2 * Modifications for Lustre
4 * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
6 * Copyright (c) 2012, 2014, Intel Corporation.
8 * Author: Eric Mei <ericm@clusterfs.com>
12 * Neil Brown <neilb@cse.unsw.edu.au>
13 * J. Bruce Fields <bfields@umich.edu>
14 * Andy Adamson <andros@umich.edu>
15 * Dug Song <dugsong@monkey.org>
17 * RPCSEC_GSS server authentication.
18 * This implements RPCSEC_GSS as defined in rfc2203 (rpcsec_gss) and rfc2078
21 * The RPCSEC_GSS involves three stages:
24 * 3/ context destruction
26 * Context creation is handled largely by upcalls to user-space.
27 * In particular, GSS_Accept_sec_context is handled by an upcall
28 * Data exchange is handled entirely within the kernel
29 * In particular, GSS_GetMIC, GSS_VerifyMIC, GSS_Seal, GSS_Unseal are in-kernel.
30 * Context destruction is handled in-kernel
31 * GSS_Delete_sec_context is in-kernel
33 * Context creation is initiated by a RPCSEC_GSS_INIT request arriving.
34 * The context handle and gss_token are used as a key into the rpcsec_init cache.
35 * The content of this cache includes some of the outputs of GSS_Accept_sec_context,
36 * being major_status, minor_status, context_handle, reply_token.
37 * These are sent back to the client.
38 * Sequence window management is handled by the kernel. The window size if currently
39 * a compile time constant.
41 * When user-space is happy that a context is established, it places an entry
42 * in the rpcsec_context cache. The key for this cache is the context_handle.
43 * The content includes:
44 * uid/gidlist - for determining access rights
46 * mechanism specific information, such as a key
50 #define DEBUG_SUBSYSTEM S_SEC
51 #include <linux/types.h>
52 #include <linux/init.h>
53 #include <linux/module.h>
54 #include <linux/slab.h>
55 #include <linux/hash.h>
56 #include <linux/mutex.h>
57 #include <linux/sunrpc/cache.h>
61 #include <obd_class.h>
62 #include <obd_support.h>
63 #include <lustre/lustre_idl.h>
64 #include <lustre_import.h>
65 #include <lustre_net.h>
66 #include <lustre_nodemap.h>
67 #include <lustre_sec.h>
70 #include "gss_internal.h"
73 #define GSS_SVC_UPCALL_TIMEOUT (20)
75 static spinlock_t __ctx_index_lock;
76 static __u64 __ctx_index;
78 __u64 gss_get_next_ctx_index(void)
82 spin_lock(&__ctx_index_lock);
84 spin_unlock(&__ctx_index_lock);
89 static inline unsigned long hash_mem(char *buf, int length, int bits)
91 unsigned long hash = 0;
106 if ((len & (BITS_PER_LONG/8-1)) == 0)
107 hash = hash_long(hash^l, BITS_PER_LONG);
110 return hash >> (BITS_PER_LONG - bits);
113 /* This compatibility can be removed once kernel 3.3 is used,
114 * since cache_register_net/cache_unregister_net are exported.
115 * Note that since kernel 3.4 cache_register and cache_unregister
118 static inline int _cache_register_net(struct cache_detail *cd, struct net *net)
120 #ifdef HAVE_CACHE_REGISTER
121 return cache_register(cd);
123 return cache_register_net(cd, net);
126 static inline void _cache_unregister_net(struct cache_detail *cd,
129 #ifdef HAVE_CACHE_REGISTER
130 cache_unregister(cd);
132 cache_unregister_net(cd, net);
135 /****************************************
136 * rpc sec init (rsi) cache *
137 ****************************************/
139 #define RSI_HASHBITS (6)
140 #define RSI_HASHMAX (1 << RSI_HASHBITS)
141 #define RSI_HASHMASK (RSI_HASHMAX - 1)
147 char nm_name[LUSTRE_NODEMAP_NAME_LENGTH + 1];
148 wait_queue_head_t waitq;
149 rawobj_t in_handle, in_token;
150 rawobj_t out_handle, out_token;
151 int major_status, minor_status;
154 #ifdef HAVE_CACHE_HEAD_HLIST
155 static struct hlist_head rsi_table[RSI_HASHMAX];
157 static struct cache_head *rsi_table[RSI_HASHMAX];
159 static struct cache_detail rsi_cache;
160 static struct rsi *rsi_update(struct rsi *new, struct rsi *old);
161 static struct rsi *rsi_lookup(struct rsi *item);
163 static inline int rsi_hash(struct rsi *item)
165 return hash_mem((char *)item->in_handle.data, item->in_handle.len,
167 hash_mem((char *)item->in_token.data, item->in_token.len,
171 static inline int __rsi_match(struct rsi *item, struct rsi *tmp)
173 return (rawobj_equal(&item->in_handle, &tmp->in_handle) &&
174 rawobj_equal(&item->in_token, &tmp->in_token));
177 static void rsi_free(struct rsi *rsi)
179 rawobj_free(&rsi->in_handle);
180 rawobj_free(&rsi->in_token);
181 rawobj_free(&rsi->out_handle);
182 rawobj_free(&rsi->out_token);
185 /* See handle_channel_req() userspace for where the upcall data is read */
186 static void rsi_request(struct cache_detail *cd,
187 struct cache_head *h,
188 char **bpp, int *blen)
190 struct rsi *rsi = container_of(h, struct rsi, h);
193 /* if in_handle is null, provide kernel suggestion */
194 if (rsi->in_handle.len == 0)
195 index = gss_get_next_ctx_index();
197 qword_addhex(bpp, blen, (char *) &rsi->lustre_svc,
198 sizeof(rsi->lustre_svc));
199 qword_addhex(bpp, blen, (char *) &rsi->nid, sizeof(rsi->nid));
200 qword_addhex(bpp, blen, (char *) &index, sizeof(index));
201 qword_addhex(bpp, blen, (char *) rsi->nm_name,
202 strlen(rsi->nm_name) + 1);
203 qword_addhex(bpp, blen, rsi->in_handle.data, rsi->in_handle.len);
204 qword_addhex(bpp, blen, rsi->in_token.data, rsi->in_token.len);
208 #ifdef HAVE_SUNRPC_UPCALL_HAS_3ARGS
209 static int rsi_upcall(struct cache_detail *cd, struct cache_head *h)
211 return sunrpc_cache_pipe_upcall(cd, h, rsi_request);
215 static int rsi_upcall(struct cache_detail *cd, struct cache_head *h)
217 return sunrpc_cache_pipe_upcall(cd, h);
221 static inline void __rsi_init(struct rsi *new, struct rsi *item)
223 new->out_handle = RAWOBJ_EMPTY;
224 new->out_token = RAWOBJ_EMPTY;
226 new->in_handle = item->in_handle;
227 item->in_handle = RAWOBJ_EMPTY;
228 new->in_token = item->in_token;
229 item->in_token = RAWOBJ_EMPTY;
231 new->lustre_svc = item->lustre_svc;
232 new->nid = item->nid;
233 memcpy(new->nm_name, item->nm_name, sizeof(item->nm_name));
234 init_waitqueue_head(&new->waitq);
237 static inline void __rsi_update(struct rsi *new, struct rsi *item)
239 LASSERT(new->out_handle.len == 0);
240 LASSERT(new->out_token.len == 0);
242 new->out_handle = item->out_handle;
243 item->out_handle = RAWOBJ_EMPTY;
244 new->out_token = item->out_token;
245 item->out_token = RAWOBJ_EMPTY;
247 new->major_status = item->major_status;
248 new->minor_status = item->minor_status;
251 static void rsi_put(struct kref *ref)
253 struct rsi *rsi = container_of(ref, struct rsi, h.ref);
255 #ifdef HAVE_CACHE_HEAD_HLIST
256 LASSERT(rsi->h.cache_list.next == NULL);
258 LASSERT(rsi->h.next == NULL);
264 static int rsi_match(struct cache_head *a, struct cache_head *b)
266 struct rsi *item = container_of(a, struct rsi, h);
267 struct rsi *tmp = container_of(b, struct rsi, h);
269 return __rsi_match(item, tmp);
272 static void rsi_init(struct cache_head *cnew, struct cache_head *citem)
274 struct rsi *new = container_of(cnew, struct rsi, h);
275 struct rsi *item = container_of(citem, struct rsi, h);
277 __rsi_init(new, item);
280 static void update_rsi(struct cache_head *cnew, struct cache_head *citem)
282 struct rsi *new = container_of(cnew, struct rsi, h);
283 struct rsi *item = container_of(citem, struct rsi, h);
285 __rsi_update(new, item);
288 static struct cache_head *rsi_alloc(void)
299 static int rsi_parse(struct cache_detail *cd, char *mesg, int mlen)
304 struct rsi rsii, *rsip = NULL;
306 int status = -EINVAL;
310 memset(&rsii, 0, sizeof(rsii));
313 len = qword_get(&mesg, buf, mlen);
316 if (rawobj_alloc(&rsii.in_handle, buf, len)) {
322 len = qword_get(&mesg, buf, mlen);
325 if (rawobj_alloc(&rsii.in_token, buf, len)) {
330 rsip = rsi_lookup(&rsii);
336 expiry = get_expiry(&mesg);
340 len = qword_get(&mesg, buf, mlen);
345 rsii.major_status = simple_strtol(buf, &ep, 10);
350 len = qword_get(&mesg, buf, mlen);
353 rsii.minor_status = simple_strtol(buf, &ep, 10);
358 len = qword_get(&mesg, buf, mlen);
361 if (rawobj_alloc(&rsii.out_handle, buf, len)) {
367 len = qword_get(&mesg, buf, mlen);
370 if (rawobj_alloc(&rsii.out_token, buf, len)) {
375 rsii.h.expiry_time = expiry;
376 rsip = rsi_update(&rsii, rsip);
381 wake_up_all(&rsip->waitq);
382 cache_put(&rsip->h, &rsi_cache);
388 CERROR("rsi parse error %d\n", status);
392 static struct cache_detail rsi_cache = {
393 .hash_size = RSI_HASHMAX,
394 .hash_table = rsi_table,
395 .name = "auth.sptlrpc.init",
396 .cache_put = rsi_put,
397 #ifndef HAVE_SUNRPC_UPCALL_HAS_3ARGS
398 .cache_request = rsi_request,
400 .cache_upcall = rsi_upcall,
401 .cache_parse = rsi_parse,
404 .update = update_rsi,
408 static struct rsi *rsi_lookup(struct rsi *item)
410 struct cache_head *ch;
411 int hash = rsi_hash(item);
413 ch = sunrpc_cache_lookup(&rsi_cache, &item->h, hash);
415 return container_of(ch, struct rsi, h);
420 static struct rsi *rsi_update(struct rsi *new, struct rsi *old)
422 struct cache_head *ch;
423 int hash = rsi_hash(new);
425 ch = sunrpc_cache_update(&rsi_cache, &new->h, &old->h, hash);
427 return container_of(ch, struct rsi, h);
432 /****************************************
433 * rpc sec context (rsc) cache *
434 ****************************************/
436 #define RSC_HASHBITS (10)
437 #define RSC_HASHMAX (1 << RSC_HASHBITS)
438 #define RSC_HASHMASK (RSC_HASHMAX - 1)
442 struct obd_device *target;
444 struct gss_svc_ctx ctx;
447 #ifdef HAVE_CACHE_HEAD_HLIST
448 static struct hlist_head rsc_table[RSC_HASHMAX];
450 static struct cache_head *rsc_table[RSC_HASHMAX];
452 static struct cache_detail rsc_cache;
453 static struct rsc *rsc_update(struct rsc *new, struct rsc *old);
454 static struct rsc *rsc_lookup(struct rsc *item);
456 static void rsc_free(struct rsc *rsci)
458 rawobj_free(&rsci->handle);
459 rawobj_free(&rsci->ctx.gsc_rvs_hdl);
460 lgss_delete_sec_context(&rsci->ctx.gsc_mechctx);
463 static inline int rsc_hash(struct rsc *rsci)
465 return hash_mem((char *)rsci->handle.data,
466 rsci->handle.len, RSC_HASHBITS);
469 static inline int __rsc_match(struct rsc *new, struct rsc *tmp)
471 return rawobj_equal(&new->handle, &tmp->handle);
474 static inline void __rsc_init(struct rsc *new, struct rsc *tmp)
476 new->handle = tmp->handle;
477 tmp->handle = RAWOBJ_EMPTY;
480 memset(&new->ctx, 0, sizeof(new->ctx));
481 new->ctx.gsc_rvs_hdl = RAWOBJ_EMPTY;
484 static inline void __rsc_update(struct rsc *new, struct rsc *tmp)
487 tmp->ctx.gsc_rvs_hdl = RAWOBJ_EMPTY;
488 tmp->ctx.gsc_mechctx = NULL;
490 memset(&new->ctx.gsc_seqdata, 0, sizeof(new->ctx.gsc_seqdata));
491 spin_lock_init(&new->ctx.gsc_seqdata.ssd_lock);
494 static void rsc_put(struct kref *ref)
496 struct rsc *rsci = container_of(ref, struct rsc, h.ref);
498 #ifdef HAVE_CACHE_HEAD_HLIST
499 LASSERT(rsci->h.cache_list.next == NULL);
501 LASSERT(rsci->h.next == NULL);
507 static int rsc_match(struct cache_head *a, struct cache_head *b)
509 struct rsc *new = container_of(a, struct rsc, h);
510 struct rsc *tmp = container_of(b, struct rsc, h);
512 return __rsc_match(new, tmp);
515 static void rsc_init(struct cache_head *cnew, struct cache_head *ctmp)
517 struct rsc *new = container_of(cnew, struct rsc, h);
518 struct rsc *tmp = container_of(ctmp, struct rsc, h);
520 __rsc_init(new, tmp);
523 static void update_rsc(struct cache_head *cnew, struct cache_head *ctmp)
525 struct rsc *new = container_of(cnew, struct rsc, h);
526 struct rsc *tmp = container_of(ctmp, struct rsc, h);
528 __rsc_update(new, tmp);
531 static struct cache_head * rsc_alloc(void)
542 static int rsc_parse(struct cache_detail *cd, char *mesg, int mlen)
545 int len, rv, tmp_int;
546 struct rsc rsci, *rscp = NULL;
548 int status = -EINVAL;
549 struct gss_api_mech *gm = NULL;
551 memset(&rsci, 0, sizeof(rsci));
554 len = qword_get(&mesg, buf, mlen);
555 if (len < 0) goto out;
557 if (rawobj_alloc(&rsci.handle, buf, len))
562 expiry = get_expiry(&mesg);
568 rv = get_int(&mesg, &tmp_int);
570 CERROR("fail to get remote flag\n");
573 rsci.ctx.gsc_remote = (tmp_int != 0);
576 rv = get_int(&mesg, &tmp_int);
578 CERROR("fail to get root user flag\n");
581 rsci.ctx.gsc_usr_root = (tmp_int != 0);
584 rv = get_int(&mesg, &tmp_int);
586 CERROR("fail to get mds user flag\n");
589 rsci.ctx.gsc_usr_mds = (tmp_int != 0);
592 rv = get_int(&mesg, &tmp_int);
594 CERROR("fail to get oss user flag\n");
597 rsci.ctx.gsc_usr_oss = (tmp_int != 0);
600 rv = get_int(&mesg, (int *) &rsci.ctx.gsc_mapped_uid);
602 CERROR("fail to get mapped uid\n");
606 rscp = rsc_lookup(&rsci);
610 /* uid, or NEGATIVE */
611 rv = get_int(&mesg, (int *) &rsci.ctx.gsc_uid);
615 CERROR("NOENT? set rsc entry negative\n");
616 set_bit(CACHE_NEGATIVE, &rsci.h.flags);
619 unsigned long ctx_expiry;
622 if (get_int(&mesg, (int *) &rsci.ctx.gsc_gid))
626 len = qword_get(&mesg, buf, mlen);
629 gm = lgss_name_to_mech(buf);
630 status = -EOPNOTSUPP;
635 /* mech-specific data: */
636 len = qword_get(&mesg, buf, mlen);
641 tmp_buf.data = (unsigned char *)buf;
642 if (lgss_import_sec_context(&tmp_buf, gm,
643 &rsci.ctx.gsc_mechctx))
646 /* currently the expiry time passed down from user-space
647 * is invalid, here we retrive it from mech. */
648 if (lgss_inquire_context(rsci.ctx.gsc_mechctx, &ctx_expiry)) {
649 CERROR("unable to get expire time, drop it\n");
652 expiry = (time_t) ctx_expiry;
655 rsci.h.expiry_time = expiry;
656 rscp = rsc_update(&rsci, rscp);
663 cache_put(&rscp->h, &rsc_cache);
668 CERROR("parse rsc error %d\n", status);
672 static struct cache_detail rsc_cache = {
673 .hash_size = RSC_HASHMAX,
674 .hash_table = rsc_table,
675 .name = "auth.sptlrpc.context",
676 .cache_put = rsc_put,
677 .cache_parse = rsc_parse,
680 .update = update_rsc,
684 static struct rsc *rsc_lookup(struct rsc *item)
686 struct cache_head *ch;
687 int hash = rsc_hash(item);
689 ch = sunrpc_cache_lookup(&rsc_cache, &item->h, hash);
691 return container_of(ch, struct rsc, h);
696 static struct rsc *rsc_update(struct rsc *new, struct rsc *old)
698 struct cache_head *ch;
699 int hash = rsc_hash(new);
701 ch = sunrpc_cache_update(&rsc_cache, &new->h, &old->h, hash);
703 return container_of(ch, struct rsc, h);
708 #define COMPAT_RSC_PUT(item, cd) cache_put((item), (cd))
710 /****************************************
712 ****************************************/
714 typedef int rsc_entry_match(struct rsc *rscp, long data);
716 static void rsc_flush(rsc_entry_match *match, long data)
718 #ifdef HAVE_CACHE_HEAD_HLIST
719 struct cache_head *ch = NULL;
720 struct hlist_head *head;
722 struct cache_head **ch;
728 write_lock(&rsc_cache.hash_lock);
729 for (n = 0; n < RSC_HASHMAX; n++) {
730 #ifdef HAVE_CACHE_HEAD_HLIST
731 head = &rsc_cache.hash_table[n];
732 hlist_for_each_entry(ch, head, cache_list) {
733 rscp = container_of(ch, struct rsc, h);
735 for (ch = &rsc_cache.hash_table[n]; *ch;) {
736 rscp = container_of(*ch, struct rsc, h);
739 if (!match(rscp, data)) {
740 #ifndef HAVE_CACHE_HEAD_HLIST
746 /* it seems simply set NEGATIVE doesn't work */
747 #ifdef HAVE_CACHE_HEAD_HLIST
748 hlist_del_init(&ch->cache_list);
754 set_bit(CACHE_NEGATIVE, &rscp->h.flags);
755 COMPAT_RSC_PUT(&rscp->h, &rsc_cache);
759 write_unlock(&rsc_cache.hash_lock);
763 static int match_uid(struct rsc *rscp, long uid)
767 return ((int) rscp->ctx.gsc_uid == (int) uid);
770 static int match_target(struct rsc *rscp, long target)
772 return (rscp->target == (struct obd_device *) target);
775 static inline void rsc_flush_uid(int uid)
778 CWARN("flush all gss contexts...\n");
780 rsc_flush(match_uid, (long) uid);
783 static inline void rsc_flush_target(struct obd_device *target)
785 rsc_flush(match_target, (long) target);
788 void gss_secsvc_flush(struct obd_device *target)
790 rsc_flush_target(target);
793 static struct rsc *gss_svc_searchbyctx(rawobj_t *handle)
798 memset(&rsci, 0, sizeof(rsci));
799 if (rawobj_dup(&rsci.handle, handle))
802 found = rsc_lookup(&rsci);
806 if (cache_check(&rsc_cache, &found->h, NULL))
811 int gss_svc_upcall_install_rvs_ctx(struct obd_import *imp,
812 struct gss_sec *gsec,
813 struct gss_cli_ctx *gctx)
815 struct rsc rsci, *rscp = NULL;
816 unsigned long ctx_expiry;
821 memset(&rsci, 0, sizeof(rsci));
823 if (rawobj_alloc(&rsci.handle, (char *) &gsec->gs_rvs_hdl,
824 sizeof(gsec->gs_rvs_hdl)))
825 GOTO(out, rc = -ENOMEM);
827 rscp = rsc_lookup(&rsci);
829 GOTO(out, rc = -ENOMEM);
831 major = lgss_copy_reverse_context(gctx->gc_mechctx,
832 &rsci.ctx.gsc_mechctx);
833 if (major != GSS_S_COMPLETE)
834 GOTO(out, rc = -ENOMEM);
836 if (lgss_inquire_context(rsci.ctx.gsc_mechctx, &ctx_expiry)) {
837 CERROR("unable to get expire time, drop it\n");
838 GOTO(out, rc = -EINVAL);
840 rsci.h.expiry_time = (time_t) ctx_expiry;
842 switch (imp->imp_obd->u.cli.cl_sp_to) {
844 rsci.ctx.gsc_usr_mds = 1;
847 rsci.ctx.gsc_usr_oss = 1;
850 rsci.ctx.gsc_usr_root = 1;
855 rscp = rsc_update(&rsci, rscp);
857 GOTO(out, rc = -ENOMEM);
859 rscp->target = imp->imp_obd;
860 rawobj_dup(&gctx->gc_svc_handle, &rscp->handle);
862 CWARN("create reverse svc ctx %p to %s: idx %#llx\n",
863 &rscp->ctx, obd2cli_tgt(imp->imp_obd), gsec->gs_rvs_hdl);
867 cache_put(&rscp->h, &rsc_cache);
871 CERROR("create reverse svc ctx: idx %#llx, rc %d\n",
872 gsec->gs_rvs_hdl, rc);
876 int gss_svc_upcall_expire_rvs_ctx(rawobj_t *handle)
878 const cfs_time_t expire = 20;
881 rscp = gss_svc_searchbyctx(handle);
883 CDEBUG(D_SEC, "reverse svcctx %p (rsc %p) expire soon\n",
886 rscp->h.expiry_time = cfs_time_current_sec() + expire;
887 COMPAT_RSC_PUT(&rscp->h, &rsc_cache);
892 int gss_svc_upcall_dup_handle(rawobj_t *handle, struct gss_svc_ctx *ctx)
894 struct rsc *rscp = container_of(ctx, struct rsc, ctx);
896 return rawobj_dup(handle, &rscp->handle);
899 int gss_svc_upcall_update_sequence(rawobj_t *handle, __u32 seq)
903 rscp = gss_svc_searchbyctx(handle);
905 CDEBUG(D_SEC, "reverse svcctx %p (rsc %p) update seq to %u\n",
906 &rscp->ctx, rscp, seq + 1);
908 rscp->ctx.gsc_rvs_seq = seq + 1;
909 COMPAT_RSC_PUT(&rscp->h, &rsc_cache);
914 static struct cache_deferred_req* cache_upcall_defer(struct cache_req *req)
918 static struct cache_req cache_upcall_chandle = { cache_upcall_defer };
920 int gss_svc_upcall_handle_init(struct ptlrpc_request *req,
921 struct gss_svc_reqctx *grctx,
922 struct gss_wire_ctx *gw,
923 struct obd_device *target,
928 struct ptlrpc_reply_state *rs;
929 struct rsc *rsci = NULL;
930 struct rsi *rsip = NULL, rsikey;
932 int replen = sizeof(struct ptlrpc_body);
933 struct gss_rep_header *rephdr;
935 int rc = SECSVC_DROP;
938 memset(&rsikey, 0, sizeof(rsikey));
939 rsikey.lustre_svc = lustre_svc;
940 rsikey.nid = (__u64) req->rq_peer.nid;
941 nodemap_test_nid(req->rq_peer.nid, rsikey.nm_name,
942 sizeof(rsikey.nm_name));
944 /* duplicate context handle. for INIT it always 0 */
945 if (rawobj_dup(&rsikey.in_handle, &gw->gw_handle)) {
946 CERROR("fail to dup context handle\n");
950 if (rawobj_dup(&rsikey.in_token, in_token)) {
951 CERROR("can't duplicate token\n");
952 rawobj_free(&rsikey.in_handle);
956 rsip = rsi_lookup(&rsikey);
959 CERROR("error in rsi_lookup.\n");
961 if (!gss_pack_err_notify(req, GSS_S_FAILURE, 0))
962 rc = SECSVC_COMPLETE;
967 cache_get(&rsip->h); /* take an extra ref */
968 init_waitqueue_head(&rsip->waitq);
969 init_waitqueue_entry(&wait, current);
970 add_wait_queue(&rsip->waitq, &wait);
973 /* Note each time cache_check() will drop a reference if return
974 * non-zero. We hold an extra reference on initial rsip, but must
975 * take care of following calls. */
976 rc = cache_check(&rsi_cache, &rsip->h, &cache_upcall_chandle);
985 read_lock(&rsi_cache.hash_lock);
986 valid = test_bit(CACHE_VALID, &rsip->h.flags);
988 set_current_state(TASK_INTERRUPTIBLE);
989 read_unlock(&rsi_cache.hash_lock);
992 unsigned long jiffies;
993 jiffies = msecs_to_jiffies(MSEC_PER_SEC *
994 GSS_SVC_UPCALL_TIMEOUT);
995 schedule_timeout(jiffies);
1000 CWARN("waited %ds timeout, drop\n", GSS_SVC_UPCALL_TIMEOUT);
1004 CDEBUG(D_SEC, "cache_check return ENOENT, drop\n");
1007 /* if not the first check, we have to release the extra
1008 * reference we just added on it. */
1010 cache_put(&rsip->h, &rsi_cache);
1011 CDEBUG(D_SEC, "cache_check is good\n");
1015 remove_wait_queue(&rsip->waitq, &wait);
1016 cache_put(&rsip->h, &rsi_cache);
1019 GOTO(out, rc = SECSVC_DROP);
1022 rsci = gss_svc_searchbyctx(&rsip->out_handle);
1024 CERROR("authentication failed\n");
1026 /* gss mechanism returned major and minor code so we return
1027 * those in error message */
1028 if (!gss_pack_err_notify(req, rsip->major_status,
1029 rsip->minor_status))
1030 rc = SECSVC_COMPLETE;
1034 cache_get(&rsci->h);
1035 grctx->src_ctx = &rsci->ctx;
1038 if (rawobj_dup(&rsci->ctx.gsc_rvs_hdl, rvs_hdl)) {
1039 CERROR("failed duplicate reverse handle\n");
1043 rsci->target = target;
1045 CDEBUG(D_SEC, "server create rsc %p(%u->%s)\n",
1046 rsci, rsci->ctx.gsc_uid, libcfs_nid2str(req->rq_peer.nid));
1048 if (rsip->out_handle.len > PTLRPC_GSS_MAX_HANDLE_SIZE) {
1049 CERROR("handle size %u too large\n", rsip->out_handle.len);
1050 GOTO(out, rc = SECSVC_DROP);
1053 grctx->src_init = 1;
1054 grctx->src_reserve_len = cfs_size_round4(rsip->out_token.len);
1056 rc = lustre_pack_reply_v2(req, 1, &replen, NULL, 0);
1058 CERROR("failed to pack reply: %d\n", rc);
1059 GOTO(out, rc = SECSVC_DROP);
1062 rs = req->rq_reply_state;
1063 LASSERT(rs->rs_repbuf->lm_bufcount == 3);
1064 LASSERT(rs->rs_repbuf->lm_buflens[0] >=
1065 sizeof(*rephdr) + rsip->out_handle.len);
1066 LASSERT(rs->rs_repbuf->lm_buflens[2] >= rsip->out_token.len);
1068 rephdr = lustre_msg_buf(rs->rs_repbuf, 0, 0);
1069 rephdr->gh_version = PTLRPC_GSS_VERSION;
1070 rephdr->gh_flags = 0;
1071 rephdr->gh_proc = PTLRPC_GSS_PROC_ERR;
1072 rephdr->gh_major = rsip->major_status;
1073 rephdr->gh_minor = rsip->minor_status;
1074 rephdr->gh_seqwin = GSS_SEQ_WIN;
1075 rephdr->gh_handle.len = rsip->out_handle.len;
1076 memcpy(rephdr->gh_handle.data, rsip->out_handle.data,
1077 rsip->out_handle.len);
1079 memcpy(lustre_msg_buf(rs->rs_repbuf, 2, 0), rsip->out_token.data,
1080 rsip->out_token.len);
1082 rs->rs_repdata_len = lustre_shrink_msg(rs->rs_repbuf, 2,
1083 rsip->out_token.len, 0);
1088 /* it looks like here we should put rsip also, but this mess up
1089 * with NFS cache mgmt code... FIXME
1092 * rsi_put(&rsip->h, &rsi_cache); */
1095 /* if anything went wrong, we don't keep the context too */
1096 if (rc != SECSVC_OK)
1097 set_bit(CACHE_NEGATIVE, &rsci->h.flags);
1099 CDEBUG(D_SEC, "create rsc with idx %#llx\n",
1100 gss_handle_to_u64(&rsci->handle));
1102 COMPAT_RSC_PUT(&rsci->h, &rsc_cache);
1107 struct gss_svc_ctx *gss_svc_upcall_get_ctx(struct ptlrpc_request *req,
1108 struct gss_wire_ctx *gw)
1112 rsc = gss_svc_searchbyctx(&gw->gw_handle);
1114 CWARN("Invalid gss ctx idx %#llx from %s\n",
1115 gss_handle_to_u64(&gw->gw_handle),
1116 libcfs_nid2str(req->rq_peer.nid));
1123 void gss_svc_upcall_put_ctx(struct gss_svc_ctx *ctx)
1125 struct rsc *rsc = container_of(ctx, struct rsc, ctx);
1127 COMPAT_RSC_PUT(&rsc->h, &rsc_cache);
1130 void gss_svc_upcall_destroy_ctx(struct gss_svc_ctx *ctx)
1132 struct rsc *rsc = container_of(ctx, struct rsc, ctx);
1134 /* can't be found */
1135 set_bit(CACHE_NEGATIVE, &rsc->h.flags);
1136 /* to be removed at next scan */
1137 rsc->h.expiry_time = 1;
1140 int __init gss_init_svc_upcall(void)
1144 spin_lock_init(&__ctx_index_lock);
1146 * this helps reducing context index confliction. after server reboot,
1147 * conflicting request from clients might be filtered out by initial
1148 * sequence number checking, thus no chance to sent error notification
1151 cfs_get_random_bytes(&__ctx_index, sizeof(__ctx_index));
1153 rc = _cache_register_net(&rsi_cache, &init_net);
1157 rc = _cache_register_net(&rsc_cache, &init_net);
1159 _cache_unregister_net(&rsi_cache, &init_net);
1163 /* FIXME this looks stupid. we intend to give lsvcgssd a chance to open
1164 * the init upcall channel, otherwise there's big chance that the first
1165 * upcall issued before the channel be opened thus nfsv4 cache code will
1166 * drop the request direclty, thus lead to unnecessary recovery time.
1167 * here we wait at miximum 1.5 seconds. */
1168 for (i = 0; i < 6; i++) {
1169 if (atomic_read(&rsi_cache.readers) > 0)
1171 set_current_state(TASK_UNINTERRUPTIBLE);
1172 LASSERT(msecs_to_jiffies(MSEC_PER_SEC) >= 4);
1173 schedule_timeout(msecs_to_jiffies(MSEC_PER_SEC / 4));
1176 if (atomic_read(&rsi_cache.readers) == 0)
1177 CWARN("Init channel is not opened by lsvcgssd, following "
1178 "request might be dropped until lsvcgssd is active\n");
1183 void gss_exit_svc_upcall(void)
1185 cache_purge(&rsi_cache);
1186 _cache_unregister_net(&rsi_cache, &init_net);
1188 cache_purge(&rsc_cache);
1189 _cache_unregister_net(&rsc_cache, &init_net);