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_import.h>
64 #include <lustre_net.h>
65 #include <lustre_nodemap.h>
66 #include <lustre_sec.h>
69 #include "gss_internal.h"
72 #define GSS_SVC_UPCALL_TIMEOUT (20)
74 static spinlock_t __ctx_index_lock;
75 static __u64 __ctx_index;
77 __u64 gss_get_next_ctx_index(void)
81 spin_lock(&__ctx_index_lock);
83 spin_unlock(&__ctx_index_lock);
88 static inline unsigned long hash_mem(char *buf, int length, int bits)
90 unsigned long hash = 0;
105 if ((len & (BITS_PER_LONG/8-1)) == 0)
106 hash = hash_long(hash^l, BITS_PER_LONG);
109 return hash >> (BITS_PER_LONG - bits);
112 /* This compatibility can be removed once kernel 3.3 is used,
113 * since cache_register_net/cache_unregister_net are exported.
114 * Note that since kernel 3.4 cache_register and cache_unregister
117 static inline int _cache_register_net(struct cache_detail *cd, struct net *net)
119 #ifdef HAVE_CACHE_REGISTER
120 return cache_register(cd);
122 return cache_register_net(cd, net);
125 static inline void _cache_unregister_net(struct cache_detail *cd,
128 #ifdef HAVE_CACHE_REGISTER
129 cache_unregister(cd);
131 cache_unregister_net(cd, net);
134 /****************************************
135 * rpc sec init (rsi) cache *
136 ****************************************/
138 #define RSI_HASHBITS (6)
139 #define RSI_HASHMAX (1 << RSI_HASHBITS)
140 #define RSI_HASHMASK (RSI_HASHMAX - 1)
146 char nm_name[LUSTRE_NODEMAP_NAME_LENGTH + 1];
147 wait_queue_head_t waitq;
148 rawobj_t in_handle, in_token;
149 rawobj_t out_handle, out_token;
150 int major_status, minor_status;
153 #ifdef HAVE_CACHE_HEAD_HLIST
154 static struct hlist_head rsi_table[RSI_HASHMAX];
156 static struct cache_head *rsi_table[RSI_HASHMAX];
158 static struct cache_detail rsi_cache;
159 static struct rsi *rsi_update(struct rsi *new, struct rsi *old);
160 static struct rsi *rsi_lookup(struct rsi *item);
162 static inline int rsi_hash(struct rsi *item)
164 return hash_mem((char *)item->in_handle.data, item->in_handle.len,
166 hash_mem((char *)item->in_token.data, item->in_token.len,
170 static inline int __rsi_match(struct rsi *item, struct rsi *tmp)
172 return (rawobj_equal(&item->in_handle, &tmp->in_handle) &&
173 rawobj_equal(&item->in_token, &tmp->in_token));
176 static void rsi_free(struct rsi *rsi)
178 rawobj_free(&rsi->in_handle);
179 rawobj_free(&rsi->in_token);
180 rawobj_free(&rsi->out_handle);
181 rawobj_free(&rsi->out_token);
184 /* See handle_channel_req() userspace for where the upcall data is read */
185 static void rsi_request(struct cache_detail *cd,
186 struct cache_head *h,
187 char **bpp, int *blen)
189 struct rsi *rsi = container_of(h, struct rsi, h);
192 /* if in_handle is null, provide kernel suggestion */
193 if (rsi->in_handle.len == 0)
194 index = gss_get_next_ctx_index();
196 qword_addhex(bpp, blen, (char *) &rsi->lustre_svc,
197 sizeof(rsi->lustre_svc));
198 qword_addhex(bpp, blen, (char *) &rsi->nid, sizeof(rsi->nid));
199 qword_addhex(bpp, blen, (char *) &index, sizeof(index));
200 qword_addhex(bpp, blen, (char *) rsi->nm_name,
201 strlen(rsi->nm_name) + 1);
202 qword_addhex(bpp, blen, rsi->in_handle.data, rsi->in_handle.len);
203 qword_addhex(bpp, blen, rsi->in_token.data, rsi->in_token.len);
207 #ifdef HAVE_SUNRPC_UPCALL_HAS_3ARGS
208 static int rsi_upcall(struct cache_detail *cd, struct cache_head *h)
210 return sunrpc_cache_pipe_upcall(cd, h, rsi_request);
214 static int rsi_upcall(struct cache_detail *cd, struct cache_head *h)
216 return sunrpc_cache_pipe_upcall(cd, h);
220 static inline void __rsi_init(struct rsi *new, struct rsi *item)
222 new->out_handle = RAWOBJ_EMPTY;
223 new->out_token = RAWOBJ_EMPTY;
225 new->in_handle = item->in_handle;
226 item->in_handle = RAWOBJ_EMPTY;
227 new->in_token = item->in_token;
228 item->in_token = RAWOBJ_EMPTY;
230 new->lustre_svc = item->lustre_svc;
231 new->nid = item->nid;
232 memcpy(new->nm_name, item->nm_name, sizeof(item->nm_name));
233 init_waitqueue_head(&new->waitq);
236 static inline void __rsi_update(struct rsi *new, struct rsi *item)
238 LASSERT(new->out_handle.len == 0);
239 LASSERT(new->out_token.len == 0);
241 new->out_handle = item->out_handle;
242 item->out_handle = RAWOBJ_EMPTY;
243 new->out_token = item->out_token;
244 item->out_token = RAWOBJ_EMPTY;
246 new->major_status = item->major_status;
247 new->minor_status = item->minor_status;
250 static void rsi_put(struct kref *ref)
252 struct rsi *rsi = container_of(ref, struct rsi, h.ref);
254 #ifdef HAVE_CACHE_HEAD_HLIST
255 LASSERT(rsi->h.cache_list.next == NULL);
257 LASSERT(rsi->h.next == NULL);
263 static int rsi_match(struct cache_head *a, struct cache_head *b)
265 struct rsi *item = container_of(a, struct rsi, h);
266 struct rsi *tmp = container_of(b, struct rsi, h);
268 return __rsi_match(item, tmp);
271 static void rsi_init(struct cache_head *cnew, struct cache_head *citem)
273 struct rsi *new = container_of(cnew, struct rsi, h);
274 struct rsi *item = container_of(citem, struct rsi, h);
276 __rsi_init(new, item);
279 static void update_rsi(struct cache_head *cnew, struct cache_head *citem)
281 struct rsi *new = container_of(cnew, struct rsi, h);
282 struct rsi *item = container_of(citem, struct rsi, h);
284 __rsi_update(new, item);
287 static struct cache_head *rsi_alloc(void)
298 static int rsi_parse(struct cache_detail *cd, char *mesg, int mlen)
302 struct rsi rsii, *rsip = NULL;
304 int status = -EINVAL;
308 memset(&rsii, 0, sizeof(rsii));
311 len = qword_get(&mesg, buf, mlen);
314 if (rawobj_alloc(&rsii.in_handle, buf, len)) {
320 len = qword_get(&mesg, buf, mlen);
323 if (rawobj_alloc(&rsii.in_token, buf, len)) {
328 rsip = rsi_lookup(&rsii);
334 expiry = get_expiry(&mesg);
338 len = qword_get(&mesg, buf, mlen);
343 status = kstrtoint(buf, 10, &rsii.major_status);
348 len = qword_get(&mesg, buf, mlen);
354 status = kstrtoint(buf, 10, &rsii.minor_status);
359 len = qword_get(&mesg, buf, mlen);
362 if (rawobj_alloc(&rsii.out_handle, buf, len)) {
368 len = qword_get(&mesg, buf, mlen);
371 if (rawobj_alloc(&rsii.out_token, buf, len)) {
376 rsii.h.expiry_time = expiry;
377 rsip = rsi_update(&rsii, rsip);
382 wake_up_all(&rsip->waitq);
383 cache_put(&rsip->h, &rsi_cache);
389 CERROR("rsi parse error %d\n", status);
393 static struct cache_detail rsi_cache = {
394 .hash_size = RSI_HASHMAX,
395 .hash_table = rsi_table,
396 .name = "auth.sptlrpc.init",
397 .cache_put = rsi_put,
398 #ifndef HAVE_SUNRPC_UPCALL_HAS_3ARGS
399 .cache_request = rsi_request,
401 .cache_upcall = rsi_upcall,
402 .cache_parse = rsi_parse,
405 .update = update_rsi,
409 static struct rsi *rsi_lookup(struct rsi *item)
411 struct cache_head *ch;
412 int hash = rsi_hash(item);
414 ch = sunrpc_cache_lookup(&rsi_cache, &item->h, hash);
416 return container_of(ch, struct rsi, h);
421 static struct rsi *rsi_update(struct rsi *new, struct rsi *old)
423 struct cache_head *ch;
424 int hash = rsi_hash(new);
426 ch = sunrpc_cache_update(&rsi_cache, &new->h, &old->h, hash);
428 return container_of(ch, struct rsi, h);
433 /****************************************
434 * rpc sec context (rsc) cache *
435 ****************************************/
437 #define RSC_HASHBITS (10)
438 #define RSC_HASHMAX (1 << RSC_HASHBITS)
439 #define RSC_HASHMASK (RSC_HASHMAX - 1)
443 struct obd_device *target;
445 struct gss_svc_ctx ctx;
448 #ifdef HAVE_CACHE_HEAD_HLIST
449 static struct hlist_head rsc_table[RSC_HASHMAX];
451 static struct cache_head *rsc_table[RSC_HASHMAX];
453 static struct cache_detail rsc_cache;
454 static struct rsc *rsc_update(struct rsc *new, struct rsc *old);
455 static struct rsc *rsc_lookup(struct rsc *item);
457 static void rsc_free(struct rsc *rsci)
459 rawobj_free(&rsci->handle);
460 rawobj_free(&rsci->ctx.gsc_rvs_hdl);
461 lgss_delete_sec_context(&rsci->ctx.gsc_mechctx);
464 static inline int rsc_hash(struct rsc *rsci)
466 return hash_mem((char *)rsci->handle.data,
467 rsci->handle.len, RSC_HASHBITS);
470 static inline int __rsc_match(struct rsc *new, struct rsc *tmp)
472 return rawobj_equal(&new->handle, &tmp->handle);
475 static inline void __rsc_init(struct rsc *new, struct rsc *tmp)
477 new->handle = tmp->handle;
478 tmp->handle = RAWOBJ_EMPTY;
481 memset(&new->ctx, 0, sizeof(new->ctx));
482 new->ctx.gsc_rvs_hdl = RAWOBJ_EMPTY;
485 static inline void __rsc_update(struct rsc *new, struct rsc *tmp)
488 tmp->ctx.gsc_rvs_hdl = RAWOBJ_EMPTY;
489 tmp->ctx.gsc_mechctx = NULL;
491 memset(&new->ctx.gsc_seqdata, 0, sizeof(new->ctx.gsc_seqdata));
492 spin_lock_init(&new->ctx.gsc_seqdata.ssd_lock);
495 static void rsc_put(struct kref *ref)
497 struct rsc *rsci = container_of(ref, struct rsc, h.ref);
499 #ifdef HAVE_CACHE_HEAD_HLIST
500 LASSERT(rsci->h.cache_list.next == NULL);
502 LASSERT(rsci->h.next == NULL);
508 static int rsc_match(struct cache_head *a, struct cache_head *b)
510 struct rsc *new = container_of(a, struct rsc, h);
511 struct rsc *tmp = container_of(b, struct rsc, h);
513 return __rsc_match(new, tmp);
516 static void rsc_init(struct cache_head *cnew, struct cache_head *ctmp)
518 struct rsc *new = container_of(cnew, struct rsc, h);
519 struct rsc *tmp = container_of(ctmp, struct rsc, h);
521 __rsc_init(new, tmp);
524 static void update_rsc(struct cache_head *cnew, struct cache_head *ctmp)
526 struct rsc *new = container_of(cnew, struct rsc, h);
527 struct rsc *tmp = container_of(ctmp, struct rsc, h);
529 __rsc_update(new, tmp);
532 static struct cache_head * rsc_alloc(void)
543 static int rsc_parse(struct cache_detail *cd, char *mesg, int mlen)
546 int len, rv, tmp_int;
547 struct rsc rsci, *rscp = NULL;
549 int status = -EINVAL;
550 struct gss_api_mech *gm = NULL;
552 memset(&rsci, 0, sizeof(rsci));
555 len = qword_get(&mesg, buf, mlen);
556 if (len < 0) goto out;
558 if (rawobj_alloc(&rsci.handle, buf, len))
563 expiry = get_expiry(&mesg);
569 rv = get_int(&mesg, &tmp_int);
571 CERROR("fail to get remote flag\n");
574 rsci.ctx.gsc_remote = (tmp_int != 0);
577 rv = get_int(&mesg, &tmp_int);
579 CERROR("fail to get root user flag\n");
582 rsci.ctx.gsc_usr_root = (tmp_int != 0);
585 rv = get_int(&mesg, &tmp_int);
587 CERROR("fail to get mds user flag\n");
590 rsci.ctx.gsc_usr_mds = (tmp_int != 0);
593 rv = get_int(&mesg, &tmp_int);
595 CERROR("fail to get oss user flag\n");
598 rsci.ctx.gsc_usr_oss = (tmp_int != 0);
601 rv = get_int(&mesg, (int *) &rsci.ctx.gsc_mapped_uid);
603 CERROR("fail to get mapped uid\n");
607 rscp = rsc_lookup(&rsci);
611 /* uid, or NEGATIVE */
612 rv = get_int(&mesg, (int *) &rsci.ctx.gsc_uid);
616 CERROR("NOENT? set rsc entry negative\n");
617 set_bit(CACHE_NEGATIVE, &rsci.h.flags);
623 if (get_int(&mesg, (int *) &rsci.ctx.gsc_gid))
627 len = qword_get(&mesg, buf, mlen);
630 gm = lgss_name_to_mech(buf);
631 status = -EOPNOTSUPP;
636 /* mech-specific data: */
637 len = qword_get(&mesg, buf, mlen);
642 tmp_buf.data = (unsigned char *)buf;
643 if (lgss_import_sec_context(&tmp_buf, gm,
644 &rsci.ctx.gsc_mechctx))
647 /* set to seconds since machine booted */
648 expiry = ktime_get_seconds();
650 /* currently the expiry time passed down from user-space
651 * is invalid, here we retrive it from mech.
653 if (lgss_inquire_context(rsci.ctx.gsc_mechctx, &ctx_expiry)) {
654 CERROR("unable to get expire time, drop it\n");
658 /* ctx_expiry is the number of seconds since Jan 1 1970.
659 * We want just the number of seconds into the future.
661 expiry += ctx_expiry - ktime_get_real_seconds();
664 rsci.h.expiry_time = expiry;
665 rscp = rsc_update(&rsci, rscp);
672 cache_put(&rscp->h, &rsc_cache);
677 CERROR("parse rsc error %d\n", status);
681 static struct cache_detail rsc_cache = {
682 .hash_size = RSC_HASHMAX,
683 .hash_table = rsc_table,
684 .name = "auth.sptlrpc.context",
685 .cache_put = rsc_put,
686 .cache_parse = rsc_parse,
689 .update = update_rsc,
693 static struct rsc *rsc_lookup(struct rsc *item)
695 struct cache_head *ch;
696 int hash = rsc_hash(item);
698 ch = sunrpc_cache_lookup(&rsc_cache, &item->h, hash);
700 return container_of(ch, struct rsc, h);
705 static struct rsc *rsc_update(struct rsc *new, struct rsc *old)
707 struct cache_head *ch;
708 int hash = rsc_hash(new);
710 ch = sunrpc_cache_update(&rsc_cache, &new->h, &old->h, hash);
712 return container_of(ch, struct rsc, h);
717 #define COMPAT_RSC_PUT(item, cd) cache_put((item), (cd))
719 /****************************************
721 ****************************************/
723 typedef int rsc_entry_match(struct rsc *rscp, long data);
725 static void rsc_flush(rsc_entry_match *match, long data)
727 #ifdef HAVE_CACHE_HEAD_HLIST
728 struct cache_head *ch = NULL;
729 struct hlist_head *head;
731 struct cache_head **ch;
737 write_lock(&rsc_cache.hash_lock);
738 for (n = 0; n < RSC_HASHMAX; n++) {
739 #ifdef HAVE_CACHE_HEAD_HLIST
740 head = &rsc_cache.hash_table[n];
741 hlist_for_each_entry(ch, head, cache_list) {
742 rscp = container_of(ch, struct rsc, h);
744 for (ch = &rsc_cache.hash_table[n]; *ch;) {
745 rscp = container_of(*ch, struct rsc, h);
748 if (!match(rscp, data)) {
749 #ifndef HAVE_CACHE_HEAD_HLIST
755 /* it seems simply set NEGATIVE doesn't work */
756 #ifdef HAVE_CACHE_HEAD_HLIST
757 hlist_del_init(&ch->cache_list);
763 set_bit(CACHE_NEGATIVE, &rscp->h.flags);
764 COMPAT_RSC_PUT(&rscp->h, &rsc_cache);
768 write_unlock(&rsc_cache.hash_lock);
772 static int match_uid(struct rsc *rscp, long uid)
776 return ((int) rscp->ctx.gsc_uid == (int) uid);
779 static int match_target(struct rsc *rscp, long target)
781 return (rscp->target == (struct obd_device *) target);
784 static inline void rsc_flush_uid(int uid)
787 CWARN("flush all gss contexts...\n");
789 rsc_flush(match_uid, (long) uid);
792 static inline void rsc_flush_target(struct obd_device *target)
794 rsc_flush(match_target, (long) target);
797 void gss_secsvc_flush(struct obd_device *target)
799 rsc_flush_target(target);
802 static struct rsc *gss_svc_searchbyctx(rawobj_t *handle)
807 memset(&rsci, 0, sizeof(rsci));
808 if (rawobj_dup(&rsci.handle, handle))
811 found = rsc_lookup(&rsci);
815 if (cache_check(&rsc_cache, &found->h, NULL))
820 int gss_svc_upcall_install_rvs_ctx(struct obd_import *imp,
821 struct gss_sec *gsec,
822 struct gss_cli_ctx *gctx)
824 struct rsc rsci, *rscp = NULL;
830 memset(&rsci, 0, sizeof(rsci));
832 if (rawobj_alloc(&rsci.handle, (char *) &gsec->gs_rvs_hdl,
833 sizeof(gsec->gs_rvs_hdl)))
834 GOTO(out, rc = -ENOMEM);
836 rscp = rsc_lookup(&rsci);
838 GOTO(out, rc = -ENOMEM);
840 major = lgss_copy_reverse_context(gctx->gc_mechctx,
841 &rsci.ctx.gsc_mechctx);
842 if (major != GSS_S_COMPLETE)
843 GOTO(out, rc = -ENOMEM);
845 if (lgss_inquire_context(rsci.ctx.gsc_mechctx, &ctx_expiry)) {
846 CERROR("unable to get expire time, drop it\n");
847 GOTO(out, rc = -EINVAL);
849 rsci.h.expiry_time = (time_t) ctx_expiry;
851 switch (imp->imp_obd->u.cli.cl_sp_to) {
853 rsci.ctx.gsc_usr_mds = 1;
856 rsci.ctx.gsc_usr_oss = 1;
859 rsci.ctx.gsc_usr_root = 1;
862 /* by convention, all 3 set to 1 means MGS */
863 rsci.ctx.gsc_usr_mds = 1;
864 rsci.ctx.gsc_usr_oss = 1;
865 rsci.ctx.gsc_usr_root = 1;
871 rscp = rsc_update(&rsci, rscp);
873 GOTO(out, rc = -ENOMEM);
875 rscp->target = imp->imp_obd;
876 rawobj_dup(&gctx->gc_svc_handle, &rscp->handle);
878 CWARN("create reverse svc ctx %p to %s: idx %#llx\n",
879 &rscp->ctx, obd2cli_tgt(imp->imp_obd), gsec->gs_rvs_hdl);
883 cache_put(&rscp->h, &rsc_cache);
887 CERROR("create reverse svc ctx: idx %#llx, rc %d\n",
888 gsec->gs_rvs_hdl, rc);
892 int gss_svc_upcall_expire_rvs_ctx(rawobj_t *handle)
894 const time64_t expire = 20;
897 rscp = gss_svc_searchbyctx(handle);
899 CDEBUG(D_SEC, "reverse svcctx %p (rsc %p) expire soon\n",
902 rscp->h.expiry_time = ktime_get_real_seconds() + expire;
903 COMPAT_RSC_PUT(&rscp->h, &rsc_cache);
908 int gss_svc_upcall_dup_handle(rawobj_t *handle, struct gss_svc_ctx *ctx)
910 struct rsc *rscp = container_of(ctx, struct rsc, ctx);
912 return rawobj_dup(handle, &rscp->handle);
915 int gss_svc_upcall_update_sequence(rawobj_t *handle, __u32 seq)
919 rscp = gss_svc_searchbyctx(handle);
921 CDEBUG(D_SEC, "reverse svcctx %p (rsc %p) update seq to %u\n",
922 &rscp->ctx, rscp, seq + 1);
924 rscp->ctx.gsc_rvs_seq = seq + 1;
925 COMPAT_RSC_PUT(&rscp->h, &rsc_cache);
930 static struct cache_deferred_req* cache_upcall_defer(struct cache_req *req)
934 static struct cache_req cache_upcall_chandle = { cache_upcall_defer };
936 int gss_svc_upcall_handle_init(struct ptlrpc_request *req,
937 struct gss_svc_reqctx *grctx,
938 struct gss_wire_ctx *gw,
939 struct obd_device *target,
944 struct ptlrpc_reply_state *rs;
945 struct rsc *rsci = NULL;
946 struct rsi *rsip = NULL, rsikey;
947 wait_queue_entry_t wait;
948 int replen = sizeof(struct ptlrpc_body);
949 struct gss_rep_header *rephdr;
951 int rc = SECSVC_DROP;
954 memset(&rsikey, 0, sizeof(rsikey));
955 rsikey.lustre_svc = lustre_svc;
956 /* In case of MR, rq_peer is not the NID from which request is received,
957 * but primary NID of peer.
958 * So we need rq_source, which contains the NID actually in use.
960 rsikey.nid = (__u64) req->rq_source.nid;
961 nodemap_test_nid(req->rq_peer.nid, rsikey.nm_name,
962 sizeof(rsikey.nm_name));
964 /* duplicate context handle. for INIT it always 0 */
965 if (rawobj_dup(&rsikey.in_handle, &gw->gw_handle)) {
966 CERROR("fail to dup context handle\n");
970 if (rawobj_dup(&rsikey.in_token, in_token)) {
971 CERROR("can't duplicate token\n");
972 rawobj_free(&rsikey.in_handle);
976 rsip = rsi_lookup(&rsikey);
979 CERROR("error in rsi_lookup.\n");
981 if (!gss_pack_err_notify(req, GSS_S_FAILURE, 0))
982 rc = SECSVC_COMPLETE;
987 cache_get(&rsip->h); /* take an extra ref */
988 init_waitqueue_head(&rsip->waitq);
989 init_waitqueue_entry(&wait, current);
990 add_wait_queue(&rsip->waitq, &wait);
993 /* Note each time cache_check() will drop a reference if return
994 * non-zero. We hold an extra reference on initial rsip, but must
995 * take care of following calls. */
996 rc = cache_check(&rsi_cache, &rsip->h, &cache_upcall_chandle);
1005 read_lock(&rsi_cache.hash_lock);
1006 valid = test_bit(CACHE_VALID, &rsip->h.flags);
1008 set_current_state(TASK_INTERRUPTIBLE);
1009 read_unlock(&rsi_cache.hash_lock);
1012 unsigned long jiffies;
1013 jiffies = msecs_to_jiffies(MSEC_PER_SEC *
1014 GSS_SVC_UPCALL_TIMEOUT);
1015 schedule_timeout(jiffies);
1017 cache_get(&rsip->h);
1020 CWARN("waited %ds timeout, drop\n", GSS_SVC_UPCALL_TIMEOUT);
1024 CDEBUG(D_SEC, "cache_check return ENOENT, drop\n");
1027 /* if not the first check, we have to release the extra
1028 * reference we just added on it. */
1030 cache_put(&rsip->h, &rsi_cache);
1031 CDEBUG(D_SEC, "cache_check is good\n");
1035 remove_wait_queue(&rsip->waitq, &wait);
1036 cache_put(&rsip->h, &rsi_cache);
1039 GOTO(out, rc = SECSVC_DROP);
1042 rsci = gss_svc_searchbyctx(&rsip->out_handle);
1044 CERROR("authentication failed\n");
1046 /* gss mechanism returned major and minor code so we return
1047 * those in error message */
1048 if (!gss_pack_err_notify(req, rsip->major_status,
1049 rsip->minor_status))
1050 rc = SECSVC_COMPLETE;
1054 cache_get(&rsci->h);
1055 grctx->src_ctx = &rsci->ctx;
1058 if (rawobj_dup(&rsci->ctx.gsc_rvs_hdl, rvs_hdl)) {
1059 CERROR("failed duplicate reverse handle\n");
1063 rsci->target = target;
1065 CDEBUG(D_SEC, "server create rsc %p(%u->%s)\n",
1066 rsci, rsci->ctx.gsc_uid, libcfs_nid2str(req->rq_peer.nid));
1068 if (rsip->out_handle.len > PTLRPC_GSS_MAX_HANDLE_SIZE) {
1069 CERROR("handle size %u too large\n", rsip->out_handle.len);
1070 GOTO(out, rc = SECSVC_DROP);
1073 grctx->src_init = 1;
1074 grctx->src_reserve_len = cfs_size_round4(rsip->out_token.len);
1076 rc = lustre_pack_reply_v2(req, 1, &replen, NULL, 0);
1078 CERROR("failed to pack reply: %d\n", rc);
1079 GOTO(out, rc = SECSVC_DROP);
1082 rs = req->rq_reply_state;
1083 LASSERT(rs->rs_repbuf->lm_bufcount == 3);
1084 LASSERT(rs->rs_repbuf->lm_buflens[0] >=
1085 sizeof(*rephdr) + rsip->out_handle.len);
1086 LASSERT(rs->rs_repbuf->lm_buflens[2] >= rsip->out_token.len);
1088 rephdr = lustre_msg_buf(rs->rs_repbuf, 0, 0);
1089 rephdr->gh_version = PTLRPC_GSS_VERSION;
1090 rephdr->gh_flags = 0;
1091 rephdr->gh_proc = PTLRPC_GSS_PROC_ERR;
1092 rephdr->gh_major = rsip->major_status;
1093 rephdr->gh_minor = rsip->minor_status;
1094 rephdr->gh_seqwin = GSS_SEQ_WIN;
1095 rephdr->gh_handle.len = rsip->out_handle.len;
1096 memcpy(rephdr->gh_handle.data, rsip->out_handle.data,
1097 rsip->out_handle.len);
1099 memcpy(lustre_msg_buf(rs->rs_repbuf, 2, 0), rsip->out_token.data,
1100 rsip->out_token.len);
1102 rs->rs_repdata_len = lustre_shrink_msg(rs->rs_repbuf, 2,
1103 rsip->out_token.len, 0);
1108 /* it looks like here we should put rsip also, but this mess up
1109 * with NFS cache mgmt code... FIXME
1112 * rsi_put(&rsip->h, &rsi_cache); */
1115 /* if anything went wrong, we don't keep the context too */
1116 if (rc != SECSVC_OK)
1117 set_bit(CACHE_NEGATIVE, &rsci->h.flags);
1119 CDEBUG(D_SEC, "create rsc with idx %#llx\n",
1120 gss_handle_to_u64(&rsci->handle));
1122 COMPAT_RSC_PUT(&rsci->h, &rsc_cache);
1127 struct gss_svc_ctx *gss_svc_upcall_get_ctx(struct ptlrpc_request *req,
1128 struct gss_wire_ctx *gw)
1132 rsc = gss_svc_searchbyctx(&gw->gw_handle);
1134 CWARN("Invalid gss ctx idx %#llx from %s\n",
1135 gss_handle_to_u64(&gw->gw_handle),
1136 libcfs_nid2str(req->rq_peer.nid));
1143 void gss_svc_upcall_put_ctx(struct gss_svc_ctx *ctx)
1145 struct rsc *rsc = container_of(ctx, struct rsc, ctx);
1147 COMPAT_RSC_PUT(&rsc->h, &rsc_cache);
1150 void gss_svc_upcall_destroy_ctx(struct gss_svc_ctx *ctx)
1152 struct rsc *rsc = container_of(ctx, struct rsc, ctx);
1154 /* can't be found */
1155 set_bit(CACHE_NEGATIVE, &rsc->h.flags);
1156 /* to be removed at next scan */
1157 rsc->h.expiry_time = 1;
1160 int __init gss_init_svc_upcall(void)
1164 spin_lock_init(&__ctx_index_lock);
1166 * this helps reducing context index confliction. after server reboot,
1167 * conflicting request from clients might be filtered out by initial
1168 * sequence number checking, thus no chance to sent error notification
1171 cfs_get_random_bytes(&__ctx_index, sizeof(__ctx_index));
1173 rc = _cache_register_net(&rsi_cache, &init_net);
1177 rc = _cache_register_net(&rsc_cache, &init_net);
1179 _cache_unregister_net(&rsi_cache, &init_net);
1183 /* FIXME this looks stupid. we intend to give lsvcgssd a chance to open
1184 * the init upcall channel, otherwise there's big chance that the first
1185 * upcall issued before the channel be opened thus nfsv4 cache code will
1186 * drop the request directly, thus lead to unnecessary recovery time.
1187 * Here we wait at minimum 1.5 seconds.
1189 for (i = 0; i < 6; i++) {
1190 if (atomic_read(&rsi_cache.readers) > 0)
1192 set_current_state(TASK_UNINTERRUPTIBLE);
1193 LASSERT(msecs_to_jiffies(MSEC_PER_SEC / 4) > 0);
1194 schedule_timeout(msecs_to_jiffies(MSEC_PER_SEC / 4));
1197 if (atomic_read(&rsi_cache.readers) == 0)
1198 CWARN("Init channel is not opened by lsvcgssd, following "
1199 "request might be dropped until lsvcgssd is active\n");
1204 void gss_exit_svc_upcall(void)
1206 cache_purge(&rsi_cache);
1207 _cache_unregister_net(&rsi_cache, &init_net);
1209 cache_purge(&rsc_cache);
1210 _cache_unregister_net(&rsc_cache, &init_net);