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/random.h>
55 #include <linux/slab.h>
56 #include <linux/hash.h>
57 #include <linux/mutex.h>
58 #include <linux/sunrpc/cache.h>
62 #include <obd_class.h>
63 #include <obd_support.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"
72 #include "gss_crypto.h"
74 #define GSS_SVC_UPCALL_TIMEOUT (20)
76 static DEFINE_SPINLOCK(__ctx_index_lock);
77 static __u64 __ctx_index;
79 unsigned int krb5_allow_old_client_csum;
81 __u64 gss_get_next_ctx_index(void)
85 spin_lock(&__ctx_index_lock);
87 spin_unlock(&__ctx_index_lock);
92 static inline unsigned long hash_mem(char *buf, int length, int bits)
94 unsigned long hash = 0;
109 if ((len & (BITS_PER_LONG/8-1)) == 0)
110 hash = hash_long(hash^l, BITS_PER_LONG);
113 return hash >> (BITS_PER_LONG - bits);
116 /****************************************
117 * rpc sec init (rsi) cache *
118 ****************************************/
120 #define RSI_HASHBITS (6)
121 #define RSI_HASHMAX (1 << RSI_HASHBITS)
122 #define RSI_HASHMASK (RSI_HASHMAX - 1)
128 char nm_name[LUSTRE_NODEMAP_NAME_LENGTH + 1];
129 wait_queue_head_t waitq;
130 rawobj_t in_handle, in_token;
131 rawobj_t out_handle, out_token;
132 int major_status, minor_status;
133 #ifdef HAVE_CACHE_HASH_SPINLOCK
134 struct rcu_head rcu_head;
138 #ifdef HAVE_CACHE_HEAD_HLIST
139 static struct hlist_head rsi_table[RSI_HASHMAX];
141 static struct cache_head *rsi_table[RSI_HASHMAX];
143 static struct cache_detail rsi_cache;
144 static struct rsi *rsi_update(struct rsi *new, struct rsi *old);
145 static struct rsi *rsi_lookup(struct rsi *item);
147 #ifdef HAVE_CACHE_DETAIL_WRITERS
148 static inline int channel_users(struct cache_detail *cd)
150 return atomic_read(&cd->writers);
153 static inline int channel_users(struct cache_detail *cd)
155 return atomic_read(&cd->readers);
159 static inline int rsi_hash(struct rsi *item)
161 return hash_mem((char *)item->in_handle.data, item->in_handle.len,
163 hash_mem((char *)item->in_token.data, item->in_token.len,
167 static inline int __rsi_match(struct rsi *item, struct rsi *tmp)
169 return (rawobj_equal(&item->in_handle, &tmp->in_handle) &&
170 rawobj_equal(&item->in_token, &tmp->in_token));
173 static void rsi_free(struct rsi *rsi)
175 rawobj_free(&rsi->in_handle);
176 rawobj_free(&rsi->in_token);
177 rawobj_free(&rsi->out_handle);
178 rawobj_free(&rsi->out_token);
181 /* See handle_channel_req() userspace for where the upcall data is read */
182 static void rsi_request(struct cache_detail *cd,
183 struct cache_head *h,
184 char **bpp, int *blen)
186 struct rsi *rsi = container_of(h, struct rsi, h);
189 /* if in_handle is null, provide kernel suggestion */
190 if (rsi->in_handle.len == 0)
191 index = gss_get_next_ctx_index();
193 qword_addhex(bpp, blen, (char *) &rsi->lustre_svc,
194 sizeof(rsi->lustre_svc));
195 qword_addhex(bpp, blen, (char *) &rsi->nid, sizeof(rsi->nid));
196 qword_addhex(bpp, blen, (char *) &index, sizeof(index));
197 qword_addhex(bpp, blen, (char *) rsi->nm_name,
198 strlen(rsi->nm_name) + 1);
199 qword_addhex(bpp, blen, rsi->in_handle.data, rsi->in_handle.len);
200 qword_addhex(bpp, blen, rsi->in_token.data, rsi->in_token.len);
204 static inline void __rsi_init(struct rsi *new, struct rsi *item)
206 new->out_handle = RAWOBJ_EMPTY;
207 new->out_token = RAWOBJ_EMPTY;
209 new->in_handle = item->in_handle;
210 item->in_handle = RAWOBJ_EMPTY;
211 new->in_token = item->in_token;
212 item->in_token = RAWOBJ_EMPTY;
214 new->lustre_svc = item->lustre_svc;
215 new->nid = item->nid;
216 memcpy(new->nm_name, item->nm_name, sizeof(item->nm_name));
217 init_waitqueue_head(&new->waitq);
220 static inline void __rsi_update(struct rsi *new, struct rsi *item)
222 LASSERT(new->out_handle.len == 0);
223 LASSERT(new->out_token.len == 0);
225 new->out_handle = item->out_handle;
226 item->out_handle = RAWOBJ_EMPTY;
227 new->out_token = item->out_token;
228 item->out_token = RAWOBJ_EMPTY;
230 new->major_status = item->major_status;
231 new->minor_status = item->minor_status;
234 #ifdef HAVE_CACHE_HASH_SPINLOCK
235 static void rsi_free_rcu(struct rcu_head *head)
237 struct rsi *rsi = container_of(head, struct rsi, rcu_head);
239 #ifdef HAVE_CACHE_HEAD_HLIST
240 LASSERT(hlist_unhashed(&rsi->h.cache_list));
242 LASSERT(rsi->h.next == NULL);
248 static void rsi_put(struct kref *ref)
250 struct rsi *rsi = container_of(ref, struct rsi, h.ref);
252 call_rcu(&rsi->rcu_head, rsi_free_rcu);
254 #else /* !HAVE_CACHE_HASH_SPINLOCK */
255 static void rsi_put(struct kref *ref)
257 struct rsi *rsi = container_of(ref, struct rsi, h.ref);
259 #ifdef HAVE_CACHE_HEAD_HLIST
260 LASSERT(hlist_unhashed(&rsi->h.cache_list));
262 LASSERT(rsi->h.next == NULL);
267 #endif /* HAVE_CACHE_HASH_SPINLOCK */
269 static int rsi_match(struct cache_head *a, struct cache_head *b)
271 struct rsi *item = container_of(a, struct rsi, h);
272 struct rsi *tmp = container_of(b, struct rsi, h);
274 return __rsi_match(item, tmp);
277 static void rsi_init(struct cache_head *cnew, struct cache_head *citem)
279 struct rsi *new = container_of(cnew, struct rsi, h);
280 struct rsi *item = container_of(citem, struct rsi, h);
282 __rsi_init(new, item);
285 static void update_rsi(struct cache_head *cnew, struct cache_head *citem)
287 struct rsi *new = container_of(cnew, struct rsi, h);
288 struct rsi *item = container_of(citem, struct rsi, h);
290 __rsi_update(new, item);
293 static struct cache_head *rsi_alloc(void)
304 static int rsi_parse(struct cache_detail *cd, char *mesg, int mlen)
308 struct rsi rsii, *rsip = NULL;
310 int status = -EINVAL;
314 memset(&rsii, 0, sizeof(rsii));
317 len = qword_get(&mesg, buf, mlen);
320 if (rawobj_alloc(&rsii.in_handle, buf, len)) {
326 len = qword_get(&mesg, buf, mlen);
329 if (rawobj_alloc(&rsii.in_token, buf, len)) {
334 rsip = rsi_lookup(&rsii);
340 expiry = get_expiry(&mesg);
344 len = qword_get(&mesg, buf, mlen);
349 status = kstrtoint(buf, 10, &rsii.major_status);
354 len = qword_get(&mesg, buf, mlen);
360 status = kstrtoint(buf, 10, &rsii.minor_status);
365 len = qword_get(&mesg, buf, mlen);
368 if (rawobj_alloc(&rsii.out_handle, buf, len)) {
374 len = qword_get(&mesg, buf, mlen);
377 if (rawobj_alloc(&rsii.out_token, buf, len)) {
382 rsii.h.expiry_time = expiry;
383 rsip = rsi_update(&rsii, rsip);
388 wake_up(&rsip->waitq);
389 cache_put(&rsip->h, &rsi_cache);
395 CERROR("rsi parse error %d\n", status);
399 static struct cache_detail rsi_cache = {
400 .hash_size = RSI_HASHMAX,
401 .hash_table = rsi_table,
402 .name = "auth.sptlrpc.init",
403 .cache_put = rsi_put,
404 .cache_request = rsi_request,
405 .cache_upcall = sunrpc_cache_pipe_upcall,
406 .cache_parse = rsi_parse,
409 .update = update_rsi,
413 static struct rsi *rsi_lookup(struct rsi *item)
415 struct cache_head *ch;
416 int hash = rsi_hash(item);
418 ch = sunrpc_cache_lookup(&rsi_cache, &item->h, hash);
420 return container_of(ch, struct rsi, h);
425 static struct rsi *rsi_update(struct rsi *new, struct rsi *old)
427 struct cache_head *ch;
428 int hash = rsi_hash(new);
430 ch = sunrpc_cache_update(&rsi_cache, &new->h, &old->h, hash);
432 return container_of(ch, struct rsi, h);
437 /****************************************
438 * rpc sec context (rsc) cache *
439 ****************************************/
441 #define RSC_HASHBITS (10)
442 #define RSC_HASHMAX (1 << RSC_HASHBITS)
443 #define RSC_HASHMASK (RSC_HASHMAX - 1)
447 struct obd_device *target;
449 struct gss_svc_ctx ctx;
450 #ifdef HAVE_CACHE_HASH_SPINLOCK
451 struct rcu_head rcu_head;
455 #ifdef HAVE_CACHE_HEAD_HLIST
456 static struct hlist_head rsc_table[RSC_HASHMAX];
458 static struct cache_head *rsc_table[RSC_HASHMAX];
460 static struct cache_detail rsc_cache;
461 static struct rsc *rsc_update(struct rsc *new, struct rsc *old);
462 static struct rsc *rsc_lookup(struct rsc *item);
464 static void rsc_free(struct rsc *rsci)
466 rawobj_free(&rsci->handle);
467 rawobj_free(&rsci->ctx.gsc_rvs_hdl);
468 lgss_delete_sec_context(&rsci->ctx.gsc_mechctx);
471 static inline int rsc_hash(struct rsc *rsci)
473 return hash_mem((char *)rsci->handle.data,
474 rsci->handle.len, RSC_HASHBITS);
477 static inline int __rsc_match(struct rsc *new, struct rsc *tmp)
479 return rawobj_equal(&new->handle, &tmp->handle);
482 static inline void __rsc_init(struct rsc *new, struct rsc *tmp)
484 new->handle = tmp->handle;
485 tmp->handle = RAWOBJ_EMPTY;
488 memset(&new->ctx, 0, sizeof(new->ctx));
489 new->ctx.gsc_rvs_hdl = RAWOBJ_EMPTY;
492 static inline void __rsc_update(struct rsc *new, struct rsc *tmp)
495 memset(&tmp->ctx, 0, sizeof(tmp->ctx));
496 tmp->ctx.gsc_rvs_hdl = RAWOBJ_EMPTY;
497 tmp->ctx.gsc_mechctx = NULL;
500 memset(&new->ctx.gsc_seqdata, 0, sizeof(new->ctx.gsc_seqdata));
501 spin_lock_init(&new->ctx.gsc_seqdata.ssd_lock);
504 #ifdef HAVE_CACHE_HASH_SPINLOCK
505 static void rsc_free_rcu(struct rcu_head *head)
507 struct rsc *rsci = container_of(head, struct rsc, rcu_head);
509 #ifdef HAVE_CACHE_HEAD_HLIST
510 LASSERT(hlist_unhashed(&rsci->h.cache_list));
512 LASSERT(rsci->h.next == NULL);
514 rawobj_free(&rsci->handle);
518 static void rsc_put(struct kref *ref)
520 struct rsc *rsci = container_of(ref, struct rsc, h.ref);
522 rawobj_free(&rsci->ctx.gsc_rvs_hdl);
523 lgss_delete_sec_context(&rsci->ctx.gsc_mechctx);
524 call_rcu(&rsci->rcu_head, rsc_free_rcu);
526 #else /* !HAVE_CACHE_HASH_SPINLOCK */
527 static void rsc_put(struct kref *ref)
529 struct rsc *rsci = container_of(ref, struct rsc, h.ref);
531 #ifdef HAVE_CACHE_HEAD_HLIST
532 LASSERT(hlist_unhashed(&rsci->h.cache_list));
534 LASSERT(rsci->h.next == NULL);
539 #endif /* HAVE_CACHE_HASH_SPINLOCK */
541 static int rsc_match(struct cache_head *a, struct cache_head *b)
543 struct rsc *new = container_of(a, struct rsc, h);
544 struct rsc *tmp = container_of(b, struct rsc, h);
546 return __rsc_match(new, tmp);
549 static void rsc_init(struct cache_head *cnew, struct cache_head *ctmp)
551 struct rsc *new = container_of(cnew, struct rsc, h);
552 struct rsc *tmp = container_of(ctmp, struct rsc, h);
554 __rsc_init(new, tmp);
557 static void update_rsc(struct cache_head *cnew, struct cache_head *ctmp)
559 struct rsc *new = container_of(cnew, struct rsc, h);
560 struct rsc *tmp = container_of(ctmp, struct rsc, h);
562 __rsc_update(new, tmp);
565 static struct cache_head * rsc_alloc(void)
576 static int rsc_parse(struct cache_detail *cd, char *mesg, int mlen)
579 int len, rv, tmp_int;
580 struct rsc rsci, *rscp = NULL;
582 int status = -EINVAL;
583 struct gss_api_mech *gm = NULL;
585 memset(&rsci, 0, sizeof(rsci));
588 len = qword_get(&mesg, buf, mlen);
589 if (len < 0) goto out;
591 if (rawobj_alloc(&rsci.handle, buf, len))
596 expiry = get_expiry(&mesg);
602 rv = get_int(&mesg, &tmp_int);
604 CERROR("fail to get remote flag\n");
607 rsci.ctx.gsc_remote = (tmp_int != 0);
610 rv = get_int(&mesg, &tmp_int);
612 CERROR("fail to get root user flag\n");
615 rsci.ctx.gsc_usr_root = (tmp_int != 0);
618 rv = get_int(&mesg, &tmp_int);
620 CERROR("fail to get mds user flag\n");
623 rsci.ctx.gsc_usr_mds = (tmp_int != 0);
626 rv = get_int(&mesg, &tmp_int);
628 CERROR("fail to get oss user flag\n");
631 rsci.ctx.gsc_usr_oss = (tmp_int != 0);
634 rv = get_int(&mesg, (int *) &rsci.ctx.gsc_mapped_uid);
636 CERROR("fail to get mapped uid\n");
640 rscp = rsc_lookup(&rsci);
644 /* uid, or NEGATIVE */
645 rv = get_int(&mesg, (int *) &rsci.ctx.gsc_uid);
649 CERROR("NOENT? set rsc entry negative\n");
650 set_bit(CACHE_NEGATIVE, &rsci.h.flags);
656 if (get_int(&mesg, (int *) &rsci.ctx.gsc_gid))
660 len = qword_get(&mesg, buf, mlen);
663 gm = lgss_name_to_mech(buf);
664 status = -EOPNOTSUPP;
669 /* mech-specific data: */
670 len = qword_get(&mesg, buf, mlen);
675 tmp_buf.data = (unsigned char *)buf;
676 if (lgss_import_sec_context(&tmp_buf, gm,
677 &rsci.ctx.gsc_mechctx))
680 /* set to seconds since machine booted */
681 expiry = ktime_get_seconds();
683 /* currently the expiry time passed down from user-space
684 * is invalid, here we retrive it from mech.
686 if (lgss_inquire_context(rsci.ctx.gsc_mechctx, &ctx_expiry)) {
687 CERROR("unable to get expire time, drop it\n");
691 /* ctx_expiry is the number of seconds since Jan 1 1970.
692 * We want just the number of seconds into the future.
694 expiry += ctx_expiry - ktime_get_real_seconds();
697 rsci.h.expiry_time = expiry;
698 rscp = rsc_update(&rsci, rscp);
705 cache_put(&rscp->h, &rsc_cache);
710 CERROR("parse rsc error %d\n", status);
714 static struct cache_detail rsc_cache = {
715 .hash_size = RSC_HASHMAX,
716 .hash_table = rsc_table,
717 .name = "auth.sptlrpc.context",
718 .cache_put = rsc_put,
719 .cache_parse = rsc_parse,
722 .update = update_rsc,
726 static struct rsc *rsc_lookup(struct rsc *item)
728 struct cache_head *ch;
729 int hash = rsc_hash(item);
731 ch = sunrpc_cache_lookup(&rsc_cache, &item->h, hash);
733 return container_of(ch, struct rsc, h);
738 static struct rsc *rsc_update(struct rsc *new, struct rsc *old)
740 struct cache_head *ch;
741 int hash = rsc_hash(new);
743 ch = sunrpc_cache_update(&rsc_cache, &new->h, &old->h, hash);
745 return container_of(ch, struct rsc, h);
750 #define COMPAT_RSC_PUT(item, cd) cache_put((item), (cd))
752 /****************************************
754 ****************************************/
756 static struct rsc *gss_svc_searchbyctx(rawobj_t *handle)
761 memset(&rsci, 0, sizeof(rsci));
762 if (rawobj_dup(&rsci.handle, handle))
765 found = rsc_lookup(&rsci);
769 if (cache_check(&rsc_cache, &found->h, NULL))
774 int gss_svc_upcall_install_rvs_ctx(struct obd_import *imp,
775 struct gss_sec *gsec,
776 struct gss_cli_ctx *gctx)
778 struct rsc rsci, *rscp = NULL;
784 memset(&rsci, 0, sizeof(rsci));
786 if (rawobj_alloc(&rsci.handle, (char *) &gsec->gs_rvs_hdl,
787 sizeof(gsec->gs_rvs_hdl)))
788 GOTO(out, rc = -ENOMEM);
790 rscp = rsc_lookup(&rsci);
792 GOTO(out, rc = -ENOMEM);
794 major = lgss_copy_reverse_context(gctx->gc_mechctx,
795 &rsci.ctx.gsc_mechctx);
796 if (major != GSS_S_COMPLETE)
797 GOTO(out, rc = -ENOMEM);
799 if (lgss_inquire_context(rsci.ctx.gsc_mechctx, &ctx_expiry)) {
800 CERROR("unable to get expire time, drop it\n");
801 GOTO(out, rc = -EINVAL);
803 rsci.h.expiry_time = ctx_expiry;
805 switch (imp->imp_obd->u.cli.cl_sp_to) {
807 rsci.ctx.gsc_usr_mds = 1;
810 rsci.ctx.gsc_usr_oss = 1;
813 rsci.ctx.gsc_usr_root = 1;
816 /* by convention, all 3 set to 1 means MGS */
817 rsci.ctx.gsc_usr_mds = 1;
818 rsci.ctx.gsc_usr_oss = 1;
819 rsci.ctx.gsc_usr_root = 1;
825 rscp = rsc_update(&rsci, rscp);
827 GOTO(out, rc = -ENOMEM);
829 rscp->target = imp->imp_obd;
830 rawobj_dup(&gctx->gc_svc_handle, &rscp->handle);
832 CWARN("create reverse svc ctx %p to %s: idx %#llx\n",
833 &rscp->ctx, obd2cli_tgt(imp->imp_obd), gsec->gs_rvs_hdl);
837 cache_put(&rscp->h, &rsc_cache);
841 CERROR("create reverse svc ctx: idx %#llx, rc %d\n",
842 gsec->gs_rvs_hdl, rc);
846 int gss_svc_upcall_expire_rvs_ctx(rawobj_t *handle)
848 const time64_t expire = 20;
851 rscp = gss_svc_searchbyctx(handle);
853 CDEBUG(D_SEC, "reverse svcctx %p (rsc %p) expire soon\n",
856 rscp->h.expiry_time = ktime_get_real_seconds() + expire;
857 COMPAT_RSC_PUT(&rscp->h, &rsc_cache);
862 int gss_svc_upcall_dup_handle(rawobj_t *handle, struct gss_svc_ctx *ctx)
864 struct rsc *rscp = container_of(ctx, struct rsc, ctx);
866 return rawobj_dup(handle, &rscp->handle);
869 int gss_svc_upcall_update_sequence(rawobj_t *handle, __u32 seq)
873 rscp = gss_svc_searchbyctx(handle);
875 CDEBUG(D_SEC, "reverse svcctx %p (rsc %p) update seq to %u\n",
876 &rscp->ctx, rscp, seq + 1);
878 rscp->ctx.gsc_rvs_seq = seq + 1;
879 COMPAT_RSC_PUT(&rscp->h, &rsc_cache);
884 static struct cache_deferred_req* cache_upcall_defer(struct cache_req *req)
888 static struct cache_req cache_upcall_chandle = { cache_upcall_defer };
890 int gss_svc_upcall_handle_init(struct ptlrpc_request *req,
891 struct gss_svc_reqctx *grctx,
892 struct gss_wire_ctx *gw,
893 struct obd_device *target,
898 struct ptlrpc_reply_state *rs;
899 struct rsc *rsci = NULL;
900 struct rsi *rsip = NULL, rsikey;
901 wait_queue_entry_t wait;
902 int replen = sizeof(struct ptlrpc_body);
903 struct gss_rep_header *rephdr;
905 int rc = SECSVC_DROP;
908 memset(&rsikey, 0, sizeof(rsikey));
909 rsikey.lustre_svc = lustre_svc;
910 /* In case of MR, rq_peer is not the NID from which request is received,
911 * but primary NID of peer.
912 * So we need rq_source, which contains the NID actually in use.
914 rsikey.nid = (__u64) req->rq_source.nid;
915 nodemap_test_nid(req->rq_peer.nid, rsikey.nm_name,
916 sizeof(rsikey.nm_name));
918 /* duplicate context handle. for INIT it always 0 */
919 if (rawobj_dup(&rsikey.in_handle, &gw->gw_handle)) {
920 CERROR("fail to dup context handle\n");
924 if (rawobj_dup(&rsikey.in_token, in_token)) {
925 CERROR("can't duplicate token\n");
926 rawobj_free(&rsikey.in_handle);
930 rsip = rsi_lookup(&rsikey);
933 CERROR("error in rsi_lookup.\n");
935 if (!gss_pack_err_notify(req, GSS_S_FAILURE, 0))
936 rc = SECSVC_COMPLETE;
941 cache_get(&rsip->h); /* take an extra ref */
943 add_wait_queue(&rsip->waitq, &wait);
946 /* Note each time cache_check() will drop a reference if return
947 * non-zero. We hold an extra reference on initial rsip, but must
948 * take care of following calls. */
949 rc = cache_check(&rsi_cache, &rsip->h, &cache_upcall_chandle);
958 cache_read_lock(&rsi_cache);
959 valid = test_bit(CACHE_VALID, &rsip->h.flags);
961 set_current_state(TASK_INTERRUPTIBLE);
962 cache_read_unlock(&rsi_cache);
965 unsigned long timeout;
967 timeout = cfs_time_seconds(GSS_SVC_UPCALL_TIMEOUT);
968 schedule_timeout(timeout);
973 CWARN("waited %ds timeout, drop\n", GSS_SVC_UPCALL_TIMEOUT);
977 CDEBUG(D_SEC, "cache_check return ENOENT, drop\n");
980 /* if not the first check, we have to release the extra
981 * reference we just added on it. */
983 cache_put(&rsip->h, &rsi_cache);
984 CDEBUG(D_SEC, "cache_check is good\n");
988 remove_wait_queue(&rsip->waitq, &wait);
989 cache_put(&rsip->h, &rsi_cache);
992 GOTO(out, rc = SECSVC_DROP);
995 rsci = gss_svc_searchbyctx(&rsip->out_handle);
997 CERROR("authentication failed\n");
999 /* gss mechanism returned major and minor code so we return
1000 * those in error message */
1001 if (!gss_pack_err_notify(req, rsip->major_status,
1002 rsip->minor_status))
1003 rc = SECSVC_COMPLETE;
1007 cache_get(&rsci->h);
1008 grctx->src_ctx = &rsci->ctx;
1011 if (gw->gw_flags & LUSTRE_GSS_PACK_KCSUM) {
1012 grctx->src_ctx->gsc_mechctx->hash_func = gss_digest_hash;
1013 } else if (!strcmp(grctx->src_ctx->gsc_mechctx->mech_type->gm_name,
1015 !krb5_allow_old_client_csum) {
1016 CWARN("%s: deny connection from '%s' due to missing 'krb_csum' feature, set 'sptlrpc.gss.krb5_allow_old_client_csum=1' to allow, but recommend client upgrade: rc = %d\n",
1017 target->obd_name, libcfs_nid2str(req->rq_peer.nid),
1019 GOTO(out, rc = SECSVC_DROP);
1021 grctx->src_ctx->gsc_mechctx->hash_func =
1022 gss_digest_hash_compat;
1025 if (rawobj_dup(&rsci->ctx.gsc_rvs_hdl, rvs_hdl)) {
1026 CERROR("failed duplicate reverse handle\n");
1030 rsci->target = target;
1032 CDEBUG(D_SEC, "server create rsc %p(%u->%s)\n",
1033 rsci, rsci->ctx.gsc_uid, libcfs_nid2str(req->rq_peer.nid));
1035 if (rsip->out_handle.len > PTLRPC_GSS_MAX_HANDLE_SIZE) {
1036 CERROR("handle size %u too large\n", rsip->out_handle.len);
1037 GOTO(out, rc = SECSVC_DROP);
1040 grctx->src_init = 1;
1041 grctx->src_reserve_len = cfs_size_round4(rsip->out_token.len);
1043 rc = lustre_pack_reply_v2(req, 1, &replen, NULL, 0);
1045 CERROR("failed to pack reply: %d\n", rc);
1046 GOTO(out, rc = SECSVC_DROP);
1049 rs = req->rq_reply_state;
1050 LASSERT(rs->rs_repbuf->lm_bufcount == 3);
1051 LASSERT(rs->rs_repbuf->lm_buflens[0] >=
1052 sizeof(*rephdr) + rsip->out_handle.len);
1053 LASSERT(rs->rs_repbuf->lm_buflens[2] >= rsip->out_token.len);
1055 rephdr = lustre_msg_buf(rs->rs_repbuf, 0, 0);
1056 rephdr->gh_version = PTLRPC_GSS_VERSION;
1057 rephdr->gh_flags = 0;
1058 rephdr->gh_proc = PTLRPC_GSS_PROC_ERR;
1059 rephdr->gh_major = rsip->major_status;
1060 rephdr->gh_minor = rsip->minor_status;
1061 rephdr->gh_seqwin = GSS_SEQ_WIN;
1062 rephdr->gh_handle.len = rsip->out_handle.len;
1063 memcpy(rephdr->gh_handle.data, rsip->out_handle.data,
1064 rsip->out_handle.len);
1066 memcpy(lustre_msg_buf(rs->rs_repbuf, 2, 0), rsip->out_token.data,
1067 rsip->out_token.len);
1069 rs->rs_repdata_len = lustre_shrink_msg(rs->rs_repbuf, 2,
1070 rsip->out_token.len, 0);
1075 /* it looks like here we should put rsip also, but this mess up
1076 * with NFS cache mgmt code... FIXME
1079 * rsi_put(&rsip->h, &rsi_cache); */
1082 /* if anything went wrong, we don't keep the context too */
1083 if (rc != SECSVC_OK)
1084 set_bit(CACHE_NEGATIVE, &rsci->h.flags);
1086 CDEBUG(D_SEC, "create rsc with idx %#llx\n",
1087 gss_handle_to_u64(&rsci->handle));
1089 COMPAT_RSC_PUT(&rsci->h, &rsc_cache);
1094 struct gss_svc_ctx *gss_svc_upcall_get_ctx(struct ptlrpc_request *req,
1095 struct gss_wire_ctx *gw)
1099 rsc = gss_svc_searchbyctx(&gw->gw_handle);
1101 CWARN("Invalid gss ctx idx %#llx from %s\n",
1102 gss_handle_to_u64(&gw->gw_handle),
1103 libcfs_nid2str(req->rq_peer.nid));
1110 void gss_svc_upcall_put_ctx(struct gss_svc_ctx *ctx)
1112 struct rsc *rsc = container_of(ctx, struct rsc, ctx);
1114 COMPAT_RSC_PUT(&rsc->h, &rsc_cache);
1117 void gss_svc_upcall_destroy_ctx(struct gss_svc_ctx *ctx)
1119 struct rsc *rsc = container_of(ctx, struct rsc, ctx);
1121 /* can't be found */
1122 set_bit(CACHE_NEGATIVE, &rsc->h.flags);
1123 /* to be removed at next scan */
1124 rsc->h.expiry_time = 1;
1127 int __init gss_init_svc_upcall(void)
1132 * this helps reducing context index confliction. after server reboot,
1133 * conflicting request from clients might be filtered out by initial
1134 * sequence number checking, thus no chance to sent error notification
1137 get_random_bytes(&__ctx_index, sizeof(__ctx_index));
1139 #ifdef HAVE_CACHE_HEAD_HLIST
1140 for (i = 0; i < rsi_cache.hash_size; i++)
1141 INIT_HLIST_HEAD(&rsi_cache.hash_table[i]);
1143 rc = cache_register_net(&rsi_cache, &init_net);
1147 #ifdef HAVE_CACHE_HEAD_HLIST
1148 for (i = 0; i < rsc_cache.hash_size; i++)
1149 INIT_HLIST_HEAD(&rsc_cache.hash_table[i]);
1151 rc = cache_register_net(&rsc_cache, &init_net);
1153 cache_unregister_net(&rsi_cache, &init_net);
1157 /* FIXME this looks stupid. we intend to give lsvcgssd a chance to open
1158 * the init upcall channel, otherwise there's big chance that the first
1159 * upcall issued before the channel be opened thus nfsv4 cache code will
1160 * drop the request directly, thus lead to unnecessary recovery time.
1161 * Here we wait at minimum 1.5 seconds.
1163 for (i = 0; i < 6; i++) {
1164 if (channel_users(&rsi_cache) > 0)
1166 schedule_timeout_uninterruptible(cfs_time_seconds(1) / 4);
1169 if (channel_users(&rsi_cache) == 0)
1171 "Init channel is not opened by lsvcgssd, following request might be dropped until lsvcgssd is active\n");
1176 void gss_exit_svc_upcall(void)
1178 cache_purge(&rsi_cache);
1179 cache_unregister_net(&rsi_cache, &init_net);
1181 cache_purge(&rsc_cache);
1182 cache_unregister_net(&rsc_cache, &init_net);