot->sl_data,
1256                    slot->sl_datalen))
1257        WARN("%s: sessions DRC could not cache compound\n", __func__);
1258    return;
1259}
1260
1261/*
1262 * Encode the replay sequence operation from the slot values.
1263 * If cachethis is FALSE encode the uncached rep error on the next
1264 * operation which sets resp->p and increments resp->opcnt for
1265 * nfs4svc_encode_compoundres.
1266 *
1267 */
1268static __be32
1269nfsd4_enc_sequence_replay(struct nfsd4_compoundargs *args,
1270              struct nfsd4_compoundres *resp)
1271{
1272    struct nfsd4_op *op;
1273    struct nfsd4_slot *slot = resp->cstate.slot;
1274
1275    dprintk("--> %s resp->opcnt %d cachethis %u \n", __func__,
1276        resp->opcnt, resp->cstate.slot->sl_cachethis);
1277
1278    /* Encode the replayed sequence operation */
1279    op = &args->ops[resp->opcnt - 1];
1280    nfsd4_encode_operation(resp, op);
1281
1282    /* Return nfserr_retry_uncached_rep in next operation. */
1283    if (args->opcnt > 1 && slot->sl_cachethis == 0) {
1284        op = &args->ops[resp->opcnt++];
1285        op->status = nfserr_retry_uncached_rep;
1286        nfsd4_encode_operation(resp, op);
1287    }
1288    return op->status;
1289}
1290
1291/*
1292 * The sequence operation is not cached because we can use the slot and
1293 * session values.
1294 */
1295__be32
1296nfsd4_replay_cache_entry(struct nfsd4_compoundres *resp,
1297             struct nfsd4_sequence *seq)
1298{
1299    struct nfsd4_slot *slot = resp->cstate.slot;
1300    __be32 status;
1301
1302    dprintk("--> %s slot %p\n", __func__, slot);
1303
1304    /* Either returns 0 or nfserr_retry_uncached */
1305    status = nfsd4_enc_sequence_replay(resp->rqstp->rq_argp, resp);
1306    if (status == nfserr_retry_uncached_rep)
1307        return status;
1308
1309    /* The sequence operation has been encoded, cstate->datap set. */
1310    memcpy(resp->cstate.datap, slot->sl_data, slot->sl_datalen);
1311
1312    resp->opcnt = slot->sl_opcnt;
1313    resp->p = resp->cstate.datap + XDR_QUADLEN(slot->sl_datalen);
1314    status = slot->sl_status;
1315
1316    return status;
1317}
1318
1319/*
1320 * Set the exchange_id flags returned by the server.
1321 */
1322static void
1323nfsd4_set_ex_flags(struct nfs4_client *new, struct nfsd4_exchange_id *clid)
1324{
1325    /* pNFS is not supported */
1326    new->cl_exchange_flags |= EXCHGID4_FLAG_USE_NON_PNFS;
1327
1328    /* Referrals are supported, Migration is not. */
1329    new->cl_exchange_flags |= EXCHGID4_FLAG_SUPP_MOVED_REFER;
1330
1331    /* set the wire flags to return to client. */
1332    clid->flags = new->cl_exchange_flags;
1333}
1334
1335__be32
1336nfsd4_exchange_id(struct svc_rqst *rqstp,
1337          struct nfsd4_compound_state *cstate,
1338          struct nfsd4_exchange_id *exid)
1339{
1340    struct nfs4_client *unconf, *conf, *new;
1341    int status;
1342    unsigned int strhashval;
1343    char dname[HEXDIR_LEN];
1344    char addr_str[INET6_ADDRSTRLEN];
1345    nfs4_verifier verf = exid->verifier;
1346    struct sockaddr *sa = svc_addr(rqstp);
1347
1348    rpc_ntop(sa, addr_str, sizeof(addr_str));
1349    dprintk("%s rqstp=%p exid=%p clname.len=%u clname.data=%p "
1350        "ip_addr=%s flags %x, spa_how %d\n",
1351        __func__, rqstp, exid, exid->clname.len, exid->clname.data,
1352        addr_str, exid->flags, exid->spa_how);
1353
1354    if (!check_name(exid->clname) || (exid->flags & ~EXCHGID4_FLAG_MASK_A))
1355        return nfserr_inval;
1356
1357    /* Currently only support SP4_NONE */
1358    switch (exid->spa_how) {
1359    case SP4_NONE:
1360        break;
1361    case SP4_SSV:
1362        return nfserr_serverfault;
1363    default:
1364        BUG(); /* checked by xdr code */
1365    case SP4_MACH_CRED:
1366        return nfserr_serverfault; /* no excuse :-/ */
1367    }
1368
1369    status = nfs4_make_rec_clidname(dname, &exid->clname);
1370
1371    if (status)
1372        goto error;
1373
1374    strhashval = clientstr_hashval(dname);
1375
1376    nfs4_lock_state();
1377    status = nfs_ok;
1378
1379    conf = find_confirmed_client_by_str(dname, strhashval);
1380    if (conf) {
1381        if (!clp_used_exchangeid(conf)) {
1382            status = nfserr_clid_inuse; /* XXX: ? */
1383            goto out;
1384        }
1385        if (!same_verf(&verf, &conf->cl_verifier)) {
1386            /* 18.35.4 case 8 */
1387            if (exid->flags & EXCHGID4_FLAG_UPD_CONFIRMED_REC_A) {
1388                status = nfserr_not_same;
1389                goto out;
1390            }
1391            /* Client reboot: destroy old state */
1392            expire_client(conf);
1393            goto out_new;
1394        }
1395        if (!same_creds(&conf->cl_cred, &rqstp->rq_cred)) {
1396            /* 18.35.4 case 9 */
1397            if (exid->flags & EXCHGID4_FLAG_UPD_CONFIRMED_REC_A) {
1398                status = nfserr_perm;
1399                goto out;
1400            }
1401            expire_client(conf);
1402            goto out_new;
1403        }
1404        /*
1405         * Set bit when the owner id and verifier map to an already
1406         * confirmed client id (18.35.3).
1407         */
1408        exid->flags |= EXCHGID4_FLAG_CONFIRMED_R;
1409
1410        /*
1411         * Falling into 18.35.4 case 2, possible router replay.
1412         * Leave confirmed record intact and return same result.
1413         */
1414        copy_verf(conf, &verf);
1415        new = conf;
1416        goto out_copy;
1417    }
1418
1419    /* 18.35.4 case 7 */
1420    if (exid->flags & EXCHGID4_FLAG_UPD_CONFIRMED_REC_A) {
1421        status = nfserr_noent;
1422        goto out;
1423    }
1424
1425    unconf = find_unconfirmed_client_by_str(dname, strhashval);
1426    if (unconf) {
1427        /*
1428         * Possible retry or client restart. Per 18.35.4 case 4,
1429         * a new unconfirmed record should be generated regardless
1430         * of whether any properties have changed.
1431         */
1432        expire_client(unconf);
1433    }
1434
1435out_new:
1436    /* Normal case */
1437    new = create_client(exid->clname, dname, rqstp, &verf);
1438    if (new == NULL) {
1439        status = nfserr_jukebox;
1440        goto out;
1441    }
1442
1443    gen_clid(new);
1444    add_to_unconfirmed(new, strhashval);
1445out_copy:
1446    exid->clientid.cl_boot = new->cl_clientid.cl_boot;
1447    exid->clientid.cl_id = new->cl_clientid.cl_id;
1448
1449    exid->seqid = 1;
1450    nfsd4_set_ex_flags(new, exid);
1451
1452    dprintk("nfsd4_exchange_id seqid %d flags %x\n",
1453        new->cl_cs_slot.sl_seqid, new->cl_exchange_flags);
1454    status = nfs_ok;
1455
1456out:
1457    nfs4_unlock_state();
1458error:
1459    dprintk("nfsd4_exchange_id returns %d\n", ntohl(status));
1460    return status;
1461}
1462
1463static int
1464check_slot_seqid(u32 seqid, u32 slot_seqid, int slot_inuse)
1465{
1466    dprintk("%s enter. seqid %d slot_seqid %d\n", __func__, seqid,
1467        slot_seqid);
1468
1469    /* The slot is in use, and no response has been sent. */
1470    if (slot_inuse) {
1471        if (seqid == slot_seqid)
1472            return nfserr_jukebox;
1473        else
1474            return nfserr_seq_misordered;
1475    }
1476    /* Normal */
1477    if (likely(seqid == slot_seqid + 1))
1478        return nfs_ok;
1479    /* Replay */
1480    if (seqid == slot_seqid)
1481        return nfserr_replay_cache;
1482    /* Wraparound */
1483    if (seqid == 1 && (slot_seqid + 1) == 0)
1484        return nfs_ok;
1485    /* Misordered replay or misordered new request */
1486    return nfserr_seq_misordered;
1487}
1488
1489/*
1490 * Cache the create session result into the create session single DRC
1491 * slot cache by saving the xdr structure. sl_seqid has been set.
1492 * Do this for solo or embedded create session operations.
1493 */
1494static void
1495nfsd4_cache_create_session(struct nfsd4_create_session *cr_ses,
1496               struct nfsd4_clid_slot *slot, int nfserr)
1497{
1498    slot->sl_status = nfserr;
1499    memcpy(&slot->sl_cr_ses, cr_ses, sizeof(*cr_ses));
1500}
1501
1502static __be32
1503nfsd4_replay_create_session(struct nfsd4_create_session *cr_ses,
1504                struct nfsd4_clid_slot *slot)
1505{
1506    memcpy(cr_ses, &slot->sl_cr_ses, sizeof(*cr_ses));
1507    return slot->sl_status;
1508}
1509
1510__be32
1511nfsd4_create_session(struct svc_rqst *rqstp,
1512             struct nfsd4_compound_state *cstate,
1513             struct nfsd4_create_session *cr_ses)
1514{
1515    struct sockaddr *sa = svc_addr(rqstp);
1516    struct nfs4_client *conf, *unconf;
1517    struct nfsd4_session *new;
1518    struct nfsd4_clid_slot *cs_slot = NULL;
1519    bool confirm_me = false;
1520    int status = 0;
1521
1522    if (cr_ses->flags & ~SESSION4_FLAG_MASK_A)
1523        return nfserr_inval;
1524
1525    nfs4_lock_state();
1526    unconf = find_unconfirmed_client(&cr_ses->clientid);
1527    conf = find_confirmed_client(&cr_ses->clientid);
1528
1529    if (conf) {
1530        cs_slot = &conf->cl_cs_slot;
1531        status = check_slot_seqid(cr_ses->seqid, cs_slot->sl_seqid, 0);
1532        if (status == nfserr_replay_cache) {
1533            dprintk("Got a create_session replay! seqid= %d\n",
1534                cs_slot->sl_seqid);
1535            /* Return the cached reply status */
1536            status = nfsd4_replay_create_session(cr_ses, cs_slot);
1537            goto out;
1538        } else if (cr_ses->seqid != cs_slot->sl_seqid + 1) {
1539            status = nfserr_seq_misordered;
1540            dprintk("Sequence misordered!\n");
1541            dprintk("Expected seqid= %d but got seqid= %d\n",
1542                cs_slot->sl_seqid, cr_ses->seqid);
1543            goto out;
1544        }
1545    } else if (unconf) {
1546        if (!same_creds(&unconf->cl_cred, &rqstp->rq_cred) ||
1547            !rpc_cmp_addr(sa, (struct sockaddr *) &unconf->cl_addr)) {
1548            status = nfserr_clid_inuse;
1549            goto out;
1550        }
1551
1552        cs_slot = &unconf->cl_cs_slot;
1553        status = check_slot_seqid(cr_ses->seqid, cs_slot->sl_seqid, 0);
1554        if (status) {
1555            /* an unconfirmed replay returns misordered */
1556            status = nfserr_seq_misordered;
1557            goto out;
1558        }
1559
1560        confirm_me = true;
1561        conf = unconf;
1562    } else {
1563        status = nfserr_stale_clientid;
1564        goto out;
1565    }
1566
1567    /*
1568     * XXX: we should probably set this at creation time, and check
1569     * for consistent minorversion use throughout:
1570     */
1571    conf->cl_minorversion = 1;
1572    /*
1573     * We do not support RDMA or persistent sessions
1574     */
1575    cr_ses->flags &= ~SESSION4_PERSIST;
1576    cr_ses->flags &= ~SESSION4_RDMA;
1577
1578    status = nfserr_jukebox;
1579    new = alloc_init_session(rqstp, conf, cr_ses);
1580    if (!new)
1581        goto out;
1582    status = nfs_ok;
1583    memcpy(cr_ses->sessionid.data, new->se_sessionid.data,
1584           NFS4_MAX_SESSIONID_LEN);
1585    memcpy(&cr_ses->fore_channel, &new->se_fchannel,
1586        sizeof(struct nfsd4_channel_attrs));
1587    cs_slot->sl_seqid++;
1588    cr_ses->seqid = cs_slot->sl_seqid;
1589
1590    /* cache solo and embedded create sessions under the state lock */
1591    nfsd4_cache_create_session(cr_ses, cs_slot, status);
1592    if (confirm_me)
1593        move_to_confirmed(conf);
1594out:
1595    nfs4_unlock_state();
1596    dprintk("%s returns %d\n", __func__, ntohl(status));
1597    return status;
1598}
1599
1600static bool nfsd4_last_compound_op(struct svc_rqst *rqstp)
1601{
1602    struct nfsd4_compoundres *resp = rqstp->rq_resp;
1603    struct nfsd4_compoundargs *argp = rqstp->rq_argp;
1604
1605    return argp->opcnt == resp->opcnt;
1606}
1607
1608static __be32 nfsd4_map_bcts_dir(u32 *dir)
1609{
1610    switch (*dir) {
1611    case NFS4_CDFC4_FORE:
1612    case NFS4_CDFC4_BACK:
1613        return nfs_ok;
1614    case NFS4_CDFC4_FORE_OR_BOTH:
1615    case NFS4_CDFC4_BACK_OR_BOTH:
1616        *dir = NFS4_CDFC4_BOTH;
1617        return nfs_ok;
1618    };
1619    return nfserr_inval;
1620}
1621
1622__be32 nfsd4_bind_conn_to_session(struct svc_rqst *rqstp,
1623             struct nfsd4_compound_state *cstate,
1624             struct nfsd4_bind_conn_to_session *bcts)
1625{
1626    __be32 status;
1627
1628    if (!nfsd4_last_compound_op(rqstp))
1629        return nfserr_not_only_op;
1630    spin_lock(&client_lock);
1631    cstate->session = find_in_sessionid_hashtbl(&bcts->sessionid);
1632    /* Sorta weird: we only need the refcnt'ing because new_conn acquires
1633     * client_lock iself: */
1634    if (cstate->session) {
1635        nfsd4_get_session(cstate->session);
1636        atomic_inc(&cstate->session->se_client->cl_refcount);
1637    }
1638    spin_unlock(&client_lock);
1639    if (!cstate->session)
1640        return nfserr_badsession;
1641
1642    status = nfsd4_map_bcts_dir(&bcts->dir);
1643    if (!status)
1644        nfsd4_new_conn(rqstp, cstate->session, bcts->dir);
1645    return status;
1646}
1647
1648static bool nfsd4_compound_in_session(struct nfsd4_session *session, struct nfs4_sessionid *sid)
1649{
1650    if (!session)
1651        return 0;
1652    return !memcmp(sid, &session->se_sessionid, sizeof(*sid));
1653}
1654
1655__be32
1656nfsd4_destroy_session(struct svc_rqst *r,
1657              struct nfsd4_compound_state *cstate,
1658              struct nfsd4_destroy_session *sessionid)
1659{
1660    struct nfsd4_session *ses;
1661    u32 status = nfserr_badsession;
1662
1663    /* Notes:
1664     * - The confirmed nfs4_client->cl_sessionid holds destroyed sessinid
1665     * - Should we return nfserr_back_chan_busy if waiting for
1666     * callbacks on to-be-destroyed session?
1667     * - Do we need to clear any callback info from previous session?
1668     */
1669
1670    if (nfsd4_compound_in_session(cstate->session, &sessionid->sessionid)) {
1671        if (!nfsd4_last_compound_op(r))
1672            return nfserr_not_only_op;
1673    }
1674    dump_sessionid(__func__, &sessionid->sessionid);
1675    spin_lock(&client_lock);
1676    ses = find_in_sessionid_hashtbl(&sessionid->sessionid);
1677    if (!ses) {
1678        spin_unlock(&client_lock);
1679        goto out;
1680    }
1681
1682    unhash_session(ses);
1683    spin_unlock(&client_lock);
1684
1685    nfs4_lock_state();
1686    nfsd4_probe_callback_sync(ses->se_client);
1687    nfs4_unlock_state();
1688
1689    nfsd4_del_conns(ses);
1690
1691    nfsd4_put_session(ses);
1692    status = nfs_ok;
1693out:
1694    dprintk("%s returns %d\n", __func__, ntohl(status));
1695    return status;
1696}
1697
1698static struct nfsd4_conn *__nfsd4_find_conn(struct svc_xprt *xpt, struct nfsd4_session *s)
1699{
1700    struct nfsd4_conn *c;
1701
1702    list_for_each_entry(c, &s->se_conns, cn_persession) {
1703        if (c->cn_xprt == xpt) {
1704            return c;
1705        }
1706    }
1707    return NULL;
1708}
1709
1710static void nfsd4_sequence_check_conn(struct nfsd4_conn *new, struct nfsd4_session *ses)
1711{
1712    struct nfs4_client *clp = ses->se_client;
1713    struct nfsd4_conn *c;
1714    int ret;
1715
1716    spin_lock(&clp->cl_lock);
1717    c = __nfsd4_find_conn(new->cn_xprt, ses);
1718    if (c) {
1719        spin_unlock(&clp->cl_lock);
1720        free_conn(new);
1721        return;
1722    }
1723    __nfsd4_hash_conn(new, ses);
1724    spin_unlock(&clp->cl_lock);
1725    ret = nfsd4_register_conn(new);
1726    if (ret)
1727        /* oops; xprt is already down: */
1728        nfsd4_conn_lost(&new->cn_xpt_user);
1729    return;
1730}
1731
1732static bool nfsd4_session_too_many_ops(struct svc_rqst *rqstp, struct nfsd4_session *session)
1733{
1734    struct nfsd4_compoundargs *args = rqstp->rq_argp;
1735
1736    return args->opcnt > session->se_fchannel.maxops;
1737}
1738
1739__be32
1740nfsd4_sequence(struct svc_rqst *rqstp,
1741           struct nfsd4_compound_state *cstate,
1742           struct nfsd4_sequence *seq)
1743{
1744    struct nfsd4_compoundres *resp = rqstp->rq_resp;
1745    struct nfsd4_session *session;
1746    struct nfsd4_slot *slot;
1747    struct nfsd4_conn *conn;
1748    int status;
1749
1750    if (resp->opcnt != 1)
1751        return nfserr_sequence_pos;
1752
1753    /*
1754     * Will be either used or freed by nfsd4_sequence_check_conn
1755     * below.
1756     */
1757    conn = alloc_conn(rqstp, NFS4_CDFC4_FORE);
1758    if (!conn)
1759        return nfserr_jukebox;
1760
1761    spin_lock(&client_lock);
1762    status = nfserr_badsession;
1763    session = find_in_sessionid_hashtbl(&seq->sessionid);
1764    if (!session)
1765        goto out;
1766
1767    status = nfserr_too_many_ops;
1768    if (nfsd4_session_too_many_ops(rqstp, session))
1769        goto out;
1770
1771    status = nfserr_badslot;
1772    if (seq->slotid >= session->se_fchannel.maxreqs)
1773        goto out;
1774
1775    slot = session->se_slots[seq->slotid];
1776    dprintk("%s: slotid %d\n", __func__, seq->slotid);
1777
1778    /* We do not negotiate the number of slots yet, so set the
1779     * maxslots to the session maxreqs which is used to encode
1780     * sr_highest_slotid and the sr_target_slot id to maxslots */
1781    seq->maxslots = session->se_fchannel.maxreqs;
1782
1783    status = check_slot_seqid(seq->seqid, slot->sl_seqid, slot->sl_inuse);
1784    if (status == nfserr_replay_cache) {
1785        cstate->slot = slot;
1786        cstate->session = session;
1787        /* Return the cached reply status and set cstate->status
1788         * for nfsd4_proc_compound processing */
1789        status = nfsd4_replay_cache_entry(resp, seq);
1790        cstate->status = nfserr_replay_cache;
1791        goto out;
1792    }
1793    if (status)
1794        goto out;
1795
1796    nfsd4_sequence_check_conn(conn, session);
1797    conn = NULL;
1798
1799    /* Success! bump slot seqid */
1800    slot->sl_inuse = true;
1801    slot->sl_seqid = seq->seqid;
1802    slot->sl_cachethis = seq->cachethis;
1803
1804    cstate->slot = slot;
1805    cstate->session = session;
1806
1807out:
1808    /* Hold a session reference until done processing the compound. */
1809    if (cstate->session) {
1810        struct nfs4_client *clp = session->se_client;
1811
1812        nfsd4_get_session(cstate->session);
1813        atomic_inc(&clp->cl_refcount);
1814        if (clp->cl_cb_state == NFSD4_CB_DOWN)
1815            seq->status_flags |= SEQ4_STATUS_CB_PATH_DOWN;
1816    }
1817    kfree(conn);
1818    spin_unlock(&client_lock);
1819    dprintk("%s: return %d\n", __func__, ntohl(status));
1820    return status;
1821}
1822
1823__be32
1824nfsd4_reclaim_complete(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate, struct nfsd4_reclaim_complete *rc)
1825{
1826    int status = 0;
1827
1828    if (rc->rca_one_fs) {
1829        if (!cstate->current_fh.fh_dentry)
1830            return nfserr_nofilehandle;
1831        /*
1832         * We don't take advantage of the rca_one_fs case.
1833         * That's OK, it's optional, we can safely ignore it.
1834         */
1835         return nfs_ok;
1836    }
1837
1838    nfs4_lock_state();
1839    status = nfserr_complete_already;
1840    if (cstate->session->se_client->cl_firststate)
1841        goto out;
1842
1843    status = nfserr_stale_clientid;
1844    if (is_client_expired(cstate->session->se_client))
1845        /*
1846         * The following error isn't really legal.
1847         * But we only get here if the client just explicitly
1848         * destroyed the client. Surely it no longer cares what
1849         * error it gets back on an operation for the dead
1850         * client.
1851         */
1852        goto out;
1853
1854    status = nfs_ok;
1855    nfsd4_create_clid_dir(cstate->session->se_client);
1856out:
1857    nfs4_unlock_state();
1858    return status;
1859}
1860
1861__be32
1862nfsd4_setclientid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
1863          struct nfsd4_setclientid *setclid)
1864{
1865    struct xdr_netobj clname = {
1866        .len = setclid->se_namelen,
1867        .data = setclid->se_name,
1868    };
1869    nfs4_verifier clverifier = setclid->se_verf;
1870    unsigned int strhashval;
1871    struct nfs4_client *conf, *unconf, *new;
1872    __be32 status;
1873    char dname[HEXDIR_LEN];
1874    
1875    if (!check_name(clname))
1876        return nfserr_inval;
1877
1878    status = nfs4_make_rec_clidname(dname, &clname);
1879    if (status)
1880        return status;
1881
1882    /*
1883     * XXX The Duplicate Request Cache (DRC) has been checked (??)
1884     * We get here on a DRC miss.
1885     */
1886
1887    strhashval = clientstr_hashval(dname);
1888
1889    nfs4_lock_state();
1890    conf = find_confirmed_client_by_str(dname, strhashval);
1891    if (conf) {
1892        /* RFC 3530 14.2.33 CASE 0: */
1893        status = nfserr_clid_inuse;
1894        if (clp_used_exchangeid(conf))
1895            goto out;
1896        if (!same_creds(&conf->cl_cred, &rqstp->rq_cred)) {
1897            char addr_str[INET6_ADDRSTRLEN];
1898            rpc_ntop((struct sockaddr *) &conf->cl_addr, addr_str,
1899                 sizeof(addr_str));
1900            dprintk("NFSD: setclientid: string in use by client "
1901                "at %s\n", addr_str);
1902            goto out;
1903        }
1904    }
1905    /*
1906     * section 14.2.33 of RFC 3530 (under the heading "IMPLEMENTATION")
1907     * has a description of SETCLIENTID request processing consisting
1908     * of 5 bullet points, labeled as CASE0 - CASE4 below.
1909     */
1910    unconf = find_unconfirmed_client_by_str(dname, strhashval);
1911    status = nfserr_resource;
1912    if (!conf) {
1913        /*
1914         * RFC 3530 14.2.33 CASE 4:
1915         * placed first, because it is the normal case
1916         */
1917        if (unconf)
1918            expire_client(unconf);
1919        new = create_client(clname, dname, rqstp, &clverifier);
1920        if (new == NULL)
1921            goto out;
1922        gen_clid(new);
1923    } else if (same_verf(&conf->cl_verifier, &clverifier)) {
1924        /*
1925         * RFC 3530 14.2.33 CASE 1:
1926         * probable callback update
1927         */
1928        if (unconf) {
1929            /* Note this is removing unconfirmed {*x***},
1930             * which is stronger than RFC recommended {vxc**}.
1931             * This has the advantage that there is at most
1932             * one {*x***} in either list at any time.
1933             */
1934            expire_client(unconf);
1935        }
1936        new = create_client(clname, dname, rqstp, &clverifier);
1937        if (new == NULL)
1938            goto out;
1939        copy_clid(new, conf);
1940    } else if (!unconf) {
1941        /*
1942         * RFC 3530 14.2.33 CASE 2:
1943         * probable client reboot; state will be removed if
1944         * confirmed.
1945         */
1946        new = create_client(clname, dname, rqstp, &clverifier);
1947        if (new == NULL)
1948            goto out;
1949        gen_clid(new);
1950    } else {
1951        /*
1952         * RFC 3530 14.2.33 CASE 3:
1953         * probable client reboot; state will be removed if
1954         * confirmed.
1955         */
1956        expire_client(unconf);
1957        new = create_client(clname, dname, rqstp, &clverifier);
1958        if (new == NULL)
1959            goto out;
1960        gen_clid(new);
1961    }
1962    /*
1963     * XXX: we should probably set this at creation time, and check
1964     * for consistent minorversion use throughout:
1965     */
1966    new->cl_minorversion = 0;
1967    gen_callback(new, setclid, rqstp);
1968    add_to_unconfirmed(new, strhashval);
1969    setclid->se_clientid.cl_boot = new->cl_clientid.cl_boot;
1970    setclid->se_clientid.cl_id = new->cl_clientid.cl_id;
1971    memcpy(setclid->se_confirm.data, new->cl_confirm.data, sizeof(setclid->se_confirm.data));
1972    status = nfs_ok;
1973out:
1974    nfs4_unlock_state();
1975    return status;
1976}
1977
1978
1979/*
1980 * Section 14.2.34 of RFC 3530 (under the heading "IMPLEMENTATION") has
1981 * a description of SETCLIENTID_CONFIRM request processing consisting of 4
1982 * bullets, labeled as CASE1 - CASE4 below.
1983 */
1984__be32
1985nfsd4_setclientid_confirm(struct svc_rqst *rqstp,
1986             struct nfsd4_compound_state *cstate,
1987             struct nfsd4_setclientid_confirm *setclientid_confirm)
1988{
1989    struct sockaddr *sa = svc_addr(rqstp);
1990    struct nfs4_client *conf, *unconf;
1991    nfs4_verifier confirm = setclientid_confirm->sc_confirm;
1992    clientid_t * clid = &setclientid_confirm->sc_clientid;
1993    __be32 status;
1994
1995    if (STALE_CLIENTID(clid))
1996        return nfserr_stale_clientid;
1997    /*
1998     * XXX The Duplicate Request Cache (DRC) has been checked (??)
1999     * We get here on a DRC miss.
2000     */
2001
2002    nfs4_lock_state();
2003
2004    conf = find_confirmed_client(clid);
2005    unconf = find_unconfirmed_client(clid);
2006
2007    status = nfserr_clid_inuse;
2008    if (conf && !rpc_cmp_addr((struct sockaddr *) &conf->cl_addr, sa))
2009        goto out;
2010    if (unconf && !rpc_cmp_addr((struct sockaddr *) &unconf->cl_addr, sa))
2011        goto out;
2012
2013    /*
2014     * section 14.2.34 of RFC 3530 has a description of
2015     * SETCLIENTID_CONFIRM request processing consisting
2016     * of 4 bullet points, labeled as CASE1 - CASE4 below.
2017     */
2018    if (conf && unconf && same_verf(&confirm, &unconf->cl_confirm)) {
2019        /*
2020         * RFC 3530 14.2.34 CASE 1:
2021         * callback update
2022         */
2023        if (!same_creds(&conf->cl_cred, &unconf->cl_cred))
2024            status = nfserr_clid_inuse;
2025        else {
2026            nfsd4_change_callback(conf, &unconf->cl_cb_conn);
2027            nfsd4_probe_callback(conf);
2028            expire_client(unconf);
2029            status = nfs_ok;
2030
2031        }
2032    } else if (conf && !unconf) {
2033        /*
2034         * RFC 3530 14.2.34 CASE 2:
2035         * probable retransmitted request; play it safe and
2036         * do nothing.
2037         */
2038        if (!same_creds(&conf->cl_cred, &rqstp->rq_cred))
2039            status = nfserr_clid_inuse;
2040        else
2041            status = nfs_ok;
2042    } else if (!conf && unconf
2043            && same_verf(&unconf->cl_confirm, &confirm)) {
2044        /*
2045         * RFC 3530 14.2.34 CASE 3:
2046         * Normal case; new or rebooted client:
2047         */
2048        if (!same_creds(&unconf->cl_cred, &rqstp->rq_cred)) {
2049            status = nfserr_clid_inuse;
2050        } else {
2051            unsigned int hash =
2052                clientstr_hashval(unconf->cl_recdir);
2053            conf = find_confirmed_client_by_str(unconf->cl_recdir,
2054                                hash);
2055            if (conf) {
2056                nfsd4_remove_clid_dir(conf);
2057                expire_client(conf);
2058            }
2059            move_to_confirmed(unconf);
2060            conf = unconf;
2061            nfsd4_probe_callback(conf);
2062            status = nfs_ok;
2063        }
2064    } else if ((!conf || (conf && !same_verf(&conf->cl_confirm, &confirm)))
2065        && (!unconf || (unconf && !same_verf(&unconf->cl_confirm,
2066                                    &confirm)))) {
2067        /*
2068         * RFC 3530 14.2.34 CASE 4:
2069         * Client probably hasn't noticed that we rebooted yet.
2070         */
2071        status = nfserr_stale_clientid;
2072    } else {
2073        /* check that we have hit one of the cases...*/
2074        status = nfserr_clid_inuse;
2075    }
2076out:
2077    nfs4_unlock_state();
2078    return status;
2079}
2080
2081/* OPEN Share state helper functions */
2082static inline struct nfs4_file *
2083alloc_init_file(struct inode *ino)
2084{
2085    struct nfs4_file *fp;
2086    unsigned int hashval = file_hashval(ino);
2087
2088    fp = kmem_cache_alloc(file_slab, GFP_KERNEL);
2089    if (fp) {
2090        atomic_set(&fp->fi_ref, 1);
2091        INIT_LIST_HEAD(&fp->fi_hash);
2092        INIT_LIST_HEAD(&fp->fi_stateids);
2093        INIT_LIST_HEAD(&fp->fi_delegations);
2094        fp->fi_inode = igrab(ino);
2095        fp->fi_id = current_fileid++;
2096        fp->fi_had_conflict = false;
2097        fp->fi_lease = NULL;
2098        memset(fp->fi_fds, 0, sizeof(fp->fi_fds));
2099        memset(fp->fi_access, 0, sizeof(fp->fi_access));
2100        spin_lock(&recall_lock);
2101        list_add(&fp->fi_hash, &file_hashtbl[hashval]);
2102        spin_unlock(&recall_lock);
2103        return fp;
2104    }
2105    return NULL;
2106}
2107
2108static void
2109nfsd4_free_slab(struct kmem_cache **slab)
2110{
2111    if (*slab == NULL)
2112        return;
2113    kmem_cache_destroy(*slab);
2114    *slab = NULL;
2115}
2116
2117void
2118nfsd4_free_slabs(void)
2119{
2120    nfsd4_free_slab(&stateowner_slab);
2121    nfsd4_free_slab(&file_slab);
2122    nfsd4_free_slab(&stateid_slab);
2123    nfsd4_free_slab(&deleg_slab);
2124}
2125
2126static int
2127nfsd4_init_slabs(void)
2128{
2129    stateowner_slab = kmem_cache_create("nfsd4_stateowners",
2130            sizeof(struct nfs4_stateowner), 0, 0, NULL);
2131    if (stateowner_slab == NULL)
2132        goto out_nomem;
2133    file_slab = kmem_cache_create("nfsd4_files",
2134            sizeof(struct nfs4_file), 0, 0, NULL);
2135    if (file_slab == NULL)
2136        goto out_nomem;
2137    stateid_slab = kmem_cache_create("nfsd4_stateids",
2138            sizeof(struct nfs4_stateid), 0, 0, NULL);
2139    if (stateid_slab == NULL)
2140        goto out_nomem;
2141    deleg_slab = kmem_cache_create("nfsd4_delegations",
2142            sizeof(struct nfs4_delegation), 0, 0, NULL);
2143    if (deleg_slab == NULL)
2144        goto out_nomem;
2145    return 0;
2146out_nomem:
2147    nfsd4_free_slabs();
2148    dprintk("nfsd4: out of memory while initializing nfsv4\n");
2149    return -ENOMEM;
2150}
2151
2152void
2153nfs4_free_stateowner(struct kref *kref)
2154{
2155    struct nfs4_stateowner *sop =
2156        container_of(kref, struct nfs4_stateowner, so_ref);
2157    kfree(sop->so_owner.data);
2158    kmem_cache_free(stateowner_slab, sop);
2159}
2160
2161static inline struct nfs4_stateowner *
2162alloc_stateowner(struct xdr_netobj *owner)
2163{
2164    struct nfs4_stateowner *sop;
2165
2166    if ((sop = kmem_cache_alloc(stateowner_slab, GFP_KERNEL))) {
2167        if ((sop->so_owner.data = kmalloc(owner->len, GFP_KERNEL))) {
2168            memcpy(sop->so_owner.data, owner->data, owner->len);
2169            sop->so_owner.len = owner->len;
2170            kref_init(&sop->so_ref);
2171            return sop;
2172        }
2173        kmem_cache_free(stateowner_slab, sop);
2174    }
2175    return NULL;
2176}
2177
2178static struct nfs4_stateowner *
2179alloc_init_open_stateowner(unsigned int strhashval, struct nfs4_client *clp, struct nfsd4_open *open) {
2180    struct nfs4_stateowner *sop;
2181    struct nfs4_replay *rp;
2182    unsigned int idhashval;
2183
2184    if (!(sop = alloc_stateowner(&open->op_owner)))
2185        return NULL;
2186    idhashval = ownerid_hashval(current_ownerid);
2187    INIT_LIST_HEAD(&sop->so_idhash);
2188    INIT_LIST_HEAD(&sop->so_strhash);
2189    INIT_LIST_HEAD(&sop->so_perclient);
2190    INIT_LIST_HEAD(&sop->so_stateids);
2191    INIT_LIST_HEAD(&sop->so_perstateid); /* not used */
2192    INIT_LIST_HEAD(&sop->so_close_lru);
2193    sop->so_time = 0;
2194    list_add(&sop->so_idhash, &ownerid_hashtbl[idhashval]);
2195    list_add(&sop->so_strhash, &ownerstr_hashtbl[strhashval]);
2196    list_add(&sop->so_perclient, &clp->cl_openowners);
2197    sop->so_is_open_owner = 1;
2198    sop->so_id = current_ownerid++;
2199    sop->so_client = clp;
2200    sop->so_seqid = open->op_seqid;
2201    sop->so_confirmed = 0;
2202    rp = &sop->so_replay;
2203    rp->rp_status = nfserr_serverfault;
2204    rp->rp_buflen = 0;
2205    rp->rp_buf = rp->rp_ibuf;
2206    return sop;
2207}
2208
2209static inline void
2210init_stateid(struct nfs4_stateid *stp, struct nfs4_file *fp, struct nfsd4_open *open) {
2211    struct nfs4_stateowner *sop = open->op_stateowner;
2212    unsigned int hashval = stateid_hashval(sop->so_id, fp->fi_id);
2213
2214    INIT_LIST_HEAD(&stp->st_hash);
2215    INIT_LIST_HEAD(&stp->st_perstateowner);
2216    INIT_LIST_HEAD(&stp->st_lockowners);
2217    INIT_LIST_HEAD(&stp->st_perfile);
2218    list_add(&stp->st_hash, &stateid_hashtbl[hashval]);
2219    list_add(&stp->st_perstateowner, &sop->so_stateids);
2220    list_add(&stp->st_perfile, &fp->fi_stateids);
2221    stp->st_stateowner = sop;
2222    get_nfs4_file(fp);
2223    stp->st_file = fp;
2224    stp->st_stateid.si_boot = boot_time;
2225    stp->st_stateid.si_stateownerid = sop->so_id;
2226    stp->st_stateid.si_fileid = fp->fi_id;
2227    stp->st_stateid.si_generation = 0;
2228    stp->st_access_bmap = 0;
2229    stp->st_deny_bmap = 0;
2230    __set_bit(open->op_share_access & ~NFS4_SHARE_WANT_MASK,
2231          &stp->st_access_bmap);
2232    __set_bit(open->op_share_deny, &stp->st_deny_bmap);
2233    stp->st_openstp = NULL;
2234}
2235
2236static void
2237move_to_close_lru(struct nfs4_stateowner *sop)
2238{
2239    dprintk("NFSD: move_to_close_lru nfs4_stateowner %p\n", sop);
2240
2241    list_move_tail(&sop->so_close_lru, &close_lru);
2242    sop->so_time = get_seconds();
2243}
2244
2245static int
2246same_owner_str(struct nfs4_stateowner *sop, struct xdr_netobj *owner,
2247                            clientid_t *clid)
2248{
2249    return (sop->so_owner.len == owner->len) &&
2250        0 == memcmp(sop->so_owner.data, owner->data, owner->len) &&
2251        (sop->so_client->cl_clientid.cl_id == clid->cl_id);
2252}
2253
2254static struct nfs4_stateowner *
2255find_openstateowner_str(unsigned int hashval, struct nfsd4_open *open)
2256{
2257    struct nfs4_stateowner *so = NULL;
2258
2259    list_for_each_entry(so, &ownerstr_hashtbl[hashval], so_strhash) {
2260        if (same_owner_str(so, &open->op_owner, &open->op_clientid))
2261            return so;
2262    }
2263    return NULL;
2264}
2265
2266/* search file_hashtbl[] for file */
2267static struct nfs4_file *
2268find_file(struct inode *ino)
2269{
2270    unsigned int hashval = file_hashval(ino);
2271    struct nfs4_file *fp;
2272
2273    spin_lock(&recall_lock);
2274    list_for_each_entry(fp, &file_hashtbl[hashval], fi_hash) {
2275        if (fp->fi_inode == ino) {
2276            get_nfs4_file(fp);
2277            spin_unlock(&recall_lock);
2278            return fp;
2279        }
2280    }
2281    spin_unlock(&recall_lock);
2282    return NULL;
2283}
2284
2285static inline int access_valid(u32 x, u32 minorversion)
2286{
2287    if ((x & NFS4_SHARE_ACCESS_MASK) < NFS4_SHARE_ACCESS_READ)
2288        return 0;
2289    if ((x & NFS4_SHARE_ACCESS_MASK) > NFS4_SHARE_ACCESS_BOTH)
2290        return 0;
2291    x &= ~NFS4_SHARE_ACCESS_MASK;
2292    if (minorversion && x) {
2293        if ((x & NFS4_SHARE_WANT_MASK) > NFS4_SHARE_WANT_CANCEL)
2294            return 0;
2295        if ((x & NFS4_SHARE_WHEN_MASK) > NFS4_SHARE_PUSH_DELEG_WHEN_UNCONTENDED)
2296            return 0;
2297        x &= ~(NFS4_SHARE_WANT_MASK | NFS4_SHARE_WHEN_MASK);
2298    }
2299    if (x)
2300        return 0;
2301    return 1;
2302}
2303
2304static inline int deny_valid(u32 x)
2305{
2306    /* Note: unlike access bits, deny bits may be zero. */
2307    return x <= NFS4_SHARE_DENY_BOTH;
2308}
2309
2310/*
2311 * Called to check deny when READ with all zero stateid or
2312 * WRITE with all zero or all one stateid
2313 */
2314static __be32
2315nfs4_share_conflict(struct svc_fh *current_fh, unsigned int deny_type)
2316{
2317    struct inode *ino = current_fh->fh_dentry->d_inode;
2318    struct nfs4_file *fp;
2319    struct nfs4_stateid *stp;
2320    __be32 ret;
2321
2322    dprintk("NFSD: nfs4_share_conflict\n");
2323
2324    fp = find_file(ino);
2325    if (!fp)
2326        return nfs_ok;
2327    ret = nfserr_locked;
2328    /* Search for conflicting share reservations */
2329    list_for_each_entry(stp, &fp->fi_stateids, st_perfile) {
2330        if (test_bit(deny_type, &stp->st_deny_bmap) ||
2331            test_bit(NFS4_SHARE_DENY_BOTH, &stp->st_deny_bmap))
2332            goto out;
2333    }
2334    ret = nfs_ok;
2335out:
2336    put_nfs4_file(fp);
2337    return ret;
2338}
2339
2340static inline void
2341nfs4_file_downgrade(struct nfs4_file *fp, unsigned int share_access)
2342{
2343    if (share_access & NFS4_SHARE_ACCESS_WRITE)
2344        nfs4_file_put_access(fp, O_WRONLY);
2345    if (share_access & NFS4_SHARE_ACCESS_READ)
2346        nfs4_file_put_access(fp, O_RDONLY);
2347}
2348
2349static void nfsd_break_one_deleg(struct nfs4_delegation *dp)
2350{
2351    /* We're assuming the state code never drops its reference
2352     * without first removing the lease. Since we're in this lease
2353     * callback (and since the lease code is serialized by the kernel
2354     * lock) we know the server hasn't removed the lease yet, we know
2355     * it's safe to take a reference: */
2356    atomic_inc(&dp->dl_count);
2357
2358    list_add_tail(&dp->dl_recall_lru, &del_recall_lru);
2359
2360    /* only place dl_time is set. protected by lock_flocks*/
2361    dp->dl_time = get_seconds();
2362
2363    nfsd4_cb_recall(dp);
2364}
2365
2366/* Called from break_lease() with lock_flocks() held. */
2367static void nfsd_break_deleg_cb(struct file_lock *fl)
2368{
2369    struct nfs4_file *fp = (struct nfs4_file *)fl->fl_owner;
2370    struct nfs4_delegation *dp;
2371
2372    BUG_ON(!fp);
2373    /* We assume break_lease is only called once per lease: */
2374    BUG_ON(fp->fi_had_conflict);
2375    /*
2376     * We don't want the locks code to timeout the lease for us;
2377     * we'll remove it ourself if a delegation isn't returned
2378     * in time:
2379     */
2380    fl->fl_break_time = 0;
2381
2382    spin_lock(&recall_lock);
2383    fp->fi_had_conflict = true;
2384    list_for_each_entry(dp, &fp->fi_delegations, dl_perfile)
2385        nfsd_break_one_deleg(dp);
2386    spin_unlock(&recall_lock);
2387}
2388
2389static
2390int nfsd_change_deleg_cb(struct file_lock **onlist, int arg)
2391{
2392    if (arg & F_UNLCK)
2393        return lease_modify(onlist, arg);
2394    else
2395        return -EAGAIN;
2396}
2397
2398static const struct lock_manager_operations nfsd_lease_mng_ops = {
2399    .fl_break = nfsd_break_deleg_cb,
2400    .fl_change = nfsd_change_deleg_cb,
2401};
2402
2403
2404__be32
2405nfsd4_process_open1(struct nfsd4_compound_state *cstate,
2406            struct nfsd4_open *open)
2407{
2408    clientid_t *clientid = &open->op_clientid;
2409    struct nfs4_client *clp = NULL;
2410    unsigned int strhashval;
2411    struct nfs4_stateowner *sop = NULL;
2412
2413    if (!check_name(open->op_owner))
2414        return nfserr_inval;
2415
2416    if (STALE_CLIENTID(&open->op_clientid))
2417        return nfserr_stale_clientid;
2418
2419    strhashval = ownerstr_hashval(clientid->cl_id, open->op_owner);
2420    sop = find_openstateowner_str(strhashval, open);
2421    open->op_stateowner = sop;
2422    if (!sop) {
2423        /* Make sure the client's lease hasn't expired. */
2424        clp = find_confirmed_client(clientid);
2425        if (clp == NULL)
2426            return nfserr_expired;
2427        goto renew;
2428    }
2429    /* When sessions are used, skip open sequenceid processing */
2430    if (nfsd4_has_session(cstate))
2431        goto renew;
2432    if (!sop->so_confirmed) {
2433        /* Replace unconfirmed owners without checking for replay. */
2434        clp = sop->so_client;
2435        release_openowner(sop);
2436        open->op_stateowner = NULL;
2437        goto renew;
2438    }
2439    if (open->op_seqid == sop->so_seqid - 1) {
2440        if (sop->so_replay.rp_buflen)
2441            return nfserr_replay_me;
2442        /* The original OPEN failed so spectacularly
2443         * that we don't even have replay data saved!
2444         * Therefore, we have no choice but to continue
2445         * processing this OPEN; presumably, we'll
2446         * fail again for the same reason.
2447         */
2448        dprintk("nfsd4_process_open1: replay with no replay cache\n");
2449        goto renew;
2450    }
2451    if (open->op_seqid != sop->so_seqid)
2452        return nfserr_bad_seqid;
2453renew:
2454    if (open->op_stateowner == NULL) {
2455        sop = alloc_init_open_stateowner(strhashval, clp, open);
2456        if (sop == NULL)
2457            return nfserr_resource;
2458        open->op_stateowner = sop;
2459    }
2460    list_del_init(&sop->so_close_lru);
2461    renew_client(sop->so_client);
2462    return nfs_ok;
2463}
2464
2465static inline __be32
2466nfs4_check_delegmode(struct nfs4_delegation *dp, int flags)
2467{
2468    if ((flags & WR_STATE) && (dp->dl_type == NFS4_OPEN_DELEGATE_READ))
2469        return nfserr_openmode;
2470    else
2471        return nfs_ok;
2472}
2473
2474static struct nfs4_delegation *
2475find_delegation_file(struct nfs4_file *fp, stateid_t *stid)
2476{
2477    struct nfs4_delegation *dp;
2478
2479    spin_lock(&recall_lock);
2480    list_for_each_entry(dp, &fp->fi_delegations, dl_perfile)
2481        if (dp->dl_stateid.si_stateownerid == stid->si_stateownerid) {
2482            spin_unlock(&recall_lock);
2483            return dp;
2484        }
2485    spin_unlock(&recall_lock);
2486    return NULL;
2487}
2488
2489static int share_access_to_flags(u32 share_access)
2490{
2491    share_access &= ~NFS4_SHARE_WANT_MASK;
2492
2493    return share_access == NFS4_SHARE_ACCESS_READ ? RD_STATE : WR_STATE;
2494}
2495
2496static __be32
2497nfs4_check_deleg(struct nfs4_file *fp, struct nfsd4_open *open,
2498        struct nfs4_delegation **dp)
2499{
2500    int flags;
2501    __be32 status = nfserr_bad_stateid;
2502
2503    *dp = find_delegation_file(fp, &open->op_delegate_stateid);
2504    if (*dp == NULL)
2505        goto out;
2506    flags = share_access_to_flags(open->op_share_access);
2507    status = nfs4_check_delegmode(*dp, flags);
2508    if (status)
2509        *dp = NULL;
2510out:
2511    if (open->op_claim_type != NFS4_OPEN_CLAIM_DELEGATE_CUR)
2512        return nfs_ok;
2513    if (status)
2514        return status;
2515    open->op_stateowner->so_confirmed = 1;
2516    return nfs_ok;
2517}
2518
2519static __be32
2520nfs4_check_open(struct nfs4_file *fp, struct nfsd4_open *open, struct nfs4_stateid **stpp)
2521{
2522    struct nfs4_stateid *local;
2523    __be32 status = nfserr_share_denied;
2524    struct nfs4_stateowner *sop = open->op_stateowner;
2525
2526    list_for_each_entry(local, &fp->fi_stateids, st_perfile) {
2527        /* ignore lock owners */
2528        if (local->st_stateowner->so_is_open_owner == 0)
2529            continue;
2530        /* remember if we have seen this open owner */
2531        if (local->st_stateowner == sop)
2532            *stpp = local;
2533        /* check for conflicting share reservations */
2534        if (!test_share(local, open))
2535            goto out;
2536    }
2537    status = 0;
2538out:
2539    return status;
2540}
2541
2542static inline struct nfs4_stateid *
2543nfs4_alloc_stateid(void)
2544{
2545    return kmem_cache_alloc(stateid_slab, GFP_KERNEL);
2546}
2547
2548static inline int nfs4_access_to_access(u32 nfs4_access)
2549{
2550    int flags = 0;
2551
2552    if (nfs4_access & NFS4_SHARE_ACCESS_READ)
2553        flags |= NFSD_MAY_READ;
2554    if (nfs4_access & NFS4_SHARE_ACCESS_WRITE)
2555        flags |= NFSD_MAY_WRITE;
2556    return flags;
2557}
2558
2559static __be32 nfs4_get_vfs_file(struct svc_rqst *rqstp, struct nfs4_file
2560*fp, struct svc_fh *cur_fh, u32 nfs4_access)
2561{
2562    __be32 status;
2563    int oflag = nfs4_access_to_omode(nfs4_access);
2564    int access = nfs4_access_to_access(nfs4_access);
2565
2566    if (!fp->fi_fds[oflag]) {
2567        status = nfsd_open(rqstp, cur_fh, S_IFREG, access,
2568            &fp->fi_fds[oflag]);
2569        if (status)
2570            return status;
2571    }
2572    nfs4_file_get_access(fp, oflag);
2573
2574    return nfs_ok;
2575}
2576
2577static __be32
2578nfs4_new_open(struct svc_rqst *rqstp, struct nfs4_stateid **stpp,
2579        struct nfs4_file *fp, struct svc_fh *cur_fh,
2580        struct nfsd4_open *open)
2581{
2582    struct nfs4_stateid *stp;
2583    __be32 status;
2584
2585    stp = nfs4_alloc_stateid();
2586    if (stp == NULL)
2587        return nfserr_resource;
2588
2589    status = nfs4_get_vfs_file(rqstp, fp, cur_fh, open->op_share_access);
2590    if (status) {
2591        kmem_cache_free(stateid_slab, stp);
2592        return status;
2593    }
2594    *stpp = stp;
2595    return 0;
2596}
2597
2598static inline __be32
2599nfsd4_truncate(struct svc_rqst *rqstp, struct svc_fh *fh,
2600        struct nfsd4_open *open)
2601{
2602    struct iattr iattr = {
2603        .ia_valid = ATTR_SIZE,
2604        .ia_size = 0,
2605    };
2606    if (!open->op_truncate)
2607        return 0;
2608    if (!(open->op_share_access & NFS4_SHARE_ACCESS_WRITE))
2609        return nfserr_inval;
2610    return nfsd_setattr(rqstp, fh, &iattr, 0, (time_t)0);
2611}
2612
2613static __be32
2614nfs4_upgrade_open(struct svc_rqst *rqstp, struct nfs4_file *fp, struct svc_fh *cur_fh, struct nfs4_stateid *stp, struct nfsd4_open *open)
2615{
2616    u32 op_share_access = open->op_share_access & ~NFS4_SHARE_WANT_MASK;
2617    bool new_access;
2618    __be32 status;
2619
2620    new_access = !test_bit(op_share_access, &stp->st_access_bmap);
2621    if (new_access) {
2622        status = nfs4_get_vfs_file(rqstp, fp, cur_fh, op_share_access);
2623        if (status)
2624            return status;
2625    }
2626    status = nfsd4_truncate(rqstp, cur_fh, open);
2627    if (status) {
2628        if (new_access) {
2629            int oflag = nfs4_access_to_omode(new_access);
2630            nfs4_file_put_access(fp, oflag);
2631        }
2632        return status;
2633    }
2634    /* remember the open */
2635    __set_bit(op_share_access, &stp->st_access_bmap);
2636    __set_bit(open->op_share_deny, &stp->st_deny_bmap);
2637
2638    return nfs_ok;
2639}
2640
2641
2642static void
2643nfs4_set_claim_prev(struct nfsd4_open *open)
2644{
2645    open->op_stateowner->so_confirmed = 1;
2646    open->op_stateowner->so_client->cl_firststate = 1;
2647}
2648
2649/* Should we give out recallable state?: */
2650static bool nfsd4_cb_channel_good(struct nfs4_client *clp)
2651{
2652    if (clp->cl_cb_state == NFSD4_CB_UP)
2653        return true;
2654    /*
2655     * In the sessions case, since we don't have to establish a
2656     * separate connection for callbacks, we assume it's OK
2657     * until we hear otherwise:
2658     */
2659    return clp->cl_minorversion && clp->cl_cb_state == NFSD4_CB_UNKNOWN;
2660}
2661
2662static struct file_lock *nfs4_alloc_init_lease(struct nfs4_delegation *dp, int flag)
2663{
2664    struct file_lock *fl;
2665
2666    fl = locks_alloc_lock();
2667    if (!fl)
2668        return NULL;
2669    locks_init_lock(fl);
2670    fl->fl_lmops = &nfsd_lease_mng_ops;
2671    fl->fl_flags = FL_LEASE;
2672    fl->fl_type = flag == NFS4_OPEN_DELEGATE_READ? F_RDLCK: F_WRLCK;
2673    fl->fl_end = OFFSET_MAX;
2674    fl->fl_owner = (fl_owner_t)(dp->dl_file);
2675    fl->fl_pid = current->tgid;
2676    return fl;
2677}
2678
2679static int nfs4_setlease(struct nfs4_delegation *dp, int flag)
2680{
2681    struct nfs4_file *fp = dp->dl_file;
2682    struct file_lock *fl;
2683    int status;
2684
2685    fl = nfs4_alloc_init_lease(dp, flag);
2686    if (!fl)
2687        return -ENOMEM;
2688    fl->fl_file = find_readable_file(fp);
2689    list_add(&dp->dl_perclnt, &dp->dl_client->cl_delegations);
2690    status = vfs_setlease(fl->fl_file, fl->fl_type, &fl);
2691    if (status) {
2692        list_del_init(&dp->dl_perclnt);
2693        locks_free_lock(fl);
2694        return -ENOMEM;
2695    }
2696    fp->fi_lease = fl;
2697    fp->fi_deleg_file = fl->fl_file;
2698    get_file(fp->fi_deleg_file);
2699    atomic_set(&fp->fi_delegees, 1);
2700    list_add(&dp->dl_perfile, &fp->fi_delegations);
2701    return 0;
2702}
2703
2704static int nfs4_set_delegation(struct nfs4_delegation *dp, int flag)
2705{
2706    struct nfs4_file *fp = dp->dl_file;
2707
2708    if (!fp->fi_lease)
2709        return nfs4_setlease(dp, flag);
2710    spin_lock(&recall_lock);
2711    if (fp->fi_had_conflict) {
2712        spin_unlock(&recall_lock);
2713        return -EAGAIN;
2714    }
2715    atomic_inc(&fp->fi_delegees);
2716    list_add(&dp->dl_perfile, &fp->fi_delegations);
2717    spin_unlock(&recall_lock);
2718    list_add(&dp->dl_perclnt, &dp->dl_client->cl_delegations);
2719    return 0;
2720}
2721
2722/*
2723 * Attempt to hand out a delegation.
2724 */
2725static void
2726nfs4_open_delegation(struct svc_fh *fh, struct nfsd4_open *open, struct nfs4_stateid *stp)
2727{
2728    struct nfs4_delegation *dp;
2729    struct nfs4_stateowner *sop = stp->st_stateowner;
2730    int cb_up;
2731    int status, flag = 0;
2732
2733    cb_up = nfsd4_cb_channel_good(sop->so_client);
2734    flag = NFS4_OPEN_DELEGATE_NONE;
2735    open->op_recall = 0;
2736    switch (open->op_claim_type) {
2737        case NFS4_OPEN_CLAIM_PREVIOUS:
2738            if (!cb_up)
2739                open->op_recall = 1;
2740            flag = open->op_delegate_type;
2741            if (flag == NFS4_OPEN_DELEGATE_NONE)
2742                goto out;
2743            break;
2744        case NFS4_OPEN_CLAIM_NULL:
2745            /* Let's not give out any delegations till everyone's
2746             * had the chance to reclaim theirs.... */
2747            if (locks_in_grace())
2748                goto out;
2749            if (!cb_up || !sop->so_confirmed)
2750                goto out;
2751            if (open->op_share_access & NFS4_SHARE_ACCESS_WRITE)
2752                flag = NFS4_OPEN_DELEGATE_WRITE;
2753            else
2754                flag = NFS4_OPEN_DELEGATE_READ;
2755            break;
2756        default:
2757            goto out;
2758    }
2759
2760    dp = alloc_init_deleg(sop->so_client, stp, fh, flag);
2761    if (dp == NULL)
2762        goto out_no_deleg;
2763    status = nfs4_set_delegation(dp, flag);
2764    if (status)
2765        goto out_free;
2766
2767    memcpy(&open->op_delegate_stateid, &dp->dl_stateid, sizeof(dp->dl_stateid));
2768
2769    dprintk("NFSD: delegation stateid=" STATEID_FMT "\n",
2770        STATEID_VAL(&dp->dl_stateid));
2771out:
2772    if (open->op_claim_type == NFS4_OPEN_CLAIM_PREVIOUS
2773            && flag == NFS4_OPEN_DELEGATE_NONE
2774            && open->op_delegate_type != NFS4_OPEN_DELEGATE_NONE)
2775        dprintk("NFSD: WARNING: refusing delegation reclaim\n");
2776    open->op_delegate_type = flag;
2777    return;
2778out_free:
2779    nfs4_put_delegation(dp);
2780out_no_deleg:
2781    flag = NFS4_OPEN_DELEGATE_NONE;
2782    goto out;
2783}
2784
2785/*
2786 * called with nfs4_lock_state() held.
2787 */
2788__be32
2789nfsd4_process_open2(struct svc_rqst *rqstp, struct svc_fh *current_fh, struct nfsd4_open *open)
2790{
2791    struct nfsd4_compoundres *resp = rqstp->rq_resp;
2792    struct nfs4_file *fp = NULL;
2793    struct inode *ino = current_fh->fh_dentry->d_inode;
2794    struct nfs4_stateid *stp = NULL;
2795    struct nfs4_delegation *dp = NULL;
2796    __be32 status;
2797
2798    status = nfserr_inval;
2799    if (!access_valid(open->op_share_access, resp->cstate.minorversion)
2800            || !deny_valid(open->op_share_deny))
2801        goto out;
2802    /*
2803     * Lookup file; if found, lookup stateid and check open request,
2804     * and check for delegations in the process of being recalled.
2805     * If not found, create the nfs4_file struct
2806     */
2807    fp = find_file(ino);
2808    if (fp) {
2809        if ((status = nfs4_check_open(fp, open, &stp)))
2810            goto out;
2811        status = nfs4_check_deleg(fp, open, &dp);
2812        if (status)
2813            goto out;
2814    } else {
2815        status = nfserr_bad_stateid;
2816        if (open->op_claim_type == NFS4_OPEN_CLAIM_DELEGATE_CUR)
2817            goto out;
2818        status = nfserr_resource;
2819        fp = alloc_init_file(ino);
2820        if (fp == NULL)
2821            goto out;
2822    }
2823
2824    /*
2825     * OPEN the file, or upgrade an existing OPEN.
2826     * If truncate fails, the OPEN fails.
2827     */
2828    if (stp) {
2829        /* Stateid was found, this is an OPEN upgrade */
2830        status = nfs4_upgrade_open(rqstp, fp, current_fh, stp, open);
2831        if (status)
2832            goto out;
2833        update_stateid(&stp->st_stateid);
2834    } else {
2835        status = nfs4_new_open(rqstp, &stp, fp, current_fh, open);
2836        if (status)
2837            goto out;
2838        init_stateid(stp, fp, open);
2839        status = nfsd4_truncate(rqstp, current_fh, open);
2840        if (status) {
2841            release_open_stateid(stp);
2842            goto out;
2843        }
2844        if (nfsd4_has_session(&resp->cstate))
2845            update_stateid(&stp->st_stateid);
2846    }
2847    memcpy(&open->op_stateid, &stp->st_stateid, sizeof(stateid_t));
2848
2849    if (nfsd4_has_session(&resp->cstate))
2850        open->op_stateowner->so_confirmed = 1;
2851
2852    /*
2853    * Attempt to hand out a delegation. No error return, because the
2854    * OPEN succeeds even if we fail.
2855    */
2856    nfs4_open_delegation(current_fh, open, stp);
2857
2858    status = nfs_ok;
2859
2860    dprintk("%s: stateid=" STATEID_FMT "\n", __func__,
2861        STATEID_VAL(&stp->st_stateid));
2862out:
2863    if (fp)
2864        put_nfs4_file(fp);
2865    if (status == 0 && open->op_claim_type == NFS4_OPEN_CLAIM_PREVIOUS)
2866        nfs4_set_claim_prev(open);
2867    /*
2868    * To finish the open response, we just need to set the rflags.
2869    */
2870    open->op_rflags = NFS4_OPEN_RESULT_LOCKTYPE_POSIX;
2871    if (!open->op_stateowner->so_confirmed &&
2872        !nfsd4_has_session(&resp->cstate))
2873        open->op_rflags |= NFS4_OPEN_RESULT_CONFIRM;
2874
2875    return status;
2876}
2877
2878__be32
2879nfsd4_renew(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
2880        clientid_t *clid)
2881{
2882    struct nfs4_client *clp;
2883    __be32 status;
2884
2885    nfs4_lock_state();
2886    dprintk("process_renew(%08x/%08x): starting\n",
2887            clid->cl_boot, clid->cl_id);
2888    status = nfserr_stale_clientid;
2889    if (STALE_CLIENTID(clid))
2890        goto out;
2891    clp = find_confirmed_client(clid);
2892    status = nfserr_expired;
2893    if (clp == NULL) {
2894        /* We assume the client took too long to RENEW. */
2895        dprintk("nfsd4_renew: clientid not found!\n");
2896        goto out;
2897    }
2898    renew_client(clp);
2899    status = nfserr_cb_path_down;
2900    if (!list_empty(&clp->cl_delegations)
2901            && clp->cl_cb_state != NFSD4_CB_UP)
2902        goto out;
2903    status = nfs_ok;
2904out:
2905    nfs4_unlock_state();
2906    return status;
2907}
2908
2909static struct lock_manager nfsd4_manager = {
2910};
2911
2912static void
2913nfsd4_end_grace(void)
2914{
2915    dprintk("NFSD: end of grace period\n");
2916    nfsd4_recdir_purge_old();
2917    locks_end_grace(&nfsd4_manager);
2918    /*
2919     * Now that every NFSv4 client has had the chance to recover and
2920     * to see the (possibly new, possibly shorter) lease time, we
2921     * can safely set the next grace time to the current lease time:
2922     */
2923    nfsd4_grace = nfsd4_lease;
2924}
2925
2926static time_t
2927nfs4_laundromat(void)
2928{
2929    struct nfs4_client *clp;
2930    struct nfs4_stateowner *sop;
2931    struct nfs4_delegation *dp;
2932    struct list_head *pos, *next, reaplist;
2933    time_t cutoff = get_seconds() - nfsd4_lease;
2934    time_t t, clientid_val = nfsd4_lease;
2935    time_t u, test_val = nfsd4_lease;
2936
2937    nfs4_lock_state();
2938
2939    dprintk("NFSD: laundromat service - starting\n");
2940    if (locks_in_grace())
2941        nfsd4_end_grace();
2942    INIT_LIST_HEAD(&reaplist);
2943    spin_lock(&client_lock);
2944    list_for_each_safe(pos, next, &client_lru) {
2945        clp = list_entry(pos, struct nfs4_client, cl_lru);
2946        if (time_after((unsigned long)clp->cl_time, (unsigned long)cutoff)) {
2947            t = clp->cl_time - cutoff;
2948            if (clientid_val > t)
2949                clientid_val = t;
2950            break;
2951        }
2952        if (atomic_read(&clp->cl_refcount)) {
2953            dprintk("NFSD: client in use (clientid %08x)\n",
2954                clp->cl_clientid.cl_id);
2955            continue;
2956        }
2957        unhash_client_locked(clp);
2958        list_add(&clp->cl_lru, &reaplist);
2959    }
2960    spin_unlock(&client_lock);
2961    list_for_each_safe(pos, next, &reaplist) {
2962        clp = list_entry(pos, struct nfs4_client, cl_lru);
2963        dprintk("NFSD: purging unused client (clientid %08x)\n",
2964            clp->cl_clientid.cl_id);
2965        nfsd4_remove_clid_dir(clp);
2966        expire_client(clp);
2967    }
2968    spin_lock(&recall_lock);
2969    list_for_each_safe(pos, next, &del_recall_lru) {
2970        dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
2971        if (time_after((unsigned long)dp->dl_time, (unsigned long)cutoff)) {
2972            u = dp->dl_time - cutoff;
2973            if (test_val > u)
2974                test_val = u;
2975            break;
2976        }
2977        list_move(&dp->dl_recall_lru, &reaplist);
2978    }
2979    spin_unlock(&recall_lock);
2980    list_for_each_safe(pos, next, &reaplist) {
2981        dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
2982        list_del_init(&dp->dl_recall_lru);
2983        unhash_delegation(dp);
2984    }
2985    test_val = nfsd4_lease;
2986    list_for_each_safe(pos, next, &close_lru) {
2987        sop = list_entry(pos, struct nfs4_stateowner, so_close_lru);
2988        if (time_after((unsigned long)sop->so_time, (unsigned long)cutoff)) {
2989            u = sop->so_time - cutoff;
2990            if (test_val > u)
2991                test_val = u;
2992            break;
2993        }
2994        dprintk("NFSD: purging unused open stateowner (so_id %d)\n",
2995            sop->so_id);
2996        release_openowner(sop);
2997    }
2998    if (clientid_val < NFSD_LAUNDROMAT_MINTIMEOUT)
2999        clientid_val = NFSD_LAUNDROMAT_MINTIMEOUT;
3000    nfs4_unlock_state();
3001    return clientid_val;
3002}
3003
3004static struct workqueue_struct *laundry_wq;
3005static void laundromat_main(struct work_struct *);
3006static DECLARE_DELAYED_WORK(laundromat_work, laundromat_main);
3007
3008static void
3009laundromat_main(struct work_struct *not_used)
3010{
3011    time_t t;
3012
3013    t = nfs4_laundromat();
3014    dprintk("NFSD: laundromat_main - sleeping for %ld seconds\n", t);
3015    queue_delayed_work(laundry_wq, &laundromat_work, t*HZ);
3016}
3017
3018static struct nfs4_stateowner *
3019search_close_lru(u32 st_id, int flags)
3020{
3021    struct nfs4_stateowner *local = NULL;
3022
3023    if (flags & CLOSE_STATE) {
3024        list_for_each_entry(local, &close_lru, so_close_lru) {
3025            if (local->so_id == st_id)
3026                return local;
3027        }
3028    }
3029    return NULL;
3030}
3031
3032static inline int
3033nfs4_check_fh(struct svc_fh *fhp, struct nfs4_stateid *stp)
3034{
3035    return fhp->fh_dentry->d_inode != stp->st_file->fi_inode;
3036}
3037
3038static int
3039STALE_STATEID(stateid_t *stateid)
3040{
3041    if (stateid->si_boot == boot_time)
3042        return 0;
3043    dprintk("NFSD: stale stateid " STATEID_FMT "!\n",
3044        STATEID_VAL(stateid));
3045    return 1;
3046}
3047
3048static inline int
3049access_permit_read(unsigned long access_bmap)
3050{
3051    return test_bit(NFS4_SHARE_ACCESS_READ, &access_bmap) ||
3052        test_bit(NFS4_SHARE_ACCESS_BOTH, &access_bmap) ||
3053        test_bit(NFS4_SHARE_ACCESS_WRITE, &access_bmap);
3054}
3055
3056static inline int
3057access_permit_write(unsigned long access_bmap)
3058{
3059    return test_bit(NFS4_SHARE_ACCESS_WRITE, &access_bmap) ||
3060        test_bit(NFS4_SHARE_ACCESS_BOTH, &access_bmap);
3061}
3062
3063static
3064__be32 nfs4_check_openmode(struct nfs4_stateid *stp, int flags)
3065{
3066        __be32 status = nfserr_openmode;
3067
3068    /* For lock stateid's, we test the parent open, not the lock: */
3069    if (stp->st_openstp)
3070        stp = stp->st_openstp;
3071    if ((flags & WR_STATE) && (!access_permit_write(stp->st_access_bmap)))
3072                goto out;
3073    if ((flags & RD_STATE) && (!access_permit_read(stp->st_access_bmap)))
3074                goto out;
3075    status = nfs_ok;
3076out:
3077    return status;
3078}
3079
3080static inline __be32
3081check_special_stateids(svc_fh *current_fh, stateid_t *stateid, int flags)
3082{
3083    if (ONE_STATEID(stateid) && (flags & RD_STATE))
3084        return nfs_ok;
3085    else if (locks_in_grace()) {
3086        /* Answer in remaining cases depends on existence of
3087         * conflicting state; so we must wait out the grace period. */
3088        return nfserr_grace;
3089    } else if (flags & WR_STATE)
3090        return nfs4_share_conflict(current_fh,
3091                NFS4_SHARE_DENY_WRITE);
3092    else /* (flags & RD_STATE) && ZERO_STATEID(stateid) */
3093        return nfs4_share_conflict(current_fh,
3094                NFS4_SHARE_DENY_READ);
3095}
3096
3097/*
3098 * Allow READ/WRITE during grace period on recovered state only for files
3099 * that are not able to provide mandatory locking.
3100 */
3101static inline int
3102grace_disallows_io(struct inode *inode)
3103{
3104    return locks_in_grace() && mandatory_lock(inode);
3105}
3106
3107static int check_stateid_generation(stateid_t *in, stateid_t *ref, int flags)
3108{
3109    /*
3110     * When sessions are used the stateid generation number is ignored
3111     * when it is zero.
3112     */
3113    if ((flags & HAS_SESSION) && in->si_generation == 0)
3114        goto out;
3115
3116    /* If the client sends us a stateid from the future, it's buggy: */
3117    if (in->si_generation > ref->si_generation)
3118        return nfserr_bad_stateid;
3119    /*
3120     * The following, however, can happen. For example, if the
3121     * client sends an open and some IO at the same time, the open
3122     * may bump si_generation while the IO is still in flight.
3123     * Thanks to hard links and renames, the client never knows what
3124     * file an open will affect. So it could avoid that situation
3125     * only by serializing all opens and IO from the same open
3126     * owner. To recover from the old_stateid error, the client
3127     * will just have to retry the IO:
3128     */
3129    if (in->si_generation < ref->si_generation)
3130        return nfserr_old_stateid;
3131out:
3132    return nfs_ok;
3133}
3134
3135static int is_delegation_stateid(stateid_t *stateid)
3136{
3137    return stateid->si_fileid == 0;
3138}
3139
3140/*
3141* Checks for stateid operations
3142*/
3143__be32
3144nfs4_preprocess_stateid_op(struct nfsd4_compound_state *cstate,
3145               stateid_t *stateid, int flags, struct file **filpp)
3146{
3147    struct nfs4_stateid *stp = NULL;
3148    struct nfs4_delegation *dp = NULL;
3149    struct svc_fh *current_fh = &cstate->current_fh;
3150    struct inode *ino = current_fh->fh_dentry->d_inode;
3151    __be32 status;
3152
3153    if (filpp)
3154        *filpp = NULL;
3155
3156    if (grace_disallows_io(ino))
3157        return nfserr_grace;
3158
3159    if (nfsd4_has_session(cstate))
3160        flags |= HAS_SESSION;
3161
3162    if (ZERO_STATEID(stateid) || ONE_STATEID(stateid))
3163        return check_special_stateids(current_fh, stateid, flags);
3164
3165    status = nfserr_stale_stateid;
3166    if (STALE_STATEID(stateid))
3167        goto out;
3168
3169    /*
3170     * We assume that any stateid that has the current boot time,
3171     * but that we can't find, is expired:
3172     */
3173    status = nfserr_expired;
3174    if (is_delegation_stateid(stateid)) {
3175        dp = find_delegation_stateid(ino, stateid);
3176        if (!dp)
3177            goto out;
3178        status = check_stateid_generation(stateid, &dp->dl_stateid,
3179                          flags);
3180        if (status)
3181            goto out;
3182        status = nfs4_check_delegmode(dp, flags);
3183        if (status)
3184            goto out;
3185        renew_client(dp->dl_client);
3186        if (filpp) {
3187            *filpp = dp->dl_file->fi_deleg_file;
3188            BUG_ON(!*filpp);
3189        }
3190    } else { /* open or lock stateid */
3191        stp = find_stateid(stateid, flags);
3192        if (!stp)
3193            goto out;
3194        status = nfserr_bad_stateid;
3195        if (nfs4_check_fh(current_fh, stp))
3196            goto out;
3197        if (!stp->st_stateowner->so_confirmed)
3198            goto out;
3199        status = check_stateid_generation(stateid, &stp->st_stateid,
3200                          flags);
3201        if (status)
3202            goto out;
3203        status = nfs4_check_openmode(stp, flags);
3204        if (status)
3205            goto out;
3206        renew_client(stp->st_stateowner->so_client);
3207        if (filpp) {
3208            if (flags & RD_STATE)
3209                *filpp = find_readable_file(stp->st_file);
3210            else
3211                *filpp = find_writeable_file(stp->st_file);
3212        }
3213    }
3214    status = nfs_ok;
3215out:
3216    return status;
3217}
3218
3219static inline int
3220setlkflg (int type)
3221{
3222    return (type == NFS4_READW_LT || type == NFS4_READ_LT) ?
3223        RD_STATE : WR_STATE;
3224}
3225
3226/*
3227 * Checks for sequence id mutating operations.
3228 */
3229static __be32
3230nfs4_preprocess_seqid_op(struct nfsd4_compound_state *cstate, u32 seqid,
3231             stateid_t *stateid, int flags,
3232             struct nfs4_stateowner **sopp,
3233             struct nfs4_stateid **stpp, struct nfsd4_lock *lock)
3234{
3235    struct nfs4_stateid *stp;
3236    struct nfs4_stateowner *sop;
3237    struct svc_fh *current_fh = &cstate->current_fh;
3238    __be32 status;
3239
3240    dprintk("NFSD: %s: seqid=%d stateid = " STATEID_FMT "\n", __func__,
3241        seqid, STATEID_VAL(stateid));
3242
3243    *stpp = NULL;
3244    *sopp = NULL;
3245
3246    if (ZERO_STATEID(stateid) || ONE_STATEID(stateid)) {
3247        dprintk("NFSD: preprocess_seqid_op: magic stateid!\n");
3248        return nfserr_bad_stateid;
3249    }
3250
3251    if (STALE_STATEID(stateid))
3252        return nfserr_stale_stateid;
3253
3254    if (nfsd4_has_session(cstate))
3255        flags |= HAS_SESSION;
3256
3257    /*
3258    * We return BAD_STATEID if filehandle doesn't match stateid,
3259    * the confirmed flag is incorrecly set, or the generation
3260    * number is incorrect.
3261    */
3262    stp = find_stateid(stateid, flags);
3263    if (stp == NULL) {
3264        /*
3265         * Also, we should make sure this isn't just the result of
3266         * a replayed close:
3267         */
3268        sop = search_close_lru(stateid->si_stateownerid, flags);
3269        /* It's not stale; let's assume it's expired: */
3270        if (sop == NULL)
3271            return nfserr_expired;
3272        *sopp = sop;
3273        goto check_replay;
3274    }
3275
3276    *stpp = stp;
3277    *sopp = sop = stp->st_stateowner;
3278
3279    if (lock) {
3280        clientid_t *lockclid = &lock->v.new.clientid;
3281        struct nfs4_client *clp = sop->so_client;
3282        int lkflg = 0;
3283        __be32 status;
3284
3285        lkflg = setlkflg(lock->lk_type);
3286
3287        if (lock->lk_is_new) {
3288            if (!sop->so_is_open_owner)
3289                return nfserr_bad_stateid;
3290            if (!(flags & HAS_SESSION) &&
3291                !same_clid(&clp->cl_clientid, lockclid))
3292                return nfserr_bad_stateid;
3293            /* stp is the open stateid */
3294            status = nfs4_check_openmode(stp, lkflg);
3295            if (status)
3296                return status;
3297        } else {
3298            /* stp is the lock stateid */
3299            status = nfs4_check_openmode(stp->st_openstp, lkflg);
3300            if (status)
3301                return status;
3302               }
3303    }
3304
3305    if (nfs4_check_fh(current_fh, stp)) {
3306        dprintk("NFSD: preprocess_seqid_op: fh-stateid mismatch!\n");
3307        return nfserr_bad_stateid;
3308    }
3309
3310    /*
3311    * We now validate the seqid and stateid generation numbers.
3312    * For the moment, we ignore the possibility of
3313    * generation number wraparound.
3314    */
3315    if (!(flags & HAS_SESSION) && seqid != sop->so_seqid)
3316        goto check_replay;
3317
3318    if (sop->so_confirmed && flags & CONFIRM) {
3319        dprintk("NFSD: preprocess_seqid_op: expected"
3320                " unconfirmed stateowner!\n");
3321        return nfserr_bad_stateid;
3322    }
3323    if (!sop->so_confirmed && !(flags & CONFIRM)) {
3324        dprintk("NFSD: preprocess_seqid_op: stateowner not"
3325                " confirmed yet!\n");
3326        return nfserr_bad_stateid;
3327    }
3328    status = check_stateid_generation(stateid, &stp->st_stateid, flags);
3329    if (status)
3330        return status;
3331    renew_client(sop->so_client);
3332    return nfs_ok;
3333
3334check_replay:
3335    if (seqid == sop->so_seqid - 1) {
3336        dprintk("NFSD: preprocess_seqid_op: retransmission?\n");
3337        /* indicate replay to calling function */
3338        return nfserr_replay_me;
3339    }
3340    dprintk("NFSD: preprocess_seqid_op: bad seqid (expected %d, got %d)\n",
3341            sop->so_seqid, seqid);
3342    *sopp = NULL;
3343    return nfserr_bad_seqid;
3344}
3345
3346__be32
3347nfsd4_open_confirm(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3348           struct nfsd4_open_confirm *oc)
3349{
3350    __be32 status;
3351    struct nfs4_stateowner *sop;
3352    struct nfs4_stateid *stp;
3353
3354    dprintk("NFSD: nfsd4_open_confirm on file %.*s\n",
3355            (int)cstate->current_fh.fh_dentry->d_name.len,
3356            cstate->current_fh.fh_dentry->d_name.name);
3357
3358    status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0);
3359    if (status)
3360        return status;
3361
3362    nfs4_lock_state();
3363
3364    if ((status = nfs4_preprocess_seqid_op(cstate,
3365                    oc->oc_seqid, &oc->oc_req_stateid,
3366                    CONFIRM | OPEN_STATE,
3367                    &oc->oc_stateowner, &stp, NULL)))
3368        goto out;
3369
3370    sop = oc->oc_stateowner;
3371    sop->so_confirmed = 1;
3372    update_stateid(&stp->st_stateid);
3373    memcpy(&oc->oc_resp_stateid, &stp->st_stateid, sizeof(stateid_t));
3374    dprintk("NFSD: %s: success, seqid=%d stateid=" STATEID_FMT "\n",
3375        __func__, oc->oc_seqid, STATEID_VAL(&stp->st_stateid));
3376
3377    nfsd4_create_clid_dir(sop->so_client);
3378out:
3379    if (oc->oc_stateowner) {
3380        nfs4_get_stateowner(oc->oc_stateowner);
3381        cstate->replay_owner = oc->oc_stateowner;
3382    }
3383    nfs4_unlock_state();
3384    return status;
3385}
3386
3387
3388/*
3389 * unset all bits in union bitmap (bmap) that
3390 * do not exist in share (from successful OPEN_DOWNGRADE)
3391 */
3392static void
3393reset_union_bmap_access(unsigned long access, unsigned long *bmap)
3394{
3395    int i;
3396    for (i = 1; i < 4; i++) {
3397        if ((i & access) != i)
3398            __clear_bit(i, bmap);
3399    }
3400}
3401
3402static void
3403reset_union_bmap_deny(unsigned long deny, unsigned long *bmap)
3404{
3405    int i;
3406    for (i = 0; i < 4; i++) {
3407        if ((i & deny) != i)
3408            __clear_bit(i, bmap);
3409    }
3410}
3411
3412__be32
3413nfsd4_open_downgrade(struct svc_rqst *rqstp,
3414             struct nfsd4_compound_state *cstate,
3415             struct nfsd4_open_downgrade *od)
3416{
3417    __be32 status;
3418    struct nfs4_stateid *stp;
3419    unsigned int share_access;
3420
3421    dprintk("NFSD: nfsd4_open_downgrade on file %.*s\n",
3422            (int)cstate->current_fh.fh_dentry->d_name.len,
3423            cstate->current_fh.fh_dentry->d_name.name);
3424
3425    if (!access_valid(od->od_share_access, cstate->minorversion)
3426            || !deny_valid(od->od_share_deny))
3427        return nfserr_inval;
3428
3429    nfs4_lock_state();
3430    if ((status = nfs4_preprocess_seqid_op(cstate,
3431                    od->od_seqid,
3432                    &od->od_stateid,
3433                    OPEN_STATE,
3434                    &od->od_stateowner, &stp, NULL)))
3435        goto out;
3436
3437    status = nfserr_inval;
3438    if (!test_bit(od->od_share_access, &stp->st_access_bmap)) {
3439        dprintk("NFSD:access not a subset current bitmap: 0x%lx, input access=%08x\n",
3440            stp->st_access_bmap, od->od_share_access);
3441        goto out;
3442    }
3443    if (!test_bit(od->od_share_deny, &stp->st_deny_bmap)) {
3444        dprintk("NFSD:deny not a subset current bitmap: 0x%lx, input deny=%08x\n",
3445            stp->st_deny_bmap, od->od_share_deny);
3446        goto out;
3447    }
3448    set_access(&share_access, stp->st_access_bmap);
3449    nfs4_file_downgrade(stp->st_file, share_access & ~od->od_share_access);
3450
3451    reset_union_bmap_access(od->od_share_access, &stp->st_access_bmap);
3452    reset_union_bmap_deny(od->od_share_deny, &stp->st_deny_bmap);
3453
3454    update_stateid(&stp->st_stateid);
3455    memcpy(&od->od_stateid, &stp->st_stateid, sizeof(stateid_t));
3456    status = nfs_ok;
3457out:
3458    if (od->od_stateowner) {
3459        nfs4_get_stateowner(od->od_stateowner);
3460        cstate->replay_owner = od->od_stateowner;
3461    }
3462    nfs4_unlock_state();
3463    return status;
3464}
3465
3466/*
3467 * nfs4_unlock_state() called after encode
3468 */
3469__be32
3470nfsd4_close(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3471        struct nfsd4_close *close)
3472{
3473    __be32 status;
3474    struct nfs4_stateid *stp;
3475
3476    dprintk("NFSD: nfsd4_close on file %.*s\n",
3477            (int)cstate->current_fh.fh_dentry->d_name.len,
3478            cstate->current_fh.fh_dentry->d_name.name);
3479
3480    nfs4_lock_state();
3481    /* check close_lru for replay */
3482    if ((status = nfs4_preprocess_seqid_op(cstate,
3483                    close->cl_seqid,
3484                    &close->cl_stateid,
3485                    OPEN_STATE | CLOSE_STATE,
3486                    &close->cl_stateowner, &stp, NULL)))
3487        goto out;
3488    status = nfs_ok;
3489    update_stateid(&stp->st_stateid);
3490    memcpy(&close->cl_stateid, &stp->st_stateid, sizeof(stateid_t));
3491
3492    /* release_stateid() calls nfsd_close() if needed */
3493    release_open_stateid(stp);
3494
3495    /* place unused nfs4_stateowners on so_close_lru list to be
3496     * released by the laundromat service after the lease period
3497     * to enable us to handle CLOSE replay
3498     */
3499    if (list_empty(&close->cl_stateowner->so_stateids))
3500        move_to_close_lru(close->cl_stateowner);
3501out:
3502    if (close->cl_stateowner) {
3503        nfs4_get_stateowner(close->cl_stateowner);
3504        cstate->replay_owner = close->cl_stateowner;
3505    }
3506    nfs4_unlock_state();
3507    return status;
3508}
3509
3510__be32
3511nfsd4_delegreturn(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3512          struct nfsd4_delegreturn *dr)
3513{
3514    struct nfs4_delegation *dp;
3515    stateid_t *stateid = &dr->dr_stateid;
3516    struct inode *inode;
3517    __be32 status;
3518    int flags = 0;
3519
3520    if ((status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0)))
3521        return status;
3522    inode = cstate->current_fh.fh_dentry->d_inode;
3523
3524    if (nfsd4_has_session(cstate))
3525        flags |= HAS_SESSION;
3526    nfs4_lock_state();
3527    status = nfserr_bad_stateid;
3528    if (ZERO_STATEID(stateid) || ONE_STATEID(stateid))
3529        goto out;
3530    status = nfserr_stale_stateid;
3531    if (STALE_STATEID(stateid))
3532        goto out;
3533    status = nfserr_bad_stateid;
3534    if (!is_delegation_stateid(stateid))
3535        goto out;
3536    status = nfserr_expired;
3537    dp = find_delegation_stateid(inode, stateid);
3538    if (!dp)
3539        goto out;
3540    status = check_stateid_generation(stateid, &dp->dl_stateid, flags);
3541    if (status)
3542        goto out;
3543    renew_client(dp->dl_client);
3544
3545    unhash_delegation(dp);
3546out:
3547    nfs4_unlock_state();
3548
3549    return status;
3550}
3551
3552
3553/*
3554 * Lock owner state (byte-range locks)
3555 */
3556#define LOFF_OVERFLOW(start, len) ((u64)(len) > ~(u64)(start))
3557#define LOCK_HASH_BITS 8
3558#define LOCK_HASH_SIZE (1 << LOCK_HASH_BITS)
3559#define LOCK_HASH_MASK (LOCK_HASH_SIZE - 1)
3560
3561static inline u64
3562end_offset(u64 start, u64 len)
3563{
3564    u64 end;
3565
3566    end = start + len;
3567    return end >= start ? end: NFS4_MAX_UINT64;
3568}
3569
3570/* last octet in a range */
3571static inline u64
3572last_byte_offset(u64 start, u64 len)
3573{
3574    u64 end;
3575
3576    BUG_ON(!len);
3577    end = start + len;
3578    return end > start ? end - 1: NFS4_MAX_UINT64;
3579}
3580
3581#define lockownerid_hashval(id) \
3582        ((id) & LOCK_HASH_MASK)
3583
3584static inline unsigned int
3585lock_ownerstr_hashval(struct inode *inode, u32 cl_id,
3586        struct xdr_netobj *ownername)
3587{
3588    return (file_hashval(inode) + cl_id
3589            + opaque_hashval(ownername->data, ownername->len))
3590        & LOCK_HASH_MASK;
3591}
3592
3593static struct list_head lock_ownerid_hashtbl[LOCK_HASH_SIZE];
3594static struct list_head lock_ownerstr_hashtbl[LOCK_HASH_SIZE];
3595static struct list_head lockstateid_hashtbl[STATEID_HASH_SIZE];
3596
3597static struct nfs4_stateid *
3598find_stateid(stateid_t *stid, int flags)
3599{
3600    struct nfs4_stateid *local;
3601    u32 st_id = stid->si_stateownerid;
3602    u32 f_id = stid->si_fileid;
3603    unsigned int hashval;
3604
3605    dprintk("NFSD: find_stateid flags 0x%x\n",flags);
3606    if (flags & (LOCK_STATE | RD_STATE | WR_STATE)) {
3607        hashval = stateid_hashval(st_id, f_id);
3608        list_for_each_entry(local, &lockstateid_hashtbl[hashval], st_hash) {
3609            if ((local->st_stateid.si_stateownerid == st_id) &&
3610                (local->st_stateid.si_fileid == f_id))
3611                return local;
3612        }
3613    }
3614
3615    if (flags & (OPEN_STATE | RD_STATE | WR_STATE)) {
3616        hashval = stateid_hashval(st_id, f_id);
3617        list_for_each_entry(local, &stateid_hashtbl[hashval], st_hash) {
3618            if ((local->st_stateid.si_stateownerid == st_id) &&
3619                (local->st_stateid.si_fileid == f_id))
3620                return local;
3621        }
3622    }
3623    return NULL;
3624}
3625
3626static struct nfs4_delegation *
3627find_delegation_stateid(struct inode *ino, stateid_t *stid)
3628{
3629    struct nfs4_file *fp;
3630    struct nfs4_delegation *dl;
3631
3632    dprintk("NFSD: %s: stateid=" STATEID_FMT "\n", __func__,
3633        STATEID_VAL(stid));
3634
3635    fp = find_file(ino);
3636    if (!fp)
3637        return NULL;
3638    dl = find_delegation_file(fp, stid);
3639    put_nfs4_file(fp);
3640    return dl;
3641}
3642
3643/*
3644 * TODO: Linux file offsets are _signed_ 64-bit quantities, which means that
3645 * we can't properly handle lock requests that go beyond the (2^63 - 1)-th
3646 * byte, because of sign extension problems. Since NFSv4 calls for 64-bit
3647 * locking, this prevents us from being completely protocol-compliant. The
3648 * real solution to this problem is to start using unsigned file offsets in
3649 * the VFS, but this is a very deep change!
3650 */
3651static inline void
3652nfs4_transform_lock_offset(struct file_lock *lock)
3653{
3654    if (lock->fl_start < 0)
3655        lock->fl_start = OFFSET_MAX;
3656    if (lock->fl_end < 0)
3657        lock->fl_end = OFFSET_MAX;
3658}
3659
3660/* Hack!: For now, we're defining this just so we can use a pointer to it
3661 * as a unique cookie to identify our (NFSv4's) posix locks. */
3662static const struct lock_manager_operations nfsd_posix_mng_ops = {
3663};
3664
3665static inline void
3666nfs4_set_lock_denied(struct file_lock *fl, struct nfsd4_lock_denied *deny)
3667{
3668    struct nfs4_stateowner *sop;
3669
3670    if (fl->fl_lmops == &nfsd_posix_mng_ops) {
3671        sop = (struct nfs4_stateowner *) fl->fl_owner;
3672        kref_get(&sop->so_ref);
3673        deny->ld_sop = sop;
3674        deny->ld_clientid = sop->so_client->cl_clientid;
3675    } else {
3676        deny->ld_sop = NULL;
3677        deny->ld_clientid.cl_boot = 0;
3678        deny->ld_clientid.cl_id = 0;
3679    }
3680    deny->ld_start = fl->fl_start;
3681    deny->ld_length = NFS4_MAX_UINT64;
3682    if (fl->fl_end != NFS4_MAX_UINT64)
3683        deny->ld_length = fl->fl_end - fl->fl_start + 1;
3684    deny->ld_type = NFS4_READ_LT;
3685    if (fl->fl_type != F_RDLCK)
3686        deny->ld_type = NFS4_WRITE_LT;
3687}
3688
3689static struct nfs4_stateowner *
3690find_lockstateowner_str(struct inode *inode, clientid_t *clid,
3691        struct xdr_netobj *owner)
3692{
3693    unsigned int hashval = lock_ownerstr_hashval(inode, clid->cl_id, owner);
3694    struct nfs4_stateowner *op;
3695
3696    list_for_each_entry(op, &lock_ownerstr_hashtbl[hashval], so_strhash) {
3697        if (same_owner_str(op, owner, clid))
3698            return op;
3699    }
3700    return NULL;
3701}
3702
3703/*
3704 * Alloc a lock owner structure.
3705 * Called in nfsd4_lock - therefore, OPEN and OPEN_CONFIRM (if needed) has
3706 * occurred.
3707 *
3708 * strhashval = lock_ownerstr_hashval
3709 */
3710
3711static struct nfs4_stateowner *
3712alloc_init_lock_stateowner(unsigned int strhashval, struct nfs4_client *clp, struct nfs4_stateid *open_stp, struct nfsd4_lock *lock) {
3713    struct nfs4_stateowner *sop;
3714    struct nfs4_replay *rp;
3715    unsigned int idhashval;
3716
3717    if (!(sop = alloc_stateowner(&lock->lk_new_owner)))
3718        return NULL;
3719    idhashval = lockownerid_hashval(current_ownerid);
3720    INIT_LIST_HEAD(&sop->so_idhash);
3721    INIT_LIST_HEAD(&sop->so_strhash);
3722    INIT_LIST_HEAD(&sop->so_perclient);
3723    INIT_LIST_HEAD(&sop->so_stateids);
3724    INIT_LIST_HEAD(&sop->so_perstateid);
3725    INIT_LIST_HEAD(&sop->so_close_lru); /* not used */
3726    sop->so_time = 0;
3727    list_add(&sop->so_idhash, &lock_ownerid_hashtbl[idhashval]);
3728    list_add(&sop->so_strhash, &lock_ownerstr_hashtbl[strhashval]);
3729    list_add(&sop->so_perstateid, &open_stp->st_lockowners);
3730    sop->so_is_open_owner = 0;
3731    sop->so_id = current_ownerid++;
3732    sop->so_client = clp;
3733    /* It is the openowner seqid that will be incremented in encode in the
3734     * case of new lockowners; so increment the lock seqid manually: */
3735    sop->so_seqid = lock->lk_new_lock_seqid + 1;
3736    sop->so_confirmed = 1;
3737    rp = &sop->so_replay;
3738    rp->rp_status = nfserr_serverfault;
3739    rp->rp_buflen = 0;
3740    rp->rp_buf = rp->rp_ibuf;
3741    return sop;
3742}
3743
3744static struct nfs4_stateid *
3745alloc_init_lock_stateid(struct nfs4_stateowner *sop, struct nfs4_file *fp, struct nfs4_stateid *open_stp)
3746{
3747    struct nfs4_stateid *stp;
3748    unsigned int hashval = stateid_hashval(sop->so_id, fp->fi_id);
3749
3750    stp = nfs4_alloc_stateid();
3751    if (stp == NULL)
3752        goto out;
3753    INIT_LIST_HEAD(&stp->st_hash);
3754    INIT_LIST_HEAD(&stp->st_perfile);
3755    INIT_LIST_HEAD(&stp->st_perstateowner);
3756    INIT_LIST_HEAD(&stp->st_lockowners); /* not used */
3757    list_add(&stp->st_hash, &lockstateid_hashtbl[hashval]);
3758    list_add(&stp->st_perfile, &fp->fi_stateids);
3759    list_add(&stp->st_perstateowner, &sop->so_stateids);
3760    stp->st_stateowner = sop;
3761    get_nfs4_file(fp);
3762    stp->st_file = fp;
3763    stp->st_stateid.si_boot = boot_time;
3764    stp->st_stateid.si_stateownerid = sop->so_id;
3765    stp->st_stateid.si_fileid = fp->fi_id;
3766    stp->st_stateid.si_generation = 0;
3767    stp->st_access_bmap = 0;
3768    stp->st_deny_bmap = open_stp->st_deny_bmap;
3769    stp->st_openstp = open_stp;
3770
3771out:
3772    return stp;
3773}
3774
3775static int
3776check_lock_length(u64 offset, u64 length)
3777{
3778    return ((length == 0) || ((length != NFS4_MAX_UINT64) &&
3779         LOFF_OVERFLOW(offset, length)));
3780}
3781
3782static void get_lock_access(struct nfs4_stateid *lock_stp, u32 access)
3783{
3784    struct nfs4_file *fp = lock_stp->st_file;
3785    int oflag = nfs4_access_to_omode(access);
3786
3787    if (test_bit(access, &lock_stp->st_access_bmap))
3788        return;
3789    nfs4_file_get_access(fp, oflag);
3790    __set_bit(access, &lock_stp->st_access_bmap);
3791}
3792
3793/*
3794 * LOCK operation
3795 */
3796__be32
3797nfsd4_lock(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3798       struct nfsd4_lock *lock)
3799{
3800    struct nfs4_stateowner *open_sop = NULL;
3801    struct nfs4_stateowner *lock_sop = NULL;
3802    struct nfs4_stateid *lock_stp;
3803    struct nfs4_file *fp;
3804    struct file *filp = NULL;
3805    struct file_lock file_lock;
3806    struct file_lock conflock;
3807    __be32 status = 0;
3808    unsigned int strhashval;
3809    int err;
3810
3811    dprintk("NFSD: nfsd4_lock: start=%Ld length=%Ld\n",
3812        (long long) lock->lk_offset,
3813        (long long) lock->lk_length);
3814
3815    if (check_lock_length(lock->lk_offset, lock->lk_length))
3816         return nfserr_inval;
3817
3818    if ((status = fh_verify(rqstp, &cstate->current_fh,
3819                S_IFREG, NFSD_MAY_LOCK))) {
3820        dprintk("NFSD: nfsd4_lock: permission denied!\n");
3821        return status;
3822    }
3823
3824    nfs4_lock_state();
3825
3826    if (lock->lk_is_new) {
3827        /*
3828         * Client indicates that this is a new lockowner.
3829         * Use open owner and open stateid to create lock owner and
3830         * lock stateid.
3831         */
3832        struct nfs4_stateid *open_stp = NULL;
3833        
3834        status = nfserr_stale_clientid;
3835        if (!nfsd4_has_session(cstate) &&
3836            STALE_CLIENTID(&lock->lk_new_clientid))
3837            goto out;
3838
3839        /* validate and update open stateid and open seqid */
3840        status = nfs4_preprocess_seqid_op(cstate,
3841                        lock->lk_new_open_seqid,
3842                                &lock->lk_new_open_stateid,
3843                    OPEN_STATE,
3844                                &lock->lk_replay_owner, &open_stp,
3845                    lock);
3846        if (status)
3847            goto out;
3848        open_sop = lock->lk_replay_owner;
3849        /* create lockowner and lock stateid */
3850        fp = open_stp->st_file;
3851        strhashval = lock_ownerstr_hashval(fp->fi_inode,
3852                open_sop->so_client->cl_clientid.cl_id,
3853                &lock->v.new.owner);
3854        /* XXX: Do we need to check for duplicate stateowners on
3855         * the same file, or should they just be allowed (and
3856         * create new stateids)? */
3857        status = nfserr_resource;
3858        lock_sop = alloc_init_lock_stateowner(strhashval,
3859                open_sop->so_client, open_stp, lock);
3860        if (lock_sop == NULL)
3861            goto out;
3862        lock_stp = alloc_init_lock_stateid(lock_sop, fp, open_stp);
3863        if (lock_stp == NULL)
3864            goto out;
3865    } else {
3866        /* lock (lock owner + lock stateid) already exists */
3867        status = nfs4_preprocess_seqid_op(cstate,
3868                       lock->lk_old_lock_seqid,
3869                       &lock->lk_old_lock_stateid,
3870                       LOCK_STATE,
3871                       &lock->lk_replay_owner, &lock_stp, lock);
3872        if (status)
3873            goto out;
3874        lock_sop = lock->lk_replay_owner;
3875        fp = lock_stp->st_file;
3876    }
3877    /* lock->lk_replay_owner and lock_stp have been created or found */
3878
3879    status = nfserr_grace;
3880    if (locks_in_grace() && !lock->lk_reclaim)
3881        goto out;
3882    status = nfserr_no_grace;
3883    if (!locks_in_grace() && lock->lk_reclaim)
3884        goto out;
3885
3886    locks_init_lock(&file_lock);
3887    switch (lock->lk_type) {
3888        case NFS4_READ_LT:
3889        case NFS4_READW_LT:
3890            filp = find_readable_file(lock_stp->st_file);
3891            if (filp)
3892                get_lock_access(lock_stp, NFS4_SHARE_ACCESS_READ);
3893            file_lock.fl_type = F_RDLCK;
3894            break;
3895        case NFS4_WRITE_LT:
3896        case NFS4_WRITEW_LT:
3897            filp = find_writeable_file(lock_stp->st_file);
3898            if (filp)
3899                get_lock_access(lock_stp, NFS4_SHARE_ACCESS_WRITE);
3900            file_lock.fl_type = F_WRLCK;
3901            break;
3902        default:
3903            status = nfserr_inval;
3904        goto out;
3905    }
3906    if (!filp) {
3907        status = nfserr_openmode;
3908        goto out;
3909    }
3910    file_lock.fl_owner = (fl_owner_t)lock_sop;
3911    file_lock.fl_pid = current->tgid;
3912    file_lock.fl_file = filp;
3913    file_lock.fl_flags = FL_POSIX;
3914    file_lock.fl_lmops = &nfsd_posix_mng_ops;
3915
3916    file_lock.fl_start = lock->lk_offset;
3917    file_lock.fl_end = last_byte_offset(lock->lk_offset, lock->lk_length);
3918    nfs4_transform_lock_offset(&file_lock);
3919
3920    /*
3921    * Try to lock the file in the VFS.
3922    * Note: locks.c uses the BKL to protect the inode's lock list.
3923    */
3924
3925    err = vfs_lock_file(filp, F_SETLK, &file_lock, &conflock);
3926    switch (-err) {
3927    case 0: /* success! */
3928        update_stateid(&lock_stp->st_stateid);
3929        memcpy(&lock->lk_resp_stateid, &lock_stp->st_stateid,
3930                sizeof(stateid_t));
3931        status = 0;
3932        break;
3933    case (EAGAIN): /* conflock holds conflicting lock */
3934        status = nfserr_denied;
3935        dprintk("NFSD: nfsd4_lock: conflicting lock found!\n");
3936        nfs4_set_lock_denied(&conflock, &lock->lk_denied);
3937        break;
3938    case (EDEADLK):
3939        status = nfserr_deadlock;
3940        break;
3941    default:
3942        dprintk("NFSD: nfsd4_lock: vfs_lock_file() failed! status %d\n",err);
3943        status = nfserr_resource;
3944        break;
3945    }
3946out:
3947    if (status && lock->lk_is_new && lock_sop)
3948        release_lockowner(lock_sop);
3949    if (lock->lk_replay_owner) {
3950        nfs4_get_stateowner(lock->lk_replay_owner);
3951        cstate->replay_owner = lock->lk_replay_owner;
3952    }
3953    nfs4_unlock_state();
3954    return status;
3955}
3956
3957/*
3958 * The NFSv4 spec allows a client to do a LOCKT without holding an OPEN,
3959 * so we do a temporary open here just to get an open file to pass to
3960 * vfs_test_lock. (Arguably perhaps test_lock should be done with an
3961 * inode operation.)
3962 */
3963static int nfsd_test_lock(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file_lock *lock)
3964{
3965    struct file *file;
3966    int err;
3967
3968    err = nfsd_open(rqstp, fhp, S_IFREG, NFSD_MAY_READ, &file);
3969    if (err)
3970        return err;
3971    err = vfs_test_lock(file, lock);
3972    nfsd_close(file);
3973    return err;
3974}
3975
3976/*
3977 * LOCKT operation
3978 */
3979__be32
3980nfsd4_lockt(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3981        struct nfsd4_lockt *lockt)
3982{
3983    struct inode *inode;
3984    struct file_lock file_lock;
3985    int error;
3986    __be32 status;
3987
3988    if (locks_in_grace())
3989        return nfserr_grace;
3990
3991    if (check_lock_length(lockt->lt_offset, lockt->lt_length))
3992         return nfserr_inval;
3993
3994    lockt->lt_stateowner = NULL;
3995    nfs4_lock_state();
3996
3997    status = nfserr_stale_clientid;
3998    if (!nfsd4_has_session(cstate) && STALE_CLIENTID(&lockt->lt_clientid))
3999        goto out;
4000
4001    if ((status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0))) {
4002        dprintk("NFSD: nfsd4_lockt: fh_verify() failed!\n");
4003        if (status == nfserr_symlink)
4004            status = nfserr_inval;
4005        goto out;
4006    }
4007
4008    inode = cstate->current_fh.fh_dentry->d_inode;
4009    locks_init_lock(&file_lock);
4010    switch (lockt->lt_type) {
4011        case NFS4_READ_LT:
4012        case NFS4_READW_LT:
4013            file_lock.fl_type = F_RDLCK;
4014        break;
4015        case NFS4_WRITE_LT:
4016        case NFS4_WRITEW_LT:
4017            file_lock.fl_type = F_WRLCK;
4018        break;
4019        default:
4020            dprintk("NFSD: nfs4_lockt: bad lock type!\n");
4021            status = nfserr_inval;
4022        goto out;
4023    }
4024
4025    lockt->lt_stateowner = find_lockstateowner_str(inode,
4026            &lockt->lt_clientid, &lockt->lt_owner);
4027    if (lockt->lt_stateowner)
4028        file_lock.fl_owner = (fl_owner_t)lockt->lt_stateowner;
4029    file_lock.fl_pid = current->tgid;
4030    file_lock.fl_flags = FL_POSIX;
4031
4032    file_lock.fl_start = lockt->lt_offset;
4033    file_lock.fl_end = last_byte_offset(lockt->lt_offset, lockt->lt_length);
4034
4035    nfs4_transform_lock_offset(&file_lock);
4036
4037    status = nfs_ok;
4038    error = nfsd_test_lock(rqstp, &cstate->current_fh, &file_lock);
4039    if (error) {
4040        status = nfserrno(error);
4041        goto out;
4042    }
4043    if (file_lock.fl_type != F_UNLCK) {
4044        status = nfserr_denied;
4045        nfs4_set_lock_denied(&file_lock, &lockt->lt_denied);
4046    }
4047out:
4048    nfs4_unlock_state();
4049    return status;
4050}
4051
4052__be32
4053nfsd4_locku(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
4054        struct nfsd4_locku *locku)
4055{
4056    struct nfs4_stateid *stp;
4057    struct file *filp = NULL;
4058    struct file_lock file_lock;
4059    __be32 status;
4060    int err;
4061                                
4062    dprintk("NFSD: nfsd4_locku: start=%Ld length=%Ld\n",
4063        (long long) locku->lu_offset,
4064        (long long) locku->lu_length);
4065
4066    if (check_lock_length(locku->lu_offset, locku->lu_length))
4067         return nfserr_inval;
4068
4069    nfs4_lock_state();
4070                                            
4071    if ((status = nfs4_preprocess_seqid_op(cstate,
4072                    locku->lu_seqid,
4073                    &locku->lu_stateid,
4074                    LOCK_STATE,
4075                    &locku->lu_stateowner, &stp, NULL)))
4076        goto out;
4077
4078    filp = find_any_file(stp->st_file);
4079    if (!filp) {
4080        status = nfserr_lock_range;
4081        goto out;
4082    }
4083    BUG_ON(!filp);
4084    locks_init_lock(&file_lock);
4085    file_lock.fl_type = F_UNLCK;
4086    file_lock.fl_owner = (fl_owner_t) locku->lu_stateowner;
4087    file_lock.fl_pid = current->tgid;
4088    file_lock.fl_file = filp;
4089    file_lock.fl_flags = FL_POSIX;
4090    file_lock.fl_lmops = &nfsd_posix_mng_ops;
4091    file_lock.fl_start = locku->lu_offset;
4092
4093    file_lock.fl_end = last_byte_offset(locku->lu_offset, locku->lu_length);
4094    nfs4_transform_lock_offset(&file_lock);
4095
4096    /*
4097    * Try to unlock the file in the VFS.
4098    */
4099    err = vfs_lock_file(filp, F_SETLK, &file_lock, NULL);
4100    if (err) {
4101        dprintk("NFSD: nfs4_locku: vfs_lock_file failed!\n");
4102        goto out_nfserr;
4103    }
4104    /*
4105    * OK, unlock succeeded; the only thing left to do is update the stateid.
4106    */
4107    update_stateid(&stp->st_stateid);
4108    memcpy(&locku->lu_stateid, &stp->st_stateid, sizeof(stateid_t));
4109
4110out:
4111    if (locku->lu_stateowner) {
4112        nfs4_get_stateowner(locku->lu_stateowner);
4113        cstate->replay_owner = locku->lu_stateowner;
4114    }
4115    nfs4_unlock_state();
4116    return status;
4117
4118out_nfserr:
4119    status = nfserrno(err);
4120    goto out;
4121}
4122
4123/*
4124 * returns
4125 * 1: locks held by lockowner
4126 * 0: no locks held by lockowner
4127 */
4128static int
4129check_for_locks(struct nfs4_file *filp, struct nfs4_stateowner *lowner)
4130{
4131    struct file_lock **flpp;
4132    struct inode *inode = filp->fi_inode;
4133    int status = 0;
4134
4135    lock_flocks();
4136    for (flpp = &inode->i_flock; *flpp != NULL; flpp = &(*flpp)->fl_next) {
4137        if ((*flpp)->fl_owner == (fl_owner_t)lowner) {
4138            status = 1;
4139            goto out;
4140        }
4141    }
4142out:
4143    unlock_flocks();
4144    return status;
4145}
4146
4147__be32
4148nfsd4_release_lockowner(struct svc_rqst *rqstp,
4149            struct nfsd4_compound_state *cstate,
4150            struct nfsd4_release_lockowner *rlockowner)
4151{
4152    clientid_t *clid = &rlockowner->rl_clientid;
4153    struct nfs4_stateowner *sop;
4154    struct nfs4_stateid *stp;
4155    struct xdr_netobj *owner = &rlockowner->rl_owner;
4156    struct list_head matches;
4157    int i;
4158    __be32 status;
4159
4160    dprintk("nfsd4_release_lockowner clientid: (%08x/%08x):\n",
4161        clid->cl_boot, clid->cl_id);
4162
4163    /* XXX check for lease expiration */
4164
4165    status = nfserr_stale_clientid;
4166    if (STALE_CLIENTID(clid))
4167        return status;
4168
4169    nfs4_lock_state();
4170
4171    status = nfserr_locks_held;
4172    /* XXX: we're doing a linear search through all the lockowners.
4173     * Yipes! For now we'll just hope clients aren't really using
4174     * release_lockowner much, but eventually we have to fix these
4175     * data structures. */
4176    INIT_LIST_HEAD(&matches);
4177    for (i = 0; i < LOCK_HASH_SIZE; i++) {
4178        list_for_each_entry(sop, &lock_ownerid_hashtbl[i], so_idhash) {
4179            if (!same_owner_str(sop, owner, clid))
4180                continue;
4181            list_for_each_entry(stp, &sop->so_stateids,
4182                    st_perstateowner) {
4183                if (check_for_locks(stp->st_file, sop))
4184                    goto out;
4185                /* Note: so_perclient unused for lockowners,
4186                 * so it's OK to fool with here. */
4187                list_add(&sop->so_perclient, &matches);
4188            }
4189        }
4190    }
4191    /* Clients probably won't expect us to return with some (but not all)
4192     * of the lockowner state released; so don't release any until all
4193     * have been checked. */
4194    status = nfs_ok;
4195    while (!list_empty(&matches)) {
4196        sop = list_entry(matches.next, struct nfs4_stateowner,
4197                                so_perclient);
4198        /* unhash_stateowner deletes so_perclient only
4199         * for openowners. */
4200        list_del(&sop->so_perclient);
4201        release_lockowner(sop);
4202    }
4203out:
4204    nfs4_unlock_state();
4205    return status;
4206}
4207
4208static inline struct nfs4_client_reclaim *
4209alloc_reclaim(void)
4210{
4211    return kmalloc(sizeof(struct nfs4_client_reclaim), GFP_KERNEL);
4212}
4213
4214int
4215nfs4_has_reclaimed_state(const char *name, bool use_exchange_id)
4216{
4217    unsigned int strhashval = clientstr_hashval(name);
4218    struct nfs4_client *clp;
4219
4220    clp = find_confirmed_client_by_str(name, strhashval);
4221    return clp ? 1 : 0;
4222}
4223
4224/*
4225 * failure => all reset bets are off, nfserr_no_grace...
4226 */
4227int
4228nfs4_client_to_reclaim(const char *name)
4229{
4230    unsigned int strhashval;
4231    struct nfs4_client_reclaim *crp = NULL;
4232
4233    dprintk("NFSD nfs4_client_to_reclaim NAME: %.*s\n", HEXDIR_LEN, name);
4234    crp = alloc_reclaim();
4235    if (!crp)
4236        return 0;
4237    strhashval = clientstr_hashval(name);
4238    INIT_LIST_HEAD(&crp->cr_strhash);
4239    list_add(&crp->cr_strhash, &reclaim_str_hashtbl[strhashval]);
4240    memcpy(crp->cr_recdir, name, HEXDIR_LEN);
4241    reclaim_str_hashtbl_size++;
4242    return 1;
4243}
4244
4245static void
4246nfs4_release_reclaim(void)
4247{
4248    struct nfs4_client_reclaim *crp = NULL;
4249    int i;
4250
4251    for (i = 0; i < CLIENT_HASH_SIZE; i++) {
4252        while (!list_empty(&reclaim_str_hashtbl[i])) {
4253            crp = list_entry(reclaim_str_hashtbl[i].next,
4254                            struct nfs4_client_reclaim, cr_strhash);
4255            list_del(&crp->cr_strhash);
4256            kfree(crp);
4257            reclaim_str_hashtbl_size--;
4258        }
4259    }
4260    BUG_ON(reclaim_str_hashtbl_size);
4261}
4262
4263/*
4264 * called from OPEN, CLAIM_PREVIOUS with a new clientid. */
4265static struct nfs4_client_reclaim *
4266nfs4_find_reclaim_client(clientid_t *clid)
4267{
4268    unsigned int strhashval;
4269    struct nfs4_client *clp;
4270    struct nfs4_client_reclaim *crp = NULL;
4271
4272
4273    /* find clientid in conf_id_hashtbl */
4274    clp = find_confirmed_client(clid);
4275    if (clp == NULL)
4276        return NULL;
4277
4278    dprintk("NFSD: nfs4_find_reclaim_client for %.*s with recdir %s\n",
4279                    clp->cl_name.len, clp->cl_name.data,
4280                clp->cl_recdir);
4281
4282    /* find clp->cl_name in reclaim_str_hashtbl */
4283    strhashval = clientstr_hashval(clp->cl_recdir);
4284    list_for_each_entry(crp, &reclaim_str_hashtbl[strhashval], cr_strhash) {
4285        if (same_name(crp->cr_recdir, clp->cl_recdir)) {
4286            return crp;
4287        }
4288    }
4289    return NULL;
4290}
4291
4292/*
4293* Called from OPEN. Look for clientid in reclaim list.
4294*/
4295__be32
4296nfs4_check_open_reclaim(clientid_t *clid)
4297{
4298    return nfs4_find_reclaim_client(clid) ? nfs_ok : nfserr_reclaim_bad;
4299}
4300
4301/* initialization to perform at module load time: */
4302
4303int
4304nfs4_state_init(void)
4305{
4306    int i, status;
4307
4308    status = nfsd4_init_slabs();
4309    if (status)
4310        return status;
4311    for (i = 0; i < CLIENT_HASH_SIZE; i++) {
4312        INIT_LIST_HEAD(&conf_id_hashtbl[i]);
4313        INIT_LIST_HEAD(&conf_str_hashtbl[i]);
4314        INIT_LIST_HEAD(&unconf_str_hashtbl[i]);
4315        INIT_LIST_HEAD(&unconf_id_hashtbl[i]);
4316        INIT_LIST_HEAD(&reclaim_str_hashtbl[i]);
4317    }
4318    for (i = 0; i < SESSION_HASH_SIZE; i++)
4319        INIT_LIST_HEAD(&sessionid_hashtbl[i]);
4320    for (i = 0; i < FILE_HASH_SIZE; i++) {
4321        INIT_LIST_HEAD(&file_hashtbl[i]);
4322    }
4323    for (i = 0; i < OWNER_HASH_SIZE; i++) {
4324        INIT_LIST_HEAD(&ownerstr_hashtbl[i]);
4325        INIT_LIST_HEAD(&ownerid_hashtbl[i]);
4326    }
4327    for (i = 0; i < STATEID_HASH_SIZE; i++) {
4328        INIT_LIST_HEAD(&stateid_hashtbl[i]);
4329        INIT_LIST_HEAD(&lockstateid_hashtbl[i]);
4330    }
4331    for (i = 0; i < LOCK_HASH_SIZE; i++) {
4332        INIT_LIST_HEAD(&lock_ownerid_hashtbl[i]);
4333        INIT_LIST_HEAD(&lock_ownerstr_hashtbl[i]);
4334    }
4335    memset(&onestateid, ~0, sizeof(stateid_t));
4336    INIT_LIST_HEAD(&close_lru);
4337    INIT_LIST_HEAD(&client_lru);
4338    INIT_LIST_HEAD(&del_recall_lru);
4339    reclaim_str_hashtbl_size = 0;
4340    return 0;
4341}
4342
4343static void
4344nfsd4_load_reboot_recovery_data(void)
4345{
4346    int status;
4347
4348    nfs4_lock_state();
4349    nfsd4_init_recdir(user_recovery_dirname);
4350    status = nfsd4_recdir_load();
4351    nfs4_unlock_state();
4352    if (status)
4353        printk("NFSD: Failure reading reboot recovery data\n");
4354}
4355
4356/*
4357 * Since the lifetime of a delegation isn't limited to that of an open, a
4358 * client may quite reasonably hang on to a delegation as long as it has
4359 * the inode cached. This becomes an obvious problem the first time a
4360 * client's inode cache approaches the size of the server's total memory.
4361 *
4362 * For now we avoid this problem by imposing a hard limit on the number
4363 * of delegations, which varies according to the server's memory size.
4364 */
4365static void
4366set_max_delegations(void)
4367{
4368    /*
4369     * Allow at most 4 delegations per megabyte of RAM. Quick
4370     * estimates suggest that in the worst case (where every delegation
4371     * is for a different inode), a delegation could take about 1.5K,
4372     * giving a worst case usage of about 6% of memory.
4373     */
4374    max_delegations = nr_free_buffer_pages() >> (20 - 2 - PAGE_SHIFT);
4375}
4376
4377/* initialization to perform when the nfsd service is started: */
4378
4379static int
4380__nfs4_state_start(void)
4381{
4382    int ret;
4383
4384    boot_time = get_seconds();
4385    locks_start_grace(&nfsd4_manager);
4386    printk(KERN_INFO "NFSD: starting %ld-second grace period\n",
4387           nfsd4_grace);
4388    ret = set_callback_cred();
4389    if (ret)
4390        return -ENOMEM;
4391    laundry_wq = create_singlethread_workqueue("nfsd4");
4392    if (laundry_wq == NULL)
4393        return -ENOMEM;
4394    ret = nfsd4_create_callback_queue();
4395    if (ret)
4396        goto out_free_laundry;
4397    queue_delayed_work(laundry_wq, &laundromat_work, nfsd4_grace * HZ);
4398    set_max_delegations();
4399    return 0;
4400out_free_laundry:
4401    destroy_workqueue(laundry_wq);
4402    return ret;
4403}
4404
4405int
4406nfs4_state_start(void)
4407{
4408    nfsd4_load_reboot_recovery_data();
4409    return __nfs4_state_start();
4410}
4411
4412static void
4413__nfs4_state_shutdown(void)
4414{
4415    int i;
4416    struct nfs4_client *clp = NULL;
4417    struct nfs4_delegation *dp = NULL;
4418    struct list_head *pos, *next, reaplist;
4419
4420    for (i = 0; i < CLIENT_HASH_SIZE; i++) {
4421        while (!list_empty(&conf_id_hashtbl[i])) {
4422            clp = list_entry(conf_id_hashtbl[i].next, struct nfs4_client, cl_idhash);
4423            expire_client(clp);
4424        }
4425        while (!list_empty(&unconf_str_hashtbl[i])) {
4426            clp = list_entry(unconf_str_hashtbl[i].next, struct nfs4_client, cl_strhash);
4427            expire_client(clp);
4428        }
4429    }
4430    INIT_LIST_HEAD(&reaplist);
4431    spin_lock(&recall_lock);
4432    list_for_each_safe(pos, next, &del_recall_lru) {
4433        dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
4434        list_move(&dp->dl_recall_lru, &reaplist);
4435    }
4436    spin_unlock(&recall_lock);
4437    list_for_each_safe(pos, next, &reaplist) {
4438        dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
4439        list_del_init(&dp->dl_recall_lru);
4440        unhash_delegation(dp);
4441    }
4442
4443    nfsd4_shutdown_recdir();
4444}
4445
4446void
4447nfs4_state_shutdown(void)
4448{
4449    cancel_delayed_work_sync(&laundromat_work);
4450    destroy_workqueue(laundry_wq);
4451    locks_end_grace(&nfsd4_manager);
4452    nfs4_lock_state();
4453    nfs4_release_reclaim();
4454    __nfs4_state_shutdown();
4455    nfs4_unlock_state();
4456    nfsd4_destroy_callback_queue();
4457}
4458
4459/*
4460 * user_recovery_dirname is protected by the nfsd_mutex since it's only
4461 * accessed when nfsd is starting.
4462 */
4463static void
4464nfs4_set_recdir(char *recdir)
4465{
4466    strcpy(user_recovery_dirname, recdir);
4467}
4468
4469/*
4470 * Change the NFSv4 recovery directory to recdir.
4471 */
4472int
4473nfs4_reset_recoverydir(char *recdir)
4474{
4475    int status;
4476    struct path path;
4477
4478    status = kern_path(recdir, LOOKUP_FOLLOW, &path);
4479    if (status)
4480        return status;
4481    status = -ENOTDIR;
4482    if (S_ISDIR(path.dentry->d_inode->i_mode)) {
4483        nfs4_set_recdir(recdir);
4484        status = 0;
4485    }
4486    path_put(&path);
4487    return status;
4488}
4489
4490char *
4491nfs4_recoverydir(void)
4492{
4493    return user_recovery_dirname;
4494}
4495

Archive Download this file



interactive