4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License version 2 only,
8 * as published by the Free Software Foundation.
10 * This program is distributed in the hope that it will be useful, but
11 * WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
13 * General Public License version 2 for more details (a copy is included
14 * in the LICENSE file that accompanied this code).
16 * You should have received a copy of the GNU General Public License
17 * version 2 along with this program; If not, see
18 * http://www.gnu.org/licenses/gpl-2.0.html
23 * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
24 * Use is subject to license terms.
26 * Copyright (c) 2011, 2017, Intel Corporation.
29 * This file is part of Lustre, http://www.lustre.org/
30 * Lustre is a trademark of Sun Microsystems, Inc.
34 * Author: Eric Mei <ericm@clusterfs.com>
37 #define DEBUG_SUBSYSTEM S_SEC
39 #include <linux/user_namespace.h>
40 #ifdef HAVE_UIDGID_HEADER
41 # include <linux/uidgid.h>
43 #include <linux/crypto.h>
44 #include <linux/key.h>
46 #include <libcfs/libcfs.h>
48 #include <obd_class.h>
49 #include <obd_support.h>
50 #include <lustre_net.h>
51 #include <lustre_import.h>
52 #include <lustre_dlm.h>
53 #include <lustre_sec.h>
55 #include "ptlrpc_internal.h"
57 static int send_sepol;
58 module_param(send_sepol, int, 0644);
59 MODULE_PARM_DESC(send_sepol, "Client sends SELinux policy status");
61 /***********************************************
63 ***********************************************/
65 static rwlock_t policy_lock;
66 static struct ptlrpc_sec_policy *policies[SPTLRPC_POLICY_MAX] = {
70 int sptlrpc_register_policy(struct ptlrpc_sec_policy *policy)
72 __u16 number = policy->sp_policy;
74 LASSERT(policy->sp_name);
75 LASSERT(policy->sp_cops);
76 LASSERT(policy->sp_sops);
78 if (number >= SPTLRPC_POLICY_MAX)
81 write_lock(&policy_lock);
82 if (unlikely(policies[number])) {
83 write_unlock(&policy_lock);
86 policies[number] = policy;
87 write_unlock(&policy_lock);
89 CDEBUG(D_SEC, "%s: registered\n", policy->sp_name);
92 EXPORT_SYMBOL(sptlrpc_register_policy);
94 int sptlrpc_unregister_policy(struct ptlrpc_sec_policy *policy)
96 __u16 number = policy->sp_policy;
98 LASSERT(number < SPTLRPC_POLICY_MAX);
100 write_lock(&policy_lock);
101 if (unlikely(policies[number] == NULL)) {
102 write_unlock(&policy_lock);
103 CERROR("%s: already unregistered\n", policy->sp_name);
107 LASSERT(policies[number] == policy);
108 policies[number] = NULL;
109 write_unlock(&policy_lock);
111 CDEBUG(D_SEC, "%s: unregistered\n", policy->sp_name);
114 EXPORT_SYMBOL(sptlrpc_unregister_policy);
117 struct ptlrpc_sec_policy * sptlrpc_wireflavor2policy(__u32 flavor)
119 static DEFINE_MUTEX(load_mutex);
120 static atomic_t loaded = ATOMIC_INIT(0);
121 struct ptlrpc_sec_policy *policy;
122 __u16 number = SPTLRPC_FLVR_POLICY(flavor);
125 if (number >= SPTLRPC_POLICY_MAX)
129 read_lock(&policy_lock);
130 policy = policies[number];
131 if (policy && !try_module_get(policy->sp_owner))
134 flag = atomic_read(&loaded);
135 read_unlock(&policy_lock);
137 if (policy != NULL || flag != 0 ||
138 number != SPTLRPC_POLICY_GSS)
141 /* try to load gss module, once */
142 mutex_lock(&load_mutex);
143 if (atomic_read(&loaded) == 0) {
144 if (request_module("ptlrpc_gss") == 0)
146 "module ptlrpc_gss loaded on demand\n");
148 CERROR("Unable to load module ptlrpc_gss\n");
150 atomic_set(&loaded, 1);
152 mutex_unlock(&load_mutex);
158 __u32 sptlrpc_name2flavor_base(const char *name)
160 if (!strcmp(name, "null"))
161 return SPTLRPC_FLVR_NULL;
162 if (!strcmp(name, "plain"))
163 return SPTLRPC_FLVR_PLAIN;
164 if (!strcmp(name, "gssnull"))
165 return SPTLRPC_FLVR_GSSNULL;
166 if (!strcmp(name, "krb5n"))
167 return SPTLRPC_FLVR_KRB5N;
168 if (!strcmp(name, "krb5a"))
169 return SPTLRPC_FLVR_KRB5A;
170 if (!strcmp(name, "krb5i"))
171 return SPTLRPC_FLVR_KRB5I;
172 if (!strcmp(name, "krb5p"))
173 return SPTLRPC_FLVR_KRB5P;
174 if (!strcmp(name, "skn"))
175 return SPTLRPC_FLVR_SKN;
176 if (!strcmp(name, "ska"))
177 return SPTLRPC_FLVR_SKA;
178 if (!strcmp(name, "ski"))
179 return SPTLRPC_FLVR_SKI;
180 if (!strcmp(name, "skpi"))
181 return SPTLRPC_FLVR_SKPI;
183 return SPTLRPC_FLVR_INVALID;
185 EXPORT_SYMBOL(sptlrpc_name2flavor_base);
187 const char *sptlrpc_flavor2name_base(__u32 flvr)
189 __u32 base = SPTLRPC_FLVR_BASE(flvr);
191 if (base == SPTLRPC_FLVR_BASE(SPTLRPC_FLVR_NULL))
193 else if (base == SPTLRPC_FLVR_BASE(SPTLRPC_FLVR_PLAIN))
195 else if (base == SPTLRPC_FLVR_BASE(SPTLRPC_FLVR_GSSNULL))
197 else if (base == SPTLRPC_FLVR_BASE(SPTLRPC_FLVR_KRB5N))
199 else if (base == SPTLRPC_FLVR_BASE(SPTLRPC_FLVR_KRB5A))
201 else if (base == SPTLRPC_FLVR_BASE(SPTLRPC_FLVR_KRB5I))
203 else if (base == SPTLRPC_FLVR_BASE(SPTLRPC_FLVR_KRB5P))
205 else if (base == SPTLRPC_FLVR_BASE(SPTLRPC_FLVR_SKN))
207 else if (base == SPTLRPC_FLVR_BASE(SPTLRPC_FLVR_SKA))
209 else if (base == SPTLRPC_FLVR_BASE(SPTLRPC_FLVR_SKI))
211 else if (base == SPTLRPC_FLVR_BASE(SPTLRPC_FLVR_SKPI))
214 CERROR("invalid wire flavor 0x%x\n", flvr);
217 EXPORT_SYMBOL(sptlrpc_flavor2name_base);
219 char *sptlrpc_flavor2name_bulk(struct sptlrpc_flavor *sf,
220 char *buf, int bufsize)
222 if (SPTLRPC_FLVR_POLICY(sf->sf_rpc) == SPTLRPC_POLICY_PLAIN)
223 snprintf(buf, bufsize, "hash:%s",
224 sptlrpc_get_hash_name(sf->u_bulk.hash.hash_alg));
226 snprintf(buf, bufsize, "%s",
227 sptlrpc_flavor2name_base(sf->sf_rpc));
229 buf[bufsize - 1] = '\0';
232 EXPORT_SYMBOL(sptlrpc_flavor2name_bulk);
234 char *sptlrpc_flavor2name(struct sptlrpc_flavor *sf, char *buf, int bufsize)
236 snprintf(buf, bufsize, "%s", sptlrpc_flavor2name_base(sf->sf_rpc));
239 * currently we don't support customized bulk specification for
240 * flavors other than plain
242 if (SPTLRPC_FLVR_POLICY(sf->sf_rpc) == SPTLRPC_POLICY_PLAIN) {
246 sptlrpc_flavor2name_bulk(sf, &bspec[1], sizeof(bspec) - 1);
247 strncat(buf, bspec, bufsize);
250 buf[bufsize - 1] = '\0';
253 EXPORT_SYMBOL(sptlrpc_flavor2name);
255 char *sptlrpc_secflags2str(__u32 flags, char *buf, int bufsize)
259 if (flags & PTLRPC_SEC_FL_REVERSE)
260 strlcat(buf, "reverse,", bufsize);
261 if (flags & PTLRPC_SEC_FL_ROOTONLY)
262 strlcat(buf, "rootonly,", bufsize);
263 if (flags & PTLRPC_SEC_FL_UDESC)
264 strlcat(buf, "udesc,", bufsize);
265 if (flags & PTLRPC_SEC_FL_BULK)
266 strlcat(buf, "bulk,", bufsize);
268 strlcat(buf, "-,", bufsize);
272 EXPORT_SYMBOL(sptlrpc_secflags2str);
274 /**************************************************
275 * client context APIs *
276 **************************************************/
279 struct ptlrpc_cli_ctx *get_my_ctx(struct ptlrpc_sec *sec)
281 struct vfs_cred vcred;
282 int create = 1, remove_dead = 1;
285 LASSERT(sec->ps_policy->sp_cops->lookup_ctx);
287 if (sec->ps_flvr.sf_flags & (PTLRPC_SEC_FL_REVERSE |
288 PTLRPC_SEC_FL_ROOTONLY)) {
291 if (sec->ps_flvr.sf_flags & PTLRPC_SEC_FL_REVERSE) {
296 vcred.vc_uid = from_kuid(&init_user_ns, current_uid());
297 vcred.vc_gid = from_kgid(&init_user_ns, current_gid());
300 return sec->ps_policy->sp_cops->lookup_ctx(sec, &vcred, create,
304 struct ptlrpc_cli_ctx *sptlrpc_cli_ctx_get(struct ptlrpc_cli_ctx *ctx)
306 atomic_inc(&ctx->cc_refcount);
309 EXPORT_SYMBOL(sptlrpc_cli_ctx_get);
311 void sptlrpc_cli_ctx_put(struct ptlrpc_cli_ctx *ctx, int sync)
313 struct ptlrpc_sec *sec = ctx->cc_sec;
316 LASSERT_ATOMIC_POS(&ctx->cc_refcount);
318 if (!atomic_dec_and_test(&ctx->cc_refcount))
321 sec->ps_policy->sp_cops->release_ctx(sec, ctx, sync);
323 EXPORT_SYMBOL(sptlrpc_cli_ctx_put);
326 * Expire the client context immediately.
328 * \pre Caller must hold at least 1 reference on the \a ctx.
330 void sptlrpc_cli_ctx_expire(struct ptlrpc_cli_ctx *ctx)
332 LASSERT(ctx->cc_ops->die);
333 ctx->cc_ops->die(ctx, 0);
335 EXPORT_SYMBOL(sptlrpc_cli_ctx_expire);
338 * To wake up the threads who are waiting for this client context. Called
339 * after some status change happened on \a ctx.
341 void sptlrpc_cli_ctx_wakeup(struct ptlrpc_cli_ctx *ctx)
343 struct ptlrpc_request *req, *next;
345 spin_lock(&ctx->cc_lock);
346 list_for_each_entry_safe(req, next, &ctx->cc_req_list,
348 list_del_init(&req->rq_ctx_chain);
349 ptlrpc_client_wake_req(req);
351 spin_unlock(&ctx->cc_lock);
353 EXPORT_SYMBOL(sptlrpc_cli_ctx_wakeup);
355 int sptlrpc_cli_ctx_display(struct ptlrpc_cli_ctx *ctx, char *buf, int bufsize)
357 LASSERT(ctx->cc_ops);
359 if (ctx->cc_ops->display == NULL)
362 return ctx->cc_ops->display(ctx, buf, bufsize);
365 static int import_sec_check_expire(struct obd_import *imp)
369 spin_lock(&imp->imp_lock);
370 if (imp->imp_sec_expire &&
371 imp->imp_sec_expire < ktime_get_real_seconds()) {
373 imp->imp_sec_expire = 0;
375 spin_unlock(&imp->imp_lock);
380 CDEBUG(D_SEC, "found delayed sec adapt expired, do it now\n");
381 return sptlrpc_import_sec_adapt(imp, NULL, NULL);
385 * Get and validate the client side ptlrpc security facilities from
386 * \a imp. There is a race condition on client reconnect when the import is
387 * being destroyed while there are outstanding client bound requests. In
388 * this case do not output any error messages if import secuity is not
391 * \param[in] imp obd import associated with client
392 * \param[out] sec client side ptlrpc security
394 * \retval 0 if security retrieved successfully
395 * \retval -ve errno if there was a problem
397 static int import_sec_validate_get(struct obd_import *imp,
398 struct ptlrpc_sec **sec)
402 if (unlikely(imp->imp_sec_expire)) {
403 rc = import_sec_check_expire(imp);
408 *sec = sptlrpc_import_sec_ref(imp);
410 CERROR("import %p (%s) with no sec\n",
411 imp, ptlrpc_import_state_name(imp->imp_state));
415 if (unlikely((*sec)->ps_dying)) {
416 CERROR("attempt to use dying sec %p\n", sec);
417 sptlrpc_sec_put(*sec);
425 * Given a \a req, find or allocate an appropriate context for it.
426 * \pre req->rq_cli_ctx == NULL.
428 * \retval 0 succeed, and req->rq_cli_ctx is set.
429 * \retval -ev error number, and req->rq_cli_ctx == NULL.
431 int sptlrpc_req_get_ctx(struct ptlrpc_request *req)
433 struct obd_import *imp = req->rq_import;
434 struct ptlrpc_sec *sec;
438 LASSERT(!req->rq_cli_ctx);
441 rc = import_sec_validate_get(imp, &sec);
445 req->rq_cli_ctx = get_my_ctx(sec);
447 sptlrpc_sec_put(sec);
449 if (!req->rq_cli_ctx) {
450 CERROR("req %p: fail to get context\n", req);
451 RETURN(-ECONNREFUSED);
458 * Drop the context for \a req.
459 * \pre req->rq_cli_ctx != NULL.
460 * \post req->rq_cli_ctx == NULL.
462 * If \a sync == 0, this function should return quickly without sleep;
463 * otherwise it might trigger and wait for the whole process of sending
464 * an context-destroying rpc to server.
466 void sptlrpc_req_put_ctx(struct ptlrpc_request *req, int sync)
471 LASSERT(req->rq_cli_ctx);
473 /* request might be asked to release earlier while still
474 * in the context waiting list.
476 if (!list_empty(&req->rq_ctx_chain)) {
477 spin_lock(&req->rq_cli_ctx->cc_lock);
478 list_del_init(&req->rq_ctx_chain);
479 spin_unlock(&req->rq_cli_ctx->cc_lock);
482 sptlrpc_cli_ctx_put(req->rq_cli_ctx, sync);
483 req->rq_cli_ctx = NULL;
488 int sptlrpc_req_ctx_switch(struct ptlrpc_request *req,
489 struct ptlrpc_cli_ctx *oldctx,
490 struct ptlrpc_cli_ctx *newctx)
492 struct sptlrpc_flavor old_flvr;
493 char *reqmsg = NULL; /* to workaround old gcc */
497 LASSERT(req->rq_reqmsg);
498 LASSERT(req->rq_reqlen);
499 LASSERT(req->rq_replen);
501 CDEBUG(D_SEC, "req %p: switch ctx %p(%u->%s) -> %p(%u->%s), "
502 "switch sec %p(%s) -> %p(%s)\n", req,
503 oldctx, oldctx->cc_vcred.vc_uid, sec2target_str(oldctx->cc_sec),
504 newctx, newctx->cc_vcred.vc_uid, sec2target_str(newctx->cc_sec),
505 oldctx->cc_sec, oldctx->cc_sec->ps_policy->sp_name,
506 newctx->cc_sec, newctx->cc_sec->ps_policy->sp_name);
509 old_flvr = req->rq_flvr;
511 /* save request message */
512 reqmsg_size = req->rq_reqlen;
513 if (reqmsg_size != 0) {
514 OBD_ALLOC_LARGE(reqmsg, reqmsg_size);
517 memcpy(reqmsg, req->rq_reqmsg, reqmsg_size);
520 /* release old req/rep buf */
521 req->rq_cli_ctx = oldctx;
522 sptlrpc_cli_free_reqbuf(req);
523 sptlrpc_cli_free_repbuf(req);
524 req->rq_cli_ctx = newctx;
526 /* recalculate the flavor */
527 sptlrpc_req_set_flavor(req, 0);
529 /* alloc new request buffer
530 * we don't need to alloc reply buffer here, leave it to the
531 * rest procedure of ptlrpc */
532 if (reqmsg_size != 0) {
533 rc = sptlrpc_cli_alloc_reqbuf(req, reqmsg_size);
535 LASSERT(req->rq_reqmsg);
536 memcpy(req->rq_reqmsg, reqmsg, reqmsg_size);
538 CWARN("failed to alloc reqbuf: %d\n", rc);
539 req->rq_flvr = old_flvr;
542 OBD_FREE_LARGE(reqmsg, reqmsg_size);
548 * If current context of \a req is dead somehow, e.g. we just switched flavor
549 * thus marked original contexts dead, we'll find a new context for it. if
550 * no switch is needed, \a req will end up with the same context.
552 * \note a request must have a context, to keep other parts of code happy.
553 * In any case of failure during the switching, we must restore the old one.
555 int sptlrpc_req_replace_dead_ctx(struct ptlrpc_request *req)
557 struct ptlrpc_cli_ctx *oldctx = req->rq_cli_ctx;
558 struct ptlrpc_cli_ctx *newctx;
564 sptlrpc_cli_ctx_get(oldctx);
565 sptlrpc_req_put_ctx(req, 0);
567 rc = sptlrpc_req_get_ctx(req);
569 LASSERT(!req->rq_cli_ctx);
571 /* restore old ctx */
572 req->rq_cli_ctx = oldctx;
576 newctx = req->rq_cli_ctx;
579 if (unlikely(newctx == oldctx &&
580 test_bit(PTLRPC_CTX_DEAD_BIT, &oldctx->cc_flags))) {
582 * still get the old dead ctx, usually means system too busy
585 "ctx (%p, fl %lx) doesn't switch, relax a little bit\n",
586 newctx, newctx->cc_flags);
588 set_current_state(TASK_INTERRUPTIBLE);
589 schedule_timeout(msecs_to_jiffies(MSEC_PER_SEC));
590 } else if (unlikely(test_bit(PTLRPC_CTX_UPTODATE_BIT, &newctx->cc_flags)
593 * new ctx not up to date yet
596 "ctx (%p, fl %lx) doesn't switch, not up to date yet\n",
597 newctx, newctx->cc_flags);
600 * it's possible newctx == oldctx if we're switching
601 * subflavor with the same sec.
603 rc = sptlrpc_req_ctx_switch(req, oldctx, newctx);
605 /* restore old ctx */
606 sptlrpc_req_put_ctx(req, 0);
607 req->rq_cli_ctx = oldctx;
611 LASSERT(req->rq_cli_ctx == newctx);
614 sptlrpc_cli_ctx_put(oldctx, 1);
617 EXPORT_SYMBOL(sptlrpc_req_replace_dead_ctx);
620 int ctx_check_refresh(struct ptlrpc_cli_ctx *ctx)
622 if (cli_ctx_is_refreshed(ctx))
628 int ctx_refresh_timeout(void *data)
630 struct ptlrpc_request *req = data;
633 /* conn_cnt is needed in expire_one_request */
634 lustre_msg_set_conn_cnt(req->rq_reqmsg, req->rq_import->imp_conn_cnt);
636 rc = ptlrpc_expire_one_request(req, 1);
637 /* if we started recovery, we should mark this ctx dead; otherwise
638 * in case of lgssd died nobody would retire this ctx, following
639 * connecting will still find the same ctx thus cause deadlock.
640 * there's an assumption that expire time of the request should be
641 * later than the context refresh expire time.
644 req->rq_cli_ctx->cc_ops->die(req->rq_cli_ctx, 0);
649 void ctx_refresh_interrupt(void *data)
651 struct ptlrpc_request *req = data;
653 spin_lock(&req->rq_lock);
655 spin_unlock(&req->rq_lock);
659 void req_off_ctx_list(struct ptlrpc_request *req, struct ptlrpc_cli_ctx *ctx)
661 spin_lock(&ctx->cc_lock);
662 if (!list_empty(&req->rq_ctx_chain))
663 list_del_init(&req->rq_ctx_chain);
664 spin_unlock(&ctx->cc_lock);
668 * To refresh the context of \req, if it's not up-to-date.
671 * - = 0: wait until success or fatal error occur
672 * - > 0: timeout value (in seconds)
674 * The status of the context could be subject to be changed by other threads
675 * at any time. We allow this race, but once we return with 0, the caller will
676 * suppose it's uptodated and keep using it until the owning rpc is done.
678 * \retval 0 only if the context is uptodated.
679 * \retval -ev error number.
681 int sptlrpc_req_refresh_ctx(struct ptlrpc_request *req, long timeout)
683 struct ptlrpc_cli_ctx *ctx = req->rq_cli_ctx;
684 struct ptlrpc_sec *sec;
685 struct l_wait_info lwi;
691 if (req->rq_ctx_init || req->rq_ctx_fini)
695 * during the process a request's context might change type even
696 * (e.g. from gss ctx to null ctx), so each loop we need to re-check
700 rc = import_sec_validate_get(req->rq_import, &sec);
704 if (sec->ps_flvr.sf_rpc != req->rq_flvr.sf_rpc) {
705 CDEBUG(D_SEC, "req %p: flavor has changed %x -> %x\n",
706 req, req->rq_flvr.sf_rpc, sec->ps_flvr.sf_rpc);
707 req_off_ctx_list(req, ctx);
708 sptlrpc_req_replace_dead_ctx(req);
709 ctx = req->rq_cli_ctx;
711 sptlrpc_sec_put(sec);
713 if (cli_ctx_is_eternal(ctx))
716 if (unlikely(test_bit(PTLRPC_CTX_NEW_BIT, &ctx->cc_flags))) {
717 if (ctx->cc_ops->refresh)
718 ctx->cc_ops->refresh(ctx);
720 LASSERT(test_bit(PTLRPC_CTX_NEW_BIT, &ctx->cc_flags) == 0);
722 LASSERT(ctx->cc_ops->validate);
723 if (ctx->cc_ops->validate(ctx) == 0) {
724 req_off_ctx_list(req, ctx);
728 if (unlikely(test_bit(PTLRPC_CTX_ERROR_BIT, &ctx->cc_flags))) {
729 spin_lock(&req->rq_lock);
731 spin_unlock(&req->rq_lock);
732 req_off_ctx_list(req, ctx);
737 * There's a subtle issue for resending RPCs, suppose following
739 * 1. the request was sent to server.
740 * 2. recovery was kicked start, after finished the request was
742 * 3. resend the request.
743 * 4. old reply from server received, we accept and verify the reply.
744 * this has to be success, otherwise the error will be aware
746 * 5. new reply from server received, dropped by LNet.
748 * Note the xid of old & new request is the same. We can't simply
749 * change xid for the resent request because the server replies on
750 * it for reply reconstruction.
752 * Commonly the original context should be uptodate because we
753 * have an expiry nice time; server will keep its context because
754 * we at least hold a ref of old context which prevent context
755 * from destroying RPC being sent. So server still can accept the
756 * request and finish the RPC. But if that's not the case:
757 * 1. If server side context has been trimmed, a NO_CONTEXT will
758 * be returned, gss_cli_ctx_verify/unseal will switch to new
760 * 2. Current context never be refreshed, then we are fine: we
761 * never really send request with old context before.
763 if (test_bit(PTLRPC_CTX_UPTODATE_BIT, &ctx->cc_flags) &&
764 unlikely(req->rq_reqmsg) &&
765 lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT) {
766 req_off_ctx_list(req, ctx);
770 if (unlikely(test_bit(PTLRPC_CTX_DEAD_BIT, &ctx->cc_flags))) {
771 req_off_ctx_list(req, ctx);
773 * don't switch ctx if import was deactivated
775 if (req->rq_import->imp_deactive) {
776 spin_lock(&req->rq_lock);
778 spin_unlock(&req->rq_lock);
782 rc = sptlrpc_req_replace_dead_ctx(req);
784 LASSERT(ctx == req->rq_cli_ctx);
785 CERROR("req %p: failed to replace dead ctx %p: %d\n",
787 spin_lock(&req->rq_lock);
789 spin_unlock(&req->rq_lock);
793 ctx = req->rq_cli_ctx;
798 * Now we're sure this context is during upcall, add myself into
801 spin_lock(&ctx->cc_lock);
802 if (list_empty(&req->rq_ctx_chain))
803 list_add(&req->rq_ctx_chain, &ctx->cc_req_list);
804 spin_unlock(&ctx->cc_lock);
807 RETURN(-EWOULDBLOCK);
809 /* Clear any flags that may be present from previous sends */
810 LASSERT(req->rq_receiving_reply == 0);
811 spin_lock(&req->rq_lock);
813 req->rq_timedout = 0;
816 spin_unlock(&req->rq_lock);
818 lwi = LWI_TIMEOUT_INTR(msecs_to_jiffies(timeout * MSEC_PER_SEC),
820 ctx_refresh_interrupt, req);
821 rc = l_wait_event(req->rq_reply_waitq, ctx_check_refresh(ctx), &lwi);
824 * following cases could lead us here:
825 * - successfully refreshed;
827 * - timedout, and we don't want recover from the failure;
828 * - timedout, and waked up upon recovery finished;
829 * - someone else mark this ctx dead by force;
830 * - someone invalidate the req and call ptlrpc_client_wake_req(),
831 * e.g. ptlrpc_abort_inflight();
833 if (!cli_ctx_is_refreshed(ctx)) {
834 /* timed out or interruptted */
835 req_off_ctx_list(req, ctx);
844 /* Bring ptlrpc_sec context up-to-date */
845 int sptlrpc_export_update_ctx(struct obd_export *exp)
847 struct obd_import *imp = exp ? exp->exp_imp_reverse : NULL;
848 struct ptlrpc_sec *sec = NULL;
849 struct ptlrpc_cli_ctx *ctx = NULL;
853 sec = sptlrpc_import_sec_ref(imp);
855 ctx = get_my_ctx(sec);
856 sptlrpc_sec_put(sec);
860 if (ctx->cc_ops->refresh)
861 rc = ctx->cc_ops->refresh(ctx);
862 sptlrpc_cli_ctx_put(ctx, 1);
868 * Initialize flavor settings for \a req, according to \a opcode.
870 * \note this could be called in two situations:
871 * - new request from ptlrpc_pre_req(), with proper @opcode
872 * - old request which changed ctx in the middle, with @opcode == 0
874 void sptlrpc_req_set_flavor(struct ptlrpc_request *req, int opcode)
876 struct ptlrpc_sec *sec;
878 LASSERT(req->rq_import);
879 LASSERT(req->rq_cli_ctx);
880 LASSERT(req->rq_cli_ctx->cc_sec);
881 LASSERT(req->rq_bulk_read == 0 || req->rq_bulk_write == 0);
883 /* special security flags according to opcode */
887 case MGS_CONFIG_READ:
889 req->rq_bulk_read = 1;
893 req->rq_bulk_write = 1;
896 req->rq_ctx_init = 1;
899 req->rq_ctx_fini = 1;
902 /* init/fini rpc won't be resend, so can't be here */
903 LASSERT(req->rq_ctx_init == 0);
904 LASSERT(req->rq_ctx_fini == 0);
906 /* cleanup flags, which should be recalculated */
907 req->rq_pack_udesc = 0;
908 req->rq_pack_bulk = 0;
912 sec = req->rq_cli_ctx->cc_sec;
914 spin_lock(&sec->ps_lock);
915 req->rq_flvr = sec->ps_flvr;
916 spin_unlock(&sec->ps_lock);
918 /* force SVC_NULL for context initiation rpc, SVC_INTG for context
920 if (unlikely(req->rq_ctx_init))
921 flvr_set_svc(&req->rq_flvr.sf_rpc, SPTLRPC_SVC_NULL);
922 else if (unlikely(req->rq_ctx_fini))
923 flvr_set_svc(&req->rq_flvr.sf_rpc, SPTLRPC_SVC_INTG);
925 /* user descriptor flag, null security can't do it anyway */
926 if ((sec->ps_flvr.sf_flags & PTLRPC_SEC_FL_UDESC) &&
927 (req->rq_flvr.sf_rpc != SPTLRPC_FLVR_NULL))
928 req->rq_pack_udesc = 1;
930 /* bulk security flag */
931 if ((req->rq_bulk_read || req->rq_bulk_write) &&
932 sptlrpc_flavor_has_bulk(&req->rq_flvr))
933 req->rq_pack_bulk = 1;
936 void sptlrpc_request_out_callback(struct ptlrpc_request *req)
938 if (SPTLRPC_FLVR_SVC(req->rq_flvr.sf_rpc) != SPTLRPC_SVC_PRIV)
941 LASSERT(req->rq_clrbuf);
942 if (req->rq_pool || !req->rq_reqbuf)
945 OBD_FREE(req->rq_reqbuf, req->rq_reqbuf_len);
946 req->rq_reqbuf = NULL;
947 req->rq_reqbuf_len = 0;
951 * Given an import \a imp, check whether current user has a valid context
952 * or not. We may create a new context and try to refresh it, and try
953 * repeatedly try in case of non-fatal errors. Return 0 means success.
955 int sptlrpc_import_check_ctx(struct obd_import *imp)
957 struct ptlrpc_sec *sec;
958 struct ptlrpc_cli_ctx *ctx;
959 struct ptlrpc_request *req = NULL;
965 sec = sptlrpc_import_sec_ref(imp);
966 ctx = get_my_ctx(sec);
967 sptlrpc_sec_put(sec);
972 if (cli_ctx_is_eternal(ctx) ||
973 ctx->cc_ops->validate(ctx) == 0) {
974 sptlrpc_cli_ctx_put(ctx, 1);
978 if (cli_ctx_is_error(ctx)) {
979 sptlrpc_cli_ctx_put(ctx, 1);
983 req = ptlrpc_request_cache_alloc(GFP_NOFS);
987 ptlrpc_cli_req_init(req);
988 atomic_set(&req->rq_refcount, 10000);
990 req->rq_import = imp;
991 req->rq_flvr = sec->ps_flvr;
992 req->rq_cli_ctx = ctx;
994 rc = sptlrpc_req_refresh_ctx(req, 0);
995 LASSERT(list_empty(&req->rq_ctx_chain));
996 sptlrpc_cli_ctx_put(req->rq_cli_ctx, 1);
997 ptlrpc_request_cache_free(req);
1003 * Used by ptlrpc client, to perform the pre-defined security transformation
1004 * upon the request message of \a req. After this function called,
1005 * req->rq_reqmsg is still accessible as clear text.
1007 int sptlrpc_cli_wrap_request(struct ptlrpc_request *req)
1009 struct ptlrpc_cli_ctx *ctx = req->rq_cli_ctx;
1014 LASSERT(ctx->cc_sec);
1015 LASSERT(req->rq_reqbuf || req->rq_clrbuf);
1017 /* we wrap bulk request here because now we can be sure
1018 * the context is uptodate.
1021 rc = sptlrpc_cli_wrap_bulk(req, req->rq_bulk);
1026 switch (SPTLRPC_FLVR_SVC(req->rq_flvr.sf_rpc)) {
1027 case SPTLRPC_SVC_NULL:
1028 case SPTLRPC_SVC_AUTH:
1029 case SPTLRPC_SVC_INTG:
1030 LASSERT(ctx->cc_ops->sign);
1031 rc = ctx->cc_ops->sign(ctx, req);
1033 case SPTLRPC_SVC_PRIV:
1034 LASSERT(ctx->cc_ops->seal);
1035 rc = ctx->cc_ops->seal(ctx, req);
1042 LASSERT(req->rq_reqdata_len);
1043 LASSERT(req->rq_reqdata_len % 8 == 0);
1044 LASSERT(req->rq_reqdata_len <= req->rq_reqbuf_len);
1050 static int do_cli_unwrap_reply(struct ptlrpc_request *req)
1052 struct ptlrpc_cli_ctx *ctx = req->rq_cli_ctx;
1057 LASSERT(ctx->cc_sec);
1058 LASSERT(req->rq_repbuf);
1059 LASSERT(req->rq_repdata);
1060 LASSERT(req->rq_repmsg == NULL);
1062 req->rq_rep_swab_mask = 0;
1064 rc = __lustre_unpack_msg(req->rq_repdata, req->rq_repdata_len);
1067 lustre_set_rep_swabbed(req, MSG_PTLRPC_HEADER_OFF);
1071 CERROR("failed unpack reply: x%llu\n", req->rq_xid);
1075 if (req->rq_repdata_len < sizeof(struct lustre_msg)) {
1076 CERROR("replied data length %d too small\n",
1077 req->rq_repdata_len);
1081 if (SPTLRPC_FLVR_POLICY(req->rq_repdata->lm_secflvr) !=
1082 SPTLRPC_FLVR_POLICY(req->rq_flvr.sf_rpc)) {
1083 CERROR("reply policy %u doesn't match request policy %u\n",
1084 SPTLRPC_FLVR_POLICY(req->rq_repdata->lm_secflvr),
1085 SPTLRPC_FLVR_POLICY(req->rq_flvr.sf_rpc));
1089 switch (SPTLRPC_FLVR_SVC(req->rq_flvr.sf_rpc)) {
1090 case SPTLRPC_SVC_NULL:
1091 case SPTLRPC_SVC_AUTH:
1092 case SPTLRPC_SVC_INTG:
1093 LASSERT(ctx->cc_ops->verify);
1094 rc = ctx->cc_ops->verify(ctx, req);
1096 case SPTLRPC_SVC_PRIV:
1097 LASSERT(ctx->cc_ops->unseal);
1098 rc = ctx->cc_ops->unseal(ctx, req);
1103 LASSERT(rc || req->rq_repmsg || req->rq_resend);
1105 if (SPTLRPC_FLVR_POLICY(req->rq_flvr.sf_rpc) != SPTLRPC_POLICY_NULL &&
1107 req->rq_rep_swab_mask = 0;
1112 * Used by ptlrpc client, to perform security transformation upon the reply
1113 * message of \a req. After return successfully, req->rq_repmsg points to
1114 * the reply message in clear text.
1116 * \pre the reply buffer should have been un-posted from LNet, so nothing is
1119 int sptlrpc_cli_unwrap_reply(struct ptlrpc_request *req)
1121 LASSERT(req->rq_repbuf);
1122 LASSERT(req->rq_repdata == NULL);
1123 LASSERT(req->rq_repmsg == NULL);
1124 LASSERT(req->rq_reply_off + req->rq_nob_received <= req->rq_repbuf_len);
1126 if (req->rq_reply_off == 0 &&
1127 (lustre_msghdr_get_flags(req->rq_reqmsg) & MSGHDR_AT_SUPPORT)) {
1128 CERROR("real reply with offset 0\n");
1132 if (req->rq_reply_off % 8 != 0) {
1133 CERROR("reply at odd offset %u\n", req->rq_reply_off);
1137 req->rq_repdata = (struct lustre_msg *)
1138 (req->rq_repbuf + req->rq_reply_off);
1139 req->rq_repdata_len = req->rq_nob_received;
1141 return do_cli_unwrap_reply(req);
1145 * Used by ptlrpc client, to perform security transformation upon the early
1146 * reply message of \a req. We expect the rq_reply_off is 0, and
1147 * rq_nob_received is the early reply size.
1149 * Because the receive buffer might be still posted, the reply data might be
1150 * changed at any time, no matter we're holding rq_lock or not. For this reason
1151 * we allocate a separate ptlrpc_request and reply buffer for early reply
1154 * \retval 0 success, \a req_ret is filled with a duplicated ptlrpc_request.
1155 * Later the caller must call sptlrpc_cli_finish_early_reply() on the returned
1156 * \a *req_ret to release it.
1157 * \retval -ev error number, and \a req_ret will not be set.
1159 int sptlrpc_cli_unwrap_early_reply(struct ptlrpc_request *req,
1160 struct ptlrpc_request **req_ret)
1162 struct ptlrpc_request *early_req;
1164 int early_bufsz, early_size;
1168 early_req = ptlrpc_request_cache_alloc(GFP_NOFS);
1169 if (early_req == NULL)
1172 ptlrpc_cli_req_init(early_req);
1174 early_size = req->rq_nob_received;
1175 early_bufsz = size_roundup_power2(early_size);
1176 OBD_ALLOC_LARGE(early_buf, early_bufsz);
1177 if (early_buf == NULL)
1178 GOTO(err_req, rc = -ENOMEM);
1180 /* sanity checkings and copy data out, do it inside spinlock */
1181 spin_lock(&req->rq_lock);
1183 if (req->rq_replied) {
1184 spin_unlock(&req->rq_lock);
1185 GOTO(err_buf, rc = -EALREADY);
1188 LASSERT(req->rq_repbuf);
1189 LASSERT(req->rq_repdata == NULL);
1190 LASSERT(req->rq_repmsg == NULL);
1192 if (req->rq_reply_off != 0) {
1193 CERROR("early reply with offset %u\n", req->rq_reply_off);
1194 spin_unlock(&req->rq_lock);
1195 GOTO(err_buf, rc = -EPROTO);
1198 if (req->rq_nob_received != early_size) {
1199 /* even another early arrived the size should be the same */
1200 CERROR("data size has changed from %u to %u\n",
1201 early_size, req->rq_nob_received);
1202 spin_unlock(&req->rq_lock);
1203 GOTO(err_buf, rc = -EINVAL);
1206 if (req->rq_nob_received < sizeof(struct lustre_msg)) {
1207 CERROR("early reply length %d too small\n",
1208 req->rq_nob_received);
1209 spin_unlock(&req->rq_lock);
1210 GOTO(err_buf, rc = -EALREADY);
1213 memcpy(early_buf, req->rq_repbuf, early_size);
1214 spin_unlock(&req->rq_lock);
1216 early_req->rq_cli_ctx = sptlrpc_cli_ctx_get(req->rq_cli_ctx);
1217 early_req->rq_flvr = req->rq_flvr;
1218 early_req->rq_repbuf = early_buf;
1219 early_req->rq_repbuf_len = early_bufsz;
1220 early_req->rq_repdata = (struct lustre_msg *) early_buf;
1221 early_req->rq_repdata_len = early_size;
1222 early_req->rq_early = 1;
1223 early_req->rq_reqmsg = req->rq_reqmsg;
1225 rc = do_cli_unwrap_reply(early_req);
1227 DEBUG_REQ(D_ADAPTTO, early_req,
1228 "error %d unwrap early reply", rc);
1232 LASSERT(early_req->rq_repmsg);
1233 *req_ret = early_req;
1237 sptlrpc_cli_ctx_put(early_req->rq_cli_ctx, 1);
1239 OBD_FREE_LARGE(early_buf, early_bufsz);
1241 ptlrpc_request_cache_free(early_req);
1246 * Used by ptlrpc client, to release a processed early reply \a early_req.
1248 * \pre \a early_req was obtained from calling sptlrpc_cli_unwrap_early_reply().
1250 void sptlrpc_cli_finish_early_reply(struct ptlrpc_request *early_req)
1252 LASSERT(early_req->rq_repbuf);
1253 LASSERT(early_req->rq_repdata);
1254 LASSERT(early_req->rq_repmsg);
1256 sptlrpc_cli_ctx_put(early_req->rq_cli_ctx, 1);
1257 OBD_FREE_LARGE(early_req->rq_repbuf, early_req->rq_repbuf_len);
1258 ptlrpc_request_cache_free(early_req);
1261 /**************************************************
1263 **************************************************/
1266 * "fixed" sec (e.g. null) use sec_id < 0
1268 static atomic_t sptlrpc_sec_id = ATOMIC_INIT(1);
1270 int sptlrpc_get_next_secid(void)
1272 return atomic_inc_return(&sptlrpc_sec_id);
1274 EXPORT_SYMBOL(sptlrpc_get_next_secid);
1276 /**************************************************
1277 * client side high-level security APIs *
1278 **************************************************/
1280 static int sec_cop_flush_ctx_cache(struct ptlrpc_sec *sec, uid_t uid,
1281 int grace, int force)
1283 struct ptlrpc_sec_policy *policy = sec->ps_policy;
1285 LASSERT(policy->sp_cops);
1286 LASSERT(policy->sp_cops->flush_ctx_cache);
1288 return policy->sp_cops->flush_ctx_cache(sec, uid, grace, force);
1291 static void sec_cop_destroy_sec(struct ptlrpc_sec *sec)
1293 struct ptlrpc_sec_policy *policy = sec->ps_policy;
1295 LASSERT_ATOMIC_ZERO(&sec->ps_refcount);
1296 LASSERT_ATOMIC_ZERO(&sec->ps_nctx);
1297 LASSERT(policy->sp_cops->destroy_sec);
1299 CDEBUG(D_SEC, "%s@%p: being destroied\n", sec->ps_policy->sp_name, sec);
1301 policy->sp_cops->destroy_sec(sec);
1302 sptlrpc_policy_put(policy);
1305 void sptlrpc_sec_destroy(struct ptlrpc_sec *sec)
1307 sec_cop_destroy_sec(sec);
1309 EXPORT_SYMBOL(sptlrpc_sec_destroy);
1311 static void sptlrpc_sec_kill(struct ptlrpc_sec *sec)
1313 LASSERT_ATOMIC_POS(&sec->ps_refcount);
1315 if (sec->ps_policy->sp_cops->kill_sec) {
1316 sec->ps_policy->sp_cops->kill_sec(sec);
1318 sec_cop_flush_ctx_cache(sec, -1, 1, 1);
1322 struct ptlrpc_sec *sptlrpc_sec_get(struct ptlrpc_sec *sec)
1325 atomic_inc(&sec->ps_refcount);
1329 EXPORT_SYMBOL(sptlrpc_sec_get);
1331 void sptlrpc_sec_put(struct ptlrpc_sec *sec)
1334 LASSERT_ATOMIC_POS(&sec->ps_refcount);
1336 if (atomic_dec_and_test(&sec->ps_refcount)) {
1337 sptlrpc_gc_del_sec(sec);
1338 sec_cop_destroy_sec(sec);
1342 EXPORT_SYMBOL(sptlrpc_sec_put);
1345 * policy module is responsible for taking refrence of import
1348 struct ptlrpc_sec * sptlrpc_sec_create(struct obd_import *imp,
1349 struct ptlrpc_svc_ctx *svc_ctx,
1350 struct sptlrpc_flavor *sf,
1351 enum lustre_sec_part sp)
1353 struct ptlrpc_sec_policy *policy;
1354 struct ptlrpc_sec *sec;
1359 LASSERT(imp->imp_dlm_fake == 1);
1361 CDEBUG(D_SEC, "%s %s: reverse sec using flavor %s\n",
1362 imp->imp_obd->obd_type->typ_name,
1363 imp->imp_obd->obd_name,
1364 sptlrpc_flavor2name(sf, str, sizeof(str)));
1366 policy = sptlrpc_policy_get(svc_ctx->sc_policy);
1367 sf->sf_flags |= PTLRPC_SEC_FL_REVERSE | PTLRPC_SEC_FL_ROOTONLY;
1369 LASSERT(imp->imp_dlm_fake == 0);
1371 CDEBUG(D_SEC, "%s %s: select security flavor %s\n",
1372 imp->imp_obd->obd_type->typ_name,
1373 imp->imp_obd->obd_name,
1374 sptlrpc_flavor2name(sf, str, sizeof(str)));
1376 policy = sptlrpc_wireflavor2policy(sf->sf_rpc);
1378 CERROR("invalid flavor 0x%x\n", sf->sf_rpc);
1383 sec = policy->sp_cops->create_sec(imp, svc_ctx, sf);
1385 atomic_inc(&sec->ps_refcount);
1389 if (sec->ps_gc_interval && policy->sp_cops->gc_ctx)
1390 sptlrpc_gc_add_sec(sec);
1392 sptlrpc_policy_put(policy);
1398 struct ptlrpc_sec *sptlrpc_import_sec_ref(struct obd_import *imp)
1400 struct ptlrpc_sec *sec;
1402 spin_lock(&imp->imp_lock);
1403 sec = sptlrpc_sec_get(imp->imp_sec);
1404 spin_unlock(&imp->imp_lock);
1408 EXPORT_SYMBOL(sptlrpc_import_sec_ref);
1410 static void sptlrpc_import_sec_install(struct obd_import *imp,
1411 struct ptlrpc_sec *sec)
1413 struct ptlrpc_sec *old_sec;
1415 LASSERT_ATOMIC_POS(&sec->ps_refcount);
1417 spin_lock(&imp->imp_lock);
1418 old_sec = imp->imp_sec;
1420 spin_unlock(&imp->imp_lock);
1423 sptlrpc_sec_kill(old_sec);
1425 /* balance the ref taken by this import */
1426 sptlrpc_sec_put(old_sec);
1431 int flavor_equal(struct sptlrpc_flavor *sf1, struct sptlrpc_flavor *sf2)
1433 return (memcmp(sf1, sf2, sizeof(*sf1)) == 0);
1437 void flavor_copy(struct sptlrpc_flavor *dst, struct sptlrpc_flavor *src)
1443 * To get an appropriate ptlrpc_sec for the \a imp, according to the current
1444 * configuration. Upon called, imp->imp_sec may or may not be NULL.
1446 * - regular import: \a svc_ctx should be NULL and \a flvr is ignored;
1447 * - reverse import: \a svc_ctx and \a flvr are obtained from incoming request.
1449 int sptlrpc_import_sec_adapt(struct obd_import *imp,
1450 struct ptlrpc_svc_ctx *svc_ctx,
1451 struct sptlrpc_flavor *flvr)
1453 struct ptlrpc_connection *conn;
1454 struct sptlrpc_flavor sf;
1455 struct ptlrpc_sec *sec, *newsec;
1456 enum lustre_sec_part sp;
1466 conn = imp->imp_connection;
1468 if (svc_ctx == NULL) {
1469 struct client_obd *cliobd = &imp->imp_obd->u.cli;
1471 * normal import, determine flavor from rule set, except
1472 * for mgc the flavor is predetermined.
1474 if (cliobd->cl_sp_me == LUSTRE_SP_MGC)
1475 sf = cliobd->cl_flvr_mgc;
1477 sptlrpc_conf_choose_flavor(cliobd->cl_sp_me,
1479 &cliobd->cl_target_uuid,
1482 sp = imp->imp_obd->u.cli.cl_sp_me;
1484 /* reverse import, determine flavor from incoming reqeust */
1487 if (sf.sf_rpc != SPTLRPC_FLVR_NULL)
1488 sf.sf_flags = PTLRPC_SEC_FL_REVERSE |
1489 PTLRPC_SEC_FL_ROOTONLY;
1491 sp = sptlrpc_target_sec_part(imp->imp_obd);
1494 sec = sptlrpc_import_sec_ref(imp);
1498 if (flavor_equal(&sf, &sec->ps_flvr))
1501 CDEBUG(D_SEC, "import %s->%s: changing flavor %s -> %s\n",
1502 imp->imp_obd->obd_name,
1503 obd_uuid2str(&conn->c_remote_uuid),
1504 sptlrpc_flavor2name(&sec->ps_flvr, str, sizeof(str)),
1505 sptlrpc_flavor2name(&sf, str2, sizeof(str2)));
1506 } else if (SPTLRPC_FLVR_BASE(sf.sf_rpc) !=
1507 SPTLRPC_FLVR_BASE(SPTLRPC_FLVR_NULL)) {
1508 CDEBUG(D_SEC, "import %s->%s netid %x: select flavor %s\n",
1509 imp->imp_obd->obd_name,
1510 obd_uuid2str(&conn->c_remote_uuid),
1511 LNET_NIDNET(conn->c_self),
1512 sptlrpc_flavor2name(&sf, str, sizeof(str)));
1515 mutex_lock(&imp->imp_sec_mutex);
1517 newsec = sptlrpc_sec_create(imp, svc_ctx, &sf, sp);
1519 sptlrpc_import_sec_install(imp, newsec);
1521 CERROR("import %s->%s: failed to create new sec\n",
1522 imp->imp_obd->obd_name,
1523 obd_uuid2str(&conn->c_remote_uuid));
1527 mutex_unlock(&imp->imp_sec_mutex);
1529 sptlrpc_sec_put(sec);
1533 void sptlrpc_import_sec_put(struct obd_import *imp)
1536 sptlrpc_sec_kill(imp->imp_sec);
1538 sptlrpc_sec_put(imp->imp_sec);
1539 imp->imp_sec = NULL;
1543 static void import_flush_ctx_common(struct obd_import *imp,
1544 uid_t uid, int grace, int force)
1546 struct ptlrpc_sec *sec;
1551 sec = sptlrpc_import_sec_ref(imp);
1555 sec_cop_flush_ctx_cache(sec, uid, grace, force);
1556 sptlrpc_sec_put(sec);
1559 void sptlrpc_import_flush_root_ctx(struct obd_import *imp)
1561 /* it's important to use grace mode, see explain in
1562 * sptlrpc_req_refresh_ctx() */
1563 import_flush_ctx_common(imp, 0, 1, 1);
1566 void sptlrpc_import_flush_my_ctx(struct obd_import *imp)
1568 import_flush_ctx_common(imp, from_kuid(&init_user_ns, current_uid()),
1571 EXPORT_SYMBOL(sptlrpc_import_flush_my_ctx);
1573 void sptlrpc_import_flush_all_ctx(struct obd_import *imp)
1575 import_flush_ctx_common(imp, -1, 1, 1);
1577 EXPORT_SYMBOL(sptlrpc_import_flush_all_ctx);
1580 * Used by ptlrpc client to allocate request buffer of \a req. Upon return
1581 * successfully, req->rq_reqmsg points to a buffer with size \a msgsize.
1583 int sptlrpc_cli_alloc_reqbuf(struct ptlrpc_request *req, int msgsize)
1585 struct ptlrpc_cli_ctx *ctx = req->rq_cli_ctx;
1586 struct ptlrpc_sec_policy *policy;
1590 LASSERT(ctx->cc_sec);
1591 LASSERT(ctx->cc_sec->ps_policy);
1592 LASSERT(req->rq_reqmsg == NULL);
1593 LASSERT_ATOMIC_POS(&ctx->cc_refcount);
1595 policy = ctx->cc_sec->ps_policy;
1596 rc = policy->sp_cops->alloc_reqbuf(ctx->cc_sec, req, msgsize);
1598 LASSERT(req->rq_reqmsg);
1599 LASSERT(req->rq_reqbuf || req->rq_clrbuf);
1601 /* zeroing preallocated buffer */
1603 memset(req->rq_reqmsg, 0, msgsize);
1610 * Used by ptlrpc client to free request buffer of \a req. After this
1611 * req->rq_reqmsg is set to NULL and should not be accessed anymore.
1613 void sptlrpc_cli_free_reqbuf(struct ptlrpc_request *req)
1615 struct ptlrpc_cli_ctx *ctx = req->rq_cli_ctx;
1616 struct ptlrpc_sec_policy *policy;
1619 LASSERT(ctx->cc_sec);
1620 LASSERT(ctx->cc_sec->ps_policy);
1621 LASSERT_ATOMIC_POS(&ctx->cc_refcount);
1623 if (req->rq_reqbuf == NULL && req->rq_clrbuf == NULL)
1626 policy = ctx->cc_sec->ps_policy;
1627 policy->sp_cops->free_reqbuf(ctx->cc_sec, req);
1628 req->rq_reqmsg = NULL;
1632 * NOTE caller must guarantee the buffer size is enough for the enlargement
1634 void _sptlrpc_enlarge_msg_inplace(struct lustre_msg *msg,
1635 int segment, int newsize)
1638 int oldsize, oldmsg_size, movesize;
1640 LASSERT(segment < msg->lm_bufcount);
1641 LASSERT(msg->lm_buflens[segment] <= newsize);
1643 if (msg->lm_buflens[segment] == newsize)
1646 /* nothing to do if we are enlarging the last segment */
1647 if (segment == msg->lm_bufcount - 1) {
1648 msg->lm_buflens[segment] = newsize;
1652 oldsize = msg->lm_buflens[segment];
1654 src = lustre_msg_buf(msg, segment + 1, 0);
1655 msg->lm_buflens[segment] = newsize;
1656 dst = lustre_msg_buf(msg, segment + 1, 0);
1657 msg->lm_buflens[segment] = oldsize;
1659 /* move from segment + 1 to end segment */
1660 LASSERT(msg->lm_magic == LUSTRE_MSG_MAGIC_V2);
1661 oldmsg_size = lustre_msg_size_v2(msg->lm_bufcount, msg->lm_buflens);
1662 movesize = oldmsg_size - ((unsigned long) src - (unsigned long) msg);
1663 LASSERT(movesize >= 0);
1666 memmove(dst, src, movesize);
1668 /* note we don't clear the ares where old data live, not secret */
1670 /* finally set new segment size */
1671 msg->lm_buflens[segment] = newsize;
1673 EXPORT_SYMBOL(_sptlrpc_enlarge_msg_inplace);
1676 * Used by ptlrpc client to enlarge the \a segment of request message pointed
1677 * by req->rq_reqmsg to size \a newsize, all previously filled-in data will be
1678 * preserved after the enlargement. this must be called after original request
1679 * buffer being allocated.
1681 * \note after this be called, rq_reqmsg and rq_reqlen might have been changed,
1682 * so caller should refresh its local pointers if needed.
1684 int sptlrpc_cli_enlarge_reqbuf(struct ptlrpc_request *req,
1685 const struct req_msg_field *field,
1688 struct req_capsule *pill = &req->rq_pill;
1689 struct ptlrpc_cli_ctx *ctx = req->rq_cli_ctx;
1690 struct ptlrpc_sec_cops *cops;
1691 struct lustre_msg *msg = req->rq_reqmsg;
1692 int segment = __req_capsule_offset(pill, field, RCL_CLIENT);
1696 LASSERT(msg->lm_bufcount > segment);
1697 LASSERT(msg->lm_buflens[segment] <= newsize);
1699 if (msg->lm_buflens[segment] == newsize)
1702 cops = ctx->cc_sec->ps_policy->sp_cops;
1703 LASSERT(cops->enlarge_reqbuf);
1704 return cops->enlarge_reqbuf(ctx->cc_sec, req, segment, newsize);
1706 EXPORT_SYMBOL(sptlrpc_cli_enlarge_reqbuf);
1709 * Used by ptlrpc client to allocate reply buffer of \a req.
1711 * \note After this, req->rq_repmsg is still not accessible.
1713 int sptlrpc_cli_alloc_repbuf(struct ptlrpc_request *req, int msgsize)
1715 struct ptlrpc_cli_ctx *ctx = req->rq_cli_ctx;
1716 struct ptlrpc_sec_policy *policy;
1720 LASSERT(ctx->cc_sec);
1721 LASSERT(ctx->cc_sec->ps_policy);
1726 policy = ctx->cc_sec->ps_policy;
1727 RETURN(policy->sp_cops->alloc_repbuf(ctx->cc_sec, req, msgsize));
1731 * Used by ptlrpc client to free reply buffer of \a req. After this
1732 * req->rq_repmsg is set to NULL and should not be accessed anymore.
1734 void sptlrpc_cli_free_repbuf(struct ptlrpc_request *req)
1736 struct ptlrpc_cli_ctx *ctx = req->rq_cli_ctx;
1737 struct ptlrpc_sec_policy *policy;
1741 LASSERT(ctx->cc_sec);
1742 LASSERT(ctx->cc_sec->ps_policy);
1743 LASSERT_ATOMIC_POS(&ctx->cc_refcount);
1745 if (req->rq_repbuf == NULL)
1747 LASSERT(req->rq_repbuf_len);
1749 policy = ctx->cc_sec->ps_policy;
1750 policy->sp_cops->free_repbuf(ctx->cc_sec, req);
1751 req->rq_repmsg = NULL;
1754 EXPORT_SYMBOL(sptlrpc_cli_free_repbuf);
1756 int sptlrpc_cli_install_rvs_ctx(struct obd_import *imp,
1757 struct ptlrpc_cli_ctx *ctx)
1759 struct ptlrpc_sec_policy *policy = ctx->cc_sec->ps_policy;
1761 if (!policy->sp_cops->install_rctx)
1763 return policy->sp_cops->install_rctx(imp, ctx->cc_sec, ctx);
1766 int sptlrpc_svc_install_rvs_ctx(struct obd_import *imp,
1767 struct ptlrpc_svc_ctx *ctx)
1769 struct ptlrpc_sec_policy *policy = ctx->sc_policy;
1771 if (!policy->sp_sops->install_rctx)
1773 return policy->sp_sops->install_rctx(imp, ctx);
1776 /* Get SELinux policy info from userspace */
1777 static int sepol_helper(struct obd_import *imp)
1779 char mtime_str[21] = { 0 }, mode_str[2] = { 0 };
1781 [0] = "/usr/sbin/l_getsepol",
1783 [2] = NULL, /* obd type */
1785 [4] = NULL, /* obd name */
1787 [6] = mtime_str, /* policy mtime */
1789 [8] = mode_str, /* enforcing mode */
1794 [1] = "PATH=/sbin:/usr/sbin",
1800 if (imp == NULL || imp->imp_obd == NULL ||
1801 imp->imp_obd->obd_type == NULL) {
1804 argv[2] = imp->imp_obd->obd_type->typ_name;
1805 argv[4] = imp->imp_obd->obd_name;
1806 spin_lock(&imp->imp_sec->ps_lock);
1807 if (imp->imp_sec->ps_sepol_mtime == 0 &&
1808 imp->imp_sec->ps_sepol[0] == '\0') {
1809 /* ps_sepol has not been initialized */
1813 snprintf(mtime_str, sizeof(mtime_str), "%lu",
1814 imp->imp_sec->ps_sepol_mtime);
1815 mode_str[0] = imp->imp_sec->ps_sepol[0];
1817 spin_unlock(&imp->imp_sec->ps_lock);
1818 ret = call_usermodehelper(argv[0], argv, envp, UMH_WAIT_PROC);
1825 static inline int sptlrpc_sepol_needs_check(struct ptlrpc_sec *imp_sec)
1829 if (send_sepol == 0 || !selinux_is_enabled())
1832 if (send_sepol == -1)
1833 /* send_sepol == -1 means fetch sepol status every time */
1836 spin_lock(&imp_sec->ps_lock);
1837 checknext = imp_sec->ps_sepol_checknext;
1838 spin_unlock(&imp_sec->ps_lock);
1840 /* next check is too far in time, please update */
1841 if (ktime_after(checknext,
1842 ktime_add(ktime_get(), ktime_set(send_sepol, 0))))
1845 if (ktime_before(ktime_get(), checknext))
1846 /* too early to fetch sepol status */
1850 /* define new sepol_checknext time */
1851 spin_lock(&imp_sec->ps_lock);
1852 imp_sec->ps_sepol_checknext = ktime_add(ktime_get(),
1853 ktime_set(send_sepol, 0));
1854 spin_unlock(&imp_sec->ps_lock);
1859 int sptlrpc_get_sepol(struct ptlrpc_request *req)
1861 struct ptlrpc_sec *imp_sec = req->rq_import->imp_sec;
1866 (req->rq_sepol)[0] = '\0';
1868 #ifndef HAVE_SELINUX
1869 if (unlikely(send_sepol != 0))
1870 CDEBUG(D_SEC, "Client cannot report SELinux status, "
1871 "it was not built against libselinux.\n");
1875 if (send_sepol == 0 || !selinux_is_enabled())
1878 if (imp_sec == NULL)
1881 /* Retrieve SELinux status info */
1882 if (sptlrpc_sepol_needs_check(imp_sec))
1883 rc = sepol_helper(req->rq_import);
1884 if (likely(rc == 0)) {
1885 spin_lock(&imp_sec->ps_lock);
1886 memcpy(req->rq_sepol, imp_sec->ps_sepol,
1887 sizeof(req->rq_sepol));
1888 spin_unlock(&imp_sec->ps_lock);
1893 EXPORT_SYMBOL(sptlrpc_get_sepol);
1895 /****************************************
1896 * server side security *
1897 ****************************************/
1899 static int flavor_allowed(struct sptlrpc_flavor *exp,
1900 struct ptlrpc_request *req)
1902 struct sptlrpc_flavor *flvr = &req->rq_flvr;
1904 if (exp->sf_rpc == SPTLRPC_FLVR_ANY || exp->sf_rpc == flvr->sf_rpc)
1907 if ((req->rq_ctx_init || req->rq_ctx_fini) &&
1908 SPTLRPC_FLVR_POLICY(exp->sf_rpc) ==
1909 SPTLRPC_FLVR_POLICY(flvr->sf_rpc) &&
1910 SPTLRPC_FLVR_MECH(exp->sf_rpc) == SPTLRPC_FLVR_MECH(flvr->sf_rpc))
1916 #define EXP_FLVR_UPDATE_EXPIRE (OBD_TIMEOUT_DEFAULT + 10)
1919 * Given an export \a exp, check whether the flavor of incoming \a req
1920 * is allowed by the export \a exp. Main logic is about taking care of
1921 * changing configurations. Return 0 means success.
1923 int sptlrpc_target_export_check(struct obd_export *exp,
1924 struct ptlrpc_request *req)
1926 struct sptlrpc_flavor flavor;
1931 /* client side export has no imp_reverse, skip
1932 * FIXME maybe we should check flavor this as well??? */
1933 if (exp->exp_imp_reverse == NULL)
1936 /* don't care about ctx fini rpc */
1937 if (req->rq_ctx_fini)
1940 spin_lock(&exp->exp_lock);
1942 /* if flavor just changed (exp->exp_flvr_changed != 0), we wait for
1943 * the first req with the new flavor, then treat it as current flavor,
1944 * adapt reverse sec according to it.
1945 * note the first rpc with new flavor might not be with root ctx, in
1946 * which case delay the sec_adapt by leaving exp_flvr_adapt == 1. */
1947 if (unlikely(exp->exp_flvr_changed) &&
1948 flavor_allowed(&exp->exp_flvr_old[1], req)) {
1949 /* make the new flavor as "current", and old ones as
1950 * about-to-expire */
1951 CDEBUG(D_SEC, "exp %p: just changed: %x->%x\n", exp,
1952 exp->exp_flvr.sf_rpc, exp->exp_flvr_old[1].sf_rpc);
1953 flavor = exp->exp_flvr_old[1];
1954 exp->exp_flvr_old[1] = exp->exp_flvr_old[0];
1955 exp->exp_flvr_expire[1] = exp->exp_flvr_expire[0];
1956 exp->exp_flvr_old[0] = exp->exp_flvr;
1957 exp->exp_flvr_expire[0] = ktime_get_real_seconds() +
1958 EXP_FLVR_UPDATE_EXPIRE;
1959 exp->exp_flvr = flavor;
1961 /* flavor change finished */
1962 exp->exp_flvr_changed = 0;
1963 LASSERT(exp->exp_flvr_adapt == 1);
1965 /* if it's gss, we only interested in root ctx init */
1966 if (req->rq_auth_gss &&
1967 !(req->rq_ctx_init &&
1968 (req->rq_auth_usr_root || req->rq_auth_usr_mdt ||
1969 req->rq_auth_usr_ost))) {
1970 spin_unlock(&exp->exp_lock);
1971 CDEBUG(D_SEC, "is good but not root(%d:%d:%d:%d:%d)\n",
1972 req->rq_auth_gss, req->rq_ctx_init,
1973 req->rq_auth_usr_root, req->rq_auth_usr_mdt,
1974 req->rq_auth_usr_ost);
1978 exp->exp_flvr_adapt = 0;
1979 spin_unlock(&exp->exp_lock);
1981 return sptlrpc_import_sec_adapt(exp->exp_imp_reverse,
1982 req->rq_svc_ctx, &flavor);
1985 /* if it equals to the current flavor, we accept it, but need to
1986 * dealing with reverse sec/ctx */
1987 if (likely(flavor_allowed(&exp->exp_flvr, req))) {
1988 /* most cases should return here, we only interested in
1989 * gss root ctx init */
1990 if (!req->rq_auth_gss || !req->rq_ctx_init ||
1991 (!req->rq_auth_usr_root && !req->rq_auth_usr_mdt &&
1992 !req->rq_auth_usr_ost)) {
1993 spin_unlock(&exp->exp_lock);
1997 /* if flavor just changed, we should not proceed, just leave
1998 * it and current flavor will be discovered and replaced
1999 * shortly, and let _this_ rpc pass through */
2000 if (exp->exp_flvr_changed) {
2001 LASSERT(exp->exp_flvr_adapt);
2002 spin_unlock(&exp->exp_lock);
2006 if (exp->exp_flvr_adapt) {
2007 exp->exp_flvr_adapt = 0;
2008 CDEBUG(D_SEC, "exp %p (%x|%x|%x): do delayed adapt\n",
2009 exp, exp->exp_flvr.sf_rpc,
2010 exp->exp_flvr_old[0].sf_rpc,
2011 exp->exp_flvr_old[1].sf_rpc);
2012 flavor = exp->exp_flvr;
2013 spin_unlock(&exp->exp_lock);
2015 return sptlrpc_import_sec_adapt(exp->exp_imp_reverse,
2019 CDEBUG(D_SEC, "exp %p (%x|%x|%x): is current flavor, "
2020 "install rvs ctx\n", exp, exp->exp_flvr.sf_rpc,
2021 exp->exp_flvr_old[0].sf_rpc,
2022 exp->exp_flvr_old[1].sf_rpc);
2023 spin_unlock(&exp->exp_lock);
2025 return sptlrpc_svc_install_rvs_ctx(exp->exp_imp_reverse,
2030 if (exp->exp_flvr_expire[0]) {
2031 if (exp->exp_flvr_expire[0] >= ktime_get_real_seconds()) {
2032 if (flavor_allowed(&exp->exp_flvr_old[0], req)) {
2033 CDEBUG(D_SEC, "exp %p (%x|%x|%x): match the middle one (%lld)\n", exp,
2034 exp->exp_flvr.sf_rpc,
2035 exp->exp_flvr_old[0].sf_rpc,
2036 exp->exp_flvr_old[1].sf_rpc,
2037 (s64)(exp->exp_flvr_expire[0] -
2038 ktime_get_real_seconds()));
2039 spin_unlock(&exp->exp_lock);
2043 CDEBUG(D_SEC, "mark middle expired\n");
2044 exp->exp_flvr_expire[0] = 0;
2046 CDEBUG(D_SEC, "exp %p (%x|%x|%x): %x not match middle\n", exp,
2047 exp->exp_flvr.sf_rpc,
2048 exp->exp_flvr_old[0].sf_rpc, exp->exp_flvr_old[1].sf_rpc,
2049 req->rq_flvr.sf_rpc);
2052 /* now it doesn't match the current flavor, the only chance we can
2053 * accept it is match the old flavors which is not expired. */
2054 if (exp->exp_flvr_changed == 0 && exp->exp_flvr_expire[1]) {
2055 if (exp->exp_flvr_expire[1] >= ktime_get_real_seconds()) {
2056 if (flavor_allowed(&exp->exp_flvr_old[1], req)) {
2057 CDEBUG(D_SEC, "exp %p (%x|%x|%x): match the oldest one (%lld)\n",
2059 exp->exp_flvr.sf_rpc,
2060 exp->exp_flvr_old[0].sf_rpc,
2061 exp->exp_flvr_old[1].sf_rpc,
2062 (s64)(exp->exp_flvr_expire[1] -
2063 ktime_get_real_seconds()));
2064 spin_unlock(&exp->exp_lock);
2068 CDEBUG(D_SEC, "mark oldest expired\n");
2069 exp->exp_flvr_expire[1] = 0;
2071 CDEBUG(D_SEC, "exp %p (%x|%x|%x): %x not match found\n",
2072 exp, exp->exp_flvr.sf_rpc,
2073 exp->exp_flvr_old[0].sf_rpc, exp->exp_flvr_old[1].sf_rpc,
2074 req->rq_flvr.sf_rpc);
2076 CDEBUG(D_SEC, "exp %p (%x|%x|%x): skip the last one\n",
2077 exp, exp->exp_flvr.sf_rpc, exp->exp_flvr_old[0].sf_rpc,
2078 exp->exp_flvr_old[1].sf_rpc);
2081 spin_unlock(&exp->exp_lock);
2083 CWARN("exp %p(%s): req %p (%u|%u|%u|%u|%u|%u) with unauthorized flavor %x, expect %x|%x(%+lld)|%x(%+lld)\n",
2084 exp, exp->exp_obd->obd_name,
2085 req, req->rq_auth_gss, req->rq_ctx_init, req->rq_ctx_fini,
2086 req->rq_auth_usr_root, req->rq_auth_usr_mdt, req->rq_auth_usr_ost,
2087 req->rq_flvr.sf_rpc,
2088 exp->exp_flvr.sf_rpc,
2089 exp->exp_flvr_old[0].sf_rpc,
2090 exp->exp_flvr_expire[0] ?
2091 (s64)(exp->exp_flvr_expire[0] - ktime_get_real_seconds()) : 0,
2092 exp->exp_flvr_old[1].sf_rpc,
2093 exp->exp_flvr_expire[1] ?
2094 (s64)(exp->exp_flvr_expire[1] - ktime_get_real_seconds()) : 0);
2097 EXPORT_SYMBOL(sptlrpc_target_export_check);
2099 void sptlrpc_target_update_exp_flavor(struct obd_device *obd,
2100 struct sptlrpc_rule_set *rset)
2102 struct obd_export *exp;
2103 struct sptlrpc_flavor new_flvr;
2107 spin_lock(&obd->obd_dev_lock);
2109 list_for_each_entry(exp, &obd->obd_exports, exp_obd_chain) {
2110 if (exp->exp_connection == NULL)
2113 /* note if this export had just been updated flavor
2114 * (exp_flvr_changed == 1), this will override the
2116 spin_lock(&exp->exp_lock);
2117 sptlrpc_target_choose_flavor(rset, exp->exp_sp_peer,
2118 exp->exp_connection->c_peer.nid,
2120 if (exp->exp_flvr_changed ||
2121 !flavor_equal(&new_flvr, &exp->exp_flvr)) {
2122 exp->exp_flvr_old[1] = new_flvr;
2123 exp->exp_flvr_expire[1] = 0;
2124 exp->exp_flvr_changed = 1;
2125 exp->exp_flvr_adapt = 1;
2127 CDEBUG(D_SEC, "exp %p (%s): updated flavor %x->%x\n",
2128 exp, sptlrpc_part2name(exp->exp_sp_peer),
2129 exp->exp_flvr.sf_rpc,
2130 exp->exp_flvr_old[1].sf_rpc);
2132 spin_unlock(&exp->exp_lock);
2135 spin_unlock(&obd->obd_dev_lock);
2137 EXPORT_SYMBOL(sptlrpc_target_update_exp_flavor);
2139 static int sptlrpc_svc_check_from(struct ptlrpc_request *req, int svc_rc)
2141 /* peer's claim is unreliable unless gss is being used */
2142 if (!req->rq_auth_gss || svc_rc == SECSVC_DROP)
2145 switch (req->rq_sp_from) {
2147 if (req->rq_auth_usr_mdt || req->rq_auth_usr_ost) {
2148 DEBUG_REQ(D_ERROR, req, "faked source CLI");
2149 svc_rc = SECSVC_DROP;
2153 if (!req->rq_auth_usr_mdt) {
2154 DEBUG_REQ(D_ERROR, req, "faked source MDT");
2155 svc_rc = SECSVC_DROP;
2159 if (!req->rq_auth_usr_ost) {
2160 DEBUG_REQ(D_ERROR, req, "faked source OST");
2161 svc_rc = SECSVC_DROP;
2166 if (!req->rq_auth_usr_root && !req->rq_auth_usr_mdt &&
2167 !req->rq_auth_usr_ost) {
2168 DEBUG_REQ(D_ERROR, req, "faked source MGC/MGS");
2169 svc_rc = SECSVC_DROP;
2174 DEBUG_REQ(D_ERROR, req, "invalid source %u", req->rq_sp_from);
2175 svc_rc = SECSVC_DROP;
2182 * Used by ptlrpc server, to perform transformation upon request message of
2183 * incoming \a req. This must be the first thing to do with an incoming
2184 * request in ptlrpc layer.
2186 * \retval SECSVC_OK success, and req->rq_reqmsg point to request message in
2187 * clear text, size is req->rq_reqlen; also req->rq_svc_ctx is set.
2188 * \retval SECSVC_COMPLETE success, the request has been fully processed, and
2189 * reply message has been prepared.
2190 * \retval SECSVC_DROP failed, this request should be dropped.
2192 int sptlrpc_svc_unwrap_request(struct ptlrpc_request *req)
2194 struct ptlrpc_sec_policy *policy;
2195 struct lustre_msg *msg = req->rq_reqbuf;
2200 LASSERT(req->rq_reqmsg == NULL);
2201 LASSERT(req->rq_repmsg == NULL);
2202 LASSERT(req->rq_svc_ctx == NULL);
2204 req->rq_req_swab_mask = 0;
2206 rc = __lustre_unpack_msg(msg, req->rq_reqdata_len);
2209 lustre_set_req_swabbed(req, MSG_PTLRPC_HEADER_OFF);
2213 CERROR("error unpacking request from %s x%llu\n",
2214 libcfs_id2str(req->rq_peer), req->rq_xid);
2215 RETURN(SECSVC_DROP);
2218 req->rq_flvr.sf_rpc = WIRE_FLVR(msg->lm_secflvr);
2219 req->rq_sp_from = LUSTRE_SP_ANY;
2220 req->rq_auth_uid = -1; /* set to INVALID_UID */
2221 req->rq_auth_mapped_uid = -1;
2223 policy = sptlrpc_wireflavor2policy(req->rq_flvr.sf_rpc);
2225 CERROR("unsupported rpc flavor %x\n", req->rq_flvr.sf_rpc);
2226 RETURN(SECSVC_DROP);
2229 LASSERT(policy->sp_sops->accept);
2230 rc = policy->sp_sops->accept(req);
2231 sptlrpc_policy_put(policy);
2232 LASSERT(req->rq_reqmsg || rc != SECSVC_OK);
2233 LASSERT(req->rq_svc_ctx || rc == SECSVC_DROP);
2236 * if it's not null flavor (which means embedded packing msg),
2237 * reset the swab mask for the comming inner msg unpacking.
2239 if (SPTLRPC_FLVR_POLICY(req->rq_flvr.sf_rpc) != SPTLRPC_POLICY_NULL)
2240 req->rq_req_swab_mask = 0;
2242 /* sanity check for the request source */
2243 rc = sptlrpc_svc_check_from(req, rc);
2248 * Used by ptlrpc server, to allocate reply buffer for \a req. If succeed,
2249 * req->rq_reply_state is set, and req->rq_reply_state->rs_msg point to
2250 * a buffer of \a msglen size.
2252 int sptlrpc_svc_alloc_rs(struct ptlrpc_request *req, int msglen)
2254 struct ptlrpc_sec_policy *policy;
2255 struct ptlrpc_reply_state *rs;
2259 LASSERT(req->rq_svc_ctx);
2260 LASSERT(req->rq_svc_ctx->sc_policy);
2262 policy = req->rq_svc_ctx->sc_policy;
2263 LASSERT(policy->sp_sops->alloc_rs);
2265 rc = policy->sp_sops->alloc_rs(req, msglen);
2266 if (unlikely(rc == -ENOMEM)) {
2267 struct ptlrpc_service_part *svcpt = req->rq_rqbd->rqbd_svcpt;
2268 if (svcpt->scp_service->srv_max_reply_size <
2269 msglen + sizeof(struct ptlrpc_reply_state)) {
2270 /* Just return failure if the size is too big */
2271 CERROR("size of message is too big (%zd), %d allowed\n",
2272 msglen + sizeof(struct ptlrpc_reply_state),
2273 svcpt->scp_service->srv_max_reply_size);
2277 /* failed alloc, try emergency pool */
2278 rs = lustre_get_emerg_rs(svcpt);
2282 req->rq_reply_state = rs;
2283 rc = policy->sp_sops->alloc_rs(req, msglen);
2285 lustre_put_emerg_rs(rs);
2286 req->rq_reply_state = NULL;
2291 (req->rq_reply_state && req->rq_reply_state->rs_msg));
2297 * Used by ptlrpc server, to perform transformation upon reply message.
2299 * \post req->rq_reply_off is set to approriate server-controlled reply offset.
2300 * \post req->rq_repmsg and req->rq_reply_state->rs_msg becomes inaccessible.
2302 int sptlrpc_svc_wrap_reply(struct ptlrpc_request *req)
2304 struct ptlrpc_sec_policy *policy;
2308 LASSERT(req->rq_svc_ctx);
2309 LASSERT(req->rq_svc_ctx->sc_policy);
2311 policy = req->rq_svc_ctx->sc_policy;
2312 LASSERT(policy->sp_sops->authorize);
2314 rc = policy->sp_sops->authorize(req);
2315 LASSERT(rc || req->rq_reply_state->rs_repdata_len);
2321 * Used by ptlrpc server, to free reply_state.
2323 void sptlrpc_svc_free_rs(struct ptlrpc_reply_state *rs)
2325 struct ptlrpc_sec_policy *policy;
2326 unsigned int prealloc;
2329 LASSERT(rs->rs_svc_ctx);
2330 LASSERT(rs->rs_svc_ctx->sc_policy);
2332 policy = rs->rs_svc_ctx->sc_policy;
2333 LASSERT(policy->sp_sops->free_rs);
2335 prealloc = rs->rs_prealloc;
2336 policy->sp_sops->free_rs(rs);
2339 lustre_put_emerg_rs(rs);
2343 void sptlrpc_svc_ctx_addref(struct ptlrpc_request *req)
2345 struct ptlrpc_svc_ctx *ctx = req->rq_svc_ctx;
2348 atomic_inc(&ctx->sc_refcount);
2351 void sptlrpc_svc_ctx_decref(struct ptlrpc_request *req)
2353 struct ptlrpc_svc_ctx *ctx = req->rq_svc_ctx;
2358 LASSERT_ATOMIC_POS(&ctx->sc_refcount);
2359 if (atomic_dec_and_test(&ctx->sc_refcount)) {
2360 if (ctx->sc_policy->sp_sops->free_ctx)
2361 ctx->sc_policy->sp_sops->free_ctx(ctx);
2363 req->rq_svc_ctx = NULL;
2366 void sptlrpc_svc_ctx_invalidate(struct ptlrpc_request *req)
2368 struct ptlrpc_svc_ctx *ctx = req->rq_svc_ctx;
2373 LASSERT_ATOMIC_POS(&ctx->sc_refcount);
2374 if (ctx->sc_policy->sp_sops->invalidate_ctx)
2375 ctx->sc_policy->sp_sops->invalidate_ctx(ctx);
2377 EXPORT_SYMBOL(sptlrpc_svc_ctx_invalidate);
2379 /****************************************
2381 ****************************************/
2384 * Perform transformation upon bulk data pointed by \a desc. This is called
2385 * before transforming the request message.
2387 int sptlrpc_cli_wrap_bulk(struct ptlrpc_request *req,
2388 struct ptlrpc_bulk_desc *desc)
2390 struct ptlrpc_cli_ctx *ctx;
2392 LASSERT(req->rq_bulk_read || req->rq_bulk_write);
2394 if (!req->rq_pack_bulk)
2397 ctx = req->rq_cli_ctx;
2398 if (ctx->cc_ops->wrap_bulk)
2399 return ctx->cc_ops->wrap_bulk(ctx, req, desc);
2402 EXPORT_SYMBOL(sptlrpc_cli_wrap_bulk);
2405 * This is called after unwrap the reply message.
2406 * return nob of actual plain text size received, or error code.
2408 int sptlrpc_cli_unwrap_bulk_read(struct ptlrpc_request *req,
2409 struct ptlrpc_bulk_desc *desc,
2412 struct ptlrpc_cli_ctx *ctx;
2415 LASSERT(req->rq_bulk_read && !req->rq_bulk_write);
2417 if (!req->rq_pack_bulk)
2418 return desc->bd_nob_transferred;
2420 ctx = req->rq_cli_ctx;
2421 if (ctx->cc_ops->unwrap_bulk) {
2422 rc = ctx->cc_ops->unwrap_bulk(ctx, req, desc);
2426 return desc->bd_nob_transferred;
2428 EXPORT_SYMBOL(sptlrpc_cli_unwrap_bulk_read);
2431 * This is called after unwrap the reply message.
2432 * return 0 for success or error code.
2434 int sptlrpc_cli_unwrap_bulk_write(struct ptlrpc_request *req,
2435 struct ptlrpc_bulk_desc *desc)
2437 struct ptlrpc_cli_ctx *ctx;
2440 LASSERT(!req->rq_bulk_read && req->rq_bulk_write);
2442 if (!req->rq_pack_bulk)
2445 ctx = req->rq_cli_ctx;
2446 if (ctx->cc_ops->unwrap_bulk) {
2447 rc = ctx->cc_ops->unwrap_bulk(ctx, req, desc);
2453 * if everything is going right, nob should equals to nob_transferred.
2454 * in case of privacy mode, nob_transferred needs to be adjusted.
2456 if (desc->bd_nob != desc->bd_nob_transferred) {
2457 CERROR("nob %d doesn't match transferred nob %d\n",
2458 desc->bd_nob, desc->bd_nob_transferred);
2464 EXPORT_SYMBOL(sptlrpc_cli_unwrap_bulk_write);
2466 #ifdef HAVE_SERVER_SUPPORT
2468 * Performe transformation upon outgoing bulk read.
2470 int sptlrpc_svc_wrap_bulk(struct ptlrpc_request *req,
2471 struct ptlrpc_bulk_desc *desc)
2473 struct ptlrpc_svc_ctx *ctx;
2475 LASSERT(req->rq_bulk_read);
2477 if (!req->rq_pack_bulk)
2480 ctx = req->rq_svc_ctx;
2481 if (ctx->sc_policy->sp_sops->wrap_bulk)
2482 return ctx->sc_policy->sp_sops->wrap_bulk(req, desc);
2486 EXPORT_SYMBOL(sptlrpc_svc_wrap_bulk);
2489 * Performe transformation upon incoming bulk write.
2491 int sptlrpc_svc_unwrap_bulk(struct ptlrpc_request *req,
2492 struct ptlrpc_bulk_desc *desc)
2494 struct ptlrpc_svc_ctx *ctx;
2497 LASSERT(req->rq_bulk_write);
2500 * if it's in privacy mode, transferred should >= expected; otherwise
2501 * transferred should == expected.
2503 if (desc->bd_nob_transferred < desc->bd_nob ||
2504 (desc->bd_nob_transferred > desc->bd_nob &&
2505 SPTLRPC_FLVR_BULK_SVC(req->rq_flvr.sf_rpc) !=
2506 SPTLRPC_BULK_SVC_PRIV)) {
2507 DEBUG_REQ(D_ERROR, req, "truncated bulk GET %d(%d)",
2508 desc->bd_nob_transferred, desc->bd_nob);
2512 if (!req->rq_pack_bulk)
2515 ctx = req->rq_svc_ctx;
2516 if (ctx->sc_policy->sp_sops->unwrap_bulk) {
2517 rc = ctx->sc_policy->sp_sops->unwrap_bulk(req, desc);
2519 CERROR("error unwrap bulk: %d\n", rc);
2522 /* return 0 to allow reply be sent */
2525 EXPORT_SYMBOL(sptlrpc_svc_unwrap_bulk);
2528 * Prepare buffers for incoming bulk write.
2530 int sptlrpc_svc_prep_bulk(struct ptlrpc_request *req,
2531 struct ptlrpc_bulk_desc *desc)
2533 struct ptlrpc_svc_ctx *ctx;
2535 LASSERT(req->rq_bulk_write);
2537 if (!req->rq_pack_bulk)
2540 ctx = req->rq_svc_ctx;
2541 if (ctx->sc_policy->sp_sops->prep_bulk)
2542 return ctx->sc_policy->sp_sops->prep_bulk(req, desc);
2546 EXPORT_SYMBOL(sptlrpc_svc_prep_bulk);
2548 #endif /* HAVE_SERVER_SUPPORT */
2550 /****************************************
2551 * user descriptor helpers *
2552 ****************************************/
2554 int sptlrpc_current_user_desc_size(void)
2558 ngroups = current_ngroups;
2560 if (ngroups > LUSTRE_MAX_GROUPS)
2561 ngroups = LUSTRE_MAX_GROUPS;
2562 return sptlrpc_user_desc_size(ngroups);
2564 EXPORT_SYMBOL(sptlrpc_current_user_desc_size);
2566 int sptlrpc_pack_user_desc(struct lustre_msg *msg, int offset)
2568 struct ptlrpc_user_desc *pud;
2570 pud = lustre_msg_buf(msg, offset, 0);
2572 pud->pud_uid = from_kuid(&init_user_ns, current_uid());
2573 pud->pud_gid = from_kgid(&init_user_ns, current_gid());
2574 pud->pud_fsuid = from_kuid(&init_user_ns, current_fsuid());
2575 pud->pud_fsgid = from_kgid(&init_user_ns, current_fsgid());
2576 pud->pud_cap = cfs_curproc_cap_pack();
2577 pud->pud_ngroups = (msg->lm_buflens[offset] - sizeof(*pud)) / 4;
2580 if (pud->pud_ngroups > current_ngroups)
2581 pud->pud_ngroups = current_ngroups;
2582 #ifdef HAVE_GROUP_INFO_GID
2583 memcpy(pud->pud_groups, current_cred()->group_info->gid,
2584 pud->pud_ngroups * sizeof(__u32));
2585 #else /* !HAVE_GROUP_INFO_GID */
2586 memcpy(pud->pud_groups, current_cred()->group_info->blocks[0],
2587 pud->pud_ngroups * sizeof(__u32));
2588 #endif /* HAVE_GROUP_INFO_GID */
2589 task_unlock(current);
2593 EXPORT_SYMBOL(sptlrpc_pack_user_desc);
2595 int sptlrpc_unpack_user_desc(struct lustre_msg *msg, int offset, int swabbed)
2597 struct ptlrpc_user_desc *pud;
2600 pud = lustre_msg_buf(msg, offset, sizeof(*pud));
2605 __swab32s(&pud->pud_uid);
2606 __swab32s(&pud->pud_gid);
2607 __swab32s(&pud->pud_fsuid);
2608 __swab32s(&pud->pud_fsgid);
2609 __swab32s(&pud->pud_cap);
2610 __swab32s(&pud->pud_ngroups);
2613 if (pud->pud_ngroups > LUSTRE_MAX_GROUPS) {
2614 CERROR("%u groups is too large\n", pud->pud_ngroups);
2618 if (sizeof(*pud) + pud->pud_ngroups * sizeof(__u32) >
2619 msg->lm_buflens[offset]) {
2620 CERROR("%u groups are claimed but bufsize only %u\n",
2621 pud->pud_ngroups, msg->lm_buflens[offset]);
2626 for (i = 0; i < pud->pud_ngroups; i++)
2627 __swab32s(&pud->pud_groups[i]);
2632 EXPORT_SYMBOL(sptlrpc_unpack_user_desc);
2634 /****************************************
2636 ****************************************/
2638 const char * sec2target_str(struct ptlrpc_sec *sec)
2640 if (!sec || !sec->ps_import || !sec->ps_import->imp_obd)
2642 if (sec_is_reverse(sec))
2644 return obd_uuid2str(&sec->ps_import->imp_obd->u.cli.cl_target_uuid);
2646 EXPORT_SYMBOL(sec2target_str);
2649 * return true if the bulk data is protected
2651 int sptlrpc_flavor_has_bulk(struct sptlrpc_flavor *flvr)
2653 switch (SPTLRPC_FLVR_BULK_SVC(flvr->sf_rpc)) {
2654 case SPTLRPC_BULK_SVC_INTG:
2655 case SPTLRPC_BULK_SVC_PRIV:
2661 EXPORT_SYMBOL(sptlrpc_flavor_has_bulk);
2663 /****************************************
2664 * crypto API helper/alloc blkciper *
2665 ****************************************/
2667 /****************************************
2668 * initialize/finalize *
2669 ****************************************/
2671 int sptlrpc_init(void)
2675 rwlock_init(&policy_lock);
2677 rc = sptlrpc_gc_init();
2681 rc = sptlrpc_conf_init();
2685 rc = sptlrpc_enc_pool_init();
2689 rc = sptlrpc_null_init();
2693 rc = sptlrpc_plain_init();
2697 rc = sptlrpc_lproc_init();
2704 sptlrpc_plain_fini();
2706 sptlrpc_null_fini();
2708 sptlrpc_enc_pool_fini();
2710 sptlrpc_conf_fini();
2717 void sptlrpc_fini(void)
2719 sptlrpc_lproc_fini();
2720 sptlrpc_plain_fini();
2721 sptlrpc_null_fini();
2722 sptlrpc_enc_pool_fini();
2723 sptlrpc_conf_fini();